Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 5.6.1 — one-step evaluation preserves bounded values

Proved
PutermanMDP.MarkovPolicies.one_step_bounded

by Shuze Chen · Oct 29, 2026 · Mathlib 0df444a (Lean v4.33.1)

bounded-valuemdpp2o-gran-per-chapterp2o-plan-bookp2o-v1policy-evaluation

Let SSS be a nonempty finite or countable state space. Suppose the real reward satisfies a uniform bound ∣r(s,a)∣≤M|r(s,a)|\le M∣r(s,a)∣≤M for all available actions. Let 0≤β≤10\le\beta\le10≤β≤1, let uuu be a bounded real function on SSS, and let ddd be an admissible randomized Markov decision rule. With rdr_drd​ and PdP_dPd​ its expected reward and transition operator,

rd+βPdu∈V,r_d+\beta P_du\in V,rd​+βPd​u∈V,

where VVV is the space of bounded real functions on SSS.

This establishes that a single policy-evaluation step remains in the value-function space.

Formalization Note Membership in VVV is expressed by existence of a uniform nonnegative real bound; no norm structure is assumed. The endpoint β=1\beta=1β=1 is included, as on the page.

Preamble
import Mathlib
import Definitions.Def_PutermanMDP_MarkovPolicies_RewardCriteria
Formal statement
namespace PutermanMDP.MarkovPolicies

/-- Lemma 5.6.1: the randomized one-step operator maps V into V. -/
theorem one_step_bounded {S A : Type*} [Countable S] [Nonempty S]
    (M : Model S A) (hM : RewardBounded M) (β : ℝ)
    (hβ0 : 0 ≤ β) (hβ1 : β ≤ 1)
    (u : S → ℝ) (hu : BoundedValue u)
    (d : (s : S) → PMF {a : A // a ∈ M.actions s}) :
    BoundedValue (oneStep M d β u) := by sorry

end PutermanMDP.MarkovPolicies
Source
Puterman, Markov Decision Processes, Wiley, 1994, p. 138 (PDF 154), Lemma 5.6.1
Read-back

What the Lean code literally says, in plain math · claude-opus-5-5

Setting and hypotheses.

  • SSS is a countable, nonempty type of states. AAA is an arbitrary type of actions.
  • MMM is a model with nonempty admissible action sets A(s)\mathcal A(s)A(s), transition PMFs p(⋅∣s,a)p(\cdot \mid s,a)p(⋅∣s,a) for a∈A(s)a \in \mathcal A(s)a∈A(s), and real rewards r(s,a)r(s,a)r(s,a).
  • MMM has bounded rewards: there is C≥0C \ge 0C≥0 with ∣r(s,a)∣≤C|r(s,a)| \le C∣r(s,a)∣≤C for all sss and all a∈A(s)a \in \mathcal A(s)a∈A(s).
  • β\betaβ is a real number with 0≤β≤10 \le \beta \le 10≤β≤1.
  • u:S→Ru : S \to \mathbb Ru:S→R is bounded: there is C′≥0C' \ge 0C′≥0 with ∣u(s)∣≤C′|u(s)| \le C'∣u(s)∣≤C′ for all sss.
  • ddd is any rule giving, for each state sss, a PMF d(⋅∣s)d(\cdot \mid s)d(⋅∣s) on A(s)\mathcal A(s)A(s).

Conclusion. The function

s  ↦  ∑a∈A(s)′d(a∣s)(r(s,a)+β∑j∈S′p(j∣s,a) u(j))s \;\mapsto\; \sum'_{a \in \mathcal A(s)} d(a \mid s)\Big( r(s,a) + \beta \sum'_{j \in S} p(j \mid s,a)\, u(j) \Big)s↦a∈A(s)∑′​d(a∣s)(r(s,a)+βj∈S∑′​p(j∣s,a)u(j))

is bounded. That is, there is some K≥0K \ge 0K≥0 with absolute value at most KKK at every state. Probabilities are converted to real numbers. Each ∑′\sum'∑′ is an unconditional infinite sum, defined to be 000 when its family is not summable. The conclusion is about the function computed with that convention. The statement gives no explicit value or formula for KKK.

Degenerate cases. Both endpoints β=0\beta = 0β=0 and β=1\beta = 1β=1 are allowed. At β=0\beta = 0β=0 the function is ∑a′d(a∣s) r(s,a)\sum'_a d(a \mid s)\, r(s,a)∑a′​d(a∣s)r(s,a). SSS may have a single element, and AAA is necessarily nonempty. ddd does not depend on time. No assumption is made on ddd beyond being a PMF on admissible actions at each state.

Human review
  • Endorsed by Community (Bot) · Oct 29, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Shuze Chen · Oct 29, 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