[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82357-en":3,"doc-seo-82357-105":31,"detail-sidebar-cat-0-en-105":93},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":28,"seo_description":14,"update_tm":29,"read_time":30},82357,687197207919,"Theodora","https://ap-avatar.wpscdn.com/avatar/a000253d6f5f7c60be?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779446848396160552",8,"Research & Report","Active rejection enables reliable generalization of universal machine-learning interatomic potentials","Universal machine learning interatomic potentials (uMLIPs) improve atomistic simulations by combining quantum accuracy with large-scale molecular dynamics, yet high-fidelity training is limited by the prohibitive cost of methods such as r2 SCAN. This scarcity creates localized reliability gaps where benchmark success does not guarantee dependable energy–force predictions for every structure. Adaptive Multi-Teacher Routing (ATR) calibrates multiple pretrained uMLIP teachers, uses inter-teacher disagreement to estimate structure–teacher reliability, and actively rejects uncertain cases. With only 0.2% r2 SCAN labels, ATR distills 2.89 million traceable pseudo-labels and trains CHGNet that outperforms baselines, improving dynamical robustness under finite-temperature molecular dynamics validations.","arXiv :2607 .09456v 1 [ cs .LG] 10 Jul 2026  \nActive rejection enables reliable generalization of universal machine-learning interatomic potentials  \nMingxiang Luo 1,2† Xinnan Mao4† Lu Wang4  \nLei Bai3 Feng Ding4* Yuqiang Li3,2*  \n1 School of Future Information Innovation, Fudan University, Shanghai, China  \n2 Shanghai Innovation Institute, Shanghai, China  \n3 Shanghai Artificial Intelligence Laboratory, Shanghai, China  \n4 Suzhou Laboratory, Suzhou, China  \nAbstract  \nUniversal machine learning interatomic potentials (uMLIPs) have revolutionized atomistic simulations by bridging quantum-mechanical accuracy with large-scale molecular dynamics. The prohibitive cost of high-accuracy calculations, such as r2 SCAN, limits high-fidelity uMLIP training to datasets that are much smaller and less diverse than the open materials space. Models trained under this constraint can exhibit localized reliability gaps, where strong benchmark performance does not ensure reliable energy–force predictions for every structure. A practical route is to mine massive medium-to low-fidelity structure repositories with multiple pretrained uMLIPs while filtering unreliable pseudolabels. Here, we propose the Adaptive Multi-Teacher Routing (ATR) framework, which reformulates high-fidelity data construction as a structure-wise decision problem under uncertainty. Using a small set of real r2 SCAN labels, ATR calibrates multiple pretrained uMLIP teachers and combines structural descriptors, teacher identity and inter-teacher disagreement to estimate the reliability of each structure–teacher pair. It selects high-confidence teacher predictions for pseudo-label generation and rejects structures for which no teacher is sufficiently reliable. With real r2 SCAN labels for only 0.2% of the candidate structures, ATR distils 2 . 89 million traceable r2 SCAN-level pseudo-labels for model pretraining. Experiments on held-out r2 SCAN structures and the MP-r2 SCAN benchmark show that a lightweight CHGNet trained on the ATR-generated dataset consistently outperforms the baseline and non-routed controls. Finite-temperature molecular dynamics validations further show that ATR pretraining improves dynamical robustness across multiple material systems, maintaining stable trajectories in cases where baseline simulations undergo catastrophic structural collapse. These results establish active rejection as an effective mechanism for converting multiple pretrained uMLIPs into a scalable and reliable data-construction system, improving the generalization of high-fidelity uMLIPs across the open materials space.  \n1 Introduction  \nMachine learning interatomic potentials (MLIPs) [1] are substantially expanding the accessible time scales, length scales and application scope of atomistic materials simulations [2, 3, 4] . Large-scale electronicstructure databases and increasingly expressive graph neural networks [5, 6, 7, 8] have enabled universal MLIPs (uMLIPs) [9, 10, 11] that support structure optimization, molecular dynamics, phase-stability assessment and property prediction across diverse materials spaces. In this sense, uMLIPs are becoming atomistic scientific foundation models [12, 13]: they encode transferable priors over potential-energy surfaces and can be reused across many systems with limited task-specific retraining. During deployment across the open materials space, uMLIPs can encounter structure-dependent reliability gaps that averaged benchmark metrics often fail to reveal. Atomistic simulation operates in a continuous physical space where elemental combinations, local coordination environments and thermodynamic states vary widely. A model trained on a relatively limited high-fidelity dataset can perform well on a benchmark yet fail on rare or out-of-distribution local environments. For interatomic potentials, such local failures are especially  \nconsequential because unreliable force predictions can destabilize molecular dynamics trajectories and lead to temperature d","cbCaijPFAReS0zwS","https://ap.wps.com/l/cbCaijPFAReS0zwS","pdf",5146723,6,1,23,"English","en",105,"# Abstract\n# Introduction","[{\"question\":\"What problem does active rejection address in universal machine-learning interatomic potentials?