Documentation

TauCeti.Analysis.Sobolev.Wkp.LineDeriv

Directional derivatives of higher-order Sobolev functions #

A function u ∈ Lᵖ(Ω) lies in W^{k+1,p}(Ω) exactly when, in every direction v, it has a weak derivative ∂_v u lying in W^{k,p}(Ω); it suffices to check the directions of one orthonormal basis. This is the description of the Sobolev scale by first derivatives, W^{k+1,p}(Ω) = {u ∈ Lᵖ(Ω) : ∂ᵢu ∈ W^{k,p}(Ω) for every i}, and it is what lets a statement about W^{k,p} be proved by induction on k one derivative at a time.

TauCeti.Wkp records the derivatives of u from the bottom up: the gradient first, and then the weak Fréchet derivative of the previous field. The fields of ∂_v u are therefore those of u with v substituted in the innermost slot, the one belonging to the gradient. At the first level this substitution is A ↦ A† v, since the gradient is identified with a vector through the inner product. Conversely, the fields of u are reassembled from those of the ∂ᵢu by inserting ⟪eᵢ, ·⟫ in the innermost slot. Neither direction uses the symmetry of weak second derivatives, which is proved separately: since weak derivatives commute, the weak Hessian is symmetric.

No boundedness or boundary regularity of Ω is assumed.

Main declarations #

References #

Substituting and inserting a direction in the innermost slot #

From W^{k+1,p} to the directional derivatives #

theorem TauCeti.Wkp.exists_hasWeakLineDerivOn_value {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] {mu : MeasureTheory.Measure E} [mu.IsAddHaarMeasure] {Omega : TopologicalSpace.Opens E} {p : ENNReal} [Fact (1 ≤ p)] (k : ℕ) (w : Wkp mu Omega p (k + 1)) (v : E) :
∃ (d : Wkp mu Omega p k), HasWeakLineDerivOn mu Omega (↑↑(value (k + 1) w)) (↑↑(value k d)) v

Directional derivatives of Sobolev functions are Sobolev. In every direction v, an element of W^{k+1,p}(Ω) has a weak derivative which is the value of an element of W^{k,p}(Ω).

theorem TauCeti.Wkp.exists_hasWeakLineDerivOn_value_and_inner_gradient_ae_eq {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] {mu : MeasureTheory.Measure E} [mu.IsAddHaarMeasure] {Omega : TopologicalSpace.Opens E} {p : ENNReal} [Fact (1 ≤ p)] (u : Wkp mu Omega p 2) (v : E) :
∃ (d : ↥(W1p mu Omega p)), HasWeakLineDerivOn mu Omega (↑↑(value 2 u)) (↑↑(W1p.value d)) v ∧ ∀ (w : E), (fun (x : E) => inner ℝ (↑↑(W1p.gradient d) x) w) =ᵐ[mu.restrict ↑Omega] fun (x : E) => inner ℝ ((↑↑(iteratedGradient 1 u) x) w) v

First derivatives of a W^{2,p} function. In every direction v, an element u of W^{2,p}(Ω) has a weak derivative ∂ᵥu which is the value of an element of W^{1,p}(Ω), whose weak derivative in a direction w is the entry ⟪D²u w, v⟫ = ∂_w ∂ᵥu of the weak Hessian.

From the directional derivatives to W^{k+1,p} #

theorem TauCeti.Wkp.exists_value_eq_of_forall_hasWeakLineDerivOn {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] {mu : MeasureTheory.Measure E} [mu.IsAddHaarMeasure] {Omega : TopologicalSpace.Opens E} {p : ENNReal} [Fact (1 ≤ p)] {ι : Type u_2} [Fintype ι] (b : OrthonormalBasis ι ℝ E) (u : ↥(MeasureTheory.Lp ℝ p (mu.restrict ↑Omega))) (k : ℕ) :
(∀ (i : ι), ∃ (d : Wkp mu Omega p k), HasWeakLineDerivOn mu Omega (↑↑u) (↑↑(value k d)) (b i)) → ∃ (w : Wkp mu Omega p (k + 1)), value (k + 1) w = u

