TRANSCENDENCE OF VALUES AT ALGEBRAIC POINTS

FOR CERTAIN HIGHER ORDER HYPERGEOMETRIC

FUNCTIONS
(UPDATED)

PIERRE-ANTOINE DESROUSSEAUX, MARVIN D. TRETKOFF,

AND PAULA TRETKOFF

Abstract. In this paper, we describe the exceptional set of a function

studied by Picard. This is the set of algebraic points at which the
function takes algebraic values. In particular, we deduce necessary and
sufficient conditions for the finiteness of the exceptional set. Some of
our results depend on recent conjectures of Pink [21],[22].

1. Introduction
The purpose of this paper is to analyze the exceptional set of a function
studied by Picard in the course of his investigation of Appell’s hypergeomet-
ric function. In particular, we describe the set of algebraic points at which
this function assumes algebraic values. Necessary and sufficient conditions
for this set to be finite are also derived. Some background for this type of
problem is provided in the present section.
One of the recurrent themes in the theory of transcendental numbers is
the problem of determining the set of algebraic numbers at which a given
transcendental function assumes algebraic values. This set has come to
be known as the exceptional set of the function. The classical work of
Hermite (1873), Lindemann (1882) and Weierstrass (1885) established that
the exceptional set associated to the exponential function exp(x), x ∈ C,
is trivial, that is, consists only of x = 0. These results stand among the
highlights of 19th century mathematics because they imply that π is a
transcendental number, thus proving the impossibility of squaring the circle,
a problem dating to the ancient Greeks.
Turning to the 20th century, we recall that C.L. Siegel (1929) suggested
studying the exceptional set of the classical (Gauss) hypergeometric func-
tion of one complex variable F = F (a, b, c; x), where a, b, c are rational
Typeset October 14, 2007 [ 13:52 ] .
1991 Mathematics Subject Classification. 11J91, 33C65.
Key words and phrases. Hypergeometric functions, transcendence, Shimura varieties.
Paula Tretkoff (formerly known by her maiden name “Paula Cohen”) acknowledges
support from NSF grant number DMS-0400942.
1
2 P-A. Desrousseaux, M.D. Tretkoff, P. Tretkoff

numbers. An important related step was taken by Th. Schneider (1937),

who showed that the exceptional set of a Weierstrass elliptic function with
algebraic invariants is void. Subsequently, Schneider (1941) showed that
at least one period of an abelian integral of the first or second kind on any
curve of positive genus defined over the algebraic numbers is transcendental.
More recently, Wüstholz [39] used his analytic subgroup theorem to obtain
the definitive result in this direction: each period is transcendental unless it
is zero.
Returning to Siegel’s question, Wolfart [35] showed that the exceptional
set of the classical hypergeometric function corresponds to a subset of com-
plex multiplication (CM or special) points on a moduli space of abelian
varieties associated to fixed parameters a, b, c. He also gave sufficient con-
ditions on a, b, c for the exceptional set to be infinite. Next, P.B. Cohen
and Wüstholz [9] gave sufficient conditions on a, b, c for the exceptional
set to be finite. Their proof, however, assumed the validity of a particular
case of the André–Oort conjecture which was subsequently established by
Edixhoven and Yafaev in [16].
Research initiated recently [15], [11], [12], [14] extends some of these re-
sults to functions of several complex variables. Here, the Appell-Lauricella
hypergeometric functions of several complex variables play the role of the
Gauss hypergeometric function of one variable. The exceptional set of the
Appell-Lauricella function is the collection of all points, each of whose coor-
dinates is an algebraic number, at which the function assumes an algebraic
value. Once again, there is a family of abelian varieties associated to an
Appell-Lauricella function and a finite to one morphism to its base space
(Shimura variety) from the domain of the Appell-Lauricella function.
At this point, an important distinction between functions of one and
several complex variables manifests itself for the first time in the theory of
transcendence. Namely, we find that the points in the exceptional set of an
Appell-Lauricella function corresponding to abelian varieties with complex
multiplication in general form a proper subset of the exceptional set. The
conditions defining this proper subset are quite subtle. This is in sharp
contrast to the situation for the Gauss hypergeometric function where, as
mentioned earlier, the entire exceptional set corresponds to CM points.
Furthermore, criteria were given in [15] for the exceptional set of an
Appell-Lauricella function not to be Zariski dense in its space of regular
points. Note that, in the case of several complex variables, the property
that replaces finiteness in the one variable case is that the set not be Zariski
dense. In particular, we derive in [15] sufficient conditions for this set not
to be Zariski dense. Our proof depends on as yet unproved conjectures of
Pink, except in some special cases. In [11], [12], [14] sufficient conditions
were given to ensure that the proper subset of the exceptional set corre-
sponding to CM points is not Zariski dense. Moreover, a still unproved case
of the André–Oort conjecture is utilized in our proofs.
 Transcendence of values at algebraic points for certain higher order hypergeometric functions. 3

