Prediction of pair interactions in mixtures by matrix completion

Area of Science:

• Computational chemistry
• Physical chemistry
• Chemical engineering

Background:

• Molecular simulations predict mixture properties using interaction models and combining rules.
• Existing combining rules often lack accuracy without experimental data adjustment.
• Adjustable binary parameters (ξij) are used to improve predictions of unlike interactions in mixtures.

Purpose of the Study:

• To develop a novel method for predicting adjustable binary parameters (ξij) for unstudied mixtures.
• To enhance the accuracy of molecular simulations for mixture properties.
• To provide predictive access to the description of unlike intermolecular interactions.

Main Methods:

• Utilized a matrix completion method (MCM) from machine learning (ML).
• Applied the method to predict ξij for molecular simulations of Henry's law constants.
• Demonstrated the prediction of ξij for unstudied mixtures with high accuracy.

Main Results:

• The MCM-based approach accurately predicts ξij for unstudied mixtures.
• Using predicted ξij significantly improves the accuracy of Henry's law constant predictions compared to default values (ξij = 1).
• The method shows a substantial increase in predictive accuracy for mixture properties.

Conclusions:

• The developed ML-based MCM provides a powerful tool for predicting intermolecular interactions in mixtures.
• This approach enhances the reliability of molecular simulations for various mixture properties.
• The generic methodology is transferable to other mixture properties and equations of state.