We are IntechOpen,

the world’s leading publisher of

Open Access books
Built by scientists, for scientists

4,000
Open access books available
116,000
International authors and editors
120M Downloads

Our authors are among the

154
Countries delivered to
TOP 1%
most cited scientists
12.2%
Contributors from top 500 universities

Selection of our books indexed in the Book Citation Index

in Web of Science™ Core Collection (BKCI)

Interested in publishing with us?

Contact [email protected]
Numbers displayed above are based on latest data collected.
For more information visit www.intechopen.com
 Provisional chapter

Operator Means and Applications

Pattrawut Chansangiam

Additional information is available at the end of the chapter

http://dx.doi.org/10.5772/46479

1. Introduction

The theory of scalar means was developed since the ancient Greek by the Pythagoreans until
the last century by many famous mathematicians. See the development of this subject in a
survey article [1]. In Pythagorean school, various means are defined via the method of pro‐
portions (in fact, they are solutions of certain algebraic equations). The theory of matrix and
operator means started from the presence of the notion of parallel sum as a tool for analyz‐
ing multi-port electrical networks in engineering; see [2]. Three classical means, namely,
arithmetic mean, harmonic mean and geometric mean for matrices and operators are then
considered, e.g., in [3], [4], [5], [6], [7]. These means play crucial roles in matrix and operator
theory as tools for studying monotonicity and concavity of many interesting maps between
algebras of operators; see the original idea in [3]. Another important mean in mathematics,
namely the power mean, is considered in [8]. The parallel sum is characterized by certain
properties in [9]. The parallel sum and these means share some common properties. This
leads naturally to the definitions of the so-called connection and mean in a seminal paper
[10]. This class of means cover many in-practice operator means. A major result of Kubo-An‐
do states that there are one-to-one correspondences between connections, operator mono‐
tone functions on the non-negative reals and finite Borel measures on the extended half-line.
The mean theoretic approach has many applications in operator inequalities (see more infor‐
mation in Section 8), matrix and operator equations (see e.g. [11], [12]) and operator entropy.
The concept of operator entropy plays an important role in mathematical physics. The rela‐
tive operator entropy is defined in [13] for invertible positive operators A, B by

S ( A | B ) = A 1/2 log ( A -1/2 BA -1/2) A 1/2. (id1)

© 2012 Chansangiam; licensee InTech. This is an open access article distributed under the terms of the
Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits
unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
2 Linear Algebra

In fact, this formula comes from the Kubo-Ando theory–S ( · | · ) is the connection corre‐
sponds to the operator monotone function t ↦ log t. See more information in Chapter
IV[14] and its references.
In this chapter, we treat the theory of operator means by weakening the original definition
of connection in such a way that the same theory is obtained. Moreover, there is a one-to-
one correspondence between connections and finite Borel measures on the unit interval.
Each connection can be regarded as a weighed series of weighed harmonic means. Hence,
every mean in Kubo-Ando's sense corresponds to a probability Borel measure on the unit
interval. Various characterizations of means are obtained; one of them is a usual property of
scalar mean, namely, the betweenness property. We provide some new properties of ab‐
stract operator connections, involving operator monotonicity and concavity, which include
specific operator means as special cases.
For benefits of readers, we provide the development of the theory of operator means. In Sec‐
tion 2, we setup basic notations and state some background about operator monotone func‐
tions which play important roles in the theory of operator means. In Section 3, we consider
the parallel sum together with its physical interpretation in electrical circuits. The arithmetic
mean, the geometric mean and the harmonic mean of positive operators are investigated
and characterized in Section 4. The original definition of connection is improved in Section 5
in such a way that the same theory is obtained. In Section 6, several characterizations and
examples of Kubo-Ando means are given. We provide some new properties of general oper‐
ator connections, related to operator monotonicity and concavity, in Section 7. Many opera‐
tor versions of classical inequalities are obtained via the mean-theoretic approach in Section
8.

2. Preliminaries

Throughout, let B (ℋ) be the von Neumann algebra of bounded linear operators acting on a
Hilbert space ℋ. Let B (ℋ)sa be the real vector space of self-adjoint operators on ℋ. Equip
B (ℋ) with a natural partial order as follows. For A, B ∈ B (ℋ)sa, we write A B if B - A is a
positive operator. The notation T ∈ B (ℋ)+ or T 0 means that T is a positive operator. The
case that T 0 and T is invertible is denoted by T > 0 or T ∈ B (ℋ)++. Unless otherwise stat‐
ed, every limit in B (ℋ) is taken in the strong-operator topology. Write An → A to indicate
that An converges strongly to A. If An is a sequence in B (ℋ)sa, the expression An ↓ A means
that An is a decreasing sequence and An → A. Similarly, An ↑ A tells us that An is increasing
and An → A. We always reserve A, B, C, D for positive operators. The set of non-negative
real numbers is denoted by ℝ+.

Remark 0.1 It is important to note that if An is a decreasing sequence in B (ℋ)sa such that
An A, then An → A if and only if An x, x → Ax, x for all x ∈ ℋ. Note first that this se‐
quence is convergent by the order completeness of B (ℋ). For the sufficiency, if x ∈ ℋ, then
 Operator Means and Applications 3
http://dx.doi.org/10.5772/46479

∥ ( An - A)1/2 x ∥ 2 = ( An - A)1/2 x, ( An - A)1/2 x = ( An - A) x, x → 0 ()

and hence ∥ ( An - A) x ∥ → 0.

The spectrum of T ∈ B (ℋ) is defined by

Sp (T ) = {λ ∈ T - λI is not invertible}. ()

Then Sp (T ) is a nonempty compact Hausdorff space. Denote by C ( Sp (T )) the C*-algebra

of continuous functions from Sp (T ) to .LetT B(H)beanormaloperatorandz: Sp(T) the inclu‐
sion. Then there exists a unique unital * -homomorphism φ : C ( Sp (T )) → B (ℋ) such that
φ (z ) = T , i.e.,

• φ is linear

• φ ( fg ) = φ ( f )φ (g ) for all f , g ∈ C ( Sp (T ))

• φ ( f̄ ) = (φ ( f ))* for all f ∈ C ( Sp (T ))

• φ (1) = I .

Moreover, φ is isometric. We call the unique isometric * -homomorphism which sends

f ∈ C ( Sp (T )) to φ ( f ) ∈ B (ℋ) the continuous functional calculus of T . We write f (T ) for
φ ( f ).

Example 0.2

• If f (t ) = a + a t + ⋯ + a t n , then f (T ) = a I + a T + ⋯ + a T n .
0 1 n 0 1 n

• If f (t ) = t̄, then f (T ) = φ ( f ) = φ (z̄ ) = φ (z )* = T *

• If f (t ) = t 1/2 for t ∈ ℝ+ and T 0, then we define T 1/2 = f (T ). Equivalently, T 1/2 is the

unique positive square root of T .

• If f (t ) = t -1/2 for t > 0 and T > 0, then we define T -1/2 = f (T ). Equivalently,

T -1/2 = (T 1/2) = (T -1) .
-1 1/2

A continuous real-valued function f on an interval I is called an operator monotone function if

one of the following equivalent conditions holds:

• A B ⇒ f ( A) f ( B ) for all Hermitian matrices A, B of all orders whose spectrums are con‐
tained in I ;

• A B ⇒ f ( A) f ( B ) for all Hermitian operators A, B ∈ B (ℋ) whose spectrums are con‐

tained in I and for an infinite dimensional Hilbert space ℋ;

• A B ⇒ f ( A) f ( B ) for all Hermitian operators A, B ∈ B (ℋ) whose spectrums are con‐

tained in I and for all Hilbert spaces ℋ.
4 Linear Algebra

This concept is introduced in [15]; see also [14], [16], [17], [18]. Every operator monotone
function is always continuously differentiable and monotone increasing. Here are examples
of operator monotone functions:
• t ↦ αt + β on ℝ, for α 0 and β ∈ ℝ,

• t ↦ - t -1 on (0, ∞ ),

• t ↦ (c - t )-1 on (a, b), for c ∉ (a, b),

• t ↦ log t on (0, ∞ ),

• t ↦ (t - 1) / log t on ℝ+, where 0 ↦ 0 and 1 ↦ 1.

The next result is called the Löwner-Heinz's inequality [15].

Theorem 0.3 For A, B ∈ B (ℋ)+ and r ∈ 0, 1 , if A B, then A r B r . That is the map t ↦ t r is

an operator monotone function on ℝ+ for any r ∈ 0, 1 .

A key result about operator monotone functions is that there is a one-to-one correspondence
between nonnegative operator monotone functions on ℝ+ and finite Borel measures on
0, ∞ via integral representations. We give a variation of this result in the next proposition.

