8.812 min read

The Spectral Theorem

The Spectral Theorem is the cornerstone of this chapter: every symmetric matrix can be diagonalized by an orthogonal matrix. Specifically, for any symmetric $n \times n$ matrix $A$ , there exists an orthogonal matrix $U$ (meaning $U^\top U = I$ ) and a diagonal matrix $\Lambda = \text{diag}(\lambda_1, \ldots, \lambda_n)$ such that $A = U \Lambda U^\top$ .

The columns $\mathbf{u}_1, \ldots, \mathbf{u}_n$ of $U$ form an orthonormal basis of $\mathbb{R}^n$ , called the principal axes of $A$ . The diagonal entries $\lambda_1, \ldots, \lambda_n$ of $\Lambda$ are called the eigenvalues of $A$ . The eigenvalues are always real (never complex) for symmetric matrices — this is a key structural fact.

Geometrically, the decomposition $A = U \Lambda U^\top$ says: rotating to the principal-axis frame (via $U^\top$ ), scaling independently along each axis (via $\Lambda$ ), then rotating back (via $U$ ). This makes symmetric matrices far easier to analyze than general matrices.

Formal View

Theorem 8.1 (Spectral Theorem) — Eigendecomposition of Symmetric Matrices

Let

A

be a real symmetric

n \times n

matrix. Then:\n1. All eigenvalues of

A

are real.\n2. Eigenvectors for distinct eigenvalues are orthogonal.\n3.

A

can be written as

A = U \Lambda U^\top

, where

U

is orthogonal (

U^\top U = I

) and

\Lambda = \operatorname{diag}(\lambda_1, \ldots, \lambda_n)

This decomposition is called the spectral decomposition or eigendecomposition of $A$ . MATLAB: `[U, Lambda] = eig(A)`.

Interactive Visualization

Eigenvector Explorer

Why This Matters

The Spectral Theorem is what makes symmetric matrices so powerful — it guarantees a clean coordinate system where everything diagonalizes.

PCA relies on the spectral theorem: the covariance matrix is symmetric, so it has orthogonal eigenvectors (principal components).
Vibration analysis: natural frequencies of a structure are eigenvalues of a symmetric stiffness matrix.
Quantum mechanics: observables are represented by symmetric (Hermitian) operators, whose eigenvalues are the measurable quantities.

Learning Resources

Eigenvectors and eigenvalues

3Blue1Brown

Visual introduction to what eigenvectors and eigenvalues mean geometrically.

17 min

Spectral theorem for symmetric matrices

MIT OpenCourseWare

Gilbert Strang proves the spectral theorem and discusses its geometric meaning.

45 min

Quiz

Question 1

The spectral theorem guarantees that every symmetric matrix has real eigenvalues.

Question 2

In the spectral decomposition $A = U \Lambda U^\top$ , what does $U^\top U = I$ tell us about $U$ ?

Common Mistakes

Confusing the spectral theorem (for symmetric matrices) with the general diagonalization theorem (which requires $n$ linearly independent eigenvectors but not necessarily orthonormal ones).
Forgetting that $U$ is orthogonal, so $U^{-1} = U^\top$ — this simplifies many calculations.