Implementing NBFiXes from CHARMM force fields

[Charles Brooks]

I have recently run into some problems that I would like some advice on fixing. These have to do with implementing NBFIXes from the CHARMM force field in the CHARMM/OpenMM interface. In particular, the NBFIXes specify a special set of LJ epsilon and sigma to be used for a specific pair of atoms but do not change the charges on those atoms (differing from what's often done with 1-4 interactions). I have implemented this using the addException method in OpenMM, whereby I loop over affected pairs and provide the values of the modified epsilon, sigma and the product of the charge on each atom of the pair. The problem comes when I compare the energy and forces I get with what are the "correct" values. For systems w/o periodic BCs I match energies and forces, no problem, however, for PME/Ewald use with cutoff and PBCs the energies and forces do not match, and my analysis suggests it is because the PBCs are somehow not being applied to the atom pairs whose energies are part of the exception. I am not certain this is the problem, as I cannot easily check what PBCs would give w/ cutoffs but not PME, since I have no direct way of turning off the reaction field correction and cannot easily compute it in CHARMM.

Any suggestions on how to deal with this would be helpful, or whether this represents a yet to be developed feature.

Charles Brooks

[Peter Eastman]

It is applying the periodic boundary conditions to the exceptions, but it isn't applying the cutoff to them, which will cause the numbers to be slightly different. For consistency it probably should. I'll fix that.

Peter

[Michael Garrahan]

Here is a unit test illustrating the problem Prof. Brooks described on the Reference platform.