OAI/Computability/MaxCut/Games/RankAdditivity.lean

GitHub source file

107 declarations, 0 contain sorry, 0 declared as axioms

Rank-additive pairs of linear maps

The rank-additive selector in Appendix A.1 has a concrete geometric meaning: the two image spaces are disjoint and the two kernels span the domain. These lemmas apply over any field and arbitrary finite dimensions, in particular over ZMod 2 with the second map equal to Y + X.

  1. theorem OAI.MaxCutGames.Appendix.RankAdditivity.ker_add_of_disjoint line 24

    Source documentation

    Disjoint images prevent cancellation between the two map values.

    theorem ker_add_of_disjoint (f g : W →ₗ[K] V)
        (h : Disjoint (LinearMap.range f) (LinearMap.range g)) :
        LinearMap.ker (f + g) = LinearMap.ker f ⊓ LinearMap.ker g := by
    …
  2. theorem OAI.MaxCutGames.Appendix.RankAdditivity.finrank_add_eq_iff line 47

    Source documentation

    The exact rank-additivity criterion used to define X ⪯ Y in Appendix A.1.

    theorem finrank_add_eq_iff (f g : W →ₗ[K] V) :
        finrank K (LinearMap.range (f + g)) =
            finrank K (LinearMap.range f) + finrank K (LinearMap.range g) ↔
          Disjoint (LinearMap.range f) (LinearMap.range g) ∧
            LinearMap.ker f ⊔ LinearMap.ker g = ⊤ := by
    …
  3. theorem OAI.MaxCutGames.Appendix.RankAdditivity.range_left_le_range_add line 76

    Source documentation

    The left image is contained in the total image for a rank-additive pair.

    theorem range_left_le_range_add (f g : W →ₗ[K] V)
        (h : finrank K (LinearMap.range (f + g)) =
          finrank K (LinearMap.range f) + finrank K (LinearMap.range g)) :
        LinearMap.range f ≤ LinearMap.range (f + g) := by
    …
  4. theorem OAI.MaxCutGames.Appendix.RankAdditivity.ker_add_le_ker_left line 94

    Source documentation

    The total kernel is contained in the left kernel for a rank-additive pair.

    theorem ker_add_le_ker_left (f g : W →ₗ[K] V)
        (h : finrank K (LinearMap.range (f + g)) =
          finrank K (LinearMap.range f) + finrank K (LinearMap.range g)) :
        LinearMap.ker (f + g) ≤ LinearMap.ker f := by
    …
  5. def OAI.MaxCutGames.Appendix.RankAdditivity.RankBelow line 101
  6. theorem OAI.MaxCutGames.Appendix.RankAdditivity.add_remainder line 106
  7. theorem OAI.MaxCutGames.Appendix.RankAdditivity.rankBelow_iff line 110
  8. theorem OAI.MaxCutGames.Appendix.RankAdditivity.range_le_of_rankBelow line 119
  9. theorem OAI.MaxCutGames.Appendix.RankAdditivity.ker_le_of_rankBelow line 128
  10. theorem OAI.MaxCutGames.Appendix.RankAdditivity.range_remainder_eq_range_on_kernel line 137
  11. theorem OAI.MaxCutGames.Appendix.RankAdditivity.mem_ker_of_apply_mem_range_remainder line 161
  12. def OAI.MaxCutGames.Appendix.CompressionCount.base line 189

    Source documentation

    Canonical rank-dim U map.

    def base : (U × K) →ₗ[F2] (U × C) :=
    …
  13. theorem OAI.MaxCutGames.Appendix.CompressionCount.base_apply line 192
  14. def OAI.MaxCutGames.Appendix.CompressionCount.compress line 198

    Source documentation

    Compress by restricting to the second domain factor and taking the second codomain coordinate.

    def compress (Y : (U × K) →ₗ[F2] (U × C)) : K →ₗ[F2] C :=
    …
  15. theorem OAI.MaxCutGames.Appendix.CompressionCount.compress_apply line 201
  16. def OAI.MaxCutGames.Appendix.CompressionCount.Fiber line 206

    Source documentation

    The actual rank-order compression fiber; subtraction agrees with addition over the binary field.

