Improving MP2 band gaps with low-scaling approximations to EOM-CCSD

TL;DR

This paper evaluates a low-scaling approximation to EOM-CCSD called P-EOM-MP2 for predicting band gaps, showing it overestimates gaps slightly but performs better than G0W0 for large-gap materials.

Contribution

The study introduces and assesses P-EOM-MP2 as a computationally efficient method for band gap prediction, highlighting its advantages over traditional GW approximation for certain materials.

Findings

P-EOM-MP2 overestimates band gaps by 0.3 eV on average.

G0W0 underestimates band gaps by 0.6 eV on average.

P-EOM-MP2 performs better for large-gap materials.

Abstract

Despite its reasonable accuracy for ground-state properties of semiconductors and insulators, second-order Moller-Plesset perturbation theory (MP2) significantly underestimates band gaps. Here, we evaluate the band gap predictions of partitioned equation-of-motion MP2 (P-EOM-MP2), which is a second-order approximation to equation-of-motion coupled-cluster theory with single and double excitations. On a test set of elemental and binary semiconductors and insulators, we find that P-EOM-MP2 overestimates band gaps by 0.3 eV on average, which can be compared to the underestimation by 0.6 eV on average exhibited by the G0W0 approximation with a PBE reference. We show that P-EOM-MP2, when interpreted as a Green's function-based theory, has a self-energy that includes all first- and second- order diagrams and a few third-order diagrams. We find that the GW approximation performs better for materials with small gaps and P-EOM-MP2 performs better for materials with large gaps, which we attribute to their superior treatment of screening and exchange, respectively.