rerun and initial run

Since nobody has replied yet, I will give you my best hunch. I don't
remember, but I'm guessing that coordinates saved in dump files are
only accurate to about 7 digits (check the number of digits by looking
at the coordinates in your dump file). Furthermore, fcm is a
measurement of force, and for lennard-jones systems, the
force-dependence on position can be very sensitive. I would expect
this to amplify the roundoff error. (In other words, I would expect
the effect of roundoff error on the forces to be larger than the
roundoff error on positions.)

I don't know how many particles are in your system. For a large
enough number of particles, you might expect these random errors in
the force to gradually cancel out when calculating fcm (scaling as

Nevertheless, I'd guess that a relative error of 3e-04 seems plausible to me.

If you are worried about this, then try increasing the number of
digits in the coordinates of your dump file using the "dump_modify"
command with the "format" argument, and see if the error goes down
accordingly. (If so, then likely this is the source of your error.)

Hope this helps