International Journal of Mathematics and Soft Computing

Vol.2, No.1 (2012), 51 - 55. ISSN 2249 - 3328

Secondary orthogonal similarity of a real matrix

and its secondary transpose
[Link]
Government Arts College(Autonomous)
Kumbakonam-612 001, Tamil Nadu, INDIA.
E-mail: profkrishnamoorthy@[Link]

[Link]
Dharmapuram Adhinam Arts College,
Mayiladuthurai-609 001, Tamil Nadu, INDIA.
E-mail: kjkdaac@[Link]

Abstract

In this paper we present some extended results of Vermeer in the context of secondary orthogonal
(respectively complex secondary unitary) Matrices.

Keywords: s-Orthogonal, s-Unitary Matrices, Similarity of Matrices, s-Orthogonal Similarity of

Matrices.
AMS Subject Classification(2010): 15B99, 15A24, 15A54.

1 Introduction
Throughout this paper we use the following notation:

Notation 1.1. Let F be a field and Mn (F ) be the algebra of n × n matrices. Let GLn (F ) denote the
set of invertible n × n matrices. Let Ons (F ) and Uns (F ) denote the set of s-orthogonal and s-unitary
matrices respectively.

Notation 1.2. The secondary transpose (conjugate secondary transpose) of A is defined by As =

V AT V (AΘ = V A∗ V ), where “V” is the fixed disjoint permutation matrix with units in its secondary
diagonal.

Definition 1.3. [4] Let A ∈ Mn (F ).

i. The matrix A is called s-symmetric if As = A.

ii. The matrix A is called s-skew symmetric if As = −A.

iii. The matrix A is called s-normal if AAΘ = AΘ A

iv. The matrix A is called s-orthogonal if AAs = As A = I.

v. The matrix A is called s-unitary if AAΘ = AΘ A = I.

Definition 1.4. [4] Let A ∈ Mn (F ) and B ∈ Mn (F ). A is said to be s-orthogonally similar (respectively

s-unitarily) to B if there exists a s-orthogonal matrix Q ∈ Mn (F ) such that A = Qs BQ(A = QΘ BQ).

Theorem 1.5. [1] Every square complex matrix is similar to a s-symmetric matrix.

51
52 S. Krishnamoorthy and K. Jaikumar

Theorem 1.6. [6] Let A ∈ Mn (F ). Then

i. there exists an X ∈ GLn (F ) such that XAX −1 = As .

ii. there exists a s-symmetric X ∈ GLn (F ) such that XAX −1 = As .

iii. every X ∈ GLn (F ) with XAX −1 = As is s-symmetric if and only if the minimal polynomial of
A is equal to its characteristic polynomial.

2 Secondary Orthogonal similarity

Lemma 2.1 ([5]). A s-unitary matrix U is a product of a s-symmetric s-unitary matrix and s-orthogonal
0
matrix O. That is, U = eiS O. It is also true that U = O0 eiS , where O0 is s-orthogonal and S 0 is real
s-symmetric.

Theorem 2.2. A ∈ Mn (C). The following assertions are equivalent.

i. A is s-unitarily similar to a complex s-symmetric matrix.

ii. There exists a s-symmetric s-unitary matrix U such that U AU Θ is s-symmetric.

iii. There exists a s-symmetric s-unitary matrix U and a s-symmetric matrix S such that A = SU.

iv. There exists a s-symmetric s-unitary matrix V such that V AV Θ = As .

Proof. (i) ⇒ (ii) : Suppose that V Θ AV = S, where V ∈ Uns (F ) and S is s-symmetric. By Lemma
2.1, V = U O with U is s-symmetric s-unitary matrix and O ∈ Ons (F ) We have A = V SV Θ =
U (OSOs )U Θ and hence U Θ AU = OSOs is s-symmetric.

(ii) ⇒ (iii) : Suppose that U Θ AU = S with U is s-symmetric s-unitary and S is s-symmetric. Then
A = U SU Θ = (U SU s )(U U s )Θ with U SU Θ s-symmetric and (U U s )Θ s-symmetric s-unitary.

(iii) ⇒ (iv) : Suppose that A = SU with S s-symmetric and U s-symmetric s-unitary. Then U AU Θ =
U S = U s S s = As .

(iv) ⇒ (i) : Suppose V AV Θ = As for some V s-symmetric s-unitary, then there exists U ∈ Uns (F )
such that V = U s U. Now (U s U )A(U s U )Θ = As and hence U AU Θ = (U s )Θ As U s = (U AU Θ )s ,
which is s-symmetric.

