[{"data":1,"prerenderedAt":26102},["ShallowReactive",2],{"lesson:\u002Fdeep-learning\u002Fneural-networks\u002Factivation-functions":3,"course-wordcounts":20319,"ref-card-index":21231,"nav:deep-learning":25913,"tikz:d5420fb011ae83d099dd67a5d362030c1674155d780d49312eac0aba17952dfa":26096,"tikz:18a1f456063a5e8203ad7077ef6598e908dbdcc9ef2a6d8b55b14f8b746b28d2":26097,"tikz:c512f14b58a1aaf7adac256ecfe9ae71c72a9496a526ddffb30eab9b3722289f":26098,"tikz:e7820baa972e3a1eaae09b6f424924be719cd1b6b1c093798d3201bc1fb63445":26099,"tikz:b849c8d32183bc5ecebac9a03d43ab42e5cdbc6624322791077f620207aeea26":26100,"tikz:62319806674704a007bcd447462d8ac34a96595c69e86c30045eb2a4b216ef73":26101},{"id":4,"title":5,"blurb":6,"body":7,"brief":20286,"category":20287,"description":20288,"draft":20289,"extension":20290,"meta":20291,"module":20293,"navigation":20294,"path":20295,"practice":20296,"rawbody":20297,"readingTime":20298,"seo":20303,"sources":20304,"status":20314,"stem":20315,"summary":20316,"topics":20317,"__hash__":20318},"course\u002F06.deep-learning\u002F02.neural-networks\u002F02.activation-functions.md","Activation Functions","",{"type":8,"value":9,"toc":20264},"minimark",[10,145,320,1160,1165,1277,3797,3882,3886,3889,3893,4029,4731,5135,5266,5565,5568,6180,6705,6708,7539,7740,7743,7780,8102,8218,8222,8240,8665,8765,8768,8825,9290,9573,9576,9579,10033,10262,10266,10359,10717,10778,11020,11213,11216,11235,11857,11984,11988,11991,11994,11998,12034,12525,12661,12944,13423,13800,14421,15044,15191,15194,15197,15200,15204,15320,15592,15776,15818,16115,16172,16674,16885,16888,16891,17397,17529,17532,17536,17547,17708,18094,18410,18683,18826,18830,18833,19021,19053,19057,19060,19161,19480,19549,19553,20011],[11,12,13,14,105,106,128,129,144],"p",{},"A layer is an affine map followed by a pointwise nonlinearity,\n",[15,16,19],"span",{"className":17},[18],"katex",[15,20,24,52,90],{"className":21,"ariaHidden":23},[22],"katex-html","true",[15,25,28,33,39,44,49],{"className":26},[27],"base",[15,29],{"className":30,"style":32},[31],"strut","height:0.6944em;",[15,34,38],{"className":35},[36,37],"mord","mathnormal","h",[15,40],{"className":41,"style":43},[42],"mspace","margin-right:0.2778em;",[15,45,48],{"className":46},[47],"mrel","=",[15,50],{"className":51,"style":43},[42],[15,53,55,59,64,69,74,78,82,87],{"className":54},[27],[15,56],{"className":57,"style":58},[31],"height:1em;vertical-align:-0.25em;",[15,60,63],{"className":61,"style":62},[36,37],"margin-right:0.0359em;","g",[15,65,68],{"className":66},[67],"mopen","(",[15,70,73],{"className":71,"style":72},[36,37],"margin-right:0.1389em;","W",[15,75,77],{"className":76},[36,37],"x",[15,79],{"className":80,"style":81},[42],"margin-right:0.2222em;",[15,83,86],{"className":84},[85],"mbin","+",[15,88],{"className":89,"style":81},[42],[15,91,93,96,100],{"className":92},[27],[15,94],{"className":95,"style":58},[31],[15,97,99],{"className":98},[36,37],"b",[15,101,104],{"className":102},[103],"mclose",")",". The affine part carries all the parameters; the\n",[107,108,110,111,127],"a",{"href":109},"\u002Fdeep-learning\u002Fneural-networks\u002Fthe-multilayer-perceptron","nonlinearity ",[15,112,114],{"className":113},[18],[15,115,117],{"className":116,"ariaHidden":23},[22],[15,118,120,124],{"className":119},[27],[15,121],{"className":122,"style":123},[31],"height:0.625em;vertical-align:-0.1944em;",[15,125,63],{"className":126,"style":62},[36,37]," supplies the entire nonlinear effect",":\nstrip it out and a deep stack collapses to a single linear map. The choice of ",[15,130,132],{"className":131},[18],[15,133,135],{"className":134,"ariaHidden":23},[22],[15,136,138,141],{"className":137},[27],[15,139],{"className":140,"style":123},[31],[15,142,63],{"className":143,"style":62},[36,37],"\ntherefore matters: it sets how gradients flow back through depth, whether units\nsaturate, and how fast the network trains. This lesson catalogs the standard\nchoices and their derivatives.",[146,147,149],"callout",{"type":148},"definition",[11,150,151,155,156,212,213,271,272,314,315,319],{},[152,153,154],"strong",{},"Definition (Activation function)."," A function ",[15,157,159],{"className":158},[18],[15,160,162,181,203],{"className":161,"ariaHidden":23},[22],[15,163,165,168,171,174,178],{"className":164},[27],[15,166],{"className":167,"style":123},[31],[15,169,63],{"className":170,"style":62},[36,37],[15,172],{"className":173,"style":43},[42],[15,175,177],{"className":176},[47],":",[15,179],{"className":180,"style":43},[42],[15,182,184,188,193,196,200],{"className":183},[27],[15,185],{"className":186,"style":187},[31],"height:0.6889em;",[15,189,192],{"className":190},[36,191],"mathbb","R",[15,194],{"className":195,"style":43},[42],[15,197,199],{"className":198},[47],"→",[15,201],{"className":202,"style":43},[42],[15,204,206,209],{"className":205},[27],[15,207],{"className":208,"style":187},[31],[15,210,192],{"className":211},[36,191],"\napplied elementwise to the pre-activation ",[15,214,216],{"className":215},[18],[15,217,219,240,262],{"className":218,"ariaHidden":23},[22],[15,220,222,226,231,234,237],{"className":221},[27],[15,223],{"className":224,"style":225},[31],"height:0.4306em;",[15,227,230],{"className":228,"style":229},[36,37],"margin-right:0.044em;","z",[15,232],{"className":233,"style":43},[42],[15,235,48],{"className":236},[47],[15,238],{"className":239,"style":43},[42],[15,241,243,247,250,253,256,259],{"className":242},[27],[15,244],{"className":245,"style":246},[31],"height:0.7667em;vertical-align:-0.0833em;",[15,248,73],{"className":249,"style":72},[36,37],[15,251,77],{"className":252},[36,37],[15,254],{"className":255,"style":81},[42],[15,257,86],{"className":258},[85],[15,260],{"className":261,"style":81},[42],[15,263,265,268],{"className":264},[27],[15,266],{"className":267,"style":32},[31],[15,269,99],{"className":270},[36,37]," to produce a layer's\noutput ",[15,273,275],{"className":274},[18],[15,276,278,296],{"className":277,"ariaHidden":23},[22],[15,279,281,284,287,290,293],{"className":280},[27],[15,282],{"className":283,"style":32},[31],[15,285,38],{"className":286},[36,37],[15,288],{"className":289,"style":43},[42],[15,291,48],{"className":292},[47],[15,294],{"className":295,"style":43},[42],[15,297,299,302,305,308,311],{"className":298},[27],[15,300],{"className":301,"style":58},[31],[15,303,63],{"className":304,"style":62},[36,37],[15,306,68],{"className":307},[67],[15,309,230],{"className":310,"style":229},[36,37],[15,312,104],{"className":313},[103],". It must be nonlinear (else depth collapses) and almost\neverywhere differentiable (so ",[107,316,318],{"href":317},"\u002Fdeep-learning\u002Fneural-networks\u002Fbackpropagation","backpropagation","\ncan pass a gradient through it).",[11,321,322,323,338,339,885,886,979,980,1127,1128,1143,1144,1159],{},"To make the requirement precise, suppose two\nlayers with ",[15,324,326],{"className":325},[18],[15,327,329],{"className":328,"ariaHidden":23},[22],[15,330,332,335],{"className":331},[27],[15,333],{"className":334,"style":123},[31],[15,336,63],{"className":337,"style":62},[36,37]," the identity: ",[15,340,342],{"className":341},[18],[15,343,345,418,521,579,634,738,836],{"className":344,"ariaHidden":23},[22],[15,346,348,352,409,412,415],{"className":347},[27],[15,349],{"className":350,"style":351},[31],"height:0.8444em;vertical-align:-0.15em;",[15,353,355,358],{"className":354},[36],[15,356,38],{"className":357},[36,37],[15,359,362],{"className":360},[361],"msupsub",[15,363,367,400],{"className":364},[365,366],"vlist-t","vlist-t2",[15,368,371,395],{"className":369},[370],"vlist-r",[15,372,376],{"className":373,"style":375},[374],"vlist","height:0.3011em;",[15,377,379,384],{"style":378},"top:-2.55em;margin-left:0em;margin-right:0.05em;",[15,380],{"className":381,"style":383},[382],"pstrut","height:2.7em;",[15,385,391],{"className":386},[387,388,389,390],"sizing","reset-size6","size3","mtight",[15,392,394],{"className":393},[36,390],"2",[15,396,399],{"className":397},[398],"vlist-s","​",[15,401,403],{"className":402},[370],[15,404,407],{"className":405,"style":406},[374],"height:0.15em;",[15,408],{},[15,410],{"className":411,"style":43},[42],[15,413,48],{"className":414},[47],[15,416],{"className":417,"style":43},[42],[15,419,421,424,465,468,509,512,515,518],{"className":420},[27],[15,422],{"className":423,"style":58},[31],[15,425,427,430],{"className":426},[36],[15,428,73],{"className":429,"style":72},[36,37],[15,431,433],{"className":432},[361],[15,434,436,457],{"className":435},[365,366],[15,437,439,454],{"className":438},[370],[15,440,442],{"className":441,"style":375},[374],[15,443,445,448],{"style":444},"top:-2.55em;margin-left:-0.1389em;margin-right:0.05em;",[15,446],{"className":447,"style":383},[382],[15,449,451],{"className":450},[387,388,389,390],[15,452,394],{"className":453},[36,390],[15,455,399],{"className":456},[398],[15,458,460],{"className":459},[370],[15,461,463],{"className":462,"style":406},[374],[15,464],{},[15,466,68],{"className":467},[67],[15,469,471,474],{"className":470},[36],[15,472,73],{"className":473,"style":72},[36,37],[15,475,477],{"className":476},[361],[15,478,480,501],{"className":479},[365,366],[15,481,483,498],{"className":482},[370],[15,484,486],{"className":485,"style":375},[374],[15,487,488,491],{"style":444},[15,489],{"className":490,"style":383},[382],[15,492,494],{"className":493},[387,388,389,390],[15,495,497],{"className":496},[36,390],"1",[15,499,399],{"className":500},[398],[15,502,504],{"className":503},[370],[15,505,507],{"className":506,"style":406},[374],[15,508],{},[15,510,77],{"className":511},[36,37],[15,513],{"className":514,"style":81},[42],[15,516,86],{"className":517},[85],[15,519],{"className":520,"style":81},[42],[15,522,524,527,567,570,573,576],{"className":523},[27],[15,525],{"className":526,"style":58},[31],[15,528,530,533],{"className":529},[36],[15,531,99],{"className":532},[36,37],[15,534,536],{"className":535},[361],[15,537,539,559],{"className":538},[365,366],[15,540,542,556],{"className":541},[370],[15,543,545],{"className":544,"style":375},[374],[15,546,547,550],{"style":378},[15,548],{"className":549,"style":383},[382],[15,551,553],{"className":552},[387,388,389,390],[15,554,497],{"className":555},[36,390],[15,557,399],{"className":558},[398],[15,560,562],{"className":561},[370],[15,563,565],{"className":564,"style":406},[374],[15,566],{},[15,568,104],{"className":569},[103],[15,571],{"className":572,"style":81},[42],[15,574,86],{"className":575},[85],[15,577],{"className":578,"style":81},[42],[15,580,582,585,625,628,631],{"className":581},[27],[15,583],{"className":584,"style":351},[31],[15,586,588,591],{"className":587},[36],[15,589,99],{"className":590},[36,37],[15,592,594],{"className":593},[361],[15,595,597,617],{"className":596},[365,366],[15,598,600,614],{"className":599},[370],[15,601,603],{"className":602,"style":375},[374],[15,604,605,608],{"style":378},[15,606],{"className":607,"style":383},[382],[15,609,611],{"className":610},[387,388,389,390],[15,612,394],{"className":613},[36,390],[15,615,399],{"className":616},[398],[15,618,620],{"className":619},[370],[15,621,623],{"className":622,"style":406},[374],[15,624],{},[15,626],{"className":627,"style":43},[42],[15,629,48],{"className":630},[47],[15,632],{"className":633,"style":43},[42],[15,635,637,640,643,683,723,726,729,732,735],{"className":636},[27],[15,638],{"className":639,"style":58},[31],[15,641,68],{"className":642},[67],[15,644,646,649],{"className":645},[36],[15,647,73],{"className":648,"style":72},[36,37],[15,650,652],{"className":651},[361],[15,653,655,675],{"className":654},[365,366],[15,656,658,672],{"className":657},[370],[15,659,661],{"className":660,"style":375},[374],[15,662,663,666],{"style":444},[15,664],{"className":665,"style":383},[382],[15,667,669],{"className":668},[387,388,389,390],[15,670,394],{"className":671},[36,390],[15,673,399],{"className":674},[398],[15,676,678],{"className":677},[370],[15,679,681],{"className":680,"style":406},[374],[15,682],{},[15,684,686,689],{"className":685},[36],[15,687,73],{"className":688,"style":72},[36,37],[15,690,692],{"className":691},[361],[15,693,695,715],{"className":694},[365,366],[15,696,698,712],{"className":697},[370],[15,699,701],{"className":700,"style":375},[374],[15,702,703,706],{"style":444},[15,704],{"className":705,"style":383},[382],[15,707,709],{"className":708},[387,388,389,390],[15,710,497],{"className":711},[36,390],[15,713,399],{"className":714},[398],[15,716,718],{"className":717},[370],[15,719,721],{"className":720,"style":406},[374],[15,722],{},[15,724,104],{"className":725},[103],[15,727,77],{"className":728},[36,37],[15,730],{"className":731,"style":81},[42],[15,733,86],{"className":734},[85],[15,736],{"className":737,"style":81},[42],[15,739,741,744,747,787,827,830,833],{"className":740},[27],[15,742],{"className":743,"style":58},[31],[15,745,68],{"className":746},[67],[15,748,750,753],{"className":749},[36],[15,751,73],{"className":752,"style":72},[36,37],[15,754,756],{"className":755},[361],[15,757,759,779],{"className":758},[365,366],[15,760,762,776],{"className":761},[370],[15,763,765],{"className":764,"style":375},[374],[15,766,767,770],{"style":444},[15,768],{"className":769,"style":383},[382],[15,771,773],{"className":772},[387,388,389,390],[15,774,394],{"className":775},[36,390],[15,777,399],{"className":778},[398],[15,780,782],{"className":781},[370],[15,783,785],{"className":784,"style":406},[374],[15,786],{},[15,788,790,793],{"className":789},[36],[15,791,99],{"className":792},[36,37],[15,794,796],{"className":795},[361],[15,797,799,819],{"className":798},[365,366],[15,800,802,816],{"className":801},[370],[15,803,805],{"className":804,"style":375},[374],[15,806,807,810],{"style":378},[15,808],{"className":809,"style":383},[382],[15,811,813],{"className":812},[387,388,389,390],[15,814,497],{"className":815},[36,390],[15,817,399],{"className":818},[398],[15,820,822],{"className":821},[370],[15,823,825],{"className":824,"style":406},[374],[15,826],{},[15,828],{"className":829,"style":81},[42],[15,831,86],{"className":832},[85],[15,834],{"className":835,"style":81},[42],[15,837,839,842,882],{"className":838},[27],[15,840],{"className":841,"style":58},[31],[15,843,845,848],{"className":844},[36],[15,846,99],{"className":847},[36,37],[15,849,851],{"className":850},[361],[15,852,854,874],{"className":853},[365,366],[15,855,857,871],{"className":856},[370],[15,858,860],{"className":859,"style":375},[374],[15,861,862,865],{"style":378},[15,863],{"className":864,"style":383},[382],[15,866,868],{"className":867},[387,388,389,390],[15,869,394],{"className":870},[36,390],[15,872,399],{"className":873},[398],[15,875,877],{"className":876},[370],[15,878,880],{"className":879,"style":406},[374],[15,881],{},[15,883,104],{"className":884},[103],". The composite is one affine map with weight ",[15,887,889],{"className":888},[18],[15,890,892],{"className":891,"ariaHidden":23},[22],[15,893,895,899,939],{"className":894},[27],[15,896],{"className":897,"style":898},[31],"height:0.8333em;vertical-align:-0.15em;",[15,900,902,905],{"className":901},[36],[15,903,73],{"className":904,"style":72},[36,37],[15,906,908],{"className":907},[361],[15,909,911,931],{"className":910},[365,366],[15,912,914,928],{"className":913},[370],[15,915,917],{"className":916,"style":375},[374],[15,918,919,922],{"style":444},[15,920],{"className":921,"style":383},[382],[15,923,925],{"className":924},[387,388,389,390],[15,926,394],{"className":927},[36,390],[15,929,399],{"className":930},[398],[15,932,934],{"className":933},[370],[15,935,937],{"className":936,"style":406},[374],[15,938],{},[15,940,942,945],{"className":941},[36],[15,943,73],{"className":944,"style":72},[36,37],[15,946,948],{"className":947},[361],[15,949,951,971],{"className":950},[365,366],[15,952,954,968],{"className":953},[370],[15,955,957],{"className":956,"style":375},[374],[15,958,959,962],{"style":444},[15,960],{"className":961,"style":383},[382],[15,963,965],{"className":964},[387,388,389,390],[15,966,497],{"className":967},[36,390],[15,969,399],{"className":970},[398],[15,972,974],{"className":973},[370],[15,975,977],{"className":976,"style":406},[374],[15,978],{}," and bias\n",[15,981,983],{"className":982},[18],[15,984,986,1081],{"className":985,"ariaHidden":23},[22],[15,987,989,992,1032,1072,1075,1078],{"className":988},[27],[15,990],{"className":991,"style":351},[31],[15,993,995,998],{"className":994},[36],[15,996,73],{"className":997,"style":72},[36,37],[15,999,1001],{"className":1000},[361],[15,1002,1004,1024],{"className":1003},[365,366],[15,1005,1007,1021],{"className":1006},[370],[15,1008,1010],{"className":1009,"style":375},[374],[15,1011,1012,1015],{"style":444},[15,1013],{"className":1014,"style":383},[382],[15,1016,1018],{"className":1017},[387,388,389,390],[15,1019,394],{"className":1020},[36,390],[15,1022,399],{"className":1023},[398],[15,1025,1027],{"className":1026},[370],[15,1028,1030],{"className":1029,"style":406},[374],[15,1031],{},[15,1033,1035,1038],{"className":1034},[36],[15,1036,99],{"className":1037},[36,37],[15,1039,1041],{"className":1040},[361],[15,1042,1044,1064],{"className":1043},[365,366],[15,1045,1047,1061],{"className":1046},[370],[15,1048,1050],{"className":1049,"style":375},[374],[15,1051,1052,1055],{"style":378},[15,1053],{"className":1054,"style":383},[382],[15,1056,1058],{"className":1057},[387,388,389,390],[15,1059,497],{"className":1060},[36,390],[15,1062,399],{"className":1063},[398],[15,1065,1067],{"className":1066},[370],[15,1068,1070],{"className":1069,"style":406},[374],[15,1071],{},[15,1073],{"className":1074,"style":81},[42],[15,1076,86],{"className":1077},[85],[15,1079],{"className":1080,"style":81},[42],[15,1082,1084,1087],{"className":1083},[27],[15,1085],{"className":1086,"style":351},[31],[15,1088,1090,1093],{"className":1089},[36],[15,1091,99],{"className":1092},[36,37],[15,1094,1096],{"className":1095},[361],[15,1097,1099,1119],{"className":1098},[365,366],[15,1100,1102,1116],{"className":1101},[370],[15,1103,1105],{"className":1104,"style":375},[374],[15,1106,1107,1110],{"style":378},[15,1108],{"className":1109,"style":383},[382],[15,1111,1113],{"className":1112},[387,388,389,390],[15,1114,394],{"className":1115},[36,390],[15,1117,399],{"className":1118},[398],[15,1120,1122],{"className":1121},[370],[15,1123,1125],{"className":1124,"style":406},[374],[15,1126],{},", and no amount of stacking escapes the affine class. A single\nnonlinear ",[15,1129,1131],{"className":1130},[18],[15,1132,1134],{"className":1133,"ariaHidden":23},[22],[15,1135,1137,1140],{"className":1136},[27],[15,1138],{"className":1139,"style":123},[31],[15,1141,63],{"className":1142,"style":62},[36,37]," between the two multiplications is what lets a network represent a\nfunction that no single affine map can. Everything below is about which ",[15,1145,1147],{"className":1146},[18],[15,1148,1150],{"className":1149,"ariaHidden":23},[22],[15,1151,1153,1156],{"className":1152},[27],[15,1154],{"className":1155,"style":123},[31],[15,1157,63],{"className":1158,"style":62},[36,37]," to\npick and what each choice costs in the backward pass.",[1161,1162,1164],"h2",{"id":1163},"the-catalog","The catalog",[11,1166,1167,1168,1225,1226,1230,1231,1276],{},"Every hidden unit in common use is one scalar function and its derivative. The\nderivative column matters most: during the backward pass the gradient at a\nunit is multiplied by ",[15,1169,1171],{"className":1170},[18],[15,1172,1174],{"className":1173,"ariaHidden":23},[22],[15,1175,1177,1181,1216,1219,1222],{"className":1176},[27],[15,1178],{"className":1179,"style":1180},[31],"height:1.0019em;vertical-align:-0.25em;",[15,1182,1184,1187],{"className":1183},[36],[15,1185,63],{"className":1186,"style":62},[36,37],[15,1188,1190],{"className":1189},[361],[15,1191,1193],{"className":1192},[365],[15,1194,1196],{"className":1195},[370],[15,1197,1200],{"className":1198,"style":1199},[374],"height:0.7519em;",[15,1201,1203,1206],{"style":1202},"top:-3.063em;margin-right:0.05em;",[15,1204],{"className":1205,"style":383},[382],[15,1207,1209],{"className":1208},[387,388,389,390],[15,1210,1212],{"className":1211},[36,390],[15,1213,1215],{"className":1214},[36,390],"′",[15,1217,68],{"className":1218},[67],[15,1220,230],{"className":1221,"style":229},[36,37],[15,1223,104],{"className":1224},[103],", so the ",[1227,1228,1229],"em",{},"size"," of ",[15,1232,1234],{"className":1233},[18],[15,1235,1237],{"className":1236,"ariaHidden":23},[22],[15,1238,1240,1244],{"className":1239},[27],[15,1241],{"className":1242,"style":1243},[31],"height:0.9463em;vertical-align:-0.1944em;",[15,1245,1247,1250],{"className":1246},[36],[15,1248,63],{"className":1249,"style":62},[36,37],[15,1251,1253],{"className":1252},[361],[15,1254,1256],{"className":1255},[365],[15,1257,1259],{"className":1258},[370],[15,1260,1262],{"className":1261,"style":1199},[374],[15,1263,1264,1267],{"style":1202},[15,1265],{"className":1266,"style":383},[382],[15,1268,1270],{"className":1269},[387,388,389,390],[15,1271,1273],{"className":1272},[36,390],[15,1274,1215],{"className":1275},[36,390]," controls how much signal\npasses through. The following table is the reference the rest of the lesson expands.",[1278,1279,1280,1380],"table",{},[1281,1282,1283],"thead",{},[1284,1285,1286,1290,1316,1371,1374,1377],"tr",{},[1287,1288,1289],"th",{},"name",[1287,1291,1292],{},[15,1293,1295],{"className":1294},[18],[15,1296,1298],{"className":1297,"ariaHidden":23},[22],[15,1299,1301,1304,1307,1310,1313],{"className":1300},[27],[15,1302],{"className":1303,"style":58},[31],[15,1305,63],{"className":1306,"style":62},[36,37],[15,1308,68],{"className":1309},[67],[15,1311,230],{"className":1312,"style":229},[36,37],[15,1314,104],{"className":1315},[103],[1287,1317,1318],{},[15,1319,1321],{"className":1320},[18],[15,1322,1324],{"className":1323,"ariaHidden":23},[22],[15,1325,1327,1330,1362,1365,1368],{"className":1326},[27],[15,1328],{"className":1329,"style":1180},[31],[15,1331,1333,1336],{"className":1332},[36],[15,1334,63],{"className":1335,"style":62},[36,37],[15,1337,1339],{"className":1338},[361],[15,1340,1342],{"className":1341},[365],[15,1343,1345],{"className":1344},[370],[15,1346,1348],{"className":1347,"style":1199},[374],[15,1349,1350,1353],{"style":1202},[15,1351],{"className":1352,"style":383},[382],[15,1354,1356],{"className":1355},[387,388,389,390],[15,1357,1359],{"className":1358},[36,390],[15,1360,1215],{"className":1361},[36,390],[15,1363,68],{"className":1364},[67],[15,1366,230],{"className":1367,"style":229},[36,37],[15,1369,104],{"className":1370},[103],[1287,1372,1373],{},"range",[1287,1375,1376],{},"saturates?",[1287,1378,1379],{},"notes",[1381,1382,1383,1745,2151,2283,2510,2630,2912,3079,3287,3470],"tbody",{},[1284,1384,1385,1389,1524,1569,1605,1608],{},[1386,1387,1388],"td",{},"sigmoid",[1386,1390,1391],{},[15,1392,1394],{"className":1393},[18],[15,1395,1397],{"className":1396,"ariaHidden":23},[22],[15,1398,1400,1404],{"className":1399},[27],[15,1401],{"className":1402,"style":1403},[31],"height:2.0908em;vertical-align:-0.7693em;",[15,1405,1407,1411,1521],{"className":1406},[36],[15,1408],{"className":1409},[67,1410],"nulldelimiter",[15,1412,1415],{"className":1413},[1414],"mfrac",[15,1416,1418,1512],{"className":1417},[365,366],[15,1419,1421,1509],{"className":1420},[370],[15,1422,1425,1486,1497],{"className":1423,"style":1424},[374],"height:1.3214em;",[15,1426,1428,1432],{"style":1427},"top:-2.314em;",[15,1429],{"className":1430,"style":1431},[382],"height:3em;",[15,1433,1435,1438,1441,1444,1447],{"className":1434},[36],[15,1436,497],{"className":1437},[36],[15,1439],{"className":1440,"style":81},[42],[15,1442,86],{"className":1443},[85],[15,1445],{"className":1446,"style":81},[42],[15,1448,1450,1454],{"className":1449},[36],[15,1451,1453],{"className":1452},[36,37],"e",[15,1455,1457],{"className":1456},[361],[15,1458,1460],{"className":1459},[365],[15,1461,1463],{"className":1462},[370],[15,1464,1467],{"className":1465,"style":1466},[374],"height:0.6973em;",[15,1468,1470,1473],{"style":1469},"top:-2.989em;margin-right:0.05em;",[15,1471],{"className":1472,"style":383},[382],[15,1474,1476],{"className":1475},[387,388,389,390],[15,1477,1479,1483],{"className":1478},[36,390],[15,1480,1482],{"className":1481},[36,390],"−",[15,1484,230],{"className":1485,"style":229},[36,37,390],[15,1487,1489,1492],{"style":1488},"top:-3.23em;",[15,1490],{"className":1491,"style":1431},[382],[15,1493],{"className":1494,"style":1496},[1495],"frac-line","border-bottom-width:0.04em;",[15,1498,1500,1503],{"style":1499},"top:-3.677em;",[15,1501],{"className":1502,"style":1431},[382],[15,1504,1506],{"className":1505},[36],[15,1507,497],{"className":1508},[36],[15,1510,399],{"className":1511},[398],[15,1513,1515],{"className":1514},[370],[15,1516,1519],{"className":1517,"style":1518},[374],"height:0.7693em;",[15,1520],{},[15,1522],{"className":1523},[103,1410],[1386,1525,1526],{},[15,1527,1529],{"className":1528},[18],[15,1530,1532,1557],{"className":1531,"ariaHidden":23},[22],[15,1533,1535,1538,1542,1545,1548,1551,1554],{"className":1534},[27],[15,1536],{"className":1537,"style":58},[31],[15,1539,1541],{"className":1540,"style":62},[36,37],"σ",[15,1543,68],{"className":1544},[67],[15,1546,497],{"className":1547},[36],[15,1549],{"className":1550,"style":81},[42],[15,1552,1482],{"className":1553},[85],[15,1555],{"className":1556,"style":81},[42],[15,1558,1560,1563,1566],{"className":1559},[27],[15,1561],{"className":1562,"style":58},[31],[15,1564,1541],{"className":1565,"style":62},[36,37],[15,1567,104],{"className":1568},[103],[1386,1570,1571],{},[15,1572,1574],{"className":1573},[18],[15,1575,1577],{"className":1576,"ariaHidden":23},[22],[15,1578,1580,1583,1586,1590,1595,1599,1602],{"className":1579},[27],[15,1581],{"className":1582,"style":58},[31],[15,1584,68],{"className":1585},[67],[15,1587,1589],{"className":1588},[36],"0",[15,1591,1594],{"className":1592},[1593],"mpunct",",",[15,1596],{"className":1597,"style":1598},[42],"margin-right:0.1667em;",[15,1600,497],{"className":1601},[36],[15,1603,104],{"className":1604},[103],[1386,1606,1607],{},"both tails",[1386,1609,1610,1744],{},[15,1611,1613],{"className":1612},[18],[15,1614,1616,1664],{"className":1615,"ariaHidden":23},[22],[15,1617,1619,1622,1654,1657,1661],{"className":1618},[27],[15,1620],{"className":1621,"style":1243},[31],[15,1623,1625,1628],{"className":1624},[36],[15,1626,63],{"className":1627,"style":62},[36,37],[15,1629,1631],{"className":1630},[361],[15,1632,1634],{"className":1633},[365],[15,1635,1637],{"className":1636},[370],[15,1638,1640],{"className":1639,"style":1199},[374],[15,1641,1642,1645],{"style":1202},[15,1643],{"className":1644,"style":383},[382],[15,1646,1648],{"className":1647},[387,388,389,390],[15,1649,1651],{"className":1650},[36,390],[15,1652,1215],{"className":1653},[36,390],[15,1655],{"className":1656,"style":43},[42],[15,1658,1660],{"className":1659},[47],"≤",[15,1662],{"className":1663,"style":43},[42],[15,1665,1667,1671],{"className":1666},[27],[15,1668],{"className":1669,"style":1670},[31],"height:1.1901em;vertical-align:-0.345em;",[15,1672,1674,1677,1741],{"className":1673},[36],[15,1675],{"className":1676},[67,1410],[15,1678,1680],{"className":1679},[1414],[15,1681,1683,1732],{"className":1682},[365,366],[15,1684,1686,1729],{"className":1685},[370],[15,1687,1690,1706,1714],{"className":1688,"style":1689},[374],"height:0.8451em;",[15,1691,1693,1696],{"style":1692},"top:-2.655em;",[15,1694],{"className":1695,"style":1431},[382],[15,1697,1699],{"className":1698},[387,388,389,390],[15,1700,1702],{"className":1701},[36,390],[15,1703,1705],{"className":1704},[36,390],"4",[15,1707,1708,1711],{"style":1488},[15,1709],{"className":1710,"style":1431},[382],[15,1712],{"className":1713,"style":1496},[1495],[15,1715,1717,1720],{"style":1716},"top:-3.394em;",[15,1718],{"className":1719,"style":1431},[382],[15,1721,1723],{"className":1722},[387,388,389,390],[15,1724,1726],{"className":1725},[36,390],[15,1727,497],{"className":1728},[36,390],[15,1730,399],{"className":1731},[398],[15,1733,1735],{"className":1734},[370],[15,1736,1739],{"className":1737,"style":1738},[374],"height:0.345em;",[15,1740],{},[15,1742],{"className":1743},[103,1410],"; not zero-centered",[1284,1746,1747,1750,1977,2048,2083,2085],{},[1386,1748,1749],{},"tanh",[1386,1751,1752],{},[15,1753,1755],{"className":1754},[18],[15,1756,1758],{"className":1757,"ariaHidden":23},[22],[15,1759,1761,1765],{"className":1760},[27],[15,1762],{"className":1763,"style":1764},[31],"height:2.2177em;vertical-align:-0.7693em;",[15,1766,1768,1771,1974],{"className":1767},[36],[15,1769],{"className":1770},[67,1410],[15,1772,1774],{"className":1773},[1414],[15,1775,1777,1966],{"className":1776},[365,366],[15,1778,1780,1963],{"className":1779},[370],[15,1781,1784,1869,1877],{"className":1782,"style":1783},[374],"height:1.4483em;",[15,1785,1786,1789],{"style":1427},[15,1787],{"className":1788,"style":1431},[382],[15,1790,1792,1825,1828,1831,1834],{"className":1791},[36],[15,1793,1795,1798],{"className":1794},[36],[15,1796,1453],{"className":1797},[36,37],[15,1799,1801],{"className":1800},[361],[15,1802,1804],{"className":1803},[365],[15,1805,1807],{"className":1806},[370],[15,1808,1811],{"className":1809,"style":1810},[374],"height:0.5904em;",[15,1812,1813,1816],{"style":1469},[15,1814],{"className":1815,"style":383},[382],[15,1817,1819],{"className":1818},[387,388,389,390],[15,1820,1822],{"className":1821},[36,390],[15,1823,230],{"className":1824,"style":229},[36,37,390],[15,1826],{"className":1827,"style":81},[42],[15,1829,86],{"className":1830},[85],[15,1832],{"className":1833,"style":81},[42],[15,1835,1837,1840],{"className":1836},[36],[15,1838,1453],{"className":1839},[36,37],[15,1841,1843],{"className":1842},[361],[15,1844,1846],{"className":1845},[365],[15,1847,1849],{"className":1848},[370],[15,1850,1852],{"className":1851,"style":1466},[374],[15,1853,1854,1857],{"style":1469},[15,1855],{"className":1856,"style":383},[382],[15,1858,1860],{"className":1859},[387,388,389,390],[15,1861,1863,1866],{"className":1862},[36,390],[15,1864,1482],{"className":1865},[36,390],[15,1867,230],{"className":1868,"style":229},[36,37,390],[15,1870,1871,1874],{"style":1488},[15,1872],{"className":1873,"style":1431},[382],[15,1875],{"className":1876,"style":1496},[1495],[15,1878,1879,1882],{"style":1499},[15,1880],{"className":1881,"style":1431},[382],[15,1883,1885,1918,1921,1924,1927],{"className":1884},[36],[15,1886,1888,1891],{"className":1887},[36],[15,1889,1453],{"className":1890},[36,37],[15,1892,1894],{"className":1893},[361],[15,1895,1897],{"className":1896},[365],[15,1898,1900],{"className":1899},[370],[15,1901,1904],{"className":1902,"style":1903},[374],"height:0.6644em;",[15,1905,1906,1909],{"style":1202},[15,1907],{"className":1908,"style":383},[382],[15,1910,1912],{"className":1911},[387,388,389,390],[15,1913,1915],{"className":1914},[36,390],[15,1916,230],{"className":1917,"style":229},[36,37,390],[15,1919],{"className":1920,"style":81},[42],[15,1922,1482],{"className":1923},[85],[15,1925],{"className":1926,"style":81},[42],[15,1928,1930,1933],{"className":1929},[36],[15,1931,1453],{"className":1932},[36,37],[15,1934,1936],{"className":1935},[361],[15,1937,1939],{"className":1938},[365],[15,1940,1942],{"className":1941},[370],[15,1943,1946],{"className":1944,"style":1945},[374],"height:0.7713em;",[15,1947,1948,1951],{"style":1202},[15,1949],{"className":1950,"style":383},[382],[15,1952,1954],{"className":1953},[387,388,389,390],[15,1955,1957,1960],{"className":1956},[36,390],[15,1958,1482],{"className":1959},[36,390],[15,1961,230],{"className":1962,"style":229},[36,37,390],[15,1964,399],{"className":1965},[398],[15,1967,1969],{"className":1968},[370],[15,1970,1972],{"className":1971,"style":1518},[374],[15,1973],{},[15,1975],{"className":1976},[103,1410],[1386,1978,1979],{},[15,1980,1982],{"className":1981},[18],[15,1983,1985,2004],{"className":1984,"ariaHidden":23},[22],[15,1986,1988,1992,1995,1998,2001],{"className":1987},[27],[15,1989],{"className":1990,"style":1991},[31],"height:0.7278em;vertical-align:-0.0833em;",[15,1993,497],{"className":1994},[36],[15,1996],{"className":1997,"style":81},[42],[15,1999,1482],{"className":2000},[85],[15,2002],{"className":2003,"style":81},[42],[15,2005,2007,2011,2042,2045],{"className":2006},[27],[15,2008],{"className":2009,"style":2010},[31],"height:0.8984em;",[15,2012,2015,2018],{"className":2013},[2014],"mop",[15,2016,1749],{"className":2017},[2014],[15,2019,2021],{"className":2020},[361],[15,2022,2024],{"className":2023},[365],[15,2025,2027],{"className":2026},[370],[15,2028,2030],{"className":2029,"style":2010},[374],[15,2031,2033,2036],{"style":2032},"top:-3.1473em;margin-right:0.05em;",[15,2034],{"className":2035,"style":383},[382],[15,2037,2039],{"className":2038},[387,388,389,390],[15,2040,394],{"className":2041},[36,390],[15,2043],{"className":2044,"style":1598},[42],[15,2046,230],{"className":2047,"style":229},[36,37],[1386,2049,2050],{},[15,2051,2053],{"className":2052},[18],[15,2054,2056],{"className":2055,"ariaHidden":23},[22],[15,2057,2059,2062,2065,2068,2071,2074,2077,2080],{"className":2058},[27],[15,2060],{"className":2061,"style":58},[31],[15,2063,68],{"className":2064},[67],[15,2066,1482],{"className":2067},[36],[15,2069,497],{"className":2070},[36],[15,2072,1594],{"className":2073},[1593],[15,2075],{"className":2076,"style":1598},[42],[15,2078,497],{"className":2079},[36],[15,2081,104],{"className":2082},[103],[1386,2084,1607],{},[1386,2086,2087,2088],{},"zero-centered; ",[15,2089,2091],{"className":2090},[18],[15,2092,2094,2141],{"className":2093,"ariaHidden":23},[22],[15,2095,2097,2100,2132,2135,2138],{"className":2096},[27],[15,2098],{"className":2099,"style":1243},[31],[15,2101,2103,2106],{"className":2102},[36],[15,2104,63],{"className":2105,"style":62},[36,37],[15,2107,2109],{"className":2108},[361],[15,2110,2112],{"className":2111},[365],[15,2113,2115],{"className":2114},[370],[15,2116,2118],{"className":2117,"style":1199},[374],[15,2119,2120,2123],{"style":1202},[15,2121],{"className":2122,"style":383},[382],[15,2124,2126],{"className":2125},[387,388,389,390],[15,2127,2129],{"className":2128},[36,390],[15,2130,1215],{"className":2131},[36,390],[15,2133],{"className":2134,"style":43},[42],[15,2136,1660],{"className":2137},[47],[15,2139],{"className":2140,"style":43},[42],[15,2142,2144,2148],{"className":2143},[27],[15,2145],{"className":2146,"style":2147},[31],"height:0.6444em;",[15,2149,497],{"className":2150},[36],[1284,2152,2153,2156,2196,2244,2277,2280],{},[1386,2154,2155],{},"ReLU",[1386,2157,2158],{},[15,2159,2161],{"className":2160},[18],[15,2162,2164],{"className":2163,"ariaHidden":23},[22],[15,2165,2167,2170,2178,2181,2184,2187,2190,2193],{"className":2166},[27],[15,2168],{"className":2169,"style":58},[31],[15,2171,2173],{"className":2172},[2014],[15,2174,2177],{"className":2175},[36,2176],"mathrm","max",[15,2179,68],{"className":2180},[67],[15,2182,1589],{"className":2183},[36],[15,2185,1594],{"className":2186},[1593],[15,2188],{"className":2189,"style":1598},[42],[15,2191,230],{"className":2192,"style":229},[36,37],[15,2194,104],{"className":2195},[103],[1386,2197,2198],{},[15,2199,2201],{"className":2200},[18],[15,2202,2204,2231],{"className":2203,"ariaHidden":23},[22],[15,2205,2207,2210,2214,2218,2221,2224,2228],{"className":2206},[27],[15,2208],{"className":2209,"style":58},[31],[15,2211,497],{"className":2212},[36,2213],"mathbf",[15,2215,2217],{"className":2216},[67],"[",[15,2219,230],{"className":2220,"style":229},[36,37],[15,2222],{"className":2223,"style":43},[42],[15,2225,2227],{"className":2226},[47],">",[15,2229],{"className":2230,"style":43},[42],[15,2232,2234,2237,2240],{"className":2233},[27],[15,2235],{"className":2236,"style":58},[31],[15,2238,1589],{"className":2239},[36],[15,2241,2243],{"className":2242},[103],"]",[1386,2245,2246],{},[15,2247,2249],{"className":2248},[18],[15,2250,2252],{"className":2251,"ariaHidden":23},[22],[15,2253,2255,2258,2261,2264,2267,2270,2274],{"className":2254},[27],[15,2256],{"className":2257,"style":58},[31],[15,2259,2217],{"className":2260},[67],[15,2262,1589],{"className":2263},[36],[15,2265,1594],{"className":2266},[1593],[15,2268],{"className":2269,"style":1598},[42],[15,2271,2273],{"className":2272},[36],"∞",[15,2275,104],{"className":2276},[103],[1386,2278,2279],{},"left only",[1386,2281,2282],{},"cheap; can die",[1284,2284,2285,2288,2331,2426,2461,2464],{},[1386,2286,2287],{},"Leaky ReLU",[1386,2289,2290],{},[15,2291,2293],{"className":2292},[18],[15,2294,2296],{"className":2295,"ariaHidden":23},[22],[15,2297,2299,2302,2308,2311,2316,2319,2322,2325,2328],{"className":2298},[27],[15,2300],{"className":2301,"style":58},[31],[15,2303,2305],{"className":2304},[2014],[15,2306,2177],{"className":2307},[36,2176],[15,2309,68],{"className":2310},[67],[15,2312,2315],{"className":2313,"style":2314},[36,37],"margin-right:0.0037em;","α",[15,2317,230],{"className":2318,"style":229},[36,37],[15,2320,1594],{"className":2321},[1593],[15,2323],{"className":2324,"style":1598},[42],[15,2326,230],{"className":2327,"style":229},[36,37],[15,2329,104],{"className":2330},[103],[1386,2332,2333],{},[15,2334,2336],{"className":2335},[18],[15,2337,2339,2363,2384,2414],{"className":2338,"ariaHidden":23},[22],[15,2340,2342,2345,2348,2351,2354,2357,2360],{"className":2341},[27],[15,2343],{"className":2344,"style":58},[31],[15,2346,497],{"className":2347},[36,2213],[15,2349,2217],{"className":2350},[67],[15,2352,230],{"className":2353,"style":229},[36,37],[15,2355],{"className":2356,"style":43},[42],[15,2358,2227],{"className":2359},[47],[15,2361],{"className":2362,"style":43},[42],[15,2364,2366,2369,2372,2375,2378,2381],{"className":2365},[27],[15,2367],{"className":2368,"style":58},[31],[15,2370,1589],{"className":2371},[36],[15,2373,2243],{"className":2374},[103],[15,2376],{"className":2377,"style":81},[42],[15,2379,86],{"className":2380},[85],[15,2382],{"className":2383,"style":81},[42],[15,2385,2387,2390,2393,2396,2399,2402,2405,2408,2411],{"className":2386},[27],[15,2388],{"className":2389,"style":58},[31],[15,2391,2315],{"className":2392,"style":2314},[36,37],[15,2394],{"className":2395,"style":1598},[42],[15,2397,497],{"className":2398},[36,2213],[15,2400,2217],{"className":2401},[67],[15,2403,230],{"className":2404,"style":229},[36,37],[15,2406],{"className":2407,"style":43},[42],[15,2409,1660],{"className":2410},[47],[15,2412],{"className":2413,"style":43},[42],[15,2415,2417,2420,2423],{"className":2416},[27],[15,2418],{"className":2419,"style":58},[31],[15,2421,1589],{"className":2422},[36],[15,2424,2243],{"className":2425},[103],[1386,2427,2428],{},[15,2429,2431],{"className":2430},[18],[15,2432,2434],{"className":2433,"ariaHidden":23},[22],[15,2435,2437,2440,2443,2446,2449,2452,2455,2458],{"className":2436},[27],[15,2438],{"className":2439,"style":58},[31],[15,2441,68],{"className":2442},[67],[15,2444,1482],{"className":2445},[36],[15,2447,2273],{"className":2448},[36],[15,2450,1594],{"className":2451},[1593],[15,2453],{"className":2454,"style":1598},[42],[15,2456,2273],{"className":2457},[36],[15,2459,104],{"className":2460},[103],[1386,2462,2463],{},"no",[1386,2465,2466,2509],{},[15,2467,2469],{"className":2468},[18],[15,2470,2472,2499],{"className":2471,"ariaHidden":23},[22],[15,2473,2475,2479,2482,2486,2489,2493,2496],{"className":2474},[27],[15,2476],{"className":2477,"style":2478},[31],"height:0.4831em;",[15,2480,2315],{"className":2481,"style":2314},[36,37],[15,2483],{"className":2484,"style":2485},[42],"margin-right:-0.1667em;",[15,2487],{"className":2488,"style":43},[42],[15,2490,2492],{"className":2491},[47],"≈",[15,2494],{"className":2495,"style":2485},[42],[15,2497],{"className":2498,"style":43},[42],[15,2500,2502,2505],{"className":2501},[27],[15,2503],{"className":2504,"style":2147},[31],[15,2506,2508],{"className":2507},[36],"0.01"," keeps a trickle",[1284,2511,2512,2515,2556,2575,2610,2612],{},[1386,2513,2514],{},"PReLU",[1386,2516,2517],{},[15,2518,2520],{"className":2519},[18],[15,2521,2523],{"className":2522,"ariaHidden":23},[22],[15,2524,2526,2529,2535,2538,2541,2544,2547,2550,2553],{"className":2525},[27],[15,2527],{"className":2528,"style":58},[31],[15,2530,2532],{"className":2531},[2014],[15,2533,2177],{"className":2534},[36,2176],[15,2536,68],{"className":2537},[67],[15,2539,2315],{"className":2540,"style":2314},[36,37],[15,2542,230],{"className":2543,"style":229},[36,37],[15,2545,1594],{"className":2546},[1593],[15,2548],{"className":2549,"style":1598},[42],[15,2551,230],{"className":2552,"style":229},[36,37],[15,2554,104],{"className":2555},[103],[1386,2557,2558,2559,2574],{},"as Leaky, ",[15,2560,2562],{"className":2561},[18],[15,2563,2565],{"className":2564,"ariaHidden":23},[22],[15,2566,2568,2571],{"className":2567},[27],[15,2569],{"className":2570,"style":225},[31],[15,2572,2315],{"className":2573,"style":2314},[36,37]," learned",[1386,2576,2577],{},[15,2578,2580],{"className":2579},[18],[15,2581,2583],{"className":2582,"ariaHidden":23},[22],[15,2584,2586,2589,2592,2595,2598,2601,2604,2607],{"className":2585},[27],[15,2587],{"className":2588,"style":58},[31],[15,2590,68],{"className":2591},[67],[15,2593,1482],{"className":2594},[36],[15,2596,2273],{"className":2597},[36],[15,2599,1594],{"className":2600},[1593],[15,2602],{"className":2603,"style":1598},[42],[15,2605,2273],{"className":2606},[36],[15,2608,104],{"className":2609},[103],[1386,2611,2463],{},[1386,2613,2614,2629],{},[15,2615,2617],{"className":2616},[18],[15,2618,2620],{"className":2619,"ariaHidden":23},[22],[15,2621,2623,2626],{"className":2622},[27],[15,2624],{"className":2625,"style":225},[31],[15,2627,2315],{"className":2628,"style":2314},[36,37]," a trained parameter",[1284,2631,2632,2635,2759,2856,2891,2894],{},[1386,2633,2634],{},"ELU",[1386,2636,2637,2652,2653,2687,2688],{},[15,2638,2640],{"className":2639},[18],[15,2641,2643],{"className":2642,"ariaHidden":23},[22],[15,2644,2646,2649],{"className":2645},[27],[15,2647],{"className":2648,"style":225},[31],[15,2650,230],{"className":2651,"style":229},[36,37]," if ",[15,2654,2656],{"className":2655},[18],[15,2657,2659,2678],{"className":2658,"ariaHidden":23},[22],[15,2660,2662,2666,2669,2672,2675],{"className":2661},[27],[15,2663],{"className":2664,"style":2665},[31],"height:0.5782em;vertical-align:-0.0391em;",[15,2667,230],{"className":2668,"style":229},[36,37],[15,2670],{"className":2671,"style":43},[42],[15,2673,2227],{"className":2674},[47],[15,2676],{"className":2677,"style":43},[42],[15,2679,2681,2684],{"className":2680},[27],[15,2682],{"className":2683,"style":2147},[31],[15,2685,1589],{"className":2686},[36]," else ",[15,2689,2691],{"className":2690},[18],[15,2692,2694,2747],{"className":2693,"ariaHidden":23},[22],[15,2695,2697,2700,2703,2706,2738,2741,2744],{"className":2696},[27],[15,2698],{"className":2699,"style":58},[31],[15,2701,2315],{"className":2702,"style":2314},[36,37],[15,2704,68],{"className":2705},[67],[15,2707,2709,2712],{"className":2708},[36],[15,2710,1453],{"className":2711},[36,37],[15,2713,2715],{"className":2714},[361],[15,2716,2718],{"className":2717},[365],[15,2719,2721],{"className":2720},[370],[15,2722,2724],{"className":2723,"style":1903},[374],[15,2725,2726,2729],{"style":1202},[15,2727],{"className":2728,"style":383},[382],[15,2730,2732],{"className":2731},[387,388,389,390],[15,2733,2735],{"className":2734},[36,390],[15,2736,230],{"className":2737,"style":229},[36,37,390],[15,2739],{"className":2740,"style":81},[42],[15,2742,1482],{"className":2743},[85],[15,2745],{"className":2746,"style":81},[42],[15,2748,2750,2753,2756],{"className":2749},[27],[15,2751],{"className":2752,"style":58},[31],[15,2754,497],{"className":2755},[36],[15,2757,104],{"className":2758},[103],[1386,2760,2761,2652,2776,2687,2809],{},[15,2762,2764],{"className":2763},[18],[15,2765,2767],{"className":2766,"ariaHidden":23},[22],[15,2768,2770,2773],{"className":2769},[27],[15,2771],{"className":2772,"style":2147},[31],[15,2774,497],{"className":2775},[36],[15,2777,2779],{"className":2778},[18],[15,2780,2782,2800],{"className":2781,"ariaHidden":23},[22],[15,2783,2785,2788,2791,2794,2797],{"className":2784},[27],[15,2786],{"className":2787,"style":2665},[31],[15,2789,230],{"className":2790,"style":229},[36,37],[15,2792],{"className":2793,"style":43},[42],[15,2795,2227],{"className":2796},[47],[15,2798],{"className":2799,"style":43},[42],[15,2801,2803,2806],{"className":2802},[27],[15,2804],{"className":2805,"style":2147},[31],[15,2807,1589],{"className":2808},[36],[15,2810,2812],{"className":2811},[18],[15,2813,2815],{"className":2814,"ariaHidden":23},[22],[15,2816,2818,2821,2824],{"className":2817},[27],[15,2819],{"className":2820,"style":1903},[31],[15,2822,2315],{"className":2823,"style":2314},[36,37],[15,2825,2827,2830],{"className":2826},[36],[15,2828,1453],{"className":2829},[36,37],[15,2831,2833],{"className":2832},[361],[15,2834,2836],{"className":2835},[365],[15,2837,2839],{"className":2838},[370],[15,2840,2842],{"className":2841,"style":1903},[374],[15,2843,2844,2847],{"style":1202},[15,2845],{"className":2846,"style":383},[382],[15,2848,2850],{"className":2849},[387,388,389,390],[15,2851,2853],{"className":2852},[36,390],[15,2854,230],{"className":2855,"style":229},[36,37,390],[1386,2857,2858],{},[15,2859,2861],{"className":2860},[18],[15,2862,2864],{"className":2863,"ariaHidden":23},[22],[15,2865,2867,2870,2873,2876,2879,2882,2885,2888],{"className":2866},[27],[15,2868],{"className":2869,"style":58},[31],[15,2871,68],{"className":2872},[67],[15,2874,1482],{"className":2875},[36],[15,2877,2315],{"className":2878,"style":2314},[36,37],[15,2880,1594],{"className":2881},[1593],[15,2883],{"className":2884,"style":1598},[42],[15,2886,2273],{"className":2887},[36],[15,2889,104],{"className":2890},[103],[1386,2892,2893],{},"left soft",[1386,2895,2896,2897],{},"smooth; pushes mean toward ",[15,2898,2900],{"className":2899},[18],[15,2901,2903],{"className":2902,"ariaHidden":23},[22],[15,2904,2906,2909],{"className":2905},[27],[15,2907],{"className":2908,"style":2147},[31],[15,2910,1589],{"className":2911},[36],[1284,2913,2914,2917,2950,3010,3058,3060],{},[1386,2915,2916],{},"GELU",[1386,2918,2919],{},[15,2920,2922],{"className":2921},[18],[15,2923,2925],{"className":2924,"ariaHidden":23},[22],[15,2926,2928,2931,2934,2937,2941,2944,2947],{"className":2927},[27],[15,2929],{"className":2930,"style":58},[31],[15,2932,230],{"className":2933,"style":229},[36,37],[15,2935],{"className":2936,"style":1598},[42],[15,2938,2940],{"className":2939},[36],"Φ",[15,2942,68],{"className":2943},[67],[15,2945,230],{"className":2946,"style":229},[36,37],[15,2948,104],{"className":2949},[103],[1386,2951,2952],{},[15,2953,2955],{"className":2954},[18],[15,2956,2958,2985],{"className":2957,"ariaHidden":23},[22],[15,2959,2961,2964,2967,2970,2973,2976,2979,2982],{"className":2960},[27],[15,2962],{"className":2963,"style":58},[31],[15,2965,2940],{"className":2966},[36],[15,2968,68],{"className":2969},[67],[15,2971,230],{"className":2972,"style":229},[36,37],[15,2974,104],{"className":2975},[103],[15,2977],{"className":2978,"style":81},[42],[15,2980,86],{"className":2981},[85],[15,2983],{"className":2984,"style":81},[42],[15,2986,2988,2991,2994,2997,3001,3004,3007],{"className":2987},[27],[15,2989],{"className":2990,"style":58},[31],[15,2992,230],{"className":2993,"style":229},[36,37],[15,2995],{"className":2996,"style":1598},[42],[15,2998,3000],{"className":2999},[36,37],"ϕ",[15,3002,68],{"className":3003},[67],[15,3005,230],{"className":3006,"style":229},[36,37],[15,3008,104],{"className":3009},[103],[1386,3011,3012],{},[15,3013,3015],{"className":3014},[18],[15,3016,3018,3030],{"className":3017,"ariaHidden":23},[22],[15,3019,3021,3024,3027],{"className":3020},[27],[15,3022],{"className":3023,"style":2478},[31],[15,3025,2492],{"className":3026},[47],[15,3028],{"className":3029,"style":43},[42],[15,3031,3033,3036,3039,3042,3046,3049,3052,3055],{"className":3032},[27],[15,3034],{"className":3035,"style":58},[31],[15,3037,2217],{"className":3038},[67],[15,3040,1482],{"className":3041},[36],[15,3043,3045],{"className":3044},[36],"0.17",[15,3047,1594],{"className":3048},[1593],[15,3050],{"className":3051,"style":1598},[42],[15,3053,2273],{"className":3054},[36],[15,3056,104],{"className":3057},[103],[1386,3059,2893],{},[1386,3061,3062,3078],{},[15,3063,3065],{"className":3064},[18],[15,3066,3068],{"className":3067,"ariaHidden":23},[22],[15,3069,3071,3075],{"className":3070},[27],[15,3072],{"className":3073,"style":3074},[31],"height:0.6833em;",[15,3076,2940],{"className":3077},[36]," standard-normal CDF",[1284,3080,3081,3084,3162,3188,3220,3222],{},[1386,3082,3083],{},"Softplus",[1386,3085,3086],{},[15,3087,3089],{"className":3088},[18],[15,3090,3092,3121],{"className":3091,"ariaHidden":23},[22],[15,3093,3095,3098,3106,3109,3112,3115,3118],{"className":3094},[27],[15,3096],{"className":3097,"style":58},[31],[15,3099,3101],{"className":3100},[2014],[15,3102,3105],{"className":3103,"style":3104},[36,2176],"margin-right:0.0139em;","log",[15,3107,68],{"className":3108},[67],[15,3110,497],{"className":3111},[36],[15,3113],{"className":3114,"style":81},[42],[15,3116,86],{"className":3117},[85],[15,3119],{"className":3120,"style":81},[42],[15,3122,3124,3127,3159],{"className":3123},[27],[15,3125],{"className":3126,"style":58},[31],[15,3128,3130,3133],{"className":3129},[36],[15,3131,1453],{"className":3132},[36,37],[15,3134,3136],{"className":3135},[361],[15,3137,3139],{"className":3138},[365],[15,3140,3142],{"className":3141},[370],[15,3143,3145],{"className":3144,"style":1903},[374],[15,3146,3147,3150],{"style":1202},[15,3148],{"className":3149,"style":383},[382],[15,3151,3153],{"className":3152},[387,388,389,390],[15,3154,3156],{"className":3155},[36,390],[15,3157,230],{"className":3158,"style":229},[36,37,390],[15,3160,104],{"className":3161},[103],[1386,3163,3164],{},[15,3165,3167],{"className":3166},[18],[15,3168,3170],{"className":3169,"ariaHidden":23},[22],[15,3171,3173,3176,3179,3182,3185],{"className":3172},[27],[15,3174],{"className":3175,"style":58},[31],[15,3177,1541],{"className":3178,"style":62},[36,37],[15,3180,68],{"className":3181},[67],[15,3183,230],{"className":3184,"style":229},[36,37],[15,3186,104],{"className":3187},[103],[1386,3189,3190],{},[15,3191,3193],{"className":3192},[18],[15,3194,3196],{"className":3195,"ariaHidden":23},[22],[15,3197,3199,3202,3205,3208,3211,3214,3217],{"className":3198},[27],[15,3200],{"className":3201,"style":58},[31],[15,3203,68],{"className":3204},[67],[15,3206,1589],{"className":3207},[36],[15,3209,1594],{"className":3210},[1593],[15,3212],{"className":3213,"style":1598},[42],[15,3215,2273],{"className":3216},[36],[15,3218,104],{"className":3219},[103],[1386,3221,2463],{},[1386,3223,3224,3225],{},"smooth ReLU; ",[15,3226,3228],{"className":3227},[18],[15,3229,3231,3278],{"className":3230,"ariaHidden":23},[22],[15,3232,3234,3237,3269,3272,3275],{"className":3233},[27],[15,3235],{"className":3236,"style":1243},[31],[15,3238,3240,3243],{"className":3239},[36],[15,3241,63],{"className":3242,"style":62},[36,37],[15,3244,3246],{"className":3245},[361],[15,3247,3249],{"className":3248},[365],[15,3250,3252],{"className":3251},[370],[15,3253,3255],{"className":3254,"style":1199},[374],[15,3256,3257,3260],{"style":1202},[15,3258],{"className":3259,"style":383},[382],[15,3261,3263],{"className":3262},[387,388,389,390],[15,3264,3266],{"className":3265},[36,390],[15,3267,1215],{"className":3268},[36,390],[15,3270],{"className":3271,"style":43},[42],[15,3273,48],{"className":3274},[47],[15,3276],{"className":3277,"style":43},[42],[15,3279,3281,3284],{"className":3280},[27],[15,3282],{"className":3283,"style":225},[31],[15,3285,1541],{"className":3286,"style":62},[36,37],[1284,3288,3289,3292,3324,3417,3465,3467],{},[1386,3290,3291],{},"Swish \u002F SiLU",[1386,3293,3294],{},[15,3295,3297],{"className":3296},[18],[15,3298,3300],{"className":3299,"ariaHidden":23},[22],[15,3301,3303,3306,3309,3312,3315,3318,3321],{"className":3302},[27],[15,3304],{"className":3305,"style":58},[31],[15,3307,230],{"className":3308,"style":229},[36,37],[15,3310],{"className":3311,"style":1598},[42],[15,3313,1541],{"className":3314,"style":62},[36,37],[15,3316,68],{"className":3317},[67],[15,3319,230],{"className":3320,"style":229},[36,37],[15,3322,104],{"className":3323},[103],[1386,3325,3326],{},[15,3327,3329],{"className":3328},[18],[15,3330,3332,3359,3398],{"className":3331,"ariaHidden":23},[22],[15,3333,3335,3338,3341,3344,3347,3350,3353,3356],{"className":3334},[27],[15,3336],{"className":3337,"style":58},[31],[15,3339,1541],{"className":3340,"style":62},[36,37],[15,3342,68],{"className":3343},[67],[15,3345,230],{"className":3346,"style":229},[36,37],[15,3348,104],{"className":3349},[103],[15,3351],{"className":3352,"style":81},[42],[15,3354,86],{"className":3355},[85],[15,3357],{"className":3358,"style":81},[42],[15,3360,3362,3365,3368,3371,3374,3377,3380,3383,3386,3389,3392,3395],{"className":3361},[27],[15,3363],{"className":3364,"style":58},[31],[15,3366,230],{"className":3367,"style":229},[36,37],[15,3369],{"className":3370,"style":1598},[42],[15,3372,1541],{"className":3373,"style":62},[36,37],[15,3375,68],{"className":3376},[67],[15,3378,230],{"className":3379,"style":229},[36,37],[15,3381,104],{"className":3382},[103],[15,3384,68],{"className":3385},[67],[15,3387,497],{"className":3388},[36],[15,3390],{"className":3391,"style":81},[42],[15,3393,1482],{"className":3394},[85],[15,3396],{"className":3397,"style":81},[42],[15,3399,3401,3404,3407,3410,3413],{"className":3400},[27],[15,3402],{"className":3403,"style":58},[31],[15,3405,1541],{"className":3406,"style":62},[36,37],[15,3408,68],{"className":3409},[67],[15,3411,230],{"className":3412,"style":229},[36,37],[15,3414,3416],{"className":3415},[103],"))",[1386,3418,3419],{},[15,3420,3422],{"className":3421},[18],[15,3423,3425,3437],{"className":3424,"ariaHidden":23},[22],[15,3426,3428,3431,3434],{"className":3427},[27],[15,3429],{"className":3430,"style":2478},[31],[15,3432,2492],{"className":3433},[47],[15,3435],{"className":3436,"style":43},[42],[15,3438,3440,3443,3446,3449,3453,3456,3459,3462],{"className":3439},[27],[15,3441],{"className":3442,"style":58},[31],[15,3444,2217],{"className":3445},[67],[15,3447,1482],{"className":3448},[36],[15,3450,3452],{"className":3451},[36],"0.28",[15,3454,1594],{"className":3455},[1593],[15,3457],{"className":3458,"style":1598},[42],[15,3460,2273],{"className":3461},[36],[15,3463,104],{"className":3464},[103],[1386,3466,2893],{},[1386,3468,3469],{},"self-gated; non-monotone",[1284,3471,3472,3475,3660,3752,3769,3771],{},[1386,3473,3474],{},"Maxout",[1386,3476,3477],{},[15,3478,3480],{"className":3479},[18],[15,3481,3483,3610],{"className":3482,"ariaHidden":23},[22],[15,3484,3486,3490,3538,3541,3598,3601,3604,3607],{"className":3485},[27],[15,3487],{"className":3488,"style":3489},[31],"height:1.2439em;vertical-align:-0.3948em;",[15,3491,3493,3499],{"className":3492},[2014],[15,3494,3496],{"className":3495},[2014],[15,3497,2177],{"className":3498},[36,2176],[15,3500,3502],{"className":3501},[361],[15,3503,3505,3529],{"className":3504},[365,366],[15,3506,3508,3526],{"className":3507},[370],[15,3509,3512],{"className":3510,"style":3511},[374],"height:0.3117em;",[15,3513,3515,3518],{"style":3514},"top:-2.55em;margin-right:0.05em;",[15,3516],{"className":3517,"style":383},[382],[15,3519,3521],{"className":3520},[387,388,389,390],[15,3522,3525],{"className":3523,"style":3524},[36,37,390],"margin-right:0.0572em;","j",[15,3527,399],{"className":3528},[398],[15,3530,3532],{"className":3531},[370],[15,3533,3536],{"className":3534,"style":3535},[374],"height:0.2861em;",[15,3537],{},[15,3539,68],{"className":3540},[67],[15,3542,3544,3549],{"className":3543},[36],[15,3545,3548],{"className":3546,"style":3547},[36,37],"margin-right:0.0269em;","w",[15,3550,3552],{"className":3551},[361],[15,3553,3555,3589],{"className":3554},[365,366],[15,3556,3558,3586],{"className":3557},[370],[15,3559,3562,3574],{"className":3560,"style":3561},[374],"height:0.8491em;",[15,3563,3565,3568],{"style":3564},"top:-2.4413em;margin-left:-0.0269em;margin-right:0.05em;",[15,3566],{"className":3567,"style":383},[382],[15,3569,3571],{"className":3570},[387,388,389,390],[15,3572,3525],{"className":3573,"style":3524},[36,37,390],[15,3575,3576,3579],{"style":1202},[15,3577],{"className":3578,"style":383},[382],[15,3580,3582],{"className":3581},[387,388,389,390],[15,3583,3585],{"className":3584},[36,390],"⊤",[15,3587,399],{"className":3588},[398],[15,3590,3592],{"className":3591},[370],[15,3593,3596],{"className":3594,"style":3595},[374],"height:0.3948em;",[15,3597],{},[15,3599,77],{"className":3600},[36,37],[15,3602],{"className":3603,"style":81},[42],[15,3605,86],{"className":3606},[85],[15,3608],{"className":3609,"style":81},[42],[15,3611,3613,3617,3657],{"className":3612},[27],[15,3614],{"className":3615,"style":3616},[31],"height:1.0361em;vertical-align:-0.2861em;",[15,3618,3620,3623],{"className":3619},[36],[15,3621,99],{"className":3622},[36,37],[15,3624,3626],{"className":3625},[361],[15,3627,3629,3649],{"className":3628},[365,366],[15,3630,3632,3646],{"className":3631},[370],[15,3633,3635],{"className":3634,"style":3511},[374],[15,3636,3637,3640],{"style":378},[15,3638],{"className":3639,"style":383},[382],[15,3641,3643],{"className":3642},[387,388,389,390],[15,3644,3525],{"className":3645,"style":3524},[36,37,390],[15,3647,399],{"className":3648},[398],[15,3650,3652],{"className":3651},[370],[15,3653,3655],{"className":3654,"style":3535},[374],[15,3656],{},[15,3658,104],{"className":3659},[103],[1386,3661,3662,3751],{},[15,3663,3665],{"className":3664},[18],[15,3666,3668],{"className":3667,"ariaHidden":23},[22],[15,3669,3671,3675],{"className":3670},[27],[15,3672],{"className":3673,"style":3674},[31],"height:0.7167em;vertical-align:-0.2861em;",[15,3676,3678,3681],{"className":3677},[36],[15,3679,3548],{"className":3680,"style":3547},[36,37],[15,3682,3684],{"className":3683},[361],[15,3685,3687,3743],{"className":3686},[365,366],[15,3688,3690,3740],{"className":3689},[370],[15,3691,3693],{"className":3692,"style":3511},[374],[15,3694,3696,3699],{"style":3695},"top:-2.55em;margin-left:-0.0269em;margin-right:0.05em;",[15,3697],{"className":3698,"style":383},[382],[15,3700,3702],{"className":3701},[387,388,389,390],[15,3703,3705],{"className":3704},[36,390],[15,3706,3708,3711],{"className":3707},[36,390],[15,3709,3525],{"className":3710,"style":3524},[36,37,390],[15,3712,3714],{"className":3713},[361],[15,3715,3717],{"className":3716},[365],[15,3718,3720],{"className":3719},[370],[15,3721,3724],{"className":3722,"style":3723},[374],"height:0.6183em;",[15,3725,3727,3731],{"style":3726},"top:-2.786em;margin-right:0.0714em;",[15,3728],{"className":3729,"style":3730},[382],"height:2.5em;",[15,3732,3736],{"className":3733},[387,3734,3735,390],"reset-size3","size1",[15,3737,3739],{"className":3738},[85,390],"⋆",[15,3741,399],{"className":3742},[398],[15,3744,3746],{"className":3745},[370],[15,3747,3749],{"className":3748,"style":3535},[374],[15,3750],{}," of the winning piece",[1386,3753,3754],{},[15,3755,3757],{"className":3756},[18],[15,3758,3760],{"className":3759,"ariaHidden":23},[22],[15,3761,3763,3766],{"className":3762},[27],[15,3764],{"className":3765,"style":187},[31],[15,3767,192],{"className":3768},[36,191],[1386,3770,2463],{},[1386,3772,3773,3774,3796],{},"learns its own shape; ",[15,3775,3777],{"className":3776},[18],[15,3778,3780],{"className":3779,"ariaHidden":23},[22],[15,3781,3783,3787,3792],{"className":3782},[27],[15,3784],{"className":3785,"style":3786},[31],"height:0.7778em;vertical-align:-0.0833em;",[15,3788,3791],{"className":3789,"style":3790},[36,37],"margin-right:0.0315em;","k",[15,3793,3795],{"className":3794},[36],"×"," params",[11,3798,3799,3800,3815,3816,3831,3832,3848,3849,3864,3865,3868,3869,3872,3873],{},"Here ",[15,3801,3803],{"className":3802},[18],[15,3804,3806],{"className":3805,"ariaHidden":23},[22],[15,3807,3809,3812],{"className":3808},[27],[15,3810],{"className":3811,"style":225},[31],[15,3813,1541],{"className":3814,"style":62},[36,37]," is the logistic sigmoid, ",[15,3817,3819],{"className":3818},[18],[15,3820,3822],{"className":3821,"ariaHidden":23},[22],[15,3823,3825,3828],{"className":3824},[27],[15,3826],{"className":3827,"style":3074},[31],[15,3829,2940],{"className":3830},[36]," and ",[15,3833,3835],{"className":3834},[18],[15,3836,3838],{"className":3837,"ariaHidden":23},[22],[15,3839,3841,3845],{"className":3840},[27],[15,3842],{"className":3843,"style":3844},[31],"height:0.8889em;vertical-align:-0.1944em;",[15,3846,3000],{"className":3847},[36,37]," are the standard-normal\nCDF and PDF, and ",[15,3850,3852],{"className":3851},[18],[15,3853,3855],{"className":3854,"ariaHidden":23},[22],[15,3856,3858,3861],{"className":3857},[27],[15,3859],{"className":3860,"style":225},[31],[15,3862,2315],{"className":3863,"style":2314},[36,37]," is a small positive constant (or a learned parameter for\nPReLU). Two families fall out of the table: the ",[152,3866,3867],{},"saturating"," units (sigmoid,\ntanh) whose derivative vanishes at both extremes, and the ",[152,3870,3871],{},"non-saturating","\nrectified units (ReLU onward) whose derivative stays bounded away from zero on at\nleast the positive half-line.",[3874,3875,3876],"sup",{},[107,3877,497],{"href":3878,"ariaDescribedBy":3879,"dataFootnoteRef":6,"id":3881},"#user-content-fn-gf-hidden",[3880],"footnote-label","user-content-fnref-gf-hidden",[1161,3883,3885],{"id":3884},"every-derivative-derived","Every derivative, derived",[11,3887,3888],{},"Backpropagation never needs the activation itself in the backward pass, only its\nderivative. Each derivative below takes one or two lines to derive.",[3890,3891,3892],"h3",{"id":1388},"Sigmoid",[11,3894,3895,3896,4028],{},"Write ",[15,3897,3899],{"className":3898},[18],[15,3900,3902,3929,3950],{"className":3901,"ariaHidden":23},[22],[15,3903,3905,3908,3911,3914,3917,3920,3923,3926],{"className":3904},[27],[15,3906],{"className":3907,"style":58},[31],[15,3909,1541],{"className":3910,"style":62},[36,37],[15,3912,68],{"className":3913},[67],[15,3915,230],{"className":3916,"style":229},[36,37],[15,3918,104],{"className":3919},[103],[15,3921],{"className":3922,"style":43},[42],[15,3924,48],{"className":3925},[47],[15,3927],{"className":3928,"style":43},[42],[15,3930,3932,3935,3938,3941,3944,3947],{"className":3931},[27],[15,3933],{"className":3934,"style":58},[31],[15,3936,68],{"className":3937},[67],[15,3939,497],{"className":3940},[36],[15,3942],{"className":3943,"style":81},[42],[15,3945,86],{"className":3946},[85],[15,3948],{"className":3949,"style":81},[42],[15,3951,3953,3957,3992],{"className":3952},[27],[15,3954],{"className":3955,"style":3956},[31],"height:1.0641em;vertical-align:-0.25em;",[15,3958,3960,3963],{"className":3959},[36],[15,3961,1453],{"className":3962},[36,37],[15,3964,3966],{"className":3965},[361],[15,3967,3969],{"className":3968},[365],[15,3970,3972],{"className":3971},[370],[15,3973,3975],{"className":3974,"style":1945},[374],[15,3976,3977,3980],{"style":1202},[15,3978],{"className":3979,"style":383},[382],[15,3981,3983],{"className":3982},[387,388,389,390],[15,3984,3986,3989],{"className":3985},[36,390],[15,3987,1482],{"className":3988},[36,390],[15,3990,230],{"className":3991,"style":229},[36,37,390],[15,3993,3995,3998],{"className":3994},[103],[15,3996,104],{"className":3997},[103],[15,3999,4001],{"className":4000},[361],[15,4002,4004],{"className":4003},[365],[15,4005,4007],{"className":4006},[370],[15,4008,4011],{"className":4009,"style":4010},[374],"height:0.8141em;",[15,4012,4013,4016],{"style":1202},[15,4014],{"className":4015,"style":383},[382],[15,4017,4019],{"className":4018},[387,388,389,390],[15,4020,4022,4025],{"className":4021},[36,390],[15,4023,1482],{"className":4024},[36,390],[15,4026,497],{"className":4027},[36,390]," and differentiate by the chain rule:",[15,4030,4033],{"className":4031},[4032],"katex-display",[15,4034,4036],{"className":4035},[18],[15,4037,4039,4098,4125,4214,4274,4462,4583],{"className":4038,"ariaHidden":23},[22],[15,4040,4042,4046,4080,4083,4086,4089,4092,4095],{"className":4041},[27],[15,4043],{"className":4044,"style":4045},[31],"height:1.0519em;vertical-align:-0.25em;",[15,4047,4049,4052],{"className":4048},[36],[15,4050,1541],{"className":4051,"style":62},[36,37],[15,4053,4055],{"className":4054},[361],[15,4056,4058],{"className":4057},[365],[15,4059,4061],{"className":4060},[370],[15,4062,4065],{"className":4063,"style":4064},[374],"height:0.8019em;",[15,4066,4068,4071],{"style":4067},"top:-3.113em;margin-right:0.05em;",[15,4069],{"className":4070,"style":383},[382],[15,4072,4074],{"className":4073},[387,388,389,390],[15,4075,4077],{"className":4076},[36,390],[15,4078,1215],{"className":4079},[36,390],[15,4081,68],{"className":4082},[67],[15,4084,230],{"className":4085,"style":229},[36,37],[15,4087,104],{"className":4088},[103],[15,4090],{"className":4091,"style":43},[42],[15,4093,48],{"className":4094},[47],[15,4096],{"className":4097,"style":43},[42],[15,4099,4101,4104,4107,4110,4113,4116,4119,4122],{"className":4100},[27],[15,4102],{"className":4103,"style":58},[31],[15,4105,1482],{"className":4106},[36],[15,4108],{"className":4109,"style":1598},[42],[15,4111,68],{"className":4112},[67],[15,4114,497],{"className":4115},[36],[15,4117],{"className":4118,"style":81},[42],[15,4120,86],{"className":4121},[85],[15,4123],{"className":4124,"style":81},[42],[15,4126,4128,4132,4168,4204,4207,4211],{"className":4127},[27],[15,4129],{"className":4130,"style":4131},[31],"height:1.1141em;vertical-align:-0.25em;",[15,4133,4135,4138],{"className":4134},[36],[15,4136,1453],{"className":4137},[36,37],[15,4139,4141],{"className":4140},[361],[15,4142,4144],{"className":4143},[365],[15,4145,4147],{"className":4146},[370],[15,4148,4151],{"className":4149,"style":4150},[374],"height:0.8213em;",[15,4152,4153,4156],{"style":4067},[15,4154],{"className":4155,"style":383},[382],[15,4157,4159],{"className":4158},[387,388,389,390],[15,4160,4162,4165],{"className":4161},[36,390],[15,4163,1482],{"className":4164},[36,390],[15,4166,230],{"className":4167,"style":229},[36,37,390],[15,4169,4171,4174],{"className":4170},[103],[15,4172,104],{"className":4173},[103],[15,4175,4177],{"className":4176},[361],[15,4178,4180],{"className":4179},[365],[15,4181,4183],{"className":4182},[370],[15,4184,4187],{"className":4185,"style":4186},[374],"height:0.8641em;",[15,4188,4189,4192],{"style":4067},[15,4190],{"className":4191,"style":383},[382],[15,4193,4195],{"className":4194},[387,388,389,390],[15,4196,4198,4201],{"className":4197},[36,390],[15,4199,1482],{"className":4200},[36,390],[15,4202,394],{"className":4203},[36,390],[15,4205],{"className":4206,"style":81},[42],[15,4208,4210],{"className":4209},[85],"⋅",[15,4212],{"className":4213,"style":81},[42],[15,4215,4217,4221,4224,4227,4262,4265,4268,4271],{"className":4216},[27],[15,4218],{"className":4219,"style":4220},[31],"height:1.0713em;vertical-align:-0.25em;",[15,4222,68],{"className":4223},[67],[15,4225,1482],{"className":4226},[36],[15,4228,4230,4233],{"className":4229},[36],[15,4231,1453],{"className":4232},[36,37],[15,4234,4236],{"className":4235},[361],[15,4237,4239],{"className":4238},[365],[15,4240,4242],{"className":4241},[370],[15,4243,4245],{"className":4244,"style":4150},[374],[15,4246,4247,4250],{"style":4067},[15,4248],{"className":4249,"style":383},[382],[15,4251,4253],{"className":4252},[387,388,389,390],[15,4254,4256,4259],{"className":4255},[36,390],[15,4257,1482],{"className":4258},[36,390],[15,4260,230],{"className":4261,"style":229},[36,37,390],[15,4263,104],{"className":4264},[103],[15,4266],{"className":4267,"style":43},[42],[15,4269,48],{"className":4270},[47],[15,4272],{"className":4273,"style":43},[42],[15,4275,4277,4281,4453,4456,4459],{"className":4276},[27],[15,4278],{"className":4279,"style":4280},[31],"height:2.3843em;vertical-align:-0.936em;",[15,4282,4284,4287,4450],{"className":4283},[36],[15,4285],{"className":4286},[67,1410],[15,4288,4290],{"className":4289},[1414],[15,4291,4293,4441],{"className":4292},[365,366],[15,4294,4296,4438],{"className":4295},[370],[15,4297,4299,4387,4395],{"className":4298,"style":1783},[374],[15,4300,4301,4304],{"style":1427},[15,4302],{"className":4303,"style":1431},[382],[15,4305,4307,4310,4313,4316,4319,4322,4357],{"className":4306},[36],[15,4308,68],{"className":4309},[67],[15,4311,497],{"className":4312},[36],[15,4314],{"className":4315,"style":81},[42],[15,4317,86],{"className":4318},[85],[15,4320],{"className":4321,"style":81},[42],[15,4323,4325,4328],{"className":4324},[36],[15,4326,1453],{"className":4327},[36,37],[15,4329,4331],{"className":4330},[361],[15,4332,4334],{"className":4333},[365],[15,4335,4337],{"className":4336},[370],[15,4338,4340],{"className":4339,"style":1466},[374],[15,4341,4342,4345],{"style":1469},[15,4343],{"className":4344,"style":383},[382],[15,4346,4348],{"className":4347},[387,388,389,390],[15,4349,4351,4354],{"className":4350},[36,390],[15,4352,1482],{"className":4353},[36,390],[15,4355,230],{"className":4356,"style":229},[36,37,390],[15,4358,4360,4363],{"className":4359},[103],[15,4361,104],{"className":4362},[103],[15,4364,4366],{"className":4365},[361],[15,4367,4369],{"className":4368},[365],[15,4370,4372],{"className":4371},[370],[15,4373,4376],{"className":4374,"style":4375},[374],"height:0.7401em;",[15,4377,4378,4381],{"style":1469},[15,4379],{"className":4380,"style":383},[382],[15,4382,4384],{"className":4383},[387,388,389,390],[15,4385,394],{"className":4386},[36,390],[15,4388,4389,4392],{"style":1488},[15,4390],{"className":4391,"style":1431},[382],[15,4393],{"className":4394,"style":1496},[1495],[15,4396,4397,4400],{"style":1499},[15,4398],{"className":4399,"style":1431},[382],[15,4401,4403],{"className":4402},[36],[15,4404,4406,4409],{"className":4405},[36],[15,4407,1453],{"className":4408},[36,37],[15,4410,4412],{"className":4411},[361],[15,4413,4415],{"className":4414},[365],[15,4416,4418],{"className":4417},[370],[15,4419,4421],{"className":4420,"style":1945},[374],[15,4422,4423,4426],{"style":1202},[15,4424],{"className":4425,"style":383},[382],[15,4427,4429],{"className":4428},[387,388,389,390],[15,4430,4432,4435],{"className":4431},[36,390],[15,4433,1482],{"className":4434},[36,390],[15,4436,230],{"className":4437,"style":229},[36,37,390],[15,4439,399],{"className":4440},[398],[15,4442,4444],{"className":4443},[370],[15,4445,4448],{"className":4446,"style":4447},[374],"height:0.936em;",[15,4449],{},[15,4451],{"className":4452},[103,1410],[15,4454],{"className":4455,"style":43},[42],[15,4457,48],{"className":4458},[47],[15,4460],{"className":4461,"style":43},[42],[15,4463,4465,4468,4574,4577,4580],{"className":4464},[27],[15,4466],{"className":4467,"style":1403},[31],[15,4469,4471,4474,4571],{"className":4470},[36],[15,4472],{"className":4473},[67,1410],[15,4475,4477],{"className":4476},[1414],[15,4478,4480,4563],{"className":4479},[365,366],[15,4481,4483,4560],{"className":4482},[370],[15,4484,4486,4541,4549],{"className":4485,"style":1424},[374],[15,4487,4488,4491],{"style":1427},[15,4489],{"className":4490,"style":1431},[382],[15,4492,4494,4497,4500,4503,4506],{"className":4493},[36],[15,4495,497],{"className":4496},[36],[15,4498],{"className":4499,"style":81},[42],[15,4501,86],{"className":4502},[85],[15,4504],{"className":4505,"style":81},[42],[15,4507,4509,4512],{"className":4508},[36],[15,4510,1453],{"className":4511},[36,37],[15,4513,4515],{"className":4514},[361],[15,4516,4518],{"className":4517},[365],[15,4519,4521],{"className":4520},[370],[15,4522,4524],{"className":4523,"style":1466},[374],[15,4525,4526,4529],{"style":1469},[15,4527],{"className":4528,"style":383},[382],[15,4530,4532],{"className":4531},[387,388,389,390],[15,4533,4535,4538],{"className":4534},[36,390],[15,4536,1482],{"className":4537},[36,390],[15,4539,230],{"className":4540,"style":229},[36,37,390],[15,4542,4543,4546],{"style":1488},[15,4544],{"className":4545,"style":1431},[382],[15,4547],{"className":4548,"style":1496},[1495],[15,4550,4551,4554],{"style":1499},[15,4552],{"className":4553,"style":1431},[382],[15,4555,4557],{"className":4556},[36],[15,4558,497],{"className":4559},[36],[15,4561,399],{"className":4562},[398],[15,4564,4566],{"className":4565},[370],[15,4567,4569],{"className":4568,"style":1518},[374],[15,4570],{},[15,4572],{"className":4573},[103,1410],[15,4575],{"className":4576,"style":81},[42],[15,4578,4210],{"className":4579},[85],[15,4581],{"className":4582,"style":81},[42],[15,4584,4586,4589,4727],{"className":4585},[27],[15,4587],{"className":4588,"style":1764},[31],[15,4590,4592,4595,4724],{"className":4591},[36],[15,4593],{"className":4594},[67,1410],[15,4596,4598],{"className":4597},[1414],[15,4599,4601,4716],{"className":4600},[365,366],[15,4602,4604,4713],{"className":4603},[370],[15,4605,4607,4662,4670],{"className":4606,"style":1783},[374],[15,4608,4609,4612],{"style":1427},[15,4610],{"className":4611,"style":1431},[382],[15,4613,4615,4618,4621,4624,4627],{"className":4614},[36],[15,4616,497],{"className":4617},[36],[15,4619],{"className":4620,"style":81},[42],[15,4622,86],{"className":4623},[85],[15,4625],{"className":4626,"style":81},[42],[15,4628,4630,4633],{"className":4629},[36],[15,4631,1453],{"className":4632},[36,37],[15,4634,4636],{"className":4635},[361],[15,4637,4639],{"className":4638},[365],[15,4640,4642],{"className":4641},[370],[15,4643,4645],{"className":4644,"style":1466},[374],[15,4646,4647,4650],{"style":1469},[15,4648],{"className":4649,"style":383},[382],[15,4651,4653],{"className":4652},[387,388,389,390],[15,4654,4656,4659],{"className":4655},[36,390],[15,4657,1482],{"className":4658},[36,390],[15,4660,230],{"className":4661,"style":229},[36,37,390],[15,4663,4664,4667],{"style":1488},[15,4665],{"className":4666,"style":1431},[382],[15,4668],{"className":4669,"style":1496},[1495],[15,4671,4672,4675],{"style":1499},[15,4673],{"className":4674,"style":1431},[382],[15,4676,4678],{"className":4677},[36],[15,4679,4681,4684],{"className":4680},[36],[15,4682,1453],{"className":4683},[36,37],[15,4685,4687],{"className":4686},[361],[15,4688,4690],{"className":4689},[365],[15,4691,4693],{"className":4692},[370],[15,4694,4696],{"className":4695,"style":1945},[374],[15,4697,4698,4701],{"style":1202},[15,4699],{"className":4700,"style":383},[382],[15,4702,4704],{"className":4703},[387,388,389,390],[15,4705,4707,4710],{"className":4706},[36,390],[15,4708,1482],{"className":4709},[36,390],[15,4711,230],{"className":4712,"style":229},[36,37,390],[15,4714,399],{"className":4715},[398],[15,4717,4719],{"className":4718},[370],[15,4720,4722],{"className":4721,"style":1518},[374],[15,4723],{},[15,4725],{"className":4726},[103,1410],[15,4728,4730],{"className":4729},[36],".",[11,4732,4733,4734,4758,4759,4801,4802,5134],{},"The first factor is ",[15,4735,4737],{"className":4736},[18],[15,4738,4740],{"className":4739,"ariaHidden":23},[22],[15,4741,4743,4746,4749,4752,4755],{"className":4742},[27],[15,4744],{"className":4745,"style":58},[31],[15,4747,1541],{"className":4748,"style":62},[36,37],[15,4750,68],{"className":4751},[67],[15,4753,230],{"className":4754,"style":229},[36,37],[15,4756,104],{"className":4757},[103],"; the second is ",[15,4760,4762],{"className":4761},[18],[15,4763,4765,4783],{"className":4764,"ariaHidden":23},[22],[15,4766,4768,4771,4774,4777,4780],{"className":4767},[27],[15,4769],{"className":4770,"style":1991},[31],[15,4772,497],{"className":4773},[36],[15,4775],{"className":4776,"style":81},[42],[15,4778,1482],{"className":4779},[85],[15,4781],{"className":4782,"style":81},[42],[15,4784,4786,4789,4792,4795,4798],{"className":4785},[27],[15,4787],{"className":4788,"style":58},[31],[15,4790,1541],{"className":4791,"style":62},[36,37],[15,4793,68],{"className":4794},[67],[15,4796,230],{"className":4797,"style":229},[36,37],[15,4799,104],{"className":4800},[103],", since\n",[15,4803,4805],{"className":4804},[18],[15,4806,4808,4826,4844,4862,4986],{"className":4807,"ariaHidden":23},[22],[15,4809,4811,4814,4817,4820,4823],{"className":4810},[27],[15,4812],{"className":4813,"style":1991},[31],[15,4815,497],{"className":4816},[36],[15,4818],{"className":4819,"style":81},[42],[15,4821,1482],{"className":4822},[85],[15,4824],{"className":4825,"style":81},[42],[15,4827,4829,4832,4835,4838,4841],{"className":4828},[27],[15,4830],{"className":4831,"style":225},[31],[15,4833,1541],{"className":4834,"style":62},[36,37],[15,4836],{"className":4837,"style":43},[42],[15,4839,48],{"className":4840},[47],[15,4842],{"className":4843,"style":43},[42],[15,4845,4847,4850,4853,4856,4859],{"className":4846},[27],[15,4848],{"className":4849,"style":1991},[31],[15,4851,497],{"className":4852},[36],[15,4854],{"className":4855,"style":81},[42],[15,4857,1482],{"className":4858},[85],[15,4860],{"className":4861,"style":81},[42],[15,4863,4865,4869,4977,4980,4983],{"className":4864},[27],[15,4866],{"className":4867,"style":4868},[31],"height:1.2484em;vertical-align:-0.4033em;",[15,4870,4872,4875,4974],{"className":4871},[36],[15,4873],{"className":4874},[67,1410],[15,4876,4878],{"className":4877},[1414],[15,4879,4881,4965],{"className":4880},[365,366],[15,4882,4884,4962],{"className":4883},[370],[15,4885,4887,4940,4948],{"className":4886,"style":1689},[374],[15,4888,4889,4892],{"style":1692},[15,4890],{"className":4891,"style":1431},[382],[15,4893,4895],{"className":4894},[387,388,389,390],[15,4896,4898,4901,4904],{"className":4897},[36,390],[15,4899,497],{"className":4900},[36,390],[15,4902,86],{"className":4903},[85,390],[15,4905,4907,4910],{"className":4906},[36,390],[15,4908,1453],{"className":4909},[36,37,390],[15,4911,4913],{"className":4912},[361],[15,4914,4916],{"className":4915},[365],[15,4917,4919],{"className":4918},[370],[15,4920,4923],{"className":4921,"style":4922},[374],"height:0.7027em;",[15,4924,4925,4928],{"style":3726},[15,4926],{"className":4927,"style":3730},[382],[15,4929,4931],{"className":4930},[387,3734,3735,390],[15,4932,4934,4937],{"className":4933},[36,390],[15,4935,1482],{"className":4936},[36,390],[15,4938,230],{"className":4939,"style":229},[36,37,390],[15,4941,4942,4945],{"style":1488},[15,4943],{"className":4944,"style":1431},[382],[15,4946],{"className":4947,"style":1496},[1495],[15,4949,4950,4953],{"style":1716},[15,4951],{"className":4952,"style":1431},[382],[15,4954,4956],{"className":4955},[387,388,389,390],[15,4957,4959],{"className":4958},[36,390],[15,4960,497],{"className":4961},[36,390],[15,4963,399],{"className":4964},[398],[15,4966,4968],{"className":4967},[370],[15,4969,4972],{"className":4970,"style":4971},[374],"height:0.4033em;",[15,4973],{},[15,4975],{"className":4976},[103,1410],[15,4978],{"className":4979,"style":43},[42],[15,4981,48],{"className":4982},[47],[15,4984],{"className":4985,"style":43},[42],[15,4987,4989,4993],{"className":4988},[27],[15,4990],{"className":4991,"style":4992},[31],"height:1.3907em;vertical-align:-0.4033em;",[15,4994,4996,4999,5131],{"className":4995},[36],[15,4997],{"className":4998},[67,1410],[15,5000,5002],{"className":5001},[1414],[15,5003,5005,5123],{"className":5004},[365,366],[15,5006,5008,5120],{"className":5007},[370],[15,5009,5012,5064,5072],{"className":5010,"style":5011},[374],"height:0.9874em;",[15,5013,5014,5017],{"style":1692},[15,5015],{"className":5016,"style":1431},[382],[15,5018,5020],{"className":5019},[387,388,389,390],[15,5021,5023,5026,5029],{"className":5022},[36,390],[15,5024,497],{"className":5025},[36,390],[15,5027,86],{"className":5028},[85,390],[15,5030,5032,5035],{"className":5031},[36,390],[15,5033,1453],{"className":5034},[36,37,390],[15,5036,5038],{"className":5037},[361],[15,5039,5041],{"className":5040},[365],[15,5042,5044],{"className":5043},[370],[15,5045,5047],{"className":5046,"style":4922},[374],[15,5048,5049,5052],{"style":3726},[15,5050],{"className":5051,"style":3730},[382],[15,5053,5055],{"className":5054},[387,3734,3735,390],[15,5056,5058,5061],{"className":5057},[36,390],[15,5059,1482],{"className":5060},[36,390],[15,5062,230],{"className":5063,"style":229},[36,37,390],[15,5065,5066,5069],{"style":1488},[15,5067],{"className":5068,"style":1431},[382],[15,5070],{"className":5071,"style":1496},[1495],[15,5073,5074,5077],{"style":1716},[15,5075],{"className":5076,"style":1431},[382],[15,5078,5080],{"className":5079},[387,388,389,390],[15,5081,5083],{"className":5082},[36,390],[15,5084,5086,5089],{"className":5085},[36,390],[15,5087,1453],{"className":5088},[36,37,390],[15,5090,5092],{"className":5091},[361],[15,5093,5095],{"className":5094},[365],[15,5096,5098],{"className":5097},[370],[15,5099,5102],{"className":5100,"style":5101},[374],"height:0.8477em;",[15,5103,5105,5108],{"style":5104},"top:-2.931em;margin-right:0.0714em;",[15,5106],{"className":5107,"style":3730},[382],[15,5109,5111],{"className":5110},[387,3734,3735,390],[15,5112,5114,5117],{"className":5113},[36,390],[15,5115,1482],{"className":5116},[36,390],[15,5118,230],{"className":5119,"style":229},[36,37,390],[15,5121,399],{"className":5122},[398],[15,5124,5126],{"className":5125},[370],[15,5127,5129],{"className":5128,"style":4971},[374],[15,5130],{},[15,5132],{"className":5133},[103,1410],". Hence",[15,5136,5138],{"className":5137},[4032],[15,5139,5141],{"className":5140},[18],[15,5142,5144,5203],{"className":5143,"ariaHidden":23},[22],[15,5145,5147,5150,5153,5185,5188,5191,5194,5197,5200],{"className":5146},[27],[15,5148],{"className":5149,"style":4045},[31],[15,5151],{"className":5152,"style":43},[42],[15,5154,5156,5159],{"className":5155},[36],[15,5157,1541],{"className":5158,"style":62},[36,37],[15,5160,5162],{"className":5161},[361],[15,5163,5165],{"className":5164},[365],[15,5166,5168],{"className":5167},[370],[15,5169,5171],{"className":5170,"style":4064},[374],[15,5172,5173,5176],{"style":4067},[15,5174],{"className":5175,"style":383},[382],[15,5177,5179],{"className":5178},[387,388,389,390],[15,5180,5182],{"className":5181},[36,390],[15,5183,1215],{"className":5184},[36,390],[15,5186,68],{"className":5187},[67],[15,5189,230],{"className":5190,"style":229},[36,37],[15,5192,104],{"className":5193},[103],[15,5195],{"className":5196,"style":43},[42],[15,5198,48],{"className":5199},[47],[15,5201],{"className":5202,"style":43},[42],[15,5204,5206,5209,5212,5215,5218,5221,5224,5227,5263],{"className":5205},[27],[15,5207],{"className":5208,"style":58},[31],[15,5210,1541],{"className":5211,"style":62},[36,37],[15,5213,68],{"className":5214},[67],[15,5216,230],{"className":5217,"style":229},[36,37],[15,5219,104],{"className":5220},[103],[15,5222],{"className":5223,"style":1598},[42],[15,5225],{"className":5226,"style":1598},[42],[15,5228,5231,5236,5239,5242,5245,5248,5251,5254,5257,5260],{"className":5229},[5230],"minner",[15,5232,68],{"className":5233,"style":5235},[67,5234],"delimcenter","top:0em;",[15,5237,497],{"className":5238},[36],[15,5240],{"className":5241,"style":81},[42],[15,5243,1482],{"className":5244},[85],[15,5246],{"className":5247,"style":81},[42],[15,5249,1541],{"className":5250,"style":62},[36,37],[15,5252,68],{"className":5253},[67],[15,5255,230],{"className":5256,"style":229},[36,37],[15,5258,104],{"className":5259},[103],[15,5261,104],{"className":5262,"style":5235},[103,5234],[15,5264],{"className":5265,"style":43},[42],[11,5267,5268,5269,5272,5273,5288,5289,5331,5332,5374,5375,5424,5425,5467,5468,5483,5484,5564],{},"Two consequences read straight off this form. The derivative is expressed in the\n",[1227,5270,5271],{},"output"," ",[15,5274,5276],{"className":5275},[18],[15,5277,5279],{"className":5278,"ariaHidden":23},[22],[15,5280,5282,5285],{"className":5281},[27],[15,5283],{"className":5284,"style":225},[31],[15,5286,1541],{"className":5287,"style":62},[36,37],", so a forward pass that cached ",[15,5290,5292],{"className":5291},[18],[15,5293,5295,5313],{"className":5294,"ariaHidden":23},[22],[15,5296,5298,5301,5304,5307,5310],{"className":5297},[27],[15,5299],{"className":5300,"style":32},[31],[15,5302,38],{"className":5303},[36,37],[15,5305],{"className":5306,"style":43},[42],[15,5308,48],{"className":5309},[47],[15,5311],{"className":5312,"style":43},[42],[15,5314,5316,5319,5322,5325,5328],{"className":5315},[27],[15,5317],{"className":5318,"style":58},[31],[15,5320,1541],{"className":5321,"style":62},[36,37],[15,5323,68],{"className":5324},[67],[15,5326,230],{"className":5327,"style":229},[36,37],[15,5329,104],{"className":5330},[103]," needs no re-exp in\nthe backward pass, only ",[15,5333,5335],{"className":5334},[18],[15,5336,5338,5362],{"className":5337,"ariaHidden":23},[22],[15,5339,5341,5344,5347,5350,5353,5356,5359],{"className":5340},[27],[15,5342],{"className":5343,"style":58},[31],[15,5345,38],{"className":5346},[36,37],[15,5348,68],{"className":5349},[67],[15,5351,497],{"className":5352},[36],[15,5354],{"className":5355,"style":81},[42],[15,5357,1482],{"className":5358},[85],[15,5360],{"className":5361,"style":81},[42],[15,5363,5365,5368,5371],{"className":5364},[27],[15,5366],{"className":5367,"style":58},[31],[15,5369,38],{"className":5370},[36,37],[15,5372,104],{"className":5373},[103],". And because ",[15,5376,5378],{"className":5377},[18],[15,5379,5381,5400],{"className":5380,"ariaHidden":23},[22],[15,5382,5384,5387,5390,5393,5397],{"className":5383},[27],[15,5385],{"className":5386,"style":2665},[31],[15,5388,1541],{"className":5389,"style":62},[36,37],[15,5391],{"className":5392,"style":43},[42],[15,5394,5396],{"className":5395},[47],"∈",[15,5398],{"className":5399,"style":43},[42],[15,5401,5403,5406,5409,5412,5415,5418,5421],{"className":5402},[27],[15,5404],{"className":5405,"style":58},[31],[15,5407,68],{"className":5408},[67],[15,5410,1589],{"className":5411},[36],[15,5413,1594],{"className":5414},[1593],[15,5416],{"className":5417,"style":1598},[42],[15,5419,497],{"className":5420},[36],[15,5422,104],{"className":5423},[103],", the product\n",[15,5426,5428],{"className":5427},[18],[15,5429,5431,5455],{"className":5430,"ariaHidden":23},[22],[15,5432,5434,5437,5440,5443,5446,5449,5452],{"className":5433},[27],[15,5435],{"className":5436,"style":58},[31],[15,5438,1541],{"className":5439,"style":62},[36,37],[15,5441,68],{"className":5442},[67],[15,5444,497],{"className":5445},[36],[15,5447],{"className":5448,"style":81},[42],[15,5450,1482],{"className":5451},[85],[15,5453],{"className":5454,"style":81},[42],[15,5456,5458,5461,5464],{"className":5457},[27],[15,5459],{"className":5460,"style":58},[31],[15,5462,1541],{"className":5463,"style":62},[36,37],[15,5465,104],{"className":5466},[103]," is a downward parabola in ",[15,5469,5471],{"className":5470},[18],[15,5472,5474],{"className":5473,"ariaHidden":23},[22],[15,5475,5477,5480],{"className":5476},[27],[15,5478],{"className":5479,"style":225},[31],[15,5481,1541],{"className":5482,"style":62},[36,37]," capped at its vertex, which\nis the source of the ",[15,5485,5487],{"className":5486},[18],[15,5488,5490],{"className":5489,"ariaHidden":23},[22],[15,5491,5493,5496],{"className":5492},[27],[15,5494],{"className":5495,"style":1670},[31],[15,5497,5499,5502,5561],{"className":5498},[36],[15,5500],{"className":5501},[67,1410],[15,5503,5505],{"className":5504},[1414],[15,5506,5508,5553],{"className":5507},[365,366],[15,5509,5511,5550],{"className":5510},[370],[15,5512,5514,5528,5536],{"className":5513,"style":1689},[374],[15,5515,5516,5519],{"style":1692},[15,5517],{"className":5518,"style":1431},[382],[15,5520,5522],{"className":5521},[387,388,389,390],[15,5523,5525],{"className":5524},[36,390],[15,5526,1705],{"className":5527},[36,390],[15,5529,5530,5533],{"style":1488},[15,5531],{"className":5532,"style":1431},[382],[15,5534],{"className":5535,"style":1496},[1495],[15,5537,5538,5541],{"style":1716},[15,5539],{"className":5540,"style":1431},[382],[15,5542,5544],{"className":5543},[387,388,389,390],[15,5545,5547],{"className":5546},[36,390],[15,5548,497],{"className":5549},[36,390],[15,5551,399],{"className":5552},[398],[15,5554,5556],{"className":5555},[370],[15,5557,5559],{"className":5558,"style":1738},[374],[15,5560],{},[15,5562],{"className":5563},[103,1410]," ceiling derived below.",[3890,5566,5567],{"id":1749},"Tanh",[11,5569,5570,5571,5805,5806,6055,6056,6179],{},"The tanh derivative follows the same pattern. With ",[15,5572,5574],{"className":5573},[18],[15,5575,5577,5601],{"className":5576,"ariaHidden":23},[22],[15,5578,5580,5583,5586,5589,5592,5595,5598],{"className":5579},[27],[15,5581],{"className":5582,"style":32},[31],[15,5584,1749],{"className":5585},[2014],[15,5587],{"className":5588,"style":1598},[42],[15,5590,230],{"className":5591,"style":229},[36,37],[15,5593],{"className":5594,"style":43},[42],[15,5596,48],{"className":5597},[47],[15,5599],{"className":5600,"style":43},[42],[15,5602,5604,5607],{"className":5603},[27],[15,5605],{"className":5606,"style":4992},[31],[15,5608,5610,5613,5802],{"className":5609},[36],[15,5611],{"className":5612},[67,1410],[15,5614,5616],{"className":5615},[1414],[15,5617,5619,5794],{"className":5618},[365,366],[15,5620,5622,5791],{"className":5621},[370],[15,5623,5625,5704,5712],{"className":5624,"style":5011},[374],[15,5626,5627,5630],{"style":1692},[15,5628],{"className":5629,"style":1431},[382],[15,5631,5633],{"className":5632},[387,388,389,390],[15,5634,5636,5666,5669],{"className":5635},[36,390],[15,5637,5639,5642],{"className":5638},[36,390],[15,5640,1453],{"className":5641},[36,37,390],[15,5643,5645],{"className":5644},[361],[15,5646,5648],{"className":5647},[365],[15,5649,5651],{"className":5650},[370],[15,5652,5655],{"className":5653,"style":5654},[374],"height:0.5935em;",[15,5656,5657,5660],{"style":3726},[15,5658],{"className":5659,"style":3730},[382],[15,5661,5663],{"className":5662},[387,3734,3735,390],[15,5664,230],{"className":5665,"style":229},[36,37,390],[15,5667,86],{"className":5668},[85,390],[15,5670,5672,5675],{"className":5671},[36,390],[15,5673,1453],{"className":5674},[36,37,390],[15,5676,5678],{"className":5677},[361],[15,5679,5681],{"className":5680},[365],[15,5682,5684],{"className":5683},[370],[15,5685,5687],{"className":5686,"style":4922},[374],[15,5688,5689,5692],{"style":3726},[15,5690],{"className":5691,"style":3730},[382],[15,5693,5695],{"className":5694},[387,3734,3735,390],[15,5696,5698,5701],{"className":5697},[36,390],[15,5699,1482],{"className":5700},[36,390],[15,5702,230],{"className":5703,"style":229},[36,37,390],[15,5705,5706,5709],{"style":1488},[15,5707],{"className":5708,"style":1431},[382],[15,5710],{"className":5711,"style":1496},[1495],[15,5713,5714,5717],{"style":1716},[15,5715],{"className":5716,"style":1431},[382],[15,5718,5720],{"className":5719},[387,388,389,390],[15,5721,5723,5753,5756],{"className":5722},[36,390],[15,5724,5726,5729],{"className":5725},[36,390],[15,5727,1453],{"className":5728},[36,37,390],[15,5730,5732],{"className":5731},[361],[15,5733,5735],{"className":5734},[365],[15,5736,5738],{"className":5737},[370],[15,5739,5742],{"className":5740,"style":5741},[374],"height:0.7385em;",[15,5743,5744,5747],{"style":5104},[15,5745],{"className":5746,"style":3730},[382],[15,5748,5750],{"className":5749},[387,3734,3735,390],[15,5751,230],{"className":5752,"style":229},[36,37,390],[15,5754,1482],{"className":5755},[85,390],[15,5757,5759,5762],{"className":5758},[36,390],[15,5760,1453],{"className":5761},[36,37,390],[15,5763,5765],{"className":5764},[361],[15,5766,5768],{"className":5767},[365],[15,5769,5771],{"className":5770},[370],[15,5772,5774],{"className":5773,"style":5101},[374],[15,5775,5776,5779],{"style":5104},[15,5777],{"className":5778,"style":3730},[382],[15,5780,5782],{"className":5781},[387,3734,3735,390],[15,5783,5785,5788],{"className":5784},[36,390],[15,5786,1482],{"className":5787},[36,390],[15,5789,230],{"className":5790,"style":229},[36,37,390],[15,5792,399],{"className":5793},[398],[15,5795,5797],{"className":5796},[370],[15,5798,5800],{"className":5799,"style":4971},[374],[15,5801],{},[15,5803],{"className":5804},[103,1410],",\nthe quotient rule gives numerator ",[15,5807,5809],{"className":5808},[18],[15,5810,5812,5859,5938,5985],{"className":5811,"ariaHidden":23},[22],[15,5813,5815,5818,5821,5850,5853,5856],{"className":5814},[27],[15,5816],{"className":5817,"style":58},[31],[15,5819,68],{"className":5820},[67],[15,5822,5824,5827],{"className":5823},[36],[15,5825,1453],{"className":5826},[36,37],[15,5828,5830],{"className":5829},[361],[15,5831,5833],{"className":5832},[365],[15,5834,5836],{"className":5835},[370],[15,5837,5839],{"className":5838,"style":1903},[374],[15,5840,5841,5844],{"style":1202},[15,5842],{"className":5843,"style":383},[382],[15,5845,5847],{"className":5846},[387,388,389,390],[15,5848,230],{"className":5849,"style":229},[36,37,390],[15,5851],{"className":5852,"style":81},[42],[15,5854,86],{"className":5855},[85],[15,5857],{"className":5858,"style":81},[42],[15,5860,5862,5865,5900,5929,5932,5935],{"className":5861},[27],[15,5863],{"className":5864,"style":3956},[31],[15,5866,5868,5871],{"className":5867},[36],[15,5869,1453],{"className":5870},[36,37],[15,5872,5874],{"className":5873},[361],[15,5875,5877],{"className":5876},[365],[15,5878,5880],{"className":5879},[370],[15,5881,5883],{"className":5882,"style":1945},[374],[15,5884,5885,5888],{"style":1202},[15,5886],{"className":5887,"style":383},[382],[15,5889,5891],{"className":5890},[387,388,389,390],[15,5892,5894,5897],{"className":5893},[36,390],[15,5895,1482],{"className":5896},[36,390],[15,5898,230],{"className":5899,"style":229},[36,37,390],[15,5901,5903,5906],{"className":5902},[103],[15,5904,104],{"className":5905},[103],[15,5907,5909],{"className":5908},[361],[15,5910,5912],{"className":5911},[365],[15,5913,5915],{"className":5914},[370],[15,5916,5918],{"className":5917,"style":4010},[374],[15,5919,5920,5923],{"style":1202},[15,5921],{"className":5922,"style":383},[382],[15,5924,5926],{"className":5925},[387,388,389,390],[15,5927,394],{"className":5928},[36,390],[15,5930],{"className":5931,"style":81},[42],[15,5933,1482],{"className":5934},[85],[15,5936],{"className":5937,"style":81},[42],[15,5939,5941,5944,5947,5976,5979,5982],{"className":5940},[27],[15,5942],{"className":5943,"style":58},[31],[15,5945,68],{"className":5946},[67],[15,5948,5950,5953],{"className":5949},[36],[15,5951,1453],{"className":5952},[36,37],[15,5954,5956],{"className":5955},[361],[15,5957,5959],{"className":5958},[365],[15,5960,5962],{"className":5961},[370],[15,5963,5965],{"className":5964,"style":1903},[374],[15,5966,5967,5970],{"style":1202},[15,5968],{"className":5969,"style":383},[382],[15,5971,5973],{"className":5972},[387,388,389,390],[15,5974,230],{"className":5975,"style":229},[36,37,390],[15,5977],{"className":5978,"style":81},[42],[15,5980,1482],{"className":5981},[85],[15,5983],{"className":5984,"style":81},[42],[15,5986,5988,5991,6026],{"className":5987},[27],[15,5989],{"className":5990,"style":3956},[31],[15,5992,5994,5997],{"className":5993},[36],[15,5995,1453],{"className":5996},[36,37],[15,5998,6000],{"className":5999},[361],[15,6001,6003],{"className":6002},[365],[15,6004,6006],{"className":6005},[370],[15,6007,6009],{"className":6008,"style":1945},[374],[15,6010,6011,6014],{"style":1202},[15,6012],{"className":6013,"style":383},[382],[15,6015,6017],{"className":6016},[387,388,389,390],[15,6018,6020,6023],{"className":6019},[36,390],[15,6021,1482],{"className":6022},[36,390],[15,6024,230],{"className":6025,"style":229},[36,37,390],[15,6027,6029,6032],{"className":6028},[103],[15,6030,104],{"className":6031},[103],[15,6033,6035],{"className":6034},[361],[15,6036,6038],{"className":6037},[365],[15,6039,6041],{"className":6040},[370],[15,6042,6044],{"className":6043,"style":4010},[374],[15,6045,6046,6049],{"style":1202},[15,6047],{"className":6048,"style":383},[382],[15,6050,6052],{"className":6051},[387,388,389,390],[15,6053,394],{"className":6054},[36,390]," over\n",[15,6057,6059],{"className":6058},[18],[15,6060,6062,6109],{"className":6061,"ariaHidden":23},[22],[15,6063,6065,6068,6071,6100,6103,6106],{"className":6064},[27],[15,6066],{"className":6067,"style":58},[31],[15,6069,68],{"className":6070},[67],[15,6072,6074,6077],{"className":6073},[36],[15,6075,1453],{"className":6076},[36,37],[15,6078,6080],{"className":6079},[361],[15,6081,6083],{"className":6082},[365],[15,6084,6086],{"className":6085},[370],[15,6087,6089],{"className":6088,"style":1903},[374],[15,6090,6091,6094],{"style":1202},[15,6092],{"className":6093,"style":383},[382],[15,6095,6097],{"className":6096},[387,388,389,390],[15,6098,230],{"className":6099,"style":229},[36,37,390],[15,6101],{"className":6102,"style":81},[42],[15,6104,86],{"className":6105},[85],[15,6107],{"className":6108,"style":81},[42],[15,6110,6112,6115,6150],{"className":6111},[27],[15,6113],{"className":6114,"style":3956},[31],[15,6116,6118,6121],{"className":6117},[36],[15,6119,1453],{"className":6120},[36,37],[15,6122,6124],{"className":6123},[361],[15,6125,6127],{"className":6126},[365],[15,6128,6130],{"className":6129},[370],[15,6131,6133],{"className":6132,"style":1945},[374],[15,6134,6135,6138],{"style":1202},[15,6136],{"className":6137,"style":383},[382],[15,6139,6141],{"className":6140},[387,388,389,390],[15,6142,6144,6147],{"className":6143},[36,390],[15,6145,1482],{"className":6146},[36,390],[15,6148,230],{"className":6149,"style":229},[36,37,390],[15,6151,6153,6156],{"className":6152},[103],[15,6154,104],{"className":6155},[103],[15,6157,6159],{"className":6158},[361],[15,6160,6162],{"className":6161},[365],[15,6163,6165],{"className":6164},[370],[15,6166,6168],{"className":6167,"style":4010},[374],[15,6169,6170,6173],{"style":1202},[15,6171],{"className":6172,"style":383},[382],[15,6174,6176],{"className":6175},[387,388,389,390],[15,6177,394],{"className":6178},[36,390],", and expanding the two squares,",[15,6181,6183],{"className":6182},[4032],[15,6184,6186],{"className":6185},[18],[15,6187,6189,6237,6316,6363,6442,6568,6692],{"className":6188,"ariaHidden":23},[22],[15,6190,6192,6195,6198,6228,6231,6234],{"className":6191},[27],[15,6193],{"className":6194,"style":58},[31],[15,6196,68],{"className":6197},[67],[15,6199,6201,6204],{"className":6200},[36],[15,6202,1453],{"className":6203},[36,37],[15,6205,6207],{"className":6206},[361],[15,6208,6210],{"className":6209},[365],[15,6211,6213],{"className":6212},[370],[15,6214,6217],{"className":6215,"style":6216},[374],"height:0.7144em;",[15,6218,6219,6222],{"style":4067},[15,6220],{"className":6221,"style":383},[382],[15,6223,6225],{"className":6224},[387,388,389,390],[15,6226,230],{"className":6227,"style":229},[36,37,390],[15,6229],{"className":6230,"style":81},[42],[15,6232,86],{"className":6233},[85],[15,6235],{"className":6236,"style":81},[42],[15,6238,6240,6243,6278,6307,6310,6313],{"className":6239},[27],[15,6241],{"className":6242,"style":4131},[31],[15,6244,6246,6249],{"className":6245},[36],[15,6247,1453],{"className":6248},[36,37],[15,6250,6252],{"className":6251},[361],[15,6253,6255],{"className":6254},[365],[15,6256,6258],{"className":6257},[370],[15,6259,6261],{"className":6260,"style":4150},[374],[15,6262,6263,6266],{"style":4067},[15,6264],{"className":6265,"style":383},[382],[15,6267,6269],{"className":6268},[387,388,389,390],[15,6270,6272,6275],{"className":6271},[36,390],[15,6273,1482],{"className":6274},[36,390],[15,6276,230],{"className":6277,"style":229},[36,37,390],[15,6279,6281,6284],{"className":6280},[103],[15,6282,104],{"className":6283},[103],[15,6285,6287],{"className":6286},[361],[15,6288,6290],{"className":6289},[365],[15,6291,6293],{"className":6292},[370],[15,6294,6296],{"className":6295,"style":4186},[374],[15,6297,6298,6301],{"style":4067},[15,6299],{"className":6300,"style":383},[382],[15,6302,6304],{"className":6303},[387,388,389,390],[15,6305,394],{"className":6306},[36,390],[15,6308],{"className":6309,"style":81},[42],[15,6311,1482],{"className":6312},[85],[15,6314],{"className":6315,"style":81},[42],[15,6317,6319,6322,6325,6354,6357,6360],{"className":6318},[27],[15,6320],{"className":6321,"style":58},[31],[15,6323,68],{"className":6324},[67],[15,6326,6328,6331],{"className":6327},[36],[15,6329,1453],{"className":6330},[36,37],[15,6332,6334],{"className":6333},[361],[15,6335,6337],{"className":6336},[365],[15,6338,6340],{"className":6339},[370],[15,6341,6343],{"className":6342,"style":6216},[374],[15,6344,6345,6348],{"style":4067},[15,6346],{"className":6347,"style":383},[382],[15,6349,6351],{"className":6350},[387,388,389,390],[15,6352,230],{"className":6353,"style":229},[36,37,390],[15,6355],{"className":6356,"style":81},[42],[15,6358,1482],{"className":6359},[85],[15,6361],{"className":6362,"style":81},[42],[15,6364,6366,6369,6404,6433,6436,6439],{"className":6365},[27],[15,6367],{"className":6368,"style":4131},[31],[15,6370,6372,6375],{"className":6371},[36],[15,6373,1453],{"className":6374},[36,37],[15,6376,6378],{"className":6377},[361],[15,6379,6381],{"className":6380},[365],[15,6382,6384],{"className":6383},[370],[15,6385,6387],{"className":6386,"style":4150},[374],[15,6388,6389,6392],{"style":4067},[15,6390],{"className":6391,"style":383},[382],[15,6393,6395],{"className":6394},[387,388,389,390],[15,6396,6398,6401],{"className":6397},[36,390],[15,6399,1482],{"className":6400},[36,390],[15,6402,230],{"className":6403,"style":229},[36,37,390],[15,6405,6407,6410],{"className":6406},[103],[15,6408,104],{"className":6409},[103],[15,6411,6413],{"className":6412},[361],[15,6414,6416],{"className":6415},[365],[15,6417,6419],{"className":6418},[370],[15,6420,6422],{"className":6421,"style":4186},[374],[15,6423,6424,6427],{"style":4067},[15,6425],{"className":6426,"style":383},[382],[15,6428,6430],{"className":6429},[387,388,389,390],[15,6431,394],{"className":6432},[36,390],[15,6434],{"className":6435,"style":43},[42],[15,6437,48],{"className":6438},[47],[15,6440],{"className":6441,"style":43},[42],[15,6443,6445,6449,6559,6562,6565],{"className":6444},[27],[15,6446],{"className":6447,"style":6448},[31],"height:1.2141em;vertical-align:-0.35em;",[15,6450,6452,6459,6494,6497,6500,6503,6506,6509,6512,6515,6553],{"className":6451},[5230],[15,6453,6455],{"className":6454,"style":5235},[67,5234],[15,6456,68],{"className":6457},[6458,3735],"delimsizing",[15,6460,6462,6465],{"className":6461},[36],[15,6463,1453],{"className":6464},[36,37],[15,6466,6468],{"className":6467},[361],[15,6469,6471],{"className":6470},[365],[15,6472,6474],{"className":6473},[370],[15,6475,6477],{"className":6476,"style":4186},[374],[15,6478,6479,6482],{"style":4067},[15,6480],{"className":6481,"style":383},[382],[15,6483,6485],{"className":6484},[387,388,389,390],[15,6486,6488,6491],{"className":6487},[36,390],[15,6489,394],{"className":6490},[36,390],[15,6492,230],{"className":6493,"style":229},[36,37,390],[15,6495],{"className":6496,"style":81},[42],[15,6498,86],{"className":6499},[85],[15,6501],{"className":6502,"style":81},[42],[15,6504,394],{"className":6505},[36],[15,6507],{"className":6508,"style":81},[42],[15,6510,86],{"className":6511},[85],[15,6513],{"className":6514,"style":81},[42],[15,6516,6518,6521],{"className":6517},[36],[15,6519,1453],{"className":6520},[36,37],[15,6522,6524],{"className":6523},[361],[15,6525,6527],{"className":6526},[365],[15,6528,6530],{"className":6529},[370],[15,6531,6533],{"className":6532,"style":4186},[374],[15,6534,6535,6538],{"style":4067},[15,6536],{"className":6537,"style":383},[382],[15,6539,6541],{"className":6540},[387,388,389,390],[15,6542,6544,6547,6550],{"className":6543},[36,390],[15,6545,1482],{"className":6546},[36,390],[15,6548,394],{"className":6549},[36,390],[15,6551,230],{"className":6552,"style":229},[36,37,390],[15,6554,6556],{"className":6555,"style":5235},[103,5234],[15,6557,104],{"className":6558},[6458,3735],[15,6560],{"className":6561,"style":81},[42],[15,6563,1482],{"className":6564},[85],[15,6566],{"className":6567,"style":81},[42],[15,6569,6571,6574,6683,6686,6689],{"className":6570},[27],[15,6572],{"className":6573,"style":6448},[31],[15,6575,6577,6583,6618,6621,6624,6627,6630,6633,6636,6639,6677],{"className":6576},[5230],[15,6578,6580],{"className":6579,"style":5235},[67,5234],[15,6581,68],{"className":6582},[6458,3735],[15,6584,6586,6589],{"className":6585},[36],[15,6587,1453],{"className":6588},[36,37],[15,6590,6592],{"className":6591},[361],[15,6593,6595],{"className":6594},[365],[15,6596,6598],{"className":6597},[370],[15,6599,6601],{"className":6600,"style":4186},[374],[15,6602,6603,6606],{"style":4067},[15,6604],{"className":6605,"style":383},[382],[15,6607,6609],{"className":6608},[387,388,389,390],[15,6610,6612,6615],{"className":6611},[36,390],[15,6613,394],{"className":6614},[36,390],[15,6616,230],{"className":6617,"style":229},[36,37,390],[15,6619],{"className":6620,"style":81},[42],[15,6622,1482],{"className":6623},[85],[15,6625],{"className":6626,"style":81},[42],[15,6628,394],{"className":6629},[36],[15,6631],{"className":6632,"style":81},[42],[15,6634,86],{"className":6635},[85],[15,6637],{"className":6638,"style":81},[42],[15,6640,6642,6645],{"className":6641},[36],[15,6643,1453],{"className":6644},[36,37],[15,6646,6648],{"className":6647},[361],[15,6649,6651],{"className":6650},[365],[15,6652,6654],{"className":6653},[370],[15,6655,6657],{"className":6656,"style":4186},[374],[15,6658,6659,6662],{"style":4067},[15,6660],{"className":6661,"style":383},[382],[15,6663,6665],{"className":6664},[387,388,389,390],[15,6666,6668,6671,6674],{"className":6667},[36,390],[15,6669,1482],{"className":6670},[36,390],[15,6672,394],{"className":6673},[36,390],[15,6675,230],{"className":6676,"style":229},[36,37,390],[15,6678,6680],{"className":6679,"style":5235},[103,5234],[15,6681,104],{"className":6682},[6458,3735],[15,6684],{"className":6685,"style":43},[42],[15,6687,48],{"className":6688},[47],[15,6690],{"className":6691,"style":43},[42],[15,6693,6695,6699,6702],{"className":6694},[27],[15,6696],{"className":6697,"style":6698},[31],"height:0.8389em;vertical-align:-0.1944em;",[15,6700,1705],{"className":6701},[36],[15,6703,1594],{"className":6704},[1593],[11,6706,6707],{},"so",[15,6709,6711],{"className":6710},[4032],[15,6712,6714],{"className":6713},[18],[15,6715,6717,6775,7172,7190,7474,7492],{"className":6716,"ariaHidden":23},[22],[15,6718,6720,6724,6757,6760,6763,6766,6769,6772],{"className":6719},[27],[15,6721],{"className":6722,"style":6723},[31],"height:1.0862em;vertical-align:-0.25em;",[15,6725,6727,6730],{"className":6726},[2014],[15,6728,1749],{"className":6729},[2014],[15,6731,6733],{"className":6732},[361],[15,6734,6736],{"className":6735},[365],[15,6737,6739],{"className":6738},[370],[15,6740,6743],{"className":6741,"style":6742},[374],"height:0.8362em;",[15,6744,6745,6748],{"style":2032},[15,6746],{"className":6747,"style":383},[382],[15,6749,6751],{"className":6750},[387,388,389,390],[15,6752,6754],{"className":6753},[36,390],[15,6755,1215],{"className":6756},[36,390],[15,6758,68],{"className":6759},[67],[15,6761,230],{"className":6762,"style":229},[36,37],[15,6764,104],{"className":6765},[103],[15,6767],{"className":6768,"style":43},[42],[15,6770,48],{"className":6771},[47],[15,6773],{"className":6774,"style":43},[42],[15,6776,6778,6782,7163,7166,7169],{"className":6777},[27],[15,6779],{"className":6780,"style":6781},[31],"height:2.4271em;vertical-align:-0.936em;",[15,6783,6785,6788,7160],{"className":6784},[36],[15,6786],{"className":6787},[67,1410],[15,6789,6791],{"className":6790},[1414],[15,6792,6794,7152],{"className":6793},[365,366],[15,6795,6797,7149],{"className":6796},[370],[15,6798,6801,6914,6922],{"className":6799,"style":6800},[374],"height:1.4911em;",[15,6802,6803,6806],{"style":1427},[15,6804],{"className":6805,"style":1431},[382],[15,6807,6809,6812,6841,6844,6847,6850,6885],{"className":6808},[36],[15,6810,68],{"className":6811},[67],[15,6813,6815,6818],{"className":6814},[36],[15,6816,1453],{"className":6817},[36,37],[15,6819,6821],{"className":6820},[361],[15,6822,6824],{"className":6823},[365],[15,6825,6827],{"className":6826},[370],[15,6828,6830],{"className":6829,"style":1810},[374],[15,6831,6832,6835],{"style":1469},[15,6833],{"className":6834,"style":383},[382],[15,6836,6838],{"className":6837},[387,388,389,390],[15,6839,230],{"className":6840,"style":229},[36,37,390],[15,6842],{"className":6843,"style":81},[42],[15,6845,86],{"className":6846},[85],[15,6848],{"className":6849,"style":81},[42],[15,6851,6853,6856],{"className":6852},[36],[15,6854,1453],{"className":6855},[36,37],[15,6857,6859],{"className":6858},[361],[15,6860,6862],{"className":6861},[365],[15,6863,6865],{"className":6864},[370],[15,6866,6868],{"className":6867,"style":1466},[374],[15,6869,6870,6873],{"style":1469},[15,6871],{"className":6872,"style":383},[382],[15,6874,6876],{"className":6875},[387,388,389,390],[15,6877,6879,6882],{"className":6878},[36,390],[15,6880,1482],{"className":6881},[36,390],[15,6883,230],{"className":6884,"style":229},[36,37,390],[15,6886,6888,6891],{"className":6887},[103],[15,6889,104],{"className":6890},[103],[15,6892,6894],{"className":6893},[361],[15,6895,6897],{"className":6896},[365],[15,6898,6900],{"className":6899},[370],[15,6901,6903],{"className":6902,"style":4375},[374],[15,6904,6905,6908],{"style":1469},[15,6906],{"className":6907,"style":383},[382],[15,6909,6911],{"className":6910},[387,388,389,390],[15,6912,394],{"className":6913},[36,390],[15,6915,6916,6919],{"style":1488},[15,6917],{"className":6918,"style":1431},[382],[15,6920],{"className":6921,"style":1496},[1495],[15,6923,6924,6927],{"style":1499},[15,6925],{"className":6926,"style":1431},[382],[15,6928,6930,6933,6962,6965,6968,6971,7006,7035,7038,7041,7044,7047,7076,7079,7082,7085,7120],{"className":6929},[36],[15,6931,68],{"className":6932},[67],[15,6934,6936,6939],{"className":6935},[36],[15,6937,1453],{"className":6938},[36,37],[15,6940,6942],{"className":6941},[361],[15,6943,6945],{"className":6944},[365],[15,6946,6948],{"className":6947},[370],[15,6949,6951],{"className":6950,"style":1903},[374],[15,6952,6953,6956],{"style":1202},[15,6954],{"className":6955,"style":383},[382],[15,6957,6959],{"className":6958},[387,388,389,390],[15,6960,230],{"className":6961,"style":229},[36,37,390],[15,6963],{"className":6964,"style":81},[42],[15,6966,86],{"className":6967},[85],[15,6969],{"className":6970,"style":81},[42],[15,6972,6974,6977],{"className":6973},[36],[15,6975,1453],{"className":6976},[36,37],[15,6978,6980],{"className":6979},[361],[15,6981,6983],{"className":6982},[365],[15,6984,6986],{"className":6985},[370],[15,6987,6989],{"className":6988,"style":1945},[374],[15,6990,6991,6994],{"style":1202},[15,6992],{"className":6993,"style":383},[382],[15,6995,6997],{"className":6996},[387,388,389,390],[15,6998,7000,7003],{"className":6999},[36,390],[15,7001,1482],{"className":7002},[36,390],[15,7004,230],{"className":7005,"style":229},[36,37,390],[15,7007,7009,7012],{"className":7008},[103],[15,7010,104],{"className":7011},[103],[15,7013,7015],{"className":7014},[361],[15,7016,7018],{"className":7017},[365],[15,7019,7021],{"className":7020},[370],[15,7022,7024],{"className":7023,"style":4010},[374],[15,7025,7026,7029],{"style":1202},[15,7027],{"className":7028,"style":383},[382],[15,7030,7032],{"className":7031},[387,388,389,390],[15,7033,394],{"className":7034},[36,390],[15,7036],{"className":7037,"style":81},[42],[15,7039,1482],{"className":7040},[85],[15,7042],{"className":7043,"style":81},[42],[15,7045,68],{"className":7046},[67],[15,7048,7050,7053],{"className":7049},[36],[15,7051,1453],{"className":7052},[36,37],[15,7054,7056],{"className":7055},[361],[15,7057,7059],{"className":7058},[365],[15,7060,7062],{"className":7061},[370],[15,7063,7065],{"className":7064,"style":1903},[374],[15,7066,7067,7070],{"style":1202},[15,7068],{"className":7069,"style":383},[382],[15,7071,7073],{"className":7072},[387,388,389,390],[15,7074,230],{"className":7075,"style":229},[36,37,390],[15,7077],{"className":7078,"style":81},[42],[15,7080,1482],{"className":7081},[85],[15,7083],{"className":7084,"style":81},[42],[15,7086,7088,7091],{"className":7087},[36],[15,7089,1453],{"className":7090},[36,37],[15,7092,7094],{"className":7093},[361],[15,7095,7097],{"className":7096},[365],[15,7098,7100],{"className":7099},[370],[15,7101,7103],{"className":7102,"style":1945},[374],[15,7104,7105,7108],{"style":1202},[15,7106],{"className":7107,"style":383},[382],[15,7109,7111],{"className":7110},[387,388,389,390],[15,7112,7114,7117],{"className":7113},[36,390],[15,7115,1482],{"className":7116},[36,390],[15,7118,230],{"className":7119,"style":229},[36,37,390],[15,7121,7123,7126],{"className":7122},[103],[15,7124,104],{"className":7125},[103],[15,7127,7129],{"className":7128},[361],[15,7130,7132],{"className":7131},[365],[15,7133,7135],{"className":7134},[370],[15,7136,7138],{"className":7137,"style":4010},[374],[15,7139,7140,7143],{"style":1202},[15,7141],{"className":7142,"style":383},[382],[15,7144,7146],{"className":7145},[387,388,389,390],[15,7147,394],{"className":7148},[36,390],[15,7150,399],{"className":7151},[398],[15,7153,7155],{"className":7154},[370],[15,7156,7158],{"className":7157,"style":4447},[374],[15,7159],{},[15,7161],{"className":7162},[103,1410],[15,7164],{"className":7165,"style":43},[42],[15,7167,48],{"className":7168},[47],[15,7170],{"className":7171,"style":43},[42],[15,7173,7175,7178,7181,7184,7187],{"className":7174},[27],[15,7176],{"className":7177,"style":1991},[31],[15,7179,497],{"className":7180},[36],[15,7182],{"className":7183,"style":81},[42],[15,7185,1482],{"className":7186},[85],[15,7188],{"className":7189,"style":81},[42],[15,7191,7193,7196,7462,7465,7468,7471],{"className":7192},[27],[15,7194],{"className":7195,"style":6781},[31],[15,7197,7199,7202,7459],{"className":7198},[36],[15,7200],{"className":7201},[67,1410],[15,7203,7205],{"className":7204},[1414],[15,7206,7208,7451],{"className":7207},[365,366],[15,7209,7211,7448],{"className":7210},[370],[15,7212,7214,7327,7335],{"className":7213,"style":6800},[374],[15,7215,7216,7219],{"style":1427},[15,7217],{"className":7218,"style":1431},[382],[15,7220,7222,7225,7254,7257,7260,7263,7298],{"className":7221},[36],[15,7223,68],{"className":7224},[67],[15,7226,7228,7231],{"className":7227},[36],[15,7229,1453],{"className":7230},[36,37],[15,7232,7234],{"className":7233},[361],[15,7235,7237],{"className":7236},[365],[15,7238,7240],{"className":7239},[370],[15,7241,7243],{"className":7242,"style":1810},[374],[15,7244,7245,7248],{"style":1469},[15,7246],{"className":7247,"style":383},[382],[15,7249,7251],{"className":7250},[387,388,389,390],[15,7252,230],{"className":7253,"style":229},[36,37,390],[15,7255],{"className":7256,"style":81},[42],[15,7258,86],{"className":7259},[85],[15,7261],{"className":7262,"style":81},[42],[15,7264,7266,7269],{"className":7265},[36],[15,7267,1453],{"className":7268},[36,37],[15,7270,7272],{"className":7271},[361],[15,7273,7275],{"className":7274},[365],[15,7276,7278],{"className":7277},[370],[15,7279,7281],{"className":7280,"style":1466},[374],[15,7282,7283,7286],{"style":1469},[15,7284],{"className":7285,"style":383},[382],[15,7287,7289],{"className":7288},[387,388,389,390],[15,7290,7292,7295],{"className":7291},[36,390],[15,7293,1482],{"className":7294},[36,390],[15,7296,230],{"className":7297,"style":229},[36,37,390],[15,7299,7301,7304],{"className":7300},[103],[15,7302,104],{"className":7303},[103],[15,7305,7307],{"className":7306},[361],[15,7308,7310],{"className":7309},[365],[15,7311,7313],{"className":7312},[370],[15,7314,7316],{"className":7315,"style":4375},[374],[15,7317,7318,7321],{"style":1469},[15,7319],{"className":7320,"style":383},[382],[15,7322,7324],{"className":7323},[387,388,389,390],[15,7325,394],{"className":7326},[36,390],[15,7328,7329,7332],{"style":1488},[15,7330],{"className":7331,"style":1431},[382],[15,7333],{"className":7334,"style":1496},[1495],[15,7336,7337,7340],{"style":1499},[15,7338],{"className":7339,"style":1431},[382],[15,7341,7343,7346,7375,7378,7381,7384,7419],{"className":7342},[36],[15,7344,68],{"className":7345},[67],[15,7347,7349,7352],{"className":7348},[36],[15,7350,1453],{"className":7351},[36,37],[15,7353,7355],{"className":7354},[361],[15,7356,7358],{"className":7357},[365],[15,7359,7361],{"className":7360},[370],[15,7362,7364],{"className":7363,"style":1903},[374],[15,7365,7366,7369],{"style":1202},[15,7367],{"className":7368,"style":383},[382],[15,7370,7372],{"className":7371},[387,388,389,390],[15,7373,230],{"className":7374,"style":229},[36,37,390],[15,7376],{"className":7377,"style":81},[42],[15,7379,1482],{"className":7380},[85],[15,7382],{"className":7383,"style":81},[42],[15,7385,7387,7390],{"className":7386},[36],[15,7388,1453],{"className":7389},[36,37],[15,7391,7393],{"className":7392},[361],[15,7394,7396],{"className":7395},[365],[15,7397,7399],{"className":7398},[370],[15,7400,7402],{"className":7401,"style":1945},[374],[15,7403,7404,7407],{"style":1202},[15,7405],{"className":7406,"style":383},[382],[15,7408,7410],{"className":7409},[387,388,389,390],[15,7411,7413,7416],{"className":7412},[36,390],[15,7414,1482],{"className":7415},[36,390],[15,7417,230],{"className":7418,"style":229},[36,37,390],[15,7420,7422,7425],{"className":7421},[103],[15,7423,104],{"className":7424},[103],[15,7426,7428],{"className":7427},[361],[15,7429,7431],{"className":7430},[365],[15,7432,7434],{"className":7433},[370],[15,7435,7437],{"className":7436,"style":4010},[374],[15,7438,7439,7442],{"style":1202},[15,7440],{"className":7441,"style":383},[382],[15,7443,7445],{"className":7444},[387,388,389,390],[15,7446,394],{"className":7447},[36,390],[15,7449,399],{"className":7450},[398],[15,7452,7454],{"className":7453},[370],[15,7455,7457],{"className":7456,"style":4447},[374],[15,7458],{},[15,7460],{"className":7461},[103,1410],[15,7463],{"className":7464,"style":43},[42],[15,7466,48],{"className":7467},[47],[15,7469],{"className":7470,"style":1598},[42],[15,7472],{"className":7473,"style":43},[42],[15,7475,7477,7480,7483,7486,7489],{"className":7476},[27],[15,7478],{"className":7479,"style":1991},[31],[15,7481,497],{"className":7482},[36],[15,7484],{"className":7485,"style":81},[42],[15,7487,1482],{"className":7488},[85],[15,7490],{"className":7491,"style":81},[42],[15,7493,7495,7498,7527,7530,7533,7536],{"className":7494},[27],[15,7496],{"className":7497,"style":2010},[31],[15,7499,7501,7504],{"className":7500},[2014],[15,7502,1749],{"className":7503},[2014],[15,7505,7507],{"className":7506},[361],[15,7508,7510],{"className":7509},[365],[15,7511,7513],{"className":7512},[370],[15,7514,7516],{"className":7515,"style":2010},[374],[15,7517,7518,7521],{"style":2032},[15,7519],{"className":7520,"style":383},[382],[15,7522,7524],{"className":7523},[387,388,389,390],[15,7525,394],{"className":7526},[36,390],[15,7528],{"className":7529,"style":1598},[42],[15,7531,230],{"className":7532,"style":229},[36,37],[15,7534],{"className":7535,"style":1598},[42],[15,7537,4730],{"className":7538},[36],[11,7540,7541,7542,7608,7609,7624,7625,7658,7659,7739],{},"Since ",[15,7543,7545],{"className":7544},[18],[15,7546,7548,7566,7599],{"className":7547,"ariaHidden":23},[22],[15,7549,7551,7554,7557,7560,7563],{"className":7550},[27],[15,7552],{"className":7553,"style":32},[31],[15,7555,1749],{"className":7556},[2014],[15,7558],{"className":7559,"style":43},[42],[15,7561,48],{"className":7562},[47],[15,7564],{"className":7565,"style":43},[42],[15,7567,7569,7572,7575,7578,7581,7584,7587,7590,7593,7596],{"className":7568},[27],[15,7570],{"className":7571,"style":58},[31],[15,7573,394],{"className":7574},[36],[15,7576,1541],{"className":7577,"style":62},[36,37],[15,7579,68],{"className":7580},[67],[15,7582,394],{"className":7583},[36],[15,7585,230],{"className":7586,"style":229},[36,37],[15,7588,104],{"className":7589},[103],[15,7591],{"className":7592,"style":81},[42],[15,7594,1482],{"className":7595},[85],[15,7597],{"className":7598,"style":81},[42],[15,7600,7602,7605],{"className":7601},[27],[15,7603],{"className":7604,"style":2147},[31],[15,7606,497],{"className":7607},[36],", tanh is a rescaled, recentered sigmoid; the\nzero-centering lifts its peak derivative to ",[15,7610,7612],{"className":7611},[18],[15,7613,7615],{"className":7614,"ariaHidden":23},[22],[15,7616,7618,7621],{"className":7617},[27],[15,7619],{"className":7620,"style":2147},[31],[15,7622,497],{"className":7623},[36]," (at ",[15,7626,7628],{"className":7627},[18],[15,7629,7631,7649],{"className":7630,"ariaHidden":23},[22],[15,7632,7634,7637,7640,7643,7646],{"className":7633},[27],[15,7635],{"className":7636,"style":225},[31],[15,7638,230],{"className":7639,"style":229},[36,37],[15,7641],{"className":7642,"style":43},[42],[15,7644,48],{"className":7645},[47],[15,7647],{"className":7648,"style":43},[42],[15,7650,7652,7655],{"className":7651},[27],[15,7653],{"className":7654,"style":2147},[31],[15,7656,1589],{"className":7657},[36],") versus the sigmoid's\n",[15,7660,7662],{"className":7661},[18],[15,7663,7665],{"className":7664,"ariaHidden":23},[22],[15,7666,7668,7671],{"className":7667},[27],[15,7669],{"className":7670,"style":1670},[31],[15,7672,7674,7677,7736],{"className":7673},[36],[15,7675],{"className":7676},[67,1410],[15,7678,7680],{"className":7679},[1414],[15,7681,7683,7728],{"className":7682},[365,366],[15,7684,7686,7725],{"className":7685},[370],[15,7687,7689,7703,7711],{"className":7688,"style":1689},[374],[15,7690,7691,7694],{"style":1692},[15,7692],{"className":7693,"style":1431},[382],[15,7695,7697],{"className":7696},[387,388,389,390],[15,7698,7700],{"className":7699},[36,390],[15,7701,1705],{"className":7702},[36,390],[15,7704,7705,7708],{"style":1488},[15,7706],{"className":7707,"style":1431},[382],[15,7709],{"className":7710,"style":1496},[1495],[15,7712,7713,7716],{"style":1716},[15,7714],{"className":7715,"style":1431},[382],[15,7717,7719],{"className":7718},[387,388,389,390],[15,7720,7722],{"className":7721},[36,390],[15,7723,497],{"className":7724},[36,390],[15,7726,399],{"className":7727},[398],[15,7729,7731],{"className":7730},[370],[15,7732,7734],{"className":7733,"style":1738},[374],[15,7735],{},[15,7737],{"className":7738},[103,1410],", and makes tanh the better of the two saturating choices.",[3890,7741,2155],{"id":7742},"relu",[11,7744,7745,7746,7779],{},"ReLU is differentiable everywhere except the kink at ",[15,7747,7749],{"className":7748},[18],[15,7750,7752,7770],{"className":7751,"ariaHidden":23},[22],[15,7753,7755,7758,7761,7764,7767],{"className":7754},[27],[15,7756],{"className":7757,"style":225},[31],[15,7759,230],{"className":7760,"style":229},[36,37],[15,7762],{"className":7763,"style":43},[42],[15,7765,48],{"className":7766},[47],[15,7768],{"className":7769,"style":43},[42],[15,7771,7773,7776],{"className":7772},[27],[15,7774],{"className":7775,"style":2147},[31],[15,7777,1589],{"className":7778},[36],", and its derivative is\njust the indicator of the active region:",[15,7781,7783],{"className":7782},[4032],[15,7784,7786],{"className":7785},[18],[15,7787,7789,7900,8063,8087],{"className":7788,"ariaHidden":23},[22],[15,7790,7792,7796,7864,7867,7873,7876,7879,7882,7885,7888,7891,7894,7897],{"className":7791},[27],[15,7793],{"className":7794,"style":7795},[31],"height:2.0574em;vertical-align:-0.686em;",[15,7797,7799,7802,7861],{"className":7798},[36],[15,7800],{"className":7801},[67,1410],[15,7803,7805],{"className":7804},[1414],[15,7806,7808,7852],{"className":7807},[365,366],[15,7809,7811,7849],{"className":7810},[370],[15,7812,7815,7830,7838],{"className":7813,"style":7814},[374],"height:1.3714em;",[15,7816,7817,7820],{"style":1427},[15,7818],{"className":7819,"style":1431},[382],[15,7821,7823,7827],{"className":7822},[36],[15,7824,7826],{"className":7825},[36,37],"d",[15,7828,230],{"className":7829,"style":229},[36,37],[15,7831,7832,7835],{"style":1488},[15,7833],{"className":7834,"style":1431},[382],[15,7836],{"className":7837,"style":1496},[1495],[15,7839,7840,7843],{"style":1499},[15,7841],{"className":7842,"style":1431},[382],[15,7844,7846],{"className":7845},[36],[15,7847,7826],{"className":7848},[36,37],[15,7850,399],{"className":7851},[398],[15,7853,7855],{"className":7854},[370],[15,7856,7859],{"className":7857,"style":7858},[374],"height:0.686em;",[15,7860],{},[15,7862],{"className":7863},[103,1410],[15,7865],{"className":7866,"style":1598},[42],[15,7868,7870],{"className":7869},[2014],[15,7871,2177],{"className":7872},[36,2176],[15,7874,68],{"className":7875},[67],[15,7877,1589],{"className":7878},[36],[15,7880,1594],{"className":7881},[1593],[15,7883],{"className":7884,"style":1598},[42],[15,7886,230],{"className":7887,"style":229},[36,37],[15,7889,104],{"className":7890},[103],[15,7892],{"className":7893,"style":43},[42],[15,7895,48],{"className":7896},[47],[15,7898],{"className":7899,"style":43},[42],[15,7901,7903,7907,8054,8057,8060],{"className":7902},[27],[15,7904],{"className":7905,"style":7906},[31],"height:3em;vertical-align:-1.25em;",[15,7908,7910,7918,8051],{"className":7909},[5230],[15,7911,7913],{"className":7912,"style":5235},[67,5234],[15,7914,7917],{"className":7915},[6458,7916],"size4","{",[15,7919,7921],{"className":7920},[36],[15,7922,7925,7976,7981],{"className":7923},[7924],"mtable",[15,7926,7929],{"className":7927},[7928],"col-align-l",[15,7930,7932,7967],{"className":7931},[365,366],[15,7933,7935,7964],{"className":7934},[370],[15,7936,7939,7952],{"className":7937,"style":7938},[374],"height:1.69em;",[15,7940,7942,7946],{"style":7941},"top:-3.69em;",[15,7943],{"className":7944,"style":7945},[382],"height:3.008em;",[15,7947,7949],{"className":7948},[36],[15,7950,497],{"className":7951},[36],[15,7953,7955,7958],{"style":7954},"top:-2.25em;",[15,7956],{"className":7957,"style":7945},[382],[15,7959,7961],{"className":7960},[36],[15,7962,1589],{"className":7963},[36],[15,7965,399],{"className":7966},[398],[15,7968,7970],{"className":7969},[370],[15,7971,7974],{"className":7972,"style":7973},[374],"height:1.19em;",[15,7975],{},[15,7977],{"className":7978,"style":7980},[7979],"arraycolsep","width:1em;",[15,7982,7984],{"className":7983},[7928],[15,7985,7987,8043],{"className":7986},[365,366],[15,7988,7990,8040],{"className":7989},[370],[15,7991,7993,8016],{"className":7992,"style":7938},[374],[15,7994,7995,7998],{"style":7941},[15,7996],{"className":7997,"style":7945},[382],[15,7999,8001,8004,8007,8010,8013],{"className":8000},[36],[15,8002,230],{"className":8003,"style":229},[36,37],[15,8005],{"className":8006,"style":43},[42],[15,8008,2227],{"className":8009},[47],[15,8011],{"className":8012,"style":43},[42],[15,8014,1589],{"className":8015},[36],[15,8017,8018,8021],{"style":7954},[15,8019],{"className":8020,"style":7945},[382],[15,8022,8024,8027,8030,8034,8037],{"className":8023},[36],[15,8025,230],{"className":8026,"style":229},[36,37],[15,8028],{"className":8029,"style":43},[42],[15,8031,8033],{"className":8032},[47],"\u003C",[15,8035],{"className":8036,"style":43},[42],[15,8038,1589],{"className":8039},[36],[15,8041,399],{"className":8042},[398],[15,8044,8046],{"className":8045},[370],[15,8047,8049],{"className":8048,"style":7973},[374],[15,8050],{},[15,8052],{"className":8053},[103,1410],[15,8055],{"className":8056,"style":43},[42],[15,8058,48],{"className":8059},[47],[15,8061],{"className":8062,"style":43},[42],[15,8064,8066,8069,8072,8075,8078,8081,8084],{"className":8065},[27],[15,8067],{"className":8068,"style":58},[31],[15,8070,497],{"className":8071},[36,2213],[15,8073,2217],{"className":8074},[67],[15,8076,230],{"className":8077,"style":229},[36,37],[15,8079],{"className":8080,"style":43},[42],[15,8082,2227],{"className":8083},[47],[15,8085],{"className":8086,"style":43},[42],[15,8088,8090,8093,8096,8099],{"className":8089},[27],[15,8091],{"className":8092,"style":58},[31],[15,8094,1589],{"className":8095},[36],[15,8097,2243],{"className":8098},[103],[15,8100,1594],{"className":8101},[1593],[11,8103,8104,8105,8138,8139,8154,8155,8170,8171,8186,8187,8170,8202,8217],{},"with the value at ",[15,8106,8108],{"className":8107},[18],[15,8109,8111,8129],{"className":8110,"ariaHidden":23},[22],[15,8112,8114,8117,8120,8123,8126],{"className":8113},[27],[15,8115],{"className":8116,"style":225},[31],[15,8118,230],{"className":8119,"style":229},[36,37],[15,8121],{"className":8122,"style":43},[42],[15,8124,48],{"className":8125},[47],[15,8127],{"className":8128,"style":43},[42],[15,8130,8132,8135],{"className":8131},[27],[15,8133],{"className":8134,"style":2147},[31],[15,8136,1589],{"className":8137},[36]," taken by convention to be ",[15,8140,8142],{"className":8141},[18],[15,8143,8145],{"className":8144,"ariaHidden":23},[22],[15,8146,8148,8151],{"className":8147},[27],[15,8149],{"className":8150,"style":2147},[31],[15,8152,1589],{"className":8153},[36]," (frameworks pick ",[15,8156,8158],{"className":8157},[18],[15,8159,8161],{"className":8160,"ariaHidden":23},[22],[15,8162,8164,8167],{"className":8163},[27],[15,8165],{"className":8166,"style":2147},[31],[15,8168,1589],{"className":8169},[36]," or ",[15,8172,8174],{"className":8173},[18],[15,8175,8177],{"className":8176,"ariaHidden":23},[22],[15,8178,8180,8183],{"className":8179},[27],[15,8181],{"className":8182,"style":2147},[31],[15,8184,497],{"className":8185},[36],";\nthe measure-zero choice is irrelevant in practice). This step-function derivative,\na hard gate of ",[15,8188,8190],{"className":8189},[18],[15,8191,8193],{"className":8192,"ariaHidden":23},[22],[15,8194,8196,8199],{"className":8195},[27],[15,8197],{"className":8198,"style":2147},[31],[15,8200,497],{"className":8201},[36],[15,8203,8205],{"className":8204},[18],[15,8206,8208],{"className":8207,"ariaHidden":23},[22],[15,8209,8211,8214],{"className":8210},[27],[15,8212],{"className":8213,"style":2147},[31],[15,8215,1589],{"className":8216},[36],", is the source of both ReLU's strength and its dead-unit\npathology.",[3890,8219,8221],{"id":8220},"leaky-relu-and-prelu","Leaky ReLU and PReLU",[11,8223,8224,8225,1594],{},"Leaky ReLU replaces the flat left branch with a shallow line of slope ",[15,8226,8228],{"className":8227},[18],[15,8229,8231],{"className":8230,"ariaHidden":23},[22],[15,8232,8234,8237],{"className":8233},[27],[15,8235],{"className":8236,"style":225},[31],[15,8238,2315],{"className":8239,"style":2314},[36,37],[15,8241,8243],{"className":8242},[4032],[15,8244,8246],{"className":8245},[18],[15,8247,8249,8276,8318,8523],{"className":8248,"ariaHidden":23},[22],[15,8250,8252,8255,8258,8261,8264,8267,8270,8273],{"className":8251},[27],[15,8253],{"className":8254,"style":58},[31],[15,8256,63],{"className":8257,"style":62},[36,37],[15,8259,68],{"className":8260},[67],[15,8262,230],{"className":8263,"style":229},[36,37],[15,8265,104],{"className":8266},[103],[15,8268],{"className":8269,"style":43},[42],[15,8271,48],{"className":8272},[47],[15,8274],{"className":8275,"style":43},[42],[15,8277,8279,8282,8288,8291,8294,8297,8300,8303,8306,8309,8312,8315],{"className":8278},[27],[15,8280],{"className":8281,"style":58},[31],[15,8283,8285],{"className":8284},[2014],[15,8286,2177],{"className":8287},[36,2176],[15,8289,68],{"className":8290},[67],[15,8292,2315],{"className":8293,"style":2314},[36,37],[15,8295,230],{"className":8296,"style":229},[36,37],[15,8298,1594],{"className":8299},[1593],[15,8301],{"className":8302,"style":1598},[42],[15,8304,230],{"className":8305,"style":229},[36,37],[15,8307,104],{"className":8308},[103],[15,8310],{"className":8311,"style":43},[42],[15,8313,48],{"className":8314},[47],[15,8316],{"className":8317,"style":43},[42],[15,8319,8321,8324,8466,8470,8473,8505,8508,8511,8514,8517,8520],{"className":8320},[27],[15,8322],{"className":8323,"style":7906},[31],[15,8325,8327,8333,8463],{"className":8326},[5230],[15,8328,8330],{"className":8329,"style":5235},[67,5234],[15,8331,7917],{"className":8332},[6458,7916],[15,8334,8336],{"className":8335},[36],[15,8337,8339,8387,8390],{"className":8338},[7924],[15,8340,8342],{"className":8341},[7928],[15,8343,8345,8379],{"className":8344},[365,366],[15,8346,8348,8376],{"className":8347},[370],[15,8349,8351,8362],{"className":8350,"style":7938},[374],[15,8352,8353,8356],{"style":7941},[15,8354],{"className":8355,"style":7945},[382],[15,8357,8359],{"className":8358},[36],[15,8360,230],{"className":8361,"style":229},[36,37],[15,8363,8364,8367],{"style":7954},[15,8365],{"className":8366,"style":7945},[382],[15,8368,8370,8373],{"className":8369},[36],[15,8371,2315],{"className":8372,"style":2314},[36,37],[15,8374,230],{"className":8375,"style":229},[36,37],[15,8377,399],{"className":8378},[398],[15,8380,8382],{"className":8381},[370],[15,8383,8385],{"className":8384,"style":7973},[374],[15,8386],{},[15,8388],{"className":8389,"style":7980},[7979],[15,8391,8393],{"className":8392},[7928],[15,8394,8396,8455],{"className":8395},[365,366],[15,8397,8399,8452],{"className":8398},[370],[15,8400,8402,8426],{"className":8401,"style":7938},[374],[15,8403,8404,8407],{"style":7941},[15,8405],{"className":8406,"style":7945},[382],[15,8408,8410,8413,8416,8420,8423],{"className":8409},[36],[15,8411,230],{"className":8412,"style":229},[36,37],[15,8414],{"className":8415,"style":43},[42],[15,8417,8419],{"className":8418},[47],"≥",[15,8421],{"className":8422,"style":43},[42],[15,8424,1589],{"className":8425},[36],[15,8427,8428,8431],{"style":7954},[15,8429],{"className":8430,"style":7945},[382],[15,8432,8434,8437,8440,8443,8446,8449],{"className":8433},[36],[15,8435,230],{"className":8436,"style":229},[36,37],[15,8438],{"className":8439,"style":43},[42],[15,8441,8033],{"className":8442},[47],[15,8444],{"className":8445,"style":43},[42],[15,8447,1589],{"className":8448},[36],[15,8450,1594],{"className":8451},[1593],[15,8453,399],{"className":8454},[398],[15,8456,8458],{"className":8457},[370],[15,8459,8461],{"className":8460,"style":7973},[374],[15,8462],{},[15,8464],{"className":8465},[103,1410],[15,8467],{"className":8468,"style":8469},[42],"margin-right:2em;",[15,8471],{"className":8472,"style":1598},[42],[15,8474,8476,8479],{"className":8475},[36],[15,8477,63],{"className":8478,"style":62},[36,37],[15,8480,8482],{"className":8481},[361],[15,8483,8485],{"className":8484},[365],[15,8486,8488],{"className":8487},[370],[15,8489,8491],{"className":8490,"style":4064},[374],[15,8492,8493,8496],{"style":4067},[15,8494],{"className":8495,"style":383},[382],[15,8497,8499],{"className":8498},[387,388,389,390],[15,8500,8502],{"className":8501},[36,390],[15,8503,1215],{"className":8504},[36,390],[15,8506,68],{"className":8507},[67],[15,8509,230],{"className":8510,"style":229},[36,37],[15,8512,104],{"className":8513},[103],[15,8515],{"className":8516,"style":43},[42],[15,8518,48],{"className":8519},[47],[15,8521],{"className":8522,"style":43},[42],[15,8524,8526,8529],{"className":8525},[27],[15,8527],{"className":8528,"style":7906},[31],[15,8530,8532,8538,8662],{"className":8531},[5230],[15,8533,8535],{"className":8534,"style":5235},[67,5234],[15,8536,7917],{"className":8537},[6458,7916],[15,8539,8541],{"className":8540},[36],[15,8542,8544,8589,8592],{"className":8543},[7924],[15,8545,8547],{"className":8546},[7928],[15,8548,8550,8581],{"className":8549},[365,366],[15,8551,8553,8578],{"className":8552},[370],[15,8554,8556,8567],{"className":8555,"style":7938},[374],[15,8557,8558,8561],{"style":7941},[15,8559],{"className":8560,"style":7945},[382],[15,8562,8564],{"className":8563},[36],[15,8565,497],{"className":8566},[36],[15,8568,8569,8572],{"style":7954},[15,8570],{"className":8571,"style":7945},[382],[15,8573,8575],{"className":8574},[36],[15,8576,2315],{"className":8577,"style":2314},[36,37],[15,8579,399],{"className":8580},[398],[15,8582,8584],{"className":8583},[370],[15,8585,8587],{"className":8586,"style":7973},[374],[15,8588],{},[15,8590],{"className":8591,"style":7980},[7979],[15,8593,8595],{"className":8594},[7928],[15,8596,8598,8654],{"className":8597},[365,366],[15,8599,8601,8651],{"className":8600},[370],[15,8602,8604,8627],{"className":8603,"style":7938},[374],[15,8605,8606,8609],{"style":7941},[15,8607],{"className":8608,"style":7945},[382],[15,8610,8612,8615,8618,8621,8624],{"className":8611},[36],[15,8613,230],{"className":8614,"style":229},[36,37],[15,8616],{"className":8617,"style":43},[42],[15,8619,2227],{"className":8620},[47],[15,8622],{"className":8623,"style":43},[42],[15,8625,1589],{"className":8626},[36],[15,8628,8629,8632],{"style":7954},[15,8630],{"className":8631,"style":7945},[382],[15,8633,8635,8638,8641,8644,8647],{"className":8634},[36],[15,8636,230],{"className":8637,"style":229},[36,37],[15,8639],{"className":8640,"style":43},[42],[15,8642,8033],{"className":8643},[47],[15,8645],{"className":8646,"style":43},[42],[15,8648,8650],{"className":8649},[36],"0.",[15,8652,399],{"className":8653},[398],[15,8655,8657],{"className":8656},[370],[15,8658,8660],{"className":8659,"style":7973},[374],[15,8661],{},[15,8663],{"className":8664},[103,1410],[11,8666,8667,8668,8701,8702,8717,8718,8733,8734,8749,8750,4730],{},"The derivative is now ",[15,8669,8671],{"className":8670},[18],[15,8672,8674,8692],{"className":8673,"ariaHidden":23},[22],[15,8675,8677,8680,8683,8686,8689],{"className":8676},[27],[15,8678],{"className":8679,"style":2665},[31],[15,8681,2315],{"className":8682,"style":2314},[36,37],[15,8684],{"className":8685,"style":43},[42],[15,8687,2227],{"className":8688},[47],[15,8690],{"className":8691,"style":43},[42],[15,8693,8695,8698],{"className":8694},[27],[15,8696],{"className":8697,"style":2147},[31],[15,8699,1589],{"className":8700},[36]," on the entire negative half-line rather than\n",[15,8703,8705],{"className":8704},[18],[15,8706,8708],{"className":8707,"ariaHidden":23},[22],[15,8709,8711,8714],{"className":8710},[27],[15,8712],{"className":8713,"style":2147},[31],[15,8715,1589],{"className":8716},[36],", so a unit sitting in the negative region still receives an ",[15,8719,8721],{"className":8720},[18],[15,8722,8724],{"className":8723,"ariaHidden":23},[22],[15,8725,8727,8730],{"className":8726},[27],[15,8728],{"className":8729,"style":225},[31],[15,8731,2315],{"className":8732,"style":2314},[36,37],"-scaled\ngradient. PReLU is identical except ",[15,8735,8737],{"className":8736},[18],[15,8738,8740],{"className":8739,"ariaHidden":23},[22],[15,8741,8743,8746],{"className":8742},[27],[15,8744],{"className":8745,"style":225},[31],[15,8747,2315],{"className":8748,"style":2314},[36,37]," is a learned parameter (one per\nchannel), so the negative slope adapts during training rather than being fixed at\n",[15,8751,8753],{"className":8752},[18],[15,8754,8756],{"className":8755,"ariaHidden":23},[22],[15,8757,8759,8762],{"className":8758},[27],[15,8760],{"className":8761,"style":2147},[31],[15,8763,2508],{"className":8764},[36],[3890,8766,2634],{"id":8767},"elu",[11,8769,8770,8771,8804,8805,8824],{},"ELU keeps the identity for ",[15,8772,8774],{"className":8773},[18],[15,8775,8777,8795],{"className":8776,"ariaHidden":23},[22],[15,8778,8780,8783,8786,8789,8792],{"className":8779},[27],[15,8781],{"className":8782,"style":2665},[31],[15,8784,230],{"className":8785,"style":229},[36,37],[15,8787],{"className":8788,"style":43},[42],[15,8790,2227],{"className":8791},[47],[15,8793],{"className":8794,"style":43},[42],[15,8796,8798,8801],{"className":8797},[27],[15,8799],{"className":8800,"style":2147},[31],[15,8802,1589],{"className":8803},[36]," but replaces the kink with a smooth exponential\nthat saturates gently toward ",[15,8806,8808],{"className":8807},[18],[15,8809,8811],{"className":8810,"ariaHidden":23},[22],[15,8812,8814,8818,8821],{"className":8813},[27],[15,8815],{"className":8816,"style":8817},[31],"height:0.6667em;vertical-align:-0.0833em;",[15,8819,1482],{"className":8820},[36],[15,8822,2315],{"className":8823,"style":2314},[36,37]," on the left:",[15,8826,8828],{"className":8827},[4032],[15,8829,8831],{"className":8830},[18],[15,8832,8834,8861,9114],{"className":8833,"ariaHidden":23},[22],[15,8835,8837,8840,8843,8846,8849,8852,8855,8858],{"className":8836},[27],[15,8838],{"className":8839,"style":58},[31],[15,8841,63],{"className":8842,"style":62},[36,37],[15,8844,68],{"className":8845},[67],[15,8847,230],{"className":8848,"style":229},[36,37],[15,8850,104],{"className":8851},[103],[15,8853],{"className":8854,"style":43},[42],[15,8856,48],{"className":8857},[47],[15,8859],{"className":8860,"style":43},[42],[15,8862,8864,8867,9058,9061,9064,9096,9099,9102,9105,9108,9111],{"className":8863},[27],[15,8865],{"className":8866,"style":7906},[31],[15,8868,8870,8876,9055],{"className":8869},[5230],[15,8871,8873],{"className":8872,"style":5235},[67,5234],[15,8874,7917],{"className":8875},[6458,7916],[15,8877,8879],{"className":8878},[36],[15,8880,8882,8980,8983],{"className":8881},[7924],[15,8883,8885],{"className":8884},[7928],[15,8886,8888,8972],{"className":8887},[365,366],[15,8889,8891,8969],{"className":8890},[370],[15,8892,8894,8905],{"className":8893,"style":7938},[374],[15,8895,8896,8899],{"style":7941},[15,8897],{"className":8898,"style":7945},[382],[15,8900,8902],{"className":8901},[36],[15,8903,230],{"className":8904,"style":229},[36,37],[15,8906,8907,8910],{"style":7954},[15,8908],{"className":8909,"style":7945},[382],[15,8911,8913,8916,8919,8922,8954,8957,8960,8963,8966],{"className":8912},[36],[15,8914,2315],{"className":8915,"style":2314},[36,37],[15,8917],{"className":8918,"style":1598},[42],[15,8920,68],{"className":8921},[67],[15,8923,8925,8928],{"className":8924},[36],[15,8926,1453],{"className":8927},[36,37],[15,8929,8931],{"className":8930},[361],[15,8932,8934],{"className":8933},[365],[15,8935,8937],{"className":8936},[370],[15,8938,8940],{"className":8939,"style":1903},[374],[15,8941,8942,8945],{"style":1202},[15,8943],{"className":8944,"style":383},[382],[15,8946,8948],{"className":8947},[387,388,389,390],[15,8949,8951],{"className":8950},[36,390],[15,8952,230],{"className":8953,"style":229},[36,37,390],[15,8955],{"className":8956,"style":81},[42],[15,8958,1482],{"className":8959},[85],[15,8961],{"className":8962,"style":81},[42],[15,8964,497],{"className":8965},[36],[15,8967,104],{"className":8968},[103],[15,8970,399],{"className":8971},[398],[15,8973,8975],{"className":8974},[370],[15,8976,8978],{"className":8977,"style":7973},[374],[15,8979],{},[15,8981],{"className":8982,"style":7980},[7979],[15,8984,8986],{"className":8985},[7928],[15,8987,8989,9047],{"className":8988},[365,366],[15,8990,8992,9044],{"className":8991},[370],[15,8993,8995,9018],{"className":8994,"style":7938},[374],[15,8996,8997,9000],{"style":7941},[15,8998],{"className":8999,"style":7945},[382],[15,9001,9003,9006,9009,9012,9015],{"className":9002},[36],[15,9004,230],{"className":9005,"style":229},[36,37],[15,9007],{"className":9008,"style":43},[42],[15,9010,2227],{"className":9011},[47],[15,9013],{"className":9014,"style":43},[42],[15,9016,1589],{"className":9017},[36],[15,9019,9020,9023],{"style":7954},[15,9021],{"className":9022,"style":7945},[382],[15,9024,9026,9029,9032,9035,9038,9041],{"className":9025},[36],[15,9027,230],{"className":9028,"style":229},[36,37],[15,9030],{"className":9031,"style":43},[42],[15,9033,1660],{"className":9034},[47],[15,9036],{"className":9037,"style":43},[42],[15,9039,1589],{"className":9040},[36],[15,9042,1594],{"className":9043},[1593],[15,9045,399],{"className":9046},[398],[15,9048,9050],{"className":9049},[370],[15,9051,9053],{"className":9052,"style":7973},[374],[15,9054],{},[15,9056],{"className":9057},[103,1410],[15,9059],{"className":9060,"style":8469},[42],[15,9062],{"className":9063,"style":1598},[42],[15,9065,9067,9070],{"className":9066},[36],[15,9068,63],{"className":9069,"style":62},[36,37],[15,9071,9073],{"className":9072},[361],[15,9074,9076],{"className":9075},[365],[15,9077,9079],{"className":9078},[370],[15,9080,9082],{"className":9081,"style":4064},[374],[15,9083,9084,9087],{"style":4067},[15,9085],{"className":9086,"style":383},[382],[15,9088,9090],{"className":9089},[387,388,389,390],[15,9091,9093],{"className":9092},[36,390],[15,9094,1215],{"className":9095},[36,390],[15,9097,68],{"className":9098},[67],[15,9100,230],{"className":9101,"style":229},[36,37],[15,9103,104],{"className":9104},[103],[15,9106],{"className":9107,"style":43},[42],[15,9109,48],{"className":9110},[47],[15,9112],{"className":9113,"style":43},[42],[15,9115,9117,9120],{"className":9116},[27],[15,9118],{"className":9119,"style":7906},[31],[15,9121,9123,9129,9287],{"className":9122},[5230],[15,9124,9126],{"className":9125,"style":5235},[67,5234],[15,9127,7917],{"className":9128},[6458,7916],[15,9130,9132],{"className":9131},[36],[15,9133,9135,9215,9218],{"className":9134},[7924],[15,9136,9138],{"className":9137},[7928],[15,9139,9141,9207],{"className":9140},[365,366],[15,9142,9144,9204],{"className":9143},[370],[15,9145,9147,9158],{"className":9146,"style":7938},[374],[15,9148,9149,9152],{"style":7941},[15,9150],{"className":9151,"style":7945},[382],[15,9153,9155],{"className":9154},[36],[15,9156,497],{"className":9157},[36],[15,9159,9160,9163],{"style":7954},[15,9161],{"className":9162,"style":7945},[382],[15,9164,9166,9169,9172],{"className":9165},[36],[15,9167,2315],{"className":9168,"style":2314},[36,37],[15,9170],{"className":9171,"style":1598},[42],[15,9173,9175,9178],{"className":9174},[36],[15,9176,1453],{"className":9177},[36,37],[15,9179,9181],{"className":9180},[361],[15,9182,9184],{"className":9183},[365],[15,9185,9187],{"className":9186},[370],[15,9188,9190],{"className":9189,"style":1903},[374],[15,9191,9192,9195],{"style":1202},[15,9193],{"className":9194,"style":383},[382],[15,9196,9198],{"className":9197},[387,388,389,390],[15,9199,9201],{"className":9200},[36,390],[15,9202,230],{"className":9203,"style":229},[36,37,390],[15,9205,399],{"className":9206},[398],[15,9208,9210],{"className":9209},[370],[15,9211,9213],{"className":9212,"style":7973},[374],[15,9214],{},[15,9216],{"className":9217,"style":7980},[7979],[15,9219,9221],{"className":9220},[7928],[15,9222,9224,9279],{"className":9223},[365,366],[15,9225,9227,9276],{"className":9226},[370],[15,9228,9230,9253],{"className":9229,"style":7938},[374],[15,9231,9232,9235],{"style":7941},[15,9233],{"className":9234,"style":7945},[382],[15,9236,9238,9241,9244,9247,9250],{"className":9237},[36],[15,9239,230],{"className":9240,"style":229},[36,37],[15,9242],{"className":9243,"style":43},[42],[15,9245,2227],{"className":9246},[47],[15,9248],{"className":9249,"style":43},[42],[15,9251,1589],{"className":9252},[36],[15,9254,9255,9258],{"style":7954},[15,9256],{"className":9257,"style":7945},[382],[15,9259,9261,9264,9267,9270,9273],{"className":9260},[36],[15,9262,230],{"className":9263,"style":229},[36,37],[15,9265],{"className":9266,"style":43},[42],[15,9268,1660],{"className":9269},[47],[15,9271],{"className":9272,"style":43},[42],[15,9274,8650],{"className":9275},[36],[15,9277,399],{"className":9278},[398],[15,9280,9282],{"className":9281},[370],[15,9283,9285],{"className":9284,"style":7973},[374],[15,9286],{},[15,9288],{"className":9289},[103,1410],[11,9291,9292,9293,9340,9341,9374,9375,9437,9438,9503,9504,9537,9538,9556,9557,9572],{},"The left derivative ",[15,9294,9296],{"className":9295},[18],[15,9297,9299],{"className":9298,"ariaHidden":23},[22],[15,9300,9302,9305,9308],{"className":9301},[27],[15,9303],{"className":9304,"style":1903},[31],[15,9306,2315],{"className":9307,"style":2314},[36,37],[15,9309,9311,9314],{"className":9310},[36],[15,9312,1453],{"className":9313},[36,37],[15,9315,9317],{"className":9316},[361],[15,9318,9320],{"className":9319},[365],[15,9321,9323],{"className":9322},[370],[15,9324,9326],{"className":9325,"style":1903},[374],[15,9327,9328,9331],{"style":1202},[15,9329],{"className":9330,"style":383},[382],[15,9332,9334],{"className":9333},[387,388,389,390],[15,9335,9337],{"className":9336},[36,390],[15,9338,230],{"className":9339,"style":229},[36,37,390]," is strictly positive and continuous, and at\n",[15,9342,9344],{"className":9343},[18],[15,9345,9347,9365],{"className":9346,"ariaHidden":23},[22],[15,9348,9350,9353,9356,9359,9362],{"className":9349},[27],[15,9351],{"className":9352,"style":225},[31],[15,9354,230],{"className":9355,"style":229},[36,37],[15,9357],{"className":9358,"style":43},[42],[15,9360,48],{"className":9361},[47],[15,9363],{"className":9364,"style":43},[42],[15,9366,9368,9371],{"className":9367},[27],[15,9369],{"className":9370,"style":2147},[31],[15,9372,1589],{"className":9373},[36]," both branches agree (",[15,9376,9378],{"className":9377},[18],[15,9379,9381,9428],{"className":9380,"ariaHidden":23},[22],[15,9382,9384,9387,9419,9422,9425],{"className":9383},[27],[15,9385],{"className":9386,"style":1243},[31],[15,9388,9390,9393],{"className":9389},[36],[15,9391,63],{"className":9392,"style":62},[36,37],[15,9394,9396],{"className":9395},[361],[15,9397,9399],{"className":9398},[365],[15,9400,9402],{"className":9401},[370],[15,9403,9405],{"className":9404,"style":1199},[374],[15,9406,9407,9410],{"style":1202},[15,9408],{"className":9409,"style":383},[382],[15,9411,9413],{"className":9412},[387,388,389,390],[15,9414,9416],{"className":9415},[36,390],[15,9417,1215],{"className":9418},[36,390],[15,9420],{"className":9421,"style":43},[42],[15,9423,48],{"className":9424},[47],[15,9426],{"className":9427,"style":43},[42],[15,9429,9431,9434],{"className":9430},[27],[15,9432],{"className":9433,"style":2147},[31],[15,9435,497],{"className":9436},[36]," from the right, ",[15,9439,9441],{"className":9440},[18],[15,9442,9444,9494],{"className":9443,"ariaHidden":23},[22],[15,9445,9447,9450,9453,9485,9488,9491],{"className":9446},[27],[15,9448],{"className":9449,"style":4010},[31],[15,9451,2315],{"className":9452,"style":2314},[36,37],[15,9454,9456,9459],{"className":9455},[36],[15,9457,1453],{"className":9458},[36,37],[15,9460,9462],{"className":9461},[361],[15,9463,9465],{"className":9464},[365],[15,9466,9468],{"className":9467},[370],[15,9469,9471],{"className":9470,"style":4010},[374],[15,9472,9473,9476],{"style":1202},[15,9474],{"className":9475,"style":383},[382],[15,9477,9479],{"className":9478},[387,388,389,390],[15,9480,9482],{"className":9481},[36,390],[15,9483,1589],{"className":9484},[36,390],[15,9486],{"className":9487,"style":43},[42],[15,9489,48],{"className":9490},[47],[15,9492],{"className":9493,"style":43},[42],[15,9495,9497,9500],{"className":9496},[27],[15,9498],{"className":9499,"style":225},[31],[15,9501,2315],{"className":9502,"style":2314},[36,37]," from the\nleft; with the common choice ",[15,9505,9507],{"className":9506},[18],[15,9508,9510,9528],{"className":9509,"ariaHidden":23},[22],[15,9511,9513,9516,9519,9522,9525],{"className":9512},[27],[15,9514],{"className":9515,"style":225},[31],[15,9517,2315],{"className":9518,"style":2314},[36,37],[15,9520],{"className":9521,"style":43},[42],[15,9523,48],{"className":9524},[47],[15,9526],{"className":9527,"style":43},[42],[15,9529,9531,9534],{"className":9530},[27],[15,9532],{"className":9533,"style":2147},[31],[15,9535,497],{"className":9536},[36]," the derivative is continuous too). Because\nthe negative branch bottoms out at ",[15,9539,9541],{"className":9540},[18],[15,9542,9544],{"className":9543,"ariaHidden":23},[22],[15,9545,9547,9550,9553],{"className":9546},[27],[15,9548],{"className":9549,"style":8817},[31],[15,9551,1482],{"className":9552},[36],[15,9554,2315],{"className":9555,"style":2314},[36,37]," instead of ",[15,9558,9560],{"className":9559},[18],[15,9561,9563],{"className":9562,"ariaHidden":23},[22],[15,9564,9566,9569],{"className":9565},[27],[15,9567],{"className":9568,"style":2147},[31],[15,9570,1589],{"className":9571},[36],", ELU can output\nnegative values, which pulls each layer's mean activation toward zero and keeps\nthe next layer's pre-activations nearer the high-gradient region.",[3890,9574,3083],{"id":9575},"softplus",[11,9577,9578],{},"Softplus is a smooth ReLU; its derivative is the sigmoid, which is why it appears\nwhenever a strictly positive, differentiable output is wanted:",[15,9580,9582],{"className":9581},[4032],[15,9583,9585],{"className":9584},[18],[15,9586,9588,9615,9642,9742,9891,10012],{"className":9587,"ariaHidden":23},[22],[15,9589,9591,9594,9597,9600,9603,9606,9609,9612],{"className":9590},[27],[15,9592],{"className":9593,"style":58},[31],[15,9595,63],{"className":9596,"style":62},[36,37],[15,9598,68],{"className":9599},[67],[15,9601,230],{"className":9602,"style":229},[36,37],[15,9604,104],{"className":9605},[103],[15,9607],{"className":9608,"style":43},[42],[15,9610,48],{"className":9611},[47],[15,9613],{"className":9614,"style":43},[42],[15,9616,9618,9621,9627,9630,9633,9636,9639],{"className":9617},[27],[15,9619],{"className":9620,"style":58},[31],[15,9622,9624],{"className":9623},[2014],[15,9625,3105],{"className":9626,"style":3104},[36,2176],[15,9628,68],{"className":9629},[67],[15,9631,497],{"className":9632},[36],[15,9634],{"className":9635,"style":81},[42],[15,9637,86],{"className":9638},[85],[15,9640],{"className":9641,"style":81},[42],[15,9643,9645,9648,9680,9683,9686,9689,9692,9724,9727,9730,9733,9736,9739],{"className":9644},[27],[15,9646],{"className":9647,"style":4045},[31],[15,9649,9651,9654],{"className":9650},[36],[15,9652,1453],{"className":9653},[36,37],[15,9655,9657],{"className":9656},[361],[15,9658,9660],{"className":9659},[365],[15,9661,9663],{"className":9662},[370],[15,9664,9666],{"className":9665,"style":6216},[374],[15,9667,9668,9671],{"style":4067},[15,9669],{"className":9670,"style":383},[382],[15,9672,9674],{"className":9673},[387,388,389,390],[15,9675,9677],{"className":9676},[36,390],[15,9678,230],{"className":9679,"style":229},[36,37,390],[15,9681,104],{"className":9682},[103],[15,9684,1594],{"className":9685},[1593],[15,9687],{"className":9688,"style":8469},[42],[15,9690],{"className":9691,"style":1598},[42],[15,9693,9695,9698],{"className":9694},[36],[15,9696,63],{"className":9697,"style":62},[36,37],[15,9699,9701],{"className":9700},[361],[15,9702,9704],{"className":9703},[365],[15,9705,9707],{"className":9706},[370],[15,9708,9710],{"className":9709,"style":4064},[374],[15,9711,9712,9715],{"style":4067},[15,9713],{"className":9714,"style":383},[382],[15,9716,9718],{"className":9717},[387,388,389,390],[15,9719,9721],{"className":9720},[36,390],[15,9722,1215],{"className":9723},[36,390],[15,9725,68],{"className":9726},[67],[15,9728,230],{"className":9729,"style":229},[36,37],[15,9731,104],{"className":9732},[103],[15,9734],{"className":9735,"style":43},[42],[15,9737,48],{"className":9738},[47],[15,9740],{"className":9741,"style":43},[42],[15,9743,9745,9749,9882,9885,9888],{"className":9744},[27],[15,9746],{"className":9747,"style":9748},[31],"height:2.1107em;vertical-align:-0.7693em;",[15,9750,9752,9755,9879],{"className":9751},[36],[15,9753],{"className":9754},[67,1410],[15,9756,9758],{"className":9757},[1414],[15,9759,9761,9871],{"className":9760},[365,366],[15,9762,9764,9868],{"className":9763},[370],[15,9765,9768,9820,9828],{"className":9766,"style":9767},[374],"height:1.3414em;",[15,9769,9770,9773],{"style":1427},[15,9771],{"className":9772,"style":1431},[382],[15,9774,9776,9779,9782,9785,9788],{"className":9775},[36],[15,9777,497],{"className":9778},[36],[15,9780],{"className":9781,"style":81},[42],[15,9783,86],{"className":9784},[85],[15,9786],{"className":9787,"style":81},[42],[15,9789,9791,9794],{"className":9790},[36],[15,9792,1453],{"className":9793},[36,37],[15,9795,9797],{"className":9796},[361],[15,9798,9800],{"className":9799},[365],[15,9801,9803],{"className":9802},[370],[15,9804,9806],{"className":9805,"style":1810},[374],[15,9807,9808,9811],{"style":1469},[15,9809],{"className":9810,"style":383},[382],[15,9812,9814],{"className":9813},[387,388,389,390],[15,9815,9817],{"className":9816},[36,390],[15,9818,230],{"className":9819,"style":229},[36,37,390],[15,9821,9822,9825],{"style":1488},[15,9823],{"className":9824,"style":1431},[382],[15,9826],{"className":9827,"style":1496},[1495],[15,9829,9830,9833],{"style":1499},[15,9831],{"className":9832,"style":1431},[382],[15,9834,9836],{"className":9835},[36],[15,9837,9839,9842],{"className":9838},[36],[15,9840,1453],{"className":9841},[36,37],[15,9843,9845],{"className":9844},[361],[15,9846,9848],{"className":9847},[365],[15,9849,9851],{"className":9850},[370],[15,9852,9854],{"className":9853,"style":1903},[374],[15,9855,9856,9859],{"style":1202},[15,9857],{"className":9858,"style":383},[382],[15,9860,9862],{"className":9861},[387,388,389,390],[15,9863,9865],{"className":9864},[36,390],[15,9866,230],{"className":9867,"style":229},[36,37,390],[15,9869,399],{"className":9870},[398],[15,9872,9874],{"className":9873},[370],[15,9875,9877],{"className":9876,"style":1518},[374],[15,9878],{},[15,9880],{"className":9881},[103,1410],[15,9883],{"className":9884,"style":43},[42],[15,9886,48],{"className":9887},[47],[15,9889],{"className":9890,"style":43},[42],[15,9892,9894,9897,10003,10006,10009],{"className":9893},[27],[15,9895],{"className":9896,"style":1403},[31],[15,9898,9900,9903,10000],{"className":9899},[36],[15,9901],{"className":9902},[67,1410],[15,9904,9906],{"className":9905},[1414],[15,9907,9909,9992],{"className":9908},[365,366],[15,9910,9912,9989],{"className":9911},[370],[15,9913,9915,9970,9978],{"className":9914,"style":1424},[374],[15,9916,9917,9920],{"style":1427},[15,9918],{"className":9919,"style":1431},[382],[15,9921,9923,9926,9929,9932,9935],{"className":9922},[36],[15,9924,497],{"className":9925},[36],[15,9927],{"className":9928,"style":81},[42],[15,9930,86],{"className":9931},[85],[15,9933],{"className":9934,"style":81},[42],[15,9936,9938,9941],{"className":9937},[36],[15,9939,1453],{"className":9940},[36,37],[15,9942,9944],{"className":9943},[361],[15,9945,9947],{"className":9946},[365],[15,9948,9950],{"className":9949},[370],[15,9951,9953],{"className":9952,"style":1466},[374],[15,9954,9955,9958],{"style":1469},[15,9956],{"className":9957,"style":383},[382],[15,9959,9961],{"className":9960},[387,388,389,390],[15,9962,9964,9967],{"className":9963},[36,390],[15,9965,1482],{"className":9966},[36,390],[15,9968,230],{"className":9969,"style":229},[36,37,390],[15,9971,9972,9975],{"style":1488},[15,9973],{"className":9974,"style":1431},[382],[15,9976],{"className":9977,"style":1496},[1495],[15,9979,9980,9983],{"style":1499},[15,9981],{"className":9982,"style":1431},[382],[15,9984,9986],{"className":9985},[36],[15,9987,497],{"className":9988},[36],[15,9990,399],{"className":9991},[398],[15,9993,9995],{"className":9994},[370],[15,9996,9998],{"className":9997,"style":1518},[374],[15,9999],{},[15,10001],{"className":10002},[103,1410],[15,10004],{"className":10005,"style":43},[42],[15,10007,48],{"className":10008},[47],[15,10010],{"className":10011,"style":43},[42],[15,10013,10015,10018,10021,10024,10027,10030],{"className":10014},[27],[15,10016],{"className":10017,"style":58},[31],[15,10019,1541],{"className":10020,"style":62},[36,37],[15,10022,68],{"className":10023},[67],[15,10025,230],{"className":10026,"style":229},[36,37],[15,10028,104],{"className":10029},[103],[15,10031,4730],{"className":10032},[36],[11,10034,10035,10036,10072,10073,10162,10163,10199,10200,10215,10216,8170,10231,10246,10247,4730],{},"As ",[15,10037,10039],{"className":10038},[18],[15,10040,10042,10060],{"className":10041,"ariaHidden":23},[22],[15,10043,10045,10048,10051,10054,10057],{"className":10044},[27],[15,10046],{"className":10047,"style":225},[31],[15,10049,230],{"className":10050,"style":229},[36,37],[15,10052],{"className":10053,"style":43},[42],[15,10055,199],{"className":10056},[47],[15,10058],{"className":10059,"style":43},[42],[15,10061,10063,10066,10069],{"className":10062},[27],[15,10064],{"className":10065,"style":8817},[31],[15,10067,86],{"className":10068},[36],[15,10070,2273],{"className":10071},[36],", ",[15,10074,10076],{"className":10075},[18],[15,10077,10079,10106,10153],{"className":10078,"ariaHidden":23},[22],[15,10080,10082,10085,10091,10094,10097,10100,10103],{"className":10081},[27],[15,10083],{"className":10084,"style":58},[31],[15,10086,10088],{"className":10087},[2014],[15,10089,3105],{"className":10090,"style":3104},[36,2176],[15,10092,68],{"className":10093},[67],[15,10095,497],{"className":10096},[36],[15,10098],{"className":10099,"style":81},[42],[15,10101,86],{"className":10102},[85],[15,10104],{"className":10105,"style":81},[42],[15,10107,10109,10112,10141,10144,10147,10150],{"className":10108},[27],[15,10110],{"className":10111,"style":58},[31],[15,10113,10115,10118],{"className":10114},[36],[15,10116,1453],{"className":10117},[36,37],[15,10119,10121],{"className":10120},[361],[15,10122,10124],{"className":10123},[365],[15,10125,10127],{"className":10126},[370],[15,10128,10130],{"className":10129,"style":1903},[374],[15,10131,10132,10135],{"style":1202},[15,10133],{"className":10134,"style":383},[382],[15,10136,10138],{"className":10137},[387,388,389,390],[15,10139,230],{"className":10140,"style":229},[36,37,390],[15,10142,104],{"className":10143},[103],[15,10145],{"className":10146,"style":43},[42],[15,10148,199],{"className":10149},[47],[15,10151],{"className":10152,"style":43},[42],[15,10154,10156,10159],{"className":10155},[27],[15,10157],{"className":10158,"style":225},[31],[15,10160,230],{"className":10161,"style":229},[36,37]," (it hugs the ReLU line); as ",[15,10164,10166],{"className":10165},[18],[15,10167,10169,10187],{"className":10168,"ariaHidden":23},[22],[15,10170,10172,10175,10178,10181,10184],{"className":10171},[27],[15,10173],{"className":10174,"style":225},[31],[15,10176,230],{"className":10177,"style":229},[36,37],[15,10179],{"className":10180,"style":43},[42],[15,10182,199],{"className":10183},[47],[15,10185],{"className":10186,"style":43},[42],[15,10188,10190,10193,10196],{"className":10189},[27],[15,10191],{"className":10192,"style":8817},[31],[15,10194,1482],{"className":10195},[36],[15,10197,2273],{"className":10198},[36]," it decays to ",[15,10201,10203],{"className":10202},[18],[15,10204,10206],{"className":10205,"ariaHidden":23},[22],[15,10207,10209,10212],{"className":10208},[27],[15,10210],{"className":10211,"style":2147},[31],[15,10213,1589],{"className":10214},[36]," smoothly. Its derivative never reaches ",[15,10217,10219],{"className":10218},[18],[15,10220,10222],{"className":10221,"ariaHidden":23},[22],[15,10223,10225,10228],{"className":10224},[27],[15,10226],{"className":10227,"style":2147},[31],[15,10229,1589],{"className":10230},[36],[15,10232,10234],{"className":10233},[18],[15,10235,10237],{"className":10236,"ariaHidden":23},[22],[15,10238,10240,10243],{"className":10239},[27],[15,10241],{"className":10242,"style":2147},[31],[15,10244,497],{"className":10245},[36],"\nexactly, so it has no dead region but also no perfectly flat gradient of ",[15,10248,10250],{"className":10249},[18],[15,10251,10253],{"className":10252,"ariaHidden":23},[22],[15,10254,10256,10259],{"className":10255},[27],[15,10257],{"className":10258,"style":2147},[31],[15,10260,497],{"className":10261},[36],[3890,10263,10265],{"id":10264},"gelu-and-swishsilu","GELU and swish\u002FSiLU",[11,10267,10268,10269,10326,10327,10342,10343,10358],{},"GELU gates the input by the probability that a standard normal falls below it,\n",[15,10270,10272],{"className":10271},[18],[15,10273,10275,10302],{"className":10274,"ariaHidden":23},[22],[15,10276,10278,10281,10284,10287,10290,10293,10296,10299],{"className":10277},[27],[15,10279],{"className":10280,"style":58},[31],[15,10282,63],{"className":10283,"style":62},[36,37],[15,10285,68],{"className":10286},[67],[15,10288,230],{"className":10289,"style":229},[36,37],[15,10291,104],{"className":10292},[103],[15,10294],{"className":10295,"style":43},[42],[15,10297,48],{"className":10298},[47],[15,10300],{"className":10301,"style":43},[42],[15,10303,10305,10308,10311,10314,10317,10320,10323],{"className":10304},[27],[15,10306],{"className":10307,"style":58},[31],[15,10309,230],{"className":10310,"style":229},[36,37],[15,10312],{"className":10313,"style":1598},[42],[15,10315,2940],{"className":10316},[36],[15,10318,68],{"className":10319},[67],[15,10321,230],{"className":10322,"style":229},[36,37],[15,10324,104],{"className":10325},[103]," with ",[15,10328,10330],{"className":10329},[18],[15,10331,10333],{"className":10332,"ariaHidden":23},[22],[15,10334,10336,10339],{"className":10335},[27],[15,10337],{"className":10338,"style":3074},[31],[15,10340,2940],{"className":10341},[36]," the standard-normal CDF and ",[15,10344,10346],{"className":10345},[18],[15,10347,10349],{"className":10348,"ariaHidden":23},[22],[15,10350,10352,10355],{"className":10351},[27],[15,10353],{"className":10354,"style":3844},[31],[15,10356,3000],{"className":10357},[36,37]," its density.