Mathematical Association of America, Taylor & Francis, Ltd. The American Mathematical Monthly

The Historical Development of Algebraic Geometry
Author(s): J. Dieudonne
Source: The American Mathematical Monthly, Vol. 79, No. 8 (Oct., 1972), pp. 827-866
Published by: Taylor & Francis, Ltd. on behalf of the Mathematical Association of America
Stable URL: https://www.jstor.org/stable/2317664
Accessed: 27-03-2019 00:59 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Mathematical Association of America, Taylor & Francis, Ltd. are collaborating with
JSTOR to digitize, preserve and extend access to The American Mathematical Monthly
This content downloaded from 186.81.145.194 on Wed, 27 Mar 2019 00:59:12 UTC
All use subject to https://about.jstor.org/terms
THE HISTORICAL DEVELOPMENT OF ALGEBRAIC GEOMETRY
J. DIEUDONNE, University of Nice, France and University of Maryland
I. THEMES AND PERIODS
Modern algebraic geometry has deservedly been considered for a long time as
an exceedingly complex part of mathematics, drawing practically on every other part
to build up its concepts and methods and increasingly becoming an indispensable tool
in many seemingly remote theories. It shares with number theory the distinction of
having one of the longest and most intricate histories among all branches of our sci-
ence, of having always attracted the efforts of the best mathematicians in each genera-
tion, and of still being one of the most active areas of research. Both are perhaps the
best candidates for the perfect mathematical theory, according to Hilbert's ideas: if we
agree with him that problems are the lifeblood of mathematics, then certainly we
may say that algebraic geometry and number theory always have had more open
problems than solved ones, and that each progress towards their solution has always
brought with it a host of new and exciting methods.
Human minds being unable to grasp complex matters as a whole, I have thought
it would be helpful to describe the history of algebraic geometry as a kind of two-
dimensional pattern, where many varied trends of thought, belonging to a few big
themes, weave their way as multicolored threads through the moving succession of
years. It should, however, be emphasized from the start that such a presentation
inevitably inflicts distortions on reality: these themes constantly react on one another
and any division of time into periods is bound to founder on the fact that periods
almost always overlap.
With these reservations, we may first group the main ideas of algebraic geometry
as follows:
(A) and (B) The twin themes of classification and tranisformation, hardly to be
separated, since the general idea behind classification of algebraic varieties is to
put together those which can be deduced from each other by some kind of "trans-
It is hardly necessary to identify Prof. Dieudonne to our readers; still a few facts may prove
interesting. Prof. Dieudonne studied at the Ecole Normale superieure from 1924-27, was a Fellow
at Princeton, Berlin, and Zurich, and received his Doctorate in 1931. He served on the faculties
at Bordeaux, Rennes, Nancy, Sao-Paulo, Michigan, Northwestern, l'Institut des Hautes Etudes
Scientifiques, and was the Dean of Faculty at Nice until his retirement. He held Visiting Professor
ships at Columbia, Johns Hopkins, Rio de Janeiro, Buenos Aires, Pisa, Maryland, Tata Institute
Bombay, Notre Dame, and Washington. His honors include the Order of the Legion of Honor, the
Order of the Academic Palms, and membership in the Academy of Sciences. He served as President
of the Mathematical Society of France in 1964-65.
Prof. Dieudonn6 has published a number of books and about 135 research articles on analysis,
topology, spectral theory, classical groups, formal Lie groups, and non-commutative rings. This
article was prepared while the author was a Visiting Professor at the University of Maryland. Editor.
827
828 J. DIEUDONNE [October
formation." Subordinate to these themes are the notion of invariant, both of al-
gebraic type and of numerical type (such as dimension, begree, genus, etc.), and the
concepts of correspondence and of morphism, which give precise meanings and
extensions to the vague idea of "transformation."
(C) Infinitely near points: a thorny problem, which has plagued generations
of mathematicians: the definition and classification of singularities, the correct
definition of "multiplicity" of intersections, later the concept of "base points" of
linear systems, and the recent introduction of rings with nilpotent elements, all
belong to that theme.
(D) Extending the scalars: a giant step forward in the search for simplicity:
the introduction of complex points and later of generic points were the forerunners
of what we now consider as perhaps the most characteristic feature of algebraic
geometry, the general idea of change of basis.
(E) Extending the space: another fruitful method for extracting understandable
results from the bewildering chaos of particular cases: projective geometry and
n-dimensional geometry paved the way for the modern concepts of "abstract"
varieties and schemes.
(F) Analysis and topology in algebraic geometry. This theme beautifully

exemplifies the cross-fertilization between various branches of mathematics. Out
of a problem of integral calculus, the computation of elliptic integrals and of their
generalizations, adelian integrals, Riemann developed the concept of Riemann
surface (the first non-trivial example of "complex manifold"), invented algebraic
topology, and he and his successors showed how these ideas completely renewed
the theory of algebraic curves and surfaces. One hundred years later, history re-
peated itself when A. Weil transferred to algebraic geometry the notion of fiber
bundle, and Serre the idea of using sheaves and their cohomology, which he and
H. Cartan had shown to be so effective for complex manifolds.
(G) Commutative algebra and algebraic geometry. As we shall see, this has
grown into the most important theme for modern algebraic geometry. Since Riemann
introduced the field of rational functions on a curve, Kronecker, Dedekind and
Weber the concepts of ideals and divisors, commutative algebra has become the
workshop where the algebraic geometer goes for his main tools: local rings, val-
uations, normalization, field theory, and the most recent and most efficient of all,
homological algebra.
Needless to say, within the scope of this article, it will be impossible to do more
than deal with a few of the highlights of our history, leaving aside a large number of
important developments which should be included in a reasonably complete survey.
1972] HISTORICAL DEVELOPMENT OF ALGEBRAIC GEOMETRY 829
II. FIRST PERIOD: "PREHISTORY"
(CA. 400 B.C.-1630 A. D.)
If it is true that the Greeks invented geometry as a deductive science, they never
(contrary to popular beliefs) made any attempt to divorce it from algebra. On the
contrary, one of their main trends was to use geometry to solve algebraic problems,
and this is best exemplified in the invention of the conics, the first curves which they
thoroughly studied after straight lines and circles. The Greeks knew simple geometric
constructions for the root of the equation x2 = ab, a and b being given as lengths of
segments, and the unknown x being considered as the side of a square; they usually
wrote the equation as a "proportion" alx = x/b. The "Delic problem" called f
construction of a length x of given cube, x3 = a2b; this was transformed by
Hippocrates of Chio (around 420 B.C.) into a "double proportion" alx = x/
for two unknown lengths x, y. Menechmus (ca. 350 B.C.) had the idea of considering
the loci given by the two equations ay = X2 and xy = ab, whose intersection has
as coordinates x, y a solution of the problem. This may seem to involve knowledge
of analytic geometry; actually the Greeks made extensive use of coordinates (in
particular for the later theory of conics by Apollonius), without however reaching
the general point of view of Descartes and Fermat (see below).
This method of solving equations by intersections of curves had in fact already
been used in the 5th century B.C., and led to the invention of many curves, both
algebraic and transcendental; of course, the distinction between the two kinds of
curves could not be perceived during that period, and more generally, there was no
attempt at classification, for which no rational basis existed. Besides planes and
spheres, the Greeks also studied some surfaces of revolution, such as cones, cylinders,
a few types of quadrics and even tori; after having discovered conics "analytically,"
Menechmus was also the first to recognize that they could be obtained as plane
sections of a cone of revolution; and a bold construction of Archytas (late 5th
century B.C.) gave a solution of the Delic problem by the intersection of a cone,
a cylinder and a torus. Finally, in his astronomical work, Eudoxus was led to de-
scribe the intersection of a sphere and a cylinder as the trajectory of a movement
conceived as the superposition of two rotations, which may be considered as the
first example of a parametric representation of a curve.
III. SECOND PERIOD: "EXPLORATION"

(1630-1795)
For once, this period has a very well-defined starting point, the independent
invention by Fermat and Descartes of "analytic geometry," which certainly also
marks the true birth of algebraic geometry. The main novelty compared to the
way the Greeks used coordinates is that the same axes are used for all curves (fixed
830 J. DIEUDONNti [October
or variable) which are being considered in a problem, and above all the fact that
the algebraic notation of Viete and Descartes opens the way to the consideration
of arbitrary equations (where the Greeks could not go beyond the third or fourth
degree). Within this frame, the distinction between algebraic and transcendental
curves immediately emerges; the concept of dimension is already clear to Fermat,
who explicitly states that a single equation defines a curve in 2 dimensions, a surface
in 3 dimensions, and already hints at the possibility of generalization to higher
dimensions. The degree of a plane curve is at once seen to be invariant with respect
to a change of coordinates, and Newton knows that it is also in-
variant under a central projection (an operation which was familiar
since the study of conic sections by the Greeks). Themes
The chief work of that period is one of exploration. Fermat A and B
shows that all curves of degree 2 are conics, and Newton classifies
all plane cubics with respect to change of coordinates and projections; Euler clas-
sifies the quadrics, and the first skew curves, given as intersection of two surfaces,
appear in the 18th century. The concept of parametric representation of a curve is
fundamental in Newton's approach to calculus, and Euler knows how to get in
certain cases a parametric representation from the cartesian
equation. A beginning is made in the elucidation of the structure Theme

of singular points and inflexion points of algebraic plane curves,
although limited to the most elementary cases, so that no general
description is yet obtained.
The problem of intersection of two algebraic plane curves is already tackled by
Newton; he and Leibniz had a clear idea of "elimination" processes expressing the
fact that two algebraic equations in one variable have a common root, and using
such a process, Newton observed that the abscissas (for instance) of the intersection
points of two curves of respective degrees m, n, are given by an equation of degree
< mn. This result was gradually improved during the 18th century, until Bezout,
using a refined elimination process, was able to prove that, in general, the equ-
ation giving the intersections had exactly the degree mn; however, no general at-
tempt was yet made during that period to attach to each intersection point an
integer measuring the "multiplicity" of the intersection, in such a way that the
sum of the multiplicities should always be mn. Bezout also generalized his elimi-
nation process to 3 dimensions, proving that the points of intersection of three
algebraic surfaces of degrees m, n, p are in general given by an equation of degree
mtip.
With the beginning of the consideration of algebraic families of algebraic curves

a problem in a sense converse to the problem of intersections appeared, namely the
determination of a curve of given degree n containing sufficiently many given points.
It should be recalled here that this (linear) problem was the starting point for the
theory of determinants, and the fact that n(n + 3)/2 points in "general position"
completely determine a curve of degree n, whereas two curves of degree n have in

general n2 common points, gave the first general example of the concept of rank
for a system of linear equations ("Cramer's paradox").
We should finally stress the fact that a number of ideas fully developed during
the next period may be traced back (in an embryonic form) to the 17th or 18th
century, as we shall see below.
IV. THIRD PERIOD: "THE GOLDEN AGE OF PROJECTIVE GEOMETRY"

(1795-1850)
Here again we have a rather sharp break with the past at the beginning of this
period. In the space of a few years, with Monge and his school and especially with
Poncelet, a new era begins with the simultaneous introduction of points at infinity
and of imaginary points: "geometry" will now, for almost 100
years, exclusively mean geometry in the complex projective plane Themes
P2(C) or the complex projective 3-dimensional space P3(C). D and E
In fact, the fundamental idea of (real) projective geometry goes
back to Desargues (17th century) who, trying to give mathematical
foundations to the methods of "perspective" used by painters
and architects, had introduced the concept of "point at infinity," and the use of
central projections as a means of getting new theorems from classical results of
Euclidean geometry; and although these ideas had inspired Pascal in his work on
conics, they had very soon dropped into oblivion, due to the outlandish language of
the author and the very limited diffusion of his book (which was for some time
believed lost). Other mathematicians in the 18th century, in particular Euler and
Stirling, had hinted at the existence of imaginary points, in order to state general
theorems without distinction of various cases. This is precisely what is brilliantly
accomplished by the new school: circles now intersect in 4 points as any two conics
should, but two of the points are imaginary and at infinity; instead of several kinds
of conics and quadrics, all nondegenerate conics (resp. quadrics) are now projectively
equivalent; instead of the 72 kinds of cubics enumerated by Newton, only 3 remain
projectively distinct; etc.
The chief beneficiaries of these new ideas are at first the theory of conics, quadrics
and of linear families of conics and quadrics; but curves and surfaces of degree 3
or 4 are also investigated in this way, revealing beautiful new theorems, such as the
cgnfigurations of the 9 inflexion points of a plane cubic, the 27 lines on a cubic
surface, the 28 bitangents to a plane quartic; the theorem of Salmon, proving the
constancy of the cross ratio of the 4 tangents to a cubic issued from a point of the
curve, was to gain even more significance later, as the first concrete example of a
"module" in Riemann's sense for an algebraic curve.
Although, with Mobius, Plucker and Cayley, projective geometry received a
sound algebraic basis by the use of homogeneous coordinates, a general tendency
of the projective school was to minimize as much as possible algebraic computations,

