AAA Abstract and Applied Analysis 1687-0409 1085-3375 Hindawi Publishing Corporation 831656 10.1155/2013/831656 831656 Research Article Iterative Algorithm for Solving a Class of Quaternion Matrix Equation over the Generalized (P,Q)-Reflexive Matrices http://orcid.org/0000-0001-7958-3410 Li Ning 1 http://orcid.org/0000-0002-0494-4928 Wang Qing-Wen 2 Anderson Douglas 1 School of Mathematics and Quantitative Economics Shandong University of Finance and Economics Jinan Shandong Province 250002 China sdu.edu.cn 2 Department of Mathematics Shanghai University Shanghai 200444 China shu.edu.cn 2013 21 10 2013 2013 18 04 2013 12 08 2013 2013 Copyright © 2013 Ning Li and Qing-Wen Wang. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The matrix equation l=1uAlXBl+s=1vCsXTDs=F, which includes some frequently investigated matrix equations as its special cases, plays important roles in the system theory. In this paper, we propose an iterative algorithm for solving the quaternion matrix equation l=1uAlXBl+s=1vCsXTDs=F over generalized (P,Q)-reflexive matrices. The proposed iterative algorithm automatically determines the solvability of the quaternion matrix equation over generalized (P,Q)-reflexive matrices. When the matrix equation is consistent over generalized (P,Q)-reflexive matrices, the sequence {X(k)} generated by the introduced algorithm converges to a generalized (P,Q)-reflexive solution of the quaternion matrix equation. And the sequence {X(k)} converges to the least Frobenius norm generalized (P,Q)-reflexive solution of the quaternion matrix equation when an appropriate initial iterative matrix is chosen. Furthermore, the optimal approximate generalized (P,Q)-reflexive solution for a given generalized (P,Q)-reflexive matrix X0 can be derived. The numerical results indicate that the iterative algorithm is quite efficient.

1. Introduction and Notations

Quaternion matrix equation is one of the topics of very active research in matrix theory and applications, and some important results have been developed. For example, Kyrchei  derived the explicit representation formulas for the minimum norm least squares solutions of quaternion matrix equations AX=B,  XA=B,  AXB=D. Yuan and Liao  derived the expressions of the least squares (j-self-conjugate) solution with the least norm of the quaternion matrix equation X-AX^B=C. Song et al.  obtained the expressions of the explicit solutions of quaternion matrix equations XF-AX=BY and XF-AX~=BY. He and Wang  studied the necessary and sufficient conditions for the existence of a solution to the quaternion matrix equation A1X+(A1X)η*+B1Y(B1)η*+C1Z(C1)η*=D1 and derived a general solution when the equation is solvable.

The definition of generalized (P,Q)-reflexive matrix can be found in . A complex matrix A is called (P,Q)-generalized reflexive (generalized anti-reflexive) if A=PAQ  (A=-PAQ), where P and Q are two generalized reflection matrices; that is, P=PH=P-1 and Q=QH=Q-1. The generalized (P,Q)-reflexive matrices have been widely used in engineering and scientific computations . Throughout, we use rm×n(P,Q) to denote the set of  m×n  generalized (P,Q)-reflexive quaternion matrices.

The iterative method is a very important method to solve matrix equations. Peng et al. [8, 9] constructed iteration methods to solve the symmetric and reflexive solutions of matrix equations A1XB1=C1,  A2XB2=C2. Ding and Chen [10, 11] developed iterative methods for solving a wide variety of matrix equations, based on the hierarchical identification principle . Peng  presented several efficient iterative methods to solve the constrained least squares solutions of linear matrix equations AXB=C and AXB+CYD=E, by using Paige's algorithm  as the frame method. Wang et al.  proposed iterative algorithms for solving the matrix equation AXB+CXTD=E. Xie et al.  presented a gradient based iterative algorithm for solving (1)l=1uAlXBl+s=1vCsXTDs=F. In , Li et al. proposed iterative algorithms to solve the minimal norm least squares solution to (1). Duan et al.  proposed iterative algorithms for the (Hermitian) positive definite solutions of some nonlinear matrix equations. Dehghan and Hajarian constructed iterative algorithms to solve several linear matrix equations over (anti-)reflexive [25, 26], generalized centrosymmetric [27, 28], and generalized bisymmetric [29, 30] matrices. Recently, Wu et al.  proposed iterative algorithms for solving various complex matrix equations. Wang et al.  derived an iterative method for finding the minimum-norm solution of the QLS problem in quaternionic quantum theory.

However, to our best knowledge, the generalized (P,Q)-reflexive solution of (1) over the quaternion algebra has not been considered so far. Due to the noncommutativity of , some well-known equalities for complex and real matrices no longer hold for quaternion matrices, which make the study of quaternion matrix equation more complex than that of real and complex equation. Motivated by the work mentioned previously and keeping the interests and wide applications of quaternion matrices in view (e.g., ), we, in this paper, consider an iterative method for the following two problems.

Problem 1.

For given matrices Al, Csp×n, Bl, Dsn×q, l=1,2,,u, s=1,2,,v, Fp×q, and the generalized reflection matrices Pn×n, Qn×n, find that Xrn×n(P,Q), such that (2)l=1uAlXBl+s=1vCsXTDs=F.

Problem 2.

When Problem 1 is consistent, let its solution set be denoted by SX. For a given (P,Q)-reflexive matrix X0rn×n(P,Q), find X^rn×n(P,Q), such that (3)X^-X0=minXSXX-X0.

The matrix equation (2) plays important roles in the system theory . Moreover, (2) obviously includes the matrix equations AXB+CXTD=F, AX+XTD=F, and AXB=C as special cases, which have been investigated in [17, 28, 36, 5153].

Throughout this paper, m×n and m×n represent the set of all m×n real matrices and the set of all m×n matrices over the quaternion algebra ={a1+a2i+a3j+a4ki2=j2=k2=ijk=-1,a1,a2,a3,a4}. We use AH,  AT,  A¯,  tr(A),  R(A),  Re(A), and vec(A) to denote the conjugate transpose, the transpose, the conjugate, the trace, the column space, the real part, and the mn×1 vector formed by the vertical concatenation of the respective columns of a matrix A; respectively. We use A to denote the Frobenius norm of A, that is, A=tr(AHA). The Kronecker matrix product and Hadamard matrix product of the matrices A and B are denoted by AB and AB. We use I to denote the identity matrix with the appropriate size.

2. Preliminaries

In this section, we provide some results which will play important roles in this paper.

Firstly, we recall the definition of real inner product space. Then we define a real inner product in the space m×n over the field .

Definition 3 (see [<xref ref-type="bibr" rid="B54">54</xref>]).

