The data structure to store the matrix for numerical factorization is a simple one. The matrix structure is after the mmd sorting algorithm and the symbolic factorization mainly composed of clustered non-zero matrix coefficients which form blocks. In this case we use a data strucutre to represent these row and column blocks. NOTE: The row blocks are simply what in the literature is specified as supernodes. We have kept the data structure as simple as possible to minimize the numerical operations and so the execution time. NOTE: When we define a system we have the possibility to define a multiplicity factor. This allows us to perform all initialization tasks ( input of incidences, symbolic factorization, block format calculation ) with all indices modulo the multiplicity factor to save time and memory. Thus the computed block format is always the same for any specified multiplicity factor only the block bounds are different. ATTENTION: The size of the permutation vector is only Dim/(Multiplicity Factor).
1.9.4