Course contentsShow
← Mathematics
Lesson 2297 of 3,12147. Linear Algebra — Applications and Advanced TopicsFree lesson

Positive Semidefinite Matrices

Study matrices where x^T A x ≥ 0 for all x, and their geometric and algebraic properties.

Positive Semidefinite Matrices

What you'll learn: You'll understand positive semidefinite matrices as a relaxation of positive definite matrices, where the quadratic form is allowed to be zero for some nonzero vectors.

The Core Idea

A positive semidefinite matrix is a symmetric matrix A where the quadratic form x^T A x ≥ 0 for all vectors x. Notice the "≥" instead of the strict ">" you saw with positive definite matrices. This means the quadratic form never dips below zero, but it can equal zero even when x ≠ 0.

Think of it like a bowl that might have a flat bottom instead of being strictly rounded—objects can rest at equilibrium (zero energy) without being at a single unique point.

Key Properties

Eigenvalue Characterization

A symmetric matrix A is positive semidefinite if and only if all its eigenvalues are ≥ 0. Some eigenvalues may be exactly zero, which distinguishes it from the positive definite case (where all eigenvalues must be strictly positive).

Geometric Interpretation

For a positive semidefinite matrix, the level surfaces of the quadratic form x^T A x = c are ellipsoids (possibly degenerate) that touch the origin. The zero eigenvalues correspond to directions where the quadratic form collapses—imagine an ellipsoid flattened into an ellipse or line.

Rank Connection

If A has rank r < n, then exactly n - r eigenvalues are zero. The matrix is singular (non-invertible) when positive semidefinite but not positive definite.

Why It Matters

Positive semidefinite matrices arise naturally in optimization, covariance matrices (which may be singular when variables are linearly dependent), and Gram matrices from inner products. They represent "non-negative energy" constraints without requiring strict positivity.

Key Takeaway: Positive semidefinite matrices satisfy x^T A x ≥ 0 for all x, have all eigenvalues ≥ 0, and generalize positive definite matrices to allow zero quadratic form values for nonzero vectors.