Open Access

On improvements of Fischer’s inequality and Hadamard’s inequality for K 0 -matrices

Journal of Inequalities and Applications20132013:460

https://doi.org/10.1186/1029-242X-2013-460

Received: 16 May 2013

Accepted: 9 September 2013

Published: 7 November 2013

Abstract

In this paper, the class of K 0 -matrices, which includes positive definite matrices, totally positive matrices, M-matrices and inverse M-matrices, is first introduced and the refinements of Fischer’s inequality and Hadamard’s inequality for K 0 -matrices are obtained. Some previous well-known results for totally nonnegative matrices can be regarded as the special case of this paper.

MSC:15A15.

Keywords

totally nonnegative matrix K 0 -matrix Fischer’s inequality Hadamard’s inequality

1 Introduction

All matrices considered in this paper are real. For an n × n matrix A , α , β n = { 1 , 2 , , n } , the submatrix of A lying in rows indexed by α and the columns indexed by β will be denoted by A [ α , β ] . If α = β , then the principal submatrix A [ α , α ] is abbreviated to A [ α ] . For any α n , let α c denote the complement of α relative to n , and let | α | denote the cardinality of α. If α = , we define det A [ ] = 1 . We use S n for the symmetric group on  n .

An n × n matrix A is called a P 0 -matrix (P-matrix) if all the principal minors of A are nonnegative (positive). A P-matrix A is called 1-minor symmetric if det A [ α , β ] det A [ β , α ] 0 , whenever | α | = | β | = 1 + | α β | . Of course, each of the P-matrices, such as positive definite matrices (PD), totally positive matrices (TP), M-matrices (M) and inverse M-matrices ( M 1 ), is 1-minor symmetric, see [1]. We have known that the following multiplicative principal minor inequalities are classical for PD, M and M 1 matrices:
  • Hadamard: det A i = 1 n a i i ;

  • Fischer: det A det A [ S ] det A [ S c ] , for S n ;

  • Koteljanskii: det A [ S T ] det A [ S T ] det A [ S ] det A [ T ] , for S , T n .

These inequalities also hold for totally nonnegative matrices (TN), see [16].

The study of multiplicative principal minor inequalities has been actively going on for many years, many authors have done various wonderful works on this topic, see [711]. In [11], Zhang and Yang improved Hadamard’s inequality for totally nonnegative matrices as follows:

If A = [ a i j ] is an n × n totally nonnegative matrix with i = 1 n a i i > 0 , then
det A min { i = 1 n a i i max 1 σ S n ( i = 1 n a i σ ( i ) a σ ( i ) i ) 1 / 2 , min k n a k k i = 1 i k n ( a i i a i k a k i a k k ) } .
(1.1)

Hadamard’s inequality for some subclasses of P 0 -matrices is an important inequality in matrix analysis, inequality (1.1) is the generalization of Hadamard’s inequality for totally nonnegative matrices. It is a noticeable problem to generalize inequality (1.1) for totally nonnegative matrices to other classes of matrices. In this paper we give some new upper bounds of Fischer’s inequality and Hadamard’s inequality for a subclass of P 0 -matrices and extend the corresponding results due to Zhang and Yang (see [11]).

2 Some lemmas

To avoid triviality, we always assume n > 1 . We will need important Sylvester’s identity for determinants (see [12]).

Lemma 2.1 [12]

Let A be an n × n matrix, α n , and suppose | α | = k ( 1 k n 1 ). Define the ( n k ) × ( n k ) matrix B = ( b i j ) , with i , j α c , by setting b i j = det A [ α { i } , α { j } ] for every i , j α c . B is called the Sylvester matrix of A associated with A [ α ] . Then Sylvester’s identity states that for each δ , γ α c , with | δ | = | γ | = l ,
det B [ δ , γ ] = ( det A [ α ] ) l 1 det A [ α δ , α γ ] .
(2.1)

For convenience, we introduce the following definition.

Definition 2.1 A P 0 -matrix (P-matrix) A is called a K 0 -matrix (K-matrix) if every principal submatrix of A satisfies Koteljanskii’s inequality.

Obviously, each principal submatrix of a K 0 -matrix (K-matrix) is a K 0 -matrix (K-matrix). Of course, each of the matrices PD, TP, M and M 1 is a K-matrix, the totally nonnegative matrices are K 0 -matrices. In fact, an evident necessary and sufficient condition for a K-matrix was given in [1].

Lemma 2.2 [1]

A P-matrix satisfies Koteljanskii’s inequality if and only if it is 1-minor symmetric.

There are K-matrices that lie in none of the classes PD, TP, M and M 1 , e.g.,
( 4 2 3 1 3 1 1 1 2 ) .

