I know people are exploring this front currently. Basically you could use results from DFT+U to train the potential; but I am also told that strongly correlatio

I know people are exploring this front currently. Basically you could use results from DFT+U to train the potential; but I am also told that strongly correlation can now be better described by SCAN functionals better than LDA+U etc so it would be even easier then.