Proposition 3 — a fractional basic solution of , covers some pair of rows fractionally
ProvedLubbecke2005.RyanFoster.exists_fractional_row_pairThis is the Ryan–Foster lemma behind branch-and-price for set-partitioning problems.
Let be a matrix with rows and columns indexed by a finite set . Let be a basic feasible solution of the system
in the sense of Bertsimas and Tsitsiklis (Definition 2.9): satisfies all constraints, and among the constraints active at (all equality rows, and those with ) there are linearly independent ones. Suppose is fractional, . Then there exist rows such that
For the sum is the row sum , so the two rows found are necessarily distinct. The pair is what Ryan–Foster branching branches on: one branch requires the rows to be covered by the same column (the sum equals ), the other by two distinct columns (the sum equals ), and the current fractional solution satisfies neither. The paper states the proposition and attributes it to Ryan and Foster (1981) without proof.
Formalization Note The paper writes "i.e., "; since has one coordinate per column, this statement reads it as (a typo correction). "Basic solution" is not defined in the paper; it is read as the textbook notion for the standard-form system, via the platform definitions LinearOptimization.stdFormSystem and LinearOptimization.IsBasicFeasibleSolution, which need no full-row-rank assumption on . The solution is any fractional basic solution, not necessarily an optimal one. Rows are Fin m, columns Fin n, is a real matrix with the property as a hypothesis, and the quantity is pairCover A lam r s. For or no fractional basic solution exists, so the statement is vacuous there, as on the page.
import Mathlib import Definitions.Def_ActiveConstraints import Definitions.Def_BasicSolution import Definitions.Def_Lubbecke2005_RyanFoster_SetPartitioning open Matrix
namespace Lubbecke2005.RyanFoster
/-- **Proposition 3** (Lübbecke–Desrosiers 2005, §7.3, p. 1020; Ryan and Foster 1981).
Let `A` be an `m × n` matrix with entries in `{0, 1}` (the columns are indexed by
`J′ = Fin n`) and let `λ` be a basic feasible solution of the standard-form system
`Aλ = 𝟏, λ ⩾ 𝟎` (Bertsimas–Tsitsiklis Definition 2.9) which is fractional, i.e.
`λ ∉ {0, 1}^{|J′|}` (the paper's `{0, 1}^m` is a typo: `λ` has one coordinate per
column). Then some rows `r, s` satisfy `0 < ∑_j a_rj a_sj λ_j < 1`. -/
theorem exists_fractional_row_pair {m n : ℕ} (A : Matrix (Fin m) (Fin n) ℝ)
(hA : IsZeroOneMatrix A) (lam : Fin n → ℝ)
(hbasic : LinearOptimization.IsBasicFeasibleSolution
(LinearOptimization.stdFormSystem A (fun _ => 1)) lam)
(hfrac : ¬ IsZeroOneVector lam) :
∃ r s : Fin m, 0 < pairCover A lam r s ∧ pairCover A lam r s < 1 := by sorry
end Lubbecke2005.RyanFoster
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.