{"id":"https://openalex.org/W7127616161","doi":"https://doi.org/10.48550/arxiv.2602.02908","title":"A Random Matrix Theory Perspective on the Consistency of Diffusion Models","display_name":"A Random Matrix Theory Perspective on the Consistency of Diffusion Models","publication_year":2026,"publication_date":"2026-02-02","ids":{"openalex":"https://openalex.org/W7127616161","doi":"https://doi.org/10.48550/arxiv.2602.02908"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2602.02908","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2602.02908","pdf_url":"https://arxiv.org/pdf/2602.02908","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2602.02908","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5063326178","display_name":"BinXu Wang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Binxu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5063822685","display_name":"Jacob A. Zavatone-Veth","orcid":"https://orcid.org/0000-0002-4060-1738","observed_orcids":["https://orcid.org/0000-0002-4060-1738"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zavatone-Veth, Jacob","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5023195984","display_name":"Cengiz Pehlevan","orcid":"https://orcid.org/0000-0001-9767-6063","observed_orcids":["https://orcid.org/0000-0001-9767-6063"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pehlevan, Cengiz","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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.9775000214576721,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.9775000214576721,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.00570000009611249,"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/T10688","display_name":"Image and Signal Denoising Methods","score":0.004699999932199717,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/diffusion-models","display_name":"diffusion models","score":0.9919999837875366},{"id":"https://openalex.org/keywords/random-matrix-theory","display_name":"random matrix theory","score":0.6669999957084656},{"id":"https://openalex.org/keywords/linear-diffusion-model","display_name":"linear diffusion model","score":0.5820000171661377},{"id":"https://openalex.org/keywords/sampling-variability","display_name":"sampling variability","score":0.7630000114440918},{"id":"https://openalex.org/keywords/reproducibility","display_name":"reproducibility","score":0.6959999799728394},{"id":"https://openalex.org/keywords/u-net","display_name":"U-Net","score":0.609000027179718},{"id":"https://openalex.org/keywords/dit","display_name":"DiT","score":0.8370000123977661}],"concepts":[{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.5619999766349792},{"id":"https://openalex.org/C2776436953","wikidata":"https://www.wikidata.org/wiki/Q5163215","display_name":"Consistency (knowledge bases)","level":2,"score":0.5437999963760376},{"id":"https://openalex.org/C64812099","wikidata":"https://www.wikidata.org/wiki/Q176604","display_name":"Random matrix","level":3,"score":0.5303000211715698},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.49559998512268066},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.4943000078201294},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.49050000309944153},{"id":"https://openalex.org/C163716315","wikidata":"https://www.wikidata.org/wiki/Q901177","display_name":"Gaussian","level":2,"score":0.4713999927043915},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.4544000029563904},{"id":"https://openalex.org/C121864883","wikidata":"https://www.wikidata.org/wiki/Q677916","display_name":"Statistical physics","level":1,"score":0.4465999901294708},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.435699999332428},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.42899999022483826},{"id":"https://openalex.org/C106487976","wikidata":"https://www.wikidata.org/wiki/Q685816","display_name":"Matrix (chemical analysis)","level":2,"score":0.41589999198913574},{"id":"https://openalex.org/C69357855","wikidata":"https://www.wikidata.org/wiki/Q163214","display_name":"Diffusion","level":2,"score":0.4131999909877777},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.3959999978542328},{"id":"https://openalex.org/C75291252","wikidata":"https://www.wikidata.org/wiki/Q1315756","display_name":"TRACE (psycholinguistics)","level":2,"score":0.3944999873638153},{"id":"https://openalex.org/C163175372","wikidata":"https://www.wikidata.org/wiki/Q3339222","display_name":"Linear model","level":2,"score":0.3792000114917755},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.3675999939441681},{"id":"https://openalex.org/C48921125","wikidata":"https://www.wikidata.org/wiki/Q10861030","display_name":"Linear regression","level":2,"score":0.36660000681877136},{"id":"https://openalex.org/C185142706","wikidata":"https://www.wikidata.org/wiki/Q1134404","display_name":"Covariance matrix","level":2,"score":0.3635999858379364},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.3379000127315521},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.3249000012874603},{"id":"https://openalex.org/C39927690","wikidata":"https://www.wikidata.org/wiki/Q11197","display_name":"Logarithm","level":2,"score":0.31779998540878296},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.30640000104904175},{"id":"https://openalex.org/C4199805","wikidata":"https://www.wikidata.org/wiki/Q2725903","display_name":"Gaussian