ARTICLE IN PRESS

Advances in Mathematics 191 (2005) 29–45

http://www.elsevier.com/locate/aim

Identities between q-hypergeometric and

hypergeometric integrals of different dimensions
V. Tarasova,c,1 and A. Varchenkob,
,2
a
St. Petersburg Branch of Steklov Mathematical Institute, Fontanka 27, St. Petersburg 191011, Russia
b
Department of Mathematics, University of North Carolina at Chapel Hill,
Chapel Hill, NC 27599-3250, USA
c
Department of Mathematical Sciences, Indiana University Purdue University at Indianapolis,
Indianapolis, IN, 46202-3216, USA
Received 26 September 2003; accepted 24 January 2004

communicated by P.Etingof

Abstract

Given complex numbers m1 ; l1 and nonnegative integers m2 ; l2 ; such that m1 þ m2 ¼ l1 þ l2 ;

for any a; b ¼ 0; y; minðm2 ; l2 Þ we define an l2 -dimensional Barnes type q-hypergeometric
integral Ia;b ðz; m; m1 ; m2 ; l1 ; l2 Þ and an l2 -dimensional hypergeometric integral
Ja;b ðz; m; m1 ; m2 ; l1 ; l2 Þ: The integrals depend on complex parameters z and m: We show
that Ia;b ðz; m; m1 ; m2 ; l1 ; l2 Þ equals Ja;b ðem ; z; l1 ; l2 ; m1 ; m2 Þ up to an explicit factor, thus
establishing an equality of l2 -dimensional q-hypergeometric and m2 -dimensional hypergeo-
metric integrals. The identity is based on the ðglk ; gln Þ duality for the qKZ and dynamical
difference equations.
r 2004 Elsevier Inc. All rights reserved.

Keywords: Hypergeometric integrals; q-hypergeometric integrals; Knizhnik–Zamolodchikov equations;

(glk , gln ) duality

Corresponding author.
E-mail addresses: vt@pdmi.ras.ru, vtarasov@math.iupui.edu (V. Tarasov), anv@email.unc.edu,
varchenko@math.unc.edu (A. Varchenko).
1
Supported in part by RFFI Grant 02-01-00085a and CRDF Grant RM1-2334-MO-02.
2
Supported in part by NSF Grant DMS-0244579.

0001-8708/$ - see front matter r 2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.aim.2004.01.007
 ARTICLE IN PRESS
30 V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45

1. Introduction

1.1. q-Hypergeometric integrals

Let k be a positive number. Let m1 ; l1 be complex numbers and m2 ; l2 nonnegative

integers such that

m1 þ m2 ¼ l1 þ l2 :

We say that an integer a is admissible with respect to m2 ; l2 if

0papminðm2 ; l2 Þ:

For a pair of admissible numbers a; b we define a function Ia;b ðz; m; m1 ; m2 ; l1 ; l2 Þ of

complex variables z; m: The function is defined as an l2 -dimensional Barnes type
q-hypergeometric integral:

Ia;b ðz; m; m1 ; m2 ; l1 ; l2 Þ
Z
¼ Fl2 ðt; z; m; m1 ; m2 Þwl2 a;a ðt; z; m1 ; m2 ÞWl2 b;b ðt; z; m1 ; m2 Þ dtl2 : ð1:1Þ
dl2 ðz;m1 ;m2 Þ

Here t ¼ ðt1 ; y; tl2 Þ and dtl2 ¼ dt1 ydtl2 : The functions Fl2 ðt; z; m; m1 ; m2 Þ;
wl2 a;a ðt; z; m1 ; m2 Þ and Wl2 b;b ðt; z; m1 ; m2 Þ are defined below. The l2 -dimensional
integration contour dl2 ðz; m1 ; m2 Þ lies in Cl2 and is also defined below.
The q-master function Fl is defined by the formula

!
X
l Y
l
Gðtu =kÞGððtu  zÞ=kÞ
Fl ðt1 ; y; tl ; z; m; m1 ; m2 Þ ¼ exp m tu =k
u¼1 u¼1
Gððtu þ m1 Þ=kÞGððtu  z þ m2 Þ=kÞ
Y Gððtu  tv þ 1Þ=kÞ
 :
1puovpl
Gððtu  tv  1Þ=kÞ

The rational weight function wla;a is defined by the formula

Y tu  tv Y l
1
wla;a ðt1 ; y; tl ; z; m1 ; m2 Þ ¼
t  tv  1 u¼1 tu þ m1
1puovpl u
" #
Y l
tu Y tu  tv  1
 Sym ;
u¼laþ1 u
t  z þ m2 1puovpl tu  tv
 ARTICLE IN PRESS
V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45 31

P
where Sym f ðt1 ; y; tl Þ ¼ sASl f ðts1 ; y; tsl Þ: The trigonometric weight function
Wlb;b is defined by the formula

Y sinðpðtu  tv Þ=kÞ Y l
epitu =k
Wlb;b ðt1 ; y; tl ; z; m1 ; m2 Þ ¼
1puovpl
sinðpðtu  tv  1Þ=kÞ u¼1 sinðpðtu þ m1 Þ=kÞ
"
Yl
epiz=k sinðptu =kÞ
 Sym
u¼lbþ1
sinðpðtu  z þ m2 Þ=kÞ
#
Y sinðpðtu  tv  1Þ=kÞ
 :
1puovpl
sinðpðtu  tv Þ=kÞ

Integral (1.1) is defined for 0oIm mo2p: With respect to other parameters we
define integral (1.1) by analytic continuation from the region where m1 ; m2 are
complex numbers with negative real parts and Re z ¼ 0: In that case we put

dl ðz; m1 ; m2 Þ ¼ fðt1 ; y; tl ÞACl j Re tu ¼ e; u ¼ 1; y; lg;

where e is a positive number less than minðRe m1 ; Re m2 Þ: In the considered

region of parameters the integrand in (1.1) is well defined on dl2 ðz; m1 ; m2 Þ for any
a; b; the integral is convergent and gives a meromorphic function of z; m1 ; m2 ; see
[TV1]. It is also known that Ia;b ðz; m; m1 ; m2 ; l1 ; l2 Þ can be analytically continued to a
nonnegative integer value of m2 ; if a; b are admissible with respect to m2 ; l2 at that
point, and the analytic continuation is given by the integral over a suitable
deformation of the imaginary plane fðt1 ; y; tl2 ÞACl2 j Re tu ¼ 0; u ¼ 1; y; l2 g;
see [MuV].

