NOMEGA: Difference between revisions
No edit summary |
No edit summary |
||
| Line 2: | Line 2: | ||
{{DEF|NOMEGA|50|for [[GW calculations]] |12|for [[ACFDT calculations]]}} | {{DEF|NOMEGA|50|for [[GW calculations]] |12|for [[ACFDT calculations]]}} | ||
Description: {{TAG|NOMEGA}} specifies the number of frequency grid points. | Description: {{TAG|NOMEGA}} specifies the number of (imaginary) frequency and imaginary time grid points. | ||
---- | ---- | ||
Typically {{TAG|NOMEGA}} should be chosen around 50-100 (for the parallel version, {{TAG|NOMEGA}} should be dividable by the number of compute nodes to obtain maximum efficiency). For quick and memory conserving calculations, it is sufficient to set {{TAG|NOMEGA}} to values around {{TAG|NOMEGA}} = 20-30, but then you must expect errors of the order of 20-50 meV for the gap, and 100-200 meV for the bottom of the conduction band. We furthermore recommend to increase {{TAG|NOMEGA}} not beyond 100 for a <math>k</math>-point sampling of 4×4×4 points/atom: the joint DOS and the self-energy tend to posses spurious fine structure related to the finite <math>k</math>-point grid. This fine structure is smoothed, when smaller values for {{TAG|NOMEGA}} are used, or if more <math>k</math>-points are used. For 6×6×6 <math>k</math>-points/atom {{TAG|NOMEGA}} can be usually increased to 200-300 without noticing problems associated with this kind of noise. | Typically {{TAG|NOMEGA}} should be chosen around 50-100 (for the parallel version, {{TAG|NOMEGA}} should be dividable by the number of compute nodes to obtain maximum efficiency). For quick and memory conserving calculations, it is sufficient to set {{TAG|NOMEGA}} to values around {{TAG|NOMEGA}} = 20-30, but then you must expect errors of the order of 20-50 meV for the gap, and 100-200 meV for the bottom of the conduction band. We furthermore recommend to increase {{TAG|NOMEGA}} not beyond 100 for a <math>k</math>-point sampling of 4×4×4 points/atom: the joint DOS and the self-energy tend to posses spurious fine structure related to the finite <math>k</math>-point grid. This fine structure is smoothed, when smaller values for {{TAG|NOMEGA}} are used, or if more <math>k</math>-points are used. For 6×6×6 <math>k</math>-points/atom {{TAG|NOMEGA}} can be usually increased to 200-300 without noticing problems associated with this kind of noise. | ||
Note that the spectral method (see {{TAG|LSPECTRAL}}) scales very favourable with respect to the number of frequency points, hence {{TAG|NOMEGA}}=30 is usually only slightly faster than {{TAG|NOMEGA}}=100. | Note that the spectral method (see {{TAG|LSPECTRAL}}) scales very favourable with respect to the number of frequency points, hence {{TAG|NOMEGA}}=30 is usually only slightly faster than {{TAG|NOMEGA}}=100. | ||
*N.B: Low-scaling GW and RPA/ACFDT calclulations require considerably less imaginary frequency (and time) points. | |||
== Related Tags and Sections == | == Related Tags and Sections == | ||
Revision as of 13:08, 12 August 2019
NOMEGA = [integer]
| Default: NOMEGA | = 50 | for GW calculations |
| = 12 | for ACFDT calculations |
Description: NOMEGA specifies the number of (imaginary) frequency and imaginary time grid points.
Typically NOMEGA should be chosen around 50-100 (for the parallel version, NOMEGA should be dividable by the number of compute nodes to obtain maximum efficiency). For quick and memory conserving calculations, it is sufficient to set NOMEGA to values around NOMEGA = 20-30, but then you must expect errors of the order of 20-50 meV for the gap, and 100-200 meV for the bottom of the conduction band. We furthermore recommend to increase NOMEGA not beyond 100 for a -point sampling of 4×4×4 points/atom: the joint DOS and the self-energy tend to posses spurious fine structure related to the finite -point grid. This fine structure is smoothed, when smaller values for NOMEGA are used, or if more -points are used. For 6×6×6 -points/atom NOMEGA can be usually increased to 200-300 without noticing problems associated with this kind of noise.
Note that the spectral method (see LSPECTRAL) scales very favourable with respect to the number of frequency points, hence NOMEGA=30 is usually only slightly faster than NOMEGA=100.
- N.B: Low-scaling GW and RPA/ACFDT calclulations require considerably less imaginary frequency (and time) points.