[gmx-users] RE: Re: potential energy

[Tom Dupree]

Thanks for the reply Mark,
I posted without thinking enough. In my workflow I have 10 ensembles. I sample one frame of co-ordinates per ps at the end of my NPT equilibration and generate new velocities for each of these frames. So they all have minor variations in initial position and velocity. My thinking was this would still allow system level deterministic effects to occur.
Since I can't see a statistically significant difference in the interaction energy plots I was wondering if the ligand internal energies combined with the interaction energies might show the difference between runs with movement and those without.
In the g_energy dialog I can get angle and dihedral forces for the whole system, but I can't see how to get those values for just one energy group (ligand in my case).
A thought that just occurred to me, should I also be using LIG-LIG interactions as well as LIG-rest?

All the best,
Tom


On 14/08/2012 11:07 AM, Tom Dupree wrote:
> Greetings all,
>
> Can I easily obtain the potential energy for a energy group rather than the whole simulation cell?

Yes. See manual 3.3 and 7.3.8. But below you imply you're already doing this. You can make custom groups - see http://www.gromacs.org/Documentation/File_Formats/Index_File

>   Does it make any sense to do so?

Doubt it.

> I am simulating protein ligand complexes and have observed movement in some cases. When I analyse the interaction energy sum of LJ and columbic of LIG-rest (or Lig-Protein + Lig-rest) I get very flat energy plots with no significant differences between the runs where movement occurs and runs where it does not.
> I am wondering if there are some energy terms I have not accounted for that may explain the differences and thought potential energy may be one. However any changes in ligand potential energy are going to be dwarfed by protein and water changes.

You can't expect to get an accurate quantitative picture by only looking at changes in internal energy, or only such changes for part of the system, because you have no estimate for how often those configurations arise. If the statistics from multiple runs of the same ensemble are significantly different, then you know you haven't converged any of the runs, and so probably can't say anything sensible about them.

Mark