But the 2Fo map around the ligand without omit is well defined.

Phase bias can do that sometimes... it might really be there, but you can't trust the final refined maps too much. What did the map look like before you placed the ligand?

Phenix defines a very big box around the target pdb-includes surrounding protein residues. Is there a way to control this?

Yes, instead of using AutoBuild for this, run phenix.refine with the default strategy (tailored appropriately for your structure if necessary) plus simulated annealing, using a model with the only ligand and any important waters deleted. The output map should be exactly what you want.

-Nat