This action might not be possible to undo. Are you sure you want to continue?
Lectures on Algebraic Cycles
Second Edition
Spencer Bloch’s 1979 Duke lectures, a milestone in modern mathematics, have been
out of print almost since their ﬁrst publication in 1980, yet they have remained
inﬂuential and are still the best place to learn the guiding philosophy of algebraic
cycles and motives. This edition, now professionally typeset, has a new preface by the
author giving his perspective on developments in the ﬁeld over the past 30 years.
The theory of algebraic cycles encompasses such central problems in mathematics
as the Hodge conjecture and the Bloch–Kato conjecture on special values of zeta
functions. The book begins with Mumford’s example showing that the Chow group of
zerocycles on an algebraic variety can be inﬁnite dimensional, and explains how
Hodge theory and algebraic Ktheory give new insights into this and other phenomena.
spencer bloch is R. M. Hutchins Distinguished Service Professor in the
Department of Mathematics at the University of Chicago.
NEW MATHEMATICAL MONOGRAPHS
Editorial Board
B´ ela Bollob´ as
William Fulton
Anatole Katok
Frances Kirwan
Peter Sarnak
Barry Simon
Burt Totaro
All the titles listed below can be obtained from good booksellers or from Cambridge
University Press. For a complete series listing visit
http://www.cambridge.org/uk/series/sSeries.asp?code=NMM
1 M. Cabanes and M. Enguehard Representation Theory of Finite Reductive Groups
2 J. B. Garnett and D. E. Marshall Harmonic Measure
3 P. Cohn Free Ideal Rings and Localization in General Rings
4 E. Bombieri and W. Gubler Heights in Diophantine Geometry
5 Y. J. Ionin and M. S. Shrikhande Combinatorics of Symmetric Designs
6 S. Berhanu, P. D. Cordaro and J. Hounie An Introduction to Involutive Structures
7 A. Shlapentokh Hilbert’s Tenth Problem
8 G. Michler Theory of Finite Simple Groups I
9 A. Baker and G. W¨ ustholz Logarithmic Forms and Diophantine Geometry
10 P. Kronheimer and T. Mrowka Monopoles and ThreeManifolds
11 B. Bekka, P. de la Harpe and A. Valette Kazhdan’s Property (T)
12 J. Neisendorfer Algebraic Methods in Unstable Homotopy Theory
13 M. Grandis Directed Algebraic Topology
14 G. Michler Theory of Finite Simple Groups II
15 R. Schertz Complex Multiplication
Lectures on Algebraic Cycles
Second Edition
SPENCER BLOCH
University of Chicago
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore,
São Paulo, Delhi, Dubai, Tokyo
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
First published in print format
ISBN13 9780521118422
ISBN13 9780511900310
© David Leibowitz 2010
2010
Information on this title: www.cambridge.org/9780521118422
This publication is in copyright. Subject to statutory exception and to the
provision of relevant collective licensing agreements, no reproduction of any part
may take place without the written permission of Cambridge University Press.
Cambridge University Press has no responsibility for the persistence or accuracy
of urls for external or thirdparty internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
eBook (EBL)
Hardback
Contents
Preface to the second edition page vii
0 Introduction 3
1 Zerocycles on surfaces 9
Appendix: On an argument of Mumford in the theory of algebraic
cycles 21
2 Curves on threefolds and intermediate jacobians 25
3 Curves on threefolds – the relative case 37
4 Ktheoretic and cohomological methods 45
5 Torsion in the Chow group 59
6 Complements on H
2
(K
2
) 69
7 Diophantine questions 79
8 Relative cycles and zeta functions 97
9 Relative cycles and zeta functions – continued 115
Bibliography 123
Index 129
Preface to the second edition
30 Years later...
Looking back over these lectures, given at Duke University in 1979, I can
say with some pride that they contain early hints of a number of important
themes in modern arithmetic geometry. Of course, the ﬂip side of that coin
is that they are now, thirty years later, seriously out of date. To bring them
up to date would involve writing several more monographs, a task best left
to mathematicians thirty years younger than me. What I propose instead is to
comment fairly brieﬂy on several of the lectures in an attempt to put the reader
in touch with what I believe are the most important modern ideas in these
areas. The section on motives just below is intended as a brief introduction
to the modern viewpoint on that subject. The remaining sections until the last
follow roughly the content of the original book, though the titles have changed
slightly to reﬂect my current emphasis. The last section, motives in physics,
represents my recent research.
In the original volume I included a quote from Charlie Chan, the great Chi
nese detective, who told his bumbling number one son “answer simple, but
question very very hard.” It seemed to me an appropriate comment on the sub
ject of algebraic cycles. Given the amazing deep new ideas introduced into the
subject in recent years, however, I think now that the question remains very
very hard, but the answer is perhaps no longer so simple...
At the end of this essay I include a brief bibliography, which is by no means
complete. It is only intended to illustrate the various ideas mentioned in the
text.
viii Preface to the second edition
Motives
Much of the recent work in this area is centered around motives and the con
struction – in fact various constructions, due to Hanamura (1995), Levine
(1998), and Voevodsky (Mazza et al. 2006; Voevodsky et al. 2000) – of a trian
gulated category of mixed motives. I will sketch Voevodsky’s construction as
it also plays a central role in his proof of the Bloch–Kato–Milnor conjecture
discussed in Lecture 5. Then I will discuss various lectures from the original
monograph.
Let k be a ﬁeld. The category Cor
k
is an additive category with objects
smooth kvarieties: Ob(Cor
k
) = Ob(Sm
k
). Morphisms Z: X → Y are ﬁnite
linear combinations of correspondences Z =
_
n
i
Z
i
where Z
i
⊂ X ×Y is closed
and the projection π
i
: Z
i
→ X is ﬁnite and surjective. Intuitively, we may think
of Z
i
as a map X → Sym
n
Y associating to x ∈ X the ﬁbre f
−1
i
(x) viewed as a
zerocycle on Y. There is an evident functor
Sm
k
→Cor
k
which is the identity on objects.
