Machine-Learning-Assisted Detection of Unexpected Spectral Features in Excited State Spectroscopy

Despite the progress in quantum chemistry methods for excited-state calculations, as highlighted by Roldao et al. (2022) and Garrett & Fortenberry (2024), subtle or unexpected spectral features often go unnoticed or are misinterpreted. Building on the notion that “unexpected failures” in TD-DFT-based spectra can reveal underlying limitations, this project proposes a machine learning system trained on both computed and experimental spectra to identify anomalies—peaks, shoulders, or deviations from theoretical expectations—and correlate them with electronic structure characteristics or method inconsistencies. Unlike existing ML models that predict spectral properties (see Molina-Aguirre et al. 2024), this approach focuses on the detection and interpretation of unexpected features, serving both as a diagnostic tool for quantum chemistry methods and as a way to discover new photophysical phenomena. The significance lies in accelerating the identification of theoretical shortcomings or new effects, guiding both method development and experimental discovery.

References:

1. Quantum-chemistry study of the ground and excited state absorption of distyrylbenzene: Multi vs single reference methods.. J. C. Roldao, E. Oliveira, Begoña Milián‐Medina, J. Gierschner, D. Roca‐Sanjuán (2022). Journal of Chemical Physics.
2. Vibrational and Rovibrational Spectroscopic Data for the Ground and First-Excited States of Phosgene (COCl2), Formic Acid (HCOOH), and Chloroformic Acid (ClCOOH).. Noah R. Garrett, R. C. Fortenberry (2024). Journal of Physical Chemistry A.
3. Photophysics of Molecular Probes for Amyloid-β Detection: Computational Insights into the Roles of Probe Linker and Functional Groups.. Gabriela Molina-Aguirre, Sayantani Chakraborty, J. Košmrlj, L. Vuković, Balazs Pinter (2024). Journal of Physical Chemistry A.