- Research
- Open access
- Published:
Extensions of interpolation between the arithmetic-geometric mean inequality for matrices
Journal of Inequalities and Applications volume 2017, Article number: 209 (2017)
Abstract
In this paper, we present some extensions of interpolation between the arithmetic-geometric means inequality. Among other inequalities, it is shown that if A, B, X are \(n\times n\) matrices, then
where \(f_{1}\), \(f_{2}\), \(g_{1}\), \(g_{2}\) are non-negative continuous functions such that \(f_{1}(t)f_{2}(t)=t\) and \(g_{1}(t)g_{2}(t)=t\) (\(t\geq0\)). We also obtain the inequality
in which m, n, s, t are real numbers such that \(m+n=s+t=1\), \(\vert \!\vert \!\vert \cdot \vert \!\vert \!\vert \) is an arbitrary unitarily invariant norm and \(p\in[0,1]\).
1 Introduction and preliminaries
Let \(\mathcal{M}_{n}\) be the \(C^{*}\)-algebra of all \(n\times n\) complex matrices and \(\langle\cdot,\cdot\rangle\) be the standard scalar product in \(\mathcal{C}^{n}\) with the identity I. The Gelfand map \(f(t)\mapsto f(A)\) is an isometrical ∗-isomorphism between the \(C^{*}\)-algebra \(C(\operatorname{sp}(A))\) of continuous functions on the spectrum \(\operatorname{sp}(A)\) of a Hermitian matrix A and the \(C^{*}\)-algebra generated by A and I.
A norm \(\vert \!\vert \!\vert \cdot \vert \!\vert \!\vert \) on \(\mathcal{M}_{n}\) is said to be unitarily invariant norm if \(\vert \!\vert \!\vert UAV \vert \!\vert \!\vert = \vert \!\vert \!\vert A \vert \!\vert \!\vert \), for all unitary matrices U and V. For \(A\in\mathcal{M}_{n}\), let \(s_{1}(A) \geq s_{2}(A) \geq\cdots \geq s_{n}(A)\) denote the singular values of A, i.e. the eigenvalues of the positive semidefinite matrix \(\vert A \vert = (A^{*}A)^{\frac{1}{2}}\) arranged in a decreasing order with their multiplicities counted. Note that \(s_{j}(A)=s_{j}(A^{*})=s_{j}( \vert A \vert )\) (\(1\leq j \leq n\)) and \(\Vert A \Vert =s_{1}(A)\). The Ky Fan norm of a matrix A is defined as \(\Vert A \Vert _{(k)}=\sum_{j=1}^{k} s_{j}(A)\) (\(1\leq k\leq n\)). The Fan dominance theorem asserts that \(\Vert A \Vert _{ (k)} \leq \Vert B \Vert _{(k)}\) for \(k = 1, 2, \ldots, n\) if and only if \(\vert \!\vert \!\vert A \vert \!\vert \!\vert \leq \vert \!\vert \!\vert B \vert \!\vert \!\vert \) for every unitarily invariant norm(see [1], p.93). The Hilbert-Schmidt norm is defined by \(\Vert A \Vert _{2}= (\sum_{j=1}^{n}s_{j}^{2}(A) )^{1/2}\), which is unitarily invariant.
The classical Cauchy-Schwarz inequality for \(a_{j}\geq0\), \(b_{j}\geq0\) (\(1\leq j\leq n\)) states that
with equality if and only if \((a_{1},\ldots, a_{n})\) and \((b_{1},\ldots, b_{n})\) are proportional [2]. Bhatia and Davis gave a matrix Cauchy-Schwarz inequality as follows:
where \(A, B, X\in\mathcal{M}_{n}\). For further information as regards the Cauchy-Schwarz inequality, see [3–5] and the references therein. Recently, Kittaneh et al. [6] extended inequality (1) to the form
where \(A, B, X\in\mathcal{M}_{n}\) and \(p\in[0,1]\). Audenaert [7] proved that, for all \(A, B\in\mathcal{M}_{n}\) and all \(p \in[0, 1]\), we have
In [8], the authors generalized inequality (3) for all \(A, B, X\in\mathcal{M}_{n}\) and all \(p \in[0, 1]\) to the form
Inequality (4) interpolates between the arithmetic-geometric mean inequality. In [6], the authors showed a refinement of inequality (4) for the Hilbert-Schmidt norm as follows:
in which \(A, B, X\in\mathcal{M}_{n}\), \(p\in[0,1]\) and \(r=\min\{p,1-p\}\). The Young inequality for every unitarily invariant norm states that \(\vert \!\vert \!\vert A^{p}B^{1-p} \vert \!\vert \!\vert \leq \vert \!\vert \!\vert pA+(1-p)B \vert \!\vert \!\vert \), where A, B are positive definite matrices and \(p\in[0,1]\) (see [9] and also [10, 11]). Kosaki [12] extended the last inequality for the Hilbert-Schmidt norm as follows:
where A, B are positive definite matrices, X is any matrix and \(p\in[0,1]\). In [13], the authors considered as a refined matrix Young inequality for the Hilbert-Schmidt norm
in which A, B are positive semidefinite matrices, \(X\in\mathcal {M}_{n}\), \(p\in[0,1]\) and \(r=\min\{p,1-p\}\).
