HYPERGEOMETRIC FUNCTIONS, CHARACTER SUMS AND

APPLICATIONS

LING LONG AND FANG-TING TU

Abstract. We summarize several aspects of hypergeometric functions based on our recent work
[9, 8, 12, 16, 17, 18, 20, 21] and our understanding of the subjects.

1. Classical hypergeometric functions and differential equations

For a discussion on the topic, please see [1]. Our approach has overlaps with [12].
1.1. Gamma and beta functions. Let N = Z>0 .
Recall the usual binomial coefficient
 
n n!
= .
k k!(n − k)!
For any a ∈ C and n ∈ N, define the rising factorial or Pochhammer symbol by

(1) (a)n := a(a + 1) · · · (a + n − 1),

and define (a)0 = 1. Note then, that
(−1)k (−n)k
 
n
(2) = .
k! k
Functions gamma function Γ(x) and beta function B(x, y) are defined as follows.
Definition 1. For Re(x) > 0, Z ∞
Γ(x) := tx−1 e−t dt.
0
The function Γ(x) can be extended to a meromorphic function with poles at non-positive integers.
For values outside of those poles, the gamma function satisfies the functional equation
(3) Γ(x + 1) = xΓ(x),
which can easily be derived using integration by parts. By (3), one has that for n ∈ Z≥0 ,
Γ(a + n)
(4) (a)n = .
Γ(a)
Theorem 1.1 (Euler’s Reflection Formula, Theorem 1.2.1 of [1]). For a ∈ C and a 6∈ Z,
π
Γ(a)Γ(1 − a) = .
sin(πa)
√ √
It follows Γ( 12 ) = π, Γ( 31 )Γ( 23 ) = 2 33π .
The gamma function also satisfies multiplication formulas.
This note is based on Fang-Ting Tu’s course on “Hypergeometric Functions” given at LSU in Fall 2020 and
Ling Long’s mini-lectures on “Hypergeometric Functions, Character Sums and Applications” given at University of
Connecticut in 2021. Comments and suggestions will be appreciated. Special thanks to Dr. Bao Pham for his inputs.
1
Theorem 1.2 (Legendre’s Duplication Formula, Theorem 1.5.1 of [1]). For a ∈ C,
 
1/2 2a− 12 1
(5) Γ(2a) (2π) = 2 Γ(a)Γ a + .
2
Stated in terms of rising factorials, Theorem 1.2 gives that for all n ∈ N,
  a + 1
2n a
(6) (a)2n = 2 .
2 n 2 n
Theorem 1.3 (Gauss’ Multiplication Formula, Theorem 1.5.2 of [1]). For m ∈ N and a ∈ C,
   
(m−1)/2 ma− 12 1 m−1
Γ(ma)(2π) =m Γ(a)Γ a + ···Γ a + .
m m
Or equivalently, for any n ∈ N,
m−1
Y 
mn i
(ma)mn = m a+ .
m n
i=0
Gamma values play important role in transcendental number theory. For example, Chudnovsky-
Chudnovsky showed a few Gamma values including Γ( 13 ) are transcendental. Meanwhile, p-adic

Gamma values rationals are algebraic for a whole congruence class of primes p. For example, Γp 13
is algebraic when p ≡ 1 mod 3, see [7, Corollary 11.7.7] a textbook by Cohen.
Theorem 1.4 (Nesterenko [22]). For any imaginary quadratic field with discriminant −d and
character (·) = −d


· , the numbers
√ Y  a (a)
π, eπ d , Γ
d
0<a<d
are algebraically independent.
√
Thus for example, Γ 13 , π, eπ 3 are algebraically independent.

It is widely believed that (3), the reflection and multiplication formulas contain all algebraic
relations among Gamma values for rational numbers. Its finite field analogue is called the Hasse
conjecture, which was proved by Yamamoto ??.
Definition 2. For Re(x) > 0, Re(y) > 0
Z 1
B(x, y) := tx−1 (1 − t)y−1 dt.
0
The assumptions on x and y can be relaxed by integrating along the Pochhammer contour path
around 0 and 1.
Definition 3. Let a, b be two points in CP 1 . Each Pochhammer contour γab is a closed curve
corresponding to a commutator of the form ABA−1 B −1 in the fundamental group of π1 (CP 1 \
{a, b, ∞}), where A, B ∈ π1 (CP 1 \ {a, b, ∞}) are loops around both of the points a, b and the
superscript −1 denotes a path taken in the opposite direction.
Example 1.1. For example, when a = 0, b = 1, a contour γ01 is a closed curve starting from a fixed
point T ∈ (0, 1), going around 0 and 1 counterclockwise (in that order) and returning to T . Then
one loops around 0 and 1 clockwise returning again to T , as indicated by the picture below.

2
Integrating over the double contour loop γ01 , the integral
Z
1
B(x, y) = tx−1 (1 − t)y−1 dt
(1 − e2πix ) (1 − e2πiy ) γ01
converges for all values of x and y. For details, see [36].
Theorem 1.5 (Schneider [25]). For any a, b ∈ Q such that a, b, a + b ∈
/ Z, then B(a, b) is transcen-
dental.
So one of Γ 51 and Γ 25 is transcendental.

The power series expansion of a given analytic function f (z) which is holomorphic at z = 0 is a
useful tool. The formula is given by
X zk
(7) f (z) = f (k) ,
k!
k≥0

where f (k) stands for the kth derivative of f in terms of z. For example, given a ∈ C× , for
f (z) = (1 − z)−a , f (k) = −a(−a − 1) · · · (−a − k + 1)(−1)k = (a)k , so
X (a)k
(8) (1 − z)−a = zk .
k!
k≥0

The next result is generalization of the above.

Theorem 1.6 (Lagrange inversion theorem for formal power series). If f (z) and g(z) are formal
power series where g(0) = 0 and g 0 (0) 6= 0, then Lagrange’s inversion theorem gives a way to write
f as a power series in g(z). In particular, one can write
∞
X
(9) f (z) = f (0) + ck g(z)k ,
k=1

where
f 0 (z)
ck = Resz ,
kg(z)k
and Resz (f ) denotes the coefficient of 1/z in the power series expansion of f .
1.2. Hypergeometric data. Let α = {a1 , ..., an }, β = {1, b2 , ..., bn } with ai , bj ∈ Q be a pair
of multi-sets of the same length. Such a pair is called primitive if ai − bj ∈ / Z for any i, j. Let
M := lcd(α ∪ β) the least positive common denominators of ai , bj ’s.
Definition 4. 1. A multiset α = {a1 , ..., an } is called defined over Q, if nj=1 (X − e2πiaj ) ∈ Z[X].
Q

I.e. for any integer c coprime to lcd(α), α and cα = {ca1 , · · · , can } mod Z.
2. The pair α, β is said to be self-dual if it is congruent to the pair −α, −β mod Z. A hyperge-
ometric datum HD = {α, β; λ} is said to be self-dual if the pair α, β is self-dual; it is defined over
Q if the pair α, β is defined over Q and λ ∈ Q× .
1.3. Classical hypergeometric functions. The classical (generalized) hypergeometric functions
n Fn−1 with complex parameters a1 , . . . , an , b1 = 1, b2 , . . . , bn , and argument z are defined by
∞ ∞
a1 a2 · · · an (a1 )k · · · (an )k k X (a1 )k · · · (an )k z k
  X
(10) n Fn−1 ; z := z = ,
b2 · · · bn (b1 )k · · · (bn )k (b2 )k · · · (bn )k k!
k=0 k=0
3
and converge when |z| < 1. It also satisfies the following inductive integral relation of Euler [1,
Equation (2.2.2)]. Namely, when Re(bn ) > Re(an+1 ) > 0,
a1 a2 · · · an
 
