Improving Force Field Accuracy by Training against Condensed-Phase Mixture Properties.

Area of Science:

• Computational Chemistry
• Chemical Physics

Background:

• Accurate molecular simulations require precise force field representations.
• Classical force fields model intermolecular interactions, like van der Waals (vdW) forces.
• Historically, vdW parameters are trained on pure substance properties, which may limit accuracy for mixtures.

Purpose of the Study:

• To investigate if including binary mixture physical property data improves force field accuracy.
• To assess the impact of retraining Lennard-Jones parameters using mixture data.
• To enhance the predictive power of molecular simulations for complex chemical systems.

Main Methods:

• Retrained Lennard-Jones parameters for the OpenFF 1.0.0 (Parsley) force field.
• Used training sets including densities and enthalpies of mixing for binary liquid mixtures.
• Incorporated densities and enthalpies of vaporization for pure liquid systems.
• Evaluated force field performance in reproducing mixture properties and solvation free energies.

Main Results:

• Retraining force field parameters against binary mixture data significantly improved accuracy.
• The enhanced force field better reproduced mixture properties, including solvation free energies.
• This approach corrected systematic errors found when training solely on pure system data.

Conclusions:

• Incorporating binary mixture data into force field training is crucial for accurate molecular simulations.
• This method provides a more robust representation of intermolecular interactions in complex systems.
• The findings advance the development of reliable computational models for chemical and biological phenomena.