{"id":"https://openalex.org/W7134925487","doi":"https://doi.org/10.48550/arxiv.2603.09952","title":"On the Width Scaling of Neural Optimizers Under Matrix Operator Norms I: Row/Column Normalization and Hyperparameter Transfer","display_name":"On the Width Scaling of Neural Optimizers Under Matrix Operator Norms I: Row/Column Normalization and Hyperparameter Transfer","publication_year":2026,"publication_date":"2026-03-10","ids":{"openalex":"https://openalex.org/W7134925487","doi":"https://doi.org/10.48550/arxiv.2603.09952"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.09952","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.09952","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.09952","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5128762490","display_name":"Ruihan Xu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Ruihan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128774715","display_name":"Jiajin Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Jiajin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5128768272","display_name":"Yiping Lu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Yiping","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.675000011920929,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.675000011920929,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11948","display_name":"Machine Learning in Materials Science","score":0.12600000202655792,"subfield":{"id":"https://openalex.org/subfields/2505","display_name":"Materials Chemistry"},"field":{"id":"https://openalex.org/fields/25","display_name":"Materials Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12535","display_name":"Machine Learning and Data Classification","score":0.04450000077486038,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/lipschitz-continuity","display_name":"Lipschitz continuity","score":0.6890000104904175},{"id":"https://openalex.org/keywords/normalization","display_name":"Normalization (sociology)","score":0.6262000203132629},{"id":"https://openalex.org/keywords/operator","display_name":"Operator (biology)","score":0.5842999815940857},{"id":"https://openalex.org/keywords/smoothness","display_name":"Smoothness","score":0.5648999810218811},{"id":"https://openalex.org/keywords/gradient-descent","display_name":"Gradient descent","score":0.5145000219345093},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.489300012588501},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.4650999903678894}],"concepts":[{"id":"https://openalex.org/C22324862","wikidata":"https://www.wikidata.org/wiki/Q652707","display_name":"Lipschitz continuity","level":2,"score":0.6890000104904175},{"id":"https://openalex.org/C136886441","wikidata":"https://www.wikidata.org/wiki/Q926129","display_name":"Normalization (sociology)","level":2,"score":0.6262000203132629},{"id":"https://openalex.org/C17020691","wikidata":"https://www.wikidata.org/wiki/Q139677","display_name":"Operator (biology)","level":5,"score":0.5842999815940857},{"id":"https://openalex.org/C102634674","wikidata":"https://www.wikidata.org/wiki/Q868473","display_name":"Smoothness","level":2,"score":0.5648999810218811},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.5145000219345093},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.4968999922275543},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.489300012588501},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.477400004863739},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.4650999903678894},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.44850000739097595},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4255000054836273},{"id":"https://openalex.org/C2777952282","wikidata":"https://www.wikidata.org/wiki/Q3354606","display_name":"Transfer operator","level":2,"score":0.37040001153945923},{"id":"https://openalex.org/C8642999","wikidata":"https://www.wikidata.org/wiki/Q4171168","display_name":"Hyperparameter","level":2,"score":0.35370001196861267},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.32820001244544983},{"id":"https://openalex.org/C122890649","wikidata":"https://www.wikidata.org/wiki/Q7834016","display_name":"Transfer matrix","level":2,"score":0.3215999901294708},{"id":"https://openalex.org/C106487976","wikidata":"https://www.wikidata.org/wiki/Q685816","display_name":"Matrix (chemical analysis)","level":2,"score":0.3188999891281128},{"id":"https://openalex.org/C81299745","wikidata":"https://www.wikidata.org/wiki/Q334269","display_name":"Transfer function","level":2,"score":0.30630001425743103},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.28029999136924744},{"id":"https://openalex.org/C134765980","wikidata":"https://www.wikidata.org/wiki/Q879126","display_name":"Bitwise operation","level":2,"score":0.2612000107765198}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.09952","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.09952","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.09952","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.09952","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"A":[0],"central":[1],"question":[2,25],"in":[3,90,172,237],"modern":[4],"deep":[5,91],"learning":[6,134],"is":[7,229],"how":[8],"to":[9,120],"design":[10],"optimizers":[11,122,182],"whose":[12],"behavior":[13],"remains":[14],"stable":[15,210],"as":[16,37,124,142],"the":[17,52,56,173,191],"network":[18,57],"width":[19],"$w$":[20],"increases.":[21],"We":[22,93,161],"address":[23],"this":[24,95],"by":[26,74,97],"interpreting":[27],"several":[28],"widely":[29],"used":[30],"neural-network":[31],"optimizers,":[32],"including":[33],"\\textrm{AdamW}":[34],"and":[35,60,67,84,117,129,146,220,239],"\\textrm{Muon},":[36],"instances":[38],"of":[39,55,64,101,159,180],"steepest":[40],"descent":[41],"under":[42],"matrix":[43],"operator":[44,79,103],"norms.":[45],"This":[46],"perspective":[47],"links":[48],"optimizer":[49,202],"geometry":[50],"with":[51,226,231],"Lipschitz":[53,66],"structure":[54],"forward":[58],"map,":[59],"enables":[61,209],"width-independent":[62,88,114,186],"control":[63],"both":[65],"smoothness":[68,115,174,187],"constants.":[69],"However,":[70],"steepest-descent":[71],"rules":[72,138],"induced":[73],"standard":[75],"$p":[76],"\\to":[77,107],"q$":[78],"norms":[80],"lack":[81],"layerwise":[82,111],"composability":[83],"therefore":[85],"cannot":[86],"provide":[87,147],"bounds":[89],"architectures.":[92],"overcome":[94],"limitation":[96],"introducing":[98],"a":[99,143,148,156,177,200],"family":[100,179],"mean-normalized":[102],"norms,":[104],"denoted":[105],"$\\pmean":[106],"\\qmean$,":[108],"that":[109,164,208,223],"admit":[110],"composability,":[112],"yield":[113],"bounds,":[116],"give":[118],"rise":[119],"practical":[121],"such":[123],"\\emph{rescaled}":[125],"\\textrm{AdamW},":[126],"row":[127,227],"normalization,":[128,228],"column":[130],"normalization.":[131],"The":[132],"resulting":[133],"rate":[135],"width-aware":[136,201],"scaling":[137],"recover":[139],"$\u03bc$P":[140],"scaling~\\cite{yang2021tensor}":[141],"special":[144],"case":[145],"principled":[149],"mechanism":[150],"for":[151],"cross-width":[152],"learning-rate":[153,211],"transfer":[154,212],"across":[155,213],"broad":[157],"class":[158],"optimizers.":[160],"further":[162],"show":[163],"\\textrm{Muon}":[165],"can":[166],"suffer":[167],"an":[168],"$\\mathcal{O}(\\sqrt{w})$":[169],"worst-case":[170],"growth":[171],"constant,":[175],"whereas":[176],"new":[178],"row-normalized":[181],"we":[183,193],"propose":[184,194],"achieves":[185],"guarantees.":[188],"Based":[189],"on":[190,205,218],"observations,":[192],"MOGA":[195],"(Matrix":[196],"Operator":[197],"Geometry":[198],"Aware),":[199],"based":[203],"only":[204],"row/column-wise":[206],"normalization":[207],"model":[214],"widths.":[215],"Large-scale":[216],"pre-training":[217],"GPT-2":[219],"LLaMA":[221],"shows":[222],"MOGA,":[224],"especially":[225],"competitive":[230],"Muon":[232],"while":[233],"being":[234],"notably":[235],"faster":[236],"large-token":[238],"low-loss":[240],"regimes.":[241]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-03-12T00:00:00"}