Empowering Quantum Chemistry through Evolutionary Neural Architecture Search

Density Functional Theory (DFT) is a widely adopted method in quantum chemistry for investigating the electronic structure of many-electron systems. However, the core Kohn-Sham (KS) equation of DFT contain exchange-correlation approximations that limit the accuracy of DFT calculations. To overcome this limitation, we focus on one-dimensional (1D) DFT and propose a data analytics approach for accurate modeling of the exchange-correlation terms. This approach involves the development of a neural architecture search framework, guided by multi-objective differential evolution (MODE) algorithm, to generate exchange-correlation potential (Vxc) models that are both highly accurate and computationally lightweight. Experimental results demonstrate that the Vxc model derived from this framework not only reduces the number of parameters but also significantly improves the accuracy of the KS equation in predicting the properties of 1D molecules, validating the effectiveness of the proposed method.