TRANSACTIONSOF THE

AMERICANMATHEMATICAL
SOCIETY
Volume 338, Number 2, August 1993

PRODUCT FORMULAS AND CONVOLUTIONSFOR

ANGULARAND RADIALSPHEROIDAL WAVEFUNCTIONS

WILLIAM C. CONNETT, CLEMENS MARKETT, AND ALAN L. SCHWARTZ

Abstract. Product formulas for angular spheroidal wave functions on [0, n\

and for radial spheroidal wave functions on [0, oo) are presented, which gen-
eralize results for the ultraspherical polynomials and functions as well as for
the Mathieu functions. Although these functions cannot be given in closed
form, the kernels of the product formulas are represented in an explicit, and
surprisingly simple way in terms of Bessel functions so that the exact range of
positivity can easily be read off. The formulas are used to introduce two families
of convolution structures on [0, n\ and [0, oo), many of which provide new
hypergroups. We proceed from the fact that the spheroidal wave functions are
eigenfunctions of Sturm-Liouville equations of confluent Heun type and em-
ploy a partial differential equation technique based on Riemann's integration
method.

1. Introduction
One of the richest families of special functions arise in a natural way when
the three dimensional Helmholtz equation ((A2 + k2)W = 0, also called the
reduced wave equation) is solved by separation of variables in a coordinate
system with concentric families of spheroids and hyperboloids as level surfaces
[E, Chapter 16, AS, Chapter 21, Mi, 3.1, MSp] (the last reference contains some
excellent figures). If W(x, 6, <p)= U(x)V(Q)exxi(±im<p), then the Helmholtz
equation separates into one of the following systems of two ordinary differential
equations:
If" + (ctnhx)U' - [h - (fccsinhx)2 + (m cschx)2]I7 = 0
(1.1a)
(0 < x < oo),

V" + (ctn 6)V + [h + (kcsind)2 - (m ese 6)2]V = 0

(1.1b)
(0<8 <n),
Received by the editors February 22, 1991 and, in revised form, May 1, 1991.
1980 Mathematics Subject Classification (1985 Revision). Primary 33E10; Secondary 34B25,
43A10, 43A62.
Key words and phrases. Spheroidal wave functions, convolution, hypergroup, generalized trans-
lation.
The work of the first and third author reported here was partially supported by the National
Science Foundation (Grant no. DMS-9005999).
This paper was initiated while the second-named author was visiting the Department of Math-
ematics and Computer Science at the University of Missouri-St. Louis. He wishes to express his
gratitude to that department for its hospitality.
©1993 American Mathematical Society
0002-9947/93 $1.00+ $.25 per page

695

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 696 W. C. CONNETT, CLEMENS MARKETT, AND A. L. SCHWARTZ

If" + (tanhx)U' -[h- (kc coshx)2 - (m sechx)2]U = 0

;i.2a)
(0 < X < oo),

V" + (ctn 6) V + [h - (kc sin 6)2 - (m csc 6)2] V = 0

(1.2b)
(0<9 <n),
where c is a parameter of the coordinate system and h and m axe separation
constants. The first system arises from using the prolate spheroidal coordinates,
the second from using the oblate spheroidal coordinates. Solutions to the first
equation in each pair are referred to as radial solutions; solutions to the sec-
ond equation in each pair are called angular (or periodic) solutions. The two
equations inside each pair are related to each other by the imaginary change of
variables, x —*id , x —»/(6 - n/2), respectively.
By a transformation of the dependent variable V(6) = sinm ÔS(8) the two
equations (1.1b) and (1.2b) can be written in the normalized form

(1.3) S" + ^-^-S' + [X-ysin29]S = 0 (0 < 6 < n)

where a = m and y = ^(kc)2 (y < 0 for prolate, y > 0 for oblate). With the
addition of the boundary conditions S'(0) = S'(n) = 0 and the normalizing
condition S(0) = 1 this becomes a Sturm-Liouville problem with a discrete
spectrum of eigenvalues. The corresponding eigenfunctions are called the an-
gular spheroidal wave functions {S^'y(6)}nem0 (No = {0,1,2,...}) and are
dealt with in this paper for all a > -1/2 and y £ R, each choice of (a, y)
leading to a different family of orthogonal functions. Moreover, we will con-
sider the radial prolate spheroidal wave functions {^f'y(x)}x>o for a > -1/2,
y < 0, which are eigenfunctions of the Sturm-Liouville equation

(1.4) 5e"' + 2a*l<9" + [A2+ (a+ l/2)2-)>sinh2xL^ =0 (0 < x < oo)

tanhx
satisfying the initial conditions ZZ"(0) = 1, Z?"(0) = 0. Equation (1.4) is
obtained from (1.1a) by the transformation U(x) = sinhm xZ?(x) (with m =
a, -A = A2+1/4).
There are some well-known particular cases covered by equations (1.3) and
(1.4). If y = 0 then (1.3) has as its eigenvalues X = n(n + 2a + 1) with
corresponding eigenfunctions P^+i/2)(cos6)/Pt+U2)(l), where P{na+l/2)is the
ultraspherical polynomial of order a + 1/2 and degree n. In this case, the
eigenfunctions of (1.4) reduce to the ultraspherical functions on the positive
half-line. When a = v- 1/2 and y / 0, (1.3) is called the Mathieu equation
if v = 0 and the associated Mathieu equation if v £ N. The counterparts
associated with (1.4) are sometimes referred to as the respective "modified"
equations (cf. Mclachlain [M]).
A great deal of work has been done with spheroidal wave functions; some
standard references are Abramowitz and Stegun [AS], Arscott [AI], Erdelyi [E],
Flammer [F], Meixner and Schäfke [MS], and Meixner, Schäfke, and Wolf
[MSW], where the extensive literature up to 1979 is cited, and Stratton, et
al. [SM]. Nevertheless, one of the most useful and far-reaching identities in the
study of orthogonal functions and expansions, the product formula, had not
been found for the spheroidal wave functions except in the cases a = -112 or