Remark. There is an alternative way to describe the integrand of integral (1.1), again
writing it down as a product of three factors. Namely, consider the functions

Xl ðt1 ; y; tl ; z; m; m1 ; m2 Þ
!
lðlþ3Þ=2
X
l
¼ ðpkÞ exp ðm  piÞ tu =k
u¼1
Y
l
 Gðtu =kÞGððm1 þ tu Þ=kÞGððtu  zÞ=kÞGððz  m2  tu Þ=kÞ
u¼1
Y
 ðtu  tv Þ sinðpðtu  tv Þ=kÞGððtu  tv þ 1Þ=kÞGððtv  tu þ 1Þ=kÞ;
1puovpl

" #
Y
l a Y
l Y tu  tv  1
pla;a ðt1 ; y; tl ; z; m2 Þ ¼ Sym ðtu  z þ m2 Þ tu ;
u¼1 u¼laþ1 1puovpl
tu  tv
 ARTICLE IN PRESS
32 V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45

"
Y
l b
Plb;b ðt1 ytl ; z; m2 Þ ¼ expðpibz=kÞ Sym sinðpðtu  z þ m2 Þ=kÞ
u¼1
#
Y
l Y sinðpðtu  tv  1Þ=kÞ
 sinðptu =kÞ :
u¼lbþ1 1puovpl
sinðpðtu  tv Þ=kÞ

Then

Fl2 ðt; z; m; m1 ; m2 Þwl2 a;a ðt; z; m1 ; m2 ÞWl2 b;b ðt; z; m1 ; m2 Þ

¼ Xl2 ðt; z; m; m1 ; m2 Þpl2 a;a ðt; z; m2 ÞPl2 b;b ðt; z; m2 Þ:

In this description of the integrand the function Xl is such that it contains all the
poles of the integrand and has no zeros anywhere but on the shifted diagonals:
ðtu  tv ÞAkZ; the functions pla;a are polynomials in t1 ; y; tl ; z; m2 ; and the functions
Plb;b are trigonometric polynomials in t1 ; y; tl ; z; m2 :

1.2. Hypergeometric integrals

Let k be a positive number. Let m1 ; l1 be complex numbers and m2 ; l2 nonnegative

integers such that
m1 þ m2 ¼ l1 þ l2 :

For a pair of admissible numbers a; b we define a function Ja;b ðz; m; m1 ; m2 ; l1 ; l2 Þ of

complex variables z; m: The function is defined as an l2 -dimensional hypergeometric
integral
Z
Ja;b ðz; m; m1 ; m2 ; l1 ; l2 Þ ¼ Cl2 ðt; z; m; m1 ; m2 Þgl2 a;a ðt; zÞ dtl2 : ð1:2Þ
gl2 b;b ðzÞ

Here t ¼ ðt1 ; y; tl2 Þ; dtl2 ¼ dt1 ; y; dtl2 : The functions Cl2 ðt; z; m; m1 ; m2 Þ and
gl2 a;a ðt; zÞ are defined below. The l2 -dimensional integration contour gl2 b;b ðzÞ lies
in Cl2 and is also defined below.
The master function Cl is defined by the formula

Cl ðt1 ; y; tl ; z; m; m1 ; m2 Þ

Y
l Y
¼ tðmþm
u
1 þm2 2lþ1Þ=k
ð1  tu Þm1 =k ðz  tu Þm2 =k ðtu  tv Þ2=k :
u¼1 1puovpl

The weight function gla;a is defined by the formula

" #
Y
l a
1 Yl
1
gla;a ðt1 ; y; tl ; zÞ ¼ Sym :
u¼1
1  tu u¼laþ1 z  tu
 ARTICLE IN PRESS
V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45 33

Fig. 1. Integration contour glb;b ðzÞ:

We define the integral in (1.2) by analytic continuation from the region

za0; 0oarg zo2p; Re m{0: ð1:3Þ

In that region the integration contour glb;b ðzÞ is shown in Fig. 1. It has the form
glb;b ðzÞ ¼ fðt1 ; y; tl ÞACl j tu ACu ; u ¼ 1; y; lg: Here Cu ; u ¼ 1; y; l; are noninter-
secting oriented loops in C: The first b loops start at infinity in the direction of z; go
around z; and return to infinity in the same direction. For 1puovpb; the loop Cu
lies inside the loop Cv : The last l  b loops start at infinity in the real positive
direction, go around 1; and return to infinity in the same direction. For b þ
1puovpl; the loop Cu lies inside the loop Cv :
If z; m are in region (1.3), we fix a univalued branch of the master function Cl over
glb;b ðzÞ by fixing the arguments of all its factors. Namely, we assume that at the
point of glb;b ðzÞ where all numbers t1 =z; y; tb =z; tbþ1 ; y; tl belong to ð0; 1Þ we have

arg tu A½0; 2pÞ; arg ð1  tu ÞAðp; pÞ; argðz  tu ÞAð0; 2pÞ; argðtu  tv ÞA½0; 2pÞ;

for u ¼ 1; y; l; v ¼ u þ 1; y; l: Integral (1.2) is convergent in region (1.3).

