{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:GIDDWI6GPXZGBOP4XNF2RZ4K6F","short_pith_number":"pith:GIDDWI6G","canonical_record":{"source":{"id":"2605.12801","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"math.NA","submitted_at":"2026-05-12T22:40:16Z","cross_cats_sorted":["cs.NA"],"title_canon_sha256":"0e39afe8acce9c0af3e4e100532a4626c408b466cb73c24dbd593c3c15da7b1d","abstract_canon_sha256":"04f207fc71768aa24fb192bf2342fc9bc85e8507f3a4adae1714e3da8306e9ad"},"schema_version":"1.0"},"canonical_sha256":"32063b23c67df260b9fcbb4ba8e78af17deaeb8535a326a7f0cc3118cdb1073e","source":{"kind":"arxiv","id":"2605.12801","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12801","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12801v1","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12801","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"pith_short_12","alias_value":"GIDDWI6GPXZG","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"GIDDWI6GPXZGBOP4","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"GIDDWI6G","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:GIDDWI6GPXZGBOP4XNF2RZ4K6F","target":"record","payload":{"canonical_record":{"source":{"id":"2605.12801","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"math.NA","submitted_at":"2026-05-12T22:40:16Z","cross_cats_sorted":["cs.NA"],"title_canon_sha256":"0e39afe8acce9c0af3e4e100532a4626c408b466cb73c24dbd593c3c15da7b1d","abstract_canon_sha256":"04f207fc71768aa24fb192bf2342fc9bc85e8507f3a4adae1714e3da8306e9ad"},"schema_version":"1.0"},"canonical_sha256":"32063b23c67df260b9fcbb4ba8e78af17deaeb8535a326a7f0cc3118cdb1073e","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:09:12.715886Z","signature_b64":"DMTM9kZDfIerwHxf0UcD2oW6r32E5ZTJqLvHdtFqZzR1I9JjQuDbeKL/0uqbtD/XyWpDyjTvCg13xT7Ki1jTAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"32063b23c67df260b9fcbb4ba8e78af17deaeb8535a326a7f0cc3118cdb1073e","last_reissued_at":"2026-05-18T03:09:12.715358Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:09:12.715358Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.12801","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:09:12Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"g+7PVrfYBsUUhu94wH0Kq2DufEjJoWBfdvEjbMSppT1DlT9zYvo+5o0lOP0cIgALU4QH956T1/mHnxhoyBK/AQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T21:00:42.293771Z"},"content_sha256":"57941589af8ead1b96ec6f3f080c31d97d59c2fd0ef8cb1fbfdc9ff12323e467","schema_version":"1.0","event_id":"sha256:57941589af8ead1b96ec6f3f080c31d97d59c2fd0ef8cb1fbfdc9ff12323e467"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:GIDDWI6GPXZGBOP4XNF2RZ4K6F","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Fast and Stable Gradient Approximation for Bilinear Forms of Hermitian Matrix Functions","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A forward-only gradient approximation reuses the Lanczos pass to stably differentiate bilinear forms u^T f(A(θ))v for Hermitian A with error proportional to the residual norm.","cross_cats":["cs.NA"],"primary_cat":"math.NA","authors_text":"Kipton Barros, Navjot Singh, Xiaoye Sherry Li","submitted_at":"2026-05-12T22:40:16Z","abstract_excerpt":"Objectives involving bilinear forms $u^\\top f(A(\\theta))v$ for Hermitian $A$ arise widely in scientific computing and probabilistic machine learning. For large matrices, Lanczos efficiently approximates these quantities, but differentiating them with respect to $\\theta$ is challenging. Existing approaches either backpropagate through the Lanczos recurrence, requiring reorthogonalization for stability, or apply Arnoldi to an augmented block matrix of twice the original size. Both introduce extra computation and orthogonalization costs that can limit performance on modern hardware. We propose a "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We propose a forward-only gradient approximation that reuses the Lanczos pass and adds very minimal overhead in most cases. We prove that its error is proportional to the Lanczos residual norm, the same quantity controlling the forward approximation. Whereas a traditional adjoint-based calculation would be unstable without reorthogonalization, the new method appears unconditionally stable in our tests.