Based on the refined Young inequality (7), Zhao and Wu [14] proved that
for \(0< p\leq\frac{1}{2}\) and
for \(\frac{1}{2}< p<1\) such that \(r=\min\{p, 1-p\}\) and \(r_{0}=\min\{ 2r, 1-2r\}\).
In this paper, we obtain some operator and unitarily invariant norms inequalities. Among other results, we obtain a refinement of inequality (5) and we also extend inequalities (2), (3) and (5) to the function \(f(t)=t^{p}\) (\(p\in\mathcal{R}\)).
2 Main results
In this section, using some ideas of [6, 15], we extend the Audenaert results for the operator norm.
Theorem 1
Let A, B, \(X\in\mathcal{M}_{n}\) and \(f_{1}\), \(f_{2}\), \(g_{1}\), \(g_{2}\) be non-negative continuous functions such that \(f_{1}(t)f_{2}(t)=t\) and \(g_{1}(t)g_{2}(t)=t\) (\(t\geq0\)). Then
Proof
It follows from
that we get the desired result. □
Corollary 2
If A, B, \(X\in\mathcal{M}_{n}\) and m, n, s, t are real numbers such that \(m+n=s+t=1\), then
In the next results, we show some generalizations of inequality (3) for the operator norm.
Corollary 3
Let \(A, B\in\mathcal{M}_{n}\) and let \(f_{1}\), \(f_{2}\), \(g_{1}\), \(g_{2}\) be non-negative continuous functions such that \(f_{1}(t)f_{2}(t)=t\) and \(g_{1}(t)g_{2}(t)=t\) (\(t\geq0\)). Then
where \(p\in[0,1]\).
Proof
Applying Theorem 1 for \(X=I\), we have
□
Corollary 4
Let \(A, B\in\mathcal{M}_{n}\) and let f, g be non-negative continuous functions such that \(f(t)g(t)=t^{2}\) (\(t\geq0\)). Then
where \(p\in[0,1]\).
Proof
Applying Theorem 1 and the Young inequality we get
□
3 Some interpolations for unitarily invariant norms
In this section, applying some ideas of [6], we generalize some interpolations for an arbitrary unitarily invariant norm.
Let \(Q_{k,n}\) denote the set of all strictly increasing k-tuples chosen from \({1, 2,\ldots, n}\), i.e. \(I \in Q_{k,n}\) if \(I = (i_{1}, i_{2}, \ldots, i_{k})\), where \(1\leq i_{1}< i_{2}<\cdots< i_{k}\leq n\). The following lemma gives some properties of the kth antisymmetric tensor powers of matrices in \(\mathcal{M}_{n}\); see [1], p.18.
Lemma 5
Let \(A, B\in\mathcal{M}_{n}\). Then
-
(a)
\((\bigwedge^{k}A)(\bigwedge^{k}B)=\bigwedge^{k}(AB)\) for \(k=1,\ldots,n\).
-
(b)
\((\bigwedge^{k}A)^{*}=\bigwedge^{k}A^{*}\) for \(k=1,\ldots,n\).
-
(c)
\((\bigwedge^{k}A)^{-1}=\bigwedge^{k}A^{-1}\) for \(k=1,\ldots,n\).
-
(d)
If \(s_{1}, s_{2}, \ldots, s_{n}\) are the singular values of A, then the singular values of \(\bigwedge^{k}A\) are \(s_{i_{1}}, s_{i_{2}}, \ldots, s_{i_{k}}\), where \(({i_{1}}, {i_{2}}, \ldots, {i_{k}})\in Q_{k,n}\).
Now, we show inequality (10) for an arbitrary unitarily invariant norm.
Theorem 6
Let \(A, B, X\in\mathcal{M}_{n}\) and \(\vert \!\vert \!\vert \cdot \vert \!\vert \!\vert \) be an arbitrary unitarily invariant norm. Then
where m, n, s, t are real numbers such that \(m+n=s+t=1\). In particular, if A, B are positive definite, then
where \(p\in[0,1]\).
