- Research
- Open access
- Published:
On the Lawson–Lim means and Karcher mean for positive invertible operators
Journal of Inequalities and Applications volume 2018, Article number: 229 (2018)
Abstract
This note aims to generalize the reverse weighted arithmetic–geometric mean inequality of n positive invertible operators due to Lawson and Lim. In addition, we make comparisons between the weighted Karcher mean and Lawson–Lim geometric mean for higher powers.
1 Introduction
Let \(B(\mathcal{H})\) be the \(C^{*}\)-algebra of all bounded linear operators on a complex separable Hilbert space \(\mathcal{H}\). \(B(\mathcal{H})^{+}\) stands for the set of positive elements in \(B(\mathcal{H})\). A linear map Φ: \(B(\mathcal{H})\rightarrow B(\mathcal{H})\) is said to be positive (strictly positive) if \(\Phi (A)\ge0\) (\(\Phi(A)>0\)) whenever \(A\ge0\) (\(A>0\)). A positive linear map is said to be normalized (unital) if \(\Phi(I)=I\). Note that a positive linear map Φ is monotone in the sense that \(A\le B\) implies \(\Phi(A)\le\Phi(B)\). \(\mathbb{P}\) stands for the convex cone of positive invertible operators. \(\Delta_{n}\) denotes the simplex of positive probability vectors in \(\mathbb{R}^{n}\) convexly spanned by the unit coordinate vectors. \(\|\cdot\|\) and \(|\!\|\cdot|\!\|\) denote the operator norm and the unitarily invariant norm, respectively.
Since the pioneering papers of Pusz and Woronowicz [18], Ando [1], and Kubo and Ando [11], an extensive theory of two-variable geometric mean has sprung up for positive operators: For two positive operators A and B, the operator geometric mean is defined by \(A\sharp B:=A^{\frac{1}{2}}(A^{-\frac{1}{2}}BA^{-\frac{1}{2}})^{\frac {1}{2}} A^{\frac{1}{2}}\) for \(A>0\). But the n-variable case for \(n> 2\) was a long standing problem and many authors studied the geometric mean of n-variable.
In 2004, Ando et al. [2] succeeded in the formulation of the geometric mean for n positive definite matrices, and they showed that it satisfies ten important properties.
Definition 1.1
(Ando–Li–Mathias geometric mean [2])
Let \(A_{i}\) \((i=1,2,\ldots,n)\) be positive definite matrices. Then the geometric mean \(G_{\mathrm{ALM}}(A_{1},A_{2},\ldots,A_{n})\) is defined by induction as follows:
-
(i)
\(G_{\mathrm{ALM}}(A_{1},A_{2})=A_{1}\#A_{2}\).
-
(ii)
Assume that the geometric mean of any \(n-1\)-tuple of operators is defined. Let
$$G_{\mathrm{ALM}}\bigl((A_{j})_{j\neq i}\bigr)=G_{\mathrm{ALM}}(A_{1}, \ldots,A_{i-1},A_{i+1},\ldots,A_{n}), $$
and let the sequences \(\{A^{(r)}_{i}\}^{\infty}_{r=0}\) be \(A^{(0)}_{i}=A_{i}\) and \(A^{(r)}_{i} =G_{\mathrm{ALM}}((A^{(r-1)}_{j})_{j\neq i})\). If there exists \(\lim_{r\rightarrow\infty}A^{(r)}_{i}\), and it does not depend on i, then the geometric mean of n-matrices is defined as
In [20], Yamazaki pointed out that the definition of the geometric mean by Ando, Li and Mathias can be extended to Hilbert space operators. Lawson and Lim [12] established a definition of the weighted version of the Ando–Li–Mathias geometric mean for n positive operators, we call it Lawson–Lim geometric mean \(G[n,t](A_{1},A_{2},\ldots ,A_{n})\); see [12] for more details. In particular, \(G[n,\frac{1}{2}]\) for \(t=\frac{1}{2}\) is the Ando-Li-Mathias geometric mean. Similarly, the weighted arithmetic mean is defined as follows:
where \(t[n]_{i}\ge0\) for all \(i=1,2,\ldots,n\) with \(\sum_{i=1}^{n} t[n]_{i}=1\). Also, the weighted harmonic mean \(H[n,t](A_{1},A_{2},\ldots,A_{n})\) is defined as
Note that the coefficient \(\{t[n]_{i}\}\) depends on n and t only; see [6, 19] for more details.
