The accuracy of DFT is determined by specifying the exchange-correlation energy functional (XC), the exact form of which is still unknown. Traditionally, approximations for XC are built on analytical solutions under conditions of low and high densities and are obtained from numerical calculations using the Quantum Monte Carlo method. However, there is no consistent and general scheme for interpolating XC and representing the functional. Most developed parameterizations primarily use a series of phenomenological rules to construct a specific XC functional.

A new class of exchange-correlation functionals has been developed in the laboratory for the basic method of quantum chemistry, density functional theory (DFT). In contrast to traditional methods, a neural network-based approach can offer a general method of parameterizing the XC functional without initial knowledge of its functional form.