This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.Lecture 6 Affine Group: the Real and Complex
Techniques in Harmonic Analysis
In this chapter we reviews complex and real techniques in harmonic
analysis.
We demonstrated that both, real and complex, techniques in harmonic
analysis have the same group-theoretical origin: the covariant
transform generated by the affine group. Moreover, they are
complemented by the wavelet construction. Therefore, there is no any
confrontation between these approaches and they can be lined up as in
Table 6.1. In other words, the binary opposition
of the real and complex methods resolves via Kant’s triad
thesis-antithesis-synthesis: complex-real-covariant.
6.1 Introduction
There are two main approaches in harmonic analysis on the real line.
The real variables technique uses various maximal functions, dyadic
cubes and, occasionally, the Poisson integral [314]. The
complex variable technique is based on the Cauchy integral and fine
properties of analytic functions [264, 265].
Both methods seem to have clear advantages. The real variable
technique:
- does not require an introduction of the imaginary unit for a
study of real-valued harmonic functions of a real variable (Occam’s
Razor: among competing hypotheses, the one with the fewest
assumptions should be selected);
- allows a straightforward generalization to several real variables.
By contrast, access to the beauty and power of analytic functions
(e.g., Möbius transformations, factorisation of zeroes,
etc. [231]) is the main reason to use the complex variable
technique. A posteriori, a multidimensional analytic version was also
discovered [252], it is based on the monogenic
Clifford-valued functions [49].
Therefore, propensity for either techniques becomes a personal choice
of a researcher. Some of them prefer the real variable method,
explicitly cleaning out any reference to analytic or harmonic
functions [314]*Ch. III, p. 88. Others,
e.g. [233, 68], happily combine the both
techniques. However, the reasons for switching between two methds at
particular places may look mysterious.
The purpose of the present paper is to revise the origins of the real
and complex variable techniques. Thereafter, we describe the common
group-theoretical root of both. Such a unification deepens our
understanding of both methods and illuminates their interaction.
Remark 1
In this paper, we consider only examples which are supported by the
affine group Aff of the real line. In the essence,
Aff is the semidirect product of the group of dilations
acting on the group of translations. Thus, our consideration can be
generalized to the semidirect product of dilations and homogeneous
(nilpotent) Lie groups, cf. [103, 201]. Other
important extensions are the group SL2(ℝ)
and associated hypercomplex algebras, see
Rems. 6, 17
and [191, 198, 197]. However, we do not aim here
to a high level of generality, it can be developed in subsequent
works once the fundamental issues are sufficiently clarified.
6.2 Two approaches to harmonic analysis
As a starting point of our discussion, we provide a schematic outline
of complex and real variables techniques in the one-dimensional harmonic
analysis. The application of complex analysis may be
summarised in the following sequence of principal steps:
- Integral transforms.
- For a function
f∈Lp(ℝ), we apply the Cauchy or Poisson integral transforms:
|
[C f] (x+i y) | | | | | | | | | | (1) |
[Pf](x, y) | | | | | | | | | | (2) |
|
An equivalent transformation on the unit circle replaces the
Fourier series ∑k ck ei kt by the Taylor series
∑k=0∞ck zk in the complex variable z=r ei t, 0≤
r<1. It is used for the Abel summation of trigonometric
series [344]*§ III.6. Some other summations methods
are in use as well [245].
- Domains.
- Above
integrals (1)–(2) map the
domain of functions from the real line to the upper half-plane,
which can be conveniently identified with the set of complex numbers
having a positive imaginary part. The larger domain allows us to
inspect functions in greater details.
- Differential operators.
- The image of
integrals (1) and (2) consists
of functions,belonging to the kernel of the Cauchy–Riemann operator
∂z and Laplace operator Δ
respectively, i.e.:
Such functions have numerous nice properties in the upper
half-plane, e.g. they are infinitely differentiable, which make
their study interesting and fruitful.
- Boundary values and SIO.
- To describe properties of the initial
function f on the real line we consider the boundary values of
[C f] (x+i y) or [Pf](x, y), i.e. their
limits as y→ 0 in some sense. The Sokhotsky–Plemelj
formula provides the boundary value of
the Cauchy integral [257]*(2.6.6):
[Cf](x,0)= | | f(x)+ | |
| ∫ | | | | dt.
(4) |
The last term is a singular integral operator defined through the principal
value in the Cauchy sense:
For the Abel summation the boundary values are replaced by the limit
as r→ 1− in the series ∑k=0∞ck (r ei t)k.
- Hardy space.
- Sokhotsky–Plemelj
formula (4) shows, that the boundary value
[Cf](x,0) may be different from f(x). The vector
space of functions f(x) such that [Cf](x,0)=f(x) is
called the Hardy space on the real line [264]*A.6.3.
Summing up this scheme: we replace a function (distribution) on the
real line by a nicer (analytic or harmonic) function on a larger
domain—the upper half-plane. Then, we trace down properties of
the extensions to its boundary values and, eventually, to the initial
function.
The real variable approach does not have a clearly designated path in
the above sense. Rather, it looks like a collection of interrelated
tools, which are efficient for various purposes. To highlight
similarity and differences between real and complex analysis, we line
up the elements of the real variable technique in the following way:
- Hardy–Littlewood maximal function
- is, probably, the most
important component [231]*§ VIII.B.1
[314]*Ch. 2 [108]*§ I.4 [54] of this technique. The maximal function
fM is defined on the real line by the identity:
fM(t)= | | | ⎧
⎪
⎨
⎪
⎩ | | | |
| ⎪
⎪ | f | ⎛
⎝ | x | ⎞
⎠ | ⎪
⎪ | dx | ⎫
⎪
⎬
⎪
⎭ | .
(6) |
- Domain
- is not apparently changed, the maximal function fM
is again defined on the real line. However, an efficient treatment
of the maximal functions requires consideration of tents [314]*§ II.2, which are parametrised by their
vertices, i.e. points (a,b), a>0, of the upper
half-plane. In other words, we repeatedly need values of all
integrals 1/2a ∫t−at+a
| f(x) | dx, rather than the single value of the
supremum
over a.
- Littlewood–Paley theory
- [68]*§ 3
and associated dyadic squares
technique [108]*Ch. VII, Thm. 1.1
[314]*§ IV.3 as well as stopping time
argument [108]*Ch. VI, Lem. 2.2 are based on bisection
of a function’s domain into two equal parts.
- SIO
- is a natural class of bounded linear operators in
Lp(ℝ). Moreover, maximal operator M:
f→ fM (32) and
singular integrals are intimately related [314]*Ch. I.
- Hardy space
- can be defined in several equivalent ways from
previous notions. For example, it is the class of such functions
that their image under maximal
operator (32) or singular
integral (5) belongs to
Lp(ℝ) [314]*Ch. III.
The following discussion will line up real variable objects along the
same axis as complex variables. We will summarize this in
Table 6.1.
6.3 Affine group and its representations
It is hard to present harmonic analysis and wavelets without touching
the affine group one way or another, e.g. through the doubling
condition on the measure,
cf. [340]. Unfortunately, many sources only mention the
group and do not use it explicitly. On the other hand, it is equally
difficult to speak about the affine group without a reference to
results in harmonic analysis: two theories are intimately
intertwined. In this section we collect fundamentals of the affine
group and its representations, which are not yet a standard background
of an analyst.
Let G=Aff be the ax+b (or the affine)
group [5]*§ 8.2, which is represented (as a
topological set) by the upper half-plane {(a,b) ∣
a∈ℝ+, b∈ℝ}. The group law is:
(a,b)· (a′,b′)=(aa′,ab′+b).
