Documentation

LeanPool.Superdiffusion.Algsuperdiff.Section4.Provider.Proportion.G2SharpArith

G2SharpArith #

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

The 𝒢₂ ε-threshold at the LOG-QUANTILE reading of θ^{-1/p} #

What changes, and why #

RatioTailArith.threshold_le_inv — the step that puts Appendix D's threshold 9 s'^{-1} C⋆^{1/p} θ^{-1/p} below the row normalizer's reciprocal — pays the quantile factor θ^{-1/p} by the crude bound θ^{-1/p} ≤ θ^{-1}. That single step is where a whole polynomial power of 1/θ enters the 𝒢₂ threshold: it is the reason the crude reading of that threshold carries the divisor θ and the reason its binding clause carries θ² rather than θ.

The honest reading is logarithmic: θ^{-1/p} ≤ 3 as soon as log(1/θ) ≤ p log 3, and log(1/θ) ≤ 1/θ − 1, so the existing moment exponent already pays for it — g2MomentExponent's rate branch is at least 64/θ, which dominates 1/θ. No new branch, no new hypothesis, no new parameter: the quantile power of 1/θ is simply not there.

Contents #

The two sharp clauses #

Writing κ₁(d), κ₂(d) for G2ThresholdArith.g2ThresholdConstOne/Two:

(1)   3κ₁(d) r² (1+c₁) C² γ              ≤ θ  c⋆² s⁵ ε²
(2)   3κ₂(d) r⁸ (1+c₁)⁴ exp(−2c⋆³/(Cγ))  ≤ θ⁴     s⁹ ε²

Each is the corresponding clause of the crude reading with one power of θ removed from the right-hand side and the constant tripled. That single power of θ is the whole content of the sharpening: against the consumer's own coupling ε² = s²θ, clause (1) reads γ ≲ θ², i.e. θ ≳ γ^{1/2} — the printed Theorem-C range — where the old θ² form read γ ≲ θ³, i.e. θ ≳ γ^{1/3}.

One polynomial power of 1/θ does remain inside D, through the moment exponent's rate branch 64r(log 3r + c₁r)/(sθ); it is exactly the power the consumer's ε² = s²θ pays for, which is why γ^{1/2} and not γ^{1/3} is the reachable exponent. No third source of 1/θ exists in the chain.

Scope #

Provider material: proved local helpers. Every transcendental step is an abstract-real private helper.

References #

1. The log-quantile step #

theorem Algsuperdiff.Section4.Provider.Proportion.rpow_neg_inv_le_three {theta p : ℝ} (ht0 : 0 < theta) (hp : 1 ≤ p) (hpth : theta⁻¹ ≤ p) :
theta ^ (-1 / p) ≤ 3

The log-quantile bound. θ^{-1/p} ≤ 3 as soon as the moment exponent clears 1/θ.

This is the step the proved RatioTailArith.threshold_le_inv pays crudely as θ^{-1/p} ≤ θ⁻¹. The honest requirement is only log(1/θ) ≤ p log 3, and log(1/θ) ≤ 1/θ − 1 ≤ p − 1 ≤ p log 3. It is the only place the quantile rpow is opened.

theorem Algsuperdiff.Section4.Provider.Proportion.threshold_le_inv_sharp {sprime theta p D : ℝ} (hs : 0 < sprime) (ht0 : 0 < theta) (hp : 1 ≤ p) (hpth : theta⁻¹ ≤ p) (hD : 0 < D) (hkey : 27 * D * (1 + Probability.ScalesConcentration.Cstar) ≤ sprime) :
9 * sprime⁻¹ * Probability.ScalesConcentration.Cstar ^ (1 / p) * theta ^ (-1 / p) ≤ D⁻¹

The Appendix-D threshold below the row normalizer's reciprocal, at the log-quantile reading.

The sharp twin of RatioTailArith.threshold_le_inv: the key condition is 27 D (1+C⋆) ≤ s', with no θ on the right, at the cost of the extra requirement θ⁻¹ ≤ p on the moment exponent — which the lane's own p already satisfies (inv_theta_le_g2MomentExponent).

2. The proved moment exponent already clears 1/θ #

theorem Algsuperdiff.Section4.Provider.Proportion.inv_theta_le_g2MomentExponent {r : ℕ} {s theta c1 : ℝ} (hr1 : 1 ≤ r) (hs0 : 0 < s) (hs1 : s ≤ 1) (hth0 : 0 < theta) (hc1 : 0 ≤ c1) :
theta⁻¹ ≤ g2MomentExponent r s theta c1

The lane's own moment exponent exceeds θ⁻¹.

g2MomentExponent's rate branch is 64r(log 3r + c₁r)/(sθ), and log 3r ≥ 1, r ≥ 1, s ≤ 1, so it is at least 64/θ. Hence the log-quantile step rpow_neg_inv_le_three is available at the unchanged moment exponent: the sharp threshold costs no new branch and no new hypothesis.