Sobolev functions from their directional derivatives. If, along every vector of an orthonormal basis, u ∈ Lᵖ(Ω) has a weak derivative which is the value of an element of W^{k,p}(Ω), then u is the value of an element of W^{k+1,p}(Ω).

theorem TauCeti.Wkp.exists_value_eq_iff_forall_exists_hasWeakLineDerivOn {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] {mu : MeasureTheory.Measure E} [mu.IsAddHaarMeasure] {Omega : TopologicalSpace.Opens E} {p : ENNReal} [Fact (1 ≤ p)] (k : ℕ) (u : ↥(MeasureTheory.Lp ℝ p (mu.restrict ↑Omega))) :
(∃ (w : Wkp mu Omega p (k + 1)), value (k + 1) w = u) ↔ ∀ (v : E), ∃ (d : Wkp mu Omega p k), HasWeakLineDerivOn mu Omega (↑↑u) (↑↑(value k d)) v

The Sobolev scale by first derivatives. A function u ∈ Lᵖ(Ω) is the value of an element of W^{k+1,p}(Ω) exactly when, in every direction, it has a weak derivative which is the value of an element of W^{k,p}(Ω).

theorem TauCeti.Wkp.exists_value_ae_eq_of_forall_hasWeakLineDerivOn {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] {mu : MeasureTheory.Measure E} [mu.IsAddHaarMeasure] {Omega : TopologicalSpace.Opens E} {p : ENNReal} [Fact (1 ≤ p)] (k : ℕ) {w : E → ℝ} (hw : MeasureTheory.MemLp w p (mu.restrict ↑Omega)) (d : E → Wkp mu Omega p k) (hd : ∀ (v : E), HasWeakLineDerivOn mu Omega w (↑↑(value k (d v))) v) :
∃ (W : Wkp mu Omega p (k + 1)), ↑↑(value (k + 1) W) =ᵐ[mu.restrict ↑Omega] w

A function w in Lᵖ(Ω) whose weak derivative on Ω in every direction v is the value of an element d v of W^{k,p}(Ω) agrees almost everywhere on Ω with the value of an element of W^{k+1,p}(Ω). This is TauCeti.Wkp.exists_value_eq_iff_forall_exists_hasWeakLineDerivOn for a function representative rather than an element of Lᵖ.

theorem TauCeti.Wkp.exists_value_one_ae_eq_of_forall_hasWeakLineDerivOn {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] {mu : MeasureTheory.Measure E} [mu.IsAddHaarMeasure] {Omega : TopologicalSpace.Opens E} {p : ENNReal} [Fact (1 ≤ p)] {w : E → ℝ} (hw : MeasureTheory.MemLp w p (mu.restrict ↑Omega)) (w' : E → E → ℝ) (hw' : ∀ (v : E), MeasureTheory.MemLp (w' v) p (mu.restrict ↑Omega)) (h : ∀ (v : E), HasWeakLineDerivOn mu Omega w (w' v) v) :
∃ (W : Wkp mu Omega p 1), ↑↑(value 1 W) =ᵐ[mu.restrict ↑Omega] w

The first-order case of TauCeti.Wkp.exists_value_ae_eq_of_forall_hasWeakLineDerivOn: a function w in Lᵖ(Ω) whose weak derivative on Ω in every direction v is a function w' v in Lᵖ(Ω) agrees almost everywhere on Ω with the value of an element of W^{1,p}(Ω).