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 SPHEROIDALWAVEFUNCTIONS 697

y = 0. It is the main purpose of this paper to establish the product formula for
the angular case in the form

(1.5) Sn"y(d)Sn*'y(<f>)
= Í San>?(cl)doe^(S) (0<6,ct><n)
Jo
where for each (6, </>),og^ = Og'Ï is an explicitly given Borel measure which
does not depend on n £ No (see Theorem 1 below). For later discussion we
also include the result in the limiting Mathieu case y = 0 which was first
stated by Meixner [Mei] and also treated by Volkmer [VI] in a different context.
In particular we will show that for each pair (a, y), the measure o^'l has
uniformly bounded total variation for 0 < 6, c¡>< n . Moreover, the exact range
of (a, y) will be given for which the measures are nonnegative. Finally, it turns
out that they are absolutely continuous with respect to Lebesgue measure if and
only if a > -1/2. The method of proof employs a partial differential equation
technique based on Riemann's integration method for solving Cauchy problems.
These ideas were used in a recent paper [CMS] to give, among other things, a
new proof of the product formula for the ultraspherical polynomials; see also the
previous work by one of us, where the techniques are extended to various other
situations (cf. [Ml, M2, M3], and for further references, [CMS]). The product
formula (1.5) can be used to define a two-parameter family of convolution
structures for expansions in spheroidal wave functions via the introduction of
appropriate generalized translation operators in the sense of Levitan [Le] (cf.
§4). Hence many standard tools from harmonic analysis on locally compact
abelian groups become available for the explication of the harmonic structure
here. The product formula leads in a natural way to a convolution for the Borel
measures on [0, %], M([0, n]). When the measure oe^ in (1.5) is positive,
it is possible to equip [0, n] with a hypergroup structure.
Along the same lines, we will also derive the product formula for the radial
prolate spheroidal functions in the form
/»OO

(1.6) 5f'y(x)5f'y(y) = f?'v(Ç)d<jx,y(Ç) (0<x,v<oo),

7o
the measures ox<y = ax]yy being independent of X £ [0, oo). (This is not the
same o as in (1.5), but context removes any danger of ambiguity.) It should
be remarked here, that unlike the angular prolate and oblate cases, the product
formula in the radial oblate case behaves somewhat differently from that in the
radial prolate case, (1.6). We have some interesting results here too, which will
be given elsewhere. The remarks about generalized translation, convolution,
and hypergroups apply here also.
In [CMS] the convolution structures associated with the eigenfunctions of the
perturbed ultraspherical equation, w"-r-(2a-l- l)(ctn 6)u' + [X-q(6)]u = 0 (a >
-1/2) are studied. Since the potential function q(Q) = ysin20 is symmetric
and real analytic on [0, n] as required in [CMS, Theorem 1.2] the existence and
uniqueness of the product formula (1.5) as well as its positivity in the prolate
case y < 0 follow. In this article we obtain explicit and simple representations
for the measures a g ' I from which the positivity for certain y > 0 can also be
determined, and which furnish additional hypergroup structures of the Jacobi
type beyond those described in [CS3]. The availability of a hypergroup structure

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 698 W. C. CONNETT, CLEMENSMARKETT, AND A. L. SCHWARTZ

allows a more refined study of various questions of harmonic analysis as, e.g.,
multiplier criteria or maximal functions (cf. [CS1, CS2]).
In addition to being useful in the harmonic analysis of eigenfunction expan-
sions, the product formulas of Theorems 1 and 2 give interesting information
about the eigenfunctions themselves. This "special functions" aspect becomes
apparent when observing that the equations (1.3) and (1.4) are singular differ-
ential equations not belonging to the hypergeometric class. In fact, if they are
thought of as equations in the complex plane (after substituting z = cos2(f7/2)
and z = cosh2(x/2), respectively), then according to the Klein-Bôcher-Ince
classification [I, Chapter XX] they have two finite regular singular points at
z = 0 and z = 1 and an irregular singular point of second species at z = oo .
Actually, the two equations are the same, but the product formulas range over
different intervals, namely (0, 1) and (1, oo), respectively. This is an equation
of the confluent Heun type [0,2, 12] (which may be reduced by a quadratic
transformation to the type [1, 1, 1], the singularities being elementary, regu-
lar, and irregular of the first species respectively). Hence the solutions lie in
the "land beyond Bessel" [A2] and cannot, in general, be represented in closed
form.
The remarkable fact is that it is still possible to write down the product
formula in closed form because of the Riemann function techniques even in
cases where there is no explicit representation of the eigenfunctions. So one way
of looking at formula (1.5), for instance, is to regard it as an integral equation
for the unknown function S„'7(6) after an appropriate choice of the variable
<f>as, e.g., 4>= n¡2 (cf. [VI] for consequences of this procedure in the Mathieu
case). Moreover, the formulas can be used to derive uniform and pointwise
estimates of the angular or radial spheroidal wave functions (cf. Theorems 3
and 4).
The spheroidal wave functions provide a rich family of examples in which
to study singular differential equations and the harmonic analysis of the corre-
sponding eigenfunction expansions. Though not in the field of classical special
function theory and, in particular, not of orthogonal polynomial type, they nev-
ertheless show a structure depending on two parameters a and y, which allows
distinct eigenfunction expansions to be organized into a scale or continuum of
expansions [CS1], and the properties of the individual eigenfunction systems
can be described by theorems about the parameters. The spheroidal wave func-
tions thus provide a convenient test for general theorems about eigenfunction
expansions.
Finally, we mention that these functions are important for various applica-
tions: they arise, for instance in signal theory where they occur naturally in the
analysis of band limited signals, see, e.g., [S, P]. We intend to study the appli-
cation of the positive convolution to these problems at another time. Further
references can be found in [D].
2. Product formulas for angular spheroidal wave functions
We begin with a few properties of the hypergeometric functions. Let (x)o =
1, and (x)„ = x(x + l)(x + 2) ■■■(x + n - 1) for n £ N, and let

