{"id":"arxiv/2609.30546","url":"https://arxiv.org/abs/2609.30546","tags":["math.NA","cs.AI","math.OC"],"score":0,"title":"Convergence guarantees for Muon: New parameter regimes and generalizations","domain":"math.NA","source":"arxiv","authors":["Arthur C. B. de Oliveira","Dhruv D. Jatkar","Guilherme S. Vicinansa","Eduardo D. Sontag"],"reviews":0,"subject":"math.NA","summary":"In this paper, we establish the first asymptotic convergence guarantees for the Muon algorithm through a more accurate proxy for the Newton-Schultz iteration than the typical matrix sign function. We prove that, for appropriate choices of hyperparameters, the iterates satisfy $\\lim_{k\\to\\infty}\\|\\nabla f(x_k)\\|=0$, and, under a global Polyak-Łojasiewicz condition, that the sequence of function values converges linearly. The key insight is that the regularization, implicit in Muon's Newton-Schulz implementation, induces a bounded preconditioner, exposing Muon as a \\emph{preconditioned Polyak…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-09-24","reactions":{}}