{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,8,1]],"date-time":"2026-08-01T01:57:31Z","timestamp":1785549451900,"version":"3.56.0"},"reference-count":63,"publisher":"IOP Publishing","issue":"2","license":[{"start":{"date-parts":[[2026,2,16]],"date-time":"2026-02-16T00:00:00Z","timestamp":1771200000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"},{"start":{"date-parts":[[2026,2,16]],"date-time":"2026-02-16T00:00:00Z","timestamp":1771200000000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/iopscience.iop.org\/info\/page\/text-and-data-mining"},{"start":{"date-parts":[[2026,2,16]],"date-time":"2026-02-16T00:00:00Z","timestamp":1771200000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/publishingsupport.iopscience.iop.org\/questions\/alternative-author-rights-policies\/"},{"start":{"date-parts":[[2026,2,16]],"date-time":"2026-02-16T00:00:00Z","timestamp":1771200000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/publishingsupport.iopscience.iop.org\/questions\/alternative-author-rights-policies\/"}],"funder":[{"name":"UCL Cosmoparticle Initiative"},{"name":"Research Capital Investment Fund"}],"content-domain":{"domain":["iopscience.iop.org"],"crossmark-restriction":false},"short-container-title":["Mach. Learn.: Sci. Technol."],"published-print":{"date-parts":[[2026,4,1]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>Training machine learning interatomic potentials (MLIPs) on total energies of molecular clusters using differential or transfer learning is becoming a popular route to extend the accuracy of correlated wave-function theory to condensed phases. A key challenge, however, lies in validation, as reference observables in finite-temperature ensembles are not available at the reference level. Here, we construct synthetic reference data from pretrained MLIPs and evaluate the generalizability of cluster-trained models on ice-Ih, considering scenarios where both energies and forces and where only energies are available for training. We study the accuracy and data-efficiency of differential, single-fidelity transfer, and multi-fidelity transfer learning against ground-truth thermodynamic observables. We find that transferring accuracy from clusters to bulk requires regularization, which is best achieved through multi-fidelity transfer learning when training on both energies and forces. By contrast, training only on energies introduces artefacts: stable trajectories and low energy errors conceal large force errors, leading to inaccurate microscopic observables. More broadly, we show that accurate reproduction of microscopic structure correlates strongly with low force errors but only weakly with energy errors, whereas global properties such as energies and densities correlate with low energy errors. This highlights the need to incorporate forces during training or to apply careful validation before production. Our results highlight the promise and pitfalls of cluster-trained MLIPs for condensed phases and provide guidelines for developing\u2014and critically, validating\u2014robust and data-efficient MLIPs.<\/jats:p>","DOI":"10.1088\/2632-2153\/ae3c57","type":"journal-article","created":{"date-parts":[[2026,1,22]],"date-time":"2026-01-22T22:53:12Z","timestamp":1769122392000},"page":"025004","update-policy":"https:\/\/doi.org\/10.1088\/crossmark-policy","source":"Crossref","is-referenced-by-count":5,"title":["The good, the bad, and the ugly of atomistic learning for \u2018clusters-to-bulk\u2019 generalization"],"prefix":"10.1088","volume":"7","author":[{"ORCID":"https:\/\/orcid.org\/0009-0001-8717-3793","authenticated-orcid":true,"given":"Miko\u0142aj