Sorry, my message was truncated, thanks to a bumpy train ride.
The last part is that the units used in the Colvars fix are those used in the LAMMPS run, i.e. for units lj, the energy is measured in units of epsilon and the force in units of epsilon/sigma.
Another thing that comes to mind is that in a LJ system you may be comparing the result of a PMF calculation with the logarithm of the RDF computed during the LAMMPS run. If you do so, be aware of the Jacobian term in the PMF (-2kBT*ln® for the distance), which is typically hidden when defining the PMF from the RDF. Please consult the Mol Phys paper for more details.