and to rely instead (beginning with Poncelet) on general heuristic "principles"
which they did not bother to justify algebraically. The remarkable
success they had in this direction was chiefly due to their skillful
use of the idea of geometric transformation, which for the first Theme B
time comes to the forefront in geometry, preparing the ground
for Klein's famous "Program" linking geometry and the theory
of groups. Most of the transformations they consider are linear: for instance, one
of their favorite 'devices in the theory of conics is to consider a conic as the locus
of two variable straight lines through two fixed points, one of them being derived
from the other by a fixed linear transformation (an idea which, in some particular
cases, goes back to Maclaurin). Similarly, in the study of the linear system of conics
through 4 fixed points, they investigate the intersections of these conics with a fixed
straight line D by considering the (linear) transformation which to a point M of D
associates the second point of intersection with D of the conic of the system which
contains M. Emboldened by the results obtained in this manner, they inaugurated
what was to become the theory of correspondences, by considering what they called
(a, ,)-correspondences, i.e., relations between two points M, M' such that to each
point M there exist a points M' related to M, and to each point M' there exist ,B
points M related to M': when M and M' vary on the same projective line, Chasles'
"correspondence principle" says that the number of points M (counted with multi-
plicities) coinciding with one of their transforms is a + ,B unless every point of the
projective line has that property, a result which it is easy to justify algebraically.
A beautiful application is the Poncelet "closure theorem" for polygons inscribed in
a conic C and circumscribed to a conic C': for a given integer n, one defines on C
a (2,2)-correspondence by assigning to M e C the nth point Mn in a sequence
Mo = M , Ml, Mn, where each side MiMi+1 is tangent to C' and the points Mi
are on C. It is easily seen that for n even, one has M = Mn if Mn/2 is a point common
to C and C', and for n odd, M = Mn if M(n- 1)/2, and the tangent to C
at that point is also tangent to C'. There are thus at least 4 points M on C such that
M = Mn, and by the correspondence principle, if there is still one more point having
that property, then M = Mn for all points on C (one uses of course the parametri-
zation of a conic by the projective line).
Later representatives of the projective school (notably Chasles in France, Steiner

and von Staudt in Germany), somewhat intoxicated by the elegance of their methods,
went so far as to insist that "pure" geometry should be entirely divorced from
algebra and even (with von Staudt) from the concept of real number. As could be
expected, such efforts did not lead very far, and probably hampered progress in the
realization of the importance of linear algebra in classical geometry; it may be,
however, that they paved the way for the later "abstract" algebraic geometry
over a field different from R or C.
In the general theory of algebraic curves (in P2(C)) and surfaces (in P3(C)),
the main problems studied before Riemann are of an enumerative character: to give
only one example of such problems, what is the number of conics tangent to 5
given conics in general position? (The correct answer is 3264.)
Chasles, and later Schubert and Zeuthen proposed half-empirical Theme C
formulas to solve these problems, based on an intuitive concept
of "intersection multiplicity" which could only be justified much
later. One of the main ideas of projective geometry, the concept
of duality, led to the introduction of new "tangential" invariants
for algebraic plane curves: the class (number of tangents through a Theme A
point), the number of inflexion points and the number of double
tangents, culminating in the famous "Plucker formulas"
m' =rn(m -1)-2d -3s,
m = m'(m' - 1) - 2d' - 3s',
s-s = 3(m' - m),
where m is the degree of the curve, ni' its class, d the number of double points,
d' the number of double tangents, s the number of cusps, s' the number of inflexion
points; no "higher singularities," either punctual or tangential, are supposed to
occur.
V. FOURTH PERIOD: "'RIEMANN AND BIRATIONAL GEOMETRY"

(1850-1866)
The importance of Riemann in the history of algebraic geometry can hardly

be overestimated, but in his two fundamental contributions, the "transcendental"
approach via abelian integrals and the introduction of the field of rational functions
on a curve, he built on basic ideas inherited from the previous period.
The origin of abelian integrals is the study of integrals of type
R(t)d t
J P(t)
where P(t) is a polynomial of degree 3 or 4 and R(t) a rational function; one of

these integrals expresses the length of an arc of an ellipse (hence the name "elliptic
integrals"). In the first half of the 18th century, Fagnano and Euler, looking for
some substitute for the classical formula expressing the sum of two arcs of a circle,
when the circle is replaced by an ellipse, found indeed that the sum
fX dt Ty dt
JaVP(t) + IP(t)
can be written
.Ja VP(t) + V(X,y),

where z is an algebraic function of x and y, and V a rational or logarithmic function
of x and y, and Euler had similar results for more general integrals.
At the beginning of his famous work of elliptic functions, Abel made a giant
step forward by showing that the Fagnano-Euler relations were special cases of a
very general theorem: he considers an arbitrary "algebraic function" y of x, defined
as a solution of a polynomial equation F(x, y) = 0; an "abelian integral" f R(x, y)dx
is an integral in which R is a rational function of x, y, in which y is replaced by the
preceding algebraic function (for instance elliptic integrals correspond to F(x, y)
= y2 - P(x)). Then, if G(x, y, a1, **, ar) = 0 is a second polynomial in x, y whose
coefficients are rational functions of some parameters a,, ..-, ar, and if (x1, y1)
(x2, Y2)' * *, (xm,ym) are the points of intersection of the two curves F = 0, G = 0, the sum
' (xI jY,) r (Xm,Ym)
V = J R(x,y)dx + + R(x,y)dx
(a,b) (a, b)
is a rational or logarithm
enough, this is little more
the roots of a polynomial.
in which V is a constant; this leads him to the realization that in that case, any sum
((xI ,y) + (Xm)Yin)
R(x, y)dx + + R(x, y)dx
a, b) (a,.b)
with arbitrary poin

afixed number 5 of v
of the (xj, yj); but, in
he showed that the n
with P of degree ? 5.
Abel, however, worked exclusively within the framework of analysis, and does
not seem to have been acquainted with projective geometry. Furthermore, he ob-
viously had no clear concept of integration in the complex plane (in 1826, Cauchy
had hardly begun his work on that subject), and with the exception of a short and
* Of course, the points xj, yj usually have complex coordinates; an integral

r(X J,yJ)
J(a,b R(x, y)dx

J a, b)
is only properly defined when the path of integration in the complex plane C with extremities a
and Xi has been fixed, and yj is the value taken by y when x varies along the path, y is a continuous
function of x and takes the value b at x = a. When the path is replaced by another one (with the
same extremities), the value of the integral is modified by a "period."
By definition, a logarithmic function of the aj has the form log S (al, ..., ar) where S is rational.
inconclusive note, he has no general discussion of the periods of his integrals.

Thus, although Abel's theorem paved the way for Jacobi's breakthrough in the
problem of inversion of hyperelliptic integrals*, Abel himself narrowly missed the
concept of integral of the first kind and the definition of the genus of a curve (his
failure to take into account the points at infinity has as a consequence the fact that
the 3 integrals he considers are not necessarily of the first kind).
When Riemann takes up the subject in 1851, the intervening years had seen the
great development by Cauchy and his school of the theory of functions of a complex
variable. Indeed, the starting point of Riemann has nothing to do with algebraic
functions, but is the extension of Cauchy's theory to the "surfaces" he introduces
in order better to deal with the so-called "multiform" functions of the most general
(not necessarily algebraic) type. This was already far beyond the contemporary
concepts, and during the 30 years following Riemann, it was the object of long and
tedious explanations by the expositors of his theory. But the way
Riemann uses this notion in order to attack the problem of
abelian integrals is much more original still. Instead of starting Theme F
(as would all his predecessors and most of his immediate succes-
sors) from an algebraic equation F(s,z) = 0 and the Riemann
surface of the algebraic function s of z which it defines, his initial object is an n-
sheeted Riemann surface without boundary and with a finite number of ramification
points**, given a priori without any reference to an algebraic equation (Riemann
* The natural idea of "inverting" the integral fx(Q(t) dt)/1P0(t)=u is to study x as a function o
u, as Abel and Jacobi had done when P has degree 3 or 4; but Jacobi realized that, due to the existence
of 4 periods, no meromorphic function of u could be a solution of the problem. Abel's theorem
finally led him to the correct conception of the problem: one considers two equations
Rx dt Sy dt rx tdt y tdt
faX
Ja / P(t)+Va =u,
/P(t) JaI/VP(t)
+ __ =v
a V/P(t)
and one "inverts" them by expressing the symmetric functions x + y and xy as functions of a
Abel's theorem yields an "addition formula" for these functions, from which one can show tha
are meromorphic and quadruply periodic.
* * The best way to define at least the part of the Riemann surface of a function s(z) (defined by
an algebraic relation F(s, z) = 0 ), containing no point at infinity, is to say that it is the subset of
C2 consisting of the pairs (s, z) satisfying the equation F(s, z) = 0 ; there is then no difficulty with
the "crossing of sheets." Ramification points are those for which OF/8s (s, z) = 0; Puiseux proved
in 1850 that if (so, zo) is such a point, the surface decomposes at that point into a finite number of
"branches" such that each branch can be represented by equations of type
z - zo = th, s-so = alt + a2t2....
where t (the "uniformizing parameter") is in a neighborhood of 0 in C and the series converges (the
integer h depending on the branch).
This description is only correct, however, when at each ramification point (so, zO) there is only
one branch; if not, the point (so, zo) must be replaced by as many points as there are branches; in
other words the points of a Riemann surface are the branches at the various points of the curve.
takes care to complete each sheet with a point at infinity, and thus avoids Abel's
difficulties with these points); then he attacks the problem in the most general
manner possible: classify the integrals of all meronmorphic functions on the surface.
The work of Cauchy and Puiseux had brought to light the general idea of "periods"
of such integrals, generally expressed (as in the example first given by Abel) as an
integral taken along an arc joining two ramification points. Here again Riemann
breaks entirely new ground: he realizes for the first time that topological concepts
are closely related to the problem, and begins by essentially creating the topological
study of compact orientable surfaces, attaching to such a surface S an invariantly
defined integer 2g, the minimal number of simple closed curves Cj on S needed to
make the complement S' of their union simply connected. Then, instead of studying
integrals of meromorphic functions, he defines directly integrals of the first and
second kinds by their periodicity properties, as functions meromorphic on S', and
tending on both sides of each C1 to limits which differ by a quantity kj constant on
Cj (a further reduction of the domain S' is needed to obtain similarly the integrals
of the third kind, having logarithmic singularities)*; integrals of the first kind are
those which have no pole on S. The existence of integrals of the three kinds is proved
by Riemann as a consequence of what he calls the "Dirichlet principle," i.e., the
existence of a harmonic function in S' taking prescribed values on the boundary
(which allows him to prescribe at will the real parts of the kj); and it is also by
an ingenious use of the same principle that Riemann obtains the fundamental re-
lation
g - 1 = w/2 - n
giving the genus in function of the number of sheets n, and the number w of ramifi-
cation points (supposed to be of a "general" type).
The meromorphic functions on S are then the integrals of the first or second
kind whose periods ki all vanish, and Riemann shows that they may be expressed
as rational functions of two of them, linked by an algebraic relation F(s, z) = 0,
thus recovering the older point of view, but immeasurably enriched
with new insights. The choice of these meromorphic functions s,
z is in a large measure arbitrary, and this leads Riemann to his Theme B
next big step forward, the general concept of birational trans-
formation between two irreducible algebraic curves, corresponding
to a biholomorphic mapping of their Riemann surfaces. Here again, Riemann was
not without predecessors: already Newton and his followers had introduced quadratic
transformations such as
xi = 1/x, y' = y/x
in the plane, and observed that they thus transformed an algebraic curve into a
* One simply joins the singularity to one of the Cj by an arc, and deletes the arc from S'.
curve of different degree. "Inversion" in the plane and in 3 dimensional space had
been intensively studied since the early 1820's, chiefly by "synthetic" geometers;
finally, the passage from a plane curve to its transform by duality (exchanging
punctual and tangential coordinates) was obviously a birational transformation
between two algebraic curves, exchanging degree and class. But the startling novelty
of Riemann's approach is of course the fact that to a class of "birationally equivalent"
irreducible algebraic curves he was able to attach his topological
invariant g, the genus of all the curves in the class. But he did not
stop there, and by an evaluation (using two different methods) Theme A
of the parameters on which a Riemann surface of genus g de-
pended, he arrived at the conclusion that classes of isomorphic
Riemann surfaces of genus g > 2 were characterized by 3g - 3 complex parameters
varying continuously (for g = 1 there is only one parameter, and none for g = 0);
the precise meaning of this result (the so-called theory of "moduli" of curves) was
to remain until very recently among the least clarified concepts of the theory.*
VI. FIFTH PERIOD: "DEVELOPMENT AND CHAOS"

