Order of per-atom model deviations in model_devi.out when using atomic = true

[TypicalBarmalei]

Hello,

I am using DeepMD + LAMMPS, and I rely on per-atom deviations (atomic = true in the model_devi command) to identify “bad” atoms.

What I want to clarify is: in which order are the per-atom deviations written to model_devi.out? - LAMMPS "internal" IDs (during the run), atoms ID as a property, or else?

In my LAMMPS setup, I write atom dumps with

dump mydump all custom N dump.lammpstrj id type x y z
dump_modify mydump sort id

which guarantees the atoms in the dump are sorted by ascending atom ID (1..N). In atomic force deviations (model_devi.out), suppose a max. deviation is achieved on j-th atom. Does this j correspond to the j - id of an atom in the dump file?

As I understand it, LAMMPS stores atoms internally in arbitrary memory order during the run, and DeepMD likely receives atoms in that internal order when evaluating energies/forces.

So my concern is: does model_devi.out list deviations in the same internal order that LAMMPS provides to DeepMD, or are they sorted by atom ID?

This matters because if the orders differ, then I cannot simply map “line i” in model_devi.out to “atom ID i” in the dump file for further analysis.

Thanks a lot!

[njzjz]

Atomic deviations are sorted by atom IDs since v2.0.0.