Smoothed Proximal Lagrangian Method for Nonlinear Constrained Programs

TL;DR

This paper proposes a single-looped smoothed proximal Lagrangian method for nonconvex constrained optimization, achieving an $ ext{O}(rac{1}{ extepsilon^2})$ iteration complexity and demonstrating practical efficiency and high solution quality.

Contribution

It introduces a novel single-looped smoothed proximal Lagrangian method with proven iteration complexity for nonconvex constrained problems.

Findings

Method is practically efficient with faster convergence.

Achieves an $ ext{O}(rac{1}{ extepsilon^2})$ iteration complexity.

Numerical experiments show improved speed and solution quality.

Abstract

This paper introduces a smoothed proximal Lagrangian method for minimizing a nonconvex smooth function over a convex domain with additional explicit convex nonlinear constraints. Two key features are 1) the proposed method is single-looped, and 2) an first-order iteration complexity of $\mathcal{O}(\epsilon^{-2})$ is established under mild regularity assumptions. The first feature suggests the practical efficiency of the proposed method, while the second feature highlights its theoretical superiority. Numerical experiments on various problem scales demonstrate the advantages of the proposed method in terms of speed and solution quality.