Class invariants and cyclotomic unit groups from

special values of modular units.

Amanda Folsom

Abstract
In this article we obtain class invariants and cyclotomic unit groups by con-
sidering specializations of modular units. We construct these modular units from
functional solutions to higher order q-recurrence equations given by Selberg in his
work generalizing the Rogers-Ramanujan identities. As a corollary, we provide a
new proof of a result of Zagier and Gupta, orginally considered by Gauss, regard-
ing the Gauss periods. These results comprise part of the author’s 2006 Ph.D.
thesis [6] in which the structure of these modular unit groups and their associated
cuspidal divisor class groups are also characterized, and a cuspidal divisor class
number formula is given in terms of products of L-functions and compared to the
classical relative class number formula within the cyclotomic fields.

1. Introduction and statement of results.

Let F` , ` ∈ N, be the modular function field with respect to the principal congruence
subgroup Γ(`) = {γ ∈ SL2 (Z) | γ ≡ 1 mod `} defined over Q(ζ` ), where ζ` = e2πi/` .
The modular functions that comprise invertible elements in the integral closure of the
ring Q[j] ⊂ F` form a group, the modular unit group of level `, where j = j(τ ) =
q −1 + 744 + 196884q + · · · is the classical modular invariant.
In this paper we define modular units r`,j (τ ), 1 ≤ j ≤ (` − 3)/2 for ` = 2k + 1 ≥ 5,
and the groups they generate

U`C := h{r`,j (τ ) | 1 ≤ j ≤ (` − 3)/2}i ⊂ U` (1)

using functional solutions to higher order q-difference equations found in the work of
Selberg [23], where in (1) and what follows, hxi denotes the group generated by x.
We will explicitly define the groups U`C and their generators r`,j , 1 ≤ j ≤ (` − 3)/2
in §3. To obtain class invariants we consider special values of the modular units r`,j ,
1 ≤ j ≤ (` − 3)/2, at points of complex multiplication, and show that only a single value
of one such modular unit is needed to generate class fields of prime moduli, to which
we compare the classical theory of complex multiplication, We then generalize these
results to class fields of more arbitrary moduli, motivated by the work of Ramachandra,
combining the theory of Kubert and Lang with that of Shimura.
With respect to the cyclotomic theory, we portray these modular units as modular
function field analogues to cyclotomic units, and show that when specialized to 0 these
modular units yield cyclotomic units that may be used to generate cyclotomic unit
groups. Dually, within the modular function fields we show how these modular units

1
may be used as generators. As a corollary, we provide a new proof of a result of Zagier
and Gupta, originally considered by Gauss. Let

S`C := {r`,j (τ ) | 1 ≤ j ≤ (` − 3)/2}. (2)

Our main results are the following.

Theorem 1.
i. For an odd integer ` = 2k + 1 > 5, the functions r`,j (τ ),
1 ≤ j ≤ (` − 3)/2 defined in (31), (32) and (34) are
modular units of level `.
ii. Further, the modular units r`,1 and r`,2 generate the field of
modular functions CF` . That is,

CF` = C(r`,1 , r`,2 ).

Theorem 2. For ` = 2k + 1 ≥ 5, the set of modular units S`C of level ` are such that
their limiting values in the cusp 0 are cyclotomic units in Q(ζ` )+ . That is,
n o
lim S`C := lim r`,j (τ ) | 1 ≤ j ≤ k − 1 ⊆ EC Q(ζ` )+ . (3)
τ →0 τ →0

Further, if ` is prime, then we have equality in (3), that is

D E
C C
EQ(ζ` )+ = lim S` .
τ →0

Theorem 3. Let K be an imaginary quadratic field with ring of integers

OK = Z[τ ], τ ∈ H an algebraic integer. Then for any prime ` ≥ 5 and any integer m
such that 1 ≤ m ≤ (` − 3)/2, the ray class field K` of modulus ` over K is given by

K` = K(r`,m (τ )).

We generalize the results of Theorem 3 to class fields of more arbitrary moduli f, with
generalized invariants Θf that reduce in a special case to powers of the (specializations
of the) modular units r`,j (τ ). In what follows, N(τ ) and T(τ ) respectively refer to the
norm and trace of τ ∈ H, and the normalized Klein form ϕ is given by
³ ¯ ´
¯ ∗ /2
ϕ z¯ ω1
ω2 = 2πie−zz σ(z|L)η(ω1 /ω2 )2 ω2−1

where z = z1 ω1 + z2 ω2 is given by the real coordinates z1 and z2 , z ∗ = z1 η1 + z2 η2 ,

and η1 , η2 are the quasi-periods of the elliptic Weierstrass ζ function of the lattice L
belonging to ω1 , ω2 .

2
Theorem 4. Let K be an imaginary quadratic field with ring of integers
OK = [τ, 1], τ ∈ H, and let f ⊆ OK be any ideal of the form f = s2 [τ + mτ , s1 ] where
s1 , s2 , mτ are integers such that (s1 , s2 ) = 1, s1 | N(τ ), s1 | mτ + T(τ ), s1 s2 - 6 = `, and
s1 , s2 prime. Then for any integers niP , 1 ≤ i ≤ m, m ∈ Z+ , and idealsQ bi ⊆ OK defined
m
by bi = [τ, bi ], with bi ∈ Z satisfying ( i=1 bi ni , `) = 1, (bi , `) = 1, bi | N(τ ), and for
any ξ ∈ K ∗ of the form ξ = sr11 τ + sr22 ∈ K ∗ with (r1 , s1 ) = (r2 , s2 ) = 1, the product
m
Y ¯
Θf := ϕ(ξ ¯fb−1
i )
12`ni
(4)
i=1

generates the ray class field of modulus f over K. That is,

K(Θf ) = Kf . (5)
Q
Further, if we do not require bi | N(τ ), we have the product Θf ∈ Kf .
A result in [21] asserts that under various conditions on the decomposition of a
modulus f = Zτ + Z, τ ∈ H, into prime ideals, and under certain hypotheses1 placed
on an ideal class b in the ray class group, where fb−1 = Zτ b−1 + Z, b ∈ Z, a simple
¡ ¯ ¢12`
quotient of the normalized Klein forms together with the specialization ϕ 1¯f
may be used to generate Kf over K:
¡¡ ¡ ¯ ¢ ¡ ¯ ¢¢12` ¡ ¯ ¢12` ¢
Kf = K ϕ 1¯fb−1 /ϕ 1¯f , ϕ 1¯f . (6)
The authors conjecture that the result (6) holds for arbitrary conductor f and every ideal
b prime to f whose ideal class in the ray class group is non trivial.
In [3], the authors form a more general product of normalized Klein forms and provide
hypotheses under which a product of such forms multiplied by an explicitly given root
of unity ζ lies in the ray class field:
Ym ³ ¯ −1 ´ni
¯
ζ ϕ ξλi ¯ τ b1i ∈ Kf . (7)
i=1

We note that an appropriate power of this product reduces in a special case (with
corrected hypotheses as pointed out in [3]) to (6) above. The authors conjecture that
with an additional set of hypotheses, the product (7) in fact generates Kf over K. We
prove in part these conjectures of [3] and [21] by a more general result here.
Theorem 5. As conjectured in [3], under the conditions of Theorem 4 and those given
in Theorem 1 of [3], ζ an explicitly given root of unity in [3], the product
m
Y ¯
Θ=ζ ϕ(ξ ¯fb−1
i )
ni

i=1

generates the ray class field modulus f over K.

1
We refer the reader to [21] for the complete theorem and hypotheses.

3
 Finally, to further develop the analogy between the modular units defining the groups
U`C and the cyclotomic units, as a corollary of the following Theorem 6, we provide a new
proof of a problem considered by Zagier and Gupta, and previously by Gauss, regarding
the Gauss periods.

Theorem 6. For an odd integer ` = 2k + 1 ≥ 3, the modular unit r`,1 (τ ) as defined in

(31) satisfies the functional equation

Mk (X) = X k + Ck−1 X k−1 + Ck−2 X k−2 + · · · + C1 X + C0

where
(q 2 ; q)j−1 Ck,k (q j )
Ck−j = (−1)b 2 c sk,j−1 (1)q j(k−1)(k+1−`)/2`
j+1
, (8)
(1 − q)j−1 Ck,k (q)

0 ≤ j ≤ k,
X∞
n2 +n (xq; q)n
Ck,k (z) = (−1)n xkn q (2k+1) 2 −kn (1 − xk q (2n+1)k ) . (9)
n=0
(q; q)n

Further, the limiting value of this functional equation in the cusp 0 gives an integral
polynomial

mk (x) = lim Mk (X) = xk + ck−1 xk−1 + ck−2 xk−2 + · · · c1 x + c0

τ →0

with coefficients given by

µ ¶
dj/2e b(k + j)/2c
ck−j = (−1) ,
b(k − j)/2c

0 ≤ j ≤ k. The roots of mk (x) are the Galois conjugates of the cyclotomic units
limτ →0 rd,1 (τ ), where d ranges over the divisors of `, d > 1.

Corollary 1. The minimal polynomial for the Gauss periods of degree 2 for odd ` ≥ 3
has coefficients of xj , 0 ≤ j ≤ k, given by
( k+1
(−1)b 2 c−j−1 ck−j j odd
dk−j = k
(−1)b 2 c−j ck−j j even

The structure of these modular unit groups and their associated cuspidal divisor class
groups are further characterized in [6], where in particular we provide a cuspidal divisor
class number formula given in terms of products of L-functions.