Proposition 0.4 A continuous function f : ℝ+ → ℝ+ is operator monotone if and only if there

exists a finite Borel measure μ on 0, 1 such that

f (x) = ∫ 1 !t x dμ (t ), x ∈ ℝ +. (id22)
0,1

Here, the weighed harmonic mean !t is defined for a, b > 0 by

-1
a !t b = (1 - t )a -1 + tb -1 (id23)

and extended to a, b 0 by continuity. Moreover, the measure μ is unique. Hence, there is a

one-to-one correspondence between operator monotone functions on the non-negative reals
and finite Borel measures on the unit interval.

Recall that a continuous function f : ℝ+ → ℝ+ is operator monotone if and only if there ex‐
ists a unique finite Borel measure ν on 0, ∞ such that

f (x) = ∫ φx (λ ) dν (λ ), x ∈ ℝ+ ()
0,∞

where

x (λ + 1)
φx (λ ) = for λ > 0, φx (0) = 1, φx (∞ ) = x. ()
x+λ
 Operator Means and Applications 5
http://dx.doi.org/10.5772/46479

t
Consider the Borel measurable function ψ : 0, 1 → 0, ∞ , t ↦ 1-t . Then, for each x ∈ ℝ+,

∫ φx (λ ) dν (λ ) =∫ φx ψ (t ) dνψ (t )
0,∞ 0,1
x
=∫ dνψ (t ) ()
0,1 x - xt + t
=∫ 1 !t x dνψ (t ).
0,1

Now, set μ = νψ. Since ψ is bijective, there is a one-to-one corresponsence between the finite
Borel measures on 0, ∞ of the form ν and the finite Borel measures on 0, 1 of the form νψ.
The map f ↦ μ is clearly well-defined and bijective.

3. Parallel sum: A notion from electrical networks

In connections with electrical engineering, Anderson and Duffin [2] defined the parallel sum
of two positive definite matrices A and B by

A : B = ( A -1 + B -1) .
-1
(id24)

The impedance of an electrical network can be represented by a positive (semi)definite ma‐

trix. If A and B are impedance matrices of multi-port networks, then the parallel sum A : B
indicates the total impedance of two electrical networks connected in parallel. This notion
plays a crucial role for analyzing multi-port electrical networks because many physical in‐
terpretations of electrical circuits can be viewed in a form involving parallel sums. This is a
starting point of the study of matrix and operator means. This notion can be extended to in‐
vertible positive operators by the same formula.

Lemma 0.5 Let A, B, C, D, An , Bn ∈ B (ℋ)++ for all n ∈ ℕ.

• If A ↓ A, then A -1↑ A -1. If A ↑ A, then A -1↓ A -1.

n n n n

• If A C and B D, then A : B C : D.

• If An ↓ A and Bn ↓ B, then An : Bn ↓ A : B.

• If An ↓ A and Bn ↓ B, then lim An : Bn exists and does not depend on the choices of An , Bn .

(1) Assume An ↓ A. Then An-1 is increasing and, for each x ∈ ℋ,

( An-1 - A -1) x, x = ( A - An ) A -1 x, An-1 x ∥ ( A - An ) A -1 x ∥ ∥ An-1 ∥ ∥ x ∥ → 0. ()

(2) Follow from (1).

6 Linear Algebra

(3) Let An , Bn ∈ B (ℋ)++ be such that An ↓ A and Bn ↓ A where A, B > 0. Then An-1↑ A -1 and
Bn-1↑ B -1. So, An-1 + Bn-1 is an increasing sequence in B (ℋ)+ such that

An-1 + Bn-1 → A -1 + B -1, ()

i.e. An-1 + Bn-1↑ A -1 + B -1. By (1), we thus have ( An-1 + Bn-1)-1↓ ( A -1 + B -1) .
-1

(4) Let An , Bn ∈ B (ℋ)++ be such that An ↓ A and Bn ↓ B. Then, by (2), An : Bn is a decreasing

sequence of positive operators. The order completeness of B (ℋ) guaruntees the existence of
the strong limit of An : Bn . Let An' and Bn' be another sequences such that An' ↓ A and Bn' ↓ B.
Note that for each n, m ∈ ℕ, we have An An + Am' - A and Bn Bn + Bm' - B. Then

An : Bn ( An + Am' - A) : ( Bn + Bm' - B ). ()

Note that as n → ∞, An + Am' - A → Am' and Bn + Bm' - B → Bm' . We have that as n → ∞,

( An + Am' - A) : ( Bn + Bm' - B) → Am' : Bm' . ()

Hence, limn→∞ An : Bn Am' : Bm' and limn→∞ An : Bn limm→∞ Am' : Bm' . By symmetry,
limn→∞ An : Bn limm→∞ Am' : Bm' .

We define the parallel sum of A, B 0 to be

A : B = lim ↓0 ( A + I ) : ( B + I ) (id30)

where the limit is taken in the strong-operator topology.

Lemma 0.6 For each x ∈ ℋ,

( A : B ) x, x = inf { Ay, y + Bz, z : y, z ∈ ℋ, y + z = x }. (id32)

First, assume that A, B are invertible. Then for all x, y ∈ ℋ,

Ay, y + B ( x - y ), x - y - ( A : B ) x, x
= Ay, y + Bx, x - 2Re Bx, y + By, y - ( B - B ( A + B )-1 B ) x, x
= ( A + B ) y, y - 2Re Bx, y + ( A + B )-1 Bx, Bx ()
2 2
= ∥ ( A + B )1/2 y ∥ - 2Re Bx, y + ∥ ( A + B )-1/2 Bx ∥
0.
 Operator Means and Applications 7
http://dx.doi.org/10.5772/46479

With y = ( A + B )-1 Bx, we have

Ay, y + B ( x - y ), x - y - ( A : B ) x, x = 0. ()

Hence, we have the claim for A, B > 0. For A, B 0, consider A + I and B + I where ↓ 0.

Remark 0.7 This lemma has a physical interpretation, called the Maxwell's minimum power
principle. Recall that a positive operator represents the impedance of a electrical network
while the power dissipation of network with impedance A and current x is the inner prod‐
uct Ax, x . Consider two electrical networks connected in parallel. For a given current in‐
put x, the current will divide x = y + z, where y and z are currents of each network, in such
a way that the power dissipation is minimum.

Theorem 0.8 The parallel sum satisfies

• monotonicity: A1 A2, B1 B2 ⇒ A1 : B1 A2 : B2.

• transformer inequality: S *( A : B )S (S * AS ) : (S * BS ) for every S ∈ B (ℋ).

• continuity from above: if An ↓ A and Bn ↓ B, then An : Bn ↓ A : B.

(1) The monotonicity follows from the formula (▭) and Lemma ▭(2).

(2) For each x, y, z ∈ ℋ such that x = y + z, by the previous lemma,

S *( A : B )Sx, x = ( A : B )Sx, Sx
ASy, Sy + S * BSz, z ()
= S * ASy, y + S * BSz, z .

Again, the previous lemma assures S *( A : B )S (S * AS ) : (S * BS ).

(3) Let An and Bn be decreasing sequences in B (ℋ)+ such that An ↓ A and Bn ↓ B. Then An : Bn
is decreasing and A : B An : Bn for all n ∈ ℕ. We have that, by the joint monotonicity of par‐
allel sum, for all > 0

An : Bn ( An + I ) : ( Bn + I ). ()

Since An + I ↓ A + I and Bn + I ↓ B + I , by Lemma 3.1.4(3) we have An : Bn ↓ A : B.

Remark 0.9 The positive operator S * AS represents the impedance of a network connected to
a transformer. The transformer inequality means that the impedance of parallel connection
with transformer first is greater than that with transformer last.

Proposition 0.10 The set of positive operators on ℋ is a partially ordered commutative sem‐
igroup with respect to the parallel sum.
8 Linear Algebra

For A, B, C > 0, we have ( A : B ) : C = A : ( B : C ) and A : B = B : A. The continuity from

above in Theorem ▭ implies that ( A : B ) : C = A : ( B : C ) and A : B = B : A for all A, B, C 0.
The monotonicity of the parallel sum means that the positive operators form a partially or‐
dered semigroup.

Theorem 0.11 For A, B, C, D 0, we have the series-parallel inequality

( A + B ) : (C + D ) A : C + B : D. (id41)

In other words, the parallel sum is concave.

For each x, y, z ∈ ℋ such that x = y + z, we have by the previous lemma that

( A : C + B : D ) x, x = ( A : C ) x, x + ( B : D ) x, x
Ay, y + Cz, z + By, y + Dz, z ()
= ( A + B ) y, y + (C + D )z, z .

