Improving physically predictive force field by adding many body effect corrections with machine learning

The ionic liquids (IL) electrolyte have been utilized in the development of safe and high-performance lithium-ion batteries. As is well known, the important determinant of the performance of a lithium-ion battery is the association of anions and lithium-ions. These properties can be predicted through molecular dynamics simulation, but developing a high-accuracy prediction model remains a challenge due to the problem that the size of lithium ions is too small. Here we developed the machine learning (ML)-based force field by adding the many-body effect correction to the physics-based predictive force field utilizing ML. The ML-based force field significantly improved the accuracy and can be used to capture the dynamics of lithium-ions and clustering of lithium-ions – the anion of IL in lithium/IL mixed electrolyte systems more precisely. Our results can predict the transport properties of lithium/IL and provides novel design path for obtaining high-performance lithium-ion batteries.