4
 2. Rogers-Ramanujan

The Rogers-Ramanujan continued fraction, defined by

q 1/5
r(τ ) = (10)
q
1 +
q2
1 +
q3
1 +
1 + .
..

convergent for τ in the upper half complex plane H has been an object of extensive study
due to a wealth of associated analytic properties, q-series identities, and combinatorial
interpretations (see [2], for example). Here instead, we present r(τ ) as an example of a
modular unit, and define the modular units r`,j , 1 ≤ j ≤ (` − 3)/2 in § 3 as higher level
analouges to the Rogers-Ramanujan function r(τ ).
Many of the fundamental properties associated to r(τ ) rely upon the fact that one
may regard r(τ ) as arising from the q-recurrence

R(z) = R(zq) + zqR(zq 2 ). (11)

That is, we may write

r(τ ) = q 1/5 R(q)/R(1). (12)

The q-recurrence (11) has a known analytic solution [20] given by

X 2
R(z) = z n q n (q; q)−1
n . (13)
n≥0

Here we use the q−Pochhammer symbol defined by


 1 k =0

 (1 − a)(1 − aq) · · · (1 − aq k−1 ) k = 1, 2, . . .
(a; q)k = −1

 [(1 − aq −1 )(1 − aq −2 ) · · · (1 − aq −k )] k = −1, −2, . . .

(1 − a)(1 − aq)(1 − aq 2 )(1 − aq 3 ) · · · k = ∞.

A celebrated result associated to r(τ ) are the Rogers-Ramanujan identities,

Y
R(1)−1 = (1 − q 5n−1 )(1 − q 5n−4 ) (14)
n≥1
Y
−1
R(q) = (1 − q 5n−2 )(1 − q 5n−3 ), (15)
n≥1

5
which give an infinite product representation for the series (13) evaluated at z = 1 and
z = q. Various proofs have emerged in the literature for the identities (14) and (15),
originally due to Rogers and (independently) Ramanujan, whose proofs rely upon the
use of Theta functions [20].2 The identities may also be interpreted combinatorially (see
for example [1]) in terms of number partitions. For example (14) may be read as saying
“the number of ways to partition a number into parts congruent to 1 and 4 mod 5 is
equal the number of partitions into parts of minimal difference 2.”
By (12), (13), (14) and (15), one obtains a q-product expansion for r(τ ). With this,
one may show that r(τ ) is in fact a modular function on Γ(5), and also a modular unit.
We also observe a parallel role played by a specialization of r(τ ) within the number
fields. By definition, the specialization of r(τ ) in the cusp 0 is easily observed:
1 1 1 √
lim r(τ ) = = (−1 + 5)/2. (16)
τ →0 1 + 1 + 1 + ···
Rewriting the special value (16) another way, we find
√
(−1 + 5)/2 = (ζ52 −ζ5−2 )(ζ5 −ζ5−1 )−1 , (17)

which we recognize as a cyclotomic unit. This special value is real, and in fact generates
the group of cyclotomic units in Q(ζ5 )+ , where we let K + denote the maximal real
subfield of a given field K. We also note this specialization may be used to generate
the field Q(ζ5 )+ , that is, Q(ζ5 )+ = Q(limτ →0 r(τ )). Rephrasing, we have a modular
unit arising from a q-recurrence that generates a modular function field, and whose
specialization in the cusp 0 generates a real cyclotomic field, as well as its cyclotomic
unit group. In what follows we consider the fields F` , ` ≥ 5, as well as specializations
at other τ in H, and provide more general proofs of these facts.

3. The modular unit groups U`C

To define higher order modular unit groups analogous to U5C = hr(τ )i generated by the
Rogers-Ramanujan continued fraction r(τ ), we seek other appropriate modular functions
on the curves X(N ). Due to the fact that r(τ ) has a continued fraction expansion
and may be viewed as originating from a certain q-recurrence, one might hope that if
analogous functions exist, that they exhibit similar traits. We indeed form such families
for curves X(N ) of higher genus g > 0 in what follows, yet will see that these functions
do not possess a continued fraction expansion but rather an expression that may be
viewed as a generalization of a continued fraction expansion. To describe this, we begin
with the observation of Sylvester that a continued fraction may be expressed as the
2
See section 4.

6
limiting value of ratios of nth order determinants. That is,
¯ ¯¯ ¯−1
¯ a0 b1 0 0 ···
¯¯ 1 0 0 0 ···
¯
b1 ¯ −1 a1 b2 0 ···
¯¯ 0 a1 b2 0 ···
¯
= ¯¯ ¯¯ ¯
0 −1 a2 b3 ··· 0 −1 a2 b3 ···
a0 + 0 0 −1 a3 ···
¯¯ 0 0 −1 a3 ···
¯
b2 ¯ . . . . ¯¯ . . . . ¯
a1 + . . . . . . . .
b3 . . . . . . . .

a2 +
a3 + .
.
.

where the successive ratios of order n determinants are the partial convergents of the
continued fraction. One finds that the Rogers-Ramanujan continued fraction may be
expressed as
¯ ¯¯ ¯−1
¯ 1 0 0 0 ···
¯¯ 1 q 0 0 ···
¯
¯
1/5 ¯
0 1 q2 0 ···
¯¯ −1 1 q2 0 ···
¯
r(τ ) = q ¯ 0 −1 1 q3 ··· ¯¯ 0 −1 1 q3 ··· ¯ (18)
0 0 −1 1 ··· ¯¯ 0 0 −1 1 ··· ¯
¯ .
.
.
.
.
.
.
. ¯¯ .
.
.
.
.
.
.
. ¯
. . . . . . . . .

We generalize this definition, and let A(n) be the order n determinant with entries ai,j ,
1 ≤ i, j ≤ n, with B (n) defined similarly. Provided the limit exists, we define the ratio
of two infinite determinants by
¯ ¯¯ ¯−1
¯ a1,1 a1,2 a1,3 a1,4 ···
¯¯ b1,1 b1,2 b1,3 b1,4 ···
¯
¯ a2,1 a2,2 a2,3 a2,4 ···
¯¯ b2,1 b2,2 b2,3 b2,4 ···
¯
lim A(n) /B (n) := ¯ a3,1 a3,2 a3,3 a3,4 ···
¯¯ b3,1 b3,2 b3,3 b3,4 ···
¯ (19)
n→∞ ¯ a4,1 a4,2 a4,3 a4,4 ···
¯¯ b4,1 b4,2 b4,3 b4,4 ···
¯
¯ .
.
.
.
.
.
.
.
.
.
.
.
¯¯ .
.
.
.
.
.
.
.
.
.
.
.
¯
.

With hopes of constructing analogous modular functions to (18) of higher level, we form
the infinite determinant Sk (z) of width k + 1
¯ sk,0 (z) sk,1 (z) sk,2 (z) ¯
¯ −1 sk,0 (zq) sk,1 (zq) sk,2· ·(zq) · sk,k−1 (z) 0 0 0 ···
¯
¯ 0 ··· sk,k−1 (zq) 0 0 ···
¯
Sk (z) := ¯ −1 sk,0 (zq 2
) sk,1 (zq 2
) sk,2 (zq 2
) · · · sk,k−1 (zq 2
) 0 · · ·
¯ (20)
¯ . . .
.
.
.
.
.
.
.
.
.
.
.
.
. ¯
. . . . . . . . .

One may deduce from (20) the higher order q-recurrences satisfied by the functions
Sk (z)
k
X
sk,m−1 (z)Sk (zq m ) = 0 (21)
m=0

where we let sk,−1 (z) = −1. We must define the functions sk,j (z) appropriately so that
(21) yields an analytic solution Sk (z).

7
We let
sk,2n (z) :=(−1)n z nk q δ(k,n) (q; q)−2
n
b k+1
2 c
X
−n−1
(22)
z j q j(n+1) (q j+1 ; q)n (q b 2 c−(j+n) ; q)n
k+1
×
j=0

sk,2n−1 (z) :=(−1)n+1 z nk−b 2 c q δ (k,n) (q; q)−2

k 0 n −1
n−1 (1 − q )
bX
2c
k
−n
(23)
z j q jn (q j+1 ; q)n (q b 2 c−(j+n)+1 ; q)n−1
k
×
j=0

for n ≥ 0 in (22), and n ≥ 1 in (23), where

¹ º
n2 + n k+1
δ(k, n) = (2k + 1) −n
2 2
2−n
¹ º
0 n k + 1 3
δ (k, n) = (2k + 1) +n ,
2 2

and

bαc = max{n ≤ α}
n∈Z
dαe = min{n ≥ α}.
n∈Z

As given in (22) and (23), the functions sk,m (z) first appear in the work of Selberg
[23], and in the case k = 2, the q-recurrence (21) reduces to (11), with S2 (z) = R(z).
With this choice of sk,j , an analytic solution to the q-recurrence (21) exists, and is given
in [23] by
∞
X n2 +n (xq; q)n Y
Sk (z) = (−1)n xkn q (2k+1) 2
−kn
(1 − xk q (2n+1)k ) (1 − zq n )−1 . (24)
(q; q)n
n=0 n≥1

Selberg uses these functions to give certain identities analogous to the Rogers-Ramanujan
identities (14) and (15), and also to give various q-continued fraction identities. In con-
trast to their use in [23], we use the functions sk,j (z) and Sk (z) to define groups of
modular units UNC . To define the modular units r`,j , we first recall the Galois action on
the modular function fields FN .
3
We note that the exponent given in [23] corresponding to δ 0(k, n) mistakenly reads
2 ¥ ¦
(2k + 1) n 2+n + n k+1
2 .

