University of Central Florida

STARS

Electronic Theses and Dissertations, 2004-2019

2006

Frames In Hilbert C*-modules

Wu Jing
University of Central Florida

Part of the Mathematics Commons

Find similar works at: https://stars.library.ucf.edu/etd
University of Central Florida Libraries http://library.ucf.edu

This Doctoral Dissertation (Open Access) is brought to you for free and open access by STARS. It has been accepted
for inclusion in Electronic Theses and Dissertations, 2004-2019 by an authorized administrator of STARS. For more
information, please contact [email protected].

STARS Citation
Jing, Wu, "Frames In Hilbert C*-modules" (2006). Electronic Theses and Dissertations, 2004-2019. 899.
https://stars.library.ucf.edu/etd/899
 Frames in Hilbert C ∗ -modules

by

Wu Jing
B.S. Ludong University, 1991
M.S. Qufu Normal University, 1994

A dissertation submitted in partial fulfillment of the requirements

for the degree of Doctor of Philosophy
in the Department of Mathematics
in the College of Sciences
at the University of Central Florida
Orlando, Florida

Summer Term
2006
Major Professor: Deguang Han

c 2006 by Wu Jing

ii
 Abstract

Since the discovery in the early 1950’s, frames have emerged as an important
tool in signal processing, image processing, data compression and sampling theory
etc. Today, powerful tools from operator theory and Banach space theory are
being introduced to the study of frames producing deep results in frame theory.
In recent years, many mathematicians generalized the frame theory from
Hilbert spaces to Hilbert C ∗ -modules and got significant results which enrich
the theory of frames. Also there is growing evidence that Hilbert C ∗ -modules
theory and the theory of wavelets and frames are tightly related to each other in
many aspects. Both research fields can benefit from achievements of the other
field. Our purpose of this dissertation is to work on several basic problems on
frames for Hilbert C ∗ -modules.
We first give a very useful characterization of modular frames which is easy
to be applied. Using this characterization we investigate the modular frames
from the operator theory point of view. A condition under which the removal of
element from a frame in Hilbert C ∗ -modules leaves a frame or a non-frame set
is also given. In contrast to the Hilbert space situation, Riesz bases of Hilbert
C ∗ -modules may possess infinitely many alternative duals due to the existence of
zero-divisors and not every dual of a Riesz basis is again a Riesz basis. We will
present several such examples showing that the duals of Riesz bases in Hilbert
C ∗ -modules are much different and more complicated than the Hilbert space
cases. A complete characterization of all the dual sequences for a Riesz basis,
and a necessary and sufficient condition for a dual sequence of a Riesz basis to
be a Riesz basis are also given. In the case that the underlying C ∗ -algebra is a
commutative W ∗ -algebra, we prove that the set of the Parseval frame generators
for a unitary group can be parameterized by the set of all the unitary operators in

iii
the double commutant of the unitary group. Similar result holds for the set of all
the general frame generators where the unitary operators are replaced by invert-
ible and adjointable operators. Consequently, the set of all the Parseval frame
generators is path-connected. We also prove the existence and uniqueness of the
best Parseval multi-frame approximations for multi-frame generators of unitary
groups on Hilbert C ∗ -modules when the underlying C ∗ -algebra is commutative.
For the dilation results of frames we show that a complete Parseval frame vector
for a unitary group on Hilbert C ∗ -module can be dilated to a complete wander-
ing vector. For any dual frame pair in Hilbert C ∗ -modules, we prove that the
pair are orthogonal compressions of a Riesz basis and its canonical dual basis for
some larger Hilbert C ∗ -module. For the perturbation of frames and Riesz bases
in Hilbert C ∗ -modules we prove that the Casazza-Christensen general perturba-
tion theorem for frames in Hilbert spaces remains valid in Hilbert C ∗ -modules.
In the Hilbert space setting, under the same perturbation condition, the pertur-
bation of any Riesz basis remains a Riesz basis. However, this no longer holds
for Riesz bases in Hilbert C ∗ -modules. We also give a complete characterization
on all the Riesz bases for Hilbert C ∗ -modules such that the perturbation (un-
der Casazza-Christensen’s perturbation condition) of a Riesz basis still remains
a Riesz basis.

iv
To the memory of my mother

v
 Acknowledgments

My first heartfelt thank goes to my advisor, Dr. Deguang Han, who guided
me into such an interesting area of research and gave me plenty of instruction,
encouragement, care and help.
I express sincere thanks to Dr. Ram N. Mohapatra. He was always there to
lend a helping hand to me during my three years’ study at UCF.
My gratitude also to Dr. David R. Larson for his generosity, insights and
help.
I want to express my appreciation to Dr. Michael Frank for many helpful
communications.
I am thankful to Dr. Lee Chow, Dr. Xin Li, Dr. Gary Richardson, Dr. David
Rollins, Dr. Qiyu Sun and Dr. Jiongmin Yong for their help and encouragement.
Having been a Graduate Teaching Assistant for three years in UCF I would
like to thank the professors, my fellow GTA’s, and the students in the classes I
taught, from all of whom I learned a lot.
Last but not least, I wish to thank my wife, Yan, and son, Shiyu, for their
love and endless support.

vi
 Table of Contents

1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 PRELIMINARIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1 Frames in Hilbert Spaces . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Frames in Hilbert Spaces . . . . . . . . . . . . . . . . . . . 9
2.1.2 Riesz Bases in Hilbert Spaces . . . . . . . . . . . . . . . . 13
2.1.3 Dilation Results of Frames . . . . . . . . . . . . . . . . . . 16
2.1.4 Structured Frames in Hilbert Spaces . . . . . . . . . . . . 17
2.2 Hilbert C ∗ -modules . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.1 C ∗ -algebras and W ∗ -algebras . . . . . . . . . . . . . . . . 19
2.2.2 Hilbert C ∗ -modules . . . . . . . . . . . . . . . . . . . . . . 23

3 HILBERT C ∗ -MODULE FRAMES . . . . . . . . . . . . . . . . . 30

3.1 Frames in Hilbert C ∗ -modules . . . . . . . . . . . . . . . . . . . . 30
3.2 An Equivalent Definition of Modular Frames . . . . . . . . . . . . 33
3.3 Characterizations of Frames and Bessel Sequences in Hilbert C ∗ -
modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.4 Removal of Elements from Frames . . . . . . . . . . . . . . . . . . 41

4 DUALS AND MODULAR RIESZ BASES . . . . . . . . . . . . 45

4.1 Characterizations of Riesz Bases in Hilbert C ∗ -modules . . . . . . 45
4.2 Duals of Riesz Bases in Hilbert C ∗ -modules . . . . . . . . . . . . 48

vii
5 STRUCTURED MODULAR FRAMES . . . . . . . . . . . . . . 58
5.1 Modular Frame Vector Parameterizations . . . . . . . . . . . . . . 58
5.2 Parseval Frame Approximations . . . . . . . . . . . . . . . . . . . 67

6 DILATIONS OF FRAMES IN HILBERT C ∗ -MODULES . . . 74

6.1 Dilation of Parseval Frame Vectors . . . . . . . . . . . . . . . . . 74
6.2 Dilation of Dual Modular Frame Pairs . . . . . . . . . . . . . . . 76
6.3 Projective Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

7 PALEY-WIENER TYPE PERTURBATION OF FRAMES AND

RIESZ BASES IN HILBERT C ∗ -MODULES . . . . . . . . . . . . . 85
7.1 Perturbation of Modular Frames . . . . . . . . . . . . . . . . . . . 85
7.2 Perturbation of Modular Riesz Bases . . . . . . . . . . . . . . . . 91

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

viii
 CHAPTER 1

INTRODUCTION

In the study of vector spaces one of the most important concepts is that of a
basis, allowing each element in the space to be written as a linear combination of
the elements in the basis. However, the conditions to a basis are very restrictive:
linear independence between the elements. This makes it hard or even impossible
to find bases satisfying extra conditions, and this is the reason that one might
look for a more flexible substitute.
Frames are such tools. A frame for a vector space equipped with an inner
product also allows each element in the space to be written as a linear combination
of the elements in the frame, but linear independence between the frame elements
is not required.
Frames for Hilbert space were formally defined by Duffin and Schaeffer ([22])
in 1952. They used frames as a tool in the study of nonharmonic Fourier series,
i.e., sequence of the type {eiλn x }n∈Z , where {λn }n∈Z is a family of real or complex
numbers. Apparently, the idea of Duffin and Schaeffer did not seem to generate
much interest outside of nonharmonic Fourier series, and the importance of the
concept was not realized by the mathematical community; at least it took 30
years before the next treatment appeared in print. In 1980 Young wrote his book
[56], which contains the basic facts about frames. Frames were presented in the
abstract setting, and again used in the context of nonharmonic Fourier series.
Then, in 1985, as the wavelet era began, Daubechies, Grossmann and Mayer
([20]) observed that frames can be used to find series expansions of functions in
L2 (R) which are very similar to the expansions using orthonormal bases. This

1
was probably the time when many mathematicians started to see the potential
of the topic. Since then, the theory of frames has been more widely studied.
Frames have been used in signal processing, image processing, data compres-
sion and sampling theory. Today, ever more use are being found for the theory
such as optics, signal detection, as well as the study of Besov spaces in Banach
space theory etc. In the other direction, powerful tools from operator theory and
Banach space theory are being introduced to the study of frames producing deep
results in frame theory. At this very moment, the theory is beginning to grow
rapidly with the host of new people entering the area.
One of the nice things about frame theory is the fact that big portions are
still underdeveloped. Also, many of the extensively developed areas, such as
Weyl-Heisenberg frames and exponential frames, still have many fundamental
open questions to challenge anyone, such as the complete classification of Weyl-
Heisenberg frames or the classification of exponential frame. Another interesting
feature of the area is the broad spectrum of people working in different parts of
it including biologists, engineers, mathematicians, etc. Although each group has
it own interests, there is an opportunity here to interact with a broad spectrum
of researchers.
Recent research also shows that frame theory has strong connections with
some famous results in other aspects of mathematics, for example, the Kadison-
Singer Conjecture in C ∗ -algebra and Naimark Dilation Theorem in operator-
valued measure theory. In frame theory the Feichtinger Conjecture states that
every bounded frame can be written as a finite union of Riesz basic sequences.
Much work has been done on this conjecture in just the last few years. This
is because the conjecture is not just interesting and important for frame theory
but also is connected to the famous Kadison-Singer Conjecture [41], which is
known to be equivalent to the paving conjecture. Recall that the Kadison-Singer
Conjecture, which is still open, states that whether every pure state on D, the
C ∗ -algebra of the diagonal operators on l2 , admits a unique extension to a (pure)
state on B(l2 ), the C ∗ -algebra of all bounded linear operators on l2 . In [12], it was

2
shown that the Kadison-Singer Conjecture implies the Feichtinger Conjecture. It
is unknown whether these two problems are equivalent, but the result in [12]
indicates that they are certainly very close. In particular, it is proved in [12] that
the Feichtinger Conjecture is equivalent to the conjectured generalization of the
Bourgain-Tzafriri Restricted-Invertibility Theorem.
In recent years, many mathematicians generalized the frame theory in Hilbert
spaces to frame theory in Hilbert C ∗ -modules and got significant results which
enrich the theory of frames. Also there is growing evidence that Hilbert C ∗ -
modules theory and the theory of wavelets and frames are tightly related to each
other in many aspects. Both research fields can benefit from achievements of the
other field.
Beside Kasparov’s Stabilization Theorem the inner structure of self-dual Hilbert
W ∗ -modules as described by Paschke in [49] has been another source of inspira-
tion for frames of Hilbert C ∗ -modules. Rephrasing his description in the context
of frames it reads as the proof of the general existence of orthogonal normal-
ized tight frames {xj }j∈J for self-dual Hilbert W ∗ -modules, where additionally
the values {hxj , xj i : j ∈ J} are projections. This point of view was already
realized by Denizeau and Havet ([21]) in 1994. They went one step further by
taking a topologically weak reconstruction formula for normalized tight frames
as a cornerstone to characterize the concept of ”quasi-bases” for Hilbert W ∗ -
modules. These ”quasi-bases” are a special example of module frames in Hilbert
C ∗ -modules. The special frames appearing from Paschke’s result are called ”or-
thogonal bases” by these authors. The two concepts were investigated by them
to the extent of tenser product properties of quasi-bases for C ∗ -correspondences
of W ∗ -algebras (cf. [21]).
Frank and Larson ([23]) defined the standard frames in Hilbert C ∗ -modules
in 1998 and got a series of result for standard frames in finitely or countably
generated Hilbert C ∗ -modules over unital C ∗ -algebras. Note that the frames
exist in abundance in finitely or countably generated Hilbert C ∗ -modules over
unital C ∗ -algebra A as well as in the C ∗ -algebras itself (see [26]). This fact allows

3
us to rely on standard decompositions for elements of Hilbert C ∗ -modules despite
the general absence of orthogonal and orthonormal Riesz bases in them.
Meanwhile, the case of Hilbert C ∗ -modules over non-unital C ∗ -algebras has
been investigated by Raeburn and Thompson ([50]), as well as by Bakić and
Guljaš ([4]) discovering standard frames even for this class of countably generated
Hilbert C ∗ -modules in a well-defined larger multiplier module.
However, many problems about frames in Hilbert C ∗ -modules still have to
be solved. For example, the well-known open problem: Does every Hilbert C ∗ -
module admit a modular frame? These problems are attracting more and more
people to enter this field.
The areas of applications indicate a large potential of problems for the inves-
tigation of which the results of frames in Hilbert C ∗ -modules could be applied.
From the point of view of applied frame theory, the advantage of the generalized
setting of Hilbert C ∗ -modules may consist in the additional degree of freedom
coming from the C ∗ -algebra A of coefficients and its special inner structure, to-
gether with the handling of the basic features of the generalized theory in almost
the same manner as for Hilbert spaces.
The aim of this manuscript is to continue the study of frames in Hilbert
C ∗ -modules. The considerations follow the line of the geometrical and operator-
theoretical approach worked by Han and Larson ([37]) in the main. However,
proofs that generalize from the Hilbert space cases, when attainable, are usually
considerably more difficult for the module case for reasons that do not occur in
the simpler Hilbert space cases.
Let’s describe the chapters in more details. Chapter 2 contains the basic
results of frames in Hilbert spaces and the basic properties of Hilbert C ∗ -modules.
In Chapter 3 we introduce the concept of frames in Hilbert C ∗ -modules. The
basic properties of modular frames are given in Section 3.1. Note that from the
definition of modular frames, it is clear that we need to compare positive elements
in the underlying C ∗ -algebra in order to test whether a sequence is a frame or
not. This usually is not a trivial task. An equivalent definition was established

4
in Section 3.2., which is much easier to be applied. Another advantage of this
equivalent definition is that it allows us to characterize modular frames from the
operator theory point of view which is the goal of Section 3.3. It is very interesting
that if we remove an element from a basis, then we must get a set which is not a
basis. But for frame this is not the case. Due to the redundancy of frame if we
remove an element from a frame we may get a new frame. In Section 3.4 we will
give a characterization of the removal of an element from a modular frame.
The aim of Chapter 4 is to characterize the modular Riesz bases and their
duals. We first give a characterization of Riesz bases in Hilbert C ∗ -modules
in Section 4.1. It is well-known that in Hilbert spaces every Riesz basis has a
unique dual which is also a Riesz basis. But in Hilbert C ∗ -modules, due to the
zero-divisors, not all Riesz bases have unique duals and not every dual is a Riesz
basis. We will present several such examples showing that the duals of Riesz
bases in Hilbert C ∗ -modules are much different and more complicated than the
Hilbert space cases. For example, a dual sequence of a Riesz basis can even not
be a Bessel sequence, and a dual Bessel sequence of a Riesz basis may not be a
Riesz basis. Several examples are provided in Section 4.2 to show the complexity
of duals of modular Riesz bases. We also characterize all the dual sequences for
a Riesz basis. And a necessary and sufficient condition for a dual sequence of a
Riesz basis to be a Riesz basis is given in Section 4.2.
The main purpose of Chapter 5 is to initiate the study of structured frames
for Hilbert C ∗ -modules. In Hilbert space frame theory, structured frames are the
ones that have attracted the most attentions. Typical examples include wavelet
frames, Gabor frames and frames induced by group representations. These frames
are the ones that have been the main focuses in the research of frame theory. In
[19], Dai and Larson introduced one class of structure frames: frames associ-
ated with a system of unitary operators. The systematic study of this kind of
structured frames can be found in the two memoirs papers [19, 37]. In this chap-
ter, we will focus our attention on the frames induced by a group of unitary
operators. More precisely, we work on two closely related issues: frame vector
parameterizations and Parseval frame approximations.

5
 In [19] the set of all wandering vectors for a unitary system was parameterized
by the set of unitary operators in the so-called local commutant of the system at
a particular fixed wandering vector. However, unlike the wandering vector case,
it was shown in [37] that the set of all the Parseval frame vectors for a unitary
group can not be parameterized by the set of all the unitary operators in the
commutant of the unitary group. This means that the Parseval frame vectors
for a representation of a countable group are not necessarily unitarily equivalent.
However, this set can be parameterized by the set of all the unitary operators in
the von Neumann algebra generated by the representation ([32, 37]). This turns
out to be a very useful result in Gabor analysis (cf. [32, 29]). Although it remains
a question whether this result is still valid in the Hilbert C ∗ -module setting, in
Section 5.1 we will prove that this result holds in Hilbert C ∗ -modules when the
underlying C ∗ -algebra is a commutative W ∗ -algebra.
In the Hilbert space frame setting, the original work on symmetric orthogonal-
ization was done by Löwidin ([47]) in the late 1970’s. The concept of symmetric
approximation of frames by Parseval frame was introduced in [27] to extend the
symmetric orthogonalization of bases by orthogonal bases in Hilbert spaces. The
existence and the uniqueness results for the symmetric approximation of frames
by Parseval frames were obtained in [27]. Following their definition, a Parseval
frame {yj }∞ ∞
j=1 is said to be a symmetric approximation of frame {xj }j=1 in Hilbert

space H if it is similar to {xj }∞

j=1 and

n
X n
X
2
kzj − xj k ≥ kyj − xj k2 (1.1)
j=1 j=1

is valid for all Parseval frames {zj }∞ ∞

j=1 of H that are similar to {xj }j=1 .

Note that in some situations the symmetric approximation fails to work when
the underlying Hilbert space is infinite dimensional since if we restrict ourselves
to the frames induced by a unitary system then the summation in (1.1) is always
infinite when the given frame is not Parseval. Instead of using the symmetric
approximations to consider the frames generated by a collection of unitary trans-
formations and some window functions, it was proposed to approximate the frame

6
generator by Parseval frame generators. The existence and uniqueness results for
such a best approximation were obtained in [32, 33]. In Section 5.2 we will prove
that this result still holds for Hilbert C ∗ -module frames when the underlying C ∗ -
algebra is commutative. It remains open whether this is true when the underlying
C ∗ -algebra is non-commutative.
In Chapter 6 we will investigate the dilation of modular frames. It is well-
known that every frame in Hilbert space is a direct summand of Riesz basis, in
other words, each frame is a compression of a Riesz basis of a larger space. In
[26] it was shown that this is still true for the modular frames. In particular, it
was prove in ([26]) that each Parseval frame of Hilbert C ∗ -modules can be dilated
to an orthonormal basis. It is natural to ask whether a complete Parseval frame
vector for a unitary group on Hilbert C ∗ -module can be dilated to a complete
wandering vector. We will answer this question affirmatively in Section 6.1.
More generally, a dual frame pair in Hilbert space can be dilated to a Riesz basis
and its dual Riesz basis (see [13]). In Section 6.2 we will see that this remains
true for Hilbert C ∗ -module frames. We want to mention here that the proof of
this result for Hilbert space frames used in [13] can not be directly applied to
Hilbert C ∗ -module frames since the adjointablity of operators is always an issue
in dealing with operators in Hilbert C ∗ -modules. In Section 6.3 we will discuss
the projective frames for future study.
Let {fj }∞ ∞
j=1 be a basis of a Banach space X, and {gj }j=1 a sequence of vectors

in X. If there exists a constant λ ∈ [0, 1) such that

X X
k cj (fj − gj )k ≤ λk cj fj k

for all finite sequence {cj } of scalars, then {gj }∞

j=1 is also a basis for X. This

result is the well-known classical Paley-Wiener Theorem on perturbation of bases

([48]).
In the last decade, many attentions have been paid to generalize the Paley-
Wiener perturbation result to the perturbation of frames in Hilbert spaces (see
[5], [11], [15] and [16]). The most general result was obtained by Casazza and
Christensen ([11]):

7
Theorem 1.1. Let {xj }j∈J be a frame for a Hilbert space H with frame bounds
C and D. Suppose that {yj }j∈J is a sequence of H and there exist λ1 , λ2 , µ ≥ 0
such that max{λ1 + √µC , λ2 } < 1. If one of the following conditions is fulfilled for
any finite scalar sequence {cj } and all x ∈ H, then {yj }j∈J is also a frame for H:
1 1 1
(1) ( j∈J |hx, xj −yj i|2 ) 2 ≤ λ1 ( j∈J |hx, xj i|2 ) 2 +λ2 ( j∈J |hx, yj i|2 ) 2 +µkxk;
P P P

1
(2) k nj=1 cj (xj − yj )k ≤ λ1 k nj=1 cj xj k + λ2 k nj=1 cj yj k + µ( nj=1 |c|2 ) 2 .
P P P P

Moreover, if {xj }j∈J is a Riesz basis for H and {yj }j∈J satisfies (2), then
{yj }j∈J is also Riesz basis for H.