Applying the previous lemma yields ( A + B ) : (C + D ) A : C + B : D.

Remark 0.12 The ordinary sum of operators represents the total impedance of two networks
with series connection while the parallel sum indicates the total impedance of two networks
with parallel connection. So, the series-parallel inequality means that the impedance of a ser‐
ies-parallel connection is greater than that of a parallel-series connection.

4. Classical means: arithmetic, harmonic and geometric means

Some desired properties of any object that is called a “mean” M on B (ℋ)+ should have are
given here.
• positivity: A, B 0 ⇒ M ( A, B ) 0;

• monotonicity: A A ', B B ' ⇒ M ( A, B ) M ( A ', B ');

• positive homogeneity: M (kA, kB ) = kM ( A, B ) for k ∈ ℝ+;

• transformer inequality: X *M ( A, B ) X M ( X * AX , X * BX ) for X ∈ B (ℋ);

• congruence invariance: X *M ( A, B ) X = M ( X * AX , X * BX ) for invertible X ∈ B (ℋ);

• concavity: M (tA + (1 - t ) B, t A ' + (1 - t ) B ') tM ( A, A ') + (1 - t )M ( B, B ') for t ∈ 0, 1 ;

• continuity from above: if An ↓ A and Bn ↓ B, then M ( An , Bn ) ↓ M ( A, B );

• betweenness: if A B, then A M ( A, B ) B;

• fixed point property: M ( A, A) = A.

 Operator Means and Applications 9
http://dx.doi.org/10.5772/46479

In order to study matrix or operator means in general, the first step is to consider three clas‐
sical means in mathematics, namely, arithmetic, geometric and harmonic means.

The arithmetic mean of A, B ∈ B (ℋ)+ is defined by

1
A ▿ B= ( A + B ). (id52)
2

Then the arithmetic mean satisfies the properties (A1)–(A9). In fact, the properties (A5) and
(A6) can be replaced by a stronger condition:

X *M ( A, B ) X = M ( X * AX , X * BX ) for all X ∈ B (ℋ).

Moreover, the arithmetic mean satisfies

affinity: M (kA + C, kB + C ) = kM ( A, B ) + C for k ∈ ℝ+.

Define the harmonic mean of positive operators A, B ∈ B (ℋ)+ by

A ! B = 2( A : B ) = lim ↓0 2( A -1 + B -1)-1 (id53)

where A ≡ A + I and B ≡ B + I . Then the harmonic mean satisfies the properties (A1)–(A9).

The geometric mean of matrices is defined in [7] and studied in details in [3]. A usage of
congruence transformations for treating geometric means is given in [19]. For a given inver‐
tible operator C ∈ B (ℋ), define

ΓC : B (ℋ)sa → B (ℋ)sa, A ↦ C * AC. ()

Then each ΓC is a linear isomorphism with inverse ΓC -1 and is called a congruence transforma‐
tion. The set of congruence transformations is a group under multiplication. Each congru‐
ence transformation preserves positivity, invertibility and, hence, strictly positivity. In fact,
ΓC maps B (ℋ)+ and B (ℋ)++ onto themselves. Note also that ΓC is order-preserving.

Define the geometric mean of A, B > 0 by

A # B = A 1/2( A -1/2 BA -1/2)

1/2
A 1/2 = ΓA 1/2 ΓA1/2-1/2( B ). (id54)

Then A # B > 0 for A, B > 0. This formula comes from two natural requirements: This defi‐
nition should coincide with the usual geometric mean in ℝ+: A # B = ( AB )1/2 provided that
AB = BA. The second condition is that, for any invertible T ∈ B (ℋ),

T *( A # B )T = (T * AT ) # (T * BT ). (id55)
10 Linear Algebra

The next theorem characterizes the geometric mean of A and B in term of the solution of a
certain operator equation.

Theorem 0.13 For each A, B > 0, the Riccati equation ΓX ( A -1) : = X A -1 X = B has a unique
positive solution, namely, X = A # B.

The direct computation shows that ( A # B ) A -1( A # B ) = B. Suppose there is another posi‐
tive solution Y 0. Then

( A -1/2 X A -1/2)2 = A -1/2 X A -1 X A -1/2 = A -1/2Y A -1Y A -1/2 = ( A -1/2Y A -1/2)2. ()

The uniqueness of positive square roots implies that A -1/2 X A -1/2 = A -1/2Y A -1/2, i.e., X = Y .

Theorem 0.14 (Maximum property of geometric mean) For A, B > 0,

A # B = max { X 0 : X A -1 X B } (id58)

where the maximum is taken with respect to the positive semidefinite ordering.

If X A -1 X B, then

( A -1/2 X A -1/2)2 = A -1/2 X A -1 X A -1/2 A -1/2 BA -1/2 ()

and A -1/2 X A -1/2 ( A -1/2 BA -1/2)

1/2
i.e. X A # B by Theorem ▭.

Recall the fact that if f : a, b → iscontinuousandAn Awith Sp(An) [a,b] foralln N, thenSp(A)
[a,b]andf(An) f(A).

Lemma 0.15 Let A, B, C, D, An , Bn ∈ B (ℋ)++ for all n ∈ ℕ.

• If A C and B D, then A # B C # D.

• If An ↓ A and Bn ↓ B, then An # Bn ↓ A # B.

• If An ↓ A and Bn ↓ B, then lim An # Bn exists and does not depend on the choices of An , Bn .

(1) The extremal characterization allows us to prove only that ( A # B )C -1( A # B ) D. In‐
deed,

= A 1/2( A -1/2 BA -1/2) A 1/2C -1 A 1/2( A -1/2 BA -1/2)

1/2 1/2
( A # B )C -1( A # B ) A 1/2
A 1/2( A -1/2 BA -1/2) A 1/2 A -1 A 1/2( A -1/2 BA -1/2)
1/2 1/2
A 1/2 ()
=B
D.
 Operator Means and Applications 11
http://dx.doi.org/10.5772/46479

(2) Assume An ↓ A and Bn ↓ B. Then An # Bn is a decreasing sequence of strictly positive op‐

erators which is bounded below by 0. The order completeness of B (ℋ) implies that this se‐
quence converges strongly to a positive operator. Since An-1 A -1, the Löwner-Heinz's
inequality assures that An-1/2 A -1/2 and hence ∥ An-1/2 ∥ ∥ A -1/2 ∥ for all n ∈ ℕ. Note also
that ∥ Bn ∥ ∥ B1 ∥ for all n ∈ ℕ. Recall that the multiplication is jointly continuous in the
strong-operator topology if the first variable is bounded in norm. So, An-1/2 Bn An-1/2 converges
strongly to A -1/2 BA -1/2. It follows that

( An-1/2 Bn An-1/2)1/2 → ( A -1/2 BA -1/2)1/2. ()

Since An1/2 is norm-bounded by ∥ A 1/2 ∥ by Löwner-Heinz's inequality, we conclude that

An1/2( An-1/2 Bn An-1/2)1/2 An1/2 → A 1/2( A -1/2 BA -1/2)

1/2
A 1/2. ()

The proof of (3) is just the same as the case of harmonic mean.

We define the geometric mean of A, B 0 by

A # B = lim ↓0 ( A + I ) # ( B + I ). (id63)

Then A # B 0 for any A, B 0.

Theorem 0.16 The geometric mean enjoys the following properties

• monotonicity: A1 A2, B1 B2 ⇒ A1 # B1 A2 # B2.

• continuity from above: An ↓ A, Bn ↓ B ⇒ An # Bn ↓ A # B.

• fixed point property: A # A = A.

• self-duality: ( A # B )-1 = A -1 # B -1.

• symmetry: A # B = B # A.

• congruence invariance: ΓC ( A) # ΓC ( B ) = ΓC ( A # B ) for all invertible C.

(1) Use the formula (▭) and Lemma ▭ (1).

(2) Follows from Lemma ▭ and the definition of the geometric mean.

(3) The unique positive solution to the equation X A -1 X = A is X = A.

(4) The unique positive solution to the equation X -1 A -1 X -1 = B is X -1 = A # B. But this

equstion is equivalent to XAX = B -1. So, A -1 # B -1 = X = ( A # B )-1.
12 Linear Algebra

(5) The equation X A -1 X = B has the same solution to the equation X B -1 X = A by taking in‐
verse in both sides.
(6) We have

ΓC ( A # B )(ΓC ( A))-1ΓC ( A # B ) = ΓC ( A # B )ΓC -1( A -1)ΓC ( A # B )

= ΓC (( A # B ) A -1( A # B )) ()
= ΓC ( B ).

