Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 2, p. 14 — under a K-cardinality constraint every policy has regret at least C√(NT/K) on a randomized instance

Proved
MNLBandit.LowerBound.theorem_2

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

assortment-optimizationlower-boundmnl-banditp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1regret

Setting. An MNL-Bandit instance with nnn products consists of a no-purchase weight v0>0v_0>0v0​>0, attraction parameters viv_ivi​ with 0≤vi≤v00\le v_i\le v_00≤vi​≤v0​, and revenues ri∈[0,1]r_i\in[0,1]ri​∈[0,1]. Offering an assortment SSS yields one customer choice drawn from the MNL model pi(S)=vi/(v0+∑j∈Svj)p_i(S)=v_i/(v_0+\sum_{j\in S}v_j)pi​(S)=vi​/(v0​+∑j∈S​vj​) for i∈S∪{0}i\in S\cup\{0\}i∈S∪{0}, and the expected revenue R(S,v)=∑i∈Srivi/(v0+∑j∈Svj)R(S,v)=\sum_{i\in S}r_iv_i/(v_0+\sum_{j\in S}v_j)R(S,v)=∑i∈S​ri​vi​/(v0​+∑j∈S​vj​). A policy may be randomized; it respects the KKK-cardinality constraint if it only offers assortments with ∣Stπ∣≤K|S^\pi_t|\le K∣Stπ​∣≤K. Its regret over TTT periods is

Regπ(T,v)=Eπ(∑t=1TR(S∗,v)−R(Stπ,v)),\mathrm{Reg}_\pi(T,v)=\mathbb E_\pi\Big(\sum_{t=1}^{T}R(S^*,v)-R(S^\pi_t,v)\Big),Regπ​(T,v)=Eπ​(t=1∑T​R(S∗,v)−R(Stπ​,v)),

where S∗S^*S∗ is an assortment of at most KKK products with maximum expected revenue.

Theorem. There is an absolute constant C>0C>0C>0 such that for every number of products n≥2n\ge2n≥2, every KKK with 1≤K≤n1\le K\le n1≤K≤n and every horizon T≥nT\ge nT≥n there is a randomized instance, a probability distribution with finite support over instances (v0,v,r)(v_0,v,r)(v0​,v,r) satisfying v0≥viv_0\ge v_iv0​≥vi​ and ri∈[0,1]r_i\in[0,1]ri​∈[0,1]. The values of v0v_0v0​ and rrr are common to every point in the support, as in the paper's constructions. Every policy π\piπ respecting the KKK-cardinality constraint has

Einstance[Regπ(T,v)]≥CnTK.\mathbb E_{\text{instance}}\big[\mathrm{Reg}_\pi(T,v)\big]\ge C\sqrt{\frac{nT}{K}} .Einstance​[Regπ​(T,v)]≥CKnT​​.

The randomized instance is chosen before the policy and may depend on nnn, KKK and TTT only. Together with the O(nTlog⁡nT)O(\sqrt{nT\log nT})O(nTlognT​) upper bound of Theorem 1 it shows that the epoch-based UCB policy is optimal up to logarithmic factors when KKK is fixed.

Formalization Note The randomized instance is a finite list of instances with weights wj≥0w_j\ge0wj​≥0 summing to 111; the paper's instances (Definitions 5.2 and E.1) are uniform over finitely many values, so a finite support is faithful, and the regret is averaged over the instance as in Lemma 5.1. All support points share the same known revenues rrr and no-purchase weight v0v_0v0​, as those constructions do. Without this clause, one common policy would be tested on instances distinguished by revenues known to the seller, weakening the lower bound. Policies are behavioural: at every history of offered sets and choices they draw the next assortment from a probability vector, which by Kuhn's theorem covers the paper's admissible policies (2.4). The page says "for any NNN and KKK"; the claim is false for n=1n=1n=1 (offering the single product whenever r1>0r_1>0r1​>0 is optimal for every vvv) and for K=0K=0K=0 (only ∅\emptyset∅ can be offered), so n≥2n\ge2n≥2 and K≥1K\ge1K≥1 are added; K≤nK\le nK≤n is the cardinality constraint's range. C>0C>0C>0 is part of the claim.

Preamble
import Mathlib
import Definitions.Def_MNLBandit_LowerBound_Setting
Formal statement
namespace MNLBandit.LowerBound

theorem theorem_2 :
    ∃ C : ℝ, 0 < C ∧ ∀ (n K T : ℕ), 2 ≤ n → 1 ≤ K → K ≤ n → n ≤ T →
      ∃ (m : ℕ) (w : Fin m → ℝ) (inst : Fin m → Instance n),
        (∀ j, 0 ≤ w j) ∧ ∑ j, w j = 1 ∧ (∀ j, (inst j).Valid) ∧
        (∀ j k, (inst j).v₀ = (inst k).v₀ ∧ (inst j).r = (inst k).r) ∧
        ∀ π : BPolicy n, IsCardinalityK K π →
          C * Real.sqrt ((n : ℝ) * T / K) ≤ ∑ j, w j * regret (inst j) K π T := by sorry

end MNLBandit.LowerBound
Source
Agrawal, Avadhanula, Goyal, Zeevi, MNL-Bandit: A Dynamic Learning Approach to Assortment Selection, arXiv:1706.03880v2, p. 14, Theorem 2; model pp. 5–7, (2.1)–(2.6)
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