Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres
Convolution is an important tool in the construction of positive definite kernels on a manifold. This contribution provides conditions on an L²-positive definite and zonal kernel on the unit sphere of Cq in order that the kernel can be recovered as a generalized convolution root of an equally positi...
Збережено в:
| Опубліковано в: : | Symmetry, Integrability and Geometry: Methods and Applications |
|---|---|
| Дата: | 2015 |
| Автори: | , |
| Формат: | Стаття |
| Мова: | English |
| Опубліковано: |
Інститут математики НАН України
2015
|
| Онлайн доступ: | https://nasplib.isofts.kiev.ua/handle/123456789/147000 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Цитувати: | Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres / V.S. Barbosa, V.A. Menegatto // Symmetry, Integrability and Geometry: Methods and Applications. — 2015. — Т. 11. — Бібліогр.: 19 назв. — англ. |
Репозитарії
Digital Library of Periodicals of National Academy of Sciences of Ukraine| id |
nasplib_isofts_kiev_ua-123456789-147000 |
|---|---|
| record_format |
dspace |
| spelling |
Barbosa, V.S. Menegatto, V.A. 2019-02-12T18:14:53Z 2019-02-12T18:14:53Z 2015 Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres / V.S. Barbosa, V.A. Menegatto // Symmetry, Integrability and Geometry: Methods and Applications. — 2015. — Т. 11. — Бібліогр.: 19 назв. — англ. 1815-0659 2010 Mathematics Subject Classification: 33C55; 41A35; 41A63; 42A82; 42A85 DOI:10.3842/SIGMA.2015.014 https://nasplib.isofts.kiev.ua/handle/123456789/147000 Convolution is an important tool in the construction of positive definite kernels on a manifold. This contribution provides conditions on an L²-positive definite and zonal kernel on the unit sphere of Cq in order that the kernel can be recovered as a generalized convolution root of an equally positive definite and zonal kernel. The first author was partially supported by CAPES. The second one by FAPESP, under grant 2014/00277-5. Special thanks goes to the anonymous referees for the careful reading of the paper and for pointing corrections and suggestions that led to this final form of the paper. en Інститут математики НАН України Symmetry, Integrability and Geometry: Methods and Applications Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres Article published earlier |
| institution |
Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| collection |
DSpace DC |
| title |
Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres |
| spellingShingle |
Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres Barbosa, V.S. Menegatto, V.A. |
| title_short |
Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres |
| title_full |
Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres |
| title_fullStr |
Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres |
| title_full_unstemmed |
Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres |
| title_sort |
generalized convolution roots of positive definite kernels on complex spheres |
| author |
Barbosa, V.S. Menegatto, V.A. |
| author_facet |
Barbosa, V.S. Menegatto, V.A. |
| publishDate |
2015 |
| language |
English |
| container_title |
Symmetry, Integrability and Geometry: Methods and Applications |
| publisher |
Інститут математики НАН України |
| format |
Article |
| description |
Convolution is an important tool in the construction of positive definite kernels on a manifold. This contribution provides conditions on an L²-positive definite and zonal kernel on the unit sphere of Cq in order that the kernel can be recovered as a generalized convolution root of an equally positive definite and zonal kernel.
|
| issn |
1815-0659 |
| url |
https://nasplib.isofts.kiev.ua/handle/123456789/147000 |
| citation_txt |
Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres / V.S. Barbosa, V.A. Menegatto // Symmetry, Integrability and Geometry: Methods and Applications. — 2015. — Т. 11. — Бібліогр.: 19 назв. — англ. |
| work_keys_str_mv |
AT barbosavs generalizedconvolutionrootsofpositivedefinitekernelsoncomplexspheres AT menegattova generalizedconvolutionrootsofpositivedefinitekernelsoncomplexspheres |
| first_indexed |
2025-11-27T07:23:10Z |
| last_indexed |
2025-11-27T07:23:10Z |
| _version_ |
1850806381957873664 |
| fulltext |
Symmetry, Integrability and Geometry: Methods and Applications SIGMA 11 (2015), 014, 13 pages
Generalized Convolution Roots
of Positive Definite Kernels on Complex Spheres
Victor S. BARBOSA and Valdir A. MENEGATTO
Instituto de Ciências Matemáticas e de Computação, Universidade de São Paulo,
Caixa Postal 668, 13560-970, São Carlos - SP, Brasil
E-mail: victorrsb@gmail.com, menegatt@icmc.usp.br
Received October 16, 2014, in final form February 10, 2015; Published online February 13, 2015
http://dx.doi.org/10.3842/SIGMA.2015.014
Abstract. Convolution is an important tool in the construction of positive definite kernels
on a manifold. This contribution provides conditions on an L2-positive definite and zonal
kernel on the unit sphere of Cq in order that the kernel can be recovered as a generalized
convolution root of an equally positive definite and zonal kernel.
Key words: positive definiteness; zonal kernels; recovery formula; convolution roots; Zernike
or disc polynomials
2010 Mathematics Subject Classification: 33C55; 41A35; 41A63; 42A82; 42A85
1 Introduction
In this paper, we deal with a specific problem involving the so-called positive definite kernels.
These kernels have importance in the resolution of issues pertaining to many areas of mathe-
matics, including approximation theory, functional analysis, statistics, etc, presenting themselves
in both theory and applications. The concept of positive definiteness usually appears in two
formats as we now introduce.
