Systems of Linear Equations

Suppose three sensors report measurements that should agree on a hidden set of unknowns. When can you actually find values that satisfy all three reports simultaneously? This is the question of solving a system of linear equations. It is arguably the single most useful computational question in all of mathematics. Systems of linear equations appear whenever a problem is built from additive, proportional relationships. Examples include balancing chemical reactions, computing currents in circuits, fitting models to data. They also include rendering graphics and solving discretized differential equations. The deep idea running through this article is simple. A system of linear equations is not merely a list of constraints to be solved by trial and error. It is a single geometric object. Each equation describes a hyperplane (a flat, infinite sheet of a certain dimension) in space. The solution set of the system is exactly the intersection of these hyperplanes. Whether a system has no solutions, exactly one solution, or infinitely many is entirely a question about how these flats meet. This geometric viewpoint explains why the algebra behaves the way it does. It does more than just show that it does. The big picture is that solving a linear system is really a question about a matrix and a vector. Writing the system as Ax=bA\mathbf{x} = \mathbf{b} converts a list of scalar equations into one vector equation. It also converts the entire theory into a study of the algebraic structure of AA. This is the doorway into all of linear algebra. Understanding Ax=bA\mathbf{x}=\mathbf{b} leads directly to rank, row reduction, vector spaces, linear independence. It eventually leads to eigenvalues and beyond. This article gives the precise definitions of a linear equation, a system, and a solution set. It proves the fundamental classification theorem (no solutions, exactly one, or infinitely many). It builds the geometric intuition of intersecting hyperplanes. It works through concrete numerical examples chosen to expose subtleties. This includes a case showing why linearity of the equations is essential to the entire theory.

Key takeaways
  • A linear equation in nn unknowns has the form a1x1+⋯+anxn=ba_1x_1+\cdots+a_nx_n=b; no products, powers, or other functions of the unknowns are allowed.
  • A system of mm linear equations in nn unknowns can be packaged as one matrix equation Ax=bA\mathbf{x}=\mathbf{b} with AA an m×nm\times n matrix.
  • The solution set of any linear system is exactly one of: empty, a single point, or an infinite set (over an infinite field).
  • Geometrically, each equation is a hyperplane, and the solution set is the intersection of all these hyperplanes.
  • Consistency (having at least one solution) is a separate question from uniqueness, and both are governed by the rank of AA versus the rank of the augmented matrix [A∣b][A\mid \mathbf{b}].
  • Key idea: solving Ax=bA\mathbf{x}=\mathbf{b} is the same problem as finding all ways to write b\mathbf{b} as a linear combination of the columns of AA.

Where this lives today

Systems of linear equations are the computational backbone of applied mathematics. In engineering, circuit analysis via Kirchhoff's laws and structural analysis via finite-element methods both reduce large physical problems to enormous, sparse linear systems solved numerically. In computer graphics and robotics, transforming coordinates, computing inverse kinematics, and solving for camera poses all involve linear systems, often with least-squares corrections for systems with no exact solution. In statistics and machine learning, linear regression is precisely the search for the best approximate solution to an inconsistent linear system Ax≈bA\mathbf{x}\approx \mathbf{b}, handled by projecting b\mathbf{b} onto the column space of AA. In cryptography, linear systems over finite fields underlie linear cryptanalysis and error-correcting codes, where the field being finite (rather than R\mathbb{R}) changes the counting in the Theorem's third case from "infinite" to a precise finite count. In physics, quantum mechanics represents states as vectors and observables as linear operators, so that eigenvalue problems (a close relative of linear systems) determine energy levels; in network science, balancing equations for flows on graphs are linear systems whose coefficient matrices are incidence matrices of graphs. The trichotomy proved here is the seed from which the entire edifice of Rank, Linear Independence, and Vector Spaces grows, and the elimination technique used in the worked examples is formalized next as Gaussian Elimination and Row Reduction, carried out rigorously over an arbitrary field.