(1866-1920)
The extraordinary wealth of new ideas and methods introduced by Riemann

provided inspiration for a steady development of algebraic geometry for over 80
years. But the grandiose synthesis he had envisioned and tried to materialize was
almost immediately broken up by his successors. During that period there will be
at least two or three schools of algebraic geometry, each using different methods,
with little in common even in the fundamental concepts. Riemann's use of analysis,
in particular in the "Dirichlet principle," exceeded the possibilities of his time, and
he had obviously neglected all the difficulties bound to the existence of singular points
on algebraic curves. The first task to which each school of algebraic geometry ad-
dressed itself was therefore the systematization of the birational theory of algebraic
plane curves, incorporating most of Riemann's results with proofs in conformity
with the principles of the school. Then, with varying success, they tried to extend
their methods to the theory of algebraic surfaces and higher dimensional algebraic
varieties.
VI a: The algebraic approach. Historically, this was the latest one, being initiated
by two fundamental papers in 1882, one by Kronecker and one by Dedekind and
Weber. But in the light of subsequent history, it is the trend which was to exert the
deepest influence on the birth of our modern concepts; in particular, just as Riemann
* One should emphasize the fact that this only describes the first half of Riemannl's paper on
abelian integrals; the second part, which solves in a masterly way the inversion problem by the intro-
duction of the general "theta functions" has been, if anything, even more influential on the develop-
ment of analysis.
had revealed the close relationship between algebraic varieties and the theory of
complex manifolds, Kronecker and Dedekind-Weber brought to light for the first
time the deep similarities between algebraic geometry and the burgeoning theory
of algebraic numbers, which were to be some of the main driving forces during the
next periods. Furthermore, this conception of algebraic geometry is for us the
clearest and simplest one, due to our familiarity with abstract algebra; but it was
precisely this "abstract" character which made it the least popular and least under-
stood one in its time.
The work of Kronecker and of his immediate followers, Lasker and Macaulay,
in the first two decades of the 20th century, was of a very general
nature, and its importance only emerged in the later periods: it
essentially consisted in setting up and consistently using an elimi- Theme G
nation method, far more flexible and powerful than the preceding
ones, with the help of which it was for the first time possible to
give a precise meaning to the concepts of dimension and of irreducible variety* and
to show that each variety (defined by an arbitrary system of algebraic equations)
in projective n-space decomposed in a unique way into a union of irreducible varieties
(in general of different dimensions).
The goal of Dedekind and Weber in their fundamental paper was quite different
and much more limited; namely, they gave purely algebraic proofs for all the algebraic
results of Riemann. They start from the fact that, for Riemann, a class of isomorphic
Riemann surfaces corresponds to a field K of rational functions, which is a finite
extension of the field C(X) of rational fractions in one indeterminate over the complex
field; what they set out to do, conversely, if a finite extension K of the field C(X) is
given abstractly, is to reconstruct a Riemann surface S such that K will be iso-
morphic to the field of rational functions on S. Their very original and fruitful
method may be presented in the following way: if the Riemann surface S was already
known, at each point zo E S, a rational function f =# 0 would have an order vz0(f),
namely the integer (positive or negative) which is the degree of the smallest power
in the Puiseux development f(u) = Sk akuk with respect to a "uniformizing param-
eter" u (equal to z - zo if zo is not a ramification point, to some power (z - Zo)1/h
if zo is a ramification point). For a fixed zo E S, the mapping f E-+ vo(ff) of K* into
Z is what is called a discrete valuation on K: we recall that this is by definition a
mapping w: K* - Z such that w(f + g) ? inf(w(f),w(g)) if f + g =# 0, and
w(fg) =w(f) + w(g), which implies w(1) = 0 and w(f ') = - w(f) (w is u
extended to K by taking w(0) = + oo by convention). What Dedekind and Weber
do is to reverse this process, and define a "point of the Riemann surface S of K"
* An irreducible variety V in Pn(C) is characterized by the property that if the product PQ of

two homogeneous polynomials is 0 in V, then one of the two polynomials P, Q must be 0 in V. The
restrictions to V of the rational functions which are defined at one point of V at least then form a
field whose transcendence degree over C is the dimension of V.
as a nontrivial discrete valuation on K (i.e., one which is not identically 0 on K*:

two proportional valuations are then identified).
Now the nontrivial discrete valuations on the field C(X) are easily determined:
one of them (the "point at infinity") wOO is such that w,,(P) = - deg(P) for any
nonzero polynomial P(X); the other ("finite points") correspond bijectively to
the points C E C, the corresponding valuation w4 being such that w4(P) is the o
of the zero C of P(X) (equal to 0 if P(() V 0). It can easily be shown that for e
discrete valuation w of C(X), there is a finite number of nonproportional valuations
vj on K such that for each j, vj/ej reduces to w on C(X), where ej is an integer
one says that the vi are the points of the Riemann surface S above w; the poin
above WO are again called points at infinity, the other finite points.
The elements f E K for which v(f) ? 0 for all finite points v of S constitute
exactly the elements of K which are integral* over the ring of polynomials C[X];
they form what we now call a Dedekind ring A, to which Dedekind's theory of
ideals may be applied.** The maximal ideals 9, of A correspond to the finite points
v y S: v is the set of f E A for which v(f) > 0; the fractionary ideals of K are the
A-modules a contained in K and for which there is an element c 0 0 in A such
that ca c A; each of them can be written uniquely as a product 1C 32 ... jr, where
the 9j are maximal ideals of A and the cj positive or negative integers. Another
way of stating this result is to say that a fractionary ideal a is the set of all f E K
such that vj(f) > cj for t < j < r, where the valuations vj correspond to the
maximal ideals j, and v(f) > 0 for the other finite valuations.
The consideration of the ideals of A, however, leaves the "points at infinity"
out of the picture. This led Dedekind and Weber to generalize the concept of ideal
and to introduce the notion of divisor on K. This is defined as a family D = (orv) of
integers aev e Z, where v runs through all points of S, and aev = 0 except for a finite
number of points: writing (av) + (f,v) = (av + f3v) defines the set 2(K) of divisors
of K as an additive group isomorphic to Z(s), in which an order relation is naturally
defined, (oc,) < (f,3) meaning that ocv _ fv for all v E S; a divisor D = (j) such that
av_ > 0 for all v E S is called positive or effective. The degree deg(D) of D = (av) is
defined as Ev s ocv (positive or negative integer); the support of D is the set of the
v E S for which cxv = 0. One of the problems considered by Riemann was the de-
termination of rational functions on a Riemann surface having poles of orders
< cp for prescribed points P (in finite number) on S. Using his bold expression
functions as sums of abelian integrals, he found that there existed rational functions
having that property for an arbitrary choice of the points P as long as Y2pcxp > g + 1,
whereas if Yp2cp < g, this was only possible for special positions of the points P.
This result was completed by his student Roch, and put in its final form by Dedekind
* Recall that an element x of a ring R is integral over a subring S if it satisfies an equation of

type xm + a, x + ... + am = O, with aj E S.
** Dedekind had developed this theory for algebraic number fields from 1870 on.
and Weber in the following way: the problem is a special case of the study of the
set L(D) of rational functions f E K satisfying the conditions
(1) v(f) > - av for all v E S
for a given divisor D = (oav); it follows from the axioms of valuations that L(D) is
a complex vector subspace of K, and it can be shown that this subspace has finite
dimension l(D).
A fractionary ideal may be described as the union of the increasing family of
spaces L(Dm), where Dm = (,) is such that the av coincide with the - ac for th
are equal to 0 for the other finite points, and to m for the points at infinity.
The relations (1) can be written in a different way. For each f c K*, there are
only a finite number of valuations v c S such that v(f) =A 0; let (f)0 (resp. (f)00)
be the positive divisor ((v(f))+) (resp. ((v(f)) )) (in the "transcendental" interper-
tation, (f)o is the "divisor of zeroes" and (f)0,, the "divisor of poles" of the rational
functionf ), atid let (f) = (1)0 -(f)0,, in the group q(F); (f) is called the principal
divisor defined by f. It can be shown that deg ((f)) = 0 by purely algebraic arguments
(in the transcendental picture, this is merely the residue theorem)*; in particular, if
v(f) > 0 for all v E S, then f cC (only constants are everywhere holomorphic on
a Riemann surface) and if in addition v(f) > 0 for some v, then f = 0. With these
definitions, the relations (1) for f # 0 are equivalent to the inequality
(2) (f) +D > 0

in the ordered group ?q(K).
Principal divisors form a subgroup gA(K) of 9(K) (isomorphic to the group

K*/C*, two elements of K* which have the same principal divisor differing by a
constant factor by the previous remarks). Divisors belonging to the same class in the
quotient group W(K) = 9(K)/1Y(K) are called (linearly) equivalent: to say that D
and D' are equivalent means therefore that there existsf =A 0 such that D' - D = (f);
it is clear that deg(D') = deg(D) and l(D') = I(D) for equivalent divisors; two
elements f, g of L(D) are such that (f) + D = (g) + D if and only if f/g is a constant,
in other words, the set ID I of positive divisors equivalent to D is identified to th

projective space P(L(D)) of dimension l(D) - 1.
The Riemann-Roch theorem is then written in the following way:
(3) I(D)-l(A-D) = deg (D) + 1 -g,
where g is the genus, and A belongs to a well-determined divisor class, called the
canonical class of K. To define it in the transcendental interpretation, one considers
on the Riemann surface S a meromorphic differentialform co: at each point P of S,
* One integrates the differential df/f on the boundary of the simply connected part S' of the Rie-
mann surface, taking into account that each arc of that boundary comes twice in the integral with
opposite orientations.
the differential form co may be written F(u)du, where u is the uniformizing parameter
in a neighborhood of P and F is meromorphic at P; if bp is the order of F at the
point P, (6p) is a canonical divisor, and it does not depend on the choice of the
uniformizing parameters. Any other meromorphic differential form may be written
fwo withfc K, hence all canonical divisors belong to the same class. There is a purely
algebraic definition of A (see section VII b), and one proves that deg(A) = 2g - 2
for g > 1 and 1(A) =g. Relation (3) implies Riemann's result on the poles of rational
functions; more generally, if deg (D) > g + 1, (3) implies l(D) > 2; if D > 0, L(D)
always contains the constant functions, and to say that l(D) > 2 means that it
contains a non constant rational function. From the definition of L(D), it follows
that l(D) = 0 if deg(D) < 0, hence, by (3), l(D) = deg(D) + 1 - g if deg(D) > 2g - 2;
in particular, for any divisor D such that deg(D) > 0, I(mD) = m * deg(D) + 1 - g
for m large enough (although one may have l(D) = 0).
VI b: The Brill-Noether theory of linear systems of points on a curve. An irredu-

cible plane curve F without singularity is identified to its Riemann surface, and a
positive divisor may therefore be identified with a system of points of F, each being
counted with a certain "multiplicity" which is a positive integer. Riemann's de-
termination of the "special" systems of at most g points of F, which may be the
poles of a rational function, had led him (by an extension of some earlier com-
putations of Abel) to define these sets as intersections with F of a family of "adjoint"
curves of smaller degree, subject to linear conditions on the coefficients of their
equations, so that such a family may be considered as given by an equation
Xjr= I RjPj(x, y) = 0 in nonhomogeneous coordinates, where the P1 are polyno
and the )j variable complex parameters. A number of points of intersection of
curves with F may be fixed (i.e., independent of the )j); as the intersection multip
city of a common point of F and of an arbitrary curve F' is immediately defined
since F has no singular point (it is the same as the intersection multiplicity of F'
and the tangent to F), we may consider for each adjoint curve F' of the family the
positive divisor D = Ep mpP - EQ mQ, where P runs through all the intersection
points of F and F', nip is the corresponding intersection multiplicity, Q runs through
the fixed intersection points and mo is the minimum value of mQ when the )j vary.
It is immediate to see that if Do is one of these divisors, corresponding to the values
A? of the parameters, then D = Do + (f), where f = (Zj,jPj)/(ZjA?Pj).
Conversely, given a divisor Do (positive or not), if 1(DO) = r > 0, the functions
f e L(DO) may be written (XJ1j1Pj (x, y))/Q(x, y), where the Pj and Q are poly-
nomials and the )j arbitrary complex numbers; the positive divisors (f) + Do where
f e L(D), are obtained by adding a fixed divisor to the variable divisor obtained
as above from the points of intersection of F and of the curve Eji jPj(x, y) =
The study of the vector spaces L(D) attached to divisors is thus essentially equiv-
alent to the study of the systems of points of intersection (with multiplicities) of F
with the curves F' of a system of curves j ljPj(x, y) = 0. It is in fact by means of
the study of such systems of points, called "linear series"-or "linear systems" on F,
that the geometric school of Clebsch, Gordan, Brill, and Max Noether described
the birational theory of algebraic plane curves after 1866. But they wanted to deal
in this way, not only with curves without singularities, but with arbitrary algebraic
curves, and linear systems of points are only easy to handle when
the curve F has no singularities, or at most "nice" singularities
such as double points with distinct tangents. One of the first Theme C
efforts of that school was therefore to establish the possibility of
finding a birational transformation of an arbitrary irreducible
algebraic curve F into a plane curve with only double points with distinct tangents;
a result proved independently by M. Noether in 1871 and equivalent to a theorem
of algebra obtained by Kronecker in 1862. In view of the extension of this result
during the later periods, it is worthwhile to note that a slightly weaker theorem may
be obtained by a succession of birational transformations of the whole projective
plane P2(C) onto itself of the type
x'/yz = y'/zx = z'/xy
(for suitable homogeneous coordinates), the so-called quadratic transformations.

Such a transformation is bijective outside the sides of the triangle having as vertices
the points (1, 0, 0), (0, 1, 0), and (0, 0, 1) but sends each point of one side (not a vertex
to the opposite vertex, and is indeterminate at a vertex: however, two points ap-
proaching a vertex along distinct lines have transforms which tend to distinct
limits on the opposite side, so that the transformation may be said to "blow up"
a vertex to the opposite side, and separates the branches of a curve having different
tangents at a vertex by transforming them to branches through different points of
the transformed curve. By repeating conveniently this process, one may show that
there is a transformed curve whose singular points are such that each has a number
of distinct tangents equal to its multiplicity. To get curves with only double points,
one uses birational transformations which are only defined on the given curve (and
not in the plane).
It is during the same period, and in the same school, that n-dimensional algebraic
geometry comes into its own for any value of n > 1 (all algebraic
varieties being considered as subvarieties of some P,1(C)). As we
shall see below, the study of algebraic varieties of dimension > 2 Theme E
was to have important repercussions on the theory of algebraic
curves, with the concept of algebraic correspondences as sub-
varieties of a product variety, and the study of abelian varieties. We only mention
here another fruitful consequence, the relation between linear series of points and
rational mappings of an irreducible curve F into a projective space Pr(C): such a
mapping can be written
0w: C P>(Pl m, P in, I Prt+ 1hom),
where the Pj are homogeneous polynomials in the homogeneous coordinates of 4
all of the same degree: if F' is the image of r by 4, the points of intersection of F by
the system of curves Ej 1jP- 0 are the inverse images by 4 of the points of inter-
section of F' by variable hyperplanes. This observation, in connection with the
theory of linear series, enables one to choose the Pi in such a way that 4 is a bi-
rational transformation and 17' has no singular points. Furthermore, the curve rF
having these properties is uniquely determined up to a birational and bijective trans-
formation (one says it is the nonsingular mnodel of the field of rational functions
of r).
VI c: Integrals of differential forms on higher dimensional varieties. As soon as

