Mock-modular forms of weight one - UCLA Department of Mathematics
Mock-modular forms of weight one - UCLA Department of Mathematics
Mock-modular forms of weight one - UCLA Department of Mathematics
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
MOCK-MODULAR FORMS OF WEIGHT ONE<br />
W. DUKE AND Y. LI<br />
Abstract. The object <strong>of</strong> this paper is to initiate a study <strong>of</strong> the Fourier coefficients <strong>of</strong><br />
a <strong>weight</strong> <strong>one</strong> mock-<strong>modular</strong> form and relate them to the complex Galois representations<br />
associated to the form’s shadow, which is a <strong>weight</strong> <strong>one</strong> newform. In this paper, our focus<br />
will be on <strong>weight</strong> <strong>one</strong> dihedral new<strong>forms</strong> <strong>of</strong> prime level p ≡ 3 (mod 4). In this case we give<br />
properties <strong>of</strong> the Fourier coefficients themselves that are similar to (and sometimes reduce<br />
to) cases <strong>of</strong> Stark’s conjectures on derivatives <strong>of</strong> L-functions. We also give a new <strong>modular</strong><br />
interpretation <strong>of</strong> certain products <strong>of</strong> differences <strong>of</strong> singular moduli studied by Gross and<br />
Zagier. Finally, we provide some numerical evidence that the Fourier coefficients <strong>of</strong> a mock<strong>modular</strong><br />
form whose shadow is exotic are similarly related to the associated complex Galois<br />
representation.<br />
1. Introduction<br />
Roughly speaking, a mock-<strong>modular</strong> form is the holomorphic part <strong>of</strong> a harmonic <strong>modular</strong><br />
form. For our purposes a harmonic <strong>modular</strong> form <strong>of</strong> <strong>weight</strong> k ∈ 1 2 Z is a Maass form for Γ 0(M)<br />
that is killed by the <strong>weight</strong> k Laplacian and that is allowed to have polar-type singularities<br />
in the cusps. Associated to such a form f is the <strong>weight</strong> 2 − k weakly holomorphic form<br />
(1) ξ k f(z) = 2iy k ∂¯z f(z).<br />
The operator ξ k is related to the <strong>weight</strong> k Laplacian ∆ k through the identity<br />
(2) ∆ k = ξ 2−k ξ k .<br />
A special class <strong>of</strong> harmonic <strong>forms</strong> have ξ k f holomorphic in the cusps and a Fourier expansions<br />
at ∞ <strong>of</strong> the shape<br />
(3) f(z) = ∑ n≥n 0<br />
c + (n)q n − ∑ n≥0<br />
c(n)β k (n, y)q −n .<br />
This expansion is unique and absolutely uniformly convergent on compact subsets <strong>of</strong> H, the<br />
upper half-plane. Here q = e 2πiz with z = x + iy ∈ H and β k (n, y) is given for n > 0 by<br />
β k (n, y) =<br />
∫ ∞<br />
y<br />
e −4πnt t −k dt<br />
while for k ≠ 1 we have β k (0, y) = y 1−k /(k − 1) and β 1 (0, y) = − log y. For such f the Fourier<br />
expansion <strong>of</strong> ξ k f is simply<br />
ξ k f(z) = ∑ c(n)q n .<br />
n≥0<br />
After Zagier, the function ∑ n c+ (n)q n is said to be mock-<strong>modular</strong> with shadow ∑ n≥0 c(n)qn .<br />
It is important to observe that a mock-<strong>modular</strong> form is only determined by its shadow up to<br />
the addition <strong>of</strong> a holomorphic form.<br />
Date: July 3, 2012.<br />
Supported by NSF grant DMS-10-01527.<br />
Supported by NSF.<br />
1
2 W. DUKE AND Y. LI<br />
Some (non-<strong>modular</strong>) mock-<strong>modular</strong> <strong>forms</strong> have Fourier coefficients that are well-known<br />
arithmetic functions. Let σ(n) = ∑ m|n<br />
m and H(n) be the Hurwitz class number. Then<br />
−8π ∑ ( )<br />
σ(n)q n σ(0) = −<br />
1<br />
24<br />
n≥0<br />
is mock-<strong>modular</strong> <strong>of</strong> <strong>weight</strong> k = 2 for the full <strong>modular</strong> group with shadow 1 and<br />
−16π ∑ ( )<br />
H(n)q n H(0) = −<br />
1<br />
12<br />
n≥0<br />
is mock-<strong>modular</strong> <strong>of</strong> <strong>weight</strong> 3/2 for Γ 0 (4) with shadow the Jacobi theta series θ(z) = ∑ n∈Z qn2 .<br />
(See [45]). Actually, these two examples will play a fundamental role in our work on the <strong>weight</strong><br />
<strong>one</strong> case.<br />
In general, the Fourier coefficients <strong>of</strong> mock-<strong>modular</strong> <strong>forms</strong> are not well understood. Especially<br />
in the case <strong>of</strong> <strong>weight</strong> 1/2, which includes the mock-theta functions <strong>of</strong> Ramanujan, there<br />
has been considerable progress made by Zwegers in his thesis [47], by Bringmann-Ono [6],<br />
and others (see [46] for a good exposition). The Fourier coefficients <strong>of</strong> mock-<strong>modular</strong> <strong>forms</strong><br />
<strong>of</strong> <strong>weight</strong> 1/2 whose shadows Shimura–lift to cusp <strong>forms</strong> attached to elliptic curves have also<br />
been shown to be quite interesting by Bruinier and Ono [10] and Bruinier [13]. We remark<br />
that mock-<strong>modular</strong> <strong>forms</strong> whose shadows are only weakly holomorphic are also <strong>of</strong> interest<br />
(see [17],[18]) but in this paper we only consider those with holomorphic shadows.<br />
The self-dual case k = 1 presents special features and difficulties. The Riemann-Roch<br />
theorem is without content when k = 2 − k, and the existence <strong>of</strong> cusp <strong>forms</strong> is a subtle<br />
issue. Furthermore, the infinite series representing the Fourier series <strong>of</strong> <strong>weight</strong> <strong>one</strong> harmonic<br />
Poincaré series are difficult to handle.<br />
The fact that interesting non-<strong>modular</strong> mock-<strong>modular</strong> <strong>forms</strong> <strong>of</strong> <strong>weight</strong> 1 exist follows from<br />
work <strong>of</strong> Kudla, Rapoport and Yang [30]. Suppose that M = p > 3 is a prime with p ≡ 3<br />
(mod 4) and that χ p (·) = ( · ) is the Legendre symbol. Let<br />
p<br />
(4) E 1 (z) = 1 2 H(p) + ∑ n≥1<br />
R p (n)q n<br />
be Hecke’s Eisenstein series <strong>of</strong> <strong>weight</strong> <strong>one</strong> for Γ 0 (p) with character χ p , where for n > 0<br />
(5) R p (n) = ∑ m|n<br />
χ p (m).<br />
It follows from [30] that Ẽ1(z) := ∑ n≥0 R+ p (n)q n is mock-<strong>modular</strong> <strong>of</strong> <strong>weight</strong> k = 1 with<br />
shadow E 1 (z), where for n > 0 we have<br />
(6) R p + (n) = −(log p)ord p (n)R p (n) − ∑<br />
log q(ord q (n) + 1)R p (n/q),<br />
χ p(q)=−1<br />
and where R p + (0) is a constant. 1 The associated harmonic form is constructed using the<br />
s-derivative <strong>of</strong> the non-holomorphic Hecke-Eisenstein series <strong>of</strong> <strong>weight</strong> 1. An arithmetic interpretation<br />
<strong>of</strong> its coefficients is given in [30]; it is shown that (−2R + (n) + 2 log pR p (n)) is the<br />
degree <strong>of</strong> a certain 0-cycle on an arithmetic curve made from elliptic curves with CM by the<br />
ring <strong>of</strong> integers in F = Q( √ −p).<br />
1 Explicitly, R + p (0) = −H(p)( L′ (0,χ p)<br />
L(0,χ p) + ζ′ (0)<br />
ζ(0) − c), where c = 1 2 γ + log(4π3/2 ) and γ is Euler’s constant.
MOCK-MODULAR FORMS OF WEIGHT ONE 3<br />
In this paper we will study the Fourier coefficients <strong>of</strong> mock-<strong>modular</strong> <strong>forms</strong> whose shadows<br />
are new<strong>forms</strong>. For simplicity we will continue to assume that M = p > 3 is a prime. To each<br />
A ∈ Cl(F ), the class group <strong>of</strong> F , <strong>one</strong> can associate a theta series ϑ A (z) defined by<br />
(7) ϑ A (z) := 1 + ∑<br />
q N(a) = ∑ r<br />
2 A (n)q n .<br />
n≥0<br />
a⊂O F<br />
[a]∈A<br />
Hecke showed that ϑ A (z) ∈ M 1 (p, χ p ), the space <strong>of</strong> <strong>weight</strong> <strong>one</strong> holomorphic <strong>modular</strong> <strong>forms</strong><br />
for Γ 0 (p) with character χ p . Let ψ be a character <strong>of</strong> Cl(F ) and consider g ψ (z) ∈ M 1 (p, χ p )<br />
defined by<br />
(8) g ψ (z) := ∑<br />
r ψ (n)q n .<br />
A∈Cl(F )<br />
ψ(A)ϑ A (z) = ∑ n≥0<br />
When ψ = ψ 0 is the trivial character, the form g ψ0 (z) is just E 1 (z) from (4), as a consequence<br />
<strong>of</strong> Dirichlet’s fundamental formula<br />
(9) R p (n) = ∑<br />
r A (n).<br />
A∈Cl(F )<br />
Otherwise, g ψ (z) is a newform in S 1 (p, χ p ), the subspace <strong>of</strong> M 1 (p, χ p ) consisting <strong>of</strong> cusp <strong>forms</strong>.<br />
Let H be the Hilbert class field <strong>of</strong> F with ring <strong>of</strong> integers O H . The following result shows<br />
that the Fourier coefficients <strong>of</strong> certain mock-<strong>modular</strong> <strong>forms</strong> <strong>of</strong> <strong>weight</strong> <strong>one</strong> with shadow g ψ (z)<br />
can be expressed in terms <strong>of</strong> logarithms <strong>of</strong> algebraic numbers in H.<br />
Theorem 1.1. Let p ≡ 3 (mod 4) be a prime with p > 3. Let ψ be a non-trivial character<br />
<strong>of</strong> Cl(F ), where F = Q( √ −p). Then there exists a <strong>weight</strong> <strong>one</strong> mock-<strong>modular</strong> form<br />
˜g ψ (z) = ∑ n≥n 0<br />
r + ψ (n)qn<br />
with shadow g ψ (z) such that the following hold.<br />
(i) When χ p (n) = 1 or n < − p+1 24 r+ ψ<br />
(n) equals to zero.<br />
(ii) The coefficients r + ψ<br />
(n) are <strong>of</strong> the form<br />
(10)<br />
∑<br />
r + ψ<br />
(n) = −β<br />
A∈Cl(F )<br />
ψ 2 (A) log |u(n, A)|,<br />
where β ∈ Q, u(n, A) are units in O H when n ≤ 0 and algebraic numbers in H when<br />
n > 0. In addition, β depends only on p and u(n, A) depends only on n and A.<br />
(iii) Let σ C ∈ Gal(H/F ) be the element associated to the class C ∈ Cl(F ) via Artin’s isomorphism.<br />
Then it acts on u(n, A) by<br />
(iv) N H/Q (u(n, A)) is an integer and satisfies<br />
for all non-zero integers n.<br />
σ C (u(n, A)) = u(n, AC −1 ).<br />
− β 2 log(|N H/Q(u(n, A))|) = R + p (n)<br />
Observe that parts (ii) and (iii) are very similar in form to Stark’s Conjectures for special<br />
values <strong>of</strong> derivatives <strong>of</strong> L-functions. In fact, they are consequences <strong>of</strong> known cases <strong>of</strong> his<br />
conjectures when n ≤ 0. Part (iv) can be viewed as a mock-<strong>modular</strong> version <strong>of</strong> Dirichlet’s<br />
identity (9). It is important to observe that ˜g ψ (z) is not necessarily unique.
4 W. DUKE AND Y. LI<br />
The pro<strong>of</strong> <strong>of</strong> this result relies heavily on the Rankin-Selberg method for computing heights<br />
<strong>of</strong> Heegner divisors as developed in [24], but entails the use <strong>of</strong> <strong>weight</strong> <strong>one</strong> harmonic <strong>forms</strong><br />
with polar singularities in cusps in place <strong>of</strong> <strong>weight</strong> <strong>one</strong> Eisenstein series and hence requires<br />
regularized inner products. In order to get to the individual mock-<strong>modular</strong> coefficients, it<br />
is necessary to consider <strong>modular</strong> curves <strong>of</strong> large prime level N and their Heegner divisors <strong>of</strong><br />
height zero. The relation <strong>of</strong> this part <strong>of</strong> the pro<strong>of</strong> to [24] has some independent interest.<br />
Zagier pointed out that his identity with Gross for the norms <strong>of</strong> differences <strong>of</strong> singular<br />
values <strong>of</strong> the <strong>modular</strong> j-function can be neatly expressed in terms <strong>of</strong> the coefficients R + p (n)<br />
given in (6). For simplicity, let −d < 0 be a fundamental discriminant not equal to −p. and<br />
set F ′ = Q( √ −d). As is well-known, the <strong>modular</strong> j-function is well-defined on ideal classes<br />
<strong>of</strong> F and F ′ and takes values in the rings <strong>of</strong> integers <strong>of</strong> their respective Hilbert class fields.<br />
Also, values <strong>of</strong> the j-function at different ideal classes are Galois conjugates <strong>of</strong> each other.<br />
For any A ∈ Cl(F ) define the quantity<br />
(11) a d,A := ∏<br />
A ′ ∈Cl(F ′ )<br />
(j(A) − j(A ′ )).<br />
The norm <strong>of</strong> a d,A to F is thus ∏ A∈Cl(F ) a d,A and is an ordinary integer. The result <strong>of</strong> Gross<br />
and Zagier [23, Theorem 1.3] is that this integer can be expressed in terms <strong>of</strong> R p + (n) as<br />
follows:<br />
(12) log ∏<br />
A∈Cl(F )<br />
|a d,A | 2/w d<br />
= − 1 4<br />
∑<br />
k∈Z<br />
δ(k) R + p ( pd−k2<br />
4<br />
),<br />
where w d is the number <strong>of</strong> roots <strong>of</strong> unity in F ′ and δ(k) = 2 if p|k and 1 otherwise.<br />
There are two pro<strong>of</strong>s <strong>of</strong> this factorization in [23]: <strong>one</strong> analytic and <strong>one</strong> algebraic. In fact,<br />
the algebraic approach, which is based on Deuring’s theory <strong>of</strong> supersingular elliptic curves<br />
over finite fields, gives the factorization <strong>of</strong> the ideal (a d,A ) in O H for each class A ∈ Cl(F ).<br />
To state it, suppose l is a rational prime such that χ p (l) ≠ 1. Then the ideal (l) factors in<br />
O H as<br />
(13) l = ∏<br />
l δ(l)<br />
A .<br />
A∈Cl(F )<br />
The l A ’s are primes in H above l uniquely labeled so that σ C (l A ) = l AC −1 for all C ∈ Cl(F )<br />
and complex conjugation sends l A to l A −1. Let A 0 be the principal class. It is shown in [23]<br />
that the order <strong>of</strong> a d,A0 at the place associated to the prime l A is given by<br />
∑<br />
(14) ord lA (a d,A0 ) = 1 δ(k) ∑ ( )<br />
pd−k<br />
r 2<br />
2<br />
A 2 .<br />
4l m<br />
m≥1<br />
k∈Z<br />
The Galois action then yields the prime factorization <strong>of</strong> the ideal (a d,A ) for any A.<br />
Our second main result gives a mock-<strong>modular</strong> interpretation <strong>of</strong> the individual values |a d,A |.<br />
It is convenient to give it as a twisted version <strong>of</strong> (12).<br />
Theorem 1.2. For any ˜g ψ (z) = ∑ n≥n 0<br />
r + ψ (n)qn given by Theorem 1.1 and −d < 0 any<br />
fundamental discriminant different from −p we have<br />
∑<br />
∑<br />
(15)<br />
ψ 2 (A) log |a d,A | 2/w d<br />
= − 1 δ(k) r + 4<br />
ψ ( pd−k2 )<br />
4<br />
A∈Cl(F )<br />
where a d,A is defined in (11).<br />
k∈Z
MOCK-MODULAR FORMS OF WEIGHT ONE 5<br />
Since p > 3 is a prime number, Cl(F ) has odd order H(p) and ψ 2 is only trivial when ψ is<br />
trivial. Thus we can express each logarithmic term in the sum on the left hand side <strong>of</strong> the<br />
following identity in terms <strong>of</strong> the coefficients <strong>of</strong> a mock-<strong>modular</strong> form with a theta series as<br />
shadow (see Corollary 9.3 below).<br />
As with Theorem 1.1, this is proved using the methods <strong>of</strong> [24] and not the analytic technique<br />
<strong>of</strong> [23], which uses the restriction to the diagonal <strong>of</strong> an Eisenstein series for a Hilbert <strong>modular</strong><br />
group. In particular, we will define a real-analytic function Φ(z), which trans<strong>forms</strong> with<br />
<strong>weight</strong> 3/2 and level 4, and use holomorphic projection to obtain an equation between a<br />
finite linear combination <strong>of</strong> r + ψ<br />
(n)’s and an infinite sum, similar to the <strong>one</strong> in [24]. We<br />
also make use <strong>of</strong> machinery from [29]. One interesting new feature is an elementary counting<br />
argument needed to construct a Green’s function evaluated at CM points. Actually, equation<br />
(15) is a particular example <strong>of</strong> a more general identity involving values <strong>of</strong> certain Borcherds<br />
lifts. Although we will not carry this out here, it will become clear that similar methods can<br />
be used to prove a level N version in this form.<br />
Let us illustrate our Theorems in the first non-trivial case, which occurs when p = 23.<br />
The class group Cl(F ) has size 3 and two non-trivial characters ψ and ψ. The Hilbert class<br />
field H is generated by X 3 − X − 1 over F . Let α = 1.32472 . . . be the unique real root<br />
<strong>of</strong> X 3 − X − 1, which is also the absolute value <strong>of</strong> the square <strong>of</strong> a unit in H. The space<br />
S 1 (23, χ 23 ) is <strong>one</strong> dimensional and spanned by the cusp form<br />
g ψ (z) = η(z)η(23z),<br />
where η(z) = q ∏ 1/24 n≥1 (1 − qn ) is the eta function. According to Theorem 1.1, there exists<br />
a mock-<strong>modular</strong> form ˜g ψ (z) having a simple pole and the following Fourier expansion<br />
˜g ψ (z) =<br />
∑<br />
r + ψ (n)qn .<br />
n≥−1<br />
χ 23 (n)≠1<br />
The condition on the principal part determines ˜g ψ (z) uniquely in this case, though this is<br />
not true in general. Using Stokes’ Theorem and Stark’s calculation at the end <strong>of</strong> [40, II], <strong>one</strong><br />
can show that<br />
r + ψ (−1) = 〈g ψ, g ψ 〉 = 3 log(α) and r + ψ<br />
(0) = − log(α).<br />
From the numerical calculations <strong>of</strong> r + ψ<br />
(n), we can predict the values <strong>of</strong> β and u(n, A) in<br />
Theorem 1.1. Let ˜β and ũ(n, A) denote the guessed values for β and u(n, A) respectively.<br />
The following table lists r + ψ<br />
(n), which are calculated numerically, and the guessed values<br />
˜β, ũ(n, A 0 ) for the principal class A 0 ∈ Cl(F ) for 1 ≤ n ≤ 23 and a few other values <strong>of</strong> n, all<br />
with χ 23 (n) ≠ 1. It also contains the norms <strong>of</strong> ũ(n, A 0 ), which agree with condition (iv) in<br />
Theorem 1.1.<br />
n r + ψ (n) ˜β ũ(n, A0 ) N H/Q (ũ(n, A 0 )) ˜β/2<br />
5 1.1001149692823391 2 π 5 5 2<br />
7 1.7161505040673007 2 π 7 7 2<br />
10 3.9614773685309742 2 5α −6 π5 −1 5 4<br />
11 0.052996471463740862 2 π 11 11 2<br />
14 1.6582443878082415 2 7α −4 π7 −1 7 4
6 W. DUKE AND Y. LI<br />
15 -1.9437136922512246 2 5απ5 −1<br />
5 4<br />
17 -4.2163115309750479 2 π 17 17 2<br />
19 -2.7119255841404505 2 π 19 19 2<br />
20 5.9051910607821988 2 5α −7 5 6<br />
21 6.7198367256215547 2 7α −10 π −1<br />
7 7 4<br />
22 -4.2709900863081686 2 11α 5 π −1<br />
11 11 4<br />
23 -3.8460181706191355 2 π 23 23<br />
28 4.2179936148444277 2 7α −5 7 6<br />
34 -2.532478252776036 2 17α 8 π −1<br />
17 17 4<br />
38 -9.942055260392833 2 19α 15 π −1<br />
19 19 4<br />
40 -14.92826076712649 2 5α 19 π 5 5 8<br />
Table 1. Coefficients <strong>of</strong> ˜g ψ (z) for g ψ (z) = η(z)η(23z) ∈ S 23 (p)<br />
Here, π l is a generator <strong>of</strong> the prime ideal l A0 in O H and is given below in terms <strong>of</strong> α.<br />
l<br />
π l<br />
5 2α 2 − α − 1<br />
7 α 2 + α − 2<br />
11 2α 2 − α<br />
17 2α 2 + 3α + 3<br />
19 3α 2 + α<br />
1<br />
23 √ −23<br />
(10α 2 + 8α + 1)<br />
Table 2. Generators <strong>of</strong> l A0 .<br />
To illustrate how Theorem 1.2 supports these predictions, consider the classical example<br />
−d = −7, which also appeared in [43]. First, we can combine equation (15) and Theorem<br />
1.1 to write (see Corollary 9.3)<br />
4 log |a d,A | 2/w d<br />
= β ∑ ( )<br />
∣ pd−k<br />
δ(k) log ∣u<br />
2<br />
, A<br />
4<br />
∣ .<br />
k∈Z<br />
For d = 7, p = 23 and A = A 0 , we can use the predicted values in Table 1 to rewrite the<br />
equation above as<br />
( )<br />
log |u(40, A0 )| + log |u(38, A 0 )| + log |u(34, A 0 )|+<br />
4 log |a 7,A0 | = 4<br />
log |u(28, A 0 )| + log |u(20, A 0 )| + log |u(10, A 0 )|<br />
= 4 log |5 3 · 7 · 17 · π17 −1 · 19 · π19 −1 · α 24 |.<br />
This agrees with the exact <strong>of</strong> value <strong>of</strong> a 7,A0 , which is 5 3 · 7 · 17 · π17 −1 · 19 · π19 −1 · α 24 . Applying<br />
the Galois action to these predicted u(n, A 0 ) shows that they also agree with Theorem 1.2.<br />
This and several other numerical examples, together with equation (14), motivate us to<br />
make a conjecture about the factorization <strong>of</strong> the fractional ideal generated by u(n, A) in<br />
Theorem 1.1. It is not hard to see that the following conjecture implies (14).<br />
Conjecture. In (ii) <strong>of</strong> Theorem 1.1 we have the following.<br />
(i) The number 2/β is an integer dividing 24H(p)h H , where h H is the class number <strong>of</strong> H.<br />
(ii) For B ∈ Cl(F ), let l B be a prime ideal above the rational prime l as in (13). Then the<br />
order <strong>of</strong> u(n, A) at the place <strong>of</strong> H corresponding to l B is<br />
∑ ( n<br />
)<br />
ord lB (u(n, A)) = 2 r<br />
β (A −1 B) 2 .<br />
l m<br />
m≥1
MOCK-MODULAR FORMS OF WEIGHT ONE 7<br />
At all other places <strong>of</strong> H, u(n, A) is a unit. In particular, u(n, A) ∈ O H for all n, A.<br />
In general (for prime level p ≡ 3 (mod 4)) , the Deligne-Serre Theorem [15] (see also [36])<br />
identifies the L-function <strong>of</strong> a newform f ∈ S 1 (p, χ p ) with the Artin L-function <strong>of</strong> an irreducible,<br />
odd, two-dimensional, complex representation ρ f <strong>of</strong> the Galois group <strong>of</strong> a normal<br />
extension K/Q. Such a ρ f gives rise to a projective representation ˜ρ f : Gal(K/Q) → PGL 2 (C)<br />
whose image is isomorphic to D 2n , S 4 , or A 5 , in which case we say that f is dihedral, octahedral,<br />
or icosahedral, respectively. The <strong>forms</strong> g ψ (z) are precisely the dihedral <strong>forms</strong> and for<br />
them K = H. Since g ψ1 (z) = g ψ2 (z) if and only if ψ 1 = ψ 2 or ψ 1 = ψ 2 , there are exactly<br />
(H(p) − 1)/2 such <strong>forms</strong>. Theorem 1.1 relates the Fourier coefficients <strong>of</strong> a mock-<strong>modular</strong><br />
form with dihedral shadow g ψ (z) to linear combinations <strong>of</strong> logarithms <strong>of</strong> algebraic numbers<br />
in the number field K = H determined by the Galois representation. Furthermore, the coefficients<br />
in these linear combinations are algebraic numbers in the field generated by the<br />
Fourier coefficients <strong>of</strong> the shadow.<br />
Non-dihedral new<strong>forms</strong> are <strong>of</strong>ten referred to as being “exotic” since their occurrence is<br />
rare and unpredictable. We have carried out some numerical calculations for mock-<strong>modular</strong><br />
<strong>forms</strong> whose shadows are certain exotic new<strong>forms</strong> and have observed that they also seem to<br />
be related to linear combinations <strong>of</strong> logarithms <strong>of</strong> algebraic numbers in the number field K<br />
determined by the associated Galois representation. For instance, when p = 283 the space<br />
S 1 (283, χ 283 ) contains a pair <strong>of</strong> octahedral new<strong>forms</strong> associated to a Galois representation <strong>of</strong><br />
Gal(K/Q), where K is the degree 2 extension <strong>of</strong> the normal closure <strong>of</strong> Q[X]/(X 4 − X − 1)<br />
with Gal(K/Q) ∼ = GL 2 (F 3 ). Similar to the dihedral case, the Petersson norms <strong>of</strong> these<br />
new<strong>forms</strong> come from the residue <strong>of</strong> a certain degree four L-function. As a consequence <strong>of</strong><br />
proven cases <strong>of</strong> Stark’s conjecture, this residue is the logarithm <strong>of</strong> a unit inside a subfield F 6<br />
<strong>of</strong> K. This determines the principal part <strong>of</strong> a mock <strong>modular</strong> form whose shadow is <strong>one</strong> <strong>of</strong><br />
these octahedral new<strong>forms</strong>. Using this hint and some numerical calculations, we noticed that<br />
the other coefficients also seem to be linear combinations <strong>of</strong> logarithms <strong>of</strong> algebraic numbers<br />
in the same subfield F 6 . Furthermore, these algebraic numbers generate ideals with nice<br />
factorizations. Thus, it is natural to expect a statement analogous to Theorem 1.1 and the<br />
Conjecture should hold for exotic new<strong>forms</strong>. In the final section <strong>of</strong> the paper we will give<br />
computational details when p = 283.<br />
We end the introduction with a brief outline <strong>of</strong> the paper. The first two sections concern<br />
general properties <strong>of</strong> <strong>weight</strong> <strong>one</strong> mock-<strong>modular</strong> <strong>forms</strong>: in §2 we deal with the existence <strong>of</strong><br />
<strong>weight</strong> <strong>one</strong> mock <strong>modular</strong> <strong>forms</strong>, while in §3 we will decompose the space <strong>of</strong> <strong>weight</strong> <strong>one</strong><br />
harmonic Maass <strong>forms</strong> into a plus space and minus space. In §4 we study the principal part<br />
at infinity <strong>of</strong> a mock-<strong>modular</strong> form with shadow g ψ (z). The pro<strong>of</strong> <strong>of</strong> Theorem 1.1 is in §5 −<br />
§7. In §5 and §6 , we will prove Theorem 5.1 and Proposition 6.1, which imply that different<br />
types <strong>of</strong> linear combinations <strong>of</strong> r + ψ<br />
(n)’s can be put into the form <strong>of</strong> the right hand side <strong>of</strong><br />
equation (10). In §7 we verify that each individual r + ψ<br />
(n) can be solved from these linear<br />
combinations. Theorem 1.2 is proved in §8 and §9. Finally, an analysis <strong>of</strong> the specific case<br />
p = 283 is given in §10.<br />
Acknowledgements. We thank Ö. Imamoḡlu for a remark made to <strong>one</strong> <strong>of</strong> us some time ago<br />
that the self-dual case <strong>of</strong> <strong>weight</strong> <strong>one</strong> mock <strong>modular</strong> <strong>forms</strong> should be especially interesting.<br />
Also, we thank J. Bruinier, S. Ehlen, A. Venkatesh, M. Viazovska, D. Zagier and S. Zwegers<br />
for their comments. After we obtained the results <strong>of</strong> this paper, we learned that there is<br />
some intersection with [42] and [19].