Moreover, the weighted arithmetic–geometric-harmonic mean inequalities holds:
Since then, another approach to generalizing the geometric mean to n-variables, depending on Riemannian trace metric, was the Karcher mean, which was studied by many researchers; see [13, 14] and the references therein. Let \(\mathbb{A}=A_{1}, A_{2},\ldots,A_{n}\in\mathbb{P}^{n}\) and \(\omega =(w_{1},w_{2},\ldots, w_{n})\in\Delta_{n}\). By computing appropriate derivatives as in [3], the ω-weighted Karcher mean of \(\mathbb{A}\), denoted by \(G_{K}(\omega;\mathbb{A})\), coincides with the unique positive definite solution of the Karcher equation
In the case of two operators, \(A_{1}, A_{2}\in\mathbb{P}\), the Karcher mean coincides with the weighted geometric mean \(A_{1}\sharp_{t} A_{2} =A_{1}^{\frac{1}{2}}(A_{1}^{-\frac{1}{2}}A_{2}A_{1}^{-\frac{1}{2}})^{t} A_{1}^{\frac{1}{2}}\). From (1.2), the Karcher mean satisfies the self-duality \(G_{K}(\omega;\mathbb{A})=G_{K}(\omega;\mathbb{A}^{-1})^{-1}\), where \(\mathbb{A}^{-1}=(A^{-1}_{1},A^{-1}_{2},\ldots,A^{-1}_{n})\).
2 Weighted arithmetic and geometric means due to Lawson and Lim
In 2006, Yamazaki [20] obtained a converse of the arithmetic–geometric mean inequality of n-operators via Kantorovich constant. Soon after, Fujii el al. [6] also proved a stronger reverse inequality of the weighted arithmetic and geometric means due to Lawson and Lim of n-operators by the Kantorovich inequality.
In this section, we present the higher-power reverse inequalities of the weighted arithmetic and geometric means due to Lawson and Lim of n-operators, and several complements of the weighted geometric mean for n-variables have been established.
Lemma 2.1
Let \(A, B\ge0\). Then the following inequality holds:
Remark 2.1
Drury [5] recently established a remarkable improvement of (2.1) when \(A, B\) are, moreover, compact. More precisely, if \(A, B\geq0\) are compact, then there exists an isometry U such that
Lemma 2.2
([3, p. 28])
Let \(A, B\ge0\). Then, for \(1\le r<+\infty\),
It is well known that \(\|A\|\le1\) is equivalent to \(A^{*}A\le I\). This fact plays an important role in the proofs of the theorems.
Theorem 2.1
For any positive integer \(n\ge2\), let \(A_{1}, A_{2}, \ldots, A_{n}\) be positive invertible operators on a Hilbert space \(\mathcal{H}\) such that \(m\le A_{i}\le M\) for \(i=1,2,\ldots,n\) and some scalars \(0< m< M\). Then, for \(p\ge2\),
Proof
Let a map \(\Psi: \mathcal{B}(\mathcal{H})\oplus\cdots\oplus\mathcal {B}(\mathcal{H}) \mapsto\mathcal{B}(\mathcal{H})\) be defined by
Then Ψ is a positive linear map such that \(\Psi(I)=I\). The condition \(0< m\le A_{i}\le M\) for \(i=1,2,\ldots,n\) implies that
By (2.3) in [16], we have
Thus,
On the other hand, by computing, we deduce
The equality above follows from the self-duality of the geometric mean (see [2, 6, 20]). □
Taking \(p=2\), (2.4) implies the following corollary.
Corollary 2.1
For any positive integer \(n\ge2\), let \(A_{1}, A_{2}, \ldots, A_{n}\) be positive invertible operators on a Hilbert space \(\mathcal{H}\) such that \(m\le A_{i}\le M\) for \(i=1,2,\ldots,n\) and some scalars \(0< m< M\). Then
Note that if \(t=\frac{1}{2}\), the inequality (2.7) reduces to Lin’s result (see [15, Theorem 3.2]). By the Löewner–Heinz inequality, we can easily get Theorem 7 in [6] from (2.7).
Theorem 2.2
For any positive integer \(n\ge2\), let \(A_{1}, A_{2}, \ldots, A_{n}\) be positive invertible operators on a Hilbert space \(\mathcal{H}\) such that \(m\le A_{i}\le M\) for \(i=1,2,\ldots,n\) and some scalars \(0< m< M\). Then, for \(1<\alpha\le2\) and \(p\ge2\alpha\),
where \(k=\frac{(m+M)^{2}}{4mM}\).