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The approach assumes A(θ) remains Hermitian so that Lanczos produces real eigenvalues and orthogonal vectors, and that the residual norm from the forward Lanczos pass is a reliable error indicator for both the value and the gradient; stability is reported from tests but the abstract does not specify the range of matrices or functions f where this holds without additional conditions.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A forward-only Lanczos gradient approximation for Hermitian matrix function bilinear forms whose error scales with the same residual norm as the forward approximation and appears stable without reorthogonalization.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A forward-only gradient approximation reuses the Lanczos pass to stably differentiate bilinear forms u^T f(A(θ))v for Hermitian A with error proportional to the residual norm.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"099caa02bb0260f37ec9f62ee35a526db3c5ef6ba5d70226b323ed21b8e36031"},"source":{"id":"2605.12801","kind":"arxiv","version":1},"verdict":{"id":"eafc3d2f-9fd8-4753-8a33-c67a8dea1976","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:28:58.810999Z","strongest_claim":"We propose a forward-only gradient approximation that reuses the Lanczos pass and adds very minimal overhead in most cases. We prove that its error is proportional to the Lanczos residual norm, the same quantity controlling the forward approximation. Whereas a traditional adjoint-based calculation would be unstable without reorthogonalization, the new method appears unconditionally stable in our tests.","one_line_summary":"A forward-only Lanczos gradient approximation for Hermitian matrix function bilinear forms whose error scales with the same residual norm as the forward approximation and appears stable without reorthogonalization.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The approach assumes A(θ) remains Hermitian so that Lanczos produces real eigenvalues and orthogonal vectors, and that the residual norm from the forward Lanczos pass is a reliable error indicator for both the value and the gradient; stability is reported from tests but the abstract does not specify the range of matrices or functions f where this holds without additional conditions.","pith_extraction_headline":"A forward-only gradient approximation reuses the Lanczos pass to stably differentiate bilinear forms u^T f(A(θ))v for Hermitian A with error proportional to the residual norm."},"references":{"count":42,"sample":[{"doi":"10.1137/080716426","year":2009,"title":"Al-Mohy and Nicholas J","work_id":"256900b9-9dc5-4470-a0e1-8b1b1c68468a","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Learning quantum systems.Nature Reviews Physics, 5:141–156, 2023","work_id":"0af8d3eb-3b7c-4734-9ac1-646988e2fd47","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1137/080741744","year":2009,"title":"Error estimates and evaluation of matrix functions via the Faber transform.SIAM Journal on Numerical Analysis, 47(5):3849–3883, 2009","work_id":"d470a323-11f4-40ba-ab88-10f3b8842689","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Matrix functions in network analysis.GAMM-Mitteilungen, 43 (3):e202000012, 2020","work_id":"1e835a40-5358-41e4-b572-0dbedcc11934","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1137/22m1526357","year":2023,"title":"Krylov-aware stochastic trace estimation.SIAM Journal on Matrix Analysis and Applications, 