Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 2.5.4 — the stationary structure theorem

Proved
MDPFinance.Stationary.structure_theorem_stationary

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

bellman-equationdynamic-programmingmarkov-decision-process

Let (SAN) hold for (I ⁣M,Δ)(\mathrm{I\!M}, \Delta)(IM,Δ). Then, for 0≤n≤N0 \leq n \leq N0≤n≤N: Jn∈I ⁣MJ_n \in \mathrm{I\!M}Jn​∈IM; for 1≤n≤N1 \leq n \leq N1≤n≤N, Jn=TJn−1J_n = T J_{n-1}Jn​=TJn−1​; and Jn=TngJ_n = T^n gJn​=Tng. Moreover, for n=1,…,Nn = 1,\dots,Nn=1,…,N there is a maximizer fn∗∈Δf_n^* \in \Deltafn∗​∈Δ of Jn−1J_{n-1}Jn−1​, and any such family assembles (via π(k):=fN−k∗\pi(k) := f^*_{N-k}π(k):=fN−k∗​ for k=0,…,N−1k=0,\dots,N-1k=0,…,N−1) into a policy attaining JNJ_NJN​.

Formalization Note. The "every sequence of maximizers defines an optimal policy" clause is formalized as: the existence of one maximizer family f1∗,…,fN∗f^*_1,\dots,f^*_Nf1∗​,…,fN∗​ whose associated policy π(k):=fN−k∗\pi(k) := f^*_{N-k}π(k):=fN−k∗​ attains JNJ_NJN​ exactly — a faithful, if index-heavy, translation of the Forward Induction Algorithm's output into a usable NNN-stage policy.

Formalization Note (moderation). Part (b) is stated as the book does: maximizers fn∗∈Δf_n^*\in\Deltafn∗​∈Δ of Jn−1J_{n-1}Jn−1​ exist for every nnn, and every sequence of maximizers defines an optimal policy (fN∗,…,f1∗)(f_N^*,\dots,f_1^*)(fN∗​,…,f1∗​) (a universally quantified statement), not merely the existence of one such family. (AN) is carried as hAN.

Preamble
import Mathlib
import Definitions.Def_MDPFinance_Stationary_Model
import Definitions.Def_MDPFinance_Stationary_Policy
import Definitions.Def_MDPFinance_Stationary_Operators
import Definitions.Def_MDPFinance_Stationary_StructureAssumption
import Definitions.Def_MDPFinance_Stationary_ValueFunction

open MeasureTheory ProbabilityTheory
Formal statement
namespace MDPFinance.Stationary

/-- Theorem 2.5.4 (Structure Theorem) (Bäuerle–Rieder, p. 41, PDF 56), under the standing
Integrability Assumption (AN). Let (SAN) be satisfied.
a) Then `J_n ∈ IM` and the Bellman equation `J_n = T J_{n-1}` holds for `1 ≤ n ≤ N`, i.e.
`J_0 = g`, `J_n(x) = sup_{a ∈ D(x)} [r(x,a) + β ∫ J_{n-1}(x') Q(dx'|x,a)]`. Moreover, `J_n = T^n
g`. b) For `n = 1, …, N` there exist maximizers `f_n^*` of `J_{n-1}` with `f_n^* ∈ Δ`, and every
sequence of maximizers `f_n^*` of `J_{n-1}` defines an optimal policy `(f_N^*, …, f_1^*)` for
the stationary `N`-stage Markov Decision Model: for every such sequence, the policy `π(k) := f^*_{N-k}` for
`k = 0, …, N-1` attains `J_N`. -/
theorem structure_theorem_stationary {E A : Type*} [MeasurableSpace E] [MeasurableSpace A]
    (M : StationaryMarkovDecisionModel E A) (IMs : Set (E → EReal)) (Delta : Set (E → A))
    (hSAN : StructureAssumption M IMs Delta) (N : ℕ) (hAN : IntegrabilityAssumption M N) :
    (∀ n ≤ N, J M n ∈ IMs) ∧
      (∀ n, 1 ≤ n → n ≤ N → J M n = T M (J M (n - 1))) ∧
      (∀ n ≤ N, J M n = TChain M n (fun x => (M.g x : EReal))) ∧
      (∀ n, 1 ≤ n → n ≤ N → ∃ f ∈ Delta, IsMaximizer M (J M (n - 1)) f) ∧
      (∀ fstar : ℕ → E → A, (∀ n, 1 ≤ n → n ≤ N → IsMaximizer M (J M (n - 1)) (fstar n)) →
        Jpi M (fun k => fstar (N - k)) N = J M N) := by sorry

end MDPFinance.Stationary
Source
Bäuerle and Rieder, Markov Decision Processes with Applications to Finance, Universitext, Springer 2011, DOI 10.1007/978-3-642-18324-9, p. 41, PDF 56, Theorem 2.5.4
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).

  • Endorsed by iris · Nov 4, 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