A.

Chambert-Loir

Part II
Lectures on height zeta functions: At the
confluence of algebraic geometry, algebraic
number theory, and analysis

Antoine Chambert-Loir

1 Introduction
1.1 Diophantine equations and geometry
1.1.1 Diophantine equations
Broadly speaking, arithmetic is the study of diophantine equations, that is, systems of polynomial
equations with integral coefficients, with a special emphasis on their solutions in rational integers.
Of course, there are numerous variants, the most obvious ones allowing to consider coefficients
and solutions in the field of rational numbers, or in more general number fields, or even in more
general fields, e.g., finite fields.
The reader should be warned that, in this generality, we are constrained by the undecidability
theorem of [32]: there is no general method, that is no algorithm, to decide whether or not any given
polynomial system has solutions in rational integers. Any mathematician working on diophantine
equations is therefore obliged to consider specific types of diophantine equations, in the hope that
such an undecidability issues do not apply within the chosen families of equations.

1.1.2 Enters geometry

At first, one is tempted to sort equations according to the degrees of the polynomials which
intervene. However, this approach is much too crude, and during the xxth century, mathematicians
were led to realize that there are profound relations between the given diophantine equation and
the geometry of its solutions in real or complex numbers. This led to considerations of geometric
invariants such as the genus of an algebraic curve, to the (essentially opposite) notions of a variety
of general type and a Fano variety, to the notion of a rational variety, etc.
In this survey, we are interested in diophantine equations having infinitely many solutions. A
natural way to describe this infinite set consists in sorting the solutions according to their size (as
integers) and in studying the asymptotic behaviour of the number of solutions of size smaller than
a growing bound.

1.1.3 The circle problem

The classical circle problem in analytic number theory is to estimate the number of integer vectors
x ∈ Zn such that s(x) 6 B, when B → ∞ and s(·) is an appropriate notion of a size of a vector
in Rn . When s(·) = k·k is a norm, for example the euclidean norm, this amounts to counting the
number of lattice points in a ball with center 0 and of radius B.

17
 A. Chambert-Loir

Since a ball is convex, the leading term is easily seen to equal the volume of this ball; in other
words,
Card{x ∈ Zn ; kxk 6 B} ∼ B n V1 ,
where V1 is the volume of the unit ball in Rn . The study of the error term, however, is of a much
more delicate nature. For n = 2, there is an easy O(B)-bound which only requires the Lipschitz
property of the boundary of the unit ball; when the norm is euclidean, one can prove a O(B 2/3 )
bound using the positivity of the curvature of the boundary; however, the conjectured O(B 1/2 )-
bound remains open.

1.2 Elementary preview of Manin’s problem

1.2.1 Projective space
As indicated above, the size of a solution of a diophantine equation can be thought of as a measure
of its complexity; for a vector x ∈ Zn , it is standard to define its size to be its euclidean norm.
However, in algebra and geometry, we are often led to consider rational functions and their
poles, which inevitably bring us to “infinity”. The appropriate context to define and study the
corresponding sizes, or as we shall now say, heights, is that of projective geometry. Recall that for
any field F , one defines the n-dimensional projective space Pn (F ) on F to be the set of lines in
the vector space F n+1 . In other words, this is the quotient set of the set F n+1 \ {0} of nonzero
vectors modulo the action of homotheties: a point in Pn (F ) can be described by a nonzero family
of n + 1 homogeneous coordinates [x0 : · · · : xn ], while we are free to multiply these coordinates by
a common nonzero element of F , still describing the same point.
To stick to the current terminology of algebraic geometry, this set Pn (F ) will be called the set
of F -rational points of the projective space Pn and the latter will be referred to as the scheme Pn .

1.2.2 Heights
Consider the case where F = Q, the field of rational numbers. Any point x ∈ Pn (Q) can be
represented by n + 1 rational numbers, not all zero; however, if we multiply these homogeneous
coordinates by a common denominator, we see that we may assume them to be integers; we then
may divide them by their greatest common divisor and obtain a system of homogeneous coordinates
[x0 : · · · : xn ] made of n + 1 coprime integers. At this point, only one choice is left to us, namely
multiplying this system by −1.
Consequently, we may define the exponential height of x as H(x) = max(|x0 | , . . . , |xn |) and its
logarithmic height as h(x) = log H(x). (Observe the notation, popularized by Serge Lang: small
“h” for logarithmic height, capital “H” for exponential height.)

1.2.3 The theorems of Northcott and Schanuel

[33] made a fundamental, albeit trivial, observation: for any real number B, there are only finitely
many points x ∈ Pn (Q) such that H(x) 6 B. Indeed, this amounts to saying that there are only
finitely many systems of coprime integers [x0 : · · · : xn ] such that |xi | 6 B for all i, considered up
to multiplication by ±1. Concretely, there are at most (2B + 1)n+1 such systems of integers, let it
be coprime, and modulo ±1!
More precisely, [37] proved that the number N (B) of such points satisfies

2n
N (B) ∼ B n+1
ζ(n + 1)

18
 A. Chambert-Loir

when B → ∞, where ζ(n + 1) is Riemann’s zeta function evaluated at n + 1. When, say, n = 1,

this can be interpreted as follows: for B → ∞, the probability that two integers from [1, B] are
coprime tends to 1/ζ(2) = 6/π 2 , a result originally due to Dirichlet; see Theorem 332 in the
classic book of [27].
The remarkable appearance of this notorious function reveals that something profound is hap-
pening before our eyes, something that would certainly appeal to any arithmetician.

1.2.4 The problem of Batyrev–Manin

More generally, we want to consider subsets of the projective space defined by polynomial equations.
Evaluating a polynomial in n + 1 variables at a system of homogeneous coordinates of a point
x ∈ Pn (F ) does not lead to a well-defined number, since the result changes very much if we
multiply the coordinates by a common nonzero factor λ. However, if our polynomial is assumed
to be homogeneous of some degree, the result is multiplied by λd ; at least we can consistently say
whether the result is zero or not.
Let thus X ⊂ Pn be a closed subscheme of Pn , namely the set of common zeroes of a family of
homogeneous polynomials in n + 1 variables. In other words, for any field F , X(F ) is the subset
of Pn (F ) consisting of points satisfying this family of equations. In fact, we will also consider the
case of locally closed subschemes of Pn , that is, subschemes of the form X \ Y , where Y is a closed
subscheme of Pn contained in a closed subscheme X.
We now define NX (B) to be the number of points x ∈ X(Q) such that H(x) 6 B. The problem
of [1] is to understand the asymptotic behaviour of NX (B), for B → ∞. In particular, one is
interested in the two numbers:

− log NX (B) + log NX (B)

βX = lim inf , βX = lim sup .
B→∞ log B B→∞ log B

Observe that −
βX 6 +
βX and that, when X = −
Pn , βX = +
βX = n + 1.

1.2.5 Heath-Brown’s conjecture

However, some problems in analytic number theory require uniform upper bounds, e.g., bounds
depending only on the degree and dimension of X. [28] conjectured that for any real number ε > 0
and any integer d > 0, there exists a constant C(d, ε) such that NX (B) 6 C(d, ε)B dim X+ε , for any
integral closed subscheme X of Pn of degree d and dimension dim X. (Degree and dimensions are
defined in such a way that when H is a general subspace of codimension dim X, (X ∩ H)(C) is a
finite set of cardinality deg X, counted with multiplicities. Here, “integral” means that X is not a
nontrivial union of two subschemes.)
As an example for the conjecture, let us consider the polynomial f = x0 x2 − x1 x3 and the
subscheme X = V (f ) in P3 defined by f . Considering lines on X, it may be seen that NX (B) 
B2.
In fact, [11] established that this conjecture is equivalent to the same statement with X assumed
to be a hypersurface defined by a single homogeneous polynomial of degree d and proved the case
d > 6. The strategy of the proof comes from [28] (from which the case d = 2 also follows) and uses
previous work of [6] about integral points on plane curves. Consequently, the only open cases are
those for which 3 6 d 6 5 and are the object of active works: to quote only two of them, the case
of smooth hypersurfaces has been completed in [10], while [36] treats the case where d > 4 and X
contains only finitely many linear spaces of codimension 1. As this is not the main topic of this
survey, we refer to these articles for more details and references.

19
 A. Chambert-Loir

1.2.6 Geometric parameters

Let us discuss the parameters at our disposal to describe the counting function NX (B) and the
− +
exponents βX and βX .
The most important of them come from geometry. First of all, there is the dimension, as was
already seen in Heath-Brown’s conjecture.
The dimension appears also in Schanuel’s theorem (precisely, the dimension plus one), but
this is not a good interpretation of the exponent. Namely, one of the insights of [1] was the
interpretation of n + 1 as describing the location of the anticanonical divisor of Pn with respect
to the cone of effective divisors. In effect, the n-differential form
d(x1 /x0 ) ∧ d(x2 /x0 ) ∧ · · · ∧ d(xn /x0 )
has a pole of multiplicity n + 1 along the hyperplane defined by the equation x0 = 0. Note that
the rank of the Néron–Severi group will also enter the final picture — this is the group of classes
of divisors where we identify two divisors if they give the same degree to any curve drawn on the
variety.
Properties of a more arithmetic nature intervene as well: the classical Hasse principle and weak
approximation, but also their refinements within the framework of the theory of Brauer-Manin
obstructions. This explains the appearance of orders of various Galois cohomology groups in
asymptotic formulas.
Were we to believe that the result only depends on the “abstract” scheme X, we would rapidly
find ourselves in contradictions. Indeed, there are many P1 that can be viewed in P2 . On lines, the
number of points of bounded height has the same order of magnitude, but the leading coefficient
can change; for example, for lines with equations, say x0 = 0, or x0 + x1 + x2 = 0, one finds
β ± = 2, but the leading coefficient in NX (B) is 2/ζ(2) in the first case, and only one half of it in
the second case. There are also “non-linear lines” in P2 , namely curves which can be parametrized
by rational functions of one variable. The equation x2 x0 − x21 defines such a curve X 0 for which
±
βX 0 = 1, reflecting its non-linearity. In other words, even if we want to think in terms of an

abstract variety X, the obtained formulas force us to take into account the embedding of X in a
projective space Pn .
There are also more subtle geometric and arithmetic caveats, related not only to X, but to its
closed subschemes; we will consider these later.

1.2.7 The height zeta function

Seeking tools and applications of analysis, we introduce∑ the height zeta function. This is noth-
ing other than the generating Dirichlet series ZX (s) = x∈X(Q) H(x)−s , where s is a complex
parameter.
This series converges for Re(s) large enough; for example, Schanuel’s theorem implies that
it converges for Re(s) > n + 1 and that it defines a holomorphic function on the corresponding
half-plane.
A first, elementary, analytic invariant related to ZX is its abscissa of convergence βX . Since this
Dirichlet series has positive coefficients, a result of Landau implies the inequality
−
βX 6 βX 6 βX
+
.
This is a very crude form of Abelian/Tauberian theorem, which we will use later on, once we
have obtained precise analytic information about ZX . Our goal will be to establish an analytic
continuation for ZX to a meromorphic function on some larger domain of the complex plane. De-
tailed information on the poles of this continuation will then imply a precise asymptotic expansion
for NX (B).

20
 A. Chambert-Loir

2 Heights
2.1 Heights over number fields
2.1.1 Absolute values on the field of rational numbers
Let us recall that an absolute value |·| on a field F is a function from F to R+ satisfying the
following properties:

• |a| = 0 if and only if a = 0 (non-degeneracy);

• |a + b| 6 |a| + |b| for any a, b ∈ F (triangular inequality);

• |ab| = |a| |b| for any a, b ∈ F (multiplicativity).

