ATB torsional parameters are written in the form k[1 + cos(n φ − δ)], which is what md++ evaluates when NTBDN = 0 — its default, and the setting used to derive every parameter here. Some torsions carry a phase shift δ of ±90°, which is how an asymmetric rotational profile is represented. At NTBDN = 1 md++ restricts δ to 0° and 180° and will refuse such a topology at start-up, naming the offending dihedral; it never evaluates it incorrectly.

The same applies to the ATB protein force field, whose φ and ψ backbone terms have used a 90° phase for some time.

Note also that gromos++ analysis programs which evaluate the older torsion form k[1 + cos(δ) cos(n φ)] — ener among them — will misreport the energy of a term with a ±90° phase, reading it as a constant. That is a limitation of those tools, not of the topology: the MD engine, GROMACS, LAMMPS and AMBER all evaluate the term as written.