Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

§2.6.4 (Katehakis–Veinott): the restart-in-state value iteration converges to μ(ξ) = max(0, ν(B, ξ)/(1 − a))

Proved
AllocationIndices.restart_value_iteration

by naimengye · Sep 24, 2026 · Mathlib 0df444a (Lean v4.33.1)

gittins-indexrestart-in-statevalue-iteration

§2.6.4. Fix a state ξ\xiξ and let μ(x)\mu(x)μ(x) be the maximal payoff obtainable from the bandit when it starts in state xxx and may be restarted in state ξ\xiξ whenever and as often as one likes; then μ(ξ)=ν(B,ξ)/(1−a)\mu(\xi) = \nu(B, \xi)/(1 - a)μ(ξ)=ν(B,ξ)/(1−a). Katehakis and Veinott (1987) compute it by value iteration: μ0(⋅)=0\mu_0(\cdot) = 0μ0​(⋅)=0, μk+1(x)=max⁡{μk(ξ), r(x)+a∑yP(y∣x)μk(y)}\mu_{k+1}(x) = \max\{\mu_k(\xi),\ r(x) + a\sum_y P(y \mid x)\mu_k(y)\}μk+1​(x)=max{μk​(ξ), r(x)+a∑y​P(y∣x)μk​(y)}, and μk(ξ)→μ(ξ)\mu_k(\xi) \to \mu(\xi)μk​(ξ)→μ(ξ) as k→∞k \to \inftyk→∞.

Formally: for a bandit process on a countable state space with bounded reward, a∈(0,1)a \in (0,1)a∈(0,1) and any state ξ\xiξ, the iteration restartIter P r a ξ satisfies

lim⁡k→∞μk(ξ)=max⁡(0, ν(B,ξ)1−a).\lim_{k \to \infty} \mu_k(\xi) = \max\Big(0,\ \frac{\nu(B, \xi)}{1 - a}\Big).k→∞lim​μk​(ξ)=max(0, 1−aν(B,ξ)​).

Why the maximum with 000. Restarting is free in the iteration (μk+1(x)≥μk(ξ)\mu_{k+1}(x) \ge \mu_k(\xi)μk+1​(x)≥μk​(ξ) with no reward and no discounting), so restarting forever is available and the restart value is never negative: for a single state with reward r(ξ)=c<0r(\xi) = c < 0r(ξ)=c<0 the iteration is identically 000 while ν/(1−a)=c/(1−a)<0\nu/(1-a) = c/(1-a) < 0ν/(1−a)=c/(1−a)<0. The book's derivation takes the supremum over τ>0\tau > 0τ>0 only, which forces at least one continuation between restarts; for a nonnegative index (in particular for nonnegative rewards) the two agree and the limit is the book's ν(B,ξ)/(1−a)\nu(B, \xi)/(1 - a)ν(B,ξ)/(1−a). Checked on random finite chains with rewards of both signs.

Preamble
import Definitions.Def_AllocationIndices_Index

open MeasureTheory ProbabilityTheory BanditAlgorithm
Formal statement
namespace AllocationIndices

theorem restart_value_iteration {S : Type*} [MeasurableSpace S] [Countable S]
    [MeasurableSingletonClass S] (P : Kernel S S) [IsMarkovKernel P] {r : S → ℝ}
    (hr : BoundedReward r) {a : ℝ} (ha0 : 0 < a) (ha1 : a < 1) (ξ : S) :
    Filter.Tendsto (fun k ↦ restartIter P r a ξ k ξ) Filter.atTop
      (nhds (max 0 (gittinsIndex P r a ξ / (1 - a)))) := by sorry

end AllocationIndices
Source
Gittins, Glazebrook and Weber, Multi-armed Bandit Allocation Indices, 2nd ed., Wiley 2011, doi:10.1002/9780470980033, §2.6.4 p. 31: the restart-in-state formulation μ(ξ) = ν(B, ξ)/(1 − a) and the value iteration μ_{k+1}(x) = max{μ_k(ξ), r(x) + a Σ_y P(y|x) μ_k(y)}, μ_0 = 0, with μ_k(ξ) → μ(ξ)
Human review
  • Endorsed by Shuze Chen · Sep 25, 2026

    Confirmed by the moderator at approval.

  • Endorsed by naimengye · Sep 25, 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