Zeros of Eisenstein Series

Stephanie Treneer
Western Washington University

May 8, 2010

Joint work with Sharon Garthwaite, Ling Long and Holly Swisher

Originated at the Women in Numbers Workshop, BIRS

The modular group
Γ := SL2 (Z) acts on H := {z ∈ C : Im(z) > 0} by
 
a b az + b
z=
c d cz + d

A fundamental domain for this action is

F := {z ∈ H : −1/2 ≤ Re(z) ≤ 1/2, |z| ≥ 1}

1 1
- €€€€ €€€€
2 2
Modular forms on Γ

A modular form of integer weight k for Γ is a holomorphic function

f : H → C with
   
az + b k a b
f = (cz + d) f (z), z ∈ H, ∈Γ
cz + d c d

We can extend f to a point at ∞ and write a Fourier series for f

at ∞ with q := e 2πiz :
∞
X
f (z) = a(n)q n
n=0
Eisenstein series
For even weight k ≥ 4,
X
Gk (z) := (cz + d)−k
c,d∈Z
(c,d)6=(0,0)

are modular forms of weight k.

We can normalize so that the constant term in the Fourier
expansion at ∞ is 1:
∞
1 2k X
Ek (z) := Gk (z) = 1 − σk−1 (n)q n
2ζ(k) Bk
n=1

More useful for us:

1 X
Ek (z) := (cz + d)−k
2
c,d∈Z
(c,d)=1
Eisenstein series

Eisenstein series can be viewed as arising from the Weierstrass ℘

function, which satisfies

(℘′ (z))2 = 4℘(z)3 − 60G4 ℘(z) − 140G6

This differential equation can be solved recursively to get

X∞
1
℘(z) = + (2k + 1)G2k+2 z 2k
z2
k=1
Zeros of Eisenstein series

How many zeros does Ek (z) have in F?

Where are the zeros located in F?

Number of zeros

Valence Formula: If f 6≡ 0 is a modular form of weight k, let

ντ (f ) be the order of f at τ ∈ F. Then

1 1 X k
ν∞ (f ) + νi (f ) + νω (f ) + ντ (f ) =
2 3 12
τ ∈F ,τ 6=i ,ω

Since Ek is holomorphic on H ∪ {∞}, it has ≈ k/12 zeros in F,

counted with multiplicity.
Location of zeros

Theorem (F.K.C. Rankin, Swinnerton-Dyer)

For all even k ≥ 4, the zeros of Ek (z) are all located on the arc

A := {z = e i θ : π/2 ≤ θ ≤ 2π/3}

1 1
- €€€€ €€€€
2 2
Outline of RSD result

Write k = 12n + s, s ∈ {4, 6, 8, 10, 0, 14}

Sufficient to show Ek (e i θ ) has at least n zeros in (π/2, 2π/3)

1 X
Define Fk (θ) := e ikθ/2 Ek (e i θ ) = (ce i θ/2 + de −i θ/2 )−k
2
c,d∈Z
(c,d)=1

Separate four terms with c 2 + d 2 = 1. Using Euler’s formula,

Fk (θ) = 2 cos(kθ/2) + R

where R is the remaining sum over terms with c 2 + d 2 > 1

Outline of RSD result

Bound |R| < 2 on [π/2, 2π/3]

The number of zeros of Fk (θ) (hence Ek (e i θ )), is at least the

number of zeros of 2 cos(kθ/2) on this interval, by an
Intermediate Value Theorem argument.

The zeros of 2 cos(kθ/2) can be easily counted. There are n

of them.
Interlacing of zeros

Theorem (Nozaki, ’08)

Any zero of Ek (e i θ ) lies between two consecutive zeros of
Ek+12 (e i θ ) on π/2 < θ < 2π/3.
Related results on Γ

Kohnen (2004) derives a closed formula for the precise

locations of the zeros of Ek (z) in terms of the Fourier
coefficients

Gekeler (2001) cites computational evidence that the

polynomials
Y
ϕk (x) = (x − j(z))
j(z) where Ek (z)=0
j(z)6=0,1728

are irreducible with full Galois group Sd where d is the degree

of ϕk (x).
Related results on Γ

Duke and Jenkins (2008) prove that the zeros of the weight k
weakly holomorphic modular forms fk,m (z) = q −m + O(q n+1 )
(with k = 12n + s as before) lie on the unit circle. They use a
circle method argument to bound the error term.
Congruence subgroup Γ(2)
   
1 0
Γ(2) := γ∈Γ:γ≡ (mod 2)
0 1
Γ(2) has genus 0 and three inequivalent cusps: 0, 1, ∞.

The fundamental domain for Γ(2) is

D := {z ∈ H : −1 ≤ Re(z) ≤ 1, |z − 1/2| ≥ 1/2, |z + 1/2| ≥ 1/2}

-1 1
Some odd weight Eisenstein series on Γ(2)

−1


Let χ(d) := d . For k ≥ 1 let

1 X
E2k+1,χ (z) := χ(d)(cz + d)−(2k+1)
2
(c,d)≡(0,1) (mod 2)
(c,d)=1

Then
 
az + b
E2k+1,χ = χ(d)(cz + d)−(2k+1) E2k+1,χ (z)
cz + d


for all z ∈ H and ca db ∈ Γ(2), and is holomorphic at the cusps.

