Compositio Mathematica 110: 215±237, 1998. 215
c 1998 Kluwer Academic Publishers. Printed in the Netherlands.
Algebraic cycles and topology of real
Enriques surfaces ?? FRED ERIC MANGOLTE1 ? and JOOST VAN HAMEL2 1Laboratoire de Mathematiques, Universite de Savoie, 73376 Le Bourget du Lac Cedex, France; e-mail: [email protected] 2Institute of Mathematics, University of Utrecht, Budapestlaan 6, 3584 CD Utrecht, The Netherlands; e-mail: [email protected]
Received:14 May 1996; accepted in ®nal form 22 October 1996
H Y R; Z=
Abstract. For a real Enriques surface Y we prove that every homology class in 1 2 can be
R represented by a real algebraic curve if and only if all connected components of Y are orientable.
Furthermore, we give a characterization of real Enriques surfaces which are Galois-Maximal and/or Y Z-Galois-Maximal and we determine the Brauer group of any real Enriques surface . Mathematics Subject Classi®cations (1991): 14C25 14P25 14J28 13A20. Key words: Algebraic cycles, real algebraic surfaces, Enriques surfaces, Galois-Maximality, Brauer group.
1. Introduction
Y C Let Y be a complex algebraic surface. Let us denote by the set of closed points
alg
H Y C ; Z
of Y endowed with the Euclidean topology and let 2 be the subgroup
Y C ; Z
of the homology group H2 generated by the fundamental classes of Y algebraic curves on Y .If is an Enriques surface, we have
alg
Y C ; Z=H YC;Z: H2 2 One of the goals of the present paper is to prove a similar property for real Enriques
surfaces with orientable real part. See Theorem 1.1 below. R
By an algebraic variety Y over we mean a geometrically integral, separated
= f ;g C = R
scheme of ®nite type over the real numbers. The Galois group G 1 of
C Y
acts on Y , the set of complex points of , via an antiholomorphic involution and
R
the real part Y is precisely the set of ®xed points under this action. An algebraic R
variety Y over will be called a real Enriques surface, a real K3-surface, etc., if
Y = Y C
the complexi®cation C is a complex Enriques surface, resp. a complex K3-surface, etc. Consider the following two classi®cation problems:
? The ®rst author was supported in part by HCM contract ERBCHBG-CT-920011 (MAP-Project) and in part by HCM contract ERBCHRX-CT93-0408. ?? The second author was supported in part by EC grant ERBCHRX-CT94-0506.
Downloaded from https://www.cambridge.org/core. IP address: 170.106.35.229, on 26 Sep 2021 at 16:44:06, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1000223408405
**INTERPRINT**/Preproof: A M/c 4: PIPS Nr.: 124150 MATHKAP comp4075.tex; 5/12/1997; 11:40; v.7; p.1
216 FREDÂ ERICÂ MANGOLTE AND JOOST VAN HAMEL
Y R
classi®cation of topological types of algebraic varieties over (the mani-
C
folds Y up to equivariant diffeomorphism),
Y R classi®cation of topological types of the real parts . For real Enriques surfaces the two classi®cations have been investigated recently by Nikulin in [Ni2]. The topological classi®cation of the real parts was completed by Degtyarev and Kharlamov, who give in [DKh1] a description of all 87 topological
types. Let us mention here that the real part of a real Enriques surface Y need not
Y R be connected and that a connected component V of is either a nonorientable
surface of genus 6 11 or it is homeomorphic to a sphere or to a torus.
C The problem of classifying Y up to equivariant diffeomorphism still lacks a satisfactory solution. In the attempts to solve this problem, equivariant (co)homology
plays an important role (see [Ni2], [NS], [DKh2]). It establishes for any algebraic
R G Y C
variety Y over a link between the action of on the (co)homology of and
R Y d the topology of Y . For example, if is of dimension , the famous inequalities
2n
X 1
H Y R; Z= 6 H G; H Y C ; Z= ; r
dim 2 dim 2 (1) = r 0
2 n
X 2
H Y R ; Z= 6 H G; H Y C ; Z;
dim even 2 dim r (2) = r 0
2n
X 1
H Y R ; Z= 6 H G; H Y C ; Z;
dim odd 2 dim r (3) = r 0 (cf. [Kr1] or [Si]) can be proven using equivariant homology.
