International Mathematical Forum, Vol. 7, 2012, no. 17, 831 - 838 Distribution for the Standard Eigenvalues of Quaternion Matrices Shahid Qaisar College of Mathematics and Statistics, Chongqing University Chongqing, 401331, P. R. China [email protected] Limin Zou College of Mathematics and Statistics, Chongqing University Chongqing, 401331, P. R. China [email protected] Abstract This paper aims to estimate and locate the standard eigenvalues of quaternion matrices. We prove that all the standard eigenvalues of a central closed matrix are located in one Gerˇsgorin ball. We shall conclude the paper with some numerical examples which will show the effectiveness of our new results. Mathematics Subject Classification: 15A18; 15A42; 15A60 Keywords: Quaternion matrices, Standard eigenvalues, Gerˇsgorin ball 1 Introduction Quaternion was introduced by the Irish mathematician Hamilton (1805-1865) in 1843[1]. Hamilton probably never thought that one day in the future the quaternions he had invented would be used in computer graphics, control the- ory, physics, and mechanics. In some differential equations problems involving the long-term stability of an oscillating system, one is sometimes interested in showing that the eigen- values of a quaternion matrix all lie in the left half-plane. And sometimes in statistics or numerical analysis one needs to show that a Hermitian quaternion matrix is positive definite. It is well known that the main obstacle in the study of quaternion matrices is the non -commutative multiplication of quaternions. So, it is not easy to 832 S. Qaisar and L.Zou study quaternion algebra problems. However, quaternion algebra theory is be- coming more and more important in recent years, and the study of quaternion matrices is still developing because of the above application [2-4]. Throughout the paper, we use a set of notation and terminology. R and C are the set of the real and complex numbers, respectively. Let H be the normed vector space and skew field of real quaternions: H = {a = a0 + a1i + a2j + a3k, a0,a1,a2,a3 ∈ R} . It is common to use H for quaternion in honor of the inventor, Hamilton. For ∗ any a = a0 + a1i + a2j + a3k ∈ H, the conjugate√ √ of a is a = a = a0 − a1i − 2 2 2 2 1/2 a2j − a3k and the modulus of a is |a| = aa¯ = aa¯ =(a0 + a1 + a2 + a3) . Let Hn×n and Hn×1 be the collections of all n × n matrices with entries in H and n-column vectors, respectively. Let In and H (n, u) be the collections of all n × n unit matrices with entries in H and quaternion unitary matrix, n×1 T T respectively. For X ∈ H ,X is the transpose of X.IfX =(x1,x2 ...xn) , − − − − T ∗ − − − then X =(x1, x2,...,xn) is the conjugate of X and X =(x1, x2,...,√xn)is the conjugate transpose of X. The modulus of X is defined to be |X| = X∗X. For an n × n matrix A =(aij)n×n (aij ∈ H), the conjugate transpose of A is − the n × n matrix A∗ = AT . The well-known Gerˇsgorin theorem is one of fundamental theorems in com- plex matrix theory. It guarantees that all the eigenvalues of a matrix are lo- cated in the union of n Gerˇsgorin disks. Recently, the Gerˇsgorin theorem has been generalized to real quaternion division algebra [3, 5]. In this paper, we prove that all the standard eigenvalues of a central closed matrix are located in one Gerˇsgorin ball. To do this, we need the following definitions and lemmas. 2 Some definitions and lemmas Definition 2.1[6] Suppose that a is a given real quaternion and R is a positive real number. The set G = {z ∈ H : |z − a|≤R} is said to be a Gerˇsgorin ball with the center a and the radiusR. n×n Definition 2.2 LetA =(aij)n×n ∈ H . The Frobinus norm of A is defined by ⎛ ⎞ 1 n n 2 ⎝ 2⎠ ∗ AF = |aij| = Tr(A A).. i=1 j=1 Lemma 2.3[4] Any n × n quaternion matrix A has exact n right eigenvalues which are complex numbers with nonnegative imaginary parts. These eigenvalues are said to be the standard eigenvalues of A. n×n Lemma 2.4[4] Let A =(aij)n×n ∈ H . Then there exists a unitary Standard eigenvalues of quaternion matrices 833 matrix U such that U ∗AU is an upper triangular matrix