    def Fiber (Z : K →ₗ[F2] C) :=
    …
  17. abbrev OAI.MaxCutGames.Appendix.CompressionCount.Parameters line 211
  18. def OAI.MaxCutGames.Appendix.CompressionCount.graph line 214
  19. def OAI.MaxCutGames.Appendix.CompressionCount.inclusion line 217
  20. def OAI.MaxCutGames.Appendix.CompressionCount.lower line 220
  21. def OAI.MaxCutGames.Appendix.CompressionCount.lift line 224

    Source documentation

    The block matrix [I+FH,FZ;H,Z].

    def lift (Z : K →ₗ[F2] C) (P : Parameters (U := U) Z) :
    …

    Mentions: OAI.MaxCutGames.Appendix.CompressionCount.Parameters

  22. theorem OAI.MaxCutGames.Appendix.CompressionCount.lift_apply line 228
  23. theorem OAI.MaxCutGames.Appendix.CompressionCount.compress_lift line 234
  24. theorem OAI.MaxCutGames.Appendix.CompressionCount.graph_injective line 239
  25. theorem OAI.MaxCutGames.Appendix.CompressionCount.inclusion_injective line 244
  26. theorem OAI.MaxCutGames.Appendix.CompressionCount.lower_surjective line 250
  27. theorem OAI.MaxCutGames.Appendix.CompressionCount.range_lift_sub_base line 257
  28. theorem OAI.MaxCutGames.Appendix.CompressionCount.range_lift line 263
  29. theorem OAI.MaxCutGames.Appendix.CompressionCount.finrank_base line 283
  30. theorem OAI.MaxCutGames.Appendix.CompressionCount.lift_rank_order line 289
  31. def OAI.MaxCutGames.Appendix.CompressionCount.toFiber line 296
  32. theorem OAI.MaxCutGames.Appendix.CompressionCount.lift_injective line 300
  33. theorem OAI.MaxCutGames.Appendix.CompressionCount.natCard_linearMap line 314
  34. theorem OAI.MaxCutGames.Appendix.CompressionCount.card_parameters line 328
  35. theorem OAI.MaxCutGames.Appendix.CompressionCount.fiber_geometry line 337

    Source documentation

    Rank additivity supplies exactly the image/kernel hypotheses used by the block argument.

    theorem fiber_geometry (Z : K →ₗ[F2] C) (Y : Fiber (U := U) Z) :
    …

    Mentions: OAI.MaxCutGames.Appendix.CompressionCount.Fiber

  36. theorem OAI.MaxCutGames.Appendix.CompressionCount.snd_mem_range line 348

    Source documentation

    Every lower coordinate of a map in the fiber lies in the compressed image. This is the step that permits taking H : U → range Z.

    theorem snd_mem_range (Z : K →ₗ[F2] C) (Y : Fiber (U := U) Z) (p : U × K) :
    …

    Mentions: OAI.MaxCutGames.Appendix.CompressionCount.Fiber

  37. def OAI.MaxCutGames.Appendix.CompressionCount.linearSection line 366

    Source documentation

    Choose a linear section of the map onto its image.

    noncomputable def linearSection (Z : K →ₗ[F2] C) : Z.range →ₗ[F2] K :=
    …
  38. theorem OAI.MaxCutGames.Appendix.CompressionCount.section_spec line 371
  39. def OAI.MaxCutGames.Appendix.CompressionCount.extractedF line 376
  40. def OAI.MaxCutGames.Appendix.CompressionCount.extractedH line 381
  41. theorem OAI.MaxCutGames.Appendix.CompressionCount.extracted_lower line 387
  42. theorem OAI.MaxCutGames.Appendix.CompressionCount.difference_ext_snd line 400

    Source documentation

    On the difference image, the lower coordinate uniquely determines the upper coordinate.

    theorem difference_ext_snd (Z : K →ₗ[F2] C) (Y : Fiber (U := U) Z)
    …

    Mentions: OAI.MaxCutGames.Appendix.CompressionCount.Fiber

  43. theorem OAI.MaxCutGames.Appendix.CompressionCount.lift_extracted line 416
  44. theorem OAI.MaxCutGames.Appendix.CompressionCount.toFiber_bijective line 440
  45. def OAI.MaxCutGames.Appendix.CompressionCount.fiberEquiv line 450