The purpose of Chapter 7 is to investigate whether the above perturbation

result remains valid for Hilbert C ∗ -modular frames. We prove that the Casazza-
Christensen general perturbation theorem for frames in Hilbert spaces remains
valid in Hilbert C ∗ -modules. In the Hilbert space setting, under the same per-
turbation condition, the perturbation of any Riesz basis remains a Riesz basis.
However, this no longer holds for Riesz bases in Hilbert C ∗ -modules. We give
a complete characterization on all the Riesz bases for Hilbert C ∗ -modules such
that the perturbation (under Casazza-Christensen’s perturbation condition) of a
Riesz basis still remains a Riesz basis.

8
 CHAPTER 2

PRELIMINARIES

2.1 Frames in Hilbert Spaces

2.1.1 Frames in Hilbert Spaces

The main feature of a basis {fk }∞

k=1 in a Hilbert space H is that every f ∈ H can

be represented as an (infinite) linear combination of the elements fk in the basis:

∞
X
f= ck (f )fk . (2.1)
k=1

The coefficients ck (f ) are unique. We now introduce the concept of frames. A

frame is also a sequence of elements {fk }∞
k=1 in H, which allows every f ∈ H to

be written as in (2.1). However, the corresponding coefficients are not necessarily

unique. Thus a frame might not be a basis.
We now give the definition of frames.

Definition 2.1. A sequence {fk }∞

k=1 of elements in Hilbert space H is a frame

for H if there exist constants A, B > 0 such that

∞
X
2
Akf k ≤ |hf, fk i|2 ≤ Bkf k2 , ∀f ∈ H. (2.2)
k=1

The numbers A, B are called frame bounds. They are not unique.
If A = B, then {fk }∞
k=1 is called a tight frame, and a Parseval frame if

A = B = 1.

9
 Particularly, if the right inequality
∞
X
|hf, fk i|2 ≤ Bkf k2 , ∀f ∈ H,
k=1

holds true, we call {fk }∞

k=1 a Bessel sequence.

We now give a few more examples of frames. They might appear quite con-
structed, but are useful for the theoretical understanding of frames.

Example 2.2. Let {ek }∞

k=1 be an orthonormal basis for a Hilbert space H.

(1) By repeating each element of {ek }∞

k=1 twice we have

{fk }∞
k=1 = {e1 , e1 , e2 , e2 , . . . }

which is a tight frame with frame bound A = B = 2.

If only e1 is repeated we get

{fk }∞
k=1 = {e1 , e1 , e2 , e3 , . . . }

which is a frame with bounds A = 1 and B = 2.

(2) Let
1 1 1 1 1
{fk }∞
k=1 = {e1 , √ e2 , √ e2 , √ e3 , √ e3 , √ e3 , . . . }.
2 2 3 3 3
Note that for each f ∈ H, we have
∞ ∞
X
2
X 1
|hf, fk i| = k|hf, √ ek i|2 = kf k2 .
k=1 k=1
k

So {fk }∞
k=1 is a Parseval frame.

Example 2.3. Let H and K be Hilbert spaces with H ⊂ K, and let {ej }∞
j=1

be an orthonormal basis for K. Let P denote the orthogonal projection from K

onto H, and let xj = P ej for all j. If x ∈ H is arbitrary, then
X
kxk2 = |hx, ej i|2 (2.3)
j

10
and
X
x= hx, ej iej . (2.4)
j

Since x = P x and xj = P ej we have hx, ej i = hx, xj i, so (2.3) becomes

X
kxk2 = |hx, xj i|2
j

and hence {xj }∞

j=1 is a Parseval frame for H. Moreover, applying P to (2.4) then

yields
X
x= hx, xj ixj (2.5)
j

for all x ∈ H. The formula (2.5) is called the reconstruction formula for {xj }.

Let {fk }∞
k=1 be a frame of Hilbert space H, then we have the corresponding

pre-frame operator, analysis operator and frame operator as follows.

The operator T : H → l2 defined by

T f = {hf, fk i}∞
k=1 ,

is called the analysis operator. The adjoint operator T ∗ : l2 → H is given by

∞
X
∗
T {ck }∞
k=1 = ck fk .
k=1

T ∗ is called pre-frame operator or the synthesis operator. By composing T and

T ∗ , we obtain the frame operator S : H → H:
∞
X
∗
Sf = T T f = hf, fk ifk .
k=1

Let state some of the important properties of S:

Proposition 2.4. Let {fk }∞

k=1 be a frame with frame operator S and frame bounds

A, B. Then the following holds:

(1) S is bounded, invertible, self-adjoint, and positive.
(2) {S −1 fk }∞ −1 −1
k=1 is a frame with bounds B , A . The frame operator for

{S −1 fk }∞
k=1 is S
−1
.

11
 The most important frame result is the following reconstruction formula. It
shows that if {fk }∞
k=1 is a frame for Hilbert space H, then every element in H has

a representation as an infinite linear combination of the frame elements. Thus it

is natural to view a frame as some kind of ”generalized basis”.

Theorem 2.5. Let {fk }∞

k=1 be a frame with frame operator S for Hilbert space

H. Then
∞
X
f= hf, S −1 fk ifk , ∀f ∈ H. (2.6)
k=1

It should be mentioned here that to every frame we can associate a canonical

Parseval frame:

Proposition 2.6. Let {fk }∞

k=1 be a frame for Hilbert space H with frame operator
1 1
S. Denote the positive square root of S −1 by S − 2 . Then {S − 2 fk }∞
k=1 is a Parseval

frame, and
∞
1 1
X
f= hf, S − 2 fk iS − 2 fk , ∀f ∈ H.
k=1

We now introduce the definition of dual frames.

Definition 2.7. Let {fk }∞

k=1 be a frame of a Hilbert space H. We call a sequence

{gk }∞ ∞
k=1 ⊆ H a dual frame of {fk }k=1 if

∞
X
f= hf, gk ifk
k=1

holds true for every f ∈ H.

In particular, {S −1 fk }∞
k=1 is called the canonical dual (or. standard dual ) of

{fk }∞ ∞
k=1 , where S is the frame operator of {fk }k=1 .

Note that the roles of a frame and its duals can be interchanged in the fol-
lowing sense.

Proposition 2.8. Assume that {fk }∞ ∞

k=1 and {gk }k=1 are Bessel sequences in

Hilbert space H. Then the following are equivalent:

12
 P∞
(1) f = k=1 hf, gk ifk , ∀f ∈ H.
P∞
(2) f = k=1 hf, fk igk , ∀f ∈ H.
P∞
(3) hf, gi = k=1 hf, fk ihgk , f i, ∀f, g ∈ H.
In case the equivalent conditions are satisfied, {fk }∞ ∞
k=1 and {gk }k=1 are dual

frames for H.

We now list some characterizations of frames in Hilbert spaces from the op-
erator theory point of view.

Theorem 2.9. A sequence {fk }∞

k=1 in Hilbert space H is a frame for H if and

only if
∞
X
T : {ck }∞
k=1 → ck fk
k=1
2
is a well-defined mapping of l onto H.

Note that the question of existence of an upper and lower frame bound, via
Theorem 2.9, is replaced by an investigation of infinite series: we need to check
P∞ ∞ 2
that k=1 ck fk converges for all {ck }k=1 ∈ l and that each f ∈ H can be
represented via such an infinite series. The following characterization of frames
involves the information about the frame bounds.

Theorem 2.10. A sequence {fk }∞

k=1 in Hilbert space H is a frame for H with

bounds A, B if and only if the following conditions are satisfied:

(1) span{fk }∞
k=1 = H;

(2) The pre-frame operator T is well defined on l2 and

∞
X ∞
X
A 2
|ck | ≤ kT {ck }∞
k=1 k
2
≤B |ck |2 , ∀{ck }∞ ⊥
k=1 ∈ (KerT ) .
k=1 k=1

2.1.2 Riesz Bases in Hilbert Spaces

There are many equivalent definitions for Riesz bases in Hilbert spaces. Here we
adopt the following:

13
Definition 2.11. A Riesz basis for a Hilbert space H is a family of the form
{U ek }∞ ∞
k=1 , where {ek }k=1 is an orthonormal basis for H and U : H → H is a

bounded bijective operator.

A Riesz basis is actually a basis. In fact, one can characterize Riesz bases in
terms of bases satisfying extra conditions:

Proposition 2.12. A sequence {fk }∞

k=1 is a Riesz basis for Hilbert space H if it

is an unconditional basis for H and

0 < inf kfk k ≤ sup kfk k < ∞.

k k

The dual basis associated to a Riesz basis is also a Riesz basis and is unique:

Proposition 2.13. If {fk }∞

k=1 is a Riesz basis for Hilbert space H, there exists

a unique sequence {gk }∞

k=1 in H satisfying

∞
X
f= hf, gk ifk , ∀f ∈ H.
k=1

{gk }∞ ∞ ∞
k=1 is also a Riesz basis, and {fk }k=1 and {gk }k=1 are biorthogonal.

Note that a Riesz basis is a frame:

Proposition 2.14. If {fk }∞

k=1 is a Riesz basis for Hilbert space H, then there

exist constants A, B such that

∞
X
2
Akf k ≤ |hf, fk i|2 ≤ Bkf k2 , ∀f ∈ H.
k=1

For a frame to be a Riesz basis, we have the following proposition:

Proposition 2.15. Let {fj }∞

j=1 be a frame for Hilbert space H. Then the follow-

ing are equivalent:

(1) {fj }∞
j=1 is a Riesz basis for H.

(2) {fj }∞
j=1 is an exact frame, i.e. it ceases to be a frame when an arbitrary

element is removed.

14
 (3) {fj }∞
j=1 is minimal, i.e. fj ∈
/ span{jk : k 6= j} for any j.
(4) {fj }∞
j=1 and {S
−1
fj }∞
j=1 are biorthogonal, where S is the frame operator

of {fj }∞
j=1 .

(5) If ∞ ∞ 2
P
j=1 cj fj = 0 for some {cj }j=1 ∈ l , then cj = 0 for all j.

(6) {fj }∞
j=1 is a basis.

We now list a characterization of Riesz bases for Hilbert spaces.

Proposition 2.16. For a sequence {fk }∞

k=1 in Hilbert space H, the following

conditions are equivalent:

(1) {fk }∞
k=1 is a Riesz basis for H.

(2) span{fk }∞
k=1 = H, and there exist constants A, B > 0 such that for any
finite sequence {ck } one has
X X X
A |ck |2 ≤ k ck fk k2 ≤ B |ck |2 .

Let’s summarize the relations between orthonormal bases, Riesz bases and
frames in Hilbert spaces as follows.

Theorem 2.17. Let {ek }∞

k=1 be an arbitrary orthonormal basis for Hilbert space

H. Then
(1) The orthonormal bases of H are the families {U ek }∞
k=1 , where linear op-

erator U : H → H is unitary.
(2) The Riesz bases of H are the families {U ek }∞
k=1 , where linear operator

U : H → H is bounded and bijective.

(3) The frames of H are the families {U ek }∞
k=1 , where linear operator U :

H → H is bounded and surjective.

15
2.1.3 Dilation Results of Frames

It turns out that Example 2.3 is a generic and serves as a model for arbitrary
Parseval frames. One can always dilate such a frame to an orthonormal basis.
One immediate consequence of the dilation is that the reconstruction formula
(2.5) always holds for a Parseval frame. We have the following dilation result.

Proposition 2.18. ([37]) Let J be a countable (or finite) index set. Suppose that
{xj }j∈J is a Parseval frame for Hilbert space H. Then there exist a Hilbert space
K ⊇ H and an orthonormal basis {ej }j∈J for K such that xj = P ej , where P is
the orthogonal projection from K onto H.

Also we can summarize the dilation results in the following way.

Proposition 2.19. Frames are precisely the inner direct summands of Riesz
bases. Parseval frames are precisely the inner direct summands of orthonormal
bases.

Example 2.20. Let {e1 , e2 , e3 } be an orthonormal basis for a 3-dimensional

Hilbert space K. Another orthonormal basis for K is then
1 1 1
{ √ (e1 + e2 + e3 ), √ (e1 − 2e2 + e3 ), √ (e1 − e3 )}.
3 6 2
Note that
1 1 1
{ √ (e1 + e2 ), √ (e1 − 2e2 ), √ e1 }
3 6 2
is a Parseval frame for H = span{e1 , e2 }.

Suppose that {uj } is a Riesz basis for a Hilbert space K ⊃ H with its unique
dual {u∗j }. If P is the orthogonal projection from K onto H then {P uj } is a
frame for H with an alternate dual {P u∗j }. In general {P u∗j } is not the canonical
dual for {P uj } unless P commutes with the frame operator of {uj } (see [37]).
So it is natural to ask whether a given frame {xj } and one of its alternate duals
{yj } can be dilated to a Riesz basis {uj } for some larger Hilbert space K so that
xj = P uj and yj = P u∗j . It was affirmatively answered in [13] as follows.

16
Proposition 2.21. Suppose that {xj } and {yj } are alternate dual frames in a
Hilbert space H. Then there is a Hilbert space K ⊃ H and a Riesz basis {uj } for
K with P uj = xj and P u∗j = yj , where {u∗j } is the (unique) dual of {uj } and P
is the orthogonal projection from K onto H.

2.1.4 Structured Frames in Hilbert Spaces

In applications the most important and practical frames are the ones that are
generated by a single vector in a Hilbert space under the action of a suitable
collection of unitary operators. Wavelet frames and Gabor frames are typical
examples. A unitary system U is a countable set of unitary operators acting on
a separable Hilbert space H that contains the identity operator. We say that a
vector φ ∈ H is a complete frame vector (resp. complete Parseval frame vector )
for U if Uφ := {U φ : U ∈ U } is a frame (resp. Parseval frame) for H. When Uφ
is an orthonormal basis for H, φ is called a complete wandering vector for U, the
set of all complete wandering vectors for U is denoted by W(U).
If U is a unitary system and φ ∈ W(U), the local commutant Cφ (U) at φ is
defined by {T ∈ B(H) : (T U − U T )φ = 0, U ∈ U}. Clearly Cφ (U) contains the
commutant U 0 of U. When U is a unitary group, it is actually the commutant of
U.
For the characterization of frame vectors for unitary systems we have the
following result.

Proposition 2.22. ([19], [32]) Suppose that φ is a complete wandering vector

for a unitary system U. Then
(1) a vector ξ ∈ H is a complete wandering vector for U if and only if there
is a (unique) unitary operator A ∈ U 0 such that ξ = Aφ.
(2) a vector η ∈ H is a complete Parseval frame vector for U if and only if
there is a (unique) co-isometry A ∈ Cφ (U) such that η = Aφ.

17
 Recall that a unitary system U is said to be group-like if

group(U) ⊂ TU := {tU : t ∈ T, U ∈ U}

and if different U and V in U are always linearly independent, where group(U)

denotes the group generated by U with respect to multiplication and T denotes
the unit circle.
We have the following parameterization of frame vectors at a fixed complete
Parseval frame vector for group-like unitary systems.

Proposition 2.23. ([32]) Let η be a complete Parseval frame vector for a group-
like unitary system U and w∗ (U) be the von Neumann algebra generated by U.
Suppose that ξ ∈ H. Then
(1) ξ is a complete Parseval frame vector for U if and only if there exists a
unitary operator A ∈ w∗ (U) such that Aη = ξ.
(2) ξ is a complete frame vector for U if and only if there exists an invertible
operator A ∈ w∗ (U) such that Aη = ξ.

Another kind of important structured frames is multi-generated frames which

are generated by some (usually finite number of) vectors under the action of a
collection of unitary operators. For example, Gabor frames and wavelet frames
are of this kind. Recall that Φ = (φ1 , . . . , φN ) with φj ∈ H is called a multi-
frame generator (resp. Parseval multi-frame generator ) for a unitary system U
if {U φj : U ∈ U, 1 ≤ j ≤ N } is a frame (resp. Parseval frame) for H.
We now define the best Parseval multi-frame approximation for a multi-frame
generator.

Definition 2.24. Let Φ = (φ1 , . . . , φN ) be a multi-frame generator for a unitary

system U. Then a Parseval multi-frame generator Ψ = (ψ1 , . . . , ψN ) for U is
called a best Parseval multi-frame approximation for Φ if the inequality
N
X N
X
2
kφj − ψj k ≤ kφj − ξj k2
j=1 j=1

is valid for all the Parseval multi-frame generator Ξ = (ξ1 , . . . , ξN ) for U.

18
 The following result was proved in [33].

Proposition 2.25. Let U be a group-like unitary system acting on a Hilbert space

H and let Φ = (φ1 , . . . , φN ) be a multi-frame generator for U. Then S −1/2 Φ is
the unique best Parseval multi-frame approximation for Φ, where S is the frame
operator for the multi-frame {U φj : U ∈ U, j = 1, . . . , N }.

2.2 Hilbert C ∗ -modules

The aim of this section is to cover the basic results of Hilbert C ∗ -modules. To
introduce the concept of Hilbert C ∗ -modules, we first introduce the definition of
C ∗ -algebras.

2.2.1 C ∗ -algebras and W ∗ -algebras

Let’s begin with

Definition 2.26. Let A be an associative algebra over the complex numbers.

The algebra A is called a normed algebra if there is associated to each element x
a real number kxk, called the norm of x, with the properties:
(1) kxk ≥ 0, and kxk = 0 if and only if x = 0.
(2) kx + yk ≤ kxk + kyk.
(3) kλxk = |λ|kxk, λ is a complex number.
(4) kxyk ≤ kxkkyk
If A is complete with respect to the norm (i.e. if A is also a Banach space),
then it is called a Banach algebra.

Note that any closed subalgebra of a Banach algebra is a Banach algebra.

19
Example 2.27. (1) Let X be a Banach space, denote by B(X) the set of all
bounded linear operators on X. Then B(X) is a Banach algebra with the
pointwise-defined operations for addition and scalar multiplication, multiplica-
tion given by (AB)(x) = A(B(x)), and the operator norm.
(2) The algebra Mn (C) of n × n-matrices with entries in C is identified with
B(Cn ). It is therefore a Banach algebra.

Definition 2.28. Let A be a unital Banach algebra and a an element of A. Then

a is invertible if there is an element b ∈ A such that

ab = ba = 1.

The set
Inv(A) = {a ∈ A : a is invertible}

is a group under multiplication.

We define the spectrum of an element a to be the set

σ(a) = {λ ∈ C : λ1 − a ∈
/ Inv(A)}.

Example 2.29. (1) Let A = C(X), where X is a compact Hausdorff space.

Then σ(f ) = f (X) for all f ∈ A.
(2) Let A = Mn (C), the algebra of all complex n × n-matrices. Then for each
A ∈ A, σ(A) is the set of eigenvalues of A.

Definition 2.30. Let A be a Banach algebra. A mapping x 7→ x∗ of A into

itself is called an involution if for all x, y ∈ A and any scalar λ ∈ C the following
conditions are satisfied:
(1) (x∗ )∗ = x.
(2) (x + y)∗ = x∗ + y ∗ .
(3) (xy)∗ = y ∗ x∗ .
(4) (λx)∗ = λx∗ .
Then A is called a Banach ∗-algebra.

20
Definition 2.31. A Banach ∗-algebra A is called a C ∗ -algebra if it satisfies

kx∗ xk = kxk2 (2.7)

for all x ∈ A.

Here are a few examples.

Example 2.32. (1) B(H), the algebra of bounded linear operator on a Hilbert
space H, is a C ∗ -algebra, where for each operator A, A∗ is the adjoint of A.
(2) Any closed ∗-subalgebra of B(H) is a C ∗ -algebra.
(3) C(X), the algebra of continuous functions on a compact space X, is an
abelian C ∗ -algebra, where f ∗ (x) ≡ f (x).
(4) C0 (X), the algebra of continuous functions on a locally compact space X
that vanish at infinity, is an abelian C ∗ -algebra, where f ∗ (x) ≡ f (x).

Definition 2.33. Let A be a unital C ∗ -algebra and a, p, u ∈ A. Then

(1) a is normal if a∗ a = aa∗ .
(2) p is a projection if p = p∗ = p2 .
(3) u is a unitary if u∗ u = uu∗ = 1.
(4) u is an isometry if u∗ u = 1.
(5) u is a co-isometry if uu∗ = 1.

