Professional Documents
Culture Documents
The Monogenic Hua-Radon Transform and Its Inverse
The Monogenic Hua-Radon Transform and Its Inverse
Contents
1. Introduction 1
2. Preliminaries 2
2.1. Clifford algebras 2
2.2. Clifford analysis 3
2.3. The Lie ball 4
3. The monogenic Hua-Radon transform 5
4. Technical lemmas 9
5. The kernel of the monogenic Hua-Radon transform 12
6. Inversion of the monogenic Hua-Radon transform 16
7. Conclusions 22
Appendix A. Computation of λα k from Theorem 5.2 22
Appendix B. Computation of the constants γα,k from Proposition 6.4 26
Acknowledgements 29
References 29
1. Introduction
The Radon transform is a well-known integral transform with many applications in both pure and
applied mathematics. It was extended to the Clifford setting by Sommen in [20, 21, 23], see also [3].
The Szegő-Radon transform, a variant of the Clifford Radon transform, was introduced in [5] by
Colombo, Sabadini and Sommen. It was originally defined as an orthogonal projection of a Hilbert
module, but it can equally be written as an integral transform over the unit sphere Sm−1 with respect
to a reproducing kernel. Like the Clifford Radon transform, it is a projection from a space of monogenic
functions in m variables over the real Clifford algebra Rm onto a space of monogenic functions in 2
variables over the complex Clifford algebra Cm .
Applying the dual transform R̃ to the reproducing kernel of the Szegő-Radon transform, i.e. taking
the integral over a Stiefel manifold of the reproducing kernel, will result in a linear combination of the
zonal spherical monogenics. Consequently, when acting on a monogenic polynomial Mk (x) of degree
k with the composition of the Szegő-Radon transform with its dual, one obtains a scalar multiple of
Mk (x), where the proportionality constant θk depends on k. As this scalar θk can be accounted for by
the Gamma operator Γx , one can invert the Szegő-Radon transform by applying the operator θ−Γx R̃.
These ideas were established in [5].
It was shown in e.g. [11, 13, 17] that monogenic functions on the unit ball admit a holomorphic
extension in the Lie ball. Following this idea, Sabadini and Sommen defined several Radon-type
transforms over the Lie sphere in [16]. They showed that each of these can be written as an integral
transform over the Lie sphere with respect to a certain kernel. One of these is the monogenic Hua-
Radon transform, which we will study in great detail.
In this paper we complete the work of Sabadini and Sommen, by determining explicitly the kernel
of the monogenic Hua-Radon transform, which is defined as an orthogonal projection onto a space of
holomorphic functions generated by a specific basis. Moreover, we will obtain an inversion for this
transform, along the lines of [5]. At this point we require that the dimension m ≥ 3, which can be
justified by the fact that most of the basis is annihilated in case m < 3. It turns out that using the dual
transform will not produce a full inversion for the monogenic Hua-Radon transform, but it will yield a
projection operator of a holomorphic function onto the term in its Fischer decomposition corresponding
to a certain power of z. Hence we will introduce a total monogenic Hua-Radon transform, for which
the outlined process leads to a complete inversion.
The structure of the paper is as follows. In Section 2 we give the necessary preliminaries on Clifford
algebras, the Lie sphere and Clifford analysis that will be needed in this paper. In Section 3 we define
the monogenic Hua-Radon transform as an orthogonal projection. Section 4 contains some technical
lemmas required in order to compute the kernel of the monogenic Hua-Radon transform in Section 5,
more precisely in Theorem 5.2. The inversion of the monogenic Hua-Radon transform will be discussed
in Section 6. Finally, the more tedious calculations involving hypergeometric series and combinatorial
identities will be performed in Appendix A and B.
2. Preliminaries
In this section we introduce all notations and preliminary results that will be useful for the paper.
We mostly follow the notations from [16].
2.1. Clifford algebras. Consider the vector space Rm with canonical orthonormal basis (e1 , e2 , . . . , em ).
The Clifford algebra Rm , respectively Cm , is the real, respectively complex, algebra generated by the
basis elements e1 , e2 , . . . , em under the relations
e2j = −1, j ∈ {1, . . . , m},
ej ek + ek ej = 0, j 6= k.
Any element α in Rm (or Cm ) is of the form
X
α= αA eA
A⊂{1,...,m}
where αA ∈ R (or αA ∈ C), A = {i1 , . . . , il } where i1 < . . . < il is a multi-index such that eA =
ei1 . . . eil and e∅ = 1. The elements of Rm (or Cm ) which are linear
Pm combinations of only the basis
vectors ej are called 1-vectors and will be underlined, e.g. v = j=1 vj ej . The scalar part α∅ of an
element α will be denoted by [α]0 .
The Hermitian conjugation is an automorphism on Cm defined for α, β ∈ Cm as
(αβ)† = β † α† ,
(α + β)† = α† + β † ,
(αA eA )† = αA e†A ,
e†j = −ej j ∈ {1, . . . , m}.
where αA stands for the complex conjugate of the complex number αA .
THE MONOGENIC HUA-RADON TRANSFORM AND ITS INVERSE 3
2π m/2
Am = ,
Γ m 2
where Γ is the gamma function. Pm
We define the scalar product of two 1-vectors u and v as hu, vi = j=1 uj vj . The wedge product of u
P
and v is defined as u ∧ v = i<j (ui vj − uj vi )ei ej . Easy calculations show
(2.1) u v = −hu, vi + u ∧ v.
We will need the following result which was proven in [5].
Lemma 2.1. Let t, s ∈ Sm−1 be such that ht, si = 0 and let τ = t + is ∈ Cm . Then τ † = −t + is and
(i) τ τ † τ = 4τ ,
(ii) τ 2 = (τ † )2 = 0,
(iii) τ τ † + τ † τ = 4.
We will be working in a complex setting, hence we will use the complexified Dirac operator
m
X
∂z = ej ∂zj .
j=1
m
Its square satisfies ∂z2 = −∆z , where ∆z = ∂z2j is the complexified Laplace operator. If we use
P
j=1
m
∂x2j . The symbol of the Dirac operator ∂z is
P
the real Laplace operator, we will denote it by ∆x =
j=1
denoted by the vector variable
m
X
z= ej zj
j=1
which is a complex-valued variable. If we are working with real variables, we will denote them by
x. By complexifying the variables, the square of the norm and the Dirac operator, we can use the
necessary results of real Clifford analysis.
Definition 2.2. A function f : Ω ⊂ Cm → Cm which is continuously differentiable in the open set
Ω is called (left) monogenic
Pm in Ω if f is holomorphic and f is in the kernel of the complexified Dirac
operator ∂z , i.e. ∂z f = j=1 ej ∂zj f = 0. The right Cm -module of (left) monogenic functions in Ω is
denoted by M(Ω).
A function f : Ω ⊂ Cm → Cm which is continuously differentiable in the open set Ω is called harmonic
in Ω if f is holomorphic and f is in the kernel of the complexified Laplace operator ∆z = −∂z2 . The
right Cm -module of harmonic functions in Ω is denoted by H(Ω).
Looking at the set of k-homogeneous polynomials Pk (Ω) with Ω ⊂ Rm we define the set of k-
homogeneous monogenics and harmonics as
Mk (Ω) = M(Ω) ∩ Pk (Ω),
Hk (Ω) = H(Ω) ∩ Pk (Ω).
4 DENIS CONSTALES, HENDRIK DE BIE, TEPPO MERTENS, AND FRANK SOMMEN
It is a well known fact, (see e.g. [8]), that for each Hk (x) ∈ Hk (Ω) there exist unique monogenic
polynomials Mk (x) ∈ Mk (Ω), Mk−1 (x) ∈ Mk−1 (Ω) such that
2.3. The Lie ball. As we are complexifying our variables, we will also complexify the space over
which we are working. To this end we will consider the Lie sphere LS m−1 and Lie ball LB(0, 1),
instead of the unit sphere Sm−1 and unit ball B(0, 1).
Remark 2.4. Another way to introduce the Lie ball (see [15, 22]) is to consider the Lie norm
Definition 2.5. The Lie sphere LS m−1 is the set of points z = x + iy ∈ Cm for which Sx,y ⊂ Sm−1
or equivalently
Since monogenic functions f (x) on B(0, 1) admit a holomorphic extension f (z) in the Lie ball
LB(0, 1) (see e.g. [11, 13, 17]), we will be interested in the following space of holomorphic functions.
