Uniform reconstruction bound for the approximant limit from uniform consistency in Hairer's regularity structures
ProvedHairer.reconstruction_limit_bound_of_uniform_consistencyUniform reconstruction bound for the approximant limit, from uniform consistency (Hairer 2014, Section 3.1). Given an approximant net built from atomic decompositions of test functions at dyadic scales , assume the per-test-function consistency estimate and the following uniform consistency: on every compact there is a constant such that every approximant of a rescaled test function S^\\delta_{s,x}\\eta (with , , ) at dyadic scale is within C\\,\\max(2^{-n},\\delta)^\\gamma of the germ \\langle \\Pi_x f(x), S^\\delta_{s,x}\\eta\\rangle. If is the pointwise limit of the net, then on every compact
\\left|\\mathrm{limval}(S^\\delta_{s,x}\\eta) - \\langle \\Pi_x f(x), S^\\delta_{s,x}\\eta\\rangle\\right| \\le C\\,\\delta^\\gammauniformly over , , . Together with consistency this yields Hairer's reconstruction theorem (Theorem 3.10): the limit is the reconstructed distribution. Formalization Note: this repairs the earlier published Hairer.reconstruction_limit_uniform_bound, whose consistency hypothesis yielded only a per-test-function constant while the conclusion needs uniform over , and -- a gap the old statement could not close. The added uniform-consistency hypothesis closes exactly this gap; the remaining proof is a routine limit passage.
import Definitions.Def_Hairer_Model set_option autoImplicit false open scoped Classical DirectSum noncomputable section
namespace Hairer
/-- **Uniform reconstruction bound for the approximant limit, from uniform
consistency (Hairer 2014, §3.1).**
Let `approx` be an approximant net: at each dyadic scale `2^{-n}`, the value
`approx ψ hψ n` is computed from some atomic decomposition of the test function
`ψ` into rescaled test balls. Assume the consistency estimate with coefficient
control holds (hypothesis `hcons`), and assume the uniform consistency estimate
`huniform`: on every compact `K` there is a constant `C` such that every
approximant of a rescaled test function `S^δ_{s,x}η` (with `x ∈ K`,
`δ ∈ (0,1]`, `η ∈ B^r_{s,0}`) at dyadic scale `2^{-n}` is within
`C·max(2^{-n},δ)^γ` of the germ `⟨Π_x f(x), S^δ_{s,x}η⟩`. Let `limval`
be the pointwise limit of the net (which exists by consistency and completeness
of `ℝ`). Then `limval` satisfies the uniform reconstruction bound: on every
compact `K` there is `C` with
`|limval(S^δ_{s,x}η) - ⟨Π_x f(x), S^δ_{s,x}η⟩| ≤ C δ^γ`
uniformly over `x ∈ K`, `δ ∈ (0,1]` and `η ∈ B^r_{s,0}`. Together with the
consistency estimate this yields the reconstruction theorem (Theorem 3.10):
the limit is the reconstructed distribution.
