Analysis On Fock Spaces, Kehe Zhu (2012)
Analysis On Fock Spaces, Kehe Zhu (2012)
Analysis On Fock Spaces, Kehe Zhu (2012)
Series Editors:
Sheldon Axler
San Francisco State University
Kenneth Ribet
University of California, Berkeley
Advisory Board:
Graduate Texts in Mathematics bridge the gap between passive study and
creative understanding, offering graduate-level introductions to advanced topics
in mathematics. The volumes are carefully written as teaching aids and highlight
characteristic features of the theory. Although these books are frequently used as
textbooks in graduate courses, they are also suitable for individual study.
123
Kehe Zhu
Department of Mathematics and Statistics
State University of New York
Albany, NY
USA
ISSN 0072-5285
ISBN 978-1-4419-8800-3 ISBN 978-1-4419-8801-0 (eBook)
DOI 10.1007/978-1-4419-8801-0
Springer New York Heidelberg Dordrecht London
Library of Congress Control Number: 2012937293
Several natural L p spaces of analytic functions have been widely studied in the past
few decades, including Hardy spaces, Bergman spaces, and Fock spaces. The terms
“Hardy spaces” and “Bergman spaces” are by now standard and well established.
But the term “Fock spaces” is a different story. I am aware of at least two other
terms that refer to the same class of spaces: Bargmann spaces and Segal–Bargmann
spaces. There is no particular reason, other than personal tradition, why I use “Fock
spaces” instead of the other variants. I have not done and do not intend to do any
research in order to justify one choice over the others.
Numerous excellent books now exist on the subject of Hardy spaces. Several
books about Bergman spaces, including some of my own, have also appeared in the
past few decades. But there has been no book on the market concerning the Fock
spaces. The purpose of this book is to fill that vacuum. There seems to be an honest
need for such a book, especially when many results are by now complete. It is at
least desirable to have the most important results and techniques summarized in one
book, so that newcomers, especially graduate students, have a convenient reference
to the subject.
There are certainly common themes to the study of the three classes of spaces
mentioned above. For example, the notions of zero sets, interpolating sets, Hankel
operators, and Toeplitz operators all make perfect sense in each of the three cases.
But needless to say, the resulting theories and results as well as the techniques
devised often depend on the underlying spaces. I will not say anything about the
various differences between the Hardy and Bergman theories; experts in these fields
are well aware of them.
What makes Fock spaces a genuinely different subject is mainly the flatness of
the domain on which these spaces are defined: the complex plane with the Euclidean
metric in our setup. Hardy and Bergman spaces are usually defined on curved
spaces, for example, bounded domains or half-spaces with a non-Euclidean metric.
Another major difference between the Fock theory and the Hardy/Bergman theory is
the behavior of the reproducing kernel in the L2 case: the Fock L2 space possesses an
exponential kernel, while the Hardy and Bergman L2 spaces both have a polynomial
kernel.
v
vi Preface
Let me mention a few particular phenomena that are unique to the analysis on
Fock spaces, as opposed to the more well-known Hardy and Bergman space settings.
First, the Fock kernel eα zw is neither bounded above nor bounded below, even
when one of the two variables is fixed. In the Hardy and Bergman theories, the
kernel function (1 − zw)α is both bounded above and bounded below when one of
the two variables is fixed. This makes many estimates in the Fock space setting
much more difficult. On the other hand, the exponential decay of e−α |z| makes it
2
much easier to prove the convergence of certain integrals and infinite series in the
Fock space setting than their Hardy and Bergman space counterparts.
Second, in the Fock space setting, there are no bounded analytic or harmonic
functions other than the trivial ones (constants). Therefore, many techniques in the
Hardy and Bergman space theories that are based on approximation by bounded
functions are no longer valid.
Third, and more technically, in the theory of Hankel and Toeplitz operators on
the Fock space, there is no “cutoff” point when characterizing membership in the
Schatten classes, while “cutoff” exists in both the Hardy and Bergman settings.
Also, for a bounded symbol function ϕ , the Hankel operator Hϕ on the Fock space
is compact if and only if Hϕ is compact. This is something unique for the Fock
spaces.
Fourth, because analysis on Fock spaces takes place on the whole complex
plane, certain techniques and methods from Fourier analysis become available. One
such example is the relationship between Toeplitz operators on the Fock space and
pseudodifferential operators on L2 (R).
And finally, I want to mention the role that Fock spaces play in quantum physics,
harmonic analysis on the Heisenberg group, and partial differential equations. In
particular, the normalized reproducing kernels in the Fock space are exactly the
so-called coherent states in quantum physics, the parametrized Berezin transform
on the Fock space provides a solution to the initial value problem on the complex
plane for the heat equation, and weighted translation operators give rise to a unitary
representation of the Heisenberg group on the Fock space.
I chose to develop the whole theory in the context of one complex variable,
although pretty much everything we do in the book can be generalized to the case
of finitely many complex variables. The case of Fock spaces of infinitely many
variables is a subject of its own and will not be discussed at all in the book.
I have tried to keep the prerequisites to a minimum. A standard graduate course
in each of real analysis, complex analysis, and functional analysis should prepare
the reader for most of the book. There are, however, several exceptions. One is
Lindelöf’s theorem which determines when a certain entire function is of finite type,
and the other is the Calderón–Vaillancourt theorem concerning the boundedness
of certain pseudodifferential operators. These two results are included in Chap. 1
without proof. Used without proof are also a couple of theorems from abstract
algebra when we characterize finite-rank Hankel and Toeplitz operators in Chaps. 6
and 7, and a couple of theorems from the general theory of interpolation when we
describe the complex interpolation spaces for Fock spaces in Chap. 2.
Preface vii
I have included some exercises at the end of each chapter. Some of these are
extensions or supplements to the main text, some are routine estimates omitted in
the main proofs, some are “lemmas” taken out of research papers, while others
are estimates or lemmas that I came up during the writing of the book that were
eventually abandoned because of better approaches found later. I have tried my best
to give a reference whenever a nontrivial result appears in the exercises.
I have tried to include as many relevant references as possible. But I am sure that
the Bibliography is not even nearly complete. I apologize in advance if your favorite
paper or reference is missing here. I did not omit it on purpose. I either overlooked
it or was not aware of it. The same is true with the brief comments I make at the end
of each chapter. I have tried my best to point the reader to sources that I consider
to be original or useful, but these comments are by no means authoritative and are
more likely biased because of my limitations in history and knowledge.
As usual, my family has been very supportive during the writing of this book.
I am very grateful to them—my wife Peijia and our sons Peter and Michael—for
their encouragement, understanding, patience, and tolerance. During the writing of
the book, I also received help from Lewis Coburn, Josh Isralowitz, Haiying Li, Alex
Schuster, Kristian Seip, Dan Stevenson, and Chunjie Wang. Thank you all!
1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.1 Entire Functions .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 3
1.2 Lattices in the Complex Plane . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 9
1.3 Weierstrass σ -Functions .. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 13
1.4 Pseudodifferential Operators . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 19
1.5 The Heisenberg Group .. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 25
1.6 Notes .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 27
1.7 Exercises.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 29
2 Fock Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 31
2.1 Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 33
2.2 Some Integral Operators .. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 43
2.3 Duality of Fock Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 53
2.4 Complex Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 59
2.5 Atomic Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 63
2.6 Translation Invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 75
2.7 A Maximum Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 81
2.8 Notes .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 87
2.9 Exercises.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 89
3 The Berezin Transform and BMO . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 93
3.1 The Berezin Transform of Operators.. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 95
3.2 The Berezin Transform of Functions.. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 101
3.3 Fixed Points of the Berezin Transform.. . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 113
3.4 Fock–Carleson Measures . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 117
3.5 Functions of Bounded Mean Oscillation .. . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 123
3.6 Notes .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 133
3.7 Exercises.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 135
4 Interpolating and Sampling Sequences . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 137
4.1 A Notion of Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 139
4.2 Separated Sequences .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 143
ix
x Contents
References .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 331
Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 341
Chapter 1
Preliminaries
This book is about certain spaces of entire functions and certain operators defined
on these spaces. So we begin by recalling some elementary results about entire
functions. The first few of these results can be found in any graduate-level complex
analysis text, and no proof is included here.
Let C denote the complex plane. If a function f is analytic on the entire complex
plane C, we say that f is an entire function. One of the fundamental results in
complex analysis is the following identity theorem.
Theorem 1.1. If f is entire and the zero set of f ,
Z( f ) = {z ∈ C : f (z) = 0},
where z = x + iy and dA(z) = dxdy is the Lebesgue area measure. The inequality in
(1.2) is the area version of the mean value theorem.
The next result is called Liouville’s theorem.
4 1 Preliminaries
If f (0) is nonzero but not necessarily 1, Jensen’s formula takes the form
N 2π
r 1
log | f (0)| = − ∑ log + log | f (reiθ )| dθ , (1.4)
k=1 |zk | 2π 0
where {z1 , · · · , zN } are zeros of f in 0 < |z| < r. More generally, if f has a zero of
order k at the origin, then Jensen’s formula takes the following form:
2π
| f (k) (0)| N
r 1
log + k logr = − ∑ log + log | f (reiθ )| dθ ,
k! k=1 |zk | 2 π 0
If a is any nonzero complex number, it is clear that E(z/a) has a unique, simple zero
at z = a.
1.1 Entire Functions 5
Theorem 1.6. Let {zn } be a sequence of nonzero complex numbers such that the
sequence {|zn |} is nondecreasing and tends to ∞. Then it is possible to choose a
sequence {pn } of nonnegative integers such that
∞ pn +1
r
∑ |zn |
<∞ (1.5)
n=1
converges uniformly on every compact subset of C, the function P is entire, and the
zeros of P are exactly {zn }, counting multiplicity.
Note that the choice pn = n − 1 will always satisfy (1.5). In many cases, however,
there are “better” choices. In particular, if {zn } is the zero sequence of an entire
function f and if there exists an integer p such that
∞
1
∑ |zn | p+1 < ∞, (1.7)
n=1
we say that f is of finite rank. If p is the smallest integer such that (1.7) is satisfied,
then f is said to be of rank p. A function with only a finite number of zeros has
rank 0. A function is of infinite rank if it is not of finite rank.
If f is of finite rank p and {zn } is the zero sequence of f , then (1.7) is satisfied
with pn = p. The product P(z) associated with this canonical choice of {pn } will be
called the standard form.
Theorem 1.7. Let f be an entire function of finite rank p. If P is the standard
product associated with the zeros of f , then there exist a nonzero integer m and
an entire function g such that
The integer m is unique, and the entire function g is unique up to an additive constant
of the form 2kπ i.
For an entire function of finite rank, we say that (1.8) is the standard factorization
of f , or the Weierstrass factorization of f .
Let f be an entire function of finite rank p. If the entire function g in the standard
factorization (1.8) of f is a polynomial of degree q, then we say that f has finite
genus. In this case, the number μ = max(p, q) is called the genus of f .
Let f be an entire function. For any r > 0, we write
In this case, the order of f is the infimum of the set of all such numbers a.
The following characterization of entire functions of finite order is traditionally
referred to as the Hadamard factorization theorem.
Theorem 1.9. An entire function f is of finite order ρ if and only if it is of finite
genus μ . Moreover, the order and genus of f satisfy the following relations: μ ≤
ρ ≤ μ + 1.
When 0 < ρ < ∞, we define
log M(r)
σ = lim sup .
r→∞ rρ
If σ < ∞, we say that f is of finite type. More specifically, we say that f is of order
ρ and type σ . If σ = ∞, we say that f is of maximum type or infinite type.
Let {zn } denote the zero sequence, excluding 0, of an entire function f . The
infimum of all positive numbers s such that
∞
1
∑ |zn |s < ∞
n=1
The following result is sometimes called Lindelöf’s theorem. This result is not so
standard in the sense that it does not appear in most elementary complex analysis
texts. See [38] for a proof.
Theorem 1.11. Suppose that ρ is a positive integer, f is an entire function of order
ρ , f (0) = 0, and {zn } is the zero sequence of f . Then f is of finite type if and only
if the following two conditions hold:
(a) n(r) = O(rρ ) as r → ∞, where n(r) is the number (counting multiplicity) of
zeros of f in |z| ≤ r.
(b) The partial sums
1
S(r) = ∑ ρ
|zn |≤r zn
are bounded in r.
Lindelöf’s theorem will be useful for us in Chap. 5 when we study zero sequences
for functions in Fock spaces. The reader should be mindful of the fact that there
are several results in complex analysis that are called Lindelöf’s theorem. In most
cases, these results are certain generalizations of the classical maximum modulus
principle.
8 1 Preliminaries
1.2 Lattices in the Complex Plane 9
The complex plane is flat, and lattices in it are easy to describe. We will need to use
rectangular lattices on several occasions later on. In this section, we fix notation and
collect basic facts about lattices in the complex plane.
The simplest lattice in C is the standard integer lattice
Z2 = {m + in : m ∈ Z, n ∈ Z},
where Z = {0, ±1, ±2, · · · , } is the integer group. All lattices we use in the book are
isomorphic to Z2 .
Let ω be any complex number, and let ω1 and ω2 be any two nonzero complex
numbers such that their ratio is not real. For any integers m and n, let ωmn = ω +
mω1 + nω2. The set
Λ = Λ (ω , ω1 , ω2 ) = {ωmn : m ∈ Z, n ∈ Z}
ω, ω + ω1 , ω + ω2 , ω + ω1 + ω2 ,
and is centered at
1
c = ω + (ω1 + ω2).
2
We shift this parallelogram so that the center becomes ω and the vertices become
1 1 1 1
ω − (ω1 + ω2 ), ω + (ω1 − ω2 ), ω + (ω2 − ω1 ), ω + (ω1 + ω2).
2 2 2 2
We denote this new parallelogram by R00 and call it the fundamental region of
Λ (ω , ω1 , ω2 ). For any integers m and n, let Rmn = R00 + ωmn , with ωmn being the
center of Rmn .
Lemma 1.12. Let Λ = Λ (ω , ω1 , ω2 ) be any lattice in C. For any positive number
δ , there exists a positive constant C such that
∑ e−δ |z−w|
2
≤C
z∈Λ
for all w ∈ C.
10 1 Preliminaries
Proof. By translation invariance, it suffices for us to prove the desired inequality for
w in the fundamental region R00 of Λ . If w is in the relatively compact set R00 , then
|w/z| < 1/2 for all but a finite number of points z ∈ Λ . For all such points z, we have
1
|z − w|2 = |z|2 |1 − (w/z)|2 ≥ |z|2 .
4
δ
Since ∑z∈Λ e− 4 |z| is obviously convergent, we obtain the desired result.
2
Lemma 1.13. With notation from above, we have
C= {Rmn : m ∈ Z, n ∈ Z},
and
C
f (z) dA(z) = ∑ f (z) dA(z)
m,n∈Z Rmn
such that the distance between any two points in each of the sublattices is at least R.
Proof. Fix a positive integer k such that k|ω1 | > R and k|ω2 | > R. For each j =
( j1 , j2 ) with 0 ≤ j1 ≤ k and 0 ≤ j2 ≤ k, let
Λ j = Λ (ω + j1 ω1 + j2 ω2 , kω1 , kω2 )
= {(ω + j1 ω1 + j2 ω2 ) + (mkω1 + nkω2 ) : m ∈ Z, n ∈ Z}.
Most lattices we use in the book are square ones. More specifically, for any given
positive parameter r, we consider the case when ω = 0, ω1 = r, and ω2 = ir. The
resulting lattice is
rZ2 = {rm + irn : m ∈ Z, n ∈ Z}.
We mention two particular cases. First, for r = π /α , where α is a positive
parameter, the resulting lattices are used in the next section when we introduce the
Weierstrass σ -functions. Second, for r = 1/N, where N is a positive integer, the
resulting lattices will be employed in Chaps. 6–8 when we characterize Hankel and
Toeplitz operators in Schatten classes.
For any two points z = x + iy and w = u + iv in rZ2 , we let γ (z, w) denote the
following path in rZ2 : we first move horizontally from z to u + iy and then vertically
from u + iy to u + iv. When z = 0, we write γ (w) in place of γ (0, w). The path γ (z, w)
is of course discrete. We use |γ (z, w)| to denote the number of points in γ (z, w) and
call it the length of γ (z, w).
The following technical lemma will play a critical role in Chap. 8.
Lemma 1.15. For any positive r and σ , there exists a positive constant C = Cr,σ
such that
z∈rZ2 w∈rZ2
for all u ∈ rZ2 , where χγ (z,w) is the characteristic function of γ (z, w).
Proof. Without loss of generality, we may assume that r = 1. Adjusting the constant
σ will then produce the general case.
Also, it is obvious that
u + γ (z, w) = γ (u + z, u + w),
Similarly, 0 lies on the vertical segment of γ (z, w) if and only if one of the following
is true:
(3) w is on the positive y-axis and z is in the third or fourth quadrant: w = ni,
z = k − mi, where n and m are nonnegative integers and k is an integer.
(4) w is on the negative y-axis and z is in the first or second quadrant: w = −ni,
z = k + mi, where n and m are nonnegative integers and k is an integer.
In each of the cases above, we have
|z − w|2 = (n + m)2 + k2 ≥ n2 + m2 + k2 .
Therefore,
∞ ∞ ∞
S≤4∑ ∑ ∑ e−σ (n
2 +m2 +k2 )
In this section we introduce several Weierstrass functions on the complex plane and
prove their periodicity or quasiperiodicity. In particular, the Weierstrass σ -function
will serve as a prototype for functions in Fock spaces and will play an important role
in our characterization of interpolating and sampling sequences for Fock spaces.
Lattices in this section are all based at the origin:
where the summation (with a prime) extends over all integers m and n with (m, n) =
(0, 0).
Proposition 1.16. The function P is an even meromorphic function in the complex
plane whose poles are exactly the points in the lattice Λ . Furthermore, P is doubly
periodic with periods ω1 and ω2 :
for all z ∈ C − Λ .
Proof. For any small δ > 0, let
1
P (z) = −2 ∑ .
m,n (z − ωmn )
3
Since {−ωmn : m ∈ Z, n ∈ Z} represents the same lattice Λ and the series above
converges absolutely (so its terms can be rearranged in any way we like), we see
that P is an odd function, and so the original function P is even.
On the other hand, for each k = 1, 2, we have
1
P (z + ωk ) = −2 ∑ .
m,n (z − ωmn + ωk )3
Since {ωmn − ωk : m ∈ Z, n ∈ Z} represents the same lattice Λ and the above series
converges absolutely for any z ∈ C − Λ , we see that P (z + ωk ) = P (z), so P is
doubly periodic with periods ω1 and ω2 .
If we integrate the equation P (z + ωk ) = P (z) on the connected region C − Λ ,
we will find a constant Ck such that P(z + ωk ) = P(z) + Ck for k = 1, 2 and all
z ∈ C − Λ . Setting z = −ωk /2 and using the fact that P is even, we obtain Ck = 0
for k = 1, 2. This shows that P is doubly periodic with periods ω1 and ω2 .
To every lattice Λ = Λ (0, ω1 , ω2 ) = {ωmn }, we associate another function ζ (z) =
ζΛ (z) as follows:
1 1 1 z
ζ (z) = + ∑ + + 2 . (1.11)
z m,n z − ωmn ωmn ωmn
The following proposition lists some of the basic properties of this function, which
should not be confused with the famous Riemann ζ -function.
Proposition 1.17. Each ζ is an odd meromorphic function with simple poles at
precisely the points of Λ . Furthermore, for k = 1, 2, we have
Break this into integration over the four sides of R and use the quasiperiodicity of
ζ . We obtain the desired result.
To every lattice Λ = Λ (0, ω1 , ω2 ) = {ωmn }, we associate yet another function
σ (z) = σΛ (z) as follows:
z z z2
σ (z) = z ∏ 1 − exp + . (1.14)
m,n ωmn ωmn 2ωmn 2
The following proposition lists some of the basic properties of the Weierstrass
σ -functions.
16 1 Preliminaries
Proposition 1.19. Each σ is an entire function whose zero set is exactly the lattice
Λ = {ωmn }. Furthermore, σ is odd and quasiperiodic in the following sense:
d
log σ (z) = ζ (z),
dz
coupled with the condition
σ (z)
lim = 1.
z→0 z
ζ (z + ωk ) = ζ (z) + ηk
in the connected region C − Λ and then exponentiate the result, we obtain a constant
ck such that
σ (z + ωk ) = ck eηk z σ (z), z ∈ C.
Let z = −ωk /2 and use the fact that σ is odd. We get ck = −eηk ωk /2 .
Finally, in this section, we consider the special case of square lattices.
For any
positive
parameter α , we consider the lattice Λ = Λ α given by ω 1 = π /α and
ω2 = π /α i. Thus,
Λα = { π /α (m + in) : m ∈ Z, n ∈ Z}.
In this particular case, we will compute the constants ηk and relate the quasiperiod-
icity of σ to a certain isometry on Fock spaces.
Proposition 1.20. Suppose σ is the Weierstrass σ -function associated
to the square
lattice
Λ α = { ω mn }, where ω mn = π / α (m + in), so that ω = π /α and ω2 =
√ √ 1
π /α i. Then η1 = πα and η2 = − πα i. Furthermore,
α
eαω mn z− 2 |ωmn | σ (z − ωmn ) = (−1)m+n+mn σ (z)
2
(1.16)
1.3 Weierstrass σ -Functions 17
1
= ζ (z).
i
Therefore,
2 η1
η2 = 2ζ (ω2 /2) = 2ζ (iω1 /2) = ζ (ω1 /2) = .
i i
√ √
This, along with (1.13), gives η1 = πα and η2 = − πα i.
To prove the translation relation in (1.16), observe that
for all positive integers m. Since σ is an odd function, it is then easy to see that the
above equation also holds for negative integers m. Similarly,
Plug in
18 1 Preliminaries
√ √
ω1 = π /α , ω2 = π /α i, η1 = πα , η2 = − πα i.
We obtain
α
σ (z + ωmn ) = (−1)m+n+mn eαω mn z+ 2 |ωmn | σ (z)
2
Proof. Property (a) follows from the quasiperiodicity of σ ; see (1.15) and (1.16).
Property (b) then follows from (a) and the fact that each point in Λα is a simple zero
of σ .
As a consequence of condition (b) above, we see that the Weierstrass σ -function
associated to Λα is of order 2 and of type α /2.
1.4 Pseudodifferential Operators 19
One of the tools we will employ in Chap. 6 when we study Toeplitz operators is
the notion of pseudodifferential operators. More specifically, Toeplitz operators on
the Fock space are unitarily equivalent to a class of pseudodifferential operators on
L2 (R). In this section, we introduce the concept of pseudodifferential operators on
the real line and collect several results in this area that will be needed later. The
references for this section are Folland’s books [92] and [93].
We begin with two well-known operators D and X defined on the space of smooth
functions on R by
1
X f (x) = x f (x), D f (x) = f (x), (1.17)
2α i
where α is any fixed positive constant. The introduction of a parameter α at this
point will facilitate and simplify our computations later in association with the Fock
spaces. The number h = π /α plays the role of Planck’s constant in quantum physics.
It is easy to verify that, as densely defined unbounded operators on L2 (R, dx),
both D and X are self-adjoint. This is an easy consequence of integration by parts.
The operators
f (D) = ∑ ak Dk , f (X) = ∑ ak X k
are perfectly and naturally defined. This easily extends to a large class of symbol
functions f . What results in are symbol calculi for the self-adjoint operators D
and X.
The notion of pseudodifferential operators arises when we try to establish a
symbol calculus for the pair of operators D and X. In other words, if we are given a
good function f (ζ , x) on R × R, we wish to define an operator f (D, X) in a natural
way. If f = aζ + bx is linear, obviously we should just define f (D, X) = aD + bX.
But we already run into problems when f is just a second-degree polynomial, say
f (ζ , x) = ζ x = xζ ,
The operators D and X do not commute, so the two products above are not equal. In
fact, it is easy to verify the following commutation relation:
1
[D, X] = DX − XD = I, (1.19)
2α i
f (ζ , x) = ∑ amn xm ζ n
and define
f (D, X) = ∑ amn X m Dn .
and define
f (D, X) = ∑ amn Dn X m .
The functional calculi defined this way are called Wick and anti-Wick correspon-
dences. They have been studied extensively in analysis and mathematical physics.
There is another important functional calculus for D and X, the John–Nirenberg
correspondence, which is especially important in partial differential equations.
We will not pursue any of the above correspondences. Instead, we focus on the
so-called Weyl pseudodifferential operators. This approach depends on a particular,
but natural, choice for the definition of σ (D, X) when σ (ζ , x) = e2π i(pζ +qx), where p
and q are real constants. Once this is done, the definition of σ (D, X) for more general
symbol functions σ (ζ , x) can be given with the help of Fourier and inverse Fourier
transforms.
1.4 Pseudodifferential Operators 21
or, equivalently,
π 2 ipq
π p
e2π i(pD+qX) f (x) = e2π iqx+ α f x+ . (1.22)
α
∂g
= 2α i(pD + qX)g, (1.23)
∂t
subject to the initial condition g(x, 0) = f (x). Rewrite the equation in (1.23) as
∂g ∂g
−p = 2α iqxg, (1.24)
∂t ∂x
and let G(t) = g(x(t),t) with x(t) = x − pt. Then by the chain rule,
∂g ∂g
G (t) = −p ,
∂t ∂x
Let t = 1. We have
g(x − p, 1) = f (x)e2α iqx−α ipq .
Replace x by x + p. We arrive at
and we define
σ (D, X) = σ (p, q)e2π i(pD+qX) dp dq. (1.26)
R R
Here, the integral is an ordinary Bochner integral whenever σ , the Fourier transform
of σ , is in L1 (R × R).
Theorem 1.23. If σ (ζ , x) and f (x) are regular enough, then we have
α x + y 2α i(x−y)ζ
σ (D, X) f (x) = σ ζ, e f (y) dy dζ . (1.27)
π R R 2
σ (D, X) f (x)
= σ (p, q)e2π i(pD+qX) f (x) dp dq
R R
π p 2π iqx+ π 2 pqi
= σ (p, q) f x + e α dp dq
R R α
π p
π 2 pq
= σ (ζ , w)e−2π i(pζ +qw) e2π iqx+
f x+ dp dq dζ dw
α
R R R R α
π p −2π ipζ
π p
= σ (ζ , w)δ x − w + e f x+ dp dζ dw
R R R 2α α
π p −2π ipζ
π p
= σ ζ,x + e f x+ dp dζ
R R 2α α
α 1
= σ ζ , (x + y) e−2α i(y−x)ζ f (y) dy dζ ,
π R R 2
It is thus also natural to simply take (1.27) as the definition of the Weyl
pseudodifferential operator σ (D, X). We remind the reader that there is a positive
parameter α built into our definition of pseudodifferential operators. To see the pre-
cise relationship between our rescaled σ (D, X) and the classical pseudodifferential
operators (as defined in Folland’s book [92], for example), we change variables and
rewrite (1.27) as follows:
x + y 2π i(x−y)ζ
σ (D, X) f1/r (rx) = σr ζ , e f (y) dy dζ , (1.29)
R R 2
where r = π /α . Here, fr (x) = f (rx) denotes the dilation of f by a positive
number r. The integral on the right-hand side of (1.29) is the classical definition
of the Weyl pseudodifferential operator with symbol σr .
The results in the three theorems below are all invariant under dilation. Therefore,
our rescaling does not alter the validity of these classical results.
The pseudodifferential operator σ (D, X) is so far only loosely defined. If σ
is sufficiently regular and f is compactly supported on R, then the integral in
(1.27) converges. For general σ , the integral in (1.27) may or may not converge,
and the definition of σ (D, X) f may only be defined for f in a certain class.
Our main concern here is the following problem: for which functions σ can the
pseudodifferential operator σ (D, X) be extended to a bounded or compact operator
on L2 (R, dx)?
Theorem 1.24. Suppose σ (ζ , x) is a function on R × R of class C3 and there exists
a positive constant C such that
n+m
∂ σ
∑ ∂ ζ n ∂ xm
(ζ , x) ≤C
n+m≤3
∂ n+m σ
∈ C0 (R × R)
∂ ζ n ∂ xm
for every pair of nonnegative integers m and n with n + m ≤ 3. Then the pseudodif-
ferential operator σ (D, X) is compact on L2 (R, dx).
24 1 Preliminaries
Although we will not use the Heisenberg group in a critical way anywhere in the
book, it is interesting to show how it fits nicely in the theory of Fock spaces. In this
brief section, we give its definition and produce a unitary representation based on
pseudodifferential operators.
The Heisenberg group H is the set C × R (or R2 × R) with the following group
operation:
(z, s) ⊕ (w,t) = (z + w, s + t − Im (zw)),
z, w = z1 w1 + · · · + zn wn .
ferential operator e2α i(pD+qX) is a unitary operator on L2 (R, dx). Furthermore, the
mapping
(p + iq,t) → u(p + iq,t) =: eα it e2α i(pD+qX)
This shows that u(z,t) preserves the group operation in the Heisenberg group H.
The mapping u(z,t) is called the Schrödinger representation of the Heisenberg
group H on L2 (R). In the next chapter, we will obtain another representation of H,
a unitary representation on the Fock space based on weighted translations.
1.6 Notes 27
1.6 Notes
The results in the first section, except Lindelöf’s theorem, are all well known and
can be found in any elementary complex analysis book. In particular, these results
can all be found in Conway’s book [67].
Lindelöf’s theorem will be needed in Chap. 3 when we study zero sequences for
Fock spaces. This is probably not a result that can be found in elementary texts. See
2.10.1 of Boas’ book [38] for a detailed proof of this result.
The section about lattices in the complex plane is completely elementary.
Whenever we really use lattices later on, we restrict our attention to square lattices,
although many arguments can easily be adapted to arbitrary lattices, even to
sequences that behave like lattices. Perhaps Lemma 1.15 looks peculiar to the reader,
but it is critical for the study of Hankel operators in Chap. 8.
Pseudodifferential operators constitute an important subject by itself, and there is
extensive literature about them. Of course, we have only touched the surface of this
vast area of modern analysis. The connection between pseudodifferential operators
and Toeplitz operators on the Fock space is both fascinating and useful. Because
of this connection, the study of Toeplitz operators on the Fock space becomes
especially interesting and fruitful. In particular, this provides us with extra and
unique tools to study Toeplitz operators on the Fock space as opposed to Toeplitz
operators on the Hardy and Bergman spaces.
Our presentation in Sect. 1.4 follows Folland’s books [92, 93] very closely. A
slight modification is made in the definition of pseudodifferential operators here in
order to incorporate the weight parameter α into everything. Note that the proof of
Theorem 1.23 depends on certain elementary facts from Fourier analysis that we are
taking for granted. It should be easy for the interested reader to make the arguments
completely rigorous.
The Heisenberg group appears very naturally in many different areas, including
Fourier analysis, harmonic analysis, and mathematical physics. The Heisenberg
group shows up in this book when we study the action of translations on Fock
spaces. Although it is possible for us to avoid the Heisenberg group, we thought
it is nice to put things in the right context.
The Weierstrass σ -functions provide a family of examples that will be very useful
to us later on when we study zero sets, interpolating sets, and sampling sets. The
book [241] contains much more information about the Weierstrass σ -functions as
well as several other important classes of entire and meromorphic functions.
28 1 Preliminaries
1.7 Exercises 29
1.7 Exercises
1. Suppose f is entire, f (z) = 0 for some z ∈ C, and |z| < r. Then log | f (z)| is
equal to
2 iθ
N r − zk z 1 2π re + z
− ∑ log +
2π 0 Re iθ − z
log | f (reiθ )| dθ ,
k=1 r(z − zk ) re
if and only if p > 2, where the summation is to exclude the possible occurrence
of 0 in the denominator.
13. Prove the commutation relation (1.19).
14. Convince yourself that the formal identity (1.28) is equivalent to the Fourier
inversion formula.
Chapter 2
Fock Spaces
In this chapter, we define Fock spaces and prove basic properties about them.
The following topics are covered in this chapter: reproducing kernel, integral
representation, duality, complex interpolation, atomic decomposition, translation
invariance, and a version of the maximum modulus principle.
α −α |z|2
dλα (z) = e dA(z),
π
where dA is the Euclidean area measure on the complex plane. A calculation with
polar coordinates shows that dλα is a probability measure.
The Fock space Fα2 consists of all entire functions f in L2 (C, dλα ). It is easy
to show that Fα2 is a closed subspace of L2 (C, dλα ). Consequently, Fα2 is a Hilbert
space with the following inner product inherited from L2 (C, dλα ):
f , gα = f (z) g(z) dλα (z).
C
|z|<R
f (z) en (z) dλα (z) = ∑ ak |z|<R
zk en (z) dλα (z).
k=0
Therefore, the condition that f , en = 0 for all n ≥ 0 implies that an = 0 for all
n ≥ 0 which in turn implies that f = 0. This shows that the system {en } is complete
in Fα2 .
34 2 Fock Spaces
Kα (z, w) = eα z w , z, w ∈ C.
The desired result then follows from the uniqueness in Riesz representation.
Recall that every closed subspace X of a Hilbert space H uniquely determines an
orthogonal projection P : H → X.
Corollary 2.3. The orthogonal projection
Pα f (z) = Pα f , Kz α = f , Pα Kz α = f , Kz α
= f (w)Kα (z, w) dλα (z).
C
Kα (w, z) α 2
kz (w) = = eα zw− 2 |z|
Kα (z, z)
denote the normalized reproducing kernel at z. Each kz is a unit vector in Fα2 . The
following change of variables formula will be used many times later in the book.
Corollary 2.4. Suppose f ≥ 0 or f ∈ L1 (C, dλα ). Then for any z ∈ C, we have
f (z ± w) dλα (w) = f (w)|kz (w)|2 dλα (w),
C C
and
f [±(z − w)]|kz (w)|2 dλα (w) = f (w) dλα (w).
C C
Proof. It is clear that
α
f (z ± w)e−α |w| dA(w)
2
f (z ± w) dλα (w) =
C π C
α
f (w)e−α |z−w| dA(w)
2
=
π C
f (w)e−α |z|
2 +α zw+α zw
= dλα (w)
C
= f (w)|kz (w)|2 dλα (w).
C
The assumption that f ≥ 0 or f ∈ L1 (C, dλα ) ensures that all integrals above make
sense. The proof of the other identity is similar.
Corollary 2.5. Suppose α > 0 and β is real. Then
β zā
e dλα (z) = eβ |a| /4α
2 2
for all a ∈ C.
36 2 Fock Spaces
p
For f ∈ Lα , we write
pα α 2 p
f pp,α = f (z)e− 2 |z| dA(z). (2.1)
2π C
On the other hand, the function f (w) = kz (w) is a unit vector in Fα∞ and kz (z) =
eα |z| /2 . Thus, we actually have
2
z∈C
for every function f ∈ Fαp . This will follow from the next proposition.
Proposition 2.9. Suppose 0 < p < ∞, f ∈ Fαp , and fr (z) = f (rz). Then:
(a) fr − f p,α → 0 as r → 1− .
(b) There is a sequence {pn } of polynomials such that pn − f p,α → 0 as n → ∞.
Proof. Suppose {gn } and g are functions in L p (X, d μ ) such that
gn (x) → g(x), n → ∞,
if and only if
lim |gn | p dμ = |g| p dμ .
n→∞ X X
This is a simple consequence of Fatou’s lemma; see Lemma 3.17 of [119] for
example. Given f ∈ Fαp , we have
pα 2 p
fr pp,α = f (rz)e−α |z| /2 dA(z)
2π C
p
pα −α |z|2 /2 −pα |z|2 (r−2 −1)/2
= f (z)e e dA(z).
2π r 2 C
Since
2 (r−2 −1)/2
e−pα |z| ≤1
for all z ∈ C and 0 < r < 1, an application of the dominated convergence theorem
shows that fr p,α → f p,α , and hence fr − f p,α → 0 as r → 1− . This proves
part (a).
Part (b) follows from part (a) if we can show that for every r ∈ (0, 1), the function
fr can be approximated by its Taylor polynomials in the norm topology of Fαp .
To this end, we fix some r ∈ (0, 1) and fix some β ∈ (r2 α , α ). It follows from
Corollary 2.8 that fr ∈ Fβ2 . Similarly, it follows from Corollary 2.8 that Fβ2 ⊂ Fαp
and there exists a positive constant C such that g p,α ≤ Cg2,β for all g ∈ Fβ2 .
Now, if pn is the nth Taylor polynomial of fr , then by Proposition 2.1,
Obviously, fα∞ is a closed subspace of Fα∞ . In fact, fα∞ is the closure in Fα∞ of the set
of all polynomials. Thus, the space fα∞ is separable while the space Fα∞ is not.
p q
Theorem 2.10. If 0 < p < q < ∞, then Fα ⊂ Fα , and the inclusion is proper and
continuous. Moreover, Fα ⊂ fα∞ , and the inclusion is proper and continuous.
p
This shows that Fαp ⊂ Fαq with f q,α ≤ (q/p)1/q f p,α for all f ∈ Fαp .
To see that the inclusion Fαp ⊂ Fαq is proper, let us assume that Fαp = Fαq . Then the
identity map I : Fαp → Fαq is bounded, one-to-one, and onto. By the open mapping
theorem, there must exist a constant C > 0 such that
for all f ∈ Fαp . On the other hand, a computation with Stirling’s formula shows that
∞
rnp e−pα r
2 /2
zn pp,α = αp r dr
0
1 np/2
np
= Γ +1
αp 2
n np/2 √
∼ n.
αe
Thus,
n n/2 1
zn p,α ∼ n 2p ,
αe
and similarly,
n n/2 1
zn q,α ∼ n 2q .
αe
It is then obvious that there is no positive constant C with the property that
Another appeal to the open mapping theorem then shows that the inclusion Fαp ⊂ fα∞
is proper.
The next result gives another useful dense subset of Fαp .
Lemma 2.11. For any positive parameters α and γ , the set of functions of the form
n n
f (z) = ∑ ck Kγ (z, wk ) = ∑ ck eγ zwk ,
k=1 k=1
Combining this with the density of the functions ∑nk=1 ck Kβ (z, uk ) in Fβ2 , we
conclude that every polynomial can be approximated in the norm topology of Fαp
by functions of the form ∑nk=1 ck Kα (z, wk ). Since the polynomials are dense in Fαp ,
we have proved the result for Fαp , 0 < p < ∞.
The proof for fα∞ is similar.
Finally, in this section, as a consequence of the pointwise estimates, we establish
the maximum order and type for functions in the Fock spaces.
Theorem 2.12. Let f ∈ Fαp with 0 < p ≤ ∞. Then f is of order less than or equal
to 2. When f is of order 2, it must be of type less than or equal to α /2.
Proof. By Corollary 2.8, there exists a positive constant C such that
for all z ∈ C. In particular, M(r) ≤ Ceα r /2 for all r > 0. It follows that the order ρ
2
of f satisfies
log log M(r)
ρ = lim sup ≤ 2.
r→∞ log r
Also, if the order of f is actually 2, then its type σ satisfies
log M(r) α
σ = lim sup ≤ ,
r→∞ r2 2
and
Qα f (z) = |Kα (z, w)| f (w) dλα (w), (2.4)
C
respectively.
We need two well-known results from the theory of integral operators. The first
one concerns the adjoint of a bounded integral operator.
Lemma 2.13. Suppose 1 ≤ p < ∞ and 1/p + 1/q = 1. If an integral operator
T f (x) = H(x, y) f (y) dμ (y)
X
T ∗ : Lq (X, d μ ) → Lq (X, d μ )
Proof. This is a standard result in real analysis. See [113] for example.
The second result is a useful criterion for the boundedness of integral operators
on L p spaces, which is usually referred to as Schur’s test.
Lemma 2.14. Suppose H(x, y) is a positive kernel and
T f (x) = H(x, y) f (y) dμ (y)
X
is the associated integral operator. Let 1 < p < ∞ with 1/p + 1/q = 1. If there exist
a positive function h(x) and positive constants C1 and C2 such that
H(x, y)h(y)q dμ (y) ≤ C1 h(x)q , x ∈ X,
X
44 2 Fock Spaces
and
H(x, y)h(x) p dμ (x) ≤ C2 h(y) p , y ∈ X,
X
then the operator T is bounded on L p (X, d μ ). Moreover, the norm of T on L p (X, d μ )
1/q 1/p
does not exceed C1 C2 .
Proof. See [250] for example.
We now consider the action of the operators Pα and Qα on the space L p (C, dλβ ).
Thus, we fix two positive parameters α and β for the rest of this section and rewrite
the integral operators Pα and Qα as follows:
α
eα zw̄+β |w|
2 −α |w|2
Pα f (z) = f (w) dλβ (w),
β C
and
α
|eα zw̄+β |w|
2 −α |w|2
Qα f (z) = | f (w) dλβ (w).
β C
It follows from Lemma 2.13 that the adjoint of Pα and Qα with respect to the integral
pairing
f , gβ = f (z)g(z) dλβ (z)
C
is given, respectively, by
α (β −α )|z|2
Pα∗ f (z) = e eα zw̄ f (w) dλβ (w), (2.5)
β C
and
α (β −α )|z|2
Q∗α f (z) = e |eα zw̄ | f (w) dλβ (w). (2.6)
β C
On the other hand, it follows from the reproducing formula in Fα2+x that
α
eα zw̄ wk e−(α +x)|w| dA(w)
2
Pα ( fx,k )(z) =
π C
α
= e(α +x)[α z/(α +x)w] wk dλα +x (w)
α +x C
k 1+k
α αz α
= = zk .
α +x α +x α +x
Therefore,
p(1+k)
α
|Pα ( fx,k )| p dλβ = |z| pk dλβ (z)
C α +x C
p(1+k)
α Γ ((pk/2) + 1)
= .
α +x β pk/2
Now, if the integral operator Pα is bounded on L p (C, dλβ ), then there exists a
positive constant C (independent of x and k) such that
p(1+k)
α Γ ((pk/2) + 1) β Γ ((pk/2) + 1)
≤C ,
α +x β pk/2 px + β (px + β ) pk/2
or
p(1+k) 1+(pk/2)
α β
≤C .
α +x β + px
Fix any x > 0 and look at what happens in the above inequality when k → ∞. We
deduce that 2
α β
≤ .
α +x β + px
Cross multiply and simplify. The result is
pα 2 ≤ 2αβ + β x.
Proof. If p > 1 and Pα is bounded on L p (C, dλβ ), then Pα∗ is bounded on Lq (C, dλβ ),
where 1/p + 1/q = 1. Applying the formula for Pα∗ from (2.5) to the constant
function f = 1 shows that the function e(β −α )|z| is in Lq (C, dλβ ). From this, we
2
deduce that
q(β − α ) < β ,
which is easily seen to be equivalent to β < pα .
Lemma 2.17. If Pα is bounded on L1 (C, dλ β ), then α = 2β .
Proof. Fix any a ∈ C and consider the function
eα zā
fa (z) = , z ∈ C.
|eα zā |
Obviously, fa ∞ = 1 for every a ∈ C. On the other hand, it follows from (2.5) and
Corollary 2.5 that
α (β −α )|a|2
Pα∗ ( fa )(a) = e |eα wā | dλβ (w)
β C
α
= e(β −α )|a| eα |a| /(4β ) .