Ask the tutor: How does row reduction actually compute the particular solution xp\mathbf{x}_p and a basis for the null space NN described in this theorem?

DefinitionLinear equation, system, and solution set

Let FF be a field (for concreteness the reader may take F=RF=\mathbb{R} or F=CF=\mathbb{C}). A linear equation in the unknowns x1,…,xnx_1,\dots,x_n over FF is an expression

a1x1+a2x2+⋯+anxn=b,a_1x_1 + a_2x_2 + \cdots + a_nx_n = b,

where a1,…,an,b∈Fa_1,\dots,a_n,b \in F are fixed scalars, called the coefficients and the constant term. No unknown may appear multiplied by another unknown, raised to a power other than 11, or as the input of a nonlinear function. A system of mm linear equations in nn unknowns is a finite collection

a11x1+a12x2+⋯+a1nxn=b1a21x1+a22x2+⋯+a2nxn=b2    ⋮am1x1+am2x2+⋯+amnxn=bm.\begin{aligned} a_{11}x_1 + a_{12}x_2 + \cdots + a_{1n}x_n &= b_1\\ a_{21}x_1 + a_{22}x_2 + \cdots + a_{2n}x_n &= b_2\\ &\;\;\vdots\\ a_{m1}x_1 + a_{m2}x_2 + \cdots + a_{mn}x_n &= b_m. \end{aligned}

Writing A=(aij)∈Fm×nA = (a_{ij}) \in F^{m\times n}, x=(x1,…,xn)T\mathbf{x} = (x_1,\dots,x_n)^{T}, and b=(b1,…,bm)T\mathbf{b} = (b_1,\dots,b_m)^{T}, the system is equivalent to the single matrix equation

Ax=b.A\mathbf{x} = \mathbf{b}.

A solution is a vector x∈Fn\mathbf{x}\in F^n satisfying all mm equations simultaneously (equivalently, satisfying Ax=bA\mathbf{x}=\mathbf{b}). The solution set is

S={x∈Fn:Ax=b}.S = \{\mathbf{x}\in F^n : A\mathbf{x}=\mathbf{b}\}.

The system is consistent if S≠∅S \neq \varnothing, and inconsistent if S=∅S=\varnothing. When b=0\mathbf{b}=\mathbf{0} the system is called homogeneous; its solution set always contains 0\mathbf{0}. The augmented matrix of the system is the m×(n+1)m\times(n+1) matrix [A∣b][A \mid \mathbf{b}] obtained by appending b\mathbf{b} as an extra column to AA.

Intuition

A single linear equation a1x1+⋯+anxn=ba_1x_1+\cdots+a_nx_n=b (with not all ai=0a_i=0) describes a hyperplane in FnF^n: a line when n=2n=2, a plane when n=3n=3, and a flat (n−1)(n-1)-dimensional slice of space in general. This is because the equation says that the vector x\mathbf{x}, when paired with the fixed coefficient vector a=(a1,…,an)\mathbf{a}=(a_1,\dots,a_n) via the dot product, always gives the same number bb; geometrically this pins x\mathbf{x} to a flat sheet perpendicular to a\mathbf{a} (over R\mathbb{R}), offset from the origin unless b=0b=0. A system of mm such equations asks for a point lying on all mm hyperplanes at once — the solution set is their intersection. Two lines in the plane either cross at one point, run parallel and never meet, or are the same line entirely; this is the entire story of 2×22\times 2 systems, and it generalizes verbatim to mm equations in nn unknowns over any field: the solution set is empty, a single point, or an affine subspace of dimension ≥1\geq 1 containing infinitely many points (when FF is infinite). No fourth possibility can occur — this is the content of the theorem below. The reason linearity matters so much is that hyperplanes are the only shapes for which "intersection of several of them" has such a rigid three-way classification. Curves given by nonlinear equations (circles, parabolas) can intersect in isolated finite sets of any size, or in complicated curves — the clean trichotomy is special to linear equations, and it is precisely what makes linear algebra tractable.

