Documentation

LeanPool.Chvatal.Weighted

The spectral coefficients in the weighted star inequality #

This file formalizes Proposition 5.3 of arXiv:2609.19123. The paper assigns each nonempty Fourier set to its largest coordinate in a fixed total order. The arguments below allow any fixed selector belonging to that set, a slightly stronger formulation which includes the paper's choice.

noncomputable def Chvatal.spectralWeights {ι : Type u_1} [Fintype ι] [DecidableEq ι] (g : Finset ι → ℝ) (select : Finset ι → ι) (i : ι) :

Proposition 5.3's coefficients λ_i: four times the nonconstant Fourier mass assigned to coordinate i. The selector is fixed independently of the weight or hereditary family under consideration.

Equations
Instances For
    theorem Chvatal.spectralWeights_nonneg {ι : Type u_1} [Fintype ι] [DecidableEq ι] (g : Finset ι → ℝ) (select : Finset ι → ι) (i : ι) :
    0 ≤ spectralWeights g select i

    Nonnegativity of every coefficient in Proposition 5.3.

    theorem Chvatal.sum_spectralWeights_mul {ι : Type u_1} [Fintype ι] [DecidableEq ι] (g : Finset ι → ℝ) (select : Finset ι → ι) (a : ι → ℝ) :
    ∑ i : ι, spectralWeights g select i * a i = 4 * ∑ S ∈ Finset.univ.erase ∅, fourier g S ^ 2 * a (select S)

    Regrouping Fourier sets according to their selected coordinate, the partition-of-mass step in Proposition 5.3.

    theorem Chvatal.sum_spectralWeights {ι : Type u_1} [Fintype ι] [DecidableEq ι] {g : Finset ι → ℝ} (hg : IsBoolean g) (hdual : dual g = g) (select : Finset ι → ι) :
    ∑ i : ι, spectralWeights g select i = 1

    The coefficients in Proposition 5.3 sum to one, by Parseval and the variance-one-quarter identity for an antipodal Boolean function.

    noncomputable def Chvatal.orderedSelector {ι : Type u_1} [LinearOrder ι] [Nonempty ι] (S : Finset ι) :
    ι

    A fixed total order gives the selector max_≺ S used in Proposition 5.3. The value at the empty set is irrelevant because the coefficients omit it.

    Equations
    Instances For
      theorem Chvatal.orderedSelector_mem {ι : Type u_1} [LinearOrder ι] [Nonempty ι] {S : Finset ι} (hS : S.Nonempty) :

      The ordered selector lies in each nonempty set, the property of the paper's max_≺ used in the weighted argument.

      theorem Chvatal.hereditary_bound_of_selected_correlation {ι : Type u_1} [Fintype ι] [DecidableEq ι] {g : Finset ι → ℝ} (hg : IsBoolean g) (hdual : dual g = g) (select : Finset ι → ι) (hcor : ∀ (f : Finset ι → ℝ), IsBoolean f → Monotone f → ∑ S ∈ Finset.univ.erase ∅, fourier g S ^ 2 * influence f (select S) ≤ covariance f g) {D : Family ι} (hD : D.IsHereditary) :
      ↑(Finset.card (D ∩ Family.oneSupport g)) ≤ ∑ i : ι, spectralWeights g select i * ↑(Finset.card (D.star i))

      The hereditary-family inequality in the proof of Proposition 5.3, deduced from the selected-coordinate spectral lower bound. This hypothesis is explicit so the counting reduction is independent of the analytic proof.

      def Chvatal.starMixture {ι : Type u_1} [Fintype ι] [DecidableEq ι] (weights : ι → ℝ) (x : Finset ι) :

      The fractional star function q(x) = ∑_i λ_i x_i from the proof of Proposition 5.3.

