OAI/Computability/UniqueGames/Repetition/AnalyticCheegerLemmas.lean

GitHub source file

24 declarations, 0 contain sorry, 0 declared as axioms

Finite threshold-overlap estimates used in the Dinur--Steurer argument. These statements are numerical inequalities, not parallel-repetition premises. The Boolean gate keeps the constraint predicate inside the overlap energy. Weights need not be normalized: a finite auxiliary measure is permitted.

  1. def OAI.UniqueGamesTheorem.Repetition.Analytic.quadraticMass line 29
  2. def OAI.UniqueGamesTheorem.Repetition.Analytic.gatedCorrelation line 32
  3. def OAI.UniqueGamesTheorem.Repetition.Analytic.gatedMinimum line 35
  4. def OAI.UniqueGamesTheorem.Repetition.Analytic.gatedMaximum line 38
  5. theorem OAI.UniqueGamesTheorem.Repetition.Analytic.quadraticMass_nonnegative line 41
  6. theorem OAI.UniqueGamesTheorem.Repetition.Analytic.gatedCorrelation_nonnegative line 48
  7. theorem OAI.UniqueGamesTheorem.Repetition.Analytic.gatedCorrelation_le_mass line 57
  8. theorem OAI.UniqueGamesTheorem.Repetition.Analytic.gatedMinimum_nonnegative line 68
  9. theorem OAI.UniqueGamesTheorem.Repetition.Analytic.gatedMinimum_le_mass line 76
  10. theorem OAI.UniqueGamesTheorem.Repetition.Analytic.gatedMinimum_add_maximum_le line 89
  11. theorem OAI.UniqueGamesTheorem.Repetition.Analytic.gatedCorrelation_sq_le_min_mul_max line 108

    Source documentation

    A gated finite Cauchy--Schwarz estimate. This is the algebraic heart of the threshold-rounding inequality, with no independence assumption.

    theorem gatedCorrelation_sq_le_min_mul_max (w f g : ι → ℝ) (gate : ι → Bool)
        (hw : ∀ i, 0 ≤ w i) :
        gatedCorrelation w f g gate ^ 2 ≤
          gatedMinimum w f g gate * gatedMaximum w f g gate := by
    …

    Mentions: OAI.UniqueGamesTheorem.Repetition.Analytic.gatedCorrelation, OAI.UniqueGamesTheorem.Repetition.Analytic.gatedMinimum, OAI.UniqueGamesTheorem.Repetition.Analytic.gatedMaximum

  12. theorem OAI.UniqueGamesTheorem.Repetition.Analytic.gatedCorrelation_sq_le_min_quadratic line 133
  13. theorem OAI.UniqueGamesTheorem.Repetition.Analytic.gatedMinimum_cheeger line 145

    Source documentation

    Sharp homogeneous Cheeger form. The expression under the square root is nonnegative by the preceding proved inequality.

    theorem gatedMinimum_cheeger (w f g : ι → ℝ) (gate : ι → Bool)
        (hw : ∀ i, 0 ≤ w i) :
        quadraticMass w f g -
            Real.sqrt (quadraticMass w f g ^ 2 - gatedCorrelation w f g gate ^ 2) ≤
          gatedMinimum w f g gate := by
    …

    Mentions: OAI.UniqueGamesTheorem.Repetition.Analytic.quadraticMass, OAI.UniqueGamesTheorem.Repetition.Analytic.gatedCorrelation, OAI.UniqueGamesTheorem.Repetition.Analytic.gatedMinimum

  14. theorem OAI.UniqueGamesTheorem.Repetition.Analytic.gatedMinimum_near_one line 159

    Source documentation

    The near-perfect form suffices for alphabet-independent repetition. The total squared mass may be at most one, rather than exactly one.

    theorem gatedMinimum_near_one (w f g : ι → ℝ) (gate : ι → Bool)
        (hw : ∀ i, 0 ≤ w i) {η : ℝ} (hη : 0 ≤ η) (hη₁ : η ≤ 1)
        (hmass : quadraticMass w f g ≤ 1)
        (hcorrelation : 1 - η ≤ gatedCorrelation w f g gate) :
        1 - Real.sqrt (2 * η) ≤ gatedMinimum w f g gate := by
    …

