OAI/Computability/MaxCut/Games/RealTarget.lean

GitHub source file

21 declarations, 0 contain sorry, 0 declared as axioms

The arbitrary-real-error target and its exact passage from encoded rational gap reductions. The same formula map and actual machine certificate are reused; choosing smaller error tolerances does not change the computation.

  1. structure OAI.MaxCutGames.Integration.RealTarget.PolynomialGapReduction line 18

    Source documentation

    A fixed-alphabet gap reduction from 3SAT, including its actual finite-machine polynomial runtime, with the two approximation thresholds expressed over ℝ.

    structure PolynomialGapReduction (ε δ : ℝ) where
    …
  2. def OAI.MaxCutGames.Integration.RealTarget.ofRational line 34

    Source documentation

    A genuine rational-error certificate supplies every weaker real threshold. The input and output encodings and finite transition program remain identical.

    def ofRational {e d : RationalError} {ε δ : ℝ}
        (p : Complexity.PolynomialGapReduction e d)
        (hε : (GapSemantics.errorValue e : ℝ) ≤ ε)
        (hδ : (GapSemantics.errorValue d : ℝ) ≤ δ) : PolynomialGapReduction ε δ where
    …

    Mentions: OAI.MaxCutGames.Integration.GapSemantics.errorValue, OAI.MaxCutGames.Integration.RealTarget.PolynomialGapReduction

  3. theorem OAI.MaxCutGames.Integration.RealTarget.for_all_real_of_dyadic line 53

    Source documentation

    It suffices to construct the full machine-backed reduction for dyadic errors. This theorem does not assert that those certificates have been built.

    theorem for_all_real_of_dyadic
        (certificates : ∀ m n : Nat, Nonempty
          (Complexity.PolynomialGapReduction
            (DyadicErrors.dyadicError m) (DyadicErrors.dyadicError n)))
        (ε δ : ℝ) (hε : 0 < ε) (hδ : 0 < δ) :
        Nonempty (PolynomialGapReduction ε δ) := by
    …

    Mentions: OAI.MaxCutGames.Integration.DyadicErrors.dyadicError, OAI.MaxCutGames.Integration.RealTarget.PolynomialGapReduction

  4. def OAI.MaxCutGames.Integration.InstanceValue.maxSatisfied line 80

    Source documentation

    The maximum number of satisfied occurrences among all vertex labelings.

    def maxSatisfied {q : Nat} (g : Instance q) : Nat :=
    …
  5. def OAI.MaxCutGames.Integration.InstanceValue.value line 86

    Source documentation

    The usual instance value: the largest satisfied fraction of its stored constraint list. The denominator is positive by the Instance invariant.

    noncomputable def value {q : Nat} (g : Instance q) : ℝ :=
    …
  6. theorem OAI.MaxCutGames.Integration.InstanceValue.countSatisfied_le_maxSatisfied line 89
  7. theorem OAI.MaxCutGames.Integration.InstanceValue.maxSatisfied_le_length line 97
  8. theorem OAI.MaxCutGames.Integration.InstanceValue.maxSatisfied_attained line 103
  9. theorem OAI.MaxCutGames.Integration.InstanceValue.value_attained line 112
  10. theorem OAI.MaxCutGames.Integration.InstanceValue.labeling_rate_le_value line 118
  11. theorem OAI.MaxCutGames.Integration.InstanceValue.satisfactionRate_le_value line 126

    Source documentation

    This value uses the same rate already defined for the encoded gap target.

    theorem satisfactionRate_le_value {q : Nat} (g : Instance q)
        (labeling : Fin g.vertices → Fin q) :
        (GapSemantics.satisfactionRate g labeling : ℝ) ≤ value g := by
    …

    Mentions: OAI.MaxCutGames.Integration.GapSemantics.satisfactionRate, OAI.MaxCutGames.Integration.InstanceValue.value

  12. theorem OAI.MaxCutGames.Integration.InstanceValue.value_nonneg line 131
  13. theorem OAI.MaxCutGames.Integration.InstanceValue.value_le_one line 134
  14. theorem OAI.MaxCutGames.Integration.InstanceValue.value_mem_unitInterval line 140
  15. theorem OAI.MaxCutGames.Integration.InstanceValue.exists_rate_ge_iff line 144

    Source documentation

    Existential real completeness is precisely a lower bound on instance value.

    theorem exists_rate_ge_iff {q : Nat} (g : Instance q) (hq : 0 < q) (a : ℝ) :
        (∃ labeling : Fin g.vertices → Fin q,
          a ≤ (countSatisfied labeling g.constraints : ℝ) / g.constraints.length) ↔
          a ≤ value g := by
    …

    Mentions: OAI.MaxCutGames.Integration.InstanceValue.value

  16. theorem OAI.MaxCutGames.Integration.InstanceValue.forall_rate_le_iff line 156

    Source documentation

    Universal real soundness is precisely an upper bound on instance value.

    theorem forall_rate_le_iff {q : Nat} (g : Instance q) (hq : 0 < q) (b : ℝ) :
        (∀ labeling : Fin g.vertices → Fin q,
          (countSatisfied labeling g.constraints : ℝ) / g.constraints.length ≤ b) ↔
          value g ≤ b := by
    …

    Mentions: OAI.MaxCutGames.Integration.InstanceValue.value

  17. theorem OAI.MaxCutGames.Integration.InstanceValue.completeAt_iff_value line 167
  18. theorem OAI.MaxCutGames.Integration.InstanceValue.soundAt_iff_value line 172
  19. theorem OAI.MaxCutGames.Integration.InstanceValue.realTarget_completeness line 178

    Source documentation

    The unchanged real target certificate gives literal val(U) ≥ 1 - ε.

    theorem realTarget_completeness {ε δ : ℝ}
        (p : RealTarget.PolynomialGapReduction ε δ) (formula : Formula)
        (hformula : formula.Satisfiable) :
        1 - ε ≤ value (p.reduce formula) := by
    …

    Mentions: OAI.MaxCutGames.Integration.RealTarget.PolynomialGapReduction, OAI.MaxCutGames.Integration.InstanceValue.value

  20. theorem OAI.MaxCutGames.Integration.InstanceValue.realTarget_soundness line 187

    Source documentation

    The unchanged real target certificate gives literal val(U) ≤ δ.

    theorem realTarget_soundness {ε δ : ℝ}
        (p : RealTarget.PolynomialGapReduction ε δ) (formula : Formula)
        (hformula : ¬ formula.Satisfiable) :
        value (p.reduce formula) ≤ δ := by
    …

    Mentions: OAI.MaxCutGames.Integration.RealTarget.PolynomialGapReduction, OAI.MaxCutGames.Integration.InstanceValue.value

  21. theorem OAI.MaxCutGames.Integration.InstanceValue.realTarget_value_guarantees line 195