JOVANA
Library Glossary Getting Started Three Levels Fields How it works Mission
Join the mission
All guides

Simultaneous spectral decomposition and the payoff

When do several normal operators share one orthonormal eigenbasis? Commuting is the answer. We close with the commutant, joint diagonalization, and the applications — PCA, quantum measurement, vibrations — that the spectral theorem powers.

One orthonormal basis for many operators

Each normal operator diagonalizes in its *own* orthonormal eigenbasis. The natural next question: when can a family share a *single* one? The simultaneous spectral decomposition answers it: a family of normal operators is simultaneously unitarily diagonalizable if and only if they pairwise commute. This is the orthonormal, spectral-theorem-strength refinement of the commuting/joint-diagonalization story from Vol I's eigentheory track.

The intuition is the same as before, sharpened by orthogonality. If B commutes with A, then B maps each spectral eigenspace of A into itself — the eigenspaces are B-invariant. Restrict B to each eigenspace, apply the spectral theorem there to get an orthonormal eigenbasis of B inside it, and because A's eigenspaces are mutually orthogonal the local bases assemble into one global orthonormal basis diagonalizing both.

Commuting normal A, B  ->  shared orthonormal eigenbasis:

  step 1  diagonalize A: spectral resolution A = sum_i lambda_i P_i
  step 2  B commutes with A  =>  B preserves each eigenspace im(P_i)
  step 3  restrict B to each eigenspace, diagonalize there (spectral thm)
  step 4  stitch local orthonormal bases -> single U diagonalizing BOTH:
              U* A U = D_A   AND   U* B U = D_B   (both diagonal)

Reason it can FAIL without commuting:
  A = [1,0;0,2]  (eigvecs e_1,e_2),  B = [0,1;1,0]  (eigvecs (1,1),(1,-1))
  AB = [0,1;2,0] != [0,2;1,0] = BA   ->  no common eigenbasis exists.
Commuting normal operators share an orthonormal eigenbasis; non-commuting ones cannot.

The commutant: who commutes with A

Turning it around, the spectral theorem describes *everything* that commutes with a normal A — the commutant. An operator commutes with A iff it preserves each eigenspace of A; equivalently, it is block-diagonal in A's eigenbasis. When A has n distinct eigenvalues the commutant is exactly the operators that are diagonal in A's basis — and these are precisely the functions of A, polynomials in A.

The payoff: PCA, quantum measurement, and vibrations

Here is why the whole track mattered. The applications of the spectral theorem are everywhere a symmetric or normal operator appears. PCA: a covariance matrix is symmetric positive semidefinite, its top eigenvector (the maximizer of the Rayleigh quotient) is the direction of greatest variance — this is Vol I's principal component analysis given its theoretical foundation. Vibrations: the eigenvalues of a symmetric stiffness operator are the squared natural frequencies, the eigenvectors the normal modes.

Quantum mechanics is where the theorem is most literal: physical observables are self-adjoint operators, their real eigenvalues are the only possible measured values, and the orthogonal spectral projections give the probabilities of each outcome. Two observables can be measured together exactly when they commute — Guide 5's simultaneous decomposition *is* the uncertainty principle's algebraic core.