Machine-learning interatomic potentials have revolutionized materials modeling at the atomic scale.Thanks to these,it is now indeed possible to perform simulations of ab initio quality over very large time and length ...Machine-learning interatomic potentials have revolutionized materials modeling at the atomic scale.Thanks to these,it is now indeed possible to perform simulations of ab initio quality over very large time and length scales.More recently,various universal machine-learning models have been proposed as an out-of-box approach avoiding the need to train and validate specific potentials for each particular material of interest.In this paper,we review and evaluate four different universal machine-learning interatomic potentials(uMLIPs),all based on graph neural network architectures which have demonstrated transferability from one chemical system to another.The evaluation procedure relies on data both from a recent verification study of density-functional-theory implementations and from the Materials Project.Through this comprehensive evaluation,we aim to provide guidance to materials scientists in selecting suitable models for their specific research problems,offer recommendations for model selection and optimization,and stimulate discussion on potential areas for improvement in current machinelearning methodologies in materials science.展开更多
Electronic and optical properties of materials are affected by atomic motion through the electron–phonon interaction:not only band gaps change with temperature,but even at absolute zero temperature,zero-point motion ...Electronic and optical properties of materials are affected by atomic motion through the electron–phonon interaction:not only band gaps change with temperature,but even at absolute zero temperature,zero-point motion causes band-gap renormalization.We present a large-scale first-principles evaluation of the zero-point renormalization of band edges beyond the adiabatic approximation.For materials with light elements,the band gap renormalization is often larger than 0.3 eV,and up to 0.7 eV.This effect cannot be ignored if accurate band gaps are sought.For infrared-active materials,global agreement with available experimental data is obtained only when non-adiabatic effects are taken into account.They even dominate zero-point renormalization for many materials,as shown by a generalized Fröhlich model that includes multiple phonon branches,anisotropic and degenerate electronic extrema,whose range of validity is established by comparison with first-principles results.展开更多
The electron–phonon interaction is central to condensed matter,e.g.through electrical resistance,superconductivity or the formation of polarons,and has a strong impact on observables such as band gaps or optical spec...The electron–phonon interaction is central to condensed matter,e.g.through electrical resistance,superconductivity or the formation of polarons,and has a strong impact on observables such as band gaps or optical spectra.The most common framework for band energy corrections is the Fröhlich model,which often agrees qualitatively with experiments in polar materials,but has limits for complex cases.A generalized version includes anisotropic and degenerate electron bands,and multiple phonons.In this work,we identify trends and outliers for the Fröhlich models on 1260 materials.We test the limits of the Fröhlich models and their perturbative treatment,in particular the large polaron hypothesis.Among our extended dataset most materials host perturbative large polarons,but there are many instances that are non-perturbative and/or localize on distances of a few bond lengths.We find a variety of behaviors,and analyze extreme cases with huge zero-point renormalization using the first-principles Allen-Heine-Cardona approach.展开更多
基金supported by the National Key Research and Development Program of China(2022YFE0141100 and 2023YFB3003005).
文摘Machine-learning interatomic potentials have revolutionized materials modeling at the atomic scale.Thanks to these,it is now indeed possible to perform simulations of ab initio quality over very large time and length scales.More recently,various universal machine-learning models have been proposed as an out-of-box approach avoiding the need to train and validate specific potentials for each particular material of interest.In this paper,we review and evaluate four different universal machine-learning interatomic potentials(uMLIPs),all based on graph neural network architectures which have demonstrated transferability from one chemical system to another.The evaluation procedure relies on data both from a recent verification study of density-functional-theory implementations and from the Materials Project.Through this comprehensive evaluation,we aim to provide guidance to materials scientists in selecting suitable models for their specific research problems,offer recommendations for model selection and optimization,and stimulate discussion on potential areas for improvement in current machinelearning methodologies in materials science.