A real inner product space is a vector space V over the real field together with an inner product defined by a mapping (4)·,·:V×V satisfying the following three axioms for all vectors x,y,zV and all scalars a.

Symmetry: x,y=y,x;

Linearity in the first argument: ax,y=ax,y,x+y,z=x,z+y,z;

Positive definiteness: x,x>0 for all x0.

In , the following real inner product was presented to solve some complex matrix equations:(5)A,B=Re[tr(BHA)], for A,Bm×n. It is easy to verify that if we let A,Bm×n, (5) also defines a real inner product in m×n over . We denote this real inner product space by (m×n,,·,·). From the equalities (6)A,A=Re[tr(AHA)]=tr(AHA)=A,Am×n, we know that the induced matrix norm by the introduced inner product ·,· is exactly the Frobenius norm. For convenience, we still use · to denote the induced matrix norm. Let Eij denote the m×n matrix whose (i,j) entry is 1, and the other elements are zeros. In inner product space (m×n,,·,·), it is easy to verify that Eij, Eiji, Eijj,  Eijk,  i=1,2,m,  j=1,2,n, is an orthonormal basis, which reveals that the dimension of the inner product space (m×n,,·,·) is 4mn.

Next, we introduce the well-known real representation of a quaternion matrix. For an arbitrary quaternion matrix M=M1+M2i+M3j+M4km×n, a map ϕ(·), from m×n to 4m×4n, can be defined as (7)ϕ(M)=[M1-M2-M3-M4M2M1-M4M3M3M4M1-M2M4-M3M2M1]. Let M and N be two arbitrary quaternion matrices with appropriate size. We know that ϕ(·) satisfies the following properties: (8)M=Nϕ(M)=ϕ(N),(9)ϕ(M+N)=ϕ(M)+ϕ(N),ϕ(MN)=ϕ(M)ϕ(N),ϕ(kM)=kϕ(M),k,(10)ϕ(MH)=ϕT(M). It follows a simple verification that ϕ(·) also satisfies the following two properties which is useful for some deduction in this paper: (11)ϕ(M)=2M,(12)ϕ(MT)=2ϕT(M)W-ϕT(M), where (13)W=[E0000E0000E0000E] and E is a matrix whose elements are all equal to one with the same size as M.

Finally, we recall some results about the commutation matrix. A commutation matrix P(m,n) is a mn×mn matrix which has the following explicit form: (14)P(m,n)=i=1mj=1nEijEijT=[EijT]i=1,,mj=1,,n. Moreover, P(m,n) is a permutation matrix and P(m,n)=PT(n,m)=P-1(n,m). We have the following lemmas on the commutation matrix.

Lemma 4 (see [<xref ref-type="bibr" rid="B24">55</xref>]).

Let A be an m×n matrix. There is a commutation matrix P(m,n) such that (15)vec(AT)=P(m,n)vec(A).

Lemma 5 (see [<xref ref-type="bibr" rid="B24">55</xref>]).

Let A be an m×n matrix and B a p×q matrix. There exist two commutation matrices P(m,p) and P(n,q) such that (16)BA=PT(m,p)(AB)P(n,q).

3. An Iterative Algorithm for Solving Problem <xref ref-type="statement" rid="problem1.1">1</xref>

In this section, we firstly construct an algorithm for solving Problem 1. Then we show that if the problem is consistent, the sequence {X(k)} generated by the introduced algorithm converges to a solution of Problem 1 within at most 4pq iteration steps.

Algorithm 6.

( 1 ) Input matrices Al,Csp×n,  Bl,  Dsn×q,  l=1,2,,u,  s=1,2,,v,  Fp×q,   generalized reflection matrices Pn×n,  Qn×n,  X(1)rn×n(P,Q).

( 2 ) Calculate (17)R(1)=F-l=1uAlX(1)Bl-s=1vCsXT(1)Ds;S(1)=12(l=1uAlHR(1)BlH+s=1v(CsHR(1)DsH)T+l=1uPAlHR(1)BlHQ+s=1vP(CsHR(1)DsH)TQ);T(1)=S(1)k:=1.

( 3 ) If R(k)=0, or R(k)0 and T(k)=0, stop; otherwise k:=k+1.

( 4 ) Calculate (18)X(k)=X(k-1)+R(k-1)2T(k-1)2T(k-1);R(k)=R(k-1)-R(k-1)2T(k-1)2(l=1uAlT(k-1)Bl+s=1vCsTT(k-1)Ds);S(k)=12(l=1uAlHR(k)BlH+s=1v(CsHR(k)DsH)T+l=1uPAlHR(k)BlHQ+s=1vP(CsHR(k)DsH)TQ)T(k)=S(k)+R(k)2R(k-1)2T(k-1).

( 5 ) Go to Step (3).

The following Lemmas show some properties of Algorithm 6.

Lemma 7.

The sequences {R(i)},  {S(i)}, and {T(i)} generated by Algorithm 6 satisfy (19)Re{tr[RH(j+1)R(i)]}=Re{tr[RH(j)R(i)]}-R(j)2T(j)2Re{tr[TH(j)S(i)]},(20)Re{tr[TH(j+1)T(i)]}=R(j+1)2R(j)2Re{tr[TH(j)T(i)]}+T(i)2R(i)2(Re{tr[RH(j+1)R(i)]}-Re{tr[RH(j+1)R(i+1)]}),fori,j=1,2,.

Proof.

One has the following: (21)Re{tr[RH(j+1)R(i)]}=Re{tr[((l=1uAlT(j)Bl+s=1vCsTT(j)Ds))HR(j)-R(j)2T(j)2×(l=1uAlT(j)Bl+s=1vCsTT(j)Ds))HR(i)]}=Re{tr[RH(j)R(i)]}-R(j)2T(j)2×Re{tr[l=1uBlHTH(j)AlHR(i)+s=1vDsHT(j)¯CsHR(i)]}=Re{tr[RH(j)R(i)]}-R(j)2T(j)2×Re{tr[TH(j)(l=1uAlHR(i)BlH+s=1v(CsHR(i)DsH)T)]}=Re{tr[RH(j)R(i)]}-R(j)2T(j)2×Re{tr[l=1uAlHR(i)BlHTH(j)×(l=1uAlHR(i)BlH+s=1v(CsHR(i)DsH)T+l=1uPAlHR(i)BlHQ+s=1vP(CsHR(i)DsH)TQl=1uAlHR(i)BlH)×2-1tr[l=1uAlHR(i)BlHTH(j)]}=Re{tr[RH(j)R(i)]}-R(j)2T(j)2Re{tr[TH(j)S(i)]}.

