Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 4(a) — V_β(f) = x(f)/(1 − β) + y(f) + ε(β, f) with ε(β, f) → 0 as β → 1

Proved
BlackwellDiscreteDP.NearOne.theorem_4a

by mikedeng1 · Oct 4, 2026 · Mathlib 0df444a (Lean v4.33.1)

average-rewardbiaslaurent-expansionmarkov-decision-processp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1

In the finite decision model, take any decision rule f∈Ff\in Ff∈F and let Q∗(f)Q^*(f)Q∗(f) be the limit matrix of Q(f)Q(f)Q(f). Then

Vβ(f(∞))=x(f)1−β+y(f)+ε(β,f),V_\beta(f^{(\infty)})=\frac{x(f)}{1-\beta}+y(f)+\varepsilon(\beta,f),Vβ​(f(∞))=1−βx(f)​+y(f)+ε(β,f),

where x(f)x(f)x(f) is the unique solution of

(I−Q(f))x=0,Q∗(f)x=Q∗(f)r(f),(I-Q(f))x=0,\qquad Q^*(f)x=Q^*(f)r(f),(I−Q(f))x=0,Q∗(f)x=Q∗(f)r(f),

y(f)y(f)y(f) is the unique solution of

(I−Q(f))y=r(f)−x(f),Q∗(f)y=0,(I-Q(f))y=r(f)-x(f),\qquad Q^*(f)y=0,(I−Q(f))y=r(f)−x(f),Q∗(f)y=0,

and ε(β,f)→0\varepsilon(\beta,f)\to0ε(β,f)→0 as β→1\beta\to1β→1.

The pole coefficient x(f)x(f)x(f) is the average income of f(∞)f^{(\infty)}f(∞) and the constant term y(f)y(f)y(f) its bias; all comparisons of stationary policies near β=1\beta=1β=1 in Theorem 4 go through this expansion.

Formalization Note x(f)x(f)x(f) and y(f)y(f)y(f) are defined as Q∗(f)r(f)Q^*(f)r(f)Q∗(f)r(f) and H(f)r(f)H(f)r(f)H(f)r(f); the statement asserts that each is the one and only solution of its system, and that Vβ(f(∞))−x(f)/(1−β)−y(f)→0V_\beta(f^{(\infty)})-x(f)/(1-\beta)-y(f)\to0Vβ​(f(∞))−x(f)/(1−β)−y(f)→0 as β→1\beta\to1β→1 from below.

Preamble
import Mathlib
import Definitions.Def_BlackwellDiscreteDP_NearOne_LimitMatrix
import Definitions.Def_BlackwellDiscreteDP_NearOne_Model
import Definitions.Def_BlackwellDiscreteDP_NearOne_GainBias
open Filter Topology Matrix
Formal statement
namespace BlackwellDiscreteDP.NearOne

/-- **Theorem 4(a).** For any `f ∈ F`,
`V_β(f) = x(f)/(1 − β) + y(f) + ε(β, f)` where `x(f)` is the unique solution of
`(I − Q(f))x = 0, Q*(f)x = Q*(f)r(f)`, `y(f)` is the unique solution of
`(I − Q(f))y = r(f) − x(f), Q*(f)y = 0`, and `ε(β, f) → 0` as `β → 1`.

Blackwell, Discrete Dynamic Programming, Ann. Math. Statist. 33(2):719–726 (1962),
DOI 10.1214/aoms/1177704593, p. 722, Theorem 4(a).

**Formalization Note.** `x(f) := Q*(f)r(f)` and `y(f) := H(f)r(f)` are defined by the closed
forms of the paper's proof; the first two conjuncts say they are exactly the solutions of the two
systems (each system has the stated vector as its only solution). `ε(β, f)` is
`V_β(f^(∞)) − x(f)/(1 − β) − y(f)`, and the limit is along `𝓝[<] 1`. -/
theorem theorem_4a {St Act : Type*} [Fintype St] [DecidableEq St] [Nonempty St]
    [Fintype Act] [Nonempty Act] (M : Model St Act)
    (f : St → Act) :
    (∀ z : St → ℝ, ((1 - M.Q f) *ᵥ z = 0 ∧ M.Qstar f *ᵥ z = M.Qstar f *ᵥ M.r f) ↔ z = M.x f) ∧
      (∀ z : St → ℝ, ((1 - M.Q f) *ᵥ z = M.r f - M.x f ∧ M.Qstar f *ᵥ z = 0) ↔ z = M.y f) ∧
      Tendsto (fun β : ℝ => M.V β (Policy.stationary f) - (1 - β)⁻¹ • M.x f - M.y f)
        (𝓝[<] 1) (𝓝 0) := by sorry

end BlackwellDiscreteDP.NearOne
Source
Blackwell, Discrete Dynamic Programming, Ann. Math. Statist. 33(2):719–726 (1962), DOI 10.1214/aoms/1177704593, p. 722, Theorem 4(a)
Human review
  • Endorsed by Shuze Chen · Oct 6, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 6, 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