Then apply Theorem ▭.

The congruence invariance asserts that ΓC is an isomorphism on B (ℋ)++ with respect to the
operation of taking the geometric mean.
Lemma 0.17 For A > 0 and B 0, the operator

( ) A C
C* B
()

is positive if and only if B - C * A -1C is positive, i.e., B C * A -1C.

By setting

X =
0
(
I - A -1C
I
, ) ()

we compute

ΓX ( ) (
A C
C* B
=
I
-C * A -1
)( )(
0 A C I - A -1C
I C* B 0 I
)
( )
()
A 0
= .
0 B - C * A -1C

Since ΓG preserves positivity, we obtain the desired result.

Theorem 0.18 The geometric mean A # B of A, B ∈ B (ℋ)+ is the largest operator

X ∈ B (ℋ)sa for which the operator

( A
X* B
X
) (id73)

is positive.
 Operator Means and Applications 13
http://dx.doi.org/10.5772/46479

By continuity argumeny, we may assume that A, B > 0. If X = A # B, then the operator (▭)
is positive by Lemma ▭. Let X ∈ B (ℋ)sa be such that the operator (▭) is positive. Then
Lemma ▭ again implies that X A -1 X B and

( A -1/2 X A -1/2)2 = A -1/2 X A -1 X A -1/2 A -1/2 BA -1/2. ()

The Löwner-Heinz's inequality forces A -1/2 X A -1/2 ( A -1/2 BA -1/2) . Now, applying ΓA 1/2
1/2

yields X A # B.
Remark 0.19 The arithmetric mean and the harmonic mean can be easily defined for multi‐
variable positive operators. The case of geometric mean is not easy, even for the case of ma‐
trices. Many authors tried to defined geometric means for multivariable positive
semidefinite matrices but there is no satisfactory definition until 2004 in [20].

5. Operator connections

We see that the arithmetic, harmonic and geometric means share the properties (A1)–(A9) in
common. A mean in general should have algebraic, order and topological properties. Kubo
and Ando [10] proposed the following definition:

Definition 0.20 A connection on B (ℋ)+ is a binary operation σ on B (ℋ)+ satisfying the fol‐
lowing axioms for all A, A ', B, B ', C ∈ B (ℋ)+:
• monotonicity: A A ', B B ' ⇒ A σ B A ' σ B '

• transformer inequality: C ( A σ B )C (CAC ) σ (CBC )

• joint continuity from above: if A , B ∈ B (ℋ)+ satisfy A ↓ A and B ↓ B, then A σ B ↓ A σ B.

n n n n n n

The term “connection" comes from the study of electrical network connections.
Example 0.21 The following are examples of connections:
• the left trivial mean ( A, B ) ↦ A and the right trivial mean ( A, B ) ↦ B
• the sum ( A, B ) ↦ A + B
• the parallel sum
• arithmetic, geometric and harmonic means
• the weighed arithmetic mean with weight α ∈ 0, 1 which is defined for each A, B 0 by
A ▿α B = (1 - α ) A + αB

• the weighed harmonic mean with weight α ∈ 0, 1 which is defined for each A, B > 0 by
-1
A !α B = (1 - α ) A -1 + α B -1 and extended to the case A, B 0 by continuity.
14 Linear Algebra

From now on, assume dim ℋ = ∞. Consider the following property:

• separate continuity from above: if A , B ∈ B (ℋ)+ satisfy A ↓ A and B ↓ B, then
n n n n
An σ B ↓ A σ B and A σ Bn ↓ A σ B.

The condition (M3') is clearly weaker than (M3). The next theorem asserts that we can im‐
prove the definition of Kubo-Ando by replacing (M3) with (M3') and still get the same theo‐
ry. This theorem also provides an easier way for checking a binary opertion to be a
connection.

Theorem 0.22 If a binary operation σ on B (ℋ)+ satisfies (M1), (M2) and (M3'), then σ satis‐
fies (M3), that is, σ is a connection.

Denote by OM (ℝ+) the set of operator monotone functions from ℝ+ to ℝ+. If a binary opera‐
tion σ has a property (A), we write σ ∈ BO ( A). The following properties for a binary opera‐
tion σ and a function f : ℝ+ → ℝ+ play important roles:
• : If a projection P ∈ B (ℋ)+ commutes with A, B ∈ B (ℋ)+, then

P ( A σ B ) = ( PA) σ (PB ) = ( A σ B ) P; ()

• : f (t ) I = I σ (tI ) for any t ∈ ℝ+.

Proposition 0.23 The transformer inequality (M2) implies

• Congruence invariance: For A, B 0 and C > 0, C ( AσB )C = (CAC ) σ (CBC );

• Positive homogeneity: For A, B 0 and α ∈ (0, ∞ ), α ( A σ B ) = (αA) σ (αB ).

For A, B 0 and C > 0, we have

C -1 (CAC ) σ (CBC ) C -1 (C -1CACC -1) σ (C -1CBCC -1) = A σ B ()

and hence (CAC ) σ (CBC ) C ( A σ B )C. The positive homogeneity comes from the congruence
invariance by setting C = αI .

Lemma 0.24 Let f : ℝ+ → ℝ+ be an increasing function. If σ satisfies the positive homogenei‐

ty, (M3') and (F), then f is continuous.

To show that f is right continuous at each t ∈ ℝ+, consider a sequence tn in ℝ+ such that
tn ↓ t. Then by (M3')

f (tn ) I = I σ tn I ↓ I σ tI = f (t ) I , ()

i.e. f (tn ) ↓ f (t ). To show that f is left continuous at each t > 0, consider a sequence tn > 0
such that tn ↑ t. Then tn-1↓ t -1 and
 Operator Means and Applications 15
http://dx.doi.org/10.5772/46479

lim tn-1 f (tn ) I = lim tn-1(I σ tn I ) = lim (tn-1 I ) σ I = (t -1 I ) σ I

()
= t -1( I σ tI ) = t -1 f (t )I

Since f is increasing, tn-1 f (tn ) is decreasing. So, t ↦ t -1 f (t ) and f are left continuous.

Lemma 0.25 Let σ be a binary operation on B (ℋ)+ satisfying (M3') and (P). If f : ℝ+ → ℝ+ is
an increasing continuous function such that σ and f satisfy (F), then f ( A) = I σ A for any
A ∈ B (ℋ ) + .

i=1 λi Pi where { Pi }i=1 is an orthogonal family of pro‐

First consider A ∈ B (ℋ)+ in the form ∑m m

jections with sum I and λi > 0 for all i = 1, ⋯ , m. Since each Pi commutes with A, we have
by the property (P) that

IσA = ∑ Pi ( I σ A) = ∑ Pi σ Pi A = ∑ Pi σ λi Pi
()
= ∑ Pi (I σ λi I ) = ∑ f (λi ) Pi = f ( A).

Now, consider A ∈ B (ℋ)+. Then there is a sequence An of strictly positive operators in the
above form such that An ↓ A. Then I σ An ↓ I σ A and f ( An ) converges strongly to f ( A).
Hence, I σ A = lim I σ An = lim f ( An ) = f ( A).

Proof of Theorem ▭: Let σ ∈ BO (M 1, M 2, M 3'). As in [10], the conditions (M1) and (M2) im‐
ply that σ satisfies (P) and there is a function f : ℝ+ → ℝ+ subject to (F). If 0 t1 t2, then by
(M1)

f (t1)I = I σ (t1I ) I σ (t2I ) = f (t2) I , ()

i.e. f (t1) f (t2). The assumption (M3') is enough to guarantee that f is continuous by Lemma
▭. Then Lemma ▭ results in f ( A) = IσA for all A 0. Now, (M1) and the fact that dim ℋ = ∞
yield that f is operator monotone. If there is another g ∈ OM (ℝ+) satisfying (F), then
f (t )I = I σ tI = g (t ) I for each t 0, i.e. f = g. Thus, we establish a well-defined map
σ ∈ BO (M 1, M 2, M 3') ↦ f ∈ OM (ℝ+) such that σ and f satisfy (F).

Now, given f ∈ OM (ℝ+), we construct σ from the integral representation (▭) in Proposition
▭. Define a binary operation σ : B (ℋ)+ × B (ℋ)+ → B (ℋ)+ by

Aσ B = ∫ A !t B dμ (t ) (id95)
0,1

where the integral is taken in the sense of Bochner. Consider A, B ∈ B (ℋ)+ and set
F t = A !t B for each t ∈ 0, 1 . Since A ∥ A ∥ I and B ∥ B ∥ I , we get
16 Linear Algebra

∥A∥ ∥B∥
A !t B ∥ A ∥ I !t ∥B∥I = I. ()
t ∥ A ∥ + (1 - t ) ∥ B ∥

By Banach-Steinhaus' theorem, there is an M > 0 such that ∥ F t ∥ M for all t ∈ 0, 1 .

