{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2025,2,21]],"date-time":"2025-02-21T23:59:51Z","timestamp":1740182391678,"version":"3.37.3"},"reference-count":63,"publisher":"IOP Publishing","issue":"4","license":[{"start":{"date-parts":[[2023,11,10]],"date-time":"2023-11-10T00:00:00Z","timestamp":1699574400000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2023,11,10]],"date-time":"2023-11-10T00:00:00Z","timestamp":1699574400000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/iopscience.iop.org\/info\/page\/text-and-data-mining"}],"funder":[{"DOI":"10.13039\/501100000038","name":"Natural Sciences and Engineering Research Council of Canada","doi-asserted-by":"crossref","id":[{"id":"10.13039\/501100000038","id-type":"DOI","asserted-by":"crossref"}]}],"content-domain":{"domain":["iopscience.iop.org"],"crossmark-restriction":false},"short-container-title":["Mach. Learn.: Sci. Technol."],"published-print":{"date-parts":[[2023,12,1]]},"abstract":"<jats:title>Abstract<\/jats:title>\n               <jats:p>Kernel models of potential energy surfaces (PESs) for polyatomic molecules are often restricted by a specific choice of the kernel function. This can be avoided by optimizing the complexity of the kernel function. For regression problems with very expensive data, the functional form of the model kernels can be optimized in the Gaussian process (GP) setting through compositional function search guided by the Bayesian information criterion. However, the compositional kernel search is computationally demanding and relies on greedy strategies, which may yield sub-optimal kernels. An alternative strategy of increasing complexity of GP kernels treats a GP as a Bayesian neural network (NN) with a variable number of hidden layers, which yields NNGP models. Here, we present a direct comparison of GP models with composite kernels and NNGP models for applications aiming at the construction of global PES for polyatomic molecules. We show that NNGP models of PES can be trained much more efficiently and yield better generalization accuracy without relying on any specific form of the kernel function. We illustrate that NNGP models trained by distributions of energy points at low energies produce accurate predictions of PES at high energies. We also illustrate that NNGP models can extrapolate in the input variable space by building the free energy surface of the Heisenberg model trained in the paramagnetic phase and validated in the ferromagnetic phase. By construction, composite kernels yield more accurate models than kernels with a fixed functional form. Therefore, by illustrating that NNGP models outperform GP models with composite kernels, our work suggests that NNGP models should be a preferred choice of kernel models for PES.<\/jats:p>","DOI":"10.1088\/2632-2153\/ad0652","type":"journal-article","created":{"date-parts":[[2023,10,24]],"date-time":"2023-10-24T22:24:05Z","timestamp":1698186245000},"page":"045027","update-policy":"https:\/\/doi.org\/10.1088\/crossmark-policy","source":"Crossref","is-referenced-by-count":4,"title":["Neural network Gaussian processes as efficient models of potential energy surfaces for polyatomic molecules"],"prefix":"10.1088","volume":"4","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-2732-7316","authenticated-orcid":true,"given":"J","family":"Dai","sequence":"first","affiliation":[]},{"given":"R V","family":"Krems","sequence":"additional","affiliation":[]}],"member":"266","published-online":{"date-parts":[[2023,11,10]]},"reference":[{"key":"mlstad0652bib1","doi-asserted-by":"publisher","DOI":"10.1063\/1.2336223","article-title":"A random-sampling high dimensional model representation neural network for building potential energy surfaces","volume":"125","author":"Manzhos","year":"2006","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib2","doi-asserted-by":"publisher","first-page":"5295","DOI":"10.1021\/jp055253z","article-title":"A nested molecule-independent neural network approach for high-quality potential fits","volume":"110","author":"Manzhos","year":"2005","journal-title":"J. Phys. Chem. A"},{"key":"mlstad0652bib3","doi-asserted-by":"publisher","first-page":"3371","DOI":"10.1021\/jp9105585","article-title":"Potential energy surfaces fitted by artificial neural networks","volume":"114","author":"Handley","year":"2010","journal-title":"J. Phys. Chem. A"},{"key":"mlstad0652bib4","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevLett.98.146401","article-title":"Generalized neural-network representation of high-dimensional potential-energy surfaces","volume":"98","author":"Behler","year":"2007","journal-title":"Phys. Rev. Lett."