Professional Documents
Culture Documents
Holger Rauhut
University of Vienna, Faculty of Mathematics, NuHAG
Nordbergstr. 15, A-1090 Vienna, Austria
rauhut@ma.tum.de
Abstract
The concept of semi-bounded generalized hypergroups (SBG hypergroups) is devel-
oped which are more special then generalized hypergroups introduced by Obata and
Wildberger and which are more general then discrete hypergroups or even discrete signed
hypergroups. The convolution of measures and functions is studied. In case of commu-
tativity we define the dual objects and prove some basic theorems of Fourier analysis.
Furthermore, we investigate the relationship between orthogonal polynomials and gener-
alized hypergroups. We discuss the Jacobi polynomials as an example.
Mathematics subject classification: 43A62, 43A99, 46J10, 05E35, 33C80
Keywords: generalized hypergroup, semi-bounded generalized hypergroup, bounded
generalized hypergroups, signed hypergroup, discrete hypergroup, convolution, dual ob-
ject, Fourier transform, orthogonal polynomials, Jacobi polynomials
1 Introduction
Locally compact hypergroups were independently introduced around the 1970s by Dunkl [4],
Jewett [7] and Spector [18]. They generalize the concepts of locally compact groups with the
purpose of doing standard harmonic analysis. Similar structures had been studied earlier in
the 1950s by Berezansky and colleagues, and even earlier in works of Delsarte and Levitan.
Later on results of harmonic analysis on hypergroups were transferred to different applica-
tions. For example a Bochner theorem is used essentially in the context of weakly stationary
processes indexed by hypergroups, see [10] and [12]. Hypergroup structure is also heavily used
in probability theory, see the monograph [2], and in approximation with respect to orthogonal
polynomial sequences, see [5] and [11]. However, not the whole set of axioms (see [2]) is used in
these application areas. So concentrating on orthogonal polynomials, Obata and Wildberger
1
2 Lasser, Obermaier and Rauhut
studied in [13] a very general concept and called it generalized hypergroups. The purpose
of the present paper is to derive results of harmonic analysis for generalized hypergroups in
more detail than in [13]. Our main interest is to include all orthogonal polynomial systems
with respect to a compactly supported orthogonalization measure in our investigations.
Proof: For (3)(5) see [13, Lemma 1.1]. Now, on the one hand we have (cn cm )cl =
P P
k,j g(n, m, k)g(k, l, j)cj and on the other hand cn (cm cl ) = k,j g(m, l, k)g(n, k, j)cj . From
the associativity of A0 and from the linear independence of the set K follows (6).
Lemma 2.3 For f, g with finite support and all n K it holds that
X X X
Ln f (m)g(m)h(m) = f (m)Lm g(n)h(m) = f (m)(Ln g)(m)h(m). (7)
m m m
We write (k) = ({k}) for a discrete measure on K. Let n denote the Dirac-measure
at n K, i.e., n (k) = 1 if k = n and n (k) = 0 else.
Definition 2.4 A positive discrete measure 6= 0 on K is called (left) Haar measure if for
all f with finite support and all n K it holds
X X
Ln f (m)(m) = f (m)(m).
m m
Theorem 2.5 A Haar measure exists if and only if K is normalized. In that case all Haar
measures are determined by = h, > 0.
Proof: Let us assume that there exists a Haar measure . Due to (A3) we get
X X X
h(n)1 (n) = g(n, m, 0)(m) = Ln 0 (m)(m) = 0 (m)(m) = (0),
m m m
which yields (n) = (0)h(n). Now, let = h. It suffices to consider f = k . By (5) we get
X X X X
Ln k (m)(m) = (k)g(k, n, m) = k (m)(m) g(k, n, m).
m m m m
In order to develop their theory further Obata and Wildberger took care of the functional
0 : A0 C defined by
!
X
0 n cn = 0 ,
n
4 Lasser, Obermaier and Rauhut
and focused on the following property. A generalized hypergroup (K, A0 ) is said to satisfy
property (B) if for all n there exists (n) 0 such that
|0 (b cn b)| (n)0 (b b) for all b A0 .