Therefore, the conclusion (19) holds. Consider (22)Re{tr[TH(j+1)T(i)]}=Re{tr[(SH(j+1)+R(j+1)2R(j)2TH(j))T(i)]}=R(j+1)2R(j)2Re{tr[TH(j)T(i)]}+Re{tr[+s=1vP(CsHR(j+1)DsH)TQ)HT12(l=1uAlHR(j+1)BlH+s=1v(CsHR(j+1)DsH)T+l=1uPAlHR(j+1)BlHQ+s=1vP(CsHR(j+1)DsH)TQ)H×T(i)tr[+s=1vP(CsHR(j+1)DsH)TQ)HT12(l=1uAlHR(j+1)BlH]}=R(j+1)2R(j)2Re{tr[TH(j)T(i)]}+Re{tr[TH(i)(l=1uAlHR(j+1)BlH+s=1v(CsHR(j+1)DsH)Tl=1uAlHR)]}=R(j+1)2R(j)2Re{tr[TH(j)T(i)]}+Re{tr[l=1uRH(j+1)AlT(i)Bl+s=1vT(i)¯CsHR(j+1)DsHl=1uRH]}=R(j+1)2R(j)2Re{tr[TH(j)T(i)]}+Re{tr[RH(j+1)(l=1uAlT(i)Bl+s=1vCsTT(i)DsRH(j+1)(l=1uAlT(i)Bl)]}=R(j+1)2R(j)2Re{tr[TH(j)T(i)]}+T(i)2R(i)2Re{tr[RH(j+1)(R(i)-R(i+1))]}=R(j+1)2R(j)2Re{tr[TH(j)T(i)]}+T(i)2R(i)2(Re{tr[RH(j+1)R(i)]}-Re{tr[RH(j+1)R(i+1)]}).

Therefore, the conclusion (20) holds.

Lemma 8.

Assume that the sequences {R(i)} and {T(i)} are generated by Algorithm 6, and then R(i),R(j)=0 and T(i),T(j)=0 for i,j=1,2,,ij.

Proof.

Since R(i),R(j)=R(j),R(i) and T(i),T(j)=T(j),T(i) for i,j=1,2,, we only need to prove that R(i),R(j)=0 and T(i),T(j)=0 for 1i<j.

Now we prove this conclusion by induction.

First, we prove (23)R(i),R(i+1)=0,T(i),T(i+1)=0for  i=1,2,. When i=1, from Lemma 7 we have (24)R(1),R(2)=Re{tr[RH(2)R(1)]}=Re{tr[RH(1)R(1)]}-R(1)2T(1)2Re{tr[TH(1)S(1)]}=R(1)2-R(1)2T(1)2T(1)2=0. From Lemma 7 and conclusion (24), we have (25)T(1),T(2)=Re{tr[TH(2)T(1)]}=R(2)2R(1)2Re{tr[TH(1)T(1)]}+T(1)2R(1)2({tr[RH(2)R(2)]}Re{tr[RH(2)R(1)]}-Re{tr[RH(2)R(2)]})=R(2)2R(1)2T(1)2-T(1)2R(1)2R(2)2=0. Now assume that conclusion (23) holds for 1it-1; from Lemma 7, we have (26)R(t),R(t+1)=Re{tr[RH(t+1)R(t)]}=Re{tr[RH(t)R(t)]}-R(t)2T(t)2Re{tr[TH(t)S(t)]}=Re{tr[RH(t)R(t)]}-R(t)2T(t)2×Re{tr[TH(t)(T(t)-R(t)2R(t-1)2T(t-1))]}=0. From Lemma 7 and conclusion (26), it can also be obtained that (27)T(t),T(t+1)=Re{tr[TH(t+1)T(t)]}=R(t+1)2R(t)2Re{tr[TH(t)T(t)]}+T(t)2R(t)2(Re{tr[RH(t+1)R(t)]}-Re{tr[RH(t+1)R(t+1)]})=R(t+1)2R(t)2T(t)2-T(t)2R(t)2R(t+1)2=0. Therefore, by the principle of induction, conclusion (23) holds.

Next, assume that R(i),R(i+r)=0 and T(i),T(i+r)=0 for i1 and r1. We will prove that (28)R(i),R(i+r+1)=0,T(i),T(i+r+1)=0. In order to prove conclusion (28), we first prove that (29)R(1),R(r+2)=0,T(1),T(r+2)=0. From Lemma 7 we have (30)R(1),R(r+2)=Re{tr[RH(r+2)R(1)]}=Re{tr[RH(r+1)R(1)]}-R(r+1)2T(r+1)2Re{tr[TH(r+1)S(1)]}=Re{tr[RH(r+1)R(1)]}-R(r+1)2T(r+1)2Re{tr[TH(r+1)T(1)]}=0. From Lemma 7 and conclusion (30), we have (31)T(1),T(r+2)=Re{tr[TH(r+2)T(1)]}=R(r+2)2R(r+1)2Re{tr[TH(r+1)T(1)]}+T(1)2R(1)2(Re{tr[RH(r+2)R(1)]}-Re{tr[RH(r+2)R(2)]})=0. Also we have (32)R(i),R(i+r+1)=Re{tr[RH(i+r+1)R(i)]}=Re{tr[RH(i+r)R(i)]}-R(i+r)2T(i+r)2Re{tr[TH(i+r)S(i)]}=-R(i+r)2T(i+r)2Re{R(i)2R(i-1)2tr[R(i)2R(i-1)2TH(i+r)×(T(i)-R(i)2R(i-1)2×T(i-1)R(i)2R(i-1)2)]}=R(i+r)2R(i)2T(i+r)2R(i-1)2×Re{tr[TH(i+r)T(i-1)]},T(i),T(i+r+1)=Re{tr[TH(i+r+1)T(i)]}=R(i+r+1)2R(i+r)2Re{tr[TH(i+r)T(i)]}+T(i)2R(i)2(Re{tr[RH(i+r+1)R(i)]}-Re{tr[RH(i+r+1)×R(i+1)RH]})=T(i)2R(i)2Re{tr[RH(i+r+1)R(i)]}=T(i)2R(i+r)2R(i)2R(i)2T(i+r)2R(i-1)2×Re{tr[TH(i+r)T(i-1)]}. Repeating the previous process (32), we have (33)R(i),R(i+r+1)==αRe{tr[TH(r+2)T(1)]},T(i),T(i+r+1)==βRe{tr[TH(r+2)T(1)]}. Combining these two relations with (30) and (31), it implies that (28) holds. So, by the principle of induction, we know that Lemma 8 holds.

Lemma 9.

When Problem 1 is consistent, let X~rn×n(P,Q) be its solution; then, for any generalized (P,Q)-reflexive initial matrix X(1)rn×n(P,Q), the sequences {R(i)}, {T(i)} and {X(i)} generated by Algorithm 6 satisfy (34)T(i),X~-X(i)=R(i)2,i=1,2,.

