Sie sind auf Seite 1von 13

NUMBER-THEORY AND ALGEBRAIC GEOMETRY

ANDRE W E I L

Mr. Chairman, Ladies and Gentlemen,


The previous speaker concluded his address with a reference to Dedekind
and Weber. It is therefore fitting that I should begin with a homage to Kronecker. There appears to have been a certain feeling of rivalry, both scientific
and personal, between Dedekind and Kronecker during their life-time; this
developed into a feud between their followers, which was carried on until the
partisans of Dedekind, fighting under the banner of the "purity of algebra",
seemed to have won the field, and to have exterminated or converted their foes.
Thus many of Kronecker's far-reaching ideas and fruitful results now lie buried
in the impressive but seldom opened volumes of his Complete Works. While
each line of Dedekind's X l t h Supplement, in its three successive and increasingly "pure" versions, has been scanned and analyzed, axiomatized and generalized, Kronecker's once famous Grundzge are either forgotten, or are thought
of merely as presenting an inferior (and less pure) method for achieving part
of the same results, viz., the foundation of ideal-theory and of the theory of
algebraic number-fields. In more recent years, it is true, the fashion has veered
to a more multiplicative and less additive approach than Dedekind's, to an
emphasis on valuations rather than ideals; but, while this trend has taken us
back to Kronecker's most faithful disciple, Hensel, it has stopped short of the
master himself.
Now it is time for us to realize that, in his Grundzge, Kronecker did not
merely intend to give his own treatment of the basic problems of ideal-theory
which form the main subject of Dedekind's life-work. His aim was a higher one.
He was, in fact, attempting to describe and to initiate a new branch of mathematics, which would contain both number-theory and algebraic geometry as
special cases. This grandiose conception has been allowed to fade out of our sight,
partly because of the intrinsic difficulties of carrying it out, partly owing to
historical accidents and to the temporary successes of the partisans of purity
and of Dedekind. It will be the main purpose of this lecture to try to rescue it
from oblivion, to revive it, and to describe the few modern results which may
be considered as belonging to the Kroneckerian program.
Let us start from the concept of a point on a variety, or, what amounts to
much the same thing, of a specialization. Take for instance a plane curve G,
defined by an irreducible equation F(X, Y) = 0, with coefficients in a field k.
A point of C is a solution (x, y) of F(X, Y) = 0, consisting of elements x] y
of some field k' containing k. In order to define the function-field on the curve,
we identify two polynomials in X, Y if they differ only by a multiple of F, i.e.,
we build the ring k[X, Y]/(F), and we take the field of fractions $ of that ring:
in particular, X and Y themselves determine the elements X = X mod F,
90