For the spectrum of normal elements in C ∗ -algebras we have the following

famous Spectral Mapping Theorem.

Theorem 2.34. Let a be a normal element of a unital C ∗ -algebra A, and f ∈

C(σ(a)). Then
σ(f (a)) = f (σ(a)).

Definition 2.35. An element a of a C ∗ -algebra A is positive if a∗ = a and

σ(a) ⊆ R+ . We write a ≥ 0 to mean that a is positive.

We have the following characterizations on positive elements.

21
Proposition 2.36. Let A be a C ∗ -algebra. The following statements are equiv-
alent:
(1) a is positive.
(2) a = b2 for some positive element b ∈ A.
(3) a = x∗ x for some x ∈ A.
(4) a∗ = a and kt − ak ≤ t for all t ≥ kak.
(5) a∗ = a and kt − ak ≤ t for some t ≥ kak.

We summarize some elementary factors about positive elements in the follow-

ing proposition.

Proposition 2.37. Let A be a C ∗ -algebra.

(1) If 0 ≤ a and 0 ≤ b, then 0 ≤ a + b.
(2) If 0 ≤ a ≤ b, then kak ≤ kbk.
(3) If A is unital and a, b are positive invertible elements in A, then a ≤ b
implies 0 ≤ b−1 ≤ a−1 .
(4) If a, b are positive elements in A, then a ≤ b implies that at ≤ bt for any
t ∈ (0, 1).
(5) If a2 ≤ b2 for all a, b ∈ A with 0 ≤ a ≤ b, the A must be commutative.
(6) If a, b ∈ A, then abb∗ a∗ ≤ kbk2 aa∗ and abb∗ a∗ ≤ kak2 bb∗ .

We now give the definition of W ∗ -algebras as follows.

Definition 2.38. A C ∗ -algebra M is called a W ∗ -algebra if it is a dual space as

a Banach space, i.e., if there exists a Banach space M∗ such that (M∗ )∗ = M,
where (M∗ )∗ is the dual Banach space of M∗ .

Definition 2.39. Let p, q be two projections of a W ∗ -algebra M. If there exists

a partial isometry u in M such that u∗ u = p and uu∗ = q, then p is said to be
equivalent to q and denote this by p ∼ q. If there exists a projection q1 (≤ q)
equivalent to p, write this by p ≺ q or q  p.

22
 For the equivalence of projections we have the famous Comparability Theo-
rem.

Theorem 2.40. Let p and q be projections of a W ∗ -algebra M. Then there exists

a central projection z ∈ M such that

pz  qz and pz 0 ≺ qz 0

where z 0 = 1 − z.

Definition 2.41. Let p be a projection of a W ∗ -algebra M. p is said to be finite

if for a projection p1 in M, p1 ≤ p and p1 ∼ p imply p1 = p.

Definition 2.42. A W ∗ -algebra is said to be finite if its identity is finite.

Note that a W ∗ -algebra M is finite if and only if every isometry in M is

unitary.
For the equivalence of projections in finite W ∗ -algebras we have the following
result.

Proposition 2.43. let M be a finite W ∗ -algebra, and let p, p1 , q and q1 be pro-

jections in M satisfying the following conditions:

p1 ≤ p, q1 ≤ q, p1 ∼ q1 and p ∼ q.

Then p − p1 ∼ q − q1 .

Particularly we have

Corollary 2.44. Let M be a finite W ∗ -algebra, and p, q be two equivalent pro-

jections in M. Then 1 − p and 1 − q are equivalent.

2.2.2 Hilbert C ∗ -modules

Hilbert C ∗ -modules form a category in between Banach spaces and Hilbert spaces.
The basic idea was to consider module over C ∗ -algebra instead of linear space

23
and to allow the inner product to take values in a more general C ∗ -algebra than
C. The structure was first used by Kaplansky [42] in 1952 and more carefully
investigated by Rieffel [51] and Paschke [49] later in 1972/73.
We give only a brief introduction to the theory of Hilbert C ∗ -modules to
make our explanations self-contained. For comprehensive accounts we refer to
the lecture note of Lance [46] and the book of Wegge-Olsen [55].
We now give the definition of Hilbert C ∗ -modules.

Definition 2.45. Let A be a C ∗ -algebra and H be a (left) A-module. Suppose

that the linear structures given on A and H are compatible, i.e. λ(ax) = a(λx)
for every λ ∈ C, a ∈ A and x ∈ H. If there exists a mapping h·, ·i : H × H → A
with the properties
(1) hx, xi ≥ 0 for every x ∈ H,
(2) hx, xi = 0 if and only if x = 0,
(3) hx, yi = hy, xi∗ for every x, y ∈ H,
(4) hax, yi = ahx, yi for every a ∈ A, and every x, y ∈ H,
(5) hx + y, zi = hx, zi + hy, zi for every x, y, z ∈ H.
Then the pair {H, h·, ·i} is called a (left-) pre-Hilbert A-module. The map h·, ·i
is said to be an A-valued inner product. If the pre-Hilbert A-module {H, h·, ·i}
1
is complete with respect to the norm kxk = khx, xik 2 then it is called a Hilbert
A-module.

Here are some important examples.

Example 2.46. The C ∗ -algebra A itself can be reorganized to become a Hilbert

A-module if we define the inner product

ha, bi = ab∗ , ∀a, b ∈ A.

The corresponding norm is just the norm on A because of the C ∗ -equation (2.7).

Example 2.47. If {Hi }nk=1 be a finite set of Hilbert A-modules over a C ∗ -algebra
A, then one can define the direct sum ⊕nk=1 Hk . The inner product on ⊕nk=1 Hk is

24
given by the formula
n
X
hx, yi := hxk , yk iHk ,
k=1
where x = x1 ⊕ x2 ⊕ · · · ⊕ xn , y = y1 ⊕ y2 ⊕ · · · ⊕ yn ∈ ⊕nk=1 Hk . Then ⊕nk=1 Hk is
a Hilbert A-module.
We denote the direct sum of n copies of a Hilbert C ∗ -module H by Hn .

Example 2.48. If {Hk }, k ∈ N is a countable set of Hilbert A-modules over

C ∗ -algebra A, then one can define their direct sum ⊕k∈N Hk . On the A-module
P
⊕k∈N Hk of all sequences x = (x1 , x2 , . . . ), xk ∈ Hk , such that the series k∈N hxk , xk i
is norm-convergent in the C ∗ -algebra A, we define the inner product by
X
hx, yi := hxk , yk iHk
k∈N

for x, y ∈ ⊕k∈N Hk .
Then ⊕k∈N Hk is a Hilbert A-module.
The direct sum of a countable number of copies of a Hilbert C ∗ -module H is
denoted by l2 (H).

Note that in Hilbert C ∗ -modules the Cauchy-Schwartz Inequality is valid.

Proposition 2.49. Let H be a Hilbert C ∗ -module, and x, y ∈ H, then

khx, yik2 ≤ khx, xik · khy, yik.

We are especially interested in finitely and countably generated Hilbert C ∗ -

modules over unital C ∗ -algebra A. A Hilbert A-module H is (algebraically) fi-
nitely generated if there exists a finite set {x1 , · · · , xn } ⊆ H such that every ele-
ment x ∈ H can be expressed as an A-linear combination x = ni=1 ai xi , ai ∈ A.
P

A Hilbert A-module H is countably generated if there exists a countable set of

generators.
Note that algebraically finitely generated Hilbert A-module over unital C ∗ -
algebra A are precisely the finitely generated projective A-modules in a pure
algebraic sense.

25
Theorem 2.50. Let A be a unital C ∗ -algebra. Every algebraically finitely gener-
ated Hilbert A-module H is an orthogonal summand of some free Hilbert A-module
An for a finite number n.

Let A be a C ∗ -algebra. The Hilbert A-module l2 (A) serves as an universal

environment for countably generated Hilbert A-module that can be described
as (special) orthogonal summands. This result was given by Kasparov ([43]) in
1980.

Theorem 2.51. Let A be a unital C ∗ -algebra. Every countably generated Hilbert

A-module H posses an embedding into l2 (A) as an orthogonal summand in such
a way that the orthogonal complement of it is isometrically isomorphic to l2 (A)
again, i.e. H ⊕ l2 (A) = l2 (A).

Remark 2.52. Note that not every Hilbert C ∗ -module has an orthonormal basis.
Though any countably generated Hilbert C ∗ -module admits a standard frame,
there are countably generated Hilbert C ∗ -modules that contain no orthonormal
basis even orthogonal Riesz basis (see Example 3.4 in [26]).

We now list some properties of operators on Hilbert C ∗ -modules.

Definition 2.53. Let H be a Hilbert A-module over a C ∗ -algebra A. A map

T : H → H (a priori neither linear nor bounded) is said to be adjointable if there
exists a map T ∗ : H → H satisfying

hx, T yi = hT ∗ x, yi

for all x, y ∈ H. Such a map T ∗ is called the adjoint of T .

By End∗A (H) we denote the set of all adjointable maps on H.

It is surprising that every adjointable operator is automatically linear and

bounded.

Proposition 2.54. Suppose that T, S are two adjointable operators on a Hilbert

A-module H over a C ∗ -algebra A, then

26
 (1) The adjoint of T is unique and adjointable with T ∗∗ = T .
(2) ST is adjointable with (ST )∗ = T ∗ S ∗ .
(3) T is a C-linear module map which is bounded with respect to the operator
norm.
(4) T ∗ S = 0 if and only if T (H) ⊥ S(H).
(5) KerT = Ker|T |, KerT ∗ = T (H)⊥ and (KerT ∗ )⊥ = T (H)⊥⊥ ⊃ T (H).

Proposition 2.55. Let H be a Hilbert A-module over a C ∗ -algebra A. Then

End∗A (H) is a C ∗ -algebra equipped with the operator norm

kT k = sup{kT xk : kxk ≤ 1}.

It should be mentioned here that, unlike B(H), End∗A (H) is not a von Neu-
mann algebra in general.
For adjointable operators we have the following Polar Decomposition Theo-
rem.

Theorem 2.56. Let H be a Hilbert C ∗ -module and T ∈ End∗ (H), then the
following are equivalent:
(1) T has a polar decomposition T = V |T |, where V ∈ End∗ (H) is a partial
isometry for which

KerV = KerT, KerV ∗ = KerT ∗ ,

V (H) = T (H), V ∗ (H) = T (H).

(2) H = Ker|T | ⊕ |T |(H) and H = KerT ∗ ⊕ T (H).

(3) Both T (H) and |T |(H) are complementable in H.

The following result will be frequently used in this manuscript.

Theorem 2.57. Suppose that H is a Hilbert C ∗ -module and T an adjoint operator

on H with closed range, then T ∗ and |T | have closed ranges and

H = Ker|T | ⊕ |T |(H) = KerT ∗ ⊕ T (H) = KerT ⊕ T ∗ (H).

27
Definition 2.58. Let H be a Hilbert C ∗ -module and M ⊆ H a submodule. Then
M is said to be complementable if H = M ⊕ N for some submodule N ⊆ H.

Note that submodules in Hilbert C ∗ -modules need not be complementable in

general. We have the following example.

Example 2.59. Let A = C[0, 1] be the set of all continuous functions on [0, 1]
with the norm closed ideal M = C0 [0, 1], where C0 [0, 1] = {f ∈ C[0, 1] : f (0) =
f (1) = 0}. In this case M is a Hilbert A-submodule with M⊥ = {0}, so
H 6= M ⊕ M⊥ and M =
6 M⊥⊥ = H.

Since it is more convenient to work with orthogonal decompositions, we would

like to describe situations where such a decomposition exists.

Proposition 2.60. Suppose that H is a Hilbert C ∗ -module and T is an ad-

jointable operator on H with closed range, then both the range and kernel of T
are complementable in H.

Corollary 2.61. Let H be a Hilbert C ∗ -module. If P ∈ End∗ (H) is an idempo-

tent, then its range is an orthogonally complementable submodule in H.

We also have

Proposition 2.62. A closed submodule of a Hilbert C ∗ -module is complementable

precisely when it is the range of an adjointable operator.

For a Hilbert A-module H over a C ∗ -algebra A, let us denote by H0 the set

of all bounded A-linear maps from H to A.

Definition 2.63. A Hilbert C ∗ -module H is called self-dual if H = H0 .

The condition of self-duality is very strong. Below we shall see that there are
quite a few self-dual modules: any Hilbert module over a C ∗ -algebra A is self-
dual if and only if A is finite dimensional. If A is a unital C ∗ -algebra, then the
Hilbert module An is obviously self-dual. Self-dual Hilbert C ∗ -modules behave
quite like Hilbert spaces. In the same way as in the case of Hilbert spaces, the
following statements can be easily checked.

28
Proposition 2.64. Let H be a self-dual Hilbert A-module over a C ∗ -algebra A,
K an arbitrary Hilbert A-module and T : H → K a bounded A-linear operator.
Then there exists an operator T ∗ : K → H such that the equality

hT x, yi = hx, T ∗ yi

holds for all x ∈ H and y ∈ K, i.e. T is adjointable.

Proposition 2.65. Let H be a self-dual Hilbert C ∗ -module and let H ⊂ K. Then

K = H ⊕ H⊥ .

We complete this chapter by the following remark.

Remark 2.66. It should mention here that by no means all results of Hilbert
space theory can be simply generalized to the situation of Hilbert C ∗ -modules.
For example,
(1) The analogue of the Riesz representation theorem for bounded A-linear
mapping is not valid for H.
(2) Since in general a Hilbert C ∗ -module H need not be self-dual, the bounded
A-linear operator on H may not have an adjoint operator.
(3) Since a Hilbert C ∗ -submodule M of the Hilbert C ∗ -module H may not be
complementable in general, the corresponding projection may not be orthogonal.

29
 CHAPTER 3

HILBERT C ∗ -MODULE FRAMES

We first introduce the definition of modular frames and list some basic and im-
portant properties of modular frames. According to this definition we need to
compare positive elements in the underlying C ∗ -algebra in order to test whether
a sequence is a frame or not. This is not a trivial task. In Section 3.2 we give
an equivalent definition of frames in Hilbert C ∗ -modules which is much easier to
be applied. Based on this equivalent definition, we characterize modular frames
from the operator theory point of view in Section 3.3.

3.1 Frames in Hilbert C ∗ -modules

Definition 3.1. Let A be a unital C ∗ -algebra and J be a finite or countable

index set. A sequence {xj }j∈J of elements in a Hilbert A-module H is said to be
a frame if there exist two constants C, D > 0 such that
X
C · hx, xi ≤ hx, xj ihxj , xi ≤ D · hx, xi (3.1)
j∈J

for every x ∈ H. The optimal constants (i.e. maximal for C and minimal for D)
are called frame bounds.
The frame {xj }j∈J is said to be tight frame if C = D, and said to be Parseval
if C = D = 1.

30
 Likewise, {xj }j∈J is called a Bessel sequence with bound D if there exists
D > 0 such that
X
hx, xj ihxj , xi ≤ Dhx, xi (3.2)
j∈J

for every x ∈ H.

A sequence {xj }j∈J is said to be a Riesz basis of H if it is a frame and a

P
generating set with the additional property that A-linear combinations j∈S aj xj
with coefficients {aj : j ∈ S} ⊆ A and S ∈ J are equal to zero if and only if in
particular every summand aj xj equals zero for j ∈ S.
We consider standard (normalized tight) frames, standard Bessel sequences
and standard Riesz bases in the main for which the sums in the inequalities (3.1)
and (3.2) always converges in norm.
It should be remarkable that following Theorem 2.50 and Theorem 2.51, it
was proved in [26] that every finitely generated or countably generated Hilbert
C ∗ -module admits a (standard) frame.
Note that we can also define the analysis operator, synthesis operator and
frame operator for modular frames as follows.
Suppose that {xj }j∈J is a frame of a finitely or countably generated Hilbert
A-module H over a unital C ∗ -algebra A.
The operator T : H → l2 (A) defined by

T x = {hx, xj i}j∈J ,

is called the analysis operator. The adjoint operator T ∗ : l2 (A) → H is given by

X
T ∗ {cj }j∈J = cj xj .
j∈J

T ∗ is called pre-frame operator or the synthesis operator. By composing T and

T ∗ , we obtain the frame operator S : H → H:
X
Sx = T ∗ T x = hx, xj ixj . (3.3)
j∈J

31
 The frame {S −1 xj }j∈J is said to be the canonical dual frame of {xj }j∈J .
The main property of frames for Hilbert spaces is the existence of the recon-
struction formula that allows a simple standard decomposition of every element
of the spaces with respect to the frame. For the frames in Hilbert C ∗ -modules,
we have the following results.

Theorem 3.2. ([23]) Let A be a unital C ∗ -algebra, H be a finitely or count-

ably generated Hilbert A-module and {xj }j∈J be a Parseval frame (not necessarily
standard) of H. Then the reconstruction formula
X
x= hx, xj ixj (3.4)
j∈J

holds for every x ∈ H in the sense of convergence with respect to the topology that
is induced by the set of semi-norms {|f (h·, ·i)|1/2 : f ∈ A∗ }. The sum converges
always in norm if and only if the frame {xj }j∈J is standard.

Also from equation (3.3) we see that

X
x= hx, S −1 xj ixj
j∈J

is valid for every x ∈ H.

More generally, we have an existence and uniqueness result that provides us
with a reconstruction formula for standard frame. Also this result guarantees the
existence of a dual for any Hilbert C ∗ -module frame.

Theorem 3.3. ([26]) Let {xj }j∈J be a standard frame in a finitely or countably
generated Hilbert A-module H over a unital A-algebra A. Then there exists a
unique operator S ∈ End∗A (H) such that
X
x= hx, S(xj )ixj (3.5)
j∈J

for every x ∈ H. The operator can be explicitly given by the formula S = G∗ G

for any adjointable invertible bounded operator G mapping H onto some other
Hilbert A-module K and realizing {G(xj ) : j ∈ J} to be a standard normalized
tight frame in K.

32
 Similar to the case of Hilbert space frames, we also have the following dilation
result for modular frames.

Proposition 3.4. ([26]) Modular frames are precisely the inner direct summands
of standard Riesz bases of An or l2 (A), where A is a C ∗ -algebra.

3.2 An Equivalent Definition of Modular Frames

Our first observation shows that the analysis operator of Bessel sequence is ad-
jointable.

Lemma 3.5. Let {xj }j∈J be a Bessel sequence of a finitely or countably generated
Hilbert A-module H over a unital C ∗ -algebra A. Then the analysis operator
T : H → l2 (A) defined by

X
Tx = hx, xj iej
j∈J

is adjointable and fulfills T ∗ ej = xj for all j.

Proof. It follows directly from the proofs of Theorem 4.1 and Theorem 4.4 in
[26].

We need the following lemma to prove our results.

Lemma 3.6. ([49]) Let M and N be Hilbert A-module over a C ∗ -algebra A and
let T : M → N be a linear map. Then the following conditions are equivalent:
(1) the operator T is bounded and A-linear;
(2) there exists a constant K ≥ 0 such that the inequality hT x, T xi ≤ Khx, xi
holds in A for all x ∈ M.

We have the following equivalent definition for Bessel sequences in Hilbert

C ∗ -modules.

33
Lemma 3.7. Let {xj }j∈J be a sequence of a finitely or countably generated Hilbert
A-module H over a unital C ∗ -algebra A. Then {xj }j∈J is a Bessel sequence with
bound D if and only if
X
k hx, xj ihxj , xik ≤ Dkxk2
j∈J

holds for all x ∈ H.

Proof. ”⇒” Obvious.

”⇐” Define a linear operator T : H → l2 (A) by
X
Tx = hx, xj iej , ∀x ∈ H.
j∈J

Then
X
kT xk2 = khT x, T xik = k hx, xj ihxj , xik ≤ Dkxk2 ,
j∈J
√
which implies that kT xk ≤ Dkxk. Hence T is bounded.
It is obvious that T is A-linear. Then by Lemma 3.6, we have

hT x, T xi ≤ Dhx, xi,
P
equivalently, j∈J hx, xj ihxj , xi ≤ Dhx, xi, as desired.

With the same argument we obtain the following equivalent definition of

frames in Hilbert C ∗ -modules.

Proposition 3.8. Let H be a finitely or countably generated Hilbert A-module

H over a unital C ∗ -algebra A and {xj }j∈J ⊆ H a sequence. Then {xj }j∈J is a
frame of H with bounds C and D if and only if
X
Ckxk2 ≤ k hx, xj ihxj , xik ≤ Dkxk2
j∈J

for all x ∈ H.

34
 One of the advantages of this equivalent definition is that it is much easier to
compare the norms of two elements than to compare two elements in C ∗ -algebras.
Using the above equivalent definition of frames we can easily prove the fol-
lowing result which will be used in the proofs of Theorem 4.9, Theorem 4.13 and
Theorem 6.2.