J","family":"Gawkowski","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0000-3289-6268","authenticated-orcid":true,"given":"Mingjia","family":"Li","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-3272-0996","authenticated-orcid":false,"given":"Benjamin X","family":"Shi","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0324-2198","authenticated-orcid":true,"given":"Venkat","family":"Kapil","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"266","published-online":{"date-parts":[[2026,2,16]]},"reference":[{"key":"mlstae3c57bib1","doi-asserted-by":"publisher","DOI":"10.1021\/jacs.4c07099","type":"journal-article","volume":"147","author":"Kov\u00e1cs","year":"2025","journal-title":"J. Am. Chem. Soc."},{"key":"mlstae3c57bib2","article-title":"A foundation model for atomistic materials chemistry","author":"Batatia","year":"2024","type":"preprint"},{"key":"mlstae3c57bib3","doi-asserted-by":"publisher","first-page":"80","DOI":"10.1038\/s41586-023-06735-9","type":"journal-article","volume":"624","author":"Merchant","year":"2023","journal-title":"Nature"},{"key":"mlstae3c57bib4","article-title":"MatterSim: a deep learning atomistic model across elements, temperatures and pressures","author":"Yang","year":"2024","type":"preprint"},{"key":"mlstae3c57bib5","doi-asserted-by":"publisher","first-page":"eadn4397","DOI":"10.1126\/sciadv.adn4397","type":"journal-article","volume":"10","author":"Unke","year":"2024","journal-title":"Sci. Adv."},{"key":"mlstae3c57bib6","doi-asserted-by":"publisher","first-page":"2353","DOI":"10.1021\/acs.jpclett.8b00242","type":"journal-article","volume":"9","author":"Bao","year":"2018","journal-title":"J. Phys. Chem. Lett."},{"key":"mlstae3c57bib7","doi-asserted-by":"publisher","DOI":"10.1063\/5.0087031","type":"journal-article","volume":"156","author":"Shi","year":"2022","journal-title":"J. Chem. Phys."},{"key":"mlstae3c57bib8","doi-asserted-by":"publisher","first-page":"e1631","DOI":"10.1002\/wcms.1631","type":"journal-article","volume":"13","author":"Bryenton","year":"2023","journal-title":"WIREs Comput. Mol. Sci."},{"key":"mlstae3c57bib9","doi-asserted-by":"publisher","first-page":"5306","DOI":"10.1021\/acs.jctc.4c00379","type":"journal-article","volume":"20","author":"Shi","year":"2024","journal-title":"J. Chem. Theory Comput."},{"key":"mlstae3c57bib10","doi-asserted-by":"publisher","first-page":"3502","DOI":"10.1021\/acs.accounts.9b00506","type":"journal-article","volume":"52","author":"Sauer","year":"2019","journal-title":"Acc. Chem. Res."},{"key":"mlstae3c57bib11","doi-asserted-by":"publisher","first-page":"628","DOI":"10.1039\/D4FD00041B","type":"journal-article","volume":"254","author":"Ye","year":"2024","journal-title":"Faraday Discuss"},{"key":"mlstae3c57bib12","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevLett.133.046401","type":"journal-article","volume":"133","author":"Della Pia","year":"2024","journal-title":"Phys. Rev. Lett."},{"key":"mlstae3c57bib13","doi-asserted-by":"publisher","first-page":"1688","DOI":"10.1038\/s41557-025-01884-y","type":"journal-article","volume":"17","author":"Shi","year":"2025","journal-title":"Nat. Chem."},{"key":"mlstae3c57bib14","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevLett.129.226001","type":"journal-article","volume":"129","author":"Daru","year":"2022","journal-title":"Phys. Rev. Lett."},{"key":"mlstae3c57bib15","doi-asserted-by":"publisher","first-page":"5372","DOI":"10.1021\/acs.jctc.5c00367","type":"journal-article","volume":"21","author":"M\u00e9sz\u00e1ros","year":"2025","journal-title":"J. Chem. Theory Comput."},{"key":"mlstae3c57bib16","doi-asserted-by":"publisher","first-page":"11710","DOI":"10.1021\/acs.jctc.5c01377","type":"journal-article","volume":"21","author":"O\u2019Neill","year":"2025","journal-title":"J. Chem. Theory Comput."