Predicting and Evaluating Crystal Structures

In a later stage of the prediction, powder diffraction data can be compared with simulated powder patterns of proposed polymorphs for identification purposes. A model structure that is close to the experimental one will produce a similar pattern and can then be further refined by means of the Rietveld method. 

If a Monte Carlo search procedure is used, it often proves more efficient to perform several short simulations rather than one long simulation for each space group. The overlap (or the lack thereof) in the identified low energy structures from a series of short runs is a good indication of the effectiveness of the search if the second or third simulation in a series does not provide any new low energy structures, it can usually be assumed that all relevant minima have been found. If only one long simulation is performed, it is much more difficult to establish the same level of confidence. 

After the energy cutoff has been applied, a large number of predicted structures may yet remain. Their number can be further reduced for the following reasons  

The same packing structure may have been predicted in different space groups. One example is the prediction of a structure in Pi, Z = 2, as well as in P2j, Z = 1 (which, on the other hand, can be taken as an indicator of the completeness of the search). Intramolecular symmetry elements may lead to the prediction of the same structure in different space groups, with the same number of independent molecules. For example, certain crystal packings of acetic acid, which is mirror-symmetric, can be predicted in both Pljc, Z = 4, and P2,2j2,  

Z = 4 (both structures have a single independent molecule in the cell). A suitable clustering of the complete set of predicted structures should thus remove duplicate structures of this type. 

---