Proposition 5 — strong duality and a primal optimizer on compact with continuous cost
OpenModelRiskOT.Duality.proposition_5Let be a compact Polish space, a Borel probability measure on , a cost satisfying (A1) that is in addition continuous on , upper semicontinuous and -integrable (A2), and . Then
and there is a primal optimizer with .
This is the first step of the proof of Theorem 1, obtained from Fenchel duality on .
Formalization Note The space is a Polish space with its Borel σ-algebra; the cost is real-valued and written curried, ; (A1) is the structure AssumptionA1; (A2) is the pair of hypotheses UpperSemicontinuous f and Integrable f μ. Values that can be infinite (, , , , ) live in EReal; an integral of an extended-real function is with lower Lebesgue integrals, and evaluates to , so a coupling with never raises the primal supremum (the paper's footnote 2 reading).
import Mathlib import Definitions.Def_ModelRiskOT_Duality_AssumptionA1 import Definitions.Def_ModelRiskOT_Duality_primalValue import Definitions.Def_ModelRiskOT_Duality_dualValue open MeasureTheory
namespace ModelRiskOT.Duality
/-- **Proposition 5** (Blanchet & Murthy, arXiv:1604.01446v2, §4.1, p. 17). Let `S` be a compact
Polish space, `c` satisfy (A1) and in addition be continuous, and `f` satisfy (A2). Then `I = J`,
and a primal optimizer `π* ∈ Φ_{μ,δ}` with `I(π*) = I` exists. -/
theorem proposition_5 {S : Type*} [TopologicalSpace S] [PolishSpace S] [MeasurableSpace S] [BorelSpace S] [CompactSpace S]
(μ : Measure S) [IsProbabilityMeasure μ] (c : S → S → ℝ) (hc : AssumptionA1 c)
(hc_cont : Continuous (fun p : S × S => c p.1 p.2))
(f : S → ℝ) (hf_usc : UpperSemicontinuous f) (hf_int : Integrable f μ)
(δ : ℝ) (hδ : 0 < δ) :
primalValue c f μ δ = dualValue c f μ δ ∧
∃ π ∈ primalFeasible c μ δ, primalObj f π = primalValue c f μ δ := by sorry
end ModelRiskOT.Duality