Using fix rigid with Langevin gives the same error.
Using fix rigid/nve without Langevin makes it work, but I then have linear trajectories.
It does mean that something the Langevinn thermostat is making the simulation go wrong.
I noticed that it has a link with the ellipsoicity of the rigid bodies, as well as with the chosen timestep.
Does anyone has an explanation?
did you see similar behavior with fix rigid instead of fix rigid/nve? What
will happen if you use fix rigid/nve without langevin?