You can use any of the existing compute molecules as a template. The idea
is to create an array on every proc that is the # of molecules long. Loop
over the atoms on each proc, and have it contribute to that array. Then
sum the array across procs. See compute msd/molecule for a specific
example.
Steve