Sei sulla pagina 1di 52

Commun. Math. Phys.

206, 639 690 (1999) Communications in


Mathematical
Physics
Springer-Verlag 1999

A Generalized Hypergeometric Function


Satisfying Four Analytic Difference Equations
of AskeyWilson Type
S. N. M. Ruijsenaars
Centre for Mathematics and Computer Science, P.O. Box 94079, 1090 GB Amsterdam, The Netherlands

Received: 21 December 1998 / Accepted: 14 April 1999

Abstract: The hypergeometric function 2 F1 can be written in terms of a contour inte-


gral involving gamma functions. We generalize this (Barnes) representation by using a
certain generalized gamma function as a building block. In this way we obtain a new
2 F1 -generalization R(a+ , a , c0 , c1 , c2 , c3 ; v, v) with various symmetry features. We
determine the analyticity properties of the R-function in all of its eight arguments, and
show that it is a joint eigenfunction of four distinct AskeyWilson type difference op-
erators, two acting on v and two on v. The AskeyWilson polynomials can be obtained
by a suitable discretization of v or v.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 639
2. The R-Function: Automorphy and Meromorphy . . . . . . . . . . . . . . . . 646
3. The R-Function: A1Es and AskeyWilson Specialization . . . . . . . . . . . 651
4. A Barnes Type Representation for Basic Hypergeometric Functions . . . . . . 656
Appendix A. The Hyperbolic Gamma Function and Related Functions . . . . . . 663
Appendix B. Analyticity Properties . . . . . . . . . . . . . . . . . . . . . . . . . 672
Appendix C. Proof of Theorem A.1 . . . . . . . . . . . . . . . . . . . . . . . . . 681
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 690

1. Introduction

In this paper the first of a series we begin a detailed study of a novel generalization of
the hypergeometric function 2 F1 (a, b, c; w). Recall the latter can be defined for |w| < 1
640 S. N. M. Ruijsenaars

by the Gauss series


X
ab 1 a(a + 1)b(b + 1) 2
2 F1 (a, b, c; w) =1+ w+ w + = Cn (a, b, c)wn ,
c 2! c(c + 1)
n=0
(1.1)
1 0(a + n) 0(b + n) 0(c)
Cn (a, b, c) . (1.2)
n! 0(a) 0(b) 0(c + n)
An obvious way to generalize this function is, therefore, to generalize the coefficients
Cn of this power series.
In particular, such generalizations can be defined within the context of basic hyper-
geometric series. Recalling the standard definitions (|q| < 1, |w| < 1)

X (a1 ; q)n (as+1 ; q)n wn
s+1 s (a1 , . . . , as+1 , b1 , . . . , bs ; q, w) , (1.3)
n=o
(b1 ; q)n (bs ; q)n (q; q)n
n
Y
(a; q)n (1 aq j 1 ), n N, (1.4)
j =1

the oldest and most widely known generalization is Heines function 2 1 (a, b, c; q, w).
Indeed, the coefficients
(a; q)n (b; q)n
Cn (a, b, c; q) , |q| < 1, (1.5)
(c; q)n (q; q)n
of its power series clearly converge to the 2 F1 -coefficients Cn (a, b, c) (1.2) as q 1.
One can also define various generalizations in terms of s+1 s with s > 1 by letting
the additional parameters depend on q in suitable ways. But such functions will not
generally satisfy q-contiguous relations and a second order q-difference equation, by
contrast to Heines 2 1 . (We refer to Ref. [1] for detailed information and references
concerning various q-world results mentioned in this paper.)
Next, we recall that the Gauss series (1.1) can be made to break off by suitably
specializing the parameters a, b, c. The largest family of polynomials obtained in this
way (the Jacobi polynomials) still depends on two continuous parameters (in addition
to the continuous variable w). As is well known, Askey and Wilson [2] generalized this
family to a five-parameter family of polynomials defined in terms of 4 3 , viz.,

pk (q, , , , ; cos v) = Nk 4 3 (q k , q k1 , eiv , eiv , , , ; q, q).


(1.6)
Here, the normalization coefficient
Nk (q, , , , ) k (; q)k ( ; q)k (; q)k (1.7)
ensures symmetry in , , , .
These polynomials are eigenfunctions of a second-order q-difference operator and
their three-term recurrence relation is explicitly known as well [2]. As a corollary of our
work, we reobtain these results in a novel way. Moreover, omitting the constant Nk in
(1.6), we show that when k is replaced by an arbitrary complex number, the resulting
4 3 -function no longer satisfies the AskeyWilson q-difference equation, though it may
be viewed as a 2 F1 -generalization.
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 641

In recent years other generalizations of the hypergeometric function were also pre-
sented and studied by Grnbaum and Haine [3], and by Nishizawa and Ueno [4].
The main object of study in this paper is a function R(a+ , a , c0 , c1 , c2 , c3 ; v, v).
It yields (in essence) the AskeyWilson polynomials when one of its variables v or
v is suitably discretized, and it may be viewed as a 2 F1 -generalization. However, its
structure is essentially different from the generalizations already mentioned. To lead
up to its definition, we begin by recalling that the hypergeometric differential equation
satisfied by 2 F1 can be transformed to a nonrelativistic Schrdinger equation via suitable
substitutions. Specifically, introducing the wave function

nr (d, d; v, v) 2 F1 ((d + d + i v)/2, (d + d i v)/2, d + 1/2; sh2 v), (1.8)

one obtains
1
(Hnr nr )(g/h, g/h; x, p/h) = (p2 + 2 (g + g)2 )nr (g/h, g/h; x, p/h),
2m
(1.9)

where Hnr is the Hamiltonian


h
Hnr (hx2 + 2[gcth(x) + gth(x)]x ). (1.10)
2m
Physically speaking, m is the particle mass, h is Plancks constant, g and g are
coupling constants with dimension [action], x and p are position and momentum, resp.,
and is a scale parameter with dimension [position]1 . Thus, the parameters d, d and
variables v, v in (1.8) are dimensionless. Our function R(a+ , a , c0 , c1 , c2 , c3 ; v, v)
generalizes the reparametrized 2 F1 -function (1.8). As it stands, it depends on four more
(dimensionless) arguments than the rhs of (1.8), but it is scale invariant in the sense that

R(a+ , a , c; v, v) = R(a+ , a , c; v, v), > 0. (1.11)

(Here and in the sequel, we use c to denote (c0 , c1 , c2 , c3 ).) Hence, one might dispense
1
with the parameter a+ (say) by taking = a+ .
For fixed parameters a+ , a > 0 and (generic) real couplings c0 , . . . , c3 , the function
R is meromorphic in the variables v, v, with poles occurring solely on the imaginary
axis. It generalizes the rhs of (1.8), since one has

lim R(, , c; v, u) = 2 F1 (c0 + iu, c0 iu, c0 + c2 + 1/2; sh2 v), (1.12)


0

where

c0 (c0 + c1 + c2 + c3 )/2, (1.13)

in a sense discussed below. The R-function is a joint eigenfunction of four independent


analytic difference operators (henceforth A1Os) of AskeyWilson type, two acting on
v and the other two on v.
The fourA1Os will be detailed in Sect. 3. Here, we only mention one of the associated
analytic difference equations (from now on A1Es), trading R for the wave function

rel (, c; v, v) R(, , c; v, v/2). (1.14)


642 S. N. M. Ruijsenaars

In terms of physical quantities, this A1E can be reformulated as

1
(Hrel rel )( h, g/h; x, p) = ch(p)rel ( h, g/h; x, p), (1.15)
m 2

where Hrel is given by

1
Hrel [V (x)(Ti h 1) + V (x)(Ti h 1) + 2 cos (g0 + g1 + g2 + g3 )],
2m 2
(1.16)

with V the potential

sh(x ig0 ) ch(x ig1 ) sh(x ig2 i/2) ch(x ig3 i/2)
V (x) ,
shx chx sh(x i/2) ch(x i/2)
(1.17)

and T the translation operator

(T F )(x) F (x ), C. (1.18)

Consequently, one obtains

1
Hrel = + Hnr + C + O( 2 ), 0, (1.19)
m 2

so that the second scale parameter with dimension [momentum]1 can be viewed as
1/mc, with c the speed of light. Moreover, (1.12) can be rewritten as a nonrelativistic
limit,

lim rel (h/mc, (g0 , . . . , g3 )/h; x, p/mc)


c
= nr ((g0 + g2 )/h, (g1 + g3 )/h; x, p/h). (1.20)

To be sure, the physical picture implied by the above notation and terminology needs
further elaboration to be convincing to a physicist reader. It is however beyond the scope
of this paper to supply the background material concerning relativistic CalogeroMoser
N-particle systems, from which our 2 F1 -generalization originated. The interested reader
is referred to our 1994 lecture notes [5], in which the function R(a+ , a , c; v, v) was
already presented and discussed in relation to the latter integrable systems (introduced
in Ref. [6] at the classical and in Ref. [7] at the quantum level).
From a mathematical viewpoint no physical interpretation is needed, of course.
Rather, a mathematician may be inclined to try and tie in the novel function R (which
has various striking symmetry properties) with the representation theory of suitably con-
structed non-compact quantum groups. The latter viewpoint will not be further discussed
here either. Indeed, we study the function R in its own right, albeit with a slight bias
towards results that are relevant to a quantum-mechanical/Hilbert space context.
By contrast to the previous 2 F1 -generalizations mentioned above, we have not found
any representation for R as an explicit power series. Instead, our definition of R pro-
ceeds in terms of a contour integral that generalizes the Barnes representation for 2 F1
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 643

(cf. e.g. Ref. [8]). For later purposes it is convenient to specify the latter in the factorized
form
Z
2 F1 (f + iu, f iu, d + 1/2; sh v) = SL (v, z)M(d; z)SR (f ; u, z)dz. (1.21)
2
C

Here, the side functions are given by

SL (v, z) exp(iz ln(sh2 v)), (1.22)

0(f + iu iz) 0(f iu iz)


SR (f ; u, z) , (1.23)
0(f + iu) 0(f iu)
and the middle function by
0(iz) 0(d + 1/2)
M(d; z) . (1.24)
2 0(d + 1/2 iz)
Taking d, f, v > 0 and u R (for ease of exposition), the contour C runs along the real
axis from to and is indented downwards near z = 0 so as to avoid the simple
pole of 0(iz).
With a suitable restriction on the parameters, the contour C can also be used to define
the function R. First, we introduce

a (a+ + a )/2, c0 (c0 + c1 + c2 + c3 )/2, (1.25)

s1 c0 + c1 a /2, s2 c0 + c2 a+ /2, s3 c0 + c3 , s4 a, (1.26)

so that

s1 + s2 + s3 + s4 = 2c0 + 2c0 . (1.27)

Next we require

sj (a, a), j = 1, 2, 3, (1.28)

c0 , c0 (0, a), v, v R. (1.29)

In view of (1.27), this yields a non-empty open subset of the coupling space R4 . (The
restrictions are imposed for expository convenience; they will be relaxed below.) Now
we set
Z
1
R(a+ , a , c; v, v) = I (a+ , a , c; v, v, z)dz. (1.30)
(a+ a )1/2 C
The integrand I may be viewed as a product of fifteen functions of the form G(a+ , a ; ),
where G is defined by
 Z  
dy sin 2yz z
G(a+ , a ; z) = exp i , |Im z| < a. (1.31)
0 y 2sha+ y sha y a+ a y
644 S. N. M. Ruijsenaars

This function (referred to as the hyperbolic gamma function) was introduced and
studied in detail in Subsect. III A of Ref. [9]. It is manifestly analytic and zero-free in
the strip |Im z| < a, and obviously has the automorphy properties

G(a , a+ ; z) = G(a+ , a ; z), G(z)


= 1/G(z), G(a+ , a ; z) = G(a+ , a ; z), > 0. (1.32)

(Since G is symmetric in a+ , a , we often suppress these parameters; recall we fix


a+ and a in (0, ).) We will need various non-obvious properties of the hyperbolic
gamma function that are summarized in Appendix A.
We are now prepared to define the integrand in (1.30). It is convenient to factorize it
as

I F (a+ , a , c0 ; v, z)K(a+ , a , c; z)F (a+ , a , c0 ; v, z), (1.33)

where
G(z + y + ib ia) G(z y + ib ia)
F (a+ , a , b; y, z) , a = (a+ + a )/2,
G(y + ib ia) G(y + ib ia)
(1.34)

1 Y
3
G(isj )
K(a+ , a , c; z) . (1.35)
G(z + ia) G(z + isj )
j =1

(Note that K is not symmetric in a+ , a , since s1 and s2 are not, cf. (1.26).)
The restrictions (1.28) ensure that the poles of K in z lie above C. Similarly, the
restrictions (1.29) ensure that the poles of the functions F in z lie below C. Using the
pole sequence symbol explained in Appendix A, we have depicted the state of affairs in
the z-plane in Fig. 1 (taking s3 < s2 < s1 and 0 < v < v). From the G-asymptotics
detailed in Appendix A one easily deduces that the integrand has asymptotics
1 1
I (z) = O(exp[2z( + )]), Re z , (1.36)
a+ a
so that R is well defined. (Specifically, one need only use (A.29), (A.30) and (A.35) to
check this.)
Next, we turn to the connection with the hypergeometric function, written in the
Barnes representation (1.21)(1.24). To this end we specialize (1.30) as
Z
R(, , c; v, u) = SL (, c0 ; v, z)M(, c; z)SR (, c0 ; u, z)dz, (1.37)
C

where

SL (, c0 ; v, z) exp(2iz ln 2)F (, , c0 ; v, z), (1.38)


SR (, c0 ; u, z) exp(2iz ln(2))F (, , c0 ; u, z), (1.39)
 1/2

M(, c; z) exp(2iz ln(4))K(, , c; z). (1.40)

Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 645

ia is3

ia is2

ia is1

C
0

v i c0 v i c0

v ic0 v ic0

Fig. 1. The contour C vs. the upward and downward pole sequences in the z-plane

The point is now that the two zero step size limits (A.36) and (A.38) entail

lim SL (, c0 ; v, z) = SL (v, z), v > 0, (1.41)


0
lim SR (, c0 ; u, z) = SR (c0 ; u, z), (1.42)
0
lim M(, c; z) = M(c0 + c2 ; z), (1.43)
0

where the right-hand-side functions are given by (1.22)(1.24).


