Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 10.2.3 — existence of an optimal stopping region via the limit of the d_n

Disproved
MDPFinance.OptimalStopping.theorem_10_2_3

by Shuze Chen · Sep 27, 2026 · Mathlib 0df444a (Lean v4.33.1)

dynamic-programmingoperations-researchoptimal-stoppingoptimizationprobability

Theorem 10.2.3 (p. 313). Suppose a stopping problem with unbounded horizon is given. Then it holds:

a) The limit d:=lim⁡n→∞dnd := \lim_{n\to\infty} d_nd:=limn→∞​dn​ exists. Define S∗:={x∈E∣d(x)≥0}S^* := \{x \in E \mid d(x) \ge 0\}S∗:={x∈E∣d(x)≥0}; then S∗=⋂nSn∗={x∈E∣J(x)=g(x)}S^* = \bigcap_n S_n^* = \{x \in E \mid J(x) = g(x)\}S∗=⋂n​Sn∗​={x∈E∣J(x)=g(x)} and f∗:=1S∗f^* := \mathbf 1_{S^*}f∗:=1S∗​ is a maximizer of JJJ. b) If Gf∗≥TGf∗G_{f^*} \ge \mathcal{T}G_{f^*}Gf∗​≥TGf∗​ and Px(τf∗<∞)=1\mathbb{P}_x(\tau_{f^*} < \infty) = 1Px​(τf∗​<∞)=1 for all x∈Ex \in Ex∈E, then τ∗:=inf⁡{n∈N0∣Xn∈S∗}\tau^* := \inf\{n \in \mathbb{N}_0 \mid X_n \in S^*\}τ∗:=inf{n∈N0​∣Xn​∈S∗} is an optimal stopping time and Gf∗(x)=Ex[Rτ∗]=V∞∗(x)G_{f^*}(x) = \mathbb{E}_x[R_{\tau^*}] = V_\infty^*(x)Gf∗​(x)=Ex​[Rτ∗​]=V∞∗​(x), x∈Ex \in Ex∈E.

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 dnd_ndn​ are monotone, which is Theorem 10.1.5 b) in another guise.

b) is conditional on two hypotheses and both are the book's. Gf∗≥TGf∗G_{f^*} \ge \mathcal{T}G_{f^*}Gf∗​≥TGf∗​ says the rule's own value is ccc-superharmonic; Px(τf∗<∞)=1\mathbb{P}_x(\tau_{f^*} < \infty) = 1Px​(τf∗​<∞)=1 says the rule actually stops. Drop either and the conclusion fails: a rule whose entry time is infinite with positive probability collects the τ=∞\tau = \inftyτ=∞ reward, which the book leaves undefined.

The triple equality Gf∗=Ex[Rτ∗]=V∞∗G_{f^*} = \mathbb{E}_x[R_{\tau^*}] = V_\infty^*Gf∗​=Ex​[Rτ∗​]=V∞∗​ 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^*}].

Preamble
import Mathlib
import Definitions.Def_MDPFinance_OptimalStopping_Stationary

open MeasureTheory Filter Topology
Formal statement
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
Source
Bäuerle and Rieder, Markov Decision Processes with Applications to Finance, Universitext, Springer 2011, DOI 10.1007/978-3-642-18324-9, p. 313 (PDF 320), Theorem 10.2.3
Human review
  • Endorsed by Community (Bot) · Oct 1, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Shuze Chen · Oct 1, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me