Proof of Theorem 5.9 — the limit (ρ*, h, f) satisfies the average cost optimality inequality
ProvedArapostathisAC.SennottACOI.acoi_limitIn the countable-state controlled Markov process of §5, assume Assumptions 5.14 and 5.16, and Assumption 5.15 with the constant . Let be a sequence in with , let be -discount optimal, and suppose that , and satisfy, for every state ,
Then for every state the series converges and
This is the step of the proof of Theorem 5.9 that passes to the limit in (5.15): the pair satisfies the average cost optimality inequality (ACOI) with the action on the right-hand side.
Formalization Note The limits are those produced by the previous milestone, written as hypotheses. The continuity of and on is part of the model. Convergence of the series is part of the conclusion. The sequence is not required to be increasing.
import Mathlib import Definitions.Def_ArapostathisAC_SennottACOI_CMP import Definitions.Def_ArapostathisAC_SennottACOI_Assumptions open MeasureTheory ProbabilityTheory Filter Topology open scoped ENNReal
namespace ArapostathisAC.SennottACOI
variable {A : Type*} [MetricSpace A] [MeasurableSpace A] [BorelSpace A]
/-- Proof of Theorem 5.9 (p. 308), the display after "we conclude that": under Assumptions
5.14–5.16 (with the constant `L` of Assumption 5.15 named), let `β_n ∈ (0, 1)` with `β_n → 1`, let
`f_{β_n} ∈ Π_SD` be `β_n`-discount optimal, and suppose `f_{β_n}(i) → f(i)` and
`h_{β_n}(i) → h(i)` for every state `i`, and `(1 − β_n) J*_{β_n}(0) → ρ*`. Then for every state `i`
the series `Σ_j P(j | i, f(i)) h(j)` converges and the average cost optimality inequality
`ρ* + h(i) ≥ c(i, f(i)) + Σ_j P(j | i, f(i)) h(j)` holds. -/
theorem acoi_limit (M : ArapostathisAC.VanishingDiscount.CMP A) (h14 : Assumption5_14 M) (L : ℕ)
(hL : ∀ β ∈ Set.Ioo (0 : ℝ) 1, ∀ i, -(L : ℝ) ≤ ArapostathisAC.VanishingDiscount.hRel M β i) (h16 : Assumption5_16 M)
(β : ℕ → ℝ) (hβ : ∀ n, β n ∈ Set.Ioo (0 : ℝ) 1) (hβ1 : Tendsto β atTop (𝓝 1))
(fβ : ℕ → StationaryPolicy M) (hfβ : ∀ n, ArapostathisAC.VanishingDiscount.IsDiscOptimal M (β n) (fβ n))
(f : StationaryPolicy M) (h : ℕ → ℝ) (ρ : ℝ)
(hf : ∀ i, Tendsto (fun n => (fβ n).1 i) atTop (𝓝 (f.1 i)))
(hh : ∀ i, Tendsto (fun n => ArapostathisAC.VanishingDiscount.hRel M (β n) i) atTop (𝓝 (h i)))
(hρ : Tendsto (fun n => (1 - β n) * (ArapostathisAC.VanishingDiscount.discValue M (β n) 0).toReal) atTop (𝓝 ρ))
(i : ℕ) :
Summable (fun j => ArapostathisAC.VanishingDiscount.prob M i (f.1 i) j * h j) ∧
M.c i (f.1 i) + ∑' j, ArapostathisAC.VanishingDiscount.prob M i (f.1 i) j * h j ≤ ρ + h i := by sorry
end ArapostathisAC.SennottACOI
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.