[sdpd] Re: McMaille V2.0

[Armel Le Bail <alb...@cristal.org>]

Joerg Bergmann wrote:
>Armel Le Bail wrote:
> > Even if not
> > fast enough, McMaille can test 1000 cells per second on a
> > processor running at 2.4GHz, corresponding to 3000 Le Bail
> > fits.
>
>A similar approach is done by the EFLECH/INDEX combination
>as available on
>www.bgmn.de/related_download.html
>On a 400 MHz Pentium II, I have measured 30000 cells per second.
>As I think, INDEX will reach  ~ 200000 cells per second
>on modern processors.

Thanks to that discussion, I tried to improve McMaille and could
obtain 20000 cells per second in cubic case (the most complex
triclinic is explored at 7000 cells/sec) by using a columnar peak
shape more simple than the previous Gaussian ones. I don't know
how INDEX is working exactly, but I would say that doing Monte
Carlo on the raw powder pattern (or on a pseudo powder pattern)
for indexing is yet too slow. The most complex problem solved to
date by McMaille is the triclinic example distributed inside of
the TREOR package (Test6.dat file in the McMaille package).
That small cell (V < 200 A**3) needed to make 74.000.000 tests
by the Monte Carlo approach, requiring 3 hours on a 2.4GHz.
This is still better than applying a grid search exploring cell
parameters in the 3-10 A range by steps of 0.01A and 60-120°
angles by steps of ~0.1° which would require ~ 700**6 tests,
corresponding to 530 centuries. Well, 3 hours instead of 530
centuries proves that Monte Carlo was right, but may be I had
chance... And I did not try to solve any 2000 A**3 triclinic cell.
Even at 200000 cells per second, this would be too long. One
month for the SDPD Round Robin 2 indexing step 1 is too short ;-).

Best,

Armel

PS - If this does not discourage you, versions 2 and 3 of McMaille
using Gaussian or columnar peak shapes, respectively, are still
available at http://www.cristal.org/McMaille/


------------------------ Yahoo! Groups Sponsor ---------------------~-->
Sell a Home for Top $
http://us.click.yahoo.com/RrPZMC/jTmEAA/MVfIAA/UIYolB/TM
---------------------------------------------------------------------~->

 

Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/