Solution Sets and Applied Linear Systems

The reduced echelon form delivers a general solution as a list of formulas. Written instead as a vector equation — the parametric vector form — that solution exposes its geometry. Solution sets are flat objects (lines, planes, and their higher-dimensional analogues), passing through the origin when the system is homogeneous and shifted off it when it is not. The same structure organizes applied systems that carry many solutions.

Homogeneous systems

A system is homogeneous if it has the form $A\mathbf{x}=0$. Every such system has the trivial solution $\mathbf{x}=0$, so the only question of interest is whether it has any others.

This is the existence and uniqueness theorem applied to a system that is automatically consistent: a solution always exists, so only uniqueness is in question, and uniqueness fails exactly when a variable is free.

Parametric vector form

Every solution is a scalar multiple of the single vector $\mathbf{v}$, so the solution set is $\mathrm{Span}\{\mathbf{v}\}$ — a line through the origin. A homogeneous system with two free variables produces two vectors and a plane through the origin. The general statement:

Nonhomogeneous systems: particular plus homogeneous

When $A\mathbf{x}=\mathbf{b}$ has many solutions, its general solution splits into a fixed part and a varying part.

The vector $\mathbf{p}$ is one particular solution (set $x_3=0$), and $x_3\mathbf{v}$ traces out the solution set of the homogeneous equation from before. The solution set of $A\mathbf{x}=\mathbf{b}$ is the homogeneous line $\mathrm{Span}\{\mathbf{v}\}$ translated by $\mathbf{p}$.

The proof is one line each way. If $A\mathbf{p}=\mathbf{b}$ and $A{\mathbf{v}}_h=0$, then $A(\mathbf{p}+{\mathbf{v}}_h)=\mathbf{b}+0=\mathbf{b}$, so every such sum is a solution; and if $A\mathbf{w}=\mathbf{b}$, then ${\mathbf{v}}_h=\mathbf{w}-\mathbf{p}$ satisfies $A{\mathbf{v}}_h=\mathbf{b}-\mathbf{b}=0$, so $\mathbf{w}=\mathbf{p}+{\mathbf{v}}_h$.¹

The mental image for a consistent $A\mathbf{x}=\mathbf{b}$ with $\mathbf{b}\ne 0$ is a point, line, or plane that does not pass through the origin. When the system is inconsistent, the solution set is empty and the theorem does not apply.

The homogeneous solution set is the shape and the particular solution is the position.

Applied systems with many solutions

Real models often produce free variables, and each carries meaning: a degree of freedom that some side condition (integer coefficients, nonnegative prices, a known total inflow) resolves to a definite value.

Equilibrium prices (Leontief exchange model)

Divide an economy into sectors. Each sector's output is distributed among the sectors as fractions summing to $1$. An equilibrium assigns a price to each sector's total output so that every sector's income equals its expenditures. For a Coal / Electric / Steel economy with the given exchange fractions, the balance equations, moved to one side, form a homogeneous system

Row reduction gives $p_C=0.94p_S$, $p_E=0.85p_S$, with $p_S$ free. The free variable is a choice of scale (the currency and overall magnitude); taking $p_S=100$ prices Coal at $94$, Electric at $85$, Steel at $100$. Every homogeneous exchange model has this structure, and Leontief proved such a nonnegative equilibrium always exists.²

Balancing a chemical reaction

Atoms are conserved, so the coefficients of a reaction must equalize each element's atom count on both sides. Represent each molecule by its vector of atom counts.

Network flow

A network is junctions joined by directed branches. The conservation rule is that flow into each junction equals flow out, and total flow into the network equals total flow out. Each junction contributes one linear equation. For a grid of one-way streets with measured inflows and outflows, labeling the unknown branch flows $x_1,\dots ,x_5$ and balancing at each intersection yields a system whose general solution carries a free variable — the flow is underdetermined by the boundary data alone.

Because the streets are one-way, all $x_i\ge 0$, and that constraint plus the free variable pins the flows to a range rather than a single value. Electrical networks fit the same mold: Kirchhoff's voltage law makes the loop currents satisfy a linear system $R\mathbf{i}=\mathbf{v}$, a matrix form of Ohm's law.³

A nutrition model

A weight-loss diet must supply exact amounts of several nutrients using a few foodstuffs, each with its own nutrient profile. With one vector per foodstuff listing grams of each nutrient per unit, and $\mathbf{b}$ the required daily totals, the diet is the solution of

where $x_i$ is the number of units of foodstuff $i$. For three ingredients and three nutrients this is a square system; row reduction gives a unique nonnegative solution when the profiles are independent. The nonnegativity requirement (a negative amount of whey is meaningless) is a real constraint that drives the choice of ingredients.

Difference equations and migration

A linear difference equation models a system sampled at discrete times by a sequence of state vectors ${\mathbf{x}}_0,{\mathbf{x}}_1,{\mathbf{x}}_2,\dots$ with

For city-versus-suburb populations where each year a fixed fraction migrates each way, the migration matrix $M$ collects those fractions, and iterating $M$ propagates the population forward.

Starting from $(600,000,400,000)$, one step gives $(582,000,418,000)$, another gives $(565,440,434,560)$, and so on. The long-run behavior of such iterations — whether the state settles to a steady distribution — is an eigenvalue question, studied alongside difference equations and Markov chains.