We focus on a stronger property than Obata and Wildberger.
An SBG hypergroup is satisfying property (B) with (n) = (n), see [13, Theorem 4.1].
It holds
(n) max(h(n)1 , 1).
By simple arguments we have
X
(m) = sup |g(n, m, k)|. (9)
n
k
If K is hermitian or commutative then (n) = (n), and if K is positive and normalized then
(n) = 1 for all n.
whenever the sum on the right hand side is finite for all k. A short calculation shows 0 =
0 = , i.e., 0 is the unit element for this convolution. For two Dirac measures we get
P
n m = k g(n, m, k)k , and supp n m is finite.
In order to investigate the convergence of the sum in (10) we introduce the spaces
( )
X
M (K) = measure on K, ||(K) = |(n)| < , kk = ||(K),
n
( )
X
M (K) = M (K), ||(K) = |(n)|(n) < , kk = ||(K).
n
Lemma 3.1 (i) If M (K) and M (K) then M (K) and k k kk kk.
(ii) If M (K) and M (K) then M (K) and k k kkkk .
Proof: Using (5) we obtain (i), and application of axiom (A3) gives (ii). For (v) we have
P
by definition kn m k = k |g(n, m, k)| (n). The second inequality is achieved analo-
gously by using (9). The assertions (iii) and (iv) are clear.
Now, we are able to compare the concept of an SBG hypergroup with that of a discrete
hypergroup, see for example [5], or a discrete signed hypergroup, see [14]. Our previous
results give the following theorem.
Theorem 3.4 (i) If K is a real, normalized and bounded generalized hypergroup then its
index set K with convolution as defined in (10) and involution as defined in (1) is a
discrete signed hypergroup.
(ii) If K is a positive and normalized SBG hypergroup then its index set K with convolution
and involution is a discrete hypergroup.
(iii) Let (K, ,) be a discrete signed hypergroup. Put K = {k , k K} and let A0 be the
vector space of all finite linear combinations of Dirac measures k K. Further, let be
the multiplication in A0 and put k = k as involution on K, which is linearly extended
to A0 . Then (K, A0 ) is a real, bounded and normalized generalized hypergroup.
6 Lasser, Obermaier and Rauhut
(iv) If (K, ,) is a discrete hypergroup then the construction in (iii) yields a positive and
normalized SBG hypergroup.
Definition 3.5 Let f and g be functions on K with finite support. The convolution of those
functions is defined by
X
(f g)(m) = f (n)(Ln g)(m)h(n). (11)
n
Lemma 3.6 If f and g have finite support then f g has finite support.
P
Proof: By definition (f g)(m) = n,k f (n)g(n, m, k)g(k)h(n). Hence, supp f g
S
nsupp f,ksupp g Mn,k , with Mn,k = {m, g(n, m, k) 6= 0}. According [13, Lemma 1.2] the set
Mn,k = {m, g(n, m, k) 6= 0} is finite for all n, k.
Theorem 3.7 If f, g are functions on K with finite support then (f g)h = (f h) (gh).
If K is commutative, then is commutative and by the last lemma we see that then also
is commutative.
For a positive discrete measure on K and 1 p < we introduce the Banach spaces
( ) !1/p
X X
p p p
l () = f : K C, |f (n)| (n) < , kf kp, = |f (n)| (n) ,
n n
l = f : K C, sup |f (n)| < , kf k = sup |f (n)|.
n n
Theorem 3.9 The convolution in (11) extends to l1 (h)l1 (h) and kf gk1,h kf k1,h kgk1,h .
Proof: First, assume f, g to have finite support and a such that a(k)(f g)(k) = |(f g)(k)|.
