OAI/Computability/MaxCut/Games/FiniteNoise.lean
7 declarations, 0 contain sorry, 0 declared as axioms
Realizing the existing Håstad noise law with a finite uniform tape
A tape contains one element of Fin D per coordinate. The corresponding noise
bit is true exactly when that element has value zero. For D > 0, its law is
exactly Foundations.Hastad.noiseWeight ((D : ℝ)⁻¹).
The construction also permits an empty coordinate type and D = 1.
- def
OAI.MaxCutGames.Reduction.FiniteNoise.realizedBitline 21 - def
OAI.MaxCutGames.Reduction.FiniteNoise.realizedNoiseline 23 - theorem
OAI.MaxCutGames.Reduction.FiniteNoise.realizedBit_eq_true_iffline 26 - theorem
OAI.MaxCutGames.Reduction.FiniteNoise.coordinate_fiber_sumline 31Source documentation
The two fibers of a coordinate have their actual normalized finite counts.
theorem coordinate_fiber_sum {D : Nat} (positive : 0 < D) (b : Bool) : (∑ z : Fin D, if realizedBit z = b then (D : ℝ)⁻¹ else 0) = if b then (D : ℝ)⁻¹ else 1 - (D : ℝ)⁻¹ := by … - theorem
OAI.MaxCutGames.Reduction.FiniteNoise.noiseWeight_eq_fiber_sumline 56Source documentation
Product distributivity computes the mass of every complete Boolean noise vector.
theorem noiseWeight_eq_fiber_sum {D : Nat} (positive : 0 < D) (μ : A → Bool) : noiseWeight ((D : ℝ)⁻¹) μ = ∑ z : A → Fin D, ∏ a, if realizedBit (z a) = μ a then (D : ℝ)⁻¹ else 0 := by … - theorem
OAI.MaxCutGames.Reduction.FiniteNoise.fiber_product_eq_indicatorline 68 - theorem
OAI.MaxCutGames.Reduction.FiniteNoise.expect_realizedNoiseline 82Source documentation
Exact pushforward law for every observable, using the existing noise weight.
theorem expect_realizedNoise {D : Nat} (positive : 0 < D) (H : (A → Bool) → ℝ) : (𝔼 z : A → Fin D, H (realizedNoise z)) = ∑ μ : A → Bool, noiseWeight ((D : ℝ)⁻¹) μ * H μ := by …Mentions: OAI.MaxCutGames.Reduction.FiniteNoise.realizedNoise