ML MB MIN: Difference between revisions
No edit summary |
No edit summary |
||
| Line 4: | Line 4: | ||
Description: {{TAG|ML_MB_MIN}} sets the minimum number of local reference configurations required for generation of a machine-learned force field. {{NB|mind|Available as of {{VASP}} 6.4.3. Previous versions do not support this tag but act as if its default was {{TAGO|ML_MB_MIN|2}}.}} | Description: {{TAG|ML_MB_MIN}} sets the minimum number of local reference configurations required for generation of a machine-learned force field. {{NB|mind|Available as of {{VASP}} 6.4.3. Previous versions do not support this tag but act as if its default was {{TAGO|ML_MB_MIN|2}}.}} | ||
---- | ---- | ||
Machine-learned force fields based on the kernel method require a reasonable amount of reference atomic environments in order to provide meaningful predictions. Typically, the on-the-fly algorithm in {{VASP}} collects these local reference configurations (also called kernel basis functions) during an MD simulation ({{TAGO|ML_MODE|train}}). However, at the start of the trajectory usually only very few local reference configurations are available. How many depends on the system size and | Machine-learned force fields based on the kernel method require a reasonable amount of reference atomic environments in order to provide meaningful predictions. Typically, the on-the-fly algorithm in {{VASP}} collects these local reference configurations (also called kernel basis functions) during an MD simulation ({{TAGO|ML_MODE|train}}). However, at the start of the trajectory usually only very few local reference configurations are available. How many depends on the system size, symmetry and number of atoms (per type). | ||
Revision as of 11:03, 16 February 2024
ML_MB_MIN = [integer]
Default: ML_MB_MIN = 3
Description: ML_MB_MIN sets the minimum number of local reference configurations required for generation of a machine-learned force field.
Mind: Available as of VASP 6.4.3. Previous versions do not support this tag but act as if its default was ML_MB_MIN = 2.
|
Machine-learned force fields based on the kernel method require a reasonable amount of reference atomic environments in order to provide meaningful predictions. Typically, the on-the-fly algorithm in VASP collects these local reference configurations (also called kernel basis functions) during an MD simulation (ML_MODE = train). However, at the start of the trajectory usually only very few local reference configurations are available. How many depends on the system size, symmetry and number of atoms (per type).
Related tags and articles
ML_LMLFF, ML_MODE, ML_MB, ML_MCONF_NEW, ML_MCONF, ML_LBASIS_DISCARD