Centroid methods variables

Both methods described above belong to a class of methods that is also called partitioning or optimization or partitioning-optimization techniques. They partition the set of objects into subsets according to some optimization criterion. Both methods use representative elements, in one case an object of the set to be clustered (the centrotype), in the other an object with real values for the variables that is not necessarily (and usually not) part of the objects to be clustered (the centroid). 

The biplot of Fig. 31.9 shows that both the centroids of the compounds and of the methods coincide with the origin (the small cross in the middle of the plot). The first two latent variables account for 83 and 14% of the inertia, respectively. Three percent of the inertia is carried by higher order latent variables. In this biplot we can only make interpretations of the bipolar axes directly in terms of the original data in X. Three prominent poles appear on this biplot DMSO, methylene-dichloride and ethylalcohol. They are called poles because they are at a large distance from the origin and from one another. They are also representative for the three clusters that have been identified already on the column-standardized biplot in Fig. 31.7. 

Points with a constant Euclidean distance from a reference point (like the center) are located on a hypersphere (in two dimensions on a circle) points with a constant Mahalanobis distance to the center are located on a hyperellipsoid (in two dimensions on an ellipse) that envelops the cluster of object points (Figure 2.11). That means the Mahalanobis distance depends on the direction. Mahalanobis distances are used in classification methods, by measuring the distances of an unknown object to prototypes (centers, centroids) of object classes (Chapter 5). Problematic with the Mahalanobis distance is the need of the inverse of the covariance matrix which cannot be calculated with highly correlating variables. A similar approach without this drawback is the classification method SIMCA based on PC A (Section 5.3.1, Brereton 2006 Eriksson et al. 2006). 

Following Fey nman s original work, several authors pmsued extensions of the effective potential idea to construct variational approximations for the quantum partition function (see, e g., Refs. 7,8). The importance of the path centroid variable in quantum activated rate processes was also explored and revealed, which gave rise to path integral quantum transition state theory and even more general approaches. The Centroid Molecular Dynamics (CMD) method for quantum dynamics simulation was also formulated. In the CMD method, the position centroid evolves classically on the efiective centroid potential. Various analysis and numerical tests for realistic systems have shown that CMD captures the main quantum effects for several processes in condensed matter such as transport phenomena. 

In PRIMA there is a separate scaling for each class this kind of scaling, where each class has its scaling parameters, i.e., the mean and the standard deviation of each variable within the class, is commonly used in SIMCA, but in PRIMA it becomes a fundamental characteristic of the method. Then, under the assumption that the variables are uncorrelated, the PRIMA distance is the squared distance of an object from the class centroid. It has all the other characteristics of SIMCA. 

In the early papers [4,8], the development of the CMD method was guided in part by the effective harmonic analysis and, in part, by physical reasoning. In Paper III, however, a mathematical justification of CMD was provided. In the latter analysis, it was shown that (1) CMD always yields a mathematically well-defined approximation to the quantum Kubo-transformed position or velocity correlation function, and (2) the equilibrium path centroid variable occupies an important role in the time correlation function because of the nature of the preaveraging procedure in CMD. Critical to the analysis of CMD and its justification was the phase-space centroid density formulation of Paper III, so that the momentum could be treated as an independent dynamical variable. The relationship between the centroid correlation function and the Kubo-transformed position correlation function was found to be unique if the centroid is taken as a dynamical variable. The analysis of Paper III will now be reviewed. For notational simplicity, the equations are restricted to a two-dimensional phase space, but they can readily be generalized. 

The essence of the extended Lagrangian method [73, 74] is to choose small enough fictitious masses m in the multidimensional version of Eq. (3.79) that the fictitious variables rapidly oscillate around the minimum of the centroid free-energy surface [Eq. (3.78)]. The centroid variables should then exhibit an adiabatic, conservative motion because there is little energy exchange between them and the fictitious degrees of... 

The governing equations were discretized using a finite volume method and solved using a general-purpose computational fluid dynamic code. The computational domains are divided into a finite number of control volumes (cells). All variables are stored at the centroid of each cell. Interpolation is used to express variable values at the control volume surface in terms of the control volume center values. Stringent numerical tests were performed to ensure that the... 

---