Adaptive Force Field Parameter Optimization for Expanding Reaction Simulations within Wide-Ranged Temperature.

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Engineering

Background:

• Simulating high-temperature chemical reactions, like polymer thermal decomposition, remains challenging.
• Accurate simulations require precise force field parameters, particularly for reactive force fields (ReaxFF).

Purpose of the Study:

• To develop an adaptive framework for automatic ReaxFF parameter optimization.
• To enhance the accuracy of chemical reaction simulations at experimental temperatures.

Main Methods:

• Utilized Random Forests and interpretable machine learning to identify influential parameters.
• Trained deep neural network (NN) models to associate parameters with reference properties.
• Employed a Genetic Algorithm (GA) with NN surrogate models and quantum mechanical targets for accelerated parameter searching.

Main Results:

• The adaptive optimized ReaxFF accurately predicted peak temperatures during resin pyrolysis.
• Achieved reasonable product composition predictions under experimental-like conditions.
• Demonstrated improved simulation accuracy at experimental temperatures.

Conclusions:

• The developed adaptive framework significantly enhances ReaxFF parameter optimization.
• This approach leads to more accurate and universal reaction simulations.
• Facilitates advancements in computational chemistry for high-temperature processes.