Fia, b; c; z) = 2Fx(a,b;c; z) = £ %^z«.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 SPHEROIDALWAVEFUNCTIONS 699

Then one has

Ha,b-,c-,l) = ^}^c-_l (Rec>Ke(a + b)),

(2.1) 4-Fia ,b;c;z) = —F(a + 1, b + 1 ; c + 1; z),

az c

(2.2) F(a,b;c;z) = (l- z)~aF (a,c-b;c;

z- 1
z
(2.3) =(l-z)-bFlc-a,b;c;
(2.4) =(l-z)c-a-bF(c-a,c-b;c;z).
Our first aim is to establish the product formula for the angular spheroidal
wave functions iS^'^f?), n £ N0, 0 < 6 < n. To this end we recall the
following formulas for the ultraspherical polynomials:
S%-°(8) = Pia+1/2)(cos0)/PiQ+1/2)(l)
= F(-n ,n + 2a+l;a+l; sin2(9/2))
and (see [CMS, (1.8) and (1.9)])

n « S^°(d)S^°(<l))= / Sï<°(Z)K°(Z,d,tl>)wa(Ç)dÇ
Kz-->) J\8-(t>\
(0<e,(p<n)
with
wa(£) = sin2a+1£,

Tia+1) (n)^2 x^e-^n-V + t-nl),

Ka(U,<t>)= \ r(a+l/2)r(l/2) ag-
io ifi^(|e-01,7t-|e + ^-7r|),
where
ipo = [cos(0 -(f))- cos¿][cos<^ - cos((9 + </>)],
coq = (sin¿;)(sin0)(sin0).
Let Jv denote the Bessel function of the first kind of order v , so that

Mz) = 2»Y(v+ l)Uz)/z» = £ ¿, + [¿kl

is an entire function that satisfies ^(0) = 1 and J^(x) = fv(ix) > 0 if x e E.

Let jv be the smallest positive zero of Jv . For detailed information on jv such
as bounds, monotonicity, and asymptotic behavior as v —► -1, see [IM] and
the references cited there.
Theorem 1. Let a >-1/2, yel, 0 < 0, <p< n, and n £ N0.
(i) For a > -1/2, the angular spheroidal wave functions satisfy
pn-\e+4>-n\
(2.6) San<y(d)San'y(<t>)= S^y(^Wa'y(^,e,<l>)wa(^)di
J\e-4>\

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 700 W. C CONNETT, CLEMENS MARKETT, AND A. L. SCHWARTZ

where
(2.7) 3T-y(i, 0,0) = Ka(Ç, 9, 4>)fa-XI2(JWo).
(ii) In the particular case of the Mathieu functions, a = —1/2, one has

sñl/2'yid)Sñl/2'y(cp) = \s-n{l2>yi\e- 0|) + l-szl/2'y(n -\e + 4>-*\

(2.8) rn-\e+<p-n\
l-K-\V+lp-7l\
+ Sn-xl2'y(ïW-xl2>y(t;,9,<j>)dcl
J\8-<t>\
where
(2.9)
l4YCOoMVW~o) ifǣ(\e-<t>\,n-\9 +
^-1/2'"(c:,0,0)=|o^
if¿í<¿i\9-(j>\,7i-\9 + 4>-n\).
(iii) If a > -1/2 and -oo < y < J¿—1/2» or a = -1/2 and y < 0, then
Z%a'yiti ,0,0) is positive, symmetric in t\, 9, and 0 and as a function of Ç, it
is supported in \9 - 0| < £ < n - \9 + 0 - n\.
(iv) If y > 0 and a > 0, then

/ \3?a>y(c¡,9,<¡>)\wa(c:)d!;<l.
J\e-<t>\
To prove Theorem 1 we will proceed (as in the ultraspherical case [CMS,
§2]) from the fact that the product on the left-hand side of (2.6) (and of (2.8))
can be considered as the unique solution of a hyperbolic initial value problem
which can be solved in terms of the initial function by Riemann's integration
method. Set
u(i,r,) = u(n,t) = San'y(Ç)Sn»y(r1)
and observe that because of the symmetry
S^y(7i-0^(-l)"S^y(í),
we may restrict our attention to (Ç, n) for which 0 <n <£, <n - n . Denote
this triangle by E. Then, since S^'y(c¡) satisfies equation (1.3) together with
S'(0) = S'(n) = 0 and S(0) = 1 one has
(Lly-L°>?)u(C,n) = 0, ((Z,n)£E),
u(Ç,0)=Sï>?(Ç), un(i,0) = 0, (0<i<n),
where
L¡'y = d2/dc¡2 + (2a + l)(cotí)rf/í/í - y sin2 f.
The solution of (2.10) requires explicit knowledge of the corresponding Rie-
mann function. For any fixed point (0, 0) 6 E let Ag^ be the characteristic
triangle with vertices P = (9, 0), ßo , and Rq , where Qe = (0 - 0 + e, e)
and Re = (8 + 0 - e, e). Then for (¿;, n) £ Ag^ , the Riemann function is
defined as the unique solution v(Ç, n) = v(Ç, n; 9, 0) = Ra'v(cl, n; 9, 0) of
the characteristic boundary value problem:

