【摘要】 针对ReaxFF反应力场，开发了一套基于PyTorch的计算软件，功能上实现了直接读取分子动力学轨迹文件并计算分子运动模拟过程中的原子所受到的势能函数等能量项。将这些能量项在分子运动模拟各个时刻的值直观地展现在科研人员面前。使用GULP也同样地计算了这些势能函数，经过对比可以看到与经GULP计算后绘制的势能函数图像重合，证明了代码是准确且可靠的。并且引入自动微分技术，使得ReaxFF可以与神经网络结合进行优化。更多还原

【Abstract】 For the ReaxFF reaction field, a set of calculation software based on PyTorch is developed, which can directly read the molecular dynamics trajectory file and calculate the energy terms such as the potential energy function of atoms in the process of molecular motion simulation. These energy terms are intuitively displayed to researchers in the simulation of molecular motion at each time value. These potential energy functions are also calculated using GULP(general utility lattice program). By comparison, it can be seen that the potential energy function images calculated by GULP coincide. It is proved that the code is accurate and reliable. And the introduction of automatic differentiation technology makes ReaxFF can be combined with neural network optimization.更多还原