Date of this Version
The Minimum Rectilinear Steiner Tree (MRST) problem is to find the minimal spanning tree of a set of points (also called terminals) in the plane that interconnects all the terminals and some extra points (called Steiner points) introduced by intermediate junctions, and in which edge lengths are measured in the L1 (Manhattan) metric. This is one of the oldest optimization problems in mathematics that has been extensively studied and has been proven to be NP-complete, thus efficient approximation heuristics are more applicable than exact algorithms.
In this thesis, we present a new heuristic to construct rectilinear Steiner trees (RSTs) with a close approximation of minimum length in Ο(n log n) time. To this end, we recursively divide a plane into a set of sub-planes of which optimal rectilinear Steiner trees (optRSTs) can be generated by a proposed exact algorithm called Const_optRST. By connecting all the optRSTs of the sub-planes, a sub-optimal MRST is eventually constructed.
We show experimentally that for topologies with up to 100 terminals, the heuristic is 1.06 to 3.45 times faster than RMST, which is an efficient algorithm based on Prim’s method, with accuracy improvements varying from 1.31 % to 10.21 %.