(11) n Fn−1 ; z = B(an , bn − an )−1
b2 · · · bn
Z 1
a1 a2 · · · an−1
 
an −1 bn −an −1
· t (1 − t) · n−1 Fn−2 ; zt dt.
0 b2 · · · bn−1
For the sake of a natural development in the finite field setting, we define in view of (8), for
general a ∈ C,
∞
−a
X (a)k
(12) 1 P0 [a; z] := (1 − z) = z k = 1 F0 [a; z].
k!
k=0
Here we are using the notation 1 P0 to indicate a relationship to periods of algebraic varieties when
a ∈ Q. Next we let
  Z 1 Z 1
a b
(13) 2 P1 ; z := b−1
t (1 − t) c−b−1
1 P0 [a; zt] dt = tb−1 (1 − t)c−b−1 (1 − zt)−a dt,
c 0 0
or one can define it using the Pochhammer contour as in [27, §1.6]. To relate it to the 2 F1 function,
one uses (11) to see that when Re(c) > Re(b) > 0,
  Z 1
a b
(14) 2 P1 ;z = tb−1 (1 − t)c−b−1 · 1 P0 [a; zt]dt
c 0
Z 1
= tb−1 (1 − t)c−b−1 · 1 F0 [a; zt]dt
0
 
a b
= B(b, c − b) · 2 F1 ; z .
c
Inductively one can define the (higher) periods n+1 Pn similarly by
 
a1 a2 · · · an+1
(15) n+1 Pn ; z :=
b2 · · · bn+1
Z 1  
an+1 −1 bn+1 −an+1 −1 a1 a2 · · · an
t (1 − t) n Pn−1 ; zt dt.
0 b2 · · · bn
In this formulation, the order of the ai ’s (resp. bj ’s) matters. Again using the beta function, one
can show that when Re(bi ) > Re(ai+1 ) > 0 for each i ≥ 1,
n+1
a1 a2 · · ·
   
an+1 Y a1 a2 · · · an+1
n+1 Fn ; z = B(ai , bi − ai )−1 · n+1 Pn ;z .
b2 · · · bn+1 b2 · · · bn+1
i=2
By the definition of n+1 Fn in (10), any given n+1 Fn function satisfies two nice properties:
1) The leading coefficient is 1;
2) The roles of the upper entries ai ’s (resp. lower entries bj ’s) are symmetric.
Clearly, the n+1 Pn period functions do not satisfy these properties in general. The hypergeometric
functions can thus be viewed as periods that are ‘normalized’ so that both properties 1) and 2) are
satisfied.
Definition 5. A hypergeometric function with α = {a1 , · · · , an }, β = {1, b2 , · · · , bn } is said to be
well-posed if ai + bi is a constant for all i ∈ [1, n].
4
 In [34] Whipple investigated well-posed series with argument ±1. He obtained a few evaluation
formulas and we will mention a few in this note.
Definition 6. A hypergeometric function F (α, β; x) with α = {a1 , · · · , an }, β = {1, b2 , · · · , bn } is
said to be k-balanced where k is a positive integer, if the following conditions are satisfied
i). x = 1
ii). one P
of the ai ’s P
is a negative integer
iii). k + ni=1 ai = ni=2 bi .
When k = 1, the series is called balanced or Saalchützian.
1.4. Hypergeometric differential equations. See [1, 35] for more details. Each n Fn−1 function
satisfies an order n ordinary Fuchsian differential equation in the variable λ with three regular
singularities at 0, 1, and ∞ [27, §2.1.2][35].
n
X Y (ai )k
Given α = {a1 , · · · , an }, β = {1, b2 · · · , bn }, let F (α, β; λ) = A(k)λk where A(k) = .
(bi )k
k≥0 i=1
d
Let θλ = λ dλ . As θλ (f g) = (θλ f )g + f (θλ g), it is a derivation. In particular, θλ λk = kλk . Then
X X
(θλ + c)( ak λk ) = (k + c)ak λk .
k≥0 k≥0

d
Lemma 1.7. Use D to denote dλ . For any integer k ≥ 1,
1
Dk = (θλ − k + 1) · · · (θλ − 1)θλ .
λk
Proof. Prove by induction. When k = 1, Df = λ1 θλ f . Assume the claim holds for k < n. When
k = n,
 
n−1 1 1
D(D f ) = θλ (θλ − n + 2) · · · (θλ − 1)θλ f
λ λn−1
 
1 1 1
= θλ ( n−1 ) (θλ − n + 2) · · · (θλ − 1)θλ f + n θλ (θλ − n + 2) · · · (θλ − 1)θλ f
λ λ λ
1 1
= (1 − n) n (θλ − n + 2) · · · (θλ − 1)θλ f + n θλ (θλ − n + 2) · · · (θλ − 1)θλ f
λ λ
1
= n (θλ − n + 1)(θλ − n + 2) · · · (θλ − 1)θλ f
λ

Lemma 1.8. For any a ∈ C, λ(θλ + a) = (θλ + a − 1)λ
Proof. Right hand side applies to a function f yields
dλf df df
(θλ + a − 1)λf = λ + (a − 1)λf = λf + λ2 + (a − 1)λf = λ2 + aλf = λ(θλ + a)f.
dλ dλ dλ


Hence λ and θλ + a do not commute.

Lemma 1.9. For any a, b ∈ C, θλ + a and θλ + b commute.
Proof. For constants a, b we have (θλ + a)(θλ + b) = θλ θλ + (a + b)θλ + ab, which is symmetric in a
and b. 
5
 k
P
Lemma 1.10. Given α = {a1 , · · · , an }, β = {1, b2 · · · , bn }, let F (α, β; λ) = k≥0 A(k)λ as before.
Let
Yn Yn
(16) Lα,β;λ := (θλ + bi − 1) − λ (θλ + ai ),
i=1 i=1
then
Lα,β;λ (F (α, β; λ)) = 0.

Proof. From the expression of A(k) in terms of Pochhammer symbol, A(k + 1)/A(k) = ni=1 (a i +k)
Q
(bi +k) .
Thus
X
Lα,β;λ F (α, β; λ) = k(k + b2 − 1) · · · (k + bn − 1)A(k)λk − (k + a1 ) · · · (k + an )A(k)λk+1
k≥0
X
= ((k + 1) · · · (k + bn )A(k + 1) − (k + a1 ) · · · (k + an )A(k)) λk+1 = 0.
k≥0


1.5. Indicial equations and characteristic exponents for singularities. Each differential
operator Lα,β;λ is Fuchsian with only three regular singularities (see [2, Definition 2.2]) at 0, 1, ∞.
 
a1 a2
For instance, 2 F1 ; λ is a solution of the following hypergeometric differential equation
b2
L{a1 ,a2 },{1,b2 };λ F = 0 which is normalized so that the coefficient of D2 is 1, where
b2 − (a1 + a2 + 1)λ a1 a2
(17) L{a1 ,a2 },{1,b2 };λ = D2 + D− .
λ(1 − λ) λ(1 − λ)
Now we recall how to define the local exponent of a normalized Fuchsian differential equation in
variable x of the form L = Dn + an−1 (x)Dn−1 + · · · + a0 (x) where ai (x) are rational functions of x.
Assume x = 0 is a singularity of L and a local solution is of the form f = xr (c0 + c1 x + x2 x2 + · · · ),
where c0 6= 0. Thus
X∞
Df = (k + r)ck xr+k−1 ,
k=0
X∞
D2 f = (k + r − 1)(k + r)ck xr+k−2 ,
k=0
···
∞
X
Dn f = (k + r − n + 1) · · · (k + r − 1)(k + r)ck xr+k−n .
k=0
So Lf = 0 implies that necessarily the combined coefficient of the lowest non-trivial power of x,
which is xr−n , has to be 0. From the above computation we know the coefficient of xr−n in Lf is
c0 times

r(r − 1) · · · (r − n + 1) + Cn−1 r(r − 1) · · · (r − n + 2) + · · · + C0 ,

where Ci = ai (x)xn−i |x=0 . Setting the above to be 0 gives the so called characteristic or indicial
equation of Lf = 0 near the singularity 0. The solutions of this degree-n polynomial of r are called
the characteristic exponents of Lf = 0 near 0. Similarly, near any other singularity a 6= ∞, the
indicial equation is
(18) r(r − 1) · · · (r − n + 1) + Cn−1 r(r − 1) · · · (r − n + 2) + · · · + C0 = 0,
6
where Ci = ai (x)(x − a)n−i |x=a . If ∞ is a singularity, a similar formula is available when the local
uniformizer is chosen as 1/x, which is omitted here.
In other words, the local exponents of Lα,β;λ can be understood as follows:
Let a be a singularity and w be the local parameter around a. The solutions around a take
the form wr f (w), where f (w) is locally holomorphic with f (0) 6= 0. The number r is called an
exponent of the solution at a. If the local exponents {r1 , . . . , rn } of the differential operator at a
are distinct, then a basis of local solutions near a can be given in the following form
wr1 f1 (w), · · · , wrn fn (w)
with fi holomorphic around a. If ei − ej ∈ Z, an additional logarithmic term is allowed.