8
 The natural action of the group Γ(1) ⊂ M2 (Z) on the fields FN given by
γ · f (τ ) = f (γτ ), (25)
where f = f (τ ) ∈ FN , may be extended to the group GL2 (Z/N Z) as follows. Given
³ ´
1 0
d ∈ (Z/N Z)∗ , let γd = 0 d act on Q(ζN ) by

γd · ζN = ζNd .
∞
X
If f ∈ FN has q-series expansion given by f (τ ) = an q n/N , the action of γd extends
n=m
to FN by
∞
X
γd · f (τ ) = (γd · an )q n/N . (26)
n=m

The matrices γd , d ∈ (Z/N Z)∗ , together with SL2 (Z/N Z), generate GL2 (Z/N Z), and
the group actions given by (25) and (26) define a Galois action on the modular function
fields, described by the following exact sequence
1 −→ GN · {±1} −→ G −→ Gal(FN /F1 ) −→ 1 (27)
where
Y
G = GL2 (Zp ) · G+
∞ (28)
p

GN = {(x) ∈ G/G+
∞ | xp ≡ 1 mod N · M2 (Zp )}. (29)

The product defining G is taken over primes p, and Zp denotes the ring of integers in
the completion Qp of the field Q at p. From (27), one finds the isomorphism
π
Gal(FN /F1 ) ∼
= GL2(Z/N Z)/{±1}. (30)

We first define using (24) the functions r`,1 and r`,2 .

Definition. For ` = 2k + 1, let
−k(k−1) Sk (1)
r`,1 (τ ) = q 2` (31)
Sk (q)

−(k+1)(k−2) Sk (1) − q k−1 Sk (q 2 )

r`,2 (τ ) = q 2` . (32)
Sk (q)

9
 For m ∈ (Z/N Z)∗ , we let
¡m 0
¢ ¡ ¢
σm := 0 1 ∈ π Gal(FN /F1 ) (33)

and for ` = 2k + 1, let

(3`−4)(`−3) η(τ )
Ψ`,q := −q 24` Sk (q).
η(`τ )
We define for 1 ≤ j ≤ (` − 3)/2 the functions
¡ ¢
r`,j (τ ) := σk+1−j Ψ`,q · Ψ−1
`,q (34)

where the Dedekind η-function η(τ ) = ∆1/24 (τ ) is the 24th root of the Discriminant
function, and is given by
Y
η(τ ) = q 1/24 (1 − q n ). (35)
n≥1

4. Theta constants

To prove Theorems 1 and 2 we will make use of the theory of the theta constants. A theta
£ ¤
characteristic is a vector χ = ²²0 ∈ R2 . Two characteristics are said to be equivalent
if their difference is in Z2 , and the space of characteristic classes is defined as
R2 modulo this equivalence relation. Given a matrix γ = (ac db) ∈ Γ(1)/{±1}, a right
group action on the space of characteristic classes is defined by
£ −ac ¤
χγ = γ t χ + bd (36)

£²¤
The theta constant with characteristic χ = ²0 ∈ R2 is defined by

£²¤ X ³1 ³ ²
´2 ³ ´ ´
² ²0
θ ²0 (τ ) = e 2 n+ 2 τ + n+ 2 2 .
n∈Z

This function converges for τ ∈ H, and satisfies the transformation rule

θ [χ] (γτ ) = κχ,γ (cτ + d)1/2 θ [χγ] (τ ) (37)

for γ ∈ Γ(1), and

³ 1 1 ´
0 2 02 0
κχ,γ = e − (a² + c² )bd − (ab² + cd² + 2bc²² ) κγ
4 8

10
where κγ is an eighth root of unity depending only on the matrix γ. As a special case
of (37), one finds for r, s ∈ Z,
£ +2r¤ ³ ´ £ ¤
²s
θ ±² 0
±² +2s (τ ) = e ± 2
θ ²²0 (τ ). (38)

The theta constants also satisfy the product identity

£ ¤ ¡ ²²0 ¢ 2 Y ¡ ²0 ¢ 1+² ¡ −²0 ¢ 1−²
θ ²²0 (τ ) = e 4 q ² /8 (1 − q n )(1 + e 2 q n− 2 )(1 + e 2 q n− 2 ), (39)
n≥1

which can be derived from the Jacobi triple product identity [5]
∞
X ∞
Y
2
q n z 2n = (1 − q 2n )(1 + q 2n−1 z 2 )(1 + q 2n−1 z −2 ). (40)
n=−∞ n=1

To prove statement i. of Theorem 1, using (24) we apply Jacobi’s triple product

formula (40) and find

Y (1 − q `n−k )(1 − q `n−(k+1) )(1 − q `n )

Sk (1) = (41)
n≥1
(1 − q n )
Y (1 − q `n−1 )(1 − q `n−(`−1) )(1 − q `n )
Sk (q) = (42)
n≥1
(1 − q n )
Y (1 − q `n−(k−1) )(1 − q `n−(k+2) )(1 − q `n )
Sk (1) − q k−1 Sk (q 2 ) = (43)
n≥1
(1 − q n )
so that
Y (1 − q `n−k )(1 − q `n−(k+1) )
r`,1 = q −k(k−1)/2` (44)
n≥1
(1 − q `n−1 )(1 − q `n−(`−1) )
Y (1 − q `n−(k−1) )(1 − q `n−(k+2) )
r`,2 = q −(k+1)(k−2)/2` . (45)
n≥1
(1 − q `n−1 )(1 − q `n−(`−1) )

Using (34), (44) and (45) we apply (39) and (37) and find for 1 ≤ m ≤ (` − 1)/2,
· 2m−1 ¸
θ ` (`τ )
1
r`,m (τ ) = e( 2` ) · `−2 ¸
k−m
(46)
θ ` (`τ )
1

We will use the following Lemma.

11
 £²¤
Lemma 1. For any odd `², ²0 and `, and any γ ∈ Γ(`), the theta constant θ ²0 (`τ )
satisfies the transformation
£²¤ £²¤
θ ²0 (`·γτ ) = νχ,γ (cz + d)1/2 θ ²0 (`τ )

where
¡ ¢ ¡ ¢ ¡ ¢
νχ,γ = e −²d(a − 1)(b` + ²0 )/4 e −²2 (b`(a − 1))/8 e −²0 d(2bc + c²0/`)/8 .

Proof. (Lemma 1) We compute

£²¤ £²¤
θ ²0 (`·γτ ) = θ ²0 (γ̃ ◦ `τ )

³ ´
a b`
where γ̃ = c/` d ∈ SL2 (Z), and find

h i
t −ac/`
χγ̃ = γ̃ χ + bd`
h 0
i
= a²b`²+ +c²d²/`0 − ac/`
+ bd`
.

For ease of notation, let γ(²) = a² + c²0 /` − ac/` and γ(²0 ) = b`² + d²0 + bd`.
³ ´ ³ ´
a b 1 + A` B`
If c d = C` 1 + D` , we find

γ(²) = ² + A`² + C²0 − C(1 + A`)

γ(²0 ) = ²0 + D`²0 + B`2 ² + B`2 (1 + D`)

If A is odd, then C must be odd, as det(γ) = 1. In this case, γ(²) ≡ ² mod 2. This
also holds if A is even, regardless of the parity of C. Similarly, if D is odd, then B is
odd, so that γ(²0 ) ≡ ²0 mod 2. Again we find that this also holds if D is even. By (37)
and (38) we have
c
θ[χ](γ̃ ◦`τ ) = κχ,γ̃ ( ` ·`τ + d)1/2 θ[χγ̃](`τ )
²
= κχ,γ̃ e( 4 (b`² + (d − 1)²0 + bd`))(cτ + d)1/2 θ[χ](`τ ).

Combining the constants gives the expression for νχ,γ and completes the proof.

12
Proof. (Theorem 1 i.)
To show that the functions r`,j are modular units of level `, we see by (44), (45) and
(34) that the r`,j are holomorphic on H. Thus it suffices to show the functions transform
correctly under Γ(`), and that they in fact lie in U` . To show transformation, by Lemma
1 and (46) it suffices to show that νχ1 ,γ = νχ2 ,γ , where
· ¸ · ¸
²1 2m−1
χ1 = 0
= `
² 1
· ¸ · ¸
²2 `−2
χ2 = = ` .
²0 1

We have
νχ1 ,γ ³ ´
d(a−1) 2 2
= e( 4 (²2 − ²1 )(1 + b`))e b`(a−1) (² 2 − ²1 )
νχ2 ,γ 8

³ ´ ³ ´
d(a−1)
= e 2` (k − m)(1 + b`) e b(a−1)2`
(k − m)(k + m + 1)
³ ´ ³ ´
dA
= e 2 (k − m)(1 + b`) e B(a−1)
2
(k − m)(k + m + 1)
³ ´
dA
= e 2 (k − m)(1 + b`) . (47)
The last equality follows from the fact that (k − m) and (k + m + 1) have opposite
parity. Finally, if b is odd, then the expression in (47) simplifies to 1. If b is even,
then ad = (1 + `A)d is odd, so that A is even, and again the expression (47) is equal
to 1. By examining (45), (45) and (34), one finds that the functions r`,j have divisors
supported on the cusps of Γ(`), a fact we prove more rigorously in Section 2.4. This
proves Theorem 1 i.
³ ´
Proof. (Theorem 2). For the involution γ = 01 −10 , we have by (37)