Of course, the usual modulus on the field of complex numbers satisfies these properties, hence
is an absolute value on C. It induces an absolute value on any of its subfields, in particular on Q.
It is called archimedean since for any a ∈ Q such that |a| 6= 0, and any T > 0, there exists an
integer n such that |na| > T . We write |·|∞ for this absolute value.
In fact, the field of rational numbers possesses many other absolute values, namely the p-adic
absolute value, where p is any prime number. It is defined as follows: Any nonzero rational
number a can be written as pm u/v, where u and v are integers not divisible by p and m is a
rational integer; the integer m depends only on a and we define |a|p = p−m ; we also set |0|p = 0.
Using uniqueness of factorization of integers into prime numbers, it is an exercise to prove that
|·|p is an absolute value on Q. In fact, not only does the triangular inequality hold, but a stronger
form is actually true: the ultrametric inequality:

• |a + b|p 6 max(|a|p , |b|p ), for any a, b ∈ Q.

2.1.2 Ostrowski’s theorem

The preceding list gives us essentially all absolute values on Q. Indeed, let |·| be an absolute value
on Q. By Ostrowski’s theorem, |·| is one of the following (mutually exclusive) absolute values:

• the trivial absolute value |·|0 , defined by |a|0 = 1 if a 6= 0 and |0|0 = 0;

• the standard archimedean absolute value |·|∞ and its powers |·|s∞ for 0 < s 6 1;

• the p-adic absolute value |·|p and its powers |·|sp for 0 < s < ∞ and some prime number p.

2.1.3 Topologies and completions

To each of these families corresponds some distance on Q (trivial, archimedean, p-adic), given by
d(a, b) = |a − b|, hence some topology. The trivial absolute value defines the discrete topology
on Q, the archimedean absolute values the usual topology of Q, as a subspace of the real numbers,
and the p-adic absolute values the so-called p-adic topology.
The standard process of Cauchy sequences now constructs for any of these absolute values a
complete field in which Q is dense and to which the absolute value uniquely extends. The obtained
field is Q∞ = R when the absolute value is the archimedean one, and is written Qp when the
absolute value is p-adic.

21
 A. Chambert-Loir

2.1.4 Product formula

A remarkable
∏ equality relates all of these absolute values, namely: for any nonzero rational number
a, p6∞ |a|p = 1. This is called the product formula. (We have written p 6 ∞ to indicate that
the set of indices p ranges
∏ over the set of prime numbers to which we adjoin the symbol ∞.) Let
us decompose a = ± p pnp as a product of a sign and of (positive or negative) prime powers. For
∏
any prime number p, we have |a|p = p−np , while |a|∞ = p pnp ; the product formula follows at
once from these formulae.
From now on, we shall ignore the trivial absolute value.

2.1.5 Number fields

Let F be a number field, that is, a finite extension of Q. Let |·|v be a (non-trivial) absolute value
on F . Its restriction to Q is an absolute value on Q, hence is given by a power of the p-adic
absolute value (for p 6 ∞), as in Ostrowski’s list. The completion furnishes a complete field Fv
which is again a finite extension of Qp ; in particular, we have a norm map N : Fv → Qp . We say
that v is normalized if |a|v = |N (a)|p .
We write Val(F ) for the set of (non-trivial) normalized valuations on F . As in the case of Q,
the product formula holds: for any nonzero a ∈ F ,
∏
|a|v = 1.
v∈Val(F )

2.1.6 Heights on the projective space

Using this machinery from algebraic number theory, we may extend the definition of the height of
a point in Pn (Q) to points in Pn (F ).
Let x be a point of Pn (F ), given by a system of homogeneous coordinates [x0 : · · · : xn ] in F ,
not all zero. We may define the (exponential) height of x as
∏
HF (x) = max(|x0 |v , |x1 |v , . . . , |xn |v )
v∈Val(F )

since the right hand side does not depend on the choice of a specific system of homogeneous
coordinates. Indeed, if we replace xi by axi , for some nonzero element of F , the right hand side
gets multiplied by ∏
|a|v = 1.
v∈Val(F )

For F = Q, this definition coincides with the one previously given. Indeed, we may assume
that the coordinates xi are coprime integers. Then, for any prime number p, |xi |p 6 1 (since the
xi s are integers), and one of them is actually equal to 1 (since they are not all divisible by p).
Consequently, the p-adic factor is equal to 1 and

HQ (x) = max(|x0 |∞ , . . . , |xn |∞ ) = H(x).

Using a bit more of algebraic number theory, one can show that for any finite extension F 0 of F ,
and any x ∈ Pn (F ),
HF 0 (x) = HF (x)d , where d = [F 0 : F ].

22
 A. Chambert-Loir

2.2 Line bundles on varieties

2.2.1 The notion of a line bundle
One of the main differences between classical and modern algebraic geometry is the change of
emphasis from “subvarieties of a projective space” to “varieties which can be embed in a projective
space”. Given an abstract variety X, the data of an embedding is essentially described by line
bundles, a notion we now have to explain.
Let X be an algebraic variety over some field F . A line bundle L on X can be thought of as a
“family” of lines L(x) (its fibres) parametrized by the points x of X. There is actually a subtle
point that not only rational points x ∈ X(F ) must be considered, but also points in X(F 0 ), for all
extensions F 0 (finite or not) of the field F . (We will try to hide these kind of complications.)
A line bundle L has sections: over an open subset U of X, such a section s induces maps
x 7→ s(x) for x ∈ U , where s(x) ∈ L(x) for any x. A section can be multiplied by a regular
function: if s is a section of L on U and f is a regular function on U , then there is a section f s
corresponding to the assignment x 7→ f (x)s(x). At the end, the set Γ(U, L) of sections of L on U ,
also written L(U ), has a natural structure of a module on the ring OX (U ) of regular functions
on U .
Sections can also be glued together: if s and s0 are sections of L over open subsets U and U 0
which coincide on U ∩ U 0 , then there is a unique section t on U ∩ U 0 which induces s on U and s0
on U 0 .
What ties all of these lines together is that for every point x ∈ X, there is an open neighbour-
hood U of x, a frame εU which is a section of L on U such that εU (x) 6= 0 for all x ∈ U and such
that any other section s of L on U can be uniquely written as f εU , where f is a regular function
on U .

2.2.2 The canonical line bundle

The sections of the trivial line bundle OX on an open set U are the ring OX (U ) of regular functions
on U .
If X is smooth and everywhere n-dimensional, it possesses a canonical line bundle ωX , defined in
such a way that its sections on an open subset U are precisely the module of n-differential forms,
defined algebraically as the n-th exterior product of the OX (U )-module of Köhler differentials
Ω1OX (U )/F .

2.2.3 The Picard group

There is a natural notion of morphism of line bundles. If f : L → M is such a morphism, it assigns
to any section s ∈ Γ(U, L) a section f (s) ∈ Γ(U, M ) so that the map s 7→ f (s) is a morphism of
OX (U )-modules (i.e., additive and compatible with multiplication by regular functions).
An isomorphism is a morphism f for which there is an “inverse morphism” g : M → L such that
g ◦ f and f ◦ g are the identical morphisms of L and M respectively.
From two line bundles L and M on X, one can construct a third one, denoted L ⊗ M , in such
a way that if ε and η are frames of L and M on an open set U , then ε ⊗ η is a frame of L ⊗ M
on U , with the obvious compatibilities suggested by the tensor product notation, namely

(f ε) ⊗ η = ε ⊗ (f η) = f (ε ⊗ η),

for any regular function on U .

23
 A. Chambert-Loir

Any line bundle L has an “inverse” L−1 for the tensor product; if ε is a frame of L on U , then
a frame of L−1 on U is ε−1 , again with the obvious compatibilities

(f ε)−1 = f −1 ε−1

for any nonvanishing regular function f on U .

These two laws are in fact compatible with the notion of isomorphism. The set of isomorphism
classes of line bundles on X is an Abelian group, which is called the Picard group and denoted
Pic(X).

2.2.4 Functoriality
Finally, if u : X → Y is a morphism of algebraic varieties and L is a line bundle on Y , there is
a line bundle u∗ L on X defined in such a way that the fibre of u∗ L at a point x ∈ X is the line
L(u(x)); similarly, if ε is a frame of L on an open set U of Y , then u∗ ε is a frame of u∗ L on u−1 (U ).
At the level of isomorphism classes, this induces a map u∗ : Pic(Y ) → Pic(X) which is a mor-
phism of Abelian groups.

2.3 Line bundles and embeddings

2.3.1 Line bundles on Pn
As a scheme over a field F , Pn parametrizes hyperplanes of the fixed vector space V = F n+1 . (In
the introduction, we identified a point x ∈ Pn (F ) with the line in V generated by any system of
homogeneous coordinates; this switch of point of view can be restored by duality.) In particular,
to a point x ∈ Pn (F ) corresponds a hyperplane H(x) in V . Considering the quotient vectorspace
V /H(x), we hence get a line L(x).
These lines L(x) form a line bundle on Pn . This line bundle admits sections s0 , . . . , sn defined
on Pn which correspond to the homogeneous coordinates. Observe however that for x ∈ Pn (F ),
si (x) is not a number, but a member of some line. However, given any generator ε(x), of this line,
there are elements xi ∈ F such that si (x) = xi ε(x) and the family [x0 : · · · : xn ] gives homogeneous
coordinates for x.
It is useful to remember the relation xj si (x) = xi sj (x), valid for any x ∈ Pn (F ) with homoge-
neous coordinates [x0 : · · · : xn ] and any couple (i, j) of indices. It will also be necessary to observe
that for any x ∈ Pn , at least one of the si (x) is nonzero — this is a reformulation of the fact that
the homogeneous coordinates of a point are not all zero.
In traditional notation, this line bundle is denoted OPn (1). Its class in the Picard group of Pn
is a generator of this group, which is isomorphic to Z. Similarly, the traditional notation for the
line bundle corresponding to an integer a ∈ Z is OPn (a).
The canonical line bundle of Pn is isomorphic to OPn (−n − 1). In fact, in the open subset Ui
of Pn where the homogeneous coordinate xi is nonzero, xj /xi defines a regular function and one
has a differential form
\
ωi = d(x0 /xi ) ∧ d(x1 /xi ) ∧ · · · ∧ d(xi /xi ) ∧ · · · ∧ d(xn /xi ),

where the hat indicates that one omits the corresponding factor.
On the intersections Ui ∩ Uj , one can check that the expressions si (x)⊗n+1 ωi and sj (x)⊗n+1 ωj
identify the one to the other if one uses the relation xi sj (x) = xj si (x) that we observed. This
implies that these expressions can be glued together as a section of OPn (n+1)⊗ωPn which vanishes
nowhere, thereby establishing the announced isomorphism.

24
 A. Chambert-Loir

2.3.2 Morphisms to a projective space

Now we translate into the language of line bundles the geometric data induced by a morphism f
from a variety X to a projective space Pn . As we have seen, Pn is given with its line bundle OPn (1)
and its sections s0 , . . . , sn which do not vanish simultaneously. By functoriality, the morphism f
furnishes a line bundle f ∗ OPn (1) on X together with (n + 1) sections f ∗ s0 , . . . , f ∗ sn which, again,
do not vanish simultaneously.
A fundamental fact in projective geometry is that this assignment defines a bijection between:
• the set of morphisms f : X → Pn ;
• the set of data (L, u0 , . . . , un ) consisting of a line bundle L on X together with n + 1 sections
u0 , . . . , un which do not vanish simultaneously.
We have only described one direction of this bijection, the other can be explained as follows. For
i ∈ {0, . . . , n}, let Ui be the open subset of X where ui 6= 0; by assumption, U0 ∪ · · · ∪ Un = X.
On Ui , ui is a frame of L and there are regular functions fi0 , . . . , fin on Ui such that, on that open
set, uj = fij ui . This allows to define a morphism of algebraic varieties fi : Ui → Pn by the formula
x 7→ [fi0 (x) : · · · : fin (x)]. It is easy to check that for any couple (i, j), the morphisms fi and fj
agree on Ui ∩ Uj , hence define a morphism f : X → Pn .
A simple example is given by the line bundle OP1 (d) on P1 and the sections s0⊗i0 ⊗ s1⊗i1 , for
i0 + i1 = d. It corresponds to the Veronese embedding of P1 in Pd defined by [x : y] 7→ [xd :
xd−1 y : · · · : xy d−1 : y d ].

