You are on page 1of 67

The Analysis of Automorphic Forms

Sean P Gomes
An essay submitted in partial fulllment of
the requirements for the degree of
B.Sc. (Honours)
Pure Mathematics
University of Sydney
November 2011
CONTENTS
Preface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
Chapter 1. Geometry of the Poincar Half-Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1. Length, Angles and Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2. The action of PSL
2
(R) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3. Classication of PSL
2
(R)transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4. 1 as a homogeneous space and decompositions of PSL
2
(R) . . . . . . . . . . . . . 12
Chapter 2. Harmonic Analysis on 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1. Invariant Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2. Eigenfunctions of . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3. Key Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4. The Green Function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5. Invariant Integral Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Chapter 3. Fuchsian Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2. PSL
2
(R) as a topological group. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3. Fuchsian Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4. Fundamental Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.5. Kloosterman Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Chapter 4. Automorphic Forms and their Spectra . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.1. Spectral Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2. Automorphic Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3. Eisenstein Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.4. The Automorphic Green Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.5. Small Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
iii
Preface
The eld of analytic number theory traditionally relied primarily on the methods of
complex analysis. The theory of analytic functions of one complex variable yields results
such as the celebrated Prime Number Theorem. The developments of abelian harmonic
analysis then took centre stage. In particular Fourier analysis on the Euclidean plane
provided a new platform from which to attack the mysteries of the primes. Prominent
results include the recent GreenTao Theorem on arithmetic progressions in primes. En-
ter the hyperbolic plane. With its highly symmetric structure, the theory of functions
which are invariant under lattice actions is key. In particular, modular forms have long
been important to the algebraic aspects of number theory, providing for example, a short
proof of Lagranges Theorem of sums of four squares and a considerably longer proof of
Fermats Last Theorem.
The study of the spectral theory of automorphic forms has been an exciting offshoot
from the study of modular forms. The eld is the study of the spectral decomposition of
automorphic functions with respect to the Laplacian, and is an exotic blend of functional
analysis, harmonic analysis, Riemannian geometry and number theory.
Much of the progress in this eld came from the 1940s onwards, with a major early
result being Selbergs trace formula, an analogue to the Poisson summation formula
from Fourier analysis. The spectral methods of automorphic forms maintain a dominant
position in analytic number theory today.
For a more detailed overview of the historic developments of analytic number theory,
one may consult the wonderful book IwaniecKowalski [10], from which this account is
drawn.
iv
Introduction
In this essay, we shall develop the basic aspects of automorphic function theory, with
a strong emphasis on the use of the machinery of harmonic analysis to establish spectral
theoretic results.
In Chapter 1, we establish basic properties of hyperbolic geometry through the use
of the Poincare upper half-plane model. Of particular importance are the group of
orientation-preserving isometries, the fractional linear transformations.
In Chapter 2, we shall discuss harmonic analysis on 1 in terms of the study of the
LaplaceBeltrami differential operator. In this context, we introduce the rudimentary
techniques of harmonic analysis on the hyperbolic plane and state the major results.
Particularly important is Theorem 2.3.3, which establishes the Fourier series expansion
of suitable functions. These Fourier series will play a key role in our subsequent analysis.
In Chapter 3, we study the Fuchsian groups. Fuchsian groups provide a rich arith-
metic structure in hyperbolic space which is analogous to the structure of lattices in C.
We construct fundamental domains for Fuchsian groups in order to provide some geo-
metric intuition. We shall conclude this chapter with a construction of the Kloosterman
sum, an exponential sum that is a key ingredient in our later results.
In Chapter 4, we begin our foray into the spectral theory of automorphic forms.
We begin by gathering some general spectral theoretic results from functional analysis.
We shall use these in conjunction with the machinery of previous chapters to determine
the discrete spectrum of the LaplaceBeltrami operator. To conclude, we shall discuss
an important conjecture due to Selberg about the spectral gap of the LaplaceBeltrami
operator on the space of all automorphic functions.
v
Acknowledgements
First and foremost I offer my sincerest gratitude to my supervisor, Dr Anne Thomas,
who has supported me throughout my thesis with her time and knowledge. Without her,
this thesis would not have been possible.
I would like to think all of my lecturers, from both my honours year and the entirety of
my degree. In particular I would like to thank Dr Daniel Daners and Dr James Parkin-
son for playing major roles in cultivating my interest in mathematical analysis.
I would also like to thank my fellow honours students. They have all been a pleasure
to work with, and I wish them all the best in their every endeavour.
Thanks to my family and friends, for their constant stream of love and support this
year and every year.
Special thanks to Rachel, for her love and support (and for keeping me well fed
during the more stressful moments this year!)
vi
CHAPTER 1
Geometry of the Poincar Half-Plane
In this chapter, we shall rst in Section 1.1 dene the Poincar Half-Plane and study
the basic aspects of its geometry. The rest of the chapter shall then be devoted to the
denition and study of an action of PSL
2
(R) on the Poincar Half-Plane which preserves
its geometric structure. Our treatment is based on that of Katok [8, Chapter 1] and
Iwaniec [7, Chapter 1].
1.1. Length, Angles and Area
We begin by establishing the basic geometry of the Poincar Half-Plane model of
the hyperbolic plane. In particular, we make the notions of length, and area precise
through the denition of a metric and a measure on the Poincar Half-Plane model re-
spectively. This is one of the most commonly-used models of the hyperbolic plane, and
the one in which we will be working for the remainder of this essay.
Denition 1.1.1. The Poincar Half-Plane is dened to be the set:
1 := z C : Im(z) > 0.
We can now dene a Riemannian metric on the half-plane by making the identica-
tion of the tangent spaces T
z
1

= C with R
2
in the obvious way and using the metric
tensor:
g
z
:=
_
1/ Im(z)
2
0
0 1/ Im(z)
2
_
. (1.1.2)
This makes 1 a Riemannian manifold. To avoid use of results from differential
geometry without proof, we instead use this denition of g
z
to motivate the following
material, arriving at the same geometry with little assumed knowledge.
Denition 1.1.3. A path is a continuous and piecewise continuously differentiable map
: [0, 1] 1. We say that is a path from (0) to (1).
Denition 1.1.4 (Hyperbolic Length). If is a path, we dene the hyperbolic length of
to be:
h() :=
_
1
0
[

(t)[
Im((t))
dt
By a change of variables, it is easy to see that the length of a path is independent of
a particular choice of parametrisation. For this reason we will often use the term path
to refer to a subset of 1that is the image of a path in the sense of Denition 1.2 rather
1
2 1. GEOMETRY OF THE POINCAR HALF-PLANE
than the piecewise continuously differentiable map itself. We will be more explicit in
situations where this leads to ambiguity.
Denition 1.1.5 (Hyperbolic Distance). For z, w 1 we dene:
(z, w) := infh() : a path from z to w.
Proposition 1.1.6. The function : 11R is a metric on 1.
Proof. Let z, w 1. For any path from z to w, h() has a non-negative integrand,
hence (z, w) 0 for all z, w 1. Moreover equality is attained if and only if

(t) = 0
for all t [0, 1], that is if and only if is constant and thus z = w.
If we dene ()(t) := (1 t) for a given path , we have h() = h() by a change
of variables and moreover, this reversal of orientation provides a bijective correspon-
dence between paths from z to w and paths from w to z. Hence for all z, w 1we have
(z, w) = (w, z).
Now let z, w, v 1. For all > 0, from the denition of hyperbolic length as an
innum, we can nd paths
1
and
2
from z to w and w to v respectively such that
h(
1
) < (z, w) + /2,
and
h(
2
) < (w, v) + /2.
Since the concatenation of these paths is a path from z to v with length h(
1
) + h(
2
)
(again by changing variables), we have:
(z, v) h(
1
) + h(
2
) < (z, w) + (w, v) + .
Letting 0 establishes the triangle inequality, and completes the proof that is a
metric on 1.
Of all the paths between two given points in the Poincar Half-Plane, we are most
interested in those which have minimal hyperbolic length.
Denition 1.1.7. A geodesic in 1 is a path between two distinct points of 1 whose
hyperbolic length is minimal amongst the set of such paths.
We now make the rst step towards nding the geodesics in 1.
Lemma 1.1.8. If 0 < a < b then (t) = i(a + (b a)t) is the unique path (up to repa-
rameterisation) from ia to ib with minimal hyperbolic length. That is, h() = (ia, ib).
Moreover, h() = log(b/a).
1.2. THE ACTION OF PSL2(R) 3
Proof. Let (t) := x(t) + iy(t) be an arbitrary path from ia to ib, where x(t) and y(t)
are real-valued functions. Now from Denition 1.1.4 above:
h() =
_
1
0
_
x

(t)
2
+ y

(t)
2
y(t)
dt

_
1
0
[y

(t)[
y(t)
dt

_
1
0
y

(t)
y(t)
dt
= log(y(1)) log(y(0))
= log(b/a)
=
_
1
0
[b a[
a + (b a)t
dt
= h().
Since equality is attained if and only if x

(t) is identically zero and y(t) is non-decreasing,


(t) = i(a + (b a)t) is indeed the unique path from ia to ib of minimal hyperbolic
length.
The hyperbolic metric from Denition 1.1.5 above motivates the denition of the
corresponding hyperbolic area, as follows. Let m denote the Lebesgue measure on the
complex plane. Then we can make the following denition:
Denition 1.1.9. For A 1, we dene the hyperbolic area (A) of A to be:
(A) :=
_
A
dm
Im(z)
2
whenever the above Lebesgue integral exists.
1.2. The action of PSL
2
(R)
We now dene an action of the group PSL
2
(R) on 1. In doing so, we introduce a
distinguished class of transformations of 1, which are exactly the orientation-preserving
isometries. The rst of the two main results of this section is Theorem 1.2.14, which
establishes that PSL
2
(R) acts isometrically on 1 with respect to the hyperbolic metric.
The second main result is Theorem 1.2.26, which nds all geodesics on 1. We shall
also, in Denition 1.2.29, extend the action of PSL
2
(R) to the compactication of 1.
Denition 1.2.1. An isometry of a metric space (M, d) is a bijective transformation
f : MM such that:
d(x, y) = d(f(x), f(y)) for all x, y M.
It is easy to see that the set of isometries of a given space forma group when equipped
with the binary operation of composition. We denote this group for the Poincar Half-
Plane by Isom(1).
4 1. GEOMETRY OF THE POINCAR HALF-PLANE
Denition 1.2.2. Given a group G and a nonempty set X, we dene a group action to
be a map : GXX denoted (g, x) g x such that:
1 x = x for all x X; and
g (h x) = (gh) x for all g, h G, x X.
Henceforth we will often omit the . Now we recall the denition of the Special
Linear Group of 2 2 real matrices:
Denition 1.2.3 (Special Linear Group). We dene
SL
2
(R) := A R
22
: det A = 1.
If we factor out the normal subgroup I we obtain the Projective Special Linear
Group over R. This is the main group of interest to us in this chapter.
Denition 1.2.4 (Projective Special Linear Group). We dene
G := PSL
2
(R) := SL
2
(R)/I.
For the remainder of this chapter, Gshall denote PSL
2
(R) unless otherwise specied.
Proposition 1.2.5. The group SL
2
(R) acts on 1 via the map:
_
a b
c d
_
z =
az + b
cz + d
.
Moreover, the only elements of SL
2
(R) that x every point of 1 are the elements I.
Proof. Let =
_
a b
c d
_
and

=
_
a

_
be elements of SL
2
(R). Now by direct
calculation:
Im( z) = Im
_
az + b
cz + d
_
= Im
_
(az + b)(c z + d)
[cz + d[
2
_
=
ad bc
[cz + d[
2
Im(z)
= [cz + d[
2
Im(z) for all z 1. (1.2.6)
so z 1 for all z 1. We now verify that the map (, z) z is a group action:
(

z) =
a
a

z+b

z+d

+ b
c
a

z+b

z+d

+ d
=
a(a

z + b

) + b(c

z + d

)
c(a

z + b

) + d(c

z + d

)
=
(aa

+ bc

)z + (ab

+ bd

)
(ca

+ dc

)z + (cb

+ dd

)
= (

) z.
1.2. THE ACTION OF PSL2(R) 5
Moreover, we have:
1 z =
_
1 0
0 1
_
z =
z + 0
0 + 1
= z.
Whence is indeed a group action. Finally suppose xes every point in 1. Then
az + b = cz
2
+ dz for all z 1.
Since the zero polynomial is the only polynomial that is identically zero on the upper
half plane, we can conclude that a = d and b = c = 0. Thus from the determinant
condition ad bc = 1 we have:
= I
as required.
A group action with the property that nontrivial group elements do not x every
element of the set that they act upon is said to be faithful. To obtain a faithful action on
1, we take the quotient of SL
2
(R) by the normal subgroup I SL
2
(R) of elements
which x every point in 1.
Proposition 1.2.7. The quotient group G = PSL
2
(R) = SL
2
(R)/I acts faithfully
on 1 via the induced map:
__
a b
c d
_
I
_
z =
az + b
cz + d
.
It is important to note that the elements of G = PSL
2
(R) are cosets of SL
2
(R), each
coset gI containing the two elements g and g of SL
2
(R). By abuse of notation
however, we will usually denote a coset in G by one of its representatives in SL
2
(R). On
the rare occasion that distinction between these two objects is required, we will be more
explicit in our notation for cosets.
Remark 1.2.8. We will frequently identify group elements g G with the correspond-
ing transformation z gz.
Before proving that G acts isometrically on 1, we discuss three important examples
of transformations of 1.
Example 1.2.9. The group element
g
1
:=
_
2 0
0 1/2
_
G
acts upon 1 by dilating points by a factor of 4 since for all z 1, we have
_
2 0
0 1/2
_
z =
2z
1/2
= 4z.
This transformation is not a Euclidean isometry, since it scales all Euclidean lengths by
a factor of 4. Moreover, g
1
has no xed points in 1. The transformation g
1
is an example
of a hyperbolic transformation.
6 1. GEOMETRY OF THE POINCAR HALF-PLANE
Example 1.2.10. The group element
g
2
:=
_
1 1
0 1
_
G
acts upon 1 by translating points 1 unit to the right since for all z 1, we have
_
1 1
0 1
_
z =
z + 1
1
= z + 1.
This transformation is a Euclidean translation and hence a Euclidean isometry, preserv-
ing Euclidean lengths, angles and areas. Moreover, g
2
has no xed points in 1. The
transformation g
2
is an example of a parabolic transformation.
Example 1.2.11. The group element
g
3
:=
_
0 1
1 0
_
G
acts upon 1by reecting points through the imaginary axis and inverting their modulus.
To see this observe that
_
0 1
1 0
_
z =
1
z
and hence
[g
3
(z)[ =
1
[z[
.
Hence g
3
has the effect of inverting the modulus of z. Since a reection in the real axis
is given by the map z z, and a 180

-rotation about the origin is given by the map


z z, we can conclude that g
3
has the effect of composing a reection in the real axis
with a 180

rotation about the origin and an inversion of modulus. The element g


3
is not
a Euclidean isometry since, for example, [2i i[ = 1, whilst
[g
3
(2i) g
3
(i)[ =

1
i

1
2i

=
1
2
.
The transformation g
3
has exactly one xed point in 1, at z = i. The transformation g
3
is an example of an elliptic transformation.
Proposition 1.2.12. For every g G, the transformation: g : z gz is conformal.
That is, the action of every g G preserves Euclidean angles.
Proof. See Katok [8, p.811].
Remark 1.2.13. It is an exercise in differential geometry to show that Euclidean angles
in tangent spaces of 1 are in fact equal to the hyperbolic angles which arise from the
metric tensor g
z
(dened in 1.1.2) and resulting inner product on tangent spaces. This
is a consequence of the matrix g
z
being diagonal. Bearing this equality in mind we will
treat the word angles in this essay as being those from standard Euclidean geometry in
C.
We now come to one of the main results of this section.
1.2. THE ACTION OF PSL2(R) 7
Theorem 1.2.14. For every g G, the transformation: g : z gz is an isometry of 1.
Hence we can regard G as a subgroup of Isom(1).
Proof. From Denition 1.1.5 above, it sufces to show that the hyperbolic length of
paths is invariant under the action of the group G. Now let
g :=
_
a b
c d
_
G
and let be a path in 1. From the denition of hyperbolic length and the chain rule, we
have:
h(g ) =
_
1
0
[(g )

(t)[
Im(g )(t)
dt
=
_
1
0
[(g

)(t)[ [

(t)[
Im(g )(t)
dt. (1.2.15)
Now using the quotient rule we obtain the following identity:
g

(z) =
a(cz + d) c(az + b)
(cz + d)
2
= (cz + d)
2
for all z 1. (1.2.16)
Furthermore, from Equation (1.2.6) in the proof of Proposition 1.2.5, we have:
Im(gz) = [cd + d[
2
Im(z) for all z 1.
Hence from Equation (1.2.15), we can conclude that:
h(g ) =
_
1
0
[

(t)[
Im(z)
dt = h()
as required.
Since G acts on 1 faithfully, we can regard G as a subgroup of Isom(1) by identifying
every group element g with the corresponding map z gz. This completes the proof.