    Source documentation

    The exact compression fiber is in bijection with two independent spaces of linear maps. No rank or cardinality claim is an assumption of this equivalence.

    noncomputable def fiberEquiv (Z : K →ₗ[F2] C) :
        Parameters (U := U) Z ≃ Fiber (U := U) Z :=
    …

    Mentions: OAI.MaxCutGames.Appendix.CompressionCount.Parameters, OAI.MaxCutGames.Appendix.CompressionCount.Fiber

  46. theorem OAI.MaxCutGames.Appendix.CompressionCount.fiber_remainder_rank line 455
  47. theorem OAI.MaxCutGames.Appendix.CompressionCount.fiber_rank line 463
  48. theorem OAI.MaxCutGames.Appendix.CompressionCount.card_fiber line 469

    Source documentation

    Lemma A.2 in arbitrary finite-dimensional block coordinates.

    theorem card_fiber (Z : K →ₗ[F2] C) : Nat.card (Fiber (U := U) Z) =
    …

    Mentions: OAI.MaxCutGames.Appendix.CompressionCount.Fiber

  49. theorem OAI.MaxCutGames.Fourier.MatrixSubspaceCount.endomorphism_range_surjective line 493

    Source documentation

    Taking the image of an endomorphism reaches every actual subspace.

    theorem endomorphism_range_surjective :
        Function.Surjective (fun f : V →ₗ[K] V => LinearMap.range f) := by
    …
  50. theorem OAI.MaxCutGames.Fourier.MatrixSubspaceCount.card_subspaces_le_endomorphisms line 499
  51. theorem OAI.MaxCutGames.Fourier.MatrixSubspaceCount.finite_color_space_inline_MatrixSubspaceCount line 509
  52. theorem OAI.MaxCutGames.Fourier.MatrixSubspaceCount.finite_endomorphisms_inline_MatrixSubspaceCount line 513
  53. theorem OAI.MaxCutGames.Fourier.MatrixSubspaceCount.finite_subspaces_inline_MatrixSubspaceCount line 517
  54. theorem OAI.MaxCutGames.Fourier.MatrixSubspaceCount.card_binary_subspaces_le line 521
  55. theorem OAI.MaxCutGames.Fourier.MatrixSubspaceCount.card_binary_subspaces_le_of_finrank_le line 528
  56. theorem OAI.MaxCutGames.Fourier.MatrixSubspaceCount.card_admissible_subspaces_le line 533
  57. theorem OAI.MaxCutGames.Fourier.MatrixSubspaceCount.card_color_admissible_subspaces_le line 541

    Source documentation

    Counting fixed colors together with actual admissible subspaces gives precisely the q * 2^(s*s) denominator budget from (5.13).

    theorem card_color_admissible_subspaces_le (P : Submodule F2 C → Prop) :
        Nat.card (C × {S : Submodule F2 C // P S}) ≤
          Nat.card C * 2 ^ (finrank F2 C * finrank F2 C) := by
    …
  58. theorem OAI.MaxCutGames.Fourier.MatrixSubspaceCount.card_color_admissible_subspaces_le_of_finrank_le line 547
  59. def OAI.MaxCutGames.Fourier.MatrixEnergy.energy line 563

    Source documentation

    Squared coefficient energy, averaged using the supplied question weights.

    noncomputable def energy (weight : Q → ℝ) (a : (q : Q) → I q → ℝ) : ℝ :=
    …
  60. def OAI.MaxCutGames.Fourier.MatrixEnergy.cutoffEnergy line 567

    Source documentation

    Squared energy retained by a coordinate projection.

    noncomputable def cutoffEnergy (weight : Q → ℝ) (keep : (q : Q) → I q → Prop)
        (a : (q : Q) → I q → ℝ) : ℝ := by
    …
  61. def OAI.MaxCutGames.Fourier.MatrixEnergy.spectralEnergy line 573

    Source documentation

    Diagonal spectral quadratic form.

    noncomputable def spectralEnergy (weight : Q → ℝ) (eigen a : (q : Q) → I q → ℝ) : ℝ :=
    …
  62. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.energy_nonneg line 576
  63. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.cutoffEnergy_nonneg line 581
  64. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.cutoffEnergy_le_energy line 597