The limits (1.41)(1.43) hold true uniformly on compacts around the contour C, but
this does not yet suffice for the limit on the lhs of (1.12) to exist. For a complete proof of
(1.12), an L1 -bound on the integrand in (1.37) that is uniform for (0, 1] (say) would
be sufficient. Unfortunately, we have been unable to supply such a bound thus far.
In this connection we mention that we are not aware of complete proofs that the basic
hypergeometric function 2 1 and other generalizations converge to 2 F1 in the obvious
limit. Of course, termwise convergence is plain by inspection, but once more a uniform
l 1 -bound is needed to dominate the convergence. For 2 1 this might be relatively simple,
since a bound of the form |Cn (q)| K n (uniformly for q [1/2, 1), say) would suffice
to obtain uniform convergence on |w| K 1  for all  > 0. But for the Askey
Wilson 4 3 -generalization discussed below far more delicate bounds seem to be needed
to control the limit.
646 S. N. M. Ruijsenaars

Rigorous control on the limiting transitions R 2 F1 and 4 3 2 F1 is the only


point left open in this article complete proofs of all other results are supplied here
or in our previous paper [9]. A comprehensive and detailed picture of the analyticity
properties of the R-function, both in its parameters a+ , a , c and in its variables v, v is
critical not only in this paper, but also for further study. To arrive at the desired results,
we need a substantial generalization of the asymptotic properties of the G-function. As
already mentioned, Appendix A summarizes various results on the G-function that were
obtained in Ref. [9], but our sharpening of the asymptotics (laid down in Theorem A.1)
involves considerable technicalities that are relegated to Appendix C.
In Appendix B we obtain analyticity results on a class of functions that includes the
a priori least accessible part of the R-function, namely, the contour integral involving
the eight z-dependent G-functions, cf. (1.30). The generalization of this contour integral
studied in Appendix B is more easily handled than its special case, though it involves
a half-space restriction on the pertinent variables that is automatically satisfied for the
R-function.
In Sect. 2 we obtain various symmetry and analyticity properties of the R-function by
exploiting the information assembled in Appendixes AC. With these properties at our
disposal, the verification that R(a+ , a , c; v, v) is a joint eigenfunction of four Askey
Wilson type A1Os is a largely computational enterprise, which is undertaken in Sect. 3.
Similarly, the analyticity features enable us to show that the dual variable v may be taken
equal to i c0 + ina for n Z, yielding polynomials Pn (ch(2 v/a+ )) for n N.
In Sect. 4 we first present a Barnes type representation for s+1 s (1.3), which in-
volves the trigonometric G-function Gt (r, a; z) from Ref. [9]. This representation was
suggested by the above one involving the hyperbolic G-function G(a+ , a ; z). It dif-
fers from and is simpler to work with than the Barnes type representation for basic
hypergeometric functions that was introduced and studied long ago by Watson [10],
cf. also Ref. [1]. (The latter uses the q-gamma function, which is closely related to
our trigonometric G-function.) In particular, the analytic continuation of s+1 s to all
of the w-plane is quite easily studied by using our new representation. Taking s = 3
and suitably reparametrizing, we obtain a trigonometric analog Rt (r, a, c; v, v) of our
hyperbolic function R(a+ , a , c; v, v). Taking once again v equal to i c0 +ina, n N,
we obtain polynomials Pn (cos 2rv) whose relation to the AskeyWilson polynomials
pn (cos v) (1.6) is quite easily established.
The function Rt (r, a, c; v, v), however, does not satisfy an AskeyWilson type differ-
ence equation. On the other hand, it fails to do so by a quite simple difference, cf. (4.51)
below. Though this may not be clear at first sight, the inhomogeneous A1E thus found
is substantially equivalent to a result obtained first by Atakishiyev and Suslov [11],
cf. also Theorem 3 in Ref. [3]. Our proof proceeds by specializing a more general re-
sult (cf. Lemma 4.2), whose proof is patterned after the proof of the A1Es satisfied by
R(a+ , a , c; v, v) (cf. Theorem 3.1).

2. The R-Function: Automorphy and Meromorphy

With the restrictions (1.28), (1.29) in effect, we have already defined the R-function by
(1.30) and (1.33)(1.35). Before relaxing the restrictions, it is convenient to collect some
symmetry properties that can be read off from the integrand (1.33).
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 647

To this end we introduce matrices



1 0 0 0 1 1 1 1
0 0 1 0 1 1 1 1 1
I , J ,
0 1 0 0
(2.1)
2 1 1 1 1
0 0 0 1 1 1 1 1
satisfying
I J = J I, K = K = K 1 , K = I, J, (2.2)
and dual couplings given by
c J c = (c0 , c1 , c2 , c3 ). (2.3)
Setting
s1 c0 + c1 a /2, s2 c0 + c2 a+ /2, s3 c0 + c3 , s4 a, (2.4)
we then have
sj = sj , j = 1, 2, 3, 4, (2.5)
cf. (1.26). We also point out the equivalences
c = c c0 = c0 c0 = c1 + c2 + c3 . (2.6)
An inspection of (1.30)(1.35) now shows that R has the parameter symmetry
R(a+ , a , c; v, v) = R(a , a+ , I c; v, v) (2.7)
and self-duality property
R(a+ , a , c; v, v) = R(a+ , a , c; v, v). (2.8)
Combining these two transformations, one obtains
R(a+ , a , c; v, v) = R(a , a+ , I c; v, v). (2.9)
From the structure of the integrand (1.33) it is also clear that R is even in v and v:
R(a+ , a , c; v, v) = R(a+ , a , c; 1 v, 2 v), 1 , 2 = +, . (2.10)
Finally, the announced scale invariance (1.11) follows via the change of variables z
z, using the scale invariance of the G-function, cf. (1.32).
Clearly, these features are preserved under analytic continuation in the parameters
a , c and variables v, v. We proceed by examining such continuation properties. Fac-
toring off the seven z-independent G-functions whose analyticity properties are known
(cf. Appendix A), we are left with the auxiliary function
Z
A(a+ , a , c; v, v) I(a+ , a ; u1 , . . . , u4 , d1 , . . . , d4 , z)dz, (2.11)
C
with
Y
4
I(u, d, z) G(z uj )/G(z dj ), (2.12)
j =1
648 S. N. M. Ruijsenaars

u1 v ic0 + ia, u3 v i c0 + ia, (2.13)


2 4

dj isj , j = 1, 2, 3, 4. (2.14)
We are rewriting the integrand in this way so as to make clear that it is of the form
(B.2) with N = 4. In the case at hand we have, using (1.27),
X
4
(uj vj ) = 4ia. (2.15)
j =1

Thus we obtain an exponential decay


1 1
|I(a+ , a ; u, d, z)| < C | exp(2z( + ))|, z > L , (2.16)
a+ a
for all a RH P , c C4 and v, v C, provided we choose L large enough, cf. (B.8),
(B.9). Moreover, the constants C can be chosen uniformly for a in RH P -compacts,
c in C4 -compacts, and v, v in C-compacts, cf. Theorem A.1.
Taking a positive again, we recall that the restrictions (1.29) ensure that the pole
sequences uj ia zkl lie in the lower half plane, whereas (1.28) ensures that the
pole sequences dj + ia + zkl lie in the upper half plane, save for the simple pole at
d4 + ia = 0. The contour C equals (, ), but for a downward indentation to avoid
the latter d-pole.
Now we are of course free to deform any finite part of C continuously without changing
the value of the integral on the rhs of (2.11), as long as we do not cross any poles.
(Whenever this happens, residues are picked up.) Doing so in suitable ways, it is also
clear from the above decay bounds that we can freely continue a , c0 , . . . , c3 , v and
v into the complex plane (keeping a RH P , of course), provided the contour still
separates the u-poles and d-poles. To be more specific, the function (2.11) we are starting
from is real-analytic in a+ , a , c0 , . . . , c3 , v and v on the intervals specified in (1.28),
(1.29), and can be continued analytically in all of its variables as long as the contour C
can be deformed so that the u-poles stay below it and the d-poles above it.
Now this much is clear from the decay bounds (2.16) and the analytic character of the
integrand in (2.11). A priori, it is not at all clear, however, whether A(a+ , a , c; v, v)
(2.11) can be continued to all of the set
D RH P 2 C6 . (2.17)
Indeed, whenever u-poles and d-poles collide, the contour gets pinched and singularities
may arise.
Collisions of the u1 -poles and u2 -poles with the d-poles can be encoded in the
hyperplanes
j
Hlm, {ila+ + ima ia v isj + ic0 = 0}, l, m N ,
= +, , j = 1, 2, 3, (2.18)

Hlm,
4
{i(l 1)a+ + i(m 1)a v + ic0 = 0}, l, m N , = +, . (2.19)
j
Similarly, u3 - and u4 -pole collisions with the d-poles occur on hyperplanes Hlm, , given
by (2.18), (2.19) with v v, c c. Let us denote the union of the collision hyperplanes
by Zaux , and introduce the set
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 649

DA D \ Zaux . (2.20)

Note that this set is open and connected, but not simply-connected.
Now some reflection makes it plausible that A can be continued to DA , but whether
or not multi-valuedness occurs is not readily guessed. Moreover, the character of the
singularities (if any) when the collision variety Zaux is approached is far from clear by
inspection. We have studied this circle of problems in the slightly more general setting
of Appendix B. The most important results of the analysis to be found there, as applied
to the case at hand, are contained in the following lemma.

Lemma 2.1. The auxiliary function A(a+ , a , c; v, v) (2.11) extends to a (one-valued)


jointly analytic function in DA (2.20). Multiplying the latter function by the E-product

Y Y
4
E(a+ , a ; v ic0 + isj )E(a+ , a ; v i c0 + i sj ), (2.21)
=+, j =1

one obtains a function that has a jointly analytic extension to D (2.17).

Proof. The assertions in this lemma easily follow from Theorem B.2, specialized to the
above setting. u
t

Of course, Appendix B contains more information than is encoded in Lemma 2.1. In


particular, the values of A in DA can be written in terms of a contour integral and a finite
residue sum, and the properties and explicit form of residues are further illuminated in
Lemma B.1. Note also that Zaux is the zero locus of the E-product (2.21), and that the
extension to D is necessarily one-valued, since D (2.17) is simply-connected.
Let us next consider the seven z-independent G-functions in the integrand (1.33).
The product G(is1 )G(is2 )G(is3 ) occurs for purposes of normalisation, but is of course
irrelevant for the analytic behavior in the variables v and v. The G-product
Y
G(a+ , a ; v ic0 + ia)(v v, c0 c0 ), (2.22)
=+,

however, gives rise to pole and zero varieties that are closely related to the hyperplanes
Hlm,
4 (2.19) and their duals. Indeed, writing the G-functions in terms of E-functions by
using (A.40), one sees that the numerator E-functions amount to the j = 4 E-functions
in (2.21).
The denominator E-functions give rise to new singularities, as compared to the (a
priori) singular locus Zaux of A. As a notational preliminary, we introduce new couplings

0 c0 a+ /2 a /2, 1 c1 a /2, 2 c2 a+ /2, 3 c3 , (2.23)

dual couplings

J = (c0 a+ /2 a /2, c1 a /2, c2 a+ /2, c3 ), (2.24)

hyperplanes

Hlm, {ila+ + ima ia v i = 0},
l, m N , = +, , = 0, 1, 2, 3, (2.25)
650 S. N. M. Ruijsenaars

and dual hyperplanes



Hlm, {ila+ + ima ia v i = 0},
l, m N , = +, , = 0, 1, 2, 3. (2.26)
For = 1, 2, 3 the hyperplanes (2.25) and (2.26) coincide with (2.18) and their duals,
as anticipated by our notation. But for = 0 the hyperplanes encode the pole locus of
the G-product (2.22), whereas (2.19) and its dual encode the zero locus.
In the following theorem we collect various symmetry and analyticity properties of
the R-function that are readily established from the above. As a preparation, we define
the zero locus

Z 3=0 =+, l,mN (Hlm, Hlm, ), (2.27)
and the open, connected, but not simply-connected set
Dren = D \ Z. (2.28)
Q
Clearly, the normalisation factor j G(isj ) yields zero and pole sets
j
Nkl, {ka+ + la + a + sj = 0}, k, l N, = +, , j = 1, 2, 3, (2.29)
with union
j
N 3j =1 =+, k,lN Nkl, , (2.30)
and it suffices to study R on the set
DR Dren \ N , (2.31)
which is also open, connected and multiply-connected.
Theorem 2.2. The renormalized R-function
Y
3
Rren (a+ , a , c; v, v) G(a+ , a ; isj ) R(a+ , a , c; v, v), (2.32)
j =1

defined by (1.30) and (1.33)(1.35) with the restrictions (1.28), (1.29) in force, has a
(one-valued) jointly analytic extension to Dren (2.28). Multiplying the latter function by
the E-product
Y
3 Y
E(a+ , a ; v + i )E(a+ , a ; v + i ), (2.33)
=0 =+,

(with , given by (2.23), (2.24)), one obtains a function that has a jointly analytic
extension to D (2.17).
The function R(a+ , a , c; v, v) has a (one-valued) jointly analytic extension to DR
(2.31). It is scale invariant in the sense that (1.11) holds, and it satisfies (2.7)(2.10).
Fixing parameters (a+ , a , c) (RH P 2 C4 ) \ N , it is meromorphic in v and v, with
poles that can be located only at the points
iv = la+ + ma a , l, m N , = 0, 1, 2, 3, (2.34)

i v = la+ + ma a , l, m N , = 0, 1, 2, 3. (2.35)
The pole order at these points is smaller than or equal to the zero order of the E-function
product (2.33).
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 651

Proof. The assertions concerning Rren follow from Lemma 2.1 and the salient properties
of the G-product (2.22), as already explained above. The first assertion about R is then
an obvious corollary. The automorphy properties were already verified, subject to the
restrictions (1.28), (1.29); evidently, they continue to DR . Finally, the pole locations
(2.34), (2.35) encode the zero locus Z (2.27) of the E-product (2.33). Hence the last
assertion readily follows. u t

3. The R-Function: A1Es and AskeyWilson Specialization


With the above analyticity properties of the R-function at our disposal, we are now going
to demonstrate that the R-function is a joint eigenfunction of four independent analytic
difference operators of AskeyWilson type. To define the latter it is convenient to use
the notation
s (z) sh(z/a ), c (z) ch(z/a ), = +, , a+ , a RH P . (3.1)
Now we introduce the A1Os,
   
A (c; z) C (c; z) Tiaz 1 + C (c; z) Tia
z

1
+ 2c (i(c0 + c1 + c2 + c3 )), = +, . (3.2)
Here, the superscript z on the shifts T (1.18) is used to indicate that they act on the
variable z that also occurs in the coefficient functions
s (z ic0 ) c (z ic1 ) s (z ic2 ia /2) c (z ic3 ia /2)
C (c; z) .
s (z) c (z) s (z ia /2) c (z ia /2)
(3.3)

Theorem 3.1. For all parameters a RH P , c C4 , the function


Rren (a+ , a , c; v, v)
(2.32) is an eigenfunction of the A1Os,
A+ (c; v), A (I c; v), A+ (c; v), A (I c; v), (3.4)
(where I, c are defined by (2.1), (2.3)), with eigenvalues
2c+ (2v), 2c (2v), 2c+ (2v), 2c (2v). (3.5)

Proof. The analyticity properties of the coefficient functions are clear by inspection.
Combining them with the analyticity properties of Rren established in Theorem 2.2, we
deduce that we need only prove the assertions for parameters a , c0 , c0 > 0, sj = sj
(a, a), j = 1, 2, 3. Moreover, in view of the symmetry properties (2.7)(2.9), we need
only show that A+ (c; v) has eigenvalue 2c+ (2v) on R(a+ , a , c; v, v) for v, v RH P .
Now with the latter restrictions on the parameters and variables in effect, the upward
z-pole sequences in the integrand I (1.33) belong to i[0, ) and the downward ones to
the right and left half planes. Thus R is given by the contour integral (1.30) whenever
the downward sequences lie in the lower half plane, too.
More generally, we may work with (1.30), provided we deform C in obvious ways
so that the downward sequences stay below it. Moreover, we may shift the infinite tails
652 S. N. M. Ruijsenaars

of the contour over i iR as the need arises. (Indeed, the exponential decay (1.36) is
uniform for Im z varying over R-compacts, cf. Theorem A.1.) Denoting the deformed
contours once again by C, we deduce that a suitable choice of C ensures that shifts of the
variables v and v in R over ia amount to shifting v and v in the integrand I (1.33) of
the R-representation (1.30).
We are now prepared to consider the salient features of the integrand in regard to
shifts of the variables v, v and z over ia . Since it consists of a product of G-functions,
these features can be derived from the G-A1Es,
G(z + ia /2)
= 2c (z), = +, , (3.6)
G(z ia /2)
cf. (A.7), (A.9). As we are going to verify the A1E
A+ (c; v)R(c; v, v) = 2c+ (2v)R(c; v, v), (3.7)
with
Z
1
R(c; v, v) = F (c0 ; v, z)K(c; z)F (c0 ; v, z)dz, (3.8)
(a+ a )1/2 C

we may and will restrict attention to shifts over ia .


