
In this paper, we propose an efficient method for some special solutions of the quaternion matrix equation AXB+CYD=E. By integrating real representation of a quaternion matrix with H-representation, we investigate the minimal norm least squares solution of the previous quaternion matrix equation over different constrained matrices and obtain their expressions. In this way, we first apply H-representation to solve quaternion matrix equation with special structure, which not only broadens the application scope of H-representation, but further expands the research idea of solving quaternion matrix equation. The algorithms only include real operations. Consequently, it is very simple and convenient, and it can be applied to all kinds of quaternion matrix equation with similar problems. The numerical example is provided to illustrate the feasibility of our algorithms.
Citation: Anli Wei, Ying Li, Wenxv Ding, Jianli Zhao. Three special kinds of least squares solutions for the quaternion generalized Sylvester matrix equation[J]. AIMS Mathematics, 2022, 7(4): 5029-5048. doi: 10.3934/math.2022280
[1] | Dong Wang, Ying Li, Wenxv Ding . The least squares Bisymmetric solution of quaternion matrix equation $ AXB = C $. AIMS Mathematics, 2021, 6(12): 13247-13257. doi: 10.3934/math.2021766 |
[2] | Fengxia Zhang, Ying Li, Jianli Zhao . The semi-tensor product method for special least squares solutions of the complex generalized Sylvester matrix equation. AIMS Mathematics, 2023, 8(3): 5200-5215. doi: 10.3934/math.2023261 |
[3] | Fengxia Zhang, Ying Li, Jianli Zhao . A real representation method for special least squares solutions of the quaternion matrix equation $ (AXB, DXE) = (C, F) $. AIMS Mathematics, 2022, 7(8): 14595-14613. doi: 10.3934/math.2022803 |
[4] | Yimeng Xi, Zhihong Liu, Ying Li, Ruyu Tao, Tao Wang . On the mixed solution of reduced biquaternion matrix equation $ \sum\limits_{i = 1}^nA_iX_iB_i = E $ with sub-matrix constraints and its application. AIMS Mathematics, 2023, 8(11): 27901-27923. doi: 10.3934/math.20231427 |
[5] | Huiting Zhang, Yuying Yuan, Sisi Li, Yongxin Yuan . The least-squares solutions of the matrix equation $ A^{\ast}XB+B^{\ast}X^{\ast}A = D $ and its optimal approximation. AIMS Mathematics, 2022, 7(3): 3680-3691. doi: 10.3934/math.2022203 |
[6] | Wenxv Ding, Ying Li, Anli Wei, Zhihong Liu . Solving reduced biquaternion matrices equation $ \sum\limits_{i = 1}^{k}A_iXB_i = C $ with special structure based on semi-tensor product of matrices. AIMS Mathematics, 2022, 7(3): 3258-3276. doi: 10.3934/math.2022181 |
[7] | Abdur Rehman, Ivan Kyrchei, Muhammad Zia Ur Rahman, Víctor Leiva, Cecilia Castro . Solvability and algorithm for Sylvester-type quaternion matrix equations with potential applications. AIMS Mathematics, 2024, 9(8): 19967-19996. doi: 10.3934/math.2024974 |
[8] | Kahraman Esen Özen . A general method for solving linear matrix equations of elliptic biquaternions with applications. AIMS Mathematics, 2020, 5(3): 2211-2225. doi: 10.3934/math.2020146 |
[9] | Xiaohan Li, Xin Liu, Jing Jiang, Jian Sun . Some solutions to a third-order quaternion tensor equation. AIMS Mathematics, 2023, 8(11): 27725-27741. doi: 10.3934/math.20231419 |
[10] | Jin Zhong, Yilin Zhang . Dual group inverses of dual matrices and their applications in solving systems of linear dual equations. AIMS Mathematics, 2022, 7(5): 7606-7624. doi: 10.3934/math.2022427 |
In this paper, we propose an efficient method for some special solutions of the quaternion matrix equation AXB+CYD=E. By integrating real representation of a quaternion matrix with H-representation, we investigate the minimal norm least squares solution of the previous quaternion matrix equation over different constrained matrices and obtain their expressions. In this way, we first apply H-representation to solve quaternion matrix equation with special structure, which not only broadens the application scope of H-representation, but further expands the research idea of solving quaternion matrix equation. The algorithms only include real operations. Consequently, it is very simple and convenient, and it can be applied to all kinds of quaternion matrix equation with similar problems. The numerical example is provided to illustrate the feasibility of our algorithms.
In this paper, we adopt the following notations. R represents the real number field; Rn stands for the set of all real column vectors with order n; Rm×n stands for the set of all m×n real matrices. C represents the complex number field; Cn stands for the set of all complex column vectors with order n; Cm×n stands for the set of all m×n complex matrices. The sets Un,U−n,Vn,Wn represent the set of all n×n tridiagonal symmetric matrices, tridiagonal skew-symmetric matrices, Brownian matrices, Generalized Rotation matrices, respectively. Q stands for the quaternion skew-field; Qn stands for the set of all quaternion column vectors with order n; Qm×n represents the set of all m×n quaternion matrices; HTQn×n, AHTQn×n, BQn×n, MQn×n represent the set of all n×n quaternion tridiagonal Hermitian matrices, quaternion tridiagonal anti-Hermitian matrices, quaternion Brownian matrices, quaternion Generalized Rotation matrices, respectively. In represents the unit matrix with order n. For matrix A, AT,AH,A† stand for the transpose, the conjugate transpose, Moore-Penrose inverse of matrix A, respectively. ⊗ represents the Kronecker product of matrices. ‖⋅‖ represents the Frobenius norm of a matrix or Euclidean norm of a vector. For C=(c1,c2,…,cn)∈Rm×n, vec(C) means the vector operator, i.e., vec(C)=(cT1,cT2,…,cTk)T.
