Hi, I completely agree with what Andrew writes. It's very easy: negative CORR means that the reflection profile after background subtraction has a negative correlation coefficient with the standard profile. How can this happen? Well, this situation occurs when the reflection is very weak and the background estimate in the pixels belonging to the reflections happens to be higher than the actual counts in those pixels. Thus the reflection profile does not look like a hill (as is the case for strong reflections) but looks like a valley! Rejecting such reflections is a very bad idea, and incidently we just published a paper (open access at http://journals.iucr.org/d/issues/2013/07/00/ba5192/index.html ) about what happens if you do this: usually the model gets worse. People do this because they think that better Rmerge/Rmeas/Rsomething and I/sigma really means that the data are better. But this is a severe misconception, and I had hoped that such a mistake would be captured during the reviewing process, but obviously this is not so ... If you want to investigate yourself, do the following: % grep -v \! XDS_ASCII.HKL | awk '{sig=$5;if (sig<0) sig=-sig;print $4/sig,$11}' > isigi_corr.dat % gnuplot gnuplot> plot 'isigi_corr.dat' us 1:2
This produces a plot with I/sigI on the x-axis, and CORR on the y axis. I attach such a plot. It shows that negative CORR is associated with negative intensity. hope that helps, Kay On Wed, 31 Jul 2013 03:00:46 +0100, <Rain Field> <rainfiel...@163.com> wrote: >Hi All, >Inspired by the "micro diffraction assembly" methods (see >http://www.nature.com/nature/journal/vaop/ncurrent/full/nature12357.html), I >checked one XDS_ASCII.HKL file and found many reflections has negative peak >profile correlation. After deleted them and rerun XSCALE, I/sigma is higher >and Rmeas is lower in the same high resolution shell than without deletion. >I am wondering why it's not a common practice to delete those reflections? >Thanks!
<<tmp.png>>