Lectures On Arithmetic Noncommutative Geometry

Lectures on Arithmetic
Noncommutative Geometry
Matilde Marcolli
And indeed there will be time
To wonder “Do I dare?” and, “Do I dare?”
Time to turn back and descend the stair.
...
Do I dare
Disturb the Universe?
...
For I have known them all already, known them all;
Have known the evenings, mornings, afternoons,
I have measured out my life with coffee spoons.
...
I should have been a pair of ragged claws
Scuttling across the floors of silent seas.
...
No! I am not Prince Hamlet, nor was meant to be;
Am an attendant lord, one that will do
To swell a progress, start a scene or two
...
At times, indeed, almost ridiculous–
Almost, at times, the Fool.
...
We have lingered in the chambers of the sea
By sea-girls wreathed with seaweed red and brown
Till human voices wake us, and we drown.
(T.S. Eliot, “The Love Song of J. Alfred Prufrock”)

Contents
Chapter 1. Ouverture 5
1. The NCG dictionary 7
2. Noncommutative spaces 8
3. Spectral triples 9
Chapter 2. Noncommutative modular curves 17
1. Modular curves 17
2. The noncommutative boundary of modular curves 24
3. Modular interpretation: noncommutative elliptic curves 24
4. Limiting modular symbols 29
5. Hecke eigenforms 40
6. Selberg zeta function 43
7. The modular complex and K-theory of C∗ -algebras 44
8. Intermezzo: Chaotic Cosmology 45
Chapter 3. Quantum statistical mechanics and Galois theory 53
1. Quantum Statistical Mechanics 54
2. The Bost–Connes system 58
3. Noncommutative Geometry and Hilbert’s 12th problem 63
4. The GL2 system 65
5. Quadratic fields 72
Chapter 4. Noncommutative geometry at arithmetic infinity 81
1. Schottky uniformization 81
2. Dynamics and noncommutative geometry 88
3. Arithmetic infinity: archimedean primes 93
4. Arakelov geometry and hyperbolic geometry 96
5. Intermezzo: Quantum gravity and black holes 100
6. Dual graph and noncommutative geometry 104
7. Arithmetic varieties and L–factors 109
8. Archimedean cohomology 115
Bibliography 125
3
CHAPTER 1
Ouverture
Noncommutative geometry, as developed by Connes starting in the

early ’80s ([15], [17], [20]), extends the tools of ordinary geometry to
treat spaces that are quotients, for which the usual “ring of functions”,
defined as functions invariant with respect to the equivalence relation,
is too small to capture the information on the “inner structure” of
points in the quotient space. Typically, for such spaces functions on the
quotients are just constants, while a nontrivial ring of functions, which
remembers the structure of the equivalence relation, can be defined
using a noncommutative algebra of coordinates, analogous to the non-
commuting variables of quantum mechanics. These “quantum spaces”
are defined by extending the Gelfan’d–Naimark correspondence
X loc.comp. Hausdorff space ⇔ C0 (X) abelian C ∗ -algebra
by dropping the commutativity hypothesis in the right hand side. The
correspondence then becomes a definition of what is on the left hand
side: a noncommutative space.
Such quotients are abundant in nature. They arise, for instance,
from foliations. Several recent results also show that noncommuta-
tive spaces arise naturally in number theory and arithmetic geometry.
The first instance of such connections between noncommutative geom-
etry and number theory emerged in the work of Bost and Connes [8],
which exhibits a very interesting noncommutative space with remark-
able arithmetic properties related to class field theory. This reveals a
very useful dictionary that relates the phenomena of spontaneous sym-
metry breaking in quantum statistical mechanics to the mathematics of
Galois theory. This space can be viewed as the space of 1-dimensional
Q-lattices up to scale, modulo the equivalence relation of commensu-
rability (cf. [29]). This space is closely related to the noncommutative
space used by by Connes to obtain a spectral realization of the zeros
of the Riemann zeta function, [22]. In fact, this is again the space
of commensurability classes of 1-dimensional Q-lattices, but with the
scale factor also taken into account.
More recently, other results that point to deep connections between
noncommutative geometry and number theory appeared in the work
5
6 1. OUVERTURE
of Connes and Moscovici [33] [34] on the modular Hecke algebras.

This shows that the Rankin–Cohen brackets, an important algebraic
structure on modular forms [94], have a natural interpretation in the
language of noncommutative geometry, in terms of the Hopf algebra of
the transverse geometry of codimension one foliations. The modular
Hecke algebras, which naturally combine products and action of Hecle
operators on modular forms, can be viewed as the “holomorphic part”
of the algebra of coordinates on the space of commensurability classes
of 2-dimensional Q-lattices constructed in joint work of Connes and
the author [29].
Cases of occurrences of interesting number theory within noncom-
mutative geometry can be found in the classification of noncommuta-
tive three spheres by Connes and Dubois–Violette [26] [27]. Here the
corresponding moduli space has a ramified cover by a noncommutative
nilmanifold, where the noncommutative analog of the Jacobian of this
covering map is expressed naturally in terms of the ninth power of the
Dedekind eta function. Another such case occurs in Connes’ calcula-
tion [24] of the explicit cyclic cohomology Chern character of a spectral
triple on SUq (2) defined by Chakraborty and Pal [12].
Other instances of noncommutative spaces that arise in the context
of number theory and arithmetic geometry can be found in the non-
commutative compactification of modular curves of [25], [70]. This
noncommutative space is again related to the noncommutative geom-
etry of Q-lattices. In fact, it can be seen as a stratum in the com-
pactification of the space of commensurability classes of 2-dimensional
Q-lattices (cf. [29]).
Another context in which noncommutative geometry provides a use-
ful tool for arithmetic geometry is in the description of the totally
degenerate fibers at “arithmetic infinity” of arithmetic varieties over
number fields, analyzed in joint work of the author with Katia Consani
([36], [37], [38], [39]).
The present text is based on a series of lectures given by the author
at Vanderbilt University in May 2004, as well as on previous series of
lectures given at the Fields Institute in Toronto (2002), at the Univer-
sity of Nottingham (2003), and at CIRM in Luminy (2004).
The main focus of the lectures is the noncommutative geometry
of modular curves (following [70]) and of the archimedean fibers of
arithmetic varieties (following [36]). A chapter on the noncommutative
space of commensurability classes of 2-dimensional Q-lattices is also
included (following [29]). The text reflects very closely the style of
the lectures. In particular, we have tried more to convey the general
picture than the details of the proofs of the specific results. Though
1. THE NCG DICTIONARY 7
many proofs have not been included in the text, the reader will find
references to the relevant literature, where complete proofs are provided
(in particular [29], [36], and [70]).
More explicitly, the text is organized as follows:
• We start by recalling a few preliminary notions of noncommu-
tative geometry (following [20]).
• The second chapter describes how various arithmetic proper-
ties of modular curves can be seen by their “noncommutative
boundary”. This part is based on the joint work of Yuri Manin
and the author. The main references are [70], [71], [72].
• We review briefly the work of Connes and the author [29] on
the noncommutative geometry of commensurability classes of
Q-lattices. The relation of the noncommutative space of com-
mensurability classes of Q-lattices to the Hilbert 12th prob-
lem of explicit class field theory is based on ongoing work of
Connes, Ramachandran and the author [31], on the original
work of Bost and Connes [8] and on Manin’s real multiplica-
tion program [62] [63].
• The noncommutative geometry of the fibers at “arithmetic in-
finity” of varieties over number fields is based on joint work of
Consani and the author, for which the references are [36], [37],
[38], [39], [40]. This chapter also contains a detailed account
of Manin’s formula for the Green function of Arakelov geom-
etry for arithmetic surfaces, based on [66], and a proposed
physical interpretation of this formula, as in [69].
1. The NCG dictionary

There is a dictionary (cf. [20]) relating concepts of ordinary geom-
etry to the corresponding counterparts in noncommutative geometry.
The entries can be arranged according to the finer structures considered
on the underlying space, roughly according to the following table.
measure theory von Neumann algebras
topology C ∗ –algebras
smooth structures smooth subalgebras
Riemannian geometry spectral triples
It is important to notice that, usually, the notions of noncommuta-

tive geometry are “richer” than the corresponding entries of the dictio-
nary on the commutative side. For instance, as Connes discovered, non-
commutative measure spaces (von Neumann algebras) come endowed
with a natural time evolution which is trivial in the commutative case.
8 1. OUVERTURE
Similarly, at the level of topology one often sees phenomena that are
closer to rigid analytic geometry. This is the case, for instance, with the
noncommutative tori Tθ , which already at the C ∗ -algebra level exhibit
moduli that behave much like moduli of one-dimensional complex tori
(elliptic curves) in the commutative case.
In the context we are going to discuss this richer structure of non-
commutative spaces is crucial, as it permits us to use tools like C ∗ -
algebras (topology) to study the properties more rigid spaces like alge-
braic or arithmetic varieties.
2. Noncommutative spaces
The way to assign the algebra of coordinates to a quotient space
X = Y / ∼ can be explained in a short slogan as follows:
• Functions on Y with f (a) = f (b) for a ∼ b. Poor!
• Functions fab on the graph of the equivalence relation. Good!
The second description leads to a noncommutative algebra, as the
product, determined by the groupoid law of the equivalence relation,
has the form of a convolution product (like the product of matrices).
For sufficiently nice quotients, even though the two notions are
not the same, they are related by Morita equivalence, which is the
suitable notion of “isomorphism” between noncommutative spaces. For
more general quotients, however, the two notions truly differ and the
second one is the only one that allows one to continue to make sense
of geometry on the quotient space.
A very simple example illustrating the above situation is the fol-
lowing (cf. [23]). Consider the topological space Y = [0, 1] × {0, 1}
with the equivalence relation (x, 0) ∼ (x, 1) for x ∈ (0, 1). By the first
method one only obtains constant functions C, while by the second
method one obtains
{f ∈ C([0, 1]) ⊗ M2 (C) : f (0) and f (1) diagonal }
which is an interesting nontrivial algebra.
The idea of preserving the information on the structure of the equiv-
alence relation in the description of quotient spaces has analogs in
Grothendieck’s theory of stacks in algebraic geometry.
2.1. Morita equivalence. In noncommutative geometry, isomor-
phisms of C ∗ -algebras are too restrictive to provide a good notion of
isomorphisms of noncommutative spaces. The correct notion is pro-
vided by Morita equivalence of C ∗ -algebras.
We have equivalent C∗ -algebras A1 ∼ A2 if there exists a bimodule
M, which is a right Hilbert A1 module with an A1 -valued inner product
3. SPECTRAL TRIPLES 9
h·, ·iA1 , and a left Hilbert A2 -module with an A2 -valued inner product
h·, ·iA2 , such that we have:
• We obtain all Ai as the closure of the span of
{hξ1 , ξ2 iAi : ξ1 , ξ2 ∈ M}.
• ∀ξ1 , ξ2 , ξ3 ∈ M we have
hξ1 , ξ2 iA1 ξ3 = ξ1 hξ2 , ξ3 iA2 .
• A1 and A2 act on M by bounded operators,
ha2 ξ, a2 ξiA1 ≤ ka2 k2 hξ, ξiA1 ha1 ξ, a1 ξiA2 ≤ ka1 k2 hξ, ξiA2
for all a1 ∈ A1 , a2 ∈ A2 , ξ ∈ M.
This notion of equivalence roughly means that one can transfer
modules back and forth between the two algebras.
2.2. The tools of noncommutative geometry. Once one iden-
tifies in a specific problem a space that, by its nature of quotient of the
type described above, is best described as a noncommutative space,
there is a large set of well developed techniques that one can use to
compute invariants and extract essential information from the geome-
try. The following is a list of some such techniques, some of which will
make their appearance in the cases treated in these notes.
• Topological invariants: K-theory
• Hochschild and cyclic cohomology
• Homotopy quotients, assembly map (Baum-Connes)
• Metric structure: Dirac operator, spectral triples
• Characteristic classes, zeta functions
We will recall the necessary notions when needed. We now begin
by taking a closer look at the analog in the noncommutative world of
Riemannian geometry, which is provided by Connes’ notion of spectral
triples.
3. Spectral triples
Spectral triples are a crucial notion in noncommutative geometry.
They provide a powerful and flexible generalization of the classical
structure of a Riemannian manifold. The two notions agree on a com-
mutative space. In the usual context of Riemannian geometry, the
definition of the infinitesimal element ds on a smooth spin manifold
can be expressed in terms of the inverse of the classical Dirac opera-
tor D. This is the key remark that motivates the theory of spectral
triples. In particular, the geodesic distance between two points on the
manifold is defined in terms of D −1 (cf. [20] §VI). The spectral triple
10 1. OUVERTURE
(A, H, D) that describes a classical Riemannian spin manifold is given

by the algebra A of complex valued smooth functions on the manifold,
the Hilbert space H of square integrable spinor sections, and the clas-
sical Dirac operator D. These data determine completely and uniquely
the Riemannian geometry on the manifold. It turns out that, when
expressed in this form, the notion of spectral triple extends to more
general non-commutative spaces, where the data (A, H, D) consist of
a C∗ -algebra A (or more generally of some smooth subalgebra of a C∗ -
algebra) with a representation in the algebra of bounded operators on
a separable Hilbert space H, and an operator D on H that verifies the
main properties of a Dirac operator.
We recall the basic setting of Connes theory of spectral triples. For
a more complete treatment see [20], [21], [32].
Definition 3.1. A spectral triple (A, H, D) consists of a C∗ -algebra

A with a representation
ρ : A → B(H)
in the algebra of bounded operators on a separable Hilbert space H, and
an operator D (called the Dirac operator) on H, which satisfies the
following properties:
(1) D is self–adjoint.
(2) For all λ ∈/ R, the resolvent (D − λ)−1 is a compact operator
on H.
(3) The commutator [D, ρ(a)] is a bounded operator on H, for all
a ∈ A0 ⊂ A, a dense involutive subalgebra of A.
The property 2. of Definition 3.1 can be regarded as a general-
ization of the ellipticity property of the standard Dirac operator on a
compact manifold. In the case of ordinary manifolds, we can consider
as subalgebra A0 the algebra of smooth functions, as a subalgebra of
the commutative C∗ -algebra of continuous functions. In fact, in the
classical case of Riemannian manifolds, property 3. is equivalent the
Lipschitz condition, hence it is satisfied by a larger class than that of
smooth functions.
Thus, the basic geometric structure encoded by the theory of spec-
tral triples is Riemannian geometry, but in more refined cases, such
as Kähler geometry, the additional structure can be easily encoded as
additional symmetries. We will see, for instance, a case (cf. [36] [40])
where the algebra involves the action of the Lefschetz operator of a
compact Kähler manifold, hence it encodes the information (at the
cohomological level) on the Kähler form.
Since we are mostly interested in the relations between noncom-

mutative geometry and arithmetic geometry and number theory, an
especially interesting feature of spectral triples is that they have an
associated family of zeta functions and a theory of volumes and inte-
gration, which is related to special values of these zeta functions. (The
following treatment is based on [20], [21].)
3.1. Volume form. A spectral triple (A, H, D) is said to be of

dimension n, or n–summable if the operator |D|−n is an infinitesimal
of order one, which means that the eigenvalues λk (|D|−n ) satisfy the
estimate λk (|D|−n ) = O(k −1 ).
For a positive compact operator T such that
k−1
X
λj (T ) = O(log k),
j=0
the Dixmier trace Trω (T ) is the coefficient of this logarithmic diver-

gence, namely
k
1 X
(1.1) Trω (T ) = lim λj (T ).
ω log k
j=1
Here the notation limω takes into account the fact that the sequence
k
1 X
S(k, T ) := λj (T )
log k j=1
is bounded though possibly non-convergent. For this reason, the usual

notion of limit is replaced by a choice of a linear form limω on the
set of bounded sequences satisfying suitable conditions that extend
analogous properties of the limit. When the sequence S(k, T ) converges
(1.1) is just the ordinary limit Trω (T ) = limk→∞ S(k, T ). So defined,
the Dixmier trace (1.1) extends to any compact operator that is an
infinitesimal of order one, since any such operator is the difference of
two positive ones. The operators for which the Dixmier trace does
not depend on the choice of the linear form limω are called measurable
operators.
On a non-commutative space the operator |D|−n generalizes the
notion of a volume form. The volume is defined as
(1.2) V = Trω (|D|−n ).
More generally, consider the algebra Ã generated by A and [D, A].
Then, for a ∈ Ã, integration with respect to the volume form |D|−n is
12 1. OUVERTURE
defined as
1
Z
(1.3) Trω (a|D|−n ).
a :=
V
The usual notion of integration on a Riemannian spin manifold M
can be recovered in this context (cf. [20], [58]) through the formula (n
even): Z
f dv = 2n−[n/2]−1 π n/2 nΓ(n/2) Trω (f |D|−n ).

M
Here D is the classical Dirac operator on M associated to the metric
that determines the volume form dv, and f in the right hand side is
regarded as the multiplication operator acting on the Hilbert space of
square integrable spinors on M .
3.2. Zeta functions. An important function associated to the
Dirac operator D of a spectral triple (A, H, D) is its zeta function
X
(1.4) ζD (z) := Tr(|D|−z ) = Tr(Π(λ, |D|))λ−z ,
λ
where Π(λ, |D|) denotes the orthogonal projection on the eigenspace

E(λ, |D|).
An important result in the theory of spectral triples ([20] §IV
Proposition 4) relates the volume (1.2) with the residue of the zeta
function (1.4) at s = 1 through the formula
(1.5) V = lim (s − 1)ζD (s) = Ress=1 Tr(|D|−s ).
s→1+
There is a family of zeta functions associated to a spectral triple

(A, H, D), to which (1.4) belongs. For an operator a ∈ Ã, we can
define the zeta functions
X
(1.6) ζa,D (z) := Tr(a|D|−z ) = Tr(a Π(λ, |D|))λ−z
λ
and
X
(1.7) ζa,D (s, z) := Tr(a Π(λ, |D|))(s − λ)−z .
λ
These zeta functions are related to the heat kernel e−t|D| by Mellin
transform
Z ∞
1
(1.8) ζa,D (z) = tz−1 Tr(a e−t|D| ) dt
Γ(z) 0
where
X
(1.9) Tr(a e−t|D| ) = Tr(a Π(λ, |D|))e−tλ =: θa,D (t).
λ
Similarly,
∞
1
Z
(1.10) ζa,D (s, z) = θa,D,s (t) tz−1 dt
Γ(z) 0
with
X
(1.11) θa,D,s (t) := Tr(a Π(λ, |D|))e(s−λ)t .
λ
Under suitable hypothesis on the asymptotic expansion of (1.11) (cf.

Theorem 2.7-2.8 of [65] §2), the functions (1.6) and (1.7) admit a
unique analytic continuation (cf. [32]) and there is an associated reg-
ularized determinant in the sense of Ray–Singer (cf. [84]):

d
(1.12) det (s) := exp − ζa,D (s, z)|z=0 .
∞ a,D dz
The family of zeta functions (1.6) also provides a refined notion of
dimension for a spectral triple (A, H, D), called the dimension spec-
trum. This is a subset Σ = Σ(A, H, D) in C with the property that
all the zeta functions (1.6), as a varies in Ã, extend holomorphically
to C \ Σ.
Examples of spectral triples with dimension spectrum not contained
in the real line can be constructed out of Cantor sets.
3.3. Index map. The data of a spectral triple determine an index
map. In fact, the self adjoint
√ operator D has a polar decomposition,
D = F |D|, where |D| = D 2 is a positive operator and F is a sign
operator, i.e. F 2 = I. Following [20] and [21] one defines a cyclic
cocycle
(1.13) τ (a0 , a1 , . . . , an ) = Tr(a0 [F, a1 ] · · · [F, an ]),
where n is the dimension of the spectral triple. For n even Tr should
be replaced by a super trace, as usual in index theory. This cocycle
pairs with the K-groups of the algebra A (with K0 in the even case and
with K1 in the odd case), and defines a class τ in the cyclic cohomology
HC n (A) of A. The class τ is called the Chern character of the spectral
triple (A, H, D).
In the case when the spectral triple (A, H, D) has discrete dimen-
sion spectrum Σ, there is a local formula for the cyclic cohomology
Chern character (1.13), analogous to the local formula for the index
in the commutative case. This is obtained by producing Hochschild
representatives (cf. [21], [32])
(1.14) ϕ(a0 , a1 , . . . , an ) = Trω (a0 [D, a1 ] · · · [D, an ] |D|−n ).
14 1. OUVERTURE
3.4. Infinite dimensional geometries. The main difficulty in

constructing specific examples of spectral triples is to produce an op-
erator D that at the same time has bounded commutators with the
elements of A and produces a non-trivial index map.
It sometimes happens that a noncommutative space does not admit
a finitely summable spectral triple, namely one such that the operator
|D|−p is of trace class for some p > 0. Obstructions to the existence of
such spectral triples are analyzed in [19]. It is then useful to consider
a weaker notion, namely that of θ-summable spectral triples. These
have the property that the Dirac operator satisfies
2
(1.15) Tr(e−tD ) < ∞ ∀t > 0.
Such spectral triples should be thought of as “infinite dimensional non-
commutative geometries”.
We’ll see examples of spectral triples that are θ-summable but not
finitely summable, because of the growth rate of the multiplicities of
eigenvalues.
3.5. Spectral triples and Morita equivalences. If (A1 , H, D)

is a spectral triple, and we have a Morita equivalence A1 ∼ A2 im-
plemented by a bimodule M which is a finite projective right Hilbert
module over A1 , then we can transfer the spectral triple from A1 to
A2 .
First consider the A1 -bimodule Ω1D generated by
{a1 [D, b1 ] : a1 , b1 ∈ A1 }.
We define a connection
∇ : M → M ⊗A1 Ω1D
by requiring that
∇(ξa1 ) = (∇ξ)a1 + ξ ⊗ [D, a1 ],
∀ξ ∈ M, ∀a1 ∈ A1 and ∀ξ1 , ξ2 ∈ M. We also require that
hξ1 , ∇ξ2 iA1 − h∇ξ1 , ξ2 iA1 = [D, hξ1 , ξ2 iA1 ].
This induces a spectral triple (A2 , H̃, D̃) obtained as follows.
The Hilbert space is given by H̃ = M ⊗A1 H. The action takes the
form
a2 (ξ ⊗A1 x) := (a2 ξ) ⊗A1 x.
The Dirac operator is given by
D̃(ξ ⊗ x) = ξ ⊗ D(x) + (∇ξ)x.
Notice that we need a Hermitian connection ∇, because commuta-

tors [D, a] for a ∈ A1 are non-trivial, hence 1 ⊗ D would not be well
defined on H̃.
3.6. K-theory of C ∗ -algebras. The K-groups are important in-
variants of C ∗ -algebras that capture information on the topology of
non-commutative spaces, much like cohomology (or more appropriately
topological K-theory) captures information on the topology of ordinary
spaces. For a C ∗ -algebra A, we have:
3.6.1. K0 (A). This group is obtained by considering idempotents
(p = p) in matrix algebras over A. Notice that, for a C ∗ -algebra, it
2
is sufficient to consider projections (P 2 = P , P = P ∗ ). In fact, we

can always replace an idempotent p by a projection P = pp∗ (1 − (p −
p∗ )2 )−1 , preserving the von Neumann equivalence. Then we consider
P ' Q if and only if P = X ∗ X Q = XX ∗ , for X = partial isometry
(X = XX ∗ X), and we impose the stable equivalence: P ∼ Q, for
P ∈ Mn (A) and Q ∈ Mm (A), if and only if there exists R a projection
with P ⊕R ' Q⊕R. We define K0 (A)+ to be the monoid of projections
modulo these equivalences and we let K0 (A) be its Grothendieck group.
3.6.2. K1 (A). Consider the group GLn (A) of invertible elements
in the matrix algebra Mn (A), and GL0n (A) the identity component of
GLn (A). The morphism

a 0
GLn (A) → GLn+1 (A) a 7→
0 1
induces
GLn (A)/GL0n (A) → GLn+1 (A)/GL0n+1 (A).
The group K1 (A) is defined as the direct limit of these morphisms.
Notice that K1 (A) is abelian even if the GLn (A)/GL0n (A) are not.
In general, K-theory is not easy to compute. This lead to two
fundamental developments in noncommutative geometry. The first is
cyclic cohomology (cf. [17], [18], [20]), introduced by Connes in 1981,
which provides cycles that pair with K-theory, and a Chern charac-
ter. The second development is the geometrically defined K-theory of
Baum–Connes [6] and the assembly map
(1.16) µ : K ∗ (X, G) → K∗ (C0 (X) o G)
from the geometric to the analytic K-theory. Much work went, in
recent years, into exploring the range of validity of the Baum–Connes
conjecture, according to which the assembly map gives an isomorphism
between these two different notions of K-theory.
CHAPTER 2
Noncommutative modular curves
The results of this section are mostly based on the joint work of
Yuri Manin and the author [70], with additional material from [71]
and [72]. We include necessary preliminary notions on modular curves
and on noncommutative tori, respectively based on the papers [68] and
[15], [85].
We first recall some aspects of the classical theory of modular
curves. We will then show that these notions can be entirely recovered
and enriched through the analysis of a noncommutative space associ-
ated to a compactification of modular curves, which includes noncom-
mutative tori as possible degenerations of elliptic curves.
1. Modular curves
Let G be a finite index subgroup of the modular group Γ = PSL(2, Z),
and let XG denote the quotient
(2.1) XG := G\H2
where H2 is the 2-dimensional real hyperbolic plane, namely the upper

half plane {z ∈ C : =z > 0} with the metric ds2 = |dz|2 /(=z)2 .
Equivalently, we identify H2 with the Poincaré disk {z : |z| < 1} with
the metric ds2 = 4|dz|2 /(1 − |z|2 )2 .
We denote the quotient map by φ : H2 → XG .
Let P denote the coset space P := Γ/G. We can write the quotient
XG equivalently as
XG = Γ\(H2 × P).
A tessellation of the hyperbolic plane H2 by fundamental domains
for the action of PSL(2, Z) is illustrated in Figure 1. The action of
PSL(2, Z) by fractional linear transformations z 7→ az+b
cz+d
is usually
written in terms of the generators S : z 7→ −1/z and T : z 7→ z + 1
(inversion and translation). Equivalently, PSL(2, Z) can be identified
with the free product PSL(2, Z) ∼ = Z/2 ∗ Z/3, with generators σ, τ of
order two and three, respectively given by σ = S and τ = ST .
17
18 2. NONCOMMUTATIVE MODULAR CURVES
Figure 1. Fundamental domains for PSL(2, Z)
An example of finite index subgroups is given by the congruence

subgroups Γ0 (N ) ⊂ Γ, of matrices

a b
c d
with c ≡ 0 mod N . A fundamental domain for Γ0 (N ) is given by
F ∪N −1 k
k=0 ST (F ), where F is a fundamental domain for Γ as in Figure
11.
The quotient space XG has the structure of a non-compact Riemann
surface. This has a natural algebro-geometric compactification, which
consists of adding the cusp points (points at infinity). The cusp points
are identified with the quotient
(2.2) G\P1 (Q) ' Γ\(P1 (Q) × P).
Thus, we write the compactification as
XG := G\(H2 ∪ P1 (Q)) ' Γ\ (H2 ∪ P1 (Q)) × P .