      Equations
      Instances For
        theorem Chvatal.sum_starMixture {ι : Type u_1} [Fintype ι] [DecidableEq ι] (weights : ι → ℝ) (D : Family ι) :
        ∑ x ∈ D, starMixture weights x = ∑ i : ι, weights i * ↑(Finset.card (D.star i))

        Counting the fractional star function on a family gives the corresponding weighted combination of its star cardinalities (Proposition 5.3).

        theorem Chvatal.sum_indicator_on_family {ι : Type u_1} [DecidableEq ι] (B D : Family ι) :
        ∑ x ∈ D, B.indicator x = ↑(Finset.card (D ∩ B))

        Summing a family indicator over another family counts their intersection, the second counting step for the defect in Proposition 5.3.

        theorem Chvatal.sum_starMixture_sub_indicator {ι : Type u_1} [Fintype ι] [DecidableEq ι] (weights : ι → ℝ) (B D : Family ι) :
        ∑ x ∈ D, (starMixture weights x - B.indicator x) = ∑ i : ι, weights i * ↑(Finset.card (D.star i)) - ↑(Finset.card (D ∩ B))

        The mean-zero defect in Proposition 5.3, summed on a hereditary family. The identity itself holds for every family.

        theorem Chvatal.sum_defect_mul {ι : Type u_1} [Fintype ι] [DecidableEq ι] (weights : ι → ℝ) (B : Family ι) (ω : Finset ι → ℝ) :
        ∑ x : Finset ι, (starMixture weights x - B.indicator x) * ω x = ∑ i : ι, weights i * ∑ x ∈ Family.star Finset.univ i, ω x - ∑ x ∈ B, ω x

        The weighted form of the same defect, obtained by interchanging finite sums. This is the finite counterpart of the layer-cake calculation in (18).

        theorem Chvatal.weighted_bound_of_hereditary_bound {ι : Type u_1} [Fintype ι] [DecidableEq ι] (weights : ι → ℝ) (B : Family ι) (hbound : ∀ (D : Family ι), D.IsHereditary → ↑(Finset.card (D ∩ B)) ≤ ∑ i : ι, weights i * ↑(Finset.card (D.star i))) (ω : Finset ι → ℝ) (hω : ∀ (x : Finset ι), 0 ≤ ω x) (hanti : Antitone ω) :
        ∑ x ∈ B, ω x ≤ ∑ i : ι, weights i * ∑ x ∈ Family.star Finset.univ i, ω x

        Finite layer-cake lifting in Proposition 5.3: domination on every hereditary family implies domination for every nonnegative decreasing weight.

        theorem Chvatal.selected_spectral_le {ι : Type u_1} [Fintype ι] [DecidableEq ι] (f g : Finset ι → ℝ) (select : Finset ι → ι) (hselect : ∀ (S : Finset ι), S.Nonempty → select S ∈ S) :
        ∑ S ∈ Finset.univ.erase ∅, fourier g S ^ 2 * influence f (select S) ≤ spectralWeight f g

        The selected-coordinate expression in Proposition 5.3 is bounded by the spectral expression in Theorem 1.2, because the selector belongs to its set.

        theorem Chvatal.weighted_bound_of_spectral_bound {ι : Type u_1} [Fintype ι] [DecidableEq ι] [Nonempty ι] {B : Family ι} (hB : B.IsMaximalIntersecting) (select : Finset ι → ι) (hselect : ∀ (S : Finset ι), S.Nonempty → select S ∈ S) (hspectral : ∀ (f : Finset ι → ℝ), IsBoolean f → Monotone f → spectralWeight f B.indicator ≤ covariance f B.indicator) (ω : Finset ι → ℝ) (hω : ∀ (x : Finset ι), 0 ≤ ω x) (hanti : Antitone ω) :
        ∑ x ∈ B, ω x ≤ ∑ i : ι, spectralWeights B.indicator select i * ∑ x ∈ Family.star Finset.univ i, ω x

        The first inequality of Proposition 5.3 deduced from the antipodal spectral bound, keeping the analytic dependency explicit.

