Message Board

Dear readers, authors and reviewers,you can add a message on this page. We will reply to you as soon as possible!

2020 Volume 45 Issue 10
Article Contents

Jing-pin HUANG, Hao XIONG, Shan-shan ZHANG. Hermite Positive Definite Solution of Nonlinear Complex System Xm-BX-C = 0[J]. Journal of Southwest China Normal University(Natural Science Edition), 2020, 45(10): 35-41. doi: 10.13718/j.cnki.xsxb.2020.10.007
Citation: Jing-pin HUANG, Hao XIONG, Shan-shan ZHANG. Hermite Positive Definite Solution of Nonlinear Complex System Xm-BX-C = 0[J]. Journal of Southwest China Normal University(Natural Science Edition), 2020, 45(10): 35-41. doi: 10.13718/j.cnki.xsxb.2020.10.007

Hermite Positive Definite Solution of Nonlinear Complex System Xm-BX-C = 0

More Information
  • Received Date: 08/03/2020
    Available Online: 20/10/2020
  • MSC: O241.6

  • Nonlinear matrix equations are widely used in fields of control theory, dynamic programming, ladder networks, and stochastic filtering. The Hermite positive definite solution of the m-th asymmetric complex system Xm-BX-C = 0 (m≥2) under the certain conditions has been discussed. The necessary and sufficient conditions for the existence of positive definite solutions of this system have been given, and the roots of two algebraic equations been used on the based on minimax eigenvalues of the coefficient matrices and Brouwer fixed point theorem, the existence interval of positive definite solution is obtained. In order to use the iterative method to find the positive definite solution of the system, a nonlinear system has been given with the same solution and symmetric structure. Then, the three iterative schemes corresponding to B are positive definite, negative definite and indeterminate are constructed respectively, and by the properties of the relative algebraic equations, the convergence of these iterations are proved. At the same time, according to the features of each iteration, the selection method of iteration initial matrix is given. Finally, a simulation example is given to illustrate the validity and feasibility of the method.
  • 加载中
  • [1] HIGHAM N J, KIM H M. Numerical Analysis of a Quadratic Matrix Equation[J]. IMA J Numer Anal, 2000, 20(4): 499-519. doi: 10.1093/imanum/20.4.499

    CrossRef Google Scholar

    [2] BAI Z Z, GAO Y H. Modified Bernoulli Iteration Methods for Quadratic Matrix Equation[J]. J Comput Math, 2007, 25(5): 498-511.

    Google Scholar

    [3] LONG J H, HU X Y, ZHANG L. Improved Newton's Method with Exact Line Searches to Solve Quadratic Matrix Equation[J]. J Comput Appl Math, 2008, 222(2): 645-654. doi: 10.1016/j.cam.2007.12.018

    CrossRef Google Scholar

    [4] KIM Y J, KIM H M. Diagonal Update Method for a Quadratic Matrix Equation[J]. Appl Math Comput, 2016, 283: 208-215. doi: 10.1016/j.amc.2016.02.016

    CrossRef Google Scholar

    [5] LIU L D. Perturbation Analysis of a Quadratic Matrix Equation Associated with an M-Matrix[J]. J Comput Appl Math, 2014, 260: 410-419. doi: 10.1016/j.cam.2013.10.011

    CrossRef Google Scholar

    [6] 蓝家新, 黄敬频, 王敏, 等.四元数矩阵方程AXB+CXD=EM自共轭解[J].西南师范大学学报(自然科学版), 2019, 44(8): 1-6.

    Google Scholar

    [7] 苏仰锋, 张开军, 刘明李.二次矩阵方程最大最小解存在的充分条件[J].复旦学报(自然科学版), 2004, 43(3): 303-306.

    Google Scholar

    [8] 尤俊丽, 廖安平, 段雪峰.二次矩阵方程X2-bX-C=0的正定解[J].大学数学, 2012, 28(3): 101-103.

    Google Scholar

    [9] LU L Z, AHMED Z, GUAN J R. Numerical Methods for a Quadratic Matrix Equation with a Nonsingular M-Matrix[J]. Appl Math Lett, 2016, 52: 46-52. doi: 10.1016/j.aml.2015.08.006

    CrossRef Google Scholar

    [10] 周磊, 肖小庆, 孙婷婷.一类非线性系统状态的网络化估计方法[J].西南大学学报(自然科学版), 2018, 40(1): 149-156.