Theorem 3.7 yields
X
k(f g)k1,h = |(f g)(k)|h(k) = (f g)h(a) = |(f h) (gh)(a)|
k
X
(n)|g(m)||f (n)|h(n)h(m) = kf k1,h kgk1,h .
n,m
By using (9) the convolution extends quite analogous to l1 (h) l1 (h) with kf gk1,h
kf k1,h kgk1,h . If K is bounded, i.e., (n) M for all n, then the last theorem gives
kf gk1,h M kf k1,h kgk1,h . For f l1 (h) define Lf g = f g. Clearly, Lf is then a
bounded operator on l1 (h) and kLf k M kf k1,h . With the norm kf k = kLf k it holds
kf gk kf k kgk . Hence, if K is bounded, then (l1 (h), k k , ) is a Banach algebra.
Proof: Since f l1 (h) the right hand side of (12) exists by Theorem 3.9 and
X
h(n)1 (n f )(m) = h(n)1 n (k)(Lk f )(m)h(k) = (Ln f )(m).
k
Lemma 3.13 For 1 p and (1/p + 1/q = 1) it holds kLn f kp,h (n)1/p (n)1/q kf kp,h .
Proof: The proof is done by using (12) and Theorem 3.12.
If K is hermitian or commutative inequality (15) becomes |(f g)(m)| (m)kf kp,h kgkq,h .
In this case we introduce the Banach space
|f (n)| |f (n)|
l () = f : K C, sup < , kf k, = sup . (16)
n (n) n (n)
4 Dual objects
We say that a generalized hypergroup (K , A0 ) is a function realization, if A0 is a dense
subalgebra of the space C(S), where S is a compact Hausdorff space. By using Gelfand
theory, Obata and Wildberger proved that for commutative generalized hypergroups (K, A0 )
satisfying (B) there is an isomorphism a a onto a function realization (K , A0 ). Moreover,
there is a positive Radon measure on S with supp = S, (S) = 1 and
Z
0 (a) = a (x)d(x) for all a A0 ,
S
and K is a complete orthogonal set for L2 (S, ), see [13, Theorem 5.1].
From now on, we assume (K, A0 ) to be commutative and A0 to be a dense subalgebra of
C(S) for some compact Hausdorff space S. The condition (B) now reads
Z Z
cn (x)|b(x)|2 d(x) (n) |b(x)|2 d(x) = (n)kbk2L2 (S,) for all b C(S),
S S
condition (B) is satisfied. The next lemma states that (n) cannot be chosen smaller.
Lemma 4.1 Let (K, A0 ) satisfy condition (B) with constants (n). Then
Proof: Let us first remark that L2 (S, ) is the completion of A0 with respect to k k2,
since K is a complete orthogonal set for L2 (S, ). The inequality
Z Z
cn (x)|b(x)|2 d(x) (n) |b(x)|2 d(x) (17)
S S
is hence valid even for all b L2 (S, ). Now, let x0 S and choose a family of neighborhoods
(Vi )iI of x0 such that Vi {x0 }. Further letR bi = Vi /kVi k2, where Vi denotes the char-
acteristic function of the set Vi . Clearly limi S cn (x)|bi (x)|2 d(x) = cn (x0 ). Since, x0 S is
arbitrarily chosen, inserting into (17) gives the assertion. Further, notice that (n) = (n) is
a valid choice by [13, Theorem 4.1].
Now, let us consider dual objects of commutative generalized hypergroups. Obata and
Wildberger already have defined characters [13, p. 74], but their definition seems to be too
weak in order to develop harmonic analysis.
10 Lasser, Obermaier and Rauhut
The elements of X b (K) are called characters and the elements of K hermitian characters.
Consider now an element x of S. It is easily seen that x (n) = cn (x) defines an element
of K. Hence K 6= . Since A0 is dense in C(S) and S is a compact Hausdorff space it follows
that for different x, y S we obtain different characters x 6= y , see also [13, Theorem 6.4].
Thus, we can identify S with a subset of K and we get the following inclusion relations
S K X b (K) . (18)
The latter relation is well known for hypergroups and signed hypergroups. In contrast to
the group case, these inclusions may be proper, as is illustrated by some known examples for
hypergroups.
