Cramer's Rule

Suppose you are handed a square linear system with numerical coefficients. You are asked only for the value of the third unknown. You do not need the entire solution vector. Gaussian elimination will get you there. But it computes far more than you asked for. It also obscures how each individual unknown depends on the data. Cramer's Rule answers the narrower question directly. It expresses each coordinate of the solution as a single ratio of two determinants. One determinant is built from the original coefficient matrix. The other comes from a matrix with a column replaced by the constants. This is not merely a computational trick. It is a precise statement about how solutions of linear systems depend algebraically on their data. The result matters because it turns an abstract existence-and-uniqueness statement into an explicit formula. That statement is: "a nonsingular system has a unique solution." Cramer's Rule reveals that each solution coordinate is a rational function. It is a function of the entries of the coefficient matrix and the constant vector. This fact has consequences well beyond hand computation. It touches perturbation theory, algebraic geometry, and the theory of linear ODEs with parameters. It also gives a clean, symbolic way to isolate one variable of interest. This can be done without back-substitution. The big picture connects three ideas already in hand. First, the determinant as a test for invertibility. Second, the Adjugate and Cofactor Matrix as an explicit formula for the inverse. Third, the interpretation of a linear system $A\mathbf{x} = \mathbf{b}$ . This system asks for coordinates of $\mathbf{b}$ in the basis given by the columns of $A$ . Cramer's Rule is the determinant-theoretic shadow of matrix inversion. Instead of computing $A^{-1}$ and multiplying, you compute a ratio of two determinants for each unknown. This article states the rule with all hypotheses spelled out. It proves the rule rigorously using the adjugate formula for the inverse. It works two nontrivial numerical examples. It shows exactly what breaks when nonsingularity fails. Finally, it situates the result historically and practically.

Key takeaways

Cramer's Rule solves $A\mathbf{x}=\mathbf{b}$ for a square, invertible matrix $A$ by ratios of determinants.
Each unknown $x_i = \det(A_i)/\det(A)$ , where $A_i$ replaces column $i$ of $A$ with $\mathbf{b}$ .
It requires $\det(A)\neq 0$ ; the rule gives the unique solution, guaranteed by Determinant and Invertibility.
The proof follows in one line once you know $A^{-1} = \frac{1}{\det A}\,\mathrm{adj}(A)$ , from The Adjugate and Cofactor Matrix.
It is computationally inefficient for large $n$ (exponential-time determinants) but theoretically indispensable.
Key idea: solving a linear system is, structurally, just applying the adjugate formula for $A^{-1}$ one column of $\mathbf b$ 's worth at a time — and reading off the answer as a determinant ratio.

Where this lives today

Cramer's Rule is rarely the algorithm of choice for large numerical linear systems — Gaussian elimination and LU decomposition dominate there because their cost grows polynomially ( $O(n^3)$ ) rather than through repeated full determinant evaluations. But the rule's real modern life is elsewhere:

Symbolic and parametric computation. When coefficients are symbols (e.g. physical constants, design parameters) rather than numbers, Cramer's Rule gives closed-form rational-function expressions for each unknown, essential in computer algebra systems and in sensitivity analysis.
Control theory and circuit analysis. Solving small linear systems (node/mesh equations in circuits, state-space equations) symbolically in terms of a parameter such as frequency, where an explicit formula for one output variable is more useful than a full numeric solve.
Differential geometry and the inverse function theorem. The proof that a $C^1$ map with invertible Jacobian is locally invertible often uses the adjugate/determinant structure underlying Cramer's Rule to control how the inverse map depends smoothly on parameters.
Algebraic geometry. Cramer's original use — passing curves through prescribed points and eliminating unknown coefficients — is the ancestor of resultant and elimination theory used in solving systems of polynomial equations.
Cryptography and coding theory. Determinant ratios appear in decoding algorithms (e.g. Reed–Solomon-related computations) and in secret-sharing schemes built on solving small linear systems over finite fields exactly.
Numerical stability analysis. Even though it is not used computationally at scale, Cramer's Rule underlies theoretical perturbation bounds: it shows explicitly how a small change in $\mathbf b$ or $A$ changes each $x_i$ , foreshadowing condition-number analysis in numerical linear algebra.

Ask the tutor: Can you show me how Cramer's Rule and Gaussian elimination give the same answer, step by step, on one shared example?

A \in M_{n\times n}(F)

Intuition

Think of the columns $\mathbf{a}_1, \dots, \mathbf{a}_n$ of $A$ as a basis of $F^n$ (they are a basis precisely because $\det A \neq 0$ , by Determinant and Invertibility). The system $A\mathbf{x} = \mathbf{b}$ asks: what are the coordinates of $\mathbf{b}$ relative to this basis? Determinants measure signed volume, and volume scales linearly in each argument separately (multilinearity). If you write $\mathbf{b} = x_1 \mathbf{a}_1 + \cdots + x_n \mathbf{a}_n$ and substitute this expansion into the $i$ -th column of $A$ , only the $x_i \mathbf{a}_i$ term survives after using alternation to kill the other terms (they duplicate columns already present) — this is exactly the volume-ratio idea: $x_i$ is the fraction of the "volume" of the parallelepiped spanned by $A$ 's columns that gets replaced when you swap in $\mathbf{b}$ for column $i$ . Geometrically in $\mathbb{R}^2$ : if $\mathbf{a}_1, \mathbf{a}_2$ span a parallelogram of signed area $\det A$ , and $\mathbf{b} = x_1\mathbf{a}_1 + x_2\mathbf{a}_2$ , then the parallelogram spanned by $\mathbf{b}, \mathbf{a}_2$ has signed area $x_1 \det A$ (because $\mathbf{a}_1$ is replaced by a combination that only contributes its $\mathbf a_1$ -component, the $\mathbf a_2$ -component vanishing against the repeated column). Ratio out $\det A$ and you recover $x_1$ .

Figure

Cramer's Rule in the plane: replacing a column rescales the parallelogram's area by exactly

x_i

Theorem

F

A \in M_{n\times n}(F)

This lemma is the content of The Adjugate and Cofactor Matrix; we cite it as already established and use it to prove Cramer's Rule.

A^{-1}

\mathbf{b} = \mathbf 0

Worked examples

Example 1 (a $3\times 3$ system with no special symmetry). Solve

\begin{cases} 2x_1 - x_2 + 3x_3 = 9 \\ 1x_1 + 4x_2 - 2x_3 = -3\\ 3x_1 + 2x_2 + x_3 = 10 \end{cases} \quad\Longleftrightarrow\quad A = \begin{pmatrix} 2 & -1 & 3 \\ 1 & 4 & -2 \\ 3 & 2 & 1 \end{pmatrix},\quad \mathbf b = \begin{pmatrix} 9 \\ -3 \\ 10 \end{pmatrix}.

Step 1: compute $\det(A)$ . Expand along the first row:

\det(A) = 2\det\begin{pmatrix}4 & -2\\ 2 & 1\end{pmatrix} - (-1)\det\begin{pmatrix}1 & -2\\3 & 1\end{pmatrix} + 3\det\begin{pmatrix}1 & 4\\3 & 2\end{pmatrix}.

Compute each $2\times2$ determinant:

\det\begin{pmatrix}4 & -2\\ 2 & 1\end{pmatrix} = 4(1) - (-2)(2) = 4 + 4 = 8,

\det\begin{pmatrix}1 & -2\\3 & 1\end{pmatrix} = 1(1) - (-2)(3) = 1 + 6 = 7,

\det\begin{pmatrix}1 & 4\\3 & 2\end{pmatrix} = 1(2) - 4(3) = 2 - 12 = -10.

\det(A) = 2(8) + 1(7) + 3(-10) = 16 + 7 - 30 = -7.