    Google Scholar

    [11] 姚合军.一类时延网络控制系统的有限时间镇定[J].西南大学学报(自然科学版), 2018, 40(7): 153-158.

    Google Scholar

    [12] VORONTSOV Y O, IKRAMOV K D. Numerical Algorithm for Solving Quadratic Matrix Equations of a Certain Class[J]. Comput Math Math Phy, 2014, 54(11): 1643-1646. doi: 10.1134/S0965542514110128

    CrossRef Google Scholar

    [13] ADAM M S I, DING J, HUANG Q L, et al. Solving a Class of Quadratic Matrix Equations[J]. Appl Math Lett, 2018, 82: 58-63. doi: 10.1016/j.aml.2018.02.017

    CrossRef Google Scholar

    [14] CHEN C R, LI R C, MA C F. Highly Accurate Doubling Algorithm for Quadratic Matrix Equation from Quasi-Birth-and-Death Process[J]. Linear Algebra Appl, 2019, 583: 1-45. doi: 10.1016/j.laa.2019.08.018

    CrossRef Google Scholar

    [15] 张千宏, 徐昌进.一个非线性差分方程组解的表现[J].西南师范大学学报(自然科学版), 2010, 35(4): 55-58.

    Google Scholar

    [16] 崔晓梅, 谭丽辉, 赵世佳.矩阵方程X-A*X-1A+B*X-2B=I正定解存在的充分条件[J].数学学报(中文版), 2014, 57(5): 973-980.

    Google Scholar

  • 加载中
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Tables(1)

Article Metrics

Article views(639) PDF downloads(115) Cited by(0)

Access History

Other Articles By Authors

Hermite Positive Definite Solution of Nonlinear Complex System Xm-BX-C = 0

Abstract: Nonlinear matrix equations are widely used in fields of control theory, dynamic programming, ladder networks, and stochastic filtering. The Hermite positive definite solution of the m-th asymmetric complex system Xm-BX-C = 0 (m≥2) under the certain conditions has been discussed. The necessary and sufficient conditions for the existence of positive definite solutions of this system have been given, and the roots of two algebraic equations been used on the based on minimax eigenvalues of the coefficient matrices and Brouwer fixed point theorem, the existence interval of positive definite solution is obtained. In order to use the iterative method to find the positive definite solution of the system, a nonlinear system has been given with the same solution and symmetric structure. Then, the three iterative schemes corresponding to B are positive definite, negative definite and indeterminate are constructed respectively, and by the properties of the relative algebraic equations, the convergence of these iterations are proved. At the same time, according to the features of each iteration, the selection method of iteration initial matrix is given. Finally, a simulation example is given to illustrate the validity and feasibility of the method.

  • 本文研究非线性复系统

    的Hermite正定解,其中m≥2是正整数,BCCn×n是Hermite矩阵且C正定.当m=2时,方程来源于Markov链的噪声Wiener-Hopf问题,在控制理论、动态规划、随机过滤等领域有广泛应用,许多学者对其求解方法和解的特征进行过研究[1-5].文献[6]讨论了一类线性系统的M自共轭解;文献[7]给出了二次系统AX 2+ BX + C = 0的最大解X 1和最小解X 2(即X 1特征值的模均大于X 2特征值的模)存在的充分条件;文献[8]在m=2,B =b>0且C对称正定时讨论了方程(1)的正定解;文献[9]在m=2,B是对角阵且C是非奇异M矩阵时讨论了一类非对称代数黎卡提方程(1)解的性质;文献[10-11]分别研究了一类Lipschitz非线性连续时滞系统基于非均匀采样测量的网络化状态估计问题,及具有状态时延的网络控制系统的有限时间镇定问题;文献[12-14]分别研究了二次系统X T DX + AX + XT B + C = 0AXA = XAXA 0+ A 1 X + A 2 X 2= X的一般解;文献[15]采用线性矩阵方程迭代法得到了一个二元非线性差分系统正平衡点的稳定性;文献[16]给出了对称非线性系统X - A * X -1 A + B * X -2 B = I存在Hermite正定解的充分条件及正定解范围.然而,目前对于高阶非对称系统的结构解研究甚少,在此本文讨论m次复系统(1)的Hermite正定解的存在性及其求解算法.

