arXiv · 2611.17026
Accelerated Transformed Primal Dual Methods for Smooth Convex Optimization Problems with Affine Equality Constraints
Abstract
Accelerated transformed primal--dual (ATPD) methods are proposed for smooth convex optimization problems with affine equality constraints. The methods combine primal--dual dynamics with a transformation designed to accelerate convergence. In the strongly convex case, ATPD achieves an accelerated linear convergence rate and matches the known lower bounds on first-order oracle complexity in terms of gradient evaluations and matrix--vector products. The analysis distinguishes these two types of computational work when assessing the complexity of the method. For general convex objectives, a homotopy perturbation strategy is introduced, and an accelerated sublinear rate for the modified Lyapunov energy is proved under a uniform boundedness condition on the primal iterates at all homotopy stages. The acceleration mechanism is based on an exponentially stable ATPD flow and a modified discrete Lyapunov function. This continuous-time perspective guides the construction and analysis of the discrete methods. Numerical experiments on policy evaluation and constrained signal reconstruction problems support the theory and illustrate the efficiency of the proposed methods.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Long Chen, Jingrong Wei. 2026-11-12. Accelerated Transformed Primal Dual Methods for Smooth Convex Optimization Problems with Affine Equality Constraints. https://arxiv.science/abs/2611.17026
Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.