A presheaf on a category ( with values in an abelian category 7is simply a
contravariant functor F: (
op
→ 7. An 7valued presheaf F on Cor
k
induces
a presheaf F¦
Sm
k
on Sm
k
. Intuitively, to lift a presheaf G from Sm
k
to Cor
k
one
needs a structure of trace maps or transfers f
∗
: G(Z) → G(X) for Z/X ﬁnite.
Presheaves on Cor
k
are referred to as presheaves with transfers.
For X ∈ Ob(Sm
k
) one has the representable sheaf Z
tr
(X) deﬁned by
Z
tr
(X)(U) = Hom
Cor
k
(U, X).
An important elaboration on this idea yields for pointed objects x
i
∈ X
i
Z
tr
((X
1
, x
1
) ∧ ∧ (X
n
, x
i
))
:= Coker
_
Z
tr
(X
1
×
¯
X
i
× X
n
) → Z
tr
(
X
i
)
_
.
In particular, one deﬁnes Z
tr
(
_
n
G
m
) by taking X
i
= A
1
− ¦0¦ and x
i
= 1.
A presheaf with transfers F is called homotopy invariant if, with obvious
notation, i
∗
0
= i
∗
1
: F(U × A
1
) → F(U). The complex of chains C
∗
(F) on a
presheaf with transfers F is the presheaf of complexes (placed in cohomologi
cal degrees [−∞, 0])
C
∗
(F) := U ·→ → F(U × ∆
n
) → → F(U × ∆
0
)
Here
(0.1) ∆
n
:= Spec k[t
0
, . . . , t
n
]/(
_
t
i
− 1)
Preface to the second edition ix
is the algebrogeometric nsimplex. The boundary maps in the complex are
the usual alternating sums of restrictions to the faces ∆
n−1
→ ∆
n
deﬁned by
setting t
i
= 0. The two restrictions
i
∗
0
, i
∗
1
: C
∗
(F)(U × A
1
) →C
∗
(F)(U)
are shown to be homotopic, so the homology presheaves H
n
(C
∗
(F)) are homo
topy invariant.
Maps f
0
, f
1
: X → Y in Cor
k
are A
1
homotopic if there exists H: X×A
1
→ Y
in Cor
k
such that f
j
= i
∗
j
H. A
1
homotopy is an equivalence relation, and A
1

homotopic maps induce homotopic maps
f
0∗
= f
1∗
: C
∗
Z
tr
(X) →C
∗
Z
tr
(Y).
Voevodsky deﬁnes
Z(q) := C
∗
Z
tr
(
_
q
G
m
)[−q], q ≥ 0.
More precisely, the above complex is viewed as a complex of presheaves on
Sm
k
and then localized for the Zariski topology. Motivic cohomology is then
deﬁned (for q ≥ 0) as the hypercohomology of this complex of Zariski sheaves:
H
p
M
(X, Z(q)) := H
p
Zar
(X, Z(q)).
One has a notion of tensor product for presheaves on the category Cor
k
, and
Z
tr
(X) ⊗Z
tr
(Y) = Z
tr
(X ×Y). In particular, Z(p) ⊗Z(q) → Z(p+q) so one gets
a product structure on motivic cohomology. In low degrees one has
H
0
M
(X, Z(0)) = Z[π
0
(X)],
H
p
M
(X, Z(0)) = (0), p > 0,
Z(1) G
m
[−1].
Another important sign that this is the right theory is the link with Milnor
Ktheory. (K
Milnor
∗
(k) is deﬁned as the quotient of the tensor algebra on k
×
by
the ideal generated by quadratic relations a ⊗ (1 − a) for a ∈ k − ¦0, 1¦.)
Theorem H
n
M
(Spec k, Z(n)) K
Milnor
n
(k).
The fact that the Zariski topology suﬃces to deﬁne motivic cohomology is
somewhat surprising because a Zariski open cover π: U → X does not yield a
resolution of Zariski sheaves
(0.2) → Z
tr
(U ×
X
U) → Z
tr
(U) → Z(X) → 0.
To remedy this, Voevodsky employs the Nisnevich topology. A morphism
π: U → X is a Nisnevich cover if for any ﬁeld K/k one has U(K) X(K).
x Preface to the second edition
To see that (0.2) becomes exact when localized for the Nisnevich topology, one
uses the fact that any ﬁnite cover of a Hensel local ring is a product of local
rings.
The actual triangulated category of eﬀective motives over k is a quotient cat
egory of the derived category D
−
(Sh
Nis
(Cor
k
)) of boundedbelow complexes of
Nisnevich sheaves on Cor
k
. One considers the smallest thick subcategory W
containing all cones of Z
tr
(X × A
1
) → Z
tr
(X), and one deﬁnes
DM
eﬀ
Nis
(k) := D
−
Sh
Nis
(Cor
k
)[W
−1
].
Said another way, one formally inverts all morphisms with cones in W. Finally,
the motive associated to a smooth kvariety X is deﬁned by
(0.3) M(X) := Z
tr
(X) ∈ DM
eﬀ
Nis
(k).
The category of geometric motives DM
eﬀ
geo
(k) is the thick subcategory in DM
eﬀ
Nis
(k)
generated by the M(X).
One has the following properties:
MayerVietoris
M(U ∩ V) → M(U) ⊕ M(V) → M(X) → M(U ∩ V)[1]
is a distinguished triangle.
K¨ unneth
M(X × Y) = M(X) ⊗ M(Y).
Vector bundle theorem
M(X) M(V)
for V/X a vector bundle.
Cancellation Assume varieties over k admit a resolution of singularities.
