[gmx-users] how small rcoulomb can be
dommert at icp.uni-stuttgart.de
Sat Jun 19 10:06:10 CEST 2010
tuning the pme to an optimal balance between accuracy and speed is indeed a nontrivial task. As already mentioned a too small cutoff in real space will require a large cutoff in reciprocal space and therefore the parallel scaling will be the problem. If you post all files necessary to generate a tpr, then I can tell you some settings for a desired accuracy.
Unfortunately the routine is not ready to push it to the master branch. There are some things to clarify. However to be able to use it you have to patch the file ewald_util.c a little bit, that it accepts negative values that you can input the parameter beta directly. But this is trivial. Just put
at the beginning of the routine and everything is fine. Then you can input just the negative value of the desired beta value at the ewald_rtol section of the mdp file and
everything is fine.
On 18.06.2010, at 18:40, Gaurav Goel wrote:
> I will re-ask my question as it didn't come out right in previous
> mail. Read below.
> On Wed, Jun 16, 2010 at 6:11 PM, Mark Abraham <mark.abraham at anu.edu.au> wrote:
>> ----- Original Message -----
>> From: Gaurav Goel <gauravgoeluta at gmail.com>
>> Date: Thursday, June 17, 2010 5:21
>> Subject: [gmx-users] how small rcoulomb can be
>> To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>>> For my NPT/NVT simulations I'm using a BKS potential for dispersion
>>> interactions (VdWtype=USER) which has a cut-off of 0.55
>>> nm. I also
>>> set rcoulomb=rlist=0.55 nm.
>>> My first question is regarding the choice of rcoulomb=0.55nm. Any
>>> thoughts on whether this rcoulomb is large enough to ascertain that
>>> real space electrostatics interactions are fairly small (in comparison
>>> to total coulombic interactions)?
>> That's around half the size of typical rcoulomb values used for PME, so
>> you'll need to adjust ewald_rtol and fourier_spacing to a considerable
>> degree in order that the electrostatics your PME model is comparably
>> accurate. If this model has been used with PME in the past, then you will
>> have some guidance in its literature (though GROMACS does not make it easy
>> to choose beta directly). This will mean the reciprocal-space processors are
>> doing lots more work than normal, and you'll want to set -npme to be half of
>> -np, or maybe zero. Even so, your parallel scaling will be noticeably worse
>> than normal PME.
> Thanks for your reply Mark. Any thoughts on instead increasing
> rcoulomb and rlist. Will that be more efficient and/or accurate rather
> than trying to adjust ewald_rtol and fourier_spacing with a very short
> rcoulomb? In literature I found that the optimal value people have
> used for beta is, beta*L=10.
> In my simulations, for rcoulomb=0.55, beta*L=15.52
> rcoulomb=0.9, beta*L= 9.49
> How can I obtain the magnitude of maximum wave vector (i.e., 'm' in
> Eq. 4.159 of manual 4.0)?
> I have a separate question in regards to (conserved) energy drift in a
> NVT simulation.
> I first equilibrated my system for 1 ns in NPT ensemble. Then, I
> turned-off the pressure coupling, further equilibrated my system for 1
> ns In NVT ensemble [N=2160, V=20.43 nm^3, T=4000K; rcoulomb=rlist=0.9,
> rvdw=0.55, nstlist=5]. Then in a production run of 3.5ns I get
> following energy drifts:
> Statistics over 3549897 steps [ 1000.0000 thru 4549.8960 ps ], 1 data sets
> All averages are exact over 3549897 steps
> Energy Average RMSD Fluct. Drift Tot-Drift
> Conserved En. -3.65264e+06 18936 73.1027 18.4782
> i.e. drift=0.16 kT/ns.
> If nstlist=10, drift= 0.5kT/ns
> Should I worry about these drifts?
>>> My second question is in regards to doing long (~15-100 ns) NVE
>>> simulations. It appears using coulombtype=PME-Switch will be the best
>>> option to accurately conserve energy. Any thoughts on how to choose
>>> rcoulomb_switch and rcoulomb?
>>> gmx-users mailing list gmx-users at gromacs.org
>>> Please search the archive at http://www.gromacs.org/search
>>> before posting!
>>> Please don't post (un)subscribe requests to the list. Use the
>>> www interface or send it to gmx-users-request at gromacs.org.
>>> Can't post? Read http://www.gromacs.org/mailing_lists/users.php
>> gmx-users mailing list gmx-users at gromacs.org
>> Please search the archive at http://www.gromacs.org/search before posting!
>> Please don't post (un)subscribe requests to the list. Use the
>> www interface or send it to gmx-users-request at gromacs.org.
>> Can't post? Read http://www.gromacs.org/mailing_lists/users.php
> gmx-users mailing list gmx-users at gromacs.org
> Please search the archive at http://www.gromacs.org/search before posting!
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
> Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Institute for Computational Physics
EMail: dommert at icp.uni-stuttgart.de
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 194 bytes
Desc: This is a digitally signed message part
More information about the gromacs.org_gmx-users