As lammps can be run on multiple processors in a parallel way, can we find the optimum no of processors for a particular system to be used for best efficiency? Since, using more processors may slow down your program. I hope my question is not too specific.
The only way to do this is empirically test it, i.e. run the same problem
on more and more procs until you see no more speed-up. The answer
will be problem dependent (model, N, what machine, etc). You can
see these kinds of tests plotted on lammps.sandia.gov/bench.html for
the fixed-size plots.