Write M(q) := M ⊗ Z(q). Then
Hom(M, N) Hom(M(q), N(q)).
The category of (not necessarily eﬀective) motives is obtained by inverting the
functor M ·→ M(1) in DM
eﬀ
.
Projective bundle theorem For V/X a rank n + 1 vector bundle
M(P(V))
n
i=0
M(X)(i)[2i].
Preface to the second edition xi
Chow motives If X, Y are smooth projective, then
(0.4) Hom(M(X), M(Y)) CH
dimX
(X × Y).
The category of Chow motives over a ﬁeld k has as objects triples
(X, p, m) with X smooth projective over k, p ∈ CH
dimX
(X × X)
Q
a projector
(i.e. p ◦ p = p) and m ∈ Z. The morphisms are given by
(0.5) Hom((X, p, m), (Y, q, n)) := q ◦ CH
dimX+n−m
(X × Y) ◦ p.
It follows from (0.4) and the existence of projectors in DM
eﬀ
Nis
(k) that the cate
gory of Chow motives embeds in DM
eﬀ
Nis
(k).
Motivic cohomology For X/k smooth, we have
(0.6) H
p
M
(X, Z(q)) Hom
DM
eﬀ
Nis
(k)
(Z
tr
(X), Z(q)[p]).
In fact, motivic cohomology is closely related to algebraic cycles, and this
relationship lies at the heart of modern cycle theory. There are a number of
ways to formulate things. I will use higher Chow groups because they relate
most naturally to arithmetic questions. Let ∆
•
be the cosimplicial variety as in
(0.1) above. Deﬁne ¸
q
(X, n) to be the free abelian group of algebraic cycles on
X × ∆
n
which are in good position with respect to all faces X × ∆
m
→ X × ∆
n
.
The complex ¸
q
(X, •) is deﬁned by taking alternating sums of pullbacks in the
usual way:
(0.7) → ¸
q
(X, 2) → ¸
q
(X, 1) → ¸
q
(X, 0) → 0.
(Here ¸
q
(X, n) is placed in cohomological degree −n.) The higher Chowgroups
are deﬁned by
(0.8) CH
q
(X, n) := H
−n
(¸
q
(X, •)).
For example, the usual Chow group CH
q
(X) = CH
q
(X, 0). Voevodsky proves
that for X smooth over a perfect ﬁeld k one has
(0.9) H
p
M
(X, Z(q)) CH
q
(X, 2q − p) = H
p
(¸
q
(X, •)[−2q]).
Beilinson and Soul´ e conjecture that the shifted chain complex
¸
q
(X, •)[−2q] ⊗ Q has cohomological support in degrees [0, 2q]. Actually,
their conjecture was formulated in terms of the γﬁltration in Ktheory, but
one has the further identiﬁcation
(0.10) H
p
M
(X, Z(q)) ⊗ Q CH
q
(X, 2q − p) ⊗ Q gr
q
γ
K
p
(X) ⊗ Q.
xii Preface to the second edition
Lecture 1: Zerocycles
The two most important ideas here are ﬁrstly the conjecture that surfaces with
geometric genus zero (p
g
= 0) should have Chow group of zerocycles repre
sentable. For S such a surface over C we expect an exact sequence
0 → Alb(S ) → CH
0
(S )
deg
−−→ Z → 0.
The group T(S ) deﬁned in Lemma 1.4 is conjectured to be zero in
this case. Secondly, for any smooth projective variety X, the Chow
group of zerocycles CH
0
(X) is conjectured (1.8) to carry a descending
ﬁltration F
∗
CH
0
(X) which is functorial for correspondences such that
the map gr
p
F
CH
0
(X)
Λ
−→ gr
p
F
CH
0
(Y) induced by an algebraic cycle
Λ ∈ CH
dimY
(X × Y) depends only on the class of Λ in cohomology.
Indeed, one may conjecture the existence of such a ﬁltration on CH
q
(X)
for any q.
These conjectures remain unproven, but a very beautiful general picture,
based on the yoga of mixed motives, has been elaborated by A. Beilinson.
Interested readers should consult the important article by Jannsen (1994) and
the literature cited there. Let me sketch brieﬂy (following Jannsen) the modern
viewpoint.
It is convenient to dualize the deﬁnition of M(X) (0.3). Assume X smooth,
projective of dimension d. Deﬁne (Hom means the internal Hom in DM)
M(X)
∗
= Hom
DM
(M(X), Z(0)).
The formula for the Chow group becomes
CH
p
(X) = H
2p
M
(X, Z(p)) = Hom
DM
(Z(0), M(X)
∗
(p)[2p]).
One of Grothendieck’s standard conjectures about algebraic cycles is that
there exist K¨ unneth projectors
π
i
∈ CH
d
(X × X)
Q
/¦homological equivalence¦
inducing the natural projections H
∗
(X) → H
i
(X) on cohomology. If we assume
further that the ideal ¦homological equiv.¦/¦rational equiv.¦ ⊂ CH
d
(X × X)
Q
is
nilpotent (nilpotence conjecture), then the pr
i
lift (noncanonically) to projec
tors pr
i,rat
∈ CH
d
(X × X)
Q
and we may use (0.5) to decompose M(X)
∗
⊗ Q =
i
h
i
[−i] noncanonically as a direct sum of Chow motives. This idea is due
to J. Murre (1993a). The hope is that DM admits a tstructure such that
(0.11) H
i
(M(X)
∗
⊗ Q) = h
i
(X).
Preface to the second edition xiii
The resulting spectral sequence
E
p,q
2
= Hom
DM
(Z(0), H
q
(M(X)
∗
( j)[p]) ⇒ Hom
DM
(Z(0), M
∗
(X)( j)[p+q])
would yield ﬁltrations on the Chow groups ⊗Q with
F
ν
CH
j
(X)
Q
2j−ν
i=0
Ext
2j−i
DM
(Q(0), h
i
(X)( j)),
gr
ν
F
CH
j
(X)
Q
Ext
ν
DM
(Q(0), H
2j−ν
(X)( j)).
