[lammps-users] (no subject)

Please continue to post on this thread to the mail list, so others
may comment.

Parallel tempering is a bit confusing with respect to what you are asking.
If you weren't doing tempering, but just a bunch of independent
runs at different Temps, then what you have below would make sense,
and you would get 1 dump file for a run at each Temp.

However, when you do tempering, the Temps are swapped between
worlds, and so for the simulation running on one world of processors
(and its dump file), the temperature will change continuously during
the simulation, as will its fix nvt params (these are changed behind
the scenes. So you will confuse yourself if you name it's dump file
snap.Temp.lammpstrj. You will see this by looking at the log
file produced by that proc - the temperature of the run will vary
all over the place as tempering proceeds.

If you want stats on all snapshots at a given Temp, then they
are scattered through all the dump files produced by all the sets
of procs. At any point in time, exactly one snapshot in one of
the files has that Temp, but you will need to look at the hi-level
tempering log file to figure out which one it is, and extract it
in a post-processing step. Like the Pizza.py suggestion I made before.


Dear Steve,

I do real tempering, I am confued by your two replies. First let me ask you a naive question. When tempering occurs, is the configuration information exchanged across processors or only temperature information exchanged? I think both ways are OK. I assume that the configuration information is exchanged in LAMMPS since your first reply question why I need to track the configuration information with the associated temperature. That is the reason I said the target temperature (Tstop) in fix NVT command for each processor keeps the same during simulation in my previous email. If this is the case, each configuration kept in one processor should be a point of the phase space of the system at Tstop which is associated with that processor. This should be guaranteed by the acceptance rule for tempering. Therefore we should use all configurations in one processor to calculate the property of the system at Tstop. However, your latest reply seems to imply that only temperature information is exchanged when tempering occurs. So please answer my naive question.

Have a nice day!


LAMMPS exchanges T between simulations, not atom coords. So it
is backwards from what you are saying.