>In amongst discussions of the finer details of integrated
>intensity vs. Rietveld, don't forget that we have used
>integrated intensities as a means to an end in a
>*structure solution* methodology. We don't see the need to fit
>the entire profile for every trial structure, when we get the
>same answer much faster using extracted intensities along with
>the covariance matrix. When we are doing refinements of high
>Tc's, GMR materials etc., we, like everyone else, do a Rietveld
>refinement - one stage is easier than two in such situations.
So, finally it is not integrated intensity vs. Rietveld, it is in fact
integrated intensity vs. raw data in *structure solution* methodology.
Even, it is not any methodology, it is *trial structure* in structure
solution methodology. Things are more clear and *not the same*...
>We've put forward an approach that for the first time shows
>how to solve structures with a high degree of internal
>flexibility by global optimisation, on a time scale
>that could be considered useful for routine applicability.
>It therefore becomes a *viable* alternative to direct methods.
Yes, why not ? What about the availability of your softwares ?
>With a recent structure solution on a molecular structure,
>we peaked at > 1000 structure refinements per second on
>a lowly SGI O2. Maybe it'll go faster if I can figure out
>the optimiser flags on the compiler. Anyone able to top that
>using a full profile ?
:-) I never solve a structure by using full profile ! Seems to
me that we are lacking some details on your compound.
You do not give the complexity of your molecular structure.
How many independent atoms ? How many x,y,z parameters ?
Also, you do not tell if 1000 trial structures are sufficient to get
THE solution.
What would be the time and number of trials for solving
a structure with LiPO3 formula where the only part of the model
known is that there are PO4 tetrahedra, 3 independent ones (36
independent x,y,z parameters), each of these tetrahedra sharing
2 oxygen atoms with the others; forming either infinite chains
or maybe cyclic entities like ...,P4O12 or P5O15 or P6O18,...
So you have a basis for building a starting model.
So I propose you officially this challenge : I send you the powder
pattern corresponding to the structure I solved two days ago. I
give you the cell and space group (a big step at which you could
have failed :-). And you will tell us if you find the result faster
than by classical direct method approach. Is that fair ?
Here are the benchmarks, on an old Pentium 90 :
Obtaining the compound : several years.
Indexing : several months because the first samples were not
well crystallized and not pure.
Extracting data by the Le Bail method with Fullprof : 789.82 sec.
20 cycles, 577 reflections extracted up to 90 2-theta degrees.
Selecting 469 reflections having no neighbouring reflection
at less than 0.01 2-theta degrees : less than one second with
the OVERLAP software (+ 2-3 minutes for preparing the data).
Preparing the SHELXS data : 2-3 minutes
Using the SHELXS direct method by the TREF option :
TREF 1000 : 4 minutes
however TREF 100 is sufficient to have the complete
solution : 3 P + 9 O atoms all in general position. Time
being 20 sec. Count on 5 or 6 time less with a Pentium II 266MHz.
Examining the proposition, looking at interatomic distances :
30 minutes for being convinced that the proposition was a
good candidate. Testing it with SHELXL : refining against
262 reflections selected because they had no neighbouring
reflection at less than 0.04 2-theta degrees, obtaining RF=30%.
Considered as convincing because PO4 were there...
Going to the Rietveld test : RB=7% after 10 cycles (~35minutes)
Going back to SHELXL, finding Li atoms...etc
I am ready to consider an exchange : send me also some of your data.
I will try to see if it was impossible to determine the structures
by conventional/standard way.
Armel
--
Armel Le Bail - Universite du Maine - Laboratoire des FluoruresCNRS UPRESA 6010 - Avenue O. Messiaen - 72085 Le Mans cedex 9 - FranceWeb : http://fluo.univ-lemans.fr:8001/