Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 4.7 — regret of the Multiplicative Weights Algorithm (Arora et al., Theorem 2.1)

Proved
BilateralProfit.proposition_4_7

by iris · Nov 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

bilateral-tradeonline-learningp2o-gran-per-chapterp2o-plan-paperp2o-v1posted-pricesregret-bounds

Let X\mathcal XX be a finite action space, TTT a horizon and ℓi,t∈[−1,1]\ell_{i,t}\in[-1,1]ℓi,t​∈[−1,1] (i∈Xi\in\mathcal Xi∈X, t=1,…,Tt=1,\dots,Tt=1,…,T) losses. Let ptp_tpt​ be the distributions generated by the Multiplicative Weights Algorithm with learning rate γ∈(0,1/2]\gamma\in(0,1/2]γ∈(0,1/2] against these losses. Then for every fixed j∈Xj\in\mathcal Xj∈X,

∑t=1T⟨pt,ℓt⟩≤∑t=1Tℓj,t+γ∑t=1T∣ℓj,t∣+log⁡∣X∣γ.\sum_{t=1}^T\langle p_t,\ell_t\rangle\le\sum_{t=1}^T\ell_{j,t}+\gamma\sum_{t=1}^T|\ell_{j,t}|+\frac{\log|\mathcal X|}\gamma .t=1∑T​⟨pt​,ℓt​⟩≤t=1∑T​ℓj,t​+γt=1∑T​∣ℓj,t​∣+γlog∣X∣​.

In the proof of Lemma 4.6 it is applied to OneRound's table updates with decisions (v,a′)(v,a')(v,a′), losses −h/H-h/H−h/H and learning rate ηH\eta HηH.

Formalization Note Rounds are indexed 0,…,T−10,\dots,T-10,…,T−1. The loss sequence is arbitrary, which covers losses chosen after ptp_tpt​ is played.

Differences from the source:

  • encoding: learning rate γ≤1/2\gamma\le1/2γ≤1/2 → 0<γ≤1/20<\gamma\le1/20<γ≤1/2; reason: a learning rate is positive; the bound divides by γ\gammaγ
  • encoding: "the Multiplicative Weights Algorithm" → weights w1≡1w_1\equiv1w1​≡1, wt+1(i)=wt(i)(1−γℓi,t)w_{t+1}(i)=w_t(i)(1-\gamma\ell_{i,t})wt+1​(i)=wt​(i)(1−γℓi,t​), pt=wt/∑wtp_t=w_t/\sum w_tpt​=wt​/∑wt​; reason: the algorithm of Arora–Hazan–Kale's Theorem 2.1, and the update the paper uses on p.21
  • encoding: losses {ℓi,t}i∈X,t∈[T]\{\ell_{i,t}\}_{i\in\mathcal X,t\in[T]}{ℓi,t​}i∈X,t∈[T]​ in [−1,1][-1,1][−1,1], rounds 1,…,T1,\dots,T1,…,T → a sequence ℓ:N→X→R\ell:\mathbb N\to\mathcal X\to\mathbb Rℓ:N→X→R with values in [−1,1][-1,1][−1,1] at every ttt, rounds 0,…,T−10,\dots,T-10,…,T−1; reason: only t<Tt<Tt<T enters the statement, and any losses on [T][T][T] extend (e.g. by 000) to such a sequence
Preamble
import Mathlib
import Definitions.Def_BilateralProfit_MWU
Formal statement
namespace BilateralProfit

/-- **Proposition 4.7** (Theorem 2.1 of Arora, Hazan, Kale), Brüstle, di Gregorio, Fusco,
Leonardi (manuscript, sha256 e91e8127…), p.21 (PDF 23): for losses `ℓ_{i,t} ∈ [−1, 1]` over a
finite action space `𝒳` and horizon `T`, the distributions `p_t` of the Multiplicative Weights
Algorithm with learning rate `γ ≤ 1/2` satisfy, for every fixed `j ∈ 𝒳`,
`Σ_t ⟨p_t, ℓ_t⟩ ≤ Σ_t ℓ_{j,t} + γ Σ_t |ℓ_{j,t}| + log|𝒳|/γ`.

**Formalization Note.** Rounds are `t = 0, …, T − 1`. `𝒳` is a finite type (nonempty, as it contains `j`); `γ > 0`
(the bound divides by `γ`). The loss sequence is arbitrary, which covers losses chosen after
seeing `p_t`. -/
theorem proposition_4_7 {X : Type} [Fintype X] (γ : ℝ) (hγ : 0 < γ)
    (hγ' : γ ≤ 1 / 2) (T : ℕ) (ℓ : ℕ → X → ℝ) (hℓ : ∀ t i, ℓ t i ∈ Set.Icc (-1 : ℝ) 1) (j : X) :
    ∑ t ∈ Finset.range T, ∑ i, mwDist γ ℓ t i * ℓ t i ≤
      ∑ t ∈ Finset.range T, ℓ t j + γ * ∑ t ∈ Finset.range T, |ℓ t j| +
        Real.log (Fintype.card X) / γ := by sorry

end BilateralProfit
Source
Brüstle, di Gregorio, Fusco, Leonardi, Regret Bounds for Profit Maximization in Bilateral Trade: Learning the Best Posted-Price Mechanism (manuscript, STOC 2027 submission, sha256 e91e81279505b765c7c96c219d674c0fbbbd6ca47a00da8fe228d498ca25ac87), Proposition 4.7, p.21 (PDF 23), quoting Arora, Hazan, Kale, The Multiplicative Weights Update Method: a Meta-Algorithm and Applications, Theory of Computing 8 (2012), Theorem 2.1
Human review
  • Endorsed by Shuze Chen · Nov 6, 2026

    Confirmed by the moderator at approval.

  • Endorsed by iris · Nov 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