8 W. DUKE AND Y. LI<br />
2. Existence <strong>of</strong> Harmonic Maass Forms <strong>of</strong> Weight One<br />
Given a cusp form <strong>of</strong> <strong>weight</strong> <strong>one</strong>, we will show the existence <strong>of</strong> a <strong>weight</strong> <strong>one</strong> mock-<strong>modular</strong><br />
form with it as shadow in this section. There are several different approaches to proving this<br />
result, such as a geometric approach in [9]. For completeness and since it might have some<br />
independent interest, in this section we prove the existence by analytically continuing <strong>weight</strong><br />
<strong>one</strong> Poincaré series via the spectral expansion.<br />
We begin with some basic definitions. Let k ∈ Z. For any function f : H → C and<br />
γ ∈ GL + 2 (R), define the <strong>weight</strong> k slash operator | k γ by<br />
(f| k γ)(z) := (det(γ))k/2 f(γz),<br />
(cz+d) k<br />
where γz is the linear fractional transformation <strong>of</strong> z under γ. We will write f|γ for f| k γ<br />
when the <strong>weight</strong> <strong>of</strong> f is understood. For M ∈ Z + let Γ 0 (M) denote the usual congruence<br />
subgroup <strong>of</strong> SL 2 (Z) <strong>of</strong> level M, namely<br />
Γ 0 (M) = {( a c d b ) ∈ SL 2(Z) ( a c d b ) ≡ ( ∗ 0 ∗<br />
∗ ) mod M}.<br />
Let ν : Z/MZ → C × be a Dirichlet character such that ν(−1) = (−1) k and ν(γ) := ν(d) for<br />
γ = ( a c d b ) ∈ Γ 0(M). Let F k (M, ν) be the space <strong>of</strong> smooth functions f : H → C such that for<br />
all γ ∈ Γ 0 (M)<br />
(f | k γ)(z) = ν(γ)f(z).<br />
Recall from (1) and (2) the differential operator ξ k and the <strong>weight</strong> k hyperbolic Laplacian<br />
∆ k . Let z = x + iy. Then ∆ k can be written as<br />
−∆ k = −y 2 (<br />
∂ 2<br />
+ ∂2<br />
∂x 2<br />
) (<br />
+ iky<br />
∂y 2<br />
∂<br />
+ i ∂<br />
∂x ∂y<br />
Then f(z) ∈ F k (M, ν) is a <strong>weight</strong> k harmonic weak Maass form <strong>of</strong> level M and character ν<br />
(or more briefly, a weakly harmonic form) if it satisfies the following properties.<br />
(i) f(z) is real-analytic.<br />
(ii) ∆ k (f) = 0.<br />
(iii) The function f(z) has at most linear exp<strong>one</strong>ntial growth at all cusps <strong>of</strong> Γ 0 (M).<br />
From now on we fix k = 1 and write | for the slash operator | 1 . Let H ! 1(M, ν) be the<br />
space <strong>of</strong> weakly harmonic <strong>modular</strong> <strong>forms</strong> <strong>of</strong> <strong>weight</strong> <strong>one</strong>, level M and character ν. Denote<br />
by M ! 1(M, ν) , M 1 (M, ν) and S 1 (M, ν) the usual subspaces <strong>of</strong> weakly holomorphic <strong>modular</strong><br />
<strong>forms</strong>, holomorphic <strong>modular</strong> <strong>forms</strong> and cusp <strong>forms</strong>, respectively. Let M ! 1(M, ν) be the space<br />
<strong>of</strong> mock-<strong>modular</strong> <strong>forms</strong> with shadows in M 1 (M, ν) and M 1 (M, ν) ⊂ M ! 1(M, ν) the subspace<br />
<strong>of</strong> mock-<strong>modular</strong> <strong>forms</strong> whose shadows are in S 1 (M, ν).<br />
From (2), it is clear that the image <strong>of</strong> H ! 1(M, ν) under ξ 1 lies in M ! 1(M, ν). Let H 1 (M, ν)<br />
be the preimage <strong>of</strong> S 1 (M, ν) under ξ 1 . Then H 1 (M, ν) is canonically isomorphic to M 1 (M, ν).<br />
We want to show that the map ξ 1 is in fact a surjection onto S 1 (M, ν).<br />
To proceed, we will construct two families <strong>of</strong> Poincaré series P m (z, s), Q m (z, s), where the<br />
first family is similar to the <strong>one</strong> used in [6]. They are a priori defined for Re(s) > 1 and will<br />
be analytically continued to Re(s) > 0 through their spectral expansions. Unlike the cases<br />
k ≥ 2, this will only be a statement about existence, and not a formula that could be used<br />
to calculate the holomorphic coefficients <strong>of</strong> the preimage explicitly. To prove the analytic<br />
continuation, we will refer to results in [31] and [35].<br />
Given any positive integer m, define<br />
(16)<br />
(17)<br />
)<br />
.<br />
φ ∗ m(z, s) := e 2πimx (4π|m|y) −1/2 M sgn(m)<br />
,s− 1 (4π|m|y),<br />
2 2<br />
ϕ ∗ m(z, s) := e 2πimx (4π|m|y) s−1/2 e −2π|m|y .
MOCK-MODULAR FORMS OF WEIGHT ONE 9<br />
Here M µ,ν (y) is the M-Whittaker function defined by<br />
(18) M µ,ν (z) := z ν+1/2 e z/2 1F 1<br />
( 1<br />
2 + µ + ν; 1 + 2ν; −z) ,<br />
with 1 F 1 (α; β; z) being the generalized hypergeometric function. Averaging them over the<br />
coset representatives <strong>of</strong> Γ ∞ \Γ 0 (M), we can define the following Poincaré series<br />
∑<br />
P m (z, s, ν) :=<br />
ν(γ)(φ ∗ m | γ)(z, s),<br />
Q m (z, s, ν) :=<br />
γ∈Γ ∞\Γ 0 (M)<br />
∑<br />
γ∈Γ ∞\Γ 0 (M)<br />
ν(γ)(ϕ ∗ m | γ)(z, s).<br />
For convenience, we shall omit ν and write P m (z, s) and Q m (z, s) instead. When Re(s) ><br />
1, both families have Fourier expansions in z that converges absolutely and uniformly on<br />
compact subset <strong>of</strong> H (see [31, Satz 6.5] for the <strong>weight</strong> zero case). In that region, both<br />
families are absolutely convergent and define a holomorphic function in s. Also, P m (z, s) is<br />
an eigenfunction <strong>of</strong> −∆ 1 with eigenvalue (s − 1/2)((1 − s) − 1/2) and belongs to F 1 (M, ν).<br />
Let ∆ ′ k be the differential operator defined by<br />
( )<br />
−∆ ′ k := −y 2 ∂ 2<br />
+ ∂2 + iky ∂ .<br />
∂x 2 ∂y 2 ∂x<br />
It is related to ∆ k by the following equation<br />
Define the space D 1 (M, ν) by<br />
∆ ′ k + k 2<br />
(<br />
1 −<br />
k<br />
2)<br />
= y k/2 ∆ k y −k/2 .<br />
D 1 (M, ν) := {y 1/2 f(z) : f(z) ∈ F 1 (M, ν) is smooth with compact support on H}<br />
Let ˜D 1 (M, ν) be the completion <strong>of</strong> D 1 (M, ν) with respect to the Petersson norm<br />
∫<br />
||g|| 2 = 〈g, g〉 := |g(z)| 2 dxdy .<br />
y 2<br />
Γ 0 (M)\H<br />
Satz 3.2 in [35] implies that −∆ ′ k has a self-adjoint extension − ˜∆ ′ k from D 1(M, ν) to ˜D 1 (M, ν).<br />
Also, − ˜∆ ′ k has a countable system <strong>of</strong> Maass cusp <strong>forms</strong> {e n(z)} n∈N ⊂ ˜D 1 (M, ν) forming the<br />
discrete spectrum. Each Maass cusp form e n (z) has eigenvalue λ n and each eigenvalue has<br />
finite multiplicity. For any f ∈ ˜D 1 (M, ν), the discrete spectrum contributes the following<br />
sum to its spectral expansion<br />
∞∑<br />
〈f, e n 〉e n (z),<br />
n=1<br />
which converges absolutely and uniformly for all z ∈ H [35, Satz 8.1].<br />
Let ι be a cusp <strong>of</strong> Γ 0 (M), σ ι ∈ GL 2 (R) the scaling matrix sending ∞ to ι, and Γ ι ⊂ Γ 0 (M)<br />
the subgroup fixing ι. One can define the <strong>weight</strong> <strong>one</strong> real analytic Eisenstein series E ι (z, s)<br />
by<br />
E ι (z, s) := y1/2<br />
2<br />
∑<br />
γ∈Γ ι\Γ 0 (M)<br />
ν(γ)<br />
cz+d (Im(σ−1 ι γz)) s−1/2 .<br />
Selberg showed that the Eisenstein series has meromorphic continuation to the whole complex<br />
plane in s. For ι ranging over the cusps <strong>of</strong> Γ 0 (M), the Eisenstein series {E ι (z, s)} ι make up
10 W. DUKE AND Y. LI<br />
the continuous spectrum <strong>of</strong> − ˜∆ ′ 1. So for any f ∈ ˜D 1 (M, ν), the contribution <strong>of</strong> the Eisenstein<br />
series to the spectral expansion <strong>of</strong> f is<br />
1<br />
4π<br />
∑<br />
ι<br />
∫ ∞<br />
−∞<br />
〈f(·), E ι (·, 1 2 + ir)〉E ι(z, 1 2 + ir)dr.<br />
If f is smooth, then the integral in r converges absolutely and uniformly for z in any fixed<br />
compact subset <strong>of</strong> H [35, Satz 12.3]. Applying the completeness theorem [35, Satz 7.2 ], we<br />
have the spectral expansion for any smooth f ∈ ˜D 1 (M, ν) in the form<br />
∞∑<br />
(19) f(z) = e n 〉e n (z) +<br />
n=1〈f, ∑ ι<br />
1<br />
4π<br />
∫ ∞<br />
−∞<br />
〈f(·), E ι (·, 1 2 + ir)〉E ι(z, 1 2 + ir)dr,<br />
where the sum over n ∈ N and the integral in r both converge uniformly and absolutely for<br />
z in any compact set <strong>of</strong> H.<br />
By setting f(z) = y 1/2 Q m (z, s) and comparing P m (z, s) to Q m (z, s) as in [31], we can<br />
deduce the following proposition.<br />
Proposition 2.1. For any positive integer m, the Poincaré series Q m (z, s) and P m (z, s)<br />
have analytic continuation to Re(s) > 0. At s = 1/2, P m (z, s) has at most a simple pole.<br />
Furthermore, the residues <strong>of</strong> P m (z, s) at s = 1/2 generate S 1 (M, ν).<br />
Pro<strong>of</strong>. First, we will prove the analytic continuation Q m (z, s). For any m > 0, the function<br />
˜Q m (z, s) := y 1/2 Q m (z, s) is square integrable for Re(s) > 1, hence contained in ˜D 1 (M, ν). So<br />
we can write out its spectral expansion<br />
(20)<br />
˜Q m (z, s) = D(z, s) + C(z, s),<br />
D(z, s) :=<br />
∞∑<br />
〈 ˜Q m (·, s), e n (·)〉e n (z),<br />
n=1<br />
C(z, s) := 1<br />
4π<br />
∑<br />
ι<br />
∫ ∞<br />
−∞<br />
〈 ˜Q m (·, s), E ι (·, 1 2 + ir)〉E ι(z, 1 2 + ir)dr,<br />
where D(z, s) and C(z, s) are the contributions from the discrete and continuous spectrum<br />
<strong>of</strong> − ˜∆ ′ 1 respectively.<br />
By Satz 5.2 and Satz 5.5 in [35], e n (z) has eigenvalue λ n ∈ [1/4, ∞) under − ˜∆ ′ 1. If<br />
λ n = 1/4, then y −1/2 e n (z) is in S 1 (M, ν). Since each eigenvalue has finite multiplicity, we<br />
can let R ≥ 0 such that {y −1/2 e n (z) : 1 ≤ n ≤ R} is an orthonormal basis <strong>of</strong> S 1 (M, ν). Note<br />
S 1 (M, ν) could be empty, in which case R = 0.<br />
For n ∈ N, let t n = √ λ n − 1/4, s n = 1/2 + it n . We can use equation (66) in [20], the<br />
bound M µ,ν (y) = O µ,ν (e y ) as y → ∞ and the vanishing property <strong>of</strong> cusp <strong>forms</strong> at all cusps<br />
to write<br />
e n (z) =<br />
∞∑<br />
u=−∞,u≠0<br />
c n,u W sgn(u)<br />
,s 1 (4π|u|y)e 2πiux ,<br />
2 n−<br />
2<br />
where W µ,ν (z) is the W -Whittaker function and c n,u are constants. Note if t n = 0, i.e.<br />
y −1/2 e n (z) is a holomorphic cusp form, then W sgn(u)/2,sn−1/2(4π|u|y) = (4π|u|y) 1/2 e −2π|u|y and
MOCK-MODULAR FORMS OF WEIGHT ONE 11<br />
c n,u = 0 for u ≤ 0. Applying the Rankin-Selberg unfolding trick to 〈 ˜Q m (·, s), e n (·)〉 gives us<br />
∫ ∞<br />
〈 ˜Q m (·, s), e n (·)〉 = (4π|m|) 1/2 e −2π|m|y (4π|m|y) s−1 c n,m W sgn(m)<br />
0<br />
= (4π|m|) 1/2 Γ ( s − 1<br />
c − it ) (<br />
2 n Γ s −<br />
1<br />
+ it )<br />
2 n<br />
(21)<br />
n,m ( )<br />
Γ s − sgn(m)<br />
2<br />
2 ,s n− 1 2<br />
(4π|m|y) dy<br />
y<br />
The last step uses the Mellin transform <strong>of</strong> the W -Whittaker function [2, eq. (8b)] and<br />
the substitution s n = 1/2 + it n . When Re(s) > 1, the sum defining D(z, s) is absolutely<br />
convergent [35, Satz 8.1 ], since y 1/2 Q m (z, s) ∈ ˜D 1 (M, ν) for Re(s) > 1. When 0 < Re(s) ≤ 1,<br />
we can write D(z, s) as<br />
D(z, s) = (4π|m|) ∑ 1/2 Γ ( s + 1 − 1<br />
c − it ) (<br />
2 n Γ s + 1 −<br />
1<br />
+ it )<br />
2 n s − sgn(m)<br />
n,m (<br />
)<br />
2<br />
e n (z) ·<br />
n∈N<br />
Γ s + 1 − sgn(m)<br />
(s − 1<br />
2<br />
2 )2 + t 2 n<br />
Since t 2 n = λ n − 1/4 and ∑ n>M λ−2 n converges [35, Satz 8.1], we can apply Cauchy-Schwarz<br />
inequality to see that the sum on the right hand side above converges absolutely on compact<br />
subsets <strong>of</strong> {s ∈ C : Re(s) > 0, s ≠ 1/2}. At s = 1/2, the first R terms in the sum produce a<br />
simple pole since t n = 0 for all 1 ≤ n ≤ R. The rest <strong>of</strong> the sum still converges absolutely. So<br />
the right hand side above gives the analytic continuation <strong>of</strong> D(z, s) to Re(s) > 0.<br />
For the continuous spectrum, the contribution C(z, s) can be treated similarly. For any<br />
cusp ι, we can write the Fourier expansion <strong>of</strong> E ι (z, s) at infinity in the following form (for<br />
ι = ∞, see [20, eq. (76)’])<br />
E ι (z, s) = y s + ψ ι (s)y 1−s + ∑ ψ ι,m (s)W sgn(m)<br />
,s− 1 (4π|m|y)e 2πimx ,<br />
2 2<br />
m≠0<br />
where ψ ι (s) and ψ ι,m (s) are products <strong>of</strong> gamma factors and Selberg-Kloosterman zeta functions.<br />
It is well-known that The Eisenstein series can be analytically continued in s to the<br />
whole complex plane. When Re(s) > 1/2, the poles <strong>of</strong> E ι (z, s) are in the interval s ∈ (1/2, 1]<br />
[35, Satz 10.3 ]. On the line Re(s) = 1/2, E ι (z 0 , s) is holomorphic in s for any fixed z 0 ∈ H<br />
[35, Satz 10.4]. So both ψ ι (s) and ψ ι,m (s) admit analytic continuation to Re(s) > 0 and are<br />
holomorphic on Re(s) = 1/2. Using the same unfolding trick above, we can evaluate<br />
〈 ˜Q m (·, s), E ι (·, 1 + ir)〉 = ψ ( 1 2 ι,m + ir) (4π|m|) 1/2 Γ ( s − 1 − ir) Γ ( s − 1 2 ( ) + ir) 2 .<br />
2<br />
Γ s − sgn(m)<br />
2<br />
Then C(z, s) can be written as<br />
C(z, s) = ∑ ι<br />
(22)<br />
= ∑ ι<br />
∫<br />
1 ∞<br />
4π −∞<br />
∫<br />
1 ∞<br />
4π −∞<br />
(<br />
ψ 1<br />
ι,m + ir) (4π|m|) 1/2 Γ ( s − 1 − ir) Γ ( s − 1 2<br />
( ) + ir) 2<br />
E<br />
2 ι (z, 1 + ir)dr<br />
Γ s − sgn(m)<br />
2<br />
2<br />
(<br />
ψ 1<br />
ι,m + ir) (4π|m|) 1/2 Γ ( s + 1 − 1 − ir) Γ ( s + 1 − 1<br />
(<br />
2<br />
)<br />
+ ir) 2<br />
2<br />
Γ s + 1 − sgn(m)<br />
2<br />
(s − sgn(m) )<br />
2<br />
·<br />
(s − 1 − ir)(s − 1 + ir) · E ι(z, 1 + ir)dr.<br />
2<br />
2 2
12 W. DUKE AND Y. LI<br />
When Re(s) > 0, C(z, s + 1) is absolutely convergent. So when Re(s) ≠ 1/2, we can apply<br />
Cauchy-Schwarz to bound expression (22) by<br />
∫ ∣<br />
∞<br />
s − sgn(m) ∣∣∣∣<br />
2<br />
(23)<br />
∣(s − 1 dr.<br />
2 )2 + r 2<br />
−∞<br />
When Re(s) > 1/2, let ˜C(z, s) be the expression (22), which gives the analytic continuation<br />
<strong>of</strong> C(z, s) in this region. When Re(s) < 1/2, we can define ˜C(z, s) in a similar fashion as in<br />
§6 <strong>of</strong> [31]. We start with (22) and Re(s) ∈ (1/2, 1/2 + ɛ) for some small ɛ > 0, then deform<br />
the contour so that it goes above 1/2−s and below s−1/2 . In the process, the following residue<br />
i<br />
i<br />
is picked up<br />
(24)<br />
(π|m|) 1/2 Γ(2s − 1)<br />
( ) ∑<br />
Γ s − sgn(m)<br />
2<br />
ι<br />
ψ ι,m (s)E ι (z, s) + ψ ι,m (1 − s)E ι (z, 1 − s).<br />
Finally we can reduce the real part <strong>of</strong> s to less than 1/2 and change the contour back to<br />
the real line. So when Re(s) < 1/2, let ˜C(z, s) be the sum <strong>of</strong> (24) and (22), which are<br />
holomorphic. The bound (23) and m ≥ 1 guarantees the existence <strong>of</strong> the limits <strong>of</strong> ˜C(z, s) as<br />
Re(s) approaches 1/2 from the left and from the right. The procedure defining ˜C(z, s) shows<br />
that the limits agree. So we define ˜C(z, s) on Re(s) = 1/2 to be this limit. Then ˜C(z, s) gives<br />
the analytic continuation <strong>of</strong> C(z, s) to Re(s) > 0. Putting these together with the analytic<br />
continuation <strong>of</strong> D(z, s), we have the analytic continuation <strong>of</strong> Q m (z, s) to Re(s) > 0.<br />
Now to analytically continue P m (z, s), we can simply compare it to Q m (z, s). Applying<br />
the power series expansions <strong>of</strong> 1 F 1 (α; β; z) and the exp<strong>one</strong>ntial map to (16) and (17), we see<br />
that for y small and 0 < Re(s) < 2,<br />
|(4π|m|y) −s+1/2 (φ ∗ m(z, s) − ϕ ∗ m(z, s))| = O m (y).<br />
So the difference P m (z, s) − Q m (z, s) defined by term-wise subtraction is a holomorphic function<br />
in s for Re(s) > 0. That means the analytic continuation <strong>of</strong> P m (z, s) to Re(s) > 0<br />
follows from that <strong>of</strong> Q m (z, s). Furthermore, they have the same poles and residues in the<br />
region Re(s) > 0. Thus, to prove the second half <strong>of</strong> the proposition, it suffices to analyze the<br />
poles and residues <strong>of</strong> Q m (z, s) at s = 1/2.<br />
Since m > 0, expression (23) is bounded by an absolute constant for all s ∈ (1/2, 2]. So<br />
˜C(z, s) does not contribute to the pole at s = 1/2. For a Maass cusp form e n (z), the right<br />
hand side <strong>of</strong> (21) vanishes at s = 1/2 if t n ≠ 0. Otherwise, y −1/2 e n (z) ∈ S 1 (M, ν) and<br />
〈 ˜Q m (·, s), e n (·)〉 has a simple pole <strong>of</strong> residue (4π|m|) 1/2 c n,m . Thus, we have<br />
∑ R<br />
(25) Res s=1/2 P m (z, s) = Res s=1/2 Q m (z, s) = (4π|m|) 1/2 c n,m y −1/2 e n (z).<br />
Since {y −1/2 e n (z) : n = 1, . . . , R} is a basis <strong>of</strong> S 1 (M, ν), the matrix<br />
{c n,m : 1 ≤ n ≤ R, 1 ≤ m ≤ K}<br />
has rank equals to R for K sufficiently large. Therefore the residues at s = 1/2 <strong>of</strong> P m (z, s)<br />
generate S 1 (M, ν).<br />
□<br />
The following theorem is an immediate consequence <strong>of</strong> the proposition above.<br />
Theorem 2.2. Using the notations above, the following map is a surjection<br />
ξ 1 : H 1 (M, ν) → S 1 (M, ν),<br />
n=1
MOCK-MODULAR FORMS OF WEIGHT ONE 13<br />
i.e. for any cusp form h(z) ∈ S 1 (M, ν), there exists ˜h(z) ∈ M 1 (M, ν) with shadow h(z).<br />
Pro<strong>of</strong>. When k = 1, equation (2) becomes ∆ 1 = ξ 1 ◦ ξ 1 . So the Poincaré series P m (z, s)<br />
satisfies<br />
(26) ∆ 1 (P m (z, s)) = ( s − 1 2) 2<br />
Pm (z, s)<br />
when Re(s) > 1. Since the difference between both sides is holomorphic in s and P m (z, s)<br />
can be analytically continued to Re(s) > 0 as in Proposition 2.1, equation (26) is valid for<br />
Re(s) > 0. At s = 1/2, suppose P m (z, s) has the following Taylor series expansion in s<br />
P m (z, s) = g −1 (z) ( s − 1 2) −1<br />
+ g0 (z) + g 1 (z) ( s − 1 2)<br />
+ Oz<br />
( (s<br />
−<br />
1<br />
2<br />
) 2<br />
)<br />
,<br />
with g j (z) ∈ F 1 (M, ν) real-analytic for j = −1, 0, 1. Since ξ 1 commutes with the slash<br />
operator and ∆ 1 (g 1 (z)) = ξ1(g 2 1 (z)) = g −1 (z), we have ξ 1 (g 1 (z)) ∈ F 1 (M, ν) is real-analytic<br />
and a preimage <strong>of</strong> g −1 (z) under ξ 1 .<br />
By considering the Fourier expansion <strong>of</strong> g 1 (z), we know that if g 1 (z) has at most linear<br />
exp<strong>one</strong>ntial growth near the cusps, then ξ 1 (g 1 (z)) has the same property. Suppose Q m (z, s)<br />
has the Laurent expansion<br />
Q m (z, s) = g −1 (z) ( s − 1 2) −1<br />
+ g<br />
′<br />
0 (z) + g ′ 1(z) ( s − 1 2)<br />
+ Oz<br />
( (s<br />
−<br />
1<br />
2<br />
near s = 1/2. Then from the spectral expansion <strong>of</strong> Q m (z, s), i.e. y −1/2 (D(z, s) + C(z, s))<br />
as in (20), it is not hard to see that g 1(z) ′ has at most linear exp<strong>one</strong>ntial growth near the<br />
cusps. The difference P m (z, s) − Q m (z, s) is a Poincaré series defined for Re(s) > 0. So the<br />
coefficient <strong>of</strong> (s − 1/2) <strong>of</strong> its Laurent series expansion around s = 1/2, say g 1(z), ′′ has at most<br />
linear exp<strong>one</strong>ntial growth near the cusps. That means the sum <strong>of</strong> g 1(z) ′ and g 1(z), ′′ which is<br />
g 1 (z) by analytic continuation, also has this property. Thus, ξ 1 (g 1 (z)) is in H 1 (M, ν) and a<br />
preimage <strong>of</strong> g −1 (z) ∈ S 1 (M, ν) under ξ 1 . Since the functions {Res s=1/2 P m (z, s) : m ≥ 1} span<br />
the space S 1 (M, ν), the map ξ 1 : H 1 (M, ν) → S 1 (M, ν) is surjective.<br />
□<br />
3. Regularized Inner Products and the Weight One Plus Space<br />
From now on, we fix M = p to be a prime number congruent to 3 modulo 4 and ν = χ p . The<br />
spaces H 1(p, ! χ p ), H 1 (p, χ p ), M 1(p, ! χ p ), M 1 (p, χ p ), S 1 (p, χ p ), M ! 1(p, χ p ), M 1 (p, χ p ) are the same<br />
as before, and we will drop χ p in these notations. In this section, we will relate the regularized<br />
inner products between g(z) ∈ S 1 (p) and f(z) ∈ M 1(p) ! to linear combinations <strong>of</strong> coefficients<br />
<strong>of</strong> a mock-<strong>modular</strong> form ˜g(z), whose shadow is g(z), via Stokes’ theorem. This technique<br />
has been used in [9], [11] and [18]. Then, we will split up the space M ! 1(p) canonically into<br />
the plus and minus spaces to facilitate further analysis. Whenever convenient, we will switch<br />
between M ! 1(p) and H 1(p).<br />
!<br />
Given f(z) ∈ M 1(p) ! and g(z) ∈ S 1 (p), the usual Petersson inner product 〈f, g〉 can be<br />
regularized as follows. Since p is prime, ( Γ 0 (p) has two inequivalent cusps, 0 and ∞. They<br />
−1/<br />
are related by the scaling matrix σ 0 =<br />
√ p<br />
√ p<br />
). Take a fundamental domain <strong>of</strong> Γ 0 (p)\H,<br />
cut <strong>of</strong>f the portion with Im(z) > Y for a large Y and intersect it with its translate under σ 0 .<br />
We will call this the truncated fundamental domain F(Y ). Now, define the regularized inner<br />
product by<br />
(27) 〈f, g〉 reg := lim<br />
Y →∞<br />
∫<br />
F(Y )<br />
f(z)g(z)y dxdy<br />
y 2 .<br />
) 2<br />
)
14 W. DUKE AND Y. LI<br />
If f(z) ∈ M 1 (p), then this is the usual Petersson inner product. Now let ĝ(z) ∈ H 1 (p) be a<br />
preimage <strong>of</strong> g(z) under ξ 1 with the following Fourier expansions<br />
where W p = (<br />
−1<br />
p<br />
ĝ(z) = ∑ n∈Z<br />
c + ∞(n)q n − ∑ n≥1<br />
c(g, n)β 1 (n, y)q −n ,<br />
(ĝ| 1 W p )(z) = ∑ c + 0 (n)q n − ∑ c(g| 1 W p , n)β 1 (n, y)q −n ,<br />
n∈Z<br />
n≥1<br />
)<br />
is the Fricke involution and it acts on f(z) ∈ H<br />
!<br />
1 (p) by<br />
( )<br />
(f| 1 W p )(z) = √ 1<br />
pz<br />
f − 1 .<br />
pz<br />
The expression for (ĝ| 1 W p )(z) follows from the commutativity between ξ 1 and the slash<br />
operator. Note that ∑ n∈Z c+ ∞(n)q n and ∑ n∈Z c+ 0 (n)q n are mock-<strong>modular</strong> <strong>forms</strong> with shadows<br />
g(z) and (g| 1 W p )(z) respectively.<br />
Suppose f(z) ∈ M 1(p) ! has the Fourier expansion ∑ n∈Z c ∞(f, n)q n and ∑ n∈Z c 0(f, n)q n at<br />
the cusp infinity and 0 respectively. Then we can express 〈f, g〉 reg in terms <strong>of</strong> these Fourier<br />
coefficients.<br />
Lemma 3.1. Let f(z) ∈ M ! 1(p) and g(z) ∈ S 1 (p). In the notations above, we have<br />
(28) 〈f, g〉 reg = ∑ n∈Z<br />
c + ∞(n)c ∞ (f, −n) + c + 0 (n)c 0 (f, −n).<br />
Pro<strong>of</strong>. Since g(z) = ξ 1 (ĝ(z)) = −2iy ∂ĝ<br />
∂z<br />
(29) 〈f, g〉 reg = − lim<br />
Y →∞<br />
and dzdz = 2idxdy, we can rewrite equation (27) as<br />
∫<br />
f(z) ∂ĝ<br />
∂z dzdz.<br />
F(Y )<br />
As an application <strong>of</strong> Stokes’ Theorem, we have<br />
∫ (<br />
f(z) ∂ĝ<br />
∂z + ∂f ) ∮<br />
∂z ĝ(z) dzdz =<br />
F(Y )<br />
∂F(Y )<br />
f(z)ĝ(z)dz,<br />
where the line integral is in the counterclockwise direction. Using the <strong>modular</strong>ity <strong>of</strong> f(z)<br />
and ĝ(z), we can cancel most <strong>of</strong> the contour integral on the right hand side, except the<br />
contributions from the horizontal segment from − 1 + iY and 1 + iY near ∞ and the arc<br />
2 2<br />
center at 0 connecting the points σ 0 (± 1 ∂f<br />
+ iY ). Also, f(z) is holomorphic, hence = 0.<br />
2 ∂z<br />
Thus, the equation above becomes<br />
∫<br />
f(z) ∂ĝ ∫ −<br />
1<br />
F(Y ) ∂z dzdz = 2 +iY<br />
f(z)ĝ(z) + (f| 1 W p )(z)(ĝ| 1 W p )(z)dz.<br />
1<br />
2 +iY<br />
Now combining this with equations (29) and the bound |β 1 (n, y)| = O(e −4πny ), we obtain<br />
equation (28)<br />
□<br />
Remark. Notice that the right hand side <strong>of</strong> equation (28) depends on the choice <strong>of</strong> ĝ(z),<br />
whereas the left hand side only depends on g(z). So if we replace ĝ(z) with h(z) ∈ M ! 1(p),<br />
then Lemma 3.1 still holds and we obtain<br />
0 = ∑ n∈Z<br />
c ∞ (h, n)c ∞ (f, −n) + c 0 (h, n)c 0 (f, −n),<br />
where h(z) has Fourier expansions ∑ n∈Z c ∞(h, n)q n and ∑ n∈Z c 0(h, n)q n at the cusp infinity<br />
and 0 respectively.
