Computational and Machine Learning Approach to Electrochemistry of Disordered Rocksalt Cathode Materials

The increasing demands in electrical energy storage require the discovery of high energy density based cathode materials of lithium-ion rechargeable batteries. Disordered rocksalt cathode with Li-excess (DRX) materials is promising candidates as these materials do not require cation chemistry to favor any particular ordering and can be synthesized with a very wide variety of elements. However, the computational modeling for DRX is difficult as it can be composed from a wide variety of chemistry with site disorder.<br/>In this report, we will demonstrate two state-of-art theoretical approaches to tackle the complex modeling of DRX systems. The first one is cluster expansion (CE) model, which extracts the effective interactions as descriptors of energetics from the information in the first-principle calculations. The well-trained CE model is used to sample the equilibrium states of DRX from Monte Carlo simulations. The CE of DRX shows success in explaining the short-range order (SRO) effects in both thermodynamics and kinetics. Another is a deep neural network (DNN) trained directly by the experimental results from electrochemistry testing. We treated the voltage profiles as a function of compositions and testing conditions. And the DNN is trained with an end-to-end learning scheme, where the dQdV information is also included and appropriately regularized. The DNN can interpolate and make predictions for the chemical space that has not yet been tested, which can accelerate the exploration of DRX and other electrode materials.