[gmx-users] gromacs energies very different from other MD engine for the very same system and conditions

Erik Lindahl lindahl at cbr.su.se
Thu Jul 29 00:26:16 CEST 2010


On Jul 27, 2010, at 5:06 PM, Ilja Khavrutskii wrote:

> I think the bond terms differ simply due to the round off error. The gro file is worse than pdb when it comes to round off in Cartesian coordinates. You can confirm this is the case by taking your gro file and converting it back to pdb using gromacs tools and then use the resulting pdb (with the same round of as the gro file now) file from gromacs in NAMD. You should see the same numbers for bonded terms then.

It's even more complex than that :-) 

To accurately compare coordinates (which affect the energy) you might even need to use PDB as input to both programs (grompp can read those too). The problem is that e.g. 15.50 and 1.550 look like a simple factor 10 difference, but remember that floating-point numbers are stored in base-2, and then the difference is no longer exactly a factor 10 (because that can't be expressed in binary).

Depending on the gromacs version, make sure you have absolutely identical coordinates, so that none of the programs have modified e.g. terminal hydrogens or oxygens. Gromacs-4.0.x does some stupid things for a few FF's there, but Gromacs-4.5 should keep them entirely clean.

I'm not saying errors are impossible, but Eric Sorin has done _very_ extensive validation of all the original Amber parts against the actual Amber package, not only for equilibrium structures but also severely distorted conformations to catch minor errors in functional forms.

In fact, he did find one minor program error related to ordering of improper torsions in TPR during his work, but that turned out to be in Amber ;-) Suffice to say, I trust his work a _lot_.



Erik Lindahl <lindahl at cbr.su.se>
Professor, Computational Structural Biology
Center for Biomembrane Research & Swedish e-Science Research Center
Department of Biochemistry & Biophysics, Stockholm University
Tel: +468164675 Cell: +46703844534

More information about the gromacs.org_gmx-users mailing list