SPARC: A Computational Framework for Prediction of Infrared Spectra for Zeolitic Catalyst Systems via Machine Learning of Born Effective Charges and Atomic Velocities

<jats:p>Molecular spectroscopy is a versatile tool for structural characterization, mechanistic illustration, and material discovery. Conventional polarization-based ab initio molecular dynamics methods for molecular spectra simulation struggle with host–guest porous materials such as zeolites. To overcome these limitations, we introduce SPARC, a framework for spectra prediction via autocorrelation of response charges and velocities. SPARC integrates two complementary models, namely, MACE for accurate potential energy surfaces (PES) and polipy4vasp for predicting rigorously defined Born effective charges (BECs) via derivative learning. It is benchmarked against conventional methods, including the static approach using harmonic approximations and the dynamic approach based on polarization analysis along trajectories. The results demonstrate that SPARC not only accurately describes the IR peak positions but also captures broad bands, significantly outperforming many available methods. Validation against transient experimental IR spectra of SSZ-13 zeolites with different Al configurations confirms its robustness. Overall, SPARC is a powerful and cost-effective tool for quantitative analysis of IR spectra and for identifying adsorbates and intermediate species in complex host–guest porous materials.</jats:p>