报告题目：Complex chemical reactions simulated by physics and machine learning
报告摘要：Hydrocarbon molecules are the main components of fuels, and the investigation of their combustion mechanisms is one of the key issues to be addressed in order to simulate engine combustion and thus drive engine design. Due to the limitation of force field accuracy, there is still much room to improve the reliability of the results of the widely used molecular mechanics methods for the simulation of combustion reactions. Due to the large amount of computational resources required for quantum chemical methods, it is not feasible to directly simulate the combustion mechanism of hydrocarbon fuels with them. In our previous work, we have developed a fragment-based quantum chemical calculation method MFCC-combustion, which achieves an efficient and accurate calculation of the energy and force of the simulated system. This method, when combined with MD simulation algorithm, enables ab initio molecular dynamics simulation (AIMD) of fuel combustion through reasonable temperature and pressure control. Recently, we have further improved the simulation efficiency of AIMD by about three orders of magnitude based on the Deep Potential model, thus achieving nanosecond-scale reactive MD simulations of hydrocarbon combustion. The development of this method is expected to provide an efficient and accurate research tool for the understanding of hydrocarbon combustion mechanism and the construction of combustion data base.