Definition 2.6. Let OL2 (LB(0, 1)) be the right Cm -module consisting of holomorphic functions f :
LB(0, 1) → Cm whose boundary value f (eiθ ω) belongs to L2 (LS m−1 ), i.e.
Z Z π
† iθ iθ
f (e ω)f (e ω)dθdS(ω) < ∞.
Sm−1 0 0
We can equip OL2 (LB(0, 1)) with the following inner product
Z Z π
hf, giOL2 (LB(0,1)) = f † (eiθ ω)g(eiθ ω)dθdS(ω)
Sm−1 0
Proposition 3.1. The functions ψτ ,0,k (z) = τ hz, τ ik are monogenic for each k ∈ N. Moreover for
each s, k ∈ N we have
where j, s, k ∈ N and M [·] stands for the monogenic projection which can be defined as follows. For
each k-homogeneous polynomial Pk (z) there exist harmonic polynomials Hk−2` (z) ∈ H(Ω) of degree
k − 2` for ` = 0, . . . , k2 such that
bX
2c
k
k
X
(3.2) Pk (z) = z j Mk−j (z).
j=0
Definition 3.2. The right Cm -module generated by {ψτj ,α,k | k ∈ N} will be denoted by Mj,α (τ ).
Definition 3.3. The monogenic Hua-Radon transform Mτ ,j is now defined as the orthogonal projec-
tion of OL2 (LB(0, 1)) onto the orthogonal direct sum α∈N Mj,α (τ ) (see [16]).
L
The aim of Sections 3 and 5 will be to write the monogenic Hua-Radon transform as an integral
transform Z πZ
Mτ ,j [f ](z) = K j (z, e−iθ ω)f (e−iθ ω)dS(ω)dθ
0 Sm−1
j
with respect to a certain reproducing kernel K that is to be determined.
In order to determine the monogenic projection of ψτ ,α,k , we will first compute its harmonic pro-
jection.
6 DENIS CONSTALES, HENDRIK DE BIE, TEPPO MERTENS, AND FRANK SOMMEN
Remark 3.4. Note that if we are working in the 2-dimensional case ψτj ,α,k = 0 for all j, k and α 6= 0, 1.
Indeed if m = 2, then τ = eiϕ (e1 ± ie2 ) for some ϕ ∈ [0, 2π). This implies
ψτ ,2s,k (z) = τ eiϕk (z1 ± iz2 )s+k (−z1 ± iz2 )s = (−1)s (−z 2 )s τ eiϕk (z1 ± iz2 )k ,
ψτ ,2s+1,k (z) = τ † τ eiϕ(k+1) (z1 ± iz2 )s+k+1 (−z1 ± iz2 )s = (−1)s (−z 2 )s τ † τ eiϕ(k+1) (z1 ± iz2 )k+1 .
We can now see that M [ψτ ,2s,k ](z) = 0 = M [ψτ ,2s+1,k ](z) whenever s 6= 0. Hence from now on we will
assume m ≥ 3.
We can project Pk in (3.1) onto each of its harmonic components Hk−2j using the following projec-
tion operator (see e.g. [1] and [18]).
Proposition 3.5. The projection operator of a k-homogeneous polynomial onto its harmonic compo-
nent of degree k − 2` is given by the following operator
bk
2 c−`
X
αj (−z 2 )j ∆j+`
z
j=0
where
(−1)j ( m m
2 + k − 2` − 1) Γ( 2 + k − 2` − j − 1)
αj = j+`
4 j!`! Γ( m
2 + k − `)
The projection operator in Proposition 3.5 projects onto the harmonic components, but we need to
project onto the monogenic components. Hence we will rewrite the previous operator using the Dirac
operator and compose it with the projection onto its monogenic component (2.2):
Proposition 3.6. The projection operator of a k-homogeneous polynomial onto its monogenic com-
ponent of degree k − 2l is given by the following operator
k−2`
X
projk2` = βj,2` z j ∂zj+2`
j=0
where
j+`
Γ( m
2 + k − 2` − j)
−1
β2j,2` = ,
4 j!`!Γ( m
2 + k − `)
j+`
Γ( m
2 + k − 2` − j − 1)
−1
β2j+1,2` = .
4 2j!`!Γ( m
2 + k − `)
Proof. Using the projection operator of a harmonic polynomial onto its monogenic component, we
get the following operator projecting a k-homogeneous polynomial onto its monogenic component of
degree k − 2`
k
bX
2 c−`
1
projk2` = 1+ z∂z αj (−1)` z 2j ∂z2(j+`)
2(k − 2`) + m − 2 j=0
∂z z 2j f = −2jz 2j−1 f + z 2j ∂z [f ]
as z 2j is a scalar and hence will commute with all ej for j = 1, . . . , m. Consequently, we can rewrite
the projection operator as follows:
THE MONOGENIC HUA-RADON TRANSFORM AND ITS INVERSE 7
k
bX
2 c−`
1
projk2` = 1+ z∂z αj (−1)` z 2j ∂z2(j+`)
2(k − 2`) + m − 2 j=0
bk
2 c−` b 2 c−` k
X 1 X
= αj (−1)l z 2j ∂z2(j+`) + z αj (−1)` ∂z z 2j ∂x2(j+`)
j=0
2(k − 2`) + m − 2 j=0
k−2`
X
= βj,2` z j ∂zj+2`
j=0
We can now use Proposition 3.6 to calculate the monogenic projection of ψτ ,α,k . The decomposition
was already known in [16], but the explicit expression for the constants was not.
α
X
M [ψτ ,α,k ](z) = µj,α,k z j ψτ ,α−j,k (z).
j=0
where
Γ( m2 + 2s + k − j) Γ(s + 1) Γ(s + k + 1)
µ2j,2s,k = (−1)j m j = 0, . . . , s,
j!Γ( 2 + 2s + k) Γ(s − j + 1) Γ(s + k − j + 1)
Γ( m + 2s + k − j − 1) Γ(s + 1) Γ(s + k + 1)
µ2j+1,2s,k = (−1)j 2 j = 0, . . . , s − 1,
2j!Γ( m
2 + 2s + k) Γ(s − j) Γ(s + k − j + 1)
m
Γ( + 2s + k + 1 − j) Γ(s + 1) Γ(s + k + 2)
µ2j,2s+1,k = (−1)j 2 m j = 0, . . . , s,
j!Γ( 2 + 2s + k + 1) Γ(s − j + 1) Γ(s + k − j + 2)
Γ( m
2 + 2s + k − j) Γ(s + 1) Γ(s + k + 2)
µ2j+1,2s+1,k = 2(−1)j m j = 0, . . . , s.
j!Γ( 2 + 2s + k + 1) Γ(s + 1 − j) Γ(s + k − j + 1)
where
j
Γ( m
2 + α + k − j)
−1
β2j,0 =
4 j!Γ( m
2 + α + k)
j
Γ( m
2 + α + k − j − 1)
−1
β2j+1,0 = .
4 2j!Γ( m
2 + α + k)
Γ(s + 1) Γ(s + k + 1)
∂z2j ψτ ,2s,k (z) = 4j ψτ ,2s−2j,k (z)
Γ(s − j + 1) Γ(s + k − j + 1)
Γ(s + 1) Γ(s + k + 1)
∂z2j+1 ψτ ,2s,k (z) = 4j ψτ ,2s−2j−1,k (z)
Γ(s − j) Γ(s + k − j + 1)
and thus the sum reduces to
M [ψτ ,2s,k ](z) = ψτ ,2s,k (z) + µ1,2s,l zψτ ,2s−1,k (z) + . . . + µ2s,2s,k z 2s ψτ ,0,k (z)
with
Γ(s + 1) Γ(s + k + 1)
µ2j,2s,k = β2j,0 4j
Γ(s − j + 1) Γ(s + k − j + 1)
m
Γ( + 2s + k − j) Γ(s + 1) Γ(s + k + 1)
= (−1)j 2 m
j!Γ( 2 + 2s + k) Γ(s − j + 1) Γ(s + k − j + 1)
Γ(s + 1) Γ(s + k + 1)
µ2j+1,2s,k = β2j+1,0 4j
Γ(s − j) Γ(s + k − j + 1)
m
Γ( + 2s + k − j − 1) Γ(s + 1) Γ(s + k + 1)
= (−1)j 2 .
2j!Γ( m
2 + 2s + k) Γ(s − j) Γ(s + k − j + 1)
The case α = 2s + 1 is done in a similar way.
