Hi Matt,
Thanks for your reply. Following Bruce's instructions, I have
re-compiled Ifeffit with a larger "max_restraint" number on a Linux
Machine. It works fine. Now, I can compare fits obtained with a set of
multiple restraints (more elegant) and with a combined (limited to 10)
set of restraints. I did combine restraints before in the way almost as
you suggested. Assuming that each restraint is squared before adding to
chi-square, I used sum as follows
restraint_sum = sqrt((restraint1)^2 + (restraint2)^2 +
(restraint3)^2...)
for group of similar atoms and bond lengths. I guess, the difference in
sums should not have a significant effect on fit. However, the
difference in fits with multiple and combined restraints is surprisingly
noticeable. I expected that the mathematical distinction should be
pretty small as you said. But it is not at least in this test. I should
explore it a bit more. The restraints combined were for similar sets of
both lengths in independent molecular units and should not be strongly
anti-correlated. I wonder if when stereochemical restraint relations are
used and a large number of independent not equally important (close to
zero elements) restraints imposed, convergence and stability of
minimization algorithm are affected by such a sparse normal-equations
matrix.
Regarding the examples with possible more than 10 restraints, I have a
data for a solution of Cu(II)complex with two almost geometrically
identical organic molecules. The crystal structure and total 28 bond
distances with uncertainties are available in the literature. This
amount can be reduced to 14 assuming 2-fold symmetry between molecules.
I wanted to apply in XAFS fit bit more rigorous independent restraints
using available from crystal structure bond distances with
uncertainties. In addition, for proper sterochemical restraints (e.g. in
"rigid-body" refinement) some bond angles and planar groups restraints
may be required, which may be important for multiple scattering paths. I
guess, to have a bit more flexibility in the amount of restraints (more
than 10) can be useful when fitting data for bio-molecular systems.
Victor
Dr Victor Streltsov
CSIRO Molecular & Health Technologies
343 Royal Parade
Parkville VIC 3052 Australia
Phone: +61 3 96627311
Fax: +61 3 96627101
victor.streltsov(a)csiro.au
>Hi Victor,
>As Bruce said, the limit of 10 restraints is compiled-in. It could
>certainly be increased.
>But I'm impressed that you have more than 10 independent restraints to
>impose! Depending on the nature of your restraints, it might be
>possible to combine some of them. That is, using one restraint of
> restraint_sum = (restraint1) + (restraint2) + (restraint3)
>(where each restraint might be an elaborate calculation) is only
>different from using three restraints if the three restraints are
>strongly anti-correlated, so that they compete against one another.
>If you're restraining different bond lengths, some of them probably
>can be combined.
>It's definitely better, and more elegant, to use multiple restraints,
>but I think the mathematical distinction may be pretty small.
>--Matt