Figure
Two lines in the plane: the three possible intersection patterns
unique solution (x*, y*) no solution parallel, distinct infinitely many solutions same line

Consistency and the classification theorem

TheoremTrichotomy for linear systems

Let FF be a field and let Ax=bA\mathbf{x}=\mathbf{b} be a system of mm linear equations in nn unknowns over FF, with solution set S⊆FnS\subseteq F^n. Then exactly one of the following holds: 1. S=∅S = \varnothing (the system is inconsistent); 2. S={x0}S = \{\mathbf{x}_0\} for a unique x0∈Fn\mathbf{x}_0 \in F^n (a unique solution); or 3. SS contains at least two distinct vectors, in which case, if FF is infinite, SS is infinite. Moreover, SS is always of the form

S=xp+N,N={x∈Fn:Ax=0},S = \mathbf{x}_p + N, \qquad N = \{\mathbf{x} \in F^n : A\mathbf{x} = \mathbf{0}\},

where xp\mathbf{x}_p is any single (particular) solution of Ax=bA\mathbf{x}=\mathbf{b}, provided S≠∅S\neq\varnothing; here NN is the solution set of the associated homogeneous system.

Proof

We use direct verification and the linearity of matrix multiplication in x\mathbf{x}, i.e. the fact that for all u,v∈Fn\mathbf{u},\mathbf{v}\in F^n and c∈Fc\in F,

A(u+v)=Au+Av,A(cu)=c(Au),A(\mathbf{u}+\mathbf{v}) = A\mathbf{u} + A\mathbf{v}, \qquad A(c\mathbf{u}) = c(A\mathbf{u}),

which follows immediately from the row-by-row definition of matrix–vector multiplication: the ii-th entry of A(u+v)A(\mathbf{u}+\mathbf{v}) is ∑jaij(uj+vj)=∑jaijuj+∑jaijvj\sum_j a_{ij}(u_j+v_j) = \sum_j a_{ij}u_j + \sum_j a_{ij}v_j, the ii-th entries of AuA\mathbf{u} and AvA\mathbf{v} added, and similarly for scalar multiples.

Step 1: Structure of NN. First we check NN is nonempty and closed under addition and scalar multiplication. Since A0=0A\mathbf{0} = \mathbf{0} (each entry is a sum of terms aij⋅0=0a_{ij}\cdot 0 = 0), we get 0∈N\mathbf{0}\in N. If n1,n2∈N\mathbf{n}_1,\mathbf{n}_2\in N then A(n1+n2)=An1+An2=0+0=0A(\mathbf{n}_1+\mathbf{n}_2) = A\mathbf{n}_1+A\mathbf{n}_2 = \mathbf{0}+\mathbf{0}=\mathbf{0}, so n1+n2∈N\mathbf{n}_1+\mathbf{n}_2\in N. If n∈N\mathbf{n}\in N and c∈Fc\in F then A(cn)=c(An)=c0=0A(c\mathbf{n}) = c(A\mathbf{n}) = c\mathbf{0} = \mathbf{0}, so cn∈Nc\mathbf{n}\in N.

Step 2: The affine structure of SS. Suppose S≠∅S\neq\varnothing and fix a particular solution xp∈S\mathbf{x}_p\in S, so Axp=bA\mathbf{x}_p=\mathbf{b}. We claim S=xp+N:={xp+n:n∈N}S = \mathbf{x}_p + N := \{\mathbf{x}_p+\mathbf{n} : \mathbf{n}\in N\}. (⊇\supseteq) If n∈N\mathbf{n}\in N, then A(xp+n)=Axp+An=b+0=bA(\mathbf{x}_p+\mathbf{n}) = A\mathbf{x}_p + A\mathbf{n} = \mathbf{b}+\mathbf{0} = \mathbf{b}, so xp+n∈S\mathbf{x}_p+\mathbf{n}\in S. (⊆\subseteq) If x∈S\mathbf{x}\in S, let n:=x−xp\mathbf{n} := \mathbf{x}-\mathbf{x}_p. Then An=Ax−Axp=b−b=0A\mathbf{n} = A\mathbf{x}-A\mathbf{x}_p = \mathbf{b}-\mathbf{b}=\mathbf{0}, so n∈N\mathbf{n}\in N, and x=xp+n∈xp+N\mathbf{x} = \mathbf{x}_p+\mathbf{n} \in \mathbf{x}_p+N. This proves the displayed formula for SS whenever S≠∅S \neq \varnothing.

