Sparse matrix methods for chemical process separation calculations on supercomputers

The author considers using the frontal method on supercomputers to solve the large, sparse linear equation systems arising in process separation calculations. The motivation is that the frontal method takes advantage of vector computers by treating parts of the sparse matrix as full submatrices. This allows arithmetic operations to be performed with full-matrix code and circumvents the difficulties inherent in indirect addressing on vector processors. Separation problems from the commercial simulators ASPEN PLUS and SPEEDUP are used as test cases. Results on a CRAY Y-MP supercomputer show that the frontal method significantly reduces simulation time, by more than an order of magnitude in many cases, compared to traditional sparse matrix methods