What Is a Symmetric Matrix?
A symmetric matrix is a square matrix $A$ that is equal to its own transpose:
$$A^T = A.$$
The transpose $A^T$ is the matrix you get by swapping rows and columns — reflecting every entry across the main diagonal. So saying $A^T = A$ is saying that this reflection changes nothing: the entry $a_{ij}$ equals the entry $a_{ji}$ for every pair of positions.
In entry form, a matrix is symmetric when
$$a_{ij} = a_{ji} \quad \text{for all } i, j.$$
A symmetric matrix must be square, and the main-diagonal entries can be anything — they sit on the mirror line, so they map to themselves. Here is a symmetric $3 \times 3$ matrix; notice each off-diagonal entry appears twice, once on each side of the diagonal:
$$A = \begin{bmatrix} 1 & 7 & 2 \ 7 & 4 & 5 \ 2 & 5 & 9 \end{bmatrix}.$$
What Are the Properties of a Symmetric Matrix?
Symmetric matrices carry a set of properties that make them the best-behaved matrices to work with:
Real eigenvalues. Every eigenvalue of a real symmetric matrix is a real number — never complex. This is the spectral theorem's headline.
Orthogonal eigenvectors. Eigenvectors for distinct eigenvalues are mutually perpendicular, which connects directly to orthogonal matrices.
Orthogonally diagonalizable. A matrix is symmetric if and only if it can be written $A = Q\Lambda Q^T$, where $Q$ is orthogonal and $\Lambda$ is diagonal — the spectral theorem.
Sum stays symmetric. The sum (or difference) of two symmetric matrices is symmetric.
The inverse is symmetric. If a symmetric matrix is invertible, its inverse is symmetric too.
Powers stay symmetric. $A^n$ is symmetric whenever $A$ is.
Product caveat. The product $AB$ of two symmetric matrices is symmetric only if $A$ and $B$ commute ($AB = BA$) — symmetry is not automatically preserved under multiplication.
The real-eigenvalues property is the one that makes symmetric matrices central to physics and statistics — it guarantees the quantities they model behave sensibly.
What Is the Difference Between Symmetric and Skew-Symmetric Matrices?
A skew-symmetric (or antisymmetric) matrix is the mirror-image idea with a sign flip: it equals the negative of its transpose, $A^T = -A$. The two sit side by side:
Feature | Symmetric matrix | Skew-symmetric matrix |
|---|---|---|
Transpose rule | $A^T = A$ | $A^T = -A$ |
Entry condition | $a_{ij} = a_{ji}$ | $a_{ij} = -a_{ji}$ |
Main diagonal | any values | all zeros (since $a_{ii} = -a_{ii}$) |
Eigenvalues | all real | zero or purely imaginary |
The forced-zero diagonal of a skew-symmetric matrix is the quickest way to tell the two apart: a skew-symmetric matrix must have zeros all down the diagonal, while a symmetric one can hold any diagonal values.
What Is the Symmetric and Skew-Symmetric Decomposition?
Here is the result that ties the two together: every square matrix splits uniquely into a symmetric part and a skew-symmetric part. For any square matrix $A$,
$$A = \underbrace{\tfrac{1}{2}\big(A + A^T\big)}{\text{symmetric}} + \underbrace{\tfrac{1}{2}\big(A - A^T\big)}{\text{skew-symmetric}}.$$
The first piece satisfies $\big(\tfrac{1}{2}(A + A^T)\big)^T = \tfrac{1}{2}(A^T + A)$ — unchanged, so symmetric. The second flips sign under transpose, so skew-symmetric. Add them and the transposes cancel back to $A$. This decomposition (sometimes called the Toeplitz decomposition) is why symmetric matrices show up even when the original data isn't symmetric — you can always extract the symmetric part.
Examples of Symmetric Matrix
The set runs from a quick verification, through the most common symmetry-test mistake, to finding an unknown entry, building a symmetric matrix from any matrix, checking skew-symmetry, and the full decomposition.
Example 1
Verify that $A = \begin{bmatrix} 3 & -1 \ -1 & 5 \end{bmatrix}$ is symmetric.
Transpose by swapping rows and columns: $A^T = \begin{bmatrix} 3 & -1 \ -1 & 5 \end{bmatrix}$, which is identical to $A$. Equivalently, $a_{12} = -1 = a_{21}$.
