The finite few-amortized-free-bits test
Lax253009.FAFTest · concepts/Lax253009/FAFTest.lean · lax-253009
No public endorsements yet.
Loading review…
Sign in with ORCIDNatural Language Statement
Theorem
Fix a small variable set and a distribution of larger sets. Each larger set carries a purported long code and a constraint predicate. The verifier reads random functions on the small set, then checks independently chosen larger tables using the CNA test with side conditions enforcing both the constraints and agreement with the reference answers.
The rejection bound combines CNA decoding errors and the many-table agreement estimate. It is uniform in the reference table, which need not be a genuine long code. The projections and constraints are explicit finite data; efficient construction from an NP input is a separate obligation.
Concept map
Evidence
Lean source view on GitHub
| 1 | import Lax253009.LongCode |
| 2 | import Lax253009.CNASoundness |
| 3 | import Lax253009.ManyTableConsistency |
| 4 | |
| 5 | /-! |
| 6 | --- |
| 7 | title: The finite few-amortized-free-bits test |
| 8 | type: theorem |
| 9 | --- |
| 10 | Fix a small variable set and a distribution of larger sets. Each larger |
| 11 | set carries a purported long code and a constraint predicate. The verifier |
| 12 | reads random functions on the small set, then checks independently |
| 13 | chosen larger tables using the CNA test with side conditions enforcing both |
| 14 | the constraints and agreement with the reference answers. |
| 15 | |
| 16 | The rejection bound combines CNA decoding errors and the many-table |
| 17 | agreement estimate. It is uniform in the reference table, which need not |
| 18 | be a genuine long code. The projections and constraints are explicit finite |
| 19 | data; efficient construction from an NP input is a separate obligation. |
| 20 | -/ |
| 21 | |
| 22 | namespace Lax253009.FAFTest |
| 23 | |
| 24 | open LongCode FiniteProbability CNASoundness |
| 25 | |
| 26 | noncomputable def condition {Ω : Type} {u w q : ℕ} |
| 27 | (ρ : Ω → Word w → Word u) (valid : Ω → Coordinate w) |
| 28 | (R : Table u) (ω : Ω) (g : Fin q → Coordinate u) : Coordinate w := |
| 29 | fun y ↦ valid ω y && decide (∀ j, g j (ρ ω y) = R (g j)) |
| 30 | |
| 31 | def Accepts {Ω : Type} {u w n s q : ℕ} |
| 32 | (ρ : Ω → Word w → Word u) (valid : Ω → Coordinate w) |
| 33 | (R : Table u) (A : Ω → Table w) |
| 34 | (ω : Fin n → Ω) (g : Fin q → Coordinate u) |
| 35 | (f : Fin n → Fin s → Coordinate w) : Prop := |
| 36 | ∀ i, AcceptsWithCondition (A (ω i)) (f i) (condition ρ valid R (ω i) g) |
| 37 | |
| 38 | axiom perfect_completeness {Ω : Type} {u w n s q : ℕ} |
| 39 | (ρ : Ω → Word w → Word u) (valid : Ω → Coordinate w) |
| 40 | (x : Word u) (y : Ω → Word w) |
| 41 | (hvalid : ∀ ω, valid ω (y ω) = true) (hproject : ∀ ω, ρ ω (y ω) = x) |
| 42 | (ω : Fin n → Ω) (g : Fin q → Coordinate u) (f : Fin n → Fin s → Coordinate w) : |
| 43 | Accepts ρ valid (evaluation x) (fun ω ↦ evaluation (y ω)) ω g f |
| 44 | |
| 45 | noncomputable def projected {Ω : Type} {u w : ℕ} |
| 46 | (ρ : Ω → Word w → Word u) (valid : Ω → Coordinate w) |
| 47 | (D : Ω → Finset (Word w)) (ω : Ω) : Finset (Word u) := |
| 48 | ((D ω).filter (fun y ↦ valid ω y = true)).image (ρ ω) |
| 49 | |
| 50 | axiom acceptance_bound {Ω : Type} [Fintype Ω] [Nonempty Ω] |
| 51 | (u w n s q : ℕ) (ρ : Ω → Word w → Word u) (valid : Ω → Coordinate w) |
| 52 | (R : Table u) (A : Ω → Table w) (D : Ω → Finset (Word w)) |
| 53 | (B k : ℕ) (p δ : ℝ) (hp : 0 ≤ p) |
| 54 | (hB : ∀ ω, (D ω).card ≤ B) |
| 55 | (hdecode : ∀ ω h, probability (BadWithCondition (s := s) (A ω) (D ω) h) ≤ δ) |
| 56 | (hpoint : ∀ x, probability (fun ω ↦ x ∈ projected ρ valid D ω) ≤ p) |
| 57 | (hsmall : (n : ℝ) * B * p ≤ 1) : |
| 58 | probability (fun z : ((Fin n → Ω) × (Fin q → Coordinate u)) × |
| 59 | (Fin n → Fin s → Coordinate w) ↦ Accepts ρ valid R A z.1.1 z.1.2 z.2) ≤ |
| 60 | n * δ + (2 : ℝ) ^ n * ((n : ℝ) * B * p) ^ (n - k) + |
| 61 | (B : ℝ) ^ n * (2 * (1 / 2 : ℝ) ^ (k + 1)) ^ q |
| 62 | |
| 63 | end Lax253009.FAFTest |
| 64 |
From Mathlib
none
Discussion
Ask a question or add context. Endorsements and structured flags are kept in the review panel above.
0 comments