Second order from the components of the gradient #

theorem TauCeti.W1p.exists_lowerOrder_eq_of_forall_hasWeakLineDerivOn {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] {mu : MeasureTheory.Measure E} [mu.IsAddHaarMeasure] {Omega : TopologicalSpace.Opens E} {p : ENNReal} [Fact (1 ≤ p)] {ι : Type u_2} [Fintype ι] (u : ↥(W1p mu Omega p)) (b : OrthonormalBasis ι ℝ E) (h : ∀ (i j : ι), ∃ (g : ↥(MeasureTheory.Lp ℝ p (mu.restrict ↑Omega))), HasWeakLineDerivOn mu Omega (fun (x : E) => inner ℝ (↑↑(gradient u) x) (b j)) (↑↑g) (b i)) :
∃ (U : Wkp mu Omega p 2), Wkp.lowerOrder 1 U = u

Second-order weak differentiability from directional derivatives. Fix an orthonormal basis e of E. If, for every pair of indices i, j, the scalar component ⟪∇u, eⱼ⟫ of the weak gradient of u ∈ W^{1,p}(Ω) has a weak derivative in Lᵖ(Ω) in the direction eᵢ, then u is the first-order part of an element of W^{2,p}(Ω).

This is how a difference-quotient argument, which produces exactly these componentwise derivatives, certifies membership in the second-order Sobolev space.

theorem TauCeti.W1p.exists_lowerOrder_eq_of_gradient_eq_of_hasWeakLineDerivOn {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] {mu : MeasureTheory.Measure E} [mu.IsAddHaarMeasure] {Omega : TopologicalSpace.Opens E} {p : ENNReal} [Fact (1 ≤ p)] {ι : Type u_2} [Fintype ι] {V : TopologicalSpace.Opens E} {w : ↥(W1p mu Omega p)} {u : ↥(W1p mu V p)} (b : OrthonormalBasis ι ℝ E) (hV : V ≤ Omega) (hgrad : ∀ᵐ (x : E) ∂mu.restrict ↑V, ↑↑(gradient w) x = ↑↑(gradient u) x) (h : ∀ (i j : ι), ∃ (g : ↥(MeasureTheory.Lp ℝ p (mu.restrict ↑Omega))), HasWeakLineDerivOn mu Omega (fun (x : E) => inner ℝ (↑↑(gradient w) x) (b j)) (↑↑g) (b i)) :
∃ (U : Wkp mu V p 2), Wkp.lowerOrder 1 U = u

Second-order weak differentiability on a subdomain. Let w ∈ W^{1,p}(Ω) and let V ≤ Ω be open. Suppose that the components ⟪∇w, eⱼ⟫ of the weak gradient of w along an orthonormal basis e have weak derivatives in Lᵖ(Ω) along every eᵢ. Then every u ∈ W^{1,p}(V) whose weak gradient agrees almost everywhere on V with that of w is the first-order part of an element of W^{2,p}(V).

Symmetry of the weak Hessian #

theorem TauCeti.Wkp.inner_iteratedGradient_one_apply_ae_eq {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] {mu : MeasureTheory.Measure E} [mu.IsAddHaarMeasure] {Omega : TopologicalSpace.Opens E} {p : ENNReal} [Fact (1 ≤ p)] (u : Wkp mu Omega p 2) (v w : E) :
(fun (x : E) => inner ℝ ((↑↑(iteratedGradient 1 u) x) v) w) =ᵐ[mu.restrict ↑Omega] fun (x : E) => inner ℝ ((↑↑(iteratedGradient 1 u) x) w) v

The weak Hessian is symmetric. For u ∈ W^{2,p}(Ω) and directions v and w, the mixed weak derivatives ⟪D²u v, w⟫ = ∂_v ∂_w u and ⟪D²u w, v⟫ = ∂_w ∂_v u agree almost everywhere on Ω. Here D²u is the highest recorded weak derivative TauCeti.Wkp.iteratedGradient 1 u.

The weak Hessian is self-adjoint. For u ∈ W^{2,p}(Ω), the weak Hessian D²u(x) = TauCeti.Wkp.iteratedGradient 1 u x is a self-adjoint endomorphism of E for almost every x ∈ Ω.