Remark 3.8. Lemma 3.7 also shows that M [ψτ ,2s,k ] = 0 = M [ψτ ,2s+1,k ] if s ≥ 1 and m = 2 just as we
have shown in Remark 3.4. This might not be obvious at first, but using Lemma 3.7 we have
2s
X
M [ψτ ,2s,k ] = µj,2s,k z j ψτ ,2s−j,k (z)
j=0
s
X s−1
X
(3.3) = µ2j,2s,k z 2j ψτ ,2s−2j,k (z) + µ2j+1,2s,k z 2j+1 ψτ ,2s−2j−1,k (z).
j=0 j=0
Xs s−1
X
(3.3) = µ2j,2s,k − 2 µ2j+1,2s,k z 2s τ eiϕk (z1 ± iz2 )k
j=0 j=0
= 2 F1 ([−s, −s − k], [−2s − k]; 1)
s 2s iϕk
− 2 F1 ([−s + 1, −s − k], [−2s − k + 1]; 1) z τ e (z1 ± iz2 )k
2s + k
THE MONOGENIC HUA-RADON TRANSFORM AND ITS INVERSE 9
4. Technical lemmas
In order to determine the kernel of the monogenic Hua-Radon transform in Section 5, we will need
the following technical lemmas. The first result was already proven in [5].
Lemma 4.1. Let t, s ∈ Rm be such that |t| = |s| = 1 and ht, si = 0, let τ = t + is ∈ Cm and ω ∈ Sm−1 .
Then for k 6= l we have
Z
hω, τ † ik hω, τ il dS(ω) = 0
Sm−1
and
Z
m Γ(k + 1)
hω, τ † ik hω, τ ik dS(ω) = (−1)k 2π 2 .
Sm−1 Γ(k + m
2)
We can write the integral in Lemma 4.1 as the inner product Am h(−1)k hx, τ ik , hx, τ il iSm−1 with
Z
1
hP, QiSm−1 = P (ω)Q(ω)dS(ω)
Am Sm−1
where P (ω) is the complex conjugate of P (ω). It has been proven (see e.g. [6]) that
k+ m
kΓ
h i
2
(4.1) 2 hHk , P i
` S m−1 = H (∂
k x )P ` (x)
Γ m
x=0
2
where Hk ∈ Hk , P` is an `-homogeneous polynomial and Hk (∂x ) is the operator obtained by substi-
tuting ∂xi for xi in H(x). Moreover if k 6= ` then
hMk , M` iSm−1 = 0
hMk , xM` iSm−1 = 0
hxMk , xM` iSm−1 = 0,
where Mk ∈ Mk and M` ∈ M` .
Now we can easily prove the following:
10 DENIS CONSTALES, HENDRIK DE BIE, TEPPO MERTENS, AND FRANK SOMMEN
Lemma 4.3. Let t, s ∈ Rm be such that |t| = |s| = 1 and ht, si = 0, let τ = t + is ∈ Cm and ω ∈ Sm−1 .
Then for k 6= ` + 1 one has
Z
hω, τ † ik hω, τ i` τ ωdS(ω) = 0
Sm−1
and
Z
m Γ(` + 2)
hω, τ † i`+1 hω, τ i` τ ωdS(ω) = (−1)` π 2 τ τ †.
Sm−1 Γ(` + 1 + m2 )
Proof. We have
Z
hω, τ † ik hω, τ i` τ ωdS(ω) = Am h(−1)`+1 hx, τ † i` τ † x, hx, τ † ik iSm−1 .
Sm−1
Using (4.1), it is easy to see that
1
h(−1)`+1 hx, τ † i` τ † x, hx, τ † ik iSm−1 = m
h(−1)`+1 hx, τ † i` τ † , ∂x hx, τ † ik iSm−1
2 k+ 2
1
= m
h(−1)`+1 hx, τ † i` τ † , khx, τ † ik−1 τ † iSm−1
2 k+ 2
1
= m
h(−1)` hx, τ † i` , khx, τ † ik−1 iSm−1 τ τ † .
2 k+ 2
The result now follows from Lemma 4.1.
Remark 4.4. Since τ † and τ have the same properties as −t ∈ Sm−1 and −t ⊥ s, we can easily replace
τ by τ † . Moreover, we have
Z Z †
† k ` k+` k † ` †
hω, τ i hω, τ i τ ωdS(ω) = − (−1) hω, τ i hω, τ i ω τ dS(ω)
Sm−1 Sm−1
(
0 for k 6= ` + 1,
= m Γ(`+2) †
(−1)` π 2 Γ(`+1+ m τ τ
) for k = ` + 1.
2
In order to prove the following lemma, we will need to use Pizzetti’s formula (see e.g. [7, 14]),
which provides an easy way to calculate integrals over the unit sphere using the Laplacian. If f is a
polynomial, then
∞
2π m/2
Z X
(4.3) f (ω) dS(ω) = (∆k f )(0).
Sm−1 4k k!Γ(k + m/2) x
k=0
This will allow us to show the following.
Lemma 4.5. Let t, s ∈ Rm be such that |t| = |s| = 1 and ht, si = 0, let τ = t + is ∈ Cm and ω ∈ Sm−1 .
Then for k 6= ` we have
Z
hω, τ † ik hω, τ i` ω τ τ † ωdS(ω) = 0
Sm−1
and
Z
m k!
(4.4) hω, τ † ik hω, τ ik ω τ τ † ωdS(ω) = (−1)k π 2 m (4τ ∧ τ † − mτ τ † − 2kτ † τ ).
Sm−1 Γ(k + 1 + 2)
THE MONOGENIC HUA-RADON TRANSFORM AND ITS INVERSE 11
Proof. To prove this, we will use Pizzetti’s formula. It can be shown, using Lemma 2.1 and some
straightforward calulations, that for j ≤ min(k, `) one has
† k−1
+ 8kj∆j−1
x (hx, τ i hx, τ i` )τ † x
+ 8`j∆xj−1 (hx, τ † ik hx, τ i`−1 )x τ
† k ` †
− 4j(j − 1)∆j−1
x (hx, τ i hx, τ i )τ τ
+ 8k 2 ∆k−1
x (hx, τ † ik−1 hx, τ ik )τ † x
+ 8k 2 ∆k−1
x (hx, τ † ik hx, τ ik−1 )x τ
− 4k(k − 1)∆k−1
x (hx, τ † ik hx, τ ik )τ † τ
+ 2k∆k−1
x (hx, τ † ik hx, τ ik )T
=(−4)k (k!)2 x τ τ † x
+ 8k 2 (−4)k−1 (k − 1)!k!hx, τ iτ † x
+ 8k 2 (−4)k−1 (k − 1)!k!hx, τ † ix τ
− 4k(k − 1)(−4)k−1 (k!)2 hx, τ † ihx, τ iτ † τ
+ 2k(−4)k−1 (k!)2 hx, τ † ik hx, τ i)T.
Hence applying the Laplace operator one more time yields
∞
2π m/2
Z X
hω, τ † il+1 hω, τ il ωτ τ † ωdS(ω) = ∆jx hx, τ † ik hx, τ ik xτ τ † x |x=0
Sm−1 j=0
4j j!Γ(j + m/2)
m k!
= (−1)k π 2 m (T − 2kτ † τ ).
Γ(k + 1 + 2)
We can now rewrite T to get a neater result. Using (2.1) we get
m
X m
X
T = ei τ τ † ei = ei (−hτ , τ † i + τ ∧ τ † )ei
i=1 i=1
m
X
= −hτ , τ † i(−m) + ei τ ∧ τ † ei = mhτ , τ † i + (4 − m)τ ∧ τ †
i=1
= 4τ ∧ τ † − m(−hτ , τ † i + τ ∧ τ † ) = 4τ ∧ τ † − mτ τ †
where we used Lemma 4.3 of [4]
† † †
Remark 4.6. As τ ∧ τ is a bivector and [τ τ ]0 = [τ τ ]0 = 2, we have that the scalar part of the
right-hand side of (4.4) yields
12 DENIS CONSTALES, HENDRIK DE BIE, TEPPO MERTENS, AND FRANK SOMMEN
m k!
(−1)k+1 4π 2 m .
Γ(k + 2)
(5.1) hψτj ,α,k , ψτj ,α0 ,k0 iOL2 = hM [ψτ ,α,k ], M [ψτ ,α0 ,k0 ]iOL2 = hM [ψτ ,α,k ], ψτ ,α0 ,k0 iOL2 .
