Functional Analysis

Semigroups of Operators

Цели урока

  • State the three axioms of a C0-semigroup and identify the generator
  • Apply the Hille-Yosida criterion and Lumer-Phillips theorem to determine generation
  • Derive solutions using Duhamel's formula for non-homogeneous evolution equations
  • Connect implicit numerical methods to Yosida approximations of semigroups

Предварительные знания

  • Banach and Hilbert spaces
  • Resolvent and spectrum of operators
  • Distributions and fundamental solutions
  • Distributions and Weak Solutions
  • Banach spaces and fixed point theorems

One formula T(t) = e^{tA} unifies the heat equation, Schrodinger equation, transport equation, and diffusion models in ML. What makes this abstraction so powerful?

  • Stable Diffusion and DALL-E: reverse diffusion is a semigroup iteration; contractive semigroup guarantees convergence of the denoising process
  • Quantum computing (IBM Q): unitary evolution U(t) = exp(-iHt) is a C0-group of unitaries; Stone's theorem characterizes its generator
  • Numerical methods: implicit Euler is one Yosida approximation step; unconditional stability comes from the Hille-Yosida resolvent bound
  • Markov chains: transition semigroup P(t) = exp(t*Q) where Q is the rate matrix; stationary distribution = kernel of Q

Hille and Yosida: A Parallel Discovery

Einar Hille and Kosaku Yosida independently published the semigroup generation criterion in 1948. Before this, there was no systematic tool for studying evolution equations in infinite dimensions. The theorem connected differential equations in Banach spaces with operator theory for the first time. Miyadera extended the result in 1952. The nonlinear theory followed: Brezis (1973) for nonlinear dissipative operators, Kato for hyperbolic systems. Today semigroup theory provides the rigorous foundation for numerical PDE software: stability analysis of implicit methods is, at its core, the Hille-Yosida theorem applied to discrete operators.

C0-Semigroups: Axioms and Key Examples

MIT uses semigroup theory to analyze diffusion in neural systems: the Gaussian heat kernel T(t)u_0(x) = (4*pi*t)^{-n/2} * integral exp(-|x-y|^2/4t) * u_0(y) dy models signal propagation through 10^6 neurons in milliseconds. Diffusion models in ML (Stable Diffusion, DALL-E) implement the reverse diffusion process as iterations of a semigroup.

Semigroup T(t)Banach spaceGenerator APDE application
Heat: exp(t*Delta)L^2(R^n)Laplacian DeltaHeat equation
Translation: f(x+t)L^p(R)Differentiation d/dxTransport equation
Schrodinger: exp(-itH)L^2(R^3)Hamiltonian HQuantum evolution
Markov: transition P(t)l^1 or C(X)Rate matrix QMarkov chains

What does the 'C0' in 'C0-semigroup' refer to?

C0-semigroup: ||T(t)u - u||_X -> 0 as t -> 0+ for each fixed u in X. Weaker than uniform continuity ||T(t) - I|| -> 0 in operator norm, but sufficient for most PDE applications.

Hille-Yosida Theorem and Generation Criteria

The central question of semigroup theory: which operators A generate a C0-semigroup? Hille (1948) and Yosida (1948) independently solved this problem. The Hille-Yosida theorem gives a complete characterization through resolvent estimates - and its proof constructs the semigroup explicitly via the Yosida approximation.

Heat semigroup satisfies Hille-Yosida

Laplacian as a generator

The Laplacian A = Delta on H = L^2(R^n) with D(A) = H^2(R^n) is the generator of the heat semigroup. Dissipativity: Re<Delta u, u> = -||nabla u||^2 <= 0 (integration by parts). Hille-Yosida: the resolvent (lambda I - Delta)^{-1} satisfies ||(lambda - Delta)^{-1}|| <= 1/lambda for lambda > 0. Explicit resolvent: R(lambda)f = integral G_lambda(x-y) f(y) dy where G_lambda is the Green's function.