£²¤ £ ²0 ¤
θ ²0 (−1/τ ) = κτ 1/2 θ −² (τ )

so that
lim r`,m (τ ) = lim r`,m (−1/τ )
τ →0 τ →∞
· 2m−1 ¸
θ ` (−1/`τ )
1
= lim e( 2` ) · `−2 ¸
k−m
τ →∞
θ ` (−1/`τ )
1
· ¸
1
θ 2m−1 (`τ )
= lim e( 2` ) · ` ¸
k−m
.
τ →∞ 1
θ 2−` (`τ )
`

13
We use the fact that

£1¤ X
θ ²0 (τ ) = q 1/8 e(²0 /4) e((n2 + n)τ /2 + n²0 /2)
n∈Z
1/8
= q (e(² /4) + e(−²0 /4) + O(|q|))
0
(48)

as τ → ∞, to conclude
2m−1 1−2m
ζ` 4
+ ζ` 4

lim r`,m (τ ) = `−2 2−`

τ →0
ζ` 4 + ζ` 4
k+1−m
− k+1−m
ζ` 2
− ζ` 2

= 1
−1
ζ`2 − ζ` 2
m−k ζ k+1−m − 1
= ζ` 2 `
ζ` − 1
1−v ζ v − 1
= ζ` 2 ` (49)
ζ` − 1
where v = k + 1 − m. To conclude the proof, we state two well known results regarding
the description of the cyclotomic unit groups [26].
Lemma 2. For N = pj , p prime, j ≥ 1,
C C
EQ(ζ N)
= h{ζN , EQ(ζN)
+ }i.

Lemma 3. For N = pj , p prime, j ≥ 1,

Dn − ζNm oE
C (1−m)/2 1
EQ(ζN)
+ = − 1, ζN | 1 < m < N/2, (m, p) = 1 .
1 − ζN

As m ranges over integers {1, 2, 3, . . . , `−1

2
− 1}, we note that v ranges over integers
`−1
{2, 3, . . . , 2 }. With v in this range, by Lemma 3 we see that the set of specializations
+
limτ →0 S`C may be used to generate the group EQ(ζ C
`)
+ of cyclotomic units in Q(ζ` ). When

` is not a prime power, there may be multiplicative dependence between the elements in
limτ →0 S`C , although these primitive cyclotomic units in Q(ζ` )+ may be used to define
a set of multiplicatively independent units as given by Ramachandra [18]. This proves
Theorem 2..
To prove statement ii. of Theorem 1, we turn to a discussion of the Siegel functions.

14
 5. Siegel functions

The Klein forms ta (τ ), defined for a vector a = (a1 , a2 ) ∈ R2 , and τ ∈ H, are given
by

ta (τ ) = e−ηa (τ )a·(τ,1)/2 σa (τ ), (50)

where σa and ηa are the classical Weierstrass functions. Using known properties
of the Weierstrass σa and ηa functions, one can verify for a lattice L that the
Klein forms satisfy the following properties

ta (λL) = λta (L) (51)

ta (γL) = taγ (L) (52)
ta+b (L) = ²(a, b)ta (L). (53)

In (51) λ ∈ C∗ , in (52) γ ∈ Γ(1), and in (53) b = (b1 , b2 ) ∈ Z2 , and the constant ²(a, b)
is given by
²(a, b) = (−1)b1 b2 +b1 +b2 e((b2 a1 − b1 a2 )/2).
The Klein forms are used to define the Siegel functions, given by

ga (τ ) = ta (τ )∆(τ )1/12 .

A major result of Kubert and Lang in their development of the theory of the modular
units is the following explicit characterization of the modular units of prime power level
`.
Theorem 7. (Kubert, Lang) Let ` = pn , where p is prime, p 6= 2, 3, n ∈ N. Then the
modular units of level ` (modulo constants) consist of products
Y
gam(a)
a

where a = ( a`1 , a`2 ) ∈ 1` Z2 , a ∈

/ Z2 , and the exponents m(a) ∈ Z satisfy the quadratic
relations
X X X
m(a)a21 ≡ m(a)a22 ≡ m(a)a1 a2 ≡ 0 mod `
a a a

Using the product expansion for the classical ∆ and σ functions, one has the following
product formula for the Siegel functions:
³ a1 ¡
´Y
∞ ³ ´³ ´
1 a1
a ¢ a1 ¡
a ¢ a1 ¡
a ¢
ga (τ )= −q 2 B2 ( `
)
e(a2 (a1 − 1)/2) 1−q ` e `2 1−q n+ ` e `2 1−q n− ` e − `2
n=1
(54)

15
where B2 (z) is the second Bernoulli polynomial defined by
1
B2 (z) = z 2 − z + .
6
One may verify by the Galois action described in Section 1.1, the product formula (54),
and the action (52), that the Galois group Gal(F` /F1 ) acts transitively on the Siegel
functions by multiplication on the indices. That is, for β ∈ Gal(F` /F1 ),

β · ga = ga·β . (55)

From (44), (45) and (54) we find the following expression for the modular units r`,1 (τ ), r`,2 (τ ):
`−1
Y
r`,1 (τ ) = g(k/`,s/`) /g(1/`,s/`) (56)
s=0
`−1
Y
r`,2 (τ ) = g(k−1/`,s/`) /g(1/`,s/`) . (57)
s=0

We let F` := CF` be the function field of the curve X(`) over C. The field F`
is Galois over C(j) with Galois group Gal(F` /C(j)) ∼
= Γ(1)/Γ(`) · {±1}. When
` > 5, the curve has positive genus, so is generated by two elements over C.

In [8], the author gives a pair of generators X2 and X3 for F` , ` ≥ 6, with

q−product expansions
Y (1 − q `n−2 )(1 − q `n−(`−2) )
X2 = q −(k−1)/` (58)
n≥1
(1 − q `n−1 )(1 − q `n−(`−1) )
Y (1 − q `n−3 )(1 − q `n−(`−3) )
X3 = q −(2k−3)/` . (59)
n≥1
(1 − q `n−1 )(1 − q `n−(`−1) )

By (58), (59), the expressions (56), (57), the product representation (54), and the
Galois action (55), we have
Proposition 1. X2 = σk−1 (r`,1
−1
), X3 = σk−1 (r`,1
−1
r`,2 ).
Proof. (Theorem 1 ii.) We have

F` ∼
= σk (F` ) = σk (C(X2 , X3 )
= C(σk (X2 ), σk (X3 ))
−1 −1
= C(r`,1 , r`,1 r`,2 )
= C(r`,1 , r`,2 ). (60)

16
However, the functions r`,j are modular units by Theorem 2 i., so that C(r`,1 , r`,2 ) ⊆
F` . By the isomorphism in (60), we must have

C(r`,1 , r`,2 ) = F` .

This concludes the proof of Theorem 1 ii.

6. Gauss periods
6.1 Gauss periods

Given an odd prime ` with factorization ` = f k + 1, f ≥ 1, there is a unique subfield

K of Q(ζ` ) with [Q(ζ` ) : K] = f . If we let {mn }, 1 ≤ n ≤ k be a set of representatives
¡ ¢k
for the cosets of (Z/`Z)× / (Z/`Z)× , the Gauss periods of degree f are defined for each
n by

TrQ(ζ` )/K (ζ`mn ), (61)

and have a common minimal polynomial Fk (x) of degree k. In the case f = 2 we note
the unique subfield of degree 2 is Q(ζ` )+ , and we may choose mn = n. In this case
Gauss explicitly described the coefficients dk−j of xj , 0 ≤ j ≤ k, of the degree k minimal
polynomial Fk (x) by
à k+j !
b 2 c
dk−j = (−1)b 2 c
k−j
(62)
b k−j
2 c .

Later Sylvester indicated how one may obtain the coefficients for composite ` and f = 2
recursively, and gave a list of the polynomials for 1 ≤ ` ≤ 36. The results of Gauss and
Sylvester may be found in [?] and [25]. The period polynomials for various `, f and k
have since been investigated. In [7], Gupta and Zagier consider the case f = 2, and
extend the definition (61) to odd `. In [7] the authors prove the reciprocal
polynomial

fk (x) = xk Fk (x−1 ) (63)

with roots {1/TrQ(ζ` )/K (ζ`n ) | 1 ≤ n ≤ k} has coefficients dk−j of xk−j in agreement
with those given by Gauss in (62). Here, we recover this original result of Gauss, and the
more general result of Gupta and Zagier by different means. In particular, we observe
this result after considering the limiting value of a functional equation satisfied by the
modular unit r`,1 . To prove Theorem 6, we will use the following lemmas. The first
follows from the fact that
1 − qa
lim = a.
τ →0 1 − q

17
Lemma 4. For integers r, j ≥ 0 and k ≥ 1,
m1X
−r−1
lim sk,2r (q j ) = (−1)r (n+r
n )(m1 −(n+r+1))
m1 −(n+1)
τ →0
n=0
m
X2 −r

lim sk,2r−1 (q j ) = (−1) r+1

(n+r
n )(m2 −(n+r))
m2 −(n+1)
τ →0
n=0

where m1 = b k+1
2
c and m2 = b k2 c.
Lemma 5. For integers m, n ≥ 0 and j ≥ 1,
(j) (j)
lim Sk (q n ) − Sk (q m ) = 0
τ →0

Lemma 6. For integers m, r ≥ 0 such that the following expressions are defined, we
have
X µr+n¶µ m−r−n−1 ¶ µ m ¶
m−2r−1
= (64)
n=0
n m−2r−n− 1 2r+1
X µr+n¶µm−r−n−1¶
m−2r µ ¶
m
= (65)
n=0
n m−2r−n 2r

Proof. (Lemma 6.)

