Postec-Linear Operators Preserving

KMS 봄학회 포항공대 2000. 4. 22. (토)

On Linear Preservers of Several

Multivariate Majorization

Contents :

(0) Abstract

(1) Definitions

(2) History

(3) Previous Results

(4) Main Results

성균관대학교 수학과

이 상 구

KMS 봄학회 포항공대 2000. 4. 22. (토)

(0) Abstract

In this paper, we characterize linear operators on real matrices that preserve multivariate majorization. We place no additional restrictions on the linear operator.

(1) Definitions

Majorization is a topic of much interest in various areas of mathematics and statistics. (1929년 Hardy, Littlewood. Polya's : convex ft. g 가 일 필요충분조건) (1923년 Schur's work on Hadamard 부등식)(1905년 경제학에서 Lorenz curve)- 대수, 해석, 기하, 수치해석, 통계, 경제등

If and are nonincreasing -vectors of nonnegative real numbers such that for with equality at , then we say that is (vector) majorized by and write. Equivalently, if two row vectors have the relationship then there is a doubly stochastic matrix such that . (이 외에 연속함수 version, 무한 수열 version 등도 있음)

Our interest is in the subject of majorization for matrices. The MATRIX VERSION of majorization is that an real matrix is multivariate majorized by if there is a DOUBLY STOCHASTIC matrix such that .

G. Dahl's Definition in Matrix majorization, Linear Algebra Appl. 288:53-73(1999)--- "Row Stochastic Matrix"

A linear operator is said to preserve a relation R on if R

implies R.

* The topic of linear preservers (부록)

(2) History

In "Majorization and inequalities in matrix theory", Linear Algebra Appl., V.199:17-67 (1994), T. Ando characterized the linear operators which preserve vector majorization (equivalently, that preserve multivariate majorization on . We shall characterize all linear operators which preserve multivariate majorization with no conditions on m, n, or T.

*. "Linear Operators Strongly Preserving Multivariate Majorization " (with Y.H. Lee) Proc. Workshops in Pure Math. V. 17, part I (1997) pp.113-118.

**. Linear operators strongly preserving multivariate majorization with $T(I)=I$. Kyungpook Math. J. 39 (1999), no. 1, 191--194. (with L. Beasley and Y.-H. Lee) (Reviewer: Chi-Kwong Li) 15A04 (15A21 15A30)

***. A proof of conjecture on strong preservers of multivariate majorization" (with L. Beasley), preprint.

C.-K. Li, B.-S. Tam. N.-K. Tsing, Linear maps preserving the set of doubly stochastic matrices, Preprint(1999)'s Main Result : TFAE

(a) T(DS(n)) = DS(n)

(b) T(P(n)) = P(n)

(c) T의 모양은 for some P, Q in P(n)

따라서

then So let 로 놓으면 조건 T(I) =I를 없앨 수 있다고 생각했음.

**** Linear Operators Preserving Multivariate Majorization " (with L. Beasley), Linear Algebra and its Applications 304 141-159 (2000) ( TODAY!! )

Linear Operators Strongly Preserving Matrix Majorization " (with Y.-H. Lee and L. Beasley), Preprint

(3) Previous Results

NOTE: 아래에서 ?? 은 실수 R 을 의미 한 답니다. 즉, 은 입니다.

(editor 의 error)

(대우순수수학 1997) Theorem 2.11 Let for . Then is a multivariate majorization strong preserver on if one of the following or a composition of the following holds :

(1) for any nonzero scalar , an invertible matrix and a permutation matrix ;

(2) for a linear functional .

(경북수학저널 1999) Theorem 3.1 If strongly preserves multivariate majorization and (i.e., preserves nonnegative matrices), then a permutation matrix .

(4) Main Results

By Birkhoff's Theorem , DS(n) is the convex hull of the set P(n) of n by n permutation matrices.

We now define two linear mappings, which are fixed throughout the paper.

Definition 2.1

Define by where

. That is, is the column vector whose th component is the th row sum of , equivalently . Define by where Thus, is the diagonal matrix whose main diagonal is the vector .

Lemma 2.2

If is any element of and is any element of , then and

D와 r이 doubly stochastic 행렬 곱한 것에 불변

Proof.