First, let us observe that F (1.34) satisfies the two A1Es,
F (b; y + ia /2, z) s+ (y + z + ib ia /2) s+ (y ib + ia /2)
= , (3.9)
F (b; y ia /2, z) s+ (y z ib + ia /2) s+ (y + ib ia /2)
F (b; y, z ia ) 1
= . (3.10)
F (b; y, z) 4s+ (y + z + ib ia )s+ (y z ib + ia )
(Indeed, these result from (3.6) with = .) Next, we use (3.9) with b = c0 , y = v, to
calculate the quotient
Q(c; v, z) (A+ (c; v)F )(c0 ; v, z)/F (c0 ; v, z). (3.11)
A critical fact is now that Q(c; v, z) can be rewritten
Q
4 4j =1 c+ (z ia /2 + isj )
2c+ (2z + 2i c0 ) + . (3.12)
s+ (v + z + ic0 ia )s+ (v z ic0 + ia )
This equality amounts to a functional equation that is not at all evident, and we postpone
its verification to the end of the proof for expository reasons.
Taking equality of Q and (3.12) for granted, we continue by demonstrating how the
equality can be exploited to prove the A1E (3.7). The denominator in (3.12) appears in
(3.10) with b = c0 , y = v. Thus we obtain
A+ (c; v)F (c0 ; v, z) = 2c+ (2z + 2i c0 )F (c0 ; v, z)
+ F (c0 ; v, z ia )5(c; z ia /2), (3.13)
where we have introduced the product

Y
4
5(c; z) 16 c+ (z + isj ). (3.14)
j =1
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 653

As a result of these calculations we obtain the identity


Z
1
A+ (c; v)R(c; v, v) = dz[2c+ (2z + 2i c0 )I (c; v, v, z)
(a+ a )1/2 C
+F (c0 ; v, z ia )5(c; z ia /2)K(c; z)F (c0 ; v, z)]. (3.15)
Now the exponential increase of c+ (2z+2i c0 ) for Re z is more than neutralized
by the exponential decay (1.36) of I (c; v, v, z). Likewise, the second term in square
brackets is O(exp[2z/a ]) for Re z , uniformly for Im z in R-compacts
(cf. Theorem A.1). Hence we are entitled to take z z + ia in the second term.
The key property of the function K(c; z) (1.35) is now that it satisfies the A1E
K(c; z + ia /2) = K(c; z ia /2)/5(c; z). (3.16)
(As before, this is easily verified from the G-A1E (3.6) with = .) Using this in the
(shifted) second term, we obtain
Z
1
A+ (c; v)R(c; v, v) = dz[2c+ (2z + 2i c0 )I (c; v, v, z)
(a+ a )1/2 C
+F (c0 ; v, z)K(c; z)F (c0 ; v, z + ia )]. (3.17)
Now from the A1E (3.10) with b = c0 , y = v we have
F (c0 ; v, z + ia ) = 4s+ (v + z + i c0 )s+ (v z i c0 )F (c0 ; v, z)
= 2[c+ (2v) c+ (2z + 2i c0 )]F (c0 ; v, z). (3.18)
Substituting this in (3.17), we finally obtain the announced A1E (3.7), cf. (3.8).
It remains to prove the functional equation announced above. From the definition
(3.11) and the A1E (3.9) we deduce that Q(c; v, z) equals
c+ (v ic1 )s+ (v ic2 ia /2)c+ (v ic3 ia /2)
s+ (v)c+ (v)s+ (v ia /2)c+ (v ia /2)
 
s+ (v z ic0 )s+ (v + ic0 ia )
s+ (v ic0 ) + (v v)
s+ (v + z + ic0 ia )
+2c+ (2i c0 ). (3.19)
Our remaining task is to show that this function equals the function (3.12).
We begin by observing that both functions are ia+ -periodic, even and meromorphic
in v. For Re v the function (3.19) has asymptotics

exp (ic1 ic2 ic3 )[exp (2z ic0 ) exp (ic0 )]
a+ a+ a+
+(i i, z z) + 2c+ (2i c0 ) = 2c+ (2z + 2i c0 ). (3.20)
Obviously, this is also true for (3.12). By virtue of Liouvilles theorem, it therefore
suffices to prove that both functions have equal residues at v-poles in a period strip.
(Indeed, the poles are all (generically) simple.)
Now the factors in (3.19) have poles for v = 0, ia+ /2, ia /2 and i(a+ + a )/2
that do not occur for (3.12). But the first square bracket function vanishes for v =
ia /2, i(a+ + a )/2 and the second one for v = ia /2, i(a+ + a )/2, so no
poles occur for v = ia /2, i(a+ + a )/2. The first and second term in (3.19) yield
654 S. N. M. Ruijsenaars

functions that do have poles for v = 0, ia+ /2, but it is straightforward to check that the
residues cancel.
We are, therefore, reduced to comparing the residues of (3.12) and (3.19) at the poles
v = z + ic0 ia . By evenness, we need only calculate the residue of (3.19) for the
upper sign pole. It reads

c+ (z + i(c0 + c1 a ))s+ (z + i(c0 + c2 a /2))c+ (z + i(c0 + c3 a /2))


41 s+ (2z + 2ic0 2ia )s+ (2z + 2ic0 ia )
s+ (2z + 2ic0 ia )s+ (z). (3.21)

Hence it equals the residue of (3.12) for v = z + ic0 ia , as announced. u


t

We continue by showing how polynomials arise upon discretizing v. Specifically, let


us consider the functions

Rn (v) R(a+ , a , c; v, i c0 + ina ), n Z, (3.22)

where we restrict the parameters by requiring

ei+ a+ > 0, + (/2, 0), a > 0, (3.23)

c0 , c0 (0, Re a), sj (Re a, Re a), j = 1, 2, 3. (3.24)

These restrictions can easily be relaxed. However, they are convenient for expository
reasons and guarantee that non-generic singularities are avoided. Indeed, they entail that
R has no poles at the pertinent v-values, cf. (2.35).
By contrast, the v-values i c0 + ina with n N always belong to the singularity
locus Zaux of the auxiliary function A(v, v) (2.11): They amount to the hyperplanes
Hn+1,1,+
4 , n N, cf. (the dual of) (2.19). As such, they encode collisions of u3 -poles
with d4 -poles, cf. (2.13), (2.14).
In order to elaborate on the special character of these v-values, we start from the R-
representation (1.30), with C the contour defined below (1.24), taking at first v, v R.
We denote the contour with the opposite indentation by C + . (Thus, C + runs along the
real axis from to , with an upward indentation at z = 0.) The pole of the integrand
I (1.33) at z = 0 is simple, and using (A.20), (1.34) and (1.35) we obtain the residue
(a+ a )1/2 /2 i. Hence we deduce
Z
1
R(v, v) = 1 + I (v, v, z)dz, v, v R. (3.25)
(a+ a )1/2 C +

Now we may continue v to Im v < c0 without v-dependent poles hitting C + . Thus the
representation (3.25) holds true in this v-region, and in view of the upward indentation we
may as well include an open disc around v = i c0 . Doing so, the factor 1/G(v+i c0 ia)
in I vanishes for v i c0 , and so we conclude

R0 (v) = 1. (3.26)

(Note that the z = 0 residue of the A-integrand I (2.12) is indeed singular for v = i c0 .)
We are now prepared for the following result.
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 655

Theorem 3.2. With the parameters a , c restricted by (3.23), (3.24), the function Rn (v)
(3.22) satisfies

Rn (v) = Pn (ch(2v/a+ )), n N, (3.27)

where Pn (u) is a polynomial of degree n in u.

Proof. As we have seen above, the relevant arguments of the R-function belong to its
analyticity domain DR (2.31). Specializing the A1E

A+ (c; v)R(c; v, v) = 2c+ (2v)R(c; v, v), (3.28)

(cf. Theorem 3.1), we obtain

C+ (c; i c0 + ina )[Rn1 (v) Rn (v)] + C+ (c; i c0 ina )[Rn+1 (v) Rn (v)]
+2c+ (2ic0 )Rn (v) = 2c+ (2v)Rn (v), n Z. (3.29)

The crux is now that we have not only

C+ (c; i c0 ) = 0, (3.30)

cf. (3.3), but also R0 = 1, cf. (3.26). Hence the polynomial character of Rn (v) follows
recursively from (3.29). (Note that none of the coefficients C+ (c; i c0 ina ) vanishes
for n N.) u t

In view of the symmetry properties (2.7)(2.10), there are various other ways to obtain
polynomials from the R-function by discretizing v or v. Staying with the polynomials
obtained above, we continue by pointing out that they satisfy the A1E

A+ (c; v)Pn (ch(2v/a+ )) = 2ch(2i(c0 + na )/a+ )Pn (ch(2 v/a+ )), (3.31)

in addition to the three-term recurrence relation (3.29). (Indeed, (3.31) is simply the A1E
(3.7) for the special v-values at issue.) It should also be observed that the A1O A (I c; v)
acts trivially on the polynomials: Since they are ia+ -periodic in v, its eigenvalue is given
by 2c (2ic0 ) for all n N. (The fourth A1O A (I c; v) in Theorem 3.1 does not act
on the polynomials, since it shifts v by ia+ .)
From the coefficients in the recurrence (3.29) and A1E (3.31) it can be read off
that we may continue a+ to all of the lower half plane. In particular, we may switch to
parameters

a a , r i/a+ , a, r > 0. (3.32)

Then we obtain polynomials Pn (r, a, c; cos(2rv)) that satisfy the A1E

A(r, a, c; v)Pn = 2ch2r(c0 + na)Pn , n N, (3.33)

and the recurrence relation

an (r, a, c)Pn+1 + bn (r, a, c)Pn + cn (r, a, c)Pn1 = 2 cos(2rv)Pn . (3.34)

Here, the A1O A(r, a, c; v) is given by


 
A C(r, a, c; v) Tiav 1 + C(r, a, c; v) Tia
v
1 + 2ch(2r c0 ), (3.35)
656 S. N. M. Ruijsenaars

sin r(v ic0 ) cos r(v ic1 ) sin r(v ic2 ia/2) cos r(v ic3 ia/2)
C(v) ,
sin rv cos rv sin r(v ia/2) cos r(v ia/2)
(3.36)
and the recurrence coefficients read
shr(2c0 + na) chr(c0 + c1 + na) shr(c0 + c2 + (n + 1/2)a)
an
shr(c0 + na) chr(c0 + na) shr(c0 + (n + 1/2)a)
chr(c0 + c3 + (n + 1/2)a)
, (3.37)
chr(c0 + (n + 1/2)a)

shr(na) chr(c0 c1 + na) shr(c0 c2 + (n 1/2)a)


cn
shr(c0 + na) chr(c0 + na) shr(c0 + (n 1/2)a)
chr(c0 c3 + (n 1/2)a)
, (3.38)
chr(c0 + (n 1/2)a)

bn 2chr(2c0 ) an cn . (3.39)

4. A Barnes Type Representation for Basic Hypergeometric Functions


The precise relation between the polynomials Pn (cos 2rv) just studied and the Askey
Wilson polynomials pn (cos v) (1.6) is not yet clear at this stage. In principle, this relation
can be determined directly by comparing the difference equations and three-term recur-
rence relations satisfied by both sets of polynomials. But the relation will be an obvious
consequence of results obtained below, and these results are of independent interest.
Therefore, we postpone a comparison of the polynomials to the end of this section. In-
deed, guided by the above integral representation involving the hyperbolic G-function
from Ref. [9], we present a novel and quite simple integral representation for the basic
hypergeometric function s+1 s (1.3), which involves the trigonometric G-function from
Ref. [9].
Taking s = 3 and choosing appropriate parameters, we obtain once more an inter-
polation of the above polynomials Pn (cos 2rv) that is essentially self-dual. But in the
present setting an AskeyWilson type difference equation only holds true when the dual
variable v is discretized in such a way that the polynomials arise. We prove the pertinent
results in basically the same way as in Sect. 3, obtaining further explicit information in
the process.
Turning to the details, we collect first of all some formulas involving the trigonometric
G-function from Ref. [9] that we have occasion to use. Throughout this section we fix
parameters r, a > 0. Then the G-function is defined by

Y
Gt (r, a; z) = (1 exp[(2m 1)ar + 2irz])1 . (4.1)
m=1

(We require positivity of r and a more for brevity than for necessity. The results that fol-
low can readily be generalized to the region Re ar > 0.) Thus Gt (r, a; ) is meromorphic
and has no zeros, while Gt has simple poles at
zj k = j /r ia(k + 1/2), j Z, k N. (4.2)
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 657

It is not hard to verify that Gt satisfies the A1E


Gt (z + ia/2)
= 1 exp 2irz, (4.3)
Gt (z ia/2)
and that it can be rewritten

!
X exp 2inrz
Gt (r, a; z) = exp , Im z > a/2. (4.4)
2nshnra
n=1

The series representation (4.4) entails that for z = x R the functions Gt and 1/Gt
have Fourier expansions of the form
X
G1
t (r, a; x) = 1 + cn (ra) exp 2inrx, x R. (4.5)
nN

Moreover, it readily yields the asymptotics

G1
t (r, a; z) = 1 + O(exp(2rIm z)), Im z , (4.6)

uniformly for Re z R.
In our previous paper Ref. [9] we obtained some further results, and we clarified
the relation to the q-gamma function. But the above information on the trigonometric
G-function (which is quite easily derived) is all we need. Our integral representation
for s+1 s involves one more ingredient, however. This is the renormalized Weierstrass
-function

s(r, a; z) (z; /2r, ia/2) exp(z2 r/ ), (4.7)

which we already employed in Ref. [9]. It is an entire, odd and /r-antiperiodic function
that obeys the A1E
s(z + ia/2)
= exp(2irz). (4.8)
s(z ia/2)
Again, these are the only properties we need.
The integrand involves quotients of G-functions, and the function

E(, z) s(z + i/2r)/s(z). (4.9)

Clearly, the latter is entire and 2i -antiperiodic in . Also, it is /r-periodic and mero-
morphic in z, with simple poles in the period lattice P = Z/r + iaZ. To be quite
precise, this is the case for not equal to a point in the lattice 2irP; for the latter -
values one infers from the A1E (4.8) that E(, z) is a multiple of exp(2ikrz), k Z.
This A1E also entails that E satisfies

E(, z) = e E(, z ia). (4.10)

Consider now the integrand

I (, , , z) E(, z)G(, , z), (4.11)


658 S. N. M. Ruijsenaars

where G denotes the following product of G-quotients:


s+1 s
Gt (ia/2) Y Gt (z + j ) Y Gt (k )
G(, , z) . (4.12)
Gt (z + ia/2) Gt (j ) Gt (z + k )
j =1 k=1

Introducing the domains

DM { Cs+1 | Im j > a(M + 1/2), j = 1, . . . , s + 1}, (4.13)


it is convenient to require at first (, , ) D0 Cs C. Then the poles of the factors
Gt ( + j ) lie in the (open) lower half plane, cf. (4.2). The poles of E(, z) in the lower
half plane are canceled by the zeros of 1/Gt (z + ia/2).
Now let 0 denote a contour from z = /2r to z = /2r such that the j -
dependent poles are below 0 and the poles at z = ina, n N, are above 0 . (For
instance, one can let 0 run along the real axis from /2r to /2r with a downward
indentation at z = 0.) Then the auxiliary function
s
Y Z
1
A(, , ) Gt (k ) I (, , , z)dz (4.14)
k=1 0

is clearly well defined and analytic in D0 Cs C.

Theorem 4.1. The function A(, , ) extends to a function that is analytic in C2s+2 .
For Re < 0 the function
s
Y
8(, , ) [2is(i/2r)]1 Gt (k ) A(, , ) (4.15)
k=1

satisfies

8(, , ) =s+1 s (e2ir1 ar , . . . , e2irs+1 ar , e2ir1 ar , . . . , e2irs ar ; e2ar , e ),


(4.16)
where s+1 s is defined by (1.3)(1.4).