Proof.

When i=1, it follows from Algorithm 6 that (35)T(1),X~-X(1)=Re{tr[(X~-X(1))HT(1)]}=Re{tr[s=1vP(CsHR(1)DsH)TQ(X~-X(1))H×(l=1uAlHR(1)BlH+s=1v(CsHR(1)DsH)T+l=1uPAlHR(1)BlHQ+s=1vP(CsHR(1)DsH)TQ)×2-1tr[s=1vP(CsHR(1)DsH)TQ(X~-X(1))Hs=1vP(CsHR(1)DsH)TQ]}=Re{tr[(X~-X(1))H(l=1uAlHR(1)BlH+s=1v(CsHR(1)DsH)T(l=1uAlHR(1)BlH)(l=1uAlHR(1)BlH]}=Re{tr[s=1vCs(X~T-XT(1)))RH(1)(s=1vCs(X~T-XT(1)))Dsl=1uAl(X~-X(1))Bl+s=1vCs(X~T-XT(1)))Dstr[s=1vCs(X~T-XT(1)))RH(1)(s=1vCs(X~T-XT(1)))Dsl=1uAl(X~-X(1))Bltr[s=1vCs(X~T-XT(1)))RH(1)(s=1vCs(X~T-XT(1)))Dsl=1uAl(X~-X(1))Bl]}=Re{tr[RH(1)R(1)]}=R(1)2. This implies that (34) holds for i=1.

Assume that (34) holds for i=t, and then, when i=t+1, (36)T(t+1),X~-X(t+1)=Re{tr[(X~-X(t+1))HT(t+1)]}=Re{R(t+1)2R(t)2T(t)tr[+R(t+1)2R(t)2T(t))(X~-X(t+1))H×(12(l=1uAlHR(t+1)BlH+s=1v(CsHR(t+1)DsH)T+l=1uPAlHR(t+1)BlHQ+s=1vP(CsHR(t+1)DsH)TQ)+R(t+1)2R(t)2T(t))]}=Re{tr[s=1v(X~-X(t+1))H×(l=1uAlHR(t+1)BlH+s=1v(CsHR(t+1)DsH)T)]}+R(t+1)2R(t)2Re{tr[(X~-X(t+1))HT(t)]}=Re{tr[s=1vCs(X~T-XT(t+1))RH(t+1)×(l=1uAl(X~-X(t+1))Bl+s=1vCs(X~T-XT(t+1))Ds)]}+R(t+1)2R(t)2×{{tr[(X~-X(t))HT(t)]}Re{tr[(X~-X(t))HT(t)]}-Re{tr[(X(t+1)-X(t))HT(t)]}}=R(t+1)2+R(t+1)2R(t)2{R(t)2T(t)2R(t)2-R(t)2T(t)2Re{tr[TH(t)T(t)]}}=R(t+1)2. Therefore, Lemma 9 holds by the principle of induction.

Remark 10.

Lemma 9 implies that if there exists a positive integer i such that R(i)0 and T(i)=0, then Problem 1 is not consistent. Therefore, the solvability of Problem 1 can be determined by Algorithm 6 in the absence of round-off errors.

From Lemmas 8 and 9, we have the following theorem.

Theorem 11.

Assume that Problem 1 is consistent, and then using Algorithm 6 for any generalized (P,Q)-reflexive initial matrix X(1)rn×n(P,Q), a solution of Problem 1 can be obtained within finite iteration steps in the absence of round-off errors.

Proof.

If R(i)0,i=1,2,,4pq, from Lemma 9 we can obtain T(i)0,i=1,2,,4pq. Thus R(4pq+1) and T(4pq+1) can be computed by Algorithm 6. According to Lemma 8, we have R(i),R(j)=0, i,j=1,2,,4pq. Considering that the inner product space (p×q,,·,·) is 4pq-dimensional, we know that the set of R(1),R(2),,R(4pq) is an orthogonal basis of the inner product space (p×q,,·,·). Also from Lemma 8, we have R(i),R(4pq+1)=0 for i=1,2,,4pq, which implies that R(4pq+1)=0; that is, X(4pq+1) is a solution of Problem 1.

4. The Solution of Problem <xref ref-type="statement" rid="problem1.2">2</xref>

In this section, we will prove that the sequence {X(k)} generated by Algorithm 6 converges to the least Frobenius norm generalized (P,Q)-reflexive solution of the quaternion matrix equation l=1uAlXBl+s=1vCsXTDs=F when an appropriate initial iterative matrix is chosen. Then, we solve Problem 2 by finding the least Frobenius norm generalized (P,Q)-reflexive solution of a new constructed quaternion matrix equation.

Lemma 12 (see [<xref ref-type="bibr" rid="B33">51</xref>]).

Assume that the consistent system of linear equations My=b has a solution y0R(MT); then, y0 is the unique least Frobenius norm solution of the system of linear equations.

Lemma 13.

If Problem 1 is consistent, then the following system of real linear equations is consistent: (37)[(l=1uϕT(Bl)ϕ(Al)-s=1v(ϕT(Ds)ϕ(Cs))P(4n,4n)+2s=1v(ϕT(Ds)ϕ(Cs))diag(vec(W))P(4n,4n))(l=1uϕT(Bl)ϕ(Q)ϕ(Al)ϕ(P)-s=1v(ϕT(Ds)ϕ(P)ϕ(Cs)ϕ(Q))P(4n,4n)+2s=1v(ϕT(Ds)ϕ(Cs))×diag(vec(W))(ϕ(P)ϕ(Q))P(4n,4n))]Y=[vec(ϕ(F))vec(ϕ(F))].

Furthermore, if the solution sets of Problem 1 and (37) are denoted by SX and SY, respectively, then, (38)vec(ϕ(SX))SY.

Proof.

If Problem 1 is consistent, let X be a solution of Problem 1, then we have l=1uAlXBl+s=1vCsXTDs=F and PXQ=X, which implies that X is a solution of quaternion matrix equations (39)l=1uAlXBl+s=1vCsXTDs=F,l=1uAlPXQBl+s=1vCs(PXQ)TDs=F.

By (8), (9), and (12), we can derive that the quaternion matrix equations (39) is equivalent to the following real matrix equations: (40)l=1uϕ(Al)ϕ(X)ϕ(Bl)-s=1vϕ(Cs)ϕT(X)ϕ(Ds)+2s=1vϕ(Cs)(ϕT(X)W)ϕ(Ds)=ϕ(F),l=1uϕ(Al)ϕ(P)ϕ(X)ϕ(Q)ϕ(Bl)-s=1vϕ(Cs)ϕ(Q)ϕT(X)ϕ(P)ϕ(Ds)+2s=1vϕ(Cs)((ϕ(Q)ϕT(X)ϕ(P))W)ϕ(Ds)=ϕ(F).

