Message Board

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

2022 Volume 47 Issue 1
Article Contents

DAI Lifang, LIANG Maolin. Solvability Conditions for a Class of Tensor Equations and Associated Optimal Approximation Problems[J]. Journal of Southwest China Normal University(Natural Science Edition), 2022, 47(1): 15-20. doi: 10.13718/j.cnki.xsxb.2022.01.003
Citation: DAI Lifang, LIANG Maolin. Solvability Conditions for a Class of Tensor Equations and Associated Optimal Approximation Problems[J]. Journal of Southwest China Normal University(Natural Science Edition), 2022, 47(1): 15-20. doi: 10.13718/j.cnki.xsxb.2022.01.003

Solvability Conditions for a Class of Tensor Equations and Associated Optimal Approximation Problems

More Information
  • Corresponding author: LIANG Maolin
  • Received Date: 21/09/2020
    Available Online: 20/01/2022
  • MSC: O241.6

  • This paper is concerned with the solution to the tensor equation A*nX=B with Hermitian X, where represents the Einstein product. Depending on the properties of Moore-Penrose generalized inverses of tensors, the solvability conditions for the existence of the Hermitian solution to the above tensor equation as well as its general solution have been derived. Meanwhile, the associated tensor approximation problem for any given tensor has been considered and the unique solution has been given. Finally, the performed numerical results demonstrate the feasibility of the proposed results.
  • 加载中
  • [1] COPPI R, BOLASCO S. Multiway Data Analysis[M]. Amsterdam: Elsevier, 1989.

    Google Scholar

    [2] QI L Q, LUO Z Y. Tensor Analysis[M]. Philadelphia: Society for Industrial and Applied Mathematics, 2017.

    Google Scholar

    [3] LAI W, RUBIN D, KREMPL E. Introduction To Continuum Mechanics [M]. Amsterdam: Elsevier, 2009.

    Google Scholar

    [4] KOLDA T G, BADER B W. Tensor Decompositions and Applications [J]. SIAM Review, 2009, 51(3): 455-500. doi: 10.1137/07070111X

    CrossRef Google Scholar

    [5] DE SILVA V, LIM L H. Tensor Rank and the Ill-Posedness of the Best Low-Rank Approximation Problem [J]. SIAM Journal on Matrix Analysis and Applications, 2008, 30(3): 1084-1127. doi: 10.1137/06066518X

    CrossRef Google Scholar

    [6] QI L Q. Eigenvalues of a Real Supersymmetric Tensor [J]. Journal of Symbolic Computation, 2005, 40(6): 1302-1324. doi: 10.1016/j.jsc.2005.05.007

    CrossRef Google Scholar

    [7] BRAZELL M, LI N, NAVASCA C, et al. Solving Multilinear Systems via Tensor Inversion [J]. SIAM Journal on Matrix Analysis and Applications, 2013, 34(2): 542-570. doi: 10.1137/100804577

    CrossRef Google Scholar

    [8] LI X T, NG M K. Solving Sparse Non-Negative Tensor Equations: Algorithms and Applications [J]. Frontiers of Mathematics in China, 2015, 10(3): 649-680. doi: 10.1007/s11464-014-0377-3

    CrossRef Google Scholar

    [9] DING W Y, WEI Y M. Solving Multi-Linear Systems with M-Tensors [J]. Journal of Scientific Computing, 2016, 68(2): 689-715. doi: 10.1007/s10915-015-0156-7

    CrossRef Google Scholar

    [10] EINSTEIN A. The Foundation of the General Theory of Relativity [M]// KOX A, KLEIN M, SCHULMANN R. The Collected Papers of Albert Einstein. Princeton: Princeton University Press, 2007: 146-200.

    Google Scholar

    [11] NI G. Hermitian Tensor And Quantum Mixed State[EB/OL]. (2019-08-23)[2021-04-01]. https://arxiv.org/pdf/1902.02640.pdf.

    Google Scholar

    [12] SUN L Z, ZHENG B D, BU C J, et al. Moore-Penrose Inverse of Tensors via Einstein Product [J]. Linear and Multilinear Algebra, 2016, 64(4): 686-698. doi: 10.1080/03081087.2015.1083933

    CrossRef Google Scholar

    [13] PENG Y X, HU X Y, ZHANG L. An Iteration Method for the Symmetric Solutions and the Optimal Approximation Solution of the Matrix Equation AXB=C [J]. Applied Mathematics and Computation, 2005, 160(3): 763-777. doi: 10.1016/j.amc.2003.11.030