\nThe product rule gives",[15,10360,10362],{"className":10361},[4032],[15,10363,10365],{"className":10364},[18],[15,10366,10368,10424,10451,10505],{"className":10367,"ariaHidden":23},[22],[15,10369,10371,10374,10406,10409,10412,10415,10418,10421],{"className":10370},[27],[15,10372],{"className":10373,"style":4045},[31],[15,10375,10377,10380],{"className":10376},[36],[15,10378,63],{"className":10379,"style":62},[36,37],[15,10381,10383],{"className":10382},[361],[15,10384,10386],{"className":10385},[365],[15,10387,10389],{"className":10388},[370],[15,10390,10392],{"className":10391,"style":4064},[374],[15,10393,10394,10397],{"style":4067},[15,10395],{"className":10396,"style":383},[382],[15,10398,10400],{"className":10399},[387,388,389,390],[15,10401,10403],{"className":10402},[36,390],[15,10404,1215],{"className":10405},[36,390],[15,10407,68],{"className":10408},[67],[15,10410,230],{"className":10411,"style":229},[36,37],[15,10413,104],{"className":10414},[103],[15,10416],{"className":10417,"style":43},[42],[15,10419,48],{"className":10420},[47],[15,10422],{"className":10423,"style":43},[42],[15,10425,10427,10430,10433,10436,10439,10442,10445,10448],{"className":10426},[27],[15,10428],{"className":10429,"style":58},[31],[15,10431,2940],{"className":10432},[36],[15,10434,68],{"className":10435},[67],[15,10437,230],{"className":10438,"style":229},[36,37],[15,10440,104],{"className":10441},[103],[15,10443],{"className":10444,"style":81},[42],[15,10446,86],{"className":10447},[85],[15,10449],{"className":10450,"style":81},[42],[15,10452,10454,10457,10460,10463,10466,10469,10472,10475,10478,10481,10484,10487,10490,10493,10496,10499,10502],{"className":10453},[27],[15,10455],{"className":10456,"style":58},[31],[15,10458,230],{"className":10459,"style":229},[36,37],[15,10461],{"className":10462,"style":1598},[42],[15,10464,3000],{"className":10465},[36,37],[15,10467,68],{"className":10468},[67],[15,10470,230],{"className":10471,"style":229},[36,37],[15,10473,104],{"className":10474},[103],[15,10476,1594],{"className":10477},[1593],[15,10479],{"className":10480,"style":8469},[42],[15,10482],{"className":10483,"style":1598},[42],[15,10485,3000],{"className":10486},[36,37],[15,10488,68],{"className":10489},[67],[15,10491,230],{"className":10492,"style":229},[36,37],[15,10494,104],{"className":10495},[103],[15,10497],{"className":10498,"style":43},[42],[15,10500,48],{"className":10501},[47],[15,10503],{"className":10504,"style":43},[42],[15,10506,10508,10512,10647,10714],{"className":10507},[27],[15,10509],{"className":10510,"style":10511},[31],"height:1.5749em;vertical-align:-0.538em;",[15,10513,10515,10518,10644],{"className":10514},[36],[15,10516],{"className":10517},[67,1410],[15,10519,10521],{"className":10520},[1414],[15,10522,10524,10635],{"className":10523},[365,366],[15,10525,10527,10632],{"className":10526},[370],[15,10528,10530,10610,10618],{"className":10529,"style":1689},[374],[15,10531,10533,10536],{"style":10532},"top:-2.551em;",[15,10534],{"className":10535,"style":1431},[382],[15,10537,10539],{"className":10538},[387,388,389,390],[15,10540,10542],{"className":10541},[36,390],[15,10543,10546],{"className":10544},[36,10545,390],"sqrt",[15,10547,10549,10601],{"className":10548},[365,366],[15,10550,10552,10598],{"className":10551},[370],[15,10553,10556,10575],{"className":10554,"style":10555},[374],"height:0.9128em;",[15,10557,10561,10564],{"className":10558,"style":10560},[10559],"svg-align","top:-3em;",[15,10562],{"className":10563,"style":1431},[382],[15,10565,10568,10571],{"className":10566,"style":10567},[36,390],"padding-left:0.833em;",[15,10569,394],{"className":10570},[36,390],[15,10572,10574],{"className":10573,"style":62},[36,37,390],"π",[15,10576,10578,10581],{"style":10577},"top:-2.8728em;",[15,10579],{"className":10580,"style":1431},[382],[15,10582,10586],{"className":10583,"style":10585},[10584,390],"hide-tail","min-width:0.853em;height:1.08em;",[10587,10588,10594],"svg",{"xmlns":10589,"width":10590,"height":10591,"viewBox":10592,"preserveAspectRatio":10593},"http:\u002F\u002Fwww.w3.org\u002F2000\u002Fsvg","400em","1.08em","0 0 400000 1080","xMinYMin slice",[10595,10596],"path",{"d":10597},"M95,702\nc-2.7,0,-7.17,-2.7,-13.5,-8c-5.8,-5.3,-9.5,-10,-9.5,-14\nc0,-2,0.3,-3.3,1,-4c1.3,-2.7,23.83,-20.7,67.5,-54\nc44.2,-33.3,65.8,-50.3,66.5,-51c1.3,-1.3,3,-2,5,-2c4.7,0,8.7,3.3,12,10\ns173,378,173,378c0.7,0,35.3,-71,104,-213c68.7,-142,137.5,-285,206.5,-429\nc69,-144,104.5,-217.7,106.5,-221\nl0 -0\nc5.3,-9.3,12,-14,20,-14\nH400000v40H845.2724\ns-225.272,467,-225.272,467s-235,486,-235,486c-2.7,4.7,-9,7,-19,7\nc-6,0,-10,-1,-12,-3s-194,-422,-194,-422s-65,47,-65,47z\nM834 80h400000v40h-400000z",[15,10599,399],{"className":10600},[398],[15,10602,10604],{"className":10603},[370],[15,10605,10608],{"className":10606,"style":10607},[374],"height:0.1272em;",[15,10609],{},[15,10611,10612,10615],{"style":1488},[15,10613],{"className":10614,"style":1431},[382],[15,10616],{"className":10617,"style":1496},[1495],[15,10619,10620,10623],{"style":1716},[15,10621],{"className":10622,"style":1431},[382],[15,10624,10626],{"className":10625},[387,388,389,390],[15,10627,10629],{"className":10628},[36,390],[15,10630,497],{"className":10631},[36,390],[15,10633,399],{"className":10634},[398],[15,10636,10638],{"className":10637},[370],[15,10639,10642],{"className":10640,"style":10641},[374],"height:0.538em;",[15,10643],{},[15,10645],{"className":10646},[103,1410],[15,10648,10650,10653],{"className":10649},[36],[15,10651,1453],{"className":10652},[36,37],[15,10654,10656],{"className":10655},[361],[15,10657,10659],{"className":10658},[365],[15,10660,10662],{"className":10661},[370],[15,10663,10666],{"className":10664,"style":10665},[374],"height:1.0369em;",[15,10667,10668,10671],{"style":4067},[15,10669],{"className":10670,"style":383},[382],[15,10672,10674],{"className":10673},[387,388,389,390],[15,10675,10677,10680,10710],{"className":10676},[36,390],[15,10678,1482],{"className":10679},[36,390],[15,10681,10683,10686],{"className":10682},[36,390],[15,10684,230],{"className":10685,"style":229},[36,37,390],[15,10687,10689],{"className":10688},[361],[15,10690,10692],{"className":10691},[365],[15,10693,10695],{"className":10694},[370],[15,10696,10699],{"className":10697,"style":10698},[374],"height:0.8913em;",[15,10700,10701,10704],{"style":5104},[15,10702],{"className":10703,"style":3730},[382],[15,10705,10707],{"className":10706},[387,3734,3735,390],[15,10708,394],{"className":10709},[36,390],[15,10711,10713],{"className":10712},[36,390],"\u002F2",[15,10715,4730],{"className":10716},[36],[11,10718,10719,10720,10777],{},"Swish (a.k.a. SiLU) is the same idea with a logistic gate, ",[15,10721,10723],{"className":10722},[18],[15,10724,10726,10753],{"className":10725,"ariaHidden":23},[22],[15,10727,10729,10732,10735,10738,10741,10744,10747,10750],{"className":10728},[27],[15,10730],{"className":10731,"style":58},[31],[15,10733,63],{"className":10734,"style":62},[36,37],[15,10736,68],{"className":10737},[67],[15,10739,230],{"className":10740,"style":229},[36,37],[15,10742,104],{"className":10743},[103],[15,10745],{"className":10746,"style":43},[42],[15,10748,48],{"className":10749},[47],[15,10751],{"className":10752,"style":43},[42],[15,10754,10756,10759,10762,10765,10768,10771,10774],{"className":10755},[27],[15,10757],{"className":10758,"style":58},[31],[15,10760,230],{"className":10761,"style":229},[36,37],[15,10763],{"className":10764,"style":1598},[42],[15,10766,1541],{"className":10767,"style":62},[36,37],[15,10769,68],{"className":10770},[67],[15,10772,230],{"className":10773,"style":229},[36,37],[15,10775,104],{"className":10776},[103],",\nso",[15,10779,10781],{"className":10780},[4032],[15,10782,10784],{"className":10783},[18],[15,10785,10787,10843,10870,10939],{"className":10786,"ariaHidden":23},[22],[15,10788,10790,10793,10825,10828,10831,10834,10837,10840],{"className":10789},[27],[15,10791],{"className":10792,"style":4045},[31],[15,10794,10796,10799],{"className":10795},[36],[15,10797,63],{"className":10798,"style":62},[36,37],[15,10800,10802],{"className":10801},[361],[15,10803,10805],{"className":10804},[365],[15,10806,10808],{"className":10807},[370],[15,10809,10811],{"className":10810,"style":4064},[374],[15,10812,10813,10816],{"style":4067},[15,10814],{"className":10815,"style":383},[382],[15,10817,10819],{"className":10818},[387,388,389,390],[15,10820,10822],{"className":10821},[36,390],[15,10823,1215],{"className":10824},[36,390],[15,10826,68],{"className":10827},[67],[15,10829,230],{"className":10830,"style":229},[36,37],[15,10832,104],{"className":10833},[103],[15,10835],{"className":10836,"style":43},[42],[15,10838,48],{"className":10839},[47],[15,10841],{"className":10842,"style":43},[42],[15,10844,10846,10849,10852,10855,10858,10861,10864,10867],{"className":10845},[27],[15,10847],{"className":10848,"style":58},[31],[15,10850,1541],{"className":10851,"style":62},[36,37],[15,10853,68],{"className":10854},[67],[15,10856,230],{"className":10857,"style":229},[36,37],[15,10859,104],{"className":10860},[103],[15,10862],{"className":10863,"style":81},[42],[15,10865,86],{"className":10866},[85],[15,10868],{"className":10869,"style":81},[42],[15,10871,10873,10876,10879,10882,10885,10888,10891,10894,10897,10930,10933,10936],{"className":10872},[27],[15,10874],{"className":10875,"style":58},[31],[15,10877,230],{"className":10878,"style":229},[36,37],[15,10880],{"className":10881,"style":1598},[42],[15,10883,1541],{"className":10884,"style":62},[36,37],[15,10886,68],{"className":10887},[67],[15,10889,230],{"className":10890,"style":229},[36,37],[15,10892,104],{"className":10893},[103],[15,10895],{"className":10896,"style":1598},[42],[15,10898,10900,10903,10906,10909,10912,10915,10918,10921,10924,10927],{"className":10899},[5230],[15,10901,68],{"className":10902,"style":5235},[67,5234],[15,10904,497],{"className":10905},[36],[15,10907],{"className":10908,"style":81},[42],[15,10910,1482],{"className":10911},[85],[15,10913],{"className":10914,"style":81},[42],[15,10916,1541],{"className":10917,"style":62},[36,37],[15,10919,68],{"className":10920},[67],[15,10922,230],{"className":10923,"style":229},[36,37],[15,10925,104],{"className":10926},[103],[15,10928,104],{"className":10929,"style":5235},[103,5234],[15,10931],{"className":10932,"style":43},[42],[15,10934,48],{"className":10935},[47],[15,10937],{"className":10938,"style":43},[42],[15,10940,10942,10945,10948,10951,10954,10957,10960,11014,11017],{"className":10941},[27],[15,10943],{"className":10944,"style":58},[31],[15,10946,1541],{"className":10947,"style":62},[36,37],[15,10949,68],{"className":10950},[67],[15,10952,230],{"className":10953,"style":229},[36,37],[15,10955,104],{"className":10956},[103],[15,10958],{"className":10959,"style":1598},[42],[15,10961,10963,10966,10969,10972,10975,10978,10981,10984,10987,10990,10993,10996,10999,11002,11005,11008,11011],{"className":10962},[5230],[15,10964,68],{"className":10965,"style":5235},[67,5234],[15,10967,497],{"className":10968},[36],[15,10970],{"className":10971,"style":81},[42],[15,10973,86],{"className":10974},[85],[15,10976],{"className":10977,"style":81},[42],[15,10979,230],{"className":10980,"style":229},[36,37],[15,10982],{"className":10983,"style":1598},[42],[15,10985,68],{"className":10986},[67],[15,10988,497],{"className":10989},[36],[15,10991],{"className":10992,"style":81},[42],[15,10994,1482],{"className":10995},[85],[15,10997],{"className":10998,"style":81},[42],[15,11000,1541],{"className":11001,"style":62},[36,37],[15,11003,68],{"className":11004},[67],[15,11006,230],{"className":11007,"style":229},[36,37],[15,11009,3416],{"className":11010},[103],[15,11012,104],{"className":11013,"style":5235},[103,5234],[15,11015],{"className":11016,"style":1598},[42],[15,11018,4730],{"className":11019},[36],[11,11021,11022,11023,11038,11039,11042,11043,11058,11059,11074,11075,11112,11113,11143,11144,11181,11182,11212],{},"Both are smooth, both are the identity in the large-",[15,11024,11026],{"className":11025},[18],[15,11027,11029],{"className":11028,"ariaHidden":23},[22],[15,11030,11032,11035],{"className":11031},[27],[15,11033],{"className":11034,"style":225},[31],[15,11036,230],{"className":11037,"style":229},[36,37]," limit, and both dip\nslightly ",[1227,11040,11041],{},"below"," zero for moderately negative ",[15,11044,11046],{"className":11045},[18],[15,11047,11049],{"className":11048,"ariaHidden":23},[22],[15,11050,11052,11055],{"className":11051},[27],[15,11053],{"className":11054,"style":225},[31],[15,11056,230],{"className":11057,"style":229},[36,37]," before returning to ",[15,11060,11062],{"className":11061},[18],[15,11063,11065],{"className":11064,"ariaHidden":23},[22],[15,11066,11068,11071],{"className":11067},[27],[15,11069],{"className":11070,"style":2147},[31],[15,11072,1589],{"className":11073},[36]," — a\nnon-monotone shape that the empirical record favors in deep Transformers. Their\nminimum sits near ",[15,11076,11078],{"className":11077},[18],[15,11079,11081,11099],{"className":11080,"ariaHidden":23},[22],[15,11082,11084,11087,11090,11093,11096],{"className":11083},[27],[15,11085],{"className":11086,"style":2478},[31],[15,11088,230],{"className":11089,"style":229},[36,37],[15,11091],{"className":11092,"style":43},[42],[15,11094,2492],{"className":11095},[47],[15,11097],{"className":11098,"style":43},[42],[15,11100,11102,11105,11108],{"className":11101},[27],[15,11103],{"className":11104,"style":1991},[31],[15,11106,1482],{"className":11107},[36],[15,11109,11111],{"className":11110},[36],"1.28"," for swish (value ",[15,11114,11116],{"className":11115},[18],[15,11117,11119,11131],{"className":11118,"ariaHidden":23},[22],[15,11120,11122,11125,11128],{"className":11121},[27],[15,11123],{"className":11124,"style":2478},[31],[15,11126,2492],{"className":11127},[47],[15,11129],{"className":11130,"style":43},[42],[15,11132,11134,11137,11140],{"className":11133},[27],[15,11135],{"className":11136,"style":1991},[31],[15,11138,1482],{"className":11139},[36],[15,11141,3452],{"className":11142},[36],") and near\n",[15,11145,11147],{"className":11146},[18],[15,11148,11150,11168],{"className":11149,"ariaHidden":23},[22],[15,11151,11153,11156,11159,11162,11165],{"className":11152},[27],[15,11154],{"className":11155,"style":2478},[31],[15,11157,230],{"className":11158,"style":229},[36,37],[15,11160],{"className":11161,"style":43},[42],[15,11163,2492],{"className":11164},[47],[15,11166],{"className":11167,"style":43},[42],[15,11169,11171,11174,11177],{"className":11170},[27],[15,11172],{"className":11173,"style":1991},[31],[15,11175,1482],{"className":11176},[36],[15,11178,11180],{"className":11179},[36],"0.75"," for GELU (value ",[15,11183,11185],{"className":11184},[18],[15,11186,11188,11200],{"className":11187,"ariaHidden":23},[22],[15,11189,11191,11194,11197],{"className":11190},[27],[15,11192],{"className":11193,"style":2478},[31],[15,11195,2492],{"className":11196},[47],[15,11198],{"className":11199,"style":43},[42],[15,11201,11203,11206,11209],{"className":11202},[27],[15,11204],{"className":11205,"style":1991},[31],[15,11207,1482],{"className":11208},[36],[15,11210,3045],{"className":11211},[36],").",[3890,11214,3474],{"id":11215},"maxout",[11,11217,11218,11219,11234],{},"Maxout is the outlier: it is not a fixed scalar shape but a piecewise-linear unit\nthat takes the max over ",[15,11220,11222],{"className":11221},[18],[15,11223,11225],{"className":11224,"ariaHidden":23},[22],[15,11226,11228,11231],{"className":11227},[27],[15,11229],{"className":11230,"style":32},[31],[15,11232,3791],{"className":11233,"style":3790},[36,37]," affine pieces,",[15,11236,11238],{"className":11237},[4032],[15,11239,11241],{"className":11240},[18],[15,11242,11244,11271,11540,11665],{"className":11243,"ariaHidden":23},[22],[15,11245,11247,11250,11253,11256,11259,11262,11265,11268],{"className":11246},[27],[15,11248],{"className":11249,"style":58},[31],[15,11251,63],{"className":11252,"style":62},[36,37],[15,11254,68],{"className":11255},[67],[15,11257,77],{"className":11258},[36,37],[15,11260,104],{"className":11261},[103],[15,11263],{"className":11264,"style":43},[42],[15,11266,48],{"className":11267},[47],[15,11269],{"className":11270,"style":43},[42],[15,11272,11274,11278,11350,11353,11474,11477,11480,11483,11486,11528,11531,11534,11537],{"className":11273},[27],[15,11275],{"className":11276,"style":11277},[31],"height:1.7873em;vertical-align:-0.8882em;",[15,11279,11282],{"className":11280},[2014,11281],"op-limits",[15,11283,11285,11341],{"className":11284},[365,366],[15,11286,11288,11338],{"className":11287},[370],[15,11289,11291,11325],{"className":11290,"style":225},[374],[15,11292,11294,11297],{"style":11293},"top:-2.3479em;margin-left:0em;",[15,11295],{"className":11296,"style":1431},[382],[15,11298,11300],{"className":11299},[387,388,389,390],[15,11301,11303,11306,11309,11312,11315,11319,11322],{"className":11302},[36,390],[15,11304,3525],{"className":11305,"style":3524},[36,37,390],[15,11307,48],{"className":11308},[47,390],[15,11310,497],{"className":11311},[36,390],[15,11313,1594],{"className":11314},[1593,390],[15,11316,11318],{"className":11317},[5230,390],"…",[15,11320,1594],{"className":11321},[1593,390],[15,11323,3791],{"className":11324,"style":3790},[36,37,390],[15,11326,11327,11330],{"style":10560},[15,11328],{"className":11329,"style":1431},[382],[15,11331,11332],{},[15,11333,11335],{"className":11334},[2014],[15,11336,2177],{"className":11337},[36,2176],[15,11339,399],{"className":11340},[398],[15,11342,11344],{"className":11343},[370],[15,11345,11348],{"className":11346,"style":11347},[374],"height:0.8882em;",[15,11349],{},[15,11351],{"className":11352,"style":1598},[42],[15,11354,11356,11362,11416,11419,11422,11425,11428,11468],{"className":11355},[5230],[15,11357,11359],{"className":11358,"style":5235},[67,5234],[15,11360,68],{"className":11361},[6458,3735],[15,11363,11365,11368],{"className":11364},[36],[15,11366,3548],{"className":11367,"style":3547},[36,37],[15,11369,11371],{"className":11370},[361],[15,11372,11374,11407],{"className":11373},[365,366],[15,11375,11377,11404],{"className":11376},[370],[15,11378,11381,11393],{"className":11379,"style":11380},[374],"height:0.8991em;",[15,11382,11384,11387],{"style":11383},"top:-2.453em;margin-left:-0.0269em;margin-right:0.05em;",[15,11385],{"className":11386,"style":383},[382],[15,11388,11390],{"className":11389},[387,388,389,390],[15,11391,3525],{"className":11392,"style":3524},[36,37,390],[15,11394,11395,11398],{"style":4067},[15,11396],{"className":11397,"style":383},[382],[15,11399,11401],{"className":11400},[387,388,389,390],[15,11402,3585],{"className":11403},[36,390],[15,11405,399],{"className":11406},[398],[15,11408,11410],{"className":11409},[370],[15,11411,11414],{"className":11412,"style":11413},[374],"height:0.3831em;",[15,11415],{},[15,11417,77],{"className":11418},[36,37],[15,11420],{"className":11421,"style":81},[42],[15,11423,86],{"className":11424},[85],[15,11426],{"className":11427,"style":81},[42],[15,11429,11431,11434],{"className":11430},[36],[15,11432,99],{"className":11433},[36,37],[15,11435,11437],{"className":11436},[361],[15,11438,11440,11460],{"className":11439},[365,366],[15,11441,11443,11457],{"className":11442},[370],[15,11444,11446],{"className":11445,"style":3511},[374],[15,11447,11448,11451],{"style":378},[15,11449],{"className":11450,"style":383},[382],[15,11452,11454],{"className":11453},[387,388,389,390],[15,11455,3525],{"className":11456,"style":3524},[36,37,390],[15,11458,399],{"className":11459},[398],[15,11461,11463],{"className":11462},[370],[15,11464,11466],{"className":11465,"style":3535},[374],[15,11467],{},[15,11469,11471],{"className":11470,"style":5235},[103,5234],[15,11472,104],{"className":11473},[6458,3735],[15,11475],{"className":11476,"style":1598},[42],[15,11478,1594],{"className":11479},[1593],[15,11481],{"className":11482,"style":8469},[42],[15,11484],{"className":11485,"style":1598},[42],[15,11487,11489,11493],{"className":11488},[36],[15,11490,11492],{"className":11491},[36],"∇",[15,11494,11496],{"className":11495},[361],[15,11497,11499,11520],{"className":11498},[365,366],[15,11500,11502,11517],{"className":11501},[370],[15,11503,11506],{"className":11504,"style":11505},[374],"height:0.1514em;",[15,11507,11508,11511],{"style":378},[15,11509],{"className":11510,"style":383},[382],[15,11512,11514],{"className":11513},[387,388,389,390],[15,11515,77],{"className":11516},[36,37,390],[15,11518,399],{"className":11519},[398],[15,11521,11523],{"className":11522},[370],[15,11524,11526],{"className":11525,"style":406},[374],[15,11527],{},[15,11529,63],{"className":11530,"style":62},[36,37],[15,11532],{"className":11533,"style":43},[42],[15,11535,48],{"className":11536},[47],[15,11538],{"className":11539,"style":43},[42],[15,11541,11543,11547,11616,11619,11623,11626,11656,11659,11662],{"className":11542},[27],[15,11544],{"className":11545,"style":11546},[31],"height:1.0248em;vertical-align:-0.2861em;",[15,11548,11550,11553],{"className":11549},[36],[15,11551,3548],{"className":11552,"style":3547},[36,37],[15,11554,11556],{"className":11555},[361],[15,11557,11559,11608],{"className":11558},[365,366],[15,11560,11562,11605],{"className":11561},[370],[15,11563,11565],{"className":11564,"style":3511},[374],[15,11566,11567,11570],{"style":3695},[15,11568],{"className":11569,"style":383},[382],[15,11571,11573],{"className":11572},[387,388,389,390],[15,11574,11576],{"className":11575},[36,390],[15,11577,11579,11582],{"className":11578},[36,390],[15,11580,3525],{"className":11581,"style":3524},[36,37,390],[15,11583,11585],{"className":11584},[361],[15,11586,11588],{"className":11587},[365],[15,11589,11591],{"className":11590},[370],[15,11592,11594],{"className":11593,"style":3723},[374],[15,11595,11596,11599],{"style":3726},[15,11597],{"className":11598,"style":3730},[382],[15,11600,11602],{"className":11601},[387,3734,3735,390],[15,11603,3739],{"className":11604},[85,390],[15,11606,399],{"className":11607},[398],[15,11609,11611],{"className":11610},[370],[15,11612,11614],{"className":11613,"style":3535},[374],[15,11615],{},[15,11617,1594],{"className":11618},[1593],[15,11620],{"className":11621,"style":11622},[42],"margin-right:1em;",[15,11624],{"className":11625,"style":1598},[42],[15,11627,11629,11632],{"className":11628},[36],[15,11630,3525],{"className":11631,"style":3524},[36,37],[15,11633,11635],{"className":11634},[361],[15,11636,11638],{"className":11637},[365],[15,11639,11641],{"className":11640},[370],[15,11642,11645],{"className":11643,"style":11644},[374],"height:0.7387em;",[15,11646,11647,11650],{"style":4067},[15,11648],{"className":11649,"style":383},[382],[15,11651,11653],{"className":11652},[387,388,389,390],[15,11654,3739],{"className":11655},[85,390],[15,11657],{"className":11658,"style":43},[42],[15,11660,48],{"className":11661},[47],[15,11663],{"className":11664,"style":43},[42],[15,11666,11668,11672,11678,11681,11730,11733,11851,11854],{"className":11667},[27],[15,11669],{"className":11670,"style":11671},[31],"height:1.7629em;vertical-align:-0.8638em;",[15,11673,11675,11676],{"className":11674},[2014],"ar",[15,11677,63],{"style":3104},[15,11679],{"className":11680,"style":1598},[42],[15,11682,11684],{"className":11683},[2014,11281],[15,11685,11687,11721],{"className":11686},[365,366],[15,11688,11690,11718],{"className":11689},[370],[15,11691,11693,11705],{"className":11692,"style":225},[374],[15,11694,11696,11699],{"style":11695},"top:-2.3723em;margin-left:0em;",[15,11697],{"className":11698,"style":1431},[382],[15,11700,11702],{"className":11701},[387,388,389,390],[15,11703,3525],{"className":11704,"style":3524},[36,37,390],[15,11706,11707,11710],{"style":10560},[15,11708],{"className":11709,"style":1431},[382],[15,11711,11712],{},[15,11713,11715],{"className":11714},[2014],[15,11716,2177],{"className":11717},[36,2176],[15,11719,399],{"className":11720},[398],[15,11722,11724],{"className":11723},[370],[15,11725,11728],{"className":11726,"style":11727},[374],"height:0.8638em;",[15,11729],{},[15,11731],{"className":11732,"style":1598},[42],[15,11734,11736,11742,11793,11796,11799,11802,11805,11845],{"className":11735},[5230],[15,11737,11739],{"className":11738,"style":5235},[67,5234],[15,11740,68],{"className":11741},[6458,3735],[15,11743,11745,11748],{"className":11744},[36],[15,11746,3548],{"className":11747,"style":3547},[36,37],[15,11749,11751],{"className":11750},[361],[15,11752,11754,11785],{"className":11753},[365,366],[15,11755,11757,11782],{"className":11756},[370],[15,11758,11760,11771],{"className":11759,"style":11380},[374],[15,11761,11762,11765],{"style":11383},[15,11763],{"className":11764,"style":383},[382],[15,11766,11768],{"className":11767},[387,388,389,390],[15,11769,3525],{"className":11770,"style":3524},[36,37,390],[15,11772,11773,11776],{"style":4067},[15,11774],{"className":11775,"style":383},[382],[15,11777,11779],{"className":11778},[387,388,389,390],[15,11780,3585],{"className":11781},[36,390],[15,11783,399],{"className":11784},[398],[15,11786,11788],{"className":11787},[370],[15,11789,11791],{"className":11790,"style":11413},[374],[15,11792],{},[15,11794,77],{"className":11795},[36,37],[15,11797],{"className":11798,"style":81},[42],[15,11800,86],{"className":11801},[85],[15,11803],{"className":11804,"style":81},[42],[15,11806,11808,11811],{"className":11807},[36],[15,11809,99],{"className":11810},[36,37],[15,11812,11814],{"className":11813},[361],[15,11815,11817,11837],{"className":11816},[365,366],[15,11818,11820,11834],{"className":11819},[370],[15,11821,11823],{"className":11822,"style":3511},[374],[15,11824,11825,11828],{"style":378},[15,11826],{"className":11827,"style":383},[382],[15,11829,11831],{"className":11830},[387,388,389,390],[15,11832,3525],{"className":11833,"style":3524},[36,37,390],[15,11835,399],{"className":11836},[398],[15,11838,11840],{"className":11839},[370],[15,11841,11843],{"className":11842,"style":3535},[374],[15,11844],{},[15,11846,11848],{"className":11847,"style":5235},[103,5234],[15,11849,104],{"className":11850},[6458,3735],[15,11852],{"className":11853,"style":1598},[42],[15,11855,4730],{"className":11856},[36],[11,11858,11859,11860,3831,11893,11964,11965,11983],{},"The gradient flows entirely through the winning piece, so maxout has no flat\nregion at all as long as one piece is increasing; with ",[15,11861,11863],{"className":11862},[18],[15,11864,11866,11884],{"className":11865,"ariaHidden":23},[22],[15,11867,11869,11872,11875,11878,11881],{"className":11868},[27],[15,11870],{"className":11871,"style":32},[31],[15,11873,3791],{"className":11874,"style":3790},[36,37],[15,11876],{"className":11877,"style":43},[42],[15,11879,48],{"className":11880},[47],[15,11882],{"className":11883,"style":43},[42],[15,11885,11887,11890],{"className":11886},[27],[15,11888],{"className":11889,"style":2147},[31],[15,11891,394],{"className":11892},[36],[15,11894,11896],{"className":11895},[18],[15,11897,11899,11955],{"className":11898,"ariaHidden":23},[22],[15,11900,11902,11906,11946,11949,11952],{"className":11901},[27],[15,11903],{"className":11904,"style":11905},[31],"height:0.5806em;vertical-align:-0.15em;",[15,11907,11909,11912],{"className":11908},[36],[15,11910,3548],{"className":11911,"style":3547},[36,37],[15,11913,11915],{"className":11914},[361],[15,11916,11918,11938],{"className":11917},[365,366],[15,11919,11921,11935],{"className":11920},[370],[15,11922,11924],{"className":11923,"style":375},[374],[15,11925,11926,11929],{"style":3695},[15,11927],{"className":11928,"style":383},[382],[15,11930,11932],{"className":11931},[387,388,389,390],[15,11933,394],{"className":11934},[36,390],[15,11936,399],{"className":11937},[398],[15,11939,11941],{"className":11940},[370],[15,11942,11944],{"className":11943,"style":406},[374],[15,11945],{},[15,11947],{"className":11948,"style":43},[42],[15,11950,48],{"className":11951},[47],[15,11953],{"className":11954,"style":43},[42],[15,11956,11958,11961],{"className":11957},[27],[15,11959],{"className":11960,"style":2147},[31],[15,11962,1589],{"className":11963},[36]," it\nrecovers ReLU exactly. The cost is ",[15,11966,11968],{"className":11967},[18],[15,11969,11971],{"className":11970,"ariaHidden":23},[22],[15,11972,11974,11977,11980],{"className":11973},[27],[15,11975],{"className":11976,"style":3786},[31],[15,11978,3791],{"className":11979,"style":3790},[36,37],[15,11981,3795],{"className":11982},[36]," the parameters, which is why it is\nrarely the default.",[11985,11986],"tikz-figure",{"hash":11987},"d5420fb011ae83d099dd67a5d362030c1674155d780d49312eac0aba17952dfa",[11,11989,11990],{},"The rectified family shares one shape below, all identity on the right and\ndiffering only in how they treat the negative side: hard zero (ReLU), a shallow\nline (Leaky\u002FPReLU), or a smooth curve that dips and returns (ELU, GELU, swish).",[11985,11992],{"hash":11993},"18a1f456063a5e8203ad7077ef6598e908dbdcc9ef2a6d8b55b14f8b746b28d2",[1161,11995,11997],{"id":11996},"the-vanishing-gradient-problem-made-quantitative","The vanishing-gradient problem, made quantitative",[11,11999,12000,12001,12017,12018,12033],{},"The backward pass through an ",[15,12002,12004],{"className":12003},[18],[15,12005,12007],{"className":12006,"ariaHidden":23},[22],[15,12008,12010,12013],{"className":12009},[27],[15,12011],{"className":12012,"style":3074},[31],[15,12014,12016],{"className":12015},[36,37],"L","-layer network multiplies one Jacobian factor per\nlayer. Along a single path of units the gradient that reaches layer ",[15,12019,12021],{"className":12020},[18],[15,12022,12024],{"className":12023,"ariaHidden":23},[22],[15,12025,12027,12030],{"className":12026},[27],[15,12028],{"className":12029,"style":32},[31],[15,12031,3791],{"className":12032,"style":3790},[36,37]," carries a\nproduct of activation derivatives,",[15,12035,12037],{"className":12036},[4032],[15,12038,12040],{"className":12039},[18],[15,12041,12043,12174],{"className":12042,"ariaHidden":23},[22],[15,12044,12046,12050,12159,12162,12165,12168,12171],{"className":12045},[27],[15,12047],{"className":12048,"style":12049},[31],"height:2.0754em;vertical-align:-0.704em;",[15,12051,12053,12056,12156],{"className":12052},[36],[15,12054],{"className":12055},[67,1410],[15,12057,12059],{"className":12058},[1414],[15,12060,12062,12147],{"className":12061},[365,366],[15,12063,12065,12144],{"className":12064},[370],[15,12066,12068,12121,12129],{"className":12067,"style":7814},[374],[15,12069,12071,12074],{"style":12070},"top:-2.296em;",[15,12072],{"className":12073,"style":1431},[382],[15,12075,12077,12082],{"className":12076},[36],[15,12078,12081],{"className":12079,"style":12080},[36],"margin-right:0.0556em;","∂",[15,12083,12085,12088],{"className":12084},[36],[15,12086,230],{"className":12087,"style":229},[36,37],[15,12089,12091],{"className":12090},[361],[15,12092,12094],{"className":12093},[365],[15,12095,12097],{"className":12096},[370],[15,12098,12101],{"className":12099,"style":12100},[374],"height:0.814em;",[15,12102,12103,12106],{"style":1469},[15,12104],{"className":12105,"style":383},[382],[15,12107,12109],{"className":12108},[387,388,389,390],[15,12110,12112,12115,12118],{"className":12111},[36,390],[15,12113,68],{"className":12114},[67,390],[15,12116,3791],{"className":12117,"style":3790},[36,37,390],[15,12119,104],{"className":12120},[103,390],[15,12122,12123,12126],{"style":1488},[15,12124],{"className":12125,"style":1431},[382],[15,12127],{"className":12128,"style":1496},[1495],[15,12130,12131,12134],{"style":1499},[15,12132],{"className":12133,"style":1431},[382],[15,12135,12137,12140],{"className":12136},[36],[15,12138,12081],{"className":12139,"style":12080},[36],[15,12141,12016],{"className":12142},[36,12143],"mathcal",[15,12145,399],{"className":12146},[398],[15,12148,12150],{"className":12149},[370],[15,12151,12154],{"className":12152,"style":12153},[374],"height:0.704em;",[15,12155],{},[15,12157],{"className":12158},[103,1410],[15,12160],{"className":12161,"style":43},[42],[15,12163],{"className":12164,"style":43},[42],[15,12166,48],{"className":12167},[47],[15,12169],{"className":12170,"style":43},[42],[15,12172],{"className":12173,"style":43},[42],[15,12175,12177,12180,12413,12416,12419,12522],{"className":12176},[27],[15,12178],{"className":12179,"style":12049},[31],[15,12181,12183,12190,12268,12271,12303,12306,12309,12363,12366,12369,12407],{"className":12182},[5230],[15,12184,12186],{"className":12185,"style":5235},[67,5234],[15,12187,68],{"className":12188},[6458,12189],"size2",[15,12191,12193,12200],{"className":12192},[2014],[15,12194,12199],{"className":12195,"style":12198},[2014,12196,12197],"op-symbol","small-op","position:relative;top:0em;","∏",[15,12201,12203],{"className":12202},[361],[15,12204,12206,12259],{"className":12205},[365,366],[15,12207,12209,12256],{"className":12208},[370],[15,12210,12213,12241],{"className":12211,"style":12212},[374],"height:0.9812em;",[15,12214,12216,12219],{"style":12215},"top:-2.4003em;margin-left:0em;margin-right:0.05em;",[15,12217],{"className":12218,"style":383},[382],[15,12220,12222],{"className":12221},[387,388,389,390],[15,12223,12225,12229,12232,12235,12238],{"className":12224},[36,390],[15,12226,12228],{"className":12227},[36,390],"ℓ",[15,12230,48],{"className":12231},[47,390],[15,12233,3791],{"className":12234,"style":3790},[36,37,390],[15,12236,86],{"className":12237},[85,390],[15,12239,497],{"className":12240},[36,390],[15,12242,12244,12247],{"style":12243},"top:-3.2029em;margin-right:0.05em;",[15,12245],{"className":12246,"style":383},[382],[15,12248,12250],{"className":12249},[387,388,389,390],[15,12251,12253],{"className":12252},[36,390],[15,12254,12016],{"className":12255},[36,37,390],[15,12257,399],{"className":12258},[398],[15,12260,12262],{"className":12261},[370],[15,12263,12266],{"className":12264,"style":12265},[374],"height:0.358em;",[15,12267],{},[15,12269],{"className":12270,"style":1598},[42],[15,12272,12274,12277],{"className":12273},[36],[15,12275,63],{"className":12276,"style":62},[36,37],[15,12278,12280],{"className":12279},[361],[15,12281,12283],{"className":12282},[365],[15,12284,12286],{"className":12285},[370],[15,12287,12289],{"className":12288,"style":1199},[374],[15,12290,12291,12294],{"style":1202},[15,12292],{"className":12293,"style":383},[382],[15,12295,12297],{"className":12296},[387,388,389,390],[15,12298,12300],{"className":12299},[36,390],[15,12301,1215],{"className":12302},[36,390],[15,12304],{"className":12305,"style":2485},[42],[15,12307],{"className":12308,"style":1598},[42],[15,12310,12312,12318,12357],{"className":12311},[5230],[15,12313,12315],{"className":12314,"style":5235},[67,5234],[15,12316,68],{"className":12317},[6458,3735],[15,12319,12321,12324],{"className":12320},[36],[15,12322,230],{"className":12323,"style":229},[36,37],[15,12325,12327],{"className":12326},[361],[15,12328,12330],{"className":12329},[365],[15,12331,12333],{"className":12332},[370],[15,12334,12337],{"className":12335,"style":12336},[374],"height:0.888em;",[15,12338,12339,12342],{"style":1202},[15,12340],{"className":12341,"style":383},[382],[15,12343,12345],{"className":12344},[387,388,389,390],[15,12346,12348,12351,12354],{"className":12347},[36,390],[15,12349,68],{"className":12350},[67,390],[15,12352,12228],{"className":12353},[36,390],[15,12355,104],{"className":12356},[103,390],[15,12358,12360],{"className":12359,"style":5235},[103,5234],[15,12361,104],{"className":12362},[6458,3735],[15,12364],{"className":12365,"style":1598},[42],[15,12367],{"className":12368,"style":1598},[42],[15,12370,12372,12375],{"className":12371},[36],[15,12373,3548],{"className":12374,"style":3547},[36,37],[15,12376,12378],{"className":12377},[361],[15,12379,12381],{"className":12380},[365],[15,12382,12384],{"className":12383},[370],[15,12385,12387],{"className":12386,"style":12336},[374],[15,12388,12389,12392],{"style":1202},[15,12390],{"className":12391,"style":383},[382],[15,12393,12395],{"className":12394},[387,388,389,390],[15,12396,12398,12401,12404],{"className":12397},[36,390],[15,12399,68],{"className":12400},[67,390],[15,12402,12228],{"className":12403},[36,390],[15,12405,104],{"className":12406},[103,390],[15,12408,12410],{"className":12409,"style":5235},[103,5234],[15,12411,104],{"className":12412},[6458,12189],[15,12414],{"className":12415,"style":1598},[42],[15,12417],{"className":12418,"style":1598},[42],[15,12420,12422,12425,12519],{"className":12421},[36],[15,12423],{"className":12424},[67,1410],[15,12426,12428],{"className":12427},[1414],[15,12429,12431,12511],{"className":12430},[365,366],[15,12432,12434,12508],{"className":12433},[370],[15,12435,12437,12486,12494],{"className":12436,"style":7814},[374],[15,12438,12439,12442],{"style":12070},[15,12440],{"className":12441,"style":1431},[382],[15,12443,12445,12448],{"className":12444},[36],[15,12446,12081],{"className":12447,"style":12080},[36],[15,12449,12451,12454],{"className":12450},[36],[15,12452,38],{"className":12453},[36,37],[15,12455,12457],{"className":12456},[361],[15,12458,12460],{"className":12459},[365],[15,12461,12463],{"className":12462},[370],[15,12464,12466],{"className":12465,"style":12100},[374],[15,12467,12468,12471],{"style":1469},[15,12469],{"className":12470,"style":383},[382],[15,12472,12474],{"className":12473},[387,388,389,390],[15,12475,12477,12480,12483],{"className":12476},[36,390],[15,12478,68],{"className":12479},[67,390],[15,12481,12016],{"className":12482},[36,37,390],[15,12484,104],{"className":12485},[103,390],[15,12487,12488,12491],{"style":1488},[15,12489],{"className":12490,"style":1431},[382],[15,12492],{"className":12493,"style":1496},[1495],[15,12495,12496,12499],{"style":1499},[15,12497],{"className":12498,"style":1431},[382],[15,12500,12502,12505],{"className":12501},[36],[15,12503,12081],{"className":12504,"style":12080},[36],[15,12506,12016],{"className":12507},[36,12143],[15,12509,399],{"className":12510},[398],[15,12512,12514],{"className":12513},[370],[15,12515,12517],{"className":12516,"style":12153},[374],[15,12518],{},[15,12520],{"className":12521},[103,1410],[15,12523,1594],{"className":12524},[1593],[11,12526,12527,12528,12543,12544,12547,12548,12637,12638,12653,12654],{},"so the magnitude of the gradient at depth ",[15,12529,12531],{"className":12530},[18],[15,12532,12534],{"className":12533,"ariaHidden":23},[22],[15,12535,12537,12540],{"className":12536},[27],[15,12538],{"className":12539,"style":32},[31],[15,12541,3791],{"className":12542,"style":3790},[36,37]," scales like the ",[152,12545,12546],{},"product"," of the\nper-layer derivatives ",[15,12549,12551],{"className":12550},[18],[15,12552,12554],{"className":12553,"ariaHidden":23},[22],[15,12555,12557,12561,12593,12596,12634],{"className":12556},[27],[15,12558],{"className":12559,"style":12560},[31],"height:1.138em;vertical-align:-0.25em;",[15,12562,12564,12567],{"className":12563},[36],[15,12565,63],{"className":12566,"style":62},[36,37],[15,12568,12570],{"className":12569},[361],[15,12571,12573],{"className":12572},[365],[15,12574,12576],{"className":12575},[370],[15,12577,12579],{"className":12578,"style":1199},[374],[15,12580,12581,12584],{"style":1202},[15,12582],{"className":12583,"style":383},[382],[15,12585,12587],{"className":12586},[387,388,389,390],[15,12588,12590],{"className":12589},[36,390],[15,12591,1215],{"className":12592},[36,390],[15,12594,68],{"className":12595},[67],[15,12597,12599,12602],{"className":12598},[36],[15,12600,230],{"className":12601,"style":229},[36,37],[15,12603,12605],{"className":12604},[361],[15,12606,12608],{"className":12607},[365],[15,12609,12611],{"className":12610},[370],[15,12612,12614],{"className":12613,"style":12336},[374],[15,12615,12616,12619],{"style":1202},[15,12617],{"className":12618,"style":383},[382],[15,12620,12622],{"className":12621},[387,388,389,390],[15,12623,12625,12628,12631],{"className":12624},[36,390],[15,12626,68],{"className":12627},[67,390],[15,12629,12228],{"className":12630},[36,390],[15,12632,104],{"className":12633},[103,390],[15,12635,104],{"className":12636},[103]," (times the weights). If each factor is\nbelow ",[15,12639,12641],{"className":12640},[18],[15,12642,12644],{"className":12643,"ariaHidden":23},[22],[15,12645,12647,12650],{"className":12646},[27],[15,12648],{"className":12649,"style":2147},[31],[15,12651,497],{"className":12652},[36],", the product shrinks geometrically with depth, and the deep layers stop\nreceiving signal.",[3874,12655,12656],{},[107,12657,394],{"href":12658,"ariaDescribedBy":12659,"dataFootnoteRef":6,"id":12660},"#user-content-fn-gf-vanish",[3880],"user-content-fnref-gf-vanish",[146,12662,12663],{"type":148},[11,12664,12665,12668,12669,12740,12741,12784,12785,2652,12787,12831,12832,12847,12848,12910,12911,11212],{},[152,12666,12667],{},"Definition (Saturating unit)."," A unit whose derivative ",[15,12670,12672],{"className":12671},[18],[15,12673,12675,12731],{"className":12674,"ariaHidden":23},[22],[15,12676,12678,12681,12713,12716,12719,12722,12725,12728],{"className":12677},[27],[15,12679],{"className":12680,"style":1180},[31],[15,12682,12684,12687],{"className":12683},[36],[15,12685,63],{"className":12686,"style":62},[36,37],[15,12688,12690],{"className":12689},[361],[15,12691,12693],{"className":12692},[365],[15,12694,12696],{"className":12695},[370],[15,12697,12699],{"className":12698,"style":1199},[374],[15,12700,12701,12704],{"style":1202},[15,12702],{"className":12703,"style":383},[382],[15,12705,12707],{"className":12706},[387,388,389,390],[15,12708,12710],{"className":12709},[36,390],[15,12711,1215],{"className":12712},[36,390],[15,12714,68],{"className":12715},[67],[15,12717,230],{"className":12718,"style":229},[36,37],[15,12720,104],{"className":12721},[103],[15,12723],{"className":12724,"style":43},[42],[15,12726,199],{"className":12727},[47],[15,12729],{"className":12730,"style":43},[42],[15,12732,12734,12737],{"className":12733},[27],[15,12735],{"className":12736,"style":2147},[31],[15,12738,1589],{"className":12739},[36]," as\n",[15,12742,12744],{"className":12743},[18],[15,12745,12747,12775],{"className":12746,"ariaHidden":23},[22],[15,12748,12750,12753,12766,12769,12772],{"className":12749},[27],[15,12751],{"className":12752,"style":58},[31],[15,12754,12756,12760,12763],{"className":12755},[5230],[15,12757,12759],{"className":12758,"style":5235},[67,5234],"∣",[15,12761,230],{"className":12762,"style":229},[36,37],[15,12764,12759],{"className":12765,"style":5235},[103,5234],[15,12767],{"className":12768,"style":43},[42],[15,12770,199],{"className":12771},[47],[15,12773],{"className":12774,"style":43},[42],[15,12776,12778,12781],{"className":12777},[27],[15,12779],{"className":12780,"style":225},[31],[15,12782,2273],{"className":12783},[36],". Sigmoid and tanh saturate in both tails; a unit is\n",[152,12786,3871],{},[15,12788,12790],{"className":12789},[18],[15,12791,12793],{"className":12792,"ariaHidden":23},[22],[15,12794,12796,12799],{"className":12795},[27],[15,12797],{"className":12798,"style":1243},[31],[15,12800,12802,12805],{"className":12801},[36],[15,12803,63],{"className":12804,"style":62},[36,37],[15,12806,12808],{"className":12807},[361],[15,12809,12811],{"className":12810},[365],[15,12812,12814],{"className":12813},[370],[15,12815,12817],{"className":12816,"style":1199},[374],[15,12818,12819,12822],{"style":1202},[15,12820],{"className":12821,"style":383},[382],[15,12823,12825],{"className":12824},[387,388,389,390],[15,12826,12828],{"className":12827},[36,390],[15,12829,1215],{"className":12830},[36,390]," stays bounded away from ",[15,12833,12835],{"className":12834},[18],[15,12836,12838],{"className":12837,"ariaHidden":23},[22],[15,12839,12841,12844],{"className":12840},[27],[15,12842],{"className":12843,"style":2147},[31],[15,12845,1589],{"className":12846},[36]," on an unbounded region\n(ReLU: ",[15,12849,12851],{"className":12850},[18],[15,12852,12854,12901],{"className":12853,"ariaHidden":23},[22],[15,12855,12857,12860,12892,12895,12898],{"className":12856},[27],[15,12858],{"className":12859,"style":1243},[31],[15,12861,12863,12866],{"className":12862},[36],[15,12864,63],{"className":12865,"style":62},[36,37],[15,12867,12869],{"className":12868},[361],[15,12870,12872],{"className":12871},[365],[15,12873,12875],{"className":12874},[370],[15,12876,12878],{"className":12877,"style":1199},[374],[15,12879,12880,12883],{"style":1202},[15,12881],{"className":12882,"style":383},[382],[15,12884,12886],{"className":12885},[387,388,389,390],[15,12887,12889],{"className":12888},[36,390],[15,12890,1215],{"className":12891},[36,390],[15,12893],{"className":12894,"style":43},[42],[15,12896,48],{"className":12897},[47],[15,12899],{"className":12900,"style":43},[42],[15,12902,12904,12907],{"className":12903},[27],[15,12905],{"className":12906,"style":2147},[31],[15,12908,497],{"className":12909},[36]," for all ",[15,12912,12914],{"className":12913},[18],[15,12915,12917,12935],{"className":12916,"ariaHidden":23},[22],[15,12918,12920,12923,12926,12929,12932],{"className":12919},[27],[15,12921],{"className":12922,"style":2665},[31],[15,12924,230],{"className":12925,"style":229},[36,37],[15,12927],{"className":12928,"style":43},[42],[15,12930,2227],{"className":12931},[47],[15,12933],{"className":12934,"style":43},[42],[15,12936,12938,12941],{"className":12937},[27],[15,12939],{"className":12940,"style":2147},[31],[15,12942,1589],{"className":12943},[36],[11,12945,12946,12947,13045,13046,13094,13095,13128,13129,13235,13236,13290,13291,13389,13390,177],{},"The sigmoid is the worst case: its derivative is small even at its\nmaximum. Maximizing ",[15,12948,12950],{"className":12949},[18],[15,12951,12953,13009,13033],{"className":12952,"ariaHidden":23},[22],[15,12954,12956,12959,12991,12994,12997,13000,13003,13006],{"className":12955},[27],[15,12957],{"className":12958,"style":1180},[31],[15,12960,12962,12965],{"className":12961},[36],[15,12963,1541],{"className":12964,"style":62},[36,37],[15,12966,12968],{"className":12967},[361],[15,12969,12971],{"className":12970},[365],[15,12972,12974],{"className":12973},[370],[15,12975,12977],{"className":12976,"style":1199},[374],[15,12978,12979,12982],{"style":1202},[15,12980],{"className":12981,"style":383},[382],[15,12983,12985],{"className":12984},[387,388,389,390],[15,12986,12988],{"className":12987},[36,390],[15,12989,1215],{"className":12990},[36,390],[15,12992,68],{"className":12993},[67],[15,12995,230],{"className":12996,"style":229},[36,37],[15,12998,104],{"className":12999},[103],[15,13001],{"className":13002,"style":43},[42],[15,13004,48],{"className":13005},[47],[15,13007],{"className":13008,"style":43},[42],[15,13010,13012,13015,13018,13021,13024,13027,13030],{"className":13011},[27],[15,13013],{"className":13014,"style":58},[31],[15,13016,1541],{"className":13017,"style":62},[36,37],[15,13019,68],{"className":13020},[67],[15,13022,497],{"className":13023},[36],[15,13025],{"className":13026,"style":81},[42],[15,13028,1482],{"className":13029},[85],[15,13031],{"className":13032,"style":81},[42],[15,13034,13036,13039,13042],{"className":13035},[27],[15,13037],{"className":13038,"style":58},[31],[15,13040,1541],{"className":13041,"style":62},[36,37],[15,13043,104],{"className":13044},[103]," over ",[15,13047,13049],{"className":13048},[18],[15,13050,13052,13070],{"className":13051,"ariaHidden":23},[22],[15,13053,13055,13058,13061,13064,13067],{"className":13054},[27],[15,13056],{"className":13057,"style":2665},[31],[15,13059,1541],{"className":13060,"style":62},[36,37],[15,13062],{"className":13063,"style":43},[42],[15,13065,5396],{"className":13066},[47],[15,13068],{"className":13069,"style":43},[42],[15,13071,13073,13076,13079,13082,13085,13088,13091],{"className":13072},[27],[15,13074],{"className":13075,"style":58},[31],[15,13077,68],{"className":13078},[67],[15,13080,1589],{"className":13081},[36],[15,13083,1594],{"className":13084},[1593],[15,13086],{"className":13087,"style":1598},[42],[15,13089,497],{"className":13090},[36],[15,13092,104],{"className":13093},[103],": set\n",[15,13096,13098],{"className":13097},[18],[15,13099,13101,13119],{"className":13100,"ariaHidden":23},[22],[15,13102,13104,13107,13110,13113,13116],{"className":13103},[27],[15,13105],{"className":13106,"style":123},[31],[15,13108,11],{"className":13109},[36,37],[15,13111],{"className":13112,"style":43},[42],[15,13114,48],{"className":13115},[47],[15,13117],{"className":13118,"style":43},[42],[15,13120,13122,13125],{"className":13121},[27],[15,13123],{"className":13124,"style":225},[31],[15,13126,1541],{"className":13127,"style":62},[36,37]," and differentiate ",[15,13130,13132],{"className":13131},[18],[15,13133,13135,13159,13180,13199],{"className":13134,"ariaHidden":23},[22],[15,13136,13138,13141,13144,13147,13150,13153,13156],{"className":13137},[27],[15,13139],{"className":13140,"style":58},[31],[15,13142,11],{"className":13143},[36,37],[15,13145,68],{"className":13146},[67],[15,13148,497],{"className":13149},[36],[15,13151],{"className":13152,"style":81},[42],[15,13154,1482],{"className":13155},[85],[15,13157],{"className":13158,"style":81},[42],[15,13160,13162,13165,13168,13171,13174,13177],{"className":13161},[27],[15,13163],{"className":13164,"style":58},[31],[15,13166,11],{"className":13167},[36,37],[15,13169,104],{"className":13170},[103],[15,13172],{"className":13173,"style":43},[42],[15,13175,48],{"className":13176},[47],[15,13178],{"className":13179,"style":43},[42],[15,13181,13183,13187,13190,13193,13196],{"className":13182},[27],[15,13184],{"className":13185,"style":13186},[31],"height:0.7778em;vertical-align:-0.1944em;",[15,13188,11],{"className":13189},[36,37],[15,13191],{"className":13192,"style":81},[42],[15,13194,1482],{"className":13195},[85],[15,13197],{"className":13198,"style":81},[42],[15,13200,13202,13206],{"className":13201},[27],[15,13203],{"className":13204,"style":13205},[31],"height:1.0085em;vertical-align:-0.1944em;",[15,13207,13209,13212],{"className":13208},[36],[15,13210,11],{"className":13211},[36,37],[15,13213,13215],{"className":13214},[361],[15,13216,13218],{"className":13217},[365],[15,13219,13221],{"className":13220},[370],[15,13222,13224],{"className":13223,"style":4010},[374],[15,13225,13226,13229],{"style":1202},[15,13227],{"className":13228,"style":383},[382],[15,13230,13232],{"className":13231},[387,388,389,390],[15,13233,394],{"className":13234},[36,390],", giving ",[15,13237,13239],{"className":13238},[18],[15,13240,13242,13260,13281],{"className":13241,"ariaHidden":23},[22],[15,13243,13245,13248,13251,13254,13257],{"className":13244},[27],[15,13246],{"className":13247,"style":1991},[31],[15,13249,497],{"className":13250},[36],[15,13252],{"className":13253,"style":81},[42],[15,13255,1482],{"className":13256},[85],[15,13258],{"className":13259,"style":81},[42],[15,13261,13263,13266,13269,13272,13275,13278],{"className":13262},[27],[15,13264],{"className":13265,"style":6698},[31],[15,13267,394],{"className":13268},[36],[15,13270,11],{"className":13271},[36,37],[15,13273],{"className":13274,"style":43},[42],[15,13276,48],{"className":13277},[47],[15,13279],{"className":13280,"style":43},[42],[15,13282,13284,13287],{"className":13283},[27],[15,13285],{"className":13286,"style":2147},[31],[15,13288,1589],{"className":13289},[36],", so the\npeak is at ",[15,13292,13294],{"className":13293},[18],[15,13295,13297,13315],{"className":13296,"ariaHidden":23},[22],[15,13298,13300,13303,13306,13309,13312],{"className":13299},[27],[15,13301],{"className":13302,"style":123},[31],[15,13304,11],{"className":13305},[36,37],[15,13307],{"className":13308,"style":43},[42],[15,13310,48],{"className":13311},[47],[15,13313],{"className":13314,"style":43},[42],[15,13316,13318,13321],{"className":13317},[27],[15,13319],{"className":13320,"style":1670},[31],[15,13322,13324,13327,13386],{"className":13323},[36],[15,13325],{"className":13326},[67,1410],[15,13328,13330],{"className":13329},[1414],[15,13331,13333,13378],{"className":13332},[365,366],[15,13334,13336,13375],{"className":13335},[370],[15,13337,13339,13353,13361],{"className":13338,"style":1689},[374],[15,13340,13341,13344],{"style":1692},[15,13342],{"className":13343,"style":1431},[382],[15,13345,13347],{"className":13346},[387,388,389,390],[15,13348,13350],{"className":13349},[36,390],[15,13351,394],{"className":13352},[36,390],[15,13354,13355,13358],{"style":1488},[15,13356],{"className":13357,"style":1431},[382],[15,13359],{"className":13360,"style":1496},[1495],[15,13362,13363,13366],{"style":1716},[15,13364],{"className":13365,"style":1431},[382],[15,13367,13369],{"className":13368},[387,388,389,390],[15,13370,13372],{"className":13371},[36,390],[15,13373,497],{"className":13374},[36,390],[15,13376,399],{"className":13377},[398],[15,13379,13381],{"className":13380},[370],[15,13382,13384],{"className":13383,"style":1738},[374],[15,13385],{},[15,13387],{"className":13388},[103,1410],", i.e. ",[15,13391,13393],{"className":13392},[18],[15,13394,13396,13414],{"className":13395,"ariaHidden":23},[22],[15,13397,13399,13402,13405,13408,13411],{"className":13398},[27],[15,13400],{"className":13401,"style":225},[31],[15,13403,230],{"className":13404,"style":229},[36,37],[15,13406],{"className":13407,"style":43},[42],[15,13409,48],{"className":13410},[47],[15,13412],{"className":13413,"style":43},[42],[15,13415,13417,13420],{"className":13416},[27],[15,13418],{"className":13419,"style":2147},[31],[15,13421,1589],{"className":13422},[36],[15,13424,13426],{"className":13425},[4032],[15,13427,13429],{"className":13428},[18],[15,13430,13432,13541,13624,13707,13790],{"className":13431,"ariaHidden":23},[22],[15,13433,13435,13439,13488,13491,13523,13526,13529,13532,13535,13538],{"className":13434},[27],[15,13436],{"className":13437,"style":13438},[31],"height:1.5019em;vertical-align:-0.7em;",[15,13440,13442],{"className":13441},[2014,11281],[15,13443,13445,13479],{"className":13444},[365,366],[15,13446,13448,13476],{"className":13447},[370],[15,13449,13451,13463],{"className":13450,"style":225},[374],[15,13452,13454,13457],{"style":13453},"top:-2.4em;margin-left:0em;",[15,13455],{"className":13456,"style":1431},[382],[15,13458,13460],{"className":13459},[387,388,389,390],[15,13461,230],{"className":13462,"style":229},[36,37,390],[15,13464,13465,13468],{"style":10560},[15,13466],{"className":13467,"style":1431},[382],[15,13469,13470],{},[15,13471,13473],{"className":13472},[2014],[15,13474,2177],{"className":13475},[36,2176],[15,13477,399],{"className":13478},[398],[15,13480,13482],{"className":13481},[370],[15,13483,13486],{"className":13484,"style":13485},[374],"height:0.7em;",[15,13487],{},[15,13489],{"className":13490,"style":1598},[42],[15,13492,13494,13497],{"className":13493},[36],[15,13495,1541],{"className":13496,"style":62},[36,37],[15,13498,13500],{"className":13499},[361],[15,13501,13503],{"className":13502},[365],[15,13504,13506],{"className":13505},[370],[15,13507,13509],{"className":13508,"style":4064},[374],[15,13510,13511,13514],{"style":4067},[15,13512],{"className":13513,"style":383},[382],[15,13515,13517],{"className":13516},[387,388,389,390],[15,13518,13520],{"className":13519},[36,390],[15,13521,1215],{"className":13522},[36,390],[15,13524,68],{"className":13525},[67],[15,13527,230],{"className":13528,"style":229},[36,37],[15,13530,104],{"className":13531},[103],[15,13533],{"className":13534,"style":43},[42],[15,13536,48],{"className":13537},[47],[15,13539],{"className":13540,"style":43},[42],[15,13542,13544,13547,13615,13618,13621],{"className":13543},[27],[15,13545],{"className":13546,"style":1670},[31],[15,13548,13550,13553,13612],{"className":13549},[36],[15,13551],{"className":13552},[67,1410],[15,13554,13556],{"className":13555},[1414],[15,13557,13559,13604],{"className":13558},[365,366],[15,13560,13562,13601],{"className":13561},[370],[15,13563,13565,13579,13587],{"className":13564,"style":1689},[374],[15,13566,13567,13570],{"style":1692},[15,13568],{"className":13569,"style":1431},[382],[15,13571,13573],{"className":13572},[387,388,389,390],[15,13574,13576],{"className":13575},[36,390],[15,13577,394],{"className":13578},[36,390],[15,13580,13581,13584],{"style":1488},[15,13582],{"className":13583,"style":1431},[382],[15,13585],{"className":13586,"style":1496},[1495],[15,13588,13589,13592],{"style":1716},[15,13590],{"className":13591,"style":1431},[382],[15,13593,13595],{"className":13594},[387,388,389,390],[15,13596,13598],{"className":13597},[36,390],[15,13599,497],{"className":13600},[36,390],[15,13602,399],{"className":13603},[398],[15,13605,13607],{"className":13606},[370],[15,13608,13610],{"className":13609,"style":1738},[374],[15,13611],{},[15,13613],{"className":13614},[103,1410],[15,13616],{"className":13617,"style":81},[42],[15,13619,4210],{"className":13620},[85],[15,13622],{"className":13623,"style":81},[42],[15,13625,13627,13630,13698,13701,13704],{"className":13626},[27],[15,13628],{"className":13629,"style":1670},[31],[15,13631,13633,13636,13695],{"className":13632},[36],[15,13634],{"className":13635},[67,1410],[15,13637,13639],{"className":13638},[1414],[15,13640,13642,13687],{"className":13641},[365,366],[15,13643,13645,13684],{"className":13644},[370],[15,13646,13648,13662,13670],{"className":13647,"style":1689},[374],[15,13649,13650,13653],{"style":1692},[15,13651],{"className":13652,"style":1431},[382],[15,13654,13656],{"className":13655},[387,388,389,390],[15,13657,13659],{"className":13658},[36,390],[15,13660,394],{"className":13661},[36,390],[15,13663,13664,13667],{"style":1488},[15,13665],{"className":13666,"style":1431},[382],[15,13668],{"className":13669,"style":1496},[1495],[15,13671,13672,13675],{"style":1716},[15,13673],{"className":13674,"style":1431},[382],[15,13676,13678],{"className":13677},[387,388,389,390],[15,13679,13681],{"className":13680},[36,390],[15,13682,497],{"className":13683},[36,390],[15,13685,399],{"className":13686},[398],[15,13688,13690],{"className":13689},[370],[15,13691,13693],{"className":13692,"style":1738},[374],[15,13694],{},[15,13696],{"className":13697},[103,1410],[15,13699],{"className":13700,"style":43},[42],[15,13702,48],{"className":13703},[47],[15,13705],{"className":13706,"style":43},[42],[15,13708,13710,13713,13781,13784,13787],{"className":13709},[27],[15,13711],{"className":13712,"style":1670},[31],[15,13714,13716,13719,13778],{"className":13715},[36],[15,13717],{"className":13718},[67,1410],[15,13720,13722],{"className":13721},[1414],[15,13723,13725,13770],{"className":13724},[365,366],[15,13726,13728,13767],{"className":13727},[370],[15,13729,13731,13745,13753],{"className":13730,"style":1689},[374],[15,13732,13733,13736],{"style":1692},[15,13734],{"className":13735,"style":1431},[382],[15,13737,13739],{"className":13738},[387,388,389,390],[15,13740,13742],{"className":13741},[36,390],[15,13743,1705],{"className":13744},[36,390],[15,13746,13747,13750],{"style":1488},[15,13748],{"className":13749,"style":1431},[382],[15,13751],{"className":13752,"style":1496},[1495],[15,13754,13755,13758],{"style":1716},[15,13756],{"className":13757,"style":1431},[382],[15,13759,13761],{"className":13760},[387,388,389,390],[15,13762,13764],{"className":13763},[36,390],[15,13765,497],{"className":13766},[36,390],[15,13768,399],{"className":13769},[398],[15,13771,13773],{"className":13772},[370],[15,13774,13776],{"className":13775,"style":1738},[374],[15,13777],{},[15,13779],{"className":13780},[103,1410],[15,13782],{"className":13783,"style":43},[42],[15,13785,48],{"className":13786},[47],[15,13788],{"className":13789,"style":43},[42],[15,13791,13793,13796],{"className":13792},[27],[15,13794],{"className":13795,"style":2147},[31],[15,13797,13799],{"className":13798},[36],"0.25.",[11,13801,13802,13803,5272,13806,13822,13823,10072,13856,14021,14022,14142,14143,14214,14215,4730],{},"So a sigmoid contributes ",[152,13804,13805],{},"at most",[15,13807,13809],{"className":13808},[18],[15,13810,13812],{"className":13811,"ariaHidden":23},[22],[15,13813,13815,13818],{"className":13814},[27],[15,13816],{"className":13817,"style":2147},[31],[15,13819,13821],{"className":13820},[36],"0.25"," per layer, and far less in the tails.\nWork the tail out numerically: at ",[15,13824,13826],{"className":13825},[18],[15,13827,13829,13847],{"className":13828,"ariaHidden":23},[22],[15,13830,13832,13835,13838,13841,13844],{"className":13831},[27],[15,13833],{"className":13834,"style":225},[31],[15,13836,230],{"className":13837,"style":229},[36,37],[15,13839],{"className":13840,"style":43},[42],[15,13842,48],{"className":13843},[47],[15,13845],{"className":13846,"style":43},[42],[15,13848,13850,13853],{"className":13849},[27],[15,13851],{"className":13852,"style":2147},[31],[15,13854,1705],{"className":13855},[36],[15,13857,13859],{"className":13858},[18],[15,13860,13862,13889,14011],{"className":13861,"ariaHidden":23},[22],[15,13863,13865,13868,13871,13874,13877,13880,13883,13886],{"className":13864},[27],[15,13866],{"className":13867,"style":58},[31],[15,13869,1541],{"className":13870,"style":62},[36,37],[15,13872,68],{"className":13873},[67],[15,13875,1705],{"className":13876},[36],[15,13878,104],{"className":13879},[103],[15,13881],{"className":13882,"style":43},[42],[15,13884,48],{"className":13885},[47],[15,13887],{"className":13888,"style":43},[42],[15,13890,13892,13895,14002,14005,14008],{"className":13891},[27],[15,13893],{"className":13894,"style":4868},[31],[15,13896,13898,13901,13999],{"className":13897},[36],[15,13899],{"className":13900},[67,1410],[15,13902,13904],{"className":13903},[1414],[15,13905,13907,13991],{"className":13906},[365,366],[15,13908,13910,13988],{"className":13909},[370],[15,13911,13913,13966,13974],{"className":13912,"style":1689},[374],[15,13914,13915,13918],{"style":1692},[15,13916],{"className":13917,"style":1431},[382],[15,13919,13921],{"className":13920},[387,388,389,390],[15,13922,13924,13927,13930],{"className":13923},[36,390],[15,13925,497],{"className":13926},[36,390],[15,13928,86],{"className":13929},[85,390],[15,13931,13933,13936],{"className":13932},[36,390],[15,13934,1453],{"className":13935},[36,37,390],[15,13937,13939],{"className":13938},[361],[15,13940,13942],{"className":13941},[365],[15,13943,13945],{"className":13944},[370],[15,13946,13949],{"className":13947,"style":13948},[374],"height:0.7463em;",[15,13950,13951,13954],{"style":3726},[15,13952],{"className":13953,"style":3730},[382],[15,13955,13957],{"className":13956},[387,3734,3735,390],[15,13958,13960,13963],{"className":13959},[36,390],[15,13961,1482],{"className":13962},[36,390],[15,13964,1705],{"className":13965},[36,390],[15,13967,13968,13971],{"style":1488},[15,13969],{"className":13970,"style":1431},[382],[15,13972],{"className":13973,"style":1496},[1495],[15,13975,13976,13979],{"style":1716},[15,13977],{"className":13978,"style":1431},[382],[15,13980,13982],{"className":13981},[387,388,389,390],[15,13983,13985],{"className":13984},[36,390],[15,13986,497],{"className":13987},[36,390],[15,13989,399],{"className":13990},[398],[15,13992,13994],{"className":13993},[370],[15,13995,13997],{"className":13996,"style":4971},[374],[15,13998],{},[15,14000],{"className":14001},[103,1410],[15,14003],{"className":14004,"style":43},[42],[15,14006,2492],{"className":14007},[47],[15,14009],{"className":14010,"style":43},[42],[15,14012,14014,14017],{"className":14013},[27],[15,14015],{"className":14016,"style":2147},[31],[15,14018,14020],{"className":14019},[36],"0.982",", and ",[15,14023,14025],{"className":14024},[18],[15,14026,14028,14084,14111,14132],{"className":14027,"ariaHidden":23},[22],[15,14029,14031,14034,14066,14069,14072,14075,14078,14081],{"className":14030},[27],[15,14032],{"className":14033,"style":1180},[31],[15,14035,14037,14040],{"className":14036},[36],[15,14038,1541],{"className":14039,"style":62},[36,37],[15,14041,14043],{"className":14042},[361],[15,14044,14046],{"className":14045},[365],[15,14047,14049],{"className":14048},[370],[15,14050,14052],{"className":14051,"style":1199},[374],[15,14053,14054,14057],{"style":1202},[15,14055],{"className":14056,"style":383},[382],[15,14058,14060],{"className":14059},[387,388,389,390],[15,14061,14063],{"className":14062},[36,390],[15,14064,1215],{"className":14065},[36,390],[15,14067,68],{"className":14068},[67],[15,14070,1705],{"className":14071},[36],[15,14073,104],{"className":14074},[103],[15,14076],{"className":14077,"style":43},[42],[15,14079,48],{"className":14080},[47],[15,14082],{"className":14083,"style":43},[42],[15,14085,14087,14090,14093,14096,14099,14102,14105,14108],{"className":14086},[27],[15,14088],{"className":14089,"style":58},[31],[15,14091,14020],{"className":14092},[36],[15,14094],{"className":14095,"style":1598},[42],[15,14097,68],{"className":14098},[67],[15,14100,497],{"className":14101},[36],[15,14103],{"className":14104,"style":81},[42],[15,14106,1482],{"className":14107},[85],[15,14109],{"className":14110,"style":81},[42],[15,14112,14114,14117,14120,14123,14126,14129],{"className":14113},[27],[15,14115],{"className":14116,"style":58},[31],[15,14118,14020],{"className":14119},[36],[15,14121,104],{"className":14122},[103],[15,14124],{"className":14125,"style":43},[42],[15,14127,2492],{"className":14128},[47],[15,14130],{"className":14131,"style":43},[42],[15,14133,14135,14138],{"className":14134},[27],[15,14136],{"className":14137,"style":2147},[31],[15,14139,14141],{"className":14140},[36],"0.0177",". The tanh is better (it\npeaks at ",[15,14144,14146],{"className":14145},[18],[15,14147,14149,14205],{"className":14148,"ariaHidden":23},[22],[15,14150,14152,14155,14187,14190,14193,14196,14199,14202],{"className":14151},[27],[15,14153],{"className":14154,"style":6723},[31],[15,14156,14158,14161],{"className":14157},[2014],[15,14159,1749],{"className":14160},[2014],[15,14162,14164],{"className":14163},[361],[15,14165,14167],{"className":14166},[365],[15,14168,14170],{"className":14169},[370],[15,14171,14173],{"className":14172,"style":6742},[374],[15,14174,14175,14178],{"style":2032},[15,14176],{"className":14177,"style":383},[382],[15,14179,14181],{"className":14180},[387,388,389,390],[15,14182,14184],{"className":14183},[36,390],[15,14185,1215],{"className":14186},[36,390],[15,14188,68],{"className":14189},[67],[15,14191,1589],{"className":14192},[36],[15,14194,104],{"className":14195},[103],[15,14197],{"className":14198,"style":43},[42],[15,14200,48],{"className":14201},[47],[15,14203],{"className":14204,"style":43},[42],[15,14206,14208,14211],{"className":14207},[27],[15,14209],{"className":14210,"style":2147},[31],[15,14212,497],{"className":14213},[36],") but still vanishes in the tails, with\n",[15,14216,14218],{"className":14217},[18],[15,14219,14221,14277,14295,14345,14363,14411],{"className":14220,"ariaHidden":23},[22],[15,14222,14224,14227,14259,14262,14265,14268,14271,14274],{"className":14223},[27],[15,14225],{"className":14226,"style":6723},[31],[15,14228,14230,14233],{"className":14229},[2014],[15,14231,1749],{"className":14232},[2014],[15,14234,14236],{"className":14235},[361],[15,14237,14239],{"className":14238},[365],[15,14240,14242],{"className":14241},[370],[15,14243,14245],{"className":14244,"style":6742},[374],[15,14246,14247,14250],{"style":2032},[15,14248],{"className":14249,"style":383},[382],[15,14251,14253],{"className":14252},[387,388,389,390],[15,14254,14256],{"className":14255},[36,390],[15,14257,1215],{"className":14258},[36,390],[15,14260,68],{"className":14261},[67],[15,14263,394],{"className":14264},[36],[15,14266,104],{"className":14267},[103],[15,14269],{"className":14270,"style":43},[42],[15,14272,48],{"className":14273},[47],[15,14275],{"className":14276,"style":43},[42],[15,14278,14280,14283,14286,14289,14292],{"className":14279},[27],[15,14281],{"className":14282,"style":1991},[31],[15,14284,497],{"className":14285},[36],[15,14287],{"className":14288,"style":81},[42],[15,14290,1482],{"className":14291},[85],[15,14293],{"className":14294,"style":81},[42],[15,14296,14298,14301,14330,14333,14336,14339,14342],{"className":14297},[27],[15,14299],{"className":14300,"style":2010},[31],[15,14302,14304,14307],{"className":14303},[2014],[15,14305,1749],{"className":14306},[2014],[15,14308,14310],{"className":14309},[361],[15,14311,14313],{"className":14312},[365],[15,14314,14316],{"className":14315},[370],[15,14317,14319],{"className":14318,"style":2010},[374],[15,14320,14321,14324],{"style":2032},[15,14322],{"className":14323,"style":383},[382],[15,14325,14327],{"className":14326},[387,388,389,390],[15,14328,394],{"className":14329},[36,390],[15,14331],{"className":14332,"style":1598},[42],[15,14334,394],{"className":14335},[36],[15,14337],{"className":14338,"style":43},[42],[15,14340,48],{"className":14341},[47],[15,14343],{"className":14344,"style":43},[42],[15,14346,14348,14351,14354,14357,14360],{"className":14347},[27],[15,14349],{"className":14350,"style":1991},[31],[15,14352,497],{"className":14353},[36],[15,14355],{"className":14356,"style":81},[42],[15,14358,1482],{"className":14359},[85],[15,14361],{"className":14362,"style":81},[42],[15,14364,14366,14369,14373,14402,14405,14408],{"className":14365},[27],[15,14367],{"className":14368,"style":4010},[31],[15,14370,14372],{"className":14371},[36],"0.96",[15,14374,14376,14379],{"className":14375},[36],[15,14377,1705],{"className":14378},[36],[15,14380,14382],{"className":14381},[361],[15,14383,14385],{"className":14384},[365],[15,14386,14388],{"className":14387},[370],[15,14389,14391],{"className":14390,"style":4010},[374],[15,14392,14393,14396],{"style":1202},[15,14394],{"className":14395,"style":383},[382],[15,14397,14399],{"className":14398},[387,388,389,390],[15,14400,394],{"className":14401},[36,390],[15,14403],{"className":14404,"style":43},[42],[15,14406,2492],{"className":14407},[47],[15,14409],{"className":14410,"style":43},[42],[15,14412,14414,14417],{"className":14413},[27],[15,14415],{"className":14416,"style":2147},[31],[15,14418,14420],{"className":14419},[36],"0.071",[1278,14422,14423,14596],{},[1281,14424,14425],{},[1284,14426,14427,14445,14498,14549],{},[1287,14428,14429,14430],{},"depth ",[15,14431,14433],{"className":14432},[18],[15,14434,14436],{"className":14435,"ariaHidden":23},[22],[15,14437,14439,14442],{"className":14438},[27],[15,14440],{"className":14441,"style":3074},[31],[15,14443,12016],{"className":14444},[36,37],[1287,14446,14447,14497],{},[15,14448,14450],{"className":14449},[18],[15,14451,14453],{"className":14452,"ariaHidden":23},[22],[15,14454,14456,14460,14464],{"className":14455},[27],[15,14457],{"className":14458,"style":14459},[31],"height:0.8413em;",[15,14461,14463],{"className":14462},[36],"0.2",[15,14465,14467,14471],{"className":14466},[36],[15,14468,14470],{"className":14469},[36],"5",[15,14472,14474],{"className":14473},[361],[15,14475,14477],{"className":14476},[365],[15,14478,14480],{"className":14479},[370],[15,14481,14483],{"className":14482,"style":14459},[374],[15,14484,14485,14488],{"style":1202},[15,14486],{"className":14487,"style":383},[382],[15,14489,14491],{"className":14490},[387,388,389,390],[15,14492,14494],{"className":14493},[36,390],[15,14495,12016],{"className":14496},[36,37,390]," (sigmoid peak)",[1287,14499,14500,14548],{},[15,14501,14503],{"className":14502},[18],[15,14504,14506],{"className":14505,"ariaHidden":23},[22],[15,14507,14509,14512,14516],{"className":14508},[27],[15,14510],{"className":14511,"style":14459},[31],[15,14513,14515],{"className":14514},[36],"1.",[15,14517,14519,14522],{"className":14518},[36],[15,14520,1589],{"className":14521},[36],[15,14523,14525],{"className":14524},[361],[15,14526,14528],{"className":14527},[365],[15,14529,14531],{"className":14530},[370],[15,14532,14534],{"className":14533,"style":14459},[374],[15,14535,14536,14539],{"style":1202},[15,14537],{"className":14538,"style":383},[382],[15,14540,14542],{"className":14541},[387,388,389,390],[15,14543,14545],{"className":14544},[36,390],[15,14546,12016],{"className":14547},[36,37,390]," (tanh peak)",[1287,14550,14551,14595],{},[15,14552,14554],{"className":14553},[18],[15,14555,14557],{"className":14556,"ariaHidden":23},[22],[15,14558,14560,14563],{"className":14559},[27],[15,14561],{"className":14562,"style":14459},[31],[15,14564,14566,14569],{"className":14565},[36],[15,14567,497],{"className":14568},[36],[15,14570,14572],{"className":14571},[361],[15,14573,14575],{"className":14574},[365],[15,14576,14578],{"className":14577},[370],[15,14579,14581],{"className":14580,"style":14459},[374],[15,14582,14583,14586],{"style":1202},[15,14584],{"className":14585,"style":383},[382],[15,14587,14589],{"className":14588},[387,388,389,390],[15,14590,14592],{"className":14591},[36,390],[15,14593,12016],{"className":14594},[36,37,390]," (ReLU active)",[1381,14597,14598,14668,14792,14918],{},[1284,14599,14600,14617,14634,14651],{},[1386,14601,14602],{},[15,14603,14605],{"className":14604},[18],[15,14606,14608],{"className":14607,"ariaHidden":23},[22],[15,14609,14611,14614],{"className":14610},[27],[15,14612],{"className":14613,"style":2147},[31],[15,14615,497],{"className":14616},[36],[1386,14618,14619],{},[15,14620,14622],{"className":14621},[18],[15,14623,14625],{"className":14624,"ariaHidden":23},[22],[15,14626,14628,14631],{"className":14627},[27],[15,14629],{"className":14630,"style":2147},[31],[15,14632,13821],{"className":14633},[36],[1386,14635,14636],{},[15,14637,14639],{"className":14638},[18],[15,14640,14642],{"className":14641,"ariaHidden":23},[22],[15,14643,14645,14648],{"className":14644},[27],[15,14646],{"className":14647,"style":2147},[31],[15,14649,497],{"className":14650},[36],[1386,14652,14653],{},[15,14654,14656],{"className":14655},[18],[15,14657,14659],{"className":14658,"ariaHidden":23},[22],[15,14660,14662,14665],{"className":14661},[27],[15,14663],{"className":14664,"style":2147},[31],[15,14666,497],{"className":14667},[36],[1284,14669,14670,14687,14758,14775],{},[1386,14671,14672],{},[15,14673,14675],{"className":14674},[18],[15,14676,14678],{"className":14677,"ariaHidden":23},[22],[15,14679,14681,14684],{"className":14680},[27],[15,14682],{"className":14683,"style":2147},[31],[15,14685,14470],{"className":14686},[36],[1386,14688,14689],{},[15,14690,14692],{"className":14691},[18],[15,14693,14695,14714],{"className":14694,"ariaHidden":23},[22],[15,14696,14698,14701,14705,14708,14711],{"className":14697},[27],[15,14699],{"className":14700,"style":1991},[31],[15,14702,14704],{"className":14703},[36],"9.8",[15,14706],{"className":14707,"style":81},[42],[15,14709,3795],{"className":14710},[85],[15,14712],{"className":14713,"style":81},[42],[15,14715,14717,14720,14723],{"className":14716},[27],[15,14718],{"className":14719,"style":4010},[31],[15,14721,497],{"className":14722},[36],[15,14724,14726,14729],{"className":14725},[36],[15,14727,1589],{"className":14728},[36],[15,14730,14732],{"className":14731},[361],[15,14733,14735],{"className":14734},[365],[15,14736,14738],{"className":14737},[370],[15,14739,14741],{"className":14740,"style":4010},[374],[15,14742,14743,14746],{"style":1202},[15,14744],{"className":14745,"style":383},[382],[15,14747,14749],{"className":14748},[387,388,389,390],[15,14750,14752,14755],{"className":14751},[36,390],[15,14753,1482],{"className":14754},[36,390],[15,14756,1705],{"className":14757},[36,390],[1386,14759,14760],{},[15,14761,14763],{"className":14762},[18],[15,14764,14766],{"className":14765,"ariaHidden":23},[22],[15,14767,14769,14772],{"className":14768},[27],[15,14770],{"className":14771,"style":2147},[31],[15,14773,497],{"className":14774},[36],[1386,14776,14777],{},[15,14778,14780],{"className":14779},[18],[15,14781,14783],{"className":14782,"ariaHidden":23},[22],[15,14784,14786,14789],{"className":14785},[27],[15,14787],{"className":14788,"style":2147},[31],[15,14790,497],{"className":14791},[36],[1284,14793,14794,14812,14884,14901],{},[1386,14795,14796],{},[15,14797,14799],{"className":14798},[18],[15,14800,14802],{"className":14801,"ariaHidden":23},[22],[15,14803,14805,14808],{"className":14804},[27],[15,14806],{"className":14807,"style":2147},[31],[15,14809,14811],{"className":14810},[36],"10",[1386,14813,14814],{},[15,14815,14817],{"className":14816},[18],[15,14818,14820,14839],{"className":14819,"ariaHidden":23},[22],[15,14821,14823,14826,14830,14833,14836],{"className":14822},[27],[15,14824],{"className":14825,"style":1991},[31],[15,14827,14829],{"className":14828},[36],"9.5",[15,14831],{"className":14832,"style":81},[42],[15,14834,3795],{"className":14835},[85],[15,14837],{"className":14838,"style":81},[42],[15,14840,14842,14845,14848],{"className":14841},[27],[15,14843],{"className":14844,"style":4010},[31],[15,14846,497],{"className":14847},[36],[15,14849,14851,14854],{"className":14850},[36],[15,14852,1589],{"className":14853},[36],[15,14855,14857],{"className":14856},[361],[15,14858,14860],{"className":14859},[365],[15,14861,14863],{"className":14862},[370],[15,14864,14866],{"className":14865,"style":4010},[374],[15,14867,14868,14871],{"style":1202},[15,14869],{"className":14870,"style":383},[382],[15,14872,14874],{"className":14873},[387,388,389,390],[15,14875,14877,14880],{"className":14876},[36,390],[15,14878,1482],{"className":14879},[36,390],[15,14881,14883],{"className":14882},[36,390],"7",[1386,14885,14886],{},[15,14887,14889],{"className":14888},[18],[15,14890,14892],{"className":14891,"ariaHidden":23},[22],[15,14893,14895,14898],{"className":14894},[27],[15,14896],{"className":14897,"style":2147},[31],[15,14899,497],{"className":14900},[36],[1386,14902,14903],{},[15,14904,14906],{"className":14905},[18],[15,14907,14909],{"className":14908,"ariaHidden":23},[22],[15,14910,14912,14915],{"className":14911},[27],[15,14913],{"className":14914,"style":2147},[31],[15,14916,497],{"className":14917},[36],[1284,14919,14920,14938,15010,15027],{},[1386,14921,14922],{},[15,14923,14925],{"className":14924},[18],[15,14926,14928],{"className":14927,"ariaHidden":23},[22],[15,14929,14931,14934],{"className":14930},[27],[15,14932],{"className":14933,"style":2147},[31],[15,14935,14937],{"className":14936},[36],"20",[1386,14939,14940],{},[15,14941,14943],{"className":14942},[18],[15,14944,14946,14965],{"className":14945,"ariaHidden":23},[22],[15,14947,14949,14952,14956,14959,14962],{"className":14948},[27],[15,14950],{"className":14951,"style":1991},[31],[15,14953,14955],{"className":14954},[36],"9.1",[15,14957],{"className":14958,"style":81},[42],[15,14960,3795],{"className":14961},[85],[15,14963],{"className":14964,"style":81},[42],[15,14966,14968,14971,14974],{"className":14967},[27],[15,14969],{"className":14970,"style":4010},[31],[15,14972,497],{"className":14973},[36],[15,14975,14977,14980],{"className":14976},[36],[15,14978,1589],{"className":14979},[36],[15,14981,14983],{"className":14982},[361],[15,14984,14986],{"className":14985},[365],[15,14987,14989],{"className":14988},[370],[15,14990,14992],{"className":14991,"style":4010},[374],[15,14993,14994,14997],{"style":1202},[15,14995],{"className":14996,"style":383},[382],[15,14998,15000],{"className":14999},[387,388,389,390],[15,15001,15003,15006],{"className":15002},[36,390],[15,15004,1482],{"className":15005},[36,390],[15,15007,15009],{"className":15008},[36,390],"13",[1386,15011,15012],{},[15,15013,15015],{"className":15014},[18],[15,15016,15018],{"className":15017,"ariaHidden":23},[22],[15,15019,15021,15024],{"className":15020},[27],[15,15022],{"className":15023,"style":2147},[31],[15,15025,497],{"className":15026},[36],[1386,15028,15029],{},[15,15030,15032],{"className":15031},[18],[15,15033,15035],{"className":15034,"ariaHidden":23},[22],[15,15036,15038,15041],{"className":15037},[27],[15,15039],{"className":15040,"style":2147},[31],[15,15042,497],{"className":15043},[36],[11,15045,15046,15047,15050,15051,15099,15100,15133,15134,15178,15179,4730,15183],{},"Ten sigmoid layers, each at their ",[1227,15048,15049],{},"best"," point, already\nattenuate the gradient by a factor of ",[15,15052,15054],{"className":15053},[18],[15,15055,15057],{"className":15056,"ariaHidden":23},[22],[15,15058,15060,15063,15066],{"className":15059},[27],[15,15061],{"className":15062,"style":4010},[31],[15,15064,497],{"className":15065},[36],[15,15067,15069,15072],{"className":15068},[36],[15,15070,1589],{"className":15071},[36],[15,15073,15075],{"className":15074},[361],[15,15076,15078],{"className":15077},[365],[15,15079,15081],{"className":15080},[370],[15,15082,15084],{"className":15083,"style":4010},[374],[15,15085,15086,15089],{"style":1202},[15,15087],{"className":15088,"style":383},[382],[15,15090,15092],{"className":15091},[387,388,389,390],[15,15093,15095],{"className":15094},[36,390],[15,15096,15098],{"className":15097},[36,390],"6",", and the realistic factor is worse\nbecause most units sit off their peak. This is why deep sigmoid\u002Ftanh stacks were\nnearly untrainable before ReLU, and why initialization that keeps pre-activations\nnear ",[15,15101,15103],{"className":15102},[18],[15,15104,15106,15124],{"className":15105,"ariaHidden":23},[22],[15,15107,15109,15112,15115,15118,15121],{"className":15108},[27],[15,15110],{"className":15111,"style":225},[31],[15,15113,230],{"className":15114,"style":229},[36,37],[15,15116],{"className":15117,"style":43},[42],[15,15119,48],{"className":15120},[47],[15,15122],{"className":15123,"style":43},[42],[15,15125,15127,15130],{"className":15126},[27],[15,15128],{"className":15129,"style":2147},[31],[15,15131,1589],{"className":15132},[36]," (where ",[15,15135,15137],{"className":15136},[18],[15,15138,15140],{"className":15139,"ariaHidden":23},[22],[15,15141,15143,15146],{"className":15142},[27],[15,15144],{"className":15145,"style":1243},[31],[15,15147,15149,15152],{"className":15148},[36],[15,15150,63],{"className":15151,"style":62},[36,37],[15,15153,15155],{"className":15154},[361],[15,15156,15158],{"className":15157},[365],[15,15159,15161],{"className":15160},[370],[15,15162,15164],{"className":15163,"style":1199},[374],[15,15165,15166,15169],{"style":1202},[15,15167],{"className":15168,"style":383},[382],[15,15170,15172],{"className":15171},[387,388,389,390],[15,15173,15175],{"className":15174},[36,390],[15,15176,1215],{"className":15177},[36,390]," is largest) matters so much; see\n",[107,15180,15182],{"href":15181},"\u002Fdeep-learning\u002Foptimization\u002Finitialization","initialization",[3874,15184,15185],{},[107,15186,15190],{"href":15187,"ariaDescribedBy":15188,"dataFootnoteRef":6,"id":15189},"#user-content-fn-gf-saturate",[3880],"user-content-fnref-gf-saturate","3",[11985,15192],{"hash":15193},"c512f14b58a1aaf7adac256ecfe9ae71c72a9496a526ddffb30eab9b3722289f",[11,15195,15196],{},"The product-of-derivatives view makes the depth penalty geometric, not additive.\nThe next figure shows the same gradient magnitude collapsing layer by layer as it\npropagates backward through a saturating stack.",[11985,15198],{"hash":15199},"e7820baa972e3a1eaae09b6f424924be719cd1b6b1c093798d3201bc1fb63445",[1161,15201,15203],{"id":15202},"relu-and-the-dead-unit-problem","ReLU and the dead-unit problem",[11,15205,15206,15207,15240,15241,15256,15257,15319],{},"ReLU sidesteps vanishing gradients on its active side outright: for ",[15,15208,15210],{"className":15209},[18],[15,15211,15213,15231],{"className":15212,"ariaHidden":23},[22],[15,15214,15216,15219,15222,15225,15228],{"className":15215},[27],[15,15217],{"className":15218,"style":2665},[31],[15,15220,230],{"className":15221,"style":229},[36,37],[15,15223],{"className":15224,"style":43},[42],[15,15226,2227],{"className":15227},[47],[15,15229],{"className":15230,"style":43},[42],[15,15232,15234,15237],{"className":15233},[27],[15,15235],{"className":15236,"style":2147},[31],[15,15238,1589],{"className":15239},[36]," the\nderivative is exactly ",[15,15242,15244],{"className":15243},[18],[15,15245,15247],{"className":15246,"ariaHidden":23},[22],[15,15248,15250,15253],{"className":15249},[27],[15,15251],{"className":15252,"style":2147},[31],[15,15254,497],{"className":15255},[36],", so the product across active layers is ",[15,15258,15260],{"className":15259},[18],[15,15261,15263,15310],{"className":15262,"ariaHidden":23},[22],[15,15264,15266,15269,15301,15304,15307],{"className":15265},[27],[15,15267],{"className":15268,"style":14459},[31],[15,15270,15272,15275],{"className":15271},[36],[15,15273,497],{"className":15274},[36],[15,15276,15278],{"className":15277},[361],[15,15279,15281],{"className":15280},[365],[15,15282,15284],{"className":15283},[370],[15,15285,15287],{"className":15286,"style":14459},[374],[15,15288,15289,15292],{"style":1202},[15,15290],{"className":15291,"style":383},[382],[15,15293,15295],{"className":15294},[387,388,389,390],[15,15296,15298],{"className":15297},[36,390],[15,15299,12016],{"className":15300},[36,37,390],[15,15302],{"className":15303,"style":43},[42],[15,15305,48],{"className":15306},[47],[15,15308],{"className":15309,"style":43},[42],[15,15311,15313,15316],{"className":15312},[27],[15,15314],{"className":15315,"style":2147},[31],[15,15317,497],{"className":15318},[36],", with no\nattenuation at all. That single property, more than any other, made very deep\nnetworks trainable.",[146,15321,15323],{"type":15322},"theorem",[11,15324,15325,15328,15329,10072,15362,15495,15496,15511,15512,15591],{},[152,15326,15327],{},"Theorem (ReLU does not saturate on the right)."," For ",[15,15330,15332],{"className":15331},[18],[15,15333,15335,15353],{"className":15334,"ariaHidden":23},[22],[15,15336,15338,15341,15344,15347,15350],{"className":15337},[27],[15,15339],{"className":15340,"style":2665},[31],[15,15342,230],{"className":15343,"style":229},[36,37],[15,15345],{"className":15346,"style":43},[42],[15,15348,2227],{"className":15349},[47],[15,15351],{"className":15352,"style":43},[42],[15,15354,15356,15359],{"className":15355},[27],[15,15357],{"className":15358,"style":2147},[31],[15,15360,1589],{"className":15361},[36],[15,15363,15365],{"className":15364},[18],[15,15366,15368,15486],{"className":15367,"ariaHidden":23},[22],[15,15369,15371,15375,15378,15450,15453,15459,15462,15465,15468,15471,15474,15477,15480,15483],{"className":15370},[27],[15,15372],{"className":15373,"style":15374},[31],"height:1.2251em;vertical-align:-0.345em;",[15,15376],{"className":15377,"style":1598},[42],[15,15379,15381,15384,15447],{"className":15380},[36],[15,15382],{"className":15383},[67,1410],[15,15385,15387],{"className":15386},[1414],[15,15388,15390,15439],{"className":15389},[365,366],[15,15391,15393,15436],{"className":15392},[370],[15,15394,15397,15414,15422],{"className":15395,"style":15396},[374],"height:0.8801em;",[15,15398,15399,15402],{"style":1692},[15,15400],{"className":15401,"style":1431},[382],[15,15403,15405],{"className":15404},[387,388,389,390],[15,15406,15408,15411],{"className":15407},[36,390],[15,15409,7826],{"className":15410},[36,37,390],[15,15412,230],{"className":15413,"style":229},[36,37,390],[15,15415,15416,15419],{"style":1488},[15,15417],{"className":15418,"style":1431},[382],[15,15420],{"className":15421,"style":1496},[1495],[15,15423,15424,15427],{"style":1716},[15,15425],{"className":15426,"style":1431},[382],[15,15428,15430],{"className":15429},[387,388,389,390],[15,15431,15433],{"className":15432},[36,390],[15,15434,7826],{"className":15435},[36,37,390],[15,15437,399],{"className":15438},[398],[15,15440,15442],{"className":15441},[370],[15,15443,15445],{"className":15444,"style":1738},[374],[15,15446],{},[15,15448],{"className":15449},[103,1410],[15,15451],{"className":15452,"style":1598},[42],[15,15454,15456],{"className":15455},[2014],[15,15457,2177],{"className":15458},[36,2176],[15,15460,68],{"className":15461},[67],[15,15463,1589],{"className":15464},[36],[15,15466,1594],{"className":15467},[1593],[15,15469],{"className":15470,"style":1598},[42],[15,15472,230],{"className":15473,"style":229},[36,37],[15,15475,104],{"className":15476},[103],[15,15478],{"className":15479,"style":43},[42],[15,15481,48],{"className":15482},[47],[15,15484],{"className":15485,"style":43},[42],[15,15487,15489,15492],{"className":15488},[27],[15,15490],{"className":15491,"style":2147},[31],[15,15493,497],{"className":15494},[36],"\nfor all ",[15,15497,15499],{"className":15498},[18],[15,15500,15502],{"className":15501,"ariaHidden":23},[22],[15,15503,15505,15508],{"className":15504},[27],[15,15506],{"className":15507,"style":225},[31],[15,15509,230],{"className":15510,"style":229},[36,37],", so a path of active ReLUs multiplies the backward gradient by\n",[15,15513,15515],{"className":15514},[18],[15,15516,15518,15582],{"className":15517,"ariaHidden":23},[22],[15,15519,15521,15525,15567,15570,15573,15576,15579],{"className":15520},[27],[15,15522],{"className":15523,"style":15524},[31],"height:1.0497em;vertical-align:-0.2997em;",[15,15526,15528,15531],{"className":15527},[2014],[15,15529,12199],{"className":15530,"style":12198},[2014,12196,12197],[15,15532,15534],{"className":15533},[361],[15,15535,15537,15558],{"className":15536},[365,366],[15,15538,15540,15555],{"className":15539},[370],[15,15541,15544],{"className":15542,"style":15543},[374],"height:0.1864em;",[15,15545,15546,15549],{"style":12215},[15,15547],{"className":15548,"style":383},[382],[15,15550,15552],{"className":15551},[387,388,389,390],[15,15553,12228],{"className":15554},[36,390],[15,15556,399],{"className":15557},[398],[15,15559,15561],{"className":15560},[370],[15,15562,15565],{"className":15563,"style":15564},[374],"height:0.2997em;",[15,15566],{},[15,15568],{"className":15569,"style":1598},[42],[15,15571,497],{"className":15572},[36],[15,15574],{"className":15575,"style":43},[42],[15,15577,48],{"className":15578},[47],[15,15580],{"className":15581,"style":43},[42],[15,15583,15585,15588],{"className":15584},[27],[15,15586],{"className":15587,"style":2147},[31],[15,15589,497],{"className":15590},[36],", independent of depth.",[146,15593,15595],{"type":15594},"proof",[11,15596,15597,5272,15600,15654,15655,15688,15689,15704,15705,15720,15721,15736,15737,15752,15753],{},[152,15598,15599],{},"Proof.",[15,15601,15603],{"className":15602},[18],[15,15604,15606,15645],{"className":15605,"ariaHidden":23},[22],[15,15607,15609,15612,15618,15621,15624,15627,15630,15633,15636,15639,15642],{"className":15608},[27],[15,15610],{"className":15611,"style":58},[31],[15,15613,15615],{"className":15614},[2014],[15,15616,2177],{"className":15617},[36,2176],[15,15619,68],{"className":15620},[67],[15,15622,1589],{"className":15623},[36],[15,15625,1594],{"className":15626},[1593],[15,15628],{"className":15629,"style":1598},[42],[15,15631,230],{"className":15632,"style":229},[36,37],[15,15634,104],{"className":15635},[103],[15,15637],{"className":15638,"style":43},[42],[15,15640,48],{"className":15641},[47],[15,15643],{"className":15644,"style":43},[42],[15,15646,15648,15651],{"className":15647},[27],[15,15649],{"className":15650,"style":225},[31],[15,15652,230],{"className":15653,"style":229},[36,37]," on ",[15,15656,15658],{"className":15657},[18],[15,15659,15661,15679],{"className":15660,"ariaHidden":23},[22],[15,15662,15664,15667,15670,15673,15676],{"className":15663},[27],[15,15665],{"className":15666,"style":2665},[31],[15,15668,230],{"className":15669,"style":229},[36,37],[15,15671],{"className":15672,"style":43},[42],[15,15674,2227],{"className":15675},[47],[15,15677],{"className":15678,"style":43},[42],[15,15680,15682,15685],{"className":15681},[27],[15,15683],{"className":15684,"style":2147},[31],[15,15686,1589],{"className":15687},[36],", whose derivative is the constant ",[15,15690,15692],{"className":15691},[18],[15,15693,15695],{"className":15694,"ariaHidden":23},[22],[15,15696,15698,15701],{"className":15697},[27],[15,15699],{"className":15700,"style":2147},[31],[15,15702,497],{"className":15703},[36],"; the\nproduct of ",[15,15706,15708],{"className":15707},[18],[15,15709,15711],{"className":15710,"ariaHidden":23},[22],[15,15712,15714,15717],{"className":15713},[27],[15,15715],{"className":15716,"style":3074},[31],[15,15718,12016],{"className":15719},[36,37]," ones is ",[15,15722,15724],{"className":15723},[18],[15,15725,15727],{"className":15726,"ariaHidden":23},[22],[15,15728,15730,15733],{"className":15729},[27],[15,15731],{"className":15732,"style":2147},[31],[15,15734,497],{"className":15735},[36]," regardless of ",[15,15738,15740],{"className":15739},[18],[15,15741,15743],{"className":15742,"ariaHidden":23},[22],[15,15744,15746,15749],{"className":15745},[27],[15,15747],{"className":15748,"style":3074},[31],[15,15750,12016],{"className":15751},[36,37],". ",[15,15754,15756],{"className":15755},[18],[15,15757,15759],{"className":15758,"ariaHidden":23},[22],[15,15760,15762,15766],{"className":15761},[27],[15,15763],{"className":15764,"style":15765},[31],"height:0.675em;",[15,15767,15771],{"className":15768},[15769,15770],"enclosing","qed",[15,15772,15775],{"className":15773},[36,15774],"amsrm","□",[11,15777,15778,15779,15782,15783,15798,15799,15814,15815,4730],{},"The flat left half causes a different failure. A unit whose pre-activation is negative for\n",[1227,15780,15781],{},"every"," training example outputs ",[15,15784,15786],{"className":15785},[18],[15,15787,15789],{"className":15788,"ariaHidden":23},[22],[15,15790,15792,15795],{"className":15791},[27],[15,15793],{"className":15794,"style":2147},[31],[15,15796,1589],{"className":15797},[36]," and has derivative ",[15,15800,15802],{"className":15801},[18],[15,15803,15805],{"className":15804,"ariaHidden":23},[22],[15,15806,15808,15811],{"className":15807},[27],[15,15809],{"className":15810,"style":2147},[31],[15,15812,1589],{"className":15813},[36],", so it receives no\ngradient and never updates: it is ",[152,15816,15817],{},"dead",[146,15819,15820],{"type":148},[11,15821,15822,15825,15826,15926,15927,3831,15969,16040,16041,16114],{},[152,15823,15824],{},"Definition (Dead ReLU)."," A ReLU unit for which ",[15,15827,15829],{"className":15828},[18],[15,15830,15832,15850,15898,15917],{"className":15831,"ariaHidden":23},[22],[15,15833,15835,15838,15841,15844,15847],{"className":15834},[27],[15,15836],{"className":15837,"style":225},[31],[15,15839,230],{"className":15840,"style":229},[36,37],[15,15842],{"className":15843,"style":43},[42],[15,15845,48],{"className":15846},[47],[15,15848],{"className":15849,"style":43},[42],[15,15851,15853,15857,15886,15889,15892,15895],{"className":15852},[27],[15,15854],{"className":15855,"style":15856},[31],"height:0.9324em;vertical-align:-0.0833em;",[15,15858,15860,15863],{"className":15859},[36],[15,15861,3548],{"className":15862,"style":3547},[36,37],[15,15864,15866],{"className":15865},[361],[15,15867,15869],{"className":15868},[365],[15,15870,15872],{"className":15871},[370],[15,15873,15875],{"className":15874,"style":3561},[374],[15,15876,15877,15880],{"style":1202},[15,15878],{"className":15879,"style":383},[382],[15,15881,15883],{"className":15882},[387,388,389,390],[15,15884,3585],{"className":15885},[36,390],[15,15887,77],{"className":15888},[36,37],[15,15890],{"className":15891,"style":81},[42],[15,15893,86],{"className":15894},[85],[15,15896],{"className":15897,"style":81},[42],[15,15899,15901,15905,15908,15911,15914],{"className":15900},[27],[15,15902],{"className":15903,"style":15904},[31],"height:0.7335em;vertical-align:-0.0391em;",[15,15906,99],{"className":15907},[36,37],[15,15909],{"className":15910,"style":43},[42],[15,15912,8033],{"className":15913},[47],[15,15915],{"className":15916,"style":43},[42],[15,15918,15920,15923],{"className":15919},[27],[15,15921],{"className":15922,"style":2147},[31],[15,15924,1589],{"className":15925},[36]," on the\nentire data distribution. Then ",[15,15928,15930],{"className":15929},[18],[15,15931,15933,15960],{"className":15932,"ariaHidden":23},[22],[15,15934,15936,15939,15942,15945,15948,15951,15954,15957],{"className":15935},[27],[15,15937],{"className":15938,"style":58},[31],[15,15940,63],{"className":15941,"style":62},[36,37],[15,15943,68],{"className":15944},[67],[15,15946,230],{"className":15947,"style":229},[36,37],[15,15949,104],{"className":15950},[103],[15,15952],{"className":15953,"style":43},[42],[15,15955,48],{"className":15956},[47],[15,15958],{"className":15959,"style":43},[42],[15,15961,15963,15966],{"className":15962},[27],[15,15964],{"className":15965,"style":2147},[31],[15,15967,1589],{"className":15968},[36],[15,15970,15972],{"className":15971},[18],[15,15973,15975,16031],{"className":15974,"ariaHidden":23},[22],[15,15976,15978,15981,16013,16016,16019,16022,16025,16028],{"className":15977},[27],[15,15979],{"className":15980,"style":1180},[31],[15,15982,15984,15987],{"className":15983},[36],[15,15985,63],{"className":15986,"style":62},[36,37],[15,15988,15990],{"className":15989},[361],[15,15991,15993],{"className":15992},[365],[15,15994,15996],{"className":15995},[370],[15,15997,15999],{"className":15998,"style":1199},[374],[15,16000,16001,16004],{"style":1202},[15,16002],{"className":16003,"style":383},[382],[15,16005,16007],{"className":16006},[387,388,389,390],[15,16008,16010],{"className":16009},[36,390],[15,16011,1215],{"className":16012},[36,390],[15,16014,68],{"className":16015},[67],[15,16017,230],{"className":16018,"style":229},[36,37],[15,16020,104],{"className":16021},[103],[15,16023],{"className":16024,"style":43},[42],[15,16026,48],{"className":16027},[47],[15,16029],{"className":16030,"style":43},[42],[15,16032,16034,16037],{"className":16033},[27],[15,16035],{"className":16036,"style":2147},[31],[15,16038,1589],{"className":16039},[36]," for all inputs, so\n",[15,16042,16044],{"className":16043},[18],[15,16045,16047,16105],{"className":16046,"ariaHidden":23},[22],[15,16048,16050,16053,16093,16096,16099,16102],{"className":16049},[27],[15,16051],{"className":16052,"style":898},[31],[15,16054,16056,16059],{"className":16055},[36],[15,16057,11492],{"className":16058},[36],[15,16060,16062],{"className":16061},[361],[15,16063,16065,16085],{"className":16064},[365,366],[15,16066,16068,16082],{"className":16067},[370],[15,16069,16071],{"className":16070,"style":11505},[374],[15,16072,16073,16076],{"style":378},[15,16074],{"className":16075,"style":383},[382],[15,16077,16079],{"className":16078},[387,388,389,390],[15,16080,3548],{"className":16081,"style":3547},[36,37,390],[15,16083,399],{"className":16084},[398],[15,16086,16088],{"className":16087},[370],[15,16089,16091],{"className":16090,"style":406},[374],[15,16092],{},[15,16094,12016],{"className":16095},[36,12143],[15,16097],{"className":16098,"style":43},[42],[15,16100,48],{"className":16101},[47],[15,16103],{"className":16104,"style":43},[42],[15,16106,16108,16111],{"className":16107},[27],[15,16109],{"className":16110,"style":2147},[31],[15,16112,1589],{"className":16113},[36],": the unit is frozen and contributes nothing. A large\nnegative bias, or a too-large gradient step that drives the weights negative,\ncan kill a unit permanently.",[11,16116,16117,16118,16171],{},"The mechanism is worth tracing through the chain rule. The gradient on a weight\n",[15,16119,16121],{"className":16120},[18],[15,16122,16124],{"className":16123,"ariaHidden":23},[22],[15,16125,16127,16130],{"className":16126},[27],[15,16128],{"className":16129,"style":11905},[31],[15,16131,16133,16136],{"className":16132},[36],[15,16134,3548],{"className":16135,"style":3547},[36,37],[15,16137,16139],{"className":16138},[361],[15,16140,16142,16163],{"className":16141},[365,366],[15,16143,16145,16160],{"className":16144},[370],[15,16146,16148],{"className":16147,"style":3511},[374],[15,16149,16150,16153],{"style":3695},[15,16151],{"className":16152,"style":383},[382],[15,16154,16156],{"className":16155},[387,388,389,390],[15,16157,16159],{"className":16158},[36,37,390],"i",[15,16161,399],{"className":16162},[398],[15,16164,16166],{"className":16165},[370],[15,16167,16169],{"className":16168,"style":406},[374],[15,16170],{}," of a ReLU unit is",[15,16173,16175],{"className":16174},[4032],[15,16176,16178],{"className":16177},[18],[15,16179,16181,16303,16386,16442,16497,16580,16604,16625],{"className":16180,"ariaHidden":23},[22],[15,16182,16184,16188,16294,16297,16300],{"className":16183},[27],[15,16185],{"className":16186,"style":16187},[31],"height:2.2074em;vertical-align:-0.836em;",[15,16189,16191,16194,16291],{"className":16190},[36],[15,16192],{"className":16193},[67,1410],[15,16195,16197],{"className":16196},[1414],[15,16198,16200,16282],{"className":16199},[365,366],[15,16201,16203,16279],{"className":16202},[370],[15,16204,16206,16257,16265],{"className":16205,"style":7814},[374],[15,16207,16208,16211],{"style":1427},[15,16209],{"className":16210,"style":1431},[382],[15,16212,16214,16217],{"className":16213},[36],[15,16215,12081],{"className":16216,"style":12080},[36],[15,16218,16220,16223],{"className":16219},[36],[15,16221,3548],{"className":16222,"style":3547},[36,37],[15,16224,16226],{"className":16225},[361],[15,16227,16229,16249],{"className":16228},[365,366],[15,16230,16232,16246],{"className":16231},[370],[15,16233,16235],{"className":16234,"style":3511},[374],[15,16236,16237,16240],{"style":3695},[15,16238],{"className":16239,"style":383},[382],[15,16241,16243],{"className":16242},[387,388,389,390],[15,16244,16159],{"className":16245},[36,37,390],[15,16247,399],{"className":16248},[398],[15,16250,16252],{"className":16251},[370],[15,16253,16255],{"className":16254,"style":406},[374],[15,16256],{},[15,16258,16259,16262],{"style":1488},[15,16260],{"className":16261,"style":1431},[382],[15,16263],{"className":16264,"style":1496},[1495],[15,16266,16267,16270],{"style":1499},[15,16268],{"className":16269,"style":1431},[382],[15,16271,16273,16276],{"className":16272},[36],[15,16274,12081],{"className":16275,"style":12080},[36],[15,16277,12016],{"className":16278},[36,12143],[15,16280,399],{"className":16281},[398],[15,16283,16285],{"className":16284},[370],[15,16286,16289],{"className":16287,"style":16288},[374],"height:0.836em;",[15,16290],{},[15,16292],{"className":16293},[103,1410],[15,16295],{"className":16296,"style":43},[42],[15,16298,48],{"className":16299},[47],[15,16301],{"className":16302,"style":43},[42],[15,16304,16306,16309,16377,16380,16383],{"className":16305},[27],[15,16307],{"className":16308,"style":7795},[31],[15,16310,16312,16315,16374],{"className":16311},[36],[15,16313],{"className":16314},[67,1410],[15,16316,16318],{"className":16317},[1414],[15,16319,16321,16366],{"className":16320},[365,366],[15,16322,16324,16363],{"className":16323},[370],[15,16325,16327,16341,16349],{"className":16326,"style":7814},[374],[15,16328,16329,16332],{"style":1427},[15,16330],{"className":16331,"style":1431},[382],[15,16333,16335,16338],{"className":16334},[36],[15,16336,12081],{"className":16337,"style":12080},[36],[15,16339,38],{"className":16340},[36,37],[15,16342,16343,16346],{"style":1488},[15,16344],{"className":16345,"style":1431},[382],[15,16347],{"className":16348,"style":1496},[1495],[15,16350,16351,16354],{"style":1499},[15,16352],{"className":16353,"style":1431},[382],[15,16355,16357,16360],{"className":16356},[36],[15,16358,12081],{"className":16359,"style":12080},[36],[15,16361,12016],{"className":16362},[36,12143],[15,16364,399],{"className":16365},[398],[15,16367,16369],{"className":16368},[370],[15,16370,16372],{"className":16371,"style":7858},[374],[15,16373],{},[15,16375],{"className":16376},[103,1410],[15,16378],{"className":16379,"style":81},[42],[15,16381,4210],{"className":16382},[85],[15,16384],{"className":16385,"style":81},[42],[15,16387,16389,16392,16424,16427,16430,16433,16436,16439],{"className":16388},[27],[15,16390],{"className":16391,"style":4045},[31],[15,16393,16395,16398],{"className":16394},[36],[15,16396,63],{"className":16397,"style":62},[36,37],[15,16399,16401],{"className":16400},[361],[15,16402,16404],{"className":16403},[365],[15,16405,16407],{"className":16406},[370],[15,16408,16410],{"className":16409,"style":4064},[374],[15,16411,16412,16415],{"style":4067},[15,16413],{"className":16414,"style":383},[382],[15,16416,16418],{"className":16417},[387,388,389,390],[15,16419,16421],{"className":16420},[36,390],[15,16422,1215],{"className":16423},[36,390],[15,16425,68],{"className":16426},[67],[15,16428,230],{"className":16429,"style":229},[36,37],[15,16431,104],{"className":16432},[103],[15,16434],{"className":16435,"style":81},[42],[15,16437,4210],{"className":16438},[85],[15,16440],{"className":16441,"style":81},[42],[15,16443,16445,16448,16488,16491,16494],{"className":16444},[27],[15,16446],{"className":16447,"style":11905},[31],[15,16449,16451,16454],{"className":16450},[36],[15,16452,77],{"className":16453},[36,37],[15,16455,16457],{"className":16456},[361],[15,16458,16460,16480],{"className":16459},[365,366],[15,16461,16463,16477],{"className":16462},[370],[15,16464,16466],{"className":16465,"style":3511},[374],[15,16467,16468,16471],{"style":378},[15,16469],{"className":16470,"style":383},[382],[15,16472,16474],{"className":16473},[387,388,389,390],[15,16475,16159],{"className":16476},[36,37,390],[15,16478,399],{"className":16479},[398],[15,16481,16483],{"className":16482},[370],[15,16484,16486],{"className":16485,"style":406},[374],[15,16487],{},[15,16489],{"className":16490,"style":43},[42],[15,16492,48],{"className":16493},[47],[15,16495],{"className":16496,"style":43},[42],[15,16498,16500,16503,16571,16574,16577],{"className":16499},[27],[15,16501],{"className":16502,"style":7795},[31],[15,16504,16506,16509,16568],{"className":16505},[36],[15,16507],{"className":16508},[67,1410],[15,16510,16512],{"className":16511},[1414],[15,16513,16515,16560],{"className":16514},[365,366],[15,16516,16518,16557],{"className":16517},[370],[15,16519,16521,16535,16543],{"className":16520,"style":7814},[374],[15,16522,16523,16526],{"style":1427},[15,16524],{"className":16525,"style":1431},[382],[15,16527,16529,16532],{"className":16528},[36],[15,16530,12081],{"className":16531,"style":12080},[36],[15,16533,38],{"className":16534},[36,37],[15,16536,16537,16540],{"style":1488},[15,16538],{"className":16539,"style":1431},[382],[15,16541],{"className":16542,"style":1496},[1495],[15,16544,16545,16548],{"style":1499},[15,16546],{"className":16547,"style":1431},[382],[15,16549,16551,16554],{"className":16550},[36],[15,16552,12081],{"className":16553,"style":12080},[36],[15,16555,12016],{"className":16556},[36,12143],[15,16558,399],{"className":16559},[398],[15,16561,16563],{"className":16562},[370],[15,16564,16566],{"className":16565,"style":7858},[374],[15,16567],{},[15,16569],{"className":16570},[103,1410],[15,16572],{"className":16573,"style":81},[42],[15,16575,4210],{"className":16576},[85],[15,16578],{"className":16579,"style":81},[42],[15,16581,16583,16586,16589,16592,16595,16598,16601],{"className":16582},[27],[15,16584],{"className":16585,"style":58},[31],[15,16587,497],{"className":16588},[36,2213],[15,16590,2217],{"className":16591},[67],[15,16593,230],{"className":16594,"style":229},[36,37],[15,16596],{"className":16597,"style":43},[42],[15,16599,2227],{"className":16600},[47],[15,16602],{"className":16603,"style":43},[42],[15,16605,16607,16610,16613,16616,16619,16622],{"className":16606},[27],[15,16608],{"className":16609,"style":58},[31],[15,16611,1589],{"className":16612},[36],[15,16614,2243],{"className":16615},[103],[15,16617],{"className":16618,"style":81},[42],[15,16620,4210],{"className":16621},[85],[15,16623],{"className":16624,"style":81},[42],[15,16626,16628,16631,16671],{"className":16627},[27],[15,16629],{"className":16630,"style":11905},[31],[15,16632,16634,16637],{"className":16633},[36],[15,16635,77],{"className":16636},[36,37],[15,16638,16640],{"className":16639},[361],[15,16641,16643,16663],{"className":16642},[365,366],[15,16644,16646,16660],{"className":16645},[370],[15,16647,16649],{"className":16648,"style":3511},[374],[15,16650,16651,16654],{"style":378},[15,16652],{"className":16653,"style":383},[382],[15,16655,16657],{"className":16656},[387,388,389,390],[15,16658,16159],{"className":16659},[36,37,390],[15,16661,399],{"className":16662},[398],[15,16664,16666],{"className":16665},[370],[15,16667,16669],{"className":16668,"style":406},[374],[15,16670],{},[15,16672,4730],{"className":16673},[36],[11,16675,16676,16677,16710,16711,16753,16754,16769,16770,16853,16854,8170,16869,16884],{},"Once ",[15,16678,16680],{"className":16679},[18],[15,16681,16683,16701],{"className":16682,"ariaHidden":23},[22],[15,16684,16686,16689,16692,16695,16698],{"className":16685},[27],[15,16687],{"className":16688,"style":2665},[31],[15,16690,230],{"className":16691,"style":229},[36,37],[15,16693],{"className":16694,"style":43},[42],[15,16696,8033],{"className":16697},[47],[15,16699],{"className":16700,"style":43},[42],[15,16702,16704,16707],{"className":16703},[27],[15,16705],{"className":16706,"style":2147},[31],[15,16708,1589],{"className":16709},[36]," on every example, the factor ",[15,16712,16714],{"className":16713},[18],[15,16715,16717,16741],{"className":16716,"ariaHidden":23},[22],[15,16718,16720,16723,16726,16729,16732,16735,16738],{"className":16719},[27],[15,16721],{"className":16722,"style":58},[31],[15,16724,497],{"className":16725},[36,2213],[15,16727,2217],{"className":16728},[67],[15,16730,230],{"className":16731,"style":229},[36,37],[15,16733],{"className":16734,"style":43},[42],[15,16736,2227],{"className":16737},[47],[15,16739],{"className":16740,"style":43},[42],[15,16742,16744,16747,16750],{"className":16743},[27],[15,16745],{"className":16746,"style":58},[31],[15,16748,1589],{"className":16749},[36],[15,16751,2243],{"className":16752},[103]," is ",[15,16755,16757],{"className":16756},[18],[15,16758,16760],{"className":16759,"ariaHidden":23},[22],[15,16761,16763,16766],{"className":16762},[27],[15,16764],{"className":16765,"style":2147},[31],[15,16767,1589],{"className":16768},[36]," for every one, so\n",[15,16771,16773],{"className":16772},[18],[15,16774,16776,16844],{"className":16775,"ariaHidden":23},[22],[15,16777,16779,16782,16785,16788,16792,16795,16835,16838,16841],{"className":16778},[27],[15,16780],{"className":16781,"style":58},[31],[15,16783,12081],{"className":16784,"style":12080},[36],[15,16786,12016],{"className":16787},[36,12143],[15,16789,16791],{"className":16790},[36],"\u002F",[15,16793,12081],{"className":16794,"style":12080},[36],[15,16796,16798,16801],{"className":16797},[36],[15,16799,3548],{"className":16800,"style":3547},[36,37],[15,16802,16804],{"className":16803},[361],[15,16805,16807,16827],{"className":16806},[365,366],[15,16808,16810,16824],{"className":16809},[370],[15,16811,16813],{"className":16812,"style":3511},[374],[15,16814,16815,16818],{"style":3695},[15,16816],{"className":16817,"style":383},[382],[15,16819,16821],{"className":16820},[387,388,389,390],[15,16822,16159],{"className":16823},[36,37,390],[15,16825,399],{"className":16826},[398],[15,16828,16830],{"className":16829},[370],[15,16831,16833],{"className":16832,"style":406},[374],[15,16834],{},[15,16836],{"className":16837,"style":43},[42],[15,16839,48],{"className":16840},[47],[15,16842],{"className":16843,"style":43},[42],[15,16845,16847,16850],{"className":16846},[27],[15,16848],{"className":16849,"style":2147},[31],[15,16851,1589],{"className":16852},[36]," regardless of the upstream gradient or the\ninputs. Nothing in the update can move ",[15,16855,16857],{"className":16856},[18],[15,16858,16860],{"className":16859,"ariaHidden":23},[22],[15,16861,16863,16866],{"className":16862},[27],[15,16864],{"className":16865,"style":225},[31],[15,16867,3548],{"className":16868,"style":3547},[36,37],[15,16870,16872],{"className":16871},[18],[15,16873,16875],{"className":16874,"ariaHidden":23},[22],[15,16876,16878,16881],{"className":16877},[27],[15,16879],{"className":16880,"style":32},[31],[15,16882,99],{"className":16883},[36,37],", so the unit cannot recover:\nthe failure is irreversible.",[11985,16886],{"hash":16887},"b849c8d32183bc5ecebac9a03d43ab42e5cdbc6624322791077f620207aeea26",[11,16889,16890],{},"The fixes all do the same thing: give the negative side a nonzero slope so a stuck\nunit still receives a gradient and can recover.",[1278,16892,16893,16942],{},[1281,16894,16895],{},[1284,16896,16897,16900,16903,16939],{},[1287,16898,16899],{},"unit",[1287,16901,16902],{},"negative branch",[1287,16904,16905,16906],{},"derivative at ",[15,16907,16909],{"className":16908},[18],[15,16910,16912,16930],{"className":16911,"ariaHidden":23},[22],[15,16913,16915,16918,16921,16924,16927],{"className":16914},[27],[15,16916],{"className":16917,"style":2665},[31],[15,16919,230],{"className":16920,"style":229},[36,37],[15,16922],{"className":16923,"style":43},[42],[15,16925,8033],{"className":16926},[47],[15,16928],{"className":16929,"style":43},[42],[15,16931,16933,16936],{"className":16932},[27],[15,16934],{"className":16935,"style":2147},[31],[15,16937,1589],{"className":16938},[36],[1287,16940,16941],{},"recovers?",[1381,16943,16944,17000,17083,17142,17286],{},[1284,16945,16946,16948,16965,16982],{},[1386,16947,2155],{},[1386,16949,16950],{},[15,16951,16953],{"className":16952},[18],[15,16954,16956],{"className":16955,"ariaHidden":23},[22],[15,16957,16959,16962],{"className":16958},[27],[15,16960],{"className":16961,"style":2147},[31],[15,16963,1589],{"className":16964},[36],[1386,16966,16967],{},[15,16968,16970],{"className":16969},[18],[15,16971,16973],{"className":16972,"ariaHidden":23},[22],[15,16974,16976,16979],{"className":16975},[27],[15,16977],{"className":16978,"style":2147},[31],[15,16980,1589],{"className":16981},[36],[1386,16983,16984,16985],{},"no — gradient is ",[15,16986,16988],{"className":16987},[18],[15,16989,16991],{"className":16990,"ariaHidden":23},[22],[15,16992,16994,16997],{"className":16993},[27],[15,16995],{"className":16996,"style":2147},[31],[15,16998,1589],{"className":16999},[36],[1284,17001,17002,17004,17063,17080],{},[1386,17003,2287],{},[1386,17005,17006,10072,17024],{},[15,17007,17009],{"className":17008},[18],[15,17010,17012],{"className":17011,"ariaHidden":23},[22],[15,17013,17015,17018,17021],{"className":17014},[27],[15,17016],{"className":17017,"style":225},[31],[15,17019,2315],{"className":17020,"style":2314},[36,37],[15,17022,230],{"className":17023,"style":229},[36,37],[15,17025,17027],{"className":17026},[18],[15,17028,17030,17054],{"className":17029,"ariaHidden":23},[22],[15,17031,17033,17036,17039,17042,17045,17048,17051],{"className":17032},[27],[15,17034],{"className":17035,"style":225},[31],[15,17037,2315],{"className":17038,"style":2314},[36,37],[15,17040],{"className":17041,"style":2485},[42],[15,17043],{"className":17044,"style":43},[42],[15,17046,48],{"className":17047},[47],[15,17049],{"className":17050,"style":2485},[42],[15,17052],{"className":17053,"style":43},[42],[15,17055,17057,17060],{"className":17056},[27],[15,17058],{"className":17059,"style":2147},[31],[15,17061,2508],{"className":17062},[36],[1386,17064,17065],{},[15,17066,17068],{"className":17067},[18],[15,17069,17071],{"className":17070,"ariaHidden":23},[22],[15,17072,17074,17077],{"className":17073},[27],[15,17075],{"className":17076,"style":225},[31],[15,17078,2315],{"className":17079,"style":2314},[36,37],[1386,17081,17082],{},"yes — small constant slope",[1284,17084,17085,17087,17122,17139],{},[1386,17086,2514],{},[1386,17088,17089,10072,17107,2574],{},[15,17090,17092],{"className":17091},[18],[15,17093,17095],{"className":17094,"ariaHidden":23},[22],[15,17096,17098,17101,17104],{"className":17097},[27],[15,17099],{"className":17100,"style":225},[31],[15,17102,2315],{"className":17103,"style":2314},[36,37],[15,17105,230],{"className":17106,"style":229},[36,37],[15,17108,17110],{"className":17109},[18],[15,17111,17113],{"className":17112,"ariaHidden":23},[22],[15,17114,17116,17119],{"className":17115},[27],[15,17117],{"className":17118,"style":225},[31],[15,17120,2315],{"className":17121,"style":2314},[36,37],[1386,17123,17124],{},[15,17125,17127],{"className":17126},[18],[15,17128,17130],{"className":17129,"ariaHidden":23},[22],[15,17131,17133,17136],{"className":17132},[27],[15,17134],{"className":17135,"style":225},[31],[15,17137,2315],{"className":17138,"style":2314},[36,37],[1386,17140,17141],{},"yes — slope adapts per channel",[1284,17143,17144,17146,17219,17268],{},[1386,17145,2634],{},[1386,17147,17148],{},[15,17149,17151],{"className":17150},[18],[15,17152,17154,17207],{"className":17153,"ariaHidden":23},[22],[15,17155,17157,17160,17163,17166,17198,17201,17204],{"className":17156},[27],[15,17158],{"className":17159,"style":58},[31],[15,17161,2315],{"className":17162,"style":2314},[36,37],[15,17164,68],{"className":17165},[67],[15,17167,17169,17172],{"className":17168},[36],[15,17170,1453],{"className":17171},[36,37],[15,17173,17175],{"className":17174},[361],[15,17176,17178],{"className":17177},[365],[15,17179,17181],{"className":17180},[370],[15,17182,17184],{"className":17183,"style":1903},[374],[15,17185,17186,17189],{"style":1202},[15,17187],{"className":17188,"style":383},[382],[15,17190,17192],{"className":17191},[387,388,389,390],[15,17193,17195],{"className":17194},[36,390],[15,17196,230],{"className":17197,"style":229},[36,37,390],[15,17199],{"className":17200,"style":81},[42],[15,17202,1482],{"className":17203},[85],[15,17205],{"className":17206,"style":81},[42],[15,17208,17210,17213,17216],{"className":17209},[27],[15,17211],{"className":17212,"style":58},[31],[15,17214,497],{"className":17215},[36],[15,17217,104],{"className":17218},[103],[1386,17220,17221],{},[15,17222,17224],{"className":17223},[18],[15,17225,17227],{"className":17226,"ariaHidden":23},[22],[15,17228,17230,17233,17236],{"className":17229},[27],[15,17231],{"className":17232,"style":1903},[31],[15,17234,2315],{"className":17235,"style":2314},[36,37],[15,17237,17239,17242],{"className":17238},[36],[15,17240,1453],{"className":17241},[36,37],[15,17243,17245],{"className":17244},[361],[15,17246,17248],{"className":17247},[365],[15,17249,17251],{"className":17250},[370],[15,17252,17254],{"className":17253,"style":1903},[374],[15,17255,17256,17259],{"style":1202},[15,17257],{"className":17258,"style":383},[382],[15,17260,17262],{"className":17261},[387,388,389,390],[15,17263,17265],{"className":17264},[36,390],[15,17266,230],{"className":17267,"style":229},[36,37,390],[1386,17269,17270,17271],{},"yes — smooth, mean toward ",[15,17272,17274],{"className":17273},[18],[15,17275,17277],{"className":17276,"ariaHidden":23},[22],[15,17278,17280,17283],{"className":17279},[27],[15,17281],{"className":17282,"style":2147},[31],[15,17284,1589],{"className":17285},[36],[1284,17287,17288,17290,17322,17378],{},[1386,17289,2916],{},[1386,17291,17292],{},[15,17293,17295],{"className":17294},[18],[15,17296,17298],{"className":17297,"ariaHidden":23},[22],[15,17299,17301,17304,17307,17310,17313,17316,17319],{"className":17300},[27],[15,17302],{"className":17303,"style":58},[31],[15,17305,230],{"className":17306,"style":229},[36,37],[15,17308],{"className":17309,"style":1598},[42],[15,17311,2940],{"className":17312},[36],[15,17314,68],{"className":17315},[67],[15,17317,230],{"className":17318,"style":229},[36,37],[15,17320,104],{"className":17321},[103],[1386,17323,17324],{},[15,17325,17327],{"className":17326},[18],[15,17328,17330,17357],{"className":17329,"ariaHidden":23},[22],[15,17331,17333,17336,17339,17342,17345,17348,17351,17354],{"className":17332},[27],[15,17334],{"className":17335,"style":58},[31],[15,17337,2940],{"className":17338},[36],[15,17340,68],{"className":17341},[67],[15,17343,230],{"className":17344,"style":229},[36,37],[15,17346,104],{"className":17347},[103],[15,17349],{"className":17350,"style":81},[42],[15,17352,86],{"className":17353},[85],[15,17355],{"className":17356,"style":81},[42],[15,17358,17360,17363,17366,17369,17372,17375],{"className":17359},[27],[15,17361],{"className":17362,"style":58},[31],[15,17364,230],{"className":17365,"style":229},[36,37],[15,17367,3000],{"className":17368},[36,37],[15,17370,68],{"className":17371},[67],[15,17373,230],{"className":17374,"style":229},[36,37],[15,17376,104],{"className":17377},[103],[1386,17379,17380,17381,17396],{},"yes — smooth, near-",[15,17382,17384],{"className":17383},[18],[15,17385,17387],{"className":17386,"ariaHidden":23},[22],[15,17388,17390,17393],{"className":17389},[27],[15,17391],{"className":17392,"style":2147},[31],[15,17394,1589],{"className":17395},[36]," slope",[11,17398,17399,17400,17415,17416,17505,17506,17521,17522],{},"Leaky ReLU and PReLU keep a literal straight line of slope ",[15,17401,17403],{"className":17402},[18],[15,17404,17406],{"className":17405,"ariaHidden":23},[22],[15,17407,17409,17412],{"className":17408},[27],[15,17410],{"className":17411,"style":225},[31],[15,17413,2315],{"className":17414,"style":2314},[36,37]," on the left so\n",[15,17417,17419],{"className":17418},[18],[15,17420,17422,17478,17496],{"className":17421,"ariaHidden":23},[22],[15,17423,17425,17428,17460,17463,17466,17469,17472,17475],{"className":17424},[27],[15,17426],{"className":17427,"style":1180},[31],[15,17429,17431,17434],{"className":17430},[36],[15,17432,63],{"className":17433,"style":62},[36,37],[15,17435,17437],{"className":17436},[361],[15,17438,17440],{"className":17439},[365],[15,17441,17443],{"className":17442},[370],[15,17444,17446],{"className":17445,"style":1199},[374],[15,17447,17448,17451],{"style":1202},[15,17449],{"className":17450,"style":383},[382],[15,17452,17454],{"className":17453},[387,388,389,390],[15,17455,17457],{"className":17456},[36,390],[15,17458,1215],{"className":17459},[36,390],[15,17461,68],{"className":17462},[67],[15,17464,230],{"className":17465,"style":229},[36,37],[15,17467,104],{"className":17468},[103],[15,17470],{"className":17471,"style":43},[42],[15,17473,48],{"className":17474},[47],[15,17476],{"className":17477,"style":43},[42],[15,17479,17481,17484,17487,17490,17493],{"className":17480},[27],[15,17482],{"className":17483,"style":2665},[31],[15,17485,2315],{"className":17486,"style":2314},[36,37],[15,17488],{"className":17489,"style":43},[42],[15,17491,2227],{"className":17492},[47],[15,17494],{"className":17495,"style":43},[42],[15,17497,17499,17502],{"className":17498},[27],[15,17500],{"className":17501,"style":2147},[31],[15,17503,1589],{"className":17504},[36]," everywhere; a dead-looking unit still receives an ",[15,17507,17509],{"className":17508},[18],[15,17510,17512],{"className":17511,"ariaHidden":23},[22],[15,17513,17515,17518],{"className":17514},[27],[15,17516],{"className":17517,"style":225},[31],[15,17519,2315],{"className":17520,"style":2314},[36,37],"-scaled\ngradient and can recover. ELU and GELU replace the kink with a smooth curve whose\nnegative-side derivative is small but strictly positive, which additionally pushes\nthe activation mean toward zero, a self-normalizing effect that helps the next\nlayer's pre-activations stay near the high-gradient region.",[3874,17523,17524],{},[107,17525,1705],{"href":17526,"ariaDescribedBy":17527,"dataFootnoteRef":6,"id":17528},"#user-content-fn-chollet-act",[3880],"user-content-fnref-chollet-act",[11985,17530],{"hash":17531},"62319806674704a007bcd447462d8ac34a96595c69e86c30045eb2a4b216ef73",[1161,17533,17535],{"id":17534},"output-units-not-hidden-units","Output units, not hidden units",[11,17537,17538,17539,17541,17542,17546],{},"The hidden-layer rules above optimize for gradient flow. The ",[152,17540,5271],{}," unit is\nchosen instead to match the target's range and the loss, and the saturating units\nthat fail in hidden layers become the right choice at the output, where they pair\nwith a log-loss that cancels the saturation. The output choice is tied to the\n",[107,17543,17545],{"href":17544},"\u002Fdeep-learning\u002Fneural-networks\u002Floss-functions-and-output-units","loss","; the table\nbelow is the standard pairing.",[1278,17548,17549,17564],{},[1281,17550,17551],{},[1284,17552,17553,17556,17559,17561],{},[1287,17554,17555],{},"task",[1287,17557,17558],{},"output activation",[1287,17560,1373],{},[1287,17562,17563],{},"paired loss",[1381,17565,17566,17594,17636,17665],{},[1284,17567,17568,17571,17574,17591],{},[1386,17569,17570],{},"regression",[1386,17572,17573],{},"none (linear)",[1386,17575,17576],{},[15,17577,17579],{"className":17578},[18],[15,17580,17582],{"className":17581,"ariaHidden":23},[22],[15,17583,17585,17588],{"className":17584},[27],[15,17586],{"className":17587,"style":187},[31],[15,17589,192],{"className":17590},[36,191],[1386,17592,17593],{},"mean squared error",[1284,17595,17596,17599,17601,17633],{},[1386,17597,17598],{},"binary classification",[1386,17600,1388],{},[1386,17602,17603],{},[15,17604,17606],{"className":17605},[18],[15,17607,17609],{"className":17608,"ariaHidden":23},[22],[15,17610,17612,17615,17618,17621,17624,17627,17630],{"className":17611},[27],[15,17613],{"className":17614,"style":58},[31],[15,17616,68],{"className":17617},[67],[15,17619,1589],{"className":17620},[36],[15,17622,1594],{"className":17623},[1593],[15,17625],{"className":17626,"style":1598},[42],[15,17628,497],{"className":17629},[36],[15,17631,104],{"className":17632},[103],[1386,17634,17635],{},"binary cross-entropy",[1284,17637,17638,17641,17644,17662],{},[1386,17639,17640],{},"multiclass",[1386,17642,17643],{},"softmax",[1386,17645,17646,17647],{},"simplex, sums to ",[15,17648,17650],{"className":17649},[18],[15,17651,17653],{"className":17652,"ariaHidden":23},[22],[15,17654,17656,17659],{"className":17655},[27],[15,17657],{"className":17658,"style":2147},[31],[15,17660,497],{"className":17661},[36],[1386,17663,17664],{},"categorical cross-entropy",[1284,17666,17667,17670,17673,17705],{},[1386,17668,17669],{},"count \u002F nonnegative",[1386,17671,17672],{},"softplus or exp",[1386,17674,17675],{},[15,17676,17678],{"className":17677},[18],[15,17679,17681],{"className":17680,"ariaHidden":23},[22],[15,17682,17684,17687,17690,17693,17696,17699,17702],{"className":17683},[27],[15,17685],{"className":17686,"style":58},[31],[15,17688,68],{"className":17689},[67],[15,17691,1589],{"className":17692},[36],[15,17694,1594],{"className":17695},[1593],[15,17697],{"className":17698,"style":1598},[42],[15,17700,2273],{"className":17701},[36],[15,17703,104],{"className":17704},[103],[1386,17706,17707],{},"Poisson \u002F MSE",[146,17709,17710],{"type":148},[11,17711,17712,17715,17716,17990,17991,18008,18009,4730,18087],{},[152,17713,17714],{},"Definition (Softmax)."," The vector activation\n",[15,17717,17719],{"className":17718},[18],[15,17720,17722,17789],{"className":17721,"ariaHidden":23},[22],[15,17723,17725,17728,17734,17737,17740,17780,17783,17786],{"className":17724},[27],[15,17726],{"className":17727,"style":58},[31],[15,17729,17731],{"className":17730},[2014],[15,17732,17643],{"className":17733},[36,2176],[15,17735,68],{"className":17736},[67],[15,17738,230],{"className":17739,"style":229},[36,37],[15,17741,17743,17746],{"className":17742},[103],[15,17744,104],{"className":17745},[103],[15,17747,17749],{"className":17748},[361],[15,17750,17752,17772],{"className":17751},[365,366],[15,17753,17755,17769],{"className":17754},[370],[15,17756,17758],{"className":17757,"style":3511},[374],[15,17759,17760,17763],{"style":378},[15,17761],{"className":17762,"style":383},[382],[15,17764,17766],{"className":17765},[387,388,389,390],[15,17767,16159],{"className":17768},[36,37,390],[15,17770,399],{"className":17771},[398],[15,17773,17775],{"className":17774},[370],[15,17776,17778],{"className":17777,"style":406},[374],[15,17779],{},[15,17781],{"className":17782,"style":43},[42],[15,17784,48],{"className":17785},[47],[15,17787],{"className":17788,"style":43},[42],[15,17790,17792,17796,17868,17871,17874,17917,17920],{"className":17791},[27],[15,17793],{"className":17794,"style":17795},[31],"height:1.1858em;vertical-align:-0.4358em;",[15,17797,17799,17802],{"className":17798},[36],[15,17800,1453],{"className":17801},[36,37],[15,17803,17805],{"className":17804},[361],[15,17806,17808],{"className":17807},[365],[15,17809,17811],{"className":17810},[370],[15,17812,17814],{"className":17813,"style":1903},[374],[15,17815,17816,17819],{"style":1202},[15,17817],{"className":17818,"style":383},[382],[15,17820,17822],{"className":17821},[387,388,389,390],[15,17823,17825],{"className":17824},[36,390],[15,17826,17828,17831],{"className":17827},[36,390],[15,17829,230],{"className":17830,"style":229},[36,37,390],[15,17832,17834],{"className":17833},[361],[15,17835,17837,17859],{"className":17836},[365,366],[15,17838,17840,17856],{"className":17839},[370],[15,17841,17844],{"className":17842,"style":17843},[374],"height:0.3281em;",[15,17845,17847,17850],{"style":17846},"top:-2.357em;margin-left:-0.044em;margin-right:0.0714em;",[15,17848],{"className":17849,"style":3730},[382],[15,17851,17853],{"className":17852},[387,3734,3735,390],[15,17854,16159],{"className":17855},[36,37,390],[15,17857,399],{"className":17858},[398],[15,17860,17862],{"className":17861},[370],[15,17863,17866],{"className":17864,"style":17865},[374],"height:0.143em;",[15,17867],{},[15,17869,16791],{"className":17870},[36],[15,17872],{"className":17873,"style":1598},[42],[15,17875,17877,17881],{"className":17876},[2014],[15,17878,17880],{"className":17879,"style":12198},[2014,12196,12197],"∑",[15,17882,17884],{"className":17883},[361],[15,17885,17887,17908],{"className":17886},[365,366],[15,17888,17890,17905],{"className":17889},[370],[15,17891,17894],{"className":17892,"style":17893},[374],"height:0.162em;",[15,17895,17896,17899],{"style":12215},[15,17897],{"className":17898,"style":383},[382],[15,17900,17902],{"className":17901},[387,388,389,390],[15,17903,3525],{"className":17904,"style":3524},[36,37,390],[15,17906,399],{"className":17907},[398],[15,17909,17911],{"className":17910},[370],[15,17912,17915],{"className":17913,"style":17914},[374],"height:0.4358em;",[15,17916],{},[15,17918],{"className":17919,"style":1598},[42],[15,17921,17923,17926],{"className":17922},[36],[15,17924,1453],{"className":17925},[36,37],[15,17927,17929],{"className":17928},[361],[15,17930,17932],{"className":17931},[365],[15,17933,17935],{"className":17934},[370],[15,17936,17938],{"className":17937,"style":1903},[374],[15,17939,17940,17943],{"style":1202},[15,17941],{"className":17942,"style":383},[382],[15,17944,17946],{"className":17945},[387,388,389,390],[15,17947,17949],{"className":17948},[36,390],[15,17950,17952,17955],{"className":17951},[36,390],[15,17953,230],{"className":17954,"style":229},[36,37,390],[15,17956,17958],{"className":17957},[361],[15,17959,17961,17981],{"className":17960},[365,366],[15,17962,17964,17978],{"className":17963},[370],[15,17965,17967],{"className":17966,"style":17843},[374],[15,17968,17969,17972],{"style":17846},[15,17970],{"className":17971,"style":3730},[382],[15,17973,17975],{"className":17974},[387,3734,3735,390],[15,17976,3525],{"className":17977,"style":3524},[36,37,390],[15,17979,399],{"className":17980},[398],[15,17982,17984],{"className":17983},[370],[15,17985,17988],{"className":17986,"style":17987},[374],"height:0.2819em;",[15,17989],{},", mapping a vector of logits\nto a probability distribution over ",[15,17992,17994],{"className":17993},[18],[15,17995,17997],{"className":17996,"ariaHidden":23},[22],[15,17998,18000,18003],{"className":17999},[27],[15,18001],{"className":18002,"style":3074},[31],[15,18004,18007],{"className":18005,"style":18006},[36,37],"margin-right:0.0715em;","K"," classes. It is the multiclass\ngeneralization of the sigmoid, and pairs with cross-entropy so the\nlog cancels the exponential and leaves the clean gradient ",[15,18010,18012],{"className":18011},[18],[15,18013,18015,18078],{"className":18014,"ariaHidden":23},[22],[15,18016,18018,18021,18069,18072,18075],{"className":18017},[27],[15,18019],{"className":18020,"style":3844},[31],[15,18022,18025],{"className":18023},[36,18024],"accent",[15,18026,18028,18060],{"className":18027},[365,366],[15,18029,18031,18057],{"className":18030},[370],[15,18032,18034,18043],{"className":18033,"style":32},[374],[15,18035,18036,18039],{"style":10560},[15,18037],{"className":18038,"style":1431},[382],[15,18040,18042],{"className":18041,"style":62},[36,37],"y",[15,18044,18045,18048],{"style":10560},[15,18046],{"className":18047,"style":1431},[382],[15,18049,18053],{"className":18050,"style":18052},[18051],"accent-body","left:-0.1944em;",[15,18054,18056],{"className":18055},[36],"^",[15,18058,399],{"className":18059},[398],[15,18061,18063],{"className":18062},[370],[15,18064,18067],{"className":18065,"style":18066},[374],"height:0.1944em;",[15,18068],{},[15,18070],{"className":18071,"style":81},[42],[15,18073,1482],{"className":18074},[85],[15,18076],{"className":18077,"style":81},[42],[15,18079,18081,18084],{"className":18080},[27],[15,18082],{"className":18083,"style":123},[31],[15,18085,18042],{"className":18086,"style":62},[36,37],[3874,18088,18089],{},[107,18090,14470],{"href":18091,"ariaDescribedBy":18092,"dataFootnoteRef":6,"id":18093},"#user-content-fn-gf-output",[3880],"user-content-fnref-gf-output",[11,18095,18096,18097,18181,18182,18393,18394,18409],{},"The cancellation is worth seeing once. With softmax output ",[15,18098,18100],{"className":18099},[18],[15,18101,18103,18160],{"className":18102,"ariaHidden":23},[22],[15,18104,18106,18109,18151,18154,18157],{"className":18105},[27],[15,18107],{"className":18108,"style":3844},[31],[15,18110,18112],{"className":18111},[36,18024],[15,18113,18115,18143],{"className":18114},[365,366],[15,18116,18118,18140],{"className":18117},[370],[15,18119,18121,18129],{"className":18120,"style":32},[374],[15,18122,18123,18126],{"style":10560},[15,18124],{"className":18125,"style":1431},[382],[15,18127,18042],{"className":18128,"style":62},[36,37],[15,18130,18131,18134],{"style":10560},[15,18132],{"className":18133,"style":1431},[382],[15,18135,18137],{"className":18136,"style":18052},[18051],[15,18138,18056],{"className":18139},[36],[15,18141,399],{"className":18142},[398],[15,18144,18146],{"className":18145},[370],[15,18147,18149],{"className":18148,"style":18066},[374],[15,18150],{},[15,18152],{"className":18153,"style":43},[42],[15,18155,48],{"className":18156},[47],[15,18158],{"className":18159,"style":43},[42],[15,18161,18163,18166,18172,18175,18178],{"className":18162},[27],[15,18164],{"className":18165,"style":58},[31],[15,18167,18169],{"className":18168},[2014],[15,18170,17643],{"className":18171},[36,2176],[15,18173,68],{"className":18174},[67],[15,18176,230],{"className":18177,"style":229},[36,37],[15,18179,104],{"className":18180},[103]," and cross-entropy loss ",[15,18183,18185],{"className":18184},[18],[15,18186,18188,18206],{"className":18187,"ariaHidden":23},[22],[15,18189,18191,18194,18197,18200,18203],{"className":18190},[27],[15,18192],{"className":18193,"style":3074},[31],[15,18195,12016],{"className":18196},[36,12143],[15,18198],{"className":18199,"style":43},[42],[15,18201,48],{"className":18202},[47],[15,18204],{"className":18205,"style":43},[42],[15,18207,18209,18212,18215,18218,18258,18261,18302,18305,18311,18314],{"className":18208},[27],[15,18210],{"className":18211,"style":15524},[31],[15,18213,1482],{"className":18214},[36],[15,18216],{"className":18217,"style":1598},[42],[15,18219,18221,18224],{"className":18220},[2014],[15,18222,17880],{"className":18223,"style":12198},[2014,12196,12197],[15,18225,18227],{"className":18226},[361],[15,18228,18230,18250],{"className":18229},[365,366],[15,18231,18233,18247],{"className":18232},[370],[15,18234,18236],{"className":18235,"style":17893},[374],[15,18237,18238,18241],{"style":12215},[15,18239],{"className":18240,"style":383},[382],[15,18242,18244],{"className":18243},[387,388,389,390],[15,18245,16159],{"className":18246},[36,37,390],[15,18248,399],{"className":18249},[398],[15,18251,18253],{"className":18252},[370],[15,18254,18256],{"className":18255,"style":15564},[374],[15,18257],{},[15,18259],{"className":18260,"style":1598},[42],[15,18262,18264,18267],{"className":18263},[36],[15,18265,18042],{"className":18266,"style":62},[36,37],[15,18268,18270],{"className":18269},[361],[15,18271,18273,18294],{"className":18272},[365,366],[15,18274,18276,18291],{"className":18275},[370],[15,18277,18279],{"className":18278,"style":3511},[374],[15,18280,18282,18285],{"style":18281},"top:-2.55em;margin-left:-0.0359em;margin-right:0.05em;",[15,18283],{"className":18284,"style":383},[382],[15,18286,18288],{"className":18287},[387,388,389,390],[15,18289,16159],{"className":18290},[36,37,390],[15,18292,399],{"className":18293},[398],[15,18295,18297],{"className":18296},[370],[15,18298,18300],{"className":18299,"style":406},[374],[15,18301],{},[15,18303],{"className":18304,"style":1598},[42],[15,18306,18308],{"className":18307},[2014],[15,18309,3105],{"className":18310,"style":3104},[36,2176],[15,18312],{"className":18313,"style":1598},[42],[15,18315,18317,18359],{"className":18316},[36],[15,18318,18320],{"className":18319},[36,18024],[15,18321,18323,18351],{"className":18322},[365,366],[15,18324,18326,18348],{"className":18325},[370],[15,18327,18329,18337],{"className":18328,"style":32},[374],[15,18330,18331,18334],{"style":10560},[15,18332],{"className":18333,"style":1431},[382],[15,18335,18042],{"className":18336,"style":62},[36,37],[15,18338,18339,18342],{"style":10560},[15,18340],{"className":18341,"style":1431},[382],[15,18343,18345],{"className":18344,"style":18052},[18051],[15,18346,18056],{"className":18347},[36],[15,18349,399],{"className":18350},[398],[15,18352,18354],{"className":18353},[370],[15,18355,18357],{"className":18356,"style":18066},[374],[15,18358],{},[15,18360,18362],{"className":18361},[361],[15,18363,18365,18385],{"className":18364},[365,366],[15,18366,18368,18382],{"className":18367},[370],[15,18369,18371],{"className":18370,"style":3511},[374],[15,18372,18373,18376],{"style":18281},[15,18374],{"className":18375,"style":383},[382],[15,18377,18379],{"className":18378},[387,388,389,390],[15,18380,16159],{"className":18381},[36,37,390],[15,18383,399],{"className":18384},[398],[15,18386,18388],{"className":18387},[370],[15,18389,18391],{"className":18390,"style":406},[374],[15,18392],{}," against a one-hot target ",[15,18395,18397],{"className":18396},[18],[15,18398,18400],{"className":18399,"ariaHidden":23},[22],[15,18401,18403,18406],{"className":18402},[27],[15,18404],{"className":18405,"style":123},[31],[15,18407,18042],{"className":18408,"style":62},[36,37],", the gradient with respect to the\nlogits collapses to",[15,18411,18413],{"className":18412},[4032],[15,18414,18416],{"className":18415},[18],[15,18417,18419,18540,18634],{"className":18418,"ariaHidden":23},[22],[15,18420,18422,18425,18531,18534,18537],{"className":18421},[27],[15,18423],{"className":18424,"style":16187},[31],[15,18426,18428,18431,18528],{"className":18427},[36],[15,18429],{"className":18430},[67,1410],[15,18432,18434],{"className":18433},[1414],[15,18435,18437,18520],{"className":18436},[365,366],[15,18438,18440,18517],{"className":18439},[370],[15,18441,18443,18495,18503],{"className":18442,"style":7814},[374],[15,18444,18445,18448],{"style":1427},[15,18446],{"className":18447,"style":1431},[382],[15,18449,18451,18454],{"className":18450},[36],[15,18452,12081],{"className":18453,"style":12080},[36],[15,18455,18457,18460],{"className":18456},[36],[15,18458,230],{"className":18459,"style":229},[36,37],[15,18461,18463],{"className":18462},[361],[15,18464,18466,18487],{"className":18465},[365,366],[15,18467,18469,18484],{"className":18468},[370],[15,18470,18472],{"className":18471,"style":3511},[374],[15,18473,18475,18478],{"style":18474},"top:-2.55em;margin-left:-0.044em;margin-right:0.05em;",[15,18476],{"className":18477,"style":383},[382],[15,18479,18481],{"className":18480},[387,388,389,390],[15,18482,16159],{"className":18483},[36,37,390],[15,18485,399],{"className":18486},[398],[15,18488,18490],{"className":18489},[370],[15,18491,18493],{"className":18492,"style":406},[374],[15,18494],{},[15,18496,18497,18500],{"style":1488},[15,18498],{"className":18499,"style":1431},[382],[15,18501],{"className":18502,"style":1496},[1495],[15,18504,18505,18508],{"style":1499},[15,18506],{"className":18507,"style":1431},[382],[15,18509,18511,18514],{"className":18510},[36],[15,18512,12081],{"className":18513,"style":12080},[36],[15,18515,12016],{"className":18516},[36,12143],[15,18518,399],{"className":18519},[398],[15,18521,18523],{"className":18522},[370],[15,18524,18526],{"className":18525,"style":16288},[374],[15,18527],{},[15,18529],{"className":18530},[103,1410],[15,18532],{"className":18533,"style":43},[42],[15,18535,48],{"className":18536},[47],[15,18538],{"className":18539,"style":43},[42],[15,18541,18543,18546,18625,18628,18631],{"className":18542},[27],[15,18544],{"className":18545,"style":3844},[31],[15,18547,18549,18591],{"className":18548},[36],[15,18550,18552],{"className":18551},[36,18024],[15,18553,18555,18583],{"className":18554},[365,366],[15,18556,18558,18580],{"className":18557},[370],[15,18559,18561,18569],{"className":18560,"style":32},[374],[15,18562,18563,18566],{"style":10560},[15,18564],{"className":18565,"style":1431},[382],[15,18567,18042],{"className":18568,"style":62},[36,37],[15,18570,18571,18574],{"style":10560},[15,18572],{"className":18573,"style":1431},[382],[15,18575,18577],{"className":18576,"style":18052},[18051],[15,18578,18056],{"className":18579},[36],[15,18581,399],{"className":18582},[398],[15,18584,18586],{"className":18585},[370],[15,18587,18589],{"className":18588,"style":18066},[374],[15,18590],{},[15,18592,18594],{"className":18593},[361],[15,18595,18597,18617],{"className":18596},[365,366],[15,18598,18600,18614],{"className":18599},[370],[15,18601,18603],{"className":18602,"style":3511},[374],[15,18604,18605,18608],{"style":18281},[15,18606],{"className":18607,"style":383},[382],[15,18609,18611],{"className":18610},[387,388,389,390],[15,18612,16159],{"className":18613},[36,37,390],[15,18615,399],{"className":18616},[398],[15,18618,18620],{"className":18619},[370],[15,18621,18623],{"className":18622,"style":406},[374],[15,18624],{},[15,18626],{"className":18627,"style":81},[42],[15,18629,1482],{"className":18630},[85],[15,18632],{"className":18633,"style":81},[42],[15,18635,18637,18640,18680],{"className":18636},[27],[15,18638],{"className":18639,"style":123},[31],[15,18641,18643,18646],{"className":18642},[36],[15,18644,18042],{"className":18645,"style":62},[36,37],[15,18647,18649],{"className":18648},[361],[15,18650,18652,18672],{"className":18651},[365,366],[15,18653,18655,18669],{"className":18654},[370],[15,18656,18658],{"className":18657,"style":3511},[374],[15,18659,18660,18663],{"style":18281},[15,18661],{"className":18662,"style":383},[382],[15,18664,18666],{"className":18665},[387,388,389,390],[15,18667,16159],{"className":18668},[36,37,390],[15,18670,399],{"className":18671},[398],[15,18673,18675],{"className":18674},[370],[15,18676,18678],{"className":18677,"style":406},[374],[15,18679],{},[15,18681,4730],{"className":18682},[36],[11,18684,18685,18686,18740,18741,8170,18756,18771,18772,18825],{},"The exponential of the softmax and the logarithm of the loss cancel, leaving a\ngradient that is just the prediction error, bounded and never vanishing even when\n",[15,18687,18689],{"className":18688},[18],[15,18690,18692],{"className":18691,"ariaHidden":23},[22],[15,18693,18695,18698],{"className":18694},[27],[15,18696],{"className":18697,"style":3844},[31],[15,18699,18701],{"className":18700},[36,18024],[15,18702,18704,18732],{"className":18703},[365,366],[15,18705,18707,18729],{"className":18706},[370],[15,18708,18710,18718],{"className":18709,"style":32},[374],[15,18711,18712,18715],{"style":10560},[15,18713],{"className":18714,"style":1431},[382],[15,18716,18042],{"className":18717,"style":62},[36,37],[15,18719,18720,18723],{"style":10560},[15,18721],{"className":18722,"style":1431},[382],[15,18724,18726],{"className":18725,"style":18052},[18051],[15,18727,18056],{"className":18728},[36],[15,18730,399],{"className":18731},[398],[15,18733,18735],{"className":18734},[370],[15,18736,18738],{"className":18737,"style":18066},[374],[15,18739],{}," saturates near ",[15,18742,18744],{"className":18743},[18],[15,18745,18747],{"className":18746,"ariaHidden":23},[22],[15,18748,18750,18753],{"className":18749},[27],[15,18751],{"className":18752,"style":2147},[31],[15,18754,1589],{"className":18755},[36],[15,18757,18759],{"className":18758},[18],[15,18760,18762],{"className":18761,"ariaHidden":23},[22],[15,18763,18765,18768],{"className":18764},[27],[15,18766],{"className":18767,"style":2147},[31],[15,18769,497],{"className":18770},[36],". Pairing the saturating output with the wrong\nloss (say squared error on a sigmoid) reintroduces a ",[15,18773,18775],{"className":18774},[18],[15,18776,18778],{"className":18777,"ariaHidden":23},[22],[15,18779,18781,18784,18816,18819,18822],{"className":18780},[27],[15,18782],{"className":18783,"style":1180},[31],[15,18785,18787,18790],{"className":18786},[36],[15,18788,1541],{"className":18789,"style":62},[36,37],[15,18791,18793],{"className":18792},[361],[15,18794,18796],{"className":18795},[365],[15,18797,18799],{"className":18798},[370],[15,18800,18802],{"className":18801,"style":1199},[374],[15,18803,18804,18807],{"style":1202},[15,18805],{"className":18806,"style":383},[382],[15,18808,18810],{"className":18809},[387,388,389,390],[15,18811,18813],{"className":18812},[36,390],[15,18814,1215],{"className":18815},[36,390],[15,18817,68],{"className":18818},[67],[15,18820,230],{"className":18821,"style":229},[36,37],[15,18823,104],{"className":18824},[103]," factor and the\nsaturation stalls learning — the reason the pairing in the table is not optional.",[1161,18827,18829],{"id":18828},"which-activation-to-use-where","Which activation to use where",[11,18831,18832],{},"Defaults, in the order you should reach for them:",[1278,18834,18835,18851],{},[1281,18836,18837],{},[1284,18838,18839,18842,18845,18848],{},[1287,18840,18841],{},"layer \u002F task",[1287,18843,18844],{},"first choice",[1287,18846,18847],{},"why",[1287,18849,18850],{},"fall back to",[1381,18852,18853,18866,18878,18907,18920,18932,18944],{},[1284,18854,18855,18858,18860,18863],{},[1386,18856,18857],{},"hidden (MLP, CNN)",[1386,18859,2155],{},[1386,18861,18862],{},"cheap, no right-side saturation",[1386,18864,18865],{},"Leaky ReLU \u002F ELU if units die",[1284,18867,18868,18871,18873,18876],{},[1386,18869,18870],{},"hidden (Transformers)",[1386,18872,2916],{},[1386,18874,18875],{},"smooth, strong empirically",[1386,18877,3291],{},[1284,18879,18880,18883,18886,18904],{},[1386,18881,18882],{},"hidden, self-normalizing",[1386,18884,18885],{},"SELU \u002F ELU",[1386,18887,18888,18889],{},"drives mean toward ",[15,18890,18892],{"className":18891},[18],[15,18893,18895],{"className":18894,"ariaHidden":23},[22],[15,18896,18898,18901],{"className":18897},[27],[15,18899],{"className":18900,"style":2147},[31],[15,18902,1589],{"className":18903},[36],[1386,18905,18906],{},"—",[1284,18908,18909,18912,18915,18918],{},[1386,18910,18911],{},"output, regression",[1386,18913,18914],{},"linear",[1386,18916,18917],{},"unbounded targets",[1386,18919,18906],{},[1284,18921,18922,18925,18927,18930],{},[1386,18923,18924],{},"output, binary",[1386,18926,1388],{},[1386,18928,18929],{},"gives a probability",[1386,18931,18906],{},[1284,18933,18934,18937,18939,18942],{},[1386,18935,18936],{},"output, multiclass",[1386,18938,17643],{},[1386,18940,18941],{},"distribution over classes",[1386,18943,18906],{},[1284,18945,18946,18949,18952,19019],{},[1386,18947,18948],{},"recurrent gates",[1386,18950,18951],{},"sigmoid + tanh",[1386,18953,18954,18955,18985,18986],{},"gating needs ",[15,18956,18958],{"className":18957},[18],[15,18959,18961],{"className":18960,"ariaHidden":23},[22],[15,18962,18964,18967,18970,18973,18976,18979,18982],{"className":18963},[27],[15,18965],{"className":18966,"style":58},[31],[15,18968,68],{"className":18969},[67],[15,18971,1589],{"className":18972},[36],[15,18974,1594],{"className":18975},[1593],[15,18977],{"className":18978,"style":1598},[42],[15,18980,497],{"className":18981},[36],[15,18983,104],{"className":18984},[103],"; state needs ",[15,18987,18989],{"className":18988},[18],[15,18990,18992],{"className":18991,"ariaHidden":23},[22],[15,18993,18995,18998,19001,19004,19007,19010,19013,19016],{"className":18994},[27],[15,18996],{"className":18997,"style":58},[31],[15,18999,68],{"className":19000},[67],[15,19002,1482],{"className":19003},[36],[15,19005,497],{"className":19006},[36],[15,19008,1594],{"className":19009},[1593],[15,19011],{"className":19012,"style":1598},[42],[15,19014,497],{"className":19015},[36],[15,19017,104],{"className":19018},[103],[1386,19020,18906],{},[146,19022,19024],{"type":19023},"remark",[11,19025,19026,19029,19030,19045,19046],{},[152,19027,19028],{},"Remark (The practical default)."," Start every hidden layer with ReLU. If you\nwatch the activation histograms and a large fraction of units sit dead at ",[15,19031,19033],{"className":19032},[18],[15,19034,19036],{"className":19035,"ariaHidden":23},[22],[15,19037,19039,19042],{"className":19038},[27],[15,19040],{"className":19041,"style":2147},[31],[15,19043,1589],{"className":19044},[36],",\nswitch to Leaky ReLU or ELU; if you are training a Transformer, use GELU. Reserve\nsigmoid and tanh for output and gating, where their bounded range is what is\nwanted, and pair sigmoid\u002Fsoftmax outputs with cross-entropy so the\nsaturation cancels.",[3874,19047,19048],{},[107,19049,15098],{"href":19050,"ariaDescribedBy":19051,"dataFootnoteRef":6,"id":19052},"#user-content-fn-stevens-act",[3880],"user-content-fnref-stevens-act",[1161,19054,19056],{"id":19055},"activations-after-relu","Activations after ReLU",[11,19058,19059],{},"Goodfellow's catalog predates the activations that now dominate large models. Two\nlater developments matter, and both were settled empirically rather than from\nfirst principles.",[11,19061,19062,19065,19066,19069,19070,19100,19101,19104,19105,19140,19141,19144,19145,19160],{},[152,19063,19064],{},"GELU and swish were found empirically."," Hendrycks & Gimpel's\n",[1227,19067,19068],{},"Gaussian Error Linear Unit"," (2016) motivated ",[15,19071,19073],{"className":19072},[18],[15,19074,19076],{"className":19075,"ariaHidden":23},[22],[15,19077,19079,19082,19085,19088,19091,19094,19097],{"className":19078},[27],[15,19080],{"className":19081,"style":58},[31],[15,19083,230],{"className":19084,"style":229},[36,37],[15,19086],{"className":19087,"style":1598},[42],[15,19089,2940],{"className":19090},[36],[15,19092,68],{"className":19093},[67],[15,19095,230],{"className":19096,"style":229},[36,37],[15,19098,104],{"className":19099},[103]," as multiplying the input\nby the probability a standard normal falls below it — a smooth, stochastic-looking\ngate rather than a hard ReLU switch. Around the same time Ramachandran, Zoph & Le\n(2017) ran an automated ",[1227,19102,19103],{},"search"," over candidate activation formulas and the winner,\n",[15,19106,19108],{"className":19107},[18],[15,19109,19111],{"className":19110,"ariaHidden":23},[22],[15,19112,19114,19117,19120,19123,19126,19129,19134,19137],{"className":19113},[27],[15,19115],{"className":19116,"style":58},[31],[15,19118,230],{"className":19119,"style":229},[36,37],[15,19121],{"className":19122,"style":1598},[42],[15,19124,1541],{"className":19125,"style":62},[36,37],[15,19127,68],{"className":19128},[67],[15,19130,19133],{"className":19131,"style":19132},[36,37],"margin-right:0.0528em;","β",[15,19135,230],{"className":19136,"style":229},[36,37],[15,19138,104],{"className":19139},[103],", they named ",[152,19142,19143],{},"Swish","; it is the SiLU of the table with a\nlearnable or fixed gain ",[15,19146,19148],{"className":19147},[18],[15,19149,19151],{"className":19150,"ariaHidden":23},[22],[15,19152,19154,19157],{"className":19153},[27],[15,19155],{"className":19156,"style":3844},[31],[15,19158,19133],{"className":19159,"style":19132},[36,37],". Neither unit is dramatically better than ReLU on\nsmall networks, but on deep Transformers the smoothness and the small negative dip\nconsistently shave a little off the loss, which is why GELU became the default in\nBERT and GPT-family models.",[11,19162,19163,19166,19167,19170,19171,19327,19328,19331,19332,19464,19465,4730],{},[152,19164,19165],{},"Gated linear units reshaped the Transformer's MLP."," Dauphin et al.\n(2017) introduced the ",[152,19168,19169],{},"gated linear unit"," (GLU), which splits a projection into\ntwo halves and lets one gate the other, ",[15,19172,19174],{"className":19173},[18],[15,19175,19177,19204,19269],{"className":19176,"ariaHidden":23},[22],[15,19178,19180,19183,19186,19189,19192,19195,19198,19201],{"className":19179},[27],[15,19181],{"className":19182,"style":58},[31],[15,19184,63],{"className":19185,"style":62},[36,37],[15,19187,68],{"className":19188},[67],[15,19190,77],{"className":19191},[36,37],[15,19193,104],{"className":19194},[103],[15,19196],{"className":19197,"style":43},[42],[15,19199,48],{"className":19200},[47],[15,19202],{"className":19203,"style":43},[42],[15,19205,19207,19210,19213,19216,19256,19259,19262,19266],{"className":19206},[27],[15,19208],{"className":19209,"style":58},[31],[15,19211,68],{"className":19212},[67],[15,19214,77],{"className":19215},[36,37],[15,19217,19219,19222],{"className":19218},[36],[15,19220,73],{"className":19221,"style":72},[36,37],[15,19223,19225],{"className":19224},[361],[15,19226,19228,19248],{"className":19227},[365,366],[15,19229,19231,19245],{"className":19230},[370],[15,19232,19234],{"className":19233,"style":375},[374],[15,19235,19236,19239],{"style":444},[15,19237],{"className":19238,"style":383},[382],[15,19240,19242],{"className":19241},[387,388,389,390],[15,19243,497],{"className":19244},[36,390],[15,19246,399],{"className":19247},[398],[15,19249,19251],{"className":19250},[370],[15,19252,19254],{"className":19253,"style":406},[374],[15,19255],{},[15,19257,104],{"className":19258},[103],[15,19260],{"className":19261,"style":81},[42],[15,19263,19265],{"className":19264},[85],"⊙",[15,19267],{"className":19268,"style":81},[42],[15,19270,19272,19275,19278,19281,19284,19324],{"className":19271},[27],[15,19273],{"className":19274,"style":58},[31],[15,19276,1541],{"className":19277,"style":62},[36,37],[15,19279,68],{"className":19280},[67],[15,19282,77],{"className":19283},[36,37],[15,19285,19287,19290],{"className":19286},[36],[15,19288,73],{"className":19289,"style":72},[36,37],[15,19291,19293],{"className":19292},[361],[15,19294,19296,19316],{"className":19295},[365,366],[15,19297,19299,19313],{"className":19298},[370],[15,19300,19302],{"className":19301,"style":375},[374],[15,19303,19304,19307],{"style":444},[15,19305],{"className":19306,"style":383},[382],[15,19308,19310],{"className":19309},[387,388,389,390],[15,19311,394],{"className":19312},[36,390],[15,19314,399],{"className":19315},[398],[15,19317,19319],{"className":19318},[370],[15,19320,19322],{"className":19321,"style":406},[374],[15,19323],{},[15,19325,104],{"className":19326},[103],". Shazeer\n(2020) swapped the sigmoid gate for a swish gate and found ",[152,19329,19330],{},"SwiGLU",",\n",[15,19333,19335],{"className":19334},[18],[15,19336,19338,19402],{"className":19337,"ariaHidden":23},[22],[15,19339,19341,19344,19347,19350,19390,19393,19396,19399],{"className":19340},[27],[15,19342],{"className":19343,"style":58},[31],[15,19345,68],{"className":19346},[67],[15,19348,77],{"className":19349},[36,37],[15,19351,19353,19356],{"className":19352},[36],[15,19354,73],{"className":19355,"style":72},[36,37],[15,19357,19359],{"className":19358},[361],[15,19360,19362,19382],{"className":19361},[365,366],[15,19363,19365,19379],{"className":19364},[370],[15,19366,19368],{"className":19367,"style":375},[374],[15,19369,19370,19373],{"style":444},[15,19371],{"className":19372,"style":383},[382],[15,19374,19376],{"className":19375},[387,388,389,390],[15,19377,497],{"className":19378},[36,390],[15,19380,399],{"className":19381},[398],[15,19383,19385],{"className":19384},[370],[15,19386,19388],{"className":19387,"style":406},[374],[15,19389],{},[15,19391,104],{"className":19392},[103],[15,19394],{"className":19395,"style":81},[42],[15,19397,19265],{"className":19398},[85],[15,19400],{"className":19401,"style":81},[42],[15,19403,19405,19408,19415,19418,19421,19461],{"className":19404},[27],[15,19406],{"className":19407,"style":58},[31],[15,19409,19411],{"className":19410},[2014],[15,19412,19414],{"className":19413},[36,2176],"swish",[15,19416,68],{"className":19417},[67],[15,19419,77],{"className":19420},[36,37],[15,19422,19424,19427],{"className":19423},[36],[15,19425,73],{"className":19426,"style":72},[36,37],[15,19428,19430],{"className":19429},[361],[15,19431,19433,19453],{"className":19432},[365,366],[15,19434,19436,19450],{"className":19435},[370],[15,19437,19439],{"className":19438,"style":375},[374],[15,19440,19441,19444],{"style":444},[15,19442],{"className":19443,"style":383},[382],[15,19445,19447],{"className":19446},[387,388,389,390],[15,19448,394],{"className":19449},[36,390],[15,19451,399],{"className":19452},[398],[15,19454,19456],{"className":19455},[370],[15,19457,19459],{"className":19458,"style":406},[374],[15,19460],{},[15,19462,104],{"className":19463},[103],", improved Transformer feed-forward blocks\nenough that it is now standard in models such as LLaMA and PaLM. The lesson's\nscalar-activation view still holds — the nonlinearity is where the depth comes\nfrom — but the modern feed-forward block gates two linear projections against each\nother rather than applying one pointwise ",[15,19466,19468],{"className":19467},[18],[15,19469,19471],{"className":19470,"ariaHidden":23},[22],[15,19472,19474,19477],{"className":19473},[27],[15,19475],{"className":19476,"style":123},[31],[15,19478,63],{"className":19479,"style":62},[36,37],[11,19481,19482,19483,19487,19488,19493,19494,19541,19542],{},"A quantitative footnote on the vanishing-gradient section: the modern answer is not\nonly ",[19484,19485,19486],"q",{},"use ReLU"," but ",[19484,19489,19490,4730],{},[152,19491,19492],{},"normalize"," Batch normalization (Ioffe & Szegedy, 2015)\nand layer normalization (Ba, Kiros & Hinton, 2016) rescale each layer's\npre-activations to keep them near the high-gradient region, so even saturating\nunits stay trainable at depth. Normalization and residual connections together are\nwhat keep deep stacks trainable despite the ",[15,19495,19497],{"className":19496},[18],[15,19498,19500],{"className":19499,"ariaHidden":23},[22],[15,19501,19503,19506,19509],{"className":19502},[27],[15,19504],{"className":19505,"style":14459},[31],[15,19507,14463],{"className":19508},[36],[15,19510,19512,19515],{"className":19511},[36],[15,19513,14470],{"className":19514},[36],[15,19516,19518],{"className":19517},[361],[15,19519,19521],{"className":19520},[365],[15,19522,19524],{"className":19523},[370],[15,19525,19527],{"className":19526,"style":14459},[374],[15,19528,19529,19532],{"style":1202},[15,19530],{"className":19531,"style":383},[382],[15,19533,19535],{"className":19534},[387,388,389,390],[15,19536,19538],{"className":19537},[36,390],[15,19539,12016],{"className":19540},[36,37,390]," arithmetic above.",[3874,19543,19544],{},[107,19545,14883],{"href":19546,"ariaDescribedBy":19547,"dataFootnoteRef":6,"id":19548},"#user-content-fn-beyond-act",[3880],"user-content-fnref-beyond-act",[1161,19550,19552],{"id":19551},"takeaways","Takeaways",[19554,19555,19556,19609,19862,19994,20005],"ul",{},[19557,19558,19559,19560,19563,19564,19608],"li",{},"The activation is the only nonlinearity in a layer; its ",[152,19561,19562],{},"derivative"," is what\nbackpropagation multiplies, so the size of ",[15,19565,19567],{"className":19566},[18],[15,19568,19570],{"className":19569,"ariaHidden":23},[22],[15,19571,19573,19576],{"className":19572},[27],[15,19574],{"className":19575,"style":1243},[31],[15,19577,19579,19582],{"className":19578},[36],[15,19580,63],{"className":19581,"style":62},[36,37],[15,19583,19585],{"className":19584},[361],[15,19586,19588],{"className":19587},[365],[15,19589,19591],{"className":19590},[370],[15,19592,19594],{"className":19593,"style":1199},[374],[15,19595,19596,19599],{"style":1202},[15,19597],{"className":19598,"style":383},[382],[15,19600,19602],{"className":19601},[387,388,389,390],[15,19603,19605],{"className":19604},[36,390],[15,19606,1215],{"className":19607},[36,390]," governs gradient flow.",[19557,19610,19611,19700,19701,3831,19716,19700,19822,19837,19838,19841,19842,19857,19858,19861],{},[15,19612,19614],{"className":19613},[18],[15,19615,19617,19664,19688],{"className":19616,"ariaHidden":23},[22],[15,19618,19620,19623,19655,19658,19661],{"className":19619},[27],[15,19621],{"className":19622,"style":1199},[31],[15,19624,19626,19629],{"className":19625},[36],[15,19627,1541],{"className":19628,"style":62},[36,37],[15,19630,19632],{"className":19631},[361],[15,19633,19635],{"className":19634},[365],[15,19636,19638],{"className":19637},[370],[15,19639,19641],{"className":19640,"style":1199},[374],[15,19642,19643,19646],{"style":1202},[15,19644],{"className":19645,"style":383},[382],[15,19647,19649],{"className":19648},[387,388,389,390],[15,19650,19652],{"className":19651},[36,390],[15,19653,1215],{"className":19654},[36,390],[15,19656],{"className":19657,"style":43},[42],[15,19659,48],{"className":19660},[47],[15,19662],{"className":19663,"style":43},[42],[15,19665,19667,19670,19673,19676,19679,19682,19685],{"className":19666},[27],[15,19668],{"className":19669,"style":58},[31],[15,19671,1541],{"className":19672,"style":62},[36,37],[15,19674,68],{"className":19675},[67],[15,19677,497],{"className":19678},[36],[15,19680],{"className":19681,"style":81},[42],[15,19683,1482],{"className":19684},[85],[15,19686],{"className":19687,"style":81},[42],[15,19689,19691,19694,19697],{"className":19690},[27],[15,19692],{"className":19693,"style":58},[31],[15,19695,1541],{"className":19696,"style":62},[36,37],[15,19698,104],{"className":19699},[103]," peaks at ",[15,19702,19704],{"className":19703},[18],[15,19705,19707],{"className":19706,"ariaHidden":23},[22],[15,19708,19710,19713],{"className":19709},[27],[15,19711],{"className":19712,"style":2147},[31],[15,19714,13821],{"className":19715},[36],[15,19717,19719],{"className":19718},[18],[15,19720,19722,19769,19787],{"className":19721,"ariaHidden":23},[22],[15,19723,19725,19728,19760,19763,19766],{"className":19724},[27],[15,19726],{"className":19727,"style":6742},[31],[15,19729,19731,19734],{"className":19730},[2014],[15,19732,1749],{"className":19733},[2014],[15,19735,19737],{"className":19736},[361],[15,19738,19740],{"className":19739},[365],[15,19741,19743],{"className":19742},[370],[15,19744,19746],{"className":19745,"style":6742},[374],[15,19747,19748,19751],{"style":2032},[15,19749],{"className":19750,"style":383},[382],[15,19752,19754],{"className":19753},[387,388,389,390],[15,19755,19757],{"className":19756},[36,390],[15,19758,1215],{"className":19759},[36,390],[15,19761],{"className":19762,"style":43},[42],[15,19764,48],{"className":19765},[47],[15,19767],{"className":19768,"style":43},[42],[15,19770,19772,19775,19778,19781,19784],{"className":19771},[27],[15,19773],{"className":19774,"style":1991},[31],[15,19776,497],{"className":19777},[36],[15,19779],{"className":19780,"style":81},[42],[15,19782,1482],{"className":19783},[85],[15,19785],{"className":19786,"style":81},[42],[15,19788,19790,19793],{"className":19789},[27],[15,19791],{"className":19792,"style":2010},[31],[15,19794,19796,19799],{"className":19795},[2014],[15,19797,1749],{"className":19798},[2014],[15,19800,19802],{"className":19801},[361],[15,19803,19805],{"className":19804},[365],[15,19806,19808],{"className":19807},[370],[15,19809,19811],{"className":19810,"style":2010},[374],[15,19812,19813,19816],{"style":2032},[15,19814],{"className":19815,"style":383},[382],[15,19817,19819],{"className":19818},[387,388,389,390],[15,19820,394],{"className":19821},[36,390],[15,19823,19825],{"className":19824},[18],[15,19826,19828],{"className":19827,"ariaHidden":23},[22],[15,19829,19831,19834],{"className":19830},[27],[15,19832],{"className":19833,"style":2147},[31],[15,19835,497],{"className":19836},[36],";\nboth ",[152,19839,19840],{},"saturate"," in the tails, and a product of ",[15,19843,19845],{"className":19844},[18],[15,19846,19848],{"className":19847,"ariaHidden":23},[22],[15,19849,19851,19854],{"className":19850},[27],[15,19852],{"className":19853,"style":3074},[31],[15,19855,12016],{"className":19856},[36,37]," small factors vanishes\ngeometrically with depth — the ",[152,19859,19860],{},"vanishing-gradient"," problem.",[19557,19863,19864,19866,19867,19909,19910,19912,19913,19993],{},[152,19865,2155],{}," has derivative ",[15,19868,19870],{"className":19869},[18],[15,19871,19873,19897],{"className":19872,"ariaHidden":23},[22],[15,19874,19876,19879,19882,19885,19888,19891,19894],{"className":19875},[27],[15,19877],{"className":19878,"style":58},[31],[15,19880,497],{"className":19881},[36,2213],[15,19883,2217],{"className":19884},[67],[15,19886,230],{"className":19887,"style":229},[36,37],[15,19889],{"className":19890,"style":43},[42],[15,19892,2227],{"className":19893},[47],[15,19895],{"className":19896,"style":43},[42],[15,19898,19900,19903,19906],{"className":19899},[27],[15,19901],{"className":19902,"style":58},[31],[15,19904,1589],{"className":19905},[36],[15,19907,2243],{"className":19908},[103],": no right-side saturation, so deep\nstacks train, but a unit driven entirely negative is ",[152,19911,15817],{}," (",[15,19914,19916],{"className":19915},[18],[15,19917,19919,19937,19984],{"className":19918,"ariaHidden":23},[22],[15,19920,19922,19925,19928,19931,19934],{"className":19921},[27],[15,19923],{"className":19924,"style":123},[31],[15,19926,63],{"className":19927,"style":62},[36,37],[15,19929],{"className":19930,"style":43},[42],[15,19932,48],{"className":19933},[47],[15,19935],{"className":19936,"style":43},[42],[15,19938,19940,19943,19975,19978,19981],{"className":19939},[27],[15,19941],{"className":19942,"style":1243},[31],[15,19944,19946,19949],{"className":19945},[36],[15,19947,63],{"className":19948,"style":62},[36,37],[15,19950,19952],{"className":19951},[361],[15,19953,19955],{"className":19954},[365],[15,19956,19958],{"className":19957},[370],[15,19959,19961],{"className":19960,"style":1199},[374],[15,19962,19963,19966],{"style":1202},[15,19964],{"className":19965,"style":383},[382],[15,19967,19969],{"className":19968},[387,388,389,390],[15,19970,19972],{"className":19971},[36,390],[15,19973,1215],{"className":19974},[36,390],[15,19976],{"className":19977,"style":43},[42],[15,19979,48],{"className":19980},[47],[15,19982],{"className":19983,"style":43},[42],[15,19985,19987,19990],{"className":19986},[27],[15,19988],{"className":19989,"style":2147},[31],[15,19991,1589],{"className":19992},[36]," forever).",[19557,19995,19996,19999,20000,3831,20002,20004],{},[152,19997,19998],{},"Leaky ReLU \u002F PReLU \u002F ELU \u002F GELU \u002F swish"," give the negative side a nonzero\nslope so no unit can be permanently frozen; ELU and GELU additionally center the\nmean; ",[152,20001,9575],{},[152,20003,11215],{}," round out the smooth and piecewise-linear ends.",[19557,20006,20007,20008,20010],{},"Hidden layers optimize for gradient flow (ReLU\u002FGELU); ",[152,20009,5271],{}," units match the\ntarget's range and loss (linear \u002F sigmoid \u002F softmax), where saturation is desired\nand cancelled by cross-entropy.",[20012,20013,20016,20021],"section",{"className":20014,"dataFootnotes":6},[20015],"footnotes",[1161,20017,20020],{"className":20018,"id":3880},[20019],"sr-only","Footnotes",[20022,20023,20024,20041,20070,20144,20159,20208,20231],"ol",{},[19557,20025,20027,10072,20030,20033,20034],{"id":20026},"user-content-fn-gf-hidden",[152,20028,20029],{},"Goodfellow",[1227,20031,20032],{},"Deep Learning",", §6.3 — Hidden Units: the catalog of ReLU and its generalizations (Leaky\u002FPReLU\u002FELU\u002FMaxout) versus the saturating sigmoid and tanh, and why rectified units are the modern default. ",[107,20035,20040],{"href":20036,"ariaLabel":20037,"className":20038,"dataFootnoteBackref":6},"#user-content-fnref-gf-hidden","Back to reference 1",[20039],"data-footnote-backref","↩",[19557,20042,20044,10072,20046,20048,20049,20064,20065],{"id":20043},"user-content-fn-gf-vanish",[152,20045,20029],{},[1227,20047,20032],{},", §8.2.5 \u002F §10.7 — the vanishing- and exploding-gradient problem: backprop multiplies per-layer Jacobians, so derivatives below ",[15,20050,20052],{"className":20051},[18],[15,20053,20055],{"className":20054,"ariaHidden":23},[22],[15,20056,20058,20061],{"className":20057},[27],[15,20059],{"className":20060,"style":2147},[31],[15,20062,497],{"className":20063},[36]," shrink the gradient geometrically with depth. ",[107,20066,20040],{"href":20067,"ariaLabel":20068,"className":20069,"dataFootnoteBackref":6},"#user-content-fnref-gf-vanish","Back to reference 2",[20039],[19557,20071,20073,10072,20075,20077,20078,20093,20094,20138,20139],{"id":20072},"user-content-fn-gf-saturate",[152,20074,20029],{},[1227,20076,20032],{},", §6.3.2 \u002F §8.4 — saturation and initialization: careful weight initialization buys precisely this, pre-activations near ",[15,20079,20081],{"className":20080},[18],[15,20082,20084],{"className":20083,"ariaHidden":23},[22],[15,20085,20087,20090],{"className":20086},[27],[15,20088],{"className":20089,"style":2147},[31],[15,20091,1589],{"className":20092},[36],", where ",[15,20095,20097],{"className":20096},[18],[15,20098,20100],{"className":20099,"ariaHidden":23},[22],[15,20101,20103,20106],{"className":20102},[27],[15,20104],{"className":20105,"style":1243},[31],[15,20107,20109,20112],{"className":20108},[36],[15,20110,63],{"className":20111,"style":62},[36,37],[15,20113,20115],{"className":20114},[361],[15,20116,20118],{"className":20117},[365],[15,20119,20121],{"className":20120},[370],[15,20122,20124],{"className":20123,"style":1199},[374],[15,20125,20126,20129],{"style":1202},[15,20127],{"className":20128,"style":383},[382],[15,20130,20132],{"className":20131},[387,388,389,390],[15,20133,20135],{"className":20134},[36,390],[15,20136,1215],{"className":20137},[36,390]," is largest. ",[107,20140,20040],{"href":20141,"ariaLabel":20142,"className":20143,"dataFootnoteBackref":6},"#user-content-fnref-gf-saturate","Back to reference 3",[20039],[19557,20145,20147,10072,20150,20153,20154],{"id":20146},"user-content-fn-chollet-act",[152,20148,20149],{},"Chollet",[1227,20151,20152],{},"Deep Learning with Python",", Ch. 4 — practical guidance on choosing activations: ReLU as the hidden-layer default, sigmoid\u002Fsoftmax at the output, and the failure modes that motivate the alternatives. ",[107,20155,20040],{"href":20156,"ariaLabel":20157,"className":20158,"dataFootnoteBackref":6},"#user-content-fnref-chollet-act","Back to reference 4",[20039],[19557,20160,20162,10072,20164,20166,20167,20185,20186,20202,20203],{"id":20161},"user-content-fn-gf-output",[152,20163,20029],{},[1227,20165,20032],{},", §6.2.2 — Output Units: sigmoid and softmax paired with the negative log-likelihood, so the loss's ",[15,20168,20170],{"className":20169},[18],[15,20171,20173],{"className":20172,"ariaHidden":23},[22],[15,20174,20176,20179],{"className":20175},[27],[15,20177],{"className":20178,"style":3844},[31],[15,20180,20182],{"className":20181},[2014],[15,20183,3105],{"className":20184,"style":3104},[36,2176]," undoes the unit's ",[15,20187,20189],{"className":20188},[18],[15,20190,20192],{"className":20191,"ariaHidden":23},[22],[15,20193,20195,20198],{"className":20194},[27],[15,20196],{"className":20197,"style":123},[31],[15,20199,20201],{"className":20200},[2014],"exp"," and the saturation that would otherwise stall learning. ",[107,20204,20040],{"href":20205,"ariaLabel":20206,"className":20207,"dataFootnoteBackref":6},"#user-content-fnref-gf-output","Back to reference 5",[20039],[19557,20209,20211,10072,20214,20217,20218,3831,20222,20225,20226],{"id":20210},"user-content-fn-stevens-act",[152,20212,20213],{},"Stevens et al.",[1227,20215,20216],{},"Deep Learning with PyTorch",", Ch. 6 — activations in practice: ",[20219,20220,20221],"code",{},"nn.Tanh",[20219,20223,20224],{},"nn.ReLU"," in a fitted network, and the effect of the choice on the training curve. ",[107,20227,20040],{"href":20228,"ariaLabel":20229,"className":20230,"dataFootnoteBackref":6},"#user-content-fnref-stevens-act","Back to reference 6",[20039],[19557,20232,20234,20235,20238,20239,20242,20243,20246,20247,20250,20251,20254,20255,20258,20259],{"id":20233},"user-content-fn-beyond-act","Primary sources: Hendrycks & Gimpel, ",[19484,20236,20237],{},"Gaussian Error Linear Units (GELUs)"," (2016); Ramachandran, Zoph & Le, ",[19484,20240,20241],{},"Searching for Activation Functions"," (2017) for Swish\u002FSiLU; Dauphin et al., ",[19484,20244,20245],{},"Language Modeling with Gated Convolutional Networks"," (2017) for the GLU; Shazeer, ",[19484,20248,20249],{},"GLU Variants Improve Transformer"," (2020) for SwiGLU; Ioffe & Szegedy, ",[19484,20252,20253],{},"Batch Normalization"," (2015) and Ba, Kiros & Hinton, ",[19484,20256,20257],{},"Layer Normalization"," (2016) for keeping pre-activations in the high-gradient region. ",[107,20260,20040],{"href":20261,"ariaLabel":20262,"className":20263,"dataFootnoteBackref":6},"#user-content-fnref-beyond-act","Back to reference 7",[20039],{"title":6,"searchDepth":20265,"depth":20265,"links":20266},2,[20267,20268,20279,20280,20281,20282,20283,20284,20285],{"id":1163,"depth":20265,"text":1164},{"id":3884,"depth":20265,"text":3885,"children":20269},[20270,20272,20273,20274,20275,20276,20277,20278],{"id":1388,"depth":20271,"text":3892},3,{"id":1749,"depth":20271,"text":5567},{"id":7742,"depth":20271,"text":2155},{"id":8220,"depth":20271,"text":8221},{"id":8767,"depth":20271,"text":2634},{"id":9575,"depth":20271,"text":3083},{"id":10264,"depth":20271,"text":10265},{"id":11215,"depth":20271,"text":3474},{"id":11996,"depth":20265,"text":11997},{"id":15202,"depth":20265,"text":15203},{"id":17534,"depth":20265,"text":17535},{"id":18828,"depth":20265,"text":18829},{"id":19055,"depth":20265,"text":19056},{"id":19551,"depth":20265,"text":19552},{"id":3880,"depth":20265,"text":20020},[],"computer-science","A layer is an affine map followed by a pointwise nonlinearity,\nh=g(Wx+b). The affine part carries all the parameters; the\nnonlinearity g supplies the entire nonlinear effect:\nstrip it out and a deep stack collapses to a single linear map. The choice of g\ntherefore matters: it sets how gradients flow back through depth, whether units\nsaturate, and how fast the network trains. This lesson catalogs the standard\nchoices and their derivatives.",false,"md",{"moduleNumber":20265,"lessonNumber":20265,"order":20292},202,"Neural Networks",true,"\u002Fdeep-learning\u002Fneural-networks\u002Factivation-functions",[],"---\ntitle: Activation Functions\nmodule: Neural Networks\nmoduleNumber: 2\nlessonNumber: 2\norder: 202\nsummary: >\n  The activation is the only nonlinear part of a layer, and the reason\n  depth adds expressive power. We catalog the standard hidden units (sigmoid, tanh,\n  ReLU and its descendants, plus GELU, softplus, swish and maxout), derive each\n  unit's derivative in full, make the vanishing-gradient problem\n  quantitative with the chain-rule product, work numeric examples, and explain\n  why the saturating units gave way to ReLU and why ReLU's own dead-unit failure\n  gave way to Leaky\u002FPReLU\u002FELU\u002FGELU.\ntopics: [Neural Networks]\nsources:\n  - book: Goodfellow\n    ref: \"§6.3 — Hidden Units\"\n  - book: Goodfellow\n    ref: \"§6.2.2 — Output Units; §8.2.5 Saturation\"\n  - book: Chollet\n    ref: \"Ch. 4 — Getting Started with Neural Networks (activations)\"\n  - book: Stevens\n    ref: \"Ch. 6 — Using a Neural Network to Fit the Data (nn.Tanh, nn.ReLU)\"\n---\n\nA layer is an affine map followed by a pointwise nonlinearity,\n$h = g(Wx + b)$. The affine part carries all the parameters; the\n[nonlinearity $g$ supplies the entire nonlinear effect](\u002Fdeep-learning\u002Fneural-networks\u002Fthe-multilayer-perceptron):\nstrip it out and a deep stack collapses to a single linear map. The choice of $g$\ntherefore matters: it sets how gradients flow back through depth, whether units\nsaturate, and how fast the network trains. This lesson catalogs the standard\nchoices and their derivatives.\n\n> **Definition (Activation function).** A function $g : \\mathbb{R} \\to \\mathbb{R}$\n> applied elementwise to the pre-activation $z = Wx + b$ to produce a layer's\n> output $h = g(z)$. It must be nonlinear (else depth collapses) and almost\n> everywhere differentiable (so [backpropagation](\u002Fdeep-learning\u002Fneural-networks\u002Fbackpropagation)\n> can pass a gradient through it).\n\nTo make the requirement precise, suppose two\nlayers with $g$ the identity: $h_2 = W_2(W_1 x + b_1) + b_2 = (W_2 W_1)x +\n(W_2 b_1 + b_2)$. The composite is one affine map with weight $W_2 W_1$ and bias\n$W_2 b_1 + b_2$, and no amount of stacking escapes the affine class. A single\nnonlinear $g$ between the two multiplications is what lets a network represent a\nfunction that no single affine map can. Everything below is about which $g$ to\npick and what each choice costs in the backward pass.\n\n## The catalog\n\nEvery hidden unit in common use is one scalar function and its derivative. The\nderivative column matters most: during the backward pass the gradient at a\nunit is multiplied by $g'(z)$, so the _size_ of $g'$ controls how much signal\npasses through. The following table is the reference the rest of the lesson expands.\n\n| name | $g(z)$ | $g'(z)$ | range | saturates? | notes |\n| --- | --- | --- | --- | --- | --- |\n| sigmoid | $\\dfrac{1}{1+e^{-z}}$ | $\\sigma(1-\\sigma)$ | $(0,1)$ | both tails | $g'\\le\\tfrac14$; not zero-centered |\n| tanh | $\\dfrac{e^{z}-e^{-z}}{e^{z}+e^{-z}}$ | $1-\\tanh^2 z$ | $(-1,1)$ | both tails | zero-centered; $g'\\le 1$ |\n| ReLU | $\\max(0,z)$ | $\\mathbf{1}[z>0]$ | $[0,\\infty)$ | left only | cheap; can die |\n| Leaky ReLU | $\\max(\\alpha z, z)$ | $\\mathbf{1}[z>0]+\\alpha\\,\\mathbf{1}[z\\le 0]$ | $(-\\infty,\\infty)$ | no | $\\alpha\\!\\approx\\!0.01$ keeps a trickle |\n| PReLU | $\\max(\\alpha z, z)$ | as Leaky, $\\alpha$ learned | $(-\\infty,\\infty)$ | no | $\\alpha$ a trained parameter |\n| ELU | $z$ if $z>0$ else $\\alpha(e^{z}-1)$ | $1$ if $z>0$ else $\\alpha e^{z}$ | $(-\\alpha,\\infty)$ | left soft | smooth; pushes mean toward $0$ |\n| GELU | $z\\,\\Phi(z)$ | $\\Phi(z)+z\\,\\phi(z)$ | $\\approx[-0.17,\\infty)$ | left soft | $\\Phi$ standard-normal CDF |\n| Softplus | $\\log(1+e^{z})$ | $\\sigma(z)$ | $(0,\\infty)$ | no | smooth ReLU; $g'=\\sigma$ |\n| Swish \u002F SiLU | $z\\,\\sigma(z)$ | $\\sigma(z)+z\\,\\sigma(z)(1-\\sigma(z))$ | $\\approx[-0.28,\\infty)$ | left soft | self-gated; non-monotone |\n| Maxout | $\\max_j (w_j^\\top x + b_j)$ | $w_{j^\\star}$ of the winning piece | $\\mathbb{R}$ | no | learns its own shape; $k\\times$ params |\n\nHere $\\sigma$ is the logistic sigmoid, $\\Phi$ and $\\phi$ are the standard-normal\nCDF and PDF, and $\\alpha$ is a small positive constant (or a learned parameter for\nPReLU). Two families fall out of the table: the **saturating** units (sigmoid,\ntanh) whose derivative vanishes at both extremes, and the **non-saturating**\nrectified units (ReLU onward) whose derivative stays bounded away from zero on at\nleast the positive half-line.[^gf-hidden]\n\n## Every derivative, derived\n\nBackpropagation never needs the activation itself in the backward pass, only its\nderivative. Each derivative below takes one or two lines to derive.\n\n### Sigmoid\n\nWrite $\\sigma(z) = (1+e^{-z})^{-1}$ and differentiate by the chain rule:\n\n$$\n\\sigma'(z) = -\\,(1+e^{-z})^{-2}\\cdot(-e^{-z})\n= \\frac{e^{-z}}{(1+e^{-z})^2}\n= \\frac{1}{1+e^{-z}}\\cdot\\frac{e^{-z}}{1+e^{-z}}.\n$$\n\nThe first factor is $\\sigma(z)$; the second is $1-\\sigma(z)$, since\n$1-\\sigma = 1 - \\frac{1}{1+e^{-z}} = \\frac{e^{-z}}{1+e^{-z}}$. Hence\n\n$$\n\\;\\sigma'(z) = \\sigma(z)\\,\\parens{1-\\sigma(z)}\\;\n$$\n\nTwo consequences read straight off this form. The derivative is expressed in the\n_output_ $\\sigma$, so a forward pass that cached $h=\\sigma(z)$ needs no re-exp in\nthe backward pass, only $h(1-h)$. And because $\\sigma \\in (0,1)$, the product\n$\\sigma(1-\\sigma)$ is a downward parabola in $\\sigma$ capped at its vertex, which\nis the source of the $\\tfrac14$ ceiling derived below.\n\n### Tanh\n\nThe tanh derivative follows the same pattern. With $\\tanh z = \\frac{e^z-e^{-z}}{e^z+e^{-z}}$,\nthe quotient rule gives numerator $(e^z+e^{-z})^2 - (e^z-e^{-z})^2$ over\n$(e^z+e^{-z})^2$, and expanding the two squares,\n\n$$\n(e^z+e^{-z})^2 - (e^z-e^{-z})^2\n= \\parens{e^{2z}+2+e^{-2z}} - \\parens{e^{2z}-2+e^{-2z}} = 4,\n$$\n\nso\n\n$$\n\\tanh'(z) = \\frac{(e^z+e^{-z})^2-(e^z-e^{-z})^2}{(e^z+e^{-z})^2}\n= 1 - \\frac{(e^z-e^{-z})^2}{(e^z+e^{-z})^2}\n= \\,1-\\tanh^2 z\\,.\n$$\n\nSince $\\tanh = 2\\sigma(2z)-1$, tanh is a rescaled, recentered sigmoid; the\nzero-centering lifts its peak derivative to $1$ (at $z=0$) versus the sigmoid's\n$\\tfrac14$, and makes tanh the better of the two saturating choices.\n\n### ReLU\n\nReLU is differentiable everywhere except the kink at $z=0$, and its derivative is\njust the indicator of the active region:\n\n$$\n\\frac{d}{dz}\\max(0,z) =\n\\begin{cases} 1 & z > 0 \\\\ 0 & z \u003C 0 \\end{cases}\n= \\mathbf{1}[z>0],\n$$\n\nwith the value at $z=0$ taken by convention to be $0$ (frameworks pick $0$ or $1$;\nthe measure-zero choice is irrelevant in practice). This step-function derivative,\na hard gate of $1$ or $0$, is the source of both ReLU's strength and its dead-unit\npathology.\n\n### Leaky ReLU and PReLU\n\nLeaky ReLU replaces the flat left branch with a shallow line of slope $\\alpha$,\n\n$$\ng(z) = \\max(\\alpha z, z) =\n\\begin{cases} z & z \\ge 0 \\\\ \\alpha z & z \u003C 0, \\end{cases}\n\\qquad\ng'(z) =\n\\begin{cases} 1 & z > 0 \\\\ \\alpha & z \u003C 0. \\end{cases}\n$$\n\nThe derivative is now $\\alpha > 0$ on the entire negative half-line rather than\n$0$, so a unit sitting in the negative region still receives an $\\alpha$-scaled\ngradient. PReLU is identical except $\\alpha$ is a learned parameter (one per\nchannel), so the negative slope adapts during training rather than being fixed at\n$0.01$.\n\n### ELU\n\nELU keeps the identity for $z>0$ but replaces the kink with a smooth exponential\nthat saturates gently toward $-\\alpha$ on the left:\n\n$$\ng(z) =\n\\begin{cases} z & z > 0 \\\\ \\alpha\\,(e^{z}-1) & z \\le 0, \\end{cases}\n\\qquad\ng'(z) =\n\\begin{cases} 1 & z > 0 \\\\ \\alpha\\,e^{z} & z \\le 0. \\end{cases}\n$$\n\nThe left derivative $\\alpha e^{z}$ is strictly positive and continuous, and at\n$z=0$ both branches agree ($g'=1$ from the right, $\\alpha e^{0}=\\alpha$ from the\nleft; with the common choice $\\alpha=1$ the derivative is continuous too). Because\nthe negative branch bottoms out at $-\\alpha$ instead of $0$, ELU can output\nnegative values, which pulls each layer's mean activation toward zero and keeps\nthe next layer's pre-activations nearer the high-gradient region.\n\n### Softplus\n\nSoftplus is a smooth ReLU; its derivative is the sigmoid, which is why it appears\nwhenever a strictly positive, differentiable output is wanted:\n\n$$\ng(z) = \\log(1+e^{z}),\n\\qquad\ng'(z) = \\frac{e^{z}}{1+e^{z}} = \\frac{1}{1+e^{-z}} = \\sigma(z).\n$$\n\nAs $z \\to +\\infty$, $\\log(1+e^z) \\to z$ (it hugs the ReLU line); as $z \\to\n-\\infty$ it decays to $0$ smoothly. Its derivative never reaches $0$ or $1$\nexactly, so it has no dead region but also no perfectly flat gradient of $1$.\n\n### GELU and swish\u002FSiLU\n\nGELU gates the input by the probability that a standard normal falls below it,\n$g(z) = z\\,\\Phi(z)$ with $\\Phi$ the standard-normal CDF and $\\phi$ its density.\nThe product rule gives\n\n$$\ng'(z) = \\Phi(z) + z\\,\\phi(z),\n\\qquad \\phi(z) = \\tfrac{1}{\\sqrt{2\\pi}}e^{-z^2\u002F2}.\n$$\n\nSwish (a.k.a. SiLU) is the same idea with a logistic gate, $g(z) = z\\,\\sigma(z)$,\nso\n\n$$\ng'(z) = \\sigma(z) + z\\,\\sigma(z)\\parens{1-\\sigma(z)} = \\sigma(z)\\parens{1 + z\\,(1-\\sigma(z))}.\n$$\n\nBoth are smooth, both are the identity in the large-$z$ limit, and both dip\nslightly _below_ zero for moderately negative $z$ before returning to $0$ — a\nnon-monotone shape that the empirical record favors in deep Transformers. Their\nminimum sits near $z\\approx -1.28$ for swish (value $\\approx -0.28$) and near\n$z\\approx -0.75$ for GELU (value $\\approx -0.17$).\n\n### Maxout\n\nMaxout is the outlier: it is not a fixed scalar shape but a piecewise-linear unit\nthat takes the max over $k$ affine pieces,\n\n$$\ng(x) = \\max_{j=1,\\dots,k}\\parens{w_j^\\top x + b_j},\n\\qquad\n\\nabla_x g = w_{j^\\star},\\quad j^\\star = \\arg\\max_j\\parens{w_j^\\top x + b_j}.\n$$\n\nThe gradient flows entirely through the winning piece, so maxout has no flat\nregion at all as long as one piece is increasing; with $k=2$ and $w_2=0$ it\nrecovers ReLU exactly. The cost is $k\\times$ the parameters, which is why it is\nrarely the default.\n\n$$\n% caption: Three activations on shared axes. Sigmoid and tanh flatten in both tails;\n% ReLU is the identity for $z>0$ and exactly $0$ for $z\u003C0$.\n\\begin{tikzpicture}[>=stealth, font=\\small]\n  \\definecolor{acc}{HTML}{2348F2}\n  \\definecolor{green}{HTML}{1F9D4D}\n  \\definecolor{red}{HTML}{C0392B}\n  % axes\n  \\draw[->, thick] (-3.6,0) -- (3.8,0) node[right, font=\\footnotesize] {input};\n  \\draw[->, thick] (0,-1.9) -- (0,2.3) node[above, font=\\footnotesize] {output};\n  \\draw[black, dashed] (-3.4,1.5) -- (3.4,1.5);\n  \\node[black, font=\\footnotesize, anchor=west] at (3.45,1.5) {1};\n  \\draw[black, dashed] (-3.4,-1.5) -- (3.4,-1.5);\n  \\node[black, font=\\footnotesize, anchor=west] at (3.45,-1.5) {-1};\n  % sigmoid, scaled so 1 -> 1.5\n  \\draw[acc, very thick] plot[domain=-3.3:3.3, samples=80] (\\x, {1.5\u002F(1+exp(-2*\\x))});\n  % tanh, scaled so 1 -> 1.5\n  \\draw[green, very thick] plot[domain=-3.3:3.3, samples=80] (\\x, {1.5*tanh(\\x)});\n  % ReLU (clip at axis range), scaled so input 1 -> 1.5\n  \\draw[red, very thick] (-3.3,0) -- (0,0) -- (1.27,1.9);\n  \\node[acc, font=\\footnotesize, anchor=east] at (-1.3,1.0) {sigmoid};\n  \\node[green, font=\\footnotesize, anchor=west] at (1.85,1.2) {tanh};\n  \\node[red, font=\\footnotesize, anchor=south east] at (1.27,1.95) {ReLU};\n\\end{tikzpicture}\n$$\n\nThe rectified family shares one shape below, all identity on the right and\ndiffering only in how they treat the negative side: hard zero (ReLU), a shallow\nline (Leaky\u002FPReLU), or a smooth curve that dips and returns (ELU, GELU, swish).\n\n$$\n% caption: The rectified family on shared axes. All are the identity for $z>0$.\n% ReLU is flat on the left; Leaky keeps a shallow line; ELU and swish curve\n% smoothly, and swish dips below $0$ before returning.\n\\begin{tikzpicture}[>=stealth, font=\\small]\n  \\definecolor{acc}{HTML}{2348F2}\n  \\definecolor{green}{HTML}{1F9D4D}\n  \\definecolor{red}{HTML}{C0392B}\n  \\definecolor{orange}{HTML}{D97A17}\n  % axes\n  \\draw[->, thick] (-2.8,0) -- (3.7,0) node[right, font=\\footnotesize] {input};\n  \\draw[->, thick] (0,-1.15) -- (0,2.4) node[above, font=\\footnotesize] {output};\n  % ReLU (blue): flat then identity\n  \\draw[acc, very thick] (-2.6,0) -- (0,0) -- (2.1,2.1);\n  % Leaky (red): shallow slope on the left\n  \\draw[red, very thick] (-2.6,-0.52) -- (0,0);\n  % ELU (green): smooth saturating left branch toward -1\n  \\draw[green, very thick] plot[domain=-2.6:0, samples=60] (\\x, {exp(\\x)-1});\n  % swish (orange): dips below 0 then returns; z*sigmoid(z)\n  \\draw[orange, very thick] plot[domain=-2.6:0, samples=60] (\\x, {\\x\u002F(1+exp(-\\x))});\n  \\node[acc, font=\\footnotesize, anchor=south east] at (2.05,2.0) {ReLU};\n  \\node[red, font=\\footnotesize, anchor=south east] at (-2.6,-0.5) {Leaky};\n  \\node[green, font=\\footnotesize, anchor=north west] at (-2.55,-0.9) {ELU};\n  \\node[orange, font=\\footnotesize, anchor=north] at (-0.55,-0.8) {swish};\n\\end{tikzpicture}\n$$\n\n## The vanishing-gradient problem, made quantitative\n\nThe backward pass through an $L$-layer network multiplies one Jacobian factor per\nlayer. Along a single path of units the gradient that reaches layer $k$ carries a\nproduct of activation derivatives,\n\n$$\n\\frac{\\partial \\mathcal{L}}{\\partial z^{(k)}}\n\\;=\\;\n\\parens{\\textstyle\\prod_{\\ell=k+1}^{L} g'\\!\\parens{z^{(\\ell)}}\\, w^{(\\ell)}}\\,\n\\frac{\\partial \\mathcal{L}}{\\partial h^{(L)}},\n$$\n\nso the magnitude of the gradient at depth $k$ scales like the **product** of the\nper-layer derivatives $g'(z^{(\\ell)})$ (times the weights). If each factor is\nbelow $1$, the product shrinks geometrically with depth, and the deep layers stop\nreceiving signal.[^gf-vanish]\n\n> **Definition (Saturating unit).** A unit whose derivative $g'(z) \\to 0$ as\n> $\\abs{z} \\to \\infty$. Sigmoid and tanh saturate in both tails; a unit is\n> **non-saturating** if $g'$ stays bounded away from $0$ on an unbounded region\n> (ReLU: $g'=1$ for all $z>0$).\n\nThe sigmoid is the worst case: its derivative is small even at its\nmaximum. Maximizing $\\sigma'(z) = \\sigma(1-\\sigma)$ over $\\sigma \\in (0,1)$: set\n$p = \\sigma$ and differentiate $p(1-p) = p - p^2$, giving $1 - 2p = 0$, so the\npeak is at $p = \\tfrac12$, i.e. $z = 0$:\n\n$$\n\\max_z \\sigma'(z) = \\tfrac12\\cdot\\tfrac12 = \\tfrac14 = 0.25 .\n$$\n\nSo a sigmoid contributes **at most** $0.25$ per layer, and far less in the tails.\nWork the tail out numerically: at $z=4$, $\\sigma(4) = \\frac{1}{1+e^{-4}} \\approx\n0.982$, and $\\sigma'(4) = 0.982\\,(1-0.982) \\approx 0.0177$. The tanh is better (it\npeaks at $\\tanh'(0)=1$) but still vanishes in the tails, with\n$\\tanh'(2) = 1-\\tanh^2 2 = 1 - 0.964^2 \\approx 0.071$.\n\n| depth $L$ | $0.25^{L}$ (sigmoid peak) | $1.0^{L}$ (tanh peak) | $1^{L}$ (ReLU active) |\n| --- | --- | --- | --- |\n| $1$ | $0.25$ | $1$ | $1$ |\n| $5$ | $9.8\\times 10^{-4}$ | $1$ | $1$ |\n| $10$ | $9.5\\times 10^{-7}$ | $1$ | $1$ |\n| $20$ | $9.1\\times 10^{-13}$ | $1$ | $1$ |\n\nTen sigmoid layers, each at their _best_ point, already\nattenuate the gradient by a factor of $10^{6}$, and the realistic factor is worse\nbecause most units sit off their peak. This is why deep sigmoid\u002Ftanh stacks were\nnearly untrainable before ReLU, and why initialization that keeps pre-activations\nnear $z=0$ (where $g'$ is largest) matters so much; see\n[initialization](\u002Fdeep-learning\u002Foptimization\u002Finitialization).[^gf-saturate]\n\n$$\n% caption: Derivatives on shared axes: the sigmoid derivative peaks at $0.25$,\n% tanh at $1$, and the ReLU derivative is a unit step.\n\\begin{tikzpicture}[>=stealth, font=\\small]\n  \\definecolor{acc}{HTML}{2348F2}\n  \\definecolor{green}{HTML}{1F9D4D}\n  \\definecolor{red}{HTML}{C0392B}\n  % axes\n  \\draw[->, thick] (-3.3,0) -- (3.5,0) node[right, font=\\footnotesize] {input};\n  \\draw[->, thick] (0,-0.2) -- (0,2.2) node[above, font=\\footnotesize] {\\texttt{derivative}};\n  \\draw[black, dashed] (-3.3,2.0) -- (3.3,2.0) node[black, right, font=\\footnotesize] {1};\n  \\draw[black, dashed] (-3.3,0.5) -- (3.3,0.5) node[black, right, font=\\footnotesize] {0.25};\n  % sigmoid' = s(1-s), peak 0.25 -> scaled to 0.5\n  \\draw[acc, very thick] plot[domain=-3.2:3.2, samples=90]\n    (\\x, {2*(1\u002F(1+exp(-2*\\x)))*(1-1\u002F(1+exp(-2*\\x)))});\n  % tanh' = 1 - tanh^2, peak 1 -> scaled to 2\n  \\draw[green, very thick] plot[domain=-3.2:3.2, samples=90]\n    (\\x, {2*(1-tanh(\\x)*tanh(\\x))});\n  % ReLU' step: 0 then 1 -> scaled to 2\n  \\draw[red, very thick] (-3.2,0) -- (0,0);\n  \\draw[red, very thick] (0,2.0) -- (3.2,2.0);\n  \\draw[red, very thick, dotted] (0,0) -- (0,2.0);\n  \\node[acc, font=\\footnotesize, anchor=south] at (1.7,0.55) {sigmoid'};\n  \\node[green, font=\\footnotesize, anchor=west] at (0.7,1.55) {tanh'};\n  \\node[red, font=\\footnotesize, anchor=south] at (2.2,2.0) {ReLU'};\n\\end{tikzpicture}\n$$\n\nThe product-of-derivatives view makes the depth penalty geometric, not additive.\nThe next figure shows the same gradient magnitude collapsing layer by layer as it\npropagates backward through a saturating stack.\n\n$$\n% caption: Gradient magnitude shrinking backward through a saturating stack: each\n% layer multiplies by $g'\\le 0.25$, so early layers receive almost nothing.\n\\begin{tikzpicture}[>=stealth, font=\\small]\n  \\definecolor{acc}{HTML}{2348F2}\n  \\definecolor{red}{HTML}{C0392B}\n  % baseline\n  \\draw[->, thick] (-0.4,0) -- (10.4,0) node[right, font=\\footnotesize] {\\texttt{layer (output at right)}};\n  \\draw[->, thick] (0,-0.1) -- (0,3.4) node[above, font=\\footnotesize] {\\texttt{gradient size}};\n  % bars: heights ~ 0.25^k from the output side, drawn left=early to right=output\n  % use explicit heights so the decay reads cleanly\n  \\draw[red, very thick, fill=red!15]  (0.4,0)  rectangle (1.2,0.10);\n  \\draw[red, very thick, fill=red!15]  (1.8,0)  rectangle (2.6,0.16);\n  \\draw[acc, very thick, fill=acc!15]  (3.2,0)  rectangle (4.0,0.30);\n  \\draw[acc, very thick, fill=acc!15]  (4.6,0)  rectangle (5.4,0.60);\n  \\draw[acc, very thick, fill=acc!15]  (6.0,0)  rectangle (6.8,1.30);\n  \\draw[acc, very thick, fill=acc!15]  (7.4,0)  rectangle (8.2,3.00);\n  \\node[font=\\scriptsize, anchor=north] at (0.8,-0.12)  {early};\n  \\node[font=\\scriptsize, anchor=north] at (7.8,-0.12)  {output};\n  \\node[red, font=\\footnotesize, anchor=south west] at (2.0,0.55)\n    {\\texttt{early layers starve}};\n  \\draw[->, black, thick] (7.2,2.7) .. controls (4.4,3.0) and (2.4,2.2) .. (1.1,1.4);\n  \\node[black, font=\\scriptsize, anchor=south] at (4.9,3.0)\n    {\\texttt{each step scales by the derivative}};\n\\end{tikzpicture}\n$$\n\n## ReLU and the dead-unit problem\n\nReLU sidesteps vanishing gradients on its active side outright: for $z>0$ the\nderivative is exactly $1$, so the product across active layers is $1^{L}=1$, with no\nattenuation at all. That single property, more than any other, made very deep\nnetworks trainable.\n\n> **Theorem (ReLU does not saturate on the right).** For $z>0$, $\\,\\frac{d}{dz}\\max(0,z)=1$\n> for all $z$, so a path of active ReLUs multiplies the backward gradient by\n> $\\prod_\\ell 1 = 1$, independent of depth.\n\n> **Proof.** $\\max(0,z)=z$ on $z>0$, whose derivative is the constant $1$; the\n> product of $L$ ones is $1$ regardless of $L$. $\\qed$\n\nThe flat left half causes a different failure. A unit whose pre-activation is negative for\n_every_ training example outputs $0$ and has derivative $0$, so it receives no\ngradient and never updates: it is **dead**.\n\n> **Definition (Dead ReLU).** A ReLU unit for which $z = w^\\top x + b \u003C 0$ on the\n> entire data distribution. Then $g(z)=0$ and $g'(z)=0$ for all inputs, so\n> $\\nabla_w \\mathcal{L}=0$: the unit is frozen and contributes nothing. A large\n> negative bias, or a too-large gradient step that drives the weights negative,\n> can kill a unit permanently.\n\nThe mechanism is worth tracing through the chain rule. The gradient on a weight\n$w_i$ of a ReLU unit is\n\n$$\n\\frac{\\partial \\mathcal{L}}{\\partial w_i}\n= \\frac{\\partial \\mathcal{L}}{\\partial h}\\cdot g'(z)\\cdot x_i\n= \\frac{\\partial \\mathcal{L}}{\\partial h}\\cdot \\mathbf{1}[z>0]\\cdot x_i .\n$$\n\nOnce $z\u003C0$ on every example, the factor $\\mathbf{1}[z>0]$ is $0$ for every one, so\n$\\partial\\mathcal{L}\u002F\\partial w_i = 0$ regardless of the upstream gradient or the\ninputs. Nothing in the update can move $w$ or $b$, so the unit cannot recover:\nthe failure is irreversible.\n\n$$\n% caption: A dead ReLU: its pre-activation stays in the flat region $z\u003C0$ where\n% the derivative is $0$, so no gradient reaches its weights and it never recovers.\n\\begin{tikzpicture}[>=stealth, font=\\small]\n  \\definecolor{acc}{HTML}{2348F2}\n  \\definecolor{red}{HTML}{C0392B}\n  \\definecolor{green}{HTML}{1F9D4D}\n  % axes\n  \\draw[->, thick] (-3.7,0) -- (3.8,0) node[right, font=\\footnotesize] {input};\n  \\draw[->, thick] (0,-1.1) -- (0,2.6) node[above, font=\\footnotesize] {output};\n  % shaded dead region (left of 0)\n  \\fill[red!14] (-3.5,-1.0) rectangle (0,2.45);\n  % ReLU curve (drawn over the shading)\n  \\draw[acc, very thick] (-3.5,0) -- (0,0) -- (2.1,2.1);\n  \\node[red, font=\\footnotesize, anchor=south] at (-1.75,1.85) {\\texttt{flat: output 0}};\n  \\node[red, font=\\footnotesize, anchor=south] at (-1.75,1.25) {\\texttt{derivative 0}};\n  % the data cloud sitting entirely in z\u003C0, below the axis to keep text clear\n  \\foreach \\p in {(-2.7,-0.55),(-2.2,-0.55),(-1.7,-0.55),(-1.2,-0.55),(-0.7,-0.55)}\n    \\fill[red] \\p circle (2.4pt);\n  \\node[red, font=\\footnotesize, anchor=north] at (-1.75,-0.75) {\\texttt{all inputs land here}};\n  \\node[green, font=\\footnotesize, anchor=west] at (0.55,2.25) {\\texttt{alive: slope 1}};\n\\end{tikzpicture}\n$$\n\nThe fixes all do the same thing: give the negative side a nonzero slope so a stuck\nunit still receives a gradient and can recover.\n\n| unit | negative branch | derivative at $z\u003C0$ | recovers? |\n| --- | --- | --- | --- |\n| ReLU | $0$ | $0$ | no — gradient is $0$ |\n| Leaky ReLU | $\\alpha z$, $\\alpha\\!=\\!0.01$ | $\\alpha$ | yes — small constant slope |\n| PReLU | $\\alpha z$, $\\alpha$ learned | $\\alpha$ | yes — slope adapts per channel |\n| ELU | $\\alpha(e^{z}-1)$ | $\\alpha e^{z}$ | yes — smooth, mean toward $0$ |\n| GELU | $z\\,\\Phi(z)$ | $\\Phi(z)+z\\phi(z)$ | yes — smooth, near-$0$ slope |\n\nLeaky ReLU and PReLU keep a literal straight line of slope $\\alpha$ on the left so\n$g'(z)=\\alpha>0$ everywhere; a dead-looking unit still receives an $\\alpha$-scaled\ngradient and can recover. ELU and GELU replace the kink with a smooth curve whose\nnegative-side derivative is small but strictly positive, which additionally pushes\nthe activation mean toward zero, a self-normalizing effect that helps the next\nlayer's pre-activations stay near the high-gradient region.[^chollet-act]\n\n$$\n% caption: ReLU versus Leaky ReLU. Both are the identity for $z>0$; for $z\u003C0$ ReLU\n% is flat (zero gradient) while Leaky ReLU keeps a small negative slope $\\alpha$.\n\\begin{tikzpicture}[>=stealth, font=\\small]\n  \\definecolor{acc}{HTML}{2348F2}\n  \\definecolor{red}{HTML}{C0392B}\n  % axes\n  \\draw[->, thick] (-3.0,0) -- (4.4,0) node[right, font=\\footnotesize] {input};\n  \\draw[->, thick] (0,-1.1) -- (0,2.3) node[above, font=\\footnotesize] {output};\n  % ReLU (blue): flat then slope 1\n  \\draw[acc, very thick] (-2.9,0) -- (0,0) -- (2.0,2.0);\n  % Leaky ReLU (red): slight negative slope then slope 1\n  \\draw[red, very thick] (-2.9,-0.58) -- (0,0) -- (2.0,2.0);\n  \\node[acc, font=\\footnotesize, anchor=south east] at (-1.4,0.08) {\\texttt{ReLU (flat)}};\n  \\node[red, font=\\footnotesize, anchor=north east] at (-1.9,-0.42) {\\texttt{Leaky (slope a)}};\n  \\node[font=\\footnotesize, anchor=west] at (2.4,1.4) {both: slope $1$};\n\\end{tikzpicture}\n$$\n\n## Output units, not hidden units\n\nThe hidden-layer rules above optimize for gradient flow. The **output** unit is\nchosen instead to match the target's range and the loss, and the saturating units\nthat fail in hidden layers become the right choice at the output, where they pair\nwith a log-loss that cancels the saturation. The output choice is tied to the\n[loss](\u002Fdeep-learning\u002Fneural-networks\u002Floss-functions-and-output-units); the table\nbelow is the standard pairing.\n\n| task | output activation | range | paired loss |\n| --- | --- | --- | --- |\n| regression | none (linear) | $\\mathbb{R}$ | mean squared error |\n| binary classification | sigmoid | $(0,1)$ | binary cross-entropy |\n| multiclass | softmax | simplex, sums to $1$ | categorical cross-entropy |\n| count \u002F nonnegative | softplus or exp | $(0,\\infty)$ | Poisson \u002F MSE |\n\n> **Definition (Softmax).** The vector activation\n> $\\softmax(z)_i = e^{z_i}\u002F\\sum_j e^{z_j}$, mapping a vector of logits\n> to a probability distribution over $K$ classes. It is the multiclass\n> generalization of the sigmoid, and pairs with cross-entropy so the\n> log cancels the exponential and leaves the clean gradient $\\hat{y}-y$.[^gf-output]\n\nThe cancellation is worth seeing once. With softmax output $\\hat{y} =\n\\softmax(z)$ and cross-entropy loss $\\mathcal{L} = -\\sum_i y_i\n\\log \\hat{y}_i$ against a one-hot target $y$, the gradient with respect to the\nlogits collapses to\n\n$$\n\\frac{\\partial \\mathcal{L}}{\\partial z_i} = \\hat{y}_i - y_i .\n$$\n\nThe exponential of the softmax and the logarithm of the loss cancel, leaving a\ngradient that is just the prediction error, bounded and never vanishing even when\n$\\hat{y}$ saturates near $0$ or $1$. Pairing the saturating output with the wrong\nloss (say squared error on a sigmoid) reintroduces a $\\sigma'(z)$ factor and the\nsaturation stalls learning — the reason the pairing in the table is not optional.\n\n## Which activation to use where\n\nDefaults, in the order you should reach for them:\n\n| layer \u002F task | first choice | why | fall back to |\n| --- | --- | --- | --- |\n| hidden (MLP, CNN) | ReLU | cheap, no right-side saturation | Leaky ReLU \u002F ELU if units die |\n| hidden (Transformers) | GELU | smooth, strong empirically | Swish \u002F SiLU |\n| hidden, self-normalizing | SELU \u002F ELU | drives mean toward $0$ | — |\n| output, regression | linear | unbounded targets | — |\n| output, binary | sigmoid | gives a probability | — |\n| output, multiclass | softmax | distribution over classes | — |\n| recurrent gates | sigmoid + tanh | gating needs $(0,1)$; state needs $(-1,1)$ | — |\n\n> **Remark (The practical default).** Start every hidden layer with ReLU. If you\n> watch the activation histograms and a large fraction of units sit dead at $0$,\n> switch to Leaky ReLU or ELU; if you are training a Transformer, use GELU. Reserve\n> sigmoid and tanh for output and gating, where their bounded range is what is\n> wanted, and pair sigmoid\u002Fsoftmax outputs with cross-entropy so the\n> saturation cancels.[^stevens-act]\n\n## Activations after ReLU\n\nGoodfellow's catalog predates the activations that now dominate large models. Two\nlater developments matter, and both were settled empirically rather than from\nfirst principles.\n\n**GELU and swish were found empirically.** Hendrycks & Gimpel's\n_Gaussian Error Linear Unit_ (2016) motivated $z\\,\\Phi(z)$ as multiplying the input\nby the probability a standard normal falls below it — a smooth, stochastic-looking\ngate rather than a hard ReLU switch. Around the same time Ramachandran, Zoph & Le\n(2017) ran an automated _search_ over candidate activation formulas and the winner,\n$z\\,\\sigma(\\beta z)$, they named **Swish**; it is the SiLU of the table with a\nlearnable or fixed gain $\\beta$. Neither unit is dramatically better than ReLU on\nsmall networks, but on deep Transformers the smoothness and the small negative dip\nconsistently shave a little off the loss, which is why GELU became the default in\nBERT and GPT-family models.\n\n**Gated linear units reshaped the Transformer's MLP.** Dauphin et al.\n(2017) introduced the **gated linear unit** (GLU), which splits a projection into\ntwo halves and lets one gate the other, $g(x) = (xW_1) \\odot \\sigma(xW_2)$. Shazeer\n(2020) swapped the sigmoid gate for a swish gate and found **SwiGLU**,\n$(xW_1)\\odot\\swish(xW_2)$, improved Transformer feed-forward blocks\nenough that it is now standard in models such as LLaMA and PaLM. The lesson's\nscalar-activation view still holds — the nonlinearity is where the depth comes\nfrom — but the modern feed-forward block gates two linear projections against each\nother rather than applying one pointwise $g$.\n\nA quantitative footnote on the vanishing-gradient section: the modern answer is not\nonly \"use ReLU\" but \"**normalize**.\" Batch normalization (Ioffe & Szegedy, 2015)\nand layer normalization (Ba, Kiros & Hinton, 2016) rescale each layer's\npre-activations to keep them near the high-gradient region, so even saturating\nunits stay trainable at depth. Normalization and residual connections together are\nwhat keep deep stacks trainable despite the $0.25^{L}$ arithmetic above.[^beyond-act]\n\n## Takeaways\n\n- The activation is the only nonlinearity in a layer; its **derivative** is what\n  backpropagation multiplies, so the size of $g'$ governs gradient flow.\n- $\\sigma' = \\sigma(1-\\sigma)$ peaks at $0.25$ and $\\tanh'=1-\\tanh^2$ peaks at $1$;\n  both **saturate** in the tails, and a product of $L$ small factors vanishes\n  geometrically with depth — the **vanishing-gradient** problem.\n- **ReLU** has derivative $\\mathbf{1}[z>0]$: no right-side saturation, so deep\n  stacks train, but a unit driven entirely negative is **dead** ($g=g'=0$ forever).\n- **Leaky ReLU \u002F PReLU \u002F ELU \u002F GELU \u002F swish** give the negative side a nonzero\n  slope so no unit can be permanently frozen; ELU and GELU additionally center the\n  mean; **softplus** and **maxout** round out the smooth and piecewise-linear ends.\n- Hidden layers optimize for gradient flow (ReLU\u002FGELU); **output** units match the\n  target's range and loss (linear \u002F sigmoid \u002F softmax), where saturation is desired\n  and cancelled by cross-entropy.\n\n[^gf-hidden]: **Goodfellow**, _Deep Learning_, §6.3 — Hidden Units: the catalog of ReLU and its generalizations (Leaky\u002FPReLU\u002FELU\u002FMaxout) versus the saturating sigmoid and tanh, and why rectified units are the modern default.\n[^gf-vanish]: **Goodfellow**, _Deep Learning_, §8.2.5 \u002F §10.7 — the vanishing- and exploding-gradient problem: backprop multiplies per-layer Jacobians, so derivatives below $1$ shrink the gradient geometrically with depth.\n[^gf-saturate]: **Goodfellow**, _Deep Learning_, §6.3.2 \u002F §8.4 — saturation and initialization: careful weight initialization buys precisely this, pre-activations near $0$, where $g'$ is largest.\n[^gf-output]: **Goodfellow**, _Deep Learning_, §6.2.2 — Output Units: sigmoid and softmax paired with the negative log-likelihood, so the loss's $\\log$ undoes the unit's $\\exp$ and the saturation that would otherwise stall learning.\n[^chollet-act]: **Chollet**, _Deep Learning with Python_, Ch. 4 — practical guidance on choosing activations: ReLU as the hidden-layer default, sigmoid\u002Fsoftmax at the output, and the failure modes that motivate the alternatives.\n[^stevens-act]: **Stevens et al.**, _Deep Learning with PyTorch_, Ch. 6 — activations in practice: `nn.Tanh` and `nn.ReLU` in a fitted network, and the effect of the choice on the training curve.\n[^beyond-act]: Primary sources: Hendrycks & Gimpel, \"Gaussian Error Linear Units (GELUs)\" (2016); Ramachandran, Zoph & Le, \"Searching for Activation Functions\" (2017) for Swish\u002FSiLU; Dauphin et al., \"Language Modeling with Gated Convolutional Networks\" (2017) for the GLU; Shazeer, \"GLU Variants Improve Transformer\" (2020) for SwiGLU; Ioffe & Szegedy, \"Batch Normalization\" (2015) and Ba, Kiros & Hinton, \"Layer Normalization\" (2016) for keeping pre-activations in the high-gradient region.\n",{"text":20299,"minutes":20300,"time":20301,"words":20302},"13 min read",12.48,748800,2496,{"title":5,"description":20288},[20305,20307,20309,20311],{"book":20029,"ref":20306},"§6.3 — Hidden Units",{"book":20029,"ref":20308},"§6.2.2 — Output Units; §8.2.5 Saturation",{"book":20149,"ref":20310},"Ch. 4 — Getting Started with Neural Networks (activations)",{"book":20312,"ref":20313},"Stevens","Ch. 6 — Using a Neural Network to Fit the Data (nn.Tanh, nn.ReLU)","available","06.deep-learning\u002F02.neural-networks\u002F02.activation-functions","The activation is the only nonlinear part of a layer, and the reason depth adds expressive power. We catalog the standard hidden units (sigmoid, tanh, ReLU and its descendants, plus GELU, softplus, swish and maxout), derive each unit's derivative in full, make the vanishing-gradient problem quantitative with the chain-rule product, work numeric examples, and explain why the saturating units gave way to ReLU and why ReLU's own dead-unit failure gave way to Leaky\u002FPReLU\u002FELU\u002FGELU.