For r ≥ 0 we have
∞ µ
X ¶
−r−1 r+n
(1 − x) = xn (66)
n=0
n
¡r+n¢ ¡ m−r−n−1 ¢
so that xn has coefficient n and xm−2r−n−1 has coefficient m−2r−n−1 , where 0 ≤ n ≤
m − 2r − 1. We find that xm−2r−1 in the ¡product¢ (1 − x −2r−2
¡ m ¢ ) has coefficient given by
m
the left hand side of (64), but also m−2r−1 = 2r+1 by (66). This gives (64), and
we argue similarly to establish (65).
Proof. (Lemma 5.)
For j = 1 we have for any r ≥ 0
b k+1
X2
c−1 ¹ º
(1) r r i(r+1) k+1
lim Sk (q ) = lim sk,0 (q ) = lim q = .
τ →0 τ →0 τ →0
i=0
2
By induction, the result follows for j > 1 from Lemma 4 and the fact that
k−1
X
(j) (j−(i+1))
Sk (q n ) = sk,i (q n )Sk (q n+i+1 ).
i=0

18
Proof. (Theorem 6.)
That the modular function r`,1 (τ ) satisfies the equation Mk (X) defined in Theorem 6
follows from (21), (24) and (31). If limτ →0 Mk (X) = mk (x) has coefficients ck−j of xk−j ,
by (8) and the Lemmas, we have for j even

X µ j + n¶µb k c − (n + 1)¶
b k2 c− 2j
j/2 2 2
ck−j = (−1)
n=0
n b k2 c − 2j − n

X µ j + n¶µb k−j c + j − n − 1¶
b k2 c− 2j
j/2 2 2 2
= (−1) k j
n=0
n b 2
c − 2
−n
µ ¶
k − b k−j+1 c
= (−1)j/2 2
j
µ ¶
j/2
k − b k−j+1
2
c
= (−1)
b k−j
2
c
µ k+j ¶
b 2 c
= (−1)j/2 k−j (67)
b 2 c

and for j odd

b k+1 c− j−1 −1 µ j−1 ¶µ ¶

2 X2
+n b k+1 c − (n + 1)
(j+1)/2 2 2
ck−j = (−1) j−1
n=0
n b k+1
2
c− 2
−n−1
b k+1 c− j−1
X −1 µ j−1 ¶µ k−j j−1 ¶
(j+1)/2
2 2
2
+n b 2 c+ 2
−n− 1
= (−1) j−1
n=0
n b k2 c − 2 −n
µ ¥ ¦¶
(j+1)/2 k − k−j+1
2
= (−1)
j
µ ¥ k−j+1 ¦¶
2
k−
= (−1)(j+1)/2 k−j
b 2 c
µ k+j ¶
b 2 c
= (−1)(j+1)/2 k−j (68)
b 2 c·
¡ m
¢ ¡m¢
We use the identity m−r
= r
, the fact that k − b k−j+1
2
c − b k−j
2
c = j, and that
¹ º ½
k−j b k2 c − 2j j even
= j−1
2 b k+12
c− 2
− 1 j odd.

19
and conclude that the limiting value limτ →0 r`,1 (τ ) satisfies the polynomial mk (x), with
coefficients given by (67) and (68), and that this polynomial is integral.
Proof. (Corollary 1) In [7], the authors prove via polynomial recurrences that for the
polynomial fk (x) as defined by (63), the coefficients dk−j of xk−j are given by (62).
Comparing (67) and (68), we find
½
(−1)j+1 ck−j k ≡ 0, 3 mod 4
dk−j = j
(−1) ck−j k ≡ 1, 2 mod 4

and conclude
d+1
mk (x) = (−1)b 2
c
fk ((−1)k x). (69)

By (49) we deduce
1
lim r`,1 (τ ) = ³ k+1 ´ (70)
τ →0 b c
TrF/F + ζ` 2

where F = Q(ζ` ). Thus the roots of mk (x) are given by

{(−1)k /TrQ(ζ` )/K (ζ`n ) | 1 ≤ n ≤ k}, (71)

which is equal to the union over the divisors d > 1 of ` of the Galois conjugates of the
cyclotomic units limτ →0 rd,1 (τ ), proving Corollary 1.

7. Class field theory

We next examine results within class field theory related to the modular units r`,j . We
note that Theorem 3 requires only one specialization of one modular function to generate
the field K` and compare the following classical result.

Theorem R. For K = Q(z) an imaginary quadratic field, and any positive integer N ,
the ray class field of conductor N over K is given by the field K · FN |z , where

FN |z = {f (z) | f ∈ FN , f (z) 6= ∞}.

That is KN = K · FN |z .

Proof. (Theorem R.) See [24], or [17].

20
Example. To illustrate Theorem 3, with τ = i, the ray class field of conductor 5 over
K = Q(i) is given by
Q(i)5 = Q(i, r(i))
where
q 1/5 q q 2 q 3 q 4
r(τ ) =
1+ 1+ 1+ 1+ 1 + · · ·
is the Rogers-Ramanujan continued fraction. The special value r(i), also determined by
Ramanujan, is given by s √ √
5+ 5 1+ 5
r(i) = − .
2 2
As another example we may take z = ρ, and find

Q(ρ)5 = Q(ρ, r(ρ)),

where p √ √
30 + 6 5 − 3 − 5
r(ρ) = e(−1/10) ,
4
also given by Ramanujan.

7.1 Shimura reciprocity

To prove Theorems 3 and 4, we will use in part the language of varieties as in [24]. Let
V and W be rational varieties defined over a common field k. A subvariety T ⊆ V × W
is called a rational map if (v, w) ∈ T ⇒ k(v, w) = k(v). A rational map is called a
rational function if W = A1 one dimensional affine space. The rational functions on V
form a field, which we denote by k(V ).
Let

G(p) = GL2 (Qp )

G∞ = GL2 (R)
Y
GA = {(x) ∈ Gp · G∞ | xp ∈ GL2 (Zp ) for almost all p.}
p

G0 = {(x) ∈ GA | x∞ = 1}
G+A = G0 G+
∞
Z = {S ⊆ G+ ∗ + ∗ +
A | S open, Q G∞ ⊆ S, S/Q G∞ compact.}

We note that a topology is placed on GA by declaring the set G, as defined in (28), to

21
be open. Let S ⊆ GA+ be an open subset of GA+ such that Q∗ G∞+ ⊆ S and S/Q∗ G∞+
is compact, and let kS ⊂ Qab be the Abelian extension of Q fixed by Q∗ det(S). Then
there is a natural variety VS corresponding to S defined over kS as follows. Given any
Fuchsian group Γ of the first kind, Γ\H∗ is a compact Riemann surface, so there exists
a nonsingular algebraic curve V defined over a subfield of C so that Γ\H∗ is biregularly
isomorphic to V . We say (V, φ) is a model of Γ\H∗ if φ is a Γ-invariant holomorphic
map, φ : H∗ −→ V , that yields such an isomorphism. If we let
FS = {h ∈ F | hσ(s) = h ∀ s ∈ S} (72)
ΓS = S ∩ GQ+ (73)
then there exists a model (VS , φS ) of ΓS \H∗ such that
VS is defined over kS (74)
FS = {f ◦ φS | f ∈ kS (VS )}. (75)
We summarize some known results in the following proposition.
Proposition 2. Each set S ∈ Z defines a corresponding subfield kS ⊆ Qab of finite
index.
Proof. (Proposition 2.) The composition
det [ (−)−1 , Q ]
GA −−−−−→ Q∗A −−−−−−−−−→ Gal(Qab /Q)
defines a homomorphism ϕ : GA → Gal(Qab /Q). We note
Q∗ · det(S) ⊆ Q∗A ,
and take the field kS so that
Gal(Qab /kS ) = ϕ(S).

Now for τ ∈ C − R imaginary quadratic, there is an embedding Eτ (see [24] 4.4)

Eτ : K ,→ M2 (Q)
such that
Eτ (K ∗ ) = {γ ∈ GL2 (Q)+ | γ(τ ) = τ },
and for k ∈ K ∗ ,
h i h i
τ τ
Eτ (k) 1 =k 1 (76)
We will use the following proposition, with the correspondence given in the sense of
Proposition 2.

22
Proposition S. (Shimura [24]) Let S ∈ Z, τ ∈ C − R imaginary quadratic, and let

W = {s ∈ KA∗ | Eτ (s) ∈ S}.

Then K · kS (φS (τ )) ⊆ K ab corresponds to the subgroup

K ∗ W ⊆ KA∗ .

Example. As an illustration of the proposition, for any N ∈ N, consider the set

SN = Q∗ {(x) ∈ G | xp ≡ 1 mod N · M2 (Zp )}.

Then ΓSN = Q∗ Γ(N ), so that (VSN , φSN ) is a model of Γ(N )\H∗ over Q. We will use
the following known result.