Hence,

∫ ∥ F t ∥ dμ (t ) ∫ M dμ (t ) < ∞. ()
0,1 0,1

So, F t is Bochner integrable. Since F t 0 for all t ∈ 0, 1 , ∫ F t dμ (t ) 0. Thus, A σ B is a well-

0,1

defined element in B (ℋ)+. The monotonicity (M1) and the transformer inequality (M2) come
from passing the monotonicity and the transformer inequality of the weighed harmonic
mean through the Bochner integral. To show (M3'), let An ↓ A and Bn ↓ B. Then
An !t B ↓ A !t B for t ∈ 0, 1 by the monotonicity and the separate continuity from above of
the weighed harmonic mean. Let ξ ∈ H . Define a bounded linear map Φ : B (ℋ) → by(T) =
T , .Foreach n N, setTn(t) = An !t BandputT(t) = A !t B.Thenforeach n N{}, Tn
isBochnerintegrableandSinceTn(t) T (t), wehavethatTn(t) ,T(t) ,asn foreach t [0,1]
.WeobtainfromthedominatedconvergencetheoremthatSo, An B A B.Similarly, A Bn A B .Thus,
satisfies (M 3').Itiseasytoseethatf(t) I = I (t I ) fort 0.Thisshowsthatthemapfissurjective.To show the
injectivity of this map, let σ1, σ2 ∈ BO (M 1, M 2, M 3') be such that σi ↦ f where, for each
t 0, I σi (tI ) = f (t ) I , i = 1, 2. Since σi satisfies the property (P), we have I σi A = f ( A) for A 0
by Lemma ▭. Since σi satisfies the congruence invariance, we have that for A > 0 and B 0,

A σi B = A 1/2( I σi A -1/2 BA -1/2) A 1/2 = A 1/2 f ( A -1/2 BA -1/2) A 1/2, i = 1, 2. ()

For each A, B 0, we obtain by (M3') that

A σ1 B = lim ↓0 A σ1 B

= lim ↓0 A 1/2(I σ1 A -1/2 BA -1/2) A 1/2

= lim ↓0 A 1/2 f ( A -1/2 BA -1/2) A 1/2

()
= lim ↓0 A 1/2(I σ2 A -1/2 BA -1/2) A 1/2
= lim ↓0 A σ2 B
= A σ2 B,

where A ≡ A + I . That is σ1 = σ2. Therefore, there is a bijection between OM (ℝ+) and

BO (M 1, M 2, M 3'). Every element in BO (M 1, M 2, M 3') admits an integral representation
(▭). Since the weighed harmonic mean possesses the joint continuity (M3), so is any element
in BO (M 1, M 2, M 3'). □
 Operator Means and Applications 17
http://dx.doi.org/10.5772/46479

The next theorem is a fundamental result of [10].

Theorem 0.26 There is a one-to-one correspondence between connections σ and operator

monotone functions f on the non-negative reals satisfying

f (t )I = I σ (tI ), t ∈ ℝ+. (id97)

There is a one-to-one correspondence between connections σ and finite Borel measures ν on

0, ∞ satisfying

t +1
Aσ B =∫ (tA : B ) dν (t ), A, B 0. (id98)
0,∞ t

Moreover, the map σ ↦ f is an affine order-isomorphism between connections and non-

negative operator monotone functions on ℝ+. Here, the order-isomorphism means that
when σ ↦ f i for i = 1, 2, A σ B A σ B for all A, B ∈ B (ℋ)+ if and only if f 1 f 2.
i 1 2

Each connection σ on B (ℋ)+ produces a unique scalar function on ℝ+, denoted by the same
notation, satisfying

(s σ t ) I = (sI ) σ (tI ), s, t ∈ ℝ+. (id99)

Let s, t ∈ ℝ+. If s > 0, then s σ t = sf (t / s ). If t > 0, then s σ t = tf (s / t ).

Theorem 0.27 There is a one-to-one correspondence between connections and finite Borel
measures on the unit interval. In fact, every connection takes the form

Aσ B =∫ A !t B dμ (t ), A, B 0 (id101)
0,1

for some finite Borel measure μ on 0, 1 . Moreover, the map μ ↦ σ is affine and order-pre‐
serving. Here, the order-presering means that when μi ↦ σi (i=1,2), if μ1(E ) μ2(E ) for all Bor‐
el sets E in 0, 1 , then A σ1 B A σ2 B for all A, B ∈ B (ℋ)+.

The proof of the first part is contained in the proof of Theorem ▭. This map is affine because
of the linearity of the map μ ↦ ∫ f dμ on the set of finite positive measures and the bijective
correspondence between connections and Borel measures. It is straight forward to show that
this map is order-preserving.

Remark 0.28 Let us consider operator connections from electrical circuit viewpoint. A gener‐
al connection represents a formulation of making a new impedance from two given impe‐
dances. The integral representation (▭) shows that such a formulation can be described as a
weighed series connection of (infinite) weighed harmonic means. From this point of view,
18 Linear Algebra

the theory of operator connections can be regarded as a mathematical theory of electrical cir‐
cuits.

Definition 0.29 Let σ be a connection. The operator monotone function f in (▭) is called the
representing function of σ. If μ is the measure corresponds to σ in Theorem ▭, the measure
μψ -1 that takes a Borel set E in 0, ∞ to μ (ψ -1(E )) is called the representing measure of σ in
the Kubo-Ando's theory. Here, ψ : 0, 1 → 0, ∞ is a homeomorphism t ↦ t / (1 - t ).

Since every connection σ has an integral representation (▭), properties of weighed harmonic
means reflect properties of a general connection. Hence, every connection σ satisfies the fol‐
lowing properties for all A, B 0, T ∈ B (ℋ) and invertible X ∈ B (ℋ):
• transformer inequality: T *( A σ B )T (T * AT ) σ (T * BT );
• congruence invariance: X *( A σ B ) X = ( X * AX ) σ ( X * BX );

• concavity: (tA + (1 - t ) B ) σ (t A ' + (1 - t ) B ') t ( A σ A ') + (1 - t )( B σ B ') for t ∈ 0, 1 .

Moreover, if A, B > 0,

A σ B = A 1/2 f ( A -1/2 BA -1/2) A 1/2 (id107)

and, in general, for each A, B 0,

A σ B = lim ↓0 A σ B (id108)

where A ≡ A + I and B ≡ B + I . These properties are useful tools for deriving operator in‐
equalities involving connections. The formulas (▭) and (▭) give a way for computing the
formula of connection from its representing function.

Example 0.30
• The left- and the right-trivial means have representing functions given by t ↦ 1 and
t ↦ t, respectively. The representing measures of the left- and the right-trivial means are
given respectively by δ0 and δ∞ where δx is the Dirac measure at x. So, the α-weighed
arithmetic mean has the representing function t ↦ (1 - α ) + αt and it has (1 - α )δ0 + αδ∞ as
the representing measure.

• The geometric mean has the representing function t ↦ t 1/2.

• The harmonic mean has the representing function t ↦ 2t / (1 + t ) while t ↦ t / (1 + t ) corr‐

sponds to the parallel sum.
Remark 0.31 The map σ ↦ μ, where μ is the representing measure of σ, is not order-pre‐
serving in general. Indeed, the representing measure of ▿ is given by μ = (δ0 + δ∞ ) / 2 while
the representing measure of ! is given by δ1. We have ! ▿ but δ1μ.
 Operator Means and Applications 19
http://dx.doi.org/10.5772/46479

6. Operator means

According to [1], a (scalar) mean is a binary operation M on (0, ∞ ) such that M (s, t ) lies be‐
tween s and t for any s, t > 0. For a connection, this property is equivalent to various proper‐
ties in the next theorem.

Theorem 0.32 The following are equivalent for a connection σ on B (ℋ)+:

• σ satisfies the betweenness property, i.e. A B ⇒ A A σ B B.

• σ satisfies the fixed point property, i.e. A σ A = A for all A ∈ B (ℋ)+.

• σ is normalized, i.e. I σ I = I .

• the representing function f of σ is normalized, i.e. f (1) = 1.

• the representing measure μ of σ is normalized, i.e. μ is a probability measure.

Clearly, (i) ⇒ (iii) ⇒ (iv). The implication (iii) ⇒ (ii) follows from the congruence invari‐
ance and the continuity from above of σ. The monotonicity of σ is used to prove (ii) ⇒ (i).
Since

IσI =∫ I !t I dμ (t ) = μ ( 0, 1 )I , ()
0,1

we obtain that (iv) ⇒ (v) ⇒ (iii).

