[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-126855-en":3,"doc-seo-126855-105":30,"detail-sidebar-cat-0-en-105":91},{"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":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":27,"seo_description":14,"update_tm":28,"read_time":29},126855,1099523882367,"Hazel","https://ap-avatar.wpscdn.com/davatar_9964176cb1d06d4a9deccf72a44ae3dc",8,"Research & Report","Leveraging Domain Adaptation for Accurate Machine Learning Predictions of New Halide Perovskites","This study combines graph neural networks (GNNs) with a low-cost bond-valence-method (BVM) structure generation workflow to build machine learning models for high-throughput screening of halide perovskites. Training uses statistical DFT/PBE formation energies (ΔHPBEf) and band gaps (EPBEg, EHSEg), with domain adaptation (DA) improving ΔHPBEf by 1.8× and EHSEg by 1.2–1.35× versus direct training. After active learning, the best model achieves MAE of 38 meV/atom for unseen materials and 0.18 eV for EHSEg. The models identify 48 stable, photovoltaic-relevant compounds, with only 8 previously reported, suggesting 40 candidates for experimental exploration.","arXiv :2401 . 10998v1 [ cond-mat .mtrl-sci ] 19 Jan 2024  \nLeveraging Domain Adaptation for Accurate Machine Learning Predictions of New  \nHalide Perovskites  \nDipannoy Das Gupta, 1, 2 Zachary J. L. Bare, 1 Suxuen Yew,3 Santosh Adhikari, 1  \nBrian DeCost,4 Qi Zhang,2 Charles Musgrave,3 and Christopher Sutton 1  \n1 Department of Chemistry and Biochemistry, University of South Carolina, Columbia, SC 29208, United States  \n2 Department of Computer Science and Engineering,  \nUniversity of South Carolina, Columbia, SC 29208, United States  \n3 Department of Chemical and Biological Engineering,  \nUniversity of Colorado, Boulder, CO 80303, United States  \n4 National Institute of Standards and Technology, Gaithersburg, MD 20899, United States  \nWe combine graph neural networks (GNN) with an inexpensive and reliable structure generation approach based on the bond-valence method (BVM) to train accurate machine learning models for screening 222,960 halide perovskites using statistical estimates of the DFT/PBE formation energy (∆HPBEf), and the PBE (EPBEg) and HSE (EHSEg) band gaps. The GNNs were fined tuned using domain adaptation (DA) from a source model, which yields a factor of 1.8 times improvement in ∆HPBEf and 1.2-1.35 times improvement in EHSEg compared to direct training (i.e., without DA) . As far as we are aware, this is the first demonstration of DA for GNNs in materials science. Moreover, after three active learning iterations – resulting in a total of 3593 training samples for ∆Hf – the best ML model produced mean absolute errors (MAE) of 38 meV/atom relative to ∆HPBEf for totally unseen materials. The ML model for predicting EHSEg achieved an MAE of 0.18 eV. Using these two ML models, 48 compounds were identified out of 222,960 candidates as both stable and that have an HSE band gap that is relevant for photovoltaic applications. For this subset, only 8 have been reported to date, indicating that 40 compounds remain unexplored to the best of our knowledge and therefore offer opportunities for potential experimental examination.  \nI. INTRODUCTION  \nHalide perovskites are a broadly used class of materials for a variety of applications, including light-emitting diodes (LEDs) and photocatalysis that are particularly promising as light-adsorbing materials for photovoltaics (PV).[1–4] One reason for the wide applicability of perovskites is their extensive chemical space,[5] which enables modulation of their optical, electronic, and thermodynamic properties by changing their compositions. Although the enormous chemical space opens up opportunities for fine-tuning the properties of potential materials, it simultaneously poses a significant challenge for effectively screening materials.  \nRecent improvements in computing power allow for the use of quantum mechanical calculations, e.g., density functional theory (DFT), for high-throughput screening of materials. Specifically for screening perovskites, Kar et al. and Schmidt et al. have examined ABX3 compounds.[6–8] However, AA’BB’X6 perovskites are less explored because their unit cells are at least twice as large as those of ABX3 perovskites, and thus, approximately 8 times more computationally expensive when assuming cubic scaling (N3 ) of DFT. Alternatively, machine learning (ML) can be used to accurately predict DFT properties of atomistic systems at orders of magnitude lower computational cost, thereby enabling the rapid evaluation of materials over a large chemical space for promising candidates.  \nThe accuracy of an ML model is significantly influenced by the selection of input features or representation. Because ML-based high-throughput screening tasks typ-  \nically cannot include features from the DFT-optimized structure, tabulated atomic properties (such as radii) have been utilized as the inputs in the ML models used to screen materials generally,[9–11] and specifically for halide perovskites.[7, 12–16] Some of us have previously reported a descriptor,[17] τ that allows for","cbCaikVJNwvggKyi","https://ap.wps.com/l/cbCaikVJNwvggKyi","pdf",3319568,1,11,"English","en",105,"# Introduction\n## Perovskites and the need for efficient screening\n## Limits of composition-based machine learning\n## Why structure-aware models matter\n## Graph representations and GNNs in materials science\n## Training set challenges for structure-based learning","[{\"question\":\"What method combines GNNs and inexpensive structure generation in this work?\",\"answer\":\"The approach couples graph neural networks with structure generation based on the bond-valence method (BVM).\"},{\"question\":\"How does domain adaptation improve model performance compared with direct training?\",\"answer\":\"Domain adaptation yields a 1.8× improvement for ΔHPBEf and a 1.2–1.35× improvement for EHSEg versus direct training without DA.\"},{\"question\":\"How are new halide perovskite candidates selected and what is the final outcome?\",\"answer\":\"Two ML models for ΔHPBEf and EHSEg are used to filter candidates, identifying 48 stable compounds relevant for photovoltaic applications, of which 40 are previously unexplored.\"}]","Leveraging Domain Adaptation for Accurate Machine Learning Predictions of New Halide Perovskites | PDF",1785935253,28,{"code":4,"msg":31,"data":32},"ok",{"site_id":24,"language":23,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"leveraging-domain-adaptation-for-accurate-machine-learning-predictions-of-new-halide-perovskites","",{"@graph":36,"@context":85},[37,54,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/leveraging-domain-adaptation-for-accurate-machine-learning-predictions-of-new-halide-perovskites/126855/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":23,"description":14,"dateModified":62,"datePublished":62,"encodingFormat":61,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-05",true,{"@type":65,"interactionType":66,"userInteractionCount":4},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What method combines GNNs and inexpensive structure generation in this work?","Question",{"text":75,"@type":76},"The approach couples graph neural networks with structure generation based on the bond-valence method (BVM).","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does domain adaptation improve model performance compared with direct training?",{"text":80,"@type":76},"Domain adaptation yields a 1.8× improvement for ΔHPBEf and a 1.2–1.35× improvement for EHSEg versus direct training without DA.",{"name":82,"@type":73,"acceptedAnswer":83},"How are new halide perovskite candidates selected and what is the final outcome?",{"text":84,"@type":76},"Two ML models for ΔHPBEf and EHSEg are used to filter candidates, identifying 48 stable compounds relevant for photovoltaic applications, of which 40 are previously unexplored.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]