Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Gaussian-surrogate separation for nonsmooth convex losses

Proved
LG2021.gaussian_surrogate_separation

by sim2000dg · Nov 14, 2026 · Mathlib 0df444a (Lean v4.33.1)

convex-analysisprobabilitystochastic-optimization

For a globally convex Lipschitz function f:Rd→Rf:\mathbb R^d\to\mathbb Rf:Rd→R, let X∼N(0,Id)X\sim N(0,I_d)X∼N(0,Id​) and define

m=Ef(X),gλ(z)=m+Ef((1−λ)X+λz)−mλ,Δr=m−min⁡∥x∥≤rf(x).m=\mathbb E f(X),\qquad g_\lambda(z)=m+\frac{\mathbb E f((1-\lambda)X+\lambda z)-m}{\lambda},\qquad \Delta_r=m-\min_{\|x\|\le r}f(x).m=Ef(X),gλ​(z)=m+λEf((1−λ)X+λz)−m​,Δr​=m−∥x∥≤rmin​f(x).

Let d≥1d\ge1d≥1, let K≥0K\ge0K≥0 be a Lipschitz constant for fff, and let 0<λ<10<\lambda<10<λ<1, R>0R>0R>0, B≥0B\ge0B≥0. Suppose

(1−(1−λ)2)B22(1−λ)2≤1,(λR)22(1−λ)2≤1,Pr⁡(∥X∥>B)≤δ,\frac{(1-(1-\lambda)^2)B^2}{2(1-\lambda)^2}\le1,\quad \frac{(\lambda R)^2}{2(1-\lambda)^2}\le1,\quad \Pr(\|X\|>B)\le\delta,2(1−λ)2(1−(1−λ)2)B2​≤1,2(1−λ)2(λR)2​≤1,Pr(∥X∥>B)≤δ, 4≤λe−2R2128d,λ2e−2 2Kdδ R2≤ε4096dd.4\le\frac{\lambda e^{-2}R^2}{128d},\qquad \frac{\lambda}{2}e^{-2}\,2K\sqrt{d\delta}\,R^2\le\frac{\varepsilon}{4096d\sqrt d}.4≤128dλe−2R2​,2λ​e−22Kdδ​R2≤4096dd​ε​.

For Z=RX/∥X∥Z=RX/\|X\|Z=RX/∥X∥ (defined to be zero when X=0X=0X=0),

Pr⁡ ⁣(gλ(Z)−f(0)≥Δ2dR−ε4096dd)≥124.\Pr\!\left(g_\lambda(Z)-f(0)\ge\frac{\Delta_{2dR}-\varepsilon}{4096d\sqrt d}\right)\ge\frac1{24}.Pr(gλ​(Z)−f(0)≥4096dd​Δ2dR​−ε​)≥241​.

No differentiability hypothesis is imposed on fff. The proof uses positive Hessians, Gaussian integration by parts and symmetric density comparison, followed by double Gaussian smoothing and a closed-event limit.

To allow formalization of the separation argument in Tor Lattimore and András György, Improved Regret for Zeroth-Order Stochastic Convex Bandits, COLT 2021, PMLR 134:2938–2964, the derivation uses the centered Gaussian Hessian score XX⊤−IXX^\top-IXX⊤−I and a direct symmetric-density comparison in the part corresponding to Appendix E. Smoothing supplies the nonsmooth case, and all tail and parameter conditions are explicit. This statement concerns separation; the full regret theorem is stated separately.

Preamble
import Mathlib
import Definitions.Def_LG2021_GaussianSeparation
import Definitions.Def_LG2021_GaussianEstimators
open LG2021 LG2021.Concentration MeasureTheory ProbabilityTheory
Formal statement
theorem LG2021.gaussian_surrogate_separation {d : ℕ} (hd : 0 < d) {f : Space d → ℝ}
    {K : NNReal} (hf : LipschitzWith K f) (hc : ConvexOn ℝ Set.univ f)
    {lambda R B delta eps : ℝ} (hl : 0 < lambda) (hl1 : lambda < 1) (hR : 0 < R)
    (hB : 0 ≤ B) (hvar : (1-(1-lambda)^2)*B^2/(2*(1-lambda)^2) ≤ 1)
    (hshift : (lambda*R)^2/(2*(1-lambda)^2) ≤ 1)
    (htail : (gaussian d).real {x : Space d | B < ‖x‖} ≤ delta)
    (hgain : 4 ≤ (lambda/2)*Real.exp (-2)*(R^2/(64*(d : ℝ))))
    (herr : ((lambda/2)*Real.exp (-2)*(2*(K : ℝ)*Real.sqrt ((d : ℝ)*delta)))*R^2 ≤
      eps/(4096*(d : ℝ)*Real.sqrt (d : ℝ))) :
    (1/24 : ℝ) ≤ (sphereMeasure d R).real {z |
      (gaussianGap (2*(d : ℝ)*R) f-eps)/(4096*(d : ℝ)*Real.sqrt (d : ℝ)) ≤
        surrogate lambda f z-f 0} := by sorry
Source
Tor Lattimore and András György, Improved Regret for Zeroth-Order Stochastic Convex Bandits, Proceedings of the 34th Conference on Learning Theory (COLT 2021), Proceedings of Machine Learning Research 134:2938–2964, 2021, Lemma 5 and Appendix E. Separation argument adapted for formalization.

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