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 SPHEROIDALWAVEFUNCTIONS 701

(Las'y-Lan<y')v(c;,n) = 0 ((tZ,n)£Ae^),
(a+l/2 a+l/2\ .„ _ ^

f a+l/2 a+l/2\ ^

v(9,<t>)= l
(the asterisk (*) denotes the adjoint operator). Let

X = xií,n;9,<t>) = [cos(0 - 0) - cos(£ - r/)][cos(¿ + if) - cos(0 + 0)],

>p=ip(cl,n;9,(t)) = [cos(0 - 0) - cos(£ + n)][cos(C- if) - cos(0 + 0)],
co = 4(sin^)(sin?7)(sin0)(sin0),

W = ^- o = -^=*
co W- 1 y/

^ (c,^,f7,0)- ^(sinö)(sin0j i^j

Lemma. For a > -1/2, y G E, and (£,, n) £ Ae $,

(2.12)
A°''(i,f/;fl,0)
i-yxi'4)A
= .#<*(£,9; 0, 0) £ F(k + 1/2 - |a|, 1/2 - \a\ ; k + 1; <P)V ]¿n) ,
ken0 [ >

in particular, for a = -1/2

R-xt2>yiZ,n;9,<f>) = Joiy/yx).
(Notice that in the ultrasphericalcase y = 0, (2.12) collapses to the k = 0 term
to yield Ra>\c¡, n; 9, 0) = •#<«(£,n; 9, 4>)F(l/2-\a\, 1/2 - |a| ; 1 ; <D).)
Proof. The substitution of the independent and dependent variables

X0 = cos(0 + 0), Y0= cos(0 - 0),

f2 13n X = cos(£ + ri), Y"= cos(£ - rç),

V{¿í>
"; *' 0) = Uin0)(sin0)J V{X'Y''X°' Y0)
leads to the selfadjoint "normal" form of problem (2.11),

(2.14) V^+{$^YT2-l)V =° ^o<X<Y<Yo),

V(X0, Y;X0, Yo)= V(X, Y0;X0, Y0)= 1.
Its solution was given by Henrici [H], (cf. also [L]) as a confluent hypergeometric
function of two auxiliary variables

Wx= (X - Xo)(Y - Y0),

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 702 W. C. CONNETT, CLEMENS MARKETT, AND A. L. SCHWARTZ

(X-Xq)(Y-Yq)
2 (X - Y)iX0 - Y0) '

namely

F(X,r;Xo,y0) = H2(l/2 + a, 1/2-a; l;W2,yWx/4)

(2.15) v- (l/2 + a)n(l/2-a)„ k
^ k\n\(k + n)\ yy2U^W) ,

which converges for \W2\ < 1 (for properties of hypergeometric functions of

several variables cf., e.g., [E, §5]). Writing the inner sum on the right of (2.15)
as an ordinary hypergeometric function and extending it analytically to the half-
plane Re W2< 1/2 by (2.2), (2.3), one arrives at

F(x,y;Xo,y0) = (i-w/2)|a|-1/2
W2 \ (yWx/4f
X £F(*+l/2-|a|,l/2-|a|;*+l;w£T)

which is valid for all X0 < X < Y < Y0. Returning to the original variables
C, n, 8, 0 via (2.13) and observing that W2= W, Wx= -% finally yields the
lemma. Notice that in the special case a = -1/2, the boundary value problem
(2.11) reduces to that for the well-known telegraph equation which was solved
by Riemann already in terms of the Bessel function Jo (cf., e.g., [G, (4.84)] or
[L]). D
Remark. In the proofs of the product formulas for the Bessel functions and for
the Laguerre functions given in [Ml], extensive use was made of the fact that
after transformation to a selfadjoint normal form, the characteristic boundary
value problems defining the corresponding Riemann functions are exactly of
the form (2.14), with y = 0 in the Bessel case and y = constant / 0 in the
Laguerre case. This explains the formal similarity of the product formulas for
the Laguerre and spheroidal wave functions (though being defined on different
intervals). In particular, one can say that the relationship between the Laguerre
and Bessel cases is the same as between the spheroidal and ultraspherical cases.

Proof of Theorem 1. (i) We discuss here the case -1/2 < a < 0 ; the case a > 0
is dealt with by similar arguments. The argument used in [CMS, Theorem 1.2]
in the ultraspherical case yields (2.6) with Z%a'y obtained from the Riemann
function (2.12) by

d_
JTa'y(í,9,ct>) = lim i-l-(sinn)2a+x-^-[(sinc:sinr1)-2"-xR°'y(Ç, r/; 0, 0)]
n->0+ [_ L dr¡'

= -22a lim i(sinf/)2a+1

r/-»0+

d
w-a-l/2 Y^F(k + a+1/2, a+1/2; k+1; 0>)( J*^
XÔ~n

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 SPHEROIDAL WAVEFUNCTIONS 703

In view of (2.1), one has

d_ (-rx/4)A
-a-x/2Y/F(k + a+l/2,a+l/2;k+l;(p)
dr¡ ¥ (k\)2
fcGNo
,ö>
= -(a + l/2)<p-a~3'2-^- Y, F(k + a + 1/2, a + 1/2 ; k + 1; <D)(-7X/4)*
dn ik\)2
k€N0

+ y/~
a_i/2 v (fc+ a+l/2)(a + l/2)
k+ 1
keN0
dö (-7^/4)*
x Fik + a + 3/2, a + 3/2; k + 2; O)
a?? ik\)2
+ ^-m^F(k+a+i/2,a+i/2;k + ,.^t^C.l^)
k&N v y x ' '

