A class of generalized Nesterov's accelerated gradient method from dynamical perspective

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Cheng, Xu, Liu, Jiaqi, Shang, Zaijiu
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911109909315584
author Cheng, Xu
Liu, Jiaqi
Shang, Zaijiu
author_facet Cheng, Xu
Liu, Jiaqi
Shang, Zaijiu
contents We propose a class of \textit{Euler-Lagrange} equations indexed by a pair of parameters ($α,r$) that generalizes Nesterov's accelerated gradient methods for convex ($α=1$) and strongly convex ($α=0$) functions from a continuous-time perspective. This class of equations also serves as an interpolation between the two Nesterov's schemes. The corresponding \textit{Hamiltonian} systems can be integrated via the symplectic Euler scheme with a fixed step-size. Furthermore, we can obtain the convergence rates for these equations ($0<α<1$) that outperform Nesterov's when time is sufficiently large for $μ$-strongly convex functions, without requiring a priori knowledge of $μ$. We demonstrate this by constructing a class of Lyapunov functions that also provide a unified framework for Nesterov's schemes for convex and strongly convex functions.
format Preprint
id arxiv_https___arxiv_org_abs_2508_12816
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A class of generalized Nesterov's accelerated gradient method from dynamical perspective
Cheng, Xu
Liu, Jiaqi
Shang, Zaijiu
Optimization and Control
Dynamical Systems
We propose a class of \textit{Euler-Lagrange} equations indexed by a pair of parameters ($α,r$) that generalizes Nesterov's accelerated gradient methods for convex ($α=1$) and strongly convex ($α=0$) functions from a continuous-time perspective. This class of equations also serves as an interpolation between the two Nesterov's schemes. The corresponding \textit{Hamiltonian} systems can be integrated via the symplectic Euler scheme with a fixed step-size. Furthermore, we can obtain the convergence rates for these equations ($0<α<1$) that outperform Nesterov's when time is sufficiently large for $μ$-strongly convex functions, without requiring a priori knowledge of $μ$. We demonstrate this by constructing a class of Lyapunov functions that also provide a unified framework for Nesterov's schemes for convex and strongly convex functions.
title A class of generalized Nesterov's accelerated gradient method from dynamical perspective
topic Optimization and Control
Dynamical Systems
url https://arxiv.org/abs/2508.12816