Hi Kay, On Fri, Feb 22, 2008 at 10:18:26AM +0100, Kay Diederichs wrote: > >By the way, I think a part of the problem may also arise that CORRECT or > >XSCALE apply too many correction factors -- try CORRECTIONS=DECAY only. > >Gives worse R-factors and worse I/sig(I) but sometimes nicer intesity > >distributions. > > Sorry, have to disagree. Just look at e.g. MODPIX.pck and you know why > the MODULATION correction is needed. The detectors are unfortunately not > ideal.
True - if the correction leads to perfect data. I could see cases where _not_ doing the corerction can be beneficial though: the reflections in the non-ideal areas of the detector are kept in so bad shape, that the scaling/merging step remove them as outliers (keeping good symmetry-related copies of those reflections in). This might be better than having a partially corrected reflection in the dataset that fails the outlier-rejection test and then adds additional noise to e.g. a small anomalous difference. But in general I agree: these corrections in CORRECT work quite well and one should always start with the defaults (as with all other software: the authors usually have a pretty good idea and reason why a default is what it is). If the structure doesn't refine well or isn't solved fully automatically, there is always a possibility and go back to the integration and change those defaults. E.g. the default in XDS is to not refine any parameters during the INTEGRATE step: this can also be changed I guess to follow e.g. changes in distance (misaligned crystal), detect cell change due to radiation damage etc etc. Cheers Clemens -- *************************************************************** * Clemens Vonrhein, Ph.D. vonrhein AT GlobalPhasing DOT com * * Global Phasing Ltd. * Sheraton House, Castle Park * Cambridge CB3 0AX, UK *-------------------------------------------------------------- * BUSTER Development Group (http://www.globalphasing.com) ***************************************************************