2 2 2
Since Pα∗ is bounded on L∞ (C), there exists a positive constant C such that
α2
β −α + ≤ 0,
4β
where x > 0 and k is a positive integer. It follows from (2.5) and the reproducing
property in Fα2+x that
α (β −α )|z|2
Pα∗ ( fx,k )(z) = eα zw̄ wk e−(β +x)|w| dA(w)
2
e
π C
2.2 Some Integral Operators 47
α (β −α )|z|2
= e e(β +x)[α z/(β +x)]w̄wk dλβ +x(w)
β +x C
α (β −α )|z|2 αz k
= e
β +x β +x
1+k
α
e(β −α )|z| zk .
2
=
β +x
We have
β Γ ((qk/2) + 1)
| fx,k |q dλβ = .
C qx + β (qx + β )qk/2
On the other hand, it follows from Lemma 2.16 and its proof that
β − q(β − α ) > 0,
so the integral
I= |Pα∗ ( fx,k )|q dλβ
C
can be evaluated as follows:
q(1+k)
α β
|z|qk e−(β −q(β −α ))|z| dA(z)
2
I=
β +x π C
q(1+k)
α β
= |z|qk dλβ −q(β −α )(z)
β +x β − q(β − α ) C
q(1+k)
α β Γ ((qk/2) + 1)
= .
β +x β − q(β − α ) (β − q(β − α ))qk/2
Therefore,
q(1+k)
α β Γ ((qk/2) + 1)
β +x β − q(β − α ) (β − q(β − α ))qk/2
is less than or equal to
Cβ Γ ((qk/2) + 1)
,
qx + β (qx + β )qk/2
48 2 Fock Spaces
Using the relation 1/p + 1/q = 1, we can change the right-hand side above to
pα − β
.
(p − 1)β + px
It follows that
Let q(x) denote the quadratic function on the left-hand side of the above inequality.
Since pα − β > 0 by Lemma 2.16, the function q(x) attains its minimum value at
pα 2 − 2β (pα − β )
x0 = .
2(pα − β )
pα 2 − 2pαβ + pβ 2 = p(α − β )2 .
It follows that x0 ≥ 0 and so h(x) ≥ h(x0 ) ≥ 0 for all real x (not just nonnegative x).
From this, we deduce that the discriminant of h(x) cannot be positive. Therefore,
(pα − 2β )2 ≤ 0.
Therefore, pα = 2β .
2.2 Some Integral Operators 49
Lemma 2.19. Suppose 2 < p < ∞ and Pα is bounded on L p (C, dλβ ). Then
pα = 2β .
Proof. If Pα is a bounded operator on L p (C, dλβ ), then Pα∗ is also bounded on
Lq (C, dλβ ), where 1 < q < 2 and 1/p + 1/q = 1. It follows from (2.5) that there
exists a positive constant C, independent of f , such that
q
e(β −α )|z|2 eα zw̄ f (w)e(α −β )|w| dλα (w) dλβ (z)
2
C
C
where g ∈ Lq (C, dλβ −q(β −α )). Recall from Lemma 2.16 that
β − q(β − α ) > 0.
for all g ∈ Lq (C, dλβ −q(β −α )). So the operator Pα is bounded on Lq (C, dλβ −q(β −α )).
Since 1 < q < 2, it follows from Lemma 2.18 that
When 1 < p < ∞, that (b) implies (c) follows from Lemmas 2.18 and 2.19, and
that (a) implies (b) is still obvious.
So we assume 1 < p < ∞ and proceed to show that condition (c) implies (a). We
do this with the help of Schur’s test (Lemma 2.14).
Let 1/p + 1/q = 1 and consider the positive function
h(z) = eδ |z| ,
2
z ∈ C,
If δ satisfies
α > qδ , (2.7)
then it follows from Corollary 2.5 that
α
|eα zw̄ | e−(α −qδ )|w| dA(w)
2
I(z) =
π C
α
= |eα zw̄ | dλα −qδ (w)
α − qδ C
α
eα |z| /4(α −qδ ) .
2 2
=
α − qδ
If we choose δ so that
α2
= qδ , (2.8)
4(α − qδ )
then we obtain
α
H(z, w)h(w)q dλβ (w) ≤ h(z)q (2.9)
C α − qδ
for all z ∈ C.
We now consider the integrals
J(w) = H(z, w)h(z) p dλβ (z), w ∈ C.
C
If δ satisfies
β − pδ > 0, (2.10)
2.2 Some Integral Operators 51
π C
α (β −α )|w|2 α 2 |w|2 /4(β −pδ )
= e e
β − pδ
α
e[(β −α )+α /4(β −pδ )]|w| .
2 2
=
β − pδ
If we choose δ so that
α2
β −α + = pδ , (2.11)
4(β − pδ )
then we obtain
α
H(z, w)h(z) p dλβ (z) ≤ h(w) p (2.12)
C β − pδ
for all w ∈ C. In view of Schur’s test and the estimates in (2.9) and (2.12), we
conclude that the operator Qα would be bounded on L p (C, dλβ ) provided that
we could choose a real δ to satisfy conditions (2.7), (2.8), (2.10), and (2.11)
simultaneously.
Under our assumption that pα = 2β , it is easy to verify that condition (2.8) is
the same as condition (2.11). In fact, we can explicitly solve for qδ and pδ in (2.8)
and (2.11), respectively, to obtain
α 2β − α
qδ = , pδ = .
2 2
The relations pα = 2β and 1/p + 1/q = 1 clearly imply that the two resulting δ ’s
above are consistent, namely,
α 2β − α
δ= = . (2.13)
2q 2p
Also, it is easy to see that the above choice of δ satisfies both (2.7) and (2.10). This
completes the proof of the theorem.
Theorem 2.21. If 1 ≤ p < ∞ and pα = 2β , then
|Pα f | p dλβ ≤ |Qα f | p dλβ ≤ 2 p | f | p dλβ
C C C
Proof. With the choice of δ in (2.13), the constants in (2.9) and (2.12) both reduce
to 2. Therefore, Schur’s test tells us that, in the case when 1 < p < ∞, the norm of
Qα on L p (C, dλβ ) does not exceed 2.
When p = 1, the desired estimate follows from Fubini’s theorem and
Corollary 2.5.
Corollary 2.22. For any α > 0 and 1 ≤ p ≤ ∞, the operator Pα is a bounded
p p p
projection from Lα onto Fα . Furthermore, Pα f p,α ≤ 2 f p,α for all f ∈ Lα .
Proof. The case 1 ≤ p < ∞ follows from Theorem 2.21. The case p = ∞ follows
from Corollary 2.5.
2.3 Duality of Fock Spaces 53
It follows easily from the usual duality of L p spaces that for any 1 ≤ p < ∞, we have
(Lαp )∗ = Lqβ , where 1/p + 1/q = 1, α and β are any positive parameters, and the
duality pairing is given by
γ
f (z)g(z)e−γ |z| dA(z).
2
f , gγ =
π C
where γ = αβ is the geometric mean of α and β .
Proof. First, assume that g ∈ Fβq and F is defined by
γ
f (z)g(z)e−γ |z| dA(z).
2
F( f ) = lim
R→∞ π |z|<R
We proceed to show that F gives rise to a bounded linear functional on Fαp . To avoid
the use of limits all over the place, we appeal to Lemma 2.11 and further assume
that g is a finite linear combination of kernel functions.
If f (z) = eγ za for some a ∈ C, then by the reproducing property of the kernel
functions Kγ (z, w) and Kα (z, w), we have
γ
f (z)g(z)e−γ |z| dA(z),
2
g(a) =
π C
and
α γ
g(a) = g a
β α
α α
eα (γ a/α )z g
z e−α |z| dA(z)
2
=
π
C β
α α
z e−α |z| dA(z).
2
= f (z) g
π C β
54 2 Fock Spaces
Therefore,
α α
z e−α |z| dA(z).
2
f g dλγ = f (z)g (2.14)
C π C β
This shows that
α α
z e−α |z| dA(z)
2
F( f ) = f (z)g
π C β
α
α α 2
− α2 |z|2
= f (z)e g z e− 2 |z| dA(z)
π C β
N
f (z) = ∑ ck eγ zak ,
k=1
where f is any finite linear combination of kernel functions, and C and C are
positive constants. This shows that F defines a bounded linear functional on Fαp .
Next, assume that F : Fαp → C is a bounded linear functional. Define a function
g on the complex plane by
g(w) = Fz eγ zw .
It is easy to show that g is entire. We are going to show that g ∈ Fβq and F( f ) =
p
f , gγ for all f in a dense subset of Fα .
To show that g ∈ Fβ , we need to show that the function g(w)e−β |w| /2 is in
q 2
It suffices for us to show that Φ defines a bounded linear functional on the space
L p (C, dA). Without loss of generality, we may assume that h has compact support
in C. In this case, the integral
h(w)eγ zw e−β |w|
2 /2
dA(w)
C
2.3 Duality of Fock Spaces 55
where
α α 2
ϕ (z) = h z e 2 |z| .
β
Since h ∈ L p (C, dA) is equivalent to ϕ ∈ Lαp and since the projection Pα maps Lαp
boundedly into Fαp , we conclude that
π
|Φ (h)| ≤ FPα (ϕ ) p,α ≤ Ch,
β
where h denotes the usual norm in L p (C, dA). This shows that the function g is
in Fβq .
Finally, if f (z) = eγ za for some a ∈ C, then by the remarks immediately following
this proof and the reproducing property in Fγ2 ,
γ
eγ za g(z)e−γ |z| dA(z) = g(a) = F( f ).
2
f , gγ = lim
R→∞ π |z|<R
for all f ∈ Fαp and g ∈ Fβq . In particular, the limit on the left-hand side of (2.16)
exists for all f ∈ Fαp and g ∈ Fβq .
Alternatively, the identity in (2.16) can be proved with the help of Taylor
expansions. Details are left to the interested reader. We now consider the case of
small exponents.
56 2 Fock Spaces
Theorem 2.24. Suppose 0 < p ≤ 1 and β > 0. Then the dual space of Fαp can be
identified with Fβ∞ under the integral pairing
γ
f (z)g(z)e−γ |z| dA(z),
2
f , gγ = lim
R→∞ π |z|<R
where γ = αβ and the limit above always exists.
Proof. First, assume that g ∈ Fβ∞ and F is defined by F( f ) = f , gγ . To show that
p
F extends to a bounded linear functional on Fα , we use (2.16) to rewrite
α
f (z)ϕ (z)e−α |z| dA(z)
2
F( f ) =
π C
α α 2
α 2
= f (z)e− 2 |z| ϕ (z)e− 2 |z| dA(z),
π C
where
α
ϕ (z) = g z
β
is in Fα∞ . It follows from this and the embedding in Theorem 2.10 (and its proof)
that
2
|F( f )| ≤ 2ϕ ∞,α f 1,α ≤ ϕ ∞,α f p,α .
p
So F extends to a bounded linear functional on Fαp , and an approximation argument
shows that the limit in the statement of the theorem always exists.
Next, suppose that F is a bounded linear functional on Fαp . As in the proof of
Theorem 2.23, we consider the function g defined on C by
g(w) = Fz eγ zw .
It follows from the boundedness of F on Fαp and the integral formula in Corollary 2.5
that
pα F p
|eγ zw e−α |z|
2 /2
|g(w)| p ≤ | p dA(z)
2π C
pα F p
|e pγ zw |e−pα |z|
2 /2
= dA(z)
2π C
= F p e pβ |w|
2 /2
.
Finally, as in the proof of Theorem 2.23, we have F( f ) = f , gγ for all functions
f of the form
N
f (z) = ∑ ck eγ zuk .
k=1
Since the set of functions of the above form is dense in Fαp , we have completed the
proof of the theorem.
Setting β = α in Theorems 2.23 and 2.24, we obtain the following special case.
Corollary 2.25. If 1 ≤ p < ∞, then the dual space of Fαp can be identified with Fαq
under the integral pairing f , gα , where 1/p + 1/q = 1. If 0 < p < 1, then the dual
space of Fαp can be identified with Fα∞ under the integral pairing f , gα .
It is interesting to observe that under the same integral pairing f , gα , the dual
space of each Fαp , 0 < p ≤ 1, can be identified with the same space Fα∞ . This differs
from the traditional Hardy and Bergman space theories.
Theorem 2.26. Suppose β > 0 and γ = αβ . Then the dual space of fα∞ can be
identified with Fβ1 under the integral pairing f , gγ .
| f , gγ | ≤ C f ∞,α
for all f ∈ Fα∞ . For any positive integer n, consider the function:
n
fn (z) = f z , z ∈ C.
n+1
58 2 Fock Spaces
It is clear that f ∈ Fα∞ implies that each fn ∈ fα∞ with fn ∞,α ≤ f ∞,α for all n.
Now,
γ
f (w)Fz eγ zw e−γ |w| dA(w)
2
f , gγ = lim
R→∞ π |w|<R
γ
fn (w)Fz eγ zw e−γ |w| dA(w)
2
= lim lim
n→∞ R→∞ π |w|<R
γ γ zw −γ |w|2
= lim F fn (w)e e dA(w)
n→∞ π C
= lim F( fn ).
n→∞
Since |F( fn )| ≤ F fn ∞,α ≤ F f ∞,α for all n, we conclude that | f , gγ | ≤
F f ∞,α for all f ∈ Fα∞ . This shows that g ∈ Fβ1 and completes the proof of the
theorem.
2.4 Complex Interpolation 59
We assume that the reader is familiar with the basic theory of complex interpolation,
including the complex interpolation of L p spaces. The book [250] provides an
elementary introduction to the subject. We will begin with the following well-known
interpolation theorem of Stein and Weiss.
Theorem 2.27. Suppose w, w0 , and w1 are positive weight functions on the complex
plane. If 1 ≤ p0 ≤ p1 ≤ ∞ and 0 ≤ θ ≤ 1, then
This result is very useful and widely known. See [216] for a proof.
Recall that Lαp is the space of Lebesgue measurable functions f on the complex
plane such that the function f (z)e−α |z| /2 is in L p (C, dA). The norm of f in Lα
2 p
p p
was defined in Sect. 2.1. With the inherited norm, Fα is the closed subspace of Lα
consisting of entire functions.
Specializing to exponential weights, we obtain the following special case of the
Stein–Weiss interpolation theorem.
Corollary 2.28. Suppose 1 ≤ p0 ≤ p1 ≤ ∞ and 0 ≤ θ ≤ 1. Then for any positive
weight parameters α0 and α1 , we have
p
Lα00 , Lαp11 θ
= Lαp ,
where
1 1−θ θ
= + , α = α0 (1 − θ ) + α1θ .
p p0 p1
Proof. Since Lαp = L p (C, dλ pα /2), it follows from the Stein–Weiss interpolation
theorem that
p0 p1
Lα1 , Lα2 θ = L p0 (C, dλ p0 α1 /2 ), L p1 (C, dλ p1 α2 /2 ) θ
= L p (C, dλ pα /2) = Lαp ,
where
1 1−θ θ
= + , α = α0 (1 − θ ) + α1θ .
p p0 p1
This proves the desired result.
60 2 Fock Spaces
Although Fαp is a closed subspace of Lαp , the Fock spaces interpolate in a way
that is much different from the containing spaces Lαp . In some sense, the Lebesgue
spaces Lαp interpolate “arithmetically,” while the Fock spaces Fαp interpolate “geo-
metrically.” We begin with the case when the weight parameter α is fixed.
Theorem 2.29. Suppose 1 ≤ p0 ≤ p1 ≤ ∞ and 0 ≤ θ ≤ 1. Then
p0 p1 p
Fα , Fα θ = Fα ,
where
1 1−θ θ
= + .
p p0 p1
Proof. The inclusion
p0 p1 p
Fα , Fα θ ⊂ Fα
p
follows from the definition of complex interpolation, the fact that each Fα k is a
pk p0 p1
closed subspace of Lα , and the fact that Lα , Lα θ = Lαp .
pOn the
other hand, if f ∈ Fαp ⊂ Lαp , then f is entire, and it follows from
Lα0 , Lα1 θ = Lαp that there exist a function F(z, ζ ) (z ∈ C and 0 ≤ Re ζ ≤ 1) and a
p
where
1 1−θ θ
= + , α = α01−θ α1θ .
p p0 p1
Proof. For any ζ ∈ C, consider the dilation operator Sζ defined by
(ζ −θ )/2
α0
Sζ f (z) = f z .
α1
Proof. It is easy to see that a necessary condition for Pα Lβp ⊂ Fγp , 1 ≤ p ≤ ∞, is that
2α > β . So for the rest of the proof, we always assume that 2α > β .
If α 2 /γ ≤ 2α − β , it follows from Corollary 2.5 that Pα maps L∞ ∞
β into Fγ .
Similarly, it follows from Fubini’s theorem and Corollary 2.5 that Pα maps L1β into
Fγ1 . By complex interpolation, Pα maps Lβp into Fγp for all 1 ≤ p ≤ ∞.
If α 2 /γ = 2α − β and f ∈ Fγp , then the function
α α
z e(β −α )|z|
2
g(z) = f
γ γ
Recall from Lemma 2.11 that the set of finite linear combinations of kernel functions
is dense in Fαp , 0 < p < ∞. In this section, we improve upon this result. We show
that every function in Fαp can actually be decomposed into an infinite series of kernel
functions.
We begin with a basic estimate for integral averages of functions in Fock spaces.
Lemma 2.32. For any positive parameters α , p, and R, there exists a positive
constant C = C(p, α , R) such that
2 p C 2 p
f (a)e−α |a| /2 ≤ 2 f (z)e−α |z| /2 dA(z)
r B(a,r)
for all entire functions f , all complex numbers a, and all r ∈ (0, R]. Here, B(a, r) is
the Euclidean disk centered at a with radius r.
Proof. Let I denote the integral above. Then
| f (z)| p e−pα |z|
2 /2
I= dA(z)
B(a,r)
| f (w + a)| p e−pα |w+a|
2 /2
= dA(w)
|w|<r
| f (w + a)e−α wa| p e−pα (|w|
2 +|a|2 )/2
= dA(w).
|w|<r
Writing the integral in polar coordinates and using the subharmonicity of the
function | f (w + a)e−α wa | p , we obtain
e−pα (|w|
2 +|a|2 )/2
I ≥ | f (a)| p dA(w)
|w|<r
r
te−pα (t
2 +|a|2 )/2
= 2π | f (a)| p dt
0
r2
= π | f (a)e−α |a| e−pα s/2 ds
2 /2
|p
0
2π
(1 − e−pα r /2 )| f (a)e−α |a| /2 | p .
2 2
=
pα
is a square lattice in the complex plane. The fundamental region of rZ2 , if we ignore
the boundary points, is the square
Moreover, the use of half-open and half-closed squares makes the decomposition
above a disjoint union. Thus,
C
f (z) d μ (z) = ∑ f (z) d μ (z),
w∈rZ2 Sr +w
C
f (z) d μ (z) ≤ ∑ f (z) d μ (z) ≤ N f (z) d μ (z)
w∈rZ2 Qr +w C
This is of course a unit vector in Fα2 . The following result is a pleasant surprise.
p
Lemma 2.33. Each ka is also a unit vector in Fα , where 0 < p ≤ ∞.
Proof. It follows from the definition of the norm in Fαp and the reproducing formula
in Fp2α /2 that
pα 2 p
ka (z)e− 2 α |z| dA(z)
1
ka pp,α =
2π C
pα − 1 pα |a|2 pα za 2 − pα |z|2
= e 2 e 2 e 2 dA(z)
2π C
pα pα
= e− 2 |a| 2 |a|
2 2
e = 1,
2.5 Atomic Decomposition 65
which proves the desired result for 0 < p < ∞. For p = ∞, observe that
α α
|ka (z)|e− 2 |z| = e− 2 |z−a| .
2 2
It follows that
α
sup |ka (z)|e− 2 |z| = 1,
2
z∈C
f pp,α ≤ ∑ |cw | p .
w∈rZ2
w∈rZ2
After interpolating between p = 1 and p = ∞, we have now shown that, for all
p ∈ (0, ∞] and {cw } ∈ l p , the function f given by (2.17) is in Fαp . Furthermore,
where C = C(p, α , r) is a positive constant and the infimum is taken over all
sequences {cw } that give rise to the representation of f in (2.17). It is interesting
to note that this part of the proof works for any positive r.
To prove the other part of the theorem, we assume that 0 < r < 1 and consider
the linear operator Tr defined on the space of entire functions as follows:
α α α
∑ eα zw− 2 |w|
2 +α iIm (wu)
f (u)e− 2 |u|
2
Tr f (z) = dA(u).
π w∈rZ2 Sr +w
It follows that
α
Dr f (z) =
π ∑ f (u)H(z, w, u) dA(u), (2.18)
w∈rZ2 Sr +w
where
α α
α 2
H(z, w, u) = eα zw− 2 |w| − eα zu− 2 |u| −α iIm (wu) e− 2 |u| +α iIm (wu) .
2 2
α
≤ Cre 4 |z−w|
2
for all u ∈ Sr , where C is a positive constant that only depends on α . Here, we used
the additional assumption that 0 < r < 1. It follows that there exists another positive
constant C, independent of r and z, such that
α
∑ e− 4 |z−w|
2
Jr (z) ≤ Cr3
w∈rZ2
α α 2 α
α
2 2
e− 4 |z−w| = e− 4 |z| e 4 wz e− 8 |w| ,
2
an application of Lemma 2.32 shows that there is yet another positive constant C,
independent of z and r, such that
α
∑ e− 4 |z−u| dA(u)
2
Jr (z) ≤ Cr
w∈rZ2 Sr +w
α
e− 4 |z−u| dA(u)
2
= Cr
C
α 4π Cr
e− 4 |u| dA(u) =
2
= Cr .
C α
68 2 Fock Spaces
This shows that there exists another positive constant C, independent of r, such that
To estimate the norm of Dr on Fα1 , first note that |Dr f (z)| is less than or equal to
α α zw− α2 |w|2 α α
∑ − eα zu− 2 |u| −α iIm (wu) | f (u)|e− 2 |u| dA(u).
2 2
e
π w∈rZ2 Sr +w
π w∈rZ2 Sr +w
where
α 2 α α
e− 2 |z| eα zw− 2 |w| − eα zu− 2 |u| −α iIm (wu) dA(z)
2 2
H(w, u) =
C
− α2 |z−w|2 α
− e− 2 |z−u| +α iIm (z−w)(u−w) dA(z)
2
= e
C
− α2 |z|2 α
− e− 2 |z−(u−w)| +α iIm z(u−w) dA(z)
2
= e
C
α 2 α 2
= e− 2 |z| 1 − eα z(u−w)− 2 |u−w| dA(z).
C
It is now clear that we can find a positive constant C = C(α ) such that H(w, u) ≤ Cr
for all w and u. It follows that
α α
|Dr f (z)|e− 2 |z| dA(z) ≤ Cr | f (u)|e− 2 |u| dA(u)
2 2
C C
2.5 Atomic Decomposition 69
By (2.19) and (2.20), if r is sufficiently small, then Dr ∞,α < 1 and Dr 1,α < 1.
By complex interpolation, we also have Dr p,α < 1 for all 1 ≤ p ≤ ∞. This shows
p
that if r is small enough, the operator Tr is invertible on Fα for all 1 ≤ p ≤ ∞. When
Tr is invertible and f ∈ Fα , we can write f = Tr g with g = Tr−1 f and obtain the
p
A simple argument with the help of Lemma 2.32 shows that the above sequence
{cw } is in l p whenever g ∈ Fαp . This completes the proof of the theorem in the case
1 ≤ p ≤ ∞.
We will complete the proof of the case 0 < p < 1 after we have proved the
following three lemmas.
Lemma 2.35. Suppose 0 < r < 1, 0 < p ≤ 1, and m is a nonnegative integer. For
any entire function f , we define a sequence
by
α α (z − w)k α 2
eα iIm z(z−w)− 2 |z−w| f (z)e− 2 |z| dA(z).
2
(S f )w,k =
π Sr +w k!
Then S maps Fαp boundedly into l p .
Proof. For any w ∈ rZ2 , z ∈ Sr + w, and 1 ≤ k ≤ m, we have
αp α 2 (z − w)
k
|(S f )w,k | p = p eα iIm z(z−w)− 2 |z−w| f (z)e−α (z−w)w
π Sr +w k!
p
α 2 α
e− 2 |z−w| − 2 |w| +α iIm (z−w)w dA(z)
2
p
− α2 |w|2 −α (z−w)w
≤ C1 r e
pk
f (z)e dA(z)
Sr +w
pα
C1 r p(2+k) e− 2 |w| sup{| f (z)e−α (z−w)w | p
2
≤ : z ∈ Sr + w}
p
pα
≤ C2 r p(2+k)−2 e− 2 |w| f (z)e−α (z−w)w dA(z)
2
Qr +w
α 2 α 2 p
= C2 r p(2+k)−2 f (z)e 2 |z−w| − 2 |z| dA(z)
Qr +w
α 2 p
≤ C3 r p(2+k)−2 f (z)e− 2 |z| dA(z).
Qr +w
70 2 Fock Spaces
Thus p
m
α 2 p α
f (z)e− 2 |z| ≤ ∑ ∑ |cw,k |k [α |z − w|] pk e− 2 |z−w|
2
,
w∈rZ2 k=0
and hence
m p
α 2 p α
f (z)e− 2 |z| dA(z) ≤ ∑ ∑ |cw,k | p |α z|k e− 2 |z|
2
dA(z)
C w∈rZ2 k=0 C
m
≤C ∑ ∑ |cw,k | p .
w∈rZ2 k=0
zk = ∑ cw kw (z),
w∈rZ2
where {cw } ∈ l p .
2.5 Atomic Decomposition 71
zk = ∑ cw kw (z),
w∈ρ Z2
where {cw } ∈ l ∞ . For w ∈ ρ Z2 , we can write w = ρ (m + in) for some integers m and
n. Since
α
kw (z) = eα zw− 2 |w| ,
2
It follows that
r k
z = ∑ cw kw (z),
ρ w ∈rZ2
where
cw = cw e−(ρ
2 −r2 )(n2 +m2 )
where
α iIm u(u−w)− α2 |u−w|2
m
[α (z − w)(u − w)]k
G = ku (z) − e ∑ k!
kw (z).
k=0
∞
(α r|z − w|)k
|G| ≤ |kw (z)| ∑ k!
,
k=m+1
72 2 Fock Spaces
π p w∈rZ 2 k=m+1 k! Sr +w
π p w∈rZ 2 Sr +w
where
p
∞
− p2α |z−w|2 (α r|z − w|)k
C(w) = e
C
∑ k!
dA(z)
k=m+1
p
∞
− p2α |z|2 (α r|z|)k
= e
C
∑ k! dA(z)
k=m+1
p
∞
pα 2 (α |z|)k
≤ r(m+1)p e− 2 |z|
C
∑ k! dA(z)
k=m+1
pα 2
≤r (m+1)p
e− 2 |z| +pα |z| dA(z).
C
Thus,
α 2 p
(I − TS) f (z)e− 2 |z| dA(z)
C
is less than or equal to
α 2 p
Cr(m+1)p+2p−2 ∑ f (u)e− 2 |u| dA(u).
w∈rZ2 Qr +w
So we can find another constant C > 0, independent of r ∈ (0, 1), such that
Since
2 2(1 − p) 2
m+3− > − 1 + 3 − = 0,
p p p
we see that there exists some r0 ∈ (0, 1) such that I − TS p,α < 1 whenever r ∈
(0, r0 ). This shows that the operator T S is invertible on Fαp whenever r ∈ (0, r0 ).
Consequently, for any r ∈ (0, r0 ), the operator T is onto, and so every function
f ∈ Fαp can be written as
m
α
∑ ∑ cw,k (z − w)k e−α zw− 2 |w| .
2
f (z) = (2.21)
w∈rZ2 k=0
u∈rZ2 ,|u|≤N
74 2 Fock Spaces
in Fαp is less than δ for all 0 ≤ k ≤ m and w ∈ rZ2 . Define an operator Ar on Fαp by
α
∑ ∑ cu,k eα iIm (wu) eα z(w+u)− 2 |w+u|
2
Ar f (z) = cw,k
w∈rZ2 ,0≤k≤m u∈rZ2 ,|u|≤N
for all f ∈ Fαp . If we choose δ such that Cδ p < 1, then I − Ar p,α < 1, and so the
operator Ar is surjective on Fαp . Since w + u ∈ rZ2 whenever w ∈ rZ2 and u ∈ rZ2 ,
the proof of Theorem 2.34 is now complete.
2.6 Translation Invariance 75
In this section, we consider the action of translations on Fock spaces and determine
three spaces that are unique under such actions: the space Fα∞ is maximal among
translation invariant Banach spaces of entire functions, the space Fα1 is minimal
among translation invariant Banach spaces of entire functions, and the space Fα2 is
the only Hilbert space of entire functions invariant under translations.
For any point a ∈ C, we define three analytic self-maps of the complex plane as
follows:
ta (z) = z + a, τa (z) = z − a, ϕa (z) = a − z.
The map ta is naturally called the translation by a, and it is clear that τa = t−a = ta−1 .
The map ϕa is the composition of the translation ta with the reflection z → −z. Note
that ϕa is its own inverse.
When making a change of variables, observe that
f ◦ ta (z) dλα (z) = f ◦ ϕa (z) dλα (z)
C C
α
f (w)e−α |a−w| dA(w)
2
=
π C
= f (w)|ka (w)|2 dλα (w).
C
Similarly,
f ◦ τa (z)|ka (z)|2 dλα (z) = f ◦ ϕa (z)|ka (z)|2 dλα (z)
C C
= f (z) dλα (z),
C
while
f ◦ ta (z)|ka (z)|2 dλα (z) = f (z + 2a) dλα (z).
C C
See Corollary 2.4. These are some of the subtle differences that can easily be
overlooked.
We can use τa and ϕa to define certain unitary operators on Fα2 . Although there is
an obvious temptation to use only one of these maps in the book, we have found that
there are situations in which one choice is more convenient than the other. Therefore,
we are going to use both in the book.
76 2 Fock Spaces
for all a ∈ C and f ∈ Fαp . Furthermore, both Wa and Ua are invertible on Fαp with
Wa−1 = W−a and Ua−1 = Ua . Consequently, Wa and Ua are both unitary operators on
Fα2 with Wa∗ = W−a and Ua∗ = Ua .
Proof. It is easy to check that
α α
e− 2 |z| |Wa f (z)| = e− 2 |z−a| | f (z − a)|,
2 2
and
α α
e− 2 |z| |Ua f (z)| = e− 2 |a−z| | f (a − z)|.
2 2
The identities
Wa f p,α = Ua f p,α = f p,α
then follow from a change of variables. See Corollary 2.4.
To see that Wa is invertible with Wa−1 = W−a , take any f ∈ Fαp and note that
α
W−aWa f (z) = e−α az− 2 |a| (Wa f )(z + a)
2
α α
= e−α az− 2 |a| eα a(z+a)− 2 |a| f (z + a − a)
2 2
= f (z).
A similar argument shows that Ua is invertible with Ua−1 = Ua . This completes the
proof of the proposition.
Although the operators Wa and Ua behave similarly in many situations, there are
sometimes reasons to pick one over the other. For example, the operators Wa almost
have a semigroup property with respect to a, while the operators Ua are all self-
adjoint. In particular, we can use the Weyl operators to obtain the following unitary
representation of the Heisenberg group. Recall that another unitary representation
was given in Chap. 1 based on Weyl pseudodifferential operators.
2.6 Translation Invariance 77
This shows that (a, θ ) → eiθ Wa is a group embedding of H into the group of unitary
operators on Fα2 .
In the rest of this section, we work with the Weyl unitary operators Wa . A similar
theory can be developed with the unitary operators Ua , which is left to the reader as
an exercise.
Proposition 2.40. The Fock space Fα∞ is maximal in the sense that if X is any
Banach space of entire functions with the following properties:
(a) Wa f X = f X for all a ∈ C and f ∈ X,
(b) the point evaluation f → f (0) is a bounded linear functional on X,
then X ⊂ Fα∞ and the inclusion is continuous.
Proof. Condition (a) implies that Wa f ∈ X for every f ∈ X and every a ∈ C.
Combining this with condition (b), we see that for every a ∈ C, the point evaluation
f → f (a) is also a bounded linear functional on X, and
α
e− 2 |a| | f (a)| = |W−a f (0)| ≤ CW−a f X = C f X ,
2
Since each kzn belongs to X and ∑ |cn | < ∞, we conclude that f ∈ X with
∞ ∞
f X ≤ ∑ |cn |kzn X = C ∑ |cn |,
n=1 n=1
where C = 1X > 0. Taking the infimum over all sequences {cn } satisfying (2.23),
we obtain another constant C > 0 such that
f X ≤ C f Fα1 , f ∈ Fα1 .
and the convergence is uniform when z and w are restricted to compact subsets of C.
In particular, the series representation for KH (z, w) in (2.24) is independent of the
choice of the orthonormal basis {en }.
It is easy to see from condition (a) and the proof of Proposition 2.38 that each Wa
is a unitary operator on H. Fix any a ∈ C and let σn = Wa en , n ≥ 1. Then {σn } is
also an orthonormal basis of H. Therefore, by (2.24), we have
∞
KH (z, w) = ∑ σn (z) σn (w)
n=1
∞
= ka (z) ka (w) ∑ en (z − a) en(w − a)
n=1
The classical maximum principle asserts that if f is an entire function and | f (z)| ≤
M for all |z| = R, then | f (z)| ≤ M for all |z| ≤ R. The purpose of this section is to
prove the following version of the maximum principle for Fock spaces.
Theorem 2.43. For any α > 0 and p ≥ 1, there exists a positive radius R = R(α , p)
such that f p,α ≤ g p,α for all entire functions f and g satisfying
Proof. Without loss of generality, we may assume that g ∈ Fαp . Otherwise, the
desired result is obvious. Under this assumption, we also have
| f | p dλα ≤ |g| p dλα < ∞,
|z|≥R |z|≥R
We fix some R > 0 and assume that | f (z)| ≤ |g(z)| for all R ≤ |z| < ∞.
We will try to compare I(r, | f | p − |g| p ) for 0 < r < R to I(ρ , |g| p − | f | p ) for
R < ρ < ∞. To this end, we let ω (z) = f (z)/g(z), which is analytic and has modulus
less than or equal to 1 in the region R < |z| < ∞. We may assume that |ω (z)| <
1 for all R < |z| < ∞. In fact, if |ω (z0 )| = 1 for some R < |z0 | < ∞, then by the
classical maximum modulus principle, the analytic function ω on R < |z| < ∞ must
be constant, which would then imply that f and g differ by a constant multiple in
the whole complex plane, from which the desired result clearly follows.
For any ρ ∈ (R, ∞), pick a point ζ (ρ ) such that |ζ (ρ )| = ρ and
We may assume that f is not identically 0, for otherwise the desired result is trivial.
Thus, 0 < |ω (ζ (ρ ))| < 1 for all ρ ∈ (R, ∞). To simplify notation, let us write ωρ =
ω (ζ (ρ )).
Since p ≥ 1, it follows from elementary calculus that
for all x ≥ 0 and y ≥ 0. We deduce from the second inequality in (2.25) that for any
0 ≤ r ≤ R < ρ < ∞, we have
p|ω | p−1 1 1 + |ω | 2
≤ = < .
1 − |ω | p 1 − |ω | 1 − |ω | 2 1 − | ω |2
Therefore,
for all 0 ≤ r ≤ R < ρ < ∞. Fix ρ and integrate both sides of (2.27) over [0, R] against
the measure re−α r dr. The result is
2
1 − e− α R
2
for all R < ρ < ∞. Divide both sides of (2.28) by γ (ρ ) and integrate both sides over
(R, ∞) against the measure ρ e−αρ dρ . The result is
2
(| f | p − |g| p) dλα ≤ CR (|g| p − | f | p ) dλα ,
|z|≤R |z|≥R
2.7 A Maximum Principle 83
where
1 − e− α R
2
CR = ∞ .
ρ e−αρ
2
α dρ
R γ (ρ )
If R is a positive radius such that CR < 1, then the integral
J= (| f | p − |g| p) dλα
C
= 0,
Since
|a − b| d(a, b) 1 − |b|2
=
1 − |a|2 1 − d 2(a, b) 1 − |a|2
for all a and b in the unit disk, we see that for all z with |z| = ρ ,
|ω (z) − ωρ | d(ω (z), ωρ ) 1 − | ω ρ |2
=
1 − |ω (z)| 2
1 − d (ω (z), ωρ ) 1 − |ω (z)|2
2
It follows that
The function H(z) = ω (R/z) is analytic from the punctured disk 0 < |z| < 1 into
the unit disk. Since H is bounded near z = 0, it has a removable singularity at z = 0.
Thus, we can think of H as analytic self-maps of the unit disk. By the classical
Schwarz lemma, we have
It follows that
d(R/z, R/ζ (ρ ))
γ (ρ ) ≤ sup .
|z|=ρ 1 − d 2(R/z, R/ζ (ρ ))
2Rρ
sup d(R/z, R/ζ (ρ )) = d(−R/ζ (ρ ), R/ζ (ρ )) = .
|z|=ρ ρ 2 + R2
2Rρ
γ (ρ ) ≤ , R < ρ < ∞.
ρ 2 − R2
2R(1 − e−α R )
2
CR ≤ ∞ .
(ρ 2 − R2)e−αρ dρ
2
α
R
The quotient above tends to 0 as R → 0+ . Therefore, CR < 1 for all sufficiently small
positive radius R. This completes the proof of the theorem.
If 0 < p < 1, the inequalities in (2.25) are replaced by
for all 0 ≤ r ≤ R < ρ < ∞. So in this case, we need to consider the function
p|ωρ | p−1 |ω (z) − ωρ |
γ (ρ ) = max : |z| = ρ , R < ρ < ∞.
1 − |ω (z)| p
Note that the function γ depends on f and g. We just need to bound γ from above by
a function that is independent of f and g. By the left inequality in (2.30), we have
Therefore, we just need to bound |ωρ | from below by a positive function that is
independent of f and g. But this is impossible, for we may have a situation like
f (z) = g(z)/N, where N is large; in this case, we have H = 1/N, and we can choose
N to be arbitrarily large.
86 2 Fock Spaces
2.8 Notes 87
2.8 Notes
There are two ways to define the Fock spaces. One way is to consider subspaces
L p (C, dλα ) consisting of entire functions. This would be similar to the definitions of
the more classical Hardy and Bergman spaces. It turns out that this is not a good way
to define the Fock spaces. The seemingly cumbersome definition of Fαp as the space
of entire functions f such that f (z)e−α |z| /2 belongs to L p (C, dA) will make the
2
statements and proofs of many results much easier and more convenient later on.
The constant α in Fαp is not essential in our theory. No generality is lost if
we choose to develop the theory with a particular choice of α , say α = 1. This
weight parameter plays the role of Planck’s constant in mathematical physics, and
it provides us with an extra level of freedom that is useful in several situations.
Although the Fock space Fα2 is a central subject in quantum physics, this book is
focused on purely mathematical analysis on Fock spaces. No serious effort is made
to show any connections or applications to physics. We refer the interested reader to
books such as [177] for applications of the Fock space in physics.
The characterization of the boundedness of Pα and Qα on L p spaces was obtained
in [74], where more precise norm estimates can also be found. See [96] for an
even more elaborate study of similar integral operators. The boundedness of the
projection Pα on Lαp for 1 ≤ p ≤ ∞ can be found in [138]. The papers [214] and
[217] also study the boundedness of Pα and Qα on L p spaces.
The study of the Heisenberg group is a small industry by itself. This is especially
so in quantum physics and harmonic analysis, where the connection of Fock spaces
to the Heisenberg group is evident. But we will not use the Heisenberg group in any
way other than the special elements Wa in it.
The paper [138] by Janson, Peetre, and Rochberg is a key reference throughout
this book. In particular, the duality, atomic decomposition, and complex inter-
polation for the Fock spaces Fαp , where 1 ≤ p ≤ ∞, were proved in [138]. Our
presentation of the case 0 < p < 1 follows [231] very closely.
The translation invariance of the Fock spaces was first considered in [138], where
it was shown that Fα1 is minimal and Fα∞ is maximal among Banach spaces of entire
functions whose norm is invariant under the action of the Heisenberg group. The
uniqueness of Fα2 among Hilbert spaces of entire functions whose norm is invariant
under the action of Wa was proved in [255].
The version of the maximum modulus principle in Sect. 2.7 was first proved in
[194], based on a technique introduced in [122] to tackle the corresponding problem
for Bergman spaces on the unit disk. That such a maximum modulus principle might
be true for the Bergman space was first conjectured by Korenblum in [141] and was
proved in [117] in the case p = 2 and in [122] when 1 ≤ p < ∞. See [232–239] for
other work concerning Korenblum’s maximum principle.
88 2 Fock Spaces
2.9 Exercises 89
2.9 Exercises
1. Show that the Fock space Fαp is a closed subspace of L p (C, dλ pα /2).
2. Show that Mz , the operator of multiplication by the coordinate function z, is a
densely defined unbounded linear operator on Fα2 . Show that the adjoint of Mz
on Fα2 is essentially the operator of differentiation. More specifically, Mz∗ f (z) =
(1/α ) f (z) for all f ∈ Fα2 .
3. Let 0 < p ≤ ∞, S be a compact subset of C, and k be a positive integer. Show
that there exists a positive constant C such that
12. Let 1 ≤ p < ∞ and let {δn } be any sequence of positive numbers decreasing
to 0. Then there exists a function f (z) = ∑ an zn in Fαp such that
α n 14 − 2p1
an = O n δn , n → ∞.
n!
90 2 Fock Spaces
∞ p
n! 2
∑ |an |
p 1
p
n− 4 + 2 < ∞.
n=0 αn
17. Let 1 < p ≤ 2 with 1/p + 1/q = 1 and f (z) = ∑ an zn is in Fαp . Then
∞ q
n! 2
∑ |an |
q 1
q
n 4 − 2 < ∞.
n=0 αn
∞ q
n! 2
∑ |an |q
q 1
n 4 − 2 < ∞,
n=0 αn
α R2
(R2 − |z|2 )α R −1 dA(z)
2
dAR (z) =
πR 2 α R 2
denote the normalized weighted area measure on the disk B(0, R), where dA is
area measure. For any entire function f , show that
lim | f (z)| p dAR (z) = | f (z)| p dλα (z).
R→∞ B(0,R) C
30. Show that the main atomic decomposition theorem remains valid if we replace
the square lattice rZ2 by any sequence {wk } in the complex plane with the
following properties: C = ∪k B(wk , r), |wk − w j | ≤ r, and |wk − w j | ≥ r/4
whenever k = j. Here, r is any sufficiently small positive radius.
31. Show that harmonic conjugation is a bounded linear operator on Lαp for
1 ≤ p ≤ ∞.
32. Characterize lacunary series in Fαp . See [226].
33. Prove an atomic decomposition for the space fα∞ .
34. Suppose f ∈ Fαp and f (a) = 0. Show that there exists a positive integer N and
at most one more point b such that
α 2 p α 2 p
f pp,α = N f (a)e− 2 |a| + f (b)e− 2 |b| .
35. Prove the analogs of Propositions 2.40, 2.41, and 2.42 when the operators Wa
are replaced by the operators Ua .