44(3):1218–1244, 2023","work_id":"a1573fb9-1aa6-47d8-b0e6-0c41c145bca3","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":42,"snapshot_sha256":"25bad2aa1de10dc1098ee862598fbc580db58620f9c4f09f8b8e7e1a2d30bfaa","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"4532b64119779b7eadd845e43a9029610bd5b1bf930968469979a04a88cbbca8"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"eafc3d2f-9fd8-4753-8a33-c67a8dea1976"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:09:12Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"vIeae35WJ+GlREzdMBtsIVQ1KYGdjcCspP85EdGtDGZjLlLTc3BzT7IXXq2nhjNuxZ6wRroVAUEG5LmdHsB6Dw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T21:00:42.294956Z"},"content_sha256":"7a5e8f08efcf9eed4b4d66b197e0610ec42d590d1244ffbba40cb38d0dd8f10b","schema_version":"1.0","event_id":"sha256:7a5e8f08efcf9eed4b4d66b197e0610ec42d590d1244ffbba40cb38d0dd8f10b"},{"event_type":"integrity_finding","subject_pith_number":"pith:2026:GIDDWI6GPXZGBOP4XNF2RZ4K6F","target":"integrity","payload":{"note":"Identifier '10.1137/18m1172319' is syntactically valid but the DOI registry (doi.org) returned 404, and Crossref / OpenAlex / internal corpus also have no record. The cited work could not be located through any authoritative source.","snippet":"Stefano Pozza and Francesco Tudisco. On the stability of network indices defined by means of matrix functions.SIAM Journal on Matrix Analysis and Applications, 39(4):1521–1546, 2018. doi: 10.1137/18M1172319","arxiv_id":"2605.12801","detector":"doi_compliance","evidence":{"doi":"10.1137/18m1172319","arxiv_id":null,"ref_index":34,"raw_excerpt":"Stefano Pozza and Francesco Tudisco. On the stability of network indices defined by means of matrix functions.SIAM Journal on Matrix Analysis and Applications, 39(4):1521–1546, 2018. doi: 10.1137/18M1172319","verdict_class":"cross_source","checked_sources":["crossref_by_doi","openalex_by_doi","doi_org_head"]},"severity":"critical","ref_index":34,"audited_at":"2026-05-19T07:07:51.198415Z","event_type":"pith.integrity.v1","detected_doi":"10.1137/18m1172319","detector_url":"https://pith.science/pith-integrity-protocol#doi_compliance","external_url":null,"finding_type":"unresolvable_identifier","evidence_hash":"1407f01f242dc5e11c1cdd4b7006bd42bcf6a5e4dbbc767c3a2a268d760eba65","paper_version":1,"verdict_class":"cross_source","resolved_title":null,"detector_version":"1.0.0","detected_arxiv_id":null,"integrity_event_id":263,"payload_sha256":"55d95ab94f00779ebdce7a3feb05498e9c2ca36005ba79f16dfc009f7a92eebc","signature_b64":"/d4XTlTIQeRTndiBORfqXN3ZUOsN691hie28r3K/DPhgKZOu3vZoPPkn75V20mRhtbRs7xZo5ojK5y+069tuAw==","signing_key_id":"pith-v1-2026-05"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-19T07:11:54Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"0VplZLQBavmsjszGl5KjNX/iXmoIqUTdHxMLIeKAxYjjVFGNQc+SEMlTI7TS1pNjg53jYxPtj09EkZoYljhHCw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T21:00:42.296502Z"},"content_sha256":"7b59ff7bf144afb0e2ad41e4e6fe4f88e384fa257a6f64acaa0be0938fb22246","schema_version":"1.0","event_id":"sha256:7b59ff7bf144afb0e2ad41e4e6fe4f88e384fa257a6f64acaa0be0938fb22246"},{"event_type":"integrity_finding","subject_pith_number":"pith:2026:GIDDWI6GPXZGBOP4XNF2RZ4K6F","target":"integrity","payload":{"note":"Identifier '10.1137/22m1503075' is syntactically valid but the DOI registry (doi.org) returned 404, and Crossref / OpenAlex / internal corpus also have no record. The cited work could not be located through any authoritative source.","snippet":"Peter Kandolf, Antti Koskela, Samuel D. Relton, and Marcel Schweitzer. Computing low- rank approximations of the Fréchet derivative of a matrix function using krylov subspace methods.SIAM Journal on Matrix Analysis and Applications, 44(3):1","arxiv_id":"2605.12801","detector":"doi_compliance","evidence":{"doi":"10.1137/22m1503075","arxiv_id":null,"ref_index":21,"raw_excerpt":"Peter