(7) |
As any other group, Aff has the left regular
representation
by shifts on functions Aff→ ℂ:
Λ(a,b): f(a′,b′) ↦ f(a,b)(a′,b′)=f | ⎛
⎜
⎜
⎝ | | , | | ⎞
⎟
⎟
⎠ | .
(8) |
A left invariant measure on Aff is dg=a−2 da db,
g=(a,b). By the definition, the left regular
representation (8) acts by unitary operators
on L2(Aff,dg). The group is not unimodular and
a right invariant measure is a−1 da db.
There are two important subgroups of the ax+b group:
A={(a,0)∈ Aff ∣ a∈ℝ+ }
and
N={(1,b)∈ Aff ∣ b∈ℝ }.
(9) |
An isometric representation of Aff on
Lp(ℝ) is given by the formula:
[p(a,b) f](x)= a | | f | ⎛
⎜
⎜
⎝ | | ⎞
⎟
⎟
⎠ | .
(10) |
Here, we identify the real line with the subgroup N or, even more
accurately, with the homogeneous space
Aff/N [92]*§ 2. This
representation is known as quasi-regular for its
similarity with (8). The action of the
subgroup N in (10) reduces to shifts,
the subgroup A acts by dilations.
Remark 2
The ax+
b group definitely escapes Occam’s Razor in
harmonic analysis, cf. the arguments against the imaginary unit in
the Introduction. Indeed, shifts are required to define convolutions
on ℝ
n, and an approximation of the identity
[314]*§ I.6.1 is a
convolution with the dilated kernel. The same scaled convolutions
define the fundamental maximal
functions
, see [314]*§ III.1.2
cf. Example 41 below. Thus, we can avoid
usage of the upper half-plane ℂ
+, but the same set will
anyway re-invent itself in the form of the ax+
b group.
The representation (10) in
L2(ℝ) is reducible and the space can be split
into irreducible subspaces. Following the philosophy presented in the
Introduction to the paper [203]*§ 1 we give the following
Definition 3
For a representation of a group G in a space V,
a generalized Hardy space
H is an
-irreducible (or -primary, as discussed in
Section 4.5) subspace of V.
Example 4
Let G=
Aff and the representation p be
defined in V=
Lp(ℝ)
by (10). Then the classical Hardy spaces
Hp(ℝ)
are p-irreducible, thus are covered
by the above definition.
Some ambiguity in picking the Hardy space out of all (well, two, as we
will see below) irreducible components is resolved by the traditional
preference.
Remark 5
We have defined the Hardy space completely in terms of
representation theory of ax+
b group. The traditional
descriptions, via the Fourier transform or analytic
extensions, will be corollaries in our approach, see
Prop. 7 and Example 7.
Remark 6
It is an interesting and important observation, that the Hardy space
in Lp(ℝ)
is invariant under the action of a
larger group SL2(ℝ)
, the group of 2× 2
matrices with real
entries and determinant equal to 1
, the group operation coincides
with the multiplication of matrices. The ax+
b group is
isomorphic to the subgroup of the upper-triangular matrices in
SL2(ℝ)
. The group SL2(ℝ)
has an isometric
representation in Lp(ℝ)
:
| | : f(x) ↦
| | f | ⎛
⎜
⎜
⎝ | | ⎞
⎟
⎟
⎠ | ,
(11) |
which produces quasi-regular
representation (10) by the restriction to
upper-triangular matrices. The Hardy space
Hp(ℝ)
is invariant under the above action
as well. Thus, SL2(ℝ)
produces a refined version in comparison with
the harmonic analysis of the ax+
b group considered in this
paper. Moreover, as representations of the ax+
b group are
connected with complex numbers, the structure of SL2(ℝ)
links
all three types of hypercomplex numbers [191]
[198]*§ 3.3.4 [197]*§ 3, see also
Rem. 17.
To clarify a decomposition of Lp(ℝ) into
irreducible subspaces of
representation (10) we need another
realization of this representation. It is called co-adjoint and is related to the
orbit method of Kirillov
[162]*§ 4.1.4 [105]*§ 6.7.1. Again,
this isometric representation can be defined on
Lp(ℝ) by the formula:
[[3]p(a,b) f](λ )= a | | e−2πi b
λ f(aλ ).
(12) |
Since a>0, there is an obvious decomposition into invariant subspaces of [3]p:
Lp(ℝ)=Lp(−∞,0)⊕Lp(0,∞).
(13) |
It is possible to demonstrate, that these components are irreducible.
This decomposition has a spatial nature, i.e., the subspaces have
disjoint supports. Each half-line can be identified with the subgroup
A or with the homogeneous space Aff/N.
The restrictions [3]+p and [3]−p of the
co-adjoint representation [3]p to invariant
subspaces (13) for p=2 are not unitary equivalent.
Any irreducible unitary representation of Aff is unitary
equivalent either to [3]+2 or [3]−2. Although
there is no intertwining operator between [3]+p and
[3]−p, the
map:
J: Lp(ℝ) →
Lp(ℝ): f(λ )↦ f(−λ ),
(14) |
has the property
[3]−p(a,−b)∘J=J∘[3]+p(a,b)
(15) |
which corresponds to the
outer automorphism (a,b)↦ (a,−b) of Aff.
As was already mentioned, for the Hilbert space L2(ℝ), representations (10)
and (12) are unitary equivalent, i.e., there is
a unitary intertwining operator between them. We may guess its
nature as follows. The eigenfunctions of the operators 2(1,b) are
e2πi ω x and the eigenfunctions of
[3]2(1,b) are δ(λ −ω ). Both sets form
“continuous bases” of L2(ℝ) and the
unitary operator which maps one to another is the Fourier transform:
F: f(x) ↦ f(λ)= | ∫ | |
e−2πi λ x f(x) dx.
(16) |
Although, the above arguments were informal, the intertwining property
F2(a,b)=[3]2(1,b)F can be directly
verified by the appropriate change of variables in the Fourier
transform. Thus, cf. [264]*Lem. A.6.2.2:
Proposition 7
The Fourier transform maps irreducible invariant subspaces
H2 and H2⊥
of (10) to irreducible invariant
subspaces L2(0,∞)=
F(
H2)
and
L2(−∞,0)=
F(
H2⊥)
of
co-adjoint representation (12). In
particular, L2(ℝ)=
H2⊕
H2⊥.
Reflection J (14) anticommutes with the Fourier
transform: F J=−J F. Thus, J also interchange
the irreducible components +p and −p of
quasi-regular representation (10) according
to (15).
Summing up, the unique rôle of the Fourier transform in harmonic
analysis is based on the following facts from the representation
theory. The Fourier transform
- intertwines shifts in quasi-regular
representation (10) to operators of
multiplication in co-adjoint
representation (12);
- intertwines dilations
in (10) to dilations
in (12);
- maps the decomposition
L2(ℝ)=H2⊕
H2⊥ into spatially separated spaces with
disjoint supports;
- anticommutes with J, which interchanges +2 and
−2.
Armed with this knowledge we are ready to proceed to harmonic analysis.
6.4 Covariant transform
We make an extension of the wavelet construction
defined in terms of group representations. See [159] for a
background in the representation theory, however, the only treated
case in this paper is the ax+b group.
Definition 8 [192, 197]
Let be a representation of
a group G in a space V and F be an operator acting from V to a space
U. We define a covariant transform
WF acting from V to the space L(
G,
U)
of
U-valued functions on G by the formula:
WF: v↦ v(g) = F((g−1) v),
v∈ V, g∈ G.