Let X be a nonempty set and K a kernel on X, that is, a complex function K with domain
X ×X. The kernel K is positive definite if
k∑
m,n=1
cmcnK(xm, xn) ≥ 0,
for all positive integer k, all subsets {x1, x2, . . . , xk} of X and any complex numbers c1, c2, . . . , ck.
As for the second concept of positive definiteness, we need to assume two facts: X is endowed
with a measure µ so that (X,µ) is a σ-finite measure space and the kernel K needs to be an
element of L2(X × X,µ × µ). As usual, we will speak of the elements in this space as if they
were functions, with equality interpreted as equality µ-a.e.. A kernel K in L2(X ×X,µ× µ) is
L2-positive definite if∫
X
(∫
X
K(x, y)f(y)dµ(y)
)
f(x)dµ(x) ≥ 0, f ∈ L2(X,µ). (1.1)
If K is the integral operator on L2(X,µ) generated by K, the previous inequality corresponds
to 〈Kf, f〉2 ≥ 0, f ∈ L2(X,µ), in which 〈·, ·〉2 is the usual inner product of L2(X,µ).
In the cases in which the setting allows a comparison between the two concepts, it is not
hard to see that they do not coincide. If one can speak of continuity in X, then a chance of
coincidence increases considerably (see [3] for additional information on a possible equivalence).
In this paper, we prefer to deal almost exclusively with L2-positive definiteness since the source
of the main problem to be discussed comes from functional analysis.
mailto:victorrsb@gmail.com
menegatt@icmc.usp.br
http://dx.doi.org/10.3842/SIGMA.2015.014
2 V.S. Barbosa and V.A. Menegatto
If an L2-positive definite kernel K from L2(X×X,µ×µ) is available, a quite general problem
is the determination of a kernel S in L2(X ×X,µ× µ) satisfying the integral relation∫
X
S(x, ξ)S(ξ, y)dµ(ξ) = K(x, y), x, y ∈ X. (1.2)
If existence is guaranteed, the uniqueness of S cannot be usually reached unless additional
requirements are added (indeed, if S is a solution, then −S is another one).
In [3], the existence of a positive definite solution for this general problem was solved in
the case in which X is a metric space, the measure µ is strictly positive and K is continuous.
The solution S itself was constructed via the square root K1/2 of the integral operator K acting
on L2(X,µ) and generated by K. As a matter of fact, the operator K1/2 was an integral operator
on L2(X,µ) itself and the generating kernel of K1/2 was a solution S, equally continuous and
positive definite. The results proved in [3] agree with those previously surveyed in [13], the main
difference between them being the setting considered in each case.
If X has an enhanced structure, the original kernel K may have additional and desirable
features. Hence, the problem specializes to the following version: is there a solution kernel S
matching all the relevant additional properties the original kernel K may have? In some cases,
the left-hand side of (1.2) generalizes the notion of convolution. For instance, if X is a compact
two-point homogeneous manifold and µ is the usual “volume” measure on X, then the integral
in (1.2) is an extension of the concept of spherical convolution described in [9, 10] when the
kernel S is zonal (isotropic). In particular, the solution of the proposed problem boils down
to the finding of a convolution root S for K having the same features as K. This is relevant
on itself because (spherical) convolution is known to be an efficient tool in the construction of
positive definite kernels. We also observe that the case in which X is a sphere, is of particular
relevancy in statistics and approximation theory (see [2, 14, 19]).
Among the many research problems collected in [4] the readers can find open questions
aligned with the material covered in the references mentioned in the previous paragraph and
with the results to be developed in Sections 2 and 3 ahead. In those sections, we will chase
a solution for (1.2) in the case in which X is the unit sphere Ω2q in Cq, µ is the usual Lebesgue
measure in Ω2q, and K is zonal, following the direct procedure adopted in [19], a method that
contemplates a Fourier analysis perspective. The procedure itself encompasses these steps:
generalized convolutions preserve zonality; zonal kernels on Ω2q have specific expansions that
are necessary and sufficient for zonality; generalized convolution acts on these expansions by
squaring coefficients; the existence of zonal generalized convolution square roots is a question
of convergence of expansions; starting from general zonal kernels, one step of generalized convo-
lution generates a continuous zonal kernel. In particular, it should become clear to the reader
that the paper deals with a standard argument that features an operation (generalized convo-
lution), a symmetry property (zonality), and a representation (series expansion) that maintains
the symmetry and is invariant under the operation.
We believe the approach developed here can be adapted to many other situations, a typical
example being some compact two-point homogeneous manifolds. The setting and the notion
of generalized spherical convolution used in the paper, can be found in [6], where a multiplier
version of the Bernstein inequality on Ω2q was obtained. In order to direct the reader for potential
applications involving the analysis on Ω2q and perhaps, to the specific material covered here,
we mention [16, 17].
The spherical approach taken in the paper is probably less general than that adopted in
the Hilbert–Schmidt theory (that was the option in [3, 13]). For the convenience of the reader
and also for possible comparisons, that more general approach will be sketched in Section 4.
However, it is valuable to mention that the technique adopted here is more explicit and allows
the study of regularity properties of the convolution roots and of the kernel.
Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres 3
2 The spherical case plus zonality
Let σq denote the surface measure on the unit sphere Ω2q of Cq. Assuming K is a zonal and
L2-positive definite kernel on Ω2q, we will seek for a kernel S in L2(Ω2q × Ω2q, σq × σq), of the
same type as K, so that∫
Ω2q
S(x, ξ)S(ξ, y)dσq(ξ) = K(x, y), x, y ∈ Ω2q.
