bond across periodic boundary

Dear all:

Recently, I applied lammps (17 Oct 2012 version) to calculate the lattice thermal conductivity of infinite polymer crystal. My simulation box was built by replicating unit cell to form a supercell. This supercell is the simulation box. My force field is General Amber and periodic boundary condition was set in the three directions.

Because my simulation system is the infinite polymer crystal, it is necessary to build a bond to go across the box boundary. But during my simulation, the error, Bond/angle/dihedral extent > half of periodic box length, always occurs. Then the simulation stops. So I am very confused that (i)Was there something wrong with my modeling process, leading to my initial structure going wrong? (ii) For this version of lammps, does minimum image principle still work? If it works, I think that the atom and its image in the neighboring box will be bonded instead of the two atoms in the two end of the box.

Best wishes!