Proposition 3.9. Suppose that H be a finitely or countably generated Hilbert

A-module H over a unital C ∗ -algebra A. Let {xj }j∈J and {yj }j∈J be two Bessel
P
sequences in H. If x = j∈J hx, yj ixj holds for any x ∈ H, then both {xj }j∈J and
P
{yj }j∈J are frames of H and x = j∈J hx, xj iyj holds for all x ∈ H.

Proof. Let’s denote the Bessel bound of {yj }j∈J by DY . For all x ∈ H we have
X X
kxk4 = kh hx, yj ixj , xik2 = k hx, yj ihxj , xik2
j∈J j∈J
X X
≤ k hx, yj ihyj , xik · k hx, xj ihxj , xik
j∈J j∈J
X
≤ DY kxk2 · k hx, xj ihxj , xik.
j∈J

It follows that
X
DY−1 kxk2 ≤ k hx, xj ihxj , xik.
j∈J

Similarly, we can show that {yj }j∈J is also a frame of H.

It follows directly from Proposition 6.3 in [26] that
X
x= hx, xj iyj
j∈J

holds true for all x ∈ H.

35
3.3 Characterizations of Frames and Bessel Sequences in
Hilbert C ∗ -modules

The aim of this section is to give some characterizations of Bessel sequences and
frames in Hilbert C ∗ -modules from the operator-theoretic point of view. These
results will be used to prove our main results in Chapter 7.
The following lemma is due to Heuser ([38]). Heuser only considered the
l2 (C)-sequence case, but his proof works in more general setting. We include the
proof here for the sake of completeness.

Lemma 3.10. Let A be a C ∗ -algebra and {cj }j∈J a sequence in A. If ∗

P
j∈J cj ξj
converges for all {ξj }j∈J ∈ l2 (A), then {cj }j∈J ∈ l2 (A).

Proof. We define a sequence of operators Fn and an operator F by

n
X ∞
X
Fn ({ξj }) = cj ξj∗ and F ({ξj }) = cj ξj∗ , ∀{ξj } ∈ l2 (A).
j=1 j=1

Observe that
n
X n
X n
X n
X
2
kFn ({ξj })k = k cj ξj∗ k2 ≤k cj c∗j k ·k ξj ξj∗ k 2
≤ k{cj }k · k ξj ξj∗ k.
j=1 j=1 j=1 j=1

It follows that Fn is bounded for each n.

Clearly, Fn → F pointwise as n → ∞, so F is bounded by the Uniform
Boundedness Theorem. Therefore kF ({ξj })k ≤ kF k·k{ξj }k for each {ξj } ∈ l2 (A).
Now fix n, and let (
c∗j , if 1 ≤ j ≤ n;
ξj =
0, otherwise.
Then {ξj } ∈ l2 (A).

36
 We compute
n
X n
X
k cj c∗j k = k cj ξj k ≤ kF k · k{ξj }k
j=1 j=1
∞ n
1 1
X X
= kF k · k ξj ξj∗ k 2 = kF k · k ξj ξj∗ k 2
j=1 j=1
n
1
X
= kF k · k cj c∗j k 2 .
j=1

1
∗ 21
Pn ∗ 2
P∞
Therefore k j=1 cj cj k ≤ kF k. It follows that k j=1 cj cj k ≤ kF k, and hence
2
{cj } ∈ l (A).

We first give a characterization of Bessel sequences in terms of operators in

Hilbert C ∗ -modules.

Proposition 3.11. Let {xj }j∈J be a sequence of a finitely or countably gener-

ated Hilbert A-module H over a unital C ∗ -algebra A. Then {xj }j∈J is a Bessel
sequence with Bessel bound D if and only if the operator U : l2 (A) → H defined
by
X
U {cj }j∈J = cj xj
j∈J
√
is a well-defined bounded operator from l2 (A) into H with kU k ≤ D.

Proof. ”⇒”. Suppose that {xj }j∈J is a Bessel sequence with bound D. We first
show that U is well-defined.

37
 For arbitrary n > m, we have
n
X m
X n
X
k cj xj − cj xj k2 = k c j x j k2
j=1 j=1 j=m+1
n
X
= sup kh cj xj , xik2
kxk=1 j=m+1
Xn
= sup k cj hxj , xik2
kxk=1 j=m+1
Xn n
X
≤ sup k hx, xj ihxj , xik · k cj c∗j k
kxk=1 j=m+1 j=m+1
n
X
≤ Dk cj c∗j k,
j=m+1
P
which implies that j∈J cj xj converges. Therefore U is well-defined.
To see the boundedness of U , we consider

kU {cj }k2 = sup khU {cj }, xik2

kxk=1
X
= sup k cj hxj , xik2
kxk=1 j∈J
X X
≤ sup k hx, xj ihxj , xik · k cj c∗j k
kxk=1 j∈J j∈J
X
≤ Dk cj c∗j k = Dk{cj }k2 .
j∈J

√
This yields that kU k ≤ D.
”⇐”. For arbitrary x ∈ H and {cj }j∈J ∈ l2 (A), we have
X X
hx, U {cj }i = hx, cj xj i = hx, xj ic∗j . (3.6)
j∈J j∈J

By Lemma 3.10, we see that {hx, xj i}j∈J ∈ l2 (A).

From (3.6), we get

hx, U {cj }i = h{hx, xj i}, {cj }i,

38
which implies that U is adjointable, with U ∗ x = {hx, xj i}j∈J , and hence U is
bounded.
Note that
X
kU ∗ xk2 = k hx, xj ihxj , xik ≤ Dkhx, xik = Dkxk2 .
j∈J

√
Consequently, kU k = kU ∗ k ≤ D, as desired.

For the case of frames in Hilbert C ∗ -modules we have the following two char-
acterizations.

Proposition 3.12. Let H be a finitely or countably generated Hilbert A-module

H over a unital C ∗ -algebra A. Suppose that {xj }j∈J is a sequence of H. Then
{xj }j∈J is a frame of H if and only if the operator U : l2 (A) → H defined by
X
U {cj }j∈J = cj xj
j∈J

is a well-defined bounded operator from l2 (A) onto H.

Proof. ”⇒”. Obvious.

”⇐”. By Proposition 3.11, {xj }j∈J is a Bessel sequence. Let D be the Bessel
bound of {xj }j∈J .
Note that for each x ∈ H, we have
X
x = U U ∗ (U U ∗ )−1 x = h(U U ∗ )−1 x, xj ixj .
j∈J

39
Now
X
kxk4 = khx, xik2 = k h(U U ∗ )−1 x, xj ihxj , xik2
j∈J
X X
≤ k h(U U ) x, xj ihxj , (U U ∗ )−1 xik · k
∗ −1
hx, xj ihxj , xik
j∈J j∈J
X
≤ Dkh(U U ∗ )−1 x, (U U ∗ )−1 xik · k hx, xj ihxj , xik
j∈J
X
= Dk(U U ∗ )−1 xk2 · k hx, xj ihxj , xik
j∈J
X
∗ −1 2
≤ Dk(U U ) k · kxk2 · k hx, xj ihxj , xik,
j∈J

which leads to the lower bound inequality of frame, that is

1 2
X
kxk ≤ k hx, xj ihxj , xik.
Dk(U U ∗ )−1 k2 j∈J

Proposition 3.13. Suppose that H is a finitely or countably generated Hilbert

A-module H over a unital C ∗ -algebra A. Let {xj }j∈J be a sequence of H, then
{xj }j∈J is a frame of H with bounds C and D if and only if
(1) span{xj : j ∈ J} = H;
(2) the operator U : l2 (A) → H defined by
X
U {cj }j∈J = cj xj
j∈J

is a well-defined bounded operator from l2 (A) into H and satisfies

√ √
Ck{cj }k ≤ kU {cj }k ≤ Dk{cj }k, ∀{cj } ∈ (KerU )⊥ . (3.7)

Proof. ”⇒”. Suppose first that {xj }j∈J is a frame. Let S be the frame operator
of {xj }j∈J . Then we have S = U U ∗ .
By Proposition 3.12, it is enough to show that
√
Ck{cj }k ≤ kU {cj }k

40
holds for all {cj } ∈ (KerU )⊥ .
Since {xj }j∈J is a frame, it follows that Rang(U ∗ ) is closed. Therefore we
have
(KerU )⊥ = Rang(U ∗ ) = Rang(U ∗ ).

As a sequence, (KerU )⊥ = {{hx, xj i}j∈J : x ∈ H}.

Now for any x ∈ H, we see that
X
k hx, xj ihxj , xik2 = khSx, xik2 ≤ kSxk2 · kxk2
j∈J
1 X
≤ kSxk2 · k hx, xj ihxj , xik.
C j∈J

≤ kSxk2 = kU U ∗ xk = kU {hx, xj i}k2 , as de-

P
Therefore Ck j∈J hx, xj ihxj , xik
sired.
”⇐”. To show that {xj }j∈J is a frame, by Proposition 3.12, it suffices to show
that Rang(U ) = H.
Since span{xj : j ∈ J} ⊆ Rang(U ), it only needs to prove that Rang(U ) is
closed.
Suppose that {un } ⊆ Rang(U ) and un → u as n → ∞. Then we can find
{vn } ⊆ (KerU )⊥ such that U vn = un .
It follows from (3.7) that {vn } is a Cauchy sequence. Suppose that vn → v
as n → ∞. Therefore un = U vn → U v = u as n → ∞. This completes the
proof.

3.4 Removal of Elements from Frames

It is obvious that if we remove an element from a basis, then we must get a set
which is not a basis. But for frame this is not the case. Due to the redundancy
of frame if we remove an element from a frame we may get a new frame. For

41
the removal of elements from frames in Hilbert spaces Christensen ([17]) gave the
following characterization.

Theorem 3.14. The removal of a vector fj from a frame {fk }∞

k=1 for Hilbert

space H leaves either a frame or an incomplete set. More precisely,

(1) If hfj , S −1 fj i =
6 1, then {fk }k6=j is a frame for H;
(2) If hfj , S −1 fj i = 1, then {fk }k6=j is incomplete;
where S is the corresponding frame operator.

We now introduce a lemma.

Lemma 3.15. Suppose that H is a finitely or countably generated Hilbert A-

module H over a unital C ∗ -algebra A. Let {xj }j∈J be a frame of H with analysis
P
operator T and frame operator S. Let x ∈ H and suppose that x = j∈J aj xj ,
where aj ∈ A for each j ∈ J. Then

aj a∗j = −1
xj ihS −1 xj , xi
P P
j∈J j∈J hx, S
− hx, S −1 xj i)(a∗j − hS −1 xj , xi).
P
+ j∈J (aj

Proof. For each j ∈ J we can write aj = (aj − hx, S −1 xj i) + hx, S −1 xj i.

Since {xj : j ∈ J} is a frame, we have x = j∈J hx, S −1 xj ixj , and so
P

X
(aj − hx, S −1 xj i)xj = 0,
j∈J

i.e. {aj − hx, S −1 xj i}j∈J ∈ KerT ∗ .

Note that {hx, S −1 xj i}j∈J = {hS −1 x, xj i}j∈J ∈ Rang(T ).
Since l2 (A) = KerT ∗ ⊕ T (H), we see that

{hx, S −1 xj i}j∈J ⊥ {aj − hx, S −1 xj i}j∈J ,

which completes the proof.

We now generalize Theorem 3.14 to the situation of frames in Hilbert C ∗ -
modules.

42
Theorem 3.16. Suppose that H is a finitely or countably generated Hilbert A-
module H over a unital C ∗ -algebra A. Let {xj }j∈J be a frame for H and 1A the
identity element of A. We have the following statements.
(1) if 1A − hxn , S −1 xn i is invertible in A, then {xj : j 6= n}j∈J is a frame for
H;
(2) if 1A − hxn , S −1 xn i is not invertible in A, then {xj : j 6= n}j∈J is not a
frame for H.

−1
P
Proof. By the frame decomposition we have xn = j∈J hxn , S xj ixj .
Define, for notational convenience, aj = hxn , S −1 xj i, for each j ∈ J. Then
P
xn = j∈J aj xj .
(1) Suppose that 1A − hxn , S −1 xn i = 1A − an is invertible.
From xn = j∈J aj xj = an xn + j6=n aj xj , we have xn = (1A −an )−1 j6=n aj xj .
P P P

Now for any x ∈ H, we see that

hx, xn ihxn , xi
= hx, (1A − an )−1 j6=n aj xj ih(1A − an )−1 j6=n aj xj , xi
P P

= hx, j6=n aj xj i((1A − an )−1 )∗ (1A − an )−1 h j6=n aj xj , xi

P P

≤ k(1A − an )−1 k2 hx, j6=n aj xj ih j6=n aj xj , xi)

P P

= k(1A − an )−1 k2 j6=n (hx, xj ia∗j ) j6=n (aj hxj , xi)

P P

≤ k(1A − an )−1 k2 k j6=n aj a∗j k2 j6=n (hx, xj ihxj , xi).

P P

Therefore
P
Chx, xi ≤ j∈J hx, xj ihxj , xi
P
= hx, xn ihxn , xi + j6=n hx, xj ihxj , xi
−1 2 ∗ 2
P P
≤ (k(1A − an ) k k j6=n aj aj k + 1) j6=n hx, xj ihxj , xi,

showing that {xj : j 6= n}j∈J satisfies the lower frame condition.

Obviously, {xj : j 6= n}j∈J also satisfies the upper frame condition.
(2) Suppose that 1A − hxn , S −1 xn i = 1A − an is not invertible in A.

43
 1
Assume on the contrary that {xj : j 6= n}j∈J is a frame, since S − 2 is invertible,
1
it follows that {S − 2 xj : j 6= n}j∈J is also a frame with frame bound C̃ and D̃.
Then

1 1
X
C̃hx, xi ≤ hx, S − 2 xj ihS − 2 xj , xi ≤ D̃hx, xi
j6=n

holds for all x ∈ H.

1
In particular, for x = S − 2 xn we have
1 1 1 1 1 1
X
C̃hS − 2 xn , S − 2 xn i ≤ hS − 2 xn , S − 2 xj ihS − 2 xj , S − 2 xn i,
j6=n

i.e. C̃hxn , S −1 xn i ≤ −1
xj ihS −1 xj , xn i.
P
j6=n hxn , S

This implies that

X
C̃an ≤ aj a∗j .
j6=n

From Lemma 3.15 we see that an = a2n + aj a∗j .

P
j6=n

Then we have C̃an ≤ an − a2n , and so C̃t ≤ t − t2 holds for any t in σ(an ), the
spectrum of an .
Since 1A − an is not invertible, it follows that 1 ∈ σ(an ). Therefore C̃ · 1 ≤
1 − 1 · 1 = 0, a contradiction. This completes the proof.

44
 CHAPTER 4

DUALS AND MODULAR RIESZ BASES

The main purpose of this chapter is to investigate the Riesz bases in Hilbert C ∗ -
modules. It is well-known that in Hilbert spaces every Riesz basis has a unique
dual which is also a Riesz basis. But in Hilbert C ∗ -modules, due to the zero-
divisors, not all Riesz bases have unique duals and not every dual is a Riesz
basis. We will present several such examples showing that the duals of Riesz
bases in Hilbert C ∗ -modules are much different and more complicated than the
Hilbert space cases. We give a complete characterization of all the dual sequences
for a Riesz basis, and a necessary and sufficient condition for a dual sequence of
a Riesz basis to be a Riesz basis.

4.1 Characterizations of Riesz Bases in Hilbert

C ∗ -modules

In this section we shall give a characterization of Riesz bases in Hilbert C ∗ -

modules which will be used in the latter part of this thesis.
Note that in Hilbert space case a frame is a Riesz basis if and only if its
analysis operator is surjective [37]. This is no longer true for Hilbert C ∗ -module
frames.
We first introduce a notation.

45
 Let Pn be the projection on l2 (A) that maps each element to its n-th compo-
nent, i.e. Pn x = {uj }j∈J , where
(
xn if j = n,
uj =
0 if j 6= n,

for each x = {xj }j∈J ∈ l2 (A).

We now prove the first main result of this chapter.

Theorem 4.1. Let {xj }j∈J be a frame of a finitely or countably generated Hilbert
A-module H over a unital C ∗ -algebra A. Then {xj }j∈J is a Riesz basis if and
only if xn 6= 0 and Pn (Rang(TX )) ⊆ Rang(TX ) for all n ∈ J, where TX is the
analysis operator of {xj }j∈J .

Proof. Suppose first that {xj }j∈J is a Riesz basis.

Note that for any a = {aj }j∈J in l2 (A), if j∈J aj xj = 0, then aj xj = 0 for
P

all j.
Therefore, if a ⊥ Rang(TX ), then a ⊥ Pn (Rang(TX )). It follows that
Pn (Rang(TX )) ⊆ Rang(TX ).
Suppose now that Pn (Rang(TX )) ⊆ Rang(TX ) for each n. We want to show
that {xj }j∈J is a Riesz basis.
P
Suppose that j∈J aj xj = 0, where aj ∈ A.
Fix an n ∈ J, then Pn TX x ∈ Rang(TX ), so there exists yn ∈ H such that
TX yn = Pn TX x.
Therefore we get
(
hx, xn i if j = n,
hyn , xj i =
0 if j 6= n.

Now for any x ∈ H we have

X
hx, an xn i = hx, xn ia∗n = hyn , xj ia∗j
j∈J
X X
= hyn , aj xj i = hyn , aj xj i = 0,
j∈J j∈J

46
which implies that an xn = 0.
P
Note that in Hilbert spaces, if {xj }j∈J is a Riesz basis and j∈J cj xj converges
for a sequence {cj } ⊆ C, then {cj } ∈ l2 . But this is not the case in the setting of
Hilbert C ∗ -modules. We have the following example.

Example 4.2. Let l∞ be the set of all bounded complex-valued sequences. For
any u = {uj }j∈N and v = {vj }j∈N in l∞ , we define

uv = {uj vj }j∈N , u∗ = {ūj }j∈N and kuk = max |uj |.

j∈N

Then A = {l∞ , k · k} is a C ∗ -algebra.

Let H = c0 be the set of all sequences converging to zero. For any u, v ∈ H
we define
hu, vi = uv ∗ = {uj v̄j }j∈N .

Then H is a Hilbert A-module.

Obviously, {ej }j∈N is an orthonormal basis of H.
√
For each j we let cj = jej+1 .
Then cj ej = 0 and so ∞
P
j=1 cj ej = 0.

But ∞ ∗
P∞
/ l2 (A).
P
j=1 cj cj = j=2 jej doesn’t converge in A. Thus {cj } ∈

Following the definition of Riesz bases in Hilbert C ∗ -modules, to test a se-

P
quence {xj }j∈J is a Riesz basis, we need to show that if j∈J cj xj = 0 for some
sequence {cj }j∈J ⊆ A, then cj xj = 0 for each j. We claim that we can restrict
the sequence {cj }j∈J in l2 (A).

Corollary 4.3. Suppose that {xj }j∈J is a frame of H, then {xj }j∈J is a Riesz
basis if and only if
6 0 for each j;
(1) xj =
(2) if j∈J cj xj = 0 for some sequence {cj }j∈J ∈ l2 (A), then cj xj = 0 for
P

each j.

Proof. See the proof of Theorem 4.1.

47
 4.2 Duals of Riesz Bases in Hilbert C ∗ -modules

The aim of this section is to have a detailed investigation on the dual sequences of
Riesz bases in Hilbert C ∗ -modules. Some of the results presented in this section
will be needed in proving Theorem 6.2.
We first introduce the following definition.

Definition 4.4. Suppose that H is a Hilbert A-module over a unital C ∗ -algebra

A. Let {xj }j∈J be a frame and {yj }j∈J a sequence of H. Then {yj }j∈J is said to
be a dual sequence of {xj }j∈J if
X
x= hx, yj ixj (4.1)
j∈J

holds for all x ∈ H, where the sum in (4.1) converges in norm. The pair {xj }j∈J
and {yj }j∈J are called a dual frame pair when {yj }j∈J is also a frame.

It should be mentioned that , contrasting to the Hilbert space situation, Riesz

bases of Hilbert C ∗ -modules may posses infinitely many dual frames due to the
existence of zero-divisors in the C ∗ -algebra of coefficients. The following three
simple examples show that the dual of Riesz bases of Hilbert C ∗ -modules are
quite different from and more complicated than the Hilbert space cases.
The first example shows that in Hilbert C ∗ -modules the dual Riesz basis of a
Riesz basis is not unique.

Example 4.5. Let A = M2×2 (C) denote the C ∗ -algebra of all 2 × 2 complex
matrices. Let H = A and for any A, B ∈ H define

hA, Bi = AB ∗ .

Then H is a Hilbert A-module.