E2k+1,χ is a modular form of weight 2k + 1 for Γ(2) with character

χ.
Some odd weight Eisenstein series on Γ(2)

Zeros of E2k+1,χ can be studied using the classical Jacobi elliptic

function cn(u), a doubly periodic generalization of cos u which
satisfies
  √ p p
κK 2Ku q cos u q 3 cos 3u q 5 cos 5u
cn = + 3
+ + ···
2π π 1+q 1+q 1 + q5
Θ22 (2z)
where K = π2 Θ23 (2z) and κ = Θ23 (2z)
, with
X X 2 /2
Θ2 (z) := q 1/8 q n(n+1)/2 , Θ3 (z) := qn
n∈Z n∈Z
Some odd weight Eisenstein series on Γ(2)
Using the Taylor series for cos u, we have
  ∞ ∞
!
κK 2Ku X (−1)k X (2r − 1)2k q (2r −1)/2
cn = u 2k
2π π (2k)! 1 + q 2r −1
k=0 r =1

and solving for cn(u), we have

∞
X ie2k π 2k+1 G2k+1 (2z) 2k
cn(u) = u
22k+1 (2k)! K 2k+1 κ
k=0

where
∞
4(−i )2k+1 X (2r − 1)2k q (2r −1)/4
G2k+1 (z) := E2k+1,χ |γ0 (z) =
e2k
r =1
1 + q (2r −1)/2

is the Fourier expansion of E2k+1,χ at the cusp 1.

Another view of cn(u)

An alternative description of cn(u) is found by solving the

differential equation
 2
d cn(u)

= (1 − cn2 (u)) 1 − λ + λcn2 (u)
du
to get

u2 u4 u6
cn(u) = 1 − + (1 + 4λ) − (1 + 44λ + 16λ2 ) + · · ·
2! 4! 6!
Θ42 (z)
Here λ(z) := Θ43 (z)
generates the field of meromorphic modular
functions for Γ(2) (analogous to j(z) for Γ). Note that λ(2z) = κ2 .
Zeros of G2k+1(z)

Equating coefficients in our two expressions for cn(u) we see that

(−1)k i π 2k+1 e2k G2k+1 (2z) (−1)k e2k G2k+1 (2z)

· =
22k+1 K 2k+1 κ Θ4k
3 (2z)G1 (2z)

is a polynomial p2k+1 (λ) of degree k − 1, having the same zeros as

G2k+1 in D.

This is expected, since this quotient is a modular function that is

holomorphic on H, but now we can easily compute the zeros!
Computational results

k p2k+1 (λ) Zeros

2 1 + 4λ −0.25
3 1 + 44λ + 16λ2 −0.0229, −2.7271
4 1 + 408λ + 912λ2 + 64λ3 −0.0025, −0.4598, −13.778
5 1 + 3688λ + 30764λ2 + 15808λ3 + 256λ4 −.00027, −.1280, −1.8792, −59.7425

For 2k + 1 ≤ 51 the λ-values of the zeros of G2k+1 are real

and lie in (−∞, 0).

The λ-zeros of G2k−1 interlace with the λ-zeros of G2k+1 .

The polynomials p2k+1 (λ) are irreducible with Galois group

Sk−1 for k ≤ 9.
Location of zeros of G2k+1

In the fundamental domain D for Γ(2), λ(z) ∈ (−∞, 0) precisely

on Re(z) = 1.

-1 1
Location of zeros of G2k+1

In order to use an RSD-type argument, we move the line

Re(z) = 1 to the arc |z + 1/2| = 1/2 using the transformation
z → −1/z. Hence we consider instead


G2k+1 |γ1 (z) := G2k+1 (z)|2k+1 01 −1
0

-1 1
Location of zeros of G2k+1

Theorem (GLST,’10)
For k ≥ 1, at least 90% of the zeros of G2k+1 (z) have real
λ-values in the range (−∞, 0].
Location of zeros of G2k+1

Recalling
1 X
E2k+1,χ (z) := χ(d)(cz + d)−(2k+1)
2
(c,d)≡(0,1) (2)

we define
X 1
Sk (α, β)(z) := .
(cz + d)k
(c,d)≡(α,β) (4)

Then
1
E2k+1,χ (z) = (S2k+1 (0, 1) + S2k+1 (2, 1) − S2k+1 (0, 3) − S2k+1 (2, 3)) .
2
Location of zeros of G2k+1

Restricting to zθ = 1/2e i θ − 1/2, we can show

G2k+1 |γ1 (zθ ) = (S(0, 3) + S(1, 0) + S(2, 1) + S(3, 2)) (e i θ ).

