Math 2.0

Resources / preprint

Training-Loss Guarantees for Muon with Finite-Step Newton--Schulz Orthogonalization CXQ7FKSKSB

URIm2://res/CXQ7FKSKSB
Kindpreprint
CreatorsAmartya Roy, Souvik Chakraborty
Date2026-10-02
Subjectsarxiv cs.LG, arxiv cs.AI, arxiv math.OC
Scopeborder

Expressions

e1: version, preprint

Date2026-10-02
IdsarXiv 2610.03306

Items in this resource

The mentions of the inventory that name this resource. Each name links to the card of its inventory row.

ItemKindWhereSource
Initialization eventstatementlemma B.1arxiv
One-step residual decreasestatementlemma 3.3arxiv
Informal statementstatementtheoremarxiv
Newton--Schulz matrix boundsstatementlemma B.3arxiv
Optimal buffer alignment per unit spectral normstatementproposition 3arxiv
Gradient-update alignmentstatementlemma 3.2arxiv
G4A8M1GEZ7statementproposition 1arxiv
Uniform NTK and ReLU remainderstatementlemma B.2arxiv
Classical Newton--Schulz polynomialsstatementproposition 2arxiv
Finite-time training guaranteestatementtheorem 3.1arxiv

Catalogue records