MOCK-MODULAR FORMS OF WEIGHT ONE 15<br />
Now, we want to split up the space H ! 1(p) and M ! 1(p) canonically into the direct sum <strong>of</strong><br />
two subspaces. This decomposition behaves nicely with respect to ξ 1 and regularized inner<br />
product. For ɛ = ±1, we define the following space<br />
(30) H !,ɛ<br />
1 (p) := {f(z) = ∑ n∈Z<br />
a(n, y)q n ∈ H ! 1(p) : a(n, y) = 0 whenever χ p (n) = −ɛ}.<br />
By imposing the same condition on the Fourier expansions, we can define H1(p), ɛ M !,ɛ<br />
1 (p),<br />
M1(p), ɛ S1(p), ɛ M !,ɛ<br />
1 (p) and M ɛ 1(p). Notice that M !,ɛ<br />
1 (p) is the same as the space A ɛ k (p, χ p)<br />
defined on page 51 <strong>of</strong> [8] for k = 1. By the definition <strong>of</strong> R p (n) in (5), it is clear that<br />
E 1 (z) ∈ M 1 + (p). Also, the non-holomorphic <strong>weight</strong> <strong>one</strong> Eisenstein series Ẽ1(z) belongs to<br />
M !,−<br />
1 (p). This can be checked from the factorization (6) and the fact that E 1 (z) ∈ M 1 + (p).<br />
Clearly, H !,ɛ<br />
1 (p) is a subspace <strong>of</strong> H 1(p) ! for ɛ = ±1. In fact, the space H 1(p) ! can be<br />
decomposed into direct sum <strong>of</strong> these subspaces using the operators defined as follows. Let<br />
U p be the standard operator on H 1(p) ! defined by<br />
(31) f(z)| 1 U p = √ 1<br />
p−1<br />
∑ (<br />
f| 1 jp<br />
)<br />
1 .<br />
p<br />
Its action on the Fourier expansion <strong>of</strong> f(z) = ∑ m a(m, y)qm ∈ H 1(p) ! is as follows<br />
(32) (f| 1 U p )(z) = ∑ ( )<br />
a pm, y q m .<br />
p<br />
m<br />
It is easy to see that U p preserves the space H 1(p). ! This is also true for the Fricke involution<br />
W p , since χ p is a real character. Applying W p twice produces a negative sign for odd <strong>weight</strong>,<br />
i.e. (f| 1 Wp 2 )(z) = −f(z). Define the following operator on H 1(p) ! for ɛ ∈ {±1}<br />
j=0<br />
(<br />
ɛi(f| 1 U p W p ) + f<br />
(33) pr ɛ (f) := 1 2<br />
Using these operators, we can decompose H 1(p) ! into the direct sum <strong>of</strong> H !,+<br />
1 (p) and H !,−<br />
1 (p).<br />
On M ! 1(p), <strong>one</strong> can define U p and W p by considering their action on the associated harmonic<br />
Maass form in H 1(p) ! and obtain the same decomposition <strong>of</strong> spaces.<br />
Lemma 3.2. For ɛ = ±1, the operator pr ɛ is the identity operator when restricted to the<br />
subspace H !,ɛ<br />
1 (p) and the zero operator when restricted to the subspace H !,−ɛ<br />
1 (p). Equivalently,<br />
we have<br />
(34) (f| 1 W p )(z) = −iɛ(f| 1 U p )(z),<br />
for all f(z) ∈ H !,ɛ<br />
1 (p). Also, we can write<br />
(35) H ! 1(p) = H !,+<br />
1 (p) ⊕ H !,−<br />
1 (p).<br />
Similarly, we have this decomposition for M ! 1(p).<br />
Remark. The decomposition above clearly holds for subspaces <strong>of</strong> H ! 1(p) or M ! 1(p), such as<br />
H 1 (p), M ! 1(p), M 1 (p), S 1 (p) and M 1 (p).<br />
Pro<strong>of</strong>. Let f(z) = ∑ n∈Z a(n, y)qn ∈ H 1(p) ! and h := f| 1 U p W p . First, we will show that<br />
f ∈ H !,ɛ<br />
1 (p) if and only if pr −ɛ (f) = 0. With some matrix calculations, such as in Lemma 3<br />
<strong>of</strong> [8], we obtain<br />
(36) h(z) = √ 1<br />
∑<br />
(f| 1 W p V p )(z) + ε p χ p (n)a(n, y)q n ,<br />
p<br />
n∈Z<br />
)<br />
.
16 W. DUKE AND Y. LI<br />
where<br />
V p =<br />
( {<br />
p 1, p ≡ 1 (mod 4)<br />
, ε<br />
1)<br />
p =<br />
i, p ≡ 3 (mod 4)<br />
Since p ≡ 3 (mod 4), we know that ε p = i. From the definition <strong>of</strong> pr −ɛ , we have<br />
(37)<br />
(pr −ɛ f)(z) = 1 (−ɛih(z) + f(z))<br />
2<br />
(<br />
= 1 −ɛi<br />
√ (f| 1 W p V p )(z) + ∑ )<br />
(1 − χ p (n)ɛ) a(n, y)q n ∈ H !,ɛ<br />
2 p<br />
1 (p).<br />
n∈Z<br />
Notice that the action V p produces an expansion with coefficients supported on multiples <strong>of</strong><br />
p. So if pr −ɛ (f) = 0, then a(n, y) = 0 whenever χ p (n) = −ɛ and f ∈ H !,ɛ<br />
1 (p) by definition.<br />
Conversely if f ∈ H !,ɛ<br />
1 (p), then<br />
pr −ɛ (f) ∈ H !,+<br />
1 (p) ∩ H !,−<br />
1 (p),<br />
which means the coefficients ξ 1 (pr −ɛ (f)) ∈ M ! 1(p) are supported on multiples <strong>of</strong> p. A lemma<br />
due to Hecke ([33, Lemma 6, p.32] ) implies that<br />
(38) M !,+<br />
1 (p) ∩ M !,−<br />
1 (p) = {0}.<br />
So ξ 1 (pr −ɛ (f)) = 0 and pr −ɛ (f) is holomorphic and also contained in M !,+<br />
1 (p) ∩ M !,−<br />
1 (p). By<br />
the same argument, pr −ɛ (f) vanishes. From this, we can deduce that H !,ɛ<br />
1 (p) is the eigenspace<br />
<strong>of</strong> iU p W p with eigenvalue ɛ, and pr ɛ restricted to H !,ɛ<br />
1 (p) is the identity. Using the relationship<br />
(f| 1 Wp 2 )(z) = −f(z), we can then deduce equation (34). It is easy to see that we have the<br />
following direct sum decomposition<br />
H 1(p) ! = H !,+<br />
1 (p) ⊕ H !,−<br />
1 (p)<br />
f = pr + (f) + pr − (f).<br />
Applying the same procedure to the harmonic Maass form associated to any mock-<strong>modular</strong><br />
form in M ! 1(p), we have the same decomposition <strong>of</strong> space<br />
M ! 1(p) = M !,+<br />
1 (p) ⊕ M !,−<br />
1 (p).<br />
The decomposition <strong>of</strong> the space H 1(p) ! above behaves well under the action <strong>of</strong> ξ 1 . By Theorem<br />
2.2, we can prove a statement about surjectivity <strong>of</strong> ξ 1 : H1(p) ɛ → S1 −ɛ (p). Furthermore,<br />
there is a nice characterization <strong>of</strong> the space S1(p) ɛ as a consequence <strong>of</strong> facts about <strong>weight</strong> <strong>one</strong><br />
<strong>modular</strong> <strong>forms</strong>.<br />
Lemma 3.3. For ɛ = ±1, we have a surjection<br />
(39) ξ 1 : H ɛ 1(p) → S −ɛ<br />
1 (p).<br />
In other words, for any g(z) ∈ S −ɛ<br />
1 (p), there exists a mock-<strong>modular</strong> form ˜g(z) ∈ M ɛ 1(p) whose<br />
shadow is g(z).<br />
In addition, S + 1 (p) contains all dihedral cusp <strong>forms</strong>. If there are 2d − non-dihedral <strong>forms</strong> in<br />
S 1 (p), then the spaces S + 1 (p) and S − 1 (p) have dimensions 1 2 (H(p)−1)+d − and d − respectively<br />
Pro<strong>of</strong>. By Theorem 2.2, the following map is surjective<br />
(40) ξ 1 : H 1 (p) → S −ɛ<br />
1 (p).<br />
□
MOCK-MODULAR FORMS OF WEIGHT ONE 17<br />
Since ξ 1 commutes with the action <strong>of</strong> U p and W p and conjugates their coefficients, it commutes<br />
with the operator pr ɛ as follows<br />
(41) ξ 1 (pr ɛ (f)) = pr −ɛ (ξ 1 (f)).<br />
The operator pr −ɛ is the identity when restricted to S1 −ɛ (p). So the preimage <strong>of</strong> S1 −ɛ (p) under<br />
ξ 1 lies in H1(p) ɛ and the map ξ 1 : H1(p) ɛ → S1 −ɛ (p) is surjective for ɛ = ±1.<br />
Now if f ∈ S 1 (p) is a dihedral newform, then it is a linear combination <strong>of</strong> theta series, whose<br />
n th coefficient is zero if χ p (n) = −1. So f ∈ S 1 + (p) by definition. If f(z) = ∑ n≥1<br />
c(f, n)qn<br />
is a octahedral or icosahedral newform, then f(z) and f(z) := f(z) = ∑ n≥1 c(f, n)qn are<br />
linearly independent new<strong>forms</strong>. When l ≠ p is a prime number, we have the relationship<br />
(42) c(f, l) = χ p (l)c(f, l).<br />
This is a consequence <strong>of</strong> the formula <strong>of</strong> the adjoint <strong>of</strong> the l th Hecke operator with respect to<br />
the Petersson inner product [34, p.21]. By the definition <strong>of</strong> S1(p), ɛ we have f + f ∈ S 1 + (p)<br />
and f − f ∈ S1 − (p). Since the number <strong>of</strong> octahedral and icosahedral new<strong>forms</strong> is set to be<br />
2d − , we obtain the formulae for the dimensions.<br />
□<br />
Remark: The spaces S ɛ 1(p) should be compared to the spaces M + and M − in (9.1.2) in<br />
[36]. If all octahedral and icosahedral new<strong>forms</strong> f(z) ∈ S 1 (p) satisfy<br />
(43) (f| 1 W p )(z) = −if(z),<br />
then M + is the span <strong>of</strong> the <strong>weight</strong> <strong>one</strong> Eisenstein series and S + 1 (p) and M − = S − 1 (p).<br />
Besides compatibility with ξ 1 , the decomposition also behaves nicely with respect to the<br />
regularized inner product. As a consequence <strong>of</strong> Lemma 3.1, we have the following proposition.<br />
Proposition 3.4. Let f(z) ∈ M !,ɛ<br />
1 (p), g(z) ∈ S ɛ′<br />
1 (p) and ˜g(z) ∈ M −ɛ′<br />
1 (p) with shadow g(z)<br />
and Fourier expansion ∑ n∈Z c+ (n)q n . If ɛ = ɛ ′ , then<br />
(44) 〈f, g〉 reg = ∑ n∈Z<br />
δ(n)c(f, −n)c + (n).<br />
Here δ(n) is 2 if p|n and 1 otherwise. If ɛ ≠ ɛ ′ , then 〈f, g〉 reg = 0.<br />
Pro<strong>of</strong>. Let ĝ(z) ∈ H −ɛ′<br />
1 (p) be the harmonic Maass form associated to ˜g(z). Then both<br />
assertions are immediate consequences <strong>of</strong> Lemma 3.1 and<br />
(f| 1 W p )(z) = −iɛ(f| 1 U p )(z), (ĝ| 1 W p )(z) = −iɛ ′ (ĝ| 1 U p )(z).<br />
Since the adjoint <strong>of</strong> W p (resp. U p ) is W −1<br />
p<br />
= −W p (resp. W p U p Wp<br />
−1 ) with respect to the<br />
regularized inner product, it is easy to check that pr ɛ is self-adjoint, which also proves the<br />
second claim.<br />
□<br />
When g(z) above is zero, i.e. ˜g(z) is <strong>modular</strong>, equation (44) reduces to an orthogonal<br />
relationship between Fourier coefficients <strong>of</strong> weakly holomorphic <strong>modular</strong> <strong>forms</strong>. In particular,<br />
we can take ˜g(z) to be cusp <strong>forms</strong> and obtain relations satisfied by {c(f, −n) : n ≥ 1}. These<br />
relations turn out to characterize the space M !,ɛ<br />
1 (p), which gives a nice characterization <strong>of</strong><br />
the space M ɛ 1(p).<br />
Proposition 3.5. Let g(z) ∈ S1(p). ɛ Then there exists a mock-<strong>modular</strong> form ˜g(z) ∈ M −ɛ<br />
1 (p)<br />
with shadow g(z) and prescribed principal part ∑ n
18 W. DUKE AND Y. LI<br />
Pro<strong>of</strong>. The necessity part follows from Proposition 3.4. Since h(z) is a cusp form, the summation<br />
in (44) is only over n < 0. When ˜g(z) ∈ M −ɛ<br />
1 (p) is <strong>modular</strong>, the sufficiency follows<br />
from an application <strong>of</strong> Serre duality [5, §3] (also see [8, Theorem 6]).<br />
When ˜g(z) is mock-<strong>modular</strong>, let ∑ n
MOCK-MODULAR FORMS OF WEIGHT ONE 19<br />
For convenience, we write E(τ, s) for E 1 (τ, s). Kr<strong>one</strong>cker’s first limit formula states that<br />
(49) ζ(2s)E(τ, s) = π<br />
s − 1 + 2π ( γ − log(2) − log( √ y|η(τ)| 2 ) ) + O(s − 1),<br />
where γ is the Euler constant. Using (49) and Rankin-Selberg unfolding trick, we can relate<br />
the inner product between dihedral new<strong>forms</strong> to logarithm <strong>of</strong> u A as follows.<br />
Proposition 4.1. Let ψ, ψ ′ be characters <strong>of</strong> Cl(F ), with ψ non-trivial. When ψ ′ = ψ or ψ,<br />
we have<br />
(50) 〈g ψ , g ψ ′〉 = 1 ∑<br />
ψ 2 (A) log |u A |.<br />
12<br />
Otherwise, 〈g ψ , g ψ ′〉 = 0.<br />
A∈Cl(F )<br />
Pro<strong>of</strong>. Since p is prime, the Eisenstein series E p (z, s) has a simple pole at s = 1 with residue<br />
, which is independent <strong>of</strong> z. So we have the relationship<br />
3<br />
π(p+1)<br />
3<br />
π(p + 1) · 〈g ψ, g ψ ′〉 = Res s=1 g ψ (z)g ψ ′(z)E p (z, s)y<br />
∫Γ dxdy .<br />
0 (p)\H<br />
y 2<br />
Now, we can use the Rankin-Selberg method to unfold the right hand side and obtain<br />
∫<br />
g ψ (z)g ψ ′(z)E p (z, s)y dxdy = Γ(s) ∑ r ψ (n)r ψ ′(n)<br />
y 2 (4π) s n s<br />
Γ 0 (p)\H<br />
Let ρ ψ : Gal(Q/Q) → GL 2 (C) be the representation induced from ψ. Then it is also the<br />
<strong>one</strong> attached to g ψ (z) via Deligne-Serre’s theorem. Up to Euler factors at p, the right hand<br />
side is L(s, ρ ψ ⊗ ρ ψ ′), the L-function <strong>of</strong> the tensor product <strong>of</strong> the representations ρ ψ and ρ ψ ′.<br />
From the character table <strong>of</strong> the dihedral group D 2H(p) , we see that<br />
ρ ψ ⊗ ρ ψ ′ = ρ ψψ ′ ⊕ ρ ψψ ′.<br />
So when ψ ′ ≠ ψ or ψ ′ , the L-function L(s, ρ ψ ⊗ρ ψ ′) is holomorphic at s = 1 and 〈g ψ , g ψ ′〉 = 0.<br />
Otherwise, we have<br />
∑ r ψ (n)r ψ (n)<br />
= ζ(s)L(s, χ p)L(s, ρ ψ 2)<br />
,<br />
n s ζ(2s)(1 + p −s )<br />
n≥1<br />
Putting these together, we obtain<br />
(51) 〈g ψ , g ψ 〉 = p<br />
2π L(1, χ p)L(1, ρ 2 ψ 2).<br />
From the theory <strong>of</strong> quadratic <strong>forms</strong>, we have<br />
√ ∑<br />
pL(s, ρψ 2) = ζ(2s) ψ 2 (A)E(τ A , s).<br />
A∈Cl(F )<br />
Since ψ is non-trivial and H(p) is odd, ψ 2 is non-trivial and (49) implies that<br />
L(1, ρ ψ 2) = −√ 2π ∑<br />
ψ 2 (A) log( √ y A |η(τ A )| 2 ),<br />
p<br />
A∈Cl(F )<br />
Along with the class number formula for p > 3<br />
L(1, χ p ) = 2πH(p)<br />
w p<br />
√ p<br />
n≥1<br />
= πH(p) √ p<br />
,
20 W. DUKE AND Y. LI<br />
we arrive at<br />
(52) 〈g ψ , g ψ 〉 = −H(p)<br />
∑<br />
A∈Cl(F )<br />
Since ψ 2 is non-trivial, we have equation (50).<br />
ψ 2 (A) log( √ y A |η(τ A )| 2 ).<br />
Remarks.<br />
i) By the same procedure, <strong>one</strong> can analyze the inner product between any <strong>weight</strong> <strong>one</strong><br />
new<strong>forms</strong>. In particular, if g(z), h(z) ∈ S 1 (p) arises from different types <strong>of</strong> Galois representations,<br />
then 〈g, h〉 = 0.<br />
ii) This proposition is really a proven case <strong>of</strong> Stark’s conjecture in the abelian, rank <strong>one</strong><br />
case.<br />
Next we prove the existence <strong>of</strong> the special pre-image.<br />
Proposition<br />
∑<br />
4.2. Let ψ be a non-trivial character <strong>of</strong> Cl(F ). Then there exists ˜g ψ (z) =<br />
n≥n 0<br />
r + ψ (n)qn ∈ M − 1 (p) with shadow g ψ (z) such that<br />
(i) When χ p (n) = 1 or n < − p+1 , the coefficient 24 r+ ψ<br />
(n) equals to zero.<br />
(ii) For n ≤ 0, the coefficients r + ψ<br />
(n) are <strong>of</strong> the form<br />
r + ψ<br />
(n) = −β<br />
∑<br />
A∈Cl(F )<br />
ψ 2 (A) log |u(n, A)|,<br />
where β ∈ Q depends only on p, and u(n, A) ∈ O × H<br />
depends only on n and A.<br />
(iii) Let σ C ∈ Gal(H/F ) be the element associated to the class C ∈ Cl(F ) via Artin’s isomorphism.<br />
Then it acts on the units u(n, A) by<br />
σ C (u(n, A)) = u(n, AC −1 ).<br />
Pro<strong>of</strong>. From Lemma 3.3, we know that the dimension <strong>of</strong> S + 1 (p), denoted by d + , is 1 2 (H(p) −<br />
1)+d − . Let {g 1 , g 2 , . . . , g d+ } be a basis <strong>of</strong> S + 1 (p). Suppose we have chosen n 1 , n 2 , . . . , n d+ > 0<br />
such that χ p (n k ) ≠ −1 and the matrix<br />
P := [c(g j , n k )] 1≤j,k≤d+<br />
is invertible. Then for any ψ, there exists ˜g ψ (z) ∈ M − 1 (p) with shadow g ψ (z) and principal<br />
part coefficients r + ψ (−n) = 0 for all positive integer n not in the set {n 1, n 2 , . . . , n d+ } by<br />
Proposition 3.5. Furthermore, these r + ψ<br />
(−n) can be uniquely determined from solving a<br />
d + × d + system <strong>of</strong> equations. We will show that such ˜g ψ (z) can be made to satisfy the<br />
proposition.<br />
First, we can choose {n j : 1 ≤ j ≤ d + } such that n j ≤ (p + 1)/24. Let {g 1 , g 2 , . . . , g d+ } to<br />
be a q-echelon basis. Then these n j ’s can be chosen to be bounded by the supremum <strong>of</strong> the<br />
set<br />
{ord ∞ h(z) : h(z) ∈ S 1 + (p)}.<br />
Now, the square <strong>of</strong> a cusp form in S 1 (p) is in S 2 (p), the space <strong>of</strong> <strong>weight</strong> 2 cusp form <strong>of</strong> trivial<br />
nebentypus on Γ 0 (p). This gives rise to a holomorphic differential form on the <strong>modular</strong> curve<br />
X 0 (p). Atkin showed in [1] that ∞ is not a Weiestrass point <strong>of</strong> the <strong>modular</strong> curve X 0 (p),<br />
i.e. ord ∞ f(z) is no more than the genus <strong>of</strong> X 0 (p). Using the Riemann-Hurwitz formula (see<br />
[37]), we know that the genus <strong>of</strong> X 0 (p) is bounded by (p + 1)/12. So the n j ’s can be chosen<br />
to be bounded by (p + 1)/24.<br />
□
MOCK-MODULAR FORMS OF WEIGHT ONE 21<br />
Let {g j (z) : 1 ≤ j ≤ (H(p) − 1)/2} be the dihedral new<strong>forms</strong>, and label the class group<br />
characters as {ψ j : 1 ≤ j ≤ H(p) − 1} such that g ψj (z) = g j (z) and ψ j = ψ H(p)−j for 1 ≤ j ≤<br />
(H(p) − 1)/2. When (H(p) + 1)/2 ≤ j ≤ d + , let g j (z) be linear combinations <strong>of</strong> non-dihedral<br />
new<strong>forms</strong> such that g j (z) has integral coefficients and the set {g j (z) : (H(p) + 1)/2 ≤ j ≤ d + }<br />
is linearly independent. Then for each non-trivial ψ, the values {r + ψ (−n k) : 1 ≤ k ≤ d + }<br />
satisfies the equation<br />
P · [δ(n k )r + ψ (−n k)] 1≤k≤d+ = R,<br />
where R is the d + × 1 matrix (〈g j , g ψ 〉) 1≤j≤d+<br />
. By Proposition 4.1, the inner product 〈g j , g ψ 〉<br />
∑<br />
is either 0 or 1<br />
12 A∈Cl(F ) ψ2 (A) log |u A |. So we have<br />
⎛<br />
⎞<br />
0.<br />
∑<br />
0<br />
1<br />
R =<br />
12 A ψ2 (A) log |u A |<br />
⎜<br />
⎟<br />
⎝<br />
0. ⎠<br />
0<br />
Let A 0 be the principal class in Cl(F ) and label the non-principal ( classes as A i for 1 ≤<br />
i ≤ H(p) − 1 such that A −1 = A H(p)−i . Let M ′ M1 0<br />
be the matrix 0 Id d−<br />
), where M 1 is an<br />
i<br />
1<br />
(H(p) − 1) × 1 (H(p) − 1) matrix defined by<br />
2 2<br />
[ψ j (A i ) + ψ j (A H(p)−i ) − 2] 1≤i,j≤(H(p)−1)/2<br />
and Id d− is the d − × d − identity matrix. The product M ′ P is the matrix<br />
⎛<br />
⎞<br />
H(p)(r A1 (n 1 ) − r A0 (n 1 )) · · · H(p)(r A1 (n d+ ) − r A0 (n d+ ))<br />
. .<br />
.. .<br />
M ′ H(p)(r<br />
P =<br />
A(H(p)−1)/2 (n 1 ) − r A0 (n 1 )) · · · H(p)(r A(H(p)−1)/2 (n d+ ) − r A0 (n d+ ))<br />
,<br />
c(g<br />
⎜<br />
(H(p)+1)/2 , n 1 ) · · · c(g (H(p)+1)/2 , n d+ )<br />
⎝<br />
. ⎟<br />
.<br />
.. .<br />
⎠<br />
c(g d+ , n 1 ) · · · c(g d+ , n d+ )<br />
which has integer coefficients. It is also non-singular since<br />
{ϑ Aj − ϑ A0 : 1 ≤ j ≤ (H(p) − 1)/2} ∪ {g j : (H(p) + 1)/2 ≤ j ≤ d + }<br />
is a basis <strong>of</strong> S 1 + (p). The product M ′ R is the d + × 1 matrix<br />
⎛ ∑<br />
A ψ2 (A) log |u A (A 1 )|<br />
where<br />
Also by (47), we have<br />
M ′ R =<br />
⎜<br />
⎝<br />
1<br />
12<br />
1<br />
12<br />
⎞<br />
∑<br />
.<br />
A ψ2 (A) log |u A (A (H(p)−1)/2 )|<br />
,<br />
⎟<br />
0.<br />
⎠<br />
0<br />
u A (B) := u A √ B u A √ B −1<br />
u 2 A<br />
σ C (u A (B)) = u AC −1(B)<br />
∈ O × H .