From Theorem 1.2.14 above, we obtain the following corollary.


Corollary 1.2.17. The group G acts by homeomorphisms on 1.
Remark 1.2.18. It can be shown that the transformations in G, together with the reec-
tion r : z z, generate Isom1. In particular, the transformations in G are precisely
the orientation-preserving isometries of 1. For a proof of this see [8, p.811].
In our construction and study of Ford polygons in Chapter 3, we will need to consider
the set of points in 1 at which a particular g G does not distort the Euclidean metric.
This motivates the following two denitions.
Denition 1.2.19. If g =
_
a b
c d
_
G, c ,= 0 and z 1, we dene:
j
g
(z) := [cz + d[
2
to be the deformation of g at z.
8 1. GEOMETRY OF THE POINCAR HALF-PLANE
Note that the set of points z 1 such that j
g
(z) = 1 is the Euclidean circle:
z 1 : [z (dc
1
)[ = [c[
1
.
Denition 1.2.20. We dene the isometric circle C
g
of g to be the set of points at which
g has deformation equal to 1.
Remark 1.2.21. As shown in the proof of Theorem 1.2.14, the quantity
j
g
(z) = [cz + d[
2
from Denition 1.2.19 is essentially a measure of how badly a given
g G fails to be a Euclidean isometry. In particular, by using (1.2.6), (1.1.2) and
Theorem 1.2.14, we can see that g does not distort the Euclidean metric on its isometric
circle C
g
.
We are now going to establish an explicit formula for the hyperbolic metric , in
Corollary 1.2.24, and characterise geodesics in 1, in Theorem 1.2.26. We rst note:
Lemma 1.2.22. Any two points z and w in 1 either lie on the same vertical line or on
some common Euclidean semicircle centred on the real axis.
Proof. If z and w are points in 1, then either they both lie on the same vertical Euclidean
line, or the Euclidean perpendicular bisector of the Euclidean interval joining z and w
meets the real axis at some point c. In the latter case, the point c is equidistant from z
and w, whence both z and w lie on some Euclidean semicircle centred at c R.
We now prove the useful Lemma 1.2.23, which will greatly simplify the computa-
tions in Corollary 1.2.24 and Theorem 1.2.26 below.
Lemma 1.2.23. If z, w 1, then there exists g G such that gz and gw lie on the
imaginary axis of 1.
Proof. Let z, w 1. From Lemma 1.2.22 above, z and w either lie on the same vertical
line or on some common semicircle centred on the real axis. If z and w lie on the same
vertical line, then Re(z) = Re(w) = k R. Hence the group element:
g :=
_
1 k
0 1
_
G = PSL
2
(R)
satises Re(gz) = Re(gw) = 0, so gz and gw lie on the imaginary axis of 1as required.
Now suppose z and w both lie on the semicircle centred at c R with radius r R
+
. In
particular, suppose that z = c + re
i
for some (0, ). Now set:
g :=
_
1

2r

c+r

2r
1

2r

cr

2r
_
1.2. THE ACTION OF PSL2(R) 9
and note that det(g) = 1, so we may consider g to be in G = PSL
2
(R). We have:
gz =
c + re
i
(c + r)
c + re
i
(c r)
=
e
i
1
e
i
+ 1
=
e
i
e
i
[e
i
+ 1[
2
which is purely imaginary. Since our choice of g did not depend on , the same argument
shows that gw also lies on the imaginary axis, as required.
This yields a convenient expression for hyperbolic distance.
Corollary 1.2.24. For z, w 1, the distance (z, w) is given by:
cosh (z, w) = 1 +
[z w[
2
2 Im(z) Im(w)
(1.2.25)
Proof. We establish the result by rst proving it for z and w on the imaginary axis and
then invoking Lemma 1.2.23.
Suppose that z = ir and w = is, where s > r. From Lemma 1.1.8 we have:
cosh (z, w) = cosh log(b/a) =
a/b + b/a
2
=
a
2
+ b
2
2ab
= 1 +
(b a)
2
2ab
as required.
Now if g =
_
a b
c d
_
G, then
[gz gw[
2
Im(gz) Im(gw)
=

az+b
cz+d

aw+b
cw+d

2
Im(gz) Im(gw)
=
[z w[
2
[cz + d[
2
[cw + d[
2
Im(gz) Im(gw)
=
[z w[
2
Im(z) Im(w)
using Equation (1.2.6). Hence the right-hand side of Equation (1.2.25) is invariant under
the action of G. Since (1.2.25) holds on the imaginary axis, and by Lemma 1.2.23 every
pair of points can be translated to the imaginary axis by an isometry, we are done.
With the help of Lemma 1.2.23, we can now completely characterise geodesics in 1.
Theorem 1.2.26. The geodesics in 1 are segments of vertical Euclidean lines and seg-
ments of Euclidean semicircles orthogonal to the real axis.
Proof. Suppose z and w are in 1, and let S be the set of all paths from z to w. If z and
w both lie on the imaginary axis then Lemma 1.1.8 asserts the existence and uniqueness
of a S of minimal length, which must be the segment of the imaginary axis joining
10 1. GEOMETRY OF THE POINCAR HALF-PLANE
z to w.
Now suppose z and w do not both lie on the imaginary axis. Then there either exists
an unique semicircle C in 1 orthogonal to the real axis and passing through z and w,
or there exists a unique vertical Euclidean line passing through z and w. From Lemma
1.2.23, we know that there exists a g G such that gz and gw lie on the imaginary axis.
Now since by Theorem 1.2.14, the map g is an isometry, g provides a length-preserving
bijection between the set of paths from z to w and the set of paths from gz to gw. Hence
we can conclude that there exists a unique geodesic S, which is the segment of the
semicircle C or line between z and w. This completes the proof.
Proposition 1.2.27. Elements of G preserve hyperbolic area.
Proof. This is a consequence of Theorem 1.2.14, which says that elements of Gpreserve
hyperbolic distance. For a complete proof see [8, p.811].
In the remainder of this essay, we will sometimes need to study the behaviour of
functions on the boundary of 1. For this reason we shall nowextend 1in the following
natural way.
Denition 1.2.28 (Extended Poincar Half Plane). The extended hyperbolic plane

1
is dened to be the topological closure of 1 in the extended complex plane.

1 := 1 R .
The boundary of the extended hyperbolic plane is

1 = R .
We can extend each of the transformations of 1 in G = PSL
2
(R) to

1 as follows:
Denition 1.2.29 (Extended PSL
2
(R)Action). Let g =
_
a b
c d
_
G.
If c = 0 then we dene gz =
_
az+b
d
for z ,=
for z = .
If c ,= 0 then we dene gz =
_

_
a
c
for z =
for z =
d
c
az+b
cz+d
for all other z

1.
Proposition 1.2.30. The map: (g, z) gz dened in Denition 1.2.29 is a group action
on

1.
Denition 1.2.31. A hyperbolic polygon is a closed subset of

1 bounded by nitely
many hyperbolic geodesic segments.
Theorem 1.2.32 (GaussBonnet). If a hyperbolic triangle A has angles , , then
(A) = ( + + ).
Proof. See [8, p.1314].
1.3. CLASSIFICATION OF PSL2(R)TRANSFORMATIONS 11
1.3. Classication of PSL
2
(R)transformations
We now classify PSL
2
(R)transformations based on the nature of their xed points
in

1. To motivate our denitions, we consider the solutions in 1 to the equation:
_
a b
c d
_
z = z (1.3.1)
where
g :=
_
a b
c d
_
G is nontrivial and c ,= 0.
From Denition 1.2.7, Equation (1.3.1) simplies to the quadratic equation:
cz
2
+ (d a)z b = 0.
Since c is assumed to be nonzero, the number of solutions is determined by the discrim-
inant:
(d a)
2
+ 4bc = a
2
+ d
2
2ad + 4bc = (a + d)
2
4 = tr(g)
2
4
where tr(g) := a + d is the trace of the chosen representative matrix.
Moreover, we have the equality gI = (g)I in G, so the quantity Tr(g) :=
[ tr(g)[ is well dened and is a convenient quantity to use in the classication of trans-
formations.
Denition 1.3.2. Let g G be a nontrivial transformation.
If Tr(g) > 2, the transformation g is hyperbolic.
If Tr(g) = 2, the transformation g is parabolic.
If Tr(g) < 2, the transformation g is elliptic.
Trace is invariant under conjugacy, so the above classes of transformations are dis-
joint and closed under conjugation. Bearing this in mind, we can use Jordan Canonical
Form to classify PSL
2
(R)transformations up to conjugacy in G, as follows.
Proposition 1.3.3. Every hyperbolic transformation is conjugate to a transformation in
the subgroup:
A :=
__
a 0
0 a
1
_
G : a R
+
_
G.
Dene
A(a) :=
_
a 0
0 a
1
_
A.
Proposition 1.3.4. Every parabolic transformation is conjugate to a transformation in
the subgroup:
N :=
__
1 t
0 1
_
G : t R
_
G.
Dene
N(t) :=
_
1 t
0 1
_
N.
12 1. GEOMETRY OF THE POINCAR HALF-PLANE
Proposition 1.3.5. Every elliptic transformation is conjugate to a transformation in the
subgroup:
K :=
__
cos sin
sin cos
_
G : R
_
G.
Dene
K() :=
_
cos sin
sin cos
_
K.
The three transformations from Examples 1.2.9, 1.2.10 and 1.2.11
above are in their respective normal forms.
Now an element g G xes some x

1 if and only if h
1
gh xes h
1
x, for all
h G. Hence conjugate elements of G x the same number of points in

1. From
Propositions 1.3.3, 1.3.4 and 1.3.5 above, we can deduce the following general results
about the xed points of elements of G.
Proposition 1.3.6. Hyperbolic transformations have exactly two distinct xed points in

1. These both lie on the boundary



1.
Proof. This follows from the fact that A(a) xes exactly 0 and in

1.
Proposition 1.3.7. Parabolic transformations have exactly one xed point in

1. This
xed point lies on the boundary

1.
Proof. This follows from the fact that N(t) xes exactly in

1.
Proposition 1.3.8. Elliptic transformations have exactly one xed point in

1. This xed
point lies in the interior of

1.
Proof. This follows from the fact that K() xes exactly i in 1.
1.4. 1 as a homogeneous space and decompositions of PSL
2
(R)
Up to this point we have worked with 1 solely using Cartesian coordinates. We can
also view 1 as a homogeneous space for the transformation group G = PSL
2
(R), and
for various aspects of harmonic analysis on 1 this is a convenient viewpoint. In this
section, we shall also develop several decompositions of PSL
2
(R).
To begin, we show that the action of G is transitive on 1, and compute the stabiliser
of i 1.
Lemma 1.4.1. The group G acts transitively on 1, and the stabiliser of i 1 is the
subgroup K from Proposition 1.3.5.
Proof. Let z and w be points in 1 and dene r := Im(w)/ Im(z). Then w rz R
and the group element:
g :=
_
r
1/2
r
1/2
(w rz)
0 r
1/2
_
G
1.4. H AS A HOMOGENEOUS SPACE AND DECOMPOSITIONS OF PSL2(R) 13
maps z to w. Hence G acts transitively on 1.
If
g =
_
a b
c d
_
G
xes i 1, we have:
ai + b = c + di.
By equating coefcients, we can deduce that:
g =
_
a b
b a
_
for some a, b R. Whence by the determinant condition, a
2
+ b
2
= 1 and we have
a = cos() and b = sin() for some R, and thus g K.
As a consequence of Lemma 1.4.1 and the Orbit-Stabiliser Theorem, we can now
identify 1 with G/K. That is, we identify the point z 1 with the coset gK of ele-
ments of G which map i to z.
In the remainder of this section, we develop three ways of factorising elements of the
transformation group G. These are the Iwasawa Decomposition, the Bruhat Decomposi-
tion and the Cartan Decomposition.
The rst of these, the Iwasawa Decomposition, provides a connection between the Carte-
sian coordinates of points of 1 and the corresponding cosets in G/K. The Iwasawa
decomposition is necesssary for developing Fourier analysis on 1.
Proposition 1.4.2 (Iwasawa Decomposition). Every nontrivial g G can be written in
the form:
g = N(t
0
)A(a
0
)K(
0
)
for some unique t
0
R, a
0
R
+
and
0
[0, ).
Proof. Let
g =
_
a b
c d
_
G = PSL
2
(R).
Given a R and [0, 2), by matrix multiplication we have:
A(a)K() =
_

a
1
sin() a
1
cos
_
where the denotes an arbitrary entry. By the polar coordinate representation of points
in R
2
, we can nd unique elements a
0
R
+
and
0
[0, ) such that the bottom row of
A(a
0
)K(
0
) agrees with the bottom row of g. Now let:
A(a
0
)K(
0
) =
_
a

c d
_
.
From the determinant condition, c and d cannot both be zero. Suppose c ,= 0. Then by
matrix multiplication:
N
_
a a

c
__
a

c d
_
=
_
a

+ c
aa

c

c d
_
=
_
a
c d
_
. (1.4.3)
14 1. GEOMETRY OF THE POINCAR HALF-PLANE
Since c is nonzero, the determinant equation:
ad cx = 1
has exactly one solution x R. Hence we in fact have:
N
_
a a

c
__
a

c d
_
=
_
a b
c d
_
.
Whence g = N(t
0
)A(a
0
)K(
0
), where t
0
:= (a a

)/c. Since c ,= 0 by assumption,


the chosen t
0
is the unique real number satisfying (1.4.3). If c = 0 and d ,= 0, we can
proceed similarly. As a
0
and
0
were already shown to be unique, this completes the
proof.
The next decomposition, the Bruhat Decomposition, is related to the study of Kloost-
erman sums, which we shall encounter in Section 3.5.
Proposition 1.4.4 (Bruhat Decomposition). Every g G can be written in one of the
two forms:
g = N(t
1
)A(a)N(t
2
) (1.4.5)
or
g = N(t
1
)A(a)N(t
2
) (1.4.6)
where t
1
, t
2
R, a R
+
and = K(/2) =
_
0 1
1 0
_
. Here, (1.4.5) factorises all
upper triangular g and (1.4.6) factorises the remaining group elements.
Proof. If g =
_
p q
0 r
_
, then the determinant condition forces r = p
1
. Hence by the
computation:
_
1 t
1
0 1
__
p 0
0 p
1
__
1 t
2
0 1
_
=
_
p pt
2
+ p
1
t
1
0 p
1
_
we can nd suitable t
1
and t
2
to represent any upper-triangular g in the required form
(1.4.6). Such a representation is not unique.
If g =
_
p q
r s
_
is not upper-triangular, then the determinant condition forces
q = (ps 1)/r. (1.4.7)
Hence by the computation:
_
1 p/r
0 1
_