According to Appell and Kampé de Feriet [1], Picard studied the func-
tion obtained from Appell’s hypergeometric function F (a, b, b0 , c; x, y) of
the complex variables x and y by fixing one of these variables. Setting
x = λ, say, where λ is a fixed number, the resulting function Fλ (y) of one
complex variable satisfies Fλ (0) = F (a, b, c; λ), the value of the Gauss hy-
pergeometric function at λ 1. Assuming that Fλ (0) 6= 0 and dividing by
this number, the resulting function Φλ (a, b, b0 , c; y) = Fλ (y)/Fλ (0) satisfies
Φλ (a, b, b0 , c; 0) = 1. As we follow the spirit of Picard’s work on Appell’s
hypergeometric function, we call this function the Picard function. In the
present paper, we use an alternate set of parameters µ = {µi }5i=0 , and there-
fore denote the Picard function by Φµ,λ = Φµ,λ (y) (see (2.12)). As in [12],
[13], our aim is to relate the exceptional set of Φµ,λ to isogeny classes of
certain abelian varieties and hence to Hecke orbits on a Shimura variety.
Several new features appear in the present paper. First of all, for each
rational ball 5-tuple µ, we distinguish two cases depending on λ and refer to
them in Theorem 4.2 as the irreducible and the reducible case, respectively.
The irreducible case can be viewed as the generic case. Indeed, for all but
finitely many µ, the full André–Oort conjecture for curves predicts that
the reducible case arises for only finitely many λ. Next, we use a recent
conjecture of Pink [21] to supplement the particular case of the André–Oort
conjecture used in [9]. Namely, the subset T may now be the Hecke orbit
of a non-special point. In the irreducible case, this enables us to treat the
situation where the fixed isogeny class is of CM type as well as to treat the
case where it is not of CM type. In the reducible case, we cannot assume
that the Hecke orbit of interest is the orbit of a point, rather it may be the
orbit of a Shimura subvariety of positive dimension. This is the origin of
the extra conditions defining the exceptional set in [12], [13]. In the present
paper, we avoid this complication by appealing to conjectures of Pink [22].
It also turns out that in the reducible case we are sometimes able to apply
the unconditional results of [16]. However, in general, Desrousseaux’s extra
conditions are necessary if we apply the André–Oort conjecture directly.
The plan of this paper is as follows. In §2, we recall some facts about
Appell-Lauricella functions and define the related Picard function. We re-
call the analytic family of abelian varieties associated to an Appell-Lauricella
function in §3. In §4, transcendence techniques are used to describe the ex-
ceptional set in terms of certain isogeny classes of abelian varieties. Finally,

1Picard studied Appell’s hypergeometric function F (a, b, b0 , c; x, y) . Among other

things, he obtained an integral expression for it whose integrand depends on x and y .
Setting x = λ yields an integral of the type studied by Pochhammer [23] in his investi-
gation of nth order ordinary differential equations whose singular points are all regular
and (n + 1) in number. It appears that Pochhammer was anticipated by Tissot. The
authoritative and comprehensive work by Schlesinger [24] in fact refers to the differential
equation under discussion as the “Tissot–Pochhammer differential equation”. We refer
the reader to Schlesinger’s book for a complete account of this fascinating period.
4 P-A. Desrousseaux, M.D. Tretkoff, P. Tretkoff

necessary and sufficient conditions for the finiteness of the exceptional set
of the Picard function are given in §§5 and 6.

2. The Picard function

In this section, we recall some well-known facts about Appell-Lauricella
hypergeometric functions of several complex variables. We also introduce a
related function studied by Picard in the course of his investigations of these
hypergeometric functions. The main results of the present paper concern the
transcendence of values of this Picard function at algebraic arguments.
The Appell–Lauricella hypergeometric functions of n > 1 complex vari-
ables are generalizations of the classical hypergeometric function of one vari-
able, see [AK], Chapter VII. Appell’s name is usually associated to the two
variable case, whereas the case of more variables is usually ascribed to Lau-
ricella.
The proper domain for these functions, which are initially defined by a
system of partial differential equations, is the weighted configuration space
of n distinct points on the projective line. The weights are in general given
by (n+3) real numbers. We assume throughout this note that these weights
µ = {µi }n+2
i=0 are rational numbers satisfying the ball (n + 3)-tuple condition
of [10]:
n+2
X
(2.1) 0 < µi < 1, i = 0, . . . , n + 2, µi = 2.
i=0
We refer to such weights as rational ball tuples.
Matching the indices of the variables xi with those of the weights µi , we
write the space of regular points for these functions as
(2.2) Qn = {x = (x2 , . . . , xn+1 ) ∈ P1 (C)n : xi 6= 0, 1, ∞; xi 6= xj , i 6= j}.
This space has the structure of a quasi-projective variety over Q and can
be identified with
±
Q0n = {(x0 , x1 , x2 , . . . , xn+1 , xn+2 ) ∈ P1 (C)n+3 : xk 6= xl ; k 6= l} Aut(P1 )
where Aut(P1 ) acts diagonally and freely. From this description, one sees
that there is a natural action of the symmetric group Sn on Q0n . Therefore,
any result in the sequel that is true for a given µ = {µi }n+2
i=0 is also true for
the rational ball tuple obtained by permuting the µi .
The Appell-Lauricella hypergeometric functions are solutions of a system
Hµ of linear partial differential equations in n variables xi , i = 2, . . . n + 1
with regular singularities along xi = 0, 1, ∞, xi = xj (j 6= i): these are also
known as the characteristic (hyper)surfaces of Hµ (see for example [34]).
The system Hµ has an (n + 1)-dimensional solution space. We denote its
monodromy group by ∆µ ; it is a subgroup of PU(1, n). The µ for which
∆µ is a lattice in PU(1, n) have been computed (see [10]). When n = 1,
there are infinitely many such lattices, although only finitely many of them
are arithmetic [32]. When n = 2, there are 58 such lattices, of which 15 are
 Transcendence of values at algebraic points for certain higher order hypergeometric functions. 5