Matrix equations can be encountered in many areas, such as system theory, control theory, stability analysis, some fields of pure and applied mathematics and so on [1,2,3]. With the rapid development of these fields, more and more scholars are interested in matrix equations and have obtained many valuable results [4,5,6]. Now, we turn our attention to quaternion matrix equation. Quaternion matrix equations and their least squares solutions are widely applied in many fields, such as computer science, quantum mechanics, control theory, field theory and so on [7,8,9]. Therefore, many people are engaged in studying theoretical properties and numerical computations of quaternion matrix equations. By means of complex representation, Jiang et al. studied algebraic algorithm for quaternion least squares problem [10] and quaternion eigenvalue problem [11]; Yuan et al. studied the quaternion least squares problems for the quaternion matrix equations AXB+CXD=E [12], X−AˆXB=C [13]. By applying the real representation of quaternion matrices, Wang et al. proposed an iterative method for solving the quaternion least squares problem [14].
Consider the generalized Sylvester matrix equation
AXB+CYD=E, | (1.1) |
If B and C are identity matrices, then the matrix Eq (1.1) reduces to the well-known Sylvester matrix Eq [15]. If C and D are identity matrices, then the matrix Eq (1.1) reduces to the well-known Stein matrix equation. It has extensive application value in robust control, feedback control, pole assignment design, neural network and so on [16,17,18]. There are many important results about their solutions, for example, [19] and [20] considered the solvability condition for the complex and real matrix Eq (1.1), respectively. For the quaternion matrix Eq (1.1), [21] derived necessary and sufficient conditions for the existence of a solution or a unique solution using the method of complex representation of quaternion matrices; [12,22] studied η-Hermitian and η-anti-Hermitian solutions to the quaternion matrix equations AXB+CXD=E, AXB+CYD=E, respectively; [23] obtained the expression of solutions of a system of quaternion matrix equations including η-Hermicity. Also, it is worth noting that a number of important results on Sylvester operators have been obtained in recent years. For instance, [24] studied some features of slice semi-regular functions SξM(Ω) on a circular domain Ω and verified the equivalence of slice semi-regular functions via Sylvester operators; [25] applied the existing results to establish some outcomes of the study of the behaviour of a class of linear operators, which include the Sylvester ones, acting on slice semi-regular functions. The name of Sylvester operator is due to the fact that, when dealing with matrices, equation Sf,g(χ)=b is usually called Sylvester equation. In the most common use, Sylvester equations are special matrices equations, introduced by Sylvester himself [26], which are used in several subjects, including similarity, commutativity, control theory and differential equation [27]. In the quaternionic setting, such equations were studied with different purposes. In this paper, we consider the least squares problem with different constrains for quaternion matrix Eq (1.1) based on the real representation of quaternion matrices together with the H-representation method, which is able to transform a matrix-valued equation into a standard vector-valued equation with independent coordinates. The related problems are described as follows.
Problem 1. Let A∈Qm×p,B∈Qp×n,C∈Qm×q,D∈Qq×n,E∈Qm×n, and
TL={(X,Y)|X∈HTQp×p,Y∈AHTQq×q,‖AXB+CYD−E‖=min}. |
Find out (XH,YA)∈TL such that
‖(XH,YA)‖=min(X,Y)∈TL‖(X,Y)‖. |
The solution (XH,YA) in Problem 1 is called the minimal norm least squares tridiagonal mixed solution.
Problem 2. Let A∈Qm×p,B∈Qp×n,C∈Qm×q,D∈Qq×n,E∈Qm×n, and
BL={(X,Y)|X∈BQp×p,Y∈BQq×q,‖AXB+CYD−E‖=min}. |
Find out (XB,YB)∈BL such that
‖(XB,YB)‖=min(X,Y)∈BL‖(X,Y)‖. |
The solution (XB,YB) in Problem 2 is called the minimal norm least squares Brownian solution.
Problem 3. Let A∈Qm×p,B∈Qp×n,C∈Qm×q,D∈Qq×n,E∈Qm×n, and
ML={(X,Y)|X∈MQp×p,Y∈MQq×q,‖AXB+CYD−E‖=min}. |
Find out (XM,YM)∈ML such that
‖(XM,YM)‖=min(X,Y)∈ML‖(X,Y)‖. |
The solution (XM,YM) in Problem 3 is called the minimal norm least squares Rotation solution.
The remaining content of this paper is organized as follows. In Section 2, we study and recall some preliminary results with regard to the real representation of a quaternion matrix, and then introduce some matrix sets with special structures. In Section 3, we give the concept of H-representation and subsequently study its properties. In Section 4, on the basis of the real representation matrix of a quaternion matrix and H-representation of matrices with special structures, operational properties, the properties of Frobenius norm and Moore-Penrose generalized inverse, we can convert Problems 1–3 into the corresponding problems of the real matrix equation over free variables, and then the unique solution (X,Y) and expressions for special solution are established. In addition, the necessary and sufficient conditions for the quaternion matrix equation to have solution with special structure are included as corollaries. In Section 5, we provide numerical algorithms for solving Problems 1–3 by the results obtained in Section 4, and afterwards we present a numerical example to verify the feasibility of our proposed method. Finally, in Section 6, we put some conclusions.
We start by recalling the usual Kronecker product.
Definition 2.1. For any two matrices A=(aij)∈Rm×n,B∈Rp×q, the Kronecker product of A and B is defined as
A⊗B:=[a11Ba12B⋯a1nBa21Ba22B⋯a2nB⋮⋮⋮am1Bam2B⋯amnB]∈Rmp×nq. |
We now turn to recall the standard representation of a quaternion.
Definition 2.2. A quaternion q∈Q is represented as
q=q1+q2i+q3j+q4k, |
where q1,q2,q3,q4∈R, and three imaginary units i,j,k satisfy
i2=j2=k2=ijk=−1,ij=k,jk=i and ki=j. |
Definition 2.3. A quaternion matrix A∈Qm×p is represented as
A=A1+A2i+A3j+A4k, |
where A1,A2,A3,A4∈Rm×p. The conjugate matrix of A is defined as
¯A=A1−A2i−A3j−A4k. |
We recall a standard norm in this setting.
Definition 2.4. [28] The Frobenius norm of A=A1+A2i+A3j+A4k is defined as
‖A‖=√‖A1‖2+‖A2‖2+‖A3‖2+‖A4‖2. |
Definition 2.5. [28] For A=A1+A2i+A3j+A4k∈Qm×p, its real representation matrix →A is defined as follows:
→A=[A1−A2−A3−A4A2A1−A4A3A3A4A1−A2A4−A3A2A1]∈R4m×4p. |
According to the matrix blocks of the real representation matrix →A, if we know a column block of →A, we know →A. For the sake of convenience, we use →Ac to represent the first column block of →A, i.e.,
→Ac=[A1A2A3A4]. |
Next, we investigate some properties of →Ac, which will be used in the sequel.