36. Suppose ω1 and ω2 are strictly positive and Lebesgue measurable weight
functions on the complex plane. If 1 ≤ p < ∞ and 1/p + 1/q = 1, then
with equal norms, where the duality pairing is given by the integral
f , gω = f (z)g(z) ω (z) dA(z),
C
and 1 1
ω (z) = ω1 (z) p ω2 (z) q
is a geometric mean of ω1 (z) and ω2 (z).
37. Suppose 1 ≤ p < ∞ and 1/p + 1/q = 1. For any positive parameters α and β ,
show that (Lαp )∗ = Lqβ under the integral pairing
γ
f (z)g(z)e−γ |z| dA(z),
2
f , gγ =
π C
g(w) = Fz (eγ zw )
is entire.
p p
40. Suppose 1 ≤ p ≤ ∞. Show that Fα is a complemented subspace of Lα , that is,
p p p p p
there exists a closed subspace Xα of Lα such that Lα = Fα ⊕ Xα . Study the case
when 0 < p < 1.
Chapter 3
The Berezin Transform and BMO
In this chapter, we study the Berezin transform on Fα2 and certain spaces of functions
of bounded mean oscillation (BMO) on the complex plane. We first consider the
Berezin symbol of a bounded linear operator on Fα2 and show that this is a Lipschitz
function in the Euclidean metric. We then consider the Berezin transform of a
function and show that there is a semigroup property with respect to the parameter
α . We also consider the action of the Berezin transform on L p spaces and the
behavior of the Berezin transform when it is iterated.
For every exponent p ∈ [1, ∞), we define a space BMO p of functions of bounded
mean oscillation, based on Euclidean disks of a fixed radius, and study the structure
of these spaces. When 1 < p < ∞, we will show that the Berezin transform of every
function in BMO p is Lipschitz in the Euclidean metric.
As is well known, the Berezin transform is closely related to the notion of
Carleson measures. So we include the discussion of Fock–Carleson measures in
this chapter as well.
Recall that for each z ∈ C, we use kz to denote the normalized reproducing kernel
at z, namely,
α 2
kz (w) = K(w, z)/ K(z, z) = eα wz− 2 |z| .
T(z) = T kz , kz , z ∈ C, (3.1)
where , is the inner product in Fα2 . We are going to call T the Berezin transform
(or sometimes the Berezin symbol) of T . In particular, if T is a bounded linear
operator on Fα2 , then the Berezin transform T is well defined and is actually real
analytic in C.
Proposition 3.1. Let L(Fα2 ) be the Banach space of all bounded linear operators on
Fα2 . Then T → T is a bounded linear mapping from L(Fα2 ) into L∞ (C). Furthermore,
the mapping is one-to-one and order preserving.
Proof. Everything is obvious except the one-to-one part. To see this, assume that
T is a bounded linear operator on Fα2 and that T kz , kz = 0 for all z ∈ C. Then
T Kz , Kz = 0 for all z ∈ C, where Kz (w) = K(w, z). The function F(z, w) =
T Kz , Kw is real analytic on C × C, holomorphic in w, and conjugate holomorphic
in z. Also, F vanishes on the diagonal of C × C. It follows from a well-known
theorem in several complex variables (see [142] for example) that F is identically
zero on C × C. Consequently, T Kz (w) = 0 for all z and w, or T Kz = 0 for all z ∈ C.
Since the set of finite linear combinations of kernel functions is dense in Fα2 , we
conclude that T = 0.
Note that the proof above concerning the one-to-one property of the Berezin
transform works for certain unbounded operators as well. More specifically, if T is
an unbounded linear operator on Fα2 such that its domain contains all finite linear
combinations of kernel functions and T Kz , Kw is real analytic, then T = 0 implies
that T = 0.
Proposition 3.2. If T is compact on Fα2 , then T(z) → 0 as z → ∞.
Proof. It is easy to see that kz → 0 weakly in Fα2 as z → ∞. This gives the desired
result.
It is a classical result in functional analysis that if T is positive and compact on a
Hilbert space H, then there exists an orthonormal set {en } in H and a nonincreasing
sequence {sn } of positive numbers such that
The numbers sn are uniquely determined by T and are called the singular values
of T .
Let T be a positive and compact operator with singular values {sn }, and let
0 < p < ∞. We say that the operator T belongs to the Schatten class S p if the
sequence {sn } belongs to l p . For a more general operator T , we say that it belongs
to the Schatten class S p if |T | = (T ∗ T )1/2 belongs to S p . If {sn } is the sequence of
singular values for |T |, we write
1/p
T S p = ∑ snp .
n
Two special cases are worth mentioning: S1 is called the trace class, and S2 is called
the Hilbert–Schmidt class. We refer the reader to [250] for more information about
the Schatten classes.
Proposition 3.3. If S is a trace-class operator or a positive operator, then
α dA(z).
tr (S) = S(z) (3.2)
π C
Furthermore, a positive operator S belongs to the trace class if and only if the
integral in (3.2) converges.
Proof. First, assume that S is positive, say S = T 2 for some T ≥ 0. Then for any
orthonormal basis {en }, it follows from Fubini’s theorem that
∞ ∞ ∞
tr (S) = ∑ Sen, en α = ∑ Ten22,α = ∑ |Ten (z)|2 dλα (z)
n=1 n=1 n=1 C
∞ ∞
=
C
∑ |Ten (z)| 2
dλα (z) =
C
∑ Ten , Kz 2α dλα (z)
n=1 n=1
∞
=
C
∑ en , T Kz 2α dλα (z) =
C
T Kz 22,α dλα (z)
n=1
= SKz , Kz α dλα (z) =
S(z)K(z, z) dλα (z)
C C
α dA(z).
= S(z)
π C
Next, assume that S is self-adjoint and belongs to the trace class. Then we can
write
|S| + S |S| − S
S= − ,
2 2
3.1 The Berezin Transform of Operators 97
where each of the two quotients above is a positive operator in the trace class. The
desired trace formula then follows from the corresponding ones for positive trace-
class operators.
Finally, an arbitrary trace-class operator S can be written as
S + S∗ S − S∗
S= +i ,
2 2i
where each of the two quotients above is a self-adjoint operator in the trace class.
The desired trace formula for S follows from the corresponding ones for self-adjoint
trace-class operators.
Lemma 3.4. Suppose T is a positive operator on a Hilbert space H and x is a
unit vector in H. Then T p x, x ≥ T x, x p for p ≥ 1 and T p x, x ≤ T x, x p for all
0 < p ≤ 1.
Proof. See Proposition 1.31 of [250].
Proposition 3.5. If p ≥ 1 and T is in the Schatten class S p , then T belongs to
L p (C, dA).
Proof. If T is in the trace class, then we can write
T = T1 − T2 + i(T3 − T4 ),
is in L1 (C, dA).
If T is a bounded linear operator on Fα2 , the function T is in L∞ (C, dA). It follows
from complex interpolation that if T is any operator in the Schatten class S p , 1 <
p < ∞, then the function T is in L p (C, dA).
Alternatively, if 1 ≤ p < ∞ and T is in the Schatten class S p , then by the
decomposition T = T1 − T2 + i(T3 − T4 ), we may assume that T is positive. But
when T is positive, it is in the Schatten class S p if and only if T p is in the trace class,
so the desired result follows from Proposition 3.3 and Lemma 3.4.
Note that we did not need the positivity of T above, while this is necessary in the
next proposition.
Proposition 3.6. Suppose 0 < p ≤ 1 and T is a positive operator on Fα2 . If T ∈
L p (C, dA), then T belongs to the Schatten class S p .
Proof. Since T is positive, it belongs to the Schatten class S p if and only if S p is in
the trace class. The desired result then follows from Proposition 3.3 and Lemma 3.4.
98 3 The Berezin Transform and BMO
Pz ( f ) = f , kz kz , f ∈ Fα2 .
It follows that
where S1 denotes the trace class as a Banach space. Note that we have just used the
well-known inequality
|tr (T S)| ≤ T SS1
we can expand the unit vector kw to an orthonormal basis of Fα2 and calculate the
trace of Pz Pw with respect to this basis to obtain
S f (z) = S f , Kz α = f , S∗ Kz α
for all z and w in C. It is easy to see that the function KS (w, z) is uniquely determined
by the following two properties:
(a) S f (z) = f (w)KS (w, z) dλα (w) for all f ∈ Fα2 and z ∈ C.
C
(b) KS ( · , z) ∈ Fα2 for all z ∈ C.
We collect in the following proposition some of the elementary properties of the
kernel function KS (w, z) induced by S.
Proposition 3.9. The mapping S → KS has the following properties:
(1) KS+T = KS + KT , KcS = cKS .
(2) KS ( · , z) ∈ Fα2 .
(3) KS∗ (w, z) = KS (z, w).
(4) KI (w, z) = K(w, z).
(5) KSn → KS pointwise whenever Sn → S weakly.
(6) |KS (w, z)| ≤ S K(w, w)K(z, z).
(7) KSn → KS uniformly on compacta whenever Sn → S in norm.
(8) KS (z, z) = K(z, z)S∗ (z).
(9) KS (w, w) ≡ 0 if and only if S = 0.
Proof. Properties (1)–(5) and (8) are direct consequences of the definition of KS
in (3.3) and the definition of the Berezin transform. Property (6) follows from (3.3)
and the Cauchy–Schwarz inequality, and it implies property (7). Since the Berezin
transform S → S is one-to-one, we see that (9) follows from (8).
100 3 The Berezin Transform and BMO
Proof. This follows from Proposition 3.3 and property (8) in Proposition 3.9.
Corollary 3.12. Let S and T be bounded linear operators on Fα2 such that ST is
trace class. Then
tr (ST ) = dλα (w) KS (z, w) KT (w, z) dλα (z).
C C
We say that a Lebesgue measurable function ϕ satisfies condition (I p), where 0 <
p < ∞, if ϕ ◦ ta ∈ L p (C, dλα ) for every a ∈ C. In particular, any function satisfying
condition (I p ) must be in L p (C, dλα ).
By a change of variables, we see that a Lebesgue measurable function ϕ on C
satisfies condition (I p ) if and only if
|K(z, a)|2 |ϕ (z)| p dλα (z) < ∞ (3.4)
C
for all a ∈ C. By the exponential form of the kernel function K(w, z), the above
condition is equivalent to
|K(z, a)||ϕ (z)| p dλα (z) < ∞, a ∈ C. (3.5)
C
We will also call f the Berezin transform of f . It is clear that we can write
α
f(z) = f (w)e−α |z−w| dA(w) =
2
f (z ± w) dλα (w). (3.7)
π C C
for z ∈ C and
1 1 1
e− s |z−w| dA(w) f (u)e− t |w−u| dA(u)
2 2
Hs Ht f (z) =
π 2 st C C
where
1 1 2 − 1 |w−u|2
I(z, u) = e− s |z−w| t dA(w).
π 2 st C
Since
1 1 1 1 1 1
− |z − w|2 − |w − u|2 = − + |w|2 − |z|2 − |u|2
s t s t s t
z u
z u
+ + w+ + w,
s t s t
we have
1 − 1 |z|2 − 1 |u|2 ( zs + ut )w 2 −( 1s + 1t )|w|2
I(z, u) = e s t e e dA(w)
π 2 st C
− 1s |z|2 − 1t |u|2 2
e 1
+ 1t ( 1s + 1t ) tz+su
s+t w e−( s + t )|w| dA(w).
1 1 2
= · s
e
π (s + t) π C
1
e− s |z| − t |u| +( s + t )| s+t | .
1 2 1 2 1 1 tz+su 2
I(z, u) =
π (s + t)
1 1
e− s+t |z−u| .
2
I(z, u) =
π (s + t)
Therefore,
1 1
f (u)e− s+t |z−u| dA(u) = Hs+t f (z).
2
Hs Ht f (z) =
π (s + t) C
Theorem 3.14. The function u(x, y,t) = Ht f (z), where z = x + iy, satisfies the heat
equation
∂ 2u ∂ 2u ∂u
+ =4 . (3.10)
∂ x2 ∂ y2 ∂t
Similarly,
∂u 2 1
(x − u) f (u, v)e− t [(x−u)
2 +(y−v)2 ]
=− 2 du dv,
∂x πt R2
and
∂ 2u 2 1
f (u, v)e− t [(x−u) +(y−v) ] du dv
2 2
=− 2
∂ x2 π t R2
4 1
(x − u)2 f (u, v)e− t [(x−u) +(y−v) ] du dv.
2 2
+ 3
π t R2
where
∂ 2u ∂ 2u
Δu = +
∂ x2 ∂ y2
It follows that
ε 1 ε 1
e− t |z−w| dA(z) < e− t |z−w| dA(w) = ε .
2 2
|I1 | ≤
πt |w−z|<δ πt C
−δ 2 /t
= 2 f ∞ e →0
as t → 0+ . It follows that
Note that in the heat equation (3.10), the value u(x, y,t) represents the
temperature at the point (x, y) ∈ C at time t. Thus, the function f (z) represents
the initial temperature distribution in the complex plane at time t = 0. With this
interpretation, the assumption that f be bounded and continuous is reasonable.
However, the initial condition in (3.11) can be shown to hold for certain functions
that are more general than bounded and continuous ones.
The following result is a direct consequence of Theorems 3.13 and 3.14.
Corollary 3.15. For any positive α and β , we have the identities
Bα Bβ = B αβ = Bβ Bα .
α +β
for every z ∈ C.
We need the following result from Fourier analysis to generalize Proposition 3.1
to the Berezin transform of functions.
Lemma 3.16 Suppose that n is a positive integer and f is a function on Rn such
that the function
x → f (x)e|tx| e−x
2
and
x2 = x21 + · · · + x2n , dx = dx1 · · · dxn .
If
f (x)P(x)e−x dx = 0
2
Rn
and
N (itx)k ∞
|tx|k
∑ ≤ ∑ = e|tx|
k=0 k! k=0 k!
106 3 The Berezin Transform and BMO
for all N ≥ 0, we apply the dominated convergence theorem to partial sums to obtain
∞
ik
eitx f (x)e−x dx = ∑ k! (tx)k f (x)e−x dx = 0
2 2
Rn k=0 Rn
for all t ∈ Rn . By the Fourier inversion theorem, we have f (x)e−x = 0, and hence
2
for all nonnegative integers m and n. The result then follows from Lemma 3.16.
In the next few results, we describe some of the mapping properties of the Berezin
transform. In particular, we will compare Bα f and Bβ f in various situations.
Theorem 3.18. Let 1 ≤ p ≤ ∞. Suppose α , β , and γ are positive weight parameters.
Then Bα Lβp ⊂ Lγp if and only if γ (2α − β ) ≥ 2αβ .
It follows that
2
α f ∞,β −α |z|2 (α − 2 ) α − β w − α − β2 |w|2
β αz
Therefore,
γ 2α f ∞,β −(α + 2γ − 2α 2 )|z|2
|Bα f (z)|e− 2 |z| ≤
2
e 2α − β .
2α − β
γ 2α 2
α+ − ≥ 0.
2 2α − β
Thus, Bα maps L∞ ∞
β into Lγ .
If f ∈ L1β , the integral
γ 2
I= Bα f (z)e− 2 |z| dA(z)
C
equals
α γ
e−(α + 2 )|z|2
f (w)|e α zw 2 −α |w|2
| e dA(w) dA(z),
π C C
π C C
2α 2 β
α− ≥ .
2α + γ 2
By complex interpolation, the Berezin transform Bα maps Lβp into Lγp for all
1 ≤ p ≤ ∞ whenever γ (2α − β ) ≥ 2αβ .
To prove the other direction, observe that
It follows from this and a change of variables that Bα f ∈ Lγp if and only if Qα f ∈
L pγ + α . Therefore, Bα Lβp ⊂ Lγp is equivalent to Qα Lβp ⊂ L pγ + α , which implies that
4 2 4 2
Pα Lβp ⊂ L pγ + α . Combining this with Theorem 2.31, we conclude that Bα Lβp ⊂ Lγp
4 2
implies that
γ α
α 2 ≤ (2α − β ) + ,
4 2
which is equivalent to γ (2α − β ) ≥ 2αβ . This completes the proof of the theorem.
Corollary 3.19. Let α > 0 and β > 0. For 1 ≤ p ≤ ∞, we have
(a) Bα : Lαp → Lβp if and only if β ≥ 2α .
(b) Bα : Lβp → Lαp if and only if 2α ≥ 3β .
Bβ f ∞ ≤ Bα f ∞
for all f .
3.2 The Berezin Transform of Functions 109
for all f .
Proof. Choose a positive γ such that 1/γ + 1/α = 1/β . By Corollary 3.15, we have
Bβ = Bγ Bα . The desired result then follows from Proposition 3.20.
Proposition 3.22. If 0 < β < α , 0 < p < ∞, and f ≥ 0. Then
α
Bα f (z) ≤ B f (z), z ∈ C.
β β
Consequently:
(a) Bβ f ∈ L∞ (C) implies that Bα f ∈ L∞ (C).
(b) Bβ f ∈ C0 (C) implies that Bα f ∈ C0 (C).
(c) Bβ f ∈ L p (C, dA) implies that Bα f ∈ L p (C, dA).
Proof. Part (a) in the case 1 ≤ p ≤ ∞ and part (b) follow from Propositions 3.21
and 3.22. Part (a) in the case 0 < p < 1 will be proved in Chap. 6.
110 3 The Berezin Transform and BMO
The following result shows that the Berezin transform commutes with each of these
maps.
Proposition 3.24. If f is a function such that the Berezin transform Bα f is well
defined, then for any a ∈ C, we have
(i) Bα ( f ◦ ta ) = (Bα f ) ◦ ta .
(ii) Bα ( f ◦ τa ) = (Bα f ) ◦ τa .
(iii) Bα ( f ◦ ϕa ) = (Bα f ) ◦ ϕa .
C f ∞
|Bnα f (z) − Bnα f (w)| ≤ √ |z − w| (3.13)
n
for all z and w in C, where C = 2 α /π .
Proof. Recall that the Berezin transform of f is
3.2 The Berezin Transform of Functions 111
α
f (u)e−α |z−u| dA(u).
2
Bα f (z) =
π C
D = Bα f (z) − Bα f (w)
can be written as
α z + w −α |u−(z−w)/2|2
− e−α |u+(z−w)/2| dA(u).
2
f u+ e
π C 2
Let (z − w)/2 = reiθ with r ≥ 0. By the rotation invariance of the area measure,
α f ∞ −α |u−r|2 2
|D| ≤ e − e−α |u+r| dA(u)
π C
α f ∞
e−α (|u| |eα (u+ū)r − e−α (u+ū)r | dA(u).
2 +r2 )
=
π C
α f ∞ ∞ −α y2 ∞
e−α (x +r ) |e−2rα x − e2rα x | dx
2 2
|D| ≤ e dy
π −∞ −∞
√ ∞ ∞
2 α f ∞ 2
e−y dy e−α (x +r ) e2rα x − e−2rα x dx
2 2
=
π −∞ 0
√ ∞
2 α f ∞
e−α (x−r) − e−α (x+r) dx
2 2
= √
π 0
√ ∞ ∞
2 α −α x2 −α x2
= √ f ∞ e dx − e dx
π −r r
√ r
2 α
e−α x dx
2
= √ f ∞
π −r
√ √
4r α 2 α
≤ √ f ∞ = √ f ∞ |z − w|.
π π
112 3 The Berezin Transform and BMO
In the theory of Bergman spaces, it follows from a theorem of Ahern, Flores, and
Rudin that a function is fixed by the Berezin transform in that context if and only
if the function is harmonic, as long as the Berezin transform of the function is well
defined. No other assumption on the function is necessary. See [1].
Therefore, it is natural to ask if the fixed points of the Berezin transform in our
context here are exactly the harmonic functions as well. It turns out that the answer
is negative in general, but positive under certain conditions.
Proposition 3.26. Suppose f is a harmonic function on C satisfying condition (I1 ).
Then f = f .
Proof. If f is harmonic, then f ◦ tz is harmonic for every z. It follows from the mean
value theorem for harmonic functions that
f ◦ tz (0) = f ◦ tz (w) dλα (w).
C
Proof. Since f is bounded, the equivalence of (b) and (c) follows from the well-
known maximum modulus principle for harmonic functions. If f is constant, then
clearly f = f . If f = f , then f(n) = f for all positive integers n. By Theorem 3.25,
there exists a positive constant C such that
C
| f (z) − f (w)| ≤ √ |z − w|
n
for all z and w in C with z = w. Let n → ∞. We see that f must be constant.
Finally, in this section, we show by an example that there are more functions than
the harmonic ones that are fixed by the Berezin transform.
Lemma 3.28. For any complex ζ , let
∞
1
eζ t−t dt.
2
I(ζ ) = √
π −∞
We have I(ζ ) = eζ
2 /4
.
114 3 The Berezin Transform and BMO
It follows that I(ζ ) = Ceζ /4 for some constant C and all ζ ∈ C. It is well known
2
where
∞
α
eζ t−α t dt.
2
I(ζ , α ) =
π −∞
This shows that the function f is fixed by the Berezin transform, but it is not
harmonic.
116 3 The Berezin Transform and BMO
3.4 Fock–Carleson Measures 117
C C
for all z ∈ C.
(c) There exists a constant C > 0 such that μ (B(z, r)) ≤ C for all z ∈ C.
Proof. Fix a positive radius r and consider the lattice rZ2 in C. Let {zn } denote any
fixed arrangement of this lattice into a sequence. For any entire function f , we set
α
| f (w)e− 2 |w| | p dμ (w).
2
I( f ) =
C
Then
α
I( f ) ≤ ∑ | f (w)e− 2 |w| | p dμ (w).
2
n B(zn ,r)
By Lemma 2.32 and the triangle inequality, there exists a constant C1 > 0 such that
α α
| f (w)e− 2 |w| | p ≤ C1 | f (u)e− 2 |u| | p dA(u)
2 2
B(zn ,2r)
for all w ∈ B(zn , r). If condition (c) holds, then we can find a positive constant C2
(independent of f ) such that
α
I( f ) ≤ C2 ∑ | f (u)e− 2 |u| | p dA(u)
2
n B(zn ,2r)
118 3 The Berezin Transform and BMO
for all entire functions f . It is clear that there exists a positive integer N such
that every point in the complex plane belongs to at most N of the disks B(zn , 2r).
Therefore,
α
| f (u)e− 2 |u| | p dA(u).
2
I( f ) ≤ C2 N
C
pα 2
μ (B(z, r)) ≤ Ce 2 r
for all z ∈ C.
It is interesting to notice that condition (c) is independent of p and α . It follows
that if condition (a) holds for some p > 0 and some α , then it holds for every p and
every α (with the constant C dependent on p and α ).
Similarly, condition (a) is independent of r. Therefore, if condition (c) holds for
some r > 0, then it holds for every r > 0 (with the constant C dependent on r).
From now on, we will call any positive Borel measure μ that satisfies any of
the equivalent conditions (a)–(c) above a Fock–Carleson measure. Similarly, we say
that a positive Borel measure μ on C is a vanishing Fock–Carleson measure if
α
| fn (z)e− 2 |z| | p dμ (z) = 0,
2
lim
n→∞ C
Proof. By the proof of Theorem 3.29, there exists a positive constant C (independent
of z) such that
pα
e− 2 |z−w|
2
μ (B(z, r)) ≤ C dμ (w)
C
eα z̄n w
fn (w) = kzn (w) = , w ∈ C,
eα |zn |
2 /2
where {zk } is a fixed arrangement into a sequence of the lattice rZ2 . If condition (iii)
holds, then z → μ (B(z, r)) is a bounded function, and for any ε > 0, there exists a
positive integer N such that μ (B(zk , r)) < ε whenever k > N. Thus, for any bounded
sequence { fn } in Fαp that converges to 0 uniformly on compact sets, we can estimate
the sequence
2 p
In = fn (w)e−α |w| /2 dμ (w)
C
N
2 p
In ≤ C ∑ fn (w)e−α |w| /2 dA(w) (3.16)
k=1 B(zk ,2r)
∞
2 p
+Cε ∑ fn (w)e−α |w| /2 dA(w),
k=N+1 B(zk ,2r)
N
2 p
lim ∑ fn (w)e−α |w| /2 dA(w) = 0.
k=1 B(zk ,2r)
n→∞
120 3 The Berezin Transform and BMO
There is a positive integer m (depending on r only) such that every point in the
complex plane belongs to at most m of the disks D(zk , 2r). Therefore,
∞
2 p 2 p
∑ fn (w)e−α |w| /2 dA(w) ≤ m fn (w)e−α |w| /2 dA(w) ≤ C,
k=N+1 B(zk ,2r) C
This shows that condition (iii) implies condition (i). The proof of the theorem is
complete.
Carefully examining the proof of Theorems 3.29 and 3.30 above, we obtain
the following characterization of Fock–Carleson and vanishing Fock–Carleson
measures.
Corollary 3.31. Suppose μ is a positive Borel measure on C, r > 0, and {zn } is
any arrangement into a sequence of the lattice rZ2 . Then
(a) μ is a Fock–Carleson measure if and only if { μ (B(zk , r))} is in l ∞ .
(b) μ is a vanishing Fock–Carleson measure if and only if the sequence
{ μ (B(zk , r))} is in c0 .
Here, l ∞ denotes the space of all bounded sequences, and c0 is the space of all
sequences tending to 0.
Let μ be a complex, regular Borel measure μ on the complex plane. Define
α 2 −α |w|2 α
e−α |z−w| dμ (w),
2
μ
(z) = |kz (w)| e dμ (w) =
π C π C
3.4 Fock–Carleson Measures 121
μ (B(z, r))
μ
r (z) = .
π r2
Thus, conditions on the function z → μ (B(z, r)) can be replaced with the corre-
sponding conditions on the averaging function μ
r .
122 3 The Berezin Transform and BMO
3.5 Functions of Bounded Mean Oscillation 123
For any positive radius r and every exponent p ∈ [1, ∞), we define BMOrp to be the
space of locally area-integrable functions f on C such that
where
1
1 p
MO p,r ( f )(z) = |f −
fr (z)| p dA .
π r2 B(z,r)
Here,
1
fr (z) = 2 f dA
πr B(z,r)
is the mean (average) of f over the Euclidean disk B(z, r). Clearly, BMOrp is a linear
space.
When p = 2, it is easy to see that
1
MO22,r ( f )(z) = | f (u) − f (v)|2 dA(u) dA(v). (3.17)
2(π r2 )2 B(z,r) B(z,r)
MO22,r ( f )(z) = |
f |2 r (z) − | fr (z)|2 . (3.18)
By Hölder’s inequality,
1
1 1
p
| fr (z) − cz | = 2 ( f − cz ) dA ≤ | f − c z | p
dA .
π r B(z,r) π r B(z,r)
2
whenever |u − v| ≤ r.
Let z and w be two arbitrary points in the complex plane. We are going to show
that (3.20) holds for some positive constant C that is independent of z and w.
If |z − w| ≤ r, then (3.20) holds with C = M. If |z − w| > r, we place points
z0 , . . . , zn on the line segment from z to w in such a way that z0 = z, zn = w, |zk −
zk+1 | = r for 0 ≤ k < n − 1, and |zn−1 − zn | ≤ r. By the triangle inequality and (3.21),
n−1
| f (z) − f (w)| ≤ ∑ | f (zk ) − f (zk+1 )| ≤ nM.
k=0
Since BOr is actually independent of the radius r, we will write BO for BOr . The
initials in BO stand for bounded oscillation. It is clear that
p
f BA p
p = sup | f | (z) < ∞,
z∈C
where |f | p is the Berezin transform of | f | p with respect to the Gaussian measure
dλα . Although the weight parameter α appears in the definition of the norm above,
the space BA p is independent of α .
The space BA p depends on p. In fact, if 1 ≤ p < q < ∞, then BAq ⊂ BA p and the
containment is strict.
We now describe the structure of BMOrp in terms of the relatively simple spaces
BO and BA p . Recall that ϕz (w) = z − w.
Theorem 3.34. Let α > 0, r > 0, and 1 ≤ p < ∞. Suppose f is a locally area-
integrable function on C. Then the following conditions are equivalent:
(a) f ∈ BMOrp .
(b) f ∈ BO + BA p .
(c) f satisfies condition (I1 ), and there exists a positive constant C such that
| f ◦ ϕz (w) − f(z)| p dλα (w) ≤ C (3.22)
C
for all z ∈ C.
(d) There exists a positive constant C such that for any z ∈ C, there is some complex
number cz with
| f ◦ ϕz (w) − cz | p dλα (w) ≤ C. (3.23)
C
p
Proof. Let f ∈ BMO2r and |z − w| ≤ r. We have
| fr (z) −
fr (w)| ≤ | fr (z) − f2r (z)| + | f2r (z) − fr (w)|
126 3 The Berezin Transform and BMO
1
≤ | f (u) −
f2r (z)| dA(u)
π r2 B(z,r)
1
+ | f (u) −
f2r (z)| dA(u).
π r2 B(w,r)
Since B(z, r) and B(w, r) are both contained in B(z, 2r), it follows from Hölder’s
inequality that the two integral summands above are both bounded by a constant
that is independent of z and w. This proves that fr belongs to BOr = BO.
On the other hand, we can show that the function g = f − fr belongs to BA p
p p
whenever f ∈ BMO2r . In fact, it follows from (3.17) that f ∈ BMO2r implies that
p
f ∈ BMOr , and it follows from the triangle inequality for L p integrals that
1 1
p 1
p
|g| r (z) =
p | f (u) − fr (u)| dA(u)p
π r B(z,r)
2
1
1
p
≤ | f (u) − f r (z)| p
dA(u)
π r2 B(z,r)
1p
1 r (u) −
+ | f f r (z)| p
dA(u)
π r2 B(z,r)
≤ f BMOrp + ωr ( fr )(z).
f = fr + ( f −
fr ) ∈ BO + BA p .
We see that condition (3.22) holds whenever f ∈ BA p . On the other hand, it follows
from Hölder’s inequality that
f ◦ ϕz − f(z)Lpp (dλα )= | f (z−w) − f(z)| p dλα (w)
C
≤ | f (z−w) − f (z−u)| p dλα (w)dλα (u).
C C
3.5 Functions of Bounded Mean Oscillation 127
This together with Lemma 3.33 shows that for any f ∈ BO,
f ◦ ϕz − f(z)Lpp (dλα ) ≤ C p [|u − w| + 1]p dλα (w) dλα (u).
C C
The integral on the right-hand side above converges. Thus, condition (3.22) holds
for all f ∈ BO as well, and we have proved that condition (b) implies condition (c).
Mimicking the proof of Lemma 3.32, we easily obtain the equivalence of
conditions (c) and (d).
Finally, if condition (3.22) holds, we can find a positive constant C such that
C
| f (w) − f(z)| p dA(w)
π r2 B(z,r)
≤ | f (w) − f(z)| p |kz (w)|2 dλα (w)
C
= | f ◦ ϕz (w) − f(z)| p dλα (w).
C
This, along with Lemma 3.32, then shows that condition (c) implies condition (a).
As a consequence of Theorem 3.34, we see that the space BMOrp is independent
of r and the Berezin transform of every function in BMOrp is well defined. Thus, we
will write BMO p for BMOrp and define a complete seminorm on BMO p by
One of the nice features of this seminorm is that it is invariant under the actions of
ta , τa , and ϕa .
The proof of Theorem 3.34 also shows that every function in BMO p satisfies
condition (I p ). In particular, BMO p ⊂ L p (C, dλα ).
Theorem 3.35. If 1 < p < ∞, then there exists a positive constant C = C(p, α ) such
that
| f(z) − f(w)| ≤ C f BMO p |z − w|
π C
128 3 The Berezin Transform and BMO
2α 2
( f (u) −
f (γ (t)))e−α |γ (t)−u| Re γ (t)(γ (t) − u) dA(u).
2
−
π C
1
2α 2 p
| f (u) − f(γ (t))| p e−α |γ (t)−u| dA(u)
2
π C
times
1
q
q −α |γ (t)−u|2
|γ (t) − u| e dA(u) . (3.24)
C
≤ C f ◦ ϕz − f(z)L p (dλα ) ,
where the last step follows from Hölder’s inequality. This shows that f − fr is a
bounded function. Since a bounded continuous function belongs to both BO and
BA p , we have f−
fr ∈ BO ∩ BA p .
Write
f = ( f − fr ) − ( f−
f− fr ),
f = fr + ( f−
fr )
BMO p = LIP + BA p ,
for all z and w. Let w = 0 and use Cauchy’s estimate. We conclude that f must be a
linear polynomial.
Conversely, if f is a linear polynomial, then f is Lipschitz in the Euclidean
metric. In particular, f ∈ BO, and so f ∈ BMO p .
Let VMOrp denote the space of locally area-integrable functions f such that
p p p p
It is clear that VMOr is a subspace of BMOr . Just like BMOr , the space VMOr is
p
also independent of r, and we will write VMO p for VMOr .
Similarly, we consider the space VOr consisting of continuous functions f such
that
lim ωr ( f )(z) = 0.
z→∞
It can be shown that VOr is independent of r, and we will write VO for VOr . The
initials in VO stand for “vanishing oscillation.”
We also consider the space VArp consisting of functions such that
1
lim | f (w)| p dA(w) = 0.
z→∞ π r2 B(z,r)
Moreover, there are two canonical decompositions for condition (iii) above:
3.6 Notes
The Berezin transform was introduced in [23] and then studied systematically in
[23–27] for a number of reproducing Hilbert spaces. It has become an indispensable
tool in the study of operators on function spaces, including Hankel operators,
Toeplitz operators, and composition operators. See [250] for applications of the
Berezin transform in the theory of Bergman spaces. In particular, the proofs of
Propositions 3.3–3.6 were adapted from the corresponding ones in [250].
In the setting of Fock spaces and when parametrized appropriately, the Berezin
transform is nothing but the heat transform. This connection with the heat equation
makes the Berezin transform on Fock spaces particularly useful. The semigroup
property of the heat transforms was first observed in [30].
The Lipschitz estimate for the Berezin transform of a bounded linear operator
on the Fock space is due to Coburn. See [54, 55]. Propositions 3.9–3.11 and
Corollary 3.12 are taken from [55], and these results will be needed in Chap. 6 when
we study Toeplitz operators on the Fock space.
Theorem 3.25, the Lipschitz estimate for the Berezin transform of a bounded
function, was first proved in [29]. Together with the semigroup property, this result
shows that the Berezin transform is a rapidly smoothing operation on bounded
functions, and consequently, a bounded function that is fixed by the Berezin
transform must be constant. On the other hand, there exist unbounded functions
fixed by the Berezin transform that are not harmonic. The example in Sect. 3.3 was
taken from [84]. This example shows the sharp contrast with the Bergman space
theory, where the fixed points of the Berezin transform are exactly the harmonic
functions; see [1].
The characterization of Fock–Carleson measures is analogous to the characteri-
zation of Carleson measures for Bergman spaces. The material in Sect. 3.4 is taken
from [132]. See [250] for the corresponding results in the Bergman space theory.
Note that the notion of Carleson measures was initially introduced in the Hardy
space setting, where a geometric characterization is much more difficult. See [76].
The notion of BMO and VMO using a fixed Euclidean radius was first introduced
in [32, 257]. This idea was then generalized to the setting of bounded symmetric
domains in [21] and to the case of strongly pseudoconvex domains in [149], with
the Euclidean metric replaced by the Bergman metric. The resulting spaces are
independent of the particular radius used, but the dependence on the exponent p
was observed and studied in [248] in the context of Bergman spaces on the unit ball.
The extension to the Fock space setting is straightforward.
The Lipschitz estimate for the Berezin transform of a function in BMO was first
proved in [21] in the context of Bergman spaces on bounded symmetric domains.
The extension to the Fock space, Theorem 3.35, was first carried out in [13].
134 3 The Berezin Transform and BMO
3.7 Exercises 135
3.7 Exercises
√
1. Show that the Lipschitz constant 2 α in Corollary 3.8 is best possible.
2. Show that the spaces BMO p and VMO p are complete under the norm
f = f BMO p + | f(0)|.
lim f (z) = L
z→∞
In this chapter, we characterize interpolating and sampling sequences for the Fock
spaces Fαp . The characterizations are based on a certain notion of uniform density
on the complex plane. So we will first spend some time discussing this geometric
notion of density which also has applications in other areas of analysis and physics.
S = B(w, r) = {z ∈ C : |z − w| < r} ,
and
S = S(w, r) = {z ∈ C : |Re z − Rew| < r/2, |Im z − Imw| < r/2} .
Proof. Fix any positive number ε . It is clear that there exist a finite number of
disjoint open squares S(w j , r j ), 1 ≤ j ≤ N, in B(0, 1) such that
0 < π − r12 + · · · + rN2 < ε .
For any w ∈ C and r > 0, it is easy to see that z ∈ S(w + rw j , rr j ) if and only if
(z − w)/r ∈ S(w j , r j ). It follows that the squares S(w + rw j , rr j ) are disjoint and
contained in B(w, r). Thus,
140 4 Interpolating and Sampling Sequences
n(Z, B(w, r)) ≥ n Z, ∪Nj=1 S(w + rw j , rr j )
N
= ∑ n(Z, S(w + rw j , rr j ))
j=1
N
n(Z, S(w + rw j , rr j ))
= ∑ (rr j )2
· (rr j )2 .
j=1
It follows that
n
n(Z, S(ζ , rr j )) r2j
≥ ∑ inf
(rr j )2
· .
π
j=1 ζ ∈C
so that
n(Z, B(w, r)) ≤ n Z, ∪Nj=1 S(w + rw j , rr j )
N
≤ ∑ n(Z, S(w + rw j , rr j ))
j=1
N
n(Z, S(w + rw j , rr j ))
= ∑ (rr j )2
· (rr j )2 .
j=1
4.1 A Notion of Density 141
It follows that
N
n(Z, S(ζ , rr j )) r2j
≤ ∑ sup (rr j )2
· .
π
j=1 ζ ∈C
Proof. The proof is similar to that of Proposition 4.1. We will not need the full
strength of the theorem and will omit its proof here. We refer the interested reader
to [36] for details.
We conclude the section with an example for which we can explicitly compute
the uniform densities.
Proposition 4.3. For any lattice
Λ = {ω + mω1 + nω2 : m ∈ Z, n ∈ Z} ,
142 4 Interpolating and Sampling Sequences
we have
1
D+ (Λ ) = D− (Λ ) = .
|Im (ω1 ω 2 )|
When r is very large, the number of points in Λ ∩ B(w, r) is roughly the area of
B(w, r) divided by the area of the fundamental region of Λ . It follows that
(π r2 )/|Im (ω1 ω 2 )| 1
D+ (Λ ) = D− (Λ ) = lim = .
r→∞ πr 2 |Im (ω1 ω 2 )|
As a special case, if r is any positive number, then the uniform densities of the
square lattice rZ2 are given by
Let Z = {zn } be a sequence of distinct points in the complex plane. We say that Z is
separated if
δ (Z) = inf {|zn − zm | : n = m} > 0.
When Z is separated, the number δ = δ (Z) will be called the separation con-
stant of Z.
The next result is a necessary condition that the values of a function in Fαp taken
on a separated sequence must satisfy.
Proposition 4.4. Let Z = {zn } be a separated sequence and 0 < p < ∞. Then there
exists a positive constant C, independent of f , such that
∞ p
−α |zn |2 /2
∑ n
f (z )e p
≤ C f p,α
n=1
for all f ∈ Fαp and all n ≥ 1, where r = δ /2. By the definition of the separation
constant, the Euclidean disks B(zn , r) are all disjoint. Therefore,
∞ ∞
∑ | f (zn )e−α |zn | /2| p ≤ C ∑ | f (z)e−α |z|
2 2 /2
| p dA(z)
n=1 n=1 B(zn ,r)
| f (z)e−α |z|
2 /2
≤C | p dA(z)
C
2π C p
= f p,α .
pα
n≥1
The following result shows that if Z is an interpolating sequence for Fαp , then
interpolation can be performed in a stable way.
Lemma 4.5. Suppose 0 < p ≤ ∞ and Z = {zn } is an interpolating sequence for Fαp .
Then there exists a positive constant C with the following property: whenever {vn }
is a sequence such that {vn e−α |zn | /2 } ∈ l p there exists a function f ∈ Fαp such that
2
l p . Let JZ denote the space of all functions f ∈ Fαp such that f (z) = 0 for all z ∈ Z.
It is clear that JZ is a closed subspace of Fαp . For any sequence v = {vk } ∈ X p , there
exists a function f ∈ Fαp such that f (zk ) = vk for all k ≥ 1. We define T (v) = f + JZ .
Then T is a well-defined linear mapping from X p into the quotient space Fαp /JZ . It is
easy to check that T has a closed graph in X p × (Fαp /JZ ). Therefore, by the closed-
graph theorem, the mapping T is continuous, which implies the desired estimate.
If Z is an interpolating sequence for Fαp , we are going to use N p (Z) = N p (Z, α )
to denote the smallest constant C satisfying the inequality in (4.3). We put N p (Z) =
N p (Z, α ) = ∞ when Z is not an interpolating sequence for Fαp . We also use the
convention that N p (0)
/ = 0.
We say that a sequence Z = {zn } of distinct points in C is a sampling sequence
for Fαp , 0 < p < ∞, if there exists a constant C > 0 such that
∞ p
− α2 |zn |2
C−1 f pp,α ≤ ∑ f (zn )e p
≤ C f p,α (4.4)
n=1
Sampling for Fα∞ requires a slightly different treatment. More specifically, we say
that an arbitrary set Z in C is a sampling set for Fα∞ if there exists a constant C > 0
such that
α
f ∞,α ≤ C sup | f (z)|e− 2 |z|
2
(4.5)
z∈Z
for all f ∈ Fα∞ . When Z is a sequence, we use the term “sampling sequence” instead
of “sampling set.”
We use M p (Z) = M p (Z, α ) to denote the smallest constant C such that
for all f ∈ Fαp . Thus, Z is a sampling set for Fα∞ if and only if M∞ (Z) < ∞, and it is a
sampling sequence for Fαp , 0 < p < ∞, if and only if M p (Z) < ∞ and f |Z p,α < ∞
for all f ∈ Fαp .
We use the convention that the empty set is not a sampling set for Fαp , which
should be easy to conceive and accept. In particular, we are going to write
/ = ∞.
M∞ (0)
Recall that for any complex number a, the Weyl unitary operator Wa is defined by
α
Wa f (z) = eα āz− 2 |a| f (z − a).
2
which allows us to translate our analysis around an arbitrary point to the origin 0.
Our next step is to show that every interpolating sequence for Fαp must be
separated, and every sampling sequence for Fαp must contain a separated sequence
that is still sampling. The following estimate will be needed for this purpose as well
as several other results.
Lemma 4.6. Suppose 0 < p < ∞, f is entire, and
For any positive radius δ , there exists a constant C = C(α , p, δ ) > 0 such that
α 2 p
|S(ζ + z)| − |S(ζ )| p ≤ C|z| p − 2 |u|
e f (u) dA(u)
B(ζ ,3δ )
It follows that
|S(ζ + z)| − |S(ζ )| = e− α2 |z|2 | fζ (z)| − | fζ (0)|
α 2
= e− 2 |z| − 1 | fζ (z)| + | fζ (z)| − | fζ (0)|
α 2
≤ 1 − e− 2 |z| | fζ (z)| + | fζ (z) − fζ (0)|
α 2 α 2
= e 2 |z| − 1 e− 2 |z| fζ (z) + | fζ (z) − fζ (0)|
α 2 α 2
= e 2 |z| − 1 e− 2 |z| fζ (z) + |z|| fζ (w)|,
where the last step follows from the mean value theorem with some w satisfying
|w| < |z|.
