I am sorry that I did not describe.the questions well. In a simple way, I could not repeat my simulations results. Even in the same computer, the same input data and the same script.
Are you running in parallel? I believe that when fix deform does its box flip operation
in a long shear run, that it perform irregular communication. That
operation is not 100% reproducible in the sense that atoms
get re-ordered depending on which message arrives first.
Re-ordering will lead to small round-off level differences
which will grow over time. Since you are running 0.5M steps,
the two trajectories could diverge dramatically. However, statistically
they should be similar if you have a well-posed simulation.
In other words, it that’s what happening, you shouldn’t worry about it.