\n",[20293],"5UF-On9-8L-FfHt9KUuSjQUVT_Mmq3fjqmztBVTRX4c",{"\u002Falgorithms\u002Ffoundations\u002Fwhat-is-an-algorithm":20320,"\u002Falgorithms\u002Ffoundations\u002Fproof-techniques":20321,"\u002Falgorithms\u002Ffoundations\u002Fasymptotic-analysis":20322,"\u002Falgorithms\u002Ffoundations\u002Fgrowth-rates-and-loop-analysis":20323,"\u002Falgorithms\u002Ffoundations\u002Frecurrences":20324,"\u002Falgorithms\u002Ffoundations\u002Famortized-analysis":20325,"\u002Falgorithms\u002Fdivide-and-conquer\u002Fmergesort":20326,"\u002Falgorithms\u002Fdivide-and-conquer\u002Fquicksort":20327,"\u002Falgorithms\u002Fdivide-and-conquer\u002Fselection":20328,"\u002Falgorithms\u002Fdivide-and-conquer\u002Ffast-multiplication":20329,"\u002Falgorithms\u002Fsorting\u002Fheaps-and-heapsort":20330,"\u002Falgorithms\u002Fsorting\u002Fsorting-lower-bounds":20331,"\u002Falgorithms\u002Fsorting\u002Flinear-time-sorting":20332,"\u002Falgorithms\u002Fsorting\u002Fexternal-sorting":20333,"\u002Falgorithms\u002Fdata-structures\u002Felementary-structures":20334,"\u002Falgorithms\u002Fdata-structures\u002Fhash-tables":20335,"\u002Falgorithms\u002Fdata-structures\u002Fbinary-search-trees":20336,"\u002Falgorithms\u002Fdata-structures\u002Favl-trees":20337,"\u002Falgorithms\u002Fdata-structures\u002Fbalanced-trees":20338,"\u002Falgorithms\u002Fdata-structures\u002Funion-find":20339,"\u002Falgorithms\u002Fdata-structures\u002Ffenwick-and-segment-trees":20340,"\u002Falgorithms\u002Fdata-structures\u002Fspatial-data-structures":20341,"\u002Falgorithms\u002Fdata-structures\u002Fskip-lists-and-probabilistic-structures":20342,"\u002Falgorithms\u002Fdata-structures\u002Fb-trees":20343,"\u002Falgorithms\u002Fdata-structures\u002Fdata-stream-algorithms":20344,"\u002Falgorithms\u002Fdata-structures\u002Fstreaming-sketches":20345,"\u002Falgorithms\u002Fsequences\u002Ftwo-pointers-and-windows":20346,"\u002Falgorithms\u002Fsequences\u002Fprefix-sums":20347,"\u002Falgorithms\u002Fsequences\u002Fmonotonic-stacks":20348,"\u002Falgorithms\u002Fsequences\u002Fbinary-search-on-the-answer":20349,"\u002Falgorithms\u002Fsequences\u002Fstring-matching":20350,"\u002Falgorithms\u002Fsequences\u002Fkmp-and-z-function":20351,"\u002Falgorithms\u002Fsequences\u002Ftries":20352,"\u002Falgorithms\u002Fsequences\u002Fsuffix-arrays-and-aho-corasick":20353,"\u002Falgorithms\u002Fgraphs\u002Frepresentations-and-traversal":20354,"\u002Falgorithms\u002Fgraphs\u002Fdepth-first-search":20355,"\u002Falgorithms\u002Fgraphs\u002Ftopological-sort-and-scc":20356,"\u002Falgorithms\u002Fgraphs\u002Fminimum-spanning-trees":20357,"\u002Falgorithms\u002Fgraphs\u002Fkruskal-and-prim":20358,"\u002Falgorithms\u002Fgraphs\u002Fshortest-paths":20359,"\u002Falgorithms\u002Fgraphs\u002Fall-pairs-and-negative-weights":20360,"\u002Falgorithms\u002Fgraphs\u002Fnetwork-flow":20361,"\u002Falgorithms\u002Fgraphs\u002Fmax-flow-min-cut":20362,"\u002Falgorithms\u002Fgraphs\u002Fbridges-and-articulation-points":20363,"\u002Falgorithms\u002Fgraphs\u002Flowest-common-ancestor":20364,"\u002Falgorithms\u002Fgraphs\u002Ftwo-sat":20365,"\u002Falgorithms\u002Fgraphs\u002Feulerian-tours":20366,"\u002Falgorithms\u002Fgraphs\u002Fbipartite-matching":20367,"\u002Falgorithms\u002Fgreedy\u002Fthe-greedy-method":20368,"\u002Falgorithms\u002Fgreedy\u002Fscheduling-and-intervals":20369,"\u002Falgorithms\u002Fgreedy\u002Fhuffman-codes":20370,"\u002Falgorithms\u002Fgreedy\u002Fmatroids":20371,"\u002Falgorithms\u002Fgreedy\u002Fstable-matching":20372,"\u002Falgorithms\u002Fdynamic-programming\u002Fprinciples":20373,"\u002Falgorithms\u002Fdynamic-programming\u002Fsequence-dp":20374,"\u002Falgorithms\u002Fdynamic-programming\u002Flongest-increasing-subsequence":20375,"\u002Falgorithms\u002Fdynamic-programming\u002Fknapsack":20376,"\u002Falgorithms\u002Fdynamic-programming\u002Fcoin-change-and-unbounded":20377,"\u002Falgorithms\u002Fdynamic-programming\u002Finterval-dp":20378,"\u002Falgorithms\u002Fdynamic-programming\u002Ftree-dp":20379,"\u002Falgorithms\u002Fdynamic-programming\u002Fbitmask-dp":20380,"\u002Falgorithms\u002Fdynamic-programming\u002Fdp-optimizations":20381,"\u002Falgorithms\u002Fdynamic-programming\u002Fdp-on-graphs":20382,"\u002Falgorithms\u002Fdynamic-programming\u002Fdigit-and-probability-dp":20383,"\u002Falgorithms\u002Fbacktracking\u002Fbacktracking-fundamentals":20384,"\u002Falgorithms\u002Fbacktracking\u002Fconstraint-search":20385,"\u002Falgorithms\u002Fbacktracking\u002Fbranch-and-bound":20386,"\u002Falgorithms\u002Fbacktracking\u002Fgraph-backtracking":20387,"\u002Falgorithms\u002Fmathematical-algorithms\u002Fnumber-theory-basics":20388,"\u002Falgorithms\u002Fmathematical-algorithms\u002Fmodular-exponentiation-and-primality":20389,"\u002Falgorithms\u002Fmathematical-algorithms\u002Fsieve-and-factorization":20390,"\u002Falgorithms\u002Fmathematical-algorithms\u002Fcombinatorics":20391,"\u002Falgorithms\u002Fmathematical-algorithms\u002Fmatrix-exponentiation":20392,"\u002Falgorithms\u002Fmathematical-algorithms\u002Ffast-fourier-transform":20393,"\u002Falgorithms\u002Fmathematical-algorithms\u002Fgradient-descent":20394,"\u002Falgorithms\u002Fcomputational-geometry\u002Fgeometric-primitives":20395,"\u002Falgorithms\u002Fcomputational-geometry\u002Fconvex-hull":20396,"\u002Falgorithms\u002Fcomputational-geometry\u002Fsweep-line":20397,"\u002Falgorithms\u002Fcomputational-geometry\u002Fpolygons-and-proximity":20398,"\u002Falgorithms\u002Fintractability\u002Fp-np-reductions":20399,"\u002Falgorithms\u002Fintractability\u002Fnp-completeness":20400,"\u002Falgorithms\u002Fintractability\u002Fcoping-with-hardness":20401,"\u002Falgorithms\u002Fintractability\u002Fapproximation-algorithms":20402,"\u002Falgorithms":20403,"\u002Fcalculus\u002Flimits-and-continuity\u002Ffunctions-and-models":20404,"\u002Fcalculus\u002Flimits-and-continuity\u002Fthe-limit-of-a-function":20405,"\u002Fcalculus\u002Flimits-and-continuity\u002Flimit-laws-and-the-precise-definition":20406,"\u002Fcalculus\u002Flimits-and-continuity\u002Fcontinuity":20407,"\u002Fcalculus\u002Fderivatives\u002Fthe-derivative-and-rates-of-change":20408,"\u002Fcalculus\u002Fderivatives\u002Fdifferentiation-rules-and-the-chain-rule":20409,"\u002Fcalculus\u002Fderivatives\u002Fimplicit-differentiation-and-related-rates":20410,"\u002Fcalculus\u002Fderivatives\u002Flinear-approximations-and-differentials":20411,"\u002Fcalculus\u002Fapplications-of-derivatives\u002Fextrema-and-the-mean-value-theorem":20412,"\u002Fcalculus\u002Fapplications-of-derivatives\u002Fhow-derivatives-shape-a-graph":20413,"\u002Fcalculus\u002Fapplications-of-derivatives\u002Fcurve-sketching-and-optimization":20414,"\u002Fcalculus\u002Fapplications-of-derivatives\u002Fnewtons-method-and-antiderivatives":20415,"\u002Fcalculus\u002Fintegrals\u002Farea-and-the-definite-integral":20416,"\u002Fcalculus\u002Fintegrals\u002Fthe-fundamental-theorem-of-calculus":20417,"\u002Fcalculus\u002Fintegrals\u002Fthe-substitution-rule":20418,"\u002Fcalculus\u002Fapplications-of-integration\u002Fareas-and-volumes":20419,"\u002Fcalculus\u002Fapplications-of-integration\u002Fwork-average-value-and-arc-length":20420,"\u002Fcalculus\u002Fapplications-of-integration\u002Fphysics-economics-and-probability":20421,"\u002Fcalculus\u002Fexponential-logarithmic-and-inverse-functions\u002Finverse-functions-logarithms-and-exponentials":20422,"\u002Fcalculus\u002Fexponential-logarithmic-and-inverse-functions\u002Fgrowth-decay-inverse-trig-and-hyperbolic-functions":20423,"\u002Fcalculus\u002Fexponential-logarithmic-and-inverse-functions\u002Flhospitals-rule":20424,"\u002Fcalculus\u002Ftechniques-of-integration\u002Fintegration-by-parts":20425,"\u002Fcalculus\u002Ftechniques-of-integration\u002Ftrigonometric-integrals-and-substitution":20426,"\u002Fcalculus\u002Ftechniques-of-integration\u002Fpartial-fractions-and-integration-strategy":20427,"\u002Fcalculus\u002Ftechniques-of-integration\u002Fapproximate-and-improper-integrals":20428,"\u002Fcalculus\u002Fparametric-and-polar\u002Fparametric-curves-and-their-calculus":20429,"\u002Fcalculus\u002Fparametric-and-polar\u002Fpolar-coordinates":20430,"\u002Fcalculus\u002Fparametric-and-polar\u002Fconic-sections":20431,"\u002Fcalculus\u002Fsequences-and-series\u002Fsequences":20432,"\u002Fcalculus\u002Fsequences-and-series\u002Fseries-and-the-integral-test":20433,"\u002Fcalculus\u002Fsequences-and-series\u002Fthe-convergence-tests":20434,"\u002Fcalculus\u002Fsequences-and-series\u002Fpower-series":20435,"\u002Fcalculus\u002Fsequences-and-series\u002Ftaylor-and-maclaurin-series":20436,"\u002Fcalculus\u002Fvectors-and-space-curves\u002Fvectors-and-the-dot-product":20437,"\u002Fcalculus\u002Fvectors-and-space-curves\u002Fthe-cross-product-lines-and-planes":20418,"\u002Fcalculus\u002Fvectors-and-space-curves\u002Fcylinders-and-quadric-surfaces":20438,"\u002Fcalculus\u002Fvectors-and-space-curves\u002Fvector-functions-and-space-curves":20439,"\u002Fcalculus\u002Fvectors-and-space-curves\u002Farc-length-curvature-and-motion":20440,"\u002Fcalculus\u002Fpartial-derivatives\u002Ffunctions-of-several-variables":20408,"\u002Fcalculus\u002Fpartial-derivatives\u002Fpartial-derivatives":20441,"\u002Fcalculus\u002Fpartial-derivatives\u002Ftangent-planes-and-the-chain-rule":20442,"\u002Fcalculus\u002Fpartial-derivatives\u002Fdirectional-derivatives-and-the-gradient":20443,"\u002Fcalculus\u002Fpartial-derivatives\u002Foptimization-and-lagrange-multipliers":20444,"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fdouble-integrals":20445,"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Ftriple-integrals-and-coordinate-systems":20446,"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fvector-fields-and-line-integrals":20447,"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fgreens-theorem-curl-and-divergence":20448,"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fsurface-integrals":20449,"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fstokes-and-the-divergence-theorem":20450,"\u002Fcalculus":20451,"\u002Fmechanics\u002Ffoundations\u002Fmeasurement-and-dimensions":20452,"\u002Fmechanics\u002Ffoundations\u002Fvector-algebra":20453,"\u002Fmechanics\u002Fkinematics\u002Fone-dimensional-motion":20454,"\u002Fmechanics\u002Fkinematics\u002Fmotion-graphs":20455,"\u002Fmechanics\u002Fkinematics\u002Fprojectile-motion":20456,"\u002Fmechanics\u002Fkinematics\u002Frelative-motion":20457,"\u002Fmechanics\u002Fkinematics\u002Fcircular-motion":20458,"\u002Fmechanics\u002Fdynamics\u002Fnewtons-laws":20459,"\u002Fmechanics\u002Fdynamics\u002Ffree-body-diagrams":20460,"\u002Fmechanics\u002Fdynamics\u002Ffriction-and-curved-motion":20461,"\u002Fmechanics\u002Fdynamics\u002Fnumerical-dynamics":20462,"\u002Fmechanics\u002Fdynamics\u002Fcenter-of-mass-systems":20463,"\u002Fmechanics\u002Fenergy\u002Fwork-and-kinetic-energy":20464,"\u002Fmechanics\u002Fenergy\u002Fpotential-energy":20465,"\u002Fmechanics\u002Fenergy\u002Fmultiparticle-work":20466,"\u002Fmechanics\u002Fenergy\u002Fmass-energy-and-binding":20467,"\u002Fmechanics\u002Fenergy\u002Fphotons-and-quantization":20468,"\u002Fmechanics\u002Fmomentum\u002Fmomentum-and-collisions":20469,"\u002Fmechanics\u002Fmomentum\u002Fcenter-of-mass-collisions":20470,"\u002Fmechanics\u002Fmomentum\u002Frocket-propulsion":20471,"\u002Fmechanics\u002Frotation\u002Frotational-inertia":20472,"\u002Fmechanics\u002Frotation\u002Frotational-dynamics":20473,"\u002Fmechanics\u002Frotation\u002Frolling-motion":20474,"\u002Fmechanics\u002Frotation\u002Fangular-momentum":20475,"\u002Fmechanics\u002Frotation\u002Frolling-resistance":20476,"\u002Fmechanics\u002Frotation\u002Fgyroscopic-precession":20477,"\u002Fmechanics\u002Fgravity-and-matter\u002Fkeplerian-orbits":20478,"\u002Fmechanics\u002Fgravity-and-matter\u002Fgravitational-fields":20479,"\u002Fmechanics\u002Fgravity-and-matter\u002Fstatic-equilibrium":20480,"\u002Fmechanics\u002Fgravity-and-matter\u002Ffluid-statics":20481,"\u002Fmechanics\u002Fgravity-and-matter\u002Ffluid-flow":20482,"\u002Fmechanics\u002Fgravity-and-matter\u002Forbital-motion":20483,"\u002Fmechanics\u002Fgravity-and-matter\u002Fstress-and-elasticity":20484,"\u002Fmechanics\u002Foscillations-waves\u002Fdamped-oscillators":20485,"\u002Fmechanics\u002Foscillations-waves\u002Ftravelling-waves":20486,"\u002Fmechanics\u002Foscillations-waves\u002Fwave-superposition":20487,"\u002Fmechanics\u002Foscillations-waves\u002Fstanding-waves":20488,"\u002Fmechanics\u002Foscillations-waves\u002Fsound-waves":20489,"\u002Fmechanics\u002Foscillations-waves\u002Fdoppler-effect":20490,"\u002Fmechanics\u002Foscillations-waves\u002Fwave-packets":20491,"\u002Fmechanics\u002Foscillations-waves\u002Fbeats-and-coupling":20492,"\u002Fmechanics\u002Foscillations-waves\u002Fsimple-harmonic-motion":20493,"\u002Fmechanics\u002Foscillations-waves\u002Fpendulum-motion":20494,"\u002Fmechanics\u002Foscillations-waves\u002Fdriven-oscillators":20495,"\u002Fmechanics\u002Foscillations-waves\u002Fwave-boundaries":20496,"\u002Fmechanics\u002Fthermodynamics\u002Fkinetic-theory-of-ideal-gases":20497,"\u002Fmechanics\u002Fthermodynamics\u002Ffirst-law-of-thermodynamics":20498,"\u002Fmechanics\u002Fthermodynamics\u002Fentropy-and-the-second-law":20499,"\u002Fmechanics\u002Fthermodynamics\u002Fthermal-processes":20500,"\u002Fmechanics\u002Fthermodynamics\u002Fphase-changes":20501,"\u002Fmechanics\u002Fthermodynamics\u002Fthermal-machines":20502,"\u002Fmechanics":20503,"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Fcharge-and-conductors":20504,"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Fcoulombs-law":20505,"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Felectric-field-and-force":20506,"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Felectric-field-maps":20507,"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Felectric-dipoles":20508,"\u002Felectricity-and-magnetism\u002Fcontinuous-charge-distributions\u002Fcontinuous-charge-fields":20509,"\u002Felectricity-and-magnetism\u002Fcontinuous-charge-distributions\u002Fgauss-law-and-conductors":20510,"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Fpoint-charge-potential":20511,"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Fpotential-gradients-and-equipotentials":20512,"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Felectrostatic-energy-and-pressure":20513,"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Flaplace-boundary-problems":20514,"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Fcontinuous-charge-potentials":20515,"\u002Felectricity-and-magnetism\u002Fcapacitance\u002Fcapacitance-fundamentals":20492,"\u002Felectricity-and-magnetism\u002Fcapacitance\u002Fcapacitor-networks":20516,"\u002Felectricity-and-magnetism\u002Fcapacitance\u002Fcapacitor-energy-and-force":20517,"\u002Felectricity-and-magnetism\u002Fcapacitance\u002Fdielectric-polarization-and-breakdown":20518,"\u002Felectricity-and-magnetism\u002Fdirect-current-circuits\u002Fcurrent-and-resistance":20488,"\u002Felectricity-and-magnetism\u002Fdirect-current-circuits\u002Fkirchhoff-network-analysis":20353,"\u002Felectricity-and-magnetism\u002Fdirect-current-circuits\u002Frc-transients":20519,"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fmagnetic-trajectories":20479,"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fhall-effect":20520,"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fmagnetic-force-on-conductors":20521,"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fmagnetic-dipoles":20522,"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fmass-spectrometry":20523,"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fmoving-charge-fields":20524,"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fbiot-savart-law":20525,"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fcircular-current-loops":20526,"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Famperes-law":20527,"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fgauss-law-for-magnetism":20528,"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fmagnetic-materials":20453,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Fmagnetic-flux":20529,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Ffaradays-law":20530,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Flenzs-law":20531,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Fmotional-emf":20532,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Feddy-currents":20533,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Fself-inductance":20534,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Fmagnetic-energy":20535,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Frl-circuits":20536,"\u002Felectricity-and-magnetism\u002Falternating-current\u002Fac-fundamentals":20471,"\u002Felectricity-and-magnetism\u002Falternating-current\u002Freactance":20470,"\u002Felectricity-and-magnetism\u002Falternating-current\u002Frlc-resonance":20537,"\u002Felectricity-and-magnetism\u002Falternating-current\u002Fac-power":20538,"\u002Felectricity-and-magnetism\u002Falternating-current\u002Ftransformers":20539,"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Fdisplacement-current":20540,"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Felectromagnetic-waves":20541,"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Felectromagnetic-momentum":20542,"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Fdipole-radiation":20543,"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Fpolarization":20544,"\u002Felectricity-and-magnetism\u002Foptics\u002Freflection-and-refraction":20545,"\u002Felectricity-and-magnetism\u002Foptics\u002Fthin-lenses":20497,"\u002Felectricity-and-magnetism\u002Foptics\u002Fspherical-mirrors":20495,"\u002Felectricity-and-magnetism":20546,"\u002Flinear-algebra\u002Flinear-systems\u002Fsystems-and-echelon-forms":20547,"\u002Flinear-algebra\u002Flinear-systems\u002Fvector-and-matrix-equations":20548,"\u002Flinear-algebra\u002Flinear-systems\u002Fsolution-sets-and-applications":20549,"\u002Flinear-algebra\u002Flinear-systems\u002Flinear-independence":20550,"\u002Flinear-algebra\u002Flinear-systems\u002Flinear-transformations":20551,"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fmatrix-operations":20552,"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fmatrix-inverse-and-invertibility":20553,"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fpartitioned-matrices-and-lu":20554,"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fsubspaces-dimension-rank":20555,"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fapplications-leontief-and-graphics":20405,"\u002Flinear-algebra\u002Fdeterminants\u002Fdeterminants-and-cofactors":20556,"\u002Flinear-algebra\u002Fdeterminants\u002Fproperties-of-determinants":20557,"\u002Flinear-algebra\u002Fdeterminants\u002Fcramer-volume-and-area":20409,"\u002Flinear-algebra\u002Fvector-spaces\u002Fvector-spaces-and-subspaces":20558,"\u002Flinear-algebra\u002Fvector-spaces\u002Fnull-and-column-spaces":20559,"\u002Flinear-algebra\u002Fvector-spaces\u002Fbases-and-independent-sets":20560,"\u002Flinear-algebra\u002Fvector-spaces\u002Fcoordinate-systems":20561,"\u002Flinear-algebra\u002Fvector-spaces\u002Fdimension-and-rank":20562,"\u002Flinear-algebra\u002Fvector-spaces\u002Fchange-of-basis":20563,"\u002Flinear-algebra\u002Fvector-spaces\u002Fdifference-equations-and-markov":20564,"\u002Flinear-algebra\u002Feigenvalues\u002Feigenvectors-and-eigenvalues":20565,"\u002Flinear-algebra\u002Feigenvalues\u002Fthe-characteristic-equation":20566,"\u002Flinear-algebra\u002Feigenvalues\u002Fdiagonalization":20567,"\u002Flinear-algebra\u002Feigenvalues\u002Feigenvectors-and-linear-transformations":20568,"\u002Flinear-algebra\u002Feigenvalues\u002Fcomplex-eigenvalues":20569,"\u002Flinear-algebra\u002Feigenvalues\u002Fdynamical-systems":20570,"\u002Flinear-algebra\u002Feigenvalues\u002Fpower-method":20571,"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Finner-product-length-orthogonality":20572,"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Forthogonal-sets-and-projections":20573,"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Fgram-schmidt-and-qr":20574,"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Fleast-squares-problems":20575,"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Fleast-squares-applications":20576,"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Finner-product-spaces":20577,"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fdiagonalizing-symmetric-matrices":20444,"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fquadratic-forms":20578,"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fconstrained-optimization":20579,"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fsingular-value-decomposition":20580,"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fsvd-applications-pca-imaging":20581,"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fnumerical-thinking-and-matrix-computation":20582,"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Flu-and-cholesky":20583,"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fconditioning-and-floating-point":20584,"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fstability-and-error-analysis":20585,"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fqr-and-numerical-least-squares":20586,"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fnumerical-eigenvalues-and-svd":20587,"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Faffine-combinations":20588,"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Faffine-independence-and-barycentric-coordinates":20589,"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Fconvex-combinations-and-convex-sets":20590,"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Fhyperplanes-and-polytopes":20591,"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Fcurves-and-surfaces":20592,"\u002Flinear-algebra":20593,"\u002Ftheory-of-computation":20594,"\u002Fcomputer-architecture\u002Ffoundations\u002Fbits-bytes-and-words":20595,"\u002Fcomputer-architecture\u002Ffoundations\u002Finteger-representation":20596,"\u002Fcomputer-architecture\u002Ffoundations\u002Finteger-arithmetic":20597,"\u002Fcomputer-architecture\u002Ffoundations\u002Ffloating-point":20598,"\u002Fcomputer-architecture\u002Ffoundations\u002Fboolean-algebra-and-bit-manipulation":20599,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fthe-machines-view":20600,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fdata-movement":20601,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Farithmetic-and-logic":20602,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fcontrol-flow":20603,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fprocedures":20604,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Farrays-structs-and-alignment":20605,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fmemory-layout-and-buffer-overflows":20606,"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Fwhat-an-isa-is":20607,"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Finstruction-formats-and-operands":20608,"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Faddressing-modes":20609,"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Fthe-y86-64-instruction-set":20610,"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Fy86-64-programming":20611,"\u002Fcomputer-architecture\u002Fdigital-logic\u002Ftransistors-gates-and-boolean-functions":20612,"\u002Fcomputer-architecture\u002Fdigital-logic\u002Fcombinational-logic-and-hcl":20613,"\u002Fcomputer-architecture\u002Fdigital-logic\u002Fmultiplexers-decoders-and-the-alu":20614,"\u002Fcomputer-architecture\u002Fdigital-logic\u002Fmemory-elements-latches-flip-flops-and-clocking":20615,"\u002Fcomputer-architecture\u002Fdigital-logic\u002Fregister-files-and-random-access-memory":20616,"\u002Fcomputer-architecture\u002Fprocessor-design\u002Fthe-fetch-decode-execute-cycle":20617,"\u002Fcomputer-architecture\u002Fprocessor-design\u002Fthe-seq-stages":20618,"\u002Fcomputer-architecture\u002Fprocessor-design\u002Fcontrol-logic-and-sequencing":20619,"\u002Fcomputer-architecture\u002Fprocessor-design\u002Fassembling-seq":20620,"\u002Fcomputer-architecture\u002Fprocessor-design\u002Ftracing-a-program":20621,"\u002Fcomputer-architecture\u002Fpipelining\u002Fpipelining-principles":20622,"\u002Fcomputer-architecture\u002Fpipelining\u002Ffrom-seq-to-pipe":20623,"\u002Fcomputer-architecture\u002Fpipelining\u002Fdata-hazards-stalling-and-forwarding":20624,"\u002Fcomputer-architecture\u002Fpipelining\u002Fcontrol-hazards-and-branch-prediction":20625,"\u002Fcomputer-architecture\u002Fpipelining\u002Fthe-complete-pipe-processor":20626,"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Fstorage-technologies-and-the-latency-gap":20627,"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Flocality":20628,"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Fcache-memories-direct-mapped":20629,"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Fset-associative-and-write-policies":20630,"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Fcache-performance-and-cache-friendly-code":20631,"\u002Fcomputer-architecture\u002Fvirtual-memory\u002Faddress-spaces-and-translation":20632,"\u002Fcomputer-architecture\u002Fvirtual-memory\u002Fpage-tables-and-page-faults":20633,"\u002Fcomputer-architecture\u002Fvirtual-memory\u002Fthe-tlb-and-multi-level-page-tables":20634,"\u002Fcomputer-architecture\u002Fexceptions-and-io\u002Fexceptional-control-flow":20635,"\u002Fcomputer-architecture\u002Fexceptions-and-io\u002Finterrupts-and-the-kernel":20636,"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fprocesses-threads-and-parallelism":20637,"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fhardware-multithreading":20638,"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fcache-coherence":20639,"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fmemory-consistency-and-synchronization":20640,"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fmulticore-organization":20641,"\u002Fcomputer-architecture\u002Fcapstone\u002Fthe-whole-machine":20642,"\u002Fcomputer-architecture\u002Fcapstone\u002Fassembling-a-complete-cpu":20643,"\u002Fcomputer-architecture":20594,"\u002Fdifferential-equations\u002Ffoundations\u002Fmodels-and-direction-fields":20644,"\u002Fdifferential-equations\u002Ffoundations\u002Fclassification-and-terminology":20645,"\u002Fdifferential-equations\u002Ffirst-order\u002Flinear-first-order-integrating-factors":20646,"\u002Fdifferential-equations\u002Ffirst-order\u002Fseparable-and-exact":20409,"\u002Fdifferential-equations\u002Ffirst-order\u002Fmodeling-first-order":20647,"\u002Fdifferential-equations\u002Ffirst-order\u002Fautonomous-and-population-dynamics":20408,"\u002Fdifferential-equations\u002Ffirst-order\u002Fexistence-uniqueness-euler":20415,"\u002Fdifferential-equations\u002Ffirst-order\u002Ffirst-order-difference-equations":20648,"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fhomogeneous-constant-coefficients":20649,"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fcomplex-and-repeated-roots":20449,"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fnonhomogeneous-undetermined-coefficients":20650,"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fvariation-of-parameters":20651,"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fmechanical-electrical-vibrations":20652,"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fhigher-order-linear":20653,"\u002Fdifferential-equations\u002Fseries-solutions\u002Fpower-series-ordinary-points":20654,"\u002Fdifferential-equations\u002Fseries-solutions\u002Fregular-singular-frobenius":20655,"\u002Fdifferential-equations\u002Fseries-solutions\u002Fbessel-and-special-functions":20656,"\u002Fdifferential-equations\u002Flaplace\u002Flaplace-definition-ivps":20657,"\u002Fdifferential-equations\u002Flaplace\u002Fstep-impulse-convolution":20658,"\u002Fdifferential-equations\u002Fsystems\u002Fmatrices-eigenvalues-review":20659,"\u002Fdifferential-equations\u002Fsystems\u002Fconstant-coefficient-systems-phase-portraits":20660,"\u002Fdifferential-equations\u002Fsystems\u002Frepeated-eigenvalues-fundamental-matrices":20661,"\u002Fdifferential-equations\u002Fnumerical\u002Feuler-and-runge-kutta":20656,"\u002Fdifferential-equations\u002Fnumerical\u002Fmultistep-systems-stability":20662,"\u002Fdifferential-equations\u002Fnonlinear\u002Fphase-plane-autonomous-stability":20663,"\u002Fdifferential-equations\u002Fnonlinear\u002Flocally-linear-and-liapunov":20664,"\u002Fdifferential-equations\u002Fnonlinear\u002Fcompeting-species-predator-prey-limit-cycles":20665,"\u002Fdifferential-equations\u002Fpdes-fourier-bvp\u002Ffourier-series":20666,"\u002Fdifferential-equations\u002Fpdes-fourier-bvp\u002Fheat-wave-laplace-equations":20667,"\u002Fdifferential-equations\u002Fpdes-fourier-bvp\u002Fsturm-liouville":20668,"\u002Fdifferential-equations\u002Fhistory-variations\u002Fcalculus-of-variations":20669,"\u002Fdifferential-equations\u002Fhistory-variations\u002Fhistorical-notes":20670,"\u002Fdifferential-equations":20671,"\u002Frelativity\u002Ffoundations\u002Fspecial-relativity-postulates":20672,"\u002Frelativity\u002Ffoundations\u002Florentz-transformation-spacetime":20673,"\u002Frelativity\u002Ffoundations\u002Ftime-dilation-length-contraction":20674,"\u002Frelativity\u002Ffoundations\u002Frelativistic-momentum-energy":20675,"\u002Frelativity\u002Ffoundations\u002Fgeneral-relativity":20554,"\u002Frelativity\u002Fspacetime-and-the-lorentz-group\u002Fminkowski-spacetime-and-the-interval":20676,"\u002Frelativity\u002Fspacetime-and-the-lorentz-group\u002Ffour-vectors-and-index-notation":20677,"\u002Frelativity\u002Fspacetime-and-the-lorentz-group\u002Fthe-lorentz-group-and-rapidity":20678,"\u002Frelativity\u002Fspacetime-and-the-lorentz-group\u002Fdoppler-aberration-and-appearance":20679,"\u002Frelativity\u002Frelativistic-dynamics\u002Ffour-momentum-force-and-accelerated-motion":20680,"\u002Frelativity\u002Frelativistic-dynamics\u002Fparticle-decays-and-two-body-kinematics":20681,"\u002Frelativity\u002Frelativistic-dynamics\u002Fcollisions-thresholds-and-the-cm-frame":20424,"\u002Frelativity\u002Frelativistic-dynamics\u002Fmandelstam-variables-and-invariants":20682,"\u002Frelativity\u002Fcovariant-electrodynamics\u002Ffour-current-and-the-four-potential":20683,"\u002Frelativity\u002Fcovariant-electrodynamics\u002Fthe-electromagnetic-field-tensor":20684,"\u002Frelativity\u002Fcovariant-electrodynamics\u002Ftransformation-of-electric-and-magnetic-fields":20685,"\u002Frelativity\u002Fcovariant-electrodynamics\u002Fcovariant-maxwell-and-the-stress-energy-tensor":20686,"\u002Frelativity\u002Fcurved-spacetime\u002Fthe-equivalence-principle-formalized":20687,"\u002Frelativity\u002Fcurved-spacetime\u002Fmanifolds-vectors-and-the-metric":20688,"\u002Frelativity\u002Fcurved-spacetime\u002Fcovariant-derivative-and-christoffel-symbols":20689,"\u002Frelativity\u002Fcurved-spacetime\u002Fgeodesics-and-the-geodesic-equation":20690,"\u002Frelativity\u002Fcurved-spacetime\u002Fcurvature-riemann-and-geodesic-deviation":20691,"\u002Frelativity\u002Fcurved-spacetime\u002Fthe-einstein-field-equations":20646,"\u002Frelativity\u002Fthe-schwarzschild-solution\u002Fthe-schwarzschild-metric":20692,"\u002Frelativity\u002Fthe-schwarzschild-solution\u002Fgeodesics-and-orbits-in-schwarzschild":20693,"\u002Frelativity\u002Fthe-schwarzschild-solution\u002Flight-bending-and-null-geodesics":20694,"\u002Frelativity\u002Ftests-of-general-relativity\u002Fperihelion-precession-of-mercury":20695,"\u002Frelativity\u002Ftests-of-general-relativity\u002Fdeflection-of-light-and-gravitational-lensing":20696,"\u002Frelativity\u002Ftests-of-general-relativity\u002Fgravitational-redshift-and-shapiro-delay":20575,"\u002Frelativity\u002Ftests-of-general-relativity\u002Frelativity-in-technology-gps":20697,"\u002Frelativity\u002Fblack-holes\u002Fhorizons-and-coordinate-singularities":20698,"\u002Frelativity\u002Fblack-holes\u002Frotating-and-charged-black-holes":20585,"\u002Frelativity\u002Fblack-holes\u002Fblack-hole-thermodynamics":20699,"\u002Frelativity\u002Fgravitational-waves\u002Flinearized-gravity-and-wave-solutions":20700,"\u002Frelativity\u002Fgravitational-waves\u002Fgeneration-and-the-quadrupole-formula":20701,"\u002Frelativity\u002Fgravitational-waves\u002Fdetection-ligo-and-the-first-events":20702,"\u002Frelativity\u002Fcosmological-bridge\u002Fthe-cosmological-principle-and-flrw-metric":20703,"\u002Frelativity\u002Fcosmological-bridge\u002Ffriedmann-equations-and-cosmic-dynamics":20704,"\u002Frelativity":20705,"\u002Fphysical-computing":20594,"\u002Fquantum-mechanics\u002Fold-quantum-theory\u002Fblackbody-radiation-and-the-planck-quantum":20706,"\u002Fquantum-mechanics\u002Fold-quantum-theory\u002Fthe-photoelectric-effect-and-the-photon":20685,"\u002Fquantum-mechanics\u002Fold-quantum-theory\u002Fx-rays-and-the-compton-effect":20707,"\u002Fquantum-mechanics\u002Fold-quantum-theory\u002Fthe-old-quantum-theory-bohr-and-sommerfeld":20708,"\u002Fquantum-mechanics\u002Fmatter-waves\u002Fde-broglie-waves-and-electron-diffraction":20709,"\u002Fquantum-mechanics\u002Fmatter-waves\u002Fwave-packets-and-the-probability-interpretation":20710,"\u002Fquantum-mechanics\u002Fmatter-waves\u002Fthe-uncertainty-principle":20711,"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fthe-schrodinger-equation-in-one-dimension":20712,"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fthe-free-particle-and-wave-packet-dynamics":20713,"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fparticle-in-infinite-and-finite-square-wells":20661,"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Foperators-expectation-values-and-the-harmonic-oscillator":20585,"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fthe-dirac-delta-potential":20714,"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fbarrier-penetration-and-quantum-tunneling":20715,"\u002Fquantum-mechanics\u002Fformalism\u002Fhilbert-space-and-dirac-notation":20716,"\u002Fquantum-mechanics\u002Fformalism\u002Fobservables-hermitian-operators-and-eigenvalues":20717,"\u002Fquantum-mechanics\u002Fformalism\u002Fthe-postulates-and-quantum-measurement":20713,"\u002Fquantum-mechanics\u002Fformalism\u002Fposition-momentum-and-continuous-spectra":20718,"\u002Fquantum-mechanics\u002Fformalism\u002Fcommutators-and-the-generalized-uncertainty-principle":20693,"\u002Fquantum-mechanics\u002Fformalism\u002Ftime-evolution-schrodinger-and-heisenberg-pictures":20446,"\u002Fquantum-mechanics\u002Foscillator-and-symmetry\u002Fladder-operators-and-the-number-states":20719,"\u002Fquantum-mechanics\u002Foscillator-and-symmetry\u002Fcoherent-and-squeezed-states":20720,"\u002Fquantum-mechanics\u002Foscillator-and-symmetry\u002Fsymmetries-generators-and-conservation-laws":20426,"\u002Fquantum-mechanics\u002Foscillator-and-symmetry\u002Fparity-time-reversal-and-discrete-symmetries":20721,"\u002Fquantum-mechanics\u002Fangular-momentum\u002Forbital-angular-momentum-and-spherical-harmonics":20722,"\u002Fquantum-mechanics\u002Fangular-momentum\u002Fthe-angular-momentum-algebra":20723,"\u002Fquantum-mechanics\u002Fangular-momentum\u002Faddition-of-angular-momenta-and-clebsch-gordan":20724,"\u002Fquantum-mechanics\u002Fcentral-potentials\u002Fthe-schrodinger-equation-in-three-dimensions":20725,"\u002Fquantum-mechanics\u002Fcentral-potentials\u002Fthe-hydrogen-atom":20726,"\u002Fquantum-mechanics\u002Fcentral-potentials\u002Fthe-isotropic-oscillator-and-hidden-symmetry":20727,"\u002Fquantum-mechanics\u002Fspin\u002Fspin-half-pauli-matrices-and-stern-gerlach":20728,"\u002Fquantum-mechanics\u002Fspin\u002Fspin-in-a-magnetic-field-precession-and-resonance":20729,"\u002Fquantum-mechanics\u002Fspin\u002Ftwo-level-systems-and-the-bloch-sphere":20685,"\u002Fquantum-mechanics\u002Fidentical-particles\u002Fidentical-particles-and-exchange-symmetry":20730,"\u002Fquantum-mechanics\u002Fidentical-particles\u002Fthe-pauli-principle-atoms-and-the-periodic-table":20731,"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Ftime-independent-perturbation-theory":20732,"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Ffine-structure-and-the-real-hydrogen-atom":20719,"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Fthe-zeeman-and-stark-effects":20407,"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Fthe-variational-method":20733,"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Fthe-wkb-approximation":20734,"\u002Fquantum-mechanics":20735,"\u002Freal-analysis\u002Ffoundations\u002Fsets-logic-functions":20669,"\u002Freal-analysis\u002Ffoundations\u002Fordered-fields-completeness":20736,"\u002Freal-analysis\u002Ffoundations\u002Fabsolute-value-bounds":20737,"\u002Freal-analysis\u002Ffoundations\u002Fintervals-uncountability":20558,"\u002Freal-analysis\u002Fsequences-series\u002Fsequences-limits":20738,"\u002Freal-analysis\u002Fsequences-series\u002Flimit-laws-monotone":20444,"\u002Freal-analysis\u002Fsequences-series\u002Flimsup-bolzano-weierstrass":20739,"\u002Freal-analysis\u002Fsequences-series\u002Fcauchy-completeness":20740,"\u002Freal-analysis\u002Fsequences-series\u002Fseries-convergence":20581,"\u002Freal-analysis\u002Fsequences-series\u002Fabsolute-conditional-rearrangement":20691,"\u002Freal-analysis\u002Fmetric-spaces\u002Fmetric-spaces-norms":20741,"\u002Freal-analysis\u002Fmetric-spaces\u002Fopen-closed-sets":20742,"\u002Freal-analysis\u002Fmetric-spaces\u002Fconvergence-completeness":20743,"\u002Freal-analysis\u002Fmetric-spaces\u002Fcompactness":20744,"\u002Freal-analysis\u002Fmetric-spaces\u002Fconnectedness":20745,"\u002Freal-analysis\u002Fcontinuity\u002Flimits-of-functions":20718,"\u002Freal-analysis\u002Fcontinuity\u002Fcontinuous-functions":20746,"\u002Freal-analysis\u002Fcontinuity\u002Fevt-ivt":20551,"\u002Freal-analysis\u002Fcontinuity\u002Funiform-continuity":20747,"\u002Freal-analysis\u002Fcontinuity\u002Fcontinuity-metric-spaces":20748,"\u002Freal-analysis\u002Fcontinuity\u002Flimits-infinity-monotone":20405,"\u002Freal-analysis\u002Fdifferentiation\u002Fthe-derivative":20749,"\u002Freal-analysis\u002Fdifferentiation\u002Fmean-value-theorem":20750,"\u002Freal-analysis\u002Fdifferentiation\u002Ftaylors-theorem":20707,"\u002Freal-analysis\u002Fdifferentiation\u002Finverse-function-1d":20434,"\u002Freal-analysis\u002Friemann-integration\u002Fdarboux-integral":20585,"\u002Freal-analysis\u002Friemann-integration\u002Fintegrability-classes":20751,"\u002Freal-analysis\u002Friemann-integration\u002Fproperties-of-the-integral":20752,"\u002Freal-analysis\u002Friemann-integration\u002Ffundamental-theorem":20570,"\u002Freal-analysis\u002Friemann-integration\u002Flog-exp-improper":20695,"\u002Freal-analysis\u002Ffunction-sequences\u002Fpointwise-uniform-convergence":20753,"\u002Freal-analysis\u002Ffunction-sequences\u002Finterchange-of-limits":20754,"\u002Freal-analysis\u002Ffunction-sequences\u002Fpower-series-weierstrass":20755,"\u002Freal-analysis\u002Ffunction-sequences\u002Fpicard-ode":20590,"\u002Freal-analysis\u002Fseveral-variables\u002Fdifferentiability-rn":20756,"\u002Freal-analysis\u002Fseveral-variables\u002Fgradient-chain-rule":20757,"\u002Freal-analysis\u002Fseveral-variables\u002Fhigher-derivatives-taylor-extrema":20758,"\u002Freal-analysis\u002Fseveral-variables\u002Finverse-implicit-theorems":20758,"\u002Freal-analysis\u002Fseveral-variables\u002Fmultiple-integrals":20759,"\u002Freal-analysis":20760,"\u002Fabstract-algebra\u002Ffoundations\u002Fsets-functions-relations":20761,"\u002Fabstract-algebra\u002Ffoundations\u002Fintegers-and-modular-arithmetic":20762,"\u002Fabstract-algebra\u002Fgroups-and-symmetry\u002Fgroup-axioms-and-first-examples":20763,"\u002Fabstract-algebra\u002Fgroups-and-symmetry\u002Fdihedral-and-symmetric-groups":20764,"\u002Fabstract-algebra\u002Fgroups-and-symmetry\u002Fmatrix-and-quaternion-groups":20765,"\u002Fabstract-algebra\u002Fgroups-and-symmetry\u002Fhomomorphisms-and-group-actions":20766,"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fsubgroups-and-substructures":20767,"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fcyclic-groups":20768,"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fgeneration-and-subgroup-lattices":20769,"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fcosets-lagrange-and-normal-subgroups":20770,"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fisomorphism-theorems":20734,"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fcomposition-series-and-the-alternating-group":20771,"\u002Fabstract-algebra\u002Fgroup-actions-and-sylow\u002Factions-and-cayleys-theorem":20763,"\u002Fabstract-algebra\u002Fgroup-actions-and-sylow\u002Fconjugation-and-the-class-equation":20680,"\u002Fabstract-algebra\u002Fgroup-actions-and-sylow\u002Fsylow-theorems":20772,"\u002Fabstract-algebra\u002Fgroup-actions-and-sylow\u002Fautomorphisms-and-simple-groups":20773,"\u002Fabstract-algebra\u002Fproducts-and-group-structure\u002Fdirect-products-and-finite-abelian-groups":20774,"\u002Fabstract-algebra\u002Fproducts-and-group-structure\u002Fsemidirect-products":20775,"\u002Fabstract-algebra\u002Fproducts-and-group-structure\u002Fnilpotent-and-solvable-groups":20776,"\u002Fabstract-algebra\u002Fproducts-and-group-structure\u002Fclassifying-small-groups":20777,"\u002Fabstract-algebra\u002Fring-theory\u002Frings-definitions-and-examples":20778,"\u002Fabstract-algebra\u002Fring-theory\u002Fideals-quotients-and-homomorphisms":20779,"\u002Fabstract-algebra\u002Fring-theory\u002Ffractions-and-the-chinese-remainder-theorem":20773,"\u002Fabstract-algebra\u002Ffactorization-and-polynomials\u002Feuclidean-domains-pids-ufds":20780,"\u002Fabstract-algebra\u002Ffactorization-and-polynomials\u002Fpolynomial-rings-over-fields":20749,"\u002Fabstract-algebra\u002Ffactorization-and-polynomials\u002Fgauss-lemma-and-unique-factorization":20781,"\u002Fabstract-algebra\u002Ffactorization-and-polynomials\u002Firreducibility-criteria-and-groebner":20782,"\u002Fabstract-algebra\u002Fmodule-theory\u002Fintroduction-to-modules":20783,"\u002Fabstract-algebra\u002Fmodule-theory\u002Ffree-modules-and-direct-sums":20784,"\u002Fabstract-algebra\u002Fmodule-theory\u002Ftensor-products-and-exact-sequences":20785,"\u002Fabstract-algebra\u002Fmodule-theory\u002Fvector-spaces-and-linear-maps":20786,"\u002Fabstract-algebra\u002Fmodules-over-pids\u002Fstructure-theorem-over-pids":20787,"\u002Fabstract-algebra\u002Fmodules-over-pids\u002Frational-canonical-form":20788,"\u002Fabstract-algebra\u002Fmodules-over-pids\u002Fjordan-canonical-form":20789,"\u002Fabstract-algebra\u002Ffield-theory\u002Ffield-extensions-and-algebraic-elements":20790,"\u002Fabstract-algebra\u002Ffield-theory\u002Fstraightedge-and-compass-constructions":20423,"\u002Fabstract-algebra\u002Ffield-theory\u002Fsplitting-fields-and-algebraic-closure":20791,"\u002Fabstract-algebra\u002Ffield-theory\u002Fseparable-and-cyclotomic-extensions":20792,"\u002Fabstract-algebra\u002Fgalois-theory\u002Fthe-galois-correspondence":20665,"\u002Fabstract-algebra\u002Fgalois-theory\u002Ffinite-fields":20793,"\u002Fabstract-algebra\u002Fgalois-theory\u002Fcyclotomic-and-abelian-extensions":20794,"\u002Fabstract-algebra\u002Fgalois-theory\u002Fgalois-groups-of-polynomials":20722,"\u002Fabstract-algebra\u002Fgalois-theory\u002Fsolvability-by-radicals-and-the-quintic":20794,"\u002Fabstract-algebra\u002Fcapstone\u002Fcommutative-algebra-and-algebraic-geometry":20795,"\u002Fabstract-algebra\u002Fcapstone\u002Frepresentation-and-character-theory":20796,"\u002Fabstract-algebra":20797,"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fatomic-spectra-rutherford":20798,"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fbohr-model-hydrogen":20799,"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fx-ray-spectra-franck-hertz":20800,"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fbohr-sommerfeld-old-quantum-theory":20801,"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fold-quantum-theory-limits-wkb":20802,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fschrodinger-3d-hydrogen":20714,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fhydrogen-wave-functions":20803,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fradial-equation-in-full":20804,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fsymmetry-degeneracy-runge-lenz":20805,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fexpectation-values-virial":20806,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fquantum-defects-alkali-spectra":20807,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Frydberg-atoms":20808,"\u002Fatomic-physics\u002Ffine-structure-and-the-dirac-atom\u002Frelativistic-kinetic-correction":20809,"\u002Fatomic-physics\u002Ffine-structure-and-the-dirac-atom\u002Fspin-orbit-thomas-precession":20548,"\u002Fatomic-physics\u002Ffine-structure-and-the-dirac-atom\u002Fdarwin-term-fine-structure-formula":20677,"\u002Fatomic-physics\u002Ffine-structure-and-the-dirac-atom\u002Fdirac-equation-hydrogen":20423,"\u002Fatomic-physics\u002Fqed-corrections-and-hyperfine-structure\u002Flamb-shift-qed":20810,"\u002Fatomic-physics\u002Fqed-corrections-and-hyperfine-structure\u002Fhyperfine-structure-21cm":20407,"\u002Fatomic-physics\u002Fqed-corrections-and-hyperfine-structure\u002Fnuclear-effects-isotope-shift":20811,"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fperiodic-table-atomic-spectra":20812,"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fcentral-field-self-consistent":20445,"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fidentical-particles-hartree-fock":20739,"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fhelium-two-electron-atom":20813,"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fls-jj-coupling-term-symbols":20814,"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fhund-rules-ground-terms":20815,"\u002Fatomic-physics\u002Fatoms-in-external-fields\u002Fzeeman-effect":20816,"\u002Fatomic-physics\u002Fatoms-in-external-fields\u002Fpaschen-back-intermediate":20817,"\u002Fatomic-physics\u002Fatoms-in-external-fields\u002Fstark-effect-polarizability":20818,"\u002Fatomic-physics\u002Fradiative-transitions-and-line-shapes\u002Ftime-dependent-perturbation-golden-rule":20819,"\u002Fatomic-physics\u002Fradiative-transitions-and-line-shapes\u002Fdipole-approximation-einstein-coefficients":20820,"\u002Fatomic-physics\u002Fradiative-transitions-and-line-shapes\u002Fselection-rules-forbidden-transitions":20821,"\u002Fatomic-physics\u002Fradiative-transitions-and-line-shapes\u002Flifetimes-and-line-shapes":20822,"\u002Fatomic-physics\u002Flasers-and-spectroscopy\u002Flaser-principles":20823,"\u002Fatomic-physics\u002Flasers-and-spectroscopy\u002Fspectroscopy-techniques":20824,"\u002Fatomic-physics\u002Flasers-and-spectroscopy\u002Fline-catalog-nist-asd":20825,"\u002Fatomic-physics\u002Fmodern-atomic-physics\u002Flaser-cooling-doppler":20826,"\u002Fatomic-physics\u002Fmodern-atomic-physics\u002Fsub-doppler-trapping":20770,"\u002Fatomic-physics\u002Fmodern-atomic-physics\u002Fbose-einstein-condensation":20827,"\u002Fatomic-physics\u002Fmodern-atomic-physics\u002Foptical-clocks-precision":20828,"\u002Fatomic-physics":20829,"\u002Fdatabases":20594,"\u002Fcategory-theory\u002Ffoundations\u002Fwhat-is-a-category":20830,"\u002Fcategory-theory\u002Ffoundations\u002Fexamples-of-categories":20831,"\u002Fcategory-theory\u002Ffoundations\u002Fspecial-morphisms":20832,"\u002Fcategory-theory\u002Ffoundations\u002Ffunctors":20761,"\u002Fcategory-theory\u002Ffoundations\u002Fnatural-transformations":20833,"\u002Fcategory-theory\u002Ffoundations\u002Fsize-and-set-theory":20834,"\u002Fcategory-theory\u002Funiversal-properties\u002Funiversal-properties":20835,"\u002Fcategory-theory\u002Funiversal-properties\u002Fproducts-and-coproducts":20836,"\u002Fcategory-theory\u002Funiversal-properties\u002Fconstructions-on-categories":20837,"\u002Fcategory-theory\u002Frepresentables-yoneda\u002Frepresentable-functors":20838,"\u002Fcategory-theory\u002Frepresentables-yoneda\u002Fyoneda-lemma":20839,"\u002Fcategory-theory\u002Frepresentables-yoneda\u002Fyoneda-consequences":20840,"\u002Fcategory-theory\u002Flimits-colimits\u002Flimits":20841,"\u002Fcategory-theory\u002Flimits-colimits\u002Fproducts-equalizers-pullbacks":20842,"\u002Fcategory-theory\u002Flimits-colimits\u002Fcolimits":20843,"\u002Fcategory-theory\u002Flimits-colimits\u002Fcomputing-limits":20844,"\u002Fcategory-theory\u002Flimits-colimits\u002Flimits-and-functors":20845,"\u002Fcategory-theory\u002Fadjunctions\u002Fadjunctions":20846,"\u002Fcategory-theory\u002Fadjunctions\u002Funits-and-counits":20847,"\u002Fcategory-theory\u002Fadjunctions\u002Fadjunctions-via-universal-arrows":20848,"\u002Fcategory-theory\u002Fadjunctions\u002Ffree-forgetful-adjunctions":20849,"\u002Fcategory-theory\u002Fadjoints-limits\u002Flimits-via-adjoints":20850,"\u002Fcategory-theory\u002Fadjoints-limits\u002Fpresheaf-limits-colimits":20851,"\u002Fcategory-theory\u002Fadjoints-limits\u002Fadjoints-preserve-limits":20852,"\u002Fcategory-theory\u002Fadjoints-limits\u002Fadjoint-functor-theorem":20843,"\u002Fcategory-theory\u002Fmonads-algebras\u002Fmonads":20853,"\u002Fcategory-theory\u002Fmonads-algebras\u002Falgebras-eilenberg-moore":20854,"\u002Fcategory-theory\u002Fmonads-algebras\u002Fkleisli-and-programming":20855,"\u002Fcategory-theory\u002Fmonads-algebras\u002Falgebras-for-endofunctors":20856,"\u002Fcategory-theory\u002Fcartesian-closed-lambda\u002Fcartesian-closed-categories":20857,"\u002Fcategory-theory\u002Fcartesian-closed-lambda\u002Flambda-calculus-correspondence":20802,"\u002Fcategory-theory\u002Fcartesian-closed-lambda\u002Ffixed-points-and-recursion":20858,"\u002Fcategory-theory":20859,"\u002Fdeep-learning\u002Fmathematical-background\u002Flinear-algebra-for-deep-learning":20860,"\u002Fdeep-learning\u002Fmathematical-background\u002Fprobability-and-information-theory":20861,"\u002Fdeep-learning\u002Fmathematical-background\u002Fnumerical-computation":20862,"\u002Fdeep-learning\u002Fmathematical-background\u002Fcalculus":20863,"\u002Fdeep-learning\u002Ffoundations\u002Fwhat-is-deep-learning":20864,"\u002Fdeep-learning\u002Ffoundations\u002Fmachine-learning-refresher":20865,"\u002Fdeep-learning\u002Ffoundations\u002Flinear-models-and-the-perceptron":20824,"\u002Fdeep-learning\u002Fneural-networks\u002Fthe-multilayer-perceptron":20866,"\u002Fdeep-learning\u002Fneural-networks\u002Factivation-functions":20302,"\u002Fdeep-learning\u002Fneural-networks\u002Funiversal-approximation":20867,"\u002Fdeep-learning\u002Fneural-networks\u002Fbackpropagation":20868,"\u002Fdeep-learning\u002Fneural-networks\u002Floss-functions-and-output-units":20869,"\u002Fdeep-learning\u002Foptimization\u002Fgradient-descent-and-sgd":20870,"\u002Fdeep-learning\u002Foptimization\u002Fmomentum-and-adaptive-methods":20871,"\u002Fdeep-learning\u002Foptimization\u002Finitialization":20872,"\u002Fdeep-learning\u002Foptimization\u002Fthe-optimization-landscape":20873,"\u002Fdeep-learning\u002Foptimization\u002Fsecond-order-and-approximate-methods":20874,"\u002Fdeep-learning\u002Fregularization\u002Fregularization-overview":20875,"\u002Fdeep-learning\u002Fregularization\u002Fdropout-and-data-augmentation":20876,"\u002Fdeep-learning\u002Fregularization\u002Fearly-stopping-and-parameter-sharing":20877,"\u002Fdeep-learning\u002Fregularization\u002Fnormalization":20878,"\u002Fdeep-learning\u002Farchitectures\u002Fconvolutional-networks":20879,"\u002Fdeep-learning\u002Farchitectures\u002Fcnn-architectures":20880,"\u002Fdeep-learning\u002Farchitectures\u002Frecurrent-networks":20881,"\u002Fdeep-learning\u002Farchitectures\u002Flstm-and-gru":20882,"\u002Fdeep-learning\u002Farchitectures\u002Fattention-and-transformers":20883,"\u002Fdeep-learning\u002Farchitectures\u002Fthe-transformer-architecture":20884,"\u002Fdeep-learning\u002Farchitectures\u002Ftransformers-in-practice":20885,"\u002Fdeep-learning\u002Farchitectures\u002Fgraph-neural-networks":20886,"\u002Fdeep-learning\u002Farchitectures\u002Fstate-space-models":20887,"\u002Fdeep-learning\u002Ftheory\u002Fgeneralization-theory":20888,"\u002Fdeep-learning\u002Ftheory\u002Fadversarial-robustness":20889,"\u002Fdeep-learning\u002Ftheory\u002Fadversarial-defenses":20890,"\u002Fdeep-learning\u002Ftheory\u002Fbayesian-and-ensemble-methods":20891,"\u002Fdeep-learning\u002Ftheory\u002Fdeep-equilibrium-models":20836,"\u002Fdeep-learning\u002Fgenerative-models\u002Flinear-factor-models":20892,"\u002Fdeep-learning\u002Fgenerative-models\u002Fautoencoders":20893,"\u002Fdeep-learning\u002Fgenerative-models\u002Fvariational-autoencoders":20894,"\u002Fdeep-learning\u002Fgenerative-models\u002Fgenerative-adversarial-networks":20895,"\u002Fdeep-learning\u002Fgenerative-models\u002Fautoregressive-and-normalizing-flows":20896,"\u002Fdeep-learning\u002Fgenerative-models\u002Fenergy-based-and-boltzmann-machines":20897,"\u002Fdeep-learning\u002Fgenerative-models\u002Fdiffusion-and-score-based-models":20898,"\u002Fdeep-learning\u002Fprobabilistic-methods\u002Fstructured-probabilistic-models":20388,"\u002Fdeep-learning\u002Fprobabilistic-methods\u002Fmonte-carlo-and-mcmc":20899,"\u002Fdeep-learning\u002Fprobabilistic-methods\u002Fapproximate-inference":20900,"\u002Fdeep-learning\u002Fpractical\u002Fpractical-methodology":20624,"\u002Fdeep-learning\u002Fpractical\u002Fhyperparameters-and-debugging":20901,"\u002Fdeep-learning\u002Fpractical\u002Frepresentation-learning":20902,"\u002Fdeep-learning\u002Fpractical\u002Ftransfer-learning":20903,"\u002Fdeep-learning\u002Fpractical\u002Fapplications":20904,"\u002Fdeep-learning\u002Fpractical\u002Fmodel-compression-and-distillation":20905,"\u002Fdeep-learning\u002Fpractical\u002Fmeta-learning-and-few-shot":20906,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Flarge-language-models":20907,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fscaling-inference-and-alignment":20908,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fseq2seq-pretraining-and-bart":20909,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Ftext-to-text-transfer-and-conditional-generation":20910,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fspeech-and-audio-models":20911,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fself-supervised-speech-and-synthesis":20912,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fai-agents":20597,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fagent-memory-retrieval-and-orchestration":20913,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fmixture-of-experts":20914,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fmultimodal-models":20915,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Ffusion-and-vision-language-models":20916,"\u002Fdeep-learning\u002Freinforcement-learning\u002Ffoundations-of-reinforcement-learning":20634,"\u002Fdeep-learning\u002Freinforcement-learning\u002Fmodel-free-prediction-and-control":20917,"\u002Fdeep-learning\u002Freinforcement-learning\u002Fdeep-q-networks":20918,"\u002Fdeep-learning\u002Freinforcement-learning\u002Fpolicy-gradients-and-actor-critic":20919,"\u002Fdeep-learning\u002Freinforcement-learning\u002Frl-from-human-feedback":20920,"\u002Fdeep-learning":20594,"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Fequilibrium-state-variables-zeroth-law":20921,"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Ffirst-law-heat-and-work":20719,"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Fsecond-law-entropy-and-the-carnot-bound":20922,"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Fthermodynamic-potentials-and-maxwell-relations":20923,"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Fstability-response-functions-and-the-third-law":20924,"\u002Fstatistical-mechanics\u002Ffoundations\u002Fclassical-statistics-and-equipartition":20925,"\u002Fstatistical-mechanics\u002Ffoundations\u002Fphase-space-and-liouvilles-theorem":20926,"\u002Fstatistical-mechanics\u002Ffoundations\u002Fensembles-and-the-equal-probability-postulate":20927,"\u002Fstatistical-mechanics\u002Ffoundations\u002Fstatistical-entropy-boltzmann-and-gibbs":20928,"\u002Fstatistical-mechanics\u002Fmicrocanonical\u002Fmicrocanonical-ensemble-and-entropy":20929,"\u002Fstatistical-mechanics\u002Fmicrocanonical\u002Fequilibrium-conditions-temperature-pressure-chemical-potential":20766,"\u002Fstatistical-mechanics\u002Fmicrocanonical\u002Fideal-gas-phase-space-and-the-sackur-tetrode-entropy":20930,"\u002Fstatistical-mechanics\u002Fmicrocanonical\u002Ftwo-state-systems-paramagnets-and-negative-temperature":20931,"\u002Fstatistical-mechanics\u002Fcanonical\u002Fcanonical-ensemble-and-the-boltzmann-distribution":20932,"\u002Fstatistical-mechanics\u002Fcanonical\u002Fpartition-function-and-the-helmholtz-free-energy":20653,"\u002Fstatistical-mechanics\u002Fcanonical\u002Fenergy-fluctuations-and-ensemble-equivalence":20446,"\u002Fstatistical-mechanics\u002Fcanonical\u002Fthe-einstein-solid-and-harmonic-systems":20933,"\u002Fstatistical-mechanics\u002Fcanonical\u002Fparamagnetism-and-the-schottky-anomaly":20934,"\u002Fstatistical-mechanics\u002Fclassical-gas\u002Fideal-gas-partition-function-and-the-gibbs-paradox":20935,"\u002Fstatistical-mechanics\u002Fclassical-gas\u002Fequipartition-and-the-virial-theorem":20580,"\u002Fstatistical-mechanics\u002Fclassical-gas\u002Fmolecular-gases-rotation-and-vibration":20936,"\u002Fstatistical-mechanics\u002Fgrand-canonical\u002Fgrand-canonical-ensemble-and-the-grand-partition-function":20937,"\u002Fstatistical-mechanics\u002Fgrand-canonical\u002Fchemical-potential-fugacity-and-number-fluctuations":20658,"\u002Fstatistical-mechanics\u002Fgrand-canonical\u002Fensemble-summary-and-the-thermodynamic-web":20585,"\u002Fstatistical-mechanics\u002Fquantum-statistics\u002Fquantum-statistics-bose-einstein-and-fermi-dirac":20938,"\u002Fstatistical-mechanics\u002Fquantum-statistics\u002Fderiving-the-quantum-distributions":20554,"\u002Fstatistical-mechanics\u002Fquantum-statistics\u002Fthe-classical-limit-and-quantum-concentration":20424,"\u002Fstatistical-mechanics\u002Fquantum-statistics\u002Fideal-quantum-gases-general-framework":20440,"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fbose-einstein-condensation-and-the-fermion-gas":20939,"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fthe-photon-gas-and-plancks-radiation-law":20940,"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fblackbody-thermodynamics-and-radiation-pressure":20941,"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fphonons-and-the-debye-model":20572,"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fbose-einstein-condensation-derived":20942,"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fthermodynamics-of-the-bose-gas-and-superfluidity":20434,"\u002Fstatistical-mechanics\u002Ffermi-gas\u002Fthe-ideal-fermi-gas-at-zero-temperature":20943,"\u002Fstatistical-mechanics\u002Ffermi-gas\u002Fsommerfeld-expansion-and-electrons-in-metals":20944,"\u002Fstatistical-mechanics\u002Ffermi-gas\u002Fwhite-dwarfs-and-the-chandrasekhar-limit":20945,"\u002Fstatistical-mechanics\u002Ffermi-gas\u002Fneutron-stars-and-nuclear-matter":20946,"\u002Fstatistical-mechanics\u002Finteractions\u002Fthe-cluster-expansion-and-virial-coefficients":20768,"\u002Fstatistical-mechanics\u002Finteractions\u002Fthe-van-der-waals-gas-and-liquid-gas-coexistence":20947,"\u002Fstatistical-mechanics\u002Finteractions\u002Fquantum-gases-with-interactions-and-exchange":20948,"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fphases-coexistence-and-classification":20949,"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fthe-ising-model-and-exact-solutions":20950,"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fmean-field-theory-and-the-weiss-model":20714,"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fcritical-exponents-and-landau-theory":20951,"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fthe-renormalization-group-idea":20692,"\u002Fstatistical-mechanics\u002Ffluctuations\u002Fthermodynamic-fluctuations-and-response":20952,"\u002Fstatistical-mechanics\u002Ffluctuations\u002Fbrownian-motion-and-the-langevin-equation":20553,"\u002Fstatistical-mechanics\u002Ffluctuations\u002Flinear-response-and-the-fluctuation-dissipation-theorem":20953,"\u002Fstatistical-mechanics":20954,"\u002Fcondensed-matter\u002Fmolecules-and-bonding\u002Fbonding-mechanisms":20955,"\u002Fcondensed-matter\u002Fmolecules-and-bonding\u002Fmolecular-orbitals-and-h2-plus":20419,"\u002Fcondensed-matter\u002Fmolecules-and-bonding\u002Fhydrogen-molecule-and-exchange":20679,"\u002Fcondensed-matter\u002Fmolecules-and-bonding\u002Fvan-der-waals-forces":20956,"\u002Fcondensed-matter\u002Fmolecular-spectra\u002Frotational-vibrational-spectra":20957,"\u002Fcondensed-matter\u002Fmolecular-spectra\u002Fanharmonicity-and-rovibrational-structure":20958,"\u002Fcondensed-matter\u002Fmolecular-spectra\u002Framan-and-electronic-bands":20959,"\u002Fcondensed-matter\u002Fmolecular-spectra\u002Flasers-and-masers":20960,"\u002Fcondensed-matter\u002Fcrystal-structure\u002Fstructure-of-solids":20961,"\u002Fcondensed-matter\u002Fcrystal-structure\u002Fbravais-lattices-and-crystal-systems":20552,"\u002Fcondensed-matter\u002Fcrystal-structure\u002Freciprocal-lattice-and-brillouin-zones":20962,"\u002Fcondensed-matter\u002Fcrystal-structure\u002Fdiffraction-and-structure-factors":20963,"\u002Fcondensed-matter\u002Flattice-dynamics\u002Fphonon-dispersion":20964,"\u002Fcondensed-matter\u002Flattice-dynamics\u002Fphonons-quantization-and-dos":20965,"\u002Fcondensed-matter\u002Flattice-dynamics\u002Fdebye-einstein-heat-capacity":20695,"\u002Fcondensed-matter\u002Flattice-dynamics\u002Fanharmonicity-and-thermal-transport":20966,"\u002Fcondensed-matter\u002Ffree-electron-fermi-gas\u002Ffree-electron-gas-and-conduction":20967,"\u002Fcondensed-matter\u002Ffree-electron-fermi-gas\u002Fsommerfeld-model-and-heat-capacity":20968,"\u002Fcondensed-matter\u002Ffree-electron-fermi-gas\u002Ftransport-and-the-hall-effect":20969,"\u002Fcondensed-matter\u002Ffree-electron-fermi-gas\u002Fscreening-and-plasmons":20970,"\u002Fcondensed-matter\u002Fband-theory\u002Fblochs-theorem-and-energy-bands":20750,"\u002Fcondensed-matter\u002Fband-theory\u002Fnearly-free-electron-model":20694,"\u002Fcondensed-matter\u002Fband-theory\u002Ftight-binding-method":20971,"\u002Fcondensed-matter\u002Fband-theory\u002Ffermi-surfaces-and-semiclassical-dynamics":20972,"\u002Fcondensed-matter\u002Fsemiconductors\u002Fsemiconductor-bands-and-junctions":20973,"\u002Fcondensed-matter\u002Fsemiconductors\u002Fintrinsic-and-extrinsic-semiconductors":20974,"\u002Fcondensed-matter\u002Fsemiconductors\u002Fcarrier-transport-and-recombination":20685,"\u002Fcondensed-matter\u002Fsemiconductors\u002Fthe-pn-junction":20975,"\u002Fcondensed-matter\u002Fsemiconductors\u002Ftransistors-and-optoelectronics":20976,"\u002Fcondensed-matter\u002Fdielectrics-and-ferroelectrics\u002Fdielectrics-and-polarization":20921,"\u002Fcondensed-matter\u002Fdielectrics-and-ferroelectrics\u002Fferroelectrics-and-piezoelectrics":20822,"\u002Fcondensed-matter\u002Fmagnetism\u002Fdiamagnetism-and-paramagnetism":20549,"\u002Fcondensed-matter\u002Fmagnetism\u002Fexchange-and-ferromagnetism":20977,"\u002Fcondensed-matter\u002Fmagnetism\u002Fantiferromagnetism-and-domains":20415,"\u002Fcondensed-matter\u002Fmagnetism\u002Fspin-waves-and-magnons":20978,"\u002Fcondensed-matter\u002Fsuperconductivity\u002Fsuperconductivity-phenomenology":20979,"\u002Fcondensed-matter\u002Fsuperconductivity\u002Flondon-theory-and-the-meissner-effect":20560,"\u002Fcondensed-matter\u002Fsuperconductivity\u002Fginzburg-landau-theory":20980,"\u002Fcondensed-matter\u002Fsuperconductivity\u002Fbcs-theory":20815,"\u002Fcondensed-matter\u002Fsuperconductivity\u002Fjosephson-and-high-tc":20981,"\u002Fcondensed-matter\u002Fnanostructures\u002Fquantum-wells-wires-and-dots":20407,"\u002Fcondensed-matter\u002Fnanostructures\u002Finteger-quantum-hall-effect":20982,"\u002Fcondensed-matter\u002Fnanostructures\u002Ffractional-quantum-hall-and-topology":20418,"\u002Fcondensed-matter\u002Fnanostructures\u002Fgraphene-and-dirac-materials":20983,"\u002Fcondensed-matter":20735,"\u002Flogic\u002Ffoundations\u002Flogic-as-a-mathematical-model":20984,"\u002Flogic\u002Fsentential-logic\u002Fformal-languages-and-well-formed-formulas":20985,"\u002Flogic\u002Fsentential-logic\u002Ftruth-assignments-and-tautologies":20986,"\u002Flogic\u002Fsentential-logic\u002Funique-readability-and-parsing":20987,"\u002Flogic\u002Fsentential-logic\u002Finduction-and-recursion":20432,"\u002Flogic\u002Fsentential-logic\u002Fexpressive-completeness-and-normal-forms":20988,"\u002Flogic\u002Fsentential-logic\u002Fboolean-circuits":20989,"\u002Flogic\u002Fsentential-logic\u002Fcompactness-and-effectiveness":20432,"\u002Flogic\u002Ffirst-order-languages\u002Ffirst-order-languages":20990,"\u002Flogic\u002Ffirst-order-languages\u002Fstructures-truth-and-satisfaction":20845,"\u002Flogic\u002Ffirst-order-languages\u002Fdefinability-and-elementary-equivalence":20991,"\u002Flogic\u002Ffirst-order-languages\u002Fterms-substitution-and-parsing":20992,"\u002Flogic\u002Fdeductive-calculus\u002Fa-deductive-calculus":20993,"\u002Flogic\u002Fdeductive-calculus\u002Fdeduction-theorem-and-derived-rules":20991,"\u002Flogic\u002Fdeductive-calculus\u002Fsoundness":20994,"\u002Flogic\u002Fdeductive-calculus\u002Fcompleteness-and-consistency":20995,"\u002Flogic\u002Fmodels-and-theories\u002Fcompactness-and-lowenheim-skolem":20996,"\u002Flogic\u002Fmodels-and-theories\u002Ftheories-elementary-classes-and-categoricity":20997,"\u002Flogic\u002Fmodels-and-theories\u002Finterpretations-between-theories":20998,"\u002Flogic\u002Fmodels-and-theories\u002Fnonstandard-analysis":20999,"\u002Flogic\u002Farithmetic-and-definability\u002Fdefinability-in-arithmetic":21000,"\u002Flogic\u002Farithmetic-and-definability\u002Fnatural-numbers-with-successor":21001,"\u002Flogic\u002Farithmetic-and-definability\u002Fpresburger-and-reducts":20922,"\u002Flogic\u002Farithmetic-and-definability\u002Fa-subtheory-and-representability":21002,"\u002Flogic\u002Fincompleteness\u002Farithmetization-of-syntax":20995,"\u002Flogic\u002Fincompleteness\u002Fincompleteness-and-undecidability":21003,"\u002Flogic\u002Fincompleteness\u002Fsecond-incompleteness-theorem":21004,"\u002Flogic\u002Fcomputability-and-representability\u002Frecursive-functions":20638,"\u002Flogic\u002Fcomputability-and-representability\u002Frepresenting-exponentiation":21005,"\u002Flogic\u002Fsecond-order-logic\u002Fsecond-order-languages":20823,"\u002Flogic\u002Fsecond-order-logic\u002Fskolem-functions-and-many-sorted-logic":21006,"\u002Flogic\u002Fsecond-order-logic\u002Fgeneral-structures":21007,"\u002Flogic":21008,"\u002Freinforcement-learning\u002Ffoundations\u002Fwhat-is-reinforcement-learning":21009,"\u002Freinforcement-learning\u002Ffoundations\u002Fa-brief-history-of-rl":21010,"\u002Freinforcement-learning\u002Ffoundations\u002Fmulti-armed-bandits":20624,"\u002Freinforcement-learning\u002Ffoundations\u002Fbandit-exploration-algorithms":21011,"\u002Freinforcement-learning\u002Ffoundations\u002Fmarkov-decision-processes":21012,"\u002Freinforcement-learning\u002Ffoundations\u002Fvalue-functions-and-optimality":21013,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fdynamic-programming":21014,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fdp-async-and-gpi":21004,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fmonte-carlo-methods":21015,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fmonte-carlo-off-policy":21016,"\u002Freinforcement-learning\u002Ftabular-methods\u002Ftemporal-difference-learning":21017,"\u002Freinforcement-learning\u002Ftabular-methods\u002Ftd-control-sarsa-and-q-learning":20915,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fn-step-bootstrapping":21018,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fn-step-off-policy-methods":21019,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fplanning-and-learning":21020,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fplanning-focusing-and-decision-time":21021,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fdecision-time-planning":21022,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fmonte-carlo-tree-search":20641,"\u002Freinforcement-learning\u002Fapproximation\u002Fon-policy-prediction":21023,"\u002Freinforcement-learning\u002Fapproximation\u002Ffeature-construction-and-nonlinear":21024,"\u002Freinforcement-learning\u002Fapproximation\u002Fon-policy-control":21025,"\u002Freinforcement-learning\u002Fapproximation\u002Faverage-reward-control":21026,"\u002Freinforcement-learning\u002Fapproximation\u002Foff-policy-and-the-deadly-triad":20832,"\u002Freinforcement-learning\u002Fapproximation\u002Fbellman-error-and-gradient-td":21027,"\u002Freinforcement-learning\u002Fapproximation\u002Feligibility-traces":21003,"\u002Freinforcement-learning\u002Fapproximation\u002Ftrue-online-and-sarsa-lambda":21028,"\u002Freinforcement-learning\u002Fapproximation\u002Fpolicy-gradient-methods":21029,"\u002Freinforcement-learning\u002Fapproximation\u002Factor-critic-and-continuous-actions":21030,"\u002Freinforcement-learning\u002Fapproximation\u002Fleast-squares-and-memory-based-methods":20652,"\u002Freinforcement-learning\u002Fapproximation\u002Fmemory-and-kernel-methods":21031,"\u002Freinforcement-learning\u002Fapproximation\u002Foff-policy-eligibility-traces":20782,"\u002Freinforcement-learning\u002Fapproximation\u002Fstable-off-policy-traces":21032,"\u002Freinforcement-learning\u002Fdeep-rl\u002Fdeep-q-networks":21033,"\u002Freinforcement-learning\u002Fdeep-rl\u002Fdqn-improvements":20614,"\u002Freinforcement-learning\u002Fdeep-rl\u002Factor-critic-and-ppo":21034,"\u002Freinforcement-learning\u002Fdeep-rl\u002Fppo-and-continuous-control":21035,"\u002Freinforcement-learning\u002Fdeep-rl\u002Fcase-studies":21036,"\u002Freinforcement-learning\u002Fdeep-rl\u002Frl-beyond-games":21037,"\u002Freinforcement-learning\u002Fdeep-rl\u002Ffrontiers":21038,"\u002Freinforcement-learning\u002Fdeep-rl\u002Freward-design-and-open-problems":20891,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fdistributional-and-rainbow":21039,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fdistributional-and-rainbow-part-2":21040,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fcontinuous-control":21041,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fcontinuous-control-part-2":20926,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmodel-based-rl":21042,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmodel-based-rl-part-2":21043,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fexploration":21044,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fexploration-part-2":20641,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Foffline-rl":20708,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Foffline-rl-part-2":21045,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fimitation-and-inverse-rl":21046,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fimitation-and-inverse-rl-part-2":21047,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmulti-agent-rl":21048,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmulti-agent-rl-part-2":21049,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fhierarchical-rl":21050,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fhierarchical-rl-part-2":21051,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Frlhf-and-language-models":21052,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fpartial-observability-pomdps":21053,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fpartial-observability-pomdps-part-2":21054,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fsafe-and-constrained-rl":21055,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fsafe-and-constrained-rl-part-2":21056,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmeta-rl-and-generalization":21057,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fpsychology-of-reinforcement":21058,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Finstrumental-conditioning-and-control":21059,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fdopamine-and-td-error":21060,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fdopamine-in-the-brain":21061,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fanimal-learning-and-cognition":21062,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fcognitive-maps-and-planning":21063,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fneuroscience-of-reinforcement":21064,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fseveral-learning-systems":21065,"\u002Freinforcement-learning":20594,"\u002Fartificial-intelligence\u002Ffoundations\u002Fwhat-is-ai":21066,"\u002Fartificial-intelligence\u002Ffoundations\u002Ffoundations-of-ai":21067,"\u002Fartificial-intelligence\u002Ffoundations\u002Fintelligent-agents":21068,"\u002Fartificial-intelligence\u002Ffoundations\u002Fagent-architectures":21069,"\u002Fartificial-intelligence\u002Fsearch\u002Funinformed-search":21070,"\u002Fartificial-intelligence\u002Fsearch\u002Fsearch-strategies-compared":21071,"\u002Fartificial-intelligence\u002Fsearch\u002Finformed-search":21072,"\u002Fartificial-intelligence\u002Fsearch\u002Fheuristic-functions":21073,"\u002Fartificial-intelligence\u002Fsearch\u002Flocal-search":21074,"\u002Fartificial-intelligence\u002Fsearch\u002Fpopulation-and-continuous-search":21075,"\u002Fartificial-intelligence\u002Fsearch\u002Fadversarial-search":21076,"\u002Fartificial-intelligence\u002Fsearch\u002Fgames-of-chance-and-imperfect-information":21077,"\u002Fartificial-intelligence\u002Fsearch\u002Fconstraint-satisfaction":21078,"\u002Fartificial-intelligence\u002Fsearch\u002Fcsp-search-and-structure":20920,"\u002Fartificial-intelligence\u002Fsearch\u002Fsearch-under-uncertainty":20776,"\u002Fartificial-intelligence\u002Fsearch\u002Fbelief-state-and-online-search":21079,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fpropositional-logic":21080,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fpropositional-inference":21081,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Ffirst-order-logic":21082,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Ffirst-order-logic-in-use":21083,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Finference-and-resolution":21084,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Ffirst-order-resolution":20902,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fclassical-planning":21085,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fplanning-graphs-and-graphplan":21086,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fplanning-in-the-real-world":21087,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fplanning-under-uncertainty":21088,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fknowledge-representation":21089,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Freasoning-systems-and-defaults":21090,"\u002Fartificial-intelligence\u002Funcertainty\u002Fprobability-and-bayes":21091,"\u002Fartificial-intelligence\u002Funcertainty\u002Fbayes-rule-and-naive-bayes":21092,"\u002Fartificial-intelligence\u002Funcertainty\u002Fbayesian-networks":21093,"\u002Fartificial-intelligence\u002Funcertainty\u002Finference-in-bayesian-networks":21094,"\u002Fartificial-intelligence\u002Funcertainty\u002Freasoning-over-time":21095,"\u002Fartificial-intelligence\u002Funcertainty\u002Ftracking-and-data-association":21096,"\u002Fartificial-intelligence\u002Funcertainty\u002Fmaking-decisions":20835,"\u002Fartificial-intelligence\u002Funcertainty\u002Fmarkov-decision-processes":21086,"\u002Fartificial-intelligence\u002Funcertainty\u002Fdecision-networks-and-game-theory":21097,"\u002Fartificial-intelligence\u002Funcertainty\u002Fgame-theory-and-mechanism-design":20375,"\u002Fartificial-intelligence\u002Flearning\u002Flearning-from-examples":21098,"\u002Fartificial-intelligence\u002Flearning\u002Ftheory-and-model-families":21099,"\u002Fartificial-intelligence\u002Flearning\u002Fprobabilistic-learning":21100,"\u002Fartificial-intelligence\u002Flearning\u002Fexpectation-maximization":21101,"\u002Fartificial-intelligence\u002Flearning\u002Freinforcement-learning":21102,"\u002Fartificial-intelligence\u002Flearning\u002Fgeneralization-and-policy-search":20883,"\u002Fartificial-intelligence\u002Flearning\u002Fknowledge-in-learning":21103,"\u002Fartificial-intelligence\u002Flearning\u002Fknowledge-based-learning-methods":21104,"\u002Fartificial-intelligence\u002Ffrontiers\u002Fvision-and-perception":21105,"\u002Fartificial-intelligence\u002Ffrontiers\u002Freconstructing-the-3d-world":21106,"\u002Fartificial-intelligence\u002Ffrontiers\u002Frobotics":21107,"\u002Fartificial-intelligence\u002Ffrontiers\u002Frobot-planning-and-control":21108,"\u002Fartificial-intelligence\u002Ffrontiers\u002Fnatural-language-in-ai":21109,"\u002Fartificial-intelligence\u002Ffrontiers\u002Fnlp-grammar-translation-and-speech":21110,"\u002Fartificial-intelligence\u002Ffrontiers\u002Fphilosophy-and-future":21111,"\u002Fartificial-intelligence\u002Ffrontiers\u002Fai-ethics-and-future":21112,"\u002Fartificial-intelligence":20594,"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fnuclear-constituents-nuclide-chart":20851,"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fnuclear-size-charge-distributions":21113,"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fnuclear-masses-binding-energy":20411,"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fsemi-empirical-mass-formula":20409,"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fnuclear-moments-multipoles":20944,"\u002Fnuclear-physics\u002Fnuclear-force-deuteron\u002Fnuclear-force-shell-overview":21114,"\u002Fnuclear-physics\u002Fnuclear-force-deuteron\u002Fthe-deuteron":20437,"\u002Fnuclear-physics\u002Fnuclear-force-deuteron\u002Fnucleon-nucleon-scattering":20747,"\u002Fnuclear-physics\u002Fnuclear-force-deuteron\u002Fmeson-theory-isospin":21115,"\u002Fnuclear-physics\u002Fnuclear-models\u002Ffermi-gas-model":21116,"\u002Fnuclear-physics\u002Fnuclear-models\u002Fliquid-drop-collective-coordinates":21117,"\u002Fnuclear-physics\u002Fnuclear-models\u002Fshell-model-single-particle":20807,"\u002Fnuclear-physics\u002Fnuclear-models\u002Fcollective-model-rotations-vibrations":21118,"\u002Fnuclear-physics\u002Fradioactive-decay\u002Fdecay-law-modes":21119,"\u002Fnuclear-physics\u002Fradioactive-decay\u002Fdecay-kinetics-equilibrium":21120,"\u002Fnuclear-physics\u002Falpha-decay\u002Falpha-decay-gamow-theory":21005,"\u002Fnuclear-physics\u002Falpha-decay\u002Falpha-fine-structure-hindrance":21121,"\u002Fnuclear-physics\u002Fbeta-decay\u002Fbeta-decay-energetics-neutrino":21122,"\u002Fnuclear-physics\u002Fbeta-decay\u002Ffermi-theory-beta-decay":20421,"\u002Fnuclear-physics\u002Fbeta-decay\u002Fweak-interaction-parity-violation":20650,"\u002Fnuclear-physics\u002Fbeta-decay\u002Fdouble-beta-decay-neutrino-mass":21123,"\u002Fnuclear-physics\u002Fgamma-decay\u002Fgamma-multipole-radiation":20751,"\u002Fnuclear-physics\u002Fgamma-decay\u002Finternal-conversion-isomers":21124,"\u002Fnuclear-physics\u002Fgamma-decay\u002Fangular-correlations-mossbauer":21125,"\u002Fnuclear-physics\u002Fnuclear-reactions\u002Freaction-kinematics-cross-sections":20646,"\u002Fnuclear-physics\u002Fnuclear-reactions\u002Fcompound-nucleus-resonances":20734,"\u002Fnuclear-physics\u002Fnuclear-reactions\u002Fdirect-reactions-optical-model":21126,"\u002Fnuclear-physics\u002Ffission\u002Ffission-barrier-dynamics":21127,"\u002Fnuclear-physics\u002Ffission\u002Fchain-reactions-reactor-physics":21128,"\u002Fnuclear-physics\u002Ffusion-nucleosynthesis\u002Ffusion-reactions-confinement":20416,"\u002Fnuclear-physics\u002Ffusion-nucleosynthesis\u002Fstellar-nucleosynthesis":20791,"\u002Fnuclear-physics\u002Ffusion-nucleosynthesis\u002Fbig-bang-nucleosynthesis":20653,"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fcharged-particle-stopping-power":21129,"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fphoton-neutron-interactions":20701,"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fradiation-detectors":20765,"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fdosimetry-radiation-biology":21130,"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fnuclear-applications-dating-medicine":21131,"\u002Fnuclear-physics":21132,"\u002Fnatural-language-processing\u002Ffoundations\u002Fwhat-is-nlp":21133,"\u002Fnatural-language-processing\u002Ffoundations\u002Fregex-and-text-normalization":21134,"\u002Fnatural-language-processing\u002Ffoundations\u002Fminimum-edit-distance":20772,"\u002Fnatural-language-processing\u002Ffoundations\u002Fn-gram-language-models":21135,"\u002Fnatural-language-processing\u002Ffoundations\u002Fsmoothing-and-backoff":21136,"\u002Fnatural-language-processing\u002Fclassification\u002Fnaive-bayes-and-sentiment":21137,"\u002Fnatural-language-processing\u002Fclassification\u002Fevaluating-classifiers":20620,"\u002Fnatural-language-processing\u002Fclassification\u002Flogistic-regression":21138,"\u002Fnatural-language-processing\u002Fclassification\u002Fsentiment-and-affect-lexicons":21139,"\u002Fnatural-language-processing\u002Fsemantics\u002Fvector-semantics-and-embeddings":20916,"\u002Fnatural-language-processing\u002Fsemantics\u002Fstatic-word-embeddings":21140,"\u002Fnatural-language-processing\u002Fsemantics\u002Fneural-language-models":21085,"\u002Fnatural-language-processing\u002Fsequences\u002Fsequence-labeling":21141,"\u002Fnatural-language-processing\u002Fsequences\u002Fcrfs-and-neural-taggers":21142,"\u002Fnatural-language-processing\u002Fsequences\u002Frnns-and-lstms":21143,"\u002Fnatural-language-processing\u002Ftransformers\u002Ftransformers-and-attention":21144,"\u002Fnatural-language-processing\u002Ftransformers\u002Fthe-transformer-architecture":21145,"\u002Fnatural-language-processing\u002Ftransformers\u002Flarge-language-models":21146,"\u002Fnatural-language-processing\u002Ftransformers\u002Fllm-pretraining-and-scaling":21147,"\u002Fnatural-language-processing\u002Ftransformers\u002Ffine-tuning-and-prompting":20598,"\u002Fnatural-language-processing\u002Ftransformers\u002Fprompting-and-alignment":21148,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fconstituency-parsing":21149,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fcky-scoring-and-evaluation":21091,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fdependency-parsing":21150,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fgraph-based-and-neural-dependency-parsing":21151,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fword-senses-and-wsd":21152,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fwsd-in-practice-and-induction":21153,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fsemantic-roles-and-information-extraction":21154,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Frelations-events-and-templates":21155,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fcoreference-and-discourse":21156,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fcoherence-and-discourse-structure":21157,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Flogical-semantics":21015,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fcompositional-semantics-and-description-logics":21158,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fsemantic-parsing":21159,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fneural-semantic-parsing":21160,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Finformation-extraction":21161,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Ftimes-events-and-templates":21162,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fdiscourse-coherence":21163,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fentity-based-and-global-coherence":21164,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fconstituency-grammars":21165,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Ftreebanks-and-lexicalized-grammars":21166,"\u002Fnatural-language-processing\u002Fapplications\u002Fmachine-translation":21167,"\u002Fnatural-language-processing\u002Fapplications\u002Fmachine-translation-decoding-and-evaluation":21168,"\u002Fnatural-language-processing\u002Fapplications\u002Fquestion-answering":21169,"\u002Fnatural-language-processing\u002Fapplications\u002Fquestion-answering-knowledge-and-llms":20869,"\u002Fnatural-language-processing\u002Fapplications\u002Fdialogue-and-chatbots":21036,"\u002Fnatural-language-processing\u002Fapplications\u002Fdialogue-systems-and-assistants":20626,"\u002Fnatural-language-processing\u002Fapplications\u002Ftext-summarization":21170,"\u002Fnatural-language-processing\u002Fapplications\u002Fabstractive-summarization-and-evaluation":21171,"\u002Fnatural-language-processing\u002Fspeech\u002Fphonetics":21172,"\u002Fnatural-language-processing\u002Fspeech\u002Facoustic-phonetics":21173,"\u002Fnatural-language-processing\u002Fspeech\u002Fautomatic-speech-recognition":20883,"\u002Fnatural-language-processing\u002Fspeech\u002Fasr-evaluation-and-applications":21174,"\u002Fnatural-language-processing":20594,"\u002Fparticle-physics\u002Ffoundations\u002Fhistorical-overview-particle-zoo":21175,"\u002Fparticle-physics\u002Ffoundations\u002Fparticle-physics-basic-concepts":20792,"\u002Fparticle-physics\u002Ffoundations\u002Ffundamental-interactions-force-carriers":21176,"\u002Fparticle-physics\u002Funits-kinematics\u002Fnatural-units-and-scales":20936,"\u002Fparticle-physics\u002Funits-kinematics\u002Ffour-vectors-invariant-mass":21177,"\u002Fparticle-physics\u002Funits-kinematics\u002Fdecay-scattering-kinematics-mandelstam":21178,"\u002Fparticle-physics\u002Funits-kinematics\u002Fcross-sections-golden-rule":21126,"\u002Fparticle-physics\u002Fsymmetries\u002Fconservation-laws-symmetries":21179,"\u002Fparticle-physics\u002Fsymmetries\u002Fdiscrete-symmetries-cpt":21180,"\u002Fparticle-physics\u002Fsymmetries\u002Fparity-violation-weak":20742,"\u002Fparticle-physics\u002Fsymmetries\u002Fsu2-su3-flavor-symmetry":20444,"\u002Fparticle-physics\u002Fquark-model\u002Feightfold-way-su3":21181,"\u002Fparticle-physics\u002Fquark-model\u002Fmeson-spectroscopy":20696,"\u002Fparticle-physics\u002Fquark-model\u002Fbaryon-spectroscopy":21182,"\u002Fparticle-physics\u002Fquark-model\u002Fcolor-confinement-exotics":20729,"\u002Fparticle-physics\u002Frelativistic-wave-equations\u002Fklein-gordon-equation":21183,"\u002Fparticle-physics\u002Frelativistic-wave-equations\u002Fdirac-equation-spinors":20982,"\u002Fparticle-physics\u002Frelativistic-wave-equations\u002Fantiparticles-hole-theory":21184,"\u002Fparticle-physics\u002Fqed\u002Ffeynman-rules-qed":21185,"\u002Fparticle-physics\u002Fqed\u002Fqed-tree-processes":20951,"\u002Fparticle-physics\u002Fqed\u002Frenormalization-running-coupling":21186,"\u002Fparticle-physics\u002Fqed\u002Felectron-g-2":20415,"\u002Fparticle-physics\u002Fweak-interaction\u002Fva-structure-weak":21187,"\u002Fparticle-physics\u002Fweak-interaction\u002Fw-z-bosons-decays":21188,"\u002Fparticle-physics\u002Fweak-interaction\u002Fckm-matrix":21189,"\u002Fparticle-physics\u002Fweak-interaction\u002Fcp-violation-kaons-b-mesons":20733,"\u002Fparticle-physics\u002Fqcd\u002Fcolor-su3-gluons":20948,"\u002Fparticle-physics\u002Fqcd\u002Fasymptotic-freedom-confinement":21190,"\u002Fparticle-physics\u002Fqcd\u002Fdeep-inelastic-scattering-partons":21191,"\u002Fparticle-physics\u002Fqcd\u002Fjets-hadronization":20808,"\u002Fparticle-physics\u002Felectroweak-higgs\u002Felectroweak-su2-u1":21192,"\u002Fparticle-physics\u002Felectroweak-higgs\u002Fspontaneous-symmetry-breaking":20853,"\u002Fparticle-physics\u002Felectroweak-higgs\u002Fhiggs-mechanism":21193,"\u002Fparticle-physics\u002Felectroweak-higgs\u002Fhiggs-boson-discovery":21194,"\u002Fparticle-physics\u002Felectroweak-higgs\u002Fstandard-model":20742,"\u002Fparticle-physics\u002Fneutrinos\u002Fneutrino-oscillations":20775,"\u002Fparticle-physics\u002Fneutrinos\u002Fneutrino-mass-pmns":21195,"\u002Fparticle-physics\u002Fneutrinos\u002Fdirac-majorana-experiments":21196,"\u002Fparticle-physics\u002Fexperiment\u002Faccelerators-luminosity":21197,"\u002Fparticle-physics\u002Fexperiment\u002Fdetectors-subsystems":20831,"\u002Fparticle-physics\u002Fexperiment\u002Fhow-discoveries-are-made":21198,"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fbeyond-standard-model":20572,"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fgrand-unified-theories":21199,"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fsupersymmetry":20675,"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fhierarchy-problem-naturalness":21200,"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fdark-matter-candidates":21201,"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fmatter-antimatter-open-questions":20998,"\u002Fparticle-physics":21202,"\u002Fastrophysics-cosmology\u002Forientation\u002Fthe-sun-and-stars":20793,"\u002Fastrophysics-cosmology\u002Forientation\u002Fstellar-death-final-states":20946,"\u002Fastrophysics-cosmology\u002Forientation\u002Fgalaxies-and-cosmology":21203,"\u002Fastrophysics-cosmology\u002Fobservational-foundations\u002Fmagnitudes-fluxes-and-the-distance-modulus":20732,"\u002Fastrophysics-cosmology\u002Fobservational-foundations\u002Fstellar-spectra-and-spectral-classification":21204,"\u002Fastrophysics-cosmology\u002Fobservational-foundations\u002Ftelescopes-and-detectors-across-the-spectrum":20792,"\u002Fastrophysics-cosmology\u002Fobservational-foundations\u002Fthe-cosmic-distance-ladder":20704,"\u002Fastrophysics-cosmology\u002Fradiation-and-matter\u002Fblackbody-radiation-and-specific-intensity":21205,"\u002Fastrophysics-cosmology\u002Fradiation-and-matter\u002Fradiative-transfer-and-the-transfer-equation":21206,"\u002Fastrophysics-cosmology\u002Fradiation-and-matter\u002Fspectral-line-formation-and-broadening":21207,"\u002Fastrophysics-cosmology\u002Fradiation-and-matter\u002Fopacity-and-the-rosseland-mean":21208,"\u002Fastrophysics-cosmology\u002Fstellar-structure\u002Fhydrostatic-equilibrium-and-the-virial-theorem":21209,"\u002Fastrophysics-cosmology\u002Fstellar-structure\u002Fthe-equations-of-stellar-structure":20997,"\u002Fastrophysics-cosmology\u002Fstellar-structure\u002Fthe-equation-of-state-and-polytropes":20928,"\u002Fastrophysics-cosmology\u002Fstellar-structure\u002Fthe-standard-solar-model":21210,"\u002Fastrophysics-cosmology\u002Fnuclear-astrophysics\u002Fthermonuclear-reaction-rates-and-the-gamow-peak":21211,"\u002Fastrophysics-cosmology\u002Fnuclear-astrophysics\u002Fhydrogen-burning-pp-chains-and-cno":21212,"\u002Fastrophysics-cosmology\u002Fnuclear-astrophysics\u002Fhelium-burning-and-the-triple-alpha-process":21213,"\u002Fastrophysics-cosmology\u002Fnuclear-astrophysics\u002Fadvanced-burning-and-neutron-capture-nucleosynthesis":21125,"\u002Fastrophysics-cosmology\u002Fism-and-star-formation\u002Fphases-of-the-interstellar-medium":21214,"\u002Fastrophysics-cosmology\u002Fism-and-star-formation\u002Fmolecular-clouds-and-gravitational-collapse":20688,"\u002Fastrophysics-cosmology\u002Fism-and-star-formation\u002Fprotostars-and-the-pre-main-sequence":20751,"\u002Fastrophysics-cosmology\u002Fstellar-evolution\u002Fthe-main-sequence-and-its-structure":20733,"\u002Fastrophysics-cosmology\u002Fstellar-evolution\u002Fpost-main-sequence-low-mass-evolution":21215,"\u002Fastrophysics-cosmology\u002Fstellar-evolution\u002Fthe-evolution-of-massive-stars":20817,"\u002Fastrophysics-cosmology\u002Fstellar-evolution\u002Fstellar-pulsation-and-the-instability-strip":20433,"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fwhite-dwarfs-and-the-chandrasekhar-limit":21216,"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fcore-collapse-supernovae":21217,"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fthermonuclear-supernovae-type-ia":20947,"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fneutron-stars-and-pulsars":21218,"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fblack-holes-schwarzschild-and-kerr":21219,"\u002Fastrophysics-cosmology\u002Fbinaries-and-gravitational-waves\u002Fbinary-systems-and-mass-transfer":21220,"\u002Fastrophysics-cosmology\u002Fbinaries-and-gravitational-waves\u002Faccreting-compact-objects":21221,"\u002Fastrophysics-cosmology\u002Fbinaries-and-gravitational-waves\u002Fgravitational-waves-from-inspiraling-binaries":20414,"\u002Fastrophysics-cosmology\u002Fbinaries-and-gravitational-waves\u002Fmultimessenger-astronomy-and-gamma-ray-bursts":21222,"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Fthe-milky-way":21223,"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Fgalaxy-morphology-and-classification":20675,"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Fgalaxy-rotation-curves-and-dark-matter":21224,"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Factive-galactic-nuclei-and-supermassive-black-holes":21225,"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Fgalaxy-clusters-and-large-scale-structure":20940,"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fthe-expanding-universe-and-hubbles-law":20596,"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fthe-frw-metric-and-cosmological-redshift":21226,"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fthe-friedmann-equations-and-cosmic-dynamics":21227,"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fcosmological-models-and-distances":20853,"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fdark-energy-and-the-accelerating-universe":20839,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fthe-thermal-history-of-the-universe":20851,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fbig-bang-nucleosynthesis":20947,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Frecombination-and-the-cosmic-microwave-background":21228,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fcmb-anisotropies-and-cosmological-parameters":20413,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fcosmic-inflation":20605,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fstructure-formation-and-the-growth-of-perturbations":21229,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fdark-matter-dark-energy-and-open-questions":20692,"\u002Fastrophysics-cosmology":20797,"\u002Fcolophon":21230,"\u002F":20594},4250,4808,3626,2682,4109,4786,3878,3875,3751,3415,4067,3153,3000,4042,5461,5808,3961,3749,4327,5067,4246,4655,4154,5436,2640,4003,3601,2158,4331,4189,2273,3252,4633,4964,4172,3131,5524,3160,4031,2309,4207,3226,2648,4842,5340,3307,5701,4977,4039,2615,3472,4460,3848,4075,4400,3382,3010,3602,3737,3740,3707,3922,5191,4043,3804,4542,4214,5062,2850,4361,3443,3627,4044,3766,4140,3860,4006,5199,4334,5234,3651,5509,5680,153,1375,1073,1093,1125,1146,1014,1132,876,1541,1189,1173,984,1402,1301,950,1268,1063,1107,1408,1161,925,1012,866,964,1090,1142,1085,1020,1207,973,980,728,764,1225,1329,796,929,801,878,774,1044,1488,1175,1130,890,814,870,154,4073,5140,4961,5127,4870,5382,5195,4955,5369,4501,5576,3824,4132,4289,4307,4570,3403,5084,5105,5201,5116,5341,5175,5368,5188,5211,5499,5155,4981,5125,5415,5255,5304,5130,5167,5552,5164,5094,5239,5036,5190,5004,5099,5035,5159,5088,5026,4937,5023,5264,5244,133,5114,5078,5043,5312,5170,5342,5139,5151,5049,5212,5013,5068,5079,5102,5121,5081,5029,5379,5854,5110,2139,3798,5055,5364,4984,4935,4895,4972,5289,5112,5156,4987,5031,5025,5149,5302,5042,5002,4979,4922,4960,5279,126,1877,1180,1129,907,958,1112,1300,1053,1250,1181,1241,1234,966,1050,734,1190,484,1082,926,733,761,571,607,798,804,952,977,731,784,645,771,1017,742,1004,1000,1562,1254,1288,1101,1011,1486,1061,856,992,1169,988,137,0,2037,1782,2384,2254,2123,2332,1643,1714,2089,1751,1367,1660,2511,1998,1892,1854,1791,2438,2487,1917,2375,2525,2266,1845,2275,1810,1631,2310,2166,2233,2113,2505,2347,2672,2112,2473,2592,2380,3013,2513,3256,3218,2194,2173,2205,2326,2081,3342,3152,1799,1670,1027,960,1095,1291,986,897,1209,1055,1817,1801,1593,1465,1196,1464,1201,1230,1435,1684,1461,1926,1500,1409,1284,1774,1869,162,1487,1122,1188,1001,1351,982,1005,979,1325,1046,943,1279,824,1008,989,1798,1277,1025,987,1043,1211,1074,981,939,1002,739,1139,1108,1013,1070,978,1458,1317,157,1357,1077,2355,1116,1037,1178,1637,1314,1109,1056,1702,1474,1071,1158,832,993,1404,1024,1068,1339,1106,1264,1248,913,1848,1328,1633,1224,1143,135,1378,959,1028,998,911,1527,1203,1266,1483,1165,990,938,965,1257,1418,1099,942,1352,956,1035,1398,1003,1094,1292,138,1721,1827,1449,1354,1148,1184,1285,1281,1213,1290,1271,1252,1274,1778,1591,1503,1437,1571,1584,1957,1117,1781,1648,1342,1667,1510,1965,1607,1365,1849,1259,1303,1356,1238,2208,1564,173,1671,1286,1227,1638,1529,668,1078,918,709,865,880,940,1534,1015,874,922,841,794,1194,822,1105,1658,1359,1296,1438,1921,1844,1570,1429,1324,1400,140,1787,1558,1654,1492,1747,2224,2002,2009,1323,1349,1785,1573,1722,1829,1353,1548,1552,1583,1624,1585,1245,1364,1514,1343,1397,1355,2211,1481,1770,160,2388,2293,2256,2552,2569,2478,2039,2578,2814,2519,2461,2587,2492,2714,3278,2654,3050,2447,2849,2238,2369,2061,2214,2602,2563,2186,2985,2749,3364,2038,2282,2409,2126,2573,2206,2176,2268,2182,2402,2705,2633,2414,2213,2801,3313,3410,3195,1952,2017,1509,2537,2645,2027,2415,2838,2356,1906,3184,2950,2807,2954,1683,1316,1034,1138,1763,1822,1705,1246,1701,1097,1104,1187,1032,1083,1228,916,1489,1033,1652,997,692,837,1023,888,864,1089,1231,1214,1675,1156,1075,1520,1309,139,1205,1051,735,1123,1072,915,567,768,825,1253,983,1007,762,1058,861,862,971,1208,1149,1145,1029,1084,927,810,838,857,807,936,949,2321,1622,1069,1113,1057,854,1958,1528,1618,2049,1432,1679,1796,1685,1346,1275,1476,1505,1610,2018,1599,1215,1838,1909,132,3902,2215,2240,3266,3208,3073,2454,2969,2451,1875,2728,1884,2371,2516,2842,1690,1904,2346,3146,1386,2607,1966,2668,1665,2885,1606,2577,3074,2869,2403,2433,2082,1939,1587,2460,2747,2032,2642,1619,3123,1993,2090,2339,3829,1737,2622,2340,2322,3828,4409,2305,3411,2510,4527,3030,3569,3043,2457,1946,2277,2044,2909,1693,1945,2093,2399,2115,2898,2742,2242,3895,3378,3376,2769,2223,3062,3262,2651,2949,2768,3128,2423,1977,2087,2866,3388,2830,2210,2489,2884,3945,2099,2713,3402,1692,2931,4195,3989,3206,4391,3004,3704,3494,2902,999,881,901,919,748,869,1018,1045,1049,1333,954,1092,1019,976,1771,1480,1396,953,1026,161,3533,2495,1818,3007,2595,3427,3537,2216,1895,2304,3396,1739,2073,1962,2203,1767,2666,2264,2276,2852,1807,3735,1560,4144,1669,1676,1972,2418,3291,1525,2040,2766,2337,2220,2800,3001,2078,1759,2836,1896,2026,1758,1543,1047,896,946,1060,1384,1482,815,1414,1322,1440,1240,1468,1098,1133,847,1009,1381,1052,1191,1258,1370,1712,1441,1199,957,1079,150,1262,1417,1368,1219,1136,1064,1463,1636,1059,931,1115,1736,1174,1376,1363,1411,1247,1746,1313,1299,1617,1102,1076,1495,1265,1193,1263,80,{"\u002Falgorithms\u002Ffoundations\u002Fwhat-is-an-algorithm":21232,"\u002Falgorithms\u002Ffoundations\u002Fproof-techniques":21237,"\u002Falgorithms\u002Ffoundations\u002Fasymptotic-analysis":21241,"\u002Falgorithms\u002Ffoundations\u002Fgrowth-rates-and-loop-analysis":21245,"\u002Falgorithms\u002Ffoundations\u002Frecurrences":21249,"\u002Falgorithms\u002Ffoundations\u002Famortized-analysis":21253,"\u002Falgorithms\u002Fdivide-and-conquer\u002Fmergesort":21257,"\u002Falgorithms\u002Fdivide-and-conquer\u002Fquicksort":21262,"\u002Falgorithms\u002Fdivide-and-conquer\u002Fselection":21266,"\u002Falgorithms\u002Fdivide-and-conquer\u002Ffast-multiplication":21270,"\u002Falgorithms\u002Fsorting\u002Fheaps-and-heapsort":21274,"\u002Falgorithms\u002Fsorting\u002Fsorting-lower-bounds":21279,"\u002Falgorithms\u002Fsorting\u002Flinear-time-sorting":21283,"\u002Falgorithms\u002Fsorting\u002Fexternal-sorting":21287,"\u002Falgorithms\u002Fdata-structures\u002Felementary-structures":21291,"\u002Falgorithms\u002Fdata-structures\u002Fhash-tables":21296,"\u002Falgorithms\u002Fdata-structures\u002Fbinary-search-trees":21300,"\u002Falgorithms\u002Fdata-structures\u002Favl-trees":21304,"\u002Falgorithms\u002Fdata-structures\u002Fbalanced-trees":21308,"\u002Falgorithms\u002Fdata-structures\u002Funion-find":21312,"\u002Falgorithms\u002Fdata-structures\u002Ffenwick-and-segment-trees":21316,"\u002Falgorithms\u002Fdata-structures\u002Fspatial-data-structures":21320,"\u002Falgorithms\u002Fdata-structures\u002Fskip-lists-and-probabilistic-structures":21324,"\u002Falgorithms\u002Fdata-structures\u002Fb-trees":21328,"\u002Falgorithms\u002Fdata-structures\u002Fdata-stream-algorithms":21332,"\u002Falgorithms\u002Fdata-structures\u002Fstreaming-sketches":21336,"\u002Falgorithms\u002Fsequences\u002Ftwo-pointers-and-windows":21340,"\u002Falgorithms\u002Fsequences\u002Fprefix-sums":21345,"\u002Falgorithms\u002Fsequences\u002Fmonotonic-stacks":21349,"\u002Falgorithms\u002Fsequences\u002Fbinary-search-on-the-answer":21353,"\u002Falgorithms\u002Fsequences\u002Fstring-matching":21357,"\u002Falgorithms\u002Fsequences\u002Fkmp-and-z-function":21361,"\u002Falgorithms\u002Fsequences\u002Ftries":21365,"\u002Falgorithms\u002Fsequences\u002Fsuffix-arrays-and-aho-corasick":21369,"\u002Falgorithms\u002Fgraphs\u002Frepresentations-and-traversal":21373,"\u002Falgorithms\u002Fgraphs\u002Fdepth-first-search":21378,"\u002Falgorithms\u002Fgraphs\u002Ftopological-sort-and-scc":21382,"\u002Falgorithms\u002Fgraphs\u002Fminimum-spanning-trees":21386,"\u002Falgorithms\u002Fgraphs\u002Fkruskal-and-prim":21390,"\u002Falgorithms\u002Fgraphs\u002Fshortest-paths":21394,"\u002Falgorithms\u002Fgraphs\u002Fall-pairs-and-negative-weights":21398,"\u002Falgorithms\u002Fgraphs\u002Fnetwork-flow":21402,"\u002Falgorithms\u002Fgraphs\u002Fmax-flow-min-cut":21406,"\u002Falgorithms\u002Fgraphs\u002Fbridges-and-articulation-points":21410,"\u002Falgorithms\u002Fgraphs\u002Flowest-common-ancestor":21414,"\u002Falgorithms\u002Fgraphs\u002Ftwo-sat":21418,"\u002Falgorithms\u002Fgraphs\u002Feulerian-tours":21422,"\u002Falgorithms\u002Fgraphs\u002Fbipartite-matching":21426,"\u002Falgorithms\u002Fgreedy\u002Fthe-greedy-method":21430,"\u002Falgorithms\u002Fgreedy\u002Fscheduling-and-intervals":21435,"\u002Falgorithms\u002Fgreedy\u002Fhuffman-codes":21439,"\u002Falgorithms\u002Fgreedy\u002Fmatroids":21443,"\u002Falgorithms\u002Fgreedy\u002Fstable-matching":21447,"\u002Falgorithms\u002Fdynamic-programming\u002Fprinciples":21451,"\u002Falgorithms\u002Fdynamic-programming\u002Fsequence-dp":21456,"\u002Falgorithms\u002Fdynamic-programming\u002Flongest-increasing-subsequence":21460,"\u002Falgorithms\u002Fdynamic-programming\u002Fknapsack":21464,"\u002Falgorithms\u002Fdynamic-programming\u002Fcoin-change-and-unbounded":21468,"\u002Falgorithms\u002Fdynamic-programming\u002Finterval-dp":21472,"\u002Falgorithms\u002Fdynamic-programming\u002Ftree-dp":21476,"\u002Falgorithms\u002Fdynamic-programming\u002Fbitmask-dp":21480,"\u002Falgorithms\u002Fdynamic-programming\u002Fdp-optimizations":21484,"\u002Falgorithms\u002Fdynamic-programming\u002Fdp-on-graphs":21488,"\u002Falgorithms\u002Fdynamic-programming\u002Fdigit-and-probability-dp":21492,"\u002Falgorithms\u002Fbacktracking\u002Fbacktracking-fundamentals":21496,"\u002Falgorithms\u002Fbacktracking\u002Fconstraint-search":21501,"\u002Falgorithms\u002Fbacktracking\u002Fbranch-and-bound":21505,"\u002Falgorithms\u002Fbacktracking\u002Fgraph-backtracking":21509,"\u002Falgorithms\u002Fmathematical-algorithms\u002Fnumber-theory-basics":21513,"\u002Falgorithms\u002Fmathematical-algorithms\u002Fmodular-exponentiation-and-primality":21518,"\u002Falgorithms\u002Fmathematical-algorithms\u002Fsieve-and-factorization":21522,"\u002Falgorithms\u002Fmathematical-algorithms\u002Fcombinatorics":21526,"\u002Falgorithms\u002Fmathematical-algorithms\u002Fmatrix-exponentiation":21530,"\u002Falgorithms\u002Fmathematical-algorithms\u002Ffast-fourier-transform":21534,"\u002Falgorithms\u002Fmathematical-algorithms\u002Fgradient-descent":21538,"\u002Falgorithms\u002Fcomputational-geometry\u002Fgeometric-primitives":21542,"\u002Falgorithms\u002Fcomputational-geometry\u002Fconvex-hull":21547,"\u002Falgorithms\u002Fcomputational-geometry\u002Fsweep-line":21551,"\u002Falgorithms\u002Fcomputational-geometry\u002Fpolygons-and-proximity":21555,"\u002Falgorithms\u002Fintractability\u002Fp-np-reductions":21559,"\u002Falgorithms\u002Fintractability\u002Fnp-completeness":21564,"\u002Falgorithms\u002Fintractability\u002Fcoping-with-hardness":21568,"\u002Falgorithms\u002Fintractability\u002Fapproximation-algorithms":21572,"\u002Falgorithms":21576,"\u002Fcalculus\u002Flimits-and-continuity\u002Ffunctions-and-models":21579,"\u002Fcalculus\u002Flimits-and-continuity\u002Fthe-limit-of-a-function":21584,"\u002Fcalculus\u002Flimits-and-continuity\u002Flimit-laws-and-the-precise-definition":21588,"\u002Fcalculus\u002Flimits-and-continuity\u002Fcontinuity":21592,"\u002Fcalculus\u002Fderivatives\u002Fthe-derivative-and-rates-of-change":21596,"\u002Fcalculus\u002Fderivatives\u002Fdifferentiation-rules-and-the-chain-rule":21601,"\u002Fcalculus\u002Fderivatives\u002Fimplicit-differentiation-and-related-rates":21605,"\u002Fcalculus\u002Fderivatives\u002Flinear-approximations-and-differentials":21609,"\u002Fcalculus\u002Fapplications-of-derivatives\u002Fextrema-and-the-mean-value-theorem":21613,"\u002Fcalculus\u002Fapplications-of-derivatives\u002Fhow-derivatives-shape-a-graph":21618,"\u002Fcalculus\u002Fapplications-of-derivatives\u002Fcurve-sketching-and-optimization":21622,"\u002Fcalculus\u002Fapplications-of-derivatives\u002Fnewtons-method-and-antiderivatives":21626,"\u002Fcalculus\u002Fintegrals\u002Farea-and-the-definite-integral":21630,"\u002Fcalculus\u002Fintegrals\u002Fthe-fundamental-theorem-of-calculus":21635,"\u002Fcalculus\u002Fintegrals\u002Fthe-substitution-rule":21639,"\u002Fcalculus\u002Fapplications-of-integration\u002Fareas-and-volumes":21643,"\u002Fcalculus\u002Fapplications-of-integration\u002Fwork-average-value-and-arc-length":21648,"\u002Fcalculus\u002Fapplications-of-integration\u002Fphysics-economics-and-probability":21652,"\u002Fcalculus\u002Fexponential-logarithmic-and-inverse-functions\u002Finverse-functions-logarithms-and-exponentials":21656,"\u002Fcalculus\u002Fexponential-logarithmic-and-inverse-functions\u002Fgrowth-decay-inverse-trig-and-hyperbolic-functions":21661,"\u002Fcalculus\u002Fexponential-logarithmic-and-inverse-functions\u002Flhospitals-rule":21665,"\u002Fcalculus\u002Ftechniques-of-integration\u002Fintegration-by-parts":21669,"\u002Fcalculus\u002Ftechniques-of-integration\u002Ftrigonometric-integrals-and-substitution":21674,"\u002Fcalculus\u002Ftechniques-of-integration\u002Fpartial-fractions-and-integration-strategy":21678,"\u002Fcalculus\u002Ftechniques-of-integration\u002Fapproximate-and-improper-integrals":21682,"\u002Fcalculus\u002Fparametric-and-polar\u002Fparametric-curves-and-their-calculus":21686,"\u002Fcalculus\u002Fparametric-and-polar\u002Fpolar-coordinates":21691,"\u002Fcalculus\u002Fparametric-and-polar\u002Fconic-sections":21695,"\u002Fcalculus\u002Fsequences-and-series\u002Fsequences":21699,"\u002Fcalculus\u002Fsequences-and-series\u002Fseries-and-the-integral-test":21704,"\u002Fcalculus\u002Fsequences-and-series\u002Fthe-convergence-tests":21708,"\u002Fcalculus\u002Fsequences-and-series\u002Fpower-series":21712,"\u002Fcalculus\u002Fsequences-and-series\u002Ftaylor-and-maclaurin-series":21716,"\u002Fcalculus\u002Fvectors-and-space-curves\u002Fvectors-and-the-dot-product":21720,"\u002Fcalculus\u002Fvectors-and-space-curves\u002Fthe-cross-product-lines-and-planes":21725,"\u002Fcalculus\u002Fvectors-and-space-curves\u002Fcylinders-and-quadric-surfaces":21729,"\u002Fcalculus\u002Fvectors-and-space-curves\u002Fvector-functions-and-space-curves":21733,"\u002Fcalculus\u002Fvectors-and-space-curves\u002Farc-length-curvature-and-motion":21737,"\u002Fcalculus\u002Fpartial-derivatives\u002Ffunctions-of-several-variables":21741,"\u002Fcalculus\u002Fpartial-derivatives\u002Fpartial-derivatives":21746,"\u002Fcalculus\u002Fpartial-derivatives\u002Ftangent-planes-and-the-chain-rule":21749,"\u002Fcalculus\u002Fpartial-derivatives\u002Fdirectional-derivatives-and-the-gradient":21753,"\u002Fcalculus\u002Fpartial-derivatives\u002Foptimization-and-lagrange-multipliers":21757,"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fdouble-integrals":21761,"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Ftriple-integrals-and-coordinate-systems":21766,"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fvector-fields-and-line-integrals":21770,"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fgreens-theorem-curl-and-divergence":21774,"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fsurface-integrals":21778,"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fstokes-and-the-divergence-theorem":21782,"\u002Fcalculus":21786,"\u002Fmechanics\u002Ffoundations\u002Fmeasurement-and-dimensions":21789,"\u002Fmechanics\u002Ffoundations\u002Fvector-algebra":21793,"\u002Fmechanics\u002Fkinematics\u002Fone-dimensional-motion":21797,"\u002Fmechanics\u002Fkinematics\u002Fmotion-graphs":21802,"\u002Fmechanics\u002Fkinematics\u002Fprojectile-motion":21806,"\u002Fmechanics\u002Fkinematics\u002Frelative-motion":21810,"\u002Fmechanics\u002Fkinematics\u002Fcircular-motion":21814,"\u002Fmechanics\u002Fdynamics\u002Fnewtons-laws":21818,"\u002Fmechanics\u002Fdynamics\u002Ffree-body-diagrams":21823,"\u002Fmechanics\u002Fdynamics\u002Ffriction-and-curved-motion":21827,"\u002Fmechanics\u002Fdynamics\u002Fnumerical-dynamics":21831,"\u002Fmechanics\u002Fdynamics\u002Fcenter-of-mass-systems":21835,"\u002Fmechanics\u002Fenergy\u002Fwork-and-kinetic-energy":21839,"\u002Fmechanics\u002Fenergy\u002Fpotential-energy":21844,"\u002Fmechanics\u002Fenergy\u002Fmultiparticle-work":21848,"\u002Fmechanics\u002Fenergy\u002Fmass-energy-and-binding":21852,"\u002Fmechanics\u002Fenergy\u002Fphotons-and-quantization":21856,"\u002Fmechanics\u002Fmomentum\u002Fmomentum-and-collisions":21860,"\u002Fmechanics\u002Fmomentum\u002Fcenter-of-mass-collisions":21865,"\u002Fmechanics\u002Fmomentum\u002Frocket-propulsion":21869,"\u002Fmechanics\u002Frotation\u002Frotational-inertia":21873,"\u002Fmechanics\u002Frotation\u002Frotational-dynamics":21878,"\u002Fmechanics\u002Frotation\u002Frolling-motion":21882,"\u002Fmechanics\u002Frotation\u002Fangular-momentum":21886,"\u002Fmechanics\u002Frotation\u002Frolling-resistance":21890,"\u002Fmechanics\u002Frotation\u002Fgyroscopic-precession":21894,"\u002Fmechanics\u002Fgravity-and-matter\u002Fkeplerian-orbits":21898,"\u002Fmechanics\u002Fgravity-and-matter\u002Fgravitational-fields":21903,"\u002Fmechanics\u002Fgravity-and-matter\u002Fstatic-equilibrium":21907,"\u002Fmechanics\u002Fgravity-and-matter\u002Ffluid-statics":21911,"\u002Fmechanics\u002Fgravity-and-matter\u002Ffluid-flow":21915,"\u002Fmechanics\u002Fgravity-and-matter\u002Forbital-motion":21919,"\u002Fmechanics\u002Fgravity-and-matter\u002Fstress-and-elasticity":21923,"\u002Fmechanics\u002Foscillations-waves\u002Fdamped-oscillators":21927,"\u002Fmechanics\u002Foscillations-waves\u002Ftravelling-waves":21932,"\u002Fmechanics\u002Foscillations-waves\u002Fwave-superposition":21936,"\u002Fmechanics\u002Foscillations-waves\u002Fstanding-waves":21940,"\u002Fmechanics\u002Foscillations-waves\u002Fsound-waves":21944,"\u002Fmechanics\u002Foscillations-waves\u002Fdoppler-effect":21948,"\u002Fmechanics\u002Foscillations-waves\u002Fwave-packets":21952,"\u002Fmechanics\u002Foscillations-waves\u002Fbeats-and-coupling":21956,"\u002Fmechanics\u002Foscillations-waves\u002Fsimple-harmonic-motion":21960,"\u002Fmechanics\u002Foscillations-waves\u002Fpendulum-motion":21964,"\u002Fmechanics\u002Foscillations-waves\u002Fdriven-oscillators":21968,"\u002Fmechanics\u002Foscillations-waves\u002Fwave-boundaries":21972,"\u002Fmechanics\u002Fthermodynamics\u002Fkinetic-theory-of-ideal-gases":21976,"\u002Fmechanics\u002Fthermodynamics\u002Ffirst-law-of-thermodynamics":21981,"\u002Fmechanics\u002Fthermodynamics\u002Fentropy-and-the-second-law":21985,"\u002Fmechanics\u002Fthermodynamics\u002Fthermal-processes":21989,"\u002Fmechanics\u002Fthermodynamics\u002Fphase-changes":21993,"\u002Fmechanics\u002Fthermodynamics\u002Fthermal-machines":21997,"\u002Fmechanics":22001,"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Fcharge-and-conductors":22004,"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Fcoulombs-law":22009,"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Felectric-field-and-force":22013,"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Felectric-field-maps":22017,"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Felectric-dipoles":22021,"\u002Felectricity-and-magnetism\u002Fcontinuous-charge-distributions\u002Fcontinuous-charge-fields":22025,"\u002Felectricity-and-magnetism\u002Fcontinuous-charge-distributions\u002Fgauss-law-and-conductors":22030,"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Fpoint-charge-potential":22034,"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Fpotential-gradients-and-equipotentials":22039,"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Felectrostatic-energy-and-pressure":22043,"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Flaplace-boundary-problems":22047,"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Fcontinuous-charge-potentials":22051,"\u002Felectricity-and-magnetism\u002Fcapacitance\u002Fcapacitance-fundamentals":22055,"\u002Felectricity-and-magnetism\u002Fcapacitance\u002Fcapacitor-networks":22060,"\u002Felectricity-and-magnetism\u002Fcapacitance\u002Fcapacitor-energy-and-force":22064,"\u002Felectricity-and-magnetism\u002Fcapacitance\u002Fdielectric-polarization-and-breakdown":22068,"\u002Felectricity-and-magnetism\u002Fdirect-current-circuits\u002Fcurrent-and-resistance":22072,"\u002Felectricity-and-magnetism\u002Fdirect-current-circuits\u002Fkirchhoff-network-analysis":22077,"\u002Felectricity-and-magnetism\u002Fdirect-current-circuits\u002Frc-transients":22081,"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fmagnetic-trajectories":22085,"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fhall-effect":22090,"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fmagnetic-force-on-conductors":22094,"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fmagnetic-dipoles":22098,"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fmass-spectrometry":22102,"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fmoving-charge-fields":22106,"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fbiot-savart-law":22111,"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fcircular-current-loops":22115,"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Famperes-law":22119,"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fgauss-law-for-magnetism":22123,"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fmagnetic-materials":22127,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Fmagnetic-flux":22131,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Ffaradays-law":22136,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Flenzs-law":22140,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Fmotional-emf":22144,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Feddy-currents":22148,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Fself-inductance":22152,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Fmagnetic-energy":22156,"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Frl-circuits":22160,"\u002Felectricity-and-magnetism\u002Falternating-current\u002Fac-fundamentals":22164,"\u002Felectricity-and-magnetism\u002Falternating-current\u002Freactance":22169,"\u002Felectricity-and-magnetism\u002Falternating-current\u002Frlc-resonance":22173,"\u002Felectricity-and-magnetism\u002Falternating-current\u002Fac-power":22177,"\u002Felectricity-and-magnetism\u002Falternating-current\u002Ftransformers":22181,"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Fdisplacement-current":22185,"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Felectromagnetic-waves":22190,"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Felectromagnetic-momentum":22194,"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Fdipole-radiation":22198,"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Fpolarization":22202,"\u002Felectricity-and-magnetism\u002Foptics\u002Freflection-and-refraction":22206,"\u002Felectricity-and-magnetism\u002Foptics\u002Fthin-lenses":22211,"\u002Felectricity-and-magnetism\u002Foptics\u002Fspherical-mirrors":22215,"\u002Felectricity-and-magnetism":22219,"\u002Flinear-algebra\u002Flinear-systems\u002Fsystems-and-echelon-forms":22222,"\u002Flinear-algebra\u002Flinear-systems\u002Fvector-and-matrix-equations":22227,"\u002Flinear-algebra\u002Flinear-systems\u002Fsolution-sets-and-applications":22231,"\u002Flinear-algebra\u002Flinear-systems\u002Flinear-independence":22235,"\u002Flinear-algebra\u002Flinear-systems\u002Flinear-transformations":22239,"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fmatrix-operations":22243,"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fmatrix-inverse-and-invertibility":22248,"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fpartitioned-matrices-and-lu":22252,"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fsubspaces-dimension-rank":22256,"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fapplications-leontief-and-graphics":22260,"\u002Flinear-algebra\u002Fdeterminants\u002Fdeterminants-and-cofactors":22264,"\u002Flinear-algebra\u002Fdeterminants\u002Fproperties-of-determinants":22269,"\u002Flinear-algebra\u002Fdeterminants\u002Fcramer-volume-and-area":22273,"\u002Flinear-algebra\u002Fvector-spaces\u002Fvector-spaces-and-subspaces":22277,"\u002Flinear-algebra\u002Fvector-spaces\u002Fnull-and-column-spaces":22282,"\u002Flinear-algebra\u002Fvector-spaces\u002Fbases-and-independent-sets":22286,"\u002Flinear-algebra\u002Fvector-spaces\u002Fcoordinate-systems":22290,"\u002Flinear-algebra\u002Fvector-spaces\u002Fdimension-and-rank":22294,"\u002Flinear-algebra\u002Fvector-spaces\u002Fchange-of-basis":22298,"\u002Flinear-algebra\u002Fvector-spaces\u002Fdifference-equations-and-markov":22302,"\u002Flinear-algebra\u002Feigenvalues\u002Feigenvectors-and-eigenvalues":22306,"\u002Flinear-algebra\u002Feigenvalues\u002Fthe-characteristic-equation":22311,"\u002Flinear-algebra\u002Feigenvalues\u002Fdiagonalization":22315,"\u002Flinear-algebra\u002Feigenvalues\u002Feigenvectors-and-linear-transformations":22319,"\u002Flinear-algebra\u002Feigenvalues\u002Fcomplex-eigenvalues":22323,"\u002Flinear-algebra\u002Feigenvalues\u002Fdynamical-systems":22327,"\u002Flinear-algebra\u002Feigenvalues\u002Fpower-method":22331,"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Finner-product-length-orthogonality":22335,"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Forthogonal-sets-and-projections":22340,"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Fgram-schmidt-and-qr":22344,"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Fleast-squares-problems":22348,"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Fleast-squares-applications":22352,"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Finner-product-spaces":22356,"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fdiagonalizing-symmetric-matrices":22360,"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fquadratic-forms":22365,"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fconstrained-optimization":22369,"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fsingular-value-decomposition":22373,"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fsvd-applications-pca-imaging":22377,"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fnumerical-thinking-and-matrix-computation":22381,"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Flu-and-cholesky":22386,"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fconditioning-and-floating-point":22390,"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fstability-and-error-analysis":22394,"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fqr-and-numerical-least-squares":22398,"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fnumerical-eigenvalues-and-svd":22402,"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Faffine-combinations":22406,"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Faffine-independence-and-barycentric-coordinates":22411,"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Fconvex-combinations-and-convex-sets":22415,"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Fhyperplanes-and-polytopes":22419,"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Fcurves-and-surfaces":22423,"\u002Flinear-algebra":22427,"\u002Ftheory-of-computation":22430,"\u002Fcomputer-architecture\u002Ffoundations\u002Fbits-bytes-and-words":22433,"\u002Fcomputer-architecture\u002Ffoundations\u002Finteger-representation":22437,"\u002Fcomputer-architecture\u002Ffoundations\u002Finteger-arithmetic":22441,"\u002Fcomputer-architecture\u002Ffoundations\u002Ffloating-point":22445,"\u002Fcomputer-architecture\u002Ffoundations\u002Fboolean-algebra-and-bit-manipulation":22449,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fthe-machines-view":22453,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fdata-movement":22458,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Farithmetic-and-logic":22462,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fcontrol-flow":22466,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fprocedures":22470,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Farrays-structs-and-alignment":22474,"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fmemory-layout-and-buffer-overflows":22478,"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Fwhat-an-isa-is":22482,"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Finstruction-formats-and-operands":22487,"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Faddressing-modes":22491,"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Fthe-y86-64-instruction-set":22495,"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Fy86-64-programming":22499,"\u002Fcomputer-architecture\u002Fdigital-logic\u002Ftransistors-gates-and-boolean-functions":22503,"\u002Fcomputer-architecture\u002Fdigital-logic\u002Fcombinational-logic-and-hcl":22508,"\u002Fcomputer-architecture\u002Fdigital-logic\u002Fmultiplexers-decoders-and-the-alu":22512,"\u002Fcomputer-architecture\u002Fdigital-logic\u002Fmemory-elements-latches-flip-flops-and-clocking":22516,"\u002Fcomputer-architecture\u002Fdigital-logic\u002Fregister-files-and-random-access-memory":22520,"\u002Fcomputer-architecture\u002Fprocessor-design\u002Fthe-fetch-decode-execute-cycle":22524,"\u002Fcomputer-architecture\u002Fprocessor-design\u002Fthe-seq-stages":22529,"\u002Fcomputer-architecture\u002Fprocessor-design\u002Fcontrol-logic-and-sequencing":22533,"\u002Fcomputer-architecture\u002Fprocessor-design\u002Fassembling-seq":22537,"\u002Fcomputer-architecture\u002Fprocessor-design\u002Ftracing-a-program":22541,"\u002Fcomputer-architecture\u002Fpipelining\u002Fpipelining-principles":22545,"\u002Fcomputer-architecture\u002Fpipelining\u002Ffrom-seq-to-pipe":22550,"\u002Fcomputer-architecture\u002Fpipelining\u002Fdata-hazards-stalling-and-forwarding":22554,"\u002Fcomputer-architecture\u002Fpipelining\u002Fcontrol-hazards-and-branch-prediction":22558,"\u002Fcomputer-architecture\u002Fpipelining\u002Fthe-complete-pipe-processor":22562,"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Fstorage-technologies-and-the-latency-gap":22566,"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Flocality":22571,"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Fcache-memories-direct-mapped":22575,"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Fset-associative-and-write-policies":22579,"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Fcache-performance-and-cache-friendly-code":22583,"\u002Fcomputer-architecture\u002Fvirtual-memory\u002Faddress-spaces-and-translation":22587,"\u002Fcomputer-architecture\u002Fvirtual-memory\u002Fpage-tables-and-page-faults":22592,"\u002Fcomputer-architecture\u002Fvirtual-memory\u002Fthe-tlb-and-multi-level-page-tables":22596,"\u002Fcomputer-architecture\u002Fexceptions-and-io\u002Fexceptional-control-flow":22600,"\u002Fcomputer-architecture\u002Fexceptions-and-io\u002Finterrupts-and-the-kernel":22605,"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fprocesses-threads-and-parallelism":22609,"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fhardware-multithreading":22614,"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fcache-coherence":22618,"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fmemory-consistency-and-synchronization":22622,"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fmulticore-organization":22626,"\u002Fcomputer-architecture\u002Fcapstone\u002Fthe-whole-machine":22630,"\u002Fcomputer-architecture\u002Fcapstone\u002Fassembling-a-complete-cpu":22635,"\u002Fcomputer-architecture":22639,"\u002Fdifferential-equations\u002Ffoundations\u002Fmodels-and-direction-fields":22642,"\u002Fdifferential-equations\u002Ffoundations\u002Fclassification-and-terminology":22646,"\u002Fdifferential-equations\u002Ffirst-order\u002Flinear-first-order-integrating-factors":22650,"\u002Fdifferential-equations\u002Ffirst-order\u002Fseparable-and-exact":22655,"\u002Fdifferential-equations\u002Ffirst-order\u002Fmodeling-first-order":22659,"\u002Fdifferential-equations\u002Ffirst-order\u002Fautonomous-and-population-dynamics":22663,"\u002Fdifferential-equations\u002Ffirst-order\u002Fexistence-uniqueness-euler":22667,"\u002Fdifferential-equations\u002Ffirst-order\u002Ffirst-order-difference-equations":22671,"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fhomogeneous-constant-coefficients":22675,"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fcomplex-and-repeated-roots":22680,"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fnonhomogeneous-undetermined-coefficients":22684,"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fvariation-of-parameters":22688,"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fmechanical-electrical-vibrations":22692,"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fhigher-order-linear":22696,"\u002Fdifferential-equations\u002Fseries-solutions\u002Fpower-series-ordinary-points":22700,"\u002Fdifferential-equations\u002Fseries-solutions\u002Fregular-singular-frobenius":22705,"\u002Fdifferential-equations\u002Fseries-solutions\u002Fbessel-and-special-functions":22709,"\u002Fdifferential-equations\u002Flaplace\u002Flaplace-definition-ivps":22713,"\u002Fdifferential-equations\u002Flaplace\u002Fstep-impulse-convolution":22718,"\u002Fdifferential-equations\u002Fsystems\u002Fmatrices-eigenvalues-review":22722,"\u002Fdifferential-equations\u002Fsystems\u002Fconstant-coefficient-systems-phase-portraits":22727,"\u002Fdifferential-equations\u002Fsystems\u002Frepeated-eigenvalues-fundamental-matrices":22731,"\u002Fdifferential-equations\u002Fnumerical\u002Feuler-and-runge-kutta":22735,"\u002Fdifferential-equations\u002Fnumerical\u002Fmultistep-systems-stability":22740,"\u002Fdifferential-equations\u002Fnonlinear\u002Fphase-plane-autonomous-stability":22744,"\u002Fdifferential-equations\u002Fnonlinear\u002Flocally-linear-and-liapunov":22749,"\u002Fdifferential-equations\u002Fnonlinear\u002Fcompeting-species-predator-prey-limit-cycles":22753,"\u002Fdifferential-equations\u002Fpdes-fourier-bvp\u002Ffourier-series":22757,"\u002Fdifferential-equations\u002Fpdes-fourier-bvp\u002Fheat-wave-laplace-equations":22762,"\u002Fdifferential-equations\u002Fpdes-fourier-bvp\u002Fsturm-liouville":22766,"\u002Fdifferential-equations\u002Fhistory-variations\u002Fcalculus-of-variations":22770,"\u002Fdifferential-equations\u002Fhistory-variations\u002Fhistorical-notes":22775,"\u002Fdifferential-equations":22779,"\u002Frelativity\u002Ffoundations\u002Fspecial-relativity-postulates":22782,"\u002Frelativity\u002Ffoundations\u002Florentz-transformation-spacetime":22787,"\u002Frelativity\u002Ffoundations\u002Ftime-dilation-length-contraction":22791,"\u002Frelativity\u002Ffoundations\u002Frelativistic-momentum-energy":22795,"\u002Frelativity\u002Ffoundations\u002Fgeneral-relativity":22799,"\u002Frelativity\u002Fspacetime-and-the-lorentz-group\u002Fminkowski-spacetime-and-the-interval":22803,"\u002Frelativity\u002Fspacetime-and-the-lorentz-group\u002Ffour-vectors-and-index-notation":22808,"\u002Frelativity\u002Fspacetime-and-the-lorentz-group\u002Fthe-lorentz-group-and-rapidity":22812,"\u002Frelativity\u002Fspacetime-and-the-lorentz-group\u002Fdoppler-aberration-and-appearance":22816,"\u002Frelativity\u002Frelativistic-dynamics\u002Ffour-momentum-force-and-accelerated-motion":22820,"\u002Frelativity\u002Frelativistic-dynamics\u002Fparticle-decays-and-two-body-kinematics":22825,"\u002Frelativity\u002Frelativistic-dynamics\u002Fcollisions-thresholds-and-the-cm-frame":22829,"\u002Frelativity\u002Frelativistic-dynamics\u002Fmandelstam-variables-and-invariants":22833,"\u002Frelativity\u002Fcovariant-electrodynamics\u002Ffour-current-and-the-four-potential":22837,"\u002Frelativity\u002Fcovariant-electrodynamics\u002Fthe-electromagnetic-field-tensor":22842,"\u002Frelativity\u002Fcovariant-electrodynamics\u002Ftransformation-of-electric-and-magnetic-fields":22846,"\u002Frelativity\u002Fcovariant-electrodynamics\u002Fcovariant-maxwell-and-the-stress-energy-tensor":22850,"\u002Frelativity\u002Fcurved-spacetime\u002Fthe-equivalence-principle-formalized":22854,"\u002Frelativity\u002Fcurved-spacetime\u002Fmanifolds-vectors-and-the-metric":22859,"\u002Frelativity\u002Fcurved-spacetime\u002Fcovariant-derivative-and-christoffel-symbols":22863,"\u002Frelativity\u002Fcurved-spacetime\u002Fgeodesics-and-the-geodesic-equation":22867,"\u002Frelativity\u002Fcurved-spacetime\u002Fcurvature-riemann-and-geodesic-deviation":22871,"\u002Frelativity\u002Fcurved-spacetime\u002Fthe-einstein-field-equations":22875,"\u002Frelativity\u002Fthe-schwarzschild-solution\u002Fthe-schwarzschild-metric":22879,"\u002Frelativity\u002Fthe-schwarzschild-solution\u002Fgeodesics-and-orbits-in-schwarzschild":22884,"\u002Frelativity\u002Fthe-schwarzschild-solution\u002Flight-bending-and-null-geodesics":22888,"\u002Frelativity\u002Ftests-of-general-relativity\u002Fperihelion-precession-of-mercury":22892,"\u002Frelativity\u002Ftests-of-general-relativity\u002Fdeflection-of-light-and-gravitational-lensing":22897,"\u002Frelativity\u002Ftests-of-general-relativity\u002Fgravitational-redshift-and-shapiro-delay":22901,"\u002Frelativity\u002Ftests-of-general-relativity\u002Frelativity-in-technology-gps":22905,"\u002Frelativity\u002Fblack-holes\u002Fhorizons-and-coordinate-singularities":22909,"\u002Frelativity\u002Fblack-holes\u002Frotating-and-charged-black-holes":22914,"\u002Frelativity\u002Fblack-holes\u002Fblack-hole-thermodynamics":22918,"\u002Frelativity\u002Fgravitational-waves\u002Flinearized-gravity-and-wave-solutions":22922,"\u002Frelativity\u002Fgravitational-waves\u002Fgeneration-and-the-quadrupole-formula":22927,"\u002Frelativity\u002Fgravitational-waves\u002Fdetection-ligo-and-the-first-events":22931,"\u002Frelativity\u002Fcosmological-bridge\u002Fthe-cosmological-principle-and-flrw-metric":22935,"\u002Frelativity\u002Fcosmological-bridge\u002Ffriedmann-equations-and-cosmic-dynamics":22940,"\u002Frelativity":22944,"\u002Fphysical-computing":22947,"\u002Fquantum-mechanics\u002Fold-quantum-theory\u002Fblackbody-radiation-and-the-planck-quantum":22950,"\u002Fquantum-mechanics\u002Fold-quantum-theory\u002Fthe-photoelectric-effect-and-the-photon":22955,"\u002Fquantum-mechanics\u002Fold-quantum-theory\u002Fx-rays-and-the-compton-effect":22959,"\u002Fquantum-mechanics\u002Fold-quantum-theory\u002Fthe-old-quantum-theory-bohr-and-sommerfeld":22963,"\u002Fquantum-mechanics\u002Fmatter-waves\u002Fde-broglie-waves-and-electron-diffraction":22967,"\u002Fquantum-mechanics\u002Fmatter-waves\u002Fwave-packets-and-the-probability-interpretation":22972,"\u002Fquantum-mechanics\u002Fmatter-waves\u002Fthe-uncertainty-principle":22976,"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fthe-schrodinger-equation-in-one-dimension":22980,"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fthe-free-particle-and-wave-packet-dynamics":22985,"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fparticle-in-infinite-and-finite-square-wells":22989,"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Foperators-expectation-values-and-the-harmonic-oscillator":22993,"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fthe-dirac-delta-potential":22997,"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fbarrier-penetration-and-quantum-tunneling":23001,"\u002Fquantum-mechanics\u002Fformalism\u002Fhilbert-space-and-dirac-notation":23005,"\u002Fquantum-mechanics\u002Fformalism\u002Fobservables-hermitian-operators-and-eigenvalues":23010,"\u002Fquantum-mechanics\u002Fformalism\u002Fthe-postulates-and-quantum-measurement":23014,"\u002Fquantum-mechanics\u002Fformalism\u002Fposition-momentum-and-continuous-spectra":23018,"\u002Fquantum-mechanics\u002Fformalism\u002Fcommutators-and-the-generalized-uncertainty-principle":23022,"\u002Fquantum-mechanics\u002Fformalism\u002Ftime-evolution-schrodinger-and-heisenberg-pictures":23026,"\u002Fquantum-mechanics\u002Foscillator-and-symmetry\u002Fladder-operators-and-the-number-states":23030,"\u002Fquantum-mechanics\u002Foscillator-and-symmetry\u002Fcoherent-and-squeezed-states":23035,"\u002Fquantum-mechanics\u002Foscillator-and-symmetry\u002Fsymmetries-generators-and-conservation-laws":23039,"\u002Fquantum-mechanics\u002Foscillator-and-symmetry\u002Fparity-time-reversal-and-discrete-symmetries":23043,"\u002Fquantum-mechanics\u002Fangular-momentum\u002Forbital-angular-momentum-and-spherical-harmonics":23047,"\u002Fquantum-mechanics\u002Fangular-momentum\u002Fthe-angular-momentum-algebra":23051,"\u002Fquantum-mechanics\u002Fangular-momentum\u002Faddition-of-angular-momenta-and-clebsch-gordan":23055,"\u002Fquantum-mechanics\u002Fcentral-potentials\u002Fthe-schrodinger-equation-in-three-dimensions":23059,"\u002Fquantum-mechanics\u002Fcentral-potentials\u002Fthe-hydrogen-atom":23064,"\u002Fquantum-mechanics\u002Fcentral-potentials\u002Fthe-isotropic-oscillator-and-hidden-symmetry":23068,"\u002Fquantum-mechanics\u002Fspin\u002Fspin-half-pauli-matrices-and-stern-gerlach":23072,"\u002Fquantum-mechanics\u002Fspin\u002Fspin-in-a-magnetic-field-precession-and-resonance":23077,"\u002Fquantum-mechanics\u002Fspin\u002Ftwo-level-systems-and-the-bloch-sphere":23081,"\u002Fquantum-mechanics\u002Fidentical-particles\u002Fidentical-particles-and-exchange-symmetry":23085,"\u002Fquantum-mechanics\u002Fidentical-particles\u002Fthe-pauli-principle-atoms-and-the-periodic-table":23090,"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Ftime-independent-perturbation-theory":23094,"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Ffine-structure-and-the-real-hydrogen-atom":23099,"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Fthe-zeeman-and-stark-effects":23103,"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Fthe-variational-method":23107,"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Fthe-wkb-approximation":23111,"\u002Fquantum-mechanics":23115,"\u002Freal-analysis\u002Ffoundations\u002Fsets-logic-functions":23118,"\u002Freal-analysis\u002Ffoundations\u002Fordered-fields-completeness":23123,"\u002Freal-analysis\u002Ffoundations\u002Fabsolute-value-bounds":23127,"\u002Freal-analysis\u002Ffoundations\u002Fintervals-uncountability":23131,"\u002Freal-analysis\u002Fsequences-series\u002Fsequences-limits":23135,"\u002Freal-analysis\u002Fsequences-series\u002Flimit-laws-monotone":23140,"\u002Freal-analysis\u002Fsequences-series\u002Flimsup-bolzano-weierstrass":23144,"\u002Freal-analysis\u002Fsequences-series\u002Fcauchy-completeness":23148,"\u002Freal-analysis\u002Fsequences-series\u002Fseries-convergence":23152,"\u002Freal-analysis\u002Fsequences-series\u002Fabsolute-conditional-rearrangement":23156,"\u002Freal-analysis\u002Fmetric-spaces\u002Fmetric-spaces-norms":23160,"\u002Freal-analysis\u002Fmetric-spaces\u002Fopen-closed-sets":23165,"\u002Freal-analysis\u002Fmetric-spaces\u002Fconvergence-completeness":23169,"\u002Freal-analysis\u002Fmetric-spaces\u002Fcompactness":23173,"\u002Freal-analysis\u002Fmetric-spaces\u002Fconnectedness":23177,"\u002Freal-analysis\u002Fcontinuity\u002Flimits-of-functions":23181,"\u002Freal-analysis\u002Fcontinuity\u002Fcontinuous-functions":23185,"\u002Freal-analysis\u002Fcontinuity\u002Fevt-ivt":23189,"\u002Freal-analysis\u002Fcontinuity\u002Funiform-continuity":23193,"\u002Freal-analysis\u002Fcontinuity\u002Fcontinuity-metric-spaces":23197,"\u002Freal-analysis\u002Fcontinuity\u002Flimits-infinity-monotone":23201,"\u002Freal-analysis\u002Fdifferentiation\u002Fthe-derivative":23205,"\u002Freal-analysis\u002Fdifferentiation\u002Fmean-value-theorem":23210,"\u002Freal-analysis\u002Fdifferentiation\u002Ftaylors-theorem":23214,"\u002Freal-analysis\u002Fdifferentiation\u002Finverse-function-1d":23218,"\u002Freal-analysis\u002Friemann-integration\u002Fdarboux-integral":23222,"\u002Freal-analysis\u002Friemann-integration\u002Fintegrability-classes":23227,"\u002Freal-analysis\u002Friemann-integration\u002Fproperties-of-the-integral":23231,"\u002Freal-analysis\u002Friemann-integration\u002Ffundamental-theorem":23235,"\u002Freal-analysis\u002Friemann-integration\u002Flog-exp-improper":23238,"\u002Freal-analysis\u002Ffunction-sequences\u002Fpointwise-uniform-convergence":23242,"\u002Freal-analysis\u002Ffunction-sequences\u002Finterchange-of-limits":23247,"\u002Freal-analysis\u002Ffunction-sequences\u002Fpower-series-weierstrass":23251,"\u002Freal-analysis\u002Ffunction-sequences\u002Fpicard-ode":23255,"\u002Freal-analysis\u002Fseveral-variables\u002Fdifferentiability-rn":23259,"\u002Freal-analysis\u002Fseveral-variables\u002Fgradient-chain-rule":23264,"\u002Freal-analysis\u002Fseveral-variables\u002Fhigher-derivatives-taylor-extrema":23268,"\u002Freal-analysis\u002Fseveral-variables\u002Finverse-implicit-theorems":23272,"\u002Freal-analysis\u002Fseveral-variables\u002Fmultiple-integrals":23276,"\u002Freal-analysis":23280,"\u002Fabstract-algebra\u002Ffoundations\u002Fsets-functions-relations":23283,"\u002Fabstract-algebra\u002Ffoundations\u002Fintegers-and-modular-arithmetic":23287,"\u002Fabstract-algebra\u002Fgroups-and-symmetry\u002Fgroup-axioms-and-first-examples":23291,"\u002Fabstract-algebra\u002Fgroups-and-symmetry\u002Fdihedral-and-symmetric-groups":23296,"\u002Fabstract-algebra\u002Fgroups-and-symmetry\u002Fmatrix-and-quaternion-groups":23300,"\u002Fabstract-algebra\u002Fgroups-and-symmetry\u002Fhomomorphisms-and-group-actions":23304,"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fsubgroups-and-substructures":23308,"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fcyclic-groups":23313,"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fgeneration-and-subgroup-lattices":23317,"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fcosets-lagrange-and-normal-subgroups":23321,"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fisomorphism-theorems":23325,"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fcomposition-series-and-the-alternating-group":23329,"\u002Fabstract-algebra\u002Fgroup-actions-and-sylow\u002Factions-and-cayleys-theorem":23333,"\u002Fabstract-algebra\u002Fgroup-actions-and-sylow\u002Fconjugation-and-the-class-equation":23338,"\u002Fabstract-algebra\u002Fgroup-actions-and-sylow\u002Fsylow-theorems":23342,"\u002Fabstract-algebra\u002Fgroup-actions-and-sylow\u002Fautomorphisms-and-simple-groups":23346,"\u002Fabstract-algebra\u002Fproducts-and-group-structure\u002Fdirect-products-and-finite-abelian-groups":23350,"\u002Fabstract-algebra\u002Fproducts-and-group-structure\u002Fsemidirect-products":23355,"\u002Fabstract-algebra\u002Fproducts-and-group-structure\u002Fnilpotent-and-solvable-groups":23359,"\u002Fabstract-algebra\u002Fproducts-and-group-structure\u002Fclassifying-small-groups":23363,"\u002Fabstract-algebra\u002Fring-theory\u002Frings-definitions-and-examples":23367,"\u002Fabstract-algebra\u002Fring-theory\u002Fideals-quotients-and-homomorphisms":23372,"\u002Fabstract-algebra\u002Fring-theory\u002Ffractions-and-the-chinese-remainder-theorem":23376,"\u002Fabstract-algebra\u002Ffactorization-and-polynomials\u002Feuclidean-domains-pids-ufds":23380,"\u002Fabstract-algebra\u002Ffactorization-and-polynomials\u002Fpolynomial-rings-over-fields":23385,"\u002Fabstract-algebra\u002Ffactorization-and-polynomials\u002Fgauss-lemma-and-unique-factorization":23389,"\u002Fabstract-algebra\u002Ffactorization-and-polynomials\u002Firreducibility-criteria-and-groebner":23393,"\u002Fabstract-algebra\u002Fmodule-theory\u002Fintroduction-to-modules":23397,"\u002Fabstract-algebra\u002Fmodule-theory\u002Ffree-modules-and-direct-sums":23402,"\u002Fabstract-algebra\u002Fmodule-theory\u002Ftensor-products-and-exact-sequences":23406,"\u002Fabstract-algebra\u002Fmodule-theory\u002Fvector-spaces-and-linear-maps":23410,"\u002Fabstract-algebra\u002Fmodules-over-pids\u002Fstructure-theorem-over-pids":23414,"\u002Fabstract-algebra\u002Fmodules-over-pids\u002Frational-canonical-form":23419,"\u002Fabstract-algebra\u002Fmodules-over-pids\u002Fjordan-canonical-form":23423,"\u002Fabstract-algebra\u002Ffield-theory\u002Ffield-extensions-and-algebraic-elements":23427,"\u002Fabstract-algebra\u002Ffield-theory\u002Fstraightedge-and-compass-constructions":23432,"\u002Fabstract-algebra\u002Ffield-theory\u002Fsplitting-fields-and-algebraic-closure":23436,"\u002Fabstract-algebra\u002Ffield-theory\u002Fseparable-and-cyclotomic-extensions":23440,"\u002Fabstract-algebra\u002Fgalois-theory\u002Fthe-galois-correspondence":23444,"\u002Fabstract-algebra\u002Fgalois-theory\u002Ffinite-fields":23449,"\u002Fabstract-algebra\u002Fgalois-theory\u002Fcyclotomic-and-abelian-extensions":23453,"\u002Fabstract-algebra\u002Fgalois-theory\u002Fgalois-groups-of-polynomials":23457,"\u002Fabstract-algebra\u002Fgalois-theory\u002Fsolvability-by-radicals-and-the-quintic":23461,"\u002Fabstract-algebra\u002Fcapstone\u002Fcommutative-algebra-and-algebraic-geometry":23465,"\u002Fabstract-algebra\u002Fcapstone\u002Frepresentation-and-character-theory":23470,"\u002Fabstract-algebra":23474,"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fatomic-spectra-rutherford":23477,"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fbohr-model-hydrogen":23482,"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fx-ray-spectra-franck-hertz":23486,"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fbohr-sommerfeld-old-quantum-theory":23490,"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fold-quantum-theory-limits-wkb":23494,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fschrodinger-3d-hydrogen":23498,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fhydrogen-wave-functions":23503,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fradial-equation-in-full":23507,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fsymmetry-degeneracy-runge-lenz":23511,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fexpectation-values-virial":23515,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fquantum-defects-alkali-spectra":23519,"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Frydberg-atoms":23523,"\u002Fatomic-physics\u002Ffine-structure-and-the-dirac-atom\u002Frelativistic-kinetic-correction":23527,"\u002Fatomic-physics\u002Ffine-structure-and-the-dirac-atom\u002Fspin-orbit-thomas-precession":23532,"\u002Fatomic-physics\u002Ffine-structure-and-the-dirac-atom\u002Fdarwin-term-fine-structure-formula":23536,"\u002Fatomic-physics\u002Ffine-structure-and-the-dirac-atom\u002Fdirac-equation-hydrogen":23540,"\u002Fatomic-physics\u002Fqed-corrections-and-hyperfine-structure\u002Flamb-shift-qed":23544,"\u002Fatomic-physics\u002Fqed-corrections-and-hyperfine-structure\u002Fhyperfine-structure-21cm":23549,"\u002Fatomic-physics\u002Fqed-corrections-and-hyperfine-structure\u002Fnuclear-effects-isotope-shift":23553,"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fperiodic-table-atomic-spectra":23557,"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fcentral-field-self-consistent":23562,"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fidentical-particles-hartree-fock":23566,"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fhelium-two-electron-atom":23570,"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fls-jj-coupling-term-symbols":23574,"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fhund-rules-ground-terms":23578,"\u002Fatomic-physics\u002Fatoms-in-external-fields\u002Fzeeman-effect":23582,"\u002Fatomic-physics\u002Fatoms-in-external-fields\u002Fpaschen-back-intermediate":23587,"\u002Fatomic-physics\u002Fatoms-in-external-fields\u002Fstark-effect-polarizability":23591,"\u002Fatomic-physics\u002Fradiative-transitions-and-line-shapes\u002Ftime-dependent-perturbation-golden-rule":23595,"\u002Fatomic-physics\u002Fradiative-transitions-and-line-shapes\u002Fdipole-approximation-einstein-coefficients":23600,"\u002Fatomic-physics\u002Fradiative-transitions-and-line-shapes\u002Fselection-rules-forbidden-transitions":23604,"\u002Fatomic-physics\u002Fradiative-transitions-and-line-shapes\u002Flifetimes-and-line-shapes":23608,"\u002Fatomic-physics\u002Flasers-and-spectroscopy\u002Flaser-principles":23612,"\u002Fatomic-physics\u002Flasers-and-spectroscopy\u002Fspectroscopy-techniques":23617,"\u002Fatomic-physics\u002Flasers-and-spectroscopy\u002Fline-catalog-nist-asd":23621,"\u002Fatomic-physics\u002Fmodern-atomic-physics\u002Flaser-cooling-doppler":23625,"\u002Fatomic-physics\u002Fmodern-atomic-physics\u002Fsub-doppler-trapping":23630,"\u002Fatomic-physics\u002Fmodern-atomic-physics\u002Fbose-einstein-condensation":23634,"\u002Fatomic-physics\u002Fmodern-atomic-physics\u002Foptical-clocks-precision":23638,"\u002Fatomic-physics":23642,"\u002Fdatabases":23645,"\u002Fcategory-theory\u002Ffoundations\u002Fwhat-is-a-category":23648,"\u002Fcategory-theory\u002Ffoundations\u002Fexamples-of-categories":23652,"\u002Fcategory-theory\u002Ffoundations\u002Fspecial-morphisms":23656,"\u002Fcategory-theory\u002Ffoundations\u002Ffunctors":23660,"\u002Fcategory-theory\u002Ffoundations\u002Fnatural-transformations":23664,"\u002Fcategory-theory\u002Ffoundations\u002Fsize-and-set-theory":23668,"\u002Fcategory-theory\u002Funiversal-properties\u002Funiversal-properties":23672,"\u002Fcategory-theory\u002Funiversal-properties\u002Fproducts-and-coproducts":23677,"\u002Fcategory-theory\u002Funiversal-properties\u002Fconstructions-on-categories":23681,"\u002Fcategory-theory\u002Frepresentables-yoneda\u002Frepresentable-functors":23685,"\u002Fcategory-theory\u002Frepresentables-yoneda\u002Fyoneda-lemma":23690,"\u002Fcategory-theory\u002Frepresentables-yoneda\u002Fyoneda-consequences":23694,"\u002Fcategory-theory\u002Flimits-colimits\u002Flimits":23698,"\u002Fcategory-theory\u002Flimits-colimits\u002Fproducts-equalizers-pullbacks":23703,"\u002Fcategory-theory\u002Flimits-colimits\u002Fcolimits":23707,"\u002Fcategory-theory\u002Flimits-colimits\u002Fcomputing-limits":23711,"\u002Fcategory-theory\u002Flimits-colimits\u002Flimits-and-functors":23715,"\u002Fcategory-theory\u002Fadjunctions\u002Fadjunctions":23719,"\u002Fcategory-theory\u002Fadjunctions\u002Funits-and-counits":23724,"\u002Fcategory-theory\u002Fadjunctions\u002Fadjunctions-via-universal-arrows":23728,"\u002Fcategory-theory\u002Fadjunctions\u002Ffree-forgetful-adjunctions":23732,"\u002Fcategory-theory\u002Fadjoints-limits\u002Flimits-via-adjoints":23736,"\u002Fcategory-theory\u002Fadjoints-limits\u002Fpresheaf-limits-colimits":23741,"\u002Fcategory-theory\u002Fadjoints-limits\u002Fadjoints-preserve-limits":23745,"\u002Fcategory-theory\u002Fadjoints-limits\u002Fadjoint-functor-theorem":23749,"\u002Fcategory-theory\u002Fmonads-algebras\u002Fmonads":23753,"\u002Fcategory-theory\u002Fmonads-algebras\u002Falgebras-eilenberg-moore":23758,"\u002Fcategory-theory\u002Fmonads-algebras\u002Fkleisli-and-programming":23762,"\u002Fcategory-theory\u002Fmonads-algebras\u002Falgebras-for-endofunctors":23766,"\u002Fcategory-theory\u002Fcartesian-closed-lambda\u002Fcartesian-closed-categories":23770,"\u002Fcategory-theory\u002Fcartesian-closed-lambda\u002Flambda-calculus-correspondence":23775,"\u002Fcategory-theory\u002Fcartesian-closed-lambda\u002Ffixed-points-and-recursion":23779,"\u002Fcategory-theory":23783,"\u002Fdeep-learning\u002Fmathematical-background\u002Flinear-algebra-for-deep-learning":23786,"\u002Fdeep-learning\u002Fmathematical-background\u002Fprobability-and-information-theory":23790,"\u002Fdeep-learning\u002Fmathematical-background\u002Fnumerical-computation":23794,"\u002Fdeep-learning\u002Fmathematical-background\u002Fcalculus":23798,"\u002Fdeep-learning\u002Ffoundations\u002Fwhat-is-deep-learning":23801,"\u002Fdeep-learning\u002Ffoundations\u002Fmachine-learning-refresher":23805,"\u002Fdeep-learning\u002Ffoundations\u002Flinear-models-and-the-perceptron":23809,"\u002Fdeep-learning\u002Fneural-networks\u002Fthe-multilayer-perceptron":23813,"\u002Fdeep-learning\u002Fneural-networks\u002Factivation-functions":23816,"\u002Fdeep-learning\u002Fneural-networks\u002Funiversal-approximation":23817,"\u002Fdeep-learning\u002Fneural-networks\u002Fbackpropagation":23821,"\u002Fdeep-learning\u002Fneural-networks\u002Floss-functions-and-output-units":23824,"\u002Fdeep-learning\u002Foptimization\u002Fgradient-descent-and-sgd":23827,"\u002Fdeep-learning\u002Foptimization\u002Fmomentum-and-adaptive-methods":23832,"\u002Fdeep-learning\u002Foptimization\u002Finitialization":23836,"\u002Fdeep-learning\u002Foptimization\u002Fthe-optimization-landscape":23839,"\u002Fdeep-learning\u002Foptimization\u002Fsecond-order-and-approximate-methods":23843,"\u002Fdeep-learning\u002Fregularization\u002Fregularization-overview":23847,"\u002Fdeep-learning\u002Fregularization\u002Fdropout-and-data-augmentation":23852,"\u002Fdeep-learning\u002Fregularization\u002Fearly-stopping-and-parameter-sharing":23856,"\u002Fdeep-learning\u002Fregularization\u002Fnormalization":23860,"\u002Fdeep-learning\u002Farchitectures\u002Fconvolutional-networks":23864,"\u002Fdeep-learning\u002Farchitectures\u002Fcnn-architectures":23869,"\u002Fdeep-learning\u002Farchitectures\u002Frecurrent-networks":23873,"\u002Fdeep-learning\u002Farchitectures\u002Flstm-and-gru":23877,"\u002Fdeep-learning\u002Farchitectures\u002Fattention-and-transformers":23881,"\u002Fdeep-learning\u002Farchitectures\u002Fthe-transformer-architecture":23885,"\u002Fdeep-learning\u002Farchitectures\u002Ftransformers-in-practice":23889,"\u002Fdeep-learning\u002Farchitectures\u002Fgraph-neural-networks":23893,"\u002Fdeep-learning\u002Farchitectures\u002Fstate-space-models":23897,"\u002Fdeep-learning\u002Ftheory\u002Fgeneralization-theory":23901,"\u002Fdeep-learning\u002Ftheory\u002Fadversarial-robustness":23906,"\u002Fdeep-learning\u002Ftheory\u002Fadversarial-defenses":23910,"\u002Fdeep-learning\u002Ftheory\u002Fbayesian-and-ensemble-methods":23914,"\u002Fdeep-learning\u002Ftheory\u002Fdeep-equilibrium-models":23918,"\u002Fdeep-learning\u002Fgenerative-models\u002Flinear-factor-models":23922,"\u002Fdeep-learning\u002Fgenerative-models\u002Fautoencoders":23927,"\u002Fdeep-learning\u002Fgenerative-models\u002Fvariational-autoencoders":23931,"\u002Fdeep-learning\u002Fgenerative-models\u002Fgenerative-adversarial-networks":23935,"\u002Fdeep-learning\u002Fgenerative-models\u002Fautoregressive-and-normalizing-flows":23939,"\u002Fdeep-learning\u002Fgenerative-models\u002Fenergy-based-and-boltzmann-machines":23943,"\u002Fdeep-learning\u002Fgenerative-models\u002Fdiffusion-and-score-based-models":23947,"\u002Fdeep-learning\u002Fprobabilistic-methods\u002Fstructured-probabilistic-models":23951,"\u002Fdeep-learning\u002Fprobabilistic-methods\u002Fmonte-carlo-and-mcmc":23956,"\u002Fdeep-learning\u002Fprobabilistic-methods\u002Fapproximate-inference":23960,"\u002Fdeep-learning\u002Fpractical\u002Fpractical-methodology":23964,"\u002Fdeep-learning\u002Fpractical\u002Fhyperparameters-and-debugging":23969,"\u002Fdeep-learning\u002Fpractical\u002Frepresentation-learning":23973,"\u002Fdeep-learning\u002Fpractical\u002Ftransfer-learning":23977,"\u002Fdeep-learning\u002Fpractical\u002Fapplications":23981,"\u002Fdeep-learning\u002Fpractical\u002Fmodel-compression-and-distillation":23985,"\u002Fdeep-learning\u002Fpractical\u002Fmeta-learning-and-few-shot":23989,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Flarge-language-models":23993,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fscaling-inference-and-alignment":23998,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fseq2seq-pretraining-and-bart":24002,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Ftext-to-text-transfer-and-conditional-generation":24006,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fspeech-and-audio-models":24010,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fself-supervised-speech-and-synthesis":24014,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fai-agents":24018,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fagent-memory-retrieval-and-orchestration":24022,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fmixture-of-experts":24026,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fmultimodal-models":24030,"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Ffusion-and-vision-language-models":24034,"\u002Fdeep-learning\u002Freinforcement-learning\u002Ffoundations-of-reinforcement-learning":24038,"\u002Fdeep-learning\u002Freinforcement-learning\u002Fmodel-free-prediction-and-control":24043,"\u002Fdeep-learning\u002Freinforcement-learning\u002Fdeep-q-networks":24047,"\u002Fdeep-learning\u002Freinforcement-learning\u002Fpolicy-gradients-and-actor-critic":24051,"\u002Fdeep-learning\u002Freinforcement-learning\u002Frl-from-human-feedback":24055,"\u002Fdeep-learning":24059,"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Fequilibrium-state-variables-zeroth-law":24061,"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Ffirst-law-heat-and-work":24065,"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Fsecond-law-entropy-and-the-carnot-bound":24069,"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Fthermodynamic-potentials-and-maxwell-relations":24073,"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Fstability-response-functions-and-the-third-law":24077,"\u002Fstatistical-mechanics\u002Ffoundations\u002Fclassical-statistics-and-equipartition":24081,"\u002Fstatistical-mechanics\u002Ffoundations\u002Fphase-space-and-liouvilles-theorem":24086,"\u002Fstatistical-mechanics\u002Ffoundations\u002Fensembles-and-the-equal-probability-postulate":24090,"\u002Fstatistical-mechanics\u002Ffoundations\u002Fstatistical-entropy-boltzmann-and-gibbs":24094,"\u002Fstatistical-mechanics\u002Fmicrocanonical\u002Fmicrocanonical-ensemble-and-entropy":24098,"\u002Fstatistical-mechanics\u002Fmicrocanonical\u002Fequilibrium-conditions-temperature-pressure-chemical-potential":24103,"\u002Fstatistical-mechanics\u002Fmicrocanonical\u002Fideal-gas-phase-space-and-the-sackur-tetrode-entropy":24107,"\u002Fstatistical-mechanics\u002Fmicrocanonical\u002Ftwo-state-systems-paramagnets-and-negative-temperature":24111,"\u002Fstatistical-mechanics\u002Fcanonical\u002Fcanonical-ensemble-and-the-boltzmann-distribution":24115,"\u002Fstatistical-mechanics\u002Fcanonical\u002Fpartition-function-and-the-helmholtz-free-energy":24120,"\u002Fstatistical-mechanics\u002Fcanonical\u002Fenergy-fluctuations-and-ensemble-equivalence":24124,"\u002Fstatistical-mechanics\u002Fcanonical\u002Fthe-einstein-solid-and-harmonic-systems":24128,"\u002Fstatistical-mechanics\u002Fcanonical\u002Fparamagnetism-and-the-schottky-anomaly":24132,"\u002Fstatistical-mechanics\u002Fclassical-gas\u002Fideal-gas-partition-function-and-the-gibbs-paradox":24136,"\u002Fstatistical-mechanics\u002Fclassical-gas\u002Fequipartition-and-the-virial-theorem":24141,"\u002Fstatistical-mechanics\u002Fclassical-gas\u002Fmolecular-gases-rotation-and-vibration":24145,"\u002Fstatistical-mechanics\u002Fgrand-canonical\u002Fgrand-canonical-ensemble-and-the-grand-partition-function":24149,"\u002Fstatistical-mechanics\u002Fgrand-canonical\u002Fchemical-potential-fugacity-and-number-fluctuations":24154,"\u002Fstatistical-mechanics\u002Fgrand-canonical\u002Fensemble-summary-and-the-thermodynamic-web":24158,"\u002Fstatistical-mechanics\u002Fquantum-statistics\u002Fquantum-statistics-bose-einstein-and-fermi-dirac":24162,"\u002Fstatistical-mechanics\u002Fquantum-statistics\u002Fderiving-the-quantum-distributions":24167,"\u002Fstatistical-mechanics\u002Fquantum-statistics\u002Fthe-classical-limit-and-quantum-concentration":24171,"\u002Fstatistical-mechanics\u002Fquantum-statistics\u002Fideal-quantum-gases-general-framework":24175,"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fbose-einstein-condensation-and-the-fermion-gas":24179,"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fthe-photon-gas-and-plancks-radiation-law":24184,"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fblackbody-thermodynamics-and-radiation-pressure":24188,"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fphonons-and-the-debye-model":24192,"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fbose-einstein-condensation-derived":24196,"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fthermodynamics-of-the-bose-gas-and-superfluidity":24200,"\u002Fstatistical-mechanics\u002Ffermi-gas\u002Fthe-ideal-fermi-gas-at-zero-temperature":24204,"\u002Fstatistical-mechanics\u002Ffermi-gas\u002Fsommerfeld-expansion-and-electrons-in-metals":24209,"\u002Fstatistical-mechanics\u002Ffermi-gas\u002Fwhite-dwarfs-and-the-chandrasekhar-limit":24213,"\u002Fstatistical-mechanics\u002Ffermi-gas\u002Fneutron-stars-and-nuclear-matter":24217,"\u002Fstatistical-mechanics\u002Finteractions\u002Fthe-cluster-expansion-and-virial-coefficients":24221,"\u002Fstatistical-mechanics\u002Finteractions\u002Fthe-van-der-waals-gas-and-liquid-gas-coexistence":24226,"\u002Fstatistical-mechanics\u002Finteractions\u002Fquantum-gases-with-interactions-and-exchange":24230,"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fphases-coexistence-and-classification":24234,"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fthe-ising-model-and-exact-solutions":24239,"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fmean-field-theory-and-the-weiss-model":24243,"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fcritical-exponents-and-landau-theory":24247,"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fthe-renormalization-group-idea":24251,"\u002Fstatistical-mechanics\u002Ffluctuations\u002Fthermodynamic-fluctuations-and-response":24255,"\u002Fstatistical-mechanics\u002Ffluctuations\u002Fbrownian-motion-and-the-langevin-equation":24260,"\u002Fstatistical-mechanics\u002Ffluctuations\u002Flinear-response-and-the-fluctuation-dissipation-theorem":24264,"\u002Fstatistical-mechanics":24268,"\u002Fcondensed-matter\u002Fmolecules-and-bonding\u002Fbonding-mechanisms":24271,"\u002Fcondensed-matter\u002Fmolecules-and-bonding\u002Fmolecular-orbitals-and-h2-plus":24276,"\u002Fcondensed-matter\u002Fmolecules-and-bonding\u002Fhydrogen-molecule-and-exchange":24280,"\u002Fcondensed-matter\u002Fmolecules-and-bonding\u002Fvan-der-waals-forces":24284,"\u002Fcondensed-matter\u002Fmolecular-spectra\u002Frotational-vibrational-spectra":24288,"\u002Fcondensed-matter\u002Fmolecular-spectra\u002Fanharmonicity-and-rovibrational-structure":24293,"\u002Fcondensed-matter\u002Fmolecular-spectra\u002Framan-and-electronic-bands":24297,"\u002Fcondensed-matter\u002Fmolecular-spectra\u002Flasers-and-masers":24301,"\u002Fcondensed-matter\u002Fcrystal-structure\u002Fstructure-of-solids":24305,"\u002Fcondensed-matter\u002Fcrystal-structure\u002Fbravais-lattices-and-crystal-systems":24310,"\u002Fcondensed-matter\u002Fcrystal-structure\u002Freciprocal-lattice-and-brillouin-zones":24314,"\u002Fcondensed-matter\u002Fcrystal-structure\u002Fdiffraction-and-structure-factors":24318,"\u002Fcondensed-matter\u002Flattice-dynamics\u002Fphonon-dispersion":24322,"\u002Fcondensed-matter\u002Flattice-dynamics\u002Fphonons-quantization-and-dos":24327,"\u002Fcondensed-matter\u002Flattice-dynamics\u002Fdebye-einstein-heat-capacity":24331,"\u002Fcondensed-matter\u002Flattice-dynamics\u002Fanharmonicity-and-thermal-transport":24335,"\u002Fcondensed-matter\u002Ffree-electron-fermi-gas\u002Ffree-electron-gas-and-conduction":24339,"\u002Fcondensed-matter\u002Ffree-electron-fermi-gas\u002Fsommerfeld-model-and-heat-capacity":24344,"\u002Fcondensed-matter\u002Ffree-electron-fermi-gas\u002Ftransport-and-the-hall-effect":24348,"\u002Fcondensed-matter\u002Ffree-electron-fermi-gas\u002Fscreening-and-plasmons":24352,"\u002Fcondensed-matter\u002Fband-theory\u002Fblochs-theorem-and-energy-bands":24356,"\u002Fcondensed-matter\u002Fband-theory\u002Fnearly-free-electron-model":24361,"\u002Fcondensed-matter\u002Fband-theory\u002Ftight-binding-method":24365,"\u002Fcondensed-matter\u002Fband-theory\u002Ffermi-surfaces-and-semiclassical-dynamics":24369,"\u002Fcondensed-matter\u002Fsemiconductors\u002Fsemiconductor-bands-and-junctions":24373,"\u002Fcondensed-matter\u002Fsemiconductors\u002Fintrinsic-and-extrinsic-semiconductors":24378,"\u002Fcondensed-matter\u002Fsemiconductors\u002Fcarrier-transport-and-recombination":24382,"\u002Fcondensed-matter\u002Fsemiconductors\u002Fthe-pn-junction":24386,"\u002Fcondensed-matter\u002Fsemiconductors\u002Ftransistors-and-optoelectronics":24390,"\u002Fcondensed-matter\u002Fdielectrics-and-ferroelectrics\u002Fdielectrics-and-polarization":24394,"\u002Fcondensed-matter\u002Fdielectrics-and-ferroelectrics\u002Fferroelectrics-and-piezoelectrics":24399,"\u002Fcondensed-matter\u002Fmagnetism\u002Fdiamagnetism-and-paramagnetism":24403,"\u002Fcondensed-matter\u002Fmagnetism\u002Fexchange-and-ferromagnetism":24408,"\u002Fcondensed-matter\u002Fmagnetism\u002Fantiferromagnetism-and-domains":24412,"\u002Fcondensed-matter\u002Fmagnetism\u002Fspin-waves-and-magnons":24416,"\u002Fcondensed-matter\u002Fsuperconductivity\u002Fsuperconductivity-phenomenology":24420,"\u002Fcondensed-matter\u002Fsuperconductivity\u002Flondon-theory-and-the-meissner-effect":24425,"\u002Fcondensed-matter\u002Fsuperconductivity\u002Fginzburg-landau-theory":24429,"\u002Fcondensed-matter\u002Fsuperconductivity\u002Fbcs-theory":24433,"\u002Fcondensed-matter\u002Fsuperconductivity\u002Fjosephson-and-high-tc":24437,"\u002Fcondensed-matter\u002Fnanostructures\u002Fquantum-wells-wires-and-dots":24441,"\u002Fcondensed-matter\u002Fnanostructures\u002Finteger-quantum-hall-effect":24446,"\u002Fcondensed-matter\u002Fnanostructures\u002Ffractional-quantum-hall-and-topology":24450,"\u002Fcondensed-matter\u002Fnanostructures\u002Fgraphene-and-dirac-materials":24454,"\u002Fcondensed-matter":24458,"\u002Flogic\u002Ffoundations\u002Flogic-as-a-mathematical-model":24461,"\u002Flogic\u002Fsentential-logic\u002Fformal-languages-and-well-formed-formulas":24465,"\u002Flogic\u002Fsentential-logic\u002Ftruth-assignments-and-tautologies":24470,"\u002Flogic\u002Fsentential-logic\u002Funique-readability-and-parsing":24474,"\u002Flogic\u002Fsentential-logic\u002Finduction-and-recursion":24478,"\u002Flogic\u002Fsentential-logic\u002Fexpressive-completeness-and-normal-forms":24482,"\u002Flogic\u002Fsentential-logic\u002Fboolean-circuits":24486,"\u002Flogic\u002Fsentential-logic\u002Fcompactness-and-effectiveness":24490,"\u002Flogic\u002Ffirst-order-languages\u002Ffirst-order-languages":24494,"\u002Flogic\u002Ffirst-order-languages\u002Fstructures-truth-and-satisfaction":24499,"\u002Flogic\u002Ffirst-order-languages\u002Fdefinability-and-elementary-equivalence":24503,"\u002Flogic\u002Ffirst-order-languages\u002Fterms-substitution-and-parsing":24507,"\u002Flogic\u002Fdeductive-calculus\u002Fa-deductive-calculus":24511,"\u002Flogic\u002Fdeductive-calculus\u002Fdeduction-theorem-and-derived-rules":24516,"\u002Flogic\u002Fdeductive-calculus\u002Fsoundness":24520,"\u002Flogic\u002Fdeductive-calculus\u002Fcompleteness-and-consistency":24524,"\u002Flogic\u002Fmodels-and-theories\u002Fcompactness-and-lowenheim-skolem":24528,"\u002Flogic\u002Fmodels-and-theories\u002Ftheories-elementary-classes-and-categoricity":24533,"\u002Flogic\u002Fmodels-and-theories\u002Finterpretations-between-theories":24537,"\u002Flogic\u002Fmodels-and-theories\u002Fnonstandard-analysis":24541,"\u002Flogic\u002Farithmetic-and-definability\u002Fdefinability-in-arithmetic":24545,"\u002Flogic\u002Farithmetic-and-definability\u002Fnatural-numbers-with-successor":24550,"\u002Flogic\u002Farithmetic-and-definability\u002Fpresburger-and-reducts":24554,"\u002Flogic\u002Farithmetic-and-definability\u002Fa-subtheory-and-representability":24558,"\u002Flogic\u002Fincompleteness\u002Farithmetization-of-syntax":24562,"\u002Flogic\u002Fincompleteness\u002Fincompleteness-and-undecidability":24567,"\u002Flogic\u002Fincompleteness\u002Fsecond-incompleteness-theorem":24571,"\u002Flogic\u002Fcomputability-and-representability\u002Frecursive-functions":24575,"\u002Flogic\u002Fcomputability-and-representability\u002Frepresenting-exponentiation":24580,"\u002Flogic\u002Fsecond-order-logic\u002Fsecond-order-languages":24584,"\u002Flogic\u002Fsecond-order-logic\u002Fskolem-functions-and-many-sorted-logic":24589,"\u002Flogic\u002Fsecond-order-logic\u002Fgeneral-structures":24593,"\u002Flogic":24597,"\u002Freinforcement-learning\u002Ffoundations\u002Fwhat-is-reinforcement-learning":24600,"\u002Freinforcement-learning\u002Ffoundations\u002Fa-brief-history-of-rl":24604,"\u002Freinforcement-learning\u002Ffoundations\u002Fmulti-armed-bandits":24608,"\u002Freinforcement-learning\u002Ffoundations\u002Fbandit-exploration-algorithms":24612,"\u002Freinforcement-learning\u002Ffoundations\u002Fmarkov-decision-processes":24616,"\u002Freinforcement-learning\u002Ffoundations\u002Fvalue-functions-and-optimality":24620,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fdynamic-programming":24624,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fdp-async-and-gpi":24628,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fmonte-carlo-methods":24632,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fmonte-carlo-off-policy":24636,"\u002Freinforcement-learning\u002Ftabular-methods\u002Ftemporal-difference-learning":24640,"\u002Freinforcement-learning\u002Ftabular-methods\u002Ftd-control-sarsa-and-q-learning":24644,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fn-step-bootstrapping":24648,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fn-step-off-policy-methods":24652,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fplanning-and-learning":24656,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fplanning-focusing-and-decision-time":24660,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fdecision-time-planning":24664,"\u002Freinforcement-learning\u002Ftabular-methods\u002Fmonte-carlo-tree-search":24668,"\u002Freinforcement-learning\u002Fapproximation\u002Fon-policy-prediction":24672,"\u002Freinforcement-learning\u002Fapproximation\u002Ffeature-construction-and-nonlinear":24677,"\u002Freinforcement-learning\u002Fapproximation\u002Fon-policy-control":24681,"\u002Freinforcement-learning\u002Fapproximation\u002Faverage-reward-control":24685,"\u002Freinforcement-learning\u002Fapproximation\u002Foff-policy-and-the-deadly-triad":24689,"\u002Freinforcement-learning\u002Fapproximation\u002Fbellman-error-and-gradient-td":24693,"\u002Freinforcement-learning\u002Fapproximation\u002Feligibility-traces":24697,"\u002Freinforcement-learning\u002Fapproximation\u002Ftrue-online-and-sarsa-lambda":24701,"\u002Freinforcement-learning\u002Fapproximation\u002Fpolicy-gradient-methods":24705,"\u002Freinforcement-learning\u002Fapproximation\u002Factor-critic-and-continuous-actions":24709,"\u002Freinforcement-learning\u002Fapproximation\u002Fleast-squares-and-memory-based-methods":24713,"\u002Freinforcement-learning\u002Fapproximation\u002Fmemory-and-kernel-methods":24717,"\u002Freinforcement-learning\u002Fapproximation\u002Foff-policy-eligibility-traces":24721,"\u002Freinforcement-learning\u002Fapproximation\u002Fstable-off-policy-traces":24725,"\u002Freinforcement-learning\u002Fdeep-rl\u002Fdeep-q-networks":24729,"\u002Freinforcement-learning\u002Fdeep-rl\u002Fdqn-improvements":24733,"\u002Freinforcement-learning\u002Fdeep-rl\u002Factor-critic-and-ppo":24737,"\u002Freinforcement-learning\u002Fdeep-rl\u002Fppo-and-continuous-control":24741,"\u002Freinforcement-learning\u002Fdeep-rl\u002Fcase-studies":24745,"\u002Freinforcement-learning\u002Fdeep-rl\u002Frl-beyond-games":24749,"\u002Freinforcement-learning\u002Fdeep-rl\u002Ffrontiers":24753,"\u002Freinforcement-learning\u002Fdeep-rl\u002Freward-design-and-open-problems":24757,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fdistributional-and-rainbow":24761,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fdistributional-and-rainbow-part-2":24766,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fcontinuous-control":24770,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fcontinuous-control-part-2":24774,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmodel-based-rl":24778,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmodel-based-rl-part-2":24782,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fexploration":24786,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fexploration-part-2":24790,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Foffline-rl":24794,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Foffline-rl-part-2":24798,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fimitation-and-inverse-rl":24802,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fimitation-and-inverse-rl-part-2":24806,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmulti-agent-rl":24810,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmulti-agent-rl-part-2":24814,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fhierarchical-rl":24818,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fhierarchical-rl-part-2":24822,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Frlhf-and-language-models":24826,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fpartial-observability-pomdps":24830,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fpartial-observability-pomdps-part-2":24834,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fsafe-and-constrained-rl":24838,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fsafe-and-constrained-rl-part-2":24842,"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmeta-rl-and-generalization":24846,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fpsychology-of-reinforcement":24850,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Finstrumental-conditioning-and-control":24855,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fdopamine-and-td-error":24859,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fdopamine-in-the-brain":24863,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fanimal-learning-and-cognition":24867,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fcognitive-maps-and-planning":24871,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fneuroscience-of-reinforcement":24875,"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fseveral-learning-systems":24879,"\u002Freinforcement-learning":24883,"\u002Fartificial-intelligence\u002Ffoundations\u002Fwhat-is-ai":24885,"\u002Fartificial-intelligence\u002Ffoundations\u002Ffoundations-of-ai":24889,"\u002Fartificial-intelligence\u002Ffoundations\u002Fintelligent-agents":24893,"\u002Fartificial-intelligence\u002Ffoundations\u002Fagent-architectures":24897,"\u002Fartificial-intelligence\u002Fsearch\u002Funinformed-search":24901,"\u002Fartificial-intelligence\u002Fsearch\u002Fsearch-strategies-compared":24906,"\u002Fartificial-intelligence\u002Fsearch\u002Finformed-search":24910,"\u002Fartificial-intelligence\u002Fsearch\u002Fheuristic-functions":24914,"\u002Fartificial-intelligence\u002Fsearch\u002Flocal-search":24918,"\u002Fartificial-intelligence\u002Fsearch\u002Fpopulation-and-continuous-search":24922,"\u002Fartificial-intelligence\u002Fsearch\u002Fadversarial-search":24926,"\u002Fartificial-intelligence\u002Fsearch\u002Fgames-of-chance-and-imperfect-information":24930,"\u002Fartificial-intelligence\u002Fsearch\u002Fconstraint-satisfaction":24934,"\u002Fartificial-intelligence\u002Fsearch\u002Fcsp-search-and-structure":24938,"\u002Fartificial-intelligence\u002Fsearch\u002Fsearch-under-uncertainty":24942,"\u002Fartificial-intelligence\u002Fsearch\u002Fbelief-state-and-online-search":24946,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fpropositional-logic":24950,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fpropositional-inference":24955,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Ffirst-order-logic":24959,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Ffirst-order-logic-in-use":24963,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Finference-and-resolution":24967,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Ffirst-order-resolution":24971,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fclassical-planning":24975,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fplanning-graphs-and-graphplan":24979,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fplanning-in-the-real-world":24983,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fplanning-under-uncertainty":24987,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fknowledge-representation":24991,"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Freasoning-systems-and-defaults":24995,"\u002Fartificial-intelligence\u002Funcertainty\u002Fprobability-and-bayes":24999,"\u002Fartificial-intelligence\u002Funcertainty\u002Fbayes-rule-and-naive-bayes":25004,"\u002Fartificial-intelligence\u002Funcertainty\u002Fbayesian-networks":25008,"\u002Fartificial-intelligence\u002Funcertainty\u002Finference-in-bayesian-networks":25012,"\u002Fartificial-intelligence\u002Funcertainty\u002Freasoning-over-time":25016,"\u002Fartificial-intelligence\u002Funcertainty\u002Ftracking-and-data-association":25020,"\u002Fartificial-intelligence\u002Funcertainty\u002Fmaking-decisions":25024,"\u002Fartificial-intelligence\u002Funcertainty\u002Fmarkov-decision-processes":25028,"\u002Fartificial-intelligence\u002Funcertainty\u002Fdecision-networks-and-game-theory":25031,"\u002Fartificial-intelligence\u002Funcertainty\u002Fgame-theory-and-mechanism-design":25035,"\u002Fartificial-intelligence\u002Flearning\u002Flearning-from-examples":25039,"\u002Fartificial-intelligence\u002Flearning\u002Ftheory-and-model-families":25044,"\u002Fartificial-intelligence\u002Flearning\u002Fprobabilistic-learning":25048,"\u002Fartificial-intelligence\u002Flearning\u002Fexpectation-maximization":25052,"\u002Fartificial-intelligence\u002Flearning\u002Freinforcement-learning":25056,"\u002Fartificial-intelligence\u002Flearning\u002Fgeneralization-and-policy-search":25059,"\u002Fartificial-intelligence\u002Flearning\u002Fknowledge-in-learning":25063,"\u002Fartificial-intelligence\u002Flearning\u002Fknowledge-based-learning-methods":25067,"\u002Fartificial-intelligence\u002Ffrontiers\u002Fvision-and-perception":25071,"\u002Fartificial-intelligence\u002Ffrontiers\u002Freconstructing-the-3d-world":25076,"\u002Fartificial-intelligence\u002Ffrontiers\u002Frobotics":25080,"\u002Fartificial-intelligence\u002Ffrontiers\u002Frobot-planning-and-control":25084,"\u002Fartificial-intelligence\u002Ffrontiers\u002Fnatural-language-in-ai":25088,"\u002Fartificial-intelligence\u002Ffrontiers\u002Fnlp-grammar-translation-and-speech":25092,"\u002Fartificial-intelligence\u002Ffrontiers\u002Fphilosophy-and-future":25096,"\u002Fartificial-intelligence\u002Ffrontiers\u002Fai-ethics-and-future":25100,"\u002Fartificial-intelligence":25104,"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fnuclear-constituents-nuclide-chart":25107,"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fnuclear-size-charge-distributions":25112,"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fnuclear-masses-binding-energy":25116,"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fsemi-empirical-mass-formula":25120,"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fnuclear-moments-multipoles":25124,"\u002Fnuclear-physics\u002Fnuclear-force-deuteron\u002Fnuclear-force-shell-overview":25128,"\u002Fnuclear-physics\u002Fnuclear-force-deuteron\u002Fthe-deuteron":25133,"\u002Fnuclear-physics\u002Fnuclear-force-deuteron\u002Fnucleon-nucleon-scattering":25137,"\u002Fnuclear-physics\u002Fnuclear-force-deuteron\u002Fmeson-theory-isospin":25141,"\u002Fnuclear-physics\u002Fnuclear-models\u002Ffermi-gas-model":25145,"\u002Fnuclear-physics\u002Fnuclear-models\u002Fliquid-drop-collective-coordinates":25150,"\u002Fnuclear-physics\u002Fnuclear-models\u002Fshell-model-single-particle":25154,"\u002Fnuclear-physics\u002Fnuclear-models\u002Fcollective-model-rotations-vibrations":25158,"\u002Fnuclear-physics\u002Fradioactive-decay\u002Fdecay-law-modes":25162,"\u002Fnuclear-physics\u002Fradioactive-decay\u002Fdecay-kinetics-equilibrium":25167,"\u002Fnuclear-physics\u002Falpha-decay\u002Falpha-decay-gamow-theory":25171,"\u002Fnuclear-physics\u002Falpha-decay\u002Falpha-fine-structure-hindrance":25176,"\u002Fnuclear-physics\u002Fbeta-decay\u002Fbeta-decay-energetics-neutrino":25180,"\u002Fnuclear-physics\u002Fbeta-decay\u002Ffermi-theory-beta-decay":25185,"\u002Fnuclear-physics\u002Fbeta-decay\u002Fweak-interaction-parity-violation":25189,"\u002Fnuclear-physics\u002Fbeta-decay\u002Fdouble-beta-decay-neutrino-mass":25193,"\u002Fnuclear-physics\u002Fgamma-decay\u002Fgamma-multipole-radiation":25197,"\u002Fnuclear-physics\u002Fgamma-decay\u002Finternal-conversion-isomers":25202,"\u002Fnuclear-physics\u002Fgamma-decay\u002Fangular-correlations-mossbauer":25206,"\u002Fnuclear-physics\u002Fnuclear-reactions\u002Freaction-kinematics-cross-sections":25210,"\u002Fnuclear-physics\u002Fnuclear-reactions\u002Fcompound-nucleus-resonances":25215,"\u002Fnuclear-physics\u002Fnuclear-reactions\u002Fdirect-reactions-optical-model":25219,"\u002Fnuclear-physics\u002Ffission\u002Ffission-barrier-dynamics":25223,"\u002Fnuclear-physics\u002Ffission\u002Fchain-reactions-reactor-physics":25228,"\u002Fnuclear-physics\u002Ffusion-nucleosynthesis\u002Ffusion-reactions-confinement":25232,"\u002Fnuclear-physics\u002Ffusion-nucleosynthesis\u002Fstellar-nucleosynthesis":25237,"\u002Fnuclear-physics\u002Ffusion-nucleosynthesis\u002Fbig-bang-nucleosynthesis":25241,"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fcharged-particle-stopping-power":25245,"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fphoton-neutron-interactions":25250,"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fradiation-detectors":25254,"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fdosimetry-radiation-biology":25258,"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fnuclear-applications-dating-medicine":25262,"\u002Fnuclear-physics":25266,"\u002Fnatural-language-processing\u002Ffoundations\u002Fwhat-is-nlp":25269,"\u002Fnatural-language-processing\u002Ffoundations\u002Fregex-and-text-normalization":25273,"\u002Fnatural-language-processing\u002Ffoundations\u002Fminimum-edit-distance":25277,"\u002Fnatural-language-processing\u002Ffoundations\u002Fn-gram-language-models":25281,"\u002Fnatural-language-processing\u002Ffoundations\u002Fsmoothing-and-backoff":25285,"\u002Fnatural-language-processing\u002Fclassification\u002Fnaive-bayes-and-sentiment":25289,"\u002Fnatural-language-processing\u002Fclassification\u002Fevaluating-classifiers":25294,"\u002Fnatural-language-processing\u002Fclassification\u002Flogistic-regression":25298,"\u002Fnatural-language-processing\u002Fclassification\u002Fsentiment-and-affect-lexicons":25302,"\u002Fnatural-language-processing\u002Fsemantics\u002Fvector-semantics-and-embeddings":25306,"\u002Fnatural-language-processing\u002Fsemantics\u002Fstatic-word-embeddings":25311,"\u002Fnatural-language-processing\u002Fsemantics\u002Fneural-language-models":25315,"\u002Fnatural-language-processing\u002Fsequences\u002Fsequence-labeling":25319,"\u002Fnatural-language-processing\u002Fsequences\u002Fcrfs-and-neural-taggers":25323,"\u002Fnatural-language-processing\u002Fsequences\u002Frnns-and-lstms":25327,"\u002Fnatural-language-processing\u002Ftransformers\u002Ftransformers-and-attention":25331,"\u002Fnatural-language-processing\u002Ftransformers\u002Fthe-transformer-architecture":25335,"\u002Fnatural-language-processing\u002Ftransformers\u002Flarge-language-models":25338,"\u002Fnatural-language-processing\u002Ftransformers\u002Fllm-pretraining-and-scaling":25341,"\u002Fnatural-language-processing\u002Ftransformers\u002Ffine-tuning-and-prompting":25345,"\u002Fnatural-language-processing\u002Ftransformers\u002Fprompting-and-alignment":25349,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fconstituency-parsing":25353,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fcky-scoring-and-evaluation":25358,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fdependency-parsing":25362,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fgraph-based-and-neural-dependency-parsing":25366,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fword-senses-and-wsd":25370,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fwsd-in-practice-and-induction":25374,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fsemantic-roles-and-information-extraction":25378,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Frelations-events-and-templates":25382,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fcoreference-and-discourse":25386,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fcoherence-and-discourse-structure":25390,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Flogical-semantics":25394,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fcompositional-semantics-and-description-logics":25398,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fsemantic-parsing":25402,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fneural-semantic-parsing":25406,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Finformation-extraction":25410,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Ftimes-events-and-templates":25414,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fdiscourse-coherence":25418,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fentity-based-and-global-coherence":25422,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fconstituency-grammars":25426,"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Ftreebanks-and-lexicalized-grammars":25430,"\u002Fnatural-language-processing\u002Fapplications\u002Fmachine-translation":25434,"\u002Fnatural-language-processing\u002Fapplications\u002Fmachine-translation-decoding-and-evaluation":25438,"\u002Fnatural-language-processing\u002Fapplications\u002Fquestion-answering":25442,"\u002Fnatural-language-processing\u002Fapplications\u002Fquestion-answering-knowledge-and-llms":25446,"\u002Fnatural-language-processing\u002Fapplications\u002Fdialogue-and-chatbots":25450,"\u002Fnatural-language-processing\u002Fapplications\u002Fdialogue-systems-and-assistants":25454,"\u002Fnatural-language-processing\u002Fapplications\u002Ftext-summarization":25458,"\u002Fnatural-language-processing\u002Fapplications\u002Fabstractive-summarization-and-evaluation":25462,"\u002Fnatural-language-processing\u002Fspeech\u002Fphonetics":25466,"\u002Fnatural-language-processing\u002Fspeech\u002Facoustic-phonetics":25471,"\u002Fnatural-language-processing\u002Fspeech\u002Fautomatic-speech-recognition":25475,"\u002Fnatural-language-processing\u002Fspeech\u002Fasr-evaluation-and-applications":25479,"\u002Fnatural-language-processing":25483,"\u002Fparticle-physics\u002Ffoundations\u002Fhistorical-overview-particle-zoo":25486,"\u002Fparticle-physics\u002Ffoundations\u002Fparticle-physics-basic-concepts":25490,"\u002Fparticle-physics\u002Ffoundations\u002Ffundamental-interactions-force-carriers":25494,"\u002Fparticle-physics\u002Funits-kinematics\u002Fnatural-units-and-scales":25498,"\u002Fparticle-physics\u002Funits-kinematics\u002Ffour-vectors-invariant-mass":25503,"\u002Fparticle-physics\u002Funits-kinematics\u002Fdecay-scattering-kinematics-mandelstam":25507,"\u002Fparticle-physics\u002Funits-kinematics\u002Fcross-sections-golden-rule":25511,"\u002Fparticle-physics\u002Fsymmetries\u002Fconservation-laws-symmetries":25515,"\u002Fparticle-physics\u002Fsymmetries\u002Fdiscrete-symmetries-cpt":25520,"\u002Fparticle-physics\u002Fsymmetries\u002Fparity-violation-weak":25524,"\u002Fparticle-physics\u002Fsymmetries\u002Fsu2-su3-flavor-symmetry":25528,"\u002Fparticle-physics\u002Fquark-model\u002Feightfold-way-su3":25532,"\u002Fparticle-physics\u002Fquark-model\u002Fmeson-spectroscopy":25537,"\u002Fparticle-physics\u002Fquark-model\u002Fbaryon-spectroscopy":25541,"\u002Fparticle-physics\u002Fquark-model\u002Fcolor-confinement-exotics":25545,"\u002Fparticle-physics\u002Frelativistic-wave-equations\u002Fklein-gordon-equation":25549,"\u002Fparticle-physics\u002Frelativistic-wave-equations\u002Fdirac-equation-spinors":25554,"\u002Fparticle-physics\u002Frelativistic-wave-equations\u002Fantiparticles-hole-theory":25558,"\u002Fparticle-physics\u002Fqed\u002Ffeynman-rules-qed":25562,"\u002Fparticle-physics\u002Fqed\u002Fqed-tree-processes":25567,"\u002Fparticle-physics\u002Fqed\u002Frenormalization-running-coupling":25571,"\u002Fparticle-physics\u002Fqed\u002Felectron-g-2":25575,"\u002Fparticle-physics\u002Fweak-interaction\u002Fva-structure-weak":25579,"\u002Fparticle-physics\u002Fweak-interaction\u002Fw-z-bosons-decays":25584,"\u002Fparticle-physics\u002Fweak-interaction\u002Fckm-matrix":25588,"\u002Fparticle-physics\u002Fweak-interaction\u002Fcp-violation-kaons-b-mesons":25592,"\u002Fparticle-physics\u002Fqcd\u002Fcolor-su3-gluons":25596,"\u002Fparticle-physics\u002Fqcd\u002Fasymptotic-freedom-confinement":25601,"\u002Fparticle-physics\u002Fqcd\u002Fdeep-inelastic-scattering-partons":25605,"\u002Fparticle-physics\u002Fqcd\u002Fjets-hadronization":25609,"\u002Fparticle-physics\u002Felectroweak-higgs\u002Felectroweak-su2-u1":25613,"\u002Fparticle-physics\u002Felectroweak-higgs\u002Fspontaneous-symmetry-breaking":25618,"\u002Fparticle-physics\u002Felectroweak-higgs\u002Fhiggs-mechanism":25622,"\u002Fparticle-physics\u002Felectroweak-higgs\u002Fhiggs-boson-discovery":25626,"\u002Fparticle-physics\u002Felectroweak-higgs\u002Fstandard-model":25630,"\u002Fparticle-physics\u002Fneutrinos\u002Fneutrino-oscillations":25634,"\u002Fparticle-physics\u002Fneutrinos\u002Fneutrino-mass-pmns":25639,"\u002Fparticle-physics\u002Fneutrinos\u002Fdirac-majorana-experiments":25643,"\u002Fparticle-physics\u002Fexperiment\u002Faccelerators-luminosity":25647,"\u002Fparticle-physics\u002Fexperiment\u002Fdetectors-subsystems":25652,"\u002Fparticle-physics\u002Fexperiment\u002Fhow-discoveries-are-made":25656,"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fbeyond-standard-model":25660,"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fgrand-unified-theories":25664,"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fsupersymmetry":25668,"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fhierarchy-problem-naturalness":25672,"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fdark-matter-candidates":25676,"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fmatter-antimatter-open-questions":25680,"\u002Fparticle-physics":25684,"\u002Fastrophysics-cosmology\u002Forientation\u002Fthe-sun-and-stars":25687,"\u002Fastrophysics-cosmology\u002Forientation\u002Fstellar-death-final-states":25692,"\u002Fastrophysics-cosmology\u002Forientation\u002Fgalaxies-and-cosmology":25696,"\u002Fastrophysics-cosmology\u002Fobservational-foundations\u002Fmagnitudes-fluxes-and-the-distance-modulus":25700,"\u002Fastrophysics-cosmology\u002Fobservational-foundations\u002Fstellar-spectra-and-spectral-classification":25705,"\u002Fastrophysics-cosmology\u002Fobservational-foundations\u002Ftelescopes-and-detectors-across-the-spectrum":25709,"\u002Fastrophysics-cosmology\u002Fobservational-foundations\u002Fthe-cosmic-distance-ladder":25713,"\u002Fastrophysics-cosmology\u002Fradiation-and-matter\u002Fblackbody-radiation-and-specific-intensity":25717,"\u002Fastrophysics-cosmology\u002Fradiation-and-matter\u002Fradiative-transfer-and-the-transfer-equation":25722,"\u002Fastrophysics-cosmology\u002Fradiation-and-matter\u002Fspectral-line-formation-and-broadening":25726,"\u002Fastrophysics-cosmology\u002Fradiation-and-matter\u002Fopacity-and-the-rosseland-mean":25730,"\u002Fastrophysics-cosmology\u002Fstellar-structure\u002Fhydrostatic-equilibrium-and-the-virial-theorem":25734,"\u002Fastrophysics-cosmology\u002Fstellar-structure\u002Fthe-equations-of-stellar-structure":25739,"\u002Fastrophysics-cosmology\u002Fstellar-structure\u002Fthe-equation-of-state-and-polytropes":25743,"\u002Fastrophysics-cosmology\u002Fstellar-structure\u002Fthe-standard-solar-model":25747,"\u002Fastrophysics-cosmology\u002Fnuclear-astrophysics\u002Fthermonuclear-reaction-rates-and-the-gamow-peak":25751,"\u002Fastrophysics-cosmology\u002Fnuclear-astrophysics\u002Fhydrogen-burning-pp-chains-and-cno":25756,"\u002Fastrophysics-cosmology\u002Fnuclear-astrophysics\u002Fhelium-burning-and-the-triple-alpha-process":25760,"\u002Fastrophysics-cosmology\u002Fnuclear-astrophysics\u002Fadvanced-burning-and-neutron-capture-nucleosynthesis":25764,"\u002Fastrophysics-cosmology\u002Fism-and-star-formation\u002Fphases-of-the-interstellar-medium":25768,"\u002Fastrophysics-cosmology\u002Fism-and-star-formation\u002Fmolecular-clouds-and-gravitational-collapse":25773,"\u002Fastrophysics-cosmology\u002Fism-and-star-formation\u002Fprotostars-and-the-pre-main-sequence":25777,"\u002Fastrophysics-cosmology\u002Fstellar-evolution\u002Fthe-main-sequence-and-its-structure":25781,"\u002Fastrophysics-cosmology\u002Fstellar-evolution\u002Fpost-main-sequence-low-mass-evolution":25786,"\u002Fastrophysics-cosmology\u002Fstellar-evolution\u002Fthe-evolution-of-massive-stars":25790,"\u002Fastrophysics-cosmology\u002Fstellar-evolution\u002Fstellar-pulsation-and-the-instability-strip":25794,"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fwhite-dwarfs-and-the-chandrasekhar-limit":25798,"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fcore-collapse-supernovae":25802,"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fthermonuclear-supernovae-type-ia":25806,"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fneutron-stars-and-pulsars":25810,"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fblack-holes-schwarzschild-and-kerr":25814,"\u002Fastrophysics-cosmology\u002Fbinaries-and-gravitational-waves\u002Fbinary-systems-and-mass-transfer":25818,"\u002Fastrophysics-cosmology\u002Fbinaries-and-gravitational-waves\u002Faccreting-compact-objects":25823,"\u002Fastrophysics-cosmology\u002Fbinaries-and-gravitational-waves\u002Fgravitational-waves-from-inspiraling-binaries":25827,"\u002Fastrophysics-cosmology\u002Fbinaries-and-gravitational-waves\u002Fmultimessenger-astronomy-and-gamma-ray-bursts":25831,"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Fthe-milky-way":25835,"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Fgalaxy-morphology-and-classification":25840,"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Fgalaxy-rotation-curves-and-dark-matter":25844,"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Factive-galactic-nuclei-and-supermassive-black-holes":25848,"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Fgalaxy-clusters-and-large-scale-structure":25852,"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fthe-expanding-universe-and-hubbles-law":25856,"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fthe-frw-metric-and-cosmological-redshift":25861,"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fthe-friedmann-equations-and-cosmic-dynamics":25865,"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fcosmological-models-and-distances":25868,"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fdark-energy-and-the-accelerating-universe":25872,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fthe-thermal-history-of-the-universe":25876,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fbig-bang-nucleosynthesis":25881,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Frecombination-and-the-cosmic-microwave-background":25885,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fcmb-anisotropies-and-cosmological-parameters":25889,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fcosmic-inflation":25893,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fstructure-formation-and-the-growth-of-perturbations":25897,"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fdark-matter-dark-energy-and-open-questions":25901,"\u002Fastrophysics-cosmology":25905,"\u002Fcolophon":25908,"\u002F":25911},{"path":21233,"title":21234,"module":21235,"summary":21236},"\u002Falgorithms\u002Ffoundations\u002Fwhat-is-an-algorithm","What Is an Algorithm?","Foundations","An algorithm is a finite, mechanical recipe that transforms inputs into outputs. We define what counts as an algorithm, how we write one down, and the three things we always ask of it: is it correct, is it fast, and can we prove it.\n",{"path":21238,"title":21239,"module":21235,"summary":21240},"\u002Falgorithms\u002Ffoundations\u002Fproof-techniques","Proof Techniques","An algorithm without a proof is a conjecture. This lesson collects the handful\nof arguments that certify the algorithms in this course — direct proof,\ncontrapositive, contradiction, ordinary and strong induction, construction, and\ndisproof by counterexample — each with a small worked\nexample and a picture. Loop invariants are a form of induction,\nrecursive correctness falls to strong induction, and the classic broken proofs\n(all horses are the same color) show where inductions go wrong.\n",{"path":21242,"title":21243,"module":21235,"summary":21244},"\u002Falgorithms\u002Ffoundations\u002Fasymptotic-analysis","Asymptotic Analysis","We measure an algorithm's running time as a function of its input size, then strip away machine-specific constants and lower-order terms to compare algorithms cleanly. This lesson defines the RAM model and the $O$, $\\Omega$, $\\Theta$, $o$, and $\\omega$ notations, proves the polynomial theorem, and shows how to rank growth rates with the limit test, L'Hôpital, base substitution, and the logarithm identities the arguments lean on.\n",{"path":21246,"title":21247,"module":21235,"summary":21248},"\u002Falgorithms\u002Ffoundations\u002Fgrowth-rates-and-loop-analysis","Growth Rates and Loop Analysis","With the asymptotic notations in hand, we rank the functions that actually arise in running times — from constant to factorial — proving the orderings between rungs, then read the running time of a loop nest straight off the page. Sequential blocks add, nested loops multiply, index scaling gives logarithms; a worked trace and a tour of cache-aware and galactic algorithms close the lesson.\n",{"path":21250,"title":21251,"module":21235,"summary":21252},"\u002Falgorithms\u002Ffoundations\u002Frecurrences","Recurrences and the Master Theorem","Recursive and divide-and-conquer algorithms describe their own running time with a recurrence: $T(n)$ in terms of $T$ on smaller inputs. We solve recurrences three ways — drawing the recursion tree, guessing-and-verifying by induction, and applying the Master Theorem — using merge sort as the running example, then handle unequal splits with Akra–Bazzi.\n",{"path":21254,"title":21255,"module":21235,"summary":21256},"\u002Falgorithms\u002Ffoundations\u002Famortized-analysis","Amortized Analysis","Some operations are occasionally expensive but cheap on average across any\nsequence. Amortized analysis bounds the average cost per operation over a\nworst-case sequence — not an expectation — so a rare costly step is paid for by\nthe many cheap ones around it. This lesson develops the aggregate, accounting,\nand potential methods on dynamic-array doubling, the binary counter, and a\nstack with multipop.\n",{"path":21258,"title":21259,"module":21260,"summary":21261},"\u002Falgorithms\u002Fdivide-and-conquer\u002Fmergesort","Divide and Conquer & Mergesort","Divide & Conquer","Divide and conquer breaks a problem into smaller copies of itself, solves\nthem recursively, and stitches the answers together. We meet the paradigm\nthrough mergesort — its merge step, its loop-invariant proof, and the\nrecursion tree that pins its cost at $\\Theta(n\\log n)$ — then count inversions\nwith the same machinery and distill the whole pattern into the master theorem.\n",{"path":21263,"title":21264,"module":21260,"summary":21265},"\u002Falgorithms\u002Fdivide-and-conquer\u002Fquicksort","Quicksort","Quicksort sorts in place by partitioning around a pivot and recursing on\neach side. We give Lomuto and Hoare partitioning with a correctness\ninvariant, see why a bad pivot costs $\\Theta(n^2)$ while a balanced one gives\n$\\Theta(n\\log n)$, and prove that randomizing the pivot makes the expected\ncost $\\Theta(n\\log n)$ on every input.\n",{"path":21267,"title":21268,"module":21260,"summary":21269},"\u002Falgorithms\u002Fdivide-and-conquer\u002Fselection","Linear-Time Selection","Finding the $k$-th smallest element looks like it should require sorting, but\nit does not. Quickselect adapts quicksort's partition to recurse on just one\nside, achieving expected $O(n)$. The median-of-medians algorithm guarantees a\ngood pivot with the groups-of-five trick, pushing the worst case down to a\nprovable $O(n)$.\n",{"path":21271,"title":21272,"module":21260,"summary":21273},"\u002Falgorithms\u002Fdivide-and-conquer\u002Ffast-multiplication","Fast Multiplication","Grade-school multiplication is $\\Theta(n^2)$, yet divide and conquer beats it.\nKaratsuba multiplies $n$-bit integers with three half-size products instead of\nfour, giving $\\Theta(n^{\\log_2 3})$, and Strassen multiplies matrices with\nseven block products instead of eight, giving $\\Theta(n^{\\log_2 7})$. Both\nspend cheap additions to save an expensive multiplication, and the master\ntheorem quantifies the savings.\n",{"path":21275,"title":21276,"module":21277,"summary":21278},"\u002Falgorithms\u002Fsorting\u002Fheaps-and-heapsort","Heaps and Heapsort","Sorting & Order Statistics","A binary heap is a tree we store flat in an array, with index arithmetic\nstanding in for pointers. We build the max-heap property bottom-up in $O(n)$\ntime, sort in place in $\\Theta(n\\log n)$ by repeatedly extracting the maximum,\nand reuse the same structure to implement a priority queue.\n",{"path":21280,"title":21281,"module":21277,"summary":21282},"\u002Falgorithms\u002Fsorting\u002Fsorting-lower-bounds","Lower Bounds for Comparison Sorting","Every sort we have seen runs in $\\Omega(n\\log n)$, and that is no accident.\nModeling a sort as a decision tree of comparisons, we show any such tree must\nhave $n!$ leaves, forcing height $\\ge \\log_2(n!) = \\Omega(n\\log n)$ — a bound\nno comparison sort beats in the worst case, on average, or with randomness.\n",{"path":21284,"title":21285,"module":21277,"summary":21286},"\u002Falgorithms\u002Fsorting\u002Flinear-time-sorting","Sorting in Linear Time","The $\\Omega(n\\log n)$ barrier only binds algorithms that compare. By instead\nusing keys as array indices we slip past it: counting sort runs in\n$\\Theta(n+k)$ and is stable, radix sort layers it digit by digit, and bucket\nsort averages $\\Theta(n)$ on uniform data. We see exactly when each applies.\n",{"path":21288,"title":21289,"module":21277,"summary":21290},"\u002Falgorithms\u002Fsorting\u002Fexternal-sorting","External Sorting","When the data dwarfs main memory, the cost that matters is no longer\ncomparisons but block transfers to and from disk. External merge sort sorts\nmemory-sized runs, then folds them together with a heap-driven $k$-way merge in\n$\\Theta(\\log_k(N\u002FM))$ passes. Larger fan-out cuts passes; replacement selection\nbuilds longer runs to cut them further.\n",{"path":21292,"title":21293,"module":21294,"summary":21295},"\u002Falgorithms\u002Fdata-structures\u002Felementary-structures","Elementary Data Structures","Data Structures","Every container is built one of two ways: **contiguous** in an array, or\n**linked** through pointers. We trade cache-friendly random access against\n$O(1)$ splicing, derive the **amortized $O(1)$** append of a doubling dynamic\narray, and assemble the two ordered access disciplines — the LIFO **stack** and\nthe FIFO **queue** (with its generalization, the **deque**) — on top of both.\n",{"path":21297,"title":21298,"module":21294,"summary":21299},"\u002Falgorithms\u002Fdata-structures\u002Fhash-tables","Hash Tables","A hash table implements the dictionary — insert, search, delete — in expected\n$O(1)$ time by scattering keys across an array with a hash function. We build\nup from direct addressing, handle collisions by chaining and by open\naddressing, analyze the load factor $\\alpha$, and see how universal hashing\nachieves its expected-time guarantee against every input.\n",{"path":21301,"title":21302,"module":21294,"summary":21303},"\u002Falgorithms\u002Fdata-structures\u002Fbinary-search-trees","Binary Search Trees","A binary search tree keeps keys ordered so that every operation follows a\nsingle root-to-leaf path. We state the BST property, trace search, insert,\nsuccessor, and all three delete cases on concrete trees, prove the inorder\nwalk sorts, and note the drawback — every operation costs $O(h)$, and a\ncarelessly built tree degrades to height $h = \\Theta(n)$, motivating balance.\n",{"path":21305,"title":21306,"module":21294,"summary":21307},"\u002Falgorithms\u002Fdata-structures\u002Favl-trees","AVL Trees","An AVL tree is the first balanced BST: at every node the two subtrees' heights\ndiffer by at most $1$. A Fibonacci-style minimal-node argument forces height\n$h \\le 1.44\\log_2 n = O(\\log n)$, so search, insert, and delete are all\n$O(\\log n)$. Insertion rebalances with at most one of four rotation cases\n(LL, RR, LR, RL); deletion may rotate all the way to the root.\n",{"path":21309,"title":21310,"module":21294,"summary":21311},"\u002Falgorithms\u002Fdata-structures\u002Fbalanced-trees","Balanced Search Trees","An ordinary BST can degrade to height $\\Theta(n)$; balanced search trees\nguarantee $h = O(\\log n)$ by maintaining invariants and repairing them after\nevery update. We meet rotations, the local restructuring primitive, then\nred-black trees, whose color invariants force logarithmic height, and finally\nB-trees, which trade tall-and-thin for short-and-wide to win on disk.\n",{"path":21313,"title":21314,"module":21294,"summary":21315},"\u002Falgorithms\u002Fdata-structures\u002Funion-find","Disjoint Sets (Union-Find)","The disjoint-set data structure tracks a partition of elements into groups,\nanswering \"are these two in the same group?\" and merging groups on demand. A\nforest of parent pointers, sped up by union by rank and path compression,\ndrives every operation to near-constant $O(\\alpha(n))$ amortized time — the\nstructure behind connectivity queries and Kruskal's minimum spanning tree.\n",{"path":21317,"title":21318,"module":21294,"summary":21319},"\u002Falgorithms\u002Fdata-structures\u002Ffenwick-and-segment-trees","Fenwick & Segment Trees","A prefix-sum array answers a range sum in $O(1)$ but pays $O(n)$ per update;\na plain array updates in $O(1)$ but pays $O(n)$ per range sum. Fenwick and\nsegment trees give us _both_ in $O(\\log n)$. The Fenwick (binary indexed) tree\nis a tiny array keyed by the low bit; the segment tree is a general balanced\ntree over canonical ranges that handles any associative aggregate and, with\nlazy propagation, range updates too.\n",{"path":21321,"title":21322,"module":21294,"summary":21323},"\u002Falgorithms\u002Fdata-structures\u002Fspatial-data-structures","Spatial Data Structures","A balanced BST orders keys on a line, but points in the plane have no single\nnatural order. Quadtrees subdivide space recursively into quadrants; k-d trees\nsplit on alternating coordinates at the median. Both make range and\nnearest-neighbour queries fast by carving the plane into boxes a query can\nprune away. Range trees nest a y-tree in an x-tree for fast orthogonal range\nreporting; interval trees index intervals to answer stabbing queries.\n",{"path":21325,"title":21326,"module":21294,"summary":21327},"\u002Falgorithms\u002Fdata-structures\u002Fskip-lists-and-probabilistic-structures","Skip Lists & Probabilistic Structures","Balanced trees achieve $O(\\log n)$ with rotations and invariants; randomization\ngives the same bound far more simply. A skip list is a layered linked list whose\nexpress lanes are chosen by coin flips, giving expected $O(\\log n)$ search and\ninsert with no rebalancing. A Bloom filter trades exactness for space: a bit\narray and a few hashes answer set membership with no false negatives and a\ntunable false-positive rate, but cannot delete.\n",{"path":21329,"title":21330,"module":21294,"summary":21331},"\u002Falgorithms\u002Fdata-structures\u002Fb-trees","B-Trees","When data lives on disk, the cost that dominates is block transfers, not\ncomparisons — and a binary tree of a billion keys is thirty reads deep. A\nB-tree of minimum degree $t$ is short and wide: $t-1$ to $2t-1$ keys per node,\nall leaves at one depth, height $O(\\log_t n)$. Insertion splits a full node on\nthe way down and pushes its median up; deletion borrows or merges to keep nodes\nfull enough. High fan-out is what minimizes disk I\u002FO.\n",{"path":21333,"title":21334,"module":21294,"summary":21335},"\u002Falgorithms\u002Fdata-structures\u002Fdata-stream-algorithms","Data-Stream Algorithms","Most of this course assumes data sits in fast memory, addressable at will.