= sx+s2 + s3.
The term-by-term differentiation is justified because all three series converge
uniformly. We show this first for a < 0. Observe that 0 < O < 1 if n £
(0, 7t/2], so that for some constants Ax and A2,
0 < F(k + a + 1/2, a + 1/2 ; k + 1 ; O)
<F(k + a+ 1/2, a + 1/2; fc+ 1; 1)
r(fc + i)r(-2a) +1/2
r(l/2-a)r(fc+l/2-a) - '
and from (2.4)
(k + a+l/2)(a+l/2)
F(k + a + 3/2, a + 3/2 ; k + 2 ; 0)(1 - d>)2a+1
fe+l
(fc+ q+l/2)(q+l/2)
F(l/2-a,k+ l/2-a;k + 2;<&)
Jfc-f-1
(2.16) < ik + a+l/2)(a+l/2) F(l/2-a, k+ l/2-a;k + 2; 1)
k+ 1
Y(k + l)Y(2a + I)
Y(k + a+l/2)Y(a+l/2)
< A2kx'2~a.

If a = 0, a very similar argument is based on the estimate

Y(k + 1)
F(k+1/2, l/2;k+l; z) < 1 log(l-z).
Y(k+l/2)Y(l/2)
Now,
dy , dx ~ ¿70
—-► 2oj0 , t.-► -2cúo , 4ct70/yo (h-»0+)
dn dn dn
and so remain bounded in the limit for any 0-0<£<0 + 0. Since, moreover,
l/ip < 1/y/o, it follows that »Si and Si axe bounded as r\ -> 0+ , and
&"•*£, 0, 0) = -2la lim {(sin n_)ln+lS2(t¡, n; 8, 0)}
n->0+

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 704 W. C. CONNETT, CLEMENS MARKETT, AND A. L. SCHWARTZ

In view of (2.16) and the fact that

hm (8in,)^V-°-1/2(l -Vr2«-1 = (¿£^.

the kernel can now be calculated to be

r*.'(? 9 ¿\ - -2« ¥^l/2 V nk + 1)r(2a + 1} (-^o/4)"

[Ç'0'Ç) (4co0)2<*k^oY(k + a + l/2)Y(a+l/2) (k\)2

Wrm r(2a+l)
- (2coo)2«Y2(a+l/2/a-l'2{VWo)-

Thus using the Legendre's duplication formula for the gamma function we ob-
tain (2.7).
(ii) In the special case a = -1/2 we have for 0 < e < n/2, (cf. [CMS,
equation (3.9)])

u(9, (b) =-((uv)\Qc + (uv)\Rc)

1 ¡■e+<j>-s
ÇV+Ç—1.
5 / [Mi,(í, fl)v(c¡, n)]\„=£d¿¡
L Je-é+e
e+<t>-e
1/
2 Je d-Ó-E
[u(t¡, n)v„(c¡, n)]\„=£dí

= Tx(e) + T2(e) + T3(e),

with v(i, n) = v(Ç,n; 9, 0) = J0(y/ñ).
Now, lim£_o+v(Qe) = limE^o+v(RE) = 1, since /0(0) = 1 , so

lim Tx(e) = \[u(9 - 0) + u(8 + 0)].

Since v(Ç, n) is uniformly bounded and u„(s) —>0 as £ -> 0+, lime^o+ ^(e)
= 0. Finally
v„it, n)=-IMVU)^
and so

Jf-'/2-^, 0,0)= lim [-\vniZ, r,)]= -IcooMVWo).

e—»0+ 4

This yields (2.9).

(iii) The general theory [CMS, Theorem 1.2(iv)] predicts the positivity of
5fa'y when a > -1/2 and y < 0, for in that case ysin20 is a nonincreas-
ing function onO<0<7r/2. The positivity is also apparent from the ex-
plicit formulas (2.7) and (2.9) since if y < 0 and v = a- 1/2, Jii^yy/o) =
J»iy/\y\\po) > 0, while if 0 < y < j2, then ^/m < jv , and so XiVfWö) > 0.
Symmetry follows since a>o and ipo are symmetric in the three variables.
(iv) If a > 0, then v = a - 1/2 > -1/2, so ^v(x) is a character for
a hypergroup. (See, for instance, the product formula for ^v given in [Sc,
equations (1.1 )—(1.3), which is in turn based on a theorem of Gegenbauer [W,

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 SPHEROIDALWAVEFUNCTIONS 705

p. 367].) Hence \fv(x)\ < \fv(0)\ = 1. Thus if y > 0 and a > 0,

/ \jra>?it,o,<t>)\waiZ)dï
J\e-4>\
rn-\e+tt>-K\
pK-\8+<j>-n\
< Ka(i,e,<p)wa(t)dt = i
'I
J\e-6\
(as can be seen by setting n = 0 in (2.5)). D

3. Product formulas for radial prolate spheroidal wave functions

Let us now turn to the product formula for the eigenfunctions of the Sturm-
Liouville equation (1.4), the radial prolate spheroidal wave functions S^f'y(x),
X, x > 0. In the following we use the same notations x, y/, œ,^ as before,
but with the new meaning (indicated sometimes by x, y instead of 0,0)

X = xi£, n ; x, y) = [cosh(x - y) - cosh(£ - r/)][cosh(£+ n) - cosh(x + y)],

yi = y/(Z, n ; x, y) = [cosh(x - y) - cosh(¿; + n)][cosh(c¡ - n) - cosh(x + y)],
co = 4(sinh(¡;)(sinhf7)(sinhx)(sinhy),
0=— , y/0= wi£, 0;x,y), w0 = (sinh£)(sinhx)(sinhy).

