雍龍泉,劉三陽(yáng),拓守恒,鄧方安,高 凱
(1.西安電子科技大學(xué) 數(shù)學(xué)與統(tǒng)計(jì)學(xué)院,西安710071;2.陜西理工學(xué)院 數(shù)學(xué)與計(jì)算機(jī)科學(xué)學(xué)院,陜西 漢中723001)
線性互補(bǔ)問(wèn)題(linear complementarity problem,LCP)是一類應(yīng)用廣泛的優(yōu)化問(wèn)題,它為線性規(guī)劃和二次規(guī)劃提供了統(tǒng)一的研究框架,因此求解線性互補(bǔ)問(wèn)題的有效算法備受關(guān)注[1-2].對(duì)線性互補(bǔ)問(wèn)題的研究,目前主要集中在理論與算法兩方面,前者主要研究其解的存在性、唯一性、穩(wěn)定性和靈敏度分析,后者主要建立其有效的求解方法和相應(yīng)的收斂性分析.求解線性互補(bǔ)問(wèn)題的算法有很多,經(jīng)典的有Lemke算法;近年來(lái)(針對(duì)單調(diào)線性互補(bǔ))出現(xiàn)了一些具有多項(xiàng)式復(fù)雜性的算法,如投影法、內(nèi)點(diǎn)法、非光滑牛頓法、光滑牛頓法和迭代法等[3-7].
絕對(duì)值方程(absolute value equation,AVE)等價(jià)于一個(gè)不可微的NP難優(yōu)化問(wèn)題.目前對(duì)于絕對(duì)值方程的研究主要集中于其解的存在性和唯一性以及建立有效的算法并進(jìn)行相應(yīng)的收斂性分析[8-15].
本文研究線性互補(bǔ)問(wèn)題與絕對(duì)值方程之間的內(nèi)在關(guān)系,給出了二者的等價(jià)轉(zhuǎn)化,并進(jìn)行了證明.本文用I表示單位矩陣,‖·‖表示2范數(shù),⊥表示兩個(gè)向量正交:即
定義1 如果對(duì)?z∈?n,都有zTMz≥0,則矩陣M∈?n×n稱為半正定矩陣;如果對(duì)?z∈?n,z≠0,都有zTMz>0,則M稱為正定矩陣.這里定義的半正定矩陣與正定矩陣不限制對(duì)稱性[5,16-17].
線性互補(bǔ)問(wèn)題即求向量z∈?n,滿足Mz+q≥0,z≥0,zT(Mz+q)=0,簡(jiǎn)記為L(zhǎng)CP(M,q).
當(dāng)矩陣M是半正定矩陣時(shí),LCP(M,q)稱為單調(diào)線性互補(bǔ)問(wèn)題[16].
引理1[5]設(shè)矩陣M∈?n×n為一半正定矩陣,對(duì)于任意的q∈?n,若LCP(M,q)是可行的,則LCP(M,q)必有解,且其解集為凸集.
引理2[5]設(shè)矩陣M∈?n×n為一正定矩陣,則對(duì)于任意的q∈?n,LCP(M,q)有唯一解.
上述判定線性互補(bǔ)問(wèn)題解存在的條件僅是充分條件,而非必要條件[17].更多關(guān)于線性互補(bǔ)問(wèn)題解存在的條件可參見(jiàn)文獻(xiàn)[5].
引理3[8]絕對(duì)值方程
定理1 線性互補(bǔ)問(wèn)題Mz+q≥0,z≥0,zT(Mz+q)=0等價(jià)于如下絕對(duì)值方程:
證明:對(duì)?a,b∈?,都有
該結(jié)論對(duì)向量也成立.因此,令a=Mz+q,b=z,則有
即
定理2 若1不是矩陣M的特征值,則LCP(M,q)等價(jià)于如下絕對(duì)值方程:
這里x=(M-I)z+q.
證明:利用定理1可知
由于1不是M 的一個(gè)特征值,故(M-I)-1存在.令x=(M-I)z+q,即z=(M-I)-1(x-q),可得
定理3 若1不是矩陣M的特征值,則LCP(M,q)等價(jià)于如下絕對(duì)值方程:
證明:令
結(jié)合引理3,有
下面給出A,b的表達(dá)式.
把z=(A-I)x-b代入Mz+q=(A+I(xiàn))x-b中,有
由于1不是M的一個(gè)特征值,故(M-I)-1存在.簡(jiǎn)單計(jì)算可得
從而
又由z=(A-I)x-b得
證畢.
