Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 3.1 — regret of GIA-FTRL in the sharp action case

Proved
IAFTRL.theorem_3_1

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

bandit-algorithmslinear-banditsonline-learning

Setting. Let A⊆Bd\mathcal A\subseteq\mathbb B^dA⊆Bd (d≥1d\ge1d≥1) be a compact action set with span(A)=Rd\mathrm{span}(\mathcal A)=\mathbb R^dspan(A)=Rd. GIA-FTRL (Algorithm 2) runs on a compact generator B\mathcal BB, meaning B⊆A⊆Conv(B)\mathcal B\subseteq\mathcal A\subseteq\mathrm{Conv}(\mathcal B)B⊆A⊆Conv(B), with k=dim⁡aff⁡(Conv B)≥1k=\dim\operatorname{aff}(\mathrm{Conv}\,\mathcal B)\ge1k=dimaff(ConvB)≥1 and a fixed enumeration of Qd\mathbb Q^dQd. The regret against a comparator uuu is RegT(u)=E[∑t≤T(At−u)⊤θt]\mathrm{Reg}_T(u)=\mathbb E\big[\sum_{t\le T}(A_t-u)^\top\theta_t\big]RegT​(u)=E[∑t≤T​(At​−u)⊤θt​], and RegT=sup⁡u∈ARegT(u)\mathrm{Reg}_T=\sup_{u\in\mathcal A}\mathrm{Reg}_T(u)RegT​=supu∈A​RegT​(u). In the corrupted stochastic regime, θt=θ+ct\theta_t=\theta+c_tθt​=θ+ct​ with ∥θ∥A≤1\|\theta\|_{\mathcal A}\le1∥θ∥A​≤1 and ∑t≤T∥ct∥A≤C\sum_{t\le T}\|c_t\|_{\mathcal A}\le C∑t≤T​∥ct​∥A​≤C almost surely, and θ\thetaθ has a unique optimal action a∗∈Aa^*\in\mathcal Aa∗∈A. All objects are those of the definition IAFTRL_GIAFTRL.

There is an absolute constant c0>0c_0>0c0​>0 with the following two properties.

  1. Adversarial regime. Every run of GIA-FTRL satisfies, for all T≥2T\ge2T≥2 and all u∈Au\in\mathcal Au∈A,
RegT(u)≤c0 dTlog⁡T.(3.8)\mathrm{Reg}_T(u)\le c_0\,d\sqrt T\log T. \tag{3.8}RegT​(u)≤c0​dT​logT.(3.8)
  1. Corrupted stochastic regime. If the instance (A,θ)(\mathcal A,\theta)(A,θ) satisfies the sharpness condition (1.3), then there are finite constants MA,θM_{\mathcal A,\theta}MA,θ​ and JθcorJ^{\mathrm{cor}}_\thetaJθcor​, independent of TTT and CCC, such that for all T≥1T\ge1T≥1, C≥0C\ge0C≥0 and u∈Au\in\mathcal Au∈A,
RegT(u)≤c0 cctr∗(A,θ) dlog⁡T+MA,θ+JθcorClog⁡+(C),(3.10)\mathrm{Reg}_T(u)\le c_0\,c^*_{\mathrm{ctr}}(\mathcal A,\theta)\,d\log T+M_{\mathcal A,\theta}+J^{\mathrm{cor}}_\theta\sqrt C\log_+(C), \tag{3.10}RegT​(u)≤c0​cctr∗​(A,θ)dlogT+MA,θ​+Jθcor​C​log+​(C),(3.10)

where log⁡+(x)=log⁡max⁡{e,x}\log_+(x)=\log\max\{e,x\}log+​(x)=logmax{e,x} and cctr∗c^*_{\mathrm{ctr}}cctr∗​ is the centered information-acquisition complexity of Definition 1.2.

This is the paper's main guarantee for compact action sets. It gives a single FTRL-based algorithm with O(dTlog⁡T)O(d\sqrt T\log T)O(dT​logT) adversarial regret and, on sharp instances, O(cctr∗dlog⁡T)O(c^*_{\mathrm{ctr}}d\log T)O(cctr∗​dlogT) stochastic regret plus an O(Clog⁡+C)O(\sqrt C\log_+C)O(C​log+​C) corruption term.

