Eigenvalues and Eigenvectors
Table of contents
1. Properties
- $ \sum \lambda = tr(A) $
- $ \prod \lambda = det(A) $
- Symmetric $S$ has real eigenvalues, and can choose orthogonal eigenvectors.
- $\lambda_1 \ne \lambda_2 \implies \mathbf{x}_1 \cdot \mathbf{x}_2 = 0$ : different values, independent vectors
- $\lambda_1 = \lambda_2 \xrightarrow{\text{?}}$ independent vectors (might or might not)
- $\mathbf{x}_i$ : orthogonal $\iff A^TA=AA^T$ (where $A$ : real)
2. Note that
- (×) $\lambda(A+B) = \lambda(A) + \lambda(B)$
- (×) $\lambda(AB) = \lambda(A) \times \lambda(B)$
3. Similar Matrix
: different matrix with same eigenvalues
\[\forall B \text{ : invertible,} \quad BAB^{-1} \text{ is similar to } A\]$ (BAB^{-1})(B\mathbf{x}) = BA\mathbf{x} = B(\lambda\mathbf{x}) = \lambda(B\mathbf{x}) $ : $\lambda$ is also an eigenvalue of $BAB^{-1}$
4. Diagonalization
(1) Definition
\[\exists P, D \quad \text{s.t.} \quad P^{-1}AP = D \quad (P \text{ : invertible, } D \text{ : diagonal})\](2) iff condition
: has a full $n$ independent eigenvectors
(3) Sketch
\[AX = X\Lambda \rightarrow A = X\Lambda X^{-1}\] \[A \begin{bmatrix} & & \\ \mathbf{x}_{1} & \cdots & \mathbf{x}_{n} \\ & & \\ \end{bmatrix} = \begin{bmatrix} & & \\ A\mathbf{x}_{1} & \cdots & A\mathbf{x}_{n} \\ & & \\ \end{bmatrix} = \begin{bmatrix} & & \\ \lambda_1\mathbf{x}_{1} & \cdots & \lambda_n\mathbf{x}_{n} \\ & & \\ \end{bmatrix} = \begin{bmatrix} & & \\ \mathbf{x}_{1} & \cdots & \mathbf{x}_{n} \\ & & \\ \end{bmatrix} \begin{bmatrix} \lambda_1 & & \\ & \ddots & \\ & & \lambda_n\\ \end{bmatrix}\](4) Uniqueness
Diagonalization is unique upto
① permutation (on diagonal $D$)
② scalar multiplication (on eigenmatrix $X$)