For large-scale problems, it may be attractive to use iterative methods
to solve the large, sparse, linear systems that arise in the equation-based
approach to process simulation. This is because, as problem sizes grow,
direct methods become extremely expensive in terms of both computation
time and storage requirements. Iterative methods, however, may not be reliable
for the indefinite and highly unsymmetric matrices encountered in a diverse
multi-unit flowsheeting problem. An extensive computational study of the
reliability of iterative linear solvers on such problems is presented here.
Nine different iterative linear solvers are considered, in connection with
incomplete factorization preconditioning. These are applied to a set of
nine test problems. Results indicate that the ordering of rows and columns
in the matrices is a crucial factor in determining the success with which
iterative methods can be employed. Results also indicate that in general
these multi-unit flowsheeting matrices are difficult to solve by iterative
methods, and that, though there is certainly potential in the use of iterative
methods, to realize this potential will require improvements in the reordering
and/or preconditioning schemes used.
Comput. Chem. Eng., 20, 1123-1132 (1996)