NUMBER-THEORY AND ALGEBRAIC GEOMETRY

91

Y = Y mod F, of $, and (X, Y) is a point of C, called generic since it does not


satisfy any relation over k except F(X, Y) = 0 and its consequences. Then any
point (x, y) of C, with coordinates in an extension kf of k, determines a homomorphism cr of the ring k[X, Y]/(F) into/c', d efined by putting <r(X) = x, <r(Y) = y,
and cr(a) = a for every a k; this homomorphism is also called a specialization
of that ring, and a generic one if it is an isomorphism of it into kf ; consequently,
(x, y) will be called a specialization of (X, Y), and will be called generic if cr
is generic.
Our homomorphism a has been so defined as to preserve the elements of the
"ground-field" k; but this restriction, usual as it is in algebraic geometry, may
well prove too narrow for some purposes. If, for example, we consider a curve
F(X, Y, t) = 0, depending upon a parameter t, where F is a polynomial in
X, Y, i with coefficients in a field k, then the coefficients of the equation of the
curve are in the field k(t). However, with our curve, we naturally associate the
surface F(X, Y, T) = 0; the curve then appears as a plane section of that surface by the plane T = t. Because of this changed point of view, the parameter
t, previously frozen by its inclusion in the field of "constants", is now liberated
and available for specialization; and so we are free now to consider as a specialization of our ring k[X, Y, t]/(F(X, Y, t)) any homomorphism of that ring
into an extension k' of k, still preserving the elements of k, but mapping X = X
mod F,Y = Y mod F, t = t mod F onto any three elements xr, yf, t' of k' satisfying F(xf, y', tf) = 0. Thus no longer restricted to the exclusive consideration
of the "generic" curve belonging to the family F(X, Y, t) = 0, we are enabled
to consider any specialization F(X, Y, t') = 0 of that curve, and the whole
surface F(X, Y, t) = 0 spanned by that family.
This shifting of our point of view necessitates a re-examination of the concept
of ground-field and of the field of definition of a variety. The previous speaker
has mentioned, as one of the main achievements of modern algebraic geometry,
the possibility of operating over quite arbitrary ground-fields. One should not
be blind, however, to the somewhat illusory nature of this achievement. As our
knowledge of algebraic curves is fairly extensive, there is, it is true, a great deal
that we can say on the curve F(X, Y, t) = 0 depending upon the parameter t,
in the example discussed above; and we should not possess that knowledge if
our methods of proof were not valid over the ground-field k(t). But as we have
pointed out, all we can say on the curve F(X, Y, t) = 0 is but part of the theory of the surface F(X, Y, T) = 0. This may be at the present moment, and it
is in fact, one of the best ways of acquiring some knowledge of the geometry on
that surface; but the fact remains that, in the final analysis, any statement on
a variety with a larger ground-field boils down to a statement on a variety (of
higher dimension, and therefore intrinsically more difficult to study) over a
smaller ground-field.
Now consider, with Kronecker, that, in most problems of algebraic geometry,
only a finite number of points and varieties occur at a time ; these will necessarily
have a common field of definition which is finitely generated over the prime

92

ANDR WEIL

field, i.e., which is generated over the prime field (the field Q of rational numbers if the characteristic is 0, and otherwise a finite field) by a finite number of
quantities (h, , tN) ; if these are considered as parameters, and are made
available for specialization, then, in the final analysis, every statement we can
make can be thought of as a theorem in algebraic geometry over an absolutely algebraic ground-field, i.e., either over a finite field or over n algebraic number-field
of finite degree. While this realization, of course, cannot in any way detract
from the methodological importance of arbitrary ground-fields as one of the
chief tools of modern algebraic geometers, it gives us some insight into the
deep meaning of Kronecker's view, according to which the absolutely algebraic
fields are the natural ground-fields of algebraic geometry, at any rate as long
as purely algebraic methods (as distinct from analytical or topological methods)
are being used. Now these are fields with strongly marked individual features,
which will undoubtedly have to be taken more and more into account as algebraic geometry develops along more Kroneckerian lines. For instance, the field
with q elements can be characterized by the fact that its elements are invariant
under the automorphism x > xq of any field containing it; this must have a
profound influence on the geometry over that field; and recent work connected
with the Riemann hypothesis ([lie]) fully confirms that expectation. Another
fact, so far an isolated one, in the same direction, is the existence of matrices,
associated with curves over a finite field, which bear a curious resemblance
with the period-matrices of abelian integrals in the classical theory (cf. [lid]).
We are now in a position to discussspecializations again from our broadened
point of view. If e.g. F(X, Y) = 0 is the equation of a curve, with coefficients
in a subring R of a field k, then any homomorphism <r of the ring R[X, Y]/(F)
into a field kf will be called a specialization of that ring; if in particular it preserves (or at least if it maps isomorphically) the elements of R, then it can be
extended to a homomorphism of k[X, Y]/(F) which preserves the elements of k.
As Kronecker realized, this affects our concept of dimension. Take for instance, instead of our curve, a hypersurface F(X\, , Xn) = 0 in n-dimensional
space, with coefficients in a subring R of a field k; let 9t denote the ring
R [Xi, , Xn]/(F). Then the dimension n 1 of that variety can be defined
as the degree of transcendency, over the ground-field k, of the function-field on
the variety, i.e., of the field of fractions of 9$, or, equivalently, as the maximum
number of successive specializations <r, af, a", , of 9? onto a ring 9', of 9t'
onto a ring 9", etc., each one of which preserves the elements of R, and none
of which is an isomorphism; the rings 9', 9", are understood to be "integral
domains" (i.e., subrings of fields). If we remove the condition that the specializations <r must preserve the elements of R, but merely require that they should
preserve the elements of the "prime ring" (the ring Z of integers if the characteristic is 0, the ring of integers mod p if it is p > 1), this gives us the dimension
over the prime field, or absolute dimension. So far, we have not crossed the
boundaries of ordinary algebraic geometry, even though we may have pushed
down the ground-field to an absolutely algebraic field. In particular, if the char-

NUMBER-THEORY AND ALGEBRAIC GEOMETRY

93

acteristic is p > 1, every homomorphism must preserve the elements of the


prime field, and so there is no temptation, nor even any possibility, for us to
cross those boundaries. However, if the characteristic is 0, there are homomorphisms which do not preserve the characteristic; as soon as we allow these
to enter the picture, we are within a wider area, where algebraic geometry and
number-theory commingle and cannot be kept apart; and, as a consequence,
the proper concept of dimension is the Kroneckerian concept. Since our sequences of specializations <r, af, can now be increased by one which changes
the characteristic from 0 to some p > 1, it follows that the Kroneckerian dimension is higher by 1 than that of algebraic geometry proper. For instance, a curve
over an algebraic number-field has the Kroneckerian dimension 2,
In this sense, the only two cases of dimension 1 are those of a curve over a
finite field, and of an algebraic number-field. In fact, it has been well known,
ever since Kronecker and Dedekind, that there are far-reaching analogies between these two cases, and these have been among the chief sources of progress
in both directions; indeed, we have reached a stage where we can deal simultaneously with large segments of both theories, not merely the more elementary
ones, but also class-field theory and part of the theory of the zeta-function. It
is true that these analogies are still incomplete at some crucial points; new concepts are clearly needed before we can transport to number-fields, even conjecturally, the facts about the Jacobian variety of a curve which have recently
led to the proof of the Riemann hypothesis ([lie], [llf]). Nevertheless, our knowledge of these topics is fairly extensive, whereas the same can hardly be said of
the problems in higher dimensions.
It is true that the theory of local rings has been extensively developed, largely
by its initiator Krull (cf. e.g. [6]), and more recently by Chevalley ([2]), I.
Cohen ([4]), and others. Such rings arise as follows: <r being, as above, a specialization, say, of the ring 9 = R[X, Y]/(F) defined by a curve F(X, Y) = 0,
it can be extended to a homomorphic mapping of the ring 9' of those elements
u/v of the field of fractions $ of 9, for which u, v are in 9? and ov ^ 0, by putting c(u/v) = au/av', 3' is the specialization-ring, and the ideal of non-units in
9', which is the kernel of <r, is the specialization-ideal; dlf is called a local ring,
and its completion, with respect to the topology defined on it by the powers of
the specialization-ideal, is a complete local ring; experience shows that it is
desirable to confine oneself to integrally closed specialization-rings, and this
leads to Zariski's fundamental concept of normality. Up to now geometers have
used only characteristic-preserving specializations; therefore all their local rings
contain a field, and have the same characteristic as their residue-class ring.
Fortunately algebraists have not confined themselves to that case, so that their
work is immediately available for the more general geometry that we are envisaging here.
We are thus led to modify the Kroneckerian view that the "true" or "natural"
ground-fields in algebraic geometry are the absolutely algebraic fields; this is so
as long as ground-fields are considered from the purely algebraic point of view,