    Source documentation

    The projection onto any set of coefficient indices is a contraction in the weighted squared norm.

    theorem cutoffEnergy_le_energy (weight : Q → ℝ) (keep : (q : Q) → I q → Prop)
        (a : (q : Q) → I q → ℝ) (hw : ∀ q, 0 ≤ weight q) :
        cutoffEnergy weight keep a ≤ energy weight a := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.cutoffEnergy, OAI.MaxCutGames.Fourier.MatrixEnergy.energy

  65. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.energy_split line 610

    Source documentation

    Retained and discarded coefficient coordinates are orthogonal: their energies add exactly to the original energy.

    theorem energy_split (weight : Q → ℝ) (keep : (q : Q) → I q → Prop) (a : (q : Q) → I q → ℝ) :
        energy weight a = cutoffEnergy weight keep a +
          cutoffEnergy weight (fun q i => ¬ keep q i) a := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.energy, OAI.MaxCutGames.Fourier.MatrixEnergy.cutoffEnergy

  66. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.spectralEnergy_le_cutoff_add line 626

    Source documentation

    A spectral multiplier bounded by one on retained indices and by η elsewhere contributes at most retained energy plus η times total energy.

    theorem spectralEnergy_le_cutoff_add
        (weight : Q → ℝ) (keep : (q : Q) → I q → Prop) (eigen a : (q : Q) → I q → ℝ) (η : ℝ)
        (hw : ∀ q, 0 ≤ weight q) (hη : 0 ≤ η)
        (hone : ∀ q i, eigen q i ≤ 1)
        (hhigh : ∀ q i, ¬ keep q i → eigen q i ≤ η) :
        spectralEnergy weight eigen a ≤ cutoffEnergy weight keep a + η * energy weight a := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.spectralEnergy, OAI.MaxCutGames.Fourier.MatrixEnergy.cutoffEnergy, OAI.MaxCutGames.Fourier.MatrixEnergy.energy

  67. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.cutoffEnergy_gt_half line 659

    Source documentation

    Unit total energy and spectral value above δ force retained energy above δ/2 when all discarded multipliers are at most δ/2.

    theorem cutoffEnergy_gt_half
        (weight : Q → ℝ) (keep : (q : Q) → I q → Prop) (eigen a : (q : Q) → I q → ℝ) (δ : ℝ)
        (hw : ∀ q, 0 ≤ weight q) (hδ : 0 ≤ δ)
        (hone : ∀ q i, eigen q i ≤ 1)
        (hhigh : ∀ q i, ¬ keep q i → eigen q i ≤ δ / 2)
        (htotal : energy weight a = 1) (hvalue : δ < spectralEnergy weight eigen a) :
        δ / 2 < cutoffEnergy weight keep a := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.energy, OAI.MaxCutGames.Fourier.MatrixEnergy.spectralEnergy, OAI.MaxCutGames.Fourier.MatrixEnergy.cutoffEnergy

  68. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.cutoffEnergy_le_twice_difference_add line 674

    Source documentation

    The elementary square bound, summed over the retained coefficients. The discarded term on the right uses the full coefficient difference, giving the contraction needed for equation (5.11) without introducing square roots.

    theorem cutoffEnergy_le_twice_difference_add
        (weight : Q → ℝ) (keep : (q : Q) → I q → Prop) (H G : (q : Q) → I q → ℝ)
        (hw : ∀ q, 0 ≤ weight q) :
        cutoffEnergy weight keep H ≤
          2 * energy weight (fun q i => H q i - G q i) + 2 * cutoffEnergy weight keep G := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.cutoffEnergy, OAI.MaxCutGames.Fourier.MatrixEnergy.energy

  69. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.deleted_energy_gt_eighth line 701
  70. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.boolean_energy_eq_mass line 711

    Source documentation

    Squared Boolean functions equal their values under a finite weighted sum.

    theorem boolean_energy_eq_mass (weight : Q → ℝ) (H : (q : Q) → I q → ℝ)
        (hbool : ∀ q i, H q i = 0 ∨ H q i = 1) :
        energy weight H = ∑ q, weight q * ∑ i, H q i := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.energy

