Lemma 8.4 — monotone and bounded convergence for the operator T_μ
ProvedBertsekasShreve.BorelFinite.lemma8_4_Tmu_limitsdynamic-programmingmonotone-convergencep2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-bookp2o-v1
Consider the finite horizon Borel model of Definition 8.1. Let be a sequence of extended real-valued, universally measurable functions on , let , and let .
- If for every and , then .
- If for every , , and , then .
- If is uniformly bounded, is bounded, and , then .
Here , the arrows denote monotone pointwise convergence, and all limits are pointwise in :
These continuity properties let infima over policies be exchanged with the operator in the dynamic programming recursion.
Formalization Note The sequence is indexed from , so the book's is Js 0. The conditions and " bounded" are imposed on , where is defined. " uniformly bounded" means for one real , all and .
Preamble
import Mathlib import Definitions.Def_BertsekasShreve_BorelFinite_Model import Definitions.Def_BertsekasShreve_BorelFinite_Policy import Definitions.Def_BertsekasShreve_BorelFinite_Operators
Formal statement
namespace BertsekasShreve.BorelFinite
open Filter Topology in
/-- **Lemma 8.4** (p. 197). Let `{J_k}` be universally measurable functions `S → R*` and
`μ ∈ U(C|S)` (the book's `J₁` is `Js 0` here).
(a) If `T_μ(J₁)(x) < ∞` for every `x` and `J_k ↓ J`, then `T_μ(J_k) ↓ T_μ(J)`.
(b) If `T_μ(J₁⁻)(x) < ∞` for every `x`, `g ≥ 0`, and `J_k ↑ J`, then `T_μ(J_k) ↑ T_μ(J)`.
(c) If `{J_k}` is uniformly bounded, `g` is bounded, and `J_k → J`, then `T_μ(J_k) → T_μ(J)`. -/
theorem lemma8_4_Tmu_limits {S C W : Type*}
[TopologicalSpace S] [MeasurableSpace S] [BorelSpace S] [BertsekasShreve.AnalyticSelection.IsBorelSpace S] [Nonempty S]
[TopologicalSpace C] [MeasurableSpace C] [BorelSpace C] [BertsekasShreve.AnalyticSelection.IsBorelSpace C] [Nonempty C]
[TopologicalSpace W] [MeasurableSpace W] [BorelSpace W] [BertsekasShreve.AnalyticSelection.IsBorelSpace W] [Nonempty W]
(M : Model S C W)
(μ : UCS M) (Js : ℕ → S → EReal) (Jl : S → EReal) (hJs : ∀ k, UMeasurable (Js k)) :
((∀ x, Tmu M μ (Js 0) x < ⊤) → (∀ x, Antitone fun k => Js k x) →
(∀ x, Tendsto (fun k => Js k x) atTop (𝓝 (Jl x))) →
(∀ x, Antitone fun k => Tmu M μ (Js k) x) ∧
∀ x, Tendsto (fun k => Tmu M μ (Js k) x) atTop (𝓝 (Tmu M μ Jl x))) ∧
((∀ x, Tmu M μ (fun y => ((negPart (Js 0) y : ENNReal) : EReal)) x < ⊤) →
(∀ z ∈ M.Γ, 0 ≤ M.g z) → (∀ x, Monotone fun k => Js k x) →
(∀ x, Tendsto (fun k => Js k x) atTop (𝓝 (Jl x))) →
(∀ x, Monotone fun k => Tmu M μ (Js k) x) ∧
∀ x, Tendsto (fun k => Tmu M μ (Js k) x) atTop (𝓝 (Tmu M μ Jl x))) ∧
((∃ b : ℝ, ∀ k x, Js k x ≤ b ∧ -(b : EReal) ≤ Js k x) →
(∃ b : ℝ, ∀ z ∈ M.Γ, M.g z ≤ b ∧ -(b : EReal) ≤ M.g z) →
(∀ x, Tendsto (fun k => Js k x) atTop (𝓝 (Jl x))) →
∀ x, Tendsto (fun k => Tmu M μ (Js k) x) atTop (𝓝 (Tmu M μ Jl x))) := by sorry
end BertsekasShreve.BorelFinite
Source
Bertsekas & Shreve, Stochastic Optimal Control: The Discrete-Time Case, Athena Scientific 1996, p. 197, Lemma 8.4
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.