    CrossRef Google Scholar

    [14] HIGHAM N J. Computing a Nearest Symmetric Positive Semidefinite Matrix [J]. Linear Algebra and Its Applications, 1988, 103: 103-118. doi: 10.1016/0024-3795(88)90223-6

    CrossRef Google Scholar

    [15] YUAN Y X, DAI H. The Nearness Problems for Symmetric Matrix with a Submatrix Constraint [J]. Journal of Computational and Applied Mathematics, 2008, 213(1): 224-231. doi: 10.1016/j.cam.2007.01.033

    CrossRef Google Scholar

    [16] HUANG G X, NOSCHESE S, REICHEL L. Regularization Matrices Determined by Matrix Nearness Problems [J]. Linear Algebra and Its Applications, 2016, 502: 41-57. doi: 10.1016/j.laa.2015.12.008

    CrossRef Google Scholar

    [17] ZHANG T, GOLUB G H. Rank-One Approximation to High Order Tensors [J]. SIAM Journal on Matrix Analysis and Applications, 2001, 23(2): 534-550. doi: 10.1137/S0895479899352045

    CrossRef Google Scholar

    [18] GANDY S, RECHT B, YAMADA I. Tensor Completion and Low-n-Rank Tensor Recovery via Convex Optimization [J]. Inverse Problems, 2011, 27(2): 025010. doi: 10.1088/0266-5611/27/2/025010

    CrossRef Google Scholar

    [19] LIANG M L, ZHENG B. Further Results on Moore-Penrose Inverses of Tensors with Application to Tensor Nearness Problems [J]. Computers & Mathematics With Applications, 2019, 77(5): 1282-1293.

    Google Scholar

    [20] BADER B, KOLDA T, MAYO J, et al. MATLAB Tensor Toolbox Version 3.2[EB/OL]. (2015-02-18)[2020-12-15]. http://www.tensortoolbox.org.

    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

Figures(1)

Article Metrics

Article views(658) PDF downloads(235) Cited by(0)

Access History

Other Articles By Authors

Solvability Conditions for a Class of Tensor Equations and Associated Optimal Approximation Problems

    Corresponding author: LIANG Maolin

Abstract: This paper is concerned with the solution to the tensor equation A*nX=B with Hermitian X, where represents the Einstein product. Depending on the properties of Moore-Penrose generalized inverses of tensors, the solvability conditions for the existence of the Hermitian solution to the above tensor equation as well as its general solution have been derived. Meanwhile, the associated tensor approximation problem for any given tensor has been considered and the unique solution has been given. Finally, the performed numerical results demonstrate the feasibility of the proposed results.

  • 张量是数值多重线性代数的主要研究对象,其在量子力学、心理测量学、化学计量学、信号处理、高阶统计等领域有重要应用[1-3]. 张量是向量和矩阵的高阶推广,它的许多性质与矩阵情形类似,但也有很大不同[4]. 张量相关问题的研究要比矩阵情形复杂得多. 目前,在张量分解、张量的低秩逼近、张量互补问题、张量特征值问题和张量方程等方面已有诸多研究成果[4-9]. 本文考虑基于Einstein积[10]的一类张量方程的求解问题.

    为了表述方便,首先介绍本文用到的一些记号和定义. m-阶I1×I2×…×Im-维复张量A=(ai1i2im),ai1i2im$ \mathbb{C} $,1≤ijIj(j=1,2,…,m). 我们用$ \mathbb{C} $I1×I2×…×Im表示所有m-阶I1×I2×…×Im-维复张量的全体. 本文中In=(ei1i2inj1j2jn)∈$ \mathbb{R} $I1×I2×…×In×I1×I2×…×In表示单位张量,它的元素定义为ei1i2inj1j2jn=$ \mathop \Pi \limits_{k = 1}^n $δikjk,这里δikjk=1如果ik=jk,否则δikjk=0.

    若张量A=(ai1i2imj1j2jn)∈$ \mathbb{C} $I1×I2×…×Im×J1×J2×…×JnB=(bj1j2jnk1k2kp)∈$ \mathbb{C} $J1×J2×…×Jn×K1×K2×…×Kp,则它们的Einstein积A*nBI1×I2×…×Im×K1×K2×…×Kp-维张量,依据文献[7]定义其元素为

    进一步,设张量S=(ai1i2imj1j2jn),T=(bk1k2kml1l2ln)且ST$ \mathbb{C} $I1×I2×…×Im×J1×J2×…×Jn,则其内积定义为

    张量的内积诱导出张量的Frobenius范数,即对于张量T有‖T‖= $\sqrt {\left\langle {\mathit{\boldsymbol{T}}, \mathit{\boldsymbol{T}}} \right\rangle } $.

