Issue with fix rigid/npt/small in LAMMPS 29Aug2024 – Unexpected Box Deformation

dmf.in (10.7 KB)
DMF.mol (989 Bytes)
lmp_29Aug_2024.out (13.5 KB)
nohup_lmp_Aug_2023.out (29.7 KB)


Hello everyone,
I am using LAMMPS (29Aug2024) to simulate 450 rigid DMF (Dimethyl Formamide) molecules with united-atom methyl groups under ambient conditions. To enforce rigidity, I am using fix rigid/npt/small with a timestep of 0.2 fs.
After 300,000 steps, the potential energy and simulation box size appear reasonable. However, as the number of timesteps increases, the simulation box unexpectedly deforms along the z-axis.
To investigate, I ran an NVT simulation, which revealed that the pressure is 135 atm instead of 1 atm in a box of 50 Angstroms.
The same simulation runs without any issues in the Aug/2023 LAMMPS version, producing expected results (density = 0.943 g/cm³, potential energy = 44.22 kJ/mol, pressure = 1.08 atm, temperature = 298.55 K).
I have attached:
nohup.out files from both LAMMPS versions showing the box size evolution.
The DMF.mol file.
My input script.
A JPEG file showing the CS2_M potential and atom numbering.
Any insights or suggestions would be greatly appreciated.
Regards
Costas Vlahos

How long does it take to run? This could be a good candidate for Git bisect (Git - git-bisect Documentation).

30.5 ns/per day in a i5 12generation with time step 1fs

So about 15 minutes to run 30k steps, which is pretty long. Can you make a more minimal reproducer? Something that runs in 30 seconds or less would be ideal.

This is almost 30 seconds with 250 DMF molecules
fast.in (10.7 KB)

The 29Aug2024 version of LAMMPS was installed using oneAPI 2025 and Intel MKL FFT, while the Aug2023 version was installed with GNU and FFTW3. However, when the Aug2023 version was reinstalled using oneAPI 2025, the box deformation issue also appeared in this setup. This suggests that the problem is related to oneAPI 2025.