Since $\det(A) = -7 \neq 0$ , Cramer's Rule applies. Step 2: $\det(A_1(\mathbf b))$ , replacing column 1 with $\mathbf b$ :

A_1(\mathbf b) = \begin{pmatrix} 9 & -1 & 3 \\ -3 & 4 & -2 \\ 10 & 2 & 1 \end{pmatrix}.

Expand along the first row:

\det\begin{pmatrix}4 & -2\\2 & 1\end{pmatrix} = 4 + 4 = 8,\qquad \det\begin{pmatrix}-3 & -2\\10 & 1\end{pmatrix} = -3 + 20 = 17,\qquad \det\begin{pmatrix}-3 & 4\\10 & 2\end{pmatrix} = -6 - 40 = -46.

\det(A_1(\mathbf b)) = 9(8) - (-1)(17) + 3(-46) = 72 + 17 - 138 = -49.

So $x_1 = \dfrac{-49}{-7} = 7$ . Step 3: $\det(A_2(\mathbf b))$ , replacing column 2:

A_2(\mathbf b) = \begin{pmatrix} 2 & 9 & 3 \\ 1 & -3 & -2 \\ 3 & 10 & 1 \end{pmatrix}.

\det\begin{pmatrix}-3 & -2\\10 & 1\end{pmatrix} = -3 + 20 = 17,\qquad \det\begin{pmatrix}1 & -2\\3 & 1\end{pmatrix} = 1 + 6 = 7,\qquad \det\begin{pmatrix}1 & -3\\3 & 10\end{pmatrix} = 10 + 9 = 19.

\det(A_2(\mathbf b)) = 2(17) - 9(7) + 3(19) = 34 - 63 + 57 = 28.

So $x_2 = \dfrac{28}{-7} = -4$ . Step 4: $\det(A_3(\mathbf b))$ , replacing column 3:

A_3(\mathbf b) = \begin{pmatrix} 2 & -1 & 9 \\ 1 & 4 & -3 \\ 3 & 2 & 10 \end{pmatrix}.

\det\begin{pmatrix}4 & -3\\2 & 10\end{pmatrix} = 40 + 6 = 46,\qquad \det\begin{pmatrix}1 & -3\\3 & 10\end{pmatrix} = 10 + 9 = 19,\qquad \det\begin{pmatrix}1 & 4\\3 & 2\end{pmatrix} = 2 - 12 = -10.

\det(A_3(\mathbf b)) = 2(46) - (-1)(19) + 9(-10) = 92 + 19 - 90 = 21.

So $x_3 = \dfrac{21}{-7} = -3$ . Check. Row 1: $2(7) - (-4) + 3(-3) = 14 + 4 - 9 = 9$ . ✓ Row 2: $7 + 4(-4) - 2(-3) = 7 - 16 + 6 = -3$ . ✓ Row 3: $3(7) + 2(-4) + (-3) = 21 - 8 - 3 = 10$ . ✓ So $(x_1,x_2,x_3) = (7,-4,-3)$ .

Example 2 (a $2\times2$ system with a complex-number twist, and extracting a single unknown). Suppose we only need $x_2$ from the complex-coefficient system

\begin{cases} (1+i)x_1 + 2x_2 = 3 - i\\ 3x_1 + (2-i)x_2 = 4 \end{cases} \quad A = \begin{pmatrix} 1+i & 2 \\ 3 & 2-i\end{pmatrix},\quad \mathbf b = \begin{pmatrix}3-i\\4\end{pmatrix}.

This illustrates that Cramer's Rule works verbatim over any field, here $\mathbb C$ . Compute $\det(A)$ :

\det(A) = (1+i)(2-i) - 2(3) = (2 - i + 2i - i^2) - 6 = (2 + i + 1) - 6 = (3+i) - 6 = -3 + i.

