weird problem with energy minimization

Hi LAMMPS Users:

I came across a weird problem about energy minimization in LAMMPS. For a polymer model I built, I did a energy minimization without saving restart files. Then I decided to re-do it and save restart files for subsequent calculations. It turns out the results of the two minimization were not same. At first I thought there might be randomization in the minimization. But after more tests, I found that all calculations without saving restart files had the same output while all those saving restart files having the same result. It seems that saving restart files is affecting calculation, although it’s hard believe for me. Can someone please take a look at this issue? The data file is big so I didn’t attach it here. If you want to do a test, the data file is available upon asking. Below I paste the input file:

For both dynamics and minimization writing a restart
file forces reneighboring to be done on that timestep.
This is so the restarted simulation will be identical
to the continued one.

But this means two simulations done with vs w/out restarting
can diverge, due
to round-off and different ordering of atoms, etc.

A hi-tolerance minimization is probably quite suceptible
to that effect.

You can set the minimization to only write a restart file
when the minimization is done, which is probably all
you care about.