求解绝对值方程组的润滑牛顿算法.pdfVIP

中文摘要 求解绝对值方程组Az+B…=6是一个ⅣP难的问题．本文提出了一个光滑 牛顿法来求解绝对值方程组．这个算法在每步迭代时解一个线性方程组．在A的 最小奇异值严格大于B的最大奇异值的条件下，算法是适定的，并且是全局收敛和 局部二次收敛的．而以前提出的算法大多是求解B=一，的情况，并且最多是线性 收敛性的． 每种情况测试了随机产生的50个例子，精度达Nlo～．200维的情况平均用时在0．1秒 以下；500维的情况平均用时在1秒以下；800维的情况平均用时在5秒以下；1000维的 情况平均用时在10秒以下． 关键词：绝对值方程组；光滑牛顿算法；全局收敛；局部二次收敛 ABSTRACT A Newtonmethodis tosolvetheNP—hardabsolutevalue smoothing proposed needstosolveone of equations：Ax+BIxI=b．Theproposedalgorithm system linear ateachiteration．Andthe iswelldefinedunderthe equations algorithm mereconditionthattheminimal valueofthematrixAis strictly singular greater thanthemaximal valueofthematrixB．Moreoverthesamecondition singular callensurethatthe is and algorithmgloballylocallyquadraticallyconvergent． Whilethe statedbeforethisthesiswere forthe case algorithms usually special whenB=-I．andwere atbest． 一 linearlyconvergent AMATLAB ofthe Newtonmethodsolvedran- implementationsmoothing 200一dimensional domlygenerated AVEs，500一dimensional and1000一dimensionalAVEs．50 AVEsforeachcase AVEs randomlygenerated solvedtoan of10～．The timeforthe200-dimensional were accuracy average was than timeforthe500一dimensionalcasewasless case less 0．1s：the average forthe waslessthan the than1s；the time 800一dimensionalcase 5s；and averag