Theorem 2. Let a > -1/2, y < 0, X>0 and x, y £ (0, oo).

ii) For a> -1/2,
fX+y
fX+y
^'y(x)^'y(y) = / S"f*>y(t¡)3fa>y(ci,x,y)(sinhcl)2a+Xdct
J\x-v\
l\x-y\

where

ra,y Y(a+l/2)Y(l/2) aft ^-'MVl^o)

(i,x,y)=<
ifc¡£(\x-y\,x+y),
10 iftë(\x-y\,x + y).
(ii) If a =-1/2,
S--x'2'y(x)S--l/2'^y) = ^-l/2'y(\x-y\) + ^Z-l/2'y(x + y)
rx+y
rX+y
+ ^-l/2'y(CW-i/2'7iC,x,y)d¿:
J\x-v\
<\x-y\

where

^-x'2-y(Ç,x,y) = i[ \\y\o)o^\(\/ÏÏm) if^£(\x-y\,x + y),

0 iftï(\x-y\,x+y).
(ni) For any 0 < x, y < oo and \x-y\ <Ç < x+y, the kernel 5?a>y(tl, x, y)
is positive and symmetric in c¡, x, and y .

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 706 W. C CONNETT, CLEMENS MARKETT, AND A. L. SCHWARTZ

Proof. The proof follows the same lines as the proof of Theorem 1. The char-
acteristic boundary value problem for the Riemann function now reads

v«-^-[^Vt+\ñ3¿iv,i
,tanhf Ji
-y[sinh2¿:-sinh2f7]v = 0 ((<*,r¡)eAx>y),
,,n /a+l/2 a+l/2\ . ,_

( a+l/2 a+l/2\
v¿-v„- [-rf-z-r1— )v = 0 (t¡ + n = x + y),
( n V tanhcl tanhiy / v ' "
v(x,y) = 1.
Its solution is given by

A-.'íí n-x vi- Asi°h^)(sinh^y+1/2/^NN-^

R K'*'*'')- i(sinhx)(sinhy)J \7o)
x ^2 F(k +1/2-\a\, l/2-|a|;/c+l;<D)(~ (k\)2
k€N0 V '

This follows immediately from the fact that the transformations

X0 = cosh(x + y), Y0= cosh(x - y), X = cosh(¿; + n), Y = cosh(¿; - n),

<* v //(sinhO(sinh??)V+'/2I//v VA
^^;x'y)=((sinhx)(Sinhy)j r(X, y ; *0. lb)
reduce problem (3.1) to the same selfadjoint problem (2.14) as was solved in
the angular spheroidal case, but now in the domain Yo < Y < X < Xq . Using
the representation (2.15) of V , the rest of the proof is straightforward and left
to the reader. D
Remark. For y = 0, the results of Theorem 2 coincide with the product formula
for ultraspherical functions due to Flensted-Jensen and Koornwinder [FK] (cf.
also [K] and, for a partial differential equation approach, [M3]).

4. Convolution structures, hypergroups, and inequalities

The measure a^l of (1.5) is defined implicitly by Theorem 1 and is given
by
doaQ>l(c\)= 3?a>y(t\,9,<?)wa(t\)dc\ (q>-1/2, yeE),
i4"1) d°ëf'yit) = A-*|(í) + 2-dôK-\e+<,-*\it)
+3r-'/2'y(t, 9, 0)¿£
(y ER)
where ôt denotes the unit point mass concentrated at t. This gives rise to
two products or convolutions on M([0, n]) which are continuous with respect
to the total variation norm || • ||. If p and v belong to M([0, n]), the first
product p#v = p#a'yv ,is defined by its action on a continuous function f :

(4.2) [* f(s)d(p#v)(s)= f f F f(cZ)do';-l(cZ)dp(9)dv

(0)-
Jo Jo Jo Jo

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 SPHEROIDALWAVEFUNCTIONS 707

The explicit form of a^'l given in (4.1) shows that the innermost integral is a
measurable function of (0,0). The product # is a continuous operation on
M([0, n]) provided that Aa'y = sup{||o-0 ^||:0 < 9, 0 < n} < oo. In fact, we
have the inequality \\p#u\\ < Aa-y\\p\\ \\v\\. Let /iA = $*S„*>y(ci)dp(t¡), then
substituting S„'y for / in (4.2) and using (1.5) we have (/i#i/)£ = p^v* , so
the product is commutative and associative. Thus if Aa-y < 1, (M([0, n]),
#"■'') is a commutative Banach algebra. There is also a convolution on an
appropriate Ll space. In particular if

ll/Hi= ll/||i,«=Jof\f(C)\wa(C)dS
<oo
define

if#g)it) = if1P'rg)ii)= f f f(8)g((f>)dcJe^)wa(d)wa((l>)dddcf>

Jo Jo
and /* = ¡¿SS'rit)fiZ)wait)dt. Examination of (4.1) shows that f#g is a
well-defined measurable function. The product satisfies

ll/#Slli<¿Q>l/||i||*lli and (ßg)An=fnS^.

