[gmx-developers] inclusion of an estimation for the absolute error of the PME into gmx4

Florian Dommert dommert at fias.uni-frankfurt.de
Sun Dec 14 13:02:55 CET 2008

Dear Gromacs developers and users,

   if I remember correctly, somebody asked a question regarding the PME
the aim was an implementation of an error approximation for gmx.
Unfortunately I am not able to find it anymore in the archive, because
I am very interested in the kind of error approximation. 

So in case the author of the mentioned thread is reading this, I have a question, 
are you estimating the absolute root mean square error of the 
electrostatic forces, or how are you going to tune the PME parameters ?

Now I have a question regarding to the kind of differentiation in the
PME routine ? Sofar gromacs uses analytical differentiation of the reciprocal energy
after backtransforming to real space. Another approach is the
differentiation in reciprocal space. Why is only the first one used in
gromacs ? Depending on the required accuracy and system size, both
methods have their advantages in terms of performance. Additionally, the
differentiation in reciprocal space is momentum conserving, isn't it ?

At the end, a last question to our dear developers. Do you mind adding a
further input parameter for Ewald/PME/PPPM in .mdp .

grompp was altered to recognize a parameter named 'ewald_beta', set to 0.0 by default,
that is choosen for the corresponding fr->ewald_beta, if (ir->ewald_beta > 0.0 ). 
In md.log the output for the Gaussian width has just changed a little
bit, that beta is directly accessible without changing the column for
the value of 1/beta, to minimize problems with existing scripts that perhaps
read the md.log.
This allows to tune the PME parameters much more easily, because
the absolute error depends strongly on beta for a given cutoff, mesh size
and interpolation order.

So the new parameter would not change anything in the current behaviour
of gromacs and just advance its flexibility :) If you like the idea,
a patch for the cvs gmx-4-0-release branch is attached.



Florian Dommert

Computational and Theoretical Softmatter & Biophysics group

Frankfurt Institute for Advanced Studies
Johann-Wolfgang-Goethe University

Ruth-Moufang-Str. 1
60438 Frankfurt am Main

Phone: +49(0)69 / 798 - 47529
Fax:   +49(0)69 / 798 - 47611

EMail: dommert at fias.uni-frankfurt.de
Home: http://fias.uni-frankfurt.de/~simbio/Florian_Dommert
-------------- next part --------------
A non-text attachment was scrubbed...
Name: beta.patch.tar.bz2
Type: application/octet-stream
Size: 28007 bytes
Desc: not available
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20081214/4a401aca/attachment.obj>

More information about the gromacs.org_gmx-developers mailing list