_
r 0
0 r
1
__
1 s/r
0 1
_
=
_
p
r s
_
together with (1.4.7), we can conclude that any g that is not upper-triangular can be
expressed in the required form (1.4.6).
The third and nal decomposition is the most convenient way of studying the Green
Function, introduced in Chapter 2.
1.4. H AS A HOMOGENEOUS SPACE AND DECOMPOSITIONS OF PSL2(R) 15
Proposition 1.4.8 (Cartan Decomposition). Every g G can be written in the form:
g = K(
0
)A(e
r/2
)K(
0
)
for some
0
[0, ), r 0 and
0
[0, ). Moreover, if g / K, then this decomposition
is unique.
Proof. If g = K(
0
) K, then we have the representation:
g = K(
0
)A(e
0
)K(0).
Suppose now that g is not in K. Let g :=
_
a b
c d
_
G = PSL
2
(R). Suppose rst that
g is not a symmetric matrix. Dene:

1
:= cot
1
_
a + d
c b
_
.
Then
cos(
1
)
sin(
1
)
= cot(
1
) =
a + d
c b
and by rearrangement:
b cos(
1
) + d sin(
1
) = a sin(
1
) + c cos(
1
).
Hence:
K(
1
)g =
_
b cos(
1
) + d sin(
1
)
a sin(
1
) + c cos(
1
)
_
is symmetric.
Since we have that either g = K(0)g is symmetric or K(
1
)g is symmetric, the
spectral theorem for symmetric matrices asserts that we can conjugate by orthogonal
matrices in order to obtain a diagonal matrix. Since the orthogonal matrices in PSL
2
(R)
are the elements of K and the diagonal matrices are the elements of A, we can conclude
that there exist
0
,
0
[0, ) and r R such that
K(
0
)
1
gK(
0
)
1
= A(e
r/2
).
Using the identity:
K(/2)A(a)K(/2) = A(a
1
)
we can ensure that r > 0. By rearrangement, this yields the desired result.
From the uniqueness of the Cartan Decomposition for g / K, and the fact that points
of 1 correspond bijectively to cosets gK G/K, we obtain an alternate coordinate
system for 1, called geodesic coordinates.
Denition 1.4.9 (Geodesic Polar Coordinates). If z 1 is such that z = i for some
nontrivial PSL
2
(R), and has Cartan Decomposition:
= K(
0
)A(e
r/2
)K(
0
),
we say that z has geodesic polar coordinates (r,
0
).
16 1. GEOMETRY OF THE POINCAR HALF-PLANE
The geodesic polar coordinates dened above will be convenient when analysing the
Green Function in Section 2.4 below.
Proposition 1.4.10. If z 1has geodesic polar coordinates (r, ), then (z, i) = r, and
the angle made between the geodesic joining z and i with the segment of the imaginary
axis between 0 and i is .
The coordinate system introduced in Proposition 1.4.10 above is analogous to the
polar coordinates on R
2
.
Having established the basic properties of the Poincar Half-Plane and the action of
PSL
2
(R) we now turn our attention towards harmonic analysis on 1.
CHAPTER 2
Harmonic Analysis on 1
In this chapter we outline harmonic analysis on 1, which roughly mirrors the theory
of classical Fourier series and the Fourier transform on R
N
. In this setting, trigono-
metric polynomials are replaced by the Whittaker functions as the functions of central
importance. We shall rst in Section 2.1 dene invariant linear operators and introduce
the important LaplaceBeltrami operator. We shall then nd expressions for the eigen-
functions of the LaplaceBeltrami operator, and state without proof the major results of
harmonic analysis on 1. Notably we will state analogues to the Fourier Inversion Theo-
rem and the Convergence Theorem for the classical Fourier series of periodic functions
in Section 2.3. The theory of Fourier series in particular will be an indispensable tool
in our study of automorphic functions in Chapter 4. We will then conclude the chapter
with a study of the Green Function on 1 in Section 2.4. The results from this section
will be important for our study of the automorphic Green Function in Section 4.4. Our
treatment will be similar to that in [6] and [7].
2.1. Invariant Operators
It will be important later to consider PSL
2
(R)invariant functions, so in this section
we develop the related notion of an invariant operator on the space of functions f : 1
C.
Let G = PSL
2
(R) act isometrically on 1, as in Chapter 1.
Denition 2.1.1. For all functions f : 1C, dene the operator T
g
as follows:
(T
g
f)(z) = f(gz) for all g G, z 1.
Denition 2.1.2. We dene an invariant linear operator T to be a linear operator on
the space of all functions f : 1C, that commutes with T
g
for every g G. That is
for all functions f : 1C, and for all g G, we have:
(T(T
g
f))(z) = (Tf)(gz)
for all z 1.
Important for the spectral theory of linear operators is the resolvent.
Denition 2.1.3. We dene the resolvent R
s
of a linear operator T to be the function
R
s
() = (I T)
1
= (s(1 s)I T)
1
for C, with = s(1 s) and s C.
The analytic structure of the resolvent encodes the spectral theory of the correspond-
ing operator. We will encounter the resolvent again in Chapter 4.
17
18 2. HARMONIC ANALYSIS ON H
In practice we shall often restrict our attention to subspaces of the space of complex-
valued functions on 1by imposing conditions on smoothness and growth. We shall now
dene the analogue of the Laplacian =

2
x
2
+

2
y
2
on R
2
.
Remark 2.1.4. We work in rectangular coordinates here, so by a slight abuse of notation,
for functions f : 1 C we dene the following:
f
x
:=


f
x
where

f : R
2
C is given by:

f(x, y) = f(x + iy).


Denition 2.1.5 (LaplaceBeltrami Operator). We dene the Laplace-Beltrami opera-
tor on 1 to be:
:= y
2
_

2
x
2
+

2
y
2
_
.
Remark 2.1.6. On every complete Riemannian manifold, a
LaplaceBeltrami operator can be dened using the metric tensor. See Buser [4, p.184].
The LaplaceBeltrami operator on 1 can also be written in terms of partial complex
derivatives:
= (z z)
2

z

z
. (2.1.7)
Proposition 2.1.8. The Laplace-Beltrami operator is an invariant differential operator.
Proof. Let g =
_
a b
c d
_
PSL
2
(R). Observe rst that the map on 1 given by z gz
is a rational function with no poles in 1, and is hence holomorphic. If we decompose
the map g into its real and imaginary components as g(z) = u(z) + iv(z), we have by
the CauchyRiemann equations:
u
x
=
v
y
and
u
y
=
v
x
. (2.1.9)
In particular, this implies that:

2
u
x
2
+

2
u
y
2
=

2
v
x
2
+

2
v
y
2
= 0 (2.1.10)
by the equality of second-order mixed partial derivatives. Now let f : 1C be a func-
tion that has second-order partial derivatives and continuous rst-order partial deriva-
tives. We use the chain rule together with Equation (2.1.9) above to obtain an expression
for (f g). First we have by the chain rule
f
x
=
f
u
u
x
+
f
v
v
x
.
2.1. INVARIANT OPERATORS 19
Thus

2
f
x
2
=
_

2
f
u
2
u
x
+

2
f
uv
_
u
x
+
_

2
f
vu
u
x
+

2
f
v
2
v
x
_
v
x
+
f
u

2
u
x
2
+
f
v

2
v
x
2
.
and similarly

2
f
y
2
=
_

2
f
u
2
u
y
+

2
f
uv
_
u
y
+
_

2
f
vu
u
y
+

2
f
v
2
v
y
_
v
y
+
f
u

2
u
y
2
+
f
v

2
v
y
2
.
Since f(u + iv) has continuous rst-order partial derivatives, its mixed second-order
partial derivatives are equal. Hence by using Equations (2.1.9) and (2.1.10) to cancel
terms we have:

2
f
x
2
+

2
f
y
2
=
_

2
f
u
2
+

2
f
v
2
_
_
_
v
x
_
2
+
_
v
y
_
2
_
.
Computing the partial derivatives of v directly from Equation (1.2.6) yields:
v
x
=
2cy(cx + d)
[cz + d[
4
and
v
y
=
[cz + d[
2
2c
2
y
2
[cz + d[
4
.
Hence:
_
v
x
_
2
+
_
v
y
_
2
=
([cz + d[
4
+ 4c
2
y
2
((cx + d)
2
+ c
2
y
2
[cz + d[
2
))
[cz + d[
8
= [cz + d[
4
.
Hence, again using Equation (1.2.6) and the denition of the LaplaceBeltrami operator,
we can conclude that:
(T
g
)(f) = y
2
[cz + d[
4
_

2
f
u
2
+

2
f
v
2
_
= v
2
_

2
f
u
2
+

2
f
v
2
_
= (T
g
)(f)
as required.
Part of the importance of the LaplaceBeltrami operator stems from the following
results.
Proposition 2.1.11. Every invariant differential operator can be written as a polynomial
in the LaplaceBeltrami operator with constant complex coefcients.
Proof. See [6, p.288].
Corollary 2.1.12. If the function f : 1Cis an eigenfunction of the LaplaceBeltrami
operator, then f is an eigenfunction of every invariant differential operator.
Having established their importance, in the next section we compute several eigen-
functions of the LaplaceBeltrami operator.
20 2. HARMONIC ANALYSIS ON H
2.2. Eigenfunctions of
For now we shall work exclusively in the rectangular coordinates
z = x+iy. Foreshadowing our generalisation of periodic functions in R
N
to automorphic
functions in 1 in Chapter 4, we search for eigenfunctions of which possess desirable
symmetries. In particular, we shall restrict our attention to the functions f(x+iy) which
are either constant or of the form f(x + iy) = F(2y)e
2ix
. This method is called
separation of variables.
We will make frequent use of the function e
2iz
in the remainder of this essay, so we
introduce some convenient notation here.
Denition 2.2.1. Dene the function e(z) as follows:
e(z) := e
2iz
.
We also make precise the notion of asymptotic equivalence, which we use to describe
the behaviour of functions for large imaginary part.
Denition 2.2.2. We say that two functions f, g : 1C are asymptotically equiva-
lent if:
lim
y
[f(x + iy)[
[g(x + iy)[
= 1
If f is asymptotically equivalent to g, we write: f g.
The eigenfunctions f : 1C of the LaplaceBeltrami operator by denition are
the non-zero solutions to the following equation:
( + )f = 0 for some C. (2.2.3)
Proposition 2.2.4. Every solution f : 1Cof (2.2.3) with ,= 1/4, which is constant
in x, is a linear combination of:
y
s
and y
1s
(2.2.5)
where s(1 s) = . In the special case of = 1/4, f is a linear combination of:
y
1/2
and y
1/2
log(y). (2.2.6)
Proof. This is a CauchyEuler equation of order 2.
In order to state the solutions of Equation (2.2.3) which are of the form f(x + iy) =
F(2y)e(x), we will need to dene the Bessel functions and state their basic properties.
We shall do this now.
Consider the modied Bessels equation:
x
2

y
x
2
+ x
y
x
(x
2
+
2
)y = 0 where C. (2.2.7)
This equation has two linearly independent solutions I

and K

, known as modied
Bessel functions of the rst and second kind respectively. A more detailed account of
Bessel functions can be found in [2, p.222223]. For our analysis, the following proper-
ties of the modied Bessel functions are sufcient.
2.2. EIGENFUNCTIONS OF 21
Proposition 2.2.8. The general solution to Equation (2.2.7) is given by linear combina-
tions of the two functions:
I

(x) :=
_
x
2
_

n=0
(
x
2
)
2n
( + n + 1)n!
,
where is the Gamma Function and
K

(x) :=

2 sin()
(I

(x) I

(x)).
Moreover, as x , we have the asymptotic equalities:
I

(x) (2x)
1/2
e
x
and
K

(x)
_

2y
_
1/2
e
x
.
Proposition 2.2.9. Every solution f : 1Cof (2.2.3) which is of the formf(x+iy) =
e(x)F(2y) is a linear combination of:
(2
1
y)
1/2
K
s1/2
(y)e(x) and (2y)
1/2
I
s1/2
(y)e(x). (2.2.10)
Proof. This result follows from the denition of the Bessel functions and the direct com-
putation of partial derivatives.
Since the two linearly independent solutions to (2.2.3) found in Proposition 2.2.9
above have very different asymptotic behaviour, we can express solutions to (2.2.3)
which are of the form f(x + iy) = e(x)F(2y) and satisfy suitable growth conditions
entirely in terms of the left hand solution in 2.2.10. This motivates the following deni-
tion.
Denition 2.2.11. The Whittaker function is given by:
W
s
(z) := 2y
1/2
K
s1/2
(2y)e(x)
in the upper half-plane. Dene:
W
s
(z) := W
s
( z)
for the lower half-plane.
We now introduce further growth constraints into our study of harmonic analysis.
Denition 2.2.12 (Little-O Notation). If f, g : 1C are functions such that:
lim
y
[f(x + iy)[
[g(x + iy)[
= 0
for all x R, we write f(z) = o(g(z)).
22 2. HARMONIC ANALYSIS ON H
Denition 2.2.13 (Big-O Notation). If M > 0 and f, g : 1Care functions such that:
[f(x + iy)[
[g(x + iy)[
M
for sufciently large y > 0 and all x R, we write f(z) = O(g(z)). We call the constant
M an implied constant in our usage of such notation.
Remark 2.2.14. Note that f(z) = o(g(z)) is a stronger condition than f(z) = O(g(z)).
From Propositions 2.2.8 and 2.2.9, we have the following result.
Corollary 2.2.15. Every solution f : 1 C of (2.2.3) which is of the form f(z) =
e(x)F(2y) and obeys the growth condition:
f(z) = o(e
2y
) (2.2.16)
is a constant multiple of the Whittaker function W
s
(z), where s(1 s) = .
2.3. Key Theorems
We shall now state the analogues on 1of two of the key theorems of classical Fourier
analysis.
Consider the space of all smooth functions on 1that decay to zero towards its boundary:
Denition 2.3.1. Dene the set (

0
(1) to be the collection of all functions f : 1C
such that:
f has continuous derivatives of all orders.
lim
|z|
f(z) = 0.
lim
Imz0
f(z) = 0.
Theorem 2.3.2 (Analogue to Fourier Inversion Theorem). Let f (

0
(1).
Let f
s
(r) :=
_
H
f(z)W
s
(rz) dz. Then:
f(z) =
1
2i
_
C
_
R
W
s
(rz)f
s
(r)
s
(r) dr ds
where C is the line in C given by Re(z) = 1/2.
Proof. [7, p.30]
Theorem 2.3.3 (Analogue to Fourier Series Expansion). Let f(z) be an eigenfunction of
with eigenvalue = s(1 s) which satises:
f(z + 1) = f(z) for all z 1
and:
f(z) = o(e
2y
).
Then we have the following series expansion:
f(z) = f
0
(y) +

nZ\{0}
a
n
W
s
(nz) (2.3.4)
2.4. THE GREEN FUNCTION 23
where the zero-th term f
0
(y) is the linear combination of functions from Proposition
2.2.4. This series converges absolutely on 1 and uniformly on compact subsets of 1.
Denition 2.3.5. We say that the series in Equation (2.3.4) is the Fourier expansion for
f.
Corollary 2.3.6. In the Fourier expansion for f : 1C, we have the following asymp-
totic bound on coefcients for all > 0:
a
n
= O(e
|n|
). (2.3.7)
where the implied constant depends only on our choice of and f.
2.4. The Green Function
In this section we will discuss the important Green Function on 1. To do this, we
shall need to study the invariant integral operators. We proceed using the coordinate
system introduced in Denition 1.4.9, the geodesic polar coordinates.
We rst note:
Proposition 2.4.1. The LaplaceBeltrami operator on 1 is given in geodesic polar co-
ordinates by:
=

2
r
2
+
1
tanh(r)

r
+
1
(2 sinh r)
2

2
.
If we replace the coordinate r by the coordinate u := (cosh(r) 1)/2 (see Equation
(1.2.25)), we obtain:
= u(u + 1)

2
u
2
+ (2u + 1)

u
+
1
16u(u + 1)

2
. (2.4.2)
We now repeat the analysis of Section 2.2 in geodesic polar coordinates.
Proposition 2.4.3. Every solution to Equation (2.2.3) that is of the form: f(z) = F(u)e
2im
in geodesic polar coordinates is a linear combination of the two functions:
F
s
(u)e
2im
and G
s
(u)e
2im
where:
F
s
(u) = F(s, 1 s; 1, u)
and
G
s
(u) =
(s)
2
4(2s)
u
s
F(s, s; 2s, u
1
) =
1
4
_
1
0
((1 ))
s1
( + u)
s
d
in terms of the hypergeometric functions F dened in [1, p.315]. We call the function G
s
the Green Function.
We now establish an important asymptotic bound for the function G
s
.
24 2. HARMONIC ANALYSIS ON H
Lemma 2.4.4. Suppose s = + it with > 0 and x ranges over R
+
. Then we have:
G
s
(x) = O(x

).
Proof. For > 0, the integral formula for G
s
converges absolutely and we have:
G
s
(u)
[u
s
[
4
_
1
0
((1 ))
s1
d =
u

4
_
1
0
((1 ))
s1
d = O(x

).