Theorem 2.3. Let A ∈ Mn (C) be non-singular. Then there are matrices R and E such that

i. If A is s-unitary then R is s-orthogonal.

ii. If A is s-unitary and s-symmetric then R is s-orthogonal and s-symmetric.

 Secondary orthogonal similarity of a real matrix and its secondary transpose 53

−1
Proof. (i) If A is s-unitary then A A = As A is s-symmetric and hence E is also s-symmetric. The
property EE = In and the s-symmetry implies that E is s-unitary. Finally, R = AE −1 is s-unitary and
real, so it is s-orthogonal.
(ii) If A is s-unitary and s-symmetric then AA = I = AA and so A = ER = RE. It follows that
s
R = EA = AE and since E (and A) are s-symmetric we see that Rs = (EA)s = As E = AE = R,
That is, R is s-symmetric.

Corollary 2.4. Let A, B ∈ Mn (R) be given.

i. A,B are real s-orthogonally similar if and only if A,B are s-unitarily similar.

ii. A,B are similar via a real s-symmetric s-orthogonal matrix if and only if A, B are similar via a
s-symmetric s-unitary matrix.

We assume that all matrices are real for the rest of the paper. We use the notation Q, Ψ for arbitrary
matrices in Ons (R), Ω for s-symmetric s-orthogonal matrices, S for a s-symmetric matrix, T for a s-skew
symmetric matrix, D for diagonal matrices and Σ for a real diagonal matrix with Σ2 = In .

We say that A has type T + D if it is the sum of a s-skew symmetric and a diagonal matrix; and A is
of type QD if A is the product of an s-orthogonal and a diagonal matrix.

We write A ' B(respectively A 'S B)if there exists an s-orthogonal (respectively s-symmetric
and s-orthogonal) matrix Q with QAQs = B. The notation [A] denotes the equivalence class of A with
respect to '. The relation 'S is not an equivalence relation.

Lemma 2.5. Let A, B ∈ Mn (R).

i. If A ' As and A ' B, then B ' B s

ii. If A 'S As and A ' B, then B 'S B s

iii. If Q ∈ On (R) and A = QAs Qs , then A = Qs As Q.

iv. If A ' B, then A + As ' B + B s and AAs ' BB s

Proof. (i) and (ii). If A = Q1 BQs1 and A = Q2 As Q2 (Q1 , Q2 are s-orthogonal matrices) then
As = Q1 B s Qs1 and so B = Qs1 AQ1 = Qs1 Q2 As Q2 Q1 = Qs1 Q2 Q1 B s Qs1 Q2 Q1 , and we conclude
B ' B s . If Q2 is s-symmetric then Qs1 Q2 Q1 and so B 'S B s .
(iii). A = QAs Qs = QAs Q−1 ⇒ Qs AQ = As , so A = (As )s = (Qs AQ)s = Qs As Q.
(iv). Obvious.

Theorem 2.6. Let A ∈ Mn (R). The following assertions are equivalent.

i. A is s-unitarily similar to a complex s-symmetric matrix.

54 S. Krishnamoorthy and K. Jaikumar

ii. A 'S As .

iii. A = ΩS, where Ω ∈ Ons (R) is s-symmetric and S is real s-symmetric.

iv. A ' Ω0 D, where Ω0 ∈ Ons (R) is s-symmetric and D is real diagonal.

Proof. (i)⇒ (ii) : According to Theorem 2.1 there exists a s-unitary secondary symmetric matrix V
with V AV Θ = As . Corollary 2.4 now ensures that A 'S As .
The equivalence of (i), (ii) and (iii) can be proved as in Theorem 2.2.
(iii) ⇒ (iv) : If A = ΩS, then write S = QDQs and we obtain A = ΩQDQs ' Qs ΩQD = Ω0 D.
(iv) ⇒ (i) If A ' Ω0 D = B then B 'S B s and so by Lemma 2.5 we have A 'S As .

Lemma 2.7. Let A ∈ Mn (R). Then

i. A ' B, where B has type T + D.

ii. A ' B, where B has type QD.

Proof. (i). Consider the Toeplitz decomposition A = 1/2(A + As ) + 1/2(A − As ) and suppose
Q(A + As )Q−1 = 2D. Then Qs AQ = D + 1/2Qs (A − As )Q is the sum of a diagonal matrix and a
secondary skew-symmetric matrix.
(ii). Consider the singular value decomposition A = Q1 DQs2 .
Then A ' B = Qs2 AQ2 = (Qs2 Q1 )D.
Lr
Lemma 2.8. Let A ∈ Mn (R) be a matrix of type T + D, where D = with di 6= dj for
i=1 di Ii
Ons (R) QAQs As then Q = ri=1 Qi , with
L
i 6= j and Ii the ki × ki identity matrix. If Q ∈ and =
Qi ∈ Ons (R).

