OAI/NumberTheory/JointDickman/Probability/IndependentKernelSplit.lean

GitHub source file

25 declarations, 0 contain sorry, 0 declared as axioms

The independent common/exclusive model and its cutoff

The three independent category bits are reindexed as three independent Boolean prime subsets. Their products are the common product and the two exclusive products. The low-exclusive union is separated exactly; the complement is the product of the two high-exclusive cell probabilities.

  1. def OAI.JointDickman.categoryToTriple line 20
  2. def OAI.JointDickman.tripleToCategory line 24
  3. theorem OAI.JointDickman.tripleToCategory_categoryToTriple line 29
  4. theorem OAI.JointDickman.categoryToTriple_tripleToCategory line 36
  5. def OAI.JointDickman.categoryTripleEquiv line 42

    Source documentation

    Three category bits at each site are exactly three Boolean vectors.

    def categoryTripleEquiv (ι : Type*) :
        (ι → SplitCategories) ≃ (ι → Bool) × (ι → Bool) × (ι → Bool) where
    …

    Mentions: OAI.JointDickman.SplitCategories

  6. theorem OAI.JointDickman.independentCategoryMass_triple line 61

    Source documentation

    The independent category weight factors into the three actual Bernoulli subset weights, without a coupling assumption.

    theorem independentCategoryMass_triple {ι : Type*} [Fintype ι] [DecidableEq ι]
        (q : ι → ℝ) (y : ι → SplitCategories) :
        bernoulliProductMass Finset.univ (fun i (_ : Fin 3) => q i) y =
          bernoulliSiteMass q ((categoryTripleEquiv ι y).1) *
          bernoulliSiteMass q ((categoryTripleEquiv ι y).2.1) *
          bernoulliSiteMass q ((categoryTripleEquiv ι y).2.2) :=
    …

    Mentions: OAI.JointDickman.SplitCategories, OAI.JointDickman.bernoulliProductMass, OAI.JointDickman.bernoulliSiteMass, OAI.JointDickman.categoryTripleEquiv

  7. theorem OAI.JointDickman.independentCategory_triple_test line 70

    Source documentation

    Exact real-test pushforward to the independent triple law.

    theorem independentCategory_triple_test {ι : Type*} [Fintype ι] [DecidableEq ι]
        (q : ι → ℝ) (F : (ι → Bool) → (ι → Bool) → (ι → Bool) → ℝ) :
        (∑ y, bernoulliProductMass Finset.univ (fun i (_ : Fin 3) => q i) y *
          F ((categoryTripleEquiv ι y).1) ((categoryTripleEquiv ι y).2.1)
            ((categoryTripleEquiv ι y).2.2)) =
          ∑ c, ∑ e₁, ∑ e₂,
            bernoulliSiteMass q c * bernoulliSiteMass q e₁ * bernoulliSiteMass q e₂ * F c e₁ e₂ := by
    …

    Mentions: OAI.JointDickman.bernoulliProductMass, OAI.JointDickman.categoryTripleEquiv, OAI.JointDickman.bernoulliSiteMass

  8. theorem OAI.JointDickman.categoryPrimeProduct_eq_retained line 83
  9. def OAI.JointDickman.independentPairMass line 96
  10. theorem OAI.JointDickman.independentPairMass_eq_triple line 100
  11. def OAI.JointDickman.lowUnionExclusiveMass line 118

    Source documentation

    The discarded mass is the genuine union event, so overlap is counted once.

    noncomputable def lowUnionExclusiveMass (w : C → ℝ) (v : E → ℝ)
        (cell : C → E → Option A) (low : E → Prop) [DecidablePred low] (a b : A) : ℝ :=
    …
  12. def OAI.JointDickman.highExclusiveCell line 125

    Source documentation

    Conditional cell probability retaining only high exclusive products.

    noncomputable def highExclusiveCell (v : E → ℝ) (cell : C → E → Option A)
        (low : E → Prop) [DecidablePred low] (c : C) (a : A) : ℝ :=
    …
  13. theorem OAI.JointDickman.highExclusiveCell_eq_sum line 130
  14. theorem OAI.JointDickman.highExclusiveCell_nonneg line 140
  15. theorem OAI.JointDickman.highExclusiveCell_le_one line 148
  16. theorem OAI.JointDickman.independentPairMass_cutoff line 157

    Source documentation

    The high-high complement factors after conditioning on the common variable.

    theorem independentPairMass_cutoff (w : C → ℝ) (v : E → ℝ)
        (cell : C → E → Option A) (low : E → Prop) [DecidablePred low] (a b : A) :
        independentPairMass w v cell a b = lowUnionExclusiveMass w v cell low a b +
          ∑ c, w c * highExclusiveCell v cell low c a * highExclusiveCell v cell low c b := by
    …

    Mentions: OAI.JointDickman.independentPairMass, OAI.JointDickman.lowUnionExclusiveMass, OAI.JointDickman.highExclusiveCell

  17. theorem OAI.JointDickman.lowUnionExclusiveMass_nonneg line 181
  18. theorem OAI.JointDickman.lowUnionExclusiveMass_le_sides line 194

    Source documentation

    The low union is dominated by the two one-sided low kernels.