Murre suggests a natural strengthening of his conjectures, based on the idea
that one should be able for i ≤ d, to ﬁnd representatives for π
i
supported on
X
i
× X ⊂ X × X, where X
i
⊂ X is a general plane section of dimension i. For
example, π
0
= ¦x¦ × X for a point x. Clearly this would imply π
i
CH
j
(X) = 0
for i < j, and since the conjectures imply
F
ν
CH
j
(X) =
2j−ν
i=0
π
i
CH
j
(X),
we could conclude further that
F
ν
CH
j
(X) = (0), ν > j.
Suppose, for example, that dimX = 2. We would get a 3step ﬁltration on
the zerocycles: CH
0
(X) = F
0
⊃ F
1
⊃ F
2
⊃ (0), with
gr
0
F
CH
0
(X)
Q
= Hom
DM
(Q(0), h
4
(X)(2)),
gr
1
F
= Ext
1
DM
(Q(0), h
3
(X)(2)),
gr
2
F
= Ext
2
DM
(Q(0), h
2
(X)(2)).
This ﬁts perfectly with the ideas in Lecture 1. Indeed, Murre has computed
gr
0
and gr
1
and he ﬁnds exactly the degree and the Albanese. Of course, gr
2
is more problematical, but note that the condition discussed in the text that
H
2
(X, Q
(1)) should be generated by divisors (which is equivalent to p
g
= 0
in characteristic zero) means h
2
(X)(2) ⊕Q(1). (This is obvious for motives
modulo homological equivalence. Assuming the nilpotence conjecture, it holds
for Chow motives as well.) In this case, gr
2
F
can be computed for X = P
2
when
it is clearly (0).
The conjectural “theorem of the hypersquare” (Proposition 1.12) can be
understood in motivic terms (Jannsen 1994, conj. 3.12) using the fact that
h
n
(X
0
× × X
n
) is a direct summand of
h
n
(X
0
×
¯
X
i
× X
n
).
Clearly wrongheaded, however, is Metaconjecture 1.10, which stated that
F
2
CH
0
(X) is controlled by the polarized Hodge structure associated to H
2
(X).
Indeed, Ext
2
= (0) in the category of Hodge structures. One may try (compare
Carlson and Hain 1989) to look at Exts in some category of variations of Hodge
xiv Preface to the second edition
structure. In the absence of parameters supporting such variations (i.e. for X
over a number ﬁeld), however, the Ext
2
term should vanish and we should
have F
2
CH
2
(X) = (0).
Lectures 2 and 3: Intermediate jacobians
The modern point of view about intermediate jacobians is to view them as
Ext
1
(Z(0), H) where H is a suitable Hodge structure, and the Ext group is
taken in the abelian category of mixed Hodge structures (Carlson 1987). In the
classic situation, H = H
2r−1
(X, Z(r)) where X is a smooth projective variety.
Note in this case that H has weight −1. An extension 0 → H → E → Z(0) → 0
would yield a mixed Hodge structure E with weights 0, −1 and Hodge ﬁltration
E
C
= F
−1
E
C
⊃ F
0
E
C
.
Let f , w ∈ E
C
be liftings of 1 ∈ Z(0) splitting the weight and Hodge ﬁltrations
respectively. The diﬀerence between them w − f gives a welldeﬁned class in
J = H
C
/(F
0
H
C
+ H
Z
) which is the intermediate jacobian. To deﬁne the class
of a codimensionr cycle Z =
_
n
i
Z
i
, let ¦Z¦ be the support of the cycle. We
have a cycle class with supports [Z] : Z → H
2r
¦Z¦
(X, Z(r)) and a diagram
(0.12)
H
2r−1
¦Z¦
(X, Z(r)) H
2r−1
(X, Z(r)) H
2r−1
(X − ¦Z¦, Z(r))
H
2r
¦Z¦
(X, Z(r)) H
2r
(X, Z(r))
Z
0
¦Z¦
H
2r
(X, Z(r))
The ﬁrst group H
2r−1
¦Z¦
(X, Z(r)) is zero by purity, and vanishing of the lower
righthand arrow will hold if Z is homologous to 0. Assuming this, we get the
desired extension of Hodge structures. (I believe this construction is due to
Deligne, though I do not have a good reference.)
For H any mixed Hodge structure with weights < 0 one has an analogous
construction. Note, however, the resulting abelian Lie group J need not be
compact. For example, Ext
1
(Z(0), Z(1)) C
×
S
1
× R. In Lecture 3, the
focus is on the case
H = H
1
(C, Z(2)) ⊗ H
1
c
(G
m
, Z)
⊗2
⊂ H
3
c
(C × (G
m
)
2
, Z(2)).
Preface to the second edition xv
Here H has weight −3, and
Ext
1
MHS
(Z(0), H) H
1
(C, C)/H
1
(C, Z(2)) H
1
(C, C
×
(1)).
At the time I was very much inspired by the work of Borel (1977) on reg
ulators for higher Kgroups of number ﬁelds. I believed that similar regula
tors could be deﬁned for arithmetic algebraic varieties more generally, and that
these regulators could be related to values of Hasse–Weil Lfunctions. This was
done in a very limited and ad hoc way in Bloch (1980, 2000), and then much
more deﬁnitively by Beilinson (1985). From the point of view of Lecture 3, the
regulator can be thought of as a relative cycle class map
H
p
M
(X, Z(q))
(0.9)
CH
q
(X, 2q−p) → Ext
1
MHS
(Z(0), H) = H
C
/(H
Z
+F
0
H
C
).
