Lemma (p. 628) — discount optimal stationary policies along α_n → 1 have a subsequence with a stationary limit point
ProvedSennott1989.AvgCost.lemma_limit_pointdiscounted-costmarkov-decision-processp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1stationary-policy
Consider a Markov decision chain on the states with finite action sets. Let be a sequence of discount factors in converging to , and for each let be a stationary policy determined by the right side of the discount optimality equation
i.e. attains the minimum at every . Then there exist a subsequence of and a stationary policy that is a limit point of : for every state there is an integer with
The policy is the candidate for average cost optimality in the paper's Theorem.
Formalization Note The conclusion is IsLimitPoint of the published Sennott (1999) development: a strictly increasing index map with for all large , for each .
Preamble
import Mathlib import Definitions.Def_Sennott1989_AvgCost_Assumptions open scoped ENNReal NNReal open Filter Topology
Formal statement
namespace Sennott1989.AvgCost
open SennottDP.Discounted
/-- Sennott (1989), §2, Lemma (unnumbered), p. 628. Let `α_n` be any sequence of discount factors
in `(0, 1)` converging to `1`, and let `f_{α_n}` be the associated discount optimal stationary
policies, i.e. `f_{α_n}` realizes the minimum in the discount optimality equation (2) at
`α = α_n`. Then there exist a subsequence `(β_n)` of `(α_n)` and a stationary policy `f` that is a
limit point of `(f_{β_n})`: for every state `i` there is an integer `N(i)` with
`f_{β_n}(i) = f(i)` for all `n ≥ N(i)`.
**Formalization Note** "There exist a subsequence and a limit point `f` of `(f_{β_n})`" is
`IsLimitPoint M fs f`: there is a strictly increasing `r : ℕ → ℕ` with `fs (r k) i = f i` for all
sufficiently large `k`, for each `i`. -/
theorem lemma_limit_point {Act : Type} (M : MDC ℕ Act) (αs : ℕ → ℝ≥0)
(hα0 : ∀ n, 0 < αs n) (hα1 : ∀ n, αs n < 1) (hlim : Tendsto αs atTop (𝓝 1))
(fs : ℕ → StationaryPolicy M)
(hfs : ∀ n, Realizes M (fs n) (αs n) (valueFn M (αs n))) :
∃ f : StationaryPolicy M, IsLimitPoint M fs f := by sorry
end Sennott1989.AvgCost
Source
Sennott, Average Cost Optimal Stationary Policies in Infinite State Markov Decision Processes with Unbounded Costs, Oper. Res. 37(4):626–633 (1989), DOI 10.1287/opre.37.4.626, §2, Lemma (unnumbered), p. 628
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.