Lemma 2.1. [28] Suppose A,B∈Qm×n,C∈Qn×p,t∈R, then we have
(i)A=B⇔→A=→B⇔→Ac=→Bc;(ii)→A+B=→A+→B,→tA=t→A,→AC=→A→C;(iii)→A+Bc=→Ac+→Bc,→tAc=t→Ac,→ACc=→A→Cc;(iv)‖A‖=12‖→A‖=‖→Ac‖. |
Proof. We only provide detailed proof of →AC=→A→C,→ACc=→A→Cc, and the rest are similarly verifiable. Suppose A=A1+A2i+A3j+A4k∈Qm×n, C=C1+C2i+C3j+C4k∈Qn×p, then
AC=(A1C1−A2C2−A3C3−A4C4)+(A1C2+A2C1+A3C4−A4C3)i+(A1C3−A2C4+A3C1+A4C2)j+(A1C4+A2C3−A3C2+A4C1)k. |
According to the Definition 2.5, we have
→A=[A1−A2−A3−A4A2A1−A4A3A3A4A1−A2A4−A3A2A1],→C=[C1−C2−C3−C4C2C1−C4C3C3C4C1−C2C4−C3C2C1],→Cc=[C1C2C3C4], |
and
→AC=[A1C1−A2C2−A3C3−A4C4−A1C2−A2C1−A3C4+A4C3−A1C3+A2C4−A3C1−A4C2−A1C4−A2C3+A3C2−A4C1A1C2+A2C1+A3C4−A4C3A1C1−A2C2−A3C3−A4C4−A1C4−A2C3+A3C2−A4C1A1C3−A2C4+A3C1+A4C2A1C3−A2C4+A3C1+A4C2A1C4+A2C3−A3C2+A4C1A1C1−A2C2−A3C3−A4C4−A1C2−A2C1−A3C4+A4C3A1C4+A2C3−A3C2+A4C1−A1C3+A2C4−A3C1−A4C2A1C2+A2C1+A3C4−A4C3A1C1−A2C2−A3C3−A4C4]=[A1−A2−A3−A4A2A1−A4A3A3A4A1−A2A4−A3A2A1][C1−C2−C3−C4C2C1−C4C3C3C4C1−C2C4−C3C2C1]=→A→C,→ACc=[A1C1−A2C2−A3C3−A4C4A1C2+A2C1+A3C4−A4C3A1C3−A2C4+A3C1+A4C2A1C4+A2C3−A3C2+A4C1]=[A1−A2−A3−A4A2A1−A4A3A3A4A1−A2A4−A3A2A1][C1C2C3C4]=→A→Cc. |
We now recall a couple of algebraic results about the structure of quaternion matrices and their real representation.
Lemma 2.2. [14] Suppose X∈Qp×p, then vec(→X)=Jvec(→Xc), where
J=[diag(I4p,…,I4p)diag(Fp,…,Fp)diag(Hp,…,Hp)diag(Sp,…,Sp)], |
and
Fp=[0−Ip00Ip000000Ip00−Ip0],Hp=[00−Ip0000−IpIp0000Ip00],Sp=[000−Ip00Ip00−Ip00Ip000]. |
Lemma 2.3. [14] Suppose X=X1+X2i+X3j+X4k∈Qp×p, then
vec(→Xc)=K[vec(X1)vec(X2)vec(X3)vec(X4)], |
where
K=[Ip0⋯000⋯000⋯000⋯000⋯0Ip0⋯000⋯000⋯000⋯000⋯0Ip0⋯000⋯000⋯000⋯000⋯0Ip0⋯00Ip⋯000⋯000⋯000⋯000⋯00Ip⋯000⋯000⋯000⋯000⋯00Ip⋯000⋯000⋯000⋯000⋯00Ip⋯0⋮⋮⋮⋮⋮⋮⋮⋮⋮⋮⋮⋮00⋯Ip00⋯000⋯000⋯000⋯000⋯Ip00⋯000⋯000⋯000⋯000⋯Ip00⋯000⋯000⋯000⋯000⋯Ip]∈Q4p2×4p2. |
Remark 2.1. Either the J in Lemma 2.2 or the K in Lemma 2.3 is just a bridge connecting the left and right sides of its equation. By the structure of real representation of quaternion matrix and its first column block, we only need to figure out the relationship between the left and right sides of its equation to express J or K.
We will now present four special matrix sets. We refer to [29] for all the concepts involved in this paper.
Definition 2.6. A tridiagonal symmetric matrix P∈Un is an n×n matrix with the following form
[x11x12⋯0x12x22⋱⋮⋮⋱⋱xn−1,n0⋯xn−1,nxnn]. |
Definition 2.7. A tridiagonal skew-symmetric matrix P∈U−n is an n×n matrix with the following form
[0x12⋯0−x120⋱⋮⋮⋱⋱xn−1,n0⋯−xn−1,n0]. |
Definition 2.8. A matrix P∈Vn is called the Brownian matrix, if
bi,j+1=bij,j>i,i,j=1,…,n−1.bi+1,j=bij,j<i, |
Specifically, the form is as follows:
[b1bn+1bn+1⋯bn+1bn+1b2nb2bn+2⋯bn+2bn+2b2nb2n+1b3⋯bn+3bn+3⋮⋮⋮⋯⋯b2nb2n+1b2n+2⋮bn−1b2n−1b2nb2n+1b2n+2⋮b3n−2bn]. |
Definition 2.9. A Generalized Rotation matrix P∈Wn is an n×n matrix with the following form
[c0c1c2⋯cn−1αcn−1c0c1cn−2αcn−2αcn−1c0⋱⋮⋮⋱⋱c1αc1αc2⋯αcn−1c0]. |
In this section, we will briefly introduce the notion of H-representation and the related properties. Besides, we will also analyze the structure of four special matrix sets mentioned above by means of H-representation and present their H-representation matrices.