(2.3)
The modular curve XΓ , for Γ = SL(2, Z) is the moduli space of
elliptic curves, with the point τ ∈ H2 parameterizing the lattice Λ =
Z ⊕ τ Z in C and the corresponding elliptic curve uniformized by
Eτ = C/Λ.
1Figures 1 and 2 are taken from Curt McMullen’s Gallery
1. MODULAR CURVES 19
The unique cusp point corresponds to the degeneration of the elliptic

curve to the cylinder C∗ , when τ → ∞ in the upper half plane.
The other modular curves, obtained as quotients XG by a con-
gruence subgroup, can also be interpreted as moduli spaces: they are
moduli spaces of elliptic curves with level structure. Namely, for ellip-
tic curves E = C/Λ, this is an additional information on the torsion
points N1 Λ/Λ ⊂ QΛ/Λ of some level N .
For instance, in the case of the principal congruence subgroups
Γ(N ) of matrices
a b
M= ∈Γ
c d
such that M ≡ Id mod N , points in the modular curve Γ(N )\H2
classify elliptic curves Et au = C/Λ, with Λ = Z + Zτ , together with
a basis {1/N, τ /N } for the torsion subgroup N1 Λ/Λ. The projection
XΓ(N ) → XΓ forgets the extra structure.
In the case of the groups Γ0 (N ), points in the quotient Γ0 (N )\H2
classify elliptic curves together with a cyclic subgroup of Eτ of order
N . This extra information is equivalent to an isogeny φ : Eτ → Eτ 0
where the cyclic group is Ker(φ). Recall that an isogeny is a morphism
φ : Eτ → Eτ 0 such that φ(0) = 0. These are implemented by the action
of GL+ 2
2 (Q) on H , namely Eτ and Eτ 0 are isogenous if and only if τ
and τ 0 in H2 are in the same orbit of GL+2 (Q).
1.1. Modular symbols. Given two points α, β ∈ H2 ∪ P1 (Q), a

real homology class {α, β}G ∈ H1 (XG , R) is defined as
Z β
{α, β}G : ω 7→ φ∗ (ω),
α
where ω are holomorphic differentials on XG , and the integration of the
pullback to H2 is along the geodesic arc connecting two cusps α and β
(cf. Figure 2).
The modular symbols {α, β}G satisfy the additivity and invariance
properties
{α, β}G + {β, γ}G = {α, γ}G ,
and
{gα, gβ}G = {α, β}G ,
for all g ∈ G.
Because of additivity, it is sufficient to consider modular symbols
of the form {0, α} with α ∈ Q,
X n
{0, α}G = − {gk (α) · 0, gk (α) · i∞}G ,
k=1
Figure 2. Geodesics between cusps define modular

symbols
where α has continued fraction expansion α = [a0 , . . . , an ], and

pk−1 (α) pk (α)
gk (α) = ,
qk−1 (α) qk (α)
with pk /qk the successive approximations, and pn /qn = α.
In the classical theory of the modular symbols of [68], [77] (cf. also
[50], [76]) cohomology classes obtained from cusp forms are evaluated
against relative homology classes given by modular symbols, namely,
given a cusp form Φ on H, obtained as pullback Φ = ϕ∗ (ω)/dz under
the quotient map ϕ : H → XG , we denote by ∆ω (s) the intersection
R g (i∞)
numbers ∆ω (s) = gss(0) Φ(z)dz, with gs G = s ∈ P. These intersection
numbers can be interpreted in terms of special values of L–functions
associated to the automorphic form which determines the cohomology
class.
We rephrase this in cohomological terms following [77]. We denote
by I and R̃ the elliptic points, namely the orbits I˜ = Γ · i and R̃ = Γ · ρ.
˜
We denote by I and R the image in XG of the elliptic points
I = G\I˜ R = G\R̃,
with ρ = eπi/3 . We use the notation
(2.4) HAB := H1 (XG \ A, B; Z).
These groups are related by the pairing

(2.5) HAB × HBA → Z.
The modular symbols {g(0), g(i∞)}, for gG ∈ P, define classes in
cusps
H . For σ and τ the generators of PSL(2, Z) with σ 2 = 1 and τ 3 = 1,
we set
(2.6) PI = hσi\P and PR = hτ i\P.
There is an isomorphism Z|P| ∼ R
= Hcusps∪I . Given the exact sequences
ι0
R π
0 → Hcusps → Hcusps R
→ Z|PR | → Z → 0
and
π
0 → Z|PI | → Hcusps∪I
R R
→I Hcusps → 0,
the image πI (x̃) ∈ Hcusps of an element x̃ = s∈P λs s in Z|P| ∼
R R
P
= Hcusps∪I
represents an element x ∈ Hcusps iff the image πR (πI (x̃)) = 0 in Z|PR | .
As proved in [77], for s = gG ∈ P, the intersection pairing • : H cusps ×
Hcusps → Z gives
{g(0), g(i∞)} • x = λs − λσs .
Thus, we write the intersection number as a function ∆x : P → R by
(2.7) ∆x (s) = λs − λσs ,
where x is given as above.
1.2. The modular complex. For x and y in H2 we denote by
hx, yi the oriented geodesic arc connecting them. Moreover, in the
decomposition PSL(2, Z) = Z/2 ∗ Z/3 as a free product, we denote by
σ the generator of Z/2 and by τ the generator of Z/3.
Definition 1.1. The modular complex is the cell complex defined
as follows.
• 0-cells: the cusps G\P1 (Q), and the elliptic points I and R.
• 1-cells: the oriented geodesic arcs G\(Γ · hi∞, ii) and G\(Γ ·
hi, ρi), where by Γ· we mean the orbit under the action of Γ.
• 2-cells: G\{Γ·E}, where E is the polygon with vertices {i, ρ, 1+
i, i∞} and sides the corresponding geodesic arcs.
• Boundary operator: ∂ : C2 → C1 is given by
gE 7→ ghi, ρi + ghρ, 1 + ii + gh1 + i, i∞i + ghi∞, ii,
for g ∈ Γ, and the boundary ∂ : C1 → C0 is given by
ghi∞, ii 7→ g(i) − g(i∞)
ghi, ρi 7→ g(ρ) − g(i).
Figure 3. The cell decomposition of the modular com-

plex
This gives a cell decomposition of H2 adapted to the action of

PSL(2, Z) and congruence subgroups, cf. Figure 3.
We have the following result [68].
Proposition 1.1. The modular complex computes the first homol-
ogy of XG :
Ker(∂ : C1 → C0 )
(2.8) H1 (XG ) ∼
= .
Im(∂ : C2 → C1 )
We can derive versions of the modular complex that compute rela-
tive homology.
Notice that we have Z[cusps] = C0 /Z[R ∪ I], hence the quotient
complex
∂ ∂˜
0 → C2 → C1 → Z[cusps] → 0,
with ∂˜ the induced boundary operator, computes the relative homology
H1 (XG , R ∪ I). The cycles are given by Z[P], as combinations of ele-
ments ghi, ρi, g ranging overPrepresentatives of P, and by the elements
⊕ag hg(i∞), g(i)i satisfying ag g(i∞) = 0. In fact, these can be rep-
resented as relative cycles in (XG , R ∪ I). Similarly, the subcomplex
∂
0 → Z[P] → Z[R ∪ I] → 0,
with Z[P] generated by the elements ghi, ρi, computes the homology
H1 (XG − cusps). The homology
H1 (XG − cusps, R ∪ I) ∼
= Z[P]
is generated by the relative cycles ghi, ρi, g ranging over representatives

of P.
cusps R∪I
With the notation (2.4), we consider the groups HR∪I , Hcusps ,
H cusps , and Hcusps .
We have a long exact sequence of relative homology
R∪I (β̃R ,β̃I )
(2.9) 0 → Hcusps → Hcusps → H0 (R) ⊕ H0 (I) → Z → 0,
R∪I
with Hcusps and Hcusps as above, and with
H0 (I) ∼
= Z[PI ], PI = hσi\P = G\I˜
H0 (R) ∼
= Z[PR ], PR = hτ i\P = G\R̃,
that is,
(2.10) 0 → Hcusps → Z[P] → Z[PR ] ⊕ Z[PI ] → Z → 0.
R∪I cusps
In the case of Hcusps and HR∪I the pairing (2.5) gives the identifi-
|P|
cation of Z[P] and Z , obtained by identifying the elements of P with
the corresponding delta functions. Thus, we can rewrite the sequence
(2.10) as
(βR ,βI )
(2.11) 0 → H cusps → Z|P| → Z|PI | ⊕ Z|PR | → Z → 0,
cusps ∼ |P|
where HR∪I =Z .
In order to understand more explicitly the map (βR , βI ) we give the
following equivalent algebraic formulation of the modular complex (cf.
[68] [77]).
R∪I
The homology group Hcusps = Z[P] is generated by the images in XG
of the geodesic segments gγ0 := ghi, ρi, with g ranging over a chosenset
of representatives of the coset space P.
cusps
We can identify (cf. [77]) the dual basis δs of HR∪I = Z|P| with
the images in XG of the paths gη0 , where for a chosen point z0 with
0 < Re(z0 ) < 1/2 and |z0 | > 1 the path η0 is given by the geodesic arcs
connecting ∞ to z0 , z0 to τ z0 , and τ z0 to 0. These satisfy
[gγ0 ] • [gη0 ] = 1
[gγ0 ] • [hη0 ] = 0,
for gG 6= hG, under the intersection pairing (2.5)
Then, in the exact sequence (2.11), the identification of H cusps with
Ker(βR , βI ) is obtained by the identification {g(0), g(i∞)}G 7→ gη0 ,
so that the relations imposed on the generators δs by the vanishing
under βI correspond to the relations δs ⊕ δσs (or δs if s = σs) and the
vanishing under βR gives another set of relations δs ⊕ δτ s ⊕ δτ 2 s (or δs
if s = τ s).
2. The noncommutative boundary of modular curves

The main idea that bridges between the algebro–geometric theory
of modular curves and noncommutative geometry consists of replacing
P1 (Q) in the classical compactification, which gives rise to a finite set
of cusps, with P1 (R). This substitution cannot be done naively, since
the quotient G\P1 (R) is ill behaved topologically, as G does not act
discretely on P1 (R).
When we regard the quotient Γ\P1 (R), or more generally Γ\(P1 (R)×
P), itself as a noncommutative space, we obtain a geometric object that
is rich enough to recover many important aspects of the classical theory
of modular curves. In particular, it makes sense to study in terms of
the geometry of such spaces the limiting behavior for certain arithmetic
invariants defined on modular curves when τ → θ ∈ R \ Q.
3. Modular interpretation: noncommutative elliptic curves
The boundary Γ\P1 (R) of the modular curve Γ\H2 , viewed itself as
a noncommutative space, continues to have a modular interpretation,
as observed originally by Connes–Douglas–Schwarz ([25]). In fact, we
can think of the quotients of S 1 by the action of rotations by an irra-
tional angle (that is, the noncommutative tori) as particular degener-
ations of the classical elliptic curves, which are “invisible” to ordinary
algebraic geometry. The quotient space Γ\P1 (R) classifies these non-
commutative tori up to Morita equivalence ([15], [85]) completes the
moduli space Γ\H2 of the classical elliptic curves. Thus, from a concep-
tual point of view it is reasonable to think of Γ\P1 (R) as the boundary
of Γ\H2 , when we allow points in this classical moduli space (that is,
elliptic curves) to have non-classical degenerations to noncommutative
tori.
Noncommutative tori are, in a sense, a prototype example of non-
commutative spaces, in as one can see there displayed the full range
of techniques of noncommutative geometry (cf. [15], [17]). As C∗ –
algebras, noncommutative tori are irrational rotation algebras. We re-
call some basic properties of noncommutative tori, which justify the
claim that these algebras behave like a noncommutative version of el-
liptic curves. We follow mostly [15] [20] and [85] for this material.
3.1. Irrational rotation and Kronecker foliation.
Definition 3.1. The irrational rotation algebra Aθ , for a given
θ ∈ R, is the universal C∗ -algebra C∗ (U, V ), generated by two unitary
operators U and V , subject to the commutation relation
(2.12) U V = e2πiθ V U.
3. MODULAR INTERPRETATION: NONCOMMUTATIVE ELLIPTIC CURVES25
The algebra Aθ can be realized as a subalgebra of bounded operators

on the Hilbert space H = L2 (S 1 ), with the circle S 1 ∼
= R/Z. For a given
θ ∈ R, we consider two operators, that act on a complete orthonormal
basis en of H as
(2.13) U en = en+1 , V en = e2πinθ en .
It is easy to check that these operators satisfy the commutation re-
lation (2.12), since, for any f ∈ H we have V U f (t) = U f (t − θ) =
e2πi(t−θ) f (t − θ), while U V f (t) = e2πit f (t − θ).
The irrational rotation algebra be described in more geometric
terms by the foliation on the usual commutative torus by lines with
irrational slope. On T 2 = R2 /Z2 one considers the foliation dx = θdy,
for x, y ∈ R/Z. The space of leaves is described as X = R/(Z + θZ) '
S 1 /θZ. This quotient is ill behaved as a classical topological space,
hence it cannot be well described by ordinary geometry.
A transversal to the foliation is given for instance by the choice
T = {y = 0}, T ∼ = S1 ∼= R/Z. Then the non-commutative torus is
obtained (cf. [15] [20]) as
(2.14) Aθ = {(fab ) a, b ∈ T in the same leaf }
where (fab ) is a power series b = n∈Z bn V n and each bn is an element
P
of the algebra C(S 1 ). The multiplication is given by
V hV −1 = h ◦ Rθ−1 ,
with
Rθ x = x + θ mod 1.
The algebra C(S ) is generated by U (t) = e2πit , hence we recover the
1
generating system (U, V ) with the relation

U V = e2πiθ V U.
What we have obtained through this description is an identification
of the irrational rotation algebra of Definition 3.1 with the crossed
product C ∗ -algebra
(2.15) Aθ = C(S 1 ) oRθ Z
representing the quotient S 1 /θZ as a non-commutative space.
3.2. Degenerations of elliptic curves. An elliptic curve Eτ over

C can be described as the quotient Eτ = C/(Z + τ Z) of the complex
plane by a 2-dimensional lattice Λ = Z+τ Z, where we can take =(τ ) >
0. It is also possible to describe the elliptic curve Eq , for q ∈ C∗ ,
q = exp(2πiτ ), |q| < 1, in terms of its Jacobi uniformization, namely


|q|

1

Figure 4. The fundamental domain for the Jacobi uni-
formization of the elliptic curve
as the quotient of C∗ by the action of the group generated by a single

hyperbolic element in P SL(2, C),
(2.16) Eq = C∗ /q Z .
The fundamental domain for the action of q Z is an annulus
{z ∈ C : |q| < z ≤ 1}
of radii 1 and |q|, and the identification of the two boundary circles is
obtained via the combination of scaling and rotation given by multi-
plication by q.
Now let us consider a degeneration where q → exp(2πiθ) ∈ S 1 ,
with θ ∈ R \ Q. We can say heuristically that in this degeneration the
elliptic curve becomes a non-commutative torus
Eq =⇒ Aθ ,
in the sense that, as we let q → exp(2πiθ), the annulus of the funda-
mental domain shrinks to a circle S 1 and we are left with a quotient
of S 1 by the infinite cyclic group generated by the irrational rotation
exp(2πiθ). Since this quotient is ill behaved as a classical quotient,
such degenerations do not admit a description within the context of
classical geometry. However, when we replace the quotient by the cor-
responding crossed product algebra C(S 1 ) oθ Z we find the irrational
rotation algebra of definition 3.1. Thus, we can consider such algebras
as non-commutative (degenerate) elliptic curves.
3. MODULAR INTERPRETATION: NONCOMMUTATIVE ELLIPTIC CURVES27
More precisely, when one considers degenerations of elliptic curves

Eτ = C∗ /q Z for q = e2πiτ , what one obtains in the limit is the suspen-
sion of a noncommutative torus. In fact, as the parameter q degenerates
to a point on the unit circle, q → e2πiθ , the “nice” quotient Eτ = C∗ /q Z
degenerates to the “bad” quotient Eθ = C∗ /e2πiτ Z , whose noncommu-
tative algebra of coordinates is Morita equivalent Aθ ⊗ C0 (R), with
ρ ∈ R the radial coordinate C∗ 3 z = eρ e2πis .
Because of the Thom isomorphism [16], the K-theory of the non-
commutative space Eθ = C0 (R2 ) o (Zθ + Z) satisfies
(2.17) K0 (Eθ ) = K1 (Aθ ) and K1 (Eθ ) = K0 (Aθ ),
which is again compatible with the identification of the Eθ (rather than

Aθ ) as degenerations of elliptic curves. In fact, for instance, the Hodge
filtration on the H 1 of an elliptic curve and the equivalence between the
elliptic curve and its Jacobian, have analogs for the noncommutative
torus Aθ in terms of the filtration on HC 0 induced by the inclusion of
K0 (cf. [17] p. 132–139, [23] §XIII), while by the Thom isomorphism,
these would again appear on the HC 1 in the case of the “noncommu-
tative elliptic curve” Eθ .
The point of view of degenerations is sometimes a useful guide-
line. For instance, one can study the limiting behavior of arithmetic
invariants defined on the parameter space of elliptic curves (on modu-
lar curves), in the limit when τ → θ ∈ R \ Q. An instance of this type
of result is the theory of limiting modular symbols of [70].
3.3. Morita equivalent NC tori. To extend the modular in-

terpretation of the quotient Γ\H2 as moduli of elliptic curves to the
noncommutative boundary Γ\P1 (R), one needs to check that points
in the same orbit of the action of the modular group PSL(2, Z) by
fractional linear transformations on P1 (R) define equivalent noncom-
mutative tori, where equivalence here is to be understood in the Morita
sense.
Connes showed in [15] (cf. also [85]) that the noncommutative tori
Aθ and A−1/θ are Morita equivalent. Geometrically in terms of the
Kronecker foliation and the description (2.14) of the corresponding al-
gebras, the Morita equivalence Aθ ' A−1/θ corresponds to changing
the choice of the transversal from T = {y = 0} to T 0 = {x = 0}.
In fact, all Morita equivalences arise in this way, by changing the
choice of the transversal of the foliation, so that Aθ and Aθ0 are Morita
equivalent if and only if θ ∼ θ 0 , under the action of PSL(2, Z).
Connes constructed in [15] explicit bimodules realizing the Morita

equivalences between non-commutative tori Aθ and Aθ0 with
aθ + b
θ0 = = gθ,
cθ + d
for
a b
g= ∈ Γ,
c d
by taking Mθ,θ0 to be the Schwartz space S(R × Z/c), with the right
action of Aθ
cθ + d
U f (x, u) = f x − ,u− 1
c
V f (x, u) = exp(2πi(x − ud/c))f (x, u)
and the left action of Aθ0

0 1
U f (x, u) = f x − , u − a
c

0 x u
V f (x, u) = exp 2πi − f (x, u).
cθ + d c
3.4. Other properties of NC elliptic curves. There are other
ways in which the irrational rotation algebra behaves much like an
elliptic curve, most notably the relation between the elliptic curve and
its Jacobian (cf. [17] and [23]) and some aspects of the theory of theta
functions, which we recall briefly.
The commutative torus T 2 = S 1 × S 1 is connected, hence the alge-
bra C(T 2 ) does not contain interesting projections. On the contrary,
the noncommutative tori Aθ contain a large family of nontrivial pro-
jections. Rieffel in [85] showed that, for a given θ irrational and for
all α ∈ (Z ⊕ Zθ) ∩ [0, 1], there exists a projection Pα in Aθ , with
Tr(Pα ) = α. A different construction of projections in Aθ , given by
Boca [7], has arithmetic relevance, in as these projections correspond
to the theta functions for noncommutative tori defined by Manin in
[64].
A method of constructing projections in C ∗ -algebras is based on
the following two steps (cf. [85] and [63]):
(1) Suppose given a bimodule A MB . If an element ξ ∈ A MB
1/2
admits an invertible ∗-invariant square root hξ, ξiB , then the
−1/2
element µ := ξhξ, ξiB satisfies µhµ, µiB = µ.
(2) Let µ ∈ A MB be a non-trivial element such that µhµ, µiB = µ.
Then the element P :=A hµ, µi is a projection.
4. LIMITING MODULAR SYMBOLS 29
In Boca’s construction, one obtains elements ξ from Gaussian ele-

ments in some Heisenberg modules, in such a way that the correspond-
ing hξ, ξiB is a quantum theta function in the sense of Manin [64]. An
introduction to the relation between the Heisenberg groups and the
theory of theta functions is given in the third volume of Mumford’s
Tata lectures on theta, [79].
4. Limiting modular symbols

We consider the action of the group Γ = PGL(2, Z) on the up-
per and lower half planes H± and the modular curves defined by the
quotient XG = G\H± , for G a finite index subgroup of Γ. Then the
noncommutative compactification of the modular curves is obtained by
extending the action of Γ on H± to the action on the full
P1 (C) = H± ∪ P1 (R),
so that we have
(2.18) X G = G\P1 (C) = Γ\(P1 (C) × P).
Due to the fact that Γ does not act discretely on P1 (R), the quotient
(2.18) makes sense as a noncommutative space
(2.19) C(P1 (C) × P) o Γ.
Here P1 (R) ⊂ P1 (C) is the limit set of the group Γ, namely the
set of accumulation points of orbits of elements of Γ on P1 (C). We
will see another instance of noncommutative geometry arising from the
action of a group of Möbius transformations of P1 (C) on its limit set,
in the context of the geometry at the archimedean primes of arithmetic
varieties.
4.1. Generalized Gauss shift and dynamics. We have de-
scribed the boundary of modular curves by the crossed product C ∗ -
algebra
(2.20) C(P1 (R) × P) o Γ.
We can also describe the quotient space Γ\(P1 (R) × P) in the following
equivalent way. If Γ = PGL(2, Z), then Γ-orbits in P1 (R) are the same
as equivalence classes of points of [0, 1] under the equivalence relation
x ∼T y ⇔ ∃n, m : T n x = T m y
where T x = 1/x−[1/x] is the classical Gauss shift of the continued frac-
tion expansion. Namely, the equivalence relation is that of having the
same tail of the continued fraction expansion (shift-tail equivalence).
A simple generalization of this classical result yields the following.
Lemma 4.1. Γ orbits in P1 (R) × P are the same as equivalence

classes of points in [0, 1] × P under the equivalence relation
(2.21) (x, s) ∼T (y, t) ⇔ ∃n, m : T n (x, s) = T m (y, t)
where T : [0, 1] × P → [0, 1] × P is the shift

1 1 −[1/x] 1
(2.22) T (x, s) = − , ·s
x x 1 0
generalizing the classical shift of the continued fraction expansion.
As a noncommutative space, the quotient by the equivalence re-
lation (2.21) is described by the C ∗ -algebra of the groupoid of the
equivalence relation
G([0, 1] × P, T ) = {((x, s), m − n, (y, t)) : T m (x, s) = T n (y, t)}
with objects G 0 = {((x, s), 0, (x, s))}.
In fact, for any T -invariant subset E ⊂ [0, 1] × P, we can consider
the equivalence relation (2.21). The corresponding groupoid C ∗ -algebra
C ∗ (G(E, T )) encodes the dynamical properties of the map T on E.
Geometrically, the equivalence relation (2.21) on [0, 1] × P is related
to the action of the geodesic flow on the horocycle foliation on the
modular curves.
4.2. Arithmetic of modular curves and noncommutative
boundary. The result of Lemma 4.1 shows that the properties of the
dynamical system T or (2.21) can be used to describe the geometry of
the noncommutative boundary of modular curves. There are various
types of results that can be obtained by this method ([70] [71]), which
we will discuss in the rest of this chapter.
(1) Using the properties of this dynamical system it is possible
to recover and enrich the theory of modular symbols on XG ,
by extending the notion of modular symbols from geodesics
connecting cusps to images of geodesics in H2 connecting ir-
rational points on the boundary P1 (R). In fact, the irrational
points of P1 (R) define limiting modular symbols. In the case
of quadratic irrationalities, these can be expressed in terms of
the classical modular symbols and recover the generators of
the homology of the classical compactification by cusps XG .
In the remaining cases, the limiting modular symbols vanishes
almost everywhere.
(2) It is possible to reinterpret Dirichlet series related to modu-
lar forms of weight 2 in terms of integrals on [0, 1] of certain
intersection numbers obtained from homology classes defined
in terms of the dynamical system. In fact, even when the

limiting modular symbol vanishes, it is possible to associate a
non-trivial cohomology class in XG to irrational points on the
boundary, in such a way that an average of the corresponding
intersection numbers give Mellin transforms of modular forms
of weight 2 on XG .
(3) The Selberg zeta function of the modular curve can be ex-
pressed as a Fredholm determinant of the Perron-Frobenius
operator associated to the dynamical system on the “bound-
ary”.
(4) Using the first formulation of the boundary as the noncom-
mutative space (2.20) we can obtain a canonical identification
of the modular complex with a sequence of K-groups of the
C ∗ -algebra. The resulting exact sequence for K-groups can be
interpreted, using the description (2.21) of the quotient space,
in terms of the Baum–Connes assembly map and the Thom
isomorphism.
All this shows that the noncommutative space C(P1 (R) × P) o Γ,
which we have so far considered as a boundary stratum of C(H2 × P) o
Γ, in fact contains a good part of the arithmetic information on the
classical modular curve itself. The fact that information on the “bulk
space” is stored in its boundary at infinity can be seen as an instance of
the physical principle of holography (bulk/boundary correspondence)
in string theory (cf. [69]). We will discuss the holography principle
more in details in relation to the geometry of the archimedean fibers
of arithmetic varieties.
4.3. Limiting modular symbols. Let γβ be an infinite geodesic

in the hyperbolic plane H with one end at i∞ and the other end at
β ∈ R r Q. Let x ∈ γβ be a fixed base point, τ be the geodesic arc
length, and y(τ ) be the point along γβ at a distance τ from x, towards
the end β. Let {x, y(τ )}G denote the homology class in XG determined
by the image of the geodesic arc hx, y(τ )i in H.
Definition 4.1. The limiting modular symbol is defined as
1
(2.23) {{∗, β}}G := lim {x, y(τ )}G ∈ H1 (XG , R),
τ
whenever such limit exists.
The limit (2.23) is independent of the choice of the initial point x
as well as of the choice of the geodesic in H ending at β, as discussed
in [70] (cf. Figure 5). We use the notation {{∗, β}}G as introduced in
[70], where ∗ in the first argument indicates the independence on the

choice of x, and the double brackets indicate the fact that the homology
class is computed as a limiting cycle.
4.3.1. Dynamics of continued fractions. As above, we consider on

[0, 1] × P the dynamical system
T : [0, 1] × P → [0, 1] × P

1 1 −[1/β] 1
(2.24) T (β, t) = − , ·t .
β β 1 0
This generalizes the classical shift map of the continued fraction

1 1
T : [0, 1] → [0, 1] T (x) = − .
x x
Recall the following basic notation regarding continued fraction ex-
pansion. Let k1 , . . . , kn be independent variables and, for n ≥ 1, let
1 Pn (k1 , . . . , kn )
[k1 , . . . , kn ] := 1 = .
k1 + k +... 1 Qn (k1 , . . . , kn )
2 kn
The Pn , Qn are polynomials with integral coefficients, which can be

calculated inductively from the relations
Qn+1 (k1 , . . . , kn , kn+1 ) = kn+1 Qn (k1 , . . . , kn ) + Qn−1 (k1 , . . . , kn−1 ),
Pn (k1 , . . . , kn ) = Qn−1 (k2 , . . . , kn ),
with Q−1 = 0, Q0 = 1. Thus, we obtain
[k1 , . . . , kn−1 , kn + xn ]

Pn−1 (k1 , . . . , kn−1 ) xn + Pn (k1 , . . . , kn ) Pn−1 Pn
= = (xn ),
Qn−1 (k1 , . . . , kn−1 ) xn + Qn (k1 , . . . , kn ) Qn−1 Qn
with the standard matrix notation for fractional linear transformations,

az + b a b
z 7→ = (z).
cz + d c d
If α ∈ (0, 1) is an irrational number, there is a unique sequence
of integers kn (α) ≥ 1 such that α is the limit of [ k1 (α), . . . , kn (α) ] as
n → ∞. Moreover, there is a unique sequence xn (α) ∈ (0, 1) such that
α = [ k1 (α), . . . , kn−1 (α), kn (α) + xn (α) ]
for each n ≥ 1. We obtain

0 1 0 1
α= ... (xn (α)).
1 k1 (α) 1 kn (α)
We set
pn (α) := Pn (k1 (α), . . . , kn (α)), qn (α) := Qn (k1 (α), . . . , kn (α))
so that pn (α)/qn (α) is the sequence of convergents to α. We also set

pn−1 (α) pn (α)
gn (α) := ∈ GL(2, Z).
qn−1 (α) qn (α)
Written in terms of the continued fraction expansion, the shift T is
given by
T : [k0 , k1 , k2 , . . .] 7→ [k1 , k2 , k3 , . . .].
The properties of the shift (2.24) can be used to extend the no-
tion of modular symbols to geodesics with irrational ends ([70]). Such
geodesics correspond to infinite geodesics on the modular curve XG ,
which exhibit a variety of interesting possible behaviors, from closed
geodesics to geodesics that approximate some limiting cycle, to geodesics
that wind around different homology class exhibiting a typically chaotic
behavior.
4.3.2. Lyapunov spectrum. A measure of how chaotic a dynamical

system is, or better of how fast nearby orbits tend to diverge, is given
by the Lyapunov exponent.
Definition 4.2. the Lyapunov exponent of T : [0, 1] → [0, 1] is
defined as
n−1
1 n 0 1 Y
(2.25) λ(β) := lim log |(T ) (β)| = lim log |T 0 (T k β)|.
n→∞ n n→∞ n
k=0
The function λ(β) is T –invariant. Moreover, in the case of the clas-

sical continued fraction shift T β = 1/β − [1/β] on [0, 1], the Lyapunov
exponent is given by
1
(2.26) λ(β) = 2 lim log qn (β),
n→∞ n
with qn (β) the successive denominators of the continued fraction ex-

pansion.
In particular, the Khintchine–Lévy theorem shows that, for almost
all β’s (with respect to the Lebesgue measure on [0, 1]) the limit (2.26)
is equal to
(2.27) λ(β) = π 2 /(6 log 2) =: λ0 .
There is, however, an exceptional set in [0, 1] of Hausdorff dimension

dimH = 1 but with Lebesgue measure zero where limit defining the
Lyapunov exponent does not exist.
As we will see later, in “good cases” the value λ(β) can be computed
from the spectrum of Perron–Frobenius operator of the shift T .
The Lyapunov spectrum is introduced (cf. [83]) by decomposing the
unit interval in level sets of the Lyapunov exponent λ(β) of (2.25). Let
Lc = {β ∈ [0, 1] |λ(β) = c ∈ R}. These sets provide a T –invariant
decomposition of the unit interval,
[
[0, 1] = Lc ∪ {β ∈ [0, 1] |λ(β) does not exist}.
c∈R
These level sets are uncountable dense T –invariant subsets of [0, 1], of
varying Hausdorff dimension [83]. The Lyapunov spectrum measures
how the Hausdorff dimension varies, as a function h(c) = dimH (Lc ).
4.3.3. Limiting modular symbols and iterated shifts. We introduce

a function ϕ : P → H cusps of the form
(2.28) ϕ(s) = {g(0), g(i∞)}G,
where g ∈ PGL(2, Z) (or PSL(2, Z)) is a representative of the coset
s ∈ P.
Then we can compute the limit (2.23) in the following way.
Theorem 4.2. Consider a fixed c ∈ R which corresponds to some
level set Lc of the Lyapunov exponent (2.26). Then, for all β ∈ Lc , the
limiting modular symbol (2.23) is computed by the limit
n
1 X
(2.29) lim ϕ ◦ T k (t0 ),
n→∞ cn
k=1
where T is the shift of (2.24) and t0 ∈ P.

Without loss of generality, one can consider the geodesic γβ in H×P
with one end at (i∞, t0 ) and the other at (β, t0 ), for ϕ(t0 ) = {0, i∞}G .
The argument given in §2.3 of [70] is based on the fact that one
can replace the homology class defined by the vertical geodesic with
one obtained by connecting the successive rational approximations to
β in the continued fraction expansion (Figure 5). Namely, one can
replace the path hx0 , yn i with the union of arcs
n
[
hx0 , y0 i ∪ hy0 , p0 /q0 i ∪ hpk−1 /qk−1 , pk /qk i ∪ hpn /qn , yn i
k=1
ζn
zn
z n+1
ζn+1
p p
α n-1
p
n+1 β n
q
q q n
n-1 n+1
Figure 5. Geodesics defining limiting modular symbols
representing the same homology class in H1 (X G , Z).