Proof. Fixing in the domain D0 and Cs , we first prove (4.16) for (2ar, 0).
In the process, we obtain a representation for 8 from which the asserted analytic con-
tinuation property of A can be read off.
We begin by introducing contours
0k = {z = x + ia(k 1/2) | x [/2r, /2r], k N }. (4.17)
Since the integrand in (4.14) is /r-periodic in z, we may shift 0 to 01 , picking up the
residue at the simple pole z = 0. From (4.7) we have

s(z) = z + O(z3 ), z 0, (4.18)


so we deduce
Z
1
8(, , ) = 1 + [2is(i/2r)] I (, , , z)dz. (4.19)
01
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 659

Next, we replace the factor E(, z) in I by e E(, zia) (recall (4.10)), and shift 01
to 02 . Using the G-A1E (4.3), the values of the G-quotients at the simple pole z = ia
are easily calculated, yielding
s+1
Y s
Y
8(, , ) = 1 + e (1 e2irj ar )/(1 e2ar ) (1 e2irk ar )
j =1 k=1
Z
+[2is(i/2r)]1 e E(, z ia)G(, , z)dz. (4.20)
02

Iterating this procedure, one readily obtains


N Qs+1 2ir ar 2ar
X j =1 (e
j ;e )n
n Q
8(, , ) = e 2ar 2ar s + RN (, , ),
(e ;e )n k=1 (e k ; e2ar )n
2ir ar
n=0
(4.21)
where the remainder term reads
Z
1 N
RN (, , ) [2is(i/2r)] e E(, z iN a)G(, , z)dz. (4.22)
0N +1

Now on the contour 0N+1 we have from (4.17),


s(x + ia/2 + i/2r)
E(, z iNa) = , x [/2r, /2r]. (4.23)
s(x + ia/2)
Thus E(, z iNa) is bounded by an N-independent constant on 0N +1 . Moreover, the
z-dependent G-functions in G(, , z) converge uniformly to 1 on 0N +1 as N ,
cf. (4.6). Since (2ar, 0), we deduce that RN (, , ) converges to 0 for N .
Thus we obtain (4.16) for (2ar, 0). Q
Next, we multiply (4.21) by 2is(i/2r) sk=1 Gt (k )1 , so that we obtain
A(, , ) on the lhs, cf. (4.15). On the rhs we get a function that is clearly analytic for
(, , ) DN +1 Cs C. Thus the theorem readily follows. u t
It should be observed that this theorem yields additional information. First, recall that
the rhs of (4.16) is analytic for Re < 0, so that the same holds true for 8(, , ). At
first sight this seems to disagree with the poles of the factor 1/s(i/2r) for Re < 0. But
for such the function E(, z) reduces to a multiple of exp(2ikrz), k N . Thus we
may replace 0 by [/2r, /2r]. Recalling now (4.5), one deduces that the integral
vanishes for the pertinent . Hence the rhs of (4.16) yields the limit value of the quotient.
For such that Re 0 and s(i/2r) = 0, however, the integral does not vanish in
general. In particular, we have
Z Z /2r Y s Y s+1

I (, , 0, z)dz = G(, , x)dx = Gt (ia/2) Gt (k )/ Gt (j ).
0 /2r r
k=1 j =1
(4.24)
Thus 8(, , ) has a simple pole at = 0, with a residue given by
s
Y s+1
Y
Gt (ia/2) Gt (k )/ Gt (j ). (4.25)
k=1 j =1
660 S. N. M. Ruijsenaars

Since A(, , ) is entire and 2i-antiperiodic in , it follows more generally from


(4.15) that the function 8(, , ) has simple poles at
= 2ark + 2il, k N, l Z, (4.26)
and is 2i-periodic in . Obviously, this has consequences for the basic hypergeometric
series (1.3): The analytic function in the unit disc defined by it has a meromorphic
continuation, with simple poles located only at
w = q k , k N. (4.27)
Furthermore, the residue at w = 1 follows from (4.25) by using
q = exp(2ar), aj = exp(2irj ar), bk = exp(2irk ar). (4.28)
In order to tie in the above function 8(, , ) with the AskeyWilson A1O (3.35),
we proceed by specializing the above variables. Specifically, for the remainder of this
section we take s = 3 and substitute
1 v + ic0 ia/2, 3 v + i c0 ia/2, (4.29)
2 4

j itj , j = 1, 2, 3, (4.30)

t1 c0 + c1 i/2r a/2, t2 c0 + c2 , t3 c0 + c3 + i/2r, t4 a/2,


(4.31)
in 8(, , ). This yields a function (c, ; v, v) that is entire in v and v and mero-
morphic in c0 , . . . , c3 , with poles that are obvious from (4.15) and the first assertion
of Theorem 4.1. Moreover, is 2i-periodic and meromorphic in , with simple poles
located only at (4.26).
It is convenient to require at first
Re c0 > 2a, Re c0 > 2a, Im v (a, a), Im v (a, a), (2ar, 0).
(4.32)
Indeed, in that case we may write
Z
(c, ; v, v) = [2is(i/2r)]1 I (c, ; v, v, z)dz, (4.33)
0
where
I = F (c0 ; v, z)K(c, ; z)F (c0 ; v, z), (4.34)

Gt (z + y + ib ia/2) Gt (z y + ib ia/2)
F (b; y, z) , (4.35)
Gt (y + ib ia/2) Gt (y + ib ia/2)

Y
4
Gt (itj )
K(c, ; z) E(; z) . (4.36)
Gt (z + itj )
j =1

Since the function (c, ; v, v) is entire in v, the AskeyWilson A1O A(c; v) (3.35) has
a well-defined action on it. Now the restrictions (4.32) ensure that the four downward
pole sequences have a distance larger than a from the real axis. Therefore, the action of
the shifts Tia amounts to taking v v ia in the integral on the rhs of (4.33). We are
now prepared for the following auxiliary result.
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 661

Lemma 4.2. With the restrictions (4.32) in effect, one has


A(c; v)(c, ; v, v) = 2 cos(2r v)(c, ; v, v)
Z
[e+2ar 1]
+ [cos(2r v) cos(2rz + 2ir c0 )]I dz. (4.37)
is(i/2r) 0
Proof. The proof of this lemma runs parallel to that of Theorem 3.1, with the G-A1E
(4.3) playing the role of the A1E (3.6) with = , the functions
c(v) cos rv, s(v) sin rv, (4.38)
corresponding to c+ (v) and s+ (v), and a to a. Indeed, as the analogs of (3.9) and (3.10)
we obtain
F (b; y + ia/2, z) s(y + z + ib ia/2) s(y ib + ia/2)
= , (4.39)
F (b; y ia/2, z) s(y z ib + ia/2) s(y + ib ia/2)
F (b; y, z ia) exp 2ir(z ib + ia)
= . (4.40)
F (b; y, z) 4s(y + z + ib ia)s(y z ib + ia)
Calculating now
Q(c; v, z) (A(c; v)F )(c0 ; v, z)/F (c0 ; v, z), (4.41)
we obtain (3.19) with c+ , s+ , a c, s, a. Equivalently, we obtain (3.19) with
a+ = i/r, a = a. (4.42)
From this one easily deduces that Q(c; v, z) equals (3.12) with (4.42) in effect.
Recalling (1.26) and (4.31), we now obtain
Z
A(c; v)(c, ; v, v) = [2is(i/2r)]1 dz[2c(2z + 2i c0 )I (c, ; v, v, z)
0
+F (c0 ; v, zia)5(c; zia/2)K(c, ; z)F (c0 ; v, z)], (4.43)

Y
4
5(c; z) 16 exp[2ir(z + ic0 ia/2)] s(z + itj ), (4.44)
j =1

as the analogs of (3.15) and (3.14). Next, we take z z+ia in the integrand of the second
term on the rhs of (4.43). (This is allowed on account of (4.32) and /r-periodicity.) As
the analog of (3.16) we calculate
K(c, ; z + ia/2)/K(c, ; z ia/2) = exp( 2irz + 2r c0 + ar)/5(c; z),
(4.45)
so we deduce
Z
A(c; v)(c, ; v, v) = [2is(i/2r)]1 dz[2c(2z + 2i c0 )I (c, ; v, v, z)
0
F (c0 ; v, z) exp( 2irz + 2r c0 + 2ar)K(c, ; z)F (c0 ; v, z + ia)]. (4.46)
Finally, using (4.40) we obtain
F (c0 ; v, z + ia) = 2 exp(2ir(z + i c0 ))[c(2v) c(2z + 2i c0 )]F (c0 ; v, z). (4.47)
Substituting this in (4.46), we obtain (4.37). u
t
662 S. N. M. Ruijsenaars

We are now prepared to define and study a trigonometric analog of the hyperbolic
function R(c; v, v): It reads
Rt (c; v, v) (c, 2ar; v, v). (4.48)
We also define
Y
4
D(c; v, v) exp(2r c0 ) Gt (itj )
j =1
1
Y
Gt (v + ic0 ia/2)Gt ( v + i c0 ia/2) .
=+,
(4.49)
Q3
Theorem 4.3. The renormalized function j =1 Gt (itj )1 Rt (c; v, v) is entire in
c0 , . . . , c3 , v and v. The function Rt satisfies the self-duality relation
Rt (c; v, v) = Rt (c; v, v), (4.50)
and the A1Es
A(c; v)Rt (c; v, v) = 2 cos(2r v)Rt (c; v, v) + D(c; v, v), (4.51)

A(c; v)Rt (c; v, v) = 2 cos(2rv)Rt (c; v, v) + D(c; v, v). (4.52)


Moreover, it has a specialization
Rt (c; v, i c0 + ina) = Pn (cos(2rv)), n N, (4.53)
where P0 (u), P1 (u), . . . are the polynomials from Theorem 3.2. Finally, one has
Rt (c; v, v) = 4 3 (e2ir(v+ic0 ) , e2ir(v+ic0 ) , e2ir(v+i c0 ) , e2ir(v+i c0 ) ,
e2r(c0 +c1 ) , e2r(c0 +c2 +a/2) , e2r(c0 +c3 +a/2) ; e2ar , e2ar) . (4.54)
Proof. The first assertion follows upon specializing the first assertion of Theorem 4.1.
From (4.33) we read off
(c, ; v, v) = (c, ; v, v), (4.55)
so (4.50) is clear from (4.48). Next, we prove (4.51), using Lemma 4.2.
To this end we need only show that the second term on the rhs of (4.37) has limit
D(c; v, v) for 2ar. Now the prefactor has limit 2r/ s 0 (ia). Using the s-A1E
(4.8) and (4.18), we obtain s 0 (ia) = ear . To determine the limit of the integral, we
recall (4.34)(4.36) and (4.9): We have
E(2ar, z) = s(z ia)/s(z) = ear e2irz , (4.56)
by virtue of (4.8). Therefore the limit of the second term becomes
Z Y
4
2r Gt (itj )
dz[cos(2r v)cos(2rz+2ir c0 )]e2irz F (c0 ; v, z) F (c0 ; v, z).
0 Gt (z + itj )
j =1
(4.57)
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 663

We may now replace 0 by [/2r, /2r] and invoke (4.5) to deduce that (4.57) equals
D(c; v, v) (4.49). Hence (4.51) follows.
Clearly, (4.52) follows by combining (4.50) and (4.51), so we proceed by proving
(4.53). The key point is that we have

D(c; v, i c0 + ina), D(c; i c0 + ina, v) = 0, n N, (4.58)

cf. (4.49). Therefore, we obtain the A1E (3.33) and recurrence (3.34) satisfied by the
polynomials Pn (cos(2rv)) when we specialize (4.51) and (4.52) to v = i c0 + ina with
n N. Hence we need only show

Rt (c; v, i c0 ) = 1 (4.59)

for (4.53) to result.


To prove (4.59), we note that we may flip the indentation of 0 in (4.33), picking
up a residue 1. Then we can let v converge to i c0 , to obtain a zero from the term
1/Gt (v + i c0 ia/2). As a consequence we obtain

(c, ; v, i c0 ) = 1, (4.60)

and so (4.59) follows by taking 2ar.


The remaining assertion (4.54) follows by combining (4.16) with the pertinent sub-
stitutions, cf. (4.29)(4.31). Hence the proof of the theorem is complete. u
t

Comparing (4.53) and (4.54) to (1.6), the relation between the polynomials
Pn (cos 2rv) and the AskeyWilson polynomials pn (cos v) can be read off: When we
take r = 1/2 in the former and substitute

q = ea , = ec0 , = ec1 , = ec2 a/2 , = ec3 a/2 , (4.61)

in the latter, we obtain

Nn1 pn (cos v) = Pn (cos v), n N. (4.62)

Appendix A. The Hyperbolic Gamma Function and Related Functions

In a previous paper [9] we presented a new approach to the theory of first order analytic
difference equations. As an application, we introduced generalized gamma functions of
hyperbolic, trigonometric and elliptic type Eulers gamma function 0(z) being of ratio-
nal type. Our trigonometric gamma function is closely related to the q-gamma function,
as detailed in Ref. [9]. We have recently become aware of the fact that our hyperbolic
gamma function is not essentially new either: It is basically equal to Kurokawas double
sine function [12,13]. In turn, the latter is a quotient of double gamma functions a
function introduced and studied in great detail by Barnes almost a century ago [14,15].
We will clarify the latter relations towards the end of this appendix. The main purpose
of this appendix is, however, to summarize and extend the results from Ref. [9] that we
need for a detailed study of our generalized hypergeometric function, which is built from
hyperbolic gamma functions. The pertinent results are mostly new within the context of
the double sine function, too, and indeed the viewpoint on this function coming from
Ref. [9] (and on the entire function E(z) introduced later on) is quite different from that
664 S. N. M. Ruijsenaars

of Barnes and later authors dealing with the double gamma and sine functions [1618,
4].
To begin with, consider the integral
Z  
dy sin 2yz z
g(a+ , a ; z), (A.1)
0 y 2sha+ y sha y a+ a y
where we take at first a , a+ (0, ). Defining the strip
S {z C||Im z| < a}, a (a+ + a )/2, (A.2)
it is clear that the integral converges absolutely and uniformly on compact subsets of S.
Therefore, g(a+ , a ; z) is analytic in S. The hyperbolic gamma function is now defined
by
G(a+ , a ; z) exp(ig(a+ , a ; z)). (A.3)
It is obviously analytic and zero-free in S. We proceed by listing further properties that
are important in this paper, referring to Sect. III A of Ref. [9] for complete proofs.
First of all, it is not obvious, but true that G extends to a meromorphic function of z.
It is obvious from (A.1) that G has the features
G(a+ , a ; z) = 1/G(a+ , a ; z), (A.4)
G(a , a+ ; z) = G(a+ , a ; z), (A.5)
G(a+ , a ; z) = G(a+ , a ; z), (0, ). (A.6)
From our viewpoint, the key property of G is that it is the unique minimal solution to
the first order A1E
G(z + ia+ /2)
= 2ch( z/a ) (A.7)
G(z ia+ /2)
that satisfies
G(0) = 1, G(z) > 0, z i(a, a). (A.8)
(The notion of minimal solution is defined in Ref. [9].) In view of the a+ a
symmetry of G, it may equivalently be defined as the unique minimal solution to
G(z + ia /2)
= 2ch( z/a+ ) (A.9)
G(z ia /2)
satisfying (A.8).
It readily follows from (A.1) that for fixed z S the function G(a+ , a ; z) is real-
analytic for a+ , a (0, ). Far stronger analyticity properties, however, can be read
off from a representation of G in terms of an infinite product of 0-functions. In order to
detail these properties, we define the quotient
a /a+ , (A.10)
the cut plane
C { C |
/ (, 0]}, (A.11)
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 665

ia

ia

Fig. 2. The zero and pole sequences of G(a+ , a ; z) for a+ , a positive

and the complex numbers


zkl ika+ + ila , k, l N. (A.12)

Provided stays in C , and (a+ , a , z) stays away from the hyperplanes z = zkl in C3 ,
with

zkl ia zkl , k, l N, (A.13)
the function G(a+ , a ; z) continues to a one-valued jointly analytic function of (a+ , a ,
+
z), vanishing solely on the hyperplanes z = zkl , with
+
zkl zkl , k, l N. (A.14)
Moreover, fixing a+ , a , the function G(a+ , a ; ) is meromorphic with poles only for

z = zkl . We symbolize the pole and zero sequences of G(a+ , a ; ) for positive a+ , a
in Fig. 2.