From (n) = L0 (n) = (0)(n) it follows (0) = 1. Furthermore, since (0) = 1 it holds
kk, 1. By Lemma 4.1 |cn (x)| (n), which implies kx k, = 1 for all x S.
For r 1 let us define the following subsets of the duals
n o
Xrb (K) = X b (K), kk, r ,
n o
Kr = K, kk, r .
If K is bounded then X b (K) = XRb (K) and K = KR , where R = supn (n) . In fact, in
that case l () = l setwise and for a character l it holds
Taking the supremum over all n K yields kk supn (n). Since (n) 1 we further
deduce
1 kk, kk R. (20)
We equip X b (K) with the topology of pointwise convergence and subsets of X b (K) with
the induced topologies. With these topologies the functions sn : X b (K) C, sn () = (n)
and their restrictions to the other duals are continuous. We only state without a proof that
the Gelfand topology on S is the topology induced by X b (K), i.e., the topology of pointwise
convergence.
5 Fourier transform
Now, due to our dual objects we are able to perform some Fourier analysis in the context of
commutative SBG hypergroups.
Generalized hypergroups 11
Definition 5.1 For M (K) we introduce the following two versions of the Fourier-
Stieltjes-transform by
X
() = (n)(n) for K,
n
X
F()() = (n)(n) for X b (K).
n
P
For x S K we write (x) = (x ) = n cn (x)(n).
Since the functions sn () = (n) are continuous on Xrb (K) for fixed n it follows that F() is
continuous on Xrb (K).
Definition 5.3 For f l1 (h) we define two versions of the Fourier transform by
X
d
f() = (f h)() = f (n)(n)h(n) for K,
n
X
F(f )() = F(f h)() = f (n)(n)h(n) for X b (K).
n
P
For x S we write f(x) = f(x ) = n f (n)cn (x)h(n).
By interpreting measures on K as functions on K we clearly have l1 (h) = {f, f h
M (K)} and hence, Lemma 5.2 immediately implies that the Fourier transform is continuous
on Xrb (K) for all r 1 and for Xrb (K) it holds
|f()| kf k1,h . (21)
p
In order to define the Fourier transform for f l2 (h) we remark that { h(n)cn , n K}
is a complete orthonormal set for L2 (S, ), see [13, Corollary 3.4]. Therefore, the series
P 2
n f (n)cn h(n) converges in L (S, ) by Parsevals identity
Z
2
X
X
X
2
| f (n)cn (x)h(n)| d(x) =
f (n)cn h(n)
= |f (n)|2 h(n) = kf k22,h . (22)
S n
n
n
2,
where convergence of the sum is understood in L2 (S, ). In (22) we already proved Plancherels
theorem.
12 Lasser, Obermaier and Rauhut
Theorem 5.4 The Fourier transform is an isometric isomorphism from l2 (h) into L2 (S, ),
in particular for f l2 (h) it holds kfk2, = kf k2,h .
which means f l2 (h). The Fourier transform on l1 (h) coincides with the one on l2 (h)
-almost everywhere and by Plancherels theorem kf k2,h = kfk2, = kF(f )|S k2, = 0. We
therefore obtain f = 0.
Let us turn our attention now to the relation of Fourier transform and convolution.
Proof: First suppose f, g l1 (h) such that f g l1 (h). Using Plancherels theorem 5.4,
Theorem 5.6 and (21) we obtain
kf gk2,h = kf[
gk2, = kfgk2, kfk,S kgk,S kf k1,h kgk1,h .
Note, that implicitly we used the commutativity of K in this proof. Immediately, we obtain
that the convolution Theorem 5.6 holds for all f, g l1 (h) with the slight adjustment that
in general (23) holds only for -almost all X b (K).
An involution on l1 (h) is given by f (n) = f (n), which is preserved by the Fourier
transform on K, i.e.,
X X
fc () = f (n)(n)h(n) = f (n)(n)h(n) = f(n) for all K.
n n
We can even extend this definition to a larger space. Let M (S) denote the space of complex
bounded Radon measures on S with the total variation as norm. For M (S) we define
the inverse Fourier-Stieltjes transform by
Z
(n) = cn (x)d(x), for all n K.
