Metraipolis › archive › dc_5ce592c992e8
Convergence guarantees for Muon: New parameter regimes and generalizations — arXiv:2609.30546
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.30546
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-08T01:39:14.779Z
- SHA-256 of the archived text
8c0ebfbd99c5ae6a7ea545e4aab5689d8af38440da314d7759d332eeadce7fdf- Stable URL of this page
https://metraipolis.com/documento/dc_5ce592c992e8
How to cite
Convergence guarantees for Muon: New parameter regimes and generalizations — arXiv:2609.30546 — arXiv (en), https://arxiv.org/abs/2609.30546 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_5ce592c992e8 (Metraipolis archive dc_5ce592c992e8, sha256 8c0ebfbd99c5ae6a7ea545e4aab5689d8af38440da314d7759d332eeadce7fdf)
BibTeX
@misc{dc5ce592c992e8,
title = {Convergence guarantees for Muon: New parameter regimes and generalizations — arXiv:2609.30546},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.30546},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_5ce592c992e8; sha256 8c0ebfbd99c5ae6a7ea545e4aab5689d8af38440da314d7759d332eeadce7fdf},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.30546. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
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 heavy-ball} method and enabling a classical Lyapunov analysis. This observation naturally motivates applying the same preconditioning structure to the Nesterov gradient evaluation. We formalize this idea by introducing \emph{Muesterov}, a Nesterov-based variant of Muon, and prove that it enjoys the same convergence guarantees, extending the theoretical framework beyond the heavy-ball setting. Numerical experiments on a scalar cross-entropy problem corroborate the theory and illuminate the joint role of the learning rate and the Newton-Schulz regularizer in controlling convergence. Preliminary numerical simulations training the nanoGPT dataset provide intuition regarding the relevance of the observations in this paper to practical applications.
Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-28 in cs.AI).
Categorie arXiv: Numerical Analysis (math.NA) ; Artificial Intelligence (cs.AI); Optimization and Control (math.OC).
Pagina del lavoro: https://arxiv.org/abs/2609.30546 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-08T01:12:14.443Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-28?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).