{"id":"arxiv/2609.11788","url":"https://arxiv.org/abs/2609.11788","tags":["math.OC"],"score":0,"title":"Optimal Recursive Composition and Dyadic Phase Laws for Gradient Descent with Predetermined Stepsizes","domain":"math.OC","source":"arxiv","authors":["Yu Liu","Kang Chen","Rujun Jiang","Tianyu Wang"],"reviews":0,"subject":"math.OC","summary":"Predetermined stepsize schedules featuring carefully chosen long steps have recently been shown to accelerate gradient descent (GD) on smooth convex functions. A prominent class of such schedules is built through recursive composition. In this paper, we characterize the convergence of these optimized recursive schedules, revealing a nonconstant log-periodic modulation across prescribed horizons. Specifically, we prove that for every $N \\geq 1$, the optimized ConPP/OBS-S schedules satisfy $$f(x_{N-1})-f_\\ast \\le \\frac{1}{2N^pΦ(\\log_2N)-1} \\frac{L}{2}\\|x_0-x_\\ast\\|^2, \\qquad…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-09-10","reactions":{}}