Functional Analysis
Distributions and Weak Solutions
Цели урока
- Define distributions as continuous linear functionals on test function spaces
- Compute distributional derivatives using <T', phi> = -<T, phi'>
- Identify fundamental solutions of classical operators and apply the convolution principle
- Understand tempered distributions and the Fourier transform rules
Предварительные знания
- Sobolev spaces and weak derivatives
- L^p spaces and duality
- Basic Fourier analysis
Why did physicists use the Dirac delta for 20 years before mathematicians provided a rigorous definition - and what does this say about the relationship between physical intuition and mathematical rigor?
- Nvidia H100 GPU signal processing: convolution with delta(t - t_0) models instantaneous impulse response; distributional convolution is the mathematical foundation
- Boundary element method (Ansys): fundamental solution -1/(4*pi*|x|) for Laplacian reduces 3D PDE to 2D boundary integral - O(N^3) becomes O(N^2)
- FNet transformer (Google 2021): FFT replaces attention; 7x faster than BERT; theoretical basis is the convolution theorem for tempered distributions
- Dirac delta in quantum mechanics: density rho(x) = q * delta^3(x - x_0) for a point charge; fundamental solution of Laplacian gives the Coulomb potential
From Dirac's Delta to Schwartz's Theory
Paul Dirac introduced the delta function in 1927 for quantum mechanics. Mathematicians rejected it as non-rigorous: no ordinary function satisfies its properties. Laurent Schwartz, working in Paris 1945-1950, built distribution theory, giving precise meaning to Dirac's objects. His book Theorie des Distributions (1950) earned him the Fields Medal. In parallel, Gelfand and Shilov in the USSR developed the theory of generalized functions. Today distributions are fundamental to PDE analysis, signal processing, and quantum field theory.
Schwartz Distributions and Test Functions
In 1950, Laurent Schwartz received the Fields Medal for distribution theory. The Dirac delta, used by physicists without rigorous justification since 1927, finally had a precise mathematical meaning. Today delta(t - t_0) appears in signal processing on Nvidia H100 GPUs: convolution with delta models an instantaneous impulse on timescales of 10^{-15} seconds.
Heaviside function and its derivative
A discontinuous function whose distributional derivative is delta
The Heaviside step function H(x) = 1 for x > 0, H(x) = 0 for x < 0, is discontinuous at 0. Its classical derivative does not exist. But its distributional derivative: <H', phi> = -<H, phi'> = -integral_0^inf phi'(x) dx = phi(0) = <delta, phi>. So H' = delta in the distributional sense. This is used in circuit theory: H(t) models a switch turning on at t=0; H'(t) = delta(t) models the instantaneous current spike.
Every distribution is infinitely differentiable - this is the key advantage over classical analysis. A single jump discontinuity (like H) has a distributional derivative (delta), which in turn has a derivative (delta'), and so on. No loss of regularity in the distributional sense.
How is the derivative of a distribution T defined in Schwartz theory?
Distributional derivative: <T', phi> = -<T, phi'>. Consistent with integration by parts for smooth functions. Every distribution is infinitely differentiable in this sense.
Fundamental Solutions and the Convolution Principle
A fundamental solution of a differential operator L is a distribution E such that L(E) = delta. Knowing E converts the equation Lu = f into a convolution: u = E * f. This is the universal construction behind boundary element methods, Green's functions in physics, and the Fourier method for constant-coefficient PDE.
| Operator L | Fundamental solution E | Space | Application |
|---|---|---|---|
| Laplacian -Delta | -1/(4*pi*|x|) | R^3 | Electrostatics, boundary element method |
| Heat operator d/dt - Delta | (4*pi*t)^{-n/2} exp(-|x|^2/4t) | R^n x R+ | Diffusion, semigroup kernel |
| Wave operator d^2/dt^2 - Delta | delta(|x|-t)/(4*pi*t) | R^3 x R | Wave propagation, Huygens principle |
| Cauchy-Riemann d/dz-bar | 1/(pi*z) | R^2 = C | Complex analysis, Cauchy integral formula |
What is the fundamental solution of a differential operator L?
Fundamental solution E: L(E) = delta. Convolution principle: u = E * f solves Lu = f. This reduces constant-coefficient PDE to a single convolution.
Tempered Distributions and Fourier Analysis
The Schwartz space S(R^n) consists of rapidly decaying smooth functions; its dual S'(R^n) is the space of tempered distributions. The Fourier transform extends to S' and maps differentiation to multiplication. This is the mathematical foundation of signal processing: every standard operation - filtering, convolution, modulation - becomes algebraic in the frequency domain.
The FNet transformer (Google, 2021) replaces the attention mechanism with a fast Fourier transform on token sequences. Each layer applies FFT - implementing convolution in the distributional sense. Speed: 7x faster than BERT at 92% quality on GLUE. The theoretical justification: FFT implements the convolution theorem F[f * g] = F[f] * F[g] for tempered distributions.
What is the Fourier transform of the Dirac delta function delta(x)?
F[delta](xi) = integral delta(x) * exp(-i*xi*x) dx = exp(-i*xi*0) = 1. Constant 1 in frequency domain - a perfect impulse contains all frequencies with equal amplitude.
Weak Solutions of PDE
A classical solution of -Delta u = f requires u in C^2. A weak solution requires only u in H^1. This extension is not a mathematical convenience - it is a necessity: the heat equation with discontinuous coefficients (two media with different conductivities) has no classical solution but always has a unique weak solution.
Schrodinger equation: weak solutions in physics
Quantum mechanics requires distributional thinking
The time-independent Schrodinger equation -hbar^2/(2m) * Delta psi + V(x)*psi = E*psi for a potential V that is a sum of delta functions (Dirac combs, crystal lattices) has no classical C^2 solution. Weak solutions in H^1 exist, and the matching conditions at each delta (continuity of psi, jump in psi') follow automatically from the weak formulation. This is the starting point for band theory in solid state physics.
What is the principal advantage of weak solutions over classical ones for PDE with discontinuous coefficients?
Weak formulation: derivatives shift onto test functions. Coefficients only need to be in L^inf; the solution only needs to be in H^1. Lax-Milgram guarantees a unique weak solution for elliptic operators.
Connections to other topics
Distribution theory is the common language of modern PDE, signal processing, and quantum physics.
- Sobolev Spaces — Distributional derivatives provide the weak formulation that defines Sobolev classes
- Distribution Theory — Schwartz's theory underlies the weak-solution machinery for PDEs
- Spectral theory of self-adjoint operators — Resolvents of differential operators are studied as tempered distributions
Итоги
- Distribution: continuous linear functional on D(Omega); Dirac delta <delta_{x_0}, phi> = phi(x_0)
- Derivative: <T', phi> = -<T, phi'>; every distribution is infinitely differentiable
- Fundamental solution E: L(E) = delta; solution of Lu = f via convolution u = E * f
- Tempered distributions S'(R^n): F[delta] = 1, F[T'] = i*xi * F[T], F[T*S] = F[T]*F[S]
- Weak solutions in H^1: shift derivatives onto test functions; valid for discontinuous coefficients
Вопросы для размышления
- Why is every distribution infinitely differentiable while ordinary L^p functions need not even be continuous?
- The Fourier transform of delta is 1 (all frequencies present equally). What is the physical meaning of this in signal processing?
- How does the convolution principle u = E * f reduce a PDE to an algebraic problem in the frequency domain?