Exchange-correlation functional development: Data-driven and physically-constrained

We present a methodology that combines data science and physical constraints for the development of new exchange-correlation functionals [1].

We aim for multi-purpose functionals that are applicable to compute a wide range of physical properties with optimal accuracy and transferability.

In this talk, we present the general methodology which lead to the MCML functional [1], its modifications as well as their performance on a number of data sets.

[1] K. Brown et al., J. Comput. Chem., vol. 42, 2004, 2021