We will say that Y is Galois-Maximal or a GM-variety if the ®rst inequality
Z Z
turns into equality, and Y will be called -Galois-Maximal,ora -GM-variety if
C inequalities (2) and (3) are equalities. When the homology of Y is torsion free,
the two notions coincide (see [Kr1, Prop. 3.6]).
R Y R 6= ;
A nonsingular projective surface Y over with is both GM and
H Y C ; Z 6= Z-GM if it is simply connected (see [Kr1, Sect. 5.3]). If 1 0, as in the case of an Enriques surface, the situation can be much more complicated. The
necessary and suf®cient conditions for a real Enriques surface Y to be a GM-variety
were found in [DKh2]; in the present paper we will give necessary and suf®cient Z conditions for Y to be -GM. See Theorem 1.2. As far as we know, this is the ®rst paper on real Enriques surfaces in which
equivariant (co)homology with integral coef®cients is studied instead of coef®cients = in Z 2.We expect that the extra information that can be obtained this way (compare for example equations (1)±(3)) will be useful in the topological classi®cation of real Enriques surfaces.
In Section 6 we demonstrate the usefulness of integral coef®cients by computing
Y Y
the Brauer group Br of any real Enriques surface . This completes the partial
Y results on the 2-torsion of Br obtained in [NS] and [Ni1]. See Theorem 1.3.
Downloaded from https://www.cambridge.org/core. IP address: 170.106.35.229, on 26 Sep 2021 at 16:44:06, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1000223408405
comp4075.tex; 5/12/1997; 11:40; v.7; p.2 REAL ENRIQUES SURFACES 217
1.1. MAIN RESULTS
alg
Y R H Y R; Z=
Let be an algebraic variety over . Denote by n 2 the subgroup
H Y R; Z=
of the homology group n 2 generated by the fundamental classes of
Y R n-dimensional Zariski-closed subsets of , see [BH] or [BCR]. We will say that these classes can be represented by algebraic cycles. The problem of determining these groups is still open for most classes of surfaces.
alg
H X C ; Z=H XC;Z For a real rational surface X we always have 2 2
alg
X R ; Z= =H XR;Z= and H1 2 12, see [Si]. For real K3-surfaces, the situa- tion is not so rigid. In most connected components of the moduli space of real K3-
alg
H X R ; Z= > k
surfaces the points corresponding to a surface X with dim 1 2
k 6
form a countable union of real analytic subspaces of codimension k for any
X R ; Z= X
dim H1 0 2 ,where 0 is any K3-surface corresponding to a point from
H X R ; Z= that component. In some components this is only true for k< dim 1 0 2 ;
these components do not contain any point corresponding to a surface X with
alg
X R ; Z= = H X R ; Z= H1 2 1 2 , see [Ma2]. For real Abelian surfaces the
situation is similar, see [Hu] or [Ma1, Ch.V].
Y R 6= ;
THEOREM 1.1. Let Y be a real Enriques surface with . If all connected
R components of the real part Y are orientable, then
alg
Y R; Z= =H YR;Z= : H1 2 12 Otherwise,
alg
Y R; Z= = H Y R ; Z= : dim H1 2 dim 1 2 1 See Theorem 4.4 for more details. In order to state further results we should mention that the set of connected
components of the real part of a real Enriques surface Y has a natural decomposition
R =Y tY into two parts Y 12. Following [DKh1] we will refer to these two parts
as the two halves of the real Enriques surface. In [Ni1] it is shown that Y is GM
R
if both halves of Y are nonempty. It follows from [DKh2, Lem. 6.3.4] that if
R Y Y R precisely one of the halves of Y is empty, then is GM if and only if is nonorientable. This result plays an important role in the proof of many of the main results of that paper (see Sect. 7 in loc. cit.). In the present paper we will see in the course of proving Theorem 1.1 that a real
Enriques surface with orientable real part is not a Z-GM-variety. In Section 5 we
also tackle the nonorientable case and combining our results with the results for = coef®cients in Z 2 that were already known we obtain the following theorem.
THEOREM 1.2. Let Y be a real Enriques surface with nonempty real part.