3. The sharp threshold and its two constants #

noncomputable def Algsuperdiff.Section4.Provider.Proportion.g2ThresholdSharp (d r : ℕ) (C : ℝ) (M : Section3.ABKModel d) (s theta c1 : ℝ) :

The θ-free ε²-threshold of the 𝒢₂ proportion lane.

g2ThresholdSharp d r s θ c₁ = 108(1+C⋆)·D, with D = g2Normalizer … the lane's own Appendix-D normalizer, unchanged. Compare the proved G2ThresholdArith.g2Threshold … = 36(1+C⋆)·D/θ: the constant is tripled and the divisor θ is gone. The two are therefore related by g2ThresholdSharp = 3θ · g2Threshold, so the sharp threshold is the smaller one exactly when θ < 1/3 — the small-level regime the printed Theorem-C range lives in.

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

    The d-only constant of the binding clause of the sharp threshold condition: three times the proved g2ThresholdConstOne.

    Equations
    Instances For

      The d-only constant of the exponentially small clause of the sharp threshold condition: three times the proved g2ThresholdConstTwo.

      Equations
      Instances For

        4. The sufficient parameter condition #

        theorem Algsuperdiff.Section4.Provider.Proportion.g2ThresholdSharp_le_of {d : ℕ} (M : Section3.ABKModel d) {C s theta c1 ep : ℝ} {r : ℕ} (hr1 : 1 ≤ r) (hC0 : 0 < C) (hs0 : 0 < s) (hs1 : s ≤ 1) (hth0 : 0 < theta) (hth1 : theta ≤ 1) (hc1 : 0 ≤ c1) (hcond1 : g2SharpConstOne d * ↑r ^ 2 * (1 + c1) * C ^ 2 * M.gamma ≤ theta * Section3.Disorder.cstar M ^ 2 * s ^ 5 * ep ^ 2) (hcond2 : g2SharpConstTwo d * ↑r ^ 8 * (1 + c1) ^ 4 * Real.exp (-(2 * (C⁻¹ * Section3.Disorder.cstar M ^ 3 * M.gamma⁻¹))) ≤ theta ^ 4 * s ^ 9 * ep ^ 2) :
        g2ThresholdSharp d r C M s theta c1 ≤ ep ^ 2

        The sharp ε-threshold is met under two closed parameter conditions.

        (1)   3κ₁(d) r² (1+c₁) C² γ              ≤ θ  c⋆² s⁵ ε²
        (2)   3κ₂(d) r⁸ (1+c₁)⁴ exp(−2c⋆³/(Cγ))  ≤ θ⁴     s⁹ ε²
        

        Each clause is the corresponding clause of the crude reading with one power of θ removed from the right-hand side and the constant tripled. Clause (1) is the binding one; at the consumer's coupling ε² = s²θ it reads γ ≲ θ², i.e. θ ≳ γ^{1/2} — the printed Theorem-C range — where the proved θ² form read γ ≲ θ³.

        theorem Algsuperdiff.Section4.Provider.Proportion.g2ThresholdSharp_le_of_gammaPow {d : ℕ} (M : Section3.ABKModel d) {C s theta c1 ep : ℝ} {r : ℕ} (hr1 : 1 ≤ r) (hC0 : 0 < C) (hs0 : 0 < s) (hs1 : s ≤ 1) (hth0 : 0 < theta) (hth1 : theta ≤ 1) (hc1 : 0 ≤ c1) (hcond1 : g2SharpConstOne d * ↑r ^ 2 * (1 + c1) * C ^ 2 * M.gamma ≤ theta * Section3.Disorder.cstar M ^ 2 * s ^ 5 * ep ^ 2) (hcond2 : 945 / 8 * g2ThresholdConstTwo d * ↑r ^ 8 * (1 + c1) ^ 4 * C ^ 7 * M.gamma ^ 7 ≤ theta ^ 4 * Section3.Disorder.cstar M ^ 21 * s ^ 9 * ep ^ 2) :
        g2ThresholdSharp d r C M s theta c1 ≤ ep ^ 2

        The second sharp clause in γ-power form. exp(−w) ≤ 5040/w⁷ at w = 2c⋆³/(Cγ) turns the exponentially small clause into

        (2')  (945/8) κ₂(d) r⁸ (1+c₁)⁴ C⁷ γ⁷ ≤ θ⁴ c⋆²¹ s⁹ ε² ,
        

        six powers of γ weaker than the binding clause (1), which is linear in γ.