By Lemma 2.32, there exists a constant C1 > 0 such that
α 2 p α 2 p
− 2 |z| − 2 |u|
e fζ (z) ≤ C1 e fζ (u) dA(u)
B(z,δ )
α 2 p
− 2 |u|
≤ C1 e fζ (u) dA(u)
B(0,2δ )
α 2 p
− 2 |u|
= C1 e f (u) dA(u).
B(ζ ,2δ )
The second inequality above follows from the triangle inequality, and the last
equality follows from a change of variables.
On the other hand, it follows from Cauchy’s integral formula that
1 fζ (u) du
fζ (w) = .
2π i |u−w|=δ (u − w)2
Consequently,
1
| fζ (w)| ≤ sup | fζ (u)| ≤ C2 sup | fζ (u)|e−α |u| /2 .
2
δ |u−w|=δ |u−w|=δ
|u + v| p ≤ 2 p (|u| p + |v| p)
With the notation S(z) = e−α |z| /2 f (z) from Lemma 4.6 and Corollary 4.7, we have
2
1 = |an | − |am | = |S(zn )| − |S(zm )| ≤ CN p (Z)|zn − zm |,
where C is a positive constant that only depends on α and p. This shows that the
sequence Z is separated.
We now proceed to show that every sampling sequence for Fαp must contain a
separated subsequence that is also a sampling sequence for Fαp . We break the proof
into two cases: 0 < p < ∞ and p = ∞.
Lemma 4.9. Suppose 0 < p < ∞ and Z = {zn } is any sequence of complex numbers.
Then the following two conditions are equivalent:
(a) There exists a positive constant C such that
∞
α 2 p
∑ f (zn )e− 2 |zn | ≤ C f pp,α
n=1
is a Fock–Carleson measure for Fαp , where δz is the unit point mass at z. Therefore,
according to (an obvious variant of) Theorem 3.29, condition (a) is equivalent to the
existence of a positive integer N such that any square S ⊂ C of side length 1 contains
at most N points from Z, which is clearly equivalent to the condition that Z is the
union of finitely many separated sequences.
An obvious consequence of the above result is that every sampling sequence for
Fαp , where 0 < p < ∞, contains a separated subsequence. The following result shows
that this is true for p = ∞ as well and we can do more than that.
Lemma 4.10. If Z = {zn } is a sampling sequence for Fα∞ , then Z contains a
separated subsequence Z that is also a sampling sequence for Fα∞ .
Proof. Fix a sufficiently small positive number ε whose exact value will be specified
later. Let z1 = z1 , discard the terms in the sequence {zn } that are within ε of z1 , and
denote the remaining terms by {z11 , z12 , · · · } with the original order. Let z2 = z11 ,
discard the terms in the sequence {z1n } that are within ε of z2 , and denote by
{z21 , z22 , · · · } the remaining terms in the original order. Continuing this process,
infinitely many times if necessary, we obtain a subsequence Z = {zn } of Z which
clearly satisfies the condition |zi − zj | ≥ ε whenever i = j. In particular, Z is
separated. Furthermore, for any zk , either it was discarded during the process above,
in which case it is within ε of some point in the sequence Z , or it eventually gets
picked as a term in Z . Either way, we have d(zk , Z ) < ε so that
Z= Z ∩ B(z , ε ) . (4.7)
z ∈Z
Given any w ∈ Z , it follows from (4.7) that there exists some z ∈ Z such that
|w − z| ≤ ε . By the triangle inequality and Corollary 4.7,
|S(w)| ≤ |S(z)| + |S(z)| − |S(w)|
≤ f |Z ∞,α + Cε f ∞,α ,
for all f ∈ Fα∞ . Since Z is a sampling sequence for Fα∞ , there exists a positive constant
c such that c f ∞,α ≤ f |Z∞,α for all f ∈ Fα∞ . Thus,
(c − Cε ) f ∞,α ≤ 2 f |Z ∞,α
for all f ∈ Fα∞ . If the value of ε was chosen such that c − Cε > 0, then there is
another positive constant C > 0 such that C f ∞,α ≤ f |Z ∞,α for all f ∈ Fα∞ ,
which means that Z is sampling for Fα∞ .
We want to show that the lemma above holds for p < ∞ as well. But the proof is
more complicated.
Lemma 4.11. Suppose 0 < p < ∞ and Z = {zk } is a sampling sequence for Fαp .
Then Z contains a separated subsequence that is also a sampling sequence for Fαp .
Proof. By Lemma 4.9, we can write Z = Z1 ∪ Z2 ∪ · · · ∪ Zn as a disjoint union of
separated sequences. We prove the result by induction on n. If n = 1, there is nothing
to prove. Thus, we assume n > 1 and proceed to show that we can find a subsequence
Z of Z such that:
(a) Z is sampling for Fα .
p
Z = (Z1 ∪ Γ ) ∪ · · · ∪ Zn−1 ,
and each of the n − 1 sequences above is separated. We will show that Z is sampling
for Fαp when ε is sufficiently small.
Since Z = Z ∪ Γ , we will be done if Γ is empty. If Γ is not empty, we write
Γ = {ζk }. For each k, there exists a point ak ∈ Z such that |ζk − ak | < ε . For i = j,
we have
|ai − a j | = |(ai − ζi ) − (a j − ζ j ) + (ζ j − ζi )|
≥ |ζi − ζ j | − |(ai − ζi ) − (a j − ζ j )|
3
≥ δ − 2ε > δ .
4
f |Γ pp,α = ∑ |S(ζk )| p
k
Since the sequence {ak } is separated with separation constant at least 3δ /4, there is
another constant C > 0, independent of ε and f , such that
p
∑|S(ζk )| − |S(ak)| ≤ C ε p f pp,α
k
so that
c − 2 pC ε p f pp,α ≤ (2 p + 1) f |Z pp,α
for all f ∈ Fαp . If the value of ε was chosen such that c − 2 pC ε > 0, then the
sequence Z is sampling for Fαp . This completes the proof of the lemma.
4.3 Stability Under Weak Convergence 151
[A, B] = inf {t : A ⊂ Bt , B ⊂ At }
and call it the Hausdorff distance between A and B. It can be verified that this is
indeed a metric. The assumption that A and B are relatively closed ensures that
[A, B] = 0 only when A = B.
Alternatively,
[A, B] = max (d ∗ (A, B), d ∗ (B, A)) ,
where
d ∗ (A, B) = sup d(z, B) = sup inf |z − w|
z∈A z∈A w∈B
nondecreasing. We use the notation W (Z) to denote the collection of weak limits of
all the translates Z + z of Z. The set W (Z) will play a crucial role in our analysis.
We first prove a certain compactness property for uniformly separated sequences
in the complex plane.
Proposition 4.12. For each n ≥ 1, let Zn be a separated sequence. If δ =
infn δ (Zn ) > 0, then there exists a subsequence {Znk } and a separated sequence Z
(possibly empty) such that {Znk } converges weakly to Z.
Proof. We write Zn = {zn1 , zn2 , · · · } with |zn1 | ≤ |zn2 | ≤ · · · . If zn1 → ∞ as n → ∞,
then for every k, we have znk → ∞ as n → ∞. In this case, {Zn } converges weakly to
the empty set.
If zn1 → ∞ as n → ∞, we can find a subsequence {Zn j } such that zn j 1 → z1 as
j → ∞. Then either zn j 2 → ∞ as j → ∞, which implies that for every k ≥ 2, we have
zn j k → ∞ as j → ∞, or {Zn j } has a subsequence whose second components converge
to some z2 ∈ C. In the latter case, the process continues.
There are now two possibilities: either the process terminates after a finite
number, say N, of iterations, which produces a subsequence of {Zn } that converges
weakly to a finite sequence Z = {z1 , · · · , zN }, or the process never stops, which via
a diagonalization argument produces a subsequence of {Zn } that converges weakly
to an infinite sequence Z = {z1 , z2 , · · · }. The condition infn δ (Zn ) > 0 ensures that
the limit sequence Z is separated as well. This proves the desired result.
The following result gives an alternative description of weak convergence for
separated sequences.
Proposition 4.13. Suppose each Zn is a separated sequence with δ = infn δ (Zn ) >
0. Write Zn = {zn1 , zn2 , · · · } with |zn1 | ≤ |zn2 | ≤ · · · . Then {Zn } converges weakly to
Z if and only if one of the following is true:
(a) Z = 0/ is the empty set, and for every k ≥ 1 we have znk → ∞ as n → ∞.
(b) Z = {z1 , · · · , zN } is a finite set, znk → zk for every 1 ≤ k ≤ N, and znk → ∞ for
every k > N.
(c) Z = {z1 , z2 , · · · } is an infinite (separated) sequence and znk → zk for every k ≥ 1.
Proof. It is clear from the definition that any one of the above conditions implies that
{Zn } converges weakly to Z. The other implication follows from Proposition 4.12
and its proof, if we start out with an arbitrary subsequence of {Zn }. Here, we use the
fact that znk → zk (where zk is either finite or infinite) if and only if each subsequence
of {z1k , z2k , · · · } converges to zk .
We now prove that any weak limit of sampling sequences for Fα∞ remains a
sampling sequence for Fα∞ .
Proposition 4.14. Suppose {Zn } converges weakly to Z. Then
f ∞,α ≤ M f |Z∞,α ,
we can write
f ∞,α 1
M∞ (Z) = sup = sup .
f ∈Fα∞ f |Z∞,α f ∞,α =1 f |Z∞,α
M −1 = inf f |Z∞,α .
f ∞,α =1
Thus, for any ε ∈ (0, 1), we can find a unit vector f ∈ Fα∞ such that
f |Z∞,α < M −1 + ε .
whenever n > N, where B̄(0, r) is the closed disk with center 0 and radius r. Here, we
may assume that ε is small enough so that none of Zn ∩ B̄(0, ε −2 ) and Z ∩ B̄(0, ε −2 )
is empty.
Let a = 1 − (δ ε 2 /2) and assume that ε and δ are small enough so that a ∈
(0, 1). If n > N and w ∈ Zn ∩ B̄(0, ε −2 ), there exists some z ∈ Z ∩ B̄(0, ε −2 ) such
that |w − z| < δ /2. It follows from the triangle inequality that
δ δ
|aw − z| ≤ a|z − w| + (1 − a)|z| < + = δ.
2 2
154 4 Interpolating and Sampling Sequences
Therefore,
α α α α
e− 2 |w| | f (aw)| = e− 2 |aw| | f (aw)|e− 2 (1−a )|w| ≤ e− 2 |aw| | f (aw)|
2 2 2 2 2
α 2
α 2 α 2
≤ e− 2 |aw| | f (aw)| − e− 2 |z| | f (z)| + e− 2 |z| | f (z)|
α
≤ f ∞,α e− 2 (1−a)|w|
2
α −4 Cα
≤ e− 2 (1−a)ε = e − 4ε .
for all |w| > ε −2 . Combining this with the last estimate in the previous paragraph,
we conclude that the function g(z) = f (az) satisfies
It follows that
g∞,α 1−ε
M∞ (Zn ) ≥ ≥ −1
g|Zn ∞,α M + 2ε
The discussion above was about the behavior of the sampling constant M∞ (Z)
under weak convergence. The following result concerns the sampling constant
M p (Z) when p < ∞. Recall that for any separated sequence Z, we use
f |Z p,α ≤ M −1 + ε ,
Thus, for any n > N and z ∈ Zn ∩ B̄(0, R), we can find some w ∈ Z such that
δ
|z − w| < , |z − w| < ε .
6
Since Z is separated with separation constant at least δ , we see that different z
correspond to different w. By Lemma 4.6, there exists a positive constant C =
C(α , p, δ ) such that
α 2 α 2 p
| f (z)|e− 2 |z| − | f (w)|e− 2 |w| ≤ Cε p | f (u)e−α |u|
2 /2
| p dA(u).
B(w,δ /2)
Sum over all z ∈ Zn ∩ B̄(0, R), observe that different z correspond to different w, and
use the facts that f is a unit vector in Fαp and δ (Z) ≥ δ . We obtain
If 1 ≤ p < ∞, we apply the version of the triangle inequality for p > 1 to get
⎡ ⎤1
p p
α
f |Zn ∩ B(0, R) p,α = ⎣ ∑ f (z)e− 2 |z| ⎦
2
z∈Zn ∩B(0,R)
⎡ ⎤1
p p
α
≤⎣ ∑ f (w)e− 2 |w| ⎦
2
w∈Z∩B(0,R)
1
p p
α
∑ f (u)e− 2 |u| dA(u)
2
+ Cε p
w∈Z B(w,δ /2)
so that
−1
lim inf M p (Zn , α ) ≥ M −1 + (1 + C1/p)ε .
n→∞
M p (Z , α ) ≤ C[Z, Z ]M p (Z)
for sequences Z that are sufficiently close to Z. Here, the constant C only depends
on p and α .
This concludes the discussion about the stability of sampling sequences under
weak convergence. Next, we consider the stability of interpolating sequences under
weak convergence.
Proposition 4.18. Suppose {Zn } converges to Z weakly. Then
By the proof of Lemma 4.8, we have δ = infn δ (Zn ) > 0. Then it follows easily from
Proposition 4.13 that the sequence Z is also separated and its separation constant is
at least δ .
With the help of Proposition 4.13, we may also assume that
n→∞
A key tool in our proof of the sufficiency of the sampling and interpolating
conditions is a special, modified Weierstrass σ -function. Thus, we let Λα = {ωmn }
denote the square lattice in C that is defined by
ωmn = π /α (m + in),
where m and n run over all integers. Recall that the Weierstrass σ -function
associated to Λα is defined by
z z 1 z2
σα (z) = z ∏
1− exp + ,
m,n ωmn ωmn 2 ωmn
2
is the fundamental region for Λα , then for any z ∈ C, there exists some w ∈ Rα
such that
α α
|σα (z)|e− 2 |z| = |σα (w)|e− 2 |w| .
2 2
The reciprocal density parameter α in Λα is critical for the Fock spaces Fαp . More
precisely, we will see that Λβ is interpolating for Fαp if and only if β < α ; and Λβ is
sampling for Fαp if and only if β > α . When β = α , Λβ is neither interpolating nor
sampling for Fαp , but is a set of uniqueness for Fαp ; see Lemma 5.7.
We will need to perturb the zeros of the Weierstrass σ -function σα (z). Let Z =
{zmn } be a sequence of distinct points in C. If there exists a constant Q > 0 (not
necessarily small!) such that |ωmn − zmn | ≤ Q for all ωmn ∈ Λα , then we say that Z
is uniformly close to Λα . For any sequence Z = {zmn } that is uniformly close to Λα ,
we define an associated function as follows:
z z 1 z2
g(z) = gZ (z) = (z − z00 ) ∏ 1 − exp + . (4.8)
m,n zmn zmn 2 ωmn 2
Here, we assume that z00 is the point of Z closest to 0. Note that both zmn and ωmn
appear in the formula above; it was not a misprint.
Lemma 4.21. Let Z be uniformly close to Λ = Λα and let g be its associated
function defined above. Then g is an entire function and the zero set of g is exactly
Z. Moreover, there exist positive constants C1 , C2 , and c such that
α
|g(z)|e− 2 |z| ≥ C1 e−c|z| log |z| d(z, Z)
2
(4.9)
and
α
|g(z)|e− 2 |z| ≤ C2 ec|z| log |z|
2
(4.10)
where
1 1
h1 (z) = exp z ∑
zmn
−
ωmn
,
|zmn |≤2|z|
and
(1 − z/zmn) exp(z/zmn )
h3 (z) = ∏ (1 − z/ωmn) exp(z/ωmn )
.
|zmn |>2|z|
for some constant C > 0 and all (m, n) = (0, 0). Using this and the elementary
estimates
N N
1
|ew | ≥ e−|w| , ∑∑ n 2 + m2
∼ log N, (4.12)
n=1 m=1
Rewrite h2 (z) as
∏ [1 − (ωmn − zmn )/(ωmn − z)]
h2 (z) = ϕ (z) ,
∏ [1 − (ωmn − zmn )/ωmn ]
where
d(z, Λ ) z − z00 1 − z/zkl
ϕ (z) = ,
d(z, Z) z 1 − z/ωkl
ωkl is the point in Λ that is closest to z, and the finite product ∏ is taken over all
(m, n) such that
Since Q satisfies |zmn − ωmn | ≤ Q for all m and n, the condition |zmn | ≤ 2|z|
implies that
|ωmn | ≤ 2|z| + Q, |ωmn − z| ≤ 3|z| + Q.
It follows that
∏ 1 − ωmn − zmn ≤ ∏ 1 + Q
ωmn |ωmn |
Q
≤ ∏ 1+ : 0 < |ωmn | ≤ 2|z| + Q .
|ωmn |
To estimate the other product ∏ in h2 (z) above, we move a few additional factors
into ϕ (z) and further assume that |z − ωmn | > Q. Therefore, we can find a positive
constant C, independent of z, such that
for all z ∈ C. Take the logarithm of the above inequality, use the fact that log(1+x) ∼
x when x is small, and observe that
1
∑ |ωmn − w| : δ < |ωmn − w| < R ∼ R
as R → ∞ (which is easily obtained with the help of polar coordinates), we see that
there are positive constants c and C such that
(1 − z/zmn ) exp(z/zmn )
1−
(1 − z/ωmn ) exp(z/ωmn )
∼ (1 − z/zmn ) exp(z/zmn ) − (1 − z/ωmn) exp(z/ωmn )
2
z2 z2 z
∼ 2 − 2 =O .
zmn ωmn ωmn
3
4.4 A Modified Weierstrass σ -Function 163
It follows that
1
log |h3 (z)| ≥ −C1 |z|2 ∑ | ω mn |
3
≥ −C2 |z|
|zmn |>2|z|
so that
|h3 (z)| ≥ Ce−c|z| , z ∈ C, (4.15)
for all z = zmn . Fix zmn , let z → zmn , and observe that d(z, Z) = |z − zmn | when z is
sufficiently close to zmn . We then obtain (4.11).
To prove (4.10), we write g = σα H, or
α α
e− 2 |z| g(z) = e− 2 |z| σ (z)H(z).
2 2
α
The quasiperiodicity of σα implies that the factor e− 2 |z| σα (z) is bounded. Rewrite
2
H = H1 H2 H3 , where H1 = h1 , H3 = h3 , and
z − z00 1 − z/zmn
H2 (z) =
z |z ∏ 1 − z/ωmn ,
mn |≤2|z|
and estimate the functions Hk the same way we did hk , the result is (4.10). This
completes the proof of the lemma.
Lemma 4.22. Let g be the function associated to Z = {zmn }. For any positive radius
R, there exists a positive constant C such that
g(z)
z − zmn ≤ C
The desired result then follows from the fact that the function g(z)/d(z, Z) is
continuous on the whole complex plane.
Lemma 4.23. Suppose Z is a sequence that is uniformly close to Λα . Then,
D+ (Z) = D− (Z) = α /π .
164 4 Interpolating and Sampling Sequences
Proof. Suppose Z = {zmn }, Λα = {ωmn }, and |zmn − ωmn | ≤ Q for all m and n,
where Q is a positive constant. When r is much larger than Q, the number of points
in Z ∩ B(w, r) is roughly the same as the number of points in Λα ∩ B(w, r). More
precisely, it is easy to see that
n(Z, B(w, r))
lim =1
r→∞ n(Λα , B(w, r))
and the convergence is uniform in w ∈ C. This clearly gives the desired result.
The following result is usually referred to as a Lagrange-type interpolation
formula.
Proposition 4.24. Let Z = {zmn } be a separated sequence in C that is uniformly
close to Λβ and let g be the function associated to Z by (4.8). If α < β , then every
function f ∈ Fα∞ can be written as
f (zmn ) g(z)
f (z) = ∑
,
m,n (zmn ) z − zmn
g
for all m and n. This, along with Lemma 4.22, shows that the series converges
uniformly on compact subsets of C.
To show that the series actually converges to f (z), we argue as follows. For each
sufficiently large r, it is easy to see that we can find a simple closed pass S = Sr
such that
d(S, Z) ≥ δ (Z)/2, d(S, 0) > r, |S| ≤ 8π r, (4.16)
where δ (Z) is the separation constant of Z. Let U be the region bounded by S. For
any z ∈ U − Z, we have by the calculus of residues that
1 f (ζ ) dζ f (z) f (zmn ) 1
= − ∑ .
2π i S (ζ − z)g(ζ ) g(z) zmn ∈U g (zmn ) z − zmn
we see that the integral on the left-hand side above tends to 0 as r → ∞. This proves
the desired expansion for f .
4.5 Sampling Sequences 165
We say that a set Z in C is a set of uniqueness for Fαp if every function in Fαp that
vanishes on Z must be identically zero. Recall that a sequence Z is a zero set for Fαp
if there exists a function f ∈ Fαp whose zero set is exactly Z. Thus, a zero sequence
is not a set of uniqueness. But we cannot say that Z is a set of uniqueness if and only
if Z is not a zero set for Fαp . It is obvious that each sampling sequence for Fαp is a
set of uniqueness for Fαp . We use the convention that the empty set is not a set of
uniqueness for Fαp , which is again easy to conceive and accept.
Recall that W (Z) is the collection of weak limits of all the translates Z + z of Z.
Lemma 4.25. A separated sequence Z is sampling for Fα∞ if and only if every A ∈
W (Z) is a set of uniqueness (and hence nonempty) for Fα∞ .
Proof. First assume that Z is a sampling sequence. Let A ∈ W (Z) be the weak
limit of some sequence An = Z + ζn , ζn ∈ C. Although the set A may not be a
sequence, it follows from the proof of Proposition 4.14 and the translation invariance
of M∞ (Z) that
M∞ (A) ≤ lim inf M∞ (An ) = M∞ (Z) < ∞,
n→∞
where M∞ (A), just as in the case of sequences, is the smallest M such that
α 2
f ∞,α ≤ M sup | f (z)|e− 2 |z| : z ∈ A
for all f ∈ Fα∞ . So A is a sampling set for Fα∞ . In particular, A is a set of uniqueness
for Fα∞ .
Next, assume that Z is not sampling for Fα∞ . Then there exists a sequence { fn } of
unit vectors in Fα∞ such that fn |Z∞,α → 0 as n → ∞. For each n, we use continuity
to find some zn ∈ C such that
1
| fn (zn )|e−α |zn |
2 /2
= .
2
Let
α
gn (z) = fn (z + zn )e−α zn z− 2 |zn | .
2
Also,
lim gn |An ∞,α = lim fn |Z∞,α = 0.
n→∞ n→∞
Using the intermediate value theorem for continuous functions, we can also find a
point ζn ∈ C such that
α + εn 1
| fn (ζn )|e− 2 |ζn |
2
= .
2
Let
α + εn
gn (z) = fn (z + ζn )e−(α +εn )ζ n z− 2 |ζn |
2
, n ≥ 1.
Then
1
gn ∞,α +εn = fn ∞,α +εn = 1, |gn (0)| = .
2
Note that
gn∞,α +ε1 ≤ gn ∞,α +εn = 1
for all n. With the help of a normal family argument and passing to a subsequence of
{gn } if necessary, we may assume that gn (z) → g(z) uniformly on compact subsets.
The limit function g is entire, and |g(0)| = 1/2. For any z ∈ C, we have
α α + εn 2
e− 2 |z| |g(z)| = lim e− 2 |z|
2
|gn (z)| ≤ lim gn ∞,α +εn = 1.
n→∞ n→∞
4.5 Sampling Sequences 167
Since Z is separated, we have infn δ (Zn ) > 0. By Proposition 4.12, {Zn } contains a
weakly convergent subsequence. Let A be the weak limit of some sequence {Znk }.
Then A ∈ W (Z).
If A is empty, it cannot be a set of uniqueness. Assume A = 0/ and fix some
point a ∈ A. For any positive integer j, there exists some point w j ∈ Znk j such that
|a − w j | < 1/ j. By Corollary 4.7, there exists a positive constant C such that
α + ε nk j 2 α + ε nk
− 2 |a| − 2 j |w j |2
e |gnk j (a)| − e |gnk j (w j )| < C|a − w j |
for all j. Letting j → ∞ leads to g(a) = 0. This shows that g ∈ Fα∞ , g(0) = 0, but g
vanishes on A. So A is not a set of uniqueness for Fα∞ . This contradicts Lemma 4.25
as we are assuming that Z is a sampling sequence for Fα∞ .
Lemma 4.27. For any fixed positive number r, the sequence {σk (r)} defined by
α r2
1
σk (r) = t k e−t dt
k! 0
It follows that
∞ ∞
1 k −t k −t
σk (r) = t e dt − t e dt
k! 0 α r2
1
= k! − Γ (k + 1, α r2 )
k!
k
(α r2 ) j
= 1 − e− α r ∑
2
,
j=0 j!
168 4 Interpolating and Sampling Sequences
where n(Z, B(w, r)) is the number of points in Z ∩ B(w, r). So the assumption
D− (Z) = α /π (1 + 2ε ) implies that there exist sequences {rn } and {wn } such that
rn → ∞ and
n(Z, B(wn , rn )) α
< , n ≥ 1.
2
rn 1+ε
Let
√ √
Rn = rn / 1 + ε Bn = B(0, rn ) = B(0, 1 + ε Rn ),
and
√
Nn = n(Z, B(wn , rn )) = n(Z, B(wn , 1 + ε Rn )).
α rn2
≤ Nn < α R2n .
1 + 2ε
In particular, Nn → ∞ as n → ∞.
To simplify the notation, we fix any n and write B = Bn , R = Rn , and N = Nn .
Let p = pn be “the” (unique up to a unimodular constant multiple) polynomial with
(Z − wn ) ∩ Bn as its zero set, normalized so that p2,α = 1.
We can write
N
αk k N
p(z) = ∑ ak fk (z), fk (z) = z, ∑ |ak |2 = 1.
k=0 k! k=0
It is easy to see that the functions { fk } are also orthogonal over the disk B:
√
fk (z) fm (z) dλα (z) = σk ( 1 + ε R)δk,m ,
B
4.5 Sampling Sequences 169
where the constants σk are from Lemma 4.27. It follows from this and Lemma 4.27
that
N
B
|p(z)|2 dλα (z) = ∑ |ak |2 B
| fk (z)|2 dλα (z)
k=0
N √ N √
= ∑ |ak |2 σk ( 1 + ε R) ≥ ∑ |ak |2 σN ( 1 + ε R)
k=0 k=0
√
1 α (1+ε )R N −t
2
= σN ( 1 + ε R) = t e dt
N! 0
(1+ε )N (1+ε )N
1 1
≥ t N e−t dt ≥ t N e−t dt
N! 0 N! N
(1+ε )N
NN N N e−N
≥ e−t dt = (1 − e−ε N ).
N! N N!
shows that there exists a constant C = C(α , ε ) > 0 (independent of N) such that
C C
|p(z)|2 dλα (z) ≥ √ ≥ √ .
B N αR
Since
α α 2 2
|p(z)|2 dλα (z) ≤ (1 + ε )R2 sup p(z)e− 2 |z| ,
B π z∈B
|z|2 = (1 + t)R2, t ≥ ε,
that
αk
| fk (z)|2 e−α |z| = (1 + t)k R2k e−α (1+t)R
2 2
k!
(α R2 )k e−α R −α tR2 +k log(1+t)
2
= e
k!
≤ e−α tR
2 +k log(1+t)
≤ e−α tR
2 +N log(1+t)
≤ e−α tR
2 +α R2 log(1+t)
.
for all 0 ≤ k ≤ N and z outside B. By the Cauchy–Schwarz inequality and the fact
that ∑Nk=0 |ak |2 = 1, we have
2
N
2 −α |z|2
= ∑ ak fk (z) e−α |z|
2
|p(z)| e
k=0
N N
∑ |ak |2 ∑ | fk (z)|2 e−α |z|
2
≤
k=0 k=0
≤ (N + 1)e−2cR ≤ (α R2 + 1)e−2cR .
2 2
≤ α R2 + 1e−cR ,
2
where Zn = Z − wn .
Finally, if we set
α
gn (z) = eα wn z− 2 |wn | pn (z − wn ),
2
then
− 32
gn ∞,α = pn ∞,α ≥ CRn
and
$
α R2n + 1e−c|Rn |
2
gn |Z∞,α = pn |Zn ∞,α ≤
4.5 Sampling Sequences 171
so that
gn |Z∞,α 5
≤ C Rn2 e−c|Rn |
2
gn ∞,α
for all n ≥ 1, where C and c are positive constants independent of n. Since Rn → ∞
as n → ∞, we conclude that
gn|Z∞,α
lim = 0.
n→∞ gn ∞,α
This contradicts with the assumption that Z is a sampling sequence for Fα∞ and
completes the proof of the lemma.
Lemma 4.29. Suppose 0 < p ≤ ∞ and Z is a sampling sequence for Fαp . Then Z is
a set of uniqueness for Fα∞ .
Proof. By Lemmas 4.10 and 4.11, we may assume that Z is separated.
The case p = ∞ is obvious. Suppose 0 < p < ∞, Z is sampling for Fαp , but Z is
not a set of uniqueness for Fα∞ . Then there exists a function f ∈ Fα∞ , not identically
zero, such that f vanishes on Z. Let g(z) = f (rz), where 0 < r < 1. Then g ∈ Fαp ,
g is not identically zero, and g vanishes on Z/r. This is impossible because by
Corollary 4.17, the sequence Z/r is sampling for Fαp when r is sufficiently close
to 1. Therefore, Z must be a set of uniqueness for Fα∞ .
Lemma 4.30. Suppose 0 < p < ∞ and Z is sampling for Fαp . Then D− (Z) > α /π .
Proof. Again, by working with a subsequence of Z if necessary, we may assume
that Z is already separated.
Recall that W (Z) consists of all weak limits of translates of Z. Since every
translation of Z is also a sampling sequence for Fαp with the same separation
constant, it follows from Proposition 4.16 that every sequence in W (Z) is sampling
for Fαp as well. Combining this with Lemmas 4.25 and 4.29, we conclude that Z is a
sampling sequence for Fα∞ . Thus, D− (Z) > α /π by Lemma 4.28.
This completes the proof for the necessity of the sampling condition D− (Z) >
α /π for Fαp . We now proceed to prove the sufficiency. This will be accomplished
with the help of the Weierstrass σ -function and its variant g(z) discussed in
the previous section. The first step is to show that every sequence contains a
subsequence that is uniformly close to a square lattice Λγ and whose uniform lower
density changes very little.
Lemma 4.31. Suppose 0 < α < β and Z is a sequence with D− (Z) = β /π . There
exists a subsequence Z of Z such that Z is uniformly close to Λγ for some
α < γ < β.
Proof. Fix γ ∈ (α , β ) and choose ε > 0 such that γ + ε < β . The condition D− (Z) =
β /π implies that there exists a positive number r such that any square of side length
r contains at least (γ + ε )r2 /π points from Z.
172 4 Interpolating and Sampling Sequences
We decompose C into the disjoint union of a sequence of squares (half open, half
closed) of side length r: C = ∪{Sk : k ≥ 1}. Since the area of each Sk is r2 and the
area of the fundamental region of Λγ is π /γ , each Sk contains r2 /(π /γ ) = γ r2 /π
points from Λγ (plus or minus a few points that can be neglected for our purpose).
But Sk contains at least (γ + ε )r2 /π points from Z. So for each k, we can choose
|Λγ ∩ Sk | points from Z to match those in Λγ ∩ Sk . We do this for each k, and the
result is a subsequence
√ of Z that is uniformly √ close to Λγ . More specifically, we
have |zmn − ωmn | ≤ 2 r for all m and n, where 2 r is the length of the diagonal of
each Sk .
We now prove that the condition D− (Z) > α /π is sufficient for a separated
sequence Z to be a sampling sequence of Fαp . For clarity, we break the proof into
three cases: 0 < p ≤ 1, 1 < p < ∞, and p = ∞.
Lemma 4.32. Suppose 1 < p < ∞ and Z is a separated sequence in C. If D− (Z) >
α /π , then Z is a sampling sequence for Fαp .
Proof. Given a function f ∈ Fαp ⊂ Fα∞ we need to estimate the integral
α 2 p
I= f (z)e− 2 |z| dA(z)
C
from above. By Lemma 4.31, we may assume that Z is uniformly close to a square
lattice Λβ with β > α . Let Ω = Rα be the fundamental region for the square lattice
Λα = {ωmn } = {−ωmn }. Then by Lemma 1.13 and a change of variables, we have
p
− α2 |z|2
I=∑ e Wωkl f (z) dA(z),
k,l Ω
where Wωkl are the Weyl unitary operators defined in Sect. 2.6.
To estimate each summand on the right-hand side above, first observe that Z + ωkl
is uniformly close to Λβ as well, with a constant Q that is independent of k and l.
Thus, we can use Proposition 4.24 to write
where gωkl is the Weierstrass σ -type function associated to the sequences Z + ωkl
and Λβ .
For ε = (β − α )/2, we can write
α
e−ε |zmn +ωkl | e− 2 |zmn | | f (zmn )|
2 2
|Wωkl f (zmn + ωkl )|
= β . (4.17)
|gωkl (zmn + ωkl )|
e− 2 |zmn +ωkl | |gωkl (zmn + ωkl )|
2
4.5 Sampling Sequences 173
Let q = p/(p − 1) so that 1/p + 1/q = 1. Then by Hölder’s inequality and (4.11) in
Lemma 4.21, we see that |Wωkl f (z)| p is less than or equal to
α p
Ch(z) ∑ e− 2 |zmn | f (zmn ) e−ε |zmn +ωkl | +c|zmn +ωkl | log |zmn +ωkl | ,
2 2
m,n
where
q qp
gωkl (z)
h(z) = hkl (z) = ∑e −ε |zmn +ωkl |2
z − zmn − ω
kl
.
m,n
By Lemmas 1.12 and 4.22, the positive function h(z) is bounded on Ω with an upper
bound that is independent of k and l. In particular, the integral
pα 2
h(z)e− 2 |z| dA(z)
Ω
k,l m,n
α p
= C ∑ e− 2 |zmn | f (zmn ) ∑ e−ε |zmn +ωkl | +c|zmn +ωkl | log |zmn +ωkl |
2 2
m,n k,l
≤C f |Z pp,α ,
which is the desired estimate. Note that the last estimate above follows from
Lemma 1.12.
Lemma 4.33. Suppose 0 < p ≤ 1 and Z is a separated sequence with D− (Z) >
α /π . Then Z is sampling for Fαp .
Proof. With notation from the proof of the previous lemma, we use the assumption
0 < p ≤ 1 to get
W f (z + ω ) p g (z) p
ωkl mn kl ωkl
|Wωkl f (z)| ≤ ∑
p
.
m,n g ωkl
(zmn + ω kl ) z − zmn − ωkl
Combining this with (4.11) and (4.17), we obtain positive constants C and c, both
independent of k and l, such that
p p
gωkl (z) − α |z |2
|Wωkl f (z)| ≤ C ∑
p e 2 mn f (zmn ) E(m, n, k, l),
z − zmn − ω
m,n kl
174 4 Interpolating and Sampling Sequences
where
E(m, n, k, l) = e−pε |zmn +ωkl | mn +ωmn | log |zmn +ωkl |
2 +c|z
.
Integrate the above inequality over Ω with respect to e−pα |z| /2 dA(z) and notice that
2
Ω
z − zmn − ω
kl
for some constant C > 0 that is independent of k and l. We obtain another constant
C > 0 such that
α p
I ≤ C ∑ ∑ e− 2 |zmn | f (zmn ) E(m, n, k, l)
2
k,l m,n
α p
= C ∑ e− 2 |zmn | f (zmn ) ∑ E(m, n, k, l)
2
m,n k,l
α p
≤ C ∑ e− 2 |zmn | f (zmn )
2
m,n
= C f |Z pp,α ,
≤ C f |Z∞,α ∑ e(m, n, k, l)
m,n
≤ C f |Z∞,α ,
This completes the proof of the sufficiency of the condition D− (Z) > α /π for Z
to be a sampling sequence of Fαp . We summarize the main results of this section as
the following two theorems.
Theorem 4.35. A set Z is sampling for Fα∞ if and only if Z contains a separated
sequence Z such that D− (Z ) > α /π .
Theorem 4.36. Let Z be a sequence in C and 0 < p < ∞. Then, Z is sampling for
Fαp if and only if Z is the union of finitely many separated sequences and Z contains
a separated subsequence Z such that D− (Z ) > α /π .
Corollary 4.37. If Z is separated and 0 < p ≤ ∞, then Z is sampling for Fαp if and
only if D− (Z) > α /π .
176 4 Interpolating and Sampling Sequences
4.6 Interpolating Sequences 177
we claim that the interpolation problem f (zkl ) = akl is solved explicitly by the
function
gmn (z − zmn )
f (z) = ∑ amn eα zmn z−α |zmn |
2
, (4.18)
m,n z − zmn
where gmn denotes the generalized Weierstrass σ -function associated with the
sequences Z − zmn and Λγ as given in (4.8) (it is easy to see that Z − zmn is uniformly
close to Λγ ). More specifically,
gmn (z − zmn )
z − zmn
178 4 Interpolating and Sampling Sequences
is equal to
z − zmn z − zmn 1 (z − zmn )2
∏ 1 − zkl − zmn exp zkl − zmn + 2 ω 2 .
(k,l)=(m,n) kl
In particular,
m,n z − zmn
m,n
α δ
≤ C ∑ e− 2 |zmn | amn e− 2 |z−zmn |
2 2
m,n
for all z ∈ C, where δ = (α − β )/2. Since Z = {zmn } is uniformly close to the square
lattice Λβ = {ωmn }, we can find another positive constant C such that
α
α 2
e− 2 |z| | f (z)| ≤ C ∑ e− 2 |zmn | amn e−σ |z−ωmn |
2 2
(4.19)
m,n
α
for all z ∈ C, where σ = δ /4. Since the sequence {e− 2 |zmn | amn } is bounded, it
2
follows from (4.19) and Lemma 1.12 that the series in (4.18) converges absolutely
to an entire function f with f (zkl ) = akl for all (k, l).
It remains for us to show that the function f defined in (4.18) belongs to Fαp . Just
as in the previous section, we break the proof into three cases: 0 < p ≤ 1, 1 < p < ∞,
and p = ∞.
The case p = ∞ is the easiest. In fact, if the sequence e−α |zmn | /2 amn is bounded,
2
m,n
m,n
Integrate term by term and use the translation invariance of the area measure. We
see that
p p
− α2 |z|2 − α2 |zmn |2
∑ amn e−pσ |z| dA(z).
2
e f (z) dA(z) ≤ C e
C m,n C
This shows that f ∈ Fαp whenever the series {amn e−α |zmn | /2 } is in l p .
2
The case 1 < p < ∞ follows from complex interpolation. In fact, examining the
arguments in the previous two paragraphs, we see that the linear operator
m,n
maps l ∞ to L∞ (C, dA) and l 1 to L1 (C, dA). Therefore, this operator maps l p to
L p (C, dA) for any 1 < p < ∞. This, along with (4.19), shows that f ∈ Fαp whenever
the sequence {amn e−α |zmn | /2 } belongs to l p .
2
The lemma above shows that the condition D+ (Z) < α /π is sufficient for a
separated sequence Z to be interpolating for Fαp . Next, we will prove that this density
condition is also necessary.
Lemma 4.40. Let 0 < p ≤ ∞. There is no sequence in C that is both sampling for
Fαp and interpolating for Fαp .
Proof. Assume the contrary and let Z be a sequence that is both sampling and
interpolating for Fαp . Then Z is separated and sampling for Fαp+ε for all sufficiently
small ε , because we have characterized sampling sequences for Fαp using the “open”
condition D− (Z) > α /π .
Fix a point ζ ∈ Z and use the assumption that Z is interpolating for Fαp to find
a function g ∈ Fαp such that g(ζ ) = 1 and g(z) = 0 for all z ∈ Z − {ζ }. Then, the
function f (z) = (z − ζ )g(z) is not identically zero, belongs to Fαp+ε , and vanishes
on Z. Thus, Z cannot possibly be sampling for Fαp+ε . This contradiction shows that
Z cannot be simultaneously sampling and interpolating for Fαp .
Lemma 4.41. Suppose 0 < p ≤ ∞ and Z is interpolating for Fαp . If Z is a set of
uniqueness for Fαp , then it must be a sampling sequence for Fαp .
Proof. Since Z is interpolating for Fαp , it must be separated by Lemma 4.8. Given
any function f ∈ Fαp , the sequence wn = f (zn ) has the property that {wn e−α |zn | /2 } ∈
2
l p . By the definition of N p (Z), there exists some function g ∈ Fαp such that g(zk ) = wk
for all k and g p,α ≤ N p (Z)g|Z p,α . Since Z is a set of uniqueness for Fαp and
f (zk ) = wk = g(zk ) for all k, we must have g = f , and so f p,α ≤ N p (Z) f |Z p,α
for all f ∈ Fαp . This says that Z is sampling for Fαp .
180 4 Interpolating and Sampling Sequences
for all n, we see that f (ζ ) = 0. The function (z − ζ ) f (z) is not identically zero,
vanishes on Z, and belongs to Fα +ε for any ε > 0. This contradicts the fact that Z is
p
p
a sampling sequence for Fα +ε .
Thus, sampling sequences for Fαp are stable under the following two operations:
deleting a finite number of points or adding any number of separated points from
outside the sequence.
Corollary 4.44. Let 0 < p ≤ ∞. If Z = {zn } is an interpolating sequence for Fαp ,
then so is Z ∪ {ζ } for any ζ ∈ Z.
Proof. By Corollary 4.42, there is a function g ∈ Fαp that is not identically zero but
vanishes on Z. By dividing out an appropriate power of z − ζ if necessary (which
preserves membership in Fαp ), we may assume that g(ζ ) = 0. Multiplying g by a
constant if necessary, we may further assume that g(ζ ) = 1.
Given a sequence {v} ∪ {vn } of values with {vn e−α |zn | /2 } ∈ l p , we can find a
2
p
function f ∈ Fα such that f (zn ) = vn for all n. The function
F(ζ ) = v, F(zn ) = vn , n ≥ 1.
where Z = {zn } and the supremum is taken over all unit vectors f in Fαp such that
f (zn ) = 0 for all n. We think of ρ p (z, Z) as some kind of distance from z to the
sequence Z. A normal family argument shows that the supremum in the definition
of ρ p (z, Z) is always attained.
By Corollary 2.8, we always have 0 ≤ ρ p (z, Z) ≤ 1. It is obvious that ρ p (z, Z) = 0
when z ∈ Z. We are going to show that ρ p (z, Z) = 0 only when z ∈ Z, provided that
Z is an interpolating sequence for Fαp .
Lemma 4.45. If Z is interpolating for Fα , where 0 < p ≤ ∞, then ρ p (z, Z) > 0 when
p
z ∈ Z.
Proof. Actually, we only need to assume that Z is not a set of uniqueness (we
already know that every interpolating sequence for Fαp is not a set of uniqueness
for Fαp ). In fact, if f is any function in Fαp that is not identically zero and vanishes
on Z, then f cannot possibly have a zero at z of infinite order. Therefore, by dividing
out a finite and nonnegative power of w − z, which does not ruin membership in Fαp ,
we arrive at a function in Fαp that vanishes on Z but has a nonzero value at z.