Proposition. With S = SN as above, the corresponding set W is given by

Y
Q∗ \W = WN = (1 + N Op )∗ ⊂ KA∗ . (77)
p

In what follows, we let τ be the variable on the upper half plane H. We let φN be
the biregular homomorphism associated to ΓSN \H∗ , and let VN be the locus of

φN (τ ) = (j(τ ), fa11 (τ ), fa21 (τ ), fa31 (τ ), fa12 (τ ), fa22 (τ ), fa32 (τ ), . . . )

1 2
where {ai }i = N
Z /Z2 − {0}, and for the lattice L = L(τ, 1), the Fricke functions fai are
defined by

g2 (L)g3 (L) ³ h τ i ´
fa1 (τ, L) = ℘ a 1 ;L
∆(L)
g2 (L)2 ³ h τ i ´2
fa (τ, L) =
2
℘ a 1 ;L
∆(L)
g3 (L) ³ h τ i ´3
fa3 (τ, L) = ℘ a 1 ;L .
∆(L)

where ℘ is the Weierstrass ℘−function, and g2 , g3 invariants of the lattice (see [17]).
Two curves are birational if and only if their function fields are isomorphic, and there
exists a birational map X : VSN −→ VN such that X ◦ φSN = φN . Thus the map X gives
an isomorphism
X
kSN (VSN ) −→ kN (VN )
hence
K · kSN (φSN (z)) ∼
= K · kN (φN (z)).

23
It is known that the field of modular functions of level N is generated by the modular
invariant j and the first Fricke functions fa1 , where a ∈ N1 Z2 , a ∈
/ Z2 . Thus, the field
ab ∗
K · FN |z ⊆ K corresponds to K W . Combining the above results, we arrive at the
following known result.
Y
Gal(K ab /KN ) ∼
= (1 + N Op )∗ . (78)
p

7.2 Galois action on UNC

Proposition 3. For fixed m, 1 < m ≤ (` − 1)/2,

n ¡ ¢ o
Gal(F/F1 (r`, `+1 −m )) ∼
= σ = (σp ) ∈ G | σp ≡ 10 nss mod ` · M2 (Z) ∼= (Z/`Z),∗
2

where ns , s ∈ Z satisfy

(m + 1)ns ≡ (1 − s)(` − 1)/2 mod `. (79)

Proof. (Theorem 3.) Let

n ¡ ¢ o
Sr := σ = (σp ) ∈ G | σp ≡ 10 nss mod ` · M2 (Z)

where ns , s satisfy (79). Assuming Proposition 3, as in Proposition S take S = Sr and

suppose u ∈ W . For any m, 1 < m ≤ (` − 1)/2, up ≡ s mod `Zp . By the property given
in (76)
Q defining the∗ embedding Eτ , we find s ≡ 1 mod `Zp , so that up ∈ 1 + `Op , hence
u ∈ p (1 + `OKp ) . To prove the converse, u ∈ 1 + `OK implies Eτ (u) ≡ 1 mod `, so
that Eτ (u) ∈ S, hence u ∈ W . Theorem 3 follows from the following proposition.
Proposition 4. For S and τ as in Proposition S, and FS as in (72), if there exists
some h ∈ FS such that h is non-singular at τ , and FS = F1 (h), then

K · kS (φS (τ )) = K · kS (h(τ )).

Proof. (Proposition 4.) We have h ∈ FS , so by (75) there exists some f ∈ kS (VS )

such that h = f ◦ φS , so that h(τ ) = f ◦ φS (τ ), which is non-singular by hypothesis.
f ∈ kS (VS ) implies kS (φS (τ ), f (φS (τ )) = kS (φS (τ )) so that h(τ ) ∈ K · kS (φS (τ )). If we
assume K · kS (h(τ )) ( K · kS (φS (τ )) then there exists s ∈ KA∗ , s ∈ / K ∗ W, such that
h(τ )[s,K] = h(τ ), where [−, K] denotes the Artin map. Shimura reciprocity implies
−1
h(τ )[s,K] = hEτ (s) (τ ),

hence Eτ (s)−1 ∈ S, hence Eτ (s) ∈ S, hence s ∈ W , which is a contradiction.

24
Proof. (Proposition 3.) By Theorem 7, each modular unit of level ` may be written as
a product of Siegel functions, which we have given previously for the functions r`,1 and
r`,2 in (56) and (57). Using (55), (34) and (42), we find

µY
`−1 ¶γ
g(m/`, s/`)
r`,γ `+1 −m =
2 g
s=0 (1/`, s/`)
`−1
Y g(m/`, s/`)γ
=
s=0
g(1/`, s/`)γ
`−1
Y g((am+cs)/`, (bm+ds)/`)
= . (80)
s=0
g((a+cs)/`, (b+ds)/`)

Let us assume γ ∈ Gal(F/F1 (r`, `+1 −m ). We first show that c ≡ 0 mod `. If we suppose
2
c 6≡ 0 mod `, then both of the sets {am + cs | 0 ≤ s ≤ ` − 1} and {a + cs | 0 ≤ s ≤ ` − 1}
contain a complete set of representatives for Z/`Z. We choose ts ≡ am + cs mod `,
rs ≡ a + cs mod `, where 0 ≤ ts , rs ≤ ` − 1, and find by (53)
g((am+cs)/`,∗) = κsm g(ts ,∗)
g((a+cs)/`,∗) = κs1 g(rs ,∗)
κsm , κs1 ∈ C. Thus, using (54), we find
`−1
`X
ord∞ r`,γ `+1 −m = B2 (ts ) − B2 (rs ) = 0.
2 2 s=0

However we have
1
ord∞ r`, `+1 −m = ((m2 − 1) − `(m − 1))
2 2
1
= (m − 1)(m − ` + 1) 6= 0
2
as 1 < m ≤ `−12
, and thus c ≡ 0 mod `.
We now argue that a ≡ ±1 mod `. We may write (am+cs)/` = am/`+bs , (a+cs) =
a/` + bs , where bs ∈ Z. Thus,
`−1
Y g(am/`,∗)
r`,γ `+1 −m =κ (81)
2
s=0
g(a/`,∗)

where κ is a constant determined by (53). By (81), we have

`−1
X ` 1
ord∞ r`,γ `+1 −m = (B2 (am/`) − B2 (a/`)) = (a2 (m2 − 1) − `a(m − 1)).
2
s=0
2 2

25
Then ord∞ r`, `+1 −m ≡ ord∞ r`,γ `+1 −m mod ` ⇔ a2 (m2 − 1) ≡ (m2 − 1) mod ` ⇔ a ≡
2 2
±1 mod `, where we use the fact that ` is prime, m 6≡ ±1 mod `. As γ is identified
with −γ under the Galois action, we have a ≡ 1 mod `.
Now d 6≡ 0 mod `, for otherwise, det(γ) ≡ 0 mod `. Thus, we find that both of the
sets {bm + ds | 0 ≤ s ≤ ` − 1} and {b + ds | 0 ≤ s ≤ ` − 1} may be used as a complete set
of representatives for Z/`Z, so for each s, there exist unique integers ts and rs using
notation as before such that 0 ≤ ts , rs ≤ ` − 1 and ts ≡ bm + ds mod `, rs ≡ b + ds
mod `. We write
1 1
`
(am + cs, bm + ds) = `
(m, ts ) + βms
1 1
`
(a + cs, b + ds) = `
(1, rs ) + β1s

where
1
βms = `
(m(a − 1) − cs, bm + ds − ts )
1
β1s = `
(a − 1 − cs, b + ds − rs ),

and βm,s , β1,s ∈ Z2 . Thus, we have

`−1
Y g(m/`, ts /`)
r`,γ `+1 −m ν
= (−1) e(µ) = (−1)ν e(µ)r`, `+1 −m
2
s=0
g(1/`, rs /`) 2

where
`−1
X
ν= (m(a−1)−cs)(bm+ds−ts )/`2 +(m(a−1)−cs)/`+(bm+ds−ts )/`
s=0
`−1
X
− (a − 1 − cs)(b + ds − rs )/`2 + (a − 1 − cs)/` + (b + ds − rs )/`
s=0
`−1
X
= (m2 − 1)(a − 1)b/`2 + (m − 1)(a − 1)ds/`2 − (a − 1)(mts − rs )/`2
s=0
− cbs(m−1)/`2 +cs(ts −rs )/`2 +(a−1)(m−1)/` + b(m−1)/` − (ts −rs )/`
= (m−1)((a−1)((m+1)b+(d−1)(`−1)/2 − bc(`−1)/2)/` + (a−1+b))+Z,

26
 `−1
1X
µ = (bm + ds − ts )m/`2 − (m(a − 1) − cs)ts /`2
2 s=0
`−1
1X
− (b + ds − rs )/`2 − (a − 1 − cs)rs /`2
2 s=0
`−1
1X
= (b(m2 −1)−(mts −rs )+ds(m−1)−(m−1)(a−1)+cs(ts −rs ))`2
2 s=0
1
= ((m − 1)((m + 1)b + (d − 1)(` − 1)/2 − (a − 1))/` + Z),
2
and
`−1
X
Z = cs(ts − rs )/`2 .
s=0
Now ts − rs = b(m − 1) + n` for some n ∈ Z, so that
Z = c(b(m − 1) + n`)(` − 1)/2`,
and c ≡ 0 mod ` implies Z ∈ Z. Thus,
(−1)ν e(µ) = (−1)Z e(Z/2)(−1)ν−Z e(µ − Z/2)
= (−1)ν−Z e(µ − Z/2)
0
= (−1)ν (m−1) e(µ0 (m − 1)/2)
where

ν 0 = (a − 1)((m + 1)b + (d − 1)(` − 1)/2 − bc(` − 1)/2)/` + (a − 1 + b)

µ0 = ((m + 1)b + (d − 1)(` − 1)/2 − (a − 1))/`.