Definition 3.1. [30] Consider a p-dimensional complex matrix subspace X⊂Cn×n over the field C. Assume that e1,e2,…,ep form a basis of X, and define H=[vec(e1) vec(e2) … vec(ep)]. If for each X∈X, we express ψ(X)=vec(X) in the form of
ψ(X)=vec(X)=H˜X |
with a p×1 vector ˜X, then H˜X is called an H-representation of ψ(X), and H is called an H-representation matrix of ψ(X).
Remark 3.1. 1) The H-representation of ψ(X) for X∈X is not unique because of the fact that the matrix H may be different owing to the basis choices of X. Apparently, when the basis of X is fixed, the H-representation matrix H, as well as ˜X, is uniquely determined; 2) ψ is used here only as a function name for the convenience of defining its inverse in the sequel.
In what follows, based on the special matrix sets defined in Section 2, we present some simple examples to elucidate Definition 3.1.
Example 3.1. Let X=U3,X=(xij)3×3, then dim(X)=5. If we select the following basis of X
e1=[100000000],e2=[010100000],e3=[000010000],e4=[000001010],e5=[000000001]. |
It is then easy to compute
ψ(X)=vec(X)=[x11 x12 0 x12 x22 x23 0 x23 x33]T, |
H=[100000100000000010000010000010000000001000001],˜X=[x11 x12 x22 x23 x33]T. |
Example 3.2. Let X=V3, then dim(X)=7. If we select the following basis of X
e1=[100000000],e2=[000100100],e3=[011000000],e4=[000010000], |
e5=[000000010],e6=[000001000],e7=[000000001]. |
Then it is easy to compute
ψ(X)=vec(X)=[b1 b6 b6 b4 b2 b7 b4 b5 b3]T, |
H=[100000001000000100000001000000010000000100001000000000100000001],˜X=[b1 b6 b4 b2 b7 b5 b3]T. |
Example 3.3. Let X=W3, then dim(X)=3. If we select the following basis of X
e1=[100010001],e2=[001α000α0],e3=[010001α00]. |
Then we can obtian
ψ(X)=vec(X)=[c0 αc2 αc1 c1 c0 αc2 c2 c1 c0]T, |
H=[1000α000α0011000α0010001100],˜X=[c0 c2 c1]T. |
In this paper, we are interested in the H-representation for X=Un/U−n/Vn/Wn. For X=Un, we select a standard basis throughout this paper as
{E11,E21,E22,E32,…,En−1,n−1,En,n−1,En,n}={Eij:1≤j≤i≤n}, |
where Eij=(elk)n×n with eij=eji=1 and the other entries are zeros. Clearly, for the above given bases, if X=Un, then for any Xn=(xij)n×n∈X, we have
˜Xn=(x11,x21,x22,x32,…,xn−1,n−1,xn,n−1,xnn)T. | (3.1) |
For X=U−n, we select a standard basis throughout this paper as
{E′21,E′32,…,E′n,n−1}={E′ij:1≤j<i≤n}, |
where E′ij=(e′lk)n×n with e′ij=−1,e′ji=1 and the other entries are all zeros. For the above given bases, if X=U−n, then for any X−n=(x′ij)n×n∈X, we have
˜X−n=(x′21,x′32,…,x′n,n−1)T. | (3.2) |
For the convenience of description, the following Z and W both represent X.
Similarly, for Z=Vn, we select a standard basis as
{F11,F21,F12,F22,F32,F23,…,Fn−1,n−1,Fn,n−1,Fn−1,n,Fn,n}={Fij,Fji:1≤i≤j≤n}, |
where Fii=(flk)n×n with fii=1, and Fij,Fji are n×n matrices with fij=1, fji=1 for ∀j>i, respectively, and the other entries are zeros. Based on above bases, for any Zn=(zij)n×n∈Z, we have
˜Zn=(z11,z21,z12,z22,z32,z23,…,zn−1,n−1,zn,n−1,zn−1,n,zn,n)T. | (3.3) |
Likewise, for W=Wn, we select a standard basis as
{D11,D21,…,Dn1}={Di1:1≤i≤n}, |
with D11=In and Di1=[Ii−1αIn−i+1], 2≤i≤n.
Based on above bases, if W=Wn, then for any Wn=(wij)n×n∈W, we have
˜Wn=(w0,wn−1,wn−2,…,w1)T. | (3.4) |
As soon as a standard basis is given, ˜Xn,˜X−n,˜Zn, and ˜Wn are uniquely determined by Xn,X−n,Zn and Wn, respectively. Thus, we can state the following definition:
Definition 3.2. We define σ1:Xn=(xij)n×n∈Un↦˜Xn, where ˜Xn is defined in (3.1), σ2:X−n=(x′ij)n×n∈U−n↦˜X−n, where ˜X−n is defined in (3.2), τ:Zn=(zij)n×n∈Vn↦˜Zn, where ˜Zn is defined in (3.3), and ϕ:Wn=(wij)n×n∈Wn↦˜Wn, where ˜Wn is defined in (3.4).
Remark 3.2. ψ,σ1,σ2,τ and ϕ are obviously invertible in the sense that for any (ν,ν1,ν2,ν3,ν4)∈Cn2×C2n−1×Cn−1×C3n−2×Cn, we have (ψ−1(ν),σ−11(ν1),σ−12(ν2),τ−1(ν3),ϕ−1(ν4))∈Cn×n×Un×U−n×Vn×Wn. It should be noted that ψ,σ1,σ2,τ and ϕ are defined on different domains.
Note that ψ(Xn) is a column vector formed by all elements of Xn, while σ1(Xn), σ2(X−n), τ(Zn), ϕ(Wn) are column vectors formed by different nonzero elements of Xn, X−n, Zn, Wn, respectively. For clarity, we denote the H-matrix in H-representation corresponding to X=Un by H1n, the H-matrix in H-representation corresponding to X=U−n by H1−n, the H-matrix in H-representation corresponding to X=Vn by H2n, the H-matrix in H-representation corresponding to X=Wn by H3n.
The following corollary is obvious from Definitions 3.1 and 3.2.
Corollary 3.1. For a n2×1 vector μ1, if ψ−1(μ1)∈Un, then there exists a (2n−1)×1 vector ν1, such that μ1=H1nν1. For a n2×1 vector μ2, if ψ−1(μ2)∈U−n, then there exists a (n−1)×1 vector ν2, such that μ2=H1−nν2. For a n2×1 vector μ3, if ψ−1(μ3)∈Vn, then there exists a (3n−2)×1 vector ν3, such that μ3=H2nν3. For a n2×1 vector μ4, if ψ−1(μ4)∈Wn, then there exists a n×1 vector ν4, such that μ4=H3nν4.