Lemma 2.3 If an n × n matrix A = [ a i j ] is a K 0 -matrix, then a i j a j i 0 for i , j n .

Proof For i , j n , i < j , we consider the submatrix A [ { i , j } ] of the K 0 -matrix A. Since A [ { i , j } ] is a K 0 -matrix, by the Hadamard’s inequality, we have
det A [ { i , j } ] = a i i a j j a i j a j i a i i a j j ,
therefore
a i j a j i 0 for  i , j n .

This completes the proof. □

Lemma 2.4 Let A be a K-matrix, B be the Sylvester matrix of A associated with A [ α ] , then B is a K-matrix.

Proof For δ α c , with | δ | = l , by Sylvester’s identity (2.1), we have
det B [ δ , δ ] = ( det A [ α ] ) l 1 det A [ α δ , α δ ] > 0 ,

this means that B is a P-matrix.

For δ , γ α c , with | δ | = | γ | = 1 + | δ γ | = k 1 , by Sylvester’s identity (2.1), we have
det B [ δ , γ ] det B [ γ , δ ] = ( det A [ α ] ) 2 k 1 2 det A [ α δ , α γ ] det A [ α γ , α δ ] > 0 .
(2.2)
Obviously,
| α δ | = | α | + | δ | = 1 + | α | + | δ γ | = 1 + | ( α δ ) ( α γ ) | , | α γ | = | α | + | γ | = 1 + | α | + | δ γ | = 1 + | ( α δ ) ( α γ ) | .
That is,
| α δ | = | α γ | = 1 + | ( α δ ) ( α γ ) | .

By Lemma 2.2 and (2.2), we conclude that B is 1-minor symmetric, therefore B is a K-matrix. □

Lemma 2.5 If A = [ a i j ] is an n × n K 0 -matrix, then
i = 1 n a i i [ i = 1 n a i σ ( i ) a σ ( i ) i ] 1 2 for each  σ S n .
(2.3)
Proof For each σ S n , by Lemma 2.3, we obtain
a i σ ( i ) a σ ( i ) i 0 for  i = 1 , 2 , , n .
Since each 2 × 2 principal minor of A is nonnegative, we have
a i i a σ ( i ) i σ ( i ) a i σ ( i ) a σ ( i ) i 0 for  i = 1 , 2 , , n ,
thus
( i = 1 n a i i ) 2 i = 1 n a i σ ( i ) a σ ( i ) i ,

hence inequality (2.3) follows. □

3 Main results

In this section, we give some new upper bounds for Fischer’s inequality and Hadamard’s inequality, and extend the corresponding results due to Zhang and Yang (see [11]).

Theorem 3.1 If A = [ a i j ] is an n × n K 0 -matrix, with i = 1 n a i i > 0 , then
det A det A [ α ] det A [ α c ] ( 1 max i α j α c a i j a j i a i i a j j ) .
(3.1)
Proof If A is singular, (3.1) is valid obviously. If A is a nonsingular K 0 -matrix, by Fisher’s inequality, we know that A is a P-matrix, therefore A is a K-matrix. Now we prove (3.1) for the K-matrix by induction on n. When n = 2 , it is very easy to see that the result is valid. We suppose that the result is valid for all m × m ( m < n and n 3 ) K-matrices, let k be the cardinality of α. Note that α and α c are symmetric in inequality (3.1). Without loss of generality, we assume k 2 . Let
max i α j α c a i j a j i a i i a j j = a i 1 j 1 a j 1 i 1 a i 1 i 1 a j 1 j 1 ,

where i 1 α and j 1 α c .

For any r α and r i 1 , let ω = α { r } , and S = ( S i j ) be the ( n k + 1 ) × ( n k + 1 ) Sylvester matrix of A associated with A [ ω ] . By Lemma 2.4, S is a K-matrix. Clearly, S r r = det A [ α ] , then, by Fisher’s inequality, we have
det S S r r det S [ α c ] = det A [ α ] det S [ α c ] ,
(3.2)
where α c = n α . It follows from Sylvester’s identity (2.1) and (3.2) that
det A = det S ( det A [ ω ] ) n k det A [ α ] det S [ α c ] ( det A [ ω ] ) n k .
(3.3)
Let β = n { r } , from Sylvester’s identity (2.1), we obtain
det S [ α c ] = ( det A [ ω ] ) n k 1 det A [ β ] .
(3.4)
Thus, by the inductive hypothesis, we have
det A [ β ] det A [ ω ] det A [ α c ] ( 1 max i ω j α c a i j a j i a i i a j j ) .
(3.5)
From (3.3), (3.4) and (3.5), it follows that
det A det A [ α ] det A [ α c ] ( 1 max i ω j α c a i j a j i a i i a j j ) .
(3.6)
Clearly,
max i ω j α c a i j a j i a i i a j j = max i α j α c a i j a j i a i i a j j .
(3.7)

