Toolbox

T1: Eigenvectors of a Symmetric Matrix

Reminder: This post contains 1240 words · 4 min read · by Xianbin

Theorem 1\textbf{Theorem 1}. Eigenvectors of a symmetric matrix with distinct eigenvalues are orthogonal.

Proof\textbf{Proof}.

(To do later).

Diagonalizability

Theorem\textbf{Theorem}. A matrix An×nA_{n\times n} is diagonalizable if and only if AA has a complete set of eigenvectors, i.e, nn linearly independent eigenvectors for AA.

By Gram-Schmidt method, we can always transform these nn linearly independent eigenvectors into nn orthonormal basis.

Spectral Theorem

Example

Let AA be an n×nn\times n symmetric real matrix. We can rewrite AA as follows.

(v1vn)(λ1λn)(v1vn)\left( \begin{matrix} \vert && \cdots && \vert \\ \textbf{v}_1 && \cdots && \textbf{v}_n\\ \vert && \cdots && \vert \end{matrix} \right) \left( \begin{matrix} \lambda_1 \\ && \dots\\ &&&& \lambda_n \end{matrix} \right) \left( \begin{matrix} -&\textbf{v}_1 &- \\ -& \vdots &-\\ -& \textbf{v}_n &- \end{matrix} \right)

where vi\textbf{v}_i are orthonormal vectors.

Theorem\textbf{Theorem}[Spectral Decomposition]. Given a matrix An×nA_{n\times n} with spectrum σ(A)={λ1,,λk}\sigma(A) = \{\lambda_1,\ldots, \lambda_k\} is diagonalizable iff there exists matrices {G1,,Gk}\{G_1,\ldots,G_k\} such that

A=λ1G1++λiGkA = \lambda_1 G_1 + \ldots + \lambda_i G_k

where GiG_i satisfies the following properties:

  1. GiGj=0G_i G_j = 0 for any iji\neq j.

  2. Gi=I\sum G_i = \textbf{I}.