noise","level":2,"score":0.30160000920295715},{"id":"https://openalex.org/C2781181686","wikidata":"https://www.wikidata.org/wiki/Q4226068","display_name":"Coherence (philosophical gambling strategy)","level":2,"score":0.2971999943256378},{"id":"https://openalex.org/C125112378","wikidata":"https://www.wikidata.org/wiki/Q176640","display_name":"Randomness","level":2,"score":0.289900004863739},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2842999994754791},{"id":"https://openalex.org/C2781395549","wikidata":"https://www.wikidata.org/wiki/Q4680762","display_name":"Adaptive sampling","level":3,"score":0.2800000011920929},{"id":"https://openalex.org/C178650346","wikidata":"https://www.wikidata.org/wiki/Q201984","display_name":"Covariance","level":2,"score":0.27889999747276306},{"id":"https://openalex.org/C129848803","wikidata":"https://www.wikidata.org/wiki/Q2564360","display_name":"Sample size determination","level":2,"score":0.2766999900341034},{"id":"https://openalex.org/C61326573","wikidata":"https://www.wikidata.org/wiki/Q1496376","display_name":"Gaussian process","level":3,"score":0.2655999958515167},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.25920000672340393},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.2574000060558319},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.25060001015663147}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2602.02908","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2602.02908","pdf_url":"https://arxiv.org/pdf/2602.02908","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2602.02908","doi":"https://doi.org/10.48550/arxiv.2602.02908","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2602.02908","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2602.02908","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2602.02908","pdf_url":"https://arxiv.org/pdf/2602.02908","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4387559876","https://openalex.org/W4391801159","https://openalex.org/W4400023626","https://openalex.org/W4404395459","https://openalex.org/W7106710713","https://openalex.org/W6966946868","https://openalex.org/W4417135962","https://openalex.org/W4415798575","https://openalex.org/W6908459910","https://openalex.org/W4379256097"],"abstract_inverted_index":{"Diffusion":[0],"models":[1],"trained":[2],"on":[3,162],"different,":[4],"non-overlapping":[5],"subsets":[6],"of":[7,39,64,83,153,194,199],"a":[8,26,48,81,88,183],"dataset":[9,106,130],"often":[10],"produce":[11],"strikingly":[12],"similar":[13],"outputs":[14],"when":[15],"given":[16],"the":[17,30,40,60,65,72,84,105,151,197],"same":[18],"noise":[19,85],"seed.":[20],"We":[21],"trace":[22],"this":[23],"consistency":[24],"to":[25,135,142,196],"simple":[27],"linear":[28,73,154],"effect:":[29],"shared":[31],"Gaussian":[32],"statistics":[33],"across":[34,121,124,177],"splits":[35],"already":[36],"predict":[37],"much":[38],"generated":[41],"images.":[42],"To":[43],"formalize":[44],"this,":[45],"we":[46,158],"develop":[47],"random":[49],"matrix":[50,137],"theory":[51,148],"(RMT)":[52],"framework":[53],"that":[54],"quantifies":[55],"how":[56,174],"finite":[57],"datasets":[58],"shape":[59],"expectation":[61],"and":[62,68,101,126,157,164,173],"variance":[63,111],"learned":[66],"denoiser":[67],"sampling":[69,77,145],"map":[70],"in":[71,167,188],"setting.":[74],"For":[75,108],"expectations,":[76],"variability":[78],"acts":[79],"as":[80],"renormalization":[82],"level":[86],"through":[87],"self-consistent":[89],"relation":[90],"$\u03c3^2":[91],"\\mapsto":[92],"\u03ba(\u03c3^2)$,":[93],"explaining":[94],"why":[95],"limited":[96],"data":[97,179,195],"overshrink":[98],"low-variance":[99],"directions":[100],"pull":[102],"samples":[103,175],"toward":[104],"mean.":[107],"fluctuations,":[109],"our":[110],"formulas":[112],"reveal":[113],"three":[114],"key":[115],"factors":[116],"behind":[117],"cross-split":[118],"disagreement:":[119],"\\textit{anisotropy}":[120],"eigenmodes,":[122],"\\textit{inhomogeneity}":[123],"inputs,":[125],"overall":[127],"scaling":[128],"with":[129],"size.":[131],"Extending":[132],"deterministic-equivalence":[133],"tools":[134],"fractional":[136],"powers":[138],"further":[139],"allows":[140],"us":[141],"analyze":[143],"entire":[144],"trajectories.":[146],"The":[147],"sharply":[149],"predicts":[150],"behavior":[152],"diffusion":[155,189],"models,":[156],"validate":[159],"its":[160],"predictions":[161],"UNet":[163],"DiT":[165],"architectures":[166],"their":[168],"non-memorization":[169],"regime,":[170],"identifying":[171],"where":[172],"deviates":[176],"training":[178],"split.":[180],"This":[181],"provides":[182],"principled":[184],"baseline":[185],"for":[186],"reproducibility":[187],"training,":[190],"linking":[191],"spectral":[192],"properties":[193],"stability":[198],"generative":[200],"outputs.":[201]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-02-06T00:00:00"}