Let Ei,j be the 2 × 2 matrix with 1 in the (i, j)-th entry and 0 elsewhere,
where 1 ≤ i, j ≥ 2.
Then {E1,1 , E2,2 } is a Riesz basis of H and it is a dual of itself.
One can check that {E1,1 + E2,1 , E2,2 } is also a dual Riesz basis of {E1,1 , E2,2 }.

48
 It is well-known that if {xj }j∈J is a Riesz basis and {yj }j∈J is a dual sequence
of {xj }j∈J in a Hilbert space H, then {yj }j∈J is a Riesz basis which is the unique
dual of {xj }j∈J . The following example shows that this is not the case in Hilbert
C ∗ -modules.

Example 4.6. Suppose H and A are the same as in Example 4.2.

Note that {ej }j∈N is an orthonormal basis of H.
Now let xj = ej and
(
e1 if j = 1,
yj =
ej + jej−1 if j 6= 1.

One can verify that

X
x= hx, yj ixj
j∈N

holds for all x ∈ H, but {yj }j∈N is not a Riesz basis, even not a Bessel sequence.

Note that even the dual sequence of a Riesz basis in Hilbert C ∗ -modules is
a Bessel sequence, it still has the chance not to be a Riesz basis. We have the
following example.

Example 4.7. Suppose H and A are the same as in Example 4.2.

Now let xj = ej and
(
e1 + e2 if j = 1, 2,
yj =
ej if j 6= 1, 2.

Then {yj }j∈N is a Bessel sequence, and satisfies

X
x= hx, yj ixj
j∈N

for all x ∈ H.
Therefore, {yj }j∈N is a frame of H. But obviously {yj }j∈N is not a Riesz basis.

The following lemma will be needed in several places in the rest of this thesis
(in particular, it will be needed in the proof of Theorem 6.2).

49
Lemma 4.8. Let {xj }j∈J be a frame of a finitely or countably generated Hilbert
A-module H over a unital C ∗ -algebra A. Suppose that {yj }j∈J and {zj }j∈J are
dual frames of {xj }j∈J with the property that either Rang(TY ) ⊆ Rang(TZ ) or
Rang(TZ ) ⊆ Rang(TY ), where TY and TZ are the analysis operators of {yj }j∈J
and {zj }j∈J respectively. Then yj = zj for all j ∈ J.

Proof. Suppose that Rang(TZ ) ⊆ Rang(TY ). Then for each x ∈ H there exists
yx ∈ H such that
TY yx = TZ x.

Applying TX∗ on the both sides we arrive at

yx = TX∗ TY yx = TX∗ TZ x = x,

and so TY x = TZ x for all x ∈ H.

Equivalently,
X X
hx, yj iej − hx, zj iej = 0,
j∈J j∈J
P
i.e. j∈J hx, yj − zj iej . Hence yj = zj for all j.

We now give a necessary and sufficient condition about the uniqueness of dual
frames in Hilbert C ∗ -modules. We also prove that if a frame has a unique dual
frame then it must be a Riesz basis.

Theorem 4.9. Suppose that H is a finitely or countably generated Hilbert A-

module over a unital C ∗ -algebra A. Let {xj }j∈J be a frame of H with analysis
operator TX , then the following statements are equivalent:
(1) {xj }j∈J has a unique dual frame;
(2) Rang(TX ) = l2 (A).
In case the equivalent conditions are satisfied, {xj }j∈J is a Riesz basis.

Proof. (2)⇒(1). Let {x∗j }j∈J be the canonical dual of {xj }j∈J with analysis oper-
−1
ator TX ∗ . Then x∗j = SX xj , where SX is the frame operator of {xj }j∈J .

50
 Let {yj }j∈J be any dual frame of {xj }j∈J with analysis operator TY , then

Rang(TY ) ⊆ l2 (A) = Rang(TX ) = Rang(TX ∗ ).

By Lemma 4.8, yj = x∗j for all j.

(1)⇒(2). Assume on the contrary that Rang(TX ) 6= l2 (A).
By Theorem 2.57, we have

l2 (A) = Rang(TX ) ⊕ KerTX∗ .

Let PX be the orthogonal projection from l2 (A) onto Rang(TX ), then

l2 (A) = PX l2 (A) ⊕ PX⊥ l2 (A).

Therefore PX⊥ l2 (A) = KerTX∗ 6= {0}.

Choose ej0 such that PX⊥ ej0 6= 0 and define an operator U : PX⊥ l2 (A) → H by

U w = hw, PX⊥ ej0 ixj0 .

Then U is an adjointable linear operator.

Now let {x∗j }j∈J be the canonical dual of {xj }j∈J with upper bound DX ∗ and
set yj = x∗j + U PX⊥ ej .
We have
X X
hx, yj ihyj , xi = hx, x∗j + U PX⊥ ej ihx∗j + U PX⊥ ej , xi
j∈J j∈J
X X
≤ 2( hx, x∗j ihx∗j , xi + hPX⊥ U ∗ x, ej ihej , PX⊥ U ∗ xi)
j∈J j∈J
X
≤ 2(DX ∗ hx, xi + hPX⊥ U ∗ x, PX⊥ U ∗ xi),
j∈J

which implies that {yj }j∈J is a Bessel sequence.

Now for any x ∈ H,
X X X
hx, U PX⊥ ej ixj = TX∗ hx, U PX⊥ ej iej = TX∗ hPX⊥ U ∗ x, ej iej
j∈J j∈J j∈J

= TX∗ PX⊥ U ∗ x = 0.

51
 P
This yields that x = j∈J hx, yj ixj for all x ∈ H. By Proposition 3.9, {yj }j∈J is
a dual frame of {xj }j∈J and is different from {x∗j }j∈J , which contradicts with the
uniqueness of the dual frame of {xj }j∈J .
To complete the proof it remains to prove that {xj }j∈J is a Riesz basis if one
of the equivalent conditions holds. Suppose now that Rang(TX ) = l2 (A).
P
If j∈J aj xj = 0 for aj ∈ A. Then
X X X
0= aj x j = aj TX∗ ej = TX∗ aj ej .
j∈J j∈J j∈J

aj ej = 0 as TX∗ is injective. Hence aj = 0 for all j.

P
Therefore j∈J

Note that xj = TX∗ ej for each j ∈ J. It follows from the injectiveness of TX∗
that xj 6= 0.

Remark 4.10. By the above theorem, Example 4.5 shows that, thought {E1,1 , E2,2 }
is a Riesz basis of H, the corresponding analysis operator is not surjective which
is different from the case in Hilbert spaces.

We also have another characterization on the uniqueness of dual frames of

Riesz bases in Hilbert C ∗ -modules.

Proposition 4.11. Let {xj }j∈J be a sequence of a finitely or countably generated

Hilbert A-module H over a unital C ∗ -algebra A, then {xj }j∈J is a Riesz basis
with unique dual frame if and only if
(1) span{xj : j ∈ J} = H;
(2) there exist C, D ≥ 0 such that

√ X √
Ck{cj }k ≤ k cj xj k ≤ Dk{cj }k, ∀{cj }j∈J ∈ l2 (A).
j∈J

Proof. It follows from Theorem 4.9 and Proposition 3.13.

We now study the dual sequences of Riesz bases in Hilbert C ∗ -modules. The
following theorem is straightforward.

52
Theorem 4.12. Suppose that {xj }j∈J is a Riesz basis of a finitely or countably
generated Hilbert A-module H over a unital C ∗ -algebra A. Let {yj }j∈J be a se-
quence of H. Then the following statements are equivalent.
(1) {yj }j∈J is a dual frame of {xj }j∈J ;
(2) {yj }j∈J is a dual Bessel sequence of {xj }j∈J ;
(3) for each j ∈ J, yj = S −1 xj + zj , where S is the frame operator of {xj }j∈J ,
and {zj }j∈J is a Bessel sequence of H satisfying hx, zj ixj = 0 for all x ∈ H and
j ∈ J.

For the case of a dual sequence of a Riesz basis to be a Riesz basis, we have
the following characterization.

Theorem 4.13. Let {xj }j∈J be a Riesz basis and {yj }j∈J a sequence of a finitely
or countably generated Hilbert A-module H over a unital C ∗ -algebra A. Then
{yj }j∈J is a dual Riesz basis of {xj }j∈J if and only if for each j ∈ J, yj =
S −1 xj + zj , where S is the frame operator of {xj }j∈J , and {zj }j∈J is a Bessel
sequence of H with the property that for each j ∈ J there exists bj ∈ A such that
zj = bj S −1 xj and hx, xj ibj xj = 0 holds for all x ∈ H.

Proof. ”⇒”. Suppose that {yj }j∈J is a dual Riesz basis of {xj }j∈J and let zj =
yj − S −1 xj .
Then it is easy to see that {zj }j∈J is a Bessel sequence of H.
Now fix an n ∈ J.
P
From yn = j∈J hyn , xj iyj we can infer that yn = hyn , xn iyn , i.e.

S −1 xn + zn = hS −1 xn + zn , xn i(S −1 xn + zn ).

Consequently, we have

zn = hzn , xn iS −1 xn + hS −1 xn , xn izn + hzn , xn izn .

To show that hS −1 xn , xn izn + hzn , xn izn = 0, it suffices to show that

hS −1 xn , xn ihzn , xi + hzn , xn ihzn , xi = 0

53
holds for all x ∈ H.
Note that
X X X X
x= hx, yj ixj = hx, S −1 xj ixj + hx, xj ixj = x + hx, zj ixj ,
j∈J j∈J j∈J j∈J
P
which implies that j∈J hx, zj ixj = 0 and so hx, zj ixj = 0 for all x ∈ H and j ∈ J.
Particularly, we have hx, zn ixn = 0 for all x ∈ H. This yields that

hx, zn ihxn , zn i = 0 and hx, zn ihxn , S −1 xn i = 0.

Equivalently, hzn , xn ihzn , xi = 0 and hS −1 xn , xn ihzn , xi = 0.

Therefore zn = bn S −1 xn , where bn = hzn , xn i.
From hxn , zn ixn = 0, we have

hy, xn ihxn , zn ihxn , xi = 0

for all x, y ∈ H, which is equivalent to hx, xn ihzn , xn ihxn , yi = 0, this implies that

hx, xn ibn xn = hx, xn ihzn , xn ixn = 0.

”⇐”. Suppose now that for each j ∈ J there exists bj ∈ A such that zj =
−1
bj S xj and hx, xj ibj xj = 0 holds for all x ∈ H. Then for all x, y ∈ H we have

hx, xj ibj hxj , yi = 0.

Equivalently,
hy, xj ib∗j hxj , xi = 0.
This implies that hy, xj ib∗j xj = 0 for all y ∈ H.
Now for arbitrary x ∈ H,
X X X
hx, yj ixj = hx, S −1 xj ixj + hx, zj ixj
j∈J j∈J j∈J
X
−1
= x+ hx, bj S xj ixj
j∈J
X
= x+ hx, S −1 xj ib∗j xj
j∈J
X
= x+ hS −1 x, xj ib∗j xj
j∈J
= x,

54
which implies that {yj }j∈J is a dual sequence of {xj }j∈J .
One can easily see that {yj }j∈J is a dual frame of {xj }j∈J by Proposition 3.9.
To complete the proof, we need to show that {yj }j∈J is a Riesz basis of H.
P
Suppose that j∈J aj yj = 0, then we have
X X
0= aj (S −1 xj + bj S −1 xj ) = aj (1 + bi )S −1 xj .
j∈J j∈J

Therefore aj (1 + bj )S −1 xj = 0, i.e. aj yj = 0 for all j.

We now show that yj 6= 0 for each j ∈ J.
Assume on the contrary that yn = 0 for some n ∈ J. Then zn = −S −1 xn . It
follows that
0 = hx, xn ibn xn = hx, xn iSzn = −hx, xn ixn

holds for all x ∈ H.

In particular, letting x = S −1 xn , we have 0 = −hS −1 xn , xn ixn = −xn , and so
xn = 0, a contradiction. This completes the proof.

Corollary 4.14. Suppose that H is a finitely or countably generated Hilbert A-

module over a unital C ∗ -algebra A. If A is commutative, then every Riesz basis
of H has a unique dual Riesz basis.

Proof. Choose an arbitrary Riesz basis {xj }j∈J of H. Suppose that {S −1 xj +zj }j∈J
is a dual Riesz basis of {xj }j∈J , where S is the frame operator of {xj }j∈J .
Then by Theorem 4.13, for each j ∈ J there exists bj ∈ A such that zj =
bj S −1 xj and hx, xj ibj xj = 0 holds for all x ∈ H.
Since A is commutative, we have bj hx, xj ixj = 0 for all x ∈ H and j ∈ J.
Let x = S −1 xj . We have

0 = bj hS −1 xj , xj ixj = bj hxj , S −1 xj ixj = bj xj ,

which yields that zj = bj S −1 xj = 0.

55
 Note that under that conditions of Corollary 4.14, though a Riesz basis has a
unique dual Riesz basis, it may have many dual frames. We have the following
example.

Example 4.15. Let A = D2×2 (C) denote the C ∗ -algebra of all 2 × 2 complex
diagonal matrices. Let H = A and for any A, B ∈ H define

hA, Bi = AB ∗ .

Then H is a Hilbert A-module.

It is obvious that A is commutative.
Let Ei,j be the 2 × 2 matrix with 1 in the (i, j)-th entry and 0 elsewhere,
where 1 ≤ i, j ≤ 2.
Then {E1,1 , E2,2 } is a Riesz basis of H, and so it has a unique dual Riesz basis
which is itself.
But the dual frame of {E1,1 , E2,2 } is not unique. For example, one can verify
that {E1,1 +αE2,2 , βE1,1 +E2,2 } is also a dual frame of {E1,1 , E2,2 } for any α, β ∈ C.

The following example shows that the converse of Corollary 4.14 is not true,
namely, if every Riesz basis of a Hilbert A-module H has a unique dual Riesz
basis, A is not necessarily commutative.

Example 4.16. Let  

a 0 0
 
H = { 0 0 0 

 : ∀a ∈ C}
0 0 0
and  
a 0 0
 
A = {
 0 b c  : ∀a, b, c, d, e ∈ C}.

0 d e

For any A, B ∈ H define

hA, Bi = AB ∗ .

56
Then H is a A-module.
Note that A is not commutative.
Let  
α 0 0
 
Eα = 
 0 0 0 .

0 0 0
Then {Eα } is a Riesz basis of H.
It is easy to see that any Riesz basis of H has the form of {Eα } for some
nonzero α ∈ C. And one can also check that every dual Riesz basis of {Eα } for
each nonzero α is unique.

We now introduce a notation. For a Hilbert A-module H, let

CA (H) = {a ∈ A : abx = bax, ∀b ∈ A, x ∈ H}.

To complete this section we pose a conjecture as follows.

Conjecture. Suppose that H is a finitely or countably generated Hilbert
A-module over a unital C ∗ -algebra A. Then every Riesz basis of H has a unique
dual Riesz basis if and only if A = CA (H).

57
 CHAPTER 5

STRUCTURED MODULAR FRAMES

The purpose of this chapter is to focus on structured frames in Hilbert C ∗ -

modules. More precisely, we work on two closely related issues: frame vector
parameterizations and Parseval frame approximations. In the case that the un-
derlying C ∗ -algebra is a commutative W ∗ -algebra, we prove that the set of the
complete Parseval frame vectors for a unitary group can be parameterized by the
set of all the unitary operators in the double commutant of the unitary group.
Similar result holds for the set of all the complete frame generators where the
unitary operators are replaced by invertible and adjointable operators. Conse-
quently, the set of all the complete Parseval frame vectors is path-connected.
We also prove the existence and uniqueness of the best Parseval multi-frame ap-
proximations for multi-frame generators of unitary groups on Hilbert C ∗ -modules
when the underlying C ∗ -algebra is commutative.

5.1 Modular Frame Vector Parameterizations

The aim of this section is to investigate the parameterizations of frame vectors

in Hilbert C ∗ -modules.
Let’s first introduce a few more notation. Let S ⊆ EndA (H), we denote its
commutant {A ∈ EndA (H) : AS = SA, S ∈ S} by S 0 . Let x ∈ H be a nonzero
vector, the local commutant Cx (S) of S at x is defined by

Cx (S) = {A ∈ EndA (H) : ASx = SAx, S ∈ S}.

58
 A unitary system U on H is a set of unitary operators acting on H which
contains the identity operator.
A vector ψ in H is called a complete frame vector (resp. complete Parseval
frame vector, complete Riesz basis vector, Bessel sequence vector ) for a unitary
system U on H if Uψ = {U ψ : U ∈ U} is a frame (resp. Parseval frame, Riesz
basis, Bessel sequence) for H. If Uψ is an orthonormal basis of H, then ψ is
called a complete wandering vector for U.
For a unitary system U on H, let lU2 (A) be the Hilbert A-module defined by
X
lU2 (A) = {{aU }U ∈U ⊆ A : aU a∗U converges in k · k}.
U ∈U

Let {eU }U denote the standard orthonormal basis of lU2 (A), where eU takes value
1A at U and 0A at everywhere else. For the case that U is a unitary group on
H, we have the left and right regular representation for each U ∈ U which are
defined by
LU eV = eU V and RU eV = eV U .

Note that L−1 ∗ −1 ∗

U = LU = LU ∗ and RU = RU = RU ∗ .

The following two propositions can be viewed as the analogue of the corre-
sponding results for Hilbert space frames obtained in [19].

Proposition 5.1. Let U be a unitary system on a finitely or countably generated

Hilbert A-module H over a unital C ∗ -algebra A. Suppose that H has orthonormal
bases and η is a complete wandering vector for U. For ξ ∈ H, we have
(1) ξ is a complete wandering vector for U if and only if there exists a unitary
T ∈ Cη (U) such that ξ = T η.
(2) ξ is a complete Riesz basis vector for U if and only if there exists an
invertible and adjointable operator T ∈ Cη (U) such that ξ = T η.
(3) ξ is a complete Parseval frame vector for U if and only if there exists a
co-isometry T ∈ Cη (U) such that ξ = T η.

59
 (4) ξ is a complete frame vector for U if and only if there exists an adjointable
operator T ∈ Cη (U) with Chx, xi ≤ hT ∗ x, T ∗ xi for some C > 0 and any x ∈ H
such that ξ = T η.
(5) ξ is a complete Bessel sequence vector for U if and only if there exists an
adjointable operator T ∈ Cη (U) such that ξ = T η.

Proof. We will prove (3). And others go similarly.

Suppose that ξ = T η for some unitary operator T ∈ Cη (U). Then for any
x ∈ H, we have
X
hx, xi = hT ∗ x, T ∗ xi = hT ∗ x, U ηihU η, T ∗ xi
U ∈U
X
= hx, T U ηihT U η, xi
U ∈U
X
= hx, U T ηihU T η, xi
U ∈U
X
= hx, U ξihU ξ, xi,
U ∈U

which implies that ξ is a complete Parseval frame vector for U.

We now assume that ξ is a complete Parseval frame vector for U.
Define two operators Tη and Tξ from H to lU2 (A) respectively by
X X
Tη x = hx, U ηieU and Tξ x = hx, U ξieU .
U ∈U U ∈U

It is easy to check that both Tη and Tξ are well-defined and adjointable.

Let T = Tξ∗ Tη . Then for any x ∈ H, we have
X
Tx = hx, U ηiU ξ
U ∈U

and
X
T ∗x = hx, U ξiU η.
U ∈U

We now show that T is a co-isometry.

60
 Indeed, for any x ∈ H, we see that
X X
hT ∗ x, T ∗ xi = h hx, U ξiU η, hx, U ξiU ηi
U ∈U U ∈U
X
= hx, U ξihU ξ, xi = hx, xi.
U ∈U

To complete the proof, it remains to prove that ξ = T η and T ∈ Cη (U).

For any V and U in U,
X
hV ξ, T U ηi = hV ξ, hU η, W ηiW ξi
W ∈U
X
= hV ξ, W ξihW η, U ηi
W ∈U
= hV ξ, U ξi,

this implies that T U η = U ξ. Particularly, we have T η = ξ. And so T U η = U ξ =

U T η, as desired.

Note that if U is a unitary system which is not a group, and if U has a complete
wandering vector, then U is not even a semigroup. We have the following result.

Proposition 5.2. Let S be a unital semigroup of unitaries on a finitely or count-

ably generated Hilbert A-module H over a unital C ∗ -algebra A. Suppose that S
has a complete wandering vector. Then S is a group.

Proof. Let η be a complete wandering vector for S and U ∈ U. For any x ∈ H,

we have
X
hx, V ηihV η, xi = hx, xi = hU ∗ x, U ∗ xi
V ∈U
X
= hU ∗ x, V ηihV η, U ∗ xi
V ∈S
X
= hx, U V ηihU V η, xi.
V ∈S

Assume on the contrary that S is not a group, then there exists U0 ∈ S such
that U0−1 ∈
/ S. Then I ∈
/ U0 S. Since U0 S ⊆ S, it follows from (5.1) that

hx, ηihη, xi = 0, ∀x ∈ H.