Combining (3.6) and (3.7), we obtain inequality (3.1). □

Example 3.1 Now, we consider a 4 × 4 K-matrix that lies in none of the classes PD, TP, M and M 1 . Let
A = ( 15 5 0 3 1 3 6 9 0 3 7 11 1 5 15 30 ) .
Then A is 1-minor symmetric and A is a P-matrix. By Lemma 2.2, we know that A is a K-matrix. Let α = { 1 , 2 } , then α c = { 3 , 4 } . By calculating, we have
det A = 36 < det A [ { 1 , 2 } ] det A [ { 3 , 4 } ] ( 1 max i { 1 , 2 } j { 3 , 4 } a i j a j i a i i a j j ) = 1 , 800 7 ,

therefore inequality (3.1) holds.

Let A [ α ] = A [ { k } ] ( k n ), by Theorem 3.1 and the induction, we can obtain the following conclusion.

Corollary 3.2 If A = [ a i j ] is an n × n K 0 -matrix, with i = 1 n a i i > 0 , then
det A min k n a k k i = 1 i k n ( a i i a i k a k i a k k ) .

If A = [ a i j ] is a totally nonnegative matrix with i = 1 n a i i > 0 , Corollary 3.2 is certainly valid, so Corollary 3.2 is the generalization of Theorem 3 in [11].

Theorem 3.3 If A = [ a i j ] is an n × n K 0 -matrix, then
det A i = 1 n a i i max 1 σ S n ( i = 1 n a i σ ( i ) a σ ( i ) i ) 1 / 2 .
(3.8)

Proof If A is singular, by Lemma 2.3 and Lemma 2.5, we know (3.8) is valid. If A is a nonsingular K 0 -matrix, then A is a K-matrix. We suppose that A is a K-matrix in the following.

If n = 2 , then equality in (3.8) holds. If n > 2 , then first we prove the following conclusion:

For any 1 σ S n , there exists k 0 σ ( k 0 ) such that
[ i = 1 n a i σ ( i ) a σ ( i ) i ] 1 2 a k 0 σ ( k 0 ) a σ ( k 0 ) k 0 i = 1 , i k 0 , σ ( k 0 ) n a i i .
(3.9)

Case 1. If i = 1 n a i σ ( i ) a σ ( i ) i = 0 . For any 1 σ S n , by Lemma 2.3, we know that (3.9) certainly holds.

Case 2. If i = 1 n a i σ ( i ) a σ ( i ) i > 0 . Let T = { k | σ ( k ) k } , then | T | 2 , it is easy to see that (3.9) is true even with the equality sign for | T | = 2 . Now we assume that | T | > 2 .

Suppose that there exists 1 σ S n , for any k σ ( k ) , such that
[ i = 1 n a i σ ( i ) a σ ( i ) i ] 1 2 > a k σ ( k ) a σ ( k ) k i = 1 , i k , σ ( k ) n a i i .
(3.10)
Then multiplying all the possible inequalities in (3.10) yields
[ i = 1 n a i σ ( i ) a σ ( i ) i ] | T | 2 > k T [ a k σ ( k ) a σ ( k ) k i = 1 , i k , σ ( k ) n a i i ] .
(3.11)
Let S = n T . If S , by (3.11) we have
[ i T a i σ ( i ) a σ ( i ) i ] | T | 2 [ i S a i i ] | T | > [ i T a i σ ( i ) a σ ( i ) i ] [ i S a i i ] | T | i T a i i | T | 2 .
(3.12)
If S = , by (3.11) we have
[ i T a i σ ( i ) a σ ( i ) i ] | T | 2 > [ i T a i σ ( i ) a σ ( i ) i ] i T a i i | T | 2 .
(3.13)
By (3.12) and (3.13), we have
[ i T a i σ ( i ) a σ ( i ) i ] 1 2 > i T a i i .
(3.14)
On the other hand, since the principal submatrix A [ T ] of A is a K-matrix, applying Lemma 2.5 to A [ T ] yields
[ i T a i σ ( i ) a σ ( i ) i ] 1 2 i T a i i ,
(3.15)

which is a contradiction to (3.14), therefore (3.9) holds.

