Documentation

TauCeti.Analysis.PDE.HeatKernel.DuhamelPotential.Basic

The Duhamel heat potential, its spatial derivatives and the Hölder estimates of its Hessian #

For a forcing term f : ℝ → E → F on a finite-dimensional real inner product space E, the Duhamel potential (or volume potential) is

w(t, x) = ∫₀ᵗ (K_{t-s} ⋆ f(s, ·))(x) ds,

where K_τ is the heat kernel. By Duhamel's principle it is the candidate solution of the inhomogeneous heat equation ∂ₜw = Δw + f with zero initial datum, complementing the heat convolution K_t ⋆ g of an initial datum g. This file proves the spatial regularity of each time slice w(t, ·), t > 0, and the temporal regularity of its Hessian, with bounds whose constants depend only on the dimension (and on α below):

The time integrand is singular at s = t. For the first derivative, the bound ‖D(K_τ ⋆ g)‖ ≤ ‖DK_τ‖_{L¹} M = O(M τ^(-1/2)) is integrable in τ. For the second derivative, ‖D²K_τ‖_{L¹} is of order τ⁻¹, which is not integrable. Since ∫ D²K_τ = 0, the value of g at the base point can be subtracted, and Hölder continuity improves the bound to O(H τ^(α/2 - 1)) (TauCeti.norm_iteratedFDeriv_heatKernel_convolution_le_of_holderWith), which is integrable for α > 0. Derivatives are then taken under the time integral with TauCeti.iteratedFDeriv_intervalIntegral. Measurability of the derivatives in the time variable comes from the parabolic scaling D^m K_τ(z) = τ^(-(n + m)/2) D^m K_1(τ^(-1/2) z).

Without the Hölder condition the second derivative of w(t, ·) need not exist, so this is the form in which D² w is controlled by the forcing.

For the Hölder estimate of D² w(t, ·) at two points at distance δ, the times t - δ² < s < t are handled by the bound above, which integrates to O(H δ^α). For s < t - δ² the mean value theorem and the bound ‖D³(K_τ ⋆ g)‖ = O(H τ^((α - 3)/2)), obtained by the same cancellation, give a contribution O(H δ ∫_{δ²}^∞ τ^((α - 3)/2) dτ) = O(H δ^α), which requires α < 1. This is the spatial half of the order-two parabolic Hölder estimate of the Duhamel potential.

The temporal half compares D² w at times t' < t, with h = t - t'. If t' ≤ h, both Hessians are O(H h^(α/2)) by the bound above. Otherwise the times s > t' - h contribute O(H h^(α/2)) by the same bound, while for s < t' - h the slices at elapsed times t' - s and t - s differ by O(H h (t' - s)^((α - 4)/2)) (TauCeti.norm_iteratedFDeriv_heatKernel_convolution_sub_le_of_holderWith, one time derivative of the kernel at the cost of a factor (t' - s)⁻¹), which integrates to O(H h^(α/2)) when α < 2. For 0 < α < 1, the two halves together give the parabolic α-Hölder estimate of D² w.

Main declarations #

References #

noncomputable def TauCeti.heatDuhamelPotential {E : Type u_1} {F : Type u_2} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] [NormedAddCommGroup F] [NormedSpace ℝ F] (f : ℝ → E → F) (t : ℝ) (x : E) :
F

The Duhamel heat potential of a forcing f, w(t, x) = ∫₀ᵗ (K_{t-s} ⋆ f(s, ·))(x) ds, where K_τ is the heat kernel.

Instances For

    The defining formula of the Duhamel heat potential.

    The Duhamel heat potential commutes with scalar multiplication, at every time.

    @[simp]

    The Duhamel heat potential vanishes at time zero.

    @[simp]
    theorem TauCeti.heatDuhamelPotential_const {E : Type u_1} {F : Type u_2} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] [NormedAddCommGroup F] [NormedSpace ℝ F] [CompleteSpace F] (c : F) {t : ℝ} (ht : 0 ≤ t) (x : E) :
    heatDuhamelPotential (fun (x : ℝ) (x_1 : E) => c) t x = t • c

    Constant forcing produces linear growth in time.

    Without completeness of F, the Duhamel heat potential vanishes.

    The sup bound. If ‖f(s, y)‖ ≤ M for almost every y, for almost every 0 < s < t, then ‖w(t, x)‖ ≤ M t.

    Joint measurability of the derivatives of heat convolutions. For jointly measurable f, some jointly strongly measurable function of (s, τ) equals D^m (K_τ ⋆ f(s, ·))(x) whenever τ > 0 and f(s, ·) is essentially bounded. Writing D^m K_τ through the parabolic scaling of K_1 makes the convolution integrand jointly measurable in (s, τ, z).

    For f jointly measurable and essentially bounded on (0, t) × E, the time integrand s ↦ (K_{t-s} ⋆ f(s, ·))(x) of the Duhamel potential is interval integrable on [0, t].

    The Duhamel heat potential is additive for jointly measurable, essentially bounded forcing on the integration interval. No spatial continuity is required.

    For f jointly measurable and essentially bounded on (0, t) × E, the time integrands s ↦ D^k (K_{t-s} ⋆ f(s, ·))(x) of the spatial derivatives of order at most one of the Duhamel potential are interval integrable on [0, t].