1870, Cayley, Clebsch and M. Noether inaugurated the study of abelian integrals
on irreducible algebraic surfaces, by considering, on a surface S
in P3(C) given by an equation F(x, y, z) =- 0 in nonhomogeneous
coordinates, double integrals of type ff R(x, y, z)dxdy, where Theme F
R is a rational function; after 1885, Picard began a thorough
investigation of the properties of these integrals, as well as of
simple integrals fP (x, y, z)dx + Q (x, y, z) dy, where P, Q are rational and the
differential is exact*. His method, which (conveniently generalized) is still very useful,
consists in looking at the sections of the surface by the planes y = const., applying
Riemann's theory to abelian integrals on these curves (which in general are irredu-
cible), and studying the way in which they depend on the parameter y; in particular,
if p is the genus of the curve for general values of y, the 2p periods of the abelian
integrals of the first kind satisfy a linear differential equation or order 2p (as functions
of y), the so called Picard-Fuchs equation, which plays an important part in the
theory. The algebraic surfaces considered by these mathematicians were usually
supposed to be without singular points, or at most to have only
"nice" singularities (double curves with distinct tangent planes
except at finitely many points and no singular points except Theme C
finitely many triple points); starting with M. Noether, many
attempts were made to prove that any algebraic surface could be
transformed into surfaces without singularities (not necessarily immersed in P3(C),
* The exact meaning of a simple integral f P (x, y, z) dx consists in assigning to each piecewise
differentiable mapping t -- (x (t), y(t), z(t) ) of an interval [a, b] c R into S (a "singular 1-simplex")
the number bsP(x (t), y(t), z(t)) x' (t) dt. Similarly, the double integral ff R(x, y, z) dxdy assigns
to each piecewise differentiable mapping (u, v)-> (x(u, v), y(u, v), z(u,v)) of a triangle T c: R2 into
S (a "singular 2-simplex") the number
ff R(x(u, v), y(u, v), z(u, v) -( ' Y) dudv.
One can then define in an obvious way the value of simple (resp. double) integrals over 1-chains
(resp. 2-chains), i.e., formal linear combinations of 1-simplices (resp. 2-simplices) with coefficients
in Z (or in R, or in C). Generalizations to higher dimensions are obvious, once one defines an n-
simplex as a piecewise differentiable mapping of the "standard n-simplex" defined by the inequalities
Xj ? 0 (1 j < n), XI + x2 + ... + x? 51 in Rn.
but in higher dimensional projective spaces), but no satisfactory proof was found
until much later.
Very early it appeared that the theory of algebraic surfaces exhibited some
features which had no counterpart in the theory of algebraic curves. Two irreducible
surfaces without singularities may be birationally equivalent without being iso-
morphic. If pg denotes the number of linearly independent double
integrals of the first kind on an irreducible surface S (i.e., integrals
which are finite over any 2-cell of S), the corresponding number Theme A
for a surface S' birationally equivalent to S is not necessarily
the same. The number p, is the obvious counterpart of the genus
of a curve; but very soon also, it was realized that the other definition of the genus
of a curve, using the "adjoints" of Riemann, also generalized to surfaces, but might
give a number Pa different from pg (see in VIIJ-a its exact definition in modern terms
pg was called the geometric genus and Pa the arithmetic genus of S, and the differenc
q = Pg- Pa (which is always ? 0) the irregularity of the surface (for instance,
Cayley found that for ruled surfaces pg = 0 and Pa < 0 in general).
It soon also became apparent that the properties of abelian integrals on a
surface or a higher dimensional variety were to a large extent subordinate to the
topological properties of the variety. H. Poincare had particularly in mind the
applications to algebraic geometry when, in 1895, he started to give mathematical
substance to Riemann's intuition of higher dimensional "Betti
numbers" by inventing the "simplicial" machinery which made
rigorous proofs possible*; algebraic varieties (and more generally Theme F
analytic varieties) are amenable to this technique due to the fact
that they are triangulable, a fact for which Poincare himself
sketched a proof, which was later made entirely rigorous by van der Waerden.
Using this machinery and the Picard technique of variable plane sections, Poincare
was able to bring to a satisfactory conclusion previous efforts by Picard and the
Italian geometers and to prove that the irregularity q of an algebraic surface without
singularity is equal to R1/2, where R1 is the first Betti number, and also equal to
the number of independent simple abelian integrals of the first kind. Around 1920,
Lefschetz considerably developed these techniques and generalized them to algebraic
varieties of arbitrary dimension, concentrating in particular on the determination
of the number of cycles on such a variety V which are homologous to cycles con-
tained in algebraic subvarieties of V: for instance, if V is a projective variety of
complex dimension n, and H a hyperplane section of V, the natural mappings
* Let us recall that to an n-chain is attached a well determined (n-1) - chain, its boundary;
n-cycles are the n-chains whose boundary is 0, and the n-th homology group Hn (M, Z) (resp.
Hn(M, R), resp. Hn(M, C) ) of a manifold M, with coefficients in Z (resp. R, C) is the quotient of
the group of n-cycles with coefficients in Z (resp. R, C) by the subgroup consisting of the boundaries
of the (n + 1)-chains. The Betti number Rp is the dimension of the real vector space Hp (M, R).
Hi(H, Z) -+ Hi(V, Z) of homology groups are bijective for 0 < i < n - 2 and sur-
jective for i = n - 1. He also showed that for an algebraic variety V, one had
R2p >O, Rp _ Rp_2for p ? n (complex dimension of V) and that the Betti
R2p 1 of odd dimension were even.
VI d: Linear systems and the Italian school. The definition of divisors, given in
VI-a, carries over to any field K finitely generated over C; on a nonsingular model
V having K as field of rational functions, the discrete nontrivial valuations of K
now correspond to irreducible subvarieties of V of codimension 1. It is still true
that deg((f)) = 0 for principal divisors, and that L(D) is a finite dimensional sub-
space of K for all divisors D. The concept of linear system of subvarieties of co-
dimension 1 may therefore be associated to the notion of divisor as in VI-b. Around
1890, the Italian school of algebraic geometry, under the leadership of a trio of
great geometers: Castelnuovo, Enriques and (slightly later) Severi, embarked upon
a program of study of algebraic surfaces (and later higher dimensional varieties)
generalizing the Brill-Noether approach via linear systems: they chiefly worked
with purely geometric methods, such as projections or intersections of curves and
surfaces in projective space, with as little use as possible of methods belonging
either to analysis and topology, or to "abstract" algebra.
These limitations implied serious difficulties in the definition of the main con-
cepts and the use of geometric methods. The chief trouble was that whereas on
curves one can work almost exclusively with positive divisors, this is not the case
any more for surfaces: for instance if pg = 0, the canonical divisor (defined as
in VY-a, but for meromorphic differential 2-forms) is not equivalent to a positive
divisor, hence does not correspond to a linear system of curves. This compelled
the Italians to introduce complicated "virtual" notions for linear systems, which
obscured the significance of much of their results.
Working under such considerable handicaps, it is amazing to see how many new
and deep results were discovered by the Italian geometers. It would be extremely
long and intricate to describe these results in their own language (see for instance [
and we shall postpone the definition of the most important notions which they
introduced until we can use the much simpler modern formulation.
Let us only mention here a few of the beautiful theorems charac-
terizing (up to birational equivalence) simple types of surfaces by Theme A
the values of the arithmetical genus Pa and new invariants defined
by Enriques, the plurigenera Pk (k ? 2): a rational surface
(i.e., birationally equivalent to a plane) is characterized by the relations Pa = 0,
P2 = 0, surfaces with Pa < -1 are ruled, whereas the surfaces such that P4 P6 = 0
are either rational or ruled; finally, a surface for which Pa = P3 = 0 and P2 = 1 is
birationally equivalent to the Enriques surface of degree 6 having the 6 edges of a
tetrahedron as double lines (it is not a rational surface, although pg = 0).
VII. SIXTH PERIOD: "NEW STRUCTURES IN ALGEBRAIC GEOMETRY"

(1920-1950)
The general trend towards the unification of mathematics by the study of the
structures underlying each theory, which started to get momentum in the 1920's,
was particularly apparent in the development of algebraic geometry; the striking
kinships between algebraic varieties and complex manifolds on the one hand,
algebraic numbers on the other, which had been discovered in earlier periods, now
became organic parts of the fundamental concepts of algebraic geometry. One of
the effects of this broadened point of view was to loosen the exclusive grip held
until then by projective and birational methods over algebraic geometry, and prepare
the way for a far more flexible approach.
VII a: Kahlerian varieties and the return to Riemann. Ever since Gauss's fundamen-
tal paper of 1826 on the theory of surfaces and Riemann's inaugural lecture of 1854
defining n-dimensional riemannian geometry, the concept of differential manifold,
defined by "maps" and differentiable "transition functions" between maps*, had
gradually become more and more precise as the fundamental topological concepts
needed to express them were defined and studied in the last part of the 19th century
and the beginning of the 20th. One of the most important developments in that
direction was the introduction of the general concept of exterior differential p-form
on a differential manifold (locally defined by expressions
X Aili2 .. ip(x)dx A dxi2 A" A dxip

il <i2 <---.< ip
in the local coordinates) and of their integrals on p-chains (generalizing the earlier
notions of "curvilinear" and "surface" integrals), due to H. Poincare and E. Cartan.
At the very beginning of his papers on algebraic topology, Poincare had pointed
out the connection between the homology of a compact differential manifold V and
the exterior differential forms on V (of which the classical Stokes' theorem is the
simplest example). This was made precise by De Rham's famous theorems in 1931,
starting from the duality between chains and forms given by the integral
< C, cl> = fcwo; due to the generalized Stokes' formula < C, do> =< bC, co> (w
b is the boundary and d the exterior derivative), this yields a duality, pairing the
real homology groups Hi(V, R) of V and the cohomology groups Hi(A)**, where A
is the "complex" of exterior differential forms
(4) 0 d A1 d A2d dAnd0 (n - dim V),

(A' is the R-vector space of the j-forms).
* If M is a differential manifold of dimension n,k: U -+ Rn, g: Y -+

U, V, of M onto Rn, the "transition function" from U to V is the mapping (only defined when
Uf V# 0)x I (y-'(x))of (Uf V)onto v(Un V).
** H'(A) is the quotient of the kernel d- (AE?l) by the image d(A- 1).
A projective algebraic variety without singularity of (complex) dimension n has

a natural underlying structure of differential manifold of dimension 2n, but in fact
it has a much richer structure. In the first place, it is a complex manifold, which
means that for the "maps" which define the differential structure
and which take their values in Cn ( = R2n), the "transition funct-
tions" are holomorphic; it follows that the space Ac of (complex) Theme F
differential p-forms for 1 ? p ? 2n decomposes naturally into a
direct sum of vector spaces Ac' corresponding to the pairs of
integers such that r + s = p; for r ? n and s < n, the forms in A23 (called forms
of type (r, s)) are those which for complex local coordinates zt, z2 , Zn, are written
(5) lAil ...Jirkij... k,(x)dZ ji A*-- A dzirA dkkl A AdZk.,

where the AJ,..k are differentiable functions with complex values (not holomorphic
in general). For r > n or s > n, one takes AtVS as reduced to 0 by convention.
But this is not the end of the story. It is possible to define on a projective complex
space, and by restriction on any complex compact submanifold of such a space
(which is necessarily an algebraic variety by a theorem of Chow) a riemannian ds2
which is kdhlerian, i.e., can be written locally as a hermitian form
ds2 = , hjkdzidik with hkl = hJk

j,k
which has the property that the corresponding exterior 2-form (which is real valued)
(6) Q= (i/2) , hjkdzk A dz3

h,j
is exact (i.e., dQ = 0).

Beginning around 1930, Hodge, in a series of remarkably original papers, showed
how to use these facts to investigate the homology of compact kahlerian varieties.
On a riemannian manifold, Beltrami had shown that it is possible to define an
operator which generalizes the usual laplacian, and therefore enables one to define
harmonic functions on the manifold. By a very imaginative generalization, Hodge
was able to define similarly, on any compact riemannian manifold, the notion of
harmonic exterior differential forms, and to prove that there existed a unique
such form in any cohomology class in any Hi(A); from that result, he deduced the
uniqueness and existence of a harmonic p-form having given periods on homo-
logically independent p-cycles, thus obtaining a complete generalization of
Riemann's fundamental result, and showing that Riemann's use of "Dirichlet's
principle" was far more than a technical device (fortunately for Hodge, the theory
of elliptic partial differential equations had advanced far enough to spare him the
difficulties which had plagued Riemann's approach). Turning next to complex
kahlerian manifolds, the space HP of harmonic p-forms with complex coefficients
splits into a direct sum of p + 1 spaces Hr's consisting of (complex) harmonic forms
of type (5), for r + s = p (with Hr's = 0 if r > n or s > n); it can be shown that
HP"' consists exactly of the holomorphic p-forms (or "differential forms of the fir
kind"), i.e., those for which in (5), s = 0 and the AJ I *Jp are holomorphic. As comp
conjugation transforms Hr,s into HSr, they have the same dimension, and this
shows that the dimension of HP, i.e., the Betti number Rp, is even when p is
On the other hand, one easily verifies that the (real) 2-form Q defined in (6) is har-
monic, as well as all its exterior powers, which proves that R2k > 1 for every in-
teger k. Finally, Of-+- Q A 4 is shown to be an injective mapping of HP into H
for p ? n - 2, from which the inequality Rp+2- Rp > 0 follows; all the Lefschet
theorems on Betti numbers of algebraic varieties are thus "explained" and shown
to belong in fact to the theory of kahlerian manifolds (there are compact kiihlerian
manifolds which are not isomorphic to projective algebraic varieties). We shall
return to the Hodge's theory when in the next period it merges into sheaf cohomology.
VII b: Abstract algebraic geometry. It is well known that, from 1900 to 1930, the
general concepts of algebra (mostly confined until then to real or complex numbers)
were developed in a completely abstract setting, the notion of algebraic structure
(such as group, ring, field, module, etc.) becoming the fundamental one and re-
legating to second place the nature of the mathematical objects on which the structure
was defined. It was therefore quite natural to think of an "abstract" extension of
algebraic geometry, in which the coefficients of the equations and the coordinates
of the points would belong to an arbitrary field. Already Dedekind and Weber, in
their 1882 paper, had observed that all their arguments only used the fact that the
basic field was algebraically closed (and of characteristic 0, a notion which had not
yet been defined then). Even notions which seem linked to analysis, such as de-
rivatives and differentials, had algebraic counterparts: a derivation in a commutative
ring A is an additive mapping x -+ Dx of A into itself such that
D(xy) = x * Dy + (Dx) * y, and a differential is an A-linear Theme