(17) |
The operator F will be called a fiducial operator
in this context (cf. the fiducial vector
in [220]).
We may drop the sup/subscripts from WF if the
functional F and/or the representation
are clear from the context.
Remark 9
We do not require that the fiducial operator F be linear.
Sometimes the positive homogeneity, i.e. F(
t v)=
tF(
v)
for
t>0
, alone can be already sufficient, see
Example 15.
Remark 10
It looks like the usefulness of the covariant transform is in the
reverse proportion to the dimension of the space U. The
covariant transform encodes properties of v in a function
WFv on G, which is a scalar-valued function if
dim
U=1
. However, such a simplicity is not always possible.
Moreover, the paper [201] gives an important example of a
covariant transform which provides a simplification even in the case
dim
U =dim
V.
We start the list of examples with the classical case of the group-theoretical
wavelet transform.
Example 11 [279, 95, 173, 5, 220, 95]
Let V be a Hilbert space with an inner product
⟨ ·,·
⟩
and be a unitary
representation of a group G in the space V. Let F:
V
→ ℂ
be the functional v↦
⟨
v,
v0
⟩
defined by a vector v0∈
V. The vector
v0 is often called the mother wavelet
in areas related
to signal processing, the vacuum state
in the quantum
framework, etc.In this set-up, transformation (1) is
the well-known expression for a wavelet transform [5]*(7.48) (or
representation coefficients):
W: v↦ v′(g) = ⟨ (g−1)v,v0
⟩ =
⟨ v,(g)v0
⟩,
v∈ V, g∈ G.
(18) |
The family of the vectors vg=(g)v0 is called
wavelets or coherent states. The image
of (3) consists of
scalar valued functions on G.
This scheme is typically carried out for a square integrable
representation with v0 being an
admissible vector [279, 95, 5, 106, 63, 87],
i.e. satisfying the condition:
0< | ⎪⎪
⎪⎪ | v′0 | ⎪⎪
⎪⎪ | 2= | ∫ | |
| ⎪
⎪ | ⟨ v0,2(g) v0
⟩ | ⎪
⎪ | 2 dg<∞.
(19) |
In this case the wavelet (covariant) transform is a map into the
square integrable functions [87] with respect to the left
Haar measure on G. The map becomes an isometry if v0 is
properly scaled. Moreover, we are able to recover the input v from its
wavelet transform through the reconstruction formula, which requires an
admissible vector as well, see Example 28 below.
The most popularized case of the above scheme is
provided by the affine group.
Example 12
For the ax+
b group,
representation (10) is square integrable for
p=2
. Any function v0, such that its Fourier transform
v
0(λ )
satisfies
is admissible in the sense of
(19) [5]*§ 12.2. The
continuous wavelet transform
is generated by
representation (10) acting on an
admissible vector v0 in
expression (3). The image of a function from
L2(ℝ)
is a function on the upper half-plane
square integrable with respect to the measure a−2 da db.
There are many examples [5]*§ 12.2 of useful
admissible vectors, say, the Mexican hat
wavelet: (1−
x2)
e−x2/2. For
sufficiently regular v
0
admissibility (5) of v0 follows by a
weaker condition
We dedicate
Section 6.8 to isometric properties of this
transform.
However, square integrable representations and admissible vectors do
not cover all interesting cases.
Example 13
For the above G=
Aff and
representation (10), we consider the operators
F±:
Lp(ℝ) → ℂ
defined by:
In L2(ℝ)
we note that
F+(
f)=⟨
f,
c
⟩
, where
c(
x)=1/π
i1/
i+
x. Computing the Fourier
transform ĉ(λ)=χ
(0,+∞)(λ)
e−λ , we see that
c ∈
H2(ℝ)
. Moreover, ĉ
does not satisfy admissibility
condition (5) for
representation (10).Then, covariant transform (1) is
Cauchy integral (1) from Lp(ℝ) to the
space of functions f′(a,b) such that
a−1/pf′(a,b) is in the Hardy space on the upper/lower half-plane
Hp(ℝ±2) [264]*§ A.6.3.
Due to inadmissibility of c(x), the
complex analysis become decoupled from the traditional wavelet
theory.
Many important objects in harmonic analysis are generated by
inadmissible mother wavelets like (7). For example,
the functionals P=1/2( F+ +F−) and Q=1/2i( F+
−F−) are defined by kernels:
which are Poisson kernel (2)
and the conjugate Poisson kernel [114]*§ 4.1
[108]*§ III.1 [231]*Ch. 5
[264]*§ A.5.3, respectively. Another interesting
non-admissible vector is the Gaussian
e−x2.
Example 14
A step in a different direction is a consideration of
non-linear operators. Take again the ax+
b group and its
representation (10).
We define F to be a homogeneous (but non-linear) functional
V→ ℝ
+:
Fm (f) = | | | | ⎪
⎪ | f(x) | ⎪
⎪ | dx.
(25) |
Covariant transform (1) becomes:
[Wpm f](a,b) = F(p( | | ) f)
= | | |
| ⎪
⎪
⎪ | a | | f | ⎛
⎝ | ax+b | ⎞
⎠ | ⎪
⎪
⎪ | dx
= | | |
| ⎪
⎪ | f | ⎛
⎝ | x | ⎞
⎠ | ⎪
⎪ | dx,
(26) |
where 1/
p+1/
q=1
, as usual.
We will see its connections with the Hardy–Littlewood maximal
functions in
Example 41.
Since linearity has clear advantages, we may prefer to reformulate the
last example using linear covariant transforms. The idea is similar
to the representation of a convex function as an envelope of linear
ones, cf. [108]*Ch. I, Lem. 6.1. To this end, we take a
collection F of linear fiducial functionals and, for a
given function f, consider the set of all covariant transforms
WF f, F∈F.
Example 15
Let us return to the setup of the previous Example for
G=
Aff and its
representation (10). Consider the unit ball B
in L∞[−1,1]
. Then, any ω∈
B defines a
bounded linear functional Fω on L1(ℝ)
:
Fω(f)= | | | f(x) ω(x) dx
= | | ∫ | | f(x) ω(x) dx.
(27) |
Of course, sup
ω∈ B Fω(
f)=
Fm(
f)
with Fm
from (11) and for all
f∈
L1(ℝ)
. Then, for the non-linear
covariant transform (12)
we have the following expression in terms of the linear covariant transforms
generated by Fω:
[W1m f](a,b) = | | [W1ω
f](a,b).
(28) |
The presence of suprimum is the price to pay for such a “linearization”.
Remark 16
The above construction is not much different to the
grand maximal function
[314]*§ III.1.2.
Although, it may look like a generalisation of covariant transform,
grand maximal function can be realised as a particular case of
Defn. 8. Indeed, let M(
V)
be a
subgroup of the group of all invertible isometries of a metric space
V. If represents a group G by isometries of V
then we can consider the group G′
generated by all finite products of
M(
V)
and (
g)
, g∈
G with the straightforward
action [2]
on V. The grand maximal functions
is produced by the covariant transform for the representation
[2]
of G′
.
