Theorem 10.2.3 — existence of an optimal stopping region via the limit of the d_n
DisprovedMDPFinance.OptimalStopping.theorem_10_2_3Theorem 10.2.3 (p. 313). Suppose a stopping problem with unbounded horizon is given. Then it holds:
a) The limit exists. Define ; then and is a maximizer of . b) If and for all , then is an optimal stopping time and , .
The route to an optimal rule that Theorem 10.2.2 does not give: 10.2.2 characterizes the value, 10.2.3 produces the stopping region. The limit exists because the are monotone, which is Theorem 10.1.5 b) in another guise.
b) is conditional on two hypotheses and both are the book's. says the rule's own value is -superharmonic; says the rule actually stops. Drop either and the conclusion fails: a rule whose entry time is infinite with positive probability collects the reward, which the book leaves undefined.
The triple equality is what "optimal" means here, so it is stated as attainment — the expected reward is the least upper bound — not as an inequality, which every stopping time satisfies by definition of the supremum.
Moderation note. As for Theorem 10.2.2 (J now the model's limit); b)'s conclusion uses the book's optimality notion (IsOptimal: stopping time, a.s. finite, attains V_∞^*) together with G_{f^*} = 𝔼_x[R_{τ^*}].
import Mathlib import Definitions.Def_MDPFinance_OptimalStopping_Stationary open MeasureTheory Filter Topology
namespace MDPFinance.OptimalStopping
/-- **Theorem 10.2.3** (p. 313), under (B). a) `d := lim_n d_n` exists; with `S^* := {d ≥ 0}`,
`S^* = ∩_n S_n^* = {J = g}` and `f^* := 1_{S^*}` is a maximizer of `J`. b) If `G_{f^*} ≥ T G_{f^*}`
and `ℙ_x(τ_{f^*} < ∞) = 1` for all `x`, then `τ^* := inf{n | X_n ∈ S^*}` is an optimal stopping
time and `G_{f^*}(x) = 𝔼_x[R_{τ^*}] = V_∞^*(x)`. -/
theorem theorem_10_2_3 {E : Type*} [MeasurableSpace E] (P : StationaryProblem E)
(Pr : E → Measure (ℕ → E)) (hPr : P.IsPathLaw Pr) (hB : P.AssumptionB Pr) :
(∃ d : E → EReal,
(∀ x : E, Tendsto (fun n => P.d n x) atTop (𝓝 (d x))) ∧
({x : E | 0 ≤ d x} = ⋂ n : ℕ, P.stopSet n) ∧
({x : E | 0 ≤ d x} = {x : E | P.Jlim x = (P.g x : EReal)}) ∧
P.IsMaximizer {x : E | 0 ≤ d x} P.Jlim) ∧
((∀ x : E, P.T (P.Gf Pr {y : E | P.Jlim y = (P.g y : EReal)}) x ≤
P.Gf Pr {y : E | P.Jlim y = (P.g y : EReal)} x) →
(∀ x : E, Pr x {w | P.ruleTime {y : E | P.Jlim y = (P.g y : EReal)} w = ⊤} = 0) →
P.IsOptimal Pr (P.ruleTime {y : E | P.Jlim y = (P.g y : EReal)}) ∧
∀ x : E, P.Gf Pr {y : E | P.Jlim y = (P.g y : EReal)} x =
P.EReward Pr (P.ruleTime {y : E | P.Jlim y = (P.g y : EReal)}) x) := by sorry
end MDPFinance.OptimalStopping
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.