Clustering model building

Figure 6.12 Proposed cluster model of smooth pea amylopectin (a) the mode of interconnection of small structural units to build up dextrin b1. Arrows trace the sub-pieces (included in boxes) of the larger dextrins (b) the fine structure of the dextrins showing how the clusters of short chains (—) are interconnected by long B-chains.261...

At low water contents (below 2 H20 per -S03 ) roughly one quarter of all OH groups of water appear to be exposed to the fluorocarbon. This corresponds to small clusters with an appreciable fraction of water molecules of type 1 and 4 in Figure 10. Simple model building leads to a cluster size of the order of 1.2 nm, assuming a spherical shape and neglecting the volumes occupied by the ions (4). This is not inconsistent with Gierke s estimate of 1.9 nm for very low water contents (26). 

Cluster analysis provides a tool for separation of very similar materials, and the success of these methods is highly dependent upon the calibration techniques used to build the model. Key elements of experimental design include the selection of samples used to teach the system and the performance of their corresponding reference tests. Proper grouping of the raw materials into appropriate cluster models is also important. To separate close materials, similar materials should be grouped together and then cross-validated against the remainder of materials in the database. 

When the partial charges included in a force field that does not include the polarizabihty are adjusted in order to reproduce the condensed phase properties, they are usually not efficient for modelling properties of the dimer or small clusters. To build accurate electrostatic models, several groups have proposed methods to obtain distributed multipoles from knowledge of the electronic wave fimction, the electron density or the electrostatic potential around a molecule [1-8]. These distributed multipole models should then be used in a force field with a polarizability model. 

---