The following result is a quantitative version of Corollary 4.44.
Lemma 4.46. Let Z = {z1 , z2 , · · · } and z0 ∈ Z. We have
1 + 2N p(Z)
N p (Z ∪ {z0 }) ≤
ρ p (z0 , Z)
for all 0 < p ≤ ∞.
Proof. We may assume that N p (Z) < ∞, that is, Z is an interpolating sequence for
Fαp . Given a sequence of values {v0 , v1 , v2 , · · · } with the l p norm of
α α α
v0 e− 2 |z0 | , v1 e− 2 |z1 | , v2 e− 2 |z2 | , · · ·
2 2 2
equal to 1, there is a function f ∈ Fαp such that f (zn ) = vn for all n ≥ 1 and
On the other hand, by Lemma 4.45, there exists a function f0 ∈ Fαp such that f0
vanishes on Z, f0 p,α ≤ 1, and
α
e− 2 |z0 | f0 (z0 ) = ρ p (z0 , Z).
2
v0 − f (z0 ) α
f0 (z)e− 2 |z0 |
2
g(z) = f (z) +
ρ p (z0 , Z)
belongs to Fαp , solves the interpolation problem g(zn ) = vn for all n ≥ 0, and satisfies
≤ N p (Z) +
ρ p (z0 , Z)
1 + f p,α
≤ N p (Z) +
ρ p (z0 , Z)
1 + N p(Z)
≤ N p (Z) +
ρ p (z0 , Z)
1 + 2N p(Z)
≤
ρ p (z0 , Z)
1 + 2N p(Z)
= g|(Z ∪ {z0 }) p,α .
ρ p (z0 , Z)
N p (Zn , α ) ≤ l0 , d(zn , Zn ) ≥ δ0 , n ≥ 1,
and ρ p (zn , Zn ) → 0 as n → ∞.
By translation invariance, we may assume that each zn = 0. Going down to a
subsequence if necessary, we may also assume that Zn converges weakly to Z ,
where Z may be empty.
By Lemma 4.18, N p (Z , α ) ≤ l0 . Also, d(0, Zn ) ≥ δ0 shows that 0 is not in Z . By
Lemma 4.45, there exists a function f ∈ Fαp such that f vanishes on Z , f p,α ≤ 1,
and f (0) = r > 0. We may further assume that
α
lim f (z)e− 2 |z| = 0.
2
(4.20)
z→∞
4.6 Interpolating Sequences 183
In fact, the above condition is automatically satisfied for f ∈ Fαp when 0 < p < ∞.
If p = ∞, we modify the construction above as follows. Pick a complex number ζ
such that ζ ∈ Z and ζ = 0. Then Z ∪ {ζ } is still an interpolating sequence for Fαp .
Thus, there exists a function g ∈ Fαp such that g vanishes on Z ∪ {ζ } and g(0) = 0.
Then the function f (z) = g(z)/(z − ζ ) belongs to Fαp , vanishes on Z , satisfies the
condition in (4.20), and f (0) = 0.
Since {Zn } converges weakly to Z , the sequence εn = f |Zn p,α converges to 0
as n → ∞, which follows easily from (4.20). Now, choose gn ∈ Fαp with gn = f on
Zn and gn p,α ≤ l0 εn and define
f (z) − gn (z)
fn (z) = .
f p,α + l0 εn
as n → ∞, we also have
r
ρ p (0, Zn ) ≥ | fn (0)| → > 0,
f p,α
which is a contradiction.
Lemma 4.48. Given positive constants l0 and α , there is a constant C = C(l0 , α ) >
0 such that if N p (Z, α ) ≤ l0 , then
log ρ p (z, Z) dA(z) ≥ −C|Q|2
Q
Lemma 4.49. Suppose 0 < p ≤ ∞ and Z is an interpolating sequence for Fαp . Then,
D+ (Z) < α /π .
Proof. We consider an arbitrary large square Q of side length R > 2 and divide it
into N = [R] × [R] squares Q j , 1 ≤ j ≤ N, each of side length s = R/[R], where [R]
denotes the integer part of R. It is clear that 1 ≤ s ≤ 2.
Since Z is interpolating for Fαp , it is separated. Thus, for each j, we can find
some point z j ∈ Q j such that d(z j , Z) ≥ δ0 , where δ0 is a positive constant that only
depends on N p (Z) and α . Let Z j = Z ∪ {z j } and use Lemmas 4.46 and 4.47 to find a
positive constant l, independent of j, such that N p (Z j ) ≤ l for all j. By Lemma 4.48,
we can find a positive constant C = C(l, α ) such that
log ρ p (z, Z j ) dA(z) ≥ −C, 1 ≤ j ≤ N.
Qj
α r2 r r
≤ − ∑ log+ − log √
2 ζ ∈Z
|z − ζ | 2 2
α r2 r r
≤ − ∑ log+ − log √ ,
2 ζ ∈Z∩Q−
|z − ζ | 2 2
where Q− is the square of side length R− 2r inside Q sharing the same center with Q
and having sides parallel to the corresponding ones of Q. In other words, Q− consists
of those points whose distance to the complement of Q exceeds r. We integrate this
inequality with respect to area measure over Q j , use Lemma 4.48, and obtain
−C ≤ log ρ p (z, Z j ) dA(z)
Qj
α r2 r r
≤ |Q j | − ∑ log+ dA(z) − |Q j | log √ .
2 ζ ∈Z∩Q− Qj |z − ζ | 2 2
where n(Z, Q− ) denotes the number of points from Z contained in Q− . This can be
rewritten as
n(Z, Q− ) α 2 r 2C R2
≤ − log √ + . (4.21)
(R − 2r)2 π π r2 2 2 π r2 (R − 2r)2
α 2 r 2C
D+ (Z) ≤ − log √ + 2 .
π π r2 2 2 πr
4.7 Notes
The main results of this chapter are due to Seip and Wallsten, and our presentation
follows their papers [206] and [209] very closely. In turn, those two papers follow
Beurling’s 1977–1978 lectures on balayage and interpolation at the Mittag–Lefler
Institute very closely. In particular, the density notion introduced in Sect. 4.1 can be
found in Beurling’s lectures [36].
We chose to follow the more classical and original arguments of estimating
certain perturbations of the Weierstrass σ -function because this is more in line
with the traditional approaches to entire functions. But we point out that there are
now more modern and more powerful techniques for interpolation and sampling
problems that work in much more general settings. For example, many ideas used in
[203,208] to characterize interpolating and sampling sequences for Bergman spaces
can be adapted to work for Fock spaces as well.
See [205] for a complete description of interpolating and sampling sequences
for Bergman spaces on the unit disk. The books [78, 119, 203] contain more details
about the Bergman space results than Seip’s original papers. The interested reader
will find many additional papers in the bibliography about various interpolation and
sampling problems.
188 4 Interpolating and Sampling Sequences
4.8 Exercises 189
4.8 Exercises
interpolation problem
% &
inf f p,α : f (zn ) = vn , n ≥ 1
for all z ∈ C.
7. Show that the incomplete gamma function has the property that
k
zj
Γ (k + 1, z) = k!e−z ∑
j=0 j!
as N → ∞.
9. Show that
1 1
∑ (n2 + m2 )3/2
∼
N
n2 +m2 >N 2
as N → ∞.
10. Let δ be a positive number. Show that for any w ∈ C, we have
1
∑ |ωmn − w|
: δ < |ωmn − w| < R ∼ R
D+ (Z) = D− (Z) = α /π .
12. Justify the last step in the proof of Lemmas 4.32 and 4.34.
13. If Z is an interpolating sequences for Fαp , then any subset of Z is also an
interpolating sequence for Fαp .
α
14. Show that |σα (ωmn )|e− 2 |ωmn | is a positive constant independent of m and n,
2
where σα is the derivative of σα .
p
15. If Z is sampling for Fα , then adding any separated sequence to Z will create a
p
sampling sequence for Fα again.
16. If Z = {zn } is a sequence in C such that
% & 2
inf |z j − zk | : j = k > √ ,
α
Show that:
(a) T is onto if and only if Z is interpolating for Fαp .
(b) T is bounded below if and only if Z is sampling for Fαp .
(c) T is one-to-one if and only if Z is a uniqueness set for Fαp .
Prove or disprove that T has closed range if and only if Z is either interpolating
or sampling for Fαp .
19. Suppose Z = {zn } is separated, 1 ≤ p < ∞, and 1/p + 1/q = 1. Then Z is an
interpolating sequence for Fαp if and only if there exists a positive constant c
such that
∞
q
∞
α
∑ ak e− 2 |z−zk | dA(z) ≥ c ∑ |ak |q
2
C k=1 k=1
21. Let μ and ν be two positive measures. If A1 and A2 are two sets that are
measurable with respect to both μ and ν . Show that
ν (A1 ) ν (A2 ) ν (A1 ∪ A2 ) ν (A1 ) ν (A2 )
min , ≤ ≤ max , .
μ (A1 ) μ (A2 ) μ (A1 ∪ A2 ) μ (A1 ) μ (A2 )
22. Make precise the word “roughly” used in the proof of Proposition 4.3.
23. For a sequence Z = {zn } of distinct points in C, show that the following
conditions are equivalent:
(a) Z is sampling for Fα2 .
(b) Atomic decomposition holds on Z.
(c) The operator
∞
∑ f (zn )eα zzn −α |zn |
2
S f (z) = (4.23)
n=1
|zmn − λmn | ≤ Q
for all (m, n), then there exists a positive constant Q = Q (α , β , Q) such that
for any (k, l), there exists some (k , l ) with the property that
∑ e−ε |zmn |
2
≤ C.
m,n
In this chapter, we study zero sets for the Fock spaces Fαp . Throughout this book, we
say that a sequence Z = {zn } ⊂ Ω is a zero set for a space X of analytic functions
in Ω if there exists a function f ∈ X, not identically zero, such that Z is exactly the
zero sequence of f , counting multiplicities.
Recall from Theorem 2.12 that every function f ∈ Fαp is of order 2. Therefore, by
Hadamard’s factorization theorem, the zero sequence {zn } of f , with the origin
removed, must satisfy
∞
1
∑ |zn |3 < ∞.
n=1
In this section, we improve upon this estimate and obtain the following necessary
condition for a sequence {zn } to be a zero set for Fαp .
Theorem 5.1. Suppose 0 < p ≤ ∞ and {zn } is the zero sequence of a function f ∈
Fαp with f (0) = 0. Then√ there exist a positive constant c and a rearrangement of
{zn } such that |zn | ≥ c n for all n.
Proof. Without loss of generality, we may assume that f (0) = 1 and p = ∞. Let
{zn } denote the zero sequence of f , repeated according to multiplicity and arranged
so that 0 < |z1 | ≤ |z2 | ≤ |z3 | ≤ · · · .
Fix any positive radius r such that f has no zero on |z| = r and let n(r) denote
the number of zeros of f in |z| < r. By Jensen’s formula,
n(r) 2π
r 1
∑ log =
|zk | 2π 0
log | f (reiθ )| dθ .
k=1
It follows that
n(r)
r α 2
∑ log |zk | ≤ 2
r + C,
k=1
n(r)
r
α
∏ |zk | ≤ exp 2
r2 + C
k=1
for all positive integers n and all r > 0 such that f has no zero on |z| = r. Since
{|zk |} is nondecreasing, we have
rn
α
≤ exp r 2
+ C ,
|zn |n 2
or
1 1 α 2 C
≤ exp r + , (5.1)
|zn | r 2n n
where n is any positive integer and r is any radius such that f has no zero on |z| = r.
There are only a countable number of radius r such that f has zeros on |z| = r.
Therefore,
√ for any positive integer n, we can choose a sequence {rk } such that rk →
n as k → ∞ and f has no zero on each |z| = rn . Combining this with (5.1), we
conclude that
1 1 α log f ∞,α
≤ √ exp + , n ≥ 1.
|zn | n 2 n
√
It is then clear that there is some positive constant c such that |zn | ≥ c n for all
n ≥ 1.
Note that the assumption f (0) = 0 is not a critical one. In fact, if f ∈ Fαp and it
has a zero of order m at the origin, then the function g defined by g(z) = f (z)/zm is
in Fαp and does not vanish at the origin.
Corollary 5.2. Suppose 0 < p ≤ ∞ and {zn } is the zero sequence of some f ∈ Fαp
with f (0) = 0. Then
∞
1
∑ |zn |r < ∞
n=1
for all n ≥ 1.
5.2 A Sufficient Condition 197
The purpose of this section is to prove the following sufficient condition for zero
sequences of Fαp .
Theorem 5.3. Suppose that {zn } is a sequence of complex numbers such that
∞
1
∑ |zn |2 < ∞. (5.2)
n=1
where E1 (z) = (1 − z)ez . By Theorem 1.6, f is entire, and {zn } is the zero sequence
of f . We will show that this function f belongs to all the Fock spaces Fαp , where
0 < p ≤ ∞ and α > 0.
If |z| < 1/2, we have
|E1 (z)| ≤ (1 + |z|)e|z| , log |E1 (z)| ≤ |z| + log(1 + |z|), (5.3)
for all z. It follows that for any positive A, there exists a positive number R such that
On the annulus 1/2 ≤ |z| ≤ R, the function |z|2 log |E1 (z)| is continuous except at
z = 1, where it tends to −∞. Hence, there is a constant B such that
1
log |E1 (z)| ≤ B|z|2 , ≤ |z| ≤ R.
2
Combining the estimates from the last three paragraphs, we conclude that
for all z ∈ C. Using (5.3) again, we can find some r1 > 0 such that
ε 2
log |E1 (z)| ≤ |z| , |z| > r1 ,
2S
where
N
1
S= ∑ |zn |2 .
n=1
Set r2 = r1 |zN |. Then |z| > r2 implies that |z/zn | > r1 for 1 ≤ n ≤ N. It follows that
N
ε
∑ log |E1 (z/zn )| ≤ 2 |z|2 , |z| > r2 .
n=1
Therefore,
∞
log | f (z)| = ∑ log |E1(z/zn )| < ε |z|2
n=1
for all |z| > r2 , or | f (z)| < eε |z| for all |z| > r2 . Since ε is arbitrary, we see that
2
sin(δ z2 )
f (z) = .
δ z2
Converting the sine function above to complex exponential functions and using
the assumption that δ < α /2, we easily check that f ∈ Fαp . Therefore, Z is a zero
sequence for Fαp .
Let Z = {eπ i/4 z : z ∈ Z} be a rotation of the sequence Z above. Then Z is also
an Fαp zero sequence. Clearly, Z and Z are disjoint. We now arrange Z ∪ Z into a
single sequence {zn } such that
If {zn } is a zero sequence for Fαp ⊂ Fα∞ , it follows from the proof of Theorem 5.1
that there exists a positive constant C such that
n
r α 2
∏ |zk | ≤ Ce 2 r
k=1
for all n ≥ 1 and r > 0. Square both sides, replace n by 8n, and integrate from 0
to ∞ with respect to the measure re−β r , where β > α . We obtain another positive
2
(8n)! 8n 1
∏ |zk |2 ≤ C
β 8n k=1
200 5 Zero Sets for Fock Spaces
for all n ≥ 1. This clearly implies that 8δ ≤ πβ . Since β can be arbitrarily close to
α , we have δ ≤ πα /8, which is a contradiction. This shows that {zn } is not an Fαp
zero set and completes the proof of the theorem.
Theorem 5.5. Let α > 0 and 0 < p ≤ ∞. There exists an Fαp zero sequence {zn }
and a subsequence {znk } which is not an Fαp zero sequence.
Proof. Fix a positive constant δ such that δ < α /2 and consider the following entire
function:
eiδ z − 1
2
f (z) = .
iδ z2
is an Fαp zero sequence. Let {zn } denote the subsequence consisting of real elements
in the above set. We proceed to show that {zn } is not an Fαp zero set.
Again, aiming to arrive at a contradiction later, we assume that g is a function in
Fαp that vanishes precisely on {zn }. It is clear that ρ1 (g) = m(g) = 2; see Sect. 1.1
for definitions and properties of these numbers. By Theorem 1.10, we always have
ρ (g) ≥ ρ1 (g), so g must be of order greater than or equal to 2. Combining this with
Theorem 2.12, we conclude that g must be of order 2. By Lindelöf’s theorem (see
Theorem 1.11), the function g must be of maximum (infinite) type since the sums
1
S(r) = ∑ z2
∼ log r, r > 1,
|zn |≤r n
are clearly unbounded. By Theorem 2.12 again, the function g cannot possibly be in
p p
Fα . This contradiction shows that {zn } is not an Fα zero set.
We now consider zero sets for different Fock spaces. The Weierstrass σ -functions
play a significant role here.
5.3 Pathological Properties 201
where the product is taken over all integers m and n with ωmn = 0.
Lemma 5.6. Let 0 < α1 < α < α2 < ∞. We have:
(a) σα ∈ Fαp2 for all 0 < p ≤ ∞.
(b) σα ∈ Fαp1 for any 0 < p ≤ ∞.
(c) σα ∈ Fα∞ .
(d) σα ∈ fα∞ , and so σα ∈ Fαp for any 0 < p < ∞.
Since the function |σα (w)|e−α |w| /2 is bounded on the relatively compact set Ωα ,
2
This clearly implies that σα ∈ Fα∞ and σα ∈ Fαp2 for all 0 < p ≤ ∞.
If S is any compact set contained in the fundamental region of Λα , then there
exists a positive constant δ such that
α
|σα (w)|e− 2 |w| ≥ δ ,
2
w ∈ S.
for all (m, n). This clearly shows that σα ∈ fα∞ . Since Fαp ⊂ fα∞ for 0 < p < ∞, we
have σα ∈ Fαp for any 0 < p < ∞. Also, Fαp1 ⊂ fα∞ for all 0 < p ≤ ∞. So σα ∈ Fαp1
for all 0 < p ≤ ∞.
Lemma 5.7. Suppose 0 < p < ∞ and f ∈ Fαp . If f (z) = 0 for all z ∈ Λα , then f is
identically zero.
Proof. By the Weierstrass factorization theorem, we can write f = hσα , where h is
an entire function. In view of the quasiperiodicity of σα , we have
α 2 p α 2 p
f (z)e− 2 |z| dA(z) = ∑ |h(z + ωmn )| p σα (z)e− 2 |z| dA(z),
C m,n Ωα
Since the function z → |h(z + ωmn )| p is subharmonic, there exists a positive constant
δ (independent of (m, n)) such that
|h(z + ωmn )| p dA(z) ≥ δ |h(z + ωmn )| p dA(z)
Ωα −D Ωα
for all (m, n). It follows that there is another positive constant C such that
α 2 p
f (z)e− 2 |z| dA(z) ≥ C |h(z)| p dA(z).
C C
Similarly, the sequence Z = Λα − {0} is a zero set for Fαp when p > 2 because
the function f (z) = σα (z)/z belongs to Fαp if and only if p > 2. However, this
sequence Z is not a zero set for Fα2 . To see this, suppose f is a function in Fα2 ,
not identically zero, such that f vanishes on Z. By Weierstrass factorization, we
have f (z) = [σα (z)/z]g(z) for some entire function g that is not identically zero.
Mimicking the proof of Lemma 5.7, we can show that
g(z) 2
|z|>1
z dA(z) < ∞.
It follows from polar coordinates and the Taylor expansion of g that this is
impossible unless g is identically zero. This actually shows that Z = Λα − {0} is
a uniqueness set for Fα2 . In the above arguments, the point 0 can be replaced by any
other point in Λα .
On the other hand, if Z is the resulting sequence when two points a and b are
removed from Λα , then the function
σα (z)
f (z) =
(z − a)(z − b)
belongs to Fα2 and has Z as its zero sequence. Therefore, Z is a zero set for Fα2 .
Consequently, it is possible to go from a uniqueness set to a zero set by removing
just one point. Equivalently, it is possible to add just a single point to a zero set of
Fα2 so that the resulting sequence becomes a uniqueness set for Fα2 . This shows how
delicate the problem of characterizing zero sets for Fαp is.
We can also show by an example that it is generally very difficult to distinguish
q
between zero sets for Fαp and Fα . More specifically, for any positive integer N with
q
N p > 2, if Z is an Fα zero set and if N points {z1 , · · · , zN } are removed from Z, then
the remaining sequence Z is an Fαp zero set. To see this, let Z be the zero sequence of
a function f ∈ Fα , not identically zero, then Z is the zero sequence of the function
q
f (z)
g(z) = ,
(z − z1 ) · · · (z − zN )
which is easily seen to be in Fαp . Therefore, zero sets for Fαp and Fαq may be different,
but they are not too much different.
Let Z be a zero sequence for Fαp and let IZ denote the set of functions f in Fαp
such that f vanishes on Z. In the classical theories of Hardy and Bergman spaces,
the space IZ is always infinite dimensional. This is no longer true for Fock spaces.
Theorem 5.9. For any 0 < p ≤ ∞ and k ∈ {1, 2, · · · } ∪ {∞}, there exists a zero set
Z for Fαp such that dim(IZ ) = k.
Proof. The case k = ∞ is trivial; any finite sequence Z will work. So we assume that
k is a positive integer in the rest of the proof.
204 5 Zero Sets for Fock Spaces
where g is an entire function. For any positive integer n, let Cn be the boundary
of
the square centered at 0 with horizontal and vertical side length (2n + 1) π /α . It
is clear that
d(Cn , Λα ) ≥ π /α /2, n ≥ 1.
This together with the assumption that F ∈ Fα∞ implies that there exists another
positive constant C such that
σα (z)h(z)
F(z) = ,
(z − a1 ) · · · (z − ak−1 )
Next, we assume that 0 < p < ∞ and k is a positive integer. In this case, we let N
denote the smallest positive integer such that N p > 2, or equivalently,
σ (z)e− α2 |z|2 p
α
dA(z) < ∞. (5.6)
|z|>1 zN
Remove any N + k − 1 points {a1 , · · · , aN+k−1 } from Λα and denote the remaining
sequence by Z. Then Z is the zero sequence of the function
σα (z)
,
(z − a1 ) · · · (z − aN+k−1 )
which belongs to Fαp in view of (5.6). In fact, if g is any polynomial of degree less
than or equal to k − 1, then it follows from (5.6) that g times the above function
belongs to IZ .
Conversely, if f is any function in Fαp that vanishes on Z, then we can write
σα (z)g(z)
f (z) = ,
(z − a1 ) · · · (z − aN+k−1 )
where g is an entire function. Since Fαp ⊂ Fα∞ , it follows from (5.5) and Cauchy’s
integral estimates that g is a polynomial with degree less than or equal to N + k − 1.
If the degree of g is j > k − 1, then
g(z) 1
∼ , z → ∞.
(z − a1 ) · · · (z − aN+k−1 ) zN+k−1− j
This together with f ∈ Fαp shows that (5.6) still holds when N is replaced by N + k −
1 − j, which contradicts our minimality assumption on N. Thus, j ≤ k − 1, which
shows that IZ is k dimensional.
The following result describes the structure of IZ when it is finite dimensional.
Theorem 5.10. Suppose Z is a zero set for Fαp and dim(IZ ) = k is a positive integer.
Then there exists a function g ∈ IZ such that IZ = gPk−1 , where Pk−1 is the set of all
polynomials of degree less than or equal to k − 1.
Proof. First, observe that if dim(IZ ) = k < ∞, then Z = Z ∪ {a1 , · · · , ak } is a
uniqueness set for Fαp for all {a1 , · · · , ak }. Here, the union in Z should be understood
in the sense of zero sequences, where multiplicities are taken into account. In fact,
if there exists a function f ∈ Fαp , not identically zero, such that f vanishes on Z ,
then the functions
f (z) f (z)
f (z), , ··· ,
z − a1 z − ak
206 5 Zero Sets for Fock Spaces
all belong to Fαp and vanish on Z. Here again, if zeros of higher multiplicity
are involved, then some obvious adjustments should be made. It is clear that the
functions listed above are linearly independent, so the dimension of IZ is at least
k + 1, a contradiction.
Next, observe that if dim(IZ ) > m, then Z = Z ∪ {a1 , · · · , am } is not a uniqueness
set for Fαp for any collection {a1 , · · · , am }. To see this, pick any m + 1 linearly
independent functions f1 , · · · , fm+1 from IZ , let
f = c1 f1 + · · · + cm+1 fm+1 ,
c1 f1 (a j ) + · · · + cm+1 fm+1 (a j ) = 0, 1 ≤ j ≤ m.
Once again, obvious adjustments should be made when there are zeros of higher
multiplicity. The homogeneous system above has m equations but m + 1 unknowns,
so it always has nonzero solutions c j , 1 ≤ j ≤ m + 1. With such a choice of c j , the
function f is not identically zero but vanishes on Z , so Z is not a uniqueness set.
It follows that if 1 ≤ j < k and Z = Z ∪ {a1 , · · · , a j }, then Z is not a uniqueness
set for Fα . We can actually show that Z is a zero set for Fα . In fact, if f is a function
p p
p
in Fα , not identically zero, such that f vanishes on Z (but not necessarily exactly on
Z ), then the conclusion of the previous paragraph shows that the number of zeros
of f in addition to those in Z cannot exceed k − j. If these additional zeros a are
divided out of f by the appropriate powers of z − a, the resulting function is still in
Fα and vanishes exactly on Z . Thus, Z is a zero set for Fα .
p p
Fix a function g ∈ IZ that has exactly Z as its zero set. If f is any function in IZ ,
not identically zero, then just as in the previous paragraph, we can show that the
zeros of f must be of the form Z = Z ∪ {a1 , · · · , a j }, where j ≤ k − 1. Thus, we
can factor f as follows: f = gPeh , where P ∈ Pk−1 and h is entire. It is clear that
dividing a polynomial out of f , whenever the division is possible, always results in
p
a function in Fα . Therefore, the function geh belongs to IK as well. It follows that
the function ge − g = g(eh − 1) belongs to IZ . If h is not constant, then by Picard’s
h
√
Proof. Fix some δ > 2/ α . For any positive integer k, let Zk denote the set of
k + 1 points evenly spaced in the first quadrant on the circle |z| = kδ , including the
end-points kδ and kδ i. Let
∞
Z= Zk = {z1 , z2 , · · · , zn , · · · }.
k=1
m
1 1 + e−iπ /k m
1 cos(π /2k)
= ∑ (kδ )2 1 − e−iπ /k ∑ (kδ )2 sin(π /2k)
=
k=1 k=1
2i m 1 2i 2i
∼− ∑ k ∼ − πδ 2 log m ∼ − πδ 2 log r
πδ 2 k=1
5.4 Notes
Theorem 5.1, the necessary condition for zero sets of Fock spaces, was obtained in
[249]. Theorem 5.3, the sufficient condition for zero sets of Fock spaces, is classical
and follows from the general theory of entire functions. The proof of Theorem 5.3
here is basically from [67].
The results in Sect. 5.3 were mostly from [249, 258]. The motivation for [249]
was Horowitz’s study of zero sets for Bergman spaces; see [127–129]. The
most intriguing results concerning zero sequences for Fock spaces are probably
Theorems 5.9 and 5.10, which were proved in [258]. One interesting problem that
remains open is the following: if p = q, do Fαp and Fαq always have different zero
sequences?
Lemma 5.7 shows that Λα is a set of uniqueness for Fαp when 0 < p < ∞. This
result as well as its proof are from [209]. Proposition 5.11, which is a little surprising
when compared to the corresponding questions in the Hardy and Bergman space
settings, is from [225].
210 5 Zero Sets for Fock Spaces
5.5 Exercises 211
5.5 Exercises
1. We say that an entire function f (z) belongs to the Nevanlinna–Fock class Fα∗ if
log+ | f (z)| dλα (z) < ∞.
C
Show that the zero sequence {zn } of any function f in Fα∗ with f (0) = 0 satisfies
the following condition:
∞
e−α |zn |
2
∑ |zn |2 < ∞.
n=1
3. Suppose Z is a zero set for Fαp and k is a positive integer. Show that the following
conditions are equivalent:
(a) dim(IZ ) ≤ k.
(b) Z ∪ {a1 , · · · , ak } is a uniqueness set for Fαp for all {a1 , · · · , ak }.
(c) Z ∪ {a1 , · · · , ak } is a uniqueness set for Fαp for some {a1 , · · · , ak }.
4. Suppose Z is a zero set for Fαp and k is a positive integer. Show that the following
conditions are equivalent:
(a) dim(IZ ) = k.
(b) For any {a1 , · · · , ak }, the sequence Z ∪ {a1, · · · , ak−1 } is not a uniqueness
set for Fαp but Z ∪ {a1, · · · , ak } is.
(c) For some {a1 , · · · , ak−1 }, the sequence Z ∪ {a1 , · · · , ak−1 } is not a unique-
ness set for Fαp , but for some {b1 , · · · , bk }, the sequence Z ∪ {b1 , · · · , bk } is
a uniqueness set for Fαp .
5. If Z is a zero set for Fαp , then the sequence remains a zero set for Fαp after any
finite number of points are removed from it.
6. Suppose 0 < p < ∞ and Z is uniformly close to Λα . Show that Z is a uniqueness
set for Fαp .
7. If Z = {zn } is a zero set for Fαp , then
∞
1
∑ |zn |2 log1+ε |zn | < ∞
n=1
for all ε > 0, provided that |zn | = 0, 1. Show that this is false in general if ε = 0.
212 5 Zero Sets for Fock Spaces
8. Suppose {zn } is the zero sequence of a function f ∈ Fαp , where f (0) = 1, 0 <
p < ∞, and {|zn |} is nondecreasing. Show that
C
αe 2
n np
1
∏ |zn | p ≤ √n n
f pp,α
k=1
belongs to F12 but the function z f (z) is no longer in F12 . See [22].
13. If Z is a zero set for Fαp and dim(IZ ) < ∞, then every function in IZ has order 2
and type α /2.
14. If Z is a zero set for Fαp and dim(IZ ) < ∞, then any two functions in IZ whose
zeros are exactly those in Z can only differ by a constant multiple. Thus, there
is essentially just one function that vanishes exactly on Z.
Chapter 6
Toeplitz Operators
There is a rich history of Toeplitz operators, especially those on the Hardy space. In
particular, Toeplitz operators on the Hardy space provide ample examples of shifts,
isometries, and Fredholm operators. They also provide motivating examples in index
theory and the theory of invariant subspaces.
In this chapter, we study Toeplitz operators on the Fock space Fα2 . Problems
considered include boundedness, compactness, and membership in the Schatten
classes. The approach here is more closely related to the theory of Toeplitz operators
on the Bergman space that was developed over the past thirty years or so.
Recall that for any fixed weight parameter α , the orthogonal projection
P : L2α → Fα2
is an integral operator,
P f (z) = K(z, w) f (w) dλα (w),
C
where
K(z, w) = eα zw
Tϕ ( f ) = P(ϕ f ), f ∈ Fα2 .
We call Tϕ the Toeplitz operator on Fα2 with symbol ϕ . It is clear that Tϕ is bounded
with Tϕ ≤ ϕ ∞ .
Proposition 6.1. For any complex numbers a and b, and for any bounded functions
ϕ and ψ , we have:
(i) Taϕ +bψ = aTϕ + bTψ .
(ii) Tϕ = Tϕ∗ .
(iii) Tϕ ≥ 0 if ϕ ≥ 0.
Proof. These follow easily from the definitions. We omit the routine details.
One of the main differences between Toeplitz operators on the Fock space and
those on Hardy and Bergman type spaces is the lack of bounded analytic and
harmonic symbols in the Fock space setting. In fact, by the maximum modulus
principle, if an analytic or harmonic function on C is bounded, it has to be a constant.
By the integral representation for the orthogonal projection P, we can write
Tϕ ( f )(z) = K(z, w) f (w)ϕ (w) dλα (w)
C
α
K(z, w) f (w)e−α |w| ϕ (w) dA(w).
2
=
π C
This motivates us to define Toeplitz operators on Fα2 with much more general
symbols.
216 6 Toeplitz Operators
Note that Tϕ is very loosely defined here, because it is not clear when the integrals
above will converge, even if the measure μ is finite, as the kernel function K(z, w)
is unbounded for any fixed z = 0.
To make things a little more precise, we say that a complex Borel measure μ on
C satisfies condition (M) if
|K(z, w)|e−α |w| d|μ |(w) < ∞
2
(6.1)
C
for all z ∈ C. Because of the exponential form of the reproducing kernel, it is clear
that the above is equivalent to
|K(z, w)|2 e−α |w| d|μ |(w) < ∞
2
(6.2)
C
for all z ∈ C. When d μ (z) = ϕ (z) dA(z), the measure μ satisfies condition (M) if
and only if the function ϕ satisfies condition (I1 ). See Sect. 3.2 for the definition of
condition (I p ).
If μ satisfies condition (M), then the Toeplitz operator Tμ above is well defined
on a dense subset of Fα2 . In fact, if
N
f (w) = ∑ ck K(w, ak )
k=1
is any finite linear combination of kernel functions in Fα2 , then it follows from
condition (M) and the Cauchy–Schwarz inequality that Tμ ( f ) is well defined. Recall
from Lemma 2.11 that the set of all finite linear combinations of kernel functions is
dense in Fα2 .
If μ satisfies condition (M), the Berezin transform of μ (see Sect. 3.4) is well
defined:
α α
|kz (w)|2 e−α |w| dμ (w) = e−α |z−w| dμ (w),
2 2
μ
(z) =
π C π C
where
α
K(z, z) = eα wz− 2 |z|
2
kz (w) = K(w, z)/
μ
(z) = Tμ kz , kz α , z ∈ C.
In the rest of this section, we focus on the case of trace-class Toeplitz operators.
We will obtain several trace formulas related to Toeplitz operators. These trace
formulas will then be used in the next two sections to study bounded and compact
Toeplitz operators.
The definition of the Berezin transform ϕ and the Toeplitz operator Tϕ requires
that the function ϕ satisfy condition (I1 ). But in the study of Toeplitz operators, we
often need to require that ϕ satisfy condition (I2 ), which is slightly stronger than
condition (I1 ).
If the Toeplitz operator Tϕ is bounded on Fα2 , we have
Tϕ f (z) = f (w)ϕ (w)K(z, w) dλα (w),
C
See Sect. 3.1 for the definition of KS (w, z) for any bounded linear operator S on Fα2 .
If we further assume that ϕ satisfies condition (I2 ), then it is also easy to check that
the uniqueness of KTϕ implies
then we have
α dA(z).
tr (Tϕ S) = ϕ (z)KS (z, z) dλα (z) = ϕ (z)S(z) (6.5)
C π C
= dλα (w) [ϕ (z)K(z, w) − H(z, w)]KS∗ (z, w) dλα (z)
C C
= dλα (w) ϕ (z)K(z, w)KS∗ (z, w) dλα (z)
C C
= dλα (w) ϕ (z)K(w, z)KS (w, z) dλα (z).
C C
Hypothesis (4) allows the application of Fubini’s theorem, which, along with the
reproducing property in Fα2 and parts (3) and (8) of Proposition 3.9, leads to the
desired trace formulas.
Taking S to be the identity operator, we obtain the following trace formula for
Toeplitz operators on the Fock space.
Corollary 6.3 Suppose ϕ satisfies condition (I2 ). If Tϕ is in the trace class and
ϕ ∈ L1 (C, dA), then
α
tr (Tϕ ) = ϕ (z)K(z, z) dλα (z) = ϕ (z) dA(z). (6.6)
C π C
Proof. When S is the identity operator, we have KS (z, w) = K(z, w), so the condition
|ϕ (z)||K(w, z)||KS (w, z)| dλα (w) dλα (z) < ∞
C C
becomes
α
|ϕ (z)|K(z, z) dλα (z) = |ϕ (z)| dA(z) < ∞.
C π C
Note that there exist symbol functions ϕ such that Tϕ is in the trace class but
ϕ ∈ L1 (C, dA). See Exercise 10.
Corollary 6.4 Suppose ϕ is bounded and compactly supported in C. Then for any
bounded linear operator S on Fα2 , the operator Tϕ S is trace class and
α dA(z).
tr (Tϕ S) = ϕ (z)KS (z, z) dλα (z) = ϕ (z)S(z) (6.7)
C π C
Proof. It is easy to see that hypotheses (1)–(3) of Theorem 6.2 are satisfied. To
check hypothesis (4) of Theorem 6.2, we write
I= |ϕ (z)| dλα (z) |KS (w, z)||K(w, z)| dλα (w).
C C
6.1 Trace Formulas 219
Thus,
α S
I ≤ S |ϕ (z)|K(z, z) dλα (z) = |ϕ | dA < ∞,
C π C
Proof. Let L denote the space of all bounded linear operators on Fα2 . Then, it is well
known that T ∗ = L and K∗ = T, with the duality pairing given by S, T = tr (ST ).
To prove (2), assume that C is not norm dense in K. By the Hahn–Banach
theorem, there must be a nonzero operator S in T such that
Tϕ , S = 0, Tϕ ∈ C.
By Corollary 6.4,
α dA(z)
0 = Tϕ , S = tr (Tϕ S) = ϕ (z)S(z)
π C
for all continuous functions ϕ with compact support in C. This implies that S = 0.
So S = 0, a contradiction which proves (2).
The proof for (1) is similar, and we omit the details here.
220 6 Toeplitz Operators
6.2 The Bargmann Transform 221
The connection between Toeplitz operators on the Fock space and pseudodifferential
operators on L2 (R, dx) is established by the Bargmann transform, and the most
elementary way to understand the Bargmann transform is via the classical Hermite
polynomials.
Recall that for any nonnegative integer n, the nth Hermite polynomial Hn (x) is
defined by
2 d
n
Hn (x) = (−1)n ex n e−x .
2
dx
The first five Hermite polynomials are given by
H0 (x) = 1,
H1 (x) = 2x,
H2 (x) = 4x2 − 2,
H3 (x) = 8x3 − 12x,
H4 (x) = 16x4 − 48x2 + 12.
In general, it is easy to check that each Hn has degree n and
Hn (x) = 2xHn−1 (x) − Hn−1 (x), n ≥ 1,
are linearly independent in the vector space of all polynomials of degree less than
or equal to N. A dimensionality argument then shows every polynomial of degree
less than or equal to N can be written as a linear combination of the polynomials in
(6.8). Therefore, the condition
f (x)hn (x) dx = 0, n ≥ 0,
R
implies that
f (x)P(x)e−α x dx = 0
2
for all polynomials P, which, according to Lemma 3.16, implies that f = 0 almost
everywhere. Thus, the set {hn } is complete in L2 (R, dx).
We now define the Bargmann transform. Let f be a function on R satisfying the
condition that f (x)e|tx|−π x is integrable with respect to dx for any real t. Then for
2
To this end, first observe that if u = x − z, where x is fixed, then d/du = −d/dz. It
follows that
dn −(x−z)2 n d
n
−u2 −u2
= e−x Hn (x).
2
n
e = (−1) n
e = e Hn (u)
dz z=0 du u=x u=x
1 ∞
2α 4 2 − α z2
π
e2α xz−α x 2 = ∑ ek (z)hk (x).
k=0
Multiply both sides by hn (x) and integrate over the real line. The desired result
Bα hn = en then follows from the fact that {hk } is orthonormal in L2 (R, dx).
Proposition 6.9. The inverse of the Bargmann transform is given by
1
2α 4 2 − α z2
[B−1
α f ](x) = f (z)e2α xz−α x 2 dλα (z), (6.10)
π C
where f ∈ Fα2 .
Proof. Fix any polynomial f ∈ Fα2 and any function g ∈ L2 (R, dx) that is compactly
supported. Since
Bα : L2 (R, dx) → Fα2 ⊂ L2α
is an isometry, we have
B−1 −1
α f , gL2 (R) = Bα Bα f , Bα gα = f , Bα gα
1
2α 4 2 − α z2
= f (z) dλα (z) g(x)e2α xz−α x 2 dx
π C R
1
2α 4 2 − α z2
= g(x) dx f (z)e2α xz−α x 2 dλα (z)
π R C
where
1
2α 4 2 − α z2
F(x) = f (z)e2α xz−α x 2 dλα (z).
π C
224 6 Toeplitz Operators
e2α i(−rD−sX) e−α x = e−2α isx+α irs−α (x−r) = e−2α isx+α irs−α x
2 2 2 +2α xr−α r2
.
is the first vector in the orthonormal basis {hn} of L2 (R, dx). By Theorem 6.8 and
its proof, Bα (h0 ) = e0 = 1, or equivalently,
2α α 2
e2α xz−2α x dx = e 2 z .
2
π R
Replacing z by −i 2π /α z and changing x to π /(2α )x, we obtain the desired
identity in (6.12).
6.2 The Bargmann Transform 225
The above lemma simply states that the Fourier transform of e−π x is e−π z . In
2 2
= σ (D, X)B−1 −1
α kz , Bα kz L2 (R)
α 2
α α
= σ p, q e2α i(pD+qX) B−1 −1
α kz , Bα kz L2 (R) dpdq.
π R R π π
Let z = r + is. By Proposition 6.10, Lemma 1.28, and the fact that each ρ (−r, −s)
is a unitary operator on L2 (R, dx), we have
= c2 e2α i(−ps+qr)ρ (−r, −s)ρ (p, q)e−α x , ρ (−r, −s)e−α x L2 (R)
2 2
where c2 = 2α /π .
226 6 Toeplitz Operators
Therefore,
α 2
α α 2α i(−ps+qr)− α (p2 +q2 )
T(z) = σ p, q e 2 dp dq
π R R π π
π2
σ (p, q)e2π i(−ps+qr)− 2α (p
2 +q2 )
= dp dq.
R R
We rewrite T(z) as
π2
e2π i(−ps+qr)− 2α (p σ (u, v)e−2π i(pu+qv) du dv.
2 +q2 )
dp dq
R R R R
π C
B−1
α ABα = X + iD = Z, B−1 ∗ ∗
α A Bα = X − iD = Z , (6.15)
where X, D, and Z are the (unbounded) operators on L2 (R, dx) defined in Sect. 1.4.
Proof. Let Cc (R) denote the space of continuous functions on R having compact
support. Then Cc (R) is dense in L2 (R, dx). Given f ∈ Cc (R), we differentiate
1
2α 4 2 − α z2
Bα f (z) = e2α xz−α x 2 f (x) dx
π R
This gives
ABα f = 2Bα X f − A∗ Bα f ,
and hence
B−1 −1 ∗
α ABα + Bα A Bα = 2X. (6.17)
ABα f = 2iBα D f + A∗ Bα f ,
and hence
B−1 −1 ∗
α ABα − Bα A Bα = 2iD. (6.18)
Bα σ (Z, Z ∗ )B−1
α = Tϕ , (6.19)
where Tϕ is the Toeplitz operator on Fα2 with symbol ϕ (z) = σ (z, z) = σ (z).
Proof. By Lemma 6.13, we have
α = ∑ cnm A A .
Bα σ (Z, Z ∗ )B−1 n ∗m
If f has the property that the function zm f (z) is also in Fα2 (all polynomials, which
are dense in Fα2 , clearly have this property), then we can write
zm f (z) = wm f (w)eα zw dλα (w).
C
Therefore,
Bα σ (Z, Z ∗ )B−1
α f (z) =
C
∑ cnm wn wm f (w)eα zw dλα (w)
= ϕ (w) f (w)eα zw dλα (w)
C
= Tϕ f (z).