The infinite product representation also yields the pole order for z equal to zkl , which
we denote by O(kl) from now on. Specifically, the latter equals the number of distinct
pairs (m, n) N2 such that zmn = zkl . In particular, all poles are simple for Im 6 = 0,
and also for positive and irrational. We have occasion to use the inequality
O(k + m, l + n) O(kl) + O(mn) 1, k, l, m, n N. (A.15)
This inequality cannot be found in Ref. [9], so we supply its simple proof here.
Clearly, (A.15) holds true when all poles are simple. Thus we need only consider the
case
= s/t, s, t N coprime. (A.16)
Then any zij , i, j N, can be uniquely written as z with {0, . . . , s 1}, and we
have
O(ij ) = [/t] + 1, (A.17)
where [] is the entier function. Hence, letting
zkl = z1 ,1 , zmn = z2 ,2 , zk+m,l+n = z3 ,3 , 1 , 2 , 3 {0, . . . , s 1},
(A.18)
666 S. N. M. Ruijsenaars

the inequality (A.15) is equivalent to


[3 /t] [1 /t] + [2 /t]. (A.19)
Now if 1 + 2 {0, . . . , s 1}, then we have 3 = 1 + 2 and 3 = 1 + 2 , so (A.19)
is clear. In case 1 +2 {s, . . . , 2s 2}, we have 3 = 1 +2 s and 3 = 1 +2 +t.
Hence (A.19) follows once more, and so the proof of (A.15) is complete.

Obviously, the pole at z00 = ia is always simple. Its residue can be determined
explicitly and reads
i
r00 = (a+ a )1/2 . (A.20)
2

More generally, the pole at zkl is simple iff the quantity
k
Y l
Y
tkl sin(ma+ /a ) sin( na /a+ ) (A.21)
m=1 n=1

is non-zero; assuming it is, the residue at the pole reads

rkl = ()kl (1/2)k+l r00 /tkl . (A.22)


Of course, corresponding properties of the zeros of G(z) follow from the relation
+
(A.4). Moreover, the latter entails that at a simple pole zkl of 1/G(z) the residue equals
rkl .
Next, we turn to the asymptotics of G(a+ , a ; z) for Re z . It so hap-
pens that we need an extension of our results obtained in Ref. [9], where we fixed
a+ , a (0, ). Here we allow a+ , a to vary over the (open) right half plane, hence-
forth denoted by RH P . Accordingly, an extensive elaboration on our previous arguments
regarding asymptotics appears inevitable. Recalling G = exp ig, we need only study
the g-asymptotics. Moreover, since g is odd in z (cf. (A.1)), it suffices to obtain the
Re z asymptotics.
Now since we take a+ , a RH P , the poles and zeros of G(z) stay in the lower and
upper half plane, and g(z) extends from the real axis (where the integral representation
(A.1) is still valid) to an analytic function in the complex plane with two wedges deleted.
(Note that g has logarithmic branch points at the poles and zeros of G.)
To be specific, let us first set
a = r exp(i ), r > 0, (/2, /2), (A.23)

max max(+ , ), min min(+ , ), (A.24)


so that
a = (a+ + a )/2 = r exp(i), r > 0, [min , max ]. (A.25)
Defining the wedge
W {iR exp(i) | R [0, ), [min , max ]}, (A.26)
it suffices to delete the two wedges
W+ ia + W, W W+ , (A.27)
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 667

in the upper and lower half planes. Indeed, g(a+ , a ; ) is clearly analytic in the domain

D C \ (W+ W ). (A.28)

In our study of the generalized hypergeometric function it is now crucial that the
function

f (a+ , a , z) g(a+ , a ; z) A(a+ , a , z), (A.29)

where A is the dominant asymptotics function


 
z2 a+ a
A(a+ , a , z) + , (A.30)
2a+ a 24 a a+

decays exponentially for Re z , locally uniformly in a+ , a and Im z, with a rate


involving the positive number

am max(r+ / cos + , r / cos ). (A.31)

To detail this asymptotics, we introduce the domain

A {am + R exp(i) | R > 0, || < m }, (A.32)

where

m min(min + /2, max + /2). (A.33)

Equivalently, m may be defined by

cot m = max(| tan + |, | tan |), m (0, /2]. (A.34)

Figure 3 may be helpful to vizualize the geometry.

Theorem A.1. Fix compacts K+ , K RH P and K R, and fix (1/2, 1). Then
there exists a positive C = C(K+ , K , K, ) such that one has

|f (a+ , a , z)| < C exp(2 Re z/am ), (A.35)

for all a+ K+ , a K and z A with Im z K.

Proof. The proof of this theorem is relegated to Appendix C. u


t

In Appendix C additional information concerning asymptotics is obtained that is of


some interest in itself. As it stands, however, the uniform bound (A.35) suffices for our
present purposes (as will become clear in Appendix B).
We proceed by detailing how the gamma function arises as a limit of our G-function:
One has

(2 )1/2
lim exp[iz ln(2)]G(1, ; z + i/2) = , (A.36)
0 0(iz + 1/2)
668 S. N. M. Ruijsenaars

W+

ia

0 A
am

ia

Fig. 3. The asymptotics domain A and the zero and pole wedges W+ , W for a+ , a RH P

the limit being uniform on compact subsets of C. (Note that the scale invariance (A.6) of
the G-function can be used to handle the case a+ 6= 1.) The well-known multiplication
formula for the gamma function can be obtained from its generalization
N
Y ia+ ia
G(a+ , a ; Nz) = G(a+ , a ; z + (N + 1 2j ) + (N + 1 2k))
2N 2N
j,k=1
(A.37)
and the limit (A.36). (In fact, the G-function satisfies a multiplication formula that is
more general than (A.37), cf. Eq. (3.25) in Ref. [9].)
A second zero step size limit of the G-function is crucial, too: One has
G(, a; z + ia)
lim = exp(( ) ln(2chz)), , R, (A.38)
a0 G(, a; z + ia)
uniformly on compacts of the cut plane C(/2), with
C(d) C \ i[d, ), d > 0. (A.39)
(Here, ln is chosen real for z real.) Note that this limit is an easy consequence of the A1E
(A.9) for Z; The branch cut occurring for / Z is due to a coalescence of
zeros and poles.
Both limits (A.36) and (A.38) are crucial to appreciate why our function R(a+ , a , c;
v, v) may be viewed as a generalization of the hypergeometric function. We do not need
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 669

any further information on the G-function to study our R-function. But in order to
obtain a clear and detailed view on the analyticity properties of the R-function in its
eight variables, we have occasion to use in addition to G a function E(a+ , a ; z) that
will be introduced and studied next.
The E-function has in particular the following properties:
(i) E is holomorphic in (a+ , a , z) as long as a /a+ stays in the cut plane C (A.11);
(ii) E is related to G by

G(a+ , a ; z) = E(a+ , a ; z)/E(a+ , a ; z). (A.40)

From these properties it is already obvious that E is non-zero unless (a+ , a , z) belongs
+
to the hyperplanes z = zkl , cf. (A.14); moreover, for a+ , a fixed, the multiplicity of
the zeros of the entire function E(a+ , a ; ) equals that of the zeros of G(a+ , a ; ).
To define E we first introduce
Z  
1 dy 1 e2iyz 2iz z2
e(a+ , a ; z) (e2a+ y + e2a y ) .
4 0 y sha+ y sha y a+ a y a+ a
(A.41)

It is straightforward to verify that e is holomorphic in (a+ , a , z) for a+ , a RH P


and Im z < Re a. Clearly, we have (cf. (A.1))

e(a+ , a ; z) e(a+ , a ; z) = ig(a+ , a ; z). (A.42)

Therefore, the function

E(a+ , a ; z) exp[e(a+ , a ; z)] (A.43)

satisfies (A.40), and is holomorphic and non-zero for a+ , a RH P and Im z < Re a.


From (A.41) it is obvious that E is symmetric under a+ a and scale-invariant:

E(a , a+ ; z) = E(a+ , a ; z), (A.44)

E(a+ , a ; z) = E(a+ , a ; z), > 0. (A.45)

The announced property (i) is now an easy consequence of scale invariance and holo-
morphy of E for (a+ , a , z) RH P RH P C, which we demonstrate next.
Indeed, the latter holomorphy property is readily obtained from the A1E

E(z + ia+ /2) (2)1/2


= exp(izK(a+ , a )), (A.46)
E(z ia+ /2) 0( aiz + 21 )

where
 
1 a+
K(a+ , a ) ln , (A.47)
2a a

which is obeyed by E(a+ , a ; z) for Im (z + ia+ /2) < Re a. We prove this A1E in a
moment, and explain first why it entails holomorphy of E(a+ , a ; z) in RH P 2 C.
670 S. N. M. Ruijsenaars

The point is that we can iterate (A.46) to get

E(a+ , a ; z + iNa+ ) = E(a+ , a ; z)


N
Y (2)1/2 exp(i[z + i(j 1/2)a+ ]K(a+ , a ))
,
j =1
0( ai [z + i(j 1/2)a+ ] + 21 )
(A.48)

for a+ , a RH P and Im (z + iNa+ ) < Re a. Now the rhs can be analytically


continued to Im z < Re a. (Indeed, E(a+ , a ; z) has this property, and by entireness
of 1/ 0(z) the product has this property, too; note that (A.47) entails analyticity of
K(a+ , a ) in RH P 2 .) Therefore, we can continue E(a+ , a ; z) analytically to Im z <
Re (a + N a+ ), and since N is arbitrary, we deduce holomorphy in RH P 2 C.
Next, we prove that E(a+ , a ; z) satisfies (A.46). This A1E is readily obtained via
(A.41): It entails
Z  2itz/a 
a+ a+ dt e 1 iz 2t
e(z + i ) e(z i ) = + e
2 2 0 t 2sht 2t a
Z
iz dy 2a y
+ (e e2a+ y )
2a 0 y
= ln((2)1/2 / 0(iz/a +1/2)) + izK(a+ , a ). (A.49)

(Cf. e.g. Eq. (A37) in Ref. [9] for the 0-representation used here.) Thus the proof of the
property (i) of E(a+ , a ; z) announced above is now complete.
In the course of the proof we have obtained the A1E (A.46). This A1E plays no
further role in this paper, inasmuch as we only need the G-A1Es (A.7) and (A.9). We
add here some further remarks on it, however, and use it to explain the relation of E to
Barnes double gamma function.
Let us point out first that the A1Es (A.7) and (A.9) satisfied by the G-function may
be obtained directly from (A.40), (A.46) and (A.44), by using the functional equation
0(w + 1/2)(w w) = / cos w. (More generally, various properties of the G-
function can be obtained as corollaries of E-function features.) Just as the G-function can
be characterized as the unique minimal solution to the A1E (A.7) (with a+ , a positive)
satisfying (A.8), we may single out the E-function as the unique minimal solution to the
A1E (A.46) satisfying

E(0) = 1, E(z) > 0, z i(, a). (A.50)

(The ambiguity in minimal solutions is of the form c exp(2 kz/a+ ) with c C and
k Z, cf. Ref. [9]. The properties (A.50) fix c and k.)
We are now prepared to tie in the E-and G-functions with Barnes double gamma
function and Kurokawas double sine function. The point is that the double gamma
function may be viewed as a minimal solution to the A1E

E(z + ia+ /2) (2)1/2 iz


= exp( ln a ). (A.51)
E(z ia+ /2) 0( aiz + 2)
1 a
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 671

Specifically, the function

E(a+ , a ; z) 1/ 02 (iz + a, (a+ , a )) (A.52)


has this property.
The latter assertion is most easily established from Shintanis paper Ref. [16]: The
A1E can be gleaned from p. 172, and from p. 179 one can deduce that the third logarith-
mic derivative of E(a+ , a ; z) equals that of E(a+ , a ; z). From this equality it already
follows that (A.52) is a minimal solution to (A.51) with a+ , a positive, cf. Theorem II.3
in Ref. [9]. Moreover, comparing (A.46) and (A.51), we deduce that we must have

E(a+ , a ; z) = c(a+ , a ) exp(2kz/a+ + ln(a+ a )z2 /4a+ a )E(a+ , a ; z),


(A.53)

with k Z. Now both E and E are invariant under a+ a . (Indeed, this property of E
follows from (A.52) and invariance of 02 (w, (a+ , a )) under a+ a , cf. e.g. p. 173
in Ref. [16].) Hence we have k = 0. Using finally E(0) = 1, we obtain

02 ((a+ + a )/2, (a+ , a )) exp(ln(a+ a )z2 /4a+ a )


E(a+ , a ; z) = . (A.54)
02 (iz + (a+ + a )/2, (a+ , a ))
From (A.40) and (A.54) we now get
02 (a iz, (a+ , a ))
G(a+ , a ; z) = . (A.55)
02 (a + iz, (a+ , a ))
The rhs can be rewritten in terms of Kurokawas double sine function S2 (z|a+ , a ),
cf. Refs. [12,4]. This yields the identity
G(a+ , a ; z) = S2 (iz + (a+ + a )/2|a+ , a ), (A.56)
already alluded to.
It seems that Shintani was actually the first author who obtained a result on the double
sine function that is not immediate from its being a quotient of double gamma functions:
He proved a representation that in terms of our G-function amounts to
  
iz2 i a+ a
G(a+ , a ; z) = exp +
2a+ a 24 a a+
1 + exp( 2 [z i(l + 1 )a ])
Y a+ 2
, Im (a /a+ ) > 0, (A.57)
l=0
1 + exp( 2
a [z + i(l + 2 )a+ ])
1

cf. p.181 in Ref. [16]. We would like to point out that this formula can be quickly rederived
by exploiting the above theorem on the Re z asymptotics of g(a+ , a ; z).
Indeed, denoting the rhs of (A.57) by G(a+ , a ; z), we first point out (following
Shintani) that it satisfies the two A1Es (A.7) and (A.9). (This is easily checked.) There-
fore G/G is elliptic in z with periods ia+ , ia . Now an inspection of the zeros and poles
of G reveals that they are simple and coincide with the simple zeros and poles of the
G-function. Hence G/G is a constant, which a priori depends on a+ , a . However, one
need only invoke the g-asymptotics for z (recall Theorem A.1 and G = exp ig),
and verify that the infinite product on the rhs of (A.57) converges to 1 for a RH P
672 S. N. M. Ruijsenaars

and z (which is routine), to deduce that this constant equals 1. Therefore, we have
now proved (A.57).
Since we have studied the G-function from a quite different perspective in Ref. [9], we
can combine (A.57) with our previous results to obtain some remarkable consequences.
In particular, using the explicit value
G(a+ , a ; i(a+ a )/2) = (a+ /a )1/2 , (A.58)
(cf. Eq. (3.68) in Ref. [9]), we obtain an identity that amounts to the transformation
formula for the Dedekind -function. (The latter identity was used by Shintani as an
ingredient of his proof.) Similarly, using G(z)G(z) = 1 (cf. (A.4)), we deduce an-
other (previously known) modular transformation formula. (The connection between
the double gamma function and elliptic functions was already pointed out and studied
in detail by Barnes in his memoir Ref. [15].)
Another bonus of the representation (A.57) is that it gives rise to an improvement on
the bound (A.35) by means of which we obtained it. More precisely, (A.57) is easily seen
to entail that one can take = 1 in (A.35), provided a+ /a does not become positive
as (a+ , a ) varies over K+ K . (Note one may switch a+ and a on the rhs of (A.57)
because of (A.5).)
This sharper bound cannot be directly derived from our proof of Theorem A.1. Indeed,
we use the special case of equal a+ and a in a telescoping argument, and in this special
case (A.35) is false for = 1. (This becomes evident in the course of the proof, cf. the
paragraph in Appendix C containing (C.55).) In this connection it is important to observe
that the rhs of (A.57) has no meaning for Im (a /a+ ) = 0. In particular, it seems quite
unlikely that (A.57) leads to useful consequences for a+ , a (0, ), which is the
critical case for quantum-mechanical applications.

