Development of a new parameter optimization scheme for a reactive force field based on a machine learning approach.

Area of Science:

• Computational Materials Science
• Chemical Engineering
• Materials Chemistry

Background:

• Reactive molecular dynamics (MD) simulations require accurate force field parameters.
• Optimizing these parameters is computationally intensive and crucial for simulation accuracy.
• Existing methods for parameter optimization can be inefficient.

Purpose of the Study:

• To develop a novel, efficient machine learning-based method for optimizing ReaxFF parameters.
• To apply the optimized parameters to simulate a relevant materials growth process.
• To validate the accuracy and utility of the developed optimization technique.

Main Methods:

• Developed a machine learning approach combining k-nearest neighbor and random forest regressors.
• Used this method to optimize ReaxFF parameters.
• Performed reactive MD simulations of α-Al2O3 crystal growth via chemical vapor deposition (CVD).

Main Results:

• Successfully optimized ReaxFF parameters using the machine learning approach.
• Reproduced the crystal structure of α-Al2O3 at 2000 K with good accuracy.
• Reactive MD simulations indicated faster growth on the (110) surface compared to the (0001) surface.

Conclusions:

• The developed machine learning technique provides an efficient way to optimize ReaxFF parameters.
• The optimized parameters enable accurate simulations of materials growth processes like CVD.
• This method can advance the understanding of chemical reactions in CVD processes.