restart file does not have the same energy as it ended

I am trying to do a two step minimization by further reducing dmax to 1.e-2 at the second minimization in a single in.script:

1: min_modify line quadratic
2: min_modify dmax 1.0e-2 line quadratic

I found that the optimized energy after the first minimization is not the same as the initial energy for the second minimization:

1: Step TotEng E_pair Press press Temp
0 -1331.5855 -1331.5855 -101568.61 -101568.61 0
32 -1346.05868134 -1346.0587 1791.1867 1791.1867 0

2: Step TotEng E_pair Press press Temp
32 -1346.05866794 -1346.0587 1792.1264 1792.1264 0
34 -1346.0587 -1346.0587 1763.8549 1763.8549 0

The step 32 not only have different energies but also different pressure. Although the difference at the fourth decimal place might not make a big difference, I’m just curious if this is due to the precision? In fact, probably due to the energy difference (increased energy), the final 2-step optimized energy is higher than using only first minimization (with dmax being the default 0.1 value).

I also tried to separate the 2-step with 2 in.scripts. Generate a restart file for the 2nd minimization. The first minimization results in the same energy and pressure as the 2-step together one. But the 2nd minimization, results in a different energy/pressure for the 32nd step (from the restart and the previous 2-step), and as a result, the final minimized energy/pressure differ too.

The log files for two steps together (lmp.o2steps), and separte (lmp.ofirst, lmp.osecond) are attached. In both cases, the stopping criteria is “linear alpha search is zero” and I am not able to get rid of it. Any suggestion about the minimization procedure?

Thank you.

lmp.o2steps (2.79 KB)

lmp.ofirst (1.58 KB)

lmp.osecond (1.57 KB)

If you used a variable charge potential such as reaxff or comb, maybe slight changes in charge are responsible. But it is difficult to comment without full input.


Yes Ray. I am using Reaxff, so it’s the variable charge potential causing this slight difference. Thanks for the answer.