1.3. Main result

Theorem 1. Let k be a generic positive number. Let m1 ; l1 be complex numbers and

m2 ; l2 nonnegative integers, such that m1 þ m2 ¼ l1 þ l2 : Let 0oIm mo2p: Then for
any a; b ¼ 1; y; minðm2 ; l2 Þ we have

Cb ðm1 ; m2 ; l1 ; l2 ÞIa;b ðz; m; m1 ; m2 ; l1 ; l2 Þ

¼ Db ðm1 ; m2 ; l1 ; l2 ÞEb ðl1 ; l2 ÞX ðz; m1 ; m2 Þ

 Y ðm; m1 ; m2 ; l1 ; l2 ÞJa;b ðem ; z; l1 ; l2 ; m1 ; m2 Þ; ð1:4Þ

 ARTICLE IN PRESS
34 V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45

where

Y
ð2piÞl2 l2 b1 Y
b1
Cb ðm1 ; m2 ; l1 ; l2 Þ ¼ sinðpðm1  jÞ=kÞ sinðpðm2  jÞ=kÞ
l2 !ðl2  bÞ!b! j¼0 j¼0
lY
2 1
Gð1 þ 1=kÞGð1 þ ðm1  jÞ=kÞ

j¼0
Gð1 þ ð j þ 1Þ=kÞ

m2Y
b1 Y
b1
1 1
Db ðm1 ; m2 ; l1 ; l2 Þ ¼ ð2iÞm2
j¼0
sinðpð j þ 1Þ=kÞ j¼0 sinðpð j þ 1Þ=kÞ
mY
2 1
Gð1 þ ðl1  jÞ=kÞ
 ;
j¼0
Gð1=kÞGð1 þ ð j þ 1Þ=kÞ

Eb ðl1 ; l2 Þ ¼ expðpiðb2  ðb  l2 Þðl1 þ l2 Þ  l2 ðl2  1Þ=2Þ=kÞ;

mY
2 1
Gðð j  m1  tÞ=kÞ
X ðt; m1 ; m2 Þ ¼ ; ð1:5Þ
j¼0
Gðð j þ 1  tÞÞ=kÞ

Y ðm; m1 ; m2 ; l1 ; l2 Þ ¼ eml2 ðl2 2m1 1Þ=2k ð1  em Þðl1 þ1Þl2 =k ; argð1  em ÞAðp; pÞ: ð1:6Þ

Remark. There is a similar theorem establishing an equality of suitable hypergeo-

metric integrals of different dimensions, see [TV4]. The factor Db ðm1 ; m2 ; l1 ; l2 Þ in the
present paper corresponds to the normalization factor Cb ðl1 ; l2 ; m1 ; m2 Þ in [TV4] and
contains the same product of sines and gamma-functions.

Example. Let m2 ¼ l2 ¼ 1: In this case formula (1.4) becomes the classical equality
of two integral representations of the Gauss hypergeometric function 2 F1 : For
instance, if a ¼ b ¼ 0; then taking a ¼ m1 =k; b ¼ ðz þ m1 Þ=k; g ¼ ð1  z 
m1 Þ=k; after simple transformations one gets
Z þiNe
1 Gðs þ bÞ
esðmpiÞ GðsÞGðs þ aÞ ds
2pi iNe Gðs þ gÞ
Z þN
m gab GðbÞ
¼ ð1  e Þ tb ðt  1Þa1 ðt  em Þbg dt
Gðg  aÞ 1
Z 1
GðbÞ GðaÞGðbÞ
¼ ua1 ð1  uÞga1 ð1  uem Þb du ¼ m
2 F1 ða; b; g; e Þ:
Gðg  aÞ 0 GðgÞ
 ARTICLE IN PRESS
V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45 35

Here it is assumed that Re g4Re a40; Re b40 and 0oeominðRe a; Re bÞ:

The second equality is obtained by the change of integration variable
u ¼ ðt  1Þ=ðt  em Þ:

Theorem 1 claims that an l2 -dimensional q-hypergeometric integral equals an

m2 -dimensional hypergeometric integral up to an explicit factor. Note that in the first
integral the numbers m1 ; m2 are shifts of arguments of the gamma-functions entering
its q-master function, while in the second integral m2 is its dimension and m1 is not
present explicitly.
It is well known that studying asymptotics of integrals with respect to their
dimension is an interesting problem appearing, for instance, in the theory of
orthogonal polynomials and in matrix models. The duality of the theorem allows us
to study asymptotics of integrals with respect to their dimension. Namely, assume
that in the 4-tuple ðm1 ; m2 ; l1 ; l2 Þ the nonnegative integer l2 tends to infinity while the
numbers m1 and m2 remain fixed. Then Ia;b ðz; m; m1 ; m2 ; l1 ; l2 Þ is a q-hypergeometric
integral of growing dimension l2 ; whose master function has fixed shifts m1 ; m2 : At
the same time, Ja;b ðem ; z; l1 ; l2 ; m1 ; m2 Þ is a hypergeometric integral of the fixed
dimension m2 ; whose master function has growing exponents l1 ; l2 : The asymptotics
of Ja;b ðem ; z; l1 ; l2 ; m1 ; m2 Þ can be calculated using the steepest descent method. An
example of such calculation is given in [TV4].
Similarly, one can assume that in the 4-tuple ðm1 ; m2 ; l1 ; l2 Þ the nonnegative integer
m2 tends to infinity while the numbers l1 and l2 are fixed. Then Ja;b ðem ; z; l1 ; l2 ; m1 ; m2 Þ
is a hypergeometric integral of growing dimension m2 ; whose master function has
fixed exponents l1 ; l2 : On the other hand, Ia;b ðz; m; m1 ; m2 ; l1 ; l2 Þ is a q-hypergeometric
integral of the fixed dimension l2 ; whose master function has growing shifts m1 ; m2 :
The asymptotics of Ia;b ðz; m; m1 ; m2 ; l1 ; l2 Þ in principle can be calculated using the
Stirling formula for asymptotics of the gamma-function.
To prove Theorem 1 we show that the matrices

