General Multiobjective Force Field Optimization Framework, with Application to Reactive Force Fields for Silicon...

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Physics

Background:

• First-principles-based force fields offer superior accuracy and transferability for molecular dynamics simulations compared to phenomenological methods.
• Optimally preparing these force fields from diverse quantum mechanical data is crucial but challenging due to large parameter spaces.

Purpose of the Study:

• To introduce GARFfield (genetic algorithm-based reactive force field optimizer method), a hybrid optimization scheme for developing accurate and transferable force fields.
• To demonstrate GARFfield's capability in generating specialized force fields for complex chemical processes.

Main Methods:

• Developed GARFfield, a hybrid multiobjective Pareto-optimal parameter development scheme.
• Utilized genetic algorithms, hill-climbing routines, and conjugate-gradient minimization for parameter optimization.
• Applied GARFfield to develop reactive force fields (ReaxFF) and electronic force fields (SiC) from quantum mechanical data.

Main Results:

• Successfully developed two distinct force fields using GARFfield: one for silicon carbide growth and another for nonadiabatic dynamics.
• Demonstrated the efficiency and flexibility of GARFfield for optimizing parameters from large quantum mechanical datasets.
• Showcased the method's applicability to demanding simulations like ReaxFF and electronic force fields.

Conclusions:

• GARFfield provides an efficient and flexible framework for the parallel optimization of force field parameters.
• The developed method facilitates the creation of accurate, transferable force fields for a wide range of chemical and physical phenomena.
• GARFfield supports various interatomic potential types, including reactive, electronic, and coarse-grain force fields.