Development of a ReaxFF force field for simulation of coal molecular structures

Accurately constructing the coal molecular structure at the atomic level is crucial for understanding its properties and behavior. However, the energetic feasibility of most existing coal molecular models has not been adequately considered during their construction process, limiting their reliability. As a reactive force field, ReaxFF is capable of describing the dynamics of chemical bonding and accurately assessing the total energy of a molecular system and the energy changes during molecular reactions. This makes ReaxFF a powerful tool for modeling the structure of coal molecules and chemical reaction processes. Although the generic ReaxFF force field can be applied to coal model construction, an optimized force field specifically tailored for coal is necessary due to the complex structural characteristics of coal. To accelerate the development of the ReaxFF force field for coal, we have written a ReaxFF parameter optimization program based on the PyTorch framework. This program uses the AdamW optimizer for parameter adjustment. The computational efficiency and accuracy of force field optimization are significantly improved with graphics processing unit acceleration technology. We used this optimizer to train against a dataset obtained from density functional theory calculations and finally developed the Coal-FF ReaxFF force field. Through energy and structural validations, we have demonstrated that the present force field can effectively reproduce the structure and energy variations of coal structural units. In contrast to the existing HCONSB force field, which has been previously used to describe the combustion behavior of coal, the Coal-FF force field exhibits higher accuracy. This force field shows promise as a valuable tool for future investigations into the complex structure of coal.