Classical results from the analysis on the sphere Ω2q and also basic Fourier analysis on Ω2q
will be needed along the way. In more than half of this section, we will detail some of them and
will prove some others not available in the literature. Recent references related to the material
to be described in this section are [1, 18] while classical ones are [5, 12].
The measure σq is invariant with respect to the group O(2q) of all unitary linear operators
on Cq in the following sense: if E is a σq-measurable subset of Ω2q and ρ is an element of O(2q)
then σq(ρ(E)) = σq(E). Zonality of a kernel K refers to the invariance property
K(ρ(x), ρ(y)) = K(x, y), x, y ∈ Ω2q, ρ ∈ O(2q).
We simplify notation by writing L2(Ω2q) := L2(Ω2q, σq), Ω2
2q := Ω2q × Ω2q and L2(Ω2
2q) :=
L2(Ω2
2q, σq×σq). The pertinent inner products in these Hilbert spaces will appear in normalized
form, that is,
〈f, g〉2 :=
1
ωq
∫
Ω2q
f(x)g(x)dσq(x), f, g ∈ L2(Ω2q),
and
〈K1,K2〉2 :=
1
ω2
q
∫
Ω2
2q
K1(x, y)K2(x, y)d(σq × σq)(x, y), K1,K2 ∈ L2
(
Ω2
2q
)
,
respectively, in which ωq := 2πq/(q − 1)! is the surface area of Ω2q.
The generalized convolution of two kernels K1 and K2 from L2(Ω2
2q) is the kernel K1 ∗ K2
given by the formula
(K1 ∗K2)(x, y) =
1
ωq
∫
Ω2q
K1(x, ξ)K2(ξ, y)dσq(ξ), x, y ∈ Ω2q. (2.1)
Clearly, the definition makes sense as long as the integral is well-defined.
From now on, in order to make the presentation clearer, we need to consider two separate
cases. For q ≥ 2, the zonality of K corresponds to the existence of a function K ′ : B[0, 1] → C
so that
K(x, y) = K ′(x · y), x, y ∈ Ω2q, (2.2)
in which · denotes the usual inner product in Cq and B[0, 1] := {z ∈ C : zz ≤ 1}. Thus, if K2
is zonal and we replace K1 with a function f that depends upon ξ only, then the convolution
K1 ∗K2 becomes the spherical convolution of K ′2 and f as defined in [7, 11].
Adapting arguments found in [8], one can see that a kernel K on Ω2q is L2-positive definite
and zonal if and only if the generating function K ′ appearing in (2.2) have a double series
representation of the form
K ′(z) =
∞∑
m,n=0
aq−2
m,n(K ′)Rq−2
m,n(z), z ∈ B[0, 1], (2.3)
4 V.S. Barbosa and V.A. Menegatto
in which aq−2
m,n(K ′) ≥ 0, m,n ∈ Z+, with convergence in L2(B[0, 1], νq−2), where
dνq−2(z) =
q − 1
π
(
1− |z|2
)q−2
dxdy, z ∈ B[0, 1].
The symbol Rq−2
m,n stands for the disk or generalized Zernike polynomial of bi-degree (m,n)
associated with the dimension q, that is,
Rq−2
m,n(z) := r|m−n|ei(m−n)θR
(q−2,|m−n|)
m∧n
(
2r2 − 1
)
, z = reiθ ∈ B[0, 1],
where R
(q−2,|m−n|)
m∧n is the Jacobi polynomial of degree m ∧ n := min{m,n} associated with the
pair of numbers q − 2 and |m− n|, normalized so that R
(q−2,|m−n|)
m∧n (1) = 1. Since
R
(q−2,|m−n|)
k =
P
(q−2,|m−n|)
k
P
(q−2,|m−n|)
k (1)
,
in which P
(q−2,|m−n|)
k is the classical Jacobi polynomial of degree k associated with q − 2 and
|m− n|, as defined in [15], then |Rq−2
m,n(z)| ≤ 1, z ∈ B[0, 1] and Rq−2
m,n(1) = 1 for all m and n. An
alternative formula for Rq−2
m,n via standard monomials is
Rq−2
m,n(z) =
m!n!(q − 2)!
(m+ q − 2)!(n+ q − 2)!
m∧n∑
j=0
(−1)j(m+ n+ q − 2− j)!
j!(m− j)!(n− j)!
zm−jzn−j .
The set {Rq−2
m,n : m,n ∈ Z+} is an orthogonal system in L2(B[0, 1], νq−2), that is,∫
B[0,1]
Rq−2
m,n(z)Rq−2
k,l (z)dνq−2(z) =
δmkδnl
hq−2
m,n
, (2.4)
where
hq−2
m,n =
m+ n+ q − 1
q − 1
(
m+ q − 2
q − 2
)(
n+ q − 2
q − 2
)
.
If K is a zonal kernel on Ω2q, we will write aq−2
m,n(K ′) to denote the Fourier coefficient of the gene-
rating function K ′, with respect to the orthogonal basis {Rq−2
m,n : m,n ∈ Z+} of L2(B[0, 1], νq−2):
aq−2
m,n(K ′) = hq−2
m,n
∫
B[0,1]
K ′(z)Rq−2
m,n(z)dνq−2(z), m, n ∈ Z+.
