Matrix Inverse via Cofactors

⚡ Key Takeaways

Matrix Inverse via Cofactors computes the inverse of a square matrix $A$ as $A^{-1} = \frac{1}{\det(A)}\operatorname{adj}(A)$ , where $\operatorname{adj}(A)$ is the adjugate — the transpose of the matrix of signed minors (cofactors). Works for any invertible $n \times n$ matrix over a field (or commutative ring with invertible determinant). Classic result formalized in the 18th–19th century; foundational in linear algebra courses, Cramer's rule, and symbolic computation.

💡 The Core Insight

The product of a row of $A$ with the corresponding column of the cofactor matrix reproduces the cofactor (Laplace) expansion of the determinant. The product with a non-corresponding column reproduces the determinant of a matrix with a repeated row — which is always zero. These two facts — "right answer on the diagonal, zero off the diagonal" — are exactly the statement $A\,\operatorname{adj}(A) = \det(A) I$ . Dividing by $\det(A)$ gives the inverse.

DefinitionMinor, Cofactor, and Adjugate

Let $A = (a_{ij})$ be an $n \times n$ matrix over a field $F$ . The $(i,j)$ minor $M_{ij}$ is the determinant of the $(n-1)\times(n-1)$ matrix obtained by deleting row $i$ and column $j$ of $A$ . The $(i,j)$ cofactor is the signed minor:

1

\quad C_{ij} = (-1)^{i+j} M_{ij}.

Derivation hint: the sign $(-1)^{i+j}$ is the parity contributed by moving entry $(i,j)$ to the top-left corner via row/column swaps in the determinant's permutation sum. The cofactor matrix is $C = (C_{ij})$ . The adjugate (classical adjoint) is its transpose:

2

\quad \operatorname{adj}(A) = C^{T}, \qquad \big(\operatorname{adj}(A)\big)_{ij} = C_{ji}.

Derivation hint: transposition is required so that the row–column pairing in $A\,\operatorname{adj}(A)$ pairs row $i$ of $A$ with the cofactors of row $i$ , reconstructing a Laplace expansion.

TheoremCofactor Inverse Formula

For any $n \times n$ matrix $A$ ,

3

\quad A\,\operatorname{adj}(A) = \operatorname{adj}(A)\,A = \det(A)\, I_n.

Derivation hint: the diagonal entries are Laplace expansions of $\det(A)$ ; off-diagonal entries are Laplace expansions of a determinant with a duplicated row, hence zero. If $\det(A) \neq 0$ , then $A$ is invertible and

4

\quad A^{-1} = \frac{1}{\det(A)}\,\operatorname{adj}(A).

Derivation hint: multiply [3] on the left or right by $\frac{1}{\det(A)}I$ and use uniqueness of the two-sided inverse. Entrywise, this reads

5

\quad \big(A^{-1}\big)_{ij} = \frac{C_{ji}}{\det(A)} = \frac{(-1)^{i+j} M_{ji}}{\det(A)}.

Derivation hint: read off entry $(i,j)$ of $\operatorname{adj}(A)/\det(A)$ using $(\operatorname{adj}A)_{ij}=C_{ji}$ from [2]; note the swapped indices.

ProofThat

A\,\operatorname{adj}(A) = \det(A)\,I

Step 1 — Laplace expansion. Fix a row $i$ . Expanding $\det(A)$ along row $i$ gives

6

\quad \det(A) = \sum_{k=1}^{n} a_{ik}\, C_{ik}.

Derivation hint: group the $n!$ terms of the Leibniz determinant sum by which column the entry from row $i$ occupies; the remaining factor is exactly $C_{ik}$ .

Step 2 — Diagonal entries. The $(i,i)$ entry of $A\,\operatorname{adj}(A)$ is

7

\quad \big(A\,\operatorname{adj}(A)\big)_{ii} = \sum_{k=1}^{n} a_{ik}\,(\operatorname{adj}A)_{ki} = \sum_{k=1}^{n} a_{ik}\,C_{ik} = \det(A).

Derivation hint: substitute $(\operatorname{adj}A)_{ki}=C_{ik}$ from [2], then apply [6].

