• Sonuç bulunamadı

Matrix trace results on the Hadamard and Khatri-Rao products

N/A
N/A
Protected

Academic year: 2021

Share "Matrix trace results on the Hadamard and Khatri-Rao products"

Copied!
6
0
0

Yükleniyor.... (view fulltext now)

Tam metin

(1)

Selçuk J. Appl. Math. Selçuk Journal of Vol. 8. No.1. pp. 3-8 , 2007 Applied Mathematics

Matrix Trace Results on the Hadamard and Khatri-Rao Products Mustafa Özel1 and Turgut Özi¸s2

1Division of Mathematics, Dokuz Eylül University, Engineering Faculty, Department

of Geophysics Tınaztepe, Buca, ˙Izmir 35160, Turkey; e-mail:mustafa.ozel@ deu.edu.tr

2Department of Mathematics, Faculty of Science, Ege University, Bornova, ˙Izmir

35100, Turkey;

e-mail:turgut.ozis@ ege.edu.tr

Received : August 13, 2006

Summary.Some matrix trace consequences involving the Hadamard and Khatri-Rao products of the matrices A and B are established by using the vec-operator. A similar conclusion between the Hadamard and Khatri-Rao products and a simple proof of the conclusion as in [7] are presented

Key words:Hadamard product; Khatri-Rao product; vec-operator. 2000 Mathematical Subject Classification 15A24, 15A69

1. Introduction

Hadamard, Khatri-Rao and Kronecker products have been studied and applied widely in matrix theory [3,4,5,7]. Also recently for partitioned matrices, some inequalities involving Khatri-Rao and Tracy-Singh products were given in [4,7]. Section 2 is devoted to some definitions of the above-mentioned products and elementary results. In section 3, a further study of the trace of partitioned matrices was carried out using the results of the Hadamard and Khatri-Rao products. A conclusion between the Hadamard and Khatri-Rao products of two matrices was established in Theorem 7. Finally, a simple proof of it was presented.

2. Basic Definitions and Theorems

In this section, we shall state some basic definitions and theorems, which will be needed in the sequel. For detail and proofs we refer to Refs. [1,2,3,4,5,6].

(2)

Definition 1Let  = [] be an  ×  matrix and = [] be a  ×  matrix.

The Kronecker product of A and B is an  ×  matrix which is defined to be the partitioned matrix

 ⊗  = []

where  = 1 2   and  = 1 2  

Definition 2 Let  = [] and  = [] be each an  ×  matrix; then the

Hadamard product of A and B is an  ×  matrix of elementwise products  ∗  = []

where  = 1 2   and  = 1 2  

Definition 3 Let  = [] be partitioned with of order × matrix

and  = [] be partitioned with of order × matrix; then the

Khatri-Rao product of A and B, which we write as A¯B, is as a matrix of order (P) × (P) defined by

 ¯  = [⊗ ]

where ⊗ is of order × 

Hadamard and Kronecker products have the relation as

(1)  ∗  = ( ⊗ )

for 2× selection matrix  such that  = . Note that  is the  ×

2matrix

(2)  = [11 22     ]

where  is the  ×  matrix of zeros except for solely one in the ( ) th

position.

Definition 4The vec operator of A, vecA , is defined as

(3)  = [1 2  ]

and is  × 1vector formed from the columns of A, where  is the columns of

A of order  × 

(3)

(4) ( ⊗ ) =   where tr(A) is the trace of A.

Theorem 2Let A and B be each  ×  matrices; then

(5) () = ()

where is the transpose of A.

Theorem 3Let , , , , and  be matrices as in the following; ( × ) ( × ) ( × ) ( × )   ( × ) then (6) () = ( ⊗ ) (7) () = ( ⊗ ) and (8) ( ) = ( ¯ )

where  = [11 22     ] is an  × 1vector and  is an  ×  identity

matrix.

Theorem 4For the matrices A, B, F, and G,

(9) ( ⊗ )( ¯ ) =  ¯ 

3. Main Results

It is hereby some trace equalities of Kronecker, Hadamard, and Khatri-Rao products of matrices were established.

Lemma 1Let J be an 2×  selection matrix; then

(10)  = ¯ 

(4)

Proof.For  = [11 22     ], we have  =  X =1 ⊗  = [11⊗ 1 22⊗ 2     ⊗ ]

where  is an  × 1 column vector which is “1” in the th and zero elsewhere is

called the unit vector. Then we write

 = [11 22     ] ¯ [1 2     ]

and (10) is obtained.

