Internal force corrections with machine learning for quantum mechanics/molecular mechanics simulations.

Area of Science:

• Computational chemistry
• Molecular dynamics simulations
• Machine learning applications in chemistry

Background:

• Ab initio quantum mechanics/molecular mechanics (QM/MM) molecular dynamics is crucial for calculating thermodynamic properties like potential of mean force.
• However, these simulations are computationally intensive and time-consuming.

Purpose of the Study:

• To develop a novel, computationally efficient method for QM/MM molecular dynamics simulations.
• To accelerate the calculation of thermodynamic properties for chemical reactions.

Main Methods:

• Implemented an internal force correction method for low-level semiempirical QM/MM simulations.
• Utilized a machine learning scheme to predict the internal force as a correction term.
• Applied the method to two reactions in aqueous solution, incorporating the ML-predicted force at each integration step.

Main Results:

• Successfully reproduced potentials of mean force at the ab initio QM/MM level.
• Achieved a computational cost reduction of approximately 2 orders of magnitude.
• Demonstrated the accuracy and efficiency of the machine learning-enhanced QM/MM approach.

Conclusions:

• The developed machine learning-based internal force correction significantly accelerates QM/MM simulations.
• This method offers a powerful tool for studying complex chemical processes computationally.
• Highlights the potential of machine learning to advance molecular simulations in chemistry.