We solve optimization problems on the ranks and inertias of the quadratic Hermitian matrix function Q-XPX* subject to a consistent system of matrix equations AX=C and XB=D. As applications, we derive necessary and sufficient conditions for the solvability to the systems of matrix equations and matrix inequalities AX=C,XB=D, and XPX*=(>,<,≥,≤)Q in the Löwner partial ordering to be feasible, respectively. The findings of this paper widely extend the known results in the literature.
1. Introduction
Throughout this paper, we denote the complex number field by ℂ. The notations ℂm×n and ℂhm×m stand for the sets of all m×n complex matrices and all m×m complex Hermitian matrices, respectively. The identity matrix with an appropriate size is denoted by I. For a complex matrix A, the symbols A* and r(A) stand for the conjugate transpose and the rank of A, respectively. The Moore-Penrose inverse of A∈ℂm×n, denoted by A†, is defined to be the unique solution X to the following four matrix equations
(1)(1)AXA=A,(2)XAX=X,(3)(AX)*=AX,(4)(XA)*=XA.
Furthermore, LA and RA stand for the two projectors LA=I-A†A and RA=I-AA† induced by A, respectively. It is known that LA=LA* and RA=RA*. For A∈ℂhm×m, its inertia
(2)𝕀n(A)=(i+(A),i-(A),i0(A))
is the triple consisting of the numbers of the positive, negative, and zero eigenvalues of A, counted with multiplicities, respectively. It is easy to see that i+(A)+i-(A)=r(A). For two Hermitian matrices A and B of the same sizes, we say A>B(A≥B) in the Löwner partial ordering if A-B is positive (nonnegative) definite.
The investigation on maximal and minimal ranks and inertias of linear and quadratic matrix function is active in recent years (see, e.g., [1–24]). Tian [21] considered the maximal and minimal ranks and inertias of the Hermitian quadratic matrix function
(3)h(X)=AXBX*A*+AXC+C*X*A*+D,
where B and D are Hermitian matrices. Moreover, Tian [22] investigated the maximal and minimal ranks and inertias of the quadratic Hermitian matrix function
(4)f(X)=Q-XPX*
such that AX=C.
The goal of this paper is to give the maximal and minimal ranks and inertias of the matrix function (4) subject to the consistent system of matrix equations
(5)AX=C,XB=D,
where Q∈ℂhn×n, P∈ℂhp×p are given complex matrices. As applications, we consider the necessary and sufficient conditions for the solvability to the systems of matrix equations and inequality
(6)AX=C,XB=D,XPX*=Q,AX=C,XB=D,XPX*>Q,AX=C,XB=D,XPX*<Q,AX=C,XB=D,XPX*≥Q,AX=C,XB=D,XPX*≤Q,
in the Löwner partial ordering to be feasible, respectively.
2. The Optimization on Ranks and Inertias of (4) Subject to (5)
In this section, we consider the maximal and minimal ranks and inertias of the quadratic Hermitian matrix function (4) subject to (5). We begin with the following lemmas.
Lemma 1 (see [3]).
Let A∈ℂhm×m, B∈ℂm×p, and C∈ℂq×m be given and denote
(7)P1=[ABB*0],P2=[AC*C0],P3=[ABC*B*00],P4=[ABC*C00].
Then
(8)maxY∈ℂp×qr[A-BYC-(BYC)*]=min{r[ABC*],r(P1),r(P2)},minY∈ℂp×qr[A-BYC-(BYC)*]=2r[ABC*]+max{w++w-,g++g-,w++g-,w-+g+},maxY∈ℂp×qi±[A-BYC-(BYC)*]=min{i±(P1),i±(P2)},minY∈ℂp×qi±[A-BYC-(BYC)*]=r[ABC*]+max{i±(P1)-r(P3),i±(P2)-r(P4)},
where
(9)w±=i±(P1)-r(P3),g±=i±(P2)-r(P4).
Lemma 2 (see [4]).
Let A∈ℂm×n, B∈ℂm×k, C∈ℂl×n, D∈ℂm×p, E∈ℂq×n, Q∈ℂm1×k, and P∈ℂl×n1 be given. Then
(10)(1)r(A)+r(RAB)=r(B)+r(RBA)=r[AB],(2)r(A)+r(CLA)=r(C)+r(ALC)=r[AC],(3)r(B)+r(C)+r(RBALC)=r[ABC0],(4)r(P)+r(Q)+r[ABLQRPC0]=r[AB0C0P0Q0],(5)r[RBALCRBDELC0]+r(B)+r(C)=r[ADBE00C00].
Lemma 3 (see [23]).