  71. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.weighted_union_bound line 724

    Source documentation

    A finite union bound for fractional removed mass dominated by indicator events. Both the event choice and the removed value may depend on the outcome.

    theorem weighted_union_bound {Ω J : Type*} [Fintype Ω] [Fintype J]
        (weight value : Ω → ℝ) (event : J → Ω → Prop)
        [∀ j, DecidablePred (event j)]
        (hw : ∀ ω, 0 ≤ weight ω) (hvalue : ∀ ω, value ω ≤ 1)
        (hcover : ∀ ω, 0 < value ω → ∃ j, event j ω) :
        (∑ ω, weight ω * value ω) ≤
          ∑ j, ∑ ω, weight ω * (if event j ω then (1 : ℝ) else 0) := by
    …
  72. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.exists_fixed_event line 749

    Source documentation

    Finite real pigeonhole principle for the one fixed color/subspace choice.

    theorem exists_fixed_event {J : Type*} [Fintype J] (mass : J → ℝ)
        (threshold budget : ℝ)
        (hcount : (Fintype.card J : ℝ) * threshold ≤ budget)
        (hmass : budget < ∑ j, mass j) : ∃ j, threshold < mass j := by
    …
  73. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.exists_fixed_event_scaled line 763

    Source documentation

    The explicit denominator in (5.13), for any supplied event-count bound.

    theorem exists_fixed_event_scaled {J : Type*} [Fintype J]
        (mass : J → ℝ) (N : ℕ) (δ : ℝ)
        (hN : 0 < N) (hδ : 0 ≤ δ) (hcard : Fintype.card J ≤ N)
        (hmass : δ / 8 < ∑ j, mass j) :
        ∃ j, δ / (8 * (N : ℝ)) < mass j := by
    …
  74. def OAI.MaxCutGames.Fourier.MatrixEnergy.linearRankOn line 788

    Source documentation

    The actual rank of a Fourier index restricted to the alphabet subspace.

    noncomputable def linearRankOn (C : Submodule F2 F) (S : F →ₗ[F2] E) : ℕ :=
    …
  75. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.linearRankOn_top line 793
  76. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.linearRankOn_range line 810

    Source documentation

    Restricting to the image of a symbol embedding has the same image rank as composition with that embedding, even when it is not injective.

    theorem linearRankOn_range {V : Type*} [AddCommGroup V] [Module F2 V]
        (S : F →ₗ[F2] E) (ι : V →ₗ[F2] F) :
        linearRankOn (LinearMap.range ι) S =
          Module.finrank F2 (LinearMap.range (S.comp ι)) := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.linearRankOn

  77. def OAI.MaxCutGames.Fourier.MatrixEnergy.rankProjection line 829

    Source documentation

    Projection onto trace characters whose index has bounded rank on C.

    noncomputable def rankProjection (C : Submodule F2 F) (r : ℕ)
        (f : (E →ₗ[F2] F) → ℝ) : (E →ₗ[F2] F) → ℝ := by
    …
  78. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.linearCoeff_sub line 837
  79. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.linearCoeff_rankProjection line 842

    Source documentation

    The genuine function-space projection acts by the claimed coefficient mask.

    theorem linearCoeff_rankProjection (C : Submodule F2 F) (r : ℕ)
        (f : (E →ₗ[F2] F) → ℝ) (S : F →ₗ[F2] E) :
        linearCoeff (rankProjection C r f) S =
          if linearRankOn C S ≤ r then linearCoeff f S else 0 := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.rankProjection, OAI.MaxCutGames.Fourier.MatrixEnergy.linearRankOn

  80. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.rankProjection_parseval line 853

    Source documentation

    Squared norm of the actual projection equals the retained coefficient mass.

    theorem rankProjection_parseval (C : Submodule F2 F) (r : ℕ)
        (f : (E →ₗ[F2] F) → ℝ) :
        (𝔼 X, rankProjection C r f X ^ 2) =
          ∑ S : F →ₗ[F2] E, if linearRankOn C S ≤ r then linearCoeff f S ^ 2 else 0 := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.rankProjection, OAI.MaxCutGames.Fourier.MatrixEnergy.linearRankOn

