next up previous contents index
Next: OMEGAMAX Up: BSE Bethe-Salpeter calculations Previous: BSE Bethe-Salpeter calculations   Contents   Index

N.B. This document is no longer maintained, please visit our wiki.


The compute time for BSE and Cassida type calculations grows with the third power of the number of included occupied and unoccupied bands

$\displaystyle (N_{\rm occ}N_{\rm virtual} N_{\rm k})^3

and the memory requirements increase quadratically

$\displaystyle (N_{\rm occ}N_{\rm virtual} N_{\rm k})^2

Please be aware that symmetry is not exploited by the BSE code, hence memory requirements can be excessive. To allow for calculations on large systems, the BSE code distributes the BSE matrix among all available cores, and uses scaLAPACK for the diagonalization.

VASP always uses the orbitals closest to the Fermi-level, and NBANDSO ( $ N_{\rm occ}$) and NBANDSV ( $ N_{\rm virtual}$) determines how many occupied and unoccupied orbitals are included. The defaults are fairly "conservative" and equal the total number of electrons/2 (this usually implies that all occupied state are included). For highly accurate results, NBANDSV often needs to be increased, whereas for large systems one is often forced to reduce both values to much smaller numbers. Sometimes qualitative results for band like Wannier-Mott excitons can be obtained even with a single conduction and valence band.

N.B. Requests for support are to be addressed to: