Documentation

LeanPool.EllipticPDE.Regularity.PointwiseEquation

Pointwise equation of a smooth representative #

A weak solution with a representative that is twice continuously differentiable on an open subset of the domain satisfies the equation there, almost everywhere and in the classical divergence form -∑ᵢⱼ ∂ⱼ(aᵢⱼ ∂ᵢu) + ∑ᵢ bᵢ ∂ᵢu + c u = f, the diffusion being C¹. This is the step the interior theory leaves to the fundamental lemma of the calculus of variations: the weak formulation tested against a function supported in the open set, integrated by parts once more with the classical derivatives of the representative in place of the weak ones, says that the residual of the equation integrates to zero against every test function, so it vanishes almost everywhere.

Two identifications feed the argument. The classical gradient of a C¹ function on an open set is a weak gradient there, by the integration by parts a test function's compact support allows, and the weak gradient on an open set is unique, so the gradient coordinates of the solution agree almost everywhere with the classical partials of the representative on every ball whose closure lies in the set, and a countable subcover of the set by such balls makes that agreement hold on the whole set.

Main declarations #

References #

James Guo, Partial Differential Equations (Course Lecture Notes), Lemma I.2.4 (p. 3); L. C. Evans, Partial Differential Equations (2nd ed.), §6.1.2 (pp. 313–315) and §6.3.1 Theorem 3 (p. 334).

Two measure-theoretic lemmas #

A function continuous on an open set and vanishing off a compact subset of it is integrable on the whole space.

From balls to the open set. A property true almost everywhere on every ball whose closure lies in an open set is true almost everywhere on the set, by a countable subcover.

The classical gradient as a weak gradient #

Classical gradient of a C¹ function on an open set is a weak gradient there. A test function supported in the set has compact support, so the integration by parts has no boundary term, and the function need only be differentiable on that support.

The pointwise equation #

theorem EllipticPdes.Regularity.weakSolution_ae_eq_of_contDiffOn {d : ℕ} (Op : Sobolev.FullEllipticOp d) {Ω : Set (EuclideanSpace ℝ (Fin d))} (hΩm : MeasurableSet Ω) (hA1 : IsC1Coeff Op.toEllipticCoeff) (u : ↥(Sobolev.H01 Ω)) (f : Sobolev.L2D Ω) (hu : ∀ (w : ↥(Sobolev.H01 Ω)), ((Op.fullBilin Ω) u) w = ∫ (x : EuclideanSpace ℝ (Fin d)) in Ω, ↑↑f x * ↑↑((↑w).ofLp 0) x) {W : Set (EuclideanSpace ℝ (Fin d))} (hWo : IsOpen W) (hWΩ : W ⊆ Ω) {u' : EuclideanSpace ℝ (Fin d) → ℝ} (hu' : u' =ᵐ[MeasureTheory.volume.restrict W] fun (x : EuclideanSpace ℝ (Fin d)) => ↑↑((↑u).ofLp 0) x) (hsm : ContDiffOn ℝ 2 u' W) :
∀ᵐ (x : EuclideanSpace ℝ (Fin d)), x ∈ W → -∑ i : Fin d, ∑ j : Fin d, Sobolev.partialD j (fun (y : EuclideanSpace ℝ (Fin d)) => Op.a y i j * Sobolev.partialD i u' y) x + ∑ i : Fin d, Op.b x i * Sobolev.partialD i u' x + Op.c x * u' x = ↑↑f x

Pointwise equation of a smooth representative. A weak solution of the Dirichlet problem whose function coordinate has a representative that is C² on an open subset of the domain satisfies the equation there almost everywhere, in the classical divergence form, the diffusion being C¹. The weak gradient of the solution on the open set is the classical gradient of the representative, the weak formulation tested against a function supported there is integrated by parts once more, and the fundamental lemma of the calculus of variations (Guo Lemma I.2.4) makes the residual vanish almost everywhere.

theorem EllipticPdes.Regularity.exists_weakSolution_interior_classical {n : ℕ} (Op : Sobolev.FullEllipticOp (n + 1)) {Ω : Set (EuclideanSpace ℝ (Fin (n + 1)))} (hΩm : MeasurableSet Ω) (hΩo : IsOpen Ω) (hΩb : Bornology.IsBounded Ω) (hb : ∀ (i : Fin (n + 1)), ∀ᵐ (x : EuclideanSpace ℝ (Fin (n + 1))) ∂MeasureTheory.volume.restrict Ω, Op.b x i = 0) (hc : ∀ᵐ (x : EuclideanSpace ℝ (Fin (n + 1))) ∂MeasureTheory.volume.restrict Ω, 0 ≤ Op.c x) (hA1 : IsC1Coeff Op.toEllipticCoeff) (hA : (k : ℕ) → IsWkInftyCoeff Op.toEllipticCoeff k) (hbc : (k : ℕ) → IsWkInftyLower Op k) (f : Sobolev.L2D Ω) (hf : ∀ (k : ℕ), ∃ (hfk : HasIteratedWeakDerivOn Ω k f) (M : ℝ), IteratedL2Bound hfk M) {V : Set (EuclideanSpace ℝ (Fin (n + 1)))} (hVc : IsCompact V) (hVΩ : V ⊆ Ω) :
∃ (u : ↥(Sobolev.H01 Ω)), (∀ (v : ↥(Sobolev.H01 Ω)), ((Op.fullBilin Ω) u) v = ∫ (x : EuclideanSpace ℝ (Fin (n + 1))) in Ω, ↑↑f x * ↑↑((↑v).ofLp 0) x) ∧ ∃ (u' : EuclideanSpace ℝ (Fin (n + 1)) → ℝ), u' =ᵐ[MeasureTheory.volume.restrict (interior V)] ↑↑((extendL2 hΩm) ((↑u).ofLp 0)) ∧ ContDiffOn ℝ (↑⊤) u' (interior V) ∧ ∀ᵐ (x : EuclideanSpace ℝ (Fin (n + 1))), x ∈ interior V → -∑ i : Fin (n + 1), ∑ j : Fin (n + 1), Sobolev.partialD j (fun (y : EuclideanSpace ℝ (Fin (n + 1))) => Op.a y i j * Sobolev.partialD i u' y) x + ∑ i : Fin (n + 1), Op.b x i * Sobolev.partialD i u' x + Op.c x * u' x = ↑↑f x

Solvability with a smooth interior representative satisfying the equation. On a bounded domain, for an operator with no transport term and a nonnegative zeroth-order coefficient, whose diffusion is C¹ and whose coefficients lie in W^{k,∞} at every order, and for a datum with weak derivatives of every order bounded in L², the Dirichlet problem has a weak solution whose class has a C^∞ representative on the interior of every compact subset of the domain, and that representative satisfies the equation there almost everywhere.