Final answer: $A^T = A$, so $A$ is symmetric. The single off-diagonal value matches its mirror.
Example 2
Is $A = \begin{bmatrix} 2 & 4 & 1 \ 4 & 6 & 3 \ 1 & 5 & 8 \end{bmatrix}$ symmetric?
Wrong attempt. A student glances at the top-left block — $a_{12} = 4 = a_{21}$ — sees the first mirror pair match, and concludes "symmetric" without checking the rest.
Check every mirror pair, not just the first. Look at $a_{23} = 3$ against $a_{32} = 5$. Those do not match, so the symmetry fails — a single mismatched pair is enough to break it.
Correct. Since $a_{23} = 3 \neq 5 = a_{32}$, the matrix is not symmetric.
Final answer: not symmetric. Symmetry requires all mirror pairs to match — one disagreement anywhere is fatal.
Example 3
Find $x$ so that $A = \begin{bmatrix} 5 & x & 2 \ 9 & 1 & 7 \ 2 & 7 & 4 \end{bmatrix}$ is symmetric.
For symmetry, $a_{12} = a_{21}$, which means $x = 9$. Check the other pairs: $a_{13} = 2 = a_{31}$ and $a_{23} = 7 = a_{32}$ already hold.
Final answer: $x = 9$. With that value, every mirror pair matches and the matrix is symmetric.
Example 4
Show that $A^T A$ is symmetric for $A = \begin{bmatrix} 1 & 2 \ 0 & 3 \end{bmatrix}$.
Compute $A^T A$:
$$A^T A = \begin{bmatrix} 1 & 0 \ 2 & 3 \end{bmatrix}\begin{bmatrix} 1 & 2 \ 0 & 3 \end{bmatrix} = \begin{bmatrix} 1 & 2 \ 2 & 13 \end{bmatrix}.$$
The result has $a_{12} = 2 = a_{21}$ — symmetric.
Final answer: $A^T A = \begin{bmatrix} 1 & 2 \ 2 & 13 \end{bmatrix}$, which is symmetric. In fact $A^T A$ is symmetric for any matrix $A$, because $(A^T A)^T = A^T (A^T)^T = A^T A$ — a fact behind least-squares regression.
Example 5
Is $B = \begin{bmatrix} 0 & 3 & -2 \ -3 & 0 & 4 \ 2 & -4 & 0 \end{bmatrix}$ symmetric, skew-symmetric, or neither?
The diagonal is all zeros, a hint toward skew-symmetric. Check: $b_{12} = 3$ and $b_{21} = -3$, so $b_{12} = -b_{21}$. Every off-diagonal pair flips sign.
Final answer: $B$ is skew-symmetric ($B^T = -B$). The all-zero diagonal is the giveaway — a symmetric matrix could have any diagonal, but a skew-symmetric one is forced to zero there.
Example 6
Decompose $A = \begin{bmatrix} 2 & 5 \ 1 & 4 \end{bmatrix}$ into symmetric and skew-symmetric parts.
First $A^T = \begin{bmatrix} 2 & 1 \ 5 & 4 \end{bmatrix}$. The symmetric part is $\tfrac{1}{2}(A + A^T)$ and the skew-symmetric part is $\tfrac{1}{2}(A - A^T)$:
$$\tfrac{1}{2}(A + A^T) = \begin{bmatrix} 2 & 3 \ 3 & 4 \end{bmatrix}, \qquad \tfrac{1}{2}(A - A^T) = \begin{bmatrix} 0 & 2 \ -2 & 0 \end{bmatrix}.$$
Final answer: $A = \begin{bmatrix} 2 & 3 \ 3 & 4 \end{bmatrix} + \begin{bmatrix} 0 & 2 \ -2 & 0 \end{bmatrix}$. Add the two and the off-diagonal $3 \pm 2$ recovers the original $5$ and $1$ — the decomposition checks out.
Why the Symmetric Matrix Earns Its Place
"Why do the matrices that describe real physical quantities almost always come out symmetric?"
The deep result here is the spectral theorem — that every real symmetric matrix has real eigenvalues and an orthonormal set of eigenvectors — credited to Augustin-Louis Cauchy (1789–1857, France), who proved the realness of the eigenvalues in the 1820s while studying the axes of quadric surfaces. That single guarantee is why symmetric matrices anchor so much applied mathematics.
