简介:最近,在矩阵是积极semidefinite和入口明智的nonnegative的地方,研究人员们对学习semidefinite编程(SDP)松驰模型感兴趣,为二次地抑制的二次的编程(QCQP)。比作基本SDP松驰,这个二倍地积极的SDP模型拥有另外的O(n2)限制,它与O(n)限制为基本模型使SDP答案复杂性比那实质地高。在这份报纸,我们证明二倍地积极的SDP模型与一套有效二次的切割等价于基本的。当QCQP对称、同类时(它代表许多古典组合并且nonconvex优化问题),甚至没有任何有效切割,二倍地积极的SDP模型等价于基本SDP。在另一方面,二倍地积极的SDP模型能帮助紧缩界限直到36%,但是不再。最后,我们设法把一些以前的结果递四次的模型。
简介:TheHermitianandskew-Hermitiansplitting(HSS)methodisanunconditionallyconvergentiterationmethodforsolvinglargesparsenon-Hermitianpositivedefinitesystemoflinearequations.BymakinguseoftheHSSiterationastheinnersolverfortheNewtonmethod,weestablishaclassofNewton-HSSmethodsforsolvinglargesparsesystemsofnonlinearequationswithpositivedefiniteJacobianmatricesatthesolutionpoints.ForthisclassofinexactNewtonmethods,twotypesoflocalconvergencetheoremsareprovedunderproperconditions,andnumericalresultsaregiventoexaminetheirfeasibilityandeffectiveness.Inaddition,theadvantagesoftheNewton-HSSmethodsovertheNewton-USOR,theNewton-GMRESandtheNewton-GCGmethodsareshownthroughsolvingsystemsofnonlinearequationsarisingfromthefinitedifferencediscretizationofatwo-dimensionalconvection-diffusionequationperturbedbyanonlinearterm.ThenumericalimplementationsalsoshowthataspreconditionersfortheNewton-GMRESandtheNewton-GCGmethodstheHSSiterationoutperformstheUSORiterationinbothcomputingtimeanditerationstep.