>input restraint: C(sp3) to C(sp3) distance ~ 1.54
>output distance: 1.54 -> OK
>output distance: 1.4x -> Could it be sp2-sp2 or a wrong atom assignment?
One can make everything "OK" simply by choosing an "appropriate" weight of
restraints. The uncertainty in the restraint's weight appropriateness makes
every restrained refinement a "know how".
>So, assuming your restraints are correct for your sample
>then you are adding information into your refinement.
>This causes the rest of the results to improve as you
>reduce the noise in the output.
This is definite for the restrained part only. For the rest it depends.
>Structure refinement is not likely to work out if you can't
>fit the peak shape or have unidentified impurities or unfitted
>texture texture etc.
The restrained refinement "works" even in worse cases as is "proved" by the
modern practice.
>It does not make sense to have the atomic
> positions soaking up systematic problems.
This soaking up is, unfortunately, unavoidable both with and without
restraints. If you tread out systematic problems from the restrained part they
are likely to concentrate in the rest.
>The problem of having jammed the wrong molecule
> into a unit cell can be identified by restraints which
> are violated. With rigid bodies it is more difficult to
> find out what the data are trying to tell you. Is this
>wrong molecule issue the problem you fear?
I prefer resolving such problems by unrestrained refinement with minimal rigid
constraints on unambiguous parts which can't be refined.
>TL;DR: Garbage in garbage out. OK.
>What has that got to do with restraints?
Pouring a glass of distilled water in a garbage-can doesn't make it cleaner.
Neither my experience nor this discussion can convince me that soft restraints
are better than free refinement with minimal rigid constraints, and I'm not the
only one [A. Immirzi, Constraints and restraints in crystal structure analysis,
J. Appl. Cryst. (2009) 362].
Nevertheless, I'm not totally against these tricks if they are applied properly
as exceptions (not as a rule), and if the uncertainties of the results are
estimated realistically. The current SDPD practice, however, consists mainly in
the opposite.
*******************************************************
Leonid A. Solovyov
Institute of Chemistry and Chemical Technology
660049, K. Marx 42, Krasnoyarsk, Russia
http://sites.google.com/site/solovyovleonid
*******************************************************
________________________________
From: Jonathan WRIGHT <wri...@esrf.fr>
To: Leonid Solovyov <l_solov...@yahoo.com>
Cc: "s...@staffmail.ed.ac.uk" <s...@staffmail.ed.ac.uk>; "rietveld_l@ill.fr"
<rietveld_l@ill.fr>; "s...@yahoogroups.com" <s...@yahoogroups.com>
Sent: Wednesday, July 31, 2013 7:48 PM
Subject: Re: [sdpd] Re: Are restraints as good as observations ?
On 31/07/2013 03:07, Leonid Solovyov wrote:
> ... There is a
> widely scattered belief that restraints always improve not only the
> restrained values, but also the “unrestrained” part of structure such as
> intermolecular distances, orientations, planarity etc.
Apart from your word "always", I certainly believe restraints can and should
improve the unrestrained output from Rietveld refinements. A couple of points:
1) You should be checking whether the restraint target value is consistent with
the experimental data by looking at whether it comes back at the end of the
refinement. For example:
input restraint: C(sp3) to C(sp3) distance ~ 1.54
output distance: 1.54 -> OK
output distance: 1.4x -> Could it be sp2-sp2 or a wrong atom assignment?
An outlier in the fit to your restraints is just as much as a problem as an
outlier in the fit to your data. You should be looking at the difference/esd,
where the esd was also input.
2) If you are terribly worried what restraints are doing then you might like to
look into "Rfree" statistics. It would be cool for them to make it into some
mainstream Rietveld software (doi:10.1524/zkri.219.12.791.55857, or
doi:10.1038/nature12070).
So, assuming your restraints are correct for your sample then you are adding
information into your refinement. This causes the rest of the results to
improve as you reduce the noise in the output.
> ... The general
> practice, however, consists in applying restraints in the cases of
> highly biased reflection intensities and peak shape misfits when the
> residuals are strongly correlated. In such cases restraints compensate
> the influence of misfits on the restrained values at the expense of
> their increased influence on the “unrestrained” part.
The restraints should improve the model, if they don't, see above. Getting one
part of a model right makes it easier to find out what is missing to get the
whole thing (about) right. If there are serious misfits this means there is
some problem to be resolved. Likelihood methods can be useful at that point:
again, it would be cool for those to be more widely available.
Structure refinement is not likely to work out if you can't fit the peak shape
or have unidentified impurities or unfitted texture texture etc. The Rietveld
video game is about solving those problems to get to learn something you didn't
already know. It does not make sense to have the atomic positions soaking up
systematic problems.
The problem of having jammed the wrong molecule into a unit cell can be
identified by restraints which are violated. With rigid bodies it is more
difficult to find out what the data are trying to tell you. Is this wrong
molecule issue the problem you fear?
TL;DR: Garbage in garbage out. OK. What has that got to do with restraints?
Jon
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Please do NOT attach files to the whole list <alan.he...@neutronoptics.com>
Send commands to <lists...@ill.fr> eg: HELP as the subject with no body text
The Rietveld_L list archive is on http://www.mail-archive.com/rietveld_l@ill.fr/
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