6.3 Boundedness
In this section, we obtain necessary and sufficient conditions for the Toeplitz
operator Tϕ to be bounded on Fα2 . These conditions are based on the Berezin
transform or the heat transform at particular time points.
The main results of the section can be summarized as follows:
(a) If Tϕ is bounded on Fα2 , then Bβ ϕ is bounded for all β ∈ (0, 2α ).
(b) If Bβ ϕ is bounded for some β > 2α , then Tϕ is bounded on Fα2 .
(c) If ϕ ≥ 0, then Tϕ is bounded on Fα2 if and only if Bα ϕ is bounded if and only if
ϕr is bounded, where r is any fixed radius. Here,
1
ϕr (z) = ϕ (w) dA(w)
π r2 B(z,r)
For any nonnegative integer n, let Pn denote the rank-one projection from Fα2 onto
the one-dimensional subspace generated by en . Thus,
Pn f = f , en en , n ≥ 0, f ∈ Fα2 .
with the usual convention that T (0) = P0 . It is clear that the series above converges
in the norm topology of Fα2 . By property (7) of Proposition 3.9, we have
∞
KT (t) (z, w) = (1 − t) ∑ t n KPn (z, w)
n=0
∞
= (1 − t) ∑ t n en (z)en (w)
n=0
= (1 − t)eα tzw ,
Recall that for each a ∈ C, we have the Weyl unitary operator Wa on Fα2 , a
weighted translation operator, defined by
α
Wa f (z) = f (z − a)ka (z) = eα za− 2 |a| f (z − a).
2
We always have
(t)
tr (Ta ) = tr [T (t)Wa∗Wa ] = tr (T (t) ) = 1
Theorem 6.15. Suppose ϕ satisfies condition (I2 ) and Tϕ is bounded on Fα2 . Then
(t)
tr (Tϕ Ta ) = Bβ ϕ (a) (6.23)
√
for all −1 < t < 2 − 1, where β = α (1 − t).
Proof. We first prove the result for a = 0. The problem is reduced to checking
hypothesis (4) of Theorem 6.2. In fact, it would then follow from (6.5) that
tr Tϕ T (t) = ϕ (z)KT (t) (z, z) dλα (z)
C
ϕ (z)eα t|z| dλα (z)
2
= (1 − t)
C
α (1 − t)
ϕ (z)e−α (1−t)|z| dA(z)
2
=
π C
= Bβ ϕ (0).
Thus, we need to estimate the integral
I(t) = |ϕ (z)| dλα (z) |K(z, w)||KT (t) (z, w)| dλα (w)
C C
= (1 − t) |ϕ (z)| dλα (z) |eα (1+t)zw | dλα (w)
C C
α (1+t)2
|z|2
= (1 − t) |ϕ (z)|e 4 dλα (z)
C
α (1 − t) (1+t)2
|ϕ (z)|eα [ 4 −1]|z|
2
= dA(z)
π C
α (1 − t) α
|ϕ (z)e− 2 |z| |e−δ (t)|z| dA(z),
2 2
=
π C
where
(1 + t)2 α α
δ (t) = α 1 − − = (1 − 2t − t 2 ).
4 2 4
1 1 1
= + .
β γ α
By Theorem 3.13, the semigroup property of the heat transform Ht , we have H1/β =
H1/γ H1/α . In terms of the parametrized Berezin transforms, we have Bβ = Bγ Bα . By
what was proved in the previous paragraph, or directly from Bα ϕ (a) = Tϕ ka , ka α ,
the boundedness of Tϕ on Fα2 implies Bα ϕ ∞ ≤ Tϕ . Since Bγ is a contraction on
L∞ , we have
Bβ ϕ ∞ = Bγ Bα ϕ ∞ ≤ Bα ϕ ∞ ≤ Tϕ .
It follows that the operators T and Tg have the same Berezin symbol. Since the
mapping S → S is one-to-one, we conclude that T = Tg .
Theorem 6.18. Let g be a symbol function on C that satisfies condition (I2 ). If there
exists some β ∈ (2α , ∞) such that Bβ g ∈ L∞ (C), then Tg is bounded on Fα2 .
Proof. Let σ (z) = B2α g(z). In view of Theorem 6.17, the Toeplitz operator Tg on
Fα2 is unitarily equivalent to the pseudodifferential operator σ (D, X) on L2 (R, dx).
We proceed to show that the pseudodifferential operator σ (D, X) is bounded.
Let γ be the positive number satisfying
1 1 1
= + .
2α β γ
Differentiating under the integral sign, we see that for any nonnegative integers n
and m, we have
∂ n+m σ
hmn (z − w, z − w)ϕ (w)e−γ |z−w| dA(w),
2
(z) =
∂ zn ∂ zm C
Proof. The equivalences of (a), (b), and (d) follow from the characterization of
Fock–Carleson measures in Sect. 3.4. In fact, when ϕ is nonnegative, we have
Tϕ f , f α = | f |2 ϕ dλα .
C
The densely defined positive operator Tϕ is bounded if and only if there exists a
constant C > 0 such that
This condition simply means that the measure dμ (z) = ϕ (z) dA(z) is Fock–Carleson.
The equivalence of (b) and (c) follows from Theorem 3.23.
As a consequence of the above theorem, we obtain the following characterization
of bounded Toeplitz operators on Fα2 induced by symbols from BMO1 .
Theorem 6.20. Suppose ϕ ∈ BMO1 . Then the following conditions are equiva-
lent:
(a) Tϕ is bounded on Fα2 .
(b) ϕ = Bα ϕ ∈ L∞ (C).
(c) Bβ ϕ ∈ L∞ (C), where β is any fixed positive weight parameter.
(d) ϕr ∈ L∞ (C), where r is any fixed positive radius.
Proof. By (3.22) of Theorem 3.34, there exists a constant C > 0 such that
ϕ ◦ ϕz − ϕ
(z)L1 (dλα ) ≤ C
6.3 Boundedness 235
|*
ϕ | − |ϕ| ∈ L∞ (C).
Therefore, ϕ ∈ L∞ (C) if and only if |* ϕ | ∈ L∞ (C). It follows from this and the
characterization of bounded Toeplitz operators with nonnegative symbols (see
Theorem 6.19) that the condition ϕ ∈ L∞ (C) implies that T|ϕ | is bounded on Fα2 .
Let ϕ = f + ig, where f and g are the real and imaginary parts of ϕ , respectively.
Since | f | ≤ |ϕ | and |g| ≤ |ϕ |, and nonnegative symbols induce positive operators,
we see that the boundedness of T|ϕ | implies that both T| f | and T|g| are bounded on
Fα2 .
Since f is real-valued, we can write f = f + − f − , where
6.4 Compactness
In this section, we discuss the compactness of Toeplitz operators on Fα2 . The main
results are parallel to those in the previous section. All conditions in this section are
in terms of membership in the space C0 (C) which consists of continuous functions
f on C such that f (z) → 0 as z → ∞. In several approximation arguments, we will
also need the space Cc (C), consisting of continuous functions f on C with compact
support. It is clear that Cc (C) is dense in C0 (C) in the supremum norm of C0 (C).
Theorem 6.21. Suppose ϕ satisfies condition (I2 ) and Tϕ is compact on Fα2 . Then
Bβ ϕ ∈ C0 (C) for all β ∈ (0, 2α ).
Proof. Recall from Theorem 6.16 and its proof that, for any β ∈ (0, 2α ), there exists
a positive constant C = C(β ) such that Bβ f ∞ ≤ CT f whenever T f is bounded
on Fα2 .
If Tϕ is compact on Fα2 , then by Theorem 6.5, there exists a sequence { fn } of
functions in Cc (C) such that
1
Tϕ − T fn ≤ , n ≥ 1.
n
Therefore,
1
Bβ ϕ − Bβ fn ∞ ≤ CTϕ − T fn <
n
for all n ≥ 1. Each fn has compact support, so Bβ fn ∈ C0 (C). Since C0 (C) is closed
in the supremum norm, we conclude that Bβ ϕ is in C0 (C) as well.
Theorem 6.22. Suppose g is a symbol function that satisfies condition (I2 ). If there
exists some β ∈ (2α , ∞) such that Bβ g ∈ C0 (C), then Tg is compact on Fα2 .
Proof. As in the proof of Theorem 6.18, the Toeplitz operator Tg on Fα2 is unitarily
equivalent to the pseudodifferential operator σ (D, X) on L2 (R, dx), where σ (z) =
B2α g(z). Furthermore, it follows from Theorem 6.18 that Tg and σ (D, X) are both
bounded operators with
γ
ϕ (w)e−γ |z−w| dA(w),
2
σ (z) = Bγ ϕ (z) =
π C
where ϕ (z) = Bβ g(z). For any pair of nonnegative integers m and n, there is a
polynomial hmn (z, z) such that
∂ m+n σ
hmn (z − w, z − w)ϕ (w)e−γ |z−w| dA(w).
2
(z) = (6.25)
∂ zm ∂ zn C
and so
|hmn (w, w)|e−γ |w| dA(w).
2
|T f (z)| ≤ f ∞
|w−z|≤R
clearly implies that T f (z) → 0 as z → ∞. Thus, T maps Cc (C) into C0 (C). Since
Cc (C) is dense in C0 (C) in the norm topology of L∞ (C), we infer from the
boundedness of T : L∞ (C) → L∞ (C) that T maps C0 (C) into C0 (C). This, along with
(6.25), shows that ∂ m+n σ /∂ zm ∂ zm is in C0 (C) for any pair of nonnegative integers
m and n. By Theorem 1.25, the pseudodifferential operator σ (D, X) is compact on
L2 (R, dx), and hence the Toeplitz operator Tg is compact on Fα2 .
Theorem 6.23. Suppose ϕ is nonnegative and satisfies condition (I1 ). Then, the
following conditions are equivalent:
(a) Tϕ is compact on Fα2 .
(b) ϕ ∈ C0 (C).
(c) Bβ ϕ ∈ C0 (C), where β is any fixed positive weight parameter.
(d) ϕr ∈ C0 (C), where r is any fixed positive radius.
Proof. The equivalence of (a), (b), and (d) follow from the characterization of
vanishing Fock–Carleson measures in Sect. 3.4. See the proof of Theorem 6.19 for
the connection to Fock–Carleson measures. The equivalence of (b) and (c) follows
from Theorem 3.23.
The rest of this section is devoted to the compactness of Toeplitz operators with
symbols in BMO1 .
Lemma 6.24. Suppose f ∈ BMO1 and f = Bα f is bounded. Then
T f Kz = Kz [P( f ◦ ϕz )] ◦ ϕz (6.26)
for all z ∈ C, where P : L2α → Fα2 is the orthogonal projection, T f is the Toeplitz
operator on Fα2 , Kz is the reproducing kernel of Fα2 , and ϕz (w) = z − w.
6.4 Compactness 239
Proof. Since BMO1 and the Berezin transform are both translation invariant, we see
that for any z ∈ C, we have
f ◦ ϕz ∈ BMO1 , Bα ( f ◦ ϕz ) ∈ L∞ (C).
for all w ∈ C.
Proof. Recall from the proof of Theorem 6.20 that if f ∈ BMO1 and f is bounded,
then |*
f | is bounded as well. By translation invariance of BMO1 and the Berezin
transform, there exists a positive constant C such that
Bα (| f ◦ ϕz |)(w) ≤ C, z, w ∈ C.
≤ | f ◦ ϕz (u)||eα wu | dλα (u)
C
≤C |eα wu | dλα (u)
C
α
= Ce 4 |w| .
2
and the set of all finite linear combinations of kernel functions is dense in Fα2 , we
see that (a) and (b) are actually equivalent.
To prove part (b), it suffices to show that
for every nonnegative integer n because the set of polynomials is dense in Fα2 .
Fix a nonnegative integer n and a point a ∈ C. Observe that
f(ϕz (a)) = f
◦ ϕz (a) = e−α |a| T f ◦ϕz Ka , Ka ,
2
where
∞
αk k k
Ka (u) = eα ua = ∑ ua.
k=1 k!
It follows that
∞
α k+ j + ,
f(ϕz (a)) = e−α |a| ∑
2
T f ◦ϕz uk , u j ak a j .
k, j=0 k! j!
can be written as
∞
α k+ j + ,
Ir (z) = ∑ k! j!
T f ◦ϕz uk , u j
|v|<r
vk+n v j dA(v)
k, j=0
∞
α 2k+n
= ∑ k!(k + n)! Tf ◦ϕz uk , uk+n |v|<r
|v|2(k+n) dA(v)
k=0
∞
α 2k+n
=π ∑ k!(k + n + 1)! Tf ◦ϕz uk , uk+n r2(k+n+1)
k=0
αn - .
= πr 2(n+1)
T f ◦ϕz 1, u + Σr,n (z) ,
n
(n + 1)!
where
∞
α 2k+n
Σr,n (z) = ∑ k!(k + n + 1)! Tf ◦ϕz uk , un+k r2k .
k=1
for any fixed r > 0. Since T f ◦ϕz ≤ C for all z ∈ C, where C is independent of z,
we see that
∞
α 2k+n uk uk+n 2k
|Σr,n (z)| ≤ C ∑ r
k=1 k!(k + n + 1)!
∞
α 2k+n k!(k + n)! 2k
=C∑ r
k=1 k!(k + n + 1)! α 2k+n
∞
(α r2 )k
∑
n
≤ Cα 2
k=1 k!
= C α 2 eα r − 1
n 2
for all r > 0, n ≥ 0, and z ∈ C. Given any ε > 0, choose a small enough positive
radius r such that
2
C α 2 eα r − 1 < ε .
n
242 6 Toeplitz Operators
where χR is the characteristic function of the open ball |z| < R and MχR is the
operator of multiplication on L2α by χR . It follows from the boundedness of T f and a
simple normal family argument that each T fR is compact. Thus, the compactness of
T f will follow if we can show that
By Schur’s test (Lemma 2.14), whenever there exists a positive function h on C such
that
|K Rf (z, u)|h(z) dλα (z) ≤ C1 h(u), u ∈ C, (6.31)
C
6.4 Compactness 243
and
|K Rf (z, u)|h(u) dλα (u) ≤ C2 h(z), z ∈ C, (6.32)
C
we then have
T f − T fR 2F 2 →L2 ≤ C1C2 . (6.33)
α α
We will arrive at constants C1 and C2 such that the product C1C2 tends to 0 as R → ∞,
which then implies the compactness of T f .
α 2
Let h(z) = K(z, z) = e 2 |z| and consider the integrals
I(z) = |K Rf (z, u)|h(u) dλα (u), z ∈ C,
C
from (6.32). It is clear that I(z) = 0 for |z| < R. For |z| ≥ R, we have
I(z) = |T f Kz (u)| K(u, u) dλα (u),
C
Making the change of variables u → ϕz (u) and simplifying the result, we get
α
α
|P( f ◦ ϕz )(u)| eα (z−u)z e− 2 |z−u| dA(u).
2
I(z) =
π C
Fix p ∈ (1, ∞) and σ ∈ (α /4, α /2). Let 1/p + 1/q = 1. By Hölder’s inequality,
α α
|P( f ◦ ϕz (u))|e−σ |u| eσ |u| |eα (z−u)z |e− 2 |z−u| dA(u)
2 2 2
I(z) =
π C
1
α p −pσ |u|2
p
≤ |P( f ◦ ϕz (u))| e dA(u)
π C
1
qσ |u|2 qα (z−u)z − q2α |z−u|2
q
× e |e |e dA(u) .
C
which is equal to
qα 2 2 qα qα 2 α
2 |z| +qσ |u| − 2 |u| 2 |z| e−q( 2 −σ )|u| dA(u).
2 2
e dA(u) = e
C C
244 6 Toeplitz Operators
On the other hand, it follows from Lemma 6.25 that for all z ∈ C, we have
α
|P( f ◦ ϕz )(u)| p e−pσ |u| ≤ Ce−p(σ − 4 )|u| ,
2 2
with the function on the right-hand side above integrable with respect to dA. This,
along with Lemma 6.26 and the dominated convergence theorem, shows that the
constants
1p
p −pσ |u|2
C1,R = sup |P( f ◦ ϕz (u))| e dA(u)
|z|≥R C
from (6.31) are slightly easier to estimate. In fact, by Lemma 6.24 and a change of
variables,
J(u) ≤ |T f Kz (u)| K(z, z) dλα (z)
C
α α 2
= |Kz (u)||P( f ◦ ϕz )(ϕz (u))|e− 2 |z| dA(z)
π C
α α
|Kz+u (u)||P( f ◦ ϕz+u)(z)|e− 2 |z+u| dA(z)
2
=
π C
α α |u|2 α 2
= e2 |P( f ◦ ϕz+u )(z)|e− 2 |z| dA(z).
π C
This proves the desired estimate in (6.31) and completes the proof of the theorem.
Corollary 6.28 Let f ∈ BMO1 , α > 0, and β > 0. Then Bα f ∈ C0 (C) if and only
if Bβ f ∈ C0 (C).
Proof. Without loss of generality, assume that 0 < α < β . If Bβ f ∈ C0 (C), then
by Proposition 3.21, Bα f ∈ C0 (C). We do not need the assumption that f ∈ BMO1
here.
If Bα f ∈ C0 (C), then by Theorem 6.27, T f is compact on Fα2 , which, according
to Theorem 6.21, implies that Bγ f ∈ C0 (C) for all 0 < γ < 2α . Repeat this process
a certain number of times, we will then get Bβ f ∈ C0 (C).
6.5 Toeplitz Operators in Schatten Classes 245
For μ ≥ 0, we are going to determine when the Toeplitz operator Tμ on Fα2 belongs
to the Schatten class S p . The case when p ≥ 1 is relatively easy and will be taken up
first.
Recall that for any bounded linear operator T on Fα2 we define the Berezin
transform T by
T(z) = T kz , kz , z ∈ C,
where kz are the normalized reproducing kernels in Fα2 . If T is positive on Fα2 , then
α
tr (T ) = T(z) dA(z).
π C
See Proposition 3.3. In particular, T is in the trace-class S1 if and only if the integral
above converges. As a consequence, we obtain the following trace formula for
Toeplitz operators on the Fock space.
Proposition 6.29. Suppose μ is a positive Borel measure on C and satisfies
condition (M). Then Tμ is in the trace-class S1 if and only if μ is finite on C.
Moreover, tr (Tμ ) = (α /π )μ (C).
Proof. Since all integrands below are nonnegative, we use Fubini’s theorem to
obtain
α
tr (Tμ ) = μ
(z) dA(z)
π C
α
eα |z| dλα (z) |eα z̄w |2 e−α (|z| +|w| ) dμ (w)
2 2 2
=
π C C
α
e−α |w| dμ (w) |eα z̄w |2 dλα (z)
2
=
π C C
α α
= dμ (w) = μ (C).
π C π
This also shows that tr (Tμ ) < ∞ if and only if μ (C) < ∞.
Lemma 6.30. If p ≥ 1 and ϕ ∈ L p (C, dA), then Tϕ ∈ S p .
Proof. If ϕ ∈ L p (C, dA), then ϕ ◦ ta ∈ L p (C, dA) by a simple change of variables.
It follows that ϕ ◦ ta ∈ L p (C, dλα ) for every a ∈ C. Thus, ϕ satisfies condition (I p).
Since p ≥ 1, ϕ also satisfies condition (I1 ) so that Tϕ is densely defined on Fα2 .
The rest is proved in exactly the same way that Proposition 7.11 in [250] was
proved.
Lemma 6.31. Suppose r > 0, μ is a positive Borel measure on C, and
μ (B(z, r))
μ
r (z) = , z ∈ C.
π r2
246 6 Toeplitz Operators
If μ
r is in L p (C, dA) for some 0 < p < ∞, then μ satisfies condition (M), and the
Toeplitz operators Tμ and Tμr are both bounded on Fα2 . Moreover, there exists a
positive constant C (independent of μ ) such that Tμ ≤ CTμr .
Proof. Let
C= μ (B(z, r)) p dA(z) < ∞.
C
When z ∈ B(a, r/2), we have B(a, r/2) ⊂ B(z, r) by the triangle inequality. It follows
that μ (B(z, r)) ≥ μ (B(a, r/2)), and so
π r2
μ (B(a, r/2)) p ≤ C, a ∈ C.
4
This shows that the function a → μ (B(a, r/2)) is bounded. By Theorem 3.29 (with
p = 2 there), the measure μ satisfies condition (M), and the Toeplitz operator Tμ is
bounded on Fα2 , which in turn implies that the function z → μ (B(z, r)) is bounded.
Thus, Tμr is bounded on Fα2 as well.
Given f ∈ Fα2 , we use Fubini’s theorem to obtain
π r2 Tμr f , f = π r2 | f (z)|2 μ
r (z) dλα (z)
C
= | f (z)|2 μ (B(z, r)) dλα (z)
C
= | f (z)|2 dλα (z) χB(z,r) (w) dμ (w)
C C
= dμ (w) | f (z)|2 χB(w,r) (z) dλα (z)
C C
α
| f (z)e−α |z|
2 /2
= dμ (w) |2 dA(z).
π C B(w,r)
Combining the above identity with Lemma 2.32, we obtain a positive constant C
such that
| f (w)|2 e−α |w| dμ (w) = Tμ f , f .
2
CTμr f , f ≥
C
Refining the arguments in the above proof then shows that μ (B(a, r/2)) → 0 as
a → ∞, which implies that Tμ is compact on Fα2 and μ r ∈ C0 (C).
For the remainder of this section, we let {an } denote any fixed arrangement of
the square lattice rZ2 into a sequence. We are now ready to characterize positive
Toeplitz operators in S p when p ≥ 1.
Theorem 6.32. Suppose μ ≥ 0, r > 0, and p ≥ 1. If μ satisfies condition (M), then
the following conditions are equivalent:
(a) The operator Tμ is in the Schatten class S p .
(b) The function μ
(z) is in L p (C, dA).
(c) The function μ (B(z, r)) is in L p (C, dA).
(d) The sequence { μ (B(an, r))} is in l p .
Proof. That (a) implies (b) follows from Proposition 3.5. The elementary inequality
μ
r (z) ≤ C μ (z) (see the proof of Theorem 3.29) shows that condition (b) implies (c).
If the averaging function μ r (z), which differs from μ (B(z, r) by a constant, is in
L p (C, dA), then it follows from Lemma 6.30 that Tμr is in S p . Combining this with
Lemma 6.31, we conclude that Tμ is in S p . This proves that (c) implies (a). Hence,
conditions (a), (b), and (c) are equivalent.
To prove that condition (d) is equivalent to the other conditions, we first assume
that condition (b) holds, which implies that the function μ (B(z, 2r)) is in L p (C, dA).
Choose a positive integer m such that each point in the complex plane belongs to at
most m of the disks B(an , r). Then
∞
m
C
μ (B(z, 2r)) p dA(z) ≥ ∑ μ (B(z, 2r)) p dA(z).
n=1 B(an ,r)
For each z ∈ B(an , r), we deduce from the triangle inequality that
Therefore, ∞
m
C
μ (B(z, 2r)) p dA(z) ≥ π r2 ∑ μ (B(an , r)) p .
n=1
where {zn } is any arrangement of the lattice (r/2)Z2 . In fact, for each point zk that
is not in the lattice {an }, the disk B(zk , r) is covered by six adjacent disks B(ak , r).
Therefore,
∞
C
μ (B(z, r/2)) p dA(z) ≤ ∑ μ (B(z, r/2)) p dA(z)
n=1 B(zn ,r/2)
∞
≤ ∑ μ (B(zn , r)) p dA(z)
n=1 B(zn ,r/2)
π r2 ∞
= ∑ μ (B(zn , r)) p < ∞.
4 n=1
This shows that condition (d) implies (c), as the equivalence of (c) to (b) implies that
if condition (c) holds for one positive radius, then it will hold for any other positive
radius. This completes the proof of the theorem.
Specializing to the case when
α
dμ (z) = ϕ (z) dA(z),
π
we obtain the following corollary concerning Toeplitz operators induced by non-
negative functions.
Corollary 6.33 Suppose ϕ ≥ 0, p ≥ 1, and r > 0. If ϕ satisfies condition (I1 ), then
the following conditions are equivalent:
(a) The Toeplitz operator Tϕ belongs to S p .
(b) The Berezin transform ϕ belongs to L p (C, dA).
(c) The averaging function
1
ϕr (z) = ϕ (w) dA(w)
π r2 B(z,r)
We now turn our attention to the case 0 < p ≤ 1, which requires new ideas and
techniques.
Lemma 6.34. Suppose μ ≥ 0, r > 0, and 0 < p ≤ 1. If μ satisfies condition (M),
then the following conditions are equivalent:
(a) The function μ
(z) is in L p (C, dA).
(b) The function μ (B(z, r)) is in L p (C, dA).
(c) The sequence { μ (B(an, r))} is in l p .
6.5 Toeplitz Operators in Schatten Classes 249
It follows that
∞
α
∑ e−α |z−an| +2α r|z−an| μ (B(an, r)).
2
μ
(z) ≤
π n=1
which is easily seen to be convergent. Thus, the condition { μ (B(an , r))} ∈ l p implies
μ
∈ L p (C, dA).
On the other hand, there exists a positive integer m such that every point in the
complex plane belongs to at most m of the disks B(an , r). Thus,
∞
m
C
μ
(z) p dA(z) ≥ ∑ μ
(z) p dA(z).
n=1 B(an ,r)
π
250 6 Toeplitz Operators
It follows that
∞
(z) p dA(z) ≥ α r2 e−4pα r ∑ μ (B(an, r)) p .
2
m μ
C n=1
Thus, μ ∈ L p (C, dA) implies that { μ (B(an , r))} ∈ l p , which proves the equivalence
of conditions (a) and (c).
That condition (a) implies (b) follows from the inequality μ (B(z, r)) ≤ C μ (z)
observed in the proof of Theorem 3.29.
To prove that condition (b) implies (c), we assume that the function μ (B(z, r))
is in L p (C, dA). Consider the lattice (r/2)Z2 and arrange it into a sequence {zn }.
There exists a positive integer m such that every point in the complex plane belongs
to at most m of the disks B(zn , r/2). Therefore,
∞
m
C
μ (B(z, r)) p dA(z) ≥ ∑ μ (B(z, r)) p dA(z).
n=1 B(zn ,r/2)
Thus,
π r2 ∞
m
C
μ (B(z, r)) p dA(z) ≥ ∑ μ (B(zn , r/2)) p.
4 n=1
By the equivalence of conditions (a) and (c), the function μ belongs to
L p (C, dA). Applying the equivalence of (a) and (c) once more, we conclude that
{ μ (B(an , r))} ∈ l p . This completes the proof of the lemma.
Lemma 6.35. Suppose μ ≥ 0, 0 < p ≤ 1, and μ satisfies condition (M). If the
function μ
belongs to L p (C, dA), then the operator Tμ belongs to S p .
Proof. Since μ
belongs to L p (C, dA) and μ dominates μ r , Lemma 6.31 shows that
*
Tμ is bounded. Thus, Tμ = μ
and the desired result follows from Proposition 3.6.
We will need the following lemma, which can be found as Proposition 1.29 in
[250].
Lemma 6.36. If 0 < p ≤ 2, then for any orthonormal basis {en } of a separable
Hilbert space H and any compact operator T on H, we have
∞ ∞
p
T S p ≤ ∑ ∑ |Ten , ek | p .
n=1 k=1
Proof. The equivalence of (b), (c), and (d) was proved in Lemma 6.34. That
condition (b) implies condition (a) was proved in Lemma 6.35. Therefore, to finish
the proof, we will show that condition (a) implies (d).
To this end, fix some large R with R > 2r and use Lemma 1.14 to partition {an }
into N sublattices such that the Euclidean distance between any two points in each
sublattice is at least R. Let {ζn } be such a sublattice and let
∞
ν= ∑ μ χn ,
n=1
so (6.36) gives us
α p p
α |z−ζn |2 α |z−ζk |2
ESpp ≤
π ∑ C
e− 2 e− 2 dν (z) . (6.37)
n=k
The last step above follows from the mean value theorem with
and so
α p pα
∞ ∞ pα pα
e− 16 R ∑ ∑ μ (B(ζ j , r)) p e− 4 |z∗ −ζn | − 4 |z∗ −ζk |
2 2 2
ESpp ≤
π n,k=1 j=1
α p pα
∞ ∞ pα 2 pα
e− 16 R ∑ μ (B(ζ j , r)) p ∑ e− 4 |z∗ −ζn | − 4 |z∗ −ζk |
2 2
= .
π j=1 n,k=1
The last series above is clearly convergent. So we can find a positive constant C2 ,
independent of R, such that
pα
∞
ESpp ≤ C2 e− 16 R ∑ μ (B(ζ j , r)) p .
2
j=1
where C3 is another positive constant. Since this holds for each of the N subse-
quences of {an }, we obtain
∞
∑ μ (B(an, r)) p ≤ C3 NTμ Spp (6.39)
n=1
Finally, an easy approximation argument shows that (6.39) holds for all positive
Borel measures μ with Tμ ∈ S p . This proves that condition (a) implies (d), and thus
completes the proof of Theorem 6.37.
Again, specializing to the case when
α
dμ (z) = ϕ (z) dA(z),
π
we obtain the following corollary concerning Toeplitz operators induced by non-
negative functions:
Corollary 6.38 Suppose ϕ ≥ 0, 0 < p ≤ 1, r > 0, and ϕ satisfies condition (I1 ).
Then the following conditions are equivalent:
(a) The Toeplitz operator Tϕ belongs to S p .
(b) The Berezin transform ϕ belongs to L p (C, dA).
(c) The averaging function
1
ϕr (z) = ϕ (w) dA(w)
π r2 B(z,r)
In this section, we consider the following problem: when does a Toeplitz operator
Tμ have finite rank on the Fock space Fα2 ? It turns out the problem is pretty tricky.
If μ has compact support in C, we will be able to determine exactly when Tμ
has finite rank. But on the other hand, we will also construct a radial function ϕ ,
not identically zero, such that Tϕ = 0 in a natural way on the Fock space. This is
something unique for the Fock space setting. In particular, in the Fock space setting,
the Berezin transform ϕ → ϕ is not one-to-one if no additional assumptions are
made about ϕ .
Let n be a positive integer and denote by P(Cn ) the algebra of all holomorphic
polynomials on Cn . For any tuple k = (k1 , · · · , kn ) of nonnegative integers, we write
and
1
n! ∑
fa (z) = sgn(σ ) f (σ (z)), z ∈ U,
σ
where the sums are taken over all permutations on {1, · · · , n}.
Let Ps (Cn ) denote the subspace of P(Cn ) consisting of all symmetric poly-
nomials. Similarly, let Pa (Cn ) denote the subspace of P(Cn ) consisting of all
antisymmetric polynomials.
Let P∗ (Cn ) denote the vector space of all conjugate linear functionals on P(Cn ).
If μ is a finite complex Borel measure with compact support in C, then the Toeplitz
256 6 Toeplitz Operators
operator Tμ is well defined on the dense set P(C) in Fα2 . Furthermore, for any f ∈
P(C), we have Tμ ( f ) ∈ P∗ (C) in the sense that
α
f (z)g(z)e−α |z| dμ (z),
2
Tμ ( f )(g) = g ∈ P(C).
π C
c1 Tμ ( f1 ) + · · · + cn Tμ ( fn ) = 0. (6.41)
Since the ck ’s are not all 0, we see that the determinant of the matrix above must
be 0.
Lemma 6.40. Suppose μ is a finite complex Borel measure on C with compact
support. If Tμ has rank less than n and
Inserting monomials fk into (6.43) and then taking finite linear combinations, we
see that (6.43) remains valid if the product f1 (z1 ) · · · fn (zn ) is replaced by any
polynomial f ∈ P(Cn ). In other words,
f (z)Δ (g1 , · · · , gn )(z) d μn (z) = 0 (6.44)
Cn
1
n! ∑
[g1 (z1 ) · · · gn (zn )]a = (sgnσ )g1 (zσ (1) ) · · · gn (zσ (n) )
σ
1
= Δ (g1 , · · · , gn )(z)
n!
either have different zeros or they have the same zeros with different multiplicities.
It follows that at least one Taylor coefficient of p differs from the corresponding
coefficient of q. Thus, there exists an elementary symmetric polynomial whose
values at z and w are different.
We can now apply the Stone–Weierstrass approximation theorem to conclude
that every function in Cs (K) can be uniformly approximated by a sequence of
functions in Φs .
The main result of this section is the following:
Theorem 6.42. Suppose μ is a compactly supported finite complex Borel measure
on C such that the rank of Tμ is less than n, where n is a positive integer. Then μ is
supported on less than n points in C.
Proof. Recall that for z = (z1 , · · · , zn ),
⎛ ⎞
1 1 ··· 1
⎜ z1 z2 · · · zn ⎟
⎜ ⎟
V (z) = det ⎜ .. .. .. .. ⎟ = ∏(zi − z j )
⎝ . . . . ⎠ i> j
zn−1
1 z2 · · · zn−1
n−1
n
for all F ∈ C(E n ), where Fs is the symmetrization of F. Thus, (6.45) holds for all
F ∈ C(E n ). Consequently, |Vp|2 dμn is the zero measure so that the support of μn is
contained in the zero variety of Vp . Since a = (a1 , · · · , an ) is contained in the support
of μn , we must have Vp(a1 , · · · , an ) = 0, which is a contradiction. This shows that μ
is supported on less than n distinct points in C.
Corollary 6.43 Let ϕ be a compactly supported and locally integrable function on
C. Then the Toeplitz operator Tϕ on Fα2 has finite rank if and only if ϕ = 0.
In the rest of this section, we present an example to show that it is necessary to
assume that the measure μ in Theorem 6.42 and ϕ in Corollary 6.43 are compactly
supported. These results will be false without this assumption. To better understand
the intricacy of the problem, we note that if ϕ is bounded, then it follows easily from
the integral representation of the projection Pα and Fubini’s theorem that
Tϕ f , g = ϕ (z) f (z)g(z) dλα (z)
C
for all polynomials f and g. A limit argument then shows that the above also holds
for all functions f and g in Fα2 .
Proposition 6.44. There exists a radial function ϕ , not identically zero, such that
Tϕ = 0 on Fα2 in the sense that
ϕ (z) f (z)g(z) dλα (z) = 0
C
π πρ πρ πρ π
− <− ≤− + ρθ ≤ < .
2 2 2 2 2
πρ
πρ
0 < cos < cos − + ρθ ≤ 1, (6.46)
2 2
and
πi πi
f (z) = exp |z|−ρ e− 2 (2−ρ )−ρθ i + |z|ρ e 2 (2−ρ )+ρθ i
πρ
= exp −(|z|−ρ + |z|ρ ) cos − + ρθ
2
πρ
+ i(|z|−ρ − |z|ρ ) sin − + ρθ .
2
In particular,
πρ
| f (z)| = exp −(|z|−ρ + |z|ρ ) cos − + ρθ
2
for z = |z|eiθ with θ ∈ [0, π ] and |z| > 0. This together with (6.46) shows that
where k is any nonnegative integer and z is restricted to the closed upper half-plane.
6.6 Finite Rank Toeplitz Operators 261
By the formula for | f (z)|, the restriction of f to the real line belongs to L1 (R, dx).
In particular, the Fourier transform of f is well defined. Let g be the function eα x
2
Since f is analytic in the upper half-plane and continuous on the closed upper half-
plane, it follows from (6.47) and contour integration around the semicircle |z| = R,
Im (z) ≥ 0, that g(x) = 0 whenever x ∈ (−∞, 0]. So the function g is supported on
(0, ∞).
By the Fourier inversion formula, we can write
∞ ∞
−α t 2 +2π itx
g(t)e−α t
2 +2π itx
f (x) = g(t)e dt = dt
−∞ 0
6.7 Notes
The systematic study of Toeplitz operators on the Fock space started in [28, 29],
where several important techniques were introduced that remain useful up to today.
For example, the use of the Berezin transform in function theoretic operator theory
began in [28].
The material in Sect. 6.1 is mostly from [30]. The Bargmann transform between
the Fock space Fα2 and L2 (R, dx) has been a well-known and very useful tool in
analysis. Our presentation in Sect. 6.2 follows Folland’s book [92] closely. Theo-
rems 6.12 and 6.14 are well known in the theory of pseudodifferential operators.
The idea of using the operators T (t) to study trace-class properties of Toeplitz
operators first appeared in [30]. Theorems 6.15–6.18, as well as their compactness
counterparts in Sect. 6.4, are all from [30]. The characterization of bounded and
compact Toeplitz operators with nonnegative symbols is very similar to the Bergman
space setting, and details are worked out in [132].
For Toeplitz operators with bounded symbols, the characterization of compact-
ness in terms of the Berezin transform is also analogous to the Bergman space
setting, which was first obtained by Axler and Zheng in [6] and later generalized
to BMO symbols by Zorborska in [259]. Our presentation here follows [15, 61]
closely.
When 1 ≤ p < ∞, the characterization of Toeplitz operators in the Schatten class
S p of the Fock space Fα2 is relatively easy and follows the Bergman space theory
very closely. However, if 0 < p < 1, there is a critical difference between the Fock
and Bergman space theories. More specifically, in the Bergman space theory, there
is a cutoff point when Schatten class Toeplitz operators are characterized using the
Berezin transform, while the cutoff disappears in the Fock space setting. The proof
of Theorem 6.37 here is simpler than the one first constructed in [132].
Theorem 6.42, the characterization of finite-rank Toeplitz operators induced by
compactly supported measures, is due to Luecking [153]. The proof in [153] is
purely algebraic and works in several different contexts, including Toeplitz opera-
tors on the Bergman space of various domains. The example in Proposition 6.44
was constructed in [105]. Note that Proposition 6.44 does not contradict with
Proposition 3.17 because the function in Proposition 6.44 is far worse than the
functions permitted in Proposition 3.17.
264 6 Toeplitz Operators
6.8 Exercises 265
6.8 Exercises
Show that the Toeplitz operator Tϕ is unitary on the Fock space F12 (α = 1)
and the Berezin transform ϕ vanishes at ∞ and belongs to L p (C, dA) for all
0 < p < ∞.
5. Recall that for any z ∈ C, we have the self-adjoint unitary operator Uz defined
by Uz f (w) = f (z − w)kz (w). Show that if Tϕ is bounded, then
Uz Tϕ Uz dλα (z) = Tψ ,
C
where ψ (w) = ϕ(−w) and the integral converges in the strong operator
topology.
6. If Tϕ is bounded, show that
Wz Tϕ Wz∗ dλα (z) = Tϕ .
C
7. Show that there exist functions ϕ such that ϕ ∈ L∞ (C) but Tϕ is not bounded
on Fα2 .
8. Show that there exist functions ϕ such that ϕ(z) → 0 as z → ∞ but Tϕ is not
compact on Fα2 .
9. Suppose ϕ is radial, that is, ϕ (z) = ϕ (|z|) for all z ∈ C. If ϕ satisfies condition
(I1 ), show that the densely defined Toeplitz operator Tϕ is diagonal with
respect to the standard basis of Fα2 . Characterize boundedness, compactness,
and membership in the Schatten classes for such Toeplitz operators in terms of
the moments of ϕ .
2 )
10. Suppose ϕ (z) = ei|z| . Show that Tϕ is in the trace class, but C |ϕ | dA = ∞.
266 6 Toeplitz Operators
α /r2
Tϕαr f (z) = Tϕ f1/r (rz)
α /r2
Tϕ : Fα2/r2 → Fα2/r2 .
16. Suppose 1 ≤ p < ∞ and Bβ ϕ ∈ L p (C, dA) for some β > 2α . Then the Toeplitz
operator Tϕ : Fα2 → Fα2 belongs to the Schatten class S p . See [30] and [61].
2
17. Let c be a complex constant and ϕ (z) = ec|z| . Show that Tϕ is bounded on Fα2
∞
if and only if B2α ϕ ∈ L (C), Tϕ is compact on Fα2 if and only if B2α ϕ ∈ C0 (C),
and Tϕ belongs to the Schatten class S p if and only if B2α ϕ ∈ L p (C, dA). See
[30].
18. Define T : Fα2 → Fα2 by T f (z) = f (−z). Show that Tϕ − T ≥ 1 for any
bounded Toeplitz operator Tϕ on Fα2 . See [30].
19. Suppose T is a finite sum of finite products of Toeplitz operators on Fα2 induced
by bounded symbols. Show that T is compact on Fα2 if and only if T ∈ C0 (C).
20. Suppose ϕ (z) = | f (z)|e−σ |z| , where f is entire and σ > 0. Show that Tϕ is
2
In this chapter, we study small Hankel operators on the Fock space Fα2 . Problems
considered in the chapter include boundedness, compactness, and membership in
the Schatten class S p . We will also determine when a small Hankel operator has
finite rank.
Recall that
P : L2α → Fα2
is the orthogonal projection. Let
2
Fα = f : f ∈ Fα2
2
and use P to denote the orthogonal projection from L2α onto F α .
Suppose ϕ is a function on C that satisfies condition (I1 ). Using the integral
representation for P (and hence P) we can define an operator hϕ on a dense subset
of Fα2 by
hϕ f (z) = P(ϕ f )(z) = K(w, z)ϕ (w) f (w) dλα (w).
C
n
f (z) = ∑ ck K(z, ak )
k=1
is a finite linear combination of reproducing kernels. The set of all such f is a dense
subspace of Fα2 .
The operator hϕ is traditionally called the small (or little) Hankel operator with
symbol ϕ . We say that hϕ is bounded on Fα2 if there exists a constant C > 0 such
that hϕ ( f )α ≤ C f α whenever f is a finite linear combination of reproducing
kernels. In this case, the domain of hϕ can be extended to the whole space Fα2 .
270 7 Small Hankel Operators
7.2 Boundedness and Compactness 271
Theorem 7.1. Suppose ϕ ∈ Fα2 . Then, hϕ is bounded on Fα2 if and only if ϕ ∈ Fα∞/2 ,
that is, there exists a constant C > 0 such that
1
ϕ Fα∞/2 ≤ hϕ ≤ ϕ Fα∞/2 .
2
Proof. First, suppose that hϕ is bounded on Fα2 . Then there exists a positive constant
C such that
|hϕ f , g| ≤ hϕ f g, f , g ∈ Fα2 ,
or
f (w)g(w)ϕ (w) dλα (w) ≤ hϕ f g, f , g ∈ Fα2 .
C
Rewrite this as
−α |z|2
e Ce
(2α z)w
ϕ (w) dλα (w) ≤ hϕ , z ∈ C.
This shows that the small Hankel operator hϕ is bounded, and we have the norm
estimate hϕ ≤ 2ϕ Fα∞/2 .
Theorem 7.2. Suppose ϕ ∈ Fα2 . Then hϕ is compact on Fα2 if and only if f ∈ fα∞/2 ,
that is,
lim e−α |z|
2 /4
ϕ (z) = 0. (7.2)
z→∞
Proof. First, assume that ϕ is an entire function that satisfies condition (7.2). Then
there exists a sequence of polynomials {pk } such that
z→∞
Our next goal is to characterize small Hankel operators induced by entire functions
that belong to the Schatten classes S p . As usual, the cases 1 ≤ p < ∞ and 0 < p < 1
require different treatments. More specifically, we use complex interpolation for the
case 1 ≤ p < ∞, and we use atomic decomposition for the case 0 < p < 1.
