While trying to use harmonic, or guassian, distance restaints in low resolution centroid only docking in the docking_protocol.linuxgccrelease I've encountered weird Scoretype: atom_pair_constraint score errors that prevent the output of docked proteins.
Using a single distance restiant taken from a crystallographic positions, AtomPair CA 132A CA 88B HARMONIC 46.0 5.0, model generation fails as the Scoretype: atom_pair_constraint score never reaches the cutoff of 1. However the distance restaint AtomPair CA 132A CA 88B HARMONIC 46.0 15.0 returns models with a CA-CA distance of ~46 angstroms, well within the first AtomPair restaint mean and SD of 46.0 and 5.0.
If it helps, my input file is:
-dock_pert 3 8
-mute core.util.prof ## dont show timing info
#-out:file:fullatom # Boolean flag to indicate full atom
This seems to be a stumper to me.
1) I'm not sure when the filter is applied - it may be that you aren't getting sufficiently close models at some earlier step, but later refinement fixes the problem so the models you see later are ok. That's not very consistent with the low_res_protocol_only flag.
2) The math for calculating constraints has a bunch of variables with sort-of-known values (by which I mean, you can figure them out, but I don't know what they are at the moment). In particular, the constraint weight and the way the SD is factored into the score may just mean that the looser constraint is necessary to get sufficiently low constraint values. An energy value of 1 does seem quite low, that basically means "totally satisfied constraints". Try plotting the constraint func based on the functional form in the constraints documentation page and seeing how your measured devations score - hopefully they really do score more than 1 with the tighter SD...