Appendix B. Analyticity Properties


In this appendix we study two complex-valued functions M and P of 2N + 2 complex
variables (a+ , a , u, d), with a varying over the (open) right half plane RH P , and u, d
over CN . These functions will be defined and shown to be jointly analytic on domains
DM and DP , with DM a proper subdomain of DP . (Throughout this appendix, domain
stands for open, connected set.) On DM the function P is given in terms of M and the
entire function E(a+ , a ; ) from Appendix A (cf. (A.43)), as follows:
P (a+ , a ; u, d) M(a+ , a ; u, d)
N
Y
E(a+ , a ; ia + uj dk ), a = (a+ + a )/2. (B.1)
j,k=1

To define M we need some preparations. The definition involves contour integrals


over the auxiliary variable z in the integrand
N
Y
I(a+ , a ; u, d, z) G(a+ , a ; z uj )/G(a+ , a ; z dj ), (B.2)
j =1

where G is the hyperbolic gamma function from Appendix A. Thus I(a+ , a ; u, d, )


is meromorphic, with u-dependent poles for
z = uj ia zkl , j = 1, . . . , N, k, l N, (B.3)
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 673

and d-dependent poles for

z = dj + ia + zkl , j = 1, . . . , N, k, l N, (B.4)

cf. (A.13), (A.14).


The u-poles (B.3) remain disjoint from the d-poles (B.4) as long as (a+ , a , u, d)
stays away from the set
jk
Z N
j,k=1 l,mN Hlm , (B.5)

where
jk
Hlm {(a+ , a , u, d) RH P 2 C2N |ila+ + ima uj + dk = 0}. (B.6)

Thus the union of hyperplanes Z equals the zero locus of the E-function product in
(B.1), cf. Appendix A.
A first restriction on the domain of M is now that Z is excluded from it. (As will be
seen, singularities arise when Z is approached.) But we need an additional restriction to
ensure that the integrand I has exponential decay for z . This requirement can
be readily studied by invoking Theorem A.1.
First, fixing (a+ , a , u, d) in RH P 2 C2N , we note that the set
   
AI N j =1 (u j + A(a+ , a )) N
j =1 (d j + A(a + , a )) (B.7)

contains an interval (L(a+ , a , u, d), ) R. (This is clear from the geometry of A,


cf. Fig. 3.) Then the bound (A.35) is easily seen to entail

N
X
iz
|I(a+ , a ; u, d, z)| < C+ | exp (uj dj ) |, z > L(a+ , a , u, d),
a+ a
j =1
(B.8)

and using (A.4) one similarly obtains



N
X
iz
|I(a+ , a ; u, d, z)| < C | exp (uj dj ) |, z < L(a+ , a , d, u).
a+ a
j =1
(B.9)

Here, the positive constants C can be chosen uniformly for a varying over compacts
in RH P and u1 , . . . , dN varying over compacts in C. Furthermore, these upper bounds
are best possible, as I satisfies lower bounds with the same characteristics.
To ensure exponential decay for z , then, we must require that (a+ , a , u, d)
belong to the domain
N
X
DP {(a+ , a , u, d) RH P 2 C2N |Im ( (uj dj )/a+ a ) > 0}. (B.10)
j =1
674 S. N. M. Ruijsenaars

The function P will be defined below on the simply-connected domain DP , whereas the
definition domain of M reads

DM DP \ Z, (B.11)

and hence is not simply-connected.


We begin by defining M on the domain

D0 {(a+ , a , u, d) DP |Im uj < Re a, Im dj > Re a, j = 1, . . . , N}. (B.12)

This is a subdomain of DM , since the imaginary part restrictions ensure that the u-poles
(B.3) lie in the lower half plane and the d-poles (B.4) in the upper one. Thus the u-poles
are separated from the d-poles by the contour

C0 (, ). (B.13)

Defining now
Z
M(a+ , a ; u, d) I(a+ , a ; u, d, z)dz, (a+ , a , u, d) D0 , (B.14)
C0

it is not hard to deduce from the above that M is jointly analytic in D0 . Indeed, the
uniform decay bounds (B.8) and (B.9), combined with the analyticity properties of the
G-functions in the integrand, entail that M is jointly continuous and separately analytic,
hence jointly analytic by virtue of Osgoods lemma (cf. Ref. [19]).
As will be made clear below, M has a one-valued jointly analytic extension to DM
(B.11), whereas the function P , defined at first by (B.1) on DM , has a (necessarily one-
valued) jointly analytic extension to DP (B.10). But we will improve considerably on
these existence assertions by making the extensions more or less explicit.
For the extension of P from DM to DP we will do so towards the end of the proof of
Theorem B.2 below. We proceed by defining M on DM . (It will not be obvious from the
definition that M is indeed analytic on DM ; this will also be demonstrated in the course
of the proof.) To this end we fix a RH P and suppress the dependence on a+ , a for
the moment.
To satisfy the decay restriction, we must require that (u, d) C2N obey
N
X
arg( (uj dj )) + (0, ), (B.15)
j =1

cf. (B.10) and (A.23). Now it is important to observe that arbitrary collisions of u-poles
and d-poles are compatible with the restriction (B.15). Indeed, one has

arg w + (0, ), w W \ {0}, (B.16)

so that the pole wedges uj + W , j = 1, . . . , N, and dk + W+ , k = 1, . . . , N, can


have arbitrarily large intersections without violating (B.15). (Recall (A.26), (A.27) and
Fig. 3.)
Fixing (a+ , a , u, d) DM , however, the u-poles are disjoint from the d-poles
by definition, even though the wedges may be arbitrarily entangled. Clearly, the pole
wedges uj + W intersect the closed upper half plane in K+ (u) triangles, with K+ (u)
{0, . . . , N}. (Here and below, triangle may reduce to point or line segment, the
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 675

latter case occurring for + = .) We now define a family CR+ , R 0, of contours, as


follows:

CR+ (, R) {z C||z| = R, Im z 0} (R, ), R 0. (B.17)

Obviously, whenever K+ (u) > 0, there exists a unique R+ (u) 0 such that at least one
u-pole lies on CR++ (u) , but all u-poles lie below CR+ for R+ > R+ (u).
Next, we modify the contours CR++ , R+ > R+ (u), depending on the (fixed) point
d CN . (For K+ (u) = 0 we modify the contour C0 (B.13) according to the following
prescription.) When no d-poles lie on the contour, we keep it unchanged. Whenever
d-poles lie on the contour, we indent it upwards so that the pertinent poles lie below
it. (Of course, the indentations should be sufficiently small so that they avoid all other
d-poles.)
Proceeding in this way, we obtain a well-defined family CR++ (d), R+ > R+ (u), of
contours. (For K+ (u) = 0 the modified contours are denoted C0+ (d).) Below each such
contour lies a certain number N+ (R+ , d) of d-poles. Denoting the residues of I(u, d, )
at these poles by D1 (u, d), . . . , DN+ (R+ ,d) (u, d), we now define

Z N+ (R+ ,d)
X
M(u, d) I(u, d, z)dz + 2 i Dj (u, d), (B.18)
CR++ (d) j =1

with R+ > R+ (u) for K+ (u) > 0 and R+ = 0 for K+ (u) = 0.


It is not hard to see that M is well defined. (That is, for K+ (u) > 0 the number on
the rhs does not depend on the choice of R+ (R+ (u), ).) Moreover, the definition
(B.18) reduces to (B.14) whenever (a+ , a , u, d) D0 . Indeed, in that case one has
K+ (u) = 0 and no d-poles lie on or below C0 , so no indentations and residues occur.
It is by no means clear, though, that M thus defined is analytic in DM , and hence
yields a one-valued analytic continuation of (B.14) to DM . Likewise, the existence of a
jointly analytic extension to DP for P (B.1) is not a matter of course. In the following
lemma we first consider the relevant analyticity properties of an arbitrary d-pole residue.
The results obtained are a crucial input for the study of M and P in Theorem B.2 below.

Lemma B.1. Fix a RH P , d CN , and

p d1 + ia + zk1 l1 , k1 , l1 N. (B.19)

Denote C with the points

p ia + zlm , l, m N , (B.20)

deleted by C(p). Letting u1 , . . . , uN C(p), the function I(a+ , a ; u, d, z) (B.2) has


a pole at z = p with a residue D(u, d) that is analytic for u C(p)N . Moreover, the
function
N
Y
5(u, d) D(u, d) E(ia + uj dk ) (B.21)
j,k=1

extends to an entire function of u1 , . . . , uN .


676 S. N. M. Ruijsenaars

Proof. Fixing u0 C(p)N , the factors G(z u0j ) in I(u, d) have no poles at z = p, so
we can find  > 0 such that the poles of these factors lie outside the circle |z p| = .
Then analyticity at u = u0 can be read off from the representation
I
1
D(u, d) = I(u, d, z)dz, (B.22)
2i |zp|=
which holds for all u such that the poles of the factors G(z uj ) lie outside |z p| = .
The proof of the last assertion involves more work. Let us first consider the (generic)
case where the factors 1/G(z dj ), j > 1, have neither a pole nor a zero for z = p.
Assuming a+ /a / Q, the pole of 1/G(z d1 ) is simple and gives rise to an I-residue
Y
rk1 l1 G(p u1 ) G(p uj )G(p + dj ), (B.23)
j >1

cf. the paragraph below (A.22). Moreover, the function G(p uj ) has simple poles at
uj = p + ia + zkl , k, l N, that are matched by simple zeros of E(ia + uj d1 ),
cf. (B.19). Therefore, entireness of 5(u, d) (B.21) is clear.
For a+ /a Q, however, the pole at z = p need not be simple, and some of the
poles of G(p uj ) are not simple. Assuming first O(k1 l1 ) = 1, so that z = p is still
a simple pole (cf. Appendix A), we obtain once again the residue (B.23). But now the
factor G(puj ) has a pole of order O(kl) for uj = p+ia +zkl . Since E(ia +uj d1 )
has a zero of order O(k + k1 , l + l1 ) for uj = p + ia + zkl , we can invoke the order
inequality (A.15) to deduce that no poles in uj occur. Hence 5(u, d) is once more entire.
Consider next the case O(k1 l1 ) > 1. Then the factor 1/G(z d1 ) has a Laurent
expansion
OX
(k1 l1 )
G(z + d1 ) = c(i, k1 , l1 )(z p)i + O(1), z p, (B.24)
i=1

and we should develop the remaining factors in a power series up to order (z p)O(k1 l1 )
to obtain the residue D(u, d). Explicitly, this yields
OX
(k1 l1 ) PN
X () l=2 jl
c(i, k1 , l1 )
i1 ! iN !j2 ! jN !
i=1 i ,... ,iN ,j2 ,... ,jN
PN1 PN
l=1 il + l=2 jl =i1
N
Y
G(i1 ) (p u1 ) G(il ) (p ul )G(jl ) (p + dl ). (B.25)
l=2

Now each differentiation increases all pole orders by 1. The pole order of the residue at
uj = p + ia + zkl equals, therefore, O(kl) + O(k1 l1 ) 1. Since E(ia + uj d1 ) has
zero order O(k + k1 , l + l1 ) at uj = p + ia + zkl , the inequality (A.15) entails once again
entireness in uj . Thus we have now proved the last assertion for all d2 , . . . , dN C
such that the factors G(z + dj ), j > 1, have neither a pole nor a zero at z = p.
We are now prepared to consider the most general situation: There may be factors
G(z + dj ), j > 1, that have a pole or a zero at z = p. Assume first that no further
poles occur, whereas at least one of these factors has a zero at z = p. Then the above
reasoning and residue formulas are still valid, the only difference being that the pertinent
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 677

factors vanish. In particular, when O(k1 l1 ) = 1, the residue (B.23) vanishes identically,
whereas (B.25) vanishes when O(k1 l1 ) 1 is smaller than the total zero order coming
from the factors G(z +dj ) for j > 1. (Of course, in these cases I has no pole at z = p,
strictly speaking.) In any case, entireness of 5(u, d) follows, since the zero orders of
the E-functions involving d1 already suffice to cancel any poles.
Next, we assume that additional poles occur for z = p. Eventually relabeling, we
may as well assume that we have
d1 + zk1 l1 = dj + zkj lj , j = 2, . . . , M N, kj , lj N, (B.26)
whereas the remaining factors G(z + dj ), j > M, are regular at z = p. (Again,
whether or not the latter are zero at z = p plays no role for the reasoning and formulas
that follow. The only significant consequence is that when the total zero order is the
total pole order, the residue vanishes and I has no pole at z = p.) When we now develop
the singular factors in a Laurent expansion at z = p as in (B.24), we readily obtain
M
Y O
X (p)
G(z + dj ) = c(i, k1 , l1 , . . . , kM , lM )(z p)i + O(1), z p, (B.27)
j =1 i=1

where
M
X
O(p) O(kj lj ), (B.28)
j =1

and where the coefficients are determined in terms of the coefficients c(i, kj , lj ) with
i = 1, . . . , O(kj lj ).
To obtain the residue D(u, d), therefore, the remaining factors should be expanded
up to order (z p)O(p) . This yields
O (p) PN
X X () l=M+1 jl
c(i, k1 , . . . , lM )
i1 ! iN !jM+1 ! jN !
i=1 i ,... ,iN ,jM+1 ,... ,jN
PN1 PN
l=1 il + l=M+1 jl =i1
M
Y N
Y
G(il ) (p ul ) G(il ) (p ul )G(jl ) (p + dl ). (B.29)
l=1 l=M+1

Thus we see that also in the most general case the residue D(u, d) is a sum of products of
meromorphic functions depending on only one of the variables u1 , . . . , uN . Clearly, the
pole order in the variable uj at uj = p+ia+zkl is less than or equal to O(kl)+O(p)1.
Now using (B.26) we may write
p + ia + zkl = 2ia + dm + zk+km ,l+lm , m = 1, . . . , M. (B.30)
Therefore, the function E(ia + uj dm ) has zero order O(k + km , l + lm ) at uj =
p + ia + zkl for m = 1, . . . , M. For the total zero order we then have, using (A.15),
M
X M
X
O(k + km , l + lm ) (O(kl) + O(km lm ) 1)
m=1 m=1
M[O(kl) 1] + O(p). (B.31)
678 S. N. M. Ruijsenaars

Thus this order is larger than or equal to the maximal pole order O(kl) 1 + O(p) of
the residue (B.29). Hence the function 5(u, d) (B.21) extends to an entire function of
u in the most general case, too. u t

Theorem B.2. The function M defined by (B.18) is jointly analytic in DM (B.11). The
function P defined by (B.1) for (a+ , a , u, d) DM has a jointly analytic extension to
DP (B.10). The functions F = M, P have the following automorphy properties:

F (a+ , a ; 1 (u), 2 (d)) = F (a+ , a ; u, d), 1 , 2 SN , (B.32)