94

ANDR WEIL

without any additional structure. However, it is now clear that the study of a
family of varieties at, or rather in the neighborhood of, a given specialization of
the parameter leads at once to the consideration of algebraic varieties over
complete local rings and their fields of fractions; some recent work by Chow
([3]) may be considered as pertaining to this subject, of which the "geometry
on a variety in the neighborhood of a subvariety" (as exemplified chiefly by
Zariski's theory of holomorphic functions ([12]) forms a natural extension. That
this does not contradict the Kroneckerian outlook, but has its root in it, is
clearly shown by the fact that the theory of local rings was originated by Hensel;
his p-adic rings, in fact, are the complete local rings attached to the specializations of the rings of integers in algebraic number-fields. Hence the local study,
say, of an algebraic curve F(X, Y) = 0 with coefficients in the ring Z of rational
integers, "at" the specialization of Z determined by a prime p, amounts to enlarging the ground-field to the p-adic field. Thus the p-adic fields appear as
another kind of "natural" ground-fieTd, and one may expect that the geometry
over such fields will acquire more and more importance as it learns to develop
its own methods. One may quote here E. Lutz's results on elliptic curves ([7]),
showing that the group of points on such a curve has a subgroup of finite index,
isomorphic to the additive group of integers in the ground-field; similar results
undoubtedly hold for Abelian varieties of any dimension. In his beautiful thesis,
Chabauty ([1]), following ideas of Skolem (LIO]), has shown how the method of
p-adic completion, with respect to a more or less arbitrary prime p, can yield
deep results about varieties over an algebraic number-field; there, as already in
Skolem's work, the problem concerns the intersection of an algebraic variety and
of a multiplicative group; by p-adic completion, the latter becomes an algebroid
variety defined by linear differential equations. Of course geometry over finite
fields may in a certain sense be obtained from the geometry over p-adic fields
by reduction modulo p, so that the latter may be said to contain all that the
former contains, and a good deal more; but little use has been made so far of
the relations between these two kinds of geometries, .and little is known about
them.
But the geometry over p-adic fields, and more generally over complete local
rings, can provide us only with local data; and the main tasks of algebraic
geometry have always been understood to be of a global nature. It is well known
that there can be no global theory of algebraic varieties unless one makes them
"complete", by adding to them suitable "points at infinity," embedding them,
for example, in projective spaces. In the theory of curves, for instance, one
would not otherwise obtain such basic facts as that the numbers of poles and
of zeros of a function are equal, or that the sum of residues of a differential is
0. One way of doing this (which, however, is effective only in the case of dimension 1) consists in considering the valuations of the field of functions on the
curve; on a given affine model F(X, Y) = 0, each simple point defines a valuation, viz., that one which assigns, to each function on the curve, the order of
the pole or zero it may have at that point; and all valuations, with a finite num-

NUMBER-THEORY AND ALGEBRAIC GEOMETRY

95

ber of exceptions, can be so obtained; the exceptions correspond to the "multiple


points" and to the "points at infinity", and give an invariant definition for
these. Correspondingly, if we apply this idea to an algebraic number-field (also
a one-dimensional problem), we obtain satisfactory formulations for global theorems, entirely analogous to the theorems on algebraic curves, provided we
allow for "archimedean" valuations with somewhat weaker properties than
those of algebraic geometry and than the p-adic valuations on number fields,
viz., those for which the completed field is the field of real or that of complex
numbers. Thus it appears that algebraic geometry over the complex numberfield is, after all, a legitimate object of study, no less necessary or useful than
geometry over p-adic fields; and so the door is opened to topology, functiontheory, differential geometry, and partial differential equations. This, at any
rate, is the logical way in which algebraic geometry over complex numbers
ought to have been bom, had mathematics consisted solely of number-theory
and algebra. That it came into being quite differently, and that it developed so
far ahead of other branches of geometry, is a historical accident; it is indeed a
fortunate one, having allowed free play to a tool which is invaluable as long as
one is aware of its limitations; I need hardly tell you that I am referring to our
spatial intuition.
We are now ready to consider in more specific terms the few known results
in the "geometry over integers", which, following Kronecker, I have been trying
to define ; and for this we must turn first of all, naturally, to Kronecker himself.
His great work on elliptic functions ([5b]), or rather its algebraic part (as distinct from the equally profound analytical theory), gives us a first example of
an investigation of that kind; this consists in the study of the equation Y2 = 1
pX2 + Xi over the ring Z\p], where p is an indeterminate, and is chiefly concerned
with the transformation of elliptic functions. Jacobi's results on this subject are
interpreted as defining, for every odd prime p, a correspondence between two
generic points (x, y, p), (x', yf, pf) of the surface Y2 = 1 pX2 + X 4 , where
%' = xnF(l/x)/F(x),
y' = G(x)/F(x)2, p' is algebraic over Q(p), and F, G are
polynomials with algebraic coefficients over Q(p). Let a- be a root of
1 - pX2 + X 4 = 0, and o-' a root of 1 - pfX2 + Xe = 0. Then Kronecker
proves the following facts. The coefficients of F, G are in the field Q(cr, <rf); if
divisibility relations are understood in the sense of integral algebraic elements
over Z[p], then </ and all the roots of G(X) are units; and F(X) is of the form
F(X) IT X*"1 + ir ytX*"'-1 + 1,
where w and the yi are integral over Z[p] ; furthermore, ir is of degree p + 1 over
Q(p), and has the norm p over that field. The main results on complex multiplication, and its application to the class-field theory of imaginary quadratic fields,
can be derived from these facts by specialization of the parameter p. It is very
probable that a reconsideration of this splendid work from a modern point of
view would not merely enrich our knowledge of elliptic function-fields, but