ðIa;b ðz; m; m1 ; m2 ; l1 ; l2 ÞÞ0pa;bpminðm2 ;l2 Þ ð1:7Þ

and

X ðz; m1 ; m2 ÞðJa;b ðem ; z; l1 ; l2 ; m1 ; m2 ÞÞ0pa;bpminðm2 ;l2 Þ

satisfy the same system of first-order linear difference equations with respect to z; see
Corollary 4 and Theorems 5 and 7. Studying asymptotics of those matrices as
Re z-  N allows us to compute the connection matrix, thus proving Theorem 1.
The fact that both matrices satisfy the same system of difference equations is based
on the duality of the qKZ and dynamical equations for glk and gln [TV3], which is a
generalization of the duality between the rational differential KZ and dynamical
equations observed in [TL]. Namely, in [TV2] a system of dynamical difference
equations was introduced. It is proved there that the dynamical difference equations
are compatible with the trigonometric KZ differential equations. In [MV]
hypergeometric solutions of the trigonometric KZ and dynamical difference
 ARTICLE IN PRESS
36 V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45

equations were presented. On the other hand, q-hypergeometric solutions of the

rational qKZ difference equations were constructed in [TV1]. It was shown in [TV3]
that the system of dynamical difference equations for glk and the system of rational
qKZ equations for gln are naturally transformed into each other under the ðglk ; gln Þ
duality. In this way one gets two sets of solutions of the same system of equations,
and in principle, these two sets of solutions can be identified. Theorem 1 is a
realization of this idea for the case of k ¼ n ¼ 2: We will discuss the case of an
arbitrary pair k; n in a separate paper.
Let us make an additional remark. In [TV3] it was introduced a system of
dynamical differential equations which is transformed under the ðglk ; gln Þ duality to
the trigonometric KZ differential equations. It is proved in [TV5] that the dynamical
differential equations are compatible with the rational qKZ difference equations. It is
shown in [TV6] that the q-hypergeometric solutions of the qKZ equations satisfy the
dynamical differential equations. In the present case of k ¼ n ¼ 2 this means that
matrices (1.7) and

Y ðm; m1 ; m2 ; l1 ; l2 ÞðJa;b ðem ; z; l1 ; l2 ; m1 ; m2 ÞÞ0pa;bpminðm2 ;l2 Þ ð1:8Þ

satisfy the same system of first order linear differential equations with respect to m:
Therefore, the connection matrix of (1.7) and (1.8) does not depend on m; which
agrees with formula (1.4). Other factors in (1.4) also have natural explanation in
terms of the q-deformation of the ðglk ; gln Þ duality.
The rest of the paper is as follows. In Section 2, we discuss the ðgl2 ; gl2 Þ duality for
the qKZ and dynamical equations. In Section 3, we describe their hypergeometric
solutions and prove Theorem 1. In Section 4, we give some facts about the dynamical
differential equations.

2. The ðgl2 ; gl2 Þ duality for KZ and dynamical equations

In this section we follow the exposition in [TV3].

2.1. The trigonometric KZ and associated dynamical difference equations

Let Eij ; i; j ¼ 1; 2; be the standard generators of the Lie algebra gl2 : Let h ¼
C  E11 "C  E22 be the Cartan subalgebra, while E12 and E21 are, respectively, the
positive and negative root vectors.
The trigonometric r-matrix rðzÞAgl#2 2 is defined by the formula

1
rðzÞ ¼ ððz þ 1ÞðE11 #E11 þ E22 #E22 Þ=2 þ zE12 #E21 þ E21 #E12 Þ:
z1

Fix a nonzero complex number k: The trigonometric Knizhnik–Zamolodchikov

ðKZÞ operators r1 ; y; rn are the following differential operators with coefficients in
 ARTICLE IN PRESS
V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45 37

Uðgl2 Þ#n acting on functions of complex variables z1 ; y; zn ; l1 ; l2 :

!
@ X
2
1X n
ðbÞ ðaÞ
Xn
ra ðz1 ; y; zn ; l1 ; l2 Þ ¼ kza  li  Eii Eii  rðza =zb ÞðabÞ : ð2:1Þ
@za i¼1 2 b¼1 b¼1
baa

ðbÞ
Here Eii ¼ 1#?# Eii #?#1; and the meaning of rðza =zb ÞðabÞ is similar. It is
bth
known that the operators r1 ; y; rn pairwise commute.
Let V1 ; y; Vn be gl2 -modules. The trigonometric KZ equations for a function
Uðz1 ; y; zn ; l1 ; l2 Þ with values in V1 #?#Vn are

ra ðz1 ; y; zn ; l1 ; l2 ÞUðz1 ; y; zn ; l1 ; l2 Þ ¼ 0; a ¼ 1; y; n: ð2:2Þ

Let Tu be the difference operator acting on functions f ðuÞ by the formula

ðTu f ÞðuÞ ¼ f ðu þ kÞ:

Introduce a series BðtÞ depending on a complex variable t

X
N Y
s
1
s s
BðtÞ ¼ 1 þ E21 E12 :
s¼1 j¼1
jðt  E11 þ E22  jÞ

For any gl2 -module V with a locally nilpotent action of E12 and finite-dimensional
weight subspaces the series BðtÞ has a well-defined action in any weight subspace
V ½mCV as a rational EndðV ½mÞ-valued function of t:
Let V1 ; y; Vn be gl2 -modules as above. Introduce the dynamical difference
operators Q1 ; Q2 acting on V1 #?#Vn -valued functions of complex variables
z1 ; y; zn ; l1 ; l2 by the formulae
Y
n
E11
ðaÞ
Q1 ðz1 ; y; zn ; l1 ; l2 Þ ¼ ðBðl1  l2 ÞÞ1 za Tl1 ;
a¼1
Y
n
E22
ðaÞ
Q2 ðz1 ; y; zn ; l1 ; l2 Þ ¼ za Bðl1  l2  kÞTl2 : ð2:3Þ
a¼1

One can see that the operators Q1 ; Q2 commute.

Proposition 2 (Tarasov and Varchenko [TV2]). One has ½ra ; Qi  ¼ 0 for all
a ¼ 1; y; n and i ¼ 1; 2:

The dynamical difference equations associated with the trigonometric KZ

equations for a function Uðz1 ; y; zn ; l1 ; l2 Þ with values in V1 #?#Vn are

Qi ðz1 ; y; zn ; l1 ; l2 ÞUðz1 ; y; zn ; l1 ; l2 Þ ¼ Uðz1 ; y; zn ; l1 ; l2 Þ; i ¼ 1; 2: ð2:4Þ

 ARTICLE IN PRESS
38 V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45

The trigonometric KZ and dynamical difference operators preserve the weight

decomposition of V1 #?#Vn : Thus the KZ and dynamical equations can be
considered as equations for a function Uðz1 ; y; zn ; l1 ; l2 Þ taking values in a given
weight subspace of V1 #?#Vn :

2.2. The qKZ equations

Let V ; W be irreducible highest weight gl2 -modules with highest weight vectors
v; w; respectively. There is a unique rational function RVW ðtÞ taking values in
EndðV #W Þ such that

½RVW ðtÞ; g#1 þ 1#g ¼ 0 for any gAgl2 ;

RVW ðtÞðE21 #E11 þ E22 #E21 þ tE21 #1Þ

¼ ðE11 #E21 þ E21 #E22 þ tE21 #1ÞRVW ðtÞ;

RVW ðtÞv#w ¼ v#w:

The function RVW ðtÞ is called the rational R-matrix for the tensor product V #W : It
comes from the representation theory of the Yangian Y ðgl2 Þ:
Let V1 ; y; Vn be irreducible highest weight gl2 -modules. Introduce the qKZ
difference operators Z1 ; y; Zn acting on V1 #?#Vn -valued functions of complex
variables z1 ; y; zn ; l1 ; l2 by the formula
ðaÞ ðaÞ
E11 E22
Za ðz1 yzn ; l1 ; l2 Þ ¼ ðRan ðza  zn ÞyRa;aþ1 ðza  zaþ1 ÞÞ1 l1 l2

 R1a ðz1  za  kÞyRa1;a ðza1  za  kÞTza : ð2:5Þ

These operators are called the qKZ operators. It is known that they pairwise
commute [FR]. The difference equations

Za ðz1 ; y; zn ; l1 ; l2 ÞUðz1 ; y; zn ; l1 ; l2 Þ ¼ Uðz1 ; y; zn ; l1 ; l2 Þ; a ¼ 1; y; n; ð2:6Þ

for a V1 #?#Vn -valued function Uðz1 ; y; zn ; l1 ; l2 Þ are called the qKZ equations.
The qKZ operators preserve the weight decomposition of V1 #?#Vn : Thus the
qKZ and dynamical differential equations can be considered as equations for a
function Uðz1 ; y; zn ; l1 ; l2 Þ taking values in a given weight subspace of
V1 #?#Vn :

2.3. The duality

For a complex number m; denote Mm the Verma module over gl2 with highest
d
weight ðm; 0Þ and highest weight vector vm : The vectors E2;1 vm ; dAZX0 ; form a basis
in Mm :
 ARTICLE IN PRESS
V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45 39

For a nonnegative integer m; denote Lm the irreducible gl2 -module with highest weight
d
ðm; 0Þ and highest weight vector vm : The vectors E2;1 vm ; d ¼ 0; y; m; form a basis in Lm :
Let m1 ; l1 be complex numbers and m2 ; l2 nonnegative integers such that
m1 þ m2 ¼ l1 þ l2 : Consider the weight subspace ðMm1 #Lm2 Þ½l1 ; l2  of the tensor
product Mm1 #Lm2 : The weight subspace has a basis

1
F a ðm1 ; m2 ; l1 ; l2 Þ ¼ E l2 a vm1 #E2;1
a
vm 2 ; a ¼ 0; y; minðm2 ; l2 Þ: ð2:7Þ
ðl2  aÞ!a! 2;1

There is a linear isomorphism

j : ðMm1 #Lm2 Þ½l1 ; l2 -ðMl1 #Ll2 Þ½m1 ; m2 ;