The most important property of the Green Function is connected to the theory of
invariant integral operators, which we now introduce.
2.5. Invariant Integral Operators
Using the linearity of the Lebesgue integral, we can dene a class of linear operators
on spaces of complex-valued functions of 1.
Denition 2.5.1. Given a space of complex-valued functions f : 1C, we dene the
integral operator L by:
(Lf)(z) :=
_
H
k(z, w)f(w) d(w)
where k : 1 1 C is such that absolute convergence is ensured. The function
k : 11C is called the kernel of the integral operator L.
Denition 2.5.2. The kernel k of an integral operator is said to be point-pair invariant
if:
k(gz, gw) = k(z, w) for all g G, z, w 1.
Proposition 2.5.3. An integral operator is invariant if and only if its kernel is point-pair
invariant.
In light of Theorem 1.2.14 and Equation 1.2.25, we will usually write point-pair
invariant kernels as k(u(z, w)), in terms of
u(z, w) :=
[z w[
2
4 Im(z) Im(w)
.
The integral operators will be important in our analysis of the spectral theory of
automorphic forms in Chapter 4. The main result is that the resolvent R
s
of is an
invariant integral operator with the Green function as its kernel. Explicitly:
(R
s
f)(z) =
_
H
G
s
(u(z, w))f(w) d(w).
The proof of this fact is not easy. A Lie algebraic approach can be found in [7,
p.3538].
Our next chapter will provide an introduction to the theory of Fuchsian groups.
CHAPTER 3
Fuchsian Groups
In this chapter, we study the Fuchsian groups, which as we explain in Section 3.1 are
roughly analogous to the lattices in the complex plane.
In Section 3.2 we will collect some relevant results about topological groups, and in
particular about PSL
2
(R). Most importantly, we will establish a characterisation of dis-
crete subgroups of PSL
2
(R) in terms of their action on 1, in Theorem 3.2.12. In Section
3.3 we will dene and provide several examples of Fuchsian groups. We will then in
Section 3.4 dene a fundamental domain for a Fuchsian group, and provide several con-
structions thereof. Our nal section will be more arithmetic in nature. We will construct
a special exponential sum over double-cosets of a Fuchsian group, and establish some
basic bounds thereof. This Kloosterman sum will prove to be a key tool in the spectral
theory that is to ensue in Chapter 4. Our work on Fuchsian groups themselves is loosely
based on that of Katok in [8, Chapter. 2], and the section on Kloosterman sums is based
on the treatment of Iwaniec in [7, Chapter. 2].
First, we shall provide some motivation for the Fuchsian groups using our well-
established intuition in C.
3.1. Motivation
Having dened a group action of PSL
2
(R) upon 1, we shall now consider the in-
duced action of certain subgroups PSL
2
(R) on 1. Of particular interest are the
complex functions which are invariant under the action of on 1.
Recall the doubly periodic functions on C.
Denition 3.1.1. A function f : CC is doubly-periodic with periods
1
,
2
C if
f(z + mw
1
+ nw
2
) = f(z) for all z C, m, n Z.
An equivalent formulation of the Denition 3.1.1 is the statement that f is invariant
with respect to an action of Z
2
on C. Also note that Z
2
can be regarded as an additive
subgroup of C that is in some sense spread out.
We will now consider the analogous class of functions on 1.
Denition 3.1.2. Suppose that PSL
2
(R). We say that f : 1C is an automor-
phic function with respect to , if
f(z) = f(z) for all .
Much more will be said about automorphic functions in Chapter 4.
25
26 3. FUCHSIAN GROUPS
For now it is natural to ask the following question: For which subgroups of
PSL
2
(R) is the theory of automorphic functions sufciently interesting?
Of course if we take = PSL
2
(R) itself, then the space of automorphic functions
consists only of constant functions, since PSL
2
(R) acts transitively on 1. At the other
extreme, if we take = I, the trivial subgroup, then the space of automorphic
functions is the space of complex-valued functions on 1 with no restrictions whatso-
ever! The subgroups of PSL
2
(R) that we shall study for the remainder of this essay are
the Fuchsian groups, whose automorphic functions possess a rich theory. In order to
dene the Fuchsian groups, we shall rst need to topologise PSL
2
(R).
3.2. PSL
2
(R) as a topological group
In this section we prove some results on topological groups that will be important for
our analysis of Fuchsian groups.
Denition 3.2.1 (Topological Group). A group G endowed with a topology T(G)
is said to be a topological group if:
The group multiplication map (g, h) gh is continuous with respect to the
product topology on GG.
The group inversion map g g
1
is continuous with respect to .
As a rst example, the additive group (R
n
, +) is a topological group with respect to
the standard Euclidean topology. Since we can view SL
2
(R) as being embedded in R
4
via the injective map:
i :
_
a b
c d
_
(a, b, c, d)
we can topologise SL
2
(R) using the subspace topology inherited from the standard Eu-
clidean topology on R
4
. Explicitly, we declare U SL
2
(R) to be open in SL
2
(R) if and
only if i(U) is open in R
4
.
Having topologised SL
2
(R) as above, there is a natural way to topologise PSL
2
(R) =
SL
2
(R)/I as a quotient space. Explicitly, we declare U PSL
2
R to be open if and
only if the union of all cosets in U is an open subset of SL
2
(R).
Proposition 3.2.2. Endowed with the topology above, the group PSL
2
(R) is a topologi-
cal group.
Proof. The group SL
2
(R) is a topological group because multiplication and inversion
of matrices are rational functions in the matrix entries and hence continuous with respect
to the chosen topology. The claim follows from the general result that closed subgroups
of topological groups are topological groups, and quotients by closed subgroups are
topological groups.
Henceforth we shall always assume PSL
2
(R) to be equipped with this topology.
Denition 3.2.3. A subset Y of a topological space (X, ) is said to be discrete if
every subset of Y is open with respect to the subspace topology on Y . In particular, a
3.2. PSL2(R) AS A TOPOLOGICAL GROUP 27
subgroup of a topological group is discrete if every subset thereof is open with respect to
the subspace topology.
For example, the subgroup (Z
n
, +) < (R
n
, +) is discrete with respect to the standard
Euclidean topology. The following lemma is immediate from the denition of conver-
gence.
Lemma 3.2.4. If X is a discrete topological space and the sequence (x
n
) in X converges
to x X, then x
n
= x for sufciently large n.
The primary reason that we will work with discrete subgroups of PSL
2
(R) is that
they are precisely the subgroups of PSL
2
(R) that act discontinuously on 1, in a sense
we will now make precise.
Denition 3.2.5. A collection of subsets M

J
of a topological space X is said to
be locally nite if for every point x X, there exists a neighbourhood N of x such that
N M

,= for only nitely many J.


Denition 3.2.6. A group G acting on the topological space X is said to act discontin-
uously if every orbit of G is locally nite in X. In the case that X is a locally compact
metrisable space, this is equivalent to the statement that for every x X and every
compact set K X, the set:
g G : gx K
is nite.
Proposition 3.2.7. Suppose a group G acts discontinuously on a locally compact topo-
logical space X. Then for each x X, the orbit:
Gx = gx : g G
has no limit points in 1.
Proof. Suppose that the orbit Gx has a limit point y 1. Then by the local compactness
of 1, there exists a compact neighbourhood of y containing innitely many distinct
points of Gx. Hence there are innitely many g G such that gx K, contradicting
our assumption that G acts discontinuously on 1.
Proposition 3.2.8. Suppose a group G acts discontinuously on a topological space X.
Then for every x X, the stabiliser:
G
x
:= G : x = x
is nite.
Proof. Suppose that the stabiliser G
x
was not nite for some x 1. In particular, this
implies that there exist innitely many g Gsuch that gx = x K an arbitrary compact
neighbourhood of x. This again contradicts our assumption that G acts discontinuously
on 1.
28 3. FUCHSIAN GROUPS
We now show that the properties dened in Denitions 3.2.3 and 3.2.6 are equivalent
in our setting. We shall need three prepatory lemmas in order to prove our main result,
Theorem 3.2.12.
Lemma 3.2.9. Suppose z
0
1 is xed and K 1 is compact. Then the set:
S := g PSL
2
(R) : gz
0
K
is compact.
Proof. Recall that PSL
2
(R) is topologised as a quotient of SL
2
(R). Hence the natural
projection map : SL
2
(R) PSL
2
(R) is continuous. Since the continuous image of a
compact set is compact, it sufces to show that the set:
S :=
__
a b
c d
_
SL
2
(R) :
az
0
+ b
cz
0
+ d
K
_
is compact as a subset of SL
2
(R). By the HeineBorel theorem in R
4
, we can establish
this by showing that S is closed and bounded when regarded as a subset of R
4
. Now
since the fractional linear transformations are continuous in their coefcients and S is
the pre-image of a closed set K, S is certainly closed. Since K is compact in 1, it is
also compact as a subset of C and hence bounded. So there exists a positive constant M
such that:

az
0
+ b
cz
0
+ d

M
for all
_
a b
c d
_
S. Furthermore, by compactness, there exists an element g of S which
minimises Imgz
0
, so there exists a positive constant L such that:
Im
_
az
0
+ b
cz
0
+ d
_
L
for all
_
a b
c d
_
S. Using Equation (1.2.6) and rearrangement, these two bounds give
us:
[cz
0
+ d[
_
Im(z
0
)
L
_
1/2
and
[az
0
+ b[ M
_
Im(z
0
)
L
_
1/2
.
Hence the coefcients a, b, c and d are bounded. This completes the proof.
The proof of the following result is from [13] due to Mariano Surez-Alvarez).
Proposition 3.2.10. Suppose G is a Hausdorff topological group. Then any discrete
subgroup H of G is closed.
3.2. PSL2(R) AS A TOPOLOGICAL GROUP 29
Proof. For each x H, the set x is closed in G. We rst prove that the collection of
all elements of H is locally nite, and then show that this implies H is closed. Suppose
for the sake of contradiction that there exists y G such that every neighbourhood
of y contains innitely many elements of H. By the denition of discreteness, there
exists an open subset U G such that U H = 1. Now by the continuity of group
multiplication and inversion in a topological group, we can nd an open set V G
such that 1 V and V
1
V U. Hence yV is an open neighbourhood of y which by
assumption must contain two distinct elements h
1
,= h
2
of H. Now:
h
1
1
h
2
= (y
1
h
1
)
1
(y
1
h
2
) V
1
V U.
So h
1
h
2
U H = 1, which is absurd. Hence no such y exists and the collection of
elements of H is locally nite in G. Now let z be an element that is not in H. From the
property of local niteness, we may choose a neighbourhood N of z that only contains
nitely many h H. Hence by taking the nite intersection of N with the complements
in G of each of these elements h H, we obtain a neighbourhood N

of z such that
N

H = . Since z is arbitrary, we can conclude that H is closed.


Lemma 3.2.11. Suppose that PSL
2
(R) acts discontinuously on 1, and z
0
1 is
xed by a nontrivial element of . Then there exists a neighbourhood N of z
0
such that
no other point in N is xed by any nontrivial elements of PSL
2
(R).
Proof. Suppose for the sake of contradiction that every neighbourhood of z
0
contains a
point other than z
0
which is xed by a nontrivial element of . In particular, this implies
that there exists a sequence (z
n
) in 1 and a sequence (
n
) in PSL
2
(R) 1 such that
z
n
z
0
and
n
z
n
= z
n
. Now since acts discontinuously, the set: : z
0

B(z
0
, 3) is nite, where > 0 is arbitrary, and B(z, r) denotes the hyperbolic ball of
radius r centred at z. Hence for sufciently large n:
(
n
z
0
, z
0
) > 3.
However, from the triangle inequality and the fact that acts by isometries, we have:
(
n
z
0
, z
0
) (
n
z
0
,
n
z
n
) + (
n
z
n
, z
0
) = (z
0
, z
n
) + (z
n
, z
0
) < 2
for sufciently large n. This yields the desired contradiction.
We now have the following theorem due to Poincar.
Theorem 3.2.12. A subgroup of PSL
2
(R) is discrete if and only if it acts discontinu-
ously on 1.
Proof. We rst show that a discrete subgroup of PSL
2
(R) acts discontinuously on 1.
Let K 1 be compact and let z 1. From Proposition 3.2.10, is closed as a subset
of PSL
2
(R). Moreover, by Lemma 3.2.9, the set g PSL
2
(R) : gz K is compact.
Hence the set:
S := g : gz K = g PSL
2
(R) : gz K
is both discrete and compact. Since S is both discrete and compact, we can conclude
that S is nite, whence acts discontinuously on 1.
30 3. FUCHSIAN GROUPS
Conversely, suppose that acts discontinuously on 1 but is not discrete as a subgroup
of PSL
2
(R). Let z
0
1 be a point which is not xed by any nontrivial element of
PSL
2
(R). (The existence of such points is guaranteed by Lemma 3.2.11.) Since is
not discrete, there exists a sequence of distinct elements (
n
) in with
n
1. By
continuity, this gives us
n
(s) s and we can conclude the intersection s K is not
nite, for any neighbourhood of s. This contradicts the denition of a discontinuous
action, hence our assumption that was not discrete is absurd. This completes the
proof.
We are now ready to dene the Fuchsian groups, which will be central to the remain-
der of this essay.
3.3. Fuchsian Groups
In this section we dene the Fuchsian groups and establish their basic properties.
Denition 3.3.1 (Fuchsian Group). A subgroup PSL
2
(R) is said to be a Fuchsian
group if is a discrete subset of PSL
2
(R).
Note that Theorem 3.2.12 provides us with the equivalent and useful characterisation
of Fuchsian groups as subgroups of PSL
2
(R) that act discontinuously on 1.
We now provide several examples of Fuchsian groups. Recall the classication of
elements of PSL
2
(R) as hyperbolic, parabolic or elliptic in Propositions 1.3.3, 1.3.4 and
1.3.5. Correspondingly we can form cyclic Fuchsian groups.
Proposition 3.3.2. We have the following three examples of Fuchsian groups:
Every cyclic subgroup of PSL
2
(R) consisting only of hyperbolic elements is a
Fuchsian group.
Every cyclic subgroup of PSL
2
(R) consisting only of parabolic elements is a
Fuchsian group.
Every nite cyclic subgroup of PSL
2
(R) consisting only of elliptic elements is a
Fuchsian group.
Proof. Since the map
1
is a homeomorphism from PSL
2
(R) to itself for
xed PSL
2
(R), any subgroup of PSL
2
(R) that is conjugate to a discrete subgroup
of PSL
2
(R) is also discrete. Using this fact and Propositions 1.3.3,1.3.4, and 1.3.5, it
sufces to prove the assertion for the three cyclic groups:

H
:=
__

n
0
0
0
n
0
_
: n Z,
0
> 0 xed
_
,

P
:=
__
1 nt
0
0 1
_
: n Z, t
0
> 0 xed
_
,
and
3.3. FUCHSIAN GROUPS 31

E
:=
__
cos(k/n) sin(k/n)
sin(k/n) cos(k/n)
_
: k Z, n N xed
_
.
To show that
H
is discrete, it sufces to observe that taking the norm on R
4
,
(
2n
0
+
2n
0
)
1/2
blows up for [n[ large. Hence only nitely many representatives for
elements of
H
lie in any given ball in R
4
about the origin. Similarly, examining the
norm (2 + n
2
t
2
0
)
1/2
shows that
P
is discrete. Finally, observe that the projection map
: SO
2
(R) K has closed kernel I. Hence K is Hausdorff, and any nite sub-
set thereof is discrete. This shows that
E
is discrete and completes the proof of the
proposition.
Remark 3.3.3. Note that innite cyclic groups consisting only of elliptic elements are
not discrete. This is an easy consequence of Proposition 3.2.8 below.
A more important example for us is the modular group.
Proposition 3.3.4 (Modular Group). The modular group:
PSL
2
(Z) :=
__
a b
c d
_
I PSL
2
(R) : a, b, c, d Z
_
is a Fuchsian group.
Proof. The discreteness of PSL
2
(Z) in PSL
2
(R) follows from the discreteness of Z
4
in
R
4
.
Of particular importance to number theory are the congruence subgroups.
Proposition 3.3.5 (Principal Congruence Subgroup). Let N be a positive integer. The
natural projection homomorphism:

N
: SL
2
(Z)SL
2
(Z/NZ)
induces a surjective homomorphism:
: PSL
2
(Z)PSL
2
(Z/NZ).
The principal congruence subgroup of level N:
(N) := ker(
N
)
is a Fuchsian group.
Proof. See [8, p.133-134].
Proposition 3.3.6 (Congruence Subgroup). Every such that
(N) PSL
2
(Z)
is a Fuchsian group, said to be a congruence subgroup of level N.
Proposition 3.3.7. Let P be a hyperbolic triangle with angles /2, /3 and /7). Let
be the subgroup of Isom(1) generated by reections in the sides of P. The subgroup
:= PSL
2
(R) of orientation-preserving elements of forms a Fuchsian group
known as the (2, 3, 7)-triangle group.
32 3. FUCHSIAN GROUPS
We now prove some of the basic results about Fuchsian groups.
Proposition 3.3.8. Suppose that is a Fuchsian group. Then for all z 1, the stabiliser

z
is nite and and cyclic.
Proof. The stabiliser
z
is nite by Proposition 3.2.8. By Lemma 1.4.1, we can nd
PSL
2
(R) such that i = z. Hence we have:

z
=
i
= K.
Now K is certainly a discrete subgroup of K. Since the conjugate sugbroup of
a cyclic subgroup is cyclic, it now sufces to prove that any discrete subgroup H of
K is cyclic. By discreteness, there exists a nontrivial element = K(
0
) H with
0
minimal amongst all elements of H. Now let

= K(
1
) H be an arbitrary element of
H. We have:
1
= m
0
+ for some unique non-negative integer mand 0 <
0
. But
K() = K(
1
)K(
0
)
m
H. Hence by minimality of
0
, we are forced to conclude that
= 0. This shows that every element of H lies in the cyclic subgroup of K generated
by K(
0
) and completes the proof.
Remark 3.3.9. From Proposition 1.2.30, the action of a Fuchsian group extends to

1,
not necessarily discontinuously.
We have the following result, the proof of which runs along similar lines to Proposi-
tion 3.3.8.
Proposition 3.3.10. Suppose that is a Fuchsian group. Then for all
s 1 = R , the stabiliser
s
is cyclic.
Remark 3.3.11. For s 1, the stabiliser
s
is not necessarily nite. For example, the
stabiliser

of the point in the Fuchsian group = PSL


2
(Z) is the innite cyclic
subgroup of PSL
2
(Z) generated by N(1).
In order to obtain more structure than is possible for arbitrary Fuchsian groups, we
now dene a special kind of Fuchsian group. These are the Fuchsian groups that are
large enough for us to have a reasonable chance of obtaining useful information about
the functions which are automorphic with respect to .
Denition 3.3.12. A Fuchsian group is said to be a Fuchsian group of the rst kind if
every point s 1 is is a limit point of some orbit z, where z 1.
An example of a Fuchsian group that is not of the rst kind is the cyclic group of
parabolic elements from Proposition 3.3.2.
We now introduce the extremely important notion of a cusp of a Fuchsian group. We
shall provide a geometric motivation for the usage of the term cusp later in Section 3.4.
Denition 3.3.13 (Cusp). If s 1is a xed point of some parabolic transformation in
, we say that the orbit s is a cusp of .
For example, the point s = is xed by the parabolic transformation N(1) in
PSL
2
(Z) and hence is a representative of a cusp thereof. We will often blur the distinc-
tion between cusps and representatives thereof.
3.4. FUNDAMENTAL DOMAINS 33
Restricting our attention to the Fuchsian groups of the rst kind for now, we shall
now further divide Fuchsian groups according to whether or not they possess cusps.
For this, we establish another characterisation of Fuchsian groups with cusps in terms
of the topological quotient space 1.
Proposition 3.3.14. If PSL
2
(R), then the relation on 1 given by:
z w w = z for some
is an equivalence relation on 1. If z w, we say that z and w are equivalent or that
z and w are congruent modulo .
In light of Proposition 3.3.14 above, we may consider the quotient topological space
1.
Denition 3.3.15. We say that a Fuchsian group is co-compact if the topological
space 1 is compact.
Proposition 3.3.16. A Fuchsian group is co-compact if and only if it does not have any
cusps.
Proof. [8, p.8490]
We now briey refer back to our motivational example of the doubly periodic func-
tions on C. The analysis of a doubly periodic function on C is often described using its
period parallelogram which tessellates C.
Such a tessellation is possible because for xed w
1
, w
2
C, the group of translations:
z z + mw
1
+ nw
2
: m, n Z
acts discontinuously on C. We will now dene an analogous construction for Fuchsian
groups, the fundamental domain.
3.4. Fundamental Domains
A convenient way of analysing the action of a particular Fuchsian group on 1 is
through the use of a fundamental domain.
Denition 3.4.1 (Fundamental Domain). Afundamental domain for the Fuchsian group
is an open connected subset F of 1 such that:
For all nontrivial and all z F, z / F.
For all z 1, the intersection z F is nonempty.
Since every -orbit in 1contains a point in F, and the points in the interior of F are
pairwise inequivalent with respect to the relation dened in Proposition 3.3.14, we can
view the family of sets:
(F) :
as a tessellation of 1.
34 3. FUCHSIAN GROUPS
Example 3.4.2. The cyclic Fuchsian group:
:= N(m) : m Z =
__
1 m
0 1
_
: m Z
_
has the strip
P := z 1 : 0 < Re(z) < 1
as a fundamental domain.
Remark 3.4.3. There certainly exists a fundamental domain for an arbitrary Fuchsian
group , as we shall construct in Proposition 3.4.6. The fundamental domain for a Fuch-
sian group is by no means unique though, even if we identify fundamental domains that
are equivalent modulo . For example we can construct a different fundamental domain
for the Fuchsian group in Example 3.4.2 by simply making a small indent in both sides
of the strip.
We shall now construct a fundamental domain for an arbitrary Fuchsian group in
two different ways.
Denition 3.4.4. Suppose z
1
, z
2
1 with z
1
,= z
2
. We dene the set:
z 1 : (z, z
1
) = (z, z
2
)
to be the perpendicular bisector of the geodesic segment joining z
1
to z
2
.
Proposition 3.4.5. The perpendicular bisector of a geodesic segment L is itself a geo-
desic segment that meets L orthogonally.
Proof. See [8, p.5354].
In particular, if and p 1, we denote the perpendicular bisector of the
geodesic segment joining p to p by L
p
(). The geodesic segment L
p
() divides the
hyperbolic plane into two open half-planes. We denote the half-plane containing p by
H
p
().
Proposition 3.4.6 (Dirichlet Normal Polygon). Let be a Fuchsian group. If p 1 is
not xed by all nontrivial , then the set:
D(p) := z 1 : (z, p) < ((z), p) for all nontrivial
is a fundamental domain for bounded by geodesics in 1.
Proof. The set D(p) can be characterised as:
D(p) =

\{1}
H
p
()
so is certainly bounded by geodesics in 1. Now let z 1 be arbitrary. Since the orbit
z does not accumulate in 1 by Proposition 3.2.7, there must exist at least one point in
1 of minimal distance from p. That is, there exists z
0
z such that:
(z
0
, p) (z
0
, p) for all .
3.4. FUNDAMENTAL DOMAINS 35
Hence z
0
D(p) and D(p) contains a point from each orbit. It remains to show
that the points in D(p) are noncongruent modulo . Suppose z
1
, z
2
D(p) are distinct
but congruent modulo . Then we have z
2
= z
1
for some nontrivial . By the
denition of D(p), this gives us:
(z
1
, p) < (z
1
, p) = (z
2
, p) < (
1
z
2
, p) = (z
1
, p)
which is absurd. Hence D(p) contains at most one one element from each orbit in 1
and is a fundamental domain as required.
The next proposition determines the Dirichlet normal polygon for the modular group.
Proposition 3.4.7 (Normal Polygon for PSL
2
(Z)). A fundamental domain for PSL
2
(Z)
is the set:
z 1 : [z[ > 1, [ Re(z)[ < 1/2.
Proof. First, we show that for k > 1, the point p = ki 1 has trivial stabiliser in
PSL
2
(Z). Suppose that:
_
a b
c d
_
ki = ki.
By expanding and equating real and imaginary parts, we obtain a = d and b = ck
2
.
Hence by the determinant condition, we have:
a
2
+ c
2
k
2
= 1.
Since a, c Z, this forces c = 0 and a = 1, whence ki has no nontrivial stabilisers in
PSL
2
(Z). Hence from Proposition 3.4.6, we have that the polygon D(ki) for PSL
2
(Z)
is a fundamental domain.
We now construct this normal polygon. Since we have N(1), N(1) PSL
2
(Z), the
polygon D(p) is contained in the intersection of half-planes given by:
H
p
(N(1)) H
p
(N(1)) = z 1 : [ Re(z)[ < 1/2.
Moreover, D(p) is contained in the half-plane H
p
(K(/2)) = z 1 : [z[ > 1. Hence
we have:
D(p) z 1 : [z[ > 1, [ Re(z)[ < 1/2 =: F. (3.4.8)
We now show that the containment in Equation 3.4.8 above is in fact equality. Suppose
for the sake of contradiction, that F is not a normal polygon for PSL
2
(Z). Then we have
some z in the set F D(p). Now since D(p) is a fundamental domain for PSL
2
(Z), there
exists a representative z D(p) for some nontrivial PSL
2
(Z). Let
=
_
a b
c d
_
.
36 3. FUCHSIAN GROUPS
Then since z F, we have:
[cz + d[
2
= c
2
[z[
2
+ 2 Re(z)cd + d
2
> c
2
+ d
2
[cd[
= ([c[ [d[)
2
+[cd[
0.
Moreover, equality cannot be obtained unless c = d = 0, which violates the determinant
condition. Since c and d are integral, this gives us [cz + d[ > 1, whence:
Im(z) < Im(z). (3.4.9)
However, we can repeat the above calculation using the pair (z,
1
(z)). The only
difference in this calculation is that z possibly lies on the boundary F. Hence we
have:
Im(z) = Im(
1
z) Im(z). (3.4.10)
Combining the inequalities from Equations 3.4.9 and 3.4.10 yields the desired contradic-
tion. Hence we can conclude that F = D(p) is a fundamental domain for the Fuchsian
group PSL
2
(Z).
The fundamental domain constructed above for PSL
2
(Z) is an extended hyperbolic
triangle with vertex at . Notice that this vertex has stabiliser

= N, generated by
the parabolic element N(1), and is hence a representative of a cusp.
In order to treat all cusps of a Fuchsian group on an equal footing, we introduce the
scaling matrices.
Proposition 3.4.11. If s

1is a representative for a cusp of with stabiliser generated
by
s
, there exists a scaling matrix
s
PSL
2
(R) such that:

s
= s

1
s

s

s
=
_
1 1
0 1
_
.
Scaling matrices are unique up to right-multiplication by elements of the subgroup N
PSL
2
(R).
Proof. Suppose rst that s = and
s
=
_
1 r
0 1
_
, for some r R
+
.
3.4. FUNDAMENTAL DOMAINS 37
We then have by direct computation that A(

r)
1

s
A(

r) = N(1), so A(

r) is a
scaling matrix for s as required.
If s ,= , by the transitivity of the action of PSL
2
(R) on

1, there exists an element
g PSL
2
(R) such that g = s. Whence from the case s = , it easily follows that the
matrix gA(a
0
) is a scaling matrix for s, given suitable choice of a
0
. We omit for brevity
the proof of the uniqueness statement.
Using scaling matrices, we will show how to construct a fundamental domain for a
Fuchsian group with a cusp at . This construction will involve the use of the isometric
circles from Denition 1.2.20. We rst prove two preliminary results. The rst is a
lemma about the effect of PSL
2
(R)-transformations on isometric circles.
Lemma 3.4.12. Suppose g PSL
2
(R) and z 1 lies strictly outside the isometric
circle C
g
. Then the point gz lies strictly inside the isometric circle C
g
1.
Proof. Since z lies outside C
g
, we have: j
g
(z) < 1. Now by Equation (1.2.6) and
Denition 1.2.19, we have:
Im(z) = Im(g
1
gz) = j
g
1(gz)j
g
(z) Im(z).
Hence we can conclude that j
g
1(gz) > 1 and that gz lies strictly inside the isometric
circle C
g1
.
We will also need:
Lemma 3.4.13. Suppose is a Fuchsian group with cusp at , whose stability group is
generated by the parabolic element:
=
_
1 1
0 1
_
.
Then for all z 1, the orbit z contains a point with maximal imaginary part.
Proof. We rst show that the set Im(gz) : g is bounded for each z 1. Fix
z 1. Suppose that (g
n
) is a sequence in with Im(g
n
z) . Then by the denition
of the hyperbolic metric we have:
(z, g
1
n
g
n
z) = (g
n
z, g
n
z) 0
as n . This contradicts Proposition 3.2.7. Hence we can conclude that no such
sequence exists and we have M := supIm(gz) : g < . It remains to show
that M is actually attained. Suppose that Im(gz) < M for every g . Since ,
every point in 1 is equivalent to a point in the closure of the strip P := z 1 :
0 < Re(z) < 1. Hence the supposition that Im(gz) < M for every g implies
that there are innitely many distinct point in the orbit z that lie in the compact set
z P : M/2 Im(z) M. Since Fuchsian groups act discontinuously on 1, this is
absurd. This completes the proof.
Our second construction for the fundamental domain, the Ford Polygon, is then as
follows.
38 3. FUCHSIAN GROUPS
Proposition 3.4.14 (Ford Standard Polygon). Suppose that is a Fuchsian group that is
not co-compact. Let a be a representative of a cusp for . If we set:
F

:= z 1 : 0 < Re(z) < 1, z lies outside every isometric circle C


g
, g 1
then the set F :=
a
(F

) is a fundamental domain for .