},{"key":"mlstad0652bib5","doi-asserted-by":"publisher","DOI":"10.1039\/c1cp21668f","article-title":"Neural network potential-energy surfaces in chemistry: a tool for large-scale simulations","volume":"13","author":"Behler","year":"2011","journal-title":"Phys. Chem. Chem. Phys."},{"key":"mlstad0652bib6","doi-asserted-by":"publisher","first-page":"1032","DOI":"10.1002\/qua.24890","article-title":"Constructing high-dimensional neural network potentials: a tutorial review","volume":"115","author":"Behler","year":"2015","journal-title":"Int. J. Quantum Chem."},{"key":"mlstad0652bib7","doi-asserted-by":"publisher","first-page":"22272","DOI":"10.1039\/C7CP04010E","article-title":"A ground state potential energy surface for HONO based on a neural network with exponential fitting functions","volume":"19","author":"Pradhan","year":"2017","journal-title":"Phys. Chem. Chem. Phys."},{"key":"mlstad0652bib8","doi-asserted-by":"publisher","DOI":"10.1063\/1.4871981","article-title":"Calculating vibrational spectra with sum of product basis functions without storing full-dimensional vectors or matrices","volume":"140","author":"Leclerc","year":"2014","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib9","doi-asserted-by":"publisher","first-page":"1012","DOI":"10.1002\/qua.24795","article-title":"Neural network-based approaches for building high dimensional and quantum dynamics-friendly potential energy surfaces","volume":"115","author":"Manzhos","year":"2014","journal-title":"Int. J. Quantum Chem."},{"key":"mlstad0652bib10","doi-asserted-by":"publisher","DOI":"10.1063\/1.4801658","article-title":"A global potential energy surface for the H2+ OH H2O + H reaction using neural networks","volume":"138","author":"Chen","year":"2013","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib11","doi-asserted-by":"publisher","first-page":"1761","DOI":"10.1021\/acs.jpcc.7b12064","article-title":"Constructing high-dimensional neural network potential energy surfaces for gas\u2013surface scattering and reactions","volume":"122","author":"Liu","year":"2018","journal-title":"J. Phys. Chem. C"},{"key":"mlstad0652bib12","doi-asserted-by":"publisher","DOI":"10.1063\/1.4966192","article-title":"Perspective: machine learning potentials for atomistic simulations","volume":"145","author":"Behler","year":"2016","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib13","doi-asserted-by":"publisher","DOI":"10.1038\/ncomms13890","article-title":"Quantum-chemical insights from deep tensor neural networks","volume":"8","author":"Sch\u00fctt","year":"2017","journal-title":"Nat. Commun."},{"key":"mlstad0652bib14","doi-asserted-by":"publisher","first-page":"3678","DOI":"10.1021\/acs.jctc.9b00181","article-title":"PhysNet: a neural network for predicting energies, forces, dipole moments and partial charges","volume":"15","author":"Unke","year":"2019","journal-title":"J. Chem. Theory Comput."},{"key":"mlstad0652bib15","doi-asserted-by":"publisher","first-page":"537","DOI":"10.1146\/annurev.physchem.50.1.537","article-title":"Constructing multidimensional molecular potential energy surfaces from ab initio data","volume":"50","author":"Hollebeek","year":"1999","journal-title":"Annu. Rev. Phys. Chem."},{"key":"mlstad0652bib16","doi-asserted-by":"publisher","first-page":"13392","DOI":"10.1039\/C9CP01883B","article-title":"Bayesian machine learning for quantum molecular dynamics","volume":"21","author":"Krems","year":"2019","journal-title":"Phys. Chem. Chem. Phys."},{"key":"mlstad0652bib17","doi-asserted-by":"publisher","first-page":"2584","DOI":"10.1063\/1.470984","article-title":"A general method for constructing multidimensional molecular potential energy surfaces from ab initio calculations","volume":"104","author":"Ho","year":"1996","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib18","doi-asserted-by":"publisher","first-page":"7223","DOI":"10.1063\/1.473683","article-title":"A fast algorithm for evaluating multidimensional potential energy surfaces","volume":"106","author":"Hollebeek","year":"1997","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib19","doi-asserted-by":"publisher","first-page":"6433","DOI":"10.1063\/1.1603219","article-title":"Reproducing kernel Hilbert space interpolation methods as a paradigm of high dimensional model representations: application to multidimensional potential energy surface construction","volume":"119","author":"Ho","year":"2003","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib20","doi-asserted-by":"publisher","first-page":"1923","DOI":"10.1021\/acs.jcim.7b00090","article-title":"Toolkit for the construction of reproducing Kernel-based representations of data: application to multidimensional potential energy surfaces","volume":"57","author":"Unke","year":"2017","journal-title":"J. Chem. Inf. Model."