Hi Brian,
I can't speak to the noise level of an anisotropic grid, or why pmemd
imposes an extra factor of 2, but the 2-3-5 rule comes from the fact
that the fast Fourier transform works recursively, and the library
used by sander (at the least) can only subdivide the problem in
factors of 2, 3, or 5. If there is another prime factor in there (7,
for example), then many FFT routines will either refuse to run, or
revert to doing an O(N^2) (non-fast) FT computation instead of the O(N
log N) FFT. AMBER just refuses to start in that case.
Cheers,
MZ
On Tue, Sep 18, 2012 at 10:25 AM, Brian Radak <radak004.umn.edu> wrote:
> So I am not a PME expert and have never quite fully understood the 2-3-5
> prime factorization rule for FFT grids. I will refrain from asking such a
> detailed and broad question here. However, it would seem that AMBER (or at
> least pmemd) is a bit overzealous in enforcing this rule and can also give
> conflicting results on the matter.
>
> Case in point:
>
> I have solvated a small molecule in a 45 A cube (set, but not constructed,
> in tleap). If I set nfft1 = nfft2 = nfft3 = 45 then pmemd gives an error
> reminding about the 2-3-5 rule *and *that these values must be even. Since
> I don't really care what the FFT grid is I then just deleted those lines
> and let pmemd choose the grid. However the result in the output is not
> isotropic, that is, nfft1 = 48 while nfft2 = nfft3 = 45. So apparently my
> original choice is only okay when pmemd makes it for me and the isotropic
> nature of the system is ignored. Am I wrong in finding this weird and
> unsatisfying? Won't the anisotropic grid in principle cause artifacts in
> the dynamics and/or affect rotational invariance, or is this really not a
> serious problem compared to all of the other numerical/statistical noise?
>
> Regards,
> Brian
>
> P.S. It is probably relevant that I'm using AMBER12 with pmemd.MPI on 8
> processors and will eventually switch to sander.MPI to perform QM/MM HREMD.
> The system is also charged, so care with the PME calculations is of higher
> importance than usual.
>
> --
> ================================ Current Address =======================
> Brian Radak : BioMaPS
> Institute for Quantitative Biology
> PhD candidate - York Research Group : Rutgers, The State
> University of New Jersey
> University of Minnesota - Twin Cities : Center for Integrative
> Proteomics Room 308
> Graduate Program in Chemical Physics : 174 Frelinghuysen Road,
> Department of Chemistry : Piscataway, NJ
> 08854-8066
> radak004.umn.edu :
> radakb.biomaps.rutgers.edu
> ====================================================================
> Sorry for the multiple e-mail addresses, just use the institute appropriate
> address.
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Sep 18 2012 - 13:00:03 PDT
