Open Access

An extension of the Golden-Thompson theorem

Journal of Inequalities and Applications20142014:14

https://doi.org/10.1186/1029-242X-2014-14

Received: 5 July 2013

Accepted: 12 December 2013

Published: 9 January 2014

Abstract

In this paper, we shall prove | tr e A + B | tr ( | e A | | e B | ) for normal matrices A, B. In particular, tr e A + B tr ( e A e B ) if A, B are Hermitian matrices, yielding the Golden-Thompson inequality.

MSC:15A16, 47A63, 15A45.

Keywords

normal matrixmajorizationGolden-Thompson inequality

1 Introduction and preliminaries

The famous Golden-Thompson inequality [14] for Hermitian matrices A, B states that tr e A + B tr ( e A e B ) . This inequality is a basic tool in quantum statistical mechanics and extensions to infinite dimension have an extensive literature [5, 6]. In this paper, we extend the classical Golden-Thompson theorem to normal matrices.

Throughout this paper, we adopt the following notation. Let M n be the set of all n × n complex matrices. For a matrix A M n , as usual, its conjugate transpose is denoted by A . A matrix A is called Hermitian if A = A , normal if A A = A A , and unitary if A A = A A = I n ( I n is the identity matrix of order n). Given a matrix A M n , the eigenvalues and singular values of A are denoted by λ 1 ( A ) , , λ n ( A ) , and s 1 ( A ) , , s n ( A ) , respectively, where | λ 1 ( A ) | | λ n ( A ) | and s 1 ( A ) s n ( A ) . In particular, when A is positive semidefinite ( A 0 ), then λ 1 ( A ) λ n ( A ) 0 . For simplicity, we denote λ ( A ) ( λ 1 ( A ) , , λ n ( A ) ) and s ( A ) ( s 1 ( A ) , , s n ( A ) ) . Recall that the singular values of a matrix A M n are defined to be the eigenvalues of | A | ( A A ) 1 / 2 , i.e., s ( A ) = λ ( | A | ) . Here s 1 ( A ) = A is the spectral norm of A. It is known that the spectral norm over M n is unitarily invariant, i.e., U A V = A for all unitary matrices U, V.

We now recall the concept of majorization (details can be found in [79]). We have the following basic majorant relations. For real vectors x = ( x 1 , , x n ) , y = ( y 1 , , y n ) in coordinates in decreasing order, we say that x is weakly majorized by y, denoted by x w y , if
j = 1 k x j j = 1 k y j , k = 1 , , n ,
and the weak log-majorant relation x wlog y means
j = 1 k x j j = 1 k y j , k = 1 , , n .

If in addition to x wlog y , 1 n x j = 1 n y j holds, we say that x is log-majorized by y, denoted briefly by symbols x log y . The following statement (see [8, 10]) is well known: x wlog y yields x w y for vectors x , y R + n .

Remark 1.1 For x = ( x 1 , , x n ) , we denote | x | = ( | x 1 | , , | x n | ) . Weyl’s majorant theorem [10] says that | λ ( A ) | log s ( A ) for A M n , that is,
( | λ 1 ( A ) | , , | λ n ( A ) | ) log ( | s 1 ( A ) | , , | s n ( A ) | ) .

The formula above implies that | λ ( A ) | w s ( A ) .

2 Lemmas

In this section, we shall propose some lemmas, laying the foundations of our main results in the next section.

Lemma 2.1 [11]

If A, B are positive semidefinite matrices, then
A B t A t B t , and λ 1 t ( A B ) λ 1 ( A t B t ) , for  t 1 .

Here note that X = s 1 ( X ) is the spectral norm of X.

Lemma 2.2 If A , B M n are normal matrices, then for any integer m 1
A B m | A m | | B | m = A m B m .
Proof Take the polar decompositions A = U | A | and B = V | B | . Here U, V are unitary matrices. Since A, B are normal, we can derive that U | A | = | A | U and V | B | = | B | V (see [12, 13]). Thus
A B = U | A | | B | V , A m B m = U m | A | m | B | m V m .
Since the norm is unitary invariant, we obtain the following:
A B m = U ( | A | | B | ) V m = ( | A | | B | ) m ,
and
A m B m = U m | A | m | B | m V m = | A | m | B | m .
From Lemma 2.1, | A | | B | m ( | A | m | B | m ) , we therefore conclude that
A B m A m B m .