F a ðm1 ; m2 ; l1 ; l2 Þ/F a ðl1 ; l2 ; m1 ; m2 Þ: ð2:8Þ

Theorem 3 (Tarasov and Varchenko [TV3]). The isomorphism j transforms the qKZ
operators acting in ðMm1 #Lm2 Þ½l1 ; l2  into the dynamical difference operators acting in
ðMl1 #Ll2 Þ½m1 ; m2 : More precisely, we have

jZ1 ðz1 ; z2 ; l1 ; l2 Þ ¼ ðGðz1  z2 ; m1 ; m2 ÞÞ1 Q1 ðl1 ; l2 ; z1 ; z2 Þj;

jZ2 ðz1 ; z2 ; l1 ; l2 Þ ¼ Gðz1  z2  k; m1 ; m2 ÞQ2 ðl1 ; l2 ; z1 ; z2 Þj;

where
mY
2 1
t þ j  m1
Gðt; m1 ; m2 Þ ¼ :
j¼0
tþjþ1

Let SðtÞ be any solution of the equation

Sðt þ kÞ ¼ Gðt; m1 ; m2 ÞSðtÞ:

For instance, one of solutions is given by SðtÞ ¼ X ðt; m1 ; m2 Þ; cf. (1.5).

Corollary 4. Let an ðMl1 #Ll2 Þ½m1 ; m2 -valued function Uðz1 ; z2 ; l1 ; l2 Þ solve the
dynamical difference equations:

Qa ðz1 ; z2 ; l1 ; l2 ÞUðz1 ; z2 ; l1 ; l2 Þ ¼ Uðz1 ; z2 ; l1 ; l2 Þ; a ¼ 1; 2:

Then the ðMm1 #Lm2 Þ½l1 ; l2 -valued function

e ðz1 ; z2 ; l1 ; l2 Þ ¼ Sðz1  z2 Þj1 ðUðl1 ; l2 ; z1 ; z2 ÞÞ

U

solves the qKZ equations

e ðz1 ; z2 ; l1 ; l2 Þ ¼ U
Za ðz1 ; z2 ; l1 ; l2 ÞU e ðz1 ; z2 ; l1 ; l2 Þ; a ¼ 1; 2:
 ARTICLE IN PRESS
40 V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45

More facts on the ðgl2 ; gl2 Þ duality for KZ and dynamical equations are given in
Section 4.

3. Hypergeometric solutions

3.1. Hypergeometric solutions of the qKZ equations

For any b ¼ 0; y; minðm2 ; l2 Þ define an ðMm1 #Lm2 Þ½l1 ; l2 -valued function

X
l2
1
I%b ðz; m; m1 ; m2 ; l1 ; l2 Þ ¼ Ia;b ðz; m; m1 ; m2 ; l1 ; l2 Þ E l2 a vm1 #E2;1
a
vm 2 :
a¼0
ðl2  aÞ!a! 2;1
a
Here the actual range of summation is until a ¼ minðm2 ; l2 Þ; since E2;1 vm 2 ¼ 0
for a4m2 :

Theorem 5. Let li ¼ emi ; i ¼ 1; 2: For any b ¼ 0; y; minðm2 ; l2 Þ the function

2 2
Ub ðz1 ; z2 ; l1 ; l2 Þ ¼ eðm1 ðm1 z1 þm2 z2 ðm1 þm2 Þ=2Þðm1 m2 Þðl2 z1 þl2 =2ÞÞ=k

 ð1  em2 m1 Þl2 =k I%b ðz2  z1 ; m2  m1 ; m1 ; m2 ; l1 ; l2 Þ ð3:1Þ

is a solution of the qKZ equations (2.6) with values in ðMm1 #Lm2 Þ½l1 ; l2 : Moreover, if
l1 =l2 is not real, than any solution of that qKZ equations is a linear combination of
functions Ub ðz1 ; z2 ; l1 ; l2 Þ with coefficients being k-periodic functions of z1 ; z2 :

The theorem is a direct corollary of the construction of q-hypergeometric solutions

of the qKZ equations given in [TV1,MuV].
We describe asymptotics of the integral Ia;b ðz; m; m1 ; m2 ; l1 ; l2 Þ as Re z-  N and
m is fixed. For a positive number k; complex numbers m; m; Im mAð0; 2pÞ; and a
nonnegative integer l consider the Selberg-type integral
Z !
Xl Yl
Al ðm; mÞ ¼ exp ðm  piÞ su Gðsu ÞGðsu  m=kÞ
dl ðmÞ u¼1 u¼1
Y l
Gðsu  sv þ 1=kÞ l
 ds : ð3:2Þ
u;v¼1
Gðsu  sv Þ
uav

The integral is defined by analytic continuation from the region where Re m is

negative. In that case

dl ðmÞ ¼ fðs1 ; y; sl ÞACl j Re su ¼ Re m=2; u ¼ 1; y; lg:

In the considered region of parameters the integrand in (3.2) is well defined on

dl ðmÞ and the integral is convergent, see [TV1]. The formula for Al ðmÞ is
 ARTICLE IN PRESS
V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45 41

well known,

Al ðm; mÞ ¼ ð2piÞl eðmpiÞðl12mÞl=2k ð1  em Þlðmlþ1Þ=k

Y
l 1
Gð1 þ ð j þ 1Þ=kÞ
 Gðð j  mÞ=kÞ;
j¼0
Gð1 þ 1=kÞ

where argð1  em ÞAðp; pÞ; see, for example, [TV1].

Remark. Other versions of Selberg-type integrals see in [FSV,TV7].

Lemma 6. Let Re z-  N and m is fixed. Then

Ia;b ðz; m; m1 ; m2 ; l1 ; l2 Þ ¼ l2 !ðl2  bÞ!b!ðpÞl2 expðmzb=kÞ

2
 ðz=kÞð2b þbðm1 m2 2l2 Þþm2 l2 Þ=k

 Al2 b ðm; m1 ÞAb ðm; m2 Þðdab þ Oðz1 ÞÞ:

The lemma follows from [TV1].