Example 1.2. For example, if we consider the differential equation (17), its indicial equation at
singularity 0 is
r(r − 1) + b2 r = r(r − 1 + b2 ) = 0,
thus the local exponent near 0 are 0, 1 − b2 . Near 1, the indicial equation is
r(r − 1) + (−b2 + a1 + a2 + 1)r = r(r − b2 + a1 + a2 ) = 0,
so the local exponents near 0 are 0, b2 − a1 − a2 .

Theorem 1.11. The local exponents of Lα,β;λ are

0, 1 − b2 , · · · , 1 − bn at λ = 0
(19) a1 , a 2 , · · · , a n at λ = ∞
0, 1, 2, · · · , n − 2, γ at λ = 1,
where
n
X n
X
(20) γ = −1 + bj − aj
j=1 j=1

See [2, §2] by Beukers and Heckman.

Information of the singular points and their characteristic exponents are often arranged in
columns as below, which is called the Riemann scheme of the differential equation. In the last
column, we indicate which the variable is in used. The Riemann scheme for L{a1 ,a2 },{1,b2 };x is as
follows:
 
0 1 ∞
P 0 0 a1 ; x ,
1 − b2 b2 − a1 − a2 a2
which is used to denote the solutions set of L{a1 ,a2 },{1,b2 };x .
Note that linear fractional transformations permuting 0, 1, ∞ are
1 1 x−1 x
x 7→ x, 1 − x, , , , .
x 1−x x x−1
When x 7→ 1 − x, we have
   
0 1 ∞ 0 1 ∞
P 0 0 a1 ; x = P  0 0 a1 ; 1 − x .
1 − b2 b2 − a1 − a2 a2 b2 − a1 − a2 1 − b2 a2
7
 Similarly
 
0 1 ∞
(21) P  0 0 a1 ; x =
1 − b2 b2 − a1 − a2 a2
   
0 1 ∞ 0 1 ∞
x  −a1  x 
P  0 a1 0 ; x−1 = (1 − x) P 0 0 a1 ; x−1 .
1 − b2 a2 b2 − a1 − a2 1 − b2 a2 − a1 b2 − a2

In the last step, when (1 − x)−a1 is pulled out, it only affects the local exponents at 1 and ∞.

Exercise 1.1. Check that the Riemann Schemes for

1−a 1−b
   
a b 1−c c−a−b
2 F1 ; z , and z (1 − z) 2 F1 ; z
c 2−c

are the same. So they satisfy the same differential operator L{a,b},{1,c},z . If c 6= 1, they are not
scalar multiple of each other.

1.6. Local solutions. Next we consider local solutions to L{a1 ,a2 },{1,b2 };z near the singularities.
(1) Around the singularity z = 0, if b2 6∈ Z, a basis of the solution space can be given by
 
a1 a2
f0 = 2 F1 ; z
b2
1 + a1 − b2 1 + a2 − b2
 
1−b2
g0 = (z) 2 F1 ; z .
2 − b2

When b2 ∈ Z, we consider
 
a1 a2
f0 (c) := 2 F1 ; z
c
1 + a1 − b2 1 + a2 − c
 
1−c
g0 (c) := (z) 2 F1 ; z
2−c

as functions of c. Then a basis of the solution space can be taken by f0 and

g0 − f0
h0 = lim .
c→b2 c − b2
For example, when b2 = 1, f0 and g0 are two linearly independent solutions when c near
1. The function h0
g0 − f0
h0 = lim
c→1 c−1
d 1−c
= z F ({1 + a1 − c, 1 + a2 − c}, {2 − c}; z) |c=1
dc
∞
"m  #
X (a1 )m (a2 )m m X 1 1 2
= log z · f0 + z + −
m!2 a1 + k − 1 a2 + k − 1 k
m=1 k=1

gives a solution, which is linearly independent to f0 .

8
 (2) Around the singularity z = 1, if a1 + a2 − b2 ∈ / Z, a basis of the solution space can be taken
by
 
a1 a2
f1 := 2 F1 ; 1−z
1 + a1 + a2 − b2
b2 − a1 b2 − a2
 
b2 −a1 −a2
g1 := (1 − z) 2 F1 ; 1−z .
1 + b2 − a1 − a1
To obtain such a basis, one can consider the differential equation under the transformation
z 7→ 1 − z. This converts the differential equation given by (17) to the equation
 
2 a1 + a2 + 1 − b2 − (a1 + a2 + 1)z a1 a2
D + D− F =0
z(1 − z) z(1 − z)
(3) Around z = ∞, and a1 − a2 ∈
/ Z, a basis of the solution space can be taken by
a1 1 + a1 − b2
 
−a1
f∞ := z 2 F1 ; 1/z
1 + a1 − a2
a2 1 + a2 − b2
 
−a2
g∞ := z 2 F1 ; 1/z .
1 + a2 − a1
Similarly, we can set z = 1/t and transform the original equation
 
1
(a1 + θz )(a2 + θz ) − (1 + θz )(b2 + θz ) F =0
z
to
((a1 − θt )(a2 − θt ) − (1 − θt )(b2 − θt )t) F = 0.
By the fact that
((a1 − θt )(a2 − θt ) − (1 − θt )(b2 − θt )t) ta
 
a+1 1
=t (θt + a + 1 − b2 )(θt + a) − (θt + a + 1 − a1 )(θt + a + 1 − aa ) ,
t
we can choose
1 + a1 − b2
 
a1 a1
t 2 F1 ; t
a1 − a2 + 1
1 + aa − b2
 
a2 a2
t 2 F1 ; t
a2 − a1 + 1
as a basis for the solution space around t = 0, i.e, z = ∞.
For general hypergeometric differential equations, the local solutions spaces can be described in
a similar way.
When no bj , j = 2, · · · , n is an integer, and no two bj differ by an integer, a fundamental set of
solutions of Lα,β;λ F = 0 is given by
F (α, β; z)
z 1−bk F ({1 + ai − bk }, {1 + bj − bk }; z), k = 2, · · · , n.
When no two ai differ by an integer, a fundamental set of solutions around z = ∞ is given by
1
f∞,k (z) := (−z)−ak F ({1 − bj + ak }, {1 − ai + ak }; ), k = 1..n,
z
9
and the relation between the solutions are
 
 
n n n
X Y Γ(a j − a )
k 
 Y Γ(b j )
F (α, β; z) =   f∞,k (z).
 Γ(aj )  Γ(bj − ak )
k=1 j=1 j=1
j6=k

See [10, (16.8.8)].

If one of bj , j = 2, · · · , n is an integer, one can find solutions through the continuity of the
hypergeometric functions as functions of the parameters bj .

1.7. Monodromy representations and Beukers-Heckman theorem. Please see [2] for full
details.
Given an order n ordinary Fuchsian differential equation L with only regular singularities x1 , · · · , xs ,
we consider the monodromy representation of the fundamental group π1 (CP 1 \ {x1 , · · · , xs }, x0 )
where x0 is any ordinary point as follows. By Cauchy, near x0 , the solution space of the homoge-
neous equation Lu = 0 is an n-dimensional vector space V (x0 ) over C. We fix a basis f1 , · · · , fn
of V (x0 ). Let L(t) : [0, 1] → CP 1 \ {x1 , · · · , xs } be a continuous function with L(0) = L(1) = x0 .
Its image is topologically a closed loop. We extend f1 , · · · , fn analytically along L when t varies
from 0 to 1. By Frobenius’ result, f1 (L(1)), · · · , fn (L(1)) form another basis of V (x0 ). Thus it can
be written as M (L)(f1 , · · · , fn )T where M (L) ∈ GLn (C) only depending on the class [L] of L in
π1 (CP 1 \ {x1 , · · · , xs }, x0 ). The map from [L] 7→ M (L) is a homomorphism, which is unique up to
conjugation by GLn (C). It is called a monodromy representation of L.
Let aj = e2πiaj , bj = e2πibj . They are roots of unity lie on the unit circle in C, under our
assumptions on α = {a1 , · · · , an } and β = {b1 , · · · , bn } with ai , bj ∈ Q.
Definition 7. Suppose 0 ≤ a1 ≤ · · · an < 1, 0 ≤ b1 ≤ · · · bn ≤ 1. We say the sets a1 =
e2πia1 , · · · , an = e2πian and b1 = e2πib1 , · · · , bn = e2πibn interlace on the unit circle if and only
if either
a1 < b1 < a2 < b2 < · · · < an < bn or b1 < a1 < · · · < bn < an .
If α, β is a primitive pair, then the monodromy group is an irreducible subgroup of GLn (C), local
2πiγ ,
monodromy matrix PnM1 nearP1nis a reflection (i.e. M1 − id has rank 1), with determinant c = e
where γ = −1 + j=1 bj − j=1 aj in equation (20).
Theorem 1.12 (Levelt, see Theorem 3.5 [2]). Let Aj , Bk be defined by
n
Y n
Y
n n−1
(22) (X − aj ) = X + A1 X + · · · + An , (X − bj ) = X n + B1 X n−1 + · · · + Bn ,
j=1 j=1

