[gmx-users] OPLS-AA Coulomb-14 energies incorrect?

[Erik Lindahl]

Hi Malcolm,

It's always nice to see comparisons!

First - I think I know what the cause of the difference might be. By 
default, the pdb2gmx program does not generate 1,4 interactions for 
hydrogens since that's the prescribed setup for the other force fields. 
  Use the -H14 flag for pdb2gmx to get the OPLS-AA standard.

This will probably be the default in future Gromacs versions (it's 
already changed in CVS), but we would probably confuse users more by 
changing the default in a point release....

The parameters have mainly been cross-validated with Tinker, the 
energies for aminoacids in the Jorgensen publications (check the header 
of the force field for references) and MacroModel.

> The Coulomb (SR) component of the GROMACS energy agrees with the
> corresponding energy from TINKER (i.e. as calculated by setting
> chg-14-scale to 0.0), but the Coulomb-14 energy is out by a sizeable
> percentage. i.e for the attached file GROMACS reports 224.739 kJ/mol
> while TINKER reports 258.3687 kJ/mol.

Again, I would guess that difference is due to not using the -H14 flag. 
You are right that the relative difference for this term is large, but 
compared to the total nonbonded energy it is very small, so it is not 
likely to have affected your simulations significantly.

> Comparing GROMACS (OPLS-AA parameter files checked out of CVS about
> a month ago) and TINKER 4.0 (with the supplied oplsaa.prm parameters) 
> on
> a couple of potential energy calculations, I found that they agree on
> the bond stretch, bend and LJ energies. The improper dihedral energies
> differ, though this may be due to a problem with TINKER. GROMACS uses
> some updated proper dihedral parameters, absent from the TINKER.

Right - the bonded parameters are essentially the Amber ones, and they 
haven't been changed in years. There are two major modifications to the 
OPLS torsion angles not present in Tinker (one by Jorgensen in 1999 and 
another set that formed the OPLS-AA/L version in 2001). I also recall 
there were some minor issues in Tinker; some uncommon charge groups 
would not end up being neutral with their charges.

Another issue you might stumble upon is that Amber and Gromacs used 
different atom orders for the improper dihedrals. I've changed this in 
our later versions, but this is not significant. The difference is only 
second order, and the improper force constants are just 
order-of-magnitude guesses anyway.  I've just uploaded a tarball with 
the latest OPLS parameters to the contrib section of www.gromacs.org if 
you want to play with it.

In all cases of doubt I've used the "original" values, i.e. Amber for 
bonded interactions and Jorgensen/Kaminski for torsions and charges. I 
also communicated rather extensively with Bill Jorgensen and Julian 
Tirado-Rives during the implementation (they were quite helpful).

Thanks for the help with checking things - and let me know if the 
problems persist with -H14.

Cheers,

Erik