Moreover if α + k 6= α0 + k 0 then the three quantities in (5.1) will vanish.
Proof. Using the homogeneity of the functions we get
Z π
0 0
hψτj ,α,k , ψτj ,α0 ,k0 iOL2 = eiθ(α +k −α−k) dθ
0
Z
†
M [ψτ ,α,k ](ω) (−ω 2 )j M [ψτ ,α0 ,k0 ](ω)dS(ω)
×
Sm−1
=hM [ψτ ,α,k ], M [ψτ ,α0 ,k0 ]iOL2 ,
where we used that ω † = −ω and −ω 2 = 1. We can now use (3.2) to decompose ψτ ,α0 ,k0 :
0
αX +k0
ψτ ,α0 ,k0 (z) = z j Mα0 +k0 −` (z)
`=0
with Mα0 +k0 −` (z) a monogenic function for each ` and Mα0 +k0 (z) = M [ψτ ,α0 ,k0 ](z). So now we have
Z π
0 0
hM [ψτ ,α,k ], ψτ ,α0 ,k0 iOL2 = eiθ(α +k −α−k) dθ
0
0 0
† αX
+k
Z
ω ` Mα0 +k0 −` (ω)dS(ω)
× M [ψτ ,α,k ](ω)
Sm−1 `=0
0
+k0
αX Z π
0 0
= eiθ(α +k −α−k) dθ
`=0 0
†
× M [ψτ ,α,k ](x) x` , Mα0 +k0 −` (x) .
Sm−1
where
∞
X †
Lα (z, e−iθ ω) = λα iθ
k M [ψτ ,α,k ](z) M [ψτ ,α,k ](e ω)
k=0
and
Γ( m
2 + 2s + k)
2
m
πAm
−1 α = 2s,
2 m
4s!(s + k)!Γ(s − 1 + 2 )Γ( m2 + s + k)
λα
k =
m
Γ( m
2 + 2s + k + 1) 2
πAm −1 α = 2s + 1.
2 16s!(s + k + 1)!Γ(s + m m
2 )Γ( 2 + s + k)
Proof. In order to prove this result, we will prove that K j,α reproduces the basis elements ψτj ,α,k and
hence it will reproduce each element of ⊕α∈N Mj,α . Moreover, due to the way K j,α is constructed,
we get that the monogenic Hua-Radon transform is indeed represented by an integral transform with
kernel K j,α . Thus we need to prove that
Z Z π
(5.2) ψτj ,α,k (z) = K j,α (z, e−iθ ω)ψτj ,α,k (eiθ ω)dS(ω)dθ.
Sm−1 0
The theorem will then follow using the orthogonality relations of ψτj ,α,k , i.e. if α 6= α0 or k 6= k 0
then hψτj ,α,k , ψτj ,α0 ,k0 iOL2 = 0, which was shown in [16]. We can rewrite equation (5.2) using the
orthogonality relations of M [ψτ ,α,k ] to the following identity
Z Z π
λα
k M [ψτ ,α,k ](z) M [ψτ ,α,k ](eiθ ω)† M [ψτ ,α,k ](eiθ ω)dS(ω)dθ = M [ψτ ,α,k ](z).
Sm−1 0
We will now show that the λkα stated in the Theorem will be the required coefficient in order to
reproduce M [ψτ ,α,k ]. We have, using Lemma 3.7
Z Z π †
M [ψτ ,α,k ](eiθ ω) M [ψτ ,α,k ](eiθ ω)dS(ω)dθ
Sm−1 0
Z Z π
†
M [ψτ ,α,k ](eiθ ω) ψτ ,α,k (eiθ ω)dS(ω)dθ
=
Sm−1 0
α Z Z π
X †
ψτ ,α−j,k (eiθ ω) (eiθ ω)−j ψτ ,α,k (eiθ ω)dS(ω)dθ.
= µj,α,k
j=0 Sm−1 0
Z Z π
Φj,α,k = e−iθ(α−j+k) (ψτ ,α−j,k (ω))† (eiθ ω)−j eiθ(α+k) ψτ ,α,k (ω)dS(ω)dθ
Sm−1 0
Z
=π (ψτ ,α−j,k (ω))† (ω)−j ψτ ,α,k (ω)dS(ω).
Sm−1
First we consider the case α = 2s even. For j = 2l we have
Z
Φ2l,2s,k = (−1)−l π (ψτ ,α−j,k (ω))† ψτ ,α,k (ω)dS(ω)
Sm−1
Z
= (−1)k−l πτ † τ hω, τ † ik−l+2s hω, τ ik−l+2s dS(ω).
Sm−1
Hence by Lemma 4.1, we get
m Γ(k − l + 2s + 1)
Φ2l,2s,k = (−1)k−l πτ † τ (−1)k−l+2s 2π 2
Γ( m
2 + k − l + 2s)
m Γ(k − l + 2s + 1)
= 2π 2 +1 τ † τ .
Γ( m
2 + k − l + 2s)
Now defining
m Γ(k − l + 2s + 1)
φ2l,2s,k = 2π 2 +1 ,
Γ( m
2 + k − l + 2s)
we have Φ2l,2s,k = φ2l,2s,k τ † τ .
For j = 2l + 1, we have by Lemma 4.3
Z
Φ2l+1,2s,k = (−1)−l+1 π (ψτ ,α−j,k (ω))† ωψτ ,α,k (ω)dS(ω)
Sm−1
Z
= (−1)k+l π hω, τ † i2s−l+k hω, τ i2s−l+k−1 τ † τ ω τ dS(ω)
Sm−1
m Γ(2s − l + k + 1) †
= −4π 2 +1 τ τ.
Γ(2s − l + k + m
2)
Write
m Γ(2s − l + k + 1)
φ2l+1,2s,k = −4π 2 +1 ,
Γ(2s − l + k + m
2)
then we have that Φ2l+1,2s,k = φ2l+1,2s,k τ † τ . Note that Φj,2s,k is a multiple of τ † τ for each j. Hence
we have
s s−1
!
X X
(5.3) α
λk M [ψτ ,α,k ](z) µ2l,2s,k φ2l,2s,k + µ2l+1,2s,k φ2l+1,2s,k τ † τ = M [ψτ ,α,k ](z).
l=0 l=0
Now note that each of the terms in the decomposition of M [ψτ ,2s,k ] shown in Lemma 3.7 is a multiple
of τ since
ψτ ,2s−2j,k (z) = hz, τ is−j+k hz, τ † is−j τ ,
ψτ ,2s−2j−1,k (z) = hz, τ is−j+k−1 hz, τ † is−j τ † τ .
Thus if we use Lemma 2.1 (i), i.e. τ τ † τ = 4τ , we have
s s−1
!
X X
4λ2s
k µ2l,2s,k φ2l,2s,k + µ2l+1,2s,k φ2l+1,2s,k =1
l=0 l=0
Summarizing the case α = 2s even, we have
THE MONOGENIC HUA-RADON TRANSFORM AND ITS INVERSE 15
s s−1
1 2s −1 X X
(λk ) = µ2l,2s,k φ2l,2s,k + µ2l+1,2s,k φ2l+1,2s,k
4
l=0 l=0
m s
2π 2 +1 Γ(s + 1)Γ(s + k + 1) Γ(2s + k − l + 1)
X
= (−1)l
Γ( m
2 + 2s + k) Γ(l + 1)Γ(s − l + 1)Γ(s + k − l + 1)
l=0
s−1
!
X Γ(2s + k − l + 1)
− (−1)l .
Γ(l + 1)Γ(s − l)Γ(s + k − l + 1)( m
2 + 2s + k − l − 1)
l=0
m Γ(2s + k − l + 2)
Φ2l,2s+1,k = (−1)k−l+1 π4τ † τ (−1)2s+k−l+1 2π 2
Γ( m
2 + 2s + k − l + 1)
m Γ(k − l + 2s + 2)
= 8π 2 +1 τ † τ .
Γ( m
2 + k − l + 2s + 1)
If we now define
m Γ(k − l + 2s + 2)
φ2l,2s+1,k = 8π 2 +1 ,
Γ( m
2 + k − l + 2s + 1)
we have Φ2l,2s+1,k = φ2l,2s+1,k τ † τ .