In this section, we solve Problems 1–3 via the real representation of quaternion matrices and H-representation. We first convert above least squares problems into corresponding problems of real matrix equation by using the real representation, then in order to reduce the size of original problems, we remove the redundancy and extract effective elements through H-representation. Finally, we obtain the solutions of Problems 1–3.
Lemma 4.1. [31] The least squares solutions of the linear system of equations Ax=b, with A∈Rm×n and b∈Rm can be represented as
x=A†b+(I−A†A)y, |
where, y∈Rn is an arbitrary vector. The minimal norm least squares solution of the linear system of equations Ax=b is A†b.
Lemma 4.2. [31] The linear system of equations Ax=b, with A∈Rm×n and b∈Rm has a solution x∈Rn if and only if
AA†b=b. |
When Ax=b is compatible, the general solution can be represented as
x=A†b+(I−A†A)y, |
where, y∈Rn is an arbitrary vector. Ax=b has a unique solution if and only if
rank(A)=n. |
In this case, the unique solution is x=A†b.
Theorem 4.3. Suppose A∈Qm×p,B∈Qp×n,C∈Qm×q,D∈Qq×n,E∈Qm×n be given. Hence the set TL of Problem 1 can be expressed as
TL={(X,Y)|(ˉXˉY)=H1G†1vec(→Ec)+H1(I5p+7q−8−G†1G1)y,∀y∈R5p+7q−8}, | (4.1) |
And then, the minimal norm least squares solution (XH,YA) of Problem 1 satisfies
(ˉXHˉYA)=H1G†1vec(→Ec). | (4.2) |
ˉX=[vec(X1)vec(X2)vec(X3)vec(X4)],ˉY=[vec(Y1)vec(Y2)vec(Y3)vec(Y4)],H1=[H1pH1−pH1−pH1−pH1−qH1qH1qH1q], |
G1=((→BTc⊗→A)JK,(→DTc⊗→C)J′K′)H1. | (4.3) |
Proof. For X=X1+X2i+X3j+X4k∈HTQp×p,Y=Y1+Y2i+Y3j+Y4k∈AHTQq×q, according to Lemmas 2.1–2.3, we have
‖AXB+CYD−E‖=‖→AXB+CYD−Ec‖=‖→AXBc+→CYDc−→Ec‖=‖vec(→A→X→Bc+→C→Y→Dc−→Ec)‖=‖(→BTc⊗→A)vec(→X)+(→DTc⊗→C)vec(→Y)−vec(→Ec)‖=‖(→BTc⊗→A)Jvec(→Xc)+(→DTc⊗→C)J′vec(→Yc)−vec(→Ec)‖=‖(→BTc⊗→A)JKˉX+(→DTc⊗→C)J′K′ˉY−vec(→Ec)‖=‖((→BTc⊗→A)JK,(→DTc⊗→C)J′K′)(ˉXˉY)−vec(→Ec)‖, |
where J′,K′ have the same structure with the J,K, respectively. Since X1∈Up,Xt∈U−p, Y1∈U−q,Yt∈Uq(t=2,3,4), in light of Corollary 3.1, we can derive
[vec(X1)vec(X2)vec(X3)vec(X4)vec(Y1)vec(Y2)vec(Y3)vec(Y4)]=[ψ(X1)ψ(X2)ψ(X3)ψ(X4)ψ(Y1)ψ(Y2)ψ(Y3)ψ(Y4)]=[H1pH1−pH1−pH1−pH1−qH1qH1qH1q][˜X1˜X2˜X3˜X4˜Y1˜Y2˜Y3˜Y4]. |
For the convenience of what follows, let us denote
H1=[H1pH1−pH1−pH1−pH1−qH1qH1qH1q],ˇX=[˜X1˜X2˜X3˜X4],ˇY=[˜Y1˜Y2˜Y3˜Y4], |
G1=((→BTc⊗→A)JK,(→DTc⊗→C)J′K′)H1. |
Then we can obtain
‖((→BTc⊗→A)JK,(→DTc⊗→C)J′K′)(ˉXˉY)−vec(→Ec)‖=‖((→BTc⊗→A)JK,(→DTc⊗→C)J′K′)H1(ˇXˇY)−vec(→Ec)‖=‖G1(ˇXˇY)−vec(→Ec)‖. |
Thus ‖AXB+CYD−E‖ assume its minimum value
‖AXB+CYD−E‖=min, |
if and only if ‖G1(ˇXˇY)−vec(→Ec)‖ does.
For the real matrix equation
G1(ˇXˇY)=vec(→Ec), |
by Lemma 4.1, its least squares solution can be represented as
(ˇXˇY)=G†1vec(→Ec)+(I5p+7q−8−G†1G1)y, y∈R5p+7q−8. | (4.4) |
Moreover, (4.1) is derived by multiplying both sides of (4.4) by the matrix H1. Meanwhile, (4.2) can be derived.
By virtue of Theorem 4.3, we can give the necessary and sufficient condition in order to prove the existence of tridiagonal mixed solution for the quaternion matrix equation AXB+CYD=E, and the expression for the tridiagonal mixed solution when (1.1) is compatible.