The complex Funk–Hecke formula in Ω2q establishes an intimate connection among everything
we have mentioned so far. If Y is a spherical harmonic of bi-degree (m,n) in q dimensions, that
is, Y is the restriction to Ω2q of a polynomial of bidegree (m,n) in Cq that belongs to the kernel
of the complex Laplacian in Cq, and K ′ is a function in L2(B[0, 1], νq−2), it states that [7, 11]
1
ωq
∫
Ω2q
K ′(x · y)Y (x)dσq(x) =
[ ∫
B[0,1]
K ′(z)Rq−2
m,n(z)dνq−2(z)
]
Y (y), y ∈ Ω2q,
as long as the integral on the left-hand side exists. In particular, the Funk–Hecke formula
simplifies to
1
ωq
∫
Ω2q
K ′(x · y)Y (x)dσq(x) =
aq−2
m,n(K ′)
hq−2
m,n
Y (y), y ∈ Ω2q.
Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres 5
We now move to the case q = 1. The zonality of a kernel K on Ω2 corresponds to the
existence of a function K ′ : Ω2 → C so that
K(x, y) = K ′(x · y), x, y ∈ Ω2, (2.5)
The characterization of the L2-positive definite and zonal kernels on Ω2 skips a little bit from the
previous representation in higher dimensions. In this particular case, the function K ′ appearing
in (2.5) have a series representation of the form
K ′(z) =
∞∑
k=−∞
ak(K
′)zk, z ∈ Ω2,
in which ak(K
′) ≥ 0, k ∈ Z, with convergence of the series in L2(Ω2). If one wants to speak
of orthogonality, integration needs to be done in accordance with the standard Fourier theory,
interpreting Ω2 as the quotient space R/2πZ, under the equivalence relation x ∼ y if and only
if x− y ∈ 2πZ. If q : R→ Ω2 is the quotient map, then (the invariant) integration on Ω2 reads
like ∫
Ω2
f(x)dσ2(x) :=
∫ 2π
0
f(q(θ))dθ,
in which dθ is the Lebesgue measure element on the interval [0, 2π). In this sense, if
Rm(z) := zm, z ∈ Ω2, m ∈ Z,
then {Rm : m ∈ Z} is an orthonormal basis of L2(Ω2q, σ2). Indeed, first observe that
1
ω2
∫
Ω2
Rm(z)Rn(z)dσ2(z) =
1
ω2
∫ 2π
0
ei(m−n)θdθ.
As so, if m = n then the integral on the right-hand side is just 1. Otherwise, we can pick
θ0 ∈ [0, 2π] so that ei(m−n)θ0 6= 1 and use a change of variables to justify the equalities
1
ω2
∫ 2π
0
ei(m−n)θdθ =
1
ω2
∫ 2π
0
ei(m−n)(θ+θ0)dθ =
1
ω2
ei(m−n)θ0
∫ 2π
0
ei(m−n)θdθ,
and arrive at
1
ω2
∫ 2π
0
ei(m−n)θdθ = 0.
To close the section, we will list some specific properties of the space
B2
(
Ω2
2q
)
:=
{
K ∈ L2
(
Ω2
2q
)
: K is zonal
}
,
some of them to be used ahead. The norm in the spaces L2(Ω2
2q) will be written as ‖ · ‖2. The
first one is of general interest and is included for completeness only. If {Kn} is a sequence
in B2(Ω2
2q) converging to K ∈ L2(Ω2
2q), ρ ∈ O(2q) and we write ρK to denote the kernel
(x, y) ∈ Ω2q × Ω2q → K(ρx, ρy), the zonality of each Kn plus the invariance of σq with respect
to ρ justify the inequality
0 ≤ ‖K − ρK‖2 ≤ ‖K −Kn‖2 + ‖Kn − ρK‖2
= ‖K −Kn‖2 + ‖ρKn − ρK‖2 = 2‖K −Kn‖2.
6 V.S. Barbosa and V.A. Menegatto
Letting n → ∞ leads to K = ρK in L2(Ω2
2q) and, consequently, we can conclude that B2(Ω2
2q)
is a closed subspace of L2(Ω2
2q).
To proceed, we will need two additional notations:
Zm,n(x, y) := Rq−2
m,n(x · y), m, n ∈ Z+, x, y ∈ Ω2q, q ≥ 2,
and
Zm(x, y) := Rm(x · y) = xmy−m, m ∈ Z, x, y ∈ Ω2.
Since q will remain fixed, the omission of the dimension q in both notations introduced above
should cause no confusion.
In the proposition below, we establish orthogonality properties of the sets {Zm,n : m,n ∈ Z+}
and {Zm : m ∈ Z} in the Hilbert space (B2(Ω2
2q), 〈·, ·〉2).
Proposition 2.1. The following assertions hold:
(i) (q ≥ 2) the Hilbert space B2(Ω2
2q) is isometrically isomorphic to L2(B[0, 1], νq−2);
(ii) (q ≥ 2) the set {(hq−2
m,n)1/2Zm,n : m,n ∈ Z+} is an orthonormal basis of B2(Ω2
2q);
(iii) the set {Zm : m ∈ Z} is an orthonormal basis of B2(Ω2
2).
Proof. Let us consider the case q ≥ 2 first. If K ∈ B2(Ω2
2q), the Funk–Hecke formula implies
that
‖K‖22 =
1
ω2
q
∫
Ω2
2q
K ′(x · y)K ′(x · y)d(σq × σq)(x, y)
=
∫
B[0,1]
K ′(z)K ′(z)dνq−2(z) = ‖K ′‖22,q,
in which ‖ · ‖2,q stands for the usual norm in the space L2(B[0, 1], νq−2). In particular, K ∈
B2(Ω2
2q)→ K ′ ∈ L2(B[0, 1], νq−2) is an isometry. This takes care of (i). Since {Rq−2
m,n : m,n ∈ Z+}
is an orthogonal basis in L2(B[0, 1], νq−2), the polarization identity reveals that its inverse image
by the isometry in (i) is an orthogonal basis in B2(Ω2
2q). Recalling (2.4), the orthonormality of
the set in (ii) follows. In order to see it is complete in B2(Ω2
2q), let K ∈ B2(Ω2
2q) and assume
that
1
ω2
q
∫
Ω2
2q
K(x, y)Rm,n(x, y)d(σq × σq)(x, y) = 0, m, n ∈ Z+.
