Id: arxiv/2607.20875 Url: https://arxiv.org/abs/2607.20875 Tags: math.OC Score: 0 Title: Implicit Primal-Dual Guarantees in Unconstrained First-Order Minimization Domain: math.OC Source: arxiv Authors: Benjamin Grimmer, Alex L. Wang Reviews: 0 Subject: math.OC Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-07-23 Page: https://delphi.science/r/arxiv/2607.20875 This work considers the design of first-order convex optimization algorithms and convergence proofs. In particular, we consider nonsmooth Lipschitz and smooth problems accessed through a subgradient or gradient oracle, respectively. For the general class of fixed-step first-order methods, prior work on Performance Estimation Problems (PEPs) has shown that structured, tight convergence proofs typically exist. Under mild conditions, we further show that any first-order method guaranteeing a bound on the primal objective gap $f(x_N)-f(x_\star)$ assuming only a bound on $\|x_0-x_\star\|$…