Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proof of Theorem 2.6, p. 19 — τq+1=exp⁡(−(τ/cγ)2/γt−1)\tau^{q+1} = \exp(-(\tau/c_\gamma)^{2/\gamma}t^{-1})τq+1=exp(−(τ/cγ​)2/γt−1) implies τ≤(a^γ/2)γ/2(tln⁡1a^t)γ/2\tau \le (\hat a\gamma/2)^{\gamma/2}(t\ln\frac1{\hat at})^{\gamma/2}τ≤(a^γ/2)γ/2(tlna^t1​)γ/2

Proved
FastRatesSVM.GeomNoise.tau_estimate

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

learning-theoryp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1real-analysis

Let cγ>0c_\gamma > 0cγ​>0, γ>0\gamma > 0γ>0 and q≥0q \ge 0q≥0, and put a^:=(cγ)2/γ(q+1)\hat a := (c_\gamma)^{2/\gamma}(q + 1)a^:=(cγ​)2/γ(q+1). For t>0t > 0t>0 consider the equation in τ>0\tau > 0τ>0

τq+1=exp⁡(−(τcγ)2/γt−1).\tau^{q+1} = \exp\Big(-\Big(\frac{\tau}{c_\gamma}\Big)^{2/\gamma} t^{-1}\Big).τq+1=exp(−(cγ​τ​)2/γt−1).

There is t0>0t_0 > 0t0​>0, depending only on cγc_\gammacγ​, γ\gammaγ and qqq, such that for every t∈(0,t0)t \in (0, t_0)t∈(0,t0​) the equation has a solution τ>0\tau > 0τ>0, and every solution satisfies

τ≤(a^γ2)γ/2(tln⁡1a^t)γ/2.\tau \le \Big(\frac{\hat a\gamma}{2}\Big)^{\gamma/2}\Big(t \ln\frac{1}{\hat a t}\Big)^{\gamma/2}.τ≤(2a^γ​)γ/2(tlna^t1​)γ/2.

This is the choice of the splitting level τ\tauτ in display (33): it balances the two terms there, and the resulting bound (C+1)τq+1(C+1)\tau^{q+1}(C+1)τq+1 decays like (tln⁡(1/t))γ(q+1)/2(t\ln(1/t))^{\gamma(q+1)/2}(tln(1/t))γ(q+1)/2, which yields every geometric noise exponent α<(q+1)γ/d\alpha < (q+1)\gamma/dα<(q+1)γ/d.

Formalization Note The page writes "Let us define τ\tauτ by … For a^:=…\hat a := \ldotsa^:=… and small ttt this definition implies …". "Small ttt" is the threshold t0t_0t0​, chosen before ttt and τ\tauτ. The existence of a solution, presupposed by "define τ\tauτ by", is stated as the first conjunct; the inequality is stated for every positive solution. For t<t0t < t_0t<t0​ one has a^t<1\hat a t < 1a^t<1, so the logarithm is positive.

Preamble
import Mathlib
Formal statement
namespace FastRatesSVM.GeomNoise

theorem tau_estimate (γ q cγ : ℝ) (hγ : 0 < γ) (hq : 0 ≤ q) (hcγ : 0 < cγ) :
    ∃ t₀ : ℝ, 0 < t₀ ∧ ∀ t : ℝ, 0 < t → t < t₀ →
      (∃ τ : ℝ, 0 < τ ∧ τ ^ (q + 1) = Real.exp (-(τ / cγ) ^ (2 / γ) * t⁻¹)) ∧
      ∀ τ : ℝ, 0 < τ → τ ^ (q + 1) = Real.exp (-(τ / cγ) ^ (2 / γ) * t⁻¹) →
        τ ≤ (cγ ^ (2 / γ) * (q + 1) * γ / 2) ^ (γ / 2) *
          (t * Real.log (1 / (cγ ^ (2 / γ) * (q + 1) * t))) ^ (γ / 2) := by sorry

end FastRatesSVM.GeomNoise
Source
Steinwart, Scovel, Fast Rates for Support Vector Machines Using Gaussian Kernels, arXiv:0708.1838v1, p. 19, §4.2, proof of Theorem 2.6, display after (33)
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