96

ANDR WEIL

would also reveal principles of great importance for any further development
of algebraic geometry over integers.
Now, coming back to the Grundzge,-take Kronecker's well-known and supposedly outmoded device for the introduction of ideals. This consists in associating with the elements OQ , a\, , a^ of a ririg the linear expression
Go + ]C?=i ufoi, or, when the homogeneous notation happens to be more suitable, the linear form X/=o ^ A > in the indeterminates Ui ; thus the Ui are new
variables adjoined to the ring, a feature which, in the eyes of orthodox Dedekindians, is a fatal blemish of this procedure. If, for instance, the a,- are in the
ring k[X, Y]/(F) determined by a plane curve F(X, Y) = 0 with coefficients in
a field k, the ideal generated by them means substantially the same as the set
of common zeros of the ai, counted with their multiplicities; and this is again
nothing else than the "fixed part" of the linear series cut by the variable linear
variety 2=o UiXi = 0 through the point 0, in the affine space of dimension
m + 1, on the model of the given curve which is the locus of the point
(aQ, , am). If we translate this into the projective language, we find ourselves
at the heart of the theory of linear series; and a slight extension of Kronecker's
idea could lead us very naturally to such thoroughly "modern" topics as, for
example, the associated form of a variety in projective space (the "Chow coordinates"). There is thus every reason to believe that the same idea will reacquire
its full meaning in number-theory as soon as the interpntration of numbertheory and algebraic geometry, which Kronecker sought to realise, has been
accomplished. Let us for instance try to define for number-fields a concept corresponding to the degree of a projective curve. If / 0 , / i , - , fm are the coordinates of a generic point of a curve, and the Ui are indeterminates, the degree
is the number of "variable" zeros of ^2iUifi(x); this must be equal to the number of fixed poles minus the number of fixed zeros; in other words, if at every
point P of the curve we put n(P) = mU<aP(fi), where P(f) indicates the order of
/ at P , then the degree of the curve is d = ^Pn(P). If we replace the fi by
numbers & belonging to an extension k of degree n of the rational number-field
Q and if ? is the point = (&>,,) in the projective ra-space, we are thus
led to consider the number H() = YLv sup^(&), where the product is taken
over all absolute values (p-adic or archimedean) of k; H() does hot change if
the & are replaced by p&, with- p k. This concept is essentially due to Siegel
[9]1; as D. G. Northcott indicates [8a], it is more convenient, for arithmetical
purposes, to introduce the number h(%) = H()Un, which depends only upon
the point and not upon the field k* We khall call h() the height of the point
. Following Kronecker, we may associate with the point , with coordinates in
k, the form F(u) = r-Nk/Q(%2 ^, where the Ui are indeterminates, and the
rational number r is so chosen that the coefficients of F(u) are rational integers
without common divisor. Then we have F(u) <$C (A( , J} w *) n (which means
1
Cf. also H. Hasse, Monatshefte fr Mathematik vol. 48 (1939) p. 205. Actually there is
a slight discrepancy between Northcott's definition of H() and that of Siegel and Hasse;
we follow the latter.