By the operation properties of vec(·),  , , and Lemma 4, it is not difficult to verify that matrix equations (40) is equivalent to(41)[(l=1uϕT(Bl)ϕ(Al)-s=1v(ϕT(Ds)ϕ(Cs))P(4n,4n)+2s=1v(ϕT(Ds)ϕ(Cs))diag(vec(W))P(4n,4n))(l=1uϕT(Bl)ϕ(Q)ϕ(Al)ϕ(P)-s=1v(ϕT(Ds)ϕ(P)ϕ(Cs)ϕ(Q))P(4n,4n)+2s=1v(ϕT(Ds)ϕ(Cs))×diag(vec(W))(ϕ(P)ϕ(Q))P(4n,4n))]×vec(ϕ(X))=[vec(ϕ(F))vec(ϕ(F))],which implies that vec(ϕ(X)) is a solution of (37). So system of real linear equations (37) is consistent.

From the previous procedure, the proof of (38) is trivial.

Theorem 14.

If X˚ is a solution of Problem 1 and X˚ can be expressed as the following form: (42)X˚=l=1uAlHGBlH+s=1v(CsHGDsH)T+l=1uPAlHGBlHQ+s=1vP(CsHGDsH)TQ,Gp×q then, X˚ is the least Frobenius norm solution of Problem 1.

Proof.

By (8), (9), (10), and (12) and Lemmas 4 and 5, we have(43)vec(ϕ(X˚))=vec(l=1uϕT(Al)ϕ(G)ϕT(Bl)+2s=1v(ϕ(Ds)ϕT(G)ϕ(Cs))W-s=1vϕ(Ds)ϕT(G)ϕ(Cs)+l=1uϕ(P)ϕT(Al)ϕ(G)ϕT(Bl)ϕ(Q)+2s=1vϕ(P)((ϕ(Ds)ϕT(G)ϕ(Cs))W)ϕ(Q)-s=1vϕ(P)ϕ(Ds)ϕT(G)ϕ(Cs)ϕ(Q))=[l=1uϕ(Bl)ϕT(Al)+2s=1vdiag(vec(W))(ϕT(Cs)ϕ(Ds))P(4p,4q)-s=1v(ϕT(Cs)ϕ(Ds))P(4p,4q),l=1uϕ(Q)ϕ(Bl)ϕ(P)ϕT(Al)+2s=1v(ϕ(Q)ϕ(P))diag(vec(W))(ϕT(Cs)ϕ(Ds))P(4p,4q)-s=1v(ϕ(Q)ϕT(Cs)ϕ(P)ϕ(Ds))P(4p,4q)][vec(ϕ(G))vec(ϕ(G))]=[(l=1uϕT(Bl)ϕ(Al)-s=1vϕT(Ds)ϕ(Cs)P(4n,4n)+2s=1v(ϕT(Ds)ϕ(Cs))diag(vec(W))P(4n,4n))(l=1uϕT(Bl)ϕ(Q)ϕ(Al)ϕ(P)-s=1vϕT(Ds)ϕ(P)ϕ(Cs)ϕ(Q)P(4n,4n)+2s=1v(ϕT(Ds)ϕ(Cs))diag(vec(W)(ϕ(P)ϕ(Q)))P(4n,4n))]T×[vec(ϕ(G))vec(ϕ(G))]R[(l=1uϕT(Bl)ϕ(Al)-s=1vϕT(Ds)ϕ(Cs)P(4n,4n)+2s=1v(ϕT(Ds)ϕ(Cs))diag(vec(W))P(4n,4n))(l=1uϕT(Bl)ϕ(Q)ϕ(Al)ϕ(P)-s=1vϕT(Ds)ϕ(P)ϕ(Cs)ϕ(Q)P(4n,4n)+2s=1v(ϕT(Ds)ϕ(Cs))diag(vec(W)(ϕ(P)ϕ(Q)))P(4n,4n))]T.By Lemma 12, vec(ϕ(X˚)) is the least Frobenius norm solution of (37).

Noting (11), we derive from Lemma 13 that X˚ is the least Frobenius norm solution of Problem 1.

Remark 15.

From Algorithm 6, we can derive that if we choose X(1)=l=1uAlHGBlH+s=1v(CsHGDsH)T+l=1uPAlHGBlHQ+s=1vP(CsHGDsH)TQ,  Gp×q, as the initial iteration matrix, then all X(k) generated by Algorithm 6 have a form of (44)X(k)=l=1uAlHGkBlH+s=1v(CsHGkDsH)T+l=1uPAlHGkBlHQ+s=1vP(CsHGkDsH)TQ,Gkp×q.

Theorem 16.

When Problem 1 is consistent, if we choose the following matrix as the initial iteration matrix (45)X(1)=l=1uAlHGBlH+s=1v(CsHGDsH)T+l=1uPAlHGBlHQ+s=1vP(CsHGDsH)TQ, where Gp×q is arbitrary, or especially, X(1)=0, then the solution X*, generated by Algorithm 6, is exactly the least Frobenius norm solution of Problem 1.

From conclusion (44) and Theorem 14, the proof of Theorem 16 is trivial.

Now we consider Problem 2. When Problem 1 is consistent, the solution set of Problem 1  SX is not empty. For a given generalized (P,Q)-reflexive matrix X0rn×n(P,Q), let X˙=X-X0,  F˙=F-l=1uAlX0Bl-s=1vCsX0TDs; then the least Frobenius norm generalized (P,Q)-reflexive solution of the following quaternion matrix equation is exactly the solution of minXSXX-X0(46)l=1uAlX˙Bl+s=1vCsX˙TDs=F˙. According to Theorem 16, if we choose the initial iteration matrix X˙(1)=l=1uAlHGBlH+s=1v(CsHGDsH)T+l=1uPAlHGBlHQ+s=1vP(CsHGDsH)TQ, where Gp×q is arbitrary, or especially, choosing    X˙(1)=0, the least Frobenius norm generalized (P,Q)-reflexive solution X˙* of (46) can be obtained. Then the solution of Problem 2 is X^=X˙*+X0.

5. Examples

In this section, we give two examples to illustrate the performance of the proposed algorithm.

Example 17.