Kandolf, Antti Koskela, Samuel D. Relton, and Marcel Schweitzer. Computing low- rank approximations of the Fréchet derivative of a matrix function using krylov subspace methods.SIAM Journal on Matrix Analysis and Applications, 44(3):1115–1139, 2023. doi: 10.1137/22M1503075","verdict_class":"cross_source","checked_sources":["crossref_by_doi","openalex_by_doi","doi_org_head"]},"severity":"critical","ref_index":21,"audited_at":"2026-05-19T07:07:51.198415Z","event_type":"pith.integrity.v1","detected_doi":"10.1137/22m1503075","detector_url":"https://pith.science/pith-integrity-protocol#doi_compliance","external_url":null,"finding_type":"unresolvable_identifier","evidence_hash":"d0315fa521710e8f32743b7b4925396bb6c5ce6cd57ba75c652b113a86a3e401","paper_version":1,"verdict_class":"cross_source","resolved_title":null,"detector_version":"1.0.0","detected_arxiv_id":null,"integrity_event_id":262,"payload_sha256":"380bc04c5ecac029f6b5dfbd998d6b6b9d63c8a79b730f22a98cc237a7acd893","signature_b64":"6CH2VyLBctFt28pvFlXQ5IVFBxSXMu6fBev5yDHhzX9Kr890EGCh77Ou+THvLjuccBa7jqcatHUnlGzrX5pDBw==","signing_key_id":"pith-v1-2026-05"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-19T07:11:54Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"pH57hh9mjTC0coToFXsS4x3gcxtmXKzFIl8+kMJ6zWD3SwDP3nsPs72MoImQMQawP3hF0djbi8KflAUSlSYXCA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T21:00:42.297144Z"},"content_sha256":"6268272e7673f20321b9e3fc446d310c70d4c6f33adbf17b0711d5529994f4d7","schema_version":"1.0","event_id":"sha256:6268272e7673f20321b9e3fc446d310c70d4c6f33adbf17b0711d5529994f4d7"},{"event_type":"integrity_finding","subject_pith_number":"pith:2026:GIDDWI6GPXZGBOP4XNF2RZ4K6F","target":"integrity","payload":{"note":"Identifier '10.1137/20m1363868' is syntactically valid but the DOI registry (doi.org) returned 404, and Crossref / OpenAlex / internal corpus also have no record. The cited work could not be located through any authoritative source.","snippet":"Alexander Davies and Nicholas J. Higham. Computingf(a)b for matrix functions f.SIAM Jour- nal on Matrix Analysis and Applications, 42(2):1084–1105, 2021. doi: 10.1137/20M1363868","arxiv_id":"2605.12801","detector":"doi_compliance","evidence":{"doi":"10.1137/20m1363868","arxiv_id":null,"ref_index":7,"raw_excerpt":"Alexander Davies and Nicholas J. Higham. Computingf(a)b for matrix functions f.SIAM Jour- nal on Matrix Analysis and Applications, 42(2):1084–1105, 2021. doi: 10.1137/20M1363868","verdict_class":"cross_source","checked_sources":["crossref_by_doi","openalex_by_doi","doi_org_head"]},"severity":"critical","ref_index":7,"audited_at":"2026-05-19T07:07:51.198415Z","event_type":"pith.integrity.v1","detected_doi":"10.1137/20m1363868","detector_url":"https://pith.science/pith-integrity-protocol#doi_compliance","external_url":null,"finding_type":"unresolvable_identifier","evidence_hash":"918d85b6ecbbab34ff9f3c86d94e78f81fee071a234933f1775575e269e8f0ae","paper_version":1,"verdict_class":"cross_source","resolved_title":null,"detector_version":"1.0.0","detected_arxiv_id":null,"integrity_event_id":261,"payload_sha256":"044585609a1a8b8fbddf9a8a3ad5b12a141256226a062dc248157d17a9d1f420","signature_b64":"lt78fgu54fz157LbSMxjJq4OqDLMRo0ZHRQZAadn/a9/LLYlWSmB4u+pcGumHYhztJI6hTrITlnfvAHQTHJ0DQ==","signing_key_id":"pith-v1-2026-05"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-19T07:11:54Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"7ggf6rCr1pyVd+ovEm10noT48DfIGL11j6S+rUOrYJQ1p5Y/azZJhV+UOh1iFIICxaFq/QJi3YdzfV7VimPWDw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T21:00:42.297821Z"},"content_sha256":"f04567efb8cb0fecfc84f394ef3bc7c8d58fcd6e90d65d118301e368d2f9faa2","schema_version":"1.0","event_id":"sha256:f04567efb8cb0fecfc84f394ef3bc7c8d58fcd6e90d65d118301e368d2f9faa2"},{"event_type":"integrity_finding","subject_pith_number":"pith:2026:GIDDWI6GPXZGBOP4XNF2RZ4K6F","target":"integrity","payload":{"note":"Identifier '10.1137/22m1526357' is syntactically valid but the DOI registry (doi.org) returned 404, and Crossref / OpenAlex / internal corpus also have no record. The cited work could not be located through any authoritative source.","snippet":"Tyler Chen and Eric Hallman. Krylov-aware stochastic trace estimation.SIAM Journal on Matrix Analysis and Applications, 44(3):1218–1244, 2023. doi: 10.1137/22M1526357","arxiv_id":"2605.12801","detector":"doi_compliance","evidence":{"doi":"10.1137/22m1526357","arxiv_id":null,"ref_index":5,"raw_excerpt":"Tyler Chen and Eric Hallman. Krylov-aware stochastic trace estimation.SIAM Journal on Matrix Analysis and Applications, 44(3):1218–1244, 2023. doi: 10.1137/22M1526357","verdict_class":"cross_source","checked_sources":["crossref_by_doi","openalex_by_doi","doi_org_head"]},"severity":"critical","ref_index":5,"audited_at":"2026-05-19T07:07:51.198415Z","event_type":"pith.integrity.v1","detected_doi":"10.1137/22m1526357","detector_url":"https://pith.science/pith-integrity-protocol#doi_compliance","external_url":null,"finding_type":"unresolvable_identifier","evidence_hash":"c5c82262c354421f1ee467a0c53297da1f005c830674f897e10ea59e82ad01b1","paper_version":1,"verdict_class":"cross_source","resolved_title":null,"detector_version":"1.0.0","detected_arxiv_id":null,"integrity_event_id":260,"payload_sha256":"a503c12f789c7bbd8b31c5a641ef7069e99cc008b4c395100d04b548730176ad","signature_b64":"upWLCG0CJKy4y1yxHY+JttidDczVpRWEhkEsy44G828cy5n9pPvrn76KeW0xtUuyi0aCNGVZihk8Fog+MZdTCw==","signing_key_id":"pith-v1-2026-05"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-19T07:11:53Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"CQXnu5gR60KKD5xbKqd00ezi+nRbK/Q2P6zwe0xn3JbdTy6zYXGkDvWR6un2R8dSppC8L4zNWUDNNYH8UFEuBg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T21:00:42.298388Z"},"content_sha256":"eb390289d911355bfdcf3de16dee30263664ec8745ab4a60e73565bece631149","schema_version":"1.0","event_id":"sha256:eb390289d911355bfdcf3de16dee30263664ec8745ab4a60e73565bece631149"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/GIDDWI6GPXZGBOP4XNF2RZ4K6F/bundle.json","state_url":"https://pith.science/pith/GIDDWI6GPXZGBOP4XNF2RZ4K6F/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/GIDDWI6GPXZGBOP4XNF2RZ4K6F/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-21T21:00:42Z","links":{"resolver":"https://pith.science/pith/GIDDWI6GPXZGBOP4XNF2RZ4K6F","bundle":"https://pith.science/pith/GIDDWI6GPXZGBOP4XNF2RZ4K6F/bundle.json","state":"https://pith.science/pith/GIDDWI6GPXZGBOP4XNF2RZ4K6F/state.json","well_known_bundle":"https://pith.science/.well-known/pith/GIDDWI6GPXZGBOP4XNF2RZ4K6F/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:GIDDWI6GPXZGBOP4XNF2RZ4K6F","merge_version":"pith-open-graph-merge-v1","event_count":6,"valid_event_count":6,"invalid_event_count":0,"equivocation_count":1,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"04f207fc71768aa24fb192bf2342fc9bc85e8507f3a4adae1714e3da8306e9ad","cross_cats_sorted":["cs.NA"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"math.NA","submitted_at":"2026-05-12T22:40:16Z","title_canon_sha256":"0e39afe8acce9c0af3e4e100532a4626c408b466cb73c24dbd593c3c15da7b1d"},"schema_version":"1.0","source":{"id":"2605.12801","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12801","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12801v1","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12801","created_at":"2026-05-18T03:09:12Z"},{"alias_kind":"pith_short_12","alias_value":"GIDDWI6GPXZG","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"GIDDWI6GPXZGBOP4","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"GIDDWI6G","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:7a5e8f08efcf9eed4b4d66b197e0610ec42d590d1244ffbba40cb38d0dd8f10b","target":"graph","created_at":"2026-05-18T03:09:12Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"We propose a forward-only gradient approximation that reuses the Lanczos pass and adds very minimal overhead in most cases. We prove that its error is proportional to the Lanczos residual norm, the same quantity controlling the forward approximation. Whereas a traditional adjoint-based calculation would be unstable without reorthogonalization, the new method appears unconditionally stable in our tests."