Number of bands NBANDS - Revision history

Vaspmaster at 11:41, 3 November 2020

2020-11-03T11:41:30Z

← Older revision		Revision as of 11:41, 3 November 2020
Line 6:		Line 6:

	----		----
	[[Category:~~Calculational issues~~]]		[[Category:Common Pitfalls]]

Kresse at 11:17, 14 October 2020

2020-10-14T11:17:42Z

← Older revision		Revision as of 11:17, 14 October 2020
Line 1:		Line 1:
	~~One should chose~~ NBANDS so that a considerable number of empty bands is included in the calculation. As a minimum ~~we require~~ one empty band. VASP will give a warning, ~~if this is not~~ the ~~case~~.		NBANDS must be chosen in such that a considerable number of empty bands is included in the calculation. As a minimum, VASP requires one empty band, otherwise VASP will give a warning. Only for large gap insulators, accurate results can be obtained without empty bands and the warning might be ignored.

	{{TAG\|NBANDS}} is ~~also~~ important ~~from a technical point of view:~~ In iterative matrix-diagonalization schemes eigenvectors close to the top of the calculated number of ~~vectors~~ converge much slower than the lowest ~~eigenvectors~~. This might result in a significant performance loss if not ~~enough~~ empty bands are included in the calculation. Therefore we recommend to set {{TAG\|NBANDS}} to ''NELECT/2 + NIONS/2''~~, this~~ is also the default setting ~~of the ''makeparam'' utility and of VASP.4.X. This setting~~ is safe in most cases. In some cases, it is also possible to decrease the number of additional bands to ''NIONS/4'' ~~for large systems without~~ performance ~~loss, but on~~ the other hand transition metals do require a much larger number of empty bands (up to ''2*NIONS'').		Choosing {{TAG\|NBANDS}} large enough is important for the following reason. In iterative matrix-diagonalization schemes, eigenvectors close to the top of the calculated number of states converge much slower than the lowest eigen states. This might result in a significant performance loss, if not sufficiently many empty bands are included in the calculation. Therefore, we recommend to set {{TAG\|NBANDS}} to ''NELECT/2 + NIONS/2''. This is also the default setting, which is safe in most cases. In some cases, it is also possible to decrease the number of additional bands to around ''NIONS/4'' slightly improving the performance. On the other hand, some transition metals and atoms with open f shells might require a much larger number of empty bands (up to ''2*NIONS'').

	To check this parameter perform several calculations for a fixed potential ({{TAG\|ICHARG}}=12) with an increasing number of bands (e.g. starting from ''NELECT/2 + NIONS/2''). An accuracy of <math> 10^{-6}</math> should be obtained in 10-15 iterations. Mind that the {{TAG\|RMM-DIIS}} scheme ({{TAG\|~~IALGO~~}}=48) is more ~~sensible~~ to the number of bands than the default CG algorithm ({{TAG\|~~IALGO~~}}=8).		To check this parameter perform several calculations for a fixed potential ({{TAG\|ICHARG}}=12) with an increasing number of bands (e.g. starting from ''NELECT/2 + NIONS/2''). An accuracy of <math> 10^{-6}</math> should be obtained in 10-15 iterations. Mind that the {{TAG\|RMM-DIIS}} scheme ({{TAG\|ALGO}}=Fast) is more sensitive to the number of bands than the default Davidson algorithm ({{TAG\|ALGO}}=Normal).

	----		----
	[[Category:Calculational issues]]		[[Category:Calculational issues]]

Karsai: Created page with "One should chose NBANDS so that a considerable number of empty bands is included in the calculation. As a minimum we require one empty band. VASP will give a warning, if this..."

2019-03-21T13:53:39Z

Created page with "One should chose NBANDS so that a considerable number of empty bands is included in the calculation. As a minimum we require one empty band. VASP will give a warning, if this..."

New page

One should chose NBANDS so that a considerable number of empty bands is included in the calculation. As a minimum we require one empty band. VASP will give a warning, if this is not the case.

{{TAG|NBANDS}} is also important from a technical point of view: In iterative matrix-diagonalization schemes eigenvectors close to the top of the calculated number of vectors converge much slower than the lowest eigenvectors. This might result in a significant performance loss if not enough empty bands are included in the calculation. Therefore we recommend to set {{TAG|NBANDS}} to ''NELECT/2 + NIONS/2'', this is also the default setting of the ''makeparam'' utility and of VASP.4.X. This setting is safe in most cases. In some cases, it is also possible to decrease the number of additional bands to ''NIONS/4'' for large systems without performance loss, but on the other hand transition metals do require a much larger number of empty bands (up to ''2*NIONS'').

To check this parameter perform several calculations for a fixed potential ({{TAG|ICHARG}}=12) with an increasing number of bands (e.g. starting from ''NELECT/2 + NIONS/2''). An accuracy of <math> 10^{-6}</math> should be obtained in 10-15 iterations. Mind that the {{TAG|RMM-DIIS}} scheme ({{TAG|IALGO}}=48) is more sensible to the number of bands than the default CG algorithm ({{TAG|IALGO}}=8).

----
[[Category:Calculational issues]]