mapping co: Z- A of the A-module of all derivations into A;
for each x E A, dx is the linear form D -+ Dx on Z, and p-forms
are defined by the usual methods of exterior algebra.
The motivation for the development of abstract algebraic geometry was therefore
a natural outcome of the progress of algebra; after 1930, a more powerful impulse
was to come from number theory, as we shall see below.
As it was apparent that a large part of the foundations of classical algebraic
geometry came from geometric intuition, more or less justified by appeals to analysis
or topology, a thorough examination of the basic concepts, from the exclusive view-
point of algebra, was necessary in order to carry out an ambitious program of
algebraic geometry over an arbitrary field. This groundwork, which at the same
time created most of modern commutative algebra, was chiefly due to E. Noether,
W. Krull, van der Waerden, and F. K. Schmidt in the period 1920-1940, and to
Zariski and A. Weil from 1940 on.
19721 HISTORICAL DEVELOPMENT OF ALGEBMIC GEOMETRY 849
The first two of these mathematicians use the geometric language very sparsely;
their results are almost always expressed in the language of rings and ideals, and it
was only after 1940 that the importance of their work was properly appreciated:
the decomposition into primary ideals in noetherian rings, the properties of integrally
closed rings, the extensive use of valuations, the notion of localization and the
fundamental properties of local rings are all due to them. (A local ring is a com-
mutative ring A in which there is only one maximal ideal. The typical example
consists of the rational functions (elements of the field C(X)) for which a given point
C E C is not a pole: they form the local ring of C(X) at the point C). A similar remark
may be made on the foundational work of Zariski, probably the deepest one in that
period; although it is usually expressed in the language of projective geometry, it
mostly belongs to local algebra and its central position in algebraic geometry was
only recognized in the next period. The contribution of F. K. Schmidt (in connection
with his work on number theory which we describe below) essentially consisted in
extending the Dedekind-Weber theory to curves defined over an algebraically closed
field of any characteristic.
The most conspicuous progress realized during that period is the successful
definition, in algebraic geometry over an arbitrary field, of the concepts of generic
point and of intersection multiplicity, due to the combined efforts of van der Waerden
and A. Weil. The Italians (not to speak of their predecessors) used these notions
with a freedom which, to their critics of the orthodox algebraic school, bordered
on recklessness. As long as the underlying field was C, the notion of "elements in
general position" could be easily justified by an appeal to continuity (although the
Italians seldom bothered to prove that these elements formed open sets in the spaces
they considered). On the other hand, Lefschetz had made the elementary but funda-
mental observation that when two subvarieties U, V of Pn(C), of complementary
dimensions r and n - r, intersect transversally in simple points, the number of
these points is equal (for convenient orientations) to the intersection nunmber
(U * V) of the cycles U, V, in the sense of algebraic topology; as this number is
known to be invariant under homology, it was quite natural to take it as the number
of intersections of U and V (counted with multiplicities) in the most general cases.
This justified the extensive use of intersection multiplicity by the Italian geometers,
in particular the "self-intersection" number (C C) of a curve on an algebraic sur-
face. (Unfortunately, the complexity of the Italian definitions was such that it was
often impossible to be sure that the same words meant the same things in two different
papers; hence the numerous controversies between geometers of that school, such
as the one which occurred as late as 1943 between Enriques and Severi, see [4]
and [10].)
These foundations of course disappeared in algebraic geometry over an arbitrary
field, and this was one of the reasons why no algebraic proofs valid over any field
(even of characteristic 0) had been found for the results obtained in the theory of
algebraic surfaces by transcendental or geometric methods. In 1926, van der Waerden
saw that to gain the freedom which Analysis gave for classical geometry over the
complex field, one had only to return to the process which had
allowed the passage from real to complex geometry, namely
enlarge the field k to which the coefficients of the equations of a Theme D
variety and the coordinates of its points are supposed to belong:
if K is any extension of k, these equations are still meaningful
when the coordinates are taken in K. Giving a general form to ideas which went
back at least to Gauss, he introduced the idea of specialization over k of any set of
elements x1, , x,m in an arbitrary extension K of k: it is a mapping which to e
xj assigns an element xJ of an extension K' of k (which may be equal to K), in
a way that for every homogeneous polynomial P E k [X1, , X,..] for which
P(x x, Xm) = 0, one also has P(x, .,x,') = 0 (van der Waerden always works in
projective spaces, or finite products of such spaces). Suppose then that V is an
irreducible algebraic variety in P,,(k), and let K be the field of rational functions
on V; one may assume that V is not contained in a hyperplane of P,,(k); for
1 < j < n, the restriction dj to V of the rational function xF-* x i/x0 (wh
x0, x, ..., xn are homogeneous coordinates of a point, x e P,,(k)) is an element of
if VK is the variety in PJ(K) defined by the same equations as V, the point (1, 4 , .,
belongs to VK. Van der Waerden calls this point a generic point of V, for it is im-
mediate to check that for any extension K' of k, any point of VK' is a specialization
of(1, 41, ., 9). Such points can then be used in the same way as the "general points"
of the Italians, despite their apparently tautological character: any theorem proved
for generic points (and of course expressible by algebraic equations (not inequalities!)
between their coordinates) is valid for arbitrary points of corresponding varieties.
Van der Waerden then proceeded to apply this new tool with great virtuosity to
many problems of algebraic geometry, and in particular to the definition of multipli-
city of intersection of two varieties in abstract algebraic geometry, which had not
yet been given a meaning except in the case of the intersection
of two curves on a surface without singularity. However, Poncelet,
as a consequence of his general vague "principle of continuity," Theme C
had already proposed to define the intersection multiplicity at
one point of two subvarieties U, V of complementary dimensions
by having V (for instance) vary continuously in such a way that for some position V'
all the intersection points with U should be simple, and counting the number of
these points which collapsed to the given point when V' tended to V; in such a way,
the total number of intersections (counted with multiplicities) would remain
constant ("principle of the conservation of number"), and it is thus that Poncelet
proved Bezout's theorem, by observing that a curve C in the plane belonged to the
continuous family of all curves of the same degree m, and that in that family there
existed curves which degenerated into a system of straight lines,-each meeting a
fixed curve F of degree n in n distinct points. Many mathematicians in the 19th
century had extensively used such arguments, and in 1912, Severi had convincingly
argued for their essential correctness. The concept introduced by van der Waerden
was based on similar ideas: under suitable conditions, the multiplicity of a solution
y (yO, -**, y) e Pn(k) of a system of equations P,,(x, y) = 0, where x is a point
of an irreducible variety V, is the number of the solutions il of the system P,(, ) = 0
where X is the generic point of V, which specialize to y when 4 specializes to x.
Using this definition, he was finally able to attach to every irreducible com-
ponent C of the intersection of two irreducible varieties V, W of an "ambient"
nonsingular variety U, an integer i(C, V W; U) _ 0, the multiplicity of C in
V n W, provided all irreducible components of V n W were "proper," i.e., had
a dimension equal to dim V + dim W - dim U.
Unfortunately, this restriction considerably reduced the usefulness of the notion
of multiplicity. Using more powerful algebraic devices, A. Weil could define an
intersection multiplicity i(C, V- W; U) when it is only supposed that C is proper
(the other components of V r) W can have larger dimensions); furthermore, he
showed that this number did not depend on the method used to define it (other,
quite different methods, were later given by Chevalley and Samuel), once it possessed
the "natural" properties similar to those of the intersection number in algebraic
topology; this he showed to be the case for his definition, and it enabled him to
develop in abstract algebraic geometry a calculus of "cycles" patterned on the
calculus of chains introduced by Poincare (irreducible subvarieties replacing
simplices). In this context, divisors on an irreducible variety of dimension n were
the cycles of dimension n - I (one also says that they have codimension 1).
Weil then went on to break away, for the first time, from projective algebraic
geometry: for his purposes (see below) he needed constructions
of algebraic varieties similar to the "gluing together" constructions
of manifolds in algebraic topology or differential geometry, which Theme E
had been familiar since the beginning of the century; he showed
that this could be done by using as "transition functions" biregular
mappings of complements of subvarieties in affine varieties (the Zariski topology
was not yet in use at that time), and he could also define in this context the notion
of "complete variety" which is the counterpart of the concept of compact space in
"abstract" algebraic geometry (in classical projective geometry, all algebraic sub-
varieties are complete).
VII c: Zeta functions and correspondences. A. Weil's work was chiefly motivated
by problems which had arisen in the early 1920's in number theory. In his thesis
of 1923, E. Artin had observed that algebraic congruences modulo a prime p, in
2 variables, i.e., of the form F(x, y) 0 (mod p), where F is a polynomial with
integral coefficients, could be interpreted as algebraic equations over the prime
field Fp = Z /p Z (and similarly the "higher congruences" in the sense of Dede
were algebraic equations over an arbitrary finite field Fq (q = pd)). He further noti
that the analogy, already exploited by Dedekind and Weber, of finite extensions of
the field C(X) with algebraic number fields, was here much closer, since the residual
fields of the valuations of a finite extension K of Fq(X) are finite fields (extension
of Fq) just as for number fields (whereas they are equal to C in classical algebraic
geometry). This enabled him to define, in complete analogy with the Riemann-
Dedekind zeta function of an algebraic number field, the zeta function of K, and
to extend to it the classical theory: functional equation and the location of the poles.
However, his treatment was entirely algebraic, without any kind of geometric inter-
pretation; a little later, F. K. Schmidt observed that a much simpler and more
natural treatment was achieved if one completely modeled the theory after Dedekind
and Weber, by introducing divisors (or "points of the abstract Riemann surface")
instead of ideals; it can then easily be shown that the zeta function can be defined by
the equation (for u = q')
-dd(logZ(u))
0 = I Nmum, Z(O) = ,
du ~~m=1
where Nm is the number of points of the curve whose coordinates belong to the
extension F,. of Fq of degree m. It turns out that this function is much simpler th
in the classical case; in fact it is a rational function
Z(u) = P2g(u)/(1 - u)(1 - qu),
where P2g is a polynomial of degree 2g (g being the genus of K). F. K. Schmidt

further discovered the remarkable fact that the functional equation
Z(l/qu) = q 1 -gu2-2gZ(u)
was nothing else but the analytic expression of the Riemann-Roch theorem!
At the same time, arithmeticians had been endeavoring to obtain an evaluation
of N1, the number of points of the nonsingular curve F corresponding to K with
coordinates in Fq, and had obtained estimates of the form N1- (q + 1) ? Cq,
with C independent of q and 1/2 < a < 1; they had observed that a = 1/2 would be
the best possible result. Hasse became interested in the problem and remarked that
the result was a consequence of the so-called "Riemann hypothesis for curves over
finite fields," namely the fact that all the zeroes of the polynomial P2g lay on t
circle I u j = ql/2 this fact implying the inequality
(7) IN,-(q + 1)| < 2g * q1/2

in an elementary way. In 1934, he succeeded in proving this result for g = 1, by
adapting to the case of finite fields ideas from the theory of complex multiplication
of elliptic functions. He and Deuring observed furthermore that an extension to
values g _ 2 would have to be based on the theory of correspondences.
This is what A. Weil proceeded to do. An irreducible correspondence between

two irreducible curves F1, F2 is an irreducible curve on the surface
Fl x F2, and in general a correspondence between F, and F2 is
a divisor onrl X 2; degenerate correspondences are those of Theme B
the form {x1} x F2 or F1 x {x2} (xi E Fi) and linear combinations
of such with integral coefficients; correspondences are called
equivalent if they differ by the sum of a principal divisor and a degenerate corre-
spondence. For 17 = r2 = 1, one defines as in set theory the composition X o Y
of tWo correspondences; it can be proved that, together with the addition of divisors,
this defines on the set of equivalence classes 9(F) a structure of ring with unit element
(the class of the diagonal A of 1 x F). The degrees d(X) and d'(X) of a corre-
spondence are defined as the integers, such that the first (resp. second) projection
of X is the cycle d(X) 17 (resp. d'(X) 1 7); on the other hand, for two correspon-
dences X, Y which intersect properly, I(X Y) is the degree of the cycle X Y.
One can then show that the integer
S(X) = d(X) + d'(X) - I(X * A)
only depends on the equivalence class 4 of X, and has the property of a tr