not arithmetic. For n between 3 and 12, there are 32 such lattices, of which
only one is not arithmetic. For n ≥ 13, the monodromy group is never a
lattice (see [10], [17], [19]).
There is a unique solution F = Fµ (x) of Hµ which is holomorphic at the
singular point x = 0 and satisfies Fµ (0) = 1. This generalizes the classical
hypergeometric function in one complex variable. For x ∈ Qn , let ω1 be
the differential form
n+1
Y
(2.3) ω1 = ω1 (µ; x) = u−µ0 (u − 1)−µ1 (u − xi )−µi du.
i=2
Even though x = 0 6∈ Qn , we set
Pn+1
(2.4) ω1 (µ; 0) = u−µ0 − i=2 µi
(u − 1)−µ1 du.
For x = (x2 , . . . , xn+1 ) ∈ Qn , we then have
Z ∞ Z ∞
±
(2.5) F = Fµ (x) = ω1 (µ; x) ω1 (µ; 0),
1 1
the normalizing constant ensuring that Fµ (0) = 1. With B(α, β) the clas-
sical Beta function, we may write this as,
Z ∞
−1
(2.6) F = Fµ (x) = B(1 − µ1 , 1 − µn+2 ) ω1 (µ; x).
1
In this paper, we will mainly be interested in the cases n = 1, 2. For
n = 1, the space of regular points is given by,
(2.7) Q1 = {x ∈ C : x 6= 0, 1}
and for n = 2, it is given by,
(2.8) Q2 = {(x, y) ∈ C2 : x, y 6= 0, 1; x 6= y}.
When n = 2, the system Hµ has a 3-dimensional solution space, and its
monodromy group ∆µ is a subgroup of PU(1, 2). For µ = {µi }4i=0 a rational
ball 5-tuple, let µ0 = {µ0i }3i=0 be the rational 4-tuple with,
(2.9) µ00 = µ0 + µ3 , µ01 = µ1 , µ02 = µ2 , µ03 = µ4 .
We assume that µ0 is a rational ball 4-tuple, so that, in particular,
(2.10) µ0 + µ3 < 1.
In [12], [13], with n = 2, it was assumed in addition that µ0 +µ2 +µ3 < 1, in
order that ω1 (µ; 0) be a differential of the first kind. In fact, his assumption
is not necessary for most of the results of the present paper. The system
Hµ0 has a 2-dimensional solution space, and its monodromy group ∆µ0 is a
subgroup of PU(1, 1).
For a fixed x = λ ∈ Q1 , we define the Picard function of a single variable
y ∈ C \ {0, 1, λ} by,
Z ∞ Z
± ∞
(2.11) Φµ,λ = Φµ,λ (y) = ω1 (µ; λ, y) ω1 (µ0 ; λ).
1 1
6 P-A. Desrousseaux, M.D. Tretkoff, P. Tretkoff

In view of (2.5), this function is also given by,

±
(2.12) Φµ,λ (y) = Fµ (λ, y) Fµ0 (λ).
For λ ∈ Q1 , we define the exceptional set Eµ,λ of Φµ,λ to be
∗
(2.13) Eµ,λ = {y ∈ Q \ {0, 1, λ} : Φµ,λ (y) ∈ Q }.
The main focus of this paper is to study criteria for the finiteness of the set
Eµ,λ .
The function Φµ,λ was studied by Picard (see [1], pp71–77). It is the
unique solution Φ = Φ(y), that extends to a holomorphic function at y = 0
and satisfies Φ(0) = 1, of the linear differential equation of order 3 given
by:
d3 Φ d2 Φ dv
(2.14) B0 (λ; y) 3
+ B1 (λ; y) 2
+ B2 (λ; y) + B3 (λ; y)Φ = 0.
dy dy dy
Here,
B0 (λ; y) = y(y − 1)(y − λ)
B1 (λ; y) = (b − c − 1)(y − 1)(y − λ) + (c − a − b0 − 2)y(y − λ) + (−b − b0 − 1)y(y − 1)
B2 (λ; y) = (b0 + 1)[(a + b + b0 + 1 − c)y + c(y − 1) + (a + 1 − b)(y − λ)]
B3 (λ; y) = ab0 (b0 + 1),
and the parameters a, b, b0 , c are related to the µi by:
µ0 = c − b − b0 ,
µ1 = a + 1 − c,
µ2 = b,
µ3 = b0 ,
(2.15) µ4 = 1 − a.
The differential equation (2.14) has regular singular points at y = 0, 1, λ, ∞.
Rh
For y 6= 0, 1, λ, ∞, a basis of solutions is given by g ω(µ; λ, y) for g 6= h in
Rh
{0, 1, λ, ∞}, where g ω(µ; x, y) is a basis of solutions to Hµ at x = λ.

3. Rational ball tuples and Shimura varieties

The contents of this section appear in previous articles but are repro-
duced here for the convenience of the reader. We recall in particular the
construction of certain analytic families of abelian varieties associated with
hypergeometric functions and we identify the Shimura varieties for these
families. Following [26], [28] (where n = 1), we prefer to call these abelian
varieties Prym varieties rather than hypergeometric varieties as in [2]. Al-
though some of our discussion is classical, we refer to [7] (n = 1), [8] (n = 2)
and [2], [27] (n ≥ 3) for more recent descriptions of this construction.
 Transcendence of values at algebraic points for certain higher order hypergeometric functions. 7

Let µ = {µi }n+2

i=0 be a rational ball (n + 3)-tuple and let N be the least
common denominator of the µi . Let K = Q(ζ), where ζ = exp(2πi/N ).
For s ∈ (Z/N Z)∗ , let σs be the Galois embedding of K which maps ζ 7→ ζ s .
By (2.1), for x ∈ Qn , n ≥ 1, the differential ω1 (µ; x) is of the first
kind on a projective non-singular curve Cµ,N = Cµ,N (x) with singular affine
model
n+1
Y
N N µ0 N µ1
w =u (u − 1) (u − xi )N µi .
i=2
n
For x ∈ Q , this curve is defined over a number field.
To every x ∈ Qn , we can associate the Prym variety Tµ,N = Tµ,N (x)
of Cµ,N (x). For proper divisors f of N there is a natural surjection from
Jac(Cµ,N ) to Jac(Cµ,f ). Let Tµ,N = Tµ,N (x) be the connected component of
the origin in the intersection of all the kernels of these endomorphisms. The
automorphism χ : (u, w) 7→ (u, ζ −1 w) of the affine model of Cµ,N induces an
action of ζ on Tµ,N which realizes the field K in the endomorphism algebra
End0 (Tµ,N ) = End(Tµ,N ) ⊗ Q and an automorphism χ∗ of the space of
differentials of the first kind H 0 (Tµ,N , Ω) on Tµ,N . For s ∈ (Z/N Z)∗ , let
Vs be the subspace of the elements of H 0 (Tµ,N , Ω) which are eigenspaces
for the action of K by χ∗ with eigenvalue ζ s , that is, on which K acts
by σs (K). A modern reference for the explicit basis of the subspaces Vs ,
s ∈ (Z/N Z)∗ is [2]. The dimension of Vs has been computed by several
authors (see, for example, [5]) and is given by
n+2
X
(3.1) rs = rs(µ) = −1 + hsµi i.
i=0

