Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proof of Theorem 3, pp. 21–22 — E|N¹ − N⁰| = Σₖ C(n,k)qᵏ(1−q)ⁿ⁻ᵏ E|2Bin(k,½) − k| ≤ √(nq)

Proved
FastCLO.LowerBound.binomial_bound

by mikedeng1 · Oct 9, 2026 · Mathlib 0df444a (Lean v4.33.1)

binomiallower-boundsp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1

Let n∈Nn \in \mathbb Nn∈N and 0≤q≤10 \le q \le 10≤q≤1. Suppose each of nnn independent draws lands in a given cell with probability qqq, and each draw in the cell is labelled 111 or 000 with probability 1/21/21/2 each. With N1,N0N^1, N^0N1,N0 the numbers of draws in the cell with label 111 and 000,

E∣N1−N0∣=∑k=0n(nk)qk(1−q)n−k E∣2 Bin(k,12)−k∣  ≤  nq.\mathbb E|N^1 - N^0| = \sum_{k=0}^n \binom nk q^k (1-q)^{n-k}\ \mathbb E\big|2\,\mathrm{Bin}(k,\tfrac12) - k\big| \;\le\; \sqrt{nq}.E∣N1−N0∣=k=0∑n​(kn​)qk(1−q)n−k E​2Bin(k,21​)−k​≤nq​.

Explicitly, the statement is

∑k=0n(nk)qk(1−q)n−k∑j=0k(kj)2−k ∣2j−k∣  ≤  nq.\sum_{k=0}^n \binom nk q^k (1-q)^{n-k} \sum_{j=0}^k \binom kj 2^{-k}\,|2j - k| \;\le\; \sqrt{nq}.k=0∑n​(kn​)qk(1−q)n−kj=0∑k​(jk​)2−k∣2j−k∣≤nq​.

The page takes q=1/ηq = 1/\etaq=1/η (Theorem 3); the proof of Theorem 7 uses the same bound with q=ζα/(η−1)q = \zeta^\alpha/(\eta-1)q=ζα/(η−1). Combined with the Jensen step it bounds the Bayes risk of every coordinate from below by 12((1+ζ)/(1−ζ))−nq\tfrac12 ((1+\zeta)/(1-\zeta))^{-\sqrt{nq}}21​((1+ζ)/(1−ζ))−nq​.

Formalization Note The expectation is written as the page's explicit mixture over kkk (the number of draws in the cell) and jjj (the number of those labelled 111).

Preamble
import Mathlib
Formal statement
namespace FastCLO.LowerBound

/-- The binomial bound (Hu, Kallus, Mao, arXiv:2011.03030v3, proof of Theorem 3, pp. 21–22): if each
of `n` draws lands in a cell with probability `q`, and each draw in the cell is labelled `1` or `0`
with probability `1/2`, then the expected absolute difference of the label counts satisfies
`E|N¹ − N⁰| = Σ_k C(n,k) q^k (1−q)^{n−k} E|2 Bin(k, 1/2) − k| ≤ √(nq)`.

Formalization Note: the expectation is written as the page's explicit mixture, with `k` the number of
draws in the cell and `j` the number of those labelled `1`. The page has `q = 1/η` (Theorem 3); the
proof of Theorem 7 uses `q = ζ^α/(η − 1)` (p. 29). -/
theorem binomial_bound (n : ℕ) (q : ℝ) (hq0 : 0 ≤ q) (hq1 : q ≤ 1) :
    ∑ k ∈ Finset.range (n + 1), (n.choose k : ℝ) * q ^ k * (1 - q) ^ (n - k) *
        (∑ j ∈ Finset.range (k + 1), (k.choose j : ℝ) / 2 ^ k * |2 * (j : ℝ) - k|) ≤
      Real.sqrt (n * q) := by sorry

end FastCLO.LowerBound
Source
Hu, Kallus, Mao, Fast Rates for Contextual Linear Optimization, arXiv:2011.03030v3, proof of Theorem 3 (A.2), pp. 21–22, the display after "Thus, letting Bin(k, 1/2) …"; used with q = ζ^α/(η−1) in the proof of Theorem 7, p. 29
Human review
  • Endorsed by Shuze Chen · Oct 9, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 9, 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