then
   −1
0 0 ··· 0 −An 0 0 ··· 0 −Bn
1 0 ··· 0 −An−1  1 0 ··· 0 −Bn−1 
−1
M0−1
   
M∞ 0
= 1 ··· 0 −An−2 
, 0
M0 =  1 ··· 0 −Bn−2 
 , M1 = M∞
 ···   ··· 
0 0 ··· 1 −A1 0 0 ··· 1 −B1
generate a hypergeometric group which is conjugate inside GLn (C) to the monodromy group of the
hypergeometric differential equation with parameters α and β.
Denote this group by H({a1 , · · · , an }, {b1 , · · · , bn }). When α and β form a primitive pair, then
H({a1 , · · · , an }, {b1 , · · · , bn }) is irreducible.
10
Remark 1. Due to a result of Pochhamer (see Proposition 2.8 [2]), the local monodromy matrix
M1 is called a (quasi)reflection as it satisfies a special property that M1 − nIn has only rank 1,
where In stands for the rank-n identity matrix.
Example 1.3. For α= { 12 , 21 },β  the two polynomials are X 2 + 2X + 1 and X 2 − 2X + 1,
= {1, 1},
0 −1 0 −1
so H{−1,−1},{1,1} = , . Note that
1 −2 1 2
   −1
1 1 1 1
H{−1,−1},{1,1} = Γ(2)
0 2 0 2
the principal level-2 congruence subgroup.
The above theorem of Levelt is a rigidity theorem. When n = 2, see Theorem 2.3.1 of [1] due to
Papperitz. Equivalently, it can be stated as
Theorem 1.13. Every second Fuchsian equation with 3 regular singularities can be transformed to
a hypereometric differential equation.
More generally,
Theorem 1.14 (Rigidity Theorem). Each order-n ordinary differential equation in variable z which
has only three regular singularities at 0, 1, ∞ and the corresponding indicial exponents as (19) is
equivalent to Lα,β;z F = 0.
a 1−a a 1−a
   
Exercise 1.2. Assume a ∈ / Z. Show that 2 F1 ; x and 2 F1 ; 1 − x are two
1 1
linear independent solutions of L{a,1−a},{1,1};x .
In [2] Beukers and Heckman gave explicit descriptions for when H({a1 , · · · , an }, {b1 , · · · , bn }) is
finite.
Theorem 1.15 (Beukers and Heckman). Let M = lcd(α ∪ β) be the least positive common denom-
inators of ai , bj ’s. The hypergeometric group H({a1 , · · · , an }; {b1 , · · · , bn }) is finite if and only if
for each k ∈ N coprime to M , the sets {ak1 , · · · , akn } and {bk1 , · · · , bkn } interlace on the unit circle.
Remark 2. When both α, β are defined over Q, we only need to check whether {a1 , · · · , an } and
{b1 , · · · , bn } interlace.
Example 1.4. According to the above theorem, the hypergeometric group for the multi-sets α =
{ 21 , 16 , 56 }, β = {1, 31 , 23 } is finite; while hypergeometric group for the multi-sets α = { 12 , 13 , 23 }, β =
{1, 16 , 56 } is not.
We plot the positions of a1 , · · · , a3 (in green color) and b1 , · · · , b3 (in red color) on the unit circle
for these two cases respectively.
When the hypergeometric group is infinite, they have the following result.
Theorem 1.16 (Beukers and Heckman). Let H({a1 , · · · , an }, {b1 , · · · , bn }) be an infinite primitive
hypergeometric group which is not a scalar shift of a finite group. Let H({a1 , · · · , an }, {b1 , · · · , bn })
be its Zariski closure. Then there are two possibilities,
I) There exists d ∈ C× such that
{da1 , · · · , dan } = {(da1 )−1 , · · · , (dan )−1 }, and {db1 , · · · , dbn } = {(db1 )−1 , · · · , (dbn )−1 }.
Ia) If c = e2πiγ = 1 where γ as in (20), then
H({da1 , · · · , dan }, {db1 , · · · , dbn }) = Spn (C);
11
 Figure 1. {α = { 21 , 61 , 56 }, β = {1, 13 , 23 }} vs {α = { 12 , 13 , 23 }, β = {1, 16 , 56 }}

Ib) if c = −1 then
H({da1 , · · · , dan }, {db1 , · · · , dbn }) = On (C).
II) For the remaining cases, SLn (C) ⊂ H({a1 , · · · , an }, {b1 , · · · , bn }).
Example 1.5. For α = { 21 , 12 }, β = {1, 1} as in Example 1.3, d = 1 and c = 1. In this case
H({−1, −1}; {1, 1}) = Sp2 (C). For α = { 12 , 13 , 23 }, β = {1, 16 , 56 }, d = 1, c = −1, the Zariski closure
of its monodromy group is isomorphic to O3 (C).
1.8. Schwarz theorem, triangle groups and arithmetic triangle groups. When α = {a, b}, β =
{1, c}, the following results are useful. There is an explicit correspondence between a hypergeomet-
ric differential equation Lα,β and a Schwarz triangle ∆(p, q, r) with p, q, r ∈ Q due to the following
theorem of Schwarz. See [3] by Beukers and [36] by Yoshida for more details.
Let H = {z ∈ C : Im z > 0} be the complex upper half-plane and P1 = C b or CP1 .

Theorem 1.17 (Schwarz, (see [35])). Fix a z0 ∈ H, let f, g be two independent solutions to the
differential equation L{a,b},{1,c};z F = 0 near at z0 , and let p = |1 − c|, q = |c − a − b|, and r = |a − b|.
If p, q, r < 1, then the Schwarz map
D: H ∪ R −→ P1 , D(z) = f (z)/g(z)
gives a bijection from H ∪ R onto a curvilinear triangle with vertices D(0), D(1), D(∞) and
corresponding angles pπ, qπ, rπ, as illustrated below.

D(0) pπ
H

D
rπ

D(∞)
D(1) qπ

The pictures above are from [12].

Remark 3. (1) By Cauchy’s fundamental theorem, the solutions f and g do not vanish si-
multaneously (at almost all points). In addition, the map D(z) is locally bijection for any
z ∈ H.
(2) A change of the basis {f,g} corresponds to a fractional linear transformation which does not
change the angles of the curvilinear triangle.
12
 (3) The image of the map D(z) at z = 1 can be determined by the Gauss evaluation formula,
see 29 below.
Note that a Schwarz triangle with angles pπ, qπ, and rπ as described in Theorem 1.17 can be
used to tile the sphere (P1 ), the Euclidean plane (C), or the hyperbolic plane (H) through reflections
along its edges, depending on whether p + q + r is equal to, greater than, or less than 1, respectively.
Therefore, each Schwarz triangle ∆(p, q, r) can be associated to the symmetry group of this tiling,
which we denote by S∆ (p, q, r).
From the Schwarz’ reflection principle [36, Proposition 7.1 and Corollary 7.2], we will see that
every element of the projective monodromy group is a product of an even number of reflections in
the edges of the curvilinear triangle. The idea is as follows: Let D be a Schwarz map.
(1) For a fixed z0 ∈ H, take a loop γ starting at z0 passing the interval (0, 1) into the lower
half-plane H := {z ∈ C : Im(z) < 0} and returning to z0 through (1, ∞). Let `∞ be the
edge connecting D(0) and D(1) and `0 be the edge connecting D(∞) and D(1). The image
of γ is a path in P1 starting at D(z0 ) in the triangle T , passing through `∞ into the mirror
image of T , say T∞ , and then passing through `0 into the mirror image of T∞ .