Since that corresponds to
1
ωq
∫
Ω2q
(
1
ωq
∫
Ω2q
K ′(x · y)Rq−2
n,m(x · y)dσq(x)
)
dσq(y) = 0, m, n ∈ Z+,
an application of the Funk–Hecke formula leads to the equality(
hq−2
m,n
)−1
aq−2
m,n(K ′) = 0, m, n ∈ Z+,
that is,∫
B[0,1]
K ′(z)Rq−2
m,n(z)dνq−2(z) = 0, m, n ∈ Z+.
Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres 7
Since {Rq−2
m,n : m,n ∈ Z+} is an orthogonal basis of L2(B[0, 1], νq−2) and this space is complete,
then K ′ = 0 in L2(B[0, 1], νq−2). In particular, K = 0 in B2(Ω2
2q) and (ii) is proved. The
orthonormality assertion in (iii) is clear. As for completeness, let K ∈ B2(Ω2
2) and assume that
1
ω2
2
∫
Ω2
2
K(x, y)Zm(x, y)d(σ2 × σ2)(x, y) = 0, m ∈ Z.
Since this equality is precisely
1
4π2
∫
Ω2
(∫
Ω2
K ′(x · y)x−mdσ2(x)
)
ymdσ2(y) = 0, m, n ∈ Z,
it follows that
am(K ′)
∫
Ω2
y−mymdσ2(y) = 0, m ∈ Z,
in which am(K ′) is the m-th usual Fourier coefficient of K ′. In other words, am(K ′) = 0, m ∈ Z,
that is, K ′ = 0. �
We close the section detaching an obvious consequence of Proposition 2.1(ii).
Lemma 2.2 (q ≥ 2). If K ′ belongs to L2(B[0, 1], νq−2), then
∞∑
m,n=0
(
hq−2
m,n
)−1
aq−2
m,n(K ′)2 is con-
vergent.
3 Convolution roots in B2(Ω2
2q)
In this section, we finally analyze the problem described in the first paragraph of the previous
section. The first steps provide technical results on the convolution operation ∗ defined in (2.1),
in the case when one or both kernels are the basic elements of Proposition 2.1. The main results
of the paper come right after that.
Lemma 3.1. The following properties hold:
(i) the convolution of a hermitian kernel from L2(Ω2
2q) with itself is L2-positive definite on Ω2q;
(ii) if K1 and K2 belong to B2(Ω2
2q), then K1 ∗K2 does so.
Proof. Write K ∗ K to denote the integral operator on L2(Ω2q) generated by K ∗ K. If f ∈
L2(Ω2q) and K is a kernel on Ω2q, then a double application of Fubini’s theorem implies that
〈K ∗ K(f), f〉2 =
1
ω3
q
∫
Ω2q
∫
Ω2q
K(x, ξ)f(x)dσq(x)
∫
Ω2q
K(ξ, y)f(y)dσq(y)dσq(ξ).
If K is hermitian, the previous equality reduces itself to
〈K ∗ K(f), f〉2 =
1
ωq
∫
Ω2q
∣∣∣∣ 1
ωq
∫
Ω2q
K(x, ξ)f(x)dσq(x)
∣∣∣∣2dσq(ξ).
This takes care of (i). As for (ii), if ρ ∈ O2q, then
(K1 ∗K2)(ρ(x), ρ(y)) =
1
ωq
∫
Ω2q
K1(x, ρ∗ξ)K2(ρ∗ξ, y)dσq(ξ), x, y ∈ Ω2q,
in which ρ∗ is the adjoint of ρ. Since the measure σq is invariant with respect to elements of O2q,
it follows that
(K1 ∗K2)(ρx, ρy) =
1
ωq
∫
Ω2q
K1(x, ρ∗ξ)K2(ρ∗ξ, y)dσq(ρ
∗ξ) = (K1 ∗K2)(x, y), x, y ∈ Ω2q,
and K is zonal. �
8 V.S. Barbosa and V.A. Menegatto
The lemma below intends to produce formulas for convolutions involving the basis elements
of B2(Ω2
2q).
Lemma 3.2. The following formulas hold:
hq−2
k,l (Zm,n ∗ Zk,l) = δkmδlnZk,l, m, n, k, l ∈ Z+,
Zm ∗ Zn = δm,nZm, m, n ∈ Z.
Proof. In the case q ≥ 2, an application of the Funk–Hecke formula leads to
(Zm,n ∗ Zk,l)(x, y) =
1
ωq
∫
Ω2q
Rq−2
m,n(x · ξ)Rq−2
k,l (ξ · y)dσq(ξ)
=
1
ωq
∫
Ω2q
Rq−2
k,l (ξ · y)Rq−2
m,n(ξ · x)dσq(ξ)
=
aq−2
m,n
(
Rq−2
k,l
)
hq−2
m,n
Rq−2
m,n(y · x), x, y ∈ Ω2q.