We want only $x_2$ , so we need only $\det(A_2(\mathbf b))$ , replacing column 2 with $\mathbf b$ :

A_2(\mathbf b) = \begin{pmatrix}1+i & 3-i\\ 3 & 4\end{pmatrix}, \qquad \det(A_2(\mathbf b)) = (1+i)(4) - (3-i)(3) = (4+4i) - (9 - 3i) = -5 + 7i.

Therefore

x_2 = \frac{-5+7i}{-3+i}.

Rationalize by multiplying by the conjugate $\overline{-3+i} = -3-i$ :

x_2 = \frac{(-5+7i)(-3-i)}{(-3+i)(-3-i)} = \frac{15 + 5i - 21i - 7i^2}{9 - i^2} = \frac{15 - 16i + 7}{9+1} = \frac{22 - 16i}{10} = \frac{11}{5} - \frac{8}{5}i.

This example demonstrates the key practical advantage of Cramer's Rule: to find a single coordinate $x_2$ , we needed only two $2\times2$ determinants ( $\det A$ and $\det A_2(\mathbf b)$ ), never solving for $x_1$ at all.

\det(A) = 0

Scope, edge cases & common traps

Situation	What happens	What to do instead
$\det(A) \neq 0$ , any field $F$	Cramer's Rule applies exactly as stated	Use it, especially to extract one unknown
$\det(A) = 0$ , system consistent	Infinitely many solutions; ratio is $0/0$ , undefined	Row-reduce; parametrize solution set
$\det(A) = 0$ , system inconsistent	No solution; ratio is (nonzero) $/0$ , undefined	Row-reduce; detect inconsistency directly
Non-square system	Rule does not apply at all — no $A_i(\mathbf b)$ makes sense the same way	Use Gaussian elimination / least squares
Large $n$ (numerically)	Technically correct but requires $n+1$ determinants of size $n$ , each costing $O(n!)$ by cofactor expansion (or $O(n^3)$ via LU-based determinant algorithms) — either way, far more arithmetic than one $O(n^3)$ Gaussian elimination that solves for all unknowns at once	Use LU decomposition / Gaussian elimination for numerical work
Symbolic / parametric coefficients	Rule is often the best choice: it isolates one unknown as an explicit rational function of parameters	Use it deliberately for symbolic derivations

The mathematician behind it

Origins · 1 minGabriel Cramer and how this was discovered

Gabriel Cramer

1704–1752 · Geneva, Switzerland — Enlightenment-era mathematics

Gabriel Cramer was a Genevan mathematician, largely self-taught in advanced analysis beyond his formal training, who held the chair of mathematics at the Academy of Geneva from a young age. He corresponded widely with the leading mathematicians of his era, including Johann and Daniel Bernoulli, and traveled extensively across Europe collecting mathematical correspondence and manuscripts, which he used to edit and publish the works of the Bernoulli brothers. His own most influential contribution appeared in an appendix on algebraic curves, where the determinant formula for solving linear systems was presented as a tool for eliminating unknowns among the coefficients of curves, not as an object of study in its own right.

History

- 1704 — Gabriel Cramer is born in Geneva.

- 1729–1750 — Cramer travels through Europe, corresponds with major mathematicians, and edits the collected works of Johann Bernoulli.

- 1750 — Cramer publishes Introduction à l'analyse des lignes courbes algébriques, whose appendix states the determinant rule for solving systems of linear equations, applied there to finding curves through prescribed points.

- Mid-18th century — Related determinant formulas appear independently in the work of Colin Maclaurin (published posthumously, 1748) and later Leibniz's earlier unpublished notes are recognized to anticipate similar ideas, though Cramer's clear general statement is the one that becomes standard.

- 1752 — Cramer dies; the eponymous rule becomes a fixture of linear algebra pedagogy over the following two centuries.

Does Cramer's Rule work for non-square systems?