Step 3: The trichotomy. If S=∅S=\varnothing we are in case (1); there is nothing further to prove, and by definition this excludes cases (2) and (3) since those require S≠∅S\neq\varnothing. Now suppose S≠∅S\neq\varnothing, so S=xp+NS = \mathbf{x}_p+N by Step 2. We show: either N={0}N=\{\mathbf{0}\}, giving S={xp}S=\{\mathbf{x}_p\} (case 2, uniqueness), or N≠{0}N\neq\{\mathbf{0}\}, giving ∣S∣≥2|S|\geq 2 and, if FF is infinite, ∣S∣=∞|S|=\infty (case 3). Case N={0}N=\{\mathbf 0\}. Then S=xp+{0}={xp}S=\mathbf{x}_p+\{\mathbf 0\}=\{\mathbf x_p\}, a single point. Uniqueness of x0=xp\mathbf{x}_0=\mathbf x_p as an element of SS is immediate since SS literally has one element. Case N≠{0}N\neq\{\mathbf 0\}. Choose n0∈N\mathbf{n}_0\in N with n0≠0\mathbf n_0\neq \mathbf 0. Then xp\mathbf x_p and xp+n0\mathbf x_p+\mathbf n_0 are two distinct elements of SS (distinct because n0≠0\mathbf n_0\neq \mathbf 0), so ∣S∣≥2|S|\ge 2. Now suppose FF is infinite. For each c∈Fc\in F, cn0∈Nc\mathbf n_0 \in N by Step 1, so xp+cn0∈S\mathbf x_p + c\mathbf n_0\in S. The map c↦xp+cn0c\mapsto \mathbf x_p+c\mathbf n_0 from FF to SS is injective: if xp+cn0=xp+c′n0\mathbf x_p+c\mathbf n_0 = \mathbf x_p+c'\mathbf n_0 then (c−c′)n0=0(c-c')\mathbf n_0=\mathbf 0, and since n0≠0\mathbf n_0\neq \mathbf 0 this forces c=c′c=c' (in a field, if (c−c′)n0=0(c-c')\mathbf{n}_0 = \mathbf 0 and some coordinate of n0\mathbf n_0 is nonzero, then c−c′=0c-c'=0 by cancellation in FF). Since FF is infinite, this exhibits an injection from an infinite set into SS, so SS is infinite.

Step 4: Exactly one case holds. Cases (1) and the pair {(2),(3)}\{(2),(3)\} are distinguished by S=∅S=\varnothing versus S≠∅S\neq\varnothing, which are mutually exclusive and exhaust all possibilities. Given S≠∅S\neq\varnothing, the two sub-cases N={0}N=\{\mathbf 0\} and N≠{0}N\neq\{\mathbf 0\} are also mutually exclusive and exhaustive, and we showed the first forces ∣S∣=1|S|=1 while the second forces ∣S∣≥2|S|\geq 2 (infinite when FF is infinite); these cannot both hold since ∣S∣=1|S|=1 and ∣S∣≥2|S|\ge2 are incompatible. Hence exactly one of (1), (2), (3) holds. ■\blacksquare

∎
CorollaryHomogeneous systems always have the trivial solution

For any homogeneous system Ax=0A\mathbf{x}=\mathbf{0}, the solution set NN is nonempty (it contains 0\mathbf{0}), so the system is always consistent, and by the Theorem, NN is either {0}\{\mathbf{0}\} alone or infinite (when FF is infinite).

