[gmx-developers] suggestion for PME for 3.3.1 for free_energy=yes

[Erik Lindahl]

Hi,

Perfect. For reference, since some of this was discussed outside the  
list:

The actual PME grid interpolation occurs in scaled relative  
coordinates along the triclinic box vectors. For this reason we made  
some tests where the fourierspacing was determined along the vectors  
rather than in orthogonal cartesian space, and this seemed to solve  
the issues David reported.

Since the cell is skewed the _volume_ per grid cell for a given  
spacing will actually be smaller in a triclinic cell compared to a  
rectangular one, so you might want to play around with those settings  
to get the last bit of performance.

Cheers,

Erik


On Apr 4, 2006, at 12:41 PM, Berk Hess wrote:

> Hi,
>
> I did some accuracy tests on a rhombic dodecahedron (3 nm).
> With rcoulomb=1, ewald_rtol=1e-6, pme_order=4.
> You can see that equal grid numbers are prefered.
>
> nx*ny*nz, nx/ny, nz, RMS error with a grid of 64^3 (in kJ/mol nm)
> 2560 16 10 28.4557
> 3072 16 12 21.8189
> 3584 16 14 11.9407
> 4096 16 16 8.80569
> 3888 18 12 20.9244
> 4536 18 14 10.1648
> 5184 18 16 5.86285
> 5832 18 18 4.54341
> 5600 20 14 9.78201
> 6400 20 16 5.21705
> 7200 20 18 3.47938
> 8000 20 20 2.70206
> 7744 22 16 4.9004
> 8712 22 18 3.15134
> 9680 22 20 2.2014
> 10648 22 22 1.74797
> 10368 24 18 3.01268
> 11520 24 20 1.97726
> 12672 24 22 1.40936
> 13824 24 24 1.12307
> 13520 26 20 1.83786
> 14872 26 22 1.19105
> 16224 26 24 0.852791
> 17576 26 26 0.749445
> 17248 28 22 1.11295
> 18816 28 24 0.729732
> 20384 28 26 0.595822
> 21952 28 28 0.50345
>
> _______________________________________________
> gmx-developers mailing list
> gmx-developers at gromacs.org
> http://www.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.
>