Here H = H
2q−1
(X × ∆
2q−p
, X × ∂∆
2q−p
; Z(q)). For details of this construc
tion, see Bloch (2000). Other constructions are given in Bloch (1986b) and
Goncharov (2005).
The quotient of this Ext group by its maximal compact subgroup is the cor
responding Ext in the category of RHodge structures. It is an Rvector space.
More generally one can associate to any mixed Hodge structure a nilpotent
matrix γ (Cattani et al., 1986, prop. 2.20), which is the obstruction to a real
splitting of the ﬁltration by weights. These invariants arise, for example, if the
curve C in Lecture 3 is allowed to degenerate, so H
1
(C, Z) is itself a mixed
Hodge structure.
Lecture 4: Cohomological methods
This chapter contains basic information about algebraic Ktheory, an important
tool in the study of algebraic cycles. I describe the “Quillen trick” and use it
to construct the Gersten resolution for Ksheaves and also Betti and ´ etale co
homology sheaves for smooth varieties. Brieﬂy, one considers Zariski sheaves
7
q
(resp. 1
q
Betti
, resp. 1
q
et
) associated to the presheaf U ·→ K
q
(U) of algebraic
Kgroups (resp. U ·→ H
q
Betti
(U, Z), resp. U ·→ H
q
et
(U, Z/nZ)). One obtains
resolutions of these sheaves which enable one to identify, for example,
H
p
(X, 7
p
) CH
p
(X),
H
p
(X, 1
p
Betti
) CH
p
(X)/¦algebraic equivalence¦.
(0.13)
I think it is fair to say that the resulting Kcohomology and the parallel
constructions for Betti and ´ etale cohomology have had important technical ap
plications but have not been the breakthrough one had hoped for at the time.
xvi Preface to the second edition
Despite the Gersten resolution, it turns out to be diﬃcult to interpret the result
ing cohomology. Finiteness results, for example, are totally lacking. One nice
application of the Betti theory (see reference [6] at the end of Lecture 4) was
to falsify a longstanding conjecture about diﬀerential forms of the second kind
on varieties of dimension ≥ 3. The spectral sequence E
p,q
2
= H
p
Zar
(X, 1
q
Betti
) →
H
p+q
Betti
(X, Z) leads to an exact sequence
(0.14) H
3
Betti
(X)
a
− → Γ(X, 1
3
)
b
− → H
2
(X, 1
2
)
c
− → H
4
Betti
(X)
Using (0.13) one can identify Ker(c) in (0.14) with the Griﬃths group of
cycles homologous to zero modulo algebraic equivalence, a group which in
some cases is known not to be ﬁnitely generated (Clemens 1983). It follows in
such cases that a is not surjective, indeed Coker(a) is inﬁnitely generated. But
Γ(X, 1
3
) is precisely the space of meromorphic 3forms of the second kind;
that is meromorphic forms which at every point diﬀer from an algebraic form
which is regular at the point by an exact algebraic form. Thus, unlike the case
of curves and surfaces, diﬀerential forms of the second kind do not necessarily
come from global cohomology classes in dimensions ≥ 3.
Lecture 5: The conjecture of Milnor–Bloch–Kato
Let F be a ﬁeld and a prime with 1/ ∈ K. The Milnor ring K
M
∗
(F)/ is gener
ated by F
×
/F
×
with relations given by Steinberg symbols f ⊗(1− f ), f 0, 1.
The conjecture in question states that the natural map to Galois cohomology
K
M
∗
(F)/ → H
∗
(F, µ
∗
)
is an isomorphism. My own contribution to this, which is explained in Lec
ture 5, is a proof that K
M
n
(F) → H
n
(F, µ
⊗n
) is surjective when F has coho
mological dimension n. For some years Voevodsky has been working on a
very diﬃcult program, using his own motivic theory and results of M. Rost,
to prove the conjecture in complete generality. The proof is now complete (for
an outline with references, see the webpage of C. Weibel), but there is still no
uniﬁed treatment and the arguments use sophisticated techniques in algebraic
homotopy theory which I have not understood.
Geometrically, the result can be formulated as follows. Let i : X → Y be a
closed immersion of varieties over a ﬁeld k of characteristic prime to , and let
j : Y − X → Y be the open immersion. For simplicity I assume µ
⊂ k so there
is no need to distinguish powers of µ
. Consider the exact sequence
H
p
(Y, Z/)
i
∗
−→ H
p
(X, Z/)
∂
− → H
p+1
(Y, j
!
Z/).
Preface to the second edition xvii
Given a cohomology class c ∈ H
p
(X) and a smooth point x ∈ X, there exists a
Zariski open neighborhood Y ⊃ U ÷ x such that 0 = ∂(c)
¸
¸
¸
U
∈ H
p+1
(U, j
!
Z/).
As an exercise, the reader might work out how this is equivalent to the con
jecture for F = k(X). Another exercise is to formulate a version of coniveau
ﬁltration as described on page 52 for the group H
p+1
(Y, j
!
Z/) in such a way
that the image of ∂ lies in F
1
.
The whole picture of motivic cohomology with ﬁnite coeﬃcients is now
quite beautiful (Voevodsky et al. 2000). Let X be a smooth, quasiprojective
scheme over an algebraically closed ﬁeld k, and let m ≥ 2 be relatively prime
to the characteristic. Let r ≥ dimX. Then
H
2r−n
M
(X, Z/mZ(r)) H
2r−n
et
(X, Z/mZ(r)).
Said another way, the cycle complexes ¸
r
(X, •)[−2r], in equation (0.9), com
pute the adic ´ etale cohomology for all prime to the characteristic, assuming
r ≥ dimX. The situation should be compared with that for abelian varieties A
where one has Tate modules T
(A) for all and these calculate H
1
(A, Z
) for
prime to the characteristic.