2.3.3 Cones in the Picard group

Let X be an algebraic variety and Pic(X)R the real vector space obtained by tensoring the
group Pic(X) with the field of real numbers. Although we shall only study cases where it is
finite-dimensional, this vector space may very well be infinite-dimensional.
The vector space Pic(X)R contains two distinguished cones. The first, Λeff , is called the effective
cone, it is generated by line bundles which have nonzero sections over X. The second, Λample , is
called the ample cone and it is generated by line bundles of the form f ∗ OPn (1), where f is an
embedding of X into a projective space Pn . (Such line bundles are called very ample, a line bundle
which has a very ample power is called ample.)
One has Λample ⊂ Λeff , because a very ample line bundle has nonzero sections, but the inclusion
is generally strict.
For X = Pn itself, one has Pic(X) ' Z, hence Pic(X)R ' R, the line bundle OPn (1) corre-
sponding to the number 1, and both cones are equal to R+ under this identification.

2.4 Metrized line bundles

2.4.1 Rewriting the formula for the height
Since an abstract variety may be embedded in many ways in a projective space, there are as many
possible definitions for a height function on it. To be able to explain what happens, we first will
rewrite the definition of the height of a point x ∈ Pn (F ) (where F is a number field) using line
bundles.
Recall that Val(F ) is the set of normalized absolute values on the number field F . Let x ∈ Pn (F )
and let [x0 : · · · : xn ] be a system of homogeneous coordinates for x. If xi 6= 0, we may write, for
any absolute value v ∈ Val(F ),
( )−1
|xi |v
max(|x0 |v , . . . , |xn |v ) = |xi |v .
max(|x0 |v , . . . , |xn |v )

25
 A. Chambert-Loir

Recalling that si is the section of OPn (1) corresponding to the ith homogeneous coordinate, we
define
|xi |v
ksi (x)kv = .
max(|x0 |v , . . . , |xn |v )
We observe that the right hand side does not depend on the choice of homogeneous coordinates.
Moreover, the product formula implies that
∏ ∏
HF (x) = ksi (x)k−1
v |xi |v = ksi (x)k−1
v .
v∈Val(F ) v∈Val(F )

In other words, we have given a formula for HF (x) as a product over Val(F ) where each factor is
well defined, independently of any choice of homogeneous coordinates.
The reader should not rush to conclusions: so far we have only exchanged the indeterminacy of
homogeneous coordinates with the choice of a specific index i, more precisely of a specific section si .

2.4.2 Metrized line bundles: an example

The notation introduced for ksi (x)k suggests that it is the v-adic norm of the vector si (x) in the
fibre of O(1) at x. A v-adic norm on a F -vector space E is a map k·kv : E → R+ satisfying the
following relations, analogous to those that the v-adic absolute value possesses:
• kekv = 0 if and only if e = 0 (non-degeneracy);
• ke + e0 kv 6 kekv + ke0 kv for any e, e0 ∈ E (triangular inequality);
• kaekv = |a|v kekv for any e ∈ E and any a ∈ F (homogeneity).
In our case, the vector space is the line E = O(1)(x) and the norm of a single nonzero vector
in E determines the norm of any other. Consequently, the formula given for ksi (x)kv (which is a
positive real number) uniquely extends to a norm on E. The formula xj si (x) = xi sj (x) implies
that the given norm does not depend on the initial choice of an index i such that si (x) 6= 0.
Moreover, when the point x varies, the so-defined norms vary continuously in the sense that the
norm of a section s on a Zariski open set U extends to a continuous function from U (FV ) (endowed
with the v-adic topology) to R+ . It suffices to check this fact on open sets which cover Pn and
over which O(1) has frames; on the set where xi 6= 0, si is such a frame and the claim follows by
observing that the given formula is continuous on Ui (Fv ).

2.4.3 Metrized line bundles: definition

We now extend the previous construction to a general definition. Let X be an algebraic variety
over a number field F and let L be a line bundle on X. A v-adic metric on L is the data of v-adic
norms on the Fv -lines L(x), when x ∈ X(Fv ), which vary continuously with the point x. This
assertion means that for any open set U and any section s of L on U , the function U (Fv ) → R+
given by x 7→ ks(x)kv is continuous. Since the absolute value of a regular function is continuous,
it suffices to check this fact for frames whose open sets of definition cover X. A particular type of
metrics is important; we call them smooth metrics. These are the metrics such that the norm of a
local frame is C ∞ if v is archimedean, and locally constant if v is ultrametric.
The construction we have given in the preceding Section therefore defines a v-adic metric on the
line bundle OPn (1) on Pn . This metric is smooth if v is ultrametric, but not if v is archimedean,
because the function (x, y) 7→ max(|x| , |y|) from R2 to R+ is not smooth. A variant of the
construction furnishes a smooth metric in that case, namely the Fubini-Study metric, defined by
replacing max(|x0 |v , . . . , |xn |v ) by (|x0 |2v + · · · + |xn |2v )1/2 .

26
 A. Chambert-Loir

2.4.4 Adelic metrics

The formula for the height features all normalized absolute values of the number field F . We thus
define an adelic metric on a line bundle L to be a family of v-adic metrics on L, for all v ∈ Val(F ).
However, to be able to define a height using such data, we need to impose a compatibility condition
of “adelic type” on all these metrics.
Let us describe this condition. Let U be an affine open subset of X and ε be a frame of L on U .
Let us represent U as a subvariety of some affine space AN defined by polynomials with coefficients
in F .
For any ultrametric absolute value v ∈ Val(F ), the subset ov ⊂ Fv consisting of elements a ∈ Fv
such that |a|v 6 1 is therefore a subring of Fv . We may thus consider the subset U (ov ) = U (Fv )∩oN
v
of U (Fv ). Although it depends on the specific choice of a representation of U as a subvariety
of AN , one can prove that two representations will define the same subsets U (ov ) up to finitely
many exceptions in Val(F ). As a consequence, for any x ∈ U (F ), one has x ∈ U (ov ) up to finitely
many exceptions v (choose a representation where x has coordinates (0, . . . , 0)).
The adelic compatibility condition can now be expressed by requiring that for all v ∈ Val(F ), up
to finitely many exceptions, kε(x)kv = 1 for any x ∈ U (ov ).

2.4.5 Heights for adelically metrized line bundles

Let X be a variety over a number field F and L a line bundle on X with an adelic metric. For any
x ∈ X(F ) and any frame s on a neighbourhood U of x, let us define
∏
HL,s (x) = ks(x)k−1
v .
v∈Val(F )

By definition of an adelic metric, almost all of the terms are equal to 1. If t is any nonzero element
of L(x), there exists a ∈ F ∗ such that t = as(x); then ktkv = |a|v ks(x)kv for any v ∈ Val(F ). In
particular, ktkv = 1 for almost all v ∈ Val(F ) and
∏ ∏ ∏
ktk−1
v = kas(x)k−1
v = |a|−1 −1
v ks(x)kv
v∈Val(F ) v∈Val(F ) v∈Val(F )
  
∏ ∏ ∏
=  |a|−1
v
 ks(x)k−1
v
 = ks(x)k−1
v = HL,s (x)
v∈Val(F ) v∈Val(F ) v∈Val(F )

where we used the product formula to establish the penultimate equality.

As a consequence, one can use any nonzero element of L(x) in the formula HL,s (x); the resulting
product does not depend on the choice of s. We write it HL (x).
For L = OPn (1) with the adelic metric constructed previously, we recover the first definition
of the height. As we shall see, other choices of adelic metrics furnish the same function, up to
a multiplicative factor which is bounded above and below. However, they can be of significative
arithmetical interest.

2.4.6 Properties of metrized line bundles

There is a natural notion of tensor product of adelically metrized line bundles, for which the v-adic
norm of a tensor product e ⊗ e0 (for e ∈ L(x) and e0 ∈ L0 (x)) is equal to kekv ke0 kv . With that
definition, one has
HL⊗L0 (x) = HL (x)HL0 (x).

27
 A. Chambert-Loir

An isometry of adelically metrized line bundles is an isomorphism which preserves the metrics.
The set of isometry classes of metrized line bundles form a group, called the Arakelov-Picard group,
c
and denoted Pic(X). c
The above formula implies that the function from X(F ) × Pic(X) to R∗+ is
linear in the second variable.
Let us moreover assume that X is projective. A v-adic metric on the trivial line bundle is
characterized by the norm of its section 1, which is a nonvanishing continuous function ρv on X(Fv ).
Consequently, an adelic metric ρ on OX is given by a family (ρv ) of such functions which are
almost all equal to 1. (The projectivity assumption on X allows to write X as a finite union of
open subsets U such that X(Fv ) is the union of the sets U (ov ).) One has
∏
Hρ (x) = ρv (x)−1
v∈Val(F )

(finite product). If ρv is identically equal to 1, we set cv = 1. Otherwise, by continuity and

compactness of X(Fv ), there is a positive real number cv such that c−1 v < ρv (x) < cv for any
x ∈ X(Fv ). Let c be the product of all cv (finite product); it follows that c−1 < Hρ (x) < c for any
x ∈ X(F ).
As a consequence, if L and L0 are two adelically metrized line bundles with the same underlying
line bundle, the quotient of the exponential height functions HL and HL0 is bounded above and
below (by a positive real number). This property is best expressed using the logarithmic heights
hL = log HL and hL0 = log HL0 : it asserts that the difference hL − hL0 is bounded.

2.4.7 Functoriality
If f : X → Y is a morphism of algebraic varieties and L is a adelically metrized line bundle on Y ,
the line bundle f ∗ L on X has a natural adelic metric: indeed, the fibre of f ∗ L at a point x is
the line L(f (x)), for which we use its given norm. This construction is compatible with isometry
c ) → Pic(X).
classes and defines a morphism of Abelian groups f ∗ : Pic(Y c At the level of heights,
it implies the equality
HL (f (x)) = Hf ∗ L (x)
for any x ∈ X(F ).