It is now clear that
(Zm,n ∗ Zk,l)(x, y) =
δmkδnl
hq−2
m,n
Rq−2
m,n(x · y), x, y ∈ Ω2q.
Moving to the case q = 1, direct computation shows that
(Zm ∗ Zn)(x, y) =
1
ω2
∫
Ω2
Zm(x, ξ)Zn(ξ, y)dσ2(ξ)
= xmy−n
1
ω2
∫
Ω2
Rm(ξ)Rn(ξ)σ2(ξ), m, n ∈ Z+, x, y ∈ Ω2.
The use of the orthonormality of {Rm : m ∈ Z} in L2(Ω2) in the previous equation implies the
second formula in the statement of the lemma. �
Theorem 3.3. If K belongs to B2(Ω2
2q) then
K ∗ Zm,n =
aq−2
m,n(K ′)
hq−2
m,n
Zm,n = Zm,n ∗K, m, n ∈ Z+, q ≥ 2,
and
K ∗ Zm = am(K ′)Zm = Zm ∗K, m, n ∈ Z+, q = 1.
Proof. In the case q ≥ 2, both equalities follow from calculations via the Funk–Hecke formula.
As for the case q = 1, it can be done by direct calculations and a change of variables. �
The formulae in the theorem below are the key step towards the main results of this section.
Theorem 3.4. If K belongs to B2(Ω2
2q) then
〈K ∗K,Zm,n〉2 =
[
aq−2
m,n(K ′)
hq−2
m,n
]2
, m, n ∈ Z+, q ≥ 2,
and
〈K ∗K,Zm〉2 = am(K ′)2, m ∈ Z, q = 1.
Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres 9
Proof. Fix m,n ∈ Z+. Introducing the equality from Lemma 3.2, we see that
〈K ∗K,Zm,n〉2 =
1
ω2
q
∫
Ω2q
∫
Ω2q
(K ∗K)(ξ, η)Zm,n(η, ξ)dσq(ξ)dσq(η)
=
hq−2
m,n
ω2
q
∫
Ω2q
∫
Ω2q
(K ∗K)(ξ, η)(Zm,n ∗ Zm,n)(η, ξ)dσq(ξ)dσq(η).
Appealing to the definition of convolution and using Fubini’s theorem twice to interchange
integration orders, we deduce that
〈K ∗K,Zm,n〉2 =
hq−2
m,n
ω2
q
∫
Ω2q
∫
Ω2q
(K ∗ Zm,n)(x, y)(Zm,n ∗K)(y, x)dσq(x)σq(y).
Applying Theorem 3.3, we obtain
〈K ∗K,Zm,n〉2 =
(
hq−2
m,n
)−1
aq−2
m,n(K ′)2‖Zm,n‖22 =
(
hq−2
m,n
)−1
aq−2
m,n(K ′)2
∥∥Rq−2
m,n
∥∥2
2,q
,
while (2.4) yields the first equality in the statement of the theorem. The other equality is proved
in a similar manner. �
Theorem 3.5. Let K belong to L2(Ω2
2q). If K = J ∗ J for some J in B2(Ω2
2q), then
∞∑
m,n=0
hq−2
m,n|〈K,Zm,n〉2| <∞, q ≥ 2,
and
∞∑
m=−∞
|〈K,Zm〉2| <∞, q = 1.
Proof. If K = J ∗ J with J ∈ B2(Ω2
2q), then Lemma 3.1 implies that K ∈ B2(Ω2
2q) as well. In
the case q ≥ 2, that allows an application of Theorem 3.4 to deduce the equality
〈K,Zm,n〉2 =
[
aq−2
m,n(J ′)
hq−2
m,n
]2
, m, n ∈ Z+,
that is,
hq−2
m,n〈K,Zm,n〉2 =
[
aq−2
m,n(J ′)
(hq−2
m,n)1/2
]2
, m, n ∈ Z+.
Since {(hq−2
m,n)1/2Rq−2
m,n : m,n ∈ Z+} is an orthonormal basis of L2(B[0, 1], νq−2) and J ′ ∈
L2(B[0, 1], νq−2), we invoke Lemma 2.2 to conclude that
∞∑
m,n=0
hq−2
m,n|〈K,Zm,n〉2| <∞.
The case q = 1 can be handled in a similar manner. �
Needless to say that, due to Lemma 3.1, the modulus sign in the convergence outcome of the
previous theorem can be removed, as long as the kernel J is hermitian.
Theorem 3.5 suggests a criterion for the existence of convolution roots, therefore, a solution
to the problem described at the beginning of Section 2.
10 V.S. Barbosa and V.A. Menegatto
Theorem 3.6. Let K be a kernel in L2(Ω2
2q). If q ≥ 2, assume that all the Fourier coefficients
〈K,Zm,n〉2 are nonnegative and that
∞∑
m,n=0
hq−2
m,n〈K,Zm,n〉2 <∞.
Otherwise, assume that all Fourier coefficients 〈K,Zm〉2 are nonnegative and that
∞∑
m=−∞
〈K,Zm〉2 <∞.
Then, there exists an L2-positive definite kernel P in B2(Ω2
2q) such that K = P ∗P . In particu-
lar, K is an L2-positive definite element of B2(Ω2
2q).
Proof. We prove the theorem in the case q ≥ 2 only. Let us consider the zonal kernel P for
which
P ′ =
∞∑
m,n=0
hq−2
m,n〈K,Zm,n〉
1/2
2 Zm,n.
The expansion of P ′ with respect to the orthonormal basis {(hq−2
m,n)1/2Zm,n : m,n ∈ Z+}
of B2(Ω2
2q) is
P ′ =
∞∑
m,n=0
(
hq−2
m,n
)1/4〈K, (hq−2
m,n
)1/2
Zm,n〉1/22
(
hq−2
m,n
)1/2
Zm,n, m, n ∈ Z+.