The subject of torsion in the Chow group seems to be important from many
points of view. I include in the bibliography a couple of relevant papers (Soul´ e
and Voisin 2005; Totaro 1997).
Lecture 6: Inﬁnitesimal methods in motivic cohomology
The inﬁnitesimal methods developed here were used also in my work on de
Rham–Witt cohomology (Bloch 1977).
It is fair to say that we still do not have an adequate notion of motivic coho
mology. That is, we do not have contravariant cohomology functors deﬁned on
singular schemes (e.g. on nonreduced schemes) with appropriate properties.
The notion of A
1
homotopy invariance which is essential in Voevodsky’s work
is not what one wants. For example, if A is a nonreduced ring, then typically
H
1
M
(Spec A, Z(1)) = A
×
A[t]
×
= H
1
M
(Spec A[t], Z(1)).
Curiously, the Kcohomology groups H
p
Zar
(X, 7
q
) discussed in Lecture 4 do
have the correct functoriality properties, and in this lecture we examine what
can be learned from the inﬁnitesimal structure of these groups.
An important step has been the work of Goodwillie (1986) computing the
Ktheory of nilpotent ideals in characteristic zero in terms of cyclic homology.
To understand what motivic cohomology of an inﬁnitesimal thickening might
mean, the reader could consult the two rather experimental papers Bloch and
xviii Preface to the second edition
Esnault (1996, 2003). More deﬁnitive results have been obtained in Krishna
and Levine (2008), Park (2007), and R¨ ulling (2007).
Assuming one has a good deﬁnition of motivic cohomology, what should
the “tangent space”
TH
p
M
(X, Z(q)) := Ker
_
H
p
(X × Spec k[ε], Z(q)) → H
p
(X, Z(q))
_
mean? (Here ε
2
= 0 and the map sends ε ·→ 0.) To begin with, one should
probably not think of TH
M
as a tangent space in the usual sense. It can be
nontrivial in situations where the motivic cohomology itself is rigid, for ex
ample for H
3
M
(k, Z(2)) with k a number ﬁeld. Better, perhaps, to think of a
nonsemistable moduli functor where jumps can occur at the boundary. For ex
ample, consider the Picard scheme of (P
1
, ¦a, b¦), that is isomorphism classes
of line bundles on P
1
with trivializations at a, b. The degreezero part is con
stant G
m
for a b, but the limit as a → b is G
a
given by degreezero line
bundles on P
1
with double order trivialization at a = b.
One important area of open questions about these groups concerns regu
lators and relations with Euclidean scissors congruence groups. The regulator
for usual motivic cohomology is closely related to volumes in hyperbolic space
(Goncharov 1999), and it seems likely that there is a similar relation between
TH
M
and Euclidean volumes. Intuitively, this is another one of those limiting
phenomena where the radius of the hyperbolic disk is allowed to go to inﬁnity
and lengths are scaled so in the limit one gets Euclidean geometry. It would be
nice to have a rigorous description of how this works.
Lecture 7: Diophantine questions
At the time of these lectures, I had expected that the Chow group of zerocycles
on a rational surface would relate in some way to the zeta function of the sur
face. As far as I can tell, that does not happen, and I have not thought further
in this direction. The reader who wants to work on diophantine questions re
garding zerocycles and Chow groups should consult ColliotTh´ el` ene (1995),
Esnault (2003), and the references given in these papers.
Lectures 8 and 9: Regulators and values of Lfunctions
The whole subject of motivic cohomology, regulators, and values of Lfunctions
remains to a large extent conjectural, but we now understand much better what
should be true (Rapoport et al. 1988; Soul´ e 1986; Bloch and Kato 1990). For
Preface to the second edition xix
constructions of the regulator, the reader can consult Goncharov (2005) and
Bloch (1986b). Concerning the basic conjecture, I am especially attracted to
the formulation given by Fontaine and PerrinRiou (Fontaine 1992; Fontaine
and PerrinRiou 1994). To understand their idea, let X be a smooth, projective
variety of dimension d over Q. (In what follows I gloss over many intractable
conjectures.) Consider a motive M = h
p
(X)(q), where h
p
(X) is a Chow motive
as in (0.11), and write M
∗
(1) := h
2d−p
(X)(d − q + 1). Write M
B
= H
p
B
(X, Q(q))
for the Betti cohomology of X(C). Let M
+
B
⊂ M
B
be the subspace invariant
under the action of conjugation. Let t
M
:= H
p
DR
(X, Q(q))/F
0
, where H
∗
DR
is
de Rham cohomology. There is a natural map α: M
+
B
⊗ R → t
M
⊗ R, and
(assuming certain conjectures) Fontaine and PerrinRiou construct an exact
sequence of motivic cohomology
0 → Hom
DM
(Q(0), M) ⊗ R → Ker α → (Ext
1
DM, f
(Q(0), M
∗
(1)) ⊗ R)
∗
→ Ext
1
DM, f
(Q(0), M) ⊗ R → Coker α → (Hom
DM
(Q, M
∗
(1)) ⊗ R)
∗
→ 0.
(0.15)
Here Hom and Ext are taken in the triangulated category of Voevodsky mo
tives over Q. The subscript f refers to behavior at ﬁnite primes. As a con
sequence of (0.15) one gets a trivialization over R of the tensor product of
determinant lines
(0.16) det(RHom
DM, f
(Q(0), M))
R
⊗ det(RHom
DM, f
(Q(0), M
∗
(1)))
R
⊗ det(α)
−1
R.
The various determinants have Qstructures (though α, itself, does not), so
one may examine in (0.16) the ratio of the real trivialization and the rational
structure. In fact, using Galois and adic cohomology, the authors actually
get a Zstructure on the left. They show that the integral conjecture in Bloch
and Kato (1990) is equivalent to this ratio being given by L
∗
(M, 0), the ﬁrst
nonvanishing term in the Taylor expansion of L(M, s) where L(M, s) is the
Hasse–Weil Lfunction associated to M.
