This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.Lecture 5 Analytic Functions
We saw in the first section that an inspiring geometry of cycles
can be recovered from the properties of SL2(ℝ).
In this section we consider a realisation of the function theory within
Erlangen approach [170, 172, 175, 176]. The
covariant transform will be our principal tool in this construction.
5.1 Induced Covariant Transform
The choice of a mother wavelet or fiducial operator F from
Section 4.1 can significantly influence the
behaviour of the covariant transform. Let G be a group and
H be its closed subgroup with the corresponding homogeneous
space X=G/H. Let be a representation of G by
operators on a space V, we denote by H the restriction
of to the subgroup H.
Definition 1
Let χ
be a representation of the subgroup H in a space U and
F:
V→
U be an intertwining operator between χ
and the representation H:
F((h) v)=F(v)χ(h), for all h∈ H,
v∈ V.
(1) |
Then the covariant transform (1) generated
by F is called the induced covariant transform
.
The following is the main motivating example.
Example 2
Consider the traditional wavelet transform as outlined in
Ex. 7. Chose a vacuum vector v0 to be a joint
eigenvector for all operators (
h)
, h∈
H, that is
(
h)
v0=χ(
h)
v0, where χ(
h)
is a complex number
depending of h. Then χ
is obviously a character of H.The image of wavelet transform (3) with such a
mother wavelet will have a property:
v(gh) = ⟨ v,(gh)v0
⟩
= ⟨ v,(g)χ(h)v0
⟩
=χ(h)v(g).
|
Thus the wavelet transform is uniquely defined by cosets on the
homogeneous space G/H. In this case we previously spoke about the
reduced wavelet transform [172].
A representation 0 is called square integrable
mod H if the induced wavelet transform [Wf0](w) of
the vacuum vector f0(x) is square integrable on X.
The image of induced covariant transform have the similar property:
v(gh)=F(((gh)−1)
v)=F((h−1)(g−1) v)
=F((g−1) v)χ(h−1).
(2) |
Thus it is enough to know the value of the covariant transform only at a
single element in every coset G/H in order to reconstruct it for
the entire group G by the representation χ. Since coherent
states (wavelets) are now parametrised by points
homogeneous space G/H they are referred sometimes as coherent
states which are not connected to a group [219], however
this is true only in a very narrow sense as explained above.
Example 3
To make it more specific we can consider the representation of SL2(ℝ)
defined on L2(ℝ)
by the formula, cf. (8):
(g): f(z) ↦ | |
f | ⎛
⎜
⎜
⎝ | | ⎞
⎟
⎟
⎠ | , g−1=
| | .
|
Let K⊂
SL2(ℝ)
be the compact subgroup of matrices
ht=
(
)
. Then for the fiducial operator
F± (7) we have
F±∘(
ht)=
e∓i tF±. Thus we can
consider the covariant transform only for points in the homogeneous
space SL2(ℝ)/
K, moreover this set can be naturally identified with
the ax+
b group. Thus we do not obtain any advantage of
extending the group in Ex. 8 from ax+
b to SL2(ℝ)
if
we will be still using the fiducial operator
F± (7).
Functions on the group G, which have the property
v(gh)=v(g)χ(h) (2),
provide a space for the representation of G induced by the
representation χ of the subgroup H. This explains the
choice of the name for induced covariant transform.
Remark 4
Induced covariant transform uses the fiducial operator F which
passes through the action of the subgroup H. This reduces
information which we obtained from this transform in some cases.
There is also a simple connection between a covariant transform and
right shifts:
Proposition 5
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 the 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 on
the group G with the associated action
B (14) on fiducial operators.
Although the above result is obvious, its infinitesimal version has
interesting consequences.
Corollary 6 ([193])
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 (14) on fiducial
operators.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 [W f](g)=F((g−1)f) for any f
satisfies:
D F(g)= 0, where
D= | | ājLXj.
|
Here LXj are the left invariant fields (Lie derivatives) on
G corresponding to Xj.
Example 7
Consider the representation (7) of
the ax+
b group with the p=1
. Let A and N be the
basis of g generating one-parameter subgroups
(
et,0)
and (0,
t)
, 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 the fiducial operator
F+ (7) consists of the null solutions to
the operator −
LA+
iLN=
i a(∂
b+
i∂
a)
,
that is in the essence the Cauchy–Riemann operator in the upper
half-plane.
Example 8
In the above setting, the function p(
x)=1/π1/
x2+1
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)
—the
Laplace operator.
Example 9
Using the setup of the previous Examples, we observe that the
Gaussian e−x2/2 is the null solution of the
operator dA + (
dN)
2=
I+
xd/
dx+
d2/
dx2. Therefore, the covariant transform with the
Gaussian as a mother wavelet consists of the null solutions to the
operator LA+(
LN)
2=
a(∂
a+
a∂
b2)
,
that is a parabolic equation related to the heat
equation (12). The second null-solution to the operator
dA + (
dN)
2 is erf(
i
x)
e−x2/2, where the error function erf satisfies
to the identity erf′(
x)=
e−x2/2. For the normalised
mother wavelet, the corresponding
wavelet transform is:
| [Wφ](a,b) | = | | |
| = | | ∫ | | φ(y) exp | ⎛
⎜
⎜
⎝ | − | | ⎞
⎟
⎟
⎠ | dy.
|
| (3) |
|
It consists of null-solutions [
Wφ](
a,
b)
to the operator
∂
a+
a∂
b2 as mentioned above. To obtain solutions
f(
x,
t)
of the heat equation ∂
t
−
k∂
x2 (12) we can use the change of
variables: f(
x,
t)=[
Wφ](2√
kt,
x)
. Then the
wavelet transform (3) mutates
to the integral formula (13).
Example 10
The 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 dyadic cubes technique, see for
example [108]*Ch. VII, Thm. 1.1 or the picture on the
front cover of this book.
The moral of the above
Examples 7–10
is: there is a significant freedom in choice of the covariant
transforms. However, some fiducial functionals has special properties,
which suggest the suitable technique (e.g., analytic, harmonic,
dyadic, etc.) following from this selection.
There is a statement which extends Corollary 6 from
differential operators to integro-differential ones. We will formulate
it for the wavelets setting.
Corollary 11
Let G be a group and be a unitary representation
of G, which can be extended to a vector space V of functions
or distributions on G.
Let a mother wavelet w∈
V′
satisfy the equation
for a fixed distribution a(
g)
and a (not necessarily
invariant) measure dg. Then any wavelet transform
F(
g)=
W f(
g)=⟨
f,(
g)
w
⟩
obeys the condition:
DF=0, where D= | ∫ | | ā(g) R(g) dg,
|
with R being the right regular representation of G.
Clearly, the Corollary 6 is a particular case of
the Corollary 11 with a distribution a,
which is a combination of derivatives of Dirac’s delta functions. The
last Corollary will be illustrated at the end of
Section 1.2.
Remark 12
We note that Corollaries 6
and 11 are true whenever we have an
intertwining property between with the right regular
representation of G.
5.2 Induced Wavelet Transform and Cauchy Integral
We again use the general scheme from
Subsection 3.2. The ax+b group is
isomorphic to a subgroups of SL2(ℝ) consisting of the
lower-triangular matrices:
F= | ⎧
⎪
⎪
⎨
⎪
⎪
⎩ | | | , a>0 | ⎫
⎪
⎪
⎬
⎪
⎪
⎭ | .
|
The corresponding homogeneous space X=SL2(ℝ)/F is one-dimensional and
can be parametrised by a real number. The natural projection
p:SL2(ℝ)→ ℝ and its
left inverse s: ℝ→ SL2(ℝ) can
be defined as follows:
Thus we calculate the corresponding map r: SL2(ℝ)→ F, see
Subsection ??:
Therefore the action of SL2(ℝ) on the real line is exactly the Möbius
map (1):
g:u↦ p(g−1*s(u)) = | | , where
g−1= | | .
|
We also calculate that
To build an induced representation we need a character of the affine group.
A generic character of F is a power of its diagonal element:
Thus the corresponding realisation of induced
representation (5) is:
κ (g): f(u) ↦ | |
f | ⎛
⎜
⎜
⎝ | | ⎞
⎟
⎟
⎠ |
where g−1= | | .
(6) |
The only freedom remaining by the scheme is in a choice of a
value of number κ and the corresponding functional
space where our representation acts. At this point we have a wider
choice of κ than it is usually assumed: it can belong to
different hypercomplex systems.
One of the important properties which would be nice to have is the
unitarity of the representation (6) with respect
to the standard inner product:
⟨ f1,f2
⟩= | ∫ | | f1(u)f2(u) du.
|
A change of variables x=au+b/cu+d in the integral suggests
the following property is necessary and sufficient for that:
A mother wavelet for an induced wavelet transform shall be an
eigenvector for the action of a subgroup H′ of SL2(ℝ),
see (1). Let us consider the most common
case of H′=K and take the infinitesimal condition with the
derived representation: dZnw0 =λ w0, since
Z (13) is the generator of the subgroup K. In
other word the restriction of w0 to a K-orbit should be given
by eλ t in the exponential coordinate t along the
K-orbit. However we usually need its expression in other “more
natural” coordinates. For example [195], an eigenvector of
the derived representation of dZn should satisfy the
differential equation in the ordinary parameter x∈ℝ:
−κ xf(x)−f′(x)(1+x2)=λ f(x).
(8) |
The equation does not have singular points, the general solution
is globally defined (up to a constant factor) by:
wλ, κ (x)= | | ⎛
⎜
⎜
⎝ | | ⎞
⎟
⎟
⎠ | |
= | (x−i)(iλ−κ )/2
|
|
(x+i)(iλ+κ )/2 |
| .
(9) |
To avoid multivalent functions we need 2π-periodicity along the
exponential coordinate on K. This implies that the parameter
m=−iλ is an integer.
Therefore the solution becomes:
wm,κ (x) = | (x+i)(m−κ )/2 |
|
(x−i)(m+κ )/2 |
| .
(10) |
The corresponding wavelets resemble the Cauchy kernel normalised to the
invariant metric in the Lobachevsky half-plane:
wm,κ (u,v;x) | = | Fκ (s(u,v)) wm,κ (x)
=
vκ/ 2 | |
|
|
Therefore the wavelet transform (3) from
function on the real line to functions on the upper half-plane is:
f(u,v) | = | ⟨ f,Fκ (u,v)wm,κ
⟩
=vκ /2 | ∫ | | f(x) | (x−(u+i v))(m−κ )/2 |
|
(x−(u−i v))(m+κ)/2 |
| dx.
|
|
|
Introduction of a complex variable z=u+i v allows to write it
as:
f(z)=(ℑ z)κ/2 | ∫ | | f(x)
| (x−z)(m−κ)/2 |
|
(x−z)(m+κ)/2 |
| dx.
(11) |
According to the general theory this wavelet transform intertwines
representations Fκ (6) on the
real line (induced by the character aκ of the subgroup
F) and Km (8) on the upper half-plane
(induced by the character ei m t of the subgroup K).
5.3 The Cauchy-Riemann (Dirac) and Laplace Operators
Ladder operators L±=±i A +B act by raising/lowering indexes of the
K-eigenfunctions wm,κ (9), see
Subsection 3.3. More explicitly [195]:
dL±κ : wm,κ
↦ − | | ( m ± κ) wm± 2,κ.
(12) |
There are two possibilities here: m±κ is zero for some
m or not. In the first case the chain (12)
of eigenfunction wm,κ terminates on one side under the
transitive action (16) of the ladder operators;
otherwise the chain is infinite in both directions. That is, the values
m=∓κ and only those correspond to the maximal (minimal)
weight function w∓κ ,κ (x)=1/(x±i)κ ∈ L2(ℝ), which are annihilated by
L±:
|
dL±κ w∓κ ,κ = (±i dAκ
+dBκ) w∓κ ,κ =0.
| | | (13) |
|
By the Cor. 6 for the mother
wavelets w∓κ ,κ , which are annihilated
by (13), the images of the respective wavelet
transforms are null solutions to the left-invariant differential
operator D±=LL±:
D±=∓iLA+LB= − | | +v(∂u±i∂v).
(14) |
This is a conformal version of the Cauchy–Riemann equation. The
second order conformal Laplace-type operators
Δ+=LL−LL+ and
Δ−=LL+LL− are:
Δ±= (v∂u− | | )2+v2∂v2
± | | .
(15) |
For the mother wavelets wm,κ in (13)
such that m=∓κ the unitarity condition
κ+κ=2, see (7), together
with m∈ℤ implies κ=∓ m=1. In such a case
the wavelet transforms (11) are:
f+(z)=(ℑ z) | | | ∫ | |
| | and
f−(z)=(ℑ z) | | | ∫ | |
| | ,
(16) |
for w−1,1 and w1,1 respectively. The first one is the Cauchy integral
formula up to the factor 2πi √ℑ z. Clearly, one integral is
the complex conjugation of another. Moreover, the minimal/maximal
weight cases can be intertwined by the following automorphism of the
Lie algebra sl2:
As explained before
f±(w) are null solutions to the operators
D± (14) and
Δ± (15). These transformations
intertwine unitary equivalent representations on the real line and
on the upper half-plane, thus they can be made unitary for proper
spaces. This is the source of two faces of the Hardy spaces: they can
be defined either as square-integrable on the real line with an analytic extension to
the half-plane, or analytic on the half-plane with
square-integrability on an infinitesimal displacement of the real line.
For the third possibility, m±κ≠ 0, there is no an
operator spanned by the derived representation of the Lie algebra
sl2 which kills the mother wavelet wm,κ.
However the remarkable Casimir operator
C=Z2−2(L−L++L+L−), which spans the
centre of the universal enveloping algebra of sl2
[321]*§ 8.1 [240]*§ X.1,
produces a second order operator which does the job. Indeed from the
identities (12) we get:
dCκ wm,κ = ( 2κ − κ2) wm,κ.
(17) |
Thus we get dCκ wm,κ =0 for κ=2 or 0.
The mother wavelet w0,2 turns to be the Poisson
kernel [114]*Ex. 1.2.17.
The associated wavelet transform
consists of null solutions of the left-invariant second-order
Laplacian, image of the Casimir operator,
cf. (15):
Another integral formula producing solutions to this equation
delivered by the mother wavelet wm,0 with the value κ=0
in (17):
f(z)= | ∫ | | f(x)
| ⎛
⎜
⎜
⎝ | | ⎞
⎟
⎟
⎠ | | dx.
(19) |
Furthermore, we can introduce higher order differential operators.
The functions w∓ 2m+1,1 are annihilated by n-th power of
operator dL±κ with 1≤ m≤ n. By
the Cor. 6 the the image of wavelet
transform (11) from a mother wavelet
∑1n am w∓ 2m,1 will consist of null-solutions of the
n-th power D±n of the conformal Cauchy–Riemann operator (14).
They are a conformal flavour of polyanalytic
functions [16].
We can similarly look for mother wavelets which are eigenvectors for other
types of one dimensional subgroups. Our consideration of subgroup
K is simplified by several facts:
- The parameter κ takes only complex values.
- The derived representation does not have singular points on the real line.
For both subgroups and N′ this will not be true. The
further consideration will be given in [195].
5.4 The Taylor Expansion
Consider an induced wavelet transform generated by a Lie group G,
its representation and a mother wavelet w which is an
eigenvector of a one-dimensional subgroup H′⊂ G. Then
by Prop. 5 the wavelet transform
intertwines with a representation H′
induced by a character of H′.
If the mother wavelet is itself in the domain of the induced wavelet
transform then the chain (16) of
H′-eigenvectors wm will be mapped to the similar chain
of their images ŵm. The corresponding derived induced
representation dH′ produces ladder operators with
the transitive action of the ladder operators on the chain of
ŵm. Then the vector space of “formal power series”:
is a module for the Lie algebra of the group G.
Coming back to the case of the group G=SL2(ℝ) and subgroup
H′=K. Images ŵm,1 of the
eigenfunctions (10) under the Cauchy integral
transform (16) are:
ŵm,1(z)=(ℑ z)1/2 | (z+i)(m−1)/2 |
|
(z−i)(m+1)/2 |
| .
|
They are eigenfunctions of the derived representation on the upper
half-plane and the action of ladder operators is
given by the same expressions (12).
In particular, the sl2-module generated by
ŵ1,1 will be one-sided since this vector is annihilated by
the lowering operator.
Since the Cauchy integral produces an unitary intertwining
operator between two representations we get the following variant of
Taylor series:
f(z)= | | cm ŵm,1(z),
where cm=⟨ f,wm,1
⟩.
|
For two other types of subgroups, representations and mother wavelets this
scheme shall be suitably adapted and detailed study will be presented
elsewhere [195].
5.5 Wavelet Transform in the Unit Disk and Other Domains
We can similarly construct an analytic function theories in unit
disks, including parabolic and hyperbolic ones [191]. This
can be done simply by an application of the Cayley transform to the
function theories in the upper half-plane. Alternatively we can apply
the full procedure for properly chosen groups and subgroups. We will
briefly outline such a possibility here, see also [170].
Elements of SL2(ℝ) can be also represented by 2× 2-matrices
with complex entries such that, cf. Example 24:
g= | | ,
g−1= | | ,
| ⎪
⎪ | α | ⎪
⎪ | 2− | ⎪
⎪ | β | ⎪
⎪ | 2=1.
|
This realisations of SL2(ℝ) (or rather SU(2,ℂ)) is more
suitable for function theory in the unit disk. It is obtained from the
form, which we used before for the upper half-plane, by means of the
Cayley transform [191]*§ 8.1.
We may identify the unit disk ⅅ with the homogeneous space
SL2(ℝ)/T for the unit circle T through
the important decomposition SL2(ℝ)∼ ⅅ×T
with K=T—the compact subgroup of SL2(ℝ):
| | = | | ⎪
⎪ | α | ⎪
⎪ |
| |
| ⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎝ | | ⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎠ |
|
| (21) |
| = | | |
|
where
x=argα,
u=βα−1, | ⎪
⎪ | u | ⎪
⎪ | <1.
|
Each element g∈SL2(ℝ) acts by the linear-fractional transformation
(the Möbius map) on ⅅ
and T
H2(T) as follows:
In the decomposition (21) the first matrix on
the right hand side acts by transformation (22) as an
orthogonal rotation of T or ⅅ; and the
second one—by transitive family of maps of the unit disk onto
itself.
The representation induced by a complex-valued character
χk(z)=z−k of T according to the
Section 3.2 is:
ρk(g): f(z) ↦
| |
f | ⎛
⎜
⎜
⎝ |
| |
| ⎞
⎟
⎟
⎠ |
where
g= | | .
(23) |
The representation 1 is unitary on square-integrable
functions and irreducible on the Hardy space on the unit circle.
We choose [173, 175] K-invariant function v0(z)≡ 1
to be a vacuum vector.
Thus the associated coherent states
v(g,z)=ρ1(g)v0(z)= (u−z)−1
|
are completely determined by the point on the unit disk
u=βα−1. The family of coherent states considered
as a function of both u and z is obviously the Cauchy
kernel [170]. The wavelet transform [170, 173]
W:L2(T)→
H2(ⅅ): f(z)↦
Wf(g)=⟨ f,vg
⟩ is the Cauchy integral:
This approach can be extended to arbitrary connected simply-connected
domain. Indeed, it is known that Möbius maps is the whole group of
biholomorphic automorphisms of the unit disk or upper half-plane. Thus
we can state the following corollary from the Riemann mapping
theorem:
Corollary 13
The group of biholomorphic automorphisms of a connected
simply connected domain with at least two points on its boundary is
isomorphic to SL2(ℝ)
.
If a domain is non-simply connected, then the group of its
biholomorphic mapping can be
trivial [257, 27]. However we may look
for a rich group acting on function spaces rather than on geometric sets.
Let a connected non-simply connected domain D be
bounded by a finite collection of non-intersecting contours
Γi, i=1,…,n. For each Γi consider the
isomorphic image Gi of the SL2(ℝ) group which is defined by the
Corollary 13. Then define the group
G=G1× G2× … × Gn and its action on
L2(∂ D)= L2(Γ1)⊕
L2(Γ2)⊕ … ⊕ L2(Γn)
through the Moebius action of Gi on L2(Γi).
Example 14
Consider an annulus
defined by
r<|
z |<
R. It is bounded by two circles: Γ
1={
z:
|
z |=
r}
and Γ
2={
z:
|
z |=
R}
. For Γ
1 the Möbius action of SL2(ℝ)
is
| | : z↦
| | ,
where | ⎪
⎪ | α | ⎪
⎪ | 2− | ⎪
⎪ | β | ⎪
⎪ | 2=1,
|
with the respective action on Γ
2. Those action can be
linearised in the spaces L2(Γ
1)
and
L2(Γ
2)
. If we consider a subrepresentation
reduced to analytic function on the annulus, then one copy of
SL2(ℝ)
will act on the part of functions analytic outside of
Γ
1 and another copy—on the part of functions analytic
inside of Γ
2.
Thus all classical objects of complex analysis (the Cauchy-Riemann
equation, the Taylor series, the Bergman space, etc.) for a rather generic domain D can
be also obtained from suitable representations similarly
to the case of the upper half-plane [170, 175].
Last modified: October 28, 2024.