22 W. DUKE AND Y. LI<br />
for any class A, B, C ∈ Cl(F ).<br />
Since M ′ P is a non-singular matrix with integer entries, its inverse, say (α i,j ) 1≤i,j≤d+ , has<br />
rational entries. Thus, r + ψ (−n i) can be written as<br />
δ(n i )r + ψ (−n i) = ∑ A<br />
(H(p)−1)/2<br />
∑<br />
ψ 2 (A) α i,j log |u A (A j )|<br />
j=1<br />
with α i,j ∈ Q for all 1 ≤ i, j ≤ d + . From this, we can choose β ∈ Q such that<br />
u(n i , A) =<br />
(H(p)−1)/2<br />
∏<br />
j=1<br />
u A (A j ) −α i,j/δ(n i )β ∈ O × H .<br />
α i,j<br />
2H(p)β ∈ Z and<br />
Then r + ψ (−n i) satisfies conditions (ii) and (iii) in the proposition. Furthermore, the unit<br />
u(n i , A) is an H(p) th power in O × H .<br />
Finally, apply Propositions 4.1 and 3.4 to the Eisenstein series E 1 (z) and the cusp form<br />
g ψ (z), we get<br />
∑<br />
δ(n)R p (n)r + ψ<br />
(−n) = 0.<br />
n≥0<br />
So we can write r + ψ<br />
(0) = −β log |u(0, A)| with<br />
u(0, A) =<br />
d +<br />
∏<br />
i=1<br />
which satisfies condition (iii) as well.<br />
u(n i , A) −δ(n i)R p(n i )/H(p) ∈ O × H ,<br />
□<br />
5. Rankin’s method and values <strong>of</strong> Green’s function I<br />
From the pro<strong>of</strong> <strong>of</strong> Proposition 4.2, we saw that information about r + ψ<br />
(n), n ≤ 0 can be<br />
retrieved from their various linear combinations coming from inner products. In this section,<br />
we will consider a different linear combination <strong>of</strong> the coefficients r + ψ<br />
(n). They can be viewed<br />
as the regularized inner product between g ψ (z) and certain weakly holomorphic <strong>forms</strong> in<br />
M !,+<br />
1 (p) lifted from <strong>modular</strong> functions <strong>of</strong> level <strong>one</strong>. Following the same procedure as in [24],<br />
we will show in Theorem 5.1 that these linear combination <strong>of</strong> r + ψ<br />
(n)’s can be expressed as the<br />
twisted sum <strong>of</strong> logarithms <strong>of</strong> the <strong>modular</strong> polynomial evaluated at certain CM points. This<br />
result is more or less a twisted version <strong>of</strong> the Gross-Zagier formula at level <strong>one</strong>, where there<br />
is no <strong>weight</strong> two cusp form or L-function. Also, the information provided by the theorem,<br />
as formulated in Corollary 5.2, will be a stepping st<strong>one</strong> to the pro<strong>of</strong> <strong>of</strong> Theorem 1.1. As a<br />
<strong>modular</strong> form on the degenerated Hilbert <strong>modular</strong> surface, the <strong>modular</strong> polynomial can be<br />
expressed in terms <strong>of</strong> a Borcherds lift <strong>of</strong> <strong>modular</strong> functions <strong>of</strong> level <strong>one</strong>. Thus, it is natural<br />
to phrase Theorem 5.1 in terms <strong>of</strong> a regularized inner product and this Borcherds lift. This<br />
prepares the way for §6, where we treat the level N case.<br />
For m ≥ 0, let j m (z) = q −m + O(q) be the unique <strong>modular</strong> function <strong>of</strong> level <strong>one</strong> with a<br />
pole <strong>of</strong> order m at infinity, e.g. j 0 (z) = 1 and j 1 (z) = j(z) − 744. For each B ∈ Cl(F ), choose<br />
an associated CM point τ B = x B + iy B ∈ H and define j lift,B<br />
m (z) ∈ M !,+<br />
1 (p) by<br />
jm<br />
lift,B (z) := j m (pz)ϑ B (z).
MOCK-MODULAR FORMS OF WEIGHT ONE 23<br />
Let M 2 (Z) be the space <strong>of</strong> 2 × 2 matrices with integer coefficients. For m ≥ 1, (τ 1 , τ 2 ) ∈<br />
(PSL 2 (Z)\H) 2 , define the function Ψ m (τ 1 , τ 2 ) by<br />
∏<br />
Ψ m (τ 1 , τ 2 ) :=<br />
(j(τ 1 ) − j(γτ 2 )) .<br />
γ∈SL 2 (Z)\M 2 (Z)<br />
det(γ)=m<br />
It is the value <strong>of</strong> the <strong>modular</strong> polynomial ϕ m (X, Y ) at X = j(τ 1 ), Y = j(τ 2 ), and also the<br />
Borcherds lift <strong>of</strong> j m (z) to a function on the degenerated Hilbert <strong>modular</strong> surface. The main<br />
result <strong>of</strong> this section is as follows.<br />
Theorem 5.1. Let ψ be a non-trivial character <strong>of</strong> Cl(F ) and m ≥ 1. Then<br />
∑<br />
( log |Ψm (τ<br />
(53) 〈jm<br />
lift,B , g ψ 〉 reg = −2 lim ψ 2 (A ′ A<br />
)<br />
′ B ′−1 + ɛ, τ A ′ B ′)| + )<br />
ɛ→0<br />
r B (m) log |y A ′ B ′−1| ,<br />
A ′ ∈Cl(F )<br />
where B ′ ∈ Cl(F ) is the unique class such that B ′2 = B.<br />
It will be clear in the pro<strong>of</strong> that equation (53) is independent <strong>of</strong> the choice <strong>of</strong> τ B . Before<br />
stating the pro<strong>of</strong>, we need a crucial result in [23]. For two distinct points τ j = x j + iy j ∈ H,<br />
the invariant hyperbolic distance d(τ 1 , τ 2 ) between them is defined by<br />
(54) cosh d(τ 1 , τ 2 ) = (x 1 − x 2 ) 2 + y1 2 + y2<br />
2 .<br />
2y 1 y 2<br />
Notice for d(τ 1 , τ 2 ) = d(γτ 1 , γτ 2 ) for all γ ∈ SL 2 (R). The Legendre function <strong>of</strong> the second<br />
kind Q s−1 (t) is defined by<br />
∫ ∞<br />
Q s−1 (t) = (t + √ t 2 − 1 cosh u) −s du, Re(s) > 1, t > 1,<br />
(55)<br />
0<br />
Q 0 (t) = 1 log ( )<br />
1 + 2<br />
2 t−1<br />
For two distinct points τ 1 , τ 2 ∈ PSL 2 (Z)\H, the following convergent series defines the automorphic<br />
Green’s function<br />
(56) G s (τ 1 , τ 2 ) := ∑<br />
g s (τ 1 , γτ 2 ), Re(s) > 1,<br />
where<br />
γ∈PSL 2 (Z)<br />
(57) g s (τ 1 , τ 2 ) := −2Q s−1 (cosh d(τ 1 , τ 2 )).<br />
Recall that E(τ, s) is defined in (48) and ϕ 1 (s) is the coefficient <strong>of</strong> y 1−s in the Fourier<br />
expansion <strong>of</strong> E(τ, s).<br />
Proposition 5.1 in [23] tells us that for distinct τ 1 , τ 2 ∈ PSL 2 (Z)\H, the values <strong>of</strong> the<br />
j-function is related to the values <strong>of</strong> the automorphic Green’s function by<br />
(58) log |j(τ 1 ) − j(τ 2 )| 2 = lim<br />
s→1<br />
(G s (τ 1 , τ 2 ) + 4πE(τ 1 , s) + 4πE(τ 2 , s) − 4πϕ 1 (s)) − 24.<br />
Apply the m th Hecke operator to τ 2 on both sides gives us [23, eq.(5.2)]<br />
( G<br />
m )<br />
(59) log |Ψ m (τ 1 , τ 2 )| 2 s (τ 1 , τ 2 ) + 4πσ 1 (m)E(τ 1 , s)+<br />
= lim<br />
s→1 4πm s σ 1−2s (m)E(τ 2 , s) − 4πσ 1 (m)ϕ 1 (s)<br />
where σ ν (m) = ∑ d|m dν and<br />
G m s (τ 1 , τ 2 ) =<br />
∑<br />
γ∈SL 2 (Z)\M 2 (Z)<br />
det(γ)=m<br />
g s (τ 1 , γτ 2 ).<br />
− 24σ 1 (m),
24 W. DUKE AND Y. LI<br />
Using this result and appropriate Rankin-Selberg unfolding trick as in [24], we can prove<br />
Theorem 5.1.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 5.1. Let ˜g ψ (z) ∈ M − 1 (p) be a mock-<strong>modular</strong> form as in Proposition 4.2<br />
with Fourier expansion<br />
˜g ψ (z) = ∑ r + ψ (n)qn<br />
n∈Z<br />
and ĝ ψ (z) the associated harmonic Maass form. For a class B ∈ Cl(F ), define<br />
Φ B (ĝ ψ , z) := Tr p 1(−i(ĝ ψ |W p )(z)ϑ B (z))<br />
= (ĝ ψ |U p )(z)ϑ B (z) + (ĝ ψ (z)ϑ B (z))|U p .<br />
This should be compared to ˜Φ s (z) in Proposition (1.2) <strong>of</strong> [24, §IV]. When N = 1, the<br />
derivative <strong>of</strong> ˜Φ s (z) at s = 1 is more or less our Φ B (ĝ ψ , z) for ψ trivial.<br />
Here, Φ B (ĝ ψ , z) trans<strong>forms</strong> like a level <strong>one</strong>, <strong>weight</strong> two <strong>modular</strong> form and has the following<br />
Fourier expansion at infinity<br />
Φ B (ĝ ψ , z) = ∑ n∈Z(b B (ĝ ψ , n, y) + a B (ĝ ψ , n))q n ,<br />
b B (ĝ ψ , n, y) = − ∑ k∈Z<br />
a B (ĝ ψ , n) = ∑ k∈Z<br />
δ(k)r ψ (k)β 1<br />
(k, y p<br />
δ(k)r + ψ (k) r B(pn − k).<br />
)<br />
r B (pn + k),<br />
By the choice <strong>of</strong> ˜g ψ (z), we have ord ∞ (˜g ψ (z)) ≥ − p+1 > −p. So the sum a 24 B(ĝ ψ , n) is zero for<br />
all n < 0. It is easy to see from the definition above that jm lift,B ∈ M !,+<br />
1 (p). So we can apply<br />
Proposition 3.4 to obtain<br />
(60)<br />
〈j lift,B<br />
m , g ψ 〉 reg = a B (ĝ ψ , m).<br />
In particular when m = 0, we have j lift,B<br />
0 = 1 lift,B = ϑ B (z) and<br />
〈ϑ B , g ψ 〉 = a B (ĝ, 0).<br />
Now, let Ẽ2(z) the non-holomorphic Eisenstein series <strong>of</strong> level <strong>one</strong>, <strong>weight</strong> two defined by<br />
Ẽ 2 (z) = 1 − 24 ∑ n≥1<br />
σ 1 (n)q n − 3<br />
πy .<br />
Then we can use Poincaré series to apply holomorphic projection to the function<br />
Φ ∗ B(ĝ ψ , z) := Φ B (ĝ ψ , z) − a B (ĝ ψ , 0)Ẽ2(z),<br />
since it satisfies the growth condition Φ ∗ B (ĝ ψ, z) = O(1/y) at the cusp infinity. For m > 0, let<br />
P m,1 (z, s) be the Poincaré series <strong>of</strong> level <strong>one</strong>, <strong>weight</strong> two defined by<br />
∑<br />
P m,1 (z, s) :=<br />
(y s e 2πimz )| 2 γ.<br />
γ∈Γ ∞\PSL 2 (Z)<br />
The limit as s goes to zero <strong>of</strong> the inner product between P m,1 (z, s) and Φ ∗ B (ĝ ψ, z) is the m th<br />
Fourier coefficient <strong>of</strong> a cusp form <strong>of</strong> <strong>weight</strong> two, level <strong>one</strong>. Since the space <strong>of</strong> such <strong>forms</strong> is<br />
trivial, we have<br />
(61) lim<br />
s→0<br />
〈Φ ∗ B(ĝ ψ , z), P m,1 (z, s)〉 = 0.
MOCK-MODULAR FORMS OF WEIGHT ONE 25<br />
Using the Fourier expansion <strong>of</strong> Φ B (ĝ ψ , z) and Rankin-Selberg unfolding, we can rewrite equation<br />
(61) as<br />
( )<br />
k<br />
(62) a B (ĝ ψ , m) + 24σ 1 (m)a B (ĝ ψ , 0) = lim δ(k)r B (pm + k)r ψ (k)φ s ,<br />
where<br />
φ s (µ) :=<br />
∫ ∞<br />
1<br />
s→0<br />
∑<br />
k≥1<br />
du<br />
(1 + µu) 1+s u .<br />
Comparing Q s−1 (t) and φ s−1 (µ), we see that φ 0 (µ) = 2Q 0 (1 + 2µ) and<br />
φ s−1 (µ) − 2Γ(2s) Q<br />
sΓ(s) 2 s−1 (1 + 2µ) = O(µ −1−s ).<br />
( ) ( )<br />
2<br />
That means we can substitute Q<br />
sΓ(s) 2 s−1 1 + 2k<br />
k<br />
for φ<br />
pm s−1 and in the limit as s approaches<br />
1. Together with (8) and (60), equation (61)<br />
pm<br />
becomes<br />
(63)<br />
〈j lift,B<br />
m , g ψ 〉 reg +24σ 1 (m)〈ϑ B , g ψ 〉 =<br />
− lim<br />
s→1<br />
∑<br />
A<br />
ψ(A) ∑ k≥1<br />
pm<br />
( ))<br />
δ(k)r B (pm + k)r A (k)<br />
(−2Q s−1 1 + 2k .<br />
pm<br />
Now the counting argument in Proposition (3.11) in [23] tells us that<br />
δ(k)r B (pm + k)r A (k) = ρ m (A, B, k),<br />
where ρ m (A, B, k) counts the number <strong>of</strong> γ ∈ M 2 (Z)/{±1} such that det(γ) = m and<br />
cosh d(τ √ AB −1 , γτ √ AB ) = 1 + 2k<br />
pm .<br />
In particular when k = 0, the number <strong>of</strong> γ ∈ M 2 (Z)/{±1} such that det(γ) = m and<br />
γτ √ AB = τ√ AB<br />
is exactly r −1 B (pm) = r B (m). Since k ≥ 1 in the summation, equation (63)<br />
becomes<br />
(64)<br />
〈jm<br />
lift,B ,g ψ 〉 reg + 24σ 1 (m)〈ϑ B , g ψ 〉 = − lim<br />
= − lim<br />
ɛ→0<br />
lim<br />
s→1<br />
∑<br />
A<br />
s→1<br />
∑<br />
A<br />
ψ(A)<br />
∑<br />
(<br />
g s τ<br />
√<br />
AB −1, γτ √ AB<br />
γ∈M 2 (Z)/{±1}<br />
det(γ)=m,<br />
γτ √ AB≠τ√ AB −1<br />
ψ(A) ( G m s (τ √ AB −1 + ɛ, τ √ AB ) − r B(m)g s (τ √ AB −1 + ɛ, τ √ AB −1 ) )<br />
where g s is defined in (57). From this expression, it is clear that the choice <strong>of</strong> these CM<br />
points do not matter.<br />
)<br />
Since g 1 (τ + ɛ, τ) = − log<br />
(1 + 4Im(τ)2 and ψ is non-trivial, we see that<br />
ɛ 2<br />
lim lim<br />
ɛ→0<br />
s→1<br />
∑<br />
A<br />
ψ(A)g s (τ √ AB −1 + ɛ, τ √ AB −1 ) = − ∑ A<br />
Also by equation (59), we have<br />
∑<br />
lim ψ(A)G m s (τ √<br />
s→1<br />
AB<br />
+ ɛ,τ √ −1 AB ) = ∑<br />
A<br />
A<br />
4πσ 1 (m) ∑ A<br />
ψ(A) log(y 2 √<br />
AB −1).<br />
ψ(A) log |Ψ m (τ √ AB −1 + ɛ, τ √ AB )|2 −<br />
ψ(A) lim<br />
s→1<br />
(E(τ √ AB −1 + ɛ, s) + E(τ √ AB , s)).<br />
)
26 W. DUKE AND Y. LI<br />
By (49) and the substitution A = A ′2 , we have<br />
lim<br />
s→1 4π ∑ A<br />
ψ(A)(E(τ √ AB −1 , s)+E(τ √ AB , s)) = −24(ψ(B)+ψ(B)) ∑ A ′ ψ 2 (A ′ ) log | √ y A ′η 2 (τ A ′)|.<br />
From (52) and<br />
H(p)ϑ B (z) = ∑ ψ ′<br />
ψ ′ (B)g ψ ′(z),<br />
it is easy to see that<br />
〈ϑ B , g ψ 〉 = −(ψ(B) + ψ(B)) ∑ A ′ ψ 2 (A ′ ) log | √ y A ′η 2 (τ A ′)|.<br />
After substituting these into equation (64) and changing A, B to A ′2 , B ′2 respectively, we<br />
obtain equation (53).<br />
□<br />
For any m ≥ 1, r ≥ 0 and τ, τ ′ ∈ H, define the level <strong>one</strong> <strong>modular</strong> function Ψ ∗ m(z; τ, τ ′ , r)<br />
by<br />
Ψ ∗ m(z; τ, τ ′ Ψ m (τ, z)<br />
, r) :=<br />
(j(τ ′ ) − j(z)) r<br />
In terms <strong>of</strong> Ψ ∗ m(z; τ, τ ′ , r), we can re-write the right hand side <strong>of</strong> (53) as<br />
−2<br />
∑<br />
ψ 2 (A ′ ) (log |Ψ ∗ m(τ A ′ B ′; τ A ′ B ′−1, τ A ′ B ′, r B(m))| + r B (m) log |y A ′ B ′j′ (τ A ′ B ′)|)<br />
A ′ ∈Cl(F )<br />
The quantity Ψ ∗ −1<br />
m(τ A ′ B ′; τA ′ B<br />
, τ ′ A ′ B ′, r B(m)) is well-defined and non-zero, since the order <strong>of</strong><br />
Ψ m (τ A ′ B ′−1, z) at z = τ A ′ B ′ is exactly r B(m) from the pro<strong>of</strong> above. By equation (47) and the<br />
fact that<br />
(j ′ (z)) 6 = j(z) 4 (j(z) − 1728) 3 ∆(z),<br />
we have<br />
σ C<br />
⎛<br />
⎝<br />
∏<br />
C ′ ∈Cl(F )<br />
y 6 A ′ B ′j′ (τ A ′ B ′)6<br />
y 6 C ′ ∆(τ C ′)<br />
⎞<br />
⎛<br />
⎠ = ⎝<br />
∏<br />
C ′ ∈Cl(F )<br />
y 6 C −1 A ′ B ′ j ′ (τ C −1 A ′ B ′)6<br />
y 6 C ′ ∆(τ C ′)<br />
⎞<br />
⎠ ∈ H.<br />
Similarly, the <strong>modular</strong> function Ψ ∗ m(z; τ A ′ B ′−1, τ A ′ B ′, r B(m)), which is defined over H, is sent<br />
to Ψ ∗ m(z; τ C −1 A ′ B ′−1, τ C −1 A ′ B ′, r B(m)) under σ C . So if we let<br />
⎛<br />
u m,B (A ′ ) := Ψ ∗ m(τ A ′ B ′; τ A ′ B ′−1, τ A ′ B ′, r B(m)) 6H(p) · ⎝<br />
⎞<br />
∏<br />
r B (m)<br />
yA 6 ′ B ′j′ (τ A ′ B ′)6 ⎠<br />
yC 6 ∆(τ ′ C ′)<br />
,<br />
then we can write<br />
with u m,B (A ′ ) ∈ H satisfying<br />
〈j lift,B<br />
m , g ψ 〉 reg = − 1<br />
3H(p)<br />
∑<br />
A ′ ∈Cl(F )<br />
C ′ ∈Cl(F )<br />
(65) σ C (u m,B (A ′ )) = u m,B (A ′ C −1 ).<br />
ψ 2 (A ′ ) log |u m,B (A ′ )|<br />
When m = 0, we have j lift,B<br />
0 (z) = ϑ B (z) and<br />
∑<br />
〈ϑ B , g ψ 〉 = − 1 ψ 2 (A ′ ) log |u<br />
12H(p)<br />
A ′ B ′u A ′ B ′−1|,<br />
A ′
MOCK-MODULAR FORMS OF WEIGHT ONE 27<br />
where u C ′ is defined in (46) for any C ′ ∈ Cl(F ). So we can let u m,B (A ′ ) = u A ′ B ′u A ′ B ′−1 when<br />
m = 0 and equation (65) still holds by (47). Since every <strong>modular</strong> function <strong>of</strong> <strong>weight</strong> <strong>one</strong> is<br />
a linear combination <strong>of</strong> j m (z), m ≥ 0, we can deduce the following corollary concerning the<br />
algebraic property <strong>of</strong> r + ψ<br />
(n) from (60) and Proposition 3.4.<br />
Corollary 5.2. Let ˜g ψ (z) = ∑ n∈Z r+ ψ (n)qn ∈ M − 1 (p) be a mock-<strong>modular</strong> form satisfying<br />
condition (i) in Proposition 4.2. For any <strong>modular</strong> function f(z) = ∑ n∈Z c(f, n)qn <strong>of</strong> level<br />
<strong>one</strong> with rational Fourier coefficients, there exists β f,B ∈ Q and u f,B (A) ∈ H × independent<br />
<strong>of</strong> the character ψ such that<br />
(66)<br />
〈f lift,B , g ψ 〉 reg = ∑ c(f, n) ∑ δ(k)r B (pn − k)r + ψ (k)<br />
n∈Z k∈Z<br />
∑<br />
= −β f,B ψ 2 (A) log |u f,B (A)|<br />
A∈Cl(F )<br />
and σ C (u f,B (A)) = u f,B (AC −1 ) for any C ∈ Cl(F ). If f has integral Fourier coefficients, then<br />
12H(p)β f,B ∈ Z.<br />
6. Rankin’s method and values <strong>of</strong> Green’s Function II<br />
In this section, we will work out the generalization <strong>of</strong> Corollary 5.2 to <strong>modular</strong> functions <strong>of</strong><br />
prime level N. From Corollary 5.2, we see that linear combinations <strong>of</strong> r + ψ<br />
(n) have the desired<br />
algebraic property. To prove Theorem 1.1, we need different linear combinations arising from<br />
regularized inner product between g ψ and other f ∈ M !,+<br />
1 (p). By considering analogues <strong>of</strong><br />
j lift,B for <strong>modular</strong> functions <strong>of</strong> level N, we can obtain these linear combinations, which will be<br />
sufficient to solve each r + ψ<br />
(n) (see §7). Other than being a step toward the pro<strong>of</strong> <strong>of</strong> Theorem<br />
1.1, this section is interesting on its own as the analogue <strong>of</strong> the Gross-Zagier formula with<br />
trivial Heegner point divisor in the Jacobian. Also, the interpretation <strong>of</strong> the regularized inner<br />
product in terms <strong>of</strong> Borcherds lifts continues to hold for <strong>modular</strong> functions <strong>of</strong> level N.<br />
Let N be a prime number different from p. For a <strong>modular</strong> function f(z) <strong>of</strong> level N and<br />
B ∈ Cl(F ), <strong>one</strong> can define f lift,N,B (z) ∈ M 1(p) ! by<br />
(67) f lift,N,B (z) := Tr Np<br />
p ((f|W N )(pz)ϑ B (Nz)),<br />
−1<br />
where W N = (<br />
N<br />
) is the Fricke involution and ϑ B (z) is the <strong>weight</strong> <strong>one</strong> theta series corresponding<br />
to the class B ∈ Cl(F ). Notice that the form f lift,N,B (z) is the same as f lift,B (z) from<br />
§5 when N = 1. The main result <strong>of</strong> this section is the following generalization <strong>of</strong> Corollary<br />
5.2.<br />
Proposition 6.1. Let f(z) is a <strong>modular</strong> function <strong>of</strong> prime level N ≠ p, with rational Fourier<br />
coefficients. There exist β f,B ∈ Q and u f,B (A) ∈ H independent <strong>of</strong> the character ψ such that<br />
∑<br />
(68) 〈f lift,N,B , g ψ 〉 reg = −β f,B ψ 2 (A) log |u f,B (A)|.<br />
A∈Cl(F )<br />
For any class C ∈ Cl(F ), the algebraic integer u f,B (A) satisfies<br />
σ C (u f,B (A)) = u f,B (AC −1 )<br />
where σ C ∈ Gal(H/F ) is associated to C ∈ Cl(F ) via Artin’s isomorphism. Furthermore, if<br />
f has integral principal part Fourier coefficients and satisfies equation (69) below, then the<br />
denominator <strong>of</strong> β f,B only depends on N and p.
28 W. DUKE AND Y. LI<br />
The pro<strong>of</strong> <strong>of</strong> Proposition 6.1 follows the same steps as in the pro<strong>of</strong> <strong>of</strong> Theorem 5.1 and<br />
Corollary 5.2. First, we will express 〈f lift,N,B , g ψ 〉 reg in terms <strong>of</strong> other regularized inner products<br />
and the limit <strong>of</strong> an infinite sum. Then, via Green’s function, we will relate the limit <strong>of</strong><br />
the infinite sum to the height pairings between Heegner divisors on J 0 (N), the Jacobian <strong>of</strong><br />
X 0 (N). Since the Heegner divisor is trivial on J 0 (N), the height pairing is the same as the<br />
value <strong>of</strong> some <strong>modular</strong> function evaluated at a Heegner point. From there, we can deduce<br />
equation (68). Before presenting the pro<strong>of</strong>, we need some notations and facts about level N<br />
<strong>modular</strong> <strong>forms</strong>. For this part, we only require N to be a prime number.<br />
Let M 2(N), ! resp. M 0(N), ! be the space <strong>of</strong> weakly holomorphic <strong>weight</strong> 2 <strong>modular</strong> <strong>forms</strong>,<br />
resp. <strong>modular</strong> functions, <strong>of</strong> level N, and trivial nebentypus. Denote the subspaces <strong>of</strong> M 2(N)<br />
!<br />
<strong>of</strong> cusp <strong>forms</strong> and holomorphic <strong>modular</strong> <strong>forms</strong> by M 2 (N), S 2 (N) respectively. Let M !,new<br />
0 (N)<br />
be the subspace <strong>of</strong> M 0(N) ! containing weakly holomorphic <strong>modular</strong> <strong>forms</strong> f(z) satisfying<br />
(69) (f|W N )(z) = −N(f|U N )(z).<br />
The following lemma shows that M ! 0(N) can be be decomposed into the direct sum <strong>of</strong><br />
M !,new<br />
0 (N) and level <strong>one</strong> <strong>modular</strong> functions.<br />
Lemma 6.2. Let f(z) be a <strong>modular</strong> function <strong>of</strong> prime level N. Then it can be written<br />
uniquely as<br />
f(z) = f 1 (Nz) + f 2 (z),<br />
such that f 1 (z) is a <strong>modular</strong> function <strong>of</strong> level 1 and f 2 (z) ∈ M !,new<br />
0 (N).<br />
Pro<strong>of</strong>. Let a, b ∈ Z. With some matrix calculations, we have<br />
( ( ))<br />
f| UN W N + a N (z) =<br />
a−1<br />
f(z) + N TrN 1 (f)(Nz)<br />
( ( ) ( ))<br />
f| UN W N + a UN W<br />
N<br />
N + b N (z) =<br />
(a−1)(b−1)<br />
f(z) + a+b+N−1 Tr N N 2<br />
N 1 (f)(Nz)<br />
Set f j (z) as follows and we are d<strong>one</strong><br />
f 1 (z) = N<br />
N+1 TrN 1 (f)(z),<br />
f 2 (z) = − N N+1 NW N ) (z) − f(z)) .<br />
Alternatively, <strong>one</strong> can characterize the space M ! 0(N) via the space S 2 (N).<br />
Definition 6.3. A set <strong>of</strong> numbers {λ m : m ≥ 1} is called a relation for S 2 (N) if<br />
(i) λ m ∈ C is non-zero for all but finitely many m.<br />
(ii) For any cusp form h(z) = ∑ m≥1 c(h, m)qm ∈ S 2 (N), the numbers {λ m } satisfy<br />
∑<br />
(70)<br />
δ N (m)λ m c(h, m) = 0.<br />
m≥1<br />
where δ N (m) = N + 1 if N|m and 1 otherwise.<br />
A relation {λ m } m≥1 is called integral if λ m ∈ Z for all m ≥ 1. Denote Λ N the set <strong>of</strong> all<br />
relations for S 2 (N).<br />
For any f(z) ∈ M !,new<br />
0 (N), <strong>one</strong> knows that the principal part coefficients, {c(f, −m) :<br />
m ≥ 1}, is a relation for S 2 (N) from works by Petersson. By considering the Fourier expansion<br />
<strong>of</strong> vector-valued Poincaré series, (See [7], [27], [32]), <strong>one</strong> can also show that the converse<br />
is true. So given λ = {λ m } ∈ Λ N , the expression<br />
∑<br />
λ m q −m<br />
m≥1<br />
□
MOCK-MODULAR FORMS OF WEIGHT ONE 29<br />
is necessarily the principal part <strong>of</strong> the Fourier expansion at infinity <strong>of</strong> a unique function<br />
f λ (z) ∈ M !,new<br />
0 (N). Alternatively, this statement follows essentially from an application <strong>of</strong><br />
Serre duality [5, §3]. Suppose λ ∈ Λ N is integral. Then the n th Fourier coefficient <strong>of</strong> f λ (z) is<br />
integral when n ≠ 0 since f λ is a rational function <strong>of</strong> j(z) and j(Nz).<br />
Let g N be the genus <strong>of</strong> Γ 0 (N). From [1], we know that for all h(z) ∈ S 2 (N), ord ∞ (h(z)) ≤<br />
dim(S 2 (N)) = g N ≤ (N + 1)/12. Using matrix computations similar to that in the pro<strong>of</strong> <strong>of</strong><br />
Lemma 6.2, <strong>one</strong> can show that<br />
(71) h| 2 U N = −h| 2 W N<br />
for all h(z) ∈ S 2 (N), implying that U N ◦ U N is the identity operator on S 2 (N). Combining<br />
these facts together gives us the following lemma about M !,new<br />
0 (N).<br />
Lemma 6.4. The space M !,new<br />
0 (N) is spanned by <strong>modular</strong> functions<br />
∞∑<br />
f m (z) = q −m + c m (n)q n ,<br />
n=−g N<br />
with m ≥ g N + 1. For these m, the coefficient c m (n) are integers when n ≠ 0. Furthermore,<br />
for all m ≥ g N + 1,<br />
f N 2 m(z) − N+1<br />
δ N (m) f m(z) = q −N 2m − N+1<br />
δ N (m) q−m + O(1).<br />
There is a similar duality statement dictating the existence <strong>of</strong> <strong>forms</strong> in M ! 2(N).<br />
Lemma 6.5. For every n ≥ 0, there exists P n,N (z) = ∑ m∈Z c(P n,N, m)q m ∈ M ! 2(N) such<br />
that<br />
(i) At the cusp infinity, P n,N (z) = q −n + O(q).<br />
(ii) (P n,N |W N )(z) = −(P n,N |U N )(z).<br />
Furthermore, if f(z) = ∑ m∈Z c(f, m)qn ∈ M !,new<br />
0 (N), then<br />
∑<br />
δ N (m)c(f, m)c(P n,N , −m) = 0.<br />
m∈Z<br />
Now, we will recall some facts about Heegner points on the <strong>modular</strong> curve from [22] and<br />
their height pairings on the Jacobian. Let X 0 (N) be the natural compactification <strong>of</strong> Y 0 (N),<br />
the open <strong>modular</strong> curve over Q classifying pairs <strong>of</strong> elliptic curves (E, E ′ ) with an order N<br />
cyclic isogeny φ : E → E ′ . The complex points <strong>of</strong> X 0 (N) has a structure <strong>of</strong> a compact<br />
Riemann surface and can be identified with H ∪ P 1 (Q) modulo the action <strong>of</strong> Γ 0 (N). Heegner<br />
points on X 0 (N) corresponds to pair <strong>of</strong> elliptic curves (E, E ′ ) having complex multiplication<br />
by the same ring O in some imaginary quadratic field F . This occurs only when there exists<br />
an O-ideal n dividing N such that O/n is cyclic <strong>of</strong> order N. In this case, E(C) is isomorphic<br />
to C/a with a an invertible O-submodule in F . Since this a can be chosen independent <strong>of</strong><br />
homothety by elements in F × , we only need to consider [a], the class <strong>of</strong> a in Pic(O). So<br />
a Heegner point on X 0 (N) can be expressed as the triplet (O, n, [a]). To find the image <strong>of</strong><br />
such a Heegner point in H, choose an oriented basis 〈ω 1 , ω 2 〉 <strong>of</strong> a such that an −1 has basis<br />
〈ω 1 , ω 2 /N〉. Then τ [a],n := ω 1 /ω 2 ∈ Γ 0 (N)\H is the image <strong>of</strong> this Heegner point.<br />
For our purpose, F = Q( √ −p) and O = O F . Then we require χ p (N) = 1 for Heegner<br />
points to exist. In that case, N splits into nn in F with<br />
n = ZN + Z bn+√ −p<br />
2<br />
.