    λmax(A),λmin(A)分别表示Hermite矩阵A的最大和最小特征值.对于正定矩阵AB,用A > B表示A - B是正定的. A *表示复矩阵A的共轭转置.

1.   正定解的存在性及算法
  • 引理1  非线性复系统(1)存在Hermite正定解的充要条件是BC = CB.

      如果方程(1)存在Hermite正定解X,则Xm- C = BX,两边取共轭转置可得

    于是

    反之,如果BC = CB,则BC可同时酉对角化,即存在酉矩阵UUn×n,使得

    其中

    Y = U * XU,则方程(1)等价于

    Y =diag(y1y2,…,yn),则方程(3)有正解Y > 0等价于下列n个代数方程均有正实根:

    f(y)=ym-by-c(c>0),则由f(0)=-c < 0,f(y)→+∞(y→+∞),以及f′(y)= mym-1-b可知f(y)在(0,+∞)至多存在一个驻点,因此∀b$\mathbb{R}$f(y)=0总存在唯一正根$\tilde{y}$.可见代数方程(4)均有唯一正解,从而方程(1)存在Hermite正定解.证毕.

    下面针对矩阵B分别为正定、负定、不定3种情况,讨论矩阵方程(1)的正定解的迭代求解方法,主要论述不同情况下迭代格式的构造及收敛性问题.

  • 引理2  设BCCn×n都是正定矩阵,b1=λmax(B),bn=λmin(B),h1=λmax(C),hn=λmin(C),则方程xm-b1x-h1=0与xm-bnx-hn=0均存在唯一正根αβ,且βα.

      正根αβ的存在唯一性由引理1已证,下面证明βα.

    xm-b1x-h1=0得$x=\sqrt[m]{{{h}_{1}}+{{b}_{1}}x}$,构造数列

    则{uk}严格递增且ukα.因此{uk}收敛,且由(5)式得$\mathop {\lim }\limits_{k \to + \infty } {u_k} = \alpha $.同理,构造数列

    则{vk}严格递增且vkβ.因此{vk}收敛,且由(6)式得$\mathop {\lim }\limits_{k \to + \infty } {v_k} = \beta $.另一方面,由(5)式和(6)式可知vkuk,根据极限保号性立即可得βα.证毕.

    现考虑矩阵方程

    其中BCCn×n是正定矩阵且BC = CB.

    引理3  方程(1)与(7)具有相同的且与B可交换的正定解.

      若X是方程(1)的正定解,则由引理1知BX = XB,因此BX能同时酉对角化,即存在酉矩阵VUn×n,使得B = VDBV*X = VDXV*,于是

    X也是方程(7)的与B可交换的正定解.反之,若X是方程(7)的且与B可交换的正定解,则由${\mathit{\boldsymbol{B}}^{\frac{1}{2}}}\mathit{\boldsymbol{X}}{\mathit{\boldsymbol{B}}^{\frac{1}{2}}} = \mathit{\boldsymbol{BX}}$可知X也是方程(1)的正定解.证毕.

    引理4  设αβ是引理2给出的正数,则矩阵方程(7)在Ω=[βIαI]内必存在正定解.

    证  建立矩阵函数

    则由b1=λmax(B),bn=λmin(B),h1=λmax(C),hn=λmin(C),∀XΩ,有

    因此F(Ω)$ \subset $Ω.根据Brouwer不动点定理,X =F(X)在[βIαI]内存在不动点$\mathit{\boldsymbol{\tilde X}}$,即方程(7)在Ω内必存在正定解.证毕.

    利用函数(8)构造迭代格式

    定理1  设m≥2是正整数,BCCn×n是正定矩阵,且BC = CBαβ是引理2给出的两个正数,则∀X 0=sIΩβsα,迭代式(9)总收敛到方程(1)的正定解.

      由引理1的证明过程可知,BC具有分解式(2),令Y = U * XU,代入方程(7)可得相应的迭代格式

    于是(9)式收敛等价于(10)式收敛.当X 0=sI (βsα)时,Y 0=sI,这时迭代式(10)等价于

    根据引理2的证明过程可知,数列{yk(i)}(i=1,2,…,n)均收敛到一个正实数,从而(9)式收敛到方程(1)的正定解.证毕.

    根据定理1的证明过程可得方程(1)的Hermite正定解表达式.

