ISSN:
1434-1948
Keywords:
Crystal packing
;
Neural networks
;
Pattern recognition
;
Organometallic compounds
;
Close packing
;
Chemistry
;
General Chemistry
Source:
Wiley InterScience Backfile Collection 1832-2000
Topics:
Chemistry and Pharmacology
Notes:
The crystal structures of hundreds of thousands of molecular compounds have now been determined with the quantitative information being stored in large databases. Chemists are generally interested in that specific part of this information which refers to individual molecules. The packing of molecules and their spatial relation to each other is not the focus of interest, even though the material properties of a molecular crystal are determined by both the packing arrangement of the molecules and their specific properties. The lack of interest in packing has at least one obvious reason. It is difficult for the non-expert to find the appropriate packing categories when looking at a principally infinite lattice. A tool performing this search for individual categories of packing in an automatic manner would therefore be of general importance. The development and application of such a tool is described in this paper. Novel strategies to extract a set of n points forming a specific n-point polyhedron from a set of p points (p 〉 n) is at its basis. Neural networks and second moment analysis are used as the methods of pattern recognition. In order to correlate packing and shape, second moment analysis of molecular shape is also used. This novel method is applied to three classes of organometallic compounds to determine whether the crystals formed by these compounds belong to any of the three classes fcc, bcc, or hcp in an idealized sense. A strong correlation is found between molecular shape and the membership or nonmembership of these classes. The corresponding program, including appropriate tools for visual representation, is available from the authors.
Additional Material:
18 Ill.
Type of Medium:
Electronic Resource