The product formula also gives rise to an inequality for the spheroidal wave
functions since (1.5) implies
\San'y(8)\|S^(0)| < ^'TC'loo (0 < 0, 0 < n),
whence HSS'l«, < Aa>y.
The second kind of convolution is based on the ideas in [CMS, §4] and uses
a renormalized measure defined by
^:J(í) = s°>yias%>yid)s°>yi<p))-xdoae:¡ic:).
Notice that S%'yii) is positive throughout [0, n] because S%-7(0) = S%-r(n) =
1 and has no zeros in the interval. So if Rak'y(£,)= S°'y(Ç)/S°'y(c;) we have
the product formula

(4.3) R«n>y(9)Ran>y(<p)
= f K'yit)dTae'l(i)
Jo
and the convolution * = *a ■y is defined by

[nfi(C)d(p
*!/)({)= r r rf(i)dxae'y^)dp(9)du(ct>).
Jo Jo Jo Jo
This operation is also commutative and associative and defines a convolution
on the space of functions / for which

f |/(0| dma>y(£,)<
oc
7o
where
dma>r(C)= [S^(i)]2wa(C)dC.
If (Tq'I is a positive measure for every 0 and 0 in [0, n], then so is tae'7^.
Moreover, if we set n = 0 in (4.3), ||Tg'£|| = 1, so * is probability preserving.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 708 W. C. CONNETT,CLEMENSMARKETT,AND A. L. SCHWARTZ

That is, if p and v are probability measures (positive with unit mass), then so
is p * v . In particular we have for the unit point masses
00*3^= ^1 (0<9,4><n).
Equation (4.3) yields the global inequality ||/î2','||oo = Rn'7(0) = 1, from which
we obtain the pointwise inequality
|Sï-y(OI<s0,,'(i) (o <£<*)•
Finally we remark that

T*f(e)=JoI*mdTÎ'lit)
is a generalized translation in the sense of Levitan [Le]. If f(¿¡) = J2n^ocn^n '7 (£) >
then T*fit) = ZnloCnRan'y(OK'y(<t>),™<i

(f*g)(Z) = fnT't'f(9)g(cf))dma'y(4>).
Jo
We summarize these observations and give some particular bounds in
Theorem 3. (i) If y £ E and a > -1/2, then M([0, n]) has a continu-
ous convolution #a'7 which satisfies \\p#a'yv\\ < Aa'y\\p\\ \\u\\ where Aa'y =
sup{||cre^||: 0 < 8, 0 < n}. Moreover HS^'l«, < Aa>y, and Aa<y satisfies the
following:
(a) // y> 0 and a > 0, then Aa-y = 1.
(b) If y >0 and a > -1/2 then Aa>y< max0<x<y\fa- \ß(-Jx)\.
(c) If y >0, then A~xl2<y< 1 + \ max0<x<7\fx(y[x)\.
id) If y<0 and q>-1/2 then Aa<y< \^a-i/2(^/\y\)\ ■
(e) lfy<0, then A~x!2>y
< 1+ pHvTJÏ)-
(ii) If y>0 and a>0, (M([0, n]), #a^y) is a Banach algebra.
(iii) If a > -1/2 and -co < y < j\_Xn, or a = -1/2 and y < 0, then
([0, n], *a'y) is a Jacobi type (a, a) hypergroup with Hermitian characters
{Rn'7}%Lo>identity So, and Haar measure m°J. Moreover
\Sï'y(Ç)\<SZ'y(i) (0<Ç<7t).
The reader is referred to [CS3] for a general discussion of Jacobi type hyper-
groups and to [CMS, §4] for a discussion of those hypergroups which arise from
a general class of differential equations which includes (1.3).
Proof, (i) In light of the discussion preceding the theorem, it only remains to
establish the estimates (a)-(e). Observe that if a > -1/2,

\ar°->iZ,d,<i>)\wait)dz
,3-4>\
< max{|^_1/2(ym)|: |0 - 0| < £ < n - \8 + 0 - n\}
by (2.7) and the argument used to prove Theorem 1, (iv). This yields (a), and
(b) follows since 0 < yio < 1 . Part (d) uses the fact that JT.-1/2 is an increasing
function; parts (c) and (e) follow from slight modifications of the above,
(ii) is a simple consequence of (i)(a).

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 SPHEROIDALWAVEFUNCTIONS 709

(iii) The hypergroup assertions follow from the remarks preceding the the-
orem. The inequality follows since in any hypergroup, the characters must be
bounded by unity. D
For the radial prolate functions, when y = 0, SÇ'°(z) has no zeros in
[0, oo), and Theorem 2 yields the product formula for Jacobi functions ob-
tained in [FK]. In this case we define
dx%>»(z)=S*«>\z)(Z70a'\x)Z7°>\y)rxdox>:Qy(z)
with doxly(z) = J?a-0(z, x, y)sinh2a+lzdz and * = *a'° analogous to the
above; there is also an analogous definition of generalized translation suggested
by Theorem 2. With this definition we obtain
Theorem 4. There is a hypergroup ([0, oo), *Q'°) with Hermitian characters
{S>!Pf"O(z)/S^oa'0(z)}x>o, identity element ôo, and Haar measure ma'° given
by dma>°(z) = LS^'V)]2 sinh2a+1zdz . Moreover, for each X>0
\Sf'°(z)\<^0a'°(z) (0<z<oo).
The authors are grateful to Hubert Kalf who pointed out that if y < 0 an
argument based on the Sturm Comparison Theorem shows that Z?Qa'y must
have zeros in (0, oo), and so these methods will not yield bounds on the kernel
that will allow the construction of a convolution.