Proof

Immediate from Step 1 of the proof above: 0∈N\mathbf 0 \in N always, so N≠∅N\ne\varnothing, and NN itself is the solution set of Ax=0A\mathbf{x}=\mathbf 0 with xp=0\mathbf{x}_p=\mathbf 0 serving as a particular solution, so the trichotomy applied to the homogeneous system rules out the "no solution" case. ■\blacksquare

∎

Worked examples

Example 1 (a fully determined asymmetric 3×33\times 3 system — unique solution). Solve over R\mathbb{R}:

2x1−x2+3x3=9x1+4x2−x3=−23x1+2x2+5x3=17\begin{aligned} 2x_1 - x_2 + 3x_3 &= 9\\ x_1 + 4x_2 - x_3 &= -2\\ 3x_1 + 2x_2 + 5x_3 &= 17 \end{aligned}

Eliminate x1x_1 using equation 2 as pivot. From equation 2: x1=−2−4x2+x3x_1 = -2 - 4x_2 + x_3. Substitute into equation 1:

2(−2−4x2+x3)−x2+3x3=9  ⟹  −4−8x2+2x3−x2+3x3=9  ⟹  −9x2+5x3=13.(∗)2(-2-4x_2+x_3) - x_2 + 3x_3 = 9 \;\Longrightarrow\; -4 -8x_2+2x_3-x_2+3x_3 = 9 \;\Longrightarrow\; -9x_2+5x_3 = 13. \quad (\ast)

Substitute into equation 3:

3(−2−4x2+x3)+2x2+5x3=17  ⟹  −6−12x2+3x3+2x2+5x3=17  ⟹  −10x2+8x3=23.(∗∗)3(-2-4x_2+x_3)+2x_2+5x_3 = 17 \;\Longrightarrow\; -6-12x_2+3x_3+2x_2+5x_3=17 \;\Longrightarrow\; -10x_2+8x_3=23. \quad (\ast\ast)

Now solve the 2×22\times2 system (∗),(∗∗)(\ast),(\ast\ast):

−9x2+5x3=13,−10x2+8x3=23.-9x_2+5x_3=13,\qquad -10x_2+8x_3=23.

Multiply the first by 88 and the second by 55: −72x2+40x3=104-72x_2+40x_3 = 104 and −50x2+40x3=115-50x_2+40x_3=115. Subtracting:

−72x2−(−50x2)=104−115⟹−22x2=−11⟹x2=12.-72x_2 - (-50x_2) = 104-115 \Longrightarrow -22x_2 = -11 \Longrightarrow x_2 = \tfrac12.

Plug into (∗)(\ast): −9(12)+5x3=13⇒−4.5+5x3=13⇒x3=3.5-9(\tfrac12)+5x_3 = 13 \Rightarrow -4.5+5x_3=13 \Rightarrow x_3 = 3.5. Then x1=−2−4(12)+3.5=−2−2+3.5=−0.5x_1 = -2-4(\tfrac12)+3.5 = -2-2+3.5 = -0.5.

Check in the original equations:

  • 2(−0.5)−0.5+3(3.5)=−1−0.5+10.5=92(-0.5)-0.5+3(3.5) = -1-0.5+10.5 = 9 ✓.
  • −0.5+4(0.5)−3.5=−0.5+2−3.5=−2-0.5+4(0.5)-3.5 = -0.5+2-3.5 = -2 ✓.
  • 3(−0.5)+2(0.5)+5(3.5)=−1.5+1+17.5=173(-0.5)+2(0.5)+5(3.5) = -1.5+1+17.5 = 17 ✓.

