A Study of Advancements in Ionic Conductivity Methods and Machine Learned Interatomic Potentials to Calculate Ionic Conductivities with a Model Solid Electrolyte (Li₄SiO₄)_x(Li₃PO₄)_1−x

Several recent publications [1-2] have advanced methodologies for realistic computations of ionic conductivity in electrolytes from molecular dynamics simulations including the effects of correlated ionic motions. At the same time, efficient machine learning techniques have been developed which enable molecular dynamics calculations to be run using large simulation cells for long simulation times, while preserving the accuracy of density functional interactions. Using the Allegro [3] machine learning software package, we report the results of a systematic study of the fidelity, convergence, and statistics of these techniques to describe the ionic conductivity of a model electrolyte alloy – (Li₄SiO₄)_x(Li₃PO₄)_1−x [4]

[1] A. Marcolongo and N. Marzari, Phy. Rev. Mat. 1, 025402 (2017).

[2] L. Gigli, D. Tisi, et al., Chem. Mater. 36, 1482 (2024).

[3] A. Musaelian, S. Batzner, et al., Nat. Comm. 14, 579 (2023).

[4] Y-W. Hu, I. D. Raistrick, and R. A. Huggins, J. Electrochem. Soc. 142, 1240 (1977).