No it shouldn’t be calculated twice. Running pair_style airebo with the LJ off and the torsion on should be equivalent to the Brenner 2002 paper with the torsion on. In principle anyway. There are some minor differences between the lammps version of airebo and the research code used in the Stuart / Harrison / Brenner groups. Harrison has a postdoc working on reconcilling these differences in Lammps and hopefully will have something availible for lammps users in the future.
Dave Schall