S(4 - l) = S(i ) for two elements of W. Furthermore, to each correspondence
X is associated another one X', deduced from X by the symmetry automorphism of
F x 1; if (, (' are the classes of X and X', one has S(Q d') _ 0, equality being only
possible for 4 = 0 in 9. This theory was first developed in 1885 by Hurwitz, using
Riemann's theory of abelian integrals, and the inequality for the trace was obtained
by Castelnuovo (of course for the classical case); using his theory of intersection
multiplicities, A. Weil was able to extend all these results to curves over arbitrary fields.
He then observed that in the Hasse problem, the number Nm was exactly I(Fm * A),
where F is the "Frobenius correspondence" which to each point of F associates its
transform by the automorphism of 1 corresponding to the automorphism t+-+ tq of
the algebraic closure of Fq; from which it follows by definition that S(Fm)
= 1 + qm - Nm, and expressing the inequality S(Q * (') ? 0 where 4 is the class of
a * A + b Fm, for arbitrary integers a, b, one gets INm - q- I f <2g . q m/2
which generalizes (7) and implies the "Riemann hypothesis."
VII d: Equivalence of divisors and abelian varieties. The introduction of varieties

of arbitrary dimension had been particularly useful because it allowed to consider as
points in a projective space of sufficiently high dimension geometric objects such as
lines, conics, etc. In 1937, Chow and van der Waerden showed quite generally that
it is possible to consider the irreducible algebraic subvarieties of given dimension
and degree in a given PA(k) as the points of some algebraic variety in a suitable
PN(k). From this result it follows that it is possible to give a precise meaning (for an
arbitrary field k) to the concepts of "specialization of cycles" and of "algebraic
family of cycles" which had been used in the classical case by the Italian school.
In particular, one can define the concept of algebraic equivalence of two divisors
D1, D2 on a nonsingular variety V as meaning that they belong to a common ir-
reducible algebraic family of divisors. Another concept of equivalence is numerical
equivalence, meaning that for any curve C on V, the intersection numbers (D1 C)
and (D2 * C) are equal. If one denotes by G, G, Ga, GI the group of divisors on V
and its subgroups formed of divisors equivalent to 0 for numerical, algebraic and
linear equivalence, one has G D Gna Ga D Gl. Severi for the classical case, and
Matsusaka for arbitrary characteristic proved that the group Gn/Ga is finite. A deeper
result, proved by Severi for complex algebraic surfaces, following earlier results of
Picard, is that the group GIG, is a free finitely generated commutative group
Zp; this result was extended by Neron for arbitrary fields and in any dimension.
Finally, it was known since Riemann that for an irreducible algebraic curve over C,
the group GaIG, was naturally endowed with a structure of g-dimensional algebraic
nonsingular variety (g being the genus of the curve) which, as a topological group,
is isomorphic to a complex torus C9/1, where F is a lattice in Cg (discrete group
isomorphic to Z 2g); this variety is called the Jacobian of the curve, and it had been
used since Clebsch to study the geometry on an algebraic curve. In general, a complex
torus Cn/F, where r is a lattice in Cn (isomorphic to Z2") can only be given the
structure of an algebraic variety if the lattice F satisfies certain bilinear relations
which had been already found by Riemann; it is then called an abelian variety.
The work of Picard and his successors proved that for an arbitrary nonsingular
algebraic variety V over C, the group Ga/Gi was again equipped with a structure of
abelian variety, called the Picard variety of V. Following his work on the Riemann
hypothesis, A. Weil developed the general theory of abelian varieties over an arbitrary
field (as "abstract" varieties), and was able to define the Jacobian of a curve. Later
work of Chow and Matsusaka proved that abelian varieties can still be imbedded in
projective space in the general case, and extended to any field the definition of the
Picard variety.
VIII. SEVENTH PERIOD: "SHEAVES AND SCHEMES"

(1950- )
After 1945, the considerable progress brought in algebraic

topology and the theory of complex manifolds by the introduction of sheaves and
spectral sequences (both due to J. Leray) completely renewed the concepts and
methods of algebraic geometry, both "classical" and "abstract," simplifying old
definitions and results and opening new ways leading to the solution of old problems.
VIII a: The Riemann-Roch theorem for higher dimensional varieties and sheaf
cohomology. The Riemann-Roch problem for an irreducible algebraic variety V is
the computation of the dimension l(D) of the vector space L(D) for an arbitrary
divisor D on V by some formula similar to the Riemann-Roch theorem for curves (3).
The Italian geometers had attacked the problem for surfaces, but succeeded only in
getting a lower bound for l(D), expressed in terms of deg(D) and birational in-
variants of the surface S, of D and of A - D (where A is a canonical divisor).
In the 1930's, study of differential geometry and in particular
of E. Cartan's method of moving frames had finally led to the
definition of vector bundles over a differential manifold M: such Theme F
a bundle is a differential manifold E with a projection p: E -+ M
such that the fibers p -(x) for any x E M are real (resp. complex)
vector spaces of fixed dimension r (the rank of E), and locally on M, E looks like
the product of M and Rr (resp. Cr); in other words each point of M has an open
neighborhood U for which there is a diffeomorphism 0 transforming p-'(U)
U x R' (resp. U x C') in such a way that 0 transforms linearly each fiber p-
into {x} x R' (resp. {x} x Cr). A section of E is a differentiable mapping s: xi-?s(x)
of M into E such that s(x) E p '(x) for every x E M. Over a complex manifold M,
one can similarly define holomorphic vector bundles by taking E as a complex
manifold, the projection p being holomorphic, the fibers p'-(x) complex vector
spaces, and 0 (in the above definition) being also holomorphic. Important e
of vector bundles are the tangent bundle T(M), where the fiber p-'(x) consists of
the tangent vectors to M at x (so that the rank is dim (M)), and the bundle of p-
covectors on M, whose sections are the exterior differential p-forms on M (see VII a).
The concept of divisor can be generalized to arbitrary complex manifolds M:
if (Ua) is an open covering of M, one considers in each Ua a meromorphic function
ha, such that in Ua n U,8, h./ha is holomorphic and # 0 everywhere; two such
systems (h), (h' ) corresponding to coverings (U,), (U ) are identified if he/hh is
holomorphic and # 0 in Ua n U for any pair (cx, A) of indices, and these classes
of systems (ha) are called divisors on M. One sees that for projective algebraic varieties
over C, this notion coincides with the old one: for instance, if M = Pn(C), and
D = Sk mkSk is a divisor on M, where each Sk is an irreducible hypersurface defined
by an equation Fk(Xo X1, X, = 0, Fk being an irreducible homogeneous poly-
nomial of degree dk, one covers Pn(C) with the n + 1 open sets Uj (0 ? j _ n),
Ui being defined by the relation xi # 0; one can then take as meromorphic function
hi in Uj the function
X + X:d [1 (Fk(XO, Xn))k

k
with d = Xk mkdk. In 1950, A. Weil observed that to a divisor D on a complex mani-

fold M was naturally attached a complex vector bundle of rank 1 (what one calls a
line bundle) B(D): with the previous notations, one "glues together" the complex
manifolds Ua x C by taking as "transition function" from Ua to U.8 the func

(x, z) t (x, (hf(x)/h,(x))z), holomorphic in (Ua n U.) x C. Furthermore, if
holomorphic section of B(D), the restrictions sx of s to Ux are such that in Ur
one has spl = (h,/hx)sx, hence there is a meromorphic function f on M such
the restriction of f to Ua is sajha for each o; for an algebraic variety M this is equivalent
to (f) + D > 0, and therefore L(D) can be interpreted as the vector space r(B(D))
of all holomorphic sections of the line bundle B(D). For instance, if M =Pn(C)
and D = H, a hyperplane in PJ(C), the transition functions for B(H) are
(X, z) i+X LkZ)
in U. r) Uk (with the notations introduced above), and F(B(H)) is the vector space
of all linear forms (x0, , xn) + 20x0 + + )nxn in Cn,
Now to each complex vector bundle E over a differential manifold M of dimension
n are attached, for each even integer 2j < n, well determined elements cj(E) of the
cohomology group H2j(M, Z) called the Chern classes of E*; when M is a complex
manifold of real dimension 2n, the Chern classes of T(M) are simply written c;
(1 < j < n) and called the Chern classes of M; the number <c, M> (where M is
considered as 2n-cycle) is the Euler-Poincare characteristic
2n
x(M)-= E ( - )jRj,
j =O
Using the interpretation of divisors by line bundles and Hodge's theory of

harmonic forms, Kodaira was able in 1951 to obtain, for compact kahlerian manifolds
of complex dimension 2, a "Riemann-Roch formula" in which the missing terms
from the formula found by the Italian geometers were expressed by means of Chern
classes; in 1952 he found a similar formula for kahlerian manifolds of dimension 3.
Meanwhile, H. Cartan and Serre had discovered that Leray's concept of sheaf
led to a remarkably simple and suggestive expression of the main results of the
theory of complex manifolds. The holomorphic functions in open sets of such a
manifold M satisfy Leray's axioms: if (9(U) is the set of the complex functions
holomorphic in the open set U a M, then, for every open covering (Va) of U, a
function f e 6)(U) is entirely determined by its restrictions f I ea E &(V9), and co
versely, given for each a an fc E (9(VY) such that fa and f6 have the same restriction
to Y,,a n Y. for all pairs (x, fl), there exists an fc ((U) such that f I Vfx = f, for all a.
* One can define the concept of direct sum of vector bundles over M by defining it locally i
an obvious way; for any differentiable map f: M' -+ M, one defines the "pullback" f*(E) of a vect
bundle E over M as the submanifold of the product M' x E consisting of the pairs (x', z) such t
f(x ) = p(z). The Chern classes of E can then be characterized by the following conditions, where o
writes c(E) for the sum ljo ( cj (E) (the sum is finite since the groups H2j, (M) are 0 for 2j > dimM
one writes by convention co (E) 1): (i) c(f*(E)) -f* (c (E)), where on the right hand sidef*:
H* (M, Z) -- H* (M', Z) is the natural mapping deduced fromf: M'-+ M.
(ii) c(E1 3 E 2 D * (DE.) = c(El) c (E2) ... c(Em) for any direct sum of vector bundles Ej over
Ml (product taken in the cohomology ring H* (M, Z) ).
(iii) c (B (H)) = ? + hn for a hyperplane H a Pn (C), hn E H2 (Pn (C), Z) being the coho
mology class orresponding to the homology class of the (2n-2)-cycle H by Poincare duality.
The sheaf thus defined is called the structural sheaf of M and written CM; one
writes H?(U, (9M) instead of ((U). More generally, for any complex vector bundle
E over M, one defines the sheaf 6(E) by replacing ((U) by the set of sections JT(U, E)
of E above U, written H0(U, O(E)); in particular one writes Qx the sheaf corre-
sponding to the complex bundle of p-covectors on M, so that H?(U, QP) is the set of
holomorphic exterior differential p-forms on U; for a divisor D on M, one writes
(x(D) instead of ((B(D)).
There are many types of sheaves other than those derived from vector bundles,
and the usefulness of sheaves derives from this versatility and from the many opera-
tions one can do with sheaves. In the first place, to a sheaf of groups 5 over M and
to each point x eM is associated a group, the stalk -F of 5 at x: for (9(E), (9(E),
consists of the equivalence classes of sections of E over neighborhoods of x for the
following relation: two sections are equivalent if they coincide on a neighborhood
of x ("germs of sections"); the general definition of jx is similar. For a sheaf of
abelian groups C and a sheaf X c 9 such that X., is a subgroup of W. for eac
one can then define a quotient sheaf /IX such that (W/X)A) = / Each s
(Cx)x (written C9,) is a local ring, and if J, 9 are two sheaves such
that Fx and gx are (9x-modules, then one can define a sheaf
SQ3 9 such that (Y?F )x = FxD6, x; one has (9x(D + D') = Theme G
(x(D) ? (9,(D') for divisors D, D'. The chief interest of sheaf theory
is that sheaves of groups may be used to replace the coefficients
in cohomology groups by "local coefficients" varying with x E M. The cohomology
groups Hj(M, ~) which one thus defines for each integer j _ 1 (one also writes
Hj(g)) have the fundamental property that for any exact sequence of sheaves of
abelian groups 0 -*X -> -* /1' -> *0, one has a "long exact sequence"
(8) 0 -* H0(X)-*H0(9) -* H0(94A) - H'((X) -* H1(9) -* H1(91$") -H 2(X)_

Once these new tools were introduced in analysis it was soon recognized that the
invariants introduced by the Italian school and by Hodge were easily expressed by
sheaf cohomology. In the first place, if M is a compact connected kaihlerian variety
of dimension n, Dolbeault and Serre proved that the corresponding space H"' of
harmonic forms of type (r, s) (see VII-a) is isomorphic to HS(f4 ); furthermore,
for any divisor D on M, Serre discovered that there is a natural duality pairing the
spaces
Hj((9M(D)) and H M (? CM( - D)) = Hn ((9m(M( - D))
"explaining" the intervention of the canonical divisor A in Riemann-Roch's theorem