For all s ∈ (Z/N Z)∗ , we have rs + r−s = (n + 1). The space H 0 (Tµ,N , Ω) is
the direct sum of the Vs ; therefore the dimension of Tµ,N is (n + 1)ϕ(N )/2.
We say that Tµ,N has (generalized) complex multiplication with (represen-
tation) type
X
(3.2) Ψ = Ψµ = rs σs .
s∈(Z/N Z)∗

Let M be a set of representatives in (Z/N Z)∗ /{±1} of those s with

rs r−s 6= 0. Let m be the cardinality of M and denote its elements by rsj ,
j = 1, . . . , m. For s ∈ (Z/N Z)∗ and x ∈ Qn , we let
n+1
Y
(s) (s) −hsµ0 i −hsµ1 i
(3.3) ω =ω (µ; x) = u (u − 1) (u − xi )−hsµi i du.
i=2

When rsj = 1, the differential form ω (sj )

is of the first kind and generates
Vsj . We can suppose that s1 = 1, and therefore that ω (s1 ) = ω1 = du/w .
Let hsj µi be the rational ball (n + 3)-tuple given by {hsj µi i}n+2i=0 . The
Prym variety of Cµ,N (hsj µi; x) is also Tµ,N (x). When n = 1, we have
rs + r−s = 2, and therefore the number rs r−s equals 1 if it is non-vanishing.
8 P-A. Desrousseaux, M.D. Tretkoff, P. Tretkoff

When n = 2, we have rs + r−s = 3, and therefore the number rs r−s equals

2 if it is non-vanishing.
The abelian variety Tµ,N is principally polarized and, as a complex torus,
has lattice L isomorphic as a Z-module to Z[ζ](n+1) . The data (K, Ψ, L)
determines a Shimura variety S . Writing
X
(3.4) Ψ= rs σs ,
s∈(Z/N Z)∗

we have
X
(3.5) dim(S) = rs r−s
s∈(Z/N Z)∗ /{±1}

The Shimura variety S is the quotient of a complex symmetric domain H(Ψ)

by an arithmetic (or modular) group Γi . We have,
Y
(3.6) H(Ψ) = Hrs ,r−s ,
s∈(Z/N Z)∗ /{±1}

where Hu,v is a point if uv = 0 and, otherwise,

© ª
(3.7) Hu,v = z ∈ Mu,v (C) : 1 − z t z positive and hermitian .
When s = 1 we have r1 = 1, r−1 = n and H1,n is the complex n-ball Bn .
For details, see [29].
To every x ∈ Qn , we can associate the point (or module) Px in S(C)
which represents the isomorphism class of TN (x). Let Z be the Zariski
closure of {Px : x ∈ Qn }; it is an irreducible closed subvariety of S with
dimension dim(Z) = n.

4. Transcendence arguments
Some of the results of this section are implicit in Lemme 2.2, Lemme
2.3, and Lemme 2.4 of [13]. One of the new features in the present paper
is that the irreducible and reducible cases of our Theorem 4.2 are treated
separately.
For two non-zero complex numbers a, b, we write a ∼ b if a/b is algebraic.
We say then that a and b are proportional over Q.
For two abelian varieties A and B , we write A=B b when A and B are
isogenous. For an abelian variety C defined over Q, let H 0 (C, ΩQ ) be
the space of differential forms which are of the first kind on C and which
are defined over Q. Denote by PC the Q-vector space generated by the
numbers, ½Z ¾
ω: ω ∈ H 0 (C; ΩQ ); γ ∈ H1 (C; Z) .
γ
If the endomorphism algebra End0 (C) of C contains a number field F ,
then we say that C is F -stable. If C is F -stable and contains no proper
non-trivial F -stable abelian subvariety, we say that C is F -irreducible.
Otherwise, we say that C is F -reducible. If F = K = Q(ζ), let VC,s be
 Transcendence of values at algebraic points for certain higher order hypergeometric functions. 9

the eigenspace of H 0 (C, Ω) on which K acts via σs (K), where the Galois
embedding σs maps ζ to ζ s . In particular σ1 is the identity embedding.
We begin with the following corollary of Theorem 5 in [38] (for more
details see [27], Prop. 1, p.6 and [26], Appendix).
Proposition 4.1. Let A and B be abelian varieties defined over Q. Then
PA ∩ PB 6= {0} if and only if there exist non-trivial simple abelian subvari-
eties A0 of A, and B 0 of B , such that A0 =B
b 0.
We apply this result to A = Tµ,N (λ, y) and B = Tµ0 ,N (λ) for λ ∈ Q,
λ 6= 0, 1, Fµ0 (λ) 6= 0, and y ∈ Eµ,λ (for notations, see §§2 and 3). In that
case, by (2.11) we have the relation
Z ∞ Z ∞
(4.1) ω1 (µ; λ, y) ∼ ω1 (µ0 ; λ),
1 1
between non-zero numbers, so that PTµ,N (λ,y) ∩ PTµ0 ,N (λ) is non-trivial and
contains a non-zero period of ω1 (µ0 ; λ). Therefore, by Proposition 4.1, it
follows that Tµ,N (λ, y) and Tµ0 ,N (λ) share a non-trivial simple factor E up
to isogeny. Let E s , s ≥ 1, be the smallest power of E that is K -stable. Ap-
plying Ex. 3 in §1 of [4] to this situation, we deduce that there are K -stable
abelian varieties C and D , which may be trivial, and a positive multiple u of
b u ×C and Tµ,N (λ, y)=E
s such that Tµ0 ,N (λ)=E b u ×D . Moreover, if Tµ0 ,N (λ)
is K -irreducible, then the abelian variety C is trivial and u = s. Since
dim(Tµ,N (λ, y)) = (3/2) dim(Tµ0 ,N (λ)), we must have dim(D) = ϕ(N )/2.
Therefore D is of CM type and Tµ,N (λ, y)=T b µ0 ,N (λ) × D . If Tµ0 ,N (λ) is
K -reducible then, as its dimension is ϕ(N ), it must split up to isogeny into
a product of two abelian varieties of CM type of dimension ϕ(N )/2. There-
fore E s and C are of CM type and there is a simple abelian variety F ,
which may equal E , such that C =F b v . It follows that Tµ0 ,N (λ)=E
b s × C,
and Tµ,N (λ, y)=E
b s × D where dim(D) = ϕ(N ).
We summarize the above discussion in the following theorem:
Theorem 4.2. Suppose that λ ∈ Q1 ∩ Q, Fµ0 (λ) 6= 0, and y ∈ Eµ,λ . We
have two possibilities:
1) Irreducible case: assume that the abelian variety Tµ0 ,N (λ) is K -irreducible.
Then, there is a fixed abelian variety A3 of CM type such that VA3 ,1 = {0},
and such that for all y ∈ Eµ,λ ,
(4.2) Tµ,N (λ, y)=T
b µ0 ,N (λ) × A3 .
2) Reducible case: assume that the abelian variety Tµ0 ,N (λ) is K -reducible,
and is therefore of CM type. Then, there are fixed abelian varieties A1 and
A2 of CM type such that VA2 ,1 = {0}, and such that for all y ∈ Eµ,λ ,
(4.3) Tµ0 ,N (λ)=A
b 1 × A2 , Tµ,N (λ, y)=A
b 1 × D,
where D is a K -stable abelian variety of dimension ϕ(N ) with VD,1 = {0}.
10 P-A. Desrousseaux, M.D. Tretkoff, P. Tretkoff