61
 In particular, if x = η, we have

hη, ηihη, ηi = 0

and hence hη, ηi = 0, therefore η = 0, a contradiction.

To prove our first main result, we need the following:

Lemma 5.3. Let G be a unitary group on a Hilbert A-module over a commutative

unital C ∗ -algebra A, then

M = N 0 = {RU : U ∈ G}0 and N = M0 = {LU : U ∈ G}0 ,

where M = {LU : U ∈ G}00 and N = {RU : U ∈ G}00 .

Proof. Note that RU LV = LV RU holds for any U, V ∈ G. Therefore to prove this

lemma it suffices to show that T S = ST for arbitrary T ∈ M0 and S ∈ N 0 .
Suppose that
X X
T eI = aU eU and SeI = bU eU
U ∈G U ∈G

for some aU , bU ∈ A.
Now for any V ∈ G, on one hand, we have

ST eV = ST LV eI = SLV T eI
X X
= SLV ( aU eU ) = S( aU eV U )
U ∈G U ∈G
X X
= S( aU RV U eI ) = aU RV U SeI
U ∈G U ∈G
X X
= aU R V U ( bW eW )
U ∈G W ∈G
X
= aU bW eW V U .
U,W ∈G

62
On the other hand

T SeV = T SRV eI = T RV SeI

X X
= T RV ( bW eW ) = T ( bW eW V )
W ∈G W ∈G
X X
= T( bW LW V eI ) = bW LW V T eI
W ∈G W ∈G
X X
= b W LW V ( aU eU )
W ∈G U ∈G
X
= bW aU eW V U .
U,W ∈G

Since A is commutative, it follows that ST eV = T SeV , and so ST = T S.

We now define a natural conjugate A-linear isomorphism π from M onto

M0 = N by
π(A)BeI = BA∗ eI , ∀A, B ∈ M.

In particular, π(A)eI = A∗ eI .
Now we are in a position to prove the parameterization of complete Parseval
frames for unitary groups.

Theorem 5.4. Let G be a unitary group on a finitely or countably generated

Hilbert A-module H over a commutative unital W ∗ -algebra A such that lG2 (A) is
self-dual. Suppose that η ∈ H be a complete Parseval frame vector for G. For ξ
in H we have
(1) ξ is a complete Parseval frame vector for G if and only if there exists a
unitary operator A ∈ G 00 such that ξ = Aη.
(2) ξ is a complete frame vector for G if and only if there exists an invertible
and adjointable operator A ∈ G 00 such that ξ = Aη.
(3) ξ is a complete Bessel sequence vector for G if and only if there exists an
adjointable operator A ∈ G 00 such that ξ = Aη.

Proof. We will prove (1). The proof of (2) and (3) is similar and we leave it to
the interested readers.

63
 Let M = {LU : U ∈ G}00 . As η is a complete Parseval frame vector for G, we
have the corresponding analysis operator Tη which is given by
X
Tη x = hx, U ηieU .
U ∈G

From the proof of Theorem 6.1, we know that Tη is an isometry with closed range
and satisfies
Tη U = LU Tη and Tη η = P eI ,

where P is the orthogonal projection from lG2 (A) onto the range of Tη , and we
also have P ∈ M0 . Note that G is unitarily equivalent to the group {LU , U ∈ G}.
Therefore, to prove this theorem is equivalent to prove the theorem for the case
that G̃ = {LU |Rang(P ) , U ∈ G} and η̃ = P eI .
Sufficiency. Suppose that we have a unitary operator A ∈ G̃ 00 such that
ξ˜ = Aη̃.
We now show that Aη̃ is a complete Parseval frame vector for G̃. For any
x ∈ Rang(P ) = Rang(Tη ), we have
X X
hx, Ũ Aη̃ihŨ Aη̃, xi = hx, LU P Aη̃ihLU P Aη̃, xi
Ũ ∈G̃ U ∈G
X
= hx, LU P AP eI ihLU P AP eI , xi
U ∈G
X
= hx, LU P AeI ihLU P AeI , xi
U ∈G
X
= hx, P LU AeI ihP LU AeI , xi
U ∈G
X
= hP x, LU AeI ihLU AeI , P xi
U ∈G
X
= hx, LU AeI ihLU AeI , xi
U ∈G
X
= hx, π(A∗ )LU eI ihπ(A∗ )LU eI , xi
U ∈G
X
= h(π(A∗ ))∗ x, eU iheU , (π(A∗ ))∗ xi
U ∈G
= h(π(A∗ ))∗ x, (π(A∗ ))∗ xi = hx, xi,

64
which means that Aη̃ is a complete Parseval frame vector for G̃.
Necessity. Let ξ˜ ∈ Rang(P ) be a complete Parseval frame vector for G̃. We
want to find a unitary operator A ∈ G̃ 00 such that ξ˜ = Aη̃.
To this aim, we first define an operator B : lG2 (A) → lG2 (A) by

eU 7−→ LU ξ, U ∈ G.

One can check that B is an adjointable operator and B ∗ eV =

P ˜
W ∈G hLW −1 LV η̃, ξieW
for any V ∈ G.
Now for any U, V ∈ G, we see that

h(BB ∗ − P )eU , eV i
X X
= h ˜ W,
hLW −1 LU η̃, ξie ˜ S i − hTη U η, eV i
hLS −1 LV η̃, ξie
W ∈G S∈G
X X
= ˜ ξ,
hLW −1 LU η̃, ξih ˜ LW −1 LV η̃i − h hU η, W ηieW , eV i
W ∈G W ∈G
X
= ˜ W ξ,
hLU η̃, LW ξihL ˜ LV η̃i − hU η, V ηi
W ∈G
= hLU η̃, LV η̃i − hU η, V ηi
= hLU P eI , LV P eI i − hU η, V ηi
= hLU Tη η, LV Tη ηi − hU η, V ηi
= hTη U η, Tη V ηi − hU η, V ηi = 0,

this lead to the fact that P = BB ∗ .

From
BLU eV = BeU V = LU V ξ˜ = LU LV ξ˜ = LU BeV ,

we see that B ∈ M0 . Hence B is a partial isometry in M0 .

Let Q = B ∗ B, then P and Q are equivalent projections in M0 .
Since lG2 (A) is self-dual, by [49], End∗A (lG2 (A)) is a W ∗ -algebra.
Let (End∗A (lG2 (A)))∗ be its predual. One can check that M and M0 are
σ(End∗A (lG2 (A)), (End∗A (lG2 (A)))∗ )-closed in End∗A (lG2 (A)), and so both M and
M0 are W ∗ -algebras (see [52]).

65
 Claim. M and M0 are finite W ∗ -algebras.

We now define φ : M → A by

φ(A) = hAeI , eI i, ∀A ∈ M.

We want to show that φ is a faithful A-value trace for M.

Since span{LU eI , U ∈ G} = lG2 (A), for any A, B ∈ M, we have

AeI = lim An eI and BeI = lim Bn eI ,

n n

where
kn
X ln
X
(n) (n)
An eI = ai LVi (n) eI and Bn eI = bj LWj (n) eI
i=1 j=1

(n) (n)
for some ai , bj ∈ A and Vi (n) , Wj (n) ∈ G.
Then
Xlm X
kn
(m) (n)
φ(AB) = hABeI , eI i = lim limh bj ai LWj (m) LVi (n) eI , eI i.
m n
j=1 i=1

While
X lm
kn X
(n) (m)
φ(BA) = lim limh ai bj LVi (n) LWj (m) eI , eI i.
n m
i=1 j=1

Note that
hLWj (m) LVi (n) eI , eI i = hLVi (n) LWj (m) eI , eI i.

Therefore φ(AB) = φ(BA).

If A ∈ M is positive and φ(A) = 0, then
1 1
hA 2 eI , A 2 eI i = hAeI , eI i = φ(A) = 0.
1
Thus A 2 eI = 0.
Now for any U ∈ G, we have
1 1 1
A 2 eU = A 2 RU eI = RU A 2 eI = 0.

66
 1
Therefore A 2 = 0, and so A = 0. Similarly we can prove that M0 is also finite.
It follows from Corollary 2.44 that I − P and I − Q are equivalent projections
in M0 . Therefore there exists a partial isometry C ∈ M0 such that CC ∗ = I − P
and C ∗ C = I − Q.
Let T = B + C. Then T is a unitary operator in M0 , and so A = (π −1 (T ))∗
is a unitary operator in M.
˜
In order to complete the proof it remains to prove that Aη̃ = ξ.
In fact,

Aη̃ = (π −1 (T ))∗ P eI = P (π −1 (T ))∗ eI

= P π(π −1 (T ))eI = P T eI
= P (B + C)eI = P BeI + P CeI
= P ξ˜ = ξ,
˜

which completes the proof.

We now can easily have the following:

Corollary 5.5. Let G be a unitary group on a finitely or countably generated

Hilbert A-module H over a commutative unital W ∗ -algebra A such that lG2 (A)
is self-dual, then the set of all complete Parseval frame vectors for G is path-
connected.

5.2 Parseval Frame Approximations

Many interesting frames are generated by some (usually finite number of) ”win-
dow” functions under the action of a collection of unitary operators. For example,
Gabor frames and wavelet frames are of this kind.

Definition 5.6. Let U be a unitary system on a Hilbert C ∗ -module H. Φ =

(φ1 , . . . , φN ), where φj ∈ H for all j, is called a multi-frame generator of length
N for U if {U φj : U ∈ U, 1 ≤ j ≤ N } is a frame.

67
Definition 5.7. Let Φ = (φ1 , . . . , φN ) be a multi-frame generator for a unitary
system U on a Hilbert C ∗ -module H. Then a Parseval multi-frame generator
Ψ = (ψ1 , . . . , ψN ) for U is called a best Parseval multi-frame approximation for
Φ if the inequality
N
X N
X
hφk − ψk , φk − ψk i ≤ hφk − ξk , φk − ξk i
k=1 k=1

is valid for all the Parseval multi-frame generator Ξ = (ξ1 , . . . , ξN ) for U.

Let Φ = {φ1 , φ2 , . . . , φN } be a multi-frame generator for a unitary system U

on a finitely or countably generated Hilbert A-module H over a unital C ∗ -algebra
A. We use TΦ to denote the analysis operator from H to lU2 ×{1,2,...,N } (A) defined
by
N X
X
TΦ x = hx, U φj ie(U,j) , ∀x ∈ H,
j=1 U ∈U

where {e(U,j) : U ∈ U, j = 1, 2, . . . , N } is the standard orthonormal basis for

2
lU×{1,2,...,N } (A).

Note that TΦ is adjointable and its adjoint operator satisfying

TΦ∗ e(U,j) = U φj , U ∈ U, j = 1, 2, . . . , N.

Lemma 5.8. Let G be a unitary group on a Hilbert A-module H over a commuta-

tive C ∗ -algebra A. Suppose that Φ = {φ1 , φ2 , . . . , φN } and Ψ = {ψ1 , ψ2 , . . . , ψN }
be two multi-frame generators for G, then
N
X N
X
hφk , φk i = hψk , ψk i.
k=1 k=1

68
Proof. We compute
N
X N X
X N X
hφk , φk i = hφk , U ψj ihU ψj , φk i
k=1 k=1 j=1 U ∈G
N X
X N X
= hU ∗ φk , ψj ihψj , U ∗ φk i
k=1 j=1 U ∈G
N X
X N X
= hψj , U ∗ φk ihU ∗ φk , ψj i
j=1 k=1 U ∈G
N
X
= hψj , ψj i.
j=1

Theorem 5.9. Let G be a unitary group on a finitely or countably generated

Hilbert A-module H over a commutative unital C ∗ -algebra A. Suppose that Φ =
1
{φ1 , φ2 , . . . , φN } is a multi-frame generator for G. Then S 2 Φ is the unique best
Parseval multi-frame approximation for Φ, where S is the frame operator for the
multi-frame {U φj : j = 1, . . . , N, U ∈ G}.

Proof. We first show that S ∈ G 0 .

For arbitrary V ∈ G and x ∈ H we have
N X
X
SV x = hV x, U φk iU φk
k=1 U ∈G
N X
X
= hx, V ∗ U φk iU φk
k=1 U ∈G

XN X
= V( hx, V ∗ U φk iV ∗ U φk )
k=1 U ∈G

XN X
= V( hx, U φk iU φk )
k=1 U ∈G
= V Sx.

This shows that S ∈ G 0 .

69
 Since End∗A (H) is a C ∗ -algebra, by the spectral decomposition for positive
1 1
elements in C ∗ -algebra, we can infer that S − 2 , S − 4 ∈ G 0 .
1 1 1
Therefore {S − 2 φ1 , S − 2 φ2 , . . . , S − 2 φN } is a complete Parseval multi-frame
generator for G.
Let Ψ = {ψ1 , ψ2 , . . . , ψN } be any Parseval multi-frame generator for G. We
claim that
N N
1 1
X X
hTS − 12 Φ S − 4 φk , TΨ S − 4 φk i = hψk , φk i,
k=1 k=1

where TS − 12 Φ and TΨ are the analysis operators with respect to the Parseval
1
multi-frame generators S − 2 Φ and Ψ respectively.
We compute
N
1 1
X
hTS − 12 Φ S − 4 φk , TΨ S − 4 φk i
k=1
N N X N X
− 41 − 12 1
X X X
= h hS φk , U S φj ie(U,j) , hS − 4 φk , V ψi ie(V,i) i
k=1 j=1 U ∈G i=1 V ∈G
N X
N X
1 1 1
X
= hS − 4 φk , U S − 2 φj ihU ψj , S − 4 φk i
k=1 j=1 U ∈G
N X
N X
1 1 1
X
= hU ψj , S − 4 φk ihS − 4 φk , U S − 2 φj i
k=1 j=1 U ∈G
N X
N X
1 1 1 1
X
= hS 4 ψj , U ∗ S − 2 φk ihU ∗ S − 2 φk , S − 4 φj i
j=1 k=1 U ∈G
N N
1
− 14
X X
= hS ψj , S
4 φj i = hψj , φj i
j=1 j=1

1
We now want to prove that S 2 Φ is a best Parseval multi–frame approximation
for Φ. We need to show that
N N
1 1
X X
hψk − φk , ψk − φk i ≥ hS − 2 φk − φk , S − 2 φk − φk i.
k=1 k=1

70
 By Lemma 5.8, it suffices to prove that
N
1 1
X
(hS − 2 φk , φk i + hφk , S − 2 φk ) − hψk , φk i − hφk , ψk i) ≥ 0.
k=1

In fact, we have
N
1 1
X
(hS − 2 φk , φk i + hφk , S − 2 φk i − hψk , φk i − hφk , ψk i)
k=1
N
1 1 1 1
X
= (hS − 4 φk , S − 4 φk i + hS − 4 φk , S − 4 φk )i
k=1
1 1 1 1
−hTS − 21 Φ S − 4 φk , TΨ S − 4 φk i − hTΨ S − 4 φk , TS − 12 Φ S − 4 φk i)
N
1 1 1 1
X
= (hTS − 21 Φ S − 4 φk , TS − 12 Φ S − 4 φk i + hTΨ S − 4 φk , TΨ S − 4 φk i
k=1
1 1 1 1
−hTS − 12 Φ S − 4 φk , TΨ S − 4 φk i − hTΨ S − 4 φk , TS − 12 Φ S − 4 φk i)
N
1 1
X
= h(TS − 21 Φ − TΨ )S − 4 φk , (TS − 12 Φ − TΨ )S − 4 φk i ≥ 0.
k=1
1
This implies that S − 2 Φ is a best Parseval multi-frame approximation for Φ.
For the uniqueness, assume that Ξ = {ξ1 , ξ2 , . . . , ξN } be another best Parseval
multi-frame approximation for Φ. Then we have
N N
1 1
X X
hξk − φk , ξk − φk i = hS − 2 φk − φk , S − 2 φk − φk i. (5.1)
k=1 k=1

By Lemma 5.8, we also have

N N
1 1
X X
hξk , ξk i = hS − 2 φk , S − 2 φk i. (5.2)
k=1 k=1

Identities (5.1) and (5.2) imply that

N
1 1
X X
(hξk , φk i + hφk , ξk i) = (hS − 2 φk , φk i + hφk , S − 2 φk i)
k=1 k=1
N
1 1
X
= 2 hS − 4 φk , S − 4 φk i.
k=1

71
 We claim that
N N
1 1 1 1
X X
hS ξk , S ξk i =
4 4 hS − 4 φk , S − 4 φk i.
k=1 k=1

In fact,
N
1 1
X
hS 4 ξk , S 4 ξk i
k=1
N X
N X
1 1 1 1
X
= hS 4 ξk , U S − 2 φj ihU S − 2 φj , S 4 ξk i
k=1 j=1 U ∈G
N X
N X
1 1
X
= hS − 4 φj , U ∗ ξk ihU ∗ ξk , S − 4 φj i
j=1 k=1 U ∈G
N
1 1
X
= hS − 4 φj , S − 4 φj i.
j=1

Then we have
N
1 1 1 1
X
hS 4 ξk − S − 4 φk , S 4 ξk − S − 4 φk i
k=1
N
1 1 1 1
X
= (hS 4 ξk , S 4 ξk i − hS 4 ξk , S − 4 φk i
k=1
1 1 1 1
−hS − 4 φk , S 4 ξk i + hS − 4 φk , S − 4 φk i)
N
1 1
X
= (2hS − 4 φk , S − 4 φk i − hξk , φk i − hφk , ξk i)
k=1
= 0.

This implies that

1 1
S 4 ξk = S − 4 φk , k = 1, 2, . . . , N.

Therefore
1
ξk = S − 2 φk , k = 1, 2, . . . , N.
1
i.e. Ξ = S − 2 Φ, as expected.

To complete this chapter we ask

72
Problem 5.10. Does Theorem 5.9 hold when the underlying C ∗ -algebra A is
non-commutative?

73
 CHAPTER 6

DILATIONS OF FRAMES IN HILBERT C ∗ -MODULES

In this chapter we investigate the dilation of frames in Hilbert C ∗ -modules. Our

first result shows that a complete Parseval frame vector for a unitary group on
Hilbert C ∗ -module can be dilated to a complete wandering vector. For any dual
frame pair in any Hilbert C ∗ -module, we prove that the pair are orthogonal
compressions of a Riesz basis and its canonical dual basis for some larger Hilbert
C ∗ -module. In other words, the dilation theorem for Hilbert space dual frame
pairs is still valid for Hilbert C ∗ -module dual frame pairs. This dilation property
remains valid even when restricted to structured frames.

6.1 Dilation of Parseval Frame Vectors

It was proved in ([26]) that each Parseval frame of Hilbert C ∗ -modules can be
dilated to an orthonormal basis. It is natural to ask whether a complete Parse-
val frame vector for a unitary group on Hilbert C ∗ -module can be dilated to a
complete wandering vector. We answer this question in the following theorem.

Theorem 6.1. Let G be a unitary group on a finitely or countably generated

Hilbert A-module H over a unital C ∗ -algebra A. Suppose that η is a complete
Parseval frame vector for G. Then there exists a Hilbert A-module H̃ ⊇ H and
a unitary group G̃ on H̃ such that G̃ has complete wandering vectors in H̃, H is
an invariant subspace of G̃ such that G̃|H = G, and the map G 7→ G|H is a group
isomorphism from G̃ onto G.

74
Proof. Let H̃ = lG2 (A).
Now for each U ∈ G, let LU be the left regular representation defined by

LU eV = eU V , ∀V ∈ G,

where eV is the characteristic function at the single point set {V }.

Let G̃ = {LU , U ∈ G}. It is easy to check that G̃ is a unitary group isomorphic
to G.
We now define an operator T : H → H̃ by
X
T (x) = hx, U ηieU .
U ∈G

One can check that T is an adjointable isometry. Also it is routine to show

that the range of T is closed in H̃. Therefore, by Proposition 2.54 and Theorem
2.57, we see that
G̃ = (T (H))⊥ ⊕ T (H).

Hence we have the orthogonal projection P from lG2 (A) onto T (H), the range of
T.
We claim that P (eU ) = T (U η) for each U ∈ G. To see this, let V ∈ G be
arbitrary, then

hT (V η), P (eU )i = hP T (V η), eU i = hT (V η), eU i

X
= h hV η, W ηieW , eU i = hV η, U ηi
W ∈G
= hT (V η), T (U η)i

We now show that LU T = T U on H for any U ∈ G.

For each V ∈ G, we have
X X
LU T (V η) = LU ( hV η, W ηieW ) = hV η, W ηieU W
W ∈G W ∈G
X X
= hU V η, U W ηieU W = hU V η, W ηieW
W ∈G W ∈G
= T U (V η).

