Publication date: 21st July 2025
Raman spectroscopy is a non-invasive and broadly accessible technique for probing atomic vibrations in solid-state materials. However, its interpretation often depends on comparisons with preselected reference systems. First-principles calculations offer a powerful alternative for interpreting experimental Raman spectra, but they become computationally demanding for systems with large unit cells, defects, or mobile ions. To address these challenges, we developed fast computational frameworks that integrate machine-learning force fields (MLFFs) [1] and machine-learned polarizability tensors to predict Raman signatures associated with mobile ions and point defects in solid-state ion conductors.
Using this ML-Raman framework, we identify low-energy Raman modes in superionic conductors and broadened peaks in disordered systems, shedding light on the conductivity mechanisms of mobile cations. Furthermore, we introduce a novel method that combines MLFFs with atomic Raman tensors to predict the vibrational signatures of ionic point defects [2]. This approach has been successfully applied to capture temperature-dependent changes in experimentally measured Raman spectra of Ni-doped SrTiO₃, which were attributed to local variations in the dominant ionic defects. Our framework establishes new synergies between theory and experiment, enhancing the understanding of dynamical properties in energy materials.