Proof
If we replace A, B and X by \(\bigwedge^{k}A\), \(\bigwedge^{k}B\) and \(\bigwedge^{k}X\), their kth antisymmetric tensor powers in inequality (9) and apply Lemma 5, then we have
which is equivalent to
Applying Lemma 5(d), we have
where \(k=1,\ldots,n\). Inequality (13) implies that
where \(k=1,\ldots,n\). Hence
Now, using the Fan dominance theorem [1], p.98, we get the desired result. □
Now, using inequality (12), Theorem 6 and the same argument in the proof of Corollaries 3 and 4, we get the following results; these inequalities are generalizations of the Audenaert inequality (3).
Corollary 7
Let \(A, B\in\mathcal{M}_{n}\), m, n, s, t be real numbers such that \(m+n=s+t=1\) and let \(\vert \!\vert \!\vert \cdot \vert \!\vert \!\vert \) be an arbitrary unitarily invariant norm. Then
where \(p\in[0,1]\).
Corollary 8
Let \(A, B\in\mathcal{M}_{n}\), m, n, s, t be real numbers such that \(m+n=s+t=2\) and let \(\vert \!\vert \!\vert \cdot \vert \!\vert \!\vert \) be an arbitrary unitarily invariant norm. Then
in which \(p\in[0,1]\).
Remark 9
If we put \(n=m=s=t=1\) in inequality (14), then we obtain the Audenaert inequality (3). Also, if we use inequality (6), Corollaries 7 and 8, then similar to Corollaries 3 and 4 we get the following inequalities:
where \(A, B\in\mathcal{M}_{n}\), m, n, s, t are real numbers such that \(m+n=s+t=1\), \(p\in[0,1]\) and
for \(A, B\in\mathcal{M}_{n}\), real numbers m, n, s, t, in which \(m+n=s+t=2\) and \(p\in[0,1]\). These inequalities are generalizations of (4) for the Hilbert-Schmidt norms.
In the following theorem, we show a refinement of inequality (15) for the Hilbert-Schmidt norm.
Theorem 10
Let \(A, B, X\in\mathcal{M}_{n}\). Then
in which m, n, s, t are real numbers such that \(m+n=s+t=1\), \(p\in[0,1]\) and \(r=\min\{p,1-p\}\).
Proof
Applying inequality (11), we deduce that
where \(p\in[0,1]\) and \(r=\min\{p,1-p\}\), and the proof is complete. □
Theorem 10 includes a special case as follows.
Corollary 11
[6], Theorem 2.5
Let \(A, B, X\in\mathcal{M}_{n}\). Then
where \(p\in[0,1]\) and \(r=\min\{p,1-p\}\).
Proof
For \(p\in[0,1]\), if we put \(m=t=p\) and \(n=s=1-p\) in Theorem 10, then we get the desired result. □
The next result is a refinement of inequality (5).
Theorem 12
Let \(A, B, X\in\mathcal{M}_{n}({\mathcal {C}})\) and let \(p\in(0, 1)\). Then
-
(i)
For \(0< p\leq\frac{1}{2}\),
$$\begin{aligned} \bigl\Vert AXB^{*} \bigr\Vert _{2}^{2}& \leq \bigl( \bigl\Vert pA^{*}AX+(1-p)XB^{*}B \bigr\Vert _{2}^{2}-r_{0} \bigl\Vert \bigl(A^{*}A\bigr)^{\frac{1}{2}}X\bigl(B^{*}B \bigr)^{\frac{1}{2}}-A^{*}AX \bigr\Vert _{2}^{2} \\ &\quad {}-(1-p)^{2} \bigl\Vert A^{*}AX-XB^{*}B \bigr\Vert _{2}^{2} \bigr)^{\frac{1}{2}} \\ &\quad {}\times \bigl( \bigl\Vert (1-p)A^{*}AX+pXB^{*}B \bigr\Vert _{2}^{2}-r_{0} \bigl\Vert \bigl(A^{*}A\bigr)^{\frac{1}{2}}X\bigl(B^{*}B \bigr)^{\frac {1}{2}}-A^{*}AX \bigr\Vert _{2}^{2} \\ &\quad {}-p^{2} \bigl\Vert A^{*}AX-XB^{*}B \bigr\Vert _{2}^{2} \bigr)^{\frac{1}{2}}. \end{aligned}$$(16) -
