>If you have twinned data then your Rfree set is already compromised, because pairs of reflections related by the twin law contribute to
each other's measured intensity. The operation described is an obvious way to recover from this state.
Wouldn't the more rigorous way be to re-solve the structure with an appropriately-selected free set? Or even better, to assign by default twin-proofed free sets from the outset in scaling, as I have heard is done in Phenix? Or maybe late changes in free sets are admittedly not ideal, but don't end up mattering that much in practice? Has someone looked at this? Maybe the structure "forgets" what it had seen in the re-assigned free set after a while?
JPK
|