},{"key":"mlstad0652bib21","doi-asserted-by":"publisher","first-page":"1386","DOI":"10.1021\/acs.jctc.9b00700","article-title":"Interpolation and extrapolation of global potential energy surfaces for polyatomic systems by Gaussian processes with composite kernels","volume":"16","author":"Dai","year":"2020","journal-title":"J. Chem. Theory Comput."},{"key":"mlstad0652bib22","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevA.107.032822","article-title":"Four-body singlet potential-energy surface for reactions of calcium monofluoride","volume":"107","author":"Sardar","year":"2023","journal-title":"Phys. Rev. A"},{"key":"mlstad0652bib23","doi-asserted-by":"publisher","DOI":"10.1063\/1.4989536","article-title":"Structure-based sampling and self-correcting machine learning for accurate calculations of potential energy surfaces and vibrational levels","volume":"146","author":"Dral","year":"2017","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib24","doi-asserted-by":"publisher","first-page":"2552","DOI":"10.1021\/acs.jpca.7b01182","article-title":"Representing global reactive potential energy surfaces using Gaussian processes","volume":"121","author":"Kolb","year":"2017","journal-title":"J. Phys. Chem. A"},{"key":"mlstad0652bib25","doi-asserted-by":"publisher","DOI":"10.1063\/1.5003074","article-title":"Neural networks vs Gaussian process regression for representing potential energy surfaces: a comparative study of fit quality and vibrational spectrum accuracy","volume":"148","author":"Kamath","year":"2018","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib26","doi-asserted-by":"publisher","DOI":"10.1063\/1.5009347","article-title":"Gaussian process regression to accelerate geometry optimizations relying on numerical differentiation","volume":"148","author":"Schmitz","year":"2018","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib27","doi-asserted-by":"publisher","first-page":"823","DOI":"10.1080\/00268976.2017.1407460","article-title":"Construction of reactive potential energy surfaces with Gaussian process regression: active data selection","volume":"116","author":"Guan","year":"2017","journal-title":"Mol. Phys."},{"key":"mlstad0652bib28","doi-asserted-by":"publisher","first-page":"237","DOI":"10.1039\/C8FD00085A","article-title":"Ab initio instanton rate theory made efficient using Gaussian process regression","volume":"212","author":"Laude","year":"2018","journal-title":"Faraday Discuss."},{"key":"mlstad0652bib29","doi-asserted-by":"publisher","first-page":"3140","DOI":"10.1021\/acs.jpca.8b00859","article-title":"Application of clustering algorithms to partitioning configuration space in fitting reactive potential energy surfaces","volume":"122","author":"Guan","year":"2018","journal-title":"J. Phys. Chem. A"},{"key":"mlstad0652bib30","doi-asserted-by":"publisher","DOI":"10.1016\/j.cpletx.2019.100022","article-title":"Multi-fidelity Gaussian process modeling for chemical energy surfaces","volume":"737","author":"Wiens","year":"2019","journal-title":"Chem. Phys. Lett."},{"key":"mlstad0652bib31","doi-asserted-by":"publisher","first-page":"3381","DOI":"10.1021\/acs.jctc.8b00298","article-title":"Assessing Gaussian process regression and permutationally invariant polynomial approaches to represent high-dimensional potential energy surfaces","volume":"14","author":"Qu","year":"2018","journal-title":"J. Chem. Theory Comput."},{"key":"mlstad0652bib32","doi-asserted-by":"publisher","DOI":"10.1063\/1.5143544","article-title":"Revisiting the Gaussian process regression for fitting high-dimensional potential energy surface and its application to the OH+HO2\u2192O2+H2O reaction","volume":"152","author":"Song","year":"2020","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib33","doi-asserted-by":"publisher","first-page":"2938","DOI":"10.3390\/molecules28072938","article-title":"Globally accurate Gaussian process potential energy surface and quantum dynamics studies on the Li(2S)+Na2\u2192LiNa+Na reaction at low collision energies","volume":"28","author":"Yang","year":"2023","journal-title":"Molecules"},{"article-title":"Global descriptors of a water molecule for machine learning of potential energy surfaces","year":"2023","author":"Albertani","key":"mlstad0652bib34"},{"key":"mlstad0652bib35","doi-asserted-by":"publisher","DOI":"10.1063\/5.0063534","article-title":"Gaussian process models of potential energy surfaces with boundary optimization","volume":"155","author":"Broad","year":"2021","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib36","doi-asserted-by":"publisher","DOI":"10.1088\/2632-2153\/ac3845","article-title":"Gradient domain machine learning with composite kernels: improving the accuracy of PES and force fields for large molecules","volume":"3","author":"Asnaashari","year":"2021","journal-title":"Mach. Learn.: Sci. Technol."