Cassetti's New Method for the Identification of Patterns in Input–Output Matrices: An Alternative Formulation

The least upper bound on the overall proportional error that results from the simplification of an input—output matrix is a useful measure of the information loss. In particular, it is proven that some of the available results on this bound can be used to reduce the computations required for an optimal introduction of zeros. Furthermore, it is shown that the matrix that solves the simplification problem for any given level of error is generally not unique, so that it is possible to impose a priori constraints on the pattern of zeros in the matrix.