Theorem 7.5.6 (Weak Duality) — the primal and dual programs are feasible and sandwich
ProvedMDPFinance.LPDuality.theorem_7_5_6Both the primal program and its dual have feasible points — a weighted multiple of the bounding function solves 's constraints, and any policy's occupation measure solves 's — and, as in classical finite-dimensional linear programming, every dual feasible value is bounded above by every primal feasible value, with both optimal values finite. This is the first genuine bridge between the Markov Decision Problem and a pair of infinite-dimensional linear programs.
Moderation note. §7.5.2's standing setting is stated: a closed linear subspace of (IsClosedSubspaceOf) with , besides the contracting model with and ; the draft's weak duality had an arbitrary set of functions.
import Mathlib import Definitions.Def_MDPFinance_LPDuality_Model import Definitions.Def_MDPFinance_LPDuality_Bounding import Definitions.Def_MDPFinance_LPDuality_LP open MeasureTheory ProbabilityTheory
namespace MDPFinance.LPDuality
/-- Theorem 7.5.6 (Weak Duality) (Bäuerle–Rieder, p. 216, PDF 227). The feasible sets `Z_P` and
`Z_D` are nonempty and it holds: `-\infty < val(D) \le val(P) < \infty`. Setting of §7.5.2: a
contracting model with a bounding function `b ≥ 1`, `∫ b dp < ∞`, and `IM` a closed linear
subspace of `IB_b` with `b ∈ IM`. -/
theorem theorem_7_5_6 {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 < ⊤) :
(ZP M IMs).Nonempty ∧ (ZD M b IMs p).Nonempty ∧
(⊥ : EReal) < valD M b IMs p ∧ valD M b IMs p ≤ valP M IMs p ∧
valP M IMs p < (⊤ : EReal) := 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.