Let A∈ℂhm×m, B∈ℂm×n, C∈ℂhn×n, Q∈ℂm×n, and P∈ℂp×n be given, and, T∈ℂm×m be nonsingular. Then
(11)(1)i±(TAT*)=i±(A),(2)i±[A00C]=i±(A)+i±(C),(3)i±[0QQ*0]=r(Q),(4)i±[ABLPLPB*0]+r(P)=i±[AB0B*0P*0P0].
Lemma 4.
Let A, C, B, and D be given. Then the following statements are equivalent.
System (5) is consistent.
Let
(12)r[AC]=r(A),[DB]=r(B),AD=CB.
In this case, the general solution can be written as
(13)X=A†C+LADB†+LAVRB,
where V is an arbitrary matrix over ℂ with appropriate size.
Now we give the fundamental theorem of this paper.
Theorem 5.
Let f(X) be as given in (4) and assume that AX=C and XB=D in (5) is consistent. Then
(14)maxAX=C,XB=Dr(Q-XPX*)=min{n+r[0PPAQCP0-D*0B*]-r(A)-r(B)-r(P),2n+r(AQA*-CPC*)-2r(A),r[Q0D0-PBD*B*0]-2r(B)},minAX=C,XB=Dr(Q-XPX*)=2n+2r[0PPAQCP0-D*0B*]-2r(A)-2r(B)-r(P)+max{s++s-,t++t-,s++t-,s-+t+},maxAX=C,XB=Di±(Q-XPX*)=min{[Q0D0-PBD*B*0]n+i±(AQA*-CPC*)-r(A),i±[Q0D0-PBD*B*0]-r(B)},(15)minAX=C,XB=Di±(Q-XPX*)=n+r[0PPAQCP0-D*0B*]-r(A)-r(B)-i±(P)+max{s±,t±},
where
(16)s±=-n+r(A)-i∓(P)+i±(AQA*-CPC*)-r[CPAQA*B*D*A*],t±=-n+r(A)-i∓(P)+i±[Q0D0-PBD*B*0]-[0PBAQCP0D*B*0].
Proof.
It follows from Lemma 4 that the general solution of (4) can be expressed as
(17)X=X0+LAVRB,
where V is an arbitrary matrix over ℂ and X0 is a special solution of (5). Then
(18)Q-XPX*=Q-(X0+LAVRB)P(X0+LAVRB)*.
Note that
(19)r[Q-(X0+LAVRB)P(X0+LAVRB)*]=r[Q(X0+LAVRB)PP(X0+LAVRB)*P]-r(P)=r[[QX0PPX0*P]+[LA0]V[0RBP]+([LA0]V[0RBP])*]-r(P),(20)i±[Q-(X0+LAVRB)P(X0+LAVRB)*]=i±[Q(X0+LAVRB)PP(X0+LAVRB)*P]-i±(P)=i±[[QX0PPX0*P]+[LA0]V[0RBP]+([LA0]V[0RBP])*]-i±(P).
Let
(21)q(V)=[QX0PPX0*P]+[LA0]V[0RBP]+([LA0]V[0RBP])*.
Applying Lemma 1 to (19) and (20) yields
(22)maxVr[q(V)]=min{r(M),r(M1),r(M2)},minVr[q(V)]=2r(M)+max{s++s-,t++t-,s++t-,s-+t+},maxVi±[q(V)]=min{i±(M1),i±(M2)},minVi±[q(V)]=r(M)+max{s±,t±},
where
(23)M=[QX0PLA0PX0*P0PRB],M1=[QX0PLAPX0*P0LA00],M2=[QX0P0PX0*PPRB0RBP0],M3=[QX0PLA0PX0*P0PRBLA000],M4=[QX0PLA0PX0*P0PRB0RBP00],s±=i±(M1)-r(M3),t±=i±(M2)-r(M4).
Applying Lemmas 2 and 3, elementary matrix operations and congruence matrix operations, we obtain
(24)r(M)=n+r[0PPAQCP0-D*0B*]-r(A)-r(B),r(M1)=2n+r(AQA*-CPC*)-2r(A)+r(P),i±(M1)=n+i±(AQA*-CPC*)-r(A)+i±(P),r(M2)=r[Q0D0-PBD*B*0]-2r(B)+r(P),i±(M2)=i±[Q0D0-PBD*B*0]-r(B)+i±(P),r(M3)=2n+r(P)-2r(A)-r(B)+r[CPAQA*B*D*A*],r(M4)=n+r(P)+r[0PBAQCP0D*B*0]-2r(B)-r(A).
Substituting (24) into (22), we obtain the results.
Using immediately Theorem 5, we can easily get the following.
Theorem 6.
Let f(X) be as given in (4), s± and let t± be as given in Theorem 5 and assume that AX=C and XB=D in (5) are consistent. Then we have the following.
