OAI/Combinatorics/SparsestCut/WeightReplication.lean

GitHub source file

14 declarations, 0 contain sorry, 0 declared as axioms

Replication of finite weights

Ceilings produce integer cluster sizes while giving one copy to every key of zero weight. The resulting counting law satisfies the multiplicative and auxiliary-mass estimates from Section 7 of OpenAI's Constant-factor hardness of uniform sparsest cut.

  1. def OAI.UniformSparsestCut.FiniteWeights.copies line 26
  2. def OAI.UniformSparsestCut.FiniteWeights.population line 28
  3. def OAI.UniformSparsestCut.FiniteWeights.countingWeights line 30
  4. theorem OAI.UniformSparsestCut.FiniteWeights.copies_zero line 34
  5. theorem OAI.UniformSparsestCut.FiniteWeights.copies_pos line 37
  6. theorem OAI.UniformSparsestCut.FiniteWeights.copies_lower line 43
  7. theorem OAI.UniformSparsestCut.FiniteWeights.copies_upper line 50
  8. theorem OAI.UniformSparsestCut.FiniteWeights.population_eq_total line 56
  9. theorem OAI.UniformSparsestCut.FiniteWeights.population_bounds line 60
  10. theorem OAI.UniformSparsestCut.FiniteWeights.population_pos line 72
  11. theorem OAI.UniformSparsestCut.FiniteWeights.counting_nonneg line 76
  12. theorem OAI.UniformSparsestCut.FiniteWeights.counting_total line 80
  13. theorem OAI.UniformSparsestCut.FiniteWeights.counting_bounds line 90

    Source documentation

    On a key of positive weight, the counting law lies within a factor of two of the original probability law.

    theorem counting_bounds (K : ℕ) (hK : 0 < K) {p : ι → ℝ}
        (hp : ∀ i, 0 ≤ p i) (hp₁ : total p = 1) (hKm : Fintype.card ι ≤ K)
        (hKp : ∀ i, 0 < p i → 1 ≤ (K : ℝ) * p i) (i : ι) (hpi : 0 < p i) :
        p i / 2 ≤ countingWeights K p i ∧ countingWeights K p i ≤ 2 * p i := by
    …

    Mentions: OAI.UniformSparsestCut.FiniteWeights.total, OAI.UniformSparsestCut.FiniteWeights.countingWeights

  14. theorem OAI.UniformSparsestCut.FiniteWeights.auxiliary_counting_bound line 110

    Source documentation

    The total counting mass of auxiliary (zero-retained-demand) vertices.

    theorem auxiliary_counting_bound (K : ℕ) (hK : 0 < K) {p : ι → ℝ} {κ : ℝ}
        (hp : ∀ i, 0 ≤ p i) (hp₁ : total p = 1) (hκ : 0 < κ)
        (hKlarge : 4 * κ * Fintype.card ι ≤ K) :
        total (auxiliaryWeights (countingWeights K p) p) ≤ 1 / (4 * κ) := by
    …

    Mentions: OAI.UniformSparsestCut.FiniteWeights.total, OAI.UniformSparsestCut.FiniteWeights.auxiliaryWeights, OAI.UniformSparsestCut.FiniteWeights.countingWeights