In particular, our convergence assumption implies that P ′ belongs to B2(Ω2
2q). Our additional
assumptions and the characterization provided in (2.3) implies that P is L2-positive definite.
At last, a help of Theorem 3.4 leads to
〈P ∗ P,Zm,n〉2 =
[
aq−2
m,n(P ′)
hq−2
m,n
]2
= 〈K,Zm,n〉2, m, n ∈ Z+.
This suffices to conclude that P ∗ P = K in L2(Ω2
2q). �
The next result reveals that, in the setting we have adopted, the existence of a zonal convo-
lution root of the kernel implies the existence of a continuous one.
Proposition 3.7. Let K belong to L2(Ω2
2q). If K = J ∗ J for some J in B2(Ω2
2q), then K is in
fact a continuous kernel.
Proof. We know already that K ∈ B2(Ω2
2q). Hence, it can be written as
K(x, y) =
∞∑
m,n=0
aq−2
m,n(K ′)Rq−2
m,n(x · y). (3.1)
Since aq−2
m,n = hq−2
m,n〈K,Zm,n〉2, m,n ∈ Z+, it follows that
∞∑
m,n=0
∣∣aq−2
m,n(K ′)Rq−2
m,n(x · y)
∣∣ ≤ ∞∑
m,n=0
hq−2
m,n|〈K,Zm,n〉2|, x, y ∈ Ω2q.
Recalling Theorem 3.5, we can apply the Weierstrass M-test to conclude that the series in (3.1)
converges uniformly in Ω2
2q. Since each Rm,n is continuous, it follows that the same series defines
a continuous kernel. �
Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres 11
The next result is a consequence of the previous proposition and Theorem 2.3 in [3].
Corollary 3.8. Let K belong to L2(Ω2
2q). If K = J ∗ J for some J in B2(Ω2
2q), then K is
positive definite in the usual sense.
This a simplified version of Theorem 3.6.
Theorem 3.9. If K is a continuous, zonal and L2-positive definite kernel on Ω2q, then there
exists an L2-positive definite kernel P in B2(Ω2
2q) such that K = P ∗ P .
Proof. We only prove the assertion in the case q ≥ 2. Write K(x, y) = K ′(x · y), x, y ∈ Ω2q,
with K ′ as described in (2.3). Applying the Funk–Hecke formula, we see that
〈K,Zm,n〉2 =
1
ω2
q
∫
Ω2
2q
K ′(x · y)Rq−2
m,n(x · y)d(σq × σq)(x, y)
=
1
ωq
aq−2
m,n(K ′)
hq−2
m,n
∫
Ω2q
Rq−2
m,n(y · y)dσq(y), m, n ∈ Z+,
that is,
〈K,Zm,n〉2 =
aq−2
m,n(K ′)
hq−2
m,n
≥ 0, m, n ∈ Z+.
Furthermore,
∞∑
m,n=0
hq−2
m,n〈K,Zm,n〉2 =
∞∑
m,n=0
aq−2
m,n(K ′) <∞.
An application of Theorem 3.6 justifies the assertion of the theorem. �
We advise the reader that the results obtained in this section can be reproduced in some
other important settings in both, real and complex versions. As a matter of fact, the approach
can be developed as long as an L2 structure similar to the one used here is available in the
setting to be considered.
4 Final remarks: the solution via integral operators
Here, for the sake of completeness, we recall the functional analysis approach via Hilbert–
Schmidt operator usually used to solve a version of the general recovery problem (1.2). One
needs the separability of L2(X,µ), the continuity of K and, in addition, both concepts of positive
definiteness need to coincide. In this case, (1.1) holds and the integral operator K generated
by K is a Hilbert–Schmidt operator.
The classical spectral and Mercer’s theories guarantee the features below.
(i) The operator K is self-adjoint and Hilbert–Schmidt.
(ii) There exists a finite or countably infinite set of positive real numbers λ1(K) ≥ λ2(K) ≥
· · · ≥ 0 and a corresponding orthonormal system {φn} of L2(X,µ) so that
K(φn) = λn(K)φn, n = 1, 2, . . . .
(iii) The sequence {λn(K)} is square-summable and converges to 0.
12 V.S. Barbosa and V.A. Menegatto
(iv) The integral operator K has an L2 representation in the form
K(f) =
∞∑
n=1
λn(K)〈f, φn〉2φn, f ∈ L2(X).
(v) The kernel K has an expansion in the form
K(x, y) =
∞∑
n=1
λn(K)φn(x)φn(y),
with convergence in L2(X ×X,µ× µ).
All this information being available, the following classical result holds.
Theorem 4.1. If K is a continuous L2-positive definite kernel on X, then there exists an
L2-positive definite kernel S in L2(X ×X,µ× µ), which can be taken continuous, so that∫
X
S(x, ξ)S(ξ, y)dµ(ξ) = K(x, y), x, y ∈ X.
The following result is also pertinent.
Corollary 4.2. Under the setting and the conditions stated in Theorem 4.1, the square root K1/2
of K is the integral operator on L2(X,µ) generated by S.
In this general context, the kernels K and S end up having the same L2 expansion structure.