The result then follows by estimating the geodesic distance τ ∼
− log =y + O(1), as y(τ ) → β and
1 1
< =yn < ,
2qn qn+1 2qn qn−1
where yn is the intersection of γβ and the geodesic with ends at pn−1 (β)/qn−1 (β)
and pn (β)/qn (β).
The matrix gk−1 (β), with

pk−1 (β) pk (β)
gk (β) = ,
qk−1 (β) qk (β)
acts on points (β, t) ∈ [0, 1] × P as the k–th power of the shift map T
of (2.24). Thus, we obtain

k −1 −1 pk−1 (β) pk (β)
ϕ(T t0 ) = {gk (β) (0), gk (β) (i∞)}G = , .
qk−1 (β) qk (β) G
4.4. Ruelle and Perron–Frobenius operators. A general prin-
ciple in the theory of dynamical systems is that one can often study
the dynamical properties of a map T (e.g. ergodicity) via the spectral
theory of an associated operator. This allows one to employ techniques
of functional analysis and derive conclusions on dynamics.
In our case, to the shift map T of (2.24), we associate the operator
∞
X 1 1 0 1
(2.30) (Lσ f )(x, t) = f , ·t
(x + k)2σ x+k 1 k
k=1
depending on a complex parameter σ.
More generally, the Ruelle transfer operator of a map T is defined

as
X
(2.31) (Lσ f )(x, t) = exp(h(y, s)) f (y, s),
(y,s)∈T −1 (x,t)
where we take h(x, t) = −2σ log |T 0 (x, t)|. Clearly this operator is well
suited for capturing the dynamical properties of the map T as it is
defined as a weighted sum over preimages. On the other hand, there
is another operator that can be associated to a dynamical system and
which typically has better spectral properties, but is less clearly related
to the dynamics. The best circumstances are when these two agree (for
a particular value of the parameter). The other operator is the Perron–
Frobenius operator P. This is defined by the relation
Z Z
(2.32) f (g ◦ T ) dµLeb = (P f ) g dµLeb .
[0,1]×P [0,1]×P
In the case of the shift T of (2.24) we have in fact that

P = Lσ |σ=1 .
4.4.1. Spectral theory of L1 . In the case of the modular group G =

Γ, the spectral theory of the Perron Frobenius operator of the Gauss
shift was studied by D.Mayer [75]. More recently, Chang and Mayer
[13] extended the results to the case of congruence subgroups. A similar
approach is used in [70] to study the properties of the shift (2.24).
The Perron–Frobenius operator
∞
X 1 1 0 1
(L1 f )(x, s) = f , ·s
(x + k)2 x+k 1 k
k=1
for the shift (2.24) has the following properties.

Theorem 4.3. On a Banach space of holomorphic functions on
D × P continuous to boundary, with D = {z ∈ C | |z − 1| < 3/2}, under
the condition (irreducibility)

∞ 0 1 0 1
(2.33) P = ∪n=0 ... (t0 ) | k1 , . . . , kn ≥ 1 ,
1 k1 1 kn
the Perron–Frobenius operator L1 has the following properties:
• L1 is a nuclear operator, of trace class.
• L1 has top eigenvalue λ = 1. This eigenvalue is simple. The
corresponding eigenfunction is (up to normalization)
1
.
(1 + x)
• The rest of the spectrum of L1 is contained in a ball of radius

r < 1.
• There is a complete set of eigenfunctions.
The irreducibility condition (2.33) is satisfied by congruence sub-
groups.
For other T -invariant subsets E ⊂ [0, 1] × P, one can also consider
operators LE,σ and PE . When the set has the property that PE =
LE,δE , for δE = dimH E the Hausdorff dimension, one can use the
spectral theory of the operator PE to study the dynamical properties
of T .
The Lyapunov exponent can be read off the spectrum of the family
of operators LE,σ .
Lemma 4.4. Let λσ denote the top eigenvalue of LE,σ . Then
d
λ(β) = λσ |σ=dimH (E) µH a.e. in E
dσ
4.4.2. The Gauss problem. Let

(2.34) mn (x) := measure of {α ∈ (0, 1) | xn(α) ≤ x }
with α = [ a1 (α), . . . , an−1 (α), an (α) + xn (α) ].
The asymptotic behavior of the measures mn is a famous problem
on the distribution of continued fractions formulated by Gauss, who
conjectured that
? 1
(2.35) m(x) = lim mn (x) = log(1 + x).
n→∞ log 2
The convergence of (2.34) to (2.35) was only proved by R. Kuzmin in
1928. Other proofs were then given by P. Lévy (1929), K. Babenko
(1978) and D. Mayer (1991). The arguments used by Babenko and
Mayer use the spectral theory on the Perron–Frobenius operator. Of
these different arguments only the latter extends nicely to the case of
the generalized Gauss shift (2.24).
The Gauss problem can be formulated in terms of a recursive rela-
tion
∞
0 0
X 1 0 1
(2.36) mn+1 (x) = (L1 mn )(x) := 2
mn .
k=1
(x + k) x + k
The right hand side of (2.36) is the image of m0n under the Gauss–
Kuzmin operator. This is nothing but the Perron–Frobenius operator
for the shift T in the case of the group Γ = PGL(2, Z).
As a consequence of Theorem 4.3, one obtains the following result.
Theorem 4.5. Let

mn (x, t) := µLeb {(y, s) : xn (y) ≤ x, gn (y)−1 (s) = t}.
Then we have
• m0n (x, t) = Ln1 (1).
• The limit m(x, t) = limn→∞ mn (x, t) exists and equals
1
m(x, t) = log(1 + x).
|P| log 2
This shows that there exists a unique T -invariant measure on [0, 1]×
P. This is uniform in the discrete set P (the counting measure) and it
is the Gauss measure of the shift of the continued fraction expansion
on [0, 1].
4.5. Two theorems on limiting modular symbols. The result

on the T -invariant measure allows us to study the general behavior of
limiting modular symbols.
A special role is played by limiting modular symbols {{∗, β}} where
β is a quadratic irrationality in R r Q.
Theorem 4.6. Let g ∈ G be hyperbolic, with eigenvalue Λg corre-
sponding to the attracting fixed point αg+ . Let Λ(g) := | log Λg |, and let
` be the period of the continued fraction expansion of β = αg+ . Then
{0, g(0)}G
{{∗, β}}G =
Λ(g)
1 X̀ −1
= {gk (β) · 0, gk−1 (β) · i∞}G .
λ(β)`
k=1
This shows that, in this case, the limiting modular symbols are
linear combinations of classical modular symbols, with coefficients in
the field generated over Q by the Lyapunov exponents λ(β) of the
quadratic irrationalities.
In terms of geodesics on the modular curve, this is the case where
the geodesic has a limiting cycle given by the closed geodesic {0, g(0)}G
(Figure 6).
There is then the “generic case”, where, contrary to the previous
example, the geodesics wind around many different cycles in such a
way that the resulting homology class averages out to zero over long
distances (Figure 7).
Figure 6. Limiting modular symbols: limiting cycle
Figure 7. Limiting modular symbols: chaotic tangling

and untangling
Theorem 4.7. For a T -invariant E ⊂ [0, 1] × P, under the irre-

ducibility condition for E,
1
Rτ (β, s) := {x, y(τ )}G
τ
converges weakly to zero. Namely, for all f ∈ L1 (E, dµH ),
Z
lim Rτ (β, s) f (β, s) dµH (β, s) = 0.
τ →∞ E
This weak convergence can be improved to strong convergence µH (E)-

almost everywhere. Thus, the limiting modular symbol satisfies
{{∗, β}}G ≡ 0 a.e. on E.
This results depends upon the properties of the operator L1 and the
result on the T -invariant measure. In fact, to get the result on the weak
convergence ([70]) one notices that the limit (2.29) computing limiting
modular symbols can be evaluated in terms of a limit of iterates of the
Perron–Frobenius operator, by
n Z n Z
1 X k 1 X
lim f (ϕ ◦ T ) = lim (Lk1 f )ϕ,
n λ0 n n λ 0 n
k=1 [0,1]×P k=1 [0,1]×P
where λ(β) = λ0 a.e. in [0, 1].

By the convergence of Lk1 1 to the density h of the T -invariant mea-
sure and of Z
k
L1 f → f dµ h,
this yields
Z Z Z
{{∗, β}} f (β, t) dµ(β, t) = f dµ ϕ h dµ
[0,1]×P [0,1]×P [0,1]×P
Z
1 X
= f dµ ϕ(s).
2#P s∈P
P
It is then easy to check that the sum s∈P ϕ(s) = 0 since each term in
the sum changes sign under the action of the inversion σ ∈ PGL(2, Z)
with σ 2 = 1, but the sum is globally invariant under σ.
The argument extends to the case of other T -invariant subsets of
[0, 1] × P for which the corresponding Perron–Frobenius operator LE,δE
has analogous properties (cf. [71]). The weak convergence proved
by this type of argument can be improved to strong convergence by
applying the strong law of large numbers to the “random variables”
ϕk = ϕ ◦ T k (cf. [71] for details). The result effectively plays the role
of an ergodic theorem for the shift T on E.
5. Hecke eigenforms
A very important question is what happens to modular forms at the
noncommutative boundary of the modular curves. There is a variety
of phenomena in the theory of modular forms that hint to the fact that
a suitable class of “modular forms” survives on the noncommutative
boundary. Zagier introduced the term “quantum modular forms” to
denote this important and yet not sufficiently understood class of exam-
ples. Some aspects of modular forms “pushed” to the noncommutative
boundary were analyzed in [70], in the form of certain averages involv-
ing modular symbols and Dirichlet series related to modular forms of
weight 2. We recall here the main results in this case.
We shall now consider the case of congruence subgroups G = Γ0 (p),
for p a prime.
Let ω be a holomorphic differential (also called a differential of the
first kind) on the modular curve XΓ0 (p) = Γ0 (p)\H. Let Φ = ϕ∗ (ω)/dz
be the pullback under the projection ϕ : H → XΓ0 (p) .
Let Φ be an eigenfunction for all the Hecke operators
d−1
XX n/d b
Tn =
0 d
d|n b=0
with (p, n) = 1,
Tn Φ = cn Φ.
5. HECKE EIGENFORMS 41
Then the L-function of ω is given by

Z ∞
(2π)s
Lω (s) = − Φ(iy) y s−1 dy
(2πi)Γ(s) 0
∞
X
Lω (s) = cn n−s .
n=1
There are many very interesting arithmetic properties of the inte-
grals of such Hecke eigenforms on modular symbols (Manin [68]).
In particular, one has ([68]) the following relation between Lω and
modular symbols:
X Z
(2.37) (σ(n) − cn ) Lω (1) = ω,
d|n, b mod d {0,b/d}
P
with σ(n) = d, from which one obtains
d|n
−(2+t)
P P R
q:(q,p)=1 q q 0 ≤q,(q,q 0 )=1 {0,q 0 /q} ω
(2.38) h (p)
ζ (1+t)
(p)
Lω (2+t)
iR
i∞
= ζ (p) (2+t) − ζ (p) (2+t)2 0 φ∗ (ω)
(p)
Here Lω is the Mellin transform of Φ with omitted Euler p–factor,
(p)
and
P ζ(s) the Riemann zeta, with corresponding ζ (p) , that is, Lω (s) =
−s
and ζ (p) (s) = n:(n,p)=1 n−s .
P
n:(n,p)=1 cn n
It is therefore interesting to study the properties of integrals on

limiting modular symbols (cf. [70]) and extensions of (2.37) (2.38).
To this purpose, in [70] a suitable class of functions on the boundary
1
P (R) × P (or just [0, 1] × P) was introduced.
These are functions of the form
X∞
(2.39) `(f, β) = f (qk (β), qk−1 (β)).
k=1
Here f is a complex valued function defined on pairs of coprime integers
(q, q 0 ) with q ≥ q 0 ≥ 1 and with f (q, q 0 ) = O(q −) for some > 0,
and qk (β) are the successive denominators of the continued fraction
expansion of β ∈ [0, 1].
The reason for this choice of functions is given by the following
classical result of Lévy (1929).
Proposition 5.1. For a function f as above, we have
Z 1 X f (q, q 0 )
0
(2.40) `(f, α)dα = .
0 q(q + q 0 )
Sums and integrals here converge absolutely and uniformly.
We can interpret the summing over pairs of successive denominators

as a property that replaces modularity, when “pushed to the bound-
ary”.
Through this class of functions it is possible recast certain averages
related to modular symbols on XG , completely in terms of function
theory on the “boundary” space Γ\(P1 (R) × P). A typical result of this
sort is the following ([70]).
Upon choosing
q + q0
Z
0
f (q, q ) = 1+t ω,
q {0,q 0 /q}G
one obtains Z 1
`(f, β) dµLeb (β) =
0
" #Z
(p) i∞
ζ(1 + t) Lω (2 + t)
− (p) φ∗ (ω),
ζ(2 + t) ζ (2 + t)2 0
which expresses the Mellin transform of a Hecke eigenform in terms of

a boundary average.
Analogous results exist for different choices of f (q, q 0 ), for Eisenstein
series twisted by modular symbols, double logarithms at roots of unity,
etc.
In fact, this result also shows that, even when the limiting modular
symbol vanishes, one can still associate nontrivial homology classes to
the geodesics with irrational ends.
In fact, for the case of G = Γ0 (p) let us consider
∞
X qn+1 (β) + qn (β) qn (β)
(2.41) C(f, β) := 1+t
0, .
n=1
q n+1 (β) q n+1 (β) Γ 0 (p)
This defines, for <(t) > 0 and for almost all β, a homology class in
H1 (XG , cusps, R) such that the integral average
Z
`(f, β) = ω
C(f,β)
on [0, 1] recovers Mellin transforms of cusp forms by

" #Z
Z 1Z (p) i∞
ζ(1 + t) Lω (2 + t)
ω dµLeb (β) = − (p) 2
φ∗ (ω).
0 C(f,β) ζ(2 + t) ζ (2 + t) 0
An estimate
n
qn+1 (β) + qn (β) qn (β) −(5+2t)nλ(β)
X
0, ∼e ϕ ◦ T k (s)
qn+1 (β)1+t qn+1 (β) G k=1
6. SELBERG ZETA FUNCTION 43
shows that the series involved are absolutely convergent.
6. Selberg zeta function

Infinite geodesics on XG are the image image of infinite geodesics on
H2 ×P with ends on P1 (R)×P. Thus, they can be coded by (ω − , ω + , s),
with (ω ± , s) = endpoints in P1 (R) × {s}, s ∈ P, ω − ∈ (−∞, −1],
ω + ∈ [0, 1].
If ω ± are not cusp points, then we have infinite continued fraction
expansions of these endpoints,
ω + = [k0 , k1 , . . . kr , kr+1 , . . .]
,
ω − = [k−1 ; k−2 , . . . , k−n , k−n−1 , . . .]
hence the corresponding geodesics are coded by (ω, s), s ∈ P, where ω
is a doubly infinite sequence
ω = . . . k−(n+1) k−n . . . k−1 k0 k1 . . . kn . . .
The equivalence relation of passing to the quotient by the group
action is implemented by the invertible (double sided) shift:

+ − 1 1 1 −[1/ω + ] 1
T (ω , ω , s) = − , , ·s .
ω+ ω + ω − − [1/ω + ] 1 0
In particular, the closed geodesics in XG correspond to the case
where the endpoints ω ± are the attractive and repelling fixed points of
a hyperbolic element in the group. This corresponds to the case where
(ω, s) is a periodic point for the shift T .
The Selberg zeta function is a suitable “generating function” for
the closed geodesics in XG . It is given by
Y∞ Y
1 − e−(s+m)`(γ) ,

ZG (s) =
m=0 γ
where γ is a primitive closed geodesic of length `(γ).

This can also be expressed in terms of the generalized Gauss shift
by the following ([13], [70])
Theorem 6.1. The Selberg zeta function for G ⊂ PGL(2, Z) of
finite index is computed by the Fredholm determinant of the Ruelle
transfer operator (2.31),
ZG (s) = det(1 − Ls ).
An analogous result holds for finite index subgroups of SL(2, Z),
where one gets det(1 − L2s ). There are also generalizations to other T -
invariant sets E, where det(1−LE,s ) corresponds to a suitable “Selberg
zeta” that only counts certain classes of closed geodesics.
Figure 8. The modular complex and the tree of

PSL(2, Z)
7. The modular complex and K-theory of C∗ -algebras

When we interpret the “boundary” P1 (R) × P with the action of Γ
as the noncommutative space C∗ -algebra C(P1 (R)×P)oΓ, it is possible
to reinterpret some of the arithmetic properties of modular curves in
terms of these operator algebras. For instance, the modular symbols
determine certain elements in the K-theory of this C∗ -algebra, and the
modular complex and the exact sequence of relative homology (2.11)
can be identified canonically with the Pimsner six terms exact sequence
for K–theory of this C∗ -algebra.
For Γ = PSL(2, Z) and G ⊂ Γ a finite index subgroup, we use the
notation X := P1 (R) × P, and Γσ := hσi = Z/2 and Γτ := hτ i = Z/3.
The group Γ acts on the tree with edges T 1 ' Γ and vertices
T 0 ' Γ/hσi ∪ Γ/hτ i. This tree is embedded in H2 with vertices at
the elliptic points and geodesic edges, as the dual graph of the trian-
gulation (Figure 3) of the modular complex (Figure 8).
For a group acting on a tree there is a Pimsner six terms exact
sequence, computing the K-theory of the crossed product C ∗ -algebra
(for A = C(X))
α α̃
K0 (A) / K0 (A o Γσ ) ⊕ K0 (A o Γτ ) / K0 (A o Γ)
O

K1 (A o Γ) o K1 (A o Γσ ) ⊕ K1 (A o Γτ ) o K1 (A)
β̃ β
A direct inspection of the maps in this exact sequence shows that

it contains a subsequence, which is canonically isomorphic to the alge-
braic presentation of the modular complex, compatibly with the cover-
ing maps between modular curves for different congruence subgroups.
8. INTERMEZZO: CHAOTIC COSMOLOGY 45
Theorem 7.1. The algebraic presentation of the modular complex

0 → Hcusps → Z[P] → Z[PR ] ⊕ Z[PI ] → Z → 0
is canonically isomorphic to the sequence
0 → Ker(β) → Z[P] → Z[PI ] ⊕ Z[PR ] → Im(β̃) → 0.
A more geometric explanation for this exact sequence is the follow-
ing. The Lie group PSL(2, R) can be identified with the circle bundle
(unit sphere bundle) over the hyperbolic plane H = PSL(2, R)/P SO(2).
Thus, the unit tangent bundle to the modular curve XG is given by the
quotient T1 XG = G\PSL(2, R).
On the other hand, we can identify P1 (R) = B\PSL(2, R), the
quotient by the upper triangular matrices B, so that we have Morita
equivalent algebras
C(P1 (R)) o G ' C(T1 XG ) o B.
The right hand side is the semidirect product of two R-actions. Thus,
by the Thom isomorphism [16] we have
K∗ (C(T1 XG ) o B) ∼= K∗ (C(T1 XG ))
where K ∗ (T1 XG ) is related to K ∗ (XG ) by the Gysin exact sequence.
A construction of crucial importance in noncommutative geometry
is the classifying space for proper actions BG of Baum–Connes [6], the
homotopy quotient X ×G EG and the µ-map
µ : K ∗ (X ×G EG) → K∗ (C(X) o G)
relating the topologically defined K-theory to the analytic K-theory of
C ∗ -algebras.
In a sense, for noncommutative spaces that are obtained as quo-
tients by “bad” equivalence relations, the homotopy quotient is a “com-
mutative shadow” from which much crucial information on the topol-
ogy can be read (cf. [6], [18]).
In our case we have EG = H2 and the µ-map is an isomorphism
(the Baum–Connes conjecture holds). By retracting EG = H2 to the
tree of PSL(2, Z) and applying the Mayer–Vietoris sequence to vertices
and edges one obtains the Pimsner six terms exact sequence.
8. Intermezzo: Chaotic Cosmology

We digress momentarily on a topic from General Relativity, which
turns out to be closely related to the “noncommutative compactifica-
tion” of the modular curve XG with G the congruence subgroup Γ0 (2)
([70]).
An important problem in cosmology is understanding how anisotropy

in the early universe affects the long time evolution of space-time. This
problem is relevant to the study of the beginning of galaxy formation
and in relating the anisotropy of the background radiation to the ap-
pearance of the universe today.
We follow [5] for a brief summary of anisotropic and chaotic cosmol-
ogy. The simplest significant cosmological model that presents strong
anisotropic properties is given by the Kasner metric
(2.42) ds2 = −dt2 + t2p1 dx2 + t2p2 dy 2 + t2p3 dz 2 ,
1 = i p2i .
P P
where the exponents pi are constants satisfying pi =P
Notice that, for pi = d log gii /d log g, the first constraint i pi = 1 is
just the condition thatPlog gij = 2αδij + βij for a traceless β, while
the second constraint i p2i = 1 amounts to the condition that, in the
Einstein equations written in terms of α and βij ,
2 2 !
dα 8π 1 dβ ij
= T 00 +
dt 3 16π dt

−3α d 3α dβij 1
e e = 8π Tij − δij Tkk ,
dt dt 3
the term T 00 is negligible with respect to the term (dβij /dt)2 /16π,
which is the “effective energy density” of the anisotropic motion of
empty space, contributing together with a matter term to the Hubble
constant.
Around 1970, Belinsky, Khalatnikov, and Lifshitz introduced a cos-
mological model (mixmaster universe) where they allowed the expo-
nents pi of the Kasner metric to depend on a parameter u,
−u
p1 = 1+u+u2
1+u
(2.43) p2 = 1+u+u2
u(1+u)
p3 = 1+u+u2
Since for fixed u the model is given by a Kasner space-time, the behav-
ior of this universe can be approximated for certain large intervals of
time by a Kasner metric. In fact, the evolution is divided into Kasner
eras and each era into cycles. During each era the mixmaster universe
goes through a volume compression. Instead of resulting in a collapse,
as with the Kasner metric, high negative curvature develops resulting
in a bounce (transition to a new era) which starts again a behavior
approximated by a Kasner metric, but with a different value of the
parameter u. Within each era, most of the volume compression is due
to the scale factors along one of the space axes, while the other scale
factors alternate between phases of contraction and expansion. These
alternating phases separate cycles within each era.
More precisely, we are considering a metric
(2.44) ds2 = −dt2 + a(t)dx2 + b(t)dy 2 + c(t)dz 2 ,
generalizing the Kasner metric (2.42). We still require that (2.44)
admits SO(3) symmetry on the space-like hypersurfaces, and that it
dt
presents a singularity at t → 0. In terms of logarithmic time dΩ = − abc ,
the mixmaster universe model of Belinsky, Khalatnikov, and Lifshitz
admits a discretization with the following properties:
1. The time evolution is divided in Kasner eras [Ωn , Ωn+1 ], for n ∈ Z.
At the beginning of each era we have a corresponding discrete value of
the parameter un > 1 in (2.43).
2. Each era, where the parameter u decreases with growing Ω, can
be subdivided in cycles corresponding to the discrete steps un , un − 1,
un − 2, etc. A change u → u − 1 corresponds, after acting with the
permutation (12)(3) on the space coordinates, to changing u to −u,
hence replacing contraction with expansion and conversely. Within
each cycle the space-time metric is approximated by the Kasner metric
(2.42) with the exponents pi in (2.43) with a fixed value of u = un −k >
1.
3. An era ends when, after a number of cycles, the parameter un
falls in the range 0 < un < 1. Then the bouncing is given by the
transition u → 1/u which starts a new series of cycles with new Kasner
parameters and a permutation (1)(23) of the space axis, in order to
have again p1 < p2 < p3 as in (2.43).
Thus, the transition formula relating the values un and un+1 of two
successive Kasner eras is
1
un+1 = ,
un − [un ]
which is exactly the shift of the continued fraction expansion, T x =
1/x − [1/x], with xn+1 = T xn and un = 1/xn .
The previous observation is the key to a geometric description of
solutions of the mixmaster universe in terms of geodesics on a modular
curve (cf. [70]).
Theorem 8.1. Consider the modular curve XΓ0 (2) . Each infinite
geodesic γ on XΓ0 (2) not ending at cusps determines a mixmaster uni-
verse.
In fact, an infinite geodesic on XΓ0 (2) is the image under the quotient
map
πΓ : H2 × P → Γ\(H2 × P) ∼ = XG ,
where Γ = PGL(2, Z), G = Γ0 (2), and P = Γ/G ∼ = P1 (F2 ) = {0, 1, ∞},
of an infinite geodesic on H2 ×P with ends on P1 (R)×P. We consider the
elements of P1 (F2 ) as labels assigned to the three space axes, according
to the identification
0 = [0 : 1] 7→ z
(2.45) ∞ = [1 : 0] 7→ y
1 = [1 : 1] 7→ x.
As we have seen, geodesics can be coded in terms of data (ω − , ω + , s)
with the action of the shift T .
The data (ω, s) determine a mixmaster universe, with the kn =
[un ] = [1/xn ] in the Kasner eras, and with the transition between
subsequent Kasner eras given by xn+1 = T xn ∈ [0, 1] and by the per-
mutation of axes induced by the transformation

−kn 1
1 0
acting on P1 (F2 ). It is easy to verify that, in fact, this acts as the
permutation 0 7→ ∞, 1 7→ 1, ∞ 7→ 0, if kn is even, and 0 7→ ∞,
1 7→ 0, ∞ 7→ 1 if kn is odd, that is, after the identification (2.45), as
the permutation (1)(23) of the space axes (x, y, z), if kn is even, or as
the product of the permutations (12)(3) and (1)(23) if kn is odd. This
is precisely what is obtained in the mixmaster universe model by the
repeated series of cycles within a Kasner era followed by the transition
to the next era.
Data (ω, s) and T m (ω, s), m ∈ Z, determine the same solution up
to a different choice of the initial time.
There is an additional time-symmetry in this model of the evolu-
tion of mixmaster universes (cf. [5]). In fact, there is an additional
parameter δn in the system, which measures the initial amplitude of
each cycle. It is shown in [5] that this is governed by the evolution of
a parameter
δn+1 (1 + un )
vn =
1 − δn+1
which is subject to the transformation across cycles vn+1 = [un ] + vn−1 .
By setting yn = vn−1 we obtain
1
yn+1 = ,
(yn + [1/xn ])
hence we see that we can interpret the evolution determined by the data
(ω + , ω − , s) with the shift T either as giving the complete evolution of
the u-parameter towards and away from the cosmological singularity,
or as giving the simultaneous evolution of the two parameters (u, v)
while approaching the cosmological singularity.
This in turn determines the complete evolution of the parameters
(u, δ, Ω), where Ωn is the starting time of each cycle. For the explicit
recursion Ωn+1 = Ωn+1 (Ωn , xn , yn ) see [5].
The result of Theorem 4.5 on the unique T -invariant measure on
[0, 1] × P
δ(s) dx
(2.46) dµ(x, s) = ,
3 log(2) (1 + x)
implies that the alternation of the space axes is uniform over the time
evolution, namely the three axes provide the scale factor responsible
for volume compression with equal frequencies.
The Perron-Frobenius operator for the shift (2.24) yields the density
of the invariant measure (2.46) satisfying L1 f = f . The top eigenvalue
ησ of Lσ is related to the topological pressure by ησ = exp(P (σ)).
This can be estimated numerically, using the technique of [4] and the
integral kernel operator representation of §1.3 of [70].
The interpretation of solutions in terms of geodesics provides a nat-
ural way to single out and study certain special classes of solutions on
the basis of their geometric properties. Physically, such special classes
of solutions exhibit different behaviors approaching the cosmological
singularity.
For instance, the data (ω + , s) corresponding to an eventually peri-
odic sequence k0 k1 . . . km . . . of some period a0 . . . a` correspond to those
geodesics on XΓ0 (2) that asymptotically wind around the closed geo-
desic identified with the doubly infinite sequence . . . a0 . . . a` a0 . . . a` . . ..
Physically, these universes exhibit a pattern of cycles that recurs peri-
odically after a finite number of Kasner eras.
Another special class of solutions is given by the Hensley Cantor
sets (cf. [72]). These are the mixmaster universes for which there is a
fixed upper bound N to the number of cycles in each Kasner era.
In terms of the continued fraction description, these solutions cor-
respond to data (ω + , s) with ω + in the Hensley Cantor set EN ⊂ [0, 1].
The set EN is given by all points in [0, 1] with all digits in the continued
fraction expansion bounded by N (cf. [51]). In more geometric terms,
these correspond to geodesics on the modular curve XΓ0 (2) that wander
only a finite distance into a cusp.
On the set EN the Ruelle and Perron–Frobenius operators are given

by
N
X 1 1 0 1
(2.47) (Lσ,N f )(x, s) = f , ·s .
(x + k)2σ x+k 1 k
k=1
This operator still has a unique invariant measure µN , whose density

satisfies L2 dimH (EN ),N f = f , with
672 log N
dimH (EN ) = 1 − − + O(1/N 2 )
π2N π4N 2
the Hausdorff dimension of the Cantor set EN . Moreover, the top
eigenvalue ησ of Lσ,N is related to the Lyapunov exponent by
d
λ(x) = ησ |σ=dimH (EN ) ,
dσ
for µN -almost all x ∈ EN .
A consequence of this characterization of the time evolution in terms
of the dynamical system (2.24) is that we can study global properties
of suitable moduli spaces of mixmaster universes. For instance, the
moduli space for time evolutions of the u-parameter approaching the
cosmological singularity as Ω → ∞ is given by the quotient of [0, 1] × P
by the action of the shift T .
Similarly when we restrict to special classes of solutions, e.g. we
can consider the “moduli space” EN × P modulo the action of T . In
this example, the dynamical system T acting on EN × P is a subshift of
finite type, and the resulting nocommutative space is a Cuntz–Krieger
algebra [42]. This is an interesting class of C ∗ -algebras, generated by
partial isometries. Another such algebra will play a fundamental role
in the geometry at arithmetic infinity, which is the topic of the next
lecture.
In the example of the mixmaster dynamics on the Hensley Cantor
sets, the shift T is described by the Markov partition
AN = {((k, t), (`, s))|Uk,t ⊂ T (U`,s )},
for k, ` ∈ {1, . . . , N }, and s, t ∈ P, with sets Uk,t = Uk × {t}, where
Uk ⊂ EN are the clopen subsets where the local inverses of T are
defined,