Step 3 — Off-diagonal entries. For $i \neq j$ ,

8

\quad \big(A\,\operatorname{adj}(A)\big)_{ij} = \sum_{k=1}^{n} a_{ik}\,(\operatorname{adj}A)_{kj} = \sum_{k=1}^{n} a_{ik}\,C_{jk}.

Derivation hint: again use [2], now pairing row $i$ entries with cofactors of row $j$ . This last sum is the Laplace expansion along row $j$ of the matrix $A'$ formed from $A$ by replacing row $j$ with a copy of row $i$ . Since $C_{jk}$ depends only on rows $\neq j$ , it equals the cofactor of $A'$ in position $(j,k)$ , so

9

\quad \sum_{k=1}^{n} a_{ik}\,C_{jk} = \det(A') = 0,

because $A'$ has two identical rows (rows $i$ and $j$ ). Derivation hint: a determinant with two equal rows vanishes — swapping them negates the determinant yet leaves it unchanged, forcing it to be $0$ (char $\neq 2$ ; the result holds in all characteristics via the alternating Leibniz formula).

Step 4 — Combine. Equations [7] and [9] give $\big(A\,\operatorname{adj}(A)\big)_{ij} = \delta_{ij}\det(A)$ , i.e. $A\,\operatorname{adj}(A)=\det(A)I$ . The identity $\operatorname{adj}(A)\,A=\det(A)I$ follows identically by expanding along columns instead of rows. $\blacksquare$

∎

⚠️ The identity is universal; the inverse is conditional

Theorem [3] holds for every square matrix, singular or not — it is a polynomial identity in the entries $a_{ij}$ . Only [4] requires $\det(A) \neq 0$ . When $\det(A)=0$ , $\operatorname{adj}(A)$ still exists but $A^{-1}$ does not; instead $A\,\operatorname{adj}(A)=0$ .

mathematician · Pierre-Simon Laplace | 1749–1827

Laplace's 1772 memoir on determinants gave the systematic cofactor expansion (Laplace expansion) that underpins this entire method. While determinants trace to Leibniz and Cramer, Laplace's recursive expansion along a row or column is the engine that connects minors to the determinant and, through the adjugate, to the inverse. The term "adjugate" and the modern $A^{-1}=\operatorname{adj}(A)/\det(A)$ packaging crystallized through Cauchy and later 19th-century algebraists.

🧠 Intuition

Think of $\det(A)$ as the signed volume scaling factor of the linear map $A$ . The inverse map must rescale volumes by $1/\det(A)$ — that explains the denominator. The numerator, $\operatorname{adj}(A)$ , encodes "how each output coordinate is built from sub-volumes." Each cofactor $C_{ij}$ measures the volume of the parallelepiped spanned by all columns except column $j$ , projected appropriately. Solving $Ax=b$ by Cramer's rule replaces column $j$ of $A$ with $b$ — the cofactors are precisely the partial derivatives of $\det$ with respect to the matrix entries: $\partial \det(A)/\partial a_{ij} = C_{ij}$ . The adjugate is therefore the gradient of the determinant, reshaped.

Figure

Cofactor structure of a 3×3 matrix: deleting row 2, column 1 to form minor M₂₁

Example2×2 inverse

Given:

10

\quad A = \begin{pmatrix} 4 & 7 \\ 2 & 6 \end{pmatrix}.

Derivation hint: a generic invertible $2\times 2$ matrix; entries chosen for a clean nonzero determinant.

Step 1 — Determinant: $\det(A) = 4\cdot 6 - 7\cdot 2 = 24 - 14 = 10.$

Step 2 — Cofactors. For $2\times 2$ , $C_{11}=a_{22}=6$ , $C_{12}=-a_{21}=-2$ , $C_{21}=-a_{12}=-7$ , $C_{22}=a_{11}=4$ . So

11

\quad C = \begin{pmatrix} 6 & -2 \\ -7 & 4 \end{pmatrix}, \qquad \operatorname{adj}(A)=C^{T}=\begin{pmatrix} 6 & -7 \\ -2 & 4 \end{pmatrix}.