75
Thus LU T = T U .
Finally we want to prove that P ∈ G̃ 0 .
Indeed, for any U, V ∈ G,

P LU (eV ) = P eU V = T (U V η) = LU T (V η) = LU P (eV ),

which implies that P LU = LU P for any U ∈ G.

By identifying H with T (H), we now complete the proof.

6.2 Dilation of Dual Modular Frame Pairs

The aim of this section is to prove the dilation theorem for dual frame pairs in
Hilbert C ∗ -modules. Our approach is different from that in [13] which involves
some results that are uncertain in the Hilbert C ∗ -module setting.
It is well known that every frame in Hilbert space is a direct summand of
Riesz basis. More generally, a dual frame pair in Hilbert space can be dilated to
a Riesz basis and its dual Riesz basis (see [13]). This remains true for Hilbert
C ∗ -module frames:

Theorem 6.2. Let {xj }j∈J and {yj }j∈J be alternate dual frames for a finitely or
countably generated Hilbert A-module H over a unital C ∗ -algebra A. Then there
exist a Hilbert A-module K ⊇ H and a Riesz basis {uj }j∈J of K which has a
unique dual {u∗j }j∈J and satisfies

P uj = xj and P u∗j = yj ∀j ∈ J,

where P is the projection from K onto H.

Proof. Let TX and TY be the analysis operators of {xj }j∈J and {yj }j∈J , and PX
and PY be the orthogonal projections from l2 (A) onto the range of TX and TY
respectively.

76
 For all x ∈ H we have

hTY x, TY SY−1 yj i = hx, TY∗ TY SY−1 yj i = hx, SY SY−1 yj i

= hx, yj i = hTY x, ej i = hTY x, PY ej i. (6.1)

Therefore PY ej = TY SY−1 yj .
Observe that for all x ∈ H we have
X X
PY TX x = PY hx, xj iej = hx, xj iPY ej
j∈J j∈J
X X
= hx, xj iTY SY−1 yj = TY SY−1 hx, xj iyj
j∈J j∈J

= TY SY−1 x. (6.2)

Note that in the third equality we use (6.1).

Let
K = H ⊕ PY⊥ l2 (A) and uj = xj ⊕ PY⊥ ej .

It is easy to see that {uj }j∈J is a Bessel sequence of K. Then we have the
corresponding analysis operator TU : K → l2 (A) given by
X
TU (x ⊕ w) = hx ⊕ w, xj ⊕ PY⊥ ej iej
j∈J

for any x ∈ H and w ∈ PY⊥ l2 (A).

Note that for all x ∈ H and w ∈ PY⊥ l2 (A) we have
X
TU (x ⊕ w) = hx ⊕ w, xj ⊕ PY⊥ ej iej
j∈J
X
= (hx, xj i + hw, PY⊥ ej i)ej = TX x + w. (6.3)
j∈J

We claim that TU is a bijection.

We first prove that the range of TU is closed.
To see this, suppose φn ∈ Rang(TU ) and φn → φ as n → ∞. Then there exist
xn ∈ H and wn ∈ PY⊥ l2 (A) such that TU (xn ⊕ wn ) = φn .

77
 It follows from identity (6.3) that TX xn + wn = φn → φ as n → ∞.
Applying TY∗ on the both sides we get

TY∗ (TX xn + wn ) = TY∗ TX xn = xn → TY∗ φ,

as n → ∞.
Since the range of TX is closed, it follows that Rang(TU ) is also closed.
To show that TU is onto, by Theorem 2.57, it is equivalent to show that TU∗
is one-to-one.
Suppose that TU∗ aj ej = 0 for some {aj }j∈J ∈ l2 (A). Then we have
P
j∈J

X X
0 = TU∗ aj ej = aj (xj ⊕ PY⊥ ej )
j∈J j∈J
X X X X
= aj x j ⊕ aj PY⊥ ej = aj xj ⊕ PY⊥ aj ej . (6.4)
j∈J j∈J j∈J j∈J

Therefore PY⊥
P P
j∈J aj ej = 0, and so aj ej ∈ Rang(PY ) = Rang(TY ). Then
j∈J
P
there exists an element z ∈ H such that TY z = j∈J aj ej .
P
From TY z = j∈J hz, yj iej , we have aj = hz, yj i for all j.
P
Identity (6.4) also implies j∈J aj xj = 0. Therefore
X X
0= aj xj = hz, yj ixj = z,
j∈J j∈J

which yields that aj = 0 for all j.

We now prove that TU is injective.
Suppose that TU (x ⊕ w) = 0 for some x ∈ H and w ∈ PY⊥ l2 (A).
By (6.3) we have TX x + w = 0, and so TX x = −w.
Applying PY on the both sides we arrive at PY TX x = PY (−w) = 0.
By (6.2) we can see that 0 = PY TX x = TY SY−1 x. Hence x = 0 as both TY and
SY−1 are injective.
By Theorem 2.57, we can infer that TU∗ is also a bijection.

78
 Now let SU = TU∗ TU .
Then SU−1 is adjointable and hence bounded. Thus SU is an invertible bounded
A-linear operator, then, by Lemma 3.6, {uj }j∈J is a frame for K.
Let {u∗j }j∈J be the canonical dual frame of {uj }j∈J and write u∗j = zj ⊕ wj .
For any x ∈ H, we have
X X
hx, zj ihzj , xi = hx ⊕ 0, zj ⊕ wj ihzj ⊕ wj , x ⊕ 0i
j∈J j∈J
≤ Dhx ⊕ 0, x ⊕ 0i = Dhx, xi,

where D is the upper bound of {u∗j }j∈J . Therefore {zj }j∈J is a Bessel sequence
of H. We denote the corresponding analysis operator by TZ .
P
We claim that x = j∈J hx, xj izj for all x ∈ H.
In fact, for every x ∈ H, we get
X
x⊕0 = hx ⊕ 0, xj ⊕ PY⊥ ej izj ⊕ wj
j∈J
X X
= hx, xj izj ⊕ hx, xj iwj ,
j∈J j∈J
P
which implies that x = j∈J hx, xj izj .

Now for any x ∈ H,

X
x⊕0 = hx ⊕ 0, zj ⊕ wj ixj ⊕ PY⊥ ej
j∈J
X X
= hx, zj ixj ⊕ hx, zj iPY⊥ ej ,
j∈J j∈J

so PY⊥ TZ x = 0 for all x ∈ H. This yields that Rang(TZ ) ⊆ Rang(TY ).

By Lemma 4.8 we can infer that zj = yj for all j.
Furthermore the analysis operator of {u∗j }j∈J is onto. Then, again by Lemma
4.8, we get the uniqueness of the dual of {uj }j∈J .
To complete the proof, it remains to prove that both {uj }j∈J and {u∗j }j∈J are
Riesz bases of K.

79
 Note that we have already proved that TU is onto. Then by Theorem 4.9,
we see that {uj }j∈J is a Riesz basis of K, and so {u∗j }j∈J is also a Riesz basis as
u∗j = SU−1 uj and SU is invertible.

We end this section by pointing out that the dilation theorem still holds when
restricted to structured frames (i.e., frames induced by unitary representations
of groups). Recall that two vectors φ, ψ ∈ H are called dual frame vectors (resp.
dual Riesz basis vectors) for a unitary group U on H if Uφ and Uψ are dual
frames (resp. dual Riesz bases) of H.
The following theorem shows that if two frames generated by unitary groups
are dual frames, they must be generated by the same unitary group.

Theorem 6.3. Let H be a Hilbert A-module over a unital C ∗ -algebra A, and

ξ, η ∈ H be complete frame vectors for unitary groups G1 , G2 on H respectively.
Suppose that π : G1 → G2 is a group isomorphism. If G1 ξ and G2 η are dual frames,
then π(U ) = U for each U ∈ G1 .

Proof. For arbitrary U ∈ G1 and x ∈ H, we have

X X
Ux = hU x, V ξiπ(V )η = hx, U −1 V ξiπ(V )η
V ∈G1 V ∈G1
X
−1
= π(U ) hx, U V ξiπ(U −1 V )η = π(U )x,
V ∈G1

which implies that π(U ) = U for all U ∈ G1 .

To complete this section we consider the dilation property of structured dual

frame pairs in Hilbert C ∗ -modules.

Theorem 6.4. Suppose that G is a unitary group on a finitely or countably

generated Hilbert A-module H over a unital C ∗ -algebra A. Let ξ and η be two
complete dual frame vectors for G. Then there exists a Hilbert A-module K ⊇ H

80
and a unitary group G̃, and a complete Riesz vector ξ˜ with a unique dual vector
η̃ for G̃ such that
P Ũ ξ˜ = U ξ and P Ũ η̃ = U η,

where P is the projection from K onto H.

Proof. Let Tξ , Tη be the analysis operator of Gξ, Gη, and Pξ , Pη be the orthogonal
projections from lG2 (A) onto the range of Tξ , Tη respectively.
Let K = H ⊕ PY⊥ lG2 (A) and Ũ = U ⊕ LU for each U ∈ G.
One can easily verify that G̃ = {Ũ : U ∈ G} is a group of unitary operators
on K.
Let ξ˜ = ξ ⊕ Pη⊥ eI , then Ũ ξ˜ = U ξ ⊕ LU Pη⊥ eI = U ξ ⊕ Pη⊥ eU .
Then, by Theorem 6.2, ξ˜ is a complete frame vector for G̃.
˜ and η̃ = S −1 ξ,
Let S be the frame operator of G̃ ξ, ˜ as desired.

6.3 Projective Frames

In this section we will discuss the characterization of projective frames. We just

list some basic observations on this topic. We will continue this work in future.

Definition 6.5. Suppose that {xn }∞

n=1 is a sequence of Banach space X. Then

{xn }∞
n=1 is called a projective frame of X if it is the projective image (i.e. apply

a bounded projection) of a (bounded unconditional) basis for a larger Banach

space.

In [13], it was proved that a sequence {xn }∞

n=1 of a Banach space X is a

projective frame if and only if there exists a sequence {yn }∞ ∗

n=1 ⊆ X such that

∞
X
x= hx, yn ixn (6.5)
n=1

holds true for all x ∈ X.

81
 We want to find a intrinsic characterization of projective frames in Banach
spaces, which does not assume the additional hypothesis of the associated dual.
In other words, we are interested in finding a characterization of a sequence
{xn }∞ ∞
n=1 in Banach space X that admits a generalized dual {yn }n=1 in the sense

of equation (6.5)
We first look at the Hilbert space case.

Theorem 6.6. In finite dimensional Hilbert spaces every generating sequence is

a projective frame.

Proof. Let H be a Hilbert space and {xn }∞

n=1 a generating sequence of H.

We can find an integer N > 0 such that {x1 , x2 , . . . , xN } generates H. Since

the dimension of H is finite, it follows that {x1 , x2 , . . . , xN } is a frame of H.
Let {x∗1 , x∗2 , . . . , x∗N } be any dual frame of {x1 , x2 , . . . , xN }.
Now let (
x∗n if 1 ≤ n ≤ N ;
yn =
0 if n > N,

One can easily check that x = ∞

P
n=1 hx, yn ixn holds true for all x ∈ H.

The following example shows that Theorem 6.6 is no longer true in general
for infinite dimensional Hilbert space.

Example 6.7. Suppose that H is a infinite-dimensional Hilbert space with an

orthonormal basis {e1 , e2 , . . . }.
For each n, let xn = ni=1 ei .
P

Then {xn }∞
n=1 is a generating sequence of H, but it is not a projective se-

quence.
Indeed, assume on the contrary that {xn }∞
n=1 is a projective frame.Then there

exist a sequence {yn }∞

n=1 ⊆ H such that for each x ∈ H we have

∞
X
x= hx, yn ixn .
n−1

82
 We first consider the case that x = e1 . We have
∞
X
e1 = he1 , yn i(e1 + e2 + · · · + en ).
n=1

It follows that
∞
X
1 = he1 , yn i
n=1
∞
X
0 = he1 , yn i
n=2
∞
X
0 = he1 , yn i
n=3
.. .. ..
. . .

which yields that he1 , y1 i = 1 and he1 , yn i = 0 for n ≥ 2.

Therefore y1 = e1 .
For the case of x = e2 , we have
∞
X
e2 = he2 , yn i(e1 + e2 + · · · + en ),
n=1

which implies that

∞
X
0 = he2 , yn i
n=1
X∞
1 = he2 , yn i
n=2
X∞
0 = he2 , yn i
n=3
.. .. ..
. . .

It follows that he2 , y1 i = −1, he2 , y2 i = 1 and he2 , yn i = 0 for all n ≥ 3.

But we already know that y1 = e1 and so he2 , y1 i = he2 , e1 i = 0, a contradic-
tion.

83
 In future we will focus on finding an intrinsic characterization of projective
frames in infinite dimensional Hilbert spaces as well as in Banach spaces, even
more generally, in Hilbert C ∗ -modules.

84
 CHAPTER 7

PALEY-WIENER TYPE PERTURBATION OF

FRAMES AND RIESZ BASES IN HILBERT
C ∗ -MODULES

In this chapter we shall see that in a finitely or countably generated Hilbert C ∗ -

module H if a sequence {yj }j∈J in H is ”sufficiently near” to a given frame in H,
then {yj }j∈J is also a frame of H. Hence it will follows that various properties of
frames {xj }j∈J in Hilbert C ∗ -module H are ”stable” in the sense that they are
conserved by every sequence {yj }j∈J ”sufficiently near” to the frame {xj }j∈J . Our
first result in this chapter extends the Casazza-Christensen’s perturbation theo-
rem of frames to Hilbert C ∗ -modules. But for the case of Riesz bases in Hilbert
C ∗ -modules, the stability of Riesz bases is quite different from that of frames.
We will give a complete characterization on all the Riesz bases for Hilbert C ∗ -
modules such that the perturbation (under Casazza-Christensen’s perturbation
condition) of a Riesz basis still remains a Riesz basis.

7.1 Perturbation of Modular Frames

In this section we concentrate on the perturbation of frames in Hilbert C ∗ -

modules. We first give a ”necessary and sufficient” perturbation theorem of
frames in any Hilbert C ∗ -module.

85
Theorem 7.1. Suppose that H is a Hilbert C ∗ -module. Let {xj }j∈J be a frame
for H with frame bounds CX and DX and {yj }j∈J be a sequence of H. Then the
following statements are equivalent:
(1) {yj }j∈J is a frame of H;
(2) There is a constant M > 0 so that for all x ∈ H we have
X X
k hx, xj − yj ihxj − yj , xik ≤ M k hx, xj ihxj , xik
j∈J j∈J

and
X X
k hx, xj − yj ihxj − yj , xik ≤ M k hx, yj ihyj , xik.
j∈J j∈J

Moreover, if {yj }j∈J is a Bessel sequence, then (1) and (2) are equivalent to
(3) There exists a constant M > 0 so that
X X
k hx, xj − yj ihxj − yj , xik ≤ M k hx, yj ihyj , xik
j∈J j∈J

holds for all x ∈ H.

Proof. (1)⇒(2). Let CY and DY be the frame bounds of {yj }j∈J .

For every x ∈ H we have
P
k j∈J hx, xj − yj ihxj − yj , xik
P
= k j∈J (hx, xj i
− hx, yj i)(hxj , xi − hyj , xi)k
P P
= k j∈J hx, xj ihxj , xi + j∈J hx, yj ihyj , xi
P P
− j∈J hx, xj ihyj , xi − j∈J hx, yj ihxj , xik
P P
≤ k2( j∈J hx, xj ihxj , xi + j∈J hx, yj ihyj , xi)k
P
≤ 2(k j∈J hx, xj ihxj , xik + DY khx, xik)
2(k j∈J hx, xj ihxj , xik + D
P P
≤ CX
Y
k j∈J hx, xj ihxj , xik)
2(1 + D
P
= )k j∈J hx, xj ihxj , xik.
Y
CX

Similarly, we have
X DX X
k hx, xj − yj ihxj − yj , xik ≤ 2(1 + )k hx, xj ihxj , xik.
j∈J
CY j∈J

86
 DY DX
Therefore we can choose M = max{2(1 + CX
), 2(1 + CY
)}.
(2)⇒(1). Given M in (2) and any x ∈ H we have

P
CX khx, xik ≤ k j∈J hx, xj ihxj , xik
P
= k j∈J (hx, xj − yj i + hx, yj i)(hxj − yj , xi + hyj , xi)k
P P
≤ 2( j∈J hx, xj − yj ihxj − yj , xi + j∈J hx, yj ihyj , xi)
P
≤ 2(M + 1)k j∈J hx, yj ihyj , xik
P
= 2(M + 1)k j∈J (hx, yj − xj i + hx, xj i)(hyj − xj , xi + hx, xj i)k
P P
≤ 4(M + 1)k j∈J hx, yj − xj ihyj − xj , xi + j∈J hx, xj ihxj , xik
P P
= 4(M + 1)k j∈J hx, xj − yj ihxj − yj , xi + j∈J hx, xj ihxj , xik
4(M + 1)2 k j∈J hx, xj ihxj , xik
P
≤
≤ 4(M + 1)2 DX khx, xik.

Therefore
CX X
khx, xik ≤ k hx, yj ihyj , xik ≤ 2(M + 1)DX khx, xik.
2(M + 1) j∈J

The moreover part follows from the proof of (2)⇒(1).

Before we prove the first main result of this section, we need the following
result which is due to Casazza and Christensen ([11]). It is a generalization of the
famous Neumann Theorem which states that an operator U on a Banach space
is invertible if kI − U k < 1.

Lemma 7.2. ([11]) Let X be a Banach space, and U : X → X a linear operator.

Assume that there exist constants λ1 , λ2 ∈ (0, 1) such that

kU x − xk ≤ λ1 kxk + λ2 kU xk, ∀x ∈ X.

Then U is bounded and invertible with

1 + λ1 1 + λ2
kU k ≤ and kU −1 k ≤ .
1 − λ2 1 − λ1

We are now in a position to prove the following theorem.

87
Theorem 7.3. Suppose that H is a finitely or countably generated Hilbert A-
module H over a unital C ∗ -algebra A. Let {xj }j∈J be a frame for H with frame
bounds C and D. Suppose that {yj }j∈J is a sequence of H and there exist
λ1 , λ2 , µ ≥ 0 such that max{λ1 + √µ , λ2 } < 1. Then {yj }j∈J is also a frame
C
for H with bounds
√ √
(1 − λ1 ) C − µ 2 (1 + λ1 ) D + µ 2
( ) and ( ),
1 + λ2 1 − λ2
if one of the following conditions is fulfilled for any finite sequence {cj }nj=1 ⊆ A
and all x ∈ H:
1
X
k hx, xj − yj ihxj − yj , xik 2 (7.1)
j∈J
1 1
X X
≤ λ1 k hx, xj ihxj , xik 2 + λ2 k hx, yj ihyj , xik 2 + µkxk;
j∈J j∈J

or
n n n n
1
X X X X
k cj (xj − yj )k ≤ λ1 k c j x j k + λ2 k cj yj k + µk cj c∗j k 2 . (7.2)
j=1 j=1 j=1 j=1

Proof. Let TX and SX denote the analysis operator and frame operator of {xj }j∈J
respectively.
Assume first that condition (7.1) holds true for all x ∈ H.
We now define an operator TY : H → l2 (A) by
X
TY x = hx, yj iej .
j∈J

Then condition (7.1) turns to be

kTX x − TY xk ≤ λ1 kTX xk + λ2 kTY xk + µkxk.

On one hand we have

(1 − λ2 )kTY xk ≤ (1 + λ1 )kTX xk + µkxk,

88
which implies that
√
1 (1 + λ1 ) D + µ
kTY xk ≤ [(1 + λ1 )kTX xk + µkxk] ≤ kxk.
1 − λ2 1 − λ2
√
Therefore {yj }j∈J is a Bessel sequence with Bessel bound ( (1+λ1−λ
1 ) D+µ 2
2
) .
On the other hand we have

(1 − λ1 )kTX xk − µkxk ≤ (1 + λ2 )kTY xk.

Therefore
√
1 (1 − λ1 ) C − µ
kTY xk ≥ [(1 − λ1 )kTX xk − µkxk] ≥ kxk,
1 + λ2 1 + λ2
which implies that {yj }j∈J is a frame.
Suppose now that condition (7.2) holds. Then for each {cj }j∈J ∈ l2 (A) we
have n n n
X 1 X X 1
k cj yj k ≤ [(1 + λ1 )k cj xj k + µk cj c∗j k 2 ],
j=1
1 − λ2 j=1 j=1

which yields that

n ∞ ∞
X 1 X X 1
k cj yj k ≤ [(1 + λ1 )k cj xj k + µk cj c∗j k 2 ].
j=1
1 − λ2 j=1 j=1

Furthermore, we obtain
∞ ∞ ∞
X 1 X X 1
k cj yj k ≤ [(1 + λ1 )k cj xj k + µk cj c∗j k 2 ].
j=1
1 − λ2 j=1 j=1

Therefore we can define a bounded operator U : H → l2 (A) by

X
U {cj } = cj yj ,
j∈J

which satisfying
√
1 ∗ (1 + λ1 ) D + µ
kU {cj }k ≤ [(1 + λ1 )kTX {cj }k + µk{cj }k] ≤ k{cj }k.
1 − λ2 1 − λ2
√
By Proposition 3.11, {yj }j∈J is a Bessel sequence with bound ( (1+λ1−λ
1 ) D+µ 2
2
).

