Theorem 7.5.7 — complementary slackness for the primal/dual pair
ProvedMDPFinance.LPDuality.theorem_7_5_7A feasible primal and dual are simultaneously optimal, with matching primal and dual values, exactly when the primal constraint they are both tied to holds with equality -almost everywhere — the infinite-dimensional analogue of complementary slackness in ordinary linear programming, and the key technical tool the strong duality theorem (the goal) uses to certify optimality of the pair it exhibits.
Moderation note. The draft had no hypotheses at all (no bounding function, no contracting model, no setting for and ), under which the integrals defining and the values are default values; §7.5.2's standing setting is now stated as in Theorem 7.5.6.
import Mathlib import Definitions.Def_MDPFinance_LPDuality_Model import Definitions.Def_MDPFinance_LPDuality_LP open MeasureTheory ProbabilityTheory
namespace MDPFinance.LPDuality
/-- Theorem 7.5.7 (Bäuerle–Rieder, p. 217, PDF 228). Let `v \in Z_P` and `\mu \in Z_D`. Then the
following statements are equivalent: (i) `v` and `\mu` are optimal and `val(P) = val(D)`. (ii)
`v(x) = Lv(x,a)` for `\mu`-almost all `(x,a) \in D`. Setting of §7.5.2 as in Theorem 7.5.6. -/
theorem theorem_7_5_7 {E A : Type*} [MeasurableSpace E] [MeasurableSpace A] (M : MarkovDecisionModel E A)
(b : E → ℝ) (cr αb : ℝ) (hb : IsBoundingFunction M b cr αb) (hb1 : ∀ x, 1 ≤ b x)
(hαb : M.β * αb < 1) (IMs : Set (E → ℝ)) (hIM : IsClosedSubspaceOf b IMs) (hbIM : b ∈ IMs)
(p : Measure E) (hpb : ∫⁻ x, ENNReal.ofReal (b x) ∂p < ⊤) (v : E → ℝ) (hv : v ∈ ZP M IMs)
(μ : Measure (E × A)) (hμ : μ ∈ ZD M b IMs p) :
(((∫ x, v x ∂p : ℝ) : EReal) = valP M IMs p ∧
((∫ xa, M.r xa ∂μ : ℝ) : EReal) = valD M b IMs p ∧ valP M IMs p = valD M b IMs p) ↔
(∀ᵐ xa ∂μ, xa ∈ M.D → v xa.1 = M.r xa + M.β * ∫ x', v x' ∂(M.Q xa)) := by sorry
end MDPFinance.LPDuality
Confirmed by the mission captain (proposal self-audit).
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.