OAI/Computability/MaxCut/Encoding/SurrogateFibers.lean
15 declarations, 0 contain sorry, 0 declared as axioms
The surrogate perturbation fixes the complementary table exactly. Splitting
the actual uniform matrix law therefore expresses surrogate acceptance as the
mean of the uniform shortcode tests on the (U,T) fibers, including zero
perturbation factors. This supplies the sampling identification needed before
applying the inverse theorem to the many useful fibers.
- def
OAI.MaxCutGames.Decoder.SurrogateFibers.shortcodeAcceptanceline 35Source documentation
Uniform shortcode sampling; neither nonzero-factor conditioning nor rank-one-matrix-uniform sampling is substituted for its two independent factors.
def shortcodeAcceptance (f : (E →ₗ[F2] K) → C) : ℝ := by …
- theorem
OAI.MaxCutGames.Decoder.SurrogateFibers.product_perturbationline 41 - theorem
OAI.MaxCutGames.Decoder.SurrogateFibers.surrogate_acceptance_eq_fiber_averageline 47 - theorem
OAI.MaxCutGames.Decoder.ActualSpectral.expect_prodline 92 - theorem
OAI.MaxCutGames.Decoder.ActualSpectral.expect_eq_uniform_sumline 97 - theorem
OAI.MaxCutGames.Decoder.ActualSpectral.eigenvalue_eq_kernelProbabilityline 104Source documentation
The finite noise law's multiplier is exactly the rational kernel event of the same gadget, cast to the real numbers.
theorem eigenvalue_eq_kernelProbability {s d : Nat} (g : SplitGadget s d) {P : Type} [AddCommGroup P] [Module F2 P] (S : Ambient s d →ₗ[F2] P) : linearNoiseEigenvalue (samplerLaw g.noise) S = (Integration.SplitGadget.kernelProbability g S : ℝ) := by …Mentions: OAI.MaxCutGames.Integration.SplitGadget.kernelProbability
- theorem
OAI.MaxCutGames.Decoder.ActualSpectral.acceptance_nestedline 114Source documentation
The four independent samples of the actual single-orbit test.
theorem acceptance_nested (S : Source) (k : Nat) {s d : Nat} (g : SplitGadget s d) (label : Fin (vertexCount S k s d) → Fin (2 ^ s)) : (acceptanceProbability S k g label : ℝ) = 𝔼 U : Question S k, 𝔼 X : Map k s d, 𝔼 i : g.NoiseIndex, 𝔼 l : Dual k, if unfolded S k s d label (U, X) = unfolded S k s d label (U, X + l.smulRight (g.noise i)) then (1 : ℝ) else 0 := by … - theorem
OAI.MaxCutGames.Decoder.ActualSpectral.acceptance_eq_equality_meanline 137Source documentation
No auxiliary side labeling is introduced: the same restored table occurs on both sides of the actual equality test.
theorem acceptance_eq_equality_mean (S : Source) (k : Nat) {s d : Nat} (g : SplitGadget s d) (label : Fin (vertexCount S k s d) → Fin (2 ^ s)) : (acceptanceProbability S k g label : ℝ) = 𝔼 U : Question S k, equalityAcceptance (samplerLaw g.noise) (fun X : Map k s d => unfolded S k s d label (U, X)) := by … - def
OAI.MaxCutGames.Decoder.ActualSpectral.fiberAcceptanceline 156 - theorem
OAI.MaxCutGames.Decoder.ActualSpectral.fiberAcceptance_le_oneline 162 - theorem
OAI.MaxCutGames.Decoder.ActualSpectral.weighted_fiber_mean_lowerline 175 - theorem
OAI.MaxCutGames.Decoder.ActualSpectral.actual_fiber_mean_lowerline 197 - def
OAI.MaxCutGames.Decoder.ActualSpectral.etaline 218 - theorem
OAI.MaxCutGames.Decoder.ActualSpectral.eta_posline 220 - theorem
OAI.MaxCutGames.Decoder.ActualSpectral.actual_good_fiber_massline 224Source documentation
At least
10 ηof the actual(U,T)pairs have shortcode acceptance at least4 η, exactly the input needed by row erasure.theorem actual_good_fiber_mass (S : Source) (k : Nat) {s d : Nat} (g : SplitGadget s d) (label : Fin (vertexCount S k s d) → Fin (2 ^ s)) (r : Nat) (hkernel : ∀ L : Ambient s d →ₗ[F2] ActualHomogeneous.E k, r ≤ Module.finrank F2 (L.comp (alphabetEmbedding s d)).range → Integration.SplitGadget.kernelProbability g L ≤ 7 / 8) (haccept : (99 : ℚ) / 100 ≤ acceptanceProbability S k g label) : 10 * eta r ≤ 𝔼 U : Question S k, 𝔼 T : ActualHomogeneous.E k →ₗ[F2] Vector d, if 4 * eta r ≤ fiberAcceptance S k s d label U T then (1 : ℝ) else 0 := by …Mentions: OAI.MaxCutGames.Integration.SplitGadget.kernelProbability, OAI.MaxCutGames.Decoder.ActualSpectral.eta, OAI.MaxCutGames.Decoder.ActualSpectral.fiberAcceptance