89
 Note that for each {cj }j∈J ∈ l2 (A) we also have
1
X X X X
k cj (xj − yj )k ≤ λ1 k c j x j k + λ2 k cj yj k + µk cj c∗j k 2 .
j∈J j∈J j∈J j∈J

−1
Then for each x ∈ H, letting {cj } = TX SX x, we get

−1 −1 −1
kx − U TX SX xk ≤ λ1 kxk + λ2 kU TX SX xk + µkTX SX xk
µ −1
≤ λ1 kxk + √ kxk + λ2 kU TX SX xk.
C
−1
By Lemma 7.2, U TX SX is invertible with

1 + λ1 + √µ
−1 C
kU TX SX k ≤
1 − λ2
and
−1 −1 1 + λ2
k(U TX SX ) k≤ .
1 − (λ1 + √µC )

Now for arbitrary x ∈ H, we have

X
−1 −1 −1 −1 −1 −1
x = U TX SX (U TX SX ) x= h(U TX SX ) x, SX xj iyj .
j∈J

It follows that

kxk4
= khx, xik2
X
−1 −1 −1
= k h(U TX SX ) x, SX xj ihyj , xik2
j∈J
X X
−1 −1 −1 −1 −1 −1
≤ k h(U TX SX ) x, SX xj ihSX xj , (U TX SX ) xik · k hx, yj ihyj , xik
j∈J j∈J
1 −1 −1 −1 −1
X
≤ kh(U TX SX ) x, (U TX SX ) xik · k hx, yj ihyj , xik
C j∈J
1 1 + λ2 2 2
X
≤ ( ) kxk · k hx, yj ihyj , xik.
C 1 − (λ1 + √µC ) j∈J

−1
Note that in the second inequality we apply the fact that {SX xj }j∈J is a frame
1 1
with frame bounds D
and C
.

90
 It follows that
√
(1 − λ1 ) C − µ 2 X
( ) kxk2 ≤ k hx, yj ihyj , xik.
1 + λ2 j∈J

7.2 Perturbation of Modular Riesz Bases

For the extension of the second part of Theorem 1.1, we first point out that if
µ = 0 in the condition (7.2) of Theorem 7.3, then {yj }j∈J is a Riesz basis provided
that {xj }j∈J is a Riesz basis.

Theorem 7.4. Let H be a finitely or countably generated Hilbert A-module H

over a unital C ∗ -algebra A and {xj }j∈J a Riesz basis for H. Suppose that {yj }j∈J
is a sequence of H and there exist λ1 , λ2 ∈ [0, 1). If
X X X
k cj (xj − yj )k ≤ λ1 k c j x j k + λ2 k cj yj k (7.3)
j∈J j∈J j∈J

holds for all finite sequence {cj }nj=1 ⊆ A, then {yj }j∈J is also a Riesz basis.

Proof. We first claim that yj 6= 0 for each j.

Assume to the contrary that there exists j0 such that yj0 = 0. Choose {cj } =
ej0 , then we have
kxj0 k ≤ λ1 kxj0 k,

which implies that xj0 = 0, a contradiction.

By Theorem 7.3, we see that {yj }j∈J is also a frame of H. We denote the
analysis operators of {xj }j∈J and {yj }j∈J by TX and TY respectively.
We now claim that Rang(TX ) = Rang(TY ).
If {cj } ∈ KerTX∗ , then we have

kTY∗ {cj }k ≤ λ2 kTY∗ {cj }k,

91
which leads to {cj } ∈ KerTY∗ .
In the same manner we can show that KerTY∗ ⊆ KerTX∗ , and so KerTX∗ =
KerTY∗ .
It follows from Proposition 3.12 that both Rang(TX∗ ) and Rang(TY∗ ) are closed,
and hence both Rang(TX ) and Rang(TY ) are closed. Now applying Theorem
15.3.8 in [55] we see that Rang(TX ) = Rang(TY ).
Then by Theorem 4.1, we can infer that {yj }j∈J is also a Riesz basis of H.

As we have seen from Lemma 4.1 that the structure of Hilbert C ∗ -module
Riesz bases is much more complicated than the Hilbert space Riesz bases. There-
fore there is no surprise that the perturbation of Riesz bases in Hilbert C ∗ -
modules could be quite different from that in Hilbert spaces. The following
example shows that the second part of Theorem 1.1 is no longer true in general
for Hilbert C ∗ -module Riesz bases.

Example 7.5. Let l∞ be the set of all bounded complex-valued sequences. For
any u = {uj }j∈N and v = {vj }j∈N in l∞ , we define

uv = {uj vj }j∈N , u∗ = {ūj }j∈N and kuk = max |uj |.

j∈N

Then A = {l∞ , k · k} is a C ∗ -algebra.

Let H = c0 be the set of all sequences converging to zero. For any u, v ∈ H
we define

hu, vi = uv ∗ = {uj v̄j }j∈N .

Then H is a Hilbert A-module.

For each j, let xj = ej . Obviously, {xj }j∈N is a Parseval Riesz basis of H.
Now let (
e1 + e2 if j = 1, 2,
yj =
ej if j 6= 1, 2,
and λ1 = 81 , λ2 = 15
16
and µ = 34 .

92
 Then one can check that condition (7.2) in Theorem 7.3 is satisfied. But
{yj }j∈J is not a Riesz basis.

Our second main result is to give a necessary and sufficient condition under
which every perturbation {yj }j∈J of a Riesz basis {xj }j∈J is also a Riesz basis in
Hilbert C ∗ -modules.

Theorem 7.6. Suppose that {xj }j∈J is a Riesz basis of H with frame bounds
C and D, where H is a finitely or countably generated Hilbert A-module over a
unital C ∗ -algebra A. Assume that there exist λ1 , λ2 ≥ 0 and µ > 0 such that
µ
max{λ1 + √ , λ2 } < 1.
C
Then the following are equivalent:
(i) Every sequence {yj }j∈J in H satisfying the following perturbation condition
is again a Riesz basis:

n n n n
1
X X X X
k cj (xj − yj )k ≤ λ1 k c j x j k + λ2 k cj yj k + µk cj c∗j k 2 (7.4)
j=1 j=1 j=1 j=1

for any c1 , c2 , . . . , cn ∈ A.
(ii) KerTX∗ = l2 (B), where TX is the analysis operator of {xj }j∈J and B =
{a ∈ A : aH = {0}}.
In case the equivalent conditions are satisfied, we also have KerTY∗ = KerTX∗
and Rang(TY ) = Rang(TX ), where TY is the analysis operator of {yj }j∈J .

Proof. From Theorem 7.3 and its proof we can infer that {yj }j∈J is a frame and
satisfies
1
X X X X
k cj (xj − yj )k ≤ λ1 k c j x j k + λ2 k cj yj k + µk cj c∗j k 2
j∈J j∈J j∈J j∈J

for all {cj } ∈ l2 (A).

”(i) ⇒ (ii)”. Suppose first that any sequence {yj }j∈J satisfying condition
(7.4) is a Riesz basis. We now show that KerTX∗ = l2 (B).

93
 Obviously, l2 (B) ⊆ KerTX∗ .
Now pick an arbitrary {aj }j∈J ∈ KerTX∗ . We need to prove that aj H = {0}
for each j.
Assume on the contrary that there exists j0 ∈ J such that aj0 H 6= {0}. We
have two cases:

Case 1 There exists j1 ∈ J such that aj0 xj1 6= 0.

kxj1 k
Choose M > 0 such that M
≤ µ.
Consider sequence {zj }j∈J given by
(
xj0 − M1 xj1 , if j = j0 ;
zj =
xj , otherwise.

One can check that {zj }j∈J satisfies condition (7.4).

Now let {cj } be a sequence such that

 M aj0 , if j = j0 ;


cj = aj0 , if j = j1 ;


 a, otherwise.
j

Observe that
X X
cj zj = aj xj = 0.
j∈J j∈J

But
cj0 zj0 = −aj0 xj1 6= 0.

Thus {zj }j∈J is not a Riesz basis, a contradiction.

Case 2 aj0 xj = 0 for all j ∈ J.

√
2
We pick z ∈ H such that aj0 z 6= 0, and N > 0 such that N
kzk ≤ µ.
Consider a sequence {zj }j∈J defined by

1
 x1 + N z, if j = 1;


zj = 1
x2 − N
z, if j = 2;


 x, otherwise.
j

94
 Note that {zj }j∈J also satisfies condition (7.4).
Letting cj = aj0 for all j, we have
X X
cj zj = aj0 xj = 0.
j∈J j∈J

But
aj0
c1 z1 = −c2 z2 = z 6= 0,
N
which contradicts the fact that {zj }j∈J is a Riesz basis.

”(ii) ⇒ (i)”. Suppose now that KerTX∗ = l2 (B) and {yj }j∈J is an arbitrary
sequence satisfying condition (7.4).
By Corollary 4.3, we consider any sequence {aj } ∈ l2 (A) such that
P
j∈J aj yj =
0.
We claim that {aj } ∈ l2 (B).
/ l2 (B). By Theorem 2.57 we have
Assume on the contrary that {aj } ∈

l2 (A) = KerTX∗ ⊕ (KerTX∗ )⊥ = l2 (B) ⊕ (l2 (B))⊥ .

Thus {aj } has a unique decomposition

(1) (2)
{aj } = {aj } ⊕ {aj },

(1) (2)
where {aj } ∈ l2 (B) and {aj } is a nonzero sequence in (l2 (B))⊥ .

95
 It follows that
X
k aj y j k
j∈J
X (1) (2)
X (2)
= k (aj + aj )yj k = k aj y j k
j∈J j∈J
X (2)
X (2)
= k aj x j − aj (xj − yj )k
j∈J j∈J
X (2)
X (2)
≥ k aj x j k −k aj (xj − yj )k
j∈J j∈J
X (2)
X (2)
X (2) (2)
≥ k aj x j k − λ1 k aj x j k − λ 2 k aj yj k − µk{aj }k
j∈J j∈J j∈J
X (2)
X (2) (2)
= (1 − λ1 )k aj x j k − λ 2 k aj yj k − µk{aj }k
j∈J j∈J
√ (2)
X (2) (2)
≥ [(1 − λ1 ) C]k{aj }k − λ2 k aj yj k − µk{aj }k
j∈J
√ (2)
X (2)
= [(1 − λ1 ) C − µ]k{aj }k − λ2 k aj yj k.
j∈J

Note that in the last inequality we apply Proposition 3.13.

It follows that
√
X (2) (1 − λ1 ) C − µ (2)
0=k aj y j k ≥ k{aj }k,
j∈J
1 + λ2

(2)
and hence aj = 0 for each j, a contradiction.
Thus we can infer that KerTY∗ = l2 (B).
To show that {yj }j∈J is a Riesz basis, it remains to show that yj 6= 0 for each
j.
Assume on the contrary that yj0 = 0 for some jo ∈ J. For any a ∈ A, let
(
a, if j = j0 ;
cj =
0, otherwise.

= 0, i.e. {cj }j∈J ∈ KerTY∗ .

P
Then j∈J cj yj

96
 Since KerTX∗ = KerTY∗ , we see that axj0 = 0 for any a ∈ A. Therefore xj0 = 0
which leads to a contradiction with the assumption that {xj }j∈J is a Riesz basis.
This completes the proof.

Remark 7.7. Case 2 in the above proof states that there exists an element a ∈ A
such that axj = 0 for all j but aH 6= {0}, where {xj }j∈J is a Riesz basis of a
Hilbert A-module H. Though this never occurs in Hilbert spaces, it could happen
in Hilbert C ∗ -modules. For example, let’s consider the C ∗ -algebra A = M2×2 (C)
of all 2 × 2 complex matrices.
Let H = A and for any x, y ∈ H define

hx, yi = xy ∗ .

Then H is a Hilbert A-module.

Choose ! !
1 0 0 1
x1 = and x2 = .
−1 0 0 −1
One can check that {x1 , x2 } is a Riesz basis of H.
Pick !
1 1
a= .
1 1

Then we have !
0 0
ax1 = ax2 = .
0 0
But, it is obvious that !
0 0
aH 6= { }.
0 0

97
 List of References

[1] J. Aiken, J. Erdos and J. Goldstein, Unitary approximation of positive oper-

ators, Illinois J. Math., 61 (1980), 61–72.
[2] J. Aiken, J. Erdos and J. Goldstein, On Löwidin orthogonalization, Internat.
J. Quantum Chem., 18 (1980), 1101–1108.
[3] A. Aldroubi, Portraits of frames, Proc. Amer. Math. Soc., 123 (1995), 1661–
1668.
[4] D. Bakić, B. Guljaš, Wigner’s theorem in Hilbert C ∗ -modules over C ∗ -
algebras of compact operators, Proc. Amer. Math. Soc., 130 (2002), 2343–
2349.
[5] R. Balan, Stability theorems for Fourier frames and wavelet Riesz bases, J.
Fourier Anal. Appl., 3 (1997), 499–504.
[6] R. Balan, Z. Landau, Hilbert frames redundancy and measure functions. in
preparation.
[7] J. Benedetto, M. Fickus, Finite normalized tight frames, Adv. Comput.
Math., 18 (2003), 357–385.
[8] D. Blecher, A new approach to Hilbert C ∗ -modules, Math. Ann., 307 (1997),
253–290.
[9] C. Boor, R. DeVore and A. Ron, Approximation from shift-invariant sub-
spaces of L2 (Rd ), Trans. Amer. Math. Soc., 341 (1994), 787–806.
[10] P. Casazza, Every frame is a sum of three (but not two) orthonormal bases
and other frame representations, J. Fourier Anal. Appl., 4 (1999), 727–732.
[11] P. Casazza, O. Christensen, Perturbation of operators and applications to
frame theory, J. Fourier Anal. Appl., 3 (1997), 543–557.
[12] P. Casazza, O. Christensen, A. Lindner and R. Vershynin, Frames and the
Feichtinger conjecture, preprint.
[13] P. Casazza, D. Han and D. Larson, Frames in Banach spaces, Contemp.
Math., 247 (1999), 149–181.
[14] P. Casazza, N. Kalton, Generalizing the Paley-Wiener perturbation theory
for Banach spaces, Proc. Amer. Math. Soc., 127 (1999), 519–527.

98
[15] O. Christensen, Frame perturbations, Proc. Amer. Math. Soc., 123 (1995),
1217–1220.
[16] O. Christensen, A Paley-Wiener theorem for frames, Proc. Amer. Math.
Soc., 123 (1995), 2199–2201.
[17] O. Christensen, An Introduction to Frames and Riesz Bases, Birkhäuser,
Boston-Basel-Berlin, 2002.
[18] O. Christensen, C. Heil, Perturbations of Banach frames and atomic decom-
positions, Math. Nachr., 185 (1997), 33–47.
[19] X. Dai, D. Larson, Wandering vectors for unitary systems and orthogonal
wavelets, Memoirs Amer. Math. Soc., 134 (1998), No. 640.
[20] I. Daubechies, A. Grossmann and Y. Meyer, Painless nonorthogonal expan-
sions, J. Math. Phys., 27 (1986), 1271–1283.
[21] Y. Denizeau, J.-F. Havet, Correspondences d’indice fini. I: Indice d’un
vecteur, J. Operator Theory, 32 (1994), 111–156.
[22] R. Duffin, A. Schaeffer, A class of nonhamonic Fourier series, Trans. Amer.
Math. Soc., 72 (1952), 341–366.
[23] M. Frank, D. Larson, Frames in Hilbert C ∗ -modules and C ∗ -algebras,
preprint, University of Huston, Houston, and Texas A&M University, College
Station, Texas, 1998.
[24] M. Frank, D. Larson, A module frame concept for Hilbert C ∗ -modules,
in Functional and Harmonic Analysis of Wavelets (San Antonio, TX, Jan.
1999), Contemp. Math., 247 (1999), 207–233.
[25] M. Frank, D. Larson, Modular frames for Hilbert C ∗ -modules and symmetric
approximation of frames, Proc. SPIE, 4119 (2000), 325–336.
[26] M. Frank, D. Larson, Frames in Hilbert C ∗ -modules and C ∗ -algebras, J.
Operator Theory, 48 (2002), 273–314.
[27] M. Frank, V. Paulsen and T. Tiballi, Symmetric approximation of frames
and bases in Hilbert spaces, Trans. Amer. Math. Soc., 354 (2002), 777–793.
[28] J. Gabardo, D, Han, Frame representations for group-like unitary operator
systems, J. Operator Theory, 49 (2003), 223–244.
[29] J. Gabardo, D. Han and D. Larson, Gabor frames and operator algebras,
Wavelet Applications in Signal and Image Processing, Proc. SPIE., 4119
(2000), 337–345.
[30] J. Goldstein, M. Levy, Linear algebra and quantum chemistry, Amer. Math.
Monthly, 98 (1991), 710-715.
[31] D. Han, Wandering vectors for irrational rotation unitary systems, Trans.
Amer. Math. Soc., 355 (1998), 309–320.

99
[32] D. Han, Approximation for Gabor and wavelet frames, Trans. Amer. Math.
Soc., 355 (2004), 3329–3342.
[33] D. Han, Tight frame approximation for multi-frames and super-frames, J.
Approx. Theory, 129 (2004), 78–93.
[34] D. Han, W. Jing and D. Larson, Duals of Riesz bases and dilation of dual
frame pairs in Hilbert C ∗ -modules, preprint (2006).
[35] D. Han, W. Jing and R. Mohapatra, Structured Parseval frames for Hilbert
C ∗ -modules, Contemp. Math., (2006), (to appear).
[36] D. Han, W. Jing and R. Mohapatra, Perturbation of frames and Riesz bases
in Hilbert C ∗ -modules, preprint (2006).
[37] D. Han, D. Larson, Frames, bases and group representations, Memoirs Amer.
Math. Soc., 147 (2000), No. 697.
[38] H. Heuser, Functional Analysis, John Wiley, New York, 1982.
[39] J. Holub, Pre-frame operators, Besselian frames, and near-Riesz bases in
Hilbert spaces, Proc. Amer. Math. Soc., 122 (1994), 779–785.
[40] K. Jensen, K. Thomsen, Elements of KK-Theory, Math. Theory Appl.,
Birkhäuser, Boston-Basel-Berlin, 1991.
[41] R. Kadison, I. Singer, Extensions of pure states, Amer. J. Math. , 81 (1959),
547–564.
[42] I. Kaplansky, Algebras of type I, Ann. Math., 56 (1952), 460-472.
[43] G. Kasparov, Hilbert C ∗ -modules: The theorem of Stinespring and
Voiculescu, J. Operator Theory, 4 (1980), 133–150.
[44] A. Khosravi, A. Moslemipour, Modular standard frames in Hilbert A-
modules, Int. Math. J., 3 (2003), 1139–1147.
[45] E. Lance, Unitary operators on Hilbert C ∗ -modules, Bull. London Math.
Soc., 26 (1994), 363–366.
[46] E. Lance, Hilbert C ∗ -modules - a Toolkit for Operator Algebraists, Lon-
don Mathematical Society Lecture Note Series v. 210, Cambridge University
Press, Cambridge, England, 1995.
[47] P.-O. Löwidin, On the nonorthogonality problem, Adv. Quantum Chem., 5
(1970), 185–199.
[48] R. Paley, N.Wiener, Fourier Transforms in Complex Domains. MAS Collo-
quium Publications 19, 1934.
[49] W. Paschke, Inner product modules over B ∗ -algebras, Trans. Amer. Math.
Soc., 182 (1973), 443–468.

100
[50] I. Raeburn, S. Thompson, Countably generated Hilbert modules, the Kas-
parov stabilization theorem, and frames in Hilbert modules, Proc. Amer.
Math. Soc., 131 (2003), 1557–1564.
[51] M. Rieffel, Morita equivalence for C ∗ -algebra and W ∗ -algebra, J. Pure
Allpied Algebra, 5 (1974), 51–96.
[52] S. Sakai, C ∗ -algebrs and W ∗ -algebras, Springer-Verlag, New York, 1998.
[53] T. Strohmer, R. Heath, Grassmannian frames with applications to coding
and communication, in preparation.
[54] R. Vale, S. Waldron, Tight frames and their symmetries, Constr. Approx.,
21 (2005), 83–112.
[55] N. Wegge-Olsen, K-theory and C ∗ -algebras - a Friendly Approach, Oxford
University Press, Oxford, England, 1993.

[56] R. Young, An Introduction to Nonharmonic Fourier Series, Academic Press,

New York, 1980.

101