Definition 0.33 A mean is a connection satisfying one, and thus all, of the properties in the
previous theorem.

Hence, every mean in Kubo-Ando's sense satisfies the desired properties (A1)–(A9) in Sec‐
tion 3. As a consequence of Theorem ▭, a convex combination of means is a mean.

Theorem 0.34 Given a Hilbert space ℋ, there exist affine bijections between any pair of the
following objects:
• the means on B (ℋ)+,

• the operator monotone functions f : ℝ+ → ℝ+ such that f (1) = 1,

• the probability Borel measures on 0, 1 .

Moreover, these correspondences between (i) and (ii) are order isomorphic. Hence, there ex‐
ists an affine order isomorphism between the means on the positive operators acting on dif‐
ferent Hilbert spaces.

Follow from Theorems ▭ and ▭.

Example 0.35 The left- and right-trivial means, weighed arithmetic means, the geometric
mean and the harmonic mean are means. The parallel sum is not a mean since its represent‐
ing function is not normalized.
20 Linear Algebra

Example 0.36 The function t ↦ t α is an operator monotone function on ℝ+ for each

α ∈ 0, 1 by the Löwner-Heinz's inequality. So it produces a mean, denoted by #α , on
B (ℋ)+. By the direct computation,

s #α t = s 1-α t α , (id127)

i.e. #α is the α-weighed geometric mean on ℝ+. So the α-weighed geometric mean on ℝ+ is
really a Kubo-Ando mean. The α-weighed geometric mean on B (ℋ)+ is defined to be the mean
corresponding to that mean on ℝ+. Since t α has an integral expression

sin απ ∞ tλ α-1
tα = ∫0 t + λ dm(λ ) (id128)
π

(see [14]) where m denotes the Lebesgue measure, the representing measure of #α is given
by

sin απ λ α-1
dμ (λ ) = dm(λ ). (id129)
π λ+1

Example 0.37 Consider the operator monotone function

t
t↦ ( , t 0, α ∈ 0, 1 . ()
1 - α )t + α

The direct computation shows that

s !α t = {((1 - α )s -1 + αt -1)-1, s, t > 0;

0, otherwise,
(id131)

which is the α-weighed harmonic mean. We define the α-weighed harmonic mean on B (ℋ)+ to
be the mean corresponding to this operator monotone function.

Example 0.38 Consider the operator monotone function f (t ) = (t - 1) / log t for t > 0, t ≠ 1,
f (0) ≡ 0 and f (1) ≡ 1. Then it gives rise to a mean, denoted by λ, on B (ℋ)+. By the direct
computation,

sλt = {
s,
0,
s-t
log s - log t
, s > 0, t > 0, s ≠ t;

s=t
otherwise,
(id133)
 Operator Means and Applications 21
http://dx.doi.org/10.5772/46479

i.e. λ is the logarithmic mean on ℝ+. So the logarithmic mean on ℝ+ is really a mean in Ku‐
bo-Ando's sense. The logarithmic mean on B (ℋ)+ is defined to be the mean corresponding to
this operator monotone function.

Example 0.39 The map t ↦ (t r + t 1-r ) / 2 is operator monotone for any r ∈ 0, 1 . This func‐
tion produces a mean on B (ℋ)+. The computation shows that

s r t 1-r + s 1-r t r
(s, t ) ↦ . ()
2

However, the corresponding mean on B (ℋ)+ is not given by the formula

A r B 1-r + A 1-r B r
( A, B ) ↦ (id135)
2

since it is not a binary operation on B (ℋ)+. In fact, the formula (▭) is considered in [21],
called the Heinz mean of A and B.

Example 0.40 For each p ∈ - 1, 1 and α ∈ 0, 1 , the map

p 1/ p ()
t ↦ (1 - α ) + αt

is an operator monotone function on ℝ+. Here, the case p = 0 is understood that we take lim‐
it as p → 0. Then

1/ p
s # p,α t = (1 - α )s p + αt p . (id137)

The corresponding mean on B (ℋ)+ is called the quasi-arithmetic power mean with parameter
( p, α ), defined for A > 0 and B 0 by

A # p,α B = A 1/2 (1 - α ) I + α ( A -1/2 BA -1/2)

p 1/ p
A 1/2. (id138)

The class of quasi-arithmetic power means contain many kinds of means: The mean #1,α is
the α-weighed arithmetic mean. The case #0,α is the α-weighed geometric mean. The case
#-1,α is the α-weighed harmonic mean. The mean # p,1/2 is the power mean or binomial mean
of order p. These means satisfy the property that

A # p,α B = B # p,1-α A. (id139)

Moreover, they are interpolated in the sense that for all p, q, α ∈ 0, 1 ,

22 Linear Algebra

( A #r , p B ) #r ,α ( A #r ,q B ) = A #r ,(1-α ) p+αq B. (id140)

Example 0.41 If σ1, σ2 are means such that σ1 σ2, then there is a family of means that interpo‐
lates between σ1 and σ2, namely, (1 - α )σ1 + ασ2 for all α ∈ 0, 1 . Note that the map
α ↦ (1 - α )σ1 + ασ2 is increasing. For instance, the Heron mean with weight α ∈ 0, 1 is de‐
fined to be h α = (1 - α ) # + α ▿ . This family is the linear interpolations between the geomet‐
ric mean and the arithmetic mean. The representing function of h α is given by

α
t ↦ (1 - α )t 1/2 + (1 + t ). ()
2

The case α = 2 / 3 is called the Heronian mean in the literature.

7. Applications to operator monotonicity and concavity

In this section, we generalize the matrix and operator monotonicity and concavity in the lit‐
erature (see e.g. [3], [22]) in such a way that the geometric mean, the harmonic mean or spe‐
cific operator means are replaced by general connections. Recall the following terminology.
A continuous function f : I → ℝ is called an operator concave function if

f (tA + (1 - t ) B ) tf ( A) + (1 - t ) f ( B ) ()

for any t ∈ 0, 1 and Hermitian operators A, B ∈ B (ℋ) whose spectrums are contained in
the interval I and for all Hilbert spaces ℋ. A well-known result is that a continuous func‐
tion f : ℝ+ → ℝ+ is operator monotone if and only if it is operator concave. Hence, the maps
t ↦ t r and t ↦ log t are operator concave for r ∈ 0, 1 . Let ℋ and be Hilbert spaces. A
map Φ : B (ℋ) → B () is said to be positive if Φ ( A) 0 whenever A 0. It is called unital if
Φ (I ) = I . We say that a positive map Φ is strictly positive if Φ ( A) > 0 when A > 0. A map Ψ
from a convex subset of B (ℋ)sa to B ()sa is called concave if for each A, B ∈ and t ∈ 0, 1 ,

Ψ (tA + (1 - t ) B ) tΨ ( A) + (1 - t )Ψ ( B ). ()

A map Ψ : B (ℋ)sa → B ()sa is called monotone if A B assures Ψ ( A) Ψ ( B ). So, in particular, the

map A ↦ A r is monotone and concave on B (ℋ)+ for each r ∈ 0, 1 . The map A ↦ log A is
monotone and concave on B (ℋ)++.

Note first that, from the previous section, the quasi-arithmetic power mean
( A, B ) ↦ A # p,α B is monotone and concave for any p ∈ - 1, 1 and α ∈ 0, 1 . In particu‐
lar, the following are monotone and concave:
 Operator Means and Applications 23
http://dx.doi.org/10.5772/46479

• any weighed arithmetic mean,

• any weighed geometric mean,

• any weighed harmonic mean,

• the logarithmic mean,

• any weighed power mean of order p ∈ - 1, 1 .

Recall the following lemma from [22].

Lemma 0.42 (Choi's inequality) If Φ : B (ℋ) → B () is linear, strictly positive and unital, then
for every A > 0, Φ ( A)-1 Φ ( A -1).

Proposition 0.43 If Φ : B (ℋ) → B () is linear and strictly positive, then for any A, B > 0

Φ ( A)Φ ( B )-1Φ ( A) Φ ( AB -1 A). (id149)

For each X ∈ B (ℋ), set Ψ ( X ) = Φ ( A)-1/2Φ ( A 1/2 X A 1/2)Φ ( A)-1/2. Then Ψ is a unital strictly
positive linear map. So, by Choi's inequality, Ψ ( A)-1 Ψ ( A -1) for all A > 0. For each A, B > 0,
we have by Lemma ▭ that

= Ψ ( A -1/2 BA -1/2)
-1
Φ ( A)1/2Φ ( B )-1Φ ( A)1/2
Ψ (( A -1/2 BA -1/2) )
-1 ()
= Φ ( A)-1/2Φ ( AB -1 A)Φ ( A)-1/2.

