openAI-math-dataset / lean /OAI /Computability /DepthThree /HardSliceExistence.lean
roseking's picture
Add lean source and binary files (part 60)
9dbc64e verified
Raw History Blame Contribute Delete
4.43 kB
import OAI.Computability.DepthThree.HardSliceProbability
import OAI.Computability.DepthThree.HardSliceSelection
namespace OAI
noncomputable section
open scoped BigOperators Classical
namespace DepthThreeLowerBound
theorem actual_hash_coefficient_window_failure_le {d r h b : ℕ}
(p : Fin r → Bool) [Fact (Irreducible (BinaryAlgebra.inputPolynomialBits p))]
(hr : 0 < r) (hb : 1 ≤ b) (hh : 0 < h) (q : ℝ)
(σ : Restriction (Fin d)) (hσ : restrictionWindow q σ)
(hbound :
(2 : ℝ) ^ Fintype.card (Live σ) / (2 : ℝ) ^ r +
sparseMomentBound (Fintype.card (Live σ)) b
(hardSliceSparseConstant b hb * Fintype.card (Live σ)) h
((2 : ℝ) ^ (-(Fintype.card (Live σ) : ℝ) / 4)) ≤
2 * (2 : ℝ) ^ (-(q * d) / 16)) :
finiteProb (fun ω : (Fin (d + r - 1) → Bool) ×
(Fin (2 * h) → BinaryAlgebra.BitQuotient p) =>
¬ corr b (fun z => sign (hashAcceptance p (2 * h) ω.1 ω.2 (fill σ z))) ≤
(2 : ℝ) ^ (-(q * d) / 16)) ≤
2 * (2 : ℝ) ^ (-(q * d) / 16) := by
have hlow : (q * d) / 2 ≤ (Fintype.card (Live σ) : ℝ) := by
simpa only [restrictionWindow, Fintype.card_fin] using hσ.1
have hthreshold : (2 : ℝ) ^ (-(Fintype.card (Live σ) : ℝ) / 8) ≤
(2 : ℝ) ^ (-(q * d) / 16) :=
Real.rpow_le_rpow_of_exponent_le (by norm_num) (by linarith)
apply le_trans _ ((actual_hash_coefficient_failure_le p hr σ b hb h hh).trans hbound)
apply finiteProb_mono
intro ω hω hsmall
exact hω (hsmall.trans hthreshold)
theorem exists_good_hash_coefficients {d r h b : ℕ}
(p : Fin r → Bool) [Fact (Irreducible (BinaryAlgebra.inputPolynomialBits p))]
(hr : 0 < r) (hb : 1 ≤ b) (hh : 0 < h) (q : ℝ)
(hq : 0 ≤ q) (hq1 : q ≤ 1)
(hbound : ∀ m : ℕ, (q * d) / 2 ≤ (m : ℝ) → (m : ℝ) ≤ 2 * (q * d) →
(2 : ℝ) ^ m / (2 : ℝ) ^ r +
sparseMomentBound m b (hardSliceSparseConstant b hb * m) h
((2 : ℝ) ^ (-(m : ℝ) / 4)) ≤ 2 * (2 : ℝ) ^ (-(q * d) / 16)) :
∃ (u : Fin (d + r - 1) → Bool) (β : Fin (2 * h) → BinaryAlgebra.BitQuotient p),
restrictionAvg q (fun σ => if ¬ goodRestriction q
((2 : ℝ) ^ (-(q * d) / 16)) b (hashAcceptance p (2 * h) u β) σ
then 1 else 0) ≤ 4 * (2 : ℝ) ^ (-(q * d) / 16) := by
have hfixed (σ : Restriction (Fin d)) (hσ : restrictionWindow q σ) :
finiteProb (fun ω : (Fin (d + r - 1) → Bool) ×
(Fin (2 * h) → BinaryAlgebra.BitQuotient p) =>
¬ corr b (fun z => sign (hashAcceptance p (2 * h) ω.1 ω.2 (fill σ z))) ≤
(2 : ℝ) ^ (-(q * d) / 16)) ≤
2 * (2 : ℝ) ^ (-(q * d) / 16) := by
apply actual_hash_coefficient_window_failure_le p hr hb hh q σ hσ
apply hbound
· simpa only [restrictionWindow, Fintype.card_fin] using hσ.1
· simpa only [restrictionWindow, Fintype.card_fin] using hσ.2
obtain ⟨ω, hω⟩ := exists_good_parameter hq hq1 b
(fun ω : (Fin (d + r - 1) → Bool) × (Fin (2 * h) → BinaryAlgebra.BitQuotient p) =>
hashAcceptance p (2 * h) ω.1 ω.2)
(by simpa only [Fintype.card_fin] using hfixed)
exact ⟨ω.1, ω.2, by simpa only [Fintype.card_fin] using hω⟩
def coefficientInputBits {r t : ℕ} (p : Fin r → Bool)
(β : Fin t → BinaryAlgebra.BitQuotient p) : Fin t → Fin r → Bool :=
fun j i => BinaryAlgebra.fieldBit
(BinaryAlgebra.coordinates (fun i => BinaryAlgebra.bitValue (p i)) (β j) i)
@[simp] theorem encode_coefficientInputBits {r t : ℕ} (p : Fin r → Bool)
(β : Fin t → BinaryAlgebra.BitQuotient p) (j : Fin t) :
BinaryAlgebra.encode (fun i => BinaryAlgebra.bitValue (p i))
(fun i => BinaryAlgebra.bitValue (coefficientInputBits p β j i)) = β j := by
simp only [coefficientInputBits, BinaryAlgebra.bitValue_fieldBit]
exact BinaryAlgebra.encode_coordinates _ _
theorem hashAcceptance_coefficientInputBits {d r t : ℕ} (p : Fin r → Bool)
[Fact (Irreducible (BinaryAlgebra.inputPolynomialBits p))]
(u : Fin (d + r - 1) → Bool) (β : Fin t → BinaryAlgebra.BitQuotient p) :
hashAcceptance p t u
(fun j => BinaryAlgebra.encode (fun i => BinaryAlgebra.bitValue (p i))
(fun i => BinaryAlgebra.bitValue (coefficientInputBits p β j i))) =
hashAcceptance p t u β := by
simp only [encode_coefficientInputBits]
end DepthThreeLowerBound
end
end OAI