Home Statistical Dictionary Symmetric Matrix

Symmetric Matrix

A square matrix that equals its own transpose -- the structural property shared by every covariance matrix, correlation matrix, and matrix of the form X'X, guaranteeing real eigenvalues and an orthogonal eigenvector basis.

Linear Algebra for StatisticsIntermediateFree

In Plain English

A symmetric matrix looks the same whether you read it by rows or by columns, flip it across its diagonal and nothing changes. This isn't a niche special case in statistics, it's the default: covariance matrices, correlation matrices, and the core matrix behind least squares regression are all automatically symmetric, which is exactly what guarantees the well-behaved eigenvalues and eigenvectors that spectral decomposition and PCA depend on.

Definition

A square matrix $A$ is symmetric if it equals its own transpose, $A = A^{⊤}$ , equivalently $a_{ij} = a_{ji}$ for every pair $i, j$ . Symmetric matrices arise constantly in statistics because any matrix of the form $X^{⊤} X$ is automatically symmetric for any $X$ , which is why covariance matrices, correlation matrices, and the normal-equations matrix in regression are all symmetric by construction, and this structure is what guarantees, via the spectral theorem, that a symmetric matrix has only real eigenvalues and an orthogonal basis of eigenvectors -- the theoretical foundation underlying spectral decomposition, PCA, and the interpretation of a covariance matrix's principal axes.

Formula

A = A^{⊤} ⟺ a_{ij} = a_{ji} \forall i, j

Notation

A^{⊤}

—the transpose of A

a_{ij}

—the (i,j) entry of A

Properties

Every covariance matrix and correlation matrix is symmetric by definition, since the covariance between variables i and j equals the covariance between variables j and i -- symmetry is not an assumption imposed on these matrices, it's a direct consequence of what covariance measures.
The spectral theorem guarantees that any real symmetric matrix has exclusively real eigenvalues (never complex) and can be diagonalized by an orthogonal matrix of eigenvectors -- this single fact is the mathematical reason PCA's principal components are always real, mutually orthogonal directions rather than some more exotic complex-valued construction.
An n x n symmetric matrix has only n(n+1)/2 independent entries rather than the full n^2, since every off-diagonal pair (i,j) and (j,i) is forced to match -- this redundancy is why storing or estimating a covariance matrix scales roughly with the square of the number of variables divided by two, not the full square.

At a Glance

Defining propertyA = A^T (mirrors across the main diagonal)

Always symmetricany matrix of the form X'X, all covariance/correlation matrices

Guarantees (spectral theorem)real eigenvalues, orthogonal eigenvector basis

Independent entriesn(n+1)/2, not n^2, for an n x n symmetric matrix

Last updated September 8, 2026← Back to the Dictionary