References

[AS] M. Abramowitz and I. A. Stegun, Handbook of mathematical functions, National Bureau of

Standards, Washington, D.C., 1964.
[Al] F. M. Arscott, Periodic differential equations, Pergamon Press, Oxford, 1964.
[A2] _, The land beyond Bessel: A survey of higher special functions, Proc. Conf. in Ordinary
and Partial Differential Equations (Dundee, 1980), (W. N. Everitt and B. D. Sleeman, eds.),
Lecture Notes in Math., vol. 846, Springer-Verlag, Berlin, 1981, pp. 26-45.
[CS1] W. C. Connett and A. L. Schwartz, The theory of ultraspherical multiplies, Mem. Amer.
Math. Soc. No. 183, 1977, pp. 1-92.
[CS2] _, A Hardy-Littlewood maximal inequality for Jacobi type hypergroups, Proc. Amer.
Math. Soc. 107 (1989), 137-143.
[CS3] _, Analysis of a class of probability preserving measure algebras on compact intervals.
Trans. Amer. Math. Soc. 317 (1990), 371-394.
[CMS] W. C. Connett, C. Markett, and A. L. Schwartz, Convolution and hypergroup structures
associated with a class of Sturm-Liouville systems, Trans. Amer. Math. Soc. 332 (1992),
365-390.
[D] J. S. Dehesa, New properties of the spheroidal wave equation, Lett. Nuovo Cimento 35
(1982), 25-28.
[E] A. Erdélyi et. al, Higher transcendental functions. Vols. I and II, McGraw-Hill, New York,
1955.
[F] C. Flammer, Spheroidal wave functions, Stanford Univ. Press, Stanford, 1957.
[FK] M. Flensted-Jensen and T. H. Koornwinder, The convolution structure for Jacobi function
expansions, Ark. Mat. 11 (1973), 245-262.
[G] P. R. Garabedian, Partial differential equations, Wiley, New York, 1964.
[H] P. Henrici, A survey of I. N. Vekua's theory of elliptic differential equations with analytic
coefficients,Z. Angew. Math. Phys. 8 (1957), 169-203.
[I] E. L. Ince, Ordinary differential equations, Dover, 1956 (1st ed., 1926).

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 710 W. C. CONNETT, CLEMENS MARKETT, AND A. L. SCHWARTZ

[IM] M. E. H. Ismail and M. E. Muldoon, On the variation with respect to a parameter of zeros
of Besseland q-Besselfunctions,J. Math. Anal. Appl. 135 (1988), 187-207.
[K] T. H. Koornwinder, Jacobi functions and analysis on noncompact semisimple Lie groups,
Special Functions: Group Theoretical Aspects and Applications, (R. A. Askey, T. H. Koorn-
winder, and W. Schempp, eds.), Reidel, Dordrecht, 1984, pp. 1-85.
[L] E. Lanckau, Die Riemannfunktionen selbstadjungierter Gleichungen, Wiss. Z. Techn. Hoch-
sch. Karl-Marx-Stadt21 (1979), 535-540.
[Le] B. M. Levitan, Generalized translation operators, Israel Program for Scientific Translations,
Jerusalem, 1964.
[Ml] C. Markett, A new proof of Watson's product formula for Laguerre polynomials via a Cauchy
problem associated with a singular differential operator, SIAM J. Math. Anal. 17 (1986),
1010-1031.
[M2] _, Norm estimates for generalized translation operators associated with a singular differ-
ential operator, Indag. Math. 46 (1984), 299-313.
[M3] _, Product formulas for Bessel, Whittaker, and Jacobi functions via the solution of an
associated Cauchy problem, Functional Analysis and Approximation, Proc. Conf. Oberwol-
fach 1983, (P. L. Butzer, R. L. Stens, and B. Sz.-Nagy, eds.), ISNM 65, Birkhäuser, Basel,
1984,pp. 449-462.
[M] N. W. Mclachlain, Theory and application of Mathieu functions, Clarendon Press, Oxford,
1947.
[Mei] J. Meixner, Integralbeziehungen zwischen Mathieuschen Funktionen, Math. Nachr. 5(1951),
371-378.
[MS] J. Meixner and F. W. Schäfke, Mathieusche Funktionen und Sphäroid Funktionen, Springer-
Verlag, Berlin, 1955.
[MSW] J. Meixner, F. W. Schäfke, and G. Wolf, Mathieu functions and spheroidal functions and
their mathematical foundations, Springer-Verlag, Berlin, 1980.
[Mi] W. Miller, Jr., Symmetry and separation of variables, Encyclopaedia of Mathematics and
its Applications, Vol. 4, Addison-Wesley, Reading, Mass., 1977.
[MSp] P. Moon and D. E. Spencer, Field theory handbook, Springer-Verlag, Berlin, 1961.
[P] M. Perlstadt, Polynomial analogs of prolate spheroidal wave functions and uncertainty, SIAM
J. Math. Anal. 17 (1986), 240-248.
[Se] A. L. Schwartz, The structure of the algebra ofHankel transforms and the algebra ofHankel-
Stieltjes transforms, Canad. J. Math. 23 (1971), 236-246.
[S] D. Slepian, Some comments on Fourier analysis, uncertainty, and modeling, SIAM Rev. 25
(1983), 379-393.
[SM] J. A. Stratton, P. M. Morse, L. J. Chu, J. D. C. Little, and F. J. Corbató, Spheroidal wave
functions, Wiley, New York, 1956.
[VI] H. Volkmer, Integralrelationen mit variablen Grenzen für spezielle Funktionen der mathe-
matischen Physik, J. Reine Angew.Math. 319 (1980), 118-132.
[V2] _, Integralgleichungen für periodische Lösungen Hillscher Differentialgleichungen, Anal-
ysis 3 (1983), 189-203.
[W] G. Watson, A treatise on the theory of Bessel functions, Cambridge Univ. Press, Cambridge,
1966.

(W. C. Connett and A. L. Schwartz) Department of Mathematics and Computer Science,

University of Missouri-St. Louis, St. Louis, Missouri, 63121
E-mail address : connett@umslvma.bitnet
E-mail address: schwartz@arch.umsl.edu

(Clemens Markett) Lehrstuhl A für Mathematik, Rheinisch-Westfälische Technische

Hochschule Aachen, Templergraben 55, D-5100 Aachen, Germany

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use