\nExternal sorting relaxed that to a re-readable disk. The streaming model goes\nfurther: items arrive one at a time, are seen once, and must be discarded, with\nonly sublinear, often polylogarithmic, memory. In exchange, the answers are\napproximate and probabilistic. We set up the model, then meet reservoir\nsampling for a uniform sample of an unknown-length stream and Morris counting\nfor an approximate tally in doubly-logarithmic space.\n",{"path":21337,"title":21338,"module":21294,"summary":21339},"\u002Falgorithms\u002Fdata-structures\u002Fstreaming-sketches","Streaming Sketches","Sampling and counting kept a random subset or a single approximate tally.\nSketches go further: fixed, tiny summaries that answer questions about a\nstream's frequencies. We meet the Count–Min sketch for point frequency\nestimation, Misra–Gries for heavy hitters, and HyperLogLog for distinct\ncounts, each trading a controlled error for space that never grows with the\nstream.\n",{"path":21341,"title":21342,"module":21343,"summary":21344},"\u002Falgorithms\u002Fsequences\u002Ftwo-pointers-and-windows","Two Pointers & Sliding Windows","Sequences & Strings","A family of array idioms that collapse an obvious $O(n^2)$ scan into a single\n$O(n)$ pass by maintaining an invariant as indices move. We meet two pointers\n(converging on a sorted array, and a fast\u002Fslow pair for in-place rewriting)\nand the sliding window (fixed and variable size, amortized $O(n)$). The\ncompanion lesson on prefix sums picks up where the window's positivity\nassumption fails.\n",{"path":21346,"title":21347,"module":21343,"summary":21348},"\u002Falgorithms\u002Fsequences\u002Fprefix-sums","Prefix Sums & Difference Arrays","Prefix sums precompute the running total once so that any range-sum query is a\nsingle subtraction, $P[r{+}1]-P[l]$, in $O(1)$. A hash map of prefix\nfrequencies then counts subarrays summing to $k$ in $O(n)$ — even with negative\nentries, where the sliding window fails. The difference-array dual turns $m$\nrange-adds into $O(m+n)$, and the whole idea lifts to 2-D rectangle sums by\ninclusion–exclusion.\n",{"path":21350,"title":21351,"module":21343,"summary":21352},"\u002Falgorithms\u002Fsequences\u002Fmonotonic-stacks","Monotonic Stacks & Queues","A **monotonic stack** keeps its contents sorted by popping every element that\nwould break the order before each push — turning a family of \"previous\u002Fnext\ngreater (or smaller) element\" questions into a single $O(n)$ scan. We trace\nthe next-greater-element routine push by push and prove its amortized bound,\nfuse two such scans to measure the **largest rectangle in a histogram** in\nlinear time, extend the idea to a **monotonic deque** that streams the\n**sliding-window maximum** in $O(n)$, and use asymmetric tie-breaking to\ncount **subarray minimums** without double-counting duplicates.\n",{"path":21354,"title":21355,"module":21343,"summary":21356},"\u002Falgorithms\u002Fsequences\u002Fbinary-search-on-the-answer","Binary Search on the Answer","Binary search locates the boundary of a **monotone predicate** $p(x)$ in\n$O(\\log(\\text{range}))$ probes; sorted arrays are only one instance. We first\nestablish the half-open `while (lo \u003C hi)` template for $\\textsc{lower\\_bound}$\nand $\\textsc{upper\\_bound}$, then generalize to \"binary search on the answer\":\nwhenever feasibility is monotone in a numeric parameter, we binary search the\nparameter itself, calling a feasibility check at each step.\n",{"path":21358,"title":21359,"module":21343,"summary":21360},"\u002Falgorithms\u002Fsequences\u002Fstring-matching","String Matching: Naive & Rabin–Karp","Given a text $T$ of length $n$ and a pattern $P$ of length $m$, find every\noccurrence of $P$ in $T$. The naive scan costs $O(nm)$ and re-reads text it has\nalready seen. Rabin–Karp fixes the first inefficiency with a **rolling hash**:\neach length-$m$ window is summarized by one number, updated in $O(1)$ per slide,\nverified on a hash match to kill collisions, for expected $O(n+m)$. A companion\nlesson removes the re-reading entirely with KMP and the Z-function.\n",{"path":21362,"title":21363,"module":21343,"summary":21364},"\u002Falgorithms\u002Fsequences\u002Fkmp-and-z-function","String Matching: KMP & the Z-Function","Two linear-time matchers that beat Rabin–Karp's expected bound with a\nworst-case guarantee and no randomness. KMP precomputes a **failure function**\n$\\pi$ so a mismatch slides the pattern by $q-\\pi[q-1]$ and the text pointer\nnever backs up, for $O(n+m)$. The **Z-function** computes the longest\nprefix-match at every position via the Z-box, giving the same bound from a\ndifferent angle; the two encodings of a string's self-overlap convert freely.\n",{"path":21366,"title":21367,"module":21343,"summary":21368},"\u002Falgorithms\u002Fsequences\u002Ftries","Tries & Prefix Trees","A **trie** stores a set of strings in a tree keyed by _characters_, so that\ninsert, search, delete, and prefix-test all run in $O(L)$ time — the length\nof the key, _independent of how many keys are stored_. Shared prefixes are\nstored once, which makes tries the natural structure for autocomplete,\nwildcard dictionaries, board word-search, and — over the alphabet $\\{0,1\\}$\n— the maximum-XOR-pair problem. Radix (Patricia) trees compress the chains.\n",{"path":21370,"title":21371,"module":21343,"summary":21372},"\u002Falgorithms\u002Fsequences\u002Fsuffix-arrays-and-aho-corasick","Suffix Arrays, LCP & Aho–Corasick","A **suffix array** sorts all $n$ suffixes of a string, indexing every substring\nat once; built in $O(n\\log n)$, it locates a pattern by binary search in\n$O(m\\log n)$. Its companion **LCP array** (Kasai's $O(n)$ algorithm) counts\ndistinct substrings and finds the longest repeated substring. **Aho–Corasick**\ngeneralises KMP to a whole dictionary: a trie of patterns plus failure links\nscans the text once in $O(\\text{text} + \\text{matches})$ to report every\noccurrence of every pattern. Manacher's algorithm finds all palindromic\nsubstrings in $O(n)$.\n",{"path":21374,"title":21375,"module":21376,"summary":21377},"\u002Falgorithms\u002Fgraphs\u002Frepresentations-and-traversal","Graph Representations and Traversal","Graphs","A graph captures _relationships_ — who connects to whom. We fix the\nvocabulary, weigh the two standard representations (adjacency list versus\nmatrix), then meet the single search skeleton behind everything that follows:\nWhatever-First-Search, and its breadth-first reading, which finds shortest\npaths by number of edges in $O(V + E)$.\n",{"path":21379,"title":21380,"module":21376,"summary":21381},"\u002Falgorithms\u002Fgraphs\u002Fdepth-first-search","Depth-First Search","Swap BFS's queue for a stack and the search plunges instead of fanning out.\nDepth-first search stamps every vertex with discovery and finish times that\nnest like parentheses, classifies each edge as tree, back, forward, or cross,\nand — through the back edge — decides in one pass whether a graph has a cycle.\nThese timestamps underpin topological sort, strong\nconnectivity, and the rest of this module.\n",{"path":21383,"title":21384,"module":21376,"summary":21385},"\u002Falgorithms\u002Fgraphs\u002Ftopological-sort-and-scc","Topological Sort and Strong Connectivity","Directed acyclic graphs model dependencies: tasks that must precede other\ntasks. A _topological order_ lays such a graph out in a line so every edge\npoints forward, and depth-first finish times yield one almost for free.\nWe then ask the harder question for graphs _with_ cycles: which vertices can\nreach each other? The answer is the strongly connected components, found by a\ntwo-pass DFS.\n",{"path":21387,"title":21388,"module":21376,"summary":21389},"\u002Falgorithms\u002Fgraphs\u002Fminimum-spanning-trees","Minimum Spanning Trees","Given a weighted network, how do we connect everything as cheaply as possible?\nThe answer is a minimum spanning tree, and one lemma — the cut property —\njustifies _every_ correct MST algorithm. We prove the cut and cycle\nproperties by exchange arguments, use them to settle uniqueness, and meet the\noldest MST algorithm, Borůvka's, whose parallel component-merging rounds fall\nstraight out of the cut rule.\n",{"path":21391,"title":21392,"module":21376,"summary":21393},"\u002Falgorithms\u002Fgraphs\u002Fkruskal-and-prim","Kruskal and Prim","The two minimum-spanning-tree algorithms you will actually implement.\nKruskal grows a forest edge by edge, cheapest first, using a union-find\nstructure to reject cycle-closing edges; Prim grows one tree outward from a\nroot with a priority queue, exactly Dijkstra rekeyed by attachment cost. Both\ntraced in full on a nine-town graph, with the edge cases, the bottleneck\nproperty, and where each one wins.\n",{"path":21395,"title":21396,"module":21376,"summary":21397},"\u002Falgorithms\u002Fgraphs\u002Fshortest-paths","Shortest Paths","Finding the cheapest route through a weighted network is one of the most-used\nalgorithms in computing, and a single operation — _relaxation_ — underlies\nevery method. We build the primitive, prove the triangle inequality and\noptimal substructure that make it work, then meet Dijkstra's algorithm: the\ngreedy solution for non-negative weights, traced vertex by vertex, with the\ncut argument that proves each extraction is final.\n",{"path":21399,"title":21400,"module":21376,"summary":21401},"\u002Falgorithms\u002Fgraphs\u002Fall-pairs-and-negative-weights","All-Pairs and Negative Weights","Dijkstra's greedy schedule breaks the moment an edge goes negative. We give it\nup for dynamic programming: Bellman-Ford derived as a DP over edge budgets,\nwith its negative-cycle detector, and Floyd-Warshall computing the distance\nbetween _every_ pair of vertices via a DP over which vertices a path may pass\nthrough. We close with Johnson's algorithm and the arbitrage problems that\nnegative cycles encode.\n",{"path":21403,"title":21404,"module":21376,"summary":21405},"\u002Falgorithms\u002Fgraphs\u002Fnetwork-flow","Network Flow","How much can flow through a network from source to sink? We build flow\nnetworks with capacity and conservation constraints, increase a flow by\npushing along augmenting paths in the residual graph, and see how reverse\nedges let the algorithm undo earlier routing. Ford-Fulkerson and its BFS refinement\nEdmonds-Karp find a maximum flow, traced end to end on a worked network.\n",{"path":21407,"title":21408,"module":21376,"summary":21409},"\u002Falgorithms\u002Fgraphs\u002Fmax-flow-min-cut","Max-Flow Min-Cut and Applications","Why is the flow found when no augmenting path remains actually optimal? The\nanswer is a duality theorem: the maximum flow equals the minimum cut. We prove\nit, read the minimum cut off the final residual graph, then derive bipartite\nmatching and a catalog of modeling reductions from the flow\nabstraction — before touching the modern algorithms that supersede\nEdmonds-Karp.\n",{"path":21411,"title":21412,"module":21376,"summary":21413},"\u002Falgorithms\u002Fgraphs\u002Fbridges-and-articulation-points","Bridges & Articulation Points","A **bridge** is an edge whose removal disconnects the graph; an **articulation\npoint** is a vertex whose removal does. Both are single points of failure in a\nnetwork. A single depth-first search computes discovery times and **low-links**,\nand two local criteria — $low[v] > disc[u]$ for bridges, $low[v] \\ge disc[u]$\nfor cut vertices — find them all in $O(V+E)$.\n",{"path":21415,"title":21416,"module":21376,"summary":21417},"\u002Falgorithms\u002Fgraphs\u002Flowest-common-ancestor","Lowest Common Ancestor & Binary Lifting","Given a rooted tree, the lowest common ancestor of $u$ and $v$ is the deepest\nnode that is an ancestor of both. A naive walk answers one query in $O(h)$;\n**binary lifting** precomputes the $2^k$-th ancestor of every node in\n$O(n\\log n)$, then answers $k$-th-ancestor and LCA queries in $O(\\log n)$ each.\nWe derive both jumps, apply them to tree distance, and compare against the\nEuler-tour + RMQ and Tarjan offline alternatives.\n",{"path":21419,"title":21420,"module":21376,"summary":21421},"\u002Falgorithms\u002Fgraphs\u002Ftwo-sat","2-SAT via Implication Graphs","A boolean formula whose every clause has exactly two literals can be solved in\n_linear_ time — even though its three-literal cousin is NP-complete. The idea\nis to read each clause as a pair of implications, build a directed graph on the\n$2n$ literals, and ask a question we already know how to answer: which literals\nshare a strongly connected component? The formula is satisfiable iff no variable\nlands in the same SCC as its own negation, and the SCCs' topological order\nyields a satisfying assignment for free.\n",{"path":21423,"title":21424,"module":21376,"summary":21425},"\u002Falgorithms\u002Fgraphs\u002Feulerian-tours","Eulerian Tours","An **Eulerian tour** uses every _edge_ of a graph exactly once. We give the\nexact parity and balance conditions under which one exists (even degree\nfor undirected graphs, in-degree equal to out-degree for directed) and Hierholzer's\n$O(E)$ algorithm that constructs one by splicing closed sub-tours. We contrast\nthis sharply with the **Hamiltonian** problem (visit every _vertex_ once),\nwhich is NP-complete: visiting edges is easy, visiting vertices is hard.\n",{"path":21427,"title":21428,"module":21376,"summary":21429},"\u002Falgorithms\u002Fgraphs\u002Fbipartite-matching","Bipartite Matching","Pairing applicants to jobs, students to slots, files to disks: all are\n**maximum bipartite matching**. We solve it combinatorially with **augmenting\npaths** (Kuhn's algorithm, $O(VE)$), speed it up to $O(E\\sqrt V)$ with\n**Hopcroft–Karp**, and uncover the structure behind it — **König's theorem**\n(max matching equals min vertex cover) and **Hall's marriage theorem** (a\nperfect matching exists iff every set has enough neighbors).\n",{"path":21431,"title":21432,"module":21433,"summary":21434},"\u002Falgorithms\u002Fgreedy\u002Fthe-greedy-method","The Greedy Method","Greedy Algorithms","A greedy algorithm builds a solution one locally-best choice at a time and\nnever looks back. We isolate the two properties that make this work — the\ngreedy-choice property and optimal substructure — prove the canonical\nactivity-selection algorithm correct with an exchange argument, watch greedy\nfail on the 0\u002F1 knapsack, and glimpse matroids as the theory\nthat says exactly when the greedy method is optimal.\n",{"path":21436,"title":21437,"module":21433,"summary":21438},"\u002Falgorithms\u002Fgreedy\u002Fscheduling-and-intervals","Scheduling & Interval Partitioning","Three classic scheduling problems all yield to greedy algorithms — and all\nthree turn on a single design decision: which key to sort by. Interval\nscheduling sorts by **finish** time to pack the most compatible jobs;\ninterval partitioning sorts by **start** time and proves the rooms needed\nequal the maximum overlap **depth**; minimizing maximum lateness sorts by\n**deadline** and is justified by an adjacent-swap exchange argument.\n",{"path":21440,"title":21441,"module":21433,"summary":21442},"\u002Falgorithms\u002Fgreedy\u002Fhuffman-codes","Huffman Codes","Huffman coding builds a\nprovably optimal prefix-free binary code by repeatedly merging the two least\nfrequent symbols. We develop prefix-free codes as binary trees, give the\nalgorithm with a priority queue, build a Huffman tree from example\nfrequencies, prove optimality with the same greedy-choice-plus-substructure\nargument, and pin the running time at $O(n\\log n)$.\n",{"path":21444,"title":21445,"module":21433,"summary":21446},"\u002Falgorithms\u002Fgreedy\u002Fmatroids","Matroids & Exchange Arguments","The capstone of the greedy module: _why_ and _when_ a greedy algorithm is\nprovably optimal. We recap the two correctness templates — **greedy-stays-ahead**\nand the **exchange argument** — then meet the **matroid** $M=(S,\\mathcal{I})$, an\nabstraction whose **exchange property** is the structure greedy needs.\nThe matroid–greedy theorem says sorting by weight and taking what stays\nindependent yields a maximum-weight basis _if and only if_ the structure is a\nmatroid. Kruskal's MST is the canonical instance; 0\u002F1 knapsack and TSP are the\ncanonical failures.\n",{"path":21448,"title":21449,"module":21433,"summary":21450},"\u002Falgorithms\u002Fgreedy\u002Fstable-matching","Stable Matching (Gale–Shapley)","Two sides each rank the other; we want a matching with no **blocking pair** — no\ntwo participants who both prefer each other to their assigned partners. The\n**Gale–Shapley deferred-acceptance** algorithm has proposers propose in\npreference order while receivers tentatively hold the best offer so far. We prove\nit terminates in $\\O(n^2)$ proposals, returns a **perfect** matching, and that\nthe matching is **stable**. A sharper asymmetry follows: deferred acceptance is\n**proposer-optimal** and **receiver-pessimal**, the structural fact behind the\nresidency match and school-choice systems.\n",{"path":21452,"title":21453,"module":21454,"summary":21455},"\u002Falgorithms\u002Fdynamic-programming\u002Fprinciples","Principles of Dynamic Programming","Dynamic Programming","Dynamic programming is recursion with memory: when a recursive solution\nre-solves the same subproblems again and again, we solve each one once and\nstore the answer. We identify the two structural conditions that make this\nwork — overlapping subproblems and optimal substructure — contrast top-down\nmemoization with bottom-up tabulation, and distil the whole method into a\nfive-step recipe.\n",{"path":21457,"title":21458,"module":21454,"summary":21459},"\u002Falgorithms\u002Fdynamic-programming\u002Fsequence-dp","Sequence Alignment & LCS","Two strings can be compared by how much of one appears inside the\nother. The longest common subsequence (LCS) and edit distance are the two\nclassic measures, and they are the _same_ dynamic program with different\ncosts. We derive the LCS recurrence by examining the last characters, fill a\nworked DP table, reconstruct the subsequence, and then show edit distance as\nthe identical $\\Theta(mn)$ pattern.\n",{"path":21461,"title":21462,"module":21454,"summary":21463},"\u002Falgorithms\u002Fdynamic-programming\u002Flongest-increasing-subsequence","Longest Increasing Subsequence","Given a sequence of numbers, how long is its longest strictly increasing\nsubsequence? A first dynamic program indexes subproblems by the element each\nsubsequence _ends at_, giving an $O(n^2)$ solution with parent-pointer\nreconstruction. A sharper idea, the patience-sorting _tails_ array searched by\nbinary search, drops the time to $O(n\\log n)$. We then fold in the\nvariants: non-decreasing, counting, Russian-doll envelopes, and bitonic.\n",{"path":21465,"title":21466,"module":21454,"summary":21467},"\u002Falgorithms\u002Fdynamic-programming\u002Fknapsack","Knapsack & Subset Problems","We start from $\\textsc{Subset-sum}$ — does some sublist hit a target $t$? — and its\ninclude\u002Fexclude recurrence over a boolean table $A(i, u)$, then bolt on values\nto get 0\u002F1 knapsack as the same machine with $\\lor$ promoted to $\\max$. We fill\nboth tables, recover the chosen items, and confront the surprise that the\n$\\Theta(nt)$ running time is only _pseudo-polynomial_ — exponential in the bit\nlength $b$, and unimprovable unless $\\mathrm{P}=\\mathrm{NP}$ since subset-sum is\n$\\textsc{NP-complete}$. The fractional variant reveals the sharp line between greedy\nand dynamic programming.\n",{"path":21469,"title":21470,"module":21454,"summary":21471},"\u002Falgorithms\u002Fdynamic-programming\u002Fcoin-change-and-unbounded","Coin Change & Unbounded Knapsack","The previous lesson let each item be taken at most once. Drop that cap — items\nmay be reused _any number of times_ — and the 0\u002F1 knapsack collapses from a\ntwo-dimensional table to a one-dimensional one, because there is no longer a\nprefix of \"already-used\" items to track. We meet **unbounded knapsack**, then\nits most famous instance, **coin change**: the minimum-coins recurrence\n$C[a] = 1 + \\min_c C[a-c]$, and the counting variant where the _order of the\nloops_ decides whether you count unordered combinations or ordered sequences —\nthe classic bug. Greed fails in general but works for canonical coin systems.\n",{"path":21473,"title":21474,"module":21454,"summary":21475},"\u002Falgorithms\u002Fdynamic-programming\u002Finterval-dp","Interval DP","Many problems ask for the best way to combine a contiguous range of items, and\nthe answer is a dynamic program over subintervals $[i,j]$ that chooses a split\npoint $k$. We derive the pattern from matrix-chain multiplication —\nparenthesising a product to minimize scalar multiplications in $O(n^3)$ — distil\nit into a reusable template filled by increasing interval length, and then meet\nits sharpest variant: the \"last operation\" trick behind Burst Balloons and\ncutting a stick, where fixing the _last_ move (not the first) makes the two\nsides independent.\n",{"path":21477,"title":21478,"module":21454,"summary":21479},"\u002Falgorithms\u002Fdynamic-programming\u002Ftree-dp","Dynamic Programming on Trees","When the subproblems of a dynamic program are _rooted subtrees_, a single\npost-order DFS solves the whole thing in $O(n)$: each node combines the\nalready-computed answers of its children. We meet the archetype — maximum-weight\nindependent set on a tree — then the \"path through a node\" pattern behind tree\ndiameter and maximum path sum, and finally **rerooting**, which computes a\nper-node answer for _every_ node as root in $O(n)$ with two passes.\n",{"path":21481,"title":21482,"module":21454,"summary":21483},"\u002Falgorithms\u002Fdynamic-programming\u002Fbitmask-dp","Bitmask DP","When a subproblem depends not on an index or a prefix but on _which subset_ of\na small ground set has been used, we can encode that subset as the bits of an\ninteger and index a DP table by it. With $n \\le \\sim 20$ the $2^n$ subsets fit\nin a table, turning $\\Theta(n!)$ brute force into $O(2^n \\cdot \\text{poly}(n))$.\nWe meet the bit tricks, the Held–Karp TSP archetype, assignment by mask,\nsubset-sum partitioning, and submask enumeration with its $3^n$ bound.\n",{"path":21485,"title":21486,"module":21454,"summary":21487},"\u002Falgorithms\u002Fdynamic-programming\u002Fdp-optimizations","DP Optimizations","A correct DP recurrence is only half the battle; its naive evaluation is often\na factor of $n$ slower than necessary. This capstone surveys five techniques,\nmonotonic-queue, the convex hull trick, divide-and-conquer optimization,\nKnuth's optimization, and SOS DP, that each exploit _structure in the\ntransition_ (a sliding window, linear costs, monotone optimal splits, the\nquadrangle inequality, or subset lattices) to shave an $O(n)$, $O(\\log n)$, or\nworse factor off the running time.\n",{"path":21489,"title":21490,"module":21454,"summary":21491},"\u002Falgorithms\u002Fdynamic-programming\u002Fdp-on-graphs","Dynamic Programming on Graphs","Many graph algorithms are dynamic programs: the subproblem is the\n_best value reachable under a restricted resource_ — intermediate vertices\nallowed, edges allowed, or a topological prefix — and edge _relaxation_ is the\nDP transition. We frame Floyd–Warshall as the archetype ($O(V^3)$ all-pairs\nshortest paths), Bellman–Ford as a DP over path length (the at-most-$K$-stops\nvariant), DAG-DP in topological order ($O(V+E)$), and Warshall's transitive\nclosure as the boolean analog.\n",{"path":21493,"title":21494,"module":21454,"summary":21495},"\u002Falgorithms\u002Fdynamic-programming\u002Fdigit-and-probability-dp","Digit & Probability DP","Two DP patterns with unusual state. _Digit DP_ counts the\nintegers in a range $[L, R]$ that satisfy a digit constraint by walking the\ndecimal places of the bound, carrying a _tight_ flag that marks when the prefix\nstill equals the bound's. _Probability\u002FExpectation DP_ replaces \"best value\" with\n\"expected value,\" using linearity of expectation to make each state an\naverage over its weighted transitions — the natural tool for expected step\ncounts and absorbing Markov chains.\n",{"path":21497,"title":21498,"module":21499,"summary":21500},"\u002Falgorithms\u002Fbacktracking\u002Fbacktracking-fundamentals","Backtracking: Subsets, Permutations & Combinations","Backtracking & Search","Backtracking builds a solution one choice at a time and abandons a partial\nsolution the moment it cannot be completed, exploring a state-space tree by\ndepth-first search. We meet the universal choose\u002Fexplore\u002Fun-choose template,\nderive the canonical enumerations — subsets ($2^n$), permutations ($n!$), and\ncombinations ($\\binom{n}{k}$) — handle duplicate elements by skipping equal\nsiblings, and see how pruning turns an exponential search into a tractable one.\n",{"path":21502,"title":21503,"module":21499,"summary":21504},"\u002Falgorithms\u002Fbacktracking\u002Fconstraint-search","Constraint Search: N-Queens & Sudoku","Many hard puzzles are **constraint satisfaction problems**: assign each\nvariable a value from its domain so that every constraint holds. Backtracking\nsolves them by assigning variables one at a time and rejecting a partial\nassignment the instant a constraint breaks. We make the rejection cheap — $O(1)$\nconflict checks for N-Queens via column and diagonal sets — and prune harder\nwith **forward checking**, **MRV** ordering, and **constraint propagation**,\nwhich is what lets an exponential search actually finish.\n",{"path":21506,"title":21507,"module":21499,"summary":21508},"\u002Falgorithms\u002Fbacktracking\u002Fbranch-and-bound","Branch & Bound and Meet in the Middle","Plain backtracking prunes a search tree by _feasibility_; for _optimization_\nproblems we can prune far more aggressively by _value_. **Branch and bound**\nkeeps the best complete solution found so far and discards any partial solution\nwhose optimistic bound cannot beat it. **Meet in the middle** splits the\ninstance in two, enumerates each half, and recombines by binary search — turning\n$2^n$ into $O(2^{n\u002F2}\\,n)$ and pushing exact search out to $n \\approx 40$.\n",{"path":21510,"title":21511,"module":21499,"summary":21512},"\u002Falgorithms\u002Fbacktracking\u002Fgraph-backtracking","Graph Backtracking: m-Coloring & Hamiltonian Paths","Two famous graph problems have no known efficient algorithm, yet yield cleanly\nto backtracking with the right pruning. **Graph $m$-coloring** assigns one of\n$m$ colors to each vertex so no edge is monochromatic; we color vertices in turn\nand reject a color the instant a neighbor already has it. **Hamiltonian\npath\u002Fcycle** asks for a walk visiting every vertex exactly once; we extend a path\ngreedily and backtrack on dead ends. Both are NP-complete, so the worst case is\nexponential — but feasibility pruning and good vertex ordering make real\ninstances tractable, and the contrast with the easy Eulerian condition shows why.\n",{"path":21514,"title":21515,"module":21516,"summary":21517},"\u002Falgorithms\u002Fmathematical-algorithms\u002Fnumber-theory-basics","Number Theory: GCD & Modular Arithmetic","Mathematical Algorithms","This lesson opens the mathematical-algorithms module with the bedrock of\ncomputational number theory. We prove Euclid's recurrence\n$\\gcd(a,b)=\\gcd(b,\\,a\\bmod b)$ and its $O(\\log\\min(a,b))$ running time, extend\nit to recover Bézout coefficients $x,y$ with $ax+by=\\gcd(a,b)$, and build\nmodular arithmetic on residue classes — including when a modular inverse\n$a^{-1}\\bmod m$ exists and how to compute it.\n",{"path":21519,"title":21520,"module":21516,"summary":21521},"\u002Falgorithms\u002Fmathematical-algorithms\u002Fmodular-exponentiation-and-primality","Modular Exponentiation & Primality","Computing $a^n \\bmod m$ naively costs $n$ multiplications; **repeated squaring**\ndoes it in $O(\\log n)$ by reading the bits of the exponent. We use this routine\nto state **Fermat's little theorem** (and the modular inverse it gives), then to\ntest primality — trial division, the probabilistic **Fermat** and **Miller–Rabin**\ntests, and the deterministic witness set that settles primality for every 64-bit\nnumber.\n",{"path":21523,"title":21524,"module":21516,"summary":21525},"\u002Falgorithms\u002Fmathematical-algorithms\u002Fsieve-and-factorization","Sieves & Factorization","The previous lesson tested one number for primality; here we ask for _all_\nprimes up to $n$ at once. The **sieve of Eratosthenes** cross-cuts composites\nin $O(n\\log\\log n)$, and a **linear sieve** does it in $O(n)$ while recording\neach number's **smallest prime factor**, which then factors any $x \\le n$ in\n$O(\\log x)$. From a factorization $x = \\prod p_i^{e_i}$ the multiplicative\nfunctions $\\tau$, $\\sigma$, and Euler's totient $\\varphi$ fall out immediately.\n",{"path":21527,"title":21528,"module":21516,"summary":21529},"\u002Falgorithms\u002Fmathematical-algorithms\u002Fcombinatorics","Combinatorics & Counting","Counting is the arithmetic of finite sets. We build up from permutations\n$n!$ and combinations $\\binom{n}{k}$, prove Pascal's rule by a bijection,\nand count multisets with stars and bars. The practical core is computing\n$\\binom{n}{k}\\bmod p$ in $O(1)$ from precomputed factorials and inverse\nfactorials. We close with inclusion–exclusion and the Chinese Remainder\nTheorem, both of which lean on the modular inverse from the previous lesson.\n",{"path":21531,"title":21532,"module":21516,"summary":21533},"\u002Falgorithms\u002Fmathematical-algorithms\u002Fmatrix-exponentiation","Matrix Exponentiation","A linear recurrence advances by a fixed linear rule, so one step is a\n**matrix–vector** product and $n$ steps are a **matrix power**. Packaging\nFibonacci, and any $k$-term recurrence, into a transition matrix lets us jump\nto the $n$-th term in $O(k^3 \\log n)$ by **exponentiation by squaring** — the\nsame doubling trick from modular exponentiation, now over matrices.\n",{"path":21535,"title":21536,"module":21516,"summary":21537},"\u002Falgorithms\u002Fmathematical-algorithms\u002Ffast-fourier-transform","Fast Fourier Transform","Multiplying two degree-$n$ polynomials by the schoolbook method costs\n$\\Theta(n^2)$. Evaluating them at the **$n$-th roots of unity** turns\nmultiplication into pointwise products, and the **Cooley–Tukey FFT** computes\nall those evaluations in $\\Theta(n\\log n)$ by splitting even and odd\ncoefficients. The inverse FFT interpolates back, giving $\\Theta(n\\log n)$\npolynomial and big-integer multiplication.\n",{"path":21539,"title":21540,"module":21516,"summary":21541},"\u002Falgorithms\u002Fmathematical-algorithms\u002Fgradient-descent","Numerical Optimization and Gradient Descent","Most of this course chases **discrete** optima over finite structures; here the\nsearch space is **continuous** and the objective $f$ is differentiable. The\n**gradient** points uphill, so stepping against it —\n$x_{t+1} = x_t - \\eta\\,\\nabla f(x_t)$ — walks downhill. **Convexity** makes every\nlocal minimum global; for convex $L$-smooth $f$ gradient descent converges at\n$O(1\u002Ft)$, and **geometrically** under strong convexity. **Newton's method** uses\nthe Hessian for local quadratic convergence, and **bisection** is the robust\nbracketing fallback for roots.\n",{"path":21543,"title":21544,"module":21545,"summary":21546},"\u002Falgorithms\u002Fcomputational-geometry\u002Fgeometric-primitives","Geometric Primitives & Orientation","Computational Geometry","Computational geometry is built on a single reliable primitive — the\n**orientation test**, a sign of a cross product that tells whether three points\nturn left, right, or lie collinear. From points-as-vectors and the dot and\ncross products we derive orientation, segment intersection, the shoelace area\nformula, and point-in-polygon tests, keeping all arithmetic **exact and\ninteger** so that no floating-point rounding can corrupt a sign.\n",{"path":21548,"title":21549,"module":21545,"summary":21550},"\u002Falgorithms\u002Fcomputational-geometry\u002Fconvex-hull","Convex Hull","The convex hull is the smallest convex polygon enclosing a point set — the\nrubber band snapped around the nails. We build it with Andrew's monotone chain,\nsorting by $(x,y)$ and sweeping a lower and upper hull while popping any\nnon-left turn via the orientation primitive, in $O(n\\log n)$. A reduction from\nsorting shows that bound is optimal, and the hull yields diameter, smallest\nenclosing rectangle, and more through rotating calipers.\n",{"path":21552,"title":21553,"module":21545,"summary":21554},"\u002Falgorithms\u002Fcomputational-geometry\u002Fsweep-line","Sweep-Line Algorithms","The plane-sweep paradigm turns a static $2$-D geometry problem into a dynamic\n$1$-D ordered-set problem: a vertical line sweeps left to right, stopping at an\n$x$-sorted **event queue** while a balanced-BST **status structure** tracks the\nobjects it currently crosses, ordered by $y$. We derive Bentley–Ottmann segment\nintersection in $O((n+k)\\log n)$, recover closest-pair in $O(n\\log n)$, and\nreduce skyline, rectangle-area, and overlap problems to $\\pm1$ event sweeps.\n",{"path":21556,"title":21557,"module":21545,"summary":21558},"\u002Falgorithms\u002Fcomputational-geometry\u002Fpolygons-and-proximity","Polygons & Proximity","Four classics that live on top of the orientation primitive and the convex\nhull. **Closest pair** falls to divide-and-conquer in $\\Theta(n\\log n)$, where a\npacking argument caps the cross-boundary combine at seven neighbours per point.\n**Point-in-polygon** is the ray-casting parity test or the winding-number count\nthat also handles self-intersecting boundaries, both with their edge caveats. The **shoelace formula**\ngives signed area as a sum of cross products, and **rotating calipers** walk the\nhull to read off diameter and width in $O(n)$.\n",{"path":21560,"title":21561,"module":21562,"summary":21563},"\u002Falgorithms\u002Fintractability\u002Fp-np-reductions","P, NP, and Reductions","Intractability","Most problems we have met so far have fast algorithms. A vast and important\nfamily seemingly does not. This lesson builds the vocabulary for that\ndivide: decision problems, the class $\\mathsf{P}$ of problems we can solve\nquickly, the class $\\mathsf{NP}$ of problems whose solutions we can _check_\nquickly, and polynomial-time reductions, the tool that lets us compare the\ndifficulty of two problems without solving either.\n",{"path":21565,"title":21566,"module":21562,"summary":21567},"\u002Falgorithms\u002Fintractability\u002Fnp-completeness","NP-Completeness","Some problems in $\\mathsf{NP}$ are universally hardest: every other problem\nin $\\mathsf{NP}$ reduces to them. This lesson defines $\\mathsf{NP}$-hard and\n$\\mathsf{NP}$-complete, states the Cook–Levin theorem that anchors the\ntheory on **SAT**, walks the web of reductions that grows from it, and gives\nthe four-step recipe for proving a brand-new problem $\\mathsf{NP}$-complete.\n",{"path":21569,"title":21570,"module":21562,"summary":21571},"\u002Falgorithms\u002Fintractability\u002Fcoping-with-hardness","Coping with NP-Hardness","An $\\mathsf{NP}$-hardness proof rules out an exact polynomial-time algorithm,\nnot the need for answers. This lesson surveys four practical responses to\nhardness: approximation algorithms with a provable ratio (worked through a\n2-approximation for vertex cover), heuristics and local search, exact\nexponential methods like branch and bound, and exploiting special structure\nin the instances you actually face.\n",{"path":21573,"title":21574,"module":21562,"summary":21575},"\u002Falgorithms\u002Fintractability\u002Fapproximation-algorithms","Approximation Algorithms","When a problem is $\\mathsf{NP}$-hard we can still ask for a solution\nprovably close to optimal. This lesson makes the approximation ratio\n$\\rho$ precise, separates absolute from relative guarantees, and proves the\nratios of four classic algorithms: greedy set cover ($H_n \\approx \\ln n$),\nthe MST-doubling $2$-approximation for metric TSP, load balancing, and the\nknapsack FPTAS. It closes with the hierarchy PTAS \u002F FPTAS and the limits of\ninapproximability.\n",{"path":21577,"title":21578,"module":6,"summary":6},"\u002Falgorithms","Algorithms",{"path":21580,"title":21581,"module":21582,"summary":21583},"\u002Fcalculus\u002Flimits-and-continuity\u002Ffunctions-and-models","Functions and Mathematical Models","Limits and Continuity","A function assigns exactly one output to each input and can be presented four ways: verbally, numerically, graphically, or by a formula. The elementary families — linear, polynomial, power, rational, trigonometric, exponential — model most elementary phenomena, and transformation, combination, and composition build every other function from them.\n",{"path":21585,"title":21586,"module":21582,"summary":21587},"\u002Fcalculus\u002Flimits-and-continuity\u002Fthe-limit-of-a-function","The Limit of a Function","The tangent and velocity problems both ask for a value a ratio approaches but never reaches — the limit. Its intuitive two-sided form splits into one-sided limits that must agree; a limit fails to exist when they disagree or when the function grows without bound, the latter producing a vertical asymptote.\n",{"path":21589,"title":21590,"module":21582,"summary":21591},"\u002Fcalculus\u002Flimits-and-continuity\u002Flimit-laws-and-the-precise-definition","Limit Laws and the ε–δ Definition","The Limit Laws reduce a limit to arithmetic on simpler limits, and direct substitution settles polynomials and rational functions outright. The 0\u002F0 forms that resist substitution yield to algebra or the Squeeze Theorem, and the ε–δ definition makes \"arbitrarily close\" precise as a pair of quantified inequalities.\n",{"path":21593,"title":21594,"module":21582,"summary":21595},"\u002Fcalculus\u002Flimits-and-continuity\u002Fcontinuity","Continuity","A function is continuous at a point when its limit there equals its value, so the graph has no break. Continuity fails in three geometric ways; it is closed under arithmetic and composition, so the elementary families and their combinations are continuous; and on a closed interval it forces the Intermediate Value Theorem, which locates roots.\n",{"path":21597,"title":21598,"module":21599,"summary":21600},"\u002Fcalculus\u002Fderivatives\u002Fthe-derivative-and-rates-of-change","The Derivative and Rates of Change","Derivatives","A single limit with three readings: the slope of the tangent line, the instantaneous velocity of a moving object, and the rate of change of one quantity with respect to another. Built from the difference quotient, extended from a value at one point to a function of x, and undefined exactly where a corner, jump, or vertical tangent appears.\n",{"path":21602,"title":21603,"module":21599,"summary":21604},"\u002Fcalculus\u002Fderivatives\u002Fdifferentiation-rules-and-the-chain-rule","Differentiation Rules and the Chain Rule","Computing every derivative from the limit definition is tedious. A short list of rules — power, constant multiple, sum, product, quotient — differentiates any polynomial or rational function by inspection. The trigonometric derivatives follow from one limit, and the chain rule extends everything to composite functions by multiplying rates along the composition.\n",{"path":21606,"title":21607,"module":21599,"summary":21608},"\u002Fcalculus\u002Fderivatives\u002Fimplicit-differentiation-and-related-rates","Implicit Differentiation and Related Rates","Not every curve is the graph of y = f(x). Implicit differentiation finds a slope from an equation in x and y directly, treating y as an unknown function and differentiating both sides. The same chain-rule idea drives related rates, where one measured rate of change forces another through a geometric constraint, and interprets the derivative as a rate across the sciences.\n",{"path":21610,"title":21611,"module":21599,"summary":21612},"\u002Fcalculus\u002Fderivatives\u002Flinear-approximations-and-differentials","Linear Approximations and Differentials","A differentiable curve looks like its tangent line under enough magnification, so the tangent is a usable stand-in for the function near the point of contact. The linear approximation and its linearization, written in the language of differentials dy and dx, estimate both function values and the measurement error propagated into a computed quantity.\n",{"path":21614,"title":21615,"module":21616,"summary":21617},"\u002Fcalculus\u002Fapplications-of-derivatives\u002Fextrema-and-the-mean-value-theorem","Extrema and the Mean Value Theorem","Applications of Derivatives","Absolute and local extrema, the Extreme Value Theorem that guarantees them, and Fermat's Theorem pinning candidates to critical numbers. The Closed Interval Method turns the search for extrema into a finite checklist. Rolle's Theorem and the Mean Value Theorem then connect a function's values to its derivative, giving the tool that most of differential calculus rests on.\n",{"path":21619,"title":21620,"module":21616,"summary":21621},"\u002Fcalculus\u002Fapplications-of-derivatives\u002Fhow-derivatives-shape-a-graph","How Derivatives Shape a Graph","The sign of the first derivative fixes where a function rises and falls, and a sign change identifies each local extremum through the First Derivative Test. The second derivative sets concavity and inflection points and gives a faster Second Derivative Test. Limits at infinity describe end behavior and the horizontal asymptotes a curve settles toward.\n",{"path":21623,"title":21624,"module":21616,"summary":21625},"\u002Fcalculus\u002Fapplications-of-derivatives\u002Fcurve-sketching-and-optimization","Curve Sketching and Optimization","A checklist that synthesizes domain, symmetry, asymptotes, monotonicity, extrema, and concavity into a hand sketch of any function, plus the slant asymptote for rational functions whose degree exceeds the denominator's. The same extremum machinery, applied to a word problem, becomes the optimization template: model one quantity, reduce it to a function of a single variable, and find its absolute extremum.\n",{"path":21627,"title":21628,"module":21616,"summary":21629},"\u002Fcalculus\u002Fapplications-of-derivatives\u002Fnewtons-method-and-antiderivatives","Newton's Method and Antiderivatives","Newton's method solves $f(x) = 0$ by repeatedly replacing the curve with its tangent line and jumping to the tangent's root, converging fast when it works and diverging when the derivative is small. Antiderivatives reverse differentiation: every antiderivative of a function differs from another by a constant, so the general antiderivative is a family of parallel curves, pinned to one by an initial condition.\n",{"path":21631,"title":21632,"module":21633,"summary":21634},"\u002Fcalculus\u002Fintegrals\u002Farea-and-the-definite-integral","Area and the Definite Integral","Integrals","The area under a curve is defined as a limit of sums of rectangle areas. The same limit — a Riemann sum taken as the mesh shrinks to zero — defines the definite integral, a single number measuring signed area, total distance, and every accumulated quantity built the same way. Its properties, comparison bounds, and reading as net area follow directly from the limit.\n",{"path":21636,"title":21637,"module":21633,"summary":21638},"\u002Fcalculus\u002Fintegrals\u002Fthe-fundamental-theorem-of-calculus","The Fundamental Theorem of Calculus","Differentiation and integration are inverse operations. Part 1 says the derivative of an area-accumulation function is the integrand; Part 2 says a definite integral equals the change in any antiderivative across the interval. Together they replace limits of Riemann sums with antiderivative lookups, define the indefinite integral, and give the Net Change Theorem for rates.\n",{"path":21640,"title":21641,"module":21633,"summary":21642},"\u002Fcalculus\u002Fintegrals\u002Fthe-substitution-rule","The Substitution Rule","Substitution runs the Chain Rule backward: spotting an inner function whose derivative also appears in the integrand lets the variable change to $u$ and collapse a composite integral to a simple one. The rule applies to indefinite and definite integrals, with two ways to handle the limits, and it yields the symmetry shortcuts that double even integrands and vanish odd ones.\n",{"path":21644,"title":21645,"module":21646,"summary":21647},"\u002Fcalculus\u002Fapplications-of-integration\u002Fareas-and-volumes","Areas Between Curves and Volumes","Applications of Integration","A definite integral computes any quantity that a limit of Riemann sums approximates. Applied to geometry it gives the area between two curves and the volume of a solid: by cross-sections, by disks and washers when the region is revolved, and by cylindrical shells when inverting the boundary is awkward.\n",{"path":21649,"title":21650,"module":21646,"summary":21651},"\u002Fcalculus\u002Fapplications-of-integration\u002Fwork-average-value-and-arc-length","Work, Average Value, Arc Length, and Surface Area","The work done by a force that varies with position, the average value of a function and the Mean Value Theorem it satisfies, the length of a curve, and the area of a surface swept out by revolving that curve. Each is a limit of Riemann sums, hence a definite integral.\n",{"path":21653,"title":21654,"module":21646,"summary":21655},"\u002Fcalculus\u002Fapplications-of-integration\u002Fphysics-economics-and-probability","Applications to Physics, Economics, and Probability","Definite integrals in physics, economics, and statistics: the force a fluid exerts on a submerged plate, the balance point of a plane region, the money consumers save at a market price, and the probability that a continuous random variable lands in an interval, together with its mean.\n",{"path":21657,"title":21658,"module":21659,"summary":21660},"\u002Fcalculus\u002Fexponential-logarithmic-and-inverse-functions\u002Finverse-functions-logarithms-and-exponentials","Inverse Functions, Logarithms, and Exponentials","Exponential, Logarithmic, and Inverse Functions","A one-to-one function has an inverse that reverses it, with a graph mirrored across y = x and a derivative given by the reciprocal-slope rule. The exponential e^x is its own derivative and the natural logarithm has derivative 1\u002Fx; logarithmic differentiation turns products, quotients, and variable powers into sums.\n",{"path":21662,"title":21663,"module":21659,"summary":21664},"\u002Fcalculus\u002Fexponential-logarithmic-and-inverse-functions\u002Fgrowth-decay-inverse-trig-and-hyperbolic-functions","Growth, Decay, Inverse Trigonometric, and Hyperbolic Functions","Any quantity whose rate of change is proportional to its size grows or decays exponentially, the single equation y' = ky behind populations, radioactive decay, cooling, and continuously compounded interest. The inverse trigonometric functions have algebraic derivatives, and the hyperbolic functions, built from e^x and e^{-x}, describe the hanging cable.\n",{"path":21666,"title":21667,"module":21659,"summary":21668},"\u002Fcalculus\u002Fexponential-logarithmic-and-inverse-functions\u002Flhospitals-rule","Indeterminate Forms and l'Hospital's Rule","When a limit produces 0\u002F0 or infinity over infinity, the value is undetermined by the forms alone. l'Hospital's Rule resolves both by replacing the ratio of functions with the ratio of their derivatives. Products, differences, and powers reduce to a quotient the rule can handle, and repeated use ranks the growth of logarithms, powers, and exponentials.\n",{"path":21670,"title":21671,"module":21672,"summary":21673},"\u002Fcalculus\u002Ftechniques-of-integration\u002Fintegration-by-parts","Integration by Parts","Techniques of Integration","The product rule for derivatives reverses into integration by parts, trading the integral of $u\\,\\d v$ for the integral of $v\\,\\d u$ whenever the second is easier. The LIATE ordering fixes which factor to differentiate. Standard cases: a polynomial against a transcendental factor, repeated parts, cyclic integrals that solve for themselves, and reduction formulas that peel an exponent down by recursion.\n",{"path":21675,"title":21676,"module":21672,"summary":21677},"\u002Fcalculus\u002Ftechniques-of-integration\u002Ftrigonometric-integrals-and-substitution","Trigonometric Integrals and Substitution","Two related techniques. Trigonometric integrals evaluate powers and products of sine, cosine, tangent, and secant by splitting off one factor and converting the rest with a Pythagorean identity, or by dropping even powers with half-angle formulas. Trigonometric substitution runs the idea in reverse: replace x by a sine, tangent, or secant to clear a radical, integrate, then read the answer back off a reference triangle.\n",{"path":21679,"title":21680,"module":21672,"summary":21681},"\u002Fcalculus\u002Ftechniques-of-integration\u002Fpartial-fractions-and-integration-strategy","Partial Fractions and Integration Strategy","Any rational function integrates in closed form: factor the denominator, split the fraction into simple pieces by partial fractions, and integrate each piece as a logarithm or an arctangent. Four denominator cases exhaust the possibilities. A four-step strategy then sorts an arbitrary integrand by its shape to the technique that fits it, and a short catalog records elementary functions whose antiderivatives are not elementary.\n",{"path":21683,"title":21684,"module":21672,"summary":21685},"\u002Fcalculus\u002Ftechniques-of-integration\u002Fapproximate-and-improper-integrals","Approximate and Improper Integrals","Two definite integrals the Fundamental Theorem cannot reach. With no antiderivative available, the Midpoint, Trapezoidal, and Simpson rules approximate the integral from sample values, each carrying a provable error bound. With an infinite interval or an integrand that blows up, the improper integral is defined as a limit that either converges or diverges; the Comparison Test settles which without evaluating it.\n",{"path":21687,"title":21688,"module":21689,"summary":21690},"\u002Fcalculus\u002Fparametric-and-polar\u002Fparametric-curves-and-their-calculus","Parametric Curves and Their Calculus","Parametric Equations and Polar Coordinates","A parametric curve gives x and y separately as functions of a third variable, recording not only a path but the direction and timing with which it is traced. Eliminating the parameter recovers a Cartesian equation; the slope, area, arc-length, and surface-area formulas run directly on the parameter, with the cycloid and astroid as worked examples.\n",{"path":21692,"title":21693,"module":21689,"summary":21694},"\u002Fcalculus\u002Fparametric-and-polar\u002Fpolar-coordinates","Polar Coordinates","Polar coordinates locate a point by a distance from the pole and an angle from the polar axis, giving circles, spirals, and flower-shaped curves short equations. Conversion between the two systems is right-triangle trigonometry, and treating a polar curve as a parametric curve in the angle yields the tangent, area, and arc-length formulas.\n",{"path":21696,"title":21697,"module":21689,"summary":21698},"\u002Fcalculus\u002Fparametric-and-polar\u002Fconic-sections","Conic Sections","Parabolas, ellipses, and hyperbolas are the plane curves cut from a double cone. Each has a focus-based geometric definition and a standard Cartesian equation. A single number, the eccentricity, ties the three together, and placing a focus at the pole gives all of them one polar equation that describes planetary orbits.\n",{"path":21700,"title":21701,"module":21702,"summary":21703},"\u002Fcalculus\u002Fsequences-and-series\u002Fsequences","Sequences","Infinite Sequences and Series","A sequence is a function on the positive integers, and its limit is defined almost exactly like a limit at infinity. The Limit Laws and Squeeze Theorem carry over from functions, monotonic and bounded sequences give a convergence criterion, and the Monotonic Sequence Theorem guarantees a limit exists without naming it.\n",{"path":21705,"title":21706,"module":21702,"summary":21707},"\u002Fcalculus\u002Fsequences-and-series\u002Fseries-and-the-integral-test","Series and the Integral Test","Adding infinitely many terms is made precise as the limit of partial sums. The two series with closed-form partial sums are geometric and telescoping; the harmonic series diverges even as its terms shrink to zero. The Integral Test compares a positive series to an improper integral, settling the p-series and supplying a remainder bound for estimating sums.\n",{"path":21709,"title":21710,"module":21702,"summary":21711},"\u002Fcalculus\u002Fsequences-and-series\u002Fthe-convergence-tests","The Convergence Tests","The comparison, alternating-series, ratio, and root tests decide convergence without a closed-form partial sum. Absolute convergence is stronger than conditional convergence and is preserved under rearrangement; an alternating series errs by less than its first omitted term. A test is chosen from the shape of the general term.\n",{"path":21713,"title":21714,"module":21702,"summary":21715},"\u002Fcalculus\u002Fsequences-and-series\u002Fpower-series","Power Series","A power series is a polynomial of infinite degree whose convergence set is an interval centered at $a$, with a radius the Ratio Test finds and endpoints that must be tested by hand. Inside that interval the series represents a function that can be differentiated and integrated term by term, generating new representations from the geometric series.\n",{"path":21717,"title":21718,"module":21702,"summary":21719},"\u002Fcalculus\u002Fsequences-and-series\u002Ftaylor-and-maclaurin-series","Taylor and Maclaurin Series","If a function equals a power series, its coefficients are forced: the nth is the nth derivative at the center over n factorial. We derive that formula, use Taylor's Inequality to prove the standard series for the exponential, sine, and cosine, record the binomial series and a reference table, and bound the error when a Taylor polynomial replaces a function.\n",{"path":21721,"title":21722,"module":21723,"summary":21724},"\u002Fcalculus\u002Fvectors-and-space-curves\u002Fvectors-and-the-dot-product","Three-Dimensional Coordinates, Vectors, and the Dot Product","Vectors and the Geometry of Space","Space needs three coordinates, so we set up the rectangular system, the distance formula, and the equation of a sphere. Vectors then package magnitude and direction into a single algebraic object with its own arithmetic. The dot product turns two vectors into a number that measures the angle between them, gives a clean test for orthogonality, and produces the projection of one vector onto another.\n",{"path":21726,"title":21727,"module":21723,"summary":21728},"\u002Fcalculus\u002Fvectors-and-space-curves\u002Fthe-cross-product-lines-and-planes","The Cross Product, Lines, and Planes","The cross product multiplies two vectors into a third perpendicular to both, with length equal to the area of the parallelogram they span. That one construction supplies the direction of a line, the normal of a plane, and, through the scalar triple product, the volume of a parallelepiped. Lines carry a point and a direction vector; planes carry a point and a normal, which fixes the angle between planes and the distance from a point to a plane.\n",{"path":21730,"title":21731,"module":21723,"summary":21732},"\u002Fcalculus\u002Fvectors-and-space-curves\u002Fcylinders-and-quadric-surfaces","Cylinders and Quadric Surfaces","A surface whose equation omits one variable is a cylinder: the graph of a plane curve swept along the missing axis. A second-degree equation in three variables is a quadric, and translation and rotation reduce every one to a short standard list. Traces — the curves cut by planes parallel to the coordinate planes — sort the six quadrics into ellipsoid, the two paraboloids, the cone, and the two hyperboloids.\n",{"path":21734,"title":21735,"module":21723,"summary":21736},"\u002Fcalculus\u002Fvectors-and-space-curves\u002Fvector-functions-and-space-curves","Vector Functions and Space Curves","A vector function assigns a vector to each value of a parameter, and as the parameter runs its tip traces a space curve. Taking limits, derivatives, and integrals component by component carries all of single-variable calculus into three dimensions. The derivative of a vector function is the tangent vector to its curve, and normalizing it gives the unit tangent that points the way along the path.\n",{"path":21738,"title":21739,"module":21723,"summary":21740},"\u002Fcalculus\u002Fvectors-and-space-curves\u002Farc-length-curvature-and-motion","Arc Length, Curvature, and Motion in Space","Integrating the speed of a vector function gives the length of its curve and a natural parameter, arc length, that depends only on the curve's shape. Curvature measures how fast the unit tangent turns, and together with the normal and binormal it builds the moving TNB frame. Reading the same vector function as a trajectory, its first two derivatives are velocity and acceleration, and acceleration splits cleanly into tangential and normal parts.\n",{"path":21742,"title":21743,"module":21744,"summary":21745},"\u002Fcalculus\u002Fpartial-derivatives\u002Ffunctions-of-several-variables","Functions of Several Variables, Limits, and Continuity","Partial Derivatives","A function of several variables assigns one number to each point of a region in the plane or in space. Domain, graph, level curve, and level surface describe it; limits and continuity extend to two variables, where a limit must agree along every path of approach, not just from the left and the right.\n",{"path":21747,"title":21744,"module":21744,"summary":21748},"\u002Fcalculus\u002Fpartial-derivatives\u002Fpartial-derivatives","A partial derivative holds every variable but one fixed and differentiates in the ordinary sense. Geometrically it is the slope of a trace curve cut from the surface by a coordinate plane. The freeze-and-differentiate rule computes the two first partials; the four second partials follow, and the two mixed ones agree under Clairaut's Theorem when they are continuous.\n",{"path":21750,"title":21751,"module":21744,"summary":21752},"\u002Fcalculus\u002Fpartial-derivatives\u002Ftangent-planes-and-the-chain-rule","Tangent Planes, Linear Approximation, and the Chain Rule","Near a point, a smooth surface looks like its tangent plane, and the plane's equation is built from the two partial derivatives. That linearization defines the total differential and the meaning of differentiability in two variables. The chain rule then propagates derivatives through composed functions, tracked by a tree diagram, and yields clean formulas for implicit differentiation.\n",{"path":21754,"title":21755,"module":21744,"summary":21756},"\u002Fcalculus\u002Fpartial-derivatives\u002Fdirectional-derivatives-and-the-gradient","Directional Derivatives and the Gradient","The partial derivatives measure slope along the two axes; the directional derivative measures slope along any chosen direction, and equals the gradient dotted with a unit vector. The gradient points in the direction of steepest increase, its length is the greatest rate, and it stands perpendicular to level curves and surfaces, which fixes the tangent plane to a level surface.\n",{"path":21758,"title":21759,"module":21744,"summary":21760},"\u002Fcalculus\u002Fpartial-derivatives\u002Foptimization-and-lagrange-multipliers","Optimization and Lagrange Multipliers","Extrema of a two-variable function sit at critical points where the gradient vanishes; the Second Derivatives Test sorts them into peaks, valleys, and saddles by the sign of a discriminant. Absolute extrema on a closed region also need the boundary. When the domain is itself a constraint curve, Lagrange multipliers set the two gradients parallel and solve the constrained problem.\n",{"path":21762,"title":21763,"module":21764,"summary":21765},"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fdouble-integrals","Double Integrals","Multiple Integrals and Vector Calculus","The double integral extends the definite integral to functions of two variables: a limit of Riemann sums that measures signed volume under a surface. Fubini's Theorem turns it into two ordinary integrations done one after the other, general regions of type I and type II fix the inner limits, polar coordinates absorb circular symmetry through the factor r, and the same machine computes mass, center of mass, and moments of a lamina.\n",{"path":21767,"title":21768,"module":21764,"summary":21769},"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Ftriple-integrals-and-coordinate-systems","Triple Integrals and Coordinate Systems","The triple integral integrates a function of three variables over a solid, as a limit of Riemann sums evaluated by three nested single integrations. Cylindrical coordinates add the factor r to handle axial symmetry, spherical coordinates add rho-squared sine-phi for radial symmetry, and the general change of variables shows both volume elements are Jacobian determinants of the coordinate map. Surface area for a graph completes the measurement toolkit.\n",{"path":21771,"title":21772,"module":21764,"summary":21773},"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fvector-fields-and-line-integrals","Vector Fields and Line Integrals","A vector field assigns a vector to every point of space; the line integral of a field along a curve accumulates its tangential component, measuring work. Conservative fields are gradients of a potential, and for them the Fundamental Theorem for Line Integrals makes the integral depend only on the endpoints. Path independence, closed-loop integrals of zero, and the component test for a potential are three faces of the same property.\n",{"path":21775,"title":21776,"module":21764,"summary":21777},"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fgreens-theorem-curl-and-divergence","Green's Theorem, Curl, and Divergence","Green's Theorem equates the line integral of a field around a positively oriented closed curve with a double integral over the enclosed region, turning a boundary computation into an area computation and vice versa. Curl measures local circulation and divergence measures local outflow; the two vector forms of Green's Theorem express the boundary integral as the integrated curl or divergence, the planar case of Stokes' and the Divergence Theorem.\n",{"path":21779,"title":21780,"module":21764,"summary":21781},"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fsurface-integrals","Parametric Surfaces and Surface Integrals","A parametric surface is the image of a two-variable vector function; its area element is the magnitude of the cross product of the two tangent vectors. The surface integral of a scalar function sums it over that area, and the flux integral of a vector field sums the field's normal component, measuring flow through the surface. Orientation by a choice of unit normal makes flux well-defined, the integral Stokes' and the Divergence Theorem operate on.\n",{"path":21783,"title":21784,"module":21764,"summary":21785},"\u002Fcalculus\u002Fmultiple-integrals-and-vector-calculus\u002Fstokes-and-the-divergence-theorem","Stokes' Theorem and the Divergence Theorem","Stokes' Theorem lifts Green's Theorem into space: the line integral of a field around the boundary of a surface equals the flux of its curl through the surface. The Divergence Theorem relates the outward flux across a closed surface to the triple integral of divergence over the solid it encloses. Together with the Fundamental Theorem of Calculus and its line-integral and Green counterparts, they are one theorem: the integral of a derivative over a region equals the integral of the field over its oriented boundary.\n",{"path":21787,"title":21788,"module":6,"summary":6},"\u002Fcalculus","Calculus",{"path":21790,"title":21791,"module":21235,"summary":21792},"\u002Fmechanics\u002Ffoundations\u002Fmeasurement-and-dimensions","Measurement and Dimensions","Every physical quantity is a number attached to a unit, and that pairing is what lets you check an equation before computing anything, since terms that add together must carry the same dimensions. We build the SI base units and the notion of dimension, then use dimensional analysis to test a proposed relation and form scaling groups — a method that fixes a formula's shape but never its numerical constants. The lesson also sets how precisely a result may be stated, through significant figures, propagated uncertainty, and order-of-magnitude checks that catch errors a raw calculator answer hides.\n",{"path":21794,"title":21795,"module":21235,"summary":21796},"\u002Fmechanics\u002Ffoundations\u002Fvector-algebra","Vector Algebra","Force, velocity, and displacement all carry a direction, so mechanics needs an arithmetic that respects it; adding magnitudes alone gives the wrong answer the moment two arrows point different ways. We set up vectors and their components in a chosen basis, then build the two products that carry most of the physics — the dot product, which extracts the part of one vector along another and yields work and power, and the cross product, which measures oriented area and yields torque and angular momentum. Rotating the axes changes the components while leaving the vector itself untouched, and the same component method resolves a force along whatever directions a constraint picks out.\n",{"path":21798,"title":21799,"module":21800,"summary":21801},"\u002Fmechanics\u002Fkinematics\u002Fone-dimensional-motion","One-Dimensional Motion","Kinematics","Motion along a line already forces the two questions the whole of kinematics repeats: how fast is the object moving now, and where will it be next? Velocity and acceleration answer the first as derivatives of position; integrating them back — the signed area under a graph — answers the second. We derive the constant-acceleration equations, mark exactly where the \"constant\" assumption is load-bearing, and see why sign, not magnitude, is what carries direction.\n",{"path":21803,"title":21804,"module":21800,"summary":21805},"\u002Fmechanics\u002Fkinematics\u002Fmotion-graphs","Motion Graphs","Draw a motion as a graph and its two most useful facts turn geometric: the slope of the position curve is the velocity, and the area under the velocity curve is the displacement. We read motion in both directions — differentiating a graph for the next rate, integrating it back to recover position — and handle the curved, piecewise, and noisy graphs that real measurements produce. Along the way we see why a velocity estimated from two positions belongs to the midpoint of their interval, not its end.\n",{"path":21807,"title":21808,"module":21800,"summary":21809},"\u002Fmechanics\u002Fkinematics\u002Fprojectile-motion","Projectile Motion","Throw an object and it seems to trace one curved path, but the motion is really two independent one-dimensional motions running at once: constant velocity across the ground and free fall in the vertical. Splitting it that way turns every projectile question — how long it stays up, how far it lands, how high it climbs, whether it clears an obstacle — into a pair of equations you already know. We derive the parabolic trajectory, work both the forward and the inverse problems, and show why the familiar $45^\\circ$ range-maximizing angle holds only when launch and landing heights match.\n",{"path":21811,"title":21812,"module":21800,"summary":21813},"\u002Fmechanics\u002Fkinematics\u002Frelative-motion","Relative Motion","A velocity is only ever measured relative to some observer, so a boat's speed through the water, over the ground, and as seen from another boat are three different vectors. Choosing the right frame — and subtracting one motion from another — collapses river crossings, crosswind headings, pursuit, and closest-approach problems into a single vector equation. We build the relative-velocity and relative-position relations for uniformly moving frames, show why acceleration is the one quantity all such observers agree on, and note where rotating frames break the simple subtraction.\n",{"path":21815,"title":21816,"module":21800,"summary":21817},"\u002Fmechanics\u002Fkinematics\u002Fcircular-motion","Circular Motion","An object going around a circle at a steady speed is still accelerating, because its velocity is forever changing direction — the fact that governs everything from a car on a curve to a satellite in orbit. We tie the angular description (angle, angular velocity, angular acceleration) to the linear one through $v=r\\omega$, split the acceleration into an inward part that turns the velocity and a tangential part that changes its speed, and extend the inward $v^2\u002Fr$ result to any curved path through its local radius of curvature. Constant angular acceleration then mirrors straight-line motion equation for equation.\n",{"path":21819,"title":21820,"module":21821,"summary":21822},"\u002Fmechanics\u002Fdynamics\u002Fnewtons-laws","Newton's Laws","Dynamics","What makes a body change its motion, and in which frames does the answer take its simplest form? Newton's three laws settle both: inertial frames are the ones where a force-free body coasts, force is whatever changes momentum, and every interaction pushes back on its source. We write the second law as $\\sum\\vec F=\\d\\vec p\u002F\\d t$, reduce it to $m\\vec a$ at constant mass, and separate what a scale actually reads — the support force — from the weight $m\\vec g$ it is so often mistaken for.\n",{"path":21824,"title":21825,"module":21821,"summary":21826},"\u002Fmechanics\u002Fdynamics\u002Ffree-body-diagrams","Free-Body Diagrams","Once several forces act on a body at once, the reliable way to predict its motion is to isolate that one body and draw every external push and pull on it — nothing more, nothing less. The free-body diagram is that discipline. We fix a system boundary, resolve $\\sum\\vec F=m\\vec a$ into components along axes chosen to fit the geometry, and solve for the unknowns a problem hands us — normal forces, tensions, friction, and the acceleration a constraint permits — seeing why internal forces drop out only when the boundary encloses both bodies that share them.\n",{"path":21828,"title":21829,"module":21821,"summary":21830},"\u002Fmechanics\u002Fdynamics\u002Ffriction-and-curved-motion","Friction and Curved Motion","Real surfaces grip before they slip, fluids push back harder the faster you move through them, and anything rounding a bend must be pulled toward the inside of the curve by something. This lesson supplies the force laws for those three cases. We bound static friction by $|f_s|\\leq\\mu_sN$ and switch to kinetic friction $\\mu_kN$ once sliding starts, model drag as a speed-dependent resistance that levels off at a terminal speed, and show that circular motion demands an inward net force $mv^2\u002Fr$ furnished by real interactions — friction, a banked normal force, tension — never by an invented outward one.\n",{"path":21832,"title":21833,"module":21821,"summary":21834},"\u002Fmechanics\u002Fdynamics\u002Fnumerical-dynamics","Numerical Dynamics","Most force laws — quadratic drag, coupled oscillators, anything nonlinear — admit no closed-form trajectory, so we advance the motion one small time step at a time and let arithmetic do what algebra cannot. This lesson turns $\\d\\vec y\u002F\\d t=f(t,\\vec y)$ into a marching rule. We derive the Euler, Euler--Cromer, midpoint, and Verlet updates, weigh their accuracy and stability, watch a drifting energy expose a bad scheme, and use step-halving and conserved quantities to separate the error of the method from the error of the model.\n",{"path":21836,"title":21837,"module":21821,"summary":21838},"\u002Fmechanics\u002Fdynamics\u002Fcenter-of-mass-systems","Center-of-Mass Systems","A firework bursts into a dozen fragments, yet one point keeps gliding along the original parabola as though nothing had happened. That point is the centre of mass, and following it collapses a many-body tangle into a single equation of motion. We define $\\vec R=\\frac1M\\sum_i m_i\\vec r_i$ and its continuous form, show that internal forces cancel so that only external ones move it, $M\\vec A_{\\rm cm}=\\sum\\vec F_{\\rm ext}$, and put the result to work on recoil, collisions viewed from the centre-of-mass frame, and rocket propulsion, where mass leaving the boundary carries momentum with it.\n",{"path":21840,"title":21841,"module":21842,"summary":21843},"\u002Fmechanics\u002Fenergy\u002Fwork-and-kinetic-energy","Work and Kinetic Energy","Energy","A constant push along a straight path is trivial to score, but real forces vary and bend along curved trajectories, and only the component along the motion transfers any energy. Work captures exactly that transfer as the line integral $W=\\int\\vec F\\cdot\\d\\vec r$, and the work-kinetic-energy theorem turns it into a statement about speed: the net work on a particle equals the change in its $\\tfrac12 mv^2$. We build work up from the dot product to the signed area under a force curve, derive the theorem from Newton's second law, and read power as its instantaneous rate $P=\\vec F\\cdot\\vec v$.\n",{"path":21845,"title":21846,"module":21842,"summary":21847},"\u002Fmechanics\u002Fenergy\u002Fpotential-energy","Potential Energy","When a force does the same work no matter which path a particle takes, that work can be stored as a function of position alone, and solving for the motion becomes bookkeeping instead of integration. We single out the forces that qualify — the conservative ones, for which $\\oint\\vec F\\cdot\\d\\vec r=0$ — define their potential energy through $\\vec F=-\\nabla U$, and use conservation of mechanical energy to read speeds, turning points, and equilibria straight off a potential curve. Friction breaks the shortcut, so we also track where mechanical energy leaks away as heat.\n",{"path":21849,"title":21850,"module":21842,"summary":21851},"\u002Fmechanics\u002Fenergy\u002Fmultiparticle-work","Multiparticle Work","A single particle has one velocity and one kinetic energy; a system of many can spin, deform, explode, and warm up while its centre of mass glides along as if nothing happened. Splitting the motion into a centre-of-mass part and an internal part separates the energy that momentum already fixes from the energy left free for relative motion, $K=\\tfrac12MV_{\\rm cm}^2+K'$. We derive the centre-of-mass work theorem, see why an explosion or a released spring can raise total kinetic energy with no external work at all, and use the reduced-mass and centre-of-mass frames to make collisions and internal transfers clean.\n",{"path":21853,"title":21854,"module":21842,"summary":21855},"\u002Fmechanics\u002Fenergy\u002Fmass-energy-and-binding","Mass-Energy and Binding","Relativity puts rest itself on the energy ledger: a mass $m$ carries energy $mc^2$ even when it sits still, so weighing a system's separated pieces and weighing the assembled whole give different answers, and the gap is binding energy. We convert freely between mass units and MeV, compute the energy that holds a nucleus together, and read the binding-energy-per-nucleon curve that explains why fusing light nuclei and splitting heavy ones both release energy. Reaction $Q$ values, thresholds, and recoil then follow from the same mass-difference accounting, once the frame and mass convention are fixed.\n",{"path":21857,"title":21858,"module":21842,"summary":21859},"\u002Fmechanics\u002Fenergy\u002Fphotons-and-quantization","Photons and Quantization","Light delivers its energy in indivisible lumps: a photon of frequency $f$ carries exactly $hf$, and this one fact explains why a dim blue lamp ejects electrons that an intense red one cannot. We fix a photon's energy and momentum from its wavelength, follow the quanta through emission, absorption, and the photoelectric threshold $K_{\\rm max}=hf-\\phi$, and watch energy and momentum conservation together produce the Compton wavelength shift when a photon scatters from an electron. The recurring discipline is unit and frame care, where a stray factor of $10^9$ or a forgotten rest energy quietly ruins an answer.\n",{"path":21861,"title":21862,"module":21863,"summary":21864},"\u002Fmechanics\u002Fmomentum\u002Fmomentum-and-collisions","Momentum and Collisions","Momentum","When two objects collide, the forces between them are too brief and too tangled to integrate directly, yet the result is fixed by one conserved quantity. Linear momentum $\\vec p=m\\vec v$ turns Newton's second law into the impulse-momentum theorem $\\vec J=\\Delta\\vec p$, and for an isolated system into a conservation law that holds through any internal collision, however dissipative. We use it to separate elastic from inelastic collisions, follow the centre of mass, and read impulse as the signed area under a force-time curve — always tracking which external impulses the chosen system and interval let us drop.\n",{"path":21866,"title":21867,"module":21863,"summary":21868},"\u002Fmechanics\u002Fmomentum\u002Fcenter-of-mass-collisions","Center-of-Mass Collisions","A two-body collision that looks asymmetric in the laboratory becomes almost trivial in the frame that rides along with the centre of mass, where the total momentum is zero and the two momenta stay equal and opposite. We build that frame, reduce the pair to a single relative coordinate carrying the reduced mass $\\mu$, and show that an elastic collision there only rotates one momentum vector while its length holds fixed. Transforming back to the laboratory then handles elastic and inelastic collisions, scattering angles, and reaction thresholds with the same construction — and shows why relative speed, not laboratory kinetic energy, measures what a collision can convert.\n",{"path":21870,"title":21871,"module":21863,"summary":21872},"\u002Fmechanics\u002Fmomentum\u002Frocket-propulsion","Rocket Propulsion","A rocket speeds up by throwing mass backward, so its own mass drops as it flies and $\\vec F=m\\vec a$ no longer applies to a fixed body. Tracking the momentum the exhaust carries across the vehicle boundary gives thrust $T=Ru_e$ and, for a force-free burn, the rocket equation $\\Delta v=u_e\\ln(m_i\u002Fm_f)$ — a logarithm that makes large velocity changes expensive in propellant and forces staging. We then add the forces a real ascent cannot ignore, gravity, drag, and steering, and show how thrust and mass-flow records are cross-checked to infer the exhaust speed.\n",{"path":21874,"title":21875,"module":21876,"summary":21877},"\u002Fmechanics\u002Frotation\u002Frotational-inertia","Rotational Inertia","Rotation","Push a wheel and a merry-go-round with the same force and they speed up at wildly different rates: the same mass resists rotation differently depending on where it sits relative to the axis. That single fact is the moment of inertia, $I=\\int r_\\perp^2\\,\\d m$, and this lesson builds it from the ground up. We tie angular motion to linear through $s=r\\theta$, $v=r\\omega$, and $a_t=r\\alpha$, derive $I$ for rods, disks, and spheres, and use the parallel- and perpendicular-axis theorems to move between axes — always naming the axis, because the same body has as many moments of inertia as it has lines to spin about.\n",{"path":21879,"title":21880,"module":21876,"summary":21881},"\u002Fmechanics\u002Frotation\u002Frotational-dynamics","Rotational Dynamics","A force applied to a wheel does nothing unless it acts off the axis: what turns a rigid body is torque, force times lever arm. This lesson makes that precise and turns it into the rotational Newton's second law, $\\sum\\tau=I\\alpha$ about a fixed axis, the exact analogue of $\\sum F=ma$. From there we get rotational work $W=\\int\\tau\\,\\d\\theta$ and power $P=\\tau\\omega$, size a motor to a load, and solve pulleys and Atwood machines where the pulley's own inertia can no longer be ignored — always insisting that every torque be measured about the same axis.\n",{"path":21883,"title":21884,"module":21876,"summary":21885},"\u002Fmechanics\u002Frotation\u002Frolling-motion","Rolling Motion","A rolling wheel is doing two things at once — translating and spinning — but the no-slip condition $v_{cm}=R\\omega$ locks them together, and that single constraint is what makes rolling tractable. We use it to split the kinetic energy into $\\tfrac12Mv_{cm}^2+\\tfrac12I\\omega^2$, find how fast a cylinder reaches the bottom of an incline, and show why the contact point is instantaneously at rest. The static friction that enforces rolling does no work; we track its direction from the tendency to slip, and mark exactly where the model breaks once the required friction exceeds $\\mu_sN$.\n",{"path":21887,"title":21888,"module":21876,"summary":21889},"\u002Fmechanics\u002Frotation\u002Fangular-momentum","Angular Momentum","A skater pulls in her arms and spins faster, with no torque acting: that is angular momentum conservation, and it lets us answer questions that would be hopeless force by force. We build $\\vec L=\\vec r\\times\\vec p$, show it obeys $\\vec\\tau_{ext}=\\d\\vec L\u002F\\d t$, and use its conservation under zero external torque to link before and after in collisions, reconfigurations, and coupled rotors without ever resolving the internal forces. The catch is bookkeeping: the origin, the system boundary, and the frame must be fixed first, and a change in total $\\vec L$ always points to an external impulse someone forgot.\n",{"path":21891,"title":21892,"module":21876,"summary":21893},"\u002Fmechanics\u002Frotation\u002Frolling-resistance","Rolling Resistance","Ideal rolling should coast forever, yet every real wheel slows down. The reason is that a deformable tire and road do not press through a single point: the contact patch spreads, the normal-force resultant shifts ahead of the axle, and that offset is a resisting moment even with no gross sliding. We package it as an equivalent force $F_{rr}=C_{rr}N$, tie the coefficient to load, surface, speed, and temperature, and use coast-down, towing, and traction tests to separate this contact loss from aerodynamic drag, bearing friction, and the adhesion limit where rolling gives way to skidding.\n",{"path":21895,"title":21896,"module":21876,"summary":21897},"\u002Fmechanics\u002Frotation\u002Fgyroscopic-precession","Gyroscopic Precession","A spinning top leans over but does not fall — it swings its axis in a slow horizontal circle instead. The paradox dissolves once torque is read as the rate of change of a vector: gravity's torque is perpendicular to the spin angular momentum, so it turns $\\vec L$ rather than toppling it. We derive the steady precession rate $\\Omega\\simeq Mgr\u002F(I_s\\omega_s)$ in the fast-top limit, state the assumptions it leans on — dominant spin, slow tilt, negligible bearing torque — and read nutation, support motion, and a decaying spin as the ways real gyroscopes depart from it.\n",{"path":21899,"title":21900,"module":21901,"summary":21902},"\u002Fmechanics\u002Fgravity-and-matter\u002Fkeplerian-orbits","Keplerian Orbits","Gravitation and Matter","Why do the planets trace ellipses rather than any other curve? Newton's inverse-square law collapses the two-body problem onto a single conic section, and the answer falls out of two conserved quantities: a central force can exert no torque, so angular momentum is fixed, and gravity is conservative, so energy is fixed. We read an orbit's size and shape straight off those invariants, recover all three of Kepler's laws, and derive escape speed, the vis-viva relation, and the timing of a pass. We also mark where the ideal ellipse breaks down — drag, oblateness, and a third body slowly move a real orbit.\n",{"path":21904,"title":21905,"module":21901,"summary":21906},"\u002Fmechanics\u002Fgravity-and-matter\u002Fgravitational-fields","Gravitational Fields","Instead of tracking the force between every pair of masses, we attach a field to the source and ask a test mass to read it off locally. That move pays off because gravity is conservative: the field is the gradient of a single scalar potential, and potentials from many sources simply add. We build the field-potential picture, use spherical symmetry and the shell theorem to get the point-mass exterior field and the zero interior field of a shell, and read tides straight out of the field's gradient. Along the way we mark exactly when the constant-$g$ and point-mass shortcuts hold and when a shape correction is needed.\n",{"path":21908,"title":21909,"module":21901,"summary":21910},"\u002Fmechanics\u002Fgravity-and-matter\u002Fstatic-equilibrium","Static Equilibrium","What does it take for a loaded structure to stay put? A body at rest needs its forces to cancel and its turning effects to cancel — $\\sum\\vec F=0$ and $\\sum\\vec\\tau=0$ about any point — and almost all of statics is the craft of turning a physical setup into those equations. We build free-body diagrams, replace supports, cables, friction, couples, and distributed loads with their idealized reactions, and locate the centre of gravity that decides whether a body tips. We also count equations against unknowns to separate a determinate problem from one that needs the material's deformation to resolve, and read every negative or inconsistent reaction as a sign that a contact or a boundary was chosen wrong.\n",{"path":21912,"title":21913,"module":21901,"summary":21914},"\u002Fmechanics\u002Fgravity-and-matter\u002Ffluid-statics","Fluid Statics","A fluid at rest cannot support a shear, so the only stress it carries is a pressure that must grow with depth to hold up the fluid above it. That single balance, $\\d p\u002F\\d z=-\\rho g$, runs the whole subject: it sets manometer readings, the force on a dam, and — integrated over a submerged boundary — Archimedes' buoyant force $F_B=\\rho g V_{\\rm disp}$. We derive these, use them to decide when a body floats and whether it floats upright, and mark where acceleration, rotation, compressibility, or capillarity forces a richer pressure model.\n",{"path":21916,"title":21917,"module":21901,"summary":21918},"\u002Fmechanics\u002Fgravity-and-matter\u002Ffluid-flow","Fluid Flow","Two accounting rules carry most of steady flow: mass cannot pile up, so the same volume crosses every section each second, and mechanical energy is conserved along a streamline when the fluid is ideal. From those we get continuity, Bernoulli's relation between pressure, speed, and height, and the results that follow — Torricelli's efflux speed, the Venturi meter, the Pitot tube. We then let go of the ideal assumptions one at a time: viscosity adds wall shear and head loss, Reynolds number decides laminar versus turbulent, and Mach number marks where a gas stops behaving as incompressible.\n",{"path":21920,"title":21921,"module":21901,"summary":21922},"\u002Fmechanics\u002Fgravity-and-matter\u002Forbital-motion","Orbital Motion","A circular orbit is nothing but free fall with enough sideways speed to keep missing the ground, and setting gravity equal to the centripetal requirement fixes that speed and the period once and for all. From the same energy bookkeeping we read off escape speed, sort orbits into bound, parabolic, and hyperbolic by the sign of their specific energy, and see why a tangential burn is the efficient way to change an orbit. We build the Hohmann transfer and its launch window, work the numbers for a geostationary orbit and an escape burn, and mark where finite thrust, perturbations, and an uncertain initial state pull a real trajectory off the ideal.\n",{"path":21924,"title":21925,"module":21901,"summary":21926},"\u002Fmechanics\u002Fgravity-and-matter\u002Fstress-and-elasticity","Stress and Elasticity","Rigid bodies are a fiction; every real material stretches, shears, or squeezes under load, and the useful question is how much. We define stress as force per area and strain as fractional deformation, then find that for small deformations the two are simply proportional — Hooke's law — with Young's, shear, and bulk moduli as the constants for stretch, twist, and volume change. From these we compute extensions, torsional twist, and stored elastic energy, and read a tensile curve for the yield, ultimate, and fracture points where linear elasticity ends. We also mark the practical limits: stress concentrations, fatigue, and the multiaxial states a single uniaxial modulus cannot capture.\n",{"path":21928,"title":21929,"module":21930,"summary":21931},"\u002Fmechanics\u002Foscillations-waves\u002Fdamped-oscillators","Damped Oscillators","Oscillations and Waves","Every real oscillator eventually stops: friction, drag, and internal loss drain its energy, so free motion is a decay rather than a permanent swing. Adding a velocity-proportional resistance to the spring-mass equation produces one dimensionless number, $b\u002F(2\\sqrt{mk})$, that decides whether the mass rings down through many cycles, returns once without overshoot, or crawls back slowly. We solve the three regimes, tie the observed decay to the power balance $b\\dot x^2$, and turn a measured ring-down into the decay rate and quality factor of the apparatus — reading damping off the data instead of assuming it.\n",{"path":21933,"title":21934,"module":21930,"summary":21935},"\u002Fmechanics\u002Foscillations-waves\u002Ftravelling-waves","Travelling Waves","A wave carries a shape, not the material: each element of a rope or air column oscillates in place while the disturbance travels through it. Writing that shape as $f(x\\mp vt)$ turns \"the pattern moves\" into a statement about the cosine's argument, and a local force balance on one string segment fixes the speed at $v=\\sqrt{T\u002F\\mu}$ — restoring stiffness over inertia, with amplitude nowhere in it. We build the sinusoidal wave and its phase, derive the wave equation from Newton's second law, and follow the energy a travelling wave transports, then check speed and power against those predictions.\n",{"path":21937,"title":21938,"module":21930,"summary":21939},"\u002Fmechanics\u002Foscillations-waves\u002Fwave-superposition","Wave Superposition","When two waves cross the same point, what does a probe read? In a linear medium the answer is arithmetic: the displacements add, $y=y_1+y_2$, and the pulses pass through each other unchanged. That one rule produces interference — reinforcement where the signs agree, cancellation where they oppose — and it guards against a common mistake, since displacement can vanish at an instant while the energy sits in transverse motion instead. We work out the signed sum, the phase bookkeeping for equal-frequency components, and why a null in the record is not a null in the wave.\n",{"path":21941,"title":21942,"module":21930,"summary":21943},"\u002Fmechanics\u002Foscillations-waves\u002Fstanding-waves","Standing Waves","Clamp a string at both ends and only certain frequencies survive: the ends must be nodes, and that single geometric demand quantizes the wave into a discrete set of modes $f_n=nv\u002F(2L)$. The travelling wave becomes a fixed pattern of nodes and antinodes — standing, not moving — because equal waves running in opposite directions superpose. We build the standing wave from its counter-propagating pieces, read the harmonic sequence off the boundary conditions (half-wavelengths for a fixed-fixed string, odd quarter-wavelengths for a closed pipe), and test the ideal model against node scans and resonance peaks.\n",{"path":21945,"title":21946,"module":21930,"summary":21947},"\u002Fmechanics\u002Foscillations-waves\u002Fsound-waves","Sound Waves","Sound is a pressure wave so small that a loud tone displaces air molecules by less than the width of an atom, yet a microphone reads it easily — because pressure, not displacement, is what the ear and the instrument sense. The acoustic impedance $Z=\\rho c$ ties pressure, density, and particle velocity together, fixes the intensity a wave carries, and sets the reference for the decibel, a logarithm that tames a $10^{12}$ range in power. We derive the sound speed from the gas's stiffness, convert between pressure and intensity levels, and treat the measurement itself — calibration, geometry, background, averaging — as part of the physics.\n",{"path":21949,"title":21950,"module":21930,"summary":21951},"\u002Fmechanics\u002Foscillations-waves\u002Fdoppler-effect","Doppler Effect","A passing siren drops in pitch not because the source changes but because motion repacks the wavefronts: an approaching source crowds its crests, a receding one stretches them, and a moving listener samples them at a different rate. For mechanical waves every velocity is measured against the medium, and one signed ratio $f_r=f_s(v-u_r)\u002F(v-u_s)$ captures both effects at once. We separate source motion, which sets crest spacing, from receiver motion, which sets arrival rate, invert the shift to recover radial velocity, and mark where the model breaks — supersonic sources, moving air, and reflected paths that carry two shifts, not one.\n",{"path":21953,"title":21954,"module":21930,"summary":21955},"\u002Fmechanics\u002Foscillations-waves\u002Fwave-packets","Wave Packets","No real signal is a single frequency: a disturbance that starts and stops is built from a band of wave numbers, and the width of that band is what makes it local. We ask how such a packet moves — carrier crests at the phase velocity $v_\\mathrm p=\\omega\u002Fk$, the envelope at the group velocity $v_\\mathrm g=\\d\\omega\u002F\\d k$ — and why the two differ once a medium is dispersive. Curvature $\\d^2\\omega\u002F\\d k^2$ spreads and chirps the packet as it travels, and the Fourier reciprocity that ties bandwidth to duration explains why a finite record, aliasing, or a coarse probe can imitate that spreading unless the sampling limits are respected.\n",{"path":21957,"title":21958,"module":21930,"summary":21959},"\u002Fmechanics\u002Foscillations-waves\u002Fbeats-and-coupling","Beats and Coupling","Add two tones a few hertz apart and the sum swells and fades at their difference frequency — a beat — though neither source is changing. We work out that envelope, then ask the mechanical version of the same question: join two oscillators and a single resonance splits into normal modes, with energy sloshing between the coordinates at their frequency difference. The lesson identifies when a slow amplitude envelope signals genuine coupling rather than two independent sources, drift, or deliberate modulation, reading it from envelope timing, spectral sidebands, and the mode shapes.\n",{"path":21961,"title":21962,"module":21930,"summary":21963},"\u002Fmechanics\u002Foscillations-waves\u002Fsimple-harmonic-motion","Simple Harmonic Motion","Any system pushed back toward equilibrium by a force proportional to its displacement obeys one equation, $\\ddot x+\\omega_0^2x=0$, and so moves sinusoidally at $\\omega_0=\\sqrt{k\u002Fm}$ whatever the amplitude. We derive that motion, follow its energy $E=mv^2\u002F2+kx^2\u002F2$ trading between kinetic and potential form at constant total, and read the elliptical phase-space orbit Hooke's law implies. Period, amplitude, velocity, and acceleration then supply redundant checks: an amplitude-dependent period or a curved force residual is the signature that the linear model has failed, and mass-loading and offset tests separate a calibration error from a real frequency shift.\n",{"path":21965,"title":21966,"module":21930,"summary":21967},"\u002Fmechanics\u002Foscillations-waves\u002Fpendulum-motion","Pendulum Motion","A pendulum keeps time only because, for small swings, gravity supplies a restoring torque proportional to the angle — and $T=2\\pi\\sqrt{L\u002Fg}$ then follows without the mass appearing at all. We derive that result, mark exactly which assumptions carry it (small angle, negligible pivot loss, a rigid support), then relax them: finite amplitude lengthens the period through an elliptic integral, and an extended body replaces $L$ with the ratio of its moment of inertia to its center-of-mass distance. How the period drifts with amplitude or pivot position is what diagnoses the geometric, damping, and distributed-mass corrections.\n",{"path":21969,"title":21970,"module":21930,"summary":21971},"\u002Fmechanics\u002Foscillations-waves\u002Fdriven-oscillators","Driven Oscillators","Drive a damped oscillator at a frequency you control and it eventually forgets its own: $m\\ddot x+b\\dot x+kx=F_0\\cos\\omega t$ settles into a steady response whose amplitude and phase depend sharply on how close the drive sits to resonance. We solve for that response, show how damping alone fixes the resonance width, the peak power, and the settling time, and treat base excitation as the same problem with a different input. The steady-state formulas hold only for constant $m$, $b$, and $k$; level-dependent peaks or hysteresis between up- and down-sweeps are how nonlinearity or an extra mode announces itself.\n",{"path":21973,"title":21974,"module":21930,"summary":21975},"\u002Fmechanics\u002Foscillations-waves\u002Fwave-boundaries","Wave Boundaries","A pulse traveling along a string does something abrupt where the string's properties change: part reflects, part transmits, and which is which is set by the impedance mismatch alone. We impose continuity of displacement and transverse force at the join to get the reflection and transmission coefficients in terms of $Z=\\sqrt{T\\mu}$, fix their signs and the polarity flip, and balance the energy. The clean result assumes linear, nondispersive segments meeting at a localized join; pulse polarity, return timing, and energy ratios are the measurements that expose a real connector's mass, loss, or distributed transition.\n",{"path":21977,"title":21978,"module":21979,"summary":21980},"\u002Fmechanics\u002Fthermodynamics\u002Fkinetic-theory-of-ideal-gases","Kinetic Theory of Ideal Gases","Thermodynamics","A gas has no springs and no gears, yet it pushes on its container with a definite pressure and stores energy in a lawful way. Kinetic theory explains both from the motion of the molecules alone: pressure is the accumulated recoil of countless elastic impacts, and temperature is the average translational kinetic energy each molecule carries. We derive $pV=\\tfrac13Nm\\overline{v^2}$ from momentum transfer, read off $\\overline{K}_{\\rm tr}=\\tfrac32kT$, and use the Maxwell–Boltzmann distribution to separate the most probable, mean, and rms speeds — each the right average for a different question — while marking where the dilute, classical assumptions stop holding.\n",{"path":21982,"title":21983,"module":21979,"summary":21984},"\u002Fmechanics\u002Fthermodynamics\u002Ffirst-law-of-thermodynamics","First Law of Thermodynamics","Heat a gas and it may warm, expand, or both; compress it and the same energy can reappear as a temperature rise. The first law settles the bookkeeping: internal energy is a state property whose change equals the heat added plus the work done on the system, $\\Delta E_{\\rm int}=Q_{\\rm in}+W_{\\rm on}$. We fix a system boundary and one sign convention, compute boundary work as $\\int p\\,\\d V$ along a path, and use calorimetry to measure heat and heat capacities. The recurring point is that heat and work are path-dependent transfers while their sum is not, so an energy ledger closes only once every boundary crossing is named.\n",{"path":21986,"title":21987,"module":21979,"summary":21988},"\u002Fmechanics\u002Fthermodynamics\u002Fentropy-and-the-second-law","Entropy and the Second Law","The first law lets energy flow either way; it never says which way heat actually goes. The second law supplies the missing arrow. Entropy, defined through the reversible transfer $\\d S=\\delta Q_{\\rm rev}\u002FT$, can only increase in an isolated system, and that single inequality fixes the direction of heat flow and caps every engine, refrigerator, and heat pump at its Carnot value. We build entropy ledgers for reservoirs and working substances, separate the entropy carried by heat from the entropy generated by irreversibility, and read the sign of the total as a hard check on any proposed thermal machine.\n",{"path":21990,"title":21991,"module":21979,"summary":21992},"\u002Fmechanics\u002Fthermodynamics\u002Fthermal-processes","Thermal Processes","Heat rarely sits still: it stretches solids, pushes real gases off their ideal isotherms, and leaks across walls by conduction, convection, and radiation. Each behavior becomes a number a designer can use. Thermal expansion sets the gaps in a bridge and the stress in a clamped rod; the van der Waals equation and a phase diagram fix when $pV=nRT$ or a latent-heat term applies; Fourier's law, Newton cooling, and Stefan–Boltzmann radiation give the rate of heat flow. We assemble these into thermal-resistance networks and transient time constants, then mark where contact resistance, phase change, or a hidden thermal bridge breaks the simple model.\n",{"path":21994,"title":21995,"module":21979,"summary":21996},"\u002Fmechanics\u002Fthermodynamics\u002Fphase-changes","Phase Changes","Add heat to ice and its temperature climbs — until it reaches $0\\ ^\\circ\\mathrm C$, where the thermometer stalls while the ice melts. That plateau is the whole subject: at a phase boundary the energy rearranges molecules, $Q=mL$, instead of raising temperature, which resumes only once one phase is gone. We stage a heating path into sensible-heat legs ($Q=mc\\Delta T$) and latent plateaus, use the Clausius–Clapeyron relation to track how a boiling point moves with pressure, and solve calorimetry by testing each coexistence endpoint — so a melt fraction that lands outside $[0,1]$ flags a wrong final-state guess rather than a real state.\n",{"path":21998,"title":21999,"module":21979,"summary":22000},"\u002Fmechanics\u002Fthermodynamics\u002Fthermal-machines","Thermal Machines","An engine, a refrigerator, and a heat pump are one machine read three ways: each shuttles heat between a hot and a cold reservoir while trading work at the boundary, and only the flow you call useful separates them. A heat engine turns part of $Q_h$ into work, $W=Q_h-Q_c$; a refrigerator spends work to pull $Q_c$ from the cold side; a heat pump counts the warm-side delivery instead. We measure each with its own ratio — efficiency or coefficient of performance — bound them all by the Carnot limit that reservoir temperatures alone set, and track how finite temperature differences, throttling, and friction generate entropy and pull real machines below that bound.\n",{"path":22002,"title":22003,"module":6,"summary":6},"\u002Fmechanics","Mechanics & Dynamics",{"path":22005,"title":22006,"module":22007,"summary":22008},"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Fcharge-and-conductors","Charge and Conductors","Electric Fields","Rub two objects together and one pulls electrons from the other; nothing is created, only moved. We define what electric charge is — conserved, additive, and quantized in units of $e$ — and why a conductor's mobile carriers rearrange until its interior field vanishes and its surface sits at one potential. We follow charge through contact, induction, and grounding, treat the field-free cavity that turns a conductor into a shield, and mark where finite conductivity and leakage set the limits of the electrostatic picture.\n",{"path":22010,"title":22011,"module":22007,"summary":22012},"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Fcoulombs-law","Coulomb's Law","Two charges at rest push or pull along the line joining them, and the whole of electrostatics is assembled by adding up such pairs. We measure that force — its inverse-square falloff, its linear dependence on each charge, the sign that says attract or repel — and write it as a vector so direction survives superposition. We work the magnitude and component forms on real numbers, check them against limiting cases and dimensions, and fix the point-charge approximation to source sizes small against every separation.\n",{"path":22014,"title":22015,"module":22007,"summary":22016},"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Felectric-field-and-force","Electric Field and Force","Rather than ask how one charge reaches across empty space to another, we credit the source with a field that fills the space and let a second charge respond to whatever field sits at its own location. Electric field is force per unit positive test charge, $\\vec E=kq\\hat r\u002Fr^2$ for a point source, and source fields add before any receiving charge is placed. We compute those fields and the force $\\vec F=q\\vec E$ they exert, then follow a charge along its parabolic path through a uniform field and into nonuniform fields where the dynamics turn position-dependent.\n",{"path":22018,"title":22019,"module":22007,"summary":22020},"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Felectric-field-maps","Electric Field Maps","A field is a vector at every point of space, and the quickest way to grasp one is to draw it. We build the two standard pictures — continuous field lines tangent to $\\vec E$, and scaled vector arrows — and read direction, magnitude, and the location of nulls straight off them. We fix what a line drawing can and cannot say: density encodes magnitude only under a stated seeding rule, and integral curves never cross at a regular point. From there we work the topology near sources, sinks, and conductor surfaces, and state the step-size and interpolation checks a numerical map must pass.\n",{"path":22022,"title":22023,"module":22007,"summary":22024},"\u002Felectricity-and-magnetism\u002Felectric-fields\u002Felectric-dipoles","Electric Dipoles","Most neutral matter carries no net charge yet still responds to an electric field, because its positive and negative charge sit slightly apart. That separation is a dipole, moment $\\vec p=q\\vec d$ pointing from the negative to the positive charge, and it is the leading term in how any neutral distribution looks from far away. We derive the torque $\\vec p\\times\\vec E$ and energy $-\\vec p\\cdot\\vec E$ a uniform field imposes, the net force a field gradient adds, and the axial and equatorial $1\u002Fr^3$ fields the pair produces, then measure how far out the point-dipole approximation still holds.\n",{"path":22026,"title":22027,"module":22028,"summary":22029},"\u002Felectricity-and-magnetism\u002Fcontinuous-charge-distributions\u002Fcontinuous-charge-fields","Continuous Charge Fields","Continuous Charge Distributions","A charged rod, ring, or disk is not a point, yet its field is still nothing but Coulomb's law added up over the charge it carries. We replace the discrete sum by an integral, with $\\d q=\\lambda\\d\\ell$, $\\sigma\\d A$, or $\\rho\\d V$, so the real work becomes geometry: writing the vector from each source element to the field point, and letting symmetry cancel the components that must cancel before any integral is attempted. We carry the line, ring, and disk fields through in full, then check each result against its near field, its far field, and its dimensions.\n",{"path":22031,"title":22032,"module":22028,"summary":22033},"\u002Felectricity-and-magnetism\u002Fcontinuous-charge-distributions\u002Fgauss-law-and-conductors","Gauss's Law and Conductors","Adding up Coulomb's law over a whole distribution is laborious; Gauss's law trades that sum for a single statement, that the flux of $\\vec E$ out of any closed surface counts the charge inside, $\\oint\\vec E\\cdot\\d\\vec A=Q_{\\rm enc}\u002F\\varepsilon_0$. The law is always true, but it hands over the field only when the source is symmetric enough to pull $E$ outside the integral. We apply it to spheres, lines, and sheets, then turn it on conductors, where the zero interior field drives every excess charge to the surface and fixes the normal-field jump $\\sigma\u002F\\varepsilon_0$, the charge induced on a cavity wall, and electrostatic shielding.\n",{"path":22035,"title":22036,"module":22037,"summary":22038},"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Fpoint-charge-potential","Point-Charge Potential","Electric Potential","The electrostatic force is conservative, so the work it does between two points\ndepends only on the endpoints. That lets us trade the vector field for a single\nscalar attached to each point, the electric potential, the potential energy a unit\ncharge would have there. We build potential from the work integral, fix the usual\nreference at infinity, and add point sources as scalars, $V=k\\sum_i q_i\u002Fr_i$,\navoiding the vector bookkeeping the field demands. Signed charges, the reference\nchoice, equipotential motion, and far-field expansions each give an independent\ncheck on a result.\n",{"path":22040,"title":22041,"module":22037,"summary":22042},"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Fpotential-gradients-and-equipotentials","Potential Gradients and Equipotentials","Given the potential everywhere, how do we recover the field? The field is the\nnegative gradient, $\\vec E=-\\nabla V$: it points down the steepest local drop in\npotential, and its magnitude is set by how fast $V$ changes, not by the shape of a\ncontour. We read off components with directional derivatives, reconstruct fields\nfrom measured potential grids using centered differences, and use closed-loop\nintegrals and grid refinement to test whether a reconstructed field is physically\nconsistent.\n",{"path":22044,"title":22045,"module":22037,"summary":22046},"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Felectrostatic-energy-and-pressure","Electrostatic Energy and Pressure","Assembling a charge configuration takes work, and that work is stored, but where\nis it kept and how much is there? We total it two ways: as a sum over the charges,\n$U=\\tfrac12\\sum_i q_iV_i$, and as an integral over the field itself,\n$u_E=\\tfrac12\\varepsilon_0E^2$, energy the field carries in every region it fills.\nDifferentiating the stored energy at fixed charge or at fixed voltage recovers the\nmechanical force on a conductor, and at a charged surface the same field scale\nappears as an outward electrostatic pressure. We work the parallel-plate case in\nfull and mark where curvature and fringing make the pressure nonuniform.\n",{"path":22048,"title":22049,"module":22037,"summary":22050},"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Flaplace-boundary-problems","Laplace Boundary Problems","Often the charges are not given, only the conductors and the voltages held on\nthem, and the potential in the empty space between has to be found. There $V$ obeys\nLaplace's equation $\\nabla^2V=0$, and the boundary data alone determine a unique solution.\nWe solve it two ways: separation of variables into boundary-matched modes, whose\nhigher spatial frequencies die away with depth into the domain, and finite-difference\nrelaxation for boundaries no analytic mode fits. Residual norms, boundary error, and\nflux balance tell us when the computed potential and its field can be trusted.\n",{"path":22052,"title":22053,"module":22037,"summary":22054},"\u002Felectricity-and-magnetism\u002Felectric-potential\u002Fcontinuous-charge-potentials","Continuous Charge Potentials","When charge is spread over a line, a surface, or a volume, the sum over point\nsources becomes an integral, $V(\\vec r)=k\\int \\d q\u002F|\\vec r-\\vec r'|$. Because\npotential is a scalar, this integral sidesteps the component algebra the field\nwould force, until the field is actually wanted through $\\vec E=-\\nabla V$. We set\nup the right density element for each geometry, choose a workable reference, handle\nthe integrable singularities that arise when the observation point sits on the\ncharge, and check every result against symmetry, dimensions, and the far-field\nmultipole limit.\n",{"path":22056,"title":22057,"module":22058,"summary":22059},"\u002Felectricity-and-magnetism\u002Fcapacitance\u002Fcapacitance-fundamentals","Capacitance Fundamentals","Capacitance","How much charge must you separate onto two conductors to hold a given voltage between\nthem? That ratio, $C=Q\u002F\\Delta V$, is fixed by the conductor geometry and the medium,\nnot by how much charge is presently stored. We compute it from the field for the\nparallel-plate, isolated-sphere, concentric-sphere, and coaxial geometries, trace how\nsurface charge and boundary conditions set each result, and see where fringing,\nguarding, and stray coupling separate the ideal formula from what a bridge measures.\n",{"path":22061,"title":22062,"module":22058,"summary":22063},"\u002Felectricity-and-magnetism\u002Fcapacitance\u002Fcapacitor-networks","Capacitor Networks","Wire several capacitors together and the source sees one equivalent capacitance — but\nwhich? The answer comes not from how the symbols are drawn but from which conductors\nshare a node: parallel branches hold a common voltage and add, $C_{\\rm eq}=\\sum_iC_i$,\nwhile series branches share a common charge and add reciprocally. We derive both rules\nfrom charge conservation on the floating internal node, then extend the node-charge\nmethod to unequal, precharged, and stray-coupled branches and carry a worked reduction\nthrough to the charge and voltage on every element.\n",{"path":22065,"title":22066,"module":22058,"summary":22067},"\u002Felectricity-and-magnetism\u002Fcapacitance\u002Fcapacitor-energy-and-force","Capacitor Energy and Force","Charging a capacitor takes work, because every increment of charge is pushed through\nthe voltage the earlier charge already established. We total that work three\nequivalent ways, $U=Q^2\u002F(2C)=Q\\Delta V\u002F2=C(\\Delta V)^2\u002F2$, locate it in the field as\na density $u=\\tfrac12\\epsilon_0E^2$, then let the plates move. Differentiating the\nstored energy at fixed charge, or the coenergy at fixed voltage, gives the mechanical\nforce; the two boundaries differ only by the work the source supplies. We work the\nparallel-plate attraction and its electrostatic pressure in full, and follow the same\ngradient into pull-in, tilt, comb drives, and traceable force calibration.\n",{"path":22069,"title":22070,"module":22058,"summary":22071},"\u002Felectricity-and-magnetism\u002Fcapacitance\u002Fdielectric-polarization-and-breakdown","Dielectric Polarization and Breakdown","Slide a dielectric between the plates and the capacitance rises — but why, and how\nhard can you drive it before the insulator fails? Bound charge answers the first:\npolarization $\\vec P$ sets up surface and volume charge that partly cancels the\napplied field, so $\\vec D=\\varepsilon_0\\vec E+\\vec P$ separates what the circuit\ncontrols from what the material contributes. We follow the field across layered\ndielectrics and interfaces, tie permittivity and loss to their frequency dependence,\nand treat dielectric strength as a measured, geometry-dependent limit rather than one\nmaterial number.\n",{"path":22073,"title":22074,"module":22075,"summary":22076},"\u002Felectricity-and-magnetism\u002Fdirect-current-circuits\u002Fcurrent-and-resistance","Current and Resistance","Direct-Current Circuits","What does it mean, physically, for charge to flow, and what sets how hard a wire resists that flow? Current counts charge crossing a surface, $I=\\int\\vec J\\cdot\\d\\vec A$, and traces back to a slow drift of many carriers, $\\vec J=nq\\vec v_d$. We establish when the linear law $V=IR$ actually holds, how resistivity and geometry combine into bulk resistance, why real sources sag under load through their internal resistance, and how the three power forms $P=IV=I^2R=V^2\u002FR$ tie electrical work to heating and component ratings.\n",{"path":22078,"title":22079,"module":22075,"summary":22080},"\u002Felectricity-and-magnetism\u002Fdirect-current-circuits\u002Fkirchhoff-network-analysis","Kirchhoff Network Analysis","Once a circuit has more than one loop, no amount of series-parallel folding will reduce it — you need the two conservation laws written as equations. Kirchhoff's junction law is charge conservation at a node; his loop law is energy conservation around a closed path. We turn a labelled network into a linear system in node voltages or mesh currents, fix the sign conventions so a negative answer just means a reversed arrow, and use power balance as an independent check that the algebra describes the circuit that was actually built.\n",{"path":22082,"title":22083,"module":22075,"summary":22084},"\u002Felectricity-and-magnetism\u002Fdirect-current-circuits\u002Frc-transients","RC Transients","How does a circuit get from one steady state to the next when a capacitor refuses to change its voltage all at once? Because a jump would demand infinite current, an RC circuit slides between states exponentially, with a single time constant $\\tau=RC$ that sets the whole schedule: charging fills as $1-e^{-t\u002F\\tau}$, discharge empties as $e^{-t\u002F\\tau}$. We solve the first-order loop equation, read the response off three numbers — the switch-instant voltage, the final dc voltage, and the Thevenin resistance the capacitor sees — and mark where source and probe resistance shift $\\tau$ or where a second storage element hides a mode a one-$\\tau$ fit misses.\n",{"path":22086,"title":22087,"module":22088,"summary":22089},"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fmagnetic-trajectories","Magnetic Trajectories","Magnetic Field","A charged particle in a magnetic field never speeds up or slows down, yet its path curves relentlessly. We work out why: the magnetic force is always perpendicular to velocity, so it does no work and bends the transverse motion into a circle of radius $r=mv_\\perp\u002F(|q|B)$ while leaving the parallel motion untouched, producing a helix. We derive the cyclotron frequency, show why it is independent of speed until relativity intervenes, and turn the geometry around: a measured curvature reads back a particle's momentum, which is how tracking detectors weigh what they cannot see.\n",{"path":22091,"title":22092,"module":22088,"summary":22093},"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fhall-effect","Hall Effect","Current tells you charge is moving, but not whether the movers are positive or negative, nor how many there are. A magnetic field settles both questions. Push current through a strip in a transverse field and the carriers pile up on one edge until a transverse electric field just balances the magnetic deflection; the sign of the resulting Hall voltage names the carrier's charge and its size counts the carriers per volume. We derive the balance $q\\vec E+q\\vec v_d\\times\\vec B=0$, read off $V_H=IB\u002F(nqt)$, and see why field-and-current reversal is what separates the real Hall signal from the offsets that mimic it.\n",{"path":22095,"title":22096,"module":22088,"summary":22097},"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fmagnetic-force-on-conductors","Magnetic Force on Conductors","A magnet pushes on a current-carrying wire even though the wire is electrically neutral. The reason is that each moving carrier feels the Lorentz force, and those microscopic pushes add up to a force the wire's supports must hold. We sum them into $\\d\\vec F=I\\,\\d\\vec\\ell\\times\\vec B$, collapse it to $\\vec F=I\\vec L\\times\\vec B$ for a straight segment in a uniform field, and see exactly when that shortcut fails and the full path integral is needed. The same law runs backward as a measurement: a force-versus-current slope weighs a magnetic field against a known length.\n",{"path":22099,"title":22100,"module":22088,"summary":22101},"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fmagnetic-dipoles","Magnetic Dipoles","A compass needle turns to point north; a current loop in a field does the same thing, and for the same reason. Both are magnetic dipoles, and a uniform field cannot push a dipole anywhere, only twist it. We package a loop's response into one vector, the magnetic moment $\\vec\\mu=IA\\hat n$, from which torque $\\vec\\tau=\\vec\\mu\\times\\vec B$ and orientation energy $U=-\\vec\\mu\\cdot\\vec B$ both follow. Stable alignment sits at the energy minimum, a field gradient is what it takes to produce a net force $\\vec F=\\nabla(\\vec\\mu\\cdot\\vec B)$, and the same moment reappears whenever anything from an electron to a planet acts magnetic.\n",{"path":22103,"title":22104,"module":22088,"summary":22105},"\u002Felectricity-and-magnetism\u002Fmagnetic-field\u002Fmass-spectrometry","Mass Spectrometry","To weigh a single atom you cannot use a scale, so you use a magnetic field instead. A charged ion of unknown mass bends in a field by an amount that depends on its momentum and charge, so if every ion enters with the same velocity, its landing position reads off its mass-to-charge ratio directly. We build the instrument in two stages: crossed electric and magnetic fields that pass only ions with $v=E\u002FB$, and a magnetic sector that bends the survivors along $r=mv\u002F(|q|B)$. Then we ask what blurs a spectral line and how reference ions turn a position into a calibrated mass.\n",{"path":22107,"title":22108,"module":22109,"summary":22110},"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fmoving-charge-fields","Moving-Charge Fields","Magnetic Sources","Every magnetic field comes from charge in motion, and the simplest source is a single point charge drifting past. We work out the field it produces — normal to both the velocity and the line of sight, falling off as the inverse square — and read off why it vanishes straight ahead of the charge and peaks broadside. Summing many such charges is the bridge to steady currents, valid while speeds stay far below $c$ and the motion changes little during the time its field takes to propagate outward.\n",{"path":22112,"title":22113,"module":22109,"summary":22114},"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fbiot-savart-law","Biot–Savart Law","A steady current is a continuous stream of current elements, and the Biot–Savart law hands each one a magnetic contribution — a right-hand cross product that falls off as the inverse square of distance. Summing the contributions along a conductor is a vector line integral, which we carry out for the straight wire to get the endpoint-angle formula. The infinite-wire field $B=\\mu_0 I\u002F2\\pi s$ falls out as the limit where both ends recede, and we mark how fast a finite wire departs from it and when a thin-filament model is safe.\n",{"path":22116,"title":22117,"module":22109,"summary":22118},"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fcircular-current-loops","Circular Current Loops","A ring of current is the simplest source with a well-defined magnetic axis, and it is the building block of every coil and electromagnet. Symmetry kills the transverse Biot–Savart contributions along that axis and leaves a single clean integral; we evaluate it to get $B_z=\\mu_0 I R^2\u002F[2(R^2+z^2)^{3\u002F2}]$, read off the centre field $\\mu_0 I\u002F2R$, and watch it fall into the $1\u002Fz^3$ tail of a magnetic dipole far away. Stacking turns just adds their axial contributions, which is what makes a solenoid out of a pile of loops.\n",{"path":22120,"title":22121,"module":22109,"summary":22122},"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Famperes-law","Ampère’s Law","When a current arrangement is symmetric enough, the Biot–Savart integral is overkill: Ampère's law, $\\oint_C\\vec B\\cdot\\d\\vec\\ell=\\mu_0 I_{\\rm enc}$, gets the field from a single line of reasoning about how much current a loop encloses. We see why the law holds for any steady current, then use cylindrical, planar, and toroidal symmetry to turn the circulation into simple algebra — the field inside and outside a wire, an infinite sheet, a solenoid, and a toroid. We also mark the catch: without symmetry the law still holds but no longer hands you the field pointwise.\n",{"path":22124,"title":22125,"module":22109,"summary":22126},"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fgauss-law-for-magnetism","Gauss’s Law for Magnetism","Electric field lines start and end on charges; magnetic field lines do neither, because no one has ever found an isolated magnetic pole. That single experimental fact is Gauss's law for magnetism: the flux of $\\vec B$ through any closed surface is zero, $\\oint\\vec B\\cdot\\d\\vec A=0$, or in differential form $\\nabla\\cdot\\vec B=0$. We work through what it says — every field line that enters a closed surface must leave it, so field lines close on themselves — and, just as important, what it does not say, since flux through an open surface is generally nonzero.\n",{"path":22128,"title":22129,"module":22109,"summary":22130},"\u002Felectricity-and-magnetism\u002Fmagnetic-sources\u002Fmagnetic-materials","Magnetic Materials","Put matter in a magnetic field and its atoms respond, each acting as a tiny current loop; the aligned moments per unit volume are the magnetization $\\vec M$, whose bound currents add to the field. Separating what we control (the free current) from what the material supplies leads to $\\vec H$ and the relation $\\vec B=\\mu_0(\\vec H+\\vec M)$. We sort materials into diamagnets, paramagnets, and ferromagnets by how $\\vec M$ answers, follow a ferromagnet around its hysteresis loop, and see why the loop's area is the energy dissipated per cycle and why a sample's shape changes the field it actually feels.\n",{"path":22132,"title":22133,"module":22134,"summary":22135},"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Fmagnetic-flux","Magnetic Flux","Electromagnetic Induction","A magnetic field threading a loop collapses to one signed number, the flux, and every induced voltage in this module turns out to be a rate of change of that number — so defining the flux and its sign comes first. We define it as the surface integral of $\\vec B$ over an oriented surface, reduce it to $BA\\cos\\theta$ for a uniform field on a flat loop, and carry the flux linkage $N\\Phi_B$ of a coil. The chosen normal fixes the sign; reversing it flips the sign without touching the field. Nonuniform fields and curved surfaces force the integral, so we also build the numerical estimate and the checks that separate a reliable value from a nominal field-times-area product.\n",{"path":22137,"title":22138,"module":22134,"summary":22139},"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Ffaradays-law","Faraday's Law","Move a magnet toward a coil, or ramp the current in a nearby circuit, and a voltage appears with no battery in sight. Faraday's law names the cause: the emf around a loop equals minus the rate of change of the magnetic flux through it, so any change of field, area, orientation, or position that alters the flux drives an emf. We separate the emf, which lives around the boundary whether or not current can flow, from the current that follows only when the path is closed; fix the single sign convention that ties flux to loop orientation; and read the emf off rotating coils and off flux sampled at discrete times.\n",{"path":22141,"title":22142,"module":22134,"summary":22143},"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Flenzs-law","Lenz's Law","The minus sign in Faraday's law is not decoration: it decides which way the induced current flows, and it always chooses the direction that fights the change that produced it. Lenz's law reads that sign off energy conservation — a current that aided the change would be free energy — and turns it into a repeatable procedure. We fix a surface normal and a positive loop direction so the sign is calculable, then work through approaching magnets, expanding loops, coupled coils, and rotating generators, using mechanical work and Joule heating as an independent check on every direction we draw.\n",{"path":22145,"title":22146,"module":22134,"summary":22147},"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Fmotional-emf","Motional EMF","Push a wire through a magnetic field and its free charges feel a sideways magnetic force that piles them up at the ends — a battery made of motion. Motional emf is that effect: the work per unit charge a moving conductor supplies is the line integral of $\\vec v\\times\\vec B$ along it, which for a rod moving perpendicular to both its length and the field collapses to $B\\ell v$. We chase where the energy comes from — the hand or motor fighting the magnetic drag, never the magnetic force itself — solve the sliding-rail circuit from both flux and carrier forces, and carry the idea into rotating rods, homopolar disks, generators, and the back emf of a motor.\n",{"path":22149,"title":22150,"module":22134,"summary":22151},"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Feddy-currents","Eddy Currents","A wire carries current along one path; a solid block of metal offers a continuum of them, and any changing flux threading that block sets charge circulating in closed loops it chooses for itself. We ask what those eddy currents do — where they heat, where they drag, and how Lenz's law fixes their direction — and why the same circulation is a feature in an induction furnace and a loss to be suppressed in a transformer core. From a representative-loop estimate we get the scaling (heating grows with the square of frequency and flux rate) and the two design levers, lamination and resistivity, that break the paths a solid conductor would otherwise hand the current.\n",{"path":22153,"title":22154,"module":22134,"summary":22155},"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Fself-inductance","Self-Inductance","A coil resists changes to its own current. Drive current through it and the flux it produces threads its own turns; change that current and Faraday's law turns the coil against the source with a back emf $\\mathcal E_L=-L\\,\\d I\u002F\\d t$. We define self-inductance as the flux linkage per ampere fixed by winding and core geometry, derive the long-solenoid value $L=\\mu_0 N^2A\u002F\\ell$, and follow the consequence that dominates circuits: because a finite voltage can only sustain a finite $\\d I\u002F\\d t$, an inductor's current cannot jump — which is why opening a switch on a live coil throws a spark.\n",{"path":22157,"title":22158,"module":22134,"summary":22159},"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Fmagnetic-energy","Magnetic Energy","Building current in a coil means working against its back emf, and that work does not vanish — it sits in the magnetic field as recoverable energy $U_B=\\tfrac12LI^2$, spread through space at density $u_B=B^2\u002F(2\\mu_0)$. We derive both forms, show they agree for a solenoid, and read a force out of the same energy: an armature is pulled toward higher inductance, and $B^2\u002F(2\\mu_0)$ doubles as a magnetic pressure. The lesson closes on the accounting a real switching event demands, where recoverable energy, copper heating, core loss, and clamp dissipation must balance a single ledger.\n",{"path":22161,"title":22162,"module":22134,"summary":22163},"\u002Felectricity-and-magnetism\u002Felectromagnetic-induction\u002Frl-circuits","RL Circuits","Put a resistor and an inductor in series and the current cannot switch on or off at will: it climbs to $V_0\u002FR$ and falls away exponentially on a single time scale $\\tau=L\u002FR$ set by how much flux the coil hoards against how fast the resistor bleeds it. We solve the turn-on and turn-off, then confront the practical sting — because the coil's current refuses to stop instantly, breaking its path throws up a large voltage, which is why real inductive circuits carry freewheel diodes and clamps that trade voltage stress against how quickly the current dies.\n",{"path":22165,"title":22166,"module":22167,"summary":22168},"\u002Felectricity-and-magnetism\u002Falternating-current\u002Fac-fundamentals","AC Fundamentals","Alternating Current","A wall socket delivers a voltage that averages to zero over each cycle, yet it still heats a filament and runs a motor. The resolution is that dissipation follows the mean of the square, not the mean, so we define the root-mean-square value that makes an alternating source the equal of a DC one for resistive heating. We show a sinusoid's RMS is its peak divided by $\\sqrt2$, work out the average power an ideal resistor draws when its current stays in phase with the applied voltage, and separate the peak, average, and RMS descriptions that a single number cannot combine.\n",{"path":22170,"title":22171,"module":22167,"summary":22172},"\u002Felectricity-and-magnetism\u002Falternating-current\u002Freactance","Reactance","A resistor obeys Ohm's law instant by instant, but a capacitor responds to how fast its voltage changes and an inductor to how fast its current changes. Under a steady sinusoid that rate-dependence collapses to a fixed quarter-cycle phase shift and a frequency-dependent amplitude ratio, the reactance. We derive $X_C=1\u002F(\\omega C)$ and $X_L=\\omega L$, adopt phasors to turn the defining derivatives into multiplication by $j\\omega$ so a single complex impedance carries amplitude and phase together, and track the energy an ideal reactance stores and returns without dissipating it. Real windings and dielectrics add loss, leakage, and self-resonance that bound where the ideal formulas hold.\n",{"path":22174,"title":22175,"module":22167,"summary":22176},"\u002Felectricity-and-magnetism\u002Falternating-current\u002Frlc-resonance","RLC Resonance","Put a resistor, inductor, and capacitor in one loop and their reactances work against each other: inductive reactance grows with frequency while capacitive reactance shrinks, and at one frequency they cancel exactly. There the branch looks purely resistive, the current peaks, and the inductor and capacitor voltages can swing far above the source. We locate that resonance at $\\omega_0=1\u002F\\sqrt{LC}$, measure how sharp the peak is with the quality factor $Q=\\omega_0L\u002FR$, tie its half-power bandwidth $R\u002FL$ to the ringdown of the unforced circuit, and read the same poles off as bandpass and peaked filters at the R, L, or C terminals.\n",{"path":22178,"title":22179,"module":22167,"summary":22180},"\u002Felectricity-and-magnetism\u002Falternating-current\u002Fac-power","AC Power","Multiply an AC load's RMS voltage by its RMS current and you get an answer in volt-amperes that the wiring must carry, but not in general the watts the load consumes. The phase between voltage and current splits that product into a part that does net work and a part that merely sloshes energy back and forth. We derive the average power $P=V_{\\rm rms}I_{\\rm rms}\\cos\\phi$, package amplitude and phase into complex power $S=P+jQ$ so that real, reactive, and apparent power form one right triangle, and see why a harmonic-rich current forces the time-domain definition $P=\\langle vi\\rangle$ in place of a single phase angle.\n",{"path":22182,"title":22183,"module":22167,"summary":22184},"\u002Felectricity-and-magnetism\u002Falternating-current\u002Ftransformers","Transformers","Two coils sharing an iron core exchange no charge, yet a changing current in one drives a voltage in the other, and the ratio of their turns sets how voltage and current trade off between the windings. That lets a transformer step a voltage up or down, isolate two circuits, and make a load look larger or smaller to the source by the square of the turns ratio. We build the ideal ratio element from Faraday's law and the dot convention, derive the reflected-impedance rule, then add the winding resistance, leakage, magnetizing current, and core loss that turn the ideal ratios into real regulation, efficiency, and a bounded voltage-frequency range.\n",{"path":22186,"title":22187,"module":22188,"summary":22189},"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Fdisplacement-current","Displacement Current","Maxwell’s Equations and Electromagnetic Waves","Ampère's law asks for the current through a surface bounded by a loop, but a charging capacitor breaks it: slide the surface off the wire and into the gap and the enclosed conduction current drops to zero, while the magnetic field around the loop plainly does not. Maxwell's repair is to count a changing electric flux as itself a source of magnetic circulation. We derive the displacement-current term $\\varepsilon_0\\,\\d\\Phi_E\u002F\\d t$, show that charge continuity demands it, compute the magnetic field it produces inside a charging capacitor, and see how it closes the Ampère–Maxwell law so that electric and magnetic fields can sustain one another as a wave.\n",{"path":22191,"title":22192,"module":22188,"summary":22193},"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Felectromagnetic-waves","Electromagnetic Waves","Once a changing electric flux can drive a magnetic field, the two curl laws feed each other: a disturbance in one regenerates the other, and the pair walks off through empty space with no medium holding it up. We take the curl of Faraday's law, land on a wave equation whose speed is fixed entirely by $\\mu_0$ and $\\varepsilon_0$, and find that $c=1\u002F\\sqrt{\\mu_0\\varepsilon_0}$ falls out of purely electric and magnetic constants. The plane-wave solution then fixes the geometry — $\\vec E$, $\\vec B$, and the propagation direction mutually perpendicular, oscillating in phase, with amplitudes locked at $E=cB$ — a set of independent predictions any real measurement must meet at once.\n",{"path":22195,"title":22196,"module":22188,"summary":22197},"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Felectromagnetic-momentum","Electromagnetic Momentum","A light beam carries no mass, yet it pushes: shine it on a surface and the surface feels a force. We trace that force back to the fields, which store energy with density $\\varepsilon_0E^2$ and carry it along the Poynting vector $\\vec S=\\vec E\\times\\vec B\u002F\\mu_0$. Because that energy also carries momentum $U\u002Fc$, an absorbed beam presses with $I\u002Fc$ and a mirror with $2I\u002Fc$. We derive the Poynting theorem as local energy conservation, tie intensity to field amplitude, and work the momentum balance carefully enough that oblique incidence, partial reflection, and finite beams all drop out of one accounting.\n",{"path":22199,"title":22200,"module":22188,"summary":22201},"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Fdipole-radiation","Dipole Radiation","Only accelerating charge radiates, and the simplest accelerator is a charge sloshing back and forth: an oscillating electric dipole. We work out the field it throws off, keeping the part that survives to large distance — the $1\u002Fr$ radiation field whose intensity goes as $\\sin^2\\theta\u002Fr^2$, zero along the dipole axis and strongest broadside. From it follow the $\\omega^4$ scaling of total radiated power, radiation resistance as the feed's view of that escaping power, and, through reciprocity, the fact that a good transmitter receives well in the same directions. The near-zone terms that fall off faster carry no net power, and we mark carefully where each description is allowed to be used.\n",{"path":22203,"title":22204,"module":22188,"summary":22205},"\u002Felectricity-and-magnetism\u002Fmaxwell-electromagnetic-waves\u002Fpolarization","Polarization","A plane wave still leaves one thing free: which way its electric field points as it oscillates. That freedom is polarization, set entirely by the relative amplitude and phase of the two transverse field components — in phase gives a line, equal amplitudes a quarter cycle apart give a circle, everything else an ellipse. We work out how a linear analyzer reads a state through Malus's law $I=I_0\\cos^2\\theta$, why that scan alone cannot tell circular light from unpolarized, and how a quarter-wave plate plus a few analyzer settings recover the full Stokes vector and the degree of polarization.\n",{"path":22207,"title":22208,"module":22209,"summary":22210},"\u002Felectricity-and-magnetism\u002Foptics\u002Freflection-and-refraction","Reflection and Refraction","Geometrical Optics","Light meeting a boundary between two transparent media splits into a reflected ray and a bent transmitted one, and predicting where those rays go is the whole starting point of geometrical optics. Fixing one convention — every angle measured from the surface normal — we get reflection's equal angles and derive Snell's law $n_1\\sin\\theta_1=n_2\\sin\\theta_2$ from wavefront timing. That single relation, applied once or twice, yields the critical angle and total internal reflection, prism deviation, the lateral shift through a window, apparent depth, and a fiber's acceptance cone; a wavelength-dependent index then adds dispersion. We mark throughout where the ray picture is trustworthy: feature sizes large against the wavelength and clean interface geometry.\n",{"path":22212,"title":22213,"module":22209,"summary":22214},"\u002Felectricity-and-magnetism\u002Foptics\u002Fthin-lenses","Thin Lenses","A lens gathers the light spreading from one point back onto another, and a single paraxial relation $1\u002Fs+1\u002Fs'=1\u002Ff$ predicts where that image lands and how large it is. We collapse two refractions into one bending plane, read image position and orientation off the three principal rays, and trace focal length back to glass and curvature through the lensmaker equation. Sign conventions carry the physics here — they separate real from virtual images and upright from inverted — so we drill them before chaining lenses in sequence and in contact. The lesson ends on how focal length is actually measured on a bench, and where finite thickness, aperture, and dispersion break the thin-lens picture.\n",{"path":22216,"title":22217,"module":22209,"summary":22218},"\u002Felectricity-and-magnetism\u002Foptics\u002Fspherical-mirrors","Spherical Mirrors","Curve a mirror and it stops merely reflecting an image and starts forming one: the same $1\u002Fs+1\u002Fs'=1\u002Ff$ that governs lenses reappears, now with $f=R\u002F2$ and reflected rays and object sharing one side of the glass. We derive the mirror equation from the reflection geometry of a single paraxial ray, then let signed distances do the sorting — real inverted images on the near branch, virtual upright ones behind the surface — and check the concave, convex, and plane-mirror limits against each other. The second half turns to how focal length is actually measured on a bench, by finite conjugates, distant targets, return imaging, and sagitta, and to the aperture and off-axis aberrations the single paraxial focus cannot capture.\n",{"path":22220,"title":22221,"module":6,"summary":6},"\u002Felectricity-and-magnetism","Electricity & Magnetism",{"path":22223,"title":22224,"module":22225,"summary":22226},"\u002Flinear-algebra\u002Flinear-systems\u002Fsystems-and-echelon-forms","Systems of Linear Equations and Row Reduction","Linear Equations in Linear Algebra","A linear system is a finite set of linear equations in shared variables. Elementary row operations rewrite it without changing its solution set, and reducing the augmented matrix to echelon form decides both existence and uniqueness. Pivot positions say whether the solution set is empty, a single point, or infinite.\n",{"path":22228,"title":22229,"module":22225,"summary":22230},"\u002Flinear-algebra\u002Flinear-systems\u002Fvector-and-matrix-equations","Vector Equations and the Matrix Equation Ax = b","The same linear system reads three equivalent ways: a system of equations, a vector equation asking whether b is a linear combination of fixed vectors, and a matrix equation Ax = b. Ax is the linear combination of A's columns weighted by x, so consistency for a given b means b lies in the span of the columns, and consistency for every b means the columns span all of R^m.\n",{"path":22232,"title":22233,"module":22225,"summary":22234},"\u002Flinear-algebra\u002Flinear-systems\u002Fsolution-sets-and-applications","Solution Sets and Applied Linear Systems","A homogeneous system Ax = 0 has a solution set that is a span through the origin; a consistent Ax = b has that same span translated by any one particular solution. Parametric vector form writes both explicitly. The structure shows up in applied systems with many solutions: equilibrium prices, balanced chemical reactions, network flows, weight-loss diets, and migration models.\n",{"path":22236,"title":22237,"module":22225,"summary":22238},"\u002Flinear-algebra\u002Flinear-systems\u002Flinear-independence","Linear Independence","A set of vectors is linearly independent when the only linear combination equal to zero is the trivial one; otherwise a dependence relation writes one vector in terms of the others. For the columns of A the question becomes whether Ax = 0 has only the trivial solution — a pivot in every column. Counting pivots settles independence, and any set with more vectors than entries is automatically dependent.\n",{"path":22240,"title":22241,"module":22225,"summary":22242},"\u002Flinear-algebra\u002Flinear-systems\u002Flinear-transformations","Linear Transformations and Their Matrices","Reading A as an action rather than an array, x maps to Ax is a transformation from R^n to R^m. The ones that preserve addition and scalar multiplication are the linear transformations, and every one is x maps to Ax for a unique standard matrix whose columns are the images of the standard basis vectors. Onto and one-to-one translate into the span and independence of those columns.\n",{"path":22244,"title":22245,"module":22246,"summary":22247},"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fmatrix-operations","Matrix Operations","Matrix Algebra","Matrices add and scale entrywise, but their product is defined so that multiplication corresponds to composition of linear maps: the columns of AB are A applied to the columns of B. From that requirement follow the row-column rule, the algebra of products (associative and distributive but not commutative), powers, and the transpose.\n",{"path":22249,"title":22250,"module":22246,"summary":22251},"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fmatrix-inverse-and-invertibility","The Inverse and the Invertible Matrix Theorem","The inverse of a square matrix is the matrix analogue of a reciprocal, defined by AA⁻¹ = I. A closed form settles the 2×2 case; the Gauss–Jordan algorithm row reduces [A | I] to [I | A⁻¹] in general; and elementary matrices record single row operations. The Invertible Matrix Theorem collects a dozen equivalent conditions for invertibility into one statement.\n",{"path":22253,"title":22254,"module":22246,"summary":22255},"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fpartitioned-matrices-and-lu","Block Matrices and the LU Factorization","Partitioning a matrix into blocks lets sums, products, and inverses be computed block by block, as if the submatrices were scalars. Block structure also underlies the LU factorization A = LU, which splits solving Ax = b into two fast triangular solves and repays the cost whenever many systems share one coefficient matrix.\n",{"path":22257,"title":22258,"module":22246,"summary":22259},"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fsubspaces-dimension-rank","Subspaces of Rⁿ, Dimension, and Rank","A subspace is a set closed under addition and scalar multiplication. Every matrix carries two: the column space of all attainable outputs Ax, and the null space of all solutions of Ax = 0. A basis measures each with a minimal spanning set, dimension counts it, and the Rank Theorem ties pivots and free variables together as rank + nullity = n.\n",{"path":22261,"title":22262,"module":22246,"summary":22263},"\u002Flinear-algebra\u002Fmatrix-algebra\u002Fapplications-leontief-and-graphics","Applications: Leontief Economics and Computer Graphics","The Leontief input–output model balances an economy through (I − C)x = d and expands the inverse as a geometric series in the consumption matrix. Computer graphics moves figures with matrix products, using homogeneous coordinates so that translation and perspective projection become matrix multiplications too.\n",{"path":22265,"title":22266,"module":22267,"summary":22268},"\u002Flinear-algebra\u002Fdeterminants\u002Fdeterminants-and-cofactors","Introduction to Determinants","Determinants","The determinant of a square matrix is defined recursively by cofactor expansion: an n-by-n determinant is a signed sum of (n-1)-by-(n-1) determinants built from the first row. The expansion can equally run along any row or down any column, and a triangular matrix has determinant equal to the product of its diagonal.\n",{"path":22270,"title":22271,"module":22267,"summary":22272},"\u002Flinear-algebra\u002Fdeterminants\u002Fproperties-of-determinants","Properties of Determinants","Row operations act on the determinant in three predictable ways, and this turns row reduction into a fast algorithm: the determinant is the product of the pivots times a sign for the interchanges. The same properties yield the invertibility test det A is nonzero, the transpose identity, and the multiplicative law det(AB) equals det A times det B.\n",{"path":22274,"title":22275,"module":22267,"summary":22276},"\u002Flinear-algebra\u002Fdeterminants\u002Fcramer-volume-and-area","Cramer's Rule, Volume, and Linear Transformations","Cramer's rule writes each unknown of an invertible system as a ratio of determinants, and the same idea gives a closed formula for the inverse through the adjugate. Geometrically the absolute determinant is the area of the parallelogram or the volume of the parallelepiped spanned by the columns, so a linear map scales every region's measure by that factor.\n",{"path":22278,"title":22279,"module":22280,"summary":22281},"\u002Flinear-algebra\u002Fvector-spaces\u002Fvector-spaces-and-subspaces","Vector Spaces and Subspaces","Vector Spaces","A vector space is any set closed under addition and scalar multiplication that obeys ten algebraic axioms. The same axioms that govern arrows in the plane govern polynomials, functions, matrices, and infinite signals, so one theory covers them all. A subspace is a subset that is a vector space in its own right, tested by three conditions, and the span of any set of vectors is the smallest subspace containing them.\n",{"path":22283,"title":22284,"module":22280,"summary":22285},"\u002Flinear-algebra\u002Fvector-spaces\u002Fnull-and-column-spaces","Null Spaces, Column Spaces, and Linear Transformations","Two subspaces sit inside every matrix. The null space collects all solutions of $Ax = 0$ and lives in the domain; the column space collects every attainable $Ax$ and lives in the codomain. One is defined implicitly by a condition, the other explicitly by a spanning set, and the same pair appears for an abstract linear transformation as its kernel and range.\n",{"path":22287,"title":22288,"module":22280,"summary":22289},"\u002Flinear-algebra\u002Fvector-spaces\u002Fbases-and-independent-sets","Linearly Independent Sets and Bases","A basis is a spanning set with no redundancy: linearly independent and still large enough to reach every vector. The spanning-set theorem shows any spanning set can be trimmed to a basis by discarding dependent vectors, and the pivot columns of a matrix give a basis for its column space. Independence and spanning are defined for abstract spaces exactly as in $\\mathbb{R}^n$.\n",{"path":22291,"title":22292,"module":22280,"summary":22293},"\u002Flinear-algebra\u002Fvector-spaces\u002Fcoordinate-systems","Coordinate Systems","Fixing a basis assigns every vector a unique list of coordinates, turning an abstract space into $\\mathbb{R}^n$. The coordinate mapping is a one-to-one linear transformation onto $\\mathbb{R}^n$ — an isomorphism — so any $n$-dimensional space is indistinguishable from $\\mathbb{R}^n$ as far as vector-space computations go. In $\\mathbb{R}^n$ the change-of-coordinates matrix $P_B$ and its inverse convert between basis coordinates and standard coordinates.\n",{"path":22295,"title":22296,"module":22280,"summary":22297},"\u002Flinear-algebra\u002Fvector-spaces\u002Fdimension-and-rank","The Dimension of a Vector Space and Rank","Every basis of a space has the same number of vectors, and that number is the dimension. Rank is the dimension of the column space, equal to the dimension of the row space and to the number of pivots. The Rank Theorem, rank plus nullity equals the number of columns, ties the four fundamental subspaces of a matrix together and adds six lines to the Invertible Matrix Theorem.\n",{"path":22299,"title":22300,"module":22280,"summary":22301},"\u002Flinear-algebra\u002Fvector-spaces\u002Fchange-of-basis","Change of Basis","Two bases give the same vector two different coordinate vectors, and a single invertible matrix converts between them. Its columns are the coordinate vectors of the old basis expressed in the new one, and its inverse reverses the conversion. In $\\mathbb{R}^n$ the change-of-coordinates matrix between two bases is found by one row reduction.\n",{"path":22303,"title":22304,"module":22280,"summary":22305},"\u002Flinear-algebra\u002Fvector-spaces\u002Fdifference-equations-and-markov","Applications: Difference Equations and Markov Chains","The solutions of an nth-order linear difference equation form an $n$-dimensional vector space, so finding $n$ independent solutions gives them all. A Markov chain evolves a probability distribution by repeated multiplication by a stochastic matrix, and a regular chain converges to a unique steady-state vector fixed by that matrix. Both applications turn a dynamic process into a subspace or a fixed-point question.\n",{"path":22307,"title":22308,"module":22309,"summary":22310},"\u002Flinear-algebra\u002Feigenvalues\u002Feigenvectors-and-eigenvalues","Eigenvectors and Eigenvalues","Eigenvalues and Eigenvectors","An eigenvector of a square matrix is a nonzero vector the matrix only stretches; its eigenvalue is the stretch factor. The eigenspace of an eigenvalue is the null space of A minus lambda times the identity, the eigenvalues of a triangular matrix are its diagonal entries, and eigenvectors for distinct eigenvalues are linearly independent.\n",{"path":22312,"title":22313,"module":22309,"summary":22314},"\u002Flinear-algebra\u002Feigenvalues\u002Fthe-characteristic-equation","The Characteristic Equation","The eigenvalues of a matrix are the roots of its characteristic polynomial det(A minus lambda I). This degree-n polynomial carries an algebraic multiplicity at each repeated root, a nonzero determinant is equivalent to zero not being an eigenvalue, and similar matrices share a characteristic polynomial and hence the same eigenvalues.\n",{"path":22316,"title":22317,"module":22309,"summary":22318},"\u002Flinear-algebra\u002Feigenvalues\u002Fdiagonalization","Diagonalization","A matrix is diagonalizable when it factors as A equals P D P inverse with D diagonal, which happens exactly when it has n linearly independent eigenvectors. The factorization computes matrix powers cheaply, distinct eigenvalues guarantee it, and a repeated eigenvalue permits it only when its eigenspace dimension equals its multiplicity.\n",{"path":22320,"title":22321,"module":22309,"summary":22322},"\u002Flinear-algebra\u002Feigenvalues\u002Feigenvectors-and-linear-transformations","Eigenvectors and Linear Transformations","Every linear transformation between finite-dimensional spaces has a matrix relative to chosen bases, built from the coordinate vectors of the images of the basis vectors. For a map from a space to itself, an eigenvector basis makes that matrix diagonal, and that change of basis is diagonalization; the matrices similar to A are the representations of the map in every basis.\n",{"path":22324,"title":22325,"module":22309,"summary":22326},"\u002Flinear-algebra\u002Feigenvalues\u002Fcomplex-eigenvalues","Complex Eigenvalues","A real matrix with no real eigenvalues still has complex ones, occurring in conjugate pairs. A real 2-by-2 matrix with eigenvalue a plus b i is similar to a rotation-scaling matrix, whose rotation angle is the argument of the eigenvalue and whose scale factor is its modulus; the modulus decides whether the trajectories close up, spiral in, or spiral out.\n",{"path":22328,"title":22329,"module":22309,"summary":22330},"\u002Flinear-algebra\u002Feigenvalues\u002Fdynamical-systems","Discrete and Continuous Dynamical Systems","Eigenvalues govern the long-term behavior of a system that evolves by x becomes A x or by x prime equals A x. An eigenvector basis decouples both kinds of system into independent scalar equations; the eigenvalues then classify the origin as attractor, repeller, saddle, or spiral, and the dominant eigenpair fixes the growth rate and limiting direction.\n",{"path":22332,"title":22333,"module":22309,"summary":22334},"\u002Flinear-algebra\u002Feigenvalues\u002Fpower-method","Iterative Estimates for Eigenvalues","When only a numerical eigenvalue is needed, iteration is preferred over the characteristic polynomial. The power method repeatedly multiplies by A to converge on the dominant eigenvalue and its eigenvector; the Rayleigh quotient sharpens the estimate for symmetric matrices; and the inverse power method targets any eigenvalue near a known guess.\n",{"path":22336,"title":22337,"module":22338,"summary":22339},"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Finner-product-length-orthogonality","Inner Product, Length, and Orthogonality","Orthogonality and Least Squares","The dot product turns the algebra of vectors in R^n into geometry: length, distance, and perpendicularity. The inner product yields the norm, the Pythagorean theorem, and the orthogonal complement, and the null space of a matrix is the orthogonal complement of its row space.\n",{"path":22341,"title":22342,"module":22338,"summary":22343},"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Forthogonal-sets-and-projections","Orthogonal Sets and Orthogonal Projections","An orthogonal basis makes coordinates trivial: each weight is a single dot product, no linear system required. Orthogonal and orthonormal bases give a direct projection formula onto a line and onto a subspace, the orthogonal decomposition and best-approximation theorems, and the matrix form U U-transpose of a projection.\n",{"path":22345,"title":22346,"module":22338,"summary":22347},"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Fgram-schmidt-and-qr","The Gram-Schmidt Process and QR Factorization","Gram-Schmidt turns any basis into an orthogonal one by repeatedly subtracting off projections onto the span already built. Normalizing the result and recording the coefficients factors the matrix as A = QR, with Q orthonormal and R upper triangular, the factorization behind stable least-squares and eigenvalue algorithms.\n",{"path":22349,"title":22350,"module":22338,"summary":22351},"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Fleast-squares-problems","Least-Squares Problems","When Ax = b has no solution, the least-squares solution makes Ax as close to b as possible. The closest Ax is the projection of b onto the column space, and the vector that produces it solves the normal equations A-transpose A x = A-transpose b. Uniqueness, the residual error, and the stabler QR route follow.\n",{"path":22353,"title":22354,"module":22338,"summary":22355},"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Fleast-squares-applications","Applications to Linear Models","Curve fitting is a least-squares problem in statistical notation. The least-squares line, polynomial fits, and multiple regression all reduce to X beta = y with a design matrix X built from the data, solved by the same normal equations.\n",{"path":22357,"title":22358,"module":22338,"summary":22359},"\u002Flinear-algebra\u002Forthogonality-least-squares\u002Finner-product-spaces","Inner Product Spaces","Promoting the four properties of the dot product to axioms defines an inner product on any vector space, including spaces of functions. Length, distance, orthogonality, Gram-Schmidt, and best approximation all carry over, along with the Cauchy-Schwarz and triangle inequalities and the integral inner product behind Fourier approximation.\n",{"path":22361,"title":22362,"module":22363,"summary":22364},"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fdiagonalizing-symmetric-matrices","Diagonalization of Symmetric Matrices","Symmetric Matrices, Quadratic Forms, and the SVD","A symmetric matrix is one that equals its own transpose. Every such matrix can be diagonalized by an orthogonal change of basis, A = PDPᵀ, with real eigenvalues and perpendicular eigenvectors. This is the Spectral Theorem, and it rewrites A as a weighted sum of rank-one projections onto its eigenvectors.\n",{"path":22366,"title":22367,"module":22363,"summary":22368},"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fquadratic-forms","Quadratic Forms","A quadratic form xᵀAx is the second-degree analogue of a linear map, attached to a symmetric matrix A. Orthogonal diagonalization changes variables to the eigenbasis, removing all cross-terms and rotating the form into standard position. The signs of the eigenvalues then classify it as definite or indefinite.\n",{"path":22370,"title":22371,"module":22363,"summary":22372},"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fconstrained-optimization","Constrained Optimization","Maximizing a quadratic form xᵀAx over the unit sphere has an exact answer: the maximum is the largest eigenvalue of A, attained at its eigenvector, and the minimum is the smallest eigenvalue. Adding orthogonality constraints peels off the eigenvalues in order, characterizing the whole spectrum by optimization.\n",{"path":22374,"title":22375,"module":22363,"summary":22376},"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fsingular-value-decomposition","The Singular Value Decomposition","The singular value decomposition factors any m×n matrix as A = UΣVᵀ, with orthogonal U and V and a nonnegative diagonal Σ of singular values. The singular values are the square roots of the eigenvalues of AᵀA, and they describe the matrix geometrically as a rotation, an axiswise stretch, and another rotation, exposing rank, the four fundamental subspaces, and a best low-rank approximation.\n",{"path":22378,"title":22379,"module":22363,"summary":22380},"\u002Flinear-algebra\u002Fsymmetric-quadratic-svd\u002Fsvd-applications-pca-imaging","Applications: Image Processing and Statistics","Principal component analysis diagonalizes the covariance matrix of a data set, producing uncorrelated variables ordered by variance. The leading components capture most of the variation, which reduces dimension, compresses images through low-rank SVD approximation, and connects directly to the singular values of the data matrix.\n",{"path":22382,"title":22383,"module":22384,"summary":22385},"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fnumerical-thinking-and-matrix-computation","Numerical Thinking and Matrix Computation","Numerical Linear Algebra","Numerical analysis builds efficient discrete algorithms for continuous problems, and its cost is dominated as much by memory traffic as by arithmetic. Block matrix calculus, flop counts, and the BLAS efficiency ratio fix the cost model; triangular and unitary matrices are the two computational building blocks every factorization rests on.\n",{"path":22387,"title":22388,"module":22384,"summary":22389},"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Flu-and-cholesky","LU and Cholesky Factorization in Practice","Gaussian elimination, read as a factorization A = LU, turns a linear system into two triangular solves. A single near-zero pivot wrecks it, so partial pivoting reorders rows to pick the largest available pivot and makes the method work for every invertible matrix. For symmetric positive-definite systems, Cholesky halves the cost and needs no pivoting.\n",{"path":22391,"title":22392,"module":22384,"summary":22393},"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fconditioning-and-floating-point","Conditioning and Floating-Point Arithmetic","A problem's condition number measures how much its answer moves when its data is perturbed, independent of any algorithm. Subtraction is ill-conditioned under cancellation, and for a linear system the amplifier is the matrix condition number κ(A). Floating-point arithmetic supplies the perturbation: every real number is rounded to within a relative machine precision, so even perfect computation inherits an error of order κ times the unit roundoff.\n",{"path":22395,"title":22396,"module":22384,"summary":22397},"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fstability-and-error-analysis","Numerical Stability and Backward Error Analysis","An algorithm is backward stable when its computed answer is the exact answer to a slightly perturbed problem. Combined with the condition number this gives the governing rule of thumb: forward error is at most condition times stability. Three cancellation case studies make the point, then the residual-based backward error applies it to Ax = b and shows why partial pivoting keeps Gaussian elimination stable.\n",{"path":22399,"title":22400,"module":22384,"summary":22401},"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fqr-and-numerical-least-squares","QR, Householder, and Numerical Least Squares","The least-squares problem reduces to the normal equations, but forming AᵀA squares the condition number and can wreck accuracy. The stable route computes a QR factorization directly on A and solves Rx = Qᵀb. Householder reflectors build that QR one column at a time using length-preserving reflections, the unconditionally backward-stable building block behind every serious least-squares solver.\n",{"path":22403,"title":22404,"module":22384,"summary":22405},"\u002Flinear-algebra\u002Fnumerical-linear-algebra\u002Fnumerical-eigenvalues-and-svd","Numerical Eigenvalue Problems and the SVD","Eigenvalues cannot be found by a formula for large matrices, so they are found by iteration. Power and inverse iteration converge to one eigenvector at a rate set by the eigenvalue gap; the QR algorithm sweeps a matrix to Schur form and, with a good shift and a Hessenberg reduction, computes the whole spectrum in cubic time. Singular values follow from the same machinery applied without ever forming AᵀA.\n",{"path":22407,"title":22408,"module":22409,"summary":22410},"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Faffine-combinations","Affine Combinations","Geometry of Vector Spaces","An affine combination is a linear combination whose weights sum to one. The affine hull of a set is the smallest flat containing it: a point, a line, a plane, or a translated subspace. Homogeneous coordinates turn every affine combination into an ordinary linear combination one dimension up.\n",{"path":22412,"title":22413,"module":22409,"summary":22414},"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Faffine-independence-and-barycentric-coordinates","Affine Independence and Barycentric Coordinates","Affine independence is linear independence for the translated or lifted points, and it guarantees each point of an affine hull a unique weight vector. Those weights are barycentric coordinates: centers of mass, ratios of triangle areas, and the interpolation rule behind smooth shading in computer graphics.\n",{"path":22416,"title":22417,"module":22409,"summary":22418},"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Fconvex-combinations-and-convex-sets","Convex Combinations and Convex Sets","A convex combination is an affine combination with nonnegative weights, and the convex hull of a set is the smallest convex set containing it. Convex sets are closed under intersection, and Carathéodory's theorem bounds how many points a convex combination in $\\mathbb{R}^n$ ever needs: at most $n+1$.\n",{"path":22420,"title":22421,"module":22409,"summary":22422},"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Fhyperplanes-and-polytopes","Hyperplanes and Polytopes","A hyperplane is a level set of a linear functional, the set where an inner product equals a constant. Hyperplanes separate disjoint convex sets and support them at their boundaries. Polytopes are convex hulls of finite point sets; their vertices are the extreme points, and a linear functional attains its extremes there.\n",{"path":22424,"title":22425,"module":22409,"summary":22426},"\u002Flinear-algebra\u002Fgeometry-of-vector-spaces\u002Fcurves-and-surfaces","Curves and Surfaces","Bézier curves are affine combinations of control points with polynomial weights, so they lie in the convex hull of those points and bend toward them. The de Casteljau algorithm evaluates them by repeated interpolation, a matrix form factors them for computation, and matching endpoints and tangents joins segments into smooth curves and surfaces.\n",{"path":22428,"title":22429,"module":6,"summary":6},"\u002Flinear-algebra","Linear Algebra",{"path":22431,"title":22432,"module":6,"summary":6},"\u002Ftheory-of-computation","Theory of Computation",{"path":22434,"title":22435,"module":21235,"summary":22436},"\u002Fcomputer-architecture\u002Ffoundations\u002Fbits-bytes-and-words","Bits, Bytes, and Words","Everything a machine stores is a string of bits grouped into bytes. We set out binary and hexadecimal, the byte as the unit of addressing, the word as the machine's natural integer size, and byte ordering — why the same four bytes read as 0x01234567 on one machine and 0x67452301 on another.\n",{"path":22438,"title":22439,"module":21235,"summary":22440},"\u002Fcomputer-architecture\u002Ffoundations\u002Finteger-representation","Integer Representation","A fixed-width byte string is just a pattern; what makes it a number is the rule we read it by. We define unsigned encoding and two's complement — where the top bit carries a negative weight — derive the ranges UMax, TMin, and TMax, and show how the same bits reinterpret between signed and unsigned, how widening sign-extends, and what truncation throws away.\n",{"path":22442,"title":22443,"module":21235,"summary":22444},"\u002Fcomputer-architecture\u002Ffoundations\u002Finteger-arithmetic","Integer Arithmetic","Fixed-width integer arithmetic is arithmetic modulo a power of two: add past the top and the result wraps. We work out unsigned and two's-complement addition and the rules that detect their overflow, why negation is a complement-plus-one, how multiplication truncates to the low-order bits and how compilers turn constant multiplies into shifts and adds, why C declares signed overflow undefined, and the bias fix that keeps shift-based signed division rounding toward zero.\n",{"path":22446,"title":22447,"module":21235,"summary":22448},"\u002Fcomputer-architecture\u002Ffoundations\u002Ffloating-point","Floating Point","IEEE-754 trades the exactness of integers for enormous range by storing numbers as sign, exponent, and fraction — scientific notation in binary. We lay out the single and double formats, the bias that encodes the exponent, the three regimes (normalized, denormalized, special), a worked encode\u002Fdecode, the four rounding modes and round-to-even at the bit level, why addition is not associative, the pitfalls of float-int conversion, and why 0.1 has no exact binary representation.\n",{"path":22450,"title":22451,"module":21235,"summary":22452},"\u002Fcomputer-architecture\u002Ffoundations\u002Fboolean-algebra-and-bit-manipulation","Boolean Algebra and Bit Manipulation","Treat a word as a vector of independent bits and the bitwise operators become an algebra. We define AND, OR, NOT, and XOR as bit vectors, build the masking idioms that set, clear, toggle, and test individual bits, extract fields with zero- and sign-extension, count set bits three ways, derive the classic x & (x - 1) family of tricks, and distinguish bitwise operators from C's short-circuiting logical operators.\n",{"path":22454,"title":22455,"module":22456,"summary":22457},"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fthe-machines-view","The Machine's View","Machine-Level Programming","The instruction set architecture is the contract a compiler writes against: the program counter, sixteen integer registers with their sub-register widths, and the condition codes. We follow one C function down through gcc to assembly, learn to read an instruction as operation plus operands, and fix the vocabulary the rest of the module uses.\n",{"path":22459,"title":22460,"module":22456,"summary":22461},"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fdata-movement","Data Movement","Most instructions a program runs simply move data. We cover the mov family and its size suffixes, the three operand forms, the full memory addressing mode D(Rb,Ri,S) and its special cases, lea for address arithmetic, and how push and pop manipulate the stack pointer %rsp on a stack that grows toward lower addresses.\n",{"path":22463,"title":22464,"module":22456,"summary":22465},"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Farithmetic-and-logic","Arithmetic and Logic","The ALU instructions that compute on register and memory values: add, sub, and imul; the unary inc\u002Fdec\u002Fneg\u002Fnot; the shifts sal\u002Fshr\u002Fsar; the bitwise and\u002For\u002Fxor; and lea reused as a fast arithmetic trick. Each binary operation also sets the condition-code flags CF, ZF, SF, and OF, which cmp and test compute without keeping a result.\n",{"path":22467,"title":22468,"module":22456,"summary":22469},"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fcontrol-flow","Control Flow","How a flat instruction stream realizes branches and loops. The conditional jumps read the condition-code flags; set instructions turn flags into a 0\u002F1 byte. We translate if\u002Felse into the standard compare-and-branch pattern, while\u002Ffor loops into the guarded-do form, and dense switches into jump tables that index a target directly.\n",{"path":22471,"title":22472,"module":22456,"summary":22473},"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fprocedures","Procedures","How a function call works at the machine level: the run-time stack, call and ret passing control through a saved return address, the System V convention that routes the first six arguments through %rdi..%r9 and the result through %rax, the caller-saved versus callee-saved split, the stack frame, and a recursive factorial traced through its frames.\n",{"path":22475,"title":22476,"module":22456,"summary":22477},"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Farrays-structs-and-alignment","Arrays, Structs, and Alignment","How aggregate data lays out in memory. Arrays as base-plus-scaled-index, the row-major ordering of multidimensional arrays, pointer arithmetic in units of the pointed-to type, struct fields at fixed byte offsets, the overlapping storage of unions, and the alignment rules that force padding into a struct.\n",{"path":22479,"title":22480,"module":22456,"summary":22481},"\u002Fcomputer-architecture\u002Fmachine-level-x86-64\u002Fmemory-layout-and-buffer-overflows","Memory Layout and Buffer Overflows","The process address space — text, data, heap, and stack — and the classic vulnerability it enables. A stack buffer that is written past its end can overwrite the saved return address and redirect ret, so we sketch the mechanism defensively and then the three standard protections: stack canaries, a non-executable stack, and address-space layout randomization.\n",{"path":22483,"title":22484,"module":22485,"summary":22486},"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Fwhat-an-isa-is","What an ISA Is","Instruction Set Architecture","The instruction set architecture is the contract that lets a compiler and a chip be written by people who never meet: the stable interface software targets and hardware implements. We separate architecture from microarchitecture, read RISC and CISC as opposite answers to where complexity should live, price out what each choice costs in decode hardware, code density, and pipeline friendliness, and see how x86-64 endures by translating its instructions into RISC-like operations on the fly.\n",{"path":22488,"title":22489,"module":22485,"summary":22490},"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Finstruction-formats-and-operands","Instruction Formats and Operands","An instruction is an opcode plus a way to name its operands. We count operands — 3-address, 2-address, 1-address accumulator, and 0-address stack machines — by writing the same C = A + B four ways, weigh register operands against memory operands, then lay out the same add byte by byte in x86-64 (REX prefix, opcode, ModRM) and in Y86-64, and what fixed versus variable length costs at fetch time.\n",{"path":22492,"title":22493,"module":22485,"summary":22494},"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Faddressing-modes","Addressing Modes","Once an operand field exists, it needs a rule for turning its bits into the data it names. That rule is the addressing mode. We walk the standard set — immediate, register, direct, register-indirect, displacement, scaled-indexed, and PC-relative — fixing the effective-address computation for each, run every mode against one concrete machine state, and price out what Y86-64 loses by keeping only base plus displacement.\n",{"path":22496,"title":22497,"module":22485,"summary":22498},"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Fthe-y86-64-instruction-set","The Y86-64 Instruction Set","Y86-64 is a teaching ISA — a stripped-down x86-64 simple enough to implement by hand yet real enough to compile to. We fix its programmer-visible state (fifteen registers, three condition codes, the PC, memory, and a status code), give the instruction set with exact byte encodings, spell out how the condition codes decide every jXX and cmovXX, and run the encoding both directions: assembly to bytes and raw bytes back to meaning.\n",{"path":22500,"title":22501,"module":22485,"summary":22502},"\u002Fcomputer-architecture\u002Finstruction-set-architecture\u002Fy86-64-programming","Y86-64 Programming","With the encodings fixed, we write real Y86-64 assembly: the .pos, .align, and .quad directives, the calling convention borrowed from x86-64, a stack set up by hand, and complete programs — an array sum and a branch-free max. We watch the assembler turn the listing into the exact byte image the processor will execute, and trace the stack across the call.\n",{"path":22504,"title":22505,"module":22506,"summary":22507},"\u002Fcomputer-architecture\u002Fdigital-logic\u002Ftransistors-gates-and-boolean-functions","Transistors, Gates, and Boolean Functions","Digital Logic","A processor is built from millions of transistor switches. We start at the MOS transistor as a voltage-controlled switch, build the CMOS inverter and NAND transistor by transistor, meet the seven standard gates with their truth tables, show that NAND alone is functionally complete, price each gate in transistors and in time, and turn any truth table into a sum-of-products circuit.\n",{"path":22509,"title":22510,"module":22506,"summary":22511},"\u002Fcomputer-architecture\u002Fdigital-logic\u002Fcombinational-logic-and-hcl","Combinational Logic and HCL","A combinational circuit is a pure Boolean function of its current inputs — no memory, no clock. We draw the line between combinational and sequential logic, do the gate-delay accounting that finds a circuit's critical path and bounds the clock, meet don't-cares, then introduce CS:APP's Hardware Control Language: bit-level operators, word-level signals, equality nets, and the case expression that compiles to a multiplexer tree.\n",{"path":22513,"title":22514,"module":22506,"summary":22515},"\u002Fcomputer-architecture\u002Fdigital-logic\u002Fmultiplexers-decoders-and-the-alu","Multiplexers, Decoders, and the ALU","The combinational building blocks that make a datapath. We build the 2:1 and 4:1 multiplexer and tie it back to HCL's case expression, the n-to-2^n decoder, a one-bit full adder (sum is XOR, carry is majority), the ripple-carry adder that chains them, and finally the ALU — a function unit that selects among add, sub, and, and xor under a control input and exposes condition flags.\n",{"path":22517,"title":22518,"module":22506,"summary":22519},"\u002Fcomputer-architecture\u002Fdigital-logic\u002Fmemory-elements-latches-flip-flops-and-clocking","Memory Elements: Latches, Flip-Flops, and Clocking","A combinational circuit holds no state; feeding a circuit's output back to its input creates memory. We build the SR latch from cross-coupled gates, the level-sensitive D latch, and the master\u002Fslave edge-triggered D flip-flop, then introduce the clock and the synchronous design discipline, the setup\u002Fhold timing window, clock skew, metastability, and the register as n flip-flops sharing one clock.\n",{"path":22521,"title":22522,"module":22506,"summary":22523},"\u002Fcomputer-architecture\u002Fdigital-logic\u002Fregister-files-and-random-access-memory","Register Files and Random-Access Memory","Storage organized for access by address. We build the register file (a small bank of registers with addressed read ports and clocked write ports, the exact structure Y86-64's decode and write-back stages use), then descend to the SRAM and DRAM cells of main memory, why one is fast and dear and the other dense and slow, and how a row decoder picks a word out of a memory array.\n",{"path":22525,"title":22526,"module":22527,"summary":22528},"\u002Fcomputer-architecture\u002Fprocessor-design\u002Fthe-fetch-decode-execute-cycle","The Fetch-Decode-Execute Cycle","Processor Design","A processor is a machine that repeats one loop forever: read the next instruction from memory, figure out what it asks for, do it, and advance. We fix the stored-program idea, lay out the datapath at a high level — PC, instruction memory, register file, ALU, data memory — and the control unit that sequences them, break the work into the six stages the rest of the module builds in hardware, and work out exactly how fetch parses variable-length instructions and computes the next PC.\n",{"path":22530,"title":22531,"module":22527,"summary":22532},"\u002Fcomputer-architecture\u002Fprocessor-design\u002Fthe-seq-stages","The SEQ Stages","The six SEQ stages, made exact. For every Y86-64 instruction — halt, nop, the moves, OPq, the jumps, call and ret, pushq and popq — we write down what Fetch, Decode, Execute, Memory, Write-back, and PC update each compute, as per-instruction stage tables with every row justified. Once the tables are filled in, the processor is fully specified; the remaining lessons turn them into wires.\n",{"path":22534,"title":22535,"module":22527,"summary":22536},"\u002Fcomputer-architecture\u002Fprocessor-design\u002Fcontrol-logic-and-sequencing","Control Logic and Sequencing","The stage tables say what each instruction needs; the control logic computes it from icode. We write the HCL for the register-port selections (srcA, srcB, dstE, dstM), the ALU function and input selection, the memory read\u002Fwrite and address, the branch condition, and the next-PC mux — each a case expression on icode that compiles to a mux — and see how one blob of combinational logic serves every instruction at once. We close by contrasting hardwired control with the microprogrammed alternative.\n",{"path":22538,"title":22539,"module":22527,"summary":22540},"\u002Fcomputer-architecture\u002Fprocessor-design\u002Fassembling-seq","Assembling SEQ","We wire the whole thing together. The functional units from digital logic and the control signals from the last lesson assemble into the complete SEQ datapath, laid out the way CS:APP draws it — six stages stacked bottom to top, Fetch at the floor and PC update at the ceiling, signals flowing up the margins. Then the timing analysis: why everything must settle in one cycle, the no-reading-back principle that makes single-cycle execution consistent, and the critical path that sets the clock. We close by walking an OPq and a ret through the assembled machine.\n",{"path":22542,"title":22543,"module":22527,"summary":22544},"\u002Fcomputer-architecture\u002Fprocessor-design\u002Ftracing-a-program","Tracing a Program","To close the module, we take a complete Y86-64 program — a loop that sums 1 through 3 — and run it through SEQ one cycle at a time, recording the PC, the fetched instruction, every stage computation, and the registers, condition codes, and memory after each cycle. Then we examine single cycles in detail: every named signal of an OPq in concrete hex, and a second program whose call and ret we trace through the stack. The traces confirm that the assembled datapath and control logic behave as a processor.\n",{"path":22546,"title":22547,"module":22548,"summary":22549},"\u002Fcomputer-architecture\u002Fpipelining\u002Fpipelining-principles","Pipelining Principles","Pipelining","A processor that runs one instruction to completion before starting the next wastes most of its hardware most of the time. Pipelining splits the work into stages separated by registers so several instructions are in flight at once. We separate throughput from latency, work the 300 ps example through one, two, and three stages, and derive the three ceilings on the gain: uneven stages, register overhead, and the dependencies between instructions.\n",{"path":22551,"title":22552,"module":22548,"summary":22553},"\u002Fcomputer-architecture\u002Fpipelining\u002Ffrom-seq-to-pipe","From SEQ to PIPE","We turn the sequential Y86-64 processor into a pipelined one by inserting pipeline registers between its stages so each cycle holds one instruction per stage. Doing it correctly forces a rearrangement: the next-PC computation must move into Fetch as a prediction, because the later stages that used to compute it are now busy with other instructions. We walk SEQ to SEQ+ to PIPE, spell out exactly what each pipeline register carries, and fix the naming discipline (D_stat versus d_stat) that keeps five in-flight instructions straight.\n",{"path":22555,"title":22556,"module":22548,"summary":22557},"\u002Fcomputer-architecture\u002Fpipelining\u002Fdata-hazards-stalling-and-forwarding","Data Hazards: Stalling and Forwarding","Overlapping instructions collide when a later one needs a value an earlier one has not finished computing: a read-after-write data hazard. We map exactly which instruction distances are dangerous, fix hazards the slow way by stalling (three bubbles), then the fast way by forwarding from five distinct sources into Decode, in a priority order that sequential semantics forces. Forwarding handles almost everything; the load-use hazard still needs exactly one stall.\n",{"path":22559,"title":22560,"module":22548,"summary":22561},"\u002Fcomputer-architecture\u002Fpipelining\u002Fcontrol-hazards-and-branch-prediction","Control Hazards and Branch Prediction","A pipeline must fetch an instruction every cycle, but after a conditional jump or a ret the next address is not yet known: a control hazard. We measure the branch penalty, weigh predict-taken against its alternatives with real loop arithmetic, watch PIPE detect a misprediction in Execute and squash the two wrong-path instructions, and meet the ret hazard, which has nothing to predict and stalls three cycles. A 2-bit counter gives a taste of dynamic prediction.\n",{"path":22563,"title":22564,"module":22548,"summary":22565},"\u002Fcomputer-architecture\u002Fpipelining\u002Fthe-complete-pipe-processor","The Complete PIPE Processor","We assemble the full pipelined Y86-64: five stages, five pipeline registers, forwarding paths, and a small control unit that decides, each cycle, whether to stall or bubble each register. The subtle part is when hazards combine: one pairing hides a genuine bug. A fourth control case reads stat and keeps exceptions precise. Performance reduces to CPI = 1 + lp + mp + rp, worked out to 1.27 with realistic frequencies, and PIPE beats SEQ by several times despite every penalty.\n",{"path":22567,"title":22568,"module":22569,"summary":22570},"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Fstorage-technologies-and-the-latency-gap","Storage Technologies and the Latency Gap","The Memory Hierarchy","No single memory is both fast and large and cheap. We survey the technologies a machine can store bits in — SRAM, DRAM, flash, and rotating disk — open up a DRAM chip to find the row buffer, work a disk access down to the millisecond, and rank everything by speed, density, and cost per bit. Then we watch the processor outrun memory decade after decade. That widening gap is the whole reason a machine stacks fast small storage on top of slow large storage into a hierarchy.\n",{"path":22572,"title":22573,"module":22569,"summary":22574},"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Flocality","Locality","A hierarchy only pays off because programs do not touch memory at random. They reuse recently-used data (temporal locality) and touch nearby data soon after (spatial locality). We make both precise and then quantitative: miss rates for stride-1 and stride-k traversals against a concrete block size, and the loop-order pair on a 2-D array where the same sum misses 16 times one way and 64 times the other — why row-major versus column-major order can change a program's speed by an order of magnitude.\n",{"path":22576,"title":22577,"module":22569,"summary":22578},"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Fcache-memories-direct-mapped","Cache Memories and Direct Mapping","A cache is fast SRAM that holds copies of recently-used blocks of main memory. We fix its organization — S sets, E lines per set, B bytes per block — and the way it dissects an address into tag, set index, and block offset, worked bit by bit on a concrete 16-byte cache. Then we run the direct-mapped (E=1) access algorithm end to end on a seven-access trace: index to a set, compare the tag, hit or miss, evict. Cold and conflict misses fall out of the structure, and a two-array ping-pong shows conflict thrashing and its padding fix.\n",{"path":22580,"title":22581,"module":22569,"summary":22582},"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Fset-associative-and-write-policies","Set-Associative Caches and Write Policies","Give each set several lines and a block has a choice of homes — fewer conflict misses, at the cost of comparing E tags in parallel and choosing a victim to evict. We re-run the direct-mapped ping-pong trace on a 2-way cache and watch the conflicts vanish, weigh LRU against random replacement, then turn to writes: write-through versus write-back with a dirty bit on a hit, write-allocate versus no-write-allocate on a miss, and a worked traffic count showing when each pairing wins.\n",{"path":22584,"title":22585,"module":22569,"summary":22586},"\u002Fcomputer-architecture\u002Fmemory-hierarchy\u002Fcache-performance-and-cache-friendly-code","Cache Performance and Cache-Friendly Code","Turn the cache mechanism into a number. Hit time, miss rate, and miss penalty combine into the average memory access time; we compute AMAT for a two-level hierarchy with real numbers, weigh the design knobs against each other, and read the memory mountain. Then we write cache-friendly code — the matrix-multiply loop-order case study (ijk versus kij, misses counted per iteration) and loop blocking, where cache-sized tiles turn evicted reuse back into hits.\n",{"path":22588,"title":22589,"module":22590,"summary":22591},"\u002Fcomputer-architecture\u002Fvirtual-memory\u002Faddress-spaces-and-translation","Address Spaces and Translation","Virtual Memory","Every process runs as if it owns a private, contiguous span of memory — its virtual address space — while the hardware maps those addresses onto a single shared physical memory. We fix virtual memory's three jobs (a cache for disk, a memory manager, a protection boundary), the page as the unit of mapping, and the MMU replacing the virtual page number while the offset passes through untouched — then run one translation end to end at the bit level and trace the control flow of a page hit against a page fault.\n",{"path":22593,"title":22594,"module":22590,"summary":22595},"\u002Fcomputer-architecture\u002Fvirtual-memory\u002Fpage-tables-and-page-faults","Page Tables and Page Faults","The page table is an array of page-table entries indexed by virtual page number; each entry's valid bit says whether the page is in DRAM, on disk, or unallocated, and its permission, reference, and dirty bits drive protection and replacement. We walk translation as a table lookup, the page fault and demand paging, the clock algorithm the OS uses to approximate LRU, memory mapping and copy-on-write (why fork is cheap), the taxonomy of bad references, and thrashing.