{"id":"https://openalex.org/W7211995705","doi":"https://doi.org/10.48550/arxiv.2609.06685","title":"RoPE attention is an exact forward-pass gradient step with softmax intact","display_name":"RoPE attention is an exact forward-pass gradient step with softmax intact","publication_year":2026,"publication_date":"2026-09-06","ids":{"openalex":"https://openalex.org/W7211995705","doi":"https://doi.org/10.48550/arxiv.2609.06685"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.06685","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.06685","pdf_url":"https://arxiv.org/pdf/2609.06685","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.06685","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5150887435","display_name":"Julie Huang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Huang, Julie","raw_affiliation_strings":["Department of Information Engineering , Oxford University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Information Engineering , Oxford University","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119958722","display_name":"Maggie Chlon","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Chlon, Maggie","raw_affiliation_strings":["Department of Information Engineering , Oxford University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Information Engineering , Oxford University","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5119958721","display_name":"Leon Chlon","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Chlon, Leon","raw_affiliation_strings":["Department of Information Engineering , Oxford University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Information Engineering , Oxford University","institution_ids":["https://openalex.org/I40120149"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I40120149"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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/T10028","display_name":"Topic Modeling","score":0.7695000171661377,"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/T10028","display_name":"Topic Modeling","score":0.7695000171661377,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.053300000727176666,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.033399999141693115,"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/positional-encoding","display_name":"positional encoding","score":0.5400000214576721},{"id":"https://openalex.org/keywords/softmax-function","display_name":"softmax","score":0.5989999771118164},{"id":"https://openalex.org/keywords/finite-differences","display_name":"finite differences","score":0.4490000009536743},{"id":"https://openalex.org/keywords/qwen2-5","display_name":"Qwen2.5","score":0.6669999957084656}],"concepts":[{"id":"https://openalex.org/C188441871","wikidata":"https://www.wikidata.org/wiki/Q7554146","display_name":"Softmax function","level":3,"score":0.723800003528595},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.6547999978065491},{"id":"https://openalex.org/C92757383","wikidata":"https://www.wikidata.org/wiki/Q382497","display_name":"Affine transformation","level":2,"score":0.5773000121116638},{"id":"https://openalex.org/C57493831","wikidata":"https://www.wikidata.org/wiki/Q3134666","display_name":"Projection (relational algebra)","level":2,"score":0.5605000257492065},{"id":"https://openalex.org/C2780992000","wikidata":"https://www.wikidata.org/wiki/Q17016113","display_name":"Generator (circuit theory)","level":3,"score":0.555899977684021},{"id":"https://openalex.org/C106487976","wikidata":"https://www.wikidata.org/wiki/Q685816","display_name":"Matrix (chemical analysis)","level":2,"score":0.5486999750137329},{"id":"https://openalex.org/C162269090","wikidata":"https://www.wikidata.org/wiki/Q1156047","display_name":"Rope","level":2,"score":0.5410000085830688},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.531000018119812},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4724999964237213},{"id":"https://openalex.org/C151376022","wikidata":"https://www.wikidata.org/wiki/Q168698","display_name":"Exponential function","level":2,"score":0.4410000145435333},{"id":"https://openalex.org/C184992742","wikidata":"https://www.wikidata.org/wiki/Q7243229","display_name":"Prime (order theory)","level":2,"score":0.4291999936103821},{"id":"https://openalex.org/C129844170","wikidata":"https://www.wikidata.org/wiki/Q41299","display_name":"Quadratic equation","level":2,"score":0.4278999865055084},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.4244999885559082},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.4072999954223633},{"id":"https://openalex.org/C103275481","wikidata":"https://www.wikidata.org/wiki/Q6787889","display_name":"Matrix representation","level":3,"score":0.4002000093460083},{"id":"https://openalex.org/C2780427248","wikidata":"https://www.wikidata.org/wiki/Q17014996","display_name":"Fundamental