(ii)
For \(\frac{1}{2}< p<1\),
$$\begin{aligned} \bigl\Vert AXB^{*} \bigr\Vert _{2}^{2}& \leq \bigl( \bigl\Vert pA^{*}AX+(1-p)XB^{*}B \bigr\Vert _{2}^{2}-r_{0} \bigl\Vert \bigl(A^{*}A\bigr)^{\frac{1}{2}}X\bigl(B^{*}B \bigr)^{\frac{1}{2}}-X\bigl(B^{*}B\bigr) \bigr\Vert _{2}^{2} \\ &\quad {}-(1-p)^{2} \bigl\Vert A^{*}AX-XB^{*}B \bigr\Vert _{2}^{2} \bigr)^{\frac{1}{2}} \\ &\quad {}\times \bigl( \bigl\Vert (1-p)A^{*}AX+pXB^{*}B \bigr\Vert _{2}^{2}-r_{0} \bigl\Vert \bigl(A^{*}A\bigr)^{\frac{1}{2}}X\bigl(B^{*}B \bigr)^{\frac {1}{2}}-X\bigl(B^{*}B\bigr) \bigr\Vert _{2}^{2} \\ &\quad {}-p^{2} \bigl\Vert A^{*}AX-XB^{*}B \bigr\Vert _{2}^{2} \bigr)^{\frac{1}{2}}, \end{aligned}$$(17)where \(r=\min\{p, 1-p\}\) and \(r_{0}=\min\{2r, 1-2r\}\).
Proof
The proof of inequality (17) is similar to that of inequality (16). Thus, we only need to prove the inequality (16).
If \(0< p\leq\frac{1}{2}\), replacing A and B by \(A^{*}A\) and \(B^{*}B\) in inequality (8), respectively, we have
Interchanging the roles of p and \(1-p\) in the inequality (18), we get
Applying inequalities (11), (18) and (19), we get the desired result. □
Corollary 13
Let \(A, B\in\mathcal{M}_{n}({\mathcal {C}})\) and \(p\in(0, 1)\). Then
-
(i)
For \(0< p\leq\frac{1}{2}\),
$$\begin{aligned} \bigl\Vert AB^{*} \bigr\Vert _{2}^{2} &\leq \bigl( \bigl\Vert pA^{*}A+(1-p)B^{*}B \bigr\Vert _{2}^{2}-r_{0} \bigl\Vert \bigl(A^{*}A\bigr)^{\frac{1}{2}}\bigl(B^{*}B \bigr)^{\frac{1}{2}}-A^{*}A \bigr\Vert _{2}^{2}\\ &\quad {} -(1-p)^{2} \bigl\Vert A^{*}A-B^{*}B \bigr\Vert _{2}^{2} \bigr)^{\frac {1}{2}} \\ &\quad {}\times \bigl( \bigl\Vert (1-p)A^{*}A+pB^{*}B \bigr\Vert _{2}^{2}-r_{0} \bigl\Vert \bigl(A^{*}A\bigr)^{\frac{1}{2}}\bigl(B^{*}B \bigr)^{\frac {1}{2}}-A^{*}A \bigr\Vert _{2}^{2}\\ &\quad {} -p^{2} \bigl\Vert A^{*}A-B^{*}B \bigr\Vert _{2}^{2} \bigr)^{\frac{1}{2}}. \end{aligned}$$ -
(ii)
For \(\frac{1}{2}< p<1\),
$$\begin{aligned} \bigl\Vert AB^{*} \bigr\Vert _{2}^{2}&\leq \bigl( \bigl\Vert pA^{*}A+(1-p)B^{*}B \bigr\Vert _{2}^{2}-r_{0} \bigl\Vert \bigl(A^{*}A\bigr)^{\frac{1}{2}}\bigl(B^{*}B \bigr)^{\frac{1}{2}}-\bigl(B^{*}B\bigr) \bigr\Vert _{2}^{2}\\ &\quad {} -(1-p)^{2} \bigl\Vert A^{*}A-B^{*}B \bigr\Vert _{2}^{2} \bigr)^{\frac {1}{2}} \\ &\quad {}\times \bigl( \bigl\Vert (1-p)A^{*}A+pB^{*}B \bigr\Vert _{2}^{2}-r_{0} \bigl\Vert \bigl(A^{*}A\bigr)^{\frac{1}{2}}\bigl(B^{*}B \bigr)^{\frac {1}{2}}-\bigl(B^{*}B\bigr) \bigr\Vert _{2}^{2}\\ &\quad {} -p^{2} \bigl\Vert A^{*}A-B^{*}B \bigr\Vert _{2}^{2} \bigr)^{\frac{1}{2}}, \end{aligned}$$where \(r=\min\{p, 1-p\}\) and \(r_{0}=\min\{2r, 1-2r\}\).
