Saved in:
Bibliographic Details
Main Authors: Gonon, Antoine, Muşat, Andreea-Alexandra, Boumal, Nicolas
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2602.11948
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908829796532224
author Gonon, Antoine
Muşat, Andreea-Alexandra
Boumal, Nicolas
author_facet Gonon, Antoine
Muşat, Andreea-Alexandra
Boumal, Nicolas
contents Muon updates weight matrices along (approximate) polar factors of the gradients and has shown strong empirical performance in large-scale training. Existing attempts at explaining its performance largely focus on single-step comparisons (on quadratic proxies) and worst-case guarantees that treat the inexactness of the polar-factor as a nuisance ``to be argued away''. We show that already on simple strongly convex functions such as $L(W)=\frac12\|W\|_{\text{F}}^2$, these perspectives are insufficient, suggesting that understanding Muon requires going beyond local proxies and pessimistic worst-case bounds. Instead, our analysis exposes two observations that already affect behavior on simple quadratics and are not well captured by prevailing abstractions: (i) approximation error in the polar step can qualitatively alter discrete-time dynamics and improve reachability and finite-time performance -- an effect practitioners exploit to tune Muon, but that existing theory largely treats as a pure accuracy compromise; and (ii) structural properties of the objective affect finite-budget constants beyond the prevailing conditioning-based explanations. Thus, any general theory covering these cases must either incorporate these ingredients explicitly or explain why they are irrelevant in the regimes of interest.
format Preprint
id arxiv_https___arxiv_org_abs_2602_11948
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Insights on Muon from Simple Quadratics
Gonon, Antoine
Muşat, Andreea-Alexandra
Boumal, Nicolas
Optimization and Control
Machine Learning
Muon updates weight matrices along (approximate) polar factors of the gradients and has shown strong empirical performance in large-scale training. Existing attempts at explaining its performance largely focus on single-step comparisons (on quadratic proxies) and worst-case guarantees that treat the inexactness of the polar-factor as a nuisance ``to be argued away''. We show that already on simple strongly convex functions such as $L(W)=\frac12\|W\|_{\text{F}}^2$, these perspectives are insufficient, suggesting that understanding Muon requires going beyond local proxies and pessimistic worst-case bounds. Instead, our analysis exposes two observations that already affect behavior on simple quadratics and are not well captured by prevailing abstractions: (i) approximation error in the polar step can qualitatively alter discrete-time dynamics and improve reachability and finite-time performance -- an effect practitioners exploit to tune Muon, but that existing theory largely treats as a pure accuracy compromise; and (ii) structural properties of the objective affect finite-budget constants beyond the prevailing conditioning-based explanations. Thus, any general theory covering these cases must either incorporate these ingredients explicitly or explain why they are irrelevant in the regimes of interest.
title Insights on Muon from Simple Quadratics
topic Optimization and Control
Machine Learning
url https://arxiv.org/abs/2602.11948