Y Y Y
(i) Suppose the two halves Y1 and 2 are nonempty. Then is GM. Moreover,
Y R is Z-GM if and only if is nonorientable.
Downloaded from https://www.cambridge.org/core. IP address: 170.106.35.229, on 26 Sep 2021 at 16:44:06, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1000223408405
comp4075.tex; 5/12/1997; 11:40; v.7; p.3
218 FREDÂ ERICÂ MANGOLTE AND JOOST VAN HAMEL
Y Y Y R
(ii) Suppose one of the halves Y1 or 2 is empty. Then is GM if and only if
Z Y R is nonorientable. Moreover, Y is -GM if and only if has at least one component of odd Euler characteristic. There are examples of all cases described in the above theorem (see [DKh1,
Figure 1]).
Y
In Section 6 we study the Brauer group Br of a real Enriques surface
Y Y using the fact that Br is isomorphic to the cohomological Brauer group
0 2
Y = H Y; G Y
Br m ,since is a nonsingular surface. In [NS] Nikulin and et
Sujatha gave various equalities and inequalities relating the dimension of the 2-
Y Y torsion of Br to other topological invariants of a real Enriques surface .Itwas
shown in [Ni1] that
; Y > s ; =
dimZ 2 Tor 2 Br 2 1
Y R where s is the number of connected components of , and that equality holds
if Y is GM. Using the results in Section 5 on equivariant homology with integral
Y
coef®cients we can compute the whole group Br . s
THEOREM 1.3. Let Y be a real Enriques surface. Let be the number of connected
R Y R 6= ;
components of Y .If is nonorientable then
2s 1
Y ' Z= :
Br 2
R 6= ;
If Y is orientable then
2s 2
Z= Z= ;
2 4 if both halves are nonempty
Y ' Br
2s
Z=
2 if one half is empty.
R =;
If Y then
Y ' Z= : Br 2
We were informed by the referee that Theorems 1.2 and 1.3 and a result similar to Theorem 1.1 have independently been obtained by V.A. Krasnov. His paper will be published in Izv. Ross. Akad. Nauk Ser. Math. 60 (1996), No. 5.
2. Equivariant homology and cohomology
= C =R
Since the group G Gal acts in a natural way on the complex points of an R
algebraic variety Y de®ned over , the best homology and cohomology theories for
R
studying the topology of Y are the ones that take this group action into account.
Y; Z= In [NS] etale cohomology H 2is used, and in [Ni1] the observation is made
et
Y C G; Z= that this is essentially the same as equivariant cohomology H ; 2 . In [DKh2] Degtyarev and Kharlamov do not use equivariant cohomology as
Downloaded from https://www.cambridge.org/core. IP address: 170.106.35.229, on 26 Sep 2021 at 16:44:06, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1000223408405
comp4075.tex; 5/12/1997; 11:40; v.7; p.4 REAL ENRIQUES SURFACES 219
such, but instead a `stabilized' form of the Hochschild-Serre spectral sequence q
2 p
X G; Z= = H G; H X; Z=
E ; 2 2. This construction, due to I. Kalinin, is
p;q
+ p
p 2
= Z= H G; M =H G; M
based on the fact that if G 2then for any group
+ p
p 1
p> M Z= H G; M =H G; M M and any 0, and if is a 2-module then even
for any p>0, so it is possible to squeeze the Hochschild±Serre spectral sequence into 1, or at most 2 diagonals. They also use the analogue of this Kalinin spectral sequence in homology. In the present paper we stick to the original equivariant cohomology supplemented with a straightforward dual construction which we call equivariant Borel±Moore homology.
First we will recall some properties of equivariant cohomology for a space with
= Z= an action of G 2. Then we will give the de®nition of equivariant Borel±Moore homology and list the properties that we are going to need. In Section 3 we give
a short treatment of the fundamental class of G-manifolds and formulate PoincareÂ
duality in the equivariant context.
G = Z=
Let X be a topological space with an action of 2. We denote the ®xed
G
X H X G; F G
point set of X by . In [Gr1] the groups ; are de®ned for a -sheaf
X G G X
F on , which is a sheaf with a -action compatible with the -action on .