Proof. By Lemma 3.1(3), QAQs = As implies QAs Qs = A and so Q(A + As )Qs = A + As , That is,
Q(2D)Qs = 2D. Hence Q commutes with the diagonal matrix 2D and this implies the result.

Theorem 2.9. Assume that A ∈ Mn (R) satisfies one of the following properties

i. A has n different real eigenvalues

ii. A + As has n different eigenvalues

iii. A has n different singular values.

If there exists Q ∈ Ons (R) with A = QAs Qs , then Q is s-symmetric and there exist at most 2n such
matrices.

Proof. (i). The s-symmetry of Q is a direct corollary of Theorem 1.5. We have to check that there exist
at most 2n such matrices Q. Let D ∈ Mn (R) be a diagonal matrix whose diagonal entries are the eigen
values of A. Suppose P, R ∈ Mn (R) are nonsingular and diagonalize A. That is, A = P DP −1 =
RDR−1 .
 Secondary orthogonal similarity of a real matrix and its secondary transpose 55

Corresponding columns of P and R are non-zero vectors in the same one-dimensional eigen space of A
so there is a diagonal matrix C ∈ Mn (R) such that R = P C. We fix one special matrix P ∈ Mn (R)
with A = P DP −1 .
Now suppose Q, Ψ ∈ Ons (R) are such that A = QAs Qs = ΨAs Ψ, so
A = Q(P DP −1 )s Qs = (Q(P s )−1 )D(Q(P s )−1 )−1 and
A = Ψ(P DP −1 )s Ψs = (Ψ(P s )−1 )D(Ψ(P s )−1 )−1 .
Note that Q(P s )−1 = P C1 for some diagonal matrix C1 ∈ Mn (R), so Q = P C1 P s is s-symmetric.
Note that Q(P s )−1 = Ψ(P s )−1 C2 for some diagonal matrix C2 ∈ Mn (R), so P s (Ψ−1 Q)(P s )−1 =
C2 and we have obtained a real diagonalization of the real s-orthogonal matrix Ψ−1 Q whose only
possible real eigen values are ±1. Thus C2 = diag(±1, ..., ±1) which shows that there are at most 2n
choices for Ψ, since Ψ = Q(P s )−1 C2−1 P s .
(ii). If Q ∈ Ons (R) and A = Qs As Q then by Lemma 2.5(iii),As = QAQs and so Q(A + As )Qs =
A + As = (A + As )s , the conclusion follows from part(i).
(iii). Fix Q ∈ Ons (R) with the property A = QAs Qs and consider a singular value decomposition
A = U DV s of A. Then A = QAs Qs = (QV )D(QU )s , so we have two singular value decomposition
of A. Distinct singular values and the uniqueness theorem for the singular value decomposition([3],
Theorem 3.1.1) ensure that QV = U P and QU = V P for some diagonal real s-orthogonal matrix P.
Then Q = U P V s = V P U s = Qs is s-symmetric and there are only 2n choices for P.

References
[1] Anna Lee, On s-Symmetric, s-Skewsymmetric and s-Orthogonal Matrices, Periodica Mathemat-
ica Hungarica, Vol.7, No.1(1976), 71-76.

[2] R. A. Horn, C. R. Johnson; Matrix Analysis, Cambridge University Press, (1985).

[3] R. A. Horn, C. R. Johnson, Topics in Matrix Analysis, Cambridge University Press,(1991).

[4] S. Krishnamoorthy, K. Jaikumar, On s-orthogonal Matrices, Global Journal of Computational

Science & Mathematics, Vol.1, No.1(2011), 1-8.

[5] S. Krishnamoorthy, K. Jaikumar, Exponential Representation of s-Orthogonal Matrices, Interna-

tional Journal of Mathematics Research, Vol.3, No.4(2011), 355-360.

[6] O. Taussky, H. Zassenhaus, On The Similarity Transformation Between a Matrix and its Trans-
pose, Pacific [Link].9(1959), 893-896.

[7] J. Vermeer, Orthogonal similarity of a real matrix and its transpose, Linear Algebra and its Appli-
cations, 428(2008), 382-392.