solving large instances of the quadratic cost partition problem on dense graphs by data-cor精选.pdf

Solving Large Instances of the Quadratic Cost Partition Problem on Dense Graphs by Data-Correcting Algorithms: a Computational Study Boris Goldengorin, Marius de Vink 20th August, 1999 Abstract The Data-Correcting Algorithm (DCA) corrects the data of a hard problem instance in such a way that we obtain an instance of a well solvable special case. For a given prescribed accuracy of the solution, the DCA uses a branch and bound scheme to make sure that the solution of the corrected instance satisfies this given prescribed accuracy. We describe the “hardness” of ran- domly generated instances of the Quadratic Cost Partition Problem (QCP) by the density of the corresponding graphs as well as by the cardinality of an optimal solution of the QCP. We study the behaviour of the DCAs through the number of search levels of the so called Preliminary Preservation Al- gorithm and average computational times. We report average computational times of the DCA for instances of dense graphs with the number of vertices up to 500 which can be solved on a standard PC within 10 minutes. Keywords: global maximum, submodular functions, quadratic program- ming Department of Econometrics and Operations Research, University of Groningen, P.O.Box 800, 9700 AV Groningen, The Netherlands. Corresponding author. Tel. +31-50-363 3794 fax: +31-50-363 3720. e-mail: B.Goldengorin@eco.rug.nl 1 1. Scope and purpose Given a graph with nonnegative weights on the edges and real numbers as weights of vertices. For every set of vertices, the difference between the sum of vertex weights of