Theorem 2.1 (i), (iii) — the discounted cost optimality equation and a discount optimal stationary policy (countable states)
ProvedArapostathisAC.SennottACOI.theorem_2_1Consider the countable-state controlled Markov process of §5: states , nonempty compact admissible action sets , a nonnegative cost and transition probabilities , both continuous in . Let and let be the optimal discounted cost over all admissible policies. Then:
- the discounted cost optimality equation (DCOE) holds: for every ,
- there is a stationary deterministic policy that is -discount optimal, i.e. for every .
This is the countable-state instance of the paper's Theorem 2.1 (i) and (iii); the paper states it for Borel state spaces under Assumptions 2.1–2.3, which hold in the §5 model (discrete state space, compact , continuity of and ). It provides the -discount optimal policies from which the proof of Theorem 5.9 starts.
Formalization Note Both sides of the DCOE are computed in , so the equation is meaningful even where is infinite; the integral of the paper is the series on the countable state space. Part (ii) of the paper's theorem (characterization of discount optimal policies) is not part of this item.
import Mathlib import Definitions.Def_ArapostathisAC_SennottACOI_CMP open MeasureTheory ProbabilityTheory Filter Topology open scoped ENNReal
namespace ArapostathisAC.SennottACOI
variable {A : Type*} [MetricSpace A] [MeasurableSpace A] [BorelSpace A]
/-- Theorem 2.1 (i), (iii) (p. 289) for the countable-state model of §5: for every discount
factor `β ∈ (0, 1)` the discounted value function satisfies the discounted cost optimality
equation (2.7), `J*_β(i) = inf_{a ∈ U(i)} {c(i, a) + β Σ_j P(j | i, a) J*_β(j)}`, computed in
`[0, ∞]`, and a `β`-discount optimal stationary deterministic policy exists. -/
theorem theorem_2_1 (M : ArapostathisAC.VanishingDiscount.CMP A) (β : ℝ) (hβ : β ∈ Set.Ioo (0 : ℝ) 1) :
(∀ i, ArapostathisAC.VanishingDiscount.discValue M β i =
⨅ a ∈ M.U i, (ENNReal.ofReal (M.c i a) +
ENNReal.ofReal β * ∑' j, M.P (i, a) {j} * ArapostathisAC.VanishingDiscount.discValue M β j)) ∧
∃ f : StationaryPolicy M, ArapostathisAC.VanishingDiscount.IsDiscOptimal M β f := by sorry
end ArapostathisAC.SennottACOI
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.