So, we have the claim.

Theorem 0.44 If Φ : B (ℋ) → B () is a positive linear map which is norm-continuous, then for
any connection σ on B ()+ and for each A, B > 0,

Φ ( A σ B ) Φ ( A) σ Φ ( B ). (id151)

If, addition, Φ is strongly continuous, then (▭) holds for any A, B 0.

First, consider A, B > 0. Assume that Φ is strictly positive. For each X ∈ B (ℋ), set

Ψ ( X ) = Φ ( B )-1/2Φ ( B 1/2 X B 1/2)Φ ( B )-1/2. ()

Then Ψ is a unital strictly positive linear map. So, by Choi's inequality, Ψ (C )-1 Ψ (C -1) for all
C > 0. For each t ∈ 0, 1 , put X t = B -1/2( A !t B ) B -1/2 > 0. We obtain from the previous prop‐
osition that
24 Linear Algebra

Φ ( A !t B ) = Φ ( B )1/2Ψ ( X t )Φ ( B )1/2
Φ ( B )1/2 Ψ ( X t-1) -1Φ ( B )1/2

= Φ ( B ) Φ ( B ((1 - t ) A -1 + t B -1) B ) Φ ( B )
-1
()
= Φ ( B ) (1 - t )Φ ( BA -1 B ) + tΦ ( B ) Φ ( B )
-1

-1
Φ ( B ) (1 - t )Φ ( B )Φ ( A)-1Φ ( B ) + tΦ ( B ) Φ ( B )
= Φ ( A) !t Φ ( B ).

For general case of Φ, consider the family Φ ( A) = Φ ( A) + I where > 0. Since the map
-1
( A, B ) ↦ A !t B = (1 - t ) A -1 + t B -1 is norm-continuous, we arrive at

Φ ( A ! t B ) Φ ( A ) ! t Φ ( B ). ()

For each connection σ, since Φ is a bounded linear operator, we have

Φ(A σ B) = Φ (∫ A !t B dμ (t )) = ∫ Φ ( A !t B ) dμ (t )
0,1 0,1
()
∫ Φ ( A) !t Φ ( B ) dμ (t ) = Φ ( A) σ Φ ( B ).
0,1

Suppose further that Φ is strongly continuous. Then, for each A, B 0,

Φ(A σ B) = Φ (lim ↓0 ( A + I ) σ ( B + I )) = lim ↓0 Φ (( A + I ) σ ( B + I ))

()
lim ↓0 Φ ( A + I ) σ Φ ( B + I ) = Φ ( A) σ Φ ( B ).

The proof is complete.

As a special case, if Φ : M n ( → M n ( is a positive linear map, then for any connection σ and
for any positive semidefinite matrices A, B ∈ M n (, we have

Φ ( AσB ) Φ ( A) σ Φ ( B ). ()

In particular, Φ ( A) # p,α Φ ( B ) Φ ( A) # p,α Φ ( B ) for any p ∈ - 1, 1 and α ∈ 0, 1 .

Theorem 0.45 If Φ1, Φ2 : B (ℋ)+ → B ()+ are concave, then the map

( A1, A2) ↦ Φ1( A1) σ Φ2( A2) (id153)

is concave for any connection σ on B ()+.

Let A1, A1', A2, A2' 0 and t ∈ 0, 1 . The concavity of Φ1 and Φ2 means that for i = 1, 2
 Operator Means and Applications 25
http://dx.doi.org/10.5772/46479

Φi (t Ai + (1 - t ) Ai') tΦi ( Ai ) + (1 - t )Φi ( Ai'). ()

It follows from the monotonicity and concavity of σ that

Φ1(t A1 + (1 - t ) A1') σ Φ2(t A2 + (1 - t ) A2')

tΦ1( A1) + (1 - t )Φ1( A1') σ tΦ2( A2) + (1 - t )Φ2( A2') ()
t Φ1( A1) σ Φ2( A2) + (1 - t ) Φ1( A1) σ Φ2( A2) .

This shows the concavity of the map ( A1, A2) ↦ Φ1( A1) σ Φ2( A2) .

In particular, if Φ1 and Φ2 are concave, then so is ( A, B ) ↦ Φ1( A) # p,α Φ2( B ) for p ∈ - 1, 1

and α ∈ 0, 1 .

Corollary 0.46 Let σ be a connection. Then, for any operator monotone functions
f , g : ℝ+ → ℝ+, the map ( A, B ) ↦ f ( A) σ g ( B ) is concave. In particular,

• the map ( A, B ) ↦ A r σ B s is concave on B (ℋ)+ for any r, s ∈ 0, 1 ,

• the map ( A, B ) ↦ ( log A) σ ( log B ) is concave on B (ℋ)++.

Theorem 0.47 If Φ1, Φ2 : B (ℋ)+ → B ()+ are monotone, then the map

( A1, A2) ↦ Φ1( A1) σ Φ2( A2) (id158)

is monotone for any connection σ on B ()+.

Let A1 A1' and A2 A2'. Then Φ1( A1) Φ1( A1') and Φ2( A2) Φ2( A2') by the monotonicity of Φ1 and
Φ2. Now, the monotonicity of σ forces Φ1( A1) σ Φ2( A2) Φ1( A1') σ Φ2( A2').

In particular, if Φ1 and Φ2 are monotone, then so is ( A, B ) ↦ Φ1( A) # p,α Φ2( B ) for

p ∈ - 1, 1 and α ∈ 0, 1 .

Corollary 0.48 Let σ be a connection. Then, for any operator monotone functions
f , g : ℝ+ → ℝ+, the map ( A, B ) ↦ f ( A) σ g ( B ) is monotone. In particular,

• the map ( A, B ) ↦ A r σ B s is monotone on B (ℋ)+ for any r, s ∈ 0, 1 ,

• the map ( A, B ) ↦ ( log A) σ ( log B ) is monotone on B (ℋ)++.

Corollary 0.49 Let σ be a connection on B (ℋ)+. If Φ1, Φ2 : B (ℋ)+ → B (ℋ)+ is monotone and
strongly continuous, then the map

( A, B ) ↦ Φ1( A) σ Φ2( B ) (id163)

26 Linear Algebra

is a connection on B (ℋ)+. Hence, the map

( A, B ) ↦ f ( A) σ g ( B ) (id164)

is a connection for any operator monotone functions f , g : ℝ+ → ℝ+.

The monotonicity of this map follows from the previous result. It is easy to see that this map
satisfies the transformer inequality. Since Φ1 and Φ2 strongly continuous, this binary opera‐
tion satisfies the (separate or joint) continuity from above. The last statement follows from
the fact that if An ↓ A, then Sp ( An ) ⊆ 0, ∥ A1 ∥ for all n and hence f ( An ) → f ( A).

8. Applications to operator inequalities

In this section, we apply Kubo-Ando's theory in order to get simple proofs of many classical
inequalities in the context of operators.

Theorem 0.50 (AM-LM-GM-HM inequalities) For A, B 0, we have

A ! B A # B A λ B A ▿ B. (id166)

It is easy to see that, for each t > 0, t ≠ 1,

2t t -1 1+t
t 1/2 . ()
1+t log t 2

Now, we apply the order isomorphism which converts inequalities of operator monotone
functions to inequalities of the associated operator connections.

Theorem 0.51 (Weighed AM-GM-HM inequalities) For A, B 0 and α ∈ 0, 1 , we have

A !α B A #α B A ▿α B. (id168)

Apply the order isomorphism to the following inequalities:

t α ()
(1 - α )t + α t 1 - α + αt, t 0.

The next two theorems are given in [23].

Theorem 0.52 For each i = 1, ⋯ , n, let Ai , Bi ∈ B (ℋ)+. Then for each connection σ
 Operator Means and Applications 27
http://dx.doi.org/10.5772/46479

n n n
∑ ( Ai σ Bi ) ∑ Ai σ ∑ Bi . (id170)
i=1 i=1 i=1

Use the concavity of σ together with the induction.

By replacing σ with appropriate connections, we get some interesting inequalities.