1 1
Uk = , ∩ EN .
k+1 k
This Markov partition determines a matrix AN , with entries (AN )kt,`s =
1 if Uk,t ⊂ T (U`,s ) and zero otherwise (cf. Figure 9).
Figure 9. The directed graph of the matrix A2
Proposition 8.2. The 3 × 3 submatrices Ak` = (A(k,t),(`,s) )s,t∈P of

the matrix AN are of the form
  

 0 0 1


  0 1 0  ` = 2m
1 0 0


Ak` =  

 0 0 1
  1 0 0  ` = 2m + 1



0 1 0

This is just the condition Uk,t ⊂ T (U`,s ) written as

0 1
· s = t,
1 `
together with the fact that the transformation

0 1
1 `
acts on P1 (F2 ) as the permutation 0 7→ ∞, 1 7→ 1, ∞ 7→ 0, when ` is
even, and ∞ 7→ 0, 0 7→ 1, 1 7→ ∞ if ` is odd.
As a non-commutative space associated to the Markov partition
we consider the Cuntz–Krieger C∗ -algebra OAN (cf. [42]), which is the
universal C∗ -algebra generated by partial isometries Skt satisfying the
relations X
∗
Skt Skt = 1,
(k,t)
X
∗ ∗
S`s S`s = A(k,t),(`,s) Skt Skt .
(k,t)
Topological invariants of this C∗ -algebra reflect dynamical properties
of the shift T .
CHAPTER 3
Quantum statistical mechanics and Galois theory
This chapter is dedicated to a brief review of the joint work of Alain

Connes and the author on quantum statistical mechanics of Q-lattices
[29], [30]. It also includes a discussion of the relation to the explicit
class field theory problem for imaginary quadratic fields, following joint
work of Alain Connes, Niranjan Ramachandran, and the author [31],
as well as a brief review of some of Manin’s idea on real quadratic fields
and noncommutative tori with real multiplication [63].
The main notion that we consider in this chapter is that of commen-
surability classes of Q-lattices, according to the following definition.
Definition 0.1. A Q-lattice in Rn consists of a pair (Λ, φ) of a
lattice Λ ⊂ Rn (a cocompact free abelian subgroup of Rn of rank n)
together with a system of labels of its torsion points given by a homo-
morphism of abelian groups
φ : Qn /Zn −→ QΛ/Λ.
A Q-lattice is invertible if φ is an isomorphism.
Two Q-lattices are commensurable,
(Λ1 , φ1 ) ∼ (Λ2 , φ2 ),
iff QΛ1 = QΛ2 and
φ1 = φ 2 mod Λ1 + Λ2
One can check that, indeed, commensurability defines an equiva-
lence relation among Q-lattices. The interesting aspect of this equiva-
lence relation is that the quotient provides another typical case which is
best described through noncommutative geometry. For this it is crucial
that we consider non-invertible Q-lattices. In fact, most Q-lattices are
not commensurable to an invertible one, while two invertible Q-lattices
are commensurable if and only if they are equal.
In the following, we denote by Ln the set of commensurability
classes of Q-lattices in Rn . The topology on this space is encoded in a
noncommutative algebra of coordinates, namely a C ∗ -algebra C ∗ (Ln ).
We will discuss in some detail only the cases n = 1 (Bost–Connes sys-
tem) and n = 2 (the case considered in [29]). In these cases, we will
53
54 3. QUANTUM STATISTICAL MECHANICS AND GALOIS THEORY
general case
invertible case
Figure 1. Generic and invertible 2-dimensional Q-

lattices
look at commensurability classes of Q-lattices up to scaling, namely

the quotients L1 /R∗+ and L2 /C∗ . The corresponding C ∗ -algebras
C ∗ (L1 /R∗+ ) and C ∗ (L2 /C∗ )
are endowed with a natural time evolution (a 1-parameter family of
automorphisms). Thus, one can consider them as quantum statistical
mechanical systems and look for equilibrium states, depending on a
thermodynamic parameter β (inverse temperature). The interesting
connection to arithmetic arises from the fact that the action of sym-
metries of the system on equilibrium states at zero temperature can
be described in terms of Galois theory. In the 1-dimensional case of
Bost–Connes, this will happen via the class field theory of Q, namely
the Galois theory of the cyclotomic field Qcycl . In the two dimensional
case, the picture is more elaborate and involves the automorphisms of
the field of modular functions.
1. Quantum Statistical Mechanics

In classical statistical mechanics a state is a probability measure µ
on the phase space that assigns to each observable f an expectation
value, in the form of an average
Z
(3.1) f dµ.
1. QUANTUM STATISTICAL MECHANICS 55
In particular, for a Hamiltonian system, the macroscopic thermody-

namic behavior is described via the Gibbs canonical ensemble. This is
the measure
1
(3.2) dµG = e−βH dµLiouville ,
Z
R −βH
normalized by Z = e dµLiouville , with a thermodynamic parameter
β = (kT )−1 , for T the temperature and k the Boltzmann constant
(which we can set equal to one in suitable units).
A quantum statistical mechanical system consists of the data of an
algebra of observables (a C ∗ -algebra A), together with a time evolution
given as a 1-parameter family of automorphisms σt ∈ Aut(A). We refer
to the pair (A, σt ) as a C ∗ -dynamical system. These data describe the
microscopic quantum mechanical evolution of the system.
The macroscopic thermodynamical properties are encoded in the
equilibrium states of the system, depending on the inverse tempera-
ture β. While the Gibbs measure in the classical case is defined in
terms of the Hamiltonian and the symplectic structure on the phase
space, the notion of equilibrium state in the quantum statistical me-
chanical setting only depends on the algebra of observables and its
time evolution, and does not involve any additional structure like the
symplectic structure or the approximation by regions of finite volume.
We first need to recall the notion of states. These can be thought
of as probability measures on noncommutative spaces.
Definition 1.1. Given a unital C ∗ -algebra A, a state on A is a
linear functional ϕ : A → C satisfying normalization and positivity,
(3.3) ϕ(1) = 1, ϕ(a∗ a) ≥ 0.
When the C ∗ -algebra A is non unital, the condition ϕ(1) = 1 is replaced
by kϕk = 1 where
(3.4) kϕk := supx∈A,kxk≤1 |ϕ(x)| .
Such states are restrictions of states on the unital C ∗ -algebra Ã ob-

tained by adjoining a unit.
Before giving the general definition of equilibrium states via the
KMS condition, we can see equilibrium states in the simple case of a
system with finitely many quantum degrees of freedom. In this case,
the algebra of observables is the algebra of operators in a Hilbert space
H and the time evolution is given by σt (a) = eitH a e−itH , where H is
a positive self-adjoint operator such that exp(−βH) is trace class for
Im z = β
iβ F(t + iβ) = ϕ(σt(b)a)
F(t) = ϕ(aσt(b))
Im z = 0 0
Figure 2. The KMS condition
any β > 0. For such a system, the analog of (3.2) is

1
ϕ(a) = Tr a e−βH

(3.5) ∀a ∈ A,
Z
with the normalization factor Z = Tr(exp(−βH)).
The Kubo–Martin–Schwinger condition (KMS) (cf. [11], [47], [48])
describing equilibrium states of more general quantum statistical me-
chanical systems generalizes (3.5) beyond the range of temperatures
where exp(−βH) is trace class.
Definition 1.2. Given a C ∗ -dynamical system (A, σt ), a state ϕ
on A satisfies the KMS condition at inverse temperature 0 < β < ∞
iff, for all a, b ∈ A, there exists a function Fa,b (z) holomorphic on the
strip 0 < =(z) < β continuous to the boundary and bounded, such that
for all t ∈ R
(3.6) Fa,b (t) = ϕ(aσt (b)) and Fa,b (t + iβ) = ϕ(σt (b)a).
KMS∞ states are weak limits of KMSβ states as β → ∞,
(3.7) ϕ∞ (a) = lim ϕβ (a) ∀a ∈ A.
β→∞
Our definition of KMS∞ state is stronger than the one often adopted
in the literature, which simply uses the existence, for each a, b ∈ A, of
a bounded holomorphic function Fa,b (z) on the upper half plane such
that Fa,b (t) = ϕ(aσt (b)). It is easy to see that this notions, which we
1. QUANTUM STATISTICAL MECHANICS 57
simply call the ground states of the system, is in fact weaker than the
notion of KMS∞ states given in Definition 1.2. For example, in the
simplest case of the trivial time evolution, all states are ground states,
while only tracial states are KMS∞ states. Another advantage of our
definition is that for all 0 < β ≤ ∞, the KMSβ states form a Choquet
simplex and we can therefore consider the set Eβ of its extremal points.
These give a good notion of points for the underlying noncommutative
space. This will be especially useful in connection to an arithmetic
structure specified by an arithmetic subalgebra AQ of A on which one
evaluates the KMS∞ states. This will play a key role in the relation
between the symmetries of the system and the action of the Galois
group on states ϕ ∈ E∞ evaluated on AQ .
1.1. Symmetries. An important role in quantum statistical me-
chanics is played by symmetries. Typically, symmetries of the algebra
A compatible with the time evolution induce symmetries of the equi-
librium states Eβ at different temperatures. Especially important are
the phenomena of symmetry breaking. In such cases, there is a global
underlying group G of symmetries of the algebra A, but in certain
ranges of temperature the choice of an equilibrium state ϕ breaks the
symmetry to a smaller subgroup Gϕ = {g ∈ G : g ∗ ϕ = ϕ}, where
g ∗ denotes the induced action on states. Various systems can exhibit
one or more phase transitions, or none at all. A typical situation in
physical systems sees a unique KMS state for all values of the parame-
ter above a certain critical temperature (β < βc ). This corresponds to
a chaotic phase such as randomly distributed spins in a ferromagnet.
When the system cools down and reaches the critical temperature, the
unique equilibrium state branches off into a larger set KMSβ and the
symmetry is broken by the choice of an extremal state in Eβ .
We will see that the case of L1 /R∗+ gives rise to a system with a
single phase transition ([8]), while in the case of L2 /C∗ the system has
multiple phase transitions.
A very important point is that we need to consider both symmetries
by automorphisms and by endomorphisms.
Automorphisms: A subgroup G ⊂ Aut(A) is compatible with σt if
for all g ∈ G and for all t ∈ R we have gσt = σt g. There is then an
induced action of G on KMS states and in particular on the set Eβ . If
u is a unitary, acting on A by
Adu : a 7→ uau∗
and satisfying σt (u) = u, then we say that Adu is an inner automor-
phism of (A, σt ). Inner automorphisms act trivially on KMS states.
Endomorphisms: Let ρσt = σt ρ be a ∗-homomorphism. Consider

the idempotent e = ρ(1). If ϕ ∈ Eβ is a state such that ϕ(e) 6= 0, then
there is a well defined pullback ρ∗ ϕ,
1
(3.8) ρ∗ (ϕ) = ϕ ◦ ρ.
ϕ(e)
Let u be an isometry compatible with the time evolution by
(3.9) σt (u) = λit u λ > 0.
One has u∗ u = 1 and uu∗ = e. We say that Adu defined by a 7→ uau∗
is an inner endomorphism of (A, σt ). The condition (3.9) ensures that
(Adu)∗ ϕ is well defined according to (3.8) and the KMS condition shows
that the induced action of an inner endomorphism on KMS states is
trivial.
One needs to be especially careful in defining the action of endo-
morphisms by (3.8). In fact, there are cases where for KMS∞ states
one finds only ϕ(e) = 0, yet it is still possible to define an interesting
action of endomorphisms by a procedure of “warming up and cooling
down”. For this to work one needs sufficiently favorable conditions,
namely that the “warming up” map
Tr(πϕ (a) e−β H )
(3.10) Wβ (ϕ)(a) =
Tr( e−β H )
gives a homeomorphism Wβ : E∞ → Eβ for all β sufficiently large. One
can then define the action by
(3.11) (ρ∗ ϕ)(a) = lim (ρ∗ Wβ (ϕ)) (a),
β→∞
for all ϕ ∈ E∞ and all a ∈ A.
2. The Bost–Connes system

We give the following geometric point of view on the BC system,
following [31].
We use the notation Sh(G, X) := G(Q)\(G(Af ) × X) for Shimura
varieties (cf. [78]). Here Af = Ẑ ⊗ Q denotes the finite adèles of Q,
with Ẑ = limn Z/nZ. The simplest case is for G = GL1 , where we
←−
consider
(3.12) Sh(GL1 , {±1}) = GL1 (Q)\(GL1 (Af ) × {±1}) = Q∗+ \A∗f .
Lemma 2.1. The quotient (3.12) parameterizes the invertible 1-
dimensional Q-lattices up to scaling.
2. THE BOST–CONNES SYSTEM 59
In fact, first notice that a 1-dimensional Q-lattice can always be

written in the form
(3.13) (Λ, φ) = (λ Z, λ ρ)
for some λ > 0 and some
(3.14) ρ ∈ Hom(Q/Z, Q/Z) = lim Z/nZ = Ẑ.
←−
Thus, the set of 1-dimensional Q-lattices up to scaling can be identified
with Ẑ. The invertible lattices correspond to the elements of Ẑ∗ , which
in turn is identified with GL+1 (Q)\GL1 (Af ).
The quotient (3.12) can be thought of geometrically as the Shimura
variety associated to the cyclotomic tower (cf. [31]). This is the tower V
of arithmetic varieties over V = Spec(Z), with Vn = Spec(Z[ζn ]), where
ζn is a primitive n-th root of unity. The group of deck transformations
is AutV (Vn ) = GL1 (Z/nZ), so that the group of deck transformations
of the tower is the projective limit
(3.15) AutV (V) = lim AutV (Vn ) = GL1 (Ẑ).
←−
n
If, instead of invertible Q-lattices up to scale, we consider the set

of commensurability classes of all 1-dimensional Q-lattices up to scale,
we find a noncommutative version of the Shimura variety (3.12) and of
the cyclotomic tower. This is described as the quotient
(3.16) Sh(nc) (GL1 , {±1}) := GL1 (Q)\(Af ×{±1}) = GL1 (Q)\A· /R∗+ ,
where A· := Af × R∗ are the adèles of Q with invertible archimedean
component.
The corresponding noncommutative algebra of coordinates is given
by the crossed product (cf. [56])
(3.17) C0 (Af ) o Q∗+ .
One obtains an equivalent description of the noncommutative space
(3.16), starting from the description (3.13) of 1-dimensional Q-lattices.
The commensurability relation is implemented by the action of N× =
Z>0 by

f (n−1 ρ) ρ ∈ nẐ
(3.18) αn (f )(ρ) =
0 otherwise.
Thus, the noncommutative algebra of coordinates of the space of com-
mensurability classes of 1-dimensional Q-lattices up to scaling can be
identified with the semigroup crossed product
(3.19) A1 := C ∗ (Q/Z) o N× ,
...
...
Figure 3. Phasors with Z/6Z discretization
where we have identified

(3.20) C(Ẑ) ' C ∗ (Q/Z),
since Ẑ is the Pontrjagin dual of Q/Z. The algebra (3.19) is Morita
equivalent to the algebra (3.17) (cf. [56]).
The algebra (3.19) is isomorphic to the algebra considered in the
work of Bost–Connes [8], which was obtained there as a Hecke algebra
for the inclusion Γ0 ⊂ Γ of the pair of groups (Γ0 , Γ) = (PZ , PQ ), where
P is the ax + b group. These have the property that the left Γ0 orbits
of any γ ∈ Γ/Γ0 are finite (same for right orbits on the left coset). The
ratio of the lengths of left and right Γ0 orbits determines a canonical
time evolution σt on the algebra (cf. [8]).
The algebra of the Bost–Connes system has an explicit presenta-
tion in terms of two sets of operators. The first type consists of phase
operators e(r), parameterized by elements r ∈ Q/Z. These phase op-
erators can be represented on the Fock space generated by occupation
numbers |ni as the operators
(3.21) e(r)|ni = α(ζrn )|ni.
Here we denote by ζa/b = ζba the abstract roots of unity generating
Qcycl and by α : Qcycl ,→ C an embedding that identifies Qcycl with the
subfield of C generated by the concrete roots of unity.
These are the phase operators used in the theory of quantum optics
and optical coherence to model the phase quantum-mechanically (cf.
[60] [61]), as well as to model the phasors in quantum computing.
They are based on the choice of a certain scale N at which the phase is
discretized (Figure 3). Namely, the quantized optical phase is defined
2. THE BOST–CONNES SYSTEM 61
as a state
m
|θm,N i = e · vN ,
N +1
where vN is a superposition of occupation states
N
1 X
vN = |ni.
(N + 1)1/2 n=0
One needs then to ensure that the results are consistent over changes
of scale.
The other operators that generate the Bost–Connes algebra can be
thought of as implementing the changes of scales in the optical phases
in a consistent way. These operators are isometries µn parameterized
by positive integers n ∈ N× = Z>0 . The changes of scale are described
by the action of the µn on the e(r) by
1 X
(3.22) µn e(r)µ∗n = e(s).
n ns=r
In addition to this compatibility condition, the operators e(r) and µn
satisfy other simple relations.
These give a presentation of the algebra of the BC system of the
form ([8], [55]):
• µ∗n µn = 1, for all n ∈ N× ,
• µk µn = µkn , for all k, n ∈ N× ,
• e(0) = 1, e(r)∗ = e(−r), and e(r)e(s) = e(r + s), for all
r, s ∈ Q/Z,
• For all n ∈ N× and all r ∈ Q/Z, the relation (3.22) holds.
This shows that the Bost–Connes algebra is isomorphic to the algebra
A1 of (3.19).
In terms of this explicit presentation, the time evolution is of the
form
(3.23) σt (µn ) = nit µn , σt (e(r)) = e(r).
The space (3.16) can be compactified by replacing A· by A, as in
[22]. This gives the quotient
(3.24) Sh(nc) (GL1 , {±1}) = GL1 (Q)\A/R∗+ .
This compactification consists of adding the trivial lattice (with a pos-
sibly nontrivial Q-structure).
The dual space to (3.24), under the duality of type II and type
III factors introduced in Connes’ thesis, is a principal R∗+ -bundle over
(3.24), whose noncommutative algebra of coordinates is obtained from
the algebra of (3.24) by taking the crossed product by the time evolu-
tion σt . The space obtained this way is the space of adèle classes
(3.25) L1 = GL1 (Q)\A → GL1 (Q)\A/R∗+
that gives the spectral realization of zeros of the Riemann ζ function
in [22]. Passing to this dual space corresponds to considering commen-
surability classes of 1-dimensional Q-lattices (no longer up to scaling).
2.1. Structure of KMS states. The Bost–Connes algebra A1
has irreducible representations on the Hilbert space H = `2 (N× ). These
are parameterized by elements α ∈ Ẑ∗ = GL1 (Ẑ). Any such element
defines an embedding α : Qcycl ,→ C and the corresponding represen-
tation is of the form
πα (e(r)) k = α(ζrk ) k
(3.26)
πα (µn ) k = nk
The Hamiltonian implementing the time evolution (3.18) on H is
(3.27) H k = log k k
Thus, the partition function of the Bost–Connes system is the Riemann
zeta function
∞
−βH
X
(3.28) Z(β) = Tr e = k −β = ζ(β).
k=1
Bost and Connes showed in [8] that KMS states have the following
structure, with a phase transition at β = 1.
• In the range β ≤ 1 there is a unique KMSβ state. Its restriction
to Q[Q/Z] is of the form
Y 1 − pβ−1
ϕβ (e(a/b)) = b−β .
1 − p−1
p prime, p|b
• For 1 < β ≤ ∞ the set of extremal KMS states Eβ can be

identified with Ẑ∗ . It has a free and transitive action of this
group induced by an action on A by automorphisms. The
extremal KMSβ state corresponding to α ∈ Ẑ∗ is of the form
1
Tr πα (x) e−βH .

(3.29) ϕβ,α (x) =
ζ(β)
• At β = ∞ the Galois group Gal(Qcycl /Q) acts on the values of
states ϕ ∈ E∞ on an arithmetic subalgebra A1,Q ⊂ A1 . These
have the property that ϕ(AQ ) ⊂ Qcycl and that the isomor-
∼
=
phism (class field theory isomorphism) θ : Gal(Qcycl /Q) → Ẑ∗
3. NONCOMMUTATIVE GEOMETRY AND HILBERT’S 12TH PROBLEM 63
intertwines the Galois action on values with the action of Ẑ∗

by symmetries, namely,
(3.30) γ −1 ϕ(x) = ϕ(θ(γ) x),
for all ϕ ∈ E∞ , for all γ ∈ Gal(Qcycl /Q) and for all x ∈ AQ .
Here the arithmetic subalgebra A1,Q can be taken to be the algebra
over Q generated by the e(r) and the µn , µ∗n . This Q-algebra can also be
obtained (as shown in [29]) as the algebra generated by the µn , µ∗n and
by homogeneous functions of weight zero on 1-dim Q-lattices obtained
as a normalization of the functions
X
(3.31) k,a (Λ, φ) = y −k
y∈Λ+φ(a)
by covolume. Namely, one considers the functions ek,a := ck k,a , where

c(Λ) is proportional to the covolume |Λ| and satisfies
√
(2π −1) c(Z) = 1.
The choice of an “arithmetic subalgebra” corresponds to endowing
the noncommutative space A with an arithmetic structure. The sub-
algebra corresponds to the rational functions and the values of KMS∞
states at elements of this subalgebra should be thought of as values
of “rational functions” at the classical points of the noncommutative
space (cf. [31]).
3. Noncommutative Geometry and Hilbert’s 12th problem

The most remarkable arithmetic feature of the result of Bost–Connes
recalled above is the Galois action on the ground states of the sys-
tem. First of all, the fact that the Galois action on the values of
states would preserve positivity (i.e. would give values of other states)
is a very unusual property. Moreover, the values of extremal ground
states on elements of the rational subalgebra generate the maximal
abelian extension Qab = Qcycl of Q. The explicit action of the Galois
group Gal(Qab /Q) ' GL1 (Ẑ) is given by automorphism of the system
(A1 , σt ). Namely, the class field theory isomorphism intertwines the
two actions of the idèles class group, as symmetry group of the system,
and of the Galois group, as permutations of the expectation values of
the rational observables.
In general, the main theorem of class field theory provides a classi-
fication of finite abelian extensions of a local or global field K in terms
of subgroups of a locally compact abelian group canonically associated
to the field. This is the multiplicative group K∗ = GL1 (K) in the local
nonarchimedean case, while in the global case it is the quotient CK /DK

of the idèle class group CK by the connected component of the identity.
Hilbert’s 12th problem can be formulated as the question of provid-
ing an explicit set of generators of the maximal abelian extension Kab
of a number field K, inside an algebraic closure K̄, and of the action of
the Galois group Gal(Kab /K). The maximal abelian extension Kab of
K has the property that
Gal(Kab /K) = Gal(K̄/K)ab .
The result that motivated Hilbert’s formulation of the explicit class
field theory problem, is the Kronecker–Weber theorem, namely the
already mentioned case of the explicit class field theory of K = Q. In
this case, the maximal abelian extension of Q can be identified with
the cyclotomic field Qcycl . Equivalently, one can say that the torsion
points of the multiplicative group C∗ (i.e. the roots of unity) generate
Qab ⊂ C.
Remarkably, the only other case for number fields where this pro-
gram has been carried out completely is that of imaginary quadratic
fields, where the construction relies on the theory of elliptic curves
with complex multiplication, and on the Galois theory of the field of
modular functions (cf. e.g. the survey [90]).
Some generalizations of the original result of Bost–Connes to other
global fields (number fields and function fields) were obtained by Harari
and Leichtnam [49], P. Cohen [14], Arledge, Laca and Raeburn [3]. A
more detailed account of various results related to the BC system and
generalizations is given in the “further developments” section of [29].
Because of this close relation to Hilbert’s 12th problem, it is clear that
finding generalizations of the Bost–Connes system to other number
fields is a very difficult problem, hence it is not surprising that, so far,
these constructions have not fully recovered the Galois properties of
the ground states of the BC system in the generalized setting.
The strongest form of the result one may wish to obtain in this
direction can be formulated as follows.
Given a number field K, we let AK denote the adèles of K and we
let A∗K = GL1 (AK ) be the group of idèles of K. As above, we write CK
for the group of idèles classes CK = A∗K /K∗ and DK for the connected
component of the identity in CK .
One wishes to construct a C ∗ -dynamical system (A, σt ) and an
“arithmetic” subalgebra AK , which satisfy the following properties:
(1) The idèles class group G = CK /DK acts by symmetries on
(A, σt ) preserving the subalgebra AK .
(2) The states ϕ ∈ E∞ , evaluated on elements of AK , satisfy:
4. THE GL2 SYSTEM 65
• ϕ(a) ∈ K̄, the algebraic closure of K in C;

• the elements of {ϕ(a) : a ∈ AK }, for ϕ ∈ E∞ generate
Kab .
(3) The class field theory isomorphism
'
θ : CK /DK −→ Gal(Kab /K)
intertwines the actions,
(3.32) α−1 ◦ ϕ = ϕ ◦ θ −1 (α),
for all α ∈ Gal(Kab /K) and for all ϕ ∈ E∞ .
This may provide a possible new approach, via noncommutative
geometry, to the explict class field theory problem.
Given a number field K with [K : Q] = n, there is an embedding
K∗ ,→ GLn (Q) of its multiplicative group in GLn (Q). Such embedding
induces an embedding of GL1 (AK,f ) into GLn (Af ), where AK,f = Af ⊗K
are the finite adèles of K.
This suggests that a possible strategy to approach the problem
stated above may be to first study quantum statistical mechanical sys-
tems corresponding to GLn analogs of the Bost–Connes system.
The main result of the joint work of Alain Connes and the author
in [29] is the construction of such system in the GL2 case and the
analysis of the arithmetic properties of its KMS states. In the case of
GL2 , one sees that the geometry of modular curves and the algebra
of modular forms appear naturally. This leads to a formulation of a
quantum statistical mechanical system related to the explicit class field
theory of imaginary quadratic fields [31].
The first case for which there is not yet a complete solution to the
explicit √
class field theory problem is the case of real quadratic fields,
K = Q( d), for some positive integer d. It is natural to ask whether
the approach outlined above, based on noncommutative geometry, may
provide any new information on this case. Recent work of Manin [62]
[63] suggests a close relation between the real quadratic case and non-
commutative geometry. We will discuss the possible relation between
his approach and the GL2 -system.
4. The GL2 system

In this section we will describe the main features of the GL2 analog
of the Bost–Connes system, according to the results of [29].
We can start with the same geometric approach in terms of Shimura
varieties (cf. [31], [30]) that we used above to introduce the BC system.
The Shimura variety associated to the tower of modular curves is

described by the adèlic quotient
(3.33)
Sh(GL2 , H± ) = GL2 (Q)\(GL2 (Af ) × H± )
= GL+ + ∗
2 (Q)\(GL2 (Af ) × H) = GL2 (Q)\GL2 (A)/C .
The tower of modular curves (cf. [31]) has base V = P1 over Q and
Vn = X(n) the modular curve corresponding to the principal congru-
ence subgroup Γ(n). The automorphisms of the projection Vn → V are
given by GL2 (Z/nZ)/{±1}, so that the group of deck transformations
of the tower V is in this case of the form
AutV (V) = lim GL2 (Z/nZ)/{±1} = GL2 (Ẑ)/{±1}.

←−
n
The inverse limit lim Γ\H over congruence subgroups Γ ⊂ SL(2, Z)

←−
gives a connected component of (3.33), while by taking congruence
subgroups in SL(2, Q) one obtains the adèlic version Sh(GL2 , H± ). The
simple reason why it is necessary to pass to the nonconnected case is
the following. The varieties in the tower are arithmetic varieties defined
over number fields. However, the number field typically changes along
the levels of the tower (Vn is defined over the cyclotomic field Q(ζn )).
Passing to nonconnected Shimura varieties allows for the definition of a
canonical model where the whole tower is defined over the same number
field.
The quotient (3.33) parameterizes invertible 2-dimensional Q-lattices
up to scaling (cf. [78]). When, instead of restricting to the invertible
case, we consider commensurability classes of 2-dimensional Q-lattices
up to scaling, we obtain a noncommutative space whose classical points
are the Shimura variety (3.33). More precisely, we have the following
(cf. [31]):
Lemma 4.1. The space of commensurability classes of 2-dimensional
Q-lattices up to scaling is described by the quotient
(3.34) Sh(nc) (GL2 , H± ) := GL2 (Q)\(M2 (Af ) × H± ).

Any 2-dimensional Q-lattice can be written in the form
(Λ, φ) = (λ(Z + Zτ ), λρ),
for some λ ∈ C∗ , some τ ∈ H, and some
ρ ∈ M2 (Ẑ) = Hom(Q2 /Z2 , Q2 /Z2 ).

Thus, the space of 2-dimensional Q-lattices up to the scale factor λ ∈

C∗ and up to isomorphisms, is given by
(3.35) M2 (Ẑ) × H mod Γ = SL(2, Z).
The commensurability relation is implemented by the partially defined
action of GL+
2 (Q) on this space, given by
g(ρ, z) = (gρ, g(z)),

where g(z) denotes action as fractional linear transformation.
Equivalently, the data (Λ, φ) of a Q-lattice in C are equivalent to
data (E, η) of an elliptic curve E = C/Λ and an Af -homomorphism
(3.36) η : Q 2 ⊗ Af → Λ ⊗ A f ,
where Λ ⊗ Af = (Λ ⊗ Ẑ) ⊗ Q, with
(3.37) Λ ⊗ Ẑ = lim Λ/nΛ,
←−
n
and Λ/nΛ = E[n] the nth torsion of E. Since the Q-lattices need not
be invertible, we do not require that η be an Af -isomorphism (cf. [78]).
The commensurability relation between Q-lattices corresponds to
the equivalence (E, η) ∼ (E 0 , η 0 ) given by an isogeny g : E → E 0 and
η 0 = (g ⊗ 1) ◦ η. Namely, the equivalence classes can be identified
with the quotient of M2 (Af ) × H± by the action of GL2 (Q), (ρ, z) 7→
(gρ, g(z)).
To associate a quantum statistical mechanical system to the space
of commensurability classes of 2-dimensional Q-lattices up to scaling,
it is convenient to use the description (3.35). Then one can consider
as noncommutative algebra of coordinates the convolution algebra of
continuous compactly supported functions on the quotient of the space
(3.38) U := {(g, ρ, z) ∈ GL+
2 (Q) × M2 (Ẑ) × H|gρ ∈ M2 (Ẑ)}
by the action of Γ × Γ,
(3.39) (g, ρ, z) 7→ (γ1 gγ2−1 , γ2 z).
One endows this algebra with the convolution product
X
(3.40) (f1 ∗ f2 )(g, ρ, z) = f1 (gs−1 , sρ, s(z))f2 (s, ρ, z)
s∈Γ\GL+
2 (Q):sρ∈M2 (Ẑ)
and the involution f ∗ (g, ρ, z) = f (g −1 , gρ, g(z)).