當(dāng)1是矩陣M的特征值時(shí),對(duì)線性互補(bǔ)問(wèn)題中的M和q乘以某個(gè)正常數(shù)λ,使得1不是矩陣λM的特征值(此時(shí)線性互補(bǔ)問(wèn)題的解不變),再應(yīng)用定理2或定理3可把線性互補(bǔ)問(wèn)題轉(zhuǎn)化為絕對(duì)值方程.
考慮LCP(M,q),其中:
M的特征值為
對(duì)于?z≠0,由于
因此矩陣M是正定矩陣,易求得該問(wèn)題的唯一解為z*=(2.5,0.5,0,2.5)T.
由于1是矩陣M的特征值,因此定理2和定理3不能直接使用.令λ=3,則λM 的特征值不為1,且LCP(λM,λq)與LCP(M,q)具有共同的最優(yōu)解z*,而LCP(λM,λq)可應(yīng)用定理2或定理3轉(zhuǎn)化為絕對(duì)值方程.
Prokopyev[21]給出了(在無(wú)任何條件下)絕對(duì)值方程轉(zhuǎn)化為線性互補(bǔ)問(wèn)題的一種方法,但該轉(zhuǎn)化增加了問(wèn)題的維數(shù);胡勝龍等[22]通過(guò)重構(gòu)技巧(在無(wú)任何條件下)把絕對(duì)值方程轉(zhuǎn)化為線性互補(bǔ)問(wèn)題,但該方法的實(shí)現(xiàn)有一定難度.
定理4(有條件轉(zhuǎn)化)若1不是A的特征值,則AVE等價(jià)于LCP(M,q):
其中:z=(A-I)x-b;M=(A+I(xiàn))(A-I)-1;q=((A+I(xiàn))(A-I)-1-I)b.
證明:由于1不是A的一個(gè)特征值,故(A-I)-1存在.結(jié)合引理3,有
令z=(A-I)x-b,即x=(A-I)-1(z+b),可得
若令
則有
即
其中:z=(A-I)x-b;M=(A+I(xiàn))(A-I)-1;q=((A+I(xiàn))(A-I)-1-I)b.
推論1 若矩陣A的奇異值(即矩陣ATA特征值的非負(fù)平方根)都大于1,則矩陣M=(A+I(xiàn))(A-I)-1是一個(gè)正定矩陣.
證明:若矩陣A的奇異值都大于1,則(A-I)-1存在(否則,?ξ≠0,使得(A-I)ξ=0,即Aξ=ξ,表明1是A的特征值,從而1也是A的奇異值,矛盾),且ATA的所有特征值都大于1,因此,對(duì)任意的z≠0都成立zT(ATA-I)z>0.而
由于(A-I)-1存在,令z=(A-I)-1v(z≠0,故v≠0),代入式(1)并整理得
即
表明矩陣M=(A+I(xiàn))(A-I)-1是一個(gè)正定矩陣.
矩陣A的奇異值都大于1?矩陣M=(A+I(xiàn))(A-I)-1是一個(gè)正定矩陣?對(duì)應(yīng)的LCP(M,q)具有唯一解?相應(yīng)的AVE具有唯一解.
綜上可見(jiàn),線性互補(bǔ)與絕對(duì)值方程具有密切的內(nèi)在聯(lián)系.基于上述對(duì)兩類問(wèn)題的轉(zhuǎn)化,在求解任意絕對(duì)值方程時(shí),都可以通過(guò)求解如下無(wú)約束優(yōu)化問(wèn)題
獲得絕對(duì)值方程的解;在求解任意線性互補(bǔ)問(wèn)題時(shí),都可以通過(guò)求解如下無(wú)約束優(yōu)化問(wèn)題
獲得線性互補(bǔ)問(wèn)題的解.該方法對(duì)矩陣A或M 可以沒(méi)有任何限制,當(dāng)目標(biāo)函數(shù)值收斂到零,則達(dá)到最優(yōu)解;當(dāng)目標(biāo)函數(shù)值收斂不到零,則原問(wèn)題可能無(wú)解;因此該方法具有較強(qiáng)的實(shí)用性.
[1]Billups S C,Murty K G.Complementarity Problems[J].Journal of Computational and Applied Mathematics,2000,124:303-318.
[2]Cottle R W,PANG Jong-shi,Stone R E.The Linear Complementarity Problems[M].San Diego:Academic Press,1992.
[3]Kojima M,Megiddo N,Noma T,et al.A Unified Approach to Interior Point Algorithms for Linear Complementary Problems:A Summary[J].Operations Research Letters,1991,10(5):247-254.
[4]Murty K G.Linear Complementarity,Linear and Nonlinear Programming[M].Berlin:Heldermann Verlag,1988.
