{"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","summary":"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 $\\|x_0-x_\\star\\|$…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-07-23","reactions":{}}