},{"key":"mlstad0652bib37","doi-asserted-by":"publisher","DOI":"10.1088\/0953-4075\/49\/22\/224001","article-title":"Efficient non-parametric fitting of potential energy surfaces for polyatomic molecules with Gaussian processes","volume":"49","author":"Cui","year":"2016","journal-title":"J. Phys. B: At. Mol. Opt. Phys."},{"key":"mlstad0652bib38","doi-asserted-by":"publisher","DOI":"10.1063\/5.0023492","article-title":"Gaussian process model of 51-dimensional potential energy surface for protonated imidazole dimer","volume":"153","author":"Sugisawa","year":"2020","journal-title":"J. Chem. Phys."},{"article-title":"Gaussian processes with spectral delta kernel for higher accurate potential energy surfaces for large molecules","year":"2021","author":"Vargas-Hern\u00e1ndez","key":"mlstad0652bib39"},{"key":"mlstad0652bib40","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevLett.115.073202","article-title":"Gaussian process model for collision dynamics of complex molecules","volume":"115","author":"Cui","year":"2015","journal-title":"Phys. Rev. Lett."},{"key":"mlstad0652bib41","doi-asserted-by":"publisher","DOI":"10.1126\/sciadv.1603015","article-title":"Machine learning of accurate energy-conserving molecular force fields","volume":"3","author":"Chmiela","year":"2017","journal-title":"Sci. Adv."},{"key":"mlstad0652bib42","doi-asserted-by":"publisher","first-page":"3887","DOI":"10.1038\/s41467-018-06169-2","article-title":"Towards exact molecular dynamics simulations with machine-learned force fields","volume":"9","author":"Chmiela","year":"2018","journal-title":"Nat. Commun."},{"key":"mlstad0652bib43","doi-asserted-by":"publisher","DOI":"10.1063\/5.0023005","article-title":"Molecular force fields with gradient-domain machine learning (GDML): comparison and synergies with classical force fields","volume":"153","author":"Sauceda","year":"2020","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib44","doi-asserted-by":"publisher","first-page":"38","DOI":"10.1016\/j.cpc.2019.02.007","article-title":"sGDML: constructing accurate and data efficient molecular force fields using machine learning","volume":"240","author":"Chmiela","year":"2019","journal-title":"Comput. Phys. Commun."},{"key":"mlstad0652bib45","doi-asserted-by":"crossref","first-page":"461","DOI":"10.1214\/aos\/1176344136","article-title":"Estimating the dimension of a model","volume":"6","author":"Schwarz","year":"1978","journal-title":"Ann. Stat."},{"key":"mlstad0652bib46","doi-asserted-by":"publisher","DOI":"10.1088\/1367-2630\/ab0099","article-title":"Bayesian optimization for the inverse scattering problem in quantum reaction dynamics","volume":"21","author":"Vargas-Hern\u00e1ndez","year":"2019","journal-title":"New J. Phys."},{"key":"mlstad0652bib47","doi-asserted-by":"publisher","first-page":"4053","DOI":"10.1021\/acs.jpca.0c01375","article-title":"Bayesian optimization for calibrating and selecting hybrid-density functional models","volume":"124","author":"Vargas-Hernandez","year":"2020","journal-title":"J. Phys. Chem. A"},{"key":"mlstad0652bib48","doi-asserted-by":"publisher","first-page":"29","DOI":"10.1007\/978-1-4612-0745-0","volume":"vol 118","author":"Neal","year":"2012"},{"article-title":"Deep neural networks as Gaussian processes","year":"2017","author":"Lee","key":"mlstad0652bib49"},{"article-title":"Gaussian process behaviour in wide deep neural networks","year":"2018","author":"Matthews","key":"mlstad0652bib50"},{"article-title":"Deep convolutional networks as shallow Gaussian processes","year":"2018","author":"Garriga-Alonso","key":"mlstad0652bib51"},{"article-title":"On the infinite width limit of neural networks with a standard parameterization","year":"2020","author":"Sohl-Dickstein","key":"mlstad0652bib52"},{"article-title":"Bayesian deep convolutional networks with many channels are Gaussian processes","year":"2018","author":"Novak","key":"mlstad0652bib53"},{"key":"mlstad0652bib54","first-page":"p III 1166\u2013III","article-title":"Structure discovery in nonparametric regression through compositional Kernel search","volume":"vol 28","author":"Duvenaud","year":"2013","edition":"ed"},{"article-title":"Additive Gaussian processes","year":"2011","author":"Duvenaud","key":"mlstad0652bib55"},{"key":"mlstad0652bib56","doi-asserted-by":"publisher","DOI":"10.1063\/5.0088821","article-title":"Quantum Gaussian process model of potential energy surface for a polyatomic molecule","volume":"156","author":"Dai","year":"2022","journal-title":"J. Chem. Phys."