Performance assessment of the effective core potentials under the fermionic neural network: First and second row...

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Materials Science

Background:

• Deep learning has advanced electronic structure calculations via neural network variational Monte Carlo (FermiNet).
• Effective core potential (ECP) schemes enhance computational efficiency but require validation within new methods.
• Comprehensive assessments of ECP performance in FermiNet are lacking.

Purpose of the Study:

• To comprehensively evaluate the performance of various effective core potentials (ECPs) within the FermiNet framework.
• To assess the accuracy of FermiNet-ECP calculations for atomic, spectral, and molecular properties of first and second-row elements.
• To identify the most suitable ECPs for accurate and efficient electronic structure calculations using FermiNet.

Main Methods:

• Extensive testing of ECPs across different chemical systems (atoms, molecules, periodic systems).
• Comparison of FermiNet calculations using various ECPs against all-electron results.
• Analysis of atomic, spectral, and molecular properties to benchmark ECP performance.
• Evaluation of correlation consistent ECP (ccECP) and energy consistent correlated electron pseudopotential (eCEPP) performance.

Main Results:

• FermiNet generally reflects the inherent quality of different ECPs.
• ccECP and eCEPP demonstrated superior overall performance among tested ECPs.
• ccECP offered slightly better spectral precision and broader elemental coverage.
• eCEPP provided systematic treatment of core polarization and consistency in shape and energy.

Conclusions:

• ECP quality is accurately represented in FermiNet calculations.
• ccECP and eCEPP are recommended for accurate electronic structure calculations with FermiNet.
• All-electron calculations face limitations due to relativistic effects and numerical instabilities for heavier elements.
• Future work should focus on improving FermiNet's capabilities, potentially incorporating relativistic effects.