Functional Analysis
Spectral theory of self-adjoint operators
Цели урока
- State the spectral theorem A = integral lambda dE(lambda) and define projection-valued measures
- Distinguish point, continuous, and residual spectrum; prove residual spectrum vanishes for self-adjoint operators
- Apply Stone's theorem connecting self-adjoint operators and unitary groups
- Use the Courant-Fischer min-max principle for variational eigenvalue estimates
Предварительные знания
- Hilbert spaces and orthogonality
- Semigroups of operators
- Resolvent and spectrum basics
Quantum measurements always yield real numbers. Wave functions are complex. How does the spectral theorem for self-adjoint operators reconcile this - and why is it the backbone of quantum mechanics?
- IBM Q 433-qubit processor: every quantum observable is a self-adjoint operator; measurement outcomes are eigenvalues; spectral theorem guarantees they are real
- MRI (magnetic resonance imaging): spectral decomposition of Laplacian on brain domain gives spatial modes; eigenvalues determine resonance frequencies
- Google PageRank: power method finds principal eigenvector of stochastic transition matrix - converges by spectral gap estimate from min-max principle
- FEM structural analysis (Abaqus, Ansys): K*u = omega^2*M*u; natural frequencies via Ritz method; min-max guarantees upper bounds
Von Neumann and the Mathematical Foundation of Quantum Mechanics
Heisenberg formulated quantum mechanics as matrix mechanics in 1925. Born and Jordan showed the position-momentum commutation relation [p,q] = -i*hbar*I. Von Neumann built the rigorous foundation in 1928-1932: operators on Hilbert spaces, the spectral theorem for unbounded self-adjoint operators, Stone's theorem (1932). The Dirac formalism (bra-ket notation) was made rigorous via the spectral theorem and rigged Hilbert spaces. Weyl's theorem on the stability of essential spectrum and Kato's perturbation theory completed the picture. Today, every quantum computing textbook is built on these foundations from the 1930s.
Spectral Theorem for Self-Adjoint Operators
In 2023 IBM Q ran quantum circuits on 433 qubits. Every quantum gate is unitary; every observable is self-adjoint. The spectral theorem, proved by von Neumann in 1929, guarantees: any self-adjoint operator A has a spectral decomposition A = integral lambda dE(lambda) where E is a projection-valued measure. Measurement outcomes are eigenvalues - always real, because A = A*.
Multiplication operator: the model of continuous spectrum
Continuous spectrum without eigenvalues
On L^2([0,1]), the multiplication operator (Mf)(x) = x*f(x) is self-adjoint. Point spectrum: empty (Mf = lambda*f requires (x-lambda)f = 0 a.e., so f = 0). Continuous spectrum: sigma_c(M) = [0,1]. The spectral measure: E(B) = multiplication by the indicator of B. This is the canonical example of a self-adjoint operator with purely continuous spectrum - no eigenvalues at all.
Why are all eigenvalues of a self-adjoint operator A = A* real?
If Av = lambda*v: lambda*||v||^2 = <Av,v> = <v,A*v> = <v,Av> = conj(lambda*||v||^2) = conj(lambda)*||v||^2. So lambda = conj(lambda): lambda is real.
Resolvent and Spectrum Decomposition
The resolvent R(lambda) = (A - lambda I)^{-1} is an analytic function of lambda outside the spectrum. Poles correspond to eigenvalues; branch cuts to the continuous spectrum. This analytic structure is the key connection between operators and complex analysis - and the foundation for contour integral methods in spectral theory.
| Operator | Point spectrum | Continuous spectrum | Setting |
|---|---|---|---|
| -d^2/dx^2 on [0,pi] (Dirichlet) | n^2, n=1,2,... | Empty | L^2([0,pi]) |
| -Delta on L^2(R^n) | Empty | [0, +inf) | L^2(R^n) |
| -d^2/dx^2 + x^2 on L^2(R) | 2n+1, n>=0 | Empty | L^2(R) |
| Multiplication by x on L^2([0,1]) | Empty | [0,1] | L^2([0,1]) |
Why is the residual spectrum of a self-adjoint operator empty?
