OAI/Probability/TraceReconstruction/ErrorRates.lean

GitHub source file

62 declarations, 0 contain sorry, 0 declared as axioms

ErrorRates

Realization of block caps, sparse scales, sharp and smoothing budgets, tree tails, fit errors, and uniform polynomial/exponential bounds.

  1. theorem OAI.TraceReconstruction.sourceWindowSize_le line 22
  2. theorem OAI.TraceReconstruction.SourceArithmeticBounds.scheduled_trials line 30

    Source comment

    We use only positive target scales below.
    theorem SourceArithmeticBounds.scheduled_trials {C N F ε : ℝ} {s b ell k : ℕ}
        (H : SourceArithmeticBounds C N F ε s b ell k) (hN : 0 < N) (hC : 0 < C) (hF : 0 < F)
        (j : ℕ) (hj : j+1 < b) :
        N^(scaleKappa b (j+2) ε+ε)/(4800*(C^(2*b)*F)) ≤
          (sourceScheduledTrials (sourceBlockWindows N b ε) (sourceArithmeticTrials C N b ε ell k) (j+1):ℝ) := by
    …

    Mentions: OAI.TraceReconstruction.SourceArithmeticBounds, OAI.TraceReconstruction.scaleKappa, OAI.TraceReconstruction.sourceScheduledTrials, OAI.TraceReconstruction.sourceBlockWindows, OAI.TraceReconstruction.sourceArithmeticTrials

  3. theorem OAI.TraceReconstruction.SourceArithmeticBounds.binomial_caps line 63
  4. theorem OAI.TraceReconstruction.sourceSparseCount_scale line 94
  5. theorem OAI.TraceReconstruction.sourceLeafCount_succ line 109
  6. theorem OAI.TraceReconstruction.sourceLeafCount_le_length line 113
  7. theorem OAI.TraceReconstruction.sourceArithmeticLength_eq_block line 123
  8. theorem OAI.TraceReconstruction.SourceArithmeticBounds.leaf_bound line 130
  9. theorem OAI.TraceReconstruction.source_sparse_weight_bound line 152

    Source documentation

    Full sparse degree, with the excluded bottom window and the exact source gap. This is the high-degree geometric gain, before constants.

    theorem source_sparse_weight_bound (N ε d W w : ℝ) (b s : ℕ) (hN : 0 < N) (hW : 0 ≤ W)
        (hw : w^2 ≤ W*N^(-scaleA b 0 ε*d)) :
        w^2*(sourceSparseCount (sourceBlockWindows N b ε) s (b-1):ℝ) ≤
          W*((2^s:ℕ):ℝ)*N^(-(scaleA b 0 ε*d-sourceInteriorExponent b ε)) := by
    …

    Mentions: OAI.TraceReconstruction.scaleA, OAI.TraceReconstruction.sourceSparseCount, OAI.TraceReconstruction.sourceBlockWindows, OAI.TraceReconstruction.sourceInteriorExponent

  10. def OAI.TraceReconstruction.sourceBadRoleFactor line 174
  11. def OAI.TraceReconstruction.sourceForcedRoleFactor line 177
  12. theorem OAI.TraceReconstruction.source_sharp_numeric line 183

    Source documentation

    Explicit finite sharp energy bound. The two undesirable pattern classes are suppressed by eta and a, and the genuine binary forced skeleton pays its full source exponent. The high tail uses the sparse, not dense count.

    theorem source_sharp_numeric (T n : ℕ→ℕ) (s b R : ℕ) (ω η a w N ε d K W Q gap : ℝ)
        (hN : 1 ≤ N) (hK : 0 ≤ K) (hW : 0 ≤ W) (_hQ : 0 ≤ Q)
        (hleaves : (SourceLeafCount T (b+1):ℝ) ≤ N) (hwone : w^2 ≤ 1)
        (hw : w^2 ≤ W*N^(-scaleA (b+1) 0 ε*d))
        (hsmall : w^2*((2^s:ℕ):ℝ)*(sourceSparseCount T s b:ℝ) ≤ 1)
        (hhalf : w^2*(sourceSparseCount T s b:ℝ) ≤ 1/2)
        (hsparse : w^2*(sourceSparseCount T s b:ℝ) ≤ Q*N^(-gap))
        (hn : ∀j≤b,hiddenBinomialCap (n j) ω ≤ K*N^(-ε/2))
        (hg : ∀j<b,hiddenBinomialCap (sourceScheduledTrials T n (j+1)) ω ≤
          K*N^(-(scaleKappa (b+1) (j+2) ε+ε)/2)) :
        sourceSharpEnergyBudget T n s b R ω η a w ≤
          sourceBadRoleFactor b R*N^R*(η^2+a^2)+
    …

    Mentions: OAI.TraceReconstruction.SourceLeafCount, OAI.TraceReconstruction.scaleA, OAI.TraceReconstruction.sourceSparseCount, OAI.TraceReconstruction.hiddenBinomialCap, OAI.TraceReconstruction.sourceScheduledTrials, OAI.TraceReconstruction.scaleKappa, OAI.TraceReconstruction.sourceSharpEnergyBudget, OAI.TraceReconstruction.sourceBadRoleFactor

