A functional that accounts for the short-range dynamic correlation energy has been derived by our research group in terms of the two-electron density, rather than the usual one-electron density or the occasional on-top density found in [1]. The two-electron density functional (TDF) is based on modelling the correlated pair density under the constraints of an electron-electron cusp and evaluating its effects on the electron-electron potential energy. Employment of the aforementioned functional, however, becomes computationally expensive as electronic systems get larger and more complex. The present work focuses on accelerating the calculation of TDF dynamic correlation energies using a supervised machine learning approach. Accordingly, the accuracy of a multi-layer perceptron algorithm for predicting the desired energies will be examined once a full hyperparameter optimization is carried out.

[1] Hollett, J. W. and Pegoretti, N. (2018). The Journal of Chemical Physics, 148, 164111.