Proof. We rst show that F

is a fundamental domain for


1
a

a
.
Suppose z F

. By Lemma 3.4.12, for any g


1
a

a
, the point gz will lie inside
some isometric circle C
g
1. Hence gz / F

, and we have that the points in F

are
pairwise inequivalent.
It remains to show that every orbit
1
a

a
z contains a representative in F

. By Lemma
3.4.13, every point z 1 is equivalent to a point z

1 of maximal imaginary part.


Moreover, from the denition of the scaling matrices, we have:
:=
_
1 1
0 1
_

1
a

a
.
Hence we can assume that 0 Re(z

) 1. Now since Im(z

) Im(gz

) for all
g
1
a

a
, we have from Equation (1.2.6) and Denition 1.2.19 that j
g
(z

) 1 for all
g
1
a

a
. This implies that z

lies on or outside every isometric circle, and is hence


an element of F

. Hence F

is a fundamental domain for


1
a

a
.
Now suppose we have two distinct points z

, w

F =
a
(F

). Then z

=
a
(z) and
w

=
a
(w) for two distinct points z and w in F

. If z

and w

are equivalent modulo


then w

= gz

for some nontrivial g . Hence by rearrangement, w =


1
a
g
a
z and w
is equivalent to z modulo
1
a

a
. Since z, w F

which is a fundamental domain for

1
a

a
, this is absurd, whence points in F are pairwise inequivalent modulo . Now let
z

=
a
(z) 1 be arbitrary. Similarly, since F

is a fundamental domain for


1
a

a
,
there exists g such that
1
a
g
a
z F

which upon rearrangement yields z

F.
Hence F is a fundamental domain for , and we are done.
The Ford Polygon will prove useful in establishing some bounds of the Kloosterman
sum in Section 3.5.
We shall now prove some key statements about fundamental domains for a Fuchsian
group. The rst establishes a numerical invariant of a Fuchsian group.
Proposition 3.4.15. Let be a Fuchsian group. Then any two fundamental domains for
have the same hyperbolic area.
Proof. Suppose F
1
and F
2
are two fundamental domains for , with (F
1
) < . We
have:
F
1
F
1

_
_

(F
2
)
_
=
_

F
1
(F
2
). (3.4.16)
Since F
2
is a fundamental domain, the nal union in Equation (3.4.16) is disjoint. Now,
from Proposition 1.2.27:
(F
1
)

(F
1
(F
2
)) =

(
1
(F
1
) F
2
) =

((F
1
) F
2
).
3.4. FUNDAMENTAL DOMAINS 39
Since F
1
is a fundamental domain for , we have:
_

(F
1
) = 1.
Hence:
(F
1
)

((F
1
) F
2
) =

((F
1
) F
2
) = (F
2
).
By symmetry, we can conclude that:
(F
1
) = (F
2
)
as required.
We can now dene the covolume of a Fuchsian group.
Denition 3.4.17. If is a Fuchsian group, the covolume (1) of is the hyperbolic
area of any fundamental domain thereof. In particular, if (1) < , we say that is
a nite-volume group.
Most of the Fuchsian groups we have discussed are nite-volume groups, an excep-
tion being the cyclic parabolic group from Proposition 3.3.2. For example, we consider
the Dirichlet polygon for PSL
2
(Z) which we found in 3.4.7. This is a hyperbolic triangle
with angles (0, /3, /3), and hence by the GaussBonnet Theorem has area /3.
There is in fact an intimate connection between nite-volume groups and Fuchsian
groups of the rst kind.
Proposition 3.4.18. A nite-volume Fuchsian group is a Fuchsian group of the rst kind.
Proof. The proof of this theorem is somewhat technical. Details can be found in [8,
p.103].
Theorem 3.4.19 (Siegels Theorem). A Fuchsian group of the rst kind which has a
hyperbolic n-gon as a fundamental domain for some n N is a nite-volume group.
Proof. The proof of of this theorem may be found in [8, p.80-84].
Remark 3.4.20. As a consequence of our constructions of the Dirichlet and Ford poly-
gons, the requirement that a polygonal fundamental domain for a Fuchsian group exists
is satised by all Fuchsian groups. However the requirement that this polygon has only
nitely many sides is a nontrivial condition. For more details on this condition see [8,
p.8084].
Since most of the Fuchsian groups listed at the start are simply congruence subgroups
of the modular group, and so are of nite index in PSL
2
(Z), we will now show how to
use the fundamental domain of a Fuchsian group to nd fundamental domains of its
nite-index subgroups. In particular, we have the following proposition, whose proof is
fairly straightforward.
40 3. FUCHSIAN GROUPS
Proposition 3.4.21. Let be a Fuchsian group with a fundamental domain F, and let
be a nite-index subgroup of . If
=
n
_
i=1

i
is a -coset decomposition of , then the set:
F

:=
n
_
i=1

i
(F)
is a fundamental domain for . Moreover, if (F) < and (F) = 0, then (F

) =
n(F).
We now examine the geometric structure of fundamental domains in some greater
detail. In the following section let be a xed Fuchsian group and let F be a xed
Dirichlet polygon thereof.
Denition 3.4.22 (Elliptic Point). Suppose that is a Fuchsian group. For z 1, we
say that the orbit z is an elliptic point of if the stabiliser
z
in is nontrivial.
As with cusps, we shall often refer to representatives of elliptic points as elliptic
points.
From Denition 3.4.1, it is immediate that any point z F with nontrivial stabiliser

z
in must lie on the boundary F.
From Theorem 3.4.6, the set F is a hyperbolic polygon, bounded by geodesic seg-
ments in 1 and possibly segments of the real axis.
Denition 3.4.23. Suppose that is a Fuchsian group with a Dirichlet polygon F. We
dene a vertex of F to be a point s F such that s is either the point of intersec-
tion of two bounding geodesics of F, or s is a representative of an elliptic point of .
Furthermore, we dene an edge of F to be a geodesic segment joining two vertices of
F.
If a vertex of a Dirichlet polygon F is not a representative of an elliptic point, it
is a representative of a cusp. Since the geodesics in 1 are precisely the vertical lines
and the semicircles orthogonal to the real axis, the angle made by two edges of F at
a cuspidal vertex is 0. This provides a geometric reason for the name cusp. The
geometric properties of Fuchsian groups are further developed in [8, Chapter. 4].
3.5. Kloosterman Sums
We shall now introduce an exponential sum called the Kloosterman sum, which will
be important for our work in Chapter 4. We rst construct a decomposition of suitable
Fuchsian groups into a union of double-cosets. This double coset decomposition will
be used to expand suitable functions into Fourier series at their cusps, which give us a
powerful tool for further analysis. This is the main reason why cusps are important, and
3.5. KLOOSTERMAN SUMS 41
why we shall consider only Fuchsian groups with cusps in Chapter 4.
In the following, we will assume that is a nite-volume Fuchsian group which is
not co-compact. In particular, this means that has at least one cusp a, from Proposition
3.3.16.
Let
B :=
__
1 n
0 1
_
: n Z
_
N
be the collection of all integral translations in PSL
2
(R).
We have the following result, which can be compared to the Bruhat decomposition
1.4.5:
Theorem 3.5.1 (Double coset decomposition). Suppose that is a nite-volume Fuch-
sian group which is not co-compact and that a, b are two cusps of , not necessarily
distinct. Then if we set
S := (c, d) R
2
: c > 0 and
_
a b
c d
_

1
a

b
for some a, b R

:= g
1
a

b
: g =
and

d/c
= B
d/c
B
for some arbitrarily chosen representative

d/c
=
_

c d
_

1
a

b
we have the disjoint union:

1
a

b
=

_
(c,d)S
_
d

d/c
(3.5.2)
where

is nonempty if and only if the cusps a and b are equivalent, and the union over
d runs over all equivalence classes of d modulo c.
Proof. We rst prove that

is nonempty if and only if the cusps a and b are equivalent.


Suppose gb = a for some g . Then by the denition of scaling matrices, we have:

1
a
g
b
=
1
a
gb =
a
a =
whence
1
a
g
b

,= . Conversely, suppose that


1
a
g
b

for some g .
Then again by the denition of scaling matrices, we have:

1
a
g
b
= g
b
=
a
gb = a
as required.
Suppose now that

is nonempty, and that w =


1
a
g
b
and w

=
1
a
g

b
are two
42 3. FUCHSIAN GROUPS
distinct elements of

. We have:
g

g
1
= (
a
w

1
b
)(
a
w
1
b
)
1
=
a
w

1
b

b
w
1

1
a
=
a
w

w
1

1
a
Hence g

g
1
a =
a
w

w
1

1
a
a =
a
w

w
1
=
a
= a and g

g
1
is in the stabiliser

a
. This gives us:
ww
1
= (
1
a
g
b
)(
1
b
g
1

a
) =
1
a
gg
1

a

1
a

a

a
= B.
Hence by symmetry,

= B

B = B

B is a double coset which is


nonempty if and only if a and b are equivalent.
The remaining elements of
1
a

b
each lie in a double coset of the form:

d/c
= B
d/c
B
for some

d/c
=
_

c d
_

1
a

b
.
Now the computation:
_
1 m
0 1
__
a
c d
__
1 n
0 1
_
=
_
a + cm
c d + cn
_
reveals that each double coset
d/c
completely determines c, whilst d is only determined
up to equivalence class modulo c. Moreover, each double coset is completely specied
by the bottom row of a representative thereof. To see this, suppose that:
w =
_
a
c d
_
and w

=
_
a


c d
_
.
We then have:
1
=
a
ww
1

1
a

a
, whence ww
1
B. By computing the
upper-left entry in the matrix B(k)w

, we can conclude that a = a

+ ck as required.
This completes the proof.
We can now dene the Kloosterman sum for .
Denition 3.5.3 (Kloosterman Sum). Let m, n Z and c R such that (c, d) S for
some d R. Given two cusps a and b of a nite-volume Fuchsian group that is not
co-compact, we dene the Kloosterman sum with modulus c and frequencies m and n
to be:
o
ab
(m, n; c) :=

e
_
m
d
c
+ n
a
c
_
(3.5.4)
where the sum is taken over double coset representatives:
_
a
c d
_
B
1
a

b
/B.
We have the following dependence on our choice of cusps and corresponding scaling
matrices.
3.5. KLOOSTERMAN SUMS 43
Proposition 3.5.5. Suppose a Fuchsian group has cusps: a, a

, b, b

, related by a

a
a and b

=
b
b for some
a
,
b
PSL
2
(R). Suppose further that the corresponding
scaling matrices are related by
a
=
a

a
N() and
b
=
b

b
N(). Then we have the
following relationship between Kloosterman sums:
o
a

b
(m, n; c) = e(m n)o
ab
(m, n; c). (3.5.6)
In the case of = PSL
2
(Z), the above reduces to the classical Kloosterman sum:
o(m, n; c) =

ad1 mod c
e
_
dm + an
c
_
.
In this case, we omit the subscripts of the Kloosterman sum, as there is a natural
choice for a cusp and a natural choice for the corresponding scaling matrix. That is,
a = b = and the scaling matrices are both the identity matrix.
The classical Kloosterman sums have connections to number theory. For example,
we have the following relationship with the Riemann Zeta function.
Proposition 3.5.7. For positive integers c, we have:
o(0, 0; c) = (c)
where is Eulers totient function, and hence:

n=1
n
2s
o(0, 0; n) =
(2s 1)
(2s)
.
In the remainder of this chapter we prove several bounds for Kloosterman sums
which will be required in our subsequent analysis.
Denition 3.5.8. If is a Fuchsian group with cusps a and b, then we dene the follow-
ing set:
(
ab
:=
_
c > 0 :
_

c
_

1
a

b
_
.
Proposition 3.5.9. Suppose that is a nite-volume Fuchsian group with cusp a. Then
the set (
aa
has a minimal positive element c
a
.
Proposition 3.5.10. For all c (
ab
, we have the bounds:
o
ab
(0, 0; c) c
2
c
1
ab
(3.5.11)
and

cM
c
1
o
ab
(0, 0; c) c
1
ab
M. (3.5.12)
where c
ab
= max c
a
, c
b
.
Proof. Without loss of generality, suppose that c
ab
= c
a
c
b
.
Suppose we have ,

, given by:
44 3. FUCHSIAN GROUPS
=
_

c d
_
and

=
_

c

_
.
If we have:

1
=
_

0
_
we have that

1
xes and hence by the denition of scaling matrices,

1
=
N(1)
k
for some k Z. In this case we have c = c

and d = d

.
If instead we have:

1
=
_

c


_
with [c

d cd

[ = [c

[ > 0, we use Proposition 3.5.9 to give us the spacing property:


[d

c
1
dc
1
[ c
a
cc

. (3.5.13)
In the case where c = c

, this equation reduces to:


[d

d[ c
a
c
1
.
Since the Kloosterman sum in question simply counts equivalence classes modulo c, it
follows that:
[o
ab
(0, 0; c)[ c(c
a
c
1
)
1
= c
1
ab
c
2
as required.
To obtain the stronger, on average result, we can enumerate all pairs (c, d) with
c M and 0 d < c by the size of dc
1
. If we sum the inequality 3.5.13 over this
enumeration, where (c

, d

) is the successor pair to (c, d), we obtain the required result


(observing that c

X).
Corollary 3.5.14. The Kloosterman sums obey the following bounds:
[o
ab
(m, n; c)[ maxc
a
, c
b