We balance the exponents as in RSD:

F2k+1 (zθ ) := (e i θ/2 )2k+1 G2k+1 |γ1 (zθ )

 
= S e2k+1 (0, 3) + S e2k+1 (1, 0) + Se2k+1 (2, 1) + S
e2k+1 (3, 2) (θ)

where
X
ek (α, β)(θ) :=
S (ce i θ/2 + de −i θ/2 )−k
(c,d)≡(α,β) (4)
(c,d)=1
Location of zeros of G2k+1

F2k+1 (zθ ) =
 
Se2k+1 (0, 3) + S
e2k+1 (1, 0) + S
e2k+1 (2, 1) + S
e2k+1 (3, 2) (θ)

As in RSD, we extract the two terms with c 2 + d 2 = 1 to create

our main term:
 
θ(2k + 1)
F2k+1 (zθ ) = −2i sin + R2k+1 (zθ )
2

Our goal is to show |R2k+1 (zθ )| < 2.

Bounding the error term

We now collect terms satisfying c 2 + d 2 = N for fixed N > 1.

The terms with N ≤ 100 are dealt with carefully

The terms with N > 100 can be easily bounded using an

appropriate integral
Bounding the error term

For each ordered pair of nonnegative integers (a, b) with a odd and
b even, define
X
P(a, b)(θ) = (ce i θ/2 + de −i θ/2 )−2k−1
(|c|,|d|)=(a,b) or (b,a)

where the sum is over those terms appearing in F2k+1 (zθ ).

For example,
   
2i θ(2k + 1)
P(3, 0) = (−3e i θ/2 )−2k−1 + (3e −i θ/2 )−2k−1 = sin
32k+1 2

Due to symmetry, each P(a, b) is purely imaginary.

Bounding the error term

We assume that 2k + 1 > 51. When b = 0, we have the cases,

(a, b) ∈ {(3, 0), (5, 0), (7, 0), (9, 0)}.

Then
 
2 (2k + 1)θ 2
|P(a, b)(θ)| = sin ≤ .
a2k+1 2 a51

The contribution from these terms is negligible.

Bounding the error term

When b 6= 0 we have
2 2
|P(a, b)| ≤ 2 + .
((a − b)2 + 4ab sin (θ/2)) ((a − b) + 4ab cos2 (θ/2))k
k 2

For fixed θ, this is worst when a and b are both small and
a − b = 1.

When a − b = 1 and θ approaches 0 or π, we have problems.

We must bound sin2 (θ/2) and cos2 (θ/2) away from 0.

Bounding the error term

2 2
|P(a, b)| ≤ 2
+ .
((a − b)2 + 4ab sin (θ/2)) ((a − b) + 4ab cos2 (θ/2))k
k 2

Note that max(cos2 (θ/2), sin2 (θ/2)) ≥ 1/2.

We will require a bound min(cos2 (θ/2), sin2 (θ/2)) > α2 > 0.

Then we have
2 2
|P(a, b)| ≤ + .
((a − b)2 + 4abα2 )k ((a − b)2 + 2ab)k
Bounding the error term

When c 2 + d 2 = N > 100 and | cos θ| < β, we can bound the

terms by

|ce i θ/2 + de −i θ/2 |2 = c 2 + 2cd cos θ + d 2 ≥ (1 − β)(c 2 + d 2 )

The√number of terms 2 2
√ with c + d = N is at most
2(2 N + 1) ≤ 5 N, so
∞
X
|R(zθ )| < E100 (α) + 5N 1/2 ((1 − β)N)−k−1/2 .
N=101

Bounding the sum with an integral yields

 
−k− 12 5 −k+1
|R(zθ )| < E100 (α) + (1 − β) · 100 .
k −1
Bounding the error term

 
−k− 12 5
|R(zθ )| < E100 (α) + (1 − β) · 100−k+1 <2
k −1

Balancing α and β to maximize the range of θ with error less than

2 yields a range of (0.05π, 0.95π).
Wrapping up the proof
 
Now we count the values θ at which 2 sin θ(2k+1) 2 = ±2 and
apply the Intermediate Value Theorem to
 
θ(2k + 1)
iF2k+1 (zθ ) = 2 sin + iR(zθ ).
2

We see that G2k+1 |γ1 (zθ ) must have at least one zero in each
interval  
Aπ (A + 2)π
, ⊂ (0.05π, 0.95π),
2k + 1 2k + 1
where A > 0 is odd.

Hence we are assured that 90% of the zeros of G2k+1 have real,
negative λ-values.
Other groups

Fricke groups:
 √ 
0 −1/ p
Γ0 (p) ∪ Γ0 (p)Wp , Wp := √
p 0

Miezaki, Nozaki and Shigezumi (2007) For p = 2, 3, all zeros

located on arc of fundamental domain

Shigezumi (2007) partial result towards cases p = 5, 7

Hahn (2007) has more general result about Fuchsian groups

of the first kind
Summary

The RSD method is pushed to its limits as we near the cusps

The connection to cn(u) could provide a different method

which would give us all the zeros

No similar results are known for groups of nonzero genus