    定义1  设张量A=(ai1i2imj1j2jn)∈$ \mathbb{C} $I1×I2×…×Im×J1×J2×…×Jn,则其共轭转置AH定义为

    若张量A$ \mathbb{C} $I1×I2×…×Im×I1×I2×…×Im满足条件AH=A,则称之为Hermitian张量.

    若定义1中A是实张量,则共轭转置退化为转置[7]. 值得一提的是,最新的研究发现,Hermitian张量在量子纠缠中有实际应用[11]. 本文考虑张量方程

    的Hermitian解,这里AB$ \mathbb{C} $P1×P2×…×Pm×I1×I2×…×In为已知张量,X$ \mathbb{C} $I1×I2×…×In×I1×I2×…×In为未知的Hermitian张量. 张量方程(1)在控制系统、连续力学等领域有实际应用[7, 11]. 例如,对于2-维泊松方程

    Ω={(xy)|0≤xy≤1},利用中心差分格式,可以离散为张量方程[7]

    这里张量A$ \mathbb{R} $N×N×N×N的非零元为

    F$ \mathbb{R} $N×N,对于没有约束条件的张量方程(1),文献[7]引入了张量逆的概念,得到了它的最小二乘解. 进一步,作为张量逆的推广形式,文献[12]提出了张量的Moore-Penrose广义逆,并讨论了张量方程(1)的可解性及其通解形式.

    在图像处理等领域的应用中,考虑方程解的特殊结构是降低算法复杂度的重要途径[13],但是对于带有约束条件的形如方程(1)的张量方程求解问题尚无研究. 借助张量的Moore-Penrose广义逆的性质,我们将建立张量方程组(1)有Hermitian解的充要条件,并得到有解时的一般解表达式. 进一步,将考虑张量方程(1)约束下的张量逼近问题.

    X0$ \mathbb{C} $I1×I2×…×In×I1×I2×…×In为一给定张量,求Hermitian张量$ {\mathop {\boldsymbol{X}}\limits^ \wedge } $$ \mathbb{C} $I1×I2×…×In×I1×I2×…×In使之满足等式

    其中Θ表示张量方程(1)的所有Hermitian解的集合.

    矩阵最佳逼近问题在有限元、控制论等领域具有实际应用,且已被深入研究[14-16]. 张量最佳逼近问题(3)可以看作是矩阵情形的直接推广形式,同时也可以看作是张量完全问题、张量低秩逼近问题等应用型问题的一般形式[17-19]. 在式(3)中,张量X0由实际测量或试验观测所得,但由于误差原因,它往往不满足所要求的特殊结构或性质,而$ {\mathop {\boldsymbol{X}}\limits^ \wedge } $是满足实际需求的张量. 若解集合Θ非空,可以证明上述张量逼近问题的解是唯一的,且可以利用已知张量的Moore-Penrose逆具体表示.

1.   主要结果及证明
  • 为了研究Hermitian张量约束下的张量方程(1)的求解问题,我们首先引入如下引理,这对得出本文的主要结果是十分重要的.

    引理1[19]  设F$ \mathbb{C} $I1×I2×…×Im×J1×J2×…×JnG$ \mathbb{C} $K1×K2×…×Kp×L1×L2×…×LqE$ \mathbb{C} $I1×I2×…×Im×L1×L2×…×Lq,则张量方程F*nZ*pG=E有解当且仅当F*nF+*mE*qG+*pG=E,此时它的通解为

    其中张量Y$ \mathbb{C} $J1×J2×…×Jn×K1×K2×…×Kp是任意的.

    引理2  设张量AB$ \mathbb{C} $P1×P2×…×Pm×I1×I2×…×In,则张量方程(1)有Hermitian解的充要条件为张量方程组

    有一般解.

      若张量方程(1)有Hermitian解,则显然张量方程组(4)有解. 反之,若M为张量方程组(4)的一个解,即A*nM=BM*nAH=BH. 令$ \mathit{\boldsymbol{Y}} = \frac{{\mathit{\boldsymbol{M}} + {\mathit{\boldsymbol{M}}^{\rm{H}}}}}{2} $,显然Y是一个Hermitian张量且满足式(1),即张量Y为它的一个Hermitian解.

    由引理2可见,求解张量方程(1)的Hermitian解等价于求解张量方程组(4)的一般解. 基于此,我们得到如下定理.

    定理1  设张量AB$ \mathbb{C} $P1×P2×…×Pm×I1×I2×…×In,则张量方程(1)有Hermitian解的充要条件为

    此时它的一般解为

    其中:$\mathit{\boldsymbol{\bar {\bar X}}}$=$ \frac{1}{2} $[A+*mB+BH*m(AH)++A+*mB*n(I1-A+*mA)+(I1-A+*mA)*nBH*m(AH)+],I1$ \mathbb{R} $I1×I2×…×In×I1×I2×…×In为单位张量,V$ \mathbb{C} $I1×I2×…×In×I1×I2×…×In是任意的Hermitian张量.