[5]韓繼業(yè),修乃華,戚厚鐸.非線性互補(bǔ)理論與算法 [M].上海:上??茖W(xué)技術(shù)出版社,2006:31-250.(HAN Jiye,XIU Naihua,QI Houduo.Theory and Algorithm for the Nonlinear Complementarity [M].Shanghai:Shanghai Science and Technology Press,2006:31-250.)
[6]YONG Longquan.Interior Point Algorithms for Monotone Linear Complementarity Problem [J].ICIC Express Letters,2011,5(3):775-781.
[7]YONG Longquan.Research on Linear Complementarity Problem and Related Application [J].Engineering Technology Press,2010(2):287-290.
[8]Mangasarian O L,Meyer R R.Absolute Value Equations[J].Linear Algebra Appl,2006,419(2/3):359-367.
[9]Rohn J.An Algorithm for Solving the Absolute Value Equation[J].Electronic Journal of Linear Algebra,2009,18:589-599.
[10]YONG Longquan,LIU Sanyang,ZHANG Shemin,et al.A New Method for Absolute Value Equations Based on Harmony Search Algorithm [J].ICIC Express Letters,Part B:Applications,2011,2(6):1231-1236.
[11]雍龍泉,拓守恒.基于凝聚函數(shù)的擬牛頓算法求解絕對(duì)值方程 [J].系統(tǒng)科學(xué)與數(shù)學(xué),2012,32(11):1427-1436.(YONG Longquan,TUO Shouheng.Quasi-Newton Method to Absolute Value Equations Based on Aggregate Function[J].Journal of Systems Science and Mathematical Sciences,2012,32(11):1427-1436.)
[12]YONG Longquan.An Iterative Method for Absolute Value Equations Problem [J].Information,2013,16(1):7-12.
[13]雍龍泉,劉三陽(yáng),拓守恒,等.具有2n個(gè)解的絕對(duì)值方程問(wèn)題 [J].吉林大學(xué)學(xué)報(bào):理學(xué)版,2013,51(3):383-388.(YONG Longquan,LIU Sanyang,TUO Shouheng,et al.Absolute Value Equations with 2n-Solutions[J].Journal of Jilin University:Science Edition,2013,51(3):383-388.)
[14]Caccetta L,BIAO Qu,ZHOU Guanglu.A Globally and Quadratically Convergent Method for Absolute Value Equations[J].Computational Optimization and Applications,2011,48(1):45-58.
[15]雍龍泉,張社民,張建科,等.絕對(duì)值方程研究進(jìn)展 [J].陜西理工學(xué)院學(xué)報(bào):自然科學(xué)版,2012,28(1):33-38.(YONG Longquan,ZHANG Shemin,ZHANG Jianke,et al.Advance in the Study of Absolute Value Equations[J].Journal of Shaanxi University of Technology:Natural Science Edition,2012,28(1):33-38.)
[16]雍龍泉,鄧方安,陳濤.單調(diào)線性互補(bǔ)問(wèn)題的一種內(nèi)點(diǎn)算法 [J].數(shù)學(xué)雜志,2009,29(5):681-686.(YONG Longquan,DENG Fang’an,CHEN Tao.An Interior Point Method for Solving Monotone Linear Complementarity Problems[J].Journal of Math,2009,29(5):681-686.)
[17]雍龍泉,鄧方安.線性互補(bǔ)問(wèn)題中矩陣正定性判別的2點(diǎn)注記 [J].吉首大學(xué)學(xué)報(bào):自然科學(xué)版,2009,30(1):33-35.(YONG Longquan,DENG Fang’an.Two Notes on Positiveness Judgement of Matrix in Linear Complementarity Problem [J].Journal of Jishou University:Natural Science Edition,2009,30(1):33-35.)
[19]魏慶舉.絕對(duì)值方程的廣義牛頓算法及其收斂性 [D].北京:北京交通大學(xué),2009.(WEI Qingju.A Generalized Newton Method and Its Convergence for Absolute Value Equations[D].Beijing:Beijing Jiaotong University,2009.)
[20]雍龍泉.絕對(duì)值方程研究綜述 [J].陜西理工學(xué)院學(xué)報(bào):自然科學(xué)版,2013,29(6):25-30.(YONG Longquan.Review of Absolute Value Equation[J].Journal of Shaanxi University of Technology:Natural Science Edition,2013,29(6):25-30.)
[21]Prokopyev O.On Equivalent Reformulations for Absolute Value Equations[J].Computational Optimization and Applications,2009,44(3):363-372.
[22]HU Shenglong,HUANG Zhenghai.A Note on Absolute Value Equations [J].Optimization Letters,2010,4(3):417-424.