Let and . Let . Then the entry of is

That is, . Similarly .

Lemma 2.4

Let and the linear operator be defined by , then preserves multivariate majorization.

MAIN THEOREM 2.5

Let be a linear operator on . Then TFAE:

(i) preserves multivariate majorization.

(ii) Either

(a) There exist matrices and such that for all or

(b) There exist matrices and a permutation matrix such that for all .

The remaining part of this article is devoted to the proof of the Main Theorem.

We state here the row version of Ando's characterization.

Lemma 2.6 (Ando, Corollary 2.7 ) Let preserve (vector) majorization, then either

(i) for some or

(ii) for some and

Lemma 2.7

Let the linear operator on preserve multivariate majorization and . Then

preserves multivariate majorization.

증명 : T는 M.M.P. 인데 S가 아니라면 모순임.

여기서 우선 T(X)의 j-th row를 생각한다.

Lemma 2.8

If preserves multivariate majorization, then defined by

the th row of T [ 0 ... 0 x 0 ... 0 ]^t

i^th row

preserves (vector) majorization.

여기서 보이는 것은 가 m.m.p.이므로 는 각각의 row vector maj.을 보존한다는 것. 그러면 이어서 T가 linear 이므로 다음을 갖는다.

Lemma 2.9

If preserves multivariate majorization, then for and there exist vectors , permutation matrices , and scalars and

with or , and , such that

이제 이 image의 모양을 단순히 하기 위해 계수의 성질들의 가능성을 줄여간다.

Lemma 2.10

Suppose for some is not a multiple of , we have

If preserves multivariate majorization then necessarily for all .

Lemma 2.12 If preserves multivariate majorization and

has at least two nonzero rows, one of which, say the th, is for some nonzero vector which is not a multiple of, and the other, say the th, is for some and , then .

이제 두 개의 모양이 동시에 나타나면 모순임을 보인다.

Lemma 2.13

for some which is not a multiple of , then does not preserve multivariate majorization.

다음의 두 Lemma는 and 임을 보여준다.

Lemma 2.15

Let be defined by

where . If preserves multivariate majorization, then .

Lemma 2.16

Let be defined by

where . If preserves multivariate majorization, then .

Now we can prove Main Theorem.

MAIN THEOREM 2.5

Let be a linear operator on . Then TFAE:

(i) preserves multivariate majorization.

(ii) Either

(a) There exist matrices and such that for all or

(b) There exist matrices and a permutation matrix such that for all .

PROOF

From the Corollary (2.14), T(X) must have one of the forms:

Either

(i) There exist matrices and such that for all or

(ii) T(X) =[] + for some , in , and for all in .

In the second case, from the previous Lemmas 2.15, 2.16 and 2.7 we have

T(X) =[] + = for some matrix and permutation matrix . QED.

(2) Linear Preserver Problems

Let K be an algebraic structure and let M(K) denote the vector space of all m*n matrices over K.

Over the last century, a great deal of effort has been devoted to the following problem. Characterize those linear operators T : M --> M which leave a function or set invariant. We call this a linear preserver problem.

The most typical and oldest type of linear preserver problem is as follows:

(1)일 때 아래 는?

F 는 scalar valued, matrix valued or set-valued function on M.

It is clear that the set of all such matrix-valued linear operators T is a multiplicative semigroup with an identity.

The study of these operators began in 1897 when Frobenius characterized the linear operators that preserve the determinant over two different sets of matrices :

Frobenius proved that the semigroup of linear operators that preserve the determinant consists of linear transformations of the form

(1.1)

(i) C 가 complex field 인 경우, det (UV) = 1

(ii) M 이 real symmetric matrices 인 경우,

U = kQ, V = Q^T and det(Q).

In 1925, Schur extended and improved the result for Case (i) as follows. Let ?(A) be the (nCk)^2 - tuple of the k-th order (i.e., kxk) minors of A in some order where k ? 3 is fixed.

Schur proved that if ?(T(A)) = S(?(A)), for a fixed nonsingular matrix S, then T has one of the two forms in (1.1) (without the restriction det (UV) = 1).

