NNP
This section contains all information to run an helium-solute interaction Neural Network Potential (NNP) calculation. [Edit on GitHub]
Keywords
Keyword descriptions
- NNP_INPUT_FILE_NAME
Type: string
Default: input.nnDescription: File containing the input information for the setup of the NNP (n2p2/RuNNer format).
- RAD_SPLINE_N
Type: integer
Default: 8192
Usage: RAD_SPLINE_N 8192Description: Number of knots per radial group in the cubic-Hermite spline tables that tabulate the radial symmetry-function product y(r) = exp(-eta*(r-rs)^2) * fcut(r). Cubic-Hermite per-evaluation error scales as O(1/n^4); the default keeps the spline residual inside the NNP regression tolerance for a radial cutoff of ~12 bohr. Models with larger cutoffs or stricter tolerances may need a larger n. Memory cost scales linearly in n per radial group.
- SCALE_FILE_NAME
Type: string
Default: scaling.dataDescription: File containing the scaling information for the symmetry functions of the NNP.
- VERLET_SKIN
Type: real
Default: -1.00000000E+000 [bohr]
Usage: VERLET_SKIN [bohr] 0.5Description: Verlet skin distance for the NNP descriptor neighbour cell-list. The neighbour-list cutoff is (symmetry-function cutoff + skin); the cell-list chain is rebuilt only when an atom drifts more than skin/2 from its rebuild-time position, analogous to the LAMMPS ‘neighbor
bin’ command. Larger skin reduces the rebuild rate but enlarges the per-atom neighbour list. A negative value (default) selects the automatic heuristic MIN(0.5 bohr, 0.1*cutoff). Useful upper bound is half the smallest perpendicular cell width.