S
Theorem 5.9 For the inverse Fourier-Stieltjes transform the following is true.
(i) For M (S) we have l () and kk, kk.
(iii) For F L1 (S, ) we have F c0 () where c0 () denotes the closure with respect to
k k, of the set of all functions with finite support. Furthermore, the image of the
inverse Fourier transform of L1 (S, ) is dense in c0 ().
yielding kf k, kf k2,h . Note that we hereby derived l2 (h) l (). Now, using this
estimation we obtain kG k, kG k2,h /2 and further
|F (k) (k)| |F (k) G(k)| + |G(k) (k)|
(k)(kF Gk1, + /2) (k) ,
which is equivalent to kF k, . Hence, F can be approximated with respect to k k,
by functions with finite support. Since all function with finite support are contained in the
image of the inverse Fourier transform of L1 (S, ) the image of L1 (S, ) is dense in c0 ().
Observe, that the last result generalizes the Riemann-Lebesgue lemma. We also have a
uniqueness theorem for the inverse Fourier transform.
Proof: Assume that 6= 0 but = 0. By [13, Theorem 5.1] A0 = {f|S , | supp f | < }
is a dense subalgebra of C(S). Hence, there is some f with finite support such that
Z
f(x)d(x) 6= 0.
S
However, we have
Z X Z X
f(x)d(x) = f (n) cn (x)d(x)h(n) = f (n)(n)h(n) = 0.
S n S n
i.e., = 0 . Another important property was shown in the proof of Theorem 5.10 above.
Suppose f has finite support and M (S). Then
Z X
f(x)d(x) = f (n)(n)h(n). (25)
S n
Proof: For = f we already know by Theorem 5.8 that = (f) = f . Now suppose
f = and let g have finite support. With (25) and (g) = g we obtain
Z Z X X Z
g(x)f(x)d(x) = g(x) f (n)cn (x)h(n)d(x) = f (n) g(x)cn (x)d(x)h(n)
S S n n S
X Z
= f (n) g(x)cn (x)d(x)h(n)
n S
X X Z
= f (n)g(n)h(n) = g(n)(n)h(n) = g(x)d(x).
n n S
Generalized hypergroups 15
(ii) Let F L1 (S, ) such that F l1 (h). Then for -almost every x S it holds
X
F (x) = F (n)cn (x)h(n) . (26)
n
Proof: (i) follows by Theorem 5.8(i). For (ii) put = F . Then = F l1 (h). With
Theorem 5.11 it holds = (F ) which is equivalent to F = (F ) in L1 (S, ). Since the right
hand side of (26) is continuous, equality holds for all x S if F is continuous.
with P0 (x) = 1 and P1 (x) = (x b)/a, where the coefficients are real numbers with c1 > 0,
cn an1 > 0, n > 1. Conversely, if we define (Pn )n=0 by (27) there is a measure with the
assumed properties, see [3].
The linearization coefficients g(n, m, k) are uniquely defined by
X n+m
X
Pn Pm = g(n, m, k)Pk = g(n, m, k)Pk . (28)
k=0 k=|nm|
The linearization coefficients are obtained recursively based on the coefficients of the tree
term recurrence relation.
(i)
an+m1 am am+1 an+m1
g(n, m, n + m) = g(n 1, m, n + m 1) = , (29)
an1 a1 a2 an1
cmn+1 cm cm1 cmn+1
g(n, m, m n) = g(n 1, m, m n + 1) = , (30)
an1 a1 a2 an1
(ii)
bn+m1 bn1
g(n, m, n + m 1) = g(n 1, m, n + m 1)
an1
an+m2
+ g(n 1, m, n + m 2) ,
an1
bmn+1 bn1
g(n, m, m n + 1) = g(n 1, m, m n + 1)
an1
cmn+2
+ g(n 1, m, m n + 2) ,
an1
which implies the recurrence formulas (i)-(iii). The second equations in (i) are proven by
induction.