So the unique solution is x0=(−12, 12, 72)\mathbf{x}_0=(-\tfrac12,\ \tfrac12,\ \tfrac72), illustrating case (2) of the Theorem: three generic planes in R3\mathbb{R}^3 meeting in exactly one point.

Example 2 (an underdetermined 2×32\times 3 system — infinitely many solutions, described via the particular-plus-homogeneous decomposition). Solve over R\mathbb{R}:

x1+2x2−x3=43x1−x2+2x3=5.\begin{aligned} x_1 + 2x_2 - x_3 &= 4\\ 3x_1 - x_2 + 2x_3 &= 5. \end{aligned}

Here m=2<n=3m=2 < n=3, so we expect (and will confirm) infinitely many solutions rather than a unique one. Eliminate x1x_1: multiply the first equation by 33 and subtract the second:

3(x1+2x2−x3)−(3x1−x2+2x3)=3(4)−5⟹7x2−5x3=7.3(x_1+2x_2-x_3) - (3x_1-x_2+2x_3) = 3(4)-5 \Longrightarrow 7x_2 - 5x_3 = 7.

This single relation between x2,x3x_2,x_3 is the residual constraint; introduce the free parameter t=x3∈Rt=x_3\in\mathbb{R}. Then

x2=7+5t7=1+57t.x_2 = \frac{7+5t}{7} = 1 + \frac{5}{7}t.

Substitute back into the first original equation to get x1x_1:

x1=4−2x2+x3=4−2(1+57t)+t=4−2−107t+t=2−37t.x_1 = 4 - 2x_2 + x_3 = 4 - 2\left(1+\tfrac57 t\right) + t = 4-2-\tfrac{10}{7}t+t = 2 - \tfrac{3}{7}t.

So the full solution set is

S={(2−37t, 1+57t, t):t∈R}.S = \left\{ \left(2-\tfrac37 t,\ 1+\tfrac57 t,\ t\right) : t\in\mathbb{R} \right\}.

To exhibit the particular-plus-homogeneous decomposition from the Theorem, take t=0t=0: xp=(2,1,0)\mathbf{x}_p = (2,1,0). Check: 2+2(1)−0=42+2(1)-0 = 4 ✓, 3(2)−1+2(0)=53(2)-1+2(0)=5 ✓. The homogeneous direction is read off from the coefficient of tt: n0=(−37,57,1)\mathbf{n}_0 = (-\tfrac37, \tfrac57, 1), and one checks directly An0=0A\mathbf n_0 = \mathbf 0:

−37+2(57)−1=−37+107−1=77−1=0,3(−37)−57+2(1)=−97−57+2=−2+2=0.-\tfrac37+2\left(\tfrac57\right)-1 = -\tfrac37+\tfrac{10}{7}-1 = \tfrac{7}{7}-1=0,\qquad 3\left(-\tfrac37\right)-\tfrac57+2(1) = -\tfrac97-\tfrac57+2 = -2+2=0.

So S=xp+{t n0:t∈R}S = \mathbf{x}_p + \{t\,\mathbf n_0 : t\in\mathbb R\}, exactly the affine-line structure predicted abstractly by the Theorem: two planes in R3\mathbb{R}^3 meeting in a line.

CounterexampleWhy linearity is essential: nonlinear "systems" need not obey the trichotomy

Drop the hypothesis that the equations are linear, keeping everything else (two equations, two unknowns, real numbers), and consider

x2+y2=5,y=x.x^2 + y^2 = 5, \qquad y = x.

