Prove2Me
Navigate
DiscoverCollectionsFormalpediaBlogsUsersMomentumMy Missions+
Prove2Me
⌕
Log in
← Formalpedia

Theorem 1, p. 13 — for f ∈ S²_{μ,L} and 0 < s ≤ 1/L, the NAG-SC high-resolution ODE satisfies f(X(t)) − f(x⋆) ≤ (2‖x₀ − x⋆‖²/s) e^{−√μ t/4}

Proved
HighResODE.NAGSCODE.theorem_1

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

accelerated-gradienthigh-resolution-odelinear-convergencenesterovp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1strongly-convex

Let f∈Sμ,L2(Rn)f\in\mathcal S^2_{\mu,L}(\mathbb R^n)f∈Sμ,L2​(Rn), that is, fff is LLL-smooth, μ\muμ-strongly convex with 0<μ≤L0<\mu\le L0<μ≤L, of class C2C^2C2 with a Lipschitz-continuous Hessian, and let x⋆x^\starx⋆ be its minimizer. Let 0<s≤1/L0<s\le1/L0<s≤1/L be a step size and x0∈Rnx_0\in\mathbb R^nx0​∈Rn. Let X=X(t)X=X(t)X=X(t) be the solution of the high-resolution ODE of NAG-SC,

X¨(t)+2μ X˙(t)+s ∇2f(X(t))X˙(t)+(1+μs)∇f(X(t))=0,X(0)=x0, X˙(0)=−2s ∇f(x0)1+μs.\ddot X(t)+2\sqrt\mu\,\dot X(t)+\sqrt s\,\nabla^2 f(X(t))\dot X(t)+\big(1+\sqrt{\mu s}\big)\nabla f(X(t))=0,\qquad X(0)=x_0,\ \dot X(0)=-\frac{2\sqrt s\,\nabla f(x_0)}{1+\sqrt{\mu s}}.X¨(t)+2μ​X˙(t)+s​∇2f(X(t))X˙(t)+(1+μs​)∇f(X(t))=0,X(0)=x0​, X˙(0)=−1+μs​2s​∇f(x0​)​.

Then for every t≥0t\ge0t≥0,

f(X(t))−f(x⋆)≤2∥x0−x⋆∥2s e−μt4.f(X(t))-f(x^\star)\le\frac{2\|x_0-x^\star\|^2}{s}\,e^{-\frac{\sqrt\mu t}{4}}.f(X(t))−f(x⋆)≤s2∥x0​−x⋆∥2​e−4μ​t​.

The function value along the high-resolution ODE of Nesterov's accelerated gradient method for strongly convex functions converges linearly, at the accelerated rate μ\sqrt\muμ​ rather than μ\muμ; with s=1/Ls=1/Ls=1/L the bound reads f(X)−f(x⋆)≤2L∥x0−x⋆∥2e−μt/4f(X)-f(x^\star)\le2L\|x_0-x^\star\|^2e^{-\sqrt\mu t/4}f(X)−f(x⋆)≤2L∥x0​−x⋆∥2e−μ​t/4. It is the continuous-time counterpart of the discrete rate for NAG-SC in Theorem 3 of the same paper.

Formalization Note. A solution is any pair of curves (X,V)(X,V)(X,V) with VVV the one-sided derivative of XXX on [0,∞)[0,\infty)[0,∞), VVV having the derivative the ODE prescribes, and the initial conditions above; existence and uniqueness (Proposition 2.1) are not assumed, so "the solution" is read as "every solution". The minimizer x⋆x^\starx⋆ is a point with f(x⋆)≤f(z)f(x^\star)\le f(z)f(x⋆)≤f(z) for all zzz. ∇2f(X)X˙\nabla^2 f(X)\dot X∇2f(X)X˙ is the Fréchet derivative of ∇f\nabla f∇f at XXX applied to X˙\dot XX˙. The Hessian's Lipschitz constant is not fixed.

Preamble
import Mathlib
import Definitions.Def_HighResODE_NAGSCODE_Setting

open scoped InnerProductSpace
Formal statement
namespace HighResODE.NAGSCODE

theorem theorem_1 {n : ℕ} (f : EuclideanSpace ℝ (Fin n) → ℝ) (μ L s : ℝ) (hf : IsS2 f μ L)
    (xs : EuclideanSpace ℝ (Fin n)) (hmin : ∀ z, f xs ≤ f z) (hs : 0 < s) (hsL : s ≤ 1 / L)
    (x0 : EuclideanSpace ℝ (Fin n)) (X V : ℝ → EuclideanSpace ℝ (Fin n))
    (hsol : IsNAGSCODE f μ s x0 X V) :
    ∀ t, 0 ≤ t →
      f (X t) - f xs ≤ 2 * ‖x0 - xs‖ ^ 2 / s * Real.exp (-(Real.sqrt μ * t / 4)) := by sorry

end HighResODE.NAGSCODE
Source
Shi, Du, Jordan & Su, Understanding the Acceleration Phenomenon via High-Resolution Differential Equations, arXiv:1810.08907v3, p. 13, Theorem 1
Human review
  • Endorsed by Shuze Chen · Oct 10, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 10, 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 paper
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me