matrix (linear differential equation)","level":2,"score":0.3375999927520752},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.3237999975681305},{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.31700000166893005},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.31360000371932983},{"id":"https://openalex.org/C2780312720","wikidata":"https://www.wikidata.org/wiki/Q5689100","display_name":"Head (geology)","level":2,"score":0.3077999949455261},{"id":"https://openalex.org/C118615104","wikidata":"https://www.wikidata.org/wiki/Q121416","display_name":"Discrete mathematics","level":1,"score":0.296099990606308},{"id":"https://openalex.org/C159694833","wikidata":"https://www.wikidata.org/wiki/Q2321565","display_name":"Iterative method","level":2,"score":0.29030001163482666},{"id":"https://openalex.org/C166437778","wikidata":"https://www.wikidata.org/wiki/Q50695","display_name":"Quadratic function","level":3,"score":0.28859999775886536},{"id":"https://openalex.org/C4263655","wikidata":"https://www.wikidata.org/wiki/Q2699958","display_name":"Matrix function","level":4,"score":0.2766999900341034},{"id":"https://openalex.org/C520416788","wikidata":"https://www.wikidata.org/wiki/Q5419229","display_name":"Exact solutions in general relativity","level":2,"score":0.26249998807907104},{"id":"https://openalex.org/C178009071","wikidata":"https://www.wikidata.org/wiki/Q93344","display_name":"Trigonometric functions","level":2,"score":0.2563000023365021},{"id":"https://openalex.org/C58166","wikidata":"https://www.wikidata.org/wiki/Q224821","display_name":"Fuzzy logic","level":2,"score":0.25270000100135803}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.06685","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.06685","pdf_url":"https://arxiv.org/pdf/2609.06685","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.06685","doi":"https://doi.org/10.48550/arxiv.2609.06685","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.06685","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:2609.06685","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.06685","pdf_url":"https://arxiv.org/pdf/2609.06685","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7211995705.pdf","grobid_xml":"https://content.openalex.org/works/W7211995705.grobid-xml"},"referenced_works_count":5,"referenced_works":["https://openalex.org/W1971713783","https://openalex.org/W3127151792","https://openalex.org/W4211077197","https://openalex.org/W4385571886","https://openalex.org/W4388979610"],"related_works":["https://openalex.org/W7213358501","https://openalex.org/W4405767785","https://openalex.org/W7167828984","https://openalex.org/W6893319214","https://openalex.org/W7165650776","https://openalex.org/W7152715660","https://openalex.org/W7203441366","https://openalex.org/W7221059447","https://openalex.org/W7172522578","https://openalex.org/W7164358502"],"abstract_inverted_index":{"We":[0,93],"derive":[1,94],"an":[2],"exact":[3,89,117],"gradient-step":[4,74],"representation":[5,75,133],"of":[6,42,101],"the":[7,39,43,49,56,65,85,99,132,136],"RoPE-softmax":[8,14],"forward":[9],"pass.":[10],"For":[11],"every":[12],"deterministic":[13],"attention":[15],"head":[16,112],"with":[17],"arbitrary":[18],"affine":[19,118],"projection":[20],"weights,":[21],"we":[22],"construct":[23],"a":[24,72,77,95,109,115],"query-dependent":[25],"effective":[26],"matrix":[27,105,142],"$\u0394M_i$":[28],"satisfying":[29],"$y_i":[30],"=":[31],"\u03bc_i":[32],"+":[33],"u_i^\\top":[34],"\u0394M_i$,":[35],"where":[36],"$\u03bc_i$":[37],"is":[38,48,143],"uniform":[40],"mean":[41],"attended":[44],"values":[45],"and":[46,106,123,134],"$u_i$":[47],"augmented":[50],"query":[51,119],"input.":[52],"The":[53,81],"construction":[54],"applies":[55],"classical":[57],"exponential":[58],"divided":[59],"difference":[60],"$\u03c1=":[61],"\u03d5_1$":[62],"to":[63,88],"retain":[64],"softmax":[66],"exactly.":[67],"Its":[68],"positive":[69],"coefficients":[70],"give":[71],"unit":[73],"on":[76,126],"query-conditioned":[78],"quadratic":[79],"objective.":[80],"same":[82],"function":[83],"connects":[84],"RoPE":[86],"generator":[87],"positional":[90],"finite":[91],"differences.":[92],"tokenwise":[96],"formula":[97],"for":[98],"error":[100],"reusing":[102],"one":[103,127,140],"query's":[104,141],"prove":[107],"that":[108],"nonconstant":[110],"finite-cache":[111],"cannot":[113],"admit":[114],"globally":[116],"readout.":[120],"Reconstruction":[121],"checks":[122],"frozen-reuse":[124],"calibration":[125],"pretrained":[128],"Qwen2.5-0.5B":[129],"layer":[130],"verify":[131],"quantify":[135],"correction":[137],"required":[138],"when":[139],"reused.":[144]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-10T00:00:00"}