What is the Yosida approximation formula for a semigroup?

Yosida approximation: T(t) = lim_{n->inf} (I - t*A/n)^{-n}. Each factor (I - t*A/n)^{-1} is the resolvent - a bounded operator. The limit converges strongly to the semigroup T(t).

Applications: Evolution Equations and Duhamel's Formula

The semigroup framework unifies the heat equation, the Schrodinger equation, and the transport equation into a single abstract Cauchy problem du/dt = Au, u(0) = u_0. For non-homogeneous equations du/dt = Au + f(t), Duhamel's formula gives the solution as a superposition of forced evolutions.

What does Duhamel's formula say for the equation du/dt = Au + f(t)?

Duhamel's formula: u(t) = T(t)u_0 + integral_0^t T(t-s)f(s) ds. Free evolution + forced response. Derivation: differentiate both sides and verify du/dt = Au + f(t).

Numerical Methods via Semigroup Theory

Stability analysis of numerical schemes for PDE is equivalent to studying discrete approximations of semigroups. The Lax equivalence theorem: consistency + stability = convergence. Stability means the discrete scheme defines a uniformly bounded family of operators - a discrete analog of the semigroup growth bound ||T(t)|| <= M*exp(omega*t).

The Crank-Nicolson scheme (theta-method with theta=1/2) is the average of implicit and explicit Euler: u^{n+1} = (I - dt/2*A)^{-1}(I + dt/2*A)u^n. The factor (I - dt/2*A)^{-1}(I + dt/2*A) is a Cayley transform. For self-adjoint A, this is exactly unitary - it conserves the H^1 norm. This is why Crank-Nicolson is the standard for Schrodinger-type equations.

Why is the implicit Euler method unconditionally stable for dissipative operators?

Implicit Euler step: (I - dt*A)^{-1}. For dissipative A (Re<Au,u> <= 0): ||(I - dt*A)^{-1}|| <= 1 by Lumer-Phillips. N steps: ||(I-dt*A)^{-N}|| <= 1 for all N. Unconditional stability.

Connections to other topics

Semigroup theory is the unified language for evolution equations in mathematics and physics.

  • Spectral theory of self-adjoint operators — Generators of C0-semigroups are described by the spectral theorem and the resolvent
  • Distributions and Weak Solutions — Semigroup evolution defines weak solutions of parabolic PDEs
  • Functional Analysis in PDEs — Operator semigroups realise the heat and wave equations abstractly

Итоги

  • C0-semigroup: T(0)=I, T(t+s)=T(t)T(s), strong continuity; growth bound ||T(t)|| <= M*exp(omega*t)
  • Generator A: A*u = lim (T(t)u - u)/t; orbits solve du/dt = Au (abstract Cauchy problem)
  • Hille-Yosida: A generates a C0-semigroup iff densely defined, closed, and ||(lambda-A)^{-n}|| <= M/(lambda-omega)^n
  • Lumer-Phillips: A dissipative (Re<Au,u> <= 0) and (I-A) dense range implies ||T(t)|| <= 1
  • Duhamel's formula: u(t) = T(t)u_0 + integral_0^t T(t-s)f(s) ds for du/dt = Au + f(t)

Вопросы для размышления

  • Why does the semigroup property T(t+s) = T(t)T(s) encode deterministic time-reversible (or irreversible) dynamics?
  • The Yosida approximation T(t) = lim(I - t*A/n)^{-n} parallels e^x = lim(1+x/n)^n. What breaks down if A is unbounded and one uses the power series directly?
  • How does the implicit Euler method derive its unconditional stability from the Hille-Yosida resolvent estimate?

Связанные уроки

  • fa-22 — Heat kernel is the fundamental solution defining the heat semigroup
  • fa-24 — Spectral theorem gives explicit formula T(t) = exp(t*A) via spectral measure
  • fa-20-fixed-point — Banach spaces are the arena for C0-semigroups
Semigroups of Operators

0

1

Sign In