Consider the quaternion matrix equation (47)A1XB1+C1XTD1+A2XB2+C2XTD2=F, with the matrices(48)A1=[2+i+2j1+2i+2j+k3j2i+j+2k2i+2j+k-1+3k1+2i+4j+2k4+i+2j+4k],A2=[i+j+ki1-2i+j+2k2+4i+2k2i+2j+k1+3i+j3+4i+j-2k2+2i+2j+k],B1=[1+5i+2j+k2+3i+3j+2k6i+4j-k2+5j2+2i+j-k4+7i+2j+2k3+j-k1+2i-2j+3k],B2=[-i+4j+k-3+2i+3j+2k2+4i-k-1+5i+2j-2  k1+6i+4j+5k2-3i+j+k-i+6j+3k1+7i-2j-4k],C1=[i+2j+k1+i-3j+2k1+5i+k2i+4j-4k1-4i+2j+k5+5i-j+k-4+2i+3j+2k1+5i+2j-4k],C2=[1+i+3j3-i+3j+k1+i+5j+3k2+4i+2j5+3i+j+4k2+i+2j+3k3+4i+3j+2k-2+i-3j+k],D1=[2+6i+6j+k1+3i+j-2k-4i+3j+8k-2-2i-4j+5k5+3i-2j+4k1+2i+j+2k3-i+2j-2k3+i-j+3k],D2=[i-5j+6k-2-8i-4j-3k3+4i+7j+2k5i+2j5+7i-2j-k-2+9i+7j+k1+3i+4j6+2j+7k],F=[6+i+2j+k-8-7i-3j+2ki-6j-4k3-2i+6k]. We apply Algorithm 6 to find the generalized (P,Q)-reflexive solution of (47), where (49)P=[0.2800.96k000.280-0.96j-0.96k0-0.28000.96j0-0.28],Q=[0.2800.96k00-100-0.96k0-0.280000-1] are two generalized reflection matrices. Choosing an initial matrix (50)X(1)=[100000.3600.48j00100-0.48j00.64]r4×4(P,Q), we obtain that the sequence {X(k)} after 21 steps is(51)X*=X(21)=[0.51783+0.066965i+0.095973j-0.025639k-0.088601+0.063970i+0.075293j+0.15420k-0.13818-0.050488i-0.073437j+0.11425k0.12924-0.077969i-0.12809j+0.12008k0.026384+0.045766i-0.22964j+0.066304k-0.20560-0.10039i+0.085294j-0.11813k0.12175+0.14209i-0.11665j+0.20690k-0.17079-0.16010i-0.17233j-0.10396k-0.026384+0.045766i-0.22964j-0.066304k-0.034186+0.089643i-0.0029795j+0.11576k-0.14209-0.12175i+0.20690j-0.11665k-0.080702+0.12324i+0.20981j+0.11867k0.55651+0.066993i-0.069276j-0.041030k-0.15435+0.0039727i+0.11952j-0.045581k0.070202-0.070130i-0.031364j+0.0024134k0.27974-0.15822i+0.10760j+0.16431k],with R(21)=2.047×10-13. The obtained results are presented in Figure 1, where r(k)=R(k).

The curve for the Frobenius norm of the residuals from Example 17.

Example 18.

Let(52)X0=[0.64-0.72i-0.75k0.48k-0.5625i-0.96k0.64+0.36i00.48j+0.48k0-0.48k1-0.96j0.361.28-0.75j0.48j+0.48k00.64+0.36i0]r4×4(P,Q).Now we consider Problem 2, that is, finding the optimal approximate generalized (P,Q)-reflexive solution of the quaternion matrix equation (47) to X0. Applying Algorithm 6 to (46) by choosing the initial matrix X˙(1)=0, we obtain that the sequence {X(k)} after 22 steps is(53)X˙*=X˙(22)=[-0.60797+0.14851i-0.031273j+0.032158k0.093923+0.086087i+0.24350j+0.44225k-0.32942-0.34127i-0.091587j+0.36047k-0.092590-0.15547i-0.19614j+0.067121k-0.087200-0.13538i+0.090255j+0.25660k-0.58966-0.32467i+0.11478j+0.12523k0.020667+0.14412i-0.17405j+0.058232k-0.26152-0.089494i+0.12345j-0.20729k0.087200-0.13538i+0.090255j-0.25660k-0.021779+0.31853i-0.55293j+0.58717k-0.14412-0.020667i+0.058232j-0.17405k0.090225-0.057613i-0.17618j-0.066567k-0.45828-0.20116i-0.047696j+0.083025k-0.78290+0.73724i+0.42471j-0.029038k-0.30730-0.22789i-0.27640j+0.079531k-0.23490+0.088756i-0.12030j-0.076818k],with R˙(22)=2.1855×10-14. The obtained results are presented in Figure 2, where r(k)=R˙(k).

So we can obtain the solution of Problem 2; that is,(54)X^=X˙*+X0=[0.032031+0.14851i-0.031273j+0.032158k0.093923-0.63391i+0.24350j-0.30775k0.31058+0.018728i-0.091587j+0.36047k-0.092590-0.15547i-0.19614j+0.067121k-0.087200-0.13538i+0.090255j-0.22340k0.41034-0.32467i-0.84522j+0.12523k0.020667+0.14412i+0.30595j+0.53823k-0.26152-0.089494i+0.12345j-0.20729k0.087200-0.13538i+0.090255j+0.22340k-0.021779-0.24397i-0.55293j-0.37283k-0.14412-0.020667i+0.53823j+0.30595k0.090225-0.057613i-0.17618j-0.066567k-0.098285-0.20116i-0.047696j+0.083025k0.49710+0.73724i-0.32529j-0.029038k0.33270+0.13211i-0.27640j+0.079531k-0.23490+0.088756i-0.12030j  -0.076818k].

The curve for the Frobenius norm of the residuals from Example 18.

6. Conclusions

In this paper, an algorithm has been presented for solving the quaternion matrix equation l=1uAlXBl+s=1vCsXTDs=F over generalized (P,Q)-reflexive matrices. It has been proven that when the quaternion matrix equation l=1uAlXBl+s=1vCsXTDs=F is consistent over generalized (P,Q)-reflexive matrices, for any generalized (P,Q)-reflexive initial iterative matrix, the sequence {X(k)} generated by Algorithm 6 will converge to a generalized (P,Q)-reflexive solution within finite iteration steps in the absence of round-off errors. We have also proven that by choosing a suitable initial iterative matrix, the sequence {X(k)} will converge to the least Frobenius norm generalized (P,Q)-reflexive solution of the quaternion matrix equation l=1uAlXBl+s=1vCsXTDs=F. Then, by using Algorithm 6, we solved Problem 2. The numerical results have shown that the presented algorithm is quite effective.

Acknowledgments

This research was supported by the grants from the Innovation Program of Shanghai Municipal Education Commission (13ZZ080), the National Natural Science Foundation of China (11171205), and the Natural Science Foundation of Shanghai (11ZR1412500).

