Search papers, labs, and topics across Lattice.
This study introduces a distributed-memory software implementation of periodic coupled-cluster theory (CCSD) that efficiently utilizes up to 12 nodes with 96 cores each, enabling dense Brillouin zone sampling with up to 216 k-points. By achieving this high sampling density, the researchers can extrapolate cohesive energies and band gaps to the thermodynamic limit with improved accuracy, yielding benchmark values for eight simple semiconductors and insulators. The results show average errors in cohesive energy of 0.1-0.2 eV and band gaps of about 0.4 eV when compared to experimental data, marking a significant advancement in the precision of CCSD calculations.
Achieving a thermodynamic limit in periodic CCSD calculations reveals benchmark cohesive energies and band gaps with unprecedented accuracy.
The high computational cost of periodic coupled-cluster theory has limited the density of Brillouin zone sampling, yielding finite-size errors that need to be removed by extrapolation. Here we report the development and application of a distributed-memory software implementation of periodic coupled-cluster theory with single and double excitations (CCSD) that runs efficiently on up to 12 nodes with 96 cores each. This new implementation allows ground-state and excited-state calculations in which the Brillouin zone is sampled with up to $6^3=216$ $k$-points, allowing us to reliably extrapolate to the thermodynamic limit. For eight simple semiconductors and insulators, we report the cohesive energy and band gap, which are converged to 0.1 eV, providing definitive benchmark numbers for the CCSD level of theory. Compared to experimental values, average errors for the cohesive energy are 0.1-0.2 eV (typically an underestimate), and average errors for the band gap are about 0.4 eV (typically an overestimate).