AX=C and XB=D have a common solution such that Q-XPX*≥0 if and only if
(25)n+r[0PPAQCP0-D*0B*]-r(A)-r(B)-i-(P)+s-≤0,n+r[0PPAQCP0-D*0B*]-r(A)-r(B)-i-(P)+t-≤0.
AX=C and XB=D have a common solution such that Q-XPX*≤0 if and only if
(26)n+r[0PPAQCP0-D*0B*]-r(A)-r(B)-i+(P)+s+≤0,n+r[0PPAQCP0-D*0B*]-r(A)-r(B)-i+(P)+t+≤0.
AX=C and XB=D have a common solution such that Q-XPX*>0 if and only if
(27)i+(AQA*-CPC*)-r(A)≥0,i+[Q0D0-PBD*B*0]-r(B)≥n.
AX=C and XB=D have a common solution such that Q-XPX*<0 if and only if
(28)i-(AQA*-CPC*)-r(A)≥0,i-[Q0D0-PBD*B*0]-r(B)≥n.
All common solutions of AX=C and XB=D satisfy Q-XPX*≥0 if and only if
(29)n+i-(AQA*-CPC*)-r(A)=0,or,i-[Q0D0-PBD*B*0]-r(B)=0.
All common solutions of AX=C and XB=D satisfy Q-XPX*≤0 if and only if
(30)n+i+(AQA*-CPC*)-r(A)=0,or,i+[Q0D0-PBD*B*0]-r(B)=0.
All common solutions of AX=C and XB=D satisfy Q-XPX*>0 if and only if
(31)n+r[0PPAQCP0-D*0B*]-r(A)-r(B)-i+(P)+s+=n,
or
(32)n+r[0PPAQCP0-D*0B*]-r(A)-r(B)-i+(P)+t+=n.
All common solutions of AX=C and XB=D satisfy Q-XPX*<0 if and only if
(33)n+r[0PPAQCP0-D*0B*]-r(A)-r(B)-i-(P)+s-=n,
or
(34)n+r[0PPAQCP0-D*0B*]-r(A)-r(B)-i-(P)+t-=n.
AX=C, XB=D, and Q=XPX* have a common solution if and only if
(35)2n+2r[0PPAQCP0-D*0B*]-2r(A)-2r(B)-r(P)+s++s-≤0,2n+2r[0PPAQCP0-D*0B*]-2r(A)-2r(B)-r(P)+t++t-≤0,2n+2r[0PPAQCP0-D*0B*]-2r(A)-2r(B)-r(P)+s++t-≤0,2n+2r[0PPAQCP0-D*0B*]-2r(A)-2r(B)-r(P)+s-+t+≤0.
Let P=I in Theorem 5, we get the following corollary.
Corollary 7.
Let Q∈ℂn×n, A, B, C, and D be given. Assume that (5) is consistent. Denote
(36)T1=[CAQB*D*],T2=AQA*-CC*,T3=[QDD*B*B],T4=[CAQA*B*D*A*],T5=[CBAQB*BD*].
Then,
(37)maxAX=C,XB=Dr(Q-XX*)=min{n+r(T1)-r(A)-r(B),2n+r(T2)-2r(A),n+r(T3)-2r(B)},minAX=C,XB=Dr(Q-XX*)=2r(T1)+max{r(T2)-2r(T4),-n+r(T3)-2r(T5),i+(T2)+i-(T3)-r(T4)-r(T5),-n+i-(T2)+i+(T3)-r(T4)-r(T5)}maxAX=C,XB=Di+(Q-XX*)=min{n+i+(T2)-r(A),i+(T3)-r(B)},maxAX=C,XB=Di-(Q-XX*)=min{n+i-(T2)-r(A),n+i-(T3)-r(B)},minAX=C,XB=Di+(Q-XX*)=r(T1)+max{i+(T2)-r(T4),i+(T3)-n-r(T5)},minAX=C,XB=Di-(Q-XX*)=r(T1)+max{i-(T2)-r(T4),i-(T3)-r(T5)}.
Remark 8.
Corollary 7 is one of the results in [24].
Let B and D vanish in Theorem 5, then we can obtain the maximal and minimal ranks and inertias of (4) subject to AX=C.
Corollary 9.
Let f(X) be as given in (4) and assume that AX=C is consistent. Then
(38)maxAX=Cr(Q-XPX*)=min{n+r[AQCP]-r(A)-r(B),2n+r(AQA*-CPC*)-2r(A),r(Q)+r(P)}minAX=Cr(Q-XPX*)=2n+2r[AQCP]-2r(A)+max{s++s-,t++t-,s++t-,s-+t+},maxAX=Ci±(Q-XPX*)=min{n+i±(AQA*-CPC*)-r(A),i±(Q)+i∓(P)},minAX=Ci±(Q-XPX*)=n+r[AQCP]-r(A)+i∓(P)+max{s±,t±},
where
(39)s±=-n+r(A)-i∓(P)+i±(AQA*-CPC*)-r[CPAQA*],t±=-n+r(A)+i±(Q)-r(P)-[AQCP].