(Source: Beukers’ Note [3])

(2) Then new Schwarz map γ · D (analytic continuation along γ) is
 
aD + b a b
γ·D = , for some ∈ GL2 (C).
cD + d c d
In the same vein, D(z) can be extended to H through any interval ` of (−∞, 0), (0, 1), (1, ∞) by
D(z) := D(z).
Then the extended D is holomorphic on H ∪ ` ∪ H and Image(D) = D(H) ∪ D(`) ∪ D(H). If D(`)
is part of a circle C, then D(H) is the mirror image of D(H) with respect to C. That is,
 
D(H) = g −1 · g · D(H) ,

where g ∈ GL2 (C) and g(C) = R. If we make analytic continuations from H to H and from H
to H (i.e. make an even number of reflections), we get a linear fractional transformation. Such
transformations form the projective monodromy group Γ.
Let S be either H, C, or P1 equipped with the hyperbolic, euclidean and spherical metric respec-
tively. For a fixed geodesic triangle ∆, denote
W (∆) := the group generated by the reflections in the edges of ∆.
[
Theorem 1.18 ([3]). For any geodesic triangle ∆, we have S = γ · ∆, where ∆ is the
γ∈W (∆)
closure of ∆ in S.
13
 The picture below shows H can be tessellated by the standard fundamental domain of SL2 (Z).
In general, the (open) triangles may overlap, so we consider special triangles, called elementary

Figure 2. A partial tessellation of H, photo courtesy of Bao Pham

triangles which give tessellation of S as well.

Definition 8. An elementary triangle is a geodesic triangle whose vertex angles are all of the form
π/n, n = 2, . . ., ∞.
Theorem 1.19 ([3]). Let ∆ be an elementary triangle. Then, for any γ ∈ W (∆), γ 6= ±1, we have
γ∆ ∩ ∆ = ∅.
For any geodesic triangle ∆, if W (∆) acts on S discretely, we can find an elementary triangle ∆et
such that W (∆et ) = W (∆).
For example, for the triangle ∆ with internal angles (2π/5, 2π/5, 2π/5), the associated ∆et has
angles (π/2, π/3, π/5).
We now assume that |1 − c| = p = 1/e1 , |c − a − b| = q = 1/e2 , and |a − b| = r = 1/e3 with
ei = 2, . . ., ∞. Let τp , τq , τr be reflections in the edges of the Schwarz triangle ∆(p, q, r).

Set gp = τq τr , gq = τp τr , gr = τp τq . Then the triangle group

Γ = (e1 , e2 , e3 ) := hgp , gq , gr : gpe1 = gqe2 = gre3 = gp gq gr = 1i
is a subgroup of Isom+ (S). The quotient space X(Γ) := Γ\S is a Riemann orbifold of genus zero
with #{∞ ∈ {e1 , e2 , e3 }} punctures.
There are three categories of triangle groups:
14
 (1) p + q + r > 1, there are only four types:
(2, 2, n), (2, 3, 3), (2, 3, 4), (2, 3, 5).
The corresponding hypergeometric functions are algebraic. For instance,
a 21 + a √ √
 
1
(1 + z)−2a + (1 − z)−2a .


2 F1 1 ; z =
2
2
(2) p + q + r = 1, there are only four cases:
(2, 2, ∞), (2, 3, 6), (2, 4, 4), (3, 3, 3).
(3) p + q + r < 1, there are infinitely many cases.

Example 1.6.
(2, 3, ∞) ' PSL2 (Z), (∞, ∞, ∞) ' Γ(2), (2, 3, 3) ' A4 .
When p + q + r < 1, we say that Γ = (e1 , e2 , e3 ) is arithmetic if Γ is commensurable with norm 1
group of certain quaternion order. There are exactly 85 arithmetic triangle groups falling into 19
classes (see the works of Takeuchi [30, 29]). The compactifications of their corresponding quotient
spaces X(Γ) are Shimura curves, which parametrize certain abelian surfaces with quaternionic
multiplication. In the work of [30], Takeuchi gives the precise quaternion algebras and orders.
From which one can realize the arithmetic triangle groups as subgroups of SL2 (R). Alternatively,
there is an explicit embedding given by Petersson [23, 6] (we will describe this embedding in a later
discussion).
Example 1.7 ([28]). Take (2, 3, ∞) ' PSL2 (Z) for example. Let J = j/1728 be the hauptmodul for
PSL2 (Z) such that J(i) = 1, J(ρ) = 0, and has a pole at cusp i∞, where ρ = e2πi/6 .
Denote
1 1  5 5 
−1/6 1/4 12 12 1/6 1/4 12 12
f (J) = J (1 − J) 2 F1 2 ; J , g(J) = J (1 − J) 2 F1 4 ; J ,
3 3
and define
ρf (J) + Cg(J)
D(J) = √ ,
f (J) − C(2 + 3)g(J)
where  1 1 
12 12 √
2 F1 2 ; 1 11 2

2


1i−ρ 3 2− 3 Γ 12 Γ 3
C= · ζ12 · 5 5 = i·
.
2i−ρ 2 7 2

4
12 12 Γ Γ
2 F1 4 ; 1 12 3
3
Slit the disk |J| < 1 along the negative real axis and use the principal branch of log J to define
J ±1/6 so that it takes positive real values on (0, 1). In a neighborhood of J = 0, |J| < 1, the
function D maps (0, 1) to the geodesic running from i to ρ. If we continue slitting the J-plane
along the negative real axis and along the positive real axis from 1 to ∞, we can continue D(J)
to a single-valued function which maps the slit plane onto the fundamental domain for PSL2 (Z) in
Figure 3.
Similarly, we can find a suitable basis to describe the Schwarz map as an inverse to the J-function
through the standard fundamental domain for PSL2 (Z). Set t = 1/J and denote
1 5   
1/4 12 12 1
f (t) = (1 − t) 2 F1 ; t , g(t) = f (t) log t + h(t) ,
1 2πi
15
 Figure 3. A fundamental domain for PSL2 (Z), photo courtesy of Bao Pham

where h(t) is a holomorphic function and vanishes at t = 0. We now choose the Schwarz map to be
g(t) 1
D(t) = − log 1728.
f (t) 2πi
Then
lim D(t) = ∞, lim D(t) = i,
t→0 t→1
the function D maps (0, 1) to the geodesic running from i∞ to i. This Schwarz map D(t) is an
inverse in a neighborhood of t = 0 to the J-function on H.

1.9. The Legendre curves.

Theorem 1.20 ([36, 28]). The Schwarz map
1 1 
2 2
; 1−λ
2 F1
1
D(λ) := i 1 1 
2 2
2 F1 ; λ
1
16
 of L{1/2,1/2};{1,1} gives an isomorphism

P1 − {0, 1, ∞} −→ Γ(2)\H,
which is the inverse map of the modular λ-function, which parameterizes isomorphism classes of
elliptic curves with level-2 structures.
Similarly, we can choose a suitable basis {f, g} of the L{1/12,5/12};{1,1} so that the corresponding
Schwarz map gives an isomorphism
P1 − {∞} −→ PSL2 (Z)\H,
which is the inverse map of the elliptic j-function.
We now elaborate some discussion behind Theorem 1.20 in which α = { 12 , 12 }, β = {1, 1}. It is
well-known that elliptic curves with level-2 structures can be written as the Legendre curves
(23) Lλ : y 2 = x(1 − x)(1 − λx).
When λ 6= 0, 1 it is an elliptic curve. It is a double over of CP 1 which ramifies only at P1 = 0, P2 =
1, P3 = λ1 , P4 = ∞ as demonstrated by the picture below. Going from right to left, first cut the
torus twice including half of each of the indicated boundaries on each torus, to get two cylinders.
Each cylinder can be realized as the sphere on the left by pinching the ends together. Gluing along
the slits gives the double cover: i.e. π : Lλ → CP 1 as a degree-2 ramified cover.

1 1
λ ∞ 0 1
λ 1 1
0 λ ∞ 0 1
1
λ ∞ 0 1

Pictures above are from [12].