2.4.8 Finiteness property

We have explained that for any real number B, Pn (Q) has only finitely many points of height
smaller than B. As was first observed by [33], this property extends to our more general setting:
if L is an adelically metrized line bundle whose underlying line bundle is ample, then the set of
points x ∈ X(F ) such that HL (x) 6 B is finite.
This can be proved geometrically on the basis of the result for the projective space over Q.
For simplicity of notation, we switch to logarithmic heights. Since L is ample, it is known that
there exists an integer d > 1 such that L⊗d is very ample, hence of the form ϕ∗ O(1) for some
embedding ϕ of X into a projective space Pn . The functoriality property of heights together with
the boundedness of the height when the underlying metrized line bundle is trivial, imply that there
is a positive real number c such that
1 1
hL (x) = hL⊗d (x) > h(ϕ(x)) + c.
d d
Consequently, it suffices to show the finiteness result on Pn . Moreover, the explicit formula given
for the height on Pn shows that
h([x0 : · · · : xn )) > sup h([xi : xj ]),
i,j

28
 A. Chambert-Loir

where the supremum runs over all couples (i, j) such that xi and xj are not both equal to 0. This
reduces us to proving the finitess assertion on P1 .
We also may assume that F is a Galois extension of Q and let G = {σ1 , . . . , σd } be its Galois
group. Let x ∈ F and let us consider the polynomial
∏
Px = (T − σ(x)) = T d + a1 T d−1 + · · · + ad ;
σ∈G

it has coefficients in Q. For v ∈ Val(F ) and σ ∈ G, the function t 7→ |σ(t)|v is a normalized

absolute value on F which is associated to the p-adic absolute value if so is v. Consequently,
∏
max(1, |σ1 (x)|v , . . . , |σd (x)|v ) 6 max(1, |x|w ),
w|p

where the notation w | p means that the product ranges over all normalized absolute values w
on F which are equivalent to the p-adic absolute value. It follows that h([1 : σ1 (x) : · · · : σd (x)) 6
dh([1 : x]).
d
∏d which associates to a point [u0 : · · · : ud ] ∈ P the coefficients [v0 : · · · : vd ]
Moreover, the map
of the polynomial i=1 (u0 T − ui ) has degree d. This implies that

h([v0 : · · · : vd ]) = dh([u0 : · · · : ud ]) + O(1).

If h(x) 6 T , this implies that the point a = [1 : a1 : · · · : ad ] of Pd (Q) is of height h(a) 6 d2 T . By

the finiteness property over Q, there are only finitely many possible polynomials Px , hence finitely
many x since x is one of the d roots of Px .

2.4.9 Rational points vs algebraic points

To keep the exposition at the simplest level, we have only considered the height of F -rational
points. However, there is a suitable notion of adelic metrics which uses not only the v-adic points
X(Fv ), but all points of X over the algebraic closure Fv of Fv , or even its completion Cv . Together
with the obvious extension to metrized line bundles of the relation between HF (x) and HF 0 (x)
when F 0 is a finite extension of F and x ∈ Pn (F ) this allows to define a (exponential) height
function HL on the whole of X(F ), as well as its logarithmic counterpart hL = log HL . The
preceding properties extend to this more general setting, the only nonobvious assertion being the
boundedness of the height when the underlying line bundle is trivial.

3 Manin’s problem
3.1 Counting functions and zeta functions
3.1.1 The counting problem
Let X be a variety over a number field F and let L be an ample line bundle with an adelic metric.
We have seen that for any real number B, the set of points x ∈ X(F ) such that HL (x) 6 B is
a finite set. Let NX (L; B) be its cardinality. We are interested in the asymptotic behaviour of
NX (L; B), when B grows to infinity. We are also interested in understanding the dependence on L.

29
 A. Chambert-Loir

3.1.2 Introducing a generating series

As is common practice in all counting problems, e.g., in combinatorics, one introduces a generating
series. For the present situation, this is a Dirichlet series, called the height zeta function, and
defined by ∑
ZX (L; s) = HL (x)−s ,
x∈X(F )

for any complex number s such that the series converges absolutely. In principle, and we will
eventually do so, one can omit the ampleness condition in that definition, but then the convergence
in some half-plane is not assured (and might actually fail).

3.1.3 Abscissa of convergence

The abscissa of convergence βX (L) is the infimum of all real numbers a such that ZX (L; s) con-
verges absolutely for Re(s) > a. Since the height zeta function is a Dirichlet series with positive
coefficients, a theorem of Landau implies that βX (L) is also the infimum of all real numbers a such
that ZX (L; a) converges.
Let us assume again that L is ample. Then, the proof of Northcott’s theorem shows that
NX (L; B) grows at most polynomially. It follows that ZX (L; s) converges for Re(s) large enough.
For instance, let us assume that NX (L; B)  B a . Let us fix a real number B > 1. There are
 B na points x ∈ X(F ) such that HL (x) 6 B n and the part of the series given by points of
heights between B n−1 and B n is bounded by B na B −(n−1)s = B n(a−s)+s . By comparison with the
geometric series, we see that ZX (L; s) converges for Re(s) > a.
This shows that the function L 7→ βX (L) is well defined for adelically metrized line bundles
whose underlying line bundle is ample. Moreover, it does not depend on the choice of an adelic
metric, so it really comes from a function on the set of ample line bundles in Pic(X). The formula
HL⊗d (x) = HL (x)d implies that βX (L⊗d ) = dβX (L) for any positive integer d. In other words,
βX is homogeneous of degree 1. One can also prove that it uniquely extends to a continuous
homogeneous function on Λample (see [1]).

3.1.4 Analytic properties of ZX (L; ·) vs. asymptotic expansions of NX (L; ·)

Despite this simple definition, very little is known about βX , let alone about the height zeta function
itself. However, there are numerous examples where ZX (L; ·) has a meromorphic continuation to
some half-plane, with a unique pole of largest real value, say αX (L), of order tX (L). By Ikehara’s
Tauberian theorem, this implies an asymptotic expansion of the form

NX (L; B) ∼ cB αX (L) (log B)tX (L)−1 .

In some cases, one can even establish terms of lower order, or prove explicit error terms.
However, these zeta functions are not as well behaved as the ones traditionally studied in alge-
braic number theory. Although some of them have a meromorphic extension to the whole complex
plane, like those of flag varieties (see §4.3.2 below, see also [23]), it happens quite often, for ex-
ample, that they have a natural boundary ; this is already the case for some toric surfaces, see [2,
Example 3.5.4]. Subtler analytic properties of the height zeta functions beyond the largest pole is
the subject of some recent investigations, see, e.g., [9].

30
 A. Chambert-Loir

3.2 The largest pole

3.2.1 The influence of the canonical line bundle
We now assume that X is smooth. One of the insights of mathematicians in the xxth century
(notably Mordell, Lang and Vojta) was that the potential density of rational points is strongly
related by the negativity of the canonical line bundle ωX with respect to the ample cone.
This is quite explicit in the case of curves. Namely, if X is a curve of genus g, three cases are
possible:
−1
1. genus g = 0, ωX ample. Then X is a conic. Two subcases are possible: either X(F ) is
empty—X has no rational point— or X is isomorphic to the projective line. Moreover, the
Hasse principle allows to decide quite effectively in which case we are.

2. genus g = 1, ωX trivial. If X has a rational point, then X is an elliptic curve, endowing

X(F ) with a structure of Abelian group. Moreover, the theorem of Mordell-Weil asserts
that X(F ) is of finite type.

3. genus g > 2, ωX ample. By Mordell’s conjecture, first proved by [24], X(F ) is a finite
set.

The counting function distinguishes very clearly the first two cases. When X is the projective
line, NX (B) grows like a polynomial in B (depending on how the height is defined), and for X an
elliptic curve, NX (B) ≈ (log B)r/2 , where r is the rank of the Abelian group X(F ).

3.2.2 Increasing the base field

The general conjectures of Lang lead us to expect that if ωX is ample, then X(F ) should not be
−1
dense in X, but that this could be expected in the opposite case where ωX is ample.
−1
However, as the case of conics already indicates, geometric invariants like the ampleness of ωX
cannot suffice to decide on the existence of rational points. For example, the conic C over Q given
by the equation x2 + y 2 + z 2 = 0 in the projective plane P2 has no rational points (a sum of three
squares of rational integers cannot be zero, unless they are all zero). However, as soon √ as the
ground field F possesses a square root of −1, then C admits a rational point P = [1 : −1 : 0]
and the usual process of intersecting with the conic C a variable line through that point P gives
us a parametrization of C(F ).
−1
Smooth projective varieties with ωX ample are called Fano varieties. In general, for such a
variety over a number field F , it is expected that there exists a finite extension F 0 of F such that
X(F 0 ) is dense in X for the Zariski topology. This seems to be a very difficult question to settle
in general, unless X has particular properties, like being rational or unirational, in which case a
dense set of points in X(F 0 ) can be parametrized by the points of a projective space Pn (F 0 ) (where
n = dim X).

3.2.3 Eliminating subvarieties

This conjecture considers the “density aspect” of the rational points. Concerning the counting
function, the situation is even more complicated since nothing forbids that most of the rational
points of small height are contained in subvarieties.
Let Y be a subscheme of X. One says that Y is strongly accumulating if the fraction
NY (L, B)/NX (L, B) tends to 1 when B → ∞, and one says that Y is weakly accumulating if the
inferior limit of this fraction is positive.

31
 A. Chambert-Loir

As a consequence, the behaviour of the counting function can only be expected to reflect the
global geometry if one doesn’t count points in accumulating subvarieties. This leads to the notation
NU (L; B) and ZU (L; s), where U is any Zariski open subset in X.
To get examples of such behaviour, it suffices to blow-up a variety at one (smooth) rational
point P . It replaces the point P by a projective space E of dimension n − 1 if X has dimension n
which is likely to possess ≈ B a points of height 6 B, while the rest could be smaller. For example,
if X is an Abelian variety (generalization of elliptic curves in higher dimension), NX (L; B) only
grows like a power of log B.

3.2.4 Definition of a geometric invariant

In order to predict on a geometric basis the abscissa of convergence of ZU (L; s), one needs to
introduce a function on Pic(X)R which is homogeneous of degree 1 and continuous. It is important
to know at that point that for Fano varieties, this vector space Pic(X)R is finite-dimensional
−1
(and identifies with the so-called Néron–Severi group). Indeed, since ωX is ample (this is the
very definition of a Fano variety), Serre’s duality and Kodaira’s vanishing theorem imply that
H1 (X, OX ) = 0, so that the Albanese variety of X is trivial.
After considering many examples ([37], [25], [39], ...), and with detailed investigations of surfaces
at hand, [1] concluded that the relevant part of Pic(X)R was not the ample cone (as the case of
the projective space naı̈vely suggests), but the effective cone Λeff .
In effect, they define for any line bundle L belonging to the interior of the effective cone a real
number αX (L) which is the least real number a such that ωX ⊗ L⊗a belongs to the effective cone
in Pic(X)R .

3.2.5 The conjectures of Batyrev and Manin

The consideration of the effective cone fits nicely with the elimination of accumulating subvarieties.
Indeed, let us assume that the line bundle L belongs to the interior of the effective cone; then
there exists a positive integer d, an ample line bundle M and an effective line bundle E on X such
that L⊗d ' M ⊗ E. (These line bundles are also called big.) For any nonzero section sE of E,
HM (x) is bounded from below where sE does not vanish, so that HL (x)  HM (x)1/d on the open
set U = X \ {sE = 0}. It follows that NU (L; B) grows at most polynomially, and that the abscissa
of convergence βU (L) of ZU (L; s) is finite.
[1] present three conjectures of increasing precision concerning the behaviour of the counting
function. Let X be a projective smooth variety over a number field F , let L be a line bundle on X
which belongs to the interior of the effective cone. The following assertions are conjectured:
1. For any ε > 0, there exists a dense open subset U ⊂ X such that βU (L) 6 αX (L) + ε (linear
growth conjecture).
2. If X is a Fano variety, then for any large enough finite extension F 0 of F , and any small
enough dense open set U ⊂ X, one has βU,F 0 (L) = α(L).
3. Same assertion, only assuming that ωX does not belong to the effective cone.

3.3 The refined asymptotic expansion

3.3.1 Logarithmic powers
In the cases evoked above, the height zeta function appears to possess a meromorphic continuation
to the left of the line Re(s) = αX (L), with a pole of some order t at s = αX (L). By Taube-
rian theory, this implies a more precise asymptotic expansion for the counting function, namely

32
 A. Chambert-Loir

NU (L; B) ≈ B αX (L) (log B)t−1 . A stronger form of the conjecture of Batyrev–Manin also pre-
−1
dicts the order of this pole, at least when L = ωX . In that case, Batyrev and Manin conjecture
that t is the dimension of the real vector space Pic(X)R . (Recall that the Picard group of a Fano
variety has finite rank.)