},{"key":"mlstae3c57bib17","doi-asserted-by":"publisher","DOI":"10.1063\/1.4930194","type":"journal-article","volume":"143","author":"Medders","year":"2015","journal-title":"J. Chem. Phys."},{"key":"mlstae3c57bib18","doi-asserted-by":"publisher","DOI":"10.1063\/1.4967719","type":"journal-article","volume":"145","author":"Reddy","year":"2016","journal-title":"J. Chem. Phys."},{"key":"mlstae3c57bib19","doi-asserted-by":"publisher","first-page":"3446","DOI":"10.1021\/acs.jctc.3c00334","type":"journal-article","volume":"19","author":"Qu","year":"2023","journal-title":"J. Chem. Theory Comput."},{"key":"mlstae3c57bib20","doi-asserted-by":"publisher","first-page":"26","DOI":"10.1021\/ct9004917","type":"journal-article","volume":"6","author":"Xie","year":"2010","journal-title":"J. Chem. Theory Comput."},{"key":"mlstae3c57bib21","doi-asserted-by":"publisher","first-page":"766","DOI":"10.1038\/s41467-021-20914-0","type":"journal-article","volume":"12","author":"Lan","year":"2021","journal-title":"Nat. Commun."},{"key":"mlstae3c57bib22","doi-asserted-by":"publisher","first-page":"1032","DOI":"10.1002\/qua.24890","type":"journal-article","volume":"115","author":"Behler","year":"2015","journal-title":"Int. J. Quantum Chem."},{"key":"mlstae3c57bib23","doi-asserted-by":"publisher","first-page":"2654","DOI":"10.1021\/ct4002202","type":"journal-article","volume":"9","author":"Del Ben","year":"2013","journal-title":"J. Chem. Theory Comput."},{"key":"mlstae3c57bib24","doi-asserted-by":"publisher","first-page":"4510","DOI":"10.1021\/acs.jctc.2c01203","type":"journal-article","volume":"19","author":"Chen","year":"2023","edition":"ed","journal-title":"J. Chem. Theory Comput."},{"key":"mlstae3c57bib25","volume":"vol 8","author":"Thrun","year":"1995","type":"conference-proceedings"},{"key":"mlstae3c57bib26","doi-asserted-by":"publisher","first-page":"1910","DOI":"10.1063\/1.443164","type":"journal-article","volume":"76","author":"Purvis III","year":"1982","journal-title":"J. Chem. Phys."},{"key":"mlstae3c57bib27","doi-asserted-by":"publisher","first-page":"479","DOI":"10.1016\/S0009-2614(89)87395-6","type":"journal-article","volume":"157","author":"Raghavachari","year":"1989","journal-title":"Chem. Phys. Lett."},{"key":"mlstae3c57bib28","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevLett.90.136401","type":"journal-article","volume":"90","author":"Zhang","year":"2003","journal-title":"Phys. Rev. Lett."},{"key":"mlstae3c57bib29","doi-asserted-by":"publisher","first-page":"2087","DOI":"10.1021\/acs.jctc.5b00099","type":"journal-article","volume":"11","author":"Ramakrishnan","year":"2015","journal-title":"J. Chem. Theory Comput."},{"key":"mlstae3c57bib30","doi-asserted-by":"publisher","DOI":"10.1063\/5.0038301","type":"journal-article","volume":"154","author":"Nandi","year":"2021","journal-title":"J. Chem. Phys."},{"key":"mlstae3c57bib31","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1021\/acs.jctc.2c01034","type":"journal-article","volume":"19","author":"Bowman","year":"2023","journal-title":"J. Chem. Theory Comput."},{"key":"mlstae3c57bib32","doi-asserted-by":"publisher","DOI":"10.1063\/1.4926879","type":"journal-article","volume":"143","author":"Pinski","year":"2015","journal-title":"J. Chem. Phys."},{"key":"mlstae3c57bib33","doi-asserted-by":"publisher","DOI":"10.1063\/1.4939030","type":"journal-article","volume":"144","author":"Riplinger","year":"2016","journal-title":"J. Chem. Phys."},{"key":"mlstae3c57bib34","doi-asserted-by":"publisher","DOI":"10.1063\/5.0021116","type":"journal-article","volume":"153","author":"Nigam","year":"2020","journal-title":"J. Chem. Phys."},{"key":"mlstae3c57bib35","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevB.99.014104","type":"journal-article","volume":"99","author":"Drautz","year":"2019","journal-title":"Phys. Rev. B"},{"key":"mlstae3c57bib36","doi-asserted-by":"publisher","first-page":"2453","DOI":"10.1038\/s41467-022-29939-5","type":"journal-article","volume":"13","author":"Batzner","year":"2022","journal-title":"Nat. Commun."