Well, okay, there is a lot here we do not understand, but my thought is that
one might redo (0.15) working directly with the cycle complexes ¸
q
(X, •)[−2q]
and ¸
d+1−q
(X, •)[−2(d + 1 − q)] which one should think of as concrete real
izations of motivic cohomology RΓ
M
(X, Z(q)) and RΓ
M
(X, Z(d + 1 − q)). The
resulting determinant metrics could perhaps be deduced (or at least interpreted)
directly from the intersection–projection map (well deﬁned in the derived cat
egory)
¸
q
(X, •)[−2q]
L
⊗ ¸
d+1−q
(X, •)[−2(d + 1 − q)] → ¸
1
(Spec Q, •)[−2d].
xx Preface to the second edition
Hanamura suggested this approach to understanding heights and biexten
sions. Of course, as it stands it is purely algebraic. It will be necessary to
take the cone over the regulator map in some fashion. The result should be
some kind of metric or related structure on the determinant of the cycle com
plex. This would ﬁt well with a conjecture of Soul´ e (1984), which says in this
context that for X regular and proper of dimension d over Spec Z, the Euler
characteristic of ¸
p
(X, •)[−2p] should be deﬁned and we should have
χ(¸
p
(X, •)[−2p]) =
(−1)
i
dim
Q
H
i
M
(X, Q(p)) = −ord
s=d−p
ζ
X
(s),
the negative of the order of zero or pole at s = d − p of the zeta function of X.
Coda: Motives in physics
The subjects of algebraic cycles and motives have enjoyed a tremendous the
oretical development over the past 30 years. At the risk of scandalizing the
reader, I would say it is high time we start looking for applications.
Dirk Kreimer has been teaching me about Feynman amplitudes and pertur
bative calculations in quantum ﬁeld theory. These are periods that arise, for
example, in computations of scattering amplitudes. They have a strong ten
dency to be multizeta numbers (Bloch et al. 2006; Broadhurst and Kreimer
1997; but cf. Belkale and Brosnan 2003). The periods in question are associ
ated to graphs. Essentially, the Kirchhoﬀ polynomial (Bloch et al. 2006) of a
graph Γ deﬁnes a hypersurface X
Γ
in projective space, and the Feynman am
plitude is a period of this hypersurface relative to a reference symplex. If in
deed these are multizeta numbers it should be the case that the cohomology
of X
Γ
has a big Tate piece. One knows if X
Γ
were smooth, then H
n
Betti
(X
Γ
, Q)
would have pure weight n, so any Tate class, i.e. any map of Hodge structures
Z(−p) → H
n
Betti
(X
Γ
, Q) would necessarily be a Hodge class, i.e. n = 2p. The
Hodge conjecture would say that such a class comes from an algebraic cycle,
i.e. a class in H
2p
M
(X
Γ
, Z(p)) via the realization map from motivic cohomology
to Betti cohomology. Unfortunately (or perhaps fortunately), the X
Γ
are highly
singular, so all one can say are that the weights of H
n
are ≤ n. There are all
kinds of possibilities for interesting cohomology classes coming via realiza
tion from motivic cohomology. For example, the “wheel with n spokes graph”
WS (n) gives rise to a hypersurface X
WS (n)
of dimension 2n − 2. The primi
tive cohomology in the middle dimension for this graph is computed (Bloch
et al. 2006) to be Q(−2) (independent of n). An appropriate generalization of
the Hodge conjecture would suggest a class in H
2n−2
M
(X
WS (n)
, Q(2)). The prob
lem of computing such motivic cohomology groups can be attacked via the
Preface to the second edition xxi
combinatorics of the graph, but what has so far proved more powerful is to
use classical algebrogeometric techniques to study the geometry of rank strat
iﬁcations associated to a homomorphism of vector bundles u: E → F over
projective space.
Ideally, knowledge of motives should provide a strong organizing force to
study complex physical phenomena. Even simple motivic invariants like weight
and Hodge level should help physicists understand the periods arising in their
computations. More sophisticated methods involving monodromy and limiting
mixed motives may give information about Landau singularities and unitarity
of the S matrix.
References for preface
[1] Be˘ılinson, A. A. 1985. Higher regulators and values of Lfunctions. J.
Soviet Math., 30, 2036–2070.
[2] Be˘ılinson, A. A., A. B. Goncharov, V. V. Schechtman, and A. N. Var
chenko. 1990. Aomoto dilogarithms, mixed Hodge structures and mo
tivic cohomology of pairs of triangles on the plane. Pp. 135–172 in
The Grothendieck Festschrift, Vol. I. Progr. Math., vol. 86. Boston:
Birkh¨ auser Boston.
[3] Belkale, Prakash, and Patrick Brosnan. 2003. Matroids, motives, and a
conjecture of Kontsevich. Duke Math. J., 116(1), 147–188.
[4] Bloch, Spencer. 1977. Algebraic Ktheory and crystalline cohomology.
Inst. Hautes
´
Etudes Sci. Publ. Math., 187–268 (1978).
[5] Bloch, S. 1980. Algebraic Ktheory and zeta functions of elliptic curves.
Pp. 511–515 in Proceedings of the International Congress of Mathemati
cians (Helsinki, 1978). Helsinki: Acad. Sci. Fennica.
[6] Bloch, S. 1994. The moving lemma for higher Chow groups. J. Algebraic
Geom., 3(3), 537–568.
[7] Bloch, Spencer. 1986a. Algebraic cycles and higher Ktheory. Adv. in
Math., 61(3), 267–304.
[8] Bloch, Spencer. 1986b. Algebraic cycles and the Be˘ılinson conjectures.