Theorem 5Let A and B be any  ×  matrices; then

(11)  ( ∗ ) = [(¯ )]( ⊗ )

Proof.Using the trace of (1), the commutative law of the trace of the product of two matrices, and (5)

(12)  ( ∗ ) = ¡( ⊗ )¢=£¡¢¤ ( ⊗ ) is obtained.

Corollary 1Let A and B be  ×  matrices; then

(13)  ( ∗ ) = () [( ⊗ ) ] and

(14)  ( ∗ ) = ()[ ¯ ( ⊗ )](2) where 2 is the 22identity matrix.

Proof.We have, using (12), (6) and (7)  ( ∗ ) =£¡2¢¤

 ( ⊗ ) 

The second equality is proved in precisely the same way starting from (13) and using (8).

Now we will give a new theorem that shows the relation between Hadamard and Khatri-Rao products of two matrices.

(5)

(15)  ( ∗ ) = (¯ )

Proof.To prove this theorem, first we write

 ( ∗ ) = [( ¯ )( ⊗ )]

using (5) in the right-hand side of (11). Then using (9) and (15) is obtained. Conversely, from Lemma 1 and (9) we write

( ¯ ) = [( ¯ )( ⊗ )] =  "Ã X =1 ⊗  ! ( ⊗ ) #

Then, using distributive law

( ¯ ) = X

()

()

is obtained and the proof is completed.

Theorem 7There exist a real matrix  of order 2×  such that

 = 

and

(16)  ∗  = ( ¯ )

for  = [] and  = [] of the same order  × , be partitioned with

 and of order  ×  as all block submatrix, respectively and

(17)  =  {      } where  is the selection matrix of order 2× 

Proof. Let  be as in (17), using the equation  =  and the proof of

Theorem1 in [7] then ( ¯ ) =h(⊗ ) i  =h¡(⊗ )  ¢  i =  ∗  is obtained.

(6)

References

1. Brewer, J.W.( 1978): Kronecker products and matrix calculus in system theory, IEEE Transactions on Circuits and Systems, 25, no.9, 772-781.

2. Graybill, F.A.(1969): Introduction to Matrices with Applications in Statistics, (Belmont, CA: Wadsworth).

3. Koning, R.H., Neudecker, H. and Wansbeek, T.(1991): Block Kronecker products and the vecb operator, Linear Algebra Appl, 149, 165-184.

4.Liu, S.(1999): Matrix results on the Khatri-Rao and Tracy-Singh products, Linear Algebra Appl., 289, 267-277.

5. Mond, B. and Peæariæ, J.E.(1998): Inequalities for the Hadamard product of matrices, SIAM J. Matrix Anal. Appl., 19, no.1, 66-70.

6.Rao, C.R. and Mitra, S.K.(1971): Generalized Inverse of Matrices and Its Applica-tions, (New York: Wiley).

7.Zhang, X., Yang, Z. and Cao, C.(2002): Inequalities involving Khatri-Rao products of positive semi-definite matrices, Applied Mathematics E-Notes, 2, 117-124.

Referanslar

Benzer Belgeler

First, advertisers are opting to place their ads in only one or two newspapers for any one specific time period, usually one Arabic language daily and/or one English language

yıldan sonra gelecek gelişmeler üzerinde­ ki düşüncelerimizi dile getirmeği daha uygun bulmakta­ yız : Devlet Güzel Sanatlar Akademisi, yurdumuzda, gü­ zel

12 Meksika, Oaxauka şehri bir kilim mağazasından 2005 (İ.Doğan, Özel

Bankaların karlılık göstergelerinden olan aktif kar- lılık, net faiz marjı ve sermaye karlılıklarını etkile- yen makroekonomik faktörlerden enflasyon oranı, GSYH’daki

Bu çalışmada, İrlanda’nın Shannon Bölgesi’nde yerel kalkınma odaklı olarak 1972 yılında kurulan Limerick Üniversitesi’nin (UL) girişimci üniversiteye

Türk insanı için vatan çok ayrı bir yere sahip olduğu için İbrahim Zeki Burdurlu, pek çok şiirinde, efsanesinde ve romanında Türk insanının vatan, bayrak ve Atatürk

Swap, “belirli bir miktar ve nitelikteki para, döviz, mali araç, alacak, mal gibi varlıklarla yükümlülüklerin, önceden belirlenen fiyat ve koşullara göre, gelecekteki

Konya Ticaret Borsası Dergisi, Nisan 2000, Yıl:3 Sayı:6 s:6­17 Konya Ticaret Borsası, Konya