 □

Lemma 2.3 If A , B M n are normal matrices, then
A B 2 = s 1 2 ( | A | | B | ) = λ 1 ( | A | 2 | B | 2 ) .
Proof Since A and B are normal, it follows from Lemma 2.2 that A B = | A | | B | . So we get
A B 2 = ( | A | | B | ) 2 = s 1 2 ( | A | | B | ) = λ 1 ( ( | A | | B | ) ( | A | | B | ) ) ,

as desired. □

Lemma 2.4 If A , B M n are normal matrices, then for integers m 2
s 1 m ( A B ) = A B m λ 1 ( | A | m | B | m ) .
Proof By Lemma 2.3, we have A B 2 = λ 1 ( | A | 2 | B | 2 ) , and
A B m = ( A B 2 ) m / 2 = λ 1 m / 2 ( | A | 2 | B | 2 ) .
Applying Lemma 2.1 to the right side above, we have the following:
λ 1 m / 2 ( | A | 2 | B | 2 ) λ 1 ( | A | m | B | m ) .

Thus we get A B m λ 1 ( | A | m | B | m ) for integers m 2 , as desired. □

Here we note that | A m | = | A | m holds for any normal matrix.

The following lemma needs the notion of the Grassmann power Λ k A (or antisymmetric tensor product), which can be found in [[8], p.18].

Lemma 2.5 If A M n , 1 k n , then for any natural number m, the following holds:
j = 1 k s j ( A m ) j = 1 k s j m ( A ) , and j = 1 n s j ( A m ) = j = 1 n s j m ( A ) .
i.e.,
s ( A m ) log s m ( A ) .
Proof For 1 k n , consider the k th antisymmetric tensor product Λ k A of A M n . It is known [[8], p.18] that Λ k ( A m ) = ( Λ k A ) m and
s 1 ( Λ k ( A m ) ) = s 1 ( ( Λ k A ) m ) = ( Λ k A ) m Λ k A m = ( s 1 ( Λ k A ) ) m .
Thus
j = 1 k s j ( A m ) j = 1 k s j m ( A ) , k = 1 , , n .
In particular,
j = 1 n s j ( A m ) = j = 1 n s j m ( A ) = | det ( A ) | m ,

which, equivalently, says that s ( A m ) log s m ( A ) . This completes the proof. □

3 Main results

In this section, we shall present the main results of this paper.

Theorem 3.1 If A , B M n are normal matrices, then
s ( e A + B ) log λ ( | e A | | e B | ) .
Proof Let A , B M n be normal matrices. It is clear that Λ k e A / m , Λ k e B / m are normal for 1 k n . By replacing A, B by Λ k e A / m , Λ k e B / m in Lemma 2.4, respectively, we can obtain the following for integers m 2 :
s 1 m ( Λ k ( e A / m e B / m ) ) = s 1 m ( Λ k e A / m Λ k e B / m ) λ 1 ( | Λ k e A | | Λ k e B | ) = λ 1 ( Λ k ( | e A | | e B | ) ) .
Here we note that | Λ k A | = Λ k | A | because | Λ k A | 2 = Λ k | A | 2 . So we obtain
j = 1 k s j m ( e A / m e B / m ) j = 1 k λ j ( | e A | | e B | ) .
From Lemma 2.5, we have
j = 1 k s j [ ( e A / m e B / m ) m ] j = 1 k s j m ( e A / m e B / m ) .
Thus,
j = 1 k s j [ ( e A / m e B / m ) m ] j = 1 k λ j ( | e A | | e B | ) .
The Lie product formula [[8], p.254] says that for any matrices A, B
lim m ( e A / m e B / m ) m = e A + B .
Thus taking m in the inequality above yields
j = 1 k s j ( e A + B ) j = 1 k λ j ( | e A | | e B | ) .
Finally we note that
j = 1 n s j ( e A + B ) = | det ( e A + B ) | = | det ( e A e B ) | = j = 1 n λ j ( | e A | | e B | ) .
Thus we get
s ( e A + B ) log λ ( | e A | | e B | ) .

This completes the proof. □

From Theorem 3.1, we know that
s ( e A + B ) log λ ( | e A | | e B | ) .
On the other hand, the following equation holds:
λ ( | e A | | e B | ) = λ ( | e A | 1 / 2 | e B | | e A | 1 / 2 ) = s ( | e A | 1 / 2 | e B | | e A | 1 / 2 ) .
The above two inequalities yield the following:
s ( e A + B ) log s ( | e A | 1 / 2 | e B | | e A | 1 / 2 ) .