3.3.2 (In)compatibilities
Some basic checks concerning that conjecture had been made by [25], for instance its compatibility
with products of varieties. However, [4] observed that the conjecture is not compatible with families
and produced a counterexample.
They consider the subscheme V of P3 × P3 defined by the equation x0 y03 + · · · + x3 y33 = 0 (where
[x0 : · · · : x3 ] are the homogeneous coordinates on the first factor, and [y0 : · · · : y3 ] are those on
the second). For fixed x ∈ P3 , the fibre Vx consisting of y ∈ P3 such that (x, y) ∈ V is a diagonal
cubic surface; in other words, V is the total space of the family of diagonal cubic surfaces.
Let p1 and p2 be the two projections from P3 × P3 to P3 . For any couple of integers (a, b), let
O(a, b) be the line bundle p∗1 OP3 (a)⊗p∗2 OP3 (b) on P3 ×P3 . The so-defined map from Z2 to Pic(P3 ×
P3 ) is an isomorphism of Abelian groups. Moreover, since V is an hypersurface in P3 × P3 , the
Lefschetz theorem implies that the restriction map induces an isomorphism from Pic(P3 × P3 )
to Pic(V ). The anticanonical line bundle ωV−1 of V corresponds to O(3, 1) (anticanonical of P3 ×P3
minus the class of the equation, that is (4, 4) − (1, 3)) and the conjecture of Batyrev–Manin
predicts that there should be ≈ B(log B) points of O(3, 1)-height 6 B in V (F ).
Let us fix some x ∈ P3 such that Vx is non-singular. This is a cubic surface, and its anticanonical
line bundle identifies with the restriction of OP3 (1). Thus, if tx = dimR Pic(Vx )R , the conjecture
predicts that there should be ≈ B(log B)tx −1 points of O(1)-height 6 B in Vx (F ).
As a consequence, if the conjecture holds for V and if tx > 2, then the fibre Vx would have
more points than what the total space seem to have. This doesn’t quite contradict the conjecture
however, because it explicitly allows to remove a finite union of subvarieties.
However, the rank of the Picard group can exhibit disordered behaviour in families; for example,
it may not be semi-continuous, and jump on a infinite union of subvarieties. This happens here,
since tx = 7 when F contains the cubic roots of unity and all the homogeneous coordinates of x
are cubes. The truth of the conjecture for V therefore requires to omit all such fibres Vx . But they
form an infinite union of disjoint subvarieties, a kind of accumulating subset which is not predicted
by the conjecture of Batyrev–Manin.
In particular, this conjecture is false, either for V , or for most cubic surfaces. By geometric
considerations, and using their previous work on toric varieties, [4] could in fact conclude that the
conjecture does not hold for V .

3.3.3 Peyre’s refinement of the conjecture

One owes to [34] a precise refinement of the previous conjecture, as well as the verification of
the refined conjecture in many important cases. Indeed, all known positive examples feature an
−1
asymptotic expansion of the form NU (ωX ; B) ∼ cB(log B)t , for some positive real number c,
which in turn is the product of four factors:
b
1. the volume of a suitable subspace X(AF )BM of the adelic space X(AF ) (where the so-called
Brauer-Manin obstruction to rational points vanishes) with respect to the Tamagawa measure
first introduced in that context by [34].

2. the cardinality of the finite Galois cohomology group H1 (Pic(XF ));

33
 A. Chambert-Loir

3. a rational number related to the location of the anticanonical line bundle in the effective
cone Λeff of Pic(X)R ;

4. the rational number 1/t!.

Let us give a short account of Peyre’s construction of a measure. The idea is to use the given
−1
adelic metric on ωX to construct a measure τX,v on the local spaces X(Fv ) for all places v of the
number field F , and then to consider a suitable (renormalized) product of these measures.

3.3.4 Local measures

Ideally, a measure on an n-dimensional analytic manifold is given in local coordinates by an
expression of the form dµ(x) = f (x) dx1 . . . dxn , where f is a positive function. When one considers
another system of local coordinates, the expression of the measure changes, and its modification is
dictated by the change-of-variables formula for multiple integrals: the appearance of the absolute
value of the Jacobian means that dµ(x) is modified as if it were “the absolute value of a differential
form”. One can deduce from this observation that if α is a local (i.e., in a chart) differential n-form,
written α = f (x)dx1 ∧ · · · ∧ dxn in coordinates, then the measure

|α|v |f (x)|v
= dx1 . . . dxn
kαkv kαkv

is well-defined, independently of the choice of α. One can therefore glue all these local measures
and obtain a measure τX,v on X(Fv ).

3.3.5 Convergence of an infinite product

Since X is proper, the adelic space X(AF ) is the (infinite) product of all X(Fv ), for v ∈ Val(F ),
endowed with the product topology. To define a measure on X(AF ) one would like to consider
the (infinite) product of the measures τX,v . However, this gives a finite measure if and only if the
product ∏
τX,v (X(Fv ))
v finite

converges absolutely. This convergence never holds, and one therefore needs to introduce conver-
gence factors to define a measure on X(AF ).
By a formula of [43], the equality τX,v (X(Fv )) = qv− dim X Card(X(kv )) holds for almost all finite
places v, where kv is the residue field of F at v, qv is its cardinality, and X(kv ) is the set of solutions
in kv of a fixed system of equations with coefficients in the ring of integers of F which defines X.
By Weil’s conjecture, established by [16], plus various cohomological computations, this implies
that
1
τX,v (X(Fv )) = 1 + Tr(Frobv | Pic(XF )R ) + O(qv−3/2 ),
qv
where Frobv is a “geometric Frobenius element” of the Galois group of F at the place v.
Let us consider the Artin L-function of the Galois-module P = Pic(XF )R : it is defined as the
infinite product
∏
L(s, P ) = Lv (s, P ), Lv (s, P ) = det(1 − qv−s Frobv |P )−1 .
v finite

34
 A. Chambert-Loir

The product converges absolutely for Re(s) > 1, defines a holomorphic function in that domain.
Moreover, L(s, P ) has a meromorphic continuation to C with a pole of order t = dim Pic(X)R at
s = 1. Let
L∗ (1, P ) = lim L(s, P )(s − 1)−t ;
s→1
it is a positive real number.
Comparing the asymptotic expansions for Lv (1, P ) and τX,v (X(Fv )), one can conclude that the
following infinite product ∏( )
τX = L∗ (1, P ) Lv (1, P )−1 τX,v
v

is absolutely convergent and defines a Radon measure on the space X(AF ).

3.3.6 Equidistribution
One important aspect of the language of metrized line bundles is that it suggests explicitly to look
at what happens when one changes the adelic metric on the canonical line bundle.
Assume for example that there is a number field F and an open subset U such that Manin’s con-
jecture (with precised constant as above) holds for any adelic metric on ωX . Then, [34] shows that
b
rational points in U (F ) of heights 6 B equidistribute towards the probability measure on X(AF )BM
proportional to τX . Namely, for any smooth function Φ on X(AF ),
∑ ∫
1 1
−1 Φ(x) −→ b
Φ(x) dτX (x).
NU (ωX ; B) x∈U (F ) τX (X(AF )BM b
) X(AF )BM
H (x)6B
ω −1
X

As we shall see in the next section, this strengthening by Peyre of the conjecture of Batyrev–
Manin is true in many remarkable cases, with quite nontrivial proofs.

4 Methods and results

4.1 Explicit counting
4.1.1 Projective space
Let F be a number field and let L be the line bundle O(1) on the projective space Pn , with an
adelic metric. [37] has given the following asymptotic expansion for the counting function on Pn :
( )n
Ress=1 ζF (s) 2r1 (2π)r2
NPn (L; B) ∼ √ (n + 1)r1 +r2 −1 B n+1 .
ζF (n + 1) |DF |

In this formula, ζF is the Dedekind zeta function of the field F , r1 and 2r2 are the number of real
embeddings and complex embeddings of F , and DF is its discriminant.
In fact, it has been shown later by many authors, see e.g. [25], that the height zeta function
ZPn (L; s) is holomorphic for Re(s) > n + 1, has a meromorphic continuation to the whole complex
plane C, with a pole of order 1 at the point s = n+1 and no other pole on the line {Re(s) = n+1}.
To prove this estimate, one can sort points [x0 : · · · : xn ] in Pn (F ) according to the class of
the fractional ideal generated by (x0 , . . . , xn ). Constructing fundamental domains for the action of
units, the enumeration of such sets can be reduced to the counting of lattice points in homothetic
sets in R(n+1)(r1 +2r2 ) whose boundary has smaller dimension. The Möbius inversion formula is
finally used to take care of the coprimality condition.

35
 A. Chambert-Loir

4.1.2 Hirzebruch surfaces and other blow-ups

There are many other cases where this explicit method works. Let us mention only some of them:

• ruled surfaces over the projective line (Hirzebruch surfaces);

• Grassmann varieties ([42]) ;

• Del Pezzo surfaces of degree 5, given as blow-ups of the plane in 4 points in general position
([8]);

• some Chow varieties of Pn , see [38].

This method suffers from an obvious drawback: 1) it is hard to take advantage of subtle geometric
properties of the situation studied; 2) it requires to know already much about the rational points,
e.g., to be able to parametrize them. However, this is so far the only way of dealing with the
(eventually singular) Del Pezzo surfaces which are not equivariant compactifications of the torus
Gm 2 or of the additive group Ga 2 .
To deal with the parametrization of the rational points, an essential tool is the universal torsor,
invented by [14], which is a quasi-projective variety lying over the given Del Pezzo surface. The
computation of a universal torsor is essentially equivalent to that of a multi-graded ring defined
in [15], the so-called Cox ring of the variety. In the case of the projective space Pn , the torsor
is An+1 \ {0}, mapping to Pn via (x0 , . . . , xn ) 7→ [x0 : · · · : xn ], and the Cox ring is the graded
polynomial ring in n + 1 variables.
In the context of Manin’s conjecture, universal torsors were first introduced by [35] for toric
varieties, in which case they are isomorphic to an open subset of an affine space. The parametriza-
tion of rational points which they allow to write proved to be very useful for the study of points
of bounded height, at least in contexts where other techniques, like harmonic analysis, do not
apply. Moreover, their computation is also related to the precise understanding of the numerically
effective cone in the Picard group.
[18] computes these universal torsors for smooth Del Pezzo surfaces (that is, blow-ups of the
projective plane in at most 8 points in general position). They were also computed for many
singular Del Pezzo surfaces, see, e.g., the article of [20] for an example leading to a case of Manin’s
conjecture. In that paper, the interested reader shall also find a list of known cases, mostly due to
Browning, de la Bretéche and Derenthal, with references to the articles; see also the more
recent preprint of [19]. Anyway, we apologize not to develop this important part of the subject
here, which would lead us to far of your topic.

4.2 The circle method of Hardy–Littlewood

The circle method was initially devised to tackle Waring’s problem, namely the decomposition
of an integer into sums of powers. More generally, it is well suited to the study of diophantine
equations in “many variables”.
Concerning our counting problem, it indeed allows to establish the conjecture of Batyrev–
Manin–Peyre for smooth complete intersections of codimension m in Pn , that is, subschemes X
of Pn defined by the vanishing of m homogeneous polynomials f1 , . . . , fm such that at every
point x ∈ X, the Jacobian matrix of the fi has rank m. Let d1 , . . . , dm denote the degrees of
the polynomials f1 , ∑. . . , fm ; then the canonical line bundle of X is the restriction to X of the line
bundle O(−n − 1 + m i=1 di ). Consequently, X is Fano if and only if d1 + · · · + dm 6 n. Therefore,
specific examples of Fano complete intersections are lines or conics in P2 , planes, quadrics or cubic
surfaces in P3 , etc.

