Computational Predictions of NMC Cathode Materials

One of the most commercially successful Li-ion battery cathodes, LiMO₂ (M=Ni,Mn,Co mixtures) has continued to be of interest to many as more of the possibly infinite combinations are tested. Yet there are still very few known phases. In this work, we attempt to explore the full ternary phase diagram through high throughput computation. A set of training data is generated from first principles using Density Functional Theory. The data is then used to train a reduced order model that is then solve in high resolution with Monte Carlo Simulations and ultimately generates a convex hull of stable maxtures. From this we can theoretically predict a collection of properties that will identify a set of promising new materials for experimental testing.

A unique aspect of this work is the incorporation of uncertainty estimation in both the DFT training data and ultimately the parameters of the reduced order model used. This is done with the use of the Bayesian Error Estimation Functional (BEEF-vdW) which allows for built in error estimation trained to experimental data. The addition of error estimation can inform the choice of what additional data points should be used to maximize the increase in model fit.