Theorem 7.4. Suppose 1 ≤ p ≤ ∞, β = α /2, and ϕ is an entire function satisfying
condition (I1 ). Then hϕ is in the Schatten class S p if and only if ϕ ∈ Fβp .
β
If we write Kw (z) = eβ zw for the reproducing kernel in Fβ2 , then it follows from
Fubini’s theorem that for polynomials f and g we have
hϕ f , g = f (z)g(z)ϕ (z) dλα (z)
C
β
= ϕ (w) dλβ (w) f (z)g(z)Kw (z) dλα (z)
C C
= ϕ (w) h β f , g dλβ (w).
C Kw
In the sense of Banach space valued integrals, we can rewrite the above as
hϕ = ϕ (w) h β dλβ (w). (7.3)
C Kw
This shows that hϕ belongs to the trace-class S1 whenever ϕ is in Fβ1 . On the other
hand, we have already shown in the previous section that hϕ is in S∞ whenever
ϕ ∈ Fβ∞ . An application of complex interpolation then shows that, for 1 ≤ p ≤ ∞,
the small Hankel operator hϕ is in the Schatten class S p whenever ϕ ∈ Fβp .
276 7 Small Hankel Operators
On the other hand, if the small Hankel operator hϕ belongs to the Schatten class
S p , where 1 ≤ p < ∞, then according to Proposition 3.5 and its proof, the function
Φ (z) = hϕ kz , kz
is in L p (C, dA), where kz are the normalized reproducing kernels of Fα2 . We compute
that
Φ (z) = ϕ (w)kz2 (w) dλα (w)
C
= e−α |z| ϕ (w)e2α zw dλα (w)
2
e−α |z|
2 /4
ϕ (z) ∈ L p (C, dA),
which in turn is equivalent to ϕ ∈ Fβp . This completes the proof of the theorem.
Note that if ϕ ∈ Fβ1 , we can also use atomic decomposition to prove that the
operator hϕ is in S1 . See the first part of the proof of the next theorem.
Theorem 7.5. Suppose 0 < p < 1, β = α /2, and ϕ is an entire function satisfying
condition (I1 ). Then hϕ is in the Schatten class S p if and only if ϕ ∈ Fβp .
∞
ϕ (z) = ∑ ck ϕk (z),
k=1
We may also assume that the sequence {zk } is dense enough to be a sampling
sequence for Fβp . Moreover, there is a constant C > 0, independent of ϕ , such that
∞
∑ |ck | p ≤ Cϕ pp,β .
k=1
7.3 Membership in Schatten Classes 277
It follows that
p
∞ ∞
hϕ Spp = ∑ ck hϕ k ≤ ∑ |ck | p hϕ k Spp .
k=1 k=1
Sp
α α
−β |zk |2 /2
=e f (β zk /α )Kzα (β zk /α ) = f (zk /2)ϕ k (z)
+ ,
= f , Kzαk /2 ϕ k (z).
α
Therefore,
hϕ k S p = hϕ k ≤ Kzαk /2 2,α ϕk 2,α = 1,
and so
∞
hϕ Spp ≤ ∑ |ck | p ≤ Cϕ pp,β .
k=1
On the other hand, if hϕ is in S p , we are going to show that ϕ ∈ Fβp . To this end,
we fix a square lattice Z = {zk } in C such that atomic decomposition holds on Z
for both Fβp and Fα2 . We also assume that 2Z is a sampling sequence for Fβp . Fix a
sufficiently large R and use Lemma 1.14 to decompose Z = Z1 ∪ · · · ∪ ZN into N
square lattices such that for each 1 ≤ k ≤ N and each pair {w1 , w2 } of distinct points
in Zk , we have |w1 − w2 | > R.
Fix an orthonormal basis {ek } for Fα2 and define an operator A on Fα2 as follows:
∞ ∞ α
∑ ck ek ∑ ck eα zzk − 2 |zk | .
2
A (z) =
k=1 k=1
By the atomic decomposition for Fα2 , the operator A is bounded and onto. Clearly,
we have A = A1 + · · · + AN , where
∞
α
∑ ck ek ∑ ck eα zzk − 2 |zk |
2
Aj (z) =
k=1 zk ∈Z j
B j : Fα2 → Fα2
278 7 Small Hankel Operators
defined by
B j( f ) = A j f , f ∈ Fα2 , 1 ≤ j ≤ N.
Since hϕ is in S p , so is the operator T = T1 + · · · + TN , where
T j = B∗j hϕ A j , 1 ≤ j ≤ N.
Write T = D + E, where D is diagonal with respect to the basis {ek } and satisfies
Dek , ek α = Tek , ek α for all k ≥ 1. If we write
α
fk (z) = eα zzk − 2 |zk | ,
2
then
∞ ∞
DSpp = ∑ |Dek , ek | p = ∑ |Tek , ek | p
k=1 k=1
∞ ∞ p
∑ |hϕ fk , f k | p = ∑ ϕ (2zk )e−α |zk |
2
=
k=1 k=1
≥ Cϕ pp,β ,
where C is a positive constant independent of ϕ . Note that the last inequality above
follows from the assumption that {2zk } is a sampling sequence for Fβp .
On the other hand, since 0 < p < 1, it follows from Lemma 6.36 that
k,l k=l
N
= ∑ ∑ |hϕ A j ek , A j el | p.
j=1 k=l
Since
hϕ A j ek , A j el = 0
unless both zk and zl are in Z j , we see that
N
ESpp ≤ ∑∑ |hϕ fk , f l | p : k = l, zk ∈ Z j , zl ∈ Z j .
j=1
∞
ϕ (z) = ∑ ci ϕi (z),
i=1
7.3 Membership in Schatten Classes 279
where
β
ϕi (z) = eβ zzi − 2 |zi |
2
and
∞
∑ |ci | p ≤ Cϕ βp ,p.
i=1
∞ N
ESpp ≤ ∑ |ci | p ∑ ∑ |hϕ i fk , f l | p : k = l, zk ∈ Z j , zl ∈ Z j .
i=1 j=1
Therefore,
∞
ESpp ≤ ∑ |ci | p ∑ e−pβ |zk −(zi /2)|
2 −pβ |z
l −(zi /2)|
2
.
i=1 |zk −zl |≥R
If 2δ is the separation constant for the sequence Z, then by Lemma 2.32, there
exists a positive constant C = C(δ , α , p) such that
−pβ [|zk − 2i |2 +|zl − 2i |2 ] e−pβ [|z|2 +|w|2 ] dA(z)dA(w).
z z
e ≤C
z z
B zk − 2i ,δ ×B zl − 2i ,δ
If (k, l) = (k , l ), then
zi
zi
zi
zi
B zk − , δ × B zl − , δ ∩ B zk − , δ × B zl − , δ = 0.
/
2 2 2 2
Also,
zi
zi
B zk − , δ × B zl − , δ ⊂ (z, w) ∈ C2 : |z − w| ≥ R − 2δ .
2 2
It follows that there exists a positive constant C, independent of large R, such that
The above double integral tends to 0 as R → ∞. This, along with the fact that
DSpp ≤ 2 p T Spp + ESpp ,
In this section, we characterize small Hankel operators on Fα2 whose range is finite
dimensional. Such operators are called finite rank operators.
We begin with an example. Suppose ϕ (z) = K(a, z) for some point a ∈ C. Then
for any function f ∈ Fα2 , we have
hϕ ( f )(z) = P(ϕ f )(z) = K(w, z)K(a, w) f (w) dλα (w)
C
= f (a)K(a, z).
So, in this case, the range of hϕ is the one-dimensional subspace spanned by the
function z → K(a, z). More generally, if
N
∂k
ϕ (z) = ∑ ck ∂ ak K(a, z)
k=0
which shows that hϕ is a finite rank operator whose range is spanned by the
following functions of z:
∂k
K(a, z), 0 ≤ k ≤ N.
∂ ak
We are going to show that these are essentially all the finite rank small Hankel
operators on Fα2 . But we first need the following elementary result from algebra.
Lemma 7.6. Let P(C) denote the ring of all complex polynomials of the variable z.
If J is an ideal in P(C) containing at least one nonzero polynomial, then there
are a finite number of complex numbers ak , 1 ≤ k ≤ N, and for each k, there
exists a nonnegative integer Nk , such that J consists of all polynomials ϕ with
the property that
ϕ (i) (ak ) = 0, 1 ≤ k ≤ N, 0 ≤ i ≤ Nk .
Proof. By a well-known fact in abstract algebra (see [146] for example), every ideal
J = (0) of P(C) is generated by a polynomial, that is, there exists a polynomial q
such that J = {pq : p ∈ P(C)}. If a1 , · · · , aN are the zeros of q, and each zero ak has
multiplicity 1 + Nk , then J has the desired form.
282 7 Small Hankel Operators
Theorem 7.7. A bounded small Hankel operator has finite rank if and only if it can
be written as hϕ , where
N Nk
ϕ (z) = ∑ ∑ cki ϕki (z). (7.5)
k=1 i=0
∂i
K(a, z)
∂ ai
evaluated at the point a = ak .
Proof. We have already proved that hϕ has finite rank if ϕ is given by (7.5).
To prove the other direction, we write the small Hankel operator as hϕ , where ϕ
is conjugate analytic. If hϕ has finite rank, then the restriction of hϕ on P(C) also
has finite rank. Consider the kernel of hϕ on P(C):
J= f ∈ P(C) : hϕ ( f ) = 0 .
In other words, J is the intersection of the kernels of finitely many linear functionals
on P(C).
Let g = ϕ ∈ Fα2 . Then the linear functional on P(C) defined by
f → f , g = hϕ ( f ), 1
vanishes on J. Combining this with the conclusion from the previous paragraph, we
can find constants cki such that
5 6
N Nk N Nk
∂i
f , g = ∑ ∑ cki f (i)
(ak ) = f , ∑ ∑ cki i K(·, ak )
∂a
k=1 i=0 k=1 i=0
N Nk
∂i
ϕ (z) = g(z) = ∑ ∑ cki ∂ ai K(ak , z),
k=1 i=0
7.5 Notes
Small Hankel operators on the Fock space were first studied in [138], where the
boundedness, compactness, and membership in Schatten classes S p for 1 ≤ p < ∞
were characterized. The case when 0 < p < 1 was taken up and settled in [231]. Our
presentation here follows [138] and [231] very closely.
284 7 Small Hankel Operators
7.6 Exercises 285
7.6 Exercises
by
hϕ ( f ) = P(ϕ f ). Show that hϕ is bounded if and only if
hϕ is bounded, hϕ is
compact if and only if hϕ is compact, and hϕ is in the Schatten class S p if and
only if
hϕ is in the Schatten class S p .
2. Suppose ϕ is an entire function. Define a bilinear form
Φ : Fα2 × Fα2 → C
by
Φ ( f , g) = ϕ f , gα = ϕ f g dλα .
C
Show that hϕ is bounded on Fα2 if and only if there exists a constant C > 0 such
that |Φ ( f , g)| ≤ C f 2,α g2,α for all f and g in Fα2 .
3. Formulate conditions for compactness and membership in Schatten classes for
hϕ on Fα2 in terms of the bilinear form Φ in the previous problem, where ϕ is
any entire function.
4. Suppose ϕ ∈ L2α . Show that hϕ = 0 if and only if ϕ ⊥ Fα2 .
5. Consider the integral transform
Vϕ (z) = hϕ kz , kz α = ϕ (w)kz (w)2 dλα (w).
C
for all z ∈ C.
7. If ϕ ∈ L p (C, dλα ) for some 1 < p < ∞, then ϕ satisfies condition (I1 ). In
particular, every function in Fα2 satisfies condition (I1 ).
8. Show that if ϕ satisfies condition (I1 ) with respect to the weight parameter
β = 3α /4, then Pα (ϕ ) satisfies condition (I1 ) with respect to the weight
parameter α .
9. Show that Theorems 7.1 and 7.2 remain valid with the weaker assumption that
ϕ is entire and satisfies condition (I1 ).
10. Verify directly that hϕ has finite rank when ϕ is a polynomial.
11. Show that hϕr S p ≤ hϕ S p for all 0 < r < 1.
Chapter 8
Hankel Operators
In this chapter, we study (big) Hankel operators Hϕ on the Fock space Fα2 .
Problems considered include, again, boundedness, compactness, and membership
in the Schatten classes. There are basically two theories here: one concerns the
simultaneous size estimates for both Hϕ and Hϕ , and one concerns the size estimates
for the single operator Hϕ . The former is similar to the situations in the more
classical Hardy and Bergman space settings, while the latter is unique to the Fock
space setting.
Hϕ ( f ) = (I − P)(ϕ f ),
P : L2α → Fα2
all complex numbers z and w. Then ϕ satisfies condition (I1 ) and Hϕ ≤
for
2π /α C.
Proof. It is easy to check that any Lipschitz function satisfies condition (I1 ) and
hence induces a well-defined Hankel operator. To estimate the norm of Hϕ , consider
the integrals
I(z) = |z − w||K(z, w)|K(w, w)|1/2 dλα (w), z ∈ C.
C
290 8 Hankel Operators
By a change of variables,
α α 2
I(z) = |z − w| eα zw− 2 |w| dA(w)
π C
α α |z|2 α
|z − w|e− 2 |z−w| dA(w)
2
= e 2
π C
α α 2 α
= e 2 |z| |w|e− 2 |w| dA(w)
2
π C
2π α |z|2
= e2 .
α
Thus,
2π
|ϕ (z) − ϕ (w)||K(z, w)|K(w, w) 1/2
dλα (w) ≤ CK(z, z)1/2
C α
for all z ∈ C. The desired norm estimate then follows from Lemma 2.14 and the
integral representation of the Hankel operator Hϕ .
Recall that for any a ∈ C, we define a unitary operator Ua on L2α by Ua f =
f ◦ ϕa ka , where ϕa (z) = a − z and ka is the normalized reproducing kernel of Fα2 at
a. It is easy to check that Ua2 = I, so Ua∗ = Ua−1 = Ua . Since Ua leaves the Fock space
Fα2 invariant, we have Ua P = PUa .
Lemma 8.2. Suppose f satisfies condition (I2 ). Then the operators T f and H f are
both densely defined on Fα2 . Moreover, we have
T f kz = Uz P( f ◦ ϕz ) = P( f ◦ ϕz ) ◦ ϕzkz (8.1)
and
H f kz = Uz (I − P)( f ◦ ϕz ) = [ f − P( f ◦ ϕz ) ◦ ϕz ]kz (8.2)
for all z ∈ C.
Proof. Since each Uz commutes with the projection P, we have
T f kz = P( f kz ) = PUz ( f ◦ ϕz ) = Uz P( f ◦ ϕz ).
$
for all z ∈ C, where MO( f ) = |
f |2 − | f|2 .
Proof. Since each kz is a unit vector, it follows from the Cauchy–Schwarz inequality
that
Similarly, we have
H f kz = f ◦ ϕz − P( f ◦ ϕz ) = f ◦ ϕz − P( f ◦ ϕz ).
= H f kz + P( f ◦ ϕz ) − P( f ◦ ϕz )(0)
= H f kz + P[ f ◦ ϕz − P( f ◦ ϕz )]
≤ H f kz + f ◦ ϕz − P( f ◦ ϕz )
= H f kz + f ◦ ϕz − P( f ◦ ϕz )
= H f kz + H f kz .
Theorem 8.4. Suppose ϕ satisfies condition (I2 ). Then the following two conditions
are equivalent:
1. Both Hϕ and Hϕ are bounded on Fα2 .
2. The function ϕ belongs to BMO2 .
Proof. First, assume that ϕ ∈ BMO2 . Then, by Corollary 3.37, we can write ϕ =
ϕ1 + ϕ2 , where the function ϕ1 satisfies the Lipschitz estimate
and the Toeplitz operator T|ϕ2 |2 is bounded. By Lemma 8.1, the Hankel operator Hϕ1
is bounded on Fα2 . On the other hand, it follows from the identity
and the boundedness of T|ϕ2 |2 that Hϕ2 is also bounded on Fα2 . Therefore, Hϕ is
bounded. Since BMO2 is closed under complex conjugation, the assumption ϕ ∈
BMO2 implies that ϕ is also in BMO2 so that Hϕ is bounded on Fα2 as well.
Next, assume that both Hϕ and Hϕ are bounded on Fα2 . Then, it follows from the
inequality (see Proposition 8.3)
1/2
MO(ϕ )(z) = | ϕ |2 (z) − |ϕ(z)|2 ≤ Hϕ kz + Hϕ kz
The purpose of this section is to show that for bounded symbol functions ϕ , the
Hankel operator Hϕ is compact on Fα2 if and only if Hϕ is compact on Fα2 . This
striking result probably reflects the lack of bounded analytic functions (except the
constants) in the complex plane, as the direct analogs for Hankel operators on the
more classical Hardy and Bergman spaces are false.
Lemma 8.7. If f ∈ L∞ (C), then |P f (z)| ≤ f ∞ eα |z|
2 /4
for all z ∈ C.
Proof. This follows directly from Corollary 2.5.
1
4
α zw+ 12 α |z|2 1 α |w|2
F(w, ϕw (z))|e | dλα (z) ≤ Ce 2 F(w, z) dλα (z)
2
C C
for all w ∈ C.
Proof. We make an obvious change of variables to rewrite the integral on the left-
hand side as
α α |w|2 α
F(w, u) e− 2 |u| dA(u).
2
e2
π C
Denote the integral above by I, apply Hölder’s inequality with exponents 4 and 4/3,
and use the assumption in (8.3). We obtain
3α |u|2 α
F(w, u)e− e− 8 |u| dA(u)
2
I= 8
C
1 3
4 4
4 − 32α |u|2 − α6 |u|2
≤ F(w, u) e dA(u) e dA(u)
C C
1
α 4
F(w, u)2 e−α |u| dA(u)
2
≤C
π C
1
4
=C F(w, z)2 dλα (z) .
C
and
1 1
| f (z) − P( f ◦ ϕw )(ϕw (z))|Kw (z)|Kw (w) 2 dλα (w) ≤ 6 f ∞ Kz (z) 2 .
C
This proves the first estimate. The second estimate follows from the triangle
inequality, the first estimate, and the fact that
1 1
|Kw (z)|Kw (w) 2 dλα (w) = 2Kz (z) 2 .
C
Proof. The proof of part (a) is exactly the same as the proof of part (b). The only
difference is in the projection that is used in the definitions of Toeplitz operators and
Hankel operators: T f = PM f and H f = (I − P)M f . Therefore, we use Q to denote
either P or I − P in the rest of the proof.
Since ka → 0 weakly in Fα2 as a → ∞, the compactness of
QM f : Fα2 → Q(L2α )
is compact.
Given h ∈ Q(L2α ) and w ∈ C, we use Lemma 8.2 to write
SR : Q(L2α ) → L2α
by
SR h(w) = χR (w)(QM f )∗ h(w), w ∈ C,
≤ α R2 QM f 2 < ∞.
296 8 Hankel Operators
where
H(w, z) = (1 − χR(w))Q( f ◦ ϕw )(ϕw (z))Kw (z).
∗
test to obtain an estimate on the norm of (QM f ) −SR .
We are going to apply Schur’s
To this end, we let h(z) = K(z, z). It follows from Lemma 8.9 that
|H(w, z)|h(w) dλα (w) ≤ 6 f ∞ h(z)
C
1
4
|H(w, z)|h(z) dλα (z) ≤ Ch(w) F(w, z)2 dλα (z)
C C
1
= Ch(w)(1 − χR(w))Q( f ◦ ϕw ) 2 .
implies that
lim (QM f )∗ − SR = 0.
R→∞
8.2 Compact Hankel Operators with Bounded Symbols 297
f◦ ϕa − P( f ◦ ϕa ) ≤ C f ◦ ϕa − P( f ◦ ϕa ) 4
1
for all a ∈ C.
Proof. It follows from Corollary 2.5 that
α
| f(w) − P f (w)| ≤ 2 f ∞ e 4 |w|
2
(8.4)
for all w ∈ C. Since the Berezin transform fixes entire functions, we have
f(w) − P f (w) = ( f (z) − P f (z))|kw (z)|2 dλα (z)
C
so that
| f(w) − P f (w)| ≤ e−α |w|
2
| f (z) − P f (z)||Kw (z)|2 dλα (z) (8.5)
C
π C
2α
f ∞ | f(w) − P f (w)|e− 4 α |w| dA(w).
3 2
≤
π C
7 C
8α
f ∞ | f (z) − P f (z)|e− 8 α |z| e− 56 α |z| dA(z).
3 2 3 2
=
7π C
Applying Hölder’s inequality (with exponents 4 and 4/3) and Lemma 8.7, we obtain
1
4
f− P f 2 ≤ C1 f ∞ | f (z) − P f (z)| e4 − 23 α |z|2
dA(z)
C
1
3 4
≤ C2 f ∞2 | f (z) − P f (z)|2 dλα (z)
C
3
= C2 f ∞2 f − P f 1/2 .
298 8 Hankel Operators
lim f ◦ ϕa − P( f ◦ ϕa ) = 0,
a→∞
lim f◦ ϕa − P( f ◦ ϕa ) = 0.
a→∞
we see that
lim f◦ ϕa − P( f◦ ϕa ) = 0,
a→∞
g ◦ ϕa − P(g ◦ ϕa) = 0,
lim
a→∞
lim g ◦ ϕa − g ◦ ϕa = 0.
a→∞
8.2 Compact Hankel Operators with Bounded Symbols 299
Since complex conjugation commutes with the Berezin transform, we also have
lim f ◦ ϕa −
f ◦ ϕa = 0,
a→∞
which implies that H f − f is compact. Using Lemma 8.12 and iteration, we conclude
that H f − f(m) is compact for every positive integer m.
On the other hand, Theorem 3.25 shows that f ∈ L∞ (C) implies
C
| f(m) (z) − f(m) (w)| ≤ √ |z − w|,
m
which, along with Lemma 8.1, shows that H f(m) → 0 as m → ∞. This, combined
with the fact that each H f − f(m) is compact, shows that H f is compact.
300 8 Hankel Operators
8.3 Membership in Schatten Classes 301
In this section, we characterize when the Hankel operators H f and H f belong to the
Schatten class S p simultaneously. Throughout the section, we fix a positive radius r
and write
1
MOr ( f )(z) = |
f |2 r (z) − | fr (z)|2 ,
2
and
1
MO( f )(z) = |
f |2 (z) − | f(z)|2 .
2
Lemma 8.14. Let 2 ≤ p < ∞. If H f and H f are both in the Schatten class S p , then
MO( f ) ∈ L p (C, dA).
Proof. If H f is in S p , then (H ∗f H f ) p/2 is in the trace class S1 , so by Proposition 3.3,
(H ∗f H f ) p/2 kz , kz dA(z) < ∞,
C
or
H f kz p dA(z) < ∞.
C
Similarly, if H f is in S p , then
H f kz p dA(z) < ∞.
C
α
≤ H ∗ H f kz , kz p/2 dA(z)
π C f
α
= H f kz p dA(z) < ∞.
π C
Lemma 8.17. Let 1 ≤ p < ∞. There exists a positive constant C = C p such that
1 + |z| p−1
| f(z) − f(0)| p dλα (z) ≤ C [MO( f )(z)] p dλα (z) (8.6)
C C |z|
for all f .
Proof. Recall from the proof of Theorem 3.35 that there exists a positive constant
C = C(α ) such that
d
f(tz/|z|) ≤ CMO( f )(tz/|z|)
dt
satisfies
1
I ≤ Cp |z| p dλα (z) MO( f ) p (tz) dt
C 0
1
= Cp dt |z| p MO( f ) p (tz) dλα (z)
0 C
1
= C |z| p e−α |z| MO( f ) p (tz) dA(z)
2
dt
0 C
1
dt
|z| p e−α |z|
2 /t 2
=C MO( f ) p (z) dA(z)
0 t 2+p C
1
= C t −(2+p)e−α |z|
2 /t 2
|z| p MO( f ) p (z) dA(z) dt
C 0
∞
= C t p e− α t
2 |z|2
|z| p MO( f ) p (z) dA(z) dt
C 1
∞
MO( f ) p (z)
= C t p e−α t dt,
2
dA(z)
C |z| |z|
|z| 1
lim = .
|z|→∞ |z| p−1 e−α |z|
2
2α
It follows that there exists another constant C > 0, independent of z, such that
∞
t p e−α t dt ≤ C 1 + |z| p−1 e−α |z|
2 2
|z|
where ϕz (w) = z − w. By Lemma 8.17 and the invariance of the Berezin transform
under the action of ϕz , there exists a positive constant C, independent of f , such that
I ≤C dA(z) ϕ (w)MO( f ◦ ϕz ) p (w) dλα (w)
C C
=C dA(z) ϕ (w)MO( f ) p (ϕz (w)) dλα (w),
C C
where ϕ (w) = (1 + |w| p−1 )/|w|. Changing variables again and applying Fubini’s
theorem, we obtain
I ≤C dA(z) ϕ (ϕz (w))MO( f ) p (w)|kz (w)|2 dλα (w)
C C
=C MO( f ) p (w) dA(w) ϕ (ϕz (w))|kw (z)|2 dλα (z)
C C
=C MO( f ) (w) dA(w)
p
ϕ (u) dλα (u).
C C
converges. It follows that I < ∞, and by Lemma 8.16, the Hankel operator H f
belongs to S p .
Next, we consider the function g = f − f. By the triangle inequality,
1 1
2
2 (z) 2 =
|g| | f (w) − f(w)|2 |kz (w)|2 dλα (w)
C
8.3 Membership in Schatten Classes 305
1
2
≤
| f (w) − f (z)| |kz (w)| dλα (w)
2 2
C
1
2
+ | f(z) − f(w)|2 |kz (w)|2 dλα (w)
C
1
2
= MO( f )(z) + | f
◦ ϕz (0) − f
◦ ϕz (w)|2 dλα (w) .
C
By assumption, the first term above is in L p (C, dA). The second term is also in
L p (C, dA). In fact, since p ≥ 2, an application of Hölder’s inequality gives
p
2
| f ◦ ϕz (w) − f ◦ ϕz (0)| dλα (w) dA(z)
2
C C
≤ dA(z) | f
◦ ϕz (w) − f
◦ ϕz (0)| p dλα (w)
C C
≤C MO( f ) p (w) dA(w).
C
The last inequality above was proved in the previous paragraph. We conclude that
√2 2 belongs to
the function |g| belongs to L p (C, dA). In other words, the function |g|
L (C, dA). By Corollary 6.33, the Toeplitz operator T|g|2 belongs to the Schatten
p/2
the operator Hg∗ Hg belongs to the Schatten class S p/2 . This shows that Hg belongs
to S p , and consequently, H f = H f + H f − f belongs to S p . The condition MO( f ) ∈
L p (C, dA) is closed under complex conjugation, so we must have H f ∈ S p as well.
Recall that Z denotes the additive integer group and
Z2 = {n + im : n, m ∈ Z}
is the lattice of integers in the complex plane. Throughout this section, we fix a
positive integer N and consider the finer lattice
1 2 n + im
Z = : n, m ∈ Z .
N N
We also consider the following two special squares in the complex plane:
1 1
SN = x + iy : 0 ≤ x < , 0 ≤ y < ,
N N
306 8 Hankel Operators
and
1 2 1 2
QN = x + iy : − ≤ x < , − ≤ y < .
N N N N
γ (ν ) = {a0 , a1 , . . . , al }
≤ N 2 JN ( f ◦ tal ).
l
≤ 4l ∑ |( f ◦ ta j )SN − ( f ◦ ta j−1 )QN |2
j=1 SN
+ |( f ◦ ta j−1 )QN − ( f ◦ ta j−1 )SN |2 dA.
N4
= JN ( f ◦ ta j−1 ).
18
308 8 Hankel Operators
Similarly,
N4
|( f ◦ ta j−1 )QN − ( f ◦ ta j−1 )SN |2 ≤ JN ( f ◦ ta j−1 ).
18
Therefore,
l
4lN 4
2
SN
|( f ◦ tal )SN − ( f ◦ ta0 )SN |2 dA ≤
9 ∑ JN ( f ◦ ta j−1 ).
j=1
∑ ∑ e−α |ν |
2 /3
sup MO( f )2 (z) ≤ C JN ( f ◦ ta ).
z∈SN ν ∈Z2 /N a∈γ (ν )
= |
f |2 (z) − | f(z)|2 + | f(z) − c|2
≥ |
f |2 (z) − | f(z)|2 .
It is clear that
Z2
C= SN + u : u ∈ ,
N
1
≤ ∑ sup{MO( f ) p (z) : z ∈ SN + u}
N 2 u∈Z 2 /N
1
= ∑ sup {MO( f ) p (u + z) : z ∈ SN }
N 2 u∈Z 2 /N
1
= ∑ sup {MO( f ◦ tu) p (z) : z ∈ SN } .
N 2 u∈Z 2 /N
310 8 Hankel Operators
Since 0 < p ≤ 2, it follows from Lemma 8.20 and Hölder’s inequality that
pα
∑ ∑
p
e− 6 |ν |
2
sup MO( f ◦ tu ) p (z) ≤ CN [JN ( f ◦ tu ◦ ta )] 2 .
z∈SN ν ∈Z2 /N a∈γ (ν )
CN pα
∑ ∑ ∑
p
e− 6 |ν |
2
I≤ [JN ( f ◦ tu+a)] 2
N2 u∈Z2 /N ν ∈Z2 /N a∈γ (ν )
CN pα
∑ ∑ ∑
p
e− 6 |ν |
2
= [JN ( f ◦ tu+a)] 2
N2 ν ∈Z2 /N a∈γ (ν ) u∈Z2 /N
CN pα
∑ ∑
p
|γ (ν )|e− 6 |ν |
2
= [JN ( f ◦ tb )] 2 ,
N2 ν ∈Z2 /N b∈Z2 /N
ν ∈Z2 /N
1 2
Iν ( f ) ≥ CN JN ( f ◦ tν ), ν∈ Z ,
N
Fix any δ ∈ (0, 1/4) and choose a positive integer N such that
α 2 +iα Im (zw)
e− 2 |z−w| = 1 + γz,w, |γz,w | < δ ,
minus
2 ( f ◦ tν (z) − f ◦ tν ) dA ( f ◦ tν (z) − f ◦ tν )γ dA ,
QN QN
It follows that
2
Iν ( f ) ≥
QN
Q ( f ◦ tν (z) − f ◦ tν (w)) dA(w) dA(z)
N
2
−2δ | f ◦ tν (z) − f ◦ tν (w)| dA(w) dA(z).
QN QN
The first integral above can be written as [9/(2N 2 )]JN ( f ◦ tν ), and according to
the Cauchy–Schwarz inequality, the second integral above is less than or equal to
(9/N 2 )JN ( f ◦ tν ). We conclude that
9 1
Iν ( f ) ≥ − 2δ JN ( f ◦ tν ).
N2 2
In the remainder of this section, we fix a positive integer N such that Lemma 8.22
holds. We will need to decompose the lattice Z2 /N into more sparse sublattices. To
this end, we fix another positive integer M whose magnitude will be specified later.
For any j = ( j1 , j2 ), where each jk ∈ {1, 2, . . ., M}, we let
ν ν
1 2
ΛM j = ν = , : νk = jk mod M, k = 1, 2 .
N N
It is clear that
Z2 M
= ΛM
j ,
N j , j =1
1 2
the sublattices ΛM
j are disjoint, and the distance between any two points in the same
ΛMj is at least M/N.
Lemma 8.23. Suppose 0 < p < ∞ and f satisfies condition (I2 ). Then the Hankel
operators H f and H f both belong to the Schatten class S p if and only if the
commutator [M f , P] = M f P − PM f belongs to the Schatten class S p .
Proof. It is easy to see that
[M f , P] = [M f , P]P + [M f , P](I − P) = H f − H ∗f .
[M f , P]P = (M f P − PM f )P = M f P − PM f P = (I − P)M f P.
So the Hankel operator H f : Fα2 → L2α is just the restriction of [M f , P] on the space
Fα2 . It follows that the membership of [M f , P] in S p implies the membership of H f
in S p . But the condition [M f , P] ∈ S p implies [M f , P] ∈ S p , so [M f , P] ∈ S p implies
that both H f and H f are in S p .
Lemma 8.24. For any 2 ≤ p < ∞, there exists a positive constant C (depending on
N but independent of f ) such that
Here E is any bounded Borel set in C. Therefore, it suffices to show that there exists
a positive constant C, independent of f and E, such that
Since ST SS p ≤ ST S p S for all bounded operators S and all T ∈ S p , and since
MχE MχE = MχE , it suffices to estimate the S p norm of the operator:
Y= ∑ MχSN +u [M f , P]MχS
N +u
= ∑ Yv ,
u,u ∈F v∈Z2 /N
where
Yv = ∑ χF×F (u, u + v)MχSN +u [M f , P]MχSN +u+v .
u∈Z2 /N
Because A(SN ) = 1/N 2 , we have for any g ∈ L2local (C, dA) that
|g(z) − g(w)|2 dA(z) dA(w)
SN SN +v
≤2 |g(z) − gSN |2 + |gSN − g ◦ tv(w)|2 dA(z) dA(w)
SN SN
2 2
= |g − gSN |2 dA + |g ◦ tv − gSN |2 dA.
N2 SN N2 SN
that
1
|g − gSN |2 dA ≤ |g − gQN |2 dA ≤ JN (g).
SN SN 2
we obtain
|g(z) − g(w)|2 dA(z) dA(w)
SN SN +v
1 4 4
≤ 2 JN (g) + N + N |γ (v)| ∑ JN (g ◦ ta)
2
N 9 a∈γ (v)
1 4
≤ N 2 + 2 + N 4 |γ (v)| ∑ JN (g ◦ ta ),
N 9 a∈γ (v)
8.3 Membership in Schatten Classes 315
where γ (v) is the discrete path in Z2 /N from 0 to v (see Sect. 1.2). Let g = f ◦ tu in
the above estimate and use (8.9). We see that
∑
p
MχSN +u [M f , P]MχSN +u+v Sp2 ≤ h(v) [JN ( f ◦ tu ◦ ta )] 2 ,
a∈γ (v)
where
α 2 2
p 1 4
p + p−2
2 2
h(v) = Be− 2 |v| N 2 + 2 + N 4 |γ (v)| .
N 9
∑ ∑
p
Yv Spp ≤ h(v) [JN ( f ◦ tu ◦ ta )] 2
u∈Z2 /N a∈γ (v)
∑ ∑
p
= h(v) [JN ( f ◦ tb )] 2 . (8.10)
u∈Z2 /N b∈γ (v)+u
Therefore,
∑ ∑
p
[JN ( f ◦ tb )] 2
u∈Z2 /N b∈γ (v)+u
p % &
= ∑ [JN ( f ◦ tb )] 2 | u ∈ Z2 /N : b ∈ γ (v) + u |
b∈Z2 /N
∑
p
= (1 + |γ (v)|) [JN ( f ◦ tb )] 2 .
b∈Z2 /N
∑
p
Yv Spp ≤ h(v)(1 + |γ (v)|) [JN ( f ◦ tb )] 2 .
b∈Z2 /N
316 8 Hankel Operators
Consequently,
Y S p ≤ ∑ Yv S p
v∈Z2 /N
⎡ ⎤1
p
∑ ∑
1 p
≤ [h(v)(1 + |γ (v)|)] ⎣ p [JN ( f ◦ tb )] ⎦ .
2
v∈Z2 /N b∈Z2 /N
From Lemma 1.12, the definition of h(v), and the elementary inequality |γ (v)| ≤
2|v|, we see that the constant
∑
1
C= [h(v)(1 + |γ (v)|)] p
v∈Z2 /N
is finite. With this constant C, the inequality in (8.7) holds for any bounded Borel
set E ⊂ C.
Lemma 8.25. Suppose 0 < p < 2 and f satisfies condition (I2 ). If both H f and
H f are in the Schatten class S p , then MO( f ) ∈ L p (C, dA). Moreover, there exists a
positive constant C, independent of f , such that
[MO( f )(z)] p dA(z) ≤ C H f Spp + H f Spp .
C
and
χQN +ν (z)[M f , P]hν (z)
ζν (z) = , ν ∈ ΛM
j .
χQN +ν [M f , P]hν
A j eν = ζν , B j eν = h ν , ν ∈ ΛM
j .
8.3 Membership in Schatten Classes 317
It is easy to check that both A j and B j extend to bounded linear operators on L2α .
In fact, since each hν is supported on QN + ν and different QN + ν are disjoint, we
have
⎛ ⎞2 2
B j ⎝ ∑ cν eν ⎠ = ∑ cν hν (w) dλα (w)
ν ∈ΛM
j
C
ν ∈ΛM
j
= ∑M |cν |2
QN +ν
|hν (w)|2 dλα (w)
ν ∈Λ j
α
= ∑M |cν |2
QN +ν π
dA(w)
ν ∈Λ j
9α
π N2 ν∑
= |cν |2 .
∈ΛN j
√ √
This shows that B j ≤ (3 α )/(N π ). A similar argument shows that A j ≤ 1.
Let W j = A∗j [M f , P]B j for each j. Then,
W j S p ≤ A j [M f , P]S p B j .
Here, we used the first identity in the proof of Lemma 8.23 and the fact that, for any
positive p and any Schatten class operators S and T , we always have
Fix a very large natural number R and consider the truncation ZR of the lattice
Z2 /N:
% &
ZR = ν = (ν1 , ν2 ) ∈ Z2 /N : |νk | ≤ R, k = 1, 2 .
and set
E j = PZ j W j PZ j − D j .
Both D j and E j are finite rank operators, so they both belong to the Schatten class
S p . Also, it follows from (8.11) that
= ∑ χQN +ν [M f , P]hν p
ν ∈Z j
p
2
= ∑ QN +ν
|(M f P − PM f )hν | dλα 2
.
ν ∈Z j
Note that
∑ [JN ( f ◦ tν )] 2 .
p
D j Spp ≥ CN
ν ∈Z j
Next, we will obtain an upper bound for E j S p , which is much more involved
than the previous estimates. We begin with the following well-known fact from
operator theory: if 0 < p ≤ 2 and T is a compact operator on a separable Hilbert
space H, then
T Spp ≤ ∑ |Ten , em | p
n,m
8.3 Membership in Schatten Classes 319
E j Spp ≤ ∑ M |E j eν , eν | p
ν ,ν ∈Λ j
= ∑ |E j eν , eν | p
ν ,ν ∈Z j ,ν =ν
[M f , P]hν , χQN +ν [M f , P]hν p
= ∑
χ QN +ν [M f , P]h ν
ν ,ν ∈Z j ,ν =ν
χQN +ν [M f , P]hν , χQN +ν [M f , P]hν p
= ∑
χ [M f , P]h
QN +ν ν
ν ,ν ∈Z j ,ν =ν
≤ ∑ χQN +ν [M f , P]hν p .
ν ,ν ∈Z j ,ν =ν
2 p
α zw− α2 |w|2 −α iIm (ν w)
2
( f (z) − f (w))e dA(w) dλα (z)
+ν Q
QN N +ν
p
and apply the Cauchy–Schwarz inequality in the inner integral. We see that E j S p
is less than or equal to [(3α )/(N π )] p times
p
2
∑ QN +ν QN +ν
| f (z) − f (w)| e 2 −α |z−w|2
dA(w) dA(z) .
ν ,ν ∈Z j ,ν =ν
z → z + ν , w → w + ν ,
and estimating the resulting exponential function with the help of the triangle
inequality, we obtain a positive constant CN such that
pα
( M−3 pα | ν − ν | 2 p
N )
2
E j Spp ≤ CN e− 2
∑ e− 5 I(ν , ν ) 2 ,
ν ,ν ∈Z j ,ν =ν
where
I(ν , ν ) = | f ◦ tν (z) − f ◦ tν (w)|2 dA(w) dA(z).
QN QN
Therefore, if we also assume N ≥ 3, the double integral I(ν , ν ) is less than or equal
to 9(l + 2)/N 2 times
| f ◦ tν (z) − ( f ◦ tν )QN |2 dA(z)
QN
+ | f ◦ tν (w) − ( f ◦ tν )QN |2 dA(w)
QN
l
+ ∑ |( f ◦ tak−1 )QN − ( f ◦ tak )QN |2 .
k=1
8.3 Membership in Schatten Classes 321
Since 0 < p/2 < 1, I(ν , ν ) p/2 is less than or equal to [9(l + 2)/N 2 ] p/2 times
p
2
| f ◦ tν − ( f ◦ tν )QN | dA 2
(8.12)
QN
p
2
+ | f ◦ tν − ( f ◦ tν )QN | dA 2
(8.13)
QN
2
l p
p
− p4α ( M−3 9(l + 2) 2
N )
2
CN e
N2
times
p
2
− α5 |ν −ν |2
∑ e
QN
| f ◦ t − ( f ◦ t )QN | dA
ν ν
2
(8.15)
ν ,ν ∈Z j ,ν =ν
p
α 2
∑ e− 5 |ν −ν |
2
+ | f ◦ tν − ( f ◦ tν )QN |2 dA (8.16)
ν ,ν ∈Z j ,ν =ν
QN
p
l 2
− α5 |ν −ν |2
+ ∑ e ∑ |( f ◦ tak−1 )QN −( f ◦ tak )QN |2 . (8.17)
ν ,ν ∈Z j ,ν =ν k=1
which is equal to
pα
( M−3
N )
2
∑
p
Ce− 4 [JN ( f ◦ tν )] 2 ,
ν ∈Z j
322 8 Hankel Operators
where
2p
N2 pα 2
C=
18 ∑ e− 6 |ν −ν |
ν ∈Z j
p
N2 2 pα 2
≤
18 ∑ e− 6 |ν −ν |
ν ∈Z2 /N
2p
N2 pα
∑ e− 6 |ν |
2
= .
18 ν ∈Z2 /N
By symmetry, we get exactly the same estimate for the quantity in (8.16).
Since 0 < p/2 < 1, we can apply Hölder’s inequality in (8.17) and reduce our
estimate to the following quantity:
pα
l pα
( M−3
N )
2 2
S j = e− 4
∑ ∑ e− 6 |ν −ν | |( f ◦ tak−1 )QN − ( f ◦ tak )QN | p .
ν ,ν ∈Z j ,ν =ν k=1
Just like the computation we performed in the proof of Lemma 8.19, we have
Thus,
pα 2
∑ ∑
p
S j ≤ CN e− 6 |ν −ν | [JN ( f ◦ tu )] 2
ν ,ν ∈Z j ,ν =ν u∈γ (ν ,ν )
pα 2
∑ ∑ [JN ( f ◦ tu)] 2 χγ (ν ,ν ) (u)
p
= CN e− 6 |ν −ν |
ν ,ν ∈Z j ,ν =ν u∈Z j
pα 2
∑ [JN ( f ◦ tu)]
p
= CN 2
∑ e− 6 |ν −ν | χγ (ν ,ν ) (u).
u∈Z j ν ,ν ∈Z j ,ν =ν
By Lemma 1.15, there exists a constant C > 0, independent of u and R, such that
pα 2
∑ e− 6 |ν −ν | χγ (ν ,ν ) (u) ≤ C
ν ,ν ∈Z j ,ν =ν
for all j, where CN is yet another constant that depends on N only. Combining this
with the earlier lower estimate for D j S p , we see that there exist two constants CN1
and CN2 , which are both independent of M and R, such that
2 2 − p4α ( M−3 )
2
∑ [JN ( f ◦ tu)] 2 .
p
M 2
H f Spp + H f Spp ≥ CN − CN M e
1 N
u∈Z j
pα
( M−3
N ) > 0,
2
CN1 − CN2 M 2 e− 4
∑ [JN ( f ◦ tu)] 2
p
p p
H f S p + H f S p ≥ C
u∈Z j
for all j ∈ {1, 2, . . . , M} × {1, 2, . . . , M}. Summing over all such j, we obtain a
constant C > 0, independent of the truncating constant R, such that
∑
p
H f Spp + H f Spp ≥ C [JN ( f ◦ tu)] 2 .
u∈ZR
Let R → ∞. We obtain
∑
p
H f Spp + H f Spp ≥ C [JN ( f ◦ tu )] 2 .
u∈Z2 /N
This, along with Lemma 8.21, completes the proof of Lemma 8.26.