36
 A. Chambert-Loir

The circle method is restricted to cases where the number of variables n is very large in compar-
ison to the degrees d1 , . . . , dm . For instance, [5] establishes the desired result for an hypersurface
of degree d in Pn provided n > 2d (d − 1). However, when it applies, it gives very strong results,
including a form of the Hasse principle and equidistribution, see [34].

4.3 Homogeneous spaces

Homogeneous spaces, namely quotients of an algebraic group by a subgroup, form a large and
interesting class of varieties. Here is a list of interesting examples:
2
• Algebraic groups themselves, like SLn viewed as the subset of Pn defined by the equa-
2
tion z n = det(A), where a point of Pn is a non-zero pair (z, A) ∈ F × Matn (F ) modulo
homotheties.

• The set VP of matrices with fixed characteristic polynomial P . Indeed, when the polyno-
mial P has distinct roots, any two such matrices are conjugate, hence VP is an homogeneous
space of GLn .

• Non singular quadrics; indeed, if Q is a non-degenerate quadratic form in n variables, the

theorem of Witt implies that the orthogonal group O(Q) acts transitively on the hypersurface
of Pn−1 defined by Q.

• Grassmann varieties: if G is the Grassmann variety of d-planes W in the n-dimensional affine

space An , it has a transitive action of the group GL(n), given by W 7→ gW ; the stabilizer
of the fixed d-plane Ad × {0} is isomorphic the subgroup of block-triangular matrices (see
below).

Various techniques allow to tackle the question of rational points of bounded height on such
varieties. These methods allow to derive asymptotic expansions for the counting function by
comparing it to the volume of a corresponding set in a real or adelic space, whose asymptotic
growth has to be established independently. They are also amenable to proving equidistribution
results, as well as establishing analyting properties of the height zeta function.

4.3.1 Point counting vs. balls

Three important papers appeared in the 90’s concerning the close problem of integral points of
bounded height and still have a great influence. Methods of ergodic theory apply usually when
one considers lattices of real or adelic semi-simple groups (that is, discrete subgroups of finite
covolume). Keywords are the equidistribution theorem of M. Ratner for unipotent flows (used
in the closely related context of integral points by [22]) or, as in [21, 26], mixing properties of
the dynamical system, implied by decay of matrix coefficients of unitary representations (theorem
of Howe–Moore and its extensions).

4.3.2 Eisenstein series

Flag varieties are classical generalizations of the projective space: they parametrize subvector
spaces W of a fixed vector space V (Grassmann variety), of, more generally, increasing families
W1 ⊂ · · · ⊂ Wm (“flags”) of subvector spaces. Let us consider for simplicity the case of a fixed
vector space V of dimension n and the Grassmann variety Grnd of subspaces of fixed dimension d.
The projective space P(V ) is recovered by considering the cases d = 1 (parametrizing lines) and
d = n − 1 (parametrizing hyperplanes). Let us fix a specific subspace W0 of dimension d in V ;

37
 A. Chambert-Loir

considering a basis of W0 and extending it to a basis of V , it is easy to observe that any other
subspace of dimension d is of the form gW0 , for some automorphism g ∈ GL(V ); moreover,
gW0 = g 0 W0 if and only if g −1 g 0 W0 = W0 , that is, if g −1 g 0 belongs to the stabilizer P of W0 for
the action of GL(V ). In the basis that we fixed, P is the set of invertible upper-triangular block
matrices having only zeroes in the lower rectangle [d + 1, . . . , n] × [1, . . . , d].
More generally, in the language of algebraic groups, generalized flag varieties are quotients of
reductive groups by parabolic subgroups. The understanding of their height zeta functions is a
consequence of the observation, due to J. Franke, that these functions were studied extensively
in the theory of automorphic forms, being nothing else than generalized Eisenstein series. Modulo
some (not so obvious) translation, the results of R. P. Langlands on these Eisenstein series readily
imply the conjectures of Batyrev–Manin and Peyre (including equidistribution) for the flag
varieties, see [25] and [34].

4.3.3 Spectral decomposition of the height zeta function

Beyond Eisenstein’s series and Langlands’s results, there are many interesting cases where the
spectral decomposition of the height zeta function can be computed, leading to a proof of Manin’s
conjecture. In fact, this has been done for equivariant compactifications of many families of
algebraic groups.
To give an idea of the principle of the proof, let us say a word on the analogous problem of
counting lattice points in large balls. Let B be a bounded measurable subset in Rn and let
n
∑B be its characteristic function. To understand the cardinality N (B) of B ∩ Z , i.e., the sum
χ
x∈Zn χB (x), one can try to use the Poisson summation formula which would relate this sum
to a sum over Zn of the Fourier transform χ bB . For this, one needs to approximate χB by a
function in a class where the Poisson formula applies, say either the Schwartz class, at least a
continuous integrable function whose Fourier transform is integrable. The value at 0 of the Fourier
transform χ bB is the volume V (B) of B. If one can bound the remaining terms, one can hope to
give an upper bound for the difference N (B) − V (B). This is at least a way to attack the circle
problem, which leads to non-trivial upper bounds.
In the next Section, we shall give more details on this spectral theoretic approach to Manin’s
conjecture in the case of compactifications of algebraic groups. As we have indicated, it relies on
the generalization of the theory of Fourier series and Fourier integrals and of the Poisson formula,
which concern the groups R and R/Z, to general locally compact topological groups. We therefore
call it the Fourier–Poisson method.

4.4 Some details on the Fourier–Poisson method

4.4.1 Algebraic groups and their compactifications
Let X be an equivariant compactification of an algebraic group G over a number field F . This
means that X is a projective variety containing G as a dense open subset and that the left and
right actions of G on itself extend to actions on X.
Then, G(F ) is a discrete subgroup of the adelic group G(AF ). Moreover, for line bundles L
on X, the height HL can often be extended to a function on G(AF ): there are functions HL,v such
that, for g ∈ G(F ) ⊂ G(AF ), ∏
HL (g) = HL,v (gv ).
v∈Val(F )

If we use the right hand side of this formula to extend HL to a function on G(AF ), we obtain
that the height zeta function ZG (L; s) is an average over the discrete group G(F ) of a function

38
 A. Chambert-Loir

HL−s on G(AF ) and one can use harmonic analysis and the spectral decomposition to under-
stand ZG (L; s).
Important cases of groups in which mathematicians have been able to use harmonic analysis to
prove the conjectures of Batyrev–Manin–Peyre are the following.

• algebraic tori : then X is called a toric variety, see [3];

• vector groups, [12];

• the Heisenberg group of upper triangular matrices, [41];

• reductive groups, embedded in the wonderful compactification defined by De Concini–

Procesi, see [40].

4.4.2 The Poisson formula

Commutative groups are of course simpler to study, because irreducible representations are one-
dimensional (i.e., are characters). Let us fix some notation. Let µ be a Haar measure on the
locally compact group G(AF ), given as product of suitable Haar measures µv on G(Fv ). Let
cv = Hom(G(Fv ), U) be the group of characters of G(Fv ) and G(A
G \ F ) = Hom(G(AF ), U) be
the group of characters of G(AF ), where U is the group of complex numbers of modulus 1. By
\
restriction, any character χ ∈ G(A F ) induces a character χv of G(Fv ), for each place v. Let G(F )
⊥

be the orthogonal of G(F ), that is the group of characters of G(AF ) which are trivial on G(F ).
This is a locally compact group and it carries a dual measure µ⊥ .
Any integrable function f ∈ L1 (G(AF )) has a Fourier transform, which is the function on G(A \ F)
\
defined as follows: for χ ∈ G(A F ),
∫
F (f ; χ) = f (g)χ(g) dµ(g).
G(AF )
∏ ∏
If f is a simple function, i.e., of the form f (g) = fv (gv ), then F (f ; χ) is a product Fv (fv ; χv )
of local Fourier transforms defined in an analogous manner, namely
∫
Fv (fv ; χv ) = fv (g)χv (g) dµv (g).
G(Fv )

The generalization of Poisson formula to this context states that if f is an integrable function
on G(AF ) such that moreover F (f ; ·) is integrable on G(F )⊥ , then
∑ ∫
f (g) = F (f ; χ) dµ⊥ (χ).
g∈G(F ) G(F )⊥

This gives a formula for the height zeta function

∫
ZG (L; s) = F (HL−s ; χ) dµ⊥ (χ),
G(F )⊥

which in many cases proved to be a key tool towards establishing the desired analytic properties
of the height zeta function.

39
 A. Chambert-Loir

4.4.3 The Fourier transform at the trivial character

Let us consider the Fourier transform of f = HL−s at the trivial character, in other words, the
function ∫
Φ(s) = F (HL−s , 1) = HL (g)−s dµ(g).
G(AF )

The decomposition of the function HL as a product∏of function on the spaces G(Fv ) implies a
similar decomposition of Φ as a the product Φ(s) = v Φv (s), where
∫
Φv (s) = Fv (HL−s , 1) = HL (g)−s dµv (g).
G(Fv )

In the above mentioned cases, these functions have been often computed explicitly, resorting to
properties of algebraic groups. However, this can also be done in a very geometric manner thanks
to the fundamental observation: the integral Φv (s) on G(Fv ) can be viewed as a geometric analogue
−1
on X(Fv ) of Igusa’s local zeta functions. Assume for simplicity that L = ωX . In Section 5 we
will explain how to prove the following facts (see § 5.2.6):

• Φv (s) converges for Re(s) > 0;

• Φv (s) has a meromorphic continuation to C;

• for almost all places v, Φv (s) can be explicitly computed (Denef’s formula);
∏
• the product Φv (s) converges for Re(s) > 1;

• this product has a meromorphic continuation which is governed by some Artin L-function.

We shall in fact explain that these facts apply in much more generality than that of algebraic
groups.

4.4.4 Conclusion of the proof

To prove Manin’s conjecture for an equivariant compactification of an algebraic group using the
Fourier method, a lot of work still remains to be done. One first has to establish a similar
meromorphic continuation for other characters (other representations in the non-commutative
cases) and, then, to integrate all these contributions over G(F )⊥ . Of course, the last step requires
that one obtains good upper bounds in the first step, so that integration is at all possible. Details
depend however very much on the specific groups involved and cannot be described here.

5 Heights and Igusa zeta functions

This final section aims at explaining the geometric computation of the Fourier transform of the
height function via the theory of Igusa zeta functions. It essentially borrows from the introduction
of our recent article [13].

5.1 Heights and measures on adelic spaces

5.1.1 Heights
Let X be a projective variety over a number field F . Let L be an effective divisor on X, and let
us endow the associated line bundle OX (L) with an adelic metric. This line bundle possesses a

40
 A. Chambert-Loir

canonical global section fL whose divisor is L. For any place v ∈ Val(F ), the function x 7→ kfL (x)kv
is continuous on X(Fv ) and vanishes precisely on L. Moreover, its sup-norm is equal to 1 for almost
all places v. ∏
Consequently, the infinite product v∈Val(F ) kfL (xv )kv converges to an element of R+ for any
adelic point x = (xv ) ∈ X(AF ). Let U = X \ L. If x is an adelic point of U , then kfL (xv )kv 6= 0 for
all v, and is in fact equal to 1 for almost all v. We thus can define the height of a point x ∈ U (AF )
to be equal to ∏
HL ((xv )) = kfL (xv )k−1
v .
v∈Val(F )

The resulting function on U (AF ) is continuous and admits a positive lower bound. Moreover, for
any B > 0, the set of points x ∈ U (AF ) such that HL (x) 6 B is compact.