\n",{"path":22597,"title":22598,"module":22590,"summary":22599},"\u002Fcomputer-architecture\u002Fvirtual-memory\u002Fthe-tlb-and-multi-level-page-tables","The TLB and Multi-Level Page Tables","A page-table read on every access would double memory traffic; a flat table for a 48-bit space would occupy 512 GB per process. The TLB fixes the first: a small set-associative cache of PTEs inside the MMU whose tag and index come from the VPN. Multi-level page tables fix the second, allocating only the sub-tables a process uses; x86-64 walks four levels with a 9+9+9+9+12 split. We trace one reference end to end through TLB, walk, and cache, and close with the overlap trick that lets the L1 cache start before translation ends.\n",{"path":22601,"title":22602,"module":22603,"summary":22604},"\u002Fcomputer-architecture\u002Fexceptions-and-io\u002Fexceptional-control-flow","Exceptional Control Flow","Exceptions & I\u002FO","Beyond the sequential, branch, and call flow a program controls itself, the hardware can divert the processor in response to events. We sort these into four classes — interrupts (asynchronous, from devices), traps (intentional syscalls), faults (recoverable, like a page fault), and aborts (unrecoverable) — then take the mechanism apart: exception numbers and the table dispatch, what the hardware pushes and why it differs from a procedure call, the divide-error \u002F page-fault \u002F general-protection trio on x86-64, the full syscall round trip with a worked write in assembly, and processes and signals as the abstractions ECF makes possible.\n",{"path":22606,"title":22607,"module":22603,"summary":22608},"\u002Fcomputer-architecture\u002Fexceptions-and-io\u002Finterrupts-and-the-kernel","Interrupts and the Kernel","An I\u002FO device signals completion by raising an interrupt, crossing the privilege boundary from user mode into the kernel. We fix that boundary, follow an interrupt from device through the interrupt controller to its vectored handler, and use the timer interrupt to drive preemptive scheduling and the context switch. Then the I\u002FO mechanics: polling versus interrupt-driven I\u002FO with a cycle count, device registers and memory-mapped I\u002FO versus port I\u002FO, DMA's full transfer walkthrough and its cache hazard, and a disk read traced end to end, from the read syscall to the completion interrupt.\n",{"path":22610,"title":22611,"module":22612,"summary":22613},"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fprocesses-threads-and-parallelism","Processes, Threads, and Parallelism","Multithreading & Multicore","Around 2004 the single core stopped getting faster, and the industry's answer was to hand programmers more cores instead. This lesson builds the vocabulary that shift demands: process versus thread and exactly which hardware state each one owns, concurrency versus parallelism, the three kinds of parallelism a machine can exploit, why Dennard scaling ended and forced the multicore turn, and Amdahl's law — the arithmetic that bounds the speedup those cores can deliver.\n",{"path":22615,"title":22616,"module":22612,"summary":22617},"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fhardware-multithreading","Hardware Multithreading","A pipeline spends much of its life waiting — on cache misses, on dependences, on branches. Hardware multithreading fills the dead cycles with instructions from another thread. We compare coarse-grained switching (change threads on a long stall), fine-grained interleaving (change every cycle), and simultaneous multithreading (mix threads inside a single cycle), work out exactly which hardware a second thread context duplicates and which it shares, and weigh when SMT pays off and when two threads just fight over one cache.\n",{"path":22619,"title":22620,"module":22612,"summary":22621},"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fcache-coherence","Cache Coherence","Give each core its own cache and the same address can live in two places at once, with copies that disagree. We reproduce the stale-copy bug with a two-core trace, then fix it the way hardware does: snooping caches that watch a shared bus and keep every line in a protocol state. We build MSI in full, upgrade it to MESI, contrast invalidation with updating, add coherence misses as the fourth C, and end with false sharing: the performance bug where cores fight over a line while never touching the same byte.\n",{"path":22623,"title":22624,"module":22612,"summary":22625},"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fmemory-consistency-and-synchronization","Memory Consistency and Synchronization","Coherence keeps cores agreeing about one location; consistency is the contract about many. We define sequential consistency, then watch real hardware break it: the store buffer lets a load slip ahead of an older store, and the classic two-thread litmus test ends with both sides reading zero. We state x86-TSO precisely, restore order with mfence, build atomic read-modify-write from the lock prefix, xchg, and cmpxchg, and write a spinlock twice — once naively, once bus-friendly — closing with what lock-free progress actually guarantees.\n",{"path":22627,"title":22628,"module":22612,"summary":22629},"\u002Fcomputer-architecture\u002Fmultithreading-and-multicore\u002Fmulticore-organization","Multicore Organization","Where everything sits on the die. A modern die gives each core private L1 and L2 caches, spreads a shared last-level cache across slices, and wires it all together with a ring or mesh; multi-socket servers add NUMA, where memory is local to one socket and every remote access pays a latency penalty. We walk the floorplan, put numbers on local versus remote latency, meet thread affinity, and account for the two shared resources — coherence traffic and LLC capacity — that decide how far a parallel program scales.\n",{"path":22631,"title":22632,"module":22633,"summary":22634},"\u002Fcomputer-architecture\u002Fcapstone\u002Fthe-whole-machine","The Whole Machine","Capstone","We take one line of C down the whole tower the course built — compiler to assembly, assembly to machine-code bytes, the bytes into the fetch–decode–execute datapath — then trace one load and one add through the pipelined, cached, translated, interruptible machine, each step cross-linked to the lesson that built it. We close with the map of the course as a stack of layers and an accounting of what we simplified: out-of-order execution, superscalar issue, and speculation past the branch predictor.\n",{"path":22636,"title":22637,"module":22633,"summary":22638},"\u002Fcomputer-architecture\u002Fcapstone\u002Fassembling-a-complete-cpu","Assembling a Complete CPU","We bolt the parts the course built — PC, instruction memory and its fetch logic, register file, ALU, condition codes, data memory, and the control unit — into one complete CPU, name the lesson that built each, wire them in a deliberate order, and power the machine on from reset. Then we assemble a real test program (sum a four-element array through a call\u002Fret procedure), give its exact bytes and memory layout, and trace it cycle by cycle to the answer 0xabcdabcdabcd. We close with how to validate such a machine, and what it takes to put two of them on one die.\n",{"path":22640,"title":22641,"module":6,"summary":6},"\u002Fcomputer-architecture","Computer Architecture",{"path":22643,"title":22644,"module":21235,"summary":22645},"\u002Fdifferential-equations\u002Ffoundations\u002Fmodels-and-direction-fields","Models, Direction Fields, and Solution Curves","A differential equation relates an unknown function to its own rates of change. Three first-order models — a falling body, a cooling object, a population under predation — share the form dy\u002Fdt = ay - b; the slope field fixes their equilibria and long-run behavior before any formula is found. Solving the linear case gives the general solution, its integral curves, and the particular solution selected by an initial condition.\n",{"path":22647,"title":22648,"module":21235,"summary":22649},"\u002Fdifferential-equations\u002Ffoundations\u002Fclassification-and-terminology","Classifying Equations: Order, Linearity, ODE vs. PDE","Every solution method targets a specific class of equation, so the first question about any differential equation is which classes it belongs to. Four independent axes sort them: ordinary versus partial, order, linear versus nonlinear, and homogeneous versus nonhomogeneous. Systems, verification of a solution by substitution, and the split between initial and boundary value problems complete the vocabulary.\n",{"path":22651,"title":22652,"module":22653,"summary":22654},"\u002Fdifferential-equations\u002Ffirst-order\u002Flinear-first-order-integrating-factors","Linear Equations and Integrating Factors","First-Order Equations","A first-order linear equation has the unknown and its derivative to the first power only. Multiplying by an integrating factor collapses the left side into a single derivative, and one integration gives the general solution in closed form. The solution exists wherever the coefficients are continuous, and for a constant coefficient it splits into a decaying transient and a steady state set by the forcing.\n",{"path":22656,"title":22657,"module":22653,"summary":22658},"\u002Fdifferential-equations\u002Ffirst-order\u002Fseparable-and-exact","Separable and Exact Equations","Two nonlinear first-order classes solve by direct integration. A separable equation splits so that each variable can be integrated on its own side, giving an implicit relation. An exact equation is the total differential of a hidden potential function, recognized by a symmetry test on its coefficients; when the test fails, an integrating factor can sometimes restore exactness. A change of variable brings homogeneous equations into the separable class.\n",{"path":22660,"title":22661,"module":22653,"summary":22662},"\u002Fdifferential-equations\u002Ffirst-order\u002Fmodeling-first-order","Modeling with First-Order Equations","A rate law is a differential equation. Each first-order model starts from one governing principle: conservation of mass for a mixing tank, proportional change for interest and radioactive decay, Newton's law of cooling, a force balance for a body falling against drag, and Kirchhoff's law for a series circuit. Setting the derivative to zero recovers the steady state, and the transient records how the initial condition relaxes toward it.\n",{"path":22664,"title":22665,"module":22653,"summary":22666},"\u002Fdifferential-equations\u002Ffirst-order\u002Fautonomous-and-population-dynamics","Autonomous Equations, Phase Lines, and Population Dynamics","An autonomous equation y' = f(y) can be analyzed qualitatively without being solved. Its constant solutions are the zeros of f, and the sign of f between them fixes whether nearby solutions rise or fall, which the phase line records as a column of arrows. The logistic and threshold models, constant- and effort-proportional harvesting, and the properties nonlinear equations lose all follow from this reading.\n",{"path":22668,"title":22669,"module":22653,"summary":22670},"\u002Fdifferential-equations\u002Ffirst-order\u002Fexistence-uniqueness-euler","Existence, Uniqueness, and Euler's Method","Existence and uniqueness can be settled before any attempt to solve. The existence-uniqueness theorem gives sufficient conditions on f, and a standard example shows what fails when they do not hold. Picard's successive approximations build the solution as the limit of an iteration, and Euler's method turns the same tangent-line idea into a numerical procedure for the equations no formula reaches.\n",{"path":22672,"title":22673,"module":22653,"summary":22674},"\u002Fdifferential-equations\u002Ffirst-order\u002Ffirst-order-difference-equations","First-Order Difference Equations","A difference equation advances a sequence one index at a time by a rule y_{n+1} = f(y_n). The linear case y_{n+1} = rho*y_n + b solves in closed form and converges to its equilibrium exactly when the ratio has magnitude below one, which underlies compound-interest and loan calculations. The logistic difference equation shows the nonlinear counterpart: an exchange of stability, a cascade of period doublings, and the onset of chaos.\n",{"path":22676,"title":22677,"module":22678,"summary":22679},"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fhomogeneous-constant-coefficients","Homogeneous Equations, the Wronskian, and Real Roots","Second-Order Linear Equations","A second-order linear homogeneous equation with constant coefficients is solved by guessing an exponential and reducing to the quadratic characteristic equation. Two solutions span every solution exactly when their Wronskian is nonzero; that condition, superposition, and Abel's formula give the full structure of the general solution for the case of two distinct real roots.\n",{"path":22681,"title":22682,"module":22678,"summary":22683},"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fcomplex-and-repeated-roots","Complex Roots, Repeated Roots, and Reduction of Order","When the characteristic equation has complex conjugate roots, Euler's formula converts the complex exponentials into a real fundamental set of decaying or growing oscillations. When it has a repeated root, one exponential is lost and reduction of order recovers the missing second solution as $t\\,e^{rt}$. The same substitution $y = v(t)y_1(t)$ finds a second solution from any known one.\n",{"path":22685,"title":22686,"module":22678,"summary":22687},"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fnonhomogeneous-undetermined-coefficients","Nonhomogeneous Equations: Undetermined Coefficients","The general solution of a nonhomogeneous linear equation is a complementary solution plus any one particular solution. When the forcing term is a polynomial, exponential, sine, or cosine, a particular solution can be found by assuming a trial form of the same shape with unknown coefficients and solving for them. The one complication is resonance, handled by multiplying the trial by a power of $t$.\n",{"path":22689,"title":22690,"module":22678,"summary":22691},"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fvariation-of-parameters","Variation of Parameters","Variation of parameters finds a particular solution of any nonhomogeneous linear equation from a fundamental set of the homogeneous one. Replacing the constants in the complementary solution by functions and imposing one convenient constraint reduces the problem to a two-by-two linear system whose solution is expressed through the Wronskian, giving an integral formula that works for forcing terms undetermined coefficients cannot touch.\n",{"path":22693,"title":22694,"module":22678,"summary":22695},"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fmechanical-electrical-vibrations","Mechanical and Electrical Vibrations","A spring-mass-damper obeys a second-order linear equation, and so does a series RLC circuit, with the same mathematics governing both. Free undamped motion is a pure sinusoid; damping adds a decaying envelope with three regimes; periodic forcing produces a transient that dies out and a steady-state oscillation whose amplitude peaks sharply near the natural frequency, the phenomenon of resonance.\n",{"path":22697,"title":22698,"module":22678,"summary":22699},"\u002Fdifferential-equations\u002Fsecond-order-linear\u002Fhigher-order-linear","Higher-Order Linear Equations","The second-order theory extends directly to order $n$: the solution space is $n$-dimensional, spanned by any $n$ solutions with nonzero Wronskian. For constant coefficients the characteristic polynomial has degree $n$, and its roots (counted with multiplicity, real and complex) build the basis by the same rules as before. Coupled oscillators are the natural application that raises the order.\n",{"path":22701,"title":22702,"module":22703,"summary":22704},"\u002Fdifferential-equations\u002Fseries-solutions\u002Fpower-series-ordinary-points","Power Series Solutions Near Ordinary Points","Series Solutions and Special Functions","A linear equation with variable coefficients has no characteristic equation. A power series substituted into the equation matches coefficients to a recurrence relation, which near an ordinary point yields two independent analytic solutions. The radius of convergence is at least the distance from the expansion point to the nearest singular point in the complex plane.\n",{"path":22706,"title":22707,"module":22703,"summary":22708},"\u002Fdifferential-equations\u002Fseries-solutions\u002Fregular-singular-frobenius","Euler Equations, Regular Singular Points, and Frobenius","The Euler equation x^2 y'' + a x y' + b y = 0 is solved outright by y = x^r, and its three root cases fix the behavior at any regular singular point. The Frobenius method multiplies x^r by a power series; the indicial equation chooses the exponents, and equal or integer-separated roots force a logarithm in the second solution. Gauss's hypergeometric equation is the archetype containing most classical functions as special cases.\n",{"path":22710,"title":22711,"module":22703,"summary":22712},"\u002Fdifferential-equations\u002Fseries-solutions\u002Fbessel-and-special-functions","Bessel's Equation, Legendre Polynomials, and Special Functions","Bessel's equation puts the Frobenius machinery through all three of its cases and produces the functions J and Y that govern anything vibrating or diffusing with circular symmetry. The gamma function extends the factorial so that Bessel functions of every order make sense; Legendre's equation, run through the hypergeometric form, yields the polynomials that play the same role in spherical geometry. Orthogonality ties both families to the eigenfunction expansions of Sturm–Liouville theory.\n",{"path":22714,"title":22715,"module":22716,"summary":22717},"\u002Fdifferential-equations\u002Flaplace\u002Flaplace-definition-ivps","The Laplace Transform: Definition, Properties, and Solving IVPs","The Laplace Transform","The Laplace transform sends a function of time to a function of a complex frequency by integrating it against the kernel e^{-st}. Differentiation in t becomes multiplication by s, so a linear constant-coefficient initial value problem turns into an algebraic equation. Existence rests on piecewise continuity and exponential order; the derivative rule folds in the initial data; and inversion runs through a transform table and partial fractions.\n",{"path":22719,"title":22720,"module":22716,"summary":22721},"\u002Fdifferential-equations\u002Flaplace\u002Fstep-impulse-convolution","Step Functions, Discontinuous Forcing, Impulses, and Convolution","The Heaviside step function and the second shifting theorem transform switches and discontinuous forcing into exponential factors on the transform. The Dirac delta idealizes an instantaneous impulse and transforms to a pure exponential. The convolution theorem inverts a product of transforms, writes the forced response as the impulse response convolved with the input, and solves Abel's tautochrone by transform.\n",{"path":22723,"title":22724,"module":22725,"summary":22726},"\u002Fdifferential-equations\u002Fsystems\u002Fmatrices-eigenvalues-review","Matrices, Linear Systems, and the Eigenvalue Toolkit","Systems of First-Order Linear Equations","Any nth-order linear equation, and any coupled collection of them, rewrites as a single first-order system x' = P(t)x + g(t). The matrix and vector algebra behind that form, the eigenvalue problem det(A - λI) = 0 that drives every solution method, and the fundamental theory — superposition, the Wronskian, Abel's theorem — together establish that n independent solutions span all solutions.\n",{"path":22728,"title":22729,"module":22725,"summary":22730},"\u002Fdifferential-equations\u002Fsystems\u002Fconstant-coefficient-systems-phase-portraits","Homogeneous Constant-Coefficient Systems and Phase Portraits","For x' = Ax with A constant, the trial x = ξe^{rt} turns the differential equation into the eigenvalue problem Aξ = rξ. The eigenvalues fix the geometry of the phase plane: real opposite signs give a saddle, real same sign a node, complex a spiral, purely imaginary a center. Worked in the plane, these cases form the eigenvalue-type classification of equilibria.\n",{"path":22732,"title":22733,"module":22725,"summary":22734},"\u002Fdifferential-equations\u002Fsystems\u002Frepeated-eigenvalues-fundamental-matrices","Repeated Eigenvalues, Fundamental Matrices, and Nonhomogeneous Systems","When a repeated eigenvalue supplies too few eigenvectors, a generalized eigenvector supplies the missing solution as ξte^{ρt} + ηe^{ρt}, giving an improper node. A fundamental set packaged as a matrix Φ(t) yields the matrix exponential e^{At}, the propagator mapping initial states to later ones. Variation of parameters solves the nonhomogeneous system x' = Ax + g(t).\n",{"path":22736,"title":22737,"module":22738,"summary":22739},"\u002Fdifferential-equations\u002Fnumerical\u002Feuler-and-runge-kutta","Euler, Improved Euler, and Runge–Kutta","Numerical Methods","Most initial value problems have no closed-form solution, so the solution is approximated on a grid. Euler's method steps along the tangent line, the improved Euler method averages two slopes, and the classical Runge–Kutta method averages four. Each added stage raises the order of accuracy at the cost of more evaluations per step, measured by how the local and global truncation errors scale with the step size.\n",{"path":22741,"title":22742,"module":22738,"summary":22743},"\u002Fdifferential-equations\u002Fnumerical\u002Fmultistep-systems-stability","Multistep Methods, Systems, and Stability","One-step methods discard everything but the last point. Multistep methods fit a polynomial to several past values and integrate it forward: the explicit Adams–Bashforth formulas, the implicit and more accurate Adams–Moulton formulas, and predictor–corrector pairs that combine them. The same rules extend verbatim to systems in vector form. A separate concern is stability: round-off can dominate truncation, and stiff equations force a tiny step for stability even when accuracy would allow a large one.\n",{"path":22745,"title":22746,"module":22747,"summary":22748},"\u002Fdifferential-equations\u002Fnonlinear\u002Fphase-plane-autonomous-stability","The Phase Plane, Critical Points, and Stability","Nonlinear Systems and Stability","Most nonlinear systems cannot be solved in closed form, so they are studied geometrically. The phase plane turns an autonomous planar system into a family of trajectories; the five archetypes of critical point follow from the eigenvalues of the coefficient matrix; the trace-determinant plane reads off type and stability directly; and epsilon-delta definitions make stability, asymptotic stability, and instability precise.\n",{"path":22750,"title":22751,"module":22747,"summary":22752},"\u002Fdifferential-equations\u002Fnonlinear\u002Flocally-linear-and-liapunov","Locally Linear Systems and Liapunov's Method","Near a critical point a nonlinear system looks linear, and the linear part is the Jacobian. The linearization fixes the type and stability of the nonlinear critical point in every case except a center or a repeated eigenvalue. Liapunov's direct method settles those cases and bounds the basin of attraction by constructing an energy-like function, without solving the system.\n",{"path":22754,"title":22755,"module":22747,"summary":22756},"\u002Fdifferential-equations\u002Fnonlinear\u002Fcompeting-species-predator-prey-limit-cycles","Population Models, Limit Cycles, and Chaos","The phase-plane methods apply directly to interacting-population models. Competing species either coexist or drive one another to extinction, decided by a single inequality among the interaction constants; the Lotka-Volterra predator-prey system produces closed population cycles. Limit cycles and the Poincaré-Bendixson theorem, the van der Pol oscillator, and the Lorenz equations with their strange attractor carry the theory into chaos.\n",{"path":22758,"title":22759,"module":22760,"summary":22761},"\u002Fdifferential-equations\u002Fpdes-fourier-bvp\u002Ffourier-series","Fourier Series and Convergence","PDEs, Fourier Series, and Boundary Value Problems","A two-point boundary value problem has nontrivial solutions only at a discrete set of eigenvalues, the same trichotomy that governs a singular linear system. For y'' + lambda y = 0 with zero endpoints the eigenfunctions are sines and cosines, and their orthogonality gives the Euler-Fourier coefficient formulas. The convergence theorem fixes when the series returns the function, the Gibbs phenomenon measures the overshoot at a jump, and even\u002Fodd symmetry produces half-range sine and cosine series.\n",{"path":22763,"title":22764,"module":22760,"summary":22765},"\u002Fdifferential-equations\u002Fpdes-fourier-bvp\u002Fheat-wave-laplace-equations","Separation of Variables: Heat, Wave, and Laplace Equations","Separation of variables replaces a partial differential equation by a pair of ordinary ones joined through a shared separation constant. Applied to the heat equation it produces the eigenvalue problem X'' + lambda X = 0, and the solution assembles as a Fourier series in the eigenfunctions. The same steps solve the wave equation, whose modes are standing waves, and Laplace's equation, the steady-state limit posed on a region rather than an interval.\n",{"path":22767,"title":22768,"module":22760,"summary":22769},"\u002Fdifferential-equations\u002Fpdes-fourier-bvp\u002Fsturm-liouville","Sturm-Liouville Theory","The eigenvalue problem behind separation of variables generalizes to the self-adjoint Sturm-Liouville form. Lagrange's identity makes the operator symmetric, and from that one fact follow real eigenvalues, orthogonal eigenfunctions, and eigenfunction expansions that behave like Fourier series. Singular problems admit Bessel and Legendre functions, and Sturm's separation and comparison theorems describe how the eigenfunctions oscillate.\n",{"path":22771,"title":22772,"module":22773,"summary":22774},"\u002Fdifferential-equations\u002Fhistory-variations\u002Fcalculus-of-variations","The Calculus of Variations","Historical Notes and the Calculus of Variations","Ordinary calculus finds the point where a function is stationary; the calculus of variations finds the whole curve where an integral is stationary. Euler's differential equation is the necessary condition for an extremal, and it becomes integrable in three cases, solving the shortest-path, minimal-surface, and brachistochrone problems. Lagrange multipliers extend the method to isoperimetric constraints, and Hamilton's principle recovers Newton's law from a single stationary integral.\n",{"path":22776,"title":22777,"module":22773,"summary":22778},"\u002Fdifferential-equations\u002Fhistory-variations\u002Fhistorical-notes","Great Problems and the People Who Solved Them","Differential equations grew out of specific problems, not a plan: the invention of calculus by Newton and Leibniz, the Bernoulli brachistochrone challenge, Euler's flood of methods, Lagrange's analytical mechanics, Gauss and Riemann's rigor, Laplace's celestial mechanics, and Poincaré's qualitative theory. Each method descends from a named problem, and reading the subject forward from those problems explains why its parts fit together.\n",{"path":22780,"title":22781,"module":6,"summary":6},"\u002Fdifferential-equations","Differential Equations",{"path":22783,"title":22784,"module":22785,"summary":22786},"\u002Frelativity\u002Ffoundations\u002Fspecial-relativity-postulates","The Postulates of Special Relativity","Foundations of Relativity","Newton's laws are the same in every inertial frame, but Maxwell's are not: the equations of electromagnetism single out one speed, c, and the nineteenth century read that as the speed of light relative to a medium, the ether. The Michelson-Morley experiment looked for Earth's motion through that medium and found nothing. Einstein's two postulates replace the ether, and their first consequence is that simultaneity is frame-dependent.\n",{"path":22788,"title":22789,"module":22785,"summary":22790},"\u002Frelativity\u002Ffoundations\u002Florentz-transformation-spacetime","The Lorentz Transformation and Spacetime","Requiring that a light sphere stay a light sphere in every inertial frame fixes the coordinate change between frames uniquely: the Lorentz transformation, with its factor gamma. Differentiating it gives relativistic velocity addition, which caps composed speeds at c. Plotting the same events on skewed spacetime axes turns the algebra into geometry, with calibration hyperbolae, an invariant interval, and a light cone that sorts events into past, future, and elsewhere.\n",{"path":22792,"title":22793,"module":22785,"summary":22794},"\u002Frelativity\u002Ffoundations\u002Ftime-dilation-length-contraction","Time Dilation, Length Contraction, and Paradoxes","A light clock and the constancy of c give the two headline effects directly: a moving clock runs slow by gamma, and a moving rod is short by the same factor. Cosmic-ray muons reaching sea level are the standing experimental proof. The relativistic Doppler effect adds the time-dilation factor to the classical shift, and the twin and pole-barn paradoxes dissolve once the relativity of simultaneity is taken seriously.\n",{"path":22796,"title":22797,"module":22785,"summary":22798},"\u002Frelativity\u002Ffoundations\u002Frelativistic-momentum-energy","Relativistic Momentum and Energy","Conserving momentum in every inertial frame forces the redefinition p = gamma m u, which diverges as the speed approaches c. Integrating the corresponding force gives the total energy E = gamma m c-squared, whose rest term m c-squared is Einstein's mass-energy equivalence. Energy and momentum join into a four-vector whose invariant length is the rest energy, giving E-squared = (pc)-squared + (m c-squared)-squared, massless particles, and nuclear binding energy.\n",{"path":22800,"title":22801,"module":22785,"summary":22802},"\u002Frelativity\u002Ffoundations\u002Fgeneral-relativity","A Taste of General Relativity","Einstein's happiest thought was that a freely falling observer feels no gravity: a uniform gravitational field is locally indistinguishable from an accelerating frame. That equivalence principle predicts that light bends near a mass, that clocks run slow deep in a gravitational well, that Mercury's orbit precesses, and that radar echoes are delayed. Every prediction has been confirmed, and pushing the redshift to its limit gives the black hole.\n",{"path":22804,"title":22805,"module":22806,"summary":22807},"\u002Frelativity\u002Fspacetime-and-the-lorentz-group\u002Fminkowski-spacetime-and-the-interval","Minkowski Spacetime and the Interval","Spacetime and the Lorentz Group","The Lorentz transformation of the foundations module is repackaged as the geometry of a four-dimensional space whose invariant is not a distance but the spacetime interval. Events, worldlines, and the metric signature define a causal structure that every observer shares. Proper time is the length of a timelike worldline, and the twin paradox becomes the statement that a straight worldline accumulates the most proper time.\n",{"path":22809,"title":22810,"module":22806,"summary":22811},"\u002Frelativity\u002Fspacetime-and-the-lorentz-group\u002Ffour-vectors-and-index-notation","Four-Vectors and Index Notation","The index calculus that the rest of the course runs on. Contravariant and covariant components, the Minkowski metric as the machine that raises and lowers indices, and the Einstein summation convention are assembled into scalar products that are the same in every frame. The four-velocity and four-acceleration follow, together with the identity that the four-velocity has constant invariant length.\n",{"path":22813,"title":22814,"module":22806,"summary":22815},"\u002Frelativity\u002Fspacetime-and-the-lorentz-group\u002Fthe-lorentz-group-and-rapidity","The Lorentz Group and Rapidity","The Lorentz transformations are the linear maps that preserve the Minkowski metric, and they form the group O(1,3). Boosts are hyperbolic rotations parametrized by rapidity, which adds along a line where velocity does not. The boost and rotation generators fix the group's local structure; its four disconnected components are set by two signs; and two non-collinear boosts compose into a boost plus a rotation, the Wigner rotation behind Thomas precession.\n",{"path":22817,"title":22818,"module":22806,"summary":22819},"\u002Frelativity\u002Fspacetime-and-the-lorentz-group\u002Fdoppler-aberration-and-appearance","Doppler, Aberration, and Appearance","Light carries a null four-momentum, and boosting it produces every optical effect of relativity at once. The covariant Doppler formula follows from the transformation of frequency, aberration from the transformation of direction, and the headlight effect from the resulting concentration of light forward. The Terrell-Penrose result shows that a fast object photographs as rotated, not contracted.\n",{"path":22821,"title":22822,"module":22823,"summary":22824},"\u002Frelativity\u002Frelativistic-dynamics\u002Ffour-momentum-force-and-accelerated-motion","Four-Momentum, Four-Force, and Accelerated Motion","Relativistic Dynamics","The four-momentum packages energy and momentum into a single vector whose invariant length is the rest mass. Its proper-time derivative is the four-force, always orthogonal to the four-velocity, and a constant orthogonal four-force produces hyperbolic motion. Constant proper acceleration gives rapidity linear in proper time, the relativistic rocket equation, and the Rindler horizon behind an eternally accelerating observer.\n",{"path":22826,"title":22827,"module":22823,"summary":22828},"\u002Frelativity\u002Frelativistic-dynamics\u002Fparticle-decays-and-two-body-kinematics","Particle Decays and Two-Body Kinematics","Conservation of four-momentum fixes the kinematics of a decay from the masses alone. In the center-of-momentum frame a parent breaks into two daughters with equal and opposite momenta and energies set by the Kallen triangle function. Boosting to the lab opens the decay into a cone, and the invariant mass built from the daughters reconstructs the parent as a peak. Worked cases: the two-photon decay of the neutral pion and a heavy two-body hadronic decay.\n",{"path":22830,"title":22831,"module":22823,"summary":22832},"\u002Frelativity\u002Frelativistic-dynamics\u002Fcollisions-thresholds-and-the-cm-frame","Relativistic Collisions and Threshold Energies","Two-body collisions run on the same conserved four-momentum as decays. The invariant s sets the total energy available in the center-of-momentum frame and therefore the threshold for producing new particles. Fixed-target energy grows only as the square root of beam energy while a collider grows linearly, which is why colliders reach high energy. Compton scattering follows as a worked photon-electron collision giving the wavelength shift.\n",{"path":22834,"title":22835,"module":22823,"summary":22836},"\u002Frelativity\u002Frelativistic-dynamics\u002Fmandelstam-variables-and-invariants","Mandelstam Variables and Lorentz Invariants","For a two-to-two process the three Mandelstam invariants s, t, and u encode all the kinematics in frame-independent form. They obey a single linear constraint, the sum of the four squared masses, so only two are independent. s is the center-of-momentum energy squared, t and u are momentum transfers tied to the scattering angle, and crossing symmetry relates one amplitude across three channels through these variables.\n",{"path":22838,"title":22839,"module":22840,"summary":22841},"\u002Frelativity\u002Fcovariant-electrodynamics\u002Ffour-current-and-the-four-potential","The Four-Current and Four-Potential","Covariant Electromagnetism","Charge density and current combine into a single four-vector whose divergence is charge conservation. The scalar and vector potentials combine likewise into the four-potential, whose gauge freedom fixes to the Lorenz condition, reducing Maxwell's equations for the potentials to a single wave equation sourced by the four-current.\n",{"path":22843,"title":22844,"module":22840,"summary":22845},"\u002Frelativity\u002Fcovariant-electrodynamics\u002Fthe-electromagnetic-field-tensor","The Electromagnetic Field Tensor","The antisymmetric derivative of the four-potential is the field-strength tensor F, gauge invariant by construction, with the electric and magnetic fields as its components. Its dual exchanges E and B, and its two contractions form the Lorentz invariants that classify a field as electric, magnetic, or radiative in every frame.\n",{"path":22847,"title":22848,"module":22840,"summary":22849},"\u002Frelativity\u002Fcovariant-electrodynamics\u002Ftransformation-of-electric-and-magnetic-fields","How E and B Transform","Transforming the field tensor under a boost gives explicit rules for the electric and magnetic fields: components along the motion are unchanged, transverse components mix and pick up a gamma. The field of a uniformly moving charge compresses transversely, and the force between a current and a moving charge shows that magnetism is the relativistic shadow of electrostatics.\n",{"path":22851,"title":22852,"module":22840,"summary":22853},"\u002Frelativity\u002Fcovariant-electrodynamics\u002Fcovariant-maxwell-and-the-stress-energy-tensor","Covariant Maxwell and the Stress–Energy Tensor","Maxwell's four equations collapse into two tensor equations, one sourced by the four-current and one an identity on the field strength, with charge conservation automatic. The Lorentz force becomes a four-vector law, and the field's energy, momentum, and stress assemble into a symmetric, conserved stress–energy tensor — the object that will source gravity.\n",{"path":22855,"title":22856,"module":22857,"summary":22858},"\u002Frelativity\u002Fcurved-spacetime\u002Fthe-equivalence-principle-formalized","The Equivalence Principle","Curved Spacetime","The equality of gravitational and inertial mass promotes to a physical principle in three graded strengths — weak, Einstein, and strong. A freely falling laboratory is locally indistinguishable from an inertial frame, but the qualifier \"locally\" is essential: the size of the patch over which gravity vanishes is set by the tidal field, which no change of frame can remove. Tidal forces are the true, coordinate-independent signature of gravity, and they are what curvature will measure.\n",{"path":22860,"title":22861,"module":22857,"summary":22862},"\u002Frelativity\u002Fcurved-spacetime\u002Fmanifolds-vectors-and-the-metric","Manifolds, Vectors, and the Metric","A manifold is a space that looks locally like flat space, described by overlapping coordinate charts. Tangent vectors are directional derivatives with the coordinate basis vectors as partial-derivative operators; one-forms live in the dual space; and the metric tensor turns a coordinate line element into an invariant length. The 2-sphere and Rindler metrics serve as worked examples, including the coordinate singularities that are artefacts of the chart, not of the geometry.\n",{"path":22864,"title":22865,"module":22857,"summary":22866},"\u002Frelativity\u002Fcurved-spacetime\u002Fcovariant-derivative-and-christoffel-symbols","Parallel Transport and the Covariant Derivative","The ordinary derivative of a vector field is not a tensor, because it subtracts vectors living in different tangent spaces. A connection supplies the missing comparison: the covariant derivative adds Christoffel-symbol correction terms that cancel the coordinate artefacts. Requiring the connection to be torsion-free and to preserve the metric fixes the Christoffel symbols uniquely in terms of derivatives of the metric, giving the Levi-Civita connection that general relativity uses.\n",{"path":22868,"title":22869,"module":22857,"summary":22870},"\u002Frelativity\u002Fcurved-spacetime\u002Fgeodesics-and-the-geodesic-equation","Geodesics and the Newtonian Limit","Free fall is geodesic motion: a freely falling particle follows the straightest possible worldline, obtained either by parallel-transporting its own tangent vector or by extremizing proper time. Both routes give the geodesic equation. Affine parameters, and conserved quantities from symmetries via Killing vectors, make it solvable. In the weak-field slow-motion limit the geodesic equation reproduces Newton's law of gravity, fixing the time-time metric component as the Newtonian potential.\n",{"path":22872,"title":22873,"module":22857,"summary":22874},"\u002Frelativity\u002Fcurved-spacetime\u002Fcurvature-riemann-and-geodesic-deviation","Curvature and the Riemann Tensor","Curvature is the failure of parallel transport to commute: carrying a vector around an infinitesimal loop returns it rotated, and the rotation per unit area is the Riemann tensor. Its symmetries cut the components to twenty in four dimensions. Geodesic deviation makes it the equation of tidal forces, and its contractions — the Ricci tensor, the Ricci scalar, and the divergence-free Einstein tensor — assemble the objects the field equation is built from.\n",{"path":22876,"title":22877,"module":22857,"summary":22878},"\u002Frelativity\u002Fcurved-spacetime\u002Fthe-einstein-field-equations","The Einstein Field Equations","The field equation is assembled from a short list of requirements: a symmetric, divergence-free, second-order geometric tensor set proportional to the stress–energy tensor, with the coefficient fixed by the Newtonian limit. The cosmological constant is the one extra term the requirements allow. The Einstein–Hilbert action gives the same equation from a variational principle, and the coupled system closes the logic of the module: matter curves spacetime, and spacetime tells matter how to move.\n",{"path":22880,"title":22881,"module":22882,"summary":22883},"\u002Frelativity\u002Fthe-schwarzschild-solution\u002Fthe-schwarzschild-metric","The Schwarzschild Metric","The Schwarzschild Solution","The first exact solution of Einstein's equation follows from two assumptions, staticity and spherical symmetry, imposed on the vacuum outside a mass. Solving the vacuum field equations fixes two metric functions and produces the Schwarzschild geometry, whose one length scale is the Schwarzschild radius $r_s = 2GM\u002Fc^2$. Birkhoff's theorem shows this is the only spherical vacuum, and the far field reduces to Newtonian gravity.\n",{"path":22885,"title":22886,"module":22882,"summary":22887},"\u002Frelativity\u002Fthe-schwarzschild-solution\u002Fgeodesics-and-orbits-in-schwarzschild","Orbits in the Schwarzschild Geometry","The two Killing symmetries of the Schwarzschild metric give a conserved energy and angular momentum per unit mass, reducing geodesic motion to a one-dimensional problem in an effective potential. The potential carries an extra attractive $1\u002Fr^3$ term absent from Newton's, which caps the centrifugal barrier, produces an innermost stable circular orbit at $6GM\u002Fc^2$, and makes bound orbits precess instead of closing.\n",{"path":22889,"title":22890,"module":22882,"summary":22891},"\u002Frelativity\u002Fthe-schwarzschild-solution\u002Flight-bending-and-null-geodesics","Null Geodesics and the Photon Sphere","Light follows null geodesics, governed by a photon effective potential with a single unstable maximum at $3GM\u002Fc^2$, the photon sphere. The impact parameter sorts rays into those that escape with a deflection and those captured, with the critical value $b_c = 3\\sqrt{3}\\,GM\u002Fc^2$ dividing them. A grazing ray bends by $4GM\u002F(c^2 b)$, twice the naive Newtonian value, and the critical impact parameter sets the edge of a black hole's shadow.\n",{"path":22893,"title":22894,"module":22895,"summary":22896},"\u002Frelativity\u002Ftests-of-general-relativity\u002Fperihelion-precession-of-mercury","The Perihelion Precession of Mercury","Tests of General Relativity","A single extra term in the Schwarzschild orbit equation, cubic in the inverse radius, keeps a bound orbit from closing. The perturbation advances the perihelion by 6πGM\u002F(c²a(1−e²)) per revolution, which for Mercury is 43 arcseconds per century — exactly the anomaly left after Newtonian planetary perturbations are subtracted. A note on frame dragging closes the lesson.\n",{"path":22898,"title":22899,"module":22895,"summary":22900},"\u002Frelativity\u002Ftests-of-general-relativity\u002Fdeflection-of-light-and-gravitational-lensing","Light Deflection and Gravitational Lensing","A light ray grazing the Sun bends by 4GM\u002F(c²b), exactly twice the value a Newtonian corpuscle would give; the extra factor is the curvature of space. The 1919 eclipse confirmed it. The same bending focuses light from distant sources into Einstein rings, multiple images, and microlensing brightenings, making lensing a direct probe of mass, including mass that emits no light.\n",{"path":22902,"title":22903,"module":22895,"summary":22904},"\u002Frelativity\u002Ftests-of-general-relativity\u002Fgravitational-redshift-and-shapiro-delay","Gravitational Redshift and the Shapiro Delay","A clock deeper in a gravitational well ticks slower, and a photon climbing out loses frequency by the ratio of the metric's time-time components. Pound and Rebka measured the 2.5×10⁻¹⁵ shift over a 22.5-metre tower. Radar signals grazing the Sun return late by about 250 microseconds, the Shapiro delay. Both probe the time part of the metric directly.\n",{"path":22906,"title":22907,"module":22895,"summary":22908},"\u002Frelativity\u002Ftests-of-general-relativity\u002Frelativity-in-technology-gps","Relativity and the Global Positioning System","A GPS satellite clock runs slow by 7 microseconds a day from its orbital speed and fast by 46 from its higher gravitational potential, a net gain of about 38 microseconds a day. Left uncorrected, the timing error would grow into kilometres of position error within a day and exceed navigation tolerance within minutes. The satellites carry a pre-launch frequency offset to cancel it.\n",{"path":22910,"title":22911,"module":22912,"summary":22913},"\u002Frelativity\u002Fblack-holes\u002Fhorizons-and-coordinate-singularities","Horizons and Coordinate Singularities","Black Holes","The Schwarzschild radius is a coordinate singularity, not a curvature singularity: the metric blows up there only because the static coordinates fail, while the geometry stays finite. Eddington–Finkelstein and Kruskal– Szekeres coordinates cross the horizon smoothly and show the light cones tipping toward the center. A freely falling observer reaches the true singularity at r=0 in finite proper time, while a distant observer sees the infall freeze and redden at the horizon.\n",{"path":22915,"title":22916,"module":22912,"summary":22917},"\u002Frelativity\u002Fblack-holes\u002Frotating-and-charged-black-holes","Rotating and Charged Black Holes","A stationary black hole is fixed by three numbers: mass, angular momentum, and charge. The Reissner–Nordström metric adds charge and splits the horizon in two; the Kerr metric adds rotation, drags inertial frames, and wraps the horizon in an ergosphere where nothing can stay still. Inside the ergosphere the Penrose process extracts rotational energy, and the no-hair theorem states that no other detail of the collapsed matter survives.\n",{"path":22919,"title":22920,"module":22912,"summary":22921},"\u002Frelativity\u002Fblack-holes\u002Fblack-hole-thermodynamics","Black-Hole Thermodynamics","The four laws of black-hole mechanics mirror the four laws of thermodynamics term for term, with horizon area playing the role of entropy and surface gravity the role of temperature. Hawking's calculation makes the analogy literal: a black hole radiates at a temperature set by its surface gravity, carries a real entropy proportional to its horizon area, and slowly evaporates. The thermal spectrum raises the information paradox.\n",{"path":22923,"title":22924,"module":22925,"summary":22926},"\u002Frelativity\u002Fgravitational-waves\u002Flinearized-gravity-and-wave-solutions","Linearized Gravity and Wave Solutions","Gravitational Waves","Weak gravity is a small perturbation of flat spacetime, and the linearized Einstein equation in the Lorenz gauge is an ordinary wave equation propagating at the speed of light. The trace-reversed perturbation carries the dynamics, residual gauge freedom fixes the transverse-traceless form, and the two physical polarizations deform a ring of freely falling masses into oscillating ellipses whose fractional size change is the strain.\n",{"path":22928,"title":22929,"module":22925,"summary":22930},"\u002Frelativity\u002Fgravitational-waves\u002Fgeneration-and-the-quadrupole-formula","The Quadrupole Formula","The retarded solution of the linearized field equation gives the field of a moving source, and conservation of mass and momentum forbids monopole and dipole radiation, leaving the mass quadrupole as the leading emitter. The quadrupole formula fixes the strain and the radiated luminosity, and applied to a compact binary it predicts the inspiral chirp of rising frequency and amplitude. The Hulse-Taylor pulsar's orbital decay confirmed it to a fraction of a percent.\n",{"path":22932,"title":22933,"module":22925,"summary":22934},"\u002Frelativity\u002Fgravitational-waves\u002Fdetection-ligo-and-the-first-events","LIGO and the First Detections","A gravitational wave is measured as a differential length change of the two arms of a kilometre-scale Michelson interferometer, a strain of order ten to the minus twenty-one that moves the mirrors by a fraction of a proton radius. GW150914 recorded the inspiral, merger, and ringdown of two black holes, fixing their masses and the energy radiated, and GW170817 with its coincident gamma-ray burst and kilonova opened multimessenger astronomy.\n",{"path":22936,"title":22937,"module":22938,"summary":22939},"\u002Frelativity\u002Fcosmological-bridge\u002Fthe-cosmological-principle-and-flrw-metric","The Cosmological Principle and the FLRW Metric","A Bridge to Cosmology","Homogeneity and isotropy restrict the spacetime of the universe to a single family of metrics: a flat cosmic-time slicing of spatial sections of constant curvature, scaled by a time-dependent factor a(t). This lesson builds the Friedmann–Lemaître–Robertson–Walker metric from those symmetries, separates comoving from proper distance, and derives cosmological redshift as the stretching of wavelengths with the scale factor.\n",{"path":22941,"title":22942,"module":22938,"summary":22943},"\u002Frelativity\u002Fcosmological-bridge\u002Ffriedmann-equations-and-cosmic-dynamics","The Friedmann Equations and Cosmic Dynamics","The Einstein equation applied to the FLRW metric with a perfect-fluid source yields the two Friedmann equations and the conservation law that ties them together. This lesson derives them, defines the critical density and the density parameters that fix the spatial geometry, works out how matter, radiation, and a cosmological constant dilute and drive the expansion, and hands off to a dedicated cosmology subject.\n",{"path":22945,"title":22946,"module":6,"summary":6},"\u002Frelativity","Relativity",{"path":22948,"title":22949,"module":6,"summary":6},"\u002Fphysical-computing","Physical Computing",{"path":22951,"title":22952,"module":22953,"summary":22954},"\u002Fquantum-mechanics\u002Fold-quantum-theory\u002Fblackbody-radiation-and-the-planck-quantum","Blackbody Radiation and the Planck Quantum","Origins of the Quantum","Millikan's oil-drop experiment fixed the electron charge as an indivisible unit, and the spectrum of thermal radiation forced a second, deeper quantum. Classical physics predicts an infinite energy density at short wavelengths; Planck removed the divergence by allowing a cavity oscillator to hold only energies that are integer multiples of hf, the first appearance of the quantum of action.\n",{"path":22956,"title":22957,"module":22953,"summary":22958},"\u002Fquantum-mechanics\u002Fold-quantum-theory\u002Fthe-photoelectric-effect-and-the-photon","The Photoelectric Effect and the Photon","Light shone on a clean metal ejects electrons, but the details defied the wave theory: the electrons' maximum energy depends on the light's frequency, not its brightness, and there is a sharp threshold frequency below which nothing happens. Einstein resolved every anomaly by treating light as a stream of energy quanta hf, each absorbed whole by one electron, and Millikan's measurement of the stopping-potential slope confirmed h to a decade before anyone expected.\n",{"path":22960,"title":22961,"module":22953,"summary":22962},"\u002Fquantum-mechanics\u002Fold-quantum-theory\u002Fx-rays-and-the-compton-effect","X-Rays and the Compton Effect","X-rays are short-wavelength electromagnetic waves produced when fast electrons are braked in a target, and their diffraction by crystals lets Bragg's law measure atomic spacings. Compton then scattered X-rays off electrons and found the wavelength shifted by an amount that only a photon carrying momentum hf\u002Fc could explain, closing the case for the particle nature of light.\n",{"path":22964,"title":22965,"module":22953,"summary":22966},"\u002Fquantum-mechanics\u002Fold-quantum-theory\u002Fthe-old-quantum-theory-bohr-and-sommerfeld","The Old Quantum Theory: Bohr, Sommerfeld, and Correspondence","Between Bohr's 1913 atom and Schrödinger's 1926 equation, physics ran on a provisional recipe: keep classical orbits, but admit only those whose action integral is a whole multiple of Planck's constant. This lesson develops the Wilson-Sommerfeld phase-integral rule, applies it to the oscillator and to the elliptical Kepler orbits of hydrogen, derives Sommerfeld's relativistic fine structure and the quantization of orbit orientation, and shows how the correspondence principle fixed intensities and selection rules. The systematic failures — helium, line intensities, the anomalous Zeeman effect — mark exactly where a theory of orbits had to give way to a theory of waves.\n",{"path":22968,"title":22969,"module":22970,"summary":22971},"\u002Fquantum-mechanics\u002Fmatter-waves\u002Fde-broglie-waves-and-electron-diffraction","De Broglie Waves and Electron Diffraction","The Wave Nature of Matter","In 1924 de Broglie proposed that every particle carries a wave of wavelength h\u002Fp. The hypothesis explains Bohr's quantized orbits as standing waves, and Davisson and Germer, then G. P. Thomson, confirmed it by diffracting electrons from crystals exactly as X-rays diffract. We derive the electron wavelength, work the Bragg analysis of the data, and give the relativistic form.\n",{"path":22973,"title":22974,"module":22970,"summary":22975},"\u002Fquantum-mechanics\u002Fmatter-waves\u002Fwave-packets-and-the-probability-interpretation","Wave Packets and the Probabilistic Wave Function","A single de Broglie wave fills all space, but a particle is localized. Adding many waves of nearby wavelength builds a wave packet that is confined and moves at the group velocity, which equals the particle velocity. Born's rule reads the squared amplitude of the wave function as a probability density, the meaning confirmed by electron interference building up one detection at a time.\n",{"path":22977,"title":22978,"module":22970,"summary":22979},"\u002Fquantum-mechanics\u002Fmatter-waves\u002Fthe-uncertainty-principle","The Uncertainty Principle and Wave-Particle Duality","The packet relations delta-k delta-x about 1 become Heisenberg's principle once momentum is hbar times wave number: position and momentum cannot both be sharp, nor energy and time. The gamma-ray microscope shows the limit is physical, not technical. It fixes the zero-point energy of a confined particle, the size of the hydrogen atom, and the natural width of spectral lines, and it frames the wave-particle duality of all matter and radiation.\n",{"path":22981,"title":22982,"module":22983,"summary":22984},"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fthe-schrodinger-equation-in-one-dimension","The Schrödinger Equation in One Dimension","Wave Mechanics in One Dimension","The wave equation for matter cannot be derived; it is postulated and judged by experiment. We build the time-dependent Schrödinger equation from the de Broglie relations, read Born's probability rule off the complex wave function, and separate the time and space dependence to get the time-independent equation whose bound-state solutions are the stationary states. The five acceptability conditions on the wave function are what force energy to be quantized.\n",{"path":22986,"title":22987,"module":22983,"summary":22988},"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fthe-free-particle-and-wave-packet-dynamics","The Free Particle and Wave-Packet Dynamics","The free particle has no bound states: its stationary solutions are non-normalizable plane waves forming a continuum. Physical states are wave packets built by superposing them, and the superposition is a Fourier transform. We delta-normalize the plane waves, assemble a Gaussian packet, solve for its exact time evolution, and read off the two facts that reconcile the wave picture with mechanics: the packet moves at the group velocity ħk\u002Fm, the classical velocity, and it spreads because its component momenta travel at different speeds.\n",{"path":22990,"title":22991,"module":22983,"summary":22992},"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fparticle-in-infinite-and-finite-square-wells","Particle in Infinite and Finite Square Wells","The infinite square well is the simplest bound-state problem: two boundary conditions quantize the energy into a ladder E_n = n² E_1, and the eigenfunctions are the standing waves of a string fixed at both ends. Relaxing the walls to a finite depth lets the wave function leak into the classically forbidden region, keeps the number of bound states finite, and turns the eigenvalue condition into a transcendental equation solved graphically.\n",{"path":22994,"title":22995,"module":22983,"summary":22996},"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Foperators-expectation-values-and-the-harmonic-oscillator","Operators, Expectation Values, and the Harmonic Oscillator","Measurable quantities are extracted from the wave function as expectation values, and each observable is represented by an operator that acts between Ψ* and Ψ — position by multiplication, momentum by a derivative, energy by the Hamiltonian. Applied to the harmonic oscillator, the machinery yields evenly spaced levels E_n = (n+½)ℏω, Gaussian- times-Hermite eigenfunctions of definite parity, and the selection rule Δn = ±1.\n",{"path":22998,"title":22999,"module":22983,"summary":23000},"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fthe-dirac-delta-potential","The Dirac-Delta Potential: A Single Bound State and Scattering","A potential concentrated at a single point is solvable in closed form and isolates the physics of matching a wave function across a discontinuity. Integrating the Schrödinger equation across the spike gives a jump condition on the derivative; the attractive delta well then supports exactly one bound state, of energy set by the strength alone, while the same spike scatters an incoming beam with a transmission that rises from zero to one. The attractive well and the repulsive barrier scatter identically yet only the well binds.\n",{"path":23002,"title":23003,"module":22983,"summary":23004},"\u002Fquantum-mechanics\u002Fwave-mechanics-1d\u002Fbarrier-penetration-and-quantum-tunneling","Barrier Penetration and Quantum Tunneling","Unbound states scatter rather than bind. A particle meeting a step is partly reflected even when it has more than enough energy to pass, and a particle meeting a barrier taller than its energy has a nonzero chance of appearing on the far side. Matching the wave function across the boundaries gives the reflection and transmission coefficients and the exponential tunneling probability that explains alpha decay, the scanning tunneling microscope, and the ammonia clock.\n",{"path":23006,"title":23007,"module":23008,"summary":23009},"\u002Fquantum-mechanics\u002Fformalism\u002Fhilbert-space-and-dirac-notation","Hilbert Space and Dirac Bra–Ket Notation","The Formalism of Quantum Mechanics","Wave mechanics is one representation of a deeper structure: quantum states are vectors in a complex inner-product space, and observables act on them as linear operators. We build that space from the axioms, introduce Dirac's kets and bras as vectors and the linear functionals that measure them, and identify the wavefunction as the components of an abstract state in the position basis. The resolution of the identity is the single algebraic tool that ties every basis, expansion, and matrix element together.\n",{"path":23011,"title":23012,"module":23008,"summary":23013},"\u002Fquantum-mechanics\u002Fformalism\u002Fobservables-hermitian-operators-and-eigenvalues","Observables, Hermitian Operators, and the Spectral Theorem","Every measurable quantity is represented by a Hermitian operator, and the reason is forced: a measurement needs real eigenvalues, orthogonal eigenvectors, and a complete eigenbasis, and Hermiticity delivers precisely those. We derive those properties from self-adjointness, state the spectral theorem, handle degeneracy, and show that two observables share an eigenbasis precisely when they commute — the algebraic condition behind compatible and incompatible measurements.\n",{"path":23015,"title":23016,"module":23008,"summary":23017},"\u002Fquantum-mechanics\u002Fformalism\u002Fthe-postulates-and-quantum-measurement","The Postulates and Quantum Measurement","With states as vectors and observables as Hermitian operators, the physical content of quantum mechanics reduces to a short list of postulates. We state them precisely, derive the Born probability rule for discrete and continuous spectra, work out projective collapse and its idempotence, compute expectation values and their variance, and state the measurement problem cleanly — the one place the postulates split unitary evolution from measurement without explaining the seam.\n",{"path":23019,"title":23020,"module":23008,"summary":23021},"\u002Fquantum-mechanics\u002Fformalism\u002Fposition-momentum-and-continuous-spectra","Position, Momentum, and Continuous Spectra","Position and momentum are the observables with no normalizable eigenstates: their spectra are continuous, their eigenkets are delta-normalized, and the two are Fourier conjugates. We derive the canonical commutator from the momentum operator, build the continuous-basis machinery (Dirac deltas replacing Kronecker deltas), show the position and momentum wavefunctions are a Fourier-transform pair, and compute expectation values in either representation.\n",{"path":23023,"title":23024,"module":23008,"summary":23025},"\u002Fquantum-mechanics\u002Fformalism\u002Fcommutators-and-the-generalized-uncertainty-principle","Commutators and the Generalized Uncertainty Principle","The commutator of two observables measures the obstruction to sharing an eigenbasis, and it bounds how sharply both can be known at once. We derive the generalized uncertainty relation from the Schwarz inequality, recover the position–momentum bound as a special case, characterize the minimum-uncertainty states that saturate it as Gaussians, and give the energy–time relation its correct reading as a lifetime bound rather than a commutator relation.\n",{"path":23027,"title":23028,"module":23008,"summary":23029},"\u002Fquantum-mechanics\u002Fformalism\u002Ftime-evolution-schrodinger-and-heisenberg-pictures","Time Evolution, Propagators, and the Heisenberg Picture","Time evolution is generated by the Hamiltonian and implemented by a unitary operator that preserves probability. We build that operator, expand a state in stationary states to see why probability densities freeze while phases wind, introduce the propagator, transfer the time dependence onto operators in the Heisenberg picture, and derive Ehrenfest's theorem — which recovers classical equations of motion for expectation values and identifies conserved quantities as observables commuting with the Hamiltonian.\n",{"path":23031,"title":23032,"module":23033,"summary":23034},"\u002Fquantum-mechanics\u002Foscillator-and-symmetry\u002Fladder-operators-and-the-number-states","Ladder Operators and the Number States","The Oscillator Algebraically, and Symmetry","The harmonic oscillator can be solved without touching a differential equation. Factoring the Hamiltonian into a lowering operator and its adjoint turns the spectrum into pure algebra: the commutator relation fixes the ladder, the vacuum condition fixes the ground state, and the energies fall out as equally spaced rungs. The same operators give the matrix elements of position and momentum for free.\n",{"path":23036,"title":23037,"module":23033,"summary":23038},"\u002Fquantum-mechanics\u002Foscillator-and-symmetry\u002Fcoherent-and-squeezed-states","Coherent and Squeezed States","A single number state never moves — its position expectation is pinned at the origin. The superposition that oscillates like a classical particle is the eigenstate of the annihilation operator: the coherent state. It is a displaced vacuum, carries Poissonian photon statistics, saturates the uncertainty bound, and traces a rigid Gaussian orbit in phase space. Squeezing deforms that circle, trading precision in one quadrature for noise in the other.\n",{"path":23040,"title":23041,"module":23033,"summary":23042},"\u002Fquantum-mechanics\u002Foscillator-and-symmetry\u002Fsymmetries-generators-and-conservation-laws","Symmetries, Generators, and Conservation Laws","Every continuous symmetry of a quantum system is a unitary operator built by exponentiating a Hermitian generator: momentum generates translations, angular momentum generates rotations, the Hamiltonian generates time evolution. When a generator commutes with the Hamiltonian, the transformation leaves the dynamics unchanged and the generator is conserved — the quantum form of Noether's theorem — and any symmetry that mixes states within a level forces degeneracy.\n",{"path":23044,"title":23045,"module":23033,"summary":23046},"\u002Fquantum-mechanics\u002Foscillator-and-symmetry\u002Fparity-time-reversal-and-discrete-symmetries","Parity, Time Reversal, and Discrete Symmetries","Parity and time reversal are symmetries no continuous generator can reach. Parity is a unitary involution whose eigenvalues label states even or odd, fixing the dipole selection rules. Time reversal is antiunitary: it conjugates i, flips momenta and spins, and for half-integer spin squares to minus one, which by Kramers' theorem makes every level of a time-reversal-invariant Hamiltonian at least doubly degenerate.\n",{"path":23048,"title":23049,"module":21888,"summary":23050},"\u002Fquantum-mechanics\u002Fangular-momentum\u002Forbital-angular-momentum-and-spherical-harmonics","Orbital Angular Momentum and Spherical Harmonics","Orbital angular momentum is the operator triple built from position and momentum. Its components fail to commute, so no state carries sharp values of more than one of them, but each commutes with the total square. Solving the common eigenvalue problem in spherical coordinates quantizes both the magnitude and the projection and produces the spherical harmonics, the angular part of every central-force wavefunction.\n",{"path":23052,"title":23053,"module":21888,"summary":23054},"\u002Fquantum-mechanics\u002Fangular-momentum\u002Fthe-angular-momentum-algebra","The Angular-Momentum Algebra and Ladder Operators","The eigenvalues of angular momentum follow from the commutation relations alone, with no reference to coordinates or wavefunctions. Raising and lowering operators built from the components generate finite multiplets, force the total quantum number to be a non-negative integer or half-integer, and fix the matrix elements of every component. The half-integer values excluded by orbital motion appear here, and they are what spin realizes.\n",{"path":23056,"title":23057,"module":21888,"summary":23058},"\u002Fquantum-mechanics\u002Fangular-momentum\u002Faddition-of-angular-momenta-and-clebsch-gordan","Addition of Angular Momenta and Clebsch–Gordan Coefficients","Two angular momenta combine into a total whose allowed magnitudes run from the difference to the sum of the parts in integer steps. The change from the uncoupled product basis to the coupled total-angular-momentum basis is carried out with the lowering operator and orthogonality, and its matrix of overlaps is the table of Clebsch–Gordan coefficients. Two spin-halves split into a triplet and a singlet, the prototype for every composite spin.\n",{"path":23060,"title":23061,"module":23062,"summary":23063},"\u002Fquantum-mechanics\u002Fcentral-potentials\u002Fthe-schrodinger-equation-in-three-dimensions","The Schrödinger Equation in Three Dimensions","Central Potentials","A central potential depends only on the distance from a force center, so the three-dimensional Schrödinger equation separates in spherical coordinates. The angular factor is a spherical harmonic; the radial factor obeys a one-dimensional equation with an effective potential whose centrifugal barrier depends on the angular-momentum quantum number. The free particle and the spherical box fix the two limiting cases through the spherical Bessel functions.\n",{"path":23065,"title":23066,"module":23062,"summary":23067},"\u002Fquantum-mechanics\u002Fcentral-potentials\u002Fthe-hydrogen-atom","The Hydrogen Atom","The Coulomb potential turns the radial equation into one whose bound states exist only for a discrete set of energies. A power-series solution truncated to keep the wavefunction normalizable forces the principal quantum number, and the energy comes out proportional to minus one over its square, recovering the Rydberg spectrum. The bound states are the associated Laguerre functions times spherical harmonics, and their energy depends on the principal number alone, giving an n-squared degeneracy larger than rotational symmetry can explain.\n",{"path":23069,"title":23070,"module":23062,"summary":23071},"\u002Fquantum-mechanics\u002Fcentral-potentials\u002Fthe-isotropic-oscillator-and-hidden-symmetry","The Isotropic Oscillator and Hidden Symmetry","The three-dimensional isotropic harmonic oscillator solves in both Cartesian and spherical bases, and the two solutions must agree on the degeneracy of every level. That agreement, and the accidental degeneracy of hydrogen, both come from a symmetry larger than rotation: the oscillator carries an SU(3) invariance built from a conserved quadrupole tensor, and the Coulomb problem carries an SO(4) invariance built from the conserved Runge–Lenz vector. These hidden symmetries pin the degeneracies that rotational invariance alone leaves unexplained.\n",{"path":23073,"title":23074,"module":23075,"summary":23076},"\u002Fquantum-mechanics\u002Fspin\u002Fspin-half-pauli-matrices-and-stern-gerlach","Spin-½, the Pauli Matrices, and Stern–Gerlach","Spin","A silver atom passing through an inhomogeneous magnetic field splits into two beams, not a smear. That single fact fixes the internal angular momentum of the electron to a two-valued quantity with no spatial wavefunction. We build the two-dimensional spin space, the Pauli matrices and their algebra, the spinor for measurement along an arbitrary axis, and the sequential Stern–Gerlach filters that expose measurement disturbance.\n",{"path":23078,"title":23079,"module":23075,"summary":23080},"\u002Fquantum-mechanics\u002Fspin\u002Fspin-in-a-magnetic-field-precession-and-resonance","Spin in a Magnetic Field: Precession and Resonance","A spin coupled to a magnetic field is the simplest nontrivial quantum dynamics. A static field makes the spin expectation precess on a cone at the Larmor frequency while the energy levels split linearly. Adding a weak oscillating field and passing to the rotating frame produces Rabi oscillations and a resonance lineshape — the physics of NMR and ESR, and the driven qubit.\n",{"path":23082,"title":23083,"module":23075,"summary":23084},"\u002Fquantum-mechanics\u002Fspin\u002Ftwo-level-systems-and-the-bloch-sphere","Two-Level Systems and the Bloch Sphere","Every two-state quantum system is a spin-½ in disguise. Its Hamiltonian is an effective magnetic field, its pure states are points on the Bloch sphere, and its unitary evolution is a rigid rotation of that sphere. The same structure produces avoided level crossings, the ammonia inversion doublet and its maser, and the qubit.\n",{"path":23086,"title":23087,"module":23088,"summary":23089},"\u002Fquantum-mechanics\u002Fidentical-particles\u002Fidentical-particles-and-exchange-symmetry","Identical Particles and Exchange Symmetry","Identical Particles","Two electrons carry no label that distinguishes one from the other, and that bare fact reshapes the state space. The exchange operator that swaps particle labels commutes with any Hamiltonian built from identical particles, so its eigenvalue is conserved, and nature admits only its two extremes: totally symmetric states for bosons and totally antisymmetric states for fermions. The antisymmetry forces a statistical correlation, the exchange \"force,\" that keeps fermions apart and draws bosons together even with no interaction between them.\n",{"path":23091,"title":23092,"module":23088,"summary":23093},"\u002Fquantum-mechanics\u002Fidentical-particles\u002Fthe-pauli-principle-atoms-and-the-periodic-table","The Pauli Principle, Atoms, and the Periodic Table","Antisymmetry packaged as a Slater determinant turns the exclusion principle into an operating rule for building atoms. Helium shows the machinery in full: the electron-electron repulsion splits into a direct Coulomb integral and an exchange integral, and the exchange term alone pushes the spin-triplet (orthohelium) below the spin-singlet (parahelium) with no magnetic interaction in sight. Screening, the aufbau order, and Hund's rules then assemble the whole periodic table from the same antisymmetry.\n",{"path":23095,"title":23096,"module":23097,"summary":23098},"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Ftime-independent-perturbation-theory","Time-Independent Perturbation Theory","Approximation Methods for Bound States","Almost no realistic Hamiltonian can be solved exactly. Perturbation theory treats a hard Hamiltonian as a solvable one plus a small correction and expands the eigenvalues and eigenstates in powers of that correction. We derive the first- and second-order energy shifts and the first-order state correction for a nondegenerate level, expose the small-denominator failure that degeneracy forces, and fix it by diagonalizing the perturbation inside the degenerate subspace to find the \"good\" zeroth-order states.\n",{"path":23100,"title":23101,"module":23097,"summary":23102},"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Ffine-structure-and-the-real-hydrogen-atom","Fine Structure and the Real Hydrogen Atom","The Bohr spectrum is only the leading term. Two relativistic corrections of order alpha-squared — the relativistic kinetic-energy correction and spin–orbit coupling, joined by the Darwin term for s states — split the hydrogen levels into fine structure that depends on the total angular momentum j. We derive each shift as a first-order perturbation, combine them into a formula depending only on n and j, and continue down the energy ladder to the Lamb shift and the hyperfine 21 cm line.\n",{"path":23104,"title":23105,"module":23097,"summary":23106},"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Fthe-zeeman-and-stark-effects","The Zeeman and Stark Effects","An atom in an external field is a perturbation problem whose good basis depends on which interaction wins. A magnetic field competes with the internal spin–orbit coupling: the weak-field limit gives the anomalous Zeeman splitting set by the Landé g-factor, the strong-field limit gives the Paschen–Back pattern in the uncoupled basis, and the intermediate regime is a matrix diagonalization. An electric field gives a quadratic shift for the nondegenerate ground state and a linear splitting for the degenerate n = 2 level.\n",{"path":23108,"title":23109,"module":23097,"summary":23110},"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Fthe-variational-method","The Variational Method","The expectation of the Hamiltonian in any trial state is an upper bound on the true ground-state energy. Minimizing that expectation over a parametrized family of trial functions turns the ground-state problem into ordinary calculus and needs no small parameter. We prove the bound, apply it to the helium atom with a screened effective charge, use a two-center trial to predict binding in the hydrogen molecular ion, and extend the method to excited states through orthogonality.\n",{"path":23112,"title":23113,"module":23097,"summary":23114},"\u002Fquantum-mechanics\u002Fapproximation-methods\u002Fthe-wkb-approximation","The WKB Approximation","When the potential varies slowly on the scale of the de Broglie wavelength, the wavefunction is locally a plane wave with a position-dependent wavelength. This semiclassical picture builds the wavefunction from the classical momentum, breaks down at the turning points where the momentum vanishes, and is repaired there by connection formulas. The result recovers the Bohr–Sommerfeld quantization rule with its half-integer correction and gives the exponential tunneling rate through a smooth barrier, the Gamow factor.\n",{"path":23116,"title":23117,"module":6,"summary":6},"\u002Fquantum-mechanics","Quantum Mechanics",{"path":23119,"title":23120,"module":23121,"summary":23122},"\u002Freal-analysis\u002Ffoundations\u002Fsets-logic-functions","Sets, Logic, and Functions","Foundations and the Real Number System","The working language of analysis: quantifiers and the proof patterns (contrapositive, contradiction, induction), sets and their operations, relations and equivalence classes, and functions with their images, injections, surjections, and bijections. Cardinality is measured by bijection, and Cantor's theorem that no set surjects onto its power set forces uncountable sets to exist.\n",{"path":23124,"title":23125,"module":23121,"summary":23126},"\u002Freal-analysis\u002Ffoundations\u002Fordered-fields-completeness","Ordered Fields and the Completeness Axiom","The real numbers are the unique ordered field with the least-upper-bound property. The field and order axioms, the exact failure of the rationals (no supremum for the set of rationals below √2), and completeness as the defining axiom of ℝ lead to the first consequences: the existence of √2, the Archimedean property, and the density of ℚ in ℝ.\n",{"path":23128,"title":23129,"module":23121,"summary":23130},"\u002Freal-analysis\u002Ffoundations\u002Fabsolute-value-bounds","Absolute Value, Bounded Sets, and Inequalities","The absolute value turns the order on ℝ into a notion of distance, with the triangle inequality as the estimate underlying most later proofs. Covered: its algebra, the triangle and reverse-triangle inequalities, and the extension of the sup\u002Finf vocabulary from sets to bounded functions.\n",{"path":23132,"title":23133,"module":23121,"summary":23134},"\u002Freal-analysis\u002Ffoundations\u002Fintervals-uncountability","Intervals, Uncountability, and Decimals","Intervals are classified, and ℝ is proved uncountable two ways: a nested-interval construction and the decimal diagonal argument. Decimal expansions are built as suprema of truncations, which pins the source of their non-uniqueness (the 0.4999… equals 0.5000… identity) and the identification of the rationals with the eventually-repeating expansions. The middle-thirds Cantor set is an uncountable set of measure zero.\n",{"path":23136,"title":23137,"module":23138,"summary":23139},"\u002Freal-analysis\u002Fsequences-series\u002Fsequences-limits","Sequences and Their Limits","Sequences and Series","A sequence is a function on the natural numbers; it converges to a limit when its terms eventually stay within any prescribed tolerance of that number. The epsilon-M definition fixes the order of the quantifiers, and from it the limit is unique, every convergent sequence is bounded, and only the tail matters. Divergence to plus or minus infinity records terms that outgrow every bound.\n",{"path":23141,"title":23142,"module":23138,"summary":23143},"\u002Freal-analysis\u002Fsequences-series\u002Flimit-laws-monotone","Limit Laws and Monotone Convergence","Limits commute with sums, products, quotients, roots, and absolute values and preserve non-strict inequalities, so a limit can be assembled from the limits of its parts without returning to epsilon and M. The squeeze lemma transfers a limit through two envelopes; the monotone convergence theorem produces a limit from boundedness alone; and the ratio test settles the geometric and factorial standard limits.\n",{"path":23145,"title":23146,"module":23138,"summary":23147},"\u002Freal-analysis\u002Fsequences-series\u002Flimsup-bolzano-weierstrass","Subsequences, Limit Superior, and Bolzano–Weierstrass","A bounded sequence need not converge, but it always has convergent subsequences, and its terms cluster between two extreme values. The limit superior and inferior are the limits of the tail suprema and infima; they always exist for a bounded sequence, coincide exactly when it converges, and are its largest and smallest subsequential limits. Bolzano–Weierstrass extracts a convergent subsequence from boundedness alone.\n",{"path":23149,"title":23150,"module":23138,"summary":23151},"\u002Freal-analysis\u002Fsequences-series\u002Fcauchy-completeness","Cauchy Sequences and the Completeness of the Reals","The Cauchy criterion tests convergence without knowing the limit: a sequence converges exactly when its terms eventually all lie within any tolerance of one another. Cauchy sequences are bounded, in the reals Cauchy and convergent are equivalent, and this completeness property is interchangeable with the least-upper-bound axiom — the single feature that separates the real line from the rationals.\n",{"path":23153,"title":23154,"module":23138,"summary":23155},"\u002Freal-analysis\u002Fsequences-series\u002Fseries-convergence","Series and Convergence Tests","A series converges when its sequence of partial sums does, so every fact about sequences transfers. Geometric and telescoping series sum in closed form; the n-th term test rejects series whose terms miss zero, though the harmonic series shows the converse fails; and the comparison test against the geometric and p-series benchmarks settles most nonnegative-term series.\n",{"path":23157,"title":23158,"module":23138,"summary":23159},"\u002Freal-analysis\u002Fsequences-series\u002Fabsolute-conditional-rearrangement","Absolute Convergence, the Ratio and Root Tests, and Rearrangements","Absolute convergence is the strong form of convergence that permits free manipulation; conditional convergence is fragile. Absolute convergence implies convergence, and the ratio and root tests detect it by comparison with the geometric series. The alternating series test supplies conditionally convergent series, Riemann's theorem rearranges any of them to any sum, and Mertens' theorem multiplies series when at least one converges absolutely.\n",{"path":23161,"title":23162,"module":23163,"summary":23164},"\u002Freal-analysis\u002Fmetric-spaces\u002Fmetric-spaces-norms","Metric Spaces, Norms, and Examples","Metric Spaces and Topology","A metric is a function $d(x,y)$ obeying four axioms: nonnegativity, identity of indiscernibles, symmetry, and the triangle inequality. The Euclidean, taxicab, sup, discrete, and great-circle metrics all qualify, as does the sup metric on $C[a,b]$. Every norm induces a metric, and strongly equivalent metrics share the same open sets.\n",{"path":23166,"title":23167,"module":23163,"summary":23168},"\u002Freal-analysis\u002Fmetric-spaces\u002Fopen-closed-sets","Open and Closed Sets, Interior, Closure","Open sets are those in which every point has room to move; closed sets are their complements. From the single ball construction come the topology axioms (arbitrary unions, finite intersections), the interior, closure, and boundary of a set, and the fact that openness is always relative to the ambient space.\n",{"path":23170,"title":23171,"module":23163,"summary":23172},"\u002Freal-analysis\u002Fmetric-spaces\u002Fconvergence-completeness","Convergence, Cauchy Sequences, and Completeness","The $\\varepsilon$-$N$ definition of a limit transfers verbatim to any metric space once $|x-y|$ is replaced by $d(x,y)$. Convergent sequences characterize closed sets and closures; Cauchy sequences and completeness capture spaces with no missing limits, with $\\mathbb{R}^n$ and $C[a,b]$ complete and $\\mathbb{Q}$ and $(0,1]$ not.\n",{"path":23174,"title":23175,"module":23163,"summary":23176},"\u002Freal-analysis\u002Fmetric-spaces\u002Fcompactness","Compactness","A set is compact if every open cover has a finite subcover. In a metric space this is equivalent to sequential compactness and to being complete and totally bounded. Compact sets are closed and bounded; the Heine–Borel theorem gives the converse in $\\mathbb{R}^n$ but nowhere else in general.\n",{"path":23178,"title":23179,"module":23163,"summary":23180},"\u002Freal-analysis\u002Fmetric-spaces\u002Fconnectedness","Connectedness","A space is connected when it cannot be split into two nonempty open pieces. The connected subsets of $\\mathbb{R}$ are precisely the intervals, path- connectedness gives a constructive sufficient condition, and connectedness is a topological invariant preserved by continuous maps, the fact behind the intermediate value theorem.\n",{"path":23182,"title":23183,"module":21582,"summary":23184},"\u002Freal-analysis\u002Fcontinuity\u002Flimits-of-functions","Limits of Functions","The limit of a function at a point is an epsilon–delta condition pinning one value L as the target of f(x) as x approaches c, mirroring the sequence definition with distance replacing index. It is stated only at cluster points of the domain, is unique when it exists, and reduces to sequential limits through the Heine criterion. The algebra of limits and one-sided limits follow from that reduction.\n",{"path":23186,"title":23187,"module":21582,"summary":23188},"\u002Freal-analysis\u002Fcontinuity\u002Fcontinuous-functions","Continuous Functions","A function is continuous at c when its limit there equals its own value, lim f(x) = f(c). The epsilon–delta and sequential forms agree; sums, products, quotients, and compositions of continuous functions are continuous; and the failures split into jump, Dirichlet, popcorn, and removable types. The topological reading is that preimages of open sets are open.\n",{"path":23190,"title":23191,"module":21582,"summary":23192},"\u002Freal-analysis\u002Fcontinuity\u002Fevt-ivt","Extreme and Intermediate Value Theorems","On a closed bounded interval a continuous function attains an absolute maximum and minimum (the extreme value theorem, compactness preserved by continuity) and takes every value between its endpoint values (the intermediate value theorem, connectedness preserved). Both proofs run through Bolzano–Weierstrass and bisection, and yield root-finding, existence of k-th roots, and fixed-point theorems.\n",{"path":23194,"title":23195,"module":21582,"summary":23196},"\u002Freal-analysis\u002Fcontinuity\u002Funiform-continuity","Uniform Continuity","Uniform continuity strengthens continuity by demanding one delta that works at every point of the domain, not a delta re-chosen at each point. It separates x^2 on a compact interval from x^2 on the whole line and from 1\u002Fx near zero; continuity on a closed bounded interval is automatically uniform; uniformly continuous functions preserve Cauchy sequences and extend to endpoints; and Lipschitz continuity is the strongest of the three, through its secant-slope bound.\n",{"path":23198,"title":23199,"module":21582,"summary":23200},"\u002Freal-analysis\u002Fcontinuity\u002Fcontinuity-metric-spaces","Continuity on Metric Spaces","The epsilon–delta definition used only distances, so continuity transfers to maps between metric spaces by replacing absolute values with the two metrics. In this generality continuity still admits a sequential form, preserves compactness and connectedness, is uniform on a compact domain, and reads topologically as preimages of open sets being open, the formulation that defines homeomorphisms.\n",{"path":23202,"title":23203,"module":21582,"summary":23204},"\u002Freal-analysis\u002Fcontinuity\u002Flimits-infinity-monotone","Limits at Infinity and Monotone Functions","Treating infinity as a cluster point extends the epsilon–delta limit to x approaching plus or minus infinity, giving horizontal asymptotes and infinite limits. For monotone functions the one-sided limits always exist as suprema and infima, the discontinuities are jumps and at most countably many, the continuity is equivalent to the image being an interval, and a strictly monotone function always has a continuous inverse.\n",{"path":23206,"title":23207,"module":23208,"summary":23209},"\u002Freal-analysis\u002Fdifferentiation\u002Fthe-derivative","The Derivative","Differentiation","The derivative is the limit of the difference quotient, the slope the secant lines approach as the second point slides into the first. Differentiability forces continuity; linearity and the product, quotient, and chain rules follow from the definition; and a continuous function can fail to be differentiable, as the absolute value does at the origin.\n",{"path":23211,"title":23212,"module":23208,"summary":23213},"\u002Freal-analysis\u002Fdifferentiation\u002Fmean-value-theorem","The Mean Value Theorem","A relative extremum in the interior forces the derivative to vanish; Rolle's theorem and the mean value theorem turn that local fact into global control. The sign of the derivative fixes monotonicity, a bounded derivative yields a Lipschitz bound, and Darboux's theorem shows derivatives have the intermediate value property even where they are discontinuous.\n",{"path":23215,"title":23216,"module":23208,"summary":23217},"\u002Freal-analysis\u002Fdifferentiation\u002Ftaylors-theorem","Taylor's Theorem","Taylor's theorem generalizes the mean value theorem: an n-times differentiable function is matched near a point by a degree-n polynomial, with a Lagrange remainder that names the error exactly through one higher derivative. Iterating the mean value theorem proves it; the second-derivative test is the order-one case; and a smooth non-analytic bump separates a Taylor series from the function it fails to represent.\n",{"path":23219,"title":23220,"module":23208,"summary":23221},"\u002Freal-analysis\u002Fdifferentiation\u002Finverse-function-1d","The Inverse Function Theorem in One Variable","A nonzero derivative certifies a local inverse and fixes its slope. A strictly monotone differentiable function has a differentiable inverse whose derivative is the reciprocal of the original; the inverse function theorem removes the monotonicity hypothesis, and the reciprocal formula constructs nth roots and the logarithm's derivative, failing exactly where the derivative vanishes.\n",{"path":23223,"title":23224,"module":23225,"summary":23226},"\u002Freal-analysis\u002Friemann-integration\u002Fdarboux-integral","Partitions, Darboux Sums, and Integrability","The Riemann Integral","The Riemann integral is defined by trapping the area under a bounded function between under- and over-estimates. Partitions cut the domain into strips; lower and upper Darboux sums bracket the area; refining a partition tightens the bracket. A function is integrable exactly when the bracket can be made arbitrarily thin, and the tagged Riemann-sum limit gives the same number.\n",{"path":23228,"title":23229,"module":23225,"summary":23230},"\u002Freal-analysis\u002Friemann-integration\u002Fintegrability-classes","Which Functions Are Integrable","The Cauchy criterion certifies whole classes of functions as integrable. Continuous functions are integrable because uniform continuity makes every oscillation cap small; monotone functions are integrable because their caps telescope to a single total jump; bounded functions with finitely many discontinuities are integrable by isolating the bad points. The Dirichlet function fails, and the Lebesgue criterion names the exact boundary.\n",{"path":23232,"title":23233,"module":23225,"summary":23234},"\u002Freal-analysis\u002Friemann-integration\u002Fproperties-of-the-integral","Properties of the Integral","The integral is a linear, order-preserving, additive operator on the integrable functions. It splits across subintervals, respects inequalities, bounds the size of a function by the integral of its absolute value, and preserves products. The mean value theorem for integrals identifies the integral with an attained average height on a fixed rectangle.\n",{"path":23236,"title":21637,"module":23225,"summary":23237},"\u002Freal-analysis\u002Friemann-integration\u002Ffundamental-theorem","The fundamental theorem ties the integral to the derivative in two forms. The evaluation form computes a definite integral from any antiderivative; the differentiation form shows the area function has derivative equal to the integrand at points of continuity. Together they make differentiation and integration inverse operations, and yield integration by parts and change of variables.\n",{"path":23239,"title":23240,"module":23225,"summary":23241},"\u002Freal-analysis\u002Friemann-integration\u002Flog-exp-improper","The Logarithm, Exponential, and Improper Integrals","The integral defines transcendental functions. The logarithm is the area under 1\u002Ft, the exponential is its inverse, and their calculus properties follow from the fundamental theorem. Improper integrals extend integration to unbounded intervals and unbounded integrands as limits of proper integrals, with a p-test, a comparison test, absolute versus conditional convergence, and the integral test linking integrals to series.\n",{"path":23243,"title":23244,"module":23245,"summary":23246},"\u002Freal-analysis\u002Ffunction-sequences\u002Fpointwise-uniform-convergence","Pointwise and Uniform Convergence","Sequences and Series of Functions","A sequence of functions has two natural notions of limit. Pointwise convergence fixes each input and takes the limit of numbers; uniform convergence demands one rate that works for every input at once. The uniform norm turns the second into a statement about a single sequence of numbers, and the uniform Cauchy criterion and the Weierstrass M-test let us certify it.\n",{"path":23248,"title":23249,"module":23245,"summary":23250},"\u002Freal-analysis\u002Ffunction-sequences\u002Finterchange-of-limits","Interchange of Limits: Continuity, Integration, Differentiation","Passing to a limit inside a continuity statement, an integral, or a derivative is an interchange of two limits, and the two limits do not always commute. Uniform convergence licenses the first two swaps: the uniform limit of continuous functions is continuous, and the limit of the integrals is the integral of the limit. Differentiation needs uniform convergence of the derivatives, and counterexamples show why each hypothesis is required.\n",{"path":23252,"title":23253,"module":23245,"summary":23254},"\u002Freal-analysis\u002Ffunction-sequences\u002Fpower-series-weierstrass","Power Series and the Weierstrass Approximation Theorem","A power series converges uniformly on every closed subinterval inside its radius of convergence, together with all of its derivatives. That makes it continuous, differentiable, and integrable term by term, so a power series defines an infinitely differentiable function. The Weierstrass approximation theorem then shows that polynomials come uniformly close to any continuous function on a closed bounded interval.\n",{"path":23256,"title":23257,"module":23245,"summary":23258},"\u002Freal-analysis\u002Ffunction-sequences\u002Fpicard-ode","Picard's Existence and Uniqueness Theorem","The Banach fixed-point theorem says a contraction of a complete metric space has exactly one fixed point, found by iterating from any start. Applied to the space of continuous functions with the uniform norm, it proves Picard's theorem: a first-order differential equation with a Lipschitz right-hand side has a unique local solution. Picard iteration constructs that solution explicitly, and worked examples show the Lipschitz condition is not optional.\n",{"path":23260,"title":23261,"module":23262,"summary":23263},"\u002Freal-analysis\u002Fseveral-variables\u002Fdifferentiability-rn","The Derivative of a Map ℝⁿ → ℝᵐ","Functions of Several Variables (Introduction)","The derivative of a map between Euclidean spaces is the linear transformation of vanishing relative error, unique when it exists and represented in coordinates by the Jacobian matrix of partial derivatives. Differentiability forces continuity through a local Lipschitz bound. Existence of the partial derivatives alone does not suffice; continuity of the partials does.\n",{"path":23265,"title":23266,"module":23262,"summary":23267},"\u002Freal-analysis\u002Fseveral-variables\u002Fgradient-chain-rule","Directional Derivatives, the Gradient, and the Chain Rule","The directional derivative measures the rate of change of a scalar field along a chosen heading and equals the derivative applied to that direction. The gradient collects these into a vector that points along steepest ascent and sits orthogonal to level sets. The chain rule composes derivatives by multiplying Jacobians, and a mean value theorem holds for scalar fields but fails for vector-valued maps.\n",{"path":23269,"title":23270,"module":23262,"summary":23271},"\u002Freal-analysis\u002Fseveral-variables\u002Fhigher-derivatives-taylor-extrema","Higher Derivatives, Taylor's Theorem, and Extrema","Iterating the derivative gives a symmetric second derivative, the Hessian, whose mixed partials agree when they are continuous. Taylor's theorem expands a smooth map to any order with a Lagrange-type remainder, and at a critical point the definiteness of the Hessian decides between a local minimum, a local maximum, and a saddle.\n",{"path":23273,"title":23274,"module":23262,"summary":23275},"\u002Freal-analysis\u002Fseveral-variables\u002Finverse-implicit-theorems","The Inverse and Implicit Function Theorems","A nonlinear map with a nonsingular Jacobian is locally invertible, with the inverse's derivative given by the inverse matrix. The contraction mapping principle supplies the local inverse; the implicit function theorem then solves a system for some variables in terms of the rest whenever the relevant Jacobian block is invertible. Worked coordinate changes show both theorems in use.\n",{"path":23277,"title":23278,"module":23262,"summary":23279},"\u002Freal-analysis\u002Fseveral-variables\u002Fmultiple-integrals","Multiple Integrals","The Riemann integral of a bounded function over a closed rectangle in Euclidean space is built from Darboux upper and lower sums on a grid of subrectangles, with the same squeeze criterion that governs the one-variable integral. Continuous integrands are integrable, and a set of content zero can be ignored. Fubini's theorem evaluates a multiple integral as an iterated one in either order, and the indicator trick extends the theory to regions bounded by curves.\n",{"path":23281,"title":23282,"module":6,"summary":6},"\u002Freal-analysis","Real Analysis",{"path":23284,"title":23285,"module":21235,"summary":23286},"\u002Fabstract-algebra\u002Ffoundations\u002Fsets-functions-relations","Sets, Functions, and Equivalence Relations","Algebra is built on three prior notions: the set, the map between sets, and the equivalence relation that reorganizes a set into disjoint classes. Sets, maps (injective, surjective, bijective), fibers and preimages, and the correspondence between equivalence relations and partitions — the one structural fact reused in every later quotient construction.\n",{"path":23288,"title":23289,"module":21235,"summary":23290},"\u002Fabstract-algebra\u002Ffoundations\u002Fintegers-and-modular-arithmetic","The Integers and Modular Arithmetic","The integers carry the template every ring later imitates: well-ordering drives induction, induction drives the division algorithm, and division drives the Euclidean algorithm, gcd, Bézout's identity, and unique factorization into primes. Quotienting by congruence mod n builds the first finite arithmetic, Z\u002FnZ, whose invertible elements form the group of units.\n",{"path":23292,"title":23293,"module":23294,"summary":23295},"\u002Fabstract-algebra\u002Fgroups-and-symmetry\u002Fgroup-axioms-and-first-examples","Group Axioms and First Examples","Groups and Symmetry","A group is a set with one associative operation that has an identity and inverses. We state the axioms, prove that the identity, inverses, and cancellation behave as expected, define the order of a group and of an element, and catalogue the running examples: the integers, the additive group of residues mod n, and the multiplicative group of units mod n.\n",{"path":23297,"title":23298,"module":23294,"summary":23299},"\u002Fabstract-algebra\u002Fgroups-and-symmetry\u002Fdihedral-and-symmetric-groups","Dihedral and Symmetric Groups","The dihedral group D_{2n} is the symmetries of a regular n-gon, generated by a rotation r and a reflection s subject to three relations. The symmetric group S_n is all permutations of n objects, written in cycle notation. Orders, generators and relations, cycle decomposition, the order of a permutation from its cycle type, and the parity that splits S_n in half.\n",{"path":23301,"title":23302,"module":23294,"summary":23303},"\u002Fabstract-algebra\u002Fgroups-and-symmetry\u002Fmatrix-and-quaternion-groups","Matrix and Quaternion Groups","Invertible matrices over a field form the general linear group GL_n(F), with the determinant-one matrices as the subgroup SL_n(F). Over a finite field the order of GL_n(F) has a clean product formula. The quaternion group Q_8 is a second small nonabelian group, distinct from the dihedral group of the same order; its multiplication and subgroup structure sharpen the contrast between the two.\n",{"path":23305,"title":23306,"module":23294,"summary":23307},"\u002Fabstract-algebra\u002Fgroups-and-symmetry\u002Fhomomorphisms-and-group-actions","Homomorphisms, Isomorphisms, and Actions","A homomorphism is a map between groups that respects the operation; an isomorphism is a bijective one, making two groups the same up to relabeling. The kernel and image measure how far a homomorphism is from injective and surjective. A group action realizes a group as permutations of a set, and actions correspond exactly to homomorphisms into a symmetric group, with orbits and stabilizers as the first tools for counting.\n",{"path":23309,"title":23310,"module":23311,"summary":23312},"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fsubgroups-and-substructures","Subgroups and Their Substructures","Subgroups and Quotients","A subgroup is a subset that is a group under the inherited operation. One test decides it: nonempty and closed under the map $(x,y) \\mapsto xy^{-1}$. From an arbitrary subset $A$ we build the centralizer, normalizer, and center, and from an action the stabilizer and kernel, all of them subgroups nested in a fixed chain inside $G$.\n",{"path":23314,"title":23315,"module":23311,"summary":23316},"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fcyclic-groups","Cyclic Groups","A cyclic group is generated by one element. Two facts organize the whole theory: the order of an element equals the order of the subgroup it generates, and cyclic groups of equal order are isomorphic, so $\\mathbb{Z}$ and $\\mathbb{Z}\u002Fn\\mathbb{Z}$ are the only ones. From there the generators ($\\varphi(n)$ of them), the subgroups (one per divisor of $n$), and a fast exponentiation algorithm all follow.\n",{"path":23318,"title":23319,"module":23311,"summary":23320},"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fgeneration-and-subgroup-lattices","Generation and the Lattice of Subgroups","The subgroup generated by a subset $A$ is the smallest subgroup containing it, described top-down as an intersection and bottom-up as the set of words in $A$ and its inverses. Collecting all subgroups and ordering them by containment produces the subgroup lattice, whose Hasse diagram shows the joins, meets, and containment relations among all subgroups.\n",{"path":23322,"title":23323,"module":23311,"summary":23324},"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fcosets-lagrange-and-normal-subgroups","Cosets, Lagrange, and Normal Subgroups","The left cosets of a subgroup partition a group into equal-sized blocks, so the order of a subgroup divides the order of the group: Lagrange's theorem. When the blocks can be multiplied consistently — exactly when the subgroup is normal — they form the quotient group $G\u002FN$. Fermat's and Euler's theorems fall out as index computations.\n",{"path":23326,"title":23327,"module":23311,"summary":23328},"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fisomorphism-theorems","The Isomorphism Theorems","Four theorems relate homomorphisms, quotients, and subgroup lattices. The first identifies the image of a homomorphism with the quotient by its kernel; the second and third compute quotients built from two subgroups and quotients of quotients; the fourth matches the subgroups of $G\u002FN$ with the subgroups of $G$ lying above $N$. Together they make quotient groups computable.\n",{"path":23330,"title":23331,"module":23311,"summary":23332},"\u002Fabstract-algebra\u002Fsubgroups-and-quotients\u002Fcomposition-series-and-the-alternating-group","Composition Series and the Alternating Group","A composition series breaks a finite group into simple quotient factors, and Jordan-Hölder says those factors are unique up to order. This turns classification into two problems: list the simple groups, and describe how to reassemble them. The sign homomorphism splits $S_n$ into even and odd permutations, defining the alternating group $A_n$, simple for $n \\ge 5$.\n",{"path":23334,"title":23335,"module":23336,"summary":23337},"\u002Fabstract-algebra\u002Fgroup-actions-and-sylow\u002Factions-and-cayleys-theorem","Actions, Orbits, and Cayley's Theorem","Group Actions and Sylow Theory","A group action turns abstract elements into permutations of a set. The action splits the set into orbits, and the orbit-stabilizer theorem ties each orbit's size to the index of a stabilizer. Applied to a group acting on itself by left multiplication, this gives Cayley's theorem: every group is a group of permutations.\n",{"path":23339,"title":23340,"module":23336,"summary":23341},"\u002Fabstract-algebra\u002Fgroup-actions-and-sylow\u002Fconjugation-and-the-class-equation","Conjugation and the Class Equation","A group acts on itself by conjugation, and the orbits are the conjugacy classes. Orbit-stabilizer turns the resulting partition into the class equation, which forces every group of prime-power order to have a nontrivial center. Conjugacy in the symmetric group is cycle type, and Burnside's lemma counts orbits by averaging fixed points.\n",{"path":23343,"title":23344,"module":23336,"summary":23345},"\u002Fabstract-algebra\u002Fgroup-actions-and-sylow\u002Fsylow-theorems","The Sylow Theorems","Lagrange's theorem forbids subgroups whose order fails to divide the group order; Sylow's theorems supply a partial converse for prime powers. A Sylow p-subgroup always exists, all of them are conjugate, and their count satisfies two congruence-and-divisibility constraints tight enough to prove many groups non-simple from their order alone.\n",{"path":23347,"title":23348,"module":23336,"summary":23349},"\u002Fabstract-algebra\u002Fgroup-actions-and-sylow\u002Fautomorphisms-and-simple-groups","Automorphisms and Simplicity of Aₙ","Conjugation makes a group act on itself and on its normal subgroups by automorphisms, giving the inner automorphism group G\u002FZ(G) and the embedding of N(H)\u002FC(H) into Aut(H). Characteristic subgroups are those every automorphism fixes, and the automorphism group of a cyclic group is its unit group. The lesson closes by proving the alternating group Aₙ is simple for n ≥ 5.\n",{"path":23351,"title":23352,"module":23353,"summary":23354},"\u002Fabstract-algebra\u002Fproducts-and-group-structure\u002Fdirect-products-and-finite-abelian-groups","Direct Products and Finite Abelian Groups","Products and Group Structure","The direct product assembles a larger group from componentwise copies of smaller ones, and a recognition theorem reverses the process when two normal subgroups meet trivially and span the group. The Fundamental Theorem of Finitely Generated Abelian Groups then classifies every such group by two equivalent invariants, invariant factors and elementary divisors.\n",{"path":23356,"title":23357,"module":23353,"summary":23358},"\u002Fabstract-algebra\u002Fproducts-and-group-structure\u002Fsemidirect-products","Semidirect Products","The semidirect product relaxes the direct product by requiring only one factor to be normal, with the other acting on it through a homomorphism into its automorphism group. This single twisting map lets abelian pieces assemble into non-abelian groups, realizes the dihedral groups as $\\mathbb{Z}_n \\rtimes \\mathbb{Z}_2$, and, with a recognition theorem, classifies groups of several small orders.\n",{"path":23360,"title":23361,"module":23353,"summary":23362},"\u002Fabstract-algebra\u002Fproducts-and-group-structure\u002Fnilpotent-and-solvable-groups","p-Groups, Nilpotent, and Solvable Groups","Finite p-groups have nontrivial center, and iterating the center upward builds the nilpotent groups, which decompose as the direct product of their Sylow subgroups. Iterating the commutator downward builds the solvable groups, whose factors are abelian. The chain cyclic, abelian, nilpotent, solvable orders these classes, and A_5 breaks the last link.\n",{"path":23364,"title":23365,"module":23353,"summary":23366},"\u002Fabstract-algebra\u002Fproducts-and-group-structure\u002Fclassifying-small-groups","Classifying Groups of Small Order","With Sylow's theorem to force normal subgroups, direct and semidirect products to assemble them, and presentations to name the result, every group up to order fifteen can be listed explicitly. Free groups make presentations precise: generators with no relations, from which any group is a quotient by the normal closure of its relations.\n",{"path":23368,"title":23369,"module":23370,"summary":23371},"\u002Fabstract-algebra\u002Fring-theory\u002Frings-definitions-and-examples","Rings: Definitions and Examples","Ring Theory","A ring carries two operations: an abelian group under addition and an associative multiplication linked by the distributive laws. The named special cases — commutative rings, integral domains, division rings, and fields — differ only in how their multiplication behaves. Standard examples include quadratic integer rings, polynomial rings, matrix rings, and group rings.\n",{"path":23373,"title":23374,"module":23370,"summary":23375},"\u002Fabstract-algebra\u002Fring-theory\u002Fideals-quotients-and-homomorphisms","Ideals, Quotient Rings, and Homomorphisms","Ring homomorphisms have kernels that absorb multiplication; such subsets are ideals, and every ideal is the kernel of the projection onto a quotient ring. The quotient construction yields the ring isomorphism theorems and classifies ideals by their quotients: R\u002FI is a field exactly when I is maximal, an integral domain exactly when I is prime.\n",{"path":23377,"title":23378,"module":23370,"summary":23379},"\u002Fabstract-algebra\u002Fring-theory\u002Ffractions-and-the-chinese-remainder-theorem","Fields of Fractions and the CRT","Rings of fractions invert a multiplicatively closed set, enlarging an integral domain into its field of fractions the way Z becomes Q. The Chinese Remainder Theorem splits a quotient by comaximal ideals into a direct product, generalizing Z\u002FmnZ ≅ Z\u002FmZ × Z\u002FnZ and explaining why the Euler function is multiplicative.\n",{"path":23381,"title":23382,"module":23383,"summary":23384},"\u002Fabstract-algebra\u002Ffactorization-and-polynomials\u002Feuclidean-domains-pids-ufds","Euclidean Domains, PIDs, and UFDs","Factorization and Polynomial Rings","Three classes of integral domain, ordered by how much of elementary arithmetic survives: Euclidean domains carry a division algorithm, principal ideal domains make every ideal a single multiple, and unique factorization domains factor every element into irreducibles in one way. We prove the chain ED implies PID implies UFD, the classes are separated by explicit counterexamples, and irreducible and prime coincide exactly in a UFD.\n",{"path":23386,"title":23387,"module":23383,"summary":23388},"\u002Fabstract-algebra\u002Ffactorization-and-polynomials\u002Fpolynomial-rings-over-fields","Polynomial Rings over Fields","When the coefficients form a field, polynomial long division works exactly as it does over the rationals, and it works with a unique quotient and remainder. That single fact makes F[x] a Euclidean domain, hence a PID and a UFD: every ideal is the multiples of one polynomial, roots correspond to linear factors, and F[x]\u002F(f) is a field precisely when f is irreducible.\n",{"path":23390,"title":23391,"module":23383,"summary":23392},"\u002Fabstract-algebra\u002Ffactorization-and-polynomials\u002Fgauss-lemma-and-unique-factorization","Gauss's Lemma and Unique Factorization","A UFD is not a field, so its polynomial ring is not a PID — yet unique factorization survives the passage from R to R[x]. Gauss's lemma supplies the passage: a polynomial that factors over the fraction field already factors over R, once content is factored out. This gives the theorem that R[x] is a UFD whenever R is, so Z[x] and Q[x,y] factor uniquely even though neither is a PID.\n",{"path":23394,"title":23395,"module":23383,"summary":23396},"\u002Fabstract-algebra\u002Ffactorization-and-polynomials\u002Firreducibility-criteria-and-groebner","Irreducibility Criteria and Gröbner Bases","Deciding whether a given polynomial is irreducible, and computing in multivariate polynomial rings. In one variable: the rational root test, reduction modulo a prime, and Eisenstein's criterion. In several variables, where division fails, a monomial order gives leading terms, a Gröbner basis restores a well-defined remainder, and Buchberger's algorithm computes it.\n",{"path":23398,"title":23399,"module":23400,"summary":23401},"\u002Fabstract-algebra\u002Fmodule-theory\u002Fintroduction-to-modules","Introduction to Modules","Module Theory","A module is an abelian group on which a ring acts, generalizing both vector spaces (when the ring is a field) and abelian groups (when the ring is the integers). Submodules, homomorphisms, quotients, and the isomorphism theorems carry over from groups, and an F[x]-module is the same datum as a vector space with a chosen linear operator — the correspondence behind the canonical forms.\n",{"path":23403,"title":23404,"module":23400,"summary":23405},"\u002Fabstract-algebra\u002Fmodule-theory\u002Ffree-modules-and-direct-sums","Generation, Direct Sums, and Free Modules","A generating set spans a module by R-linear combinations; a direct sum decomposes it into independent pieces; a free module has a basis and the universal property that a homomorphism is determined by arbitrary values on that basis. Rank is well defined over a commutative ring, torsion blocks a basis, and every module is a quotient of a free one — a presentation by generators and relations.\n",{"path":23407,"title":23408,"module":23400,"summary":23409},"\u002Fabstract-algebra\u002Fmodule-theory\u002Ftensor-products-and-exact-sequences","Tensor Products and Exact Sequences","The tensor product builds a module in which elements of two modules can be multiplied, characterized by a universal property turning bilinear maps into linear ones; extension of scalars is its guiding case. Exact sequences track how a module is assembled from a submodule and a quotient, when that assembly splits, and which modules — projective, injective, flat — make the Hom and tensor functors preserve exactness.\n",{"path":23411,"title":23412,"module":23400,"summary":23413},"\u002Fabstract-algebra\u002Fmodule-theory\u002Fvector-spaces-and-linear-maps","Vector Spaces and Linear Maps","A vector space is a module over a field, and the field hypothesis removes every pathology a general module can have: every vector space is free, so it has a basis, a well-defined dimension, and a coordinate isomorphism with F^n. Linear maps become matrices, change of basis becomes similarity, every space pairs with a dual of the same dimension, and the determinant is the unique alternating multilinear normalized form.\n",{"path":23415,"title":23416,"module":23417,"summary":23418},"\u002Fabstract-algebra\u002Fmodules-over-pids\u002Fstructure-theorem-over-pids","The Structure Theorem for Modules over a PID","Modules over PIDs and Canonical Forms","Every finitely generated module over a principal ideal domain splits as a free part plus a direct sum of cyclic torsion pieces, in two canonical ways: invariant factors, tied together by a divisibility chain, and elementary divisors, one prime power at a time. Existence follows from the stacked-basis theorem, both lists are unique, and the case $R = \\mathbb{Z}$ is the classification of finitely generated abelian groups.\n",{"path":23420,"title":23421,"module":23417,"summary":23422},"\u002Fabstract-algebra\u002Fmodules-over-pids\u002Frational-canonical-form","Rational Canonical Form","A linear operator turns its vector space into a module over the polynomial ring $F[x]$, with $x$ acting as the operator. The structure theorem's invariant factors then become polynomials, each cyclic summand becomes a companion matrix, and the block-diagonal assembly is the rational canonical form. It is unique, it is computed inside the base field, and two matrices are similar exactly when their rational canonical forms agree.\n",{"path":23424,"title":23425,"module":23417,"summary":23426},"\u002Fabstract-algebra\u002Fmodules-over-pids\u002Fjordan-canonical-form","Jordan Canonical Form","When the base field contains all the eigenvalues, the elementary divisors of an operator are powers of linear polynomials, and each cyclic summand becomes a Jordan block: an eigenvalue on the diagonal with ones just above it. Stacking the blocks gives the Jordan canonical form, unique up to reordering, as close to diagonal as the operator allows. Diagonalizability reads off the minimal polynomial, and the block sizes are counted by ranks of powers of the operator minus the eigenvalue.\n",{"path":23428,"title":23429,"module":23430,"summary":23431},"\u002Fabstract-algebra\u002Ffield-theory\u002Ffield-extensions-and-algebraic-elements","Field Extensions and Algebraic Elements","Field Theory","A field extension makes a larger field K into a vector space over a smaller field F, and its degree [K:F] is that dimension. Adjoining a root of an irreducible polynomial builds a simple extension F(α) isomorphic to F[x]\u002F(m), whose degree is the degree of the minimal polynomial. The tower law makes these degrees multiply, which turns algebra over fields into bookkeeping with integers.\n",{"path":23433,"title":23434,"module":23430,"summary":23435},"\u002Fabstract-algebra\u002Ffield-theory\u002Fstraightedge-and-compass-constructions","Straightedge-and-Compass Constructions","The lengths a straightedge and compass can build from a unit form a field closed under square roots, and every constructible number lies in a tower of quadratic extensions. So its degree over the rationals is a power of two. That single obstruction settles three problems the Greeks left open: doubling the cube, trisecting a general angle, and squaring the circle are all impossible.\n",{"path":23437,"title":23438,"module":23430,"summary":23439},"\u002Fabstract-algebra\u002Ffield-theory\u002Fsplitting-fields-and-algebraic-closure","Splitting Fields and Algebraic Closure","The splitting field of a polynomial is the smallest extension in which it factors into linear pieces, obtained by adjoining all its roots. Every polynomial has one, its degree is at most n factorial, and any two splitting fields are isomorphic. Pushing this to all polynomials at once gives the algebraic closure, a field in which every polynomial splits and which is unique up to isomorphism.\n",{"path":23441,"title":23442,"module":23430,"summary":23443},"\u002Fabstract-algebra\u002Ffield-theory\u002Fseparable-and-cyclotomic-extensions","Separable Extensions and Cyclotomic Fields","A polynomial is separable when its roots are distinct, detected by whether it shares a factor with its formal derivative. Over perfect fields — characteristic zero and finite fields — every irreducible is separable, and the existence and uniqueness of the finite fields follow. Cyclotomic polynomials package the roots of unity by order, are irreducible over the rationals, and give the cyclotomic field its degree phi(n).\n",{"path":23445,"title":23446,"module":23447,"summary":23448},"\u002Fabstract-algebra\u002Fgalois-theory\u002Fthe-galois-correspondence","The Galois Correspondence","Galois Theory","Galois theory attaches to a field extension its group of symmetries and shows that, for the right extensions, the subgroups of that group are in exact order-reversing correspondence with the intermediate fields. The automorphism group, Artin's theorem, the characterization of Galois extensions, and the Fundamental Theorem together turn questions about fields into questions about finite groups.\n",{"path":23450,"title":23451,"module":23447,"summary":23452},"\u002Fabstract-algebra\u002Fgalois-theory\u002Ffinite-fields","Finite Fields","Every finite field has prime-power order, is the splitting field of $x^{p^n} - x$, and is unique up to isomorphism. Its extension over the prime field is Galois with cyclic group generated by the Frobenius map $x \\mapsto x^p$, so the Galois correspondence reduces the subfield lattice to the divisor lattice of $n$. Möbius inversion counts the irreducible polynomials of each degree, and cyclic error-correcting codes are one application.\n",{"path":23454,"title":23455,"module":23447,"summary":23456},"\u002Fabstract-algebra\u002Fgalois-theory\u002Fcyclotomic-and-abelian-extensions","Cyclotomic and Abelian Extensions","The Galois group of the $n$th cyclotomic field over $\\mathbb{Q}$ is the unit group $(\\mathbb{Z}\u002Fn\\mathbb{Z})^\\times$, which makes cyclotomic fields the worked catalogue of abelian extensions of $\\mathbb{Q}$. The isomorphism identifies subfields with subgroups, realizes every finite abelian group as a Galois group over $\\mathbb{Q}$, and leads to Kronecker–Weber. Composites of Galois extensions and the primitive element theorem supply the machinery.\n",{"path":23458,"title":23459,"module":23447,"summary":23460},"\u002Fabstract-algebra\u002Fgalois-theory\u002Fgalois-groups-of-polynomials","Galois Groups of Polynomials","Ordering the roots of a separable polynomial embeds its Galois group in the symmetric group $S_n$, and the group is transitive exactly when the polynomial is irreducible. The discriminant decides membership in $A_n$; for cubics and quartics the resolvent cubic pins the group down; and reduction modulo a prime produces elements of prescribed cycle type, the standard tool for computing Galois groups over $\\mathbb{Q}$.\n",{"path":23462,"title":23463,"module":23447,"summary":23464},"\u002Fabstract-algebra\u002Fgalois-theory\u002Fsolvability-by-radicals-and-the-quintic","Solvability by Radicals and the Quintic","A polynomial is solvable by radicals exactly when its Galois group is solvable. Cyclic extensions are radical extensions once roots of unity are present, which turns a radical tower into a solvable subnormal series. Since $S_n$ is solvable only for $n \\le 4$, the general quintic has no radical formula, and an explicit quintic with Galois group $S_5$ has roots provably not expressible in radicals.\n",{"path":23466,"title":23467,"module":23468,"summary":23469},"\u002Fabstract-algebra\u002Fcapstone\u002Fcommutative-algebra-and-algebraic-geometry","A Glimpse of Commutative Algebra and Algebraic Geometry","Capstone: Where Algebra Goes Next","Commutative algebra reads geometry off the ring of polynomial functions. The dictionary runs through Noetherian rings and the ascending chain condition, Hilbert's Basis Theorem, affine algebraic sets and the two maps connecting ideals to zero sets, radicals, the Zariski topology, and Hilbert's Nullstellensatz, which over an algebraically closed field makes radical ideals and algebraic sets the same object.\n",{"path":23471,"title":23472,"module":23468,"summary":23473},"\u002Fabstract-algebra\u002Fcapstone\u002Frepresentation-and-character-theory","A Glimpse of Representation and Character Theory","Representation theory studies a group by the ways it can act linearly on a vector space. Representations are equivalent to modules over the group ring; Maschke's theorem gives complete reducibility, the Wedderburn consequences bound the irreducible degrees, and character theory reduces a representation to a trace invariant governed by the orthogonality relations and displayed in the character table of a small group.\n",{"path":23475,"title":23476,"module":6,"summary":6},"\u002Fabstract-algebra","Abstract Algebra",{"path":23478,"title":23479,"module":23480,"summary":23481},"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fatomic-spectra-rutherford","Atomic Spectra and Rutherford's Nucleus","Early Atomic Models and the Old Quantum Theory","Atoms emit light only at sharp, reproducible wavelengths, and by 1890 those wavelengths were captured by the Rydberg-Ritz formula. Neither empirical regularity had a mechanical explanation. Rutherford's alpha-scattering experiment supplied the missing structure: the atom's positive charge and nearly all its mass sit in a tiny central nucleus, with the electrons far outside.\n",{"path":23483,"title":23484,"module":23480,"summary":23485},"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fbohr-model-hydrogen","The Bohr Model of Hydrogen","Bohr grafted three quantum postulates onto Rutherford's nuclear atom: certain orbits do not radiate, radiation accompanies a jump between them, and quantization must match classical physics for large orbits. Quantizing the angular momentum fixes the orbit radii and energies, reproduces the Rydberg-Ritz formula, and predicts the Rydberg constant from fundamental constants alone.\n",{"path":23487,"title":23488,"module":23480,"summary":23489},"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fx-ray-spectra-franck-hertz","X-Ray Spectra and the Franck-Hertz Experiment","Two 1913-14 experiments confirmed the Bohr-Rutherford atom independently of optical spectra. Moseley found that the square root of a characteristic X-ray frequency is linear in atomic number, fixing Z as nuclear charge and ordering the periodic table. Franck and Hertz measured discrete atomic energy levels directly by scattering electrons through a mercury vapor.\n",{"path":23491,"title":23492,"module":23480,"summary":23493},"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fbohr-sommerfeld-old-quantum-theory","The Bohr-Sommerfeld Old Quantum Theory","Bohr fixed the hydrogen levels with a single quantum number by quantizing angular momentum. Sommerfeld replaced that ad hoc rule with a general prescription: quantize the action of each separable coordinate. The rule produces elliptical orbits, a second (azimuthal) quantum number, space quantization, and — once the relativistic mass variation is included — a fine-structure splitting that matches experiment to order alpha squared.\n",{"path":23495,"title":23496,"module":23480,"summary":23497},"\u002Fatomic-physics\u002Fearly-models-and-old-quantum-theory\u002Fold-quantum-theory-limits-wkb","Limits of the Old Quantum Theory and the WKB Bridge","The old quantum theory works only where the classical motion is separable into independent periodic coordinates. It fails for helium, forbids the correct zero angular momentum of the hydrogen ground state, and misses the half-integer in the oscillator and in molecular spectra. The WKB quantization condition, derived from the Schrodinger equation, is the modern descendant of the Sommerfeld rule and repairs the half-integer through the Maslov correction.\n",{"path":23499,"title":23500,"module":23501,"summary":23502},"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fschrodinger-3d-hydrogen","The Schrödinger Equation in Three Dimensions and Hydrogen","The Quantum Hydrogen Atom","Extending the Schrödinger equation to three dimensions and separating it in spherical coordinates produces three ordinary differential equations, one per coordinate. Their boundary conditions generate the quantum numbers n, ℓ, and mℓ, quantize the angular momentum to √(ℓ(ℓ+1))ℏ with projections mℏ, and fix the bound-state energies of hydrogen at −Z²(13.6 eV)\u002Fn².\n",{"path":23504,"title":23505,"module":23501,"summary":23506},"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fhydrogen-wave-functions","Hydrogen Wave Functions and Orbitals","The hydrogen wave functions factor into a radial part Rₙℓ(r) and an angular spherical harmonic Yℓm(θ,φ). Squaring gives the probability cloud; the radial distribution P(r) = r²|ψ|² peaks at the Bohr radius for the ground state and at the Bohr orbits for excited states. The angular part fixes the s, p, and d orbital shapes that govern chemical bonding.\n",{"path":23508,"title":23509,"module":23501,"summary":23510},"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fradial-equation-in-full","Solving the Radial Equation in Full","The hydrogen radial equation is solved from the differential equation up. The substitution u = rR turns it into a one-dimensional problem with a centrifugal barrier; matching the asymptotic behaviour at the origin and at infinity peels off the factors r^(ℓ+1) and e^(−r\u002Fna₀); a Frobenius series for the remainder must terminate, and that termination condition yields the quantization n ≥ ℓ+1 with E = −Z²Ry\u002Fn². The surviving polynomials are the associated Laguerre functions, whose degree n−ℓ−1 counts the radial nodes.\n",{"path":23512,"title":23513,"module":23501,"summary":23514},"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fsymmetry-degeneracy-runge-lenz","Accidental Degeneracy and the Runge-Lenz Symmetry","Hydrogen energies depend only on n, so states of different ℓ at the same n are degenerate. This is not a coincidence but the mark of a hidden symmetry: the quantum Runge-Lenz vector is conserved for the 1\u002Fr potential alone, and together with angular momentum it generates the group SO(4). The Casimir invariant of that group reproduces E = −Z²Ry\u002Fn² and its representations count the n² states. Any departure from 1\u002Fr breaks the symmetry and lifts the ℓ-degeneracy.\n",{"path":23516,"title":23517,"module":23501,"summary":23518},"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fexpectation-values-virial","Expectation Values, the Virial Theorem, and Scaling","The radial matrix elements ⟨r^k⟩ of hydrogenic states are the raw material of every later correction. This lesson derives ⟨1\u002Fr⟩ from the virial theorem, builds the full family ⟨r⟩, ⟨r²⟩, ⟨1\u002Fr²⟩, ⟨1\u002Fr³⟩ from Kramers' recursion and the Feynman-Hellmann theorem, and reads off their scaling with n, ℓ, and Z. The virial balance ⟨T⟩ = −½⟨V⟩ = −E fixes the energy budget of every bound state.\n",{"path":23520,"title":23521,"module":23501,"summary":23522},"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Fquantum-defects-alkali-spectra","Quantum Defects and Alkali Spectra","An alkali atom is one valence electron outside a closed-shell core, and to a good approximation it is hydrogen with a modified quantum number. Core penetration makes low-ℓ states more bound than the Coulomb formula predicts, and the shortfall is captured by a single number per ℓ, the quantum defect δℓ. The spectrum then follows the Rydberg formula with n replaced by the effective n − δℓ, and the sodium D-line doublet is the worked case.\n",{"path":23524,"title":23525,"module":23501,"summary":23526},"\u002Fatomic-physics\u002Fquantum-hydrogen-atom\u002Frydberg-atoms","Rydberg Atoms","A Rydberg atom is an atom excited to a very high principal quantum number, and every hydrogenic property becomes exaggerated by a power of n. Size grows as n², binding falls as n⁻², radiative lifetime lengthens as n³, and the static polarizability explodes as n⁷. The levels crowd toward the ionization limit, and the enormous dipole interaction between two Rydberg atoms produces the blockade that underlies neutral-atom quantum computing.\n",{"path":23528,"title":23529,"module":23530,"summary":23531},"\u002Fatomic-physics\u002Ffine-structure-and-the-dirac-atom\u002Frelativistic-kinetic-correction","The Relativistic Kinetic-Energy Correction","Fine Structure and the Dirac Atom","The Bohr energies treat the electron as slowly moving, but its speed is of order αc, so the kinetic energy needs a relativistic correction. Expanding √(p²c²+m²c⁴) to order (v\u002Fc)² produces the perturbation −p⁴\u002F8m³c², whose first-order shift on a hydrogenic state is evaluated with the trick p²=2m(E−V). The result depends on n and ℓ, is smaller than the gross structure by α²≈5×10⁻⁵, and is one of the three pieces that combine into the fine-structure formula.\n",{"path":23533,"title":23534,"module":23530,"summary":23535},"\u002Fatomic-physics\u002Ffine-structure-and-the-dirac-atom\u002Fspin-orbit-thomas-precession","Spin-Orbit Coupling and Thomas Precession","In the electron's rest frame the nucleus orbits it, and the resulting current produces a magnetic field that couples to the electron's spin moment. The interaction is ξ(r) L·S, with ξ built from the Coulomb potential and the radial expectation ⟨1\u002Fr³⟩. A relativistic subtlety, Thomas precession, halves the naive coefficient because the electron's rest frame is accelerating. The result splits each ℓ≥1 level into a j=ℓ±½ doublet and makes (n, ℓ, j, mⱼ) the good quantum numbers.\n",{"path":23537,"title":23538,"module":23530,"summary":23539},"\u002Fatomic-physics\u002Ffine-structure-and-the-dirac-atom\u002Fdarwin-term-fine-structure-formula","The Darwin Term and the Fine-Structure Formula","The third fine-structure correction, the Darwin term, is a contact interaction proportional to ∇²V that acts only on s-states, physically a smearing of the electron over a Compton wavelength. Adding the relativistic, spin-orbit, and Darwin shifts, the ℓ-dependence cancels and the total collapses to a formula in n and j alone. The n=2 shell splits into 2S₁\u002F₂, 2P₁\u002F₂, 2P₃\u002F₂, with the two j=½ levels exactly degenerate, a coincidence the Dirac theory explains.\n",{"path":23541,"title":23542,"module":23530,"summary":23543},"\u002Fatomic-physics\u002Ffine-structure-and-the-dirac-atom\u002Fdirac-equation-hydrogen","The Dirac Equation for Hydrogen","The fine-structure formula was assembled from three perturbations; the Dirac equation produces it in one stroke and exactly. A first-order relativistic wave equation forces a four-component spinor, from which spin s=½, the g-factor of 2, the spin-orbit term, and antiparticles all emerge automatically. Its exact Coulomb spectrum depends only on n and j, and expanding in Zα reproduces the perturbative result, including the 2S₁\u002F₂–2P₁\u002F₂ degeneracy that sets up the Lamb shift.\n",{"path":23545,"title":23546,"module":23547,"summary":23548},"\u002Fatomic-physics\u002Fqed-corrections-and-hyperfine-structure\u002Flamb-shift-qed","The Lamb Shift and QED Radiative Corrections","QED Corrections and Hyperfine Structure","The Dirac equation makes the 2S₁\u002F₂ and 2P₁\u002F₂ levels of hydrogen exactly degenerate. Lamb and Retherford measured a splitting of about 1058 MHz that the Dirac theory cannot produce. The gap comes from the electron's coupling to the quantized electromagnetic field: self-energy, vacuum polarization, and the anomalous magnetic moment. Welton's vacuum-fluctuation estimate reproduces the size and shows why the effect lands almost entirely on s-states, and the same radiative corrections make hydrogen the most stringent test of QED.\n",{"path":23550,"title":23551,"module":23547,"summary":23552},"\u002Fatomic-physics\u002Fqed-corrections-and-hyperfine-structure\u002Fhyperfine-structure-21cm","Hyperfine Structure and the 21 cm Line","The proton carries a magnetic moment, and it interacts with the magnetic field the electron produces at the nucleus. For s-states that interaction is the Fermi contact term, proportional to the electron density at the origin and to the dot product of the nuclear and electronic spins. Coupling I and J into F = I + J splits each level by a Landé interval rule; in hydrogen's ground state it produces the F = 0\u002FF = 1 doublet whose 1420 MHz, 21 cm transition maps neutral hydrogen across the galaxy.\n",{"path":23554,"title":23555,"module":23547,"summary":23556},"\u002Fatomic-physics\u002Fqed-corrections-and-hyperfine-structure\u002Fnuclear-effects-isotope-shift","Nuclear Size, Moments, and Isotope Shifts","A real nucleus has a finite size, a mass that changes between isotopes, and, when its spin is at least one, an electric quadrupole moment. Each leaves a fingerprint in the atomic spectrum: the volume shift from s-electrons sampling the charge distribution, the mass and field isotope shifts that separate on a King plot, the quadrupole interaction that breaks the Landé interval rule, and the hyperfine anomaly from the magnetization distribution. Atomic spectroscopy reads nuclear properties out of these shifts.\n",{"path":23558,"title":23559,"module":23560,"summary":23561},"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fperiodic-table-atomic-spectra","The Periodic Table and Atomic Spectra","Many-Electron Atoms","Identical electrons demand antisymmetric wave functions, which is the Pauli exclusion principle: no two electrons share all four quantum numbers. Filling shells in order of increasing energy — shifted by penetration and shielding — builds the periodic table and its recurring ionization pattern. Selection rules govern optical spectra, and an external field splits lines by the Zeeman effect.\n",{"path":23563,"title":23564,"module":23560,"summary":23565},"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fcentral-field-self-consistent","The Central-Field Approximation and the Self-Consistent Field","The N-electron Hamiltonian does not separate because every pair of electrons repels. The central-field approximation replaces that pairwise repulsion with an averaged spherical potential each electron feels, restoring hydrogen-like orbitals labelled by n and ℓ. The Thomas-Fermi statistical model fixes the shape of the screened charge from Fermi-gas thermodynamics; the Hartree self-consistent field determines it exactly by iterating orbitals against the potential they generate until the two agree.\n",{"path":23567,"title":23568,"module":23560,"summary":23569},"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fidentical-particles-hartree-fock","Exchange, Slater Determinants, and Hartree-Fock","A product wave function ignores that electrons are identical fermions. Enforcing antisymmetry writes the state as a Slater determinant, which vanishes whenever two electrons share a spin-orbital — the exclusion principle made algebraic. The energy of a determinant carries a new term with no classical analogue, the exchange integral, nonzero only for parallel spins; it lowers the energy of aligned electrons and carves a Fermi hole around each one. Adding the exchange operator to the mean field gives the Hartree-Fock equations, and what they still miss defines the correlation energy.\n",{"path":23571,"title":23572,"module":23560,"summary":23573},"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fhelium-two-electron-atom","Helium: the Prototype Two-Electron Atom","Helium is the smallest atom the Schrödinger equation cannot solve exactly, and the smallest that shows every many-electron effect. Ignoring the electron repulsion overbinds the ground state by 30 eV; first-order perturbation theory and a one-parameter variational calculation with an effective charge close most of the gap. The excited configurations split into para (singlet) and ortho (triplet) states separated by the exchange integral, with the triplet lower — and the absence of a 1s² triplet is the Pauli principle in its plainest form.\n",{"path":23575,"title":23576,"module":23560,"summary":23577},"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fls-jj-coupling-term-symbols","LS and jj Coupling; Term Symbols","A configuration is not a single energy level. The residual electrostatic repulsion and the spin-orbit interaction split it, and which one dominates fixes the coupling scheme. In light atoms the electrostatic term wins: orbital and spin angular momenta couple separately into L and S, then into J, giving Russell- Saunders term symbols. In heavy atoms spin-orbit wins and each electron's j forms first. The Pauli principle prunes the allowed terms of equivalent electrons, the Landé interval rule spaces the fine-structure multiplet, and the scheme crosses over from LS to jj down a column.\n",{"path":23579,"title":23580,"module":23560,"summary":23581},"\u002Fatomic-physics\u002Fmany-electron-atoms\u002Fhund-rules-ground-terms","Hund's Rules and Ground-State Terms","A configuration allows several terms; Hund's three rules pick the ground one. Maximize the spin S first, then the orbital L, then set J to |L−S| for a less-than-half shell and L+S for a more-than-half shell. The first two rules come from exchange lowering the energy of apart-kept electrons; the third comes from the sign of the spin-orbit coupling, which flips as a shell passes half-filling and turns the multiplet from normal to inverted. Worked ground terms for carbon, nitrogen, oxygen, and iron show the rules in action.\n",{"path":23583,"title":23584,"module":23585,"summary":23586},"\u002Fatomic-physics\u002Fatoms-in-external-fields\u002Fzeeman-effect","The Zeeman Effect","Atoms in External Fields","A magnetic field couples to the atom through its magnetic moment, splitting each level into equally spaced sublevels labelled by the projection of the total angular momentum. When spin is present the spacing is not the classical one: it carries the Landé g-factor, a projection of the spin and orbital moments onto the total angular momentum. We derive the weak-field Hamiltonian from minimal coupling, evaluate the shift with the projection theorem, and read off the polarization of the emitted components.\n",{"path":23588,"title":23589,"module":23585,"summary":23590},"\u002Fatomic-physics\u002Fatoms-in-external-fields\u002Fpaschen-back-intermediate","The Paschen-Back and Intermediate-Field Regimes","When the magnetic interaction grows past the fine-structure coupling, spin and orbital angular momentum decouple and precess independently about the field. The anomalous Zeeman pattern reverts to a simple triplet, the Paschen-Back effect. Between the two limits neither coupling dominates and the level positions follow from diagonalizing the combined spin-orbit and Zeeman Hamiltonian. We build the two-by-two problem for a single valence electron, solve it in closed form, and show both limits emerge from one expression.\n",{"path":23592,"title":23593,"module":23585,"summary":23594},"\u002Fatomic-physics\u002Fatoms-in-external-fields\u002Fstark-effect-polarizability","The Stark Effect and Field Ionization","An electric field shifts atomic levels by coupling to the electron's position. Parity forbids a first-order shift for a non-degenerate state, so most atoms respond only at second order through their polarizability, a quadratic Stark shift. Hydrogen is the exception: its accidental degeneracy admits a permanent dipole and a linear shift, cleanest in parabolic coordinates. At large fields the Coulomb well develops a saddle, and Rydberg states field-ionize at a threshold that falls as the fourth power of the principal quantum number.\n",{"path":23596,"title":23597,"module":23598,"summary":23599},"\u002Fatomic-physics\u002Fradiative-transitions-and-line-shapes\u002Ftime-dependent-perturbation-golden-rule","Time-Dependent Perturbation Theory and the Golden Rule","Radiative Transitions and Spectral Lines","An atom in a weak oscillating field makes transitions between its stationary states. First-order time-dependent perturbation theory gives the transition amplitude as a Fourier component of the perturbation at the Bohr frequency, and the resulting probability is a sinc-squared resonance that sharpens as the field acts longer. For a two-level system the same coupling produces Rabi oscillations; for a transition into a continuum the long-time limit collapses the sinc-squared into a delta function and yields Fermi's golden rule, a constant transition rate set by the coupling strength and the density of final states.\n",{"path":23601,"title":23602,"module":23598,"summary":23603},"\u002Fatomic-physics\u002Fradiative-transitions-and-line-shapes\u002Fdipole-approximation-einstein-coefficients","The Dipole Approximation and Einstein Coefficients","The coupling between an atom and light is the interaction of the electron with the electromagnetic field. Because an optical wavelength dwarfs the atom, the spatial variation of the field across the atom can be dropped, leaving the electric-dipole interaction and its matrix element. That matrix element defines the oscillator strength, which obeys the Thomas-Reiche-Kuhn sum rule. Einstein's three rate coefficients (absorption, stimulated emission, spontaneous emission) follow from detailed balance with thermal radiation, fixing the ratio of spontaneous to stimulated rates and its steep growth with frequency.\n",{"path":23605,"title":23606,"module":23598,"summary":23607},"\u002Fatomic-physics\u002Fradiative-transitions-and-line-shapes\u002Fselection-rules-forbidden-transitions","Selection Rules and Forbidden Transitions","The dipole matrix element vanishes for most pairs of states, and the pattern of which survive is the set of selection rules. Parity forces the orbital angular momentum to change by one; the angular integral of three spherical harmonics restricts the magnetic quantum number to change by zero or one; the photon's spin restricts the total angular momentum. When the dipole element vanishes, higher multipoles (magnetic dipole and electric quadrupole) can still drive the transition at rates smaller by powers of the fine-structure constant, and states with no allowed decay become metastable.\n",{"path":23609,"title":23610,"module":23598,"summary":23611},"\u002Fatomic-physics\u002Fradiative-transitions-and-line-shapes\u002Flifetimes-and-line-shapes","Lifetimes, Line Widths, and Line Shapes","A spectral line is never infinitely sharp. The finite lifetime of the excited state gives every line a natural Lorentzian width set by the total decay rate, the Fourier transform of an exponentially damped emission. Thermal motion adds a Gaussian Doppler width that usually dominates in a gas; collisions add a further Lorentzian pressure width; the observed profile is the Voigt convolution of the Gaussian and Lorentzian parts. Strong driving fields broaden the line further through saturation. Each mechanism has a distinct dependence on temperature, density, and intensity that lets it be identified and, where possible, removed.\n",{"path":23613,"title":23614,"module":23615,"summary":23616},"\u002Fatomic-physics\u002Flasers-and-spectroscopy\u002Flaser-principles","Population Inversion, Gain, and the Laser","Lasers and Spectroscopy","A laser is an optical amplifier placed inside a resonant cavity. Amplification requires that stimulated emission outrun absorption, which requires more atoms in the upper level than the lower one — a population inversion that the Einstein relations forbid in thermal equilibrium and that no two-level pump can produce. Three- and four-level schemes reach it by routing atoms through auxiliary states. The gain coefficient sets how strongly a weak beam grows, the cavity fixes the threshold and selects a comb of longitudinal modes, and gain saturation clamps the steady-state inversion at its threshold value.\n",{"path":23618,"title":23619,"module":23615,"summary":23620},"\u002Fatomic-physics\u002Flasers-and-spectroscopy\u002Fspectroscopy-techniques","Spectroscopic Techniques and Frequency Combs","A tunable laser turns spectroscopy from photographing a spectrum into interrogating a single transition, but at room temperature the Doppler width buries the natural linewidth under a thousandfold-broader Gaussian. Saturated absorption and two-photon spectroscopy defeat the first-order Doppler shift by selecting the zero-velocity class or cancelling the shift between counter-propagating photons, recovering natural-width features. Laser-induced fluorescence pushes sensitivity to single atoms, and the optical frequency comb converts an optical frequency into a countable radio-frequency beat, giving absolute frequency measurement across the visible spectrum.\n",{"path":23622,"title":23623,"module":23615,"summary":23624},"\u002Fatomic-physics\u002Flasers-and-spectroscopy\u002Fline-catalog-nist-asd","Reading Real Spectra with the NIST Database","Every quantity computed in this course — energy levels, transition frequencies, oscillator strengths, lifetimes — is tabulated for real atoms in the NIST Atomic Spectra Database. This lesson reads that data as physics: how levels are labelled by term symbols and energies in wavenumbers, how a transition list encodes wavelength, Einstein coefficient, and line strength, how a Grotrian diagram is reconstructed from the tables, and how a measured spectrum is matched to catalog lines. The residual between computed and tabulated positions is the running score of atomic theory.\n",{"path":23626,"title":23627,"module":23628,"summary":23629},"\u002Fatomic-physics\u002Fmodern-atomic-physics\u002Flaser-cooling-doppler","Laser Cooling and Optical Molasses","Modern Atomic Physics","A near-resonant laser beam pushes an atom because every absorbed photon delivers one unit of momentum and the subsequent spontaneous emission averages to zero. Two counter-propagating red-detuned beams turn that push into friction: the Doppler shift brings a moving atom closer to resonance with the beam it moves against, so the net force opposes the velocity. Six beams give optical molasses in three dimensions. The random recoil of spontaneous emission heats against the friction, and the balance sets the Doppler cooling limit. Adding a magnetic-field gradient makes the force position-dependent as well, giving the magneto-optical trap.\n",{"path":23631,"title":23632,"module":23628,"summary":23633},"\u002Fatomic-physics\u002Fmodern-atomic-physics\u002Fsub-doppler-trapping","Sub-Doppler Cooling and Atom Traps","Optical molasses cools multilevel atoms below the Doppler limit. A polarization gradient plus optical pumping makes an atom repeatedly climb a light-shift hill and be pumped to the valley, losing kinetic energy each cycle — Sisyphus cooling. The floor is the recoil limit, one photon momentum of residual motion. Below it, cooling must avoid scattering photons: conservative magnetic and optical-dipole traps hold the atoms while forced evaporation removes the hot tail, driving the phase-space density up toward quantum degeneracy.\n",{"path":23635,"title":23636,"module":23628,"summary":23637},"\u002Fatomic-physics\u002Fmodern-atomic-physics\u002Fbose-einstein-condensation","Bose-Einstein Condensation of Atomic Gases","Below a critical temperature a gas of identical bosons places a macroscopic fraction of its atoms in the single lowest-energy state. The transition occurs when the thermal de Broglie wavelength grows to the interparticle spacing, so the phase- space density reaches order unity. The critical temperature follows from the Bose-Einstein distribution and the density of states, the condensate fraction grows as one minus (T\u002FTc) to the three-halves, and the condensate reveals itself in time-of-flight as a sharp bimodal peak in the momentum distribution. The 1995 rubidium and sodium experiments realized it in dilute trapped gases.\n",{"path":23639,"title":23640,"module":23628,"summary":23641},"\u002Fatomic-physics\u002Fmodern-atomic-physics\u002Foptical-clocks-precision","Optical Atomic Clocks and Precision Measurement","An atomic clock counts the oscillations of a field locked to an atomic transition. The cesium microwave standard defines the second through the 9.19 GHz ground-state hyperfine transition, interrogated by Ramsey's separated-oscillatory-field method whose fringe width is set by the free-precession time. Optical clocks replace the microwave transition with an optical one five orders of magnitude higher in frequency, raising the quality factor and the fractional stability in proportion. Lattice and single-ion clocks reach fractional uncertainties near ten-to-the-minus- eighteen by trapping the atoms at a magic wavelength that cancels the light shift, and at that level they measure the gravitational redshift over centimetres of height.\n",{"path":23643,"title":23644,"module":6,"summary":6},"\u002Fatomic-physics","Atomic Physics",{"path":23646,"title":23647,"module":6,"summary":6},"\u002Fdatabases","Databases",{"path":23649,"title":23650,"module":21235,"summary":23651},"\u002Fcategory-theory\u002Ffoundations\u002Fwhat-is-a-category","Categories, Objects, and Arrows","A category is objects, arrows between them, a rule for composing arrows, and an identity arrow on every object, subject to associativity and the unit laws. The axioms mention no elements: arrows need not be functions, and an object is known only through the arrows into and out of it. Isomorphism, commutative diagrams, duality, and the terminal object are the first consequences.\n",{"path":23653,"title":23654,"module":21235,"summary":23655},"\u002Fcategory-theory\u002Ffoundations\u002Fexamples-of-categories","A Zoo of Categories","The axioms admit two very different kinds of model: large categories of structured sets and their structure-preserving maps (Set, Mon, Grp, Top, Vect), and small categories that are themselves single algebraic objects — a monoid as a one-object category, a poset as a thin category. The awkward cases Rel and Pfn have sets as objects but relations and partial functions as arrows, and a typed programming language presents its types and programs as a category.\n",{"path":23657,"title":23658,"module":21235,"summary":23659},"\u002Fcategory-theory\u002Ffoundations\u002Fspecial-morphisms","Isomorphisms, Monos, and Epis","Injectivity and surjectivity mention elements, so a general category re-expresses them by cancellation: monomorphisms cancel on the left, epimorphisms on the right. Sections and retractions are the split versions with an explicit one-sided inverse. Mono plus epi does not force an isomorphism, and subobjects are equivalence classes of monos into a fixed object.\n",{"path":23661,"title":23662,"module":21235,"summary":23663},"\u002Fcategory-theory\u002Ffoundations\u002Ffunctors","Functors: Maps Between Categories","A functor sends objects to objects and arrows to arrows while preserving composition and identities. Covariant and contravariant functors, the standard stock (forgetful, free, hom, and powerset), and the classification by faithfulness, fullness, and essential surjectivity all follow. Functors compose, so categories and functors form a category themselves.\n",{"path":23665,"title":23666,"module":21235,"summary":23667},"\u002Fcategory-theory\u002Ffoundations\u002Fnatural-transformations","Natural Transformations and Functor Categories","A natural transformation is a map between two parallel functors: one component arrow per object, subject to a commuting square for every arrow of the source. Naturality is verified for the determinant, the double dual, and list operations; functors and natural transformations form the functor category [C, D]; and vertical and horizontal composition satisfy the Godement interchange law.\n",{"path":23669,"title":23670,"module":21235,"summary":23671},"\u002Fcategory-theory\u002Ffoundations\u002Fsize-and-set-theory","Size: Small, Large, Locally Small","The objects of Set do not form a set, and pretending otherwise reproduces the classical paradoxes. Classes make the small\u002Flarge distinction precise, with locally small and essentially small as the intermediate notions. Cantor's theorem shows Set and its algebraic relatives are large, and the function-based axiomatization of sets is the one category theory prefers to ZFC.\n",{"path":23673,"title":23674,"module":23675,"summary":23676},"\u002Fcategory-theory\u002Funiversal-properties\u002Funiversal-properties","Universal Properties, Initial and Terminal Objects","Universal Properties and Basic Constructions","A universal property characterizes an object by a for-all\u002Fexists-unique condition on the arrows into or out of it, and any two objects satisfying the same property are isomorphic by a unique isomorphism. Initial and terminal objects are the simplest cases; the free vector space, the discrete topology, and the ring of integers show the pattern at work.\n",{"path":23678,"title":23679,"module":23675,"summary":23680},"\u002Fcategory-theory\u002Funiversal-properties\u002Fproducts-and-coproducts","Products and Coproducts","The product of two objects is a wedge of projections through which every other wedge factors uniquely; the coproduct is the dual, built from injections. In Set these are the cartesian product and the disjoint union, in a poset the meet and join, and in abelian groups the two coincide. The mediating-arrow discipline established here is the template for all limits.\n",{"path":23682,"title":23683,"module":23675,"summary":23684},"\u002Fcategory-theory\u002Funiversal-properties\u002Fconstructions-on-categories","Opposite, Product, Slice, and Comma Categories","Categories are themselves mathematical structures, and the standard algebraic constructions apply: opposites, products, subcategories, slices, and the comma category that subsumes them. The opposite category yields the duality principle, halving the subject's proofs; slice and comma categories repackage every universal property as an initial or terminal object.\n",{"path":23686,"title":23687,"module":23688,"summary":23689},"\u002Fcategory-theory\u002Frepresentables-yoneda\u002Frepresentable-functors","Hom-Functors and Representables","Representables and the Yoneda Lemma","Fixing an object A of a locally small category produces a set-valued functor, the hom-functor A(A,-), that records every map out of A. A functor is representable when it is naturally isomorphic to such a hom-functor. We define the covariant and contravariant hom-functors, collect the standard representables (identity, forgetful, powerset), and read maps as generalized elements of varying shape.\n",{"path":23691,"title":23692,"module":23688,"summary":23693},"\u002Fcategory-theory\u002Frepresentables-yoneda\u002Fyoneda-lemma","The Yoneda Lemma","The Yoneda lemma computes the natural transformations out of a representable presheaf: they form a set in natural bijection with X(A). The proof fixes a single degree of freedom, the image of the identity arrow, and shows naturality forces everything else. We prove the bijection, verify naturality in both variables, and read off that a natural transformation out of a representable is just one element.\n",{"path":23695,"title":23696,"module":23688,"summary":23697},"\u002Fcategory-theory\u002Frepresentables-yoneda\u002Fyoneda-consequences","The Yoneda Embedding and Its Uses","Three corollaries turn the Yoneda lemma into working machinery. A representation of a presheaf is the same thing as a universal element; the Yoneda embedding of a category into its presheaf category is full and faithful; and two objects are isomorphic exactly when their representables are. Together they justify constructing arrows by constructing natural transformations between hom-functors, and they contain Cayley's theorem as the one-object case.\n",{"path":23699,"title":23700,"module":23701,"summary":23702},"\u002Fcategory-theory\u002Flimits-colimits\u002Flimits","Cones and Limits","Limits and Colimits","A diagram is a functor from a small shape category; a cone over it is an object with compatible legs to every node; and a limit is the terminal cone, the one every other cone factors through uniquely. Products and terminal objects reappear as limits over particular shapes, and the whole construction is unique up to a single isomorphism.\n",{"path":23704,"title":23705,"module":23701,"summary":23706},"\u002Fcategory-theory\u002Flimits-colimits\u002Fproducts-equalizers-pullbacks","Equalizers and Pullbacks","The equalizer of a parallel pair is the universal arrow that makes the two composites agree; the pullback of a cospan is the universal commutative square. In Set they are solution sets and fibered products, every equalizer is monic, monics are stable under pullback, and products plus equalizers together generate all limits.\n",{"path":23708,"title":23709,"module":23701,"summary":23710},"\u002Fcategory-theory\u002Flimits-colimits\u002Fcolimits","Colimits: Coproducts, Coequalizers, Pushouts","Colimits are limits in the opposite category: cocones replace cones, and the universal cocone is initial rather than terminal. Coproducts glue objects side by side, coequalizers impose relations and produce quotients, pushouts glue along a shared part, and in Set every colimit is a quotient of a disjoint union. Directed colimits admit a clean elementwise description.\n",{"path":23712,"title":23713,"module":23701,"summary":23714},"\u002Fcategory-theory\u002Flimits-colimits\u002Fcomputing-limits","Computing Limits in Concrete Categories","In Set the limit of any diagram is the set of threads: choice functions through the nodes that commute with every edge. In Pos, Mon, and Top the recipe is the same limit downstairs plus the unique structure that makes the projections structure-preserving — pointwise order, componentwise operations, the topology generated by the projections. The pattern is what \"the forgetful functor creates limits\" means concretely.\n",{"path":23716,"title":23717,"module":23701,"summary":23718},"\u002Fcategory-theory\u002Flimits-colimits\u002Flimits-and-functors","Preservation, Reflection, and Creation of Limits","A functor preserves limits if it sends limit cones to limit cones, reflects them if it recognizes them, and creates them if limits downstairs lift uniquely upstairs. Representable functors preserve all limits, forgetful functors from algebra create them, and limits in functor categories are computed pointwise, one evaluation at a time.\n",{"path":23720,"title":23721,"module":23722,"summary":23723},"\u002Fcategory-theory\u002Fadjunctions\u002Fadjunctions","Adjoint Functors via Hom-Set Bijections","Adjunctions","An adjunction is a natural bijection between two hom-sets: maps out of $F(A)$ in one category correspond to maps into $G(B)$ in the other. We give the definition, spell out the naturality axioms that make the correspondence compatible with composition, and work the flagship examples — free vector spaces, free groups, discrete and indiscrete topologies, and currying.\n",{"path":23725,"title":23726,"module":23722,"summary":23727},"\u002Fcategory-theory\u002Fadjunctions\u002Funits-and-counits","Units, Counits, and the Triangle Identities","The whole hom-set bijection of an adjunction is generated by two natural transformations: the unit, obtained by transposing identity maps on one side, and the counit, by transposing them on the other. Two triangle identities are all they must satisfy, and any pair satisfying them determines a unique adjunction. The same correspondence specializes to order-preserving maps between posets and to free constructions.\n",{"path":23729,"title":23730,"module":23722,"summary":23731},"\u002Fcategory-theory\u002Fadjunctions\u002Fadjunctions-via-universal-arrows","Adjunctions from Universal Arrows","The unit component at a single object is an initial object of a comma category, and this universal property alone rebuilds the whole adjunction. A functor has a left adjoint exactly when every object admits such a universal arrow, and the left adjoint is assembled from them one object at a time. We prove the equivalence of all three formulations of adjointness.\n",{"path":23733,"title":23734,"module":23722,"summary":23735},"\u002Fcategory-theory\u002Fadjunctions\u002Ffree-forgetful-adjunctions","Free Constructions and Free–Forgetful Adjunctions","Free monoids, free groups, and free vector spaces are left adjoints to forgetful functors, and the universal mapping property is all one needs to prove it. Some forgetful functors also have right adjoints (co-free constructions like the indiscrete topology), producing three-functor chains. Contravariant adjunctions, symmetric in their two functors, close the lesson with the pattern behind duality and representation theorems.\n",{"path":23737,"title":23738,"module":23739,"summary":23740},"\u002Fcategory-theory\u002Fadjoints-limits\u002Flimits-via-adjoints","Limits as Adjoints and as Representables","Adjoints, Representables, and Limits Together","A cone on a diagram is a natural transformation from a constant diagram, so a limit is a representation of the cone functor and, equivalently, a value of the right adjoint to the diagonal functor. We prove both rephrasings, derive uniqueness and functoriality of limits from them, and record the dual statement that a colimit is the left adjoint to the diagonal.\n",{"path":23742,"title":23743,"module":23739,"summary":23744},"\u002Fcategory-theory\u002Fadjoints-limits\u002Fpresheaf-limits-colimits","Limits and Colimits of Presheaves","Representables preserve limits, and limits in a functor category are computed one object at a time, so a presheaf category is complete and cocomplete with all its structure inherited pointwise from Set. The Yoneda embedding then preserves limits but not colimits, and the density theorem repairs the colimit side: every presheaf is a canonical colimit of representables.\n",{"path":23746,"title":23747,"module":23739,"summary":23748},"\u002Fcategory-theory\u002Fadjoints-limits\u002Fadjoints-preserve-limits","Right Adjoints Preserve Limits (RAPL)","A functor with a left adjoint preserves every limit that exists, and dually a functor with a right adjoint preserves colimits. The proof is a four-line chain of natural isomorphisms through the adjunction and the continuity of representables. The theorem yields product-and-exponential arithmetic in Set, another proof that limits commute with limits, and a standard test for proving that a functor has no adjoint.\n",{"path":23750,"title":23751,"module":23739,"summary":23752},"\u002Fcategory-theory\u002Fadjoints-limits\u002Fadjoint-functor-theorem","The Adjoint Functor Theorem","RAPL makes limit preservation necessary for having a left adjoint; the adjoint functor theorems identify when it is sufficient. For ordered sets no extra hypothesis is needed. In general the candidate adjoint is a limit over a comma category that may be large, and the general adjoint functor theorem tames it with a weakly initial set. We prove GAFT in full and apply it to free groups and, through the special adjoint functor theorem, the Stone–Čech compactification.\n",{"path":23754,"title":23755,"module":23756,"summary":23757},"\u002Fcategory-theory\u002Fmonads-algebras\u002Fmonads","Monads from Adjunctions","Monads and Algebras","A monad on a category is an endofunctor equipped with a unit and a multiplication satisfying associativity and unit laws — the data of a monoid, written internally to the category of endofunctors. Every adjunction induces one, and the list, exception, and state constructions that model computational effects are all monads on Set.\n",{"path":23759,"title":23760,"module":23756,"summary":23761},"\u002Fcategory-theory\u002Fmonads-algebras\u002Falgebras-eilenberg-moore","Algebras for a Monad","An algebra for a monad is an object with a structure map that interacts correctly with the unit and multiplication. The algebras form the Eilenberg–Moore category, whose free–forgetful adjunction induces the monad back; a comparison functor relates any other inducing adjunction to it, and for the list monad the algebras are exactly monoids.\n",{"path":23763,"title":23764,"module":23756,"summary":23765},"\u002Fcategory-theory\u002Fmonads-algebras\u002Fkleisli-and-programming","The Kleisli Category and Monads in Programming","The Kleisli category of a monad has the same objects as the base but takes arrows A to TB, composed by mapping and flattening. These arrows are effectful programs, Kleisli composition is the bind of functional programming, and the Kleisli adjunction is the initial resolution of the monad, with Eilenberg–Moore at the terminal end.\n",{"path":23767,"title":23768,"module":23756,"summary":23769},"\u002Fcategory-theory\u002Fmonads-algebras\u002Falgebras-for-endofunctors","Algebras for an Endofunctor and Recursion","Dropping the monad laws leaves algebras for a bare endofunctor, whose initial objects are the least fixed points of the functor by Lambek's lemma. The natural numbers, lists, and trees are initial algebras; the unique map out of an initial algebra is the fold of functional programming; and the Smyth–Plotkin fixed-point technique builds Scott domains the same way.\n",{"path":23771,"title":23772,"module":23773,"summary":23774},"\u002Fcategory-theory\u002Fcartesian-closed-lambda\u002Fcartesian-closed-categories","Cartesian Closed Categories","Cartesian Closed Categories and Typed Lambda Calculus","A cartesian closed category has a terminal object, binary products, and for every pair of objects an exponential object that internalizes the hom-set as an object of the category. The defining data is an evaluation arrow and a currying operation, packaged by the adjunction between product-with-A and exponential-by-A. Set, Boolean and Heyting algebras, functor categories, and Cat are all cartesian closed.\n",{"path":23776,"title":23777,"module":23773,"summary":23778},"\u002Fcategory-theory\u002Fcartesian-closed-lambda\u002Flambda-calculus-correspondence","Typed Lambda Calculus and CCCs","The typed lambda calculus and the cartesian closed category are two presentations of the same theory. Types become objects, terms with one free variable become arrows, product types become products, and function types become exponentials, with abstraction matching currying and application matching evaluation. Building the category of a lambda theory and the internal language of a category are mutually inverse up to equivalence.\n",{"path":23780,"title":23781,"module":23773,"summary":23782},"\u002Fcategory-theory\u002Fcartesian-closed-lambda\u002Ffixed-points-and-recursion","Fixed Points in Cartesian Closed Categories","The untyped lambda calculus has a fixed-point combinator; the typed calculus cannot, and Lawvere's fixed-point theorem explains why: any point-surjection onto an exponential forces every endomap to have a fixed point, which is the abstract form of Cantor's diagonal argument. Recursion is recovered instead by restricting to omega-complete partially ordered objects, where every continuous endomap has a least fixed point built by iterating from bottom. This gives While loops a semantics.\n",{"path":23784,"title":23785,"module":6,"summary":6},"\u002Fcategory-theory","Category Theory",{"path":23787,"title":22429,"module":23788,"summary":23789},"\u002Fdeep-learning\u002Fmathematical-background\u002Flinear-algebra-for-deep-learning","Mathematical Background","Every quantity a network touches is a tensor, and every layer is a matrix acting on one. This lesson compiles the linear algebra deep learning actually uses: products and norms, the system $Ax=b$ and when it is solvable, the two decompositions (eigen and SVD) that diagonalize a transformation, and the pseudoinverse that solves what cannot be solved exactly. It then derives PCA as the worked example that ties it all together.\n",{"path":23791,"title":23792,"module":23788,"summary":23793},"\u002Fdeep-learning\u002Fmathematical-background\u002Fprobability-and-information-theory","Probability & Information Theory","This lesson assembles the probabilistic vocabulary a network is trained in (random variables, densities, the chain rule, expectation and covariance, the handful of distributions that recur everywhere) and then the information theory that turns a probabilistic model into a loss: self-information, entropy, and the KL divergence whose asymmetry is the cross-entropy objective itself.\n",{"path":23795,"title":23796,"module":23788,"summary":23797},"\u002Fdeep-learning\u002Fmathematical-background\u002Fnumerical-computation","Numerical Computation","Machine learning runs on finite-precision arithmetic, where every number is approximated and every operation rounds. This lesson sets the numerical ground rules: overflow and underflow and the standard stabilizations, the condition number that measures how much a problem amplifies error, and the gradient-based optimization (first and second order, constrained and unconstrained) that every training loop runs.\n",{"path":23799,"title":21788,"module":23788,"summary":23800},"\u002Fdeep-learning\u002Fmathematical-background\u002Fcalculus","This lesson assembles the differential calculus used in training networks: the gradient and directional derivative, the Jacobian and Hessian, and the chain rule in scalar, vector, and matrix form. From the chain rule it derives back-propagation as a single sweep over the computational graph, tabulates the matrix-calculus identities that recur in layer gradients, reads optimization off a second-order Taylor expansion, and ends with why reverse-mode automatic differentiation is the algorithm every framework runs.\n",{"path":23802,"title":23803,"module":21235,"summary":23804},"\u002Fdeep-learning\u002Ffoundations\u002Fwhat-is-deep-learning","What Is Deep Learning?","Deep learning is representation learning by composition: stack simple differentiable layers, define a loss, and let gradient descent discover the features a human would otherwise have to engineer by hand. We set up the whole vocabulary (model, loss, optimizer, data), the training loop that ties them together, and the three reasons the approach became practical.\n",{"path":23806,"title":23807,"module":21235,"summary":23808},"\u002Fdeep-learning\u002Ffoundations\u002Fmachine-learning-refresher","A Machine-Learning Refresher","The statistical framework the networks live in: data drawn from an unknown distribution, a loss to minimize, and the central question of generalization: will it work on data we have not seen? We set up empirical risk, capacity, the bias–variance tradeoff, and maximum likelihood.\n",{"path":23810,"title":23811,"module":21235,"summary":23812},"\u002Fdeep-learning\u002Ffoundations\u002Flinear-models-and-the-perceptron","Linear Models & the Perceptron","The simplest learners (linear regression, logistic regression, the perceptron) already contain the whole template: a weighted sum, a loss, a gradient step. They also fail on the XOR problem, which no linear model can solve — the limitation that motivates deep learning.\n",{"path":109,"title":23814,"module":20293,"summary":23815},"The Multilayer Perceptron","Stacking linear layers with a nonlinearity between them removes the limitation that stopped the perceptron. We build the multilayer perceptron in explicit matrix form (the forward pass, its dimensions, a worked XOR network with concrete weights) and prove why the nonlinearity is essential: without it the deepest stack collapses to a single hyperplane.\n",{"path":20295,"title":5,"module":20293,"summary":20316},{"path":23818,"title":23819,"module":20293,"summary":23820},"\u002Fdeep-learning\u002Fneural-networks\u002Funiversal-approximation","Universal Approximation","One hidden layer with a non-polynomial activation can approximate any continuous function on a compact set to arbitrary accuracy: the universal approximation theorem. We prove it constructively (two sigmoids make a bump; sums of bumps make any curve), then show the limitation: existence is not efficiency. Depth-separation results exhibit functions a deep net represents with $O(n)$ units that a shallow net needs $\\exp(n)$ units to match.\n",{"path":317,"title":23822,"module":20293,"summary":23823},"Backpropagation","Backpropagation is the chain rule run backward over a computational graph. We formalize the graph, derive the four backprop equations for an MLP, present the forward and backward passes as algorithms, and work a tiny two-layer net by hand with explicit numbers. The result: one scalar loss, reverse-mode autodiff, and a gradient for every parameter at twice the cost of a forward pass.\n",{"path":17544,"title":23825,"module":20293,"summary":23826},"Loss Functions & Output Units","The last layer is where a network's hidden representation meets the task. Choosing an output unit and a loss is not two independent choices; maximum likelihood fixes the pair. We derive the standard couplings (linear\u002FMSE, sigmoid\u002FBCE, softmax\u002Fcross-entropy), show why softmax and cross-entropy were built to cancel into the residual $\\hat y - y$, and prove why squared error is the wrong loss for a saturating classifier.\n",{"path":23828,"title":23829,"module":23830,"summary":23831},"\u002Fdeep-learning\u002Foptimization\u002Fgradient-descent-and-sgd","Gradient Descent & SGD","Optimization","Training is descent on the empirical risk: step the parameters against the gradient. We derive the minibatch gradient as an unbiased estimator whose variance falls as $1\u002FB$, derive the learning-rate ceiling from the smoothness-stability bound $\\eta \u003C 2\u002FL$, and lay out the schedules (step, exponential, cosine, warmup) that anneal it over training.\n",{"path":23833,"title":23834,"module":23830,"summary":23835},"\u002Fdeep-learning\u002Foptimization\u002Fmomentum-and-adaptive-methods","Momentum & Adaptive Methods","Plain gradient descent zig-zags across ravines and moves slowly along flat valleys, because one global learning rate cannot suit a surface with wildly different curvature in different directions. Two fixes address the two problems: momentum accumulates a velocity that damps the oscillation and accelerates the drift, and adaptive methods give every parameter its own learning rate scaled by the history of its gradients. Adam fuses both, and is the default optimizer of modern deep learning.\n",{"path":15181,"title":23837,"module":23830,"summary":23838},"Weight Initialization","The initial weights determine whether training can succeed before the first gradient step. Initialize every weight equal and all hidden units compute the same function forever; initialize too small or too large and the signal vanishes or explodes as it crosses depth. A single variance condition, $n_{\\text{in}}\\mathrm{Var}(W)=1$, fixes both, and reading it off the forward and backward passes yields Xavier and He initialization directly.\n",{"path":23840,"title":23841,"module":23830,"summary":23842},"\u002Fdeep-learning\u002Foptimization\u002Fthe-optimization-landscape","The Optimization Landscape","The loss of a deep network is a non-convex surface in millions of dimensions, so local search carries no global guarantee, yet it works. We classify critical points by the eigenvalues of the Hessian, show that in high dimension nearly all of them are saddle points rather than bad local minima, and read off the practical terrain — plateaus, cliffs, ill-conditioning, and the sharp-versus-flat distinction that ties the geometry of a minimum to how well it generalizes.\n",{"path":23844,"title":23845,"module":23830,"summary":23846},"\u002Fdeep-learning\u002Foptimization\u002Fsecond-order-and-approximate-methods","Second-Order & Approximate Methods","Newton's method reads the curvature of the loss off its Hessian and jumps to the minimum of the local quadratic in a single step, rescaling away the ill-conditioning that slows first-order descent. We derive it, then explain the three obstacles that keep it out of deep learning: a $d \\times d$ Hessian for $d$ in the billions, an attraction to saddle points, and minibatch noise. The alternative is approximation (conjugate gradients, BFGS and L-BFGS, the natural gradient and Hessian-free methods), each buying some of Newton's curvature information without ever forming or inverting $H$.\n",{"path":23848,"title":23849,"module":23850,"summary":23851},"\u002Fdeep-learning\u002Fregularization\u002Fregularization-overview","Regularization Overview","Regularization","Regularization is any modification to a learning algorithm meant to lower test error at the possible expense of training error. We derive the bias–variance decomposition that explains why it helps, set up the two parameter-norm penalties, $L^2$ weight decay and $L^1$, derive their update rules and eigenbasis shrinkage, show geometrically why $L^1$ alone produces sparse weights (soft-thresholding), distinguish weight decay from loss-added $L^2$ under AdamW, and read both penalties through the two lenses that recur across the chapter: a norm-ball constraint via KKT, and a prior via MAP estimation.\n",{"path":23853,"title":23854,"module":23850,"summary":23855},"\u002Fdeep-learning\u002Fregularization\u002Fdropout-and-data-augmentation","Dropout & Data Augmentation","Two of the most effective regularizers add no penalty term at all; they perturb the computation instead. Dropout multiplies hidden units by a random Bernoulli mask, training an exponential ensemble of thinned subnetworks that share weights; inverted scaling collapses that ensemble into one cheap forward pass at test time. Data augmentation enlarges the training set with label-preserving transforms, injecting the invariances the task demands, and noise injection (input, weight, label smoothing, Mixup) generalizes the same idea into a continuous family.\n",{"path":23857,"title":23858,"module":23850,"summary":23859},"\u002Fdeep-learning\u002Fregularization\u002Fearly-stopping-and-parameter-sharing","Early Stopping & Parameter Sharing","Two cheap regularizers that cost no extra term in the loss. Early stopping treats training time itself as a hyperparameter (watch the validation curve, halt at its minimum, keep the best checkpoint), and for a quadratic objective it is provably equivalent to $L^2$ weight decay. Parameter sharing goes the other way: it constrains many weights to be _equal_, the prior behind every convolution and every recurrent step, and the reason a CNN has orders of magnitude fewer parameters than the dense net it replaces.\n",{"path":23861,"title":23862,"module":23850,"summary":23863},"\u002Fdeep-learning\u002Fregularization\u002Fnormalization","Normalization","Normalization layers standardize activations to zero mean and unit variance inside the network, then hand the model a learnable scale and shift to undo the constraint when it pays to. Batch normalization does this across the batch and must keep separate train-time and test-time statistics; layer, instance, and group norm change only the axes they average over. The result is faster, better-conditioned optimization and a free dose of regularizing batch noise.\n",{"path":23865,"title":23866,"module":23867,"summary":23868},"\u002Fdeep-learning\u002Farchitectures\u002Fconvolutional-networks","Convolutional Networks","Architectures","A convolutional network replaces the dense layer's all-to-all weight matrix with a small kernel slid across the input. Three structural commitments (sparse connectivity, parameter sharing, and translation equivariance) collapse the parameter count by orders of magnitude and bake the right prior for images directly into the architecture. We derive the convolution arithmetic, the output geometry, pooling, and the receptive field, then assemble the canonical stack.\n",{"path":23870,"title":23871,"module":23867,"summary":23872},"\u002Fdeep-learning\u002Farchitectures\u002Fcnn-architectures","CNN Architectures","Six landmark networks, each contributing exactly one idea: LeNet's conv-pool stack, AlexNet's ReLU-and-dropout scale, VGG's $3\\times3$ uniformity, Inception's multi-scale module, ResNet's residual skip, and DenseNet's dense connectivity. The common thread is the degradation problem (why plain deeper nets train worse, not just overfit) and the residual block that solved it by keeping a $+1$ path open for the gradient.\n",{"path":23874,"title":23875,"module":23867,"summary":23876},"\u002Fdeep-learning\u002Farchitectures\u002Frecurrent-networks","Recurrent Networks","A recurrent network folds a sequence into a fixed-size hidden state, reusing one set of weights at every time step, the architectural prior that the same rule applies wherever it lands in time. Unrolling the recurrence exposes a deep feed-forward graph; backpropagation through it sums gradient contributions across all steps and chains a product of Jacobians, and that product is why long-range gradients vanish or explode. That failure motivates gated architectures.\n",{"path":23878,"title":23879,"module":23867,"summary":23880},"\u002Fdeep-learning\u002Farchitectures\u002Flstm-and-gru","LSTM & GRU","A plain recurrent network propagates its hidden state through a repeated weight-matrix multiply, and the Jacobian product that results vanishes or explodes long before a useful gradient can reach the early steps. Gated RNNs fix this with an additive memory path: a cell state that is carried forward almost unchanged, past which the gradient flows along a near-identity highway. We derive that highway, give the full LSTM and GRU equations, and compare the two.\n",{"path":23882,"title":23883,"module":23867,"summary":23884},"\u002Fdeep-learning\u002Farchitectures\u002Fattention-and-transformers","Attention & Transformers","Attention replaces fixed wiring with content-based routing: every position reads from every other through a soft, learned dot-product lookup. We derive scaled dot-product attention and its $\\sqrt{d_k}$ correction, build it into multi-head self-attention, inject order with positional encodings, and stack the whole thing into the Transformer block that displaced recurrence and convolution alike.\n",{"path":23886,"title":23887,"module":23867,"summary":23888},"\u002Fdeep-learning\u002Farchitectures\u002Fthe-transformer-architecture","The Transformer Architecture","The Transformer is the architecture built around the attention mechanism. This first part assembles the full encoder–decoder of \"Attention Is All You Need\" — embeddings and positional encoding, stacked self-attention and feed-forward sublayers wrapped in residual connections and LayerNorm, masked decoding and cross-attention — works through causal masking and the three modern families (encoder-only, decoder-only, encoder–decoder), and accounts for where the parameters and the $O(n^2)$ compute actually go.\n",{"path":23890,"title":23891,"module":23867,"summary":23892},"\u002Fdeep-learning\u002Farchitectures\u002Ftransformers-in-practice","Transformers in Practice","The Transformer makes no assumption about what a token represents. This part follows the architecture out of language: image patches feed a plain encoder (the Vision Transformer), the decoder-only half scales into the GPT line of large language models, and one substrate covers translation, retrieval, and multimodal grounding. We work the ViT patch arithmetic and a GPT parameter count by hand, then close on the empirical scaling laws — power-law loss, the Chinchilla compute-optimal balance, and emergent behavior — that made scale the dominant lever.\n",{"path":23894,"title":23895,"module":23867,"summary":23896},"\u002Fdeep-learning\u002Farchitectures\u002Fgraph-neural-networks","Graph Neural Networks","A graph neural network learns on data with no grid and no canonical ordering: atoms in a molecule, users in a social network, road segments in a map. The unifying idea is message passing — each node repeatedly aggregates its neighbors' states and updates its own — built to respect the one symmetry graphs demand, permutation equivariance. We derive the message-passing framework, specialize it into GCN, GraphSAGE, GAT, and GIN, read off graph-level outputs, and bound what message passing can and cannot tell apart.\n",{"path":23898,"title":23899,"module":23867,"summary":23900},"\u002Fdeep-learning\u002Farchitectures\u002Fstate-space-models","State-Space Models and Mamba","A state-space model carries a continuous linear hidden state through a sequence, and that linearity buys two equivalent algorithms from one set of weights: a recurrence that runs in linear time with constant memory, and a global convolution that trains in parallel. Long-range memory comes from how the transition matrix is initialized (HiPPO) and parameterized (S4's diagonal-plus-low-rank form). Mamba breaks the convolution on purpose, making the parameters input-dependent so the model can select what to remember, recovered at speed by a hardware-aware parallel scan.\n",{"path":23902,"title":23903,"module":23904,"summary":23905},"\u002Fdeep-learning\u002Ftheory\u002Fgeneralization-theory","Generalization Theory","Theory & Frontiers","Classical learning theory bounds the gap between training and test error by a model's capacity (VC dimension, Rademacher complexity), and predicts that a model with more parameters than data should overfit catastrophically. Modern networks do the opposite: they interpolate, even fit pure noise, and still generalize. We derive the classical bounds, work the bias-variance decomposition, show why the bounds go vacuous, and survey what replaced them: double descent, the interpolation threshold, margin and norm-based bounds, and the implicit bias of the optimizer itself.\n",{"path":23907,"title":23908,"module":23904,"summary":23909},"\u002Fdeep-learning\u002Ftheory\u002Fadversarial-robustness","Adversarial Robustness","A trained network can be fooled by a perturbation too small for a human to see: add a carefully aimed vector of magnitude $\\epsilon$ to a correctly classified image and the prediction flips. We derive the fast gradient sign method as the first-order-optimal step inside an $L_\\infty$ ball, explain the linearity hypothesis that makes high-dimensional models so easy to push around, build up to projected gradient descent, and frame adversarial training as a min-max robust-optimization problem with its own accuracy cost. Defenses beyond training continue in the next lesson.\n",{"path":23911,"title":23912,"module":23904,"summary":23913},"\u002Fdeep-learning\u002Ftheory\u002Fadversarial-defenses","Adversarial Defenses","Defending a network against an adversary is far harder than attacking one. This lesson covers the defense side: certified guarantees via randomized smoothing, the transferability that makes black-box attacks possible, and the recurring failure of gradient masking, where a defense hides the attacker's gradient instead of moving the decision boundary. It ends with the adaptive-attack discipline (BPDA, EOT, transfer) that every robustness claim must be tested against.\n",{"path":23915,"title":23916,"module":23904,"summary":23917},"\u002Fdeep-learning\u002Ftheory\u002Fbayesian-and-ensemble-methods","Bayesian & Ensemble Methods","A trained network returns a single point prediction and, with the softmax, a confidence, but that confidence is usually miscalibrated, collapsing to near- certainty even on inputs the model has never seen. This lesson covers uncertainty estimation for networks: the two kinds of uncertainty, the Bayesian posterior over weights and its tractable stand-ins (MC dropout, deep ensembles), and how to check whether a model's reported confidences match observed frequencies.