Remark 17
It is instructive to compare action (11) of
the large SL2(ℝ)
group on the mother wavelet 1/
x+
i
for the Cauchy integral and the principal case
ω(
x)=χ
[−1,1](
x)
(the characteristic function of
[−1,1]
) for functional (27). The
wavelet 1/
x+
i is an eigenvector for all matrices
(
)
, which form the one-parameter compact subgroup
K⊂
SL2(ℝ)
. The respective covariant transform (i.e., the
Cauchy integral) maps functions to the homogeneous space SL2(ℝ)/
K,
which is the upper half-plane with the Möbius (linear-fractional)
transformations of complex numbers [191]
[198]*§ 3.3.4 [197]*§ 3. By contrast,
the mother wavelet χ
[−1,1] is an eigenvector for all
matrices (
)
, which form the one-parameter subgroup
A∈
SL2(ℝ)
. The covariant transform (i.e., the averaging) maps
functions to the homogeneous space SL2(ℝ)/
A, which can be
identified with a set of double numbers with corresponding Möbius
transformations [191] [198]*§ 3.3.4
[197]*§ 3. Conformal geometry of double numbers is
suitable for real variables technique, in particular,
tents [314]*§ II.2 make a Möbius-invariant
family.
6.5 The contravariant transform
Define the left action Λ of a group G on a space of
functions over G by:
Λ(g): f(h) ↦ f(g−1h).
(29) |
For example, in the case of the affine group it is (8).
An object invariant under the left action
Λ is called left invariant.
In particular, let L and L′ be two left invariant spaces of
functions on G. We say that a pairing ⟨ ·,·
⟩:
L× L′ → ℂ is left invariant if
⟨ Λ(g)f,Λ(g) f′
⟩= ⟨ f,f′
⟩, for all
f∈ L, f′∈ L′, g∈ G.
(30) |
Remark 18
-
We do not require the pairing to be linear in general, in some
cases it is sufficient to have only homogeneity, see
Example 35.
- If the pairing is invariant on space L× L′ it is not
necessarily invariant (or even defined) on large spaces of
functions.
- In some cases, an invariant pairing on G can be obtained
from an invariant functional l by the formula
⟨ f1,f2
⟩=l(f1 f2).
For a representation of G in V and w0∈ V,
we construct a function w(g)=(g)w0 on G. We assume
that the pairing can be extended in its second component to this
V-valued functions. For example, such an extension can be defined
in the weak sense.
Definition 19 [192, 197]
Let ⟨ ·,·
⟩
be a left invariant pairing on
L×
L′
as above, let be a representation of
G in a space V, we define the function
w(
g)=(
g)
w0 for w0∈
V such that w(
g)∈
L′
in
a suitable sense. The contravariant transform
Mw0 is a map L →
V defined
by the pairing:
Mw0: f ↦ ⟨ f,w
⟩,
where f∈ L.
(31) |
We can drop out sup/subscripts in Mw0 as we
did for WF.
Example 20 (Haar paring)
The most used example of an invariant pairing on
L2(
G,
dµ)×
L2(
G,
dµ)
is the integration with respect to the Haar measure:
⟨ f1,f2
⟩= | ∫ | | f1(g) f2(g) dg.
(32) |
If is a square integrable representation of G and w0 is an
admissible vector, see Example 7, then this
pairing can be extended to w(
g)=(
g)
w0. The
contravariant transform is known in this setup as the
reconstruction formula
, cf. [5]*(8.19):
Mw0 f = | ∫ | | f(g) w(g) dg,
where w(g)=(g) w0.
(33) |
It is possible to use different admissible vectors v0 and
w0 for wavelet
transform (3) and reconstruction
formula (19), respectively, cf.
Example 40.
Let either
- be not a square integrable representation (even modulo
a subgroup); or
- w0 be an inadmissible vector of a square integrable
representation .
A suitable invariant pairing in this case is not associated with
integration over the Haar measure on G. In this
case we speak about a Hardy pairing. The following example explains
the name.
Example 21 (Hardy pairing)
Let G be the ax+
b group and its representation
(10) in Example 8. An
invariant pairing on G, which is not generated by the Haar
measure a−2da db, is:
⟨ f1,f2
⟩H=
| | | |
f1(a,b) f2(a,b) | | .
(34) |
For this pairing, we can consider functions 1/π
i1/
x+
i or e−x2, which are not admissible vectors in the
sense of square integrable representations. For example, for
v0=1/π
i1/
x+
i we obtain:
In other words, it expresses the boundary values at a=0
of the Cauchy integral
[−
Cf](
x+
i a)
.
Here is an important example of non-linear pairing.
Example 22
Let G=
Aff and an invariant homogeneous functional
on G be given by the L∞-version of Haar
functional (18):
⟨ f1,f2
⟩∞= | | ⎪
⎪ | f1(g)f2(g) | ⎪
⎪ | .
(35) |
Define the following two functions on ℝ
:
v0+(t)= | ⎧
⎨
⎩ | |
| and
v0*(t)= | ⎧
⎪
⎨
⎪
⎩ | |
|
(36) |
The respective contravariant transforms are generated by
representation ∞
(10) are:
|
[ Mv0+f](t) | = | f+(t)=⟨ f(a,b),∞(a,b)
v0+(t)
⟩∞=
| | ⎪
⎪ | f(a,t) | ⎪
⎪ | , |
| (37) |
[ Mv0*f](t) | = | f*(t)=⟨ f(a,b),∞(a,b)
v0*(t)
⟩∞=
| | ⎪
⎪ | f(a,b) | ⎪
⎪ | .
|
| (38) |
|
Transforms (25)
and (26) are the vertical
and
non-tangential maximal
functions
[231]*§ VIII.C.2, respectively.
Example 23
Consider again G=
Aff equipped now with an
invariant linear functional, which is a Hardy-type modification
(cf. (21)) of
L∞-functional (23):
⟨ f1,f2
⟩ | | = | | | | (f1(a,b) f2(a,b)),
(39) |
where lim is the upper limit. Then, the covariant
transform MH for this pairing for functions v+ and
v* (24) becomes:
|
[ Mv0+Hf](t) | = | ⟨ f(a,b),∞(a,b)
v0+(t)
⟩ | | =
| | f(a,t), |
| (40) |
[ Mv0*Hf](t) | = | ⟨ f(a,b),∞(a,b)
v0*(t)
⟩ | | =
| | f(a,b).
|
| (41) |
|
They are the normal
and
non-tangential
upper limits from the upper-half plane to
the real line, respectively.
Note the obvious inequality ⟨ f1,f2
⟩∞ ≥
⟨ f1,f2
⟩∞H between
pairings (23) and (27),
which produces the corresponding relation between respective
contravariant transforms.
There is an explicit duality between the covariant transform and the
contravariant transform. Discussion of the grand maximal function in the
Rem. 17 shows usefulness of the covariant
transform over a family of fiducial functionals. Thus, we shall not
be surprised by the contravariant transform over a family of
reconstructing vectors as well.
Definition 24
Let w:
Aff →
L1(ℝ)
be a function. We define a new function 1
w on
Aff with values in L1(ℝ)
via the point-wise action [1
w](
g)=1(
g)
w(
g)
of
∞
(10). If
sup
g||
w(
g)||
1< ∞
, then, for
f∈
L1(
Aff)
, we define the extended
contravariant transform
by:
[Mw f](x)= | ∫ | | f(g) [1 w](g) dg.
(42) |
Note, that (30) reduces to the
contravariant transform (19) if we start from
the constant function w(g)=w0.
Definition 25
We call a function r on ℝ
a nucleus
if:
-
r is supported in [−1,1],
- | r |<1/2 almost everywhere, and
- ∫ℝ r(x) dx=0, cf. (6).
Clearly, for a nucleus r, the function s=1(a,b) r has
the following properties:
- s is supported in a ball centred at b and radius a,
- | s |<1/2a almost everywhere, and
- ∫ℝ s(x) dx=0.
In other words, s=1(a,b) r is an atom, cf. [314]*§ III.2.2 and any atom may be
obtained in this way from some nucleus and certain (a,b)∈Aff.
Example 26
Let f(
g)=∑
j λ
j δ
gj(
g)
with ∑
j
| λ
j |<∞
be a countable sum of point masses on
Aff. If all values of w(
gj)
are nucleuses,
then (30) becomes:
[Mw f](x)= | ∫ | | f(g) [1 w](g) dg
= | | λj sj,
(43) |
where sj=1(
gj)
w(
gj)
are atoms.
The right-hand side of (31) is known as an atomic
decomposition
of a function h(
x)=[
Mw
f](
x)
, see [314]*§ III.2.2.
6.6 Intertwining properties of covariant transforms
The covariant transform has obtained its name because of the following property.
Theorem 27 [192, 197]
Covariant transform (1)
intertwines and the left regular representation
Λ
(2) on L(
G,
U)
:
Corollary 28
The image space W(
V)
is invariant under the
left shifts on G.
The covariant transform is also a natural source of relative
convolutions [171, 204], which are
operators Ak=∫G k(g)(g) dg obtained by integration a
representation of a group G with a suitable kernel
k on G. In particular, inverse wavelet transform
Mw0 f (19) can be defined
from the relative convolution Af as well: Mw0 f= Af
w0.
Corollary 29
Covariant transform (1) intertwines
the operator of convolution K (with kernel k) and the operator
of relative convolution Ak, i.e. K W=
W Ak.
If the invariant pairing is defined by integration with respect to the Haar
measure, cf. Example 28, then we can show an
intertwining property for the contravariant transform as well.
Proposition 30 [173]*Prop. 2.9
Inverse wavelet transform
Mw0 (19) intertwines
left regular representation Λ
(2)
on L2(
G)
and :
Corollary 31
The image Mw0(
L(
G))⊂
V of a left
invariant space L(
G)
under the inverse wavelet
transform Mw0 is invariant under the representation
.
Remark 32
It is an important observation, that the above intertwining property
is also true for some contravariant transforms which are not based
on pairing (18). For example, in the case of the
affine group all pairings (21),
(27) and
(non-linear!) (23) satisfy
to (20) for the respective representation
p (10).
There is also a simple connection between a covariant transform and
right shifts.
Proposition 33 [193, 197]
Let G be a Lie group and be a representation of
G in a space V. Let [
Wf](
g)=
F((
g−1)
f)
be a
covariant transform defined by a fiducial operator F:
V →
U.
Then the right shift [
Wf](
gg′)
by g′
is the covariant transform
[
W′f](
g)=
F′((
g−1)
f)]
defined by the fiducial operator
F′=
F∘(
g−1)
. In other words the covariant transform intertwines right shifts R(g): f(h) ↦
f(hg) on the group G with the associated action
on fiducial operators:
R(g) ∘ WF=WB(g)F, g∈ G.
(47) |
Although the above result is obvious, its infinitesimal version has
interesting consequences. Let G be a Lie group with a Lie algebra
g and be a smooth representation of G.
We denote by dB the derived representation of the
associated representation B (46) on
fiducial operators.
Corollary 34 [193, 197]
Let a fiducial operator F be a null-solution, i.e. A F=0
,
for the operator A=∑
j aj dXjB, where
Xj∈
g and aj are constants. Then the covariant
transform [
WF f](
g)=
F((
g−1)
f)
for any f
satisfies
D (WF f)= 0, where
D= | | ājLXj.
|
Here, LXj are the left invariant fields (Lie derivatives) on
G corresponding to Xj.
Example 35
Consider representation (10) of
the ax+
b group with the p=1
. Let A and N be the
basis of g generating one-parameter subgroups A
and N (9), respectively. Then, the derived representations are:
[dA f](x)= −f(x)−xf′(x), [dNf](x)=−f′(x).
|
The corresponding left invariant vector fields on ax+
b group are:
The mother wavelet 1/
x+
i in (7) is a
null solution of the operator
Therefore, the image of the covariant transform with fiducial
operator F+ (7) consists of the null solutions
to the operator −
LA+
iLN=
i
a(∂
b+
i∂
a)
, that is in the essence
Cauchy–Riemann operator
∂
z (3) in the upper
half-plane.
Example 36
In the above setting, the function
p(
x)=1/π1/
x2+1
(23) is a
null solution of the operator:
(dA)2 − dA
+(dN)2
=2I+4x | | +(1+x2) | | .
|
The covariant transform with the mother wavelet p(
x)
is the
Poisson integral, its values are null solutions to the operator
(
LA)
2−
LA+(
LN)
2
=
a2(∂
b2+∂
a2)
, which is
Laplace operator Δ
(3).
Example 37
Fiducial functional Fm (11) is a null
solution of the following functional equation:
Consequently, the image of wavelet transform
Wpm (12) consists of functions which
solve the equation:
(I−R ( | | , | | )−R ( | | ,− | | ))f=0
or
f(a,b)=f( | | a, b+ | | a)+f( | | a, b− | | a).
|
The last relation is the key to the stopping time
argument [108]*Ch. VI, Lem. 2.2
and the dyadic squares technique, see for
example [314]*§ IV.3, [108]*Ch. VII, Thm. 1.1 or the picture on the
front cover of the latter book.
The moral of the above
Examples 7–10
is: there is a significant freedom in choice of covariant
transforms. However, some fiducial functionals have special properties,
which suggest the suitable technique (e.g., analytic, harmonic,
dyadic, etc.) following from this choice.
6.7 Composing the covariant and the contravariant transforms
From Props. 6, 30 and
Rem. 32 we deduce the following
Corollary 38
The composition Mw ∘
WF of a covariant
Mw and contravariant WF transforms is a
map V→
V, which commutes with , i.e.,
intertwines with itself.In particular for the affine group and
representation (10), Mw ∘
WF commutes with shifts and dilations of the real line.
Since the image space of Mw ∘ WF is an
Aff-invariant space, we shall be interested in the smallest
building blocks with the same property. For the Hilbert spaces, any
group invariant subspace V can be decomposed into a direct
integral V=⊕∫Vµ dµ of irreducible subspaces Vµ, i.e. Vµ
does not have any non-trivial invariant
subspace [159]*§ 8.4. For representations in Banach
spaces complete reducibility may not occur and we shall look for
primary subspace, i.e. space which is not a direct sum of two
invariant subspaces [159]*§ 8.3. We already identified
such subspaces as generalized Hardy spaces in Defn. 3. They
are also related to covariant functional
calculus [182] [197]*§ 6.
For irreducible Hardy spaces, we can use the following general
principle, which has several different formulations,
cf. [159]*Thm. 8.2.1:
Lemma 39 (Schur)
[5]*Lem. 4.3.1
Let be a continuous unitary irreducible representation
of G on the Hilbert space H. If a bounded operator T:
H
→
T commutes
with (
g)
, for all g ∈
G,
then T =
k I, for some λ ∈ ℂ
.
Remark 40
A revision of proofs of the Schur’s Lemma, even in different
formulations, show that the result is related to the existence of
joint invariant subspaces for all operators (
g)
, g∈
G.
In the case of classical wavelets,
the relation between wavelet transform (3)
and inverse wavelet transform (19) is
suggested by their names.
Example 41
For an irreducible square integrable representation and admissible
vectors v0 and w0, there is the
relation [5]*(8.52):
as an immediate consequence from the Schur’s lemma. Furthermore,
square integrability condition (19) ensures
that k≠ 0
. The exact value of the constant k depends on
v0, w0 and the Duflo–Moore
operator [87] [5]*§ 8.2.
It is of interest here, that two different vectors can be used as
analysing vector in (3) and for the
reconstructing formula (19). Even a
greater variety can be achieved if we use additional fiducial
operators and invariant pairings.
For the affine group, recall the decomposition from
Prop. 7 into invariant subspaces
L2(ℝ)=H2⊕
H2⊥ and the fact, that the restrictions
+2 and −2 of
2 (10) on H2 and
H2⊥ are not unitary equivalent. Then, Schur’s
lemma implies:
Corollary 42
Any bounded linear operator T:
L2(ℝ)→
L2(ℝ)
commuting with 2
has the form k1
IH2 ⊕
k2 IH2⊥ for some
constants k1, k2∈ℂ
. Consequently, the Fourier
transform maps T to the operator of multiplication by
k1χ
(0,+∞)+
k2χ
(−∞,0).
Of course, Corollary 42 is applicable to the
composition of covariant and contravariant transforms. In
particular, the constants k1 and k2 may have zero values:
for example, the zero value occurs for
W (3) with an admissible vector
v0 and non-tangential limit
Mv0*H (29)—because a square
integrable function f(a,b) on Aff vanishes for
a→ 0.
Example 43
The composition of contravariant transform
Mv0* (26) with
covariant transform W∞ (12) is:
|
[Mv0* W∞f](t) | = |
| | ⎧
⎪
⎨
⎪
⎩ | | | |
| ⎪
⎪ | f | ⎛
⎝ | x | ⎞
⎠ | ⎪
⎪ | dx | ⎫
⎪
⎬
⎪
⎭ |
| (50) |
| = |
| | | ⎧
⎪
⎨
⎪
⎩ | | | |
| ⎪
⎪ | f | ⎛
⎝ | x | ⎞
⎠ | ⎪
⎪ | dx | ⎫
⎪
⎬
⎪
⎭ | .
|
| |
|
Thus, Mv0* W∞f coincides with
Hardy–Littlewood maximal function
fM (32), which contains important
information on the original function f
[231]*§ VIII.B.1. Combining Props. 6
and 30 (through
Rem. 32), we deduce that the operator M:
f↦
fM commutes with p: pM=
M
p. Yet, M is non-linear and
Cor. 42 is not applicable in this case.
Example 44
Let the mother wavelet v0(
x)=δ(
x)
be the Dirac delta
function, then the wavelet transform Wδ generated
by ∞
(10) on
C(ℝ)
is [
Wδf](
a,
b)=
f(
b)
.
Take the reconstruction vector w0(
t)=(1−χ
[−1,1](
t))/
t/π
and
consider the respective inverse wavelet transform Mw0
produced by Hardy pairing (21). Then, the
composition of both maps is:
The last expression is the Hilbert transform
H=
Mw0∘
Wδ, which is an example of a singular integral
operator
(SIO) [314]*§ I.5
[257]*§ 2.6 defined through the principal
value (5) (in the sense of Cauchy). By
Cor. 42 we know that H=
k1
IH2 ⊕
k2 IH2⊥ for some
constants k1, k2∈ℂ
. Furthermore, we can
directly check that H J= −
JH , for the reflection J
from (14), thus k1=−
k2. An evaluation of
H on a simple function from H2 (say, the
Cauchy kernel 1/
x+
i) gives the value of the constant
k1=−
i. Thus, H=(−
i IH2) ⊕
(
i IH2⊥)
.
In fact, the previous reasons imply the following
Proposition 45
[313]*§ III.1.1 Any bounded linear operator on
L2(ℝ)
commuting with quasi-regular
representation 2
(10) and
anticommuting with reflection J (14) is a
constant multiple of Hilbert transform (51).
Example 46
Consider the covariant transform Wq defined by the inadmissible
wavelet q(
t)
(24), the conjugated
Poisson kernel. Its composition with the
contravariant transform Mv0+H (28)
is
[Mv0+H∘ Wq f](t)=
| |
| | ∫ | | | | dx
(52) |
We can see that this composition satisfies to
Prop. 45, the constant factor can again be
evaluated from the Cauchy kernel f(
x)=1/
x+
i and
is equal to 1
. Of course, this is a classical
result [114]*Thm. 4.1.5 in harmonic analysis
that (52) provides an alternative
expression for Hilbert transform (51).
Example 47
Let W be a covariant transfrom generated either by the
functional F± (7) (i.e. the Cauchy integral)
or 1/2 (
F+ −
F−)
(i.e. the Poisson integral) from the
Example 9. Then, for contravariant
transform Mv0+H (25)
the composition Mv0+H W becomes the normal
boundary value of the Cauchy/Poisson integral, respectively. The
similar composition Mv0*H W for
reconstructing vector v0* (24) turns to be
the non-tangential limit of the Cauchy/Poisson integrals.
The maximal function and SIO are often treated as elementary
building blocks of harmonic analysis. In particular, it is common to
define the Hardy space as a closed subspace of
Lp(ℝ) which is mapped to
Lp(ℝ) by either the maximal
operator (50) or by the
SIO (51) [314]*§ III.1.2
and § III.4.3 [89]. From this
perspective, the coincidence of both characterizations seems to be
non-trivial. On the contrast, we presented both the maximal operator
and SIO as compositions of certain co- and contravariant
transforms. Thus, these operators act between certain
Aff-invariant subspaces, which we associated with
generalized Hardy spaces in Defn. 3. For the
right choice of fiducial functionals, the coincidence of the
respective invariant subspaces is quite natural.
The potential of the group-theoretical approach is not limited to the
Hilbert space L2(ℝ). One of possibilities is
to look for a suitable modification of Schur’s Lemma 39,
say, to Banach spaces. However, we can proceed with the affine group
without such a generalisation. Here is an illustration to a classical
question of harmonic analysis: to identify the class of functions on
the real line such that Mv0*H W becomes the
identity operator on it.
Proposition 48
Let B
be the space of bounded uniformly continuous functions
on the real line. Let F:
B→ ℝ
be
a fiducial functional such that:
| | F(
∞(1/a,0) f )= 0, for all f∈
B such that f(0)=0
(53) |
and F(∞(1,
b)
f)
is a continuous function of
b∈ℝ
for a given f∈
B.Then, Mv0*H∘ WF is a constant multiple
of the identity operator on B.
Proof.
First of all we note that Mv0*H WF is a
bounded operator on B. Let
v(a,b)*=∞(a,b) v*. Obviously,
v(a,b)*(0)=v*(−b/a) is an eigenfunction for
operators Λ(a′,0), a′∈ℝ+ of the left
regular representation of Aff:
Λ(a′,0) v(a,b)*(0)= v(a,b)*(0).
(54) |
This and the left invariance of
pairing (30) imply that
Mv0*H∘ Λ (1/a,0)=Mv0*H for
any (a,0)∈Aff. Then, applying intertwining
properties (44) we obtain that
[Mv0*H ∘ WF f](0) | = | [Mv0*H∘ Λ (1/a,0)∘ WF f](0) |
| = | [Mv0*H∘ WF ∘ ∞(1/a,0)f](0).
|
|
Using the limit a→ 0 (33)
and the continuity of F∘ ∞(1,b) we conclude that
the linear functional l:f↦ [Mv0*H∘ WF
f](0) vanishes for any f∈B such that f(0)=0.
Take a function f1∈B such that f1(0)=1 and
define c=l(f1). From linearity of l, for any f∈
B we have:
l(f)=l(f−f(0)f1+f(0)f1)=l(f−f(0)f1)+f(0)l(f1)=cf(0).
|
Furthermore, using intertwining
properties (44) and (20):
[Mv0*H∘ WF
f](t) | = | [∞(1,−t) ∘ Mv0*H ∘ WF
f](0) |
| = | [Mv0*H ∘ WF ∘ ∞(1,−t) f](0) |
| = | l ( ∞(1,−t) f) |
| = | c[ ∞(1,−t) f](0) |
| = | cf(t).
|
|
This completes the proof.
To get the classical statement we need the following lemma.
Lemma 49
For a non-zero w(
t)∈
L1(ℝ)
, define
the fiducial functional on B:
F(f)= | ∫ | | f(t) w(t) dt.
(55) |
Then F satisfies the conditions (and thus the conclusions) of
Prop. 46.
Proof.
Let f be a continuous bounded function such that
f(0)=0. For ε>0 chose
- δ>0 such that | f(t) |<ε for
all | t |<δ;
- M>0 such that
∫| t |>M| w(t) | dt<ε.
Then, for a<δ/M, we have the estimation:
| = | | ⎪
⎪
⎪
⎪
⎪
⎪ | ∫ | | f | ⎛
⎝ | at | ⎞
⎠ | w(t) dt | ⎪
⎪
⎪
⎪
⎪
⎪ |
|
| ≤ | ⎪
⎪
⎪
⎪
⎪
⎪ | ∫ | | f | ⎛
⎝ | at | ⎞
⎠ | w(t) dt | ⎪
⎪
⎪
⎪
⎪
⎪ |
+ | ⎪
⎪
⎪
⎪
⎪
⎪ | ∫ | | f | ⎛
⎝ | at | ⎞
⎠ | w(t) dt | ⎪
⎪
⎪
⎪
⎪
⎪ |
|
| ≤ | ε ( | ⎪⎪
⎪⎪ | w | ⎪⎪
⎪⎪ | 1 + | ⎪⎪
⎪⎪ | f | ⎪⎪
⎪⎪ | ∞) .
|
|
|
Finally, for a uniformly continuous function g for
ε>0 there is δ>0 such that
| g(t+b)−g(t) |<ε for all b<δ and
t∈ℝ. Then
| ⎪
⎪ | F(∞(1,b) g )−F(g) | ⎪
⎪ | = | ⎪
⎪
⎪
⎪
⎪
⎪ | ∫ | |
(g(t+b)−g(t)) w(t) dt | ⎪
⎪
⎪
⎪
⎪
⎪ | ≤ ε | ⎪⎪
⎪⎪ | w | ⎪⎪
⎪⎪ | 1.
|
This proves the continuity of F(∞(1,b) g ) at
b=0 and, by the group property, at any other point as well.
Remark 50
A direct evaluation shows, that the constant c=
l(
f1)
from the
proof of Prop. 46 for fiducial
functional (35) is equal to
c=∫
ℝ w(
t)
dt. Of course, for non-trivial
boundary values we need c≠ 0
. On the other hand,
admissibility condition (6) requires
c=0
. Moreover, the classical harmonic analysis and the
traditional wavelet construction are two “orthogonal” parts of the
same covariant transform theory in the following sense. We can
present a rather general bounded function w=
wa+
wp as a sum of
an admissible mother wavelet wa and a suitable multiple wp
of the Poisson kernel. An extension of this technique to unbounded
functions leads to Calderón–Zygmund decomposition
[314]*§ I.4.
The table integral ∫ℝ dx/x2+1=π tells
that the “wavelet”
p(t)=1/π1/1+t2 (23) is in
L1(ℝ) with c=1, the corresponding wavelet
transform is the Poisson integral. Its boundary behaviour from
Prop. 46 is the classical result,
cf. [108]*Ch. I, Cor. 3.2. The comparison of our
arguments with the traditional proofs, e.g. in [108],
does not reveal any significant distinctions. We simply made an
explicit usage of the relevant group structure, which is implicitly
employed in traditional texts anyway, cf. [54]. Further
demonstrations of this type can be found
in [4, 92].
6.8 Transported norms
If the functional F and the representation
in (1) are both linear, then the resulting
covariant transform is a linear map. If WF is injective,
e.g. due to irreducibility of ,
then WF transports a norm ||·|| defined on
V to a norm ||·||F defined on the image space
WF V by the simple rule:
| ⎪⎪
⎪⎪ | u | ⎪⎪
⎪⎪ | F:= | ⎪⎪
⎪⎪ | v | ⎪⎪
⎪⎪ | ,
where the unique
v∈ V is defined by u=WF v.
(56) |
By the very definition, we have the following
Proposition 51
-
WF is an isometry (V,||·||)→
(WF V, ||·||F).
- If the representation acts on
(V,||·||) by isometries then ||·||F is
left invariant.
A touch of non-triviality occurs if the transported norm can be
naturally expressed in the original terms of G.
Example 52
It is common to consider a unitary square integrable representation and an
admissible mother wavelet f∈
V. In this case,
wavelet transform (3) becomes an isometry to
square integrable functions on G with respect to a Haar
measure [5]*Thm. 8.1.3. In particular, for the
affine group and setup of Example 8, the wavelet
transform with an admissible vector is a multiple of an isometry map
from L2(ℝ)
to the functions on the upper
half-plane, i.e., the ax+
b group, which are square integrable
with respect to the Haar measure a−2 da db.
A reader expects that there are other interesting examples of the
transported norms, which are not connected to the Haar integration.
Example 53
In the setup of Example 9, consider the space
Lp(ℝ)
with
representation (10) of Aff and
Poisson kernel p(
t)
(23) as an inadmissible
mother wavelet. The norm transported by WP to the image
space on Aff is [264]*§ A.6.3:
| ⎪⎪
⎪⎪ | u | ⎪⎪
⎪⎪ | p= | | ⎛
⎜
⎜
⎝ | | ⎪
⎪ | u(a,b) | ⎪
⎪ | p | | ⎞
⎟
⎟
⎠ | | .
(57) |
In the theory of Hardy spaces, the Lp-norm on the
real line and transported norm (57) are
naturally intertwined, cf. [264]*Thm. A.3.4.1(iii), and
are used interchangeably.
The second possibility to transport a norm from V to a function
space on G uses an contravariant transform Mv:
| ⎪⎪
⎪⎪ | u | ⎪⎪
⎪⎪ | v:= | ⎪⎪
⎪⎪ | Mv u | ⎪⎪
⎪⎪ | .
(58) |
Proposition 54
-
The contravariant transform Mv is an isometry
(L,||·||v)→ (V,||·||).
- If the composition Mv ∘ WF=c I is a
multiple of the identity on V then transported norms
||·||v (58) and
||·||F (56) differ only by a
constant multiplier.
The above result is well-known for traditional wavelets.
Example 55
In the setup of Example 40, for a square integrable
representation and two admissible mother wavelets v0 and
w0 we know that Mw0Wv0=
k
I (49), thus transported norms
(56) and (58)
differ by a constant multiplier. Thus,
norm (58) is also provided by the
integration with respect to the Haar measure on G.
In the theory of Hardy spaces the result is also classical.
Example 56
For the fiducial functional F with
property (33) and the contravariant
transform Mv0*H (29),
Prop. 46 implies Mv0*H∘
WF=
c I. Thus, the norm transported to Aff by
Mv0*H from Lp(ℝ)
up to
factor coincides with (57). In other words, the
transition to the boundary limit on the Hardy space is an isometric
operator. This is again a classical result of the harmonic analysis,
cf. [264]*Thm. A.3.4.1(ii).
The co- and contravariant transforms can be used to transport norms in
the opposite direction: from a classical space of functions on G
to a representation space V.
Example 57
Let V be the space of σ
-finite signed measures of a
bounded variation on the upper half-plane. Let the ax+
b group
acts on V by the representation adjoint to
[1(
a,
b)
f](
x,
y)=
a−1f(
x−
b/
a,
y/
a)
on
L2(ℝ
+2)
,
cf. (8). If the mother wavelet v0 is
the indicator function of the square { 0<
x<1, 0<
y<1}
, then the
covariant transform of a measure µ
is
µ′(
a,
b)=
a−1µ(
Qa,b)
, where Qa,b is the
square {
b<
x<
b+
a, 0<
y<
a}
. If we request that
µ′(
a,
b)
is a bounded function on the affine group, then
µ
is a Carleson measure [108]*§ I.5. A norm
transported from L∞(
Aff)
to the
appropriate subset of V becomes the Carleson norm of measures. Indicator function of a
tent taken as a mother wavelet will lead to an
equivalent definition.
It was already mentioned in Rem. 17 and
Example 39 that we may be interested to
mix several different covariant and contravariant transforms. This
motivate the following statement.
Proposition 58
Let (
V,||·||)
be a normed space and be a
continuous representation of a topological locally compact group
G on V. Let two fiducial operators F1 and F2
define the respective covariant transforms W1 and
W2 to the same image space W=
W1 V=
W2
V. Assume, there exists an contravariant transform
M:
W →
V such that M∘
W1=
c1 I and M∘
W2=
c2 I. Define by
||·||
M the norm on U transpordef from V
by M. Then
| ⎪⎪
⎪⎪ | W1 v1+ W2 v2 | ⎪⎪
⎪⎪ | M= | ⎪⎪
⎪⎪ | c1 v1 +c2
v2 | ⎪⎪
⎪⎪ | , for any v1, v2∈ V.
(59) |
Proof.
Indeed:
| | =
| ⎪⎪
⎪⎪ | M∘W1 v1+ M∘W2
v2 | ⎪⎪
⎪⎪ |
|
| =
| ⎪⎪
⎪⎪ | c1 v1+ c2 v2 | ⎪⎪
⎪⎪ | ,
|
|
|
by the definition of transported
norm (58) and the assumptions
M∘Wi=ci I.
Although the above result is simple, it does have important consequences.
Corollary 59 (Orthogonality Relation)
Let be a square integrable representation of a group
G in a Hilbert space V. Then, for any two admissible mother
wavelets f and f′
there exists a constant c such that:
| ∫ | | ⟨ v,(g)f
⟩ | | dg
=c ⟨ v,v′
⟩ for any v1,v2∈ V.
(60) |
Moreover, the constant c=
c(
f′,
f)
is a sesquilinear form of
vectors f′
and f.
Proof.
We can derive (60)
from (59) as follows. Let Mf be
the inverse wavelet transform (19)
defined by the admissible vector f, then Mf∘
Wf=I on V providing the right scaling of f.
Furthermore, Mf∘ Wf′=cI
by (49) for some complex constant
c. Thus, by (59):
| ⎪⎪
⎪⎪ | Wf v +Wf′v′ | ⎪⎪
⎪⎪ | M= | ⎪⎪
⎪⎪ | v+cv′ | ⎪⎪
⎪⎪ | .
|
Now, through the polarisation identity [164]*Problem 476
we get the equality (60) of inner products.
The above result is known as the orthogonality relation in the theory of wavelets, for some
further properties of the constant c
see [5]*Thm. 8.2.1.
Here is an application of Prop. 58 to harmonic
analysis, cf. [114]*Thm. 4.1.7:
Corollary 60
The covariant transform Wq with conjugate Poisson
kernel q (24) is a
bounded map from (
L2(ℝ),||·||)
to
(
L(
Aff), ||·||
2)
with norm
||·||
2 (57). Moreover:
| ⎪⎪
⎪⎪ | Wq f | ⎪⎪
⎪⎪ | 2= | ⎪⎪
⎪⎪ | f | ⎪⎪
⎪⎪ | , for all f ∈ L2(ℝ).
|
Proof.
As we establish in Example 46 for
contravariant transform
Mv0+H (28),
Mv0+H ∘ Wq=−i I and i I on
H2 and H2⊥, respectively.
Take the unique presentation f=u+u⊥, for u∈
H2 and u⊥∈ H2⊥. Then,
by (59)
| ⎪⎪
⎪⎪ | Wq f | ⎪⎪
⎪⎪ | 2= | ⎪⎪
⎪⎪ | −i u+i u ⊥ | ⎪⎪
⎪⎪ | = | ⎪⎪
⎪⎪ | u+u⊥ | ⎪⎪
⎪⎪ | = | ⎪⎪
⎪⎪ | f | ⎪⎪
⎪⎪ | .
|
This completes the proof.
90
Covariant scheme
|
Complex variable
|
Real variable
|
Covariant transform is WF:
v↦ v(g) = F((g−1) v). In particular, the wavelet transform for the mother wavelet v0 is
v′(g) = ⟨ v,(g)v0
⟩. | The Cauchy
integral is generated by the mother wavelet
1/2πi1/x+i. The Poisson integral is generated by the mother wavelet
1/π1/x2+1 | The averaging operator
f′(b)=1/2a∫b−ab+a f(t) dt is
defined by the mother wavelet χ[−1,1](t), to average the modulus of f(t) we use all elements of the unit
ball in
L∞[−1,1]. |
The covariant transform maps vectors to functions on G
or, in the induced case, to functions on the homogeneous space
G/H. | Functions are mapped from the real line to the upper
half-plane parametrised by either the ax+b-group or the
homogeneous space SL2(ℝ)/K. | Functions are mapped from the real
line to the upper half-plane parametrised by either the
ax+b-group or the homogeneous
space SL2(ℝ)/A. |
Annihilating action on the mother wavelet produces
functional relation on the image of the covariant transform | The
operator −dA −i
dN=I+(x+i)d/dx annihilates the mother
wavelet 1/2πi1/x+i, thus the image of
wavelet transform is in the kernel of the Cauchy–Riemann operator
−LA+iLN=i
a(∂b+i∂a). Similarly, for the Laplace
operator. | The mother wavelet v0=χ[−1,1] satisfies the equality
χ[−1,1]=χ[−1,0]+χ[0,1], where both terms are
again scaled and shifted v0. The image of the wavelet
transform is suitable for the stopping time argument and the dyadic squares technique.
|
An invariant pairing
⟨ ·,·
⟩ generates the contravariant transform [Mw0 f]
⟨ f(g),(g)w0
⟩ for | The contravariant transform with the
invariant Hardy pairing on the ax+b group produces boundary
values of functions on the real line. | The covariant transform with the invariant sup pairing
produces the vertical and non-tangential maximal functions.
|
The composition Mv ∘ WF of the
covariant and contravariant transforms is a multiple of the
identity on irreducible components. | SIO is a composition of the Cauchy
integral and its boundary value. | The Hardy–Littlewood maximal function is the composition of the
averaging operator and the contravariant transform from the invariant
sup pairing.
|
The Hardy space is an invariant subspace of the group
representation. | The Hardy space consists of the limiting values
of the Cauchy integral. SIO is bounded on this space. | The
Hardy–Littlewood maximal operator is bounded on
the Hardy space Hp .
|
Figure 6.1: The correspondence between different elements of
harmonic analysis. |
Last modified: October 28, 2024.