Performance assessment of the effective core potentials under the Fermionic neural network: first and second row elements

Mengsa Wang,Yuzhi Zhou,Han Wang

DOI: https://doi.org/10.48550/arXiv.2403.08287

2024-03-13

Computational Physics

Abstract:The rapid development of deep learning techniques has driven the emergence of a neural network-based variational Monte Carlo method (referred to as FermiNet), which has manifested high accuracy and strong predictive power in the electronic structure calculations of atoms, molecules as well as some periodic systems. Recently, the implementation of the effective core potential (ECP) scheme in it further facilitates more efficient calculations in practice. But there still lack a more comprehensive assessment on the ECP's performance under the FermiNet. In this work, we set sail to fill this gap by conducting extensive tests on the first two row elements regarding their atomic spectral and molecular properties. Our major finding is that in general the qualities of ECPs have been correctly reflected under the FermiNet, and a more recently built ECP, ccECP, seems to prevail on the overall performance. Meanwhile, a variation of the transferability between different ECPs has also been observed in the results of the hydrides. On the other hand, the high accuracy of the all-electron calculations is hindered by the absence of relativistic effects as one gets to the second row. Meanwhile the numerical instabilities are more often seen in the all-electron calculations, which could be another source of errors. Finally, with more in-depth discussions, we generate possible directions for developing and improving the FermiNet in the near future.