OAI/NumberTheory/EgyptianFractions/SecondMomentAssembly.lean

GitHub source file

9 declarations, 0 contain sorry, 0 declared as axioms

  1. def OAI.Problem337.weightedCorrelation line 9

    Source documentation

    Finite weighted pair correlation, with no probability-library dependency.

    def weightedCorrelation {Ω I : Type*} [Fintype Ω]
        (w : Ω → ℝ) (z : Ω → I → ℂ) (i j : I) : ℂ :=
    …
  2. theorem OAI.Problem337.norm_sum_sq_eq_sum_pair line 13
  3. theorem OAI.Problem337.weighted_second_moment_identity line 22
  4. theorem OAI.Problem337.second_moment_exceptional_bound line 33

    Source documentation

    Exceptional pairs cost their cardinality; all other pairs cost the common correlation bound. The assumptions deliberately expose the exact analytic interface.

    theorem second_moment_exceptional_bound {Ω I : Type*} [Fintype Ω] [Fintype I]
        (w : Ω → ℝ) (z : Ω → I → ℂ) (E : Finset (I × I)) (δ : ℝ) (hδ : 0 ≤ δ)
        (hall : ∀ i j, ‖weightedCorrelation w z i j‖ ≤ 1)
        (hgood : ∀ i j, (i, j) ∉ E → ‖weightedCorrelation w z i j‖ ≤ δ) :
        (∑ ω, w ω * ‖∑ i, z ω i‖ ^ 2) ≤
          (E.card : ℝ) + (Fintype.card I : ℝ) ^ 2 * δ := by
    …

    Mentions: OAI.Problem337.weightedCorrelation

  5. theorem OAI.Problem337.weightedCorrelation_norm_le_one line 56

    Source documentation

    Correlations of unit-bounded variables under nonnegative weights of mass one are themselves bounded by one.

    theorem weightedCorrelation_norm_le_one {Ω I : Type*} [Fintype Ω]
        (w : Ω → ℝ) (z : Ω → I → ℂ) (hw : ∀ ω, 0 ≤ w ω)
        (hmass : ∑ ω, w ω = 1) (hz : ∀ ω i, ‖z ω i‖ ≤ 1) (i j : I) :
        ‖weightedCorrelation w z i j‖ ≤ 1 := by
    …

    Mentions: OAI.Problem337.weightedCorrelation

  6. theorem OAI.Problem337.boolean_second_moment_bound line 75

    Source documentation

    The random-products second-moment reduction: after the separated-pair correlation estimate, the only loss is the exponentially small Hamming exception.

    theorem boolean_second_moment_bound {Ω : Type*} [Fintype Ω] (m : ℕ)
        (w : Ω → ℝ) (z : Ω → (Fin m → Bool) → ℂ)
        (hw : ∀ ω, 0 ≤ w ω) (hmass : ∑ ω, w ω = 1)
        (hz : ∀ ω i, ‖z ω i‖ ≤ 1) (δ : ℝ) (hδ : 0 ≤ δ)
        (hgood : ∀ i j, (m : ℝ) / 4 ≤ (hammingDist i j : ℝ) →
          ‖weightedCorrelation w z i j‖ ≤ δ) :
        (∑ ω, w ω * ‖∑ i, z ω i‖ ^ 2) / (4 : ℝ) ^ m ≤
          Real.exp (-(m : ℝ) / 16) + δ := by
    …

    Mentions: OAI.Problem337.weightedCorrelation

  7. theorem OAI.Problem337.weighted_exposure_bound line 103

    Source documentation

    Averaging over exposed variables costs at most the mass of bad exposures plus the bound that holds uniformly on the good exposures.

    theorem weighted_exposure_bound {Ω : Type*} [Fintype Ω]
        (w : Ω → ℝ) (f : Ω → ℂ) (B : Finset Ω) (a b : ℝ)
        (hw : ∀ ω, 0 ≤ w ω) (hmass : ∑ ω, w ω = 1)
        (ha : 0 ≤ a) (hbad : ∑ ω ∈ B, w ω ≤ b)
        (hall : ∀ ω, ‖f ω‖ ≤ 1) (hgood : ∀ ω, ω ∉ B → ‖f ω‖ ≤ a) :
        ‖∑ ω, (w ω : ℂ) * f ω‖ ≤ a + b := by
    …
  8. theorem OAI.Problem337.random_second_moment_absorption line 127

    Source documentation

    The numerical absorption at the end of the random-products argument.

    theorem random_second_moment_absorption (m V : ℝ) (hlarge : 50 ≤ min m V) :
        Real.exp (-m / 16) + Real.exp (-V / 20) + Real.exp (-V / 5) ≤
          Real.exp (-min m V / 100) := by
    …
  9. theorem OAI.Problem337.boolean_random_second_moment_bound line 157

    Source documentation

    Final second-moment estimate once the good/bad pair-exposure estimates are supplied. All constants agree with the random-products lemma.

    theorem boolean_random_second_moment_bound {Ω : Type*} [Fintype Ω] (m : ℕ) (V : ℝ)
        (w : Ω → ℝ) (z : Ω → (Fin m → Bool) → ℂ)
        (hw : ∀ ω, 0 ≤ w ω) (hmass : ∑ ω, w ω = 1)
        (hz : ∀ ω i, ‖z ω i‖ ≤ 1) (hlarge : 50 ≤ min (m : ℝ) V)
        (hgood : ∀ i j, (m : ℝ) / 4 ≤ (hammingDist i j : ℝ) →
          ‖weightedCorrelation w z i j‖ ≤ Real.exp (-V / 20) + Real.exp (-V / 5)) :
        (∑ ω, w ω * ‖∑ i, z ω i‖ ^ 2) / (4 : ℝ) ^ m ≤
          Real.exp (-min (m : ℝ) V / 100) := by
    …

    Mentions: OAI.Problem337.weightedCorrelation