30 W. DUKE AND Y. LI<br />
Let A ∈ Pic(O F ) = Cl(F ). A point τ ∈ H corresponding to the Heegner point (O, n, A)<br />
must satisfy an equation<br />
Aτ 2 + Bτ + C = 0,<br />
with B 2 − 4AC = −p, N|A, B ≡ b n (mod 2N) and gcd(A/N, B, NC) = 1. We will denote<br />
this image by τ(A, n). Notice τ(A, n) is well-defined up to the action <strong>of</strong> Γ 0 (N).<br />
Let J 0 (N) be the Jacobian <strong>of</strong> X 0 (N). Its complex points is a compact Riemann surface<br />
and can be viewed as the set <strong>of</strong> divisors <strong>of</strong> degree zero modulo the set <strong>of</strong> principal divisors<br />
on X 0 (N). Let 〈, 〉 C be the height symbol on X 0 (N)(C). It is a unique real-valued function<br />
defined on the set <strong>of</strong> divisors <strong>of</strong> degree zero and satisfies<br />
〈∑<br />
ni (x i ), b〉<br />
= ∑ n i log |f(x i )| 2<br />
C<br />
i<br />
if b = (f) is a principal divisor.<br />
For n ≥ 0 and a class B ∈ Cl(F ), denote the inner product<br />
(72) ρ N,B,ψ (n) := 〈j lift,B[n]<br />
n<br />
+ j lift,B[n]−1<br />
n , g ψ 〉 reg .<br />
In particular for j 0 (z) = 1, the inner product ρ N,B,ψ (0) becomes 〈T N (ϑ B ), g ψ 〉, where T N is<br />
the N th Hecke operator on M ! 1(p). Also, notice that<br />
(73) T N (ϑ B )(z) = ϑ B (Nz) + (ϑ B |U N )(z) = ϑ B[n] (z) + ϑ B[n] −1(z).<br />
By Corollary 5.2, we know that ρ N,B,ψ (n) can be written in the form <strong>of</strong> the right hand<br />
side <strong>of</strong> (68) and satisfies the conditions in Proposition 6.1. The following lemma relates<br />
〈f lift,N,B , g ψ 〉 reg to ρ N,B,ψ (n) and an infinite sum analogous to the right hand side <strong>of</strong> (62).<br />
Lemma 6.6. Suppose f(z) ∈ M !,new<br />
0 (N). In the notations above, we have<br />
(74) 〈f lift,N,B , g ψ 〉 reg = ∑ m ′ ≥1<br />
c(f, −Nm ′ )ρ N,B,ψ (m ′ ) + C f ρ N,B,ψ (0) − Σ f,N,B,ψ ,<br />
where<br />
∑<br />
Σ f,N,B,ψ := − lim<br />
s→1<br />
C f =<br />
m≥1<br />
N∤m<br />
(N + 1) lim<br />
(<br />
c(f, −m) ∑ k≥1<br />
s→1<br />
∑<br />
m ′ ≥1<br />
c(f, −Nm ′ ) ∑ k ′ ≥1<br />
Nc(f, 0) + 24 ∑ m ′ ≥1<br />
c(f, −Nm ′ )σ 1 (m ′ )<br />
)<br />
δ(k)r B (pm + Nk)r ψ (k)2Q s−1<br />
(1 + 2kN +<br />
pm<br />
)<br />
δ(k ′ )r B (k ′ )r ψ (Nk ′ − pm ′ )2Q s−1<br />
(−1 + 2k′ N<br />
,<br />
pm ′<br />
)<br />
.<br />
Pro<strong>of</strong>. Let ˜g ψ (z) ∈ M − 1 (p) be a mock-<strong>modular</strong> form as in Proposition 4.2 with Fourier expansion<br />
˜g ψ (z) = ∑ n∈Z r+ ψ (n)qn and ĝ ψ (z) the associated harmonic Maass form. By Theorem<br />
5.1, we have the following identities<br />
)<br />
ρ N,B,ψ (n) = −24σ 1 (n)ρ N,B,ψ (0) + lim δ(k)c(T N (ϑ B ), pm + k)r ψ (k)2Q s−1<br />
(1 + 2k ,<br />
pm<br />
ρ N,B,ψ (n) = ∑ k∈Z<br />
s→1<br />
∑<br />
k≥1<br />
r + ψ (k) ( (<br />
r pn−k<br />
)<br />
B N + rB (pNn − Nk) ) δ(k).
Similar to the pro<strong>of</strong> <strong>of</strong> Theorem 5.1, we define<br />
MOCK-MODULAR FORMS OF WEIGHT ONE 31<br />
Φ N,B (ĝ, z) := Tr Np<br />
N (−i(ĝ|W p)(Nz)ϑ B (z))<br />
= (ĝ|U p )(Nz)ϑ B (z) + (ĝ(z)ϑ B (Nz))|U p .<br />
When N = 1, this is the same as Φ B (ĝ ψ , z) defined in the pro<strong>of</strong> <strong>of</strong> Theorem 5.1. It trans<strong>forms</strong><br />
like a <strong>modular</strong> form <strong>of</strong> level N, <strong>weight</strong> two and has the following Fourier expansion at infinity<br />
Φ N,B (ĝ ψ , z) = ∑ m∈Z(b N,B (ĝ ψ , m, y) + a N,B (ĝ ψ , m))q m ,<br />
b N,B (ĝ ψ , m, y) = − ∑ k∈Z<br />
a N,B (ĝ ψ , m) = ∑ k∈Z<br />
Suppose f(z) has the Fourier expansion<br />
at infinity. Define the sum Σ ′ f,N,B,ψ by<br />
δ(k)r ψ (k)β 1<br />
(<br />
k, Ny<br />
p<br />
r + ψ (k) r B(pm − Nk)δ(k).<br />
f(z) = ∑ m∈Z<br />
c(f, m)q m<br />
(75) Σ ′ f,N,B,ψ := ∑ m∈Z<br />
c(f, −m)δ N (m)a N,B (ĝ ψ , m).<br />
)<br />
r B (pm + Nk),<br />
Since f lift,N,B (z) ∈ M !,+<br />
1 (p), we can apply Proposition 3.4 to obtain<br />
⎛ ∑<br />
〈f lift,N,B , g ψ 〉 reg = ∑ r + ψ (k)r ⎞<br />
B(pm − Nk)δ(k)−<br />
c(f, −m) ⎜<br />
k∈Z<br />
⎝<br />
m∈Z N ∑ r + ψ (k)r ( pm−Nk<br />
)<br />
⎟<br />
⎠<br />
B N δ(k) 2<br />
k∈Z<br />
= −N ∑ c(f, −Nm ′ ) ∑ ( ( )<br />
)<br />
r + ψ (k) pm<br />
r ′ −k<br />
B + r<br />
N B (pNm ′ − Nk) δ(k)<br />
m ′ ∈Z<br />
k∈Z<br />
+ ∑ m∈Z<br />
δ N (m)c(f, −m)a N,B (ĝ ψ , m)<br />
(76)<br />
= −N ∑ m ′ ≥0<br />
c(f, −Nm ′ )ρ N,B,ψ (m ′ ) + Σ ′ f,N,B,ψ<br />
The sum over m ′ ∈ Z changes to be over only m ′ ≥ 0, since r + ψ<br />
(k) = 0 for all k ≤ −p by the<br />
choice <strong>of</strong> ˜g ψ (z).<br />
For n ≥ 0, let P n,N (z) = q −n + O(q) ∈ M 2(N) ! be the weakly holomorphic <strong>modular</strong> form<br />
as in Lemma 6.5. When n = 0, we can take P 0,N (z) ∈ M 2 (N) to be<br />
P 0,N (z) =<br />
Consider Φ ∗ N,B (ĝ ψ, z), defined by<br />
(77)<br />
Φ ∗ N,B(ĝ ψ , z) := Φ N,B (ĝ ψ , z) − ∑ n≥0<br />
(N−1)Ẽ2(Nz) N − (N−1)Ẽ2(z).<br />
1<br />
a N,B (ĝ ψ , −n)P n,N (z) + Nρ N,B,ψ(0)<br />
N 2 − 1<br />
(Ẽ2(Nz) − Ẽ2(z)).<br />
Now it is easy to check that Φ ∗ N,B (ĝ ψ, z) is O(1/y) at the cusp infinity. At the cusp 0, some<br />
calculations show that<br />
Φ N,B (ĝ ψ , z)|W N = −Φ N,B (ĝ ψ , z)|U N + Φ 1,B[n] (ĝ ψ , z) + Φ 1,B[n] −1(ĝ ψ , z)
32 W. DUKE AND Y. LI<br />
From the pro<strong>of</strong> <strong>of</strong> Theorem 5.1, we know that Φ 1,B[n] (ĝ ψ , z) + Φ 1,B[n] −1(ĝ ψ , z) has a constant<br />
term ρ N,B,ψ (0) and no pole. So Φ ∗ N,B (ĝ ψ, z) is also O(1/y) at the cusp 0.<br />
For m ≥ 1, let P m,N (z, s) be the Poincaré series <strong>of</strong> level <strong>one</strong>, <strong>weight</strong> two defined by<br />
∑<br />
P m,N (z, s) := (y s e 2πimz )| 2 γ.<br />
It is characterized by the property that<br />
γ∈Γ ∞\Γ 0 (N)<br />
lim 〈h(z), P m,N(z, s)〉 = mc(h, m)<br />
s→0<br />
for any h(z) = ∑ m≥1 c(h, m)qm ∈ S 2 (N). Since {c(f, −m) : m ≥ 1} ∈ Λ N , we know that<br />
〈<br />
(78) lim h(z), ∑ 〉<br />
δ N (m)<br />
c(f, −m)P<br />
s→0 m m,N(z, s) = 0,<br />
m≥1<br />
δ N (m)<br />
m c(f, −m)P m,N(z, s) ∈ S 2 (N) is 0.<br />
for any h(z) ∈ S 2 (N). So lim s→0<br />
∑m≥1<br />
Now we can consider the inner product between Φ ∗ N,B (ĝ ψ, z) and this linear combination<br />
<strong>of</strong> Poincaré series and obtain<br />
∑<br />
δ<br />
4π lim N (m)<br />
c(f,<br />
s→0 m −m)〈Φ∗ N,B(ĝ ψ , z), P m,N (z, s)〉 = 0.<br />
m≥1<br />
Since Φ ∗ N,B (ĝ ψ, z) is O(1/y) at both cusps, we can apply Rankin-Selberg unfolding as in the<br />
pro<strong>of</strong> <strong>of</strong> Theorem 5.1. The limit on the left hand side above then breaks up into two parts.<br />
The first part is<br />
⎛<br />
a N,B (ĝ ψ , m) − ∑ a N,B (ĝ ψ , −n)c(P n,N , m)<br />
∑<br />
⎜<br />
n≥0<br />
⎟<br />
(79)<br />
m≥1<br />
δ N (m)c(f, −m) ⎜<br />
⎝<br />
− 24Nρ N,B,ψ(0)<br />
N 2 − 1<br />
(<br />
σ1<br />
( m<br />
N<br />
⎞<br />
)<br />
− σ1 (m) ) ⎟<br />
⎠<br />
Since (f|W N )(z) = −N(f|U N )(z) and c(P n,N , 0) = 0 for n ≥ 1, Lemma 6.5 tells us that for<br />
n ≥ 0<br />
∑<br />
δ N (m)c(f, −m)c(P n,N , m) = −δ N (n)c(f, n).<br />
m≥1<br />
So expression (79) becomes<br />
Σ ′ f,N,B,ψ − 24Nρ N,B,ψ(0)<br />
N 2 − 1<br />
∑<br />
δ N (m)c(f, −m) ( (<br />
σ m<br />
)<br />
1 N − σ1 (m) ) .<br />
m≥1<br />
The second part involves the limit <strong>of</strong> an infinite sum, which can be evaluated as<br />
Σ f,N,B,ψ − (N + 1) ∑ m ′ ≥1<br />
c(f, −Nm ′ )(ρ N,B,ψ (m ′ ) + 24σ 1 (m ′ )ρ N,B,ψ (0)).<br />
Adding the last two expressions together, which yields 0, and substituting into (76), we<br />
obtain equation (74). The constant C f appears since<br />
∑ ( (<br />
(80) c(f, 0) = 24<br />
m<br />
)<br />
N 2 Nσ1<br />
−1<br />
N − σ1 (m) ) δ N (m)c(f, −m)<br />
m≥1<br />
from applying Lemma 6.5 to P 0,N (z) ∈ M 2(N) ! and f(z) ∈ M !,new<br />
0 (N). Notice if c(f, −m) ∈ Z<br />
for all m ≥ 1, then the denominator <strong>of</strong> c(f, 0) is independent <strong>of</strong> f.<br />
□
MOCK-MODULAR FORMS OF WEIGHT ONE 33<br />
Pro<strong>of</strong> <strong>of</strong> Proposition 6.1. If f(z) = f 1 (Nz) with f 1 (z) a <strong>modular</strong> function <strong>of</strong> level <strong>one</strong>, then<br />
(f|W N )(z) = (f|U N )(z) and<br />
f lift,N,B (z) = f lift,B<br />
1 (z) = f 1 (pz)ϑ B (z).<br />
This case <strong>of</strong> the proposition follows from Corollary 5.2. By Lemma 6.2, it is enough to prove<br />
the proposition for f(z) ∈ M !,new<br />
0 (N). In that case, some calculations show that<br />
(81) f lift,N,B (z) = −N(f|U N )(pz)ϑ B (Nz) + (f(pz)ϑ B (z)) |U N .<br />
Let χ p (N) = ɛ. It is easy to see that f lift,N,B (z) ∈ M !,ɛ<br />
1 (p). Since g ψ (z) ∈ S + 1 (p), Proposition<br />
3.4 tells us that 〈f lift,N,B , g ψ 〉 reg = 0 when ɛ = −1. So we can suppose χ p (N) = ɛ = 1.<br />
By Lemma 6.4, it is easy to see that M !,new<br />
0 (N) is generated by the set S 1 ∪ S 2 , where<br />
(82)<br />
S 1 := {f m (z) : m ≥ 1 + g N , N 2 ∤ m}<br />
S 2 := {f N 2 m(z) − N+1<br />
δ N (m) f m(z) : m ≥ 1 + g N }.<br />
Let f(z) = f N 2 m(z) − N+1<br />
δ N (m) f m(z) ∈ S 2 . We can uniquely write the integer m as N r m ′ with<br />
r ≥ 0, m ′ ≥ −1 and gcd(N, m ′ ) = 1. Then at the cusp infinity, we have<br />
f(z) = q −N r+2 m ′ − N+1<br />
δ N (N r m ′ ) q−N r m ′ + O(1).<br />
From equation (81), it follows that the −n th Fourier coefficient <strong>of</strong> f lift,N,B (z) can be written<br />
as<br />
(<br />
( ))<br />
( )<br />
pN r m ′<br />
c(f lift,N,B pN<br />
, −n) = −N r r+1 m ′ −n<br />
B − N+1 r N<br />
−n<br />
N<br />
δ N (N r m ′ ) B N<br />
(<br />
)<br />
+ r B (−Nn + pN r+2 m ′ ) − N+1 r δ N (N r m ′ ) B(−Nn + pN r m ′ ) .<br />
When r = 0, we can rewrite the equation above as<br />
c(f lift,N,B , −n) = −(N + 1) ( ( )<br />
r B pm ′ − n N + rB (pm ′ − Nn) )<br />
( )<br />
)<br />
pNm<br />
+<br />
(r ′ −n<br />
B + r<br />
N<br />
B (N(pNm ′ − n))<br />
(<br />
)<br />
= −(N + 1)c T N (j lift,B<br />
m<br />
), −n + c (j ′<br />
Nm ′(pz)T N (ϑ B )(z), −n) .<br />
Notice that j pNm ′(pz)T N (ϑ B )(z) = j lift,B[n]<br />
Nm<br />
(z) + j lift,B[n]−1<br />
′<br />
Nm<br />
(z). Since this is true for all n ≥ 1<br />
′<br />
and the principal part uniquely determines a form in M !,+<br />
f lift,N,B (z) = −(N + 1)T N<br />
(<br />
j lift,B<br />
m ′<br />
)<br />
(z)<br />
+ j lift,B[n]<br />
Nm ′<br />
1 (p) up to <strong>forms</strong> in M + 1 (p), we have<br />
(z) + j lift,B[n]−1 (z) + g f (z)<br />
for some g f (z) ∈ M + 1 (p) with integral linear combination <strong>of</strong> c(f, n)’s as Fourier coefficients.<br />
Since T N is self-adjoint with respect to the regularized inner product and<br />
T N (g ψ )(z) = (ψ([n]) + ψ([n]))g ψ (z)<br />
for N prime with χ p (N) = 1, we can rewrite 〈f lift,N,B , g ψ 〉 reg as<br />
〈f lift,N,B , g ψ 〉 reg = −(N + 1)(ψ([n]) + ψ([n]))〈j lift,B<br />
m ′ , g ψ 〉 reg + ρ N,B,ψ (Nm ′ ) + 〈g f , g ψ 〉.<br />
Nm ′
34 W. DUKE AND Y. LI<br />
When r ≥ 1, the situation is similar. In this case, we have<br />
( )<br />
f lift,N,B (z) = − (N + 1)T N j lift,B<br />
N r m<br />
(z) + (Nj ′ N r−1 m ′(pz) + j N r+1 m ′(pz))T N(ϑ B )<br />
+ g f (z),<br />
〈f lift,N,B , g ψ 〉 reg = − (N + 1)(ψ([n]) + ψ([n]))〈j lift,B<br />
m ′ , g ψ 〉 reg + Nρ N,B,ψ (N r−1 m ′ )<br />
+ ρ N,B,ψ (N r+1 m ′ ) + 〈g f , g ψ 〉,<br />
for some g f (z) ∈ M 1 + (p) with integral linear combination <strong>of</strong> c(f, n)’s as Fourier coefficients.<br />
By Proposition 4.1 and Corollary 5.2, the proposition holds for f(z) ∈ S 2 .<br />
Now, we will prove the proposition when f(z) is in the integral span <strong>of</strong> S 1 and show that<br />
the denominator <strong>of</strong> the constant β f,B in (68) is independent <strong>of</strong> f. This will finish the pro<strong>of</strong><br />
<strong>of</strong> Proposition 6.1. By Lemma 6.2 and Lemma 6.6, it is enough to show that Σ f,N,B,ψ can<br />
also be written in the form <strong>of</strong> the right hand side <strong>of</strong> equation (68). To accomplish that,<br />
we will relate Σ f,N,B,ψ to the automorphic Green’s function using the counting arguments in<br />
Proposition (3.11) from [24]. In their notation, let R N be a subset <strong>of</strong> M 2 (Z) containing all<br />
matrices whose lower left entry is divisible by N. For z, z ′ ∈ H and k ≥ 0, let ρ m N (z, z′ , k) be<br />
the number <strong>of</strong> γ = ( a c d b ) ∈ R N/{±1} such that det(γ) = m and<br />
cosh d(z, γz ′ ) = 1 + 2Nk<br />
pm .<br />
Define the automorphic Green’s function <strong>of</strong> level N to be<br />
∑<br />
G N,s (z, z ′ ) :=<br />
g s (z, γz ′ ) +<br />
γ∈Γ 0 (N)/{±1}<br />
z≠γz ′<br />
where<br />
∑<br />
γ∈Γ 0 (N)/{±1}<br />
z=γz ′<br />
g s (z) := lim<br />
w→z<br />
(g s (z, w) − log |2πiη(z) 4 (z − w)| 2<br />
g s (z),<br />
= − log |2π(z − z)η(z) 4 | 2 + 2 Γ′ Γ′<br />
(s) − 2 (1).<br />
Γ Γ<br />
When gcd(N, m) = 1, the m th Hecke operator T m acts on z ′ in G N,s (z, z ′ ) and defines<br />
G m N,s (z, z′ ). It can be written as<br />
∑<br />
G m N,s(z, z ′ ) :=<br />
g s (z)<br />
γ∈R N /{±1}<br />
det γ=m<br />
= −2 ∑ k>0<br />
γ∈R N /{±1}<br />
det γ=m<br />
z=γz ′<br />
z≠γz ′ g s (z, γz ′ ) + ∑<br />
ρ m N(z, z ′ , k)Q s−1<br />
(1 + 2Nk<br />
pm<br />
)<br />
+ ρ m N(z, z ′ , 0)g s (z).<br />
Notice that ρ m N (z, z′ , k) could be non-zero when k is not an integer. When z, z ′ ∈ H are<br />
Heegner points <strong>of</strong> discriminant −p, then ρ m N (z, z′ , k) is necessarily supported on integral k’s.<br />
For j = 1, 2, let A j ∈ Cl(F ) and τ j ∈ Γ 0 (N)\H be the image <strong>of</strong> the Heegner points<br />
x j = (O F , A j , [n]). Let a j be integral ideals in the class A j such that n|a j , N(a j ) = A j . Then<br />
by Proposition (3.11) in [24], the counting function ρ m N (τ 2, τ 1 , k) satisfies<br />
ρ m N(τ 2 , τ 1 , k) =#{(α, β) ∈ (a −1<br />
1 a −1<br />
2 × a −1<br />
1 a −1<br />
2 n)/{±1} | N F/Q (α) = Nk+pm<br />
A 1 A 2<br />
,<br />
N F/Q (β) = Nk<br />
A 1 A 2<br />
, A 1 A 2 (α − β) ≡ 0 (mod d)},<br />
=r A1 A −1(Nk<br />
+ pm)r A<br />
2 1 A 2 [n] −1(k)δ(k)
MOCK-MODULAR FORMS OF WEIGHT ONE 35<br />
for k ≥ 0, where d = ( √ −p) is the different <strong>of</strong> F . The first equality is established by the<br />
bijection<br />
γ = ( a c d b ) ∈ R N/{±1} ↦→ α = cτ 1τ 2 + dτ 2 − aτ 1 − b,<br />
β = cτ 1 τ 2 + dτ 2 − aτ 1 − b.<br />
The Fricke involution W N sends x j to the Heegner point (O F , A j [n] −1 , [n] −1 ). Set A ′ 1 =<br />
A 1 [n] −1 and a ′ 1 ∈ A ′ 1 a class such that n|a ′ 1, then the same counting argument establishes<br />
ρ m′<br />
N (τ 2 , τ ′ 1, k ′ ) =#{(α, β) ∈ ((a ′ 1) −1 ā −1<br />
2 n × (a ′ 1) −1 a −1<br />
2 )/{±1} | N F/Q (α) = Nk′<br />
A 1 A 2<br />
,<br />
N F/Q (β) = Nk′ −pm ′<br />
A 1 A 2<br />
, A 1 A 2 (α − β) ≡ 0 (mod d)},<br />
=r A ′<br />
1 A 2<br />
(Nk ′ − pm)r A ′<br />
1 A −1<br />
2 [n]−1 (k ′ )δ(k ′ )<br />
for k ′ ≥ pm ′ /N. Apply these counting arguments to G m N,s (τ 2, τ 1 ) and G m′<br />
N,s (τ 2, τ 1), ′ we obtain<br />
G m N,s(τ 2 , τ 1 ) = − 2 ∑ ( )<br />
r A1 A −1(Nk<br />
+ pm)r A<br />
2 1 A 2 [n] −1(k)δ(k)Q s−1 1 + 2Nk +<br />
pm<br />
(83)<br />
k≥1<br />
r A1 A −1(m)g<br />
s(τ 2 ),<br />
2<br />
∑<br />
( )<br />
G m′<br />
N,s(τ 2 , τ 1) ′ = − 2 r A ′<br />
1 A 2<br />
(Nk ′ − pm ′ )r A ′<br />
1 A −1 (k ′ )δ(k ′ )Q<br />
2 [n]−1 s−1 −1 + 2Nk′ +<br />
pm ′<br />
(84)<br />
k ′ >pm ′ /N<br />
r A ′<br />
1 A −1<br />
2 [n]−1 ( m′<br />
N )g s(τ 2 ).<br />
Since N ∤ m ′ in the definition <strong>of</strong> G m′<br />
N,s (z, z′ ), the term r A ′<br />
1 A −1 ( m′ )g<br />
2 [n]−1 N s(τ 2 ) above vanishes.<br />
By Proposition (2.23) in [24, p.242], we can relate the Green’s function to the height pairing<br />
between (x 2 ) − (∞), (x 1 ) − (0) and W N ((x 1 ) − (0)) on the Jacobian as follows<br />
( )<br />
G<br />
m<br />
N,s (τ 2 , τ 1 ) + 4πσ 1 (m)E N (W N τ 2 , s)<br />
〈(x 2 ) − (∞), T m ((x 1 ) − (0))〉 C = lim<br />
s→1 + 4πm s σ 1−2s (m)E N (τ 1 , s) + σ 1(m)κ N<br />
s−1<br />
−σ 1 (m)(λ N + 2κ N ),<br />
( )<br />
G<br />
m<br />
N,s(τ ′<br />
2 , τ 1) ′ + 4πσ 1 (m ′ )E N (τ 2 , s)<br />
〈(x 2 ) − (∞), T m ′W N ((x 1 ) − (0))〉 C = lim<br />
s→1<br />
+ 4π(m ′ ) s σ 1−2s (m ′ )E N (W N τ 1 , s) + σ 1(m ′ )κ N<br />
s−1<br />
−σ 1 (m ′ )(λ N + 2κ N ).<br />
Here E N (z, s) is the Eisenstein series defined in (48) and κ N , λ N are explicit constants dependent<br />
on N.<br />
For A ′ , B ∈ Cl(F ), substitute A 1 = A ′√ B[n], A 2 = A ′√ B −1 [n] and A ′ 1 = A ′√ B[n] −1 into<br />
equations (83) and (84). Denote τ 1 = τ A1 ,n, τ 2 = τ A2 ,n and τ 1 ′ = τ ′ A 1 [n] −1 ,¯n by τ 1(A ′ , B), τ 2 (A ′ , B)<br />
and τ 1(A ′ ′ , B) respectively. Summing together equations (83), (84) over m, m ′ and A ′ with a<br />
character ψ 2 gives us<br />
Σ f,N,B,ψ = ∑ m≥1<br />
N∤m<br />
c(f, −m)<br />
∑<br />
A ′ ∈Cl(F )<br />
− (N + 1) ∑ m ′ ≥1<br />
c(f, −Nm ′ )<br />
ψ 2 (A ′ )(G m N,s(τ 2 (A ′ , B), τ 1 (A ′ , B)) − r B (m)g s (τ 2 (A ′ , B)))<br />
∑<br />
A ′ ∈Cl(F )<br />
ψ 2 (A ′ )G m N,s(τ 2 (A ′ , B), τ ′ 1(A ′ , B))<br />
We can rewrite the sum above in terms <strong>of</strong> height pairings as follows<br />
(85) Σ f,N,B,ψ = ∑<br />
ψ 2 (A ′ )〈(x 2 (A ′ , B)) − (∞), T f ((x 1 (A ′ , B)) − (0))〉 C + U f,B,ψ ,<br />
A ′ ∈Cl(F )
36 W. DUKE AND Y. LI<br />
where x j (A ′ , B) is the preimage <strong>of</strong> the Heegner point τ j (A ′ , B) on X 0 (N) and<br />
⎛<br />
⎞<br />
∑<br />
T f := ⎜<br />
⎝ c(f, −m)T m − (N + 1) ∑ c(f, −Nm ′ )T m ′W N<br />
⎟<br />
⎠ ,<br />
m≥1<br />
m ′ ≥1<br />
N∤m<br />
N∤m ′ (<br />
U f,B,ψ = c f,1ψ(B[n]) + c f,2 ψ(B −1 [n]) + c f,3 ψ(B[n] −1 ) + c f,4 ψ(B −1 [n] −1 ) ∑<br />
ψ 2 (A ′ ) log |u<br />
12H(p)(N 2 A ′|)<br />
− 1)<br />
A ′<br />
for some constants c f,j ∈ Q determined by f(z). Since f(z) ∈ S 1 , the coefficient c(f, −m)<br />
vanishes for all m ≥ 1 satisfying N 2 |m. So we have N ∤ m ′ in the summation above. Using<br />
(49) and equation (2.16) in [24, p. 240], <strong>one</strong> can show that the c f,j ’s are integral linear<br />
combinations <strong>of</strong> c(f, −m), m ≥ 1. Since f is in the integral span <strong>of</strong> S 1 , Lemma 6.4 implies<br />
that c(f, −m) ∈ Z for all m ≥ 1. Then the denominator <strong>of</strong> constant in the definition <strong>of</strong> U f,B,ψ<br />
is independent <strong>of</strong> f.<br />
Given any newform h(z) ∈ S 2 (N), we know that (h|W N )(z) = −(h|U N )(z). It is then<br />
easy to check that T f (h(z)) = 0. Since N is prime, S 2 (N) is spanned by new<strong>forms</strong>. So T f<br />
annihilates any h(z) ∈ S 2 (N). Then T f ((x 1 (A, B)) − (0)) is a trivial divisor on J 0 (N) defined<br />
over H, since the actions <strong>of</strong> the Hecke operators and Fricke involution on the Jacobian are<br />
the same as those on S 2 (N). So it is the divisor <strong>of</strong> a rational function φ f,A,B on X 0 (N),<br />
defined over H. By the axioms defining height pairings, we can rewrite Σ f,N,B,ψ as<br />
Σ f,N,B,ψ =<br />
∑<br />
ψ 2 (A) log |u f,B (A)| + U f,B,ψ ,<br />
where<br />
A∈Cl(F )<br />
u f,B (A) = φ f,A,B(τ 2 (f, A, B))<br />
∈ H.<br />
φ f,A,B (∞)<br />
Given any σ C ∈ Gal(H/F ) associated to C ∈ Cl(F ), it sends the Heegner point x 1 (A, B) to<br />
the Heegner point x 1 (AC −1 , B). Since the Galois action commutes with the Hecke action, we<br />
see that<br />
φ σ C<br />
f,A,B (z)<br />
φ f,AC −1 ,B(z) ∈ H×<br />
is a constant. So σ C (u f,B (A)) = u f,B (AC −1 ). Thus, the proposition holds for all f(z) in the<br />
integral span S 1 .<br />
□<br />
7. Pro<strong>of</strong> <strong>of</strong> Theorem 1.1<br />
From §5 and §6, we know that the regularized inner product between f lift,N,B (z) ∈ M !,+<br />
1 (p)<br />
and g ψ (z) can be put into a form quite similar to (10) and satisfies condition (iii) in Theorem<br />
1.1, whenever f(z) ∈ M 0(N) ! has rational Fourier coefficients and N is either 1 or a prime<br />
number different from p. For a mock-<strong>modular</strong> form ˜g ψ (z) = ∑ n∈Z r+ ψ (n)qn as in Proposition<br />
4.2, (76) gives rise to the following equation involving a linear combination <strong>of</strong> r + ψ (n)’s<br />
∑<br />
δ(k)r B (pm − Nk)r + ψ (k) =〈f lift,N,B , g ψ 〉 reg<br />
(86)<br />
m∈Z<br />
δ N (m)c(f, −m) ∑ k∈Z<br />
+ N ∑ m ′ ≥0<br />
c(f, −Nm ′ )ρ N,B,ψ (m ′ ).