The time evolution is given by
(3.41) σt (f )(g, ρ, z) = det(g)it f (g, ρ, z).
For ρ ∈ M2 (Ẑ) let
(3.42) Gρ := {g ∈ GL+
2 (Q) : gρ ∈ M2 (Ẑ)}
and consider the Hilbert space Hρ = `2 (Γ\Gρ ).

A 2-dimensional Q-lattice L = (Λ, φ) = (ρ, z) determines a repre-
sentation of the Hecke algebra by bounded operators on Hρ , setting
X
(3.43) (πL (f )ξ) (g) = f (gs−1 , sρ, s(z)) ξ(s).
s∈Γ\Gρ
In particular, when the Q-lattice L = (Λ, φ) is invertible one obtains

Hρ ∼
= `2 (Γ\M2+ (Z)).
In this case, the Hamiltonian implementing the time evolution (3.41)
is given by the operator
(3.44) H m = log det(m) m .
Thus, in the special case of invertible Q-lattices (3.43) yields a positive
energy representation. In general for Q-lattices which are not commen-
surable to an invertible one, the corresponding Hamiltonian H is not
bounded below.
The Hecke algebra (3.40) admits a C ∗ -algebra completion A2 , where
the norm is the sup over all representations πL .
The partition function for this GL2 system is given by
X ∞
X
−β
(3.45) Z(β) = det(m) = σ(k) k −β = ζ(β)ζ(β − 1),
m∈Γ\M2+ (Z) k=1
P
where σ(k) = d|k d. This suggests the fact that one expects two phase
transitions to take place, at β = 1 and β = 2, respectively.
The set of components of Sh(GL2 , H± ) is given by
(3.46) π0 (Sh(GL2 , H± )) = Sh(GL1 , {±1}).
Similarly, on the level of the corresponding noncommutative spaces
(3.34) and (3.16), the identification (3.46) gives the compatibility be-
tween the GL1 and the GL2 system (cf. [31]). At the level of the
classical commutative spaces, this is given by the map
(3.47) det × sign : Sh(GL2 , H± ) → Sh(GL1 , {±1}),
which corresponds in fact to passing to the set π0 of connected compo-
nents.
4.1. KMS states of the GL2 -system. The main result of [29]
on the structure of KMS states is the following.
Theorem 4.2. The KMSβ states of the GL2 -system have the fol-
lowing properties:
(1) In the range β ≤ 1 there are no KMS states.
(2) In the range β > 2 the set of extremal KMS states is given by
the classical Shimura variety
(3.48) Eβ ∼
= GL2 (Q)\GL2 (A)/C∗ .
This shows that the extremal KMS states at sufficiently low tem-
perature are parameterized by the invertible Q-lattices. The explicit
expression for these extremal KMSβ states is obtained as
1 X
(3.49) ϕβ,L (f ) = f (1, mρ, m(z)) det(m)−β
Z(β) +
m∈Γ\M2 (Z)
where L = (ρ, z) is an invertible Q-lattice.

As the temperature rises, and we let β → 2 from above, all the
different phases of the system merge, which is strong evidence for the
fact that in the intermediate range 1 < β ≤ 2 the system should have
only a single KMS state.
The symmetry group of A2 (including both automorphisms and
endomorphisms) can be identified with the group
(3.50) GL2 (Af ) = GL+
2 (Q)GL2 (Ẑ).
Here the group GL2 (Ẑ) acts by automorphisms,

(3.51) θγ (f )(g, ρ, z) = f (g, ργ, z),
as the group of deck transformations of coverings of modular curves.
The novelty with respect to the BC case is that we also have an
action of GL+
2 (Q) by endomorphisms
(
f (g, ρm̃−1 , z) ρ ∈ m M2 (Ẑ)
(3.52) θm (f )(g, ρ, z) =
0 otherwise
where m̃ = det(m)m−1 .
The subgroup Q∗ ,→ GL2 (Af ) acts by inner, hence the group of
symmetries of the set of extremal states Eβ is of the form
(3.53) S = Q∗ \GL2 (Af ).
In the case of E∞ states (defined as weak limits) the action of
GL+2 (Q) is more subtle to define. In fact, (3.52) does not directly
induce a nontrivial action on E∞ . However, there is a nontrivial action
induced by the action on Eβ states for sufficiently large β. The action

on the KMS∞ states is obtained by a “warming up and cooling down
procedure”, as in (3.10) and (3.11).
Finally, the Galois action on the extremal KMS∞ states is described
by the following result [29].
Theorem 4.3. There exists an arithmetic subalgebra A2,Q of un-
bounded multiplier of the C ∗ -algebra A2 , such that the following holds.
For ϕ∞,L ∈ E∞ with L = (ρ, τ ) generic, the values on arithmetic ele-
ments satisfy
(3.54) ϕ∞,L (A2,Q ) ⊂ Fτ ,
where Fτ is the embedding of the modular field in C given by evaluation
at τ ∈ H. There is an isomorphism
'
(3.55) θϕ : Gal(Fτ /Q) −→ Q∗ \GL2 (Af ),
that intertwines the Galois action on the values of the state with the
action of symmetries,
(3.56) γ −1 ϕ(f ) = ϕ(θϕ (γ)f ), ∀f ∈ A2,Q , ∀γ ∈ Gal(Fτ /Q).
Here recall that the modular field F is the field of modular functions
over Qab , namely the union of the fields FN of modular functions of
level N rational over the cyclotomic field Q(ζn ), that is, such that the
q-expansion in powers of q 1/N = exp(2πiτ /N ) has all coefficients in
Q(e2πi/N ).
It has explicit generators given by the Fricke functions ([88], [57]).
If ℘ is the Weierstrass ℘-function, which gives the parameterization
z 7→ (1, ℘(z; τ, 1), ℘0 (z; τ, 1))
of the elliptic curve
y 2 = 4x3 − g2 (τ )x − g3 (τ )
by the quotient C/(Z+Zτ ), then the Fricke functions are homogeneous
functions of 1-dimensional lattices of weight zero, parameterized by
v ∈ Q2 /Z2 , of the form
g2 (τ )g3 (τ )
(3.57) fv (τ ) = −27 35 ℘(λτ (v); τ, 1),
∆(τ )
where ∆ = g23 − 27g32 is the discriminant and λτ (v) := v1 τ + v2 .
For generic τ (such that j(τ ) is transcendental), evaluation of the
modular functions at the point τ ∈ H gives an embedding Fτ ,→ C.
There is a corresponding isomorphism
'
(3.58) θτ : Gal(Fτ /Q) → Aut(F ).
The isomorphism (3.55) is given by

(3.59) θϕ (γ) = ρ−1 θτ (γ) ρ,
for θτ as in (3.58). In fact, the group of automorphisms Aut(F ) has a
completely explicit description, due to a result of Shimura [88], which
identifies it with the quotient
Aut(F ) ∼
= Q∗ \GL2 (Af ).
We still need to explain what is the arithmetic subalgebra that
appears in Theorem 4.3.
We consider continuous functions on the quotient of (3.38) by the
action (3.39), which have finite support in the variable g ∈ Γ\GL+
2 (Q).
For convenience we adopt the notation
f(g,ρ) (z) = f (g, ρ, z)
so that f(g,ρ) ∈ C(H). Let pN : M2 (Ẑ) → M2 (Z/N Z) be the canonical
projection. We say that f is of level N if
f(g,ρ) = f(g,pN (ρ)) ∀(g, ρ).
Then f is completely determined by the functions
f(g,m) ∈ C(H), for m ∈ M2 (Z/N Z).
Notice that the invariance
f (gγ, ρ, z) = f (g, γρ, γ(z)),
for all γ ∈ Γ and for all (g, ρ, z) ∈ U , implies that we have
(3.60) f(g,m)|γ = f(g,m) , ∀γ ∈ Γ(N ) ∩ g −1 Γg.
so that f is invariant under a congruence subgroup.
Elements f of A2,Q are characterized by the following properties.
• The support of f in Γ\GL+ 2 (Q) is finite.
• The function f is of finite level with
f(g,m) ∈ F ∀(g, m).
• The function f satisfies the cyclotomic condition:
f(g,α(u)m) = cycl(u) f(g,m) ,
for all g ∈ GL+ ∗
2 (Q) diagonal and all u ∈ Ẑ , with

u 0
α(u) = .
0 1
Here cycl : Ẑ∗ → Aut(F ) denotes the action of the Galois group Ẑ∗ '
Gal(Qab /Q) on the coefficients of the q-expansion in powers of q 1/N =
exp(2πiτ /N ).
If we took only the first two conditions, this would allow the algebra
A2,Q to contain the cyclotomic field Qab ⊂ C, but this would prevent
the existence of states satisfying the desired Galois property. In fact, if
the subalgebra contains scalar elements in Qcycl , the sought for Galois
property would not be compatible with the C-linearity of states. The
cyclotomic condition then forces the spectrum of the elements of A2,Q
to contain all Galois conjugates of any root of unity that appears in the
coefficients of the q-series, so that elements of A2,Q cannot be scalars.
The algebra A2,Q defined by the properties above is a subalgebra
of unbounded multipliers of A2 , which is globally invariant under the
group of symmetries Q∗ \GL2 (Af ).
5. Quadratic fields
The first essential step in the direction of generalizing the BC sys-
tem to other number fields and exploring a possible approach to the
explicit class field theory problem via noncommutative geometry is the
construction of a system that recovers√the explicit class field theory
for imaginary quadratic fields K = Q( −d), for d a positive integer.
This case is being investigated as part of ongoing joint work of Alain
Connes, Niranjan Ramachandran, and the author [31].
A system for imaginary quadratic fields can be constructed by con-
sidering 1-dimensional K-lattices. These are lattices in K with a homo-
morphism φ : K/O → QΛ/Λ, where O is the ring of integers of K. A
choice of a generator τ ∈ H such that K = Q(τ ) realizes 1-dimensional
K-lattices as a particular set of 2-dimensional Q-lattices. The com-
mensurability relation for 1-dimensional K-lattices can be defined as a
commensurability of the underlying 2-dimensional Q-lattices where the
isomorphism is realized by multiplication by an element in K∗ , viewed
as embedded in GL+ 2 (Q).
The set of commensurability classes of 1-dimensional K-lattices up
to scale is then described by the quotient
(3.61) AK,f /K∗ ,
where AK,f = Af ⊗K are the finite adèles of K, while the set of invertible
1-dimensional K-lattices up to scale can be identified with the idèle
class group
(3.62) CK /DK = A∗K,f /K∗ .
5. QUADRATIC FIELDS 73
The noncommutative algebra of coordinates of the quotient (3.61)

is obtained as in the case of the GL2 -system as the convolution algebra
of functions on
UK = {(x, ρ) ∈ K∗ × Ô : xρ ∈ Ô},
with Ô = Ẑ ⊗ O, endowed with the time evolution
(3.63) σt (f )(x, ρ, λ) = N (x)−it f (x, ρ, λ),
where N : K ∗ → Q∗ is the norm map.
This quantum statistical mechanical system has properties that are,
in a sense, intermediate between the BC system and the GL2 -system.
The symmetry group is the group of idèles classes A∗K,f , with the
subgroup K∗ acting by inner, so that the induced action on KMS states
is by the idèle class group (3.62). As in the case of the GL2 -system,
only the subgroup Ô/O ∗ , with O ∗ the group of units, acts by automor-
phisms, while the full action of A∗K,f /K∗ also involves endomorphisms.
This, in particular, shows the appearance of the class number of K,
as one can see from the commutative diagram
1 / Ô ∗ /O ∗ / A∗K,f /K∗ / Cl(O) / 1

(3.64) ' ' '

1 / Gal(Kab /H) / Gal(Kab /K) / Gal(H/K) / 1,
where H is the Hilbert class field of K, i.e. , its maximal abelian unram-
ified extension. The ideal class group Cl(O) is naturally isomorphic to
the Galois group Gal(H/K). The case of class number one is analogous
to the BC system, as was already observed (cf. [49]).
The same definition of the arithmetic algebra A2,Q for the GL2 -
system provides a subalgebra of the system for the imaginary quadratic
field K. In fact, one considers functions in A2,Q whose finite support
in the g variable lies in K∗ embedded in GL+ 2 (Q), and then takes the
restriction to the set of 1-dimensional K-lattices, by evaluation at τ ∈ H
and restriction to ρ ∈ Ô, embedded in M2 (Ẑ).
Notice that, because of the fact that the variable z ∈ H is now set
to be z = τ , we obtain a subalgebra as in the BC case and not an
algebra of unbounded multipliers as in the GL2 -case.
With this choice of the arithmetic subalgebra, one obtains a result
analogous to Theorem 4.3. Namely, since now the 2-dimensional Q-
lattices L = (ρ, τ ) we are considering have a fixed τ ∈ H ∩ K, we are no
longer in the generic case of Theorem 4.3. For a CM point τ ∈ H ∩ K
evaluation of elements f in the modular field at τ no longer gives an
embedding. The image Fτ ⊂ C is in this case a copy of the maximal

abelian extension of K (cf. [88])
(3.65) F → Fτ ' Kab ⊂ C.
The explicit action of the Galois group Gal(Kab /K) is obtained through
the action of automorphisms of the modular field via Shimura reci-
procity [88], as described in the following diagram with exact rows:
θ
1 / K∗ / GL1 (AK,f ) / Gal(K ab /K) / 1
(3.66) qτ

σ
1 / Q∗ / GL2 (Af ) / Aut(F ) / 1.
Here θ is the class field theory isomorphism and qτ is the embedding
determined by the choice of τ ∈ H with K = Q(τ ). Thus, the explicit
Galois action is given by
θ(γ)(f (τ )) = f σ(qτ (γ)) (τ ),
where f 7→ f α denotes the action of α ∈ Aut(F ).
This provides the intertwining between the symmetries of the quan-
tum statistical mechanical system and the Galois action on the image
of states on the elements of the arithmetic subalgebra.
The structure of KMS states at positive temperatures is similar to
the Bost–Connes case (cf. [31]).
The next fundamental question in the direction of generalizations
of the BC system to other number fields is how to approach the more
case of real quadratic fields. We give a brief outline of Manin’s ideas on
real multiplication [62] [63], and suggest how they may combine with
the GL2 -system described above.
5.1. Real multiplication. Recently, Manin developed a theory of

real multiplication for noncommutative tori [63], aimed at providing a
setting, within noncommutative geometry, where to treat the problem
of abelian extensions of real quadratic fields on a somewhat similar
footing as the known case of imaginary quadratic fields, for which the
theory of elliptic curves with complex multiplication provides the right
geometric setup.
The first entry in the dictionary developed in [63] between elliptic
curves with complex multiplication and noncommutative tori with real
multiplication consists of a parallel between lattices and pseudolattices
in C,
(C, lattice Λ) ! (R, pseudo-lattice L).
By a pseudo-lattice one means the data of a free abelian group of

rank two, with an injective homomorphism to a 1-dimensional complex
vector space, such that the image lies in an oriented real line.
This aims at generalizing the well known equivalence between the
category of elliptic curves and the category of lattices, realized by the
period functor, to a setting that includes noncommutative tori.
As we have seen previously, any 2-dimensional lattice is, up to iso-
morphism, of the form Λτ = Z + Zτ , for τ ∈ C r R and non-trivial
morphisms between such lattices are given by the action of matrices
M2 (Z) by fractional linear transformations. Thus, the moduli space of
lattices up to isomorphism is given by the quotient of P1 (C) r P1 (R)
by PGL2 (Z).
A pseudo-lattice is, up to isomorphism, of the form Lθ = Z + Zθ,
for θ ∈ R r Q, and non-trivial morphisms of pseudo-lattices are again
given by matrices in M2 (Z) acting by fractional linear transformations
on P1 (R). The moduli space of pseudo-lattices is given by the quotient
of P1 (R) by the action of PGL(2, Z). Since the action does not give
rise to a nice classical quotient, this moduli space should be treated as
a noncommutative space.
As described at lenght in the previous chapter, the resulting space
represents a component in the “boundary” of the classical moduli space
of elliptic curves, which parameterizes those degenerations from lattices
to pseudo-lattices, that are invisible to the usual algebro-geometric
setting.
The cusp, i.e. the orbit of P1 (Q), corresponds to the degenerate
case where the image in C of the rank two free abelian group has rank
one.
This gives another entry in the dictionary, regarding the moduli
spaces:
H/PSL(2, Z) ! C(P1 (R)) o PGL(2, Z).
In the correspondence of pseudo-lattices and noncommutative tori,
the group of invertible morphisms of pseudo-lattices corresponds to
isomorphisms of noncommutative tori realized by strong Morita equiv-
alences. In this context a “morphism” is not given as a morphism of
algebras but as a map of the category of modules, obtained by tensoring
with a bimodule. The notion of Morita equivalences as morphisms fits
into the more general context of correspondences for operator algebras
as in [20] §V.B, as well as in the algebraic approach to noncommutative
spaces of [86].
The category of noncommutative tori is defined by considering as
morphisms the isomorphism classes of (C ∗ -)bimodules that are range of
projections. The functor from noncommutative tori to pseudo-lattices

(cf. [62] §3.3 and [63] §1.4) is then given on objects by
(3.67) Tθ2 7→ (K0 (Aθ ), HC0 (Aθ ), τ : K0 (Aθ ) → HC0 (Aθ )) .
Here Aθ is the algebra of the noncommutative torus, HC0 (Aθ ) =

Aθ /[Aθ , Aθ ], with τ the universal trace, and the orientation is deter-
mined by the cone of positive elements in K0 . On morphisms the
functor is given by
(3.68) Mθ,θ0 7→ ([E] 7→ [E ⊗Aθ Mθ,θ0 ]) ,
where Mθ,θ0 are the bimodules constructed by Connes in [15]. A crucial

point in this definition is the fact that, for noncommutative tori, finite
projective modules are classified by the value of a unique normalized
trace (cf. [15] [85]).
The functor of (3.67) (3.68) is weaker than an equivalence of cate-
gories. For instance, it maps trivially all ring homomorphisms that act
trivially on K0 . However, this correspondence is sufficient to develop
a theory of real multiplication for noncommutative tori parallel to the
theory of complex multiplication for elliptic curves (cf. [62] [63]).
For lattices/elliptic curves, the typical situation is that End(Λ) = Z,
but there are exceptional lattices for which End(Λ) ) Z. In this case,
there exists a complex quadratic field K such that Λ is isomorphic to
a lattice in K. More precisely, the endomorphism group is given by
End(Λ) = Z + f O, where O is the ring of integers of K and the integer
f ≥ 1 is called the conductor. Such lattices are said to have complex
multiplication. The elliptic curve EK with EK (C) = C/O is endowed
with a complex multiplication map, given on the universal cover by
x 7→ ax, a ∈ O.
Similarly, there is a parallel situation for pseudo-lattices: End(L) )
Z happens when there exists a real quadratic field K such that L is
isomorphic to a pseudolattice contained in K. In this case, one also
has End(L) = Z + f O. Such pseudo-lattices are said to have real
multiplication. The pseudo-lattices Lθ that have real multiplication
correspond to values of θ ∈ R r Q that are quadratic irrationalities.
These are characterized by the having eventually periodic continued
fraction expansion. The “real multiplication map” is given by tensoring
with a bimodule: in the case of θ with periodic continued fraction
expansion, there is an element g ∈ PGL(2, Z) such that gθ = θ, to
which we can associate an Aθ –Aθ bimodule E.
An analog of isogenies for noncommutative tori is obtained by con-
sidering morphisms of pseudolattices Lθ 7→ Lnθ which correspond to a
Morita morphism given by Aθ viewed as an Anθ –Aθ bimodule, where

Anθ ,→ Aθ by U 7→ U n , V 7→ V .
By considering isogenies, one can enrich the dictionary between
moduli space of elliptic curves and moduli space of Morita equivalent
noncommutative tori. This leads one to consider the whole tower of
modular curves parameterizing elliptic curves with level structure and
the corresponding tower of noncommutative modular curves described
in the previous chapter (cf. [70]),
(3.69) H/G ! C(P1 (R) × P) o Γ,
for Γ = PGL(2, Z) (or PSL(2, Z)) and G ⊂ Γ a finite index subgroup,
with P = G\Γ the coset space.
In the problem of constructing the maximal abelian extension of
complex quadratic fields, one method is based on evaluating at the tor-
sion points of the elliptic curve EK a power of the Weierstrass function.
This means considering the corresponding values of a coordinate on the
projective line EK /O ∗ . The analogous object in the noncommutative
setting, replacing this projective line, should be (cf. [63]) a crossed
product of functions on K by the ax + b group with a ∈ O ∗ and b ∈ O,
for a real quadratic field K.
A different method to construct abelian extensions of a complex
quadratic field K is via Stark numbers. Following the notation of [63],
if j : Λ → V is an injective homomorphism of a free abelian group of
rank two to a one-dimensional complex vector space, and λ0 ∈ Λ ⊗ Q,
then one can consider the zeta function
X 1
(3.70) ζ(Λ, λ0, s) = .
λ∈Λ
|j(λ0 + λ)|2s
Stark proved (cf. [89]) that the numbers
(3.71) S(Λ, λ0 ) = exp ζ 0 (Λ, λ0 , 0)
are algebraic units generating certain abelian extensions of K. The
argument in this case is based on a direct computational tool (the
Kronecker second limit formula) and upon reducing the problem to
the theory of complex multiplication. There is no known independent
argument for the Stark conjectures, while the analogous question is
open for the case of real quadratic fields.
For a real quadratic field K, instead of zeta functions of the form
(3.70), the conjectural Stark units are obtained from zeta functions of
the form (in the notation of [63])
X sgn(l0 + l)0
(3.72) ζ(L, l0 , s) = sgn l00 N (a)s ,
|N (l0 + l)|s
where l 7→ l0 is the action of the nontrivial element in Gal(K/Q) and

N (l) = ll0 . The element l0 ∈ O is chosen so that the ideal a = (L, l0 )
and (l0 )a−1 are coprime wih La−1 , and the summation over l ∈ L in
(3.72) is restricted by taking only one representative from each coset
class (l0 + l), for units satisfying (l0 + L) = (l0 + L), i.e. ≡ 1
mod La−1 . Then the Stark numbers are given as in (3.71) by (cf. [63])
(3.73) S(L, l0 ) = exp ζ 0 (L, l0 , 0).
In [63], Manin develops an approach to the computation of sums
(3.72) based on a version of theta functions for pseudo-lattices, which
are obtained by averaging theta constants of complex lattices along
geodesics with ends at a pair of conjugate quadratic irrationalities θ, θ 0
in R r Q.
This procedure fits into a general philosophy, according to which
one can recast part of the arithmetic theory of modular curves in terms
of the noncommutative boundary (3.69) by studying the limiting be-
havior when τ → θ ∈ R \ Q along geodesics, or some averaging along
such geodesics. In general, some nontrivial result is obtained when ap-
proaching θ along a path that corresponds to a geodesic in the modular
curve that spans a limiting cycle, which is the case precisely when the
endpoint θ is a quadratic irrationality. An example of this type of be-
havior is the theory of “limiting modular symbols” developed in [70],
[71].
5.2. Pseudolattices and the GL2 -system. The noncommuta-
tive space (3.34) of the GL2 -system also admits a compactification,
now given by adding the boundary P1 (R) to H± , as in the noncommu-
tative compactification of modular curves of [70],
Sh(nc) (GL2 , H± ) := GL2 (Q)\(M2 (Af ) × P1 (C))
(3.74)
= GL2 (Q)\M2 (A)/C∗ ,
where P1 (C) = H± ∪ P1 (R).
This corresponds to adding to the space of commensurability classes
of 2-dimensional Q-lattices the pseudolattices in the sense of [63], here
considered together with a Q-structure. It seems then that Manin’s
real multiplication program may fit in with the boundary of the non-
commutative space of the GL2 -system. The crucial question in this
respect becomes the construction of an arithmetic algebra associated
to the noncommutative modular curves. The results illustrated in the
previous chapter, regarding identities involving modular forms at the
boundary of the classical modular curves and limiting modular sym-
bols, as well as the still mysterious phenomenon of “quantum modular
forms” identified by Zagier, point to the fact that there should exist
a rich class of objects replacing modular forms, when “pushed to the
boundary”.
Regarding the role of modular forms, notice that, in the case of
(3.74), we can again consider the dual system. This is a C∗ -bundle
(3.75) L2 = GL2 (Q)\M2 (A).
On this dual space modular forms appear naturally instead of modular
functions and the algebra of coordinates contains the modular Hecke
algebra of Connes–Moscovici ([33], [34]) as arithmetic subalgebra.
Thus, the noncommutative geometry of the space of Q-lattices mod-
ulo the equivalence relation of commensurability provides a setting that
unifies several phenomena involving the interaction of noncommutative
geometry and number theory. These include the Bost–Connes system,
the noncommutative space underlying the construction of the spec-
tral realization of the zeros of the Riemann zeta function in [22], the
modular Hecke algebras of [33] [34], and the noncommutative modular
curves of [70].
CHAPTER 4
Noncommutative geometry at arithmetic infinity
This chapter is based on joint work of Katia Consani and the au-
thor ([36], [37], [38], [39], [40]) that proposes a model for the dual
graph of the maximally degenerate fibers at the archimedean places
of an arithmetic surface in terms of a noncommutative space (spec-
tral triple) related to the action of a Schottky group on its limit set.
This description of ∞-adic geometry provides a compatible setting that
combines Manin’s result [66] on the Arakelov Green function for arith-
metic surfaces in terms of hyperbolic geometry and for a cohomologi-
cal construction of Consani [35] associated to the archimedean fibers
of arithmetic varieties, related to Deninger’s calculation of the local
L-factors as regularized determinants (cf. [43], [44]).
1. Schottky uniformization
Topologically a compact Riemann surface X of genus g is obtained
by gluing the sides of a 4g-gon. Correspondingly, the fundamental
group has a presentation
Y
π1 (X) = ha1 , . . . , ag , b1 , . . . , bg | [ai , bi ] = 1i,
i
where the generators ai and bi label the sides of the polygon.

In the genus g = 1 case, the parallelogram is the fundamental do-
main of the π1 (X) ' Z2 action on the plane C, so that X = C/(Z+Zτ )
is an elliptic curve.
For genus at least g ≥ 2, the hyperbolic plane H2 admits a tessel-
lation by regular 4g-gons, and the action of the fundamental group by
deck transformation is realized by the action of a subgroup π1 (X) '
G ⊂ PSL(2, R) by isometries of H2 . This endows the compact Riemann
surface X with a hyperbolic metric and a Fuchsian uniformization
X = G\H2 .
Another, less well known, type of uniformization of compact Rie-
mann surfaces is Schottky uniformization. We recall briefly some gen-
eral facts on Schottky groups.
81
82 4. NONCOMMUTATIVE GEOMETRY AT ARITHMETIC INFINITY
Figure 1. Limit set of a Schottky group
1.1. Schottky groups. A Schottky group of rank g is a discrete

subgroup Γ ⊂ PSL(2, C), which is purely loxodromic and isomorphic to
a free group of rank g. The group PSL(2, C) acts on P1 (C) by fractional
linear transformations
(az + b)
γ : z 7→ .
(cz + d)
Thus, Γ also acts on P1 (C).
We denote by ΛΓ the limit set of the action of Γ. This is the
smallest non–empty closed Γ–invariant subset of P1 (C). This set can
also be described as the closure of the set of the attractive and repelling
fixed points z ± (γ) of the loxodromic elements γ ∈ Γ. In the case g = 1
the limit set consists of two points, but for g ≥ 2 the limit set is usually
a fractal of some Hausdorff dimension 0 ≤ δ = dimH (ΛΓ ) < 2 (cf. e.g.
Figure 11).
We denote by ΩΓ the domain of discontinuity of Γ, that is, the
complement of ΛΓ in P1 (C). The quotient
(4.1) X(C) = Γ\ΩΓ
is a Riemann surface of genus g and the covering ΩΓ → X(C) is called
a Schottky uniformization of X(C).
1Figures 1 and 3 are taken from “Indra’s pearls” by Mumford, Series, and
Wright, [81]
1. SCHOTTKY UNIFORMIZATION 83
Figure 2. Schottky uniformization for g = 2
Every compact Riemann surface X(C) admits a Schottky uniformiza-

tion.
Let {γi }gi=1 be a set of generators of the Schottky group Γ. We write
γi+g = γi−1 . There are 2g Jordan curves γk on the sphere at infinity
P1 (C), with pairwise disjoint interiors Dk , such that the elements γk
are given by fractional linear transformations that map the interior
of Ck to the exterior of Cj with |k − j| = g. The curves Ck give a
marking of the Schottky group. The markings are circles in the case of
classical Schottky groups. A fundamental domain for the action of a
classical Schottky group Γ on P1 (C) is the region exterior to 2g-circles
(cf. Figure 2).
1.2. Schottky and Fuchsian. Notice that, unlike Fuchsian uni-
formization, where the covering H2 is the universal cover, in the case
of Schottky uniformization ΩΓ is very far from being simply connected,
in fact it is the complement of a Cantor set.
The relation between Fuchsian and Schottky uniformization is given
by passing to the covering that corresponds to the normal subgroup
N ha1 , . . . , ag i of π1 (X) generated by half the generators {a1 , . . . , ag },
Γ ' π1 (X)/N ha1 , . . . , ag i,
with a corresponding covering map
J /Ω
H2 A Γ
AA }}
AA }}
πG AAA }} π
~} Γ
}
X
At the level of moduli, there is a corresponding map between Te-
ichmüller space Tg and Schottky space Sg , which depends on 3g − 3
complex moduli.
Figure 3. Classical and Fuchsian Schottky groups
1.3. Surface with boundary: simultaneous uniformization.

