[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-126626-en":3,"doc-seo-126626-105":31,"detail-sidebar-cat-0-en-105":92},{"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},126626,549768064622,"Anda","https://ap-avatar.wpscdn.com/davatar_6f874abed73319feea01a86fa6f0fab8",8,"Research & Report","Machine Learning Force Fields for Molecular Liquids - Ethylene Carbonate / Ethyl Methyl Carbonate - Binary Solvent - Supplementary Information","Supplementary information details the model-locality validation and parameter selection for ML force fields applied to the EC:EMC binary solvent (ethylene carbonate / ethyl methyl carbonate). A locality test defines Rcut by quantifying when DFT force contributions saturate, guiding the force regularization choice σF. Training data generation uses OPLS sampling with NPT or NVT ensembles depending on density, followed by DFT recomputation. An iterative GAP-MD training loop adds configurations targeting density fluctuations and MSD changes, including non-equilibrium volume-expansion scans using LAMMPS.","Supplementary Information for Machine Learning Force Fields for Molecular Liquids: Ethylene Carbonate / Ethyl Methyl Carbonate  \nBinary Solvent  \nIoan-Bogdan Magd˘au ∗1, Daniel J. Arismendi-Arrieta2 , Holly E. Smith3 , Clare P. Grey3 ,  \nKersti Hermansson2 , and G´abor Cs´anyi 1  \n1 Engineering Laboratory, University of Cambridge, Trumpington Street, Cambridge, CB2 1PZ, United Kingdom  \n2 Department of Chemistry–˚Angstr¨om Laboratory, Uppsala University, Box 538, 75121 Uppsala, Sweden  \n3Yusuf Hamid Department of Chemistry, University of Cambridge, Lensfield Road,  \nCambridge, CB2 1EW, United Kingdom  \n1 Locality Test  \nRcut is an important parameter in our ML models and defines the extent of the atomic neighborhood used for constructing the geometric descriptors. This parameter determines the length-scale of the interactions that can be captured by the short-range model. Contributions from atoms outside this cut-off are not a function of the local environment and appear as noise in the model. This force noise determine the maximum accuracy the model can achieve and prescribes a suitable value for the regularization parameter σF [1] .  \nSupplementary Figure 1: Locality test. Panel A illustrates the locality test: the atoms inside the yellow sphere are kept fixed while the others are simulated with OPLS at 2000 K. The distribution of DFT forces experienced by the central atom at different cut-off radii of the fixed sphere is shown with a violin plot in Panel B. Panel C reports the standard deviation of the force distribution at each radius.  \nAppropriate values for Rcut and σF can be found by performing the locality test, illustrated in Supplementary Figure 1, and introduced in earlier work [2, 3] . We select a molecular configuration from the liquid state and choose a central atom, fixing the positions of all of its neighbours within a sphere of radius Rcut. All other  \n∗ [i.b.magdau@gmail.com](i.b.magdau@gmail.com)  \natoms are allowed to move freely at 2000 K. The DFT properties are computed for ten different configurations sampled with OPLS dynamics [4, 5] . We repeat the same exercise for five values of Rcut and monitor the DFT force distribution on the central atom as shown in Supplementary Figure 1B. The standard deviation of the central force decreases quickly with Rcut and saturates at around 5 ˚A to a value smaller than 0 . 1 eV ˚A−1 which implies the force is largely determined by the environment inside this cutoff. From an abundance of caution we chose a slightly larger cutoff of 6 ˚A for all our ML models and we set the force regularization parameter to σF = 0 . 1 eV ˚A−1 .  \n2 OPLS Training Set  \nSupplementary Figure 2: Sampling the original OPLS training set. Panel A shows the density and temperature of the OPLS-sampled 12-molecule configurations used to construct the fixed training set. Panel B shows the MSD of the trajectories employed for sampling.  \nWe used OPLS to sample our first data set at the target composition EC:EMC (3:7 M) . As shown in Supplementary Figure 2, for densities above the equilibrium value of ≈ 1 g cm −3, we used the NPT ensemble which allows the volume to fluctuate. At lower densities, even small negative external pressures result in liquid vaporization, so we use the NVT ensemble instead, which maintains the systems in the liquid state. We constructed training sets of different sizes: ranging from 200 configurations by sampling every 50 ps to 1000 configurations by sampling every 10 ps, and recomputed all properties with DFT. We could not achieve MD-stable GAP models by training only on the OPLS training sets and iterative training was required.  \n3 Iterative Training  \nSupplementary Figure 3 illustrates the iterative training protocol. At each generation we use the working model to run a series of small-size GAP-MD simulation (12 molecules up to Gen13 and 17 to 25 molecules thereafter, see Table 2 from the main text) . These simulations span between 20 to 100 ps and are initialized with ","cbCaio28zbKK8Xmg","https://ap.wps.com/l/cbCaio28zbKK8Xmg","pdf",19748210,2,1,19,"English","en",105,"# Locality Test\n## OPLS Training Set\n## Iterative Training","[{\"question\":\"What does the locality test establish for the ML force-field models?\",\"answer\":\"It determines the cut-off radius Rcut by measuring how DFT force contributions change as the atomic neighborhood grows. Once the force distribution standard deviation saturates, atoms outside the cut-off add noise rather than meaningful signal.\"},{\"question\":\"How are training sets generated from OPLS, and when are NPT vs NVT used?\",\"answer\":\"OPLS is used to sample molecular configurations at the target EC:EMC composition, with NPT for densities above the equilibrium density to allow volume fluctuation. For lower densities, NVT is used to prevent liquid vaporization triggered by negative external pressures.\"},{\"question\":\"What is the iterative training strategy in the protocol?\",\"answer\":\"The working model runs short GAP-MD simulations across generations, then new atomic configurations are selected—especially near large density fluctuations or MSD changes—to augment the next training set. Additional physics-inspired configurations are added at later generations to address remaining model shortcomings.\"}]","Machine Learning Force Fields for Molecular Liquids - Ethylene Carbonate / Ethyl Methyl Carbonate - Binary Solvent - Supplementary Information | PDF",1785933875,48,{"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":87,"head_meta":89,"extra_data":91,"updated_unix":29},"machine-learning-force-fields-for-molecular-liquids-ethylene-carbonate-ethyl-methyl-carbonate-binary-solvent-supplementary-information","",{"@graph":37,"@context":86},[38,54,69],{"@type":39,"itemListElement":40},"BreadcrumbList",[41,45,48,51],{"item":42,"name":43,"@type":44,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":46,"name":47,"@type":44,"position":20},"https://docshare.wps.com/document/","Document",{"item":49,"name":12,"@type":44,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":44,"position":53},"https://docshare.wps.com/document/machine-learning-force-fields-for-molecular-liquids-ethylene-carbonate-ethyl-methyl-carbonate-binary-solvent-supplementary-information/126626/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":24,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":42,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-28","2026-08-05",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"What does the locality test establish for the ML force-field models?","Question",{"text":76,"@type":77},"It determines the cut-off radius Rcut by measuring how DFT force contributions change as the atomic neighborhood grows. Once the force distribution standard deviation saturates, atoms outside the cut-off add noise rather than meaningful signal.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How are training sets generated from OPLS, and when are NPT vs NVT used?",{"text":81,"@type":77},"OPLS is used to sample molecular configurations at the target EC:EMC composition, with NPT for densities above the equilibrium density to allow volume fluctuation. For lower densities, NVT is used to prevent liquid vaporization triggered by negative external pressures.",{"name":83,"@type":74,"acceptedAnswer":84},"What is the iterative training strategy in the protocol?",{"text":85,"@type":77},"The working model runs short GAP-MD simulations across generations, then new atomic configurations are selected—especially near large density fluctuations or MSD changes—to augment the next training set. Additional physics-inspired configurations are added at later generations to address remaining model shortcomings.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":93},[94,98,102,106,111,116,121,124,129,132,136],{"id":21,"doc_module":4,"doc_module_name":47,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":47,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":47,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":107,"doc_module":4,"doc_module_name":47,"category_name":108,"show_sort_weight":109,"slug":110},5,"Comic",60,"comic",{"id":112,"doc_module":4,"doc_module_name":47,"category_name":113,"show_sort_weight":114,"slug":115},6,"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":22,"doc_module":4,"doc_module_name":47,"category_name":137,"show_sort_weight":107,"slug":138},"General","general"]