Derivation hint: the $2\times2$ adjugate is "swap diagonal, negate off-diagonal," which is exactly $C^T$ here.

Step 3 — Inverse:

12

\quad \boxed{\,A^{-1} = \frac{1}{10}\begin{pmatrix} 6 & -7 \\ -2 & 4 \end{pmatrix} = \begin{pmatrix} 0.6 & -0.7 \\ -0.2 & 0.4 \end{pmatrix}\,}

Derivation hint: scale $\operatorname{adj}(A)$ by $1/\det(A)=1/10$ .

Check: $A A^{-1}=\begin{pmatrix}4(0.6)+7(-0.2)&4(-0.7)+7(0.4)\\2(0.6)+6(-0.2)&2(-0.7)+6(0.4)\end{pmatrix}=\begin{pmatrix}1&0\\0&1\end{pmatrix}.$ ✓ (dimensionless)

Example3×3 inverse

Given:

13

\quad A = \begin{pmatrix} 1 & 2 & 3 \\ 0 & 1 & 4 \\ 5 & 6 & 0 \end{pmatrix}.

Derivation hint: a standard textbook matrix with an easy-to-track structure.

Step 1 — Determinant (expand along row 1):

14

\quad \det(A) = 1(1\cdot0-4\cdot6) - 2(0\cdot0-4\cdot5) + 3(0\cdot6-1\cdot5) = -24 + 40 - 15 = 1.

Derivation hint: each bracket is a $2\times2$ minor; signs alternate $+,-,+$ along the row.

Step 2 — Cofactor matrix. Compute all nine signed minors:

15

\quad C = \begin{pmatrix} -24 & 20 & -5 \\ 18 & -15 & 4 \\ 5 & -4 & 1 \end{pmatrix}.

Derivation hint: e.g. $C_{11}=+\det\!\begin{pmatrix}1&4\\6&0\end{pmatrix}=-24$ , $C_{12}=-\det\!\begin{pmatrix}0&4\\5&0\end{pmatrix}=20$ , $C_{23}=-\det\!\begin{pmatrix}1&2\\5&6\end{pmatrix}=4$ ; apply signs $(-1)^{i+j}$ throughout.

Step 3 — Adjugate (transpose) and inverse:

16

\quad \operatorname{adj}(A)=C^{T}=\begin{pmatrix} -24 & 18 & 5 \\ 20 & -15 & -4 \\ -5 & 4 & 1 \end{pmatrix}.

Derivation hint: transpose [15]; rows of $C$ become columns of $\operatorname{adj}(A)$ . Since $\det(A)=1$ :

17

\quad \boxed{\,A^{-1}=\begin{pmatrix} -24 & 18 & 5 \\ 20 & -15 & -4 \\ -5 & 4 & 1 \end{pmatrix}\,}

Derivation hint: dividing by $\det(A)=1$ leaves $\operatorname{adj}(A)$ unchanged.

Check (first column): $A \cdot (-24,20,-5)^T = (-24+40-15,\,0+20-20,\,-120+120+0)^T = (1,0,0)^T.$ ✓ (dimensionless)

⚠️ Frequent errors

Forgetting the transpose. $A^{-1}=C/\det(A)$ is wrong; it must be $C^{T}/\det(A)$ . The mistake is invisible for symmetric cofactor matrices, so test cases hide it.
Sign of the checkerboard. The pattern $(-1)^{i+j}$ starts with $+$ in the top-left and alternates. Index from $1$ , not $0$ .
Index swap in [5]. Entry $(i,j)$ of $A^{-1}$ uses $C_{ji}$ , not $C_{ij}$ — the transpose again.
" $\det(A)=0$ means undefined cofactors." No — cofactors and $\operatorname{adj}(A)$ are perfectly defined; only the division fails.
Efficiency. Cofactor expansion of an $n\times n$ inverse costs $O(n!\cdot n)$ naively — catastrophically slow. Use it for theory and tiny matrices, not numerics.