    If f is jointly measurable and essentially bounded on (0, t) × E, and f(s, ·) is α-Hölder for almost every 0 < s < t, with α > 0, then the time integrands s ↦ D^k (K_{t-s} ⋆ f(s, ·))(x) of the spatial derivatives of order at most two of the Duhamel potential are interval integrable on [0, t].

    The first derivative of the Duhamel potential. For f jointly measurable and essentially bounded on (0, t) × E, the derivatives of w(t, ·) of order at most one are computed under the time integral: D^k w(t, x) = ∫₀ᵗ D^k (K_{t-s} ⋆ f(s, ·))(x) ds.

    For f jointly measurable and essentially bounded on (0, t) × E, the Duhamel potential w(t, ·) is C¹.

    The second derivative of the Duhamel potential. If f is jointly measurable and essentially bounded on (0, t) × E, and f(s, ·) is α-Hölder with constant H for almost every 0 < s < t, with α > 0, then the derivatives of w(t, ·) of order at most two are computed under the time integral: D^k w(t, x) = ∫₀ᵗ D^k (K_{t-s} ⋆ f(s, ·))(x) ds.

    If f is jointly measurable and essentially bounded on (0, t) × E, and f(s, ·) is α-Hölder for almost every 0 < s < t, with α > 0, then the Duhamel potential w(t, ·) is C².

    The gradient bound for the Duhamel potential. There is a constant C, depending only on the dimension, such that ‖D w(t, x)‖ ≤ C M √t whenever f is jointly measurable and ‖f(s, y)‖ ≤ M for almost every y, for almost every 0 < s < t.

    theorem TauCeti.exists_norm_iteratedFDeriv_two_heatDuhamelPotential_le {E : Type u_1} {F : Type u_2} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] [NormedAddCommGroup F] [NormedSpace ℝ F] {α : NNReal} (hα : 0 < α) :
    ∃ (C : ℝ), ∀ {f : ℝ → E → F} {M t : ℝ} {H : NNReal}, MeasureTheory.StronglyMeasurable (Function.uncurry f) → (∀ᵐ (s : ℝ) ∂MeasureTheory.volume.restrict (Set.Ioo 0 t), ∀ᵐ (y : E), ‖f s y‖ ≤ M) → (∀ᵐ (s : ℝ) ∂MeasureTheory.volume.restrict (Set.Ioo 0 t), HolderWith H α (f s)) → 0 < t → ∀ (x : E), ‖iteratedFDeriv ℝ 2 (heatDuhamelPotential f t) x‖ ≤ C * ↑H * t ^ (↑α / 2)

    The Hessian bound for the Duhamel potential. For α > 0 there is a constant C, depending only on α and the dimension, such that ‖D² w(t, x)‖ ≤ C H t^(α / 2) whenever f is jointly measurable and essentially bounded on (0, t) × E and f(s, ·) is α-Hölder with constant H for almost every 0 < s < t. The bound on f does not enter the estimate.

    The Hölder estimate of the Hessian #

    The Hölder estimate for the Hessian of the Duhamel potential. For 0 < α < 1 there is a constant C, depending only on α and the dimension, such that the Hessian D² w(t, ·) is α-Hölder with constant C H whenever f is jointly measurable and essentially bounded on (0, t) × E and f(s, ·) is α-Hölder with constant H for almost every 0 < s < t. Neither the bound on f nor the time t enters the estimate.

    The temporal Hölder estimate of the Hessian #

    theorem TauCeti.exists_holderOnWith_iteratedFDeriv_two_heatDuhamelPotential {E : Type u_1} {F : Type u_2} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] [NormedAddCommGroup F] [NormedSpace ℝ F] [CompleteSpace F] {α : NNReal} (hα₀ : 0 < α) (hα₂ : α < 2) :
    ∃ (C : NNReal), ∀ {f : ℝ → E → F} {M T : ℝ} {H : NNReal}, MeasureTheory.StronglyMeasurable (Function.uncurry f) → (∀ᵐ (s : ℝ) ∂MeasureTheory.volume.restrict (Set.Ioo 0 T), ∀ᵐ (y : E), ‖f s y‖ ≤ M) → (∀ᵐ (s : ℝ) ∂MeasureTheory.volume.restrict (Set.Ioo 0 T), HolderWith H α (f s)) → ∀ (x : E), HolderOnWith (C * H) (α / 2) (fun (t : ℝ) => iteratedFDeriv ℝ 2 (heatDuhamelPotential f t) x) (Set.Icc 0 T)

    The temporal Hölder estimate for the Hessian of the Duhamel potential. For 0 < α < 2 there is a constant C, depending only on α and the dimension, such that at every point x the Hessian t ↦ D² w(t, ·)(x) is α/2-Hölder in time on [0, T] with constant C H, whenever f is jointly measurable and essentially bounded on (0, T) × E and f(s, ·) is α-Hölder with constant H for almost every 0 < s < T. Neither the bound on f nor the horizon T enters the estimate. Together with TauCeti.exists_holderWith_iteratedFDeriv_two_heatDuhamelPotential this is the parabolic α-Hölder estimate of D² w on the slab E × [0, T], for 0 < α < 1.