Lattice Dynamics and Raman Spectroscopy of Solid-State Ion Conductors with Machine Learning*

Solid-state ion conductors (SSICs) are attracting increased interest for next-generation energy storage technologies. Recent reports highlighted that host lattice vibrations enable fast conduction of mobile ions by taking them into strongly anharmonic domains of the potential energy surface. Here, we investigate the role of lattice dynamics in SSICs by combining machine-learning molecular dynamics (MLMD) and ML-based Raman calculations, using AgI and Na₃PS₄ as model systems for (super-)ionic conductors. Specifically, we investigate the accuracy of ML-generated force fields for diffusion coefficients and vibrational properties of these systems. Applying a Δ-ML model [1] allows us to compute Raman spectra of these materials with molecular dynamics at reduced computational cost and to compare them to experimental data.