  81. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.rankProjection_contraction line 865

    Source documentation

    Contraction for the actual rank projection with the uniform function norm.

    theorem rankProjection_contraction (C : Submodule F2 F) (r : ℕ)
        (f : (E →ₗ[F2] F) → ℝ) :
        (𝔼 X, rankProjection C r f X ^ 2) ≤ 𝔼 X, f X ^ 2 := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.rankProjection

  82. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.rankProjection_orthogonal line 878

    Source documentation

    The projected part is orthogonal to the discarded part in the actual uniform function inner product.

    theorem rankProjection_orthogonal (C : Submodule F2 F) (r : ℕ)
        (f : (E →ₗ[F2] F) → ℝ) :
        (𝔼 X, rankProjection C r f X * (f X - rankProjection C r f X)) = 0 := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.rankProjection

  83. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.linear_energy_parseval line 888
  84. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.linear_difference_energy_parseval line 893
  85. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.linear_deleted_energy_gt_eighth line 901

    Source documentation

    Equation (5.11) now for actual functions on a binary linear-map space, using the actual restriction rank and trace Fourier coefficients.

    theorem linear_deleted_energy_gt_eighth
        (weight : Q → ℝ) (C : Submodule F2 F) (r : ℕ)
        (H G : Q → (E →ₗ[F2] F) → ℝ) (δ : ℝ)
        (hw : ∀ q, 0 ≤ weight q)
        (hlarge : δ / 2 < cutoffEnergy weight (fun _ S => linearRankOn C S ≤ r)
          (fun q S => linearCoeff (H q) S))
        (hsmall : cutoffEnergy weight (fun _ S => linearRankOn C S ≤ r)
          (fun q S => linearCoeff (G q) S) ≤ δ / 8) :
        δ / 8 < ∑ q, weight q * (𝔼 X, (H q X - G q X) ^ 2) := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.cutoffEnergy, OAI.MaxCutGames.Fourier.MatrixEnergy.linearRankOn

  86. def OAI.MaxCutGames.Fourier.MatrixEnergy.colorIndicator line 916

    Source documentation

    Color indicator of one actual labeling.

    noncomputable def colorIndicator {X Y : Type*} (label : X → Y) (y : Y) : X → ℝ := by
    …
  87. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.colorIndicator_partition line 920
  88. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.colorIndicator_parseval line 927

    Source documentation

    Parseval for all color indicators sums to exactly one, with the correct uniform primal normalization and counting measure on Fourier indices.

    theorem colorIndicator_parseval {Y : Type*} [Fintype Y]
        (label : (E →ₗ[F2] F) → Y) :
        (∑ y, ∑ S : F →ₗ[F2] E, linearCoeff (colorIndicator label y) S ^ 2) = 1 := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.colorIndicator

  89. def OAI.MaxCutGames.Fourier.MatrixEnergy.labelingCoeff line 937

    Source documentation

    Fourier coefficient array for a family of labelings, including color as part of the component index.

    noncomputable def labelingCoeff {Y : Type*}
        (label : Q → (E →ₗ[F2] F) → Y) (qy : Q × Y) (S : F →ₗ[F2] E) : ℝ :=
    …
  90. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.labelingCoeff_energy_one line 941
  91. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.labeling_lowRank_energy line 957

    Source documentation

    The low-rank Fourier-energy conclusion (5.5), for genuine binary linear maps and genuine labeling coefficients. The spectral premise can be obtained from the noise diagonalization; the bound is on actual restriction rank.

    theorem labeling_lowRank_energy {Y : Type*} [Fintype Y]
        (weight : Q → ℝ) (label : Q → (E →ₗ[F2] F) → Y)
        (C : Submodule F2 F) (r : ℕ) (eigen : Q → (F →ₗ[F2] E) → ℝ) (δ : ℝ)
        (hw : ∀ q, 0 ≤ weight q) (hnorm : ∑ q, weight q = 1) (hδ : 0 ≤ δ)
        (hone : ∀ q S, eigen q S ≤ 1)
        (hhigh : ∀ q S, r < linearRankOn C S → eigen q S ≤ δ / 2)
        (hvalue : δ < spectralEnergy (fun qy : Q × Y => weight qy.1)
          (fun qy S => eigen qy.1 S) (labelingCoeff label)) :
        δ / 2 < cutoffEnergy (fun qy : Q × Y => weight qy.1)
          (fun _ S => linearRankOn C S ≤ r) (labelingCoeff label) := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.linearRankOn, OAI.MaxCutGames.Fourier.MatrixEnergy.spectralEnergy, OAI.MaxCutGames.Fourier.MatrixEnergy.labelingCoeff, OAI.MaxCutGames.Fourier.MatrixEnergy.cutoffEnergy

