Lemma 5.2, p. 7 —
ProvedCurvatureSubmod.LocalSearch.lemma_5_2local-searchmatroidp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1submodular
Let be a finite set, monotone increasing and submodular, a linear function with arbitrary real weights , the Filmus–Ward potential of , and
Let and be bases of a matroid on , indexed according to Lemma 5.1 so that is a base for all . Then
If no swap improves , the sum is nonpositive and is a joint -approximation for and ; this is the analysis of the non-oblivious local search of Figure 4.
Formalization Note The indexing is a bijection (Set.BijOn) with ; the sum runs over . As in (5.4), is assumed monotone and nonnegative, the setting in which the paper uses the lemma, and excludes the term.
Preamble
import Mathlib import Definitions.Def_NonmonotoneSubmod_Shared_Submodular import Definitions.Def_CurvatureSubmod_LocalSearch_Setting
Formal statement
namespace CurvatureSubmod.LocalSearch
/-- Lemma 5.2, p. 7: `g(A) + ℓ(A) ≥ (1 − e^{−1}) g(B) + ℓ(B) + Σ_i [ψ(A) − ψ(A − a_i + b_i)]`
for bases indexed as in Lemma 5.1. -/
theorem lemma_5_2 {X : Type} [Fintype X] [DecidableEq X] (g : Finset X → ℝ)
(hg0 : ∀ A, 0 ≤ g A) (hgmono : MonoInc g) (hgsub : NonmonotoneSubmod.Shared.Submodular g)
(w : X → ℝ) (M : Matroid X) (A B : Finset X)
(hA : M.IsBase (↑A : Set X)) (hB : M.IsBase (↑B : Set X))
(π : X → X) (hπ : Set.BijOn π (↑A : Set X) (↑B : Set X))
(hswap : ∀ x ∈ A, M.IsBase (↑(insert (π x) (A.erase x)) : Set X)) :
(1 - Real.exp (-1)) * g B + linFun w B
+ ∑ a ∈ A, (psi g w A - psi g w (insert (π a) (A.erase a)))
≤ g A + linFun w A := by sorry
end CurvatureSubmod.LocalSearch
Source
Sviridenko, Vondrák, Ward, Optimal approximation for submodular and supermodular optimization with bounded curvature (SODA 2015 version, Oct. 9, 2014), p. 7, Lemma 5.2
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.