NUMBER-THEORY AND ALGEBRAIC GEOMETRY

97

that every coefficient of F is at most equal in absolute value to the corresponding


one in the right-hand side) : hence, if n0 and hQ are given, there is at most a finite
number of points for which n < nQ, h() < hQ. This is Northcott's lemma
([8a]; cf. [llh]), which is at the bottom of the application of the "infinite descent" to elliptic curves, and, more generally, to Abelian varieties over algebraic
number-fields ([lia]; cf. [8b] and [lie]).
The height of a variable point on a curve or on a variety can best be studied
by means of the theory of distributions; this is the only chapter of Kroneckerian
geometry which has been developed beyond the rudiments. Let us first consider
a curve C, defined over an algebraic number-field k; if it is rational, i.e., if its
function-field is the field k(t) generated over k by a single variable t, every function f(t) on it can be written as f(t) T]X'( ~ a%)mi> where 7 is a constant, the
ai are the poles and zeros of the function, and the integers mi are their multiplicities (counted positively for a zero, negatively for a pole). If the curve is not
rational, such a representation is not possible, except in a merely symbolical
manner, or else by means of transcendental multi-valued functions which cannot
be used for arithmetical purposes. Let us, however, consider for a moment a
definite embedding of k in the field of Complex numbers, so that C is defined
over that field; and consider merely absolute values. Then one can attach to
each point A of C a continuous real-valued function dA(M) on G, with
0 <! dA(M) < 1, which is 0 when M = A and only then, in such a way that if
a function f(M) belonging to the function-field of G (over k or even over the
field of complex numbers) has the zeros and poles A i with the multiplicities mi,
then
y(M)JldAi(M)mi,

f(M) =

where y(M) is an inessential factor in the sense that there are constants 71,
72, both > 0, such that 71 < y(M) < y2 for all M. This can easily be verified
by elementary topological methods. It can also be proved by an algebraic argument, which remains valid if the field of complex numbers is replaced by the
algebraic closure of the p-adic field, and also if the curve C is replaced by a
variety. Reduced to its essential features, this argument can be described as
follows. If y is a variety in an affine space, defined over the complex numberfield or over a p-adic field, and if it does not contain the origin, then there is a
polynomial P(Xi, , Xn ) , vanishing on V and not at 0, with coefficients in
the ground-field; this means that all points of V must satisfy an equation
y ' i . . . y'n

1 Y\
1

j^,aVl

...

VnA.\

A.

where all terms in the right-hand side are of degree > 1; therefore, if (xi, , xn)
is such a point, sup* | xi | cannot be arbitrarily small, and precisely it must be
> 1 or > ( 2 I avi vn |) _ 1 ; here | | denotes of course the ordinary orthep-adic
absolute value, as the case may be.
So far we have considered only one absolute value, ordinary or p-adic, at a
time, and so we have obtained, in this sense, merely "local" results; global re-

98

ANDR WEIL

suits come from the consideration of all absolute values simultaneously; or else,
what amounts to the same thing, one can treat the archimedean absolute values
separately, in the manner indicated above, and then deal simultaneously with
all the others. This is done by remarking that, if a variety V is defined, over an
algebraic number-field k, and does not go through 0, its points must, as above,
satisfy an equation
oto = JLfttvi vn Xi

Xn

whose coefficients are algebraic integers in ft; and then if (xi, , xn) is a point
on V, with algebraic coordinated, the G.C.D. of the numerators of the fractional principal ideals (xi), , (xn) must divide the principal ideal (a0) of ft.
Out of this very simple fact one derives all the known results of the theory of
distributions,- one of whose main results is the following "theorem of decomposition":
Let G be a curve, defined over an algebraic number-field ft. One can attach
to each algebraic point A on G a function aA(M), defined at all algebraic points
M of G, whose value at M is an ideal of the algebraic number-fieldft(A, M), so
that the following properties hold: aA(M) is 0 when M A and only then;
and whenever / is a function on G, having the zeros and poles Ai with the multiplicities mi, then the principal fractional ideal (f(M)) has the expression