Substituting the second equation into the first: x2+x2=5x^2+x^2=5, so x2=52x^2=\tfrac52, giving x=±5/2x=\pm\sqrt{5/2}. Hence the solution set is exactly

S={(5/2, 5/2), (−5/2, −5/2)},S = \left\{ \left(\sqrt{5/2},\ \sqrt{5/2}\right),\ \left(-\sqrt{5/2},\ -\sqrt{5/2}\right) \right\},

a set of size exactly 2 — neither 00, nor 11, nor infinite. This is impossible for a genuinely linear system by the Theorem, whose proof relied essentially on linearity: Step 1 used A0=0A\mathbf 0=\mathbf 0 and additivity/homogeneity of x↦Ax\mathbf x\mapsto A\mathbf x, and Step 3's infinitude argument used the fact that scaling a nonzero homogeneous solution by every field element cc gives a different solution for each cc — a fact with no analogue once one of the equations is quadratic. The circle x2+y2=5x^2+y^2=5 is not a hyperplane, and its intersection with the line y=xy=x is two points precisely because a line can cross a circle in at most two points, a fundamentally different geometric fact from "two lines meet in at most one point unless equal." This shows the linearity hypothesis cannot be weakened without destroying the trichotomy.

Scope, edge cases & common traps

SituationWhat can happenCommon trap
m<nm<n (fewer equations than unknowns)Consistent systems typically have infinitely many solutions (as in Example 2), but can still be inconsistentAssuming "more unknowns than equations" guarantees solutions exist
m>nm>n (more equations than unknowns)Typically inconsistent (over-determined), but can be consistent with a unique solution if equations are redundantAssuming extra equations always cause contradiction
m=nm=nCan be any of the three cases; unique solution iff AA is invertibleAssuming a square system automatically has a unique solution
FF finite (e.g. F2\mathbb F_2)Case (3) of the Theorem still gives ∣S∣≥2|S|\ge 2, but SS need not be infinite — it is instead exactly ∣F∣dim⁡N|F|^{\dim N}Quoting "infinitely many solutions" over finite fields
Homogeneous systemAlways consistent (0∈N\mathbf 0 \in N); question is only whether N={0}N=\{\mathbf 0\}Forgetting the trivial solution always exists

The mathematician behind it

Origins · 1 minCarl Friedrich Gauss and how this was discovered
CF
Carl Friedrich Gauss
1777–1855 · German mathematician, astronomer, and physicist

Gauss developed and systematized the elimination method for solving linear systems in the course of his astronomical and geodetic work — most famously in predicting the orbit of the asteroid Ceres in 1801 from a handful of positional observations, a problem that reduced to solving linear (and least-squares) systems by hand. Although elimination techniques for small systems were known to Chinese mathematicians over a thousand years earlier (the "Fangcheng" procedure in the Nine Chapters on the Mathematical Art), Gauss's systematic treatment, applied at scale to real observational data, cemented the method's central role in the mathematical sciences and gave it the name still used today.

History
- c. 200 BCE–100 CE: The Nine Chapters on the Mathematical Art describes a systematic elimination procedure for solving linear systems, essentially equivalent to modern Gaussian elimination.
- 1750: Gabriel Cramer publishes what becomes known as Cramer's Rule, solving square linear systems via determinants.
- 1801: Gauss applies elimination-based methods to compute the orbit of Ceres from observational data, popularizing systematic solution techniques in Europe.
- 1888: Giuseppe Peano gives an early axiomatic treatment of vector spaces, providing the abstract setting in which solution sets of linear systems are properly understood as affine subspaces.
- 1947: John von Neumann and Herman Goldstine analyze the numerical stability of Gaussian elimination on early electronic computers, launching the modern era of large-scale computational linear algebra.
What exactly makes an equation "linear"?

An equation is linear in unknowns x1,…,xnx_1,\dots,x_n if it can be written as a1x1+⋯+anxn=ba_1x_1+\cdots+a_nx_n=b for fixed scalars ai,ba_i,b, with each unknown appearing to the first power, un-multiplied by any other unknown, and not inside any other function like sin⁡\sin, log⁡\log, or a square root. Equivalently, the left side must be expressible as a dot product of a fixed coefficient vector with the vector of unknowns. Equations like xy=1xy=1, x2+y=3x^2+y=3, or sin⁡(x)=y\sin(x)=y are not linear, and systems containing them do not obey the trichotomy theorem proved here, as the counterexample block demonstrates explicitly.