F (a+ , a ; d, u) = F (a+ , a ; u, d), (B.33)
F (a , a+ ; u, d) = F (a+ , a ; u, d), (B.34)
F (a+ , a ; u, d) = F (a+ , a ; u, d), (0, ), (B.35)
F (a+ , a ; u1 + r, . . . , dN + r) = F (a+ , a ; u, d), r R. (B.36)

Proof. Clearly, the variable transformations in (B.32)(B.36) leave the domains DP ,


DM and D0 invariant. For F = M and variables in D0 , the automorphy properties are
clear from (B.14), (B.2) and the G-features (A.4)(A.6). For F = P and variables in
D0 they then follow from (B.1) and the E-features (A.44), (A.45).
Now the automorphy relations are preserved under analytic continuation. (In fact,
one may take C and r C.) Therefore it remains to prove the first two assertions.
To this end we begin by claiming that we may just as well define M by
Z N (R ,u)
X
M(u, d) I(u, d, z)dz 2 i Uj (u, d), (B.37)
CR (u) j =1

instead of (B.18). Here, the notation will be largely clear from context: The pole wedges
dj + W+ intersect the closed lower half plane in K (d) triangles, the contour CR (u)
is defined by taking contours CR in the closed lower half plane as starting point (i.e.,
with Im z 0 on the rhs of (B.17)), one should take R > R (d) in (B.37) so that all
d-poles lie above CR , indentations to avoid u-poles are downwards, and the functions
U1 , . . . , UN (R ,u) denote the residues of I at the u-poles lying above the contour.
To see that the right-hand-sides of (B.18) and (B.37) indeed yield the same number,
one need only deform the contour CR++ (d) to CR (u) in the obvious way. Then all of the
N+ (R+ , u) d-poles in (B.18) and all of the N (R , u) u-poles in (B.37) are passed,
whilst no other d- and u-poles are encountered. Thus the residues Dj in (B.18) are
canceled and the residues Uj in (B.37) appear.
With the alternative definition (B.37) of M at our disposal, we are now going to prove
that M is analytic in each of the variables a+ , a , u1 , . . . , dN . Joint analyticity is then
a consequence of Hartogs theorem (cf. e.g. Theorem 2.2.8 in Ref. [20]).
Our task is, then, to demonstrate that for each of the 2N + 2 variables there exists a
disc around an arbitrary fixed point in DM so that M is analytic in this disc. To avoid
notational confusion, we denote the fixed point by (a+ 0 , a 0 , u0 , d 0 ).

Beginning with u1 , we start from (B.18) with u, d u0 , d 0 and parameters a+ 0 , a0

understood. Now we choose 1 > 0 sufficiently small so that the following properties
hold true for u1 varying over |u1 u01 | 1 :
(i) The points (a+0 , a 0 , u , u0 , . . . , d 0 ) stay in D ;
1 2 N M
(ii) The z-poles u1 ia 0 zkl 0 stay below C + (d 0 ) for all k, l N.
R+
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 679

(Since DM is an open set, (i) can be achieved by letting u1 vary over |u1 u01 |  with
 > 0 small enough. Then only finitely many points u1 ia 0 zkl 0 can cross the contour
+
CR+ (d 0 ) as u1 varies, so an eventual decrease of  ensures that none of them does. Now
take 1 = /2.)
A moments thought suffices to see that M is indeed analytic in u1 as u1 varies over
the disc |u1 u01 | < 1 : The d-poles and contour occurring in (B.18) are constant on
the disc, the contour integral is analytic in u1 due to (i), (ii) and uniform convergence
for z , and each of the residues Dj (u1 , u02 , . . . , u0N ) is analytic in u1 by virtue
of (i). (Indeed, since (a+ 0 , a 0 , u , u0 , . . . , d 0 ) stays away from the hyperplanes H 1k
1 2 N lm
(B.6), we may invoke Lemma B.1.)
Of course, this reasoning can be repeated for u2 , . . . , uN . To prove separate analytic-
ity in d1 , . . . , dN , we take (B.37) as a starting point, and adapt the argument accordingly.
(In particular, Lemma B.1 has an obvious analog for an arbitrary residue U (u, d) at a
u-pole.)
We continue by proving analyticity in a+ , taking (B.18) with u, d u0 , d 0 as a
starting point. Making the dependence on a+ , a explicit, it can be rewritten
Z
M(a+ , a ; u , d ) =
0 0 0 0
I(a+ 0
, a0
; u0 , d 0 , z)dz
CR++ (a+
0 ,a 0 ;d 0 )

N+ (a+ I
0 ,a 0 ;R ,d 0 )
X +

+ I(a+
0
, a
0
; u0 , d 0 , z)dz. (B.38)
j =1 0j (a+
0 ,a 0 ;d 0 )

Here, the contour 0j is a circle around the pertinent d-pole, whose radius j is chosen
sufficiently small so that all other d-poles and all of the u-poles lie outside 0j .
Now we choose + > 0 small enough so that when a+ varies over |a+ a+ 0| ,
+
the following properties hold true:
(i) The points (a+ , a 0 , u0 , d 0 ) stay in D ;
M
(ii) None of the u-poles and d-poles cross the contours CR++ (a+
0 , a 0 ; d 0 ) and 0 (a 0 , a 0 ;
j +
d 0 ), with j = 1, . . . , N+ (a+ 0 , a 0 ; R , d 0 ).
+

(As before, openness of DM ensures one can find  > 0 such that for |a+ a+ 0 |  all
pertinent points stay in DM . Only finitely many points can then cross the finitely many
fixed contours, so by decreasing  one may ensure none does. Now take + = /2.)
When we now replace I(a+ 0 , a 0 ; u0 , d 0 , z) by I(a , a 0 ; u0 , d 0 , z) on the rhs of
+
(B.38), then the integrals are well-defined for a+ varying over the disc |a+ a+ 0|< ,
+
and yield analytic functions in this disc. Thus the rhs yields a function R(a+ 0 , a 0 , u0 , d 0 ;

a+ ) that is defined and analytic in a+ for |a+ a+ 0|< .
+
We claim that the latter function coincides with M(a+ , a 0 ; u0 , d 0 ) as already defined

via (B.18) (with parameters a+ , a understood). Indeed, this is plain for a+ = a+


0 0.

Assuming next a+ 6 = a+ 0 , the asserted equality is easily verified whenever all of the

N+ (a+ , a ; R+ , d ) d-poles entering (B.38) are simple.


0 0 0
As we have already seen in the proof of Lemma B.1, the poles need not be simple,
however. First, there may be coinciding points

dj0i + zk0i li , i = 1, . . . , M, M {2, . . . , N}, 1 j1 < < jM N. (B.39)


680 S. N. M. Ruijsenaars

This can happen for arbitrary a 0 RH P , of course. But when a 0 /a 0 Q, additional


+
multiplicity may arise from zj k being equal to zlm
0 0 for j 6 = l and (hence) k 6 = m. In the

latter case one obtains O(j k) distinct points whenever a+ 6 = a+ 0 . Similarly, a pair of

distinct ki1 , ki2 in (B.39) gives rise to distinct points for a+ 6 = a+ . 0

Whenever one of these cases occurs, R(a+ 0 , a 0 , u0 , d 0 ; a ) is no longer mani-


+
festly equal to M(a+ , a 0 ; u0 , d 0 ) as defined by (B.18), as N (a , a 0 ; R , d 0 ) >
+ + +
N+ (a+ 0 , a 0 ; R , d 0 ). But we may replace all of the circle integrals entering the def-
+
inition of R by 2i times the sum of the residues of the d-poles they enclose, and when
we do, we obtain the previous definition (B.18) of M, up to relabeling.
The upshot is that we have proved analyticity in a+ . The same arguments yield
analyticity in a , so we deduce joint analyticity of M in DM , as announced.
It now follows from (B.1) and joint analyticity of E(a+ , a ; z) in RH P 2 C (cf. Ap-
pendix A) that P is also jointly analytic in DM . It remains to prove that P has a jointly
analytic extension to DP . Just as for M, we do so by exhibiting an extension to DP , and
then proving that the extended function is in fact analytic in each of its variables, hence
jointly analytic by Hartogs theorem.
In order to detail the extension, we fix a point (a+ 0 , a 0 , u0 , d 0 ) D Z. Then we
P
can partition the index set {1, . . . , N} into two subsets I= and I6= , depending on whether
u0j equals one of the points

dk0 + zlm
0
, k = 1, . . . , N, l, m N , (B.40)

or not. Since the fixed point belongs to Z (B.5), we have |I= | 1.


Next, for j I= we let uj vary over a punctured disc 0 < |uj u0j | j , whereas
we keep uj = u0j for j I6= . Choosing the j s small enough, we can ensure that the
points (a+0 , a 0 , u, d 0 ) stay in D .
M
Now for the subset of DM thus determined, we may invoke (B.18) with d d 0 ,
provided we choose R+ larger than all of the R+ (u) involved. Doing so, it follows from
previous arguments that the contour integral yields an analytic function of uj in the disc
|uj u0j | < j for all j I= . Multiplying the integral by the E-product (cf. (B.1)), we
therefore obtain an analytic function of uj that vanishes for u = u0 .
From Lemma B.1 we now deduce that the limit

P (u0 , d 0 ) lim P (u, d 0 ) (B.41)


uu0

exists and equals

N+ (R+ ,d 0 ) N
X Y
2i lim Dj (u, d 0 ) E(ia 0 + uj dk0 ), (B.42)
uu0
j =1 j,k=1

with Dj given by the general formula (B.29) up to relabeling and substitutions. Moreover,
Lemma B.1 entails that the function P on DP thus obtained is analytic in u1 , . . . , uN .
The first assertion of the theorem has already been proved, so we may invoke the
automorphy property P (u, d) = P (d, u) for arguments in DM . It entails that the
extended function P on DP is also analytic in d1 , . . . , dN . We continue by proving that
it is analytic in a+ as well.
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 681

To this end we reconsider the local behavior near the above point (a+
0 , a 0 , u0 , d 0 ) in

DP Z. We let a+ vary over a disc
D+ {|a+ a+
0
| + }, + > 0, (B.43)
with + small enough so that:
(1) The points (a+ , a0 , u, d 0 ) stay in D as a varies over D and u over |u u0 |
P + + j j j
j /2, j I= ;
(2) The points (a+ , a0 , u, d 0 ) stay in D as a varies over D and u over |u u0 | =
M + + j j j
j /4, j I= .
(Since DP is open, one can achieve (1) by letting a+ vary over D+ (B.43) with +
sufficiently small. Now (a+ 0 , a 0 , u, d 0 ) stays away from Z as u varies over |u u0 | =
j j j
j /4, j I= , so an eventual shrinking of + ensures (2).)
We are now prepared to examine the function

  I
1 |I= | Y duj
P (a+ , a
0
; u, d 0 ), (B.44)
2 i |u j u0 |= /4 (u u0 )
j j j
j I= j

with, as before, uj = u0j for j I6= . Due to (2), it is analytic in a+ for |a+ a+
0|< .
+
Invoking (1), the analyticity in uj already proved, and Cauchys formula, one sees that
it is also equal to P (a+ , a
0 ; u0 , d 0 ). Hence the announced analyticity in a follows.
+
Analyticity in a can be proved analogously. u t

Appendix C. Proof of Theorem A.1


The geometry of the asymptotics domain A (A.32) plays a crucial role in the following
proof. In particular, using from now on the notation
= Re z, = Im z, (C.1)
we note that we have
z A, > 0 z ita A, t [0, /r cos ], = +, , (C.2)
whereas z ita does not generally belong to A for t larger than specified. Moreover,
we have
z A, 0 > am + | tan |, = +, . (C.3)
(These assertions easily follow from (A.32)(A.34), cf. also Fig. 3.)
Next, we observe that for z in the half strip
1 4
H S { > am , || < r cos }, (C.4)
2 5
the function g(a+ , a ; z) is given by the integral (A.1). We are going to reduce the
asymptotics for arbitrary ||-values to the asymptotics in H S by exploiting the A1Es
satisfied by g. To be specific, the latter entail the relations
f (a+ , a , z) = f (a+ , a , z + ia ) + i ln(1 + exp[2(z + ia /2)/a ]),
(C.5)
682 S. N. M. Ruijsenaars

with , = +, , whenever the points z, z + ia belong to D (A.28). (Indeed, for


positive a+ , a this follows from Eqs. (3.1)(3.4) in Ref. [9], and these A1Es can be
analytically continued.)
Turning now to the details, let us choose a point (a+ , a , ) in K+ K K and
fix {+, } such that

r cos = min(r+ cos + , r cos ). (C.6)

(Here and below, we use tildes whenever the quantities (A.23)(A.25) refer to the fixed
points a+ , a .) Next, we define discs

D (d) {a C | |a a | d r cos }, d (0, 1/2), = +, , (C.7)

(so that D (d) RH P ), and an interval

I (b) { R | | | br cos }, b (0, 1/2), (C.8)

where d and b will be further restricted shortly. By virtue of compactness of K and K,


we need only show there exists a positive C (depending on D (d), I (b) and ) such
that the bound (A.35) holds for all a D (d) and z A with Im z I (b).
We proceed by doing so, assuming 0 from now on. (The proof for the case
0 involves a few obvious changes.) This assumption entails that there exists a
unique N N such that

1 2
N r cos r cos ( , ]. (C.9)
3 3
With and N fixed by (C.6) and (C.9), resp., we are now going to choose d and b such
that for all a D (d) and all z A with I (b) we have

z ila A, l = 1/2, . . . , N 1/2, (N > 0) (C.10)

z iNa H S. (C.11)

It is not obvious that such a choice can be made. In order to prove its feasibility,
we begin by observing that (C.10) and (C.11) are implied by the (more convenient)
requirements

1
Nr cos > r cos , (C.12)
2
3
Nr cos < r cos , (C.13)
4
3 4
r cos < r cos . (C.14)
4 5
Indeed, for N = 0 (C.10) is vacuous, while for N > 0 (C.10) follows from (C.12),
cf. the first paragraph of this appendix. Considering next (C.11), we deduce from (C.4)
that it amounts to
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 683

1
+ Nr sin > am , (C.15)
2

4
| Nr cos | < r cos . (C.16)
5
Now (C.16) is clear from (C.12)(C.14), so we are left with (C.15). Since + i A
by assumption, we have > am , so (C.15) is plain for N = 0.
Finally, to prove that (C.15) is implied by (C.12)(C.14) for N > 0, we first note
that (C.9) and (C.8) entail > 0 on I (b). Thus we can use (C.3) to infer that (C.15) is
implied by
1
am + | tan | > am + N r | sin |. (C.17)
2
Now (C.17) is evident for = 0, whereas for 6= 0 it can be rewritten
1
Nr cos > am cos /| sin |. (C.18)
2
Using (C.12), we deduce that we need only verify

r cos > am cos /| sin |, (C.19)

which amounts to

| sin 2 | < 2am cos /r . (C.20)

On account of the definition (A.31) of am , the rhs is 2, and so (C.20) holds true.
Summarizing, we have shown that (C.10) and (C.11) are implied by (C.12)(C.14).
We continue by restricting d and b so that (C.12)(C.14) hold for all a D (d) and
I (b). First, from (C.6) we deduce

r cos r cos , (C.21)

and (C.7) yields

r cos r cos [1 d, 1 + d]. (C.22)

