When multicomponent, multistage separation problems are solved on parallel
computers by successive linearization methods, the solution of a large
sparse linear equation system becomes a computational bottleneck, since
other parts of the calculation are more easily parallelized. When the standard
problem formulation is used, this system has a block-tridiagonal form.
It is shown how this structure can be used in parallelizing the sparse
matrix computation. By reformulating the problem so that it has a bordered-block-bidiagonal
superstructure, it can be made even more amenable to parallelization. These
strategies permit the use of a two-level hierarchy of parallelism that
provides substantial improvements in computational performance on parallel
machines.
AIChE J., 40, 65-72 (1994)