This statement repairs `Hairer.reconstruction_limit_uniform_bound`, whose
hypothesis `hcons` gave only a per-test-function constant `C(φ)` while the
conclusion needs `C` uniform over `x ∈ K`, `δ` and `η` — a gap the old
statement could not close. The added uniform-consistency hypothesis `huniform`
closes exactly this gap; the remaining proof is a routine limit passage. -/
theorem reconstruction_limit_bound_of_uniform_consistency
{d : ℕ} {s : Fin d → ℕ} (hs : IsScaling s)
{A : Set ℝ} {E : A → Type} [∀ a : A, NormedAddCommGroup (E a)]
[∀ a : A, NormedSpace ℝ (E a)]
{G : Subgroup (ModelSpace A E ≃ₗ[ℝ] ModelSpace A E)} {one : ModelSpace A E}
(hT : IsRegularityStructure A E G one)
{r : ℕ} {Pi : Pt d → ModelSpace A E →ₗ[ℝ] Distrib d}
{Gam : Pt d → Pt d → ModelSpace A E ≃ₗ[ℝ] ModelSpace A E}
(hmod : IsModel s r G Pi Gam)
{α : ℝ} (hα : IsLeast A α) (hαneg : α < 0)
{γ : ℝ} (hγ : 0 < γ)
{f : Pt d → ModelSpace A E} (hf : IsModelled s γ Gam f)
(approx : (ψ : Pt d → ℝ) → ψ ∈ testFunctions d → ℕ → ℝ)
(happrox : ∀ (ψ : Pt d → ℝ) (hψ : ψ ∈ testFunctions d) (n : ℕ),
∃ (m : ℕ) (x : Fin m → Pt d) (c : Fin m → ℝ) (η : Fin m → (Pt d → ℝ)),
(∀ i, IsTestBall s r (η i)) ∧
(∀ y, ψ y = ∑ i, c i * scaledTest s ((2 : ℝ) ^ (-(n : ℤ))) (x i) (η i) y) ∧
approx ψ hψ n = ∑ i, c i * (Pi (x i) (f (x i))).eval
(scaledTest s ((2 : ℝ) ^ (-(n : ℤ))) (x i) (η i)))
(hcons : ∀ (φ : Pt d → ℝ) (hφ : φ ∈ testFunctions d),
∀ (K : Set (Pt d)) (M : ℝ), IsCompact K →
∃ C : ℝ,
∀ (δ δ' : ℝ) (p p' : ℕ)
(x : Fin p → Pt d) (c : Fin p → ℝ) (η : Fin p → (Pt d → ℝ))
(x' : Fin p' → Pt d) (c' : Fin p' → ℝ) (η' : Fin p' → (Pt d → ℝ)),
0 < δ → δ ≤ 1 → 0 < δ' → δ' ≤ 1 →
(∀ i, x i ∈ K) → (∀ i, x' i ∈ K) →
(∑ i, |c i|) ≤ M → (∑ i, |c' i|) ≤ M →
(∀ i, IsTestBall s r (η i)) → (∀ i, IsTestBall s r (η' i)) →
(∀ y, φ y = ∑ i, c i * scaledTest s δ (x i) (η i) y) →
(∀ y, φ y = ∑ i, c' i * scaledTest s δ' (x' i) (η' i) y) →
|∑ i, c i * (Pi (x i) (f (x i))).eval (scaledTest s δ (x i) (η i))
- ∑ i, c' i * (Pi (x' i) (f (x' i))).eval (scaledTest s δ' (x' i) (η' i))|
≤ C * (max δ δ') ^ γ)
(huniform : ∀ (K : Set (Pt d)), IsCompact K → ∃ C : ℝ,
∀ (x : Pt d), x ∈ K → ∀ (δ : ℝ), ∀ hδ : 0 < δ, ∀ _ : δ ≤ 1,
∀ (η : Pt d → ℝ), ∀ hη : IsTestBall s r η, ∀ (n : ℕ),
|approx (scaledTest s δ x η)
(scaledTest_mem s hδ x ⟨hη.smooth, hη.compactSupport⟩) n
- (Pi x (f x)).eval (scaledTest s δ x η)|
≤ C * (max ((2 : ℝ) ^ (-(n : ℤ))) δ) ^ γ)
(limval : (ψ : Pt d → ℝ) → ψ ∈ testFunctions d → ℝ)
(hlim : ∀ (ψ : Pt d → ℝ) (hψ : ψ ∈ testFunctions d),
Filter.Tendsto (approx ψ hψ) Filter.atTop (nhds (limval ψ hψ))) :
∀ K : Set (Pt d), IsCompact K → ∃ C : ℝ, ∀ (x : Pt d), x ∈ K →
∀ (δ : ℝ), ∀ hδ : 0 < δ, ∀ _ : δ ≤ 1, ∀ (η : Pt d → ℝ), ∀ hη : IsTestBall s r η,
|limval (scaledTest s δ x η) (scaledTest_mem s hδ x ⟨hη.smooth, hη.compactSupport⟩)
- (Pi x (f x)).eval (scaledTest s δ x η)| ≤ C * δ ^ γ := by
sorry
end Hairer