Hence, it is expectable that S will have the very same properties K has, as long as the properties
are attached to the Hilbert space structure of L2(X,µ). In a certain sense, the results in Section 3
ratify this in the spherical setting, at least when the property to be preserved is zonality. We
are unaware of specific papers addressing the analysis of a recovery property similar to the one
considered here on Ω2q via this Functional Analysis approach.
Acknowledgements
The first author was partially supported by CAPES. The second one by FAPESP, under grant
2014/00277-5. Special thanks goes to the anonymous referees for the careful reading of the paper
and for pointing corrections and suggestions that led to this final form of the paper.
References
[1] Bezubik A., Strasburger A., On spherical expansions of smooth SU(n)-zonal functions on the unit sphere
in Cn, J. Math. Anal. Appl. 404 (2013), 570–578, arXiv:1112.0648.
[2] Estrade A., Istas J., Ball throwing on spheres, Bernoulli 16 (2010), 953–970, arXiv:0810.4004.
[3] Ferreira J.C., Menegatto V.A., Eigenvalues of integral operators defined by smooth positive definite kernels,
Integral Equations Operator Theory 64 (2009), 61–81.
[4] Gneiting T., Strictly and non-strictly positive definite functions on spheres, Bernoulli 19 (2013), 1327–1349,
arXiv:1111.7077.
[5] Koornwinder T.H., The addition formula for Jacobi polynomials. II. The Laplace type integral representation
and the product formula, Math. Centrum Amsterdam, Report TW133, 1976.
[6] Levesley J., Kushpel A., A multiplier version of the Bernstein inequality on the complex sphere, J. Comput.
Appl. Math. 240 (2013), 184–191, arXiv:1204.6147.
[7] Menegatto V.A., Oliveira C.P., Annihilating properties of convolution operators on complex spheres, Anal.
Math. 31 (2005), 13–30.
http://dx.doi.org/10.1016/j.jmaa.2013.03.035
http://arxiv.org/abs/1112.0648
http://dx.doi.org/10.3150/09-BEJ241
http://arxiv.org/abs/0810.4004
http://dx.doi.org/10.1007/s00020-009-1680-3
http://dx.doi.org/10.3150/12-BEJSP06
http://arxiv.org/abs/1111.7077
http://dx.doi.org/10.1016/j.cam.2012.09.034
http://dx.doi.org/10.1016/j.cam.2012.09.034
http://arxiv.org/abs/1204.6147
http://dx.doi.org/10.1007/s10476-005-0002-5
http://dx.doi.org/10.1007/s10476-005-0002-5
Generalized Convolution Roots of Positive Definite Kernels on Complex Spheres 13
[8] Menegatto V.A., Peron A.P., Positive definite kernels on complex spheres, J. Math. Anal. Appl. 254 (2001),
219–232.
[9] Platonov S.S., Approximations on compact symmetric spaces of rank 1, Sb. Math. 188 (1997), 753–769.
[10] Platonov S.S., On some problems in the theory of the approximation of functions on compact homogeneous
manifolds, Sb. Math. 200 (2009), 845–885.
[11] Quinto E.T., Injectivity of rotation invariant Radon transforms on complex hyperplanes in Cn, in Integral
Geometry (Brunswick, Maine, 1984), Contemp. Math., Vol. 63, Amer. Math. Soc., Providence, RI, 1987,
245–260.
[12] Rudin W., Function theory in the unit ball of Cn, Classics in Mathematics, Springer-Verlag, Berlin, 2008.
[13] Schaback R., A unified theory of radial basis functions: Native Hilbert spaces for radial basis functions II,
J. Comput. Appl. Math. 121 (2000), 165–177.
[14] Schreiner M., Locally supported kernels for spherical spline interpolation, J. Approx. Theory 89 (1997),
172–194.
[15] Szegő G., Orthogonal polynomials, American Mathematical Society, Colloquium Publications, Vol. 23,
4th ed., Amer. Math. Soc., Providence, R.I., 1975.
[16] Thirulogasanthar K., Saad N., Honnouvo G., 2D-Zernike polynomials and coherent state quantization of
the unit disc, arXiv:1303.5483.
[17] Torre A., Generalized Zernike or disc polynomials: an application in quantum optics, J. Comput. Appl.
Math. 222 (2008), 622–644.
[18] Wünsche A., Generalized Zernike or disc polynomials, J. Comput. Appl. Math. 174 (2005), 135–163.
[19] Ziegel J., Convolution roots and differentiability of isotropic positive definite functions on spheres, Proc.
Amer. Math. Soc. 142 (2014), 2063–2077, arXiv:1201.5833.
http://dx.doi.org/10.1006/jmaa.2000.7264
http://dx.doi.org/10.1070/SM1997v188n05ABEH000233
http://dx.doi.org/10.1070/SM2009v200n06ABEH004021
http://dx.doi.org/10.1090/conm/063/876322
http://dx.doi.org/10.1016/S0377-0427(00)00345-9
http://dx.doi.org/10.1006/jath.1997.3037
http://arxiv.org/abs/1303.5483
http://dx.doi.org/10.1016/j.cam.2007.12.009
http://dx.doi.org/10.1016/j.cam.2007.12.009
http://dx.doi.org/10.1016/j.cam.2004.04.004
http://dx.doi.org/10.1090/S0002-9939-2014-11989-7
http://dx.doi.org/10.1090/S0002-9939-2014-11989-7
http://arxiv.org/abs/1201.5833
1 Introduction
2 The spherical case plus zonality
3 Convolution roots in B2(2q2)
4 Final remarks: the solution via integral operators
References
|