1
c
2
(3.5.15)
and

cM
c
1
[o
ab
(m, n; c)[ maxc
a
, c
b

1
M (3.5.16)
Proof. This corollary follows by taking modulus signs inside the sum and applying
Proposition 3.5.10.
Proposition 3.5.17. Suppose is a nite-volume Fuchsian group with cusp a. We have
the following:
[
a

a
: Im(
1
a
z > Y )[ < 1 +
10
c
a
Y
.
Proof. By conjugation, it sufces to prove the Proposition for the case where a =
and
a
= N(1). Let F be the Ford polygon for and without loss of generality assume
that z F. If
a

a

a
: Im(
1
a
z > Y ), we have [cz + d[ 1 by the
denition of the Ford polygon. Hence by Equation 1.2.6, we have:
y > Y,
3.5. KLOOSTERMAN SUMS 45
and by taking real and imaginary parts, we get:
c < y
1/2
Y
1/2
and [cx + d[ < y
1/2
Y
1/2
.
Using the second of these inequalities combined with the spacing property 3.5.13 we can
bound the number of pairs (c, d) with C c < 2C above by:
1 + 8c
1
a
Cy
1/2
Y
1/2
10c
1
a
Cy
1/2
Y
1/2
.
We now sum this bound over C = 2
n
y
1/2
Y
1/2
, n Z
+
. This sum is equal to
10c
1
a
Y
1
and is an upper bound for the number of suitable cosets
a
which do not x
. To complete the proof we simply add 1 to account for the coset
a
.
We have the following deep bound due to A. Weil for Kloosterman sums with respect
to the modular group.
Theorem 3.5.18. Suppose that = PSL
2
(Z). Then:
[o(m, n; c)[ gcd(m, n, c)
1/2
c
1/2
(c).
CHAPTER 4
Automorphic Forms and their Spectra
In this nal chapter we study automorphic functions, and amongst them, the automor-
phic forms. Our primary objective is to obtain a spectral decomposition for the space of
square-integrable automorphic functions with respect to the LaplaceBeltrami operator.
Due to space limitations we cannot hope to provide a thorough account here. Rather,
we shall provide an illustration of how the techniques of harmonic analysis can be used
in conjunction with the theory of Fuchsian groups to obtain deep results concerning the
spectral theory of square-integrable automorphic forms. Our treatment of this topic runs
along similar lines to that in [7] and [10, Chapter 15].
In Section 4.2, we will rst dene automorphic functions and introduce the method
of averaging images, an important tool for constructing such functions. We shall then
in Section 4.3 study the Eisenstein series, whose theory leads us to a decomposition
of the space of square-integrable automorphic functions into two orthogonal subspaces,
which have very different spectral theory. These subspaces shall give rise to the discrete
and continuous spectra of the LaplaceBeltrami operator on L
2
(1). In Section
4.3, we will prove the discrete part of the spectral theorem using the theory of compact
operators. Then in Section 4.4 we shall revisit the Green Function in the context of
automorphic functions. This will lead to a better understanding of the resolvent, which
will be necesssary for our nal section, in which we discuss an important conjecture of
Selberg concerning the point spectrum in the case where is a congruence subgroup.
4.1. Spectral Theory
The spectral theory of an operator on a function space is covered extensively in [11].
We summarise here the results that will be needed for this chapter.
Denition 4.1.1. Let T be a continuous linear operator on a Banach space X. The
spectrum of T is dened to be the set:
(T) := C : I T is not invertible.
There are several ways in which a linear operator can fail to be invertible, and ac-
cordingly we can divide the spectrum of a linear operator into the point spectrum, the
continuous spectrum and the residual spectrum as follows.
Denition 4.1.2. We dene the following subsets of the spectrum (T) of a linear oper-
ator T.
46
4.2. AUTOMORPHIC FORMS 47
The point spectrum is the set:
C : I T is not injective.
The continuous spectrum is the set:
C : I T has non-closed range which is dense in X.
The residual spectrum is the set:
C : I T has non-closed range which is not dense in X.
The elements of the point spectrum are the eigenvalues of the operator T.
In the case that we are working in a Hilbert space H, we have the following important
results.
Theorem 4.1.3. A symmetric non-negative operator on a Hilbert space admits a self-
adjoint extension.
Proposition 4.1.4. Let Lbe a symmetric operator on H with nite-dimensional eigenspaces
that commutes with another symmetric operator . Then there exists a maximal or-
thonormal system of eigenvectors of L in H which are also eigenvectors of .
Finally we state the HilbertSchmidt Theorem, a powerful way of nding eigenfunc-
tion expansions with respect to integral operators of a certain kind.
Denition 4.1.5. An integral operator is said to be of HilbertSchmidt type if its region
of integration is an open connected set F and its kernel k is square integrable over F F.
Theorem 4.1.6 (HilbertSchmidt Theorem). If L is a nonzero HilbertSchmidt operator
with a real kernel, then we have:
L has a nonempty and purely discrete spectrum.
Eigenspaces of L are nite-dimensional.
The range of L in L
2
(F) is spanned by the eigenfunctions of L.
If u
j

j0
is a maximal orthonormal system of eigenfunctions of L, the series:
f(z) =

j=0
f, u
j
u
j
(z)
is absolutely and uniformly convergent.
4.2. Automorphic Forms
The most important result in this section is the Poincar series, which gives us a
simple method for constructing automorphic functions.
Recalling Denition 3.1.2, a function f : 1C is said to be an automorphic func-
tion with respect to the nite-volume Fuchsian group if it obeys the invariance:
f(z) = f(z) for all .
48 4. AUTOMORPHIC FORMS AND THEIR SPECTRA
Example 4.2.1. With respect to the Fuchsian group of integral translations B, the func-
tion:
f(x + iy) := e(x)
is automorphic.
In the following work we shall assume that is a xed nite-volume Fuchsian group
that is not co-compact. Furthermore, we take F to be a xed polygonal fundamental
domain for .
As automorphic functions are completely specied by their values on each orbit
in 1, we can view an automorphic function f as being dened on the quotient space
1. We shall not explore the geometry of 1 in this essay, but this quotient as a
Riemannanian orbifold obtained by gluing congruent sides of our polygonal fundamental
domain together.
We denote the space of all automorphic functions with respect to by /(1).
A common technique for constructing automorphic functions is the the method of
averaging images.
Proposition 4.2.2. Suppose that is a nite-volume Fuchsian group and p : 1C is
a function which decays sufciently rapidly so that:
f(z) :=

p(z)
is a convergent series for all z 1. Then f /(1).
Proof. Suppose that p obeys the required growth constraints (which depend on ). We
can then treat the sum formally:
f(

z) =

p(

z)
=

p(
1

z)
=

p(z)
= f(z)
for all

.
As an extension of the idea above, we can also construct automorphic functions by
summing over cosets in .
Proposition 4.2.3. Suppose that is a nite-volume Fuchsian group, and is an innite
subgroup thereof. Suppose further, that p : 1C is a function which is automorphic
with respect to . Then
f(z) :=

p(z)
4.2. AUTOMORPHIC FORMS 49
where the summation is taken over all right cosets in , is automorphic with respect
to . Again we require that p satises certain growth constraints in order to ensure that
the series dening f converges.
Proof. If we treat the sum formally as in the proof of Proposition 4.2.2, the result follows
from the automorphy of p.
Using Proposition 4.2.3 above, we can dene the Poincar series.
Corollary 4.2.4. Let be a Fuchsian group that is not co-compact and hence has a
cusp. Let p : 1C be a function which is Binvariant and satises a suitable growth
condition. Then the function:
E
a
(z[p) =

a\
p(
1
a
z)
is automorphic with respect to .
Denition 4.2.5. In Corollary 4.2.4 above, we say that E
a
(z[p) is the Poincar series
of p with respect to at the cusp a.
Now let a be a cusp for and
a
be a scaling matrix for this cusp. From the denition
of a scaling matrix, we have the following Fourier series expansion at the cusp a.
Proposition 4.2.6. Suppose that f : 1Clies in /(1). Then we have the following
Fourier series expansion:
f(
a
z) =

nZ
f
an
(y)e(nx) (4.2.7)
with coefcients given by:
f
an
(y) =
_
1
0
f(
a
z)e(nx) dx.
If f is holomorphic, then the series (4.2.7) converges absolutely and uniformly on com-
pact subsets of 1.
Proof. This result follows from the classical Fourier series expansion.
Of particular interest are the automorphic functions which are also eigenfunctions of
the LaplaceBeltrami operator (see Denition 2.1.5.)
Denition 4.2.8 (Automorphic Forms). An automorphic form is an automorphic func-
tion f /(1) which is also an eigenfunction of the LaplaceBeltrami operator.
Explicitly:
( + )f = 0 for some C.
We denote the space of automorphic forms with eigenvalue = s(1 s) C by
/
s
(1) /(1).
50 4. AUTOMORPHIC FORMS AND THEIR SPECTRA
Automorphic forms are often called Maass forms, however we shall maintain the
nomenclature of Iwaniec, so as to emphasise their inherent symmetry.
If f : 1C is an automorphic form, we can make the Fourier series (4.2.7) more
explicit.
Theorem 4.2.9. Suppose f /
s
(1) satises the growth condition:
f(
a
z) = o(e
2y
).
Then we have the Fourier expansion of f at the cusp a given by:
f(
a
z) = f
a
(y) +

nZ\0

f
a
W
s
(nz) (4.2.10)
where the zero-th term f
a
(y) is given by a linear combination of the functions from
Proposition 2.2.9. The nonzero coefcients in (4.2.10) are bounded by:

f
a
(n) = O(e
|n|
) (4.2.11)
for any > 0, where the implied constant depends on and f.
Proof. This is a direct application of Theorem 2.3.3 and Corollary 2.3.6.
The asymptotic bound on Fourier coefcients in (4.2.11) gives us the following as-
ymptotic bound of automorphic forms in cuspidal zones.
Corollary 4.2.12. If the function f /
s
(1) has a Fourier expansion at the cusp a
given by (4.2.10), then we have:
f(
a
z) = f
a
(y) + O(e
2y
).
The most convenient setting for our subsequent analysis of automorphic forms is the
space:
L
2
(1) :=
_
f /(1) :
_
F
[f(z)[
2
d <
_
.
In particular, since is a nite-volume group, any bounded automorphic function is in
L
2
(1). We shall compute the point spectrum of on L
2
(1) in the remainder of
this chapter. First we must consider the Eisenstein series.
4.3. Eisenstein Series
In this section, we dene one of the most important Poincar series, the Eisenstein
series. The Eisenstein series is important for our study of the automorphic Green Func-
tion, and consequently our analysis of the discrete spectrum. Our main result is the
decomposition of L
2
(1) into two orthogonal subspaces in Corollary 4.3.12.
We rst obtain sufcient conditions for the convergence of the Poincar series from
Proposition 4.2.4 given a particular weight function.
4.3. EISENSTEIN SERIES 51
Proposition 4.3.1. Suppose : R
+
C is a smooth function obeying the growth con-
straint:
(y) = O(y(log y)
2
) as y 0. (4.3.2)
Then the Poincar series generated by the weight function:
p(z) = (y)e(mz)
converges absolutely on 1 for any non-negative integer m. We write the resulting auto-
morphic function as E
am
(z[).
Proof. This follows readily from 3.5.17 using a dyadic decomposition of 1 into hori-
zontal strips.
Applying the above proposition to the weight function (y) := y
s
, we obtain the
Eisenstein series.
Denition 4.3.3 (Eisenstein Series). For a nite-volume Fuchsian group with cusp a and
for s C with Re(s) > 1, we dene the Eisenstein series:
E
a
(z, s) :=

a\
Im(
1
a
z)
s
.
By Propositions 4.2.4 and 4.3.1, the Eisenstein series is automorphic with respect to .
We now show that the Eisenstein series is an automorphic form.
Proposition 4.3.4. The Eisenstein series E
a
(z, s) is an automorphic formin z with eigen-
value = s(1 s).
Proof. This follows from the fact that y
s
is an eigenfunction of the LaplaceBeltrami
operator and use of the chain rule.
Note however, that E
a
(z, s) is not square-integrable over F.
Another important class of weighted Poincar series constructed using Proposition
4.3.1 is the class of incomplete Eisenstein series.
Denition 4.3.5 (Incomplete Eisenstein Series). If is compactly supported, then we
dene the incomplete Eisenstein series by:
E
a
(z[) :=

a\
(Im(
1
a
z)).
By Propositions 4.2.4 and 4.3.1, the incomplete Eisenstein series are automorphic
with respect to . Furthermore, since is compactly supported and (F) < , the in-
complete Eisenstein series E
a
(z[) is bounded and hence square-integrable. We denote
the space of all incomplete Eisenstein series by c(1).
In the following section, we denote the space of smooth bounded automorphic func-
tions by B(1), and the space of incomplete Eisenstein series by c(1).
Proposition 4.3.6. We have the inclusions:
c(1) B(1) L
2
(1) /(1).
52 4. AUTOMORPHIC FORMS AND THEIR SPECTRA
The above proposition leads us to examine the orthogonal complement to c(1) in
B(1).
Proposition 4.3.7. Suppose f(z) is an automorphic function which is absolutely inte-
grable over F. Suppose E
a
(z[) is the incomplete Eisenstein series dened using the
cusp a and the compactly supported function : R
+
C. Then we have:
f, E
a
([) =
_

0
f
a
(y)(y)y
2
dy (4.3.8)
where f
a
is the zero-th term in the Fourier expansion 4.2.6 of f at a.
Proof. From the denition of the inner-product we have:
f, E
a
([) =
_
F
f(z)E
a
(z[) d(z)
=
_
F
f(z)

a\
(Im(
a
z)) d(z)
=

a\
_

1
a
F
f(
a
z)(y) d(z)
where the last line follows from automorphy of f by changing the order of summation
and integration. Now, from our denition of scaling matrices, the sets
1
a
F tessellate
the strip
P := z 1 : 0 < Re(z) < 1
as runs over
a
. Hence we obtain:
_
P
f(
a
z)(y) d(z) =
_

0
__
1
0
f(
a
z)dx
_
(y)y
2
dy
by change of variables and the denition of the measure . Now the inner integral on the
right-hand side is simply the zero-th term of the Fourier expansion in Proposition 4.2.6.
This completes the proof.
Denition 4.3.9. We dene ((1) to be the space of all smooth bounded automorphic
functions which have vanishing zero-th Fourier coefcients with respect to every cusp.
Intuitively speaking, these are the smooth and bounded automorphic functions that
vanish at every cusp. Amongst the elements of ((1) are the cusp forms.
Denition 4.3.10. We dene cusp forms to be elements of ((1) that are also auto-
morphic forms. We write:
(
s
(1) := ((1) /
s
(1).
Proposition 4.3.11. Cusp forms are bounded functions on 1. Furthermore, the Fourier
coefcients of a cusp form are bounded at any cusp.
FromProposition 4.3.7, we have the following orthogonal decomposition of L
2
(1).
4.3. EISENSTEIN SERIES 53
Corollary 4.3.12.
L
2
(1) = ((1) c(1).
The spectral theory of the LaplaceBeltrami operator is decidedly different on the
orthogonal subspaces ((1) and c(1). We will show in the following section that
the spectrum of restricted to ((1) is purely discrete. It turns out that the spectrum
of restricted to c(1) is continuous except for on a nite-dimensional subspace,
however we will not have the space to prove that result in this essay. For details on this,
the interested reader is referred to [7, Chapter 7].
The spectral resolution on ((1) will utilise the theory of compact operators and
in particular the Hilbert-Schmidt Theorem.
Before we prove the discrete part of our spectral theorem, we rst observe some
properties of the LaplaceBeltrami operator. It is convenient here to work over the space:
T(1) := f B(1) : f B(1).
Proposition 4.3.13. The subspace T(1) is dense in L
2
(1).
Consequently, there is an unique self-adjoint continuous linear operator on L
2
(1)
that agrees with on T(1).
From Stokes theorem on 1, we have the following useful identity.
Lemma 4.3.14. If f, g T(1) then we have:
f, g =
_
F
fg dx dy.
This gives us the following result.
Proposition 4.3.15. Suppose f, g T(1). Then we have:
f, g = f, g
and
f, f 0.
That is, the LaplaceBeltrami operator on L
2
(1) is symmetric and non-negative.
Proof. This proposition follows directly from Lemma 4.3.14.
As an upshot of Proposition 4.3.15, the eigenvalue corresponding to any eigenfunc-
tion of in T(1) is real and non-negative.
We now recall the invariant integral operators from Section 2.5. An invariant integral
operator L is given by:
(Lf)(z) =
_
H
k(z, w)f(w)d(w)
where the kernel k is point-pair invariant and obeys growth condition such that absolute
convergence is ensured.
54 4. AUTOMORPHIC FORMS AND THEIR SPECTRA
When dealing with spaces of automorphic functions, it is convenient to work with a
kernel that obeys a symmetry similar to automorphy of complex-valued functions. We
can construct such automorphic kernels using the method of averaging images.
Denition 4.3.16 (Automorphic Kernel). Given a kernel k(z, w) for an invariant integral
operator, we dene the automorphic kernel:
K(z, w) =