No. The rule is stated only for square systems $A\mathbf x = \mathbf b$ with $A$ an $n\times n$ matrix, because the construction of $A_i(\mathbf b)$ depends on $A$ having exactly as many columns as $\mathbf b$ has entries, and the proof relies on $A$ being invertible, which requires squareness. For a non-square (over- or under-determined) system, there is no invertible coefficient matrix to appeal to; instead one uses row reduction to find the general solution, or least-squares methods when an exact solution does not exist. There is no direct determinant-ratio analogue for rectangular systems, though related ideas (e.g. the Moore–Penrose pseudoinverse) generalize some of the same intuition.

Is Cramer's Rule ever actually the fastest method to solve a system?

For very small systems, $n=2$ or $n=3$ , it is often the quickest method by hand, since each determinant is a short explicit sum. For symbolic systems where entries are parameters rather than numbers, it can also outperform elimination, because elimination on symbolic entries can produce fraction pile-ups mid-computation while Cramer's Rule produces one clean rational expression per unknown. For large numerical systems, however, its cost (computing $n+1$ determinants of size $n$ ) is prohibitively worse than the $O(n^3)$ cost of Gaussian elimination or LU decomposition, so it is not used in practice at scale.

Can Cramer's Rule be used to determine whether a system has a solution?

Not reliably on its own. If $\det(A) \neq 0$ , the system always has a unique solution, given directly by the rule. If $\det(A) = 0$ , the rule breaks down entirely and gives no information by itself about whether the system is consistent; as the counterexample above shows, both a consistent (infinite-solution) system and an inconsistent system can have $\det(A)=0$ , and distinguishing them requires checking whether $\mathbf b$ lies in the column space of $A$ , typically via row reduction or rank comparison, not via Cramer's Rule.

How does Cramer's Rule relate to matrix inversion?

It is essentially matrix inversion in disguise. Since $\mathbf x = A^{-1}\mathbf b$ and $A^{-1} = \frac{1}{\det A}\mathrm{adj}(A)$ , expanding this matrix-vector product entrywise and recognizing each entry as a cofactor expansion of a determinant is exactly the content of the proof of Cramer's Rule. In other words, Cramer's Rule is what you get when you apply the adjugate formula for the inverse, column by column of the "input" $\mathbf b$ , and interpret the resulting sum as a determinant.

Does Cramer's Rule work over fields other than the real numbers?

Yes, entirely unchanged. The proof uses only field axioms: multilinearity and alternation of the determinant, the cofactor expansion, and division by the nonzero scalar $\det(A)$ , all of which make sense in any field $F$ (rationals, complex numbers, finite fields, etc.). The complex-number worked example above illustrates this directly; the same formula and proof apply verbatim over $\mathbb{F}_p$ for a prime $p$ , which is why Cramer's Rule appears in coding-theory and cryptographic contexts over finite fields.

What is the computational complexity of solving a system via Cramer's Rule?

Using cofactor expansion, computing a single $n\times n$ determinant costs $O(n!)$ arithmetic operations in the worst case, and Cramer's Rule requires computing $n+1$ such determinants ( $\det A$ and each $\det A_i(\mathbf b)$ ), so naive complexity is $O(n \cdot n!)$ . Using more efficient determinant algorithms based on LU decomposition, each determinant can be computed in $O(n^3)$ , bringing total cost to $O(n^4)$ for all $n+1$ determinants — still worse than the $O(n^3)$ cost of solving the whole system at once via Gaussian elimination. This gap widens rapidly as $n$ grows, which is why the rule is a theoretical and small-scale tool rather than a numerical workhorse.

What exactly does

A_i(\mathbf b)

mean, and is the order of operations (which column, which vector) important?

$A_i(\mathbf b)$ is formed by taking the original matrix $A$ and replacing its $i$ -th column — and only its $i$ -th column — with the constant vector $\mathbf b$ from the right-hand side of $A\mathbf x = \mathbf b$ ; all other columns of $A$ stay exactly as they were. The index $i$ must match the unknown being solved for: to find $x_2$ , replace column $2$ , not row $2$ and not column $1$ . Swapping which column is replaced, or replacing a row instead, produces a different (generally unrelated) determinant and an incorrect value for $x_i$ .