[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [sdpd] Important McMaille update

Hi,

McMaille for indexing powder diffraction patterns needs now two
times less Monte Carlo (MC) events for obtaining the same result.
This is equivalent to having increased the speed by a factor 2.

List of latest improvements :

   -  A second criterium for selecting cells has been introduced
(the first was that Rp < Rmax) : if the expected number of peaks
is explained by a cell proposal, then that cell is "refined" also by
MC even if Rp > Rmax. 2 times more solutions are gathered in
the same time thanks to that small modification (one line of code).
This shows how decisive algorithmic development was still possible...
    - Dynamical change of Rmax occurs if too many cell proposals
are retained (use negative Rmax in the .dat file for this option). (*)
    - Pressing the K keystroke (capital letter - for Kill) will stop the
program a few seconds later (a test is made every 30000 MC events),
saving the current results. (*)
     - New logo and icon.
     - Output .prf file for viewing by WINPLOTR (pseudo pattern with
Gaussian peak shape). (*)
     - Possibility for individual peak width (use W negative and give triplets
of (2-theta, I, W), instead of (2-theta, I) doublets. (*)
     - New example files (samples 1-3 from the SDPD Round Robin 2).
     - Output of the M(20) and F(20) classic figures of merit for the "best"
refined cells.
     - Elimination of some redundant solutions (test if Volume and cell
parameters do not differ by more than 1%). (*)
     - Presentation of 2 sorted lists of "best" cells : by increasing Rp, and
by increasing volume.
     - No crash if the data file name does not exist, a second chance is 
given. (*)
     - Cell parameters and zeropoint refinement by using the routine CELREF
from Laugier and Filhol.
     - The user may decide for the number of unindexed reflections.
     - Etc

(*) = on demand from Lachlan Cranswick

These improvements lead to a more user friendly program, and to
much less time for obtaining solution(s), at least for the example
test files which produce correct answers with regularity (though
the total time depends on your chance when using a Monte Carlo
process ;-). Less than 10 minutes for small (<1000A**3) triclinic cases
is feasible (for 20 peak positions, 2 tolerated unindexed peaks, max
a, b, c = 20 A). A few seconds are needed for cubic, tetragonal,
hexagonal and rhombohedral. Time is in the range 10 seconds to
5 minutes for orthorhombic and monoclinic (2.4GHz processor).

But still no result with bad data...

http://www.cristal.org/McMaille/ or http://www.cristal.org/McMaille/

Best,

Armel



------------------------ Yahoo! Groups Sponsor ---------------------~-->
Sell a Home for Top $
http://us.click.yahoo.com/RrPZMC/jTmEAA/jd3IAA/UIYolB/TM
---------------------------------------------------------------------~->

 

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