Where symmetric matrices do real work:
Statistics and machine learning. A covariance matrix is symmetric, because the covariance of $X$ with $Y$ equals the covariance of $Y$ with $X$. Principal component analysis — the workhorse of dimensionality reduction — is the spectral decomposition of a symmetric covariance matrix.
Physics and engineering. Stress, strain, and inertia tensors are symmetric matrices; their real eigenvalues are the principal stresses or moments of inertia that engineers actually design around.
Quadratic forms and optimisation. The matrix of second derivatives (the Hessian) is symmetric, and its eigenvalues tell you whether a critical point is a minimum, maximum, or saddle.
Graph theory and networks. The adjacency matrix of an undirected graph is symmetric, because an edge between two nodes runs both ways.
Where Students Trip Up on the Symmetric Matrix
Mistake 1: Checking only one mirror pair
Where it slips in: Verifying symmetry on a $3 \times 3$ or larger matrix.
Don't do this: Confirm one off-diagonal pair matches and call the whole matrix symmetric.
The correct way: Check every mirror pair — $a_{12}$ vs $a_{21}$, $a_{13}$ vs $a_{31}$, $a_{23}$ vs $a_{32}$, and so on. A single mismatched pair breaks symmetry.
Mistake 2: Confusing symmetric with skew-symmetric
Where it slips in: Classifying a matrix with an all-zero diagonal.
Don't do this: Assume an all-zero diagonal means symmetric.
The correct way: An all-zero diagonal points toward skew-symmetric ($A^T = -A$), not symmetric. A symmetric matrix can have any diagonal; the forced-zero diagonal belongs to the skew case. The second-guesser who knows both definitions still mixes the sign — anchor on "skew flips the sign, and its diagonal must vanish."
Mistake 3: Assuming the product of symmetric matrices is symmetric
Where it slips in: Multiplying two symmetric matrices and expecting a symmetric result.
Don't do this: Treat $AB$ as symmetric just because $A$ and $B$ are.
The correct way: $AB$ is symmetric only if $A$ and $B$ commute ($AB = BA$). In general $(AB)^T = B^T A^T = BA \neq AB$. The memorizer who learned "symmetric stays symmetric" overgeneralises from sums (which do stay symmetric) to products (which need not).
Key Takeaways
A symmetric matrix equals its own transpose, $A^T = A$, so $a_{ij} = a_{ji}$ — its entries mirror across the diagonal.
Real symmetric matrices have real eigenvalues and orthogonal eigenvectors, and are orthogonally diagonalizable ($A = Q\Lambda Q^T$).
A skew-symmetric matrix satisfies $A^T = -A$ and must have an all-zero diagonal.
Every square matrix splits uniquely into a symmetric part $\tfrac{1}{2}(A + A^T)$ and a skew-symmetric part $\tfrac{1}{2}(A - A^T)$.
The product of two symmetric matrices is symmetric only when they commute.
Symmetric matrices model covariance, stress tensors, Hessians, and undirected graphs — the reason PCA and finite element analysis depend on them.
Practice These Before Moving On
Is $\begin{bmatrix} 1 & 6 \ 6 & 2 \end{bmatrix}$ symmetric? Justify with the transpose.
Find $y$ so that $\begin{bmatrix} 0 & 4 \ y & 0 \end{bmatrix}$ is skew-symmetric.
Decompose $\begin{bmatrix} 3 & 7 \ 1 & 2 \end{bmatrix}$ into symmetric and skew-symmetric parts.
Answer to Question 1: $A^T = A$ (the $6$s mirror), so yes, symmetric. Answer to Question 2: $y = -4$ (skew needs $a_{21} = -a_{12}$). Answer to Question 3: symmetric part $\begin{bmatrix} 3 & 4 \ 4 & 2 \end{bmatrix}$, skew-symmetric part $\begin{bmatrix} 0 & 3 \ -3 & 0 \end{bmatrix}$. If Question 2 tempted you toward $y = 4$, return to Mistake 2.
Want a live Bhanzu trainer to walk through more symmetric matrix problems? Book a free demo class — online globally.
Was this article helpful?
Your feedback helps us write better content