SUPERCOMPUTING STRATEGIES FOR THE DESIGN AND ANALYSIS OF COMPLEX SEPARATION SYSTEMS

The solution of large, sparse, linear equation systems is a critical c omputational step in the analysis and design of interlinked separation columns. If the modeling equations for a separation system are groupe d by equilibrium stage, the linear systems take on an almost-block-tri diagonal form. We study here the use of the frontal approach to solve these linear systems on supercomputers. The frontal approach is potent ially attractive because it exploits vector processing architectures b y treating parts of the sparse matrix as full submatrices, thereby all owing arithmetic operations to be performed with easily vectorizable f ull-matrix code. The performance of the frontal method for different m atrix orderings and different numbers of components is considered. Nin e interlinked distillation systems are used as test problems. Results indicate that the frontal approach provides substantial savings in com putation time, an order of magnitude in some cases, compared to tradit ional sparse matrix techniques.