For j = 2l + 1, we have by Lemma 4.3
Z
Φ2l+1,2s+1,k = (−1)−l+1 π (ψτ ,α−j,k (ω))† ωψτ ,α,k (ω)dS(ω)
Sm−1
Z
= (−1) k+l+1
π hω, τ † i2s−l+k hω, τ i2s−l+k+1 τ † ω τ † τ dS(ω)
Sm−1
m Γ(2s − l + k + 2)
= −4π 2 +1 τ †τ .
Γ(2s − l + k + 1 + m2 )
We can now define
m Γ(2s − l + k + 2)
φ2l+1,2s+1,k = −4π 2 +1 ,
Γ(2s − l + k + 1 + m2)
so that Φ2l+1,2s+1,k = φ2l+1,2s+1,k τ † τ .
Summarizing the case α = 2s + 1 odd, we have
s s
1 2s+1 −1 X X
(λk ) = µ2l,2s+1,k φ2l,2s+1,k + µ2l+1,2s+1,k φ2l+1,2s+1,k
4
l=0 l=0
m s
8π 2 +1
Γ(s + 1)Γ(s + k + 2) Γ(2s + k − l + 2)
X
= (−1)l
Γ( m
2 + 2s + k + 1) Γ(l + 1)Γ(s − l + 1)Γ(s + k − l + 2)
l=0
s
!
X
l Γ(2s + k − l + 2)Γ( m2 + 2s + k − l)
− (−1) .
Γ(l + 1)Γ(s − l + 1)Γ(s + k − l + 1)Γ( m
2 + 2s + k − l + 1)
l=0
16 DENIS CONSTALES, HENDRIK DE BIE, TEPPO MERTENS, AND FRANK SOMMEN
After some lengthy and straightforward calculations which are contained in Appendix A, we get the
result stated in the theorem.
(|x||y|)k x ∧ y m2 −1
m
−1
(6.1) Ck (x, y) = (k + m − 2)Ck2 (t) + (m − 2) C (t)
m−2 |x||y| k−1
with t = hx, yi/(|x||y|). Moreover they exhibit the following properties:
(i) Ck (x, y) is a homogeneous polynomial of degree k in x and y,
Pm
(ii) ∂x Ck (x, y) = 0 and Ck (x, y)∂y = 0, where Ck (x, y)∂y = j=1 (∂yj Ck (x, y))ej ,
Q2r
(iii) For every σ ∈ Spin(m) = { i=1 σi |σi ∈ Sm−1 } the functions Ck (x, y) have the invariance property
The functions M [ψτ ,α,k ] that we defined earlier in Section 3, meet (i) and (ii). But they do not
satisfy property (iii). However they do admit the following Lemma.
Lemma 6.2. Let σ ∈ Spin(m), then it holds that
M [ψτ ,α,k ](x) = ψτ ,α,k (x) + µ1 xψτ ,α−1,k (x) + . . . + µα xα ψτ ,0,k (x).
Hence it suffices to look at σψτ ,α,k (σxσ)σ. Now observing that σσ = σσ = 1 and that hσxσ, τ i =
hx, στ σi, we get
Z Z
(6.2) M [ψτ ,α,k ](x)M [ψτ ,α,k ](y)† dS(s)dS(t) = γα,k Cα+k (x, y)
Sm−1 Sm−2
Proof. If we define the integral on the left of (6.2) as Lα,k (x, y), then we see that this integral is a
homogeneous polynomial of degree α + k in both x and y. Moreover it is left monogenic in x and right
monogenic in y.
Furthermore, let σ ∈ Spin(m), then by using Lemma 6.2 we get
Z Z
σLk (σxσ, σyσ)σ = σM [ψτ ,α,k ](σxσ)M [ψτ ,α,k ](σyσ)† σdS(s)dS(t)
Sm−1 Sm−2
Z Z
= M [ψστ σ,α,k ](x)M [ψστ σ,α,k ](y)† dS(s)dS(t)
Sm−1 Sm−2
Z Z
= M [ψτ ,α,k ](x)M [ψτ ,α,k ](y)† dS(s)dS(t).
Sm−1 Sm−2
Hence Lα,k (x, y) is Spin-invariant, so all the conditions of Lemma 6.1 are fulfilled. Now noting that
Lα,k (x, x) is the sum of a scalar aα,k (x) and a bivector bα,k (x), Cα+k (x, x) is a scalar, we get
aα,k (x) + bα,k (x) = (λ + µe1...m )Cα+k (x, x).
If we use the fact that m ≥ 3 (see Remark 3.4), we see that bα,k (x) = 0 and µ = 0. Hence we have
m
s!(s + k)!Γ + s − 1 m
2
Γ(m − 1)Γ(2s + k + 1)
γ2s,k = 2 Γ + s + k (m − 2) ,
m 2 Γ(2s + k + m − 1)
Γ 2 + 2s + k
s!(s + k + 1)!Γ m
2 +s m Γ(m − 1)Γ(2s + k + 2)
γ2s+1,k = 2 Γ + s + k (m − 2) .
Γ m + 2s + k 2 Γ(2s + k + m)
2
Proof. As the equality in Proposition 6.3 must hold for each x and y, it suffices to calculate γα,k for
x = y = ω ∈ Sm−1 and take the scalar part of the equation. The right-hand side of (6.2) will now be
equal to
α + k + m − 2 m/2−1
γα,k Cα+k (1).
m−2
First we make the following observation for the left hand side of (6.2). There exists σ ∈ Spin(m) such
that τ = σ(e1 + ie2 )σ.
For ease of notation let us write η = e1 + ie2 . Using Lemma 6.2 we get
(6.3) M [ψτ ,α,k ](ω)M [ψτ ,α,k ](ω)† = σM [ψη,α,k ](σωσ)M [ψη,α,k ](σωσ)† σ.
The integral in (6.2) can be interpreted as the mean of (6.3) over the Stiefel manifold. But using the
notation τ = σησ, we can also interpret it as a mean over the Spin-group Spin(m) (see e.g. [5]). Hence
using Lemma 6.2 we get
Z Z
1
M [ψτ ,α,k ](ω)M [ψτ ,α,k ](ω)† dS(s)dS(t)
Am Am−1 Sm−1 Sm−2
Z
1
(6.4) = M [ψσησ,α,k ](ω)M [ψσησ,α,k ](ω)† dS(σ)
Vol(Spin(m)) Spin(m)
Z
1
= σM [ψη,α,k ](σωσ)M [ψη,α,k ](σωσ)† σdS(σ).
Vol(Spin(m)) Spin(m)
18 DENIS CONSTALES, HENDRIK DE BIE, TEPPO MERTENS, AND FRANK SOMMEN
Since the action of the Spin element on a vector is a rotation, it is easy to see that it is an automorphism
on k-vectors, see [9, 10]. Hence we get that the scalar part of (6.4) is equal to
Z
1 h i
M [ψη,α,k ](σωσ)M [ψη,α,k ](σωσ)† dS(σ)
Vol(Spin(m)) Spin(m) 0
Z
1 h i
= M [ψη,α,k ](ω)M [ψη,α,k ](ω))† dS(ω).
Am Sm−1 0
In order to continue, we need to know the parity of α − j and α − l, so we will distinguish the cases
where α is even and where α is odd. Suppose α = 2s even, then we have
α
X
µj,α,k µl,α,k ω j ψη,α−j,k (ω)ψη,α−l,k (ω)† (ω † )l
j,l=0
s X
X s
= µ2j,α,k µ2l,α,k (−1)j+l ψη,2s−2j,k (ω)ψη,2s−2l,k (ω)†
j=0 l=0
s−1 X
X s
+ µ2j+1,α,k µ2l,α,k (−1)j+l ωψη,2(s−j−1)+1,k (ω)ψη,2s−2l,k (ω)†
j=0 l=0
s X
X s−1
+ µ2j,α,k µ2l+1,α,k (−1)j+l+1 ψη,2s−2j,k (ω)ψη,2s−2l−1,k (ω)† ω
j=0 l=0
s−1 X
X s−1
+ µ2j+1,α,k µ2l+1,α,k (−1)j+l+1 ωψη,2s−2j−1,k (ω)ψη,2s−2l−1,k (ω)† ω.
j=0 l=0
where we used that ω 2 = −1 and ω † = −ω for ω ∈ Sm−1 . So now we have 4 different double sums we
need to integrate. We can calculate the integrals using the previous lemmas and then we can take the
scalar part in order to simplify the next calculations.