},{"key":"mlstae3c57bib37","author":"Batatia","year":"2022","type":"preprint"},{"key":"mlstae3c57bib38","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevX.14.021036","type":"journal-article","volume":"14","author":"Bochkarev","year":"2024","journal-title":"Phys. Rev. X"},{"key":"mlstae3c57bib39","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevLett.131.028001","type":"journal-article","volume":"131","author":"Darby","year":"2023","journal-title":"Phys. Rev. Lett."},{"key":"mlstae3c57bib40","doi-asserted-by":"publisher","first-page":"120","DOI":"10.1039\/D4FD00107A","type":"journal-article","volume":"256","author":"Kaur","year":"2024","journal-title":"Faraday Discuss."},{"key":"mlstae3c57bib41","article-title":"Materials project trajectory (MPTRJ) dataset","author":"Deng","year":"2023","type":"other"},{"key":"mlstae3c57bib42","doi-asserted-by":"publisher","DOI":"10.1002\/anie.202413637","type":"journal-article","volume":"64","author":"Bocus","year":"2025","journal-title":"Angew. Chem. Int. Edn."},{"key":"mlstae3c57bib43","doi-asserted-by":"crossref","DOI":"10.1038\/s41524-025-01779-z","type":"preprint","article-title":"Fast and Fourier features for transfer learning of interatomic potentials","author":"Novelli","year":"2025"},{"key":"mlstae3c57bib44","doi-asserted-by":"crossref","DOI":"10.1038\/s41524-025-01727-x","type":"preprint","article-title":"Fine-tuning foundation models of materials interatomic potentials with frozen transfer learning","author":"Radova","year":"2025"},{"key":"mlstae3c57bib45","doi-asserted-by":"crossref","DOI":"10.1038\/s41467-025-65662-7","type":"preprint","article-title":"PET-MAD, a lightweight universal interatomic potential for advanced materials modeling","author":"Mazitov","year":"2025"},{"key":"mlstae3c57bib46","doi-asserted-by":"crossref","DOI":"10.1088\/2632-2153\/ae040b","type":"preprint","article-title":"Multi-fidelity learning for interatomic potentials: low-level forces and high-level energies are all you need","author":"Messerly","year":"2025"},{"key":"mlstae3c57bib47","doi-asserted-by":"publisher","DOI":"10.1088\/2632-2153\/adc222","type":"journal-article","volume":"6","author":"Cui","year":"2025","journal-title":"Mach. Learn.: Sci. Technol."},{"key":"mlstae3c57bib48","doi-asserted-by":"publisher","DOI":"10.1039\/D5SC01325A","type":"journal-article","volume":"16","author":"Pia","year":"2025","journal-title":"Chem. Sci."},{"key":"mlstae3c57bib49","author":"Tuckerman","year":"2010","type":"book"},{"key":"mlstae3c57bib50","volume":"vol 35","author":"Batatia","year":"2022","type":"conference-proceedings"},{"key":"mlstae3c57bib51","doi-asserted-by":"publisher","DOI":"10.1063\/5.0215869","type":"journal-article","volume":"161","author":"Litman","year":"2024","journal-title":"J. Chem. Phys."},{"key":"mlstae3c57bib52","doi-asserted-by":"publisher","DOI":"10.1088\/1361-648X\/aa680e","type":"journal-article","volume":"29","author":"Hjorth Larsen","year":"2017","journal-title":"J. Phys.: Condens. Matter"},{"key":"mlstae3c57bib53","doi-asserted-by":"publisher","DOI":"10.1021\/acs.jpclett.4c02925","type":"journal-article","volume":"15","author":"Stolte","year":"2024","journal-title":"J. Phys. Chem. Lett."},{"key":"mlstae3c57bib54","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.1308560110","type":"journal-article","volume":"110","author":"Ceriotti","year":"2013","journal-title":"Proc. Natl Acad. Sci."},{"key":"mlstae3c57bib55","doi-asserted-by":"publisher","first-page":"2996","DOI":"10.1021\/acs.jpclett.5c00168","type":"journal-article","volume":"16","author":"Dasgupta","year":"2025","journal-title":"J. Phys. Chem. Lett."