RE: Factorization in PMEGridSize

From: JC Gumbart (gumbart_at_ks.uiuc.edu)
Date: Wed Jun 07 2006 - 09:25:24 CDT

That would be fine.

 

In my experience the size of the PME grid doesn’t contribute greatly to the
computational cost, provided it can be easily factored. So don’t worry
about a density being higher than absolutely necessary (at least 1 grid
point per A^3).

 

  _____

From: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] On Behalf
Of Edelmiro Moman
Sent: Wednesday, June 07, 2006 4:37 AM
To: namd-l_at_ks.uiuc.edu
Subject: namd-l: Factorization in PMEGridSize

 

Hi,

In the PME section of the input file of the Ubiquitin tutorial, one can
read:

Furthermore, for speed in computing Fast Fourier Transforms, PMEGridSizeX
should be chosen so that it can be factorized by 2, 3, or 5. If your
cellBasisVector1 = (60, 0, 0), a good choice for PMEGridSizeX might be 64,
since 60 Å / 64 = 0.9375 Å and $64=2^{6}$.

I am not a mathematician and I am not very familiar FFTs. I have the
following situation:

cellBasisVector1 = (86, 0, 0)
cellBasisVector2 = (0, 73, 0)
cellBasisVector3 = (0, 0, 108)

Should I make, for instance:

PMEGridSizeX = 90 (2*3^2*5)
PMEGridSizeY = 80 (2^4*5)
PMEGridSizeZ = 120 (2^3*3*5)

?

Miro


image001.gif

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:42:09 CST