We may
Q assume m is odd, Q`−1for otherwise, we replace m by0 ` − m which is odd, and note
that `−1 g
s=0 (m/`, s/`) = s=0 g((`−m)/`, s/`) . The quantity µ (m − 1) ∈ Z ⇔ (m + 1)b ≡
(1 − d)(` − 1)/2 mod `, as a ≡ 1 mod `. Proposition 3 now follows.

Proof. (Theorem 4.) To prove Theorem 4, for fixed ` ∈ Z+ , and any m ∈ Z+ , any
bi , ni ∈ Z, we define the product

m
Y µ ¶12`ni
σbi L
h(L) = gxi (82)
i=1
N(σbi )

27
 ³ ´
where N(σ) = det(σ), σbi = 10 b0i , and where xi ∈ 1` Z2 \ Z2 . We now show that h(L)
is modular with respect to a particular subgroup of Γ = SL2 (Z). Homogeneity and
the existence of a q-expansion follow from known properties of the Klein forms ta (51),
(52), (53) and by definition (50). In what follows, to ease notation we may write
Qm Q ³ ´ Q
αβ
i=1 = . If we take γ = γ δ ∈ Γ(` bi ), we have

Y µ
¶12`ni Y µ σb L ¶12`ni
σbi γL
h(γL) = gxi = gxi γi i
N(σbi ) N(σbi )
Y µ ¶ 12`n Y µ ¶12`ni
σbi L i
σbi L
= gxi γi = (²(xi , γi )gxi
N(σbi ) N(σbi )
Y
12`ni
= ²xi (γi ) · h(L),

³ α −1 ´
βb
where γi = γbi δi ∈ Γ(`), and ²xi (γi ) is a 2`-th root of unity (see [16], Chapter 2).
Q
We Q find that the product ²xi (γi )12`ni = 1, so that h is invariant under the action of
Γ(` bi ). By examining the q-expansion of h(L), we verifyQthat the coefficients in fact
lie in Q(ζ` ), so that h(L) is modular with respect
Q to Γ(` bi ). We now describe the
Galois action on h(L), and let G = GL2 (Z/` bi Z)/±1.
Proposition 5. Let σ ∈ G. Then the Galois action of G on h is given by
Ym µ 0 ¶
σ
σbi L 12`ni
h(L) = gxi γi0 σδ (83)
i=1
N(σb0 i )
³ ´ Q ³ ´ Q
a b 1 0 ∗
where σ = γσδ , γ = c d ∈ SL2 (Z/` i bi Z), σδ =
, δ ∈ (Z/` 0 δ i bi Z) , and
Q ³ ´
γi ∈ SL2 (Z/` i bi Z), σbi are defined by σbi = γi σbi γ = ∗0 ∗∗ .
0 0 0 0−1

Proof. (Proposition 5.) As discussed in Section 1.1, GL2 (Z/N Z) is generated by SL2 (Z/N Z)
³ ´
and the matrices γd = 10 d0 , d ∈ (Z/N Z)∗ . We now show that a decomposition

³ ´
∗ ∗
γi0−1 σbi γ = σb0 i = 0 ∗ (84)

Q Q
exists. If γ ∈ SL2 (Z/`Q bi Z), then a and c are relatively prime mod ` bi . We let
gi = g.c.d.(a, bi ) mod ` bi , and decompose a = ai gi , bi = b0i gi . Then there exist Xi and

28
Yi = yi s1 such that
Y
Yi ai + Xi b0i ≡ 1 mod ` bi (85)

where we use the fact that (bi , `) = 1. If we let Mi = gi d − Yi bc, and Ni = −Xi b, then
the matrix ³ ´
Mi Ni
Y
γi0−1 = −cb 0 a
i i
∈ SL 2 (Z/` bi Z)

gives
³ ´
gi Yi b
σb0 i = 0 b0i .

Proposition 5 follows by applying (84), the action of G on the Siegel functions (55), and
noting that N(σbi ) = N(σb0 i ).
We note that replacing x by x + a for some a ∈ Z2 changes gx by a root of unity,
12`
so that gx+a = gx12` . Thus, we may choose x = (x1 , x2 ) with 0 ≤ `xi < `. Next, we
determine Gal(F/F1 (h)).
Proposition 6. The Galois group Gal(F/F1 (h)) is given by
½ Y Y ³a ´ ¯ Q ¾
bp ¯ 1
ap ≡1 mod s1 , bp ≡0 mod gcd(a
s i b
Gal(F/F1 (h)) = σ = σp = p
cp dp ∈G¯ p ,bi ) (86)
cp ≡0 mod s2 , dp ≡1 mod s2
p p .

Assuming Proposition 6, we conclude the proof of Theorem 4. Let

n Y Y ³a ´ ¯ s
Q
b
o
∈G¯
bp 1
ap ≡1 mod s1 , bp ≡0 mod gcd(a i
Sh := σ = σp = cp
p
dp cp ≡0 mod s2 , dp ≡1 mod s2
p ,bi )

p p

denote the set given in (86). Using the notation as in Proposition S, let S = Sh , and
suppose u ∈ W . Then Eτ (u) ∈ Sh implies
hτ i h i
Eτ (up ) 1 = acppττ+bp
+dp
.

The action given in (76) also implies dp = ap − cp T(τ ), bp = −cp N(τ ), where N and T
denote the norm and trace respectively. Then

up = cp (τ − T(τ )) + ap
= cp (τ + mτ ) − cp (T(τ ) + mτ ) + ap

Now ap = dp + cp T(τ ) implies ap ≡ 1 mod s2 , as cp , dp − 1 ≡ 0 mod s2 . Since ap ≡ 1

mod s1 , we have ap ≡ 1 mod `. We use this, and also the fact that s1 | T(τ ) + mτ to
conclude up ∈ 1 + fOp , hence u ∈ 1 + fOK .

29
 Conversely, let u ∈ 1+fOK . If u = 1+s2 (τ +mτ )M +`N , then c ≡ 0 mod s2 and d =
1+s2 mτ +`N ≡ 1 mod s2 . Since a = cT(τ )+d, we find a = 1+s2 M (τ +T(τ ))+`N ≡ 1
mod s1 again using the fact that s1 | T(τ )+mτ . Finally, with hypotheses stated on N(τ )
and the fact that b = −cN(τ ), we conclude q(u) ∈ Sh . This shows that W = 1 + fOK .
We now determine kS . For any s ∈ S, one finds a determinant of the form det(s) =
1 + s1 M1 + s2 M2 + `M3 . Since (s1 , s2 ) = 1, this determinant may be made arbitrary, so
that kS = Q. We now apply Proposition 4, to conclude Kf = K(h(τ )).
Proof. (Proposition 6.) We will establish necessary and sufficient conditions
for the equality of h and hσ by comparing their divisors. We set xi =
(r1 s2 bi /`, r2 s1 /`), T xi = xi γi0 σδ , where s1 s2 = `, r1 , r2 ∈ Z, and first examine
ord∞ h − ord∞ hσ
Y ³X ´
≡ 12`2 bi ni (B2 (xi1 )/2bi − gi2 B2 (T xi1 )/2bi )
X Y ³
2
≡ 6ni ` bj (1 − gi2 )/6 + x2i1 (1 − a2 ) − xi1 (1 − gi a) + xi2 gi bi c
j6=i
´
− (xi2 bi c)2 − 2xi1 xi2 abi c
X Y ³ ´
2 2 2 2
≡ ni bj 6bi (r1 s2 ) (1 − a )−6(r2 s1 bi c) −12acbi r1 r2 s1 s2
j6=i

where the congruence is taken mod `. If we assume first that s2 6= 1, then this difference
congruent to 0 implies Y X
s2 | 6(r2 s1 )2 bj c2 bi ni
hence s2 | c2 . Since ` is square free, we must have s2 | c. We observe similarly s1 | a2 − 1.
We next compare the first Fourier coefficients in the q-expansions of h and hσ , denoted
by a(h) and a(hσ ) respectively. We find
Y 12`ni
a(h)/a(hσ ) = (e(xi2 (xi1 −1)/2)/e((Yi bB2 (T xi1 )/b0i + T xi2 (T xi1 −1))/2))
³ X³ ´ ´
= e 6` xi2 (xi1 − 1)−Yi bB2 (T xi1 )/b0i −T xi2 (T xi1 −1) ni .

Returning to the definition of xi and T xi , we see that this holds provided

X
6` (Yi b((a0i xi1 )2 +(b0i cxi2 )2 +1/6)/b0i +a0i δbXi x2i1 +b0i cδ(gi d−Yi bc)x2i2 )ni ∈ Z,
which holds if and only if
X Y Y
(Yi b bj (6aa0i b2i (r1 s2 )2 + 6bi b0i c2 (r2 s1 )2 + gi `2 )ni + 6 bj (a0i δbXi (r1 s2 )2 b2i
j6=i j
Y
+ b0i cδ(gi d 2
− Yi bc)(r2 s1 ) )ni ≡ 0 mod ` bj .
j

30
This implies Y X
s1 | bj ab(r1 s2 )2 (Yi a0i + Xi b0i δ)ni bi .
j

We now suppose s1 - b. Then there is some d1 6= 1, d1 | s1 such that

X
d1 | δ ni bi

where we return to the relation given in (85), and use the fact that s1 | Yi . Thus, s1 | b.
We note also that we must have
Y X Y
bj | Yi b bj ni gi `2
j j6=i
Q
a condition that forces bi | b.
Finally,
Q we determine conditions under which orda h = orda hσ for all cusps a of
Γ(` bi ). For each cusp a, there exists some
³ ´ Y
â b̂
γ̂a = ĉ dˆ ∈ SL2 (Z/` bi Z)

such that γa−1 ∞ = a. We first assume a 6= 0, and as above decompose â = âi gˆi , bi = bˆi gˆi ,
with gˆi = g.c.d.(â, bi ). There exists Ŷi , X̂i such that
Y
Ŷi âi + X̂i bˆi ≡ 1 mod ` bi

so that the matrix

³ aˆ ´
i X̂i b̂
γ̂i = ĉbˆi gˆi d−
ˆ Yˆi b̂ĉ

satisfies ³ ´
∗ ∗
γ̂i −1 σbi γˆa = σˆbi = 0 ∗ .