(/m) = ow ifouW1,
where c(M) is an inessential fractional ideal in the sense that both c(M) and
c(M)~~x divide a fixed natural integer. Furthermore, exactly the same result.
holds for every nonsingular projective variety V of any dimension r, except
that, of course, the ideal-valued functions a(M) are then attached, not to the
points of V, but to the subvarieties of V of dimension r 1.
As we have said above, this becomes a truly global result if we combine it
with the corresponding result over complex numbers. When this is done, one
finds inequalities for the height of a variable point on a, projective variety, which
is found to depend essentially only upon the class of the divisors in the linear
series determined on the variety by its hyperplane sections. In particular, let
G be a curve of degree d in a projective space; let G' be a curve, birationally
equivalent to G, of degree d', in the same or in another projective space; let
M, Mf be corresponding points on G, C, with algebraic coordinates; then, to
every e, there are constants 71, 72, both > 0, such that
lld e

yih(M)

< h(M')lld' < y2h(M)lld+t

for all pairs of corresponding points M, Mf on ; in this sense, the "order of


magnitude" of h(M)1,d is independent of the projective model chosen for C.
This is the decisive inequality for Siegel's proof of the fact that a nonrational
curve can have at most a finite number of points with integral coordinates in a
given algebraic number-field ([9]; cf. [Uh]). The same approach also leads very
simply to Northcott's inequalities ([8]; cf. [Uh]); these contain as special cases

NUMBER-THEORY AND ALGEBRAIC GEOMETRY

99

the inequalities by which it was first proved that the points on an Abelian variety, with coordinates in a given algebraic number-field, form a finitely generated
group ([lia]; cf. [lie]), so that a thoroughly "modernized" version of that proof
could now be given.
I should like to conclude with a brief discussion of a very interesting conjecture, due, I believe, to Hasse. As we have said, from the Kroneckerian point of
view the fields of dimension 1 are the number-fields and the function-fields of
curves over finite fields; to each one of these there belongs a zeta-function, the
properties of which may be said to epitomize in analytic garb some of the more
important properties of the field. It is therefore reasonable to guess that similar
functions can be attached to fields of higher dimension, and in the first place
to the fields of dimension 2, i.e., to the curves over an algebraic number-field,
and to the surfaces over a finite field. Consider the latter problem first: let S
be a surface over the finite field ft of q elements; and define NP, for each v, as
the number of points on the surface with coordinates in the extension kv of
degree v of the ground-field; the analogy with curves, as well as the consideration
of some special cases, makes it very natural ([llg]) to introduce the function
Z(q~), where Z(U) is defined by Z(0) = 1, d log Z(U)/dU = J^iN^U^1,
and to expect that this will have the essential properties of a zeta-function over
a finite field; i.e., that it is a rational function of U, that it satisfies a functional
equation, and that it satisfies a suitably modified Riemann hypothesis; even
the first property seems exceedingly difficult to prove at present, except in
special cases. Now, suppose that we have on S a family of curves C(t) depending
upon a parameter t; for simplicity assume that C(t) depends rationally upon t,
and that no two curves C(t) have a point in common. If we give to t a value
which is algebraic over ft, C(i) will be defined over k(t), and a zeta-function will
be attached to it, defined in a manner similar to that employed for S. As the
number of points on S with coordinates inftis obviously the sum of the same
numbers for all the curves C(t), it follows at once that Z(U) is the product of
the zeta-functions attached to the curves C(t), provided that we take only one
representative for each set of curves conjugate to each other over ft. Now this
definition may at once be transported to number-fields: if G is a curve over the
algebraic number-field K, given by an equation F(X, Y) = 0, then, for almost
all prime ideals p of K, the equation F = 0, reduced modulo p, will define
a curve of the same genus as G over the finite field of q = N($) elements; this
has a zeta-function; and we are thus led to consider the product of these zetafunctions for all p, which is precisely the function previously defined by Hasse,
of which he conjectured that it can be continued analytically over the whole
plane, that it is meromorphic, and that it satisfies a functional equation. In a
few simple cases, this function can actually be computed; e.g., for the curve
Y2 = X* 1 it can be expressed in terms of Hecke's L-functions for the field
ft(\/l) ; this example also shows that such functions have infinitely many poles,
which is a clear indication of the very considerable difficulties that one may
expect in their study.