Pp. 65–79 in The Lefschetz Centennial Conference, Part I (Mexico City,
1984). Contemp. Math., vol. 58. Providence, R.I.: American Mathemat
ical Society.
[9] Bloch, Spencer, and H´ el` ene Esnault. 1996. The coniveau ﬁltration and
nondivisibility for algebraic cycles. Math. Ann., 304(2), 303–314.
[10] Bloch, Spencer J. 2000. Higher Regulators, Algebraic KTheory, and
xxii Preface to the second edition
Zeta Functions of Elliptic Curves. CRM Monograph Series, vol. 11.
Providence, R.I.: American Mathematical Society.
[11] Bloch, Spencer, and H´ el` ene Esnault. 2003. An additive version of higher
Chow groups. Ann. Sci.
´
Ecole Norm. Sup. (4), 36(3), 463–477.
[12] Bloch, Spencer, H´ el` ene Esnault, and Dirk Kreimer. 2006. On motives
associated to graph polynomials. Comm. Math. Phys., 267(1), 181–225.
[13] Bloch, Spencer, and Kazuya Kato. 1990. Lfunctions and Tamagawa
numbers of motives. Pp. 333–400 in The Grothendieck Festschrift, Vol.
I. Progr. Math., vol. 86. Boston: Birkh¨ auser Boston.
[14] Bloch, Spencer, and Igor Kˇr´ıˇ z. 1994. Mixed Tate motives. Ann. of Math.
(2), 140(3), 557–605.
[15] Borel, Armand. 1977. Cohomologie de SL
n
et valeurs de fonctions zeta
aux points entiers. Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4), 4(4), 613–
636.
[16] Broadhurst, D. J., and D. Kreimer. 1997. Association of multiple zeta
values with positive knots via Feynman diagrams up to 9 loops. Phys.
Lett. B, 393(3–4), 403–412.
[17] Carlson, James A. 1987. The geometry of the extension class of a mixed
Hodge structure. Pp. 199–222 in Algebraic Geometry, Bowdoin, 1985
(Brunswick, Maine, 1985). Proc. Sympos. Pure Math., vol. 46. Provi
dence, R.I.: American Mathematical Society.
[18] Carlson, James A., and Richard M. Hain. 1989. Extensions of variations
of mixed Hodge structure. Ast´ erisque, 9, 39–65. Actes du Colloque de
Th´ eorie de Hodge (Luminy, 1987).
[19] Cattani, Eduardo, Aroldo Kaplan, and Wilfried Schmid. 1986. Degener
ation of Hodge structures. Ann. of Math. (2), 123(3), 457–535.
[20] Clemens, Herbert. 1983. Homological equivalence, modulo algebraic
equivalence, is not ﬁnitely generated. Inst. Hautes
´
Etudes Sci. Publ.
Math., 19–38 (1984).
[21] ColliotTh´ el` ene, JeanLouis. 1995. L’arithm´ etique des z´ erocycles (ex
pos´ e aux Journ´ ees arithm´ etiques de Bordeaux, septembre 93). Journal
de th´ eorie des nombres de Bordeaux, 51–73.
[22] Deligne, Pierre, and Alexander B. Goncharov. 2005. Groupes fondamen
taux motiviques de Tate mixte. Ann. Sci.
´
Ecole Norm. Sup. (4), 38(1),
1–56.
[23] Esnault, H´ el` ene. 2003. Varieties over a ﬁnite ﬁeld with trivial Chow
group of 0cycles have a rational point. Invent. Math., 151(1), 187–191.
[24] Fontaine, JeanMarc. 1992. Valeurs sp´ eciales des fonctions L des motifs.
Ast´ erisque, exp. no. 751, 4, 205–249. S´ eminaire Bourbaki, vol. 1991/92.
Preface to the second edition xxiii
[25] Fontaine, JeanMarc, and Bernadette PerrinRiou. 1994. Autour des con
jectures de Bloch et Kato: cohomologie galoisienne et valeurs de fonc
tions L. Pp. 599–706 in Motives (Seattle, 1991). Proc. Sympos. Pure
Math., vol. 55. Providence, R.I.: American Mathematical Society.
[26] Friedlander, Eric M. 2005. Motivic complexes of Suslin and Voevodsky.
Pp. 1081–1104 in Handbook of KTheory. Vol. 1, 2. Berlin: Springer.
[27] Goncharov, Alexander. 1999. Volumes of hyperbolic manifolds and mixed
Tate motives. J. Amer. Math. Soc., 12(2), 569–618.
[28] Goncharov, Alexander B. 2005. Regulators. Pp. 295–349 in Handbook
of KTheory. Vol. 1, 2. Berlin: Springer.
[29] Goodwillie, Thomas G. 1986. Relative algebraic Ktheory and cyclic
homology. Ann. of Math. (2), 124(2), 347–402.
[30] Hanamura, Masaki. 1995. Mixed motives and algebraic cycles. I. Math.
Res. Lett., 2(6), 811–821.
[31] Hanamura, Masaki. 1999. Mixed motives and algebraic cycles. III. Math.
Res. Lett., 6(1), 61–82.
[32] Hanamura, Masaki. 2000. Homological and cohomological motives of
algebraic varieties. Invent. Math., 142(2), 319–349.
[33] Jannsen, Uwe. 1994. Motivic sheaves and ﬁltrations on Chow groups.
Pp. 245–302 in Motives (Seattle, 1991). Proc. Sympos. Pure Math.,
vol. 55. Providence, R.I.: American Mathematical Society.
[34] Kahn, Bruno. 1996. Applications of weighttwo motivic cohomology.
Doc. Math., 1(17), 395–416.
[35] Krishna, Amalendu, and Marc Levine. 2008. Additive higher Chow
groups of schemes. J. Reine Angew. Math., 619, 75–140.
[36] Levine, Marc. 