MOCK-MODULAR FORMS OF WEIGHT ONE 37<br />
In this section, we will show that these linear combinations <strong>of</strong> r + ψ<br />
(n)’s are sufficient to determine<br />
˜g ψ (z) up to some element in S1 − (p). Then, <strong>one</strong> can choose an appropriate ˜g ψ (z)<br />
satisfying Theorem 1.1. The following lemma is key to the pro<strong>of</strong> <strong>of</strong> Theorem 1.1. Recall that<br />
δ(n) is the same as in equation (12) and δ N (m) appears in Definition 6.3(ii).<br />
Lemma 7.1. Let N 0 = 1 and {N j : 1 ≤ j ≤ (p − 1)/2} be a set <strong>of</strong> primes congruent to 1<br />
modulo N. Suppose {d(n)} n≥1 is a set <strong>of</strong> complex numbers such that d(n) = 0 for all n with<br />
χ p (n) = 1 and they satisfy the following equations<br />
∑<br />
m∈Z<br />
δ N (m)c(f, −m) ∑ n≥1<br />
∑<br />
δ(n)d(n)r B (pm − n) = 0,<br />
n≥1<br />
δ(n)d(n)r B (pm − Nn) = 0,<br />
for all classes B ∈ Cl(F ) and f(z) ∈ M !,new<br />
0 (N j ), 1 ≤ j ≤ (p − 1)/2. Then the power series<br />
D(z) := ∑ n≥1 d(n)qn is a <strong>weight</strong> <strong>one</strong> cusp form in S − 1 (p).<br />
Pro<strong>of</strong>. Define the sum s j,B (m) for m ≥ 1 and the series φ j,B (z) by<br />
s j,B (m) := ∑ n≥1<br />
φ j,B (z) := ∑ m≥1<br />
s j,B (m)q m .<br />
δ(n)d(n)r B (pm − N j n),<br />
First, it is not hard to see that φ j,B (z) ∈ S 2 (N j ). When j = 0, this is clear since φ j,B (z) is identically<br />
0. When 1 ≤ j ≤ (p−1)/2, N j is a prime and a power series h(z) = ∑ m≥1 c(h, m)qm ∈<br />
CJqK is in S 2 (N j ) if and only if<br />
∑<br />
δ Nj (m)c(f, −m)c(h, m) = 0,<br />
m∈Z<br />
for all f(z) = ∑ m∈Z c(f, m)qm ∈ M !,new<br />
0 (N j ). The sufficiency is a consequence <strong>of</strong> Lemma 6.4.<br />
Next, it is easy to see that<br />
(87)<br />
φ j,B (z) = (ϑ B (z)D(N j z))|U p + ϑ B (z)(D|U p )(N j z),<br />
= 1 p<br />
k=0<br />
p−1<br />
∑ ( (ϑ B<br />
) )<br />
z+k<br />
+ ϑ<br />
p B (z) D<br />
(<br />
Nj z+k<br />
p<br />
)<br />
where U p acts formally on power series in CJqK and we used N j ≡ 1 (mod p) to reach the<br />
second step. Set<br />
M :=<br />
(p+1)/2<br />
∏<br />
j ′ =1<br />
M j := M N j<br />
.<br />
N j ′,
38 W. DUKE AND Y. LI<br />
Then φ j,B (pM j z) is a <strong>modular</strong> form <strong>of</strong> <strong>weight</strong> two, level pM for all 0 ≤ j ≤ (p − 1)/2. Define<br />
the following matrices with entries in CJqK by<br />
( )<br />
)<br />
L :=<br />
(ϑ B M j z + k + ϑ<br />
p B (pM j z)<br />
( ( ))<br />
X := D Mz + k p<br />
0≤k≤p−1<br />
R := (pφ j,B (pM j z)) 0≤j≤(p−1)/2<br />
.<br />
0≤j≤(p−1)/2<br />
0≤k≤p−1.<br />
The dimensions <strong>of</strong> L, X and R are (p + 1)/2 × p, p × 1 and (p + 1)/2 × 1 respectively. Then<br />
we can rewrite equation (87) in the following matrix expression<br />
(88) L · X = R.<br />
Notice that matrix L has rank (p + 1)/2. This can be seen by considering the product<br />
(<br />
)<br />
( )<br />
∑<br />
1 /p<br />
L ·<br />
p e−2πikk′ = δ(k ′ ) r B (n)q M jn<br />
,<br />
0≤k,k ′ ≤p−1<br />
n≡k ′ mod p<br />
0≤j≤(p−1)/2<br />
0≤k ′ ≤p−1.<br />
which, after permuting columns, is composed <strong>of</strong> a (p + 1)/2 × (p − 1)/2 zero submatrix and<br />
a (p + 1)/2 × (p + 1)/2 submatrix <strong>of</strong> the form<br />
(<br />
)<br />
∑<br />
L ′ := δ(n ν ) r B (n)q M jn<br />
.<br />
n≡n ν mod p<br />
0≤j,ν≤(p−1)/2<br />
Here n 0 = 0 and 0 < n 1 < n 2 < · · · < n (p−1)/2 < p are all the residues modulo p. The<br />
determinant <strong>of</strong> this submatrix is a nonzero power series in q with leading term<br />
⎛<br />
⎞<br />
(p−1)/2<br />
∏<br />
⎝ r B (n B,ν ) ⎠ q ∑ 1≤j≤(p−1)/2 M jn B,j<br />
,<br />
ν=1<br />
where {n B,ν } 1≤ν≤(p−1)/2 is a set <strong>of</strong> positive integers such that<br />
(i) 0 < n B,1 < n B,2 < · · · < n B,(p−1)/2 ,<br />
(ii) r B (n B,ν ) ≠ 0 for all 1 ≤ ν ≤ (p − 1)/2,<br />
(iii) r B (n) = 0 if n ≡ n B,ν ′ (mod p) for some ν ′ and n < n B,ν ′.<br />
So the submatrix L ′ has full rank in M (p+1)/2,(p+1)/2 (CJqK) and the null space <strong>of</strong> L is spanned<br />
by the column vectors<br />
(<br />
1<br />
p e−2πikk′ /p<br />
)<br />
0≤k≤p−1<br />
with k ′ running over the (p − 1)/2 non-residues<br />
modulo p. This also shows that L has rank (p + 1)/2 in M (p−1)/2,(p−1) (C) after substituting<br />
q = e 2πiz for all z in a dense subset U ∈ H.<br />
Let γ = ( )<br />
a b<br />
pMc d ∈ Γ0 (pM), then applying | 2 γ to both sides <strong>of</strong> equation (87) gives us a new<br />
equation with the same right hand side. Let c denote the multiplicative inverse <strong>of</strong> c modulo<br />
p and we have the following standard transformation <strong>of</strong> ϑ B (z)<br />
{<br />
(<br />
)<br />
) (<br />
ϑ B<br />
(M j z + k | a b<br />
) χ<br />
p 1 pMc d = p (a + ck)ϑ B M j z + (a+ck)dk , if a + ck ≢ 0 mod p<br />
p<br />
i √ pχ p (−c)ϑ B (pM j z),<br />
if a + ck ≡ 0 mod p<br />
(<br />
ϑ B (pM j z) | a b<br />
)<br />
χ √ )<br />
1 pMc d = p(−c)i p<br />
ϑ<br />
p B<br />
(M j z + cd .<br />
p<br />
Using these properties, equation (88) trans<strong>forms</strong> into<br />
L · T · X ′ = R,
where<br />
T := (T kk ′) 0≤k,k ′ ≤p−1,<br />
T kk ′ := χp(−c)i √ p<br />
(δ cd(k) ′ − gp(k) ) +<br />
p<br />
δ ′ α(β) :=<br />
{<br />
1, if α = β<br />
0, otherwise<br />
MOCK-MODULAR FORMS OF WEIGHT ONE 39<br />
p−1<br />
∑<br />
{ p<br />
g p (k) := 1 (1 + χ<br />
2<br />
p (n))e 2πink/p =<br />
, if p | k<br />
2<br />
i √ pχ p(k)<br />
, if p ∤ k<br />
n=0<br />
2<br />
( ( ) )<br />
X ′ := D Mz + k′ |<br />
p 1 γ<br />
0≤k ′ ≤p−1<br />
{<br />
χp (a + ck ′ )(δ ′ (a+ck ′ )dk ′(k′ ) − gp(k) ),<br />
p<br />
if k ≢ −ac mod p<br />
iχ p(−c)<br />
2 √ g p p(k), if k ≡ −ac mod p<br />
So when q = e 2πiz for any z ∈ H, the column vector X − T · X ′ with entries in C is in)<br />
the null<br />
space <strong>of</strong> L. If z ∈ U, the null space is spanned by the column vectors<br />
with k ′ ∈ Z and χ p (k ′ ) = −1.<br />
Now define a 1 × p row vector V by<br />
(<br />
1<br />
V := (g p p(k) + 1) .<br />
2<br />
)0≤k≤p−1<br />
It is orthogonal to the null space <strong>of</strong> L when z ∈ U. So for all z ∈ U,<br />
Simple calculation shows that<br />
V · X = V · T · X ′ .<br />
V · T = χ p (d)V,<br />
V · X = χ p (d)D(Mz),<br />
V · X ′ = χ p (d)(D(Mz)| 1 γ).<br />
(<br />
1<br />
p e−2πikk′ /p<br />
0≤k≤p−1<br />
So we know that (D(Mz)| 1 γ) = χ p (d)D(Mz) for all z ∈ U. Since U ⊂ H is dense and both<br />
sides are holomorphic, we have (D(Mz)| 1 γ) = χ p (d)D(Mz) for all z ∈ H. This is true for<br />
any γ ∈ Γ 0 (pM), or equivalently<br />
for all ( )<br />
a Mb<br />
pc d ∈ Γ0 (p), b ∈ Z.<br />
Let Γ 0 (p, M) := {γ = ( a b<br />
(<br />
(D| a Mb<br />
1 pc d<br />
)<br />
)(z) = χp (d)D(z)<br />
c d ) : p|a, M|b}. It is not too hard to see that Γ 0(p) is generated<br />
by Γ 0 (p, M) and T = ( 1 1 1 ) ∈ Γ 0 (p). Since D(z) is also invariant under the action <strong>of</strong> T , it<br />
has level p. By the shape <strong>of</strong> the Fourier expansion <strong>of</strong> D(z), it is in S1 − (p).<br />
□<br />
Corollary 7.2. Let N 0 = 1 and N j be distinct primes congruent to 1 modulo p for 1 ≤ j ≤<br />
(p − 1)/2. Then the space M !,+<br />
1 (p) is spanned by the set<br />
{f lift,N j,B (z) : B ∈ Cl(F ), f(z) ∈ M !,+<br />
0 (N j ), 0 ≤ j ≤ (p − 1)/2}.<br />
Pro<strong>of</strong>. Denote the subspace <strong>of</strong> M !,+<br />
1 (p) spanned by the set above by M !,+,lift<br />
1 (p). Suppose<br />
there exists f(z) ∈ M !,+<br />
1 (p) not in M !,+,lift<br />
1 (p). Then by the perfect pairing between M !,+<br />
1 (p)<br />
and S1 − (p), there exists<br />
D(z) = ∑ d(n)q n ∈ CJqK<br />
n≥1<br />
such that
40 W. DUKE AND Y. LI<br />
(i) d(n) = 0 for all n with χ p (n) = 1,<br />
(ii) ∑ n∈Z c(H, n)d(n)δ(n) = 0 for all H(z) = ∑ n∈Z c(H, n)qn ∈ M !,+,lift<br />
1 (p),<br />
(iii) D(z) ∉ S1 − (p).<br />
In particular for H(z) = f lift,N,B (z), condition (ii) can be rewritten as<br />
(89) ∑ m∈Z<br />
c(f, −m) ∑ k∈Z<br />
Also, when H(z) = j lift,B[n]<br />
m ′<br />
∑<br />
δ(k)d(k)<br />
δ(k)d(k)r B (pm − Nk) = N ∑ m ′ ∈Z<br />
k∈Z<br />
So equation (89) can be written as<br />
∑<br />
c(f, −Nm ′ ) ∑ k∈Z<br />
(z) + j lift,B[n]−1<br />
m<br />
(z), we have<br />
(r ′<br />
( )<br />
)<br />
B + r B (pNm ′ − Nk) = 0.<br />
pm ′ −k<br />
N<br />
m∈Z<br />
δ N (m)c(f, −m) ∑ k∈Z<br />
δ(k)d(k)r B (pm − Nk) = 0.<br />
δ(k)d(k)r B<br />
(<br />
By Lemma 7.1, D(z) ∈ S − 1 (p). This contradicts condition (iii) and we must have<br />
M !,+,lift<br />
1 (p) = M !,+<br />
1 (p).<br />
)<br />
pm ′ −k<br />
.<br />
N<br />
With Proposition 6.1 and Lemma 7.1, we are ready to give the pro<strong>of</strong> <strong>of</strong> Theorem 1.1.<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 1.1. Denote the dimension and the q-echelon basis <strong>of</strong> S − 1 (p) by d − and the<br />
basis by<br />
{h t (z) ∈ S − 1 (p) : h t (z) = q nt + O(q n d − +1<br />
), 1 ≤ t ≤ d − }.<br />
Choose a mock-<strong>modular</strong> form ˜g ψ (z) = ∑ n≥n 0<br />
r + ψ (n)qn such that it has a fixed principal part as<br />
in Proposition 4.2 and r + ψ (n t) satisfies conditions (ii) and (iii) in Theorem 1.1 for 1 ≤ t ≤ d − .<br />
Note that if S1 − (p) = ∅, then there is only <strong>one</strong> ˜g ψ (z) with a fixed principal part. We claim<br />
that this ˜g ψ (z) is a desired choice.<br />
Consider all f(z) ∈ M !,new<br />
0 (N j ), 1 ≤ j ≤ (p − 1)/2 with rational Fourier coefficients and<br />
j m (z) for all m ≥ 0. They generate the complex vector space M !,new<br />
0 (N j ) and M 0(1) ! respectively.<br />
They also give rise to equation (86) and equation (66), which form a system <strong>of</strong> equations<br />
satisfied by {r + ψ (n) : n ≥ n 0}. Suppose there is another set <strong>of</strong> solutions {c + (n) : n ≥ n 0 }<br />
with r + ψ (n) = c+ (n) whenever n ≤ 0, χ p (n) = 1, or n = n t for t = 1, . . . , d − . Then by Lemma<br />
7.1, the power series<br />
D(z) := ∑ n≥1<br />
(r + ψ (n) − c+ (n))q n<br />
is a cusp form in S1 − (p). However, since r + ψ (n t) = c + (n t ) for all 1 ≤ t ≤ d − , the cusp<br />
form D(z) must be zero in S1 − (p). So with {r + ψ (n) : n ≤ 0 or n = n t, 1 ≤ t ≤ d − } fixed,<br />
{r + ψ (n) : n ≥ d − + 1} is the unique solution to this system <strong>of</strong> equations. From Corollary<br />
7.2 and Proposition 3.5, it is clear that r + ψ<br />
(n) can be solved from this system <strong>of</strong> equations,<br />
i.e. the r + ψ<br />
(n)’s are a rational linear combination <strong>of</strong> these regularized inner products. By<br />
Corollary 5.2 and Proposition 6.1, we can write<br />
r + ψ (n) = −β ∑<br />
n ψ 2 (A) log |u(n, A)|,<br />
A∈Cl(F )<br />
□
MOCK-MODULAR FORMS OF WEIGHT ONE 41<br />
with β n ∈ Q, u(n, A) ∈ H independent <strong>of</strong> ψ and σ C (u(n, A)) = u(n, AC −1 ) for all C ∈ Cl(F ).<br />
Now to prove that we can choose β n independent <strong>of</strong> n, it is enough to show that there is a<br />
choice <strong>of</strong> β n whose denominators do not depend on n.<br />
When n > p · g N /N, it is not hard to see from equation (86) that we can explicitly express<br />
r + ψ (n) in terms <strong>of</strong> these regularized inner products and the r+ ψ<br />
(k)’s with k < n. If n = pn′<br />
with n ′ ∈ Z, then substituting f = j n ′(z) into equation (66) gives us<br />
r + ψ (n) = 〈f lift,B , g ψ 〉 reg − ∑ k l such that kN + l ≡ 0<br />
(mod p) and set m = (l + Nk)/p. Notice that the conditions on N and m forces N ∤ m and<br />
χ p (N) = 1. With such a choice <strong>of</strong> N and m, let f m ∈ S 1 as in (82). Then δ(n)δ N (m)r B (pm −<br />
Nn) = 1 and equation (86) becomes<br />
r + ψ (n) = − ∑ (<br />
gN<br />
)<br />
∑<br />
δ(k) δ N (m ′′ )c(f m , −m ′′ )r B (pm ′′ − Nk) r + ψ (k)<br />
k>0 m ′′ =1<br />
− ∑ ( )<br />
∑<br />
δ(k) δ N (m ′′ )c(f m , −m ′′ )r B (pm ′′ − Nk) r + ψ (k)<br />
k≤0 m ′′ ≤g N<br />
+ 〈f lift,N,B<br />
m , g ψ 〉 reg + Nc(f m , 0)ρ N,B,ψ (0).<br />
When k > 0 in the summation on the right hand side, the coefficient <strong>of</strong> r + ψ<br />
(k) is an integer<br />
by Lemma 6.4. Similarly when k ≤ 0, the denominator <strong>of</strong> the coefficient <strong>of</strong> r + ψ<br />
(k) divides<br />
(N 2 − 1). Thus after choosing β n for n ≤ p · g N /N , we can choose the other β n such that<br />
their denominators are independent <strong>of</strong> n.<br />
Now choose some β and u(n, A) for all n, A such that r + ψ<br />
(n) is expressed in terms <strong>of</strong> β and<br />
u(n, A) for all non-trivial character ψ as in equation (10). Define r + A<br />
(n) by<br />
(<br />
r + 1<br />
A<br />
(n) := R + H(p) p (n) +<br />
∑<br />
)<br />
ψ(A)r + ψ (n)<br />
= 1<br />
H(p)<br />
= 1<br />
H(p)<br />
ψ non-trivial<br />
⎛<br />
⎞<br />
∏<br />
⎝R p + (n) + β log<br />
u(n, A ′ )<br />
⎠ − β log |u(n, √ A)|<br />
∣A ′ ∈Cl(F ) ∣<br />
(<br />
R<br />
+<br />
p (n) + β log |N H/F (u(n, A))| ) − β log |u(n, √ A)|,<br />
where the last step follows from the transitive action <strong>of</strong> Gal(H/F ). Clearly, the generating<br />
series ∑ n∈Z r+ A (n)qn is a mock-<strong>modular</strong> form with shadow ϑ A (z). When n ≠ 0, there exists<br />
some b ∈ Q and c n ∈ Q independent <strong>of</strong> A such that<br />
1<br />
H(p)<br />
(<br />
R<br />
+<br />
p (n) + β log |N H/F (u(n, A))| ) = b log |c n |.<br />
This follows from equation (6). Thus, we could choose a different β ′ ∈ Q and u ′ (n, A) ∈ H<br />
such that they satisfy condition (iii) and<br />
−β ′ log |u ′ (n, A)| = b log |c n | − β log |u(n, A)|.<br />
Then r + A (n) = −β′ log |u ′ (n, √ A)| and it is not hard to see that β ′ and u ′ (n, A) satisfies<br />
condition (ii) and (iii) in Theorem 1.1. When ψ is a non-trivial character, ψ 2 is non-trivial
42 W. DUKE AND Y. LI<br />
and we have<br />
r + ψ (n) = −β ∑ A<br />
ψ 2 (A) log |u(n, A)| = −β ′ ∑ A<br />
ψ 2 (A) log |u ′ (n, A)|.<br />
By the choice <strong>of</strong> β ′ and u ′ (n, A), we also have<br />
(90)<br />
∑<br />
R p + (n) = r + A (n)<br />
A∈Cl(F )<br />
∑<br />
= −β ′ log |u ′ (n, A)|<br />
A∈Cl(F )<br />
= − β′<br />
2 log |N H/Q(u ′ (n, A))|.<br />
This is an analogue <strong>of</strong> equation (9) for mock-<strong>modular</strong> <strong>forms</strong> as menti<strong>one</strong>d in the introduction.<br />
□<br />
8. Counting CM points<br />
In preparation for the pro<strong>of</strong> <strong>of</strong> Theorem 1.2, we will count the number <strong>of</strong> CM points on a<br />
hyperbolic circle in terms <strong>of</strong> the representation numbers <strong>of</strong> positive definite binary quadratic<br />
<strong>forms</strong>. As before, such a counting argument is needed to construct a Green’s function at<br />
special points. This construction follows ideas <strong>of</strong> [23], but in the counting argument given<br />
there <strong>one</strong> also sums over all classes <strong>of</strong> a given discriminant. Even when <strong>one</strong> discriminant is<br />
prime, which we are assuming, their pro<strong>of</strong> involves quite an intricate application <strong>of</strong> algebraic<br />
number theory. Surprisingly, the refined version we need for a fixed class has a completely<br />
elementary pro<strong>of</strong> using the classical theory <strong>of</strong> composition <strong>of</strong> binary quadratic <strong>forms</strong>. It<br />
has the further advantage that it applies without extra effort when the other discriminant<br />
is not fundamental. Although we will not give details here, the argument adapts to give a<br />
corresponding refinement <strong>of</strong> the level N case in [24].<br />
Good references for the theory <strong>of</strong> binary quadratic <strong>forms</strong> are the books by Buell [3] and<br />
Cox [14]. Let −d < 0 be a discriminant, not necessarily fundamental, and Q −d the set <strong>of</strong><br />
positive definite integral binary quadratic <strong>forms</strong><br />
Q(x, y) = ax 2 + bxy + cy 2<br />
with −d = b 2 − 4ac, where a, b, c ∈ Z. For any Q ∈ Q −d the associated CM point is defined<br />
to be<br />
τ Q = −b + √ −d<br />
∈ H,<br />
2a<br />
where H is the upper half plane. Clearly Q(τ Q , 1) = 0. The <strong>modular</strong> group Γ = PSL 2 (Z) acts<br />
on Q ∈ Q −d by a linear change <strong>of</strong> variables, which induces linear fractional transformation<br />
on τ Q . Let w Q be the number <strong>of</strong> stabilizers <strong>of</strong> Q. The number <strong>of</strong> equivalence classes <strong>of</strong><br />
quadratic <strong>forms</strong> is the Hurwitz class number H(d). Those classes represented by primitive<br />
<strong>forms</strong> (those with gcd(a, b, c) = 1) comprise a finite abelian group under composition, the<br />
class group. When −d is fundamental, this class group is canonically isomorphic to the ideal<br />
class group <strong>of</strong> Q( √ −d) by sending [Q] to the class A ∈ Cl(Q( √ −d)) containing the fractional<br />
ideal Z + Zτ Q . For primitive Q ∈ Q −d , let Q 2 denote a representative <strong>of</strong> the square class <strong>of</strong><br />
Q.<br />
Associated to Q is the counting function<br />
r Q (k) = 1 2 #{(x, y) ∈ Z2 | Q(x, y) = k},
MOCK-MODULAR FORMS OF WEIGHT ONE 43<br />
which is finite and can be non-zero only for non-negative integers k. Clearly r Q (k) is a class<br />
invariant. When −d is fundamental and A ∈ Cl(Q( √ −d)) is the class associated to [Q], the<br />
counting function r Q (k) is the same as r A (k), the number <strong>of</strong> integral ideals in the class A<br />
having norm equals to k.<br />
The invariant hyperbolic distance d(τ, τ ′ ) between two points τ = x + iy and τ ′ = x ′ + iy ′<br />
in H is defined by (54). The goal now is to count the CM points τ Q ′ <strong>of</strong> a given discriminant<br />
−d that are at a fixed hyperbolic distance from a fixed CM point τ Q <strong>of</strong> a (possibly) different<br />
negative discriminant −D. The number we will study is given by<br />
(91) ρ Q (k, d) = #{Q ′ ∈ Q −d | cosh d(τ Q ′, τ Q ) = k √<br />
dD<br />
}.<br />
Since the hyperbolic distance is invariant under linear fractional transformation, the quantity<br />
ρ Q (k, d) depends only on the class <strong>of</strong> Q. In case −D is the negative <strong>of</strong> a prime we will give<br />
a formula for ρ Q in terms <strong>of</strong> the counting function r Q 2 for the squared class <strong>of</strong> the quadratic<br />
form Q.<br />
Proposition 8.1. Let Q be a positive definite binary quadratic form <strong>of</strong> discriminant −D<br />
where D = p ≡ 3 (mod 4) is a prime. Suppose that −d < 0 is a discriminant and that k ≥ 0.<br />
Then<br />
(92) ρ Q (k, d) = δ(k) r Q 2( k2 −pd<br />
4<br />
).<br />
Our pro<strong>of</strong> <strong>of</strong> this result is an entirely elementary exercise in the classical theory <strong>of</strong> binary<br />
quadratic <strong>forms</strong>. It relies on a remarkable algebraic identity given in (97) below for a particular<br />
quadratic form in the square class, whose existence can be coaxed out <strong>of</strong> the work <strong>of</strong><br />
Dirichlet [16].<br />
Lemma 8.2. Every class <strong>of</strong> primitive positive definite binary quadratic <strong>forms</strong> <strong>of</strong> discriminant<br />
−D contains a representative <strong>of</strong> the form<br />
(93) Q(x, y) = Ax 2 + Bxy + ACy 2<br />
with −D = B 2 − 4A 2 C, A > 0 and gcd(A, B) = 1. Furthermore, the class <strong>of</strong> the square <strong>of</strong> Q<br />
is represented by the form<br />
(94) Q 2 (x, y) = A 2 x 2 + Bxy + Cy 2 .<br />
Pro<strong>of</strong>. It is a well–known result <strong>of</strong> Gauss ([21, §228], see also [3, Prop 4.2 p.50] ) that a<br />
primitive form <strong>of</strong> discriminant −D properly represents a positive integer A prime to −D and<br />
is thus, upon an appropriate change <strong>of</strong> variables, equivalent to a form <strong>of</strong> the shape (A, b, c)<br />
with b 2 − 4Ac = −D. In particular we have that gcd(A, b) = 1.<br />
Next, by means <strong>of</strong> a simple translation transformation x ↦→ x + ty, y ↦→ y we will arrange<br />
that (A, b, c) ∼ (A, B, AC) as desired. This transformation leaves A al<strong>one</strong> and changes b to<br />
B = b + 2At. We now choose t to force<br />
Since c = b2 +D<br />
4A<br />
B 2 = (b + 2At) 2 ≡ −D (mod 4A 2 ).<br />
this is equivalent to solving for t the congruence<br />
tb ≡ c<br />
(mod A),<br />
which is possible since gcd(b, A) = 1.<br />
The fact that (A 2 , B, C) represents the square class <strong>of</strong> (A, B, AC) is a consequence <strong>of</strong> a<br />
classical result <strong>of</strong> Dirichlet on the convolution <strong>of</strong> “united” <strong>forms</strong> (see [3, p.57]). □
Thus<br />
(An<br />
− (2A 2 x + By) )( An + (2A 2 x + By) ) ≡A 2 n 2 − (2A 2 x + By) 2<br />
44 W. DUKE AND Y. LI<br />
Turning now to the pro<strong>of</strong> <strong>of</strong> Proposition 8.1, it is easily checked using (54) that for<br />
Q ′ (x, y) = ax 2 + bxy + cy 2 ∈ Q −d<br />
and Q ∈ Q −D as in (93) above we have<br />
√<br />
(95)<br />
dD cosh d(τQ ′, τ Q ) = 2Ac + 2ACa − Bb.<br />
It follows that the statement <strong>of</strong> Theorem 8.1 is equivalent to the equality<br />
(96) #{(a, b, c) ∈ Z 3 | b 2 − 4ac = −d and 2Ac + 2ACa − Bb = k} = δ(k) r Q 2( k2 −dD<br />
4<br />
),<br />
when D = p. Note that a > 0 for any (a, b, c) in the set since k ≥ 0.<br />
In order to prove this we will establish a direct bijection between the solutions to the<br />
relevant equations. A calculation verifies the truth <strong>of</strong> the following key identity:<br />
(97) 4 Q 2 (x, y) = (2Ac + 2ACa − Bb) 2 − (B 2 − 4A 2 C)(b 2 − 4ac)<br />
where Q 2 is as in (94) above and<br />
x =c − Ca<br />
y =Ba − Ab.<br />
Thus every solution (a, b, c) from (96) gives rise to a solution (x, y) <strong>of</strong><br />
(98) Q 2 (x, y) = k2 − dD<br />
.<br />
4<br />
In fact, there is no need to assume that D = p for this part <strong>of</strong> the argument.<br />
This assumption simplifies our treatment <strong>of</strong> the converse, to which we now turn. Suppose<br />
we are given a solution (x, y) <strong>of</strong> (98). The result is trivial when d = pm 2 and k = pm for<br />
some integer m, so assume otherwise. Then (−x, −y) is another distinct solution. Observe<br />
that since p = 4A 2 C − B 2 we have<br />
4A 2 Q 2 (x, y) ≡ A 2 (4A 2 x 2 + 4Bxy + 4Cy 2 ) ≡ (2A 2 x + By) 2<br />
(mod p).<br />
Since Q is assumed to be primitive we know that p ∤ A.<br />
If p ∤ k we can find an integer a such that<br />
An = ±(2A 2 x + By) + pa<br />
≡A 2 n 2 − 4A 2 Q 2 (x, y) ≡ 0<br />
(mod p).<br />
in precisely <strong>one</strong> <strong>of</strong> the two cases <strong>of</strong> ±. Suppose as we may that An = 2A 2 x + By + pa. Thus<br />
we have<br />
a = An − 2A2 x − By<br />
.<br />
p<br />
Then since p = 4A 2 C − B 2 we have<br />
k = 2A(x + Ca) + 2ACa − B Ba − y<br />
A .<br />
Since gcd(A, B) = 1 we must have that A | (Ba − y) and we may define<br />
b =(Ba − y)/A<br />
c =x + Ca.
