Documentation

LeanPool.EllipticPDE.Regularity.CoeffWkInfty

W^{k,∞} coefficients #

Guo, Partial Differential Equations (Course Lecture Notes), Theorem VIII.3.2 (Higher Interior Regularity, p. 65) runs the induction over a_{ij} ∈ W^{k+1,∞}(Ω) and b_i, c ∈ W^{k,∞}(Ω), where Evans, Partial Differential Equations (2nd ed.), §6.3.1, Theorem 2 asks for C^{m+1}. The development here asks W^{k+1,∞} for a_{ij} and W^{k,∞} for b_i, c (higher_interior_regularity). This file states the hypothesis: weak derivatives up to order k, each essentially bounded, with no continuity assumed anywhere.

Choice of a new weak-derivative predicate #

HasWeakDerivOn is typed on Lp ℝ 2 (volume.restrict V) classes, which suits a solution on a bounded domain. A coefficient is a bounded measurable function on all of EuclideanSpace ℝ (Fin d), and on an unbounded domain such a function need not be L², so it has no Lp 2 class to represent it. HasWeakPartial below is the same integration-by-parts identity stated for plain functions, which is where a locally integrable coefficient lives.

Indexing of the derivative family #

The iterated derivative is indexed by a List (Fin d) of directions rather than by a multi-index in Fin d →₀ ℕ. Guo's proof expands D^α through the Leibniz rule over β ≤ α, and both indexings support that; a list is chosen because one step of the recursion is cons, matching D_step, whereas a finitely-supported function would need the order of differentiation to be quotiented out before the recursion could be stated. Equality of mixed partials is a theorem about the family rather than part of its definition, so nothing here presumes it.

The family is data, not an existential: D α i j names the chosen representative of the order-α.length derivative of the (i,j) entry. D_nil pins the empty list to the coefficient itself and D_step makes each successive entry a weak derivative of its parent.

Main declarations #

Statements this file does not yet supply #

The bridge IsCkCoeff A k → IsWkInftyCoeff A k and the difference-quotient bound under the weaker hypothesis. The latter cannot be had in the everywhere-pointwise form that IsC1Coeff.abs_diffQuot_coeff_le currently has: that proof is the classical mean value inequality, which needs a derivative at every point, and recovering a pointwise bound from an essentially bounded weak derivative is the statement that W^{1,∞} functions have Lipschitz representatives. Mathlib has Rademacher's theorem in the opposite direction (LipschitzWith.ae_differentiableAt) and not this one. The route that avoids it is mollification: (∇a) * ρ_ε inherits the essential bound of ∇a, the classical inequality applies to the smooth a * ρ_ε, and the bound passes to the limit almost everywhere. That yields the bound a.e. rather than everywhere, which is all its consumers integrate against.

f' is the weak k-th partial derivative of f: integration by parts against every smooth compactly supported test function holds with no boundary term. Stated for plain functions, so that a bounded measurable coefficient on EuclideanSpace ℝ (Fin d) is in scope where the Lp 2-typed HasWeakDerivOn is not.

Equations
  • One or more equations did not get rendered due to their size.
Instances For

    A W^{k,∞} ellipticity bundle, in the sense of Guo, Partial Differential Equations (Course Lecture Notes), Theorem VIII.3.2 (p. 65): every coefficient entry has weak derivatives up to order k, each measurable and essentially bounded, with no continuity assumed. The derivative family is indexed by a list of directions, one cons per differentiation.

    • D : List (Fin d) → Fin d → Fin d → EuclideanSpace ℝ (Fin d) → ℝ

      The chosen representative of the iterated weak derivative of the (i,j) entry along a list of directions.

    • D_nil (i j : Fin d) : self.D [] i j = fun (x : EuclideanSpace ℝ (Fin d)) => A.a x i j

      The empty list of directions is the coefficient entry itself.

    • D_meas (i j : Fin d) (α : List (Fin d)) : α.length ≤ k → Measurable (self.D α i j)

      Every entry of the family is measurable.

    • D_step (i j l : Fin d) (α : List (Fin d)) : α.length < k → HasWeakPartial l (self.D α i j) (self.D (l :: α) i j)

      Each successive entry is a weak partial derivative of its parent, up to order k.

    • bound : ℕ → ℝ

      The uniform bound on the derivatives of each order. Only the values at m ≤ k are constrained by ess_bdd.

    • bound_nonneg (m : ℕ) : 0 ≤ self.bound m

      Every bound is nonnegative.

    • ess_bdd (i j : Fin d) (α : List (Fin d)) : α.length ≤ k → ∀ᵐ (x : EuclideanSpace ℝ (Fin d)), |self.D α i j x| ≤ self.bound α.length

      Each derivative of order m ≤ k is bounded by bound m almost everywhere. The bound is essential rather than pointwise, which is what W^{k,∞} asserts.

    Instances For

      An order-k bundle is an order-l bundle for every l ≤ k: the family and its bounds are inherited, and each hypothesis is asked of fewer orders.

      Equations
      • hA.mono hlk = { D := hA.D, D_nil := ⋯, D_meas := ⋯, D_step := ⋯, bound := hA.bound, bound_nonneg := ⋯, ess_bdd := ⋯ }
      Instances For

        The order-zero bound applies to the coefficient entries themselves.