Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Section 5, p. 1028 — on a fixed realisation, any policy's greatest prevailing charge is at least the Gittins policy's

Proved
WeberGittins.Suboptimality.greatest_prevailing_charge_ge_gittins

by mikedeng1 · Oct 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

banditsgittins-indexinterleavingp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1prevailing-charge

Fix a realisation of the states through which each of kkk bandits evolves. Along it, the prevailing charge of bandit iii after it has been played mmm times is a number Hi(m)H_i(m)Hi​(m), and m↦Hi(m)m \mapsto H_i(m)m↦Hi​(m) is nonincreasing (it is a running minimum of fair charges). A policy, on this realisation, is a sequence of arms a(0),a(1),…a(0), a(1), \dotsa(0),a(1),…; write

nia(t)=#{ s<t:a(s)=i }n_i^a(t) = \#\{\, s < t : a(s) = i \,\}nia​(t)=#{s<t:a(s)=i}

for the number of plays of bandit iii before time ttt, so that Hi(nia(t))H_i(n_i^a(t))Hi​(nia​(t)) is the prevailing charge of bandit iii at time ttt. The Gittins policy plays, at every time, a bandit whose current prevailing charge is greatest: a∗a^*a∗ is a sequence with Hi(nia∗(t))≤Ha∗(t)(na∗(t)a∗(t))H_i(n_i^{a^*}(t)) \le H_{a^*(t)}(n_{a^*(t)}^{a^*}(t))Hi​(nia∗​(t))≤Ha∗(t)​(na∗(t)a∗​(t)) for all iii and ttt.

Then for every sequence aaa and every time ttt,

max⁡iHi(nia∗(t))  ≤  max⁡iHi(nia(t)).\max_i H_i\big(n_i^{a^*}(t)\big) \;\le\; \max_i H_i\big(n_i^{a}(t)\big).imax​Hi​(nia∗​(t))≤imax​Hi​(nia​(t)).

In words: if one follows a policy that differs from the Gittins policy, the greatest prevailing charge at time ttt is at least as great as it would have been under the Gittins policy. Combined with (5), this gives the suboptimality bound of Theorem 2.

Formalization Note The statement is deterministic. The paper's "fixed realisation" is encoded by arbitrary nonincreasing sequences Hi:N→RH_i : \mathbb{N} \to \mathbb{R}Hi​:N→R (the prevailing-charge stacks, which along a realisation are m↦min⁡l≤mγi(ωi(l))m \mapsto \min_{l \le m} \gamma_i(\omega_i(l))m↦minl≤m​γi​(ωi​(l))), and "the Gittins policy" by any greedy interleaving of these stacks (the published IsGreedyChargeStackInterleaving, ties broken arbitrarily). The statement is proved for all nonincreasing stacks, which include the running-minimum stacks. The play counts are the published stackPullCountBefore. Bandits are indexed by Fin k\mathrm{Fin}\,kFink; at k=0k = 0k=0 both maxima are the empty supremum 000.

Preamble
import Definitions.Def_GittinsChargeInterleaving

open BanditAlgorithm
Formal statement
namespace WeberGittins.Suboptimality
theorem greatest_prevailing_charge_ge_gittins {k : ℕ} (H : Fin k → ℕ → ℝ)
    (hH : ∀ i, Antitone (H i)) (astar : ℕ → Fin k)
    (hastar : IsGreedyChargeStackInterleaving H astar) (a : ℕ → Fin k) (t : ℕ) :
    (⨆ i : Fin k, H i (stackPullCountBefore astar i t)) ≤
      ⨆ i : Fin k, H i (stackPullCountBefore a i t) := by sorry
end WeberGittins.Suboptimality
Source
Weber, On the Gittins index for multiarmed bandits, Ann. Appl. Probab. 2 (1992), p. 1028, Section 5, the paragraph after eq. (5)
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