Unified Ergodic Primal-Dual Gap Rates with Unhalved Primal Stepsizes

Abstract

We study ergodic primal-dual gap rates for first-order primal-dual methods applied to \[ x f(x)+g(x)+h(Ax), \] where f is smooth and convex, g and h are proper, closed, convex functions, and A is linear. Standard gap-rate proofs often impose the halved smooth-stepsize condition τ 1/L, even though the corresponding convergence theory allows the larger range τ<2/L. We introduce a residual-to-gap transfer principle: positive residual terms in the one-step gap inequality are controlled by the decrease of a Lyapunov function. This yields O(1/K) ergodic primal-dual gap bounds with the unhalved primal stepsize τ<2/L for Condat--Vũ, PD3O, AFBA/PDDY, and PAPC/PDFP2O, under their algorithm-dependent product conditions. We also give a two-dimensional counterexample showing that the fully separated rectangle τ<2/L, τη\|A\|2<4/3 cannot hold in the general three-function setting.

0

Turn this paper into a full lesson

ArcXiv compiles a staged curriculum from this paper: 8-12 lessons across beginner → advanced, synthesised section guides, visuals, flashcards, a quiz, exercises, and on-demand deep dives per section. Grounded in the abstract, never invented.

Discussion (0)

Sign in to join the discussion.

Loading comments…