k(z, z).
Suppose rst that k(u(z, w)) is the smooth and compactly supported kernel of an
invariant integral operator L. We have the following immediate result.
Proposition 4.3.17. The operator L maps B(1) into B(1).
Proof. By point-pair invariance of k and automorphy of f, we have:
(Lf)(z) =
_
H
k(z, w)f(w) d(w) =
_
H
k(z, w)f(w) d(w) = (Lf)(z)
for all .
Moreover, the subspace ((1) is preserved by L.
Proposition 4.3.18. If L is an invariant integral operator, then the subspace ((1) of
B(1) is Linvariant.
Proof. This is a straightforward calculation, using the Fourier expansion of elements of
((1).
Despite their symmetry, the automorphic kernels K(z, w) have the seemingly serious
problem that they are unbounded as z and w approach the same cusp. To see this, it
sufces to consider the case where this cusp is with stabiliser generated by N(1). In
this case, we can use a similar technique as to in the proof of Lemma 3.4.13 to show that
there exist arbitrarily many for which k(u(z, w)) does not vanish.
The way around the bad behaviour of automorphic kernels in cuspidal zones is by
subtracting the parts which cause problems.
Denition 4.3.19. We dene the principal part of the automorphic kernel K at the cusp
a by:
H
a
:=

a\
__
R
k(z,
a
N(t)
1
a
w) dt
_
. (4.3.20)
Lemma 4.3.21. For z, w 1, we have the asymptotic bound:
H
a
(
a
z, w) = O(1 + Im(z)) (4.3.22)
which is uniform in w.
4.3. EISENSTEIN SERIES 55
Proof. By a change of variables and the point-pair invariance of the kernel k, we have:
H
a
(
a
z) =

a\
__
R
k(
a
z,
a
N(t)
1
a
w)
_
dt
=

a\
__
R
k(z, N(t)
1
a
w)
_
dt
=

B\
1
a
a
__
R
k(z, N(t)w)
_
dt
=

B\
1
a
a
__
R
k(z, w + t)
_
dt
But since k is compactly supported, we can restrict t and to values such that u(z, t +
w) M (where M > 0 is some constant) without changing the value of this expres-
sion. Now using Equation (1.2.25) and by considering real and imaginary parts, we get
Im(z) = O(Im(w)) and Im(w) = O(Im(z)). Hence, the restriction of values that
t can take shows that the integral is bounded above by O(Im(z)). Hence Proposition
3.5.17 leads us to the conclusion that:
H
a
(
a
z,
a
w) = O
_
1 +
1
Im(z)
_
O(Im(z)) = O(1 + Im(z)).

Proposition 4.3.23. For xed z 1 and xed f ((1), the inner product:
H
a
(z, ), f = 0.
Proof. By a change of variables and using the trick from the proof of Proposition 4.3.7
to change a sum of integrals over a fundamental domain to an integral over the strip P,
we have:
H
a
(z, ), f =
_
F
H
a
(
a
z, w)f(w)d(w)
=
_
F
_
_

a\
_
R
k(z, N(t)
1
a
w) dt
_
_
f(w)d(w)
=
_
R
__
P
k(z, N(t)w)f(w) d(w)
_
dt
=
_

0
__
R
k(z, t + iv) dt
___
1
0
f(
a
w) du
_
v
2
dv
= 0
from the denition of the space ((1).
56 4. AUTOMORPHIC FORMS AND THEIR SPECTRA
As a consequence of Proposition 4.3.23 above, we can dene the compact part

K
of the kernel K(z, w) by:

K(z, w) = K(z, w)

a
H
a
(z, w). (4.3.24)
The resulting modied integral operator

L satises the following corollary.
Corollary 4.3.25. For f ((1), we have Lf =

Lf.
Proof. This follows directly from Proposition 4.3.23.
Proposition 4.3.26. If F is a polygonal fundamental domain for with noncongruent
cuspidal vertices, then the modied kernel

K is bounded on F F.
Proof. The proof of this proposition is quite involved, and invokes the EulerMaclaurin
Formula. See [7, p.7374] for details.
We are now ready to compute the spectrum of on ((1).
From Denition 4.1.5 we have that the modied integral operator

L is of Hilbert-
Schmidt type. However, the HilbertSchmidt theorem only gives us an eigenfunction
expansion for functions in the range of our chosen HilbertSchmidt operator. We need
then, a particular choice of Lsuch that

Lhas a range of interest to us. Recall the resolvent
operator R
s
from Denition 2.1.3. Dene the integral operator L by:
L := R
s
R
a
(4.3.27)
where a > s 2.
Proposition 4.3.28. For L as dened above, the modied operator

L is bounded on
L
2
(F) and has dense range in the subspace ((1). Moreover, the subspace ((1)
T(1) is contained in the range of

L.
Using the Hilbert-Schmidt theorem and Proposition 4.3.28 above, we obtain the fol-
lowing result.
Proposition 4.3.29. The operator

L maps the subspace ((1) T(1) densely into
itself. This restricted operator has a purely discrete spectrum with nite-dimensional
eigenspaces. There is a complete orthonormal system of eigenfunctions of

L in ((1),
and given any choice of complete orthonormal system of eigenfunctions u
j
of

L in
((1), an arbitrary function f ((1) T(1) has the expansion:
f(z) =

j0
f, u
j
u
j
(z)
which converges absolutely and uniformly on compact subsets of 1.
Proof. This proposition is a direct application of the HilbertSchmidt Theorem.
4.3. EISENSTEIN SERIES 57
Now

L is an invariant integral operator and hence commutes with . So since

L
is a symmetric operator with eigenspaces of nite dimension that commutes with ,
by Proposition 4.1.4 there exists a maximal orthonormal system of eigenfunctions of L
which are also eigenfunctions of . Hence we have the discrete part of our spectral
theorem.
Theorem 4.3.30. The LaplaceBeltrami operator has a purely discrete spectrum in
((1). Furthermore the space ((1) is spanned by cusp forms, and given any
f T(1) ((1), the series expansion:
f(z) =

j0
f, u
j
u
j
(z)
converges absolutely and uniformly on compact subsets of 1.
Proof. Follows from Proposition 4.3.29 above and the subsequent discussion.
Having established the discrete part of our spectral theorem, we now turn our at-
tention towards the elements of the point spectrum themselves.
In order to prove some bounds for elements of this spectrum, we will need to state
the Fourier expansions for the Eisenstein series, and subsequently the automorphic Green
function.
We now provide the explicit Fourier expansion of the Eisenstein series.
Theorem 4.3.31 (Fourier Expansion of Eisenstein Series). Let be a nite-volume
Fuchsian group with cusps at a and b. Let s C with Re(s) > 1. Then we have
the Fourier expansion:
E
a
(
b
z, s) =
ab
y
s
+
ab
(s)y
1s
+

nZ\{0}

ab
(n, s)W
s
(nz) (4.3.32)
where

ab
(s) =
1/2
(s 1/2)
(s)

c
c
2s
o
ab
(0, 0; c), (4.3.33)

ab
(n, s) =
s
(s)
1
[n[
s1

c
c
2s
o
ab
(0, n; c), (4.3.34)
and W
s
(z) is the Whittaker function 2.2.11.
Proof. This result is a rather lengthy application of the Poisson summation formula. For
a complete proof see [7, p.65-66].
In order to further analyse the spectrum of on ((1), we shall need to return to
the study of the Green function from Section 2.4, with a view towards better understand-
ing the resolvent R
s
.
58 4. AUTOMORPHIC FORMS AND THEIR SPECTRA
4.4. The Automorphic Green Function
In this section we shall dene the automorphic Green Function and state its Fourier
expansion. We will also state without proof a deep result concerning the analytic contin-
uation of the automorphic Green Function.
We construct the automorphic Green function on 1 using the method of averaging
images from Proposition 4.2.2.
Denition 4.4.1 (Automorphic Green Function). For z, z

1 that are inequivalent


modulo , we dene the automorphic Green Function as follows:
G
s
(z/z

) :=

G
s
(z, z

) (4.4.2)
where G
s
is the Green function dened in Proposition 2.4.3.
Proposition 4.4.3. For Re(s) > 1, the series (4.4.2) converges absolutely.
Proof. This follows from the bound in Lemma 2.4.4.
Proposition 4.4.4. The automorphic Green function is an automorphic form in each
variable.
Proof. This follows from the fact that G
s
is an eigenfunction of .
Proposition 4.4.5. The resolvent R
s
is the right inverse to the operator ( + s(1 s))
on the space B(1).
The Fourier expansion of the automorphic Green Function uses the theory of the
Green Function of an ordinary differential equation. Full details can be found in [7,
p.7881,p.8386]. The main result holds on the restricted domain:
D
ab
= (z, z

) 11 : y

> y, y

y > c(ab)
2

where c(a, b) denotes the minimal element of the set (


ab
.
Theorem4.4.6 (Fourier Expansion of the Automorphic Green Function). Suppose Re(s) >
1 and (z, z

) D
ab
. We have the following Fourier series expansion:
G
s
(
a
z/
b
z

) = (2s 1)
1
(y
s
(y

)
1s

ab
+
ab
(s)(yy

)
1s
)
+

nZ\{0}
(4[n[)
1
W
s
(nz

)V
s
(nz)
ab
+ (2s 1)
1
y
1s

mZ\{0}

ab
(m, s)W
s
(mz

)
+ (2s 1)
1
(y

)
1s

nZ\{0}

ab
(n, s)W
s
(nz)
+

mn=0
Z
s
(m, n)W
s
(mz

)W
s
(nz)
4.5. SMALL EIGENVALUES 59
where
ab
(s) and
ab
(n, s) are the Fourier coefcients of the Eisenstein series in Equa-
tions (4.3.33) and (4.3.34), and Z
s
(m, n) is the zeta-function of the Kloosterman sum,
dened by:
Z
s
(m, n) :=
_
2
1
[mn[
1/2

c
c
1
o
ab
(m, n; c)I
2s1
(4c
1
[mn[
1/2
) (for mn 0)
2
1
[mn[
1/2

c
c
1
o
ab
(m, n; c)J
2s1
(4c
1
(mn)
1/2
) (for mn > 0)
(4.4.7)
where c runs over the set (
ab
(see Denition 3.5.8), I is the modied Bessel function from
2.2.8, and J is the Bessel Function, dened in [2, p.199].
Theorem 4.4.6 is very deep and uses results from [7, Chapters 89].
Proposition 4.4.8. The zeta-function Z
s
(m, n) converges absolutely for Re(s) > 1.
The zeta-function Z
s
(m, n) encodes the spectral theory of automorphic forms in a
way that is supercially similar to the way the Riemann zeta-function encodes informa-
tion about the primes.
Using the Fourier expansion of the Eisenstein series, we can obtain the following
analytic continuation of Z
s
(m, n) to a meromorphic function in s.
Theorem 4.4.9. Suppose the discrete spectrum of the LaplaceBeltrami operator on
L
2
(1) consists of eigenvalues:

j
= s
j
(1 s
j
).
Then the function Z
s
(m, n) has an analytic continuation in s to all of C, with simple
poles at s = s
j
. There is a reection formula for this analytic continuation about the
line Re(s) = 1/2.
Proof. See [7, p.137].
4.5. Small Eigenvalues
For arbitrary Fuchsian groups , there can be arbitrarily small eigenvalues in the
spectrum of ((1). This is proven by Chavel in [5, Chapter X, Section 5].
For congruence subgroups however, we have very different spectral behaviour.
We shall conclude this essay by studying the special case in which is a congruence
subgroup of PSL
2
(Z) (recall Denition 3.3.6). We have the following conjecture due to
Selberg.
Conjecture 4.5.1 (Selbergs Eigenvalue Conjecture). Let be a congruence subgroup
of PSL
2
(R) and let f /
s
(1). Then the corresponding eigenvalue satises:
= s(1 s) 1/4.
This conjecture is one of the major unsolved problems in the study of automorphic
forms, and has number theoretic applications to such problems as the prime geodesic
problem. This problem is discussed in, for example, [10, Chapter 15], and [12].
60 4. AUTOMORPHIC FORMS AND THEIR SPECTRA
Using the analytic continuation of the automorphic Green function, we can prove the
following bound, rst proven by Selberg himself. The particular proof that we present is
due to Iwaniec in [7].
Theorem 4.5.2 (Selbergs Theorem). Let be a congruence subgroup of PSL
2
(R). Let
f /
s
(1). Then the corresponding eigenvalue satises:
= s(1 s) 3/16. (4.5.3)
Proof. Let s = + it, with , t R. We show that Z
s
(m, n) converges absolutely for
> 3/4. Let = 3/4 + for any xed > 0. We shall assume mn 0 (the case
mn > 0 is similar). From Proposition 2.2.8, we have that:
I

(x) (x/2)

( + 1)
as x 0. Hence as c , Weils bound 3.5.18 gives us:
[c
1
o
ab
(m, n; c)I
2s1
(4[mn[
1/2
c
1
)[ = O(c
1
g
1/2
c
1/2
(c)(2[mn[
1/2
c
1
)
21
)
= O(c
1/22
(c))
= O(c
12
(c))
= O(c
1
)
where g = gcd(m, n, c) and where the last line follows from [3, p.296]. Hence the series
in Equation (4.4.7) converges, and by Theorem 4.4.9, all eigenvalues = s(1 s) of
satisfy 1/4 Re(s) 3/4. Hence we have:
= Re() = Re(s) Re(1 s) Im(s) Im(1 s) 3/16 + Im(s)
2
3/16.

This is the rst nontrivial step towards Selbergs conjecture. Unfortunately it turns
out that the machinery developed in this essay cannot really do better (see [12]). In or-
der to improve Theorem 4.5.2, we would need a way of detecting cancellations in the
Kloosterman sums.
Recent progress has been made into improving the bound in Theorem 4.5.2 some-
what. In [12], the developments prior to 1995 are discussed. The current best known
result was proven by Kim in 2002, in [9].
Theorem 4.5.4 (Kim). Let be a congruence subgroup of PSL
2
(R). Let f /
s
(1).
Then the corresponding eigenvalue satises:
= s(1 s) 975/4096 0.238. (4.5.5)
References
[1] Lars V. Ahlfors. Complex Analysis, Third Edition. McGraw-Hill, Inc., 1979.
[2] Andrews-Askey-Roy. Special Functions. Cambridge University Press., 1999.
[3] Tom M. Apostol. Introduction to Analytic Number Theory. SpringerVerlag, 1976.
[4] Peter Buser. Geometry and Spectra of Compact Riemann Surfaces. Birkhauser.,
1992.
[5] Isaac Chavel. Eigenvalues in Riemannian Geometry. Academic Press, Inc., 1984.
[6] Sigurdur Helgason. Groups and Geometric Analysis. Academic Press, Inc., 1984.
[7] Henryk Iwaniec. Introduction to the Spectral Theory of Automorphic Forms. Re-
vista Matemtica Iberoamericana., 1995.
[8] Svetlana Katok. Fuchsian Groups. The University of Chicago Press, Ltd., London.,
1992.
[9] Henry H. Kim. Functoriality for the exterior square of gl
4
and the symmetric fourth
of gl
2
. J. Amer. Math. Soc., 2002.
[10] Henryk Iwaniec-Emmanuel Kowalski. Analytic Number Theory. American Mathe-
matical Society Colloquium Publications, Volume 53., 2003.
[11] Walter Rudin. Functional Analysis. McGraw-Hill, Inc., 1973.
[12] Peter Sarnak. Selbergs Eigenvalue Conjecture, on the occasion of Robert Osser-
mans retirement. In Notices of the AMS. Volume 42. Number 11., 1995.
[13] Mariano Surez-Alvarez. http://math.stackexchange.com/questions/29515/,
12/10/11.
61

You might also like