    Mentions: OAI.UniqueGamesTheorem.Repetition.Analytic.quadraticMass, OAI.UniqueGamesTheorem.Repetition.Analytic.gatedCorrelation, OAI.UniqueGamesTheorem.Repetition.Analytic.gatedMinimum

  15. theorem OAI.UniqueGamesTheorem.Repetition.Analytic.gatedMinimum_unit_mass line 180

    Source documentation

    Unit-mass form for the joint question-pair and auxiliary-coordinate law.

    theorem gatedMinimum_unit_mass (w f g : ι → ℝ) (gate : ι → Bool)
        (hw : ∀ i, 0 ≤ w i) (hf : ∀ i, 0 ≤ f i) (hg : ∀ i, 0 ≤ g i)
        (hmass : quadraticMass w f g = 1) :
        1 - Real.sqrt (2 * (1 - gatedCorrelation w f g gate)) ≤
          gatedMinimum w f g gate := by
    …

    Mentions: OAI.UniqueGamesTheorem.Repetition.Analytic.quadraticMass, OAI.UniqueGamesTheorem.Repetition.Analytic.gatedCorrelation, OAI.UniqueGamesTheorem.Repetition.Analytic.gatedMinimum

  16. def OAI.UniqueGamesTheorem.Repetition.dsRate line 212

    Source documentation

    The scalar rate appearing in the alphabet-independent repetition estimate.

    noncomputable def dsRate (gap : ℝ) : ℝ := 1 - gap ^ 2 / 16
  17. theorem OAI.UniqueGamesTheorem.Repetition.dsRate_nonneg line 214
  18. theorem OAI.UniqueGamesTheorem.Repetition.dsRate_lt_one line 219
  19. theorem OAI.UniqueGamesTheorem.Repetition.exists_positive_power_lt line 225

    Source documentation

    A scalar contraction admits a strictly positive exponent for every positive target error. This lemma makes no assertion about any game value.

    theorem exists_positive_power_lt {rate error : ℝ}
        (hrate₀ : 0 ≤ rate) (hrate₁ : rate < 1) (herror : 0 < error) :
        ∃ n : ℕ, 1 ≤ n ∧ rate ^ n < error := by
    …
  20. theorem OAI.UniqueGamesTheorem.Repetition.exists_dsRate_pow_le line 237

    Source documentation

    Select the repetition count before choosing any alphabet.

    theorem exists_dsRate_pow_le {gap error : ℝ}
        (hgap₀ : 0 < gap) (hgap₁ : gap < 1) (herror : 0 < error) :
        ∃ n : ℕ, 1 ≤ n ∧ dsRate gap ^ n ≤ error := by
    …

    Mentions: OAI.UniqueGamesTheorem.Repetition.dsRate

  21. theorem OAI.UniqueGamesTheorem.Repetition.dsRate_one_div_eight_hundred line 244
  22. theorem OAI.UniqueGamesTheorem.Repetition.dsRate_one_div_fifteen line 248
  23. theorem OAI.UniqueGamesTheorem.Repetition.exists_final_repetition_count line 254

    Source documentation

    The final soundness exponent depends only on error, even when the later construction chooses its alphabet after this exponent.

    theorem exists_final_repetition_count {error : ℝ} (herror : 0 < error) :
        ∃ n : ℕ, 1 ≤ n ∧ (1 - (1 : ℝ) / 10240000) ^ n ≤ error := by
    …
  24. theorem OAI.UniqueGamesTheorem.Repetition.exists_preliminary_repetition_count line 260

    Source documentation

    The preliminary constant-gap specialization of the same scalar rate.

    theorem exists_preliminary_repetition_count {error : ℝ} (herror : 0 < error) :
        ∃ n : ℕ, 1 ≤ n ∧ (1 - (1 : ℝ) / 3600) ^ n ≤ error := by
    …