  13. def OAI.TraceReconstruction.sourceSmoothingPrefactor line 264
  14. theorem OAI.TraceReconstruction.hiddenSmoothingBound_scale line 267
  15. theorem OAI.TraceReconstruction.sourceLeafCount_positive line 306
  16. theorem OAI.TraceReconstruction.sourceFitTail_simple line 314
  17. theorem OAI.TraceReconstruction.windowCollection_card_bound line 336
  18. theorem OAI.TraceReconstruction.collectionCount_lower line 352
  19. theorem OAI.TraceReconstruction.sourceEffectiveHiddenTail_simple line 365
  20. theorem OAI.TraceReconstruction.SourceArithmeticBounds.fit_exponent line 394

    Source documentation

    The literal centering fluctuation has the source epsilon exponent. This does not treat nested window starts as independent.

    theorem SourceArithmeticBounds.fit_exponent {C N F ε : ℝ} {s b ell k j : ℕ}
        (H : SourceArithmeticBounds C N F ε s b ell k) (hC : 0<C) (hF : 0<F) (hN : 1 ≤ N) (he : 0 ≤ ε)
        (hκ : ∀v<b,3*ε ≤ scaleKappa b (v+1) ε) (hj : j+1<b) :
        N^ε/512 ≤ 2*((sourceBlockWindows N b ε j:ℝ)/8)^2 /
          (((sourceBlockWindows N b ε (j+1):ℝ)+1)*(sourceArithmeticTrials C N b ε ell k j:ℝ)) := by
    …

    Mentions: OAI.TraceReconstruction.SourceArithmeticBounds, OAI.TraceReconstruction.scaleKappa, OAI.TraceReconstruction.sourceBlockWindows, OAI.TraceReconstruction.sourceArithmeticTrials

  21. def OAI.TraceReconstruction.SourcePolyUpper line 442

    Source documentation

    A convenience for the manuscript's fixed-B polynomial losses. This predicate is only used to prove bounds on the literal finite constants.

    def SourcePolyUpper (f : ℕ→ℝ) : Prop :=
    …
  22. theorem OAI.TraceReconstruction.SourcePolyUpper.const line 445
  23. theorem OAI.TraceReconstruction.SourcePolyUpper.depth line 450
  24. theorem OAI.TraceReconstruction.SourcePolyUpper.mono line 453
  25. theorem OAI.TraceReconstruction.SourcePolyUpper.add line 458
  26. theorem OAI.TraceReconstruction.SourcePolyUpper.mul line 469
  27. theorem OAI.TraceReconstruction.SourcePolyUpper.pow line 480
  28. theorem OAI.TraceReconstruction.sourceEpsilon_inverse_poly line 487
  29. theorem OAI.TraceReconstruction.totalScoreExponent_poly line 494
  30. theorem OAI.TraceReconstruction.SourcePolyUpper.ceil line 502
  31. theorem OAI.TraceReconstruction.sourceCancellationOrder_poly line 508
  32. theorem OAI.TraceReconstruction.sourceCancellationOrder_large line 526
  33. def OAI.TraceReconstruction.SourceExpUpper line 541

    Source documentation

    Literal finite-block factors are subpower uniformly over the joint deletion regime when their logarithms grow polynomially in the depth.

    def SourceExpUpper (f : ℕ→ℝ→ℝ) : Prop :=
    …
  34. theorem OAI.TraceReconstruction.SourcePolyUpper.expUpper line 545
  35. theorem OAI.TraceReconstruction.SourceExpUpper.const line 555
  36. theorem OAI.TraceReconstruction.SourceExpUpper.mono line 558
  37. theorem OAI.TraceReconstruction.SourceExpUpper.mul line 563
  38. theorem OAI.TraceReconstruction.SourceExpUpper.pow_variably line 582
  39. theorem OAI.TraceReconstruction.SourceExpUpper.pow line 596
  40. theorem OAI.TraceReconstruction.sourceOverlapConstant_expUpper line 601
  41. theorem OAI.TraceReconstruction.initialOverlapLower_inverse_expUpper line 608
  42. theorem OAI.TraceReconstruction.SourceExpUpper.uniform line 622

    Source documentation

    All indices are inside the same physical n,q instance.

    theorem SourceExpUpper.uniform (f : ℕ→ℝ→ℝ) (hf : SourceExpUpper f)
        (B : ℕ) (ζ δ : ℝ) (hB : 2 ≤ B) (hζ : ζ∈Set.Ioo 0 1) (hδ : 0<δ)
        {n q : ℕ→ℝ} (hn : Tendsto n atTop atTop) (hq : ∀t,q t∈Set.Ioc 0 (1/2:ℝ))
        (hH : Tendsto (fun t=>(q t)^3*Real.log (n t)) atTop atTop) :
        ∀ᶠt in atTop,∀i,f (sourceDepth B ζ ((q t)^3*Real.log (n t))) (q t) ≤
          (sourceTargetScale B (sourceDepth B ζ ((q t)^3*Real.log (n t))) (n t) i)^(
            δ*sourceEpsilon B (sourceDepth B ζ ((q t)^3*Real.log (n t)))) := by
    …