  92. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.dependent_linear_energy_parseval line 987
  93. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.dependent_remainder_cutoff_le_eighth line 998
  94. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.dependent_deleted_energy_from_level line 1028

    Source documentation

    The actual removed probability energy in Section 5.3, allowing the map domain to depend on the complete question and color index.

    theorem dependent_deleted_energy_from_level
        (weight : J → ℝ) (C : Submodule F2 F) (r : ℕ)
        (H G : (j : J) → (E j →ₗ[F2] F) → ℝ) (κ δ : ℝ)
        (hw : ∀ j, 0 ≤ weight j) (hκ : 0 ≤ κ) (hκδ : κ ≤ δ / 8)
        (hlarge : δ / 2 < cutoffEnergy weight (fun _ S => linearRankOn C S ≤ r)
          (fun j S => linearCoeff (H j) S))
        (hlevel : ∀ j, (∑ S : F →ₗ[F2] E j,
          if linearRankOn C S ≤ r then linearCoeff (G j) S ^ 2 else 0) ≤
            κ * (𝔼 X, G j X ^ 2))
        (hmass : (∑ j, weight j * (𝔼 X, G j X ^ 2)) ≤ 1) :
        δ / 8 < ∑ j, weight j * (𝔼 X, (H j X - G j X) ^ 2) := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.cutoffEnergy, OAI.MaxCutGames.Fourier.MatrixEnergy.linearRankOn

  95. def OAI.MaxCutGames.Fourier.MatrixEnergy.boolReal line 1049

    Source documentation

    Real value of a Boolean indicator.

    def boolReal (b : Bool) : ℝ := if b then 1 else 0
  96. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.boolReal_sq line 1051
  97. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.boolReal_le_one line 1054
  98. def OAI.MaxCutGames.Fourier.MatrixEnergy.labelBool line 1058

    Source documentation

    Boolean indicator of one color of a labeling.

    noncomputable def labelBool {X Y : Type*} (label : X → Y) (y : Y) : X → Bool := by
    …
  99. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.labelBool_real line 1062
  100. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.removed_boolReal line 1069

    Source documentation

    Deletion's actual Boolean removed function is exactly the difference of the original and remainder indicators.

    theorem removed_boolReal {X Row Witness : Type*}
        (S : Deletion.RestrictionSystem X Row Witness) (H : X → Bool) (p d : ℕ) (x : X) :
        boolReal (Deletion.removed S H p d x) =
          boolReal (H x) - boolReal (Deletion.remainder S H p d x) := by
    …

    Mentions: OAI.MaxCutGames.Fourier.Deletion.RestrictionSystem, OAI.MaxCutGames.Fourier.MatrixEnergy.boolReal, OAI.MaxCutGames.Fourier.Deletion.removed, OAI.MaxCutGames.Fourier.Deletion.remainder

  101. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.removed_difference_sq line 1078
  102. def OAI.MaxCutGames.Fourier.MatrixEnergy.questionEventMass line 1091

    Source documentation

    An event probability: first sample the question with the supplied weights, then sample uniformly in its own finite space.

    noncomputable def questionEventMass
        (weight : Q → ℝ) (B : (q : Q) → Ω q → Prop) : ℝ := by
    …
  103. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.dependent_removed_union_bound line 1096
  104. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.fixed_target_of_removed_mass line 1133

    Source documentation

    One target is fixed before sampling the question and its map.

    theorem fixed_target_of_removed_mass
        (weight : Q → ℝ) (d : (q : Q) → Y → Ω q → ℝ)
        (B : (q : Q) → Y → A → Ω q → Prop)
        (hw : ∀ q, 0 ≤ weight q) (hd : ∀ q y X, d q y X ≤ 1)
        (hc : ∀ q y X, 0 < d q y X → ∃ a, B q y a X)
        (δ : ℝ) (N : ℕ) (hδ : 0 ≤ δ) (hN : 0 < N)
        (hcount : Fintype.card (Y × A) ≤ N)
        (hremoved : δ / 8 < ∑ q, weight q * ∑ y, 𝔼 X, d q y X) :
        ∃ y a, δ / (8 * (N : ℝ)) <
          questionEventMass weight (fun q X => B q y a X) := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.questionEventMass

  105. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.boolean_subpartition_energy_le_one line 1151

    Source documentation

    A Boolean subpartition has total squared mass at most one. Colors are summed with counting measure, not averaged.

    theorem boolean_subpartition_energy_le_one [∀ q, Nonempty (Ω q)]
        (weight : Q → ℝ) (label : (q : Q) → Ω q → Y)
        (G : (q : Q) → Y → Ω q → Bool)
        (hw : ∀ q, 0 ≤ weight q) (hnorm : ∑ q, weight q = 1)
        (hG : ∀ q y X, G q y X = true → label q X = y) :
        (∑ q, weight q * ∑ y, 𝔼 X, boolReal (G q y X) ^ 2) ≤ 1 := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.boolReal

  106. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.remainder_subpartition_energy_le_one line 1181

    Source documentation

    The actual whole-row remainder of labeling indicators satisfies the mass bound needed in the deleted-energy argument, with no remaining mass hypothesis.

    theorem remainder_subpartition_energy_le_one [∀ q, Nonempty (Ω q)]
        {Row Witness : Q → Type*}
        (S : (q : Q) → Deletion.RestrictionSystem (Ω q) (Row q) (Witness q))
        (weight : Q → ℝ) (label : (q : Q) → Ω q → Y) (p d : ℕ)
        (hw : ∀ q, 0 ≤ weight q) (hnorm : ∑ q, weight q = 1) :
        (∑ q, weight q * ∑ y, 𝔼 X,
          boolReal (Deletion.remainder (S q) (labelBool (label q) y) p d X) ^ 2) ≤ 1 := by
    …

    Mentions: OAI.MaxCutGames.Fourier.Deletion.RestrictionSystem, OAI.MaxCutGames.Fourier.MatrixEnergy.boolReal, OAI.MaxCutGames.Fourier.Deletion.remainder, OAI.MaxCutGames.Fourier.MatrixEnergy.labelBool

  107. theorem OAI.MaxCutGames.Fourier.MatrixEnergy.fixed_badRow_target_of_removed_mass line 1205

    Source documentation

    Equation (5.13) for the actual whole-row deletion system. Only the color and target subspace are selected. Neither the question, row value, nor witness vanishing subspace is part of the counted target.

    theorem fixed_badRow_target_of_removed_mass
        (weight : Q → ℝ) (H : (q : Q) → C → (E q →ₗ[F2] C) → Bool)
        (r p d : ℕ) (δ : ℝ)
        (hw : ∀ q, 0 ≤ weight q) (hδ : 0 ≤ δ)
        (hremoved : δ / 8 < ∑ q, weight q * ∑ y, 𝔼 X,
          boolReal (Deletion.removed (MatrixRestrictions.system r) (H q y) p d X)) :
        ∃ (y : C) (C' : Submodule F2 C), Module.finrank F2 (C ⧸ C') ≤ r ∧
          δ / (8 * (Nat.card C : ℝ) * (2 : ℝ) ^
            (Module.finrank F2 C * Module.finrank F2 C)) <
          questionEventMass weight (fun q (X : E q →ₗ[F2] C) =>
            Deletion.BadRow (MatrixRestrictions.system r) (H q y) p d
              ⟨C', C'.mkQ.comp X⟩) := by
    …

    Mentions: OAI.MaxCutGames.Fourier.MatrixEnergy.boolReal, OAI.MaxCutGames.Fourier.Deletion.removed, OAI.MaxCutGames.Fourier.MatrixRestrictions.system, OAI.MaxCutGames.Fourier.MatrixEnergy.questionEventMass, OAI.MaxCutGames.Fourier.Deletion.BadRow