As we will observe in §5, the CM type of A3 can be given explicitly in

terms of the rational ball 5-tuple µ.
We have the following partial converse of Theorem 4.2:
2 R∞
RProposition 4.3. For (λ, y) ∈ Q2 ∩ Q , suppose that 1 ω(µ0 , λ) 6= 0 and
∞
1 ω(µ; λ, y) 6= 0.
1) If we have an isogeny of the form

(4.4) Tµ,N (λ, y)=T

b µ0 ,N (λ) × A3 ,

with Tµ0 ,N (λ) of CM type, then

Z ∞ Z ∞
(4.5) ω1 (µ; λ, y) ∼ ω1 (µ0 ; λ).
1 1

2) Let A1 , A2 and D be K -stable abelian varieties. Suppose that A1 ,

A2 are of CM type and that VA2 ,1 = {0}. Moreover, suppose that D has
dimension ϕ(N ) and that VD,1 = {0}. If we have an isogeny of the form

(4.6) Tµ0 ,N (λ)=A

b 1 × A2 , Tµ,N (λ, y)=A
b 1 × D,

then
Z ∞ Z ∞
(4.7) ω1 (µ; λ, y) ∼ ω1 (µ0 ; λ).
1 1

(µ) (µ0 )
Proof: We begin by proving part 1). As r1 = r1 = 1, the isogeny
in (4.4) induces a K -equivariant map on differential 1-forms which sends
∗
ω1 (µ0 ; λ) to α · ω1 (µ; λ, y) where α ∈ Q . As Tµ0 ,N (λ) is of CM type, by
Proposition 5 in [27], the non-zero 0
∗ R periods of ω1 (µR; λ) are0 all proportional
over Q . By (4.4), the numbers γ ω1 (µ; λ, y) and γ ω1 (µ ; λ) are therefore
proportional over Q, where γ is a Pochhammer cycle between 1 and ∞.
Replacing the Pochhammer cycle by the line integral from 1 to ∞ multiplies
∗
the period by an element of Q , and so (4.5) follows. The proof of part 2)
is analogous to the proof of part 1).
Let λ ∈ Q be a fixed number satisfying the conditions of Theorem 4.2.
In the irreducible case, the points of the exceptional set Eµ,λ of the Picard
function correspond to abelian varieties Tµ,N (λ, y) in a fixed isogeny class
represented by Tµ0 ,N (λ) × A3 . In §5, this will allow us to apply a conjec-
ture of Pink in [21]. In the reducible case, the isogeny class of the abelian
variety D may change as y ∈ Eµ,λ varies. There is a subset of Eµ,λ , de-
fined in Théorème 2 of [13] by introducing a second condition on another
Picard function, for which this isogeny class is fixed and of CM type. One
may then apply the André–Oort conjecture. In some cases, this conjecture
suffices without the second condition. In other cases, we may apply another
conjecture of Pink in [22]. We will return to this in §6.
 Transcendence of values at algebraic points for certain higher order hypergeometric functions. 11