\n",{"path":23919,"title":23920,"module":23904,"summary":23921},"\u002Fdeep-learning\u002Ftheory\u002Fdeep-equilibrium-models","Deep Equilibrium Models","A deep network need not be a fixed stack of layers; it can be a single weight-tied layer iterated to convergence, its output defined implicitly as the fixed point $z^\\star = f_\\theta(z^\\star, x)$. The forward pass becomes root-finding and the backward pass becomes implicit differentiation, so training costs O(1) memory regardless of effective depth. We derive both passes from the implicit function theorem and close the course on defining a layer by a fixed-point condition rather than an explicit stack.\n",{"path":23923,"title":23924,"module":23925,"summary":23926},"\u002Fdeep-learning\u002Fgenerative-models\u002Flinear-factor-models","Linear Factor Models","Generative Models","The simplest generative models share one template: a latent variable drawn from a fixed prior, run through a linear decoder, plus noise. Probabilistic PCA, factor analysis, independent component analysis, and sparse coding are all this template with a different prior on the latents and a different noise model. We derive each marginal, see why ICA needs non-Gaussianity to identify its sources, and show how sparse coding learns Gabor-like dictionary atoms.\n",{"path":23928,"title":23929,"module":23925,"summary":23930},"\u002Fdeep-learning\u002Fgenerative-models\u002Fautoencoders","Autoencoders","An autoencoder is a network trained to copy its input to its output through a narrow channel; the useful product is the bottleneck representation $h$, not the reconstruction. We derive the undercomplete autoencoder and prove its linear case recovers PCA, then trade the bottleneck for explicit regularization (sparse, denoising, contractive) and show how a denoising autoencoder learns the low-dimensional manifold the data lives on.\n",{"path":23932,"title":23933,"module":23925,"summary":23934},"\u002Fdeep-learning\u002Fgenerative-models\u002Fvariational-autoencoders","Variational Autoencoders","An autoencoder compresses, but its latent space has gaps: sample a point between two encodings and the decoder produces noise. The variational autoencoder fixes this by training a probabilistic encoder against a prior, so the latent space becomes a smooth, samplable density. We derive the evidence lower bound it maximizes, the reparameterization trick that lets gradients flow through a random sample, and the closed-form Gaussian regularizer that pulls the posterior toward the prior.\n",{"path":23936,"title":23937,"module":23925,"summary":23938},"\u002Fdeep-learning\u002Fgenerative-models\u002Fgenerative-adversarial-networks","Generative Adversarial Networks","A generative adversarial network trains two networks against each other: a generator that turns noise into samples, and a discriminator that tries to tell real data from forgeries. The game has a clean theory: the optimal discriminator is a likelihood ratio, and at equilibrium the generator minimizes the Jensen–Shannon divergence to the data, with a global optimum exactly when its distribution matches the data. We derive that result, fix the saturating loss that breaks training, and catalogue the failure modes (mode collapse, instability, vanishing gradients) and the architectural fixes.\n",{"path":23940,"title":23941,"module":23925,"summary":23942},"\u002Fdeep-learning\u002Fgenerative-models\u002Fautoregressive-and-normalizing-flows","Autoregressive Models & Normalizing Flows","Two families that provide exact likelihoods, each at a cost. Autoregressive models factor the joint by the probability chain rule and learn each conditional with a masked network: exact $\\log p(x)$, but sampling proceeds one coordinate at a time. Normalizing flows push a simple base density through an invertible map and read $\\log p(x)$ off the change-of-variables formula, trading architectural freedom for a cheap Jacobian determinant via triangular coupling layers.\n",{"path":23944,"title":23945,"module":23925,"summary":23946},"\u002Fdeep-learning\u002Fgenerative-models\u002Fenergy-based-and-boltzmann-machines","Energy-Based & Boltzmann Machines","Energy-based models replace an explicit density with a scalar energy and a Boltzmann normalization, $p(x) = e^{-E(x)}\u002FZ$: simple to specify, but with an intractable partition function $Z$. The Boltzmann machine and its restricted variant make the energy bilinear so the hidden units factorize, and contrastive divergence sidesteps $Z$ by replacing the model expectation with a few Gibbs steps started at the data. We close on the undirected deep models (DBNs and DBMs) and how they differ from the directed VAE.\n",{"path":23948,"title":23949,"module":23925,"summary":23950},"\u002Fdeep-learning\u002Fgenerative-models\u002Fdiffusion-and-score-based-models","Diffusion and Score-Based Models","Corrupt a data point with Gaussian noise in small steps until only noise remains, then train a network to undo one step at a time. We derive the forward process and its closed-form marginal, reduce the variational bound to the single noise-prediction objective that makes diffusion trainable, and show the score-matching view that unifies it with Langevin sampling and the continuous SDE. The lesson closes with DDIM fast sampling, classifier-free guidance, and the latent diffusion that powers modern text-to-image systems.\n",{"path":23952,"title":23953,"module":23954,"summary":23955},"\u002Fdeep-learning\u002Fprobabilistic-methods\u002Fstructured-probabilistic-models","Structured Probabilistic Models","Probabilistic Methods","A joint distribution over $n$ variables is a table with exponentially many entries; nobody can store it, fit it, or sample from it directly. Structure fixes this: a graph whose missing edges encode conditional independencies that factor the joint into small local pieces. We build the two dialects, directed (Bayesian networks) and undirected (Markov random fields), read independence off the graph, and connect the machinery to the latent-variable and energy-based models that power deep generative learning.\n",{"path":23957,"title":23958,"module":23954,"summary":23959},"\u002Fdeep-learning\u002Fprobabilistic-methods\u002Fmonte-carlo-and-mcmc","Monte Carlo & MCMC","Most quantities of interest in a probabilistic model are integrals nobody can compute in closed form: expectations, marginals, partition functions. Monte Carlo replaces the integral with an average over samples; importance sampling reweights samples from a tractable proposal; and when even sampling the target is hard, Markov-chain Monte Carlo builds a chain whose stationary distribution _is_ the target. We derive Metropolis–Hastings and Gibbs, analyze mixing, and close on the partition-function gradient that powers energy-based learning.\n",{"path":23961,"title":23962,"module":23954,"summary":23963},"\u002Fdeep-learning\u002Fprobabilistic-methods\u002Fapproximate-inference","Approximate Inference","In a latent-variable model the quantity we need, the posterior $p(h\\mid v)$ over hidden causes, is almost never computable, because its normalizer is an intractable sum over configurations. Approximate inference reframes the problem as optimization: maximize the evidence lower bound, a tractable functional whose gap to the true log-evidence equals a KL divergence. From that single bound fall expectation–maximization, mean-field variational inference, MAP, and the learned encoders behind variational autoencoders.\n",{"path":23965,"title":23966,"module":23967,"summary":23968},"\u002Fdeep-learning\u002Fpractical\u002Fpractical-methodology","Practical Methodology","Practical Deep Learning","Knowing the algorithms is half the job; the other half is a disciplined loop. Fix a goal and a metric, stand up an end-to-end baseline, then read the train\u002Fvalidation gap to decide whether the next move is more data or a bigger model. We detail that loop: choosing metrics under class imbalance, default baselines by data type, extrapolating the data a target needs, and guarding the data pipeline against the leaks and label bugs that corrupt every gradient. Hyperparameter tuning, debugging, and deployment continue in the sequel.\n",{"path":23970,"title":23971,"module":23967,"summary":23972},"\u002Fdeep-learning\u002Fpractical\u002Fhyperparameters-and-debugging","Hyperparameters & Debugging","The tuning half of the methodology loop. The learning rate is the one hyperparameter that dominates, so we tune it first, on a log scale, coarse to fine, and prefer random search to grid when only a few dials matter. Then an ordered debugging playbook — overfit one batch, check the loss at initialization against ln C, watch the gradient norm, gradient-check against centered finite differences — and, after launch, monitoring for train-test skew and distribution drift with confidence-based abstention.\n",{"path":23974,"title":23975,"module":23967,"summary":23976},"\u002Fdeep-learning\u002Fpractical\u002Frepresentation-learning","Representation Learning","A good representation makes a hard task easy by changing coordinates: it disentangles the factors of variation, spends its bits as a distributed code, and respects the low-dimensional manifold the data lives on. We make those three properties precise, recover the manifold hypothesis, and close on the first method that turned them into training practice — greedy layer-wise unsupervised pretraining — before the sequel picks up how the field learned to reuse those features.\n",{"path":23978,"title":23979,"module":23967,"summary":23980},"\u002Fdeep-learning\u002Fpractical\u002Ftransfer-learning","Transfer Learning","A representation learned once can be reused everywhere. We cover the main mechanisms of reuse: feature extraction versus fine-tuning, the generic-to-specific gradient of features that sets the freeze boundary, the learning-rate discipline that keeps borrowed weights from being erased, domain adaptation when only the input distribution shifts, and the modern arc from supervised transfer to self-supervised foundation models.\n",{"path":23982,"title":23983,"module":23967,"summary":23984},"\u002Fdeep-learning\u002Fpractical\u002Fapplications","Applications","We survey large-scale training (the hardware, the two axes of parallelism, mixed precision, and the compression tricks that shrink a model after it is trained), then specialize the same gradient loop to vision, language, speech, and recommendation. Each domain is a different prior bolted onto one optimizer: convolutional invariance for pixels, distributed word vectors for tokens, sequence transduction for audio, low-rank factorization for the user–item matrix.\n",{"path":23986,"title":23987,"module":23967,"summary":23988},"\u002Fdeep-learning\u002Fpractical\u002Fmodel-compression-and-distillation","Model Compression and Distillation","A trained network and a deployable one are rarely the same object. This lesson is the toolkit for closing that gap: knowledge distillation transfers a large teacher's soft, information-rich logits into a small student; pruning deletes the weights that contribute least; quantization swaps 32-bit floats for 8- or 4-bit integers; and low-rank factorization replaces a fat matrix with two thin ones. We derive each method, show what it costs in accuracy, and lay out which combinations win on which hardware.\n",{"path":23990,"title":23991,"module":23967,"summary":23992},"\u002Fdeep-learning\u002Fpractical\u002Fmeta-learning-and-few-shot","Meta-Learning and Few-Shot Learning","A deep network trained on one example per class overfits. Meta-learning targets this few-shot regime by training across a distribution of tasks so that a new task is learnable from a handful of examples. We formalize the $N$-way $K$-shot episode, then derive the two dominant families: metric methods that learn an embedding where distance classifies (Prototypical Networks), and optimization methods that learn an initialization a few gradient steps can adapt (MAML). We close on the link to transfer learning and to the in-context few-shot behavior of large language models.\n",{"path":23994,"title":23995,"module":23996,"summary":23997},"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Flarge-language-models","Large Language Models","Large Models & Agents","A large language model is a decoder-only Transformer trained on one objective, next-token prediction, then scaled until new behavior appears. This first part builds the object itself: the equivalence between next-token prediction and lossless compression, subword tokenization (BPE, WordPiece, Unigram, SentencePiece) worked on a real sentence, the four pretraining objectives and the attention masks that distinguish them, and the three model families (encoder-only, decoder-only, encoder--decoder) with their parameter budgets. Scaling, decoding, the KV cache, and alignment continue in part two.\n",{"path":23999,"title":24000,"module":23996,"summary":24001},"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fscaling-inference-and-alignment","Scaling, Inference, and Alignment of Language Models","Once a language model is built, three questions remain: how does it improve as it grows, how is it decoded and served affordably, and how is a raw next-token predictor turned into an assistant. We derive the Kaplan power laws and the Chinchilla compute-optimal balance, trace emergent abilities and in-context learning, catalog the decoding strategies from greedy to nucleus sampling, work the KV cache that makes generation quadratic instead of cubic, cover parameter-efficient adaptation by low-rank updates (LoRA), and close on the alignment stack: instruction tuning, RLHF, and DPO.\n",{"path":24003,"title":24004,"module":23996,"summary":24005},"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fseq2seq-pretraining-and-bart","Denoising Sequence-to-Sequence Pretraining: BART","BERT corrupts and reconstructs; GPT predicts the next token. Sequence-to-sequence pretraining unifies both by training a full encoder–decoder as a denoising autoencoder: corrupt the text with a noise function, then reconstruct the original through a bidirectional encoder and an autoregressive decoder. This first part derives the denoising objective, catalogs BART's five noise functions (with a worked Poisson-infilling budget), proves BART specializes to both BERT and GPT, and traces a dimension-annotated forward pass through its encoder--decoder. T5, PEGASUS, fine-tuning, and decoding continue in part two.\n",{"path":24007,"title":24008,"module":23996,"summary":24009},"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Ftext-to-text-transfer-and-conditional-generation","Text-to-Text Transfer and Conditional Generation","BART reconstructs a corrupted document; T5 pushes the same denoising idea into a single interface where every task is a string-to-string map. This second part covers T5's span corruption with sentinel tokens (with a worked token budget), PEGASUS's summarization-matched gap sentences and the MASS midpoint, supervised fine-tuning and beam-search decoding with a length penalty, the exposure-bias failure modes of autoregressive decoding, and a theorem showing why a bidirectional encoder--decoder strictly dominates a decoder-only model when the output is conditioned on a full input.\n",{"path":24011,"title":24012,"module":23996,"summary":24013},"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fspeech-and-audio-models","Speech Recognition: Front-Ends and Alignment","Speech is a long, high-rate sequence whose label is short and unaligned, so the whole subject turns on bridging that mismatch. This first part builds the spectral front-ends that compress a waveform into frames (STFT, mel spectrogram, MFCC, with a worked frame-count), derives CTC's marginalization over alignments and its forward-backward recursion with a two-frame numeric example, and contrasts it with attention-based seq2seq (LAS) and the RNN transducer. Self-supervised and weakly-supervised models, and text-to-speech, continue in part two.\n",{"path":24015,"title":24016,"module":23996,"summary":24017},"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fself-supervised-speech-and-synthesis","Self-Supervised Speech Models and Synthesis","The recognition front-ends and alignment losses of part one all need transcribed audio, which is scarce. This second part removes that dependence: wav2vec 2.0 learns speech representations from unlabeled audio by a masked contrastive objective, HuBERT swaps the contrast for masked prediction of clustered units, and Whisper trades curation for scale with weakly-supervised web audio and a multitask token interface. We close with text-to-speech (the same length mismatch run backwards) and a tour of speech foundation models, discrete audio codecs, and neural TTS.\n",{"path":24019,"title":24020,"module":23996,"summary":24021},"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fai-agents","AI Agents: Tools and Reasoning","A language model that only emits text is a function from prompt to prompt; an agent closes the loop, letting that model act on an environment, read back the result, and decide again. This first part formalizes the agent as a policy over interaction histories, builds out tool calling and the executor trust boundary, the ReAct interleaving of reasoning and action (with concrete traces), and search over thoughts: chain-of-thought, self-consistency, least-to-most, and Tree of Thoughts. Memory, retrieval, reflection, and multi-agent orchestration continue in part two.\n",{"path":24023,"title":24024,"module":23996,"summary":24025},"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fagent-memory-retrieval-and-orchestration","Agent Memory, Retrieval, and Orchestration","An agent's reasoning and tool use only matter if it can remember what it learned and coordinate work larger than one context window. This second part builds the systems around the loop: short-term scratchpad versus long-term vector store, retrieval-augmented generation with a worked softmax over passage scores, reflection (Reflexion, Self-Refine), and multi-agent orchestration. It closes on the failure modes that bound agents — invalid tool calls, horizon-error compounding, context overflow, non-terminating loops — and the benchmarks that score the full loop.\n",{"path":24027,"title":24028,"module":23996,"summary":24029},"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fmixture-of-experts","Mixture-of-Experts","A mixture-of-experts layer replaces one feed-forward network with many and a router that sends each token to only a few of them, so the parameter count and the per-token compute become separate dials. We derive the gated output, sparse top-$k$ routing softmax, the load-balancing loss that stops the router from collapsing onto a single expert, and expert\u002Ftoken capacity with dropping, then work the dimension-annotated tensor shapes and FLOP arithmetic. We trace the architectures from the sparsely-gated LSTM through GShard, Switch Transformer, and Mixtral, cover distributed expert parallelism, and close on the training dynamics, failure modes, and serving costs of a sparse model.\n",{"path":24031,"title":24032,"module":23996,"summary":24033},"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Fmultimodal-models","Multimodal Contrastive Learning","A multimodal model places images, text, and audio in one representation space, so a picture and its caption land close together. This first part builds the contrastive route: the shared embedding space and its residual modality gap, the Vision Transformer image encoder (patch embedding, CLS token, position embeddings, with shapes), the symmetric InfoNCE loss that trains the CLIP dual encoder from a batch similarity matrix (with a worked numeric step), and zero-shot classification as a softmax over class-prompt embeddings. Fusion and vision-language models continue in part two.\n",{"path":24035,"title":24036,"module":23996,"summary":24037},"\u002Fdeep-learning\u002Flarge-models-and-agents\u002Ffusion-and-vision-language-models","Fusion and Vision-Language Models","A contrastive model compares modalities but never lets one read another. This second part builds the fusion route: early, late, and cross-attention fusion, then the three designs that connect a frozen vision encoder to a frozen language model — Flamingo's zero-initialized gated cross-attention, BLIP-2's Q-Former, and LLaVA's linear projector. We work the token-budget arithmetic that separates them, name the object-hallucination and fine-detail failure modes, cover the contrastive-then- instruction-tune recipe and its retrieval\u002Fcaptioning\u002FVQA benchmarks, and close on natively multimodal models.\n",{"path":24039,"title":24040,"module":24041,"summary":24042},"\u002Fdeep-learning\u002Freinforcement-learning\u002Ffoundations-of-reinforcement-learning","Foundations of Reinforcement Learning","Reinforcement Learning","Reinforcement learning is the third paradigm: an agent learns to act by interacting with an environment that returns rewards, not labels. We formalize the interaction as a Markov decision process, define the value functions that rank states and actions, and derive the Bellman expectation and optimality equations that every method downstream solves. Dynamic programming gives the exact answer when the model is known, and its convergence rests on a single fact: the Bellman operator is a contraction.\n",{"path":24044,"title":24045,"module":24041,"summary":24046},"\u002Fdeep-learning\u002Freinforcement-learning\u002Fmodel-free-prediction-and-control","Model-Free Prediction and Control","When the dynamics are unknown, an agent cannot plan against a model; it must learn directly from sampled experience. We build prediction and control from two estimators of the same return: Monte Carlo averages whole episodes, while temporal-difference learning bootstraps from its own next estimate. We trace the bias-variance contrast between them, derive SARSA and Q-learning as the on-policy and off-policy forms of control, unify everything through n-step returns and eligibility traces, and close on the deadly triad that makes off-policy bootstrapping with function approximation diverge.\n",{"path":24048,"title":24049,"module":24041,"summary":24050},"\u002Fdeep-learning\u002Freinforcement-learning\u002Fdeep-q-networks","Deep Q-Networks","A Deep Q-Network replaces the tabular action-value function with a neural approximator $Q(s,a;\\theta)$ and trains it by regression toward a bootstrapped target. Naive online Q-learning with a network diverges, so DQN adds two stabilizers: an experience-replay buffer that decorrelates samples, and a periodically-frozen target network that holds the regression target still. We derive the loss, give the full algorithm and the Atari pipeline, and then layer on Double DQN, the dueling split, prioritized replay, and the Rainbow combination.\n",{"path":24052,"title":24053,"module":24041,"summary":24054},"\u002Fdeep-learning\u002Freinforcement-learning\u002Fpolicy-gradients-and-actor-critic","Policy Gradients and Actor-Critic Methods","Value-based reinforcement learning learns what each state is worth and acts greedily; policy-gradient methods skip the detour and optimize a parameterized policy directly by ascending the gradient of expected return. The policy gradient theorem makes this tractable through the log-derivative trick, turning an intractable gradient of an expectation into an expectation of a gradient. REINFORCE realizes the idea but suffers high variance; baselines, the advantage function, and actor-critic learning reduce it, and trust-region methods (TRPO, PPO) keep each update from destroying the policy it just learned.\n",{"path":24056,"title":24057,"module":24041,"summary":24058},"\u002Fdeep-learning\u002Freinforcement-learning\u002Frl-from-human-feedback","Reinforcement Learning from Human Feedback","Many objectives we want from a model, that it be helpful and harmless, are hard to write down but easy to judge by comparison. RLHF turns that asymmetry into a training signal: fit a reward model to pairwise human preferences under the Bradley-Terry likelihood, then fine-tune the policy to maximize that reward under a KL penalty toward a reference. We derive the reward loss, the KL-regularized RL objective and its closed-form optimum, then show how DPO inverts that optimum to collapse the whole pipeline into one supervised log-sigmoid loss, and survey IPO, KTO, RLAIF, and GRPO.\n",{"path":24060,"title":20032,"module":6,"summary":6},"\u002Fdeep-learning",{"path":24062,"title":24063,"module":21979,"summary":24064},"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Fequilibrium-state-variables-zeroth-law","Equilibrium, State Variables, and the Zeroth Law","Thermodynamics describes a many-body system by a handful of macroscopic variables and the equilibrium relations among them. This lesson fixes the vocabulary: systems and the walls that separate them, state variables versus path-dependent process quantities, quasi-static and reversible idealizations, and the zeroth law, whose transitivity of thermal equilibrium is what lets temperature exist as a number. The ideal-gas thermometer turns that number into a scale, and an equation of state ties the variables into a surface.\n",{"path":24066,"title":24067,"module":21979,"summary":24068},"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Ffirst-law-heat-and-work","The First Law: Internal Energy, Heat, and Work","The first law is energy conservation for a system that exchanges energy as both heat and work. Internal energy is a state function with an exact differential; heat and work are path-dependent process quantities. This lesson states $\\d U=\\delta Q+\\delta W$, computes compression work as an area on the $P$–$V$ plane, defines the heat capacities $C_V$ and $C_P$ and the enthalpy that makes $C_P$ natural, and works the isothermal and adiabatic processes of an ideal gas, including the adiabat $PV^\\gamma=\\text{const}$.\n",{"path":24070,"title":24071,"module":21979,"summary":24072},"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Fsecond-law-entropy-and-the-carnot-bound","The Second Law, Carnot Cycles, and Entropy","The second law forbids the free conversion of heat into work. This lesson states the Kelvin and Clausius forms, proves them equivalent, and analyzes the Carnot cycle to get the efficiency bound $1-T_c\u002FT_h$. Carnot's theorem makes that bound universal and defines the thermodynamic temperature scale. The Clausius inequality $\\oint \\delta Q\u002FT\\le 0$ then constructs entropy as a state function, $\\d S=\\delta Q_{\\rm rev}\u002FT$, whose non-decrease in isolated systems is the arrow of time.\n",{"path":24074,"title":24075,"module":21979,"summary":24076},"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Fthermodynamic-potentials-and-maxwell-relations","Thermodynamic Potentials and Maxwell Relations","The fundamental relation $\\d U=T\\,\\d S-P\\,\\d V+\\mu\\,\\d N$ packages the first and second laws into one exact differential. Legendre transforms swap each conjugate pair to produce the Helmholtz, enthalpy, Gibbs, and grand potentials, each minimized under its own natural variables. Equality of mixed second partials of these potentials gives the Maxwell relations, which convert unmeasurable entropy derivatives into measurable ones from the equation of state.\n",{"path":24078,"title":24079,"module":21979,"summary":24080},"\u002Fstatistical-mechanics\u002Fthermodynamics\u002Fstability-response-functions-and-the-third-law","Response Functions, Stability, and the Third Law","Response functions — heat capacities, compressibilities, thermal expansion — are the second derivatives of the potentials and the quantities an experiment actually measures. This lesson derives the general relation $C_P-C_V=TV\\alpha^2\u002F\\kappa_T$, shows that convexity of the potentials forces the stability conditions $C_V>0$ and $\\kappa_T>0$, and states the third law: entropy approaches a constant as $T\\to0$, so heat capacities and expansion coefficients vanish there and absolute zero is unattainable.\n",{"path":24082,"title":24083,"module":24084,"summary":24085},"\u002Fstatistical-mechanics\u002Ffoundations\u002Fclassical-statistics-and-equipartition","Classical Statistics and Equipartition","Microstates, Phase Space, and Statistical Entropy","A liter of gas holds on the order of a trillion trillion molecules, far too many to track by their equations of motion. Classical statistical mechanics replaces the trajectories with a single probability law, the Boltzmann distribution, and reads the measurable properties of matter off it: the Maxwell speed distribution, the average energy per degree of freedom, and the heat capacities of gases and solids — together with the low-temperature failures that forced the quantum revision.\n",{"path":24087,"title":24088,"module":24084,"summary":24089},"\u002Fstatistical-mechanics\u002Ffoundations\u002Fphase-space-and-liouvilles-theorem","Phase Space, Trajectories, and Liouville's Theorem","A classical system of N particles is one point in a 6N-dimensional phase space, and its evolution is a single trajectory driven by Hamilton's equations. This lesson builds that geometric picture, introduces the phase-space density of an ensemble, and proves Liouville's theorem: the density is carried by the flow as an incompressible fluid, so phase-space volume is conserved. The stationary densities of equilibrium follow as functions of the conserved quantities alone.\n",{"path":24091,"title":24092,"module":24084,"summary":24093},"\u002Fstatistical-mechanics\u002Ffoundations\u002Fensembles-and-the-equal-probability-postulate","Ensembles and the Postulate of Equal a Priori Probabilities","An ensemble is a probability distribution over the microstates of a system. This lesson states the single postulate on which equilibrium statistical mechanics rests — that an isolated system in equilibrium is equally likely to be in any of its accessible microstates — and works out its consequences: the accessible phase-space volume, the overwhelming dominance of the most probable macrostate as the particle number grows, and the ergodic hypothesis that lets a time average be replaced by an ensemble average.\n",{"path":24095,"title":24096,"module":24084,"summary":24097},"\u002Fstatistical-mechanics\u002Ffoundations\u002Fstatistical-entropy-boltzmann-and-gibbs","Statistical Entropy: Boltzmann and Gibbs","Entropy is the logarithm of the number of accessible microstates. This lesson builds the two statistical entropies — Boltzmann's S = k ln Omega for an isolated system and Gibbs's S = -k sum p ln p for any ensemble — proves they agree for a uniform distribution, and connects both to Shannon's measure of missing information. The second law emerges as the drift toward maximum multiplicity, and maximizing the Gibbs entropy under constraints previews the canonical distribution.\n",{"path":24099,"title":24100,"module":24101,"summary":24102},"\u002Fstatistical-mechanics\u002Fmicrocanonical\u002Fmicrocanonical-ensemble-and-entropy","The Microcanonical Ensemble and Statistical Entropy","The Microcanonical Ensemble","An isolated system holds its energy, volume, and particle number fixed, and the fundamental postulate assigns equal probability to every microstate on its energy shell. This lesson builds the microcanonical distribution, defines the enclosed phase-space volume $\\Gamma(E)$, the surface density of states $\\omega(E)=\\d\\Gamma\u002F\\d E$, and the shell count $\\Omega(E)$, shows their logarithms agree to $O(\\ln N)$ for large $N$, and reads the Boltzmann entropy $S=k\\ln\\Omega$ off the count. The measure factors $h^{3N}$ and $N!$ enter here and make $S$ extensive.\n",{"path":24104,"title":24105,"module":24101,"summary":24106},"\u002Fstatistical-mechanics\u002Fmicrocanonical\u002Fequilibrium-conditions-temperature-pressure-chemical-potential","Thermal, Mechanical, and Diffusive Equilibrium","Two isolated subsystems that can exchange energy, volume, or particles reach equilibrium at the partition that maximizes their combined entropy. Setting the derivative of the total entropy to zero identifies the statistical definitions $1\u002FT=(\\partial S\u002F\\partial E)$, $P\u002FT=(\\partial S\u002F\\partial V)$, and $-\\mu\u002FT=(\\partial S\u002F\\partial N)$, shows heat flows from hot to cold as an entropy increase, and recovers the fundamental relation $\\d S=(\\d E+P\\,\\d V-\\mu\\,\\d N)\u002FT$ from pure counting.\n",{"path":24108,"title":24109,"module":24101,"summary":24110},"\u002Fstatistical-mechanics\u002Fmicrocanonical\u002Fideal-gas-phase-space-and-the-sackur-tetrode-entropy","The Ideal Gas, Phase-Space Volume, and the Sackur–Tetrode Entropy","The monatomic ideal gas is the first system whose microcanonical count can be done in closed form. The momentum integral is the volume of a $3N$-dimensional ball of radius $\\sqrt{2mE}$, the configuration integral is $V^N$, and together they give the Sackur–Tetrode entropy $S=Nk[\\ln(V\u002FN\\lambda^3)+5\u002F2]$ with the thermal wavelength $\\lambda=h\u002F\\sqrt{2\\pi mkT}$. The formula matches the measured entropy of helium, fixes the classical regime $n\\ll n_Q$, and shows why the $N!$ is needed for extensivity.\n",{"path":24112,"title":24113,"module":24101,"summary":24114},"\u002Fstatistical-mechanics\u002Fmicrocanonical\u002Ftwo-state-systems-paramagnets-and-negative-temperature","Two-State Systems, Paramagnets, and Negative Temperature","The ideal two-state paramagnet has a multiplicity counted by the binomial coefficient, an entropy that is an inverted dome in the energy, and a temperature read from the slope $1\u002FT=\\partial S\u002F\\partial E$. Because the energy is bounded above, the slope changes sign past the entropy maximum: a population-inverted spin system has a negative absolute temperature, which is hotter than any positive temperature. Nuclear-spin experiments and lasers realize the inverted state.\n",{"path":24116,"title":24117,"module":24118,"summary":24119},"\u002Fstatistical-mechanics\u002Fcanonical\u002Fcanonical-ensemble-and-the-boltzmann-distribution","The Canonical Ensemble and the Boltzmann Distribution","The Canonical Ensemble","A system held at fixed temperature by contact with a heat reservoir is described by the canonical ensemble. Expanding the reservoir entropy to first order in the system energy gives the Boltzmann distribution $p_i\\propto e^{-\\beta E_i}$, and the same law follows from maximizing the Gibbs entropy at fixed mean energy. Both routes identify $\\beta=1\u002Fk_BT$ and fix the probability of every microstate from the temperature alone.\n",{"path":24121,"title":24122,"module":24118,"summary":24123},"\u002Fstatistical-mechanics\u002Fcanonical\u002Fpartition-function-and-the-helmholtz-free-energy","The Partition Function and the Helmholtz Free Energy","The normalizing sum of the Boltzmann distribution, the partition function $Z=\\sum_i e^{-\\beta E_i}$, is a generating function for the thermodynamics. The mean energy is $-\\partial\\ln Z\u002F\\partial\\beta$, and the Gibbs entropy of the canonical distribution collapses to the bridge relation $F=-k_BT\\ln Z$. From $F$ every thermodynamic quantity follows by differentiation, and $Z$ factorizes over independent degrees of freedom.\n",{"path":24125,"title":24126,"module":24118,"summary":24127},"\u002Fstatistical-mechanics\u002Fcanonical\u002Fenergy-fluctuations-and-ensemble-equivalence","Energy Fluctuations and the Equivalence of Ensembles","In the canonical ensemble the energy fluctuates, and the second derivative of $\\ln Z$ gives its variance. The fluctuation–response identity $\\langle\\Delta E^2\\rangle = k_BT^2C_V$ ties the spread of the energy to the heat capacity, and the relative fluctuation falls as $1\u002F\\sqrt{N}$. In the thermodynamic limit the canonical energy distribution is a sharp spike, and the canonical and microcanonical ensembles predict the same thermodynamics.\n",{"path":24129,"title":24130,"module":24118,"summary":24131},"\u002Fstatistical-mechanics\u002Fcanonical\u002Fthe-einstein-solid-and-harmonic-systems","Harmonic Systems: The Einstein Solid and Vibrational Heat Capacity","A quantum harmonic oscillator has a geometric partition function summed in closed form, giving a mean energy $\\hbar\\omega(\\tfrac12+\\langle n\\rangle)$ with the Bose occupation factor. Modeling a solid as $3N$ independent oscillators yields a heat capacity that rises from zero and saturates at the Dulong–Petit value $3Nk_B$. The Einstein temperature sets the crossover, and the model's exponential low-temperature falloff, too steep against the observed $T^3$, motivates the Debye theory.\n",{"path":24133,"title":24134,"module":24118,"summary":24135},"\u002Fstatistical-mechanics\u002Fcanonical\u002Fparamagnetism-and-the-schottky-anomaly","Paramagnetism, Two-Level Systems, and the Schottky Anomaly","A magnetic moment in a field is a two-level system whose partition function is a hyperbolic cosine. The magnetization of a spin-$\\tfrac12$ paramagnet is $N\\mu\\tanh(\\mu B\u002Fk_BT)$, generalizing to the Brillouin function for spin $J$; it gives Curie's law $\\chi\\propto 1\u002FT$ at high temperature and saturates at low temperature. A finite level gap produces the Schottky heat-capacity peak, and the temperature dependence of the entropy on the field is the basis of adiabatic demagnetization cooling.\n",{"path":24137,"title":24138,"module":24139,"summary":24140},"\u002Fstatistical-mechanics\u002Fclassical-gas\u002Fideal-gas-partition-function-and-the-gibbs-paradox","The Ideal Gas Partition Function and the Gibbs Paradox","The Classical Ideal Gas","The classical monatomic ideal gas built from the partition function. The single-particle sum is $z_1=V\u002F\\lambda^3$ with the thermal de Broglie wavelength $\\lambda$; the $N$-particle partition function is $z_1^N\u002FN!$, and the $N!$ is forced by indistinguishability. From $Z$ the ideal-gas law, $U=\\tfrac32 Nk_BT$, and the Sackur–Tetrode entropy follow. The $N!$ makes the entropy extensive and resolves the Gibbs paradox: mixing identical gases produces no entropy change.\n",{"path":24142,"title":24143,"module":24139,"summary":24144},"\u002Fstatistical-mechanics\u002Fclassical-gas\u002Fequipartition-and-the-virial-theorem","Equipartition and the Virial Theorem","The equipartition theorem derived from the canonical ensemble: every phase-space coordinate that enters the Hamiltonian quadratically carries a mean energy $\\tfrac12 k_BT$. The generalized form $\\langle x_i\\,\\partial H\u002F\\partial x_j\\rangle = k_BT\\,\\delta_{ij}$ contains equipartition and the classical virial theorem as special cases. Equipartition fixes the classical heat capacities, fails by quantum freeze-out when a level gap exceeds $k_BT$, and shifts for a relativistic gas whose energy is linear rather than quadratic in momentum.\n",{"path":24146,"title":24147,"module":24139,"summary":24148},"\u002Fstatistical-mechanics\u002Fclassical-gas\u002Fmolecular-gases-rotation-and-vibration","Molecular Gases: Rotational and Vibrational Degrees of Freedom","The internal partition function of a diatomic gas factorizes into translational, rotational, vibrational, and electronic parts. The rigid rotor gives a rotational temperature $\\theta_{\\rm rot}$; the harmonic bond gives a vibrational temperature $\\theta_{\\rm vib}$. Each mode contributes to the heat capacity only above its characteristic temperature, producing the diatomic $C_V$ staircase from $\\tfrac32 R$ to $\\tfrac52 R$ to $\\tfrac72 R$. Homonuclear molecules carry a symmetry number, and hydrogen splits into ortho and para species.\n",{"path":24150,"title":24151,"module":24152,"summary":24153},"\u002Fstatistical-mechanics\u002Fgrand-canonical\u002Fgrand-canonical-ensemble-and-the-grand-partition-function","The Grand Canonical Ensemble","Grand Canonical Ensemble","When a system exchanges both energy and particles with a reservoir, the reservoir fixes its temperature and its chemical potential. Expanding the reservoir entropy to first order in the exchanged energy and particle number gives the Gibbs factor $e^{-\\beta(E-\\mu N)}$, and summing it over every microstate of every particle number gives the grand partition function $\\Xi$. The grand potential $\\Phi = -k_BT\\ln\\Xi = -PV$ generates the mean particle number, energy, entropy, and pressure by differentiation.\n",{"path":24155,"title":24156,"module":24152,"summary":24157},"\u002Fstatistical-mechanics\u002Fgrand-canonical\u002Fchemical-potential-fugacity-and-number-fluctuations","Chemical Potential, Fugacity, and Number Fluctuations","The chemical potential is the energy to add one particle at fixed entropy and volume, equal to the slope of the free energy in the particle number. For the classical ideal gas $\\mu=k_BT\\ln(n\\lambda^3)$ is large and negative, and the fugacity $z=n\\lambda^3$ is small. The grand ensemble makes the particle number fluctuate; its variance $\\langle\\Delta N^2\\rangle=k_BT(\\partial N\u002F\\partial\\mu)$ equals $k_BT\\,N^2\\kappa_T\u002FV$, tying density fluctuations to the isothermal compressibility. Equality of $\\mu$ is the condition for diffusive equilibrium and phase coexistence.\n",{"path":24159,"title":24160,"module":24152,"summary":24161},"\u002Fstatistical-mechanics\u002Fgrand-canonical\u002Fensemble-summary-and-the-thermodynamic-web","The Three Ensembles and the Thermodynamic Web","The microcanonical, canonical, and grand canonical ensembles hold different variables fixed and generate different potentials — the entropy $S$, the Helmholtz free energy $F$, and the grand potential $\\Phi$ — linked by Legendre transforms that trade each fixed variable for its conjugate. Each successive ensemble lets one more quantity fluctuate. In the thermodynamic limit the three agree, the relative fluctuations vanishing as $1\u002F\\sqrt{N}$; the ideal gas gives the same equation of state in all three. The choice of ensemble is a matter of convenience, set by which sum is easiest.\n",{"path":24163,"title":24164,"module":24165,"summary":24166},"\u002Fstatistical-mechanics\u002Fquantum-statistics\u002Fquantum-statistics-bose-einstein-and-fermi-dirac","Quantum Statistics — Bose-Einstein and Fermi-Dirac","Quantum Statistics","Quantum particles of the same kind are genuinely indistinguishable: no label survives an overlap of their wave functions. Counting states with that constraint replaces the Boltzmann distribution with two quantum laws — the Bose-Einstein distribution for integer-spin particles, which clump into shared states, and the Fermi-Dirac distribution for half-integer-spin particles, which exclude one another. Both reduce to Boltzmann in the dilute, hot limit, and a de Broglie criterion says exactly when.\n",{"path":24168,"title":24169,"module":24165,"summary":24170},"\u002Fstatistical-mechanics\u002Fquantum-statistics\u002Fderiving-the-quantum-distributions","Deriving the Quantum Distributions from the Grand Ensemble","The Bose-Einstein and Fermi-Dirac distributions follow from one observation: in the occupation-number representation the single-particle modes are independent, so the grand partition function factorizes into one factor per mode. A boson mode sums a geometric series over all occupancies; a fermion mode sums two terms. Differentiating each factor gives the mean occupation $1\u002F(e^{\\beta(\\varepsilon-\\mu)}\\mp 1)$, the Maxwell-Boltzmann limit when occupancies are small, and the occupation fluctuations that distinguish bunching from anti-bunching.\n",{"path":24172,"title":24173,"module":24165,"summary":24174},"\u002Fstatistical-mechanics\u002Fquantum-statistics\u002Fthe-classical-limit-and-quantum-concentration","The Classical Limit and Quantum Concentration","When every single-particle level is nearly empty, both quantum distributions collapse to the Maxwell-Boltzmann form, and the fugacity equals the ratio of the number density to the quantum concentration $n_Q = 1\u002F\\lambda^3$. The gas is classical when $n \\ll n_Q$, degenerate when $n \\gtrsim n_Q$. The chemical potential is large and negative in the classical regime and rises through zero as the gas degenerates. The leading quantum correction to the ideal-gas law is a second virial term that lowers the pressure for bosons and raises it for fermions — a statistical attraction and repulsion with no interaction behind it.\n",{"path":24176,"title":24177,"module":24165,"summary":24178},"\u002Fstatistical-mechanics\u002Fquantum-statistics\u002Fideal-quantum-gases-general-framework","Ideal Quantum Gases: The General Framework","Every ideal quantum gas is handled by one calculation. The sum over single-particle modes becomes an energy integral weighted by a density of states $g(\\varepsilon)\\propto\\varepsilon^{1\u002F2}$, and the number and pressure reduce to the Bose and Fermi functions $g_\\nu(z)$ and $f_\\nu(z)$ of the fugacity. An integration by parts fixes $PV=\\tfrac23 U$ for a nonrelativistic gas and $PV=\\tfrac13 U$ for an ultrarelativistic one, independent of statistics. Specializing the density of states and the chemical potential then produces the photon gas, phonons, the Bose gas, and the Fermi gas as four branches of the same framework.\n",{"path":24180,"title":24181,"module":24182,"summary":24183},"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fbose-einstein-condensation-and-the-fermion-gas","Bose-Einstein Condensation and the Fermion Gas","Bosonic Systems","Below a critical temperature a boson gas drops a macroscopic fraction of its particles into the single ground state — Bose-Einstein condensation, the mechanism behind superfluid helium and the dilute-atom condensates cooled to nanokelvin. The same statistics applied to a photon gas reproduces Planck's blackbody spectrum. Fermions do the opposite: forbidden from sharing states, they fill every level up to the Fermi energy, and that filled sea governs the electrons in metals and the pressure that holds up a white dwarf.\n",{"path":24185,"title":24186,"module":24182,"summary":24187},"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fthe-photon-gas-and-plancks-radiation-law","The Photon Gas and Planck's Radiation Law","Electromagnetic radiation in equilibrium with cavity walls is a gas of non-conserved bosons, and non-conservation forces the chemical potential to zero. Counting standing-wave modes with two polarizations and weighting each by the Bose occupation gives the Planck spectral energy density. Its low-frequency tail reproduces the classical Rayleigh-Jeans law and the ultraviolet catastrophe; the Bose factor cuts the divergence off at high frequency and the peak obeys Wien's displacement law.\n",{"path":24189,"title":24190,"module":24182,"summary":24191},"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fblackbody-thermodynamics-and-radiation-pressure","Blackbody Thermodynamics and Radiation Pressure","Integrating the Planck spectrum over all frequencies gives the total energy density proportional to the fourth power of temperature — the Stefan-Boltzmann law — and the isotropy of a relativistic gas fixes the radiation pressure at one third of the energy density. From the free energy follow the entropy and heat capacity, both proportional to T cubed, and the adiabatic law for radiation. The results govern the pressure inside stars and the cooling of the cosmic microwave background as the universe expands.\n",{"path":24193,"title":24194,"module":24182,"summary":24195},"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fphonons-and-the-debye-model","Phonons and the Debye Model","The vibrations of a crystal lattice are quantized into phonons — bosons of zero chemical potential, counted exactly like cavity photons but with three polarizations, a finite sound speed, and a total of 3N modes. The Debye model replaces the true dispersion by a linear one cut off at a frequency that enforces that count. It gives the correct low-temperature T-cubed heat capacity the Einstein model missed and recovers the Dulong-Petit value at high temperature.\n",{"path":24197,"title":24198,"module":24182,"summary":24199},"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fbose-einstein-condensation-derived","Bose-Einstein Condensation Derived","For a gas of conserved bosons the excited states can hold only a finite number of particles at fixed temperature, set by the Bose function at unit fugacity. When the total exceeds that ceiling the surplus collapses into the single ground state, which the continuum density-of-states integral misses and which must be restored by hand. This fixes the critical temperature, the condensate fraction, and the fact that a uniform gas condenses only in three or more dimensions.\n",{"path":24201,"title":24202,"module":24182,"summary":24203},"\u002Fstatistical-mechanics\u002Fbose-systems\u002Fthermodynamics-of-the-bose-gas-and-superfluidity","Thermodynamics of the Bose Gas and Superfluidity","The energy and pressure of the ideal Bose gas follow from the Bose function at the order above the density, and below the critical temperature the pressure depends on temperature alone because the condensate carries none. The heat capacity rises to a cusp at the transition. Real superfluid helium departs from the ideal gas because interactions matter: the Landau criterion ties frictionless flow to the phonon-roton excitation spectrum, and the two-fluid model carries a second sound.\n",{"path":24205,"title":24206,"module":24207,"summary":24208},"\u002Fstatistical-mechanics\u002Ffermi-gas\u002Fthe-ideal-fermi-gas-at-zero-temperature","The Ideal Fermi Gas at Zero Temperature","Degenerate Fermi Gas","At absolute zero a gas of non-interacting fermions fills every single-particle state up to the Fermi energy and leaves the rest empty, a filled Fermi sphere in momentum space. This lesson computes the Fermi momentum, energy, and temperature from the density, the density of states, the total ground-state energy, and the degeneracy pressure that grows as $n^{5\u002F3}$. Numerical Fermi energies for metals set the scale: they are electron-volts, so room temperature is deep in the degenerate regime.\n",{"path":24210,"title":24211,"module":24207,"summary":24212},"\u002Fstatistical-mechanics\u002Ffermi-gas\u002Fsommerfeld-expansion-and-electrons-in-metals","The Sommerfeld Expansion and Electrons in Metals","Turning on a small temperature blurs the Fermi step over a shell of width $k_BT$ around $\\epsilon_F$. The Sommerfeld expansion turns integrals over the Fermi function into a power series in $(k_BT\u002F\\epsilon_F)^2$, giving the shift of the chemical potential and a heat capacity linear in $T$. This resolves the old puzzle of the missing electronic heat capacity, predicts the combined $C=\\gamma T+AT^3$ of a metal, and gives the temperature-independent Pauli paramagnetism of the electron gas.\n",{"path":24214,"title":24215,"module":24207,"summary":24216},"\u002Fstatistical-mechanics\u002Ffermi-gas\u002Fwhite-dwarfs-and-the-chandrasekhar-limit","White Dwarfs and the Chandrasekhar Limit","A white dwarf is held up against its own gravity by the degeneracy pressure of its electrons. Balancing that pressure against gravity gives a mass-radius relation $R\\propto M^{-1\u002F3}$: heavier white dwarfs are smaller and denser. As the density rises the electrons turn relativistic, the pressure softens from $n^{5\u002F3}$ to $n^{4\u002F3}$, and the star can no longer support itself above a critical mass. This lesson derives that Chandrasekhar mass, about $1.4\\,M_\\odot$, and what lies beyond it.\n",{"path":24218,"title":24219,"module":24207,"summary":24220},"\u002Fstatistical-mechanics\u002Ffermi-gas\u002Fneutron-stars-and-nuclear-matter","Neutron Stars and Dense Matter","When a collapsing core passes nuclear density, electron capture converts the matter to neutrons and their degeneracy pressure takes over. The same balance that fixes a white dwarf, rescaled by the neutron mass, gives a neutron star of a few solar masses in a ten-kilometre radius. General relativity is no longer a correction: the Tolman-Oppenheimer-Volkoff equation replaces the Newtonian balance and sets a maximum mass around two solar masses. This lesson rescales the Fermi-gas argument, states where it breaks, and places the compact objects in one stability sequence.\n",{"path":24222,"title":24223,"module":24224,"summary":24225},"\u002Fstatistical-mechanics\u002Finteractions\u002Fthe-cluster-expansion-and-virial-coefficients","The Cluster Expansion and Virial Coefficients","Interacting Gases","A real gas departs from $PV=Nk_BT$ because its molecules interact. The configuration integral factors through the Mayer function $f_{ij}=e^{-\\beta u_{ij}}-1$, and expanding it in powers of density produces the virial expansion $PV\u002FNk_BT = 1 + B_2(T)n + B_3(T)n^2 + \\cdots$. The second virial coefficient $B_2(T)=-\\tfrac12\\int f\\,\\d^3r$ is a single integral over the pair potential; it is positive for a hard core, negative for an attractive well, and vanishes at the Boyle temperature where the two balance.\n",{"path":24227,"title":24228,"module":24224,"summary":24229},"\u002Fstatistical-mechanics\u002Finteractions\u002Fthe-van-der-waals-gas-and-liquid-gas-coexistence","The van der Waals Gas and Liquid-Gas Coexistence","Resumming the second virial coefficient $B_2=b-a\u002Fk_BT$ into an equation of state gives the van der Waals model $(P+a\u002Fv^2)(v-b)=k_BT$, the simplest theory of a fluid that condenses. Below the critical temperature its isotherms develop a mechanically unstable loop; the Maxwell equal-area construction replaces the loop with a coexistence tie line. The critical point sits at $v_c=3b$, $k_BT_c=8a\u002F27b$, $P_c=a\u002F27b^2$, and the model predicts universal but incorrect critical exponents because it ignores fluctuations.\n",{"path":24231,"title":24232,"module":24224,"summary":24233},"\u002Fstatistical-mechanics\u002Finteractions\u002Fquantum-gases-with-interactions-and-exchange","Quantum Gases with Interactions and Statistical Exchange","A quantum gas has a nonzero second virial coefficient even with no forces between the particles: symmetrization alone produces an effective statistical interaction, attractive for bosons and repulsive for fermions, with range the thermal wavelength $\\lambda$. This lesson derives that exchange contribution $B_2=\\mp\\lambda^3\u002F2^{5\u002F2}g$, writes it as a statistical potential $v_s(r)=-k_BT\\ln(1\\pm e^{-2\\pi r^2\u002F\\lambda^2})$, and shows how real interactions add on top through the Beth-Uhlenbeck phase-shift formula, reducing at low temperature to a single scattering length.\n",{"path":24235,"title":24236,"module":24237,"summary":24238},"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fphases-coexistence-and-classification","Phases, Coexistence, and the Classification of Transitions","Phase Transitions","A phase transition is a point where the free energy of a substance loses analyticity, so a small change in temperature or pressure produces a qualitative change of state. This lesson maps the coexistence curves of a pure substance, derives the Clausius-Clapeyron relation between the slope of a coexistence line and its latent heat, and separates first-order transitions (discontinuous entropy and density) from continuous ones (a vanishing order parameter and divergent response). The Ehrenfest scheme, the order parameter, and the triple and critical points fix the vocabulary the rest of the module builds on.\n",{"path":24240,"title":24241,"module":24237,"summary":24242},"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fthe-ising-model-and-exact-solutions","The Ising Model and Exact Results","The Ising model reduces cooperative ordering to spins on a lattice coupled to their neighbors, and the same Hamiltonian describes uniaxial magnets, the liquid-gas critical point through the lattice gas, and binary alloys. This lesson solves the one-dimensional chain exactly with the transfer matrix, shows by a domain-wall argument why one dimension has no ordered phase at any positive temperature, contrasts the survival of order in two dimensions, and quotes Onsager's exact two-dimensional results: the critical temperature, the logarithmically divergent heat capacity, and the magnetization exponent one eighth.\n",{"path":24244,"title":24245,"module":24237,"summary":24246},"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fmean-field-theory-and-the-weiss-model","Mean-Field Theory and Spontaneous Symmetry Breaking","Mean-field theory replaces the neighbors of each spin by their average, turning the interacting Ising model into a single spin in a self-consistent field. The resulting equation m = tanh(beta J z m + beta h) has only the zero solution above a critical temperature and gains a nonzero root below it, giving spontaneous magnetization and a mean-field critical temperature k T_c = J z. The Bragg-Williams free energy turns single-welled above T_c and double-welled below, the picture of spontaneous symmetry breaking. The approximation is exact in high dimension and fails below the upper critical dimension four, quantified by the Ginzburg criterion.\n",{"path":24248,"title":24249,"module":24237,"summary":24250},"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fcritical-exponents-and-landau-theory","Critical Exponents, Scaling, and Landau Theory","Near a continuous transition every singular quantity follows a power law in the reduced temperature, and the exponents alpha, beta, gamma, delta, nu, and eta encode the transition more sharply than T_c itself. Landau theory expands the free energy in the order parameter and delivers the mean-field exponents in a few lines. They disagree with experiment and with the exact two-dimensional Ising values, but the exponents are not independent: the scaling relations of Rushbrooke, Widom, Fisher, and Josephson tie them together, and the correlation length sets the length scale that organizes universality classes.\n",{"path":24252,"title":24253,"module":24237,"summary":24254},"\u002Fstatistical-mechanics\u002Fphase-transitions\u002Fthe-renormalization-group-idea","Scaling and the Renormalization-Group Idea","At a critical point fluctuations exist on every length scale, so the system looks the same after coarse-graining. The renormalization group makes this self-similarity a computation: group spins into blocks, integrate out the short scales, and track how the couplings change. The transformation has fixed points, and the flow near a critical fixed point separates relevant couplings that grow from irrelevant ones that shrink, which is why only dimension and symmetry survive to set the exponents. The one-dimensional Ising decimation carries the whole scheme through in closed form and reproduces the absence of a finite-temperature transition.\n",{"path":24256,"title":24257,"module":24258,"summary":24259},"\u002Fstatistical-mechanics\u002Ffluctuations\u002Fthermodynamic-fluctuations-and-response","Thermodynamic Fluctuations and Response Functions","Fluctuations and Response","Thermodynamic variables are sharp only on average; a macroscopic system in equilibrium fluctuates about its mean values. Einstein inverted Boltzmann's $S=k_B\\ln\\Omega$ into a Gaussian probability for a fluctuation, $w\\propto e^{\\Delta S\u002Fk_B}$, and the second moments it predicts reproduce the response functions: $\\langle\\Delta E^2\\rangle=k_BT^2C_V$, $\\langle\\Delta V^2\\rangle=k_BTV\\kappa_T$, $\\langle\\Delta M^2\\rangle=k_BT\\chi_T$. The variances diverge where the responses diverge, at a critical point, producing critical opalescence and the breakdown of the thermodynamic description.\n",{"path":24261,"title":24262,"module":24258,"summary":24263},"\u002Fstatistical-mechanics\u002Ffluctuations\u002Fbrownian-motion-and-the-langevin-equation","Brownian Motion and the Langevin Equation","A pollen grain in water executes a random walk driven by molecular collisions. Einstein tied its diffusion constant to its mobility, $D=\\mu_{\\mathrm{mob}}k_BT$, turning a visible motion into a measurement of Avogadro's number. The Langevin equation splits the collisions into a systematic drag and a random force whose strength is fixed by the drag through $\\langle\\xi(t)\\xi(t')\\rangle=2\\gamma k_BT\\,\\delta(t-t')$ — the first fluctuation–dissipation relation. The mean-square displacement grows ballistically at short times and linearly, $\\langle r^2\\rangle=2dDt$, at long times, and the Stokes–Einstein relation $D=k_BT\u002F6\\pi\\eta a$ closes the loop to Perrin's experiments.\n",{"path":24265,"title":24266,"module":24258,"summary":24267},"\u002Fstatistical-mechanics\u002Ffluctuations\u002Flinear-response-and-the-fluctuation-dissipation-theorem","Linear Response and the Fluctuation-Dissipation Theorem","A system driven by a weak external field responds through a generalized susceptibility $\\chi(\\omega)$ whose imaginary part measures dissipation. The Wiener–Khinchin theorem makes the power spectrum of equilibrium fluctuations the Fourier transform of their correlation function, and the fluctuation–dissipation theorem ties the two together: $S_x(\\omega)=(2k_BT\u002F\\omega)\\,\\chi''(\\omega)$, so the spectrum of spontaneous fluctuations is fixed by the dissipative response. The Johnson–Nyquist noise of a resistor, $\\langle V^2\\rangle=4k_BTR\\,\\Delta f$, is the canonical example, and Onsager reciprocity closes the subject.\n",{"path":24269,"title":24270,"module":6,"summary":6},"\u002Fstatistical-mechanics","Statistical Mechanics",{"path":24272,"title":24273,"module":24274,"summary":24275},"\u002Fcondensed-matter\u002Fmolecules-and-bonding\u002Fbonding-mechanisms","Bonding Mechanisms","Molecules and Chemical Bonding","A molecule forms when the total energy of two atoms drops below the energy of the separated pair. This lesson works through the four mechanisms that produce that minimum: the ionic bond from charge transfer, the covalent bond from shared electron wave functions, the metallic bond, and the weak dipole-dipole and hydrogen bonds, computing bond lengths and dissociation energies for NaCl, H₂, and H₂⁺.\n",{"path":24277,"title":24278,"module":24274,"summary":24279},"\u002Fcondensed-matter\u002Fmolecules-and-bonding\u002Fmolecular-orbitals-and-h2-plus","The Molecular-Orbital Method and H₂⁺","The hydrogen molecule ion is the two-center problem that fixes the language of chemical bonding. This lesson builds the molecular orbital as a linear combination of atomic orbitals, minimizes the energy through the variational secular equation, and reduces the result to three two-center integrals: the overlap, the Coulomb term, and the exchange (resonance) integral. The bonding and antibonding levels, their potential-energy curves, and the charge piled between the nuclei follow from those integrals.\n",{"path":24281,"title":24282,"module":24274,"summary":24283},"\u002Fcondensed-matter\u002Fmolecules-and-bonding\u002Fhydrogen-molecule-and-exchange","The Hydrogen Molecule, Exchange, and Hybridization","Adding the second electron turns the one-electron ion into the two-electron hydrogen molecule, where electron-electron repulsion and the Pauli principle govern the bond. This lesson contrasts the Heitler-London valence-bond and molecular-orbital wave functions, derives the singlet-triplet splitting as an exchange energy, shows why naive molecular orbitals fail at dissociation, and builds the sp, sp², and sp³ hybrids that fix the directed geometry of covalent bonds.\n",{"path":24285,"title":24286,"module":24274,"summary":24287},"\u002Fcondensed-matter\u002Fmolecules-and-bonding\u002Fvan-der-waals-forces","Van der Waals Forces","The bond of last resort acts between all atoms, even closed-shell noble gases with no permanent moment. This lesson separates the three van der Waals contributions — Keesom orientation, Debye induction, and London dispersion — derives the London 1\u002Fr⁶ attraction from the coupled-oscillator and second-order perturbation pictures, and assembles the Lennard-Jones potential to compute the equilibrium spacing and cohesive energy of the noble-gas crystals, argon in particular.\n",{"path":24289,"title":24290,"module":24291,"summary":24292},"\u002Fcondensed-matter\u002Fmolecular-spectra\u002Frotational-vibrational-spectra","Rotational and Vibrational Spectra of Molecules","Molecular Spectra","A diatomic molecule stores energy in three well-separated ledgers: electronic, vibrational, and rotational. Quantizing the rigid rotor gives levels spaced as ℓ(ℓ+1); quantizing the bond as a harmonic oscillator gives equally spaced vibrational levels. Their combination produces the P and R branches of an infrared absorption band, from which the bond length and force constant are read directly.\n",{"path":24294,"title":24295,"module":24291,"summary":24296},"\u002Fcondensed-matter\u002Fmolecular-spectra\u002Fanharmonicity-and-rovibrational-structure","Anharmonicity and Rovibrational Structure","The rigid rotor and harmonic oscillator are first approximations. A real bond follows the Morse potential, whose levels converge toward dissociation; a real rotor stretches centrifugally; and vibration couples to rotation, so the rotational constant depends on the vibrational level. This lesson works out the anharmonic and centrifugal corrections, the Birge-Sponer route to the dissociation energy, the isotope shift, and the thermal band envelope.\n",{"path":24298,"title":24299,"module":24291,"summary":24300},"\u002Fcondensed-matter\u002Fmolecular-spectra\u002Framan-and-electronic-bands","Raman Scattering and Electronic Bands","Not every vibration absorbs in the infrared. Raman scattering reaches modes that modulate the polarizability, giving Stokes and anti-Stokes lines whose intensity ratio measures temperature, and the mutual-exclusion rule pairs it with infrared absorption. Electronic transitions add the vibronic structure of band spectra, governed by the Franck-Condon principle, and the radiative fates of an excited state are sorted by the Jablonski diagram into fluorescence and phosphorescence.\n",{"path":24302,"title":24303,"module":24291,"summary":24304},"\u002Fcondensed-matter\u002Fmolecular-spectra\u002Flasers-and-masers","Lasers, Masers, and Stimulated Emission","Einstein's three radiative processes — absorption, spontaneous emission, and stimulated emission — and the coefficients that relate them. Stimulated emission produces coherent photons, and inverting the level populations turns it into net amplification. We build the ruby three-level laser and the helium-neon four-level laser, and show why the fourth level makes inversion easy.\n",{"path":24306,"title":24307,"module":24308,"summary":24309},"\u002Fcondensed-matter\u002Fcrystal-structure\u002Fstructure-of-solids","The Structure of Solids","Crystal Structure","A crystal is a unit cell repeated in three dimensions. We classify the common cubic lattices, compute the Coulomb energy of an ionic crystal through the Madelung constant, and show how the divergent naive lattice sum is tamed by cubic shells. The cohesive energy that results predicts melting points and connects the diatomic bond of an earlier lesson to the bulk solid.\n",{"path":24311,"title":24312,"module":24308,"summary":24313},"\u002Fcondensed-matter\u002Fcrystal-structure\u002Fbravais-lattices-and-crystal-systems","Bravais Lattices, Bases, and Crystal Structures","A crystal is a Bravais lattice decorated by a basis. This lesson separates the two, builds primitive and Wigner-Seitz cells, enumerates the seven crystal systems and fourteen Bravais lattices, and fixes the language of point and space groups. Miller indices label planes and directions, and the packing fractions of the close-packed, cubic, and diamond structures follow from the geometry.\n",{"path":24315,"title":24316,"module":24308,"summary":24317},"\u002Fcondensed-matter\u002Fcrystal-structure\u002Freciprocal-lattice-and-brillouin-zones","The Reciprocal Lattice and Brillouin Zones","Every periodic crystal has a dual lattice in wavevector space. This lesson defines the reciprocal lattice through the condition b_i dot a_j equals two pi delta, derives its properties, shows the reciprocal of fcc is bcc, links reciprocal vectors to families of lattice planes, and builds the first Brillouin zone as the Wigner-Seitz cell of the reciprocal lattice, including the higher zones.\n",{"path":24319,"title":24320,"module":24308,"summary":24321},"\u002Fcondensed-matter\u002Fcrystal-structure\u002Fdiffraction-and-structure-factors","X-ray and Neutron Diffraction","A crystal diffracts radiation whose wavelength matches its atomic spacing. This lesson derives the Bragg condition, the equivalent Laue condition 2k dot G equals G squared, and the Ewald-sphere construction, then computes the geometric structure factor that produces systematic absences for bcc and fcc, the atomic form factor, and the powder method. It closes on why neutrons and electrons complement X-rays.\n",{"path":24323,"title":24324,"module":24325,"summary":24326},"\u002Fcondensed-matter\u002Flattice-dynamics\u002Fphonon-dispersion","The Harmonic Crystal and Phonon Dispersion","Lattice Dynamics","Atoms in a crystal vibrate about their equilibrium sites, and expanding the potential to second order turns the whole lattice into a set of coupled harmonic oscillators. This lesson sets up the harmonic approximation and the dynamical matrix, solves the monatomic linear chain for its dispersion omega(k) = 2 sqrt(K\u002FM) times the absolute sine of ka over two, explains why wavevectors outside the first Brillouin zone are redundant, and extends the chain to two atoms per cell to produce acoustic and optical branches with a frequency gap.\n",{"path":24328,"title":24329,"module":24325,"summary":24330},"\u002Fcondensed-matter\u002Flattice-dynamics\u002Fphonons-quantization-and-dos","Phonons, Density of States, and Crystal Momentum","Quantizing the normal modes of a harmonic crystal turns each vibrational mode into a quantum oscillator whose excitations are phonons. This lesson counts phonons with Bose-Einstein statistics, defines crystal momentum and the normal versus Umklapp distinction in momentum conservation, builds the density of states with its van Hove singularities, and shows how inelastic neutron scattering measures a dispersion curve point by point.\n",{"path":24332,"title":24333,"module":24325,"summary":24334},"\u002Fcondensed-matter\u002Flattice-dynamics\u002Fdebye-einstein-heat-capacity","Thermal Properties — Einstein and Debye Models","The lattice heat capacity follows the classical Dulong-Petit value at high temperature but collapses toward zero as T approaches zero, a purely quantum effect. This lesson derives that behavior from the Einstein model of a single frequency, then the Debye model of a linear phonon spectrum with a cutoff, obtaining the Debye T-cubed law at low temperature and the Debye interpolation across all temperatures, and closes with thermal expansion and the Gruneisen parameter.\n",{"path":24336,"title":24337,"module":24325,"summary":24338},"\u002Fcondensed-matter\u002Flattice-dynamics\u002Fanharmonicity-and-thermal-transport","Anharmonicity, Thermal Expansion, and Heat Conduction","A perfectly harmonic crystal neither expands when heated nor resists heat flow. Both effects come from the cubic and higher terms the harmonic approximation discards. This lesson derives thermal expansion from an asymmetric interatomic potential, treats phonon-phonon scattering as the decay channel these terms open, shows why Umklapp processes are what make lattice thermal conductivity finite, and traces the temperature dependence of the conductivity and the phonon mean free path.\n",{"path":24340,"title":24341,"module":24342,"summary":24343},"\u002Fcondensed-matter\u002Ffree-electron-fermi-gas\u002Ffree-electron-gas-and-conduction","Conduction and the Free-Electron Gas","Free-Electron Fermi Gas","Drude's classical free-electron model gets Ohm's law right but the resistivity, its temperature dependence, and the heat capacity wrong. Replacing the Maxwell-Boltzmann distribution with the Fermi-Dirac distribution and treating electron-lattice collisions as wave scattering repairs all three: the Fermi energy, Fermi speed, and a mean free path set by thermal lattice vibrations.\n",{"path":24345,"title":24346,"module":24342,"summary":24347},"\u002Fcondensed-matter\u002Ffree-electron-fermi-gas\u002Fsommerfeld-model-and-heat-capacity","The Sommerfeld Model: Ground State and Heat Capacity","Quantizing the free-electron gas in a box fills a Fermi sphere in k-space. The density of states grows as the square root of energy in three dimensions, and the Fermi energy, temperature, and wavevector follow for real metals. The Sommerfeld expansion shows only a thermal shell of width k_BT near E_F is excited, giving an electronic heat capacity linear in T that sits beneath the phonon T-cubed term.\n",{"path":24349,"title":24350,"module":24342,"summary":24351},"\u002Fcondensed-matter\u002Ffree-electron-fermi-gas\u002Ftransport-and-the-hall-effect","Transport, Wiedemann–Franz, and the Hall Effect","The relaxation-time picture displaces the Fermi sphere under an applied field and gives the electrical conductivity ne-squared-tau over m. The same electrons carry heat, and their ratio yields the Wiedemann–Franz law with the universal Lorenz number. A magnetic field bends the carriers into cyclotron orbits and produces the Hall voltage, whose sign reveals the charge of the carriers.\n",{"path":24353,"title":24354,"module":24342,"summary":24355},"\u002Fcondensed-matter\u002Ffree-electron-fermi-gas\u002Fscreening-and-plasmons","Screening, Plasmons, and the Limits of Free Electrons","A mobile electron gas rearranges to screen any foreign charge, turning the bare Coulomb potential into a short-ranged Yukawa form over the Thomas–Fermi length. Displaced collectively, the gas rings at the plasma frequency, whose quantum is the plasmon and whose value sets the reflectivity edge of metals. A ledger of free-electron successes and failures then motivates band theory.\n",{"path":24357,"title":24358,"module":24359,"summary":24360},"\u002Fcondensed-matter\u002Fband-theory\u002Fblochs-theorem-and-energy-bands","Bloch's Theorem and Energy Bands","Band Theory","An electron in a periodic potential has stationary states that are plane waves modulated by a lattice-periodic envelope. This lesson proves Bloch's theorem two ways, defines crystal momentum and the band index, counts the allowed wavevectors from Born–von Kármán boundary conditions, and sets up the extended, reduced, and repeated-zone descriptions of a band.\n",{"path":24362,"title":24363,"module":24359,"summary":24364},"\u002Fcondensed-matter\u002Fband-theory\u002Fnearly-free-electron-model","The Nearly-Free-Electron Model","A weak periodic potential leaves the free-electron parabola almost intact except near Brillouin-zone boundaries, where two nearly degenerate plane waves mix. This lesson solves the resulting two-by-two secular problem, shows the gap of size twice the potential component opening at each boundary, identifies the two standing waves that pile charge on and between the ions, and works the exactly solvable Kronig–Penney model.\n",{"path":24366,"title":24367,"module":24359,"summary":24368},"\u002Fcondensed-matter\u002Fband-theory\u002Ftight-binding-method","The Tight-Binding Method","The opposite limit to nearly-free electrons builds bands from atomic orbitals. A Bloch sum of one orbital per site gives a dispersion set by the hopping integral between neighbours; the band widens from a sharp atomic level as the atoms approach. This lesson derives the s-band cosine dispersion, extends it to p-bands, and introduces Wannier functions as the localized dual of Bloch states.\n",{"path":24370,"title":24371,"module":24359,"summary":24372},"\u002Fcondensed-matter\u002Fband-theory\u002Ffermi-surfaces-and-semiclassical-dynamics","Fermi Surfaces, Effective Mass, and Metals vs Insulators","Filling the bands settles which crystals conduct. A filled band carries no current, so a crystal with filled bands and a gap is an insulator, while a partly filled band makes a metal. This lesson derives the no-current theorem for a filled band, defines the Fermi surface and Harrison's construction, introduces holes and the effective mass from band curvature, and states the semiclassical equations of motion that lead to Bloch oscillations.\n",{"path":24374,"title":24375,"module":24376,"summary":24377},"\u002Fcondensed-matter\u002Fsemiconductors\u002Fsemiconductor-bands-and-junctions","Band Theory and Semiconductors","Semiconductors","The periodic lattice splits atomic levels into allowed energy bands separated by forbidden gaps. Whether the highest occupied band is full or partly full, and how wide the gap above it is, sorts every solid into conductor, insulator, or semiconductor. Doping adds donor or acceptor levels inside the gap, and a p-n junction built from doped regions gives the diode, the solar cell, the LED, and the transistor.\n",{"path":24379,"title":24380,"module":24376,"summary":24381},"\u002Fcondensed-matter\u002Fsemiconductors\u002Fintrinsic-and-extrinsic-semiconductors","Carrier Statistics: Intrinsic and Extrinsic Semiconductors","The number of mobile electrons and holes in a semiconductor follows from the density of states near each band edge and the Fermi-Dirac tail that reaches into it. This lesson derives the effective densities of states, the intrinsic concentration and its exponential gap dependence, the law of mass action, the temperature march of the Fermi level, and the freeze-out, saturation, and intrinsic regimes of a doped crystal.\n",{"path":24383,"title":24384,"module":24376,"summary":24385},"\u002Fcondensed-matter\u002Fsemiconductors\u002Fcarrier-transport-and-recombination","Carrier Transport and Recombination","Carriers move by drift in a field and by diffusion down a concentration gradient, the two tied together by the Einstein relation. This lesson derives mobility and its scattering-limited temperature dependence, the drift and diffusion currents, the continuity equations, band-to-band and trap-assisted recombination, and the minority-carrier lifetime and diffusion length that set the length scale of every junction device.\n",{"path":24387,"title":24388,"module":24376,"summary":24389},"\u002Fcondensed-matter\u002Fsemiconductors\u002Fthe-pn-junction","The p-n Junction in Depth","Joining p-type and n-type silicon aligns their Fermi levels and leaves a depletion region of fixed charge with a built-in potential. This lesson derives the space-charge field and potential from Poisson's equation in the depletion approximation, the built-in voltage from Fermi-level alignment, the Shockley diode equation from minority-carrier diffusion, junction and diffusion capacitance, and the avalanche and Zener breakdown mechanisms.\n",{"path":24391,"title":24392,"module":24376,"summary":24393},"\u002Fcondensed-matter\u002Fsemiconductors\u002Ftransistors-and-optoelectronics","Transistors and Optoelectronic Devices","Two junctions in series make a bipolar transistor whose thin base gives current gain; a gate over an oxide makes a MOSFET whose inversion channel switches digital logic. Run in reverse, a junction converts photons to current. This lesson derives the transistor current gain and the MOSFET channel current, then treats the LED, the diode laser, and the illuminated solar-cell characteristic.\n",{"path":24395,"title":24396,"module":24397,"summary":24398},"\u002Fcondensed-matter\u002Fdielectrics-and-ferroelectrics\u002Fdielectrics-and-polarization","Dielectrics, Polarization, and the Local Field","Dielectrics and Ferroelectrics","An insulator responds to an electric field by polarizing. This lesson builds the macroscopic polarization and the dielectric constant, sorts the microscopic response into electronic, ionic, and orientational polarizability, and corrects the field an atom actually feels to the Lorentz local field E + P\u002F3 epsilon-0. The Clausius-Mossotti relation links the measured permittivity to the atomic polarizability, and the frequency dependence of each mechanism explains why the static and optical dielectric constants differ.\n",{"path":24400,"title":24401,"module":24397,"summary":24402},"\u002Fcondensed-matter\u002Fdielectrics-and-ferroelectrics\u002Fferroelectrics-and-piezoelectrics","Ferroelectrics, Piezoelectrics, and Structural Transitions","Some crystals carry a polarization with no applied field and switch it under a reversing field, tracing a hysteresis loop. This lesson develops the ferroelectric transition through the perovskite BaTiO3 displacive instability and its soft transverse-optical mode, builds the Landau free-energy theory of first- and second-order polar transitions, derives the Curie-Weiss divergence of the dielectric constant, and closes with piezoelectricity and pyroelectricity and their devices.\n",{"path":24404,"title":24405,"module":24406,"summary":24407},"\u002Fcondensed-matter\u002Fmagnetism\u002Fdiamagnetism-and-paramagnetism","Diamagnetism and Paramagnetism","Magnetism in Solids","Every solid responds to a magnetic field. Filled shells give a small negative diamagnetic susceptibility from induced Larmor currents; localized moments give a positive Curie paramagnetism described by the Brillouin function, with the ground-state moment fixed by Hund's rules. The conduction electrons add a temperature-independent Pauli paramagnetism from the thermal shell near the Fermi surface, partly cancelled by Landau diamagnetism of their orbital motion.\n",{"path":24409,"title":24410,"module":24406,"summary":24411},"\u002Fcondensed-matter\u002Fmagnetism\u002Fexchange-and-ferromagnetism","Exchange and Ferromagnetism","Magnetic ordering at hundreds of kelvin cannot be dipolar; it is an exchange effect, the Coulomb repulsion sorted by the Pauli principle into a spin-dependent energy captured by the Heisenberg Hamiltonian. Weiss molecular-field theory replaces the exchange field by an average proportional to the magnetization, giving a self-consistent equation whose solution is spontaneous magnetization below a Curie temperature and a Curie–Weiss susceptibility above it. Itinerant ferromagnetism follows from the Stoner criterion on the band density of states.\n",{"path":24413,"title":24414,"module":24406,"summary":24415},"\u002Fcondensed-matter\u002Fmagnetism\u002Fantiferromagnetism-and-domains","Antiferromagnetism, Ferrimagnetism, and Domains","A negative exchange coupling orders neighboring spins antiparallel. Two-sublattice molecular-field theory gives a Néel temperature marked by a cusp in the susceptibility, and unequal sublattices leave a net moment — ferrimagnetism, the magnetism of the ferrites. A ferromagnet breaks into domains to reduce its magnetostatic energy, separated by Bloch walls whose width is set by the competition between exchange and magnetocrystalline anisotropy, and the irreversible motion of those walls produces the hysteresis loop.\n",{"path":24417,"title":24418,"module":24406,"summary":24419},"\u002Fcondensed-matter\u002Fmagnetism\u002Fspin-waves-and-magnons","Spin Waves and Magnons","The lowest excitations of a ferromagnet are not single flipped spins but collective precessions in which every moment tips slightly and its phase advances along the crystal. These spin waves have a quadratic dispersion at long wavelength, quantize into magnons obeying Bose statistics, and their thermal population removes magnetization as the Bloch T-to-the-three-halves law. Antiferromagnetic magnons disperse linearly, and inelastic neutron scattering measures both.\n",{"path":24421,"title":24422,"module":24423,"summary":24424},"\u002Fcondensed-matter\u002Fsuperconductivity\u002Fsuperconductivity-phenomenology","Superconductivity: Phenomenology and BCS","Superconductivity","Below a critical temperature some materials lose all resistance and expel magnetic flux — the Meissner effect that defines the state. The isotope effect points to lattice vibrations, and BCS theory binds electrons into Cooper pairs through phonon exchange. The paired condensate opens an energy gap, quantizes magnetic flux, and drives the Josephson effects.\n",{"path":24426,"title":24427,"module":24423,"summary":24428},"\u002Fcondensed-matter\u002Fsuperconductivity\u002Flondon-theory-and-the-meissner-effect","London Theory and the Meissner Effect","A perfect conductor freezes the field it was cooled in; a superconductor expels it. The distinction needs a constitutive law beyond zero resistance — the two London equations — whose solution is exponential flux decay over the penetration depth. The same rigidity follows from a macroscopic condensate wave function, and the thermodynamics of the critical field fixes the condensation energy, the latent heat, and the specific-heat jump.\n",{"path":24430,"title":24431,"module":24423,"summary":24432},"\u002Fcondensed-matter\u002Fsuperconductivity\u002Fginzburg-landau-theory","Ginzburg–Landau Theory, Vortices, and Type-II","A complex order parameter and a free-energy expansion turn the superconducting transition into a Landau theory. Two lengths emerge — the coherence length and the penetration depth — whose ratio kappa sorts superconductors into type I and type II. Type-II materials admit flux as an Abrikosov lattice of vortices, each threading exactly one quantum h\u002F2e, between a lower and an upper critical field.\n",{"path":24434,"title":24435,"module":24423,"summary":24436},"\u002Fcondensed-matter\u002Fsuperconductivity\u002Fbcs-theory","Microscopic BCS Theory","A phonon-mediated attraction, however weak, binds two electrons above the Fermi sea — the Cooper problem shows the sea is unstable. The BCS variational ground state pairs all electrons near the Fermi surface and, through a self-consistent gap equation, opens an energy gap. Weak-coupling solution gives the exponential T_c and the universal ratios 2 Delta(0) = 3.53 k_B T_c and Delta C \u002F C_n = 1.43.\n",{"path":24438,"title":24439,"module":24423,"summary":24440},"\u002Fcondensed-matter\u002Fsuperconductivity\u002Fjosephson-and-high-tc","Josephson Effects and Unconventional Superconductors","Two superconductors joined by a thin barrier carry a supercurrent set by their phase difference — the dc Josephson effect — and oscillate at 2eV\u002Fh under a voltage. A two-junction loop turns flux quantization into a magnetometer of single-quantum sensitivity. The cuprates superconduct in CuO2 planes with a doping-dependent dome, d-wave pairing, and a pseudogap that lie outside the phonon picture.\n",{"path":24442,"title":24443,"module":24444,"summary":24445},"\u002Fcondensed-matter\u002Fnanostructures\u002Fquantum-wells-wires-and-dots","Quantum Wells, Wires, and Dots","Nanostructures","When a crystal is shrunk until one or more of its dimensions approaches the electron wavelength, the continuous bands of the bulk break into discrete subbands. Confining in one direction gives a quantum well with a step-like density of states, in two directions a quantum wire with inverse-square-root singularities, and in all three a quantum dot whose levels are sharp like an atom's. This lesson derives the density of states in each case and applies it to size-tunable dot emission and the Coulomb blockade of a single-electron transistor.\n",{"path":24447,"title":24448,"module":24444,"summary":24449},"\u002Fcondensed-matter\u002Fnanostructures\u002Finteger-quantum-hall-effect","The 2D Electron Gas and the Integer Quantum Hall Effect","A two-dimensional electron gas in a strong perpendicular magnetic field has its continuous density of states collapse into macroscopically degenerate Landau levels. As the field is swept, the Hall resistance locks onto exact plateaus at h over an integer times e squared, while the longitudinal resistance drops to zero. This lesson derives the Landau levels and their degeneracy, explains the plateaus through disorder-localized states and current-carrying edge channels, and states why the von Klitzing constant is now a resistance standard.\n",{"path":24451,"title":24452,"module":24444,"summary":24453},"\u002Fcondensed-matter\u002Fnanostructures\u002Ffractional-quantum-hall-and-topology","The Fractional Quantum Hall Effect and Topological Order","When the lowest Landau level is only partly filled, the non-interacting theory predicts no gap, yet a plateau appears at filling one-third. It is a many-body effect: Coulomb repulsion selects a correlated ground state, the Laughlin wavefunction, whose excitations carry a fraction of the electron charge. This lesson builds the Laughlin state, introduces composite fermions that map the fractional effect onto an integer one, and explains how the quantum Hall effect brought the Chern number and topology into condensed-matter physics.\n",{"path":24455,"title":24456,"module":24444,"summary":24457},"\u002Fcondensed-matter\u002Fnanostructures\u002Fgraphene-and-dirac-materials","Graphene and Dirac Materials","Graphene is one atomic layer of carbon on a honeycomb lattice. A tight-binding calculation on its two-atom basis gives valence and conduction bands that touch at the corners of the Brillouin zone, where the dispersion is linear and the electrons behave as massless two-dimensional Dirac particles with a fixed speed. This lesson derives the Dirac cones, the Berry phase of pi and the sublattice chirality, the anomalous half-integer quantum Hall effect that follows, and how opening a gap in a Dirac cone points toward topological insulators.\n",{"path":24459,"title":24460,"module":6,"summary":6},"\u002Fcondensed-matter","Condensed Matter Physics",{"path":24462,"title":24463,"module":21235,"summary":24464},"\u002Flogic\u002Ffoundations\u002Flogic-as-a-mathematical-model","Logic as a Mathematical Model of Deduction","Symbolic logic models deductive reasoning the way probability theory models chance: it keeps the form of a correct deduction and discards its content. A deduction is valid when its conclusion follows from the form of the premises alone, independent of what the non-logical words mean. Two models carry the subject — coarse sentential logic and fine first-order logic — and four questions organize it: logical consequence, methods of proof, the gap between provable and true, and the link between logic and computability. Tuples, relations, functions, equivalence classes, and cardinality supply the set-theoretic vocabulary every later chapter uses.\n",{"path":24466,"title":24467,"module":24468,"summary":24469},"\u002Flogic\u002Fsentential-logic\u002Fformal-languages-and-well-formed-formulas","Formal Languages and Well-Formed Formulas","Sentential Logic","The language of sentential logic has an alphabet of sentence symbols, five connectives, and two parentheses, with formation rules that pick out the well-formed formulas. The wffs are the least set of expressions closed under the five formula-building operations, and every such generated set carries an induction principle.\n",{"path":24471,"title":24472,"module":24468,"summary":24473},"\u002Flogic\u002Fsentential-logic\u002Ftruth-assignments-and-tautologies","Truth Assignments, Tautologies, and Consequence","A truth assignment fixes the sentence symbols true or false, and a recursion extends it uniquely to every formula. Satisfaction, tautologies, and tautological implication — one formula following semantically from others — rest on that extension, and the truth-table procedure decides implication for finite premise sets.\n",{"path":24475,"title":24476,"module":24468,"summary":24477},"\u002Flogic\u002Fsentential-logic\u002Funique-readability-and-parsing","Unique Readability and a Parsing Algorithm","Parentheses keep a formula from being read two ways. The parenthesis lemmas and a top-down parsing algorithm recover a formula's structure and yield unique readability: every wff has exactly one formation tree, which is what makes the truth recursion well defined.\n",{"path":24479,"title":24480,"module":24468,"summary":24481},"\u002Flogic\u002Fsentential-logic\u002Finduction-and-recursion","Induction and Recursion on Formulas","Two principles govern any set generated from initial elements by operations: prove a property of all its members by covering the initial elements and the closure steps, and define a function on it by recursion on structure. The recursion theorem needs the set to be freely generated, and unique readability supplies that condition for the well-formed formulas.\n",{"path":24483,"title":24484,"module":24468,"summary":24485},"\u002Flogic\u002Fsentential-logic\u002Fexpressive-completeness-and-normal-forms","Sentential Connectives and Normal Forms","Every formula computes a Boolean function of its atoms, and Post's theorem gives the converse: every Boolean function is realized by a wff in disjunctive normal form, so the five connectives are more than enough. Minimal complete sets follow, down to the single connectives NAND and NOR, together with a method for proving a set of connectives incomplete.\n",{"path":24487,"title":24488,"module":24468,"summary":24489},"\u002Flogic\u002Fsentential-logic\u002Fboolean-circuits","Switching Circuits","A memoryless two-valued circuit computes a Boolean function, so every formula names a gate network and every network a formula. Cost and delay are read off the formula by recursion, and tautological equivalence and normal forms design and simplify circuits realizing a given specification.\n",{"path":24491,"title":24492,"module":24468,"summary":24493},"\u002Flogic\u002Fsentential-logic\u002Fcompactness-and-effectiveness","Compactness and Effectiveness","The compactness theorem reduces satisfiability of an infinite set of formulas to its finite subsets, proved by extension to a maximal finitely satisfiable set and applied to color infinite graphs. Effectiveness fixes what \"decidable\" and \"effectively enumerable\" mean and settles the decidability of tautologyhood.\n",{"path":24495,"title":24496,"module":24497,"summary":24498},"\u002Flogic\u002Ffirst-order-languages\u002Ffirst-order-languages","First-Order Languages","First-Order Languages and Structures","Sentential logic cannot see inside a simple statement, so it misses valid arguments that turn on quantifiers and predicates. A first-order language adds a quantifier, variables, and a chosen vocabulary of predicate, function, and constant symbols. Terms and well-formed formulas are built by recursion over this alphabet, and a variable occurs free or bound according to the quantifiers that reach it.\n",{"path":24500,"title":24501,"module":24497,"summary":24502},"\u002Flogic\u002Ffirst-order-languages\u002Fstructures-truth-and-satisfaction","Structures, Truth, and Satisfaction","A structure interprets a language: a nonempty universe plus a meaning for every predicate, function, and constant symbol. Tarski's recursion defines when a structure satisfies a formula under a variable assignment, and hence when a sentence is true. From satisfaction we recover logical implication, validity, and logical equivalence for first-order logic.\n",{"path":24504,"title":24505,"module":24497,"summary":24506},"\u002Flogic\u002Ffirst-order-languages\u002Fdefinability-and-elementary-equivalence","Definability and Elementary Equivalence","Fix a structure and ask which relations a formula can pick out: the definable ones. A set of sentences picks out a class of structures, the elementary classes. Homomorphisms and isomorphisms compare structures, and the homomorphism theorem shows isomorphic structures satisfy the same sentences. Automorphisms bound what first-order logic can distinguish, giving a tool for proving relations undefinable.\n",{"path":24508,"title":24509,"module":24497,"summary":24510},"\u002Flogic\u002Ffirst-order-languages\u002Fterms-substitution-and-parsing","Parsing, Substitution, and Substitutability","Every recursion on first-order syntax rests on unique readability. A parenthesis-counting function proves that terms and formulas decompose in exactly one way, and a parsing algorithm recovers the decomposition. Substituting a term for a free variable can capture it under a quantifier; the substitutability condition rules that out, and the substitution lemma trades syntactic substitution for a change of assignment.\n",{"path":24512,"title":24513,"module":24514,"summary":24515},"\u002Flogic\u002Fdeductive-calculus\u002Fa-deductive-calculus","A Deductive Calculus for First-Order Logic","The Deductive Calculus and Its Metatheorems","A proof must be finite and mechanically checkable. A Hilbert-style calculus meets both demands: six schemas of logical axioms, a single rule of inference (modus ponens), and the syntactic consequence relation they generate. Substitution and substitutability are defined by recursion, and the bridge theorem reduces deducibility to tautological implication from the axioms.\n",{"path":24517,"title":24518,"module":24514,"summary":24519},"\u002Flogic\u002Fdeductive-calculus\u002Fdeduction-theorem-and-derived-rules","The Deduction Theorem and Derived Rules","Raw deductions from axioms are unusable by hand. The generalization theorem, the deduction theorem, contraposition, reductio ad absurdum, and rule T reduce the calculus to the moves of ordinary mathematics, each proved once to license a block of axiom-level steps. Generalization on constants and alphabetic variants handle the quantifier and substitution bookkeeping.\n",{"path":24521,"title":24522,"module":24514,"summary":24523},"\u002Flogic\u002Fdeductive-calculus\u002Fsoundness","The Soundness Theorem","Soundness is the easy half of the match between proof and truth. Whatever the calculus deduces is logically implied, by an induction on deduction length that rests on one lemma: every logical axiom is valid. The only hard case, quantifier instantiation, needs the substitution lemma. The contrapositive corollary states that every satisfiable set is consistent.\n",{"path":24525,"title":24526,"module":24514,"summary":24527},"\u002Flogic\u002Fdeductive-calculus\u002Fcompleteness-and-consistency","The Completeness Theorem","Gödel's completeness theorem is the deep converse of soundness: whatever is logically implied can be deduced. Equivalently, every consistent set has a model. The Henkin proof manufactures that model out of syntax alone: add witnessing constants, extend to a maximal consistent set, and read a term model off the formulas it contains. Compactness and the enumerability theorem drop out.\n",{"path":24529,"title":24530,"module":24531,"summary":24532},"\u002Flogic\u002Fmodels-and-theories\u002Fcompactness-and-lowenheim-skolem","Compactness and the Löwenheim–Skolem Theorems","Models, Compactness, and Theories","A set of first-order sentences has a model whenever each of its finite subsets does. This compactness theorem follows from completeness and yields the finiteness limitation, the downward and upward Löwenheim–Skolem theorems, models of every infinite cardinality, and nonstandard models of arithmetic.\n",{"path":24534,"title":24535,"module":24531,"summary":24536},"\u002Flogic\u002Fmodels-and-theories\u002Ftheories-elementary-classes-and-categoricity","Theories, Elementary Classes, and Categoricity","A theory is a set of sentences closed under logical consequence. Theories correspond to classes of models; a theory may be complete, axiomatizable, or finitely axiomatizable, and completeness together with axiomatizability yields decidability. The Łoś–Vaught test derives completeness from categoricity in a cardinal, applied to dense linear orders and to algebraically closed fields.\n",{"path":24538,"title":24539,"module":24531,"summary":24540},"\u002Flogic\u002Fmodels-and-theories\u002Finterpretations-between-theories","Interpretations Between Theories","An interpretation translates the vocabulary of one theory into formulas of another, relativizing quantifiers to a definable domain and mapping symbols to defining formulas. Defined function symbols meet a noncreativity criterion; the syntactic translation of formulas carries theoremhood forward, and a faithful interpretation transfers decidability and undecidability between theories.\n",{"path":24542,"title":24543,"module":24531,"summary":24544},"\u002Flogic\u002Fmodels-and-theories\u002Fnonstandard-analysis","Nonstandard Analysis","Compactness builds a model of the real ordered field containing infinite elements and nonzero infinitesimals. The transfer principle carries every first-order truth from the reals to this extension, the standard-part map collapses finite hyperreals back onto the reals, and continuity and the derivative are rederived by working with infinitely small quantities directly.\n",{"path":24546,"title":24547,"module":24548,"summary":24549},"\u002Flogic\u002Farithmetic-and-definability\u002Fdefinability-in-arithmetic","The Structure of Arithmetic and Definability","Number Theory and Definability","Number theory is the theory of one fixed structure, the natural numbers under successor, order, addition, multiplication, and exponentiation. Every number is named by a numeral, and a relation is definable when a single formula picks out exactly its tuples. The central gap separates the sentences true in that structure from those any reasonable set of axioms can prove.\n",{"path":24551,"title":24552,"module":24548,"summary":24553},"\u002Flogic\u002Farithmetic-and-definability\u002Fnatural-numbers-with-successor","Natural Numbers with Successor","The weakest reduct keeps only zero and successor. Its models are a standard chain together with disjoint copies of the integers, which makes the theory categorical in every uncountable power, hence complete and decidable. A quantifier-elimination procedure gives a practical decision method and shows a subset is definable if and only if it is finite or cofinite.\n",{"path":24555,"title":24556,"module":24548,"summary":24557},"\u002Flogic\u002Farithmetic-and-definability\u002Fpresburger-and-reducts","Reducts: Order, Addition, and Multiplication","Adding order to the successor reduct keeps decidability and makes the theory finitely axiomatizable; adding addition gives Presburger arithmetic, still decidable by quantifier elimination once congruence predicates are included, with definable sets exactly the eventually periodic ones. Multiplication is the break point: neither addition nor order can define it, and once it joins addition the theory stops being decidable.\n",{"path":24559,"title":24560,"module":24548,"summary":24561},"\u002Flogic\u002Farithmetic-and-definability\u002Fa-subtheory-and-representability","A Subtheory of Number Theory and Representability","A finite set of eleven axioms, the recursion equations for successor, order, addition, multiplication, and exponentiation, already proves every true quantifier-free and existential sentence. Representability asks a theory to prove the right instances of a formula rather than merely make them true, and a relation is defined to be recursive exactly when some consistent finite theory represents it. Church's thesis identifies that with decidability, and closure under composition, minimization, and primitive recursion builds the catalog the incompleteness proofs need.\n",{"path":24563,"title":24564,"module":24565,"summary":24566},"\u002Flogic\u002Fincompleteness\u002Farithmetization-of-syntax","Arithmetization of Syntax","Arithmetization and the Incompleteness Theorems","Gödel numbering assigns a natural number to every symbol, expression, formula, and deduction, turning statements about syntax into statements about numbers. The syntactic operations — substitution, \"is a wff\", \"is an axiom\", \"d codes a deduction of a\" — come out primitive recursive and hence representable in the subtheory, which lets a formula of arithmetic talk about formulas, including itself.\n",{"path":24568,"title":24569,"module":24565,"summary":24570},"\u002Flogic\u002Fincompleteness\u002Fincompleteness-and-undecidability","Incompleteness, Undecidability, and Church's Theorem","The fixed-point lemma manufactures a sentence that talks about its own Gödel number. Pointed at truth it gives Tarski's theorem — arithmetic truth is not arithmetically definable; pointed at provability it gives Gödel's first incompleteness theorem and the undecidability of the theory of the natural numbers, and, applied to validity, Church's theorem that first-order logic is undecidable. The set of theorems of a recursive theory is only recursively enumerable — the gap between provable and true.\n",{"path":24572,"title":24573,"module":24565,"summary":24574},"\u002Flogic\u002Fincompleteness\u002Fsecond-incompleteness-theorem","The Second Incompleteness Theorem","Consistency of a recursively axiomatized theory is itself an arithmetic sentence, built from a provability predicate. When the theory is strong enough to formalize its own reflection and modus ponens — the Hilbert–Bernays–Löb derivability conditions — it cannot prove that sentence unless it is inconsistent. Löb's theorem is the companion result, and set theory is the case that closes Hilbert's program.\n",{"path":24576,"title":24577,"module":24578,"summary":24579},"\u002Flogic\u002Fcomputability-and-representability\u002Frecursive-functions","Recursive Functions and Church's Thesis","Recursive Functions and Representability","The recursive functions are the formal counterpart of the effectively computable ones: built from three initial functions by composition, primitive recursion, and minimization, and equivalently the functions representable in a finitely axiomatized arithmetic. Church's thesis identifies the class with effective calculability; Kleene's normal form theorem and the unsolvable halting problem place the recursive sets strictly inside the recursively enumerable ones.\n",{"path":24581,"title":24582,"module":24578,"summary":24583},"\u002Flogic\u002Fcomputability-and-representability\u002Frepresenting-exponentiation","Representing Exponentiation and the β-Function","Coding finite sequences by prime-power exponents already assumes exponentiation, so representing exponentiation from addition and multiplication alone needs a different encoder. Gödel's β-function, built from a pairing function and the Chinese remainder theorem, reads back arbitrary finite sequences using only plus and times. This represents exponentiation in the addition-multiplication arithmetic and closes the last gap in the representability of every recursive syntactic operation.\n",{"path":24585,"title":24586,"module":24587,"summary":24588},"\u002Flogic\u002Fsecond-order-logic\u002Fsecond-order-languages","Second-Order Languages","Second-Order Logic and Beyond","Second-order logic quantifies over relations and functions, not just individuals. Second-order Peano arithmetic and the second-order theory of the reals become categorical, and finiteness is definable by a single sentence. Compactness, completeness, and the Löwenheim–Skolem theorems all fail for the standard semantics.\n",{"path":24590,"title":24591,"module":24587,"summary":24592},"\u002Flogic\u002Fsecond-order-logic\u002Fskolem-functions-and-many-sorted-logic","Skolem Functions and Many-Sorted Logic","Skolem functions replace existential quantifiers with named witnesses, putting any first-order formula into a prenex form with all existentials — now over functions — pulled to the front. The Skolemized formula is equisatisfiable with the original, which reduces satisfiability to universal sentences and, through Herbrand expansions, to sentential logic. Many-sorted logic then adds several universes at once and reduces cleanly to ordinary one-sorted logic.\n",{"path":24594,"title":24595,"module":24587,"summary":24596},"\u002Flogic\u002Fsecond-order-logic\u002Fgeneral-structures","General (Henkin) Structures","General semantics reinterprets second-order logic by letting the predicate and function quantifiers range over a designated collection of relations and functions rather than all of them. Recast as many-sorted first-order logic with comprehension axioms, general second-order logic recovers a sound and complete calculus together with compactness and Löwenheim–Skolem, giving up the categoricity of the standard semantics. The ω-models of analysis show the trade.\n",{"path":24598,"title":24599,"module":6,"summary":6},"\u002Flogic","Logic",{"path":24601,"title":24602,"module":21235,"summary":24603},"\u002Freinforcement-learning\u002Ffoundations\u002Fwhat-is-reinforcement-learning","What Is Reinforcement Learning?","Reinforcement learning is learning what to do — how to map situations to actions — so as to maximize a numerical reward signal, discovered by trial and error rather than told. We set up the agent–environment loop, separate it from supervised and unsupervised learning, name the four elements (policy, reward, value, and an optional model), and train a tic-tac-toe player with a temporal-difference value update.\n",{"path":24605,"title":24606,"module":21235,"summary":24607},"\u002Freinforcement-learning\u002Ffoundations\u002Fa-brief-history-of-rl","A Brief History of Reinforcement Learning","The origins of reinforcement learning. Three threads — trial-and-error learning from animal psychology, optimal control and dynamic programming, and temporal-difference learning — ran independently for decades and merged around 1989 into the modern field. Replacing the lookup table with a neural network then produced deep reinforcement learning: DQN, AlphaGo, AlphaZero, MuZero, and RLHF.\n",{"path":24609,"title":24610,"module":21235,"summary":24611},"\u002Freinforcement-learning\u002Ffoundations\u002Fmulti-armed-bandits","Multi-Armed Bandits","A bandit is reinforcement learning stripped to a single decision, repeated: no state, no consequences, only the tension between exploiting the arm that looks best and exploring the ones that might be better. We build up the whole toolkit — sample-average value estimates, the incremental update rule, ε-greedy, optimistic initialization, UCB, and gradient bandits — and use it to study exploration in isolation, the one problem that carries over to the full setting.\n",{"path":24613,"title":24614,"module":21235,"summary":24615},"\u002Freinforcement-learning\u002Ffoundations\u002Fbandit-exploration-algorithms","Bandit Exploration Algorithms","Better ways to explore than picking at random. Upper-confidence-bound selection explores by optimism about what it hasn't measured; gradient bandits learn action preferences by stochastic gradient ascent on reward. We then add context to get the contextual bandit, the bridge to full RL, and measure everything by regret — where UCB1 and Thompson sampling reach the logarithmic optimum that fixed-ε greedy cannot.\n",{"path":24617,"title":24618,"module":21235,"summary":24619},"\u002Freinforcement-learning\u002Ffoundations\u002Fmarkov-decision-processes","Markov Decision Processes","A Markov decision process is the formal interface between an agent and its environment: at each step the agent reads a state, chooses an action, and receives a reward and a next state. We fix that loop, the dynamics function that governs it, and the Markov property that makes the state sufficient; then turn goals into a scalar reward and rewards into a discounted return, with one notation that covers both episodic and continuing tasks.\n",{"path":24621,"title":24622,"module":21235,"summary":24623},"\u002Freinforcement-learning\u002Ffoundations\u002Fvalue-functions-and-optimality","Value Functions and Optimality","A value function scores how good a state (or state–action pair) is under a policy: the expected return from there onward. Its defining property is the Bellman equation, a self-consistency condition linking a state's value to its successors' values, which we derive from the return and the dynamics. Pushing the same idea to the best-achievable value gives the Bellman optimality equations, whose solution yields an optimal policy — and whose intractability is what the rest of the course is about.\n",{"path":24625,"title":21454,"module":24626,"summary":24627},"\u002Freinforcement-learning\u002Ftabular-methods\u002Fdynamic-programming","Tabular Solution Methods","Dynamic programming computes optimal policies when a perfect model of the MDP is given, by turning the Bellman equations into assignment statements. We build up iterative policy evaluation (the expected update), the policy improvement theorem, and the two classic algorithms that alternate them — policy iteration and value iteration — worked on the gridworld, a two-state MDP, Jack's car rental, and the gambler's problem.\n",{"path":24629,"title":24630,"module":24626,"summary":24631},"\u002Freinforcement-learning\u002Ftabular-methods\u002Fdp-async-and-gpi","Dynamic Programming: Asynchronous DP and Generalized Policy Iteration","Policy and value iteration both sweep the entire state set on every pass, which is impossible once the state space is huge. This lesson loosens the schedule: asynchronous DP updates states in any order, generalized policy iteration names the alternation of evaluation and improvement that underlies nearly every RL method, and a look at efficiency and the curse of dimensionality places DP among the alternatives. We close past Sutton & Barto with prioritized sweeping, neuro-dynamic programming, value-iteration networks, and MuZero.\n",{"path":24633,"title":24634,"module":24626,"summary":24635},"\u002Freinforcement-learning\u002Ftabular-methods\u002Fmonte-carlo-methods","Monte Carlo Methods","Monte Carlo methods learn value functions and optimal policies from complete sampled episodes, with no model of the environment: they simply average the returns that actually followed each state. We build prediction (first-visit and every-visit averaging), see why estimating action values forces the exploration question, and answer it two ways on-policy — exploring starts and epsilon-soft control. Throughout, Monte Carlo samples one whole trajectory to termination and never bootstraps.\n",{"path":24637,"title":24638,"module":24626,"summary":24639},"\u002Freinforcement-learning\u002Ftabular-methods\u002Fmonte-carlo-off-policy","Monte Carlo Methods: Off-Policy Learning","On-policy Monte Carlo can only reach the best exploring policy, not the true optimum. Off-policy methods remove that ceiling by learning about a greedy target policy from data generated by a soft behavior policy, corrected with importance sampling. We derive the importance-sampling ratio, weigh ordinary against weighted estimators on real numbers, give the incremental off-policy algorithm, sharpen it with discounting-aware sampling, and close by placing Monte Carlo on the model\u002Fbootstrap map beside DP and temporal-difference learning.\n",{"path":24641,"title":24642,"module":24626,"summary":24643},"\u002Freinforcement-learning\u002Ftabular-methods\u002Ftemporal-difference-learning","Temporal-Difference Learning","Temporal-difference learning is the one idea most central to reinforcement learning: learn a value directly from experience, like Monte Carlo, but update each guess toward the next guess before the episode ends, like dynamic programming. We derive the TD(0) prediction rule and its reward-prediction error, contrast its one-step backup with MC and DP, work the driving-home and random-walk examples, and show the batch-updating optimality that makes TD approximate the certainty-equivalence estimate.\n",{"path":24645,"title":24646,"module":24626,"summary":24647},"\u002Freinforcement-learning\u002Ftabular-methods\u002Ftd-control-sarsa-and-q-learning","TD Control: Sarsa, Q-learning, and Double Learning","With TD prediction in hand, control follows the generalized-policy-iteration pattern with TD as the evaluation step. We build Sarsa (on-policy), Q-learning (off-policy, targeting the optimal policy), and Expected Sarsa that spans the two, then confront the maximization bias every max-based method inherits and fix it with Double Q-learning. We close past Sutton & Barto, following each one-step tabular update into its deep-RL descendant — DQN, Double DQN, and Rainbow.\n",{"path":24649,"title":24650,"module":24626,"summary":24651},"\u002Freinforcement-learning\u002Ftabular-methods\u002Fn-step-bootstrapping","n-Step Bootstrapping","Monte Carlo waits for the full return; one-step TD bootstraps after a single reward. Between them lies a whole spectrum, indexed by one integer n: look ahead n real rewards, then bootstrap from the value n steps out. The n-step return unifies the previous two lessons, and — on the random walk — an intermediate n beats both extremes. We build the n-step return, the n-step TD update, the backup-diagram spectrum, and n-step Sarsa for control.\n",{"path":24653,"title":24654,"module":24626,"summary":24655},"\u002Freinforcement-learning\u002Ftabular-methods\u002Fn-step-off-policy-methods","n-Step Bootstrapping: Off-Policy Methods","Taking the n-step family off-policy raises the same importance-sampling questions Monte Carlo did, now over a window of exactly n actions. We reweight n-step returns by the policy ratio, watch the ratio product inflate variance on real numbers, then build the tree-backup algorithm that learns off-policy with no ratios at all — and finally n-step Q(sigma), one algorithm whose per-step switch recovers Sarsa, tree backup, and Expected Sarsa as special cases.\n",{"path":24657,"title":24658,"module":24626,"summary":24659},"\u002Freinforcement-learning\u002Ftabular-methods\u002Fplanning-and-learning","Planning and Learning","Planning and learning are the same operation run on two kinds of experience. A model turns states and actions into simulated transitions; planning backs up values over that simulated experience exactly as learning backs them up over real experience. We build the Dyna architecture that interleaves acting, model-learning, direct RL, and planning in one loop, trace a single Dyna-Q step by hand, and patch the architecture for when the model goes stale.\n",{"path":24661,"title":24662,"module":24626,"summary":24663},"\u002Freinforcement-learning\u002Ftabular-methods\u002Fplanning-focusing-and-decision-time","Planning: Focusing Updates and Decision-Time Search","Dyna plans by replaying remembered transitions, but sampling them uniformly wastes most of the effort. This lesson sharpens planning: prioritized sweeping works backward from states whose value just changed, expected versus sample updates weigh thoroughness against cost, and trajectory sampling and real-time DP focus updates on the states the policy actually visits. We trace Dyna forward to model-based deep RL, then turn to decision-time planning — heuristic search, rollouts, and Monte Carlo Tree Search.\n",{"path":24665,"title":24666,"module":24626,"summary":24667},"\u002Freinforcement-learning\u002Ftabular-methods\u002Fdecision-time-planning","Decision-Time Planning","Planning need not build a global policy. Decision-time planning runs a fresh lookahead every time a state arrives and returns just one action, then throws the work away. We start from real-time dynamic programming — asynchronous value iteration on the states the agent actually visits — then move through heuristic search and rollout algorithms, each a one-step policy improvement applied on the fly to the current state.\n",{"path":24669,"title":24670,"module":24626,"summary":24671},"\u002Freinforcement-learning\u002Ftabular-methods\u002Fmonte-carlo-tree-search","Monte Carlo Tree Search","Monte Carlo Tree Search is a rollout algorithm with memory: it accumulates value estimates across simulations and steers later ones toward promising branches. We work through the four steps — selection, expansion, simulation, backup — the UCT selection rule computed on real numbers, the asymmetric growing tree, and the full pseudocode. We close past Sutton & Barto with the lineage from UCT to AlphaGo, AlphaZero, and MuZero, where a learned network stands in for the leaf value and the rollout.\n",{"path":24673,"title":24674,"module":24675,"summary":24676},"\u002Freinforcement-learning\u002Fapproximation\u002Fon-policy-prediction","On-Policy Prediction with Approximation","Approximate Solution Methods","Every tabular method so far stored one number per state, which fails once the state space is large or continuous. We replace the table with a parameterized value function $\\hat v(s,\\mathbf{w})$, define the mean squared value error it should minimize under the on-policy distribution, and derive stochastic- and semi-gradient learning rules — the semi-gradient TD(0) update that bootstraps and so is not a true gradient. Linear methods make the analysis clean and give the TD fixed point; feature construction (polynomials, Fourier basis, coarse and tile coding, RBFs) supplies the vectors $\\mathbf{x}(s)$, and neural networks are the nonlinear bridge to deep RL.\n",{"path":24678,"title":24679,"module":24675,"summary":24680},"\u002Freinforcement-learning\u002Fapproximation\u002Ffeature-construction-and-nonlinear","Feature Construction and Nonlinear Approximation","Linear methods are only as good as the feature vectors $\\mathbf{x}(s)$ fed to them, and this lesson builds those vectors. Polynomials and the Fourier basis turn a state's coordinates into smooth global features; coarse coding, tile coding, and radial basis functions cover a continuous space with overlapping local receptive fields whose size sets the reach of generalization. Then we stop designing features by hand: a neural network learns the representation itself by gradient descent, trading the convergence guarantees of the linear case for expressiveness — the bridge to deep reinforcement learning.\n",{"path":24682,"title":24683,"module":24675,"summary":24684},"\u002Freinforcement-learning\u002Fapproximation\u002Fon-policy-control","On-Policy Control with Approximation","Prediction learned a value function from features; control learns to act. We carry semi-gradient methods over to action values $\\hat q(s,a,\\mathbf{w})$, giving episodic semi-gradient Sarsa and its n-step form, and solve Mountain Car by descending a cost-to-go surface. In the continuing case, function approximation makes discounting unable to affect which policy is best, so we replace it with the average-reward setting — the differential return, differential value functions, and differential semi-gradient Sarsa.\n",{"path":24686,"title":24687,"module":24675,"summary":24688},"\u002Freinforcement-learning\u002Fapproximation\u002Faverage-reward-control","Average-Reward Control for Continuing Tasks","With function approximation, discounting has no effect on a continuing task: averaged over the on-policy distribution, the discounted objective equals the average reward times a policy-independent constant, so $\\gamma$ cannot change which policy is best. This lesson replaces discounting with the average-reward setting — the long-run reward rate $r(\\pi)$, the differential return that measures each state's transient advantage over that rate, differential value functions and TD error, and differential semi-gradient Sarsa, the control method for continuing tasks that never invokes a discount factor.\n",{"path":24690,"title":24691,"module":24675,"summary":24692},"\u002Freinforcement-learning\u002Fapproximation\u002Foff-policy-and-the-deadly-triad","Off-Policy Methods and the Deadly Triad","Off-policy learning with function approximation is where the convergence guarantees of reinforcement learning fail. We extend the tabular off-policy updates to semi-gradient form with per-step importance sampling, show Baird's counterexample driving the weights to infinity, and identify the cause: the deadly triad of function approximation, bootstrapping, and off-policy training — any two are safe, all three can diverge. The divergence is not caused by sampling noise: a fully synchronous dynamic-programming update blows up just the same, which is what makes the triad a structural hazard rather than a fluke.\n",{"path":24694,"title":24695,"module":24675,"summary":24696},"\u002Freinforcement-learning\u002Fapproximation\u002Fbellman-error-and-gradient-td","Value-Function Geometry and Gradient-TD Methods","Why does the deadly triad diverge, and how do you stop it? This lesson develops the geometry that explains the failure: value functions as vectors, the projection operator onto the representable subspace, and the split between the Bellman error, the value error, and the projected Bellman error: the three objectives have different minimizers. The projected Bellman error is the learnable one, and Gradient-TD methods (GTD2, TDC) do true stochastic gradient descent on it, staying stable even off-policy at $O(d)$ cost. Emphatic TD reweights states instead, and a survey of variance-reduction techniques closes the gap between stability and usable learning.\n",{"path":24698,"title":24699,"module":24675,"summary":24700},"\u002Freinforcement-learning\u002Fapproximation\u002Feligibility-traces","Eligibility Traces","n-step methods unify TD and Monte Carlo by storing the last n feature vectors; eligibility traces do the same job with a single short-term memory vector. The λ-return averages every n-step return under a geometric weighting; the forward view looks ahead to that average, and the backward view produces nearly the same updates online through a decaying trace vector. We build the λ-return, TD(λ) with its trace, the two ways λ recovers TD(0) and Monte Carlo, a note on the exact equivalence of true online TD(λ), and Sarsa(λ) for control.\n",{"path":24702,"title":24703,"module":24675,"summary":24704},"\u002Freinforcement-learning\u002Fapproximation\u002Ftrue-online-and-sarsa-lambda","True Online TD(λ) and Sarsa(λ)","Plain TD(λ) makes the forward and backward views nearly agree; this lesson closes the gap. True online TD(λ) uses a dutch trace and a small correction term to produce exactly the same weight sequence as the online λ-return algorithm, at the same memory and only a constant factor more compute — the sharpest statement of the forward\u002Fbackward duality. The whole apparatus then lifts to control unchanged: Sarsa(λ) threads a single delayed reward back along an entire trajectory in one sweep, and the λ-weighting reappears in modern deep RL as generalized advantage estimation.\n",{"path":24706,"title":24707,"module":24675,"summary":24708},"\u002Freinforcement-learning\u002Fapproximation\u002Fpolicy-gradient-methods","Policy Gradient Methods","Every method so far learned values and read a policy off them. Policy gradient methods drop the intermediary: parameterize the policy directly and climb the performance gradient. We build the softmax-in-preferences parameterization, prove the policy gradient theorem that makes the gradient computable without the unknown state distribution, and derive REINFORCE and its variance-cutting state-value baseline — the launch point for the bootstrapping actor-critic that follows.\n",{"path":24710,"title":24711,"module":24675,"summary":24712},"\u002Freinforcement-learning\u002Fapproximation\u002Factor-critic-and-continuous-actions","Actor-Critic Methods and Continuous Actions","REINFORCE with a baseline learns a value function but never bootstraps; this lesson adds the bootstrapping critic that completes the actor-critic architecture. The critic scores each transition into a single TD error that steers both the actor's policy step and its own value step, trading a little bias for much lower variance and fully online, continuing-task learning. The policy gradient theorem carries over unchanged to the average-reward setting, a Gaussian policy handles real-valued actions with self-tuning exploration, and the natural policy gradient leads straight to TRPO, PPO, and the deep actor-critic methods that train today's agents.\n",{"path":24714,"title":24715,"module":24675,"summary":24716},"\u002Freinforcement-learning\u002Fapproximation\u002Fleast-squares-and-memory-based-methods","Least-Squares TD","Semi-gradient TD spends one cheap step per example and needs many examples; this lesson makes the opposite tradeoff. Least-Squares TD (LSTD) accumulates the matrices $\\mathbf{A}$ and $\\mathbf{b}$ and solves the TD fixed point $\\mathbf{w} = \\mathbf{A}^{-1}\\mathbf{b}$ directly, using the Sherman-Morrison identity to maintain the inverse in $O(d^2)$ — the most data-efficient linear TD method, at a quadratic cost. We work a solve by hand, weigh the quadratic cost against semi-gradient TD's cheap steps, and note that LSTD never forgets — a problem in control, where least-squares policy iteration is the natural extension.\n",{"path":24718,"title":24719,"module":24675,"summary":24720},"\u002Freinforcement-learning\u002Fapproximation\u002Fmemory-and-kernel-methods","Memory-Based and Kernel Methods","Least-squares TD spent more compute to extract more from each example; this lesson drops the parametric form entirely. Memory-based methods store training examples untouched and answer a query locally at retrieval time — nearest neighbor, weighted average, locally weighted regression — so accuracy grows with the data and effort concentrates where the agent actually goes. Kernel-based methods weight stored examples by a similarity kernel $k(s,s')$, and every linear method turns out to be a kernel method. Interest and emphasis, finally, make the on-policy weighting itself a design choice, aiming scarce approximation capacity at the states that matter.\n",{"path":24722,"title":24723,"module":24675,"summary":24724},"\u002Freinforcement-learning\u002Fapproximation\u002Foff-policy-eligibility-traces","Off-Policy Eligibility Traces","Eligibility traces meet off-policy learning and function approximation — the corner where stability gets hard. We first let the bootstrapping and discounting parameters vary with state, so a single generalized return covers episodic and continuing tasks and folds termination into the discount. Then we fold the per-decision importance ratio into the trace with a control-variate correction, and build Watkins's Q(λ) and its importance-sampling-free successor Tree-Backup(λ) — all correct in expectation, but still semi-gradient, so the deadly triad and its fixes wait for the next lesson.\n",{"path":24726,"title":24727,"module":24675,"summary":24728},"\u002Freinforcement-learning\u002Fapproximation\u002Fstable-off-policy-traces","Stable Off-Policy Methods with Traces","Off-policy traces get the expected target right, but with $\\lambda \u003C 1$ they bootstrap, so off-policy plus bootstrapping plus function approximation is the deadly triad and the weights can diverge. This lesson carries the two one-step fixes to traces: GTD(λ) and GQ(λ) add a second weight vector and a gradient correction for true gradient descent on the projected Bellman error, while Emphatic TD(λ) reweights updates through a followon trace and interest to recover the on-policy stability. It closes with the implementation reality that traces are cheap because they are sparse, and with Retrace and V-trace — the clipped-ratio descendants that make off-policy traces work at deep-RL scale.\n",{"path":24730,"title":24049,"module":24731,"summary":24732},"\u002Freinforcement-learning\u002Fdeep-rl\u002Fdeep-q-networks","Deep Reinforcement Learning","Deep Q-networks replace the linear value function with a neural network $Q(s,a;\\mathbf{w})$ and confront the fact that a nonlinear approximator, off-policy bootstrapping, and correlated online data — the deadly triad — make naive Q-learning diverge. DQN counters this empirically with two stabilizers: an experience replay buffer that decorrelates and reuses samples, and a periodically-frozen target network that fixes the bootstrap target. We derive the DQN loss and gradient, walk through the Atari convolutional architecture and its results, and then add the three refinements that define modern value-based deep RL — Double DQN, dueling networks, and prioritized experience replay.\n",{"path":24734,"title":24735,"module":24731,"summary":24736},"\u002Freinforcement-learning\u002Fdeep-rl\u002Fdqn-improvements","DQN Improvements: Double, Dueling, and Prioritized Replay","Three refinements that turn plain DQN into the standard modern value-based agent, each touching a different part of the system. Double DQN fixes the maximization bias in the target by splitting action selection from evaluation; dueling networks restructure the network around a state value and per-action advantages; prioritized replay changes which transitions are learned from. We close with Rainbow, which combines them, and the distributional view that predicts the whole return distribution rather than its mean.\n",{"path":24738,"title":24739,"module":24731,"summary":24740},"\u002Freinforcement-learning\u002Fdeep-rl\u002Factor-critic-and-ppo","Actor–Critic and GAE","Make the actor and the critic deep networks and the policy-gradient architecture becomes modern deep RL. We build the neural actor-critic, the advantage estimate that replaces the raw return, and Generalized Advantage Estimation as a λ-blend of n-step advantages, then the parallel-worker methods A3C and A2C that decorrelate on-policy data. The step-size constraints — trust regions, PPO, and the continuous-control family — follow in the next lesson.\n",{"path":24742,"title":24743,"module":24731,"summary":24744},"\u002Freinforcement-learning\u002Fdeep-rl\u002Fppo-and-continuous-control","PPO and Continuous Control","Keeping the policy-gradient step from destroying the policy, and the algorithms that result. Trust-region optimization bounds each update by a KL constraint; PPO keeps that goal but replaces the second-order machinery with a first-order clip on the probability ratio, which is why it is the modern default and the optimizer inside RLHF. We then tour the off-policy continuous-control family — DDPG, TD3, and SAC — and where actor-critic went at scale, from OpenAI Five to language-model alignment.\n",{"path":24746,"title":24747,"module":24731,"summary":24748},"\u002Freinforcement-learning\u002Fdeep-rl\u002Fcase-studies","Case Studies: Learning to Play","The game-playing systems that turned reinforcement learning from a theory into a track record: Samuel's checkers player, TD-Gammon, Watson's Daily-Double wagering, a reinforcement-learning memory controller, DQN, and AlphaGo through AlphaGo Zero. Read as a set they draw one line — a value function, learned by self-play or interaction, refined by search, carried by a deep network — that runs from a 1959 checkers program to superhuman Go.\n",{"path":24750,"title":24751,"module":24731,"summary":24752},"\u002Freinforcement-learning\u002Fdeep-rl\u002Frl-beyond-games","Reinforcement Learning Beyond Games","The same value-and-reward machinery, pointed at problems with no opponent. Web personalization as a contextual bandit and then a full MDP for life-time value; thermal soaring, where a glider learns to climb on turbulent air and reward design does most of the work; and the industrial-scale systems that carried the same design past Sutton & Barto — AlphaStar, OpenAI Five, GT Sophy, and RLHF, where the reward itself is learned from human preference.\n",{"path":24754,"title":24755,"module":24731,"summary":24756},"\u002Freinforcement-learning\u002Fdeep-rl\u002Ffrontiers","Frontiers: Beyond the Standard MDP","The standard MDP fixes three things — state, reward, and single-step actions — and this lesson loosens two of them. We generalize the value function into a general value function that predicts any signal, and use those predictions as auxiliary tasks that shape representations; we extend actions in time with the options framework; and we treat state as a construction the agent builds from a stream of observations. Reward design and the open problems follow in the next lesson.\n",{"path":24758,"title":24759,"module":24731,"summary":24760},"\u002Freinforcement-learning\u002Fdeep-rl\u002Freward-design-and-open-problems","Reward Design and Open Problems","How to design a reward signal that encodes the intended goal — sparse reward, shaping, and reward hacking — and the problems the whole tabular, approximate, and deep arc leaves unsolved. We close with how the frontiers were pushed after Sutton & Barto: auxiliary tasks, learned options, intrinsic-motivation bonuses, learned world models, and offline RL, then the two concerns of reward hacking and safety that any real-world agent must address.\n",{"path":24762,"title":24763,"module":24764,"summary":24765},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fdistributional-and-rainbow","Sharpening DQN: Improvements and the Distributional Idea","Modern Deep Reinforcement Learning","In the years after the 2015 DQN paper, a stream of focused improvements each fixed one weakness of the baseline without disturbing its frame. This lesson recaps five that keep the scalar $Q$-value — Double DQN, multi-step returns, dueling networks, prioritized replay, and NoisyNets, each changing a different slot of the same Q-learning loop — then develops the sixth, distributional RL, which changes the objective itself: learn the whole return distribution $Z(s,a)$. We build the distributional Bellman equation and the C51 categorical algorithm, projection step and all, worked end to end on real numbers. A companion lesson takes up QR-DQN, Rainbow, and the modern distributional line.\n",{"path":24767,"title":24768,"module":24764,"summary":24769},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fdistributional-and-rainbow-part-2","Distributional RL and Rainbow","A companion to the DQN improvements lesson. C51 fixed the return atoms and learned their probabilities; QR-DQN does the reverse — fix the probabilities, learn the values — which removes the projection and trains with a quantile loss. We cover why the distribution helps even when you act on the mean, then assemble Rainbow: all six improvements in one Q-learning loop, with the component ablation that shows each one's real weight. The distributional line then runs on through IQN, FQF, and Agent57, the first agent to beat the human baseline on all 57 Atari games.\n",{"path":24771,"title":24772,"module":24764,"summary":24773},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fcontinuous-control","Continuous Control: DDPG and TD3","When actions are real-valued, the $\\arg\\max_a Q(s,a)$ in Q-learning becomes an optimization problem on every step. This lesson builds the off-policy actor-critic family that sidesteps it: the deterministic policy gradient and DDPG, which replaces the max with a learned actor, and the three fixes of TD3 that counter the value overestimation DDPG inherits. A companion lesson takes up SAC's maximum-entropy objective and the methods built on this template.\n",{"path":24775,"title":24776,"module":24764,"summary":24777},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fcontinuous-control-part-2","Continuous Control: SAC and Beyond","A companion to the DDPG and TD3 lesson. Where those actors are deterministic and explore with bolted-on noise, soft actor-critic (SAC) changes the objective itself: maximize return plus the entropy of the policy, so exploration becomes intrinsic and the agent stays robust. We develop the maximum-entropy objective, the reparameterized squashed-Gaussian actor, and automatic temperature tuning, then survey the methods built on this off-policy template — distributional critics (D4PG), critic ensembles (REDQ), and control from pixels (DrQ, RAD).\n",{"path":24779,"title":24780,"module":24764,"summary":24781},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmodel-based-rl","Model-Based Deep RL: Sample Efficiency and PETS","A model turns experience into imagined planning. This lesson makes the sample-efficiency case for learning a dynamics model, works through why a learned model's errors compound over the planning horizon, and builds the most direct model-based method: PETS plans online with a probabilistic ensemble under model-predictive control, distrusting the model exactly where its members disagree. A companion lesson takes up latent world models (Dreamer) and MuZero.\n",{"path":24783,"title":24784,"module":24764,"summary":24785},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmodel-based-rl-part-2","Model-Based Deep RL: World Models, Dreamer, and MuZero","A companion to the PETS lesson. PETS plans in the environment's native state space; these methods change what the model represents. World Models and Dreamer learn a compact latent state and do almost all their learning by imagining inside it, with value gradients flowing through the differentiable dynamics. MuZero predicts neither states nor pixels — only the reward, value, and policy that MCTS reads — and plans with search against that learned model, AlphaZero without the rules. We close with MBPO, TD-MPC, and EfficientZero.\n",{"path":24787,"title":24788,"module":24764,"summary":24789},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fexploration","Exploration in Deep RL: Novelty as Reward","When the state space is enormous and reward is rare, ε-greedy amounts to a random walk that almost never reaches the first reward. This lesson scales the bandit's exploration ideas up to deep RL through the dominant approach — manufacture a reward for novelty and let the agent chase it: optimism and pseudo-counts from density models, and intrinsic motivation and curiosity (the Intrinsic Curiosity Module and Random Network Distillation). A companion lesson takes up posterior sampling, Go-Explore, and the modern methods.\n",{"path":24791,"title":24792,"module":24764,"summary":24793},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fexploration-part-2","Exploration in Deep RL: Posterior Sampling and Go-Explore","A companion to the novelty-as-reward lesson. Pseudo-counts and curiosity reward the unfamiliar after the agent stumbles into it; this lesson covers two ideas that go further. Bootstrapped DQN keeps an ensemble that approximates a posterior over value functions and explores by committing to one sampled hypothesis per episode — the deep, directed exploration ε-greedy cannot manage. Go-Explore remembers and returns to the frontier, defeating detachment and derailment to solve Montezuma's Revenge. We close with episodic memory (Never Give Up), Agent57, and model-based exploration.\n",{"path":24795,"title":24796,"module":24764,"summary":24797},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Foffline-rl","Offline RL: The Problem and Value-Based Fixes","Offline reinforcement learning learns a policy from a fixed logged dataset with no further environment interaction — off-policy learning pushed to the extreme, and it breaks for the extreme version of the same reason. Bootstrapping queries the value function at out-of-distribution actions the data never covers, those errors are optimistic, and with no online feedback to correct them they compound through the Bellman backup. This lesson sets up the failure and off-policy evaluation, then builds the first two families of pessimistic fixes: policy constraint (BCQ) and conservative value estimation (CQL). A companion lesson takes up implicit methods, model-based offline RL, and Decision Transformer.\n",{"path":24799,"title":24800,"module":24764,"summary":24801},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Foffline-rl-part-2","Offline RL: Implicit Methods, Sequence Models, and Beyond","A companion to the offline-RL problem lesson. Policy constraint and conservative value estimation both still query a learned value function; implicit methods (IQL) avoid querying it off the data at all, using an in-sample expectile backup. We then build pessimism into a learned model (MOPO, COMBO) and drop bootstrapping entirely with Decision Transformer's return-conditioned sequence modeling, closing with offline-to-online fine-tuning, diffusion planners, and the offline view of RLHF. The one rule throughout: without online correction, be pessimistic about what you cannot verify.\n",{"path":24803,"title":24804,"module":24764,"summary":24805},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fimitation-and-inverse-rl","Imitation Learning: Cloning, DAgger, and Inverse RL","When a reward is hard to specify but an expert is easy to watch, learn from demonstrations instead. Behavioral cloning treats control as supervised learning of the expert's state-to-action map, and fails through compounding error: small mistakes carry the agent off the expert's distribution, where it was never trained. DAgger fixes the mismatch by querying the expert on the learner's own states. Inverse RL instead recovers the reward the expert seems to optimize — an ill-posed problem that maximum-entropy IRL disambiguates. A companion lesson casts imitation as adversarial occupancy matching (GAIL, AIRL).\n",{"path":24807,"title":24808,"module":24764,"summary":24809},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fimitation-and-inverse-rl-part-2","Imitation as Adversarial Matching: GAIL and AIRL","A companion to the imitation-learning lesson. If the point of recovering a reward is only to re-run RL and match the expert, you can skip the reward and match the behavior directly. GAIL casts imitation as a GAN — a discriminator separating expert from learner state-action pairs supplies the reward a policy-gradient method optimizes — matching occupancy measures without ever naming a reward. AIRL reads a transferable reward back out of the discriminator. We compare all four methods and close with reward models in RLHF, scaled cloning, and diffusion policies.\n",{"path":24811,"title":24812,"module":24764,"summary":24813},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmulti-agent-rl","Multi-Agent RL: Markov Games and Centralized Training","With more than one learning agent in an environment, each agent's world becomes non-stationary because the others are changing too. This lesson builds the Markov-game generalization of the MDP, diagnoses non-stationarity as the central obstacle, shows why the naive baselines fail, and develops the dominant fix — centralized training with decentralized execution (MADDPG, VDN, QMIX). A companion lesson takes up self-play, the landmark game-playing systems, and the equilibrium concepts that define what \"solved\" means.\n",{"path":24815,"title":24816,"module":24764,"summary":24817},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmulti-agent-rl-part-2","Multi-Agent RL: Self-Play and Solution Concepts","A companion to the Markov-games lesson. In the purely competitive setting, an agent can generate its own training curriculum by playing against copies of itself — self-play, the method behind AlphaGo, OpenAI Five, and AlphaStar. We develop why self-play produces an ever-improving opponent, the systems it built, and then the equilibrium solution concepts (Nash, correlated, coarse-correlated) that define what \"solved\" means once there is an opponent, closing with PSRO, MAPPO, and the language-model-agent frontier.\n",{"path":24819,"title":24820,"module":24764,"summary":24821},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fhierarchical-rl","Hierarchical RL: Options and the Option-Critic","Flat RL cannot explore a long horizon: reaching reward through hundreds of primitive actions is exponentially unlikely, and every credit-assignment update crawls one step at a time. Hierarchy breaks one hard long-horizon problem into many short ones. This lesson develops temporal abstraction — the options framework and its semi-Markov view, and learning options end to end with the option-critic. A companion lesson takes up goal-conditioned manager\u002Fworker hierarchies (FeUdal Networks and HIRO), hindsight relabeling, and unsupervised skill discovery.\n",{"path":24823,"title":24824,"module":24764,"summary":24825},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fhierarchical-rl-part-2","Hierarchical RL: Goal-Conditioned Hierarchies and Skills","A companion to the options lesson. Options package a behavior; goal-conditioned hierarchies instead give the top level an explicit language of goals — a manager proposes a target state or a latent direction, and a worker is rewarded for reaching it (FeUdal Networks, HIRO). We develop that architecture, the hindsight relabeling that lets it learn from sparse reward, and unsupervised skill discovery (DIAYN) that learns a repertoire of behaviors with no reward at all. The shared idea throughout: shorten the horizon by inserting a level that decides less often.\n",{"path":24827,"title":24828,"module":24764,"summary":24829},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Frlhf-and-language-models","RLHF and Language Models","A language model trained to predict the next token is fluent but not helpful, honest, or harmless — the objective it was optimized for is not the objective we want. RLHF closes that gap by turning the one thing humans do reliably, comparing two outputs, into a reward. We build the three-stage pipeline: supervised fine-tuning, a Bradley-Terry reward model fit to preference pairs, then PPO against that reward with a KL penalty keeping it near the reference policy. We then cover reward hacking and why the KL penalty matters, Direct Preference Optimization, which folds the reward model into a single classification loss, and the RLAIF and verifiable-reward variants. This pipeline is what makes the largest models usable as assistants.\n",{"path":24831,"title":24832,"module":24764,"summary":24833},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fpartial-observability-pomdps","Partial Observability: POMDPs and the Belief State","Drop the assumption that the agent sees the state. It sees an observation, a partial and noisy function of a hidden state, and one observation is no longer a Markov signal. This lesson builds the POMDP tuple, shows that the belief state — the posterior over hidden states — is a sufficient statistic that turns a POMDP back into an MDP over beliefs, and works the Bayes-filter belief update step by step. A companion lesson explains why exact planning is intractable and develops the deep-RL answer of recurrent, history-based policies.\n",{"path":24835,"title":24836,"module":24764,"summary":24837},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fpartial-observability-pomdps-part-2","Partial Observability: Planning and Recurrent Policies","A companion to the belief-state lesson. In principle a POMDP reduces to an MDP over beliefs; in practice two obstacles block that. Exact planning over the belief simplex is intractable — the value function is piecewise-linear-and-convex with a number of pieces that can explode — and computing the belief needs a model the agent rarely has. This lesson develops the intractability, the point-based approximations that address it, and the deep-RL answer: make the policy a function of history with a recurrent network (DRQN, R2D2), with frame-stacking, attention, and world-model latents as learned beliefs.\n",{"path":24839,"title":24840,"module":24764,"summary":24841},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fsafe-and-constrained-rl","Safe and Constrained RL: The CMDP and Policy Methods","Maximizing a scalar reward is not the same as behaving well: a capable optimizer will find and exploit any gap between the reward and what its designer actually meant, a failure called specification gaming or reward hacking. The remedy is to add explicit cost constraints — the constrained MDP — maximizing return subject to an expected-cost budget. This lesson builds the core toolkit: the CMDP itself, Lagrangian primal-dual methods that learn a multiplier on the constraint (RCPO), and constrained policy optimization (CPO) with its trust-region cost bound. A companion lesson covers risk-sensitivity, safe exploration, and the alignment framing.\n",{"path":24843,"title":24844,"module":24764,"summary":24845},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fsafe-and-constrained-rl-part-2","Safe RL: Risk, Safe Exploration, and Alignment","A companion to the constrained-MDP lesson. Constraining the mean cost is not enough: a policy safe on average can be catastrophic in the tail, and a policy safe at convergence can violate its limits wildly while learning. This lesson optimizes the tail with risk-sensitive objectives (CVaR), then makes exploration itself safe with shields, Lyapunov methods, and safety layers that project unsafe actions onto the feasible set — closing with benchmarks, safe RLHF, robustness, and the alignment framing that ties safety back to the problem of incompletely specified reward.\n",{"path":24847,"title":24848,"module":24764,"summary":24849},"\u002Freinforcement-learning\u002Fmodern-deep-rl\u002Fmeta-rl-and-generalization","Meta-RL and Generalization","An agent that masters one task often fails on the next; it has overfit to a single environment. This lesson treats fast adaptation as a meta-problem over a distribution of tasks: meta-train so that a few episodes at meta-test time suffice. We cover the two families — optimization-based (MAML learns an initialization) and context-based (RL-squared and PEARL infer a latent task) — the exploration cost of adaptation, and the parallel problem of generalization: why deep RL memorizes environments and what fixes it (domain randomization, procedural generation, augmentation, regularization). It closes on foundation models and sequence-model agents as the generalist endpoint.\n",{"path":24851,"title":24852,"module":24853,"summary":24854},"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fpsychology-of-reinforcement","The Psychology of Reinforcement","Reinforcement Learning in Minds and Brains","Reinforcement learning is both an engineering method and a theory of how animals learn. The prediction\u002Fcontrol split of the algorithms mirrors the psychologist's split between classical and instrumental conditioning. We trace the correspondence: the Rescorla–Wagner model as a prediction-error rule that explains blocking, its real-time TD extension, Thorndike's Law of Effect behind trial-and-error control, and the habitual\u002Fgoal-directed distinction that maps onto model-free versus model-based learning.\n",{"path":24856,"title":24857,"module":24853,"summary":24858},"\u002Freinforcement-learning\u002Fminds-and-brains\u002Finstrumental-conditioning-and-control","The Psychology of Reinforcement: Instrumental Control","Classical conditioning was prediction; instrumental conditioning is control. Thorndike's Law of Effect is trial-and-error control — selection plus association, search plus memory — and Skinner's shaping and schedules are reward engineering. The habitual\u002Fgoal-directed distinction maps onto model-free versus model-based control, dissociated by outcome devaluation and arbitrated by uncertainty. Delayed reinforcement is the credit-assignment problem, and the stimulus traces and secondary reinforcers of animal-learning theory are eligibility traces and value functions.\n",{"path":24860,"title":24861,"module":24853,"summary":24862},"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fdopamine-and-td-error","Dopamine and the TD Error","The TD error was invented as an algorithm; a decade later it turned out to closely describe the firing of the brain's dopamine neurons. We follow Schultz's experiments — dopamine fires at an unpredicted reward, shifts to the earliest predictive cue, and dips below baseline when a predicted reward is withheld — and match each result to the TD error term by term. We then read the basal ganglia as a neural actor–critic with dopamine as its shared training signal, and close on addiction as a hijacking of that signal.\n",{"path":24864,"title":24865,"module":24853,"summary":24866},"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fdopamine-in-the-brain","Dopamine in the Brain: The Neural Actor–Critic","If phasic dopamine is a TD error, where does it go and what does it change? We follow the axons into the basal ganglia, read the corticostriatal synapse as the place where state, action, and error meet, and map the ventral and dorsal striatum onto the critic and the actor of an actor–critic. Addiction becomes a broken cancellation in the same learning signal, and distributional dopamine extends the scalar RPE into a population code.\n",{"path":24868,"title":24869,"module":24853,"summary":24870},"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fanimal-learning-and-cognition","Animal Learning and Cognition","Three classic associative phenomena turn out to be reinforcement-learning mechanisms seen in behavior. Blocking says learning is driven by prediction error, not co-occurrence, and reduces to least-squares regression fitting a collinear feature. Higher-order conditioning and conditioned reinforcement make a value estimate a secondary reinforcer — bootstrapping in an animal. Delayed reinforcement is the credit-assignment problem, and the stimulus traces and goal gradients of Pavlov and Hull are eligibility traces and TD-learned value functions.\n",{"path":24872,"title":24873,"module":24853,"summary":24874},"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fcognitive-maps-and-planning","Cognitive Maps and Model-Based Learning","Tolman's rats learned the layout of a maze with no reward, then used it the moment food appeared — latent learning, a cognitive map, and the behavioral face of model-based reinforcement learning. The map is learned by system identification (stimulus–stimulus associations), which fills in whether or not reward is present, and queried by planning, which re-solves a route from a single changed reward. The successor representation sits between cache and model, and hippocampal predictive maps and scaled-up world models carry the same idea into brain and machine.\n",{"path":24876,"title":24877,"module":24853,"summary":24878},"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fneuroscience-of-reinforcement","The Neuroscience of Reinforcement","The dopamine story is one contact point between reinforcement learning and the brain; this lesson fills in the surrounding neuroscience so the mapping stands on its own. We build a working primer of neurons, synapses, and neuromodulation; separate four signals that casual usage conflates — reward, reinforcement, value, and prediction error; and read the actor and critic as corticostriatal synapses updated by two- and three-factor rules, grounded in spike-timing-dependent and reward-modulated plasticity.\n",{"path":24880,"title":24881,"module":24853,"summary":24882},"\u002Freinforcement-learning\u002Fminds-and-brains\u002Fseveral-learning-systems","The Brain's Several Learning Systems","The actor's three-factor rule has an ancestor in Klopf's hedonistic neuron — a single cell as a reinforcement-seeking agent — and a bacterium's run-and-twiddle shows the Law of Effect with no synapses at all. Teams of such neurons implement policy gradient collectively, the broadcast reward replacing backpropagation. And the brain is not only model-free: outcome devaluation, prefrontal value coding, and hippocampal forward sweeps localize a model-based system. The recurring conclusion is that the brain is several interacting learning systems, not one algorithm.\n",{"path":24884,"title":24041,"module":6,"summary":6},"\u002Freinforcement-learning",{"path":24886,"title":24887,"module":21235,"summary":24888},"\u002Fartificial-intelligence\u002Ffoundations\u002Fwhat-is-ai","What Is Artificial Intelligence?","Eight definitions of AI fall into a two-by-two grid: think versus act, and measure success against human performance versus an ideal standard of rationality. We work through all four schools — the Turing test, cognitive modelling, the laws of thought, and the rational agent — and adopt the last as the frame for the whole course: AI is the study and design of rational agents.\n",{"path":24890,"title":24891,"module":21235,"summary":24892},"\u002Fartificial-intelligence\u002Ffoundations\u002Ffoundations-of-ai","The Foundations of AI","Where the rational-agent idea came from and what surrounds it. AI inherited its core tools from eight older disciplines — philosophy, mathematics, economics, neuroscience, psychology, computer engineering, control theory, and linguistics. Its history runs in cycles of boom and winter, from the 1956 Dartmouth workshop through expert systems to the statistical turn. And the deep-learning era — AlexNet, the Transformer, GPT-3, AlphaGo — is a new way of computing the agent function at scale, not a new definition of AI.\n",{"path":24894,"title":24895,"module":21235,"summary":24896},"\u002Fartificial-intelligence\u002Ffoundations\u002Fintelligent-agents","Intelligent Agents","An agent perceives an environment through sensors and acts on it through actuators; its behavior is an agent function mapping percept sequences to actions. A rational agent chooses, for each percept sequence, the action that maximizes its expected performance measure given its knowledge. We build the first half of the vocabulary the whole course rests on — the agent function, rationality, PEAS task specifications, and the six axes along which task environments vary.\n",{"path":24898,"title":24899,"module":21235,"summary":24900},"\u002Fartificial-intelligence\u002Ffoundations\u002Fagent-architectures","Agent Architectures","How to build a program that computes a good agent function without storing an astronomically large lookup table. Four skeleton architectures in order of increasing power — simple reflex, model-based, goal-based, and utility-based — plus the learning agent that improves any of them, the scale of world representations (atomic, factored, structured) they rest on, and how a modern language-model agent fits the same frame.\n",{"path":24902,"title":24903,"module":24904,"summary":24905},"\u002Fartificial-intelligence\u002Fsearch\u002Funinformed-search","Uninformed Search","Search","A goal-based agent that cannot see which action is best turns the problem into a state space — an initial state, a set of actions, a transition model, a goal test, and a path cost — and searches for a sequence of actions reaching the goal. We build the state-space formulation on the 8-puzzle and route-finding, give the one TREE-SEARCH \u002F GRAPH-SEARCH skeleton every algorithm specializes, and measure strategies by completeness, optimality, and complexity. This lesson develops the first two frontier disciplines — breadth-first and uniform-cost search; the rest follow in the next lesson.\n",{"path":24907,"title":24908,"module":24904,"summary":24909},"\u002Fartificial-intelligence\u002Fsearch\u002Fsearch-strategies-compared","Search Strategies Compared","Breadth-first and uniform-cost search pay for optimality in memory. This lesson develops the strategies that trade memory for depth: depth-first search, which keeps only the current path; depth-limited and iterative-deepening search, which fix DFS's failure on infinite paths; and bidirectional search, which meets in the middle for a square-root saving. It closes by lining up all six uninformed strategies against completeness, optimality, and complexity, and tracing where the algorithms came from and where they went.\n",{"path":24911,"title":24912,"module":24904,"summary":24913},"\u002Fartificial-intelligence\u002Fsearch\u002Finformed-search","Informed Search and A*","An informed search uses a heuristic $h(n)$, an estimate of the cost from a node to the goal, to decide what to expand next. Greedy best-first search follows the heuristic blindly and gives up optimality; A* corrects it by ranking nodes on $f(n) = g(n) + h(n)$, and is optimal when the heuristic is admissible (tree search) or consistent (graph search). This lesson defines the heuristic, builds best-first search, and proves why A* is optimal, with the contour picture that explains its pruning. Where good heuristics come from is the next lesson.\n",{"path":24915,"title":24916,"module":24904,"summary":24917},"\u002Fartificial-intelligence\u002Fsearch\u002Fheuristic-functions","Heuristic Functions and Memory-Bounded Search","A* is only as good as its heuristic, so this lesson answers where good heuristics come from: relaxed problems, whose exact solution cost is an admissible heuristic, and pattern databases, which precompute subproblem costs. It measures heuristic quality with dominance and the effective branching factor, then tackles A*'s memory problem with IDA*, RBFS, and SMA*. It closes with modern heuristic search — weighted A*, learned and disjoint pattern-database heuristics, and bidirectional A*.\n",{"path":24919,"title":24920,"module":24904,"summary":24921},"\u002Fartificial-intelligence\u002Fsearch\u002Flocal-search","Local Search and Optimization","When the path to a goal is irrelevant and only the final state matters, we can discard the search tree entirely and keep just the current state, moving to a better neighbor at each step. This lesson builds the state-space landscape metaphor, works through hill climbing and the three obstacles that defeat it (local maxima, ridges, plateaus), then develops the first escapes: random restarts and simulated annealing with its temperature schedule. The population-based methods and continuous-space calculus follow in the next lesson.\n",{"path":24923,"title":24924,"module":24904,"summary":24925},"\u002Fartificial-intelligence\u002Fsearch\u002Fpopulation-and-continuous-search","Population and Continuous Search","Single-state local search escapes a trap by restarting or tolerating downhill moves. This lesson develops the alternatives that keep several states at once — local beam search, which shares successors across parallel threads, and genetic algorithms, which recombine two parents through crossover and mutation — then crosses into continuous spaces, where calculus replaces the finite neighbor set: gradient ascent, line search, and Newton's method. It closes with the industrial descendants of these methods and the loop they all share.\n",{"path":24927,"title":24928,"module":24904,"summary":24929},"\u002Fartificial-intelligence\u002Fsearch\u002Fadversarial-search","Adversarial Search and Games","When another agent plans against you, search becomes a game. We formalize two-player, zero-sum, perfect-information games as search problems, define the minimax value that optimal play backs up through the game tree, and give the MINIMAX algorithm that computes it. Alpha–beta pruning then cuts the cost of that search roughly in half in the exponent without changing the answer, and a heuristic evaluation function plus a cutoff test turns the exact algorithm into a real-time player that copes with the horizon effect.\n",{"path":24931,"title":24932,"module":24904,"summary":24933},"\u002Fartificial-intelligence\u002Fsearch\u002Fgames-of-chance-and-imperfect-information","Games of Chance and Imperfect Information","Minimax and alpha–beta assume a deterministic game both players can see in full. Drop either assumption and search must change. This lesson adds chance nodes and the expectiminimax value for games with dice, then belief-state reasoning for partially observable games — Kriegspiel and card games — where averaging over clairvoyance both helps and misleads. It closes with the line from Deep Blue's alpha–beta to AlphaGo's learned evaluation and Monte Carlo tree search, and the provable-pruning and self-play research around each end of that story.\n",{"path":24935,"title":24936,"module":24904,"summary":24937},"\u002Fartificial-intelligence\u002Fsearch\u002Fconstraint-satisfaction","Constraint Satisfaction Problems","A constraint satisfaction problem replaces the black-box state with a factored one: variables, domains, and constraints. That structure supports inference before any search runs. This lesson defines the CSP on map coloring, Sudoku, and scheduling, then develops constraint propagation: node and arc consistency, the AC-3 algorithm that makes a whole network arc-consistent, and the way one deleted value cascades across the graph to prune impossible options ahead of search.\n",{"path":24939,"title":24940,"module":24904,"summary":24941},"\u002Fartificial-intelligence\u002Fsearch\u002Fcsp-search-and-structure","CSP Search and Structure","Propagation prunes a CSP but rarely finishes it, so we search. This lesson builds backtracking search over partial assignments and the general-purpose heuristics that make it fast — MRV, degree, least-constraining-value, forward checking, MAC, and intelligent backtracking. It then shows how the shape of the constraint graph controls difficulty: tree-structured problems fall in linear time, cutset conditioning handles the rest, and min-conflicts local search solves a million queens in a constant number of steps.\n",{"path":24943,"title":24944,"module":24904,"summary":24945},"\u002Fartificial-intelligence\u002Fsearch\u002Fsearch-under-uncertainty","Search Under Uncertainty","Classical search assumes the agent knows the state it is in and exactly what each action does. Drop the second assumption and a plan can no longer be a fixed sequence of actions. This lesson develops the first response: AND-OR search over nondeterministic actions, which returns a branching contingency plan rather than a straight line. We build it on the erratic vacuum world, show how OR nodes (the agent's choices) alternate with AND nodes (nature's outcomes), trace the recursion that finds a plan, and handle the case where the only solution is a cyclic \"try, try again.\"\n",{"path":24947,"title":24948,"module":24904,"summary":24949},"\u002Fartificial-intelligence\u002Fsearch\u002Fbelief-state-and-online-search","Belief-State and Online Search","When the agent cannot see the full state, a plan can no longer test where it actually is — it must reason over the set of states it might be in. This lesson develops belief-state search, from sensorless (conformant) planning that coerces an unknown world into a goal, through the predict-observe-update cycle of contingent planning with percepts, to online search in unknown environments, where the agent must act in order to learn. It closes with LRTA*, which refines its own heuristic as it explores, one step from reinforcement learning.\n",{"path":24951,"title":24952,"module":24953,"summary":24954},"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fpropositional-logic","Logical Agents and Propositional Logic","Logic and Planning","A knowledge-based agent keeps a store of sentences and acts by asking it what to do. To make \"asking\" mean something we need entailment — the relation $KB \\models \\alpha$ that holds when every model of the knowledge base is a model of the query. Propositional logic gives a syntax and a truth-table semantics for which entailment is decidable. This first part builds the foundations: the agent loop, the Wumpus World, models and entailment, the connectives and truth tables, theorem proving by refutation, and the resolution rule with its CNF conversion — a single complete inference procedure for all of propositional logic.\n",{"path":24956,"title":24957,"module":24953,"summary":24958},"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fpropositional-inference","Propositional Inference and Logical Agents","Model checking and resolution decide entailment, but both can blow up. This part turns propositional logic into a practical engine and a working agent. Horn clauses give linear-time forward and backward chaining — the basis of logic programming. DPLL and WalkSAT make satisfiability testing fast in the common case. Then we make the agent situated: time-indexed fluents, the frame problem and its solution by successor-state axioms, a hybrid agent that deduces a safe map and plans a route through it, and SATPlan, which finds a plan by asking a SAT solver for a satisfying model.\n",{"path":24960,"title":24961,"module":24953,"summary":24962},"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Ffirst-order-logic","First-Order Logic","Propositional logic can only say that facts hold; it cannot talk about the objects a fact is about, or state a rule once and have it cover every object. First-order logic fixes this by committing to a world of objects, relations, and functions. This first part builds the language from the ground up: the ontology it commits to, the model that gives a sentence a truth value, the syntax of terms and sentences, the two quantifiers with their standard mistakes, and equality.\n",{"path":24964,"title":24965,"module":24953,"summary":24966},"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Ffirst-order-logic-in-use","First-Order Logic in Use","With the language of first-order logic in hand, this part is about using it well. Database semantics trades expressive power for the convenience of a single intended model; higher-order logic shows what first-order logic gives up for decidability. Then we put the language to work: the Tell\u002FAsk interface, the kinship domain axiomatized from scratch, and the seven-step knowledge-engineering process applied to a digital circuit.\n",{"path":24968,"title":24969,"module":24953,"summary":24970},"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Finference-and-resolution","Inference in First-Order Logic","Propositional inference lifts to first-order logic once we can make terms match. Unification is that machinery: the algorithm that finds the substitution making two expressions identical, and the basis of generalized modus ponens. This first part builds the lifted inference rules and the two chaining algorithms they drive — forward chaining, the data-driven procedure behind production systems and Datalog, and backward chaining, the goal-driven procedure behind Prolog.\n",{"path":24972,"title":24973,"module":24953,"summary":24974},"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Ffirst-order-resolution","First-Order Resolution","Chaining is complete only for Horn knowledge bases. General first-order sentences — with disjunctive conclusions and negations — need a single sound and complete rule: resolution. This part converts arbitrary sentences to CNF by skolemizing away the existentials, lifts the resolution rule with unification, and proves entailment by refuting the negated goal. The result is the proof procedure Gödel's completeness theorem guarantees will find any entailment, together with the search strategies that make it usable.\n",{"path":24976,"title":24977,"module":24953,"summary":24978},"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fclassical-planning","Classical Planning","Classical planning represents a problem in a factored language, PDDL: states are sets of ground fluents, and actions are lifted schemas with a precondition and an effect. That structure turns planning into search — forward through states or backward through goals — and lets a program read heuristics straight off the schemas by relaxing the problem. This first part develops the representation, the two search directions, and the domain-independent heuristics that come from ignoring preconditions or delete lists.\n",{"path":24980,"title":24981,"module":24953,"summary":24982},"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fplanning-graphs-and-graphplan","Planning Heuristics and GraphPlan","Every relaxation heuristic can be inaccurate, and none can tell how far apart subgoals sit. The planning graph is a polynomial-size structure that does better: leveled off the problem, it yields admissible distance estimates and a record of which actions and fluents cannot coexist. This part builds the graph, reads heuristics from it, extracts plans with GraphPlan, and closes with the other classical approaches — SATPlan and partial-order planning — and the representational trade that makes all of it work.\n",{"path":24984,"title":24985,"module":24953,"summary":24986},"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fplanning-in-the-real-world","Planning and Acting in the Real World","Classical planning's clean theory rests on four assumptions: time is ignored, actions are atomic, the world is deterministic and fully observable, and the agent is alone. This first part drops the first two. We add durations and resource constraints — turning a plan into a schedule, solved by the critical-path method and, once resources contend, by NP-hard job-shop scheduling — and let a planner reason at multiple levels of abstraction through high-level actions and their angelic reachable sets.\n",{"path":24988,"title":24989,"module":24953,"summary":24990},"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fplanning-under-uncertainty","Planning Under Uncertainty","Classical planning assumed the world was deterministic, fully observable, and the agent alone. This part drops the last two assumptions. When the agent cannot see or predict the world, planning moves into belief-state space: sensorless plans that coerce the world into the goal without sensing, contingent plans that branch on what is sensed, and online agents that monitor and replan when execution diverges. Then we add other agents — joint plans, the coordination problem, and the conventions that let a team act without constant negotiation.\n",{"path":24992,"title":24993,"module":24953,"summary":24994},"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Fknowledge-representation","Knowledge Representation","First-order logic gives you the language; this lesson is about what to say in it. This first part builds the content: a general upper ontology from the top down, categories as first-class objects with taxonomies and inheritance, physical composition and the count-noun\u002Fmass-noun split, events and time reified through the event calculus, and belief modeled with modal logic — the machinery for representing the world an agent reasons about.\n",{"path":24996,"title":24997,"module":24953,"summary":24998},"\u002Fartificial-intelligence\u002Flogic-and-planning\u002Freasoning-systems-and-defaults","Reasoning Systems and Default Logic","Having represented the world, this part is about reasoning with it at scale. Semantic networks give a graphical notation with fast inheritance; description logics keep subsumption and classification tractable by design. Then we confront the fact that most useful rules hold only by default: circumscription and default logic give a logical account of nonmonotonic reasoning, and truth maintenance systems retract conclusions cleanly when the beliefs beneath them change.\n",{"path":25000,"title":25001,"module":25002,"summary":25003},"\u002Fartificial-intelligence\u002Funcertainty\u002Fprobability-and-bayes","Quantifying Uncertainty","Uncertainty","Logic breaks down in any domain where the rules have exceptions you cannot enumerate — the qualification problem. Probability replaces truth values with degrees of belief that obey Kolmogorov's axioms, and the full joint distribution becomes a knowledge base from which any query is answered by summing entries: marginalization, conditioning, and normalization. Independence factors that joint into smaller pieces — the first step toward a calculus of rational belief that an agent can actually compute with.\n",{"path":25005,"title":25006,"module":25002,"summary":25007},"\u002Fartificial-intelligence\u002Funcertainty\u002Fbayes-rule-and-naive-bayes","Bayes' Rule and Naive Bayes","Bayes' rule inverts a causal model into a diagnostic one, turning \"how a cause produces its symptoms\" into \"which cause explains what I observed.\" Ignoring the prior is the base-rate fallacy behind overconfident test results. Conditional independence then lets several pieces of evidence combine by multiplying likelihood ratios instead of building an exponential joint, giving the naive Bayes model and pointing directly at Bayesian networks.\n",{"path":25009,"title":25010,"module":25002,"summary":25011},"\u002Fartificial-intelligence\u002Funcertainty\u002Fbayesian-networks","Bayesian Networks","A Bayesian network is a directed acyclic graph of random variables in which each node carries a conditional probability table for itself given its parents. That structure factors the full joint distribution into a product of local terms, turning an exponential table into a linear one, and it makes the conditional independences of the domain explicit. We build the canonical burglary–alarm network, read compactness and d-separation off the graph, run exact inference by variable elimination, and, where that is intractable, estimate answers by sampling.\n",{"path":25013,"title":25014,"module":25002,"summary":25015},"\u002Fartificial-intelligence\u002Funcertainty\u002Finference-in-bayesian-networks","Bayesian Networks: Inference and Relational Models","When exact inference is intractable, sampling estimates the posterior instead: prior and rejection sampling, likelihood weighting, and Gibbs\u002FMCMC, whose error shrinks as one over the square root of the sample count. The same graphical idea then lifts from a fixed set of variables to whole populations — relational and open-universe probability models write dependencies once and unroll them over objects — and we close by placing probability against the rule-based, Dempster–Shafer, and fuzzy alternatives it displaced.\n",{"path":25017,"title":25018,"module":25002,"summary":25019},"\u002Fartificial-intelligence\u002Funcertainty\u002Freasoning-over-time","Probabilistic Reasoning over Time","A world that changes needs a state variable at every point in time. The Markov assumption cuts the dependence on history down to the previous slice, leaving a transition model and a sensor model that define a temporal Bayesian network. Four recursive tasks fall out — filtering, prediction, smoothing, and the most likely explanation — each a message passed along the sequence. We ground them in hidden Markov models and their matrix form, sketch the Kalman filter for continuous state, and reach dynamic Bayesian networks with particle filtering as the general approximate method.\n",{"path":25021,"title":25022,"module":25002,"summary":25023},"\u002Fartificial-intelligence\u002Funcertainty\u002Ftracking-and-data-association","Reasoning over Time: Tracking and Data Association","Dynamic Bayesian networks generalize HMMs and Kalman filters to arbitrarily many state variables per slice, and when exact inference blows up, particle filtering approximates the belief state with a population of weighted samples that propagate, reweight, and resample. Tracking several objects at once adds the data-association problem — which observation came from which object — whose combinatorics defeat any exact filter, so particle filters and MCMC keep many hypotheses alive. We close with SLAM and learned state-space models.\n",{"path":25025,"title":25026,"module":25002,"summary":25027},"\u002Fartificial-intelligence\u002Funcertainty\u002Fmaking-decisions","Making Decisions: Utility Theory","A rational agent chooses the action that maximizes expected utility — the probability of each outcome weighted by how much the agent wants it. We derive the utility function from six axioms on preferences, so maximizing expected utility is forced by consistency rather than assumed; look at risk aversion in the utility-of-money curve; package one-shot choices into decision networks; and quantify what an observation is worth with the value of information.\n",{"path":25029,"title":24618,"module":25002,"summary":25030},"\u002Fartificial-intelligence\u002Funcertainty\u002Fmarkov-decision-processes","When an agent must act repeatedly in a stochastic world, a fixed plan is useless — it needs a policy, an action for every state. The Markov decision process makes this precise with a transition model, a reward, and a discount factor; the Bellman equation characterizes the optimal state utilities, and value iteration and policy iteration solve it. Partial observability lifts the problem to belief states, and bandits, Monte-Carlo tree search, and scalable POMDP solvers extend it — this is the model-known half of reinforcement learning.\n",{"path":25032,"title":25033,"module":25002,"summary":25034},"\u002Fartificial-intelligence\u002Funcertainty\u002Fdecision-networks-and-game-theory","Decision Analysis: Multi-Attribute Utility and Decision Networks","Decision analysis takes the single-agent utility framework and makes it practical: utility over several attributes, dominance and additive value functions, influence diagrams that fold Bayesian networks together with decision and utility nodes, and the value of information that tells an agent which questions are worth asking. Structure in an agent's preferences — dominance, preferential and utility independence — collapses an exponential utility table into a few one-dimensional functions, the same move that made Bayesian networks compact.\n",{"path":25036,"title":25037,"module":25002,"summary":25038},"\u002Fartificial-intelligence\u002Funcertainty\u002Fgame-theory-and-mechanism-design","Game Theory and Mechanism Design","When outcomes depend on other rational agents, single-agent utility maximization no longer suffices. Game theory studies decisions among agents — normal-form games, dominant strategies, Nash and maximin equilibria, and repeated games — and mechanism design runs the logic backwards, engineering rules (auctions, VCG) so that self-interested play produces a good collective outcome. Algorithmic game theory then asks whether equilibria can be computed, what selfishness costs society, and how the mechanisms deployed at internet scale actually behave.\n",{"path":25040,"title":25041,"module":25042,"summary":25043},"\u002Fartificial-intelligence\u002Flearning\u002Flearning-from-examples","Learning from Examples","Learning","An agent that improves with experience does not need its designer to anticipate every situation. Inductive learning takes that ambition and narrows it to one tractable problem: from labelled input-output pairs, recover a function that predicts the output for inputs never seen. This first part builds the foundation around a single organizing question — generalization — through decision trees and information gain, and the training\u002Fvalidation\u002Ftest discipline for evaluating and choosing hypotheses. A second part takes up the theory of learning and the main model families.\n",{"path":25045,"title":25046,"module":25042,"summary":25047},"\u002Fartificial-intelligence\u002Flearning\u002Ftheory-and-model-families","The Theory of Learning and Model Families","Cross-validation measures generalization but does not explain it. This part supplies the theory — PAC learning, sample complexity, and the VC dimension — that says when a hypothesis consistent with enough data is probably approximately correct, and why an unrestricted hypothesis space can never generalize. It then surveys the model families a practitioner reaches for: linear regression and gradient descent, the perceptron and logistic regression, support vector machines and the kernel trick, and ensembles by bagging and boosting — closing with what deep learning changed about the classical picture.\n",{"path":25049,"title":25050,"module":25042,"summary":25051},"\u002Fartificial-intelligence\u002Flearning\u002Fprobabilistic-learning","Learning Probabilistic Models","A [Bayesian network](\u002Fartificial-intelligence\u002Funcertainty\u002Fbayesian-networks) is useless until its numbers are filled in, and those numbers come from data. This first part casts learning itself as probabilistic inference: hypotheses carry a prior, data update it to a posterior, and predictions average over what remains. From that frame fall the standard estimators — maximum likelihood by counting, MAP with a conjugate prior, full Bayesian updating — for the case where every variable is observed. A second part takes up the harder case of hidden variables and the EM algorithm.\n",{"path":25053,"title":25054,"module":25042,"summary":25055},"\u002Fartificial-intelligence\u002Flearning\u002Fexpectation-maximization","Learning with Hidden Variables: The EM Algorithm","Complete data can be learned by counting; real data usually hide some variables — the disease behind the symptoms, the cluster behind the points. This part develops the expectation-maximization algorithm, which learns those models by alternating an expected completion of the missing data with a re-estimation of the parameters. It works the idea through mixtures of Gaussians, Bayesian networks, and hidden Markov models, proves the monotone-likelihood guarantee from the evidence lower bound, and traces the line from EM to variational inference and the variational autoencoder.\n",{"path":25057,"title":24041,"module":25042,"summary":25058},"\u002Fartificial-intelligence\u002Flearning\u002Freinforcement-learning","Reinforcement learning is an MDP with the model unknown: the agent knows neither how its actions move the world nor which states are rewarded, and must recover good behaviour from experienced transitions and rewards alone. This first part builds the classical tabular theory — passive learning (fix a policy, learn its value, by direct estimation, adaptive dynamic programming, and temporal differences) and active learning (choose actions, trade exploration against exploitation, and learn control with Q-learning and SARSA). A second part lifts it off the lookup table with function approximation and policy search.\n",{"path":25060,"title":25061,"module":25042,"summary":25062},"\u002Fartificial-intelligence\u002Flearning\u002Fgeneralization-and-policy-search","Reinforcement Learning: Generalization and Policy Search","Tabular reinforcement learning stores one number per state, which is hopeless for backgammon or chess. This part lifts RL off the lookup table with function approximation, so that updating one state generalizes to related ones, then turns to policy search — representing and optimizing the policy directly, up to the REINFORCE policy gradient and correlated sampling. It closes with the bridge to deep reinforcement learning (deep Q-networks, actor-critic, PPO), the classic applications, and the hand-off to the dedicated RL subject.\n",{"path":25064,"title":25065,"module":25042,"summary":25066},"\u002Fartificial-intelligence\u002Flearning\u002Fknowledge-in-learning","Knowledge in Learning","Pure induction learns a function from labelled examples while knowing almost nothing to begin with. This first part brings prior knowledge into the loop by recasting learning as logical inference — hypotheses, examples, and classifications as sentences. It develops current-best-hypothesis search, the version space and its general\u002Fspecific boundary maintained by candidate elimination, and states the three entailment constraints that fix how background knowledge enters. A second part builds the three knowledge-based methods those constraints define.\n",{"path":25068,"title":25069,"module":25042,"summary":25070},"\u002Fartificial-intelligence\u002Flearning\u002Fknowledge-based-learning-methods","Knowledge-Based Learning: EBL, Relevance, and ILP","Once learning is cast as logical inference, three methods follow from the three ways prior knowledge can enter. Explanation-based learning generalizes a single example by explaining it with the domain theory, gaining speed but nothing new. Relevance-based learning uses determinations to shrink the hypothesis space and converge from fewer examples. Inductive logic programming learns genuinely new first-order rules — top-down with FOIL, bottom-up by inverting resolution, even inventing new predicates — and connects to modern statistical relational and neuro-symbolic learning.\n",{"path":25072,"title":25073,"module":25074,"summary":25075},"\u002Fartificial-intelligence\u002Ffrontiers\u002Fvision-and-perception","Vision and Perception","Frontiers","Perception connects an agent to the physical world. We follow one modality — vision — from the physics of image formation (the pinhole camera, perspective projection, lenses, shading, color) through the early operations that turn a pixel array into edges, texture, and motion, and into recognition by appearance. The recurring problem is inversion: a camera collapses a 3-D world onto a 2-D grid, and an agent that wants to act must build the scene back up. Rebuilding the scene is the subject of the companion lesson.\n",{"path":25077,"title":25078,"module":25074,"summary":25079},"\u002Fartificial-intelligence\u002Ffrontiers\u002Freconstructing-the-3d-world","Vision: Reconstructing the 3D World","A camera collapses a three-dimensional world onto a flat grid; this lesson inverts that collapse. We build the camera projection matrix (intrinsics and extrinsics), triangulate a point from two views, then work through the toolbox of depth cues — motion parallax, binocular stereopsis, multiple views, texture, shading, and contour — that turn an ambiguous image back into a scene. We add structural recognition (pictorial-structure \"cardboard people\"), the task-driven use of vision in cars and robots, and the shift from hand-built pipelines to learned deep-vision networks.