3.2. Hypergeometric solutions of the trigonometric KZ and difference dynamical

equations

For b ¼ 0; y; minðm2 ; l2 Þ; define an Ml1 #Ll2 ½m1 ; m2 -valued function

X
m2
1
J%b ðz; m; l1 ; l2 ; m1 ; m2 Þ ¼ Ja;b ðz; m; l1 ; l2 ; m1 ; m2 Þ E m2 a vl1 #E2;1
a
vl 2 :
a¼0
ðm2  aÞ!a! 2;1

a
Here the actual range of summation is until a ¼ minðm2 ; l2 Þ; since E2;1 vl2 ¼ 0 for
a4l2 :

Theorem 7. For any b ¼ 0; y; minðm2 ; l2 Þ the function

ðl ðl1 m2 Þþl2 m2 m22 m1 l1 þl12 =2Þ=k l2 ðl1 m1 þl2 =2Þ=k
Ub ðz1 ; z2 ; l1 ; l2 Þ ¼ z1 1 z2

 ðz1  z2 Þl1 l2 =k J%b ðz2 =z1 ; l2  l1 ; l1 ; l2 ; m1 ; m2 Þ

is a solution of the KZ equations (2.2) and difference dynamical equations (2.4) with
values in ðMl1 #Ll2 Þ½m1 ; m2 :

The theorem is a direct corollary of [MV].

We describe asymptotics of the integral Ja;b ðem ; z; l1 ; l2 ; m1 ; m2 Þ as Re z-  N and
m is fixed. For a positive number k; a complex number l and a nonnegative integer m
 ARTICLE IN PRESS
42 V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45

Fig. 2. The contour gm :

consider the Selberg-type integral

Z Pm Y
m Y
Bm ðlÞ ¼ e s =k
u¼1 u ðsu Þ1l=k ðsu  sv Þ2=k dsm :
gm u¼1 1puovpm

The integration contour gm has the form

gm ¼ fðs1 ; y; sm ÞACm j su ACu ; u ¼ 1; y; mg;

see Fig. 2. Here Cu ; u ¼ 1; y; m; are nonintersecting oriented loops in C: The loops

start at þN; go around 0, and return to þN: For uov; the loop Cu lies inside the
loop Cv : We fix a univalued branch of the integrand by assuming that at the point of
gm where all numbers s1 ; y; sm are negative we have argðsu Þ ¼ 0 for u ¼ 1; y; m;
and argðsu  sv Þ ¼ 0 for 1puovpm:
The formula for Bm ðlÞ is well known

Y
m 1
Gð1  1=kÞ
Bm ðlÞ ¼ ð2piÞm kmðm1lÞ=k ; ð3:3Þ
j¼0
Gð1 þ ðl  jÞ=kÞGð1  ð j þ 1Þ=kÞ

for example, cf. [TV2,MTV].

Lemma 8. Let Re z-  N and m is fixed. Then

Ja;b ðem ; z; l1 ; l2 ; m1 ; m2 Þ ¼ ðm2  bÞ!b!epiðm2 bÞð2bl2 Þ=k

2
 embðzþl1 2m2 þbÞ=k ð1  em Þð2b þbðl1 l2 2m2 Þþm2 l2 Þ=k
2
 ðzÞð2b þbðl1 l2 2m2 Þþm2 ðm2 l1 1ÞÞ=k

 Bm2 b ðl1 ÞBb ðl2 Þðdab þ Oðz1 ÞÞ:

The proof is straightforward.

3.3. Proof of Theorem 1

Theorems 5 and 7, and Corollary 4 imply that for any b ¼ 0; y; minðm2 ; l2 Þ the
functions I%b ðz; m; m1 ; m2 ; l1 ; l2 Þ and X ðz; m1 ; m2 Þj1 ðJ%b ðem ; z; l1 ; l2 ; m1 ; m2 ÞÞ satisfy the
 ARTICLE IN PRESS
V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45 43

same first-order difference equation with respect to z with step k: Hence, for any
a ¼ 0; y; minðm2 ; l2 Þ one has

X ðz; m1 ; m2 ÞJa;b ðem ; z; l1 ; l2 ; m1 ; m2 Þ

X2 ;l2 Þ
minðm
¼ Ia;c ðz; m; m1 ; m2 ; l1 ; l2 ÞGb;c ðz; m; m1 ; m2 ; l1 ; l2 Þ; ð3:4Þ
c¼0

the connection coefficients Gb;c ðz; m; m1 ; m2 ; l1 ; l2 Þ being k-periodic functions of z and

holomorphic functions of m in the strip 0oIm mo2p: Taking into account
asymptotics of the integrals Ia;c ðz; m; m1 ; m2 ; l1 ; l2 Þ and Ja;b ðem ; z; l1 ; l2 ; m1 ; m2 Þ as
Re z-  N and m is fixed, see Lemmas 6 and 8, one can compute the connection
coefficients and obtain formula (1.4). Theorem 1 is proved. &

Remark. One can see from formula (1.4) that all connection coefficients
Gb;c ðz; m; m1 ; m2 ; l1 ; l2 Þ in (3.4) as functions of m are proportional to the same
function Y ðm; m1 ; m2 ; l1 ; l2 Þ: This fact, which is pure computational in the given proof
of Theorem 1, can be observed independently in advance, because Theorems 7
and 12, and Corollary 11 imply that the functions I%b ðz; m; m1 ; m2 ; l1 ; l2 Þ and
Y ðm; m1 ; m2 ; l1 ; l2 Þj1 ðJ%b ðem ; z; l1 ; l2 ; m1 ; m2 ÞÞ satisfy the same first-order differential
equation with respect to m:

4. Differential dynamical operators

Introduce the dynamical differential operators D1 ; D2 with coefficients in Uðgl2 Þ#n

acting on functions of complex variables z1 ; y; zn ; l1 ; l2 by the formula

@ Ee2 X n
ðaÞ
Di ðz1 ; y; zn ; l1 ; l2 Þ ¼ kli þ ii  za Eii
@li 2 a¼1
X
2 X ðaÞ ðbÞ li 0
 Eij Eji  ðE e12  Ee22 Þ:
e21 E
j¼1 1paobpn
li  li 0

Pn ðaÞ
Here Eekl ¼ a¼1 Ekl ; and i0 is supplementary to i; that is, fi; i0 g ¼ f1; 2g:

Proposition 9 (Tarasov and Varchenko [TV5]). One has ½Za ; Di  ¼ 0 for all
a ¼ 1; y; n and i ¼ 1; 2; where Z1 ; y; Zn are the qKZ operators (2.5).

The differential equations

Di ðz1 ; y; zn ; l1 ; l2 ÞUðz1 ; y; zn ; l1 ; l2 Þ ¼ 0; i ¼ 1; 2; ð4:1Þ

 ARTICLE IN PRESS
44 V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45

for a V1 #?#Vn -valued function Uðz1 ; y; zn ; l1 ; l2 Þ are called the dynamical

differential equations associated with the qKZ equations.

Theorem 10 (Tarasov and Varchenko [TV3]). The isomorphism j; see (2.8),

transforms the dynamical differential operators acting in ðMm1 #Lm2 Þ½l1 ; l2  into the
trigonometric KZ operators (2.1) acting in ðMl1 #Ll2 Þ½m1 ; m2 :

jDa ðz1 ; z2 ; l1 ; l2 Þ ¼ ra ðl1 ; l2 ; z1 ; z2 Þj; a ¼ 1; 2:

Corollary 11. Let an ðMl1 #Ll2 Þ½m1 ; m2 -valued function Uðz1 ; z2 ; l1 ; l2 Þ solve the
trigonometric KZ equations:

ra ðz1 ; z2 ; l1 ; l2 ÞUðz1 ; z2 ; l1 ; l2 Þ ¼ 0; a ¼ 1; 2:

Then the ðMm1 #Lm2 Þ½l1 ; l2 -valued function

e ðz1 ; z2 ; l1 ; l2 Þ ¼ j1 ðUðl1 ; l2 ; z1 ; z2 ÞÞ

U

solves the system of the dynamical differential equations (4.1):

e ðz1 ; z2 ; l1 ; l2 Þ ¼ 0;
Da ðz1 ; z2 ; l1 ; l2 ÞU a ¼ 1; 2:

The next statement describes q-hypergeometric solutions of the dynamical

differential equations.

Theorem 12. For any b ¼ 0; y; minðm2 ; l2 Þ the function Ub ðz1 ; z2 ; l1 ; l2 Þ defined by

(3.1) is a solution of Eq. (4.1) with values in ðMm1 #Lm2 Þ½l1 ; l2 :

The theorem follows from [TV6].

Acknowledgments

The authors thank Y. Markov for useful discussions.

References

[FR] I.B. Frenkel, N.Yu. Reshetikhin, Quantum affine algebras and holonomic difference equations,
Comm. Math. Phys. 146 (1992) 1–60.
[FSV] G. Felder, L. Stevens, A. Varchenko, Elliptic Selberg integrals and conformal blocks, preprint,
2002, 1–13; math.QA/0210040.
[MTV] Y. Markov, V. Tarasov, A. Varchenko, The determinant of a hypergeometric period matrix,
Houston J. Math. 24 (2) (1998) 197–220.
[MV] Y. Markov, A. Varchenko, Hypergeometric solutions of trigonometric KZ equations satisfy
dynamical difference equations, Adv. Math. 166 (1) (2002) 100–147.
 ARTICLE IN PRESS
V. Tarasov, A. Varchenko / Advances in Mathematics 191 (2005) 29–45 45

[MuV] E. Mukhin, A. Varchenko, The quantized Knizhnik–Zamolodchikov equation in tensor

products of irreducible slð2Þ-modules, Calogero–Moser–Sutherland models, Montreal, QC,
1997, CRM Series Math. Phys., Springer, New York, 2000.
[TL] V. Toledano-Laredo, A Kohno–Drinfeld theorem for quantum Weyl groups, Duke Math.
J. 112 (3) (2002) 421–451.
[TV1] V. Tarasov, A. Varchenko, Geometry of q-hypergeometric functions as a bridge between
Yangians and quantum affine algebras, Invent. Math. 128 (1997) 501–588.
[TV2] V. Tarasov, A. Varchenko, Difference equations compatible with trigonometric KZ differential
equations, Internat. Math. Res. Notices (15) (2000) 801–829.
[TV3] V. Tarasov, A. Varchenko, Duality for Knizhnik–Zamolodchikov and dynamical equations,
Acta Appl. Math. 73 (1–2) (2002) 141–154.
[TV4] V. Tarasov, A. Varchenko, Identities for hypergeometric integrals of different dimensions,
preprint, 2003, pp. 1–9; math.QA/0305224.
[TV5] V. Tarasov, A. Varchenko, Dynamical differential equations compatible with rational qKZ
equations, preprint, 2004, pp. 1–7; math. QA/0403416.
[TV6] V. Tarasov, A. Varchenko, Hypergeometric solutions of rational qKZ equations satisfy
dynamical differential equations, in preparation.
[TV7] V. Tarasov, A. Varchenko, Selberg-type integrals associated with sl3 ; Lett. Math. Phys. 65 (3)
(2003) 173–185.