For given λ, it has a unique up to scalar holomorphic differential 1-form
dx
ωλ := p .
x(1 − x)(1 − λx)
With ωλ , one can compute periods of Eλ as follows. Its first homology group H1 (Lλ , Z) is a
rank-2 Z-module with two generators, say H1 (Lλ , Z) = γ1 Z ⊕ γ2 Z, one of the generators, say γ1 can
be chosen as the blue circle γ01 , which is homotopic to γ1/λ∞ . The other generator can be chosen
as γ0∞ . Namely a set of generators for H1 (Lλ , Z) can be assemble from π −1 (γPi Pj ) where the γPi Pj
are paths on CP 1 connecting branched points Pi , Pj where 1 ≤ i < j ≤ 4.
The periods of Lλ is the lattice
Z Z
Λ(λ) = Z ωλ ⊕ Z ωλ .
γ1 γ2
 0   
γ1 a b γ1
If a different basis {γ10 , γ20 } of H1 (Lλ , Z) is chosen, then = ,
γ20 c d γ2
Z Z  
d c dτ + c
τ0 = ωλ / ωλ = ·τ = ,
0
γ2 0
γ1 b a bτ +a
17
  
d c
where ∈ Γ(2). Upon relabelling, we can assume
b a
R
γ ωλ
τ=R2 ∈H
γ1 ωλ
and its isomorphism class is determined by the value of the modular λ(τ ) evaluated at τ , which is
invariant under linear transformation by elements in Γ(2).
Question 1. How likely is the period ratio τ ∈ Q?
The answer is not very likely, unless Lλ admits complex multiplication (CM), which is a funda-
mental result by Schneider, see [24].
Theorem 1.21 (Chowla and Selberg). If E is an elliptic
√ curve whose endomorphism ring over C
is an order of an imaginary quadratic field K = Q( −d) with fundamental discriminant −d, then
all periods of E are algebraic multiples of a particular transcendental number
  Y
1  a  n(a)
4hK
(24) ω−d := Γ Γ ,
2 d
0<a<d
where Γ (·) stands for the Gamma function, n is the number of torsion elements in K,  is the
primitive quadratic Dirichlet character modulo d, that is, the quadratic character attached to K
over Q, and hK is the class number of K
See [26] by Selberg and Chowla, [14] by Gross, or [37, (97)].

Γ( 13 ) 3/2
 

1
1 Γ( 4 ) 1
Exercise 1.3. Verify that ω−4 = Γ 2 Γ 3 and ω−3 = Γ 2 Γ( 23 )
.
(4)
One take-way from this theorem is that ratios alike might come from CM background.
Exercise 1.4. Compute ω−d for d = 7, 8, 11, 19, 43, 67, 163.

We now return to the discussion of periods.

Question 2. How is the above picture related to the differential equation L{1/2,1/2},{1,1}λ F = 0?
Z
For any γ ∈ H1 (Lλ , Z), define pγ (λ) := ωλ . If we vary γ analytically, pγ (λ) becomes a function
γ Z Z
of λ. This means by analytic continuation, τ (λ) = ωλ / ωλ is also considered as function of
γ2 γ1
λ. The differential equation L{1/2,1/2},{1,1}λ F = 0 is called the Picard-Fuchs equation of Lλ , by
which it means L{1/2,1/2};{1,1} ωλ is an exact form on Lλ .
Exercise 1.5. Verify that L{1/2,1/2};{1,1} ωλ is an exact form on Lλ .
Z Z
It follows that both ωλ and ωλ are solutions of L{1/2,1/2},{1,1};λ F = 0, which are linearly
γ1 γ2
independent as functions of λ. R
For example if we compute γ01 ωλ , up to a scalar, by Euler integral formula, it agrees with
1 1  1 1 
R1 2 2 2 2
2 0 ωλ = 2 · 2 P 1 ; λ = 2π · 2 F1 ; λ . The other linearly independent solution is
1 1
1 1 
F 2 2 ; 1 − λ (see Exercise 1.2). This explains the claim in Theorem 1.20 that the Schwarz
2 1
1
map is the inverse of λ-function.
18
 In the literature, there is another model of the Legendre curve which is more convenient for
elliptic integrals. Given k ∈ C r {0, ±1}, let
L(k) : y 2 = (1 − x2 )(1 − k 2 x2 )
which is now a double cover of CP 1 ramified at ±1, ±1/k. Similar to the above, one can perform
branch cuts along two non-intersecting lines between 1 and −1 as well as between 1/k and −1/k.
The 1st de Rham cohomology HDR 1 (L(k)/C) is isomorphic to the differentials on L(k) with at most

a double pole at infinity (see [15, Appendix 1] by Katz). It is a 2-dimensional vector space over C
generated by w1 (k) = dx 2 2 dx
y and w2 (k) = (1 − k x ) y . Here w1 (k) is holomorphic. Typically there is
1 (L(k)/C) unless L(k) admits complex
no a priori ‘functorial’ choice for the second generator of HDR
multiplication (CM), see [15]. Like the above model, a loopRγ on L(k) can be obtained from the
path from −1 to 1 wrapping around twice. The two periods L wi (k), i = 1, 2 can be computed by
Z 1 1 1 
π
K(k) = w1 (k) = 2 F1 2 2 ; k 2
0 2 1
and  1
1 1 
−2
Z
π 2 2
E(k) = w2 (k) = 2 F1 ; k ,
0 2 1
which are complete elliptic integrals of first and second kind respectively, see [4, Chapter 1] by
Borwein-Borwein.
Exercise 1.6. Verify that
d
E(k) = k(1 − k 2 ) K(k) + (1 − k 2 )K(k).
dk
They satisfy the Legendre relation (Theorem 1.6 [4])

π p
(25) E(k)K(k 0 ) + E(k 0 )K(k) − K(k)K(k 0 ) =
, k0 = 1 − k2 .
2
Remark 4. The Legendre relation is an expression for the Wronskian of the differential equation
L{ 1 , 1 },{1,1};k2 .
2 2

When k is a singular moduli value, namely L(k) admits CM, using the Clausen formula (see (28)
below) and based on the Chowla and Selberg theorem 1.21, Borwein-Borwein gave a method to
prove Ramanujan formulas for 1/π in [4]. Similar method was used by Chudnovsky-Chudnovsky [5],
see [38] by Zudilin or [12, §7.2] for reviews of Ramanujan’s formulas for 1/π. One of Ramanujan’s
1/π formulas is
∞
X ( 1 )3 2 k 1 4
(26) (6k + 1) = .
k!3 4k π
k=0

1.10. Hypergeometric formulas. To obtain hypergeometric formulas, there are a variety of ways.
Here we only summarize only a few that are relevant to our later discussion. Note that we omit
the convergence conditions here. Careful readers can check them out from [1].

1.10.1. From rigidity. For instance, as a consequence of (21) plus the uniqueness of the local holo-
morphic solution near 0 with exponent 0, one has the following identify due to Pfaff
a c−b
   
a b −a x
(27) 2 F1 ; x = (1 − x) 2 F1 ; .
c c x−1
19
Another example is the Clausen formula

 2  
a b 2a 2b a+b
(28) 2 F1 1 ; z = 3 F2 1 ; z .
a+b+ 2 2a + 2b a + b + 2

Note again both hand sides take value 1 at z = 0. The scheme for the 2 F1 on the left is
 
0 1 ∞
P 0 0 a ; z
1 1
2 −a−b 2 b

Its symmetric square is

 
0 1 ∞
 0 0 2a ; z 
P 1
 −a−b 1
,
2 2b 
1
1 − 2a − 2b 2 a + b

which coincides with the Riemann scheme of the right hand side.
This technique is also known as “pull-back transformation between hypergeometric differential
equations”, see [31] by Vidunas. A general pull-back transformation of higher degree converts a
hypergeometric differential equation to a Fuchsian equation with several singularities. A pull-back
of second order hypergeometric equations to a Fuchsian equation with three singularities gives rise
to an algebraic transformation of 2 F1 -functions.

Example 1.8. The identities of the local systems

   
0 1 ∞ 0 −1 1 ∞
a 2
P  0 0 2 ; z  =P  0 0 0 a ; z
1 a+1
− b b − a 1 − 2b b − a b − a a + 1
2 2
  
0 1 ∞ 2  a 0 1 ∞
2z  2z 2z 
=P  0 0 0 a ; z+1 = 2 − P  0 0 a ; z+1
z+1
1 − 2b b − a b − a a + 1 1 − 2b b−a b

give one of the Kummer quadratic formulas:

a a+1   
a b 2z
(1 + z)a 2 F1 2
1
2 ; z 2 = 2 F1 ;
2 + b 2b z + 1

equivalently,
" 2 #
   a a+1 
a b z −a 2 2 z
2 F1 ; z = 1− F
2 1 1 ; .
2b 2 2 + b 2−z

To double check we see that as formal power series near z = 0, both hand sides start with constant
1.
20
 Similarly, we have
   
0 1 ∞ −1 1 ∞
2 0 0 2a ; z 
P 0
 0 a ; z  =P 
1 1 1 1
−a−b b − a−b − a − b 2b
2 2
2 2

0 1 ∞
1+z 
=P  0 0 2a ; 2
1 1
−a−b − a − b 2b
2 2

0 1 ∞
1−z 
=P 0 0 2a ; 2 ,
1 1
2 −a−b 2 − a − b 2b
and hence there are constants C1 , C2 such that
   √   √ 
a b 2a 2b 1− z 2a 2b 1+ z
2 F1 1 ; z = C1 · 2 F 1 1 ; + C2 · 2 F1 ; .
2 a+b+ 2 2 a + b + 12 2
√ √
   
2a 2b 1− z 2a 2b 1+ z
Exercise 1.7. 1) Can 2 F1 1 ; 2 and 2 F1 ; 2 be scalar multi-
a+b+ 2 a + b + 12
ple of each other?
2) What information would be helpful in order to reach the following conclusion
1 1
!    √   √ 
2, a + b + 2 a b 2a 2b 1− z 2a 2b 1+ z
2Γ 1 1 2 F1 1 ; z = 2 F1 1 ; + 2 F1 1 ; ,
a + 2, b + 2 2 a+b+ 2 2 a+b+ 2 2
 
where Γ ab11 ,··· ,ar
,··· ,bs = Γ(a 1 )···Γ(ar )
Γ(b1 )···Γ(bs ) ?