Through the following, we would like to obtain upper bound for \(\vert \!\vert \!\vert AXB^{*} \vert \!\vert \!\vert \), for every unitary invariant norm.
The following lemma has been shown in [16], and it is considered as a refined matrix Young inequality for every unitary invariant norm.
Lemma 14
Let \(A, B, X\in\mathcal{M}_{n}\) such that A, B are positive semidefinite. Then, for \(0\leq p\leq1\), we have
where \(r_{0}=\min\{p,1-p\}\).
Proposition 15
Let \(A, B, X\in\mathcal{M}_{n}\). Then
where \(p\in[0,1]\) and \(r_{0}=\min\{p,1-p\}\).
Proof
In inequality (20), if we replace A by \(A^{*}A\) and B by \(B^{*}B\), then we have
Interchanging p with \(1-p\) in inequality (21), we get
Now applying inequalities (11), (21) and (22) we get the desired inequality. □
4 Conclusions
Our application of the methods based on the Audenaert results is presented in this paper to the operator norm and so are some interpolations for an arbitrary unitarily invariant norm. Moreover, we refine some previous inequalities as regards the Cauchy-Schwarz inequality for the operator and Hilbert-Schmidt norms.
References
Bhatia, R: Matrix Analysis. Springer, New York (1997)
Mitrinović, DS, Pečarić, JE, Fink, AM: Classical and New Inequalities in Analysis. Kluwer Academic, Dordrecht (1993)
Bakherad, M: Some reversed and refined Callebaut inequalities via Kontorovich constant. Bull. Malays. Math. Sci. Soc. (2016). doi:10.1007/s40840-016-0364-9
Bakherad, M, Moslehian, MS: Reverses and variations of Heinz inequality. Linear Multilinear Algebra 63(10), 1972-1980 (2015)
Bakherad, M, Moslehian, MS: Complementary and refined inequalities of Callebaut inequality for operators. Linear Multilinear Algebra 63(8), 1678-1692 (2015)
Al-khlyleh, M, Kittaneh, F: Interpolating inequalities related to a recent result of Audenaert. Linear Multilinear Algebra 65(5), 922-929 (2017)
Audenaert, KMR: Interpolating between the arithmetic-geometric mean and Cauchy-Schwarz matrix norm inequalities. Oper. Matrices 9, 475-479 (2015)
Zou, L, Jiang, Y: A note on interpolation between the arithmetic-geometric mean and Cauchy-Schwarz matrix norm inequalities. J. Math. Inequal. 10(4), 1119-1122 (2016)
Ando, T: Matrix Young inequality. Oper. Theory, Adv. Appl. 75, 33-38 (1995)
Bakherad, M, Krnic, M, Moslehian, MS: Reverse Young-type inequalities for matrices and operators. Rocky Mt. J. Math. 46(4), 1089-1105 (2016)
Hajmohamadi, M, Lashkaripour, R, Bakherad, M: Some extensions of the Young and Heinz inequalities for matrices. Bull. Iranian Math. Soc. (in press)
Kosaki, H: Arithmetic-geometric mean and related inequalities for operators. J. Funct. Anal. 156, 429-451 (1998)
Hirzallah, O, Kittaneh, F: Matrix Young inequalities for the Hilbert-Schmidt norm. Linear Algebra Appl. 308, 77-84 (2000)
Zho, J, Wu, J: Operator inequalities involving improved Young and its reverse inequalities. J. Math. Anal. Appl. 421, 1779-1789 (2015)
Hajmohamadi, M, Lashkaripour, R, Bakherad, M: Some generalizations of numerical radius on off-diagonal part of \(2\times2\) operator matrices. J. Math. Inequal. (in press)
Sababheh, M: Interpolated inequalities for unitarily invariant norms. Linear Algebra Appl. 475, 240-250 (2005)
Acknowledgements
The first author would like to thank the Tusi Mathematical Research Group (TMRG).
Author information
Authors and Affiliations
Corresponding author
Additional information
Funding
The authors declare that there is no source of funding for this research.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
The authors contributed equally to the manuscript and read and approved the final manuscript.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
About this article
Cite this article
Bakherad, M., Lashkaripour, R. & Hajmohamadi, M. Extensions of interpolation between the arithmetic-geometric mean inequality for matrices. J Inequal Appl 2017, 209 (2017). https://doi.org/10.1186/s13660-017-1485-x
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s13660-017-1485-x