Properties of the information value decomposition

The information value decomposition approximates a positive-valued matrix by a sequence of reduced rank matrices. A rank K approximating matrix is closest to the original matrix in the sense of minimizing the discrimination between the original matrix and the approximation, over rank K matrices. The information value decomposition is analogous to the singular value decomposition with discrimination used for the discrepancy measure instead of squared error. Several properties of the information value decomposition correspond to properties of the singular value decomposition. These properties are discussed