[phenixbb] on the convergence of the Phenix refine

[Pavel Afonine]

Refinement involves many steps some of them do not guarantee monotonic 
decrease in R-factor. So what you observe is possible and in fact 
expectable just by the construction of the procedure.

If you would like to learn more

http://journals.iucr.org/d/issues/2012/04/00/ba5180/ba5180.pdf

and follow references therein.

So if you don't observe R-factor jumping up by more than 2-3% then I 
would say you have noting to worry about.

Pavel

On 4/21/15 7:20 PM, Smith Liu wrote:
> Dear All,
> Suppose I have a refine of 10 cycles, it get stablized from cycle 7. 
> But sometimes I noticed that from the end of the cycle 10 to the end 
> of the whole refinement, the stablization from 7 to cycle 10 can be 
> destroyed a little, which makes me concern whether the final 
> refinement can be regarded as convergent or not.
> Will you please explain why the stablization from 7 to cycle 10 can be 
> destroyed a little from the end of cycle 10 to the end of the whole 
> refinement?
> More critically, in the other situation, suppose I have a refine of 10 
> cycles, it get relatively stablized from cycle 5, from the end of 
> cycle 5 to the end of cycle 10 the Rfree increase very slightly each 
> cycle of the refine, which makes at the end of cycle 10 the Rfree is 
> slightly higher than the Rfree at the start of the whole refinement. 
> But from the end of cylce 10 to the end of the whole refinement, there 
> is a slight decrease of Rfree, which makes the Rfree at the end of the 
> whole refinement slightly lower than Rfree at the start of the whole 
> refinement.
> It seems to increase the cycles of the refinement would not help further.
> Will you please explain the above phenomenon and to see whether the 
> refine result can be acceptable, or do you have any suggestions for 
> improvement?
> Smith

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20150421/33f5e01e/attachment-0001.htm>