







摘 要:通過(guò)引入互補(bǔ)函數(shù)將張量絕對(duì)值問(wèn)題重新表述為張量互補(bǔ)問(wèn)題.針對(duì)重構(gòu)的張量互補(bǔ)問(wèn)題,建立了自適應(yīng)非精確LM算法,并證明了算法的收斂性.數(shù)值實(shí)驗(yàn)結(jié)果表明所提出的算法是有效的.
關(guān)鍵詞:張量絕對(duì)值方程;非精確LM方法;收斂性分析;數(shù)值實(shí)驗(yàn)
中圖分類號(hào):O241文獻(xiàn)標(biāo)志碼:A
這個(gè)例子主要用于觀察算法1的迭代過(guò)程.隨機(jī)生成對(duì)稱張量B∈S[4,4]和隨機(jī)向量x*∈R4且張量B和向量x*元素取值均在[0,1].為了確保方程只有唯一的解,計(jì)算b=Ax*m-1-|x*|[m-1].這里取a=3,從而計(jì)算出c=4.899 8.為了檢驗(yàn)本文算法1的有效性,將其與文獻(xiàn)[9]中算法3.1進(jìn)行對(duì)比發(fā)現(xiàn),要達(dá)到相同的精度,本文的算法1在迭代次數(shù)上并不具有優(yōu)勢(shì),但CPU時(shí)間是占優(yōu)勢(shì)的,參見(jiàn)表3中的數(shù)值結(jié)果.分析其原因,文獻(xiàn)[9]中算法3.1是精確LM算法,每一迭代步需要精確求解LM方程,這是比較耗費(fèi)時(shí)間的.另外從表4可以看到‖H(xk)‖隨著迭代次數(shù)k的增加會(huì)快速趨于0.此外,‖Ψ(xk)‖隨著迭代次數(shù)k的增加亦會(huì)快速趨于0.這說(shuō)明了算法具有良好的收斂性.
此外,選擇不同的b值來(lái)測(cè)試算法的收斂結(jié)果.實(shí)驗(yàn)中式(9)中的參數(shù)a取維15,數(shù)值結(jié)果見(jiàn)表5.其中xs表示方程的解,Iter表示迭代次數(shù).
3 結(jié) 論
本文提出了一個(gè)求解連續(xù)張量絕對(duì)值方程的非精確自適應(yīng)LM算法.分析了算法的全局收斂性和局部二次收斂性.并通過(guò)數(shù)值實(shí)例來(lái)驗(yàn)證所做的理論分析及有效性和可行性.數(shù)值結(jié)果表明,本文所提出的算法是有效的.
參 考 文 獻(xiàn)
[1] "MANGASARIAN O" L,MEYER R R.Absolute value equations[J].Linear Algebra Appl,2006,419:359-367.
[2]PROKOPYEV O.On equivalent reformulations for absolute value equations[J].Comput" Optim Appl,2009,44:363-372.
[3]NOOR M" A,IQBAL J,AL-SAID E.Residual iterative method for solving absolute value equations[J].Abstr" Appl" Anal,2012,21:121-129.
[4]NOOR M" A,IQBAL J,KHATTRI S,et al.A new iterative method for solving absolute value equations[J].Int" J Phys" Sci,2011,6:1793-1797.
[5]NOOR M" A,IQBAL J,NOOR K" I,et al.On an iterative method for solving absolute value equations[J].Optim" Lett,2012,6:1027-1033.
[6]DING W,WEI Y.Solving multilinear systems with M-tensors[J].J" Sci" Comput,2016,68:689-715.
[7]WANG X,CHE M,WEI Y.Neural networks based approach solving multi-linear systems with M-tensors[J].Neurocomputing,2019,351:33-42.
[8]LI X T,NG M K.Solving sparse non-negative tensor equations,algorithms and applications[J].Front" Math" China,2015,10:649-680.
[9]LV C" Q,MA C" F.A Levenberg-Marquardt method for solving semi-symmetric tensor equations [J].J Comput Appl Math,2018,332:13-25.
[10]DU S,ZHANG L,CHEN C,et al.Tensor absolute value equations[J].Sci" China Math,2018,61:1695-1710.
[11]SONG Y,QI L.Tensor complementarity problem and semi-positive tensors[J].J Optim Theory Appl,2016,169:1069-1078.
[12]HUANG Z,QI L.Formulating an n-person noncooperative game as a tensor complementarity problem[J].Comput Optim Appl,2017,66:557-576.
[13]LUO Z Y,QI L,XIU N H.The sparse solutions to Z-tensor complementarity problems[J].Optim Lett,2017,11:471-482.
[14]SONG Y,YU G.Properties of solution set of tensor complementarity problem[J].J Optim Theory Appl,2016,170:85-96.
[15]CHEN H,CHEN Y,LI G,et al.A semidefinite program approach for computing the maximum eigenvalue of a class of structured tensors and its applications in hypergraphs and copositivity test[J].Numer Linear Algebra Appl,2018,25(1):e2125.
[16]PROKOPYEV O.On equivalent reformulations for absolute value equations[J].Comput" Optim" Appl,2009,44:363-372.
[17]CLARKE F H.Optimization and Nonsmooth Analysis[M].New York:Wiley,1983.
[18]DAN H,YAMASHITA N,F(xiàn)UKUSHIMA M.Convergence properties of the" inexact Levenberg-Marquardt type under local error bound conditions[J].Optim Meth Soft,2001,17:605-626.
The inexact LM method for tensor absolute value equation
Ma Changfeng, Xie Yajun
(School of Big Data; Key Laboratory of Data Science and Intelligent Computing,
Fuzhou University of International Studies and Trade, Fuzhou 350202," China)
Abstract: In this paper," the tensor absolute value equations is reformulated into tensor complementary" problem by introducing complementary function. For the reformulated tensor complementary problem, an adaptive inexact Levenberg-Marquardt (LM) algorithm is proposed. And convergence theorem of the algorithm is proved. Numerical experiments show that the proposed algorithm is effective.
Keywords: tensor absolute value equation; inexact LM algorithm; convergence analysis; numerical experiment
[責(zé)任編校 陳留院 趙曉華] .