Formalization Note.

  • O(⋅)O(\cdot)O(⋅) is the absolute constant c0c_0c0​. Since Tlog⁡T=0\sqrt T\log T=0T​logT=0 at T=1T=1T=1, (3.8) is stated for T≥2T\ge2T≥2.
  • The bound on RegT=sup⁡uRegT(u)\mathrm{Reg}_T=\sup_u\mathrm{Reg}_T(u)RegT​=supu​RegT​(u) is written as a bound for every comparator.
  • The constants MMM and JJJ may depend on ddd, A\mathcal AA, B\mathcal BB, the enumeration and θ\thetaθ.
Preamble
import Mathlib
import Definitions.Def_IAFTRL_LinEnv
import Definitions.Def_IAFTRL_GIAFTRL
Formal statement
namespace IAFTRL

open MeasureTheory

/-- **Theorem 3.1 (Regret in the sharp action case).** Zhao–Tan, *Best-of-Both-Worlds FTRL for
Linear Bandits* (STOC submission version), Theorem 3.1, p.25: "For linear bandits with compact
action sets, GIA-FTRL guarantees (3.8) `Reg_T = O(d√T log T)` in the adversarial regime.
Simultaneously, if the linear bandit instance satisfies the sharpness condition in (1.3),
GIA-FTRL guarantees `Reg_T = O(c*_ctr(𝒜, θ) d log T + M_{𝒜,θ} + J^cor_θ √C log₊(C))` (3.10) in
the corrupted stochastic regime, where `M_{𝒜,θ}` and `J^cor_θ` are finite parameter-dependent
constants independent of `T, C`."

GIA-FTRL is run on a compact generator `ℬ` of the compact action set `𝒜 ⊆ 𝔹^d`
(`span 𝒜 = ℝ^d`, `k = dim aff 𝒦 ≥ 1`) with an enumeration `e` of `ℚ^d`. `Reg_T = sup_{u∈𝒜}
Reg_T(u)` is written as a bound for every comparator `u ∈ 𝒜`. `O(·)` is an absolute constant
`c₀`; (3.8) is stated for `T ≥ 2` (at `T = 1`, `√T log T = 0`). `log₊ x = log max{e, x}`. -/
theorem theorem_3_1 : ∃ c₀ : ℝ, 0 < c₀ ∧
    ∀ {d : ℕ}, 1 ≤ d → ∀ (𝒜 ℬ : Set (Fin d → ℝ)) (e : ℕ → Fin d → ℚ),
      IsCompact 𝒜 → IsCompact ℬ → IsGenerator 𝒜 ℬ → (∀ a ∈ 𝒜, a ⬝ᵥ a ≤ 1) →
      Submodule.span ℝ 𝒜 = ⊤ → 1 ≤ kdim ℬ → Function.Surjective e →
      (∀ {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P] (T : ℕ)
          (R : GRun d Ω), 2 ≤ T → IsGIAFTRLRun P 𝒜 ℬ e T R →
          ∀ u ∈ 𝒜, gRegretVs P T R u ≤ c₀ * ((d : ℝ) * Real.sqrt T * Real.log T)) ∧
      (∀ (θ astar : Fin d → ℝ), setNorm 𝒜 θ ≤ 1 → astar ∈ 𝒜 →
        (∀ a ∈ 𝒜, a ≠ astar → astar ⬝ᵥ θ < a ⬝ᵥ θ) → IsSharp 𝒜 θ astar →
        ∃ M J : ℝ, ∀ {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
          (T : ℕ) (C : ℝ) (R : GRun d Ω), 1 ≤ T → 0 ≤ C →
          IsGIAFTRLRun P 𝒜 ℬ e T R → IsCorruptedG P 𝒜 θ C T R →
          ∀ u ∈ 𝒜, gRegretVs P T R u ≤
            c₀ * (cstarCtr 𝒜 θ astar * d * Real.log T) + M +
              J * Real.sqrt C * Real.log (max (Real.exp 1) 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), Section 3.2, Theorem 3.1, p. 25 (proof in Appendix F, pp. 120–123)

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