Kyrchei I. Explicit representation formulas for the minimum norm least squares solutions of some quaternion matrix equations Linear Algebra and its Applications 2013 438 1 136 152 10.1016/j.laa.2012.07.049 MR2993371 ZBL1255.15022 Yuan S. Liao A. Least squares solution of the quaternion matrix equation X-AX^B=C with the least norm Linear and Multilinear Algebra 2011 59 9 985 998 10.1080/03081087.2010.509928 MR2826067 ZBL1228.65072 Song C. Chen G. Wang X. On solutions of quaternion matrix equations XF-AX=BY and XF-AX˜=BY Acta Mathematica Scientia B 2012 32 5 1967 1982 10.1016/S0252-9602(12)60153-2 MR2960763 He Z.-H. Wang Q.-W. A real quaternion matrix equation with applications Linear and Multilinear Algebra 2013 61 6 725 740 10.1080/03081087.2012.703192 MR3005652 Chen H.-C. Sameh A. H. A matrix decomposition method for orthotropic elasticity problems SIAM Journal on Matrix Analysis and Applications 1989 10 1 39 64 10.1137/0610004 MR976151 ZBL0669.73010 Chen H.-C. Generalized reflexive matrices: special properties and applications SIAM Journal on Matrix Analysis and Applications 1998 19 1 140 153 10.1137/S0895479895288759 MR1609944 ZBL0910.15005 Datta L. Morgera S. D. On the reducibility of centrosymmetric matrices—applications in engineering problems Circuits, Systems, and Signal Processing 1989 8 1 71 96 10.1007/BF01598746 MR998028 Peng Z.-h. Hu X.-y. Zhang L. An efficient algorithm for the least-squares reflexive solution of the matrix equation A1XB1=C1, A2XB2=C2 Applied Mathematics and Computation 2006 181 2 988 999 10.1016/j.amc.2006.01.071 MR2269978 ZBL1115.65048 Peng Y.-X. Hu X.-Y. Zhang L. An iterative method for symmetric solutions and optimal approximation solution of the system of matrix equations A1XB1=C1, A2XB2=C2 Applied Mathematics and Computation 2006 183 2 1127 1137 10.1016/j.amc.2006.05.124 MR2294070 ZBL1134.65032 Ding F. Chen T. Gradient based iterative algorithms for solving a class of matrix equations IEEE Transactions on Automatic Control 2005 50 8 1216 1221 10.1109/TAC.2005.852558 MR2156053 Ding F. Chen T. On iterative solutions of general coupled matrix equations SIAM Journal on Control and Optimization 2006 44 6 2269 2284 10.1137/S0363012904441350 MR2248183 ZBL1115.65035 Ding F. Chen T. Hierarchical gradient-based identification of multivariable discrete-time systems Automatica 2005 41 2 315 325 10.1016/j.automatica.2004.10.010 MR2157667 ZBL1073.93012 Peng Z.-Y. A matrix LSQR iterative method to solve matrix equation AXB=C International Journal of Computer Mathematics 2010 87 8 1820 1830 10.1080/00207160802516875 MR2665753 ZBL1195.65056 Peng Z.-y. New matrix iterative methods for constraint solutions of the matrix equation AXB=C Journal of Computational and Applied Mathematics 2010 235 3 726 735 10.1016/j.cam.2010.07.001 MR2719811 ZBL1206.65145 Peng Z.-Y. Solutions of symmetry-constrained least-squares problems Numerical Linear Algebra with Applications 2008 15 4 373 389 10.1002/nla.578 MR2413779 ZBL1212.65181 Paige C. C. Bidiagonalization of matrices and solutions of the linear equations SIAM Journal on Numerical Analysis 1974 11 197 209 MR0341842 10.1137/0711019 Wang M. Cheng X. Wei M. Iterative algorithms for solving the matrix equation AXB+CXTD=E Applied Mathematics and Computation 2007 187 2 622 629 10.1016/j.amc.2006.08.169 MR2323068 ZBL1121.65048 Xie L. Ding J. Ding F. Gradient based iterative solutions for general linear matrix equations Computers & Mathematics with Applications 2009 58 7 1441 1448 10.1016/j.camwa.2009.06.047 MR2555281 ZBL1189.65083 Li Z.-Y. Wang Y. Zhou B. Duan G.-R. Least squares solution with the minimum-norm to general matrix equations via iteration Applied Mathematics and Computation 2010 215 10 3547 3562 10.1016/j.amc.2009.10.052 MR2578938 ZBL1186.65047 Duan X. Liao A. Tang B. On the nonlinear matrix equation X-i=1mA*iXδiAi=Q Linear Algebra and its Applications 2008 429 1 110 121 10.1016/j.laa.2008.02.014 MR2419144 ZBL1148.15012 Duan X. Liao A. On the nonlinear matrix equation X+A*X-qA=Q(q1) Mathematical and Computer Modelling 2009 49 5-6 936 945 10.1016/j.mcm.2008.10.009 MR2495010 ZBL1165.15302 Duan X. Liao A. On Hermitian positive definite solution of the matrix equation X-i=1mA*iXrAi=Q Journal of Computational and Applied Mathematics 2009 229 1 27 36 10.1016/j.cam.2008.10.018 MR2522496 ZBL1170.15005 Duan X. Li C. Liao A. Solutions and perturbation analsis for the nonlinear matrix equation X+i=1mAi*X-1Ai=I Applied Mathematics and Computation 2011 218 8 4458 4466 10.1016/j.amc.2011.10.026 MR2862116 Duan X. Liao A. On the existence of Hermitian positive definite solutions of the matrix equation Xs+A*X-tA=Q Linear Algebra and its Applications 2008 429 4 673 687 10.1016/j.laa.2008.03.019 MR2428122 ZBL1143.15011 Dehghan M. Hajarian M. Finite iterative algorithms for the reflexive and anti-reflexive solutions of the matrix equation A1X1B1+A2X2B2=C Mathematical and Computer Modelling 2009 49 9-10 1937 1959 10.1016/j.mcm.2008.12.014 MR2532100 ZBL1171.15310 Dehghan M. Hajarian M. Two iterative algorithms for solving coupled matrix equations over reflexive and anti-reflexive matrices Computational & Applied Mathematics 2012 31 2 353 371 10.1590/S1807-03022012000200008 MR2991771 ZBL1262.65054 Dehghan M. Hajarian M. Analysis of an iterative algorithm to solve the generalized coupled Sylvester matrix equations Applied Mathematical Modelling 2011 35 7 3285 3300 10.1016/j.apm.2011.01.022 MR2785279 ZBL1227.65037 Hajarian M. Dehghan M. The generalized centro-symmetric and least squares generalized centro-symmetric solutions of the matrix equation AYB+CYTD=E Mathematical Methods in the Applied Sciences 2011 34 13 1562 1579 10.1002/mma.1459 MR2850542 ZBL1228.65066 Dehghan M. Hajarian M. The general coupled matrix equations over generalized bisymmetric matrices Linear Algebra and its Applications 2010 432 6 1531 1552 10.1016/j.laa.2009.11.014 MR2580446 ZBL1187.65042 Dehghan M. Hajarian M. An iterative method for solving the generalized coupled Sylvester matrix equations over generalized bisymmetric matrices Applied Mathematical Modelling. Simulation and Computation for Engineering and Environmental Systems 2010 34 3 639 654 10.1016/j.apm.2009.06.018 MR2563346 ZBL1185.65054 Wu A.