Calculation of condition numbers of sparse matrices on the nCUBE 2

In this note we discuss the development and implementation of an efficient and highly parallelizable algorithm for the calculation of some of the most used condition numbers for large sparse matrices. In the process the inverse matrix could be also evaluated. A number of numerical experiments are carried out on an nCUBE 2 parallel computer. Symmetric and nonsymmetric matrices are used with dimensions ranging from 500 to 3000. The maximum speed up obtained in the tests is approximately 113 when 128 processors are employed. This result shows the high degree of parallelization that can be achieved by the proposed algorithm.