MOCK-MODULAR FORMS OF WEIGHT ONE 45<br />
Then we have 2Ac + 2ACa − Bb = k. A computation also shows that b 2 − 4ac = −d. If p | k<br />
then we get two distinct triples (a, b, c) for ±(x, y) which both satisfy these. Obviously the<br />
definition <strong>of</strong> (a, b, c) inverts the map we started with to get (x, y), at least for those pairs<br />
that correspond to an (a, b, c). This finishes the pro<strong>of</strong> <strong>of</strong> Proposition 8.1.<br />
To give a level N version <strong>of</strong> Proposition 8.1, we can start with the more general identity<br />
4N Q ∗ (x, y) = (2ANc + 2ACNa − Bb) 2 − (B 2 − 4A 2 NC)(b 2 − 4Nac)<br />
where Q ∗ (x, y) = A 2 Nx 2 + Bxy + Cy 2 and<br />
x =c − Ca<br />
y =Ba − Ab.<br />
9. Pro<strong>of</strong> <strong>of</strong> Theorem 1.2<br />
In this section we will prove Theorem 1.2. It will be deduced as a corollary <strong>of</strong> a more<br />
general identity for special values <strong>of</strong> certain Borcherds lifts <strong>of</strong> <strong>weight</strong> 1/2 weakly holomorphic<br />
<strong>forms</strong>. We again apply the Rankin method to <strong>forms</strong> with poles. We remark that the method<br />
also yields another pro<strong>of</strong> <strong>of</strong> (12), whose analytic pro<strong>of</strong> in [23] uses Hecke-Eisenstein series <strong>of</strong><br />
<strong>weight</strong> <strong>one</strong>. In fact, the method can be generalized to higher levels in order to obtain some<br />
refinements <strong>of</strong> the type we have been considering <strong>of</strong> certain results <strong>of</strong> [25] on height pairings<br />
<strong>of</strong> Heegner divisors.<br />
Let M<br />
1/2 ! be the space <strong>of</strong> weakly holomorphic <strong>modular</strong> <strong>forms</strong> <strong>of</strong> <strong>weight</strong> 1/2 and level 4<br />
satisfying Kohnen’s plus space condition. It has a canonical basis {f d } d≥0 with d ≡ 0, 3<br />
(mod 4) and Fourier expansions<br />
f d (z) = q −d + ∑ n≥1<br />
c(f d , m)q m .<br />
Let f(z) ∈ M<br />
1/2 ! be a weakly holomorphic form with integral Fourier coefficients c(f, n). In<br />
[4], Borcherds constructed an infinite product Ψ f (z) using c(f, n) as exp<strong>one</strong>nts, and showed<br />
that it is a <strong>modular</strong> form <strong>of</strong> <strong>weight</strong> c(f, 0) and some character. The divisors <strong>of</strong> Ψ f (z) are<br />
supported on cusps and imaginary quadratic irrationals. In particular, if τ is a quadratic<br />
irrational <strong>of</strong> discriminant −D < 0, then its multiplicity in Ψ f (z) is<br />
ord τ (Ψ f ) = ∑ k>0<br />
c(f, −Dk 2 ).<br />
For example, when f(z) = f d (z) with d > 0, the Borcherds product Ψ fd (z) equals to<br />
∏<br />
(99)<br />
(j(z) − j(τ q )) 1/wq ,<br />
q∈Q −d /Γ<br />
where Γ = PSL 2 (Z). Note that when −d is fundamental, w d , the number <strong>of</strong> roots <strong>of</strong> unity in<br />
Q( √ −d), is equal to 2w q for all q ∈ Q −d .<br />
Given f(z) ∈ M ! 1/2 , define a <strong>modular</strong> form f lift,θ (z) ∈ M ! 1(p) by<br />
(100) f lift,θ (z) := U 4 (f(pz)θ(z)),<br />
where U 4 is the standard U-operator. It is easy to verify that f lift,θ (z) ∈ M !,+<br />
1 (p) from its<br />
Fourier expansion.<br />
For ψ a non-trivial character <strong>of</strong> Cl(F ), let<br />
˜g ψ (z) = ∑ n∈Z<br />
r + ψ (n)qn ∈ M − 1 (p)
46 W. DUKE AND Y. LI<br />
be a mock-<strong>modular</strong> form with shadow g ψ (z). By Proposition 3.4, the regularized inner<br />
product 〈f lift,θ , g ψ 〉 reg can be expressed in terms <strong>of</strong> r + ψ<br />
(n) as<br />
(101) 〈f lift,θ , g ψ 〉 reg = ∑ ∑<br />
( )<br />
−pm − k<br />
r + 2<br />
ψ<br />
c(f, m)δ(k).<br />
4<br />
m∈Z<br />
k∈Z<br />
The following theorem relates this inner product, hence the linear combination <strong>of</strong> r + ψ<br />
(n), to<br />
the values <strong>of</strong> the Borcherds lift Ψ f (z).<br />
Theorem 9.1. Let f(z) ∈ M<br />
1/2 ! be a weakly holomorphic <strong>modular</strong> form with integral Fourier<br />
coefficients c(f, n), and Ψ f (z) its Borcherds lift. Suppose ψ is a non-trivial character <strong>of</strong><br />
Cl(F ). Then we have<br />
∑<br />
(102) 〈f lift,θ , g ψ 〉 reg = −2 lim ψ 2 (A) ( log |Ψ f (τ A + ɛ)| 2 + C f log |y A | ) ,<br />
ɛ→0<br />
A∈Cl(F )<br />
where C f = ∑ k∈Z c(f, −pk2 ) is the constant term <strong>of</strong> f lift,θ (z), and τ A is a CM point associated<br />
to the class A.<br />
Remark. Recall that jm<br />
lift,B (z) = j m (pz)ϑ B (z) for B ∈ Cl(F ). When B = A 0 is the principal<br />
class in Cl(F ), we can write<br />
2j lift,A 0<br />
m (z) = j m (pz)U 4 (θ(pz)θ(z))<br />
= (j m (4z)θ(z)) lift,θ<br />
= ∑ t∈Z<br />
f lift,θ<br />
4m−t<br />
(z). 2<br />
t 2 ≤4m<br />
Suppose r A0 (m) = 0, then m is not a perfect square and the right hand side <strong>of</strong> equation (53)<br />
becomes<br />
−2<br />
∑<br />
ψ 2 (A) log |Ψ m (τ A , τ A )| .<br />
A∈Cl(F )<br />
By Kr<strong>one</strong>cker’s identity, this is the same as<br />
⎛<br />
−2<br />
∑<br />
A∈Cl(F )<br />
∑<br />
ψ 2 (A) ⎜<br />
⎝ log |Ψ f4m−t 2<br />
(τ A )| ⎟<br />
⎠ .<br />
t∈Z<br />
t 2 ≤4m<br />
Thus, this case <strong>of</strong> Theorem 5.1 is a consequence <strong>of</strong> Theorem 9.1.<br />
The plan <strong>of</strong> the pro<strong>of</strong> goes as follows. First, notice that both sides <strong>of</strong> equation (102) are<br />
additive with respect to f(z). So it suffices to prove the theorem when f(z) = f d (z) for all<br />
d ≥ 0 and d ≡ 3 or 0 modulo 4. For convenience, we denote<br />
(103) a(d, ψ) := 〈f lift,θ<br />
d<br />
, g ψ 〉 reg .<br />
Then, we define a function Φ(z) analogous to G D (z) in [29], where the inner product between<br />
G D (z) and a <strong>weight</strong> k+1/2 level 4 eigenform g(z) gives special values <strong>of</strong> L-function associated<br />
to the Shimura lift <strong>of</strong> g(z). For us, Φ(z) is non-holomorphic, trans<strong>forms</strong> with <strong>weight</strong> 3/2 and<br />
level 4, and satisfies Kohnen’s plus space condition. Since there is no holomorphic cusp form<br />
<strong>of</strong> <strong>weight</strong> 3/2, level 4 satisfying the plus space condition, the inner product between the<br />
holomorphic projection <strong>of</strong> (almost) Φ(z) and the Poincaré series with an s-variable giving<br />
rise to such cusp <strong>forms</strong> is zero as s approaches zero. This gives us the identity relating finite<br />
⎞
MOCK-MODULAR FORMS OF WEIGHT ONE 47<br />
linear combinations <strong>of</strong> r + ψ<br />
(n) to the limit <strong>of</strong> an infinite sum. The finite linear combination<br />
will be exactly the right hand side <strong>of</strong> equation (101). Then, we will relate the limit <strong>of</strong> the<br />
infinite sum to Green’s function, hence Ψ fd (z), via the counting argument in §8. In some<br />
sense, the pro<strong>of</strong> is in the same spirit as Zagier’s pro<strong>of</strong> <strong>of</strong> Borcherds theorem in [44].<br />
The following lemma is analogous to Lemma 6.6 and relates a(d, ψ) with the limit <strong>of</strong> an<br />
infinite sum involving r ψ (n).<br />
Lemma 9.2. Let d ≥ 1 be an integer congruent to 0 or 3 modulo 4. Then we have<br />
(<br />
a(0, ψ)H(d) ∑<br />
( ) ( ) )<br />
(104) a(d, ψ) − = lim δ(k)ϱ s−1 −1 + k2 −pd + k<br />
2<br />
r ψ ,<br />
H(0) s→1 pd 4<br />
k∈Z<br />
where the function ϱ s (µ) is defined by<br />
ϱ s (µ) :=<br />
∫ ∞<br />
1<br />
du<br />
(µu + 1) 1+s<br />
2 u , µ > 0.<br />
Pro<strong>of</strong>. Let ˜g ψ (z) ∈ M − 1 (p) be a mock-<strong>modular</strong> form with shadow g ψ (z) and<br />
ord ∞˜g ψ (z) ≥ − p 4 .<br />
Forms satisfying these conditions exist by Proposition 4.2. The restriction on the order <strong>of</strong><br />
˜g ψ (z) at infinity simplifies the pro<strong>of</strong> and is not important. But the requirement that ˜g ψ (z)<br />
be in the minus space is crucial for the validity <strong>of</strong> the statement.<br />
Denote ĝ ψ (z) ∈ H − 1 (p) the associated harmonic Maass form and define the function Φ(z)<br />
by<br />
Φ(z) = Tr 4p<br />
4 ((ĝ ψ |U p )(4z)θ(pz)),<br />
where Tr 4p<br />
4 is the trace down operator from level 4p to level 4. The function Φ(z) trans<strong>forms</strong><br />
like a <strong>weight</strong> 3/2 <strong>modular</strong> form <strong>of</strong> level 4 and should be compared to G D (z) in [29], which<br />
was defined by applying the trace down operator Tr 4D<br />
4 to the product <strong>of</strong> the holomorphic<br />
<strong>weight</strong> k Eisenstein series and θ(|D|z). If <strong>one</strong> replaces the <strong>weight</strong> k holomorphic Eisenstein<br />
series by the <strong>weight</strong> <strong>one</strong> real-analytic Eisenstein series with an s variable, and consider the<br />
derivative with respect to s, then it is something quite similar to this Φ(z).<br />
Using the coset representatives<br />
Γ 0 (4p)\Γ 0 (4) =<br />
p−1<br />
⊔<br />
µ=0<br />
(<br />
1<br />
4 1<br />
) (<br />
1 µ<br />
1<br />
) ⊔<br />
(<br />
1<br />
1)<br />
,<br />
and the fact that (ĝ ψ |U p )(z) = −i(ĝ ψ |W p )(z), We can calculate the Fourier expansion <strong>of</strong><br />
Φ(z) = Tr 4p<br />
4 ((ĝ ψ |U p )(4z)θ(pz)) as<br />
Φ(z) = ∑ n∈Z(b(n, y) + a(n))q n ,<br />
b(n, y) = − ∑<br />
a(n) = ∑ k∈Z<br />
k 2 −4m=pn<br />
δ(k)r + ψ<br />
δ(k)r ψ (m)β 1<br />
(m, 4y p<br />
( pn − k<br />
2<br />
Since p ≡ 3 (mod 4), Φ(z) satisfies Kohnen’s plus space condition and n ≡ 0 or 3 (mod 4).<br />
Since ord ∞˜g ψ (z) ≥ −p/4, the right hand side <strong>of</strong> equation (101) equals to a(d) for f(z) = f d (z).<br />
4<br />
)<br />
.<br />
)<br />
,
48 W. DUKE AND Y. LI<br />
So we have<br />
(105) a(d) = a(d, ψ).<br />
Let F(z) be the <strong>weight</strong> 3/2 Eisenstein series studied in [28], which has the following Fourier<br />
expansion<br />
∞∑<br />
∑ ∞<br />
F(z) = H(n)q n + y −1/2 1<br />
16π β 3/2(m 2 , y)q −m2 ,<br />
n=0<br />
m=−∞<br />
and satisfies Kohnen’s plus space condition. For all n ∈ Z, notice that b(n, y)q n is exp<strong>one</strong>ntially<br />
decaying as y → ∞. Also, a(n) vanishes for all n < 0. Thus, the function<br />
Φ ∗ (z) := Φ(z) − a(0)F(z)<br />
H(0)<br />
is O(y −1/2 ) as y → ∞. The same decaying property holds at the other two cusps <strong>of</strong> Γ 0 (4) as<br />
well, since Φ ∗ (z) satisfies Kohnen’s plus space condition. So we can consider the holomorphic<br />
projection <strong>of</strong> Φ ∗ (z) to the Kohnen plus space S + 3/2 (Γ 0(4)). Define the <strong>weight</strong> 3/2 Poincaré<br />
series by<br />
∑<br />
(106)<br />
P d (z, s) :=<br />
j(γ, z) −3 e 2dπiγz Im(γz) s/2 ,<br />
where for γ ∈ Γ 0 (4)<br />
γ∈Γ ∞\Γ 0 (4)<br />
j(γ, z) := θ(γz)<br />
θ(z) .<br />
This series converges absolutely for Re(s) > 1 and can be analytically continued to Re(s) ≥<br />
0. As s → 0, the inner product 〈Φ ∗ (z), P d (z, s)〉 is the d th coefficient <strong>of</strong> a cusp form in<br />
S + 3/2 (Γ 0(4)), since Φ ∗ (z) is already in the plus space. Given S + 3/2 (Γ 0(4)) = {0}, we know the<br />
limit is zero and obtain the following equation after applying Rankin-Selberg unfolding,<br />
( Γ(<br />
1+s<br />
(107) lim<br />
) (<br />
2<br />
a(d) − a(0)H(d) ) ∫ ∞<br />
)<br />
+ b(d, y)e −4πdy 1/2+s/2 dy<br />
y = 0.<br />
s→0 (4πd) 1/2+s/2 H(0)<br />
y<br />
After some manipulations, we have<br />
∫ ∞<br />
After substituting µ = k2<br />
pd<br />
(108) −<br />
∫ ∞<br />
Here, we used the fact<br />
0<br />
0<br />
β 1 (d, µy)e −4πdy 1/2+s/2 dy<br />
y<br />
y<br />
− 1 and µ =<br />
pk2<br />
d<br />
b(d, y)e −4πdy 1/2+s/2 dy<br />
y<br />
y<br />
0<br />
=<br />
Γ(<br />
1+s<br />
2 )<br />
(4πd) 1/2+s/2 ϱ s (µ) .<br />
− 1, we arrive at the following expression<br />
∑<br />
( ) −pd + k<br />
2<br />
δ(k)r ψ ϱ s<br />
(−1 + k2<br />
4<br />
pd<br />
1+s<br />
Γ(<br />
2<br />
= )<br />
(4πd) 1+s<br />
2<br />
k∈Z<br />
β 1 (d, αy) = β 1 (dα, y)<br />
for all α, y, d ∈ R >0 . Now substituting (108) and (105) into (107) gives us equation (104)<br />
Pro<strong>of</strong> <strong>of</strong> Theorem 9.1. When d = 0, f d (z) = θ(z) is the Jacobi theta function and θ lift,θ (z) =<br />
2ϑ A0 (z) is twice the <strong>weight</strong> <strong>one</strong> theta series associated to the principal class A 0 . The<br />
Borcherds lift <strong>of</strong> θ(z) is η 2 (z) and equation (102) becomes<br />
〈2ϑ A0 , g ψ 〉 = −4 ∑<br />
ψ 2 (A) log ∣ √ y A η 2 (τ A ) ∣ A∈Cl(F )<br />
)<br />
.<br />
□
MOCK-MODULAR FORMS OF WEIGHT ONE 49<br />
This is justified by Proposition 4.1 and the facts that<br />
1 ∑<br />
〈ϑ A0 , g ψ 〉 =<br />
〈g ψ ′, g ψ 〉,<br />
H(p)<br />
ψ ′ :Cl(F )→C × ∣ √ y A η 2 (τ A ) ∣ = ∣ √ y A η 2 (τ A ) ∣ .<br />
When d > 0 is congruent to 0 or 3 modulo 4, we need to study the right hand side<br />
<strong>of</strong> equation (104), which will yield the Green’s function. The procedure here follows part<br />
<strong>of</strong> §5 in [23] and is analogous to the pro<strong>of</strong> <strong>of</strong> Theorem 5.1. Recall that Q s−1 (t) is the<br />
Legendre function <strong>of</strong> the second kind defined in (55). By comparing it with ϱ s (µ), we see<br />
that ϱ 0 (µ) = 2Q 0 ( √ µ + 1) and<br />
Q s−1 ( √ µ + 1) −<br />
( )<br />
Substituting 22−s Γ(2s)<br />
Q |k|<br />
sΓ(s) 2 s−1<br />
√ pd<br />
equation (104) yields<br />
a(d, ψ) −<br />
a(0, ψ)H(d)<br />
H(0)<br />
⎛<br />
= lim ⎝2<br />
s→1<br />
sΓ(s)2<br />
2 2−s Γ(2s) ϱ s−1(µ) = O(µ −1/2−s/2 ).<br />
( )<br />
for ϱ s−1 −1 + k2 in the limit on the right hand side <strong>of</strong><br />
∑<br />
A∈Cl(F )<br />
pd<br />
ψ(A) ∑<br />
k> √ pd<br />
δ(k)r A<br />
( −pd + k<br />
2<br />
4<br />
)<br />
2Q s−1<br />
( k<br />
√ pd<br />
) ⎞ ⎠ .<br />
The factor <strong>of</strong> 2 next to the summation comes from the contribution k < − √ pd.<br />
Let Q A ∈ Q −p be the quadratic form such that Q A (τ A , 1) = 0. By Proposition 8.1 we have<br />
( ) ( )<br />
ρ QA (k, d) = δ(k)r Q 2<br />
A<br />
= δ(k)r A 2 ,<br />
k 2 −pd<br />
4<br />
k 2 −pd<br />
4<br />
where ρ QA (k, d) is defined in (91) and counts the number <strong>of</strong> quadratic <strong>forms</strong> Q ′ ∈ Q −d such<br />
that cosh d(τ QA , τ Q ′) = √ k<br />
pd<br />
and is independent <strong>of</strong> the choice <strong>of</strong> τ A . So equation (104) becomes<br />
a(d, ψ) −<br />
a(0, ψ)H(d)<br />
H(0)<br />
= (−2) lim<br />
= (−2) lim<br />
∑<br />
s→1<br />
A∈Cl(F )<br />
∑<br />
s→1<br />
A∈Cl(F )<br />
= (−2) lim ⎜<br />
⎝ lim<br />
ɛ→0<br />
⎛<br />
⎛<br />
ψ(A 2 ) ⎝ ∑<br />
ψ 2 (A) ∑<br />
∑<br />
s→1<br />
A∈Cl(F )<br />
k> √ pd<br />
⎞<br />
( )<br />
k<br />
−2Q s−1<br />
√ pd<br />
ρ QA (k, d) ⎠<br />
g s (τ A , τ Q )<br />
Q∈Q −d<br />
τ Q ≠τ A<br />
ψ 2 (A) ∑<br />
⎞<br />
g s (τ A + ɛ, τ Q ) ⎟<br />
⎠<br />
Q∈Q −d<br />
τ Q ≠τ A<br />
where g s (τ 1 , τ 2 ) is defined in (57).<br />
As long as d is not <strong>of</strong> the form pm 2 for any integer m, the condition τ Q ≠ τ A is satisfied<br />
by all Q ∈ Q −d . Otherwise, there exists exactly <strong>one</strong> Q ∈ Q −d such that τ Q = τ A . Since<br />
c(f d , 0) = 0, the number <strong>of</strong> such Q ∈ Q −d can also be written as<br />
1<br />
C 2 f d<br />
= ∑ c(f d , −pm 2 ) = 1.<br />
m>0
50 W. DUKE AND Y. LI<br />
To treat both cases simultaneously, we can write<br />
⎛<br />
⎞<br />
∑<br />
g s (τ A + ɛ, τ Q ) = ⎝ ∑<br />
g s (τ A + ɛ, τ Q ) ⎠ − 1C 2 f d<br />
g s (τ A + ɛ, τ A )<br />
Q∈Q −d<br />
Q∈Q −d<br />
disc(τ Q )≠−p<br />
=<br />
⎛<br />
⎝<br />
∑<br />
Q∈Q −d /Γ<br />
⎞<br />
G s (τ A + ɛ, τ Q ) ⎠ − 1C 2 f d<br />
g s (τ A + ɛ, τ A ),<br />
where G s (τ 1 , τ 2 ) is the automorphic Green’s function defined in (56). After substituting this<br />
into equation (109), it becomes<br />
⎛<br />
⎞<br />
a(0, ψ)H(d)<br />
∑ ∑<br />
a(d, ψ) − =(−2) lim ⎝lim ψ 2 (A) G s (τ A + ɛ, τ Q ) ⎠ +<br />
H(0)<br />
ɛ→0 s→1<br />
A∈Cl(F ) Q∈Q −d /Γ<br />
(109)<br />
⎛<br />
⎞<br />
C fd lim ⎝ ∑<br />
ψ 2 (A)g 1 (τ A + ɛ, τ A ) ⎠ .<br />
ɛ→0<br />
A∈Cl(F )<br />
The first summand above can be evaluated using the identity (58). Substituting τ 1 = τ A +ɛ,<br />
τ 2 = τ Q into equation (58) and summing them together over A ∈ Cl(F ) and Q ∈ Q −d /Γ with<br />
coefficient ψ 2 (A) gives us<br />
⎛<br />
⎞<br />
∑<br />
ψ 2 (A) log |Ψ fd (τ A + ɛ)| 2 = lim ⎝ ∑ ∑<br />
ψ 2 (A) G s (τ A + ɛ, τ Q ) + 4πE(τ A + ɛ, s) ⎠ .<br />
s→1<br />
A<br />
A<br />
By (49), we have<br />
⎛<br />
lim ⎝ ∑<br />
s→1<br />
A<br />
ψ 2 (A)<br />
So the first summand is<br />
⎛<br />
−2 ⎝lim<br />
∑<br />
Q∈Q −d /Γ<br />
∑<br />
ɛ→0<br />
A∈Cl(F )<br />
⎞<br />
Q∈Q −d /Γ<br />
4πE(τ A , s) ⎠ = −24H(d) ∑ A<br />
a(0, ψ)H(d)<br />
= −<br />
2H(0)<br />
ψ 2 (A) log |Ψ fd (τ A + ɛ)| 2 ⎞<br />
(<br />
1 + 4y2<br />
⎠ −<br />
ψ 2 (A) log ∣ ∣ √ y A η 2 (τ A ) ∣ ∣<br />
a(0, ψ)H(d)<br />
.<br />
H(0)<br />
)<br />
Substitute g 1 (τ + ɛ, τ) = − log with τ = τ<br />
ɛ 2 A into the second summand, we obtain<br />
⎛<br />
⎞<br />
a(0, ψ)H(d)<br />
∑<br />
a(d, ψ) − = − 2 ⎝lim ψ 2 (A) log |Ψ fd (τ A + ɛ)| 2 ⎠<br />
a(0, ψ)H(d)<br />
−<br />
H(0)<br />
ɛ→0<br />
H(0)<br />
− 2C fd<br />
A∈Cl(F )<br />
∑<br />
A∈Cl(F )<br />
ψ 2 (A) log |y A |.<br />
After canceling the term − a(0,ψ)H(d)<br />
H(0)<br />
, equation (109) becomes equation (102) for f = f d , d > 0.<br />
□
MOCK-MODULAR FORMS OF WEIGHT ONE 51<br />
We can now easily deduce Theorem 1.2. It is convenient to give a slightly more general<br />
version stated in terms <strong>of</strong> theta series and quadratic <strong>forms</strong>.<br />
Corollary 9.3. Let p be a prime number congruent to 3 modulo 4, and −d a discriminant<br />
not <strong>of</strong> the form −pm 2 for any integer m. For a class B ∈ Cl(F ), suppose ˜ϑ B 2(z) =<br />
∑<br />
n∈Z r+ B<br />
(n)q n ∈ M − 2 1 (p) has shadow ϑ B 2(z) and ord ∞ ( ˜ϑ B 2) > −p/4. Then<br />
∣<br />
(110) log<br />
∣<br />
∏<br />
q∈Q −d /Γ<br />
(j(τ B ) − j(τ q )) 1/wq ∣ ∣∣∣∣∣<br />
= − 1 4<br />
∑<br />
Remark. When −d is fundamental, w d = 2w q for all q ∈ Q −d .<br />
k∈Z<br />
δ(k) r + B 2 ( pd−k2<br />
4<br />
).<br />
Pro<strong>of</strong>. Let f(z) = f d (z) with d not <strong>of</strong> the form pm 2 . Using the Borcherds lift <strong>of</strong> f d (z) in (99),<br />
the right hand side <strong>of</strong> (102) can be rewritten as<br />
∣<br />
−4<br />
∑<br />
∏<br />
∣∣∣∣∣<br />
ψ 2 (A) log<br />
(j(τ A ) − j(τ q )) 1/wq .<br />
∣<br />
A∈Cl(F )<br />
q∈Q −d /Γ<br />
For each non-trivial character ψ, choose ˜g ψ (z) ∈ M − 1 (p) having shadow g ψ (z) and r + ψ (n) = 0<br />
for all n ≤ −p/4. For these ˜g ψ (z), equation (101) becomes<br />
〈f lift,θ<br />
d<br />
, g ψ 〉 reg = ∑ ( ) pd − k<br />
r + 2<br />
ψ<br />
δ(k).<br />
4<br />
k∈Z<br />
Combining these two together and dividing both sides by -4, we have<br />
∣<br />
∑<br />
( ) pd − k<br />
(111) − 1 r + 2<br />
4 ψ<br />
δ(k) =<br />
∑<br />
∣∣∣∣∣ ∏<br />
ψ 2 (A) log<br />
4<br />
(j(A) − j(τ q )) 1/wq ,<br />
∣<br />
k∈Z<br />
A∈Cl(F )<br />
which is precisely equation (15) when −d is fundamental.<br />
Consider the mock-<strong>modular</strong> form ˜F (z) ∈ M − 1 (p) defined by<br />
⎛<br />
˜F (z) := 1<br />
H(p)<br />
⎜<br />
⎝ E+ 1 (z) +<br />
∑<br />
ψ:Cl(F )→C ×<br />
ψ non-trivial<br />
q∈Q −d /Γ<br />
⎞<br />
ψ(B −2 )˜g ψ (z) ⎟<br />
⎠ − ˜ϑ B 2(z).<br />
∑<br />
Since ϑ B 2(z) = 1<br />
H(p) ψ ψ(B−2 )g ψ (z), ˜F (z) has shadow F (z) = 0, hence is <strong>modular</strong>. Also, it<br />
has Fourier expansion in the form<br />
˜F (z) =<br />
∑<br />
c + (n)q n .<br />
From the pro<strong>of</strong> <strong>of</strong> Lemma 9.2, we know that<br />
∑ (<br />
c +<br />
k∈Z<br />
pd−k 2<br />
4<br />
n>−p/4<br />
χ p(n)≠1<br />
)<br />
δ(k) = 〈f lift,θ<br />
d<br />
, F 〉 reg = 0.