-G. Zeng X. Duan G.-R. Wu W.-J. Iterative solutions to the extended Sylvester-conjugate matrix equations Applied Mathematics and Computation 2010 217 1 130 142 10.1016/j.amc.2010.05.029 MR2672571 ZBL1223.65032 Wu A.-G. Feng G. Duan G.-R. Wu W.-J. Iterative solutions to coupled Sylvester-conjugate matrix equations Computers & Mathematics with Applications 2010 60 1 54 66 10.1016/j.camwa.2010.04.029 MR2651883 ZBL1198.65083 Wu A.-G. Li B. Zhang Y. Duan G.-R. Finite iterative solutions to coupled Sylvester-conjugate matrix equations Applied Mathematical Modelling 2011 35 3 1065 1080 10.1016/j.apm.2010.07.053 MR2740850 ZBL1211.15024 Wu A.-G. Lv L. Hou M.-Z. Finite iterative algorithms for a common solution to a group of complex matrix equations Applied Mathematics and Computation 2011 218 4 1191 1202 10.1016/j.amc.2011.05.096 MR2831627 ZBL1229.65071 Wu A.-G. Lv L. Hou M.-Z. Finite iterative algorithms for extended Sylvester-conjugate matrix equations Mathematical and Computer Modelling 2011 54 9-10 2363 2384 10.1016/j.mcm.2011.05.045 MR2834638 ZBL1235.65040 Wang M. Wei M. Feng Y. An iterative algorithm for least squares problem in quaternionic quantum theory Computer Physics Communications 2008 179 4 203 207 10.1016/j.cpc.2008.02.016 MR2585244 ZBL1197.81138 Zhang F. Quaternions and matrices of quaternions Linear Algebra and its Applications 1997 251 21 57 10.1016/0024-3795(95)00543-9 MR1421264 ZBL0873.15008 Le Bihan N. Mars J. Singular value decomposition of quaternion matrices: a new tool for vector-sensor signal processing Signal Processing 2004 84 7 1177 1199 2-s2.0-2642566022 10.1016/j.sigpro.2004.04.001 Farid F. O. Wang Q.-W. Zhang F. On the eigenvalues of quaternion matrices Linear and Multilinear Algebra 2011 59 4 451 473 10.1080/03081081003739204 MR2802526 ZBL1237.15016 Took C. C. Mandic D. P. Zhang F. On the unitary diagonalisation of a special class of quaternion matrices Applied Mathematics Letters 2011 24 11 1806 1809 10.1016/j.aml.2011.04.038 MR2812216 Sangwine S. J. Le Bihan N. Quaternion singular value decomposition based on bidiagonalization to a real or complex matrix using quaternion Householder transformations Applied Mathematics and Computation 2006 182 1 727 738 10.1016/j.amc.2006.04.032 MR2292081 ZBL1109.65037 De Leo S. Scolarici G. Right eigenvalue equation in quaternionic quantum mechanics Journal of Physics A 2000 33 15 2971 2995 10.1088/0305-4470/33/15/306 MR1766502 ZBL0954.81008 Wang Q.-W. Liu X. Yu S.-W. The common bisymmetric nonnegative definite solutions with extreme ranks and inertias to a pair of matrix equations Applied Mathematics and Computation 2011 218 6 2761 2771 10.1016/j.amc.2011.08.018 MR2838182 ZBL1267.15017 Wang Q.-W. Zhou Y. Zhang Q. Ranks of the common solution to six quaternion matrix equations Acta Mathematicae Applicatae Sinica 2011 27 3 443 462 10.1007/s10255-011-0083-9 MR2810515 Wang Q.-W. Chang H.-X. Lin C.-Y. P -(skew)symmetric common solutions to a pair of quaternion matrix equations Applied Mathematics and Computation 2008 195 2 721 732 10.1016/j.amc.2007.05.021 MR2381251 ZBL1149.15011 Wang Q.-W. van der Woude J. W. Chang H.-X. A system of real quaternion matrix equations with applications Linear Algebra and its Applications 2009 431 12 2291 2303 10.1016/j.laa.2009.02.010 MR2563022 ZBL1180.15019 Wang Q. Yu S. Xie W. Extreme ranks of real matrices in solution of the quaternion matrix equation AXB=C with applications Algebra Colloquium 2010 17 2 345 360 MR2654356 ZBL1188.15016 Fletcher L. R. Kautsky J. Nichols N. K. Eigenstructure assignment in descriptor systems IEEE Transactions on Automatic Control 1986 AC-31 12 1138 1141 2-s2.0-0022863919 Dai L. Singular Control Systems 1989 118 Berlin, Germany Springer x+332 Lecture Notes in Control and Information Sciences 10.1007/BFb0002475 MR986970 Frank P. M. Fault diagnosis in dynamic systems using analytical and knowledge-based redundancy: a survey and some new results Automatica 1990 26 3 459 474 2-s2.0-0025421966 10.1016/0005-1098(90)90018-D 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 Applied Mathematics and Computation 2005 160 3 763 777 10.1016/j.amc.2003.11.030 MR2113117 ZBL1068.65056 Xie L. Liu Y. Yang H. Gradient based and least squares based iterative algorithms for matrix equations AXB+CXTD=F Applied Mathematics and Computation 2010 217 5 2191 2199 10.1016/j.amc.2010.07.019 MR2727965 ZBL1210.65097 Piao F. Zhang Q. Wang Z. The solution to matrix equation AX+XTC=B Journal of the Franklin Institute 2007 344 8 1056 1062 10.1016/j.jfranklin.2007.05.002 MR2389931 ZBL1171.15015 Zhang X. D. Matrix Analysis and Applications 2004 Beijing, China Tsinghua University Press Horn R. A. Johnson C. R. Topics in Matrix Analysis 1991 Cambridge, UK Cambridge University Press viii+607 10.1017/CBO9780511840371 MR1091716