Theorem 8.26. Suppose 0 < p < ∞, r > 0, N is any positive integer, and f satisfies
condition (I2 ). Then the following conditions are equivalent:
(a) The operators H f and H f both belong to the Schatten class S p .
(b) The function
is in L p (C, dA).
(c) The function
is in L p (C, dA).
324 8 Hankel Operators
belongs to l p .
Proof. That (a) implies (b) follows from Lemmas 8.14 and 8.25. Lemmas 8.15 and
8.18 show that condition (b) implies (a). So (a) and (b) are equivalent.
By the double integral representations for MO( f ) and MOr ( f ), it is easy for us
to find a positive constant C = C(α , r) such that
This is possible because of the triangle inequality for the Euclidean metric.
Consider the function:
1
2
Fr (z) = | f (u) − f (v)|2 dA(u) dA(v) .
B(z,r) B(z,r)
Since MOr ( f ) and Fr differ only by a multiplicative constant, condition (c) implies
that Fr ∈ L p (C, dA).
Let
I= Fr (z) p dA(z).
C
Since the complex plane is the disjoint union of SN + ν , ν ∈ Z2 /N, it follows from
the mean value theorem and (8.18) that
1
I= ∑ Fr (z) p dA(z) = ∑ Fr (ζν ) p
N 2 ν ∈Z
ν ∈Z2 /N SN +ν 2 /N
p
1 2
= 2 ∑ | f (u) − f (v)| dA(u) dA(v)
2
N ν ∈Z2 /N B(ζν ,r) B(ζν ,r)
p
1 2
≥ 2 ∑ | f (u) − f (v)| dA(u) dA(v)
2
N ν ∈Z2 /N QN +ν QN +ν
1
∑ [JN ( f ◦ tν )] 2 .
p
=
N 2 ν ∈Z 2 /N
8.3 Membership in Schatten Classes 325
8.4 Notes
The study of Hankel operators on the Fock space goes back to [28] at least, where the
compactness was studied for Hankel operators induced by bounded symbols. This
compactness problem is equivalent to the symbol calculus for Toeplitz operators
with bounded symbols modulo compact operators.
The introduction of BMO (and VMO) defined with a fixed radius into the study
of Hankel and Toeplitz operators was first made in [257] in the context of Bergman
spaces in the unit disk. The extension to Fock spaces was first carried out in [32].
One of the unique features of the Fock space theory is the following: when f is
bounded, the Hankel operator H f is compact on Fα2 if and only if H f is compact. This
result is due to Berger and Coburn [28,29], and it is not true for Hankel operators on
the Bergman space or the Hardy space. A partial explanation for this difference is
probably the lack of bounded analytic or harmonic functions on the entire complex
plane.
The material in Sect. 8.3 concerning membership of the Hankel operators H f in
Schatten classes is mostly from [131, 242]. Again, there is a key difference between
the Fock and Bergman theories. In the Bergman space setting, there is a cutoff point
when the invariant mean oscillation MO( f ) is used to describe the membership of
H f and H f in S p , while in the Fock space setting, this cutoff point disappears because
of the exponential decay of the Fock kernel e−α |z| .
2
328 8 Hankel Operators
8.5 Exercises 329
8.5 Exercises
1. Show that on the space Fα2 , we have Wa = eiTψ for any a ∈ C, where ψ (z) =
2Im (az).
2. Show that H f and H f both belong to the Schatten S p if and only if the sequence
{MOr ( f )(ν ) : ν ∈ Z2 /N} belongs to l p , where r > 0 and N is any positive integer.
3. For f ∈ L∞ (C), show that H f is Hilbert–Schmidt if and only if H f is Hilbert–
Schmidt. See [12].
4. Show that Theorems 8.4 and 8.5 remain valid with the weaker assumption that
ϕ ∈ L2α .
5. Show that Hϕ∗ Hϕ = T|ϕ |2 − Tϕ Tϕ .
6. If f kz 2 ≤ C as for all z ∈ C, show that H f and H f are both bounded. Similarly,
if f kz → 0 as z → ∞, then H f and H f are both compact.
α
7. Show that | f (z) − P f (z)| ≤ 2 f ∞ e 4 |z| for almost all z ∈ C and f ∈ L∞ (C).
2
8. Define and study Hankel operators on the Fock space Fαp when 1 ≤ p ≤ ∞.
References
20. W. Bauer, Y.J. Lee, Commuting toeplitz operators on the Segal–Bargmann space. J. Funct.
Anal. 260, 460–489 (2011)
21. D. Bekolle, C. Berger, L. Coburn, K. Zhu, BMO in the Bergman metric on bounded symmetric
domains. J. Funct. Anal. 93, 310–350 (1990)
22. C. Bénéteau, B. Carswell, S. Kouchedian, Extremal problems in the Fock space. Comput.
Methods Funct. Theory 10, 189–206 (2010)
23. F.A. Berezin, Covariant and contravariant symbols of operators. Math. USSR-Izv. 6,
1117–1151 (1972)
24. F.A. Berezin, Quantization. Math. USSR-Izv. 8, 1109–1163 (1974)
25. F.A. Berezin, Quantization in complex symmetric spaces. Math. USSR-Izv. 9, 341–379
(1975)
26. F.A. Berezin, General concept of quantization. Comm. Math. Phys. 40, 153–174 (1975)
27. F.A. Berezin, The relation between co- and contra-variant symbols of operators on classical
complex symmetric spaces. Soviet Math. Dokl. 19, 786–789 (1978)
28. C. Berger, L. Coburn, Toeplitz operators and quantum mechanics. J. Funct. Anal. 68, 273–299
(1986)
29. C. Berger, L. Coburn, Toeplitz operators on the Segal–Bargmann space. Trans. Amer. Math.
Soc. 301, 813–829 (1987)
30. C. Berger, L. Coburn, Heat flow and Berezin–Toeplitz estimates. Amer. J. Math. 116, 563–590
(1994)
31. C. Berger, L. Coburn, A symbol calculus for Toeplitz operators. Proc. Nat. Acad. Sci. USA
83, 3072–3073 (1986)
32. C. Berger, L. Coburn, K. Zhu, Toeplitz operators and function theory in n-dimensions.
Springer Lect. Notes Math. 1256, 28–35 (1987)
33. J. Bergh, J. Löfstrom, Interpolation Spaces. Grundlehren Math. Wiss. 223, Springer, (1976)
34. S. Bergman, The Kernel Function and Conformal Mapping, Math. Surveys V, (American
Mathematical Society, Providence, RI, 1950)
35. B. Berndtsson, J. Ortega-Cerdá, On interpolation and sampling in Hilbert spaces of analytic
functions. J. Reine Angew. Math. 464, 109–128 (1995)
36. A. Beurling, The Collected Works of Arne Beurling, vol. 2, (Harmonic Analysis, Boston,
1989)
37. O. Blasco, A. Galbis, On taylor coefficients of entire functions integrable against exponential
weights. Math. Nachr. 223, 5–21 (2001)
38. R. Boas, Entire Functions, (Academic Press, New York, 1954)
39. H. Bommier-Hato, H. Youssfi, Hankel operators on weighted Fock space. Integr. Equ. Oper.
Theory 59, 1–17 (2007)
40. H. Bommier-Hato, H. Youssfi, Hankel operators and the Stieltjes moment problem. J. Funct.
Anal. 258, 978–998 (2010)
41. A. Borichev, R. Dhuez, K. Kellay, Sampling and interpolation in large Bergman and Fock
spaces. J. Funct. Anal. 242, 563–606 (2007)
42. A. Borichev, Y. Lyubarskii, Riesz bases of reproducing kernels in Fock-type spaces. (English
summary) J. Inst. Math. Jussieu 9, 449–461 (2010)
43. O. Bratteli, D. Robinson, Operator Algebras and Quantum Statistical Mechanics, I, (Springer,
New York, 1979)
44. O. Bratteli, D. Robinson, Operator Algebras and Quantum Statistical Mechanics, II,
(Springer, New York, 1981)
45. S. Brekke, K. Seip, Density theorems for sampling and interpolation in the Bargmann–Fock
spaces III. Math. Scand. 73, 112–126 (1993)
46. B. Carswell, B. MacCluer, A. Schuster, Composition operators on the Fock space. Acta. Sci.
Math. (Szeged) 69, 871–887 (2003)
47. J. Cheeger, M. Gromov, M. Taylor, Finite propagation speed, kernel estimates for functions
of the Laplace operator. J. Differ. Geom. 17, 15–53 (1982)
48. X. Chen, K. Guo, Analytic hilbert spaces over the complex plane. J. Math. Anal. Appl. 268,
684–700 (2002)
References 333
49. X. Chen, S. Hou, A Beurling type theorem for the Fock space. Proc. Amer. Math. Soc. 131,
2791–2795 (2003)
50. H. Cho, B. Choe, H. Koo, Linear combinations of composition operators on the Fock–Sobolev
spaces, preprint, (2011)
51. H. Cho, K. Zhu, Fock–Sobolev spaces and their Carleson measures, to appear in J. Funct.
Anal.
52. B. Choe, On higher dimensional Luecking’s theorem. J. Math. Soc. Japan 61, 213–224 (2009)
53. B. Choe, K. Izuchi, H. Koo, Linear sums of two composition operators on the Fock space.
J. Math. Anal. Appl. 369, 112–119 (2010)
54. L. Coburn, A sharp berezin lipschitz estimate. Proc. Amer. Math. Soc. 135, 1163–1168 (2007)
55. L. Coburn, A Lipschitz estimate for Berezin’s operator calculus. Proc. Amer. Math. Soc. 133,
127–131 (2005)
56. L. Coburn, The Bargmann isometry and Gabor–Daubechies wavelet localization operators, in
Systems, Approximation, Singular Integral Operators, and Related Topics, (Bordeaux, 2000),
169–178; Oper. Theory Adv. Appl. 129, (Birkhauser, Basel, 2001)
57. L. Coburn, On the Berezin–Toeplitz calculus. Proc. Amer. Math. Soc. 129, 3331–3338 (2001)
58. L. Coburn, The measure algebra of the Heisenberg group. J. Funct. Anal. 161, 509–525 (1999)
59. L. Coburn, Berezin–Toeplitz quantization, in Algebraic Methods in Operator Theory,
(Birkhauser, Boston, 1994) pp. 101–108
60. L. Coburn, Deformation estimates for the Berezin–Toeplitz quantization. Comm. Math. Phys.
149, 415–424 (1992)
61. L. Coburn, J. Isralowitz, B. Li, Toeplitz operators with BMO symbols on the Segal–Bargmann
space. Trans. Amer. Math. Soc. 363, 3015–3030 (2011)
62. L. Coburn, B. Li, Directional derivative estimates for Berezin’s operator calculus. Proc. Amer.
Math. Soc. 136, 641–649 (2008)
63. L. Coburn, J. Xia, Toeplitz algebras and Rieffel deformation. Comm. Math. Phys. 168, 23–38
(1995)
64. M. Christ, On the ∂ equation in weighted L2 norms in C1 . J. Geom. Anal. 1, 193–230 (1991)
65. R. Coifman, R. Rochberg, Representation theorems for holomorphic and harmonic functions
in L p . Astérisque 77, 11–66 (1980)
66. R. Coifman, R. Rochberg, G. Weiss, Factorization theorems for Hardy spaces in several
complex variables. Ann. Math. 103, 611–635 (1976)
67. J. Conway, Functions of One Complex Variable, (Springer, New York, 1973)
68. I. Daubechies, Time-frequency localization operators–a geometric phase space approach.
IEEE Trans. Inform. Th. 34, 605–612 (1988)
69. I. Daubechies, The wavelet transform, time-frequency localization, and signal analysis. IEEE
Trans. Inform. Th. 36, 961–1005 (1990)
70. I. Daubechies, Ten Lectures on Wavelets, (Society for Industrial and Applied Mathematics,
Philladephia, 1992)
71. I. Daubechies, A. Grossmann, Frames in the Bargmann space of entire functions. Comm.
Pure. Appl. Math. 41, 151–164 (1988)
72. I. Daubechies, A. Grossmann, Y. Meyer, Painless nonorthogonal expansions. J. Math. Phys.
27, 1271–1283 (1986)
73. I. Daubechies, T. Paul, Time-frequency localization operators– a geometric phase space
approach II. Inverse Probl. 4, 661–680 (1988)
74. M. Dostanić, K. Zhu, Integral operators induced by the Fock kernel. Integr. Equat. Operat.
Theor. 60, 217–236 (2008)
75. R.J. Duffin, A.C. Schaeffer, A class of nonharmonic Fourier series. Trans. Amer. Math. Soc.
72, 341–366 (1952)
76. P. Duren, Theory of H p Spaces, 2nd edn, (Dover Publications, New York, 2000)
77. P. Duren, B. Romberg, A. Shields, Linear functionals on H p spaces with 0 < p < 1. J. Reine
Angew. Math. 238, 32–60 (1969)
78. P. Duren, A. Schuster, Bergman Spaces, (American Mathematical Society, Providence, RI,
2004)
334 References
79. P. Duren, A. Schuster, D. Vukotić, On Uniformly Discrete Sets in the Unit Disk, in
Quadrature Domains and Applications, Oper. Theory Adv. Appl. 156, (Birkhaüser, Basel,
2005) pp. 131–150
80. P. Duren, G. Taylor, Mean growth and coefficients of H p functions. Illinois J. Math. 14,
419–423 (1970)
81. A.E. Dzhrbashyan, Integral representation and continuous projections in certain spaces of
harmonic functions, Mat. Sb. 121, 259–271 (1983); (Russian). English translation: Math.
USSR-Sb 49, 255–267 (1984)
82. M. Englis, Toeplitz Operators on Bergman-Type Spaces, Ph.D. thesis, (MU CSAV, Prague,
1991)
83. M. Englis, Some density theorems for Toeplitz operators on Bergman spaces. Czechoslovak
Math. J. 40, 491–502 (1990)
84. M. Englis, Functions invariant under the Berezin transform. J. Funct. Anal. 121, 233–254
(1994)
85. M. Englis, Compact Toeplitz operators via the Berezin transform on bounded symmetric
domains. Integr. Equat. Operat. Theor. 33, 326–355 (1999)
86. M. Englis, Berezin transform on the harmonic Fock space. J. Math. Anal. Appl. 367, 75–97
(2010)
87. M. Englis, Toeplitz operators and localization operators. Trans. Amer. Math. Soc. 361,
1039–1052 (2009)
88. H.G. Feichtinger, On a new segal algebra. Monatsh. f. Math. 92, 269–289 (1981)
89. E. Fischer, Über algebraische Modulsysteme und lineare homogene partielle Differentialgle-
ichungen mit konstanten Koeffizienten. J. Reine Angew. Math. 141, 48–81 (1911)
90. V. Fock, Verallgemeinerung und Lösung der Diracschen statistischen Gleichung. Z. Physik
49, 339–357 (1928)
91. V. Fock, Konfigurationsraum und zweite Quantelung. Z. Phys. 75, 622–647 (1932)
92. G. Folland, Harmonic Analysis in Phase Space, Ann. Math. Studies 122, (Princeton
University Press, Princeton, NJ, 1989)
93. G. Folland, Fourier Analysis and its Applications, Brooks/Cole Publishing Company, (Pacific
Grove, California, 1992)
94. F. Forelli, W. Rudin, Projections on spaces of holomorphic functions in balls. Indiana Univ.
Math. J. 24, 593–602 (1974)
95. O. Furdui, Norm calculations of composition operators on Fock spaces. Acta Sci. Math.
(Szeged) 74, 281–288 (2008)
96. O. Furdui, On a class of integral operators. Integr. Equat. Operat. Theor. 60, 469–483 (2008)
97. D. Gabor, Theory of communication. J. Inst. Elect. Eng. 93, 429–457 (1946)
98. D. Garling, P. Wojtaszczyk, Some Bargmann spaces of analytic functions. Lect. Notes Pure
Appl. Math. 172, 123–138 (1995)
99. H.F. Gautrin, Toeplitz operators in Bargmann spaces. Integr. Equat. Operat. Theor. 11,
173–185 (1988)
100. I.C. Gohberg, M.G. Krein, Introduction to the Theory of Linear Nonselfadjoint Operators,
(Nauka, Moscow, 1965); (Russian). English translation: Trans. Math. Monographs 18,
(American Mathematical Society, Providence, RI, 1969)
101. A. Grishin, A. Russakovskii, Free interpolation by entire functions. J. Sov. Math. 48, 267–275
(1990)
102. K. Grochenig, H. Razafinjatovo, On Landau’s necessary conditions for sampling and
interpolation of band-limited functions. J. London Math. Soc. 54, 557–565 (1996)
103. K. Grochenig, D. Walnut, A Riesz basis for the Bargmann–Fock space related to sampling
and interpolation. Ark. Math. 30, 283–295 (1992)
104. A. Grossmann, J. Morlet, Decomposition of hardy functions into square integrable wavelets
of constant shape. SIAM J. Math. Anal. 15, 723–736 (1984)
105. S.M. Grudsky, N.L. Vasilevski, Toeplitz operators on the Fock space: radial component
effects. Integr. Equat. Operat. Theor. 44, 10–37 (2002)
106. W. Gryc, T. Kemp, Duality in Segal-Bargmann spaces. J. Funct. Anal. 261, 1591–1623 (2011)
References 335
107. V. Guillemin, Toeplitz operators in n-dimensions. Integr. Equat. Operat. Theor. 7, 145–205
(1984)
108. K. Guo, Quasi-invariant subspaces generated by polynomials with nonzero leading terms.
Studia Math. 164, 231–241 (2004)
109. K. Guo, Homogeneous quasi-invariant subspaces of the Fock space. J. Aust. Math. Soc. 75,
399–407 (2003)
110. K. Guo, K. Izuchi, Composition operators on Fock type spaces. Acta Sci. Math. (Szeged) 74,
807–828 (2008)
111. K. Guo, D. Zheng, Invariant subspaces, quasi-invariant subspaces, and Hankel operators.
J. Funct. Anal. 187, 308–342 (2001)
112. B. Hall, W. Lewkeeratiyutkul, Holomorphic sobolev spaces and the generalized Segal–
Bargmann transform. J. Funct. Anal. 217, 192–220 (2004)
113. P. Halmos, V. Sunder, Bounded Integral Operator on L2 Spaces, (Springer, Berlin, 1978)
114. G. Hardy, J. Littlewood, Some new properties of fourier constants. Math. Ann. 97, 159–209
(1926)
115. G. Hardy, J. Littlewood, Some properties of conjugate functions. J. Reine Angew. Math. 167,
405–423 (1931)
116. W.W. Hastings, A carleson measure theorem for Bergman spaces. Proc. Amer. Math. Soc. 52,
237–241 (1975)
117. W. Hayman, On a conjecture of korenblum. Anal. (Munich) 19, 195–205 (1999)
118. H. Hedenmalm, An invariant subspace of the Bergman space having the co-dimension 2
property. J. Reine Angew. Math. 443, 1–9 (1993)
119. H. Hedenmalm, B. Korenblum, K. Zhu, Theory of Bergman Spaces, (Springer, New York,
2000)
120. H. Hedenmalm, S. Richter, K. Seip, Interpolating sequences and invariant subspaces of given
index in the Bergman space. J. Reine Angew. Math. 477, 13–30 (1996)
121. H. Hedenmalm, K. Zhu, On the failure of optimal factorization for certain weighted Bergman
spaces. Complex Variables Theor. Appl. 19, 165–176 (1992)
122. A. Hinkkanen, On a maximum principle in Bergman space. J. Anal. Math. 79, 335–344 (1999)
123. F. Holland, R. Rochberg, Bergman kernels and Hankel forms on generalized Fock spaces.
Contemp. Math. 232, 189–200 (1999)
124. F. Holland, R. Rochberg, Bergman kernel asymptotics for generalized Fock spaces. J. Anal.
Math. 83, 207–242 (2001)
125. L. Hörmander, The Analysis of Linear Partial Differential Operators, IV, (Springer, Berlin,
1985)
126. L. Hörmander, An Introduction to Complex Analysis in Several Complex Variables, 3rd edn.
(Van Nostrand, Princeton, NJ, 1990)
127. C. Horowitz, Zeros of functions in the Bergman spaces, Ph. D. thesis, (University of
Michigan, Ann Arbor, 1974)
128. C. Horowitz, Zeros of functions in the Bergman spaces. Duke Math. J. 41, 693–710 (1974)
129. C. Horowitz, Factorization theorems for functions in the Bergman spaces. Duke Math. J. 44,
201–213 (1977)
130. R. Howe, Quantum mechanics and partial differential equations. J. Funct. Anal. 38, 188–254
(1980)
131. J. Isralowitz, Schatten p class Hankel operators on the Segal–Bargmann space H 2 (Cn , dμ )
for 0 < p < 1. J. Operat. Theor. 66, 145–160 (2011)
132. J. Isralowitz, K. Zhu, Toeplitz operators on the Fock space. Integr. Equat. Operat. Theor. 66,
593–611 (2010)
133. K. Izuchi, Cyclic vectors in the Fock space over the complex plane. Proc. Amer. Math. Soc.
133, 3627–3630 (2005)
134. K. Izuchi, K. Izuchi, Polynomials having leading terms over C2 in the Fock space. J. Funct.
Anal. 225, 439–479 (2005)
135. J. Janas, Unbounded Toeplitz operators in the Segal–Bargmann space. Studia Math. 99, 8799
(1991)
336 References
136. J. Janas, J. Stochel, Unbounded Toeplitz operators in the Segal–Bargmann space II. J. Funct.
Anal. 126, 418-447 (1994)
137. S. Janson, P. Jones, Interpolation between H p spaces: the complex method. J. Funct. Anal.
48, 58–80 (1982)
138. S. Janson, J. Peetre, R. Rochberg, Hankel forms and the Fock space. Revista Mat. Ibero-Amer.
3, 61–138 (1987)
139. J.R. Klauder, B.S. Skagerstam, Coherent States–Applications in Physics and Mathematical
Physics, (World Scientific Press, Singapore, 1985)
140. W. Knirsch, G. Schneider, Continuity and Schatten–von Neumann p-class membership of
Hankel operators with anti-holomorphic symbols on generalized Fock spaces. J. Math. Anal.
Appl. 320, 403–414 (2006)
141. B. Korenblum, A maximum principle for the Bergman space. Publ. Mat. 35, 479–486 (1991)
142. S. Krantz, Function Theory of Several Complex Variables, 2nd edn. (American Mathematical
Society, Providence, RI, 2001)
143. O. Kures, K. Zhu, A class of integral operators on the unit ball of Cn . Integr. Equ. Oper.
Theory 56, 71–82 (2006)
144. H. Landau, Necessary density conditions for sampling and interpolation of certain entire
functions. Acta Math. 117, 37–52 (1967)
145. H. Landau, Sampling, data transmission, and the Nyquist rate. Proc. IEEE 55, 1701–1706
(1967)
146. S. Lang, Algebra, Graduate Texts in Mathematics, 211, (Springer, New York, 2002)
147. B. Levin, Lectures on Entire Functions, Transl. Math. Monographs 150, (American Mathe-
matical Society, Providence, RI, 1996)
148. B. Levin, Y. Lyubarskii, Interpolation by special classes of entire functions and related
expansions in exponential series, Izv. Akad. Nauk SSSR Ser. Mat. 39, 657–702 (1975);
(Russian). English translation: Math. USSR-Izv 9, 621–662 (1975)
149. H. Li, BMO, VMO, and Hankel operators on the Bergman space of strongly pseudo-convex
domains. J. Funct. Anal. 106, 375–408 (1992)
150. N. Lindholm, Sampling in weighted L p spaces of entire functions in Cn and estimates of the
Bergman kernel. J. Funct. Anal. 182, 390–426 (2001)
151. D. Luecking, Forward and reverse Carleson inequalities for functions in Bergman spaces and
their derivatives. Amer. J. Math. 107, 85–111 (1985)
152. D. Luecking, Trace ideal criteria for Toeplitz operators. J. Funct. Anal. 73, 345–368 (1987)
153. D. Luecking, Finite rank Toeplitz operators on the Bergman space. Proc. Amer. Math. Soc.
136, 1717–1723 (2008)
154. W. Lusky, On the Fourier series of unbounded harmonic functions. J. London Math. Soc. 61,
568–580 (2000)
155. Y. Lyubarskii, Frames in the Bargmann space of entire functions. Adv. Soviet Math. 429,
107–113 (1992)
156. T. MacGregor, K. Zhu, Coefficient multipliers between the Bergman and Hardy spaces.
Mathematika 42, 413–426 (1995)
157. M. Morris, Understanding Quantum Physics, (Prentice Hall, New Jersey, 1990)
158. Y. Lyubarskii, K. Seip, Sampling and interpolation of entire functions and exponential
systems in convex domains. Ark. Mat. 32, 157–193 (1994)
159. Y. Lyubarskii, K. Seip, Complete interpolating sequences for Paley–Wiener spaces and
Muckenhoupt’s A p -condition. Rev. Mat. Iberoamer. 13, 361–376 (1997)
160. N. Marco, X. Massaneda, J. Ortega-Cerdá, Interpolating and sampling sequences for entire
functions. Geom. Funct. Anal. 13, 862–914 (2003)
161. X. Massaneda, P. Thomas, Interpolating sequences for Bargmann–Fock spaces in Cn . Indag.
Math. 11, 115–127 (2000)
162. A. Nakamura, F. Ohya, H. Watanabe, On some properties of functions in weighted Bergman
spaces. Proc. Fac. Sci. Tokai Univ. 15, 33–44 (1979)
163. D. Newman, H. Shapiro, Certain Hilbert spaces of entire functions. Bull. Amer. Math. Sco.
72, 971–977 (1966)
References 337
164. D. Newman, H. Shapiro, Fischer Spaces of Entire Functions, in 1968 Entire Functions and
Related Parts of Analysis, (American Mathematical Society, Providence, RI, 1966)
165. D. Newman, H. Shaprio, A Hilbert Space of Entire Functions Related to the Operational
Calculus, unpublished manuscript, (1964, 1971)
166. J. von Neumann, Foundations of Quantum Mechanics, (Princeton University Press, Princeton,
1955)
167. V.L. Oleinik, Carleson measures and the heat equation. J. Math. Sci. 101, 3133–3188 (2000)
168. J. Ortega-Cerdá, K. Seip, Beurling type density theorems for sampling and interpolation in
weighted L p spaces of entire functions. J. Anal. Math. 75, 247–266 (1998)
169. J. Ortega-Cerdá, K. Seip, Multipliers for entire functions and an interpolation problem of
Beurling. J. Funct. Anal. 161, 400–415 (1999)
170. J. Ortega-Cerdá, K. Seip, Fourier frames. Ann. Math. 155, 789–806 (2002)
171. J. Peetre, Paracommutators and Minimal Spaces, in Operators and Function Theory ed. S.C.
Power (Reidel, Dordrecht, 1985) pp. 163–224
172. J. Peetre, Invariant function spaces and Hankel operators–a rapid survey. Exposition. Math.
5, 5–16 (1987)
173. V.V. Peller, Hankel operators of the class S p and their applications (rational approximation,
Gaussian processes, the problem of majorizing operators). Mat. Sb. 113, 538–581 (1980);
(Russian). English translation: Math. USSR-Sb. 41, 443–479 (1980)
174. V.V. Peller, Hankel Operators and Their Applications, (Springer, New York, 2003)
175. L. Peng, Paracommutators of Schatten–Von Neumann class S p , 0 < p < 1. Math Scand. 61,
68–92 (1987)
176. A.M. Perelomov, On the completeness of a system of coherent states. Theor. Math. Phys. 6,
156–164 (1971)
177. A.M. Perelomov, Generalized Coherent States and Their Applications, (Springer, Berlin,
1986)
178. S. Pichorides, On the best values of the constants in the theorems of M. Riesz, Zygmund, and
Kolmogorov. Studia Math. 44, 165–179 (1972)
179. J. Pool, Mathematical aspects of the Weyl correspondence. J. Math. Phys. 7, 66–76 (1966)
180. S.C. Power, Hankel operators on Hilbert space. Bull. London Math. Soc. 12, 422–442 (1980)
181. S.C. Power, Hankel operators on Hilbert space. Res. Notes Math. 64, 87 (1982)
182. S.C. Power, Finite rank multivariable Hankel forms. Lin. Algebra Appl. 48, 237–244 (1982)
183. D. Quillen, On the representation of Hermitian forms as sums of squares. Invent. Math. 5,
237–242 (1968)
184. E. Ramirez de Arellano, N. Vasilevski, Toeplitz operators on the Fock space with presymbols
discontinuous on a thick set. Math. Nachr. 180, 299–315 (1996)
185. E. Ramirez de Arellano, N. Vasilevski, Bargmann projection, three-valued functions, and
corresponding Toeplitz operators. Contemp. Math. 212, 185–196 (1998)
186. R. Rochberg, Trace ideal criteria for Hankel operators and commutators. Indiana Univ. Math.
J. 31, 913–925 (1982)
187. R. Rochberg, Decomposition Theorems for Bergman Spaces and Their Applications, in
Operators and Function Theory ed. by S.C. Power (Reidel, Dorddrecht, 1985) pp. 225–277
188. C. Rondeaux, Classes de Schatten d’opérateurs pseudo-différentiels. Ann. Sci. École Norm.
Sup. 4, 67–81 (1984)
189. G. Rozenblum, N. Shirokov, Finite rank Bergman–Toeplitz and Bargmann–Toeplitz operators
in many dimensions. Complex Anal. Oper. Theor. 4, 767-775 (2010)
190. W. Rudin, Functional Analysis, 2nd edn. (McGraw–Hill, New York, 1991)
191. W. Rudin, Function Theory in the Unit Ball of Cn , (Springer, New York, 1980)
192. Sangadji, K. Stroethoff, Compact Toeplitz operators on generalized Fock spaces. Acta Sci.
Math. (Szeged) 64, 657–669 (1998)
193. A. Schuster, On Seip’s description of sampling sequences for Bergman spaces. Complex
Variables 42, 347–367 (2000)
194. A. Schuster, The maximum principle for the Bergman space and the Möbius pseudodistance
for the annulus. Proc. Amer. Math. Soc. 134, 3525–3530 (2006)
338 References
195. A. Schuster, K. Seip, A Carleson type condition for interpolation in Bergman spaces. J. Reine
Angew. Math. 497, 223–233 (1998)
196. A. Schuster, K. Seip, Weak conditions for interpolation in holomorphic spaces. Publ. Mat. 44,
277–293 (2000)
197. A. Schuster, D. Varolin, Sampling sequences for Bergman spaces, 0 < p < 1. Complex
Variables 47, 243–253 (2002)
198. I.E. Segal, Lectures at the Summer Seminar on Applied Mathematics, (Boulder, Colorado,
1960)
199. I.E. Segal, Mathematical Problems of Relativistic Physics, (American Mathematical Society,
Providence, RI, 1963)
200. I.E. Segal, The Complex Wave Representation of the Free Boson Field, in Topics in
Functional Analysis: Essays dedicated to M.G. Krein on the occasion of his 70th birthday,
ed. by I. Gohberg, M. Kac (Academic Press, New York, 1978) pp. 321–343
201. K. Seip, Reproducing formulas and double orthogonality in Bargmann and Bergman spaces.
SIAM J. Math. Anal. 22, 856–876 (1991)
202. K. Seip, Regular sets of sampling and interpolation for weighted Bergman spaces. Proc. Amer.
Math. Soc. 117, 213–220 (1993)
203. K. Seip, Interpolation and Sampling in Spaces of Analytic Functions, (American
Mathematical Society, Providence, RI, 2004)
204. K. Seip, Density theorems for sampling and interpolation in the Bargmann–Fock space
I. J. Reine Angew. Math. 429, 91–106 (1992)
205. K. Seip, Beurling type density theorems in the unit disk. Invent. Math. 113, 21–39 (1993)
206. K. Seip, Density theorems for sampling and interpolation in the Bargmann–Fock space. Bull.
Amer. Math. Soc. 26, 322–328 (1992)
207. K. Seip, On Korenblum’s density condition for the zero sequences of A−α . J. Analyse Math.
67, 307–322 (1995)
208. K. Seip, Interpolating and sampling in small Bergman spaces, to appear in Collect. Math.
209. K. Seip, R. Wallstén, Density theorems for sampling and interpolation in the Bargmann–Fock
space II. J. Reine Angew. Math. 429, 107–113 (1992)
210. K. Seip, H. Youssfi, Hankel operators on Fock spaces and related Bergman kernel estimates.
preprint, (2010)
211. S. Semmes, Trace ideal criteria for Hankel operators and applications to Besov spaces. Integr.
Equat. Operat. Theor. 7, 241–281 (1984)
212. M. Shubin, Pseudodifferential Operators and Spectral Theory, (Springer, Berlin, 1987)
213. B. Simon, Trace Ideals and Their Applications, London Math. Soc. Lecture Notes Series 35,
(Cambridge University Press, Cambridge, 1979)
214. P. Sjögren, Un contre-exemple pour le noyau reproduisant de la mesure gaussienne dans
le plan complexe, Seminaire Paul Krée (Equations aux dérivées partienlles en dimension
infinite) 1975/76, Paris
215. E. Stein, Interpolation of linear operators. Trans. Amer. Math. Soc. 83, 482–492 (1956)
216. E. Stein, G. Weiss, Interpolation of operators with change of measures. Trans. Amer. Math.
Soc. 87, 159–172 (1958)
217. R. Strichartz, L p contractive projections and the heat semigroup for differential forms.
J. Funct. Anal. 65, 348–357 (1986)
218. K. Stroethoff, Hankel and Toeplitz operators on the Fock space. Mich. Math. J. 39, 3–16
(1992)
219. R. Supper, Zeros of functions of finite order. J. Inequal. Appl. 7, 49–60 (2002)
220. M. Tatari, S. Vaezpour, A. Pishinian, On some properties of Fock space Fα2 by frame theory.
Int. J. Contemp. Math. Sci. 5, 1107–1114 (2010)
221. D. Timotin, Cp estimates for certain kernels: the case 0 < p < 1, J. Funct. Anal. 72, 368–380
(1987)
222. H. Triebel, Interpolation Theory, Function Spaces, and Differential Operators, (VEB, Berlin,
1977)
223. J.Y. Tung, Fock spaces, Ph.D. thesis, (University of Michigan, Ann Arbor, 2005)
References 339
224. J.Y. Tung, Taylor coefficients of functions in Fock spaces. J. Math. Anal. Appl. 318, 397–409
(2006)
225. J.Y. Tung, Zero sets and interpolating sets in Fock spaces. Proc. Amer. Math. Soc. 134,
259–263 (2005)
226. J.Y. Tung, On Taylor coefficients and multipliers in Fock spaces. Contemp. Math. 454,
135–147 (2008)
227. G. Valiron, Sur la formule d’interpolation de Lagrange. Bull. Sci. Math. 49, 181–192 (1925)
228. N. Vasilevski, V. Kisil, E. Ramirez, R. Trujilo, Toeplitz operators with discontinuous
presymbols in the Fock space. Dokl. Math. 52, 345–347 (1995)
229. D. Vukotić, A sharp estimate for A p functions in Cn . Proc. Amer. Math. Soc. 117, 753–756
(1993)
230. D. Vukotić, On the coefficient multipliers of Bergman spaces. J. London Math. Soc. 50,
341–348 (1994)
231. R. Wallsten, The S p -criterion for Hankel forms on the Fock space, 0 < p < 1. Math. Scand.
64, 123–132 (1989)
232. C. Wang, Some results on Korenblum’s maximum principle J. Math. Anal. Appl. 373,
393–398 (2011)
233. C. Wang, Domination in the Bergman space and Korenblum’s constant. Integr. Equat. Operat.
Theor. 61, 423–432 (2008)
234. C. Wang, Behavior of the constant in Korenblum’s maximum principle. Math. Nachr. 281,
447–454 (2008)
235. C. Wang, On a maximum principle for Bergman spaces with small exponents. Integr. Equat.
Operat. Theor. 59, 597–601 (2007)
236. C. Wang, On Korenblum’s maximum principle. Proc. Amer. Math. Soc. 134, 2061–2066
(2006)
237. C. Wang, An upper bound on Korenblum’s maximum principle. Integr. Equat. Operat. Theor.
49, 561–563 (2004)
238. C. Wang, On Korenblum’s constant. J. Math. Anal. Appl. 296, 262–264 (2004)
239. C. Wang, Refining the constant in a maximum principle for the Bergman space. Proc. Amer.
Math. Soc. 132, 853–855 (2004)
240. E.T. Whittaker, On the functions which are represented by the expansions of interpolation
theory. Proc. R. Soc. Edinburgh 35, 181–194 (1915)
241. E.T. Whittaker, G.N. Watson, A Course of Modern Analysis, 4th edn. (Cambridge University
Press, Cambridge, 1996)
242. J. Xia, D. Zheng, Standard deviation and Schatten class Hankel operators on the Segal–
Bargmann space. Indiana Univ. Math. J. 53, 1381–1399 (2004)
243. J. Xia, D. Zheng, Two-variable Berezin transform and Toeplitz operators on the Fock space,
preprint
244. R.M. Young, An Introduction to Nonharmonic Fourier Series, (Academic Press, New York,
1980)
245. K. Zhu, Positive Toeplitz operators on weighted Bergman spaces of bounded symmetric
domains. J. Operat. Theor. 20, 329–357 (1988)
246. K. Zhu, A Forelli–Rudin type theorem. Complex Variables 16, 107–113 (1991)
247. K. Zhu, Schatten class Hankel operators on the Bergman space of the unit ball. Amer. J. Math.
113, 147–167 (1991)
248. K. Zhu, BMO and Hankel operators on Bergman spaces. Pacific J. Math. 155, 377–397 (1992)
249. K. Zhu, Zeros of functions in Fock spaces. Complex Variables 21, 87–98 (1993)
250. K. Zhu, Operator Theory in Function Spaces, 2nd edn. (American Mathematical Society,
Providence, RI, 2007)
251. K. Zhu, Spaces of Holomorphic Functions in the Unit Ball, (Springer, New York, 2005)
252. K. Zhu, Interpolating and recapturing in reproducing Hilbert spaces. Bull. Hong Kong Math.
Soc. 1, 21–33 (1997)
253. K. Zhu, Evaluation operators on the Bergman space. Math Proc. Cambridge Philos. Soc. 117,
513–523 (1995)
340 References
254. K. Zhu, Interpolating sequences for the Bergman space. Michigan Math. J. 41, 73–86 (1994)
255. K. Zhu, Invariance of Fock spaces under the action of the Heisenberg group. Bull. Sci. Math.
135, 467–474 (2011)
256. K. Zhu, Duality of Bloch spaces and norm convergence of Taylor series. Michigan Math. J.
38, 89–101 (1991)
257. K. Zhu, VMO, ESV, and Toeplitz operators on the Bergman space. Trans. Amer. Math. Soc.
302, 617–646 (1987)
258. K. Zhu, Maximal zero sequences for Fock spaces, preprint, (2011)
259. N. Zorboska, Toeplitz operators with BMO symbols and the Berezin transform. Int. J. Math.
Math. Sci. 46, 2929–2945 (2003)
Index
I
ideal, 281
N
identity operator, 20
Nevanlinna–Fock class, 211
identity theorem, 3
normalized reproducing kernel, 35
infinite order, 6
infinite rank, 5
infinite type, 6
initial condition, 102 O
integer group, 9 odd function, 14
integer lattice, 9 optimal rate of growth, 36
integral operator, 43 order, 6
integral pairing, 53 orthogonal projection, 34
integral representation, 35 orthonormal basis, 33
intermediate value theorem, 166
interpolating sequence for Fαp , 143
inverse Bargmann transform, 223 P
inverse Fourier transform, 20 parallelogram, 9
isometry, 76 parametrized Bargmann transform, 222
iterates of the Berezin transform, 110 parametrized Berezin transforms, 105
pathological properties, 199
period, 13
J periodicity, 13
Jensen’s formula, 4 permutation, 255
John–Nirenberg correspondence, 20 permutation invariance, 255
permutation invariant, 257
perturbation, 154
K Planck’s constant, 19, 87
Korenblum’s maximum principle, 87 pseudodifferential operator, 19
L Q
Lagrange-type interpolation formula, 164 quantum physics, 19
Laplacian, 104 quasi-periodic, 15
lattice, 9, 142, 277, 312 quasi-periodicity, 13, 201, 202
344 Index
R trace, 98
rank, 5 trace class, 96
rank of an operator, 258 trace class operator, 96
rank-one operator, 98 trace formula, 213
rank-two operator, 98 translation, 75
rate of growth, 36 translation invariance, 10, 75, 145
relatively closed set, 151 translation operator, 76
reproducing formula, 36
reproducing kernel, 34
Riemann ζ -function, 14
U
Riesz representation, 34
uniform density, 139
uniformly close, 160
S uniqueness sequence, 165
sampling sequence for Fαp , 144 uniqueness set, 165
sampling set, 145 unit mass, 22
Schatten class, 96, 97, 275 unitary operator, 26, 76
Schatten class Hankel operator, 301 unitary representation, 25, 76
Schatten class operator, 96 upper density, 139
Schatten class Toeplitz operator, 96
Schatten classes, 24
Schrödinger representation, 26
Schur’s test, 43 V
Schwarz lemma, 84 Vandermonde determinant, 258
semi-group property, 101 vanishing average, 130
separated sequence, 143 vanishing Carleson measure, 118
separation constant, 143 vanishing Fock–Carleson measure, 118
set of uniqueness, 165 vanishing mean oscillation, 130
small Hankel operator, 267 vanishing oscillation, 130
square lattice, 11, 16 vertices, 9
stability, 151
stable interpolation, 144
standar factorization, 5
Stein–Weiss interpolation theorem, 59 W
Stirling’s formula, 169 weak convergence, 151
Stone–Weierstrass theorem, 258 weak limit, 165
strong convergence, 151 Weierstrass σ -function, 13, 201
strong limit, 165 Weierstrass factorization, 4
strong operator topology, 265 Weierstrass factorization theorem, 202
sub-lattice, 10 Weierstrass functions, 15
symbol, 19 Weierstrass product, 16, 197
symbol calculus, 19 weighted translation operator, 76
symbol function, 19 Weyl pseudodifferential operator, 20
symmetric function, 255 Wick correspondence, 20
symmetric polynomial, 255 Wyle operator, 76
symmetrization, 255
T Z
telescoping decomposition, 306 zero sequence, 193
Toeplitz operator, 213 zero set, 193