By Fisher’s inequality (3.1) and (3.9), we have
det A det A [ { k 0 , σ ( k 0 ) } ] det A [ { k 0 , σ ( k 0 ) } c ] [ a k 0 k 0 a σ ( k 0 ) σ ( k 0 ) a σ ( k 0 ) k 0 a k 0 σ ( k 0 ) ] i = 1 , i k 0 , σ ( k 0 ) a i i i = 1 n a i i [ i = 1 n a i σ ( i ) a σ ( i ) i ] 1 2 ,

therefore (3.8) holds. □

If A = [ a i j ] is a totally nonnegative matrix, Theorem 3.3 is certainly valid too, so Theorem 3.3 is the generalization of Theorem 4 in [11].

Example 3.2 Now we consider the previous K-matrix. Let
A = ( 4 2 3 1 3 1 1 1 2 ) .
By calculating we have
det A = 2 i = 1 3 a i i max 1 σ S 3 [ i = 1 3 a i σ ( i ) a σ ( i ) i ] 1 2 = 24 max { 4 , 9 , 6 , 4 , 6 } = 15 ,

therefore inequality (3.8) holds.

By Corollary 3.2 and Theorem 3.3, we get the following result.

Corollary 3.4 If A = [ a i j ] is an n × n K 0 -matrix, with i = 1 n a i i > 0 , then
det A min { i = 1 n a i i max 1 σ S n ( i = 1 n a i σ ( i ) a σ ( i ) i ) 1 / 2 , min k n a k k i = 1 i k n ( a i i a i k a k i a k k ) } .

If A is a totally nonnegative matrix, Corollary 3.4 is valid, so we obtain the result in [11].

Corollary 3.5 [11]

If A = [ a i j ] is an n × n totally nonnegative matrix, with i = 1 n a i i > 0 , then
det A min { i = 1 n a i i max 1 σ S n ( i = 1 n a i σ ( i ) a σ ( i ) i ) 1 / 2 , min k n a k k i = 1 i k n ( a i i a i k a k i a k k ) } .

Declarations

Acknowledgements

The author expresses his deep gratitude to the referees for their many very valuable suggestions and comments. The research of this paper was supported by the Natural Science Foundation of Shandong Province of China (ZR2010AL017).

Authors’ Affiliations

(1)
School of Mathematics and Information Sciences, Weifang University

References

  1. Fallat SM, Johnson CR: Determinantal inequalities: ancient history and recent advances. Contemp. Math. 2000, 259: 199–212.MathSciNetView ArticleGoogle Scholar
  2. Ando T: Totally positive matrices. Linear Algebra Appl. 1987, 90: 165–219.MATHMathSciNetView ArticleGoogle Scholar
  3. Carlson D: Weakly sign-symmetric matrices and some determinantal inequalities. Colloq. Math. 1967, 17: 123–129.MATHMathSciNetGoogle Scholar
  4. Gantmacher FR, Krein MG: Oszillationsmatrizen, Oszillationskerne und kleine Schwingungen Mechanischer Systeme. Akademie Verlag, Berlin; 1960.MATHGoogle Scholar
  5. Johnson CR, Smith RL: Almost principal minors of inverse M -matrices. Linear Algebra Appl. 2001, 337: 253–265. 10.1016/S0024-3795(01)00352-4MATHMathSciNetView ArticleGoogle Scholar
  6. Koteljanskii DM: A property of sign-symmetric matrices. Usp. Mat. Nauk 1953, 8: 163–167. (in Russian). English transl.: Am. Math. Soc. Transl. 27, 19–24 (1963)Google Scholar
  7. Chen S: Inequalities for M -matrices and inverse M -matrices. Linear Algebra Appl. 2007, 426: 610–618. 10.1016/j.laa.2007.05.040MATHMathSciNetView ArticleGoogle Scholar
  8. Elhashash A, Szyld DB: Generalizations of M -matrices which may not have a nonnegative inverse. Linear Algebra Appl. 2008, 49: 2435–2450.MathSciNetView ArticleGoogle Scholar
  9. Fallat SM, Gekhtman MI: Multiplicative principal-minor inequalities for totally nonnegative matrices. Adv. Appl. Math. 2003, 30: 442–470. 10.1016/S0196-8858(02)00506-7MATHMathSciNetView ArticleGoogle Scholar
  10. Fallat SM, Johnson CR: Hadamard powers and totally positive matrices. Linear Algebra Appl. 2007, 423: 420–427. 10.1016/j.laa.2007.01.012MATHMathSciNetView ArticleGoogle Scholar
  11. Zhang X, Yang S: An improvement of Hadamard’s inequality for totally nonnegative matrices. SIAM J. Matrix Anal. Appl. 1993, 14: 705–711. 10.1137/0614050MATHMathSciNetView ArticleGoogle Scholar
  12. Horn RA, Johnson CR: Matrix Analysis. Cambridge University Press, New York; 1985.MATHView ArticleGoogle Scholar

Copyright

© Li; licensee Springer. 2013

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.