\n",{"path":25081,"title":25082,"module":25074,"summary":25083},"\u002Fartificial-intelligence\u002Ffrontiers\u002Frobotics","Robotics","A robot is an agent with a body: sensors that read the physical world and effectors that push back on it. This lesson grounds the abstract AI machinery in that body. We build up the hardware (range finders, proprioception, degrees of freedom), then cast perception as probabilistic filtering — the kinematic motion and sensor models, Monte Carlo localization, the extended Kalman filter, and simultaneous localization and mapping (SLAM). The companion lesson takes the estimated pose forward into planning and control.\n",{"path":25085,"title":25086,"module":25074,"summary":25087},"\u002Fartificial-intelligence\u002Ffrontiers\u002Frobot-planning-and-control","Robotics: Planning and Control","A robot that knows where it is still has to decide how to move, and then make a slipping, sensing-imperfect body actually go there. This lesson takes the pose estimate forward: planning motion in configuration space with cell decomposition and sampling-based roadmaps (PRMs and RRTs), planning under uncertainty with most-likely-state and online replanning, closing the loop with P\u002FPD\u002FPID control and potential fields, and finally the software architectures — subsumption, three-layer, and pipeline — that assemble it all, plus the learning-based turn in modern robotics.\n",{"path":25089,"title":25090,"module":25074,"summary":25091},"\u002Fartificial-intelligence\u002Ffrontiers\u002Fnatural-language-in-ai","Natural Language for AI Agents","Language is how agents acquire the knowledge already written down and how they communicate with the humans they serve. This lesson gives the classical AI account of language as a source of information: n-gram language models and the information-seeking tasks built on them — text classification, information retrieval (BM25, the inverted index, PageRank), and information extraction with finite-state templates and hidden Markov models. Throughout, we point to the dedicated NLP subject for the modern deep-learning treatment; the companion lesson takes up grammar, translation, and speech.\n",{"path":25093,"title":25094,"module":25074,"summary":25095},"\u002Fartificial-intelligence\u002Ffrontiers\u002Fnlp-grammar-translation-and-speech","Language for AI Agents: Grammar, Translation, and Speech","N-gram models see only a local window; they cannot say why \"black dog\" is well-formed English and \"dog black\" is not, because that is a fact about structure. This lesson takes up structure: phrase-structure and probabilistic context-free grammars, syntactic analysis by chart parsing and CYK, augmented grammars and compositional semantics, then the two major statistical successes — machine translation and speech recognition — cast as noisy-channel problems. It closes with the bridge from n-grams to transformers and where the classical account sits relative to modern NLP.\n",{"path":25097,"title":25098,"module":25074,"summary":25099},"\u002Fartificial-intelligence\u002Ffrontiers\u002Fphilosophy-and-future","Philosophy, Ethics, and the Future of AI","Two questions have shadowed the field since its founding: can machines act intelligently (weak AI), and can they really think (strong AI)? We work through Turing's objections and their rebuttals — the arguments from disability, mathematics, and informality — then the strong-AI debate: the mind-body problem, functionalism and the brain prosthesis, Searle's Chinese Room and the systems reply, and consciousness and qualia. The companion lesson turns from what AI can do to what it should, and closes the course.\n",{"path":25101,"title":25102,"module":25074,"summary":25103},"\u002Fartificial-intelligence\u002Ffrontiers\u002Fai-ethics-and-future","The Ethics and Future of AI","Having asked whether machines can act intelligently and really think, we turn to whether we should build them at all. This lesson works through the six ethical risks — lost jobs, autonomous weapons, surveillance and privacy, biased decisions, the safety of superintelligence, and the erosion of accountability — then the value-alignment problem in the LLM era, and where the classical agent components could go next. It closes the course by tying search, logic, probability, and learning into a single picture of intelligence as rational agency.\n",{"path":25105,"title":25106,"module":6,"summary":6},"\u002Fartificial-intelligence","Artificial Intelligence",{"path":25108,"title":25109,"module":25110,"summary":25111},"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fnuclear-constituents-nuclide-chart","Nuclear Composition and Ground-State Properties","Nuclear Properties","The nucleus is a bound assembly of Z protons and N neutrons packed to a radius R = R0 A^(1\u002F3) at a nearly constant density of about 10^17 kg\u002Fm^3. We fix the vocabulary of nuclides, derive nuclear size from mirror-nuclide and electron-scattering data, read the binding-energy-per-nucleon curve, and model it with the liquid-drop semiempirical mass formula.\n",{"path":25113,"title":25114,"module":25110,"summary":25115},"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fnuclear-size-charge-distributions","Nuclear Size, Shape, and Charge Distributions","Elastic electron scattering resolves the nucleus by its de Broglie wavelength. The measured cross section is the Mott point-charge cross section modulated by a form factor, and that form factor is the Fourier transform of the charge density. Diffraction minima fix the radius, the small-angle slope fixes the mean-square radius, and the fitted Woods-Saxon profile gives a central density and a skin thickness. Mirror-nucleus Coulomb energies, muonic-atom X-rays, and optical isotope shifts give independent radii that all track R = R0 A^(1\u002F3).\n",{"path":25117,"title":25118,"module":25110,"summary":25119},"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fnuclear-masses-binding-energy","Nuclear Masses, Mass Excess, and Separation Energies","The atomic mass unit fixes the scale, and the mass excess collects the small binding-driven deviation from the integer mass number. Penning-trap cyclotron frequencies now measure masses to parts in a billion, and every decay and reaction Q-value is a difference of these masses. One- and two-nucleon separation energies read the binding difference between neighbouring nuclides directly, showing the even-odd pairing stagger and the sharp drops at magic numbers, and their vanishing marks the neutron and proton drip lines that bound the chart of the nuclides.\n",{"path":25121,"title":25122,"module":25110,"summary":25123},"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fsemi-empirical-mass-formula","The Semi-Empirical Mass Formula and the Valley of Stability","Five physical terms reproduce nuclear binding across the chart: a volume term from saturation, a surface term from the deficit of edge neighbours, a Coulomb term from the electrostatic self-energy of a charged sphere, an asymmetry term from the Pauli cost of unequal proton and neutron filling, and a pairing term. The formula is quadratic in Z at fixed A, so isobars lie on a mass parabola whose minimum sets the most stable charge and whose slope dictates the direction of beta decay. The same competition between surface and Coulomb energy defines the fissility parameter and the onset of fission.\n",{"path":25125,"title":25126,"module":25110,"summary":25127},"\u002Fnuclear-physics\u002Fnuclear-properties\u002Fnuclear-moments-multipoles","Nuclear Spin, Magnetic Dipole, and Electric Quadrupole Moments","The ground state of a nucleus carries a definite spin and parity, a magnetic dipole moment of order the nuclear magneton, and, when its spin exceeds one-half, an electric quadrupole moment that measures its shape. The single-particle Schmidt lines predict the magnetic moment of an odd-A nucleus from the last unpaired nucleon, and the measured moments fall between them. The quadrupole moment distinguishes prolate from oblate deformation, and hyperfine structure is the experimental handle that fixes the spin and the moments from an atomic spectrum.\n",{"path":25129,"title":25130,"module":25131,"summary":25132},"\u002Fnuclear-physics\u002Fnuclear-force-deuteron\u002Fnuclear-force-shell-overview","The Nuclear Force and the Shell Model","The Nuclear Force","The strong force between nucleons is short-range, charge-independent, saturated, and repulsive at its core, about a hundred times stronger than Coulomb. Yukawa explained it as an exchange of massive mesons, tying the force's range to the meson mass through the uncertainty principle. Layered on top, an independent-particle shell model with strong spin-orbit coupling reproduces the magic numbers 2, 8, 20, 28, 50, 82, 126.\n",{"path":25134,"title":25135,"module":25131,"summary":25136},"\u002Fnuclear-physics\u002Fnuclear-force-deuteron\u002Fthe-deuteron","The Deuteron and the Tensor Force","The deuteron is the only bound two-nucleon state: one shallow level at 2.22 MeV, no excited states. A square-well fit fixes a depth near 35 MeV over a 2 fm range, yet the wavefunction leaks so far past the edge that most of the probability lies outside the force. Its spin-1 ground state, magnetic moment close to the sum of the free-nucleon moments, and small but nonzero electric quadrupole moment together force a D-state admixture and a non-central tensor force.\n",{"path":25138,"title":25139,"module":25131,"summary":25140},"\u002Fnuclear-physics\u002Fnuclear-force-deuteron\u002Fnucleon-nucleon-scattering","Nucleon-Nucleon Scattering and the Interaction's Structure","Scattering probes the nuclear force above threshold. Partial-wave analysis reduces low-energy data to a single s-wave phase shift, and the effective-range expansion packages that into a scattering length and an effective range. The triplet channel binds (the deuteron) while the singlet is only virtual, which together explain the anomalously large free neutron-proton cross section. Comparing pp, nn, and np results establishes charge symmetry and charge independence, and polarization experiments expose the spin-orbit and tensor pieces.\n",{"path":25142,"title":25143,"module":25131,"summary":25144},"\u002Fnuclear-physics\u002Fnuclear-force-deuteron\u002Fmeson-theory-isospin","Meson Exchange, the Yukawa Potential, and Isospin","Yukawa's massive-field propagator turns the range of the nuclear force into a meson mass: the exchanged quantum's Compton wavelength is the range. One-pion exchange fixes the long-range tail, complete with the tensor structure the deuteron demanded, while heavier mesons build the intermediate attraction and the repulsive core. Charge independence becomes an isospin symmetry, the force is diagonalized by the total isospin through a tau-dot-tau interaction, and the whole picture sits inside QCD as a residual color force between color-neutral nucleons.\n",{"path":25146,"title":25147,"module":25148,"summary":25149},"\u002Fnuclear-physics\u002Fnuclear-models\u002Ffermi-gas-model","The Fermi Gas Model","Nuclear Models","Treating the nucleus as two degenerate Fermi gases of protons and neutrons confined in a common well fixes the Fermi momentum near 250 MeV\u002Fc and the Fermi energy near 33 MeV from the nuclear density alone. The average kinetic energy per nucleon is about 20 MeV, the well depth is the Fermi energy plus the separation energy, and unequal proton and neutron Fermi levels reproduce the asymmetry term of the mass formula.\n",{"path":25151,"title":25152,"module":25148,"summary":25153},"\u002Fnuclear-physics\u002Fnuclear-models\u002Fliquid-drop-collective-coordinates","The Liquid-Drop Model and Collective Deformation","Deforming a charged liquid drop into a spheroid raises its surface energy and lowers its Coulomb energy; the two effects compete through the deformation parameter to set a stability minimum and a fission barrier. The ratio of Coulomb to twice the surface energy is the fissility Z-squared over A, which crosses one near 49 and marks the point where the sphere is unstable. The same surface tension that restores small deformations quantizes into collective vibrations, carrying the static mass formula into dynamic collective motion.\n",{"path":25155,"title":25156,"module":25148,"summary":25157},"\u002Fnuclear-physics\u002Fnuclear-models\u002Fshell-model-single-particle","The Shell Model: Single-Particle States and Spin-Orbit Coupling","A harmonic-oscillator well reproduces the first three magic numbers but fails above twenty; adding a strong inverted spin-orbit term that drives the stretched j equals l plus one-half level down closes the gaps at 28, 50, 82, and 126. The filled shells couple to zero, so the last unpaired nucleon fixes the ground-state spin and parity, and its single-particle magnetic moment falls on the Schmidt lines. Configuration mixing sets the limits of the extreme single-particle model.\n",{"path":25159,"title":25160,"module":25148,"summary":25161},"\u002Fnuclear-physics\u002Fnuclear-models\u002Fcollective-model-rotations-vibrations","The Collective Model: Rotations, Vibrations, and Deformed Nuclei","Deformed nuclei rotate with energies proportional to I times I plus one, giving the ground-state band its characteristic level ratios, while near-spherical nuclei vibrate in quantized surface phonons that build one- and two-phonon multiplets. The Nilsson model tracks single-particle levels as the well deforms, moments of inertia fall between the rigid and irrotational limits, backbending marks the sudden alignment of a broken pair, and giant resonances are the bulk dipole and quadrupole modes of the whole nucleus.\n",{"path":25163,"title":25164,"module":25165,"summary":25166},"\u002Fnuclear-physics\u002Fradioactive-decay\u002Fdecay-law-modes","Radioactivity and Decay Modes","Radioactive Decay","Unstable nuclei decay at a rate proportional to how many remain, giving the exponential law N(t) = N0 e^(-lambda t) with half-life t = 0.693\u002Flambda. We work through the three common modes: alpha decay as Coulomb-barrier tunneling with the Geiger-Nuttall rule, beta decay whose continuous spectrum demands the neutrino, and gamma de-excitation, and follow a decay chain across the chart of nuclides.\n",{"path":25168,"title":25169,"module":25165,"summary":25170},"\u002Fnuclear-physics\u002Fradioactive-decay\u002Fdecay-kinetics-equilibrium","Serial Decay, the Bateman Equations, and Radioactive Equilibrium","A radioactive parent that decays into a radioactive daughter obeys a coupled pair of rate equations whose solution is the Bateman formula. Depending on the half-life ordering the chain settles into secular equilibrium (equal activities), transient equilibrium (a fixed activity ratio), or no equilibrium. Constant production under irradiation drives the activity toward a saturation value equal to the production rate, competing decay modes split the total decay constant into partial constants, and the natural decay series in secular equilibrium underpin radiometric dating.\n",{"path":25172,"title":25173,"module":25174,"summary":25175},"\u002Fnuclear-physics\u002Falpha-decay\u002Falpha-decay-gamow-theory","Alpha Decay and the Gamow Theory of Tunneling","Alpha Decay","The alpha Q-value turns positive above mass number 150 because the emitted helium-4 is exceptionally tightly bound. Emission proceeds by quantum tunneling through the Coulomb barrier: a WKB integral from the nuclear surface to the outer turning point gives the Gamow factor, and multiplying its penetrability by the assault frequency yields half-lives spanning more than twenty orders of magnitude. The leading term reproduces the Geiger-Nuttall relation, log t½ proportional to the daughter charge over the square root of Q.\n",{"path":25177,"title":25178,"module":25174,"summary":25179},"\u002Fnuclear-physics\u002Falpha-decay\u002Falpha-fine-structure-hindrance","Fine Structure, Angular Momentum, and Hindrance Factors","A single parent emits several alpha groups of slightly different energy, each feeding a distinct level of the daughter, so the alpha spectrum maps the daughter's low-lying states. Emission with orbital angular momentum L raises the barrier by a centrifugal term and is allowed only when angular-momentum and parity selection rules permit. Comparing the measured partial half-life to the Gamow estimate defines a hindrance factor near unity for even-even ground-state transitions and large for odd-A decays that must rearrange the unpaired nucleon.\n",{"path":25181,"title":25182,"module":25183,"summary":25184},"\u002Fnuclear-physics\u002Fbeta-decay\u002Fbeta-decay-energetics-neutrino","Beta Decay Energetics and the Neutrino","Beta Decay and the Weak Interaction","Beta decay converts a neutron into a proton or the reverse, adjusting Z at fixed A along an isobaric mass parabola. We write the three processes (beta-minus, beta-plus, electron capture), reduce every Q-value to a difference of neutral atomic masses, and read the continuous electron spectrum as the fingerprint of a third, nearly massless particle. Pauli's neutrino, its detection by Reines and Cowan, and the endpoint bound on its mass close the lesson.\n",{"path":25186,"title":25187,"module":25183,"summary":25188},"\u002Fnuclear-physics\u002Fbeta-decay\u002Ffermi-theory-beta-decay","Fermi's Theory: Kurie Plots and ft Values","Fermi treated beta decay as a point-contact weak transition and read its rate from the golden rule. The electron spectrum then follows from phase space and the Coulomb Fermi function; the Kurie plot straightens it to a line whose intercept is the endpoint. Integrating the spectrum gives the comparative half-life ft, whose logarithm sorts transitions into superallowed, allowed, and forbidden classes governed by the Fermi and Gamow-Teller selection rules.\n",{"path":25190,"title":25191,"module":25183,"summary":25192},"\u002Fnuclear-physics\u002Fbeta-decay\u002Fweak-interaction-parity-violation","The Weak Interaction and Parity Violation","Beta decay violates mirror symmetry. The Wu experiment on polarized cobalt-60 showed electrons emitted preferentially against the nuclear spin, a pseudoscalar correlation forbidden if parity were conserved. The result fixes the weak charged current as left-handed V minus A, forces neutrinos to be left-handed and antineutrinos right-handed (measured by Goldhaber), and places beta decay within the electroweak theory as W-boson exchange turning a down quark into an up quark.\n",{"path":25194,"title":25195,"module":25183,"summary":25196},"\u002Fnuclear-physics\u002Fbeta-decay\u002Fdouble-beta-decay-neutrino-mass","Double Beta Decay and Neutrino Mass","For even-A isobars the pairing term splits the mass parabola into two curves, and a handful of even-even nuclides sit below their odd-odd neighbor yet above the next even-even one: single beta decay is forbidden but second-order double beta decay is allowed. The two-neutrino mode is a standard-model process with the longest measured lifetimes in nature; the neutrinoless mode would require the neutrino to be its own antiparticle and its rate measures the effective Majorana mass, the sharpest probe of the absolute neutrino mass scale.\n",{"path":25198,"title":25199,"module":25200,"summary":25201},"\u002Fnuclear-physics\u002Fgamma-decay\u002Fgamma-multipole-radiation","Multipole Radiation and Selection Rules","Gamma Decay","Gamma decay carries a nucleus from an excited state to a lower one by emitting a photon of definite angular momentum and parity. We correct the photon energy for nuclear recoil, expand the radiation field into electric and magnetic multipoles, and read off how the transition rate collapses with each increase in multipole order. The Weisskopf single-particle estimates set the scale, and angular-momentum and parity conservation fix which multipole dominates.\n",{"path":25203,"title":25204,"module":25200,"summary":25205},"\u002Fnuclear-physics\u002Fgamma-decay\u002Finternal-conversion-isomers","Internal Conversion and Isomers","A nucleus can shed excitation energy without emitting a photon by handing it directly to an atomic electron. We define the internal-conversion coefficient, trace its growth with atomic number, multipole order, and decreasing energy, and treat the electron-only E0 transitions and internal pair formation. When the lowest allowed multipole is high and the energy low, the gamma rate falls so far that the excited state survives as a metastable isomer.\n",{"path":25207,"title":25208,"module":25200,"summary":25209},"\u002Fnuclear-physics\u002Fgamma-decay\u002Fangular-correlations-mossbauer","Angular Correlations and the Mössbauer Effect","Two gammas emitted in cascade are not independent in direction: detecting the first selects magnetic substates of the intermediate level and makes the second anisotropic, so the correlation function fixes the intermediate spin. The same nuclear resonance that recoil normally destroys is recovered when the emitter is locked in a lattice, giving the Mössbauer effect and its part-in-a-trillion resolution of isomer shifts and hyperfine fields.\n",{"path":25211,"title":25212,"module":25213,"summary":25214},"\u002Fnuclear-physics\u002Fnuclear-reactions\u002Freaction-kinematics-cross-sections","Nuclear Reactions, Fission, and Fusion","Nuclear Reactions","A nuclear reaction X(x, y)Y is governed by its Q value and its cross section, the effective target area for a given process. Splitting the curve of binding energy near iron in either direction releases energy: fission of heavy nuclei by neutron capture and a chain reaction, and fusion of light nuclei that powers the Sun and needs Lawson's density-confinement criterion to be practical.\n",{"path":25216,"title":25217,"module":25213,"summary":25218},"\u002Fnuclear-physics\u002Fnuclear-reactions\u002Fcompound-nucleus-resonances","The Compound Nucleus and Resonance Reactions","Low-energy reactions proceed through a long-lived intermediate state whose decay forgets how it formed. Bohr's independence hypothesis factorizes the cross section into a formation step and a branching ratio, an isolated level gives the single-level Breit-Wigner line shape with total width Γ tied to the lifetime by Γτ = ħ, and at high excitation overlapping levels merge into a statistical continuum described by evaporation spectra and the Hauser-Feshbach average.\n",{"path":25220,"title":25221,"module":25213,"summary":25222},"\u002Fnuclear-physics\u002Fnuclear-reactions\u002Fdirect-reactions-optical-model","Direct Reactions and the Optical Model","A complex optical potential replaces the many-body target by a single particle moving in an average field whose imaginary part removes flux into non-elastic channels, reproducing the diffraction pattern of elastic scattering. Direct reactions bypass the compound nucleus, transferring a nucleon in one step: stripping and pickup deposit or remove a single nucleon, the angle of the first peak in the distorted-wave angular distribution fixes the transferred orbital angular momentum, and its magnitude gives the spectroscopic factor.\n",{"path":25224,"title":25225,"module":25226,"summary":25227},"\u002Fnuclear-physics\u002Ffission\u002Ffission-barrier-dynamics","The Fission Barrier and Fragment Energetics","Nuclear Fission","Fission is the large-amplitude collective deformation of a heavy nucleus into two fragments. The liquid-drop model sets a barrier from the competition between rising surface energy and falling Coulomb energy under quadrupole deformation, with the fissility parameter Z²\u002FA measuring how close a nucleus is to instability. Bohr-Wheeler theory separates spontaneous from neutron-induced fission, the fragment mass yield is double-humped and asymmetric, about 200 MeV is released per event, and shell corrections add a second minimum that produces fission isomers.\n",{"path":25229,"title":25230,"module":25226,"summary":25231},"\u002Fnuclear-physics\u002Ffission\u002Fchain-reactions-reactor-physics","Chain Reactions and Reactor Physics","A self-sustaining chain reaction is a fixed point of neutron bookkeeping: the multiplication factor k counts the neutrons in one generation per neutron in the last, and criticality is k = 1. The four-factor formula tracks a neutron through fast fission, resonance escape, thermal utilization, and reproduction; moderation slows fission neutrons to the thermal energies where the fission cross section is largest; and the small delayed-neutron fraction sets the timescale that makes a reactor controllable. Breeding converts fertile U-238 and Th-232 into new fissile fuel.\n",{"path":25233,"title":25234,"module":25235,"summary":25236},"\u002Fnuclear-physics\u002Ffusion-nucleosynthesis\u002Ffusion-reactions-confinement","Fusion Reactions and Confinement","Fusion and Nucleosynthesis","Light nuclei release energy when they fuse because binding per nucleon rises steeply toward the iron peak, but the Coulomb barrier suppresses the rate at reactor temperatures. The thermonuclear rate is a convolution of the Maxwell distribution with the tunneling probability, sharply peaked at the Gamow energy. The deuterium-tritium reaction has the lowest barrier and largest cross section; sustained energy gain requires the Lawson triple product of density, temperature, and confinement time, reached by magnetic or inertial confinement.\n",{"path":25238,"title":25239,"module":25235,"summary":25240},"\u002Fnuclear-physics\u002Ffusion-nucleosynthesis\u002Fstellar-nucleosynthesis","Stellar Nucleosynthesis","Main-sequence stars burn hydrogen to helium through the proton-proton chain and the CNO cycle, both releasing 26.7 MeV per helium nucleus. Helium burning bridges the mass-5 and mass-8 gaps by the triple-alpha process through the Beryllium-8 and Hoyle resonances, and successive carbon-to-silicon burning stages climb to the iron peak, where fusion stops. The elements beyond iron are built by slow and rapid neutron capture, and the solar neutrino flux confirms the reactions directly.\n",{"path":25242,"title":25243,"module":25235,"summary":25244},"\u002Fnuclear-physics\u002Ffusion-nucleosynthesis\u002Fbig-bang-nucleosynthesis","Big-Bang Nucleosynthesis","In the first three minutes the expanding universe forged the light elements. The weak interaction froze the neutron-to-proton ratio near one in six when the reaction rate fell below the expansion rate, and free-neutron decay lowered it to about one in seven before the deuterium bottleneck broke. Almost every surviving neutron ended in helium-4, fixing the primordial helium mass fraction near 0.25, with trace deuterium, helium-3, and lithium-7. The deuterium abundance measures the cosmic baryon density.\n",{"path":25246,"title":25247,"module":25248,"summary":25249},"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fcharged-particle-stopping-power","Stopping Power and the Range of Charged Particles","Radiation and Applications","A heavy charged particle loses energy in a dense sequence of small Coulomb collisions with atomic electrons, at a rate the Bethe-Bloch formula fixes from the particle's charge and speed and the medium's electron density and mean excitation energy. The rate scales as the inverse square of the speed, so most energy is deposited at the end of the track in the Bragg peak, and integrating the reciprocal rate gives a sharp range. Electrons differ: they also radiate, and above a critical energy bremsstrahlung dominates. Fast particles above the phase velocity of light in the medium emit Cherenkov radiation.\n",{"path":25251,"title":25252,"module":25248,"summary":25253},"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fphoton-neutron-interactions","Interactions of Photons and Neutrons","Photons are removed from a beam in single events, so their intensity falls exponentially with a linear attenuation coefficient built from three processes: the photoelectric effect at low energy, Compton scattering at intermediate energy, and pair production above twice the electron rest energy, each with its own atomic-number and energy dependence. Neutrons carry no charge and interact only with nuclei, moderating by elastic scattering and being captured with a cross section that rises as one over speed away from resonances.\n",{"path":25255,"title":25256,"module":25248,"summary":25257},"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fradiation-detectors","Radiation Detectors and Nuclear Spectroscopy","Every detector converts the energy a radiation deposits into a measurable electrical signal. Gas counters read the ionization directly, in three operating regions set by the applied voltage; scintillators convert the energy to light read out by a photomultiplier; semiconductor detectors collect electron-hole pairs and give the best energy resolution because so many carriers are made per event. The resolution is governed by the number of independent charge carriers, and the pulse-height spectrum of a gamma line shows a full-energy photopeak, a Compton continuum with its edge, and escape peaks.\n",{"path":25259,"title":25260,"module":25248,"summary":25261},"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fdosimetry-radiation-biology","Dosimetry, Radiation Biology, and Protection","Absorbed dose is the energy deposited per unit mass, measured in gray. Equal absorbed doses do unequal biological damage because densely ionizing radiation deposits its energy along short tracks: weighting the dose by a radiation factor gives the equivalent dose, and weighting by tissue sensitivity gives the effective dose, both in sieverts. Deterministic effects have a threshold and a severity that grows with dose; stochastic effects are assumed to follow a linear-no-threshold probability. Natural background dominates the dose to the population, and protection rests on time, distance, and shielding.\n",{"path":25263,"title":25264,"module":25248,"summary":25265},"\u002Fnuclear-physics\u002Fradiation-matter-applications\u002Fnuclear-applications-dating-medicine","Applications — Dating, Analysis, and Nuclear Medicine","Charged particles lose energy continuously and stop at a well-defined range with a Bragg peak, while gamma rays are attenuated exponentially. These interactions define radiation detectors and dosimetry (gray and sievert) and drive the applications: neutron activation analysis, magnetic resonance imaging, PET, and radiometric dating with carbon-14 and long-lived rock clocks.\n",{"path":25267,"title":25268,"module":6,"summary":6},"\u002Fnuclear-physics","Nuclear Physics",{"path":25270,"title":25271,"module":21235,"summary":25272},"\u002Fnatural-language-processing\u002Ffoundations\u002Fwhat-is-nlp","What Is Natural Language Processing?","Natural language processing is the computational treatment of human language: reading it, representing it, and generating it. We set up why the problem is hard — ambiguity at every level, from sound to intent — trace the field from ELIZA's pattern-matching through statistical methods to today's neural models, lay out the linguistic levels and task families the course covers, and fix the vocabulary of tokens, types, and corpora the rest of the notes rely on.\n",{"path":25274,"title":25275,"module":21235,"summary":25276},"\u002Fnatural-language-processing\u002Ffoundations\u002Fregex-and-text-normalization","Regular Expressions and Text Normalization","Before any model touches text, the text has to be found and cleaned. Regular expressions give an algebra for describing string patterns; tokenization, case folding, and stemming turn raw characters into the units a model counts; and byte-pair encoding builds a subword vocabulary that spells out any word. Measuring how far apart two strings are — minimum edit distance — is the next lesson.\n",{"path":25278,"title":25279,"module":21235,"summary":25280},"\u002Fnatural-language-processing\u002Ffoundations\u002Fminimum-edit-distance","Minimum Edit Distance","Much of language processing needs to measure how similar two strings are — a speller ranking corrections, a diff tool, a coreference resolver. Minimum edit distance counts the insertions, deletions, and substitutions that turn one string into another, computed by a dynamic-programming table. We fill the table for intention to execution, backtrace to recover the alignment, and see how the same machinery generalizes to weighted edits, Viterbi, and biological sequence alignment.\n",{"path":25282,"title":25283,"module":21235,"summary":25284},"\u002Fnatural-language-processing\u002Ffoundations\u002Fn-gram-language-models","N-Gram Language Models","A language model assigns a probability to a sequence of words and, equivalently, predicts the next word from its history. The n-gram model makes this tractable by truncating the history to the last few words, estimates the resulting conditional probabilities by counting, and is scored by perplexity. We build the model from the chain rule, work a bigram example on a small corpus, and read perplexity as a branching factor. The next lesson covers the zero counts that break this model and the smoothing that repairs them.\n",{"path":25286,"title":25287,"module":21235,"summary":25288},"\u002Fnatural-language-processing\u002Ffoundations\u002Fsmoothing-and-backoff","Smoothing and Backoff","Every finite corpus is missing good word sequences it simply never saw, so a raw n-gram model assigns them probability zero and breaks. Smoothing repairs the zeros: add-one and add-k shave mass off seen events, backoff and interpolation fall back on shorter contexts, and Kneser-Ney — worked here by hand — replaces raw frequency with how many contexts a word completes. We close on web-scale stupid backoff and the neural models that dissolve the zero problem rather than patch it.\n",{"path":25290,"title":25291,"module":25292,"summary":25293},"\u002Fnatural-language-processing\u002Fclassification\u002Fnaive-bayes-and-sentiment","Naive Bayes and Sentiment Classification","Text Classification","Text classification assigns a category to a document — positive or negative, spam or not, one topic among many. Naive Bayes is a generative solution: apply Bayes' rule, assume the words are conditionally independent given the class, and the winning class is the one maximizing the product of a prior and per-word likelihoods. We train it by counting with add-one smoothing, work a full sentiment example by hand, sharpen it for sentiment (binary counts, negation, lexicons), and place it among the transformer classifiers that came after.\n",{"path":25295,"title":25296,"module":25292,"summary":25297},"\u002Fnatural-language-processing\u002Fclassification\u002Fevaluating-classifiers","Evaluating Classifiers","A trained classifier is only useful once we can measure how good it is. We build the confusion matrix, see why accuracy misleads on unbalanced data, and define precision, recall, and the F-measure that balances them. Multi-class tasks need macro- versus micro-averaging; reliable estimates need cross-validation. We close on statistical significance — the paired bootstrap test for whether one system's lead over another is significant.\n",{"path":25299,"title":25300,"module":25292,"summary":25301},"\u002Fnatural-language-processing\u002Fclassification\u002Flogistic-regression","Logistic Regression","Logistic regression is the discriminative counterpart to naive Bayes: instead of modelling how a document is generated, it learns weights that directly separate the classes. We build it from the sigmoid, derive the cross-entropy loss from maximum likelihood, learn the weights by stochastic gradient descent, regularize to curb overfitting, and generalize to many classes with the softmax. The two-class model is already a one-neuron network, so this is the bridge to neural language models.\n",{"path":25303,"title":25304,"module":25292,"summary":25305},"\u002Fnatural-language-processing\u002Fclassification\u002Fsentiment-and-affect-lexicons","Sentiment and Affect Lexicons","A sentiment lexicon is a list of words annotated with the affective meaning they carry — positive or negative, or scores along valence, arousal, and dominance. We fix what \"emotion\" means (basic-emotion versus dimensional models), survey the standard lexicons, and then build lexicons three ways: by human labeling with best-worst scaling, by semi-supervised induction from seed words over an embedding space, and by supervised learning from starred reviews. We close on connotation frames, which record the sentiment a verb implies about each of its arguments.\n",{"path":25307,"title":25308,"module":25309,"summary":25310},"\u002Fnatural-language-processing\u002Fsemantics\u002Fvector-semantics-and-embeddings","Vector Semantics and Embeddings","Semantics","Vector semantics represents a word's meaning as a point in space, derived from the company the word keeps. This first part builds the count-based side: the distributional hypothesis, co-occurrence matrices in their term-document and word-word forms, cosine as the similarity measure, and the two weightings — tf-idf and PPMI — that fix what raw counts get wrong. The result is a sparse, interpretable vector for every word, and the setup for the dense embeddings of the next lesson.\n",{"path":25312,"title":25313,"module":25309,"summary":25314},"\u002Fnatural-language-processing\u002Fsemantics\u002Fstatic-word-embeddings","Static Word Embeddings: word2vec and After","Count-based vectors are long and sparse; embeddings are the short, dense alternative. This lesson builds them with word2vec's skip-gram and negative sampling — a classifier whose learned weights are the vectors — derives its gradient, and works one update by hand. It then reads relations off the analogy parallelogram, surveys the papers that framed the static-embedding era (word2vec, GloVe, the SGNS-as-PPMI equivalence, fastText, ELMo), and closes on the biases embeddings inherit and the single-vector-per-word ceiling that contextual models break.\n",{"path":25316,"title":25317,"module":25309,"summary":25318},"\u002Fnatural-language-processing\u002Fsemantics\u002Fneural-language-models","Neural Networks and Neural Language Models","A neural network is a stack of units, each a weighted sum passed through a non-linearity — a single unit on its own is logistic regression. We build the network up from that unit: the activation functions that give it power, the XOR problem that forces a hidden layer, the feedforward forward pass in matrix form, and the Bengio-style feedforward neural language model that concatenates word embeddings and predicts the next word with a softmax. Training is cross-entropy minimized by gradient descent, with backpropagation supplying the gradient. Embeddings let the model share statistical strength across similar words, avoiding the sparsity that limits n-gram models.\n",{"path":25320,"title":25321,"module":21701,"summary":25322},"\u002Fnatural-language-processing\u002Fsequences\u002Fsequence-labeling","Sequence Labeling: POS and NER","Sequence labeling assigns one tag to every token in a sentence. This first part sets up the task through its two canonical cases — part-of-speech tagging over the Penn Treebank tagset, and named-entity recognition reframed as token labeling with the BIO scheme — then builds the hidden Markov model, the classic probabilistic tagger. The HMM tags by Bayesian inference: transition and emission probabilities under two Markov assumptions, reducing tagging to an argmax over tag sequences. That argmax is exponential to enumerate, which sets up the Viterbi decoder, the CRF, and neural taggers of the next lesson.\n",{"path":25324,"title":25325,"module":21701,"summary":25326},"\u002Fnatural-language-processing\u002Fsequences\u002Fcrfs-and-neural-taggers","Viterbi Decoding, CRFs, and Neural Taggers","The HMM reduced tagging to an argmax over exponentially many tag sequences. This lesson builds the decoder that makes it tractable — the Viterbi dynamic program, worked through a full numeric trace on real WSJ probabilities — then keeps that same decoder while replacing the HMM's rigid tables. The linear-chain conditional random field is a discriminative log-linear model whose global feature functions can inspect any part of the input, which is why CRFs win for NER. Finally it traces the shift to neural taggers (biLSTM-CRF, character-aware NER, ELMo), where hand-built features become learned representations while the Viterbi decoder carries over unchanged.\n",{"path":25328,"title":25329,"module":21701,"summary":25330},"\u002Fnatural-language-processing\u002Fsequences\u002Frnns-and-lstms","RNNs and LSTMs","A feedforward neural language model sees a fixed window of words and can look no further back. The recurrent neural network removes that limit: it carries a hidden state across time, so each word is read in the context of everything before it. We build the RNN from its one recurrent equation, use it as a language model, train it by backpropagation through time, and diagnose the vanishing-gradient problem that makes plain RNNs forget. The LSTM fixes the forgetting with a cell state and three gates, and the encoder-decoder stacks two RNNs into a sequence-to-sequence model — and its single-vector bottleneck is the problem attention was invented to remove.\n",{"path":25332,"title":25333,"module":22183,"summary":25334},"\u002Fnatural-language-processing\u002Ftransformers\u002Ftransformers-and-attention","Transformers and Self-Attention","Recurrence forced language models to read one word at a time and to squeeze every dependency through a chain of hidden states. Self-attention removes the recurrence: at every layer each position compares itself to every other and reads a weighted mixture of them, in a single parallel step. This first part builds the attention operation from the ground up — the soft lookup, queries and keys and values, the scaled dot-product, the numeric trace, the matrix form, and the causal mask — and sets up the full transformer architecture that follows.\n",{"path":25336,"title":23887,"module":22183,"summary":25337},"\u002Fnatural-language-processing\u002Ftransformers\u002Fthe-transformer-architecture","This part takes the scaled dot-product attention of the previous lesson and assembles the full transformer architecture around it: multi-head attention so several relations can be read at once, the transformer block of residual connections and layer norm that makes deep stacks trainable, positional embeddings that restore word order, the decoder-only language model, and the encoder, decoder, and encoder-decoder shapes — closing with the 2017 paper and the pre-norm, FlashAttention, and RoPE refinements that scaled it up.\n",{"path":25339,"title":23995,"module":22183,"summary":25340},"\u002Fnatural-language-processing\u002Ftransformers\u002Flarge-language-models","A large language model is a decoder-only transformer trained on one objective — predict the next token. This first part assembles the inference side: the language-modeling head that turns a hidden state into a distribution over the vocabulary, autoregressive generation, and the decoding strategies — greedy, beam, and sampling with temperature, top-k, and nucleus — that read text back out of that distribution. Training the distribution at web scale comes next.\n",{"path":25342,"title":25343,"module":22183,"summary":25344},"\u002Fnatural-language-processing\u002Ftransformers\u002Fllm-pretraining-and-scaling","Large Language Models: Pretraining and Scaling","A language model's next-token distribution is only as good as the parameters behind it. This part is where those parameters come from: self-supervised pretraining on web-scale text with teacher forcing and cross-entropy, the scaling laws that make test loss a predictable power law in parameters, data, and compute, the KV cache that keeps long-context inference affordable, and how a finished model is evaluated by perplexity and benchmarks — closing with the Kaplan, Chinchilla, GPT-3, and emergence papers behind the scaling story.\n",{"path":25346,"title":25347,"module":22183,"summary":25348},"\u002Fnatural-language-processing\u002Ftransformers\u002Ffine-tuning-and-prompting","Fine-Tuning and Prompting","A pretrained transformer is a general-purpose knowledge source; a task is what you do with it. There are two ways to adapt one, and this first part covers the one that updates the weights: fine-tuning. A bidirectional encoder like BERT is pretrained by masked language modeling, then a small task head is bolted on and the whole thing is trained on labelled data for classification, sequence labeling, or span-based question answering — with parameter-efficient variants (adapters, LoRA) that touch only a sliver of the weights. Prompting, the family that leaves the weights frozen, comes next.\n",{"path":25350,"title":25351,"module":22183,"summary":25352},"\u002Fnatural-language-processing\u002Ftransformers\u002Fprompting-and-alignment","Prompting and Alignment","Fine-tuning adapts a model by changing its weights. The second family of adaptation changes nothing: a large frozen model performs a task from an instruction and a few examples placed in its context. This part covers prompting and in-context learning, chain-of-thought that elicits reasoning, and the two training stages — instruction tuning and RLHF — that turn a fluent base predictor into an aligned assistant, closing with the BERT, LoRA, chain-of-thought, InstructGPT, and retrieval-augmentation papers behind the modern adaptation pipeline.\n",{"path":25354,"title":25355,"module":25356,"summary":25357},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fconstituency-parsing","Constituency Parsing","Linguistic Structure","A constituency parse groups a sentence into nested phrases described by a context-free grammar. We build the CFG formalism, read the phrase structure of English off a treebank, confront the structural ambiguity that makes parsing hard, convert to Chomsky normal form, and then solve it with CKY — the dynamic-programming chart that fills a triangular table bottom-up. Probabilistic and neural span parsers, evaluation, and shallow parsing follow in the companion lesson.\n",{"path":25359,"title":25360,"module":25356,"summary":25361},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fcky-scoring-and-evaluation","CKY Scoring, Evaluation, and Shallow Parsing","The CKY chart returns every parse but does not say which is correct. Disambiguation needs a score on trees. This lesson attaches probabilities to a grammar (the PCFG and lexicalization), replaces the grammar with a neural span scorer over a pretrained encoder, states the self-attentive results that made it the state of the art, evaluates parsers against a treebank with PARSEVAL, and closes with chunking and shallow parsing for tasks that need only the flat phrases.\n",{"path":25363,"title":25364,"module":25356,"summary":25365},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fdependency-parsing","Dependency Parsing","A dependency parse throws away phrases and keeps only directed, labeled arcs from heads to their dependents, so the subject and object of a verb hang off the verb directly. We fix the formalism (rooted trees, typed Universal-Dependency relations, projectivity), then build the first parser family: transition-based arc-standard and arc-eager parsing, a greedy stack-and-buffer machine trained from an oracle. Graph-based and neural dependency parsing follow in the companion lesson.\n",{"path":25367,"title":25368,"module":25356,"summary":25369},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fgraph-based-and-neural-dependency-parsing","Graph-Based and Neural Dependency Parsing","Greedy transition parsing commits locally; the graph-based family scores whole trees instead. This lesson scores every candidate head-dependent edge and extracts the maximum spanning tree with Chu-Liu\u002FEdmonds, develops the biaffine neural scorer that made graph-based parsing the accuracy leader, evaluates parsers with the unlabeled and labeled attachment scores (UAS and LAS), and closes on where the two parser families sit and what they feed downstream.\n",{"path":25371,"title":25372,"module":25356,"summary":25373},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fword-senses-and-wsd","Word Senses and Disambiguation","A word is not an atom of meaning: \"bass\" names a fish, a voice, and an instrument, and one static embedding blurs them into a single point. This lesson pulls those senses apart. We define polysemy and the relations that organize senses — synonymy, antonymy, hyponymy, meronymy — build them into WordNet's synset graph, measure similarity along that graph, and then solve the core of word sense disambiguation: the most-frequent-sense baseline, the Lesk gloss-overlap algorithm, feature-based classifiers, and the nearest-neighbor method over BERT embeddings. WSD variants, embeddings, and evaluation follow in the companion lesson.\n",{"path":25375,"title":25376,"module":25356,"summary":25377},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fwsd-in-practice-and-induction","WSD in Practice and Word Sense Induction","Beyond core word sense disambiguation lie the variants and loose ends: the sense-inventory-free Word-in-Context task, retrofitting static embeddings to a thesaurus, discovering senses without a fixed inventory (word sense induction), the gloss-aware and bi-encoder neural systems that hold the state of the art, and how WSD and its cousins are evaluated. Together they connect one-vector-per-word embeddings to sense-aware contextual representations.\n",{"path":25379,"title":25380,"module":25356,"summary":25381},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fsemantic-roles-and-information-extraction","Semantic Roles and Information Extraction","Semantic roles answer \"who did what to whom\" for a single event, abstracting away the syntax that expresses it. We show why syntax alone is not enough, generalize over diathesis alternations with thematic roles, number a predicate's arguments with PropBank and group predicates into frames with FrameNet, tag each argument automatically with semantic role labeling, and factor predicates into primitives. Information extraction scales the idea to a corpus in the companion lesson.\n",{"path":25383,"title":25384,"module":25356,"summary":25385},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Frelations-events-and-templates","Relations, Events, and Templates","Semantic roles answer \"who did what\" for one predicate; information extraction scales the idea to a whole corpus. This lesson turns unstructured text into structured data: relation extraction pulls entity-relation-entity triples out of sentences by patterns, supervision, and distant supervision; event and temporal extraction place those facts on a timeline; and template filling and knowledge-base population assemble them into a database a downstream system can query.\n",{"path":25387,"title":25388,"module":25356,"summary":25389},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fcoreference-and-discourse","Coreference and Discourse","A text is more than a bag of sentences: entities recur under different names. Coreference resolution links every mention to the discourse entity it evokes — the linguistic background of pronouns, definite NPs, and names; mention detection; the mention-pair, mention-ranking, and entity-based architectures; a neural end-to-end span model that scores candidate antecedents; features, evaluation by the CoNLL F1, gender bias, and the neural coreference lineage. Discourse coherence follows in the companion lesson.\n",{"path":25391,"title":25392,"module":25356,"summary":25393},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fcoherence-and-discourse-structure","Coherence and Discourse Structure","Coherence is what makes a run of sentences a discourse rather than an arbitrary collection. This lesson develops coherence relations and Rhetorical Structure Theory trees, discourse-structure parsing, Centering and the entity grid for entity-based coherence, and representation-learning models of local coherence, measured in part over the coreference chains recovered in the companion lesson.\n",{"path":25395,"title":25396,"module":25356,"summary":25397},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Flogical-semantics","Logical Representations of Meaning","A meaning representation turns a sentence into a formal structure a machine can check against a world and reason over. We set the desiderata a good representation must meet, ground truth in a model, build up first-order logic for sentences with its connectives, quantifiers, and inference, and reify events with the neo-Davidsonian event variable to escape fixed predicate arity. The compositional lambda calculus, quantifier scope, and description logics follow in the companion lesson.\n",{"path":25399,"title":25400,"module":25356,"summary":25401},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fcompositional-semantics-and-description-logics","Compositional Semantics and Description Logics","How do you compute a logical form from a sentence automatically? This lesson builds the compositional machinery: the lambda calculus that assembles a formula from a parse tree one beta-reduction at a time, the quantifier-scope ambiguity a single syntax tree leaves open, and the decidable description logics — TBox, ABox, subsumption, role restrictions — behind the Web Ontology Language, closing with how the map from string to logical form can be learned.\n",{"path":25403,"title":25404,"module":25356,"summary":25405},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fsemantic-parsing","Semantic Parsing","Turning a sentence into a structured, executable meaning, the grammar-based way. We take the logical forms defined earlier and build them compositionally: a rule-based parser that walks a syntax tree applying lambda terms, then Combinatory Categorial Grammar (CCG), which fuses syntax and semantics so one lexicalized derivation produces both — including supertagging and A* parsing. Learned and neural semantic parsers follow in the companion lesson.\n",{"path":25407,"title":25408,"module":25356,"summary":25409},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fneural-semantic-parsing","Learned and Neural Semantic Parsing","Hand-writing a lexicon of lambda terms does not scale, so this lesson learns the parser instead. We cover the two supervision regimes (from logical forms and from denotations), Abstract Meaning Representation as a rooted concept graph, neural sequence-to-sequence parsing with constrained decoding and copy mechanisms, executable text-to-SQL and knowledge-based question answering, the practical systems that made learned parsers accurate, and how the task is evaluated.\n",{"path":25411,"title":25412,"module":25356,"summary":25413},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Finformation-extraction","Information Extraction","Information extraction turns free text into a database, and the first step is relation extraction: pulling entity-relation-entity triples out of sentences. We cover all five families — hand-built patterns, supervised classifiers, semi-supervised bootstrapping, distant supervision, and unsupervised Open IE — with worked bootstrapping and distant-supervision traces, then the neural and LLM systems that extended them. Times, events, and templates follow in the companion lesson.\n",{"path":25415,"title":25416,"module":25356,"summary":25417},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Ftimes-events-and-templates","Extracting Times, Events, and Templates","Once relation extraction has produced typed triples, the information-extraction pipeline still has to place facts in time and assemble them into records. This lesson detects and normalizes temporal expressions to ISO 8601 values, detects events and orders them on a timeline with the 13 Allen relations, and fills slot-and-filler templates — flat and hierarchical — for stereotyped situations, closing the loop from text to a queryable database.\n",{"path":25419,"title":25420,"module":25356,"summary":25421},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fdiscourse-coherence","Discourse Coherence","A text is more than a set of sentences. What binds a run of sentences into a discourse is coherence, and one of its sources is structured relations between clauses. This lesson develops relational coherence — RST and the PDTB models of coherence relations — and discourse-structure parsing: EDU segmentation and shift-reduce RST parsing, then PDTB relation classification. Entity-based and global coherence follow in the companion lesson.\n",{"path":25423,"title":25424,"module":25356,"summary":25425},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fentity-based-and-global-coherence","Entity-Based and Global Coherence","A text coheres not only through relations between clauses but by staying about the same entities and the same topic, and by obeying the macro-structure of its genre. This lesson develops Centering Theory and the entity grid for entity-based coherence, representation-learning models of local coherence, and global coherence — topic segmentation, narrative and argumentation structure, and scientific discourse — then the neural models that learn each.\n",{"path":25427,"title":25428,"module":25356,"summary":25429},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Fconstituency-grammars","Constituency Grammars","A constituency grammar is the declarative theory of sentence structure that a parser operates on. We build the context-free grammar formalism from its four parts, show how derivations become parse trees, and work through the phrase structure of English — noun phrases, verb phrases and their subcategorization frames, agreement, coordination, and long-distance dependencies. The treebank, normal-form, and lexicalized views follow in the companion lesson.\n",{"path":25431,"title":25432,"module":25356,"summary":25433},"\u002Fnatural-language-processing\u002Flinguistic-structure\u002Ftreebanks-and-lexicalized-grammars","Treebanks and Lexicalized Grammars","Where does a grammar come from, and how is it prepared for a parser? We read a context-free grammar off the Penn Treebank, normalize it to Chomsky Normal Form for the CKY chart, then invert the phrase-structure emphasis with lexicalized grammars — Combinatory Categorial Grammar and its slash categories — and close with the grammar's fate in the neural era: span scoring, self-attention, and grammar induction.\n",{"path":25435,"title":25436,"module":23983,"summary":25437},"\u002Fnatural-language-processing\u002Fapplications\u002Fmachine-translation","Machine Translation","Machine translation is the task that built the modern toolkit: the encoder-decoder was invented for it, attention was invented to fix its fixed-context bottleneck, and both were later folded into the general transformer. We work through why translation is hard (word order, morphology, lexical and structural divergences), the sequence-to-sequence model and its attention mechanism, transformer-based NMT with cross-attention, subword tokenization with a shared vocabulary, beam-search decoding, and evaluation by BLEU and its successors chrF, BERTScore, and COMET — closing on multilingual and low-resource translation and backtranslation.\n",{"path":25439,"title":25440,"module":23983,"summary":25441},"\u002Fnatural-language-processing\u002Fapplications\u002Fmachine-translation-decoding-and-evaluation","Machine Translation: Decoding, Evaluation, and Scale","Having built the transformer translation model, we now decode from it and measure the output. Beam search turns the decoder's per-step distributions into a single output string; length normalization keeps it from favoring short translations. We then score translations automatically — BLEU with its n-gram precision, clipping, and brevity penalty, worked through by hand, then its successors chrF, BERTScore, and COMET — and close on the parts of MT that scale beyond one language pair: multilingual and low-resource translation, backtranslation, gender bias, and the lineage from the Transformer to massively multilingual models like NLLB-200.\n",{"path":25443,"title":25444,"module":23983,"summary":25445},"\u002Fnatural-language-processing\u002Fapplications\u002Fquestion-answering","Question Answering","A question-answering system takes a natural-language question and returns an answer, not a ranked list of documents. Almost every modern system is built on one pattern: retrieve then read. We start with the information-retrieval machinery that finds candidate text — tf-idf and BM25 term weighting, a worked ranking example, the inverted index, and dense embedding retrieval — then build the retriever-reader pipeline that extracts an answer span with BERT and trace a full retrieve-and-read example end to end.\n",{"path":25447,"title":25448,"module":23983,"summary":25449},"\u002Fnatural-language-processing\u002Fapplications\u002Fquestion-answering-knowledge-and-llms","Question Answering: Knowledge Bases and Language Models","The retrieve-and-read pipeline extracts an answer span from prose, but not all knowledge lives in prose. This part covers the rest of the QA stack: entity linking (Wikification) that grounds a question's entities to a knowledge base, knowledge-based QA by semantic parsing a question into an executable query, and the modern default — closed-book QA and retrieval-augmented generation with a large language model — closing on the DPR\u002FRAG\u002Ffusion-in-decoder lineage and how factoid answers are scored by exact match and F1.\n",{"path":25451,"title":25452,"module":23983,"summary":25453},"\u002Fnatural-language-processing\u002Fapplications\u002Fdialogue-and-chatbots","Dialogue and Chatbots","Conversation is the most natural interface to a machine and one of the hardest to build. We set up what makes human dialogue work — turns, speech acts, grounding, and the local structure of adjacency pairs — then trace the two traditions that answer it: chatbots built to chat (ELIZA's pattern-matching, corpus retrieval, and seq2seq generation with its blandness problem) and task-oriented systems built to get something done (the GUS frame-and-slot architecture and the modern NLU \u002F state-tracker \u002F policy \u002F NLG pipeline that accumulates a frame across turns).\n",{"path":25455,"title":25456,"module":23983,"summary":25457},"\u002Fnatural-language-processing\u002Fapplications\u002Fdialogue-systems-and-assistants","Dialogue Systems: LLM Assistants, Evaluation, and Design","Two dialogue traditions — chatbots built to chat and task-oriented frame systems built to get something done — met in the aligned LLM assistant. Instruction tuning plus RLHF fold chit-chat and task dialogue into one model; the LaMDA \u002F InstructGPT \u002F ChatGPT lineage fills in how. The lesson then turns to evaluation (human ratings and acute-eval for chatbots, task success and slot error rate for task systems), user-centered design with Wizard-of-Oz prototyping, and the ethical stakes of building agents people talk to.\n",{"path":25459,"title":25460,"module":23983,"summary":25461},"\u002Fnatural-language-processing\u002Fapplications\u002Ftext-summarization","Text Summarization","Summarization compresses a document to its essential meaning, by either selecting sentences to keep (extractive) or writing new ones (abstractive). This part fixes the task and its flavors — single vs. multi-document, generic vs. query-focused, extractive vs. abstractive — then works through extractive summarization in full: scoring by position and centrality, the TextRank\u002FLexRank graph algorithm run as PageRank over a sentence-similarity graph with a worked iteration, and supervised sentence selection.\n",{"path":25463,"title":25464,"module":23983,"summary":25465},"\u002Fnatural-language-processing\u002Fapplications\u002Fabstractive-summarization-and-evaluation","Abstractive Summarization and Evaluation","Extractive methods can only reuse the source's own sentences; to compress within a sentence or paraphrase, a summarizer has to generate. This part covers abstractive summarization: the sequence-to-sequence approach, the pointer-generator's copy switch and coverage mechanism, pretrained summarizers (BART, PEGASUS) and zero-shot LLM prompting, the long-document and factuality problems, and ROUGE evaluation with a worked example and its limits — closing on the abstractive lineage from See 2017 through faithfulness metrics.\n",{"path":25467,"title":25468,"module":25469,"summary":25470},"\u002Fnatural-language-processing\u002Fspeech\u002Fphonetics","Phonetics","Speech","Before a recognizer can read speech it has to know what speech is. This first part covers the linguistic substrate: phones and their transcription in the IPA and ARPAbet; articulatory phonetics — how the vocal tract shapes airflow into consonants and vowels; and prosody — stress, tune, and the F0 contour. The acoustic side — the waveform, its spectrum, formants, and the spectrogram — is the second part.\n",{"path":25472,"title":25473,"module":25469,"summary":25474},"\u002Fnatural-language-processing\u002Fspeech\u002Facoustic-phonetics","Acoustic Phonetics","Articulation is the cause; the acoustic signal is the effect, and the effect is all a microphone ever gets. This part follows the sound out of the mouth: waves, sampling and the Nyquist limit, F0 and the pitch track, the mel scale, the spectrum and Fourier analysis, the source-filter model that explains why each vowel carries its own formants, and the spectrogram the log-mel front end of every ASR system sits directly on top of — closing with neural TTS, wav2vec, HuBERT, and Whisper, where phonetics went in neural speech.\n",{"path":25476,"title":25477,"module":25469,"summary":25478},"\u002Fnatural-language-processing\u002Fspeech\u002Fautomatic-speech-recognition","Automatic Speech Recognition","Speech recognition maps an acoustic waveform to a string of words, and once the waveform is turned into a sequence of log-mel spectrogram frames the problem is the same sequence-to-sequence transduction the rest of the course already solved. This first part builds the feature front end (framing, the DFT, the mel filterbank, the log), then the modern architectures: the attention-based encoder-decoder, the CTC alignment trick that collapses repeated and blank frames, and RNN-T for streaming. Training-data advances, evaluation, TTS, and the other speech tasks come next.\n",{"path":25480,"title":25481,"module":25469,"summary":25482},"\u002Fnatural-language-processing\u002Fspeech\u002Fasr-evaluation-and-applications","ASR Evaluation and Speech Applications","A recognizer turns a waveform into text; this part scores that text and puts the same machinery to other uses. It opens with the self-supervised and weakly- supervised systems (wav2vec 2.0, HuBERT, Whisper) that made ASR error rates fall. Word error rate reuses the edit distance from the first module, run over words. Text-to-speech runs the whole pipeline in reverse — text to mel spectrogram to waveform. And a family of smaller tasks — wake-word detection, speaker recognition and diarization, language identification — reuse the same log-mel front end without the decoder.\n",{"path":25484,"title":25485,"module":6,"summary":6},"\u002Fnatural-language-processing","Natural Language Processing",{"path":25487,"title":25488,"module":21235,"summary":25489},"\u002Fparticle-physics\u002Ffoundations\u002Fhistorical-overview-particle-zoo","From the Electron to the Particle Zoo","A timeline of the subject, from J. J. Thomson's electron in 1897 to the Higgs boson in 2012. The electron, photon, nucleus, proton, and neutron gave a tidy picture that Yukawa's meson prediction and the muon–pion confusion complicated; strange particles in cosmic rays and the accelerator-era flood of hadrons then produced a \"particle zoo\" that only the quark model organized.\n",{"path":25491,"title":25492,"module":21235,"summary":25493},"\u002Fparticle-physics\u002Ffoundations\u002Fparticle-physics-basic-concepts","Basic Concepts and Particle Classification","Every particle has an antiparticle of equal mass and opposite charge, a consequence of the Dirac equation confirmed by the positron. Feynman diagrams track interactions in spacetime; the material particles sort into leptons and the composite hadrons built from quarks, with baryons carrying three quarks and mesons a quark-antiquark pair.\n",{"path":25495,"title":25496,"module":21235,"summary":25497},"\u002Fparticle-physics\u002Ffoundations\u002Ffundamental-interactions-force-carriers","Fundamental Interactions and Force Carriers","Four interactions account for every force in nature: strong, electromagnetic, weak, and gravitational, in decreasing strength. Each is carried by a boson — the gluon, photon, W and Z, and the graviton — with a range fixed by the carrier's mass through the Yukawa relation, and a coupling constant that itself varies with distance.\n",{"path":25499,"title":25500,"module":25501,"summary":25502},"\u002Fparticle-physics\u002Funits-kinematics\u002Fnatural-units-and-scales","Natural Units and Scales","Units and Kinematics","Setting $\\hbar = c = 1$ collapses mass, momentum, and energy into a single unit, the GeV, and turns lengths and times into inverse energies through the conversion $\\hbar c = 197.3$ MeV·fm. This lesson fixes the natural-unit conventions used for the rest of the course, converts cross sections between barns and GeV$^{-2}$, and shows how to restore factors of $\\hbar$ and $c$ by dimensional analysis.\n",{"path":25504,"title":25505,"module":25501,"summary":25506},"\u002Fparticle-physics\u002Funits-kinematics\u002Ffour-vectors-invariant-mass","Four-Vectors and Invariant Mass","The energy and momentum of a particle form a four-vector whose square is the frame-independent quantity $p^2 = m^2$. This lesson develops the metric and four-vector products, the invariant mass of a multiparticle system, the center-of-momentum and laboratory frames, and the description of collinear boosts by rapidity, whose additivity replaces the awkward velocity-addition law.\n",{"path":25508,"title":25509,"module":25501,"summary":25510},"\u002Fparticle-physics\u002Funits-kinematics\u002Fdecay-scattering-kinematics-mandelstam","Decay, Scattering, and Mandelstam Variables","Two-body decay in the rest frame fixes the daughter momenta from the three masses alone; production thresholds follow from the minimum invariant mass. This lesson works both, then introduces the Mandelstam invariants $s$, $t$, $u$ for $2\\to2$ scattering, proves the identity $s+t+u=\\sum m_i^2$, and maps the physical regions and the crossing that relates channels.\n",{"path":25512,"title":25513,"module":25501,"summary":25514},"\u002Fparticle-physics\u002Funits-kinematics\u002Fcross-sections-golden-rule","Cross Sections and the Golden Rule","The cross section measures how often a scattering happens and the decay width how fast a particle disintegrates. This lesson defines both, relates event rate to luminosity through $R=\\mathcal L\\,\\sigma$ and lifetime to width through $\\tau=\\hbar\u002F\\Gamma$, and states Fermi's golden rule with Lorentz-invariant phase space, giving the master formulas that turn an amplitude $\\mathcal M$ into a measurable rate for $1\\to2$ decay and $2\\to2$ scattering.\n",{"path":25516,"title":25517,"module":25518,"summary":25519},"\u002Fparticle-physics\u002Fsymmetries\u002Fconservation-laws-symmetries","Conservation Laws and Symmetries","Symmetries and Conservation Laws","Which decays occur is decided by conservation laws, each tied by Noether's theorem to a symmetry of physical law. Energy, charge, baryon number, and lepton number are conserved universally; strangeness, isospin, and parity hold in the strong and electromagnetic interactions but break in the weak one, whose parity and CP violation distinguish matter from antimatter.\n",{"path":25521,"title":25522,"module":25518,"summary":25523},"\u002Fparticle-physics\u002Fsymmetries\u002Fdiscrete-symmetries-cpt","Discrete Symmetries — C, P, T, and CPT","Parity reflects space, charge conjugation swaps particle for antiparticle, and time reversal runs the clock backward. Each assigns multiplicative quantum numbers that act as selection rules — intrinsic parities, the photon's C = −1, the C-parity argument fixing the pion's two-photon decay. Their product CPT is a theorem of any local relativistic field theory, forcing particle and antiparticle to share mass and lifetime.\n",{"path":25525,"title":25526,"module":25518,"summary":25527},"\u002Fparticle-physics\u002Fsymmetries\u002Fparity-violation-weak","Parity Violation and the Weak Force","The tau–theta puzzle forced a choice: two particles with identical mass but opposite parity, or one particle whose decay violates parity. Lee and Yang proposed the latter, Wu's polarized cobalt-60 confirmed it, and the violation proved maximal. The charged weak force couples only to left-handed chirality — the Goldhaber experiment showed the neutrino is left-handed — which is why the mirror image of a weak decay is something nature never produces.\n",{"path":25529,"title":25530,"module":25518,"summary":25531},"\u002Fparticle-physics\u002Fsymmetries\u002Fsu2-su3-flavor-symmetry","Isospin, SU(2), and Flavor SU(3)","The near-equal masses of the proton and neutron, and of the three pions, signal a continuous internal symmetry of the strong force: isospin, an SU(2) whose ladder operators move between the members of a multiplet. Adding strangeness enlarges it to an approximate SU(3) flavor symmetry, and the Gell-Mann–Nishijima relation Q = I3 + Y\u002F2 places every hadron on a weight diagram in the isospin–hypercharge plane — the language in which the quark model is written.\n",{"path":25533,"title":25534,"module":25535,"summary":25536},"\u002Fparticle-physics\u002Fquark-model\u002Feightfold-way-su3","The Eightfold Way and SU(3) Flavor","The Quark Model","Gell-Mann and Ne'eman's classification of the hadrons into geometric multiplets, read as representations of an approximate flavor SU(3). The fundamental triplet (u, d, s) and its antitriplet combine into the meson nonet from 3⊗3̄ = 8⊕1 and the baryon octet and decuplet from 3⊗3⊗3, and the empty corner of the decuplet forecast the Ω⁻.\n",{"path":25538,"title":25539,"module":25535,"summary":25540},"\u002Fparticle-physics\u002Fquark-model\u002Fmeson-spectroscopy","Meson Multiplets and Quantum Numbers","Mesons as quark–antiquark bound states. The spin singlet and triplet, orbital excitations, and the assignment of J^PC from the quark spins and orbital angular momentum, giving the pseudoscalar and vector nonets. The η–η' and ω–φ mixing problems, and the charmonium and bottomonium spectra read as heavy-quark positronium.\n",{"path":25542,"title":25543,"module":25535,"summary":25544},"\u002Fparticle-physics\u002Fquark-model\u002Fbaryon-spectroscopy","Baryon Multiplets, Spin, and the Color Puzzle","Baryons as three-quark states, with a wavefunction factored into space, spin, flavor, and color. The spin-3\u002F2 Δ⁺⁺ = uuu forces a totally symmetric state that the Pauli principle forbids, and the resolution is an antisymmetric color factor — the first evidence for color. The octet and decuplet spin content, and baryon magnetic moments as a quantitative test of the model.\n",{"path":25546,"title":25547,"module":25535,"summary":25548},"\u002Fparticle-physics\u002Fquark-model\u002Fcolor-confinement-exotics","Color, Confinement, and Exotic Hadrons","Color as the gauged SU(3) charge, and the requirement that every physical hadron be a color singlet — which selects q-qbar mesons and qqq baryons as the simplest states. The R-ratio of e⁺e⁻ annihilation measures three colors directly. Beyond the simplest singlets lie glueballs, tetraquarks, and pentaquarks, and the recent XYZ states, read as either compact multiquarks or loose hadronic molecules.\n",{"path":25550,"title":25551,"module":25552,"summary":25553},"\u002Fparticle-physics\u002Frelativistic-wave-equations\u002Fklein-gordon-equation","The Klein-Gordon Equation","Relativistic Wave Equations","Quantizing the relativistic energy relation $E^2 = p^2 + m^2$ produces the Klein-Gordon equation for a scalar field. Its plane-wave solutions come in positive- and negative-energy branches, and the conserved density it supplies is not positive-definite — the two difficulties that first drove physicists to seek a first-order equation. The static Klein-Gordon equation with a point source gives the Yukawa potential, and the free equation gives the scalar propagator that later modules attach to exchanged lines.\n",{"path":25555,"title":25556,"module":25552,"summary":25557},"\u002Fparticle-physics\u002Frelativistic-wave-equations\u002Fdirac-equation-spinors","The Dirac Equation and Spinors","Dirac demanded a wave equation first order in time to fix the Klein-Gordon density problem. Factorizing $E^2 = p^2 + m^2$ into a linear form forces the coefficients to be anticommuting matrices — the gamma matrices of the Clifford algebra — so the wavefunction becomes a four-component spinor. The plane-wave solutions split into two particle and two antiparticle states, spin appears automatically with the correct $g = 2$ magnetic moment, and the chirality projectors that the weak interaction later needs fall straight out of the fifth gamma matrix.\n",{"path":25559,"title":25560,"module":25552,"summary":25561},"\u002Fparticle-physics\u002Frelativistic-wave-equations\u002Fantiparticles-hole-theory","Antiparticles and Hole Theory","The negative-energy solutions of the Dirac equation refuse to go away, so they must mean something. Dirac read them as a filled sea of occupied negative-energy states whose holes are positive-energy antiparticles, predicting the positron before its discovery. The picture works for fermions but not bosons, and the Feynman-Stückelberg interpretation replaces it: an antiparticle is a negative-energy solution propagating backward in time, equivalent to a positive-energy antiparticle going forward. Crossing symmetry ties incoming particles to outgoing antiparticles in a single amplitude.\n",{"path":25563,"title":25564,"module":25565,"summary":25566},"\u002Fparticle-physics\u002Fqed\u002Ffeynman-rules-qed","Feynman Rules for QED","Quantum Electrodynamics","Quantum electrodynamics computes a process by summing diagrams, each a term in a power series in the coupling. Every diagram translates into an amplitude by a fixed dictionary: spinors and polarization vectors for external lines, propagators for internal lines, and the vertex factor $ie\\gamma^\\mu$ for each photon-fermion junction. Squaring the amplitude and feeding it to Fermi's golden rule produces a cross section or decay rate, with each extra vertex costing one power of $\\alpha$.\n",{"path":25568,"title":25569,"module":25565,"summary":25570},"\u002Fparticle-physics\u002Fqed\u002Fqed-tree-processes","Tree-Level QED Processes","The Feynman rules become numbers on the reference reactions of QED. Muon pair production $e^+e^-\\to\\mu^+\\mu^-$ sets the scale with its $1+\\cos^2\\theta$ distribution and $4\\pi\\alpha^2\u002F3s$ total cross section, and its ratio to hadron production counts colors. Compton scattering gives the Klein-Nishina formula and the Thomson limit; Bhabha scattering shows $s$- and $t$-channel interference. Casimir's trick turns every spin-averaged square into a trace of gamma matrices.\n",{"path":25572,"title":25573,"module":25565,"summary":25574},"\u002Fparticle-physics\u002Fqed\u002Frenormalization-running-coupling","Renormalization and the Running Coupling","Beyond tree level, QED loops diverge. The three primitive one-loop diagrams — vacuum polarization, electron self-energy, and vertex correction — carry ultraviolet divergences that regularization exposes as logarithms of a cutoff. Renormalization absorbs them into the measured mass, charge, and field normalization, leaving finite predictions. The surviving physical content is that the coupling runs: vacuum polarization screens charge, so $\\alpha$ grows from $1\u002F137$ at low energy to about $1\u002F128$ at the $Z$ mass.\n",{"path":25576,"title":25577,"module":25565,"summary":25578},"\u002Fparticle-physics\u002Fqed\u002Felectron-g-2","The Anomalous Magnetic Moment","The Dirac equation predicts $g=2$; loops shift it. Schwinger's one-loop vertex correction gives the anomaly $a=(g-2)\u002F2=\\alpha\u002F2\\pi$, and the QED series continues to five loops. The electron $a_e$ agrees with theory to better than a part in a billion, the most precise confrontation of theory and experiment in physics. The muon $a_\\mu$, heavier and so more sensitive to virtual heavy states, is dominated by hadronic uncertainty and sits at the center of a long-running comparison with the Standard Model prediction.\n",{"path":25580,"title":25581,"module":25582,"summary":25583},"\u002Fparticle-physics\u002Fweak-interaction\u002Fva-structure-weak","The V–A Charged Weak Current","The Weak Interaction","Fermi modelled beta decay as a four-fermion contact interaction, but a coupling with dimensions of inverse mass squared makes cross sections grow without bound and the theory fails near 300 GeV. The cure is a heavy mediator: the $W$ boson, whose propagator collapses to Fermi's contact term at low energy and fixes $G_F\u002F\\sqrt2 = g^2\u002F8M_W^2$. Parity violation dictates the current's form — vector minus axial-vector, coupling only to left-chiral fields — and universality of the coupling ties muon decay, beta decay, and pion decay to one constant. Pion decay's helicity suppression of the electron channel is the sharpest test.\n",{"path":25585,"title":25586,"module":25582,"summary":25587},"\u002Fparticle-physics\u002Fweak-interaction\u002Fw-z-bosons-decays","The W and Z Bosons","The contact theory hides a massive mediator. The charged $W^\\pm$ carries the current that changes flavour; the neutral $Z^0$ carries a current that does not. Both were found at CERN's proton–antiproton collider in 1983 at the masses the electroweak theory demanded. Their decay widths partition into leptonic and hadronic channels, and the $Z$ carries a decisive extra: an invisible width from decays to neutrinos that counts the number of light generations at exactly three. Beta decay and muon decay are re-read at the parton level as $W$ exchange.\n",{"path":25589,"title":25590,"module":25582,"summary":25591},"\u002Fparticle-physics\u002Fweak-interaction\u002Fckm-matrix","Quark Mixing and the CKM Matrix","The quark eigenstates the weak force acts on are not the mass eigenstates. Cabibbo captured this with one rotation angle; the GIM mechanism added a fourth quark to cancel dangerous flavour-changing neutral currents and predicted charm before its discovery. Three generations promote the rotation to the unitary Cabibbo–Kobayashi–Maskawa matrix — three angles and one irreducible complex phase, the sole source of Standard-Model CP violation. The Wolfenstein parametrization exposes its steep hierarchy, and unitarity closes into a triangle whose area measures the phase.\n",{"path":25593,"title":25594,"module":25582,"summary":25595},"\u002Fparticle-physics\u002Fweak-interaction\u002Fcp-violation-kaons-b-mesons","CP Violation in Kaons and B Mesons","The neutral kaon is its own laboratory for CP. Weak box diagrams mix $K^0$ and its antiparticle into short- and long-lived states that should be pure CP eigenstates decaying to two and three pions. In 1964 Cronin and Fitch caught the long-lived kaon decaying to two pions — CP is violated, at the two-per-mille level of $\\epsilon$. Direct violation ($\\epsilon'$) followed, and the $B$ factories turned the CKM phase into a large, clean time-dependent asymmetry measuring $\\sin 2\\beta$. The effect is real but far too small to explain why the universe is made of matter.\n",{"path":25597,"title":25598,"module":25599,"summary":25600},"\u002Fparticle-physics\u002Fqcd\u002Fcolor-su3-gluons","Color SU(3), Gluons, and the QCD Lagrangian","Quantum Chromodynamics","Color is the exact gauged SU(3) charge of the strong force. Gauging it forces eight massless gluons in the adjoint representation and, because the gauge group is non-abelian, three- and four-gluon self-couplings absent from QED. This lesson builds the QCD Lagrangian from the covariant derivative and the non-abelian field strength, states the Feynman rules with their color factors, and computes the Casimir invariants that set the strength of quark-gluon and gluon-gluon coupling.\n",{"path":25602,"title":25603,"module":25599,"summary":25604},"\u002Fparticle-physics\u002Fqcd\u002Fasymptotic-freedom-confinement","Asymptotic Freedom and Confinement","The QCD beta function is negative: gluon self-interaction antiscreens color, so the coupling weakens at short distance (asymptotic freedom) and strengthens at long distance (confinement). This lesson computes the one-loop beta coefficient, solves for the running of alpha_s and the emergent scale Lambda_QCD, and reads the strong-coupling regime as the linear quark-antiquark potential of a color flux tube that breaks by pair creation.\n",{"path":25606,"title":25607,"module":25599,"summary":25608},"\u002Fparticle-physics\u002Fqcd\u002Fdeep-inelastic-scattering-partons","Deep Inelastic Scattering and the Parton Model","Scattering electrons hard off a proton resolves pointlike constituents. This lesson sets up the deep-inelastic kinematics, defines the structure functions F1 and F2, and reads Bjorken scaling as the signature of free spin-half partons. The Callan-Gross relation fixes the parton spin, the structure function becomes a charge-weighted sum of parton distributions, and the slow logarithmic scaling violations expose the gluon through DGLAP evolution.\n",{"path":25610,"title":25611,"module":25599,"summary":25612},"\u002Fparticle-physics\u002Fqcd\u002Fjets-hadronization","Jets, Hadronization, and Testing QCD","Quarks and gluons produced in a collision fragment into collimated sprays of hadrons — jets — whose directions track the underlying partons. This lesson reads two-jet events as the quark and antiquark of electron-positron annihilation, three-jet events as direct evidence of the radiated gluon, and the hadronization step as the flux tube breaking into color singlets. Jet algorithms and event-shape variables turn the pattern into precision measurements of alpha_s.\n",{"path":25614,"title":25615,"module":25616,"summary":25617},"\u002Fparticle-physics\u002Felectroweak-higgs\u002Felectroweak-su2-u1","The Electroweak Theory","Electroweak Unification and the Higgs","The electromagnetic and weak interactions are two faces of a single gauge theory built on $SU(2)_L \\times U(1)_Y$. Left-handed fermions sit in weak-isospin doublets and right-handed fermions in singlets, each carrying a hypercharge fixed by the Gell-Mann–Nishijima relation $Q = T_3 + Y\u002F2$. The four gauge fields $W^{1,2,3}$ and $B$ mix: the charged combinations $W^\\pm$ mediate the charged current, while $W^3$ and $B$ rotate through the Weinberg angle into the massless photon and the massive $Z$. The single angle $\\theta_W$ ties the couplings, the boson masses, and the neutral-current strengths together.\n",{"path":25619,"title":25620,"module":25616,"summary":25621},"\u002Fparticle-physics\u002Felectroweak-higgs\u002Fspontaneous-symmetry-breaking","Spontaneous Symmetry Breaking","A symmetry of the Lagrangian need not be a symmetry of the ground state. When the lowest-energy configuration sits away from the symmetric point, the symmetry is spontaneously broken and the vacuum is one of a degenerate family. Breaking a continuous global symmetry produces one massless scalar — a Goldstone boson — for every broken generator, the flat direction along the vacuum manifold. The Mexican-hat potential and the ferromagnet below its Curie point are the working pictures.\n",{"path":25623,"title":25624,"module":25616,"summary":25625},"\u002Fparticle-physics\u002Felectroweak-higgs\u002Fhiggs-mechanism","The Higgs Mechanism","Gauging a spontaneously broken symmetry converts the would-be Goldstone bosons into the longitudinal polarizations of the gauge fields, which thereby acquire mass. Applied to $SU(2)_L \\times U(1)_Y$ with a single Higgs doublet, three of the four scalar degrees of freedom are eaten by the $W^\\pm$ and $Z$; the fourth survives as the physical Higgs boson, and the photon stays massless. Fermion masses come from Yukawa couplings to the same field, each mass proportional to its coupling times the vacuum expectation value $v \\approx 246$ GeV.\n",{"path":25627,"title":25628,"module":25616,"summary":25629},"\u002Fparticle-physics\u002Felectroweak-higgs\u002Fhiggs-boson-discovery","The Higgs Boson","The Higgs boson is produced at the LHC chiefly through gluon fusion, with vector-boson fusion and associated production as cleaner but rarer channels. It decays most often to $b\\bar b$ and $WW^\\ast$, but the discovery rested on two rare clean modes, $H \\to \\gamma\\gamma$ and $H \\to ZZ^\\ast \\to 4\\ell$, whose narrow invariant-mass peaks emerged over smooth backgrounds. ATLAS and CMS announced a boson near 125 GeV in 2012; its measured spin-parity $0^+$ and its couplings, which scale with particle mass, identify it as the Standard Model Higgs.\n",{"path":25631,"title":25632,"module":25616,"summary":25633},"\u002Fparticle-physics\u002Felectroweak-higgs\u002Fstandard-model","The Standard Model","The Standard Model combines the quark model, quantum chromodynamics, and the electroweak theory. SU(3) symmetry sorts the hadrons and predicted the omega; color explains why only colorless quark combinations exist; QCD gives asymptotic freedom and confinement; and spontaneous symmetry breaking through the Higgs field gives the weak bosons their mass.\n",{"path":25635,"title":25636,"module":25637,"summary":25638},"\u002Fparticle-physics\u002Fneutrinos\u002Fneutrino-oscillations","Neutrino Oscillations","Neutrino Physics","Neutrinos are produced and detected in flavour states, but they propagate as mass states, and the two bases are misaligned. A flavour therefore evolves coherently into a superposition of other flavours with a probability set by the mass-squared splitting and the ratio L\u002FE. This lesson derives the two-flavour oscillation formula, applies it to the solar and atmospheric neutrino deficits, shows how the SNO neutral-current measurement resolved the solar problem, and works out the MSW resonance that amplifies mixing inside the Sun.\n",{"path":25640,"title":25641,"module":25637,"summary":25642},"\u002Fparticle-physics\u002Fneutrinos\u002Fneutrino-mass-pmns","Neutrino Mass and the PMNS Matrix","Three-flavour mixing promotes the single oscillation angle to the unitary Pontecorvo–Maki–Nakagawa–Sakata matrix, parametrised by three angles and a Dirac CP phase. This lesson decomposes the PMNS matrix into three rotations, records the measured angles and mass-squared splittings, lays out the normal and inverted mass orderings, contrasts the large leptonic mixing with the near-diagonal CKM matrix, and collects the absolute-mass bounds from beta decay and cosmology.\n",{"path":25644,"title":25645,"module":25637,"summary":25646},"\u002Fparticle-physics\u002Fneutrinos\u002Fdirac-majorana-experiments","Dirac, Majorana, and Neutrino Experiments","A neutral fermion can carry a mass term forbidden to every charged particle, so the neutrino may be its own antiparticle. This lesson contrasts the Dirac and Majorana mass terms and their state content, derives the seesaw mechanism that ties a tiny light mass to a heavy right-handed partner, presents neutrinoless double-beta decay as the decisive lepton-number test, surveys the reactor, accelerator, solar, and atmospheric sources on a baseline–energy map, and explains why neutrino mass is physics beyond the original Standard Model.\n",{"path":25648,"title":25649,"module":25650,"summary":25651},"\u002Fparticle-physics\u002Fexperiment\u002Faccelerators-luminosity","Accelerators, Colliders, and Luminosity","Accelerators and Detectors","Fixed-target machines waste energy in the center-of-mass motion of the whole system, so the reachable $\\sqrt s$ grows only as the square root of the beam energy, while colliders put the full beam energy into the collision. Circular electron machines are limited by synchrotron radiation scaling as $E^4\u002Fm^4R$; proton machines are limited by bending fields. Luminosity, set by beam current and focusing, converts a cross section into an event rate through $R=\\mathcal L\\,\\sigma$, and integrated luminosity sets the total event count.\n",{"path":25653,"title":25654,"module":25650,"summary":25655},"\u002Fparticle-physics\u002Fexperiment\u002Fdetectors-subsystems","Particle Detectors and Subsystems","A detector reads a collision by the energy particles deposit as they cross matter. Charged particles ionize at the Bethe-Bloch rate, radiate in the field of nuclei above a critical energy, and emit Cherenkov light above a velocity threshold; electrons and photons build electromagnetic showers over a radiation length, and hadrons build wider showers over a nuclear interaction length. The onion of tracker, electromagnetic and hadronic calorimeters, and outer muon chambers turns these processes into momentum, energy, and identity, with neutrinos inferred from missing transverse momentum.\n",{"path":25657,"title":25658,"module":25650,"summary":25659},"\u002Fparticle-physics\u002Fexperiment\u002Fhow-discoveries-are-made","From Collisions to Discoveries","A discovery is a peak that survives statistics. Events are reconstructed into invariant masses, a signal accumulates as a bump over a smooth background, and its significance is judged by a p-value; the field's threshold is five sigma. The expected yield is a product — luminosity times cross section times branching ratio times acceptance and efficiency — that must be balanced by a trigger and data-reduction chain against an overwhelming rate. Worked reconstructions of $Z\\to\\ell\\ell$, the $J\u002F\\psi$, and the Higgs show the same peak-over-background logic at three scales.\n",{"path":25661,"title":25662,"module":25662,"summary":25663},"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fbeyond-standard-model","Beyond the Standard Model","The Standard Model leaves the four interactions ununified and the neutrinos massless, both now known to be wrong. Grand unification predicts the couplings merge near ten-to-the-sixteen GeV and the proton decays; supersymmetry pairs each particle with a superpartner; and the confirmed oscillation of neutrinos proves they carry mass, the first crack in the model.\n",{"path":25665,"title":25666,"module":25662,"summary":25667},"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fgrand-unified-theories","Grand Unified Theories and Proton Decay","The Standard Model gauge group is a product of three factors with three independent couplings. A grand unified theory embeds them in a single simple group — SU(5) is the minimal choice — so that one coupling runs into all three and the fractional quark charges follow from a tracelessness condition. The same embedding places quarks and leptons in shared multiplets, mediates baryon-number violation through superheavy gauge bosons, and predicts the proton decays with a lifetime that Super-Kamiokande has pushed past ten-to-the-thirty-four years.\n",{"path":25669,"title":25670,"module":25662,"summary":25671},"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fsupersymmetry","Supersymmetry","Supersymmetry relates fermions and bosons, pairing every Standard Model particle with a superpartner whose spin differs by one half. The pairing makes the scalar and fermion loop corrections to the Higgs mass cancel, removing the quadratic sensitivity to high scales; it sharpens the meeting of the three gauge couplings; and, when R-parity is conserved, it leaves the lightest superpartner stable and neutral, a natural dark-matter candidate. The LHC has excluded gluinos and light squarks below roughly two TeV.\n",{"path":25673,"title":25674,"module":25662,"summary":25675},"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fhierarchy-problem-naturalness","The Hierarchy Problem and Naturalness","The electroweak scale sits sixteen orders of magnitude below the Planck scale, and nothing in the Standard Model protects that gap. The Higgs mass squared picks up quadratic corrections proportional to the highest scale in the theory, so keeping it at the observed value requires the bare mass and its counterterm to cancel to some thirty significant figures. Naturalness treats that cancellation as a symptom of missing physics. Supersymmetry, compositeness, and extra dimensions each remove the quadratic sensitivity, but the LHC has found none of them at the predicted scale.\n",{"path":25677,"title":25678,"module":25662,"summary":25679},"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fdark-matter-candidates","Dark Matter and Particle Candidates","Flat galactic rotation curves, gravitational lensing, the cosmic microwave background, and structure formation all require about five times more matter than the visible baryons, none of it interacting electromagnetically. A stable weakly interacting particle of roughly weak-scale mass freezes out of the early universe with close to the observed abundance — the WIMP miracle — and is the leading candidate, with axions and sterile neutrinos as alternatives. Direct, indirect, and collider searches have so far only tightened the limits.\n",{"path":25681,"title":25682,"module":25662,"summary":25683},"\u002Fparticle-physics\u002Fbeyond-standard-model\u002Fmatter-antimatter-open-questions","Matter-Antimatter Asymmetry and Open Questions","The universe is made of matter, with about one extra baryon for every billion photons and no antimatter regions. Sakharov identified the three conditions any dynamical explanation must meet: baryon-number violation, C and CP violation, and a departure from thermal equilibrium. The Standard Model contains all three in principle, but its CP violation falls short by some ten orders of magnitude, so baryogenesis requires new physics — leptogenesis being the leading route. A closing survey collects the open questions and the experiments aimed at them.\n",{"path":25685,"title":25686,"module":6,"summary":6},"\u002Fparticle-physics","Particle Physics",{"path":25688,"title":25689,"module":25690,"summary":25691},"\u002Fastrophysics-cosmology\u002Forientation\u002Fthe-sun-and-stars","The Sun and the Life of Stars","Orientation","The Sun is the one star close enough to study in detail: its luminosity fixes a surface temperature of 5780 K, and the proton-proton fusion cycle in its 1.5-million-kelvin core supplies its power. Measuring other stars needs the magnitude scale, parallax, and the distance ladder; plotting luminosity against temperature builds the Hertzsprung-Russell diagram, on which a star's mass sets its lifetime and its evolutionary track off the main sequence.\n",{"path":25693,"title":25694,"module":25690,"summary":25695},"\u002Fastrophysics-cosmology\u002Forientation\u002Fstellar-death-final-states","Cataclysmic Events and the Final States of Stars","A star's death is set by its mass. In close binaries, matter poured across the Roche lobe onto a white dwarf produces novae and, at the Chandrasekhar limit of 1.4 solar masses, a Type Ia supernova; a massive star fusing to an iron core collapses into a Type II supernova. The remnant is a white dwarf held by electron degeneracy, a neutron star held by neutron degeneracy, or, above the neutron-star limit, a black hole inside its Schwarzschild radius.\n",{"path":25697,"title":25698,"module":25690,"summary":25699},"\u002Fastrophysics-cosmology\u002Forientation\u002Fgalaxies-and-cosmology","Galaxies, Cosmology, and the Evolving Universe","Galaxies come in elliptical, spiral, and irregular forms, and their redshifts obey Hubble's law, evidence that space itself is expanding. The critical density and the density parameter decide whether the universe is open, flat, or closed; baryons, dark matter, and dark energy each contribute. The cosmic microwave background and primordial helium anchor the Big Bang, whose thermal history runs from inflation through nucleosynthesis to the atoms of today.\n",{"path":25701,"title":25702,"module":25703,"summary":25704},"\u002Fastrophysics-cosmology\u002Fobservational-foundations\u002Fmagnitudes-fluxes-and-the-distance-modulus","Magnitudes, Fluxes, and the Distance Modulus","Observational Foundations","The brightness of a star reaches us as a radiant flux that falls off as the inverse square of distance. The magnitude scale encodes flux logarithmically through the Pogson ratio; the apparent and absolute magnitudes differ by the distance modulus, which converts a measured brightness into a distance. The bolometric correction folds a filtered magnitude into a total luminosity, and the difference of two magnitudes in different bands, the color index, measures surface temperature.\n",{"path":25706,"title":25707,"module":25703,"summary":25708},"\u002Fastrophysics-cosmology\u002Fobservational-foundations\u002Fstellar-spectra-and-spectral-classification","Stellar Spectra and Spectral Classification","A stellar spectrum is a continuum crossed by absorption lines whose strengths are set by the temperature of the atmosphere. The Boltzmann factor governs how atoms populate excited states, and the Saha equation governs how they ionize; their product explains why each line, such as the hydrogen Balmer series, peaks in strength at a characteristic temperature. This behavior orders stars into the OBAFGKM sequence, and the luminosity classes of the MK system add a second dimension for surface gravity.\n",{"path":25710,"title":25711,"module":25703,"summary":25712},"\u002Fastrophysics-cosmology\u002Fobservational-foundations\u002Ftelescopes-and-detectors-across-the-spectrum","Telescopes and Detectors Across the Spectrum","A telescope collects light in proportion to its collecting area and resolves detail down to the diffraction limit set by its aperture and the observing wavelength. The atmosphere blurs and blocks large parts of the spectrum, which drives the choice between ground and space and between refractors, reflectors, and radio dishes. CCDs record the light with high quantum efficiency, and interferometry synthesizes an aperture as large as the separation of two telescopes.\n",{"path":25714,"title":25715,"module":25703,"summary":25716},"\u002Fastrophysics-cosmology\u002Fobservational-foundations\u002Fthe-cosmic-distance-ladder","The Cosmic Distance Ladder","No single method measures distances from the nearest stars to the far reaches of the universe. Instead a ladder of overlapping techniques, each calibrated by the one below it, extends the scale rung by rung: trigonometric parallax, main-sequence fitting, pulsating variables, the tip of the red-giant branch, the Tully-Fisher relation, and Type Ia supernovae. Each rung inherits the uncertainty of every rung beneath it, so the whole chain sets the accuracy of the Hubble constant.\n",{"path":25718,"title":25719,"module":25720,"summary":25721},"\u002Fastrophysics-cosmology\u002Fradiation-and-matter\u002Fblackbody-radiation-and-specific-intensity","Blackbody Radiation and Specific Intensity","Radiation and Matter","Specific intensity is the fundamental measure of a radiation field: energy per unit area, time, frequency, and solid angle. It is conserved along a ray in empty space, and its angular moments give the mean intensity, flux, and radiation pressure. In thermal equilibrium the intensity equals the Planck function, whose limits and integrals reproduce the Rayleigh-Jeans law, the Wien law, Stefan-Boltzmann, and Wien's displacement law.\n",{"path":25723,"title":25724,"module":25720,"summary":25725},"\u002Fastrophysics-cosmology\u002Fradiation-and-matter\u002Fradiative-transfer-and-the-transfer-equation","Radiative Transfer and the Transfer Equation","Along a ray, matter adds intensity through emission and removes it through absorption. Measuring path length in optical depth turns this into the transfer equation, whose formal solution superposes an attenuated background on the source function integrated along the line of sight. In local thermodynamic equilibrium the source function is the Planck function, and the Eddington-Barbier relation shows that the emergent intensity samples the source function at optical depth of order unity, explaining absorption lines and solar limb darkening.\n",{"path":25727,"title":25728,"module":25720,"summary":25729},"\u002Fastrophysics-cosmology\u002Fradiation-and-matter\u002Fspectral-line-formation-and-broadening","Spectral-Line Formation and Broadening","A spectral line is a bound-bound transition whose strength is set by an oscillator strength and whose shape is set by three broadening mechanisms: the Lorentzian natural and collisional wings, the Gaussian thermal Doppler core, and their Voigt convolution. Equivalent width measures the total absorption, and the curve of growth relates it to the number of absorbers through a linear, saturated, and damping regime, turning line strengths into abundances.\n",{"path":25731,"title":25732,"module":25720,"summary":25733},"\u002Fastrophysics-cosmology\u002Fradiation-and-matter\u002Fopacity-and-the-rosseland-mean","Opacity Sources and the Rosseland Mean","Stellar opacity comes from four processes: bound-bound line absorption, bound-free photoionization, free-free absorption, and electron scattering. The bound-free and free-free terms follow a Kramers law, electron scattering sets a frequency-flat floor, and the negative hydrogen ion dominates cool photospheres. The Rosseland mean averages these harmonically, weighting transparent frequencies because they carry the flux, and its value fixes the radiative temperature gradient and decides where a star becomes convective.\n",{"path":25735,"title":25736,"module":25737,"summary":25738},"\u002Fastrophysics-cosmology\u002Fstellar-structure\u002Fhydrostatic-equilibrium-and-the-virial-theorem","Hydrostatic Equilibrium and the Virial Theorem","Stellar Structure","A star holds itself up by balancing the inward pull of gravity against an outward pressure gradient. This balance, hydrostatic equilibrium, fixes a lower bound on the central pressure and, combined with the gravitational potential energy, yields the virial theorem. The virial relation gives a star a negative heat capacity, so that losing energy makes it hotter, and sets the Kelvin-Helmholtz timescale over which contraction alone can power the Sun.\n",{"path":25740,"title":25741,"module":25737,"summary":25742},"\u002Fastrophysics-cosmology\u002Fstellar-structure\u002Fthe-equations-of-stellar-structure","The Equations of Stellar Structure","A static star is described by four coupled first-order differential equations in the interior mass or radius: mass conservation, hydrostatic equilibrium, energy generation, and energy transport. Closed with an equation of state, opacity, and reaction rates, and subject to central and surface boundary conditions, they determine the structure uniquely from mass and composition, the Vogt-Russell theorem. Energy moves by radiation until the temperature gradient exceeds the Schwarzschild limit, where convection takes over.\n",{"path":25744,"title":25745,"module":25737,"summary":25746},"\u002Fastrophysics-cosmology\u002Fstellar-structure\u002Fthe-equation-of-state-and-polytropes","The Equation of State and Polytropes","Stellar pressure comes from gas, radiation, and, at high density, degenerate electrons. When pressure depends on density as a power law, hydrostatic equilibrium reduces to the Lane-Emden equation, whose solutions describe polytropes of index n. The relativistic degenerate case, n equal to three, gives a mass independent of radius, the Chandrasekhar mass. Eddington's standard model treats a radiation-supported star as an n equal to three polytrope and yields the quartic relating radiation fraction to mass.\n",{"path":25748,"title":25749,"module":25737,"summary":25750},"\u002Fastrophysics-cosmology\u002Fstellar-structure\u002Fthe-standard-solar-model","The Standard Solar Model","The standard solar model integrates the structure equations for one solar mass and calibrates the composition and convection parameter to reproduce the Sun's observed luminosity, radius, and age. Helioseismology tests the model's sound speed through the Sun's acoustic p-mode oscillations, and the model predicts a neutrino flux by production channel. The measured deficit, the solar-neutrino problem, is resolved by matter-enhanced flavor oscillation, confirmed when SNO measured the total flux across all flavors.\n",{"path":25752,"title":25753,"module":25754,"summary":25755},"\u002Fastrophysics-cosmology\u002Fnuclear-astrophysics\u002Fthermonuclear-reaction-rates-and-the-gamow-peak","Thermonuclear Reaction Rates and the Gamow Peak","Nuclear Astrophysics","Stellar fusion proceeds only by quantum tunneling through the Coulomb barrier, because thermal energies are a thousand times smaller than the barrier height. The reaction rate is an integral over the Maxwell–Boltzmann distribution and the tunneling probability, whose product is sharply peaked at the Gamow energy. The astrophysical S-factor isolates the nuclear physics from the barrier penetration, and the steep temperature dependence follows from the width and position of the Gamow peak.\n",{"path":25757,"title":25758,"module":25754,"summary":25759},"\u002Fastrophysics-cosmology\u002Fnuclear-astrophysics\u002Fhydrogen-burning-pp-chains-and-cno","Hydrogen Burning: pp Chains and the CNO Cycle","Four protons fuse into one helium-4 nucleus, releasing 26.7 MeV, through two competing networks. The pp chain begins with a weak-interaction bottleneck and branches three ways; the CNO cycle uses carbon, nitrogen, and oxygen as catalysts and is limited by nitrogen-14 proton capture. Their steep and gentle temperature dependences cross near 1.8e7 K, which divides pp-powered lower-main-sequence stars from CNO-powered upper-main-sequence stars.\n",{"path":25761,"title":25762,"module":25754,"summary":25763},"\u002Fastrophysics-cosmology\u002Fnuclear-astrophysics\u002Fhelium-burning-and-the-triple-alpha-process","Helium Burning and the Triple-Alpha Process","Helium fuses to carbon in two steps through the unbound beryllium-8 nucleus and a resonant excited state of carbon-12, the Hoyle state, whose existence was predicted from the observed carbon abundance. The rate scales as roughly the fortieth power of temperature, and in a degenerate low-mass core this drives the runaway helium flash. A competing alpha capture on carbon-12 sets the carbon-to-oxygen ratio and the composition of the resulting white dwarf.\n",{"path":25765,"title":25766,"module":25754,"summary":25767},"\u002Fastrophysics-cosmology\u002Fnuclear-astrophysics\u002Fadvanced-burning-and-neutron-capture-nucleosynthesis","Advanced Burning, the Iron Peak, and the s\u002Fr Processes","Massive stars burn carbon, neon, oxygen, and silicon in ever-shorter stages, building an onion-shell interior and reaching nuclear statistical equilibrium at the iron peak, where the binding-energy-per-nucleon curve turns over and fusion can release no more energy. Elements beyond iron form by neutron capture: the slow s-process in AGB stars tracks the valley of stability, while the rapid r-process in supernovae and neutron-star mergers builds the heaviest nuclei far from it.\n",{"path":25769,"title":25770,"module":25771,"summary":25772},"\u002Fastrophysics-cosmology\u002Fism-and-star-formation\u002Fphases-of-the-interstellar-medium","The Phases of the Interstellar Medium","The Interstellar Medium","The gas between the stars separates into distinct thermal phases, from cold molecular clouds at 10 K to a diffuse million-degree corona, held near a common pressure by a balance of photoelectric heating and radiative cooling. Neutral hydrogen is traced by the 21-cm hyperfine line, dust reddens and extinguishes starlight along a characteristic wavelength law, and the ultraviolet output of hot stars carves ionized Strömgren spheres out of the surrounding gas.\n",{"path":25774,"title":25775,"module":25771,"summary":25776},"\u002Fastrophysics-cosmology\u002Fism-and-star-formation\u002Fmolecular-clouds-and-gravitational-collapse","Molecular Clouds and Gravitational Collapse","Stars form in cold, dense molecular clouds when self-gravity overcomes thermal and magnetic support. The virial theorem fixes the Jeans mass and length at which a clump becomes unstable, the free-fall time sets how fast it collapses, and a fragmentation cascade — cut off at a minimum mass by the onset of opacity — turns one cloud into a whole cluster, imprinting the stellar initial mass function.\n",{"path":25778,"title":25779,"module":25771,"summary":25780},"\u002Fastrophysics-cosmology\u002Fism-and-star-formation\u002Fprotostars-and-the-pre-main-sequence","Protostars and Pre-Main-Sequence Evolution","A collapsing core becomes optically thick and forms a protostar that grows by accretion through a disk while driving bipolar outflows. The newborn star appears on the birthline and contracts down the fully convective Hayashi track, then crosses the radiative Henyey track to the zero-age main sequence, powered by gravitational contraction until hydrogen ignites. Below about 0.08 solar masses degeneracy halts contraction before ignition, dividing stars from brown dwarfs.\n",{"path":25782,"title":25783,"module":25784,"summary":25785},"\u002Fastrophysics-cosmology\u002Fstellar-evolution\u002Fthe-main-sequence-and-its-structure","The Main Sequence and Its Structure","Stellar Evolution","A star settles onto the zero-age main sequence when core hydrogen ignition halts contraction. Homology scaling of the structure equations reproduces the mass–luminosity relation, and the burning mode splits the sequence into an upper branch with a convective core and a lower branch with a convective envelope. The main-sequence lifetime falls steeply with mass, and the turnoff of a coeval cluster serves as a clock.\n",{"path":25787,"title":25788,"module":25784,"summary":25789},"\u002Fastrophysics-cosmology\u002Fstellar-evolution\u002Fpost-main-sequence-low-mass-evolution","Post-Main-Sequence Evolution of Low-Mass Stars","When a low-mass star exhausts core hydrogen, burning moves to a shell, the core contracts, and the envelope swells into a red giant. A degenerate helium core ignites in a flash, settles onto the horizontal branch, and after a second contraction the star climbs the asymptotic giant branch with two burning shells. Thermal pulses and dredge-up enrich the surface, and mass loss ejects a planetary nebula, leaving a carbon–oxygen white dwarf.\n",{"path":25791,"title":25792,"module":25784,"summary":25793},"\u002Fastrophysics-cosmology\u002Fstellar-evolution\u002Fthe-evolution-of-massive-stars","The Evolution of Massive Stars","Stars above about eight solar masses burn through hydrogen, helium, carbon, neon, oxygen, and silicon in stages that grow shorter as neutrino losses accelerate contraction. The interior becomes an onion of concentric burning shells around an inert iron core. Radiation pressure near the Eddington limit drives fierce winds that can strip the hydrogen envelope entirely, and silicon burning builds an iron core toward the threshold of collapse.\n",{"path":25795,"title":25796,"module":25784,"summary":25797},"\u002Fastrophysics-cosmology\u002Fstellar-evolution\u002Fstellar-pulsation-and-the-instability-strip","Stellar Pulsation and the Instability Strip","Radial pulsation is a standing sound wave whose period scales inversely with the square root of the mean density. The kappa mechanism, an opacity valve seated in the helium partial-ionization zone, turns a star into a heat engine that pumps the oscillation. Stars in the instability strip pulsate as Cepheids, RR Lyrae, and Mira variables, and the Cepheid period–luminosity relation calibrates the distance ladder.\n",{"path":25799,"title":24215,"module":25800,"summary":25801},"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fwhite-dwarfs-and-the-chandrasekhar-limit","Stellar Death and Compact Remnants","A white dwarf is held up by the degeneracy pressure of its electrons, a quantum-mechanical stiffness that survives to zero temperature. Filling the Fermi sea sets a pressure that scales as density to the five-thirds power when the electrons are slow and only four-thirds when they are relativistic. The softer relativistic law produces the inverted mass-radius relation and a maximum mass, the Chandrasekhar limit near 1.4 solar masses, above which no cold equilibrium exists. Cooling and crystallization then turn the white-dwarf population into a clock for the Galactic disk.\n",{"path":25803,"title":25804,"module":25800,"summary":25805},"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fcore-collapse-supernovae","Core-Collapse Supernovae","When a massive star builds an iron core past the Chandrasekhar mass, degeneracy fails and the core collapses in less than a second. Photodisintegration and electron capture remove pressure support and neutronize the matter; the collapse halts abruptly at nuclear density, launching a shock that stalls and is revived by neutrino heating. The event is a Type II or stripped-envelope Ib\u002FIc supernova, and the neutrinos from SN 1987A confirmed the picture directly.\n",{"path":25807,"title":25808,"module":25800,"summary":25809},"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fthermonuclear-supernovae-type-ia","Thermonuclear Supernovae","A carbon-oxygen white dwarf driven toward the Chandrasekhar mass ignites its degenerate fuel and unbinds itself in a thermonuclear runaway, the Type Ia supernova. The light curve is powered by the radioactive decay of nickel-56 to cobalt-56 to iron-56, and the Phillips relation between peak brightness and decline rate makes these events standardizable candles. Their near-uniform luminosity turns them into the distance indicators that revealed cosmic acceleration.\n",{"path":25811,"title":25812,"module":25800,"summary":25813},"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fneutron-stars-and-pulsars","Neutron Stars and Pulsars","A neutron star is held up by neutron degeneracy and the repulsive nuclear force, with a maximum mass, the Tolman-Oppenheimer-Volkoff limit, set by an uncertain dense-matter equation of state. Its rotating magnetic dipole sweeps a beam past Earth as a pulsar, and magnetic braking traces a track across the period-period- derivative diagram. Millisecond pulsars, magnetars, glitches, and the orbital decay of the Hulse-Taylor binary follow from the same structure.\n",{"path":25815,"title":25816,"module":25800,"summary":25817},"\u002Fastrophysics-cosmology\u002Fstellar-death-and-compact-remnants\u002Fblack-holes-schwarzschild-and-kerr","Black Holes, Schwarzschild and Kerr","Above the neutron-star mass limit gravity wins completely and the remnant is a black hole. The Schwarzschild solution gives the event horizon, gravitational redshift, and time dilation; the innermost stable circular orbit sets the efficiency of accretion. Rotating Kerr black holes drag spacetime and carry an ergosphere. Stellar-mass black holes are found in X-ray binaries, and the Event Horizon Telescope has imaged the shadow of a supermassive one.\n",{"path":25819,"title":25820,"module":25821,"summary":25822},"\u002Fastrophysics-cosmology\u002Fbinaries-and-gravitational-waves\u002Fbinary-systems-and-mass-transfer","Binary Systems and Mass Transfer","Binaries and Gravitational Waves","Most stars are born in pairs, and a binary is the only setting where a stellar mass can be measured directly. Visual, spectroscopic, and eclipsing binaries each expose a different combination of the orbital elements, and together they calibrate the mass-luminosity relation. When one star swells to fill its Roche lobe, gas streams through the inner Lagrange point onto its companion. Conservative transfer widens or shrinks the orbit depending on the mass ratio, and the sign of that response explains the Algol paradox.\n",{"path":25824,"title":25825,"module":25821,"summary":25826},"\u002Fastrophysics-cosmology\u002Fbinaries-and-gravitational-waves\u002Faccreting-compact-objects","Accreting Compact Objects","Gas falling onto a compact object converts gravitational binding energy into radiation with an efficiency set by the depth of the potential well, up to tens of percent of the rest mass for a neutron star or black hole. Angular momentum forces the flow into a disk, and viscous dissipation gives a temperature profile that falls as radius to the minus three-quarters, producing a multicolor blackbody spectrum. Radiation pressure caps the steady luminosity at the Eddington limit. Unstable nuclear burning of the accreted fuel powers classical novae on white dwarfs and Type I X-ray bursts on neutron stars.\n",{"path":25828,"title":25829,"module":25821,"summary":25830},"\u002Fastrophysics-cosmology\u002Fbinaries-and-gravitational-waves\u002Fgravitational-waves-from-inspiraling-binaries","Gravitational Waves from Inspiraling Binaries","A time-varying mass quadrupole radiates gravitational waves, ripples in spacetime that stretch and squeeze a ring of free masses along two polarizations. The radiated power drains a binary's orbital energy, shrinking the orbit and sweeping the wave frequency upward in a chirp whose rate fixes the chirp mass. Laser interferometers with kilometre arms measure the resulting strain of order ten to the minus twenty-one. The first detection, GW150914, matched a template for two merging black holes near thirty solar masses each.\n",{"path":25832,"title":25833,"module":25821,"summary":25834},"\u002Fastrophysics-cosmology\u002Fbinaries-and-gravitational-waves\u002Fmultimessenger-astronomy-and-gamma-ray-bursts","Multimessenger Astronomy and Gamma-Ray Bursts","Gamma-ray bursts split into two populations: long bursts from the collapse of massive stars and short bursts from merging compact objects. The compactness problem forces the emitting plasma to move at ultra-relativistic speed, beaming the radiation into a narrow jet. The neutron-star merger GW170817 tied a gravitational chirp to a short gamma-ray burst, a radioactive kilonova, and a broadband afterglow, confirming that mergers forge r-process elements. A merger with a measured redshift is a standard siren that reads the Hubble constant from gravitational data alone.\n",{"path":25836,"title":25837,"module":25838,"summary":25839},"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Fthe-milky-way","The Milky Way Galaxy","Galaxies and Dark Matter","The Galaxy resolves into a thin disk of gas and young stars, a central bar and bulge, and a diffuse old halo studded with globular clusters. Star counts and the reddening of distant light map these components, while the differential rotation of the disk — encoded in the Oort constants and the flat rotation curve — measures the enclosed mass and reveals more than the stars can account for. Spiral arms are density waves, not material structures, and the innermost stellar orbits around Sgr A* weigh a four-million-solar-mass black hole.\n",{"path":25841,"title":25842,"module":25838,"summary":25843},"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Fgalaxy-morphology-and-classification","Galaxy Morphology and Classification","Galaxies sort along the Hubble tuning fork from smooth ellipticals through lenticulars to grand-design and barred spirals, with irregulars off the end. The light of a spheroid follows the de Vaucouleurs quarter-power law while a disk fades exponentially, and the general Sérsic profile interpolates between them. Virial scaling relations — Tully–Fisher for disks, Faber–Jackson and the fundamental plane for spheroids — tie luminosity to internal motions, and the Schechter function fixes the abundance of galaxies as a function of luminosity.\n",{"path":25845,"title":25846,"module":25838,"summary":25847},"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Fgalaxy-rotation-curves-and-dark-matter","Galaxy Rotation Curves and Dark Matter","The rotation curves of disk galaxies stay flat far beyond the light, demanding an extended halo whose density falls as the inverse square of radius. Decomposing the curve into disk, bulge, and halo, and fitting isothermal or NFW profiles, quantifies the missing mass. Gravitational lensing weighs the same mass without dynamics, the mass-to-light ratio climbs from stars to clusters, and the Bullet Cluster separates the collisionless dark matter from the colliding gas — evidence that MOND strains to match.\n",{"path":25849,"title":25850,"module":25838,"summary":25851},"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Factive-galactic-nuclei-and-supermassive-black-holes","Active Galactic Nuclei","A small fraction of galaxies pour out enormous luminosity from a region smaller than the solar system. Accretion onto a supermassive black hole, limited by the Eddington balance of radiation pressure and gravity, powers the Seyferts, quasars, radio galaxies, and blazars — one engine seen from different angles through an obscuring torus. Relativistic jets produce apparent superluminal motion, reverberation mapping and stellar dynamics weigh the central mass, and the M–sigma relation ties that mass to the host bulge.\n",{"path":25853,"title":25854,"module":25838,"summary":25855},"\u002Fastrophysics-cosmology\u002Fgalaxies\u002Fgalaxy-clusters-and-large-scale-structure","Galaxy Clusters and Large-Scale Structure","Galaxies gather into groups and rich clusters bound by a common dark halo and filled with hot X-ray gas. Three independent probes — the virial theorem, the hydrostatic X-ray temperature, and gravitational lensing — agree on a mass that dwarfs the stars. On the largest scales galaxies trace a cosmic web of filaments, walls, and voids, quantified by the two-point correlation function, whose baryon acoustic oscillation bump provides a standard ruler for cosmology.\n",{"path":25857,"title":25858,"module":25859,"summary":25860},"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fthe-expanding-universe-and-hubbles-law","The Expanding Universe and Hubble's Law","Cosmic Expansion and Dynamics","The universe is homogeneous and isotropic on large scales, so its expansion is captured by a single function of time, the scale factor. Comoving coordinates stay fixed while proper distances grow in proportion to the scale factor, producing Hubble's law and a cosmological redshift that measures stretched space rather than a Doppler shift. A Newtonian energy argument reproduces the dynamics, and the same finite, expanding cosmos resolves Olbers' paradox.\n",{"path":25862,"title":25863,"module":25859,"summary":25864},"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fthe-frw-metric-and-cosmological-redshift","The FRW Metric and Cosmological Redshift","The geometry of a homogeneous, isotropic universe is fixed by symmetry to the Robertson-Walker metric, with the entire freedom reduced to a scale factor and a single curvature constant selecting an open, flat, or closed space. From the metric the null geodesic of light gives comoving distance, the exact cosmological redshift, and the distinction between the proper distance we cannot measure and the redshift we can.\n",{"path":25866,"title":22942,"module":25859,"summary":25867},"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fthe-friedmann-equations-and-cosmic-dynamics","The scale factor obeys the Friedmann equation, the acceleration equation, and the fluid equation, only two of which are independent. An equation of state fixes how each component behaves under expansion, so radiation dilutes as the inverse fourth power of the scale factor, matter as the inverse cube, and vacuum energy not at all. The critical density defines the density parameters, and the deceleration parameter encodes whether gravity or dark energy is winning.\n",{"path":25869,"title":25870,"module":25859,"summary":25871},"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fcosmological-models-and-distances","Cosmological Models and Distances","Integrating the Friedmann equation for particular mixtures gives the benchmark models, from the matter-only Einstein-de Sitter universe to the concordance Lambda-CDM, each with its own scale-factor history and age. Because the redshift is the only direct observable, several distance measures diverge at high redshift, and the angular-diameter distance even turns over so that the most distant objects look larger. The horizon and lookback time set what is causally and observationally reachable.\n",{"path":25873,"title":25874,"module":25859,"summary":25875},"\u002Fastrophysics-cosmology\u002Fcosmology-expansion-and-dynamics\u002Fdark-energy-and-the-accelerating-universe","Dark Energy and the Accelerating Universe","In 1998 two teams found that distant Type Ia supernovae are fainter than a decelerating universe predicts, revealing that the expansion is accelerating and that a component with negative pressure dominates the energy budget. The simplest candidate is the cosmological constant, or vacuum energy, with an equation of state near minus one. It works observationally but leaves two deep puzzles: why the vacuum energy is a hundred and twenty orders of magnitude smaller than expected, and why it is comparable to the matter density just now.\n",{"path":25877,"title":25878,"module":25879,"summary":25880},"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fthe-thermal-history-of-the-universe","The Thermal History of the Universe","The Hot Big Bang","Running the expansion backward compresses and heats the universe, so its past is a sequence of thermal epochs set by temperature. Temperature scales as the inverse scale factor; species stay in equilibrium while their interaction rate exceeds the expansion rate and freeze out when it drops below. The effective degrees of freedom count the relativistic species and step down through mass thresholds, and neutrino decoupling just before electron-positron annihilation leaves a relic neutrino background slightly cooler than the photons.\n",{"path":25882,"title":25883,"module":25879,"summary":25884},"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fbig-bang-nucleosynthesis","Big Bang Nucleosynthesis","In the first three minutes the weak interactions freeze out the neutron-to-proton ratio, and once deuterium survives photodissociation a fast reaction network converts nearly all free neutrons into helium-4. The primordial abundances of deuterium, helium-3, helium-4, and lithium-7 depend on a single free parameter, the baryon-to-photon ratio, so measuring them fixes the baryon density. The predictions match observation across nine decades of abundance, with a persistent discrepancy in lithium-7.\n",{"path":25886,"title":25887,"module":25879,"summary":25888},"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Frecombination-and-the-cosmic-microwave-background","Recombination and the Cosmic Microwave Background","As the universe cooled through a few thousand kelvin the free electrons bound to protons, and the Saha equation tracks the falling ionization fraction. Once the plasma neutralized, photons stopped scattering and streamed freely from a spherical surface of last scattering at redshift about 1100. Those photons are the cosmic microwave background, an almost perfect blackbody at 2.725 kelvin with a dipole from our motion through it.\n",{"path":25890,"title":25891,"module":25879,"summary":25892},"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fcmb-anisotropies-and-cosmological-parameters","CMB Anisotropies and Cosmological Parameters","The cosmic microwave background carries temperature fluctuations at the ten-parts-per-million level, imprinted by sound waves in the photon-baryon plasma before recombination. Decomposed into spherical harmonics, the fluctuations form an angular power spectrum whose acoustic peaks encode the geometry and contents of the universe: the first peak fixes spatial flatness, the odd-even peak ratio the baryon density, and the third peak the dark-matter density. Polarization adds an independent channel, and the Planck measurements pin the concordance parameters.\n",{"path":25894,"title":25895,"module":25879,"summary":25896},"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fcosmic-inflation","Cosmic Inflation","The hot Big Bang leaves three initial-condition puzzles unexplained: why causally disconnected patches share a temperature, why the geometry is so nearly flat, and why no magnetic monopoles are seen. A brief epoch of accelerated expansion driven by a slowly rolling scalar field solves all three by stretching a small causal patch across the observable universe. The same accelerated expansion freezes quantum fluctuations into a near-scale-invariant spectrum of density perturbations, seeding all later structure.\n",{"path":25898,"title":25899,"module":25879,"summary":25900},"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fstructure-formation-and-the-growth-of-perturbations","Structure Formation and the Growth of Perturbations","The near-uniform early universe grew its galaxies and clusters by gravitational instability acting on the tiny inflationary perturbations. In an expanding background the growth is slowed to a power law rather than the exponential of a static medium; perturbations stall during radiation domination and grow with the scale factor once matter dominates. The transfer function turns the primordial spectrum into the processed matter power spectrum, and cold dark matter builds structure from the bottom up.\n",{"path":25902,"title":25903,"module":25879,"summary":25904},"\u002Fastrophysics-cosmology\u002Fthe-hot-big-bang\u002Fdark-matter-dark-energy-and-open-questions","Dark Matter, Dark Energy, and Open Questions","Five independent lines of evidence converge on a universe whose energy budget is dominated by dark energy and dark matter, with ordinary baryons a small remainder. The candidate particles for dark matter range from WIMPs to axions to sterile neutrinos, each with its own detection strategy. The concordance model fits the data with six parameters but leaves the nature of dark energy, the Hubble tension, small-scale structure, and the matter-antimatter asymmetry unexplained.\n",{"path":25906,"title":25907,"module":6,"summary":6},"\u002Fastrophysics-cosmology","Astrophysics & Cosmology",{"path":25909,"title":25910,"module":6,"summary":6},"\u002Fcolophon","Colophon",{"path":16791,"title":25912,"module":6,"summary":6},"Study Notes",[25914,25927,25936,25950,25963,25974,25999,26012,26029,26038,26056,26082],{"module":23788,"moduleNumber":25915,"slug":25916,"lessons":25917},1,"mathematical-background",[25918,25920,25922,25924],{"title":22429,"path":23787,"lessonNumber":25915,"topics":25919,"summary":23789},[23788],{"title":23792,"path":23791,"lessonNumber":20265,"topics":25921,"summary":23793},[23788],{"title":23796,"path":23795,"lessonNumber":20271,"topics":25923,"summary":23797},[23788],{"title":21788,"path":23799,"lessonNumber":25925,"topics":25926,"summary":23800},4,[23788],{"module":21235,"moduleNumber":20265,"slug":25928,"lessons":25929},"foundations",[25930,25932,25934],{"title":23803,"path":23802,"lessonNumber":25915,"topics":25931,"summary":23804},[21235],{"title":23807,"path":23806,"lessonNumber":20265,"topics":25933,"summary":23808},[21235],{"title":23811,"path":23810,"lessonNumber":20271,"topics":25935,"summary":23812},[21235],{"module":20293,"moduleNumber":20271,"slug":25937,"lessons":25938},"neural-networks",[25939,25941,25943,25945,25947],{"title":23814,"path":109,"lessonNumber":25915,"topics":25940,"summary":23815},[20293],{"title":5,"path":20295,"lessonNumber":20265,"topics":25942,"summary":20316},[20293],{"title":23819,"path":23818,"lessonNumber":20271,"topics":25944,"summary":23820},[20293],{"title":23822,"path":317,"lessonNumber":25925,"topics":25946,"summary":23823},[20293],{"title":23825,"path":17544,"lessonNumber":25948,"topics":25949,"summary":23826},5,[20293],{"module":23830,"moduleNumber":25925,"slug":25951,"lessons":25952},"optimization",[25953,25955,25957,25959,25961],{"title":23829,"path":23828,"lessonNumber":25915,"topics":25954,"summary":23831},[23830],{"title":23834,"path":23833,"lessonNumber":20265,"topics":25956,"summary":23835},[23830],{"title":23837,"path":15181,"lessonNumber":20271,"topics":25958,"summary":23838},[23830],{"title":23841,"path":23840,"lessonNumber":25925,"topics":25960,"summary":23842},[23830],{"title":23845,"path":23844,"lessonNumber":25948,"topics":25962,"summary":23846},[23830],{"module":23850,"moduleNumber":25948,"slug":25964,"lessons":25965},"regularization",[25966,25968,25970,25972],{"title":23849,"path":23848,"lessonNumber":25915,"topics":25967,"summary":23851},[23850],{"title":23854,"path":23853,"lessonNumber":20265,"topics":25969,"summary":23855},[23850],{"title":23858,"path":23857,"lessonNumber":20271,"topics":25971,"summary":23859},[23850],{"title":23862,"path":23861,"lessonNumber":25925,"topics":25973,"summary":23863},[23850],{"module":23867,"moduleNumber":25975,"slug":25976,"lessons":25977},6,"architectures",[25978,25980,25982,25984,25986,25988,25990,25993,25996],{"title":23866,"path":23865,"lessonNumber":25915,"topics":25979,"summary":23868},[23867],{"title":23871,"path":23870,"lessonNumber":20265,"topics":25981,"summary":23872},[23867],{"title":23875,"path":23874,"lessonNumber":20271,"topics":25983,"summary":23876},[23867],{"title":23879,"path":23878,"lessonNumber":25925,"topics":25985,"summary":23880},[23867],{"title":23883,"path":23882,"lessonNumber":25948,"topics":25987,"summary":23884},[23867],{"title":23887,"path":23886,"lessonNumber":25975,"topics":25989,"summary":23888},[23867],{"title":23891,"path":23890,"lessonNumber":25991,"topics":25992,"summary":23892},7,[23867],{"title":23895,"path":23894,"lessonNumber":25994,"topics":25995,"summary":23896},8,[23867],{"title":23899,"path":23898,"lessonNumber":25997,"topics":25998,"summary":23900},9,[23867],{"module":23904,"moduleNumber":25991,"slug":26000,"lessons":26001},"theory",[26002,26004,26006,26008,26010],{"title":23903,"path":23902,"lessonNumber":25915,"topics":26003,"summary":23905},[23904],{"title":23908,"path":23907,"lessonNumber":20265,"topics":26005,"summary":23909},[23904],{"title":23912,"path":23911,"lessonNumber":20271,"topics":26007,"summary":23913},[23904],{"title":23916,"path":23915,"lessonNumber":25925,"topics":26009,"summary":23917},[23904],{"title":23920,"path":23919,"lessonNumber":25948,"topics":26011,"summary":23921},[23904],{"module":23925,"moduleNumber":25994,"slug":26013,"lessons":26014},"generative-models",[26015,26017,26019,26021,26023,26025,26027],{"title":23924,"path":23923,"lessonNumber":25915,"topics":26016,"summary":23926},[23925],{"title":23929,"path":23928,"lessonNumber":20265,"topics":26018,"summary":23930},[23925],{"title":23933,"path":23932,"lessonNumber":20271,"topics":26020,"summary":23934},[23925],{"title":23937,"path":23936,"lessonNumber":25925,"topics":26022,"summary":23938},[23925],{"title":23941,"path":23940,"lessonNumber":25948,"topics":26024,"summary":23942},[23925],{"title":23945,"path":23944,"lessonNumber":25975,"topics":26026,"summary":23946},[23925],{"title":23949,"path":23948,"lessonNumber":25991,"topics":26028,"summary":23950},[23925],{"module":23954,"moduleNumber":25997,"slug":26030,"lessons":26031},"probabilistic-methods",[26032,26034,26036],{"title":23953,"path":23952,"lessonNumber":25915,"topics":26033,"summary":23955},[23954],{"title":23958,"path":23957,"lessonNumber":20265,"topics":26035,"summary":23959},[23954],{"title":23962,"path":23961,"lessonNumber":20271,"topics":26037,"summary":23963},[23954],{"module":23967,"moduleNumber":26039,"slug":26040,"lessons":26041},10,"practical",[26042,26044,26046,26048,26050,26052,26054],{"title":23966,"path":23965,"lessonNumber":25915,"topics":26043,"summary":23968},[23967],{"title":23971,"path":23970,"lessonNumber":20265,"topics":26045,"summary":23972},[23967],{"title":23975,"path":23974,"lessonNumber":20271,"topics":26047,"summary":23976},[23967],{"title":23979,"path":23978,"lessonNumber":25925,"topics":26049,"summary":23980},[23967],{"title":23983,"path":23982,"lessonNumber":25948,"topics":26051,"summary":23984},[23967],{"title":23987,"path":23986,"lessonNumber":25975,"topics":26053,"summary":23988},[23967],{"title":23991,"path":23990,"lessonNumber":25991,"topics":26055,"summary":23992},[23967],{"module":23996,"moduleNumber":26057,"slug":26058,"lessons":26059},11,"large-models-and-agents",[26060,26062,26064,26066,26068,26070,26072,26074,26076,26078,26080],{"title":23995,"path":23994,"lessonNumber":25915,"topics":26061,"summary":23997},[23996],{"title":24000,"path":23999,"lessonNumber":20265,"topics":26063,"summary":24001},[23996],{"title":24004,"path":24003,"lessonNumber":20271,"topics":26065,"summary":24005},[23996],{"title":24008,"path":24007,"lessonNumber":25925,"topics":26067,"summary":24009},[23996],{"title":24012,"path":24011,"lessonNumber":25948,"topics":26069,"summary":24013},[23996],{"title":24016,"path":24015,"lessonNumber":25975,"topics":26071,"summary":24017},[23996],{"title":24020,"path":24019,"lessonNumber":25991,"topics":26073,"summary":24021},[23996],{"title":24024,"path":24023,"lessonNumber":25994,"topics":26075,"summary":24025},[23996],{"title":24028,"path":24027,"lessonNumber":25997,"topics":26077,"summary":24029},[23996],{"title":24032,"path":24031,"lessonNumber":26039,"topics":26079,"summary":24033},[23996],{"title":24036,"path":24035,"lessonNumber":26057,"topics":26081,"summary":24037},[23996],{"module":24041,"moduleNumber":26083,"slug":26084,"lessons":26085},12,"reinforcement-learning",[26086,26088,26090,26092,26094],{"title":24040,"path":24039,"lessonNumber":25915,"topics":26087,"summary":24042},[24041],{"title":24045,"path":24044,"lessonNumber":20265,"topics":26089,"summary":24046},[24041],{"title":24049,"path":24048,"lessonNumber":20271,"topics":26091,"summary":24050},[24041],{"title":24053,"path":24052,"lessonNumber":25925,"topics":26093,"summary":24054},[24041],{"title":24057,"path":24056,"lessonNumber":25948,"topics":26095,"summary":24058},[24041],"\u003Csvg style=\"width:100%;max-width:342.911px;height:auto\" xmlns=\"http:\u002F\u002Fwww.w3.org\u002F2000\u002Fsvg\" viewBox=\"-75 -75 257.183 139.444\">\u003Cg stroke=\"currentColor\" style=\"stroke-miterlimit:10;stroke-width:.4\">\u003Cg style=\"stroke-width:.8\">\u003Cpath fill=\"none\" d=\"M-58.537 6.714h207.95\"\u002F>\u003Cpath stroke=\"none\" d=\"m152.014 6.714-4.16-2.08 1.56 2.08-1.56 2.08\"\u002F>\u003Cg transform=\"translate(111.853 1.905)\">\u003Cpath d=\"M45.991 6.714L44.213 6.714L44.213 6.417Q44.487 6.417 44.655 6.370Q44.823 6.323 44.823 6.155L44.823 4.019Q44.823 3.804 44.766 3.708Q44.709 3.612 44.596 3.591Q44.483 3.569 44.237 3.569L44.237 3.273L45.436 3.187L45.436 6.155Q45.436 6.323 45.582 6.370Q45.729 6.417 45.991 6.417L45.991 6.714M44.549 1.792Q44.549 1.601 44.684 1.470Q44.819 1.339 45.014 1.339Q45.135 1.339 45.239 1.402Q45.342 1.464 45.405 1.568Q45.467 1.671 45.467 1.792Q45.467 1.987 45.336 2.122Q45.206 2.257 45.014 2.257Q44.815 2.257 44.682 2.124Q44.549 1.991 44.549 1.792M48.420 6.714L46.565 6.714L46.565 6.417Q46.838 6.417 47.006 6.370Q47.174 6.323 47.174 6.155L47.174 4.019Q47.174 3.804 47.112 3.708Q47.049 3.612 46.930 3.591Q46.811 3.569 46.565 3.569L46.565 3.273L47.756 3.187L47.756 3.921Q47.870 3.706 48.063 3.538Q48.256 3.370 48.495 3.278Q48.733 3.187 48.987 3.187Q50.155 3.187 50.155 4.265L50.155 6.155Q50.155 6.323 50.325 6.370Q50.495 6.417 50.764 6.417L50.764 6.714L48.909 6.714L48.909 6.417Q49.182 6.417 49.350 6.370Q49.518 6.323 49.518 6.155L49.518 4.280Q49.518 3.898 49.397 3.669Q49.276 3.441 48.924 3.441Q48.612 3.441 48.358 3.603Q48.104 3.765 47.957 4.034Q47.811 4.304 47.811 4.601L47.811 6.155Q47.811 6.323 47.981 6.370Q48.151 6.417 48.420 6.417L48.420 6.714M53.092 8.265L51.237 8.265L51.237 7.972Q51.506 7.972 51.674 7.927Q51.842 7.882 51.842 7.706L51.842 3.882Q51.842 3.675 51.686 3.622Q51.530 3.569 51.237 3.569L51.237 3.273L52.459 3.187L52.459 3.652Q52.690 3.429 53.004 3.308Q53.319 3.187 53.659 3.187Q54.131 3.187 54.536 3.433Q54.940 3.679 55.172 4.095Q55.405 4.511 55.405 4.987Q55.405 5.362 55.256 5.691Q55.108 6.019 54.838 6.271Q54.569 6.523 54.225 6.657Q53.881 6.792 53.522 6.792Q53.233 6.792 52.961 6.671Q52.690 6.550 52.483 6.339L52.483 7.706Q52.483 7.882 52.651 7.927Q52.819 7.972 53.092 7.972L53.092 8.265M52.483 4.050L52.483 5.890Q52.635 6.179 52.897 6.359Q53.159 6.538 53.467 6.538Q53.752 6.538 53.975 6.400Q54.198 6.261 54.350 6.030Q54.502 5.800 54.581 5.528Q54.659 5.257 54.659 4.987Q54.659 4.655 54.534 4.298Q54.409 3.941 54.161 3.704Q53.913 3.468 53.565 3.468Q53.241 3.468 52.946 3.624Q52.651 3.780 52.483 4.050M56.612 5.761L56.612 4.019Q56.612 3.804 56.549 3.708Q56.487 3.612 56.368 3.591Q56.248 3.569 56.002 3.569L56.002 3.273L57.248 3.187L57.248 5.737L57.248 5.761Q57.248 6.073 57.303 6.235Q57.358 6.398 57.508 6.468Q57.659 6.538 57.979 6.538Q58.409 6.538 58.682 6.200Q58.956 5.862 58.956 5.417L58.956 4.019Q58.956 3.804 58.893 3.708Q58.831 3.612 58.711 3.591Q58.592 3.569 58.346 3.569L58.346 3.273L59.592 3.187L59.592 5.972Q59.592 6.183 59.655 6.278Q59.717 6.374 59.836 6.396Q59.956 6.417 60.202 6.417L60.202 6.714L58.979 6.792L58.979 6.171Q58.811 6.460 58.530 6.626Q58.248 6.792 57.928 6.792Q56.612 6.792 56.612 5.761M61.272 5.753L61.272 3.562L60.569 3.562L60.569 3.308Q60.924 3.308 61.166 3.075Q61.409 2.843 61.520 2.495Q61.631 2.148 61.631 1.792L61.913 1.792L61.913 3.265L63.088 3.265L63.088 3.562L61.913 3.562L61.913 5.737Q61.913 6.058 62.032 6.286Q62.151 6.515 62.432 6.515Q62.612 6.515 62.729 6.392Q62.846 6.269 62.899 6.089Q62.952 5.909 62.952 5.737L62.952 5.265L63.233 5.265L63.233 5.753Q63.233 6.007 63.127 6.247Q63.022 6.487 62.825 6.640Q62.627 6.792 62.370 6.792Q62.053 6.792 61.801 6.669Q61.549 6.546 61.411 6.312Q61.272 6.077 61.272 5.753\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg style=\"stroke-width:.8\">\u003Cpath fill=\"none\" d=\"M43.893 60.774V-56.128\"\u002F>\u003Cpath stroke=\"none\" d=\"m43.893-58.728-2.08 4.16 2.08-1.56 2.08 1.56\"\u002F>\u003Cg transform=\"translate(-12.514 -70.73)\">\u003Cpath d=\"M44.131 5.019Q44.131 4.515 44.387 4.083Q44.643 3.652 45.079 3.400Q45.514 3.148 46.014 3.148Q46.401 3.148 46.743 3.292Q47.084 3.437 47.346 3.698Q47.608 3.960 47.750 4.296Q47.893 4.632 47.893 5.019Q47.893 5.511 47.629 5.921Q47.366 6.331 46.936 6.562Q46.506 6.792 46.014 6.792Q45.522 6.792 45.088 6.560Q44.655 6.327 44.393 5.919Q44.131 5.511 44.131 5.019M46.014 6.515Q46.471 6.515 46.723 6.292Q46.975 6.069 47.063 5.718Q47.151 5.366 47.151 4.921Q47.151 4.491 47.057 4.153Q46.963 3.816 46.709 3.609Q46.456 3.402 46.014 3.402Q45.366 3.402 45.122 3.818Q44.877 4.234 44.877 4.921Q44.877 5.366 44.965 5.718Q45.053 6.069 45.305 6.292Q45.557 6.515 46.014 6.515M49.061 5.761L49.061 4.019Q49.061 3.804 48.998 3.708Q48.936 3.612 48.817 3.591Q48.698 3.569 48.452 3.569L48.452 3.273L49.698 3.187L49.698 5.737L49.698 5.761Q49.698 6.073 49.752 6.235Q49.807 6.398 49.957 6.468Q50.108 6.538 50.428 6.538Q50.858 6.538 51.131 6.200Q51.405 5.862 51.405 5.417L51.405 4.019Q51.405 3.804 51.342 3.708Q51.280 3.612 51.161 3.591Q51.041 3.569 50.795 3.569L50.795 3.273L52.041 3.187L52.041 5.972Q52.041 6.183 52.104 6.278Q52.166 6.374 52.286 6.396Q52.405 6.417 52.651 6.417L52.651 6.714L51.428 6.792L51.428 6.171Q51.260 6.460 50.979 6.626Q50.698 6.792 50.377 6.792Q49.061 6.792 49.061 5.761M53.721 5.753L53.721 3.562L53.018 3.562L53.018 3.308Q53.373 3.308 53.616 3.075Q53.858 2.843 53.969 2.495Q54.081 2.148 54.081 1.792L54.362 1.792L54.362 3.265L55.538 3.265L55.538 3.562L54.362 3.562L54.362 5.737Q54.362 6.058 54.481 6.286Q54.600 6.515 54.881 6.515Q55.061 6.515 55.178 6.392Q55.295 6.269 55.348 6.089Q55.401 5.909 55.401 5.737L55.401 5.265L55.682 5.265L55.682 5.753Q55.682 6.007 55.577 6.247Q55.471 6.487 55.274 6.640Q55.077 6.792 54.819 6.792Q54.502 6.792 54.250 6.669Q53.998 6.546 53.860 6.312Q53.721 6.077 53.721 5.753M58.284 8.265L56.428 8.265L56.428 7.972Q56.698 7.972 56.866 7.927Q57.034 7.882 57.034 7.706L57.034 3.882Q57.034 3.675 56.877 3.622Q56.721 3.569 56.428 3.569L56.428 3.273L57.651 3.187L57.651 3.652Q57.881 3.429 58.196 3.308Q58.510 3.187 58.850 3.187Q59.323 3.187 59.727 3.433Q60.131 3.679 60.364 4.095Q60.596 4.511 60.596 4.987Q60.596 5.362 60.448 5.691Q60.299 6.019 60.030 6.271Q59.760 6.523 59.416 6.657Q59.073 6.792 58.713 6.792Q58.424 6.792 58.153 6.671Q57.881 6.550 57.674 6.339L57.674 7.706Q57.674 7.882 57.842 7.927Q58.010 7.972 58.284 7.972L58.284 8.265M57.674 4.050L57.674 5.890Q57.827 6.179 58.088 6.359Q58.350 6.538 58.659 6.538Q58.944 6.538 59.166 6.400Q59.389 6.261 59.541 6.030Q59.694 5.800 59.772 5.528Q59.850 5.257 59.850 4.987Q59.850 4.655 59.725 4.298Q59.600 3.941 59.352 3.704Q59.104 3.468 58.756 3.468Q58.432 3.468 58.137 3.624Q57.842 3.780 57.674 4.050M61.803 5.761L61.803 4.019Q61.803 3.804 61.741 3.708Q61.678 3.612 61.559 3.591Q61.440 3.569 61.194 3.569L61.194 3.273L62.440 3.187L62.440 5.737L62.440 5.761Q62.440 6.073 62.495 6.235Q62.549 6.398 62.700 6.468Q62.850 6.538 63.170 6.538Q63.600 6.538 63.873 6.200Q64.147 5.862 64.147 5.417L64.147 4.019Q64.147 3.804 64.084 3.708Q64.022 3.612 63.903 3.591Q63.784 3.569 63.538 3.569L63.538 3.273L64.784 3.187L64.784 5.972Q64.784 6.183 64.846 6.278Q64.909 6.374 65.028 6.396Q65.147 6.417 65.393 6.417L65.393 6.714L64.170 6.792L64.170 6.171Q64.002 6.460 63.721 6.626Q63.440 6.792 63.120 6.792Q61.803 6.792 61.803 5.761M66.463 5.753L66.463 3.562L65.760 3.562L65.760 3.308Q66.116 3.308 66.358 3.075Q66.600 2.843 66.711 2.495Q66.823 2.148 66.823 1.792L67.104 1.792L67.104 3.265L68.280 3.265L68.280 3.562L67.104 3.562L67.104 5.737Q67.104 6.058 67.223 6.286Q67.342 6.515 67.623 6.515Q67.803 6.515 67.920 6.392Q68.038 6.269 68.090 6.089Q68.143 5.909 68.143 5.737L68.143 5.265L68.424 5.265L68.424 5.753Q68.424 6.007 68.319 6.247Q68.213 6.487 68.016 6.640Q67.819 6.792 67.561 6.792Q67.245 6.792 66.993 6.669Q66.741 6.546 66.602 6.312Q66.463 6.077 66.463 5.753\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cpath fill=\"none\" d=\"M-52.846-35.966h193.478\" style=\"stroke-dasharray:3.0,3.0\"\u002F>\u003Cg transform=\"translate(101.695 -40.101)\">\u003Cpath d=\"M47.487 6.714L44.694 6.714L44.694 6.417Q45.756 6.417 45.756 6.155L45.756 1.987Q45.327 2.202 44.647 2.202L44.647 1.905Q45.666 1.905 46.182 1.394L46.327 1.394Q46.401 1.413 46.420 1.491L46.420 6.155Q46.420 6.417 47.487 6.417\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cpath fill=\"none\" d=\"M-52.846 49.393h193.478\" style=\"stroke-dasharray:3.0,3.0\"\u002F>\u003Cg transform=\"translate(101.695 45.257)\">\u003Cpath d=\"M46.245 5.265L43.991 5.265L43.991 4.714L46.245 4.714L46.245 5.265M50.319 6.714L47.526 6.714L47.526 6.417Q48.588 6.417 48.588 6.155L48.588 1.987Q48.159 2.202 47.479 2.202L47.479 1.905Q48.498 1.905 49.014 1.394L49.159 1.394Q49.233 1.413 49.252 1.491L49.252 6.155Q49.252 6.417 50.319 6.417\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cpath fill=\"none\" stroke=\"var(--tk-accent)\" d=\"m-50 6.655 2.376-.01 2.377-.012 2.377-.015 2.377-.017 2.377-.02 2.377-.025 2.377-.029 2.377-.033 2.377-.04 2.377-.047 2.377-.055 2.377-.065 2.377-.077 2.377-.09 2.377-.106 2.377-.125 2.377-.147 2.377-.172 2.377-.201 2.377-.236 2.377-.275 2.377-.32 2.377-.372 2.377-.43 2.377-.498 2.377-.573 2.377-.654 2.377-.747L18.932.42l2.377-.95 2.377-1.061 2.377-1.176 2.377-1.288 2.377-1.4 2.377-1.504 2.377-1.598 2.377-1.673 2.377-1.734 2.377-1.766 2.377-1.784 2.377-1.767 2.377-1.73 2.377-1.675 2.377-1.602 2.377-1.503 2.377-1.4 2.377-1.291 2.377-1.175 2.377-1.062 2.377-.947 2.377-.848 2.377-.748 2.377-.655 2.377-.572 2.377-.495 2.377-.432 2.377-.373 2.377-.321 2.377-.277 2.377-.233 2.377-.202 2.377-.17 2.377-.148 2.377-.128 2.377-.1 2.377-.094 2.377-.077 2.377-.062 2.377-.057 2.377-.046 2.377-.04 2.377-.034 2.377-.025 2.377-.029 2.377-.02 2.377-.014 2.377-.014 2.377-.014 2.377-.01\" style=\"stroke-width:1.2\"\u002F>\u003Cpath fill=\"none\" stroke=\"var(--tk-good)\" d=\"m-50 49.277 2.376-.021 2.377-.025 2.377-.03 2.377-.035 2.377-.04 2.377-.049 2.377-.056 2.377-.068 2.377-.08 2.377-.093 2.377-.11 2.377-.131 2.377-.154 2.377-.179 2.377-.215 2.377-.248 2.377-.293 2.377-.344 2.377-.405 2.377-.468 2.377-.552 2.377-.64 2.377-.744 2.377-.867 2.377-.99 2.377-1.148 2.377-1.306 2.377-1.494 2.377-1.69L21.31 34.9l2.377-2.121 2.377-2.348 2.377-2.577 2.377-2.805 2.377-3.008 2.377-3.197 2.377-3.347 2.377-3.465 2.377-3.542 2.377-3.542 2.377-3.543 2.377-3.465 2.377-3.35 2.377-3.197 2.377-3.01 2.377-2.799 2.377-2.577 2.377-2.352 2.377-2.121 2.377-1.904 2.377-1.69 2.377-1.493 2.377-1.31 2.377-1.145 2.377-.994 2.377-.862 2.377-.746 2.377-.642 2.377-.552 2.377-.47 2.377-.403 2.377-.343 2.377-.296 2.377-.247 2.377-.213 2.377-.182 2.377-.153 2.377-.13 2.377-.11 2.377-.095 2.377-.078 2.377-.067 2.377-.058 2.377-.048 2.377-.041 2.377-.035 2.377-.03 2.377-.025 2.377-.02\" style=\"stroke-width:1.2\"\u002F>\u003Cpath fill=\"none\" stroke=\"var(--tk-warn)\" d=\"M-50 6.714h93.893l36.135-54.06\" style=\"stroke-width:1.2\"\u002F>\u003Cg fill=\"var(--tk-accent)\" stroke=\"var(--tk-accent)\">\u003Cg transform=\"translate(-68.903 -26.453)\">\u003Cpath d=\"M44.174 6.706L44.174 5.484Q44.174 5.456 44.206 5.425Q44.237 5.394 44.260 5.394L44.366 5.394Q44.436 5.394 44.452 5.456Q44.514 5.777 44.653 6.017Q44.791 6.257 45.024 6.398Q45.256 6.538 45.565 6.538Q45.803 6.538 46.012 6.478Q46.221 6.417 46.358 6.269Q46.495 6.120 46.495 5.874Q46.495 5.620 46.284 5.454Q46.073 5.288 45.803 5.234L45.182 5.120Q44.776 5.042 44.475 4.786Q44.174 4.530 44.174 4.155Q44.174 3.788 44.375 3.566Q44.577 3.343 44.901 3.245Q45.225 3.148 45.565 3.148Q46.030 3.148 46.327 3.355L46.549 3.171Q46.573 3.148 46.604 3.148L46.655 3.148Q46.686 3.148 46.713 3.175Q46.741 3.202 46.741 3.234L46.741 4.218Q46.741 4.249 46.715 4.278Q46.690 4.308 46.655 4.308L46.549 4.308Q46.514 4.308 46.487 4.280Q46.459 4.253 46.459 4.218Q46.459 3.819 46.207 3.599Q45.956 3.378 45.557 3.378Q45.202 3.378 44.918 3.501Q44.635 3.624 44.635 3.929Q44.635 4.148 44.836 4.280Q45.038 4.413 45.284 4.456L45.909 4.569Q46.338 4.659 46.647 4.956Q46.956 5.253 46.956 5.667Q46.956 6.237 46.557 6.515Q46.159 6.792 45.565 6.792Q45.014 6.792 44.663 6.456L44.366 6.769Q44.342 6.792 44.307 6.792L44.260 6.792Q44.237 6.792 44.206 6.761Q44.174 6.730 44.174 6.706M49.342 6.714L47.565 6.714L47.565 6.417Q47.838 6.417 48.006 6.370Q48.174 6.323 48.174 6.155L48.174 4.019Q48.174 3.804 48.118 3.708Q48.061 3.612 47.948 3.591Q47.834 3.569 47.588 3.569L47.588 3.273L48.788 3.187L48.788 6.155Q48.788 6.323 48.934 6.370Q49.081 6.417 49.342 6.417L49.342 6.714M47.901 1.792Q47.901 1.601 48.036 1.470Q48.170 1.339 48.366 1.339Q48.487 1.339 48.590 1.402Q48.694 1.464 48.756 1.568Q48.819 1.671 48.819 1.792Q48.819 1.987 48.688 2.122Q48.557 2.257 48.366 2.257Q48.166 2.257 48.034 2.124Q47.901 1.991 47.901 1.792M49.842 7.323Q49.842 7.042 50.053 6.831Q50.264 6.620 50.549 6.530Q50.393 6.405 50.315 6.216Q50.237 6.027 50.237 5.827Q50.237 5.472 50.467 5.179Q50.100 4.839 50.100 4.370Q50.100 4.019 50.303 3.749Q50.506 3.480 50.827 3.333Q51.147 3.187 51.491 3.187Q52.010 3.187 52.381 3.468Q52.745 3.097 53.291 3.097Q53.471 3.097 53.598 3.224Q53.725 3.351 53.725 3.530Q53.725 3.636 53.647 3.714Q53.569 3.792 53.459 3.792Q53.350 3.792 53.274 3.716Q53.198 3.640 53.198 3.530Q53.198 3.429 53.237 3.378Q53.245 3.370 53.248 3.364Q53.252 3.359 53.252 3.355Q52.877 3.355 52.557 3.609Q52.877 3.948 52.877 4.370Q52.877 4.640 52.760 4.857Q52.643 5.073 52.438 5.232Q52.233 5.390 51.991 5.472Q51.748 5.554 51.491 5.554Q51.272 5.554 51.059 5.495Q50.846 5.437 50.651 5.316Q50.557 5.456 50.557 5.636Q50.557 5.843 50.694 5.995Q50.831 6.148 51.038 6.148L51.733 6.148Q52.221 6.148 52.633 6.232Q53.045 6.316 53.325 6.573Q53.604 6.831 53.604 7.323Q53.604 7.687 53.284 7.919Q52.963 8.152 52.522 8.253Q52.081 8.355 51.725 8.355Q51.370 8.355 50.926 8.253Q50.483 8.152 50.163 7.919Q49.842 7.687 49.842 7.323M50.346 7.323Q50.346 7.519 50.491 7.667Q50.635 7.816 50.848 7.905Q51.061 7.995 51.301 8.042Q51.541 8.089 51.725 8.089Q51.967 8.089 52.297 8.011Q52.627 7.933 52.864 7.759Q53.100 7.585 53.100 7.323Q53.100 6.917 52.690 6.808Q52.280 6.698 51.717 6.698L51.038 6.698Q50.768 6.698 50.557 6.876Q50.346 7.054 50.346 7.323M51.491 5.288Q52.213 5.288 52.213 4.370Q52.213 3.448 51.491 3.448Q50.764 3.448 50.764 4.370Q50.764 5.288 51.491 5.288M56.018 6.714L54.163 6.714L54.163 6.417Q54.436 6.417 54.604 6.370Q54.772 6.323 54.772 6.155L54.772 4.019Q54.772 3.804 54.709 3.708Q54.647 3.612 54.528 3.591Q54.409 3.569 54.163 3.569L54.163 3.273L55.354 3.187L55.354 3.921Q55.467 3.706 55.661 3.538Q55.854 3.370 56.092 3.278Q56.331 3.187 56.584 3.187Q57.545 3.187 57.721 3.898Q57.905 3.569 58.233 3.378Q58.561 3.187 58.940 3.187Q60.116 3.187 60.116 4.265L60.116 6.155Q60.116 6.323 60.284 6.370Q60.452 6.417 60.721 6.417L60.721 6.714L58.866 6.714L58.866 6.417Q59.139 6.417 59.307 6.372Q59.475 6.327 59.475 6.155L59.475 4.280Q59.475 3.894 59.350 3.667Q59.225 3.441 58.873 3.441Q58.569 3.441 58.313 3.603Q58.057 3.765 57.909 4.034Q57.760 4.304 57.760 4.601L57.760 6.155Q57.760 6.323 57.930 6.370Q58.100 6.417 58.370 6.417L58.370 6.714L56.514 6.714L56.514 6.417Q56.788 6.417 56.956 6.370Q57.123 6.323 57.123 6.155L57.123 4.280Q57.123 3.894 56.998 3.667Q56.873 3.441 56.522 3.441Q56.217 3.441 55.961 3.603Q55.706 3.765 55.557 4.034Q55.409 4.304 55.409 4.601L55.409 6.155Q55.409 6.323 55.579 6.370Q55.748 6.417 56.018 6.417L56.018 6.714M61.166 5.019Q61.166 4.515 61.422 4.083Q61.678 3.652 62.114 3.400Q62.549 3.148 63.049 3.148Q63.436 3.148 63.778 3.292Q64.120 3.437 64.381 3.698Q64.643 3.960 64.786 4.296Q64.928 4.632 64.928 5.019Q64.928 5.511 64.664 5.921Q64.401 6.331 63.971 6.562Q63.541 6.792 63.049 6.792Q62.557 6.792 62.123 6.560Q61.690 6.327 61.428 5.919Q61.166 5.511 61.166 5.019M63.049 6.515Q63.506 6.515 63.758 6.292Q64.010 6.069 64.098 5.718Q64.186 5.366 64.186 4.921Q64.186 4.491 64.092 4.153Q63.998 3.816 63.745 3.609Q63.491 3.402 63.049 3.402Q62.401 3.402 62.157 3.818Q61.913 4.234 61.913 4.921Q61.913 5.366 62 5.718Q62.088 6.069 62.340 6.292Q62.592 6.515 63.049 6.515M67.272 6.714L65.495 6.714L65.495 6.417Q65.768 6.417 65.936 6.370Q66.104 6.323 66.104 6.155L66.104 4.019Q66.104 3.804 66.047 3.708Q65.991 3.612 65.877 3.591Q65.764 3.569 65.518 3.569L65.518 3.273L66.717 3.187L66.717 6.155Q66.717 6.323 66.864 6.370Q67.010 6.417 67.272 6.417L67.272 6.714M65.831 1.792Q65.831 1.601 65.965 1.470Q66.100 1.339 66.295 1.339Q66.416 1.339 66.520 1.402Q66.623 1.464 66.686 1.568Q66.748 1.671 66.748 1.792Q66.748 1.987 66.618 2.122Q66.487 2.257 66.295 2.257Q66.096 2.257 65.963 2.124Q65.831 1.991 65.831 1.792M69.588 6.792Q69.108 6.792 68.700 6.548Q68.291 6.304 68.053 5.890Q67.815 5.476 67.815 4.987Q67.815 4.495 68.073 4.079Q68.331 3.663 68.762 3.425Q69.194 3.187 69.686 3.187Q70.307 3.187 70.756 3.624L70.756 1.995Q70.756 1.780 70.694 1.685Q70.631 1.589 70.514 1.568Q70.397 1.546 70.151 1.546L70.151 1.249L71.373 1.163L71.373 5.972Q71.373 6.183 71.436 6.278Q71.498 6.374 71.616 6.396Q71.733 6.417 71.983 6.417L71.983 6.714L70.733 6.792L70.733 6.308Q70.268 6.792 69.588 6.792M69.655 6.538Q69.995 6.538 70.288 6.347Q70.581 6.155 70.733 5.859L70.733 4.027Q70.584 3.753 70.323 3.597Q70.061 3.441 69.748 3.441Q69.123 3.441 68.840 3.888Q68.557 4.335 68.557 4.995Q68.557 5.640 68.809 6.089Q69.061 6.538 69.655 6.538\" fill=\"var(--tk-accent)\" stroke=\"var(--tk-accent)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg fill=\"var(--tk-good)\" stroke=\"var(--tk-good)\">\u003Cg transform=\"translate(56.17 -31.365)\">\u003Cpath d=\"M44.756 5.753L44.756 3.562L44.053 3.562L44.053 3.308Q44.409 3.308 44.651 3.075Q44.893 2.843 45.004 2.495Q45.116 2.148 45.116 1.792L45.397 1.792L45.397 3.265L46.573 3.265L46.573 3.562L45.397 3.562L45.397 5.737Q45.397 6.058 45.516 6.286Q45.635 6.515 45.916 6.515Q46.096 6.515 46.213 6.392Q46.331 6.269 46.383 6.089Q46.436 5.909 46.436 5.737L46.436 5.265L46.717 5.265L46.717 5.753Q46.717 6.007 46.612 6.247Q46.506 6.487 46.309 6.640Q46.112 6.792 45.854 6.792Q45.538 6.792 45.286 6.669Q45.034 6.546 44.895 6.312Q44.756 6.077 44.756 5.753M47.534 5.882Q47.534 5.398 47.936 5.103Q48.338 4.808 48.889 4.689Q49.440 4.569 49.932 4.569L49.932 4.280Q49.932 4.054 49.817 3.847Q49.702 3.640 49.504 3.521Q49.307 3.402 49.077 3.402Q48.651 3.402 48.366 3.507Q48.436 3.534 48.483 3.589Q48.530 3.644 48.555 3.714Q48.581 3.784 48.581 3.859Q48.581 3.964 48.530 4.056Q48.479 4.148 48.387 4.198Q48.295 4.249 48.190 4.249Q48.084 4.249 47.993 4.198Q47.901 4.148 47.850 4.056Q47.799 3.964 47.799 3.859Q47.799 3.441 48.188 3.294Q48.577 3.148 49.077 3.148Q49.409 3.148 49.762 3.278Q50.116 3.409 50.344 3.663Q50.573 3.917 50.573 4.265L50.573 6.066Q50.573 6.198 50.645 6.308Q50.717 6.417 50.846 6.417Q50.971 6.417 51.039 6.312Q51.108 6.206 51.108 6.066L51.108 5.554L51.389 5.554L51.389 6.066Q51.389 6.269 51.272 6.427Q51.155 6.585 50.973 6.669Q50.791 6.753 50.588 6.753Q50.358 6.753 50.206 6.581Q50.053 6.409 50.022 6.179Q49.862 6.460 49.553 6.626Q49.245 6.792 48.893 6.792Q48.381 6.792 47.957 6.569Q47.534 6.347 47.534 5.882M48.221 5.882Q48.221 6.167 48.448 6.353Q48.674 6.538 48.967 6.538Q49.213 6.538 49.438 6.421Q49.663 6.304 49.797 6.101Q49.932 5.898 49.932 5.644L49.932 4.812Q49.666 4.812 49.381 4.866Q49.096 4.921 48.825 5.050Q48.553 5.179 48.387 5.386Q48.221 5.593 48.221 5.882M53.612 6.714L51.756 6.714L51.756 6.417Q52.030 6.417 52.198 6.370Q52.366 6.323 52.366 6.155L52.366 4.019Q52.366 3.804 52.303 3.708Q52.241 3.612 52.122 3.591Q52.002 3.569 51.756 3.569L51.756 3.273L52.948 3.187L52.948 3.921Q53.061 3.706 53.254 3.538Q53.448 3.370 53.686 3.278Q53.924 3.187 54.178 3.187Q55.346 3.187 55.346 4.265L55.346 6.155Q55.346 6.323 55.516 6.370Q55.686 6.417 55.956 6.417L55.956 6.714L54.100 6.714L54.100 6.417Q54.373 6.417 54.541 6.370Q54.709 6.323 54.709 6.155L54.709 4.280Q54.709 3.898 54.588 3.669Q54.467 3.441 54.116 3.441Q53.803 3.441 53.549 3.603Q53.295 3.765 53.149 4.034Q53.002 4.304 53.002 4.601L53.002 6.155Q53.002 6.323 53.172 6.370Q53.342 6.417 53.612 6.417L53.612 6.714M58.331 6.714L56.475 6.714L56.475 6.417Q56.748 6.417 56.916 6.370Q57.084 6.323 57.084 6.155L57.084 1.995Q57.084 1.780 57.022 1.685Q56.959 1.589 56.840 1.568Q56.721 1.546 56.475 1.546L56.475 1.249L57.698 1.163L57.698 3.866Q57.823 3.655 58.010 3.505Q58.198 3.355 58.424 3.271Q58.651 3.187 58.897 3.187Q60.065 3.187 60.065 4.265L60.065 6.155Q60.065 6.323 60.235 6.370Q60.405 6.417 60.674 6.417L60.674 6.714L58.819 6.714L58.819 6.417Q59.092 6.417 59.260 6.370Q59.428 6.323 59.428 6.155L59.428 4.280Q59.428 3.898 59.307 3.669Q59.186 3.441 58.834 3.441Q58.522 3.441 58.268 3.603Q58.014 3.765 57.868 4.034Q57.721 4.304 57.721 4.601L57.721 6.155Q57.721 6.323 57.891 6.370Q58.061 6.417 58.331 6.417\" fill=\"var(--tk-good)\" stroke=\"var(--tk-good)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\">\u003Cg transform=\"translate(10.896 -59.016)\">\u003Cpath d=\"M46.604 6.714L44.245 6.714L44.245 6.417Q44.569 6.417 44.811 6.370Q45.053 6.323 45.053 6.155L45.053 1.812Q45.053 1.640 44.811 1.593Q44.569 1.546 44.245 1.546L44.245 1.249L46.838 1.249Q47.170 1.249 47.557 1.335Q47.944 1.421 48.291 1.595Q48.639 1.769 48.858 2.050Q49.077 2.331 49.077 2.698Q49.077 3.023 48.875 3.288Q48.674 3.554 48.368 3.730Q48.061 3.905 47.733 3.995Q48.100 4.116 48.360 4.386Q48.620 4.655 48.670 5.019L48.764 5.714Q48.834 6.163 48.932 6.394Q49.030 6.624 49.327 6.624Q49.573 6.624 49.706 6.407Q49.838 6.191 49.838 5.929Q49.858 5.855 49.940 5.835L50.022 5.835Q50.116 5.859 50.116 5.952Q50.116 6.183 50.018 6.398Q49.920 6.612 49.743 6.747Q49.565 6.882 49.334 6.882Q48.737 6.882 48.319 6.595Q47.901 6.308 47.901 5.737L47.901 5.042Q47.901 4.761 47.748 4.546Q47.596 4.331 47.346 4.218Q47.096 4.105 46.823 4.105L45.795 4.105L45.795 6.155Q45.795 6.319 46.039 6.368Q46.284 6.417 46.604 6.417L46.604 6.714M45.795 1.812L45.795 3.851L46.725 3.851Q47.045 3.851 47.313 3.798Q47.581 3.745 47.780 3.618Q47.979 3.491 48.096 3.259Q48.213 3.027 48.213 2.698Q48.213 2.046 47.817 1.796Q47.420 1.546 46.725 1.546L46.198 1.546Q45.979 1.546 45.887 1.589Q45.795 1.632 45.795 1.812M50.377 4.960Q50.377 4.480 50.610 4.064Q50.842 3.648 51.252 3.398Q51.663 3.148 52.139 3.148Q52.870 3.148 53.268 3.589Q53.666 4.030 53.666 4.761Q53.666 4.866 53.573 4.890L51.123 4.890L51.123 4.960Q51.123 5.370 51.245 5.726Q51.366 6.081 51.637 6.298Q51.909 6.515 52.338 6.515Q52.702 6.515 52.998 6.286Q53.295 6.058 53.397 5.706Q53.405 5.659 53.491 5.644L53.573 5.644Q53.666 5.671 53.666 5.753Q53.666 5.761 53.659 5.792Q53.596 6.019 53.457 6.202Q53.319 6.386 53.127 6.519Q52.936 6.652 52.717 6.722Q52.498 6.792 52.260 6.792Q51.889 6.792 51.551 6.655Q51.213 6.519 50.946 6.267Q50.678 6.015 50.528 5.675Q50.377 5.335 50.377 4.960M51.131 4.652L53.092 4.652Q53.092 4.347 52.991 4.056Q52.889 3.765 52.672 3.583Q52.456 3.402 52.139 3.402Q51.838 3.402 51.608 3.589Q51.377 3.777 51.254 4.068Q51.131 4.359 51.131 4.652M58.635 6.714L54.252 6.714L54.252 6.417Q54.573 6.417 54.817 6.370Q55.061 6.323 55.061 6.155L55.061 1.812Q55.061 1.640 54.817 1.593Q54.573 1.546 54.252 1.546L54.252 1.249L56.838 1.249L56.838 1.546Q55.827 1.546 55.827 1.812L55.827 6.155Q55.827 6.327 55.918 6.372Q56.010 6.417 56.229 6.417L56.916 6.417Q57.389 6.417 57.698 6.292Q58.006 6.167 58.184 5.937Q58.362 5.706 58.454 5.380Q58.545 5.054 58.588 4.601L58.870 4.601L58.635 6.714M60.366 4.905L60.366 1.812Q60.366 1.640 60.122 1.593Q59.877 1.546 59.557 1.546L59.557 1.249L61.940 1.249L61.940 1.546Q61.620 1.546 61.375 1.595Q61.131 1.644 61.131 1.812L61.131 4.882Q61.131 5.605 61.475 6.095Q61.819 6.585 62.518 6.585Q62.983 6.585 63.354 6.355Q63.725 6.124 63.934 5.732Q64.143 5.339 64.143 4.882L64.143 2.027Q64.143 1.546 63.334 1.546L63.334 1.249L65.245 1.249L65.245 1.546Q64.436 1.546 64.436 2.027L64.436 4.905Q64.436 5.425 64.182 5.882Q63.928 6.339 63.487 6.610Q63.045 6.882 62.518 6.882Q62.108 6.882 61.727 6.739Q61.346 6.597 61.034 6.327Q60.721 6.058 60.543 5.691Q60.366 5.323 60.366 4.905\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fsvg>\u003Cfigcaption class=\"tikz-cap\">Three activations on shared axes. Sigmoid and tanh flatten in both tails; ReLU is the identity for \u003Cspan class=\"katex\">\u003Cspan class=\"katex-html\" aria-hidden=\"true\">\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.5782em;vertical-align:-0.0391em;\">\u003C\u002Fspan>\u003Cspan class=\"mord mathnormal\" style=\"margin-right:0.044em;\">z\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003Cspan class=\"mrel\">>\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003C\u002Fspan>\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.6444em;\">\u003C\u002Fspan>\u003Cspan class=\"mord\">0\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan> and exactly \u003Cspan class=\"katex\">\u003Cspan class=\"katex-html\" aria-hidden=\"true\">\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.6444em;\">\u003C\u002Fspan>\u003Cspan class=\"mord\">0\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan> for \u003Cspan class=\"katex\">\u003Cspan class=\"katex-html\" aria-hidden=\"true\">\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.5782em;vertical-align:-0.0391em;\">\u003C\u002Fspan>\u003Cspan class=\"mord mathnormal\" style=\"margin-right:0.044em;\">z\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003Cspan class=\"mrel\"><\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003C\u002Fspan>\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.6444em;\">\u003C\u002Fspan>\u003Cspan class=\"mord\">0\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>.\u003C\u002Ffigcaption>","\u003Csvg style=\"width:100%;max-width:343.991px;height:auto\" xmlns=\"http:\u002F\u002Fwww.w3.org\u002F2000\u002Fsvg\" viewBox=\"-75 -75 257.993 125.768\">\u003Cg stroke=\"currentColor\" style=\"stroke-miterlimit:10;stroke-width:.4\">\u003Cg style=\"stroke-width:.8\">\u003Cpath fill=\"none\" d=\"M-32.12 9.559h182.344\"\u002F>\u003Cpath stroke=\"none\" d=\"m152.824 9.559-4.16-2.08 1.56 2.08-1.56 2.08\"\u002F>\u003Cg transform=\"translate(109.008 1.905)\">\u003Cpath d=\"M49.647 9.559L47.869 9.559L47.869 9.262Q48.143 9.262 48.311 9.215Q48.479 9.168 48.479 9L48.479 6.864Q48.479 6.649 48.422 6.553Q48.365 6.457 48.252 6.436Q48.139 6.414 47.893 6.414L47.893 6.118L49.092 6.032L49.092 9Q49.092 9.168 49.238 9.215Q49.385 9.262 49.647 9.262L49.647 9.559M48.205 4.637Q48.205 4.446 48.340 4.315Q48.475 4.184 48.670 4.184Q48.791 4.184 48.895 4.246Q48.998 4.309 49.061 4.413Q49.123 4.516 49.123 4.637Q49.123 4.832 48.992 4.967Q48.861 5.102 48.670 5.102Q48.471 5.102 48.338 4.969Q48.205 4.836 48.205 4.637M52.076 9.559L50.221 9.559L50.221 9.262Q50.494 9.262 50.662 9.215Q50.830 9.168 50.830 9L50.830 6.864Q50.830 6.649 50.768 6.553Q50.705 6.457 50.586 6.436Q50.467 6.414 50.221 6.414L50.221 6.118L51.412 6.032L51.412 6.766Q51.526 6.551 51.719 6.383Q51.912 6.215 52.151 6.123Q52.389 6.032 52.643 6.032Q53.811 6.032 53.811 7.110L53.811 9Q53.811 9.168 53.981 9.215Q54.151 9.262 54.420 9.262L54.420 9.559L52.565 9.559L52.565 9.262Q52.838 9.262 53.006 9.215Q53.174 9.168 53.174 9L53.174 7.125Q53.174 6.743 53.053 6.514Q52.932 6.286 52.580 6.286Q52.268 6.286 52.014 6.448Q51.760 6.610 51.613 6.879Q51.467 7.149 51.467 7.446L51.467 9Q51.467 9.168 51.637 9.215Q51.807 9.262 52.076 9.262L52.076 9.559M56.748 11.110L54.893 11.110L54.893 10.817Q55.162 10.817 55.330 10.772Q55.498 10.727 55.498 10.551L55.498 6.727Q55.498 6.520 55.342 6.467Q55.186 6.414 54.893 6.414L54.893 6.118L56.115 6.032L56.115 6.496Q56.346 6.274 56.660 6.153Q56.975 6.032 57.315 6.032Q57.787 6.032 58.192 6.278Q58.596 6.524 58.828 6.940Q59.061 7.356 59.061 7.832Q59.061 8.207 58.912 8.536Q58.764 8.864 58.494 9.116Q58.225 9.368 57.881 9.502Q57.537 9.637 57.178 9.637Q56.889 9.637 56.617 9.516Q56.346 9.395 56.139 9.184L56.139 10.551Q56.139 10.727 56.307 10.772Q56.475 10.817 56.748 10.817L56.748 11.110M56.139 6.895L56.139 8.735Q56.291 9.024 56.553 9.204Q56.815 9.383 57.123 9.383Q57.408 9.383 57.631 9.245Q57.854 9.106 58.006 8.875Q58.158 8.645 58.236 8.373Q58.315 8.102 58.315 7.832Q58.315 7.500 58.190 7.143Q58.065 6.786 57.817 6.549Q57.569 6.313 57.221 6.313Q56.897 6.313 56.602 6.469Q56.307 6.625 56.139 6.895M60.268 8.606L60.268 6.864Q60.268 6.649 60.205 6.553Q60.143 6.457 60.024 6.436Q59.904 6.414 59.658 6.414L59.658 6.118L60.904 6.032L60.904 8.582L60.904 8.606Q60.904 8.918 60.959 9.080Q61.014 9.243 61.164 9.313Q61.315 9.383 61.635 9.383Q62.065 9.383 62.338 9.045Q62.611 8.707 62.611 8.262L62.611 6.864Q62.611 6.649 62.549 6.553Q62.486 6.457 62.367 6.436Q62.248 6.414 62.002 6.414L62.002 6.118L63.248 6.032L63.248 8.817Q63.248 9.028 63.311 9.123Q63.373 9.219 63.492 9.241Q63.611 9.262 63.858 9.262L63.858 9.559L62.635 9.637L62.635 9.016Q62.467 9.305 62.186 9.471Q61.904 9.637 61.584 9.637Q60.268 9.637 60.268 8.606M64.928 8.598L64.928 6.407L64.225 6.407L64.225 6.153Q64.580 6.153 64.822 5.920Q65.065 5.688 65.176 5.340Q65.287 4.993 65.287 4.637L65.569 4.637L65.569 6.110L66.744 6.110L66.744 6.407L65.569 6.407L65.569 8.582Q65.569 8.903 65.688 9.131Q65.807 9.360 66.088 9.360Q66.268 9.360 66.385 9.237Q66.502 9.114 66.555 8.934Q66.608 8.754 66.608 8.582L66.608 8.110L66.889 8.110L66.889 8.598Q66.889 8.852 66.783 9.092Q66.678 9.332 66.481 9.485Q66.283 9.637 66.026 9.637Q65.709 9.637 65.457 9.514Q65.205 9.391 65.067 9.157Q64.928 8.922 64.928 8.598\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg style=\"stroke-width:.8\">\u003Cpath fill=\"none\" d=\"M47.549 42.279v-98.407\"\u002F>\u003Cpath stroke=\"none\" d=\"m47.549-58.728-2.08 4.16 2.08-1.56 2.08 1.56\"\u002F>\u003Cg transform=\"translate(-12.514 -73.575)\">\u003Cpath d=\"M47.787 7.864Q47.787 7.360 48.043 6.928Q48.299 6.496 48.735 6.245Q49.170 5.993 49.670 5.993Q50.057 5.993 50.399 6.137Q50.740 6.282 51.002 6.543Q51.264 6.805 51.406 7.141Q51.549 7.477 51.549 7.864Q51.549 8.356 51.285 8.766Q51.022 9.176 50.592 9.407Q50.162 9.637 49.670 9.637Q49.178 9.637 48.744 9.405Q48.311 9.172 48.049 8.764Q47.787 8.356 47.787 7.864M49.670 9.360Q50.127 9.360 50.379 9.137Q50.631 8.914 50.719 8.563Q50.807 8.211 50.807 7.766Q50.807 7.336 50.713 6.998Q50.619 6.661 50.365 6.454Q50.111 6.246 49.670 6.246Q49.022 6.246 48.778 6.663Q48.533 7.079 48.533 7.766Q48.533 8.211 48.621 8.563Q48.709 8.914 48.961 9.137Q49.213 9.360 49.670 9.360M52.717 8.606L52.717 6.864Q52.717 6.649 52.654 6.553Q52.592 6.457 52.473 6.436Q52.354 6.414 52.108 6.414L52.108 6.118L53.354 6.032L53.354 8.582L53.354 8.606Q53.354 8.918 53.408 9.080Q53.463 9.243 53.613 9.313Q53.764 9.383 54.084 9.383Q54.514 9.383 54.787 9.045Q55.061 8.707 55.061 8.262L55.061 6.864Q55.061 6.649 54.998 6.553Q54.936 6.457 54.817 6.436Q54.697 6.414 54.451 6.414L54.451 6.118L55.697 6.032L55.697 8.817Q55.697 9.028 55.760 9.123Q55.822 9.219 55.942 9.241Q56.061 9.262 56.307 9.262L56.307 9.559L55.084 9.637L55.084 9.016Q54.916 9.305 54.635 9.471Q54.354 9.637 54.033 9.637Q52.717 9.637 52.717 8.606M57.377 8.598L57.377 6.407L56.674 6.407L56.674 6.153Q57.029 6.153 57.272 5.920Q57.514 5.688 57.625 5.340Q57.736 4.993 57.736 4.637L58.018 4.637L58.018 6.110L59.194 6.110L59.194 6.407L58.018 6.407L58.018 8.582Q58.018 8.903 58.137 9.131Q58.256 9.360 58.537 9.360Q58.717 9.360 58.834 9.237Q58.951 9.114 59.004 8.934Q59.057 8.754 59.057 8.582L59.057 8.110L59.338 8.110L59.338 8.598Q59.338 8.852 59.233 9.092Q59.127 9.332 58.930 9.485Q58.733 9.637 58.475 9.637Q58.158 9.637 57.906 9.514Q57.654 9.391 57.516 9.157Q57.377 8.922 57.377 8.598M61.940 11.110L60.084 11.110L60.084 10.817Q60.354 10.817 60.522 10.772Q60.690 10.727 60.690 10.551L60.690 6.727Q60.690 6.520 60.533 6.467Q60.377 6.414 60.084 6.414L60.084 6.118L61.307 6.032L61.307 6.496Q61.537 6.274 61.852 6.153Q62.166 6.032 62.506 6.032Q62.979 6.032 63.383 6.278Q63.787 6.524 64.020 6.940Q64.252 7.356 64.252 7.832Q64.252 8.207 64.104 8.536Q63.955 8.864 63.686 9.116Q63.416 9.368 63.072 9.502Q62.729 9.637 62.369 9.637Q62.080 9.637 61.809 9.516Q61.537 9.395 61.330 9.184L61.330 10.551Q61.330 10.727 61.498 10.772Q61.666 10.817 61.940 10.817L61.940 11.110M61.330 6.895L61.330 8.735Q61.483 9.024 61.744 9.204Q62.006 9.383 62.315 9.383Q62.600 9.383 62.822 9.245Q63.045 9.106 63.197 8.875Q63.350 8.645 63.428 8.373Q63.506 8.102 63.506 7.832Q63.506 7.500 63.381 7.143Q63.256 6.786 63.008 6.549Q62.760 6.313 62.412 6.313Q62.088 6.313 61.793 6.469Q61.498 6.625 61.330 6.895M65.459 8.606L65.459 6.864Q65.459 6.649 65.397 6.553Q65.334 6.457 65.215 6.436Q65.096 6.414 64.850 6.414L64.850 6.118L66.096 6.032L66.096 8.582L66.096 8.606Q66.096 8.918 66.151 9.080Q66.205 9.243 66.356 9.313Q66.506 9.383 66.826 9.383Q67.256 9.383 67.529 9.045Q67.803 8.707 67.803 8.262L67.803 6.864Q67.803 6.649 67.740 6.553Q67.678 6.457 67.559 6.436Q67.440 6.414 67.194 6.414L67.194 6.118L68.440 6.032L68.440 8.817Q68.440 9.028 68.502 9.123Q68.565 9.219 68.684 9.241Q68.803 9.262 69.049 9.262L69.049 9.559L67.826 9.637L67.826 9.016Q67.658 9.305 67.377 9.471Q67.096 9.637 66.776 9.637Q65.459 9.637 65.459 8.606M70.119 8.598L70.119 6.407L69.416 6.407L69.416 6.153Q69.772 6.153 70.014 5.920Q70.256 5.688 70.367 5.340Q70.479 4.993 70.479 4.637L70.760 4.637L70.760 6.110L71.936 6.110L71.936 6.407L70.760 6.407L70.760 8.582Q70.760 8.903 70.879 9.131Q70.998 9.360 71.279 9.360Q71.459 9.360 71.576 9.237Q71.694 9.114 71.746 8.934Q71.799 8.754 71.799 8.582L71.799 8.110L72.080 8.110L72.080 8.598Q72.080 8.852 71.975 9.092Q71.869 9.332 71.672 9.485Q71.475 9.637 71.217 9.637Q70.901 9.637 70.649 9.514Q70.397 9.391 70.258 9.157Q70.119 8.922 70.119 8.598\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cpath fill=\"none\" stroke=\"var(--tk-accent)\" d=\"M-26.429 9.559H47.55l59.75-59.751\" style=\"stroke-width:1.2\"\u002F>\u003Cpath fill=\"none\" stroke=\"var(--tk-warn)\" d=\"M-26.429 24.354 47.55 9.56\" style=\"stroke-width:1.2\"\u002F>\u003Cpath fill=\"none\" stroke=\"var(--tk-good)\" d=\"m-26.429 35.9 1.254-.097 1.254-.1 1.254-.102 1.254-.108 1.253-.114 1.254-.12 1.254-.125 1.254-.128 1.254-.136 1.254-.143 1.253-.148 1.254-.153 1.254-.163 1.254-.167 1.254-.177 1.254-.185 1.253-.194 1.254-.199 1.254-.21 1.254-.222 1.254-.228 1.254-.242 1.254-.25 1.253-.262 1.254-.276 1.254-.285 1.254-.302 1.254-.313 1.254-.324 1.253-.344 1.254-.356 1.254-.373 1.254-.39 1.254-.41 1.254-.424 1.253-.446 1.254-.464 1.254-.487 1.254-.51 1.254-.529 1.254-.555 1.253-.58 1.254-.607 1.254-.634 1.254-.66 1.254-.695 1.254-.723 1.253-.754 1.254-.791 1.254-.823 1.254-.865 1.254-.899 1.254-.942 1.253-.985 1.254-1.027 1.254-1.076 1.254-1.124 1.254-1.173 1.254-1.226\" style=\"stroke-width:1.2\"\u002F>\u003Cpath fill=\"none\" stroke=\"var(--tk-hi)\" d=\"m-26.429 14.673 1.254.124 1.254.124 1.254.125 1.254.125 1.253.125 1.254.125 1.254.125 1.254.124 1.254.123 1.254.122 1.253.12 1.254.12 1.254.116 1.254.115 1.254.111 1.254.108 1.253.104 1.254.099 1.254.096 1.254.09 1.254.084 1.254.078 1.254.07 1.253.065 1.254.055 1.254.048 1.254.038 1.254.028 1.254.016 1.253.005 1.254-.006 1.254-.019 1.254-.032 1.254-.046 1.254-.062 1.253-.078 1.254-.093 1.254-.112 1.254-.13 1.254-.147 1.254-.167 1.253-.188 1.254-.208 1.254-.23 1.254-.252 1.254-.275 1.254-.298 1.253-.322 1.254-.348 1.254-.371 1.254-.398 1.254-.424 1.254-.45 1.253-.475 1.254-.504 1.254-.532 1.254-.558 1.254-.586 1.254-.611\" style=\"stroke-width:1.2\"\u002F>\u003Cg fill=\"var(--tk-accent)\" stroke=\"var(--tk-accent)\">\u003Cg transform=\"translate(33.09 -60.438)\">\u003Cpath d=\"M50.260 9.559L47.901 9.559L47.901 9.262Q48.225 9.262 48.467 9.215Q48.709 9.168 48.709 9L48.709 4.657Q48.709 4.485 48.467 4.438Q48.225 4.391 47.901 4.391L47.901 4.094L50.494 4.094Q50.826 4.094 51.213 4.180Q51.600 4.266 51.947 4.440Q52.295 4.614 52.514 4.895Q52.733 5.176 52.733 5.543Q52.733 5.868 52.531 6.133Q52.330 6.399 52.024 6.575Q51.717 6.750 51.389 6.840Q51.756 6.961 52.016 7.231Q52.276 7.500 52.326 7.864L52.420 8.559Q52.490 9.008 52.588 9.239Q52.686 9.469 52.983 9.469Q53.229 9.469 53.361 9.252Q53.494 9.036 53.494 8.774Q53.514 8.700 53.596 8.680L53.678 8.680Q53.772 8.704 53.772 8.797Q53.772 9.028 53.674 9.243Q53.576 9.457 53.399 9.592Q53.221 9.727 52.990 9.727Q52.393 9.727 51.975 9.440Q51.557 9.153 51.557 8.582L51.557 7.887Q51.557 7.606 51.404 7.391Q51.252 7.176 51.002 7.063Q50.752 6.950 50.479 6.950L49.451 6.950L49.451 9Q49.451 9.164 49.695 9.213Q49.940 9.262 50.260 9.262L50.260 9.559M49.451 4.657L49.451 6.696L50.381 6.696Q50.701 6.696 50.969 6.643Q51.236 6.590 51.436 6.463Q51.635 6.336 51.752 6.104Q51.869 5.871 51.869 5.543Q51.869 4.891 51.473 4.641Q51.076 4.391 50.381 4.391L49.854 4.391Q49.635 4.391 49.543 4.434Q49.451 4.477 49.451 4.657M54.033 7.805Q54.033 7.325 54.266 6.909Q54.498 6.493 54.908 6.243Q55.319 5.993 55.795 5.993Q56.526 5.993 56.924 6.434Q57.322 6.875 57.322 7.606Q57.322 7.711 57.229 7.735L54.779 7.735L54.779 7.805Q54.779 8.215 54.901 8.571Q55.022 8.926 55.293 9.143Q55.565 9.360 55.994 9.360Q56.358 9.360 56.654 9.131Q56.951 8.903 57.053 8.551Q57.061 8.504 57.147 8.489L57.229 8.489Q57.322 8.516 57.322 8.598Q57.322 8.606 57.315 8.637Q57.252 8.864 57.113 9.047Q56.975 9.231 56.783 9.364Q56.592 9.496 56.373 9.567Q56.154 9.637 55.916 9.637Q55.545 9.637 55.207 9.500Q54.869 9.364 54.602 9.112Q54.334 8.860 54.184 8.520Q54.033 8.180 54.033 7.805M54.787 7.496L56.748 7.496Q56.748 7.192 56.647 6.901Q56.545 6.610 56.328 6.428Q56.111 6.246 55.795 6.246Q55.494 6.246 55.264 6.434Q55.033 6.621 54.910 6.913Q54.787 7.204 54.787 7.496M62.291 9.559L57.908 9.559L57.908 9.262Q58.229 9.262 58.473 9.215Q58.717 9.168 58.717 9L58.717 4.657Q58.717 4.485 58.473 4.438Q58.229 4.391 57.908 4.391L57.908 4.094L60.494 4.094L60.494 4.391Q59.483 4.391 59.483 4.657L59.483 9Q59.483 9.172 59.574 9.217Q59.666 9.262 59.885 9.262L60.572 9.262Q61.045 9.262 61.354 9.137Q61.662 9.012 61.840 8.782Q62.018 8.551 62.110 8.225Q62.201 7.899 62.244 7.446L62.526 7.446L62.291 9.559M64.022 7.750L64.022 4.657Q64.022 4.485 63.778 4.438Q63.533 4.391 63.213 4.391L63.213 4.094L65.596 4.094L65.596 4.391Q65.276 4.391 65.031 4.440Q64.787 4.489 64.787 4.657L64.787 7.727Q64.787 8.450 65.131 8.940Q65.475 9.430 66.174 9.430Q66.639 9.430 67.010 9.200Q67.381 8.969 67.590 8.577Q67.799 8.184 67.799 7.727L67.799 4.871Q67.799 4.391 66.990 4.391L66.990 4.094L68.901 4.094L68.901 4.391Q68.092 4.391 68.092 4.871L68.092 7.750Q68.092 8.270 67.838 8.727Q67.584 9.184 67.143 9.455Q66.701 9.727 66.174 9.727Q65.764 9.727 65.383 9.584Q65.002 9.442 64.690 9.172Q64.377 8.903 64.199 8.536Q64.022 8.168 64.022 7.750\" fill=\"var(--tk-accent)\" stroke=\"var(--tk-accent)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\">\u003Cg transform=\"translate(-99.819 9.138)\">\u003Cpath d=\"M52.268 9.559L47.885 9.559L47.885 9.262Q48.205 9.262 48.449 9.215Q48.694 9.168 48.694 9L48.694 4.657Q48.694 4.485 48.449 4.438Q48.205 4.391 47.885 4.391L47.885 4.094L50.471 4.094L50.471 4.391Q49.459 4.391 49.459 4.657L49.459 9Q49.459 9.172 49.551 9.217Q49.643 9.262 49.861 9.262L50.549 9.262Q51.022 9.262 51.330 9.137Q51.639 9.012 51.817 8.782Q51.994 8.551 52.086 8.225Q52.178 7.899 52.221 7.446L52.502 7.446L52.268 9.559M53.092 7.805Q53.092 7.325 53.324 6.909Q53.557 6.493 53.967 6.243Q54.377 5.993 54.854 5.993Q55.584 5.993 55.983 6.434Q56.381 6.875 56.381 7.606Q56.381 7.711 56.287 7.735L53.838 7.735L53.838 7.805Q53.838 8.215 53.959 8.571Q54.080 8.926 54.352 9.143Q54.623 9.360 55.053 9.360Q55.416 9.360 55.713 9.131Q56.010 8.903 56.111 8.551Q56.119 8.504 56.205 8.489L56.287 8.489Q56.381 8.516 56.381 8.598Q56.381 8.606 56.373 8.637Q56.311 8.864 56.172 9.047Q56.033 9.231 55.842 9.364Q55.651 9.496 55.432 9.567Q55.213 9.637 54.975 9.637Q54.604 9.637 54.266 9.500Q53.928 9.364 53.660 9.112Q53.393 8.860 53.242 8.520Q53.092 8.180 53.092 7.805M53.846 7.496L55.807 7.496Q55.807 7.192 55.705 6.901Q55.604 6.610 55.387 6.428Q55.170 6.246 54.854 6.246Q54.553 6.246 54.322 6.434Q54.092 6.621 53.969 6.913Q53.846 7.204 53.846 7.496M56.967 8.727Q56.967 8.243 57.369 7.948Q57.772 7.653 58.322 7.534Q58.873 7.414 59.365 7.414L59.365 7.125Q59.365 6.899 59.250 6.692Q59.135 6.485 58.938 6.366Q58.740 6.246 58.510 6.246Q58.084 6.246 57.799 6.352Q57.869 6.379 57.916 6.434Q57.963 6.489 57.988 6.559Q58.014 6.629 58.014 6.704Q58.014 6.809 57.963 6.901Q57.912 6.993 57.820 7.043Q57.729 7.094 57.623 7.094Q57.518 7.094 57.426 7.043Q57.334 6.993 57.283 6.901Q57.233 6.809 57.233 6.704Q57.233 6.286 57.621 6.139Q58.010 5.993 58.510 5.993Q58.842 5.993 59.195 6.123Q59.549 6.254 59.778 6.508Q60.006 6.762 60.006 7.110L60.006 8.911Q60.006 9.043 60.078 9.153Q60.151 9.262 60.279 9.262Q60.404 9.262 60.473 9.157Q60.541 9.051 60.541 8.911L60.541 8.399L60.822 8.399L60.822 8.911Q60.822 9.114 60.705 9.272Q60.588 9.430 60.406 9.514Q60.225 9.598 60.022 9.598Q59.791 9.598 59.639 9.426Q59.486 9.254 59.455 9.024Q59.295 9.305 58.986 9.471Q58.678 9.637 58.326 9.637Q57.815 9.637 57.391 9.414Q56.967 9.192 56.967 8.727M57.654 8.727Q57.654 9.012 57.881 9.198Q58.108 9.383 58.401 9.383Q58.647 9.383 58.871 9.266Q59.096 9.149 59.231 8.946Q59.365 8.743 59.365 8.489L59.365 7.657Q59.100 7.657 58.815 7.711Q58.529 7.766 58.258 7.895Q57.986 8.024 57.820 8.231Q57.654 8.438 57.654 8.727M62.940 9.559L61.143 9.559L61.143 9.262Q61.412 9.262 61.580 9.217Q61.748 9.172 61.748 9L61.748 4.840Q61.748 4.625 61.686 4.530Q61.623 4.434 61.506 4.413Q61.389 4.391 61.143 4.391L61.143 4.094L62.365 4.008L62.365 7.774L63.463 6.887Q63.670 6.707 63.670 6.559Q63.670 6.493 63.617 6.450Q63.565 6.407 63.494 6.407L63.494 6.110L65.029 6.110L65.029 6.407Q64.498 6.407 63.901 6.887L63.291 7.383L64.365 8.782Q64.502 8.957 64.610 9.065Q64.717 9.172 64.852 9.217Q64.987 9.262 65.213 9.262L65.213 9.559L63.588 9.559L63.588 9.262Q63.830 9.262 63.830 9.110Q63.830 9.032 63.787 8.961Q63.744 8.891 63.662 8.782L62.861 7.735L62.334 8.161L62.334 9Q62.334 9.168 62.502 9.215Q62.670 9.262 62.940 9.262L62.940 9.559M66.014 10.856Q66.127 10.934 66.303 10.934Q66.592 10.934 66.813 10.721Q67.033 10.508 67.158 10.207L67.447 9.559L66.174 6.672Q66.092 6.496 65.947 6.452Q65.803 6.407 65.533 6.407L65.533 6.110L67.252 6.110L67.252 6.407Q66.830 6.407 66.830 6.590Q66.830 6.602 66.846 6.672L67.783 8.797L68.615 6.887Q68.654 6.797 68.654 6.719Q68.654 6.579 68.553 6.493Q68.451 6.407 68.311 6.407L68.311 6.110L69.662 6.110L69.662 6.407Q69.408 6.407 69.215 6.532Q69.022 6.657 68.916 6.887L67.471 10.207Q67.358 10.461 67.192 10.684Q67.026 10.907 66.797 11.049Q66.569 11.192 66.303 11.192Q66.006 11.192 65.766 11Q65.526 10.809 65.526 10.520Q65.526 10.364 65.631 10.262Q65.737 10.161 65.885 10.161Q65.990 10.161 66.070 10.207Q66.151 10.254 66.197 10.332Q66.244 10.411 66.244 10.520Q66.244 10.641 66.184 10.729Q66.123 10.817 66.014 10.856\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg fill=\"var(--tk-good)\" stroke=\"var(--tk-good)\">\u003Cg transform=\"translate(-69.022 34.607)\">\u003Cpath d=\"M52.733 9.559L47.885 9.559L47.885 9.262Q48.205 9.262 48.449 9.215Q48.694 9.168 48.694 9L48.694 4.657Q48.694 4.485 48.449 4.438Q48.205 4.391 47.885 4.391L47.885 4.094L52.619 4.094L52.854 5.942L52.572 5.942Q52.490 5.246 52.315 4.926Q52.139 4.606 51.791 4.498Q51.444 4.391 50.725 4.391L49.861 4.391Q49.643 4.391 49.551 4.434Q49.459 4.477 49.459 4.657L49.459 6.575L50.092 6.575Q50.518 6.575 50.725 6.512Q50.932 6.450 51.020 6.254Q51.108 6.059 51.108 5.637L51.389 5.637L51.389 7.805L51.108 7.805Q51.108 7.383 51.020 7.190Q50.932 6.996 50.725 6.934Q50.518 6.871 50.092 6.871L49.459 6.871L49.459 9Q49.459 9.172 49.551 9.217Q49.643 9.262 49.861 9.262L50.787 9.262Q51.369 9.262 51.723 9.180Q52.076 9.098 52.281 8.901Q52.486 8.704 52.600 8.366Q52.713 8.028 52.807 7.446L53.084 7.446L52.733 9.559M58.045 9.559L53.662 9.559L53.662 9.262Q53.983 9.262 54.227 9.215Q54.471 9.168 54.471 9L54.471 4.657Q54.471 4.485 54.227 4.438Q53.983 4.391 53.662 4.391L53.662 4.094L56.248 4.094L56.248 4.391Q55.236 4.391 55.236 4.657L55.236 9Q55.236 9.172 55.328 9.217Q55.420 9.262 55.639 9.262L56.326 9.262Q56.799 9.262 57.108 9.137Q57.416 9.012 57.594 8.782Q57.772 8.551 57.863 8.225Q57.955 7.899 57.998 7.446L58.279 7.446L58.045 9.559M59.776 7.750L59.776 4.657Q59.776 4.485 59.531 4.438Q59.287 4.391 58.967 4.391L58.967 4.094L61.350 4.094L61.350 4.391Q61.029 4.391 60.785 4.440Q60.541 4.489 60.541 4.657L60.541 7.727Q60.541 8.450 60.885 8.940Q61.229 9.430 61.928 9.430Q62.393 9.430 62.764 9.200Q63.135 8.969 63.344 8.577Q63.553 8.184 63.553 7.727L63.553 4.871Q63.553 4.391 62.744 4.391L62.744 4.094L64.654 4.094L64.654 4.391Q63.846 4.391 63.846 4.871L63.846 7.750Q63.846 8.270 63.592 8.727Q63.338 9.184 62.897 9.455Q62.455 9.727 61.928 9.727Q61.518 9.727 61.137 9.584Q60.756 9.442 60.444 9.172Q60.131 8.903 59.953 8.536Q59.776 8.168 59.776 7.750\" fill=\"var(--tk-good)\" stroke=\"var(--tk-good)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg fill=\"var(--tk-hi)\" stroke=\"var(--tk-hi)\">\u003Cg transform=\"translate(-25.613 31.85)\">\u003Cpath d=\"M47.830 9.551L47.830 8.329Q47.830 8.301 47.861 8.270Q47.893 8.239 47.916 8.239L48.022 8.239Q48.092 8.239 48.108 8.301Q48.170 8.621 48.309 8.862Q48.447 9.102 48.680 9.243Q48.912 9.383 49.221 9.383Q49.459 9.383 49.668 9.323Q49.877 9.262 50.014 9.114Q50.151 8.965 50.151 8.719Q50.151 8.465 49.940 8.299Q49.729 8.133 49.459 8.079L48.838 7.965Q48.432 7.887 48.131 7.631Q47.830 7.375 47.830 7Q47.830 6.633 48.031 6.411Q48.233 6.188 48.557 6.090Q48.881 5.993 49.221 5.993Q49.686 5.993 49.983 6.200L50.205 6.016Q50.229 5.993 50.260 5.993L50.311 5.993Q50.342 5.993 50.369 6.020Q50.397 6.047 50.397 6.079L50.397 7.063Q50.397 7.094 50.371 7.123Q50.346 7.153 50.311 7.153L50.205 7.153Q50.170 7.153 50.143 7.125Q50.115 7.098 50.115 7.063Q50.115 6.664 49.863 6.444Q49.611 6.223 49.213 6.223Q48.858 6.223 48.574 6.346Q48.291 6.469 48.291 6.774Q48.291 6.993 48.492 7.125Q48.694 7.258 48.940 7.301L49.565 7.414Q49.994 7.504 50.303 7.801Q50.611 8.098 50.611 8.512Q50.611 9.082 50.213 9.360Q49.815 9.637 49.221 9.637Q48.670 9.637 48.319 9.301L48.022 9.614Q47.998 9.637 47.963 9.637L47.916 9.637Q47.893 9.637 47.861 9.606Q47.830 9.575 47.830 9.551M52.725 9.528L51.654 6.672Q51.588 6.493 51.457 6.450Q51.326 6.407 51.069 6.407L51.069 6.110L52.748 6.110L52.748 6.407Q52.299 6.407 52.299 6.606Q52.303 6.621 52.305 6.639Q52.307 6.657 52.307 6.672L53.100 8.766L53.811 6.856Q53.776 6.762 53.776 6.717Q53.776 6.672 53.740 6.672Q53.674 6.493 53.543 6.450Q53.412 6.407 53.158 6.407L53.158 6.110L54.748 6.110L54.748 6.407Q54.299 6.407 54.299 6.606Q54.303 6.625 54.305 6.643Q54.307 6.661 54.307 6.672L55.139 8.887L55.893 6.887Q55.916 6.829 55.916 6.758Q55.916 6.598 55.779 6.502Q55.643 6.407 55.475 6.407L55.475 6.110L56.861 6.110L56.861 6.407Q56.627 6.407 56.449 6.534Q56.272 6.661 56.190 6.887L55.205 9.528Q55.151 9.637 55.037 9.637L54.979 9.637Q54.865 9.637 54.822 9.528L53.963 7.254L53.108 9.528Q53.069 9.637 52.947 9.637L52.893 9.637Q52.779 9.637 52.725 9.528M59.135 9.559L57.358 9.559L57.358 9.262Q57.631 9.262 57.799 9.215Q57.967 9.168 57.967 9L57.967 6.864Q57.967 6.649 57.910 6.553Q57.854 6.457 57.740 6.436Q57.627 6.414 57.381 6.414L57.381 6.118L58.580 6.032L58.580 9Q58.580 9.168 58.727 9.215Q58.873 9.262 59.135 9.262L59.135 9.559M57.694 4.637Q57.694 4.446 57.828 4.315Q57.963 4.184 58.158 4.184Q58.279 4.184 58.383 4.246Q58.486 4.309 58.549 4.413Q58.611 4.516 58.611 4.637Q58.611 4.832 58.481 4.967Q58.350 5.102 58.158 5.102Q57.959 5.102 57.826 4.969Q57.694 4.836 57.694 4.637M59.678 9.551L59.678 8.329Q59.678 8.301 59.709 8.270Q59.740 8.239 59.764 8.239L59.869 8.239Q59.940 8.239 59.955 8.301Q60.018 8.621 60.156 8.862Q60.295 9.102 60.528 9.243Q60.760 9.383 61.069 9.383Q61.307 9.383 61.516 9.323Q61.725 9.262 61.861 9.114Q61.998 8.965 61.998 8.719Q61.998 8.465 61.787 8.299Q61.576 8.133 61.307 8.079L60.686 7.965Q60.279 7.887 59.979 7.631Q59.678 7.375 59.678 7Q59.678 6.633 59.879 6.411Q60.080 6.188 60.404 6.090Q60.729 5.993 61.069 5.993Q61.533 5.993 61.830 6.200L62.053 6.016Q62.076 5.993 62.108 5.993L62.158 5.993Q62.190 5.993 62.217 6.020Q62.244 6.047 62.244 6.079L62.244 7.063Q62.244 7.094 62.219 7.123Q62.194 7.153 62.158 7.153L62.053 7.153Q62.018 7.153 61.990 7.125Q61.963 7.098 61.963 7.063Q61.963 6.664 61.711 6.444Q61.459 6.223 61.061 6.223Q60.705 6.223 60.422 6.346Q60.139 6.469 60.139 6.774Q60.139 6.993 60.340 7.125Q60.541 7.258 60.787 7.301L61.412 7.414Q61.842 7.504 62.151 7.801Q62.459 8.098 62.459 8.512Q62.459 9.082 62.061 9.360Q61.662 9.637 61.069 9.637Q60.518 9.637 60.166 9.301L59.869 9.614Q59.846 9.637 59.811 9.637L59.764 9.637Q59.740 9.637 59.709 9.606Q59.678 9.575 59.678 9.551M64.916 9.559L63.061 9.559L63.061 9.262Q63.334 9.262 63.502 9.215Q63.670 9.168 63.670 9L63.670 4.840Q63.670 4.625 63.608 4.530Q63.545 4.434 63.426 4.413Q63.307 4.391 63.061 4.391L63.061 4.094L64.283 4.008L64.283 6.711Q64.408 6.500 64.596 6.350Q64.783 6.200 65.010 6.116Q65.237 6.032 65.483 6.032Q66.651 6.032 66.651 7.110L66.651 9Q66.651 9.168 66.820 9.215Q66.990 9.262 67.260 9.262L67.260 9.559L65.404 9.559L65.404 9.262Q65.678 9.262 65.846 9.215Q66.014 9.168 66.014 9L66.014 7.125Q66.014 6.743 65.893 6.514Q65.772 6.286 65.420 6.286Q65.108 6.286 64.854 6.448Q64.600 6.610 64.453 6.879Q64.307 7.149 64.307 7.446L64.307 9Q64.307 9.168 64.477 9.215Q64.647 9.262 64.916 9.262\" fill=\"var(--tk-hi)\" stroke=\"var(--tk-hi)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fsvg>\u003Cfigcaption class=\"tikz-cap\">The rectified family on shared axes. All are the identity for \u003Cspan class=\"katex\">\u003Cspan class=\"katex-html\" aria-hidden=\"true\">\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.5782em;vertical-align:-0.0391em;\">\u003C\u002Fspan>\u003Cspan class=\"mord mathnormal\" style=\"margin-right:0.044em;\">z\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003Cspan class=\"mrel\">>\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003C\u002Fspan>\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.6444em;\">\u003C\u002Fspan>\u003Cspan class=\"mord\">0\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>. ReLU is flat on the left; Leaky keeps a shallow line; ELU and swish curve smoothly, and swish dips below \u003Cspan class=\"katex\">\u003Cspan class=\"katex-html\" aria-hidden=\"true\">\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.6444em;\">\u003C\u002Fspan>\u003Cspan class=\"mord\">0\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan> before returning.