To better visualize geometrically the Schottky uniformization of a com-
pact Riemann surface, we can relate it to a simultaneous uniformization
of the upper and lower half planes that yields two Riemann surfaces
with boundary, joined at the boundary.
A Schottky group that is specified by real parameters so that it lies
in PSL(2, R) is called a Fuchsian Schottky group (cf. Figure 3). Viewed
as a group of isometries of the hyperbolic plane H2 , or equivalently of
the Poincaré disk, a Fuchsian Schottky group G produces a quotient
G\H2 , which is topologically a Riemann surface with boundary.
A quasi-circle for Γ is a Jordan curve C in P1 (C) which is invariant
under the action of Γ. In particular, such curve contains the limit set
ΛΓ . It was proved by Bowen that, if X(C) is a Riemann surface of
genus g ≥ 2 with Schottky uniformization, then there exists always a
quasi-circle for Γ.
We have then P1 (C) \ C = Ω1 ∪ Ω2 and, for πΓ : ΩΓ → X(C) the
covering map,
Ĉ = πΓ (C ∩ ΩΓ ) ⊂ X(C)
is a set of curves on X(C) that disconnect the Riemann surface in the
union of two surfaces with boundary, uniformized respectively by Ω1
and Ω2 .
There exist conformal maps
'
αi : Ω i → U i , U1 ∪ U2 = P1 (C) \ P1 (R)
with Ui ' H2 = upper and lower half planes in P1 (C), and with
Gi := {αi γαi −1 : γ ∈ Γ̃} ' Γ
Fuchsian Schottky groups Gi ⊂ PSL(2, R). Here Γ̃ ⊂ SL(2, R) is the
Γ-stabilizer of each of the two connected components in P1 (C) \ C.
Figure 4. Fuchsian Schottky groups: Riemann surfaces

with boundary
The compact Riemann surface X(C) is thus obtained as

X(C) = X1 ∪∂X1 =Ĉ=∂X2 X2 ,
with Xi = Ui /Gi Riemann surfaces with boundary Ĉ (cf. Figure 4).

In the case where X(C) has a real structure ι : X → X, and the
fixed point set F ix(ι) = X(R) of the involution ι is nonempty, we have
in fact Ĉ = X(R), and the quasi-circle is given by P1 (R).
Notice that, in the case of a Fuchsian Schottky group, the Hausdorff
dimension dimH ΛΓ of the limit set is in fact bounded above by 1, since
ΛΓ is contained in the rectifiable quasi-circle P1 (R).
1.4. Hyperbolic Handlebodies. The action of a rank g Schot-

tky group Γ ⊂ PSL(2, C) on P1 (C), by fractional linear transforma-
tions, extends to an action by isometries on real hyperbolic 3-space H3 .
For a classical Schottky group, a fundamental domain in H3 is given
by the region external to 2g half spheres over the circles Ck ⊂ P1 (C)
(cf. Figure 5).
The quotient
(4.2) XΓ = H3 /Γ
is topologically a handlebody of genus g filling the Riemann surface
X(C).
Figure 5. Genus two: fundamental domain in H3
Metrically, XΓ is a real hyperbolic 3-manifold of infinite volume,

having X(C) as its conformal boundary at infinity X(C) = ∂XΓ .
We denote by XΓ the compactification obtained by adding the con-
formal boundary at infinity,
(4.3) XΓ = (H3 ∪ ΩΓ )/Γ.
In the genus zero case, we just have the sphere P1 (C) as the con-
formal boundary at infinity of H3 , thought of as the unit ball in the
Poincaré model.
In the genus one case we have a solid torus H3 /q Z , for q ∈ C∗ acting
as
q(z, y) = (qz, |q|y)
in the upper half space model, with conformal boundary at infinity the
Jacobi uniformized elliptic curve C∗ /q Z .
In this case, the limit set consists of the point {0, ∞}, the domain
of discontinuity is C∗ and a fundamental domain is the annulus {|q| <
|z| ≤ 1} (exterior of two circles).
The relation of Schottky uniformization to the usual Euclidean uni-
formization of complex tori X = C/(Z + τ Z) is given by q = exp(2πiτ ).
We shall return to the genus one case where we discuss a physical
interpretation of Manin’s result on the Green function. To our pur-
poses, however, the most interesting case is when the genus is g ≥ 2.
In this case, the limit set ΛΓ is a Cantor set with an interesting dy-
namics of the action of Γ. It is the dynamics of the Schottky group on
its limit set that generates an interesting noncommutative space.
Figure 6. Handlebody of genus 2: fundamental do-

mains in H3
1.5. Geodesics in XΓ . The hyperbolic handlebody XΓ has infinite

volume, but it contains a region of finite volume, which is a deformation
retract of XΓ . This is called the “convex core” of XΓ and is obtained by
taking the geodesic hull of the limit set ΛΓ in H3 and then the quotient
by Γ.
We identify different classes of infinite geodesics in XΓ .
• Closed geodesics: since Γ is purely loxodromic, for all γ ∈ Γ
there exist two fixed points {z ± (γ)} ∈ P1 (C). The geodesics
in H3 ∪ P1 (C) with ends at two such points {z ± (γ)}, for some
γ ∈ Γ, correspond to closed geodesics in the quotient XΓ .
• Bounded geodesics: The images in XΓ of geodesics in H3 ∪P1 (C)

having both ends on the limit set ΛΓ are geodesics that remain
confined within the convex core of XΓ .
• Unbounded geodesics: these are the geodesics in XΓ that even-

tually wander off the convex core towards the conformal bound-
ary X(C) at infinity. They correspond to geodesics in H3 ∪
P1 (C) with at least one end at a point of ΩΓ .
In the genus one case, there is a unique primitive closed geodesic,
namely the image in the quotient of the geodesic in H3 connecting 0
and ∞. The bounded geodesics are those corresponding to geodesics
in H3 originating at 0 or ∞.
The most interesting case is that of genus g ≥ 2, where the bounded

geodesics form a complicated tangle inside XΓ . Topologically, this is
a generalized solenoid, namely it is locally the product of a line and a
Cantor set.
2. Dynamics and noncommutative geometry

Since the uniformizing group Γ is a free group, there is a simple way
of obtaining a coding of the bounded geodesics in XΓ . The set of such
geodesics can be identified with ΛΓ ×Γ ΛΓ , by specifying the endpoints
in H3 ∪ P1 (C) modulo the action of Γ.
If {γi }gi=1 is a choice of generators of Γ and γi+g = γi−1 , i = 1, . . . , g,
we can introduce a subshift of finite type (S, T ) where
(4.4)
S = {. . . a−m . . . a−1 a0 a1 . . . a` . . . |ai ∈ {γi }2g −1
i=1 , ai+1 6= ai , ∀i ∈ Z}
is the set of doubly infinite words in the generators and their inverses,
with the admissibility condition that no cancellations occur. The map
T is the invertible shift
(4.5) T (. . . a−m . . . a−1 a0 a1 . . . a` . . .) = . . . a−m+1 . . . a0 a1 a2 . . . a`+1 . . .
Then we can pass from the discrete dynamical system (S, T ) to its
suspension flow and obtain the mapping torus
(4.6) ST := S × [0, 1]/(x, 0) ∼ (T x, 1).
Topologically this space is a solenoid, that is, a bundle over S 1 with

fiber a Cantor set.
2.1. Homotopy quotient. The space ST of (4.6) has a natural

interpretation in noncommutative geometry as the homotopy quotient
in the sense of Baum–Connes [6] of the noncommutative space given
by the C ∗ -algebra
(4.7) C(S) oT Z
describing the action of the shift (4.5) on the totally disconnected

space (4.4). The noncommutative space (4.7) parameterizes bounded
geodesics in the handlebody XΓ .
The homotopy quotient is given by
S ×Z R = S T .
2. DYNAMICS AND NONCOMMUTATIVE GEOMETRY 89
The K-theory of the C ∗ -algebra (4.7) can be computed via the

Pimsner–Voiculescu six terms exact sequence,
δ=1−T
K1 (A) / K0 (C(S)) / K0 (C(S))
O

K1 (C(S)) o K1 (C(S)) o K0 (A)
where A = C(S) oT Z. Here, since the space S is totally disconnected,

we have K1 (C(S)) = 0 and K0 (C(S)) = C(S, Z) (locally constant
integer valued functions). Thus the exact sequence becomes
δ=1−T
(4.8) 0 → K1 (A) → C(S, Z) → C(S, Z) → K0 (A) → 0,
and we obtain K1 (A) = Z = Ker(δ) and K0 (A) = Coker(δ). In dy-
namical systems language, these are respectively the invariants and
coinvariants of the invertible shift T (cf. [10] [82]).
In terms of the homotopy quotient, one can describe this exact
sequence more geometrically in terms of the Thom isomorphism and
the µ-map
µ : K ∗+1 (ST ) ∼
= H ∗+1 (ST , Z) → K∗ (C(S) oT Z).
Thus, we obtain:
K1 (A) ' H 0 (ST ) = Z
K0 (A) ' H 1 (ST )
The H 1 (ST ) can be identified with the Čech cohomology group given
by the homotopy classes [ST , U (1)], by mapping f 7→ [exp(2πitf (x))],
for f ∈ C(S, Z)/δ.
2.2. Filtration. The identification

H 1 (ST , Z) ∼
= K0 (C(S) oT Z)
of the cohomology of ST with the K0 -group of the crossed product
C∗ -algebra for the action of T on S endows H 1 (ST , Z) with a filtration.
Theorem 2.1. The first cohomology of ST is a direct limit
H 1 (ST ) = lim Fn ,
−→
n
of free abelian groups F0 ,→ F1 ,→ · · · Fn ,→ · · · of ranks rank F0 = 2g

and rank Fn = 2g(2g − 1)n−1 (2g − 2) + 1, for n ≥ 1.
In fact, by the Pimsner-Voiculescu six terms exact sequence, the

group K0 (C(S) oT Z) can be identified with the cokernel of the map
1 − T acting as f 7→ f − f ◦ T on the Z module C(S, Z) ' K0 (C(S)).
Then the filtration is given by the submodules of functions depend-
ing only on the a0 . . . an coordinates in the doubly infinite words de-
scribing points in S. Namely, one first shows that
H 1 (ST , Z) = C(S, Z)/δ(C(S, Z)) = P/δP
where P denotes the Z-module of locally constant Z-valued functions
that depend only on “future coordinates”. This can be identified with
functions on the limit set ΛΓ , since each point in ΛΓ is described by
an infinite (to the right) admissible sequence in the generators γi and
their inverses. We have P ' C(ΛΓ , Z).
The module P clearly has a filtration by the submodules Pn of
functions of the first n + 1 coordinates. These have rank Pn = 2g(2g −
1)n .
We set Fn := Pn /δPn−1 . On can show that there are induced
injections Fn ,→ Fn+1 and that
H 1 (ST ) = lim Fn .
−→
n
Moreover, we have rankFn = θn − θn−1 + 1, where θn is the number

of admissible words of length n + 1. All the Z modules Fn and the
quotients Fn /Fn−1 are torsion free (cf. [82]).
2.3. Hilbert space and grading. It is convenient to consider the

complex vector space
PC = C(ΛΓ , Z) ⊗ C
and the corresponding exact sequence computing the cohomology with
C coefficients:
δ
(4.9) 0 → C → PC −→ PC → H 1 (ST , C) → 0
The complex vector space PC sits in the Hilbert space
PC ⊂ L = L2 (ΛΓ , dµ),
where µ is the Patterson–Sullivan measure on the limit set, satisfying
γ ∗ dµ = |γ 0 |dimH (ΛΓ ) dµ,
with dimH (ΛΓ ) the Hausdorff dimension.
This gives a Hilbert space L, together with a filtration Pn by finite
dimensional subspaces. In such setting, it is natural to consider a
2. DYNAMICS AND NONCOMMUTATIVE GEOMETRY 91
corresponding grading operator,

X
(4.10) D= nΠ̂n ,
n
where Πn denotes the orthogonal projection onto Pn and Π̂n = Πn −

Πn−1 .
2.4. Cuntz–Krieger algebra. There is a noncommutative space
that encodes nicely the dynamics of the Schottky group Γ on its limit
set ΛΓ . Consider the 2g × 2g matrix A that gives the admissibility
condition for sequences in S: this is the matrix with {0, 1} entries
satisfying Aij = 1 for |i − j| 6= g and Aij = 0 otherwise.
The Cuntz–Krieger algebra OA associated to this matrix is the uni-
versal C ∗ -algebra generated by partial isometries Si , i = 1, . . . , 2g,
satisfying the relations
X X
Sj Sj∗ = 1 Si∗ Si = Aij Sj Sj∗ .
j j
Recall that a partial isometry is an operator S satisfying S = SS ∗ S.

This algebra is related to the Schottky group by the following result.
Proposition 2.2. There is an isomorphism
(4.11) OA ∼ = C(ΛΓ ) o Γ.
Up to a stabilization (tensoring with compact operators), the alge-
bra has another crossed product description as
(4.12) OA ' FA oT Z,
with FA an AF-algebra (a direct limit of finite dimensional C ∗ -algebras).
Consider the cochain complex of Hilbert spaces
δ
0→C→L→L→H→0
determined by the Pimsner–Voiculescu sequence (4.8).
Proposition 2.3. The C ∗ -algebra OA admits a faithful represen-
tation on the Hilbert space L = L2 (ΛΓ , dµ).
This is obtained as follows.
For dH = dimH (ΛΓ ) the Hausdorff dimension, consider the opera-
tors
(4.13) P i f = χ γi f Ti f = |(γi−1 )0 |dH /2 f ◦ γi−1 ,
where {γi } are the generators of Γ and their inverses and
(4.14) Tγ f = |γ 0 |dH /2 f ◦ γ.
Then the operators

X
(4.15) Si = Aij Ti∗ Pj
j
are partial isometries on L satisfying the Cuntz–Krieger relations for

the matrix A of the subshift of finite type (4.4). This gives the repre-
sentation of OA .
2.5. Spectral triple for Schottky groups. On the Hilbert space

H = L⊕L consider the diagonal action of the algebra OA and the Dirac
operator D defined as
P
D|L⊕0 = n (n + 1) (Π̂n ⊕ 0)
(4.16) P
D|0⊕L = − n n (0 ⊕ Π̂n ).
The choice of the sign in (4.16) is not optimal from the point of
view of the K-homology class. A better choice would be

0 1
F = .
1 0
This would require in turn a modification of |D|. A construction along
these lines is being considered in joint work of the author with Alina
Vdovina and Gunther Cornelissen. In our setting here, the reason
for the choice (4.16) in [36] is the formula (4.91) relating D to the
“logarithm of Frobenius” at arithmetic infinity.
Theorem 2.4. For a Schottky group Γ with dimH (ΛΓ ) < 1, the
data (OA , H, D), for H = L⊕L with the diagonal action of OA through
the representation (4.15) and the Dirac operator (4.16), define a non-
finitely summable, θ-summable spectral triple.
The key point of this result is the compatibility relation between the
algebra and the Dirac operator, namely the fact that the commutators
alg
[D, a] are bounded operators, for all a ∈ OA , the involutive subal-
gebra generated algebraically by the Si subject to the Cuntz–Krieger
relations.
This follows by an estimate on the norm of the commutators k[D, Si ]k
and k[D, Si∗ ]k, in terms of the Poincaré series of the Schottky group
(dH < 1)
X
|γ 0 |s , s = 1 > dH ,
γ∈Γ
where the Hausdorff dimension dH is the exponent of convergence of

the Poincaré series.
3. ARITHMETIC INFINITY: ARCHIMEDEAN PRIMES 93
The dimension of the n-th eigenspace of D is 2g(2g − 1)n−1 (2g − 2)

for n ≥ 1, 2g for n = 0 and 2g(2g − 1)−n−1 (2g − 2) for n ≤ −1, so the
spectral triple is not finitely summable, since |D|z is not of trace class.
It is θ summable, since the operator exp(−tD 2 ) is of trace class, for all
t > 0.
Using the description (4.12) of the noncommutative space as crossed
product of an AF algebra by the action of the shift, FA oT Z, one may
be able to find a 1-summable spectral triple. Here the dense subalgebra
should not contain any of the group elements. In fact, by the result of
[19], the group Γ is a free group, hence its growth is too fast to support
finitely summable spectral triples on its group ring.
3. Arithmetic infinity: archimedean primes

An algebraic number field K, which is an extension of Q with [K :
Q] = n admits n = r1 + 2r2 embeddings
(4.17) α : K ,→ C.
These can be subdivided into r1 real embeddings K ,→ R and r2 pairs of
complex conjugate embeddings. The embeddings (4.17) are called the
archimedean primes of the number field. The set of archimedean primes
is often referred to as “arithmetic infinity”, a terminology borrowed
from the case of the unique embedding of Q ,→ R, which is called the
“infinite prime”.
A general strategy in arithmetic geometry is to adapt the tools of
classical algebraic geometry to the arithmetic setting. In particular,
over Q the set of primes Spec(Z) is the analog in arithmetic geometry
of the affine line. It becomes clear then that some compactification is
necessary, at least in order to have a well behaved form of intersec-
tion theory in arithmetic geometry. Namely, we need to pass from the
“affine” Spec(Z) to the projective case. The compactification is ob-
tained by adding the infinite prime to the set of finite primes. A goal
of arithmetic geometry then becomes developing a setting that treats
the infinite prime and the finite primes of equal footing.
More generally, for a number field K with OK its ring of inte-
gers, the set of primes Spec(OK ) is compactified by adding the set
of “archimedean primes”
(4.18) Spec(OK ) = Spec(OK ) ∪ {α : K ,→ C}.
3.1. Arithmetic surfaces. Let X be a smooth projective alge-

braic curve defined over Q. Then, by clearing denominators one ob-
tains an equation with Z coefficients. This determines a scheme XZ
over Spec(Z),
XZ ⊗Z Spec(Q) = X,
where the closed fiber of XZ at a prime p ∈ Spec(Z) is the reduction
X mod p. Thus, viewed as an arithmetic varities, an algebraic curve
becomes a 2-dimensional fibration over the affine line Spec(Z).
One can also consider reductions of X (defined over Z) modulo
n
p for some prime p ∈ Spec(Z). The limit as n → ∞ defines a p–
adic completion of XZ . This can be thought of as an “infinitesimal
neighborhood” of the fiber at p.
The picture is more complicated at arithmetic infinity, since one
does not have a suitable notion of “reduction mod ∞” available to
define the closed fiber. On the other hand, one does have the analog of
the p–adic completion at hand. This is given by the Riemann surface
(smooth projective algebraic curve over C) determined by the equation
of the algebraic curve X over Q, under the embedding of Q ⊂ C,
X(C) = X ⊗Q Spec(C)
with the absolute value | · | at the infinite prime replacing the p-adic
valuations.
Similarly, for K a number field and OK its ring of integers, a smooth
proper algebraic curve X over K determines a smooth minimal model
XOK , which defines an arithmetic surface XOK over Spec(OK ). The
closed fiber X℘ of XOK over a prime ℘ ∈ OK is given by the reduction
mod ℘.
When Spec(OK ) is compactified by adding the archimedean primes,
one also obtains n Riemann surfaces Xα (C), obtained from the equation
defining X over K under the embeddings α : K ,→ C. Of these Riemann
surfaces, r1 carry a real involution.
Thus, the picture of an arithmetic surface over Spec(OK ) is as fol-
lows:
/ XSpec(O ) / XSpec(O ) o
X℘ K K
? _ ???

℘ / Spec(OK ) / Spec(OK ) o ?_ α
where we do not have an explicit geometric description of the closed
fibers over the archimedean primes (Figure 7).
Formally, one can enlarge the group of divisors on the arithmetic
surface by adding formal real
P linear combinations of irreducible “closed
vertical fibers at infinity” α λα Fα . Here the fibers Fα are only treated
as formal symbols, and no geometric model of such fibers is provided.
3. ARITHMETIC INFINITY: ARCHIMEDEAN PRIMES 95
2 3 5 7 11 oo
Figure 7. Arithmetic surface over Spec(Z)
The remarkable fact is that Hermitian geometry on the Riemann sur-

faces Xα (C) is sufficient to specify the contribution of such divisors to
intersection theory on the arithmetic surface, even without an explicit
knowledge of the closed fiber.
The main idea of Arakelov geometry is that it is sufficient to work
with the “infinitesimal neighborhood” Xα (C) of the fibers Fα , to have
well defined intersection indices.
If one thinks, by analogy, to the case of the classical geometry of
a degeneration of algebraic curves over a disk ∆, with a special fiber
over 0 the analogous statement would be saying that the geometry of
the special fiber is completely determined by the generic fiber. This is
a very strong statement on the form of the degeneration: for instance
blowing up points on the special fiber is not seen by just looking at the
generic fiber. Investigating this analogy leads one to expect that the
fiber at infinity should behave like the totally degenerate case. This is
the case where one has maximal degeneration, where all the compo-
nents of the closed fiber are P1 ’s and the geometry of the degeneration
is completely encoded by the dual graph, which describes in a purely
combinatorial way how these P1 ’s are joined. The dual graph has a ver-
tex for each component of the closed fiber and an edge for each double
point.
The local intersection multiplicities of two finite, horizontal, irre-
ducible divisors D1 , D2 on XOK is given by
[D1 , D2 ] = [D1 , D2 ]f in + [D1 , D2 ]inf
where the first term counts the contribution from the finite places (i.e.
what happens over Spec(OK )) and the second term is the contribution
of the archimedean primes, i.e. the part of the intersection that happens
over arithmetic infinity.
While the first term is computed in algebro geometric terms, from

the local equations for the divisors Di at P , the second term is defined
as a sum of values of Green functions gα on the Riemann surfaces
Xα (C),
X X
α α
[D1 , D2 ]inf = − α ( gα (P1,β , P2,γ )),
α β,γ
at points
α
{Pi,β | β = 1, . . . , [K(Di ) : K]} ⊂ Xα (C),
for a finite extension K(Di ) of K determined by Di . Here α = 1 for
real embeddings and = 2 for complex embeddings.
For a detailed account of these notions of Arakelov geometry, one
can refer to [41] [59].
Further evidence for the similarity between the archimedean and
the totally degenerate fibers came from an explicit computation of the
Green function at the archimedean places derived by Manin [66] in
terms of a Schottky uniformization of the Riemann surface Xα (C).
Such uniformization has an analog at a finite prime, in terms of p-adic
Schottky groups, only in the totally degenerate case. Another source
of evidence comes from a cohomological theory of the local factors at
the archimedean primes, developed by Consani [35], which shows that
the resulting description of the local factor as regularized determinant
at the archimedean primes resembles mostly the case of the totally
degenerate reduction at a finite prime.
We will present both results in the light of the noncommutative
space (OA , H, D) introduced in the previous section. As showed by
Consani and the author in [36] [37] [38] [39] the noncommutative
geometry of this space is naturally related to both Manin’s result on
the Arakelov Green function and the cohomological construction of
Consani.
4. Arakelov geometry and hyperbolic geometry

In this section we give a detailed account of Manin’s result [66] on
the relation between the Arakelov Green function on a Riemann surface
X(C) with Schottky uniformization and geodesics in the 3-dimensional
hyperbolic handlebody XΓ . Our exposition follows closely the seminal
paper [66].
P
4.1. Arakelov Green function. Given a divisor A = x mx (x)
with support |A| on a smooth compact Riemann surface X(C), and
a choice of a positive real–analytic 2-form dµ on X(C), the Green
4. ARAKELOV GEOMETRY AND HYPERBOLIC GEOMETRY 97
function gµ,A = gA is a real analytic function on on X(C)\|A|, uniquely

determined by the following conditions:
• Laplace equation: gA satisfies ¯
P ∂ ∂ gA = πi (deg(A) dµ − δA ),
with δA the δ–current ϕ 7→ x mx ϕ(x).
• Singularities: if z is a local coordinate in a neighborhood of x,
then gA − mx log |z| is locally real analytic.
R
• Normalization: gA satisfies X gA dµ = 0.
P
If B = y ny (y) is another divisor, such that |A| ∩ |B| = ∅, then
P
the expression gµ (A, B) := y ny gµ,A (y) is symmetric and biadditive
in A, B. Generally, such expression gµ depends on µ, where the choice
of µ is equivalent to the choice of a real analytic Riemannian metric on
X(C), compatible with the complex structure.
However, in the special case of degree zero divisors, deg A = deg B =
0, the gµ (A, B) = g(A, B) are conformal invariants.
In the case on the Riemann sphere P1 (C), if wA is a meromorphic
function with Div(wA ) = A, we have
dwA
Y Z
ny
(4.19) g(A, B) = log |wA (y)| = Re ,
y∈|B| γB w A
where γB is a 1–chain with boundary B.

In the case of degree zero divisors A, B on a Riemann surface of
higher genus, the formula (4.19) can be generalized, replacing the loga-
rithmic differential dwA /wA with a differential of the third kind (mero-
morphic differential with nonvanishing residues) ωA with purely imag-
inary periods and residues mx at x. This gives
Z
(4.20) g(A, B) = Re ωA .
γB
Thus, one can explicitly compute g(A, B) from a basis of differen-

tials of the third kind with purely imaginary periods.
4.2. Cross ratio and geodesics. The basic step leading to the
result of Manin, expressing the Arakelov Green function in terms of
geodesics in the hyperbolic handlebody XΓ , is a very simple classical
fact of hyperbolic geometry, namely the fact that the cross ratio of four
points on P1 (C) can be expressed in terms of geodesics in the ‘interior’
H3 :
(4.21) log |ha, b, c, di| = −ordist (a ∗ {c, d}, b ∗ {c, d}) .
Here, following [66], ordist denotes the oriented distance, and we use
the notation a ∗ {c, d} to indicate the point on the geodesic {c, d} in H3
b*{c,d}
a {c,d}
b
a*{c,d}
Figure 8. Cross ratio and geodesic length
with endpoints c, d ∈ P1 (C), obtained as the intersection of {c, d} with

the unique geodesic from a that cuts {c, d} at a right angle (Figure 8).
4.3. Differentials and Schottky uniformization. The next im-

portant step in Manin’s result [66] is to show that, if X(C) = Γ\ΩΓ is
a Riemann surface with a Schottky uniformization, then one obtains a
basis of differentials of the third kind with purely imaginary periods,
by taking suitable averages over the group Γ of expressions involving
the cross ratio of points on P1 (C).
We denote by C(|γ) a set of representatives of Γ/(γ Z ), by C(ρ|γ)
a set of representatives for (ρZ ) \ Γ/(γ Z ), and by S(γ) the conjugacy
class of γ in Γ.
Let wA be a meromorphic function on P1 (C) with divisor A =
(a) − (b), such that the support |A| is contained in the complement of
an open neighborhood of ΛΓ . We use the notation
(a − b)(c − d)
(4.22) ha, b, c, di :=
(a − d)(c − b)
for the cross-ratio of points a, b, c, d ∈ P1 (C).
For a fixed choice of a base point z0 ∈ ΩΓ , the series
X
(4.23) ν(a)−(b) := d logha, b, γz, γz0 i
γ∈Γ
4. ARAKELOV GEOMETRY AND HYPERBOLIC GEOMETRY 99
gives the lift to ΩΓ of a differential of the third kind on the Riemann sur-
face X(C), endowed with the choice of Schottky uniformization. These
differentials have residues ±1 at the images of a and b in X(C), and
they have vanishing ak periods, where {ak , bk }k=1...g are the generators
of the homology of X(C).
Similarly, we obtain lifts of differentials of the first kind on X(C),
by considering the series
X
(4.24) ωγ = d loghhz + (γ), hz − (γ), z, z0 i,
h∈C(|γ)
where we denote by {z + (γ), z − (γ)} ⊂ ΛΓ the pair of the attractive and

repelling fixed points of γ ∈ Γ.
The series (4.23) and (4.24) converge absolutely on compact sets
K ⊂ ΩΓ , whenever dimH ΛΓ < 1. Moreover, they do not depend on
the choice of the base point z0 ∈ ΩΓ .
In particular, given a choice {γk }gk=1 of generators of the Schottky
group Γ, we obtain by (4.24) a basis of holomorphic differentials ωγk ,
that satisfy
√
Z
(4.25) ωγ` = 2π −1 δk` .
ak
One can then use a linear combination of the holomorphic differ-

entials ωγk to correct the meromorphic differentials ν(a)−(b) in such a
way that the resulting meromorphic differentials have purely imaginary
bk –periods. Let X` (a, b) be coefficients such that the differentials of the
third kind
X
(4.26) ω(a)−(b) := ν(a)−(b) − X` (a, b)ωg`
`
have purely imaginary bk –periods. The coefficients X` (a, b) satisfy the

system of equations
(4.27)
g Z
X X
X` (a, b) Re τkl = Re ν(a)−(b) = log |ha, b, z + (h), z − (h)i|.
`=1 bk h∈S(gk )
Thus, one obtains ([66], cf. also [93]) from (4.20) and (4.26) that
the Arakelov Green function for X(C) with Schottky uniformization
can be computed as
P
g((a) − (b), (c) − (d)) = h∈Γ log |ha, b, hc, hdi|
(4.28)
− g`=1 X` (a, b) h∈S(g` ) log |hz + (h), z − (h), c, di|.
P P
Notice that this result seems to indicate that there is a choice of