Remark 10.
Corollary 9 is one of the results in [22].
ChuD. L.HungY. S.WoerdemanH. J.Inertia and rank characterizations of some matrix expressions20093131187122610.1137/080712945MR2558819HeZ. H.WangQ. W.Solutions to optimization problems on ranks and inertias of a matrix function with applications201221962989300110.1016/j.amc.2012.09.024MR2991998LiuY.TianY.Max-min problems on the ranks and inertias of the matrix expressions A-BXC±(BXC)* with applications2011148359362210.1007/s10957-010-9760-8MR2769881ZBL1223.90077MarsagliaG.StyanG. P. H.Equalities and inequalities for ranks of matrices1974/752269292MR038484010.1080/03081087408817070ZhangX.WangQ. W.LiuX.Inertias and ranks of some Hermitian matrix functions with applications201210132935110.2478/s11533-011-0117-9MR2863801WangQ. W.ZhouY.ZhangQ.Ranks of the common solution to six quaternion matrix equations201127344346210.1007/s10255-011-0083-9MR2810515ZhangQ.WangQ. W.The (P,Q)-(skew)symmetric extremal rank solutions to a system of quaternion matrix equations2011217229286929610.1016/j.amc.2011.04.011MR2804035ZBL1222.15022LianD. Z.WangQ. W.TangY.Extreme ranks of a partial banded block quaternion matrix expression subject to some matrix equations with applications2011182333346MR2785788ZBL1227.15015ZhangH. S.WangQ. W.Ranks of submatrices in a general solution to a quaternion system with applications201148596999010.4134/BKMS.2011.48.5.969MR2867185ZBL1229.15017WangQ. W.JiangJ.Extreme ranks of (skew-)Hermitian solutions to a quaternion matrix equation201020552573MR2735973ZBL1207.15016KhanI. A.WangQ. W.SongG. J.Minimal ranks of some quaternion matrix expressions with applications201021752031204010.1016/j.amc.2010.07.004MR2727951ZBL1204.15005WangQ.YuS.XieW.Extreme ranks of real matrices in solution of the quaternion matrix equation AXB=C with applications2010172345360MR2654356ZBL1188.15016WangQ. W.ZhangH. S.SongG. J.A new solvable condition for a pair of generalized Sylvester equations200918289301MR2519916ZBL1190.15019WangQ. W.YuS. W.ZhangQ.The real solutions to a system of quaternion matrix equations with applications20093762060207910.1080/00927870802317590MR2530763WangQ. W.LiC. K.Ranks and the least-norm of the general solution to a system of quaternion matrix equations20094305-61626164010.1016/j.laa.2008.05.031MR2490703ZBL1158.15010WangQ. W.ZhangH. S.YuS. W.On solutions to the quaternion matrix equation AXB+CYD=E200817343358MR2430860ZBL1154.15019WangQ. W.SongG. J.LiuX.Maximal and minimal ranks of the common solution of some linear matrix equations over an arbitrary division ring with applications2009162293308MR2502772ZBL1176.15020WangQ. W.SongG. J.LinC. Y.Rank equalities related to the generalized inverse AT,S(2) with applications2008205137038210.1016/j.amc.2008.08.016MR2466641ZBL1159.15006WangQ. W.YuS. W.LinC. Y.Extreme ranks of a linear quaternion matrix expression subject to triple quaternion matrix equations with applications2008195273374410.1016/j.amc.2007.05.018MR2381252ZBL1149.15012WangQ. W.SongG. J.LinC. Y.Extreme ranks of the solution to a consistent system of linear quaternion matrix equations with an application200718921517153210.1016/j.amc.2006.12.039MR2332107ZBL1124.15010TianY.Formulas for calculating the extremum ranks and inertias of a four-term quadratic matrix-valued function and their applications2012437383585910.1016/j.laa.2012.03.021MR2921740ZBL1252.15026TianY.Solving optimization problems on ranks and inertias of some constrained nonlinear matrix functions via an algebraic linearization method201275271773410.1016/j.na.2011.09.003MR2847451ZBL1236.65070TianY.Maximization and minimization of the rank and inertia of the Hermitian matrix expression A−BX−(BX)* with applications2011434102109213910.1016/j.laa.2010.12.010MR2781681ZBL1211.15022WangQ. W.ZhangX.HeZ. H.On the Hermitian structures of the solution to a pair of matrix equations2013611739010.1080/03081087.2012.663370