100

ANDR WEIL
REFERENCES

l . C . CHABATJTY, Sur les quations diophantiennes lies aux units d'un corps de nombres
algbriques fini, Annali di Matematica (IV) vol. 17 (1938) p. 127.
2. C. CHEVALLEY, On the theory of local rings, Ann. of Math. vol. 44 (1943) p. 690.
3. W. L. CHOW, Algebraic systems of positive cycles in an algebraic variety, Amer. J.
Math. vol. 72 (1950) p. 247.
4. I. S. COHEN, On the structure and ideal theory of complete local rings, Trans. Amer.
Math. Soc. vol. 59 (1946) p. 54.
5. L. KRONECKER, a) Grundzuge einer arithmetischen Theorie der algebraischen Grssen,
Werke, vol. II, Leipzig, Teubner, 1897, pp. 237-387.
b) Zur- Theorie der elliptischen Funktionen, Werke, vol. IV, Leipzig-Berlin, Teubner,
1929, pp. 345-495.
6. W. KRULL, Dimensionstheorie in Stellenringen, J. Reine Angew. Math. vol. 179 (1938)
p. 204.
7. E. LUTZ, Sur Vquation y2 = xz Ax B dans les corps ^-adiques, J. Reine Angew.
Math. vol. 177 (1937) p. 238.
8. D. G. NORTHCOTT, a) An inequality in the theory of arithmetic on algebraic varieties,
Proc. Cambridge Philos. Soc. vol. 45 (1949) p. 502.
b) A further inequality in the theory of arithmetic on algebraic varieties, ibid. p. 510.
9. C. L. SIEGEL, ber einige Anwendungen diophantischer Approximationen, Abhandlungen der Preussischen Akademie der Wissenschaften (Jahrgang 1929, no. 1), Berlin, 1930.
10. TH. SKOLEM, Einige Stze ber ip-adische Potenzreihen mit Anwendung auf gewisse
exponentielle Gleichungen, Math. Ann. vol. 111 (1935) p. 399.
11. A. WEIL, a) L'arithmtique sur les courbes algbriques, Acta Math. vol. 52 (1928) p.
281 (Thse, Paris 1928).
b) Arithmtique et gomtrie sur les varits algbriques, Actualits Scientifiques et
Industrielles, no. 206 (Exp. math, la mm. de J. Herbrand XI), Paris, Hermann et Cie,
1935.
c) Arithmetic of algebraic varieties (Russian), Uspehi Matematieeskih Nauk, vol. 3
(1937) p. 101.
d) Sur les fonctions algbriques corps de constantes fini, C. R. Acad. Sci. Paris vol. 210
(1940) p. 592.
e) Sur les courbes algbriques et les varits qui s'en dduisent, Actualits Scientifiques
et Industrielles, no. 1041 (Pubi. Inst. Math. Strasb., VII), Paris, Hermann et Cie, 1948.
f) Varits abliennes et courbes algbriques, Actualits Scientifiques et Industrielles,
no. 1064 (Pubi. Inst. Math. Strasb. VIII), Paris, Hermann et Cie, 1948.
g) Numbers of solutions of equations in finite fields, Bull. Amer. Math. Soc. vol. 55 (1949)
p. 497.
h) Arithmetic on algebraic varieties, Ann. of Math. vol. 53 (1951) pp. 412-444,
12. O. ZARISKI, Theory and applications of holomorphic functions on algebraic varieties,
Memoirs of Amer. Math. Soc, no. 5 (1951).
UNIVERSITY OF CHICAGO,
C H I C A G O , I I I . , IT. S.

A.

INTERNATIONAL CONGRESS
OF

MATHEMATICIANS

Cambridge, Massachusetts, U. S. A.
1950

CONFERENCE I N ANALYSIS

Committee
Marston Morse (Chairman)
L. V. Ahlfors

G. C. Evans

Salomon Bochner

EINAR HILLB

Das könnte Ihnen auch gefallen