52 W. DUKE AND Y. LI<br />
for any d. By the same argument in the pro<strong>of</strong> <strong>of</strong> Theorem 9.1, <strong>one</strong> can show that equation<br />
(12) holds when d ≠ pm 2 for any integer m. So in that case, the equation above becomes<br />
∣<br />
− 1 ∑<br />
( ) pd − k<br />
r + 2<br />
1 ∑<br />
B<br />
δ(k) = ψ(B −2 ) ∑ ∣∣∣∣∣<br />
∏<br />
ψ 2 (A) log<br />
4<br />
2 4<br />
H(p)<br />
− j(τ q ))<br />
k∈Z<br />
ψ<br />
A ∣q∈Q −d /Γ(j(A) 1/wq ∣ ∣∣∣∣∣ ∏<br />
= log<br />
(j(B) − j(τ q )) 1/wq .<br />
∣<br />
q∈Q −d /Γ<br />
10. Case p = 283: Some numerical calculations<br />
Finally, we will present another numerical example to demonstrate that statements similar<br />
to the Conjecture in §1 should be true for octahedral new<strong>forms</strong>. These calculations were<br />
conducted in SAGE [41]. When p = 283, Cl(F ) has order 3 and the space S 1 (p) is 3-<br />
dimensional spanned by dihedral newform h(z) and octahedral new<strong>forms</strong> f ± (z) = f 1 (z) ±<br />
√ −2f2 (z), where<br />
h(z) = q + q 4 − q 7 + q 9 + O(q 10 ),<br />
f 1 (z) = q − q 4 + 2q 6 − q 7 − q 9 + O(q 10 ),<br />
f 2 (z) = q 2 − q 3 − q 5 + O(q 10 ).<br />
By Lemma 3.3, h(z), f 1 (z) ∈ S + 1 (p) and f 2 (z) ∈ S − 1 (p). By Deligne-Serre Theorem, f ± (z)<br />
arises from ρ : Gal(K/Q) → GL 2 (C), where K is the degree 2 extension <strong>of</strong> the normal closure<br />
<strong>of</strong> Q[X]/(X 4 − X − 1) such that Gal(K/Q) ∼ = GL 2 (F 3 ).<br />
Let F 6 be the subfield <strong>of</strong> K fixed by a subgroup <strong>of</strong> Gal(K/Q) isomorphic to Z/8Z, and H 3<br />
the cubic subfield <strong>of</strong> F 6 . Explicitly, we can write<br />
H 3 = Q[Y ]/(Y 3 + 4Y + 1),<br />
F 6 = Q[X]/(X 6 − 3X 5 + 6X 4 − 7X 3 + 10X 2 − 7X + 6),<br />
□<br />
with the embedding<br />
H 3 ↩→ F 6<br />
Y ↦→ X 2 − X + 1.<br />
Fix a complex embedding <strong>of</strong> F 6 ↩→ C such that X = t, Y = θ with t, θ ∈ C having positive<br />
imaginary part. Since F 6 is totally complex, the unit group <strong>of</strong> F 6 has rank 2 and is generated<br />
by<br />
u 1 = t 2 − t + 1, u 2 = t 3 + t − 1.<br />
As in the case <strong>of</strong> dihedral new<strong>forms</strong>, <strong>one</strong> can express the Petersson norms <strong>of</strong> f + (z) and f − (z)<br />
as<br />
〈f + , f + 〉 = 〈f − , f − 〉 = p + 1<br />
12<br />
Res s=1<br />
ζ(s)L(s, F 6 )<br />
L(s, H 3 )ζ(2s)(1 + p −s ) .<br />
Notice the ratio L(s, F 6 )/L(s, H 3 ) is the L-function <strong>of</strong> the quadratic Hecke character <strong>of</strong> H 3<br />
associated to the Hilbert quadratic norm residue symbol <strong>of</strong> F 6 /H 3 . So it is holomorphic at<br />
s = 1 and the right hand side can be evaluated to be 8 log |u 1 u 2 2|.
MOCK-MODULAR FORMS OF WEIGHT ONE 53<br />
By Proposition 3.5, there exists mock <strong>modular</strong> <strong>forms</strong> ˜f 1 (z) ∈ M − 1 (p) and ˜f 2 (z) ∈ M + 1 (p)<br />
with shadows f 1 , f 2 respectively such that<br />
∑<br />
˜f 1 (z) = c + 1 (−4)q −4 + c + 1 (−1)q −1 + c + 1 (0) +<br />
c + 1 (n)q n ,<br />
˜f 2 (z) = c + 2 (−2)q −2 + c + 2 (0) +<br />
Using Proposition 3.4, we find that<br />
∑<br />
n>0,χ 283 (n)≠−1<br />
n>0,χ 283 (n)≠1<br />
c + 2 (n)q n .<br />
−c + 1 (−4) = c + 1 (−1) = c + 2 (−2) = 2 log |u 1 u 2 2|.<br />
Since M 1 + (p) ⊂ M + 1 (p) and M1 − (p) ⊂ M − 1 (p) are both non-empty, the principal part <strong>of</strong> ˜f j (z)<br />
does not determine it uniquely. However, once c + 1 (2) is chosen, then ˜f 1 is fixed. Similarly,<br />
once c + 2 (0), c + 2 (1) and c + 2 (4) are chosen, ˜f2 is fixed. Since we expect c + j (n) to be logarithms<br />
<strong>of</strong> absolute values <strong>of</strong> algebraic numbers in F 6 , it is natural to choose c + 1 (2), c + 2 (0), c + 2 (1) and<br />
c + 2 (4) this way and study the other coefficients. So we will write<br />
c + j (n) = β j log |u j (n)|,<br />
where β j ∈ Q and u j (n) ∈ C × is some complex number. Then we fix ˜f j (z) by letting<br />
β 1 = 2, β 2 = u 2 (0) = u 2 (1) = 1 and<br />
u 2 (4) = u 2 1(2) = − 7 16 t5 + 5 4 t4 − 7 8 t3 + 47<br />
16 t2 − 13<br />
16 t + 19<br />
8 .<br />
From the numerical calculations, we can make predictions <strong>of</strong> the algebraicity <strong>of</strong> u j (n). In the<br />
table below, we list c + j (l) for j = 1, 2 and various primes l. Also, we list the predicted values<br />
˜β j and fractional ideals generated by ũ j (n) in F 6 .<br />
When χ p (l) ≠ 1, the ideal (l) splits into L 0 L 1 in H 3 , and L 1 splits into l l,1 l l,2 in F 6 .<br />
When χ p (l) = 1 and c(f 1 , l) = ±1, the ideal (l) splits into l 1 l 2 in F 6 . When χ p (l) = 1 and<br />
c(f 1 , l) = 0, the ideal (l) splits into L l,0 L l,1 in H 3 and L l,1 splits into l l,1 l l,2 in F 6 , where<br />
l l,j has order 4 in Cl(F 6 ), the class group <strong>of</strong> F 6 . Since F 6 and H 3 have class numbers 8 and<br />
2 respectively, the fractional ideals in Table 3 and Table 4 are all principal. For example,<br />
the value we chose for u 2 1(2) generates the fractional ideal (l 2,1 /l 2,2 ) 4 , where l 2,1 and l 2,2 have<br />
order 8 in Cl(F 6 ). So it is necessary to make ˜β 1 = 1/2. The numerical pattern also justifies<br />
this choice <strong>of</strong> ˜f j (z).<br />
Table 3. Coefficients <strong>of</strong> ˜f 1 (z)<br />
l c(f 2 , l) c + 1 (l) ˜β1 (ũ 1 (l))<br />
2 1 1.2075349695016218 1/2 (l 2,1 /l 2,2 ) 4<br />
3 -1 -0.44226603950742649 1/2 (l 3,1 /l 3,2 ) 4<br />
5 -1 -3.9855512247433431 1/2 (l 5,1 /l 5,2 ) 4<br />
17 0 -3.2181607607379323 1/2 (l 17,1 /l 17,2 ) 4<br />
19 1 5.3481233955176073 1/2 (l 19,1 /l 19,2 ) 4<br />
31 1 -1.7192005338244623 1/2 (l 31,1 /l 31,2 ) 4<br />
37 0 0.32541651822318252 1/2 (l 37,1 /l 37,2 ) 4
54 W. DUKE AND Y. LI<br />
43 1 -4.6200896216743352 1/2 (l 43,1 /l 43,2 ) 4<br />
47 -1 -1.0203031328088645 1/2 (l 47,1 /l 47,2 ) 4<br />
53 0 5.8419201851710110 1/2 (l 53,1 /l 53,2 ) 4<br />
67 0 -3.9318486618330462 1/2 (l 67,1 /l 67,2 ) 4<br />
79 0 7.5720154112893967 1/2 (l 79,1 /l 79,2 ) 4<br />
107 0 -0.81774052769784944 1/2 (l 107,1 /l 107,2 ) 4<br />
109 -1 4.8808523053451562 1/2 (l 109,1 /l 109,2 ) 4<br />
283 0 6.86483405137284 1/2 (l 283,1 /l 283,2 ) 4<br />
Table 4. Coefficients <strong>of</strong> ˜f 2 (z)<br />
l c(f 1 , l) c + 2 (l) ˜β2 (ũ 2 (l))<br />
7 -1 -3.27983974462451 1 l 7,1 /l 7,2<br />
11 1 -2.56257986300244 1 l 11,1 /l 11,2<br />
13 1 -5.57196302179201 1 l 13,1 /l 13,2<br />
23 -1 1.01652189648251 1 l 23,1 /l 23,2<br />
29 -1 1.54494007675715 1 l 29,1 /l 29,2<br />
41 1 0.771808245755645 1 l 41,1 /l 41,2<br />
71 0 -4.99942007705695 1 (l 71,1 /l 71,2 ) 2<br />
73 0 -1.64986308549260 1 (l 73,1 /l 73,2 ) 2<br />
83 -2 8.97062724307569 1 1<br />
89 -1 -0.399183274865547 1 l 89,1 /l 89,2<br />
101 0 5.99108448704487 1 (l 101,1 /l 101,2 ) 2<br />
283 1 4.48531362153791 1 1<br />
643 2 2.32782185606303e-10 1 1<br />
773 2 5.08403073372794e-9 1 1<br />
859 -2 -8.97062719191082 1 1<br />
References<br />
1. Atkin,A.O.L. Modular Forms <strong>of</strong> Weight 1 and Supersingular Reduction U.S.-Japan Seminar on Modular<br />
Functions, Ann Arbor, 1975<br />
2. Buchholz, Herbert The confluent hypergeometric function with special emphasis on its applications. Translated<br />
from the German by H. Lichtblau and K. Wetzel. Springer Tracts in Natural Philosophy, Vol. 15<br />
Springer-Verlag New York Inc., New York 1969 xviii+238 pp<br />
3. Buell, Duncan A. Binary quadratic <strong>forms</strong>. Classical theory and modern computations. Springer-Verlag,<br />
New York, 1989. x+247 pp.<br />
4. Borcherds, Richard E. Automorphic <strong>forms</strong> on O s+2,2 (R) and infinite products. Invent. Math. 120 (1995),<br />
no. 1, 161-213.<br />
5. Borcherds, Richard E. The Gross-Kohnen-Zagier theorem in higher dimensions. Duke Math. J. 97 (1999),<br />
no. 2, 219-233.<br />
6. Bringmann, Kathrin; Ono, Ken, Lifting cusp <strong>forms</strong> to Maass <strong>forms</strong> with an application to partitions.<br />
Proc. Natl. Acad. Sci. USA 104 (2007), no. 10, 3725-3731<br />
7. Bruinier, Jan Hendrik Borcherds products on O(2, l) and Chern classes <strong>of</strong> Heegner divisors. Lecture<br />
Notes in <strong>Mathematics</strong>, 1780. Springer-Verlag, Berlin, 2002. viii+152 pp.<br />
8. Bruinier, Jan Hendrik; Bundschuh, Michael On Borcherds products associated with lattices <strong>of</strong> prime<br />
discriminant. Rankin memorial issues. Ramanujan J. 7 (2003), no. 1-3, 49-61.<br />
9. Bruinier, Jan Hendrik; Funke, Jens On two geometric theta lifts. Duke Math. J. 125 (2004), no. 1, 45-90.<br />
10. Bruinier, Jan Hendrik; Ono, Ken, Heegner Divisors, L-Functions and Harmonic Weak Maass Forms,<br />
Annals <strong>of</strong> Math. 172 (2010), 2135–2181<br />
11. Bruinier, Jan H.; Ono, Ken; Rhoades, Robert C. Differential Operators for harmonic weak Maass <strong>forms</strong><br />
and the vanishing <strong>of</strong> Hecke eigenvalues. Math. Ann. 342 (2008), no. 3, 673693.
MOCK-MODULAR FORMS OF WEIGHT ONE 55<br />
12. Bruinier, Jan Hendrik; Yang, Tonghai CM-values <strong>of</strong> Hilbert <strong>modular</strong> functions. Invent. Math. 163 (2006),<br />
no. 2, 229-288.<br />
13. Bruinier, Jan Hendrik Harmonic Maass <strong>forms</strong> and periods, preprint (2011)<br />
14. Cox, David A. Primes <strong>of</strong> the form x 2 + ny 2 . Fermat, class field theory and complex multiplication. A<br />
Wiley-Interscience Publication. John Wiley & Sons, Inc., New York, 1989. xiv+351 pp.<br />
15. Deligne, Pierre; Serre, Jean-Pierre Formes modulaires de poids 1. (French) Ann. Sci. cole Norm. Sup. (4)<br />
7 (1974), 507-530 (1975).<br />
16. L. Dirichlet, Werke II.<br />
17. Duke, W.; Imamoḡlu.Ö. ; Tóth, Á . Cycle integrals <strong>of</strong> the j-function and mock <strong>modular</strong> <strong>forms</strong>. Ann. <strong>of</strong><br />
Math. (2) 173 (2011), no. 2, 947-981.<br />
18. Duke, W.; Imamoḡlu.Ö. ; Tóth, Á . Real quadratic analogs <strong>of</strong> traces <strong>of</strong> singular moduli. Int. Math. Res.<br />
Not. IMRN 2011, no. 13, 3082-3094<br />
19. Ehlen, Stephan On CM values <strong>of</strong> Borcherds products and <strong>weight</strong> <strong>one</strong> harmonic weak Maass <strong>forms</strong>, in<br />
preparation.<br />
20. Fay, John D. Fourier coefficients <strong>of</strong> the resolvent for a Fuchsian group. J. Reine Angew. Math. 293/294<br />
(1977), 143-203.<br />
21. C.F. Gauss, Werke, I.<br />
22. Gross, Benedict H. Heegner points on X 0 (N). Modular <strong>forms</strong> (Durham, 1983), 87-105, Ellis Horwood<br />
Ser. Math. Appl.: Statist. Oper. Res., Horwood, Chichester, 1984.<br />
23. Gross, Benedict H.; Zagier, Don B. On singular moduli. J. Reine Angew. Math. 355 (1985), 191—220.<br />
24. Gross, Benedict H.; Zagier, Don B. Heegner points and derivatives <strong>of</strong> L-series. Invent. Math. 84 (1986),<br />
no. 2, 225-320.<br />
25. Gross, B.; Kohnen, W.; Zagier, D. Heegner points and derivatives <strong>of</strong> L-series. II. Math. Ann. 278 (1987),<br />
no. 1-4, 497-562.<br />
26. Hecke, Erich Mathematische Werke. (German) With introductory material by B. Schoeneberg, C. L.<br />
Siegel and J. Nielsen. Third edition. Vandenhoeck & Ruprecht, Gttingen, 1983. 960 pp.<br />
27. Hejhal, Dennis A. The Selberg trace formula for PSL(2, R). Vol. 2. Lecture Notes in <strong>Mathematics</strong>, 1001.<br />
Springer-Verlag, Berlin, 1983. viii+806 pp.<br />
28. Hirzebruch, F.; Zagier, D. Intersection numbers <strong>of</strong> curves on Hilbert <strong>modular</strong> surfaces and <strong>modular</strong> <strong>forms</strong><br />
<strong>of</strong> Nebentypus. Invent. Math. 36 (1976), 57-113.<br />
29. Kohnen, W.; Zagier, D. Values <strong>of</strong> L-series <strong>of</strong> <strong>modular</strong> <strong>forms</strong> at the center <strong>of</strong> the critical strip. Invent.<br />
Math. 64 (1981), no. 2, 175-198.<br />
30. Kudla, Stephen S.; Rapoport, Michael; Yang, Tonghai On the derivative <strong>of</strong> an Eisenstein series <strong>of</strong> <strong>weight</strong><br />
<strong>one</strong>. Internat. Math. Res. Notices 1999, no. 7, 347-385.<br />
31. Neunhöffer, H. Über die analytische Fortsetzung von Poincaréreihen. (German) S.-B. Heidelberger Akad.<br />
Wiss. Math.-Natur. Kl. 1973, 33-90.<br />
32. Niebur, Douglas A class <strong>of</strong> nonanalytic automorphic functions. Nagoya Math. J. 52 (1973), 133-145.<br />
33. Ogg, A. Survey on Modular functions <strong>of</strong> One Variable Notes by F. van Oystaeyen. Modular functions <strong>of</strong><br />
<strong>one</strong> variable, I (Proc. Internat. Summer School, Univ. Antwerp, Antwerp, 1972), pp. 1-35. Lecture Notes<br />
in Math. Vol. 320, Springer, Berlin, 1973.<br />
34. Ribet, K. Galois Representation Attached to Eigen<strong>forms</strong> with Nebentypus Modular functions <strong>of</strong> <strong>one</strong><br />
variable, V (Proc. Second Internat. Conf., Univ. Bonn, Bonn, 1976), pp. 17-51. Lecture Notes in Math.,<br />
Vol. 601, Springer, Berlin, 1977.<br />
35. Roelcke, Walter Das Eigenwertproblem der automorphen Formen in der hyperbolischen Ebene. I, II.<br />
(German) Math. Ann. 167 (1966), 292-337; ibid. 168 1966 261-324.<br />
36. Serre, J.-P. Modular <strong>forms</strong> <strong>of</strong> <strong>weight</strong> <strong>one</strong> and Galois representations. Algebraic number fields: L-functions<br />
and Galois properties (Proc. Sympos., Univ. Durham, Durham, 1975), pp. 193-268. Academic Press,<br />
London, 1977.<br />
37. Shimura, Goro Introduction to the arithmetic theory <strong>of</strong> automorphic functions. Reprint <strong>of</strong> the 1971 original.<br />
Publications <strong>of</strong> the Mathematical Society <strong>of</strong> Japan, 11. Kan Memorial Lectures, 1. Princeton University<br />
Press, Princeton, NJ, 1994. xiv+271 pp.<br />
38. Siegel, Carl Ludwig Lectures on advanced analytic number theory. Notes by S. Raghavan. Tata Institute<br />
<strong>of</strong> Fundamental Research Lectures on <strong>Mathematics</strong>, No. 23 Tata Institute <strong>of</strong> Fundamental Research,<br />
Bombay 1965 iii+331+iii pp.
56 W. DUKE AND Y. LI<br />
39. Stark, H. M. Class fields and <strong>modular</strong> <strong>forms</strong> <strong>of</strong> <strong>weight</strong> <strong>one</strong>. Modular functions <strong>of</strong> <strong>one</strong> variable, V (Proc.<br />
Second Internat. Conf., Univ. Bonn, Bonn, 1976), pp. 277-287. Lecture Notes in Math., Vol. 601, Springer,<br />
Berlin, 1977.<br />
40. Stark, H. M. L-functions at s = 1.I– IV. Advances in Math. 7 1971 301-343 (1971), 17 (1975), no. 1,<br />
60-92., 22 (1976), no. 1, 64-84, 35 (1980), no. 3, 197-235.<br />
41. Stein, William A. et al. Sage <strong>Mathematics</strong> S<strong>of</strong>tware (Version 5.0.1), The Sage Development Team, 2012,<br />
http://www.sagemath.org.<br />
42. Viazovska, Maryna. CM Values <strong>of</strong> Higher Green’s Functions, preprint 2011 (see arXiv:1110.4654)<br />
43. Zagier, Don. Letter to Gross<br />
44. Zagier, Don Traces <strong>of</strong> singular moduli. Motives, polylogarithms and Hodge theory, Part I (Irvine, CA,<br />
1998), 211-244, Int. Press Lect. Ser., 3, I, Int. Press, Somerville, MA, 2002.<br />
45. Zagier, Don. Nombres de classes et formes modulaires de poids 3/2. (French) C. R. Acad. Sci. Paris Sr.<br />
A-B 281 (1975), no. 21, Ai, A883-A886.<br />
46. Zagier, Don Ramanujan’s mock theta functions and their applications (after Zwegers and Ono-<br />
Bringmann). Seminaire Bourbaki. Vol. 2007/2008. Astrisque No. 326 (2009), Exp. No. 986, vii-viii,<br />
143-164 (2010).<br />
47. Zwegers , S.P., <strong>Mock</strong> Theta Functions, Utrecht PhD Thesis, (2002) ISBN 90-393-3155-3<br />
<strong>UCLA</strong> <strong>Mathematics</strong> <strong>Department</strong>, Box 951555, Los Angeles, CA 90095-1555<br />
E-mail address: wdduke@ucla.edu<br />
<strong>UCLA</strong> <strong>Mathematics</strong> <strong>Department</strong>, Box 951555, Los Angeles, CA 90095-1555<br />
E-mail address: yingkun@ucla.edu