Theorem 8.7 — the rank of w.r.t. L&P, simple disjunctive, strengthened L&P and MIG cuts is at most
OpenDisjunctive.CutCorrespondence.rank_bound_four_cut_families_v2Let be the LP relaxation of a mixed 0-1 program with 0-1 variables , where, as throughout Chapter 8, the system contains the rows () and (). Then the rank of with respect to each of the following cut families is at most :
(a) unstrengthened lift-and-project cuts; (b) simple disjunctive cuts; (c) strengthened lift-and-project cuts; (d) mixed integer Gomory cuts (strengthened simple disjunctive cuts).
That is, for each family there is an ordering of such that applying the family's elementary closure for the disjunctions in turn produces .
Formalization Note. The retired version was refuted through part (d): its PiBar applied the Gomory strengthening to the surplus of an arbitrary row (with read as a set of row positions), which yields cuts invalid for the integer hull. The corrected definitions strengthen only nonbasic positions whose row is a bound with (""), and Theorem 6.4's now omits the multipliers of from (the retired equalled ). The chapter's standing assumption is stated as the book states it, as rows of the system (IsMixedZeroOneSystem), instead of the weaker Poly A b ⊆ box. The rank is formalized, as in the retired version, by one generic predicate HasRankAtMost over an enumeration of ; closures of the representation-dependent families are intersected over all linear representations of the current polyhedron.
import Mathlib import Definitions.Def_Disjunctive_CutCorrespondence_Basic import Definitions.Def_Disjunctive_CutCorrespondence_Cglp_v2 import Definitions.Def_Disjunctive_CutCorrespondence_Tableau_v2 import Definitions.Def_Disjunctive_CutCorrespondence_Rank_v2
namespace Disjunctive.CutCorrespondence
/-- Theorem 8.7 (Balas, *Disjunctive Programming*, §8.4, p. 105), the goal theorem of this
mission: for the LP relaxation `P = {x : Ãx ≥ b̃}` of a mixed 0-1 program, the rank of `P` with
respect to each of (a) unstrengthened lift-and-project cuts, (b) simple disjunctive cuts,
(c) strengthened lift-and-project cuts, and (d) mixed integer Gomory cuts (strengthened simple
disjunctive cuts) is at most `p := |N'|`, the number of 0-1 variables.
Corrected from the retired version: the MIG strengthening (`PiBar`) is applied only to nonbasic
positions whose row is the bound `x_j ≥ 0` of a 0-1 variable (the retired `PiBar` strengthened
the surplus of arbitrary rows, producing invalid "cuts"), Theorem 6.4's `Gamma` now removes the
multipliers of `x_j ≥ 0` from `uÃ_j, vÃ_j`, and the standing assumption of the chapter is stated
as the book does: the system `Ãx ≥ b̃` contains the rows `x_j ≥ 0` (all `j`) and `-x_j ≥ -1`
(`j ∈ N'`). -/
theorem rank_bound_four_cut_families_v2 {m n : ℕ} (A : Matrix (Fin m) (Fin n) ℝ) (b : Fin m → ℝ)
(Nprime : Finset (Fin n)) (hsys : IsMixedZeroOneSystem A b Nprime) :
HasRankAtMost SplitConvexify (Poly A b) Nprime Nprime.card ∧
HasRankAtMost SimpleDisjClosureOfSet (Poly A b) Nprime Nprime.card ∧
HasRankAtMost (fun S k => StrengthenedLPClosureOfSet S k Nprime) (Poly A b) Nprime
Nprime.card ∧
HasRankAtMost (fun S k => MIGClosureOfSet S k Nprime) (Poly A b) Nprime Nprime.card := by sorry
end Disjunctive.CutCorrespondence