      根据张量Moore-Penrose广义逆的性质和引理1可知,张量方程(1)有解的充要条件为

    此时

    这里Z$ \mathbb{C} $I1×I2×…×In×I1×I2×…×In为任意张量. 当式(7)成立时,结合式(4)可知A*nBH=B*nAH,即知式(5)成立.

    进一步,将式(8)代入式(4)的第二个方程X*nAH=BH并整理得

    这是关于变量Z的张量方程. 在条件(7)成立时,该方程总是有解的,且由引理1知其一般解为

    其中W$ \mathbb{C} $I1×I2×…×In×I1×I2×…×In为任意张量. 将式(9)代入式(8)可得

    结合引理2的证明过程可得式(6)成立. 命题得证.

      根据Moore-Penrose广义逆的性质有〈$\mathit{\boldsymbol{\bar {\bar X}}}$,(I1A+*mA)*nV*n(I1A+*mA)〉=0,故$\mathit{\boldsymbol{\bar {\bar X}}}$为张量方程(1)的Hermitian极小范数解.

    接下来考虑张量的最佳逼近问题(3). 首先引入如下引理.

    引理3[19]  设E$ \mathbb{C} $I1×I2×…×Im×J1×J2×…×JnF$ \mathbb{C} $I1×I2×…×Im×I1×I2×…×ImG$ \mathbb{C} $J1×J2×…×Jn×J1×J2×…×Jn且满足

    则‖EF*mE*nG‖=$ \mathop {\min }\limits_{\boldsymbol{H} \in \mathbb{C} ^{ {I_1} \times {I_2} \times \cdots \times {I_m} \times J_1 \times J_2 \times \cdots \times {J_n}} } \, $EF*mH*nG‖当且仅当F*m(EH)*nG=O.

    根据引理3,我们可以证明张量逼近问题(3)的解是唯一的,并给出解的具体形式.

    定理2  给定张量X0$ \mathbb{C} $I1×I2×…×In×I1×I2×…×In. 若定理1中的条件(5)成立,则张量最佳逼近问题(3)有唯一Hermitian解$ {\mathop {\boldsymbol{X}}\limits^ \wedge } $,即

    成立,此时

    其中张量$\mathit{\boldsymbol{\bar {\bar X}}}$的具体形式见定理1.

      当定理1中的条件(5)成立时,张量方程(1)的解集合Θ是非空的,且容易验证它是一个闭凸集,这说明最佳逼近问题(3)有唯一解. 由式(6)和式(3)可得

    因为I1A+*mA为正交投影张量,故满足引理3中的假设条件,从而有

    当且仅当

    由式(12)和式(13)可得式(10),而式(6)和式(14)说明最佳逼近问题(10)的唯一解为

    利用$\mathit{\boldsymbol{\bar {\bar X}}}$的具体表达式和张量Moore-Penrose广义逆的性质,化简式(15)即知式(11)成立.

2.   数值试验
  • 本节通过数值例子验证所得结论的可行性. 接下来的所有实验数据均是通过配置了Inter($ \mathbb{R} $) Core(TM) i5-4200M CPU与4.00 G内存的电脑上的MATLAB软件编程实现,其中张量积的运算用到了张量工具包[20].

      考虑离散的2-维泊松方程的Hermitian解,即张量方程(2),这里取张量F=eyes(NN)为单位张量. 另外,给定张量U0$ \mathbb{R} $N×N随机产生,即U0=randn(NN),容易验证,已知张量AF满足定理1中的条件(5),故相应的张量逼近问题

    有唯一解$ {\mathop {\boldsymbol{U}}\limits^ \wedge } $. 对于不同正整数N和张量U0,根据定理2,可以得到上述最佳逼近问题的解(图 1). 图 1展示了最佳逼近解的对称性.

3.   总结
  • 本文考虑了基于Einstein积的张量方程A*nX=B关于Hermitian张量X的可解性问题. 利用张量Moore-Penrose广义逆的性质,得到了上述问题有解的充要条件,并得到了它的一般解表达式. 另外,对于任意给定张量,在假定上述条件成立时,讨论了相应的张量最佳逼近问题,证明了解的唯一性,并得到了它的具体表达式. 最后给出实际应用实例验证了本文所得结果的可行性,这些结果对张量相关理论的完善具有重要意义.

Figure (1)  Reference (20)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return