Schottky uniformization involved as additional data for Arakelov ge-
ometry at arithmetic infinity. However, as we remarked previously, at
least in the case of archimedean primes that are real embeddings (as
is the case of arithmetic infinity for Q), the Schottky uniformization
is determined by the real structure, by splitting X(C) along the real
locus X(R), when the latter is nonempty.
4.4. Green function and geodesics. By combining the basic

formula (4.21) with the formula (4.28) for the Green function on a
Riemann surface with Schottky uniformization one can replace each
term appearing in (4.28) with a corresponding term which computes
the oriented geodesic length of a certain arc of geodesic in XΓ :
(4.29) P
g((a) − (b), (c) − (d)) = − h∈Γ ordist(a ∗ {hc, hd}, b ∗ {hc, hd})
+ g`=1 X` (a, b) h∈S(g` ) ordist(z + (h) ∗ {c, d}, z − (h) ∗ {c, d}).
P P
The coefficients X` (a, b) can also be expressed in terms of geodesics,

using the equation (4.27).
5. Intermezzo: Quantum gravity and black holes
Anti de Sitter space AdSd+1 is a highly symmetric space–time,
which satisfies Einstein’s equations with constant curvature R < 0.
Physically it describes empty space with a negative cosmological con-
stant. In order to avoid time–like closed geodesics, it is customary to
pass to the universal cover AdS ˜ d+1 , whose boundary at infinity of the
is a compactification of d–dimensional Minkowski space–time. When
passing to Euclidean signature AdS ˜ d+1 becomes the real hyperbolic
d+1
space H .
The 3 + 1 dimensional Anti de Sitter space is a well known example
of space–time in general relativity. Topologically AdS 3+1 is of the form
S 1 × R3 , while metrically it is realized by the hyperboloid −u2 − v 2 +
x2 + y 2 + z 2 = 1 in R5 , with the metric element ds2 = −du2 − dv 2 +
dx2 + dy 2 + dz 2 . The universal cover is topologically R4 . In the context
of quantum gravity, it is especially interesting to consider the case of
the 2 + 1 dimensional Anti de Sitter space AdS2+1 and its Euclidean
counterpart, the real 3-dimensional hyperbolic space H3 .
The holography principle postulates the existence of an explicit
correspondence between gravity on a bulk space which is asymptotically
˜ d+1 (e.g. a space obtained as a global quotient of AdS
AdS ˜ d+1 by a
discrete group of isometries) and field theory on its conformal boundary
at infinity.
5. INTERMEZZO: QUANTUM GRAVITY AND BLACK HOLES 101
The relation (4.21), that identifies the Green function

g((a) − (b), (c) − (d))
on P1 (C) with the oriented length of a geodesic arc in H3 , can be
thought of as an instance of this holography principle, when we in-
terpret one side as geodesic propagator on the bulk space in a semi-
classical approximation, and the other side (the Green function) as
the two point correlation function of the boundary field theory. No-
tice that, because of the prescribed behavior of the Green function at
the singularities given by the points of the divisor, our “four-points”
g((a) − (b), (c) − (d)), when a → c and b → d, gives the two point cor-
relator with a logarithmic divergence which is intrinsic and does not
depend on a choice of cut-off functions (unlike the regularization often
used in the physics literature).
In [69] we showed that the relation (4.29) between Arakelov Green
functions and configurations of geodesics in the hyperbolic handlebody,
proved by Manin in [66], provides in fact precisely the correspondence
prescribed by the holography principle, for a class of 2 + 1 dimen-
sional space–times known as Euclidean Krasnov black holes. These in-
clude the Bañados–Teitelboim–Zanelli black holes: an important class
of space–times in the context of (2 + 1)-dimensional quantum gravity.
5.1. Bañados–Teitelboim–Zanelli black hole. We consider the

case of a hyperbolic handlebody of genus one (a solid torus), with con-
formal boundary at infinity given by an elliptic curve.
Recall that we can describe elliptic curves via the Jacobi uniformiza-
tion. (We have already encountered it in the previous lectures, in the
context of non-commutative elliptic curves.) Let Xq (C) = C∗ /q Z be
such description of an elliptic curve, where q is a hyperbolic element of
PSL(2, C) with fixed points {0, ∞} on the sphere at infinity P1 (C) of
H3 , that is, q ∈ C∗ with |q| < 1. The action of q on P1 (C) extends to
an action on H̄3 = H3 ∪ P1 (C) by
q : (z, y) 7→ (qz, |q|y).
It is easy to see that the quotient by this action
(4.30) Xq = H3 /q Z
is topologically a solid torus, compactified at infinity by the conformal
boundary Xq (C).
The space Xq is well known in the physics literature as the Euclidean
Bañados–Teitelboim–Zanelli black hole, where the parameter q ∈ C∗ is
Figure 9. Minkowskian BTZ black hole
written in the form

2π(i|r− | − r+ )
q = exp ,
`
with
12
√
r±M ` ± M 2 `2 + J .
=
2
Here M and J are the mass and angular momentum of the rotating
black hole, and −1/`2 is the cosmological constant. The corresponding
black hole in Minkowskian signature is illustrated in Figure 92.
In the case of the elliptic curve Xq (C) = C∗ /q Z , a formula of

Alvarez-Gaume, Moore, and Vafa gives the operator product expan-
sion of the path integral for bosonic field theory as
∞
!
Y
g(z, 1) = log |q|B2 (log |z|/ log |q|)/2 |1 − z| |1 − q n z| |1 − q n z −1 | .
n=1
This is in fact the Arakelov Green function on Xq (C). In terms of

geodesics in the Euclidean BTZ black hole this becomes (cf. [66])
X
1 `γ0 (z̄, 1̄) X
g(z, 1) = − `(γ0 ) B2 + `γ1 (0̄, z̄n ) + `γ1 (0̄, z̃n ).
2 `(γ0 ) n≥0 n≥1
2Figure 9 is taken from [2].

5. INTERMEZZO: QUANTUM GRAVITY AND BLACK HOLES 103
oo
3
H
X
q
-
1
-
x
x -
1 1 -
x
0
1
x
Figure 10. Geodesics in Euclidean BTZ black holes

oo
3 Xq
Η
δ0
-
0
-
xn
γ
0
-x
-x
0 1 - -x 1
1 0 2 x
n
q x
Figure 11. Geodesics in Euclidean BTZ black holes
Here we are using the notation x̄ = x ∗ {0, ∞}; z̄n = q n z ∗ {1, ∞},
z̃n = q n z −1 ∗ {1, ∞} as in [66], as illustrated in Figure 10 and 11.
These terms describe gravitational properties of the Euclidean BTZ
black hole. For instance, `(γ0 ) measures the black hole entropy. The
whole expression is a combination of geodesic propagators.
5.2. Krasnov black holes. The problem of computing the bosonic

field propagator on an algebraic curve XC can be solved by providing
differentials of the third kind with purely imaginary periods
X
ω(a)−(b) := ν(a)−(b) − Xl (a, b)ωgl ,
l
hence it can be related directly to the problem of computing the Arakelov

Green function.
Differentials as above then determine all the higher correlation func-

tions
m X̀
X
G(z1 , . . . , zm ; w1 , . . . , w` ) = qi hφ(zi , z̄i )φ(wj , w̄j )iqj0 ,
j=1 i=1
for qi a system of charges at positions zi interacting with qj0 charges at

positions wj , from the basic two point correlator Gµ (a − b, z) given by
the Green function expressed in terms of the differentials ω(a)−(b) (z).
When we use a Schottky uniformization, we obtain the differentials
ω(a)−(b) as in (4.26).
The bulk space corresponding to the conformal boundary X(C) is
given by the hyperbolic handlebody XΓ . As in the case of the BTZ
black hole, it is possible to interpret these real hyperbolic 3-manifolds
as analytic continuations to Euclidean signature of Minkowskian black
holes that are global quotients of AdS2+1 . This is not just the effect
of the usual rotation from Minkowskian to Euclidean signature, but
a more refined form of “analytic continuation” which is adapted to
the action of the Schottky group, and which was introduced by Kirill
Krasnov [53] [54] in order to deal with this class of space-times.
The formula (4.29) then gives the explicit bulk/boundary corre-
spondence of the holography principle for this class of space-times:
each term in the Bosonic field propagator for XC is expressed in terms
of geodesics in the Euclidean Krasnov black hole XΓ = H3 /Γ.
6. Dual graph and noncommutative geometry

Manin’s result on the Arakelov Green function and hyperbolic ge-
ometry suggests a geometric model for the dual graph of the mysterious
fiber at arithmetic infinity.
In fact, the result discussed above on the Green function has an
analog, due to Drinfel’d and Manin [46] in the case of a finite prime
with a totally split fiber. This is the case where the p-adic completion
admits a Schottky uniformization by a p-adic Schottky group.
6.1. Schottky–Mumford curves. If K is a given finite extension

of Qp , we denote by O ⊂ K its ring of integers, by m ⊂ O the maximal
ideal, and by k the residue field k = O/m. This is a finite field of
cardinality q = card(O/m).
It is well known that a curve X over a finite extension K of Qp ,
which is k-split degenerate, for k the residue field, admits a p-adic
uniformization by a p-adic Schottky group Γ acting on the Bruhat-Tits
tree ∆K .
6. DUAL GRAPH AND NONCOMMUTATIVE GEOMETRY 105
. . . . .. .. . .
.. .
. .
. .
. .
. .
. .
.
.
.. . .
. .
. .
. .
. .
. .
. .
. .
. .
.
. . ..
. .
. .
. .
. .
. .
. . . .
. . . . .
Figure 12. Mumford curve of genus one: Jacobi–Tate
uniformized elliptic curve
The Bruhat-Tits tree is obtained by considering free O-modules of

rank 2, with the equivalence relation that M1 ∼ M2 iff ∃λ ∈ K ∗ , such
that M1 = λM2 . The vertices of the tree consist of these equivalence
classes. A distance is introduced on this set by prescribing that if
{M1 }, {M2 } ∈ ∆0K , with M1 ⊃ M2 , then
M1 /M2 ' O/ml ⊕ O/mk , l, k ∈ N
and we set the distance to be d({M1 }, {M2 }) = |l − k|. We form the
tree ∆K , by adding an edge between any two pairs of vertices with
d({M1 }, {M2 }) = 1. This gives a connected, locally finite tree with
q + 1 edges departing from each vertex. The group PGL(2, K) acts
transitively (on the left) by isometries.
The Bruhat-Tits tree ∆K is the analog of the 3-dimensional real
hyperbolic space H3 at the infinite primes. The set of ends of ∆K is
identified with P1 (K), just as we have P1 (C) = ∂H3 in the case at
infinity.
A p-adic Schottky group Γ is a discrete subgroup of PGL(2, K)

which consists of hyperbolic elements (the eigenvalues of in K have
different valuation), and which is isomorphic to a free group in g gen-
erators. We still denote by ΛΓ ⊂ P1 (K) the limit set, that is, the
closure of the set of points in P1 (K) that are fixed points of some
Figure 13. Genus two: special fibers and dual graphs
γ ∈ Γ \ {1}. As in the case at infinity, we have card(ΛΓ ) < ∞ if and

only if Γ = (γ)Z , for some γ ∈ Γ (the genus one case). We denote by
ΩΓ = ΩΓ (K) the complement ΩΓ = P1 (K) \ ΛΓ , that is, the domain of
discontinuity of Γ.
In the case of genus g ≥ 2 the quotient
XΓ := ΩΓ /Γ
is a Schottky–Mumford curve, with p-adic Schottly uniformization.

The case of genus one gives the Jacobi-Tate uniformization of the el-
liptic curve (Figure 12).
A path in ∆K , infinite in both directions and with no back-tracking,
is called an axis of ∆K . Any two points z1 , z2 ∈ P1 (K) uniquely define
their connecting axis with endpoints at z1 and z2 in ∂∆K . The unique
axis of ∆K whose ends are the fixed points of an hyperbolic element γ
is called the axis of γ. The element γ acts on its axis as a translation.
We denote by ∆0Γ ⊂ ∆K the smallest subtree containing the axes of all
elements of Γ.
This subtree is Γ-invariant, with set of ends ΛΓ . The quotient ∆0Γ /Γ
is a finite graph, which is the dual graph of the closed fiber of the
minimal smooth model over O (k-split degenerate semi-stable curve)
of XΓ . Figure 13 shows the special fiber and the corresponding dual
graph for all the possible cases of maximal degenerations special for
genus two. Figure 14 shows the corresponding trees ∆0Γ .
6. DUAL GRAPH AND NONCOMMUTATIVE GEOMETRY 107
a
b b
a b
a
a a
b b b b
a a
a
b b
a
b b
c
a a c
b
b
a c c a
c b b
c b c
a a a a
b a c b
b c a c
c b a b c b c a
a
a a a
b c b a c
a
a b c ab
c b c a
c b a
c
Figure 14. Genus two: trees and dual graphs
For each n ≥ 0, we can also consider a subgraph ∆K,n of the Bruhat-

Tits tree ∆K defined by setting
∆0K,n := {v ∈ ∆0K : d(v, ∆0Γ ) ≤ n},
where d is the distance on ∆K and d(v, ∆0Γ ) := inf{d(v, ṽ) : ṽ ∈ (∆0Γ )0 },

and
∆1K,n := {w ∈ ∆1K : s(w), r(w) ∈ ∆0K,n }.
In particular, we have ∆K,0 = ∆0Γ .

For all n ∈ N, the graph ∆K,n is invariant under the action of the
Schottky group Γ on ∆, and the finite graph ∆K,n /Γ gives the dual
graph of the reduction XK ⊗ O/mn+1 .
For a more detailed account of Schottky–Mumford curves see [80]
and [67].
6.2. Model of the dual graph. The dictionary between the case
of Mumford curves and the case at arithmetic infinity is then summa-
rized as follows:
Bruhat–Tits ∆K Hyperbolic 3-space H3

1
P (K) = ∂∆K P1 (C) = ∂H3
Schottky Γ ⊂ PGL(2, K) Schottky Γ ⊂ PSL(2, C)
paths in ∆K geodesics in H3
Mumford curve ΩΓ /Γ Riemann surface ΩΓ /Γ
0
tree ∆Γ convex core in H3
graph ∆K /Γ handlebody XΓ
0
dual graph ∆Γ /Γ bounded geodesics in XΓ
Since we can identify bounded geodesics in XΓ with infinite geodesics

in H3 with endpoints on ΛΓ ⊂ P1 (C), modulo the action of Γ, these
are parameterized by the complement of the diagonal in ΛΓ ×Γ ΛΓ .
This quotient is identified with the quotient of the totally disconnected
space S of (4.4) by the action of the invertible shift T of (4.5). Thus we
obtain the following model for the dual graph of the fiber at infinity:
• The solenoid ST of (4.6) is a geometric model of the dual graph
of the fiber at infinity of an arithmetic surface.
• The noncommutative space OA , representing the algebra of
coordinates on the quotient ΛΓ /Γ, corresponds to the set of
“vertices of the dual graph”(set of components of the fiber at
infinity), while the noncommutative space (4.7), corresponding
to the quotient ΛΓ ×Γ ΛΓ gives the set of “edges of the dual
graph”.
Moreover, using noncommutative geometry, one can also give a no-
tion of “reduction mod ∞” analogous to the reduction maps mod pm
defined by the graphs ∆K,n in the case of Mumford curves.
In fact, the reduction map corresponds to the paths connecting ends
of the graph ∆K,n /Γ to corresponding vertices of the graph ∆0Γ /Γ =
∆K,0 /Γ. The analog at arithmetic infinity consists then of geodesics
in XΓ which are the image of geodesics in H3 starting at some point
x0 ∈ H3 ∪ ΩΓ and having the other end at a point of ΛΓ . These are
parameterized by the set
ΛΓ ×Γ (H3 ∪ ΩΓ ).
Thus, in terms of noncommutative geometry, the reduction mod ∞
corresponds to a compactification of the homotopy quotient
ΛΓ ×Γ H3 = ΛΓ ×Γ EΓ
where
BΓ = H3 /Γ = XΓ .
Thus, we can view ΛΓ /Γ as the quotient of a foliation on the ho-
motopy quotient with contractible leaves H3 . The reduction mod ∞ is
7. ARITHMETIC VARIETIES AND L–FACTORS 109
then given by the µ-map

µ : K ∗+1 (ΛΓ ×Γ H3 ) → K∗ (C(ΛΓ ) o Γ).
This shows that the noncommutative space (OA , H, D) is closely re-
lated ot the geometry of the fiber at arithmetic infinity of an algebraic
variety. One can ask then what arithmetic information is captured
by the Dirac operator D of this spectral triples. We’ll see in the next
section that (as proved in [36]), the Dirac operator gives another impor-
tant arithmetic invariant, namely the local L-factor at the archimedean
prime.
7. Arithmetic varieties and L–factors

An important invariant of arithmetic varieties is the L-function.
This is written as a product of contributions from the finite primes and
the archimedean primes,
Y
(4.31) L℘ (H m (X), s).
℘∈Spec OK
We do not plan to give a detailed account on the subject, but we refer

the interested reader to [87]. Here we only try to convey some basic
ideas.
The reason why one needs to consider also the contribution of the
archimedean primes can be seen in the case of the “affine line” Spec(Z),
where one has the Riemann zeta function, which is written as the Euler
product
Y
(4.32) ζ(s) = (1 − p−s )−1 .
p
However, to have a nice functional equation, one needs to consider the

product
(4.33) ζ(s) Γ(s/2)π −s/2,
which includes a contribution of the archimedean prime, expressed in
terms of the Gamma function
Z ∞
(4.34) Γ(s) = e−t ts−1 dt.
0
An analogy with ordinary geometry suggests to think of the func-
tional equation as a sort of “Poincaré duality”, which holds for a com-
pact manifold, hence the need to “compactify” arithmetic varieties by
adding the archimedean primes and the corresponding archimedean
fibers.
When one looks at an arithmetic variety over a finite prime ℘ ∈

Spec(OK ), the fact that the reduction lives over a residue field of posi-
tive characteristic implies that there is a special operator, the geometric
Frobenius F r℘∗ , acting on a suitable cohomology theory (étale cohomol-
ogy), induced by the Frobenius automorphism φ℘ of Gal(F̄p /Fp ).
The local L-factors of (4.31) at finite primes encodes the action of
the geometric Frobenius in the form [87]
−1
(4.35) L℘ (H m (X), s) = det 1 − F r℘∗ N (℘)−s |H m (X̄, Q` )I℘ .
Here we are considering the action of the geometric Frobenius F r℘∗
on the inertia invariants H m (X̄, Q` )I℘ of the étale cohomology. An in-
troduction to étale cohomology and a precise definition of these arith-
metic structures is beyond the purpose of these notes. In fact, our pri-
mary concern will only be the contribution of the archimedean primes
to (4.31), where the construction will be based on ordinary de Rham
cohomology. Thus, we only give a quick and somewhat heuristic ex-
planation of (4.35). We refer to [87] [91] for a detailed and rigorous
account and for the precise hypothesis under which the following holds.
For X a smooth projective algebraic variety (in any dimension)
defined over Q, the notation X̄ denotes
X̄ := X ⊗ Spec(Q̄),
where Q̄ is an algebraic closure. For ` a prime, the cohomology H ∗ (X̄, Q` )
is a finite dim Q` -vector space satisfying
(4.36) H i (X(C), C) ' H i (X̄, Q` ) ⊗ C.
The absolute Galois group Gal(Q̄/Q) acts on H ∗ (X̄, Q` ).
Similarly, we can consider H ∗ (X̄, Q` ) for X defined over a number
field K. For ℘ ∈ Spec(OK ) and ` a prime such that (`, q) = 1, where q
is the cardinality of the residue field at ℘, the inertia invariants
(4.37) H ∗ (X̄, Q` )I℘ ⊂ H ∗ (X̄, Q` )
are the part of the `-adic cohomology where the inertia group at ℘ acts
trivially. The latter is defined as
(4.38) I℘ = Ker(D℘ → Gal(F̄p /Fp )),
with D℘ = {σ ∈ Gal(Q̄/Q) | σ(℘) = ℘}. The Frobenius automorphism
of Gal(F̄p /Fp ) lifts to φ℘ ∈ D℘ /I℘ and induces the geometric Frobe-
nius F℘∗ := (φ−1 ∗ ∗ I℘
℘ ) acting on H (X̄, Q` ) . We use the notation N in
(4.35) for the norm map. Thus, we can write the local L-factor (4.35)
equivalently as
Y m (X I℘
(4.39) L℘ (H m (X), s) = (1 − λq −s )− dim H Γ )λ
,
∗)
λ∈Spec(F r℘
I
where H m (XΓ )λ℘ is the eigenspace of the Frobenius with eigenvalue λ.
For our purposes, what is most important to retain from the dis-
cussion above is that the local L-factor (4.35) depends upon the data
H ∗ (X̄, Q` )I℘ , F r℘∗

(4.40)
of a vector space, which has a cohomological interpretation, together
with a linear operator.
7.1. Archimedean L-factors. Since étale cohomology satisfies
the compatibility (4.36), if we again resort to the general philosophy,
according to which we can work with the smooth complex manifold
X(C) and gain information on the “closed fiber” at airhtmetic infin-
ity, we are lead to expect that the contribution of the archimedean
primes to the L-function may be expressed in terms of the cohomology
H ∗ (X(C), C), or equivalently in terms of de Rham cohomology.
In fact, Serre showed ([87]) that the expected contribution of the
archimedean primes depends upon the Hodge structure
(4.41) H m (X(C)) = ⊕p+q=m H p,q (X(C))
and is again expressed in terms of Gamma functions, as in the case of
(4.33). Namely, one has a product of Gamma functions according to
the Hodge numbers hp,q ,
(4.42)
hp,q
( Q
∗ p,q ΓC (s − min(p, q))
L(H , s) = Q hp,q hp+ p−
ΓR (s − p + 1)h
Q
p<q ΓC (s − p) p ΓR (s − p)
where the two cases correspond, respectively, to the complex and the
real embeddings. Here hp,± is the dimension of the ±(−1)p -eigenspace
of the involution on H p,p induced by the real structure and
(4.43) ΓC (s) := (2π)−s Γ(s), ΓR (s) := 2−1/2 π −s/2 Γ(s/2).
One of the general ideas in arithmetic geometry is that one should
always seek a unified picture of what happens at the finite and at the
infinite primes. In particular, there should be a suitable reformula-
tion of the local factors (4.35) and (4.42) where both formulae can be
expressed in the same way.
Seeking a unified description of local L-factors at finite and infinite
primes, Deninger in [43], [44], [45] expressed both (4.35) and (4.42) as
infinite determinants.
Recall that the Ray-Singer determinant of an operator T with pure

point spectrum with finite multiplicities {mλ }λ∈Spec(T ) is defined as

d
(4.44) det(s − T ) := exp − ζT (s, z)|z=0
∞ dz
where the zeta function of T is defined as
X
(4.45) ζT (s, z) = mλ (s − λ)−z .
λ∈Spec(T )
Suitable conditions for the convergence of these expressions in the case

of the local factors are described in [65].
Deninger showed that (4.39) can be written equivalently in the form
(4.46) L℘ (H m (X), s)−1 = det∞ (s − Θq ),
for an operator with spectrum
(4.47)
2πi log q ∗
Spec(s − Θq ) = (s − αλ ) + n : n ∈ Z, λ ∈ Spec(F rv ) ,
log q 2πi
with multiplicities dλ and with q αλ = λ.
Moreover, the local factor (4.42) at infinity can be written similarly
in the form
−1
q 1
(4.48) L(H (X), s) = det (s − Φ)|Hm
∞ 2π
where Hm is an infinite dimensional vector space and Φ is a linear oper-
ator with spectrum Spec(Φ) = Z and finite multiplicites. This operator
is regarded as a “logarithm of Frobenius” at arithmetic infinity.
Given Deninger’s formulae (4.46) and (4.48), it is natural to ask for
a cohomological interpretation of the data
(4.49) (Hm , Φ)
somewhat analogous to (4.40).
7.2. Arithmetic surfaces: L-factor and Dirac operator. Let
us now return to the special case of arithmetic surfaces, in the case of
genus g ≥ 2.
At an archimedean prime, we consider the Riemann surface Xα (C)
with a Schottky uniformization
X(C) = ΩΓ /Γ.
In the case of a real embedding we can assume that the choice of Schot-
tky uniformization is the one that corresponds to the real structure,
obtained by cutting X(C) along the real locus X(R).
Consider the spectral triple (OA , H, D) associated to the Schottky

group Γ acting on its limit set ΛΓ . Since the spectral triple is not
finitely summable, we do not have zeta functions of the spectral triple
in the form Tr(a|D|z ). However, we can consider the restriction of D
to a suitable subspace of H, where it becomes of trace class.
In particular, consider the zeta function
X
(4.50) ζπV ,D (s, z) := Tr (πV Π(λ, D)) (s − λ)−z
λ∈Spec(D)
where πV is the orthogonal projection of H onto the subspace V of

0 ⊕ L ⊂ H defined by
g
X
(4.51) Π n πV Π n = Sin Si∗ n ,
i=1
on L, where Πn are the spectral projections of the Dirac operator,

Πn = Π(−n, D), for n ≥ 0.
In the case of an arithmetic surface the interesting local factor is the
one for the first cohomology, L(H 1 (X), s). This can be computed ([36],
[38]) from the zeta function (4.50) of the spectral triple (OA , H, D).
Theorem 7.1. The local L-factor (4.42) is given by
−1
1 d s
(4.52) L(H (X), s) = exp − ζπV , D , z |z=0 .
dz 2π 2π
In case of a real embedding, the same holds, with the projection πV,F̄∞ =id
onto +1 eigenspace of the involution F̄∞ induced by the real structure
on V.
This result shows in particular that, for the special case of arith-
metic surfaces with X(C) of genus g ≥ 2, the pair
(4.53) (V ⊂ L, D|V )
is a possible geometric construction of the pair (H1 , Φ) of (4.49). In
particular, the Dirac operator of the spectral triple has an arithmetic
meaning, in as it recovers the “logarithm of Frobenius”
(4.54) D|V = Φ.
If we look more closely at the subspace V of the Hilbert space H,
we see that it has a simple geometric interpretation in terms of the
geodesics in the handlebody XΓ . Remeber that the filtered subspace P
of L = L2 (ΛΓ , dµ) describes 1-cochains on the mapping torus ST , with
P/δP = H 1 (ST ).
The mapping torus ST is a copy of the tangle of bounded geodesics

inside XΓ . Among these geodesics there are g “fundamental” closed
geodesics that correspond to the generators of Γ (they correspond to
geodesics in H3 connecting the fixed points {z ± (γi )}, for i = 1, . . . , g).
Topologically, these are the g core handles of the handlebody XΓ and
they generate the homology H1 (XΓ ).
Consider the cohomology H 1 (ST ). Suppose that we wish to find
elements of H 1 (ST ) supported on these fundamental closed geodesics.
This is impossible, because forms on ST are defined by functions in
C(S, Z), that are supported on some clopen set covering the totally
disconnected set S, which contains no isolated points. However, it is
possible to choose a sequence of 1-cochains on ST whose supports are
smaller and smaller clopen sets containing the infinite word in S corre-
sponding to one of the g fundamental geodesics. The finite dimensional
subspace Vn = V ∩ Pn
Vn ⊂ P n dim Vn = 2g
gives representatives of exactly such cohomology classes in H 1 (ST ).

(We get 2g instead of g because we take into account the two possible
choices of orientation.)
Thus, this gives us, in the case of arithmetic surfaces, a cohomologi-
cal interpretation of the space H1 = V in the pair (4.49). Moreover, the
Schottky uniformization also provides us with a way of expressing this
cohomology H1 = V in terms of the de Rham cohomology H 1 (X(C)).
In fact, we have already seen in the calculation of the Green function
that, under the hypothesis that dimH (ΛΓ ) < 1, to each generator γi of
the Schottky group we can associate a holomorphic differential on the
Riemann surface X(C) by (4.24). The map
X
γi 7→ ωγi = d loghhz + (γi ), hz − (γi ), z, z0 i
h∈C(|γi )
thus gives an identification
(4.55) V ' ⊕n∈Z≥0 H 1 (X(C)).
We’ll see in the next section that, in fact, the right hand side of (4.55)
is a particular case of a more general construction that works for arith-
metic varieties in any dimensions and that gives a cohomological inter-
pretation of (4.49).
8. ARCHIMEDEAN COHOMOLOGY 115
Figure 15. Vanishing cycles
8. Archimedean cohomology
Consani gave in [35], for general arithmetic varieties (in any dimen-
sion), a cohomological interpretation of the pair (Hm , Φ) in Deninger’s
calculation of the archimedean L-factors as regularized determinants.
Her construction was motivated by the analogy between geometry
at arithmetic infinity and the classical geometry of a degeneration over
a disk. She introduced a double complex of differential forms with
an endomorphism N representing the “logarithm of the monodromy”
around the special fiber at arithmetic infinity, which is modelled on
(a resolution of) the complex of nearby cycles in the geometric case.
The definition of the complex of nearby cycles and of its resolution,
on which the following construction is modelled is rather technical.
What is easier to visualize geometrically is the related complex of the
vanishing cycles of a geometric degeneration (Figure 15).
We describe here the construction of [35] using the notation of [40].
We construct the cohomology theory underlying (4.49) in several steps.
In the following we let X = X(C) be a complex compact Kähler
manifold.
Step 1: We begin by considering a doubly infinite graded complex
(4.56) C · = Ω· (X) ⊗ C[U, U −1 ] ⊗ C[~, ~−1 ],

where Ω· (X) is the de Rham complex of differential forms on X, while

U and ~ are formal variables, with U of degree two and ~ of degree
zero.
On this complex we consider differentials
√
(4.57) d0 := ~ d, d00 = −1(∂¯ − ∂),
C C
with total differential δC = + d0C d00C .