We easily derive
Z 1 Qn1
1 ai
h(n) = g(n, n, 0) = Pn2 (x) d(x) = Qi=1
n . (31)
i=1 ci
Let K = {Pn , n N0 }, A0 be the set of polynomials with complex coefficients in one real
variable and be the complex conjugation .
Generalized hypergroups 17
(iii) (K, A0 ) is an SBG hypergroup if and only if the sequences (an ), (bn ) and (cn ) are
bounded.
P
Now, we are looking for an OPS (Rn )nN0 with k gR (n, m, k) = 1 for all n, m N0 ,
which is equivalent to the existence of x0 R with Rn (x0 ) = 1 for all n N0 .
Theorem 6.4 Suppose to have compact support and let (Pn )nN0 be an arbitrary orthogonal
polynomial sequence with respect to . Denote by [d, e] the smallest interval containing S.
Choose x0 R \ (d, e) and define Rn (x) = Pn (x)/Pn (x0 ), n N0 . Then K = {Rk , k N0 } is
a normalized SBG hypergroup.
18 Lasser, Obermaier and Rauhut
Proof: Let (Qn )nN0 be the monic orthogonal polynomials with respect to as in the
proof of Lemma 6.3. Then
Qn+1 (x0 )
0 < 1, n 0.
(x0 bn b )Qn (x0 )
Hence, |a
n | < |(x0 bn b )/Q1 (x0 )|, which shows the boundedness of (an ). Finally,
an + bn + cn = 1 yields the boundedness of (cn ). By Theorem 6.2 (iii) the proof is complete.
Now, let us examine the duals of an SBG polynomial hypergroup. We define the sets
[
Dr = {z C, |Pn (z)| r(n) for all n N0 }, D = Dr , (32)
r1
[
Drs = Dr R and D s = Drs . (33)
r1
Furthermore we define for some z C the function z (n) = Pn (z) for all n N0 . Then the
following theorem holds.
D X b (N0 ), z 7 z and c0 ,
Ds N x 7 x
are homeomorphisms.
c0 are bounded.
(iii) X b (N0 ) and N
We would like to mention the the question wether the dual of an SBG polynomial hyper-
group is compact is still open.
7 Jacobi polynomials
(,)
The Jacobi polynomials Pn are orthogonal with respect to the measure
with supp = [1, 1] = S. According to Theorem 6.4 they form a normalized SBG polyno-
mial hypergroup when normalizing at a point x0 / (1, 1).
In case x0 = 1 the three term recurrence relation coefficients are given by
2( + 1)
a = , b = ,
++2 ++2
(n + + + 1)(n + + 1)( + + 2)
an = ,
(2n + + + 2)(2n + + + 1)( + 1)
( + + 2)( + )
bn = 1 ,
2( + 1) (2n + + + 2)(2n + + )
n(n + )( + + 2)
cn = ,
(2n + + + 1)(2n + + )( + 1)
(,)
see [9]. The corresponding normalized polynomials are denoted by Rn and we compute
(,)
where a = + + 1 and b = , then (Rn )nN0 constitutes a discrete hypergroup, see
[6, Theorem 1].
By switching the normalization point x0 to 1 and denoting the corresponding polyno-
(,) (,) (,)
mials by Sn we have Sn (x) = Rn (x), see also [6, p.585]. Hence, when (, ) V
(,)
then Sn constitute a discrete signed hypergroup and when (, ) W they form a discrete
hypergroup.
The remaining region is G = {(, ), 1 < , < 1/2}. Making use of Theorem
(,)
6.4, (34) and (n) max(h(n)1 , 1) we deduce for (, ) G both {Rn , n N0 } and
(,)
{Sn , n N0 } form an SBG polynomial hypergroup which is not bounded.
For the ultraspherical polynomials, i.e., = we will determine (n) explicitly for
1 < < 1/2.
(,)
Theorem 7.1 Let 1 < < 1/2. For corresponding with Rn it holds (0) = (1) = 1
and for n 2
2n1 n
!