(3) (one has written QM = (9M(A)). By definition, the geometric genus of M can be
written
(9) Pg = dim(H0(Rn)) and also pg = dim(H'(CM))
by the isomorphism of Hr'S an

exterior forms of all degrees <
(10) Pa = dim H ((QM) - dim H n-((9M) + + (- 1) n ~'dim H'((m)
and the plurigenera are given by
(11) Pk = dimHO((M(kA)).
In 1937, Eger and Todd introduced, on an algebraic nonsingular projective

variety M of complex dimension n, "canonical" equivalence classes of algebraic
cycles of dimension n -j, which later were recognized to correspond exactly via
Poincare duality, to the Chern classes cj of M; furthermore, Todd discovered
that the arithmetic genus of M could be computed by the formula
(12) (-l)npa + 1 =<Tn(c1, ,cn),M>,
where Tn is a polynomial with rational coefficients

the following device: in the power series
n j
,l 1-exp(y3z)
one considers the coefficient of zn, which is a symmetric polynomial in the variables
yj, and one expresses it in terms of the elementary symmetric functions of the vj; then
one replaces each elementary symmetric function aj by cj. For instance, the first
three Todd polynomials are
T1(c1) c1/2, T2(C1, C2) =(C2 + c)/12,
T3(C1,C2,C3) = c2c,/24.
In 1954, Hirzebruch generalized both Todd's result and the Riemann-Roch

formulas of Kodaira by proving that for any divisor D on M, the expression
dim HO((M(D)) - dim H'((9M(D)) + * * + ( - 1) dimHn((QM(D))
could be expressed as KP(f. cl, I .9Cn), M>, where f is the first Chern class of the
bundle B(D), and P a polynomial which is obtained by the same device as above,
starting from the power series
ef z I1 jZ
efz 1 -exp(7jz)
It was later recognized that in fact, Hirzebruch's formula was a particular case
of a much more general theorem valid for all differential manifolds, the Atiyah-Singer
index formula.
The Hirzebruch formula enables one to solve the Riemann-Roch problem when
all cohomology groups Hj((QM(D)) are reduced to 0 for j > 1. Kodaira found suf-
ficient conditions for this fact to hold; for instance, it is true when one replaces D
by D + mH where H is the intersection of M and a hyperplane (in the projective
space where M is imbedded) and m > 0 is large enough. He has also obtained a
fundamental criterion for a compact kahlerian manifold M to be isomorphic to a
projective algebraic variety: there must exist on M a kahlerian metric such that the
cohomology class of the form Q (equation (6)) in H2(M, R) belongs to H2(M,Q).
Vm b: The Serre varieties. In 1942, Zariski began a deep study of singu-

larities of projective algebraic varieties over any field, in view
of proving a desingularization theorem (which he succeeded to
do for dimension ? 3 and over a field of characteristic 0); for Theme C
that purpose, he used for the first time the general theory of
valuations*, developed 10 years earlier by Krull. In the course
of this work, he introduced the generalization of the "abstract Riemann surface" of
Dedekind-Weber for an arbitrary field K of algebraic functions over a field k, de-
fining it to be the set V of all valuations of K which vanish on k*; but in addition,
using ideas introduced a few years earlier by M. Stone, he defined on V (by purely
algebraic considerations) a topology for which V became quasi-compact, although
that topology is not Hausdorff in general: for instance, in the case of dimension 1,
considered by Dedekind-Weber, the closed sets are V and all the finite subsets of V.
By 1950 A. Weil observed that this "Zariski topology" could be defined on his
"abstract varieties" (see VII-b); not only did it appreciably improve the exposition
of the theory by allowing one to use a "geometric" language, but it also made
possible a definition of vector bundles modeled on the classical one, and to extend
to abstract varieties the relations between divisors and line
bundles (see VIII-a). Going one step further. Serre, in 1955, had
the idea to transfer in the same way the theory of sheaves to Theme G
abstract varieties, using the Zariski topology instead of the usual
one in Leray's definition. At the same time, he observed that the
concept of sheaf made possible a much simpler definition of "abstract varieties,"
using the general idea of "ringed space" of H. Cartan, i.e., a topological space X
on which is given a sheaf of rings Ox9; the advantage of this kind of structure is that
it lends itself very easily to "gluing" ringed spaces along open subsets, the verification
of the conditions of compatibility being usually trivial. In Serre's case the "pieces"
which are glued together are affine varieties over an algebraically closed field k of
* The only difference between the definition of a general valuation and the definition of a
discrete valuation (see VI-a) is that the valuation may take its value in an arbitrary totally ordered
group. For instance, the group Z x Z may be totally ordered by writing (m, n) < (m', n') if either
m < m', or m=m', and n < n' ("lexicographic ordering"); one may then define on C(X, Y) a valua-
tion with value in that totally ordered group by taking for w(P), where P is a polynomial $ 0, the
smallest (m, n) in Z x Z for which the term in Xm Yn in P has a nonzero coefficient.
arbitrary characteristic: such a variety X is a (Zariski) closed set of some kn (i.e.,

defined by polynomial equations), and Ox is the sheaf of rings such that for each
open set U c X, (C(U) = H?(U, OX) consists of the restrictions to U of the rational
functions P(X)/Q(X) on kn which are defined (i.e., Q(x) # 0) at every x E U. Of
course cohomology groups Hj(F) can still be defined when Y is a sheaf of modules
over the rings (C)x; they are vector spaces over k and Serre computed the groups
Hi(CM(mH)) for M = Pn(k) and H a hyperplane (m e Z); he also extended to ar-
bitrary fields and to projective varieties his duality theorem; but when k has charac-
teristic p > 0, most of the results obtained in the classical case by the methods of
Lefschetz and Hodge fail to generalize: for instance, the dimension of Hr(Wx) and
of H(.Qx) for a projective variety X are not necessarily equal. Nevertheless,
Grothendieck and Washnitzer were able independently to extend Hirzebruch's
formula to fields k of arbitrary characteristic, and Grothendieck, by the intro-
duction of his "K-theory," gave a far reaching generalization of that formula.
Finally, when k is the complex field, Serre showed that the cohomology groups
obtained by using the Zariski topology coincided with the classical ones.
Being chiefly interested in cohomology, Serre did not dwell at length on the general
properties of his varieties; these were investigated in detail by Chevalley almost
simultaneously (in a different language, which we do not repro-
duce here). One of the points which should be emphasized
is that with Serre and still more with Chevalley, birational geom- Theme B
etry fades out of the picture and the concept of morphism comes
to the fore. Until then, the center of interest was the theory of
complete varieties, and it is only seldom that a correspondence between two such
varieties X, Y, even if it assigns only one point of Y to a point of X (a (1, n)-corre-
spondence in classical language), is defined at every point of X. A morphism f: X -Y,
where X and Y are Serre varieties, is on the contrary a mapping of X into Y, which
is continuous for the Zariski topologies and such that for every point x e X and
every affine neighborhood V of y = f(x), there is an affine neighborhood U of x
such that f(U) c V and, for every function s e H0(V, (Cy), the function x + s(f(x))
defined in U, belongs to H0(U, Ox). The main results of Chevalley are general
theorems on morphisms and studies of special types of morphisms using results of
commutative algebra going back to E. Noether and Krull. It had been known for
a long time that the image f(X) of X by a morphismf: X -+ Y was not even locally
closed in Y in general; Chevalley showed however that when X is irreducible, f(X)
always contains a set which is open and dense in the subspace f(X) of Y. Another
of Chevalley's results is that if X and Y are irreducible, and for each x e X one
writes e(x) the maximum of the dimensions of the irreducible components of
f-'(f(x)) which contain x, then the mapping x e(x) is upper semi-continuous
in X (in other words, when x' is close enough to x, e(x') is never < e(x)).
Chevalley also showed how important concepts introduced by Zariski in the
1940's, and which A. Weil had already used in his theory of abstract varieties, led to
very suggestive theorems on morphisms. For projective varieties, Zariski had ob-
served that the "regularity" properties of a point x e X were linked very closely
to the structure of the local ring C., of the variety X at that point: x only be
to one irreducible component if C. has no zero divisors, and x is simple if (O,
regular local ring (i.e., C. is an integral domain whose field of fractions has a t
cendence degree over the base field k (always assumed to be algebraically closed)
equal to the dimension over k of the vector space m,/m2, where m, is the m
ideal of Ox). A property, of which Zariski was the first to grasp the geometric s
ficance, is the fact for Ox to be integrally closed in its field of fractions, in wh
case x is said to be normal. Zariski showed that simple (or normal) points of an
irreducible variety formed an open dense set, and that the complement of the set
of normal points has codimension at least 2. Furthermore, Zariski defined for
each projective irreducible variety X its "normalization;" this can easily be extended
to Serre varieties: for any finite extension L of the field of rational functions K of X,
there is a variety X' and a morphism p: X' -+ X such that for each affine open set
U of X, p '(U) is an affine open set of X' and the ring H0(p '(U), O9x) is the integral
closure in L of the ring H0(U, Ox); X' is called the normalization of X in L, and
simply the normalization of X if L = K. The normalization of X is of course bi-
rationally equivalent to X, and its singular points form a subvariety of codimension
> 2; in particular, if X is a curve, X' has no singular points, and this is the simplest
"desingularization" of a curve (valid in every characteristic).
The climax of Zariski's investigations on normal varieties had been his "main
theorem" expressed in the language of birational correspondences; Chevalley
showed that it implies a far more intuitive result about morphisms: suppose X
and Y are irreducible and normal varieties, f: X -+ Y is a morphism such that
f(X) is dense in Y and each set f - 1(y) is finite for y e Y. Then f factorizes in
X 1 Y' P Y where Y' is the normalization of Y in the field of rational functions
of X, and g is an isomorphism of X onto an open subvariety of Y'.
Finally, Chevalley defined the notion of complete variety in a much simpler
way than before: X is complete if, for every variety Y, the second projection
X x Y-+ Y is a closed mapping.
The interest of Chevalley in such theorems was spurred by the theory of algebraic
groups, which he and A. Borel brought to a high level of development during the
1950's; in that theory, both affine and complete varieties play an important part
and the preceding theorems are powerful tools.
VIII c: Schemes and topologies. Until the 1950's, no one seems to have tried to
give an intrinsic definition of an affine variety over an algebraically closed field k,
independent of any imbedding of the variety in some "affine space" kn, although
the tools to do so were available since the 1890's. In his work on invariant theory,
Hilbert had proved his famous "Nullstellensatz," one of the forms of which is
that the maximal ideals of the algebra of polynomials k[X1, ,Xn] are in one-t
one correspondence with th

sponding to the ideal generated by the polynomials Xi- l, Xn- Cn. Just as
Riemann attached to a projective curve the field of rational functions on that curve,
so one may attach to an affine variety V c kn the ring R(V) of the restrictions to V
of all polynomial functions on kn; this ring is a finitely generated algebra over k,
which has no nilpotent elements (one says it is reduced); and by Hilbert's Null-
stellensatz, the points of V are in one-to-one correspondence with the maximal
ideals of R(V). Conversely, it is readily seen that any reduced and finitely generated
k-algebra has the form R(V) for an affine variety determined up to isomorphism.
Furthermore, when V is irreducible, it is even possible to define the sheaf (Ov directly
from the ring R(V): for any open (Zariski) subset U of V which is defined as the set
of points x such that f(x) : 0 for some f e R(V), one defines ((U) as the ring of
rational functions of type g/ftm for g e R(V) and m a positive integer, and it is easy
to see that this defines completely OX. Finally, if V, W are two affine varieties over
we have seen above that to a morphism f: V -+ W corresponds a k-algebra homo
morphism R(f): R(W) -+ R(V); but the converse is also true, for Hilbert's Null-
stellensatz implies that for any such homomorphism 4: R(W) - R(V), the inver
image 4) `(m) of a maximal ideal of R(V) is again a maximal ideal in R(W), and
m i-* J `(m) is the morphism corresponding to 0. In the language of categorie
which was beginning to be used in the late 1950's, the category of affine varieties
over k was equivalent to the dual of the category of reduced finitely generated
(commutative) k-algebras.
Following a suggestion of Cartier, A. Grothendieck undertook around 1957 a
gigantic program aiming at a vast generalization of algebraic geometry, absorbing
all previous developments and starting from the category of all commutative rings
(with unit) instead of reduced finitely generated algebras over an algebraically
closed field. If one wanted to define a category which would be equivalent to the dual
of the category of all commutative rings, a nontrivial modification was needed from
the start, since if 4): A -+ B is a homomorphism of rings (sending unit element on
unit element), the inverse image 0 1'(m) of a maximal ideal of B is not in general
a maximal ideal of A, whereas the inverse image 04 1 (3) of a prime ideal of B is
always a prime ideal of A. It was thus necessary to take as the set replacing the
affine variety the spectrum of A, i.e., the set Spec(A) of all prime ideals of A; closed
sets in Spec(A) are defined as sets of prime ideals containing a given (arbitrary)
ideal of A, hence a "Zariski topology" for which, however, finite sets are no longer
closed in general; finally, using work of Chevalley and Uzkov on localization dating
from the 1940's, it is possible to give a meaning to g/fln even when f is a zero-
divisor of A, hence to define the sheaf (9x on X = Spec (A) in the same way as for
afiine varieties. The ringed spaces thus obtained are called affine schemes and they
form a category equivalent to the dual of the category of all commutative rings;
finally, the usual "gluing process" for ringed spaces yields the category of schemes
by replacing affine varieties by affine schemes.
The experience of the last 10 years has convinced the specialists that, in spite
of the much greater amount of commutative algebra techniques which it requires,
the theory of schemes is the context in which the problems of algebraic geometry
are best understood and attacked. Among the features which distinguish it from
previous conceptual frames for algebraic geometry, let us mention only the few
following ones:
(1) The notion of generic point, which had disappeared from the Serre-Chevalley
theory, is now reintroduced in a natural way: for instance, if A is an integral domain,
its (unique) generic point is the prime ideal (0) in Spec(A); its "generic" property is
expressed by the fact that its closure is the whole space Spec(A), and thus con-
tinuity arguments in the Italian style (but in the Zariski topology!) are now again
available.
(2) The predominance of "relative" versus "absolute" notions, or, put in a
different way, the fact that most of the times what is studied is not a scheme but a
morphism of schemes f: X -+ S, where S is often quite arbitrary (one also says
that the study of such morphisms, for fixed S, is the study of "S-schemes"). This is
particularly apparent when it comes to imposing finiteness conditions (without any
such condition, there is very little likelihood of ever getting any deep result):
Grothendieck has shown that, except for cohomological notions, one may usually
allow the "base scheme" S to be free from finiteness assumptions (such as being
noetherian, or of finite dimension, etc.), and the results only depend on finiteness
conditions for the morphism f; this allows considerable freedom in the "change
of bases" (see below).
(3) Given two "S-schemes" f: X -+ S, g: Y -+ S, there is an essentially unique
triplet consisting in an S-scheme X x sY and two morphisms Pt: X x sY -X,
P2: X x sY -? Y such that f o Pt = g ? P2' which is the "categorical" product of
X and Y over S: this means that, given two morphisms u: Z -+ X, v: Z Y such
that f o u = g o v, there is a unique morphism w: Z-+ X x sY such that u Pi o w
and v = P2 a w (there is no similar result for Serre varieties; it easily follows from
the existence of the tensor product B0 A C of arbitrary A-algebras, where A is any
ring).
Most of the time this fundamental process is applied to study the morphism
f: X -+ S by replacing the "base" S by another one Y, in such a way that the new
morphism P2, which is now written f(y): X(y) -+ Y (the notation X(y) replacing
X x sY) can be more easily handled. This "change of base" is probably the most
powerful tool in the theory of schemes, generalizing in a bewildering variety of
ways the old idea of "extending the scalars." To give only one example, consider
at any point s E S the residual field k(s) = -s/ms of the local ring Cs at that p
then Xs = X x s Spec(k(s)) has as underlying space the "fiber" f -1(s) in X and
(provided f satisfies finiteness conditions) it can be considered as a "variety" over
the field k(s) (in a slightly more general sense than with Serre). In this way, an S-
scheme X may be considered as a "family of varieties" Xs parametrized by S
(generalizing the old Picard method (see VI-c)) and many properties of S-schemes
may be obtained by a study of the fibers Xs.
(4) It may seem strange at first that one should consider affine schemes
Spec(A) even when A has nilpotent elements other than 0; but in fact, this also
corresponds to geometric facts which were not taken into account by older
theories. For instance, consider the parabola y2 _ x = 0 in C2 and the mapping
which projects it on the x-axis; in the language of schemes, we consider the affine
schemes U = Spec(C[X, y]/(y2 _ X)), V = Spec(C[X]) and the morphism
p: U -+ V which corresponds to the natural injection C[X] _+ C[X, y]/(y2 _
which sends X onto the class of X. A maximal ideal (X - C) in C[X] is identified w
the point CE C, and the fiber V, = p-1(C) is the affine scheme Spec (C[Y]/(Y
now, if C $ 0, the ring C[Y]/(y2 _ C) is isomorphic to the direct sum of two
isomorphic to C, corresponding to the fact that the fiber has two distinct points;
but if C = 0, C[Y]/(Y2) has nilpotent elements: the two points have become "in
finitely near" one another. It turns out that this is a general phenomenon: nilpotent
elements in the local rings of a scheme are the algebraic counterpart of "infinitesimal"
properties, and their presence allows a much more natural and flexible treatment
of these properties than in classical algebraic geometry (see e.g. [8]).
(5) If we return to the concept of affine Serre variety, corresponding to a reduced
finitely generated algebra A over an algebraically closed field k, the points of the
variety are not all points of Spec(A), but only the closed ones, corresponding to all
homomorphisms A -+ k which are k-homomorphisms, i.e., such that the compositio
with the natural mapping k -+ A gives the identity on k; similarly, if one wants to
consider the points of variety "with coordinates in a field K extension of k" (see
VI1-b), one has to consider homomorphisms A -+ K which by composition k -+ A -+
give the homomorphism defining the extension K of k. This idea has been greatly
generalized by Grothendieck: for an S-scheme X -+ S the "points of X in an arbitra
S-scheme T" (or more briefly the "T-points" of X) are by definition the morphisms
T -+ X which, composed with X -+ S, give the structural morphism T -+S; if we
denote by Mors(T, X) the set of these "S-morphisms," it can easily be shown that
T H-+ Mors(T, X) is a functor from the category of S-schemes to the category of sets,
and that the knowledge of that functor entirely determines the S-scheme X, which
is said to "represent" the functor. This idea has become a very fruitful principle
allowing the definition of schemes by the functor which they "represent," which is
generally much easier (provided one has general theorems establishing the "repre-
sentability" of functors); in particular, one transfers in that way to the theory of
schemes many classical constructions such as projective spaces, Grassmannians,
Chow varieties, Picard varieties, and one is able to give a general meaning to the
concept of "moduli" introduced by Riemann for curves.
(6) It was early recognized that the Zariski topology on schemes had some
unpleasant features regarding "vector bundles:" natural definitions of S-schemes
X -+ S, which in classical geometry gave vector bundles X over S, did not have in
general the property of being "locally" products of a (Zariski) neighborhood and

