Orthogonal Diagonalization
You'll learn when a matrix can be diagonalized using an orthogonal matrix, and why this special property belongs to symmetric matrices.
The Core Idea
Remember that diagonalization means writing a matrix A as A = PDP⁻¹, where P contains eigenvectors and D is diagonal. Orthogonal diagonalization is a premium upgrade: we find P whose columns are orthonormal eigenvectors, making P an orthogonal matrix.
The magic? When P is orthogonal, P⁻¹ = Pᵀ (its inverse equals its transpose), so we get the elegant form:
A = PDPᵀ
This is much cleaner computationally and numerically stable.
When Can We Orthogonally Diagonalize?
Here's the beautiful theorem: A matrix is orthogonally diagonalizable if and only if it is symmetric (meaning A = Aᵀ).
Why does symmetry matter? Symmetric matrices have two crucial properties:
- All eigenvalues are real
- Eigenvectors corresponding to different eigenvalues are automatically orthogonal
Even when eigenvalues repeat, you can always find an orthogonal basis of eigenvectors for each eigenspace (using Gram-Schmidt if needed).
The Process
To orthogonally diagonalize a symmetric matrix A:
- Find all eigenvalues (they'll be real)
- Find eigenvectors for each eigenvalue
- If an eigenspace has dimension > 1, apply Gram-Schmidt within it
- Normalize all eigenvectors to length 1
- Assemble these orthonormal eigenvectors as columns of P
An Analogy
Think of regular diagonalization as finding "natural axes" for your transformation. Orthogonal diagonalization ensures these axes are perpendicular to each other—like the standard x, y, z axes, but possibly rotated. Only symmetric matrices respect this geometric harmony.
Key Takeaway: A matrix can be orthogonally diagonalized (written as A = PDPᵀ with orthogonal P) if and only if it is symmetric, guaranteeing real eigenvalues and orthogonal eigenvectors.