I have been testing whether I could use Modeller to minimize a
protein-protein complex.
Modeller's strength is satisfying a bunch of distance restraints. It's
not a great general purpose dynamics package - for that you'd be better
off using something like CHARMM or GROMACS. Your script looks fine but
you haven't given Modeller any restraints between the receptor and
ligand, so there's nothing to pull them together. In a regular MD
package you might expect solvent, dispersion, or electrostatics to do
this, but Modeller doesn't model any of those interactions.