For A = A*: if lambda is not an eigenvalue, ker(A - lambda I) = 0. By self-adjointness, ker(A* - conj(lambda)I) = ker(A - lambda I) = 0, which means im(A - lambda I) is dense. So lambda is not in the residual spectrum.
Stone's Theorem and Min-Max Principle
Stone's theorem (1932) establishes a one-to-one correspondence between self-adjoint operators and unitary groups: H is self-adjoint if and only if U(t) = e^{-iHt} is a C0-group of unitary operators. The min-max (Courant-Fischer) principle characterizes eigenvalues variationally - the foundation of the Ritz method in structural mechanics.
FEM eigenvalue problem via min-max
Structural mechanics: natural frequencies
In structural analysis (Abaqus, Ansys), natural frequencies omega_n of a structure satisfy K*u = omega^2*M*u where K is the stiffness matrix and M is the mass matrix. Both are self-adjoint. Min-max: omega_1^2 = min (u^T K u)/(u^T M u). The Ritz method approximates this: minimize over a finite-dimensional subspace spanned by shape functions. Convergence is guaranteed by the min-max principle - the Ritz values are upper bounds for the true eigenvalues.
What does Stone's theorem establish about self-adjoint operators?
Stone's theorem: H = H* (self-adjoint) if and only if {e^{-iHt}} is a C0-group of unitary operators. The generator is iH. Proof uses the spectral theorem: e^{-iHt} = integral e^{-i*lambda*t} dE(lambda).
Compact Self-Adjoint Operators and Hilbert-Schmidt Theorem
Compact self-adjoint operators are the most tractable: they have a complete orthonormal basis of eigenvectors, and their eigenvalues accumulate only at 0. This is the precise analog of eigendecomposition for symmetric matrices, extended to infinite dimensions. Google PageRank computes the principal eigenvector of a stochastic matrix; MRI reconstructs images from eigenfunctions of the Laplacian.
Kernel methods in machine learning: the kernel matrix K_{ij} = k(x_i, x_j) is a finite approximation of a compact self-adjoint integral operator. Mercer's theorem: if k(x,y) is continuous and positive definite, the operator with kernel k is compact self-adjoint with non-negative eigenvalues. The eigenfunctions {e_k} form the RKHS basis, and eigenvalues decay controls the smoothness of functions in the RKHS.
What does the Hilbert-Schmidt theorem guarantee for compact self-adjoint operators?
Hilbert-Schmidt theorem: compact K = K* => H = direct sum of eigenspaces. Eigenvalues: countable, accumulate only at 0. Eigenvectors: complete orthonormal basis. Generalizes symmetric matrix diagonalization.
Connections to other topics
Spectral theory of self-adjoint operators is central to mathematical physics and numerical methods.
- Semigroups of Operators — Spectrum of a self-adjoint generator dictates stability and long-time behaviour
- C*-algebras and von Neumann algebras — The spectral theorem extends via GNS to algebras of observables
- Operator Algebras and Quantum Mechanics — Self-adjoint operators model observables in quantum theory
Итоги
- Self-adjoint A = A*: all eigenvalues real; eigenvectors orthogonal; no residual spectrum
- Spectral theorem: A = integral lambda dE(lambda); Borel functional calculus f(A) = integral f(lambda) dE(lambda)
- Stone's theorem: H self-adjoint iff e^{-iHt} is C0-group of unitaries
- Compact self-adjoint: Hilbert-Schmidt theorem gives complete ONB of eigenvectors; eigenvalues accumulate at 0
- Min-max (Courant-Fischer): lambda_n = min_{V, codim n-1} max_{u perp V} <Au,u>/||u||^2
Вопросы для размышления
- Why is the distinction between 'symmetric' and 'self-adjoint' crucial for unbounded operators, and why does it matter for quantum mechanics?
- The continuous spectrum contains no eigenvectors, yet the spectral theorem still decomposes A via a PVM. How does this work conceptually?
- How does the min-max principle give upper bounds for eigenvalues in the Ritz method, and why are they upper bounds and not lower bounds?