Reconstruction of Nuclear Ensemble Approach Electronic Spectra Using Probabilistic Machine Learning

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Spectroscopy

Background:

• Accurate prediction of molecular electronic spectra is crucial for understanding photophysical and photochemical processes.
• The Nuclear Ensemble Approach (NEA) is a common computational strategy, but relies on subjective broadening parameters.
• Limitations in computational resources often restrict the number of sampled configurations, impacting spectral accuracy.

Purpose of the Study:

• To develop and validate a novel, data-driven approach for reconstructing NEA spectra.
• To eliminate the need for subjective broadening parameters (δ) in NEA spectral predictions.
• To enhance the accuracy and objectivity of theoretical molecular electronic spectra.

Main Methods:

• Developed the Gaussian Mixture Models-Nuclear Ensemble Approach (GMM-NEA).
• Utilized Gaussian Mixture Models (GMMs), a probabilistic machine learning algorithm.
• Implemented an algorithm for detecting anomalous quantum mechanical computations (outliers).

Main Results:

• GMM-NEA effectively reconstructs spectra without phenomenological broadening (δ).
• The GMM-NEA approach outperforms other data-driven models, especially for small datasets.
• An outlier detection algorithm was successfully used to improve spectral shape and uncertainty estimation.

Conclusions:

• GMM-NEA offers a more objective and accurate method for predicting molecular electronic spectra.
• The developed approach addresses limitations of traditional NEA by removing subjective parameters.
• Applied GMM-NEA to predict the photolysis rate of HgBrOOH, relevant to atmospheric chemistry.