X 2 Y Y
(n) = |g(n, n, k)| = Qn1 ak + ck 1 < 4h(n)1 1.
k k=1 ak k=n k=1
Proof: Make use of Theorem 6.4, (34) and (n) max(h(n)1 , 1) to show the correspon-
dence with an SBG hypergroup which is not bounded.
It is clear that (0) = (1) = 1. We use Lemma 6.1 to deduce for n m 2 that
g(n, m, n m + 2j 1) = 0, j = 1, 2, . . . , m, g(n, m, n m), g(n, m, n m) > 0 and
g(n, m, n m + 2j) < 0, j = 1, 2, . . . , m 1. Hence for all n, m 2
X
|g(n, m, k)| = 2(g(n, m, |n m|) + g(n, m, n + m)) 1.
k
Finally, we derive
2n1 n
!
X 2 Y Y
(n) = |g(n, n, k)| = Qn1 ak + ck 1 for all n 2.
k k=1 ak k=n k=1
Generalized hypergroups 21
Q 2n1 Qn
By using k=n ak < k=1 ck this yields h(n)1 (n) < 4h(n)1 1. With (34) we get the
last assertion.
Now it is easy to determine the dual objects of the generalized hypergroups generated by
ultraspherical polynomials.
(,)
Theorem 7.2 Let 1 < . Then the duals of the generalized hypergroup {Rn , n N0 }
of ultraspherical polynomials coincide,
c0 = X b (N0 ) [1, 1] .
S = N
Proof: We have to show S X b (K). Assume z C \ [1, 1]. From [16, (8.21.9)] we
deduce that Rn (z) grows exponentially with n. From (35) we know that (n) grows only
polynomially. Hence, there does not exist a constant r such that |Rn (z)| r(n) which
/ X b (N0 ).
means z
One might have the question what happens in the case (, ) G when choosing the
normalization point c
/ [1, 1]. Surprisingly, [1, Theorem 2] immediately yields the following
theorem.
(,)
Theorem 7.3 Let (, ) G and choose c R \ [1, 1]. The Jacobi polynomials Tn nor-
(,)
malized at c (i.e., Tn (c) = 1) constitute a normalized and bounded generalized hypergroup,
i.e., a discrete signed hypergroup. The duals are given by
p p
S = [1, 1], c0 = [|c|, |c|],
N X b (N0 ) = {z C, |z + z 2 1| |c| + c2 1}.
Hereby, the branch of z 2 1 is chosen such that |z + z 2 1| 1.
References
[1] R. Askey, G. Gasper, Linearization of the product of Jacobi polynomials III, Can.
J. Math. 23 (1971), 332338.
[4] C. Dunkl, The measure algebra of a locally compact hypergroup, Trans. Amer.
Math. Soc. 179 (1973), 331348.
22 Lasser, Obermaier and Rauhut
[6] G. Gasper, Linearization of the product of Jacobi polynomials II, Can. J. Math. 22
(1970), 582593.
[8] R. Larsen, Banach Algebras, an Introduction, Marcel Dekker Inc., New York, 1973.
[9] R. Lasser, Orthogonal polynomials and hypergroups, Rend. Mat. 3 (1983), 185
209.
[14] K.A. Ross, Signed hypergroups a survey, Cont. Math. 183 (1995), 319329.
[16] G. Szego, Orthogonal polynomials, Amer. Math. Soc., New York, 1959.
[18] R. Spector, Apercu de la Theorie des Hypergroupes, Lecture Notes in Math., Vol.
497 (Analyse Harmonique sur les Groupes de Lie, Sem. Nancy-Strasbourg 1973
1975), Springer, Berlin, 1975.
[19] W. van Assche, Orthogonal polynomials in the complex plane and on the real line,
in: M.E.H. Ismail et al. (Eds.), Special Functions, q-Series and Related Topics, pp.
211245, Fields Institute Communications 14, Amer. Math. Soc., 1997.
[20] W. van Assche, Asymptotic properties of orthogonal polynomials from their recur-
rence Formula I, J. Approx. Theory 44 (1985), 258276.