5. The irreducible case

We assume throughout this section that λ ∈ Q1 ∩ Q, Fµ0 (λ) 6= 0, and
that the abelian variety Tµ0 ,N (λ) is K -irreducible. By Theorem 4.2, the
exceptional Eµ,λ is then a subset of the set,
(5.1) E µ,λ = {y ∈ Q : (λ, y) ∈ Q2 and Tµ,N (λ, y)=T
b µ0 ,N (λ) × A3 }.
By Proposition 4.3, when Tµ0 ,N (λ) is of CM type, we see that the comple-
ment of Eµ,λ in E µ,λ corresponds to the possible zeros of Φµ,λ in Q. In
the notation of §3, let Z2 be the complex surface in S given by the Zariski
closure of the set {P(x,y) : (x, y) ∈ Q2 }. For λ 6= 0, 1, let Cλ be the complex
curve in S given by the Zariski closure of the set {P(λ,y) : y 6= λ, y ∈ Q1 }.
Consider Shimura morphisms ϕ1 : T → S , ϕ2 : T → T 0 and a point
t ∈ T 0 . An irreducible component of ϕ1 (ϕ−1
0 0
2 (t )), or of its image under a
Hecke operator, is called a weakly special subvariety of S . For a detailed
discussion of weakly special subvarieties see [21].
The following is a special case of Conjecture 1.6 in [21]:
Conjecture 5.1. Let S be a Shimura variety over C and Λ ⊂ S the Hecke
orbit of a point P0 ∈ S . Let C ⊂ S be a closed algebraic curve such that the
set C ∩ Λ has infinite cardinality. Then C is a weakly special curve in S .
As remarked in [21], it is already known that if C is a weakly special
subvariety of S then C ∩ Λ is infinite if it is non-empty, so it is the other
direction of the conjecture that is unknown in general. Points on a Shimura
variety in the same Hecke orbit correspond to isogenous abelian varieties.
Applying part 1) of Theorem 4.2, Proposition 4.3 and Conjecture 5.1, with
P0 the module in S corresponding to Tµ0 ,N (λ)×A3 , we deduce the following:
Theorem 5.2. Let λ ∈ Q1 ∩ Q with Fµ0 (λ) 6= 0. Suppose that the abelian
variety Tµ0 ,N (λ) is K -irreducible. Let Eµ,λ be the exceptional set of the
Picard function. If we assume the validity Conjecture 5.1, then Eµ,λ is
finite if Cλ is a not a weakly special curve in S . If Tµ0 ,N (λ) is of CM type
and Cλ is a weakly special curve in S , then Eµ,λ is of infinite cardinality.
In the case where P0 is a special point and, therefore, corresponds to
an abelian variety with complex multiplication, Conjecture 5.1 is a known
particular case of the André–Oort conjecture. This uses the fact, proved in
[21], that a weakly special curve containing a CM point is a special curve.
This particular case was proved in [16]. We therefore have the following
unconditional result:
Theorem 5.3. Let λ ∈ Q1 ∩ Q with Fµ0 (λ) 6= 0. Suppose that the abelian
variety Tµ0 ,N (λ) is K -irreducible and of CM type. Then Cλ is a special
curve in S if and only if Eµ,λ is of infinite cardinality.
If Tµ0 ,N (λ) is K -irreducible and of CM type, its endomorphism algebra
is then a field which is a totally imaginary quadratic extension of K . The
full strength of the André–Oort conjecture for curves (proved subject to the
12 P-A. Desrousseaux, M.D. Tretkoff, P. Tretkoff

Generalized Riemann Hypothesis in [33]) predicts that there are infinitely

many such λ if and only if the monodromy group ∆µ0 is arithmetic.
We conclude this section by remarking that the CM type of A3 can be
given explicitly in terms of the rational ball 5 tuple µ. For (λ, y) ∈ Q2 , let
Ψµ denote the generalized CM type of Tµ,N (λ, y) and Ψµ0 the generalized
CM type of Tµ0 ,N (λ). By (3.1), we have
X
(5.2) Ψµ = rs(µ) σs
s∈(Z/N Z)∗

where, for all s ∈ (Z/N Z)∗ , we have

(5.3) rs(µ) = −1 + hsµ0 i + hsµ3 i + hsµ1 i + hsµ2 i + hsµ4 i,
(µ) (µ)
and rs + r−s = 3. Again using (3.1),
X 0
(5.4) Ψµ0 = rs(µ ) σs ,
s∈(Z/N Z)∗

where, for all s ∈ (Z/N Z)∗ , we have

0
(5.5) rs(µ ) = −1 + hs(µ0 + µ3 )i + hsµ1 i + hsµ2 i + hsµ4 i,
(µ0 ) (µ0 )
and rs + r−s = 2. Therefore, the CM type of A3 is given by
X
(5.6) Ψ03 = rs(03) σs ,
s∈(Z/N Z)∗

where, for all s ∈ (Z/N Z)∗ , we have

(5.7) rs(03) = hsµ0 i + hsµ3 i − hs(µ0 + µ3 )i,
(03) (03)
and rs + r−s = 1. Here, we used the relation Ψµ = Ψµ0 + Ψ03 .
We denote by S 0 (03) the set of those points (x, y) ∈ P1 (C)×P1 (C) which
lie on the line y = 0 but which do not lie on any other characteristic line
of Hµ . As µ0 + µ3 < 1, the points of S 0 (03) are in the semi-stable com-
pactification Qsst
2 of Q2 constructed in [10]. In the same way as described
in §§4 and 5 of [8] for the case where ∆µ is a lattice, we can associate to
the points of S 0 (03) the points in S which are the modules corresponding
to an abelian variety Tµ0 ,N (λ) × A03 , with A03 isogenous to A3 .

6. The reducible case

We assume throughout this section that λ ∈ Q1 ∩ Q, Fµ0 (λ) 6= 0, and
that the abelian variety Tµ0 ,N (λ) is K -reducible and, therefore, of CM type.
The full strength of the André–Oort conjecture for curves, proved subject
to the Generalized Riemann Hypothesis in [33], predicts that there are only
finitely many such λ if and only if ∆µ0 is not arithmetic. By Theorem 4.2,
there are fixed abelian varieties A1 and A2 of CM type such that, for all
y ∈ Eµ,λ ,
(6.1) Tµ0 ,N (λ)=A
b 1 × A2 , Tµ,N (λ, y)=A
b 1 × D,
 Transcendence of values at algebraic points for certain higher order hypergeometric functions. 13

where D is a K -stable abelian variety of dimension ϕ(N ). There is no

reason to assume a priori that D is fixed as y varies, even though this
sometimes happens as we shall see below. As in §5, we shall use the fact
that points on a Shimura variety in the same Hecke orbit correspond to
isogenous abelian varieties.
Denote by ΨD the generalized CM type of D and by Ψi the CM type of
Ai , i = 1, 2. These types give the representation of K on the holomorphic
1-forms of these abelian varieties. With Ψµ as in (5.2) and Ψ03 as in (5.6),
we have from (6.1) that
(6.2) Ψµ = Ψ1 + Ψ2 + Ψ03 = Ψ1 + ΨD ,
therefore
X
(6.3) ΨD = Ψ2 + Ψ03 = rsD σs ,
s∈(Z/N Z)∗

where rsD + r−s D = 2. If Ψ = Ψ , then we have Ψ = 2Ψ , and by the

2 03 D 03
arguments of [29] we can assume that D is always isogenous to A203 . We
may then apply the particular case of the André–Oort conjecture proved in
[16]. This is the case in the following example: when the monodromy group
∆µ of Fµ (x, y) is a lattice in PU(1, 2) and µ0 + µ2 + µ3 < 1, the point
(x, y) = (0, 0) is a stable point (in the sense of [10]). The analytic family
of Prym varieties Tµ,N (x, y) can therefore be extended to the stable point
(0, 0), as explained in §5 of [8]. Moreover, the corresponding abelian variety
Tµ,N (0, 0) is of CM type and isogenous to A1 × A203 as described in Part 1),
Théorème 2 of [8]. We see from this same reference that we can take A1 to
have CM type conjugate to:
X
(6.4) (hs(1 − µ1 )i + hs(1 − µ4 )i) σs .
s∈(Z/N Z)∗

