Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

p. 123: the indices computed by the adaptive greedy algorithm AG(A, r) for a SFABP are nondecreasing along its order and equal the Gittins indices

Proved
AllocationIndices.adaptive_greedy_indices_eq_gittins

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

achievable-regionadaptive-greedy-algorithmgittins-index

p. 123. The algorithm calculates νij\nu_{i_j}νij​​s, with riN=νiN≥νiN−1≥⋯≥νi1r_{i_N} = \nu_{i_N} \ge \nu_{i_{N-1}} \ge \cdots \ge \nu_{i_1}riN​​=νiN​​≥νiN−1​​≥⋯≥νi1​​. These are in fact the Gittins indices.

Formally: for a bandit process on E={1,…,N}E = \{1, \dots, N\}E={1,…,N} with kernel PPP, rewards rrr and discount factor a∈(0,1)a \in (0, 1)a∈(0,1), let AiS=E[1+a+⋯+aTiS−1]A_i^S = \mathbb{E}[1 + a + \cdots + a^{T_i^S - 1}]AiS​=E[1+a+⋯+aTiS​−1] be the coefficients (5.3). For every output (σ,y)(\sigma, y)(σ,y) of the adaptive greedy algorithm AG(A,r)AG(A, r)AG(A,r) (any tie-breaking), the indices νik=∑j≥kyˉSj\nu_{i_k} = \sum_{j \ge k} \bar y_{S_j}νik​​=∑j≥k​yˉ​Sj​​ satisfy νi1≤νi2≤⋯≤νiN\nu_{i_1} \le \nu_{i_2} \le \cdots \le \nu_{i_N}νi1​​≤νi2​​≤⋯≤νiN​​, and νi=ν(B,i)\nu_i = \nu(B, i)νi​=ν(B,i), the Gittins index of state iii, for every state iii. (That νiN=riN\nu_{i_N} = r_{i_N}νiN​​=riN​​ is immediate from AiE=1A_i^E = 1AiE​=1 and is not restated.)

Preamble
import Definitions.Def_AllocationIndices_Achievable

open MeasureTheory ProbabilityTheory BanditAlgorithm Finset
Formal statement
namespace AllocationIndices

/-- **The adaptive greedy indices are the Gittins indices** (p. 123, Eqs. (5.13)–(5.14)). For a
bandit process on the finite state space `E = Fin N` with transition kernel `P`, rewards `r` and
discount factor `a ∈ (0, 1)`, let `A` be the matrix (5.3) of expected discounted return times.
Every output `(σ, y)` of the adaptive greedy algorithm `AG(A, r)` lists the states in
nondecreasing order of its indices, `ν_{i_1} ≤ ν_{i_2} ≤ ⋯ ≤ ν_{i_N}`, and these indices are
the Gittins indices: `ν_i = ν(B, i)` for every state `i`. -/
theorem adaptive_greedy_indices_eq_gittins {N : ℕ} (P : Kernel (Fin N) (Fin N))
    [IsMarkovKernel P] (r : Fin N → ℝ) {a : ℝ} (ha0 : 0 < a) (ha1 : a < 1)
    (σ : Equiv.Perm (Fin N)) (y : Fin N → ℝ) (h : IsAdaptiveGreedy (conservationCoeff P a) r σ y) :
    (∀ j k : Fin N, j ≤ k → greedyIndex σ y (σ j) ≤ greedyIndex σ y (σ k)) ∧
    ∀ i, greedyIndex σ y i = gittinsIndex P r a i := by sorry

end AllocationIndices
Source
Gittins, Glazebrook and Weber, Multi-armed Bandit Allocation Indices, 2nd ed., Wiley 2011, doi:10.1002/9780470980033, §5.3 pp. 122-123, the adaptive greedy algorithm AG(A, r) and the identification (5.13)-(5.14) of its output with the Gittins indices
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