Abstract:We propose a new framework to design and analyze accelerated methods that solve general monotone equation (ME) problems $F(x)=0$. Traditional approaches include generalized steepest descent methods and inexact Newton-type methods. If $F$ is uniformly monotone and twice differentiable, these methods achieve local convergence rates while the latter methods are globally convergent thanks to line search and hyperplane projection. However, a global rate is unknown for these methods. The variational inequality methods can be applied to yield a global rate that is expressed in terms of $\|F(x)\|$ but these results are restricted to first-order methods and a Lipschitz continuous operator. It has not been clear how to obtain global acceleration using high-order Lipschitz continuity. This paper takes a continuous-time perspective where accelerated methods are viewed as the discretization of dynamical systems. Our contribution is to propose accelerated rescaled gradient systems and prove that they are equivalent to closed-loop control systems. Based on this connection, we establish the properties of solution trajectories. Moreover, we provide a unified algorithmic framework obtained from discretization of our system, which together with two approximation subroutines yields both existing high-order methods and new first-order methods. We prove that the $p^{th}$-order method achieves a global rate of $O(k^{-p/2})$ in terms of $\|F(x)\|$ if $F$ is $p^{th}$-order Lipschitz continuous and the first-order method achieves the same rate if $F$ is $p^{th}$-order strongly Lipschitz continuous. If $F$ is strongly monotone, the restarted versions achieve local convergence with order $p$ when $p \geq 2$. Our discrete-time analysis is largely motivated by the continuous-time analysis and demonstrates the fundamental role that rescaled gradients play in global acceleration for solving ME problems.

Understanding the acceleration phenomenon via high-resolution differential equations

Acceleration via symplectic discretization of high-resolution differential equations

A Differential Equation for Modeling Nesterov's Accelerated Gradient Method: Theory and Insights

Accelerated optimization algorithms and ordinary differential equations: the convex non Euclidean case

Comparative analysis of accelerated gradient algorithms for convex optimization: high and super resolution ODE approach

A unified Euler--Lagrange system for analyzing continuous-time accelerated gradient methods

Analysis Accelerated Mirror Descent via High-resolution ODEs

Direct Runge-Kutta Discretization Achieves Acceleration.

A dynamical systems perspective on Nesterov acceleration

Accelerated Gradient Algorithms with Adaptive Subspace Search for Instance-Faster Optimization

A Continuous-Time Perspective on Global Acceleration for Monotone Equation Problems

Lyapunov Analysis For Monotonically Forward-Backward Accelerated Algorithms

On the connections between optimization algorithms, Lyapunov functions, and differential equations: theory and insights

A variational perspective on accelerated methods in optimization

Accelerated gradient methods for nonconvex optimization: Escape trajectories from strict saddle points and convergence to local minima

Asymptotic Analysis via Stochastic Differential Equations of Gradient Descent Algorithms in Statistical and Computational Paradigms

A Lyapunov analysis of accelerated methods in optimization

An SDE Perspective on Stochastic Inertial Gradient Dynamics with Time-Dependent Viscosity and Geometric Damping

Acceleration in First Order Quasi-strongly Convex Optimization by ODE Discretization