We also have an inner product
(4.58) hα ⊗ U r ⊗ ~k , η ⊗ U s ⊗ ~t i := hα, ηi δr,s δk,t
where hα, ηi is the usual Hodge inner product of forms,
Z
(4.59) hα, ηi = α ∧ ∗ C(η̄),
X
√
with C(η) = ( −1 )p−q , for η ∈ Ωp,q (X).
Step 2: We use the Hodge filtration
0
(4.60) F p Ωm (X) := ⊕p0 +q=m, p0 ≥p Ωp ,q (X)
to define linear subspaces of (4.56) of the form
M
(4.61) Cm,2r = F m+r−k Ωm (X) ⊗ U r ⊗ ~k
p+q=m
k≥max{0,2r+m}
and the Z-graded vector space

(4.62) C· = ⊕·=m+2r Cm,2r .
Step 3: We pass to a real vector space by considering
(4.63) T · = (C· )c=id ,
where c denotes complex conjugation.
In terms of the intersection of the Hodge filtrations
(4.64) γ · = F · ∩ F̄ ·
we can write (4.63) as
T · = ⊕·=m+2r T m,2r
where
M
(4.65) T m,2r = γ m+r−k Ωm (X) ⊗ U r ⊗ ~k .
p+q=m
k≥max{0,2r+m}
The Z-graded complex vector space C· is a subcomplex of C · with

respect to the differential d0C , and for P ⊥ the orthogonal projection
onto C· in the inner product (4.58), we obtain a second differential
d00 = P ⊥ d00C . Similarly, d0 = d0C and d00 = P ⊥ d00C define differentials on


k

m, .
Tp,q

2r−2k+ |p−q|+m=0

r
2r+m−k=0
Figure 16. Cutoffs defining the complex at arithmetic

infinity
the Z-graded real vector space T · , since the inner product (4.58) is real
on real forms and induces an inner product on T · . We write δ = d0 + d00
for the total differential.
We can describe the real vector spaces T · in terms of certain cutoffs
on the indices of the complex C · . Namely, for
(4.66) Λp,q = {(r, k) ∈ Z2 : k ≥ κ(p, q, r)}
with

|p − q| + 2r + m
(4.67) κ(p, q, r) := max 0, 2r + m,
2
(cf. Figure 16), we identify T · as a real vector space with the span
(4.68) T · = Rhα ⊗ U r ⊗ ~k i,
¯ with ξ ∈ Ωp,q (X).
where (r, k) ∈ Λp,q for α = ξ + ξ,
8.1. Operators. The complex (T · , δ) has some interesting struc-

tures given by the action of certain linear operators.
We have the operators N and Φ that correspond to the “logarithm
of the monodromy” and the “logarithm of Frobenius”. These are of
the form
∂
(4.69) N = U~ Φ = −U
∂U
and they satisfy [N, d ] = [N, d ] = 0 and [Φ, d0 ] = [Φ, d00 ] = 0, hence
0 00
they induce operators in cohomology.

Moreover, there is another important operator, which corresponds
to the Lefschetz operator on forms,
(4.70) L : η ⊗ U r ⊗ ~k 7→ η ∧ ω ⊗ U r−1 ⊗ ~k ,
where ω is the Kähler form on the manifold X. This satisfies [L, d0 ] =
[L, d00 ] = 0, so it also descends on the cohomology.
The pairs of operators N and Φ or L and Φ satisfy the interesting
commutation relations
[Φ, N ] = −N, [Φ, L] = L
that can be seen as an action of the ring of differential operators
C[P, Q]/(P Q − QP = Q).
8.2. SL(2, R) Representations. Another important piece of the
structure of (T · , δ) are two involutions
(4.71) S : α ⊗ U r ⊗ ~2r+m+` 7→ α ⊗ U −(r+m) ⊗ ~`
(4.72) S̃ : α ⊗ U r ⊗ ~k 7→ C(∗α) ⊗ U r−(n−m) ⊗ ~k .

These maps, together with the nilpotent operators N and L define
two representations of SL(2, R). In terms of

s 0
ν(s) := s ∈ R∗
0 s−1

1 t
(4.73) u(t) := t∈R
0 1

0 1
w := ,
−1 0
the representations σ L and σ R are given by
σ L (ν(s)) = s−n+m σ R (ν(s)) = s2r+m
(4.74) σ L (u(t)) = exp(t L) σ R (u(t)) = exp(t N )
√
σ L (w) = ( −1)n C S̃ σ R (w) = C S.
Of these representations, σ L extends to an action by bounded op-
erators on the Hilbert completion of T · in the inner product (4.58),
while the action of the subgroup ν(s), s ∈ R∗ of SL(2, R) via the rep-
resentation σ R on this Hilbert space is by unbounded densely defined
operators.
8.3. Renormalization group and monodromy. This very gen-
eral structure, which exists for varieties in any dimension, also has in-
teresting connections to noncommutative geometry. For instance, we
can see that in fact the map N does play the role of the “logarithm
of the monodromy” using an analog in our context of the theory of
renormalization à la Connes–Kreimer [28].
In the classical case of a geometric degeneration on a disk, the
monodromy around the special fiber is defined as the map
√
(4.75) T = exp(−2π −1 Res0 (∇))
where
(4.76) N = Res0 (∇)
is the residue at zero of the connection, acting as an endomorphism of
the cohomology.
In our setting, we consider loops φµ with values in the group G =
Aut(TC· , δ), depending on a “mass parameter” µ ∈ C∗ . Here TC· is the
complexification of the real vector space T · . The Birkhoff decomposi-
tion of a loop φµ consists in the multiplicative decomposition
(4.77) φµ (z) = φ−
µ (z)
−1 +
φµ (z),
for z ∈ ∂∆ ⊂ P1 (C), where ∆ is a small disk centered at zero. Of the
two terms in the right hand side of (4.77), φ+
µ extends to a holomorphic
function on ∆ and φµ to a holomorphic functions on P1 (C) r ∆ with
−
values in G. We normalize (4.77) by requiring that φ− µ (∞) = 1.

By analogy with the Connes–Kreimer theory of renormalization we
require the following properties of (4.77):
• The time evolution
(4.78) θt : a 7→ e−tΦ a etΦ
acts by scaling
(4.79) φλµ () = θt φµ (),
for λ = et ∈ R∗+ and ∈ ∂∆.
• The term φ− −
µ = φ in the Birkhoff decomposition is indepen-
dent of the energy scale µ.
The residue of a loop φµ , as in [28], is given by
d
φ− (1/z)−1 |z=0

(4.80) Res φ =
dz
and the beta function of renormalization is

d
(4.81) β = Υ Res φ, θt |t=0 .
with Υ =
dt
In our setting, there is a natural choice of the time evolution, given
by the “geodesic flow” associated to the “Dirac operator” Φ, namely
(4.82) θt (a) = e−tΦ a etΦ .
This gives
d
(4.83) θt (a)|t=0 = [a, Φ].
Υ(a) =
dt
There is a scattering formula ([28]), by which one can reconstruct
−
φ from the residue. Namely, we can write
X
φ− (z)−1 = 1 + dk z −k ,
k≥1
with
Z
(4.84) dk = θ−s1 (β) · · · θ−sk (β) ds1 · · · dsk .
s1 ≥···≥sk ≥0
The renormalization group is given by

(4.85) ρ(λ) = lim φ− () θt (φ− ()−1 ),
→0
for λ = et ∈ R∗+ .
Thus, we only need to specify the residue in order to have the
corresponding renormalization theory associated to (TC· , δ). By analogy
to the case of the geometric degeneration (4.76) it is natural to require
that Res φ = N . We then have ([40]):
Proposition 8.1. A loop φµ in G = Aut(TC· , δ) with Res φµ = N ,
subject to (4.79) and with φ− independent of µ, satisfies
z
µ
φµ (z) = exp N
z
with Birkhoff decomposition
µz − 1

φµ (z) = exp(−N/z) exp N .
z
In fact, by (4.83) and Res φµ = N we have β = [N, Φ] = N and
θt (N ) = et N , hence the scattering formula (4.84) gives
φ− (z) = exp(−N/z)
and the scaling property (4.79) determines

z
µ
φµ (z) = exp N .
z
The part of the Birkhoff decomposition that is regular at z = 0
satisfies φ+ N
µ (0) = µ .
The renormalization group is of the form
(4.86) ρ(λ) = λN = exp(tN ),
which, through the representation σ R of SL(2, R), corresponds to the
horocycle flow on SL(2, R)

1 t
ρ(λ) = u(t) = .
0 1
The Birkhoff decomposition (4.77) gives a trivialization of a princi-
pal G-bundle over P1 (C). We can consider the associated vector bundle
Eµ· with fiber TC· .
Moreover, we obtain a Fuchsian connection ∇µ on this bundle,
(4.87) ∇µ : Eµ· → Eµ· ⊗O∆ Ω·∆ (log 0),
where the notation Ω·∆ (log 0) denotes forms with logarithmic poles at
0. This has the form
d µz − 1

1
∇µ = N + dz,
z dz z
with local gauge potentials
log π(γ) dz +
−φ+ (z)−1 φ (z) + φ+ (z)−1 dφ+ (z)
z
with respect to the monodromy representation
π : π1 (∆∗ ) = Z → G
given by
√
(4.88) π(γ) = exp(−2π −1 N )
for γ the generator of π1 (∆∗ ). This corresponds to (4.75) in the classical
geometric case. This Fuchsian connection has residue
Resz=0 ∇µ = N
as in the geometric case (4.76).
8.4. Local factor and archimedean cohomology. In [35] Con-

sani showed that the data (Hm , Φ) of (4.49) can be identified with
H· (T · , δ)N =0 , Φ ,

where H· (T · , δ) is the hypercohomology (the cohomology with respect

to the total differential δ) of the complex T · and H· (T · , δ)N =0 is the
kernel of the map induced by N on cohomology. The operator Φ is
the one induced on cohomology by that of (4.69). She called H m '
H· (T · , δ)N =0 the archimedean cohomology.
This can also be viewed (cf. [35]) as a piece of the cohomology of
the cone of the monodromy N . This is the complex
(4.89) Cone(N )· = T · ⊕ T · [+1]
with differential
δ −N
D= .
0 δ
The complex (4.89) inherits a positive definite inner product from
T , which descends on cohomology. The representation σ L of SL(2, R)
·
on T · induces a representation on Cone(N )· . The corresponding rep-

resentation dσ L : g → End(T · ) of the Lie algebra g = sl(2, R) extends
to a representation of the universal enveloping algebra U (g) on T · and
on Cone(N )· . This gives a representation in the algebra of bounded
operators on the Hilbert completion of Cone(N )· in the inner product.
Theorem 8.2. The triple
(A, H, D) = (U (g), H· (Cone(N )), Φ)
has the properties that D = D ∗ and that (1 + D 2 )−1/2 is a compact
operator. The commutators [D, a] are bounded operators for all a ∈
U (g) and the triple is 1+ -summable.
Thus, (A, H, D) has most of the properties of a spectral triple,
confirming the fact that the logarithm of Frobenius Φ should be thought
of as a Dirac operator. However, we are not dealing here with an
involutive subalgebra of a C ∗ -algebra.
In any case, the structure is sufficient to consider zeta functions
for this “spectral triple”. In particular, we can recover the alternating
products of the local L-factors at infinity from a zeta function of the
spectral triple.
Theorem 8.3. Consider the zeta function
ζa,Φ (z) = Tr(a|Φ|−z )
with a = σ L (w). This gives

2n
Y m
det (s) = L(H m (X), s)(−1) .
∞ σ L (w),Φ
m=0
8.5. Arithmetic surfaces: homology and cohomology. In

the particular case of arithmetic surfaces, there is an identification
([35], [36])
(4.90) H· (Cone(N )) ' H· ⊕ Ȟ· ,
where H· is the archimedean cohomology and Ȟ· is its dual under the
involution S of (4.71).
We can then extend the identification
'
U : H1 → V ⊂ L
of (4.55), by considering a subspace W of the homology H1 (ST ) with
W ' Ȟ· . The homology H1 (ST ) can also be computed as a direct limit
H1 (ST , Z) = lim KN ,
−→
N
where the KN are free abelian of rank (2g − 1)N + 1 for N even and
(2g − 1)N + (2g − 1) for N odd. The Z-module KN is generated by
the closed geodesics represented by periodic sequences in S of period
N + 1. These need not be primitive closed geodesics. In terms of
primitive closed geodesics we can write equivalently
H1 (ST , Z) = ⊕∞
N =0 RN
where the RN are free abelian groups with

1 X
rk RN = µ(d)rk KN/d ,
N
d|N
P
with µ(d) the Möbius function satisfying d|N µ(d) = δN,1 .
The pairing of homology and cohomology is given by
h·, ·i : Fn × KN → Z h[f ], xi = N · f (x̄).
This determines a graded subspace W ⊂ H1 (ST , Z) dual to V ⊂
H 1 (ST ). With the identification
U
H1 / V ⊂ H 1 (ST )
S <,>

U
Ȟ1 / W ⊂ H1 (ST ),
we can identify the Dirac operator of (4.16) with the logarithm of

Frobenius
(4.91) D|V⊕V = ΦH1 ⊕Ȟ1 .
Bibliography
[1] N. Alling, N. Greenleaf, Foundations of the theory of Klein surfaces, Lecture

Notes in Mathematics Vol. 219, Springer Verlag 1971.
[2] S. Aminneborg, I. Bengtsson, D. Brill, S. Holst, P. Peldan, Black holes and
wormholes in 2 + 1 dimensions, Class. Quantum Grav. 15 (1998) 627–644.
[3] J. Arledge, M. Laca, I. Raeburn, Semigroup crossed products and Hecke
algebras arising from number fields, Doc. Math. 2 (1997) 115–138.
[4] K. I. Babenko, On a problem of Gauss. Dokl. Akad. Nauk SSSR, Tom 238
(1978) No. 5, 1021–1024.
[5] J. D. Barrow. Chaotic behaviour and the Einstein equations. In: Classical
General Relativity, eds. W. Bonnor et al., Cambridge Univ. Press, Cam-
bridge, 1984, 25–41.
[6] P. Baum, A. Connes, Geometric K-theory for Lie groups and foliations.
Preprint IHES 1982; l’Enseignement Mathematique, t. 46, 2000, 1-35.
[7] F.P. Boca, Projections in rotation algebras and theta functions, Commun.
Math. Phys. 202 (1999) 325–357.
[8] J.B. Bost, A. Connes, Hecke algebras, Type III factors and phase transitions
with spontaneous symmetry breaking in number theory, Selecta Math. (New
Series) Vol.1 (1995) N.3, 411–457.
[9] R. Bowen, Hausdorff dimension of quasi–circles, Publ.Math. IHES 50 (1979)
11–25.
[10] M. Boyle, D. Handelman, Orbit equivalence, flow equivalence, and ordered
cohomology, Israel J. Math. 95 (1996) 169–210.
[11] O. Bratteli, D.W. Robinson, Operator algebras and quantum statistical me-
chanics I,II, Springer Verlag, 1981.
[12] P. Chakraborty, A. Pal, Equivariant spectral triples on the quantum SU(2)
group, K-Theory 28 (2003), no. 2, 107–126.
[13] C.H.Chang and D.Mayer, Thermodynamic formalism and Selberg’s zeta
function for modular groups, Regular and chaotic dynamics 15 (2000) N.3
281–312.
[14] P.B. Cohen, A C ∗ -dynamical system with Dedekind zeta partition function
and spontaneous symmetry breaking, Journal de Théorie des Nombres de
Bordeaux 11 (1999) 15–30.
[15] A. Connes, C ∗ algèbres et géométrie differentielle. C.R. Acad. Sci. Paris,
Ser. A-B , 290 (1980) 599–604.
[16] A. Connes, An analogue of the Thom isomorphism for crossed products of a
C ∗ -algebra by an action of R, Adv. in Math. 39 (1981), no. 1, 31–55.
[17] A. Connes, Non–commutative differential geometry, Publ.Math. IHES N.62
(1985) 257–360.
125
126 BIBLIOGRAPHY
[18] A. Connes, Cyclic cohomology and the transverse fundamental class of a

foliation. In: Geometric methods in operator algebras (Kyoto, 1983). Pitman
Res. Notes in Math., 123, Longman, Harlow 1986, 52–144.
[19] A. Connes, Compact metric spaces, Fredholm modules, and hyperfiniteness,
Ergod. Th. Dynam. Sys. (1989) 9, 207–220.
[20] A. Connes, Noncommutative geometry, Academic Press, 1994.
[21] A. Connes, Geometry from the spectral point of view. Lett. Math. Phys. 34
(1995), no. 3, 203–238.
[22] A. Connes, Trace formula in Noncommutative Geometry and the zeros of
the Riemann zeta function. Selecta Mathematica. New Ser. 5 (1999) 29–106.
[23] A. Connes, A short survey of noncommutative geometry, J. Math. Phys. 41
(2000), no. 6, 3832–3866.
[24] A. Connes, Cyclic cohomology, Quantum group Symmetries and the Local
Index Formula for SUq (2), J. Inst. Math. Jussieu 3 (2004), no. 1, 17–68.
[25] A. Connes, M. Douglas, A. Schwarz, Noncommutative geometry and Matrix
theory: compactification on tori. J. High Energy Phys. (1998) no. 2, Paper
3, 35 pp. (electronic)
[26] A. Connes, M. Dubois-Violette, Noncommutative finite-dimensional man-
ifolds. I. spherical manifolds and related examples, Comm. Math. Phys.
Vol.230 (2002) N.3, 539–579.
[27] A. Connes, M. Dubois-Violette, Moduli space and structure of noncommu-
tative 3-spheres, Lett. Math. Phys., Vol.66 (2003) N.1-2, 91–121.
[28] A. Connes, D. Kreimer, Renormalization in quantum field theory and
the Riemann-Hilbert problem. II. The β-function, diffeomorphisms and the
renormalization group. Comm. Math. Phys. 216 (2001), no. 1, 215–241.
[29] A. Connes, M. Marcolli, Quantum Statistical Mechanics of Q-lattices, (From
Physics to Number Theory via Noncommutative Geometry, Part I), preprint
arXiv:math.NT/0404128.
[30] A. Connes, M. Marcolli, Q-lattices: quantum statistical mechanics and Ga-
lois theory, to appear in Journal of Geometry and Physics.
[31] A. Connes, M. Marcolli, N. Ramachandran KMS states and complex multi-
plication, preprint 2004.
[32] A. Connes, H. Moscovici, The local index formula in noncommutative geom-
etry. Geom. Funct. Anal. 5 (1995), no. 2, 174–243.
[33] A. Connes, H. Moscovici, Modular Hecke algebras and their Hopf symmetry,
Moscow Math. Journal, Vol.4 (2004) N.1, 67–109.
[34] A. Connes, H. Moscovici, Rankin-Cohen Brackets and the Hopf Algebra of
Transverse Geometry, Moscow Math. Journal, Vol.4 (2004) N.1, 111–130.
[35] C. Consani, Double complexes and Euler L–factors, Compositio Math. 111
(1998) 323–358.
[36] C. Consani, M. Marcolli, Noncommutative geometry, dynamics and ∞-adic
Arakelov geometry, to appear in Selecta Mathematica
[37] C. Consani, M. Marcolli, Triplets spectreaux in geometrie d’Arakelov,
C.R.Acad.Sci. Paris, Ser. I 335 (2002) 779–784.
[38] C. Consani, M. Marcolli, New perspectives in Arakelov geometry, to appear
in proceedings of CNTA7 meeting Montreal 2002.
[39] C. Consani, M. Marcolli, Spectral triples from Mumford curves, International
Mathematics Research Notices 36 (2003) 1945–1972.
BIBLIOGRAPHY 127
[40] C. Consani, M. Marcolli, Archimedean cohomology revisited, preprint 2004.

[41] G. Cornell and J.H. Silverman (eds.) Arithmetic geometry, Springer-Verlag,
New York, 1986
[42] J. Cuntz, W. Krieger, A class of C ∗ –algebras and topological Markov chains,
Invent. Math. 56 (1980) 251–268.
[43] C. Deninger, On the Γ–factors attached to motives, Invent. Math. 104 (1991)
245–261.
[44] C. Deninger, Local L-factors of motives and regularized determinants. Invent.
Math. 107 (1992), no. 1, 135–150.
[45] C. Deninger, Motivic L-functions and regularized determinants, in “Mo-
tives”, Proceedings of Symposia in Pure Mathematics, Vol. 55 (1994) Part
I, 707–743.
[46] V. Drinfeld, Yu.I. Manin, Periods of p-adic Schottky groups. J. Reine Angew.
Math. 262/263 (1973), 239–247.
[47] R. Haag, Local Quantum Physics, Springer, Berlin 1992.
[48] R. Haag, N. M. Hugenholtz, M. Winnink On the equilibrium states in quan-
tum statistical mechanics, Comm. Math. Phys. 5 (1967), 215, 236.
[49] D. Harari, E. Leichtnam, Extension du phénomène de brisure spontanée
de symétrie de Bost–Connes au cas des corps globaux quelconques, Selecta
Math. (New Series) Vol.3 (1997) 205–243.
[50] G. Harder, General aspects in the theory of modular symbols, Seminar on
number theory, Paris 1981–82 (Paris, 1981/1982), 73–88, Progr. Math., 38,
Birkhäuser Boston, Boston, MA, 1983.
[51] D. Hensley, Continued fraction Cantor sets, Hausdorff dimension, and func-
tional analysis, J. Number Theory 40 (1992) 336–358.
[52] I. M. Khalatnikov, E. M. Lifshitz, K. M. Khanin, L. N. Schur, Ya. G. Sinai.
On the stochasticity in relativistic cosmology. J. Stat. Phys., 38:1/2 (1985),
97–114.
[53] K. Krasnov, Holography and Riemann Surfaces, Adv. Theor. Math. Phys. 4
(2000) 929–979
[54] K. Krasnov, Analytic Continuation for Asymptotically AdS 3D Gravity, gr-
qc/0111049
[55] M. Laca, Semigroups of ∗ -endomorphisms, Dirichlet series, and phase tran-
sitions. J. Funct. Anal. 152 (1998), no. 2, 330–378.
[56] M. Laca, From endomorphisms to automorphisms and back: dilations and
full corners, J. London Math. Soc. (2) 61 (2000) 893–904.
[57] S. Lang, Elliptic Functions, (Second Edition), Graduate Texts in Mathemat-
ics, Vol.112, Springer-Verlag 1987.
[58] G. Landi, An introduction to noncommutative spaces and their geometries,
Lecture Notes in Physics, Vol. m-51, Springer Verlag 1997.
[59] S. Lang, Introduction to Arakelov Theory, Springer-Verlag, New York, 1988.
[60] R. Loudon, The Quantum Theory of Light, third edition, Oxford University
Press, 2000.
[61] L. Mandel, E. Wolf, Optical Coherence and Quantum Optics, Cambridge
University Press, 1995.
[62] Yu.I. Manin, Von Zahlen und Figuren, arXiv:math.AG/0201005.
[63] Yu.I. Manin, Real Multiplication and noncommutative geometry,
arXiv:math.AG/0202109.
128 BIBLIOGRAPHY
[64] Yu.I. Manin, Theta functions, quantum tori, and Heisenberg groups, Lett.
in Math. Phys. 56 (2001) 295–320.
[65] Yu.I. Manin, Lectures of zeta functions and motives (according to Deninger
and Kurokawa). Columbia University Number Theory Seminar (New York,
1992). Astérisque No. 228 (1995), 4, 121–163.
[66] Yu.I. Manin, Three–dimensional hyperbolic geometry as ∞–adic Arakelov
geometry, Invent. Math. 104 (1991) 223–244.
[67] Yu.I. Manin, p-adic automorphic functions. Journ. of Soviet Math., 5 (1976)
279-333.
[68] Yu.I. Manin, Parabolic points and zeta functions of modular curves, Math.
USSR Izvestija, vol. 6 N. 1 (1972) 19–64. Selected Papers, World Scientific,
1996, 202–247.
[69] Yu.I. Manin, M. Marcolli, Holography principle and arithmetic of algebraic
curves, Adv. Theor. Math. Phys. Vol.3 (2001) N.5, 617–650.
[70] Yu.I. Manin, M. Marcolli, Continued fractions, modular symbols, and non-
commutative geometry, Selecta Mathematica (New Series) Vol.8 N.3 (2002)
475-520.
[71] M. Marcolli, Limiting modular symbols and the Lyapunov spectrum, Journal
of Number Theory, Vol.98 N.2 (2003) 348-376.
[72] M. Marcolli, Modular curves, C ∗ -algebras and chaotic cosmology, preprint.
[73] K. Matsuzaki, M. Taniguchi, Hyperbolic manifolds and Kleinian groups, Ox-
ford Univ. Press, 1998.
[74] D.H. Mayer, Relaxation properties of the mixmaster universe, Phys. Lett. A
121 (1987), no. 8-9, 390–394.
[75] D.H. Mayer, Continued fractions and related transformations, 1991.
[76] B. Mazur, Courbes elliptiques et symboles modulaires, Séminaire Bourbaki
vol.1971/72, Exposés 400–417, LNM 317, Springer Verlag 1973, pp. 277–294.
[77] L. Merel, Intersections sur les courbes modulaires, Manuscripta Math., 80
(1993) 283–289.
[78] J.S. Milne, Canonical models of Shimura curves, manuscript, 2003
(www.jmilne.org)
[79] D. Mumford (with M.Nori and P.Norman), Tata lectures on theta, III,
Progress in Mathematics Vol. 97, Birkhäuser 1991.
[80] D. Mumford, An analytic construction of degenerating curves over complete
local rings, Compositio Math. 24 (1972) 129–174.
[81] D. Mumford, C. Series, D. Wright, Indra’s Pearls. The Vision of Felix Klein.
Cambridge University Press, New York, 2002.
[82] W. Parry, S. Tuncel, Classification problems in ergodic theory, London Math.
Soc. Lecture Notes Series 67, 1982.
[83] M. Pollicott, H. Weiss, Multifractal analysis of Lyapunov exponent for con-
tinued fraction and Manneville-Pomeau transformations and applications to
Diophantine approximation, Comm. Math. Phys. 207 (1999), no. 1, 145–171.
[84] D.B. Ray, I.M. Singer, Analytic torsion for complex manifolds. Ann. of Math.
(2) 98 (1973), 154–177.
[85] M.A. Rieffel, C∗ -algebras associated to irrational rotations, Pacific J. Math.
93 (1981) 415-429.
[86] A.L. Rosenberg, Noncommutative schemes. Compositio Math. 112 (1998)
N.1, 93–125.
BIBLIOGRAPHY 129
[87] J. P. Serre, Facteurs locaux des fonctions zêta des variétés algébriques
(définitions et conjectures). Sém. Delange-Pisot-Poitou, exp. 19, 1969/70.
[88] G. Shimura, Arithmetic Theory of Automorphic Functions, Iwanami Shoten
and Princeton 1971.
[89] H.M. Stark, L-functions at s = 1. IV. First derivatives at s = 0, Adv. Math.
35 (1980) 197–235.
[90] P. Stevenhagen, Hilbert’s 12th problem, complex multiplication and Shimura
reciprocity, Advanced Studies in Pure Math. 30 (2001) “Class Field Theory
– its centenary and prospect” pp. 161–176.
[91] A. Weil, Basic Number Theory, Springer 1974.
[92] R.O. Wells, Differential analysis on complex manifolds. Springer–Verlag,
1980.
[93] A. Werner, Local heights on Mumford curves. Math. Ann. 306 (1996), no. 4,
819–831.
[94] D. Zagier, Modular forms and differential operators, in K. G. Ramanathan
memorial issue. Proc. Indian Acad. Sci. Math. Sci. 104 (1994), no. 1, 57–75.

Lectures On Arithmetic Noncommutative Geometry

Uploaded by

Copyright:

Available Formats

Lectures On Arithmetic Noncommutative Geometry

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lectures On Arithmetic Noncommutative Geometry

Uploaded by

Copyright:

Available Formats

Lectures on Arithmetic

(T.S. Eliot, “The Love Song of J. Alfred Prufrock”)

Noncommutative geometry, as developed by Connes starting in the

of Connes and Moscovici [33] [34] on the modular Hecke algebras.

1. The NCG dictionary

It is important to notice that, usually, the notions of noncommuta-

(A, H, D) that describes a classical Riemannian spin manifold is given

Definition 3.1. A spectral triple (A, H, D) consists of a C∗ -algebra

Since we are mostly interested in the relations between noncom-

3.1. Volume form. A spectral triple (A, H, D) is said to be of

the Dixmier trace Trω (T ) is the coefficient of this logarithmic diver-

is bounded though possibly non-convergent. For this reason, the usual

where Π(λ, |D|) denotes the orthogonal projection on the eigenspace

There is a family of zeta functions associated to a spectral triple

Under suitable hypothesis on the asymptotic expansion of (1.11) (cf.

3.4. Infinite dimensional geometries. The main difficulty in

3.5. Spectral triples and Morita equivalences. If (A1 , H, D)

Notice that we need a Hermitian connection ∇, because commuta-

is sufficient to consider projections (P 2 = P , P = P ∗ ). In fact, we

Noncommutative modular curves

where H2 is the 2-dimensional real hyperbolic plane, namely the upper

Figure 1. Fundamental domains for PSL(2, Z)

An example of finite index subgroups is given by the congruence

The unique cusp point corresponds to the degeneration of the elliptic

1.1. Modular symbols. Given two points α, β ∈ H2 ∪ P1 (Q), a

Figure 2. Geodesics between cusps define modular

where α has continued fraction expansion α = [a0 , . . . , an ], and

These groups are related by the pairing

Figure 3. The cell decomposition of the modular com-

This gives a cell decomposition of H2 adapted to the action of

is generated by the relative cycles ghi, ρi, g ranging over representatives

2. The noncommutative boundary of modular curves

The algebra Aθ can be realized as a subalgebra of bounded operators

generating system (U, V ) with the relation

3.2. Degenerations of elliptic curves. An elliptic curve Eτ over

as the quotient of C∗ by the action of the group generated by a single

More precisely, when one considers degenerations of elliptic curves

(2.17) K0 (Eθ ) = K1 (Aθ ) and K1 (Eθ ) = K0 (Aθ ),

which is again compatible with the identification of the Eθ (rather than

3.3. Morita equivalent NC tori. To extend the modular in-

Connes constructed in [15] explicit bimodules realizing the Morita

In Boca’s construction, one obtains elements ξ from Gaussian ele-

4. Limiting modular symbols

Lemma 4.1. Γ orbits in P1 (R) × P are the same as equivalence

in terms of the dynamical system. In fact, even when the

4.3. Limiting modular symbols. Let γβ be an infinite geodesic

[70], where ∗ in the first argument indicates the independence on the

4.3.1. Dynamics of continued fractions. As above, we consider on

The Pn , Qn are polynomials with integral coefficients, which can be

4.3.2. Lyapunov spectrum. A measure of how chaotic a dynamical

The function λ(β) is T –invariant. Moreover, in the case of the clas-

with qn (β) the successive denominators of the continued fraction ex-

There is, however, an exceptional set in [0, 1] of Hausdorff dimension

4.3.3. Limiting modular symbols and iterated shifts. We introduce

where T is the shift of (2.24) and t0 ∈ P.

Figure 5. Geodesics defining limiting modular symbols

representing the same homology class in H1 (X G , Z).

More generally, the Ruelle transfer operator of a map T is defined

In the case of the shift T of (2.24) we have in fact that

by covolume. Namely, one considers the functions ek,a := ck k,a , where