},{"article-title":"Neural tangents: fast and easy infinite neural networks in python","year":"2020","author":"Novak","key":"mlstad0652bib57"},{"key":"mlstad0652bib58","doi-asserted-by":"publisher","first-page":"2914","DOI":"10.1038\/s41467-021-23103-1","article-title":"Spectral bias and task-model alignment explain generalization in kernel regression and infinitely wide neural networks","volume":"12","author":"Canatar","year":"2021","journal-title":"Nat. Commun."},{"key":"mlstad0652bib59","doi-asserted-by":"publisher","first-page":"5284","DOI":"10.1021\/acs.jctc.6b00765","article-title":"Ab initio potential for H3O+\u2192H++H2O : a step to a many-body representation of the hydrated proton?","volume":"12","author":"Yu","year":"2016","journal-title":"J. Chem. Theory Comput."},{"key":"mlstad0652bib60","doi-asserted-by":"publisher","first-page":"33","DOI":"10.1088\/2058-7058\/12\/4\/23","article-title":"Quantum phase transitions","volume":"12","author":"Sachdev","year":"1999","journal-title":"Phys. World"},{"key":"mlstad0652bib61","doi-asserted-by":"publisher","first-page":"4415","DOI":"10.1063\/1.480988","article-title":"On the long-range and short-range behavior of potentials from reproducing kernel Hilbert space interpolation","volume":"112","author":"Sold\u00e1n","year":"2000","journal-title":"J. Chem. Phys."},{"key":"mlstad0652bib62","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevB.82.085116","article-title":"Momentum average approximation for models with boson-modulated hopping: role of closed loops in the dynamical generation of a finite quasiparticle mass","volume":"82","author":"Berciu","year":"2010","journal-title":"Phys. Rev. B"},{"key":"mlstad0652bib63","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevLett.121.255702","article-title":"Extrapolating quantum observables with machine learning: inferring multiple phase transitions from properties of a single phase","volume":"121","author":"Vargas-Hern\u00e1ndez","year":"2018","journal-title":"Phys. Rev. Lett."}],"container-title":["Machine Learning: Science and Technology"],"original-title":[],"link":[{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ad0652","content-type":"text\/html","content-version":"am","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ad0652\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ad0652","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ad0652\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ad0652\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"syndication"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ad0652\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ad0652\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"similarity-checking"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ad0652\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2023,11,10]],"date-time":"2023-11-10T07:34:40Z","timestamp":1699601680000},"score":1,"resource":{"primary":{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ad0652"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2023,11,10]]},"references-count":63,"journal-issue":{"issue":"4","published-online":{"date-parts":[[2023,11,10]]},"published-print":{"date-parts":[[2023,12,1]]}},"URL":"https:\/\/doi.org\/10.1088\/2632-2153\/ad0652","relation":{},"ISSN":["2632-2153"],"issn-type":[{"type":"electronic","value":"2632-2153"}],"subject":[],"published":{"date-parts":[[2023,11,10]]},"assertion":[{"value":"Neural network Gaussian processes as efficient models of potential energy surfaces for polyatomic molecules","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 2023 The Author(s). Published by IOP Publishing Ltd","name":"copyright_information","label":"Copyright Information"},{"value":"2023-04-21","name":"date_received","label":"Date Received","group":{"name":"publication_dates","label":"Publication dates"}},{"value":"2023-10-24","name":"date_accepted","label":"Date Accepted","group":{"name":"publication_dates","label":"Publication dates"}},{"value":"2023-11-10","name":"date_epub","label":"Online publication date","group":{"name":"publication_dates","label":"Publication dates"}}]}}