Differential Equations

Variational Methods for PDEs

Цели урока

Understand weak derivatives and Sobolev spaces H^k
Apply the Lax-Milgram theorem to prove existence of weak solutions
Connect elliptic regularity to smoothness of data and domain
Extend to nonlinear problems via monotone operators

Предварительные знания

Functional analysis (Hilbert spaces)
Integration by parts
Second-order PDEs

Optimal Control

How can one define a meaningful solution to a PDE with discontinuous coefficients - for example, at the interface between two different materials?

FEM in Ansys uses weak solutions for heterogeneous material problems (steel-concrete interfaces)
Image processing: variational functionals of Rudin-Osher-Fatemi type minimize H^1 energy for denoising
Machine learning: Barron's 1993 approximation theory connects neural networks to H^1 function classes
Quantum chemistry: the Schrodinger equation is solved in H^1 for molecules of any complexity

Sobolev, Lax, and Milgram: Functional Analysis for PDEs

Sergei Sobolev introduced generalized derivatives in 1938 while working on wave theory. The mathematical community was initially skeptical about extending the notion of a function. In 1954 Peter Lax and Arthur Milgram proved the existence theorem for a broad class of problems - and FEM acquired a rigorous foundation. Claude Barre in 1950 and subsequent authors developed trace theory, allowing boundary conditions to be imposed correctly for H^1 functions.

Sobolev Spaces and Weak Derivatives

Classical derivatives require smoothness - discontinuous functions do not have them. Sergei Sobolev introduced the weak derivative in 1938: a function u has weak derivative Du = v if for all smooth compactly supported phi the relation int(u * D*phi) = -int(v * phi) holds. This allows differentiating step functions and modeling discontinuous media.

The weak derivative is not merely a technical device. In problems with discontinuous coefficients (material interfaces) the classical solution does not exist, while the weak solution exists and is unique. This is precisely what makes FEM applicable to real engineering problems.

What does the Sobolev embedding H^k -> C^m guarantee?

Lax-Milgram Theorem and Existence of Solutions

The Lax-Milgram theorem (1954) is the primary tool for proving existence and uniqueness of weak solutions. It handles non-symmetric operators - unlike the Riesz representation theorem, which requires symmetry. This is what makes it applicable to convection-diffusion equations and Navier-Stokes.

Elliptic Equation with Variable Coefficients

Lax-Milgram applied to -div(A(x) nabla u) = f

Problem: -div(A(x) nabla u) + c(x)u = f in Omega, u = 0 on boundary. Bilinear form: a(u,v) = int(A(x) nabla u cdot nabla v + c(x)uv). When A(x) >= lambda > 0 and c(x) >= 0, coercivity holds: a(u,u) >= lambda*||nabla u||^2 >= (lambda/C_P^2)*||u||_{H^1}^2. The Lax-Milgram theorem guarantees a unique solution in H^1_0 without any smoothness requirement on A(x).

Why is the Lax-Milgram theorem needed when the Riesz representation theorem already exists?

Elliptic Regularity

A weak solution exists in H^1. But how smooth is it? Regularity theory answers: the smoothness of the data (f, boundary) determines the smoothness of the solution. For smooth right-hand side and smooth boundary the weak solution is classical. This elliptic regularity property is the key distinction between elliptic and hyperbolic PDEs.

Regularity breaks at corner points of the boundary (L-shaped domain): the solution can develop a singularity of the form r^{pi/omega}, where omega is the interior angle. This is why adaptive FEM refines near re-entrant corners.

Galerkin Method and Projection

The general FEM framework as a Galerkin method

The Galerkin method: find u_h in a finite-dimensional subspace V_h subset V such that a(u_h, v_h) = F(v_h) for all v_h in V_h. Cea's lemma gives: ||u - u_h||_V <= (M/alpha) * min_{w in V_h} ||u - w||_V. FEM produces a quasi-optimal approximation with constant M/alpha. Choosing the subspace V_h is the art of element design.

Why does the Poisson solution develop a singularity at a re-entrant corner of an L-shaped domain?

Weak Solutions for Nonlinear PDEs

Nonlinear PDEs are harder: Lax-Milgram does not apply directly. For monotone operators the Browder-Minty theorem provides existence. For nonlinear elasticity and viscosity, compactness arguments (Rellich theorem) are used. The 3D Navier-Stokes equation remains an open Millennium Prize problem.

3D Navier-Stokes: the existence of smooth solutions for arbitrary initial data is unproven. This is one of the seven Millennium Prize Problems of the Clay Institute (1 million dollars reward). Weak solutions (Leray, 1934) exist, but their uniqueness and smoothness in 3D remain open.

Why cannot Lax-Milgram be applied directly to the nonlinear p-Laplacian?

Connections to Other Areas

Variational methods and Sobolev spaces are the foundation of modern PDE theory and computational mathematics.

Finite Element Method — Related topic
Image Processing — Related topic
Neural Networks and Approximation — Related topic
Navier-Stokes Equation — Related topic

Итоги

Weak derivatives generalize classical ones to discontinuous functions; H^1_0 is the right class for the Poisson problem
Lax-Milgram: boundedness plus coercivity of the bilinear form guarantees a unique weak solution
Elliptic regularity: f in L^2 and smooth boundary give u in H^2 - the weak solution is classical
For nonlinear operators the Browder-Minty theorem applies; 3D Navier-Stokes regularity remains open

Вопросы для размышления

Why does a discontinuous coefficient A(x) not prevent existence of the weak solution but breaks the classical one?
How does the Sobolev embedding theorem explain why all H^1 functions in 1D are continuous but not in 2D?
What do the Galerkin method in FEM and stochastic gradient descent in machine learning have in common?

Связанные уроки

de-25-fem — FEM is a direct application of the Lax-Milgram theorem
de-26-optimal-control — Functional analysis is needed for variational optimal control
de-28-wave-equation — Hyperbolic PDEs also have weak formulations in Sobolev spaces