5. Concluding Remarks

11] K. Levenberg. A method for the solution of certain problems in least squares. Quart. 20 of the conjugate gradient procedure and factorizing when conjugate gradient time reaches a given threshold, we should be able to keep the average number of conjugate gradient iterations below 5 and still reduce total cpu time. This paper described preliminary results of ongoing research. More computational experience is called for. Our future work will include implementation of improved data structures, a rank-1 update scheme for the factorization and specialized variants for diierent problem classes. We plan to test the procedure on other NP-complete problems, such as graph partitioning, graph coloring, the traveling salesman problem, inductive inference and linear ordering. This approach has been shown to work well for small global VLSI routing problems 13]. With the improved data structures we plan to evaluate its applicability to real-world VLSI routing.