[phenixbb] unit cell discrepancies

Dale Tronrud det102 at uoxray.uoregon.edu
Thu Apr 21 12:59:00 PDT 2011

On 04/21/11 12:26, Jacob Keller wrote:
>>  The location of the spots on the images give a wealth of information
>> about the cell constants leading to very precise values, if you know
>> your wave length and crystal to detector distance.  Intensities, not so much.
> I think that there is a fair amount of imprecision here. For example,
> what is the usual precision of the beamline-defined crystal-detector
> distance, and of the wavelength? Even 1mm/100mm total distance could
> have its effect on cell parameters, and I don't think refinement of
> the distance based on trigonometric distortion has a very sharp
> minimum, although please correct me if I am wrong (how is the detector
> distance calibrated, anyway?). Even having an oblong crystal might
> change the distance by ~0.5mm or so. But this is theoretical, and I
> think you have much more empirical, real knowledge--what are the
> levels of precision on these fronts, and how precisely do we know cell
> parameters from experiment (I thought it was pretty imprecise)?
   My understanding of data collection is that the least accurate contributors
all lead to scaling errors.  This means that the cell angles and ratios of the
cell lengths are pretty precise.  We are left with an uncertainty in the
isotropic scale of the crystal.  When I've done detailed comparisons of models
from two different crystals (even isomorphous ones) I've superimposed them by
a rotation, translation, AND scale to minimize systematic errors.  It's not
perfect because the bond length restraints ensure that the distortions are not
distributed uniformly, but it does reduce the background level of differences.

   You are worried about the 1% error in cell scale that might result from
a measurement error in the data collection set up.  I don't see that refining
the cell constant along with the other parameters is going to give anything
better.  The worst case I've see is a model that was refined with A and B set
10% too large (The PDB entry has been obsoleted but I can dig it up if you
want.)  When I refined the model with the correct cell the R factor dropped
by about 3%.  Measurable, but 10% is huge and 3%, when in the mid-twenties, is
not.  What resolution you ask?  It depends on which cell constants you believe!
Let's just say it was around 2.5 A.  This problem was uncovered by a survey
of the PDB using a packing analysis check.  It stood out as one of the loosest
packed models.

Dale Tronrud

> Jacob
> _______________________________________________
> phenixbb mailing list
> phenixbb at phenix-online.org
> http://phenix-online.org/mailman/listinfo/phenixbb

More information about the phenixbb mailing list