Thus, we can get the following corollary by using the Fan Dominance Principle [[10], p.56].

Corollary 3.2 If A , B M n are normal matrices, then
| e A + B | | | e A | 1 / 2 | e B | | e A | 1 / 2 | ,

for all unitarily invariant norms | | .

From Theorem 3.1, we can also have the following result.

Theorem 3.3 If A , B M n are normal matrices, then
| λ ( e A + B ) | log λ ( | e A | | e B | ) .

Proof By Weyl’s majorant theorem we have | λ ( A ) | log s ( A ) . Hence Theorem 3.1 implies the desired inequality in Theorem 3.3. □

Note that Theorem 3.3 strengthens the Golden-Thompson inequality:
| tr ( e A + B ) | tr ( e A e B )

for Hermitian matrices A, B.

Theorem 3.4 If A, B are normal matrices, then
| tr ( e A + B ) | tr ( | e A | | e B | ) .
Proof Because x log y implies x w y , it follows from Theorem 3.3 that
| λ ( e A + B ) | w λ ( | e A | | e B | ) .
Taking the traces above, we have
| tr ( e A + B ) | tr ( | e A | | e B | ) .

So we get the desired inequality. This completes the proof. □

Of course, Theorem 3.4 is an extension of Golden-Thompson inequality:
tr ( e A + B ) tr ( e A e B )

for Hermitian matrices A, B.

Declarations

Acknowledgements

The authors would like to thank the referees for reading this work carefully, providing valuable suggestions and comments, which have significantly improved this article. This work is supported by National Natural Science Foundation of China (Grant No. 61379001).

Authors’ Affiliations

(1)
School of Mathematics and System Science, Beihang University

References

  1. Golden S: Lower bounds for the Helmholtz function. Phys. Rev. 1965, 137: B1127-B1128. 10.1103/PhysRev.137.B1127MathSciNetView ArticleMATHGoogle Scholar
  2. Hiai F: Equality cases in matrix norm inequalities of Golden-Thompson type. Linear Multilinear Algebra 1994, 35: 239-249.MathSciNetView ArticleMATHGoogle Scholar
  3. Symanzik K: Proofs and refinements of an inequality of Feynman. J. Math. Phys. 1965, 6: 1155-1156. 10.1063/1.1704383View ArticleGoogle Scholar
  4. Thompson CJ: Inequality with applications in statistical mechanics. J. Math. Phys. 1965, 6: 1812-1813. 10.1063/1.1704727View ArticleMathSciNetGoogle Scholar
  5. Araki H: Golden-Thompson and Peierls-Bogoliubov inequalities for a general von Neumann algebra. Commun. Math. Phys. 1973, 34: 167-178. 10.1007/BF01645678MathSciNetView ArticleMATHGoogle Scholar
  6. Ruskai MB: Inequalities for traces on von Neumann algebras. Commun. Math. Phys. 1972, 26: 280-289. 10.1007/BF01645523MathSciNetView ArticleMATHGoogle Scholar
  7. Ando T: Majorizations, doubly stochastic matrices, and comparison of eigenvalues. Linear Algebra Appl. 1989, 118: 163-248.MathSciNetView ArticleMATHGoogle Scholar
  8. Bhatia R: Matrix Analysis. Springer, Berlin; 1997.View ArticleMATHGoogle Scholar
  9. Marshall AW, Olkin I: Inequalities: Theory of Majorization and Its Applications. Academic Press, New York; 1979.MATHGoogle Scholar
  10. Zhan X Lecture Notes in Mathematics 1790. In Matrix Inequalities. Springer, Berlin; 2002.View ArticleGoogle Scholar
  11. Furuta T: Norm inequalities equivalent to Lowner-Heinz theorem. Rev. Math. Phys. 1989, 1: 135-137. 10.1142/S0129055X89000079MathSciNetView ArticleMATHGoogle Scholar
  12. Horn RA, Johnson CR: Matrix Analysis. Cambridge University Press, Cambridge; 1985.View ArticleMATHGoogle Scholar
  13. Zhang F Universitext. In Matrix Theory: Basic Results and Techniques. Springer, New York; 1999.View ArticleGoogle Scholar

Copyright

© Li and Zhao; licensee Springer. 2014

This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.