Documentation

LeanPool.Superdiffusion.Algsuperdiff.Section4.Provider.ExcessDecay.OneStepDatumPriceOdd

OneStepDatumPriceOdd #

Ported from scottnarmstrong/Superdiffusion at 7f20476385d4af418f110303085ad0740c9eba46. Module imports, support namespace and local option directives adapted for Lean Pool.

Pricing the odd-class defect: the split, the reduction, and the refutation #

OneStepBoundaryOdd compresses the boundary branch's whole datum contribution into the single scalar oddClassDefect x m n V c A recorded as its residue the pricing

  oddClassDefect x m n V c A ≤ C(d) · (3^{n-2})^{3/2} · [∇h]_{C^{0,1/2}(U₂)} ,

read as the manuscript's second boundary-Schauder leg. This module settles the shape of that residue.

1. The universally quantified reading is refuted #

oddClassDefect x m n V c A is stated for an arbitrary odd affine datum (c,A) ∈ 𝕃_odd. At the zero competitor the defect is exactly the seminorm of that datum (oddClassDefect_zero_competitor), so on any window carrying the affine nondegeneracy the defect is strictly positive for every nonzero odd affine datum (oddClassDefect_zero_competitor_pos) — while the datum side of the pricing may be 0 (take h ≡ 0; note that hVodd already forces a continuous competitor to vanish on the met face, so h ≡ 0 on the face is the compatible datum, not an extra assumption). The pricing therefore cannot hold for an arbitrary member of the odd class: it must be read existentially, at the best odd affine datum, exactly as the manuscript's min over 𝕃_odd does.

2. The best odd datum: the split at the met face #

At a single met face {yᵢ = a} the odd class is the line ℝ·(yᵢ − a) (OddReflectionMap.slope_eq_zero_of_ne_of_meets). The canonical odd affine datum attached to (c,A) is therefore its normal part Aᵢ·(yᵢ − a) (oddAffineIntercept, oddAffineSlope), and the complement — the even part evenAffineIntercept, evenAffineSlope — has vanishing i-slope (evenAffineSlope_apply_self), i.e. it is constant along the reflection.

oddClassDefect_oddPart_le prices the defect at the odd part of any affine datum by that datum's own excess gap plus the seminorm of the even part; at an affine minimizer the gap vanishes and the reduction is exact:

  oddClassDefect x m n V (odd part of ℓ*) ≤ ‖even part of ℓ*‖_{L²(U₂)} .

3. The exact remaining inequality (reported, not proved) #

Let V be odd about the met face and harmonic on the doubled window, and let ℓ* be an affine minimizer of V on U₂. What remains is

  ‖even part of ℓ*‖_{L²(U₂)}  ≤  C(d) · affineExcessRaw U₂ V .        (★)

Both sides are L²(U₂) quantities of the same homogeneity in V, and neither mentions h; (★) is exactly the statement that odd harmonic functions cannot approximate a nonzero even affine function on the half-box better than a fixed fraction. It is true — a competitor odd about the face vanishes there, while an even affine function does not, and harmonicity across the face (the doubled window puts the face in the interior) caps the transition rate at the window scale, so the near-face slab already costs a fixed fraction of ‖even part‖. Its formalization needs interior gradient estimates for harmonic functions at the face together with a reverse-Chebyshev bound for affine functions, and is not attempted here.

(★) prices the defect by the excess itself, not by [∇h]. That is the architectural finding of this module: under hVodd the competitor vanishes on the met face, i.e. the datum has already been subtracted, so the manuscript's [∇h]_{C^{0,1/2}(U₂)} leg cannot enter through oddClassDefect; it belongs to the (still missing) reduction from the true competitor v = h on ∂□_m to an odd one.

1. The odd/even split of an affine datum at a met upper face #

The intercept of the normal part Aᵢ·(yᵢ − a) of the affine datum (c,A) at the met upper i-face {yᵢ = a}, a = (1/2)·3^m, in the affineLift parametrization.

Equations
Instances For

    The slope of the normal part: A with every tangential coordinate erased.

    Equations
    Instances For

      The intercept of the even part, in the affineEval parametrization.

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

        The even part is constant along the reflection: its normal slope vanishes.

        The normal part pairs with a vector through its i-coordinate only.

        The normal part is the affine function Aᵢ·(yᵢ − a).

        theorem Algsuperdiff.Section4.Provider.ExcessDecay.Schauder.isOddAffineData_oddAffineDatum {d : ℕ} {x : HCPolySupport.Vec d} {m k : ℤ} {i : Fin d} (hkm : k < m) (hup : MeetsUpperFace x m k i) (hother : ∀ (j : Fin d), j ≠ i → ¬MeetsUpperFace x m k j ∧ ¬MeetsLowerFace x m k j) (A : HCPolySupport.Vec d) :

        The normal part lies in the odd class. In the one-met-face regime the canonical odd affine datum attached to (c,A) is admissible for OneStepBoundaryOdd.

        2. The reduction: the defect at the normal part #

        The affine datum splits: ℓ = (normal part) + (even part), pointwise.

        The reduction. The defect at the normal part of an affine datum is at most that datum's own gap to the excess minimum plus the seminorm of the even part.

        The reduction at a minimizer. If (c,A) attains the excess minimum on U₂, the defect at its normal part is at most the seminorm of its even part. This is the exact residue: what remains's is the bound of the right side by C(d) · affineExcessRaw U₂ V.

        3. The refutation of the universally quantified pricing #