GraphMath

Diagonalization

Expressing a matrix through its eigenvector basis

What does diagonalization reveal about a transformation?

If A has enough linearly independent eigenvectors, those eigenvectors form a basis in which A acts by independent scalings. The factorization A = CDC⁻¹ first changes to the eigenvector basis, then applies the diagonal matrix D and finally returns to the original coordinates. For a real symmetric matrix, the eigenvectors can be chosen orthonormally and the factorization becomes A = QDQᵀ.

Key ideas

Diagonalization replaces one coupled transformation by separate actions along eigenvector directions.

  • If A has a basis of eigenvectors, then A = CDC⁻¹, where the columns of C are eigenvectors and D contains the corresponding eigenvalues
  • C⁻¹ changes coordinates to the eigenvector basis, D scales each eigenvector coordinate and C returns to the original basis
  • For uniform scaling, every nonzero vector is an eigenvector and any invertible C is valid
  • Upper-triangular and general non-symmetric matrices may be diagonalizable even when their eigenvectors are not orthogonal
  • A real symmetric matrix can be orthogonally diagonalized as A = QDQᵀ
  • For symmetric matrices, the columns of Q are orthonormal eigenvectors and Q⁻¹ = Qᵀ
  • A projection has eigenvalues 1 and 0, corresponding to preserved and collapsed directions
  • A reflection has eigenvalues 1 and −1, corresponding to fixed and reversed directions

Each example compares the direct action of A with the three-step route through the eigenvector basis.

Why does A = CDC⁻¹ simplify the action of A?

In the standard basis, A may mix the coordinate directions. The matrix C⁻¹ rewrites a vector in the eigenvector basis, where each coordinate lies along an eigenvector. The diagonal matrix D then scales those coordinates independently by their eigenvalues, and C converts the result back. The same transformation is therefore represented as a sequence of a basis change, diagonal scaling and the inverse basis change.

Related chapters

Chapter contents

The PDF page links below are best-effort: most browsers support them, but some viewers may ignore the page hint.

Topic Pages
Uniform scaling 1
Upper-triangular 2
General non-symmetric 3
Symmetric positive λ 4
Symmetric negative λ 5–6
Symmetric mixed-sign λ 6–7
Projection 7–8
Reflection 8–9
Diagonalizable matrices with λ ∈ ℝ 10–12

What is special about orthogonal diagonalization?

For a real symmetric matrix, eigenvectors belonging to distinct eigenvalues are orthogonal and can be normalized to form an orthogonal matrix Q. Because Q⁻¹ = Qᵀ, the general factorization A = CDC⁻¹ becomes A = QDQᵀ. Geometrically, Qᵀ rotates or reflects coordinates into an orthonormal eigenvector basis, D applies independent scalings and Q returns to the original coordinates without introducing shear.

Was this chapter helpful?

Quick feedback helps us improve the site.