Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma G.9 — direct corruption contribution of GIA-FTRL

Proved
IAFTRL.lemma_G_9

by Canzhe Zhao · Nov 17, 2026 · Mathlib 0df444a (Lean v4.33.1)

bandit-algorithmslinear-banditsonline-learning

Setting. A GIA-FTRL run of T≥1T\ge1T≥1 rounds in the corrupted regime with budget C≥0C\ge0C≥0: ∥θ∥A≤1\|\theta\|_{\mathcal A}\le1∥θ∥A​≤1, ∥θt∥A≤1\|\theta_t\|_{\mathcal A}\le1∥θt​∥A​≤1, and ∑t∥θt−θ∥A≤C\sum_t\|\theta_t-\theta\|_{\mathcal A}\le C∑t​∥θt​−θ∥A​≤C almost surely.

Claim.

I4=∑tE⟨pt−qt,x⊤ctvec⟩≤32 dC.I_4=\sum_t\mathbb E\langle p_t-q_t,x^\top c^{\mathrm{vec}}_t\rangle\le32\,d\sqrt C .I4​=t∑​E⟨pt​−qt​,x⊤ctvec​⟩≤32dC​.

The paper states the slightly smaller constant 162 d16\sqrt2\,d162​d.

Proof idea.

  1. Marginal domination gives pt−qt=Γt(p~t−qt)p_t-q_t=\Gamma_t(\tilde p_t-q_t)pt​−qt​=Γt​(p~​t​−qt​) for a probability measure p~t\tilde p_tp~​t​.
  2. On K\mathcal KK the range of x⊤ctvecx^\top c^{\mathrm{vec}}_tx⊤ctvec​ is at most 2κt2\kappa_t2κt​, where κt=∥ctvec∥A≤2\kappa_t=\|c^{\mathrm{vec}}_t\|_{\mathcal A}\le2κt​=∥ctvec​∥A​≤2.
  3. With Γt≤4d/t\Gamma_t\le4d/\sqrt tΓt​≤4d/t​, it remains to show ∑tκt/t≤4C\sum_t\kappa_t/\sqrt t\le4\sqrt C∑t​κt​/t​≤4C​. Split at m=⌊C/4⌋m=\lfloor C/4\rfloorm=⌊C/4⌋.
Preamble
import Mathlib
import Definitions.Def_IAFTRL_LinEnv
import Definitions.Def_IAFTRL_GIAFTRL
import Definitions.Def_IAFTRL_GTerms
Formal statement
namespace IAFTRL

open MeasureTheory

/-- **Lemma G.9 (Direct corruption contribution).** Zhao–Tan, *Best-of-Both-Worlds FTRL for Linear Bandits* (STOC submission version), Lemma G.9, p.137: "For
`κ_t = ‖c_t^vec‖_𝒜` with `0 ≤ κ_t ≤ 2` and `Σ_t κ_t ≤ C` almost surely,
`I₄ := Σ_t E[⟨p_t − q_t, xᵀc_t^vec⟩] ≤ 16√2 d √C`."

Stated with the slightly larger constant `32 d`: for every GIA-FTRL run of `T ≥ 1` rounds in the
corrupted regime with budget `C ≥ 0`, `I₄ ≤ 32 d √C`. -/
theorem lemma_G_9 {d : ℕ} (hd : 1 ≤ d) (𝒜 ℬ : Set (Fin d → ℝ)) (e : ℕ → Fin d → ℚ)
    (h𝒜 : IsCompact 𝒜) (hℬ : IsCompact ℬ) (hgen : IsGenerator 𝒜 ℬ)
    (hball : ∀ a ∈ 𝒜, a ⬝ᵥ a ≤ 1) (hspan : Submodule.span ℝ 𝒜 = ⊤) (hk : 1 ≤ kdim ℬ)
    (θ : Fin d → ℝ) {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
    (T : ℕ) (C : ℝ) (R : GRun d Ω) (hT : 1 ≤ T) (hC : 0 ≤ C) (hR : IsGIAFTRLRun P 𝒜 ℬ e T R)
    (hcor : IsCorruptedG P 𝒜 θ C T R) :
    corrI4 P ℬ e T R θ ≤ 32 * d * Real.sqrt C := by sorry

end IAFTRL
Source
Zhao–Tan, Best-of-Both-Worlds FTRL for Linear Bandits: The Interplay of Information Acquisition, Corruption, and Geometry (STOC submission version; arXiv:2611.01299), Appendix G.5, Lemma G.9, (G.148)–(G.151)

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