CounterexampleThe method cannot rescue a singular matrix

Let $A=\begin{pmatrix}1&2\\2&4\end{pmatrix}$ , so $\det(A)=1\cdot4-2\cdot2=0$ . The adjugate is $\operatorname{adj}(A)=\begin{pmatrix}4&-2\\-2&1\end{pmatrix}$ , and indeed

18

\quad A\,\operatorname{adj}(A)=\begin{pmatrix}1&2\\2&4\end{pmatrix}\begin{pmatrix}4&-2\\-2&1\end{pmatrix}=\begin{pmatrix}0&0\\0&0\end{pmatrix}=\det(A)\,I.

Derivation hint: identity [3] forces the product to be $0\cdot I$ ; you cannot divide by $\det(A)=0$ , confirming $A$ has no inverse. Attempting $\operatorname{adj}(A)/\det(A)$ divides by zero — there is no inverse, consistent with rows being linearly dependent.

connection

The cofactor formula sits at the crossroads of several core results. It is the engine behind Cramer's Rule, since $x_j = \det(A_j)/\det(A)$ is exactly $(\operatorname{adj}(A)\,b)_j/\det(A)$ . It depends entirely on the Laplace Expansion of the determinant and on properties of the Determinant as an alternating multilinear form. The identity $\partial\det(A)/\partial a_{ij}=C_{ij}$ links it to Jacobi's Formula for the derivative of a determinant. For practical computation it is contrasted with Gaussian Elimination and LU Decomposition, which compute inverses in $O(n^3)$ . Over rings rather than fields, the adjugate appears in the Cayley–Hamilton Theorem, where $\operatorname{adj}(xI-A)$ generates the characteristic polynomial's coefficients.

🔬 Where this appears

⏳ History

- 1683 — Seki Takakazu (Japan) independently develops determinant-like quantities for solving simultaneous equations.

- 1693 — Gottfried Leibniz writes to L'Hôpital describing determinants and the systematic elimination of variables.

- 1750 — Gabriel Cramer publishes Cramer's Rule, expressing solutions as ratios of determinants — the conceptual seed of the adjugate.

- 1772 — Pierre-Simon Laplace introduces the cofactor (Laplace) expansion of a determinant along any row or column.

- 1812 — Augustin-Louis Cauchy gives the first rigorous, unified theory of determinants, proves the product rule

\det(AB)=\det A\det B

, and systematizes minors and cofactors.

- 1841 — Carl Gustav Jacob Jacobi develops the calculus of determinants and the functional determinant (Jacobian), formalizing

\partial\det/\partial a_{ij}=C_{ij}

.

- 1858 — Arthur Cayley founds matrix algebra proper in A Memoir on the Theory of Matrices, defining the inverse and stating what becomes the Cayley–Hamilton theorem.

- 1880s — Ferdinand Georg Frobenius consolidates rank, adjugate, and determinant theory into the modern framework.

- 20th century — recognition that cofactor methods, while exact, are computationally infeasible for large

n

, driving the dominance of elimination-based algorithms.

open-problem · Boundaries and refinements

The adjugate behaves richly near singularity: if $A$ has rank $n-1$ , then $\operatorname{rank}(\operatorname{adj}(A))=1$ ; if $\operatorname{rank}(A)\le n-2$ , then $\operatorname{adj}(A)=0$ . This makes $\operatorname{adj}$ a sensitive but non-continuous-in-rank probe of degeneracy. Numerically, computing inverses via cofactors amplifies round-off and scales factorially, so the genuine open practical questions concern stable, fast exact inversion over rings (where one cannot pivot freely by division): Bareiss-style fraction-free elimination and division-free determinant/adjugate algorithms (e.g., Berkowitz, Mahajan–Vinay) aim to compute $\operatorname{adj}(A)$ in polynomial arithmetic operations without divisions — relevant to complexity classes like $\mathsf{GapL}$ and the algebraic complexity of the determinant. Whether the determinant/adjugate admit polynomial-size formulas (vs. circuits) ties into Valiant's $\mathsf{VP}$ vs. $\mathsf{VNP}$ program, one of the central open problems of algebraic complexity.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also