Unraveling the Temperature Effects on the Raman Spectra of V₂O₅ by Combining Machine Learning and Molecular Dynamics

Raman spectroscopy serves as a powerful operando tool for probing the atomic structure in catalysts and functional materials, yet interpreting the spectra acquired under complex working conditions often requires the aid of reliable computational approaches. We considered the important catalyst V₂O₅ and experimentally observed an anomalous blue shift of a Raman mode near 400 cm^-1 with increasing temperature. To explain this phenomenon, a Raman calculation method that integrates graph neural network machine learning with molecular dynamics was developed, which enables the accurate reproduction of the experimental Raman spectra. Calculation results reveal that the blue shift originates from the local contraction of a specific V-V interatomic distance under anisotropic thermal expansion. This leads to an increase in the vibrational frequency of a mode dominated by an oxygen atom bridging the two V atoms. This work not only provides mechanistic insight into temperature-dependent Raman responses of V₂O₅ but also establishes an effective computational framework for interpreting Raman spectra recorded under realistic conditions, such as those of catalysts in operation.