MAXDIS: Difference between revisions

Latest revision as of 08:50, 10 June 2026

MAXDIS = [real]

Default: MAXDIS

= 0.0 (switched off)

Description: This tag sets the maximum distance that an atom is allowed to travel (in Angstrom) between two ab-initio steps before the charge density is reset to atomic an atomic charge density.

At each ionic step, the maximum Cartesian displacement of any atom (using minimum-image convention) is compared to MAXDIS. If exceeded, the charge density extrapolation controlled by IWAVPR is skipped and atomic densities are used instead. Setting MAXDIS=0.0 disables this reset entirely.

MAXDIS is particularly important in on-the-fly machine learning force field (ML_MODE=train) calculations, where many machine-learning-driven ionic steps are executed between successive ab initio evaluations. This allows atoms to travel a considerable distance between two DFT calculations, making the charge density from the previous ab initio step a poor initial guess for the next one — potentially causing slow or problematic electronic convergence.

@@ Line 7: / Line 7: @@
 At each ionic step, the maximum Cartesian displacement of any atom (using minimum-image convention) is compared to MAXDIS. If exceeded, the charge density extrapolation controlled by {{TAG|IWAVPR}} is skipped and atomic densities are used instead. Setting {{TAG|MAXDIS}}=0.0 disables this reset entirely.
-{{TAG|MAXDIS}} is particularly important in [[:Category:Machine-learned force fields|on-the-fly machine learning force field (MLFF)]] calculations, where many ML-driven ionic steps are executed between successive ab initio evaluations. This allows atoms to travel a considerable distance between two DFT calculations, making the charge density from the previous ab initio step a poor initial guess for the next one — potentially causing slow or problematic electronic convergence.
+{{TAG|MAXDIS}} is particularly important in [[:Category:Machine-learned force fields|on-the-fly machine learning force field]] ({{TAG|ML_MODE}}=train) calculations, where many machine-learning-driven ionic steps are executed between successive ab initio evaluations. This allows atoms to travel a considerable distance between two DFT calculations, making the charge density from the previous ab initio step a poor initial guess for the next one — potentially causing slow or problematic electronic convergence.
 == Related tags and articles ==
-{{TAG|IVAPR}}, {{TAG|IBRION}}, {{TAG|POTIM}}, {{TAG|ML_LMLFF}}, {{TAG|ML_MODE}}
+{{TAG|IWAVPR}}, {{TAG|IBRION}}, {{TAG|POTIM}}, {{TAG|ML_LMLFF}}, {{TAG|ML_MODE}}
 [[Category:INCAR tag]]