Similarly, we decompose gi â + Yi bĉ = Âi gˆi 0 , ĉbi = Ĉi gˆi 0 so that there exists a matrix
³ ´ Y
0 Âi ∗
γ̂i = Ĉi ∗ ∈ SL2 (Z/` bi Z)

satisfying
0 −1
³ ´
0 ∗ ∗
γ̂i σ γˆa = σˆbi =
b0i 0 ∗ .

31
We let
T̂ xi = xi γ̂i = (âi xi1 + ĉbˆi xi2 , ∗)
T̂ x0i = T xi γ̂i 0 = (Âi T xi1 + Ĉi T xi2 , ∗)
so that
Y
hγˆa = gT̂ xi (σˆbi (L)/N(σ̂bi ))
Y
(hσ )γˆa = gT̂ x0 (σ̂b0 i (L)/N(σ̂b0 i )).
i

Thus, we find
orda h − orda hσ
γa
= ord∞ hγa − ord∞ hσ
Y ³ X ĝ 2 gˆi 0 2 ´
i
= 12`2 bi B2 (T̂ xi1 )ni − B2 (T̂ x0i1 )ni
2bi 2bi
X Y Y
≡ 6ni bj ((r1 s2 bi â)2 + (r2 s1 bi ĉ)2 ) − 6ni bj `2 gˆi 0 2 B2 (ÂT xi1 + ĈT xi1 )
i j6=i j6=i
X Y
≡ 6ni bj ((r1 s2 bi â)2 + (r2 s1 bi ĉ)2 )
i j6=i
Y
− 6ni bj ((Âgˆi 0 `T xi1 )2 + (Ĉ gˆi 0 `T xi2 )2
j6=i
Y
+ 2ÂĈ gˆi 0 2 T xi1 T xi2 − gˆi 0 2 ÂT xi1 − gˆi 0 2 ĈT xi2 ) + bj ni `2 gˆi 0 2
j6=i
X Y
≡ 6ni bj (((r1 s2 bi â)2 + (r2 s1 bi ĉ)2 ) − ((Âgˆi 0 )2 ((r1 s2 ai bi )2 + (r2 s1 cb0i )2 )
i j6=i

+ (Ĉ gˆi 0 )2 ((δXi br1 s2 bi )2 + (δ(gi d − Yi bc)r2 s1 )2 ))

+ 2ÂĈ gˆi 0 2 ((r1 bi s2 )2 ai Xi bδ + (r2 s1 )2 b0i cδ(gi d − Yi bc)))
X Y
≡ 6ni bj ((r1 s2 bi â)2 + (r2 s1 bi ĉ)2 ) − ((gi â + Yi bĉ)2 (r1 s2 ab0i )2
i j6=i

+ (ĉb0i )2 (δr2 s1 gi d)2 )

X Y
≡ 6ni bi ((âbi r1 s2 )2 + (ĉbi s1 r2 )2 ) − ((âbi r1 s2 )2 + (ĉbi s1 r2 dδ)2 )
i j6=i
X Y
≡ 6ni bi ((âbi r1 s2 )2 (1 − (dδ)2 ))
i j6=i
Y X
≡6 bj (r2 s1 ĉ)2 (1 − (dδ)2 ) ni bi
j i

(87)

32
where the congruence is taken mod `. Now for a = 0, we let
³ ´ ³ ´
0 1 0 1
γ0 = −1 0 γ̂i = −1 0

³ ´ ³ ´
bi 0 −b0i 0
σˆbi = 0 1 σˆbi 0 = 0 −gi .

Following the above argument in this case with ĝi = 1, Ĉi = 1, bˆi = bi , âi 0 = 0, and
ĉ = −1, we find
Y X
ord0 h − ord0 hσ ≡ 6 bj (r2 s1 )2 (1 − (dδ)2 ) ni bi . (88)
j i

Assuming this sum is congruent to 0 mod `, with hypotheses as stated on `, bi and ni ,

we must have s2 | (dδ)2 − 1. Thus we may conclude that Gal(F/F1 (h)) ⊆ Sh . By (87),
we find Sh ⊆ Gal(F/F1 (h)), which proves Proposition 6.

References

[1] G. Andrews, “An analytic proof of the Rogers-Ramanujan-Gordon identities,”

Amer. J. Math. 88 (1966), 844-846.

[2] G. Andrews, “An introduction to Ramanujan’s ‘lost’ notebook,” Amer. Math.

Monthly 86 (1979), no. 2, 89-108.

[3] S. Bettner and R. Schertz, “Lower powers of elliptic units,” J. Théor. Nombres
Bordeaux 13 (2001) no. 2, 339-351.

[4] W. Duke, “Continued Fractions and Modular Functions,” Bull. Amer. Math. Soc.
(N.S.) 42 (2005), no. 2, 137-162.

[5] H. M. Farkas and I. Kra, Theta constants, Riemann surfaces and the modular group.
Graduate Studies in Mathematics, 37. American Mathematical Society, Providence,
RI, (2001).

[6] A. Folsom, “Modular Units,” UCLA Ph.D. Thesis, 2006.

[7] S. Gupta and D. Zagier, “On the coefficients of the minimal polynomials of Gaussian
periods,” Math. Comp., 60 (1993), no. 201, 385-398.

[8] N. Ishida, “Generators and equations for modular function fields of principal con-
gruence subgroups,” Acta Arith. 85 (1998), no. 3, 197-207.

[9] D. S. Kubert and S. Lang, “Units in the modular function field. I,” Math. Ann. 218
(1975), no. 1, 67-96.

33
[10] D. S. Kubert and S. Lang, “Units in the modular function field. II. A full set of
units,” Math. Ann. 218 (1975), no. 2, 175-189

[11] D. S. Kubert and S. Lang, “Units in the modular function field. III. Distribution
relations,” Math. Ann. 218 (1975), no. 3, 273-285.

[12] D. S. Kubert and S. Lang, “Units in the modular function field,” Modular functions
of one variable V. (Proc. Second Internat. Conf., Univ. Bonn, Bonn, 1976), Lecture
Notes in Math., Vol. 601, Springer, Berlin, (1977), 247-275.

[13] D. S. Kubert and S. Lang, “Distributions on toroidal groups,” Math. Z. 148 (1976),
no. 1, 33-51.

[14] D. S. Kubert and S. Lang, “Units in the modular function field. IV. The Siegel
functions are generators,” Math. Ann. 227 (1977), no. 3, 223-242.

[15] D. S. Kubert and S. Lang, “The p-primary component of the cuspidal divisor class
group on the modular curve X(p),” Math. Ann. 234 (1978), no. 1, 25-44.

[16] D. S Kubert and S. Lang, Modular Units, Grundlehren der Mathematischen

Wissenschaften [Fundamental Principles of Mathematical Science], 244. Springer-
Verlag, New York-Berlin, (1981).

[17] S. Lang, Elliptic Functions. Addison-Wesley Publishing Co., Reading, MA, (1973).

[18] K. Ramachandra, “Some applications of Kronecker’s limit formulas,” Ann. of Math.

(2) 80, (1964) 104-148.

[19] L.J. Rogers, “Second Memoir on the Expansion of Certain Infinite Products,” Proc.
London Math. Soc. 25 (1894), 318-343.

[20] L.J. Rogers and S. Ramanujan, “Proof of certain identities in combinatory analysis,”
Cambr. Phil. Soc. Proc. 19 (1919), 211-216.

[21] R. Schertz, “Construction of Ray Class Fields by Elliptic Units,” J. Théor. Nombres
Bordeaux 9 (1997), no. 2, 383-394.

[22] I. Schur, “Ein Beitrag zur additiven Zahlentheorie und zur Theorie der Ket-
tenbrüche,” Sitzungsber. Preuss. Akad. Wiss. Phys.-Math. Klasse (1917) 302-321.

[23] A. Selberg, “Über einge aritheoremetische Identitäten,” Avh. Norske Vidensk.

Akad. Oslo, I 1936, Nr. 8, 23 S.(1936); reprinted in Collected Papers, Vol. I,
Springer-Verlag, Berlin, (1989).

34
[24] G. Shimura, Introduction to the Aritheoremetic Theory of Automorphic Functions.
Publications of the Mathematical Society of Japan, 11. Iwanami Shoten, Publishers,
and Princeton University Press, (1971).

[25] J.J. Sylvester, “On certain ternary cubic equations” Amer. J. Math., v. 2 (1879),
357-381; reprinted in Collected Papers, Vol. 3, Cambridge, (1909), 325-339.

[26] L.C. Washington, Introduction to Cyclotomic Fields, 2nd Ed. Graduate Texts in
Mathematics vol. 83, Springer Verlag, (1997).

35