{"paper":{"title":"Force-Aware Neural Tangent Kernels for Scalable and Robust Active Learning of MLIPs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Mixed parameter-coordinate derivatives of the NTK yield natural similarity metrics for force-aware active learning in MLIPs.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Eszter Varga-Umbrich, Jules Tilly, Olivier Peltre, Paul Duckworth, Shikha Surana, Zachary Weller-Davies","submitted_at":"2026-05-13T17:08:37Z","abstract_excerpt":"Active learning for machine-learning interatomic potentials (MLIPs) must address several challenges to be practical: scaling to large candidate pools, leveraging energy-force supervision, and maintaining robustness when candidate pools are biased relative to the target distribution. In this work, we jointly address these challenges. We first introduce a linearly scaling acquisition framework based on chunked feature-space posterior-variance shortlisting. By avoiding materialisation of the candidate and train set kernels, this approach enables screening of ~200k structures within hours and appl"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"we demonstrate the effectiveness of the joint energy-force NTK on the OC20 dataset, where force-aware acquisition is crucial: it achieves the lowest energy and force MAE and RMSE across all metrics and distribution splits.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The mixed parameter-coordinate derivatives of the NTK yield effective natural similarity metrics for vector-field force prediction in pretrained MLIPs without requiring additional fitting or validation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Force-aware NTKs and chunked acquisition enable scalable, robust active learning for MLIPs, achieving lowest energy and force errors on OC20 and remaining competitive on other benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Mixed parameter-coordinate derivatives of the NTK yield natural similarity metrics for force-aware active learning in MLIPs.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"1d83d92efc6ccc1f4ce3b64c7ecd06e3fba990b7b1f67ecc39c212f6ea53bd1b"},"source":{"id":"2605.13788","kind":"arxiv","version":1},"verdict":{"id":"5b1c0eb5-5329-4239-89f7-9d5d93b5652b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:07:33.136069Z","strongest_claim":"we demonstrate the effectiveness of the joint energy-force NTK on the OC20 dataset, where force-aware acquisition is crucial: it achieves the lowest energy and force MAE and RMSE across all metrics and distribution splits.","one_line_summary":"Force-aware NTKs and chunked acquisition enable scalable, robust active learning for MLIPs, achieving lowest energy and force errors on OC20 and remaining competitive on other benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The mixed parameter-coordinate derivatives of the NTK yield effective natural similarity metrics for vector-field force prediction in pretrained MLIPs without requiring additional fitting or validation.","pith_extraction_headline":"Mixed parameter-coordinate derivatives of the NTK yield natural similarity metrics for force-aware active learning in MLIPs."},"references":{"count":123,"sample":[{"doi":"10.1021/acs.chemrev.4c00572","year":2025,"title":"2024 , journal =","work_id":"75e7d7fb-17e5-44ed-b142-679ca5232de4","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1103/physrevb.87.184115","year":2013,"title":"On representing chemical environments","work_id":"e9773ddd-1a8c-4ce0-81d4-f4cff1f037ff","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"\\ Elena , author Dávid P","work_id":"29bfbd6f-762e-4448-8f07-4f24bf553e51","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Ilyes Batatia, Dávid Péter Kovács, Gregor N. C. Simm, Christoph Ortner, and Gábor Csányi. Mace: Higher order equivariant message passing neural networks for fast and accurate force fields, 2023 b . UR","work_id":"c256ce9a-079f-49e3-ac8e-5c51a3d9dcf6","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1038/s41467-022-29939-5","year":2022,"title":"E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials","work_id":"f52ce8d5-e753-4360-a60f-60c8faae7c16","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":123,"snapshot_sha256":"7cc1416fd29e94a58d518b6249caab5e97ddec752fdfeab004c37d88e9c5b732","internal_anchors":1},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}