Boosting ReaxFF Reactive Force Field Optimization with Adaptive Sampling.

In ReaxFF reactive force field conventional optimizations, the quality of the initial guesses plays a crucial role in determining the accuracy of the parametrization, particularly in high-dimensional spaces. To address this, we propose an adaptive sampling method that efficiently identifies high-quality initial guesses through uniform sampling followed by iterative refinement. Using this framework, we applied three optimization approaches to parametrize the Cu/H/O ReaxFF force field. The developed force field was used to study copper surface reconstruction with water molecules, revealing a stable bilayer structure driven by OH intrusion, which aligns closely with experimental observations. This adaptive sampling approach serves as a powerful tool for efficiently developing reliable ReaxFF reactive force field, enabling high-precision modeling of chemical reactions at the nanoscale.