Hence we have
 
1+d
r cos r cos , (C.23)
1d
so that (C.14) can be satisfied by requiring
 
3 1+d 4
< . (C.24)
4 1d 5
Next, we use (C.9), (C.7) and (C.8) to obtain

N r cos N r cos | | N|r cos r cos |


1 (b + N d + 1/3)
> r cos ( b Nd) r cos , (C.25)
3 (1 d)
684 S. N. M. Ruijsenaars

and, likewise,
(b + Nd + 2/3)
Nr cos r cos . (C.26)
(1 d)
Thus we can satisfy (C.12) and (C.13) by requiring
(b + Nd + 1/3)/(1 d) < 1/2, (C.27)
(b + Nd + 2/3)/(1 d) < 3/4. (C.28)
Clearly, we can choose d and b such that the restrictions (C.24), (C.27) and (C.28)
are satisfied. To be specific, let us take
d = 1/36(N + 1), b = 1/18. (C.29)
Then the latter restrictions are easily checked, so we are now in the position to exploit
(C.10) and (C.11).
Indeed, consider the A1E (C.5) satisfied by f (a+ , a , z), choosing = , a
D (d) and z A with I (b), and fixing via (C.6). When we now iterate (C.5) N
times, we obtain
N
X 1
f (z) = f (z iNa ) i ln(1 + exp[2(z i(j )a )/a ]). (C.30)
2
j =1

The crux is that this reduces the argument of f to H S (due to (C.11)), whereas the N
ln-terms are of the form ln(1 + exp[2w/a ]) with w A (due to (C.10)).
To bound the latter function, we use (A.32) to write
Re (w/a ) = am cos /r + R cos( )/r . (C.31)
The second term on the rhs is positive, so we deduce
Re (w/a ) > 1, w A. (C.32)
Clearly, this entails
| exp(2w/a )| < exp(2 ), w A, (C.33)
so we can use the elementary estimate
| ln(1 + x)| < c1 |x|, |x| < exp(2 ), (C.34)
to obtain the desired bound
| ln(1 + exp(2w/a ))| < c1 exp(2Re (w/a ))
< c1 exp(2[|Im w|| sin |/r Re w/am ]). (C.35)
For each of the N logarithmic terms in (C.30) we can now use (C.35) to obtain
a bound of the form Cj (a+ , a , ) exp(2/am ), with Cj continuous on D+ (d)
D (d) I (b), and hence bounded. Thus we obtain
|f (z) f (z iNa )| C(d, b) exp(2 /am ), (C.36)
for all a D (d) and z A with I (b).
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 685

A moments thought now shows that the proof of Theorem A.1 is complete once we
succeed in showing

|f (a+ , a , z)| < C(a+ , a , ) exp(2 /am ), z H S, (C.37)

with C(a+ , a , ) continuous for a+ , a RH P . Indeed, the argument z iN a in


(C.36) remains in H S as (a+ , a ) varies over D+ (d) D (d) and the imaginary part
of z A varies over I (b), cf. (C.11). Therefore, (C.37) entails that |f (z iN a )| is ma-
jorized by exp(2 Re z/am ) times a function CN (a+ , a , ) that is also continuous
for a+ , a RH P , hence bounded on D+ (d) D (d).
Our main tools for obtaining (C.37) are a comparison argument and a suitable use
of the integral representation (A.1). The comparison argument exploits the fact that the
function g(a, a; z) admits a second integral representation from which its asymptotics
can be easily established, cf. Eqs. (3.41)(3.45) in Ref. [9]. (Here, however, the com-
parison parameter (a+ + a )/2 differs from loc. cit. Eq. (3.52).) Turning to the details,
in the present case the cited formulas entail the identity

a2
g(a, a; z) = A(a+ , a , z) (a+ , a ) + D1 (a+ , a , z). (C.38)
a+ a
Here, A is given by (A.30), and and D1 are defined by
 
a+ a
(a+ , a ) 2 , (C.39)
48 a a+

Z
a2 tet
D1 (a+ , a , z) dt , a = (a+ + a )/2. (C.40)
a+ a z/a sht

Choosing at first z R, we now write


Z
a2 1 dy
g(a+ , a ; z) = g(a, a; z) + I (y) sin 2yz, (C.41)
a+ a 2 0 y
where
1 a2
I (y) . (C.42)
sha+ y sha y a+ a sh2 ay
Clearly, we have

I (y) = 4/ + O(y 2 ), y 0. (C.43)

Thus we can use the sine transform


Z  
ay sin 2yz z
dy = cth(z/a) , Re a > 0, z R, (C.44)
0 sh2 ay 2a ash2 ( z/a)
(cf. e.g. Eq. (2.66) in Ref. [9]), to obtain
Z
1 dy
I (y) sin 2yz = D2 (a+ , a , z) + (a+ , a ) + D3 (a+ , a , z), (C.45)
2 0 y
686 S. N. M. Ruijsenaars

with
D2 (z) [cth(z/a) 1 ( z/a)/sh2 ( z/a)], (C.46)
Z
D3 (z) dyJ (y)e2iyz , (C.47)

 
1 4a 2 y 2
J (y) I (y) . (C.48)
4iy sh2 ay
The point is now that J (y) is O(y) for y 0 and O(y exp(2r|y| cos )) for
y . Therefore, we may take |Im z| < r cos in the integral (C.47). Combining
the above formulas, we obtain
X
3
f (a+ , a , z) = Dj (a+ , a , z), (C.49)
j =1

so we need only show that D1 , D2 and D3 satisfy a bound of the form (C.37), with C
continuous.
In order to prove this, we begin by pointing out that one has
r/ cos am . (C.50)
Indeed, one readily verifies the stronger inequality
r/ cos (r+ / cos + + r / cos )/2, (C.51)
where equality holds iff + = .
Using (C.50), we now obtain
Re (z/a) = (Re z cos + Im z sin )/r
am cos 4
> cos | sin |
2r 5
1 2 1
> = , z H S. (C.52)
2 5 10
Hence we have a 1 H S RH P . Also, setting
w z/a, (C.53)
we deduce
|(1 exp(2w))1 | < (1 exp(/5))1 c2 , w a 1 H S. (C.54)
Using this numerical majorization, the desired bound on D2 (z) (C.46) is readily obtained.
To handle D1 , we need only study
Z
h(w) 2 te2t /(1 e2t )dt, w a 1 H S. (C.55)
w

Using the elementary integral


Z
yey dy = (R + 1)eR , (C.56)
R
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 687

we first rewrite h(w) as


Z
h(w) = 2 te4t /(1 e2t )dt + we2w + e2w /2. (C.57)
w

To bound the integral, we put

w = sei , s |w|, (C.58)

and change the contour to the ray xei , x s. (The arc from to ei yields zero
contribution.) Then we obtain the integral
Z
ei x exp(4xei )(1 exp(2xei ))1 dx. (C.59)
s

Using (C.54), we infer that its modulus is majorized by


Z
c2
c2 x exp(4x cos )dx = 2
(4s cos + 1) exp(4s cos ), (C.60)
s 16 cos
cf. (C.56). Now we have

s cos = Re w = Re (z/a) = r 1 ( cos + sin ). (C.61)

Therefore, the first term in (C.57) gives rise to a term in D1 (a+ , a , z) that is bounded
by the rhs of (C.37) for all < 2. Obviously, this holds true for the third term in (C.57)
as well, provided we take 1. For the second one, however, we must take < 1 so
as to counterbalance the factor linear in w.
It remains to prove (C.37) with f replaced by D3 . To this end we make the key
observation that
Z
4iD3 (z) = exp(2dz) duM(z, d, u), (C.62)

with
e2iuz 1
M(z, d, u) (
u + id sha+ (u + id)sha (u + id)
a2 1
[1 (a+ a )2 (u + id)2 ]), (C.63)
a+ a sh a(u + id)
2 12
for all z H S and d [0, /am ). Indeed, (C.62) is evident for d = 0. To check
independence of d, we should first of all verify that the integrand M is non-singular for
u R and d (0, /am ), and we proceed by doing so.
First, the factor sha(u + id) vanishes iff u + id belongs to i a 1 Z, which amounts
to

u = k sin /r, d = k cos /r, k Z. (C.64)

Since we have d (0, /am ) and /am cos /r (recall (C.50)), it follows that
sha(u+id) is non-zero. Likewise, since 1/am cos /r , we deduce sha (u+ir) 6 =
0. Therefore, the function (C.63) has no poles on the contour in (C.62).
688 S. N. M. Ruijsenaars

Next, we note
M(z, d, u) = O(u exp[2(|| r cos )|u|]), u . (C.65)
Now for z H S we have 5|| 4r cos , so the integral (C.62) is well defined, and we
may shift contours to deduce independence of d [0, /am ).
Fixing (1/2, 1), we now let
d = /am . (C.66)
Then we obtain the bound (with + i H S)
Z exp( 85 r|u| cos )
|D3 (a+ , a , + i)| < exp(2 /am ) du
(u2 + d 2 )1/2
(|sha+ (u + id)sha (u + id)|1
|a 2 |
+
|a+ a sh2 a(u + id)|
1
(1 + |a+ a |2 (u2 + d 2 ))). (C.67)
12
Next, we split up the integration region R in three subsets: Letting
1
u0 + max(| tan + |, | tan |), (C.68)
2 am
we consider successively u > u0 , u < u0 and u [u0 , u0 ].
For u > u0 we have
|Re (a(u + id))| = r|u cos d sin | r(u0 cos d| sin |)
1
r( cos + | sin | | sin |)
2 am am
1
r cos . (C.69)
2
Likewise, we obtain
1
|Re (a (u + id))| r cos , u > u0 . (C.70)
2
Therefore, the integral over {u > u0 } can be majorized by
Z
du 8
4 exp( ru cos )| exp(2a(u + id))|
1/2 u 5
Y |a 2 |
( [1 exp(r cos )]1 + [1 exp(r cos )]2
|a+ a |
=+,
1 2
(1 + |a+ a |2 (u2 + 2 ))). (C.71)
12 am
Now we have
| exp(2a(u + id))| = exp(2ru cos + 2r sin /am )
< exp(2ru cos ) exp(| sin 2|). (C.72)
Generalized Hypergeometric Function Satisfying Equations of AskeyWilson Type 689

Thus, (C.71) is bounded above by


Z du 2
exp( ru cos )(C1 (a+ , a ) + C2 (a+ , a )u2 ), (C.73)
1/2 u 5

with Cj (a+ , a ) continuous and positive. Finally, (C.73) can be bounded by


Z 2
du exp( ru cos )(2C1 (a+ , a ) + uC2 (a+ , a ))
0 5
= 5C1 (a+ , a )/(r cos ) + 25C2 (a+ , a )/(4r 2 cos2 )
C> (a+ , a ), (C.74)

with C> (a+ , a ) continuous and positive.


Likewise, the integral over {u < u0 } in (C.67) can be majorized by a continuous,
positive function C< (a+ , a ). Thus we are left with the integral over [u0 , u0 ]. On this
interval we bound the u-dependent terms in (C.67) by their maxima. Then the integral
is bounded by

8
2u0 d 1 exp( ru0 cos )(max |sha+ (u + id)sha (u + id)|1
5
|a 2 | 1
+ max |sha(u + id)|2 (1 + |a+ a |2 (u20 + d 2 ))). (C.75)
|a+ a | 12

Now we have already seen that the sh-factors stay away from 0 for d am 1 (1/2, 1)
and u R. But as 1 in (C.66), the distance to at least one of the zeros goes to 0
(cf. the paragraph containing (C.64)). Therefore, (C.75) yields a function C[ ] (a+ , a , )
which is continuous in a+ , a for (1/2, 1), but which diverges as 1.
The upshot is that we obtain the desired bound

|D3 (a+ , a , z)| < C3 (a+ , a , ) exp(2 /am ), z H S, (C.76)

where

C3 C> + C< + C[ ] (C.77)

is continuous in a+ , a . Therefore, the estimate (C.37) on f (a+ , a , z) follows, which


completes the proof of Theorem A.1.

Acknowledgements. Some of the results of this paper were first reported in our Banff Summer School lecture
notes (1994). We would like to thank the organizers G. Semenoff and L. Vinet for their invitation. We also
outlined the main results in a lecture series at the Workshop on Invariant Differential Operators, Special
Functions and Representation Theory (R.I.M.S., Kyoto, 1997). We are indebted to the Organizing Committee,
in particular to M. Kashiwara and T. Oshima, for inviting us, and to the R.I.M.S. for its hospitality and financial
support. This paper was completed during our stay at the Universidad de Chile in Santiago. We gratefully
acknowledge the financial support of Fundacin Andes, and the fine hospitality and support provided by the
Departamento de Matemticas de la Facultad de Ciencias.We also thank J.F. van Diejen for his invitation and
for useful discussions.
690 S. N. M. Ruijsenaars

References
1. Gasper, G., Rahman, M.: Basic hypergeometric series. Encyclopedia of Mathematics and its Applications,
35, Cambridge: Cambridge Univ. Press, 1990
2. Askey, R., Wilson, J.: Some basic hypergeometric orthogonal polynomials that generalize Jacobi poly-
nomials. Mem. Am. Math. Soc. 319, (1985)
3. Grnbaum, F.A., Haine, L.: Some functions that generalize the AskeyWilson polynomials. Commun.
Math. Phys. 184, 173202 (1997)
4. Nishizawa, M., Ueno, K.: Integral solutions of q-difference equations of the hypergeometric type with
|q| = 1. To appear in Proceedings of the 1996 Workshop Infinite analysis, IIAS, Japan
5. Ruijsenaars, S.N.M.: Systems of CalogeroMoser type. In: Proceedings of the 1994 Banff summer
school Particles and fields, eds. G. Semenoff, L. Vinet. CRM Series in Mathematical Physics, Berlin
HeidelbergNew York: Springer-Verlag, 1999, pp. 251352
6. Ruijsenaars, S.N.M., Schneider, H.: A new class of integrable systems and its relation to solitons. Ann.
Phys. (N.Y.) 170, 370405 (1986)
7. Ruijsenaars, S.N.M.: Complete integrability of relativistic CalogeroMoser systems and elliptic function
identities. Commun. Math. Phys. 110, 191213 (1987)
8. Whittaker, E.T., Watson, G.N.: A course of modern analysis. Cambridge: Cambridge Univ. Press, 1973
9. Ruijsenaars, S.N.M.: First order analytic difference equations and integrable quantum systems. J. Math.
Phys. 38, 10691146 (1997)
10. Watson, G.N.: The continuation of functions defined by generalized hypergeometric series. Trans. Camb.
Phil. Soc. 21, 281299 (1910)
11. Atakishiyev, N.M., Suslov, S.K.: Difference hypergeometric functions. In: Progress in Approximation
Theory, A.A. Gonchar and E.B. Saff, eds., BerlinHeidelbergNew York: Springer-Verlag, 1992, pp. 1
35
12. Kurokawa, N.: Multiple sine functions and Selberg zeta functions. Proc. Jap. Acad., Ser. A 67, 6164
(1991)
13. Kurokawa, N.: Gamma factors and Plancherel measures. Proc. Jap. Acad., Ser. A 68, 256260 (1992)
14. Barnes, E.W.: The genesis of the double gamma functions. Proc. London Math. Soc. 31, 358381 (1900)
15. Barnes, E.W.: The theory of the double gamma function. Phil. Trans. Royal Soc. (A) 196, 265387 (1901)
16. Shintani, T.: On a Kronecker limit formula for real quadratic fields. J. Fac. Sci. Univ. Tokyo, Sect. 1A 24,
167199 (1977)
17. Nishizawa, M.: On a q-analogue of the multiple gamma functions. Lett. Math. Phys. 37, 201209 (1996)
18. Ueno, K., Nishizawa, M.: The multiple gamma function and its q-analogue. In: Quantum groups and
quantum spaces, Banach Center Publications Vol. 40, Inst. of Math., Polish Ac. Sciences, Warszawa,
1997, pp. 429441
19. Gunning, R.C., Rossi, H.: Analytic functions of several complex variables. Englewood Cliffs, N.J.:
Prentice-Hall, 1965
20. Hrmander, L.: An introduction to complex analysis in several variables. London: Van Nostrand, 1966

Communicated by T. Miwa

Potrebbero piacerti anche