• Rewriting the terms of the first sum, gives us
s−l+k s−l
µ2j,α,k µ2l,α,k (−1)j+l hω, ηis−j+k hω, η † is−j ηη † hω, ηi hω, η † i
= µ2j,α,k µ2l,α,k (−1)j+l+k hω, ηi2s−j−l+k hω, η † i2s−j−l+k ηη †
Hence using Lemma 4.1 we get
Z
1
µ2j,α,k µ2l,α,k (−1)j+l ψη,2s−2j,k (ω)ψη,2s−2l,k (ω)† dS(ω) =
Am Sm−1
Z
1
= µ2j,α,k µ2l,α,k (−1)j+l+k hω, ηi2s−j−l+k hω, η † i2s−j−l+k ηη † dS(ω)
Am Sm−1
1 m Γ(2s − j − l + k + 1)
= µ2j,α,k µ2l,α,k 2π 2 ηη †
Am Γ 2s − j − l + k + m 2
Hence the scalar part will be equal to
1 m Γ(2s − j − l + k + 1)
2µ2j,α,k µ2l,α,k 2π 2 .
Am Γ 2s − j − l + k + m
2
THE MONOGENIC HUA-RADON TRANSFORM AND ITS INVERSE 19
s−l+k s−l
µ2j+1,α,k µ2l,α,k (−1)j+l ωhω, ηis−j+k hω, η † is−j−1 η † ηη † hω, ηi hω, η † i .
Z
1
4µ2j+1,α,k µ2l,α,k (−1)j+l+k hω, ηi2s−j−l+k hω, η † i2s−j−l+k−1 ωη † dS(ω)
Am Sm−1
1 m Γ(2s − j − l + k + 1)
=− 4µ2j+1,α,k µ2l,α,k π 2 ηη † .
Am Γ 2s − j − l + k + m
2
1 m Γ(2s − j − l + k + 1)
− 8µ2j+1,α,k µ2l,α,k π 2 .
Am Γ 2s − j − l + k + m
2
s−l+k s−l−1
µ2j,α,k µ2l+1,α,k (−1)j+l+1 hω, ηis−j+k hω, η † is−j ηη † ηhω, ηi hω, η † i ω.
Z
1
4µ2j,α,k µ2l+1,α,k (−1)j+l+k hω, ηi2s−j−l+k−1 hω, η † i2s−j−l+k ηωdS(ω)
Am Sm−1
1 m Γ(2s − j − l + k + 1)
=− 4µ2j,α,k µ2l+1,α,k π 2 ηη † .
Am Γ 2s − j − l + k + m
2
1 m Γ(2s − j − l + k + 1)
− 8µ2j,α,k µ2l+1,α,k π 2 .
Am Γ 2s − j − l + k + m
2
Note that this is equal to the terms of the second sum where we interchange j and l.
• Lastly, for the terms of the fourth sum, we have
s−l+k s−l−1
µ2j+1,α,k µ2l+1,α,k (−1)j+l+1 ωhω, ηis−j+k hω, η † is−j−1 ηη † ηη † hω, ηi hω, η † i ω.
Using Lemma 4.5 and taking the scalar part, the terms of this sum become
1 m (2s − j − l + k − 1)!
µ2j+1,α,k µ2l+1,α,k 16π 2 .
Am Γ(2s − j − l + k − 1 + m2)
So now we have
20 DENIS CONSTALES, HENDRIK DE BIE, TEPPO MERTENS, AND FRANK SOMMEN
2s + k + m − 2 m2 −1
γ2s,k C2s+k (1)
m−2
Z Z
1 †
= M [ψτ ,α,k ](ω)M [ψτ ,α,k ](ω) dS(s)dS(t)
Am Am−1 Sm−1 Sm−2 0
s s
4 m X X Γ(2s − j − l + k + 1)
= π2 µ2j,α,k µ2l,α,k
2s − j − l + k + m
Am j=0 l=0
Γ 2
s−1 X
s
X Γ(2s − j − l + k + 1)
−4 µ2j+1,α,k µ2l,α,k
Γ 2s − j − l + k + m
j=0 l=0 2
s−1 X
s−1
X (2s − j − l + k − 1)!
+4 µ2j+1,α,k µ2l+1,α,k
j=0 l=0
Γ(2s − j − l + k − 1 + m
2)
s Xs
m X Γ(2s − j − l + k + 1)
=2Γ µ2j,α,k µ2l,α,k
2s − j − l + k + m
2 j=0
Γ 2
l=0
s−1 X
s
X Γ(2s − j − l + k + 1)
−4 µ2j+1,α,k µ2l,α,k
Γ 2s − j − l + k + m
j=0 l=0 2
s−1 X
s−1
X (2s − j − l + k − 1)!
+4 µ2j+1,α,k µ2l+1,α,k .
j=0 l=0
Γ(2s − j − l + k − 1 + m
2)
After doing the necessary calculations which are contained in Appendix B, we end up with the
result stated in the Proposition.
R̃[Mτ ,j (z n M` (z))] = 0 if n 6= j
j
and for f (z) = z Ml (z) we have
(` + 1)!(m − 2)!
ϑj,` = .
2(` + m − 2)!
THE MONOGENIC HUA-RADON TRANSFORM AND ITS INVERSE 21
Proof. We have
Z Z π
R̃[Mτ ,j (z n M` )] = R̃ K j (z, e−iθ ω)(eiθ ω)n M` (eiθ ω)dS(ω)dθ
m−1 0
"ZS #
Z π X
j α −iθ iθ −j iθ n iθ
= R̃ z L (z, e ω)(e ω) (e ω) M` (e ω)dS(ω)dθ
Sm−1 0 α∈N
Z Z π X
z j R̃ Lα (z, e−iθ ω) (eiθ ω)n−j M` (eiθ ω)dS(ω)dθ.
=
Sm−1 0 α∈N
Using Proposition 6.3, we get
"∞ #
X
−iθ †
α α iθ
R̃ L (z, e ω) = R̃ λk M [ψτ ,α,k ](z)M [ψτ ,α,k ](e ω)
k=0
∞
X
†
λα iθ
= k R̃ M [ψτ ,α,k ](z)M [ψτ ,α,k ](e ω)
k=0
X∞
−iθ
= λα
k γα,k Cα+k (z, e ω).
k=0
Hence
∞ Z
XX π
(6.5) R̃[Mτ ,j (z j M` )] =z j eiθ(n+`−j−α−k) dθ
α∈N k=0 0
Z
× λα
k γα,k Cα+k (z, ω)ω
n−j
M` (ω)dS(ω).
Sm−1
Using Remark 4.2, we get that (6.5) is non-zero if and only if n = j. If n = j, we can use the
reproducing properties of the zonal spherical monogenics so that
!
X X
j
R̃[Mτ ,j (z M` )] = z πAm j
λ2s
k γ2s,k + λ2s+1
k γ2s+1,k M` (z)
2s+k=` 2s+1+k=`
Using this result we can write ϑj,l calculated in Theorem 6.6 as an operator:
22 DENIS CONSTALES, HENDRIK DE BIE, TEPPO MERTENS, AND FRANK SOMMEN
ϑ2j,` = ϑ2j,−Γz ,
ϑ2j+1,` = ϑ2j+1,Γz −m+1 .
This yields an inversion formula for the monogenic Hua-Radon transform for functions of the type
z j M` (z) given in the following Theorem:
Theorem 6.7. For any j, ` ∈ N, we have
z 2j M` (z) = ϑ−1 2j
2j,−Γz R̃[Mτ ,2j (z M` (z))],
7. Conclusions
In Theorem 5.2 we completed the computations of the kernel of the monogenic Hua-Radon trans-
form. Following the techniques used to invert the Szegő-Radon transform in [5], we obtained a pro-
jection operator mapping a holomorphic function onto the term of the form z j Mk (z) in its Fischer
decomposition. This process leads to a full inversion formula for the total monogenic Hua-Radon
transform defined in Theorem 6.7.
THE MONOGENIC HUA-RADON TRANSFORM AND ITS INVERSE 23
Appendix A. Computation of λα
k from Theorem 5.2
s s
X Γ(2s + k − l + 1) X (2s + k − l)! s!
(−1)l = (−1)l
Γ(l + 1)Γ(s − l + 1)Γ(s + k − l + 1) l!(s − l)!(s + k − l)! s!
l=0 l=0
s
X s 2s + k − l
= (−1)l ,
l s+k−l
l=0
s s
X Γ(2s + k − l + 2) X (2s + k − l + 1)! s!