5.1.2 Local measures (II)

The measure τX,v on X(Fv ) defined in § 3.3.4 gives finite volume to X(Fv ) and the open set U (Fv )
has full measure. We modify the construction as follows so as to define a Radon measure on U (Fv )
whose total mass (unless L(Fv ) = ∅) is now infinite:
1
dτ(X,L),v (x) = dτX,v (x).
kfL (x)kv

For example, in the important case where X is an equivariant compactification of an algebraic

group G, U = G and −L is a canonical divisor, this measure is a Haar measure on the locally
compact group G(Fv ).

5.1.3 Definition of a global Tamagawa measure

The product of these local measures does not converge in general, and to use them to construct a
measure on the adelic space U (AF ), it is necessary to introduce convergence factors, i.e., a family
(λv ) of positive real numbers such that the product
∏
λv τ(X,L),v (U (ov ))
v finite

converges absolutely. The limit will thus be a positive real number (unless some factor U (ov ) is
empty, which does not happen if we remove an adequate finite set of places in the product). The
existence of such factors is a mere triviality, since one could take for λv the inverse of τ(X,L),v (U (ov )).
Of course, we claim for a meaningful definition of a family (λv ), based on geometric or arithmetic
invariants of U .
The condition under which we can produce such factors is the following: X is proper, smooth,
geometrically connected, and the two cohomology groups H1 (X, OX ) and H2 (X, OX ) vanish. Let
us assume that this holds and let us define
∗
M 0 = H0 (UF , O ∗ )/F and M 1 = H1 (UF , O ∗ )/torsion;

in words, M 0 is the Abelian group of invertible functions on UF modulo constants, and M 1 is the
Picard group of UF modulo torsion. By the very definition as a cohomology group of UF , they
possess a canonical action of the Galois group ΓF of F /F . Moreover, M 0 is a free Z-module of finite
rank. Indeed, to an invertible function on UF , one may attach its divisor, which is supported by
the complementary subset, that is LF . This gives a morphism from M 0 to the free Abelian group

41
 A. Chambert-Loir

generated by the irreducible components of LF and this map is injective because X is normal. It
follows that M 0 is a free Z-module of finite rank. Moreover, under the vanishing assertion above,
then M 1 is a free Z-module of finite rank too. We thus can consider the Artin L-functions of these
two ΓF -modules, L(s, M 0 ) and L(s, M 1 ).
Using Weil’s conjectures, proved by [16], to estimate the volumes U (ov ), in a similar manner to
what Peyre had done to define the global measure τX on X(AF ), we prove that the family (λv )
given by
Lv (1, M 0 )
λv =
Lv (1, M 1 )
if v is a finite place of F and λv = 1 if v is archimedean is a family of convergence factors. In other
words, the infinite product of measures

L∗ (1, M 1 ) ∏ Lv (1, M 0 )
dτ (xv )
L∗ (1, M 0 ) Lv (1, M 1 ) (X,L),v
v∈Val(F )

defines a Radon measure on the locally compact space U (AF ). We denote this measure by τ(X,L)
and call it the Tamagawa measure on U (AF ).

5.1.4 Examples for the Tamagawa measure

This construction coincides with Peyre’s construction if L = ∅, since then M 0 = 0 (there are no
non-constant invertible functions on X) and M 1 = Pic(XF )/torsion.
It also coincides with the classical constructions for algebraic groups. Let us indeed assume that
X is an equivariant compactification of an algebraic group G and that −L is a canonical divisor,
with G = U = X \ L. We have mentionned that τ(X,L),v is a Haar measure on G(Fv ). If G is
semi-simple, or if G is nilpotent, then M 0 = M 1 = 0; in that case λv = 1, according to the fact
that the global measure does not need any convergence factor. However, if G is a torus, then
M 0 is the group of characters of GF and M 1 = 0; our definition coincides with Ono’s. In all of
these cases, the global measure we construct is of course a Haar measure on the locally compact
group G(AF ).
It also recovers some cases of homogeneous spaces G/H, where G and H are semisimple groups
over Q, under the assumption that H has no nontrivial characters, like those studied by [7]. In fact,
the definition of Tamagawa measures on such homogeneous spaces does not need any convergence
factors, as can be seen by the fact that M 0 = 0 (since invertible functions on G are already
constants) and M 1 = 0 (because the Picard group of G/H is given by characters of H).
As a last example, let us recall that [35] had shown that the universal torsors over toric vari-
eties do not need any convergence factors, because the convergence factor of the fibers (a torus)
compensates the one of the base (a projective toric variety). In our approach, this is accounted by
the fact, discovered by [14], that such torsors have neither non-constant invertible functions, nor
non-trivial line bundles, hence M 0 = M 1 = 0 and λv = 1.

5.2 An adelic-geometric analogue of Igusa’s local zeta functions

As we explained above, when using the Fourier method to elucidate the number of points of
bounded height on some algebraic varieties, we are lead to establish the meromorphic continuation
of the function of a complex variable:
∫
Z : s 7→ HL (x)−s dτ(X,L) (x).
U (AF )

42
 A. Chambert-Loir

(This function was denoted Φ in §4.4.) By definition, the measure τ(X,L) is (up to the convergence
factors and the global factor coming from Artin L-functions) the product of the local measures
τ(X,L),v on the analytic manifolds U (Fv ). Similarly, the integrated function x 7→ HL (x)−s is the
product of the local functions x 7→ kfL (x)ksv . Consequently, provided it converges absolutely, this
adelic integral decomposes as a product of integrals over local fields,
∫
L∗ (1, M 1 ) ∏ Lv (1, M 0 )
Z(s) = ∗ Z v (s), Z v (s) = kfL (x)ksv dτ(X,L),v (x).
L (1, M 0 ) Lv (1, M 1 ) U (Fv )
v∈Val(F )

5.2.1 Geometric Igusa zeta functions

Igusa’s theory of local zeta functions studies the analytic properties of integrals of the form
∫
IΦ (s) = Φ(x) |f (x)|s−1
v dx,
Fvn

where f is a polynomial and Φ is a smooth function with compact support on the affine n-space
over Fv . (See [30] for a very good survey on this theory.)
Our integrals Zv (s) are straightforward generalizations of Igusa’s local zeta functions. Indeed,
since τ(X,L),v = kfL k−1
v τX,v , we have
∫ ∫
s−1
Zv (s) = kfL (x)kv dτX,v (x) = kfL (x)ks−1
v dτX,v (x)
U (Fv ) X(Fv )

because L(Fv ) has measure 0 in X(Fv ). In our case, we integrate on a compact analytic manifold,
rather than a test function on the affine space; as always in Differential geometry, partitions of
unity and local charts convert a global integral into a finite sum of local ones. The absolute value
of a polynomial has been replaced by the norm of a global section of a line bundle; again, in
local coordinates, the function kfL kv is of the form |f |v ϕ, where f is a local equation of L and
ϕ a continuous non-vanishing function. However, because of these slight differences, we call such
integrals geometric Igusa zeta functions.
From now on, we will assume in this text that over the algebraic closure F , L is a divisor with
simple normal crossings. This means that the irreducible components of the divisor LF are smooth
and meet transversally; in particular, any intersection of part of these irreducible components is
either empty, or a smooth subvariety of the expected codimension. This assumption is only here for
the convenience of the computation. Indeed, by the theorem on resolution of singularities of [29],
there exists a proper birational morphism π : Y → X, with Y smooth, which is an isomorphism
on U and such that the inverse image of L (as a Cartier divisor) satisfies this assumption. We
may replace (X, L) by (Y, π ∗ L) without altering the definitions of Z, Zv , etc.. Let A be the set
of irreducible components of L; for α∑∈ A , let Lα be the corresponding component, and dα its
multiplicity in L. We thus have L = α∈A dα Lα .
This induces a stratification of X indexed by subsets A ⊂ A , the stratum XA being given by

XA = {x ∈ X ; x ∈ Lα ⇔ α ∈ A}.

In the sequel, we shall moreover assume that for any α ∈ A , Lα is geometrically irreducible.
This does not cover all cases, but the general case can be treated using a similar analysis, the
Dedekind zeta function of F being replaced by Dedekind zeta functions of finite extensions Fα
defined by the components Lα ; I refer the interested reader to [13] for details.

43
 A. Chambert-Loir

5.2.2 Local analysis

Let x be a point in XA (Fv ). The “normal crossings” assumption on L implies that on a small
enough analytic neighbourhood Ωx of x, one may find local equations xα of Lα , for α ∈ A, and
complete these local equations into a system of local coordinates ((xα )α∈A ; y1 , . . . , ys ), so that
s + Card(A) = dim X. In this neighbourhood of x, the measure τ can be written ω(x; y) dx dy,
with ω a positive continuous function, and there is a continous and positive function ϕ such that
∏
kfL kv = ϕ · |xα |dvα
α∈A

on Ωx . If Φx is a continuous function with compact support on Ωx , we thus have

∫ ∫ ∏ ∏
Φx kfL ks−1 τ(X,L),v = (ϕs−1 ωΦx ) |xα |(s−1)d
v
α
dxα dy1 . . . dys .
X(Fv ) Ωx α∈A α∈A

By comparison with the integral ∫

Φ(u) |u|sv du
Fv

which converges for Re(s) > −1 if Φ is compactly supported on Fv , we conclude that

∫
Φx kfL ks−1 τ(X,L),v
X(Fv )

converges absolutely for Re(s) > 0.

Introducing a partition of unity, we conclude that for any place v ∈ Val(F ), the integral Zv (s)
converges for Re(s) > 0 and defines a holomorphic function on this half-plane.

5.2.3 Denef ’s formula

Our understanding of the infinite product of the functions Zv (s) relies in the explicit computation
of almost all of them. The formula below is a straightforward generalization of a formula which
[17] used to prove that, when f is a polynomial with coefficients in the number field F and Φ is
the characteristic function of the unit polydisk in Fvn , then Igusa’s local zeta function IΦ (s) is a
rational function of qv−s whose degree is bounded when v varies within the set of finite places of F .
Assuming that the whole situation has “good reduction” at a place v, we obtain the following
formula, where kv is the residue field of F at v and qv is its cardinality:
∑ ∏ qv − 1
Zv (s) = qv− dim X Card(XA (kv )). 1+dα (s−1)
.
A⊂A α∈A qv −1

As in [17], this is a consequence of the fact that for each point x̃ ∈ X(kv ), in the local analytic
description of the previous paragraph, we may find local coordinates xα and y such that ϕ ≡ ω = 1
which parametrize the open unit polydisk with “center x̃.” That this is at all possible is more or
less what is meant by “having good reduction”.

5.2.4 Meromorphic continuation

Let us analyse the various terms of this formula. The stratum XA has codimension Card(A) in X
(or is empty); this implies easily that

qv− dim X Card(XA (kv )) = O(qv− Card(A) ).

