The inverse of a square matrix is the matrix analogue of reciprocal numbers. For a nonzero scalar \(a\), the inverse is \(a^{-1}=1/a\). For a square matrix \(A\), the inverse \(A^{-1}\) is defined by
\[ AA^{-1}=A^{-1}A=I. \]
Not every matrix has an inverse. Understanding when inversion is possible, how to compute it, and when direct inversion is numerically risky is essential.
Existence and Uniqueness
For \(A\in\mathbb{R}^{n\times n}\), the following are equivalent:
- \(A\) is invertible.
- \(\det A\ne 0\).
- \(\operatorname{rank}(A)=n\).
- The linear system \(A\mathbf{x}=\mathbf{b}\) has a unique solution for every \(\mathbf{b}\).
If an inverse exists, it is unique.
Geometric Meaning
A matrix defines a linear map. Invertibility means the map is bijective: no direction is collapsed, and every target vector has exactly one preimage.
In 2D and 3D, \(\det A\) measures signed area or volume scaling. If \(\det A=0\), area or volume collapses to zero and no inverse can exist.
Closed Form in 2x2
For \(A=\begin{bmatrix}a&b\\c&d\end{bmatrix}\),
\[ \det A = ad-bc. \]
If \(ad-bc\ne0\), then
\[ A^{-1}=\frac{1}{ad-bc} \begin{bmatrix} d & -b\\ -c & a \end{bmatrix}. \]
Example
\[ A=\begin{bmatrix}4&7\\2&3\end{bmatrix}, \qquad \det A=4\cdot3-7\cdot2=-2. \]
Hence
\[ A^{-1}=-\frac{1}{2} \begin{bmatrix} 3 & -7\\ -2 & 4 \end{bmatrix}. \]
Gauss-Jordan Method
For practical computation, Gauss-Jordan elimination is the standard constructive method. Start with the augmented matrix
\[ [A\mid I], \]
then apply elementary row operations until the left block becomes \(I\):
\[ [A\mid I]\ \longrightarrow\ [I\mid A^{-1}]. \]
Allowed row operations:
- Swap two rows.
- Multiply a row by a nonzero scalar.
- Add a multiple of one row to another row.
Worked Gauss-Jordan Example
With \(A=\begin{bmatrix}4&7\\2&3\end{bmatrix}\):
\[ \left[\begin{array}{cc|cc} 4&7&1&0\\ 2&3&0&1 \end{array}\right] \xrightarrow{R_2\leftarrow 2R_2-R_1} \left[\begin{array}{cc|cc} 4&7&1&0\\ 0&-1&-1&2 \end{array}\right] \]
\[ \xrightarrow{R_2\leftarrow -R_2} \left[\begin{array}{cc|cc} 4&7&1&0\\ 0&1&1&-2 \end{array}\right] \xrightarrow{R_1\leftarrow R_1-7R_2} \left[\begin{array}{cc|cc} 4&0&-6&14\\ 0&1&1&-2 \end{array}\right] \]
\[ \xrightarrow{R_1\leftarrow \frac{1}{4}R_1} \left[\begin{array}{cc|cc} 1&0&-\frac{3}{2}&\frac{7}{2}\\ 0&1&1&-2 \end{array}\right] = [I\mid A^{-1}], \]
so
\[ A^{-1}=\begin{bmatrix} -\frac{3}{2} & \frac{7}{2}\\ 1 & -2 \end{bmatrix} =- \frac{1}{2} \begin{bmatrix} 3 & -7\\ -2 & 4 \end{bmatrix}. \]
Adjugate Formula
Another exact formula is
\[ A^{-1}=\frac{1}{\det A}\,\operatorname{adj}(A), \]
where \(\operatorname{adj}(A)\) is the transpose of the cofactor matrix. This is important theoretically, but for larger matrices it is usually less practical than elimination or factorization methods.
Using the Inverse to Solve Systems
For \(A\mathbf{x}=\mathbf{b}\) with invertible \(A\):
\[ \mathbf{x}=A^{-1}\mathbf{b}. \]
Conceptually this is clean. Numerically, direct solve methods (for example Gaussian elimination or LU decomposition) are often preferred over computing \(A^{-1}\) explicitly.
Core Inverse Identities
For invertible matrices of compatible sizes:
\[ (AB)^{-1}=B^{-1}A^{-1},\qquad (A^{\mathsf T})^{-1}=(A^{-1})^{\mathsf T},\qquad (A^k)^{-1}=(A^{-1})^k\ (k\in\mathbb{N}). \]
The order reversal in \((AB)^{-1}=B^{-1}A^{-1}\) is a frequent source of mistakes.
Numerical Stability and Conditioning
In floating-point arithmetic, inversion can amplify errors when \(A\) is ill-conditioned. A common indicator is the condition number \(\kappa(A)=\|A\|\,\|A^{-1}\|\): large \(\kappa(A)\) means high sensitivity.
Practical consequences:
- If \(\det A\) is very close to zero, numerical inversion is unstable.
- Solving \(A\mathbf{x}=\mathbf{b}\) via decomposition is often more robust.
- Pivoting in elimination is essential for stability.
Common Pitfalls
- Trying to invert non-square matrices in the usual sense.
- Assuming \(AB=BA\) while manipulating inverse expressions.
- Dividing by \(\det A\) without checking if it is zero or numerically tiny.
- Computing explicit inverses where a direct solver is better.
Rectangular Matrices: What Changes?
Non-square matrices do not have a two-sided inverse. In least-squares settings, one typically uses the Moore-Penrose pseudoinverse \(A^+\) instead. This is a different object with different guarantees.