Documentation

LeanPool.Superdiffusion.Algsuperdiff.Section3.Provider.Diffusivity.ApproximateRecurrence.Closure.GammaTenEnvelope

GammaTenEnvelope #

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

The pathwise envelope of a monotone family with a uniform fourth moment #

ABK26, Step 2 of l.approximate.recurrence.formula.

The obstruction this module removes #

LocalizationFluctuationPerCube.perCube_localizationFluctuation_le consumes the Besov envelope of bfF_z in the pathwise form

  hpos1 hpos2 :  forall N, [ gauged bfF_z ]_{B^s_{2,2}, <= N}  <=  Bg

--- a single real Bg, valid at every depth cutoff, at the sample point where the per-cube bound is being read. The oscillation input of Step 2 is annealed instead: the pathwise fold cannot be instantiated at the recurrence's own data, and what is available is a depth fold in the mean,

  E[ [ u ]_{<= N}^4 ]  <=  4 (A 3^{-B})^4      (uniformly in N).

The two are reconciled here. The finite-depth positive q = 2 seminorms are monotone in the depth cutoff (their square is a sum of nonnegative terms over range (N+1)), so the fourth powers form a monotone family; monotone convergence turns a bound on E[u_N^4] that is uniform in N into the same bound on the fourth moment of the pointwise envelope, and in particular the envelope is finite almost surely. That envelope is the pathwise Bg the per-cube bound asks for, and its fourth moment is the oscillation leg the annealed Hoelder fold consumes.

What is proved #

The device built on them is Closure.GammaTenEnvelopeInputGrid.exists_grid_pathwise_envelope_of_monotone: a monotone, measurable, nonnegative family whose fourth moments are bounded by a single C uniformly in the depth cutoff admits a measurable pathwise envelope B with

  (a.e. w)  forall N, u N w <= B w ,        0 <= B ,        E[B^4] <= C ,

run simultaneously at every cell of the mesh. The almost-everywhere quantifier is the exact strength the consumer needs: the per-cube bound is read inside a Bochner integral, where integral_mono_ae applies.

Binders #

Nonnegativity of the family, and finiteness of the supremum of its fourth powers wherever a member is to be dominated. No smallness gate, no geometry, nothing about the corrector or the coefficient field.

Scope #

There is no sorry, no admit, no custom axiom and no set_option maxHeartbeats.

References #

noncomputable def Algsuperdiff.Section3.Provider.Diffusivity.ApproximateRecurrence.Closure.envelopeSup {Omega : Type u_1} (u : ℕ → Omega → ℝ) (w : Omega) :

The pointwise envelope of a family of nonnegative reals, made total.

The supremum of the fourth powers is taken, so that it exists with no boundedness hypothesis, and the fourth root is taken after reading it back in ℝ.

Equations
Instances For
    theorem Algsuperdiff.Section3.Provider.Diffusivity.ApproximateRecurrence.Closure.pow_envelopeSup {Omega : Type u_1} (u : ℕ → Omega → ℝ) (w : Omega) :
    envelopeSup u w ^ 4 = (⨆ (N : ℕ), ENNReal.ofReal (u N w ^ 4)).toReal

    The fourth power of the envelope is the supremum itself, read in ℝ.

    theorem Algsuperdiff.Section3.Provider.Diffusivity.ApproximateRecurrence.Closure.le_envelopeSup_of_ne_top {Omega : Type u_1} {u : ℕ → Omega → ℝ} {w : Omega} (hw : ⨆ (N : ℕ), ENNReal.ofReal (u N w ^ 4) ≠ ⊤) (hnn : ∀ (N : ℕ), 0 ≤ u N w) (N : ℕ) :
    u N w ≤ envelopeSup u w

    Every member of the family is below the envelope, wherever the supremum is finite.