= f ;g
Writing G 1 , this just means that we are given an isomorphism of sheaves
F ! F & & =
& : satisfying id. Now de®ne
p p G
X G; =R X;
H ; G the pth right derived functor of the -invariant global sections functor. We have
natural mappings
p
p p G
H X G; F ! H X; F ;
e : ; F
which are the edge morphisms of the Hochschild±Serre spectral sequence
q p+q
2 p
X G; F =H G; H X; F H X G; F
E ; ;
p;q
Z=
For us, the most important G-sheaves will be the constant sheaf 2andthe
Zk k 2 Z
constant sheaves constructed from the G-modules for . Here we de®ne
k G z =
Z , to be the group of integers, equipped with an action of de®ned by
k
z Ak Z= Zk
1 . We will use the notation to denote either 2or , and we will
Ak k sometimes use A instead of if is even.
There is a cup-product
p q p+q
X G; Ak H X G; Al ! H X G; Ak + l
H ; ; ;
f X ! Y and a pull-back f for any continuous equivariant mapping : , which both
have the usual properties.
p p
H G; M = H G; M ;
If X is a point, pt; which is cohomology of the
M H G; Z= group G with coef®cients in . Recall that as a graded ring, 2 is
Downloaded from https://www.cambridge.org/core. IP address: 170.106.35.229, on 26 Sep 2021 at 16:44:06, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1000223408405
comp4075.tex; 5/12/1997; 11:40; v.7; p.5
220 FREDÂ ERICÂ MANGOLTE AND JOOST VAN HAMEL
= [ ] isomorphic to the polynomial ring Z 2 ,where is the nontrivial element in
1
G; Z= H 2 . By abuse of notation, we will also use the notation for the nontrivial
1 2 2
G; Z ' Z= H G; Z ' element in H 1 2and for the nontrivial element in
1
2 H G; Z 2 = Z 2. This notation is justi®ed by the fact that 1 maps to
1 2 2
G; Z= 2 H G; Z H 2 under the reduction modulo 2 mapping and maps to
2 2
2 H G; Z=
2 .
! H G; Z= !
The constant mapping X pt induces a mapping 2
p G p G
X G; Z= H X ; Z= ,! H X G; Z=
H ; 2 and we have a natural injection 2 ; 2 , X
so cup-product gives us for any G-space a mapping
G G
X ; Z= H G; Z= ! H X G; Z= ; H 2 2 ; 2
which is well-known to be an isomorphism. Taking the inverse of this isomorphism
G
H X ; Z=
and sending to the unit element in 2 we obtain a surjective homo-
G G
X G; Z= ! H X ; Z= A = Z
morphism of rings H ; 2 2 and we de®ne for
= k 2 Z
or Z 2andany the homomorphism of rings
G
H X G; Ak ! H X ; Z= : ; 2
to be the composite mapping
i
G
-
X G; Ak H X G; Ak
H ; ;
G G
mod 2
- -
X G; Z= H X ; Z= ;
H ; 2 2
G
i X ,! X
where i is induced by the inclusion : . Note that coincides with the 0
mapping in [Kr3]. It is clear from the de®nition that
f ! =f !:
We use the notation
n;p n p G
H X G; Ak ! H X Z= ; : ; ; 2
for the mapping induced by . In Section 5, we will need one technical lemma which can easily be proven using the Hochschild±Serre spectral sequence.
G 2
G X 6= ; e
LEMMA 2.1. Let X be a -space with .Thenif is not surjective on
Ak
2 G 1 1
H X; Ak ! 2 H X G; Ak e ! 6=
,thereisaclass ; 1 such that 0,
k
A 1
! = but 0.
Downloaded from https://www.cambridge.org/core. IP address: 170.106.35.229, on 26 Sep 2021 at 16:44:06, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1000223408405
comp4075.tex; 5/12/1997; 11:40; v.7; p.6 REAL ENRIQUES SURFACES 221
The homology theory we are going to use is the natural dual to equivariant cohomology. For an extensive treatment of its properties, see [vH]. Here we will give a short account without proofs.
IntherestofthissectionweassumeX to be a locally compact space of ®nite
= Z= Ak
cohomological dimension with an action of G 2, and will be as above.
Ak We de®ne the equivariant Borel±Moore homology of X with coef®cients in
by