Corollary 4.4. Let A∈Qm×p,B∈Qp×n,C∈Qm×q,D∈Qq×n,E∈Qm×n be given, and G1 be defined as in (4.3).Then (1.1) has a solution X∈HTQp×p,Y∈AHTQq×q, if and only if
(G1G†1−I4mn)vec(→Ec)=0. | (4.5) |
If (4.5) holds, the solutions set of (1.1) can be represented as
ST={(X,Y)|(ˉXˉY)=H1G†1vec(→Ec)+H1(I5p+7q−8−G†1G1)y, ∀y∈R5p+7q−8}. | (4.6) |
Moreover, (1.1) has unique tridiagonal mixed solution (X′H,Y′A), if and only if
rank(G1)=5p+7q−8, |
and the unique tridiagonal mixed solution (X′H,Y′A) satisfies
(ˉX′HˉY′A)=H1G†1vec(→Ec). | (4.7) |
Proof. According to the proof of Theorem 4.3, Lemma 4.2 and the definition of Moore-Penrose generalized inverse, we have
‖AXB+CYD−E‖=‖G1(ˇXˇY)−vec(→Ec)‖=‖G1G†1G1(ˇXˇY)−vec(→Ec)‖=‖G1G†1vec(→Ec)−vec(→Ec)‖=‖(G1G†1−I4mn)vec(→Ec)‖, |
thus (1.1) has tridiagonal mixed solution (X,Y) if and only if
‖AXB+CYD−E‖=0⟺‖(G1G†1−I4mn)vec(→Ec)‖=0⟺(G1G†1−I4mn)vec(→Ec)=0. |
So we get the formula in (4.5). Under the condition that (4.5) is established, the solution (X,Y) of (1.1) satisfies
G1(ˇXˇY)=vec(→Ec). |
Moreover, the solution (X,Y) of (1.1) satisfies
(ˇXˇY)=G†1vec(→Ec)+(I5p+7q−8−G†1G1)y, ∀y∈R5p+7q−8. |
Similarly, we can deduce (4.6) by multiplying both sides of the above equation by the matrix H1. At the same time, the unique tridiagonal mixed solution (4.7) can also be obtained.
In what follows, we concentrate on Problems 2 and 3. By Theorem 3.1, for (X,Y) with special structure, we can give its H-representation matrix, which will help us extract effective elements and reduce the complexity of operations. Based on the above ideas, the following conclusions can be easily obtained.
Theorem 4.5. Suppose A∈Qm×p,B∈Qp×n,C∈Qm×q,D∈Qq×n,E∈Qm×n. Hence the set BL of Problem 2 can be represented as
BL={(X,Y)|(ˉXˉY)=H2G†2vec(→Ec)+H2(I12p+12q−16−G†2G2)y,∀y∈R12p+12q−16}, | (4.8) |
and then, the minimal norm least squares solution (XB,YB) of Problem 2 satisfies
(ˉXBˉYB)=H2G†2vec(→Ec), | (4.9) |
where
H2=[H2pH2pH2pH2pH2qH2qH2qH2q],G2=((→BTc⊗→A)JK,(→DTc⊗→C)J′K′)H2. |
Corollary 4.6. Let A∈Qm×p,B∈Qp×n,C∈Qm×q,D∈Qq×n,E∈Qm×n be given. G2 is defined in Theorem 4.5. Then (1.1) has a solution X∈BQp×p,Y∈BQq×q, if and only if
(G2G†2−I4mn)vec(→Ec)=0. | (4.10) |
If (4.10) holds, the Brownian solutions set of (1.1) can be represented as
SB={(X,Y)|(ˉXˉY)=H2G†2vec(→Ec)+H2(I12p+12q−16−G†2G2)y, ∀y∈R12p+12q−16}, |
furthermore, (1.1) has unique Brownian solution (X′B,Y′B), if and only if
rank(G2)=12p+12q−16, |
and the unique Brownian solution (X′B,Y′B) satisfies
(ˉX′BˉY′B)=H2G†2vec(→Ec). | (4.11) |
Remark 4.1. When X∈BQp×p, Y∈BQq×q, according to Theorem 4.5, we can give H2, G2, then Corollary 4.6 can be obtained by a proof method similar to Corollary 4.4.
Similar to Theorem 4.5, when X∈MQp×p, Y∈MQq×q, we can give H3, G3 for studying Problem 3.
Theorem 4.7. Suppose A∈Qm×p,B∈Qp×n,C∈Qm×q,D∈Qq×n,E∈Qm×n. Hence the set ML of Problem 3 can be expressed as
ML={(X,Y)|(ˉXˉY)=H3G†3vec(→Ec)+H3(I4p+4q−G†3G3)y, ∀y∈R4p+4q}, | (4.12) |
and then, the minimal norm least squares solution (XM,YM) of Problem 3 satisfies
(ˉXMˉYM)=H3G†3vec(→Ec), | (4.13) |
where
H3=[H3pH3pH3pH3pH3qH3qH3qH3q],G3=((→BTc⊗→A)JK,(→DTc⊗→C)J′K′)H3. |
Corollary 4.8. Let A∈Qm×p,B∈Qp×n,C∈Qm×q,D∈Qq×n,E∈Qm×n be given. G3 is defined in Theorem 4.7. Then (1.1) has a solution X∈MQp×p,Y∈MQq×q, if and only if
(G3G†3−I4mn)vec(→Ec)=0. | (4.14) |
If (4.14) holds, the Rotation solutions set of (1.1) can be expressed as
SM={(X,Y)|(ˉXˉY)=H3G†3vec(→Ec)+H3(I4p+4q−G†3G3)y, ∀y∈R4p+4q}, |
in addition, (1.1) has unique Rotation solution (X′M,Y′M), if and only if
rank(G3)=4p+4q, |
and the unique Rotation solution (X′M,Y′M) satisfies
(ˉX′MˉY′M)=H3G†3vec(→Ec). | (4.15) |
In this section, on the basis of the discussions in Section 4, we propose the algorithms of solving Problems 1–3, and then give a numerical example to prove the feasibility of the proposed algorithms.
Algorithm 5.1. (Problem 1)
(1) Input A,B,C,D,E∈Qn×n, output →BTc,→DTc,→A,→C,vec(→Ec),
(2) Input J,K,H1n,H1−n, output H1,G1,
(3) According to (4.2), calculate the minimal norm least squares solution (XH,YA) of Problem 1.
Algorithm 5.2. (Problem 2)
(1) Input A,B,C,D,E∈Qn×n, output →BTc,→DTc,→A,→C,vec(→Ec),
(2) Input J,K,H2n, output H2,G2,
(3) According to (4.9), calculate the minimal norm least squares solution (XB,YB) of Problem 2.
Algorithm 5.3. (Problem 3)
(1) Input A,B,C,D,E∈Qn×n, output →BTc,→DTc,→A,→C,vec(→Ec),
(2) Input J,K,H3n, output H3,G3,
(3) According to (4.13), calculate the minimal norm least squares solution (XM,YM) of Problem 3.
