Spectral theorem
The spectral theorem is the fundamental result that decomposes self-adjoint (or normal) operators on a Hilbert space into sums or integrals of orthogonal projections onto eigenspaces. In finite dimensions, it states that every Hermitian matrix can be diagonalized by a unitary transformation — its eigenvectors form an orthonormal basis, and the matrix acts by scaling each eigenvector by its eigenvalue. In infinite dimensions, the theorem generalizes to the functional calculus, which allows functions of operators to be defined rigorously and unifies the disparate-looking theories of diagonalization, Fourier transforms, and quantum observables.
The Finite-Dimensional Case
For an n × n Hermitian matrix A, the spectral theorem asserts the existence of n orthonormal eigenvectors v₁, ..., vₙ with real eigenvalues λ₁, ..., λₙ. The matrix can be written as A = Σ λᵢ Pᵢ, where Pᵢ is the orthogonal projection onto the eigenspace of λᵢ. Equivalently, A = UΛU*, where U is unitary and Λ is diagonal.
This is not merely a computational convenience. The spectral decomposition reveals that a self-adjoint operator is completely determined by its eigenvalues and the geometry of its eigenspaces. The spectral theorem is the reason why quantum mechanical observables have real eigenvalues (measurable quantities) and orthogonal eigenstates (distinguishable outcomes).
The Infinite-Dimensional Spectral Theorem
In infinite dimensions, the situation is richer. A self-adjoint operator on a Hilbert space may have no eigenvalues at all — consider the position operator in quantum mechanics, whose spectrum is purely continuous. The spectral theorem resolves this by replacing the discrete sum with a spectral integral.
For a self-adjoint operator A, there exists a unique spectral measure E (a projection-valued measure on the real line) such that:
A = ∫ λ dE(λ)
This integral converges in the strong operator topology. The spectral measure encodes not just eigenvalues but the entire spectrum: point spectrum (eigenvalues), continuous spectrum, and residual spectrum. The functional calculus allows us to define f(A) = ∫ f(λ) dE(λ) for any bounded Borel function f, turning operator theory into a form of generalized measure theory.
Functional Calculus and Applications
The spectral theorem underlies the rigorous definition of functions of operators. exp(A), log(A), and √A are all defined via the spectral integral, provided the function is defined on the spectrum of A. This is essential for:
- Quantum dynamics: The time evolution operator U(t) = exp(-iHt/ℏ) is defined via the spectral theorem
- Semigroup theory: The heat semigroup exp(-tL) and the Schrödinger semigroup are spectral constructions
- Stochastic processes: The Karhunen-Loève theorem uses the spectral decomposition of covariance operators
The Spectral Theorem as a Bridge
The spectral theorem connects three apparently distinct areas of mathematics: linear algebra (diagonalization), harmonic analysis (Fourier transforms), and measure theory (spectral integrals). In fact, the Fourier transform is itself a spectral decomposition: the operator -d²/dx² on L²(ℝ) is self-adjoint, and its eigenfunctions are the plane waves e^{ikx}. The Fourier transform diagonalizes this operator, and the spectral measure is Lebesgue measure on the frequency axis.
This unity is not accidental. The spectral theorem reveals that the apparent diversity of "diagonalization procedures" in mathematics is actually a single phenomenon viewed through different lenses. What changes is not the theorem but the spectrum: discrete for matrices, continuous for differential operators, and mixed for the operators that appear in quantum field theory.
The spectral theorem is often taught as a technical result about operators, but it is better understood as a statement about the nature of mathematical representation itself. Every self-adjoint operator is a coordinate system in disguise: the spectral measure picks out the "natural" coordinates in which the operator acts by simple scaling. The theorem tells us that the complexity of an operator is never irreducible — it is always a misalignment between the operator's intrinsic geometry and the coordinates we have chosen. This is a profound metaphysical claim dressed as functional analysis: nothing is fundamentally complicated, only poorly represented.