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The approach assumes A(θ) remains Hermitian so that Lanczos produces real eigenvalues and orthogonal vectors, and that the residual norm from the forward Lanczos pass is a reliable error indicator for both the value and the gradient; stability is reported from tests but the abstract does not specify the range of matrices or functions f where this holds without additional conditions."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"A forward-only Lanczos gradient approximation for Hermitian matrix function bilinear forms whose error scales with the same residual norm as the forward approximation and appears stable without reorthogonalization."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"A forward-only gradient approximation reuses the Lanczos pass to stably differentiate bilinear forms u^T f(A(θ))v for Hermitian A with error proportional to the residual norm."}],"snapshot_sha256":"099caa02bb0260f37ec9f62ee35a526db3c5ef6ba5d70226b323ed21b8e36031"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"4532b64119779b7eadd845e43a9029610bd5b1bf930968469979a04a88cbbca8"},"paper":{"abstract_excerpt":"Objectives involving bilinear forms $u^\\top f(A(\\theta))v$ for Hermitian $A$ arise widely in scientific computing and probabilistic machine learning. For large matrices, Lanczos efficiently approximates these quantities, but differentiating them with respect to $\\theta$ is challenging. Existing approaches either backpropagate through the Lanczos recurrence, requiring reorthogonalization for stability, or apply Arnoldi to an augmented block matrix of twice the original size. Both introduce extra computation and orthogonalization costs that can limit performance on modern hardware. We propose a ","authors_text":"Kipton Barros, Navjot Singh, Xiaoye Sherry Li","cross_cats":["cs.NA"],"headline":"A forward-only gradient approximation reuses the Lanczos pass to stably differentiate bilinear forms u^T f(A(θ))v for Hermitian A with error proportional to the residual norm.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"math.NA","submitted_at":"2026-05-12T22:40:16Z","title":"Fast and Stable Gradient Approximation for Bilinear Forms of Hermitian Matrix Functions"},"references":{"count":42,"internal_anchors":0,"resolved_work":42,"sample":[{"cited_arxiv_id":"","doi":"10.1137/080716426","is_internal_anchor":false,"ref_index":1,"title":"Al-Mohy and Nicholas J","work_id":"256900b9-9dc5-4470-a0e1-8b1b1c68468a","year":2009},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Learning quantum systems.Nature Reviews Physics, 5:141–156, 2023","work_id":"0af8d3eb-3b7c-4734-9ac1-646988e2fd47","year":2023},{"cited_arxiv_id":"","doi":"10.1137/080741744","is_internal_anchor":false,"ref_index":3,"title":"Error estimates and evaluation of matrix functions via the Faber transform.SIAM Journal on Numerical Analysis, 47(5):3849–3883, 2009","work_id":"d470a323-11f4-40ba-ab88-10f3b8842689","year":2009},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Matrix functions in network analysis.GAMM-Mitteilungen, 43 (3):e202000012, 2020","work_id":"1e835a40-5358-41e4-b572-0dbedcc11934","year":2020},{"cited_arxiv_id":"","doi":"10.1137/22m1526357","is_internal_anchor":false,"ref_index":5,"title":"Krylov-aware stochastic trace estimation.SIAM Journal on Matrix Analysis and Applications, 44(3):1218–1244, 2023","work_id":"a1573fb9-1aa6-47d8-b0e6-0c41c145bca3","year":2023}],"snapshot_sha256":"25bad2aa1de10dc1098ee862598fbc580db58620f9c4f09f8b8e7e1a2d30bfaa"},"source":{"id":"2605.12801","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T19:28:58.810999Z","id":"eafc3d2f-9fd8-4753-8a33-c67a8dea1976","model_set":{"reader":"grok-4.3"},"one_line_summary":"A forward-only Lanczos gradient approximation for Hermitian matrix function bilinear forms whose error scales with the same residual norm as the forward approximation and appears stable without reorthogonalization.