    Mentions: OAI.TraceReconstruction.SourceExpUpper, OAI.TraceReconstruction.sourceDepth, OAI.TraceReconstruction.sourceTargetScale, OAI.TraceReconstruction.sourceEpsilon

  43. def OAI.TraceReconstruction.sourceMomentEnvelope line 645
  44. def OAI.TraceReconstruction.sourceCapEnvelope line 648
  45. def OAI.TraceReconstruction.sourceWeightEnvelope line 651
  46. theorem OAI.TraceReconstruction.sourceMomentEnvelope_pos line 653
  47. theorem OAI.TraceReconstruction.sourceCapEnvelope_pos line 657
  48. theorem OAI.TraceReconstruction.sourceMomentEnvelope_expUpper line 661
  49. theorem OAI.TraceReconstruction.sourceCapEnvelope_expUpper line 672
  50. theorem OAI.TraceReconstruction.sourceWeightEnvelope_expUpper line 683
  51. theorem OAI.TraceReconstruction.sourceMotif_expUpper line 688
  52. theorem OAI.TraceReconstruction.sourceSmoothingPrefactor_expUpper line 692
  53. theorem OAI.TraceReconstruction.sourceSkeletonPrefactor_expUpper line 709
  54. theorem OAI.TraceReconstruction.SourceArithmeticBounds.fit_tail line 731
  55. theorem OAI.TraceReconstruction.SourceArithmeticBounds.metadata_tail line 742
  56. theorem OAI.TraceReconstruction.source_clipping_moment_power line 764

    Source documentation

    Quantitative clipping without any default-limit increase or limiting interchange: the power is the finite even moment actually used by the oracle.

    theorem source_clipping_moment_power (T : ℕ→ℕ) (s b k : ℕ) (N gap L : ℝ)
        (hN : 0<N)
        (hhalf : ((6*(2*k:ℕ):ℝ)*L)^2*sourceSparseCount T s b ≤ 1/2)
        (hscale : ((6*(2*k:ℕ):ℝ)*L)^2*sourceSparseCount T s b ≤ N^(-gap/2)/8) :
        Real.sqrt (6*(4*sourceCenteredBudget T s b k L)^k) ≤ 3*N^(-gap*(k:ℝ)/4) := by
    …

    Mentions: OAI.TraceReconstruction.sourceSparseCount, OAI.TraceReconstruction.sourceCenteredBudget

  57. theorem OAI.TraceReconstruction.SourceExpUpper.add line 799
  58. def OAI.TraceReconstruction.sourceScaleInverseEnvelope line 826
  59. theorem OAI.TraceReconstruction.sourceScaleInverseEnvelope_expUpper line 828
  60. theorem OAI.TraceReconstruction.source_log_scale_lower line 837
  61. theorem OAI.TraceReconstruction.SourceExpUpper.uniform_log line 847
  62. theorem OAI.TraceReconstruction.SourceExpUpper.uniform_tail line 895

    Source documentation

    Uniform exponential domination at every working scale, with the full depth and the varying known deletion probability retained.

    theorem SourceExpUpper.uniform_tail {f g h : ℕ→ℝ→ℝ}
        (hf : SourceExpUpper f) (hg : SourceExpUpper g) (hh : SourceExpUpper h)
        (hf0 : ∀D q,q∈Set.Ioc 0 (1/2:ℝ) → 0<f D q)
        (_hg0 : ∀D q,q∈Set.Ioc 0 (1/2:ℝ) → 0≤g D q)
        (hh0 : ∀D q,q∈Set.Ioc 0 (1/2:ℝ) → 0≤h D q)
        (B : ℕ) (ζ : ℝ) (hB : 2≤B) (hζ : ζ∈Set.Ioo 0 1)
        {n q : ℕ→ℝ} (hn : Tendsto n atTop atTop) (hq : ∀t,q t∈Set.Ioc 0 (1/2:ℝ))
        (hH : Tendsto (fun t=>(q t)^3*Real.log (n t)) atTop atTop) :
        ∀ᶠt in atTop,∀i,
          g (sourceDepth B ζ ((q t)^3*Real.log (n t))) (q t)*
            Real.exp (-(sourceTargetScale B (sourceDepth B ζ ((q t)^3*Real.log (n t))) (n t) i)^
              (sourceEpsilon B (sourceDepth B ζ ((q t)^3*Real.log (n t))))/
    …

    Mentions: OAI.TraceReconstruction.SourceExpUpper, OAI.TraceReconstruction.sourceDepth, OAI.TraceReconstruction.sourceTargetScale, OAI.TraceReconstruction.sourceEpsilon