with diagonal entries h1 + k1i, ···,hn + kni, where the ht + kti’s are the standard eigenvalues of A, i.e., kt ≥ 0,t=1, 2, ···,n. We denote the set of the standard eigenvalues of A by λ (A). The upper triangular matrix is said to be the Schur upper triangular matrix of A. n×n Lemma 2.5[7] LetA =(aij)n×n ∈ H . Suppose that the standard eigen- values of A are λ1,λ2 , ···,λn. Then n 2 4 ∗ 2 2 2 |λi| ≤ AF −AA F + A F . i=1 n×n Lemma 2.6[4] If A =(aij)n×n ∈ H is in triangular form, then every diagonal entries is a right eigenvalue of A. Definition 2.7[8] A quaternion matrix A is said to be a central closed matrix if there exists a invertible matrix P such that −1 P AP = diag (λ1,λ2, ···,λn) , where λi’s are real numbers. By Lemma 2.3 and Lemma 2.6, we see that λi’s are standard eigenvalues of A. Recall that the trace of a square complex matrix is the sum of main diagonal entries of the matrix, which is equal to the sum of the eigenvalues of the matrix. This is not always true in quaternion case. Example 2.8 Let 1 i A = . jk √ 1+ 3 1 Then√ the trace of A is√ 1 + k.√ The standard eigenvalues of A are λ = 2 + −1+ 3 1− 3 1+ 3 2 i and λ2 = 2 + 2 i. Obviously, Tr(A) = λ1 + λ2. Fortunately we can show that the trace of central closed matrix is equal to the sum of the standard eigenvalues of the matrix. Lemma 2.9[8] Let A be an n × n central closed matrix and suppose that the standard eigenvalues of A are λ1,λ2, ···,λn. Then n Tr(A)= λi. i=1 It turns out that if A and B are n × n complex matrices, then Tr(AB)= Tr(BA). However, quaternion matrices are different from complex matrices. Example 2.10 Let i 0 k 0 A = and B = . Then Tr(AB) = Tr(BA). 0 i 0 k 834 S. Qaisar and L.Zou The following is the generalization of the result Tr(AB)=Tr(BA)of complex matrices to quaternion matrices, which will be used in the distribution for the standard eigenvalues of quaternion matrices. n×n n×n Lemma 2.11 Let A =(aij)n×n ∈ H and B =(bij)n×n ∈ H . Then Re (Tr(AB)) = Re (Tr(BA)) . For quaternion matrices, we know that left row rank is equal to right column rank. Both of them are equal to rank (A). It is easy to show that if an n × n central closed matrix A hassnon-zero standard eigenvalues, then rank (A) ≥ s. Lemma 2.12 Let A be an n × n central closed matrix and suppose that the standard eigenvalues of A are λ1,λ2, ···,λn and λ1,λ2, ···,λs = 0 . Then 2 4 ∗ 2 2 2 |Tr(A)| ≤ rank (A) · AF −AA F + A F . Proof. By Lemma 2.9, we have s 2 | |2 Tr(A) = λi . i=1 Using Lemma 2.5, it is easy to see the following s 2 s s ≤ · | |2 ≤ · | |2 ≤ · 4 − ∗2 22 λi s λi rank (A) λi rank (A) A F AA F + A F . i=1 i=1 i=1 So the lemma is proved. Lemma 2.13 Let A be an n × n central closed matrix and λ be a standard ∗ 2 2 2 eigenvalue of A. Suppose that F (A)=AA F −A F , then F (A)=F (λI − A) . Proof. It is easy to see that F (A) ∈ R, λ ∈ R and F (λI − A) ∈ R. Since ∗ 2 2 2 F (λI − A)=(λI − A)(λI − A) − (λI − A) F F ∗ ∗ = Tr((λI − A)(λI − A) (λI − A)(λI − A) ) ∗ ∗ −Tr((λI − A) (λI − A) (λI − A)(λI − A)) , we have F (λI − A)=F (A)+Tr(2λ (A∗AA + A∗A∗A) − λ (A∗AA∗ + AAA∗ + AA∗A∗ + AA∗A)) . By Lemma 2.11, we have Re (F (λI − A)) = Re (F (A)) , that is F (A)=F (λI − A) . So the lemma is proved. Standard eigenvalues of quaternion matrices 835 3 Main Results We now focus on the location of the standard eigenvalues of quaternion ma- trices. Theorem 3.1 Let A be an n × n central closed matrix. Then all the standard eigenvalues λ of A are located in the following Gerˇsgorin ball: Tr(A) G (A)= z ∈ H : z − ≤ R (A) , n where n − 1 n − 1 2n − 1 R (A)= · η + η2 − F (A), 2n − 1 n n2 2 2 |Tr(A)| 2 2 η = A − ,F(A)=AA∗ − A2 . F n F F Proof. Let B = λI − A. Then rank (B)=rank (λI − A) ≤ n − 1. By Lemma 2.12, we have 2 4 4 |Tr(B)| ≤ rank (B) · BF − F (B) ≤ (n − 1) BF − F (B). Using Lemma 2.13, we have F (B)=F (A) . Thus 2 4 |Tr(λI − A)| ≤ (n − 1) λI − AF − F (A). (1) We compute 2 |Tr(λI − A)| = Tr(λI − A) · Tr(λI − A) 2 = n2λ2 − λnT r A − λnT r (A)+|Tr(A)| , 4 ∗ 2 λI − AF =[Tr((λI − A) · (λI − A) )] 2 = nλ2 − λT r (A∗) − λT r (A)+Tr(AA∗) 2 ∗ 2 2 = nλ − λT r (A ) − λT r (A)+AF .