\u003C\u002Ffigcaption>","\u003Csvg style=\"width:100%;max-width:320.149px;height:auto\" xmlns=\"http:\u002F\u002Fwww.w3.org\u002F2000\u002Fsvg\" viewBox=\"-75 -75 240.112 87.344\">\u003Cg stroke=\"currentColor\" style=\"stroke-miterlimit:10;stroke-width:.4\">\u003Cg style=\"stroke-width:.8\">\u003Cpath fill=\"none\" d=\"M-58.537 2.28h190.88\"\u002F>\u003Cpath stroke=\"none\" d=\"M134.942 2.28 130.782.2l1.56 2.08-1.56 2.08\"\u002F>\u003Cg transform=\"translate(103.318 1.905)\">\u003Cpath d=\"M37.455 2.281L35.677 2.281L35.677 1.984Q35.951 1.984 36.119 1.937Q36.287 1.890 36.287 1.722L36.287-0.414Q36.287-0.629 36.230-0.725Q36.173-0.821 36.060-0.842Q35.947-0.864 35.701-0.864L35.701-1.160L36.900-1.246L36.900 1.722Q36.900 1.890 37.046 1.937Q37.193 1.984 37.455 1.984L37.455 2.281M36.013-2.641Q36.013-2.832 36.148-2.963Q36.283-3.094 36.478-3.094Q36.599-3.094 36.703-3.031Q36.806-2.969 36.869-2.865Q36.931-2.762 36.931-2.641Q36.931-2.446 36.800-2.311Q36.669-2.176 36.478-2.176Q36.279-2.176 36.146-2.309Q36.013-2.442 36.013-2.641M39.884 2.281L38.029 2.281L38.029 1.984Q38.302 1.984 38.470 1.937Q38.638 1.890 38.638 1.722L38.638-0.414Q38.638-0.629 38.576-0.725Q38.513-0.821 38.394-0.842Q38.275-0.864 38.029-0.864L38.029-1.160L39.220-1.246L39.220-0.512Q39.334-0.727 39.527-0.895Q39.720-1.063 39.959-1.155Q40.197-1.246 40.451-1.246Q41.619-1.246 41.619-0.168L41.619 1.722Q41.619 1.890 41.789 1.937Q41.959 1.984 42.228 1.984L42.228 2.281L40.373 2.281L40.373 1.984Q40.646 1.984 40.814 1.937Q40.982 1.890 40.982 1.722L40.982-0.153Q40.982-0.535 40.861-0.764Q40.740-0.992 40.388-0.992Q40.076-0.992 39.822-0.830Q39.568-0.668 39.421-0.399Q39.275-0.129 39.275 0.168L39.275 1.722Q39.275 1.890 39.445 1.937Q39.615 1.984 39.884 1.984L39.884 2.281M44.556 3.832L42.701 3.832L42.701 3.539Q42.970 3.539 43.138 3.494Q43.306 3.449 43.306 3.273L43.306-0.551Q43.306-0.758 43.150-0.811Q42.994-0.864 42.701-0.864L42.701-1.160L43.923-1.246L43.923-0.781Q44.154-1.004 44.468-1.125Q44.783-1.246 45.123-1.246Q45.595-1.246 46-1Q46.404-0.754 46.636-0.338Q46.869 0.078 46.869 0.554Q46.869 0.929 46.720 1.258Q46.572 1.586 46.302 1.838Q46.033 2.090 45.689 2.224Q45.345 2.359 44.986 2.359Q44.697 2.359 44.425 2.238Q44.154 2.117 43.947 1.906L43.947 3.273Q43.947 3.449 44.115 3.494Q44.283 3.539 44.556 3.539L44.556 3.832M43.947-0.383L43.947 1.457Q44.099 1.746 44.361 1.926Q44.623 2.105 44.931 2.105Q45.216 2.105 45.439 1.967Q45.662 1.828 45.814 1.597Q45.966 1.367 46.044 1.095Q46.123 0.824 46.123 0.554Q46.123 0.222 45.998-0.135Q45.873-0.492 45.625-0.729Q45.377-0.965 45.029-0.965Q44.705-0.965 44.410-0.809Q44.115-0.653 43.947-0.383M48.076 1.328L48.076-0.414Q48.076-0.629 48.013-0.725Q47.951-0.821 47.832-0.842Q47.712-0.864 47.466-0.864L47.466-1.160L48.712-1.246L48.712 1.304L48.712 1.328Q48.712 1.640 48.767 1.802Q48.822 1.965 48.972 2.035Q49.123 2.105 49.443 2.105Q49.873 2.105 50.146 1.767Q50.419 1.429 50.419 0.984L50.419-0.414Q50.419-0.629 50.357-0.725Q50.294-0.821 50.175-0.842Q50.056-0.864 49.810-0.864L49.810-1.160L51.056-1.246L51.056 1.539Q51.056 1.750 51.119 1.845Q51.181 1.941 51.300 1.963Q51.419 1.984 51.666 1.984L51.666 2.281L50.443 2.359L50.443 1.738Q50.275 2.027 49.994 2.193Q49.712 2.359 49.392 2.359Q48.076 2.359 48.076 1.328M52.736 1.320L52.736-0.871L52.033-0.871L52.033-1.125Q52.388-1.125 52.630-1.358Q52.873-1.590 52.984-1.938Q53.095-2.285 53.095-2.641L53.377-2.641L53.377-1.168L54.552-1.168L54.552-0.871L53.377-0.871L53.377 1.304Q53.377 1.625 53.496 1.853Q53.615 2.082 53.896 2.082Q54.076 2.082 54.193 1.959Q54.310 1.836 54.363 1.656Q54.416 1.476 54.416 1.304L54.416 0.832L54.697 0.832L54.697 1.320Q54.697 1.574 54.591 1.814Q54.486 2.054 54.289 2.207Q54.091 2.359 53.834 2.359Q53.517 2.359 53.265 2.236Q53.013 2.113 52.875 1.879Q52.736 1.644 52.736 1.320\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg style=\"stroke-width:.8\">\u003Cpath fill=\"none\" d=\"M35.357 7.971v-65.686\"\u002F>\u003Cpath stroke=\"none\" d=\"m35.357-60.315-2.08 4.16 2.08-1.56 2.08 1.56\"\u002F>\u003Cg transform=\"translate(-21.25 -66.329)\">\u003Cpath d=\"M37.220 2.320Q36.755 2.320 36.390 2.070Q36.025 1.820 35.820 1.416Q35.615 1.011 35.615 0.554Q35.615 0.211 35.740-0.110Q35.865-0.430 36.097-0.680Q36.330-0.930 36.634-1.069Q36.939-1.207 37.294-1.207Q37.806-1.207 38.212-0.887L38.212-2.047L37.791-2.047Q37.580-2.071 37.541-2.285L37.541-2.375Q37.580-2.582 37.791-2.606L38.603-2.606Q38.802-2.582 38.853-2.375L38.853 1.722L39.279 1.722Q39.486 1.746 39.525 1.953L39.525 2.043Q39.486 2.258 39.279 2.281L38.462 2.281Q38.263 2.258 38.212 2.043L38.212 1.914Q38.017 2.105 37.755 2.213Q37.494 2.320 37.220 2.320M37.259 1.761Q37.619 1.761 37.871 1.492Q38.123 1.222 38.212 0.847L38.212 0.015Q38.154-0.172 38.027-0.323Q37.900-0.473 37.722-0.561Q37.544-0.649 37.349-0.649Q37.041-0.649 36.791-0.479Q36.541-0.309 36.396-0.024Q36.252 0.261 36.252 0.562Q36.252 1.011 36.537 1.386Q36.822 1.761 37.259 1.761M43.115 0.793L40.673 0.793Q40.728 1.070 40.925 1.293Q41.123 1.515 41.400 1.638Q41.677 1.761 41.962 1.761Q42.435 1.761 42.658 1.472Q42.666 1.461 42.722 1.355Q42.779 1.250 42.828 1.207Q42.877 1.164 42.970 1.152L43.115 1.152Q43.306 1.172 43.365 1.386L43.365 1.441Q43.298 1.742 43.068 1.939Q42.837 2.136 42.525 2.228Q42.212 2.320 41.908 2.320Q41.423 2.320 40.984 2.092Q40.544 1.863 40.277 1.463Q40.009 1.062 40.009 0.570L40.009 0.511Q40.009 0.043 40.255-0.360Q40.502-0.762 40.910-0.996Q41.318-1.231 41.787-1.231Q42.291-1.231 42.644-1.008Q42.998-0.785 43.181-0.397Q43.365-0.008 43.365 0.496L43.365 0.554Q43.306 0.769 43.115 0.793M40.681 0.242L42.709 0.242Q42.662-0.168 42.423-0.420Q42.185-0.672 41.787-0.672Q41.392-0.672 41.086-0.410Q40.779-0.149 40.681 0.242M44.072 2.043L44.072 1.953Q44.130 1.746 44.322 1.722L45.033 1.722L45.033-0.606L44.322-0.606Q44.127-0.629 44.072-0.848L44.072-0.934Q44.130-1.145 44.322-1.168L45.423-1.168Q45.623-1.149 45.673-0.934L45.673-0.606Q45.935-0.891 46.291-1.049Q46.646-1.207 47.033-1.207Q47.326-1.207 47.560-1.073Q47.794-0.938 47.794-0.672Q47.794-0.504 47.685-0.387Q47.576-0.270 47.408-0.270Q47.255-0.270 47.140-0.381Q47.025-0.492 47.025-0.649Q46.650-0.649 46.336-0.448Q46.021-0.246 45.847 0.088Q45.673 0.422 45.673 0.801L45.673 1.722L46.619 1.722Q46.826 1.746 46.865 1.953L46.865 2.043Q46.826 2.258 46.619 2.281L44.322 2.281Q44.130 2.258 44.072 2.043M48.712 2.043L48.712 1.953Q48.763 1.746 48.959 1.722L49.998 1.722L49.998-0.606L49.025-0.606Q48.826-0.629 48.775-0.848L48.775-0.934Q48.826-1.145 49.025-1.168L50.392-1.168Q50.587-1.149 50.638-0.934L50.638 1.722L51.552 1.722Q51.748 1.746 51.798 1.953L51.798 2.043Q51.748 2.258 51.552 2.281L48.959 2.281Q48.763 2.258 48.712 2.043M49.744-2.145L49.744-2.199Q49.744-2.371 49.880-2.492Q50.017-2.614 50.193-2.614Q50.365-2.614 50.502-2.492Q50.638-2.371 50.638-2.199L50.638-2.145Q50.638-1.969 50.502-1.848Q50.365-1.727 50.193-1.727Q50.017-1.727 49.880-1.848Q49.744-1.969 49.744-2.145M53.998 2.058L53.111-0.606L52.791-0.606Q52.591-0.629 52.541-0.848L52.541-0.934Q52.591-1.145 52.791-1.168L53.951-1.168Q54.146-1.149 54.197-0.934L54.197-0.848Q54.146-0.629 53.951-0.606L53.669-0.606L54.462 1.769L55.252-0.606L54.974-0.606Q54.775-0.629 54.724-0.848L54.724-0.934Q54.775-1.145 54.974-1.168L56.134-1.168Q56.330-1.145 56.380-0.934L56.380-0.848Q56.330-0.629 56.134-0.606L55.814-0.606L54.927 2.058Q54.884 2.172 54.783 2.246Q54.681 2.320 54.556 2.320L54.365 2.320Q54.248 2.320 54.144 2.248Q54.041 2.176 53.998 2.058M56.994 1.168Q56.994 0.722 57.408 0.465Q57.822 0.207 58.363 0.107Q58.904 0.008 59.4120Q59.412-0.215 59.277-0.367Q59.142-0.520 58.935-0.596Q58.728-0.672 58.517-0.672Q58.173-0.672 58.013-0.649L58.013-0.590Q58.013-0.422 57.894-0.307Q57.775-0.192 57.611-0.192Q57.435-0.192 57.320-0.315Q57.205-0.438 57.205-0.606Q57.205-1.012 57.586-1.121Q57.966-1.231 58.525-1.231Q58.794-1.231 59.062-1.153Q59.330-1.074 59.554-0.924Q59.779-0.774 59.916-0.553Q60.052-0.332 60.052-0.055L60.052 1.664Q60.052 1.722 60.580 1.722Q60.775 1.742 60.826 1.953L60.826 2.043Q60.775 2.258 60.580 2.281L60.435 2.281Q60.091 2.281 59.863 2.234Q59.634 2.187 59.490 2Q59.029 2.320 58.322 2.320Q57.986 2.320 57.681 2.179Q57.377 2.039 57.185 1.777Q56.994 1.515 56.994 1.168M57.634 1.176Q57.634 1.449 57.877 1.605Q58.119 1.761 58.404 1.761Q58.623 1.761 58.855 1.703Q59.087 1.644 59.250 1.506Q59.412 1.367 59.412 1.144L59.412 0.554Q59.130 0.554 58.714 0.611Q58.298 0.668 57.966 0.806Q57.634 0.945 57.634 1.176M62.033 1.176L62.033-0.606L61.283-0.606Q61.084-0.629 61.033-0.848L61.033-0.934Q61.084-1.145 61.283-1.168L62.033-1.168L62.033-1.918Q62.084-2.125 62.283-2.153L62.427-2.153Q62.623-2.125 62.673-1.918L62.673-1.168L64.033-1.168Q64.224-1.149 64.283-0.934L64.283-0.848Q64.228-0.629 64.033-0.606L62.673-0.606L62.673 1.144Q62.673 1.761 63.248 1.761Q63.498 1.761 63.662 1.576Q63.826 1.390 63.826 1.144Q63.826 1.051 63.898 0.980Q63.970 0.910 64.072 0.898L64.216 0.898Q64.416 0.922 64.466 1.129L64.466 1.176Q64.466 1.500 64.283 1.763Q64.099 2.027 63.806 2.174Q63.513 2.320 63.193 2.320Q62.681 2.320 62.357 2.010Q62.033 1.699 62.033 1.176M65.697 2.043L65.697 1.953Q65.748 1.746 65.943 1.722L66.982 1.722L66.982-0.606L66.009-0.606Q65.810-0.629 65.759-0.848L65.759-0.934Q65.810-1.145 66.009-1.168L67.377-1.168Q67.572-1.149 67.623-0.934L67.623 1.722L68.537 1.722Q68.732 1.746 68.783 1.953L68.783 2.043Q68.732 2.258 68.537 2.281L65.943 2.281Q65.748 2.258 65.697 2.043M66.728-2.145L66.728-2.199Q66.728-2.371 66.865-2.492Q67.002-2.614 67.177-2.614Q67.349-2.614 67.486-2.492Q67.623-2.371 67.623-2.199L67.623-2.145Q67.623-1.969 67.486-1.848Q67.349-1.727 67.177-1.727Q67.002-1.727 66.865-1.848Q66.728-1.969 66.728-2.145M70.982 2.058L70.095-0.606L69.775-0.606Q69.576-0.629 69.525-0.848L69.525-0.934Q69.576-1.145 69.775-1.168L70.935-1.168Q71.130-1.149 71.181-0.934L71.181-0.848Q71.130-0.629 70.935-0.606L70.654-0.606L71.447 1.769L72.236-0.606L71.959-0.606Q71.759-0.629 71.709-0.848L71.709-0.934Q71.759-1.145 71.959-1.168L73.119-1.168Q73.314-1.145 73.365-0.934L73.365-0.848Q73.314-0.629 73.119-0.606L72.798-0.606L71.912 2.058Q71.869 2.172 71.767 2.246Q71.666 2.320 71.541 2.320L71.349 2.320Q71.232 2.320 71.128 2.248Q71.025 2.176 70.982 2.058M77.084 0.793L74.642 0.793Q74.697 1.070 74.894 1.293Q75.091 1.515 75.369 1.638Q75.646 1.761 75.931 1.761Q76.404 1.761 76.627 1.472Q76.634 1.461 76.691 1.355Q76.748 1.250 76.796 1.207Q76.845 1.164 76.939 1.152L77.084 1.152Q77.275 1.172 77.334 1.386L77.334 1.441Q77.267 1.742 77.037 1.939Q76.806 2.136 76.494 2.228Q76.181 2.320 75.877 2.320Q75.392 2.320 74.953 2.092Q74.513 1.863 74.246 1.463Q73.978 1.062 73.978 0.570L73.978 0.511Q73.978 0.043 74.224-0.360Q74.470-0.762 74.878-0.996Q75.287-1.231 75.755-1.231Q76.259-1.231 76.613-1.008Q76.966-0.785 77.150-0.397Q77.334-0.008 77.334 0.496L77.334 0.554Q77.275 0.769 77.084 0.793M74.650 0.242L76.677 0.242Q76.630-0.168 76.392-0.420Q76.154-0.672 75.755-0.672Q75.361-0.672 75.054-0.410Q74.748-0.149 74.650 0.242\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg style=\"stroke-dasharray:3.0,3.0\">\u003Cpath fill=\"none\" d=\"M-58.537-54.625h187.789\"\u002F>\u003Cg transform=\"translate(97.427 -54.328)\">\u003Cpath d=\"M38.951 2.281L36.158 2.281L36.158 1.984Q37.220 1.984 37.220 1.722L37.220-2.446Q36.791-2.231 36.111-2.231L36.111-2.528Q37.130-2.528 37.646-3.039L37.791-3.039Q37.865-3.020 37.884-2.942L37.884 1.722Q37.884 1.984 38.951 1.984\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg style=\"stroke-dasharray:3.0,3.0\">\u003Cpath fill=\"none\" d=\"M-58.537-11.946h187.789\"\u002F>\u003Cg transform=\"translate(97.427 -11.649)\">\u003Cpath d=\"M37.478 2.449Q36.775 2.449 36.375 2.049Q35.974 1.648 35.830 1.039Q35.685 0.429 35.685-0.270Q35.685-0.793 35.755-1.256Q35.826-1.719 36.019-2.131Q36.212-2.543 36.570-2.791Q36.927-3.039 37.478-3.039Q38.029-3.039 38.386-2.791Q38.744-2.543 38.935-2.133Q39.127-1.723 39.197-1.254Q39.267-0.785 39.267-0.270Q39.267 0.429 39.125 1.037Q38.982 1.644 38.582 2.047Q38.181 2.449 37.478 2.449M37.478 2.191Q37.951 2.191 38.183 1.756Q38.416 1.320 38.470 0.781Q38.525 0.242 38.525-0.399Q38.525-1.395 38.341-2.088Q38.158-2.781 37.478-2.781Q37.111-2.781 36.890-2.543Q36.669-2.305 36.574-1.948Q36.478-1.590 36.453-1.219Q36.427-0.848 36.427-0.399Q36.427 0.242 36.482 0.781Q36.537 1.320 36.769 1.756Q37.002 2.191 37.478 2.191M40.322 1.816Q40.322 1.633 40.459 1.496Q40.595 1.359 40.787 1.359Q40.978 1.359 41.111 1.492Q41.244 1.625 41.244 1.816Q41.244 2.015 41.111 2.148Q40.978 2.281 40.787 2.281Q40.595 2.281 40.459 2.144Q40.322 2.008 40.322 1.816M45.548 2.281L42.388 2.281L42.388 2.074Q42.388 2.047 42.412 2.015L43.763 0.617Q44.142 0.230 44.390-0.059Q44.638-0.348 44.812-0.705Q44.986-1.063 44.986-1.453Q44.986-1.801 44.853-2.094Q44.720-2.387 44.466-2.565Q44.212-2.742 43.857-2.742Q43.498-2.742 43.207-2.547Q42.916-2.352 42.771-2.024L42.826-2.024Q43.009-2.024 43.134-1.903Q43.259-1.781 43.259-1.590Q43.259-1.410 43.134-1.281Q43.009-1.153 42.826-1.153Q42.646-1.153 42.517-1.281Q42.388-1.410 42.388-1.590Q42.388-1.992 42.609-2.328Q42.830-2.664 43.195-2.852Q43.560-3.039 43.962-3.039Q44.443-3.039 44.859-2.852Q45.275-2.664 45.527-2.303Q45.779-1.942 45.779-1.453Q45.779-1.094 45.625-0.791Q45.470-0.489 45.218-0.229Q44.966 0.031 44.617 0.316Q44.267 0.601 44.099 0.754L43.169 1.594L43.884 1.594Q45.259 1.594 45.298 1.554Q45.369 1.476 45.412 1.291Q45.455 1.105 45.498 0.816L45.779 0.816L45.548 2.281M47.048 1.402L46.986 1.402Q47.127 1.754 47.451 1.965Q47.775 2.176 48.162 2.176Q48.755 2.176 49.005 1.742Q49.255 1.308 49.255 0.672Q49.255 0.078 49.086-0.369Q48.916-0.817 48.416-0.817Q48.119-0.817 47.914-0.737Q47.709-0.656 47.607-0.565Q47.505-0.473 47.390-0.340Q47.275-0.207 47.224-0.192L47.154-0.192Q47.068-0.215 47.048-0.293L47.048-2.942Q47.080-3.039 47.154-3.039Q47.169-3.039 47.177-3.037Q47.185-3.035 47.193-3.031Q47.779-2.781 48.377-2.781Q48.959-2.781 49.576-3.039L49.599-3.039Q49.642-3.039 49.669-3.014Q49.697-2.989 49.697-2.949L49.697-2.871Q49.697-2.840 49.673-2.817Q49.377-2.465 48.955-2.268Q48.533-2.071 48.072-2.071Q47.724-2.071 47.345-2.176L47.345-0.680Q47.564-0.875 47.839-0.973Q48.115-1.071 48.416-1.071Q48.873-1.071 49.242-0.823Q49.611-0.574 49.818-0.170Q50.025 0.234 50.025 0.679Q50.025 1.168 49.769 1.576Q49.513 1.984 49.082 2.217Q48.650 2.449 48.162 2.449Q47.767 2.449 47.412 2.258Q47.056 2.066 46.845 1.732Q46.634 1.398 46.634 0.984Q46.634 0.804 46.752 0.691Q46.869 0.578 47.048 0.578Q47.166 0.578 47.257 0.631Q47.349 0.683 47.402 0.775Q47.455 0.867 47.455 0.984Q47.455 1.168 47.341 1.285Q47.228 1.402 47.048 1.402\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cpath fill=\"none\" stroke=\"var(--tk-accent)\" d=\"m-55.691 2.187 2.045-.015 2.046-.017 2.046-.019 2.046-.022 2.045-.026 2.046-.03 2.046-.034 2.046-.038 2.045-.046 2.046-.051 2.046-.06 2.046-.07 2.045-.079 2.046-.09 2.046-.103 2.046-.12 2.045-.136 2.046-.155L-16.822.9l2.045-.201 2.046-.229 2.046-.26 2.046-.29 2.045-.33 2.046-.37 2.046-.411 2.046-.458 2.045-.505 2.046-.554 2.046-.601 2.046-.65 2.045-.694 2.046-.732 2.046-.762 2.045-.78 2.046-.788 2.046-.775 2.046-.745 2.045-.693 2.046-.622 2.046-.528 2.046-.414 2.045-.287 2.046-.146 2.046-.002 2.046.145 2.045.284 2.046.415 2.046.525 2.046.618 2.045.695 2.046.742 2.046.775 2.045.79 2.046.78 2.046.76 2.046.739 2.045.692 2.046.647 2.046.609 2.046.548 2.045.507 2.046.46 2.046.409 2.046.366 2.045.34 2.046.294 2.046.248 2.045.23L87.51.9l2.046.178 2.046.153 2.045.138 2.046.115 2.046.106 2.046.094 2.045.067 2.046.067 2.046.073 2.046.05 2.045.04 2.046.045 2.046.033 2.046.018 2.045.034 2.046.023 2.046.016 2.045.017 2.046.023\" style=\"stroke-width:1.2\"\u002F>\u003Cpath fill=\"none\" stroke=\"var(--tk-good)\" d=\"m-55.691 1.904 2.045-.057 2.046-.069 2.046-.076 2.046-.088 2.045-.102 2.046-.12 2.046-.138 2.046-.157L-37.28.92l2.046-.21 2.046-.239 2.046-.276 2.045-.317 2.046-.36 2.046-.415 2.046-.479 2.045-.539 2.046-.618 2.046-.71 2.045-.81 2.046-.915 2.046-1.035 2.046-1.172 2.045-1.315 2.046-1.485 2.046-1.648 2.046-1.83 2.045-2.017 2.046-2.215 2.046-2.407 2.046-2.6 2.045-2.776 2.046-2.93 2.046-3.042 2.045-3.12 2.046-3.15 2.046-3.1 2.046-2.983 2.045-2.77 2.046-2.487 2.046-2.112 2.046-1.658 2.045-1.145 2.046-.586 2.046-.003 2.046.577 2.045 1.138 2.046 1.652 2.046 2.105 2.046 2.48 2.045 2.774 2.046 2.978 2.046 3.096 2.045 3.146 2.046 3.124 2.046 3.056 2.046 2.926 2.045 2.775 2.046 2.59 2.046 2.425 2.046 2.223 2.045 2.017 2.046 1.83 2.046 1.648 2.046 1.485 2.045 1.32 2.046 1.172 2.046 1.036 2.045.915 2.046.812 2.046.71 2.046.62 2.045.546 2.046.472 2.046.416 2.046.363 2.045.322 2.046.271 2.046.241 2.046.208 2.045.18 2.046.159 2.046.134 2.046.12 2.045.102 2.046.09 2.046.078 2.045.067 2.046.057\" style=\"stroke-width:1.2\"\u002F>\u003Cpath fill=\"none\" stroke=\"var(--tk-warn)\" d=\"M-55.691 2.28h91.048M35.357-54.625h91.05\" style=\"stroke-width:1.2\"\u002F>\u003Cpath fill=\"none\" stroke=\"var(--tk-warn)\" d=\"M35.357 2.28v-56.905\" style=\"stroke-dasharray:1.2,2.0;stroke-width:1.2\"\u002F>\u003Cg fill=\"var(--tk-accent)\" stroke=\"var(--tk-accent)\">\u003Cg transform=\"translate(32.999 -20.738)\">\u003Cpath d=\"M35.638 2.273L35.638 1.051Q35.638 1.023 35.669 0.992Q35.701 0.961 35.724 0.961L35.830 0.961Q35.900 0.961 35.916 1.023Q35.978 1.344 36.117 1.584Q36.255 1.824 36.488 1.965Q36.720 2.105 37.029 2.105Q37.267 2.105 37.476 2.045Q37.685 1.984 37.822 1.836Q37.959 1.687 37.959 1.441Q37.959 1.187 37.748 1.021Q37.537 0.855 37.267 0.801L36.646 0.687Q36.240 0.609 35.939 0.353Q35.638 0.097 35.638-0.278Q35.638-0.645 35.839-0.867Q36.041-1.090 36.365-1.188Q36.689-1.285 37.029-1.285Q37.494-1.285 37.791-1.078L38.013-1.262Q38.037-1.285 38.068-1.285L38.119-1.285Q38.150-1.285 38.177-1.258Q38.205-1.231 38.205-1.199L38.205-0.215Q38.205-0.184 38.179-0.155Q38.154-0.125 38.119-0.125L38.013-0.125Q37.978-0.125 37.951-0.153Q37.923-0.180 37.923-0.215Q37.923-0.614 37.671-0.834Q37.419-1.055 37.021-1.055Q36.666-1.055 36.382-0.932Q36.099-0.809 36.099-0.504Q36.099-0.285 36.300-0.153Q36.502-0.020 36.748 0.023L37.373 0.136Q37.802 0.226 38.111 0.523Q38.419 0.820 38.419 1.234Q38.419 1.804 38.021 2.082Q37.623 2.359 37.029 2.359Q36.478 2.359 36.127 2.023L35.830 2.336Q35.806 2.359 35.771 2.359L35.724 2.359Q35.701 2.359 35.669 2.328Q35.638 2.297 35.638 2.273M40.806 2.281L39.029 2.281L39.029 1.984Q39.302 1.984 39.470 1.937Q39.638 1.890 39.638 1.722L39.638-0.414Q39.638-0.629 39.582-0.725Q39.525-0.821 39.412-0.842Q39.298-0.864 39.052-0.864L39.052-1.160L40.252-1.246L40.252 1.722Q40.252 1.890 40.398 1.937Q40.544 1.984 40.806 1.984L40.806 2.281M39.365-2.641Q39.365-2.832 39.500-2.963Q39.634-3.094 39.830-3.094Q39.951-3.094 40.054-3.031Q40.158-2.969 40.220-2.865Q40.283-2.762 40.283-2.641Q40.283-2.446 40.152-2.311Q40.021-2.176 39.830-2.176Q39.630-2.176 39.498-2.309Q39.365-2.442 39.365-2.641M41.306 2.890Q41.306 2.609 41.517 2.398Q41.728 2.187 42.013 2.097Q41.857 1.972 41.779 1.783Q41.701 1.594 41.701 1.394Q41.701 1.039 41.931 0.746Q41.564 0.406 41.564-0.063Q41.564-0.414 41.767-0.684Q41.970-0.953 42.291-1.100Q42.611-1.246 42.955-1.246Q43.474-1.246 43.845-0.965Q44.209-1.336 44.755-1.336Q44.935-1.336 45.062-1.209Q45.189-1.082 45.189-0.903Q45.189-0.797 45.111-0.719Q45.033-0.641 44.923-0.641Q44.814-0.641 44.738-0.717Q44.662-0.793 44.662-0.903Q44.662-1.004 44.701-1.055Q44.709-1.063 44.712-1.069Q44.716-1.074 44.716-1.078Q44.341-1.078 44.021-0.824Q44.341-0.485 44.341-0.063Q44.341 0.207 44.224 0.424Q44.107 0.640 43.902 0.799Q43.697 0.957 43.455 1.039Q43.212 1.121 42.955 1.121Q42.736 1.121 42.523 1.062Q42.310 1.004 42.115 0.883Q42.021 1.023 42.021 1.203Q42.021 1.410 42.158 1.562Q42.294 1.715 42.502 1.715L43.197 1.715Q43.685 1.715 44.097 1.799Q44.509 1.883 44.789 2.140Q45.068 2.398 45.068 2.890Q45.068 3.254 44.748 3.486Q44.427 3.719 43.986 3.820Q43.544 3.922 43.189 3.922Q42.834 3.922 42.390 3.820Q41.947 3.719 41.627 3.486Q41.306 3.254 41.306 2.890M41.810 2.890Q41.810 3.086 41.955 3.234Q42.099 3.383 42.312 3.472Q42.525 3.562 42.765 3.609Q43.005 3.656 43.189 3.656Q43.431 3.656 43.761 3.578Q44.091 3.500 44.328 3.326Q44.564 3.152 44.564 2.890Q44.564 2.484 44.154 2.375Q43.744 2.265 43.181 2.265L42.502 2.265Q42.232 2.265 42.021 2.443Q41.810 2.621 41.810 2.890M42.955 0.855Q43.677 0.855 43.677-0.063Q43.677-0.985 42.955-0.985Q42.228-0.985 42.228-0.063Q42.228 0.855 42.955 0.855M47.482 2.281L45.627 2.281L45.627 1.984Q45.900 1.984 46.068 1.937Q46.236 1.890 46.236 1.722L46.236-0.414Q46.236-0.629 46.173-0.725Q46.111-0.821 45.992-0.842Q45.873-0.864 45.627-0.864L45.627-1.160L46.818-1.246L46.818-0.512Q46.931-0.727 47.125-0.895Q47.318-1.063 47.556-1.155Q47.794-1.246 48.048-1.246Q49.009-1.246 49.185-0.535Q49.369-0.864 49.697-1.055Q50.025-1.246 50.404-1.246Q51.580-1.246 51.580-0.168L51.580 1.722Q51.580 1.890 51.748 1.937Q51.916 1.984 52.185 1.984L52.185 2.281L50.330 2.281L50.330 1.984Q50.603 1.984 50.771 1.939Q50.939 1.894 50.939 1.722L50.939-0.153Q50.939-0.539 50.814-0.766Q50.689-0.992 50.337-0.992Q50.033-0.992 49.777-0.830Q49.521-0.668 49.373-0.399Q49.224-0.129 49.224 0.168L49.224 1.722Q49.224 1.890 49.394 1.937Q49.564 1.984 49.834 1.984L49.834 2.281L47.978 2.281L47.978 1.984Q48.252 1.984 48.419 1.937Q48.587 1.890 48.587 1.722L48.587-0.153Q48.587-0.539 48.462-0.766Q48.337-0.992 47.986-0.992Q47.681-0.992 47.425-0.830Q47.169-0.668 47.021-0.399Q46.873-0.129 46.873 0.168L46.873 1.722Q46.873 1.890 47.043 1.937Q47.212 1.984 47.482 1.984L47.482 2.281M52.630 0.586Q52.630 0.082 52.886-0.350Q53.142-0.781 53.578-1.033Q54.013-1.285 54.513-1.285Q54.900-1.285 55.242-1.141Q55.584-0.996 55.845-0.735Q56.107-0.473 56.250-0.137Q56.392 0.199 56.392 0.586Q56.392 1.078 56.128 1.488Q55.865 1.898 55.435 2.129Q55.005 2.359 54.513 2.359Q54.021 2.359 53.587 2.127Q53.154 1.894 52.892 1.486Q52.630 1.078 52.630 0.586M54.513 2.082Q54.970 2.082 55.222 1.859Q55.474 1.636 55.562 1.285Q55.650 0.933 55.650 0.488Q55.650 0.058 55.556-0.280Q55.462-0.617 55.209-0.824Q54.955-1.031 54.513-1.031Q53.865-1.031 53.621-0.615Q53.377-0.199 53.377 0.488Q53.377 0.933 53.464 1.285Q53.552 1.636 53.804 1.859Q54.056 2.082 54.513 2.082M58.736 2.281L56.959 2.281L56.959 1.984Q57.232 1.984 57.400 1.937Q57.568 1.890 57.568 1.722L57.568-0.414Q57.568-0.629 57.511-0.725Q57.455-0.821 57.341-0.842Q57.228-0.864 56.982-0.864L56.982-1.160L58.181-1.246L58.181 1.722Q58.181 1.890 58.328 1.937Q58.474 1.984 58.736 1.984L58.736 2.281M57.294-2.641Q57.294-2.832 57.429-2.963Q57.564-3.094 57.759-3.094Q57.880-3.094 57.984-3.031Q58.087-2.969 58.150-2.865Q58.212-2.762 58.212-2.641Q58.212-2.446 58.082-2.311Q57.951-2.176 57.759-2.176Q57.560-2.176 57.427-2.309Q57.294-2.442 57.294-2.641M61.052 2.359Q60.572 2.359 60.164 2.115Q59.755 1.871 59.517 1.457Q59.279 1.043 59.279 0.554Q59.279 0.062 59.537-0.354Q59.794-0.770 60.226-1.008Q60.658-1.246 61.150-1.246Q61.771-1.246 62.220-0.809L62.220-2.438Q62.220-2.653 62.158-2.748Q62.095-2.844 61.978-2.865Q61.861-2.887 61.615-2.887L61.615-3.184L62.837-3.270L62.837 1.539Q62.837 1.750 62.900 1.845Q62.962 1.941 63.080 1.963Q63.197 1.984 63.447 1.984L63.447 2.281L62.197 2.359L62.197 1.875Q61.732 2.359 61.052 2.359M61.119 2.105Q61.459 2.105 61.752 1.914Q62.044 1.722 62.197 1.426L62.197-0.406Q62.048-0.680 61.787-0.836Q61.525-0.992 61.212-0.992Q60.587-0.992 60.304-0.545Q60.021-0.098 60.021 0.562Q60.021 1.207 60.273 1.656Q60.525 2.105 61.119 2.105M64.541-0.942Q64.541-0.969 64.572-1.016Q64.877-1.270 65.048-1.635Q65.220-2 65.220-2.399L65.220-2.473Q65.087-2.352 64.900-2.352Q64.709-2.352 64.572-2.485Q64.435-2.617 64.435-2.817Q64.435-3.012 64.572-3.141Q64.709-3.270 64.900-3.270Q65.107-3.270 65.236-3.133Q65.365-2.996 65.421-2.795Q65.478-2.594 65.478-2.399Q65.478-1.942 65.283-1.528Q65.087-1.114 64.748-0.824Q64.720-0.797 64.685-0.797Q64.634-0.797 64.587-0.844Q64.541-0.891 64.541-0.942\" fill=\"var(--tk-accent)\" stroke=\"var(--tk-accent)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg fill=\"var(--tk-good)\" stroke=\"var(--tk-good)\">\u003Cg transform=\"translate(23.45 -41.324)\">\u003Cpath d=\"M36.220 1.320L36.220-0.871L35.517-0.871L35.517-1.125Q35.873-1.125 36.115-1.358Q36.357-1.590 36.468-1.938Q36.580-2.285 36.580-2.641L36.861-2.641L36.861-1.168L38.037-1.168L38.037-0.871L36.861-0.871L36.861 1.304Q36.861 1.625 36.980 1.853Q37.099 2.082 37.380 2.082Q37.560 2.082 37.677 1.959Q37.794 1.836 37.847 1.656Q37.900 1.476 37.900 1.304L37.900 0.832L38.181 0.832L38.181 1.320Q38.181 1.574 38.076 1.814Q37.970 2.054 37.773 2.207Q37.576 2.359 37.318 2.359Q37.002 2.359 36.750 2.236Q36.498 2.113 36.359 1.879Q36.220 1.644 36.220 1.320M38.998 1.449Q38.998 0.965 39.400 0.670Q39.802 0.375 40.353 0.256Q40.904 0.136 41.396 0.136L41.396-0.153Q41.396-0.379 41.281-0.586Q41.166-0.793 40.968-0.912Q40.771-1.031 40.541-1.031Q40.115-1.031 39.830-0.926Q39.900-0.899 39.947-0.844Q39.994-0.789 40.019-0.719Q40.044-0.649 40.044-0.574Q40.044-0.469 39.994-0.377Q39.943-0.285 39.851-0.235Q39.759-0.184 39.654-0.184Q39.548-0.184 39.457-0.235Q39.365-0.285 39.314-0.377Q39.263-0.469 39.263-0.574Q39.263-0.992 39.652-1.139Q40.041-1.285 40.541-1.285Q40.873-1.285 41.226-1.155Q41.580-1.024 41.808-0.770Q42.037-0.516 42.037-0.168L42.037 1.633Q42.037 1.765 42.109 1.875Q42.181 1.984 42.310 1.984Q42.435 1.984 42.503 1.879Q42.572 1.773 42.572 1.633L42.572 1.121L42.853 1.121L42.853 1.633Q42.853 1.836 42.736 1.994Q42.619 2.152 42.437 2.236Q42.255 2.320 42.052 2.320Q41.822 2.320 41.669 2.148Q41.517 1.976 41.486 1.746Q41.326 2.027 41.017 2.193Q40.709 2.359 40.357 2.359Q39.845 2.359 39.421 2.136Q38.998 1.914 38.998 1.449M39.685 1.449Q39.685 1.734 39.912 1.920Q40.138 2.105 40.431 2.105Q40.677 2.105 40.902 1.988Q41.127 1.871 41.261 1.668Q41.396 1.465 41.396 1.211L41.396 0.379Q41.130 0.379 40.845 0.433Q40.560 0.488 40.289 0.617Q40.017 0.746 39.851 0.953Q39.685 1.160 39.685 1.449M45.076 2.281L43.220 2.281L43.220 1.984Q43.494 1.984 43.662 1.937Q43.830 1.890 43.830 1.722L43.830-0.414Q43.830-0.629 43.767-0.725Q43.705-0.821 43.586-0.842Q43.466-0.864 43.220-0.864L43.220-1.160L44.412-1.246L44.412-0.512Q44.525-0.727 44.718-0.895Q44.912-1.063 45.150-1.155Q45.388-1.246 45.642-1.246Q46.810-1.246 46.810-0.168L46.810 1.722Q46.810 1.890 46.980 1.937Q47.150 1.984 47.419 1.984L47.419 2.281L45.564 2.281L45.564 1.984Q45.837 1.984 46.005 1.937Q46.173 1.890 46.173 1.722L46.173-0.153Q46.173-0.535 46.052-0.764Q45.931-0.992 45.580-0.992Q45.267-0.992 45.013-0.830Q44.759-0.668 44.613-0.399Q44.466-0.129 44.466 0.168L44.466 1.722Q44.466 1.890 44.636 1.937Q44.806 1.984 45.076 1.984L45.076 2.281M49.794 2.281L47.939 2.281L47.939 1.984Q48.212 1.984 48.380 1.937Q48.548 1.890 48.548 1.722L48.548-2.438Q48.548-2.653 48.486-2.748Q48.423-2.844 48.304-2.865Q48.185-2.887 47.939-2.887L47.939-3.184L49.162-3.270L49.162-0.567Q49.287-0.778 49.474-0.928Q49.662-1.078 49.888-1.162Q50.115-1.246 50.361-1.246Q51.529-1.246 51.529-0.168L51.529 1.722Q51.529 1.890 51.699 1.937Q51.869 1.984 52.138 1.984L52.138 2.281L50.283 2.281L50.283 1.984Q50.556 1.984 50.724 1.937Q50.892 1.890 50.892 1.722L50.892-0.153Q50.892-0.535 50.771-0.764Q50.650-0.992 50.298-0.992Q49.986-0.992 49.732-0.830Q49.478-0.668 49.332-0.399Q49.185-0.129 49.185 0.168L49.185 1.722Q49.185 1.890 49.355 1.937Q49.525 1.984 49.794 1.984L49.794 2.281M53.169-0.942Q53.169-0.969 53.201-1.016Q53.505-1.270 53.677-1.635Q53.849-2 53.849-2.399L53.849-2.473Q53.716-2.352 53.529-2.352Q53.337-2.352 53.201-2.485Q53.064-2.617 53.064-2.817Q53.064-3.012 53.201-3.141Q53.337-3.270 53.529-3.270Q53.736-3.270 53.865-3.133Q53.994-2.996 54.050-2.795Q54.107-2.594 54.107-2.399Q54.107-1.942 53.912-1.528Q53.716-1.114 53.377-0.824Q53.349-0.797 53.314-0.797Q53.263-0.797 53.216-0.844Q53.169-0.891 53.169-0.942\" fill=\"var(--tk-good)\" stroke=\"var(--tk-good)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\">\u003Cg transform=\"translate(50.562 -60.438)\">\u003Cpath d=\"M38.068 2.281L35.709 2.281L35.709 1.984Q36.033 1.984 36.275 1.937Q36.517 1.890 36.517 1.722L36.517-2.621Q36.517-2.793 36.275-2.840Q36.033-2.887 35.709-2.887L35.709-3.184L38.302-3.184Q38.634-3.184 39.021-3.098Q39.408-3.012 39.755-2.838Q40.103-2.664 40.322-2.383Q40.541-2.102 40.541-1.735Q40.541-1.410 40.339-1.145Q40.138-0.879 39.832-0.703Q39.525-0.528 39.197-0.438Q39.564-0.317 39.824-0.047Q40.084 0.222 40.134 0.586L40.228 1.281Q40.298 1.730 40.396 1.961Q40.494 2.191 40.791 2.191Q41.037 2.191 41.169 1.974Q41.302 1.758 41.302 1.496Q41.322 1.422 41.404 1.402L41.486 1.402Q41.580 1.426 41.580 1.519Q41.580 1.750 41.482 1.965Q41.384 2.179 41.207 2.314Q41.029 2.449 40.798 2.449Q40.201 2.449 39.783 2.162Q39.365 1.875 39.365 1.304L39.365 0.609Q39.365 0.328 39.212 0.113Q39.060-0.102 38.810-0.215Q38.560-0.328 38.287-0.328L37.259-0.328L37.259 1.722Q37.259 1.886 37.503 1.935Q37.748 1.984 38.068 1.984L38.068 2.281M37.259-2.621L37.259-0.582L38.189-0.582Q38.509-0.582 38.777-0.635Q39.044-0.688 39.244-0.815Q39.443-0.942 39.560-1.174Q39.677-1.406 39.677-1.735Q39.677-2.387 39.281-2.637Q38.884-2.887 38.189-2.887L37.662-2.887Q37.443-2.887 37.351-2.844Q37.259-2.801 37.259-2.621M41.841 0.527Q41.841 0.047 42.074-0.369Q42.306-0.785 42.716-1.035Q43.127-1.285 43.603-1.285Q44.334-1.285 44.732-0.844Q45.130-0.403 45.130 0.328Q45.130 0.433 45.037 0.457L42.587 0.457L42.587 0.527Q42.587 0.937 42.709 1.293Q42.830 1.648 43.101 1.865Q43.373 2.082 43.802 2.082Q44.166 2.082 44.462 1.853Q44.759 1.625 44.861 1.273Q44.869 1.226 44.955 1.211L45.037 1.211Q45.130 1.238 45.130 1.320Q45.130 1.328 45.123 1.359Q45.060 1.586 44.921 1.769Q44.783 1.953 44.591 2.086Q44.400 2.219 44.181 2.289Q43.962 2.359 43.724 2.359Q43.353 2.359 43.015 2.222Q42.677 2.086 42.410 1.834Q42.142 1.582 41.992 1.242Q41.841 0.902 41.841 0.527M42.595 0.219L44.556 0.219Q44.556-0.086 44.455-0.377Q44.353-0.668 44.136-0.850Q43.919-1.031 43.603-1.031Q43.302-1.031 43.072-0.844Q42.841-0.656 42.718-0.365Q42.595-0.074 42.595 0.219M50.099 2.281L45.716 2.281L45.716 1.984Q46.037 1.984 46.281 1.937Q46.525 1.890 46.525 1.722L46.525-2.621Q46.525-2.793 46.281-2.840Q46.037-2.887 45.716-2.887L45.716-3.184L48.302-3.184L48.302-2.887Q47.291-2.887 47.291-2.621L47.291 1.722Q47.291 1.894 47.382 1.939Q47.474 1.984 47.693 1.984L48.380 1.984Q48.853 1.984 49.162 1.859Q49.470 1.734 49.648 1.504Q49.826 1.273 49.918 0.947Q50.009 0.621 50.052 0.168L50.334 0.168L50.099 2.281M51.830 0.472L51.830-2.621Q51.830-2.793 51.586-2.840Q51.341-2.887 51.021-2.887L51.021-3.184L53.404-3.184L53.404-2.887Q53.084-2.887 52.839-2.838Q52.595-2.789 52.595-2.621L52.595 0.449Q52.595 1.172 52.939 1.662Q53.283 2.152 53.982 2.152Q54.447 2.152 54.818 1.922Q55.189 1.691 55.398 1.299Q55.607 0.906 55.607 0.449L55.607-2.406Q55.607-2.887 54.798-2.887L54.798-3.184L56.709-3.184L56.709-2.887Q55.900-2.887 55.900-2.406L55.900 0.472Q55.900 0.992 55.646 1.449Q55.392 1.906 54.951 2.177Q54.509 2.449 53.982 2.449Q53.572 2.449 53.191 2.306Q52.810 2.164 52.498 1.894Q52.185 1.625 52.007 1.258Q51.830 0.890 51.830 0.472M57.869-0.942Q57.869-0.969 57.900-1.016Q58.205-1.270 58.377-1.635Q58.548-2 58.548-2.399L58.548-2.473Q58.416-2.352 58.228-2.352Q58.037-2.352 57.900-2.485Q57.763-2.617 57.763-2.817Q57.763-3.012 57.900-3.141Q58.037-3.270 58.228-3.270Q58.435-3.270 58.564-3.133Q58.693-2.996 58.750-2.795Q58.806-2.594 58.806-2.399Q58.806-1.942 58.611-1.528Q58.416-1.114 58.076-0.824Q58.048-0.797 58.013-0.797Q57.962-0.797 57.916-0.844Q57.869-0.891 57.869-0.942\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fsvg>\u003Cfigcaption class=\"tikz-cap\">Derivatives on shared axes: the sigmoid derivative peaks at \u003Cspan class=\"katex\">\u003Cspan class=\"katex-html\" aria-hidden=\"true\">\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.6444em;\">\u003C\u002Fspan>\u003Cspan class=\"mord\">0.25\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>, tanh at \u003Cspan class=\"katex\">\u003Cspan class=\"katex-html\" aria-hidden=\"true\">\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.6444em;\">\u003C\u002Fspan>\u003Cspan class=\"mord\">1\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>, and the ReLU derivative is a unit step.\u003C\u002Ffigcaption>","\u003Csvg style=\"width:100%;max-width:596.912px;height:auto\" xmlns=\"http:\u002F\u002Fwww.w3.org\u002F2000\u002Fsvg\" viewBox=\"-75 -75 447.684 132.974\">\u003Cg stroke=\"currentColor\" style=\"stroke-miterlimit:10;stroke-width:.4\">\u003Cg style=\"stroke-width:.8\">\u003Cpath fill=\"none\" d=\"M-42.693 38.202h304.69\"\u002F>\u003Cpath stroke=\"none\" d=\"m264.597 38.202-4.16-2.08 1.56 2.08-1.56 2.08\"\u002F>\u003Cg stroke=\"none\" font-family=\"cmtt8\" font-size=\"8\">\u003Cg transform=\"translate(299.641 1.889)\">\u003Cpath d=\"M-30.863 37.964L-30.863 37.874Q-30.812 37.667-30.617 37.643L-29.511 37.643L-29.511 33.874L-30.617 33.874Q-30.812 33.850-30.863 33.636L-30.863 33.546Q-30.812 33.339-30.617 33.315L-29.121 33.315Q-28.929 33.339-28.871 33.546L-28.871 37.643L-27.769 37.643Q-27.570 37.667-27.519 37.874L-27.519 37.964Q-27.570 38.179-27.769 38.202L-30.617 38.202Q-30.812 38.179-30.863 37.964M-26.660 37.089Q-26.660 36.643-26.246 36.386Q-25.832 36.128-25.291 36.028Q-24.750 35.929-24.242 35.921Q-24.242 35.706-24.376 35.554Q-24.511 35.401-24.718 35.325Q-24.925 35.249-25.136 35.249Q-25.480 35.249-25.640 35.272L-25.640 35.331Q-25.640 35.499-25.759 35.614Q-25.878 35.729-26.042 35.729Q-26.218 35.729-26.333 35.606Q-26.449 35.483-26.449 35.315Q-26.449 34.909-26.068 34.800Q-25.687 34.690-25.128 34.690Q-24.859 34.690-24.591 34.768Q-24.324 34.847-24.099 34.997Q-23.875 35.147-23.738 35.368Q-23.601 35.589-23.601 35.866L-23.601 37.585Q-23.601 37.643-23.074 37.643Q-22.878 37.663-22.828 37.874L-22.828 37.964Q-22.878 38.179-23.074 38.202L-23.218 38.202Q-23.562 38.202-23.791 38.155Q-24.019 38.108-24.164 37.921Q-24.625 38.241-25.332 38.241Q-25.667 38.241-25.972 38.100Q-26.277 37.960-26.468 37.698Q-26.660 37.436-26.660 37.089M-26.019 37.097Q-26.019 37.370-25.777 37.526Q-25.535 37.682-25.250 37.682Q-25.031 37.682-24.798 37.624Q-24.566 37.565-24.404 37.427Q-24.242 37.288-24.242 37.065L-24.242 36.475Q-24.523 36.475-24.939 36.532Q-25.355 36.589-25.687 36.727Q-26.019 36.866-26.019 37.097M-22.476 39.323Q-22.476 39.214-22.425 39.122Q-22.375 39.030-22.283 38.979Q-22.191 38.929-22.085 38.929Q-21.976 38.929-21.884 38.979Q-21.792 39.030-21.742 39.122Q-21.691 39.214-21.691 39.323L-21.835 39.323Q-21.835 39.460-21.812 39.460Q-21.554 39.460-21.367 39.268Q-21.179 39.077-21.093 38.811L-20.882 38.202L-22.019 35.315L-22.347 35.315Q-22.542 35.292-22.597 35.073L-22.597 34.987Q-22.539 34.776-22.347 34.753L-21.187 34.753Q-20.992 34.776-20.941 34.987L-20.941 35.073Q-20.992 35.292-21.187 35.315L-21.453 35.315Q-21.117 36.163-20.873 36.817Q-20.628 37.472-20.628 37.561L-20.621 37.561Q-20.621 37.503-20.529 37.210Q-20.437 36.917-20.244 36.333Q-20.050 35.749-19.910 35.315L-20.187 35.315Q-20.398 35.292-20.437 35.073L-20.437 34.987Q-20.386 34.772-20.187 34.753L-19.035 34.753Q-18.828 34.776-18.789 34.987L-18.789 35.073Q-18.828 35.292-19.035 35.315L-19.355 35.315L-20.531 38.811Q-20.699 39.311-21.025 39.665Q-21.351 40.018-21.812 40.018Q-22.085 40.018-22.281 39.807Q-22.476 39.597-22.476 39.323M-15.062 36.714L-17.503 36.714Q-17.449 36.991-17.251 37.214Q-17.054 37.436-16.777 37.559Q-16.500 37.682-16.214 37.682Q-15.742 37.682-15.519 37.393Q-15.511 37.382-15.455 37.276Q-15.398 37.171-15.349 37.128Q-15.300 37.085-15.207 37.073L-15.062 37.073Q-14.871 37.093-14.812 37.307L-14.812 37.362Q-14.878 37.663-15.109 37.860Q-15.339 38.057-15.652 38.149Q-15.964 38.241-16.269 38.241Q-16.753 38.241-17.193 38.013Q-17.632 37.784-17.900 37.384Q-18.167 36.983-18.167 36.491L-18.167 36.432Q-18.167 35.964-17.921 35.561Q-17.675 35.159-17.267 34.925Q-16.859 34.690-16.390 34.690Q-15.886 34.690-15.533 34.913Q-15.179 35.136-14.996 35.524Q-14.812 35.913-14.812 36.417L-14.812 36.475Q-14.871 36.690-15.062 36.714M-17.496 36.163L-15.468 36.163Q-15.515 35.753-15.753 35.501Q-15.992 35.249-16.390 35.249Q-16.785 35.249-17.091 35.511Q-17.398 35.772-17.496 36.163M-14.105 37.964L-14.105 37.874Q-14.046 37.667-13.855 37.643L-13.144 37.643L-13.144 35.315L-13.855 35.315Q-14.050 35.292-14.105 35.073L-14.105 34.987Q-14.046 34.776-13.855 34.753L-12.753 34.753Q-12.554 34.772-12.503 34.987L-12.503 35.315Q-12.242 35.030-11.886 34.872Q-11.531 34.714-11.144 34.714Q-10.851 34.714-10.617 34.848Q-10.382 34.983-10.382 35.249Q-10.382 35.417-10.492 35.534Q-10.601 35.651-10.769 35.651Q-10.921 35.651-11.037 35.540Q-11.152 35.429-11.152 35.272Q-11.527 35.272-11.841 35.473Q-12.156 35.675-12.330 36.009Q-12.503 36.343-12.503 36.722L-12.503 37.643L-11.558 37.643Q-11.351 37.667-11.312 37.874L-11.312 37.964Q-11.351 38.179-11.558 38.202L-13.855 38.202Q-14.046 38.179-14.105 37.964\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(299.641 1.889)\">\u003Cpath d=\"M-2.698 38.843Q-3.229 38.534-3.629 38.046Q-4.030 37.557-4.241 36.964Q-4.452 36.370-4.452 35.753Q-4.452 35.136-4.243 34.548Q-4.034 33.960-3.637 33.479Q-3.241 32.999-2.706 32.675Q-2.635 32.651-2.604 32.651L-2.514 32.651Q-2.416 32.663-2.350 32.729Q-2.284 32.796-2.284 32.897Q-2.284 33.042-2.385 33.104Q-2.834 33.393-3.161 33.809Q-3.487 34.225-3.649 34.718Q-3.811 35.210-3.811 35.753Q-3.811 36.159-3.715 36.546Q-3.620 36.932-3.442 37.268Q-3.264 37.604-3.004 37.888Q-2.745 38.171-2.397 38.401Q-2.284 38.479-2.284 38.616Q-2.284 38.714-2.350 38.784Q-2.416 38.854-2.514 38.866L-2.604 38.866Q-2.663 38.866-2.698 38.843M0.556 38.241Q0.084 38.241-0.301 37.997Q-0.686 37.753-0.909 37.343Q-1.131 36.932-1.131 36.475Q-1.131 36.132-1.006 35.809Q-0.881 35.487-0.651 35.233Q-0.420 34.979-0.114 34.835Q0.193 34.690 0.556 34.690Q0.919 34.690 1.232 34.837Q1.544 34.983 1.767 35.229Q1.990 35.475 2.117 35.796Q2.244 36.116 2.244 36.475Q2.244 36.932 2.019 37.345Q1.794 37.757 1.410 37.999Q1.025 38.241 0.556 38.241M0.556 37.682Q1.021 37.682 1.312 37.288Q1.603 36.893 1.603 36.409Q1.603 36.116 1.468 35.848Q1.334 35.581 1.093 35.415Q0.853 35.249 0.556 35.249Q0.252 35.249 0.013 35.415Q-0.225 35.581-0.360 35.848Q-0.495 36.116-0.495 36.409Q-0.495 36.889-0.202 37.286Q0.091 37.682 0.556 37.682M3.423 37.347L3.423 35.315L3.002 35.315Q2.794 35.292 2.752 35.073L2.752 34.987Q2.798 34.776 3.002 34.753L3.818 34.753Q4.013 34.776 4.064 34.987L4.064 37.315Q4.064 37.550 4.234 37.616Q4.404 37.682 4.689 37.682Q4.896 37.682 5.091 37.606Q5.287 37.530 5.412 37.380Q5.537 37.229 5.537 37.018L5.537 35.315L5.115 35.315Q4.904 35.292 4.865 35.073L4.865 34.987Q4.904 34.776 5.115 34.753L5.927 34.753Q6.127 34.776 6.177 34.987L6.177 37.643L6.603 37.643Q6.810 37.667 6.849 37.874L6.849 37.964Q6.810 38.179 6.603 38.202L5.787 38.202Q5.587 38.179 5.537 37.979Q5.134 38.241 4.627 38.241Q4.392 38.241 4.177 38.200Q3.962 38.159 3.796 38.057Q3.630 37.956 3.527 37.778Q3.423 37.600 3.423 37.347M8.127 37.097L8.127 35.315L7.377 35.315Q7.177 35.292 7.127 35.073L7.127 34.987Q7.177 34.776 7.377 34.753L8.127 34.753L8.127 34.003Q8.177 33.796 8.377 33.768L8.521 33.768Q8.716 33.796 8.767 34.003L8.767 34.753L10.127 34.753Q10.318 34.772 10.377 34.987L10.377 35.073Q10.322 35.292 10.127 35.315L8.767 35.315L8.767 37.065Q8.767 37.682 9.341 37.682Q9.591 37.682 9.755 37.497Q9.919 37.311 9.919 37.065Q9.919 36.972 9.992 36.901Q10.064 36.831 10.166 36.819L10.310 36.819Q10.509 36.843 10.560 37.050L10.560 37.097Q10.560 37.421 10.377 37.684Q10.193 37.948 9.900 38.095Q9.607 38.241 9.287 38.241Q8.775 38.241 8.451 37.931Q8.127 37.620 8.127 37.097M11.244 39.745L11.244 39.659Q11.287 39.440 11.494 39.417L11.916 39.417L11.916 35.315L11.494 35.315Q11.287 35.292 11.244 35.073L11.244 34.987Q11.291 34.776 11.494 34.753L12.310 34.753Q12.505 34.772 12.556 34.987L12.556 35.057Q12.767 34.889 13.031 34.802Q13.294 34.714 13.564 34.714Q13.904 34.714 14.201 34.858Q14.498 35.003 14.712 35.255Q14.927 35.507 15.043 35.819Q15.158 36.132 15.158 36.475Q15.158 36.940 14.931 37.350Q14.705 37.761 14.318 38.001Q13.931 38.241 13.462 38.241Q12.943 38.241 12.556 37.874L12.556 39.417L12.982 39.417Q13.189 39.440 13.228 39.659L13.228 39.745Q13.189 39.956 12.982 39.979L11.494 39.979Q11.291 39.956 11.244 39.745M13.419 37.682Q13.728 37.682 13.978 37.513Q14.228 37.343 14.373 37.061Q14.517 36.780 14.517 36.475Q14.517 36.186 14.392 35.907Q14.267 35.628 14.035 35.450Q13.802 35.272 13.502 35.272Q13.181 35.272 12.919 35.458Q12.658 35.643 12.556 35.944L12.556 36.796Q12.646 37.163 12.867 37.423Q13.087 37.682 13.419 37.682M16.162 37.347L16.162 35.315L15.740 35.315Q15.533 35.292 15.490 35.073L15.490 34.987Q15.537 34.776 15.740 34.753L16.556 34.753Q16.752 34.776 16.802 34.987L16.802 37.315Q16.802 37.550 16.972 37.616Q17.142 37.682 17.427 37.682Q17.634 37.682 17.830 37.606Q18.025 37.530 18.150 37.380Q18.275 37.229 18.275 37.018L18.275 35.315L17.853 35.315Q17.642 35.292 17.603 35.073L17.603 34.987Q17.642 34.776 17.853 34.753L18.666 34.753Q18.865 34.776 18.916 34.987L18.916 37.643L19.341 37.643Q19.548 37.667 19.587 37.874L19.587 37.964Q19.548 38.179 19.341 38.202L18.525 38.202Q18.326 38.179 18.275 37.979Q17.873 38.241 17.365 38.241Q17.130 38.241 16.916 38.200Q16.701 38.159 16.535 38.057Q16.369 37.956 16.265 37.778Q16.162 37.600 16.162 37.347M20.865 37.097L20.865 35.315L20.115 35.315Q19.916 35.292 19.865 35.073L19.865 34.987Q19.916 34.776 20.115 34.753L20.865 34.753L20.865 34.003Q20.916 33.796 21.115 33.768L21.259 33.768Q21.455 33.796 21.505 34.003L21.505 34.753L22.865 34.753Q23.056 34.772 23.115 34.987L23.115 35.073Q23.060 35.292 22.865 35.315L21.505 35.315L21.505 37.065Q21.505 37.682 22.080 37.682Q22.330 37.682 22.494 37.497Q22.658 37.311 22.658 37.065Q22.658 36.972 22.730 36.901Q22.802 36.831 22.904 36.819L23.048 36.819Q23.248 36.843 23.298 37.050L23.298 37.097Q23.298 37.421 23.115 37.684Q22.931 37.948 22.638 38.095Q22.345 38.241 22.025 38.241Q21.513 38.241 21.189 37.931Q20.865 37.620 20.865 37.097\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(299.641 1.889)\">\u003Cpath d=\"M28.595 37.089Q28.595 36.643 29.009 36.386Q29.423 36.128 29.964 36.028Q30.505 35.929 31.013 35.921Q31.013 35.706 30.878 35.554Q30.744 35.401 30.537 35.325Q30.330 35.249 30.119 35.249Q29.775 35.249 29.615 35.272L29.615 35.331Q29.615 35.499 29.496 35.614Q29.377 35.729 29.212 35.729Q29.037 35.729 28.921 35.606Q28.806 35.483 28.806 35.315Q28.806 34.909 29.187 34.800Q29.568 34.690 30.127 34.690Q30.396 34.690 30.664 34.768Q30.931 34.847 31.156 34.997Q31.380 35.147 31.517 35.368Q31.654 35.589 31.654 35.866L31.654 37.585Q31.654 37.643 32.181 37.643Q32.377 37.663 32.427 37.874L32.427 37.964Q32.377 38.179 32.181 38.202L32.037 38.202Q31.693 38.202 31.464 38.155Q31.236 38.108 31.091 37.921Q30.630 38.241 29.923 38.241Q29.587 38.241 29.283 38.100Q28.978 37.960 28.787 37.698Q28.595 37.436 28.595 37.089M29.236 37.097Q29.236 37.370 29.478 37.526Q29.720 37.682 30.005 37.682Q30.224 37.682 30.457 37.624Q30.689 37.565 30.851 37.427Q31.013 37.288 31.013 37.065L31.013 36.475Q30.732 36.475 30.316 36.532Q29.900 36.589 29.568 36.727Q29.236 36.866 29.236 37.097M33.634 37.097L33.634 35.315L32.884 35.315Q32.685 35.292 32.634 35.073L32.634 34.987Q32.685 34.776 32.884 34.753L33.634 34.753L33.634 34.003Q33.685 33.796 33.884 33.768L34.029 33.768Q34.224 33.796 34.275 34.003L34.275 34.753L35.634 34.753Q35.826 34.772 35.884 34.987L35.884 35.073Q35.830 35.292 35.634 35.315L34.275 35.315L34.275 37.065Q34.275 37.682 34.849 37.682Q35.099 37.682 35.263 37.497Q35.427 37.311 35.427 37.065Q35.427 36.972 35.500 36.901Q35.572 36.831 35.673 36.819L35.818 36.819Q36.017 36.843 36.068 37.050L36.068 37.097Q36.068 37.421 35.884 37.684Q35.701 37.948 35.408 38.095Q35.115 38.241 34.794 38.241Q34.283 38.241 33.959 37.931Q33.634 37.620 33.634 37.097\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(299.641 1.889)\">\u003Cpath d=\"M41.162 37.964L41.162 37.874Q41.220 37.667 41.412 37.643L42.123 37.643L42.123 35.315L41.412 35.315Q41.216 35.292 41.162 35.073L41.162 34.987Q41.220 34.776 41.412 34.753L42.513 34.753Q42.712 34.772 42.763 34.987L42.763 35.315Q43.025 35.030 43.380 34.872Q43.736 34.714 44.123 34.714Q44.416 34.714 44.650 34.848Q44.884 34.983 44.884 35.249Q44.884 35.417 44.775 35.534Q44.666 35.651 44.498 35.651Q44.345 35.651 44.230 35.540Q44.115 35.429 44.115 35.272Q43.740 35.272 43.425 35.473Q43.111 35.675 42.937 36.009Q42.763 36.343 42.763 36.722L42.763 37.643L43.709 37.643Q43.916 37.667 43.955 37.874L43.955 37.964Q43.916 38.179 43.709 38.202L41.412 38.202Q41.220 38.179 41.162 37.964M45.802 37.964L45.802 37.874Q45.853 37.667 46.048 37.643L47.087 37.643L47.087 35.315L46.115 35.315Q45.916 35.292 45.865 35.073L45.865 34.987Q45.916 34.776 46.115 34.753L47.482 34.753Q47.677 34.772 47.728 34.987L47.728 37.643L48.642 37.643Q48.837 37.667 48.888 37.874L48.888 37.964Q48.837 38.179 48.642 38.202L46.048 38.202Q45.853 38.179 45.802 37.964M46.834 33.776L46.834 33.722Q46.834 33.550 46.970 33.429Q47.107 33.307 47.283 33.307Q47.455 33.307 47.591 33.429Q47.728 33.550 47.728 33.722L47.728 33.776Q47.728 33.952 47.591 34.073Q47.455 34.194 47.283 34.194Q47.107 34.194 46.970 34.073Q46.834 33.952 46.834 33.776M49.669 38.850Q49.669 38.550 49.818 38.288Q49.966 38.026 50.216 37.866Q50.033 37.612 50.033 37.292Q50.033 37.007 50.185 36.745Q49.935 36.421 49.935 36.003Q49.935 35.643 50.127 35.347Q50.318 35.050 50.636 34.882Q50.955 34.714 51.318 34.714Q51.525 34.714 51.728 34.774Q51.931 34.835 52.095 34.936Q52.478 34.675 52.951 34.675Q53.189 34.675 53.371 34.806Q53.552 34.936 53.552 35.163Q53.552 35.311 53.451 35.417Q53.349 35.522 53.193 35.522Q53.060 35.522 52.964 35.444Q52.869 35.366 52.837 35.241Q52.693 35.249 52.494 35.339Q52.697 35.651 52.697 36.003Q52.697 36.284 52.585 36.516Q52.474 36.749 52.279 36.927Q52.084 37.104 51.832 37.202Q51.580 37.300 51.318 37.300Q50.931 37.300 50.599 37.112Q50.587 37.112 50.578 37.184Q50.568 37.257 50.568 37.292Q50.568 37.413 50.634 37.520Q50.701 37.628 50.822 37.667Q50.837 37.663 50.851 37.661Q50.865 37.659 50.888 37.659Q50.904 37.659 50.935 37.667Q50.966 37.675 50.974 37.675L51.568 37.675Q52.349 37.675 52.890 37.917Q53.431 38.159 53.431 38.850Q53.431 39.151 53.252 39.378Q53.072 39.604 52.775 39.749Q52.478 39.893 52.158 39.960Q51.837 40.026 51.552 40.026Q51.162 40.026 50.720 39.903Q50.279 39.780 49.974 39.513Q49.669 39.245 49.669 38.850M50.209 38.843Q50.209 39.061 50.449 39.202Q50.689 39.343 51.013 39.409Q51.337 39.475 51.552 39.475Q51.767 39.475 52.091 39.409Q52.416 39.343 52.656 39.202Q52.896 39.061 52.896 38.843Q52.896 38.554 52.671 38.415Q52.447 38.276 52.166 38.243Q51.884 38.210 51.537 38.210L50.919 38.210Q50.736 38.210 50.574 38.290Q50.412 38.370 50.310 38.516Q50.209 38.663 50.209 38.843M51.318 36.745Q51.619 36.745 51.837 36.526Q52.056 36.307 52.056 36.003Q52.056 35.847 52.002 35.716Q51.947 35.585 51.841 35.479Q51.736 35.374 51.605 35.319Q51.474 35.264 51.318 35.264Q51.013 35.264 50.794 35.483Q50.576 35.702 50.576 36.003Q50.576 36.300 50.798 36.522Q51.021 36.745 51.318 36.745M53.748 37.964L53.748 37.874Q53.791 37.667 53.998 37.643L54.419 37.643L54.419 33.874L53.998 33.874Q53.791 33.850 53.748 33.636L53.748 33.546Q53.791 33.339 53.998 33.315L54.814 33.315Q55.009 33.339 55.060 33.546L55.060 35.097Q55.521 34.714 56.119 34.714Q56.466 34.714 56.705 34.854Q56.943 34.995 57.058 35.253Q57.173 35.511 57.173 35.866L57.173 37.643L57.599 37.643Q57.806 37.667 57.845 37.874L57.845 37.964Q57.806 38.179 57.599 38.202L56.205 38.202Q56.009 38.179 55.959 37.964L55.959 37.874Q56.009 37.663 56.205 37.643L56.533 37.643L56.533 35.897Q56.533 35.589 56.443 35.431Q56.353 35.272 56.060 35.272Q55.791 35.272 55.562 35.403Q55.334 35.534 55.197 35.763Q55.060 35.991 55.060 36.257L55.060 37.643L55.486 37.643Q55.693 37.667 55.732 37.874L55.732 37.964Q55.693 38.179 55.486 38.202L53.998 38.202Q53.791 38.179 53.748 37.964M59.123 37.097L59.123 35.315L58.373 35.315Q58.173 35.292 58.123 35.073L58.123 34.987Q58.173 34.776 58.373 34.753L59.123 34.753L59.123 34.003Q59.173 33.796 59.373 33.768L59.517 33.768Q59.712 33.796 59.763 34.003L59.763 34.753L61.123 34.753Q61.314 34.772 61.373 34.987L61.373 35.073Q61.318 35.292 61.123 35.315L59.763 35.315L59.763 37.065Q59.763 37.682 60.337 37.682Q60.587 37.682 60.752 37.497Q60.916 37.311 60.916 37.065Q60.916 36.972 60.988 36.901Q61.060 36.831 61.162 36.819L61.306 36.819Q61.505 36.843 61.556 37.050L61.556 37.097Q61.556 37.421 61.373 37.684Q61.189 37.948 60.896 38.095Q60.603 38.241 60.283 38.241Q59.771 38.241 59.447 37.931Q59.123 37.620 59.123 37.097M63.209 38.866L63.123 38.866Q63.017 38.854 62.949 38.782Q62.880 38.710 62.880 38.616Q62.880 38.475 62.986 38.409Q63.322 38.194 63.591 37.905Q63.861 37.616 64.044 37.268Q64.228 36.921 64.318 36.542Q64.408 36.163 64.408 35.753Q64.408 35.214 64.244 34.722Q64.080 34.229 63.761 33.815Q63.443 33.401 63.002 33.112Q62.880 33.030 62.880 32.897Q62.880 32.800 62.949 32.731Q63.017 32.663 63.123 32.651L63.209 32.651Q63.263 32.651 63.298 32.675Q63.685 32.897 64.025 33.245Q64.365 33.593 64.585 33.987Q64.806 34.382 64.927 34.827Q65.048 35.272 65.048 35.753Q65.048 36.237 64.927 36.688Q64.806 37.139 64.582 37.536Q64.357 37.932 64.033 38.266Q63.709 38.600 63.306 38.843Q63.236 38.866 63.209 38.866\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003Cg style=\"stroke-width:.8\">\u003Cpath fill=\"none\" d=\"M-31.312 41.047v-96.984\"\u002F>\u003Cpath stroke=\"none\" d=\"m-31.312-58.537-2.08 4.16 2.08-1.56 2.08 1.56\"\u002F>\u003Cg stroke=\"none\" font-family=\"cmtt8\" font-size=\"8\">\u003Cg transform=\"translate(-27.625 -102.25)\">\u003Cpath d=\"M-31.074 38.850Q-31.074 38.550-30.925 38.288Q-30.777 38.026-30.527 37.866Q-30.710 37.612-30.710 37.292Q-30.710 37.007-30.558 36.745Q-30.808 36.421-30.808 36.003Q-30.808 35.643-30.617 35.347Q-30.425 35.050-30.107 34.882Q-29.789 34.714-29.425 34.714Q-29.218 34.714-29.015 34.774Q-28.812 34.835-28.648 34.936Q-28.265 34.675-27.792 34.675Q-27.554 34.675-27.373 34.806Q-27.191 34.936-27.191 35.163Q-27.191 35.311-27.292 35.417Q-27.394 35.522-27.550 35.522Q-27.683 35.522-27.779 35.444Q-27.875 35.366-27.906 35.241Q-28.050 35.249-28.250 35.339Q-28.046 35.651-28.046 36.003Q-28.046 36.284-28.158 36.516Q-28.269 36.749-28.464 36.927Q-28.660 37.104-28.912 37.202Q-29.164 37.300-29.425 37.300Q-29.812 37.300-30.144 37.112Q-30.156 37.112-30.166 37.184Q-30.175 37.257-30.175 37.292Q-30.175 37.413-30.109 37.520Q-30.042 37.628-29.921 37.667Q-29.906 37.663-29.892 37.661Q-29.878 37.659-29.855 37.659Q-29.839 37.659-29.808 37.667Q-29.777 37.675-29.769 37.675L-29.175 37.675Q-28.394 37.675-27.853 37.917Q-27.312 38.159-27.312 38.850Q-27.312 39.151-27.492 39.378Q-27.671 39.604-27.968 39.749Q-28.265 39.893-28.585 39.960Q-28.906 40.026-29.191 40.026Q-29.582 40.026-30.023 39.903Q-30.464 39.780-30.769 39.513Q-31.074 39.245-31.074 38.850M-30.535 38.843Q-30.535 39.061-30.294 39.202Q-30.054 39.343-29.730 39.409Q-29.406 39.475-29.191 39.475Q-28.976 39.475-28.652 39.409Q-28.328 39.343-28.087 39.202Q-27.847 39.061-27.847 38.843Q-27.847 38.554-28.072 38.415Q-28.296 38.276-28.578 38.243Q-28.859 38.210-29.207 38.210L-29.824 38.210Q-30.007 38.210-30.169 38.290Q-30.332 38.370-30.433 38.516Q-30.535 38.663-30.535 38.843M-29.425 36.745Q-29.125 36.745-28.906 36.526Q-28.687 36.307-28.687 36.003Q-28.687 35.847-28.742 35.716Q-28.796 35.585-28.902 35.479Q-29.007 35.374-29.138 35.319Q-29.269 35.264-29.425 35.264Q-29.730 35.264-29.949 35.483Q-30.167 35.702-30.167 36.003Q-30.167 36.300-29.945 36.522Q-29.722 36.745-29.425 36.745M-26.843 37.964L-26.843 37.874Q-26.785 37.667-26.593 37.643L-25.882 37.643L-25.882 35.315L-26.593 35.315Q-26.789 35.292-26.843 35.073L-26.843 34.987Q-26.785 34.776-26.593 34.753L-25.492 34.753Q-25.292 34.772-25.242 34.987L-25.242 35.315Q-24.980 35.030-24.625 34.872Q-24.269 34.714-23.882 34.714Q-23.589 34.714-23.355 34.848Q-23.121 34.983-23.121 35.249Q-23.121 35.417-23.230 35.534Q-23.339 35.651-23.507 35.651Q-23.660 35.651-23.775 35.540Q-23.890 35.429-23.890 35.272Q-24.265 35.272-24.580 35.473Q-24.894 35.675-25.068 36.009Q-25.242 36.343-25.242 36.722L-25.242 37.643L-24.296 37.643Q-24.089 37.667-24.050 37.874L-24.050 37.964Q-24.089 38.179-24.296 38.202L-26.593 38.202Q-26.785 38.179-26.843 37.964M-22.414 37.089Q-22.414 36.643-22 36.386Q-21.585 36.128-21.044 36.028Q-20.503 35.929-19.996 35.921Q-19.996 35.706-20.130 35.554Q-20.265 35.401-20.472 35.325Q-20.679 35.249-20.890 35.249Q-21.234 35.249-21.394 35.272L-21.394 35.331Q-21.394 35.499-21.513 35.614Q-21.632 35.729-21.796 35.729Q-21.972 35.729-22.087 35.606Q-22.203 35.483-22.203 35.315Q-22.203 34.909-21.822 34.800Q-21.441 34.690-20.882 34.690Q-20.613 34.690-20.345 34.768Q-20.078 34.847-19.853 34.997Q-19.628 35.147-19.492 35.368Q-19.355 35.589-19.355 35.866L-19.355 37.585Q-19.355 37.643-18.828 37.643Q-18.632 37.663-18.582 37.874L-18.582 37.964Q-18.632 38.179-18.828 38.202L-18.972 38.202Q-19.316 38.202-19.544 38.155Q-19.773 38.108-19.917 37.921Q-20.378 38.241-21.085 38.241Q-21.421 38.241-21.726 38.100Q-22.031 37.960-22.222 37.698Q-22.414 37.436-22.414 37.089M-21.773 37.097Q-21.773 37.370-21.531 37.526Q-21.289 37.682-21.003 37.682Q-20.785 37.682-20.552 37.624Q-20.320 37.565-20.158 37.427Q-19.996 37.288-19.996 37.065L-19.996 36.475Q-20.277 36.475-20.693 36.532Q-21.109 36.589-21.441 36.727Q-21.773 36.866-21.773 37.097M-16.710 38.241Q-17.175 38.241-17.541 37.991Q-17.906 37.741-18.111 37.337Q-18.316 36.932-18.316 36.475Q-18.316 36.132-18.191 35.811Q-18.066 35.491-17.833 35.241Q-17.601 34.991-17.296 34.852Q-16.992 34.714-16.636 34.714Q-16.125 34.714-15.718 35.034L-15.718 33.874L-16.140 33.874Q-16.351 33.850-16.390 33.636L-16.390 33.546Q-16.351 33.339-16.140 33.315L-15.328 33.315Q-15.128 33.339-15.078 33.546L-15.078 37.643L-14.652 37.643Q-14.445 37.667-14.406 37.874L-14.406 37.964Q-14.445 38.179-14.652 38.202L-15.468 38.202Q-15.667 38.179-15.718 37.964L-15.718 37.835Q-15.914 38.026-16.175 38.134Q-16.437 38.241-16.710 38.241M-16.671 37.682Q-16.312 37.682-16.060 37.413Q-15.808 37.143-15.718 36.768L-15.718 35.936Q-15.777 35.749-15.904 35.598Q-16.031 35.448-16.208 35.360Q-16.386 35.272-16.582 35.272Q-16.890 35.272-17.140 35.442Q-17.390 35.612-17.535 35.897Q-17.679 36.182-17.679 36.483Q-17.679 36.932-17.394 37.307Q-17.109 37.682-16.671 37.682M-13.710 37.964L-13.710 37.874Q-13.660 37.667-13.464 37.643L-12.425 37.643L-12.425 35.315L-13.398 35.315Q-13.597 35.292-13.648 35.073L-13.648 34.987Q-13.597 34.776-13.398 34.753L-12.031 34.753Q-11.835 34.772-11.785 34.987L-11.785 37.643L-10.871 37.643Q-10.675 37.667-10.625 37.874L-10.625 37.964Q-10.675 38.179-10.871 38.202L-13.464 38.202Q-13.660 38.179-13.710 37.964M-12.679 33.776L-12.679 33.722Q-12.679 33.550-12.542 33.429Q-12.406 33.307-12.230 33.307Q-12.058 33.307-11.921 33.429Q-11.785 33.550-11.785 33.722L-11.785 33.776Q-11.785 33.952-11.921 34.073Q-12.058 34.194-12.230 34.194Q-12.406 34.194-12.542 34.073Q-12.679 33.952-12.679 33.776M-6.570 36.714L-9.011 36.714Q-8.957 36.991-8.759 37.214Q-8.562 37.436-8.285 37.559Q-8.007 37.682-7.722 37.682Q-7.250 37.682-7.027 37.393Q-7.019 37.382-6.962 37.276Q-6.906 37.171-6.857 37.128Q-6.808 37.085-6.714 37.073L-6.570 37.073Q-6.378 37.093-6.320 37.307L-6.320 37.362Q-6.386 37.663-6.617 37.860Q-6.847 38.057-7.160 38.149Q-7.472 38.241-7.777 38.241Q-8.261 38.241-8.701 38.013Q-9.140 37.784-9.408 37.384Q-9.675 36.983-9.675 36.491L-9.675 36.432Q-9.675 35.964-9.429 35.561Q-9.183 35.159-8.775 34.925Q-8.367 34.690-7.898 34.690Q-7.394 34.690-7.041 34.913Q-6.687 35.136-6.503 35.524Q-6.320 35.913-6.320 36.417L-6.320 36.475Q-6.378 36.690-6.570 36.714M-9.003 36.163L-6.976 36.163Q-7.023 35.753-7.261 35.501Q-7.500 35.249-7.898 35.249Q-8.292 35.249-8.599 35.511Q-8.906 35.772-9.003 36.163M-5.765 37.964L-5.765 37.874Q-5.722 37.667-5.515 37.643L-5.093 37.643L-5.093 35.315L-5.515 35.315Q-5.722 35.292-5.765 35.073L-5.765 34.987Q-5.718 34.776-5.515 34.753L-4.699 34.753Q-4.503 34.776-4.453 34.987L-4.453 35.073L-4.460 35.097Q-4.234 34.917-3.960 34.815Q-3.687 34.714-3.394 34.714Q-3.046 34.714-2.808 34.854Q-2.570 34.995-2.455 35.253Q-2.339 35.511-2.339 35.866L-2.339 37.643L-1.914 37.643Q-1.707 37.667-1.667 37.874L-1.667 37.964Q-1.707 38.179-1.914 38.202L-3.308 38.202Q-3.503 38.179-3.554 37.964L-3.554 37.874Q-3.503 37.663-3.308 37.643L-2.980 37.643L-2.980 35.897Q-2.980 35.589-3.070 35.431Q-3.160 35.272-3.453 35.272Q-3.722 35.272-3.951 35.403Q-4.179 35.534-4.316 35.763Q-4.453 35.991-4.453 36.257L-4.453 37.643L-4.027 37.643Q-3.820 37.667-3.781 37.874L-3.781 37.964Q-3.820 38.179-4.027 38.202L-5.515 38.202Q-5.722 38.179-5.765 37.964M-0.390 37.097L-0.390 35.315L-1.140 35.315Q-1.339 35.292-1.390 35.073L-1.390 34.987Q-1.339 34.776-1.140 34.753L-0.390 34.753L-0.390 34.003Q-0.339 33.796-0.140 33.768L0.004 33.768Q0.200 33.796 0.250 34.003L0.250 34.753L1.610 34.753Q1.801 34.772 1.860 34.987L1.860 35.073Q1.805 35.292 1.610 35.315L0.250 35.315L0.250 37.065Q0.250 37.682 0.825 37.682Q1.075 37.682 1.239 37.497Q1.403 37.311 1.403 37.065Q1.403 36.972 1.475 36.901Q1.547 36.831 1.649 36.819L1.793 36.819Q1.993 36.843 2.043 37.050L2.043 37.097Q2.043 37.421 1.860 37.684Q1.676 37.948 1.383 38.095Q1.090 38.241 0.770 38.241Q0.258 38.241-0.066 37.931Q-0.390 37.620-0.390 37.097\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(-27.625 -102.25)\">\u003Cpath d=\"M7.525 38.003L7.525 37.089Q7.552 36.882 7.763 36.858L7.931 36.858Q8.095 36.882 8.154 37.042Q8.357 37.682 9.084 37.682Q9.291 37.682 9.519 37.647Q9.748 37.612 9.916 37.497Q10.084 37.382 10.084 37.179Q10.084 36.968 9.861 36.854Q9.638 36.741 9.365 36.698L8.666 36.585Q7.525 36.374 7.525 35.651Q7.525 35.362 7.669 35.173Q7.814 34.983 8.054 34.876Q8.294 34.768 8.550 34.729Q8.806 34.690 9.084 34.690Q9.334 34.690 9.527 34.720Q9.720 34.749 9.884 34.827Q9.962 34.710 10.091 34.690L10.169 34.690Q10.267 34.702 10.330 34.764Q10.392 34.827 10.404 34.921L10.404 35.628Q10.392 35.722 10.330 35.788Q10.267 35.854 10.169 35.866L10.002 35.866Q9.908 35.854 9.841 35.788Q9.775 35.722 9.763 35.628Q9.763 35.249 9.068 35.249Q8.720 35.249 8.402 35.331Q8.084 35.413 8.084 35.659Q8.084 35.925 8.755 36.034L9.459 36.155Q9.943 36.237 10.293 36.485Q10.642 36.733 10.642 37.179Q10.642 37.569 10.406 37.811Q10.169 38.054 9.820 38.147Q9.470 38.241 9.084 38.241Q8.505 38.241 8.107 37.987Q8.037 38.112 7.988 38.169Q7.939 38.225 7.834 38.241L7.763 38.241Q7.548 38.218 7.525 38.003M11.802 37.964L11.802 37.874Q11.853 37.667 12.048 37.643L13.087 37.643L13.087 35.315L12.115 35.315Q11.916 35.292 11.865 35.073L11.865 34.987Q11.916 34.776 12.115 34.753L13.482 34.753Q13.677 34.772 13.728 34.987L13.728 37.643L14.642 37.643Q14.837 37.667 14.888 37.874L14.888 37.964Q14.837 38.179 14.642 38.202L12.048 38.202Q11.853 38.179 11.802 37.964M12.834 33.776L12.834 33.722Q12.834 33.550 12.970 33.429Q13.107 33.307 13.283 33.307Q13.455 33.307 13.591 33.429Q13.728 33.550 13.728 33.722L13.728 33.776Q13.728 33.952 13.591 34.073Q13.455 34.194 13.283 34.194Q13.107 34.194 12.970 34.073Q12.834 33.952 12.834 33.776M19.025 38.202L15.959 38.202Q15.853 38.202 15.775 38.120Q15.697 38.038 15.697 37.936L15.697 37.827Q15.697 37.710 15.783 37.636L18.279 35.315L16.470 35.315L16.470 35.608Q16.419 35.827 16.224 35.850L16.080 35.850Q15.888 35.831 15.830 35.608L15.830 34.987Q15.888 34.776 16.080 34.753L18.982 34.753Q19.091 34.753 19.169 34.831Q19.248 34.909 19.248 35.018L19.248 35.132Q19.248 35.245 19.158 35.323L16.662 37.643L18.630 37.643L18.630 37.249Q18.681 37.042 18.880 37.018L19.025 37.018Q19.220 37.042 19.271 37.249L19.271 37.964Q19.220 38.179 19.025 38.202M23.189 36.714L20.748 36.714Q20.802 36.991 21 37.214Q21.197 37.436 21.474 37.559Q21.752 37.682 22.037 37.682Q22.509 37.682 22.732 37.393Q22.740 37.382 22.796 37.276Q22.853 37.171 22.902 37.128Q22.951 37.085 23.044 37.073L23.189 37.073Q23.380 37.093 23.439 37.307L23.439 37.362Q23.373 37.663 23.142 37.860Q22.912 38.057 22.599 38.149Q22.287 38.241 21.982 38.241Q21.498 38.241 21.058 38.013Q20.619 37.784 20.351 37.384Q20.084 36.983 20.084 36.491L20.084 36.432Q20.084 35.964 20.330 35.561Q20.576 35.159 20.984 34.925Q21.392 34.690 21.861 34.690Q22.365 34.690 22.718 34.913Q23.072 35.136 23.255 35.524Q23.439 35.913 23.439 36.417L23.439 36.475Q23.380 36.690 23.189 36.714M20.755 36.163L22.783 36.163Q22.736 35.753 22.498 35.501Q22.259 35.249 21.861 35.249Q21.466 35.249 21.160 35.511Q20.853 35.772 20.755 36.163\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003Cpath fill=\"var(--tk-soft-warn)\" stroke=\"var(--tk-warn)\" d=\"M-19.93 38.202v-2.846H2.831v2.846ZM19.903 38.202v-4.553h22.763v4.553Zm22.763-4.553\" style=\"stroke-width:1.2\"\u002F>\u003Cpath fill=\"var(--tk-soft-accent)\" stroke=\"var(--tk-accent)\" d=\"M59.737 38.202v-8.536h22.762v8.536ZM99.571 38.202V21.13h22.762v17.072ZM139.405 38.202V1.213h22.762v36.989ZM179.238 38.202v-85.358h22.763v85.358Zm22.763-85.358\" style=\"stroke-width:1.2\"\u002F>\u003Cg transform=\"translate(14.196 11.808)\">\u003Cpath d=\"M-31.039 36.667Q-31.039 36.346-30.914 36.057Q-30.789 35.768-30.563 35.545Q-30.338 35.321-30.042 35.201Q-29.747 35.081-29.429 35.081Q-29.101 35.081-28.839 35.181Q-28.578 35.280-28.402 35.462Q-28.226 35.645-28.132 35.903Q-28.038 36.161-28.038 36.493Q-28.038 36.585-28.120 36.606L-30.375 36.606L-30.375 36.667Q-30.375 37.255-30.092 37.638Q-29.808 38.021-29.241 38.021Q-28.919 38.021-28.651 37.828Q-28.383 37.635-28.294 37.320Q-28.287 37.279-28.212 37.265L-28.120 37.265Q-28.038 37.289-28.038 37.361Q-28.038 37.368-28.044 37.395Q-28.157 37.792-28.528 38.031Q-28.899 38.270-29.323 38.270Q-29.760 38.270-30.160 38.062Q-30.560 37.853-30.799 37.486Q-31.039 37.119-31.039 36.667M-30.369 36.397L-28.554 36.397Q-28.554 36.120-28.651 35.868Q-28.749 35.615-28.947 35.459Q-29.145 35.304-29.429 35.304Q-29.706 35.304-29.919 35.462Q-30.133 35.621-30.251 35.876Q-30.369 36.131-30.369 36.397M-27.392 37.474Q-27.392 37.142-27.168 36.915Q-26.944 36.688-26.600 36.560Q-26.257 36.431-25.884 36.379Q-25.512 36.326-25.208 36.326L-25.208 36.073Q-25.208 35.868-25.315 35.688Q-25.423 35.509-25.604 35.406Q-25.785 35.304-25.994 35.304Q-26.400 35.304-26.636 35.396Q-26.547 35.433-26.501 35.517Q-26.455 35.601-26.455 35.703Q-26.455 35.799-26.501 35.878Q-26.547 35.956-26.628 36.001Q-26.708 36.045-26.797 36.045Q-26.947 36.045-27.048 35.948Q-27.149 35.850-27.149 35.703Q-27.149 35.081-25.994 35.081Q-25.782 35.081-25.532 35.145Q-25.283 35.208-25.081 35.327Q-24.879 35.447-24.753 35.632Q-24.626 35.816-24.626 36.059L-24.626 37.635Q-24.626 37.751-24.565 37.847Q-24.503 37.942-24.391 37.942Q-24.281 37.942-24.216 37.848Q-24.151 37.754-24.151 37.635L-24.151 37.187L-23.885 37.187L-23.885 37.635Q-23.885 37.905-24.112 38.070Q-24.339 38.236-24.620 38.236Q-24.828 38.236-24.965 38.082Q-25.102 37.929-25.125 37.713Q-25.272 37.980-25.554 38.125Q-25.836 38.270-26.161 38.270Q-26.438 38.270-26.722 38.195Q-27.005 38.120-27.198 37.941Q-27.392 37.761-27.392 37.474M-26.776 37.474Q-26.776 37.648-26.676 37.778Q-26.575 37.908-26.419 37.978Q-26.264 38.048-26.100 38.048Q-25.881 38.048-25.672 37.951Q-25.464 37.853-25.336 37.672Q-25.208 37.491-25.208 37.265L-25.208 36.537Q-25.532 36.537-25.898 36.628Q-26.264 36.719-26.520 36.931Q-26.776 37.142-26.776 37.474M-21.718 38.202L-23.454 38.202L-23.454 37.922Q-23.225 37.922-23.076 37.888Q-22.928 37.853-22.928 37.713L-22.928 35.864Q-22.928 35.594-23.035 35.533Q-23.143 35.471-23.454 35.471L-23.454 35.191L-22.425 35.116L-22.425 35.823Q-22.295 35.515-22.053 35.316Q-21.810 35.116-21.492 35.116Q-21.273 35.116-21.103 35.240Q-20.932 35.365-20.932 35.577Q-20.932 35.714-21.031 35.813Q-21.130 35.912-21.263 35.912Q-21.400 35.912-21.499 35.813Q-21.598 35.714-21.598 35.577Q-21.598 35.437-21.499 35.338Q-21.790 35.338-21.989 35.534Q-22.189 35.731-22.282 36.025Q-22.374 36.319-22.374 36.599L-22.374 37.713Q-22.374 37.922-21.718 37.922L-21.718 38.202M-18.679 38.202L-20.282 38.202L-20.282 37.922Q-20.057 37.922-19.908 37.888Q-19.759 37.853-19.759 37.713L-19.759 34.094Q-19.759 33.824-19.867 33.762Q-19.975 33.701-20.282 33.701L-20.282 33.420L-19.206 33.345L-19.206 37.713Q-19.206 37.850-19.055 37.886Q-18.905 37.922-18.679 37.922L-18.679 38.202M-17.750 39.337Q-17.620 39.405-17.483 39.405Q-17.312 39.405-17.162 39.316Q-17.011 39.227-16.900 39.082Q-16.789 38.937-16.710 38.769L-16.447 38.202L-17.616 35.676Q-17.691 35.529-17.821 35.497Q-17.951 35.464-18.184 35.464L-18.184 35.184L-16.663 35.184L-16.663 35.464Q-17.011 35.464-17.011 35.611Q-17.008 35.632-17.006 35.649Q-17.004 35.666-17.004 35.676L-16.146 37.535L-15.374 35.864Q-15.340 35.796-15.340 35.717Q-15.340 35.604-15.424 35.534Q-15.507 35.464-15.620 35.464L-15.620 35.184L-14.424 35.184L-14.424 35.464Q-14.643 35.464-14.815 35.568Q-14.988 35.673-15.080 35.864L-16.416 38.769Q-16.587 39.139-16.857 39.385Q-17.127 39.631-17.483 39.631Q-17.753 39.631-17.972 39.465Q-18.190 39.299-18.190 39.036Q-18.190 38.899-18.098 38.810Q-18.006 38.722-17.866 38.722Q-17.729 38.722-17.640 38.810Q-17.551 38.899-17.551 39.036Q-17.551 39.139-17.604 39.217Q-17.657 39.296-17.750 39.337\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.210\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(210.188 11.253)\">\u003Cpath d=\"M-31.039 36.719Q-31.039 36.377-30.904 36.078Q-30.769 35.779-30.529 35.555Q-30.290 35.331-29.972 35.206Q-29.654 35.081-29.323 35.081Q-28.878 35.081-28.479 35.297Q-28.079 35.512-27.844 35.890Q-27.610 36.267-27.610 36.719Q-27.610 37.060-27.752 37.344Q-27.894 37.628-28.138 37.835Q-28.383 38.041-28.692 38.156Q-29.001 38.270-29.323 38.270Q-29.753 38.270-30.155 38.069Q-30.557 37.867-30.798 37.515Q-31.039 37.163-31.039 36.719M-29.323 38.021Q-28.721 38.021-28.497 37.643Q-28.273 37.265-28.273 36.633Q-28.273 36.021-28.508 35.662Q-28.742 35.304-29.323 35.304Q-30.375 35.304-30.375 36.633Q-30.375 37.265-30.150 37.643Q-29.924 38.021-29.323 38.021M-26.441 37.368L-26.441 35.864Q-26.441 35.594-26.549 35.533Q-26.657 35.471-26.968 35.471L-26.968 35.191L-25.860 35.116L-25.860 37.348L-25.860 37.368Q-25.860 37.648-25.809 37.792Q-25.758 37.935-25.616 37.992Q-25.474 38.048-25.187 38.048Q-24.934 38.048-24.729 37.908Q-24.524 37.768-24.408 37.542Q-24.291 37.317-24.291 37.067L-24.291 35.864Q-24.291 35.594-24.399 35.533Q-24.507 35.471-24.818 35.471L-24.818 35.191L-23.710 35.116L-23.710 37.529Q-23.710 37.720-23.657 37.802Q-23.604 37.884-23.504 37.903Q-23.403 37.922-23.187 37.922L-23.187 38.202L-24.264 38.270L-24.264 37.706Q-24.374 37.888-24.519 38.011Q-24.664 38.134-24.850 38.202Q-25.037 38.270-25.238 38.270Q-26.441 38.270-26.441 37.368M-22.073 37.361L-22.073 35.464L-22.712 35.464L-22.712 35.242Q-22.395 35.242-22.177 35.032Q-21.960 34.822-21.860 34.512Q-21.759 34.203-21.759 33.895L-21.492 33.895L-21.492 35.184L-20.416 35.184L-20.416 35.464L-21.492 35.464L-21.492 37.348Q-21.492 37.624-21.388 37.823Q-21.284 38.021-21.024 38.021Q-20.867 38.021-20.761 37.917Q-20.655 37.812-20.605 37.659Q-20.556 37.505-20.556 37.348L-20.556 36.934L-20.289 36.934L-20.289 37.361Q-20.289 37.587-20.388 37.797Q-20.487 38.007-20.672 38.139Q-20.856 38.270-21.085 38.270Q-21.523 38.270-21.798 38.033Q-22.073 37.795-22.073 37.361M-17.835 39.559L-19.465 39.559L-19.465 39.279Q-19.236 39.279-19.088 39.244Q-18.939 39.210-18.939 39.070L-18.939 35.724Q-18.939 35.553-19.076 35.512Q-19.212 35.471-19.465 35.471L-19.465 35.191L-18.385 35.116L-18.385 35.522Q-18.163 35.321-17.876 35.218Q-17.589 35.116-17.281 35.116Q-16.854 35.116-16.490 35.329Q-16.126 35.543-15.912 35.907Q-15.699 36.271-15.699 36.691Q-15.699 37.136-15.938 37.500Q-16.177 37.864-16.570 38.067Q-16.963 38.270-17.408 38.270Q-17.674 38.270-17.922 38.170Q-18.170 38.069-18.358 37.888L-18.358 39.070Q-18.358 39.207-18.209 39.243Q-18.061 39.279-17.835 39.279L-17.835 39.559M-18.358 35.871L-18.358 37.481Q-18.225 37.734-17.982 37.891Q-17.739 38.048-17.462 38.048Q-17.134 38.048-16.881 37.847Q-16.628 37.645-16.495 37.327Q-16.362 37.009-16.362 36.691Q-16.362 36.462-16.427 36.233Q-16.492 36.004-16.620 35.806Q-16.748 35.608-16.943 35.488Q-17.138 35.369-17.370 35.369Q-17.664 35.369-17.932 35.498Q-18.201 35.628-18.358 35.871M-14.489 37.368L-14.489 35.864Q-14.489 35.594-14.596 35.533Q-14.704 35.471-15.015 35.471L-15.015 35.191L-13.908 35.116L-13.908 37.348L-13.908 37.368Q-13.908 37.648-13.856 37.792Q-13.805 37.935-13.663 37.992Q-13.521 38.048-13.234 38.048Q-12.981 38.048-12.776 37.908Q-12.571 37.768-12.455 37.542Q-12.339 37.317-12.339 37.067L-12.339 35.864Q-12.339 35.594-12.447 35.533Q-12.554 35.471-12.865 35.471L-12.865 35.191L-11.758 35.116L-11.758 37.529Q-11.758 37.720-11.705 37.802Q-11.652 37.884-11.551 37.903Q-11.450 37.922-11.235 37.922L-11.235 38.202L-12.312 38.270L-12.312 37.706Q-12.421 37.888-12.566 38.011Q-12.711 38.134-12.898 38.202Q-13.084 38.270-13.286 38.270Q-14.489 38.270-14.489 37.368M-10.121 37.361L-10.121 35.464L-10.760 35.464L-10.760 35.242Q-10.442 35.242-10.225 35.032Q-10.008 34.822-9.907 34.512Q-9.806 34.203-9.806 33.895L-9.540 33.895L-9.540 35.184L-8.463 35.184L-8.463 35.464L-9.540 35.464L-9.540 37.348Q-9.540 37.624-9.435 37.823Q-9.331 38.021-9.071 38.021Q-8.914 38.021-8.808 37.917Q-8.702 37.812-8.653 37.659Q-8.603 37.505-8.603 37.348L-8.603 36.934L-8.336 36.934L-8.336 37.361Q-8.336 37.587-8.436 37.797Q-8.535 38.007-8.719 38.139Q-8.904 38.270-9.133 38.270Q-9.570 38.270-9.845 38.033Q-10.121 37.795-10.121 37.361\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.210\"\u002F>\u003C\u002Fg>\u003Cg fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\">\u003Cg fill=\"var(--tk-warn)\" stroke=\"none\" font-family=\"cmtt8\" font-size=\"8\">\u003Cg transform=\"translate(60.438 -20.96)\">\u003Cpath d=\"M-27.800 36.714L-30.242 36.714Q-30.187 36.991-29.990 37.214Q-29.792 37.436-29.515 37.559Q-29.238 37.682-28.953 37.682Q-28.480 37.682-28.257 37.393Q-28.250 37.382-28.193 37.276Q-28.136 37.171-28.087 37.128Q-28.039 37.085-27.945 37.073L-27.800 37.073Q-27.609 37.093-27.550 37.307L-27.550 37.362Q-27.617 37.663-27.847 37.860Q-28.078 38.057-28.390 38.149Q-28.703 38.241-29.007 38.241Q-29.492 38.241-29.931 38.013Q-30.371 37.784-30.638 37.384Q-30.906 36.983-30.906 36.491L-30.906 36.432Q-30.906 35.964-30.660 35.561Q-30.414 35.159-30.005 34.925Q-29.597 34.690-29.128 34.690Q-28.625 34.690-28.271 34.913Q-27.917 35.136-27.734 35.524Q-27.550 35.913-27.550 36.417L-27.550 36.475Q-27.609 36.690-27.800 36.714M-30.234 36.163L-28.207 36.163Q-28.253 35.753-28.492 35.501Q-28.730 35.249-29.128 35.249Q-29.523 35.249-29.830 35.511Q-30.136 35.772-30.234 36.163M-26.660 37.089Q-26.660 36.643-26.246 36.386Q-25.832 36.128-25.291 36.028Q-24.750 35.929-24.242 35.921Q-24.242 35.706-24.376 35.554Q-24.511 35.401-24.718 35.325Q-24.925 35.249-25.136 35.249Q-25.480 35.249-25.640 35.272L-25.640 35.331Q-25.640 35.499-25.759 35.614Q-25.878 35.729-26.042 35.729Q-26.218 35.729-26.333 35.606Q-26.449 35.483-26.449 35.315Q-26.449 34.909-26.068 34.800Q-25.687 34.690-25.128 34.690Q-24.859 34.690-24.591 34.768Q-24.324 34.847-24.099 34.997Q-23.875 35.147-23.738 35.368Q-23.601 35.589-23.601 35.866L-23.601 37.585Q-23.601 37.643-23.074 37.643Q-22.878 37.663-22.828 37.874L-22.828 37.964Q-22.878 38.179-23.074 38.202L-23.218 38.202Q-23.562 38.202-23.791 38.155Q-24.019 38.108-24.164 37.921Q-24.625 38.241-25.332 38.241Q-25.667 38.241-25.972 38.100Q-26.277 37.960-26.468 37.698Q-26.660 37.436-26.660 37.089M-26.019 37.097Q-26.019 37.370-25.777 37.526Q-25.535 37.682-25.250 37.682Q-25.031 37.682-24.798 37.624Q-24.566 37.565-24.404 37.427Q-24.242 37.288-24.242 37.065L-24.242 36.475Q-24.523 36.475-24.939 36.532Q-25.355 36.589-25.687 36.727Q-26.019 36.866-26.019 37.097M-22.597 37.964L-22.597 37.874Q-22.539 37.667-22.347 37.643L-21.636 37.643L-21.636 35.315L-22.347 35.315Q-22.542 35.292-22.597 35.073L-22.597 34.987Q-22.539 34.776-22.347 34.753L-21.246 34.753Q-21.046 34.772-20.996 34.987L-20.996 35.315Q-20.734 35.030-20.378 34.872Q-20.023 34.714-19.636 34.714Q-19.343 34.714-19.109 34.848Q-18.875 34.983-18.875 35.249Q-18.875 35.417-18.984 35.534Q-19.093 35.651-19.261 35.651Q-19.414 35.651-19.529 35.540Q-19.644 35.429-19.644 35.272Q-20.019 35.272-20.333 35.473Q-20.648 35.675-20.822 36.009Q-20.996 36.343-20.996 36.722L-20.996 37.643L-20.050 37.643Q-19.843 37.667-19.804 37.874L-19.804 37.964Q-19.843 38.179-20.050 38.202L-22.347 38.202Q-22.539 38.179-22.597 37.964M-18.125 37.964L-18.125 37.874Q-18.074 37.667-17.878 37.643L-16.773 37.643L-16.773 33.874L-17.878 33.874Q-18.074 33.850-18.125 33.636L-18.125 33.546Q-18.074 33.339-17.878 33.315L-16.382 33.315Q-16.191 33.339-16.132 33.546L-16.132 37.643L-15.031 37.643Q-14.832 37.667-14.781 37.874L-14.781 37.964Q-14.832 38.179-15.031 38.202L-17.878 38.202Q-18.074 38.179-18.125 37.964M-13.984 39.323Q-13.984 39.214-13.933 39.122Q-13.882 39.030-13.791 38.979Q-13.699 38.929-13.593 38.929Q-13.484 38.929-13.392 38.979Q-13.300 39.030-13.250 39.122Q-13.199 39.214-13.199 39.323L-13.343 39.323Q-13.343 39.460-13.320 39.460Q-13.062 39.460-12.875 39.268Q-12.687 39.077-12.601 38.811L-12.390 38.202L-13.527 35.315L-13.855 35.315Q-14.050 35.292-14.105 35.073L-14.105 34.987Q-14.046 34.776-13.855 34.753L-12.695 34.753Q-12.500 34.776-12.449 34.987L-12.449 35.073Q-12.500 35.292-12.695 35.315L-12.960 35.315Q-12.625 36.163-12.380 36.817Q-12.136 37.472-12.136 37.561L-12.128 37.561Q-12.128 37.503-12.037 37.210Q-11.945 36.917-11.751 36.333Q-11.558 35.749-11.417 35.315L-11.695 35.315Q-11.906 35.292-11.945 35.073L-11.945 34.987Q-11.894 34.772-11.695 34.753L-10.542 34.753Q-10.335 34.776-10.296 34.987L-10.296 35.073Q-10.335 35.292-10.542 35.315L-10.863 35.315L-12.039 38.811Q-12.207 39.311-12.533 39.665Q-12.859 40.018-13.320 40.018Q-13.593 40.018-13.789 39.807Q-13.984 39.597-13.984 39.323\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(60.438 -20.96)\">\u003Cpath d=\"M-5.362 37.964L-5.362 37.874Q-5.311 37.667-5.116 37.643L-4.010 37.643L-4.010 33.874L-5.116 33.874Q-5.311 33.850-5.362 33.636L-5.362 33.546Q-5.311 33.339-5.116 33.315L-3.620 33.315Q-3.428 33.339-3.370 33.546L-3.370 37.643L-2.268 37.643Q-2.069 37.667-2.018 37.874L-2.018 37.964Q-2.069 38.179-2.268 38.202L-5.116 38.202Q-5.311 38.179-5.362 37.964M-1.159 37.089Q-1.159 36.643-0.745 36.386Q-0.331 36.128 0.210 36.028Q0.752 35.929 1.259 35.921Q1.259 35.706 1.125 35.554Q0.990 35.401 0.783 35.325Q0.576 35.249 0.365 35.249Q0.021 35.249-0.139 35.272L-0.139 35.331Q-0.139 35.499-0.258 35.614Q-0.377 35.729-0.541 35.729Q-0.717 35.729-0.832 35.606Q-0.948 35.483-0.948 35.315Q-0.948 34.909-0.567 34.800Q-0.186 34.690 0.373 34.690Q0.642 34.690 0.910 34.768Q1.177 34.847 1.402 34.997Q1.627 35.147 1.763 35.368Q1.900 35.589 1.900 35.866L1.900 37.585Q1.900 37.643 2.427 37.643Q2.623 37.663 2.673 37.874L2.673 37.964Q2.623 38.179 2.427 38.202L2.283 38.202Q1.939 38.202 1.710 38.155Q1.482 38.108 1.337 37.921Q0.877 38.241 0.169 38.241Q-0.166 38.241-0.471 38.100Q-0.776 37.960-0.967 37.698Q-1.159 37.436-1.159 37.089M-0.518 37.097Q-0.518 37.370-0.276 37.526Q-0.034 37.682 0.252 37.682Q0.470 37.682 0.703 37.624Q0.935 37.565 1.097 37.427Q1.259 37.288 1.259 37.065L1.259 36.475Q0.978 36.475 0.562 36.532Q0.146 36.589-0.186 36.727Q-0.518 36.866-0.518 37.097M3.025 39.323Q3.025 39.214 3.076 39.122Q3.127 39.030 3.218 38.979Q3.310 38.929 3.416 38.929Q3.525 38.929 3.617 38.979Q3.709 39.030 3.759 39.122Q3.810 39.214 3.810 39.323L3.666 39.323Q3.666 39.460 3.689 39.460Q3.947 39.460 4.134 39.268Q4.322 39.077 4.408 38.811L4.619 38.202L3.482 35.315L3.154 35.315Q2.959 35.292 2.904 35.073L2.904 34.987Q2.962 34.776 3.154 34.753L4.314 34.753Q4.509 34.776 4.560 34.987L4.560 35.073Q4.509 35.292 4.314 35.315L4.048 35.315Q4.384 36.163 4.628 36.817Q4.873 37.472 4.873 37.561L4.880 37.561Q4.880 37.503 4.972 37.210Q5.064 36.917 5.257 36.333Q5.451 35.749 5.591 35.315L5.314 35.315Q5.103 35.292 5.064 35.073L5.064 34.987Q5.115 34.772 5.314 34.753L6.466 34.753Q6.673 34.776 6.712 34.987L6.712 35.073Q6.673 35.292 6.466 35.315L6.146 35.315L4.970 38.811Q4.802 39.311 4.476 39.665Q4.150 40.018 3.689 40.018Q3.416 40.018 3.220 39.807Q3.025 39.597 3.025 39.323M10.439 36.714L7.998 36.714Q8.052 36.991 8.250 37.214Q8.447 37.436 8.724 37.559Q9.002 37.682 9.287 37.682Q9.759 37.682 9.982 37.393Q9.990 37.382 10.046 37.276Q10.103 37.171 10.152 37.128Q10.201 37.085 10.294 37.073L10.439 37.073Q10.630 37.093 10.689 37.307L10.689 37.362Q10.623 37.663 10.392 37.860Q10.162 38.057 9.849 38.149Q9.537 38.241 9.232 38.241Q8.748 38.241 8.308 38.013Q7.869 37.784 7.601 37.384Q7.334 36.983 7.334 36.491L7.334 36.432Q7.334 35.964 7.580 35.561Q7.826 35.159 8.234 34.925Q8.642 34.690 9.111 34.690Q9.615 34.690 9.968 34.913Q10.322 35.136 10.505 35.524Q10.689 35.913 10.689 36.417L10.689 36.475Q10.630 36.690 10.439 36.714M8.005 36.163L10.033 36.163Q9.986 35.753 9.748 35.501Q9.509 35.249 9.111 35.249Q8.716 35.249 8.410 35.511Q8.103 35.772 8.005 36.163M11.396 37.964L11.396 37.874Q11.455 37.667 11.646 37.643L12.357 37.643L12.357 35.315L11.646 35.315Q11.451 35.292 11.396 35.073L11.396 34.987Q11.455 34.776 11.646 34.753L12.748 34.753Q12.947 34.772 12.998 34.987L12.998 35.315Q13.259 35.030 13.615 34.872Q13.970 34.714 14.357 34.714Q14.650 34.714 14.884 34.848Q15.119 34.983 15.119 35.249Q15.119 35.417 15.009 35.534Q14.900 35.651 14.732 35.651Q14.580 35.651 14.464 35.540Q14.349 35.429 14.349 35.272Q13.974 35.272 13.660 35.473Q13.345 35.675 13.171 36.009Q12.998 36.343 12.998 36.722L12.998 37.643L13.943 37.643Q14.150 37.667 14.189 37.874L14.189 37.964Q14.150 38.179 13.943 38.202L11.646 38.202Q11.455 38.179 11.396 37.964M16.005 38.003L16.005 37.089Q16.033 36.882 16.244 36.858L16.412 36.858Q16.576 36.882 16.634 37.042Q16.837 37.682 17.564 37.682Q17.771 37.682 18 37.647Q18.228 37.612 18.396 37.497Q18.564 37.382 18.564 37.179Q18.564 36.968 18.341 36.854Q18.119 36.741 17.845 36.698L17.146 36.585Q16.005 36.374 16.005 35.651Q16.005 35.362 16.150 35.173Q16.294 34.983 16.535 34.876Q16.775 34.768 17.031 34.729Q17.287 34.690 17.564 34.690Q17.814 34.690 18.007 34.720Q18.201 34.749 18.365 34.827Q18.443 34.710 18.572 34.690L18.650 34.690Q18.748 34.702 18.810 34.764Q18.873 34.827 18.884 34.921L18.884 35.628Q18.873 35.722 18.810 35.788Q18.748 35.854 18.650 35.866L18.482 35.866Q18.388 35.854 18.322 35.788Q18.255 35.722 18.244 35.628Q18.244 35.249 17.548 35.249Q17.201 35.249 16.882 35.331Q16.564 35.413 16.564 35.659Q16.564 35.925 17.236 36.034L17.939 36.155Q18.423 36.237 18.773 36.485Q19.123 36.733 19.123 37.179Q19.123 37.569 18.886 37.811Q18.650 38.054 18.300 38.147Q17.951 38.241 17.564 38.241Q16.986 38.241 16.587 37.987Q16.517 38.112 16.468 38.169Q16.419 38.225 16.314 38.241L16.244 38.241Q16.029 38.218 16.005 38.003\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(60.438 -20.96)\">\u003Cpath d=\"M24.525 38.003L24.525 37.089Q24.552 36.882 24.763 36.858L24.931 36.858Q25.095 36.882 25.154 37.042Q25.357 37.682 26.084 37.682Q26.291 37.682 26.519 37.647Q26.748 37.612 26.916 37.497Q27.084 37.382 27.084 37.179Q27.084 36.968 26.861 36.854Q26.638 36.741 26.365 36.698L25.666 36.585Q24.525 36.374 24.525 35.651Q24.525 35.362 24.669 35.173Q24.814 34.983 25.054 34.876Q25.294 34.768 25.550 34.729Q25.806 34.690 26.084 34.690Q26.334 34.690 26.527 34.720Q26.720 34.749 26.884 34.827Q26.962 34.710 27.091 34.690L27.169 34.690Q27.267 34.702 27.330 34.764Q27.392 34.827 27.404 34.921L27.404 35.628Q27.392 35.722 27.330 35.788Q27.267 35.854 27.169 35.866L27.002 35.866Q26.908 35.854 26.841 35.788Q26.775 35.722 26.763 35.628Q26.763 35.249 26.068 35.249Q25.720 35.249 25.402 35.331Q25.084 35.413 25.084 35.659Q25.084 35.925 25.755 36.034L26.459 36.155Q26.943 36.237 27.293 36.485Q27.642 36.733 27.642 37.179Q27.642 37.569 27.406 37.811Q27.169 38.054 26.820 38.147Q26.470 38.241 26.084 38.241Q25.505 38.241 25.107 37.987Q25.037 38.112 24.988 38.169Q24.939 38.225 24.834 38.241L24.763 38.241Q24.548 38.218 24.525 38.003M29.384 37.097L29.384 35.315L28.634 35.315Q28.435 35.292 28.384 35.073L28.384 34.987Q28.435 34.776 28.634 34.753L29.384 34.753L29.384 34.003Q29.435 33.796 29.634 33.768L29.779 33.768Q29.974 33.796 30.025 34.003L30.025 34.753L31.384 34.753Q31.576 34.772 31.634 34.987L31.634 35.073Q31.580 35.292 31.384 35.315L30.025 35.315L30.025 37.065Q30.025 37.682 30.599 37.682Q30.849 37.682 31.013 37.497Q31.177 37.311 31.177 37.065Q31.177 36.972 31.250 36.901Q31.322 36.831 31.423 36.819L31.568 36.819Q31.767 36.843 31.818 37.050L31.818 37.097Q31.818 37.421 31.634 37.684Q31.451 37.948 31.158 38.095Q30.865 38.241 30.544 38.241Q30.033 38.241 29.709 37.931Q29.384 37.620 29.384 37.097M32.837 37.089Q32.837 36.643 33.252 36.386Q33.666 36.128 34.207 36.028Q34.748 35.929 35.255 35.921Q35.255 35.706 35.121 35.554Q34.986 35.401 34.779 35.325Q34.572 35.249 34.361 35.249Q34.017 35.249 33.857 35.272L33.857 35.331Q33.857 35.499 33.738 35.614Q33.619 35.729 33.455 35.729Q33.279 35.729 33.164 35.606Q33.048 35.483 33.048 35.315Q33.048 34.909 33.429 34.800Q33.810 34.690 34.369 34.690Q34.638 34.690 34.906 34.768Q35.173 34.847 35.398 34.997Q35.623 35.147 35.759 35.368Q35.896 35.589 35.896 35.866L35.896 37.585Q35.896 37.643 36.423 37.643Q36.619 37.663 36.669 37.874L36.669 37.964Q36.619 38.179 36.423 38.202L36.279 38.202Q35.935 38.202 35.707 38.155Q35.478 38.108 35.334 37.921Q34.873 38.241 34.166 38.241Q33.830 38.241 33.525 38.100Q33.220 37.960 33.029 37.698Q32.837 37.436 32.837 37.089M33.478 37.097Q33.478 37.370 33.720 37.526Q33.962 37.682 34.248 37.682Q34.466 37.682 34.699 37.624Q34.931 37.565 35.093 37.427Q35.255 37.288 35.255 37.065L35.255 36.475Q34.974 36.475 34.558 36.532Q34.142 36.589 33.810 36.727Q33.478 36.866 33.478 37.097M36.900 37.964L36.900 37.874Q36.959 37.667 37.150 37.643L37.861 37.643L37.861 35.315L37.150 35.315Q36.955 35.292 36.900 35.073L36.900 34.987Q36.959 34.776 37.150 34.753L38.252 34.753Q38.451 34.772 38.502 34.987L38.502 35.315Q38.763 35.030 39.119 34.872Q39.474 34.714 39.861 34.714Q40.154 34.714 40.388 34.848Q40.623 34.983 40.623 35.249Q40.623 35.417 40.513 35.534Q40.404 35.651 40.236 35.651Q40.084 35.651 39.968 35.540Q39.853 35.429 39.853 35.272Q39.478 35.272 39.164 35.473Q38.849 35.675 38.675 36.009Q38.502 36.343 38.502 36.722L38.502 37.643L39.447 37.643Q39.654 37.667 39.693 37.874L39.693 37.964Q39.654 38.179 39.447 38.202L37.150 38.202Q36.959 38.179 36.900 37.964M42.580 37.979L41.693 35.315L41.373 35.315Q41.173 35.292 41.123 35.073L41.123 34.987Q41.173 34.776 41.373 34.753L42.533 34.753Q42.728 34.772 42.779 34.987L42.779 35.073Q42.728 35.292 42.533 35.315L42.252 35.315L43.044 37.690L43.834 35.315L43.556 35.315Q43.357 35.292 43.306 35.073L43.306 34.987Q43.357 34.776 43.556 34.753L44.716 34.753Q44.912 34.776 44.962 34.987L44.962 35.073Q44.912 35.292 44.716 35.315L44.396 35.315L43.509 37.979Q43.466 38.093 43.365 38.167Q43.263 38.241 43.138 38.241L42.947 38.241Q42.830 38.241 42.726 38.169Q42.623 38.097 42.580 37.979M48.681 36.714L46.240 36.714Q46.294 36.991 46.492 37.214Q46.689 37.436 46.966 37.559Q47.244 37.682 47.529 37.682Q48.002 37.682 48.224 37.393Q48.232 37.382 48.289 37.276Q48.345 37.171 48.394 37.128Q48.443 37.085 48.537 37.073L48.681 37.073Q48.873 37.093 48.931 37.307L48.931 37.362Q48.865 37.663 48.634 37.860Q48.404 38.057 48.091 38.149Q47.779 38.241 47.474 38.241Q46.990 38.241 46.550 38.013Q46.111 37.784 45.843 37.384Q45.576 36.983 45.576 36.491L45.576 36.432Q45.576 35.964 45.822 35.561Q46.068 35.159 46.476 34.925Q46.884 34.690 47.353 34.690Q47.857 34.690 48.210 34.913Q48.564 35.136 48.748 35.524Q48.931 35.913 48.931 36.417L48.931 36.475Q48.873 36.690 48.681 36.714M46.248 36.163L48.275 36.163Q48.228 35.753 47.990 35.501Q47.752 35.249 47.353 35.249Q46.959 35.249 46.652 35.511Q46.345 35.772 46.248 36.163\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003Cg style=\"stroke-width:.8\">\u003Cpath fill=\"none\" d=\"M173.548-38.62C93.88-47.157 36.975-24.395 2.201-2.995\"\u002F>\u003Cpath stroke=\"none\" d=\"m-.013-1.632 4.632-.409-2.418-.953.238-2.59\"\u002F>\u003C\u002Fg>\u003Cg stroke=\"none\" font-family=\"cmtt8\" font-size=\"7\">\u003Cg transform=\"translate(76.199 -90.447)\">\u003Cpath d=\"M-28.239 36.900L-30.375 36.900Q-30.328 37.142-30.155 37.337Q-29.982 37.532-29.740 37.640Q-29.497 37.747-29.248 37.747Q-28.834 37.747-28.639 37.494Q-28.632 37.484-28.583 37.392Q-28.533 37.300-28.490 37.262Q-28.448 37.224-28.366 37.214L-28.239 37.214Q-28.072 37.231-28.020 37.419L-28.020 37.467Q-28.079 37.730-28.280 37.903Q-28.482 38.076-28.755 38.156Q-29.029 38.236-29.295 38.236Q-29.719 38.236-30.104 38.036Q-30.488 37.836-30.722 37.486Q-30.957 37.136-30.957 36.705L-30.957 36.654Q-30.957 36.244-30.741 35.891Q-30.526 35.539-30.169 35.334Q-29.812 35.129-29.401 35.129Q-28.960 35.129-28.651 35.324Q-28.342 35.519-28.181 35.859Q-28.020 36.199-28.020 36.640L-28.020 36.691Q-28.072 36.879-28.239 36.900M-30.369 36.418L-28.595 36.418Q-28.636 36.059-28.844 35.838Q-29.053 35.618-29.401 35.618Q-29.747 35.618-30.015 35.847Q-30.283 36.076-30.369 36.418M-27.241 37.228Q-27.241 36.838-26.879 36.613Q-26.517 36.387-26.043 36.300Q-25.570 36.213-25.125 36.206Q-25.125 36.018-25.243 35.885Q-25.361 35.751-25.542 35.685Q-25.724 35.618-25.908 35.618Q-26.209 35.618-26.349 35.639L-26.349 35.690Q-26.349 35.837-26.453 35.938Q-26.558 36.038-26.701 36.038Q-26.855 36.038-26.956 35.931Q-27.057 35.823-27.057 35.676Q-27.057 35.321-26.723 35.225Q-26.390 35.129-25.901 35.129Q-25.666 35.129-25.431 35.198Q-25.197 35.266-25.001 35.398Q-24.804 35.529-24.685 35.722Q-24.565 35.915-24.565 36.158L-24.565 37.662Q-24.565 37.713-24.104 37.713Q-23.933 37.730-23.888 37.915L-23.888 37.994Q-23.933 38.181-24.104 38.202L-24.230 38.202Q-24.531 38.202-24.731 38.161Q-24.931 38.120-25.057 37.956Q-25.460 38.236-26.079 38.236Q-26.373 38.236-26.640 38.113Q-26.906 37.990-27.074 37.761Q-27.241 37.532-27.241 37.228M-26.681 37.235Q-26.681 37.474-26.469 37.611Q-26.257 37.747-26.007 37.747Q-25.816 37.747-25.613 37.696Q-25.409 37.645-25.267 37.524Q-25.125 37.402-25.125 37.207L-25.125 36.691Q-25.372 36.691-25.736 36.741Q-26.100 36.790-26.390 36.912Q-26.681 37.033-26.681 37.235M-23.393 36.691Q-23.393 36.271-23.179 35.909Q-22.965 35.546-22.601 35.338Q-22.237 35.129-21.817 35.129Q-21.332 35.129-21 35.225Q-20.668 35.321-20.668 35.676Q-20.668 35.823-20.768 35.931Q-20.867 36.038-21.017 36.038Q-21.167 36.038-21.272 35.938Q-21.376 35.837-21.376 35.690L-21.376 35.639Q-21.516 35.618-21.810 35.618Q-22.097 35.618-22.331 35.767Q-22.565 35.915-22.699 36.163Q-22.832 36.411-22.832 36.691Q-22.832 36.972-22.682 37.218Q-22.531 37.464-22.282 37.606Q-22.032 37.747-21.745 37.747Q-21.472 37.747-21.361 37.657Q-21.250 37.566-21.147 37.400Q-21.044 37.235-20.942 37.221L-20.795 37.221Q-20.703 37.231-20.646 37.291Q-20.590 37.351-20.590 37.440Q-20.590 37.481-20.607 37.515Q-20.703 37.771-20.880 37.929Q-21.058 38.086-21.299 38.161Q-21.540 38.236-21.817 38.236Q-22.241 38.236-22.605 38.028Q-22.969 37.819-23.181 37.467Q-23.393 37.115-23.393 36.691M-20.104 37.994L-20.104 37.915Q-20.067 37.734-19.886 37.713L-19.517 37.713L-19.517 34.415L-19.886 34.415Q-20.067 34.394-20.104 34.206L-20.104 34.128Q-20.067 33.947-19.886 33.926L-19.171 33.926Q-19 33.947-18.956 34.128L-18.956 35.485Q-18.553 35.150-18.030 35.150Q-17.726 35.150-17.517 35.273Q-17.309 35.396-17.208 35.621Q-17.107 35.847-17.107 36.158L-17.107 37.713L-16.734 37.713Q-16.553 37.734-16.519 37.915L-16.519 37.994Q-16.553 38.181-16.734 38.202L-17.955 38.202Q-18.125 38.181-18.170 37.994L-18.170 37.915Q-18.125 37.730-17.955 37.713L-17.667 37.713L-17.667 36.185Q-17.667 35.915-17.746 35.777Q-17.825 35.639-18.081 35.639Q-18.317 35.639-18.517 35.753Q-18.717 35.868-18.836 36.067Q-18.956 36.267-18.956 36.500L-18.956 37.713L-18.583 37.713Q-18.402 37.734-18.368 37.915L-18.368 37.994Q-18.402 38.181-18.583 38.202L-19.886 38.202Q-20.067 38.181-20.104 37.994\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.210\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(76.199 -90.447)\">\u003Cpath d=\"M-12.205 38.028L-12.205 37.228Q-12.181 37.047-11.997 37.026L-11.850 37.026Q-11.706 37.047-11.655 37.187Q-11.477 37.747-10.842 37.747Q-10.660 37.747-10.460 37.717Q-10.260 37.686-10.114 37.585Q-9.967 37.484-9.967 37.306Q-9.967 37.122-10.161 37.023Q-10.356 36.924-10.595 36.886L-11.207 36.787Q-12.205 36.602-12.205 35.970Q-12.205 35.717-12.079 35.551Q-11.952 35.386-11.742 35.292Q-11.532 35.198-11.308 35.163Q-11.084 35.129-10.842 35.129Q-10.623 35.129-10.454 35.155Q-10.284 35.181-10.141 35.249Q-10.072 35.146-9.960 35.129L-9.891 35.129Q-9.806 35.139-9.751 35.194Q-9.697 35.249-9.686 35.331L-9.686 35.950Q-9.697 36.032-9.751 36.090Q-9.806 36.148-9.891 36.158L-10.038 36.158Q-10.120 36.148-10.178 36.090Q-10.237 36.032-10.247 35.950Q-10.247 35.618-10.855 35.618Q-11.159 35.618-11.438 35.690Q-11.717 35.762-11.717 35.977Q-11.717 36.209-11.129 36.305L-10.513 36.411Q-10.090 36.483-9.784 36.700Q-9.478 36.917-9.478 37.306Q-9.478 37.648-9.685 37.860Q-9.891 38.072-10.197 38.154Q-10.503 38.236-10.842 38.236Q-11.347 38.236-11.696 38.014Q-11.758 38.123-11.800 38.173Q-11.843 38.223-11.935 38.236L-11.997 38.236Q-12.185 38.216-12.205 38.028M-7.953 37.235L-7.953 35.676L-8.610 35.676Q-8.784 35.656-8.828 35.464L-8.828 35.389Q-8.784 35.204-8.610 35.184L-7.953 35.184L-7.953 34.528Q-7.909 34.347-7.735 34.323L-7.608 34.323Q-7.437 34.347-7.393 34.528L-7.393 35.184L-6.203 35.184Q-6.036 35.201-5.985 35.389L-5.985 35.464Q-6.032 35.656-6.203 35.676L-7.393 35.676L-7.393 37.207Q-7.393 37.747-6.890 37.747Q-6.672 37.747-6.528 37.585Q-6.385 37.423-6.385 37.207Q-6.385 37.125-6.321 37.064Q-6.258 37.002-6.169 36.992L-6.043 36.992Q-5.868 37.013-5.824 37.194L-5.824 37.235Q-5.824 37.518-5.985 37.749Q-6.145 37.980-6.402 38.108Q-6.658 38.236-6.938 38.236Q-7.386 38.236-7.670 37.964Q-7.953 37.693-7.953 37.235M-2.215 36.900L-4.351 36.900Q-4.303 37.142-4.130 37.337Q-3.958 37.532-3.715 37.640Q-3.472 37.747-3.223 37.747Q-2.809 37.747-2.614 37.494Q-2.608 37.484-2.558 37.392Q-2.509 37.300-2.466 37.262Q-2.423 37.224-2.341 37.214L-2.215 37.214Q-2.047 37.231-1.996 37.419L-1.996 37.467Q-2.054 37.730-2.256 37.903Q-2.457 38.076-2.731 38.156Q-3.004 38.236-3.271 38.236Q-3.695 38.236-4.079 38.036Q-4.464 37.836-4.698 37.486Q-4.932 37.136-4.932 36.705L-4.932 36.654Q-4.932 36.244-4.717 35.891Q-4.501 35.539-4.144 35.334Q-3.787 35.129-3.377 35.129Q-2.936 35.129-2.626 35.324Q-2.317 35.519-2.156 35.859Q-1.996 36.199-1.996 36.640L-1.996 36.691Q-2.047 36.879-2.215 36.900M-4.344 36.418L-2.570 36.418Q-2.611 36.059-2.820 35.838Q-3.028 35.618-3.377 35.618Q-3.722 35.618-3.990 35.847Q-4.259 36.076-4.344 36.418M-1.510 39.552L-1.510 39.477Q-1.473 39.285-1.292 39.265L-0.923 39.265L-0.923 35.676L-1.292 35.676Q-1.473 35.656-1.510 35.464L-1.510 35.389Q-1.469 35.204-1.292 35.184L-0.577 35.184Q-0.406 35.201-0.362 35.389L-0.362 35.451Q-0.177 35.304 0.053 35.227Q0.284 35.150 0.520 35.150Q0.817 35.150 1.077 35.276Q1.337 35.403 1.525 35.623Q1.713 35.844 1.813 36.117Q1.914 36.390 1.914 36.691Q1.914 37.098 1.716 37.457Q1.518 37.816 1.179 38.026Q0.841 38.236 0.431 38.236Q-0.024 38.236-0.362 37.915L-0.362 39.265L0.011 39.265Q0.192 39.285 0.226 39.477L0.226 39.552Q0.192 39.737 0.011 39.757L-1.292 39.757Q-1.469 39.737-1.510 39.552M0.393 37.747Q0.663 37.747 0.882 37.599Q1.101 37.450 1.227 37.204Q1.354 36.958 1.354 36.691Q1.354 36.438 1.244 36.194Q1.135 35.950 0.932 35.794Q0.728 35.639 0.465 35.639Q0.185 35.639-0.044 35.801Q-0.273 35.963-0.362 36.226L-0.362 36.972Q-0.283 37.293-0.090 37.520Q0.103 37.747 0.393 37.747\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.210\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(76.199 -90.447)\">\u003Cpath d=\"M6.389 38.028L6.389 37.228Q6.413 37.047 6.597 37.026L6.744 37.026Q6.888 37.047 6.939 37.187Q7.117 37.747 7.752 37.747Q7.934 37.747 8.134 37.717Q8.334 37.686 8.480 37.585Q8.627 37.484 8.627 37.306Q8.627 37.122 8.433 37.023Q8.238 36.924 7.999 36.886L7.387 36.787Q6.389 36.602 6.389 35.970Q6.389 35.717 6.515 35.551Q6.642 35.386 6.852 35.292Q7.062 35.198 7.286 35.163Q7.510 35.129 7.752 35.129Q7.971 35.129 8.140 35.155Q8.310 35.181 8.453 35.249Q8.522 35.146 8.634 35.129L8.703 35.129Q8.788 35.139 8.843 35.194Q8.897 35.249 8.908 35.331L8.908 35.950Q8.897 36.032 8.843 36.090Q8.788 36.148 8.703 36.158L8.556 36.158Q8.474 36.148 8.416 36.090Q8.357 36.032 8.347 35.950Q8.347 35.618 7.739 35.618Q7.435 35.618 7.156 35.690Q6.877 35.762 6.877 35.977Q6.877 36.209 7.465 36.305L8.081 36.411Q8.504 36.483 8.810 36.700Q9.116 36.917 9.116 37.306Q9.116 37.648 8.909 37.860Q8.703 38.072 8.397 38.154Q8.091 38.236 7.752 38.236Q7.247 38.236 6.898 38.014Q6.836 38.123 6.794 38.173Q6.751 38.223 6.659 38.236L6.597 38.236Q6.409 38.216 6.389 38.028M10.080 36.691Q10.080 36.271 10.294 35.909Q10.507 35.546 10.871 35.338Q11.235 35.129 11.656 35.129Q12.141 35.129 12.473 35.225Q12.804 35.321 12.804 35.676Q12.804 35.823 12.705 35.931Q12.606 36.038 12.456 36.038Q12.305 36.038 12.201 35.938Q12.097 35.837 12.097 35.690L12.097 35.639Q11.957 35.618 11.663 35.618Q11.376 35.618 11.141 35.767Q10.907 35.915 10.774 36.163Q10.641 36.411 10.641 36.691Q10.641 36.972 10.791 37.218Q10.941 37.464 11.191 37.606Q11.440 37.747 11.728 37.747Q12.001 37.747 12.112 37.657Q12.223 37.566 12.326 37.400Q12.428 37.235 12.531 37.221L12.678 37.221Q12.770 37.231 12.826 37.291Q12.883 37.351 12.883 37.440Q12.883 37.481 12.866 37.515Q12.770 37.771 12.592 37.929Q12.415 38.086 12.174 38.161Q11.933 38.236 11.656 38.236Q11.232 38.236 10.868 38.028Q10.504 37.819 10.292 37.467Q10.080 37.115 10.080 36.691M13.662 37.228Q13.662 36.838 14.024 36.613Q14.387 36.387 14.860 36.300Q15.334 36.213 15.778 36.206Q15.778 36.018 15.660 35.885Q15.542 35.751 15.361 35.685Q15.180 35.618 14.995 35.618Q14.694 35.618 14.554 35.639L14.554 35.690Q14.554 35.837 14.450 35.938Q14.346 36.038 14.202 36.038Q14.048 36.038 13.948 35.931Q13.847 35.823 13.847 35.676Q13.847 35.321 14.180 35.225Q14.513 35.129 15.002 35.129Q15.238 35.129 15.472 35.198Q15.706 35.266 15.903 35.398Q16.099 35.529 16.219 35.722Q16.338 35.915 16.338 36.158L16.338 37.662Q16.338 37.713 16.800 37.713Q16.971 37.730 17.015 37.915L17.015 37.994Q16.971 38.181 16.800 38.202L16.673 38.202Q16.373 38.202 16.173 38.161Q15.973 38.120 15.846 37.956Q15.443 38.236 14.824 38.236Q14.530 38.236 14.264 38.113Q13.997 37.990 13.830 37.761Q13.662 37.532 13.662 37.228M14.223 37.235Q14.223 37.474 14.435 37.611Q14.647 37.747 14.896 37.747Q15.087 37.747 15.291 37.696Q15.494 37.645 15.636 37.524Q15.778 37.402 15.778 37.207L15.778 36.691Q15.532 36.691 15.168 36.741Q14.804 36.790 14.513 36.912Q14.223 37.033 14.223 37.235M17.415 37.994L17.415 37.915Q17.459 37.734 17.630 37.713L18.598 37.713L18.598 34.415L17.630 34.415Q17.459 34.394 17.415 34.206L17.415 34.128Q17.459 33.947 17.630 33.926L18.939 33.926Q19.107 33.947 19.158 34.128L19.158 37.713L20.122 37.713Q20.296 37.734 20.341 37.915L20.341 37.994Q20.296 38.181 20.122 38.202L17.630 38.202Q17.459 38.181 17.415 37.994M23.810 36.900L21.674 36.900Q21.722 37.142 21.894 37.337Q22.067 37.532 22.310 37.640Q22.552 37.747 22.802 37.747Q23.215 37.747 23.410 37.494Q23.417 37.484 23.467 37.392Q23.516 37.300 23.559 37.262Q23.602 37.224 23.684 37.214L23.810 37.214Q23.978 37.231 24.029 37.419L24.029 37.467Q23.971 37.730 23.769 37.903Q23.567 38.076 23.294 38.156Q23.021 38.236 22.754 38.236Q22.330 38.236 21.946 38.036Q21.561 37.836 21.327 37.486Q21.093 37.136 21.093 36.705L21.093 36.654Q21.093 36.244 21.308 35.891Q21.523 35.539 21.881 35.334Q22.238 35.129 22.648 35.129Q23.089 35.129 23.398 35.324Q23.708 35.519 23.868 35.859Q24.029 36.199 24.029 36.640L24.029 36.691Q23.978 36.879 23.810 36.900M21.681 36.418L23.455 36.418Q23.414 36.059 23.205 35.838Q22.997 35.618 22.648 35.618Q22.303 35.618 22.034 35.847Q21.766 36.076 21.681 36.418M24.965 38.028L24.965 37.228Q24.989 37.047 25.174 37.026L25.321 37.026Q25.464 37.047 25.516 37.187Q25.693 37.747 26.329 37.747Q26.510 37.747 26.710 37.717Q26.910 37.686 27.057 37.585Q27.204 37.484 27.204 37.306Q27.204 37.122 27.009 37.023Q26.814 36.924 26.575 36.886L25.963 36.787Q24.965 36.602 24.965 35.970Q24.965 35.717 25.092 35.551Q25.218 35.386 25.428 35.292Q25.639 35.198 25.863 35.163Q26.086 35.129 26.329 35.129Q26.548 35.129 26.717 35.155Q26.886 35.181 27.030 35.249Q27.098 35.146 27.211 35.129L27.279 35.129Q27.365 35.139 27.419 35.194Q27.474 35.249 27.484 35.331L27.484 35.950Q27.474 36.032 27.419 36.090Q27.365 36.148 27.279 36.158L27.132 36.158Q27.050 36.148 26.992 36.090Q26.934 36.032 26.924 35.950Q26.924 35.618 26.315 35.618Q26.011 35.618 25.733 35.690Q25.454 35.762 25.454 35.977Q25.454 36.209 26.042 36.305L26.657 36.411Q27.081 36.483 27.387 36.700Q27.693 36.917 27.693 37.306Q27.693 37.648 27.486 37.860Q27.279 38.072 26.973 38.154Q26.668 38.236 26.329 38.236Q25.823 38.236 25.475 38.014Q25.413 38.123 25.370 38.173Q25.328 38.223 25.235 38.236L25.174 38.236Q24.986 38.216 24.965 38.028\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.210\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(76.199 -90.447)\">\u003Cpath d=\"M32.557 37.994L32.557 34.415L32.188 34.415Q32.007 34.394 31.970 34.206L31.970 34.128Q32.007 33.947 32.188 33.926L32.903 33.926Q33.074 33.947 33.118 34.128L33.118 35.451Q33.303 35.304 33.533 35.227Q33.764 35.150 34 35.150Q34.297 35.150 34.557 35.276Q34.817 35.403 35.005 35.623Q35.193 35.844 35.293 36.117Q35.394 36.390 35.394 36.691Q35.394 37.098 35.196 37.457Q34.998 37.816 34.659 38.026Q34.321 38.236 33.911 38.236Q33.456 38.236 33.118 37.915L33.118 37.994Q33.074 38.185 32.903 38.202L32.776 38.202Q32.609 38.181 32.557 37.994M33.873 37.747Q34.143 37.747 34.362 37.599Q34.581 37.450 34.707 37.204Q34.834 36.958 34.834 36.691Q34.834 36.438 34.724 36.194Q34.615 35.950 34.412 35.794Q34.208 35.639 33.945 35.639Q33.665 35.639 33.436 35.801Q33.207 35.963 33.118 36.226L33.118 36.972Q33.197 37.293 33.390 37.520Q33.583 37.747 33.873 37.747M35.924 39.183Q35.924 39.087 35.969 39.007Q36.013 38.927 36.093 38.882Q36.174 38.838 36.266 38.838Q36.362 38.838 36.442 38.882Q36.522 38.927 36.567 39.007Q36.611 39.087 36.611 39.183L36.485 39.183Q36.485 39.303 36.505 39.303Q36.731 39.303 36.895 39.135Q37.059 38.968 37.134 38.735L37.319 38.202L36.324 35.676L36.037 35.676Q35.866 35.656 35.818 35.464L35.818 35.389Q35.869 35.204 36.037 35.184L37.052 35.184Q37.223 35.204 37.267 35.389L37.267 35.464Q37.223 35.656 37.052 35.676L36.820 35.676Q37.114 36.418 37.327 36.990Q37.541 37.563 37.541 37.641L37.548 37.641Q37.548 37.590 37.628 37.334Q37.708 37.077 37.877 36.567Q38.047 36.056 38.170 35.676L37.927 35.676Q37.742 35.656 37.708 35.464L37.708 35.389Q37.753 35.201 37.927 35.184L38.935 35.184Q39.116 35.204 39.151 35.389L39.151 35.464Q39.116 35.656 38.935 35.676L38.655 35.676L37.626 38.735Q37.479 39.173 37.194 39.482Q36.908 39.791 36.505 39.791Q36.266 39.791 36.095 39.607Q35.924 39.422 35.924 39.183\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.210\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(76.199 -90.447)\">\u003Cpath d=\"M44.113 37.235L44.113 35.676L43.457 35.676Q43.283 35.656 43.238 35.464L43.238 35.389Q43.283 35.204 43.457 35.184L44.113 35.184L44.113 34.528Q44.158 34.347 44.332 34.323L44.459 34.323Q44.629 34.347 44.674 34.528L44.674 35.184L45.863 35.184Q46.031 35.201 46.082 35.389L46.082 35.464Q46.034 35.656 45.863 35.676L44.674 35.676L44.674 37.207Q44.674 37.747 45.176 37.747Q45.395 37.747 45.539 37.585Q45.682 37.423 45.682 37.207Q45.682 37.125 45.745 37.064Q45.809 37.002 45.897 36.992L46.024 36.992Q46.198 37.013 46.243 37.194L46.243 37.235Q46.243 37.518 46.082 37.749Q45.921 37.980 45.665 38.108Q45.409 38.236 45.128 38.236Q44.681 38.236 44.397 37.964Q44.113 37.693 44.113 37.235M46.841 37.994L46.841 37.915Q46.878 37.734 47.060 37.713L47.429 37.713L47.429 34.415L47.060 34.415Q46.878 34.394 46.841 34.206L46.841 34.128Q46.878 33.947 47.060 33.926L47.774 33.926Q47.945 33.947 47.989 34.128L47.989 35.485Q48.393 35.150 48.916 35.150Q49.220 35.150 49.428 35.273Q49.637 35.396 49.738 35.621Q49.838 35.847 49.838 36.158L49.838 37.713L50.211 37.713Q50.392 37.734 50.426 37.915L50.426 37.994Q50.392 38.181 50.211 38.202L48.991 38.202Q48.820 38.181 48.775 37.994L48.775 37.915Q48.820 37.730 48.991 37.713L49.278 37.713L49.278 36.185Q49.278 35.915 49.199 35.777Q49.121 35.639 48.864 35.639Q48.628 35.639 48.429 35.753Q48.229 35.868 48.109 36.067Q47.989 36.267 47.989 36.500L47.989 37.713L48.362 37.713Q48.543 37.734 48.577 37.915L48.577 37.994Q48.543 38.181 48.362 38.202L47.060 38.202Q46.878 38.181 46.841 37.994M53.567 36.900L51.431 36.900Q51.479 37.142 51.652 37.337Q51.824 37.532 52.067 37.640Q52.310 37.747 52.559 37.747Q52.973 37.747 53.168 37.494Q53.174 37.484 53.224 37.392Q53.273 37.300 53.316 37.262Q53.359 37.224 53.441 37.214L53.567 37.214Q53.735 37.231 53.786 37.419L53.786 37.467Q53.728 37.730 53.526 37.903Q53.325 38.076 53.051 38.156Q52.778 38.236 52.511 38.236Q52.087 38.236 51.703 38.036Q51.318 37.836 51.084 37.486Q50.850 37.136 50.850 36.705L50.850 36.654Q50.850 36.244 51.065 35.891Q51.281 35.539 51.638 35.334Q51.995 35.129 52.405 35.129Q52.846 35.129 53.156 35.324Q53.465 35.519 53.626 35.859Q53.786 36.199 53.786 36.640L53.786 36.691Q53.735 36.879 53.567 36.900M51.438 36.418L53.212 36.418Q53.171 36.059 52.962 35.838Q52.754 35.618 52.405 35.618Q52.060 35.618 51.792 35.847Q51.523 36.076 51.438 36.418\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.210\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(76.199 -90.447)\">\u003Cpath d=\"M59.570 38.236Q59.164 38.236 58.844 38.017Q58.524 37.799 58.345 37.445Q58.166 37.091 58.166 36.691Q58.166 36.390 58.275 36.110Q58.384 35.830 58.588 35.611Q58.791 35.392 59.058 35.271Q59.324 35.150 59.635 35.150Q60.083 35.150 60.439 35.430L60.439 34.415L60.069 34.415Q59.885 34.394 59.851 34.206L59.851 34.128Q59.885 33.947 60.069 33.926L60.780 33.926Q60.955 33.947 60.999 34.128L60.999 37.713L61.372 37.713Q61.553 37.734 61.587 37.915L61.587 37.994Q61.553 38.181 61.372 38.202L60.657 38.202Q60.483 38.181 60.439 37.994L60.439 37.881Q60.268 38.048 60.039 38.142Q59.810 38.236 59.570 38.236M59.605 37.747Q59.919 37.747 60.139 37.512Q60.360 37.276 60.439 36.948L60.439 36.220Q60.387 36.056 60.276 35.924Q60.165 35.792 60.010 35.715Q59.854 35.639 59.683 35.639Q59.413 35.639 59.194 35.787Q58.976 35.936 58.849 36.185Q58.723 36.435 58.723 36.698Q58.723 37.091 58.972 37.419Q59.222 37.747 59.605 37.747M64.728 36.900L62.592 36.900Q62.640 37.142 62.812 37.337Q62.985 37.532 63.228 37.640Q63.470 37.747 63.720 37.747Q64.133 37.747 64.328 37.494Q64.335 37.484 64.385 37.392Q64.434 37.300 64.477 37.262Q64.520 37.224 64.602 37.214L64.728 37.214Q64.896 37.231 64.947 37.419L64.947 37.467Q64.889 37.730 64.687 37.903Q64.485 38.076 64.212 38.156Q63.939 38.236 63.672 38.236Q63.248 38.236 62.864 38.036Q62.479 37.836 62.245 37.486Q62.011 37.136 62.011 36.705L62.011 36.654Q62.011 36.244 62.226 35.891Q62.441 35.539 62.799 35.334Q63.156 35.129 63.566 35.129Q64.007 35.129 64.316 35.324Q64.626 35.519 64.786 35.859Q64.947 36.199 64.947 36.640L64.947 36.691Q64.896 36.879 64.728 36.900M62.599 36.418L64.373 36.418Q64.332 36.059 64.123 35.838Q63.915 35.618 63.566 35.618Q63.221 35.618 62.952 35.847Q62.684 36.076 62.599 36.418M65.565 37.994L65.565 37.915Q65.617 37.734 65.784 37.713L66.406 37.713L66.406 35.676L65.784 35.676Q65.613 35.656 65.565 35.464L65.565 35.389Q65.617 35.204 65.784 35.184L66.748 35.184Q66.922 35.201 66.967 35.389L66.967 35.676Q67.196 35.427 67.507 35.288Q67.818 35.150 68.156 35.150Q68.413 35.150 68.618 35.268Q68.823 35.386 68.823 35.618Q68.823 35.765 68.727 35.868Q68.631 35.970 68.484 35.970Q68.351 35.970 68.250 35.873Q68.149 35.775 68.149 35.639Q67.821 35.639 67.546 35.815Q67.271 35.991 67.119 36.283Q66.967 36.575 66.967 36.907L66.967 37.713L67.794 37.713Q67.975 37.734 68.009 37.915L68.009 37.994Q67.975 38.181 67.794 38.202L65.784 38.202Q65.617 38.181 65.565 37.994M69.626 37.994L69.626 37.915Q69.670 37.734 69.841 37.713L70.751 37.713L70.751 35.676L69.899 35.676Q69.725 35.656 69.681 35.464L69.681 35.389Q69.725 35.204 69.899 35.184L71.096 35.184Q71.267 35.201 71.311 35.389L71.311 37.713L72.111 37.713Q72.282 37.734 72.326 37.915L72.326 37.994Q72.282 38.181 72.111 38.202L69.841 38.202Q69.670 38.181 69.626 37.994M70.528 34.329L70.528 34.282Q70.528 34.131 70.648 34.025Q70.768 33.919 70.921 33.919Q71.072 33.919 71.191 34.025Q71.311 34.131 71.311 34.282L71.311 34.329Q71.311 34.483 71.191 34.589Q71.072 34.695 70.921 34.695Q70.768 34.695 70.648 34.589Q70.528 34.483 70.528 34.329M74.251 38.007L73.475 35.676L73.194 35.676Q73.020 35.656 72.976 35.464L72.976 35.389Q73.020 35.204 73.194 35.184L74.210 35.184Q74.380 35.201 74.425 35.389L74.425 35.464Q74.380 35.656 74.210 35.676L73.963 35.676L74.657 37.754L75.348 35.676L75.105 35.676Q74.931 35.656 74.886 35.464L74.886 35.389Q74.931 35.204 75.105 35.184L76.120 35.184Q76.291 35.204 76.336 35.389L76.336 35.464Q76.291 35.656 76.120 35.676L75.840 35.676L75.064 38.007Q75.026 38.106 74.938 38.171Q74.849 38.236 74.739 38.236L74.572 38.236Q74.469 38.236 74.379 38.173Q74.288 38.110 74.251 38.007M76.872 37.228Q76.872 36.838 77.234 36.613Q77.597 36.387 78.070 36.300Q78.544 36.213 78.988 36.206Q78.988 36.018 78.870 35.885Q78.752 35.751 78.571 35.685Q78.390 35.618 78.205 35.618Q77.904 35.618 77.764 35.639L77.764 35.690Q77.764 35.837 77.660 35.938Q77.556 36.038 77.412 36.038Q77.258 36.038 77.158 35.931Q77.057 35.823 77.057 35.676Q77.057 35.321 77.390 35.225Q77.723 35.129 78.212 35.129Q78.448 35.129 78.682 35.198Q78.916 35.266 79.113 35.398Q79.309 35.529 79.429 35.722Q79.548 35.915 79.548 36.158L79.548 37.662Q79.548 37.713 80.010 37.713Q80.181 37.730 80.225 37.915L80.225 37.994Q80.181 38.181 80.010 38.202L79.883 38.202Q79.583 38.202 79.383 38.161Q79.183 38.120 79.056 37.956Q78.653 38.236 78.034 38.236Q77.740 38.236 77.474 38.113Q77.207 37.990 77.040 37.761Q76.872 37.532 76.872 37.228M77.433 37.235Q77.433 37.474 77.645 37.611Q77.857 37.747 78.106 37.747Q78.297 37.747 78.501 37.696Q78.704 37.645 78.846 37.524Q78.988 37.402 78.988 37.207L78.988 36.691Q78.742 36.691 78.378 36.741Q78.014 36.790 77.723 36.912Q77.433 37.033 77.433 37.235M81.281 37.235L81.281 35.676L80.625 35.676Q80.451 35.656 80.406 35.464L80.406 35.389Q80.451 35.204 80.625 35.184L81.281 35.184L81.281 34.528Q81.326 34.347 81.500 34.323L81.627 34.323Q81.797 34.347 81.842 34.528L81.842 35.184L83.031 35.184Q83.199 35.201 83.250 35.389L83.250 35.464Q83.202 35.656 83.031 35.676L81.842 35.676L81.842 37.207Q81.842 37.747 82.344 37.747Q82.563 37.747 82.707 37.585Q82.850 37.423 82.850 37.207Q82.850 37.125 82.913 37.064Q82.977 37.002 83.065 36.992L83.192 36.992Q83.366 37.013 83.411 37.194L83.411 37.235Q83.411 37.518 83.250 37.749Q83.089 37.980 82.833 38.108Q82.577 38.236 82.296 38.236Q81.849 38.236 81.565 37.964Q81.281 37.693 81.281 37.235M84.487 37.994L84.487 37.915Q84.532 37.734 84.703 37.713L85.612 37.713L85.612 35.676L84.761 35.676Q84.586 35.656 84.542 35.464L84.542 35.389Q84.586 35.204 84.761 35.184L85.957 35.184Q86.128 35.201 86.172 35.389L86.172 37.713L86.972 37.713Q87.143 37.734 87.188 37.915L87.188 37.994Q87.143 38.181 86.972 38.202L84.703 38.202Q84.532 38.181 84.487 37.994M85.390 34.329L85.390 34.282Q85.390 34.131 85.509 34.025Q85.629 33.919 85.783 33.919Q85.933 33.919 86.053 34.025Q86.172 34.131 86.172 34.282L86.172 34.329Q86.172 34.483 86.053 34.589Q85.933 34.695 85.783 34.695Q85.629 34.695 85.509 34.589Q85.390 34.483 85.390 34.329M89.112 38.007L88.336 35.676L88.056 35.676Q87.881 35.656 87.837 35.464L87.837 35.389Q87.881 35.204 88.056 35.184L89.071 35.184Q89.242 35.201 89.286 35.389L89.286 35.464Q89.242 35.656 89.071 35.676L88.825 35.676L89.519 37.754L90.209 35.676L89.966 35.676Q89.792 35.656 89.748 35.464L89.748 35.389Q89.792 35.204 89.966 35.184L90.982 35.184Q91.152 35.204 91.197 35.389L91.197 35.464Q91.152 35.656 90.982 35.676L90.701 35.676L89.925 38.007Q89.888 38.106 89.799 38.171Q89.710 38.236 89.601 38.236L89.433 38.236Q89.331 38.236 89.240 38.173Q89.149 38.110 89.112 38.007M94.451 36.900L92.315 36.900Q92.362 37.142 92.535 37.337Q92.708 37.532 92.950 37.640Q93.193 37.747 93.442 37.747Q93.856 37.747 94.051 37.494Q94.058 37.484 94.107 37.392Q94.157 37.300 94.200 37.262Q94.242 37.224 94.324 37.214L94.451 37.214Q94.618 37.231 94.669 37.419L94.669 37.467Q94.611 37.730 94.410 37.903Q94.208 38.076 93.935 38.156Q93.661 38.236 93.395 38.236Q92.971 38.236 92.586 38.036Q92.202 37.836 91.968 37.486Q91.733 37.136 91.733 36.705L91.733 36.654Q91.733 36.244 91.949 35.891Q92.164 35.539 92.521 35.334Q92.878 35.129 93.289 35.129Q93.730 35.129 94.039 35.324Q94.348 35.519 94.509 35.859Q94.669 36.199 94.669 36.640L94.669 36.691Q94.618 36.879 94.451 36.900M92.321 36.418L94.095 36.418Q94.054 36.059 93.846 35.838Q93.637 35.618 93.289 35.618Q92.943 35.618 92.675 35.847Q92.407 36.076 92.321 36.418\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.210\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fsvg>\u003Cfigcaption class=\"tikz-cap\">Gradient magnitude shrinking backward through a saturating stack: each layer multiplies by \u003Cspan class=\"katex\">\u003Cspan class=\"katex-html\" aria-hidden=\"true\">\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.9463em;vertical-align:-0.1944em;\">\u003C\u002Fspan>\u003Cspan class=\"mord\">\u003Cspan class=\"mord mathnormal\" style=\"margin-right:0.0359em;\">g\u003C\u002Fspan>\u003Cspan class=\"msupsub\">\u003Cspan class=\"vlist-t\">\u003Cspan class=\"vlist-r\">\u003Cspan class=\"vlist\" style=\"height:0.7519em;\">\u003Cspan style=\"top:-3.063em;margin-right:0.05em;\">\u003Cspan class=\"pstrut\" style=\"height:2.7em;\">\u003C\u002Fspan>\u003Cspan class=\"sizing reset-size6 size3 mtight\">\u003Cspan class=\"mord mtight\">\u003Cspan class=\"mord mtight\">′\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003Cspan class=\"mrel\">≤\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003C\u002Fspan>\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.6444em;\">\u003C\u002Fspan>\u003Cspan class=\"mord\">0.25\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>, so early layers receive almost nothing.\u003C\u002Ffigcaption>","\u003Csvg style=\"width:100%;max-width:346.705px;height:auto\" xmlns=\"http:\u002F\u002Fwww.w3.org\u002F2000\u002Fsvg\" viewBox=\"-75 -75 260.028 128.392\">\u003Cg stroke=\"currentColor\" style=\"stroke-miterlimit:10;stroke-width:.4\">\u003Cg style=\"stroke-width:.8\">\u003Cpath fill=\"none\" d=\"M-58.537 15.25H152.26\"\u002F>\u003Cpath stroke=\"none\" d=\"m154.859 15.25-4.16-2.08 1.56 2.08-1.56 2.08\"\u002F>\u003Cg transform=\"translate(111.853 1.905)\">\u003Cpath d=\"M48.836 15.249L47.058 15.249L47.058 14.952Q47.332 14.952 47.500 14.905Q47.668 14.858 47.668 14.690L47.668 12.554Q47.668 12.339 47.611 12.243Q47.554 12.147 47.441 12.126Q47.328 12.104 47.082 12.104L47.082 11.808L48.281 11.722L48.281 14.690Q48.281 14.858 48.427 14.905Q48.574 14.952 48.836 14.952L48.836 15.249M47.394 10.327Q47.394 10.136 47.529 10.005Q47.664 9.874 47.859 9.874Q47.980 9.874 48.084 9.937Q48.187 9.999 48.250 10.103Q48.312 10.206 48.312 10.327Q48.312 10.522 48.181 10.657Q48.050 10.792 47.859 10.792Q47.660 10.792 47.527 10.659Q47.394 10.526 47.394 10.327M51.265 15.249L49.410 15.249L49.410 14.952Q49.683 14.952 49.851 14.905Q50.019 14.858 50.019 14.690L50.019 12.554Q50.019 12.339 49.957 12.243Q49.894 12.147 49.775 12.126Q49.656 12.104 49.410 12.104L49.410 11.808L50.601 11.722L50.601 12.456Q50.715 12.241 50.908 12.073Q51.101 11.905 51.340 11.813Q51.578 11.722 51.832 11.722Q53 11.722 53 12.800L53 14.690Q53 14.858 53.170 14.905Q53.340 14.952 53.609 14.952L53.609 15.249L51.754 15.249L51.754 14.952Q52.027 14.952 52.195 14.905Q52.363 14.858 52.363 14.690L52.363 12.815Q52.363 12.433 52.242 12.204Q52.121 11.976 51.769 11.976Q51.457 11.976 51.203 12.138Q50.949 12.300 50.802 12.569Q50.656 12.839 50.656 13.136L50.656 14.690Q50.656 14.858 50.826 14.905Q50.996 14.952 51.265 14.952L51.265 15.249M55.937 16.800L54.082 16.800L54.082 16.507Q54.351 16.507 54.519 16.462Q54.687 16.417 54.687 16.241L54.687 12.417Q54.687 12.210 54.531 12.157Q54.375 12.104 54.082 12.104L54.082 11.808L55.304 11.722L55.304 12.187Q55.535 11.964 55.849 11.843Q56.164 11.722 56.504 11.722Q56.976 11.722 57.381 11.968Q57.785 12.214 58.017 12.630Q58.250 13.046 58.250 13.522Q58.250 13.897 58.101 14.226Q57.953 14.554 57.683 14.806Q57.414 15.058 57.070 15.192Q56.726 15.327 56.367 15.327Q56.078 15.327 55.806 15.206Q55.535 15.085 55.328 14.874L55.328 16.241Q55.328 16.417 55.496 16.462Q55.664 16.507 55.937 16.507L55.937 16.800M55.328 12.585L55.328 14.425Q55.480 14.714 55.742 14.894Q56.004 15.073 56.312 15.073Q56.597 15.073 56.820 14.935Q57.043 14.796 57.195 14.565Q57.347 14.335 57.425 14.063Q57.504 13.792 57.504 13.522Q57.504 13.190 57.379 12.833Q57.254 12.476 57.006 12.239Q56.758 12.003 56.410 12.003Q56.086 12.003 55.791 12.159Q55.496 12.315 55.328 12.585M59.457 14.296L59.457 12.554Q59.457 12.339 59.394 12.243Q59.332 12.147 59.213 12.126Q59.093 12.104 58.847 12.104L58.847 11.808L60.093 11.722L60.093 14.272L60.093 14.296Q60.093 14.608 60.148 14.770Q60.203 14.933 60.353 15.003Q60.504 15.073 60.824 15.073Q61.254 15.073 61.527 14.735Q61.800 14.397 61.800 13.952L61.800 12.554Q61.800 12.339 61.738 12.243Q61.675 12.147 61.556 12.126Q61.437 12.104 61.191 12.104L61.191 11.808L62.437 11.722L62.437 14.507Q62.437 14.718 62.500 14.813Q62.562 14.909 62.681 14.931Q62.800 14.952 63.047 14.952L63.047 15.249L61.824 15.327L61.824 14.706Q61.656 14.995 61.375 15.161Q61.093 15.327 60.773 15.327Q59.457 15.327 59.457 14.296M64.117 14.288L64.117 12.097L63.414 12.097L63.414 11.843Q63.769 11.843 64.011 11.610Q64.254 11.378 64.365 11.030Q64.476 10.683 64.476 10.327L64.758 10.327L64.758 11.800L65.933 11.800L65.933 12.097L64.758 12.097L64.758 14.272Q64.758 14.593 64.877 14.821Q64.996 15.050 65.277 15.050Q65.457 15.050 65.574 14.927Q65.691 14.804 65.744 14.624Q65.797 14.444 65.797 14.272L65.797 13.800L66.078 13.800L66.078 14.288Q66.078 14.542 65.972 14.782Q65.867 15.022 65.670 15.175Q65.472 15.327 65.215 15.327Q64.898 15.327 64.646 15.204Q64.394 15.081 64.256 14.847Q64.117 14.612 64.117 14.288\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg style=\"stroke-width:.8\">\u003Cpath fill=\"none\" d=\"M46.738 46.548V-56.128\"\u002F>\u003Cpath stroke=\"none\" d=\"m46.738-58.728-2.08 4.16 2.08-1.56 2.08 1.56\"\u002F>\u003Cg transform=\"translate(-12.514 -79.266)\">\u003Cpath d=\"M46.976 13.554Q46.976 13.050 47.232 12.618Q47.488 12.187 47.924 11.935Q48.359 11.683 48.859 11.683Q49.246 11.683 49.588 11.827Q49.929 11.972 50.191 12.233Q50.453 12.495 50.595 12.831Q50.738 13.167 50.738 13.554Q50.738 14.046 50.474 14.456Q50.211 14.866 49.781 15.097Q49.351 15.327 48.859 15.327Q48.367 15.327 47.933 15.095Q47.500 14.862 47.238 14.454Q46.976 14.046 46.976 13.554M48.859 15.050Q49.316 15.050 49.568 14.827Q49.820 14.604 49.908 14.253Q49.996 13.901 49.996 13.456Q49.996 13.026 49.902 12.688Q49.808 12.351 49.554 12.144Q49.300 11.937 48.859 11.937Q48.211 11.937 47.967 12.353Q47.722 12.769 47.722 13.456Q47.722 13.901 47.810 14.253Q47.898 14.604 48.150 14.827Q48.402 15.050 48.859 15.050M51.906 14.296L51.906 12.554Q51.906 12.339 51.843 12.243Q51.781 12.147 51.662 12.126Q51.543 12.104 51.297 12.104L51.297 11.808L52.543 11.722L52.543 14.272L52.543 14.296Q52.543 14.608 52.597 14.770Q52.652 14.933 52.802 15.003Q52.953 15.073 53.273 15.073Q53.703 15.073 53.976 14.735Q54.250 14.397 54.250 13.952L54.250 12.554Q54.250 12.339 54.187 12.243Q54.125 12.147 54.006 12.126Q53.886 12.104 53.640 12.104L53.640 11.808L54.886 11.722L54.886 14.507Q54.886 14.718 54.949 14.813Q55.011 14.909 55.131 14.931Q55.250 14.952 55.496 14.952L55.496 15.249L54.273 15.327L54.273 14.706Q54.105 14.995 53.824 15.161Q53.543 15.327 53.222 15.327Q51.906 15.327 51.906 14.296M56.566 14.288L56.566 12.097L55.863 12.097L55.863 11.843Q56.218 11.843 56.461 11.610Q56.703 11.378 56.814 11.030Q56.925 10.683 56.925 10.327L57.207 10.327L57.207 11.800L58.383 11.800L58.383 12.097L57.207 12.097L57.207 14.272Q57.207 14.593 57.326 14.821Q57.445 15.050 57.726 15.050Q57.906 15.050 58.023 14.927Q58.140 14.804 58.193 14.624Q58.246 14.444 58.246 14.272L58.246 13.800L58.527 13.800L58.527 14.288Q58.527 14.542 58.422 14.782Q58.316 15.022 58.119 15.175Q57.922 15.327 57.664 15.327Q57.347 15.327 57.095 15.204Q56.843 15.081 56.705 14.847Q56.566 14.612 56.566 14.288M61.129 16.800L59.273 16.800L59.273 16.507Q59.543 16.507 59.711 16.462Q59.879 16.417 59.879 16.241L59.879 12.417Q59.879 12.210 59.722 12.157Q59.566 12.104 59.273 12.104L59.273 11.808L60.496 11.722L60.496 12.187Q60.726 11.964 61.041 11.843Q61.355 11.722 61.695 11.722Q62.168 11.722 62.572 11.968Q62.976 12.214 63.209 12.630Q63.441 13.046 63.441 13.522Q63.441 13.897 63.293 14.226Q63.144 14.554 62.875 14.806Q62.605 15.058 62.261 15.192Q61.918 15.327 61.558 15.327Q61.269 15.327 60.998 15.206Q60.726 15.085 60.519 14.874L60.519 16.241Q60.519 16.417 60.687 16.462Q60.855 16.507 61.129 16.507L61.129 16.800M60.519 12.585L60.519 14.425Q60.672 14.714 60.933 14.894Q61.195 15.073 61.504 15.073Q61.789 15.073 62.011 14.935Q62.234 14.796 62.386 14.565Q62.539 14.335 62.617 14.063Q62.695 13.792 62.695 13.522Q62.695 13.190 62.570 12.833Q62.445 12.476 62.197 12.239Q61.949 12.003 61.601 12.003Q61.277 12.003 60.982 12.159Q60.687 12.315 60.519 12.585M64.648 14.296L64.648 12.554Q64.648 12.339 64.586 12.243Q64.523 12.147 64.404 12.126Q64.285 12.104 64.039 12.104L64.039 11.808L65.285 11.722L65.285 14.272L65.285 14.296Q65.285 14.608 65.340 14.770Q65.394 14.933 65.545 15.003Q65.695 15.073 66.015 15.073Q66.445 15.073 66.718 14.735Q66.992 14.397 66.992 13.952L66.992 12.554Q66.992 12.339 66.929 12.243Q66.867 12.147 66.748 12.126Q66.629 12.104 66.383 12.104L66.383 11.808L67.629 11.722L67.629 14.507Q67.629 14.718 67.691 14.813Q67.754 14.909 67.873 14.931Q67.992 14.952 68.238 14.952L68.238 15.249L67.015 15.327L67.015 14.706Q66.847 14.995 66.566 15.161Q66.285 15.327 65.965 15.327Q64.648 15.327 64.648 14.296M69.308 14.288L69.308 12.097L68.605 12.097L68.605 11.843Q68.961 11.843 69.203 11.610Q69.445 11.378 69.556 11.030Q69.668 10.683 69.668 10.327L69.949 10.327L69.949 11.800L71.125 11.800L71.125 12.097L69.949 12.097L69.949 14.272Q69.949 14.593 70.068 14.821Q70.187 15.050 70.468 15.050Q70.648 15.050 70.765 14.927Q70.883 14.804 70.935 14.624Q70.988 14.444 70.988 14.272L70.988 13.800L71.269 13.800L71.269 14.288Q71.269 14.542 71.164 14.782Q71.058 15.022 70.861 15.175Q70.664 15.327 70.406 15.327Q70.090 15.327 69.838 15.204Q69.586 15.081 69.447 14.847Q69.308 14.612 69.308 14.288\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cpath fill=\"var(--tk-soft-warn)\" stroke=\"none\" d=\"M-52.846 43.702V-54.46h99.584v98.162ZM46.738-54.46\"\u002F>\u003Cpath fill=\"none\" stroke=\"var(--tk-accent)\" d=\"M-52.846 15.25h99.584l59.751-59.751\" style=\"stroke-width:1.2\"\u002F>\u003Cg fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\">\u003Cg fill=\"var(--tk-warn)\" stroke=\"none\" font-family=\"cmtt8\" font-size=\"8\">\u003Cg transform=\"translate(-81.668 -57.948)\">\u003Cpath d=\"M47.058 15.011L47.058 14.921Q47.109 14.714 47.304 14.690L48.187 14.690L48.187 12.362L47.332 12.362Q47.133 12.339 47.082 12.120L47.082 12.034Q47.133 11.823 47.332 11.800L48.187 11.800L48.187 11.347Q48.187 10.882 48.593 10.601Q49 10.319 49.480 10.319Q49.793 10.319 50.037 10.440Q50.281 10.562 50.281 10.843Q50.281 11.007 50.172 11.124Q50.062 11.241 49.898 11.241Q49.750 11.241 49.629 11.136Q49.508 11.030 49.508 10.882L49.425 10.882Q49.273 10.882 49.136 10.942Q49 11.003 48.914 11.118Q48.828 11.233 48.828 11.378L48.828 11.800L49.859 11.800Q50.054 11.819 50.105 12.034L50.105 12.120Q50.054 12.339 49.859 12.362L48.828 12.362L48.828 14.690L49.707 14.690Q49.902 14.714 49.953 14.921L49.953 15.011Q49.902 15.226 49.707 15.249L47.304 15.249Q47.109 15.226 47.058 15.011M51.433 15.011L51.433 14.921Q51.484 14.714 51.679 14.690L52.785 14.690L52.785 10.921L51.679 10.921Q51.484 10.897 51.433 10.683L51.433 10.593Q51.484 10.386 51.679 10.362L53.175 10.362Q53.367 10.386 53.425 10.593L53.425 14.690L54.527 14.690Q54.726 14.714 54.777 14.921L54.777 15.011Q54.726 15.226 54.527 15.249L51.679 15.249Q51.484 15.226 51.433 15.011M55.636 14.136Q55.636 13.690 56.050 13.433Q56.465 13.175 57.006 13.075Q57.547 12.976 58.054 12.968Q58.054 12.753 57.920 12.601Q57.785 12.448 57.578 12.372Q57.371 12.296 57.160 12.296Q56.816 12.296 56.656 12.319L56.656 12.378Q56.656 12.546 56.537 12.661Q56.418 12.776 56.254 12.776Q56.078 12.776 55.963 12.653Q55.847 12.530 55.847 12.362Q55.847 11.956 56.228 11.847Q56.609 11.737 57.168 11.737Q57.437 11.737 57.705 11.815Q57.972 11.894 58.197 12.044Q58.422 12.194 58.558 12.415Q58.695 12.636 58.695 12.913L58.695 14.632Q58.695 14.690 59.222 14.690Q59.418 14.710 59.468 14.921L59.468 15.011Q59.418 15.226 59.222 15.249L59.078 15.249Q58.734 15.249 58.506 15.202Q58.277 15.155 58.133 14.968Q57.672 15.288 56.965 15.288Q56.629 15.288 56.324 15.147Q56.019 15.007 55.828 14.745Q55.636 14.483 55.636 14.136M56.277 14.144Q56.277 14.417 56.519 14.573Q56.761 14.729 57.047 14.729Q57.265 14.729 57.498 14.671Q57.730 14.612 57.892 14.474Q58.054 14.335 58.054 14.112L58.054 13.522Q57.773 13.522 57.357 13.579Q56.941 13.636 56.609 13.774Q56.277 13.913 56.277 14.144M60.675 14.144L60.675 12.362L59.925 12.362Q59.726 12.339 59.675 12.120L59.675 12.034Q59.726 11.823 59.925 11.800L60.675 11.800L60.675 11.050Q60.726 10.843 60.925 10.815L61.070 10.815Q61.265 10.843 61.316 11.050L61.316 11.800L62.675 11.800Q62.867 11.819 62.925 12.034L62.925 12.120Q62.871 12.339 62.675 12.362L61.316 12.362L61.316 14.112Q61.316 14.729 61.890 14.729Q62.140 14.729 62.304 14.544Q62.468 14.358 62.468 14.112Q62.468 14.019 62.541 13.948Q62.613 13.878 62.715 13.866L62.859 13.866Q63.058 13.890 63.109 14.097L63.109 14.144Q63.109 14.468 62.925 14.731Q62.742 14.995 62.449 15.142Q62.156 15.288 61.836 15.288Q61.324 15.288 61 14.978Q60.675 14.667 60.675 14.144M65.289 14.690Q65.289 14.468 65.455 14.302Q65.621 14.136 65.851 14.136Q66 14.136 66.127 14.214Q66.254 14.292 66.328 14.417Q66.402 14.542 66.402 14.690Q66.402 14.917 66.236 15.083Q66.070 15.249 65.851 15.249Q65.625 15.249 65.457 15.081Q65.289 14.913 65.289 14.690M65.289 12.354Q65.289 12.132 65.455 11.966Q65.621 11.800 65.851 11.800Q66 11.800 66.127 11.878Q66.254 11.956 66.328 12.081Q66.402 12.206 66.402 12.354Q66.402 12.581 66.236 12.747Q66.070 12.913 65.851 12.913Q65.625 12.913 65.457 12.745Q65.289 12.577 65.289 12.354\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(-81.668 -57.948)\">\u003Cpath d=\"M78.610 15.288Q78.137 15.288 77.753 15.044Q77.368 14.800 77.145 14.390Q76.923 13.979 76.923 13.522Q76.923 13.179 77.048 12.856Q77.173 12.534 77.403 12.280Q77.634 12.026 77.940 11.882Q78.247 11.737 78.610 11.737Q78.973 11.737 79.286 11.884Q79.598 12.030 79.821 12.276Q80.044 12.522 80.171 12.843Q80.298 13.163 80.298 13.522Q80.298 13.979 80.073 14.392Q79.848 14.804 79.464 15.046Q79.079 15.288 78.610 15.288M78.610 14.729Q79.075 14.729 79.366 14.335Q79.657 13.940 79.657 13.456Q79.657 13.163 79.522 12.895Q79.387 12.628 79.147 12.462Q78.907 12.296 78.610 12.296Q78.305 12.296 78.067 12.462Q77.829 12.628 77.694 12.895Q77.559 13.163 77.559 13.456Q77.559 13.937 77.852 14.333Q78.145 14.729 78.610 14.729M81.477 14.394L81.477 12.362L81.055 12.362Q80.848 12.339 80.805 12.120L80.805 12.034Q80.852 11.823 81.055 11.800L81.872 11.800Q82.067 11.823 82.118 12.034L82.118 14.362Q82.118 14.597 82.288 14.663Q82.458 14.729 82.743 14.729Q82.950 14.729 83.145 14.653Q83.341 14.577 83.466 14.427Q83.591 14.276 83.591 14.065L83.591 12.362L83.169 12.362Q82.958 12.339 82.919 12.120L82.919 12.034Q82.958 11.823 83.169 11.800L83.981 11.800Q84.180 11.823 84.231 12.034L84.231 14.690L84.657 14.690Q84.864 14.714 84.903 14.921L84.903 15.011Q84.864 15.226 84.657 15.249L83.841 15.249Q83.641 15.226 83.591 15.026Q83.188 15.288 82.680 15.288Q82.446 15.288 82.231 15.247Q82.016 15.206 81.850 15.104Q81.684 15.003 81.581 14.825Q81.477 14.647 81.477 14.394M86.180 14.144L86.180 12.362L85.430 12.362Q85.231 12.339 85.180 12.120L85.180 12.034Q85.231 11.823 85.430 11.800L86.180 11.800L86.180 11.050Q86.231 10.843 86.430 10.815L86.575 10.815Q86.770 10.843 86.821 11.050L86.821 11.800L88.180 11.800Q88.372 11.819 88.430 12.034L88.430 12.120Q88.376 12.339 88.180 12.362L86.821 12.362L86.821 14.112Q86.821 14.729 87.395 14.729Q87.645 14.729 87.809 14.544Q87.973 14.358 87.973 14.112Q87.973 14.019 88.046 13.948Q88.118 13.878 88.219 13.866L88.364 13.866Q88.563 13.890 88.614 14.097L88.614 14.144Q88.614 14.468 88.430 14.731Q88.247 14.995 87.954 15.142Q87.661 15.288 87.341 15.288Q86.829 15.288 86.505 14.978Q86.180 14.667 86.180 14.144M89.298 16.792L89.298 16.706Q89.341 16.487 89.548 16.464L89.969 16.464L89.969 12.362L89.548 12.362Q89.341 12.339 89.298 12.120L89.298 12.034Q89.344 11.823 89.548 11.800L90.364 11.800Q90.559 11.819 90.610 12.034L90.610 12.104Q90.821 11.937 91.085 11.849Q91.348 11.761 91.618 11.761Q91.958 11.761 92.255 11.905Q92.552 12.050 92.766 12.302Q92.981 12.554 93.096 12.866Q93.212 13.179 93.212 13.522Q93.212 13.987 92.985 14.397Q92.759 14.808 92.372 15.048Q91.985 15.288 91.516 15.288Q90.997 15.288 90.610 14.921L90.610 16.464L91.036 16.464Q91.243 16.487 91.282 16.706L91.282 16.792Q91.243 17.003 91.036 17.026L89.548 17.026Q89.344 17.003 89.298 16.792M91.473 14.729Q91.782 14.729 92.032 14.560Q92.282 14.390 92.427 14.108Q92.571 13.827 92.571 13.522Q92.571 13.233 92.446 12.954Q92.321 12.675 92.089 12.497Q91.856 12.319 91.555 12.319Q91.235 12.319 90.973 12.505Q90.712 12.690 90.610 12.991L90.610 13.843Q90.700 14.210 90.921 14.470Q91.141 14.729 91.473 14.729M94.216 14.394L94.216 12.362L93.794 12.362Q93.587 12.339 93.544 12.120L93.544 12.034Q93.591 11.823 93.794 11.800L94.610 11.800Q94.805 11.823 94.856 12.034L94.856 14.362Q94.856 14.597 95.026 14.663Q95.196 14.729 95.481 14.729Q95.688 14.729 95.884 14.653Q96.079 14.577 96.204 14.427Q96.329 14.276 96.329 14.065L96.329 12.362L95.907 12.362Q95.696 12.339 95.657 12.120L95.657 12.034Q95.696 11.823 95.907 11.800L96.719 11.800Q96.919 11.823 96.969 12.034L96.969 14.690L97.395 14.690Q97.602 14.714 97.641 14.921L97.641 15.011Q97.602 15.226 97.395 15.249L96.579 15.249Q96.380 15.226 96.329 15.026Q95.927 15.288 95.419 15.288Q95.184 15.288 94.969 15.247Q94.755 15.206 94.589 15.104Q94.423 15.003 94.319 14.825Q94.216 14.647 94.216 14.394M98.919 14.144L98.919 12.362L98.169 12.362Q97.969 12.339 97.919 12.120L97.919 12.034Q97.969 11.823 98.169 11.800L98.919 11.800L98.919 11.050Q98.969 10.843 99.169 10.815L99.313 10.815Q99.509 10.843 99.559 11.050L99.559 11.800L100.919 11.800Q101.110 11.819 101.169 12.034L101.169 12.120Q101.114 12.339 100.919 12.362L99.559 12.362L99.559 14.112Q99.559 14.729 100.134 14.729Q100.384 14.729 100.548 14.544Q100.712 14.358 100.712 14.112Q100.712 14.019 100.784 13.948Q100.856 13.878 100.958 13.866L101.102 13.866Q101.302 13.890 101.352 14.097L101.352 14.144Q101.352 14.468 101.169 14.731Q100.985 14.995 100.692 15.142Q100.399 15.288 100.079 15.288Q99.567 15.288 99.243 14.978Q98.919 14.667 98.919 14.144\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(-81.668 -57.948)\">\u003Cpath d=\"M108.360 15.327Q107.927 15.327 107.594 15.089Q107.262 14.851 107.042 14.462Q106.821 14.073 106.714 13.636Q106.606 13.198 106.606 12.800Q106.606 12.409 106.716 11.966Q106.825 11.522 107.042 11.142Q107.259 10.761 107.593 10.520Q107.927 10.280 108.360 10.280Q108.915 10.280 109.315 10.679Q109.716 11.077 109.913 11.663Q110.110 12.249 110.110 12.800Q110.110 13.354 109.913 13.942Q109.716 14.530 109.315 14.929Q108.915 15.327 108.360 15.327M108.360 14.769Q108.661 14.769 108.878 14.552Q109.094 14.335 109.221 14.007Q109.348 13.679 109.409 13.333Q109.469 12.987 109.469 12.706Q109.469 12.456 109.407 12.130Q109.344 11.804 109.214 11.513Q109.083 11.222 108.870 11.032Q108.657 10.843 108.360 10.843Q107.985 10.843 107.733 11.155Q107.481 11.468 107.364 11.901Q107.247 12.335 107.247 12.706Q107.247 13.104 107.360 13.585Q107.473 14.065 107.721 14.417Q107.969 14.769 108.360 14.769\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003Cg fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\">\u003Cg fill=\"var(--tk-warn)\" stroke=\"none\" font-family=\"cmtt8\" font-size=\"8\">\u003Cg transform=\"translate(-75.293 -39.099)\">\u003Cpath d=\"M48.601 15.288Q48.136 15.288 47.771 15.038Q47.406 14.788 47.201 14.384Q46.996 13.979 46.996 13.522Q46.996 13.179 47.121 12.858Q47.246 12.538 47.478 12.288Q47.711 12.038 48.015 11.899Q48.320 11.761 48.675 11.761Q49.187 11.761 49.593 12.081L49.593 10.921L49.172 10.921Q48.961 10.897 48.922 10.683L48.922 10.593Q48.961 10.386 49.172 10.362L49.984 10.362Q50.183 10.386 50.234 10.593L50.234 14.690L50.660 14.690Q50.867 14.714 50.906 14.921L50.906 15.011Q50.867 15.226 50.660 15.249L49.843 15.249Q49.644 15.226 49.593 15.011L49.593 14.882Q49.398 15.073 49.136 15.181Q48.875 15.288 48.601 15.288M48.640 14.729Q49 14.729 49.252 14.460Q49.504 14.190 49.593 13.815L49.593 12.983Q49.535 12.796 49.408 12.645Q49.281 12.495 49.103 12.407Q48.925 12.319 48.730 12.319Q48.422 12.319 48.172 12.489Q47.922 12.659 47.777 12.944Q47.633 13.229 47.633 13.530Q47.633 13.979 47.918 14.354Q48.203 14.729 48.640 14.729M54.496 13.761L52.054 13.761Q52.109 14.038 52.306 14.261Q52.504 14.483 52.781 14.606Q53.058 14.729 53.343 14.729Q53.816 14.729 54.039 14.440Q54.047 14.429 54.103 14.323Q54.160 14.218 54.209 14.175Q54.258 14.132 54.351 14.120L54.496 14.120Q54.687 14.140 54.746 14.354L54.746 14.409Q54.679 14.710 54.449 14.907Q54.218 15.104 53.906 15.196Q53.593 15.288 53.289 15.288Q52.804 15.288 52.365 15.060Q51.925 14.831 51.658 14.431Q51.390 14.030 51.390 13.538L51.390 13.479Q51.390 13.011 51.636 12.608Q51.883 12.206 52.291 11.972Q52.699 11.737 53.168 11.737Q53.672 11.737 54.025 11.960Q54.379 12.183 54.562 12.571Q54.746 12.960 54.746 13.464L54.746 13.522Q54.687 13.737 54.496 13.761M52.062 13.210L54.090 13.210Q54.043 12.800 53.804 12.548Q53.566 12.296 53.168 12.296Q52.773 12.296 52.467 12.558Q52.160 12.819 52.062 13.210M55.453 15.011L55.453 14.921Q55.511 14.714 55.703 14.690L56.414 14.690L56.414 12.362L55.703 12.362Q55.508 12.339 55.453 12.120L55.453 12.034Q55.511 11.823 55.703 11.800L56.804 11.800Q57.004 11.819 57.054 12.034L57.054 12.362Q57.316 12.077 57.672 11.919Q58.027 11.761 58.414 11.761Q58.707 11.761 58.941 11.895Q59.175 12.030 59.175 12.296Q59.175 12.464 59.066 12.581Q58.957 12.698 58.789 12.698Q58.636 12.698 58.521 12.587Q58.406 12.476 58.406 12.319Q58.031 12.319 57.717 12.520Q57.402 12.722 57.228 13.056Q57.054 13.390 57.054 13.769L57.054 14.690L58 14.690Q58.207 14.714 58.246 14.921L58.246 15.011Q58.207 15.226 58 15.249L55.703 15.249Q55.511 15.226 55.453 15.011M60.093 15.011L60.093 14.921Q60.144 14.714 60.340 14.690L61.379 14.690L61.379 12.362L60.406 12.362Q60.207 12.339 60.156 12.120L60.156 12.034Q60.207 11.823 60.406 11.800L61.773 11.800Q61.968 11.819 62.019 12.034L62.019 14.690L62.933 14.690Q63.129 14.714 63.179 14.921L63.179 15.011Q63.129 15.226 62.933 15.249L60.340 15.249Q60.144 15.226 60.093 15.011M61.125 10.823L61.125 10.769Q61.125 10.597 61.261 10.476Q61.398 10.354 61.574 10.354Q61.746 10.354 61.883 10.476Q62.019 10.597 62.019 10.769L62.019 10.823Q62.019 10.999 61.883 11.120Q61.746 11.241 61.574 11.241Q61.398 11.241 61.261 11.120Q61.125 10.999 61.125 10.823M65.379 15.026L64.492 12.362L64.172 12.362Q63.972 12.339 63.922 12.120L63.922 12.034Q63.972 11.823 64.172 11.800L65.332 11.800Q65.527 11.819 65.578 12.034L65.578 12.120Q65.527 12.339 65.332 12.362L65.050 12.362L65.843 14.737L66.633 12.362L66.355 12.362Q66.156 12.339 66.105 12.120L66.105 12.034Q66.156 11.823 66.355 11.800L67.515 11.800Q67.711 11.823 67.761 12.034L67.761 12.120Q67.711 12.339 67.515 12.362L67.195 12.362L66.308 15.026Q66.265 15.140 66.164 15.214Q66.062 15.288 65.937 15.288L65.746 15.288Q65.629 15.288 65.525 15.216Q65.422 15.144 65.379 15.026M68.375 14.136Q68.375 13.690 68.789 13.433Q69.203 13.175 69.744 13.075Q70.285 12.976 70.793 12.968Q70.793 12.753 70.658 12.601Q70.523 12.448 70.316 12.372Q70.109 12.296 69.898 12.296Q69.554 12.296 69.394 12.319L69.394 12.378Q69.394 12.546 69.275 12.661Q69.156 12.776 68.992 12.776Q68.816 12.776 68.701 12.653Q68.586 12.530 68.586 12.362Q68.586 11.956 68.967 11.847Q69.347 11.737 69.906 11.737Q70.175 11.737 70.443 11.815Q70.711 11.894 70.935 12.044Q71.160 12.194 71.297 12.415Q71.433 12.636 71.433 12.913L71.433 14.632Q71.433 14.690 71.961 14.690Q72.156 14.710 72.207 14.921L72.207 15.011Q72.156 15.226 71.961 15.249L71.816 15.249Q71.472 15.249 71.244 15.202Q71.015 15.155 70.871 14.968Q70.410 15.288 69.703 15.288Q69.367 15.288 69.062 15.147Q68.758 15.007 68.566 14.745Q68.375 14.483 68.375 14.136M69.015 14.144Q69.015 14.417 69.258 14.573Q69.500 14.729 69.785 14.729Q70.004 14.729 70.236 14.671Q70.468 14.612 70.631 14.474Q70.793 14.335 70.793 14.112L70.793 13.522Q70.511 13.522 70.095 13.579Q69.679 13.636 69.347 13.774Q69.015 13.913 69.015 14.144M73.414 14.144L73.414 12.362L72.664 12.362Q72.465 12.339 72.414 12.120L72.414 12.034Q72.465 11.823 72.664 11.800L73.414 11.800L73.414 11.050Q73.465 10.843 73.664 10.815L73.808 10.815Q74.004 10.843 74.054 11.050L74.054 11.800L75.414 11.800Q75.605 11.819 75.664 12.034L75.664 12.120Q75.609 12.339 75.414 12.362L74.054 12.362L74.054 14.112Q74.054 14.729 74.629 14.729Q74.879 14.729 75.043 14.544Q75.207 14.358 75.207 14.112Q75.207 14.019 75.279 13.948Q75.351 13.878 75.453 13.866L75.597 13.866Q75.797 13.890 75.847 14.097L75.847 14.144Q75.847 14.468 75.664 14.731Q75.480 14.995 75.187 15.142Q74.894 15.288 74.574 15.288Q74.062 15.288 73.738 14.978Q73.414 14.667 73.414 14.144M77.078 15.011L77.078 14.921Q77.129 14.714 77.324 14.690L78.363 14.690L78.363 12.362L77.390 12.362Q77.191 12.339 77.140 12.120L77.140 12.034Q77.191 11.823 77.390 11.800L78.758 11.800Q78.953 11.819 79.004 12.034L79.004 14.690L79.918 14.690Q80.113 14.714 80.164 14.921L80.164 15.011Q80.113 15.226 79.918 15.249L77.324 15.249Q77.129 15.226 77.078 15.011M78.109 10.823L78.109 10.769Q78.109 10.597 78.246 10.476Q78.383 10.354 78.558 10.354Q78.730 10.354 78.867 10.476Q79.004 10.597 79.004 10.769L79.004 10.823Q79.004 10.999 78.867 11.120Q78.730 11.241 78.558 11.241Q78.383 11.241 78.246 11.120Q78.109 10.999 78.109 10.823M82.363 15.026L81.476 12.362L81.156 12.362Q80.957 12.339 80.906 12.120L80.906 12.034Q80.957 11.823 81.156 11.800L82.316 11.800Q82.511 11.819 82.562 12.034L82.562 12.120Q82.511 12.339 82.316 12.362L82.035 12.362L82.828 14.737L83.617 12.362L83.340 12.362Q83.140 12.339 83.090 12.120L83.090 12.034Q83.140 11.823 83.340 11.800L84.500 11.800Q84.695 11.823 84.746 12.034L84.746 12.120Q84.695 12.339 84.500 12.362L84.179 12.362L83.293 15.026Q83.250 15.140 83.148 15.214Q83.047 15.288 82.922 15.288L82.730 15.288Q82.613 15.288 82.509 15.216Q82.406 15.144 82.363 15.026M88.465 13.761L86.023 13.761Q86.078 14.038 86.275 14.261Q86.472 14.483 86.750 14.606Q87.027 14.729 87.312 14.729Q87.785 14.729 88.008 14.440Q88.015 14.429 88.072 14.323Q88.129 14.218 88.177 14.175Q88.226 14.132 88.320 14.120L88.465 14.120Q88.656 14.140 88.715 14.354L88.715 14.409Q88.648 14.710 88.418 14.907Q88.187 15.104 87.875 15.196Q87.562 15.288 87.258 15.288Q86.773 15.288 86.334 15.060Q85.894 14.831 85.627 14.431Q85.359 14.030 85.359 13.538L85.359 13.479Q85.359 13.011 85.605 12.608Q85.851 12.206 86.259 11.972Q86.668 11.737 87.136 11.737Q87.640 11.737 87.994 11.960Q88.347 12.183 88.531 12.571Q88.715 12.960 88.715 13.464L88.715 13.522Q88.656 13.737 88.465 13.761M86.031 13.210L88.058 13.210Q88.011 12.800 87.773 12.548Q87.535 12.296 87.136 12.296Q86.742 12.296 86.435 12.558Q86.129 12.819 86.031 13.210\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(-75.293 -39.099)\">\u003Cpath d=\"M95.610 15.327Q95.177 15.327 94.844 15.089Q94.512 14.851 94.292 14.462Q94.071 14.073 93.964 13.636Q93.856 13.198 93.856 12.800Q93.856 12.409 93.966 11.966Q94.075 11.522 94.292 11.142Q94.509 10.761 94.843 10.520Q95.177 10.280 95.610 10.280Q96.165 10.280 96.565 10.679Q96.966 11.077 97.163 11.663Q97.360 12.249 97.360 12.800Q97.360 13.354 97.163 13.942Q96.966 14.530 96.565 14.929Q96.165 15.327 95.610 15.327M95.610 14.769Q95.911 14.769 96.128 14.552Q96.344 14.335 96.471 14.007Q96.598 13.679 96.659 13.333Q96.719 12.987 96.719 12.706Q96.719 12.456 96.657 12.130Q96.594 11.804 96.464 11.513Q96.333 11.222 96.120 11.032Q95.907 10.843 95.610 10.843Q95.235 10.843 94.983 11.155Q94.731 11.468 94.614 11.901Q94.497 12.335 94.497 12.706Q94.497 13.104 94.610 13.585Q94.723 14.065 94.971 14.417Q95.219 14.769 95.610 14.769\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003Cpath fill=\"var(--tk-warn)\" stroke=\"none\" d=\"M-27.684 30.899a2.4 2.4 0 1 0-4.8 0 2.4 2.4 0 0 0 4.8 0M-13.458 30.899a2.4 2.4 0 1 0-4.8 0 2.4 2.4 0 0 0 4.8 0M.7689999999999999 30.899a2.4 2.4 0 1 0-4.8 0 2.4 2.4 0 0 0 4.8 0M14.995000000000001 30.899a2.4 2.4 0 1 0-4.8 0 2.4 2.4 0 0 0 4.8 0M29.221999999999998 30.899a2.4 2.4 0 1 0-4.8 0 2.4 2.4 0 0 0 4.8 0m-2.4 0\"\u002F>\u003Cg fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\">\u003Cg fill=\"var(--tk-warn)\" stroke=\"none\" font-family=\"cmtt8\" font-size=\"8\">\u003Cg transform=\"translate(-92.293 29.761)\">\u003Cpath d=\"M47.144 14.136Q47.144 13.690 47.558 13.433Q47.972 13.175 48.513 13.075Q49.054 12.976 49.562 12.968Q49.562 12.753 49.427 12.601Q49.293 12.448 49.086 12.372Q48.879 12.296 48.668 12.296Q48.324 12.296 48.164 12.319L48.164 12.378Q48.164 12.546 48.045 12.661Q47.925 12.776 47.761 12.776Q47.586 12.776 47.470 12.653Q47.355 12.530 47.355 12.362Q47.355 11.956 47.736 11.847Q48.117 11.737 48.675 11.737Q48.945 11.737 49.213 11.815Q49.480 11.894 49.705 12.044Q49.929 12.194 50.066 12.415Q50.203 12.636 50.203 12.913L50.203 14.632Q50.203 14.690 50.730 14.690Q50.925 14.710 50.976 14.921L50.976 15.011Q50.925 15.226 50.730 15.249L50.586 15.249Q50.242 15.249 50.013 15.202Q49.785 15.155 49.640 14.968Q49.179 15.288 48.472 15.288Q48.136 15.288 47.832 15.147Q47.527 15.007 47.336 14.745Q47.144 14.483 47.144 14.136M47.785 14.144Q47.785 14.417 48.027 14.573Q48.269 14.729 48.554 14.729Q48.773 14.729 49.006 14.671Q49.238 14.612 49.400 14.474Q49.562 14.335 49.562 14.112L49.562 13.522Q49.281 13.522 48.865 13.579Q48.449 13.636 48.117 13.774Q47.785 13.913 47.785 14.144M51.433 15.011L51.433 14.921Q51.484 14.714 51.679 14.690L52.785 14.690L52.785 10.921L51.679 10.921Q51.484 10.897 51.433 10.683L51.433 10.593Q51.484 10.386 51.679 10.362L53.175 10.362Q53.367 10.386 53.425 10.593L53.425 14.690L54.527 14.690Q54.726 14.714 54.777 14.921L54.777 15.011Q54.726 15.226 54.527 15.249L51.679 15.249Q51.484 15.226 51.433 15.011M55.679 15.011L55.679 14.921Q55.730 14.714 55.925 14.690L57.031 14.690L57.031 10.921L55.925 10.921Q55.730 10.897 55.679 10.683L55.679 10.593Q55.730 10.386 55.925 10.362L57.422 10.362Q57.613 10.386 57.672 10.593L57.672 14.690L58.773 14.690Q58.972 14.714 59.023 14.921L59.023 15.011Q58.972 15.226 58.773 15.249L55.925 15.249Q55.730 15.226 55.679 15.011\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(-92.293 29.761)\">\u003Cpath d=\"M64.356 15.011L64.356 14.921Q64.407 14.714 64.602 14.690L65.641 14.690L65.641 12.362L64.669 12.362Q64.469 12.339 64.419 12.120L64.419 12.034Q64.469 11.823 64.669 11.800L66.036 11.800Q66.231 11.819 66.282 12.034L66.282 14.690L67.196 14.690Q67.391 14.714 67.442 14.921L67.442 15.011Q67.391 15.226 67.196 15.249L64.602 15.249Q64.407 15.226 64.356 15.011M65.387 10.823L65.387 10.769Q65.387 10.597 65.524 10.476Q65.661 10.354 65.837 10.354Q66.009 10.354 66.145 10.476Q66.282 10.597 66.282 10.769L66.282 10.823Q66.282 10.999 66.145 11.120Q66.009 11.241 65.837 11.241Q65.661 11.241 65.524 11.120Q65.387 10.999 65.387 10.823M68.055 15.011L68.055 14.921Q68.098 14.714 68.305 14.690L68.727 14.690L68.727 12.362L68.305 12.362Q68.098 12.339 68.055 12.120L68.055 12.034Q68.102 11.823 68.305 11.800L69.122 11.800Q69.317 11.823 69.368 12.034L69.368 12.120L69.360 12.144Q69.587 11.964 69.860 11.862Q70.134 11.761 70.427 11.761Q70.774 11.761 71.012 11.901Q71.251 12.042 71.366 12.300Q71.481 12.558 71.481 12.913L71.481 14.690L71.907 14.690Q72.114 14.714 72.153 14.921L72.153 15.011Q72.114 15.226 71.907 15.249L70.512 15.249Q70.317 15.226 70.266 15.011L70.266 14.921Q70.317 14.710 70.512 14.690L70.841 14.690L70.841 12.944Q70.841 12.636 70.751 12.478Q70.661 12.319 70.368 12.319Q70.098 12.319 69.870 12.450Q69.641 12.581 69.505 12.810Q69.368 13.038 69.368 13.304L69.368 14.690L69.794 14.690Q70.001 14.714 70.040 14.921L70.040 15.011Q70.001 15.226 69.794 15.249L68.305 15.249Q68.098 15.226 68.055 15.011M72.302 16.792L72.302 16.706Q72.344 16.487 72.552 16.464L72.973 16.464L72.973 12.362L72.552 12.362Q72.344 12.339 72.302 12.120L72.302 12.034Q72.348 11.823 72.552 11.800L73.368 11.800Q73.563 11.819 73.614 12.034L73.614 12.104Q73.825 11.937 74.089 11.849Q74.352 11.761 74.622 11.761Q74.962 11.761 75.259 11.905Q75.555 12.050 75.770 12.302Q75.985 12.554 76.100 12.866Q76.216 13.179 76.216 13.522Q76.216 13.987 75.989 14.397Q75.762 14.808 75.376 15.048Q74.989 15.288 74.520 15.288Q74.001 15.288 73.614 14.921L73.614 16.464L74.040 16.464Q74.247 16.487 74.286 16.706L74.286 16.792Q74.247 17.003 74.040 17.026L72.552 17.026Q72.348 17.003 72.302 16.792M74.477 14.729Q74.786 14.729 75.036 14.560Q75.286 14.390 75.430 14.108Q75.575 13.827 75.575 13.522Q75.575 13.233 75.450 12.954Q75.325 12.675 75.093 12.497Q74.860 12.319 74.559 12.319Q74.239 12.319 73.977 12.505Q73.716 12.690 73.614 12.991L73.614 13.843Q73.704 14.210 73.925 14.470Q74.145 14.729 74.477 14.729M77.219 14.394L77.219 12.362L76.798 12.362Q76.591 12.339 76.548 12.120L76.548 12.034Q76.594 11.823 76.798 11.800L77.614 11.800Q77.809 11.823 77.860 12.034L77.860 14.362Q77.860 14.597 78.030 14.663Q78.200 14.729 78.485 14.729Q78.692 14.729 78.887 14.653Q79.083 14.577 79.208 14.427Q79.333 14.276 79.333 14.065L79.333 12.362L78.911 12.362Q78.700 12.339 78.661 12.120L78.661 12.034Q78.700 11.823 78.911 11.800L79.723 11.800Q79.923 11.823 79.973 12.034L79.973 14.690L80.399 14.690Q80.606 14.714 80.645 14.921L80.645 15.011Q80.606 15.226 80.399 15.249L79.583 15.249Q79.384 15.226 79.333 15.026Q78.930 15.288 78.423 15.288Q78.188 15.288 77.973 15.247Q77.759 15.206 77.593 15.104Q77.427 15.003 77.323 14.825Q77.219 14.647 77.219 14.394M81.923 14.144L81.923 12.362L81.173 12.362Q80.973 12.339 80.923 12.120L80.923 12.034Q80.973 11.823 81.173 11.800L81.923 11.800L81.923 11.050Q81.973 10.843 82.173 10.815L82.317 10.815Q82.512 10.843 82.563 11.050L82.563 11.800L83.923 11.800Q84.114 11.819 84.173 12.034L84.173 12.120Q84.118 12.339 83.923 12.362L82.563 12.362L82.563 14.112Q82.563 14.729 83.137 14.729Q83.387 14.729 83.552 14.544Q83.716 14.358 83.716 14.112Q83.716 14.019 83.788 13.948Q83.860 13.878 83.962 13.866L84.106 13.866Q84.305 13.890 84.356 14.097L84.356 14.144Q84.356 14.468 84.173 14.731Q83.989 14.995 83.696 15.142Q83.403 15.288 83.083 15.288Q82.571 15.288 82.247 14.978Q81.923 14.667 81.923 14.144M85.555 15.050L85.555 14.136Q85.583 13.929 85.794 13.905L85.962 13.905Q86.126 13.929 86.184 14.089Q86.387 14.729 87.114 14.729Q87.321 14.729 87.550 14.694Q87.778 14.659 87.946 14.544Q88.114 14.429 88.114 14.226Q88.114 14.015 87.891 13.901Q87.669 13.788 87.395 13.745L86.696 13.632Q85.555 13.421 85.555 12.698Q85.555 12.409 85.700 12.220Q85.844 12.030 86.085 11.923Q86.325 11.815 86.581 11.776Q86.837 11.737 87.114 11.737Q87.364 11.737 87.557 11.767Q87.751 11.796 87.915 11.874Q87.993 11.757 88.122 11.737L88.200 11.737Q88.298 11.749 88.360 11.812Q88.423 11.874 88.434 11.968L88.434 12.675Q88.423 12.769 88.360 12.835Q88.298 12.901 88.200 12.913L88.032 12.913Q87.938 12.901 87.872 12.835Q87.805 12.769 87.794 12.675Q87.794 12.296 87.098 12.296Q86.751 12.296 86.432 12.378Q86.114 12.460 86.114 12.706Q86.114 12.972 86.786 13.081L87.489 13.202Q87.973 13.284 88.323 13.532Q88.673 13.780 88.673 14.226Q88.673 14.616 88.436 14.858Q88.200 15.101 87.850 15.194Q87.501 15.288 87.114 15.288Q86.536 15.288 86.137 15.034Q86.067 15.159 86.018 15.216Q85.969 15.272 85.864 15.288L85.794 15.288Q85.579 15.265 85.555 15.050\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(-92.293 29.761)\">\u003Cpath d=\"M93.938 15.011L93.938 14.921Q93.989 14.714 94.184 14.690L95.290 14.690L95.290 10.921L94.184 10.921Q93.989 10.897 93.938 10.683L93.938 10.593Q93.989 10.386 94.184 10.362L95.680 10.362Q95.872 10.386 95.930 10.593L95.930 14.690L97.032 14.690Q97.231 14.714 97.282 14.921L97.282 15.011Q97.231 15.226 97.032 15.249L94.184 15.249Q93.989 15.226 93.938 15.011M98.141 14.136Q98.141 13.690 98.555 13.433Q98.969 13.175 99.510 13.075Q100.052 12.976 100.559 12.968Q100.559 12.753 100.425 12.601Q100.290 12.448 100.083 12.372Q99.876 12.296 99.665 12.296Q99.321 12.296 99.161 12.319L99.161 12.378Q99.161 12.546 99.042 12.661Q98.923 12.776 98.759 12.776Q98.583 12.776 98.468 12.653Q98.352 12.530 98.352 12.362Q98.352 11.956 98.733 11.847Q99.114 11.737 99.673 11.737Q99.942 11.737 100.210 11.815Q100.477 11.894 100.702 12.044Q100.927 12.194 101.063 12.415Q101.200 12.636 101.200 12.913L101.200 14.632Q101.200 14.690 101.727 14.690Q101.923 14.710 101.973 14.921L101.973 15.011Q101.923 15.226 101.727 15.249L101.583 15.249Q101.239 15.249 101.010 15.202Q100.782 15.155 100.637 14.968Q100.177 15.288 99.469 15.288Q99.134 15.288 98.829 15.147Q98.524 15.007 98.333 14.745Q98.141 14.483 98.141 14.136M98.782 14.144Q98.782 14.417 99.024 14.573Q99.266 14.729 99.552 14.729Q99.770 14.729 100.003 14.671Q100.235 14.612 100.397 14.474Q100.559 14.335 100.559 14.112L100.559 13.522Q100.278 13.522 99.862 13.579Q99.446 13.636 99.114 13.774Q98.782 13.913 98.782 14.144M102.052 15.011L102.052 14.921Q102.094 14.714 102.302 14.690L102.723 14.690L102.723 12.362L102.302 12.362Q102.094 12.339 102.052 12.120L102.052 12.034Q102.098 11.823 102.302 11.800L103.118 11.800Q103.313 11.823 103.364 12.034L103.364 12.120L103.356 12.144Q103.583 11.964 103.856 11.862Q104.130 11.761 104.423 11.761Q104.770 11.761 105.009 11.901Q105.247 12.042 105.362 12.300Q105.477 12.558 105.477 12.913L105.477 14.690L105.903 14.690Q106.110 14.714 106.149 14.921L106.149 15.011Q106.110 15.226 105.903 15.249L104.509 15.249Q104.313 15.226 104.262 15.011L104.262 14.921Q104.313 14.710 104.509 14.690L104.837 14.690L104.837 12.944Q104.837 12.636 104.747 12.478Q104.657 12.319 104.364 12.319Q104.094 12.319 103.866 12.450Q103.637 12.581 103.501 12.810Q103.364 13.038 103.364 13.304L103.364 14.690L103.790 14.690Q103.997 14.714 104.036 14.921L104.036 15.011Q103.997 15.226 103.790 15.249L102.302 15.249Q102.094 15.226 102.052 15.011M108.091 15.288Q107.626 15.288 107.260 15.038Q106.895 14.788 106.690 14.384Q106.485 13.979 106.485 13.522Q106.485 13.179 106.610 12.858Q106.735 12.538 106.968 12.288Q107.200 12.038 107.505 11.899Q107.809 11.761 108.165 11.761Q108.677 11.761 109.083 12.081L109.083 10.921L108.661 10.921Q108.450 10.897 108.411 10.683L108.411 10.593Q108.450 10.386 108.661 10.362L109.473 10.362Q109.673 10.386 109.723 10.593L109.723 14.690L110.149 14.690Q110.356 14.714 110.395 14.921L110.395 15.011Q110.356 15.226 110.149 15.249L109.333 15.249Q109.134 15.226 109.083 15.011L109.083 14.882Q108.887 15.073 108.626 15.181Q108.364 15.288 108.091 15.288M108.130 14.729Q108.489 14.729 108.741 14.460Q108.993 14.190 109.083 13.815L109.083 12.983Q109.024 12.796 108.897 12.645Q108.770 12.495 108.593 12.407Q108.415 12.319 108.219 12.319Q107.911 12.319 107.661 12.489Q107.411 12.659 107.266 12.944Q107.122 13.229 107.122 13.530Q107.122 13.979 107.407 14.354Q107.692 14.729 108.130 14.729\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(-92.293 29.761)\">\u003Cpath d=\"M114.809 15.011L114.809 14.921Q114.852 14.714 115.059 14.690L115.481 14.690L115.481 10.921L115.059 10.921Q114.852 10.897 114.809 10.683L114.809 10.593Q114.852 10.386 115.059 10.362L115.876 10.362Q116.071 10.386 116.122 10.593L116.122 12.144Q116.583 11.761 117.180 11.761Q117.528 11.761 117.766 11.901Q118.005 12.042 118.120 12.300Q118.235 12.558 118.235 12.913L118.235 14.690L118.661 14.690Q118.868 14.714 118.907 14.921L118.907 15.011Q118.868 15.226 118.661 15.249L117.266 15.249Q117.071 15.226 117.020 15.011L117.020 14.921Q117.071 14.710 117.266 14.690L117.594 14.690L117.594 12.944Q117.594 12.636 117.505 12.478Q117.415 12.319 117.122 12.319Q116.852 12.319 116.624 12.450Q116.395 12.581 116.259 12.810Q116.122 13.038 116.122 13.304L116.122 14.690L116.548 14.690Q116.755 14.714 116.794 14.921L116.794 15.011Q116.755 15.226 116.548 15.249L115.059 15.249Q114.852 15.226 114.809 15.011M122.497 13.761L120.055 13.761Q120.110 14.038 120.307 14.261Q120.505 14.483 120.782 14.606Q121.059 14.729 121.344 14.729Q121.817 14.729 122.040 14.440Q122.048 14.429 122.104 14.323Q122.161 14.218 122.210 14.175Q122.259 14.132 122.352 14.120L122.497 14.120Q122.688 14.140 122.747 14.354L122.747 14.409Q122.680 14.710 122.450 14.907Q122.219 15.104 121.907 15.196Q121.594 15.288 121.290 15.288Q120.805 15.288 120.366 15.060Q119.927 14.831 119.659 14.431Q119.391 14.030 119.391 13.538L119.391 13.479Q119.391 13.011 119.637 12.608Q119.884 12.206 120.292 11.972Q120.700 11.737 121.169 11.737Q121.673 11.737 122.026 11.960Q122.380 12.183 122.563 12.571Q122.747 12.960 122.747 13.464L122.747 13.522Q122.688 13.737 122.497 13.761M120.063 13.210L122.091 13.210Q122.044 12.800 121.805 12.548Q121.567 12.296 121.169 12.296Q120.774 12.296 120.468 12.558Q120.161 12.819 120.063 13.210M123.454 15.011L123.454 14.921Q123.512 14.714 123.704 14.690L124.415 14.690L124.415 12.362L123.704 12.362Q123.509 12.339 123.454 12.120L123.454 12.034Q123.512 11.823 123.704 11.800L124.805 11.800Q125.005 11.819 125.055 12.034L125.055 12.362Q125.317 12.077 125.673 11.919Q126.028 11.761 126.415 11.761Q126.708 11.761 126.942 11.895Q127.177 12.030 127.177 12.296Q127.177 12.464 127.067 12.581Q126.958 12.698 126.790 12.698Q126.637 12.698 126.522 12.587Q126.407 12.476 126.407 12.319Q126.032 12.319 125.718 12.520Q125.403 12.722 125.229 13.056Q125.055 13.390 125.055 13.769L125.055 14.690L126.001 14.690Q126.208 14.714 126.247 14.921L126.247 15.011Q126.208 15.226 126.001 15.249L123.704 15.249Q123.512 15.226 123.454 15.011M130.989 13.761L128.548 13.761Q128.602 14.038 128.800 14.261Q128.997 14.483 129.274 14.606Q129.552 14.729 129.837 14.729Q130.309 14.729 130.532 14.440Q130.540 14.429 130.596 14.323Q130.653 14.218 130.702 14.175Q130.751 14.132 130.844 14.120L130.989 14.120Q131.180 14.140 131.239 14.354L131.239 14.409Q131.173 14.710 130.942 14.907Q130.712 15.104 130.399 15.196Q130.087 15.288 129.782 15.288Q129.298 15.288 128.858 15.060Q128.419 14.831 128.151 14.431Q127.884 14.030 127.884 13.538L127.884 13.479Q127.884 13.011 128.130 12.608Q128.376 12.206 128.784 11.972Q129.192 11.737 129.661 11.737Q130.165 11.737 130.518 11.960Q130.872 12.183 131.055 12.571Q131.239 12.960 131.239 13.464L131.239 13.522Q131.180 13.737 130.989 13.761M128.555 13.210L130.583 13.210Q130.536 12.800 130.298 12.548Q130.059 12.296 129.661 12.296Q129.266 12.296 128.960 12.558Q128.653 12.819 128.555 13.210\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003Cg fill=\"var(--tk-good)\" stroke=\"var(--tk-good)\">\u003Cg fill=\"var(--tk-good)\" stroke=\"none\" font-family=\"cmtt8\" font-size=\"8\">\u003Cg transform=\"translate(19.182 -62.463)\">\u003Cpath d=\"M47.144 14.136Q47.144 13.690 47.558 13.433Q47.972 13.175 48.513 13.075Q49.054 12.976 49.562 12.968Q49.562 12.753 49.427 12.601Q49.293 12.448 49.086 12.372Q48.879 12.296 48.668 12.296Q48.324 12.296 48.164 12.319L48.164 12.378Q48.164 12.546 48.045 12.661Q47.925 12.776 47.761 12.776Q47.586 12.776 47.470 12.653Q47.355 12.530 47.355 12.362Q47.355 11.956 47.736 11.847Q48.117 11.737 48.675 11.737Q48.945 11.737 49.213 11.815Q49.480 11.894 49.705 12.044Q49.929 12.194 50.066 12.415Q50.203 12.636 50.203 12.913L50.203 14.632Q50.203 14.690 50.730 14.690Q50.925 14.710 50.976 14.921L50.976 15.011Q50.925 15.226 50.730 15.249L50.586 15.249Q50.242 15.249 50.013 15.202Q49.785 15.155 49.640 14.968Q49.179 15.288 48.472 15.288Q48.136 15.288 47.832 15.147Q47.527 15.007 47.336 14.745Q47.144 14.483 47.144 14.136M47.785 14.144Q47.785 14.417 48.027 14.573Q48.269 14.729 48.554 14.729Q48.773 14.729 49.006 14.671Q49.238 14.612 49.400 14.474Q49.562 14.335 49.562 14.112L49.562 13.522Q49.281 13.522 48.865 13.579Q48.449 13.636 48.117 13.774Q47.785 13.913 47.785 14.144M51.433 15.011L51.433 14.921Q51.484 14.714 51.679 14.690L52.785 14.690L52.785 10.921L51.679 10.921Q51.484 10.897 51.433 10.683L51.433 10.593Q51.484 10.386 51.679 10.362L53.175 10.362Q53.367 10.386 53.425 10.593L53.425 14.690L54.527 14.690Q54.726 14.714 54.777 14.921L54.777 15.011Q54.726 15.226 54.527 15.249L51.679 15.249Q51.484 15.226 51.433 15.011M55.847 15.011L55.847 14.921Q55.898 14.714 56.093 14.690L57.133 14.690L57.133 12.362L56.160 12.362Q55.961 12.339 55.910 12.120L55.910 12.034Q55.961 11.823 56.160 11.800L57.527 11.800Q57.722 11.819 57.773 12.034L57.773 14.690L58.687 14.690Q58.883 14.714 58.933 14.921L58.933 15.011Q58.883 15.226 58.687 15.249L56.093 15.249Q55.898 15.226 55.847 15.011M56.879 10.823L56.879 10.769Q56.879 10.597 57.015 10.476Q57.152 10.354 57.328 10.354Q57.500 10.354 57.636 10.476Q57.773 10.597 57.773 10.769L57.773 10.823Q57.773 10.999 57.636 11.120Q57.500 11.241 57.328 11.241Q57.152 11.241 57.015 11.120Q56.879 10.999 56.879 10.823M61.133 15.026L60.246 12.362L59.925 12.362Q59.726 12.339 59.675 12.120L59.675 12.034Q59.726 11.823 59.925 11.800L61.086 11.800Q61.281 11.819 61.332 12.034L61.332 12.120Q61.281 12.339 61.086 12.362L60.804 12.362L61.597 14.737L62.386 12.362L62.109 12.362Q61.910 12.339 61.859 12.120L61.859 12.034Q61.910 11.823 62.109 11.800L63.269 11.800Q63.465 11.823 63.515 12.034L63.515 12.120Q63.465 12.339 63.269 12.362L62.949 12.362L62.062 15.026Q62.019 15.140 61.918 15.214Q61.816 15.288 61.691 15.288L61.500 15.288Q61.383 15.288 61.279 15.216Q61.175 15.144 61.133 15.026M67.234 13.761L64.793 13.761Q64.847 14.038 65.045 14.261Q65.242 14.483 65.519 14.606Q65.797 14.729 66.082 14.729Q66.554 14.729 66.777 14.440Q66.785 14.429 66.842 14.323Q66.898 14.218 66.947 14.175Q66.996 14.132 67.090 14.120L67.234 14.120Q67.425 14.140 67.484 14.354L67.484 14.409Q67.418 14.710 67.187 14.907Q66.957 15.104 66.644 15.196Q66.332 15.288 66.027 15.288Q65.543 15.288 65.103 15.060Q64.664 14.831 64.396 14.431Q64.129 14.030 64.129 13.538L64.129 13.479Q64.129 13.011 64.375 12.608Q64.621 12.206 65.029 11.972Q65.437 11.737 65.906 11.737Q66.410 11.737 66.763 11.960Q67.117 12.183 67.300 12.571Q67.484 12.960 67.484 13.464L67.484 13.522Q67.425 13.737 67.234 13.761M64.800 13.210L66.828 13.210Q66.781 12.800 66.543 12.548Q66.304 12.296 65.906 12.296Q65.511 12.296 65.205 12.558Q64.898 12.819 64.800 13.210M69.535 14.690Q69.535 14.468 69.701 14.302Q69.867 14.136 70.097 14.136Q70.246 14.136 70.373 14.214Q70.500 14.292 70.574 14.417Q70.648 14.542 70.648 14.690Q70.648 14.917 70.482 15.083Q70.316 15.249 70.097 15.249Q69.871 15.249 69.703 15.081Q69.535 14.913 69.535 14.690M69.535 12.354Q69.535 12.132 69.701 11.966Q69.867 11.800 70.097 11.800Q70.246 11.800 70.373 11.878Q70.500 11.956 70.574 12.081Q70.648 12.206 70.648 12.354Q70.648 12.581 70.482 12.747Q70.316 12.913 70.097 12.913Q69.871 12.913 69.703 12.745Q69.535 12.577 69.535 12.354\" fill=\"var(--tk-good)\" stroke=\"var(--tk-good)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(19.182 -62.463)\">\u003Cpath d=\"M81.325 15.050L81.325 14.136Q81.352 13.929 81.563 13.905L81.731 13.905Q81.895 13.929 81.954 14.089Q82.157 14.729 82.884 14.729Q83.091 14.729 83.319 14.694Q83.548 14.659 83.716 14.544Q83.884 14.429 83.884 14.226Q83.884 14.015 83.661 13.901Q83.438 13.788 83.165 13.745L82.466 13.632Q81.325 13.421 81.325 12.698Q81.325 12.409 81.469 12.220Q81.614 12.030 81.854 11.923Q82.094 11.815 82.350 11.776Q82.606 11.737 82.884 11.737Q83.134 11.737 83.327 11.767Q83.520 11.796 83.684 11.874Q83.762 11.757 83.891 11.737L83.969 11.737Q84.067 11.749 84.130 11.812Q84.192 11.874 84.204 11.968L84.204 12.675Q84.192 12.769 84.130 12.835Q84.067 12.901 83.969 12.913L83.802 12.913Q83.708 12.901 83.641 12.835Q83.575 12.769 83.563 12.675Q83.563 12.296 82.868 12.296Q82.520 12.296 82.202 12.378Q81.884 12.460 81.884 12.706Q81.884 12.972 82.555 13.081L83.259 13.202Q83.743 13.284 84.093 13.532Q84.442 13.780 84.442 14.226Q84.442 14.616 84.206 14.858Q83.969 15.101 83.620 15.194Q83.270 15.288 82.884 15.288Q82.305 15.288 81.907 15.034Q81.837 15.159 81.788 15.216Q81.739 15.272 81.634 15.288L81.563 15.288Q81.348 15.265 81.325 15.050M85.434 15.011L85.434 14.921Q85.485 14.714 85.680 14.690L86.786 14.690L86.786 10.921L85.680 10.921Q85.485 10.897 85.434 10.683L85.434 10.593Q85.485 10.386 85.680 10.362L87.177 10.362Q87.368 10.386 87.427 10.593L87.427 14.690L88.528 14.690Q88.727 14.714 88.778 14.921L88.778 15.011Q88.727 15.226 88.528 15.249L85.680 15.249Q85.485 15.226 85.434 15.011M91.352 15.288Q90.880 15.288 90.495 15.044Q90.110 14.800 89.887 14.390Q89.665 13.979 89.665 13.522Q89.665 13.179 89.790 12.856Q89.915 12.534 90.145 12.280Q90.376 12.026 90.682 11.882Q90.989 11.737 91.352 11.737Q91.716 11.737 92.028 11.884Q92.341 12.030 92.563 12.276Q92.786 12.522 92.913 12.843Q93.040 13.163 93.040 13.522Q93.040 13.979 92.815 14.392Q92.591 14.804 92.206 15.046Q91.821 15.288 91.352 15.288M91.352 14.729Q91.817 14.729 92.108 14.335Q92.399 13.940 92.399 13.456Q92.399 13.163 92.264 12.895Q92.130 12.628 91.889 12.462Q91.649 12.296 91.352 12.296Q91.048 12.296 90.809 12.462Q90.571 12.628 90.436 12.895Q90.302 13.163 90.302 13.456Q90.302 13.937 90.594 14.333Q90.887 14.729 91.352 14.729M93.548 16.792L93.548 16.706Q93.591 16.487 93.798 16.464L94.219 16.464L94.219 12.362L93.798 12.362Q93.591 12.339 93.548 12.120L93.548 12.034Q93.594 11.823 93.798 11.800L94.614 11.800Q94.809 11.819 94.860 12.034L94.860 12.104Q95.071 11.937 95.335 11.849Q95.598 11.761 95.868 11.761Q96.208 11.761 96.505 11.905Q96.802 12.050 97.016 12.302Q97.231 12.554 97.346 12.866Q97.462 13.179 97.462 13.522Q97.462 13.987 97.235 14.397Q97.009 14.808 96.622 15.048Q96.235 15.288 95.766 15.288Q95.247 15.288 94.860 14.921L94.860 16.464L95.286 16.464Q95.493 16.487 95.532 16.706L95.532 16.792Q95.493 17.003 95.286 17.026L93.798 17.026Q93.594 17.003 93.548 16.792M95.723 14.729Q96.032 14.729 96.282 14.560Q96.532 14.390 96.677 14.108Q96.821 13.827 96.821 13.522Q96.821 13.233 96.696 12.954Q96.571 12.675 96.339 12.497Q96.106 12.319 95.805 12.319Q95.485 12.319 95.223 12.505Q94.962 12.690 94.860 12.991L94.860 13.843Q94.950 14.210 95.171 14.470Q95.391 14.729 95.723 14.729M101.235 13.761L98.794 13.761Q98.848 14.038 99.046 14.261Q99.243 14.483 99.520 14.606Q99.798 14.729 100.083 14.729Q100.555 14.729 100.778 14.440Q100.786 14.429 100.843 14.323Q100.899 14.218 100.948 14.175Q100.997 14.132 101.091 14.120L101.235 14.120Q101.427 14.140 101.485 14.354L101.485 14.409Q101.419 14.710 101.188 14.907Q100.958 15.104 100.645 15.196Q100.333 15.288 100.028 15.288Q99.544 15.288 99.104 15.060Q98.665 14.831 98.397 14.431Q98.130 14.030 98.130 13.538L98.130 13.479Q98.130 13.011 98.376 12.608Q98.622 12.206 99.030 11.972Q99.438 11.737 99.907 11.737Q100.411 11.737 100.764 11.960Q101.118 12.183 101.302 12.571Q101.485 12.960 101.485 13.464L101.485 13.522Q101.427 13.737 101.235 13.761M98.802 13.210L100.829 13.210Q100.782 12.800 100.544 12.548Q100.305 12.296 99.907 12.296Q99.512 12.296 99.206 12.558Q98.899 12.819 98.802 13.210\" fill=\"var(--tk-good)\" stroke=\"var(--tk-good)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(19.182 -62.463)\">\u003Cpath d=\"M107.094 15.011L107.094 14.921Q107.145 14.714 107.344 14.690L108.161 14.690L108.161 11.507Q107.782 11.815 107.329 11.815Q107.098 11.815 107.048 11.585L107.048 11.495Q107.098 11.280 107.294 11.257Q107.622 11.257 107.876 11.019Q108.130 10.780 108.270 10.433Q108.341 10.304 108.497 10.280L108.552 10.280Q108.747 10.300 108.798 10.515L108.798 14.690L109.614 14.690Q109.813 14.714 109.864 14.921L109.864 15.011Q109.813 15.226 109.614 15.249L107.344 15.249Q107.145 15.226 107.094 15.011\" fill=\"var(--tk-good)\" stroke=\"var(--tk-good)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fsvg>\u003Cfigcaption class=\"tikz-cap\">A dead ReLU: its pre-activation stays in the flat region \u003Cspan class=\"katex\">\u003Cspan class=\"katex-html\" aria-hidden=\"true\">\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.5782em;vertical-align:-0.0391em;\">\u003C\u002Fspan>\u003Cspan class=\"mord mathnormal\" style=\"margin-right:0.044em;\">z\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003Cspan class=\"mrel\"><\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003C\u002Fspan>\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.6444em;\">\u003C\u002Fspan>\u003Cspan class=\"mord\">0\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan> where the derivative is \u003Cspan class=\"katex\">\u003Cspan class=\"katex-html\" aria-hidden=\"true\">\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.6444em;\">\u003C\u002Fspan>\u003Cspan class=\"mord\">0\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>, so no gradient reaches its weights and it never recovers.\u003C\u002Ffigcaption>","\u003Csvg style=\"width:100%;max-width:390.358px;height:auto\" xmlns=\"http:\u002F\u002Fwww.w3.org\u002F2000\u002Fsvg\" viewBox=\"-75 -75 292.768 116.682\">\u003Cg stroke=\"currentColor\" style=\"stroke-miterlimit:10;stroke-width:.4\">\u003Cg style=\"stroke-width:.8\">\u003Cpath fill=\"none\" d=\"M-22.951 6.714h207.95\"\u002F>\u003Cpath stroke=\"none\" d=\"m187.599 6.714-4.16-2.08 1.56 2.08-1.56 2.08\"\u002F>\u003Cg transform=\"translate(128.925 1.905)\">\u003Cpath d=\"M64.505 6.714L62.727 6.714L62.727 6.417Q63.001 6.417 63.169 6.370Q63.337 6.323 63.337 6.155L63.337 4.019Q63.337 3.804 63.280 3.708Q63.223 3.612 63.110 3.591Q62.997 3.569 62.751 3.569L62.751 3.273L63.950 3.187L63.950 6.155Q63.950 6.323 64.096 6.370Q64.243 6.417 64.505 6.417L64.505 6.714M63.063 1.792Q63.063 1.601 63.198 1.470Q63.333 1.339 63.528 1.339Q63.649 1.339 63.753 1.402Q63.856 1.464 63.919 1.568Q63.981 1.671 63.981 1.792Q63.981 1.987 63.850 2.122Q63.719 2.257 63.528 2.257Q63.329 2.257 63.196 2.124Q63.063 1.991 63.063 1.792M66.934 6.714L65.079 6.714L65.079 6.417Q65.352 6.417 65.520 6.370Q65.688 6.323 65.688 6.155L65.688 4.019Q65.688 3.804 65.626 3.708Q65.563 3.612 65.444 3.591Q65.325 3.569 65.079 3.569L65.079 3.273L66.270 3.187L66.270 3.921Q66.384 3.706 66.577 3.538Q66.770 3.370 67.009 3.278Q67.247 3.187 67.501 3.187Q68.669 3.187 68.669 4.265L68.669 6.155Q68.669 6.323 68.839 6.370Q69.009 6.417 69.278 6.417L69.278 6.714L67.423 6.714L67.423 6.417Q67.696 6.417 67.864 6.370Q68.032 6.323 68.032 6.155L68.032 4.280Q68.032 3.898 67.911 3.669Q67.790 3.441 67.438 3.441Q67.126 3.441 66.872 3.603Q66.618 3.765 66.471 4.034Q66.325 4.304 66.325 4.601L66.325 6.155Q66.325 6.323 66.495 6.370Q66.665 6.417 66.934 6.417L66.934 6.714M71.606 8.265L69.751 8.265L69.751 7.972Q70.020 7.972 70.188 7.927Q70.356 7.882 70.356 7.706L70.356 3.882Q70.356 3.675 70.200 3.622Q70.044 3.569 69.751 3.569L69.751 3.273L70.973 3.187L70.973 3.652Q71.204 3.429 71.518 3.308Q71.833 3.187 72.173 3.187Q72.645 3.187 73.050 3.433Q73.454 3.679 73.686 4.095Q73.919 4.511 73.919 4.987Q73.919 5.362 73.770 5.691Q73.622 6.019 73.352 6.271Q73.083 6.523 72.739 6.657Q72.395 6.792 72.036 6.792Q71.747 6.792 71.475 6.671Q71.204 6.550 70.997 6.339L70.997 7.706Q70.997 7.882 71.165 7.927Q71.333 7.972 71.606 7.972L71.606 8.265M70.997 4.050L70.997 5.890Q71.149 6.179 71.411 6.359Q71.673 6.538 71.981 6.538Q72.266 6.538 72.489 6.400Q72.712 6.261 72.864 6.030Q73.016 5.800 73.094 5.528Q73.173 5.257 73.173 4.987Q73.173 4.655 73.048 4.298Q72.923 3.941 72.675 3.704Q72.427 3.468 72.079 3.468Q71.755 3.468 71.460 3.624Q71.165 3.780 70.997 4.050M75.126 5.761L75.126 4.019Q75.126 3.804 75.063 3.708Q75.001 3.612 74.882 3.591Q74.762 3.569 74.516 3.569L74.516 3.273L75.762 3.187L75.762 5.737L75.762 5.761Q75.762 6.073 75.817 6.235Q75.872 6.398 76.022 6.468Q76.173 6.538 76.493 6.538Q76.923 6.538 77.196 6.200Q77.469 5.862 77.469 5.417L77.469 4.019Q77.469 3.804 77.407 3.708Q77.344 3.612 77.225 3.591Q77.106 3.569 76.860 3.569L76.860 3.273L78.106 3.187L78.106 5.972Q78.106 6.183 78.169 6.278Q78.231 6.374 78.350 6.396Q78.469 6.417 78.716 6.417L78.716 6.714L77.493 6.792L77.493 6.171Q77.325 6.460 77.044 6.626Q76.762 6.792 76.442 6.792Q75.126 6.792 75.126 5.761M79.786 5.753L79.786 3.562L79.083 3.562L79.083 3.308Q79.438 3.308 79.680 3.075Q79.923 2.843 80.034 2.495Q80.145 2.148 80.145 1.792L80.427 1.792L80.427 3.265L81.602 3.265L81.602 3.562L80.427 3.562L80.427 5.737Q80.427 6.058 80.546 6.286Q80.665 6.515 80.946 6.515Q81.126 6.515 81.243 6.392Q81.360 6.269 81.413 6.089Q81.466 5.909 81.466 5.737L81.466 5.265L81.747 5.265L81.747 5.753Q81.747 6.007 81.641 6.247Q81.536 6.487 81.339 6.640Q81.141 6.792 80.884 6.792Q80.567 6.792 80.315 6.669Q80.063 6.546 79.925 6.312Q79.786 6.077 79.786 5.753\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cg style=\"stroke-width:.8\">\u003Cpath fill=\"none\" d=\"M62.407 38.012v-94.14\"\u002F>\u003Cpath stroke=\"none\" d=\"m62.407-58.728-2.08 4.16 2.08-1.56 2.08 1.56\"\u002F>\u003Cg transform=\"translate(-12.514 -70.73)\">\u003Cpath d=\"M62.645 5.019Q62.645 4.515 62.901 4.083Q63.157 3.652 63.593 3.400Q64.028 3.148 64.528 3.148Q64.915 3.148 65.257 3.292Q65.598 3.437 65.860 3.698Q66.122 3.960 66.264 4.296Q66.407 4.632 66.407 5.019Q66.407 5.511 66.143 5.921Q65.880 6.331 65.450 6.562Q65.020 6.792 64.528 6.792Q64.036 6.792 63.602 6.560Q63.169 6.327 62.907 5.919Q62.645 5.511 62.645 5.019M64.528 6.515Q64.985 6.515 65.237 6.292Q65.489 6.069 65.577 5.718Q65.665 5.366 65.665 4.921Q65.665 4.491 65.571 4.153Q65.477 3.816 65.223 3.609Q64.969 3.402 64.528 3.402Q63.880 3.402 63.636 3.818Q63.391 4.234 63.391 4.921Q63.391 5.366 63.479 5.718Q63.567 6.069 63.819 6.292Q64.071 6.515 64.528 6.515M67.575 5.761L67.575 4.019Q67.575 3.804 67.512 3.708Q67.450 3.612 67.331 3.591Q67.212 3.569 66.966 3.569L66.966 3.273L68.212 3.187L68.212 5.737L68.212 5.761Q68.212 6.073 68.266 6.235Q68.321 6.398 68.471 6.468Q68.622 6.538 68.942 6.538Q69.372 6.538 69.645 6.200Q69.919 5.862 69.919 5.417L69.919 4.019Q69.919 3.804 69.856 3.708Q69.794 3.612 69.675 3.591Q69.555 3.569 69.309 3.569L69.309 3.273L70.555 3.187L70.555 5.972Q70.555 6.183 70.618 6.278Q70.680 6.374 70.800 6.396Q70.919 6.417 71.165 6.417L71.165 6.714L69.942 6.792L69.942 6.171Q69.774 6.460 69.493 6.626Q69.212 6.792 68.891 6.792Q67.575 6.792 67.575 5.761M72.235 5.753L72.235 3.562L71.532 3.562L71.532 3.308Q71.887 3.308 72.130 3.075Q72.372 2.843 72.483 2.495Q72.594 2.148 72.594 1.792L72.876 1.792L72.876 3.265L74.052 3.265L74.052 3.562L72.876 3.562L72.876 5.737Q72.876 6.058 72.995 6.286Q73.114 6.515 73.395 6.515Q73.575 6.515 73.692 6.392Q73.809 6.269 73.862 6.089Q73.915 5.909 73.915 5.737L73.915 5.265L74.196 5.265L74.196 5.753Q74.196 6.007 74.091 6.247Q73.985 6.487 73.788 6.640Q73.591 6.792 73.333 6.792Q73.016 6.792 72.764 6.669Q72.512 6.546 72.374 6.312Q72.235 6.077 72.235 5.753M76.798 8.265L74.942 8.265L74.942 7.972Q75.212 7.972 75.380 7.927Q75.548 7.882 75.548 7.706L75.548 3.882Q75.548 3.675 75.391 3.622Q75.235 3.569 74.942 3.569L74.942 3.273L76.165 3.187L76.165 3.652Q76.395 3.429 76.710 3.308Q77.024 3.187 77.364 3.187Q77.837 3.187 78.241 3.433Q78.645 3.679 78.878 4.095Q79.110 4.511 79.110 4.987Q79.110 5.362 78.962 5.691Q78.813 6.019 78.544 6.271Q78.274 6.523 77.930 6.657Q77.587 6.792 77.227 6.792Q76.938 6.792 76.667 6.671Q76.395 6.550 76.188 6.339L76.188 7.706Q76.188 7.882 76.356 7.927Q76.524 7.972 76.798 7.972L76.798 8.265M76.188 4.050L76.188 5.890Q76.341 6.179 76.602 6.359Q76.864 6.538 77.173 6.538Q77.458 6.538 77.680 6.400Q77.903 6.261 78.055 6.030Q78.208 5.800 78.286 5.528Q78.364 5.257 78.364 4.987Q78.364 4.655 78.239 4.298Q78.114 3.941 77.866 3.704Q77.618 3.468 77.270 3.468Q76.946 3.468 76.651 3.624Q76.356 3.780 76.188 4.050M80.317 5.761L80.317 4.019Q80.317 3.804 80.255 3.708Q80.192 3.612 80.073 3.591Q79.954 3.569 79.708 3.569L79.708 3.273L80.954 3.187L80.954 5.737L80.954 5.761Q80.954 6.073 81.009 6.235Q81.063 6.398 81.214 6.468Q81.364 6.538 81.684 6.538Q82.114 6.538 82.387 6.200Q82.661 5.862 82.661 5.417L82.661 4.019Q82.661 3.804 82.598 3.708Q82.536 3.612 82.417 3.591Q82.298 3.569 82.052 3.569L82.052 3.273L83.298 3.187L83.298 5.972Q83.298 6.183 83.360 6.278Q83.423 6.374 83.542 6.396Q83.661 6.417 83.907 6.417L83.907 6.714L82.684 6.792L82.684 6.171Q82.516 6.460 82.235 6.626Q81.954 6.792 81.634 6.792Q80.317 6.792 80.317 5.761M84.977 5.753L84.977 3.562L84.274 3.562L84.274 3.308Q84.630 3.308 84.872 3.075Q85.114 2.843 85.225 2.495Q85.337 2.148 85.337 1.792L85.618 1.792L85.618 3.265L86.794 3.265L86.794 3.562L85.618 3.562L85.618 5.737Q85.618 6.058 85.737 6.286Q85.856 6.515 86.137 6.515Q86.317 6.515 86.434 6.392Q86.552 6.269 86.604 6.089Q86.657 5.909 86.657 5.737L86.657 5.265L86.938 5.265L86.938 5.753Q86.938 6.007 86.833 6.247Q86.727 6.487 86.530 6.640Q86.333 6.792 86.075 6.792Q85.759 6.792 85.507 6.669Q85.255 6.546 85.116 6.312Q84.977 6.077 84.977 5.753\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003Cpath fill=\"none\" stroke=\"var(--tk-accent)\" d=\"M-20.106 6.714h82.513l56.905-56.906\" style=\"stroke-width:1.2\"\u002F>\u003Cpath fill=\"none\" stroke=\"var(--tk-warn)\" d=\"M-20.106 23.216 62.407 6.714l56.905-56.906\" style=\"stroke-width:1.2\"\u002F>\u003Cg fill=\"var(--tk-accent)\" stroke=\"var(--tk-accent)\">\u003Cg fill=\"var(--tk-accent)\" stroke=\"none\" font-family=\"cmtt8\" font-size=\"8\">\u003Cg transform=\"translate(-90.117 -6.476)\">\u003Cpath d=\"M63.981 6.714L62.798 6.714Q62.602 6.691 62.552 6.476L62.552 6.386Q62.602 6.179 62.798 6.155L63.071 6.155L63.071 2.386L62.798 2.386Q62.602 2.362 62.552 2.148L62.552 2.058Q62.602 1.851 62.798 1.827L64.368 1.827Q64.755 1.827 65.110 1.997Q65.466 2.167 65.688 2.478Q65.911 2.788 65.911 3.187Q65.911 3.413 65.829 3.622Q65.747 3.831 65.604 3.999Q65.462 4.167 65.270 4.288Q65.466 4.441 65.581 4.657Q65.696 4.874 65.696 5.112L65.696 5.691Q65.696 6.234 65.895 6.234Q65.942 6.234 65.968 6.105Q65.993 5.976 65.993 5.874Q66.044 5.667 66.239 5.644L66.384 5.644Q66.579 5.663 66.630 5.874L66.630 5.929Q66.630 6.261 66.425 6.527Q66.219 6.792 65.903 6.792Q65.489 6.792 65.272 6.485Q65.055 6.179 65.055 5.745L65.055 5.171Q65.055 4.882 64.846 4.714Q64.637 4.546 64.352 4.546L63.712 4.546L63.712 6.155L63.981 6.155Q64.180 6.179 64.231 6.386L64.231 6.476Q64.180 6.691 63.981 6.714M63.712 2.386L63.712 3.987L64.286 3.987Q64.520 3.987 64.747 3.892Q64.973 3.796 65.122 3.612Q65.270 3.429 65.270 3.187Q65.270 2.944 65.122 2.761Q64.973 2.577 64.749 2.482Q64.524 2.386 64.286 2.386L63.712 2.386M70.165 5.226L67.723 5.226Q67.778 5.503 67.975 5.726Q68.173 5.948 68.450 6.071Q68.727 6.194 69.012 6.194Q69.485 6.194 69.708 5.905Q69.716 5.894 69.772 5.788Q69.829 5.683 69.878 5.640Q69.927 5.597 70.020 5.585L70.165 5.585Q70.356 5.605 70.415 5.819L70.415 5.874Q70.348 6.175 70.118 6.372Q69.887 6.569 69.575 6.661Q69.262 6.753 68.958 6.753Q68.473 6.753 68.034 6.525Q67.594 6.296 67.327 5.896Q67.059 5.495 67.059 5.003L67.059 4.944Q67.059 4.476 67.305 4.073Q67.552 3.671 67.960 3.437Q68.368 3.202 68.837 3.202Q69.341 3.202 69.694 3.425Q70.048 3.648 70.231 4.036Q70.415 4.425 70.415 4.929L70.415 4.987Q70.356 5.202 70.165 5.226M67.731 4.675L69.759 4.675Q69.712 4.265 69.473 4.013Q69.235 3.761 68.837 3.761Q68.442 3.761 68.136 4.023Q67.829 4.284 67.731 4.675M71.122 6.476L71.122 6.386Q71.180 6.179 71.372 6.155L71.739 6.155L71.739 2.386L71.372 2.386Q71.180 2.362 71.122 2.148L71.122 2.058Q71.180 1.851 71.372 1.827L72.899 1.827Q73.094 1.851 73.145 2.058L73.145 2.148Q73.094 2.362 72.899 2.386L72.380 2.386L72.380 6.155L74.212 6.155L74.212 5.515Q74.262 5.308 74.458 5.280L74.602 5.280Q74.802 5.308 74.852 5.515L74.852 6.476Q74.802 6.691 74.602 6.714L71.372 6.714Q71.180 6.691 71.122 6.476M75.712 5.112L75.712 2.386L75.403 2.386Q75.192 2.362 75.153 2.148L75.153 2.058Q75.192 1.851 75.403 1.827L76.665 1.827Q76.872 1.851 76.915 2.058L76.915 2.148Q76.872 2.362 76.665 2.386L76.352 2.386L76.352 5.058Q76.352 5.323 76.458 5.599Q76.563 5.874 76.776 6.054Q76.989 6.234 77.266 6.234Q77.548 6.234 77.757 6.054Q77.966 5.874 78.071 5.599Q78.177 5.323 78.177 5.058L78.177 2.386L77.864 2.386Q77.657 2.362 77.618 2.148L77.618 2.058Q77.657 1.851 77.864 1.827L79.130 1.827Q79.337 1.851 79.376 2.058L79.376 2.148Q79.337 2.362 79.130 2.386L78.817 2.386L78.817 5.112Q78.817 5.534 78.614 5.923Q78.411 6.312 78.053 6.552Q77.696 6.792 77.266 6.792Q76.614 6.792 76.163 6.280Q75.712 5.769 75.712 5.112\" fill=\"var(--tk-accent)\" stroke=\"var(--tk-accent)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(-90.117 -6.476)\">\u003Cpath d=\"M86.770 7.355Q86.239 7.046 85.839 6.558Q85.438 6.069 85.227 5.476Q85.016 4.882 85.016 4.265Q85.016 3.648 85.225 3.060Q85.434 2.472 85.831 1.991Q86.227 1.511 86.762 1.187Q86.833 1.163 86.864 1.163L86.954 1.163Q87.052 1.175 87.118 1.241Q87.184 1.308 87.184 1.409Q87.184 1.554 87.083 1.616Q86.634 1.905 86.307 2.321Q85.981 2.737 85.819 3.230Q85.657 3.722 85.657 4.265Q85.657 4.671 85.753 5.058Q85.848 5.444 86.026 5.780Q86.204 6.116 86.464 6.400Q86.723 6.683 87.071 6.913Q87.184 6.991 87.184 7.128Q87.184 7.226 87.118 7.296Q87.052 7.366 86.954 7.378L86.864 7.378Q86.805 7.378 86.770 7.355M88.223 6.476L88.223 6.386Q88.274 6.179 88.469 6.155L89.352 6.155L89.352 3.827L88.497 3.827Q88.298 3.804 88.247 3.585L88.247 3.499Q88.298 3.288 88.497 3.265L89.352 3.265L89.352 2.812Q89.352 2.347 89.759 2.066Q90.165 1.784 90.645 1.784Q90.958 1.784 91.202 1.905Q91.446 2.027 91.446 2.308Q91.446 2.472 91.337 2.589Q91.227 2.706 91.063 2.706Q90.915 2.706 90.794 2.601Q90.673 2.495 90.673 2.347L90.591 2.347Q90.438 2.347 90.302 2.407Q90.165 2.468 90.079 2.583Q89.993 2.698 89.993 2.843L89.993 3.265L91.024 3.265Q91.219 3.284 91.270 3.499L91.270 3.585Q91.219 3.804 91.024 3.827L89.993 3.827L89.993 6.155L90.872 6.155Q91.067 6.179 91.118 6.386L91.118 6.476Q91.067 6.691 90.872 6.714L88.469 6.714Q88.274 6.691 88.223 6.476M92.598 6.476L92.598 6.386Q92.649 6.179 92.844 6.155L93.950 6.155L93.950 2.386L92.844 2.386Q92.649 2.362 92.598 2.148L92.598 2.058Q92.649 1.851 92.844 1.827L94.341 1.827Q94.532 1.851 94.591 2.058L94.591 6.155L95.692 6.155Q95.891 6.179 95.942 6.386L95.942 6.476Q95.891 6.691 95.692 6.714L92.844 6.714Q92.649 6.691 92.598 6.476M96.802 5.601Q96.802 5.155 97.216 4.898Q97.630 4.640 98.171 4.540Q98.712 4.441 99.219 4.433Q99.219 4.218 99.085 4.066Q98.950 3.913 98.743 3.837Q98.536 3.761 98.325 3.761Q97.981 3.761 97.821 3.784L97.821 3.843Q97.821 4.011 97.702 4.126Q97.583 4.241 97.419 4.241Q97.243 4.241 97.128 4.118Q97.012 3.995 97.012 3.827Q97.012 3.421 97.393 3.312Q97.774 3.202 98.333 3.202Q98.602 3.202 98.870 3.280Q99.137 3.359 99.362 3.509Q99.587 3.659 99.723 3.880Q99.860 4.101 99.860 4.378L99.860 6.097Q99.860 6.155 100.387 6.155Q100.583 6.175 100.634 6.386L100.634 6.476Q100.583 6.691 100.387 6.714L100.243 6.714Q99.899 6.714 99.671 6.667Q99.442 6.620 99.298 6.433Q98.837 6.753 98.130 6.753Q97.794 6.753 97.489 6.612Q97.184 6.472 96.993 6.210Q96.802 5.948 96.802 5.601M97.442 5.609Q97.442 5.882 97.684 6.038Q97.927 6.194 98.212 6.194Q98.430 6.194 98.663 6.136Q98.895 6.077 99.057 5.939Q99.219 5.800 99.219 5.577L99.219 4.987Q98.938 4.987 98.522 5.044Q98.106 5.101 97.774 5.239Q97.442 5.378 97.442 5.609M101.841 5.609L101.841 3.827L101.091 3.827Q100.891 3.804 100.841 3.585L100.841 3.499Q100.891 3.288 101.091 3.265L101.841 3.265L101.841 2.515Q101.891 2.308 102.091 2.280L102.235 2.280Q102.430 2.308 102.481 2.515L102.481 3.265L103.841 3.265Q104.032 3.284 104.091 3.499L104.091 3.585Q104.036 3.804 103.841 3.827L102.481 3.827L102.481 5.577Q102.481 6.194 103.055 6.194Q103.305 6.194 103.469 6.009Q103.634 5.823 103.634 5.577Q103.634 5.484 103.706 5.413Q103.778 5.343 103.880 5.331L104.024 5.331Q104.223 5.355 104.274 5.562L104.274 5.609Q104.274 5.933 104.091 6.196Q103.907 6.460 103.614 6.607Q103.321 6.753 103.001 6.753Q102.489 6.753 102.165 6.443Q101.841 6.132 101.841 5.609M105.927 7.378L105.841 7.378Q105.735 7.366 105.667 7.294Q105.598 7.222 105.598 7.128Q105.598 6.987 105.704 6.921Q106.040 6.706 106.309 6.417Q106.579 6.128 106.762 5.780Q106.946 5.433 107.036 5.054Q107.126 4.675 107.126 4.265Q107.126 3.726 106.962 3.234Q106.798 2.741 106.479 2.327Q106.161 1.913 105.719 1.624Q105.598 1.542 105.598 1.409Q105.598 1.312 105.667 1.243Q105.735 1.175 105.841 1.163L105.927 1.163Q105.981 1.163 106.016 1.187Q106.403 1.409 106.743 1.757Q107.083 2.105 107.303 2.499Q107.524 2.894 107.645 3.339Q107.766 3.784 107.766 4.265Q107.766 4.749 107.645 5.200Q107.524 5.652 107.300 6.048Q107.075 6.444 106.751 6.778Q106.427 7.112 106.024 7.355Q105.954 7.378 105.927 7.378\" fill=\"var(--tk-accent)\" stroke=\"var(--tk-accent)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003Cg fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\">\u003Cg fill=\"var(--tk-warn)\" stroke=\"none\" font-family=\"cmtt8\" font-size=\"8\">\u003Cg transform=\"translate(-121.344 21.039)\">\u003Cpath d=\"M62.630 6.476L62.630 6.386Q62.688 6.179 62.880 6.155L63.247 6.155L63.247 2.386L62.880 2.386Q62.688 2.362 62.630 2.148L62.630 2.058Q62.688 1.851 62.880 1.827L64.407 1.827Q64.602 1.851 64.653 2.058L64.653 2.148Q64.602 2.362 64.407 2.386L63.887 2.386L63.887 6.155L65.719 6.155L65.719 5.515Q65.770 5.308 65.966 5.280L66.110 5.280Q66.309 5.308 66.360 5.515L66.360 6.476Q66.309 6.691 66.110 6.714L62.880 6.714Q62.688 6.691 62.630 6.476M70.165 5.226L67.723 5.226Q67.778 5.503 67.975 5.726Q68.173 5.948 68.450 6.071Q68.727 6.194 69.012 6.194Q69.485 6.194 69.708 5.905Q69.716 5.894 69.772 5.788Q69.829 5.683 69.878 5.640Q69.927 5.597 70.020 5.585L70.165 5.585Q70.356 5.605 70.415 5.819L70.415 5.874Q70.348 6.175 70.118 6.372Q69.887 6.569 69.575 6.661Q69.262 6.753 68.958 6.753Q68.473 6.753 68.034 6.525Q67.594 6.296 67.327 5.896Q67.059 5.495 67.059 5.003L67.059 4.944Q67.059 4.476 67.305 4.073Q67.552 3.671 67.960 3.437Q68.368 3.202 68.837 3.202Q69.341 3.202 69.694 3.425Q70.048 3.648 70.231 4.036Q70.415 4.425 70.415 4.929L70.415 4.987Q70.356 5.202 70.165 5.226M67.731 4.675L69.759 4.675Q69.712 4.265 69.473 4.013Q69.235 3.761 68.837 3.761Q68.442 3.761 68.136 4.023Q67.829 4.284 67.731 4.675M71.305 5.601Q71.305 5.155 71.719 4.898Q72.134 4.640 72.675 4.540Q73.216 4.441 73.723 4.433Q73.723 4.218 73.589 4.066Q73.454 3.913 73.247 3.837Q73.040 3.761 72.829 3.761Q72.485 3.761 72.325 3.784L72.325 3.843Q72.325 4.011 72.206 4.126Q72.087 4.241 71.923 4.241Q71.747 4.241 71.632 4.118Q71.516 3.995 71.516 3.827Q71.516 3.421 71.897 3.312Q72.278 3.202 72.837 3.202Q73.106 3.202 73.374 3.280Q73.641 3.359 73.866 3.509Q74.091 3.659 74.227 3.880Q74.364 4.101 74.364 4.378L74.364 6.097Q74.364 6.155 74.891 6.155Q75.087 6.175 75.137 6.386L75.137 6.476Q75.087 6.691 74.891 6.714L74.747 6.714Q74.403 6.714 74.175 6.667Q73.946 6.620 73.802 6.433Q73.341 6.753 72.634 6.753Q72.298 6.753 71.993 6.612Q71.688 6.472 71.497 6.210Q71.305 5.948 71.305 5.601M71.946 5.609Q71.946 5.882 72.188 6.038Q72.430 6.194 72.716 6.194Q72.934 6.194 73.167 6.136Q73.399 6.077 73.561 5.939Q73.723 5.800 73.723 5.577L73.723 4.987Q73.442 4.987 73.026 5.044Q72.610 5.101 72.278 5.239Q71.946 5.378 71.946 5.609M75.290 6.476L75.290 6.386Q75.341 6.179 75.536 6.155L76.001 6.155L76.001 2.386L75.536 2.386Q75.341 2.362 75.290 2.148L75.290 2.058Q75.341 1.851 75.536 1.827L76.282 1.827Q76.477 1.847 76.528 2.058L76.528 4.882L77.673 3.827L77.376 3.827Q77.169 3.804 77.130 3.585L77.130 3.499Q77.169 3.288 77.376 3.265L78.825 3.265Q79.028 3.288 79.075 3.499L79.075 3.585Q79.032 3.804 78.825 3.827L78.458 3.827L77.512 4.691L78.665 6.155L79.001 6.155Q79.208 6.179 79.251 6.386L79.251 6.476Q79.208 6.691 79.001 6.714L77.833 6.714Q77.637 6.691 77.587 6.476L77.587 6.386Q77.637 6.179 77.833 6.155L77.993 6.155L77.130 5.050L76.528 5.601L76.528 6.155L76.993 6.155Q77.184 6.179 77.243 6.386L77.243 6.476Q77.184 6.691 76.993 6.714L75.536 6.714Q75.341 6.691 75.290 6.476M79.735 7.835Q79.735 7.726 79.786 7.634Q79.837 7.542 79.928 7.491Q80.020 7.441 80.126 7.441Q80.235 7.441 80.327 7.491Q80.419 7.542 80.469 7.634Q80.520 7.726 80.520 7.835L80.376 7.835Q80.376 7.972 80.399 7.972Q80.657 7.972 80.844 7.780Q81.032 7.589 81.118 7.323L81.329 6.714L80.192 3.827L79.864 3.827Q79.669 3.804 79.614 3.585L79.614 3.499Q79.673 3.288 79.864 3.265L81.024 3.265Q81.219 3.288 81.270 3.499L81.270 3.585Q81.219 3.804 81.024 3.827L80.759 3.827Q81.094 4.675 81.339 5.329Q81.583 5.984 81.583 6.073L81.591 6.073Q81.591 6.015 81.682 5.722Q81.774 5.429 81.968 4.845Q82.161 4.261 82.302 3.827L82.024 3.827Q81.813 3.804 81.774 3.585L81.774 3.499Q81.825 3.284 82.024 3.265L83.177 3.265Q83.384 3.288 83.423 3.499L83.423 3.585Q83.384 3.804 83.177 3.827L82.856 3.827L81.680 7.323Q81.512 7.823 81.186 8.177Q80.860 8.530 80.399 8.530Q80.126 8.530 79.930 8.319Q79.735 8.109 79.735 7.835\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(-121.344 21.039)\">\u003Cpath d=\"M91.020 7.355Q90.489 7.046 90.089 6.558Q89.688 6.069 89.477 5.476Q89.266 4.882 89.266 4.265Q89.266 3.648 89.475 3.060Q89.684 2.472 90.081 1.991Q90.477 1.511 91.012 1.187Q91.083 1.163 91.114 1.163L91.204 1.163Q91.302 1.175 91.368 1.241Q91.434 1.308 91.434 1.409Q91.434 1.554 91.333 1.616Q90.884 1.905 90.557 2.321Q90.231 2.737 90.069 3.230Q89.907 3.722 89.907 4.265Q89.907 4.671 90.003 5.058Q90.098 5.444 90.276 5.780Q90.454 6.116 90.714 6.400Q90.973 6.683 91.321 6.913Q91.434 6.991 91.434 7.128Q91.434 7.226 91.368 7.296Q91.302 7.366 91.204 7.378L91.114 7.378Q91.055 7.378 91.020 7.355M92.739 6.515L92.739 5.601Q92.766 5.394 92.977 5.370L93.145 5.370Q93.309 5.394 93.368 5.554Q93.571 6.194 94.298 6.194Q94.505 6.194 94.733 6.159Q94.962 6.124 95.130 6.009Q95.298 5.894 95.298 5.691Q95.298 5.480 95.075 5.366Q94.852 5.253 94.579 5.210L93.880 5.097Q92.739 4.886 92.739 4.163Q92.739 3.874 92.884 3.685Q93.028 3.495 93.268 3.388Q93.509 3.280 93.764 3.241Q94.020 3.202 94.298 3.202Q94.548 3.202 94.741 3.232Q94.934 3.261 95.098 3.339Q95.177 3.222 95.305 3.202L95.384 3.202Q95.481 3.214 95.544 3.277Q95.606 3.339 95.618 3.433L95.618 4.140Q95.606 4.234 95.544 4.300Q95.481 4.366 95.384 4.378L95.216 4.378Q95.122 4.366 95.055 4.300Q94.989 4.234 94.977 4.140Q94.977 3.761 94.282 3.761Q93.934 3.761 93.616 3.843Q93.298 3.925 93.298 4.171Q93.298 4.437 93.969 4.546L94.673 4.667Q95.157 4.749 95.507 4.997Q95.856 5.245 95.856 5.691Q95.856 6.081 95.620 6.323Q95.384 6.566 95.034 6.659Q94.684 6.753 94.298 6.753Q93.719 6.753 93.321 6.499Q93.251 6.624 93.202 6.681Q93.153 6.737 93.048 6.753L92.977 6.753Q92.762 6.730 92.739 6.515M96.848 6.476L96.848 6.386Q96.899 6.179 97.094 6.155L98.200 6.155L98.200 2.386L97.094 2.386Q96.899 2.362 96.848 2.148L96.848 2.058Q96.899 1.851 97.094 1.827L98.591 1.827Q98.782 1.851 98.841 2.058L98.841 6.155L99.942 6.155Q100.141 6.179 100.192 6.386L100.192 6.476Q100.141 6.691 99.942 6.714L97.094 6.714Q96.899 6.691 96.848 6.476M102.766 6.753Q102.294 6.753 101.909 6.509Q101.524 6.265 101.302 5.855Q101.079 5.444 101.079 4.987Q101.079 4.644 101.204 4.321Q101.329 3.999 101.559 3.745Q101.790 3.491 102.096 3.347Q102.403 3.202 102.766 3.202Q103.130 3.202 103.442 3.349Q103.755 3.495 103.977 3.741Q104.200 3.987 104.327 4.308Q104.454 4.628 104.454 4.987Q104.454 5.444 104.229 5.857Q104.005 6.269 103.620 6.511Q103.235 6.753 102.766 6.753M102.766 6.194Q103.231 6.194 103.522 5.800Q103.813 5.405 103.813 4.921Q103.813 4.628 103.678 4.360Q103.544 4.093 103.303 3.927Q103.063 3.761 102.766 3.761Q102.462 3.761 102.223 3.927Q101.985 4.093 101.850 4.360Q101.716 4.628 101.716 4.921Q101.716 5.402 102.009 5.798Q102.302 6.194 102.766 6.194M104.962 8.257L104.962 8.171Q105.005 7.952 105.212 7.929L105.634 7.929L105.634 3.827L105.212 3.827Q105.005 3.804 104.962 3.585L104.962 3.499Q105.009 3.288 105.212 3.265L106.028 3.265Q106.223 3.284 106.274 3.499L106.274 3.569Q106.485 3.402 106.749 3.314Q107.012 3.226 107.282 3.226Q107.622 3.226 107.919 3.370Q108.216 3.515 108.430 3.767Q108.645 4.019 108.761 4.331Q108.876 4.644 108.876 4.987Q108.876 5.452 108.649 5.862Q108.423 6.273 108.036 6.513Q107.649 6.753 107.180 6.753Q106.661 6.753 106.274 6.386L106.274 7.929L106.700 7.929Q106.907 7.952 106.946 8.171L106.946 8.257Q106.907 8.468 106.700 8.491L105.212 8.491Q105.009 8.468 104.962 8.257M107.137 6.194Q107.446 6.194 107.696 6.025Q107.946 5.855 108.091 5.573Q108.235 5.292 108.235 4.987Q108.235 4.698 108.110 4.419Q107.985 4.140 107.753 3.962Q107.520 3.784 107.219 3.784Q106.899 3.784 106.637 3.970Q106.376 4.155 106.274 4.456L106.274 5.308Q106.364 5.675 106.585 5.935Q106.805 6.194 107.137 6.194M112.649 5.226L110.208 5.226Q110.262 5.503 110.460 5.726Q110.657 5.948 110.934 6.071Q111.212 6.194 111.497 6.194Q111.969 6.194 112.192 5.905Q112.200 5.894 112.257 5.788Q112.313 5.683 112.362 5.640Q112.411 5.597 112.505 5.585L112.649 5.585Q112.841 5.605 112.899 5.819L112.899 5.874Q112.833 6.175 112.602 6.372Q112.372 6.569 112.059 6.661Q111.747 6.753 111.442 6.753Q110.958 6.753 110.518 6.525Q110.079 6.296 109.811 5.896Q109.544 5.495 109.544 5.003L109.544 4.944Q109.544 4.476 109.790 4.073Q110.036 3.671 110.444 3.437Q110.852 3.202 111.321 3.202Q111.825 3.202 112.178 3.425Q112.532 3.648 112.716 4.036Q112.899 4.425 112.899 4.929L112.899 4.987Q112.841 5.202 112.649 5.226M110.216 4.675L112.243 4.675Q112.196 4.265 111.958 4.013Q111.719 3.761 111.321 3.761Q110.927 3.761 110.620 4.023Q110.313 4.284 110.216 4.675\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(-121.344 21.039)\">\u003Cpath d=\"M118.064 5.601Q118.064 5.155 118.478 4.898Q118.892 4.640 119.433 4.540Q119.974 4.441 120.482 4.433Q120.482 4.218 120.347 4.066Q120.213 3.913 120.006 3.837Q119.799 3.761 119.588 3.761Q119.244 3.761 119.084 3.784L119.084 3.843Q119.084 4.011 118.965 4.126Q118.846 4.241 118.681 4.241Q118.506 4.241 118.390 4.118Q118.275 3.995 118.275 3.827Q118.275 3.421 118.656 3.312Q119.037 3.202 119.596 3.202Q119.865 3.202 120.133 3.280Q120.400 3.359 120.625 3.509Q120.849 3.659 120.986 3.880Q121.123 4.101 121.123 4.378L121.123 6.097Q121.123 6.155 121.650 6.155Q121.846 6.175 121.896 6.386L121.896 6.476Q121.846 6.691 121.650 6.714L121.506 6.714Q121.162 6.714 120.933 6.667Q120.705 6.620 120.560 6.433Q120.099 6.753 119.392 6.753Q119.056 6.753 118.752 6.612Q118.447 6.472 118.256 6.210Q118.064 5.948 118.064 5.601M118.705 5.609Q118.705 5.882 118.947 6.038Q119.189 6.194 119.474 6.194Q119.693 6.194 119.926 6.136Q120.158 6.077 120.320 5.939Q120.482 5.800 120.482 5.577L120.482 4.987Q120.201 4.987 119.785 5.044Q119.369 5.101 119.037 5.239Q118.705 5.378 118.705 5.609M122.943 7.378L122.857 7.378Q122.752 7.366 122.683 7.294Q122.615 7.222 122.615 7.128Q122.615 6.987 122.721 6.921Q123.056 6.706 123.326 6.417Q123.596 6.128 123.779 5.780Q123.963 5.433 124.053 5.054Q124.142 4.675 124.142 4.265Q124.142 3.726 123.978 3.234Q123.814 2.741 123.496 2.327Q123.178 1.913 122.736 1.624Q122.615 1.542 122.615 1.409Q122.615 1.312 122.683 1.243Q122.752 1.175 122.857 1.163L122.943 1.163Q122.998 1.163 123.033 1.187Q123.420 1.409 123.760 1.757Q124.099 2.105 124.320 2.499Q124.541 2.894 124.662 3.339Q124.783 3.784 124.783 4.265Q124.783 4.749 124.662 5.200Q124.541 5.652 124.316 6.048Q124.092 6.444 123.767 6.778Q123.443 7.112 123.041 7.355Q122.971 7.378 122.943 7.378\" fill=\"var(--tk-warn)\" stroke=\"var(--tk-warn)\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003Cg stroke=\"none\" font-family=\"cmr8\" font-size=\"8\">\u003Cg transform=\"translate(71.82 -37.834)\">\u003Cpath d=\"M63.559 6.714L63.278 6.714L63.278 1.995Q63.278 1.780 63.216 1.685Q63.153 1.589 63.036 1.568Q62.919 1.546 62.673 1.546L62.673 1.249L63.895 1.163L63.895 3.652Q64.372 3.187 65.071 3.187Q65.552 3.187 65.960 3.431Q66.368 3.675 66.604 4.089Q66.841 4.503 66.841 4.987Q66.841 5.362 66.692 5.691Q66.544 6.019 66.274 6.271Q66.005 6.523 65.661 6.657Q65.317 6.792 64.958 6.792Q64.637 6.792 64.339 6.644Q64.040 6.495 63.833 6.234L63.559 6.714M63.919 4.042L63.919 5.882Q64.071 6.179 64.331 6.359Q64.591 6.538 64.903 6.538Q65.329 6.538 65.596 6.319Q65.864 6.101 65.979 5.755Q66.094 5.409 66.094 4.987Q66.094 4.339 65.846 3.890Q65.598 3.441 65.001 3.441Q64.665 3.441 64.376 3.599Q64.087 3.757 63.919 4.042\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(71.82 -37.834)\">\u003Cpath d=\"M67.603 5.019Q67.603 4.515 67.859 4.083Q68.115 3.652 68.551 3.400Q68.986 3.148 69.486 3.148Q69.873 3.148 70.215 3.292Q70.556 3.437 70.818 3.698Q71.080 3.960 71.222 4.296Q71.365 4.632 71.365 5.019Q71.365 5.511 71.101 5.921Q70.838 6.331 70.408 6.562Q69.978 6.792 69.486 6.792Q68.994 6.792 68.560 6.560Q68.127 6.327 67.865 5.919Q67.603 5.511 67.603 5.019M69.486 6.515Q69.943 6.515 70.195 6.292Q70.447 6.069 70.535 5.718Q70.623 5.366 70.623 4.921Q70.623 4.491 70.529 4.153Q70.435 3.816 70.181 3.609Q69.927 3.402 69.486 3.402Q68.838 3.402 68.594 3.818Q68.349 4.234 68.349 4.921Q68.349 5.366 68.437 5.718Q68.525 6.069 68.777 6.292Q69.029 6.515 69.486 6.515M72.474 5.753L72.474 3.562L71.771 3.562L71.771 3.308Q72.127 3.308 72.369 3.075Q72.611 2.843 72.722 2.495Q72.834 2.148 72.834 1.792L73.115 1.792L73.115 3.265L74.291 3.265L74.291 3.562L73.115 3.562L73.115 5.737Q73.115 6.058 73.234 6.286Q73.353 6.515 73.635 6.515Q73.814 6.515 73.931 6.392Q74.049 6.269 74.101 6.089Q74.154 5.909 74.154 5.737L74.154 5.265L74.435 5.265L74.435 5.753Q74.435 6.007 74.330 6.247Q74.224 6.487 74.027 6.640Q73.830 6.792 73.572 6.792Q73.256 6.792 73.004 6.669Q72.752 6.546 72.613 6.312Q72.474 6.077 72.474 5.753M77.084 6.714L75.228 6.714L75.228 6.417Q75.502 6.417 75.670 6.370Q75.838 6.323 75.838 6.155L75.838 1.995Q75.838 1.780 75.775 1.685Q75.713 1.589 75.594 1.568Q75.474 1.546 75.228 1.546L75.228 1.249L76.451 1.163L76.451 3.866Q76.576 3.655 76.763 3.505Q76.951 3.355 77.177 3.271Q77.404 3.187 77.650 3.187Q78.818 3.187 78.818 4.265L78.818 6.155Q78.818 6.323 78.988 6.370Q79.158 6.417 79.427 6.417L79.427 6.714L77.572 6.714L77.572 6.417Q77.845 6.417 78.013 6.370Q78.181 6.323 78.181 6.155L78.181 4.280Q78.181 3.898 78.060 3.669Q77.939 3.441 77.588 3.441Q77.275 3.441 77.021 3.603Q76.767 3.765 76.621 4.034Q76.474 4.304 76.474 4.601L76.474 6.155Q76.474 6.323 76.644 6.370Q76.814 6.417 77.084 6.417L77.084 6.714M80.353 6.249Q80.353 6.066 80.490 5.929Q80.627 5.792 80.818 5.792Q81.010 5.792 81.142 5.925Q81.275 6.058 81.275 6.249Q81.275 6.448 81.142 6.581Q81.010 6.714 80.818 6.714Q80.627 6.714 80.490 6.577Q80.353 6.441 80.353 6.249M80.353 3.722Q80.353 3.538 80.490 3.402Q80.627 3.265 80.818 3.265Q81.010 3.265 81.142 3.398Q81.275 3.530 81.275 3.722Q81.275 3.921 81.142 4.054Q81.010 4.187 80.818 4.187Q80.627 4.187 80.490 4.050Q80.353 3.913 80.353 3.722\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(71.82 -37.834)\">\u003Cpath d=\"M86.063 6.706L86.063 5.484Q86.063 5.456 86.094 5.425Q86.126 5.394 86.149 5.394L86.255 5.394Q86.325 5.394 86.341 5.456Q86.403 5.777 86.542 6.017Q86.680 6.257 86.913 6.398Q87.145 6.538 87.454 6.538Q87.692 6.538 87.901 6.478Q88.110 6.417 88.247 6.269Q88.384 6.120 88.384 5.874Q88.384 5.620 88.173 5.454Q87.962 5.288 87.692 5.234L87.071 5.120Q86.665 5.042 86.364 4.786Q86.063 4.530 86.063 4.155Q86.063 3.788 86.264 3.566Q86.466 3.343 86.790 3.245Q87.114 3.148 87.454 3.148Q87.919 3.148 88.216 3.355L88.438 3.171Q88.462 3.148 88.493 3.148L88.544 3.148Q88.575 3.148 88.602 3.175Q88.630 3.202 88.630 3.234L88.630 4.218Q88.630 4.249 88.604 4.278Q88.579 4.308 88.544 4.308L88.438 4.308Q88.403 4.308 88.376 4.280Q88.348 4.253 88.348 4.218Q88.348 3.819 88.096 3.599Q87.844 3.378 87.446 3.378Q87.091 3.378 86.807 3.501Q86.524 3.624 86.524 3.929Q86.524 4.148 86.725 4.280Q86.927 4.413 87.173 4.456L87.798 4.569Q88.227 4.659 88.536 4.956Q88.844 5.253 88.844 5.667Q88.844 6.237 88.446 6.515Q88.048 6.792 87.454 6.792Q86.903 6.792 86.552 6.456L86.255 6.769Q86.231 6.792 86.196 6.792L86.149 6.792Q86.126 6.792 86.094 6.761Q86.063 6.730 86.063 6.706M91.286 6.714L89.454 6.714L89.454 6.417Q89.727 6.417 89.895 6.370Q90.063 6.323 90.063 6.155L90.063 1.995Q90.063 1.780 90.001 1.685Q89.938 1.589 89.819 1.568Q89.700 1.546 89.454 1.546L89.454 1.249L90.677 1.163L90.677 6.155Q90.677 6.323 90.844 6.370Q91.012 6.417 91.286 6.417L91.286 6.714M91.731 5.019Q91.731 4.515 91.987 4.083Q92.243 3.652 92.678 3.400Q93.114 3.148 93.614 3.148Q94.001 3.148 94.343 3.292Q94.684 3.437 94.946 3.698Q95.208 3.960 95.350 4.296Q95.493 4.632 95.493 5.019Q95.493 5.511 95.229 5.921Q94.966 6.331 94.536 6.562Q94.106 6.792 93.614 6.792Q93.122 6.792 92.688 6.560Q92.255 6.327 91.993 5.919Q91.731 5.511 91.731 5.019M93.614 6.515Q94.071 6.515 94.323 6.292Q94.575 6.069 94.663 5.718Q94.751 5.366 94.751 4.921Q94.751 4.491 94.657 4.153Q94.563 3.816 94.309 3.609Q94.055 3.402 93.614 3.402Q92.966 3.402 92.721 3.818Q92.477 4.234 92.477 4.921Q92.477 5.366 92.565 5.718Q92.653 6.069 92.905 6.292Q93.157 6.515 93.614 6.515M97.860 8.265L96.005 8.265L96.005 7.972Q96.274 7.972 96.442 7.927Q96.610 7.882 96.610 7.706L96.610 3.882Q96.610 3.675 96.454 3.622Q96.298 3.569 96.005 3.569L96.005 3.273L97.227 3.187L97.227 3.652Q97.458 3.429 97.772 3.308Q98.087 3.187 98.427 3.187Q98.899 3.187 99.303 3.433Q99.708 3.679 99.940 4.095Q100.173 4.511 100.173 4.987Q100.173 5.362 100.024 5.691Q99.876 6.019 99.606 6.271Q99.337 6.523 98.993 6.657Q98.649 6.792 98.290 6.792Q98.001 6.792 97.729 6.671Q97.458 6.550 97.251 6.339L97.251 7.706Q97.251 7.882 97.419 7.927Q97.587 7.972 97.860 7.972L97.860 8.265M97.251 4.050L97.251 5.890Q97.403 6.179 97.665 6.359Q97.927 6.538 98.235 6.538Q98.520 6.538 98.743 6.400Q98.966 6.261 99.118 6.030Q99.270 5.800 99.348 5.528Q99.427 5.257 99.427 4.987Q99.427 4.655 99.302 4.298Q99.177 3.941 98.928 3.704Q98.680 3.468 98.333 3.468Q98.009 3.468 97.714 3.624Q97.419 3.780 97.251 4.050\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(71.82 -37.834)\">\u003Cpath d=\"M100.943 4.960Q100.943 4.480 101.176 4.064Q101.408 3.648 101.818 3.398Q102.228 3.148 102.705 3.148Q103.435 3.148 103.834 3.589Q104.232 4.030 104.232 4.761Q104.232 4.866 104.139 4.890L101.689 4.890L101.689 4.960Q101.689 5.370 101.810 5.726Q101.932 6.081 102.203 6.298Q102.475 6.515 102.904 6.515Q103.267 6.515 103.564 6.286Q103.861 6.058 103.963 5.706Q103.971 5.659 104.057 5.644L104.139 5.644Q104.232 5.671 104.232 5.753Q104.232 5.761 104.225 5.792Q104.162 6.019 104.023 6.202Q103.885 6.386 103.693 6.519Q103.502 6.652 103.283 6.722Q103.064 6.792 102.826 6.792Q102.455 6.792 102.117 6.655Q101.779 6.519 101.512 6.267Q101.244 6.015 101.094 5.675Q100.943 5.335 100.943 4.960M101.697 4.652L103.658 4.652Q103.658 4.347 103.557 4.056Q103.455 3.765 103.238 3.583Q103.021 3.402 102.705 3.402Q102.404 3.402 102.174 3.589Q101.943 3.777 101.820 4.068Q101.697 4.359 101.697 4.652\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003Cg transform=\"translate(71.82 -37.834)\">\u003Cpath d=\"M110.910 6.714L108.117 6.714L108.117 6.417Q109.179 6.417 109.179 6.155L109.179 1.987Q108.750 2.202 108.070 2.202L108.070 1.905Q109.089 1.905 109.605 1.394L109.750 1.394Q109.824 1.413 109.843 1.491L109.843 6.155Q109.843 6.417 110.910 6.417\" fill=\"currentColor\" stroke=\"currentColor\" class=\"tikz-text\" style=\"stroke-width:0.240\"\u002F>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fg>\u003C\u002Fsvg>\u003Cfigcaption class=\"tikz-cap\">ReLU versus Leaky ReLU. Both are the identity for \u003Cspan class=\"katex\">\u003Cspan class=\"katex-html\" aria-hidden=\"true\">\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.5782em;vertical-align:-0.0391em;\">\u003C\u002Fspan>\u003Cspan class=\"mord mathnormal\" style=\"margin-right:0.044em;\">z\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003Cspan class=\"mrel\">>\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003C\u002Fspan>\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.6444em;\">\u003C\u002Fspan>\u003Cspan class=\"mord\">0\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>; for \u003Cspan class=\"katex\">\u003Cspan class=\"katex-html\" aria-hidden=\"true\">\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.5782em;vertical-align:-0.0391em;\">\u003C\u002Fspan>\u003Cspan class=\"mord mathnormal\" style=\"margin-right:0.044em;\">z\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003Cspan class=\"mrel\"><\u003C\u002Fspan>\u003Cspan class=\"mspace\" style=\"margin-right:0.2778em;\">\u003C\u002Fspan>\u003C\u002Fspan>\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.6444em;\">\u003C\u002Fspan>\u003Cspan class=\"mord\">0\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan> ReLU is flat (zero gradient) while Leaky ReLU keeps a small negative slope \u003Cspan class=\"katex\">\u003Cspan class=\"katex-html\" aria-hidden=\"true\">\u003Cspan class=\"base\">\u003Cspan class=\"strut\" style=\"height:0.4306em;\">\u003C\u002Fspan>\u003Cspan class=\"mord mathnormal\" style=\"margin-right:0.0037em;\">α\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>\u003C\u002Fspan>.\u003C\u002Ffigcaption>",1785117762584]