Proof
Let a map \(\Psi: \mathcal{B}(\mathcal{H})\oplus\cdots\oplus\mathcal {B}(\mathcal{H}) \mapsto\mathcal{B}(\mathcal{H})\) be defined as in the proof of Theorem 2.1. By (2.5) and the Löewner–Heinz inequality, we have
that is,
By (2.3) in [16], we have
On the other hand, by (2.7),
By computing, we deduce
We obtain the desired result. □
Putting \(\alpha=2\) in the inequality (2.8) implies the following.
Corollary 2.2
Under the same conditions as in Theorem 2.2, then, for \(p\ge4\),
Remark 2.2
When \(\frac{M}{m}\le2+\sqrt{3}\), we have \(\frac {(m+M)^{2p}}{16m^{p}M^{p}}\ge\frac{(k(M^{2}+m^{2}))^{p}}{16M^{p}m^{p}}\), it is easy to see that (2.11) is sharper than (2.4) for \(p\ge4\).
Next, we show the complements of the weighted geometric mean due to Lawson and Lim by virtue of the following lemma (see Corollary 2.12 in [10]) and we will generalize Lemma 2.8 and Lemma 2.9 in [19] in two following theorems.
Lemma 2.3
([10])
For any integer \(n\ge2\), let \(A_{1},A_{2},\ldots,A_{n}\) be positive invertible operators in \(\mathbb{P}\) such that \(m\le A_{i}\le M\) for all \(i=1,2,\ldots,n\) and some scalars \(0< m\le M\). Then
where \(K(m,M,p)=\frac{(p-1)^{p-1}}{p^{p}}\frac {(M^{p}-m^{p})^{p}}{(M-m)(mM^{p}-Mm^{p})^{p-1}}\) is the generalized Kantorovich constant.
Proof
By Corollary 2.6 in [17],
Let the map \(\Phi: \mathcal{B}(\mathcal{H})\oplus\cdots\oplus\mathcal {B}(\mathcal{H}) \mapsto\mathcal{B}(\mathcal{H})\) be defined as Ψ in the proof of Theorem 2.1. Then, for \(p\ge1\),
□
Theorem 2.3
For any integer \(n\ge2\), let \(A_{1},A_{2},\ldots,A_{n}\) be positive invertible operators in \(\mathbb{P}\) such that \(m\le A_{i}\le M \) for all \(i=1,2,\ldots,n\) and some scalars \(0< m\le M\). Then
and
Proof
By the arithmetic–geometric mean inequality and (2.12), it follows that
For \(p\in(1,2]\), it follows from (2.7) and the Löewner–Heinz inequality that
Combining the inequalities above, we have
For \(p\in[2,\infty)\), from (2.4) and (2.13), we obtain
□
In the following remark, we present the case of \(p\ge1\) for the Ando–Li–Mathias geometric mean.
Remark 2.3
Let \(t=\frac{1}{2}\) in Theorem 2.3. Then
and
Theorem 2.4
For any integer \(n\ge2\), let \(A_{1},A_{2},\ldots,A_{n}\) be positive invertible operators in \(\mathbb{P}\) such that \(m\le A_{i}\le M \) for all \(i=1,2,\ldots,n\) and some scalars \(0< m\le M\). Then
for all \(1<\frac{p}{q}\le2\) and \(p\ge1\), and
for all \(\frac{p}{q}\ge2\) and \(p\ge1\).
Proof
For each \(0< q\le p\), it follows from the arithmetic–geometric mean inequality (1.1) and (2.12) that
On the other hand, for \(1<\frac{p}{q}\le2\), from (2.7) and \(m^{q}\le A_{i}^{q}\le M^{q}\), it follows that
Combining the two inequalities above, we obtain
By the Löewner–Heinz inequality and \(p\ge1\), it follows that
Similarly, for all \(\frac{p}{q}\ge2\), from (2.4) we have
Combining with (2.14), we obtain
It follows from \(p\ge1\) that
This completes the proof. □
3 Comparisons between the weighted Karcher mean and the Lawson–Lim geometric mean
In the final section, we make comparisons between the weighted Karcher mean and the Lawson–Lim geometric mean for higher powers. This is a fascinating work because the order relation can be preserved between higher-power operators by the Kantorovich constant.
Lemma 3.1
([8])
Let \(0< m\le A\le M\) and \(A\le B\). Then
Lemma 3.2
([9])
Let A and B be positive invertible operators on a Hilbert space \(\mathcal{H}\) satisfying \(B\ge A>0\) and \(0< m\le A\le M\). Then
holds for any \(p\ge1\), where \(\mathrm{K}(m,M,p)\) is the generalized Kantorovich constant or the Ky Fan–Furuta constant.
