Paperlane

no. 16 · Tuesday 17 November 2026

Augmented Lagrangian methods with Nesterov acceleration for convex optimization with nonlinear inequality constraints

Xin He

math.OC · 34 pp

Abstract

In this paper, we develop Nesterov accelerated augmented Lagrangian methods for composite convex optimization with nonlinear inequality constraints. The methods are derived through a compatible discretization of inertial primal-dual dynamics, yielding a unified framework that couples difference extrapolation of the constraint mapping with a projected multiplier. Under convexity and weighted summability of the primal stationarity errors, we prove boundedness of the primal-dual iterates and simultaneous $\mathcal O(k^{-2})$ last-iterate rates for the objective residual and constraint violation. Under strict parameter conditions, the primal-dual sequence converges to a Karush-Kuhn-Tucker pair, and both residuals improve to $o(k^{-2})$. These results require no strong convexity of the objective or constraints. Numerical experiments examine the performance of the proposed methods.

comments
34 pages. This paper is an expanded version of the discrete-time part (Section 3) of arXiv:2609.01415v2
fields
Optimization and Control, Numerical Analysis
pdf ↗ page arXiv:2611.20134 ↗