一个非凸二次规模基于投影SDP松驰的分支切割算法.pdfVIP

℄SDP - NP-hard KKT ℄   KKT ℄℄ -  ℄ : , , Æ, , i SDP - A BRANCH-AND-CUT ALGORITHM USING PROJECTION SDP RELAXATION FOR NON-CONVEX QUADRATIC PROGRAMMING ABSTRACT It is well-known that solving a non-convex quadratic programming for a global minimum is an NP-hard problem. A non-convex quadratic pro- gramming problem(NCQP)may have many local solutions which are quite different from each other in the most cases. The branch-and-bound tech- nique is the most commonly used method for a global solution. The KKT- branching method is the best branching method. A common bounding way of an NCQP is to lift the problem into a higher dimensional space by introducing extra variables, which can result in their sizes being sig- nificant increased. But low-dimensional relaxations using projection tech- niques can make up for the drawback. Since the good performance of the KKT-branching and low-dimensional relaxations in practice, we present a Branch-and-cut algorithm for box-constrained non-convex quadratic pro- grams and apply the algorithm to solve some randomly generated problems originated from the literature. The numerical results demonstrate that our algorithm is effective. KEY WORDS: non-convex quadratic programming, Semidefinite pro- gramming, projection techniques, Branch-and-bound, global algorithm ii