See the works of Goursat, Vidūnas [13, 31, 32, etc.] for more examples and details.
1.10.2. From definition. Euler integral formula immediately implies the Gauss evaluation formula
(29)   Z 1  
a b 1 −a b−1 c−b−1 B(b, c − a − b) c, c − a − b
2 F1 ; 1 = (1 − t) t (1 − t) dt = =Γ .
c B(b, c − b) 0 B(b, c − b) c − a, c − b
Another way to prove the Pfaff formula (27) is to use Euler’s integral formula and then change
variable which we recall here (following the proof of Theorem 2.2.5 of [1]).
  Z 1
a b 1
2 F1 ; x = (1 − xt)−a tb−1 (1 − t)c−b−1 dt
c B(b, c − b) 0
Z 1
t7→1−s 1
= (1 − x + xs)−a (1 − s)b−1 sc−b−1 ds
B(b, c − b) 0
(1 − x)−a
Z 1
xs −a c−b−1
= (1 − ) s (1 − s)b−1 ds
B(b, c − b) 0 x−1
a c−b
 
−a x
= (1 − x) 2 F1 ; .
c x−1
The Euler formula below is a consequence of the Pfaff formula iterated twice.
c−a c−b
   
a b c−a−b
(30) 2 F1 ; x = (1 − x) 2 F1 ; x .
c c
Exercise 1.8. Derive (30) from (27).
21
1.10.3. From comparing coefficients. Obtaining evaluating formulas from comparing coefficients of
known identities is commonly used. For example the Pfaff-Saalschütz formula [1, Thm. 2.2.6]
follows from comparing coefficients on both sides of (30). It states that for a positive integer n and
a, b, c ∈ C,

(31)
−n
   
a b (c − a)n (c − b)n c − a + n, c − b + n, c, c − a − b
3 F2 ; 1 = =Γ .
c 1+a+b−n−c (c)n (c − a − b)n c − a, c − b, c + n, c − a − b + n

By Definition 6, the left hand side is a balanced series. In some sense, transformation formula like
(30) and evaluation formula (31) are two equivalent ways to convey the same information.

Exercise 1.9. Assume (31), show that (30) holds as an equality between two formal power series.

In Euler transformation (30), if we use variable d, e, f in place of a, b, c, namely

f −d f −e
   
f −d−e d e
(1 − x) 2 F1 ; x = 2 F1 ; x
f f

and assume c − a − b = f − d − e, then

f −d f −e c−a c−b
       
a b d e
2 F1 ; x 2 F1 ; x = 2 F1 ; x 2 F1 ; x
c f c f

From comparing coefficients and relabelling the variables, one gets the following formula relating
two balanced 4 F3 series
−n a b c −n a d−b d−c
   
(e − a)n (f − a)n
(32) 4 F3 ; 1 = 4 F3 ; 1 ,
d e f (e)n (f )n d a+1−n−e a+1−n−f

where a + b + c − n + 1 = d + e + f. See Theorem 3.3.3 [1].

(c)k
Let f → ∞ while keeping f − c fixed. This means (f )k → 1 for any k ≤ n, which is a fixed finite
−n a b c −n a b
   
positive integer. Under this assumption, 4 F3 ; 1 7→ 3 F2 ; 1 . Dealing
d e f d e
with the right hand side similarly leads to

−n a b −n a d−b
   
(e − a)n
(33) F
3 2 ; 1 = F
3 2 ; 1 .
d e (e)n d a+1−n−e

Long, Osburn and Swisher used (33) and the Pfaff-Saalschütz formula (31) to prove a conjecture
of Kimoto and Wakayam, see [19] for more information.

Exercise 1.10. This is related to (32). Check that

ι : (−n, a, b, c ; d, e, f ) 7→ (−n, a, d − a, d − c ; d, a + 1 − n − e, a + 1 − n, f )

is an involution on the 7-tuples of parameters in which the upper parameters and lower parameters
are separated by “;”. In addition, one can permute the upper (resp. lower) parameters. Say you
permute a, b on the left and then apply ι again, what will you get? Do you get a new 4 F3 (1) series?

See [33] by Whipple for relations among 3 F2 (1) series and [11] by Formichella, Green and Stade
for Coxeter group actions on balanced 4 F3 (1) series.
22
1.10.4. From local expansion agreements. The idea behind the method is very straightforward,
namely two formal power series are identical if their coefficients are the same.
We will illustrate how to use it to prove the following Kummer quadratic transformation formula
 1+c c   
−c 2 −b 2 −4x b c
(34) (1 − x) 2 F1 ; = 2 F1 ; x .
c − b + 1 (1 − x)2 c−b+1
We now outline a proof using following [12] by noting
(a)n
(35) (a)n−r = (−1)r .
(1 − a − n)r

Proof. To begin, note that

 
−c − 2k (c + 2k)n−k (35) (c + 2k)n (−n)k
(36) = (−1)n−k = (−1)n−k
n−k (1)n−k (1 − c − 2k − n)k (1)n
Γ(c + 2k + n)Γ(1 − c − 2k − n) (−n)k
= (−1)n−k
Γ(c + 2k)Γ(1 − c − k − n) n!
reflection Γ(c + 2k + n)Γ(c + k + n) (−n) k (c)n+k (−n)k
= (−1)n = (−1)n
Γ(c + 2k)Γ(c + 2k + n) n! (c)2k n!
(c)n (c + n)k (−n)k (6) (c)n (c + n)k (−n)k
= (−1)n = (−1)n k c c+1 .
(c)2k n! 4 ( 2 )k ( 2 )k n!
The left hand side of (34) can be expanded as
X ( 1+c − b)k ( c )k
2 2
(−4x)k (1 − x)−c−2k
k!(c − b + 1)k
k≥0
X ( 1+c − b)k ( c )k X −c − 2k 
k
= 2 2
(−4x) (−x)i
k!(c − b + 1)k i
k≥0 i≥0
1+c c
2 − b)k ( 2 )k k −c − 2k
X (  
n=k+i
= 4 (−x)n
k!(c − b + 1)k n−k
k,n≥0

(36) X ( 1+c − b)k ( c )k (c)n (c + n)k (−n)k

= 2 2
4k k c c+1 xn
k!(c − b + 1)k 4 ( 2 )k ( 2 ) k n!
k,n≥0
 1+c
−b c+n −n
X (c)n 
= 3 F2
2
c+1 ; 1 xn .
n≥0
n! 2 c − b + 1

By the Pfaff-Saalschütz formula (31), the above equals

X (c)n (b)n ( 1−c − n)n reflection
X (b)n (c)n
1+c
2
xn = xn .
n!( 2 ) n (b − c − n)n n!(c − b + 1)n
n≥0 n≥0


Exercise 1.11. Prove the following Bailey cubic transformation formula:
a 1−a
   a a+1 
3 −a 3 3 −27x
(37) 2 F1 4a+5 ; x = (1 − 4x) 2 F1 4a+5 ; (1 − 4x)3 .
6 6
23
1.10.5. From specializing values in known identities. For example, if we let x = −1 in (34), then
1+c c 
− b
the left hand side becomes 2−c 2 F1 2 2 ; 1 which can be evaluated by the Gauss
c−b+1
evaluation formula (29). As a consequence, it leads to the Kummer’s evaluation formula (See
Corollary 3.1.2 of [1])
   1+c c 
b c −c 2 −b 2
2 F1 ; −1 = 2 2 F1 ; 1
c−b+1 c−b+1
(38) 1
!
− 1 + c − b, 2c + 1
 