(−1)l = (−1)l
Γ(l + 1)Γ(s − l + 1)Γ(s + k − l + 2) l!(s − l)!(s + k − l + 1)! s!
l=0 l=0
s
X s 2s + k − l + 1
= (−1)l .
l s+k−l+1
l=0
Note that the last sum is just the first sum by shifting k by 1. This leads to the following:
s
X s
l 2s + k − l
(A.1) (−1) = 1.
l s+k−l
l=0
n n−1 n−1
= +
j j j−1
s
X s 2s + k − l s 2s + k s s+k
(−1)l = + (−1)s
l s+k−l 0 s+k s k
l=0
s−1
X
l s−1 s−1 2s + k − l
+ (−1) +
l l−1 s+k−l
l=1
s−1 2s + k s−1 s − 1 s+k
= − (−1)
0 s+k s−1 k
s−1
X
l s−1 s−1 2s + k − l
+ (−1) +
l l−1 s+k−l
l=1
s−1
X s−1 2s + k − l 2s + k − l − 1
= (−1)l −
l s+k−l s+k−l−1
l=0
s−1
X
l s−1 2s + k − l − 1
(A.2) = (−1) .
l s+k−l
l=0
24 DENIS CONSTALES, HENDRIK DE BIE, TEPPO MERTENS, AND FRANK SOMMEN
Hence we see that (A.2) is equal to sum (A.1) under the transform s 7→ s − 1 and k 7→ k + 1. Hence
repeating this process allows us to lower the upper index of the sum, so that we end up with
s Xs−s
X s 2s + k − l s−s 2s + k − l − s
(−1)l = (−1)l
l s+k−l l s+k−l
l=0 l=0
0 s+k−0
=
0 s+k−0
=1.
m s−1 !
8π 2 +1 s!(s + k)!
−1
X s s−1 2s + k − l
(λ2s
k ) = 1− l
(−1) ,
Γ( m
2 + 2s + k)
m
2 + 2s + k − l − 1 l s+k−l
l=0
m s !
32π 2 +1 s!(s + k + 1)!
X s+1 s 2s + k − l + 1
(λ2s+1 )−1 = 1− (−1)l m .
k
Γ( m
2 + 2s + k + 1) 2 + 2s + k − l l s+k−l
l=0
We see that for the remaining sums, if we shift s by 1 and k by −1 in the first sum, we get the second
sum. Hence it suffices to calculate the first sum.
Proof. First of all, we see that we can rewrite this series as a multiple of a hypergeometric function
3 F2 :
s−1
X s s−1 2s + k − l
(−1)l m
2 + 2s + k − l − 1 l s+k−l
l=0
sΓ(2s + k + 1)
= m
2 + 2s + k − 1 Γ(s + k + 1)Γ(s + 1)
m m
× 3 F2 ([1 − s, −k − s, − − 2s − k + 1], [−k − 2s, − − 2s − k + 2]; 1) .
| 2 {z 2 }
:=φ
In order to determine the value of this hypergeometric function, we will use the following result from
[12, p873, appendix, equation (II)]:
(1 + a − d − N − e + b)N
(A.3) 3 F2 ([a, b, −N ], [d, e]); 1) =(−1)N
(d)N
× 3 F2 ([e − a, e − b, −N ], [e − b − a + d, e]); 1).
m
N =s−1 b = −k − s a=− − 2s − k + 1
2
m
d=− − 2s − k + 2 = a + 1 e = −k − 2s = b − N − 1,
2
we get that
(1)N
φ = (−1)N 3 F2 ([b − N − a − 1, −N − 1, −N ], [−N, b − N − 1]; 1)
(a + 1)N
N
(−1)N (1)N X (b − N − a − 1)j (−N − 1)j (−N )j
=
(a + 1)N j=0 (−N )j (b − N − 1)j j!
N +1
(−1)N (1)N X (b − N − a − 1)j (−N − 1)j (b − N − a − 1) N +1
= − (−1)N +1
(a + 1)N j=0
(b − N − 1) j j! (b − N − 1) N +1
N
(−1) N ! N +1 (b − N − a − 1)N +1
= F
2 1 ([−N − 1, b − c − N − 1], [b − N − 1]; 1) − (−1) .
(a + 1)N (b − N − 1)N +1
We can now use Gauss hypergeometric theorem [2, Chapter 15] so that we end up with
(−1)N N !
(a)N +1 (b − N − a − 1)N +1
φ= − (−1)N +1
(a + 1)N (b − N − 1)N +1 (b − N − 1)N +1
N
(−1) N ! (a)N +1 (1 + a − b)N +1
= −
(b − N − 1)N +1 (a + 1)N (a + 1)N
−N ! (1 + a − b)N +1
= a−
(1 − b)N +1 (a + 1)N
!
−m
(s − 1)! m 2 −s+2 s
= + 2s + k − 1 +
−m
2 − 2s − k + 2 s−1
(k + s + 1)s 2
!
m
Γ(s)Γ(s + k + 1) m 2 −1 s
= + 2s + k − 1 − m .
Γ(2s + k + 1) 2 2 + s + k s−1
Thus now our sum becomes
s−1
X s s−1 2s + k − l
(−1)l m
2 + 2s + k − l − 1 l s+k−l
l=0
!
m
1 m 2 −1
= m
+ 2s + k − 1 − m
s
2 + 2s + k − 1 2 2 +s+k s−1
(m
2 − 1)s
=1 − m
( 2 + s + k)s
Γ(s − 1 + m m
2 )Γ( 2 + s + k)
=1 − .
Γ( m m
2 − 1)Γ( 2 + 2s + k)
26 DENIS CONSTALES, HENDRIK DE BIE, TEPPO MERTENS, AND FRANK SOMMEN
s X
s
X Γ(2s − j − l + k + 1)
µ2j µ2l
Γ 2s − j − l + k + m
j=0 l=0 2
s m
(s!)2 ((s + k)!)2 X j Γ( 2 + 2s + k − j)
(B.1) = (−1)
Γ( m 2
2 + 2s + k) j=0 j!(s − j)!(s + k − j)!
s
X Γ( m2 + 2s + k − l) Γ(2s − j − l + k + 1)
(B.2) × (−1)l ,
l!(s − l)!(s + k − l)! Γ 2s − j − l + k + m
2
l=0
s−1 X
s
X Γ(2s − j − l + k + 1)
µ2j+1 µ2l
Γ 2s − j − l + k + m
j=0 l=0 2
s−1 m
(s!)2 ((s + k)!)2 X j Γ( 2 + 2s + k − j − 1)
(B.3) = (−1)
Γ( m 2
2 + 2s + k) j=0 2j!(s − j − 1)!(s + k − j)!
s
X (−1)l Γ( m
2 + 2s + k − l)Γ(2s − j − l + k + 1)
(B.4) × m ,
l=0
l!(s − l)!(s + k − l)!Γ 2s − j − l + k + 2
s−1 X
s−1
X Γ(2s − j − l + k + 1)
µ2j+1 µ2l+1
Γ 2s − j − l + k + m
j=0 l=0 2
s−1 m
(s!)2 ((s + k)!)2 X j Γ( 2 + 2s + k − j − 1)
(B.5) = m (−1)
Γ( 2 + 2s + k)2 j=0 2j!(s − j − 1)!(s + k − j)!
s−1
X (−1)l Γ( m
2 + 2s + k − l − 1)Γ(2s − j − l + k)
(B.6) × .
l=0
2l!(s − l − 1)!(s + k − l)!Γ 2s − j − l + k − 1 + m
2
Now note that the right-hand sides of (B.2) and (B.4) are identical and if we replace s by s − 1 and
k by k + 1, the right-hand side of (B.2) transforms into (B.6). Moreover, if we write out the sums in
the odd case, the summations with respect to l are equal to (B.2) (if necessary change k to k + 1).
Moreover (B.3) and (B.5) can be achieved by changing s to s − 1 and k to k + 1 in (B.1) and the same
can be done for the remaining sums in the odd case. Hence it suffices to only calculate sum (B.2) and
(B.1). We have the following:
Proposition B.1.
s
X Γ( m2 + 2s + k − l) Γ(2s − j − l + k + 1)
(−1)l = 1.
l!(s − l)!(s + k − l)! Γ 2s − j − l + k + m
2
l=0
Proof. We have
s
X Γ( m2 + 2s + k − l) Γ(2s − j − l + k + 1)
(−1)l =
l!(s − l)!(s + k − l)! Γ 2s − j − l + k + m
2
l=0
Γ( m
2 + 2s + k) Γ(2s − j + k + 1)
φ,
s!(s + k)! Γ m2 + 2s + k − j
where
with
m
a = −s − k b = j − k − 2s − +1 N =s
2
m
d = −2s + j − k = a − N + j e=− − 2s − k + 1 = b − j.
