Theorem 4.3 — the Lagrangian dual bound does not exceed
ProvedShorNonsmooth.Decomposition.dual_bound_le_optimumConsider the problem (4.185)–(4.186): over subject to , , where is compact. Let
and . If is the optimum value of (4.185)–(4.186), then
is the lower estimate of used in branch-and-bound methods for discrete and mixed discrete-continuous programs (where is partially discrete); computing it is a nonsmooth concave maximization.
Formalization Note The optimum is attained (a least element of on the feasible part of ), and is assumed to be attained, as the book's "max" presupposes. The minimum in (4.187) is assumed to be attained for every , as the book's "min" presupposes (no continuity is assumed; for a partially discrete the functions need not be continuous). is an arbitrary compact set; its partial discreteness plays no role.
import Mathlib import Definitions.Def_ShorNonsmooth_Decomposition_PenaltyDual
namespace ShorNonsmooth.Decomposition
/-- Shor (1985), **Theorem 4.3** (p. 148): for problem (4.185)–(4.186), `min f₀(x)` over `x ∈ X ⊂ E_N`
subject to `f_i(x) ≤ 0`, with `X` compact, let `Φ(u) = min_{x ∈ X}[f₀(x) + Σ u_i f_i(x)]` (4.187) and
`Q = max_{u ≥ 0} Φ(u)`. If `f*` is the optimum value of (4.185)–(4.186), then `Q ≤ f*`.
The minimum in (4.187) is assumed to be attained for every `u ≥ 0` (the book writes `min`), the optimum
`f*` is attained, and `Q` is assumed to be attained, as the book's `max` presupposes. -/
theorem dual_bound_le_optimum {N m : ℕ}
(X : Set (EuclideanSpace ℝ (Fin N))) (hX : IsCompact X)
(f₀ : EuclideanSpace ℝ (Fin N) → ℝ) (f : Fin m → EuclideanSpace ℝ (Fin N) → ℝ)
(hmin : ∀ u : Fin m → ℝ, (∀ i, 0 ≤ u i) →
∃ x ∈ X, ∀ x' ∈ X, f₀ x + ∑ i, u i * f i x ≤ f₀ x' + ∑ i, u i * f i x')
(fstar : ℝ) (hfstar : IsLeast (f₀ '' (X ∩ feasibleSet f)) fstar)
(Q : ℝ) (hQ : IsGreatest (dualFn X f₀ f '' {u | ∀ i, 0 ≤ u i}) Q) :
Q ≤ fstar := by sorry
end ShorNonsmooth.Decomposition
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.