Revealing Fast Ionic Conduction in Solid Electrolytes through Machine Learning Accelerated Raman Calculations

TL;DR

This work exploits the resulting low-frequency, diffusive Raman scattering as a spectral signature of fast ionic conduction and develops a machine learning-accelerated computational pipeline to identify promising solid electrolytes based on this feature, demonstrating the predictive power of this approach for sodium-ion conductors.

Abstract

Fast ionic conduction is a defining property of solid electrolytes for all-solid-state batteries. Previous studies have suggested that liquid-like cation motion associated with fast ionic transport can disrupt crystalline symmetry, thereby lifting Raman selection rules. Here, we exploit the resulting low-frequency, diffusive Raman scattering as a spectral signature of fast ionic conduction and develop a machine learning-accelerated computational pipeline to identify promising solid electrolytes based on this feature. By overcoming the steep computational barriers to calculating Raman spectra of strongly disordered materials at finite temperatures, we achieve near-ab initio accuracy and demonstrate the predictive power of our approach for sodium-ion conductors, revealing clear Raman signatures of liquid-like ion conduction. This work highlights how machine learning can bridge atomistic simulations and experimental observables, enabling data-efficient discovery of fast-ion conductors.

Figures (15)

• Figure 1: Machine learning-enabled insights into fast-ion conduction. (A) Workflow integrating machine-learned force fields and polarizabilities to simulate vibrational spectra and identify signatures of fast-ion conduction using Raman spectroscopy. (B) Schematic comparison of two ion transport regimes: thermally activated hopping (left) versus liquid-like fast conduction (middle), characterized by distinct structural dynamics and diffusion behavior. Mean squared displacement (right) distinguishes fast and slow conduction regimes based on their time dependence.
• Figure 2: Performance of the machine-learning approach for AgI. (A) Learning curve showing $R^2$ scores of the machine-learning predictions as a function of the training set size $N_T$. (B and C) Scatter plots showing predictions versus ab initio values for diagonal and off-diagonal components of the dielectric tensor. (D) Computed Raman spectra at 500 K obtained from both machine-learning-predicted and ab initio polarizabilities. An experimentally measured spectrum at 500 K is shown for comparison brenner_etal_2020.
• Figure 3: (A--C) Raman spectra of pristine and W-doped Na$_3$SbS$_4$ at 300 K as well as $\gamma$-Na$_3$PS$_4$ at 900 K. The low-frequency regions are magnified to highlight the differences. (D--F) Schematic sketches of the structures and conduction mechanisms of these materials.
• Figure S1: Forces predicted by the ML force fields compared to DFT reference values. (A) AgI. (B) Na$_3$SbS$_4$. (C) Na$_{2.94}$Sb$_{0.94}$W$_{0.06}$S$_4$. (D) $\gamma$-Na$_3$PS$_4$. Root mean squared errors (RMSE) are given in units of eV Å$^{-1}$.
• Figure S2: Atom-resolved vibrational density of states (VDOS). (A) AgI. (B) Na$_3$SbS$_4$. (C) Na$_{2.94}$Sb$_{0.94}$W$_{0.06}$S$_4$. (D) $\gamma$-Na$_3$PS$_4$.