Therefore, Theorem 4.2 and Proposition 4.3 of this paper, together with
the particular case of the André–Oort conjecture proved in [16], imply the
following:
Theorem 6.1. Let λ ∈ Q1 ∩Q with Fµ0 (λ) 6= 0 and suppose in addition that
Tµ0 ,N (λ)=A
b 1 × A03 . Then the exceptional set Eµ,λ of the Picard function
is infinite if and only if Cλ is a special curve in S . Moreover, there are
infinitely many λ ∈ Q1 ∩ Q with Tµ0 ,N (λ)=A b 1 × A03 if and only if the
Zariski closure of the image of S 0 (03) in S is a special curve.
In the case where the isogeny class of D is not fixed, we may again apply
conjectures of Pink. Let SD be the Shimura variety associated to the data
(K, ΨD , Z[ζ]). From (6.3), the dimension of SD is given by
X
(6.5) dim(SD ) = rsD r−s
D
.
s∈(Z/N Z)∗ /{±1}

Following [22], we define the special closure of an irreducible subvariety Z

of S to be the smallest special subvariety containing Z . Moreover, we say
14 P-A. Desrousseaux, M.D. Tretkoff, P. Tretkoff

that Z is Hodge generic if its special closure is an irreducible component

of S , so that Z is not contained in any special subvariety of S of positive
codimension.
Conjecture 6.2. (see [22]) Consider a Shimura variety S over C and an
irreducible closed subvariety Z of S . Let SZ be the special closure of Z .
Then the intersection of Z with the union of all special subvarieties of S of
dimension strictly less than dim(SZ ) − dim(Z) is not Zariski dense in Z .
In fact, we only need this conjecture in the following weaker form:
Conjecture 6.3. Consider a Shimura variety S over C and an irreducible
closed subvariety Z . Let SZ be the special closure of Z . Then the inter-
section of Z with the Hecke orbit of a special subvariety of S of dimension
strictly less than dim(SZ ) − dim(Z) is not Zariski dense in Z .
We deduce the following from Theorem 4.2 and Proposition 4.3 (using (6.1)):
Theorem 6.4. Let λ ∈ Q1 ∩ Q with Fµ0 (λ) 6= 0. Let Sλ be the special
closure of Cλ . If dim(Sλ ) = 1, then Eµ,λ is infinite. Now assume Conjecture
6.3. If dim(Sλ ) > dim(SD ) + 1, then Eµ,λ is finite.
For example, if S is the special closure of Cλ we have the following:
Proposition 6.5. Suppose Cλ is Hodge generic in S . Then Eµ,λ is finite.
Proof: We must verify the condition
(6.6) dim(Sλ ) > dim(SD ) + 1,
of Theorem 6.4. As Sλ = S by hypothesis, we have
X (µ)
(6.7) dim(Sλ ) = rs(µ) r−s = 2m,
s∈(Z/N Z)∗ /{±1}

(µ) (µ)
where m, as in §3, is the number of s ∈ (Z/N Z)∗ with rs r−s 6= 0. On
the other hand, from (6.1) and (6.3) we see that r1D = 0, so that
(6.8) dim(SD ) ≤ m − 1.
The inequality in (6.6) now follows from (6.7) and (6.8).
In [12], [13], additional natural conditions were imposed on y ∈ Eµ,λ in
order to restrict D to a fixed isogeny class of CM type, thereby allowing
for the application of [16]. The main idea is to utilize the periods of the
differentials of the first kind in the eigensubspace V−1 for Tµ,N (λ, y). This
is in the spirit of the results of Cohen, Shiga, and Wolfart in §5 of [27].
Arguments similar to those of §4 of the present article are then applied to
these periods. These arguments imply that A2 and D in (6.1) share a simple
factor up to isogeny. It follows that D=A b 2 × A03 . We refer to [12], [13] for
details.
 Transcendence of values at algebraic points for certain higher order hypergeometric functions. 15

