Yes, I used this way before, except I had a run 1 instead of run 0. Indeed, this seems to be the most elegant way to go. Thanks a lot.
But something unexpected (to me) happened when I ran with the reset_timestep. My final energy was slightly different than before (with a continuous run). The difference is really small (much less than one percent), so this is just about curiosity. What could lead to this difference?
First maybe, a more general question: if you run the same input file, with exactly the same options, number of cpus, machine, binary, random seed number, etc, are you always guaranteed to get the same result? Or is there some “ghost” out there, or random process, involved?
In addition to this, I ran some time ago two relaxation processes. In that case I know I used different binaries - one compiled with Intel, the other one with PGI. At the end of the process, the difference in total energy was almost 2%, which is noticable. I should say I am still running with an older version of the code and maybe this has been addressed before, or even this may be something I should know already; but could the compiler (or the optimization options) yield that kind of difference? If so, what should we trust? Probably this is not related to LAMMPS only, but more generally to any code used by the community. I’d be glad to get your opinion on this.
2008/8/29 Aidan Thompson <[email protected]>