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"A forward-only gradient approximation reuses the Lanczos pass to stably differentiate bilinear forms u^T f(A(θ))v for Hermitian A with error proportional to the residual norm.","strongest_claim":"We propose a forward-only gradient approximation that reuses the Lanczos pass and adds very minimal overhead in most cases. We prove that its error is proportional to the Lanczos residual norm, the same quantity controlling the forward approximation. Whereas a traditional adjoint-based calculation would be unstable without reorthogonalization, the new method appears unconditionally stable in our tests.","weakest_assumption":"The approach assumes A(θ) remains Hermitian so that Lanczos produces real eigenvalues and orthogonal vectors, and that the residual norm from the forward Lanczos pass is a reliable error indicator for both the value and the gradient; stability is reported from tests but the abstract does not specify the range of matrices or functions f where this holds without additional conditions."}},"verdict_id":"eafc3d2f-9fd8-4753-8a33-c67a8dea1976"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:57941589af8ead1b96ec6f3f080c31d97d59c2fd0ef8cb1fbfdc9ff12323e467","target":"record","created_at":"2026-05-18T03:09:12Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"04f207fc71768aa24fb192bf2342fc9bc85e8507f3a4adae1714e3da8306e9ad","cross_cats_sorted":["cs.NA"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"math.NA","submitted_at":"2026-05-12T22:40:16Z","title_canon_sha256":"0e39afe8acce9c0af3e4e100532a4626c408b466cb73c24dbd593c3c15da7b1d"},"schema_version":"1.0","source":{"id":"2605.12801","kind":"arxiv","version":1}},"canonical_sha256":"32063b23c67df260b9fcbb4ba8e78af17deaeb8535a326a7f0cc3118cdb1073e","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"32063b23c67df260b9fcbb4ba8e78af17deaeb8535a326a7f0cc3118cdb1073e","first_computed_at":"2026-05-18T03:09:12.715358Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:09:12.715358Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"DMTM9kZDfIerwHxf0UcD2oW6r32E5ZTJqLvHdtFqZzR1I9JjQuDbeKL/0uqbtD/XyWpDyjTvCg13xT7Ki1jTAg==","signature_status":"signed_v1","signed_at":"2026-05-18T03:09:12.715886Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.12801","source_kind":"arxiv","source_version":1}}},"equivocations":[{"signer_id":"pith.science","event_type":"integrity_finding","target":"integrity","event_ids":["sha256:6268272e7673f20321b9e3fc446d310c70d4c6f33adbf17b0711d5529994f4d7","sha256:7b59ff7bf144afb0e2ad41e4e6fe4f88e384fa257a6f64acaa0be0938fb22246","sha256:eb390289d911355bfdcf3de16dee30263664ec8745ab4a60e73565bece631149","sha256:f04567efb8cb0fecfc84f394ef3bc7c8d58fcd6e90d65d118301e368d2f9faa2"]}],"invalid_events":[],"applied_event_ids":["sha256:57941589af8ead1b96ec6f3f080c31d97d59c2fd0ef8cb1fbfdc9ff12323e467","sha256:7a5e8f08efcf9eed4b4d66b197e0610ec42d590d1244ffbba40cb38d0dd8f10b"],"state_sha256":"3cd48d444058b985b55a7c25b649692153547381e71cc9f90c9f1bb76825b094"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"HZF430ALdBYxMGW8VJQF3oNNLfkPuhuVf9aTu0k9kAU1WjxwZBqRCzbf+WczD7Jbw9H2jTcUjjpnZKJQUKpECg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-21T21:00:42.302058Z","bundle_sha256":"565adba4dfc89954df096422e2441f7dcc96e1d6ea5e180ce13c06a14fb89342"}}