Machine Learning\nfor High-Throughput Configuration\nSampling of Li−La−Ti−O Disordered Solid-State\nElectrolyte

Most solid-state lithium electrolytes are disordered\nionic crystalline\nmaterials that possess crystallographic sites that can be vacant or\noccupied by different ions. The presence of these partially occupied\nsites enables lithium diffusion through their lattice and makes such\nmaterials promising for developing all-solid batteries. High-throughput\ncomputational screening of such materials must bypass costly DFT sampling\nof disordered configurations and, therefore, commonly relies on the\ncomputationally efficient Coulomb approximation to find just a few\nrepresentative low-energy ionic configurations, for which DFT is then\nused to quickly predict a number of important materials’ properties,\nsuch as the electrochemical stability window. This work demonstrates,\nusing the Li−La−Ti−O solid electrolyte (LLTO)\nas an example, that the Coulomb approximation fails to correctly detect\nthe most stable arrangement of Li and La ions in the LLTO, which has\na noticeable impact on the accuracy of subsequent computational prediction\nof the electrochemical stability window of the material. The analysis\nherein shows that the sampling problem arises from the relatively\nmodest geometry relaxation of the LLTO lattice. A kernel ridge regression\nmachine learning (ML) method employing the smooth overlap of atomic\npositions as a structure descriptor (SOAP-KRR) leads to significant\nimprovements in detecting the most stable configurations of the LLTO.\nThe universal ML potential based on the multiple atomic cluster expansion\nis also found to be reliable but to a lesser extent than SOAP-KRR.\nRemarkably, accurate energies can be obtained with SOAP-RKK trained\non as few as 40 LLTO structures, making this method promising for\ndesigning force matching ML potentials that can serve as a computationally\ninexpensive alternative to the costly DFT structure relaxation in\nhigh-throughput screening of large data sets of ionic materials.