−c c b + 1, 2
= 2 Γ c+1 =Γ .
c
2 ,1 + 2 − b
1 + c, 2c − b + 1
Exercise 1.12. Work out the last equality.
By Definition (5), the left hand side is a well-posed series evaluated at −1.
1.10.6. From Bailey transform. There are other nice methods described in the textbooks including
Bailey transform [27, 2.4], we will leave the interested readers to check the details. But we would
like to mention as a consequence, one can prove the following formula originally discovered by
Whipple in [34]. It says
 a 
a 1+ 2 c d e f g
(39) 7 F6 a ; 1
2 1+a−c 1+a−d 1+a−e 1+a−f 1+a−g
   
1 + a − e, 1 + a − f, 1 + a − g, 1 + a − e − f − g a e f g
=Γ · 4 F3 ; 1 ,
1 + a, 1 + a − f − g, 1 + a − e − f, 1 + a − e − g e+f +g−a 1+a−c 1+a−d
when both sides terminate. Again the left hand side is a well-posed series while the 4 F3 (1) series
on the right hand side is 2-balanced.
1.10.7. From formulas to identities. As the audience may already notice, these hypergeometric
formulas give a rich source of identities. For instance, if we let a, b = −n in Gauss evaluation 29,
we will get immediately
n     
X n m n+m
(40) = .
k k n
k=0
n  2
X n
Exercise 1.13. Guess a formula for (−1)k and then prove your claim.
k
k=0

References
[1] George E. Andrews, Richard Askey, and Ranjan Roy. Special functions, volume 71 of Encyclopedia of Mathematics
and its Applications. Cambridge University Press, Cambridge, 1999.
[2] F. Beukers and G. Heckman. Monodromy for the hypergeometric function n Fn−1 . Invent. Math., 95(2):325–354,
1989.
[3] Frits Beukers. Notes of differential equations and hypergeometric functions. unpublished notes.
[4] Jonathan M. Borwein and Peter B. Borwein. Pi and the AGM, volume 4 of Canadian Mathematical Society Series
of Monographs and Advanced Texts. John Wiley & Sons, Inc., New York, 1998. A study in analytic number theory
and computational complexity, Reprint of the 1987 original, A Wiley-Interscience Publication.
[5] D. V. Chudnovsky and G. V. Chudnovsky. Approximations and complex multiplication according to Ramanujan.
In Ramanujan revisited (Urbana-Champaign, Ill., 1987), pages 375–472. Academic Press, Boston, MA, 1988.
[6] Pete L. Clark and John Voight. Algebraic curves uniformized by congruence subgroups of triangle groups. Trans.
Amer. Math. Soc., 371(1):33–82, 2019.
[7] Henri Cohen. Number theory. Vol. II. Analytic and modern tools, volume 240 of Graduate Texts in Mathematics.
Springer, New York, 2007.
24
 [8] Alyson Deines, Jenny G. Fuselier, Ling Long, Holly Swisher, and Fang-Ting Tu. Generalized Legendre curves
and quaternionic multiplication. J. Number Theory, 161:175–203, 2016.
[9] Alyson Deines, Jenny G. Fuselier, Ling Long, Holly Swisher, and Fang-Ting Tu. Hypergeometric series, truncated
hypergeometric series, and Gaussian hypergeometric functions. In Directions in number theory, volume 3 of Assoc.
Women Math. Ser., pages 125–159. Springer, [Cham], 2016.
[10] NIST Digital Library of Mathematical Functions. http://dlmf.nist.gov/, Release 1.1.0 of 2020-12-15. F. W. J.
Olver, A. B. Olde Daalhuis, D. W. Lozier, B. I. Schneider, R. F. Boisvert, C. W. Clark, B. R. Miller, B. V.
Saunders, H. S. Cohl, and M. A. McClain, eds.
[11] Marc Formichella, R. M. Green, and Eric Stade. Coxeter group actions on 4 F3 (1) hypergeometric series. Ra-
manujan J., 24(1):93–128, 2011.
[12] Jenny Fuselier, Ling Long, Ravi Ramakrishna, Holly Swisher, and Fang-Ting Tu. Hypergeometric functions over
finite fields. Memoirs of AMS, to apper.
[13] Édouard Goursat. Sur l’équation différentielle linéaire, qui admet pour intégrale la série hypergéométrique. Ann.
Sci. École Norm. Sup., 10:3–142, 1881.
[14] Benedict H. Gross. On the periods of abelian integrals and a formula of Chowla and Selberg. Invent. Math.,
45(2):193–211, 1978. With an appendix by David E. Rohrlich.
[15] Nicholas M. Katz. p-adic properties of modular schemes and modular forms. In Modular functions of one vari-
able, III (Proc. Internat. Summer School, Univ. Antwerp, Antwerp, 1972), pages 69–190. Lecture Notes in
Mathematics, Vol. 350, 1973.
[16] Wen-Ching Winnie Li, Ling Long, and Fang-Ting Tu. A Whipple 7 F6 formula revisted. arXiv 2103.08858, 2021.
[17] Ling Long. Hypergeometric evaluation identities and supercongruences. Pacific J. Math., 249(2):405–418, 2011.
[18] Ling Long. Some numeric hypergeometric supercongruences. In Vertex operator algebras, number theory and
related topics, volume 753 of Contemp. Math., pages 139–156. Amer. Math. Soc., Providence, RI, 2020.
[19] Ling Long, Robert Osburn, and Holly Swisher. On a conjecture of Kimoto and Wakayama. Proc. Amer. Math.
Soc., 144(10):4319–4327, 2016.
[20] Ling Long and Ravi Ramakrishna. Some supercongruences occurring in truncated hypergeometric series. Adv.
Math., 290:773–808, 2016.
[21] Ling Long, Fang-Ting Tu, Noriko Yui, and Wadim Zudilin. Supercongruences for rigid hypergeometric calabi–yau
threefolds. ArXiv e-prints, 2017.
[22] Yu. V. Nesterenko. Modular functions and transcendence questions. Mat. Sb., 187(9):65–96, 1996.
[23] Hans Petersson. über die eindeutige bestimmung und die erweiterungsfähigkeit von gewissen grenzkreisgruppen.
Abh. Math. Sem. Univ. Hamburg, 12(1):180–199, 1937.
[24] Theodor Schneider. Arithmetische Untersuchungen elliptischer Integrale. Math. Ann., 113(1):1–13, 1937.
[25] Theodor Schneider. Zur Theorie der Abelschen Funktionen und Integrale. J. Reine Angew. Math., 183:110–128,
1941.
[26] A. Selberg and S. Chowla. On Epstein’s zeta-function. J. Reine Angew. Math., 227:86–110, 1967.
[27] Lucy Joan Slater. Generalized hypergeometric functions. Cambridge University Press, Cambridge, 1966.
[28] P. F. Stiller. Classical automorphic forms and hypergeometric functions. J. Number Theory, 28(2):219–232, 1988.
[29] Kisao Takeuchi. Arithmetic triangle groups. J. Math. Soc. Japan, 29(1):91–106, 1977.
[30] Kisao Takeuchi. Commensurability classes of arithmetic triangle groups. J. Fac. Sci. Univ. Tokyo Sect. IA Math.,
24(1):201–212, 1977.
[31] Raimundas Vidūnas. Transformations of some Gauss hypergeometric functions. J. Comput. Appl. Math., 178(1-
2):473–487, 2005.
[32] Raimundas Vidūnas. Algebraic transformations of Gauss hypergeometric functions. Funkcial. Ekvac., 52(2):139–
180, 2009.
[33] F. J. W. Whipple. A Group of Generalized Hypergeometric Series: Relations Between 120 Allied Series of the
Type Formula. Proc. London Math. Soc. (2), 23(2):104–114, 1924.
[34] F. J. W. Whipple. On Well-Poised Series, Generalized Hypergeometric Series having Parameters in Pairs, each
Pair with the Same Sum. Proc. London Math. Soc. (2), 24(4):247–263, 1925.
[35] Masaaki Yoshida. Fuchsian differential equations. Aspects of Mathematics, E11. Friedr. Vieweg & Sohn, Braun-
schweig, 1987. With special emphasis on the Gauss-Schwarz theory.
[36] Masaaki Yoshida. Hypergeometric functions, my love. Aspects of Mathematics, E32. Friedr. Vieweg & Sohn,
Braunschweig, 1997. Modular interpretations of configuration spaces.
[37] Don Zagier. Elliptic modular forms and their applications. In The 1-2-3 of modular forms, Universitext, pages
1–103. Springer, Berlin, 2008.
[38] Wadim Zudilin. Ramanujan-type formulae for 1/π: a second wind? 54:179–188, 2008.

25