        theorem Chvatal.weighted_star_mixture_bound {ι : Type u_1} [Fintype ι] [DecidableEq ι] [Nonempty ι] {B : Family ι} (hB : B.IsMaximalIntersecting) (select : Finset ι → ι) (hselect : ∀ (S : Finset ι), S.Nonempty → select S ∈ S) (ω : Finset ι → ℝ) (hω : ∀ (x : Finset ι), 0 ≤ ω x) (hanti : Antitone ω) :
        ∑ x ∈ B, ω x ≤ ∑ i : ι, spectralWeights B.indicator select i * ∑ x ∈ Family.star Finset.univ i, ω x

        Proposition 5.3's first inequality, proved from the antipodal case of Theorem 1.2. It holds for any fixed selector belonging to each nonempty index.

        theorem Chvatal.star_mixture_le_max {ι : Type u_1} [Fintype ι] [DecidableEq ι] [Nonempty ι] (weights : ι → ℝ) (hweights : ∀ (i : ι), 0 ≤ weights i) (hsum : ∑ i : ι, weights i = 1) (ω : Finset ι → ℝ) :
        ∑ i : ι, weights i * ∑ x ∈ Family.star Finset.univ i, ω x ≤ Finset.univ.sup' ⋯ fun (i : ι) => ∑ x ∈ Family.star Finset.univ i, ω x

        A convex combination of star weights is at most their maximum, the second inequality in equation (18).

        theorem Chvatal.kleitman_weighted_bound {ι : Type u_1} [Fintype ι] [DecidableEq ι] [LinearOrder ι] [Nonempty ι] {B : Family ι} (hB : B.IsMaximalIntersecting) (ω : Finset ι → ℝ) (hω : ∀ (x : Finset ι), 0 ≤ ω x) (hanti : Antitone ω) :
        (∀ (i : ι), 0 ≤ spectralWeights B.indicator orderedSelector i) ∧ ∑ i : ι, spectralWeights B.indicator orderedSelector i = 1 ∧ ∑ x ∈ B, ω x ≤ ∑ i : ι, spectralWeights B.indicator orderedSelector i * ∑ x ∈ Family.star Finset.univ i, ω x ∧ ∑ i : ι, spectralWeights B.indicator orderedSelector i * ∑ x ∈ Family.star Finset.univ i, ω x ≤ Finset.univ.sup' ⋯ fun (i : ι) => ∑ x ∈ Family.star Finset.univ i, ω x

        Proposition 5.3 in the paper's exact ordered-selector formulation: the spectral coefficients are nonnegative, sum to one, and satisfy both inequalities in equation (18). The total order represents the paper's fixed permutation.

        theorem Chvatal.exists_weighted_star_bound {ι : Type u_1} [Fintype ι] [DecidableEq ι] [Nonempty ι] {A : Family ι} (hA : A.IsIntersecting) (ω : Finset ι → ℝ) (hω : ∀ (x : Finset ι), 0 ≤ ω x) (hanti : Antitone ω) :
        ∃ (i : ι), ∑ x ∈ A, ω x ≤ ∑ x ∈ Family.star Finset.univ i, ω x

        The final conclusion of Proposition 5.3 for one intersecting family: its total nonnegative decreasing weight is bounded by some full-cube star.

        theorem Chvatal.exists_largest_weighted_intersecting_star {ι : Type u_1} [Fintype ι] [DecidableEq ι] [Nonempty ι] (ω : Finset ι → ℝ) (hω : ∀ (x : Finset ι), 0 ≤ ω x) (hanti : Antitone ω) :
        ∃ (i : ι), (Family.star Finset.univ i).IsIntersecting ∧ ∀ (A : Family ι), A.IsIntersecting → ∑ x ∈ A, ω x ≤ ∑ x ∈ Family.star Finset.univ i, ω x

        The final maximum-attainment statement of Proposition 5.3: one star simultaneously dominates every intersecting family for a fixed nonnegative decreasing weight, and that star is itself intersecting.