[gmx-developers] Gromacs 3.3.4 PME energies keep changing with longer cutoffs

[Florian Dommert]

Hi,

 Berk is completely right, so I would also suggest to compare forces instead of energies. An appropriate set of parameters is hard to determine, since you have a bunch of possiblities to obtain an accurate force calculation. I would suggest to calculate reference forces with the help of a standard Ewald sum that has a large cutoff and a splitting parameter of about 3 and K=128. Since you can not input beta directly some small hacking of the code is necessary. src/gmxlib/ewald_util.c contains the routine calc_ewaldcoeff. Just change it that you input a negative value for ewald_rtol that is then given out as the corresponding positive beta. So something like this at the beginning of the routine:

if(dtol<0.0)
	return -dtol;

Then you can input the splitting parameter directly and play around with it.

Flo

On 18.02.2010, at 08:12, hess at sbc.su.se wrote:

> Hi,
> 
> I would like to add to this discussion that the energy is probably
> not the best quantity to look at. If errors in pair energies
> are uncorrelated (which they are not really), the error in the total
> energy will scale as the square root of the number of particles
> and the relative error will therefore seemingly decrease with system
> size. You should look at the error in the forces.
> But in both cases there is still the issue that you probably
> don't now what value of the error is acceptable.
> 
> Berk
> 
>> I now observe you're using 3.3.4. The general nature of my comments
>> still applies, but your npme is the number of cpus. You'll get
>> substantially better performance from 4.0.7 unless you need 3.3.4 for
>> some kind of scientific continuity.
>> 
>> Mark
>> --
>> gmx-developers mailing list
>> gmx-developers at gromacs.org
>> http://lists.gromacs.org/mailman/listinfo/gmx-developers
>> Please don't post (un)subscribe requests to the list. Use the
>> www interface or send it to gmx-developers-request at gromacs.org.
>> 
> 
> -- 
> gmx-developers mailing list
> gmx-developers at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-developers
> Please don't post (un)subscribe requests to the list. Use the 
> www interface or send it to gmx-developers-request at gromacs.org.

--
Florian Dommert
Dipl.-Phys.

Institute for Computational Physics

University Stuttgart

Pfaffenwaldring 27
70569 Stuttgart

Phone: +49(0)711/685-6-3613
Fax:   +49-(0)711/685-6-3658 

EMail: dommert at icp.uni-stuttgart.de
Home: http://www.icp.uni-stuttgart.de/~icp/Florian_Dommert

-------------- next part --------------
A non-text attachment was scrubbed...
Name: PGP.sig
Type: application/pgp-signature
Size: 194 bytes
Desc: This is a digitally signed message part
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20100218/c7b61bf0/attachment.sig>