Can a linear system have exactly 5 solutions?

No, not over an infinite field such as R\mathbb R or C\mathbb C. By the classification theorem, the solution set of a linear system is either empty, a single point, or infinite; there is no way to get a finite set of size 2,3,52,3,5, or any number greater than 11. This is a genuinely special feature of linearity — the proof shows that once you find two distinct solutions, you automatically get an entire infinite family by scaling the difference vector. Over a finite field, however, finite sets larger than one point certainly occur, since the field itself has finitely many scalars to scale by.

Is every consistent system with more unknowns than equations guaranteed to have infinitely many solutions?

Not automatically — it depends on whether the equations are independent, but the intuition is usually correct once made precise. If m<nm<n and the system is consistent, then (as will be proved using rank in the theory of row reduction) the solution set is a coset of a subspace NN of dimension at least n−m≥1n-m\ge 1, so N≠{0}N\neq\{\mathbf 0\} and by the Theorem the system has infinitely many solutions over an infinite field. The subtlety is entirely in confirming consistency first; an inconsistent system with m<nm<n, such as x+y=1, x+y=2x+y=1,\ x+y=2, still has no solutions at all.

What is the difference between the solution set SS and the null space / solution set NN of the homogeneous system?

NN is the solution set of the associated homogeneous system Ax=0A\mathbf x=\mathbf 0 and is always a genuine vector subspace of FnF^n (closed under addition and scalar multiplication, containing 0\mathbf 0), as proved in Step 1 above. SS, the solution set of the original system Ax=bA\mathbf x=\mathbf b, is generally not a subspace at all when b≠0\mathbf b\neq \mathbf 0 — it does not contain 0\mathbf 0 unless 0\mathbf 0 happens to be a solution. Instead SS is a translate (an affine copy) of NN, namely xp+N\mathbf x_p+N for any fixed particular solution xp\mathbf x_p, which is the precise relationship proved in the Theorem.

Does the trichotomy theorem tell us how to compute the solution set, or only that it has one of three shapes?

Only the shape — the Theorem is a structural, existence-and-classification statement, proved purely from the additivity and homogeneity of matrix–vector multiplication, without reference to any particular computational method. Actually finding xp\mathbf x_p and a description of NN (for instance, a list of vectors whose combinations produce all of NN) requires an algorithm, the standard one being Gaussian elimination and row reduction. That computational machinery is the natural next topic once the qualitative classification here is understood.

Why does the article insist on stating that FF is a field, rather than just working with real numbers throughout?

Stating the field explicitly matters because the proof of the trichotomy theorem uses field axioms directly: the existence of additive inverses (to form x−xp\mathbf x - \mathbf x_p), multiplicative cancellation (to show c=c′c=c' from (c−c′)n0=0(c-c')\mathbf n_0=\mathbf 0), and — crucially, for the "infinite" conclusion — the cardinality of FF itself. Every step of the proof goes through verbatim for Q\mathbb{Q}, C\mathbb{C}, or any infinite field; only the finiteness conclusion changes over a finite field like F2\mathbb F_2, where case (3) still gives at least two solutions but not infinitely many. Making the hypothesis explicit prevents silently assuming properties special to R\mathbb R.

How does the geometric picture of intersecting hyperplanes relate to the algebraic picture of Ax=bA\mathbf x=\mathbf b?

Each row of AA, together with the corresponding entry of b\mathbf b, defines one linear equation, hence one hyperplane in FnF^n; the solution set SS is by definition the intersection of all mm hyperplanes, since a vector solves the system exactly when it satisfies every row simultaneously. The algebraic classification (empty, single point, or infinite affine subspace) is thus a direct translation of the geometric fact that finitely many hyperplanes either fail to have a common point, meet in exactly one point, or meet in a common flat of dimension at least one. This dictionary between rows-of-AA and hyperplanes is the geometric content underlying every proof in this article.

Where this leads