Hi Eckhard,
if the xray data reach real atomic resolution (say <1A), one can nicely refine cofactors w/o restraints.
or you can increase X-ray term weight (or decrease restraints weights) such that everything is relaxed or even unrestrained, depending how high is your "<1A" resolution.
In intermediate cases, the existence of multiple copies allows to use the "SAME" option in SHELX refinement to restrain identical molecules to each other, somewhat like NCS, but on bonds and angles.
Is there a possibility to do something similar with PHENIX? I have structures with up to 16 or more copies in the ASU (and several copies per chain), and 1.1A resolution. This would give me a very stable way to access unbiased information about stereochemistry.
No, there is no such option in phenix.refine. I'm not convinced that implementing it is worth the effort. Any restraints introduce bias, and weight between X-ray and restraints term is the indirect measure of that. In your case the solution that might work is that you need to lower restraints weight such that the model fits the data best while maintaining still "acceptable" geometry. That you can do in phenix.refine. Please let me know if you have any questions or need help with this! Pavel