Low dose data

The background counts per pixel are printed in INIT.LP, or can be estimated by visualizing the frames with XDS-Viewer or adxv.

With the current version of XDS, the background of a data frame is used for scaling adjacent frames relative to each other in the INTEGRATE step.

This seems to work very well down to an average of 0.5 counts/pixel. If, however, the crystal is exposed so shortly/weakly that the average drops significantly below this value, scaling becomes impossible - the ratio of averages which are around zero is obviously undefined. The keyword DATA_RANGE_FIXED_SCALE_FACTOR has been introduced to handle this situation.

Another problem is that below 0.5 counts/pixel, the I/sigma values become overly optimistic. Whether this is due to the intensities being too high or the sigmas too low has not yet been investigated; likely it's the latter. This effect is not mitigated by use of DATA_RANGE_FIXED_SCALE_FACTOR .

CC1/2 (printed out in CORRECT.LP since version of March 15, 2012), on the other hand, appears to be more realistic.

It is therefore recommended that

• exposure time and flux is high enough to ensure at least 0.5 counts per pixel, on average.
• fine-slicing is performed according to the findings in the open-access paper [1] which means that the OSCILLATION_RANGE should be no less than half the REFLECTING_RANGE_E.S.D. ; choosing OSCILLATION_RANGE about the same as REFLECTING_RANGE_E.S.D. should result in data that are close to optimal.

Data frames which have been collected too finely/weakly may be added together using the merge2cbf program of the XDS package.

All of this is currently only relevant for the Pilatus detector; a typical low background on other types of detectors is more on the order of 10 counts/pixel (Holton and Fraenkel (2010) Acta Cryst. (2010). D66, 393–408; [2]).

See also: Difficult datasets