    推论1  设m≥2是正整数,BCCn×n是正定矩阵,且BC = CB,则矩阵

    是方程(1)的正定解,其中U由分解式(2)所给出,pi(i=1,2,…,n)是数列(11)的收敛点.

      因为pi是数列(11)的收敛点,所以pim-bipi-ci=0(i=1,2,…,n).把矩阵(12)代入方程(1)左边,得

    $\mathit{\boldsymbol{\tilde X}}$是方程(1)的正定解.

    注1  根据前面的讨论可知,由定理1或推论1求出的正定解$\mathit{\boldsymbol{\tilde X}}$满足矩阵不等式βI$\mathit{\boldsymbol{\tilde X}}$αI,其中αβ是引理2给出的正数.

    注2  当矩阵阶数n较大时,对矩阵BC进行同时酉对角化较为困难,因此不便利用推论1中的公式(12)来计算方程(1)的正定解.此时采用迭代公式(9)计算,只需估计αβ的近似值就可选取初始矩阵X 0=αIX 0=βI,大大降低计算难度.实际上根据引理2知

    因此,可选取X 0=($\sqrt[m]{{{b_1}\;\sqrt[m]{{{h_1}}} + {h_1}}}$) I作为迭代式(9)的初始矩阵.

  • 首先指出,当B是负定矩阵,C是正定矩阵时,引理2的结论仍然成立.事实上,当B < 0C > 0时,有

    g(x)=xm+dx-h(dh>0),则由g(0)=-h < 0,$\mathop {\lim }\limits_{x \to + \infty } g\left( x \right) = + \infty $,以及∀x∈(0,+∞),g′(x)=mxm-1+d>0,可知方程xm-b1x-h1=0与xm-bnx-hn=0分别存在唯一正根μγ.因此有μm=b1μ+h1γm=bnγ+hn.倘若γ>μ>0,则γm>μm>0,所以有

    另一方面,(h1-hn)≥0,bn(μ-γ)≥0⇒(h1-hn)+bn(μ-γ)≥0,与(13)式矛盾,从而γμ.这时,为迭代求解方程(1),建立拟牛顿迭代格式如下:

    定理2  设m≥2是正整数,B < 0,C >0且BC = CB,则∀X 0=tIγtμ,迭代式(14)总收敛到方程(1)的正定解.

      类似于定理1的证明过程,考虑到方程ym+diy-ci=0(dici>0)的牛顿迭代

    所产生的数列{yk(i)}(i=1,2,…,n)分别收敛到区间($0, \frac{{{c_i}}}{{{d_i}}}$)内唯一正实数,且当γtμt∈ ($0, \frac{{{c_i}}}{{{d_i}}}$),于是迭代式(14)总收敛到方程(1)的正定解.

    注3  由定理2可知,当B < 0,C >0且BC = CB时,方程(1)的正定解$\mathit{\boldsymbol{\hat X}}$满足不等式γI$\mathit{\boldsymbol{\hat X}}$μI,其中μγ分别是xm-b1x-h1=0与xm-bnx-hn=0的唯一正根.

  • B是不定矩阵,C是正定矩阵时,不妨设

    利用前面1.2的方法,容易知道引理2的结论仍然成立.也就是说方程xm-b1x-h1=0与xm-bnx-hn=0分别存在唯一正根στ,且τσ.这时可把方程(1)改写为

    根据方程(16)建立新的拟牛顿迭代格式

    定理3  设m≥2是正整数,B是Hermite矩阵,C > 0BC = CB,则∀X 0=tIτtσ,迭代式(17)总收敛到方程(1)的正定解.

      由前面的讨论可知,我们只需考虑方程ym-biy-ci=0(ci>0,bi$\mathbb{R}$)的牛顿迭代

    对于初值y0(i)=t的收敛性.

    bi < 0时,由(15)式可知对∀t$\left( 0, \frac{{{c}_{i}}}{\left| {{b}_{i}} \right|} \right)$,迭代式(18)收敛;

    bi=0时,显然对∀t∈(0,$\sqrt[m]{{{c}_{i}}}$),迭代式(18)收敛;

    bi>0时,由注2可知对∀t∈(0,$\sqrt[m]{{{b}_{i}}\ \sqrt[m]{{{c}_{i}}}+{{c}_{i}}}$),迭代式(18)收敛.