Example 5.1. Consider the quaternion matrix equation AXB+CYD=E, where
A=rand(m,p)+rand(m,p)i+rand(m,p)j+rand(m,p)k, |
B=rand(p,n)+rand(p,n)i+rand(p,n)j+rand(p,n)k, |
C=rand(m,q)+rand(m,q)i+rand(m,q)j+rand(m,q)k, |
D=rand(q,n)+rand(q,n)i+rand(q,n)j+rand(q,n)k. |
Denote Xs=Xs1+Xs2i+Xs3j+Xs4k,Ys=Ys1+Ys2i+Ys3j+Ys4k.
(i) For s=1. Then
X1=X11+X12i+X13j+X14k∈HTQp×p, |
Y1=Y11+Y12i+Y13j+Y14k∈AHTQq×q. |
Let AX1B+CY1D=E.
(ii) For s=2. Then
X2=X21+X22i+X23j+X24k∈BQp×p, |
Y2=Y21+Y22i+Y23j+Y24k∈BQq×q. |
Let AX2B+CY2D=E.
(iii) For s=3. Then
X3=X31+X32i+X33j+X34k∈MQp×p, |
Y3=Y31+Y32i+Y33j+Y34k∈MQq×q. |
Let AX3B+CY3D=E.
In all cases, the quaternion matrix Eq (1.1) have the unique solutions (XH,YA), (XB,YB), (XM,YM), respectively. Of course, for s∈{1,2,3}, (Xs,Ys) is also the minimal norm least squares solution of the quaternion matrix Eq (1.1) over X∈HTQp×p/BQp×p/MQp×p and Y∈AHTQq×q/BQq×q/MQq×q. By Algorithms 5.1–5.3, for s∈{1,2,3}, we compute (Xs′,Ys′). Let m=p=n=q=2K and the error ε=log10(‖(Xs′,Ys′)−(Xs,Ys)‖). The relation between K and the error ε is shown in Figure 1.
According to Figure 1, we obtain that the errors ε are all no more than -9 for s∈{1,2,3}, which confirms the difference between the numerical solution and the exact solution is tiny. In other words, these three figures of Figure 1 are very similar, which is consistent with the actual situation. Therefore, our proposed algorithms are very feasible.
In this paper, by combining the real representation of quaternion matrices with H-representation, we convert the least squares problem of the quaternion matrix Eq (1.1) into a corresponding problem of the real matrix equation over free variables. Then we derive the expression of the minimal norm least squares solution for the quaternion matrix Eq (1.1) over different constrained matrices as in Problems 1–3. Our resulting expressions are expressed only by real matrices, and the algorithms only involve real operations. The final example shows that our proposed method is feasible and convenient to analyze such a matrix problem with special structures.
This work was supported by National Natural Science Foundation of China under grant 62176112; the Natural Science Foundation of Shandong under grant ZR2020MA053.
The authors declare that they have no competing interests.
[1] |
J. Z. Liu, Z. H. Huang, L. Zhu, Z. J. Huang, Theorems on Schur complement of block diagonally dominant matrices and their application in reducing the order for the solution of large scale linear systems, Linear Algebra Appl., 435 (2011), 3085–3100. http://dx.doi.org/10.1016/j.laa.2011.05.023 doi: 10.1016/j.laa.2011.05.023
![]() |
[2] |
A. G. Wu, Y. M. Fu, G. R. Duan, On solutions of matrix equations v−AVF=BW and v−aˉVF=BW, Math. Comput. Model., 47 (2008), 1181–1197. http://dx.doi.org/10.1016/j.mcm.2007.06.024 doi: 10.1016/j.mcm.2007.06.024
![]() |
[3] |
H. M. Zhang, Reduced-rank gradient-based algorithms for generalized coupled Sylvester matrix equations and its applications, Comput. Math. Appl., 70 (2015), 2049–2062. http://dx.doi.org/10.1016/j.camwa.2015.08.013 doi: 10.1016/j.camwa.2015.08.013
![]() |
[4] |
M. Dehghani-Madiseh, M. Dehghan, Generalized solution sets of the interval generalized Sylvester matrix equation ∑pi=1AiXi+∑qj=1YjBj=C and some approaches for inner and outer estimations, Comput. Math. Appl., 68 (2014), 1758–1774. http://dx.doi.org/10.1016/j.camwa.2014.10.014 doi: 10.1016/j.camwa.2014.10.014
![]() |
[5] |
A. Navarra, P. L. Odell, D. M. Young, A representation of the general common solution to the matrix equations A1XB1=C1 and A2XB2=C2 with applications, Comput. Math. Appl., 41 (2001), 929–935. http://dx.doi.org/10.1016/S0898-1221(00)00330-8 doi: 10.1016/S0898-1221(00)00330-8
![]() |
[6] |
X. P. Sheng, A relaxed gradient based algorithm for solving generalized coupled Sylvester matrix equations, J. Franklin I., 355 (2018), 4282–4297. http://dx.doi.org/10.1016/j.jfranklin.2018.04.008 doi: 10.1016/j.jfranklin.2018.04.008
![]() |
[7] |
S. L. Adler, Scattering and decay theory for quaternionic quantum mechanics, and the structure of induced T nonconservation, Phys. Rev. D, 37 (1988), 3564–3662. http://dx.doi.org/10.1103/PhysRevD.37.3654 doi: 10.1103/PhysRevD.37.3654
![]() |
[8] | N. L. Bihan, S. J. Sangwine, Color image decomposition using quaternion singular value decomposition, 2003 International Conference on Visual Information Engineering VIE 2003. IET, 2003,113–116. |
[9] |
C. E. Moxey, S. J. Sangwine, T. A. Ell, Hypercomplex correlation techniques for vector images, IEEE T. Signal Proces., 51 (2003), 1941–1953. http://dx.doi.org/10.1109/TSP.2003.812734 doi: 10.1109/TSP.2003.812734
![]() |
[10] |
T. S. Jiang, L. Chen, Algebraic algorithms for least squares problem in quaternionic quantum theory, Comput. Phys. Commun., 176 (2007), 481–485. http://dx.doi.org/10.1016/j.cpc.2006.12.005 doi: 10.1016/j.cpc.2006.12.005
![]() |
[11] |
