Proposition 6.3(a) — a uniformly -stage optimal policy exists iff the DP infima are attained
ProvedBertsekasShreve.Generalized.prop6_3a_uniform_optimal_iff_attaineddynamic-programmingfinite-horizonp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-bookp2o-v1
Consider the generalized abstract model of Section 6.1 under conditions A.1–A.4 and the exact selection assumption, and let be a positive integer. There exists a uniformly -stage optimal policy in if and only if the infimum in
is attained (by some ) for each and .
The exact selection assumption is what allows pointwise minimizers to be assembled into a selector in the restricted class .
Preamble
import Mathlib import Definitions.Def_BertsekasShreve_Generalized_Model import Definitions.Def_BertsekasShreve_Generalized_Assumptions import Definitions.Def_BertsekasShreve_Generalized_Optimality
Formal statement
namespace BertsekasShreve.Generalized
open Filter Topology
/-- Proposition 6.3(a), p. 96. Under A.1–A.4 and the exact selection assumption, a uniformly
`N`-stage optimal policy exists if and only if the infimum in
`T^{k+1}(J₀)(x) = inf_{u ∈ U(x)} H[x, u, T^k(J₀)]` is attained for each `x ∈ S` and
`k = 0, 1, …, N − 1`. `N` is a positive integer (p. 28). -/
theorem prop6_3a_uniform_optimal_iff_attained {S C : Type*} (P : Model S C)
(hA1 : P.A1) (hA2 : P.A2) (hA3 : P.A3) (hA4 : P.A4) (hES : P.ExactSelection)
(N : ℕ) (hN : 0 < N) :
(∃ π : P.Policy, P.IsUniformlyNStageOptimal N π) ↔
∀ x, ∀ k, k < N → ∃ u ∈ P.U x, P.H x u (P.T^[k] P.J0) = P.T^[k + 1] P.J0 x := by sorry
end BertsekasShreve.Generalized
Source
Bertsekas & Shreve, Stochastic Optimal Control: The Discrete-Time Case, Athena Scientific 1996, p. 96, Proposition 6.3(a)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.