In 1949, Dieudonné showed that if F is any field and if T is a semilinear operator mapping M(F) onto itself, which holds the cone det (A) = 0 invariant, then T is of the form

where ? is an automorphism of the field and a_ij is the (i,j) entry of A.

We will call those operators T in (1.1)

(U,V)-operators.

Other interesting results of this type include:

(a) [59] If M is is the set of all nxn Hermitian matrices,

and F(A) = Inertia (A), then T is a (U,V)-operator

where U = V*, the conjugate transpose of U, for

some nonsingular matrix V ? M(C).

(b) [43] If M = M (C) and F(A) = { x*Ax| x ? C , x*x = 1},

then T is a (U,V)-operator where U = V* for some

unitary matrix U.

Consequences and recent work of this type can be found in [3, 17, 22, 23, 24, 28, 48, 53, 54] for preservers of generalized matrix functions, in [33, 39, 40, 50, 55] for inertia preservers, and in [44, 45, 46, 47] for numerical range and radius preservers.

We note that many linear preserver problems can be reduced to the problem of characterizing linear operators on M(K) which map the set of rank 1 matrices into itself.

A semiring (Gregory and Pullman [32] or Kim [41]) is a binary system (S, +, ?) such that (S, +) is an Abelian monoid (identity 0), (S, ?) is a monoid (identity 1), ? distributes over +, 0 ? s = s ? 0 = 0 for all s in S, and 1 $ 0. If 0 is the only element to have an additive inverse, then S is antinegative.

The Second type of linear preserver problem is:

(2) Let 일 때 는 s.t

Some examples of this type of problem with

appropriate characterizations.

(a) [51] If M = M(C), B is the set of unitary

matrices in M(C) and T(B) ⊂ B, then T is a

(U,V)-operator for some unitary matrices U and V.

(b) [4, 11, 13, 53] Suppose M = M(C) and B is the

set of all rank-k matrices. If T(B)⊂B, then T is

a (U,V)-operator for some nonsingular matrices

U and V.

(c) [19] Suppose M = M(S) where S is an

antinegative semiring with no zero divisors, and B

is the set of all idempotent matrices in M(S). If

T(B) = B, then T is a (U,V)-operator where

U = V for some invertible matrix V in M(S).

We may note that, in the first example, we were simply characterizing the structure of all linear operators T that map the unitary group into itself.

There is an extensive literature concerning the characterization of linear operators of type [II]. Further results may be found in [1, 2, 6, 7, 8, 9, 30, 49] for sets defined by rank, in [15, 16, 18, 21, 29] for sets defined by graph theoretic properties, i.e., irreducibility or strong connectivity, term rank, etc., and in [10, 20, 25, 35, 56, 57] for characterization of preservers of other sets.

The last and 3rd type of linear preserver problem is: (3) 인 는?

Some of the major results are:

(a) [5, 61] Suppose M = M (C) and A ~ D if A

and D commute. If T is a nonsingular linear

operator on M satisfying T(A) ~ T(D) whenever

A ~ D, then T is of the form

for some nonsingular matrix X ? M (C) ; l ? C

and linear functional f on M.

(b) [62] Suppose M = M (C) and A ~ D if A and

D are similar.

If T is a linear operator on M satisfying

T(A)~T(D) whenever A~D, then T is of the form

for some nonsingular matrix X ? M (C), l ? C and

linear functional f on M.

Further studies of the above type are found in [36, 61].

In [26] and [27], Chan, Lim and Tan gave the following characterization for those linear operators that preserve idempotent and tripotent matrices.

THE END

**********************

참고 자료:

A. W. Marshall and I. Olkin, Inequalities: Theory of Majorization and Its Applications, Academic Press, 1979

T. Ando, Majorization, doubly stochastic matrices and comparison of eigenvalues, Linear Algebra Appl. V.118:163-248 (1989).

T. Ando, Majorization and inequalities in matrix theory, Linear Algebra Appl., V.199:17-67 (1994)

J. V. Bondar, Comments and complements to: Inequalities: Theory of Majorization and Its Appl. by Albert W. Marshall and Ingram Olkin, Linear

Algebra Appl. V.199:115-130(1994)

G. Dahl, Matrix majorization, Linear Algebra Appl. 288:53-73(1999)

*************************