Physics-informed machine learning for spectroscopic property prediction

Overview

This innovative solution leverages physics-informed machine learning to enhance the traditional time-dependent density functional theory (TD-DFT) approaches in predicting the spectroscopic properties of large molecular systems and their environmental effects. By automating the integration of TD-DFT data, this method composes the spectra of extensive molecules from smaller, local fragments and predicts solvent and environment effects in a cost-efficient manner. This approach promises significant advancements in understanding the spectroscopic properties of complex molecular systems, such as polymers and biological filaments, by providing more accurate and scalable predictions.

Technical specifications

Key features:

• Utilizes machine learning to automate the combination of TD-DFT data for large molecule spectra prediction
• Predicts solvent and environmental effects using a physics-informed machine learning framework
• Employs a fragmentation approach to manage complex molecular assemblies by analyzing representative local fragments
• Offers a back-propagation mechanism to refine data preparation and enhance predictive accuracy

Technology readiness level

Currently, the technology is at TRL 2, indicating that the principles have been observed, and the concept is being formulated. Future validation steps include developing a comprehensive training set of molecular classes and refining the fragmentation approaches for proxy subsystem selection.