    又因为

    其中bn=λmin(B) < 0,hn=λmin(C)>0.令δ=min$\left\{ \frac{{{h}_{n}}}{\left| {{b}_{n}} \right|}, \sqrt[m]{{{h}_{n}}} \right\}$,则当t∈(0,δ)时,迭代式(18)的所有数列{yk(i)}都收敛.所以选取X 0=tI,迭代式(17)总收敛到方程(1)的正定解.

2.   数值算例
  • 例1  给定两个n阶Hermite正定矩阵

    试求方程X 5- BX - C = 0的Hermite正定解.

      对任意正整数n直接验证可知BC = CB,因此由引理1知方程X 5- BX - C = 0存在Hermite正定解.

    对于B > 0C > 0,可采用(9)式建立迭代格式.根据Getschgorin圆盘定理,B的特征值bi分布在圆盘|λ-2|≤2内的实轴上,C的特征值ci分布在圆盘|λ-5|≤2.4内的实轴上.由于$\sqrt[5]{2\cdot \sqrt[5]{5}+5}\approx 1.5$,根据定理1的注2,迭代初始矩阵可选取为X 0=1.5 I,因此有

    n=5时,计算结果为

    用error=‖ Xk5- BXk- C表示第k次近似解的余项,iter表示迭代次数,time表示运行时间,当n=100,500,1 000时,计算结果见表 1.

    $\mathit{\boldsymbol{\widetilde B}}$=- B,则$\mathit{\boldsymbol{\widetilde B}}$是一个Hermite负定矩阵.对于$\mathit{\boldsymbol{\widetilde B}}$ < 0C > 0可采用迭代(14)来求方程X 5-$\mathit{\boldsymbol{\widetilde B}}$- C = 0的正定解.根据BC的特征值分布和定理2证明过程中(15)式可知,迭代初始矩阵可选取为X 0=1.25 I,因此有

    n=5时,计算结果为

    n=100,500,1 000时,计算结果见表 1.

    表 1结果显示,本文所给3种迭代用较少迭代次数均能达到预设误差精度,表明敛速较高.

    给出一个不定矩阵如下:

    对于不定矩阵$\mathit{\boldsymbol{\mathord{\buildrel{\lower3pt\hbox{$\scriptscriptstyle\frown$}} \over B} }}$C > 0,可采用迭代式(17)来求方程X 5- $\mathit{\boldsymbol{\mathord{\buildrel{\lower3pt\hbox{$\scriptscriptstyle\frown$}} \over B} X}}$ - C = 0的正定解.由于$\mathit{\boldsymbol{\mathord{\buildrel{\lower3pt\hbox{$\scriptscriptstyle\frown$}} \over B} }}$C的特征值范围是-1.5≤λ($\mathit{\boldsymbol{\mathord{\buildrel{\lower3pt\hbox{$\scriptscriptstyle\frown$}} \over B} }}$)≤2.5,2.6≤λ(C)≤7.4,根据定理3的证明过程可知,迭代初始矩阵可选取为X 0=δI,其中δ=min $\left\{ {\frac{{{h_n}}}{{\left| {{b_n}} \right|}}, \sqrt[m]{{{h_n}}}} \right\} \approx {\rm{min}}\left\{ {\frac{{2.6}}{{1.5}}, \sqrt[5]{{2.6}}} \right\} \approx 1.2$,因此有

    n=5时,计算结果为

    n=100,500,1 000时,计算结果见表 1.

3.   结论
  • 非线性系统(1)是一个非对称m次矩阵方程,本文在BC为Hermite矩阵且C正定的条件下讨论了它的正定解.给出了方程(1)有正定解的充要条件及其解的存在区间.针对矩阵B分别为正定、负定、不定3种情况建立相应的迭代格式,并结合代数方程的特点证明了它们的收敛性.其中迭代式(9)充分利用了迭代序列的严格单调性,迭代式(14)根据矩阵函数求导恒正的特点建立了拟牛顿算法,迭代式(17)通过把BX转化成${\mathit{\boldsymbol{X}}^{\frac{1}{2}}}\mathit{\boldsymbol{B}}{\mathit{\boldsymbol{X}}^{\frac{1}{2}}}$克服了对不定矩阵B开方的障碍.根据3种迭代的特点分析给出初始矩阵的选取方法.数值算例表明,本文所给方法对求解非线性系统(1)可行、有效.

Table (1) Reference (16)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return