a "typical fiber" (one says that they are not "locally trivial" for the Zariski to-
pology). However, Serre observed that in important cases, a mild "extension of the
base" T -> S, where T is an "etale covering" of S (which corresponds in classical
geometry to an unramified covering with finitely many sheets) was enough to restore
"local triviality." Starting from this remark, Grothendieck conceived the idea of
replacing the Zariski topology on S by a new structure, called "etale topology,"
which is not any more a topology in the usual sense; essentially it consists in re-
placing the usual open subsets of S (or rather their natural injections U -*S) by
etale coverings of S (one may say that the open sets are now "out of the space"
instead of being parts of it). The important fact is that he was able to transfer to
this new concept the definition of sheaves and of sheaf cohomology, and to show
that this "etale cohomology" can partly remedy to the defects of the usual (Zariski)
sheaf cohomology for varieties over a field of characteristic p > 0.
IX. OPEN PROBLEMS
To have some idea of the dozens of problems on which algebraic geometers are
now working, one may consult for instance the various reports in [18], [19], or [20].
We will conclude by mentioning very briefly some of the most conspicuous ones.
(1) The famous problem of "desingularization" of algebraic varieties over a
field k has been solved by Hironaka in all dimensions, when k has characteristic 0,
and this result has become a very powerful tool in many problems of algebraic
geometry, both classical and "abstract." For fields of characteristic p > 0, the
problem is still open in dimensions > 3; for dimension 2, the desingularization
theorem has been proved by Abhyankar in all characteristics.
(2) The problem of Riemann's "moduli" has attracted much attention during
the last 20 years, both in classical and in abstract geometry: the general idea is to
prove the existence of a variety (or scheme) whose points would correspond to
isomorphism classes of curves of a given genus over a given field; the most compre-
hensive results to date are those of Mumford, who has proved the existence of such
a scheme; but much remains to be done regarding the properties of that scheme.
One has similar results when curves of given genus are replaced by abelian varieties
of given dimension; but already for algebraic surfaces, very little progress has been
made on similar problems. Even when one considers "local" problems, i.e., how
algebraic structures depending on parameters may "deform" in the neighborhood
of a point in the parameter space, the results are far from final.
(3) In spite of the progresses brought by "etale cohomology" (and other similar
theories based on other types of "Grothendieck topologies"), the cohomological
properties of varieties over a field of characteristic p > 0 are not yet well understood,
and nothing has yet satisfactorily replaced the abelian integrals in that case. Central
in these problems are the "Weil conjectures" which he formulated as extensions
to algebraic varieties of arbitrary dimension of his work on the zeta function of
algebraic curves over finite fields; some of them have been proved by Grothendieck
866 J. DIEUDONNE
and M. Artin, using etale cohomology, but the extension of the "Riemann hypothe-
sis" has up to now resisted all efforts.
(4) In classical algebraic geometry, the theory of integrals of "second" of "third"
kinds on projective algebraic varieties of arbitrary dimension is still incomplete,
although much advanced recently by the work of Leray, Hodge-Atiyah and Griffiths
on the concept of "residue." Generalizations of the Hodge theory to non compact
algebraic varieties (over C) with singularities have recently been started by Deligne
and others.
(5) One would expect that the precise knowledge of divisors under various
"'equivalence" concepts (see VII-d) should extend to "cycles" of arbitrary con-
dimension, but even in the classical case that theory is still in an embryonic stage.
(6) Finally, the beautiful results of Castelnuovo and Enriques on the characteri-
zation of classes of surfaces by properties of their invariants have been greatly
extended by Kodaira and Shafarevich [11], and generalized by Mumford to sur-
faces over an algebraically closed field of characteristic p > 0 [19], but much remains
to be done, and practically no comparable results have been obtained in higher
dimenslons.
References
1. M. Baldassari, Algebraic varieties, Ergeb. der Math., Heft 12, Springer, Berlin-Gottingen-
Heidelberg, 1956.
2. J. Dieudonn&, Algebraic geometry, Advances in Math., 3 (1969) 233-321.
3. , Fondements de la geometrie algebrique moderne, Advances in Math., 3 (1969) 322-
413.
4. F. Enriques, Sui sistemi continui di curve appartenenti ad una superficie algebrica, Comm.
Math. Helv., 15 (1943) 227-237.
5. F. Hirzebruch, Topological methods in algebraic geometry, Springer, Berlin-Heidelberg-
New York, 3rd ed., 1966.
6. S. Lefschetz, L'Analysis Situs et la Geometrie algebrique, Gauthier-Villars, Paris, 1924.
7. D. Mumford, Geometric invariant theory, Erg. der Math. Heft 34, Springer, Berlin-Heidel-
berg-New York, 1965.
8. , Lectures on curves on an algebraic surface, Princeton Univ. Press, Princeton, 1966.
9. , Abelian varieties, Oxford Univ. Press, Oxford, 1970.
10. F. Severi, Intorno ai sistemi continui di curve sopra una superficie algebrica, Com
Helv., 15 (1943) 238-248.
11. I. Shafarevich et al. Algebraic surfaces, Proc. Steklov Inst. of Math., Amer. Math. Soc., 1967.
12. B. L. van der Waerden, Einfuhrung in die algebraische Geometrie, Springer, Berlin, 1939.
13. A. Weil, Foundations of algebraic geometry, Amer. Math. Soc. Coll. Publ., 29 (1946).
14. , Sur les courbes algebriques et les varietes qui s'en deduisent, Hermann, Paris, 1948.
15. , Introduction a l'etude des varietes kahleriennes, Hermann, Paris, 1958.
16. 0. Zariski, Algebraic surfaces, Erg. der Math. 2nd ed., Springer, Berlin-Heidelberg-New
York, 1971.
17. , An introduction to the theory of algebraic surfaces, Lecture Notes in Math., 83,
Springer, Berlin-Heidelberg-New York, 1969.
18. Dix exposes sur la cohomologie des schemas, North-Holland, Amsterdam-London, 1968.
19. Global Analysis (Papers in honor of K. Kodaira), Princeton Univ. Press, 1969.
20. Actes du Congres international des mathematiciens, Nice, 1970, vol. I et II, Gauthier-Villars,
Paris, 1971.

Mathematical Association of America, Taylor & Francis, Ltd. The American Mathematical Monthly

Uploaded by

Copyright:

Available Formats

Mathematical Association of America, Taylor & Francis, Ltd. The American Mathematical Monthly

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mathematical Association of America, Taylor & Francis, Ltd. The American Mathematical Monthly

Uploaded by

Copyright:

Available Formats

The Historical Development of Algebraic Geometry

J. DIEUDONNE, University of Nice, France and University of Maryland

I. THEMES AND PERIODS

(F) Analysis and topology in algebraic geometry. This theme beautifully

II. FIRST PERIOD: "PREHISTORY"

(CA. 400 B.C.-1630 A. D.)

III. SECOND PERIOD: "EXPLORATION"

equation. A beginning is made in the elucidation of the structure Theme

With the beginning of the consideration of algebraic families of algebraic curves

completely determine a curve of degree n, whereas two curves of degree n have in

IV. THIRD PERIOD: "THE GOLDEN AGE OF PROJECTIVE GEOMETRY"

of the projective school was to minimize as much as possible algebraic computations,

Later representatives of the projective school (notably Chasles in France, Steiner

m' =rn(m -1)-2d -3s,

m = m'(m' - 1) - 2d' - 3s',

s-s = 3(m' - m),

V. FOURTH PERIOD: "'RIEMANN AND BIRATIONAL GEOMETRY"

The importance of Riemann in the history of algebraic geometry can hardly

where P(t) is a polynomial of degree 3 or 4 and R(t) a rational function; one of

.Ja VP(t) + V(X,y),

with arbitrary poin

* Of course, the points xj, yj usually have complex coordinates; an integral

J(a,b R(x, y)dx

inconclusive note, he has no general discussion of the periods of his integrals.

z - zo = th, s-so = alt + a2t2....

xi = 1/x, y' = y/x

VI. FIFTH PERIOD: "DEVELOPMENT AND CHAOS"

The extraordinary wealth of new ideas and methods introduced by Riemann

* An irreducible variety V in Pn(C) is characterized by the property that if the product PQ of

as a nontrivial discrete valuation on K (i.e., one which is not identically 0 on K*:

* Recall that an element x of a ring R is integral over a subring S if it satisfies an equation of

(1) v(f) > - av for all v E S

(2) (f) +D > 0

Principal divisors form a subgroup gA(K) of 9(K) (isomorphic to the group

in other words, the set ID I of positive divisors equivalent to D is identified to th

(3) I(D)-l(A-D) = deg (D) + 1 -g,

VI b: The Brill-Noether theory of linear systems of points on a curve. An irredu-

(for suitable homogeneous coordinates), the so-called quadratic transformations.

0w: C P>(Pl m, P in, I Prt+ 1hom),

where the Pj are homogeneous polynomials in the homogeneous coordinates of 4

VI c: Integrals of differential forms on higher dimensional varieties. As soon as

ff R(x(u, v), y(u, v), z(u, v) -( ' Y) dudv.

VII. SIXTH PERIOD: "NEW STRUCTURES IN ALGEBRAIC GEOMETRY"

X Aili2 .. ip(x)dx A dxi2 A" A dxip

(4) 0 d A1 d A2d dAnd0 (n - dim V),

* If M is a differential manifold of dimension n,k: U -+ Rn, g: Y -+

A projective algebraic variety without singularity of (complex) dimension n has

(5) lAil ...Jirkij... k,(x)dZ ji A*-- A dzirA dkkl A AdZk.,

ds2 = , hjkdzidik with hkl = hJk

(6) Q= (i/2) , hjkdzk A dz3

is exact (i.e., dQ = 0).

D(xy) = x * Dy + (Dx) * y, and a differential is an A-linear Theme

Z(u) = P2g(u)/(1 - u)(1 - qu),

where P2g is a polynomial of degree 2g (g being the genus of K). F. K. Schmidt

(7) IN,-(q + 1)| < 2g * q1/2

This is what A. Weil proceeded to do. An irreducible correspondence between

S(X) = d(X) + d'(X) - I(X * A)

only depends on the equivalence class 4 of X, and has the property of a tr

VII d: Equivalence of divisors and abelian varieties. The introduction of varieties

(8) 0 -* H0(X)-H0(9) - H0(94A) - H'((X) -* H1(9) -* H1(91$") -H 2(X)_