Title：Acceleratingmaterials research through machine learning

Time：Apr 28, 2021 9:30-10:30

Location: Room 209, School of Materails Science and Engineering

 Tencent Meeting , Meeting ID: 642 747 534

Speaker：Tim Mueller

Host： Prof. JIANG Qing

Abstract：

Machine learning has the potential to transform computational materials researchby accelerating the calculation of material properties by orders of magnitude.I will present three examples of how this can be done at the atomic scale.  In the first, I will demonstrate how machinelearning, when combined with the cluster expansion approach, can be used tocreate highly accurate models of complex substitutional alloys.  I will present several applications of thisapproach to problems in catalysis, including the prediction of the structures andproperties of ternary alloy nanoparticles and the construction of novelcatalytic activity maps of alloy phase diagrams.  Such catalytic activity maps can be used torapidly identify the synthesis conditions that are likely to produce highlystable and active catalysts, an important step towards rational catalysisdesign.  In the second example, I willillustrate how an emerging class of machine-learned interatomic potentialsknown as "moment tensor potentials", trained on the fly, enables identificationof candidate coating materials for solid state lithium ion batteries byaccelerating molecular dynamics calculations by about seven orders ofmagnitude.  This approach results incalculated activation energies for lithium-ion diffusion that are in much betteragreement with experimental values than those calculated using only densityfunctional theory.  Using this method aspart of a computational screen, we have identified novel candidate coatingmaterials for use in all-solid-state lithium-ion batteries.  In the final example, I will demonstrate howmoment-tensor potentials, trained on the fly, can be used with geneticalgorithms to dramatically accelerate the search for low-energy structures ofsmall atomic clusters.  This approach hasenabled us to discover more than two dozen new low-energy structures for smallaluminum clusters.