Physics-Based Machine Learning Models Predict Carbon Dioxide Solubility in Chemically Reactive Deep Eutectic Solvents

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Materials Science

Background:

• Carbon dioxide (CO2) is a major greenhouse gas driving global warming.
• Deep eutectic solvents (DESs) offer a sustainable and eco-friendly medium for CO2 capture.
• Chemically reactive DESs show superior CO2 absorption compared to non-reactive DESs, but lack accurate predictive models for solubility.

Purpose of the Study:

• To develop accurate machine learning (ML) models for predicting CO2 solubility in chemically reactive DESs.
• To leverage physics-driven input features derived from quantum chemical methods.
• To enable precise prediction of CO2 solubility, facilitating the design of advanced CO2 capture materials.

Main Methods:

• Collected 214 data points on CO2 solubility in 149 chemically reactive DESs from published literature.
• Utilized σ-profile descriptors, calculated via the COSMO-RS method, as physics-driven input features for ML models.
• Trained and evaluated various ML models, including artificial neural networks (ANNs).

Main Results:

• The developed ML models, particularly the ANN, demonstrated high accuracy in predicting CO2 solubility.
• The best performing ANN model achieved an average absolute relative deviation of 2.94% on testing datasets.
• COSMO-RS-derived σ-profile features proved effective for predicting bond formation, despite not explicitly modeling reaction profiles.

Conclusions:

• Machine learning models can accurately predict CO2 solubility in chemically reactive DESs.
• The developed models can significantly accelerate the design and application of effective DES-based CO2 capture systems.
• This work provides a valuable computational tool for advancing sustainable CO2 capture strategies.