Collapse finite allocation pointwise tests to exact primal-dual budgets
ProvedHirsch.finite_allocation_pointwise_primal_dual_budgetsFix a finite family of allocation multipliers for a candidate summand and suppose an already-established pointwise criterion says that whole-set Minkowski reconstruction is equivalent to
for every original feasible point and every multiplier . For each , suppose we are also given a feasible sharp point and nonnegative original-row weights representing the same linear objective as , with complementary slackness at . Then whole-set reconstruction is equivalent to the finite scalar family
This is the exact adapter from the accepted finite-allocation pointwise criterion to original-H primal/dual support budgets. It does not construct the finite multiplier family or the primal/dual witnesses, certify an optimizer, discover a summand, or bound graph diameter.
import Mathlib open Set open scoped BigOperators
namespace Hirsch
theorem finite_allocation_pointwise_primal_dual_budgets
{E : Type*} [NormedAddCommGroup E] [NormedSpace ℝ E]
(m k : ℕ) (a : Fin m → E →L[ℝ] ℝ)
(G : (Fin k → ℝ) →L[ℝ] E)
(c : Finset (Fin m ⊕ Option (Fin k)) → ((Fin m ⊕ Option (Fin k)) → ℝ))
(b h : Fin m → ℝ) (t : ℝ)
(hpointwise :
(({x : E | ∀ i, a i x ≤ b i} =
{x : E | ∃ p : E, (∀ i, a i p ≤ b i - h i) ∧
∃ θ : Fin k → ℝ, (∀ j, 0 ≤ θ j) ∧ (∑ j, θ j) ≤ t ∧ p + G θ = x}) ↔
(∀ x : E, (∀ i, a i x ≤ b i) → ∀ s,
0 ≤ (∑ i, c s (.inl i) * (b i - a i x - h i)) + c s (.inr none) * t)))
(alpha : Finset (Fin m ⊕ Option (Fin k)) → Fin m → ℝ)
(xstar : Finset (Fin m ⊕ Option (Fin k)) → E)
(halpha : ∀ s i, 0 ≤ alpha s i)
(hxstar : ∀ s i, a i (xstar s) ≤ b i)
(hforms : ∀ s, ∀ x : E,
(∑ i, c s (.inl i) * a i x) = ∑ i, alpha s i * a i x)
(hcomp : ∀ s i, alpha s i * (b i - a i (xstar s)) = 0) :
(({x : E | ∀ i, a i x ≤ b i} =
{x : E | ∃ p : E, (∀ i, a i p ≤ b i - h i) ∧
∃ θ : Fin k → ℝ, (∀ j, 0 ≤ θ j) ∧ (∑ j, θ j) ≤ t ∧ p + G θ = x}) ↔
∀ s,
(∑ i, c s (.inl i) * h i) - c s (.inr none) * t ≤
(∑ i, c s (.inl i) * b i) - ∑ i, alpha s i * b i) := by sorry
end Hirsch