\",\"answer\":\"Active rejection addresses structure-dependent reliability gaps where strong benchmark metrics do not guarantee accurate energy–force predictions for every structure during atomistic simulations.\"},{\"question\":\"How does the ATR framework decide which pseudo-labels to trust?\",\"answer\":\"ATR calibrates multiple pretrained uMLIP teachers using a small set of real r2 SCAN labels, then combines structural descriptors, teacher identity, and inter-teacher disagreement to estimate reliability for each structure–teacher pair.\"},{\"question\":\"What is the impact of ATR pretraining on molecular dynamics performance?\",\"answer\":\"ATR pretraining with a CHGNet trained on ATR-generated pseudo-labels improves dynamical robustness, maintaining stable trajectories in cases where baseline simulations can undergo catastrophic structural collapse.\"}]","Active rejection enables reliable generalization of universal machine-learning interatomic potentials | PDF",1784179868,58,{"code":4,"msg":32,"data":33},"ok",{"site_id":25,"language":24,"slug":34,"title":13,"keywords":35,"description":14,"schema_data":36,"social_meta":88,"head_meta":90,"extra_data":92,"updated_unix":29},"active-rejection-enables-reliable-generalization-of-universal-machine-learning-interatomic-potentials","",{"@graph":37,"@context":87},[38,55,70],{"@type":39,"itemListElement":40},"BreadcrumbList",[41,45,49,52],{"item":42,"name":43,"@type":44,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":46,"name":47,"@type":44,"position":48},"https://docshare.wps.com/document/","Document",2,{"item":50,"name":12,"@type":44,"position":51},"https://docshare.wps.com/document/research-report/",3,{"item":53,"name":13,"@type":44,"position":54},"https://docshare.wps.com/document/active-rejection-enables-reliable-generalization-of-universal-machine-learning-interatomic-potentials/82357/",4,{"url":53,"name":13,"@type":56,"author":57,"headline":13,"publisher":59,"fileFormat":62,"inLanguage":24,"description":14,"dateModified":63,"datePublished":64,"encodingFormat":62,"isAccessibleForFree":65,"interactionStatistic":66},"DigitalDocument",{"name":9,"@type":58},"Person",{"url":42,"name":60,"@type":61},"DocShare","Organization","application/pdf","2026-07-29","2026-07-16",true,{"@type":67,"interactionType":68,"userInteractionCount":20},"InteractionCounter",{"@type":69},"ViewAction",{"@type":71,"mainEntity":72},"FAQPage",[73,79,83],{"name":74,"@type":75,"acceptedAnswer":76},"What problem does active rejection address in universal machine-learning interatomic potentials?","Question",{"text":77,"@type":78},"Active rejection addresses structure-dependent reliability gaps where strong benchmark metrics do not guarantee accurate energy–force predictions for every structure during atomistic simulations.","Answer",{"name":80,"@type":75,"acceptedAnswer":81},"How does the ATR framework decide which pseudo-labels to trust?",{"text":82,"@type":78},"ATR calibrates multiple pretrained uMLIP teachers using a small set of real r2 SCAN labels, then combines structural descriptors, teacher identity, and inter-teacher disagreement to estimate reliability for each structure–teacher pair.",{"name":84,"@type":75,"acceptedAnswer":85},"What is the impact of ATR pretraining on molecular dynamics performance?",{"text":86,"@type":78},"ATR pretraining with a CHGNet trained on ATR-generated pseudo-labels improves dynamical robustness, maintaining stable trajectories in cases where baseline simulations can undergo catastrophic structural collapse.","https://schema.org",{"og:url":53,"og:type":89,"og:title":13,"og:site_name":60,"og:description":14},"article",{"robots":91,"canonical":53},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":94},[95,99,103,107,112,116,121,124,129,132,136],{"id":21,"doc_module":4,"doc_module_name":47,"category_name":96,"show_sort_weight":97,"slug":98},"Story & Novel",90,"story-novel",{"id":48,"doc_module":4,"doc_module_name":47,"category_name":100,"show_sort_weight":101,"slug":102},"Literature",80,"literature",{"id":54,"doc_module":4,"doc_module_name":47,"category_name":104,"show_sort_weight":105,"slug":106},"Exam",70,"exam",{"id":108,"doc_module":4,"doc_module_name":47,"category_name":109,"show_sort_weight":110,"slug":111},5,"Comic",60,"comic",{"id":20,"doc_module":4,"doc_module_name":47,"category_name":113,"show_sort_weight":114,"slug":115},"Technology",50,"technology",{"id":117,"doc_module":4,"doc_module_name":47,"category_name":118,"show_sort_weight":119,"slug":120},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":47,"category_name":12,"show_sort_weight":122,"slug":123},30,"research-report",{"id":125,"doc_module":4,"doc_module_name":47,"category_name":126,"show_sort_weight":127,"slug":128},9,"Religion & Spirituality",20,"religion-spirituality",{"id":127,"doc_module":4,"doc_module_name":47,"category_name":130,"show_sort_weight":127,"slug":131},"World Cup","world-cup",{"id":133,"doc_module":4,"doc_module_name":47,"category_name":134,"show_sort_weight":133,"slug":135},10,"Lifestyle","lifestyle",{"id":137,"doc_module":4,"doc_module_name":47,"category_name":138,"show_sort_weight":108,"slug":139},19,"General","general"]