T. S. Jiang, L. Chen, An algebraic method for Schr ¨o dinger equations in quaternionic quantum mechanics, Comput. Phys. Commun., 178 (2008), 795–799. http://dx.doi.org/10.1016/j.cpc.2008.01.038 doi: 10.1016/j.cpc.2008.01.038
![]() |
[12] |
S. F. Yuan, Q. W. Wang, Two special kinds of least squares solutions for the quaternion matrix equation AXB+CXD=E, Electronic J. Linear Al., 23 (2012), 257–274. http://dx.doi.org/10.1017/is011004009jkt155 doi: 10.1017/is011004009jkt155
![]() |
[13] |
S. F. Yuan, A. P. Liao, Least squares solution of the quaternion matrix equation X−AˆXB=C with the least norm, Linear Multilinear A., 59 (2011), 985–998. http://dx.doi.org/10.1080/03081087.2010.509928 doi: 10.1080/03081087.2010.509928
![]() |
[14] |
M. H. Wang, M. S. Wei, Y. Feng, An iterative algorithm for least squares problem in quaternionic quantum theory, Comput. Phys. Commun., 179 (2008), 203–207. http://dx.doi.org/10.1016/j.cpc.2008.02.016 doi: 10.1016/j.cpc.2008.02.016
![]() |
[15] |
J. D. Gardiner, A. J. Laub, J. J. Amato, C. B. Moler, Solution of the Sylvester matrix equation AXBT+CXDT=E, ACM T. Math. Software, 18 (1992), 223–231. http://dx.doi.org/10.1145/146847.146929 doi: 10.1145/146847.146929
![]() |
[16] |
R. K. Cavin Iii, S. P. Bhattacharyya, Robust and well-conditioned eigenstructure assignment via Sylvester's equation, Optim. Contr. Appl. Met., 4 (1983), 205–212. http://dx.doi.org/10.1002/oca.4660040302 doi: 10.1002/oca.4660040302
![]() |
[17] |
G. L. Chen, J. W. Xia, G. M. Zhuang, Improved passivity analysis for neural networks with Markovian jumping parameters and interval time-varying delays, Neurocomputing, 155 (2015), 253–260. http://dx.doi.org/10.1016/j.neucom.2014.12.023 doi: 10.1016/j.neucom.2014.12.023
![]() |
[18] |
G. R. Duan, Eigenstructure assignment and response analysis in descriptor linear systems with state feedback control, Int. J. Control, 69 (1998), 663–694. http://dx.doi.org/10.1080/002071798222622 doi: 10.1080/002071798222622
![]() |
[19] |
S. K. Mitra, The matrix equation AXB+CXD=E, SIAM J. Appl. Math., 32 (1977), 823–825. http://dx.doi.org/10.1080/002071798222622 doi: 10.1080/002071798222622
![]() |
[20] |
Y. Tian, The solvability of two linear matrix equations, Linear Multilinear A., 48 (2000), 123–147. http://dx.doi.org/10.1080/03081080008818664 doi: 10.1080/03081080008818664
![]() |
[21] |
L. P. Huang, The matrix equation AXB−GXD=E over the quaternion field, Linear Algebra Appl., 234 (1996), 197–208. http://dx.doi.org/10.1016/0024-3795(94)00103-0 doi: 10.1016/0024-3795(94)00103-0
![]() |
[22] |
S. F. Yuan, Q. W. Wang, X. Zhang, Least-squares problem for the quaternion matrix equation AXB+CYD=E over different constrained matrices, Int. J. Comput. Math., 90 (2013), 565–576. http://dx.doi.org/10.1080/00207160.2012.722626 doi: 10.1080/00207160.2012.722626
![]() |
[23] |
Y. Zhang, R. H. Wang, The exact solution of a system of quaternion matrix equations involving η-Hermicity, Appl. Math. Comput., 222 (2013), 201–209. http://dx.doi.org/10.1016/j.amc.2013.07.025 doi: 10.1016/j.amc.2013.07.025
![]() |
[24] |
A. Altavilla, C. de Fabritiis, Equivalence of slice semi-regular functions via Sylvester operators, Linear Algebra Appl., 607 (2020), 151–189. http://dx.doi.org/10.1016/j.laa.2020.08.009 doi: 10.1016/j.laa.2020.08.009
![]() |
[25] |
A. Altavilla, C. de Fabritiis, Applications of the Sylvester operator in the space of slice semi-regular functions, Concr. Operators, 7 (2020), 1–12. http://dx.doi.org/10.1515/conop-2020-0001 doi: 10.1515/conop-2020-0001
![]() |
[26] | J. J. Sylvester, Sur l'equations en matrices px=xq, C. R. Acad. Sci. Paris, 99 (1884). |
[27] |
R. Bhatia, P. Rosenthal, How and why to solve the operator equation AX−XB=Y, B. Lond. Math. Soc., 29 (1997), 1–21. http://dx.doi.org/10.1112/S0024609396001828 doi: 10.1112/S0024609396001828
![]() |
[28] |
F. X. Zhang, M. S. Wei, Y. Li, J. L. Zhao, An efficient real representation method for least squares problem of the quaternion constrained matrix equation AXB+CYD=E, Int. J. Comput. Math., 98 (2021), 1408–1419. http://dx.doi.org/10.1080/00207160.2020.1821001 doi: 10.1080/00207160.2020.1821001
![]() |
[29] | J. L. Chen, X. H. Chen, Special matrices, Tsinghua University Press, 2001. |
[30] |
W. H. Zhang, B. S. Chen, H-Representation and applications to generalized Lyapunov equations and linear stochastic systems, IEEE T. Automat. Contr., 57 (2012), 3009–3022. http://dx.doi.org/10.1109/TAC.2012.2197074 doi: 10.1109/TAC.2012.2197074
![]() |
[31] | G. H. Golub, C. F. Van Loan, Matrix computations, 4 Eds., Baltimore MD: The Johns Hopkins University Press, 2013. |
1. | Xueling Fan, Ying Li, Zhihong Liu, Jianli Zhao, Solving Quaternion Linear System Based on Semi-Tensor Product of Quaternion Matrices, 2022, 14, 2073-8994, 1359, 10.3390/sym14071359 | |
2. | Xueling Fan, Ying Li, Jianhua Sun, Jianli Zhao, Solving quaternion linear system $$AXB=E$$ based on semi-tensor product of quaternion matrices, 2023, 17, 2662-2033, 10.1007/s43037-023-00251-8 |