Symmetric Matrix: A matrix A where A = A^T (equal to its transpose)
Spectral Theorem: Every symmetric matrix A can be orthogonally diagonalized as A = QDQ^T
Orthogonal Matrix: Q^T Q = QQ^T = I (columns are orthonormal vectors)
Real Eigenvalues: Symmetric matrices have only real eigenvalues
- Verify symmetry: Check that A = A^T
- Find eigenvalues: Solve det(A - λI) = 0
- Find eigenvectors: For each eigenvalue, solve (A - λI)v = 0
- Orthogonalize: Apply Gram-Schmidt within each eigenspace
- Normalize: Make eigenvectors unit vectors
- Form matrices: Q from orthonormal eigenvectors, D from eigenvalues
• Spectral decomposition: A = QDQ^T where Q^T = Q^(-1)
• Characteristic equation: det(A - λI) = 0
• Eigenvector equation: Av = λv
• Orthogonality: vᵢ · vⱼ = 0 for i ≠ j
Spectral Decomposition: Express A as A = QDQ^T where Q is orthogonal and D is diagonal
- Verify A is symmetric: A = A^T
- Find characteristic polynomial det(A - λI) = 0
- Find eigenvalues λ
- Find eigenvectors v for each λ
- Normalize eigenvectors to form orthonormal basis
- Form Q from orthonormal eigenvectors, D from eigenvalues
A =
det(A - λI) = det
= (4-λ)(1-λ) - (2)(2) = (4-λ)(1-λ) - 4
= 4 - 4λ - λ + λ² - 4 = λ² - 5λ
= λ(λ - 5)
λ(λ - 5) = 0 → λ₁ = 0, λ₂ = 5
(A - 0I)v = 0 →
This gives: 4x + 2y = 0 and 2x + y = 0, so y = -2x
Eigenvector: v₁ =
(A - 5I)v = 0 →
This gives: -x + 2y = 0, so x = 2y
Eigenvector: v₂ =
||v₁|| = √(1² + (-2)²) = √5
u₁ = v₁/||v₁|| =
||v₂|| = √(2² + 1²) = √5
u₂ = v₂/||v₂|| =
u₁ · u₂ = (1/√5)(2/√5) + (-2/√5)(1/√5) = 2/5 - 2/5 = 0 ✓
A = QDQ^T with Q orthogonal and D diagonal containing eigenvalues
• Symmetry verification: Essential first step
• Orthogonal eigenvectors: Automatic for distinct eigenvalues
• Normalization: Ensures Q is orthogonal
3x3 Symmetric Matrix: Same process extended to higher dimensions
A^T =
det(A - λI) = det
= (3-λ)[(2-λ)(2-λ) - 1] = (3-λ)[(2-λ)² - 1]
= (3-λ)[4 - 4λ + λ² - 1] = (3-λ)[λ² - 4λ + 3]
= (3-λ)(λ - 1)(λ - 3) = -(λ - 3)²(λ - 1)
-(λ - 3)²(λ - 1) = 0 → λ₁ = 1, λ₂ = 3 (multiplicity 2)
(A - I)v = 0 →
This gives: x + y = 0 and z = 0, so y = -x and z = 0
Eigenvector: v₁ =
(A - 3I)v = 0 →
This gives: -x + y = 0, so x = y
Eigenspace is span{
v₁ ·
v₁ ·
A = QDQ^T with repeated eigenvalue handled properly
• Repeated eigenvalues: Use Gram-Schmidt within eigenspace
• Orthogonality: Eigenspaces of different eigenvalues are orthogonal
• Diagonalization: Symmetric matrices are always diagonalizable
Verification: Confirm spectral decomposition by computing QDQ^T and comparing to A
Q =
Q^T =
QD =
QDQ^T =
=
=
QDQ^T = A, confirming our spectral decomposition is correct
• Orthogonal matrix: Q^T = Q^(-1) for symmetric matrices
• Verification: Essential to confirm correctness
• Matrix multiplication: Row by column rule
Quadratic Form: f(x) = x^T Ax where A is symmetric matrix of coefficients
f(x,y) = [x y]
So A =
A = QDQ^T where Q =
Let [x y]^T = Q[u v]^T, so [u v]^T = Q^T[x y]^T
Then f(x,y) = x^T Ax = x^T QDQ^T x = (Q^T x)^T D (Q^T x) = [u v] D [u v]^T
= [u v]
In the new coordinate system (u,v), the quadratic form becomes f(u,v) = 5v²
This is a parabolic cylinder opening along the v-axis
Since one eigenvalue is 0, the form is positive semidefinite
The quadratic form represents a parabolic cylinder aligned with the eigenvector directions
• Principal axes: Change to coordinate system of eigenvectors
• Quadratic form simplification: Eliminates cross terms
• Classification: Based on eigenvalue signs
Matrix Powers via Spectral Theorem: A^n = QD^nQ^T is extremely efficient for symmetric matrices
Q =
Since D is diagonal, D⁶ =
First compute QD⁶:
Then compute (QD⁶)Q^T:
=
=
A⁶ computed efficiently using spectral decomposition
• Power efficiency: A^n = QD^nQ^T avoids n-1 matrix multiplications
• Diagonal powers: Just raise diagonal elements to power
• Orthogonal property: Q^T = Q^(-1) simplifies computation