44
 A. Chambert-Loir

More precisely, since Lα is geometrically irreducible and Xα is an open subset of it, it follows from
[16] that
qv− dim X Card(Xα (kv )) = qv−1 + O(qv−3/2 ).
(In fact, the estimates of [31] suffice for that.) Similarly, X∅ = X \ L = U , and
qv− dim X Card(X∅ (kv )) = 1 + O(qv−1/2 ).
In fact, the vanishing assumption on the cohomology groups H1 (X, OX ) and H2 (X, OX ) implies
that the remainder is even O(qv−1 ). By an integration formula of [43], this expression is precisely
equal to τ(X,L),v (U (ov )).
We therefore obtain that for any ε > 0, there exists δ > 0 such that if Re(s − 1) > −δ, then
∏ ( )
Zv (s) = τ(X,L),v (U (ov )) (1 − qv−1−dα (s−1) )−1 1 + O(qv−1−ε ) .
α∈A

If we multiply this estimate by the convergence factors λv that we have introduced, and which
satisfy λv = 1 + O(qv−1 ), we obtain that the infinite product
∏ ∏
λv Zv (s) ζF,v (1 + dα (s − 1))−1
v finite α∈A

converges absolutely and uniformly for Re(s − 1) > −δ/2; it defines a holomorphic function Φ(s)
on that half-plane. We have denoted by ζF,v the local factor at v of the Dedekind zeta function of
the number field F . Then, the equality
L∗ (1, M 1 ) ∏
Z(s) = λv Zv (s)
L∗ (1, M 0 )
v∈Val(F )
( )
L∗ (1, M 1 ) ∏ ∏
= ∗ ζF (1 + dα (s − 1)) Φ(s) Zv (s)
L (1, M 0 )
α∈A v archimedean

shows that Z converges absolutely, and defines a holomorphic function on {Re(s) > 1}. Since the
Dedekind zeta function ζF has a pole of order 1 at s = 1, Z(s) admits a meromorphic continuation
on the half-plane {Re(s) > 1 − δ/2}, whose only pole is at s = 1, with multiplicity Card(A ).

5.2.5 The leading term

Moreover,
( )
L∗ (1, M 1 ) ∏ ∏
lim (s − 1) Card(A )
Z(s) = ∗ d−1
α ζF∗ (1)Card(A ) Φ(1) Zv (1).
s→1 L (1, M 0 )
α∈A v archimedean

By definition of Φ, one has

∏ ∏
Φ(1) = λv ζF,v (1)−1 Zv (1)
v finite α∈A
( )
∏ ∏
−1
= λv ζF,v (1) τX,v (X(Fv )).
v finite α∈A

Comparing the Galois modules M 0 and M 1 for X and U , one can conclude that
∏
lim (s − 1)Card(A ) Z(s) = τX (X(AF )) d−1
α . (5.1)
s→1
α∈A

45
 A. Chambert-Loir

5.2.6 Conclusion of the computation

Finally, we have shown the following: The integral
∫
s 7→ HL (x)−s dτ(X,L) (x)
U (AF )

converges for Re(s) > 1, defines a holomorphic function on this half-plane. It has a meromorphic
continuation on some half-plane {Re(s) > 1 − δ} (for some δ > 0) whose only pole is at s = 1, has
order Card(A ) and its asymptotic behaviour at s = 1 is given by Equation (5.1).

5.3 Application to volume estimates

Let us finally derive our volume estimates from the analytic property of the function Z(s). We are
interested in the volume function defined by

V (B) = τ(X,L) ({x ∈ U (AF ) ; HL (x) 6 B})

for B > 0, more precisely in its asymptotic behaviour when B → ∞.

Its Mellin-Stieltjes transform is given by
∫ ∞ ∫
−s
B dV (B) = HL (x)−s dτ(X,L) (x) = Z(s).
0 U (AF )

By a slight generalization of Ikehara’s Tauberian theorem, we conclude that V (B) satisfies the
following asymptotic expansion:
( ∏ )−1
V (B) ∼ (a − 1)! dα τX (X(AF ))B(log B)a−1 ,
α∈A

where a = Card(A ) is the number of irreducible components of the divisor L.

Our argument, applied to different metrizations, also implies that when B → ∞, the probability
measure
1
1 dτ (x)
V (B) HL (x)6B (X,L)
on U (AF ) (viewed as a subset of X(AF )) converges to the measure
1
dτX (x)
τX (X(AF ))

on X(AF ).

Acknowledgments
I would like to thank Yuri Tschinkel for his comments on a first version of this survey. Our
collaboration on the topics presented here began more than ten years ago. Let it be the occasion
to thank him heartily for having shared his views and projects with me.
I also thank the referee, as well as R. de la Bretéche, H. Oh and W. Veys for their suggestions.

46
 A. Chambert-Loir

References
[1] V. V. Batyrev & Yu. I. Manin (1990), Sur le nombre de points rationnels de hauteur bornée
des variétés algébriques. Math. Ann., 286, 27–43.

[2] V. V. Batyrev & Yu. Tschinkel (1995), Rational points on bounded height on compactifi-
cations of anisotropic tori. Internat. Math. Res. Notices, 12, 591–635.

[3] V. V. Batyrev & Yu. Tschinkel (1996), Height zeta functions of toric varieties. Journal
Math. Sciences, 82 (1), 3220–3239.

[4] V. V. Batyrev & Yu. Tschinkel (1996), Rational points on some Fano cubic bundles. C.
R. Acad. Sci. Paris Sér. I Math., 323, 41–46.

[5] B. J. Birch (1962), Forms in many variables. Proc. London Math. Soc., 265A, 245–263.

[6] E. Bombieri & J. PILA (1989), The number of integral points on arcs and ovals. Duke Math.
J., 59 (2), 337–357.

[7] M. Borovoi & Z. Rudnick (1995), Hardy-Littlewood varieties and semisimple groups. Invent.
Math., 119 (1), 37–66.

[8] R. de la Bretèche (2002), Nombre de points de hauteur bornée sur les surfaces de del Pezzo
de degré 5. Duke Math. J., 113 (3), 421–464.

[9] R. De la Bretèche & P. Swinnerton-Dyer (2007), Fonction zêta des hauteurs associée
à une certaine surface cubique. Bull. Soc. Math. France, 135 (1), 65–92.

[10] T. D. Browning & D. R. Heath-Brown (2006), The density of rational points on non-
singular hypersurfaces. II. Proc. London Math. Soc. (3), 93 (2), 273–303. With an appendix
by J. M. Starr.

[11] T. D. Browning, D. R. Heath-Brown & P. Salberger (2006), Counting rational points

on algebraic varieties. Duke Math. J., 132 (3), 545–578.

[12] A. Chambert-Loir & Yu. Tschinkel (2002), On the distribution of points of bounded height
on equivariant compactifications of vector groups. Invent. Math., 148, 421–452.

[13] A. Chambert-Loir & Yu. Tschinkel (2008), Tamagawa measures and volume estimates in
analytic and adelic geometry. Work in progress.

[14] J.-L. Colliot-Thélène & J.-J. Sansuc (1987), La descente sur les variétés rationnelles. II.
Duke Math. J., 54 (2), 375–492.

[15] D. A. Cox (1995), The homogeneous coordinate ring of a toric variety. J. Algebraic Geometry,
4 (1), 17–50.

[16] P. Deligne (1974), La conjecture de Weil, I. Publ. Math. Inst. Hautes Études Sci., 43, 273–
307.

[17] J. Denef (1987), On the degree of Igusa’s local zeta function. Amer. J. Math., 109, 991–1008.

[18] U. Derenthal (2006), On the Cox ring of Del Pezzo surfaces. ArXiv:math/060311
[math.AG].

47
 A. Chambert-Loir

[19] U. Derenthal (2008), Counting integral points on universal torsors. ArXiv:0810.4122

[math.NT].

[20] U. Derenthal & Y Tschinkel Universal torsors over del Pezzo surfaces and rational points.
In Equidistribution in number theory, an introduction, volume 237 de NATO Sci. Ser. IIMath.
Phys. Chem., p. 169-196, Springer, Dordrecht.

[21] A. Eskin & C. McMullen (1993), Mixing, counting and equidistribution on Lie groups. Duke
Math. J., 71 (1), 181–209.

[22] A. Eskin, S. Mozes & N. Shah (1996), Unipotent flows and counting lattice points on ho-
mogeneous varieties. Ann. of Math., 143 (2), 253–299.

[23] D. Essouabri (2005), Analytic continuation of a class of zeta functions of heights. (Pro-
longements analytiques d’une classe de fonctions zêta des hauteurs et applications.). Bull.
Soc. Math. France, 133 (2), 297–329.

[24] G. Faltings (1983), Endlichkeitsätze für abelsche Varietäten über Zahlkörpern. Invent. Math.,
73 (3), 349–366.

[25] J. Franke, Yu. I. Manin & Yu. Tschinkel (1989), Rational points of bounded height on
Fano varieties. Invent. Math., 95 (2), 421–435.

[26] A. Gorodnik, F. Maucourant & H. Oh (2009), Manin’s and Peyre’s conjectures on rational
points and adelic mixing. Ann. Sci. École Norm. Sup. To appear.

[27] G. H. Hardy & E. M. Wright (1979), An introduction to the theory of numbers, The
Clarendon Press Oxford University Press, New York, fifth edition.

[28] D. R. Heayh-Brown (2002),The density of rational points on curves and surfaces, Ann. of
Math. (2), 155 (2), 553-595.

[29] H. Hironaka (1964), Resolution of singularities of an algebraic variety over a field of char-
acteristic zero. I, II. Ann. of Math. (2), 79, 109–203, 205–326.

[30] J.-I. Igusa (2000), An introduction to the theory of local zeta functions, AMS/IP Studies in
Advanced Mathematics 14, American Mathematical Society, Providence, RI.

[31] S. Lang & A. Weil (1954), Number of points of varieties in finite fields. Amer. J. Math., 76,
819–827.

[32] J. V. Matijasevič (1971), Diophantine representation of enumerable predicates. Izv. Akad.

Nauk SSSR Ser. Mat., 35, 3–30.

[33] D. G. Northcott (1950), Periodic points on an algebraic variety. Ann. of Math., 51, 167–
177.

[34] E. Peyre (1995), Hauteurs et mesures de Tamagawa sur les variétés de Fano. Duke Math.
J., 79, 101–218.

[35] P. Salberger (1998), Tamagawa measures on universal torsors and points of bounded
height on Fano varieties. Nombre et répartition des points de hauteur bornée, E. Peyre,
Astérisque 251, 91–258.

48
 A. Chambert-Loir

[36] P. Salberger (2007), On the density of rational and integral points on algebraic varieties. J.
Reine Angew.Math., 606, 123–147.

[37] S. Schanuel (1979), Heights in number fields. Bull. Soc. Math. France, 107, 433–449.

[38] W. M. Schmidt (1993), Northcott’s theorem on heights. I. A general estimate. Monatsh.

Math., 115 (1-2), 169–181.

[39] J.-P. Serre (1997), Lectures on the Mordell-Weil theorem, Aspects of Mathematics, Friedr.
Vieweg & Sohn, Braunschweig, third . Translated from the French and edited by Martin
Brown from notes by Michel Waldschmidt, With a foreword by Brown and Serre.

[40] J. Shalika, R. Takloo-Bighash & Y. Tschinkel (2007), Rational points on compactifica-

tions of semi-simple groups. J. Amer. Math. Soc., 20 (4), 1135–1186 (electronic).

[41] J. A. Shalika & Y. Tschinkel (2004), Height zeta functions of equivariant compactifications
of the Heisenberg group. Contributions to automorphic forms, geometry, and number theory,
743–771, Johns Hopkins Univ. Press, Baltimore, MD.

[42] J. Thunder (1992), An asymptotic estimate for heights of algebraic subspaces. Trans. Amer.
Math. Soc., 331, 395–424.

[43] A. Weil (1982), Adeles and algebraic groups, Progr. Math. 23, Birkhäuser.

IRMAR, Université de Rennes 1

Campus de Beaulieu
35042 Rennes Cedex
France
email:[email protected]

49