基金This work has been supported by the Fonds de la Recherche Scientifique(FRS-FNRS Belgium)through the PdR Grant No.T.0238.13-AIXPHOthe PdR Grant No.T.0103.19-ALPS+7 种基金the Fonds de Recherche du Québec Nature et Technologie(FRQ-NT)the Natural Sciences and Engineering Research Council of Canada(NSERC)under grants RGPIN-2016-06666Computational resources have been provided by the supercomputing facilities of the Universitécatholique de Louvain(CISM/UCL)the Consortium des Equipements de Calcul Intensif en Fédération Wallonie Bruxelles(CECI)funded by the FRS-FNRS under Grant No.2.5020.11the Tier-1 supercomputer of the Fédération Wallonie-Bruxelles,infrastructure funded by the Walloon Region under the grant agreement No.1117545as well as the Canadian Foundation for Innovation,the Ministère de l’Éducation des Loisirs et du Sport(Québec),Calcul Québec,and Compute Canada.This work was supported by the Center for Computational Study of Excited-State Phenomena in Energy Materials(C2SEPEM)at the Lawrence Berkeley National Laboratory,which is funded by the U.S.Department of Energy,Office of Science,Basic Energy Sciences,Materials Sciences and Engineering Division under Contract No.DE-AC02-05CH11231as part of the Computational Materials Sciences Program(advanced algorithms/codes)and by the National Science Foundation under grant DMR-1926004(basic theory and formalism)This research used resources of the National Energy Research Scientific Computing Center(NERSC),a DOE Office of Science User Facility supported by the Office of Science of the U.S.Department of Energy under Contract No.DE-AC02-05CH11231.
文摘Electronic and optical properties of materials are affected by atomic motion through the electron–phonon interaction:not only band gaps change with temperature,but even at absolute zero temperature,zero-point motion causes band-gap renormalization.We present a large-scale first-principles evaluation of the zero-point renormalization of band edges beyond the adiabatic approximation.For materials with light elements,the band gap renormalization is often larger than 0.3 eV,and up to 0.7 eV.This effect cannot be ignored if accurate band gaps are sought.For infrared-active materials,global agreement with available experimental data is obtained only when non-adiabatic effects are taken into account.They even dominate zero-point renormalization for many materials,as shown by a generalized Fröhlich model that includes multiple phonon branches,anisotropic and degenerate electronic extrema,whose range of validity is established by comparison with first-principles results.
基金This work has been supported by the Fonds de la Recherche Scientifique(FRS-FNRS Belgium)through the PdR Grant No.T.0103.19-ALPSthe Federal government of Belgium through the EoS Project ID 40007525.ZZ+7 种基金PMMCM acknowledge financial support by the Netherlands Sector Plan program 2019-2023the research program“Materials for the Quantum Age”(QuMAt,registration number 024.005.006)part of the Gravitation program of the Dutch Ministry of Education,Culture and Science(OCW)This project has received funding from the European Union’s Horizon 2020 research and innovation program under grant agreement No.951786-NOMAD CoEComputational resources have been provided by the CISM/UCLouvain and the CECI funded by the FRS-FNRS Belgium under Grant No.2.502011,as well as the Tier-1 supercomputer of the Fédération Wallonie-Bruxelles,funded by the Walloon Region under grant agreement No.1117545We acknowledge a PRACE award granting access to MareNostrum4 at Barcelona Supercomputing Center(BSC),Spain(OptoSpin project id.2020225411)Moreover,we also acknowledge a PRACE Tier-1 award in the DECI-16 call for project REM-EPI on Archer and Archer2 EPCC in Edinburgh.This work was sponsored by NWO-Domain Science for the use of supercomputer facilities.
文摘The electron–phonon interaction is central to condensed matter,e.g.through electrical resistance,superconductivity or the formation of polarons,and has a strong impact on observables such as band gaps or optical spectra.The most common framework for band energy corrections is the Fröhlich model,which often agrees qualitatively with experiments in polar materials,but has limits for complex cases.A generalized version includes anisotropic and degenerate electron bands,and multiple phonons.In this work,we identify trends and outliers for the Fröhlich models on 1260 materials.We test the limits of the Fröhlich models and their perturbative treatment,in particular the large polaron hypothesis.Among our extended dataset most materials host perturbative large polarons,but there are many instances that are non-perturbative and/or localize on distances of a few bond lengths.We find a variety of behaviors,and analyze extreme cases with huge zero-point renormalization using the first-principles Allen-Heine-Cardona approach.
