Applications: Leontief Economics and Computer Graphics

The Leontief input–output model, recognized with a Nobel Prize in economics, balances supply against demand across an entire economy through a single matrix equation, solved with the inverse and a series expansion in powers of a matrix. Computer graphics animates figures by multiplying their vertex coordinates by transformation matrices, and a change of coordinates lets even translation and perspective, which are not linear, join the same matrix arithmetic.

The Leontief input–output model

Divide an economy into $n$ sectors that produce goods or services. Let the production vector $x\in {\mathbb{R}}^n$ list each sector's yearly output, and let the final demand vector $d\in {\mathbb{R}}^n$ list what a non-productive open sector (consumers, government, exports) demands. Producing goods itself consumes goods: making output creates intermediate demand for inputs. The model seeks a production level that exactly balances the two demands,

The model's assumption is that each sector has a fixed unit consumption vector: the inputs, from every sector, needed per unit of its output. Collecting these as columns forms the consumption matrix $C=[c_1\cdots c_n]$. If sector $j$ plans output $x_j$, its intermediate demand is $x_j{c}_j$, so the total intermediate demand is

Solving the model

When a solution exists

The economics guarantees the equation is solvable and the answer is meaningful.

Nonnegativity matters: a production vector with a negative entry would be economically meaningless. The column-sum condition is the natural one — a sector that consumed a full unit or more of inputs per unit of output could never run a surplus.

The inverse as a series

The proof of the Leontief solvability theorem also yields a way to compute $(I-C{)}^{-1}$ and an economic reading of it. Imagine meeting the demand $d$ in rounds. First produce $d$; this needs inputs $Cd$. Producing $Cd$ needs $C(Cd)=C^{2}d$; producing that needs $C^{3}d$; and so on. The total production is

The bracketed sum is $(I-C{)}^{-1}$, justified by the finite identity

When the column sums of $C$ are all below $1$, the powers $C^m$ approach the zero matrix, so $C^{m+1}\to 0$ and the right side tends to $I$. Thus

and in practice the powers shrink fast enough that a few terms suffice.

The entries of $(I-C{)}^{-1}$ carry economic meaning: column $j$ lists the extra output each sector must produce to satisfy a one-unit increase in final demand for sector $j$. If final demand shifts from $d$ to $d+\Delta d$, production shifts by $\Delta x=(I-C{)}^{-1}\Delta d$, so the inverse is the multiplier that propagates a demand change through the whole economy.

Computer graphics

A graphical object is stored as a list of vertices, connected by line segments. Because the standard transformations map line segments to line segments, moving an object reduces to transforming its vertices and reconnecting them.

For example, the shear $A=\left[\begin{array}{cc}1& .25\\ 0& 1\end{array}\right]$ slants a figure by shifting each point's $x$-coordinate in proportion to its height, turning an upright letter into an italic one; following it with the $x$-scaling $S=\left[\begin{array}{cc}.75& 0\\ 0& 1\end{array}\right]$ compensates for the added width. The composite transformation is the single matrix $SA$.

Homogeneous coordinates

A translation $(x,y)\mapsto (x+h,y+k)$ is not linear (it moves the origin), so it is not multiplication by a $2\times 2$ matrix. To address this, lift the plane one level.

With the extra coordinate, translation becomes a matrix product:

Any linear map on ${\mathbb{R}}^2$ with matrix $A$ becomes the block matrix $\left[\begin{array}{cc}A& 0\\ 0& 1\end{array}\right]$ in homogeneous coordinates, so rotations, reflections, and scalings keep their familiar forms while translation joins them. The three basic $3\times 3$ transforms:

Composite transformations

Screen motion usually chains several transforms, and composition is matrix multiplication in reverse order of application.

Building the composite once and reusing it on every vertex is why graphics hardware multiplies small matrices at high speed.

Three dimensions and perspective

The same lift works one dimension higher: $(x,y,z)$ gets homogeneous coordinates $(x,y,z,1)$, and $4\times 4$ matrices carry out rotation about an axis, translation, and scaling in space. More generally $(X,Y,Z,H)$ with $H\ne 0$ represents the point $(X/H,Y/H,Z/H)$, so every nonzero scalar multiple of $(x,y,z,1)$ names the same point. That freedom is what makes perspective a matrix operation.

To display a three-dimensional scene on a flat screen, project each point onto a viewing plane through the viewer's eye. Put the screen at the $xy$-plane and the eye at $(0,0,d)$ on the positive $z$-axis. A point $(x,y,z)$ maps to the image point $(x^{\ast },y^{\ast },0)$ where the line from the eye through the point crosses the screen. Similar triangles give

Scaling the image coordinates by $1-z/d$, the point $(x,y,0,1-z/d)$ is a homogeneous representative of the projection, so a single matrix performs it:

Dividing the first three entries by the fourth recovers the screen coordinates $(x^{\ast },y^{\ast },0)$. Continuous, realistic animation is a stream of such $4\times 4$ products — rotations, translations, and projections — which is why graphics processors implement $4\times 4$ matrix arithmetic directly in hardware.

Common ground

Both models reduce to matrix operations. Leontief economics solves $(I-C)x=d$ with the inverse, read as a demand multiplier and computed as a series in powers of $C$; computer graphics composes transformation matrices, with homogeneous coordinates extending matrix multiplication to translation and perspective. Powers of a matrix and its inverse reappear in eigenvalues and the dynamical systems they govern.¹