[gmx-users] PME accuracy
lindahl at stanford.edu
Wed May 7 18:41:01 CEST 2003
The alpha parameter is calculated so that the "switched off" real-space
interaction is PME_rtol fraction of the nonswitched interaction at the
cutoff, just as it is in Tom Darden's original reference code.
The fourierspacing value is the largest gridspacing you can tolerate.
The actual grid used is calculated to optimize the dimensions for FFTW
- you'll see a line about it when running grompp. In other words, if
you only increase the gridspacing marginally it sounds quite normal
that the actual grid dimensions didn't change.
On Wednesday, May 7, 2003, at 06:47 America/Los_Angeles, Jason
> Hi gmx-users,
> Unfortunately, as I think it was David who pointed it
> out to me: there is no analytical error estimate for
> smooth PME yet. But as the error-estimates for Ewald, P3M
> and regular PME would suggest, the error varies with several
> things including the number of charges. Keep this in mind
> as well as the fact that the Ewald parameter is probably
> being adjusted "behind the scenes" in order to deliver
> the specified "ewald_rtol".
> If anyone knows of a way to find out what Ewald parameter
> (alpha) is being used, let me know. Also I am not very clear
> on what is being done, if anything, to the PME parameters
> behind the scenes. For example I have on occasion reduced the
> Fourier spacing to find no change was actually made in the
> mdout.mdp file. Is that possible? or am I mistaken?
> best regards,
> Jason de Joannis, Ph.D.
> Chemistry Department, Emory University
> 1515 Pierce Dr. NE, Atlanta, GA 30322
> Phone: (404) 712-2983
> Email: jdejoan at emory.edu
> Quoting Vincent Ballenegger <vcb25 at cam.ac.uk>:
>> Dear gmx-users,
>> According to the gromacs manual, the PME method is such that "with a
>> grid spacing of 0.1nm and cubic interpolation (i.e. pme_order=4), the
>> electrostatic forces have an accuracy of 2-3e-4".
>> I did a small test and obtained the following results for the electric
>> force between two ions (charge +e and -e) a distance 2.35nm apart
>> the diagonal of a cubic box of 3.5nm:
>> * Ewald method with fourier_nx=25:
>> force = -10.824 "exact result"
>> * PME method with fourier_spacing=0.1, pme_order=4:
>> force = -7.789 error: 28% !!!
>> * PME method with fourier_spacing=0.05, pme_order=4: (very slow!)
>> force = -10.932 error: 1%
>> * PME method with fourier_spacing=0.1, pme_order=6:
>> force = -10.716 error: 1%
>> Where is the claimed accuracy of 3e-4?
>> gmx-users mailing list
>> gmx-users at gromacs.org
>> Please don't post (un)subscribe requests to the list. Use the
>> www interface or send it to gmx-users-request at gromacs.org.
> gmx-users mailing list
> gmx-users at gromacs.org
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
Erik Lindahl, MSc, PhD <lindahl at stanford.edu>
D109, Fairchild Building
Dept. Structural Biology, Stanford University School of Medicine
Tel. 650-7250754 Fax. 650-7238464
More information about the gromacs.org_gmx-users