Sparse Matrix Inversion on Illiac IV

摘要

The inversion of a large (of order 500 to 1000) sparse matrix is discussed. This report describes an algorithm for the inversion of irregularly sparse matrices which are stored packed and with tags for the non-zero elements. The processing of giant sparse matrices is made difficult by the fact that only non-zero elements are stored in memory. This difficulty may be overcome by using the Expand Row Routine and Pack Row Routine. The algorithms for these routines and overall flow chart are shown for a simple example. Furthermore in order to reduce operation time, the scheme is used in which the operations are eliminated for the rows having zeros in the pivot column. (Author)