},{"key":"mlstae3c57bib56","doi-asserted-by":"publisher","first-page":"50","DOI":"10.1039\/D3FD00113J","type":"journal-article","volume":"249","author":"Kapil","year":"2023","journal-title":"Faraday Discuss."},{"key":"mlstae3c57bib57","article-title":"The dark side of the forces: assessing non-conservative force models for atomistic machine learning","author":"Bigi","year":"2025","type":"preprint"},{"key":"mlstae3c57bib58","article-title":"Resolving the body-order paradox of machine learning interatomic potentials","author":"Chong","year":"2025","type":"preprint"},{"key":"mlstae3c57bib59","doi-asserted-by":"publisher","DOI":"10.1002\/sstr.202500226","type":"journal-article","volume":"6","author":"Sosa","year":"2025","journal-title":"Small Struct."},{"key":"mlstae3c57bib60","doi-asserted-by":"publisher","DOI":"10.1063\/1.5086544","type":"journal-article","volume":"150","author":"Pinski","year":"2019","journal-title":"J. Chem. Phys."},{"key":"mlstae3c57bib61","author":"Zhang","year":"2024","type":"preprint"},{"key":"mlstae3c57bib62","doi-asserted-by":"publisher","first-page":"6020","DOI":"10.1021\/acs.jctc.4c00498","type":"journal-article","volume":"20","author":"Slootman","year":"2024","journal-title":"J. Chem. Theory Comput."},{"key":"mlstae3c57bib63","type":"dataset","article-title":"Data","author":"Gawkowski","year":"2025","unstructured":"Gawkowski M J 2025 Data (available at: https:\/\/github.com\/mikolgawk\/G-B-U )"}],"container-title":["Machine Learning: Science and Technology"],"original-title":[],"link":[{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae3c57","content-type":"text\/html","content-version":"am","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae3c57\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae3c57","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae3c57\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae3c57\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"syndication"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae3c57\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae3c57\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"similarity-checking"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae3c57\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,2,16]],"date-time":"2026-02-16T11:43:20Z","timestamp":1771242200000},"score":1,"resource":{"primary":{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae3c57"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,2,16]]},"references-count":63,"journal-issue":{"issue":"2","published-online":{"date-parts":[[2026,2,16]]},"published-print":{"date-parts":[[2026,4,1]]}},"URL":"https:\/\/doi.org\/10.1088\/2632-2153\/ae3c57","relation":{},"ISSN":["2632-2153"],"issn-type":[{"value":"2632-2153","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,2,16]]},"assertion":[{"value":"The good, the bad, and the ugly of atomistic learning for \u2018clusters-to-bulk\u2019 generalization","name":"article_title","label":"Article Title"},{"value":"Machine Learning: Science and Technology","name":"journal_title","label":"Journal Title"},{"value":"paper","name":"article_type","label":"Article Type"},{"value":"\u00a9 2026 The Author(s). Published by IOP Publishing Ltd","name":"copyright_information","label":"Copyright Information"},{"value":"2025-09-24","name":"date_received","label":"Date Received","group":{"name":"publication_dates","label":"Publication dates"}},{"value":"2026-01-22","name":"date_accepted","label":"Date Accepted","group":{"name":"publication_dates","label":"Publication dates"}},{"value":"2026-02-16","name":"date_epub","label":"Online publication date","group":{"name":"publication_dates","label":"Publication dates"}}]}}