Inconsistent Energy Values Between DeepMD Model Inference and LAMMPS

[hdjani]

Hello everyone,

I’m encountering a persistent issue with DeepMD-kit and LAMMPS, and I’d appreciate any insights or suggestions. Here’s the problem:

• Model Training: I’ve trained a DeepMD model that appears to work correctly during inference (using dp test or dp eval), yielding expected energy values.
  ASE Validation: The model also gives correct results when tested with the ASE (Atomic Simulation Environment) interface.
  -LAMMPS Issue: However, when I run the same model in LAMMPS, the calculated energies are incorrect. I’ve tried:
  Reinstalling different versions of DeepMD-kit.
  Testing on multiple computer platforms (same issue persists).
  Attached Details:
  LAMMPS Input Script (lammps.in).
  Screenshot of Energy values from DeepMD inference (correct).
  Screenshot of Energy values from LAMMPS (incorrect)
  lammps-energy.pdf
  mode-inference.pdf
  lammps.in.pdf

Has anyone encountered similar discrepancies between DeepMD and LAMMPS?
Could this be related to unit conventions, periodic boundary conditions, or how forces are mapped in LAMMPS?
Are there specific checks I should perform in my LAMMPS script?
Thank you in advance for your help!

Hania

[njzjz]

I found you used fix box/relax, which produces additional energy.