MSDGW F
MSDGW_F = [real]
Default: MSDGW_F = -1
Description: A positive value of MSDGW_F triggers the mixed stochastic-deterministic compression algorithm of Altman and co-workers.[1]
MSDGW_F is the constant energy ratio of the compression algorithm. If set to a positive value, energies beyond the protected space defined by MSDGW_NP are subdivided into energy bins of width and replaced by other energies , such that . The original orbitals are replaced by MSDGW_NXI randomly linear combined orbitals. Larger values of MSDGW_F increase the compression level at expense of accuracy. The same holds true for smaller values of MSDGW_NXI.
This compression algorithm has been developed to reduce the large number of virtual states required for the calculation of the screened interaction in GW calculations. It has been demonstrated[1] thay one can reduce the virtual manifold by more than 50 per cent and speed up the GW step by a factor of 2 or more with an resulting error of only 50 meV or less on the quasi particle band gap. Comparable results can be achieved for Constrained Random Phase Approximation (cRPA) calculations.
Mind: Avaliable as of VASP.6.5.2.
Use cases
- Recommended for ALGO=EVG0W[R|RK]|CRPA[R|RK]:
The compression can be used for all type of GW calculations regardless if the calculation is performed in steps or in the all-in-one mode. Following lines in stdout and OUTCAR indicate that the compression has been performed:
=> avg. energy ratio F (%): 1.00 bands after compression: 240
To test different compression settings, it is recommended perform the GW/CRPA calculation in steps and to set MSDGW_F only in the actual GW step. This avoids repeating the expensive exact diagonalization of the Kohn-Sham hamiltonian.
Caveats
Care must be taken for GW/CRPA calculations that include the long-wave limit stored in WAVEDER. After band compression, this limit must be re-calculated by setting LOPTICS in combination with LPEAD.
Related tags and articles
MSDGW_NXI, MSDGW_SEED, MSDGW_NP