2
Using (A.3), we have
(1 + a + b − d − e − N )N
φ = (−1)N 3 F2 ([e − a, e − b, −N ], [e − a − b + d, e]; 1)
(d)N
(1)N
= (−1)N 3 F2 ([e − a, −j, −N ], [−N, e]; 1)
(d)N
j
(−1)N N ! X (e − a)l (−j)l (−N )l
=
(d)N (−N )l (e)l l!
l=0
j
Γ(−d + N + 1)N ! X (e − a)l (−j)l
=
Γ(−d + 1) (e)l l!
l=0
Γ(−d + N + 1)N !
= 2 F1 ([e − a, −j], [e]; 1)
Γ(−d + 1)
where we used that j ≤ s = N and hence the sum will terminate at j instead of N . Now using
Chu-Vandermonde identity (see [2]) we have
(a)j
2 F1 ([e − a, −j], [e]; 1) =
(e)j
(−s − k)j
=
(− m2 − 2s − k + 1)j
a(a + 1) . . . (a + j − 1)
=
e(e + 1) . . . (e + j − 1)
(−a)(−a − 1) . . . (−a − j + 1)
=
((−e)(−e − 1) . . . (−e − j + 1)
Γ(−a + 1)Γ(−e − j + 1)
=
Γ(−a − j + 1)Γ(−e + 1)
as −a − j + 1 ≥ 0, −a ≥ 0, −e ≥ 0 and −e − j + 1 ≥ 0. Thus for our total sum, we have
s
X Γ( m2 + 2s + k − l) Γ(2s − j − l + k + 1)
(−1)l
l!(s − l)!(s + k − l)! Γ 2s − j − l + k + m
2
l=0
Γ( m
2 + 2s + k) Γ(2s − j + k + 1) Γ(−d + N + 1)N ! Γ(−a + 1)Γ(−e − j + 1)
=
s!(s + k)! Γ m2 + 2s + k − j
Γ(−d + 1) Γ(−a − j + 1)Γ(−e + 1)
Γ(−e + 1) Γ(−d + 1) Γ(−d + N + 1)N ! Γ(−a + 1)Γ(−e − j + 1)
=
N !Γ(−a + 1) Γ(−e − j + 1) Γ(−d + 1) Γ(−a − j + 1)Γ(−e + 1)
Γ(−d − N + 1)
=
Γ(−a − j + 1)
Γ(−d − N + 1)
=
Γ(−d − N + 1)
=1.
28 DENIS CONSTALES, HENDRIK DE BIE, TEPPO MERTENS, AND FRANK SOMMEN
Now combining the results of Proposition B.1 and Proposition B.2 we get the following:
Proposition B.3. The constants γα,k are given by
s!(s + k)!Γ m
2 +s−1 m Γ(m − 1)Γ(2s + k + 1)
γ2s,k = 2 Γ + s + k (m − 2) ,
m 2 Γ(2s + k + m − 1)
Γ 2 + 2s + k
s!(s + k + 1)!Γ m
2 +s−1 m Γ(m − 1)Γ(2s + k + 1)
γ2s+1,k = 4 2 Γ + s + k (m − 2) .
m 2 Γ(2s + k + m − 1)
Γ + 2s + k
2
and thus
THE MONOGENIC HUA-RADON TRANSFORM AND ITS INVERSE 29
Acknowledgements
The authors would like to thank Roy Oste for his advice on hypergeometric series, which was
especially useful in establishing Proposition A.2. The work of HDB is supported by the Research
Foundation Flanders (FWO) under Grant EOS 30889451.
References
[1] S. Ben Saı̈d, B. Ørsted, Segal-Bargmann transforms associated with finite Coxeter groups. Mathematische Annalen
334 (2006), 281–323.
[2] R. F. Boisvert, C. W. Clark, D. W. Lozier, F. W. Olver, NIST Handbook of Mathematical Functions. Cambridge
University Press, 2010.
[3] J. Bureš, R. Lávička, V. Souček, Elements of Quaternionic Analysis and Radon transform. Textos de Matematica
42 Departamento de Matemática da Universidade de Coimbra (2009)
[4] L. Cnudde, H. De Bie, Slice Fourier Transform and Convolutions. Annali di Matematica Pura ed Applicati 196
(2017), 837–862.
[5] F. Colombo, I. Sabadini, F. Sommen, On the Szegő–Radon projection of monogenic functions. Advances in applied
mathematics 74 (2016), 1–22.
[6] F. Dai, Y. Xu, Approximation Theory and Harmonic Analysis on Spheres and Balls, Springer, London, 2013.
[7] H. De Bie, F. Sommen, Spherical Harmonics and Integration in Superspace. Journal of Physics A-Mathematical
and Theoretical 40 (2007), 7193–7212.
[8] R. Delanghe, F. Sommen, V. Souček, Clifford algebra and spinor-valued functions. Mathematics and its Applications
53 (1992) Kluwer Academic Publishers Group, Dordrecht.
[9] T. Friedrich, Dirac Operators in Riemannian Geometry. Graduate Studies in Mathematics 25 (2000), American
Mathematical Society.
[10] J. E. Gilbert, M.Murray, Clifford algebras and Dirac operators in harmonic analysis. Cambridge studies in Advanced
Mathematics 26 (1991), Cambridge University Press.
[11] L.K. Hua, Harmonic Analysis of Functions of Several Complex variables in Classical Domains. Translation of
Mathematical Monographs 6 (1959) AMS.
[12] R. Jagannathan, V. Rajeswari, J. Raynal, K.S. Rao, J. Van der Jeugt, Group theoretical basis for the terminating
3F2(l) series. Journal of Physics A: Mathematical and General 25 (1992), 861–876.
[13] M. Morimoto, Analytic Functionals on the Lie Sphere. J. Math 3 (1980), 1–35.
[14] P. Pizzetti, Sulla media dei valori che una funzione dei punti dello spazio assume alla superficie di una sfera. Rend.
Lincei, 18 (1909), 182–185
[15] I. Sabadini, F. Sommen, A Fourier–Borel Transform for Monogenic Functionals on the Lie ball. Journal of the
Mathematical Society of Japan 68 (2016), 1487–1504.
30 DENIS CONSTALES, HENDRIK DE BIE, TEPPO MERTENS, AND FRANK SOMMEN
[16] I. Sabadini, F. Sommen, Radon type transforms for holomorphic functions in the Lie ball. Journal of Geometric
analysis 29 (2019), 2709–2737.
[17] J. Siciak, Holomorphic Continuation of Harmonic Functions. Ann. Polon. Math. 29 (1974), 67–73.
[18] S. L. Sobolev, Introduction to the Theory of Cubature Formulas. Izdat. ”Nauka” (1974) Moscow.
[19] F. Sommen, Plane waves, biregular functions and hypercomplex Fourier analysis. Rend. Circ. Math. Palermo 9
(1986), 205–219.
[20] F. Sommen, An extension of the Radon transform to Clifford analysis. Complex Var. Theory Apply 8 (1987),
243–266.
[21] F. Sommen, Radon and X-ray transforms in Clifford analysis. Complex Var. Theory Apply 11 (1989), 49–70.
[22] F. Sommen, Spherical Monogenics on the Lie Sphere. Journal of Functional Analysis 92 (1990), 372–402.
[23] F. Sommen, Clifford analysis and integral geometry. Clifford Algebras and Their Applications in Mathematical
Physics, Kluwer, Dordrecht (1992), 293–311.
Department of Electronics and Information Systems, Faculty of Engineering and Architecture, Ghent
University, Krijgslaan 281, 9000 Gent, Belgium.
Email address: Denis.Constales@UGent.be
Department of Electronics and Information Systems, Faculty of Engineering and Architecture, Ghent
University, Krijgslaan 281, 9000 Gent, Belgium.
Email address: Hendrik.DeBie@UGent.be
Department of Electronics and Information Systems, Faculty of Engineering and Architecture, Ghent
University, Krijgslaan 281, 9000 Gent, Belgium.
Email address: Teppo.Mertens@UGent.be
Department of Electronics and Information Systems, Faculty of Engineering and Architecture, Ghent
University, Krijgslaan 281, 9000 Gent, Belgium.
Email address: Franciscus.Sommen@UGent.be