References
[1] Appell, P., Kampé de Fériet, M.J.: Fonctions hypergéométriques et hypersphériques.
Polynômes d’Hermite, Gauthier-Villars 1926.
[2] Archinard, N.: Hypergeometric abelian varieties. Canad. J. Math. 55 (5) (2003), 897–
932.
[3] Baker, A.: Transcendental number theory. Cambridge Mathematical Library, 1990
(paperback edition), first published in 1975 by Cambridge U. Press.
[4] Bertrand, D.: Endomorphismes de groupes algébriques; applications arithmétiques.
In: Approximations diophantiennes et nombres transcendants, Prog. Math. 31, pp.
1–45 1983.
[5] Chevalley, Cl., Weil, A.: Uber das Verhalten der Integrale 1. Gattung bei Automor-
phismen des Funktionkörpers. Abh. Hamburger Math. Sem. 10 (1934), 358–361.
[6] Cohen, P.B.: Humbert surfaces and transcendence properties of automorphic func-
tions. Rocky Mountain J. Math., 26 (3) (1996), 987–1001.
[7] Cohen, P.B., Wolfart, J: Modular embeddings for some non-arithmetic Fuchsian
groups. Acta Arithmetica, LVI (1990), 93–110.
[8] Cohen, P.B., Wolfart, J.: Fonctions hypergéométriques en plusieurs variables et es-
paces des modules de variétés abéliennes. Ann. scient. Éc. Norm. Sup., 4e série, t.26,
(1993), 665–690.
[9] Cohen, P.B., Wüstholz, G: Application of the André–Oort conjecture to some ques-
tions in transcendence. In: Panorama of Number Theory, A view from Baker’s garden
(ed. by G. Wüstholz), pp. 89–106. Cambridge U. Press, Cambridge 2002.
[10] Deligne, P., Mostow, G.D.: Monodromy of hypergeometric functions and nonlattice
integral monodromy. Publ. I.H.E.S. 63 (1986), 5–89.
[11] Desrousseaux, P-A.: Valeurs exceptionnelles de fonctions hypergéométriques
d’Appell. Ramanujan J. 8 (3) (2004), 331–355.
[12] Desrousseaux, P-A: Fonctions hypergéométriques de Lauricella, périodes de variétés
abéliennes et transcendance. Comptes rendus mathém. de l’Acad. des Sciences, La
Soc. royale de Canada 26 (4) (2004), 110–117.
[13] Desrousseaux, P-A: Valeurs d’une fonction de Picard en des points algébriques, J. of
Number Theory 125 (2007), 95–116.
[14] Desrousseaux, P-A: Exceptional sets of Lauricella hypergeometric functions, Journal
of Algebra, Number Theory and Applications 5 (3), p429–467.
[15] Desrousseaux, P-A, Tretkoff, M.D., Tretkoff, P.: Zariski-density of exceptional sets
for hypergeometric functions, preprint 2005, to appear in Forum Mathematicum.
[16] Edixhoven, S., Yafaev, A: Subvarieties of Shimura varieties. Annals of Math. 157
(2003), 621–645.
[17] Felikson, A.A.: On Thurston signatures. (Russian) Uspekhi Mat. Nauk. 52 (4) (1997),
217–218; translation in Russian Math. Surveys 52 (4) (1997), 826–827.
[18] Koblitz, N., Rohrlich, D.: Simple factors in the Jacobian of a Fermat curve, Can. J.
Math. 30 (1978), 1183–1205.
[19] Mostow, G.D.: Generalized Picard lattices arising from half-integral conditions, Publ.
de l’I.H.E.S. 63 (1986), 91–106.
[20] Noot, R.: Correspondances de Hecke, action de Galois et la Conjecture d’André–Oort
[d’après Edixhoven et Yafaev]. Séminaire Bourbaki, 57ème année, 2004/5, Exp. No.
942 (novembre 2004).
[21] Pink, R.: A Combination of the Conjectures by Mordell-Lang and André-Oort. In:
Geometric Methods in Algebra and Number Theory (Bogomolov, F., Tschinkel, Y.,
Eds.) Progress in Math. 235, Basel: Birkhäuser (2005), 251–282.
[22] Pink, R.: A Common Generalization of the Conjectures of André–Oort. Manin–
Mumford, and Mordell–Lang, preprint, April 17, 2005.
16 P-A. Desrousseaux, M.D. Tretkoff, P. Tretkoff

[23] Pochhammer, L.: Ueber hypergeometrische Funktionen nter Ordnung, J. f. die reine
und angewandte Mathematik, 71, 316–352.
[24] Schlesinger, L.: Handbuch der Theorie der Linearen Differentialgleichungen, Zweiter
Band, Teubner, Stuttgart 1897; reprinted by Johnson Reprint Corp., New York, 1968.
[25] Schneider, Th.: Arithmetische Untersuchungen elliptisher Integrale. Math. Ann. 113
(1937), 1–13.
[26] Shiga, H., Tsutsui, T., Wolfart, J.: Triangle Fuchsian differential equations with
apparent singularities (with an appendix by P. B. Cohen). Osaka J. Math. 41 (2004),
625–658.
[27] Shiga, H., Wolfart, J: Criteria for complex multiplication and transcendence proper-
ties of automorphic functions. J. Reine Angew. Math. 463 (1995), 1–25.
[28] Shiga, H., Wolfart, J.: Algebraic values of Schwarz triangle functions. Technical Re-
ports of Math. Sciences Chiba Univ. 20 (5) 2004.
[29] Shimura, G: On analytic families of polarized abelian varieties and automorphic func-
tions. Annals of Math. 78 (1963), 149–192.
[30] Shimura, G., Taniyama, Y: Complex multiplication of abelian varieties and its appli-
cations to number theory. Publ. Math. Soc. Japan 6, 1961.
[31] Siegel, C-L.: Über einige Anwendungen diophantischer Approximationen. Ges.
Werke, Bd I, pp209–266. Berlin-Heidelberg-New York: Springer 1966.
[32] Takeuchi, K: Arithmetic triangle groups. J. Math. Soc. Japan 29 (1977), 91–106.
[33] Yafaev, A: A conjecture of Yves André, arxiv.org/abs/math.NT/0302125.
[34] Yoshida, M: Fuchsian differential equations. Aspects of math., Vieweg, 1987.
[35] Wolfart, J: Werte hypergeometrische Funktionen. Invent. Math. 92 (1988), 187–216.
[36] Wolfart, J., Wüstholz, G: Der Überlagerungsradius gewisser algeraische Kurven und
die Werte der Betafunktionen an rationalen Stellen. Math. Ann. 273 (1985), 1–15.
[37] Wüstholz, G.: Zum Periodenproblem. Invent. Math. 78 (1984), 381–391.
[38] Wüstholz, G.: Algebraic groups, Hodge theory and transcendence. In: Proc. of the
Intern. Congress of Math. 1, pp 476–483, Berkeley, California, USA, 1986.
[39] Wüstholz, G: Algebraische Punkte auf analytischen Untergruppen algebraischer
Gruppen, Annals of Math. 129 (1989), 501–517.

Current address: Department of Mathematics, Texas A&M University, TAMU 3368,

TX 77843-3368, USA, and UMR 8524 au CNRS, Université des Sciences et Technologies
de Lille, Villeneuve d’Ascq, 59655, FRANCE.
All correspondence to go to: Paula Tretkoff, Department of Mathematics,
Texas A&M University, TAMU 3368, TX 77843-3368, USA, [email protected]
E-mail address: [email protected]
E-mail address: [email protected]
E-mail address: [email protected]