Let me pontificate a bit on restraints and on R-free etc. The problem with powder diffraction measurements is that they often leave us information-starved. We just don't get enough data to investigate what we want to learn from the experiment, along with all the things we [think we] already know. No one needs to measure the flatness of a benzene ring in a 50-atom asymmetric unit from a powder.
Building restraints into a model in the early stages is an excellent way to ensure that model stays chemically reasonable while we tinker with getting all the experimental terms under control. It is not unusual to have to hit the project over the head with restraints to make them matter (weighting them on the order of the number of real data points or even more). In the end it is nice when the weights can be set to zero, which allows us to see a fully relaxed fit to the data, but some problems are just too complex, even with the best of data. (I have a Faujasite refinement I sometimes expound upon as an example...). The way that I would describe a situation where constraints are required is this: the paucity of data allow a wide range of models to be fit with approximately the same goodness of fit. However, many of those models are not chemically reasonable (non-flat benzene rings for example) so we use restraints to make sure that we select a model that at worst is plausible. Note that even when we don't use restraints, almost no one ever does a uniqueness test to prove that their model is the only one that can fit the data and false minima are always possible. Every time when one looks at a published structure, with or without restraints, one must think that there could be another structure that fits the data as well, if not better. An important test when restraints are used is to see what happens as their weights are made smaller. The model may become unreasonable, since there are not enough data to constrain everything, but the fit to the [real] data (Rwp, etc.) should not get significantly better. If that is not true, then the restraints are in conflict with the experiment and something is wrong. To perform R-free with powder data, one must excise multiple complete peaks from the pattern, say a few sections making up 10-20% of pattern. Thus one has even less data. This makes the information problem above even worse. Macromolecular crystallographers can afford to "throw away" data because they fit molecular fragment envelopes, mostly not individual atoms. Sorry, no free lunch here. Brian
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ Please do NOT attach files to the whole list <alan.he...@neutronoptics.com> Send commands to <lists...@ill.fr> eg: HELP as the subject with no body text The Rietveld_L list archive is on http://www.mail-archive.com/rietveld_l@ill.fr/ ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