(1) Cauchy-Schwarz's inequality: For Ai , Bi ∈ B (ℋ)sa,

n n n
∑ Ai2 # Bi2 ∑ Ai2 # ∑ Bi2. (id171)
i=1 i=1 i=1

(2) Hölder's inequality: For Ai , Bi ∈ B (ℋ)+ and p, q > 0 such that 1 / p + 1 / q = 1,

n n n
∑ Aip #1/ p Biq ∑ Aip #1/ p ∑ Biq . (id172)
i=1 i=1 i=1

(3) Minkowski's inequality: For Ai , Bi ∈ B (ℋ)++,

(∑ ( Ai + Bi )-1) (∑ Ai-1) + (∑ Bi-1) .

n -1 n -1 n -1
(id173)
i=1 i=1 i=1

Theorem 0.53 Let Ai , Bi ∈ B (ℋ)+, i = 1, ⋯ , n, be such that

A1 - A2 - ⋯ - An 0 and B1 - B2 - ⋯ - Bn 0. ()

Then

n n n
A1 σ B1 - ∑ Ai σ Bi ( A1 - ∑ Ai ) σ ( B1 - ∑ Bi ). (id175)
i=2 i=2 i=2

Substitute A1 to A1 - A2 - ⋯ - An and B1 to B1 - B2 - ⋯ - Bn in (▭).

Here are consequences.

(1) Aczél's inequality: For Ai , Bi ∈ B (ℋ)sa, if

A12 - A22 - ⋯ - An2 0 and B12 - B22 - ⋯ - Bn2 0, ()

then
28 Linear Algebra

A12 # B12 - ∑ Ai2 # Bi2 ( A12 - ∑ Ai2) # ( B12 - ∑ Bi2).

n n n
(id176)
i=2 i=2 i=2

(2) Popoviciu's inequality: For Ai , Bi ∈ B (ℋ)+ and p, q > 0 such that 1 / p + 1 / q = 1, if p, q > 0
are such that 1 / p + 1 / q = 1 and

A1p - A2p - ⋯ - Anp 0 and B1q - B2q - ⋯ - Bnq 0, ()

then

A1p #1/ p B1q - ∑ Aip #1/ p Biq ( A1p - ∑ Aip ) #1/ p ( B1q - ∑ Biq).
n n n
(id177)
i=2 i=2 i=2

(3) Bellman's inequality: For Ai , Bi ∈ B (ℋ)++, if

A1-1 - A2-1 - ⋯ - An-1 > 0 and B1-1 - B2-1 - ⋯ - Bn-1 > 0, ()

then

( A1-1 + B1-1) - ∑ ( Ai + Bi )-1 ( A1-1 - ∑ Ai-1) + ( B1-1 - ∑ Bi-1) .

n -1 n -1 n -1
(id178)
i=2 i=2 i=2

The mean-theoretic approach can be used to prove the famous Furuta's inequality as fol‐
lows. We cite [24] for the proof.

Theorem 0.54 (Furuta's inequality) For A B 0, we have

( B r A p B r )1/q B
( p+2r )/q
(id180)
A
( p+2r )/q
( A r B p A r )1/q

where r 0, p 0, q 1 and (1 + 2r )q p + 2r.

By the continuity argument, assume that A, B > 0. Note that (▭) and (▭) are equivalent. In‐
deed, if (▭) holds, then (▭) comes from applying (▭) to A -1 B -1 and taking inverse on both
sides. To prove (▭), first consider the case 0 p 1. We have B p+2r = B r B p B r B r A p B r and
the Löwner-Heinz's inequality (LH) implies the desired result. Now, consider the case p 1
and q = ( p + 2r ) / (1 + 2r ), since (▭) for q > ( p + 2r ) / (1 + 2r ) can be obtained by (LH). Let
f (t ) = t 1/q and let σ be the associated connection (in fact, σ = #1/q ). Must show that, for any
r 0,
 Operator Means and Applications 29
http://dx.doi.org/10.5772/46479

B -2r σ A p B. (id181)

1
For 0 r 2, we have by (LH) that A 2r B 2r and

(1-1/q )
B -2r σ A p A -2r σ A p = A -2r A p/q = A B = B -2r σ B p . ()

1
Now, set s = 2r + 2 and q1 = ( p + 2s ) / (1 + 2s ) 1. Let f 1(t ) = t / 1 and consider the associated
1q

connection σ1. The previous step, the monotonicity and the congruence invariance of con‐
nections imply that

= B -r B - 2r +1 σ1 ( B r A p B r ) B -r
( )
B -2s σ1 A p
-1/q1
B -r ( B r A p B r ) σ1 ( B r A p B r ) B -r
()
= B -r ( B r A p B r )
1/q
B -r
B -r B 1+2r B -r
= B.

Note that the above result holds for A, B 0 via the continuity of a connection. The desired
equation (▭) holds for all r 0 by repeating this process.

Acknowledgement

The author thanks referees for article processing.

Author details

Pattrawut Chansangiam1

1 King Mongkut's Institute of Technology Ladkrabang,, Thailand

References

[1] Anderson, W. & Duffin, R. (1969). Series and parallel addition of matrices, Journal of
Mathematical Analysis and Applications, Vol. 26, 576–594.
[2] Anderson, W. & Morley, T. & Trapp, G. (1990). Positive solutions to X=A-BX -1 B,
Linear Algebra and its Applications, Vol. 134, 53–62.
30 Linear Algebra

[3] Ando, T. (1979). Concavity of certain maps on positive definite matrices and applica‐
tions to Hadamard products, Linear Algebra and its Applications, Vol. 26, 203–241.
[4] Ando, T. (1983). On the arithmetic-geometric-harmonic mean inequality for positive
definite matrices, Linear Algebra and its Applications, Vol. 52-53, 31–37.
[5] Ando, T. & Li, C. & Mathias, R. (2004). Geometric Means, Linear Algebra and its Appli‐
cations, Vol. 385, 305–334.
[6] Bhagwat, K. & Subramanian, A. (1978). Inequalities between means of positive oper‐
ators, Mathematical Proceedings of the Cambridge Philosophical Society, Vol. 83, 393–401.
[7] Bhatia, R. (1997). Matrix Analysis, Springer-Verlag New York Inc., New York.
[8] Bhatia, R. (2006). Interpolating the arithmetic-geometric mean inequality and its op‐
erator version, Linear Algebra and its Applications, Vol. 413, 355–363.
[9] Bhatia, R. (2007). Positive Definite Matrices, Princeton University Press, New Jersey.
[10] Donoghue, W. (1974). Monotone matrix functions and analytic continuation, Springer-
Verlag New York Inc., New York.
[11] Fujii, J. (1978). Arithmetico-geometric mean of operators, Mathematica Japonica, Vol.
23, 667–669.
[12] Fujii, J. (1979). On geometric and harmonic means of positive operators, Mathematica
Japonica, Vol. 24, No. 2, 203–207.
[13] Fujii, J. (1992). Operator means and the relative operator entropy, Operator Theory:
Advances and Applications, Vol. 59, 161–172.
[14] Furuta, T. (1989). A proof via operator means of an order preserving inequality, Line‐
ar Algebra and its Applications, Vol. 113, 129–130.
[15] Hiai, F. (2010). Matrix analysis: matrix monotone functions, matrix means, and majo‐
rizations, Interdisciplinary Information Sciences, Vol. 16, No. 2, 139–248,
[16] Hiai, F. & Yanagi, K. (1995). Hilbert spaces and linear operators, Makino Pub. Ltd.
[17] Kubo, F. & Ando, T. (1980). Means of positive linear operators, Mathematische Annal‐
en, Vol. 246, 205–224.
[18] Lawson, J. & Lim, Y. (2001). The geometric mean, matrices, metrices and more, The
American Mathematical Monthly, Vol. 108, 797–812.
[19] Lim, Y. (2008). On Ando–Li–Mathias geometric mean equations, Linear Algebra and its
Applications, Vol. 428, 1767–1777.
[20] Löwner, C. (1934). Über monotone matrix funktionen. Mathematische Zeitschrift, Vol.
38, 177–216.
[21] Mond, B. & Pec ˇaric ´, J. & S ˇunde, J. & Varos ˇanec, S. (1997). Operator versions of
some classical inequalities, Linear Algebra and its Applications, Vol. 264, 117–126.
 Operator Means and Applications 31
http://dx.doi.org/10.5772/46479

[22] Nishio, K. & Ando, T. (1976). Characterizations of operations derived from network

connections. Journal of Mathematical Analysis and Applications, Vol. 53, 539–549.

[23] Pusz, W. & Woronowicz, S. (1975). Functional calculus for sesquilinear forms and the

purification map, Reports on Mathematical Physics, Vol. 8, 159–170.

[24] Toader, G. & Toader, S. (2005). Greek means and the arithmetic-geometric mean,

RGMIA Monographs, Victoria University, (ONLINE: http://rgmia.vu.edu.au/mono‐

graphs).
32 Linear Algebra