    theorem lowUnionExclusiveMass_le_sides (w : C → ℝ) (v : E → ℝ)
        (cell : C → E → Option A) (low : E → Prop) [DecidablePred low]
        (hw : ∀ c, 0 ≤ w c) (hv : ∀ e, 0 ≤ v e) (a b : A) :
        lowUnionExclusiveMass w v cell low a b ≤
          lowExclusiveMass w v cell low a b + lowExclusiveMass w v cell low b a := by
    …

    Mentions: OAI.JointDickman.lowUnionExclusiveMass, OAI.JointDickman.lowExclusiveMass

  19. def OAI.JointDickman.lowUnionExclusiveKernel line 217
  20. theorem OAI.JointDickman.lowUnionExclusiveKernel_nonneg line 221
  21. theorem OAI.JointDickman.lowUnionExclusiveKernel_weighted_rows line 230

    Source documentation

    Local marginal and conditional common-cell bounds control the actual union kernel, including the overlap of its two low-exclusive events.

    theorem lowUnionExclusiveKernel_weighted_rows [Fintype A]
        (w : C → ℝ) (v : E → ℝ) (cell : C → E → Option A)
        (low : E → Prop) [DecidablePred low]
        (hw : ∀ c, 0 ≤ w c) (hv : ∀ e, 0 ≤ v e) (hv1 : ∑ e, v e = 1)
        {m H : ℝ} (hm : 0 < m)
        (hmarginal : ∀ a, independentCellMarginal w v cell a ≤ H * m)
        (hcommon : ∀ e, low e → ∀ b, optionCellMass w (fun c => cell c e) b ≤ H * m)
        (a : A) :
        (∑ b, m * lowUnionExclusiveKernel w v cell low (fun _ => m) a b) ≤
          2 * lowExclusiveProbability v low * H := by
    …

    Mentions: OAI.JointDickman.independentCellMarginal, OAI.JointDickman.optionCellMass, OAI.JointDickman.lowUnionExclusiveKernel, OAI.JointDickman.lowExclusiveProbability

  22. def OAI.JointDickman.primeProductCell line 268
  23. theorem OAI.JointDickman.independentPrimeCellMass_eq_pair line 272

    Source documentation

    Exact cell-pair probabilities under the independent prime-category law.

    theorem independentPrimeCellMass_eq_pair {A : Type*} [DecidableEq A]
        (Q : Finset ℕ) (q : Q → ℝ) (cell : ℕ → Option A) (a b : A) :
        (∑ y, if cell (independentSplitProducts Q y).1 = some a ∧
            cell (independentSplitProducts Q y).2 = some b then
          bernoulliProductMass Finset.univ (fun p (_ : Fin 3) => q p) y else 0) =
          independentPairMass (bernoulliSiteMass q) (bernoulliSiteMass q)
            (primeProductCell Q cell) a b := by
    …

    Mentions: OAI.JointDickman.independentSplitProducts, OAI.JointDickman.bernoulliProductMass, OAI.JointDickman.independentPairMass, OAI.JointDickman.bernoulliSiteMass, OAI.JointDickman.primeProductCell

  24. theorem OAI.JointDickman.independentPairKernel_cutoff line 302

    Source documentation

    Normalize the exact cutoff identity by the actual output cell masses.

    theorem independentPairKernel_cutoff {C E D R : Type*}
        [Fintype C] [Fintype E] [DecidableEq D] [DecidableEq R]
        (w : C → ℝ) (v : E → ℝ) (cell : C → E → Option (D × R))
        (low : E → Prop) [DecidablePred low] (μ : D → ℝ) (a b : D × R) :
        independentPairMass w v cell a b / (μ a.1 * μ b.1) =
          lowUnionExclusiveKernel w v cell low (fun a : D × R => μ a.1) a b +
          exclusiveProductKernel w μ (highExclusiveCell v cell low) a b := by
    …

    Mentions: OAI.JointDickman.independentPairMass, OAI.JointDickman.lowUnionExclusiveKernel, OAI.JointDickman.exclusiveProductKernel, OAI.JointDickman.highExclusiveCell

  25. theorem OAI.JointDickman.independentPrimeKernel_cutoff line 314

    Source documentation

    The independent category kernel has exactly the low-union plus high-exclusive-product decomposition needed by residue flattening.

    theorem independentPrimeKernel_cutoff {D R : Type*} [DecidableEq D] [DecidableEq R]
        (Q : Finset ℕ) (q : Q → ℝ) (cell : ℕ → Option (D × R))
        (low : (Q → Bool) → Prop) [DecidablePred low] (μ : D → ℝ) (a b : D × R) :
        (∑ y, if cell (independentSplitProducts Q y).1 = some a ∧
            cell (independentSplitProducts Q y).2 = some b then
          bernoulliProductMass Finset.univ (fun p (_ : Fin 3) => q p) y else 0) /
            (μ a.1 * μ b.1) =
          lowUnionExclusiveKernel (bernoulliSiteMass q) (bernoulliSiteMass q)
            (primeProductCell Q cell) low (fun a : D × R => μ a.1) a b +
          exclusiveProductKernel (bernoulliSiteMass q) μ
            (highExclusiveCell (bernoulliSiteMass q) (primeProductCell Q cell) low) a b := by
    …

    Mentions: OAI.JointDickman.independentSplitProducts, OAI.JointDickman.bernoulliProductMass, OAI.JointDickman.lowUnionExclusiveKernel, OAI.JointDickman.bernoulliSiteMass, OAI.JointDickman.primeProductCell, OAI.JointDickman.exclusiveProductKernel, OAI.JointDickman.highExclusiveCell