The original cell refines well and gives an excellent Rmerge: For all data to 60° 2θ, Rmerge = 0.02. However, refinement of the structure model is difficult and gives very high residuals. Furthermore, the chemist disagrees with the structure!
Close inspection of the diffraction images reveals a number of very weak, diffuse spots, some of which are relatively close to strong reflections. The presence of a twin component would be a reasonable explanation.
The first step in this operation is to pick spots from the entire data set. Parameters should be chosen to include these weak, diffuse spots. In particular, the neighbor distance should be small enough to prevent these reflections from being rejected, and the sigma level should be reduced. CrystalClear's pixel measuring tool can be used to determine the typical peak separation.
We're only going to use a few hundred reflections for indexing; therefore we first sort the reflection list by I/σ(I). This brings stronger reflections from the entire data collection process to the top of the list so that indexing considers a large portion of the reciprocal lattice. TwinSolve also has an algorithm that permits us to remove possible Kα2 reflections. We also reduce the tolerance for a correct solution, anticipating that some reflections belong to a second component of the twin.
The triclinic subcell determined by this process is 1/3 the volume of the original. This implies Z'=0.5 instead of 1.5.
1827 spots out of 2722 were indexed in this pass. The next step is to attempt to index the remaining reflections. No constraints are placed on this indexing. It is conceivable that they arise from a completely different unit cell, though this is a rare occurrence.
The second indexing solution is identical to the first in terms of cell dimensions. The new component is added to the list. The next step is to determine the twin law, the equation that relates the two twin components.
The twin law indicates that the two components are related by a 180° rotation around the normal to the 010 axis. Unless the twinning is caused by a split crystal, interpenetrating growth, or a satellite, twin laws usually make this type of geometric sense, with rotations of 30, 60, 90, 120, 180°, etc. around normals to primary axes.
The combined spot position prediction from the two twin components should account for most of the observed diffraction spots. If an appreciable number of reflections remain unindexed, the possibility of a third twin component should be explored. In this case, all 2722 indexing reflections are accounted for by two components.
For integration, two options are available. If there is minimal possibility of peak overlap and one component dominates the other, the integration can focus on the principle component. However, to treat overlap rigorously, both components should be integrated.
After integration, scale and merge, the data are ready to be passed to the structure solution and refinement software. The standard output is a SHELX HKLF5 twin format file where the relative contribution of all twin components for each reflection is specified.
Results are summarized in CIF format. A constrained unit cell refinement can then be performed, linking the unit cell dimensions for the two indexing solutions together to produce the final experimental cell parameters.
The data from TwinSolve allow the solution and refinement of the structure (SHELX twin refinement). Note that the molecular structure hasn't changed, regardless of what the chemist wanted!