Theorem 3.1
For any positive integer \(n\ge2\), let \(A_{1}, A_{2}, \ldots, A_{n}\) be positive invertible operators on a Hilbert space \(\mathcal{H}\) such that \(m\le A_{i}\le M\) for \(i=1,2,\ldots,n\) and some scalars \(0< m< M\). Then
and
Proof
By the Löewner–Heinz inequality and (2.7), we have
It follows from Lemma 3.2 and (3.3) that
The inequality (3.2) follows from Lemma 3.2 and (2.4) that
□
Remark 3.1
When \(p=1, t=\frac{1}{2}\), the inequality (3.1) reduces to the first inequality of Theorem 5.1 in [7].
References
Ando, T.: Topics on operator inequalities. Hokkaido Univ. Lecture Note, 1978
Ando, T., Li, C.K., Mathias, R.: Geometric means. Linear Algebra Appl. 385, 305–334 (2004)
Bhatia, R.: Positive Definite Matrices. Princeton University Press, Princeton (2007)
Bhatia, R., Kittaneh, F.: Notes on matrix arithmetic–geometric mean inequalities. Linear Algebra Appl. 308, 203–211 (2000)
Drury, S.W.: On a question of Bhatia and Kittaneh. Linear Algebra Appl. 437, 1955–1960 (2012)
Fujii, J.I., Fujii, M., Nakamura, M., Pečarić, J., Seo, Y.: A reverse inequality for the weighted geometric mean due to Lawson–Lim. Linear Algebra Appl. 427, 272–284 (2007)
Fujii, J.I., Seo, Y.: On the Ando–Li–Mathias mean and the Karcher mean of positive definite matrices. Linear Multilinear Algebra 63, 636–649 (2015)
Fujii, M., Izumino, S., Nakamoto, R., Seo, Y.: Operator inequalities related to Cauchy–Schwarz and Hölder–McCarthy inequalities. Nihonkai Math. J. 8, 117–122 (1997)
Furuta, T.: Operator inequalities associated with Hölder–McCarthy and Kantorovich inequalities. J. Inequal. Appl. 2, 137–148 (1998)
Furuta, T., Mićić, J., Pečarić, J., et al.: Mond–Pečarić Method in Operator Inequalities [M]. Element (2005)
Kubo, F., Ando, T.: Means of positive linear operators. Math. Ann. 246, 205–224 (1980)
Lawson, J., Lim, Y.: A general framework for extending means to higher orders. Colloq. Math. 113, 191–221 (2008)
Lawson, J., Lim, Y.: Karcher means and Karcher equations of positive definite operators. Trans. Am. Math. Soc. 1, 1–22 (2014)
Lim, Y., Pálfia, M.: Matrix power means and the Karcher mean. J. Funct. Anal. 262, 1498–1514 (2012)
Lin, M.: Squaring a reverse AM-GM inequality. Stud. Math. 215, 187–194 (2013)
Lin, M.: On an operator Kantorovich inequality for positive linear maps. J. Math. Anal. Appl. 402, 127–132 (2013)
Mićić, J., Pečarić, J., Seo, Y.: Complementary inequalities to inequalities of Jensen and Ando based on the Mond–Pečarić method. Linear Algebra Appl. 318, 87–107 (2000)
Pusz, W., Woronowicz, S.L.: Functional calculus for sesquilinear forms and the purification map. Rep. Math. Phys. 8, 159–170 (1975)
Seo, Y.: Generalized Pólya–Szegö type inequalities for some non-commutative geometric means. Linear Algebra Appl. 438, 1711–1726 (2013)
Yamazaki, T.: An extension of Kantorovich inequality to n-operators via the geometric mean by Ando–Li–Mathias. Linear Algebra Appl. 416, 688–695 (2006)
Acknowledgements
The authors wish to thank the editor and anonymous referees for the careful reading of the manuscript and useful comments.
Funding
This research is supported by the Science and Technology Research Program of Chongqing Municipal Education Commission (Grant No. KJ1713330), Research Foundation of Chongqing University of Science and Technology (Grant No. ck2017zkyb025) and the National Natural Science Foundation of China (Grant No. 61601068).
Author information
Authors and Affiliations
Contributions
All authors contributed equally to the writing of this paper. All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare that they have no competing interests.
Additional information
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
Liao, W., Long, P., Ren, Z. et al. On the Lawson–Lim means and Karcher mean for positive invertible operators. J Inequal Appl 2018, 229 (2018). https://doi.org/10.1186/s13660-018-1817-5
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s13660-018-1817-5