Re: PME and r-RESPA issue in NAMD

From: Thomas C. Bishop (
Date: Fri Jun 20 2008 - 09:56:25 CDT

This brings up a related question/point.

In NAMD NVE is really NVE so N, V, and E (theoretically) conserved.
Computationally this is not the case so the proposed tests are to
determine the behavior of the system numerically.

In AMBER if I'm not mistaken when PME is turned on there is a long range
component to VDW that is random and thus NVE is not NVE even
theoretically. (this may not be true in all versions of AMBER)

Correct me if I'm wrong or verify if I'm right.


On Sat, 2008-06-07 at 10:50 -0400, Axel Kohlmeyer wrote:
> On Sat, 7 Jun 2008, sudipta sinha wrote:
> SS> Hi All,
> SS>
> SS> Could anyone suggest me how we determine PMEGridSizeX,
> SS> PMEGridSizeY, PMEGridSizeZ in NAMD. Are these values choosen as
> SS> boxl/PMEGridSizeX, boxl/PMEGridSizeY, boxl/PMEGridSizeZ are approximately
> SS> equal to 1 or less than one and I am curious to know about the integer
> SS> factor (2, 3, 5) of grid points. Another thing I want know which combination
> please check out the literature on PME. there are bunch of papers (don't
> have the references here, but paul crozier mentioned them a while ago
> on the lammps mailing list, so you may search their archive)
> investigating the accuracy of PME and related methods. using one grid
> point per angstrom was found to be reaching a good balance between
> real space and k-space calculation at acceptable accuracy and the
> fact that the total number of grid points has to be multiples of
> 2,3,5 is a limitation of fast fourier transforms (in principle FFTW
> supports roots up to 13, but the smaller ones are more efficient).
> SS> of r-RESPA gives the best performance for parallel running. We are using
> SS> timestep=2fs, fullectrostaticfrequency=2, nonbonded=1
> r-RESPA is a completely different story. whatever parameters you choose
> it is always a compromise between accuracy & energy conservation
> and performance. _you_ have to determine what is an acceptable
> energy conservation and accuracy and based on that you can choose
> the respa parameters to so they stay within those boundaries and
> give the best performance. since this tends to be dependent on the
> kind of system you are running, you have to make tests by yourself.
> cheers,
> axel.
> SS>
> SS> Thanks
> SS> Sudipta
> SS>

*  Thomas C. Bishop   *
*Office: 504-862-3370 *
*CCS:    504-862-8391 *
*Fax:    504-862-8392 *

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:49:35 CST