Re: Rutherford Lab Occult Methods for Powder Diffraction? Re:

Armel Le Bail ( armel@fluo.univ-lemans.fr )
Tue, 21 Apr 1998 22:18:05 +0200

>While not having the cited journal at hand, I am in no
>doubt that algorithms appropriately convoluted with
>genetic/Artificial Life methods have great power and
>strength. Odds are that much data of insufficient quality
>for "conventional methods" to solve, will be solvable
>by applying various innovative forms of genetic/A-life methods.

I have myself absolutely no doubt about the efficiency of the 15
or so methods proposed in the 10 last years (including GA of
course) which are able to locate a fragment of known geometry
in a cell. They are quite good methods. The point in the
previous "discussion" was that the classical methods remain
also good (Patterson, direct methods). The point was that
a method has to be choosen depending of the problem kind.

Just a remark about data of insufficient quality. It is clear
that any method that will locate a fragment or a full molecule
will not need of a lot of data for that. Two very recent examples
using synchrotron data are quite exemplary :

Formula S.G. ia Nxyz m method Nhkl constraints

C10H16O Cmcm 27 81 1 Grid-search 87 full restraint
Fluorescein Pn21a 25 75 1 MC 138 99

ia = number of independent atoms
xyz = number of refined atomic coordinates
m = number of independent molecules to be located
MC=Monte Carlo

So, thanks to these methods for locating the molecule, and now ?
Let me recall that if you try to publish in Acta Cryst. a structure
with a ratio Nhkl/Nxyz < 10 you may have some problem. The two
above cases show Nhkl/Nxyz < 2. The consequence is that
refining the structure was quasi impossible. Restraining the
model at the point it was done allows me to conclude that very
little true information has come from these studies. Modifying
some torsion angles in such cases would have almost no effect
on the R factors...

I hope you have found this "discussion" of some interest. New
is beautiful, so that such studies are now easily published,
even in prestigious periodicals. But I think that crystallographers
will never place results with Nhkl/Nxyz < 2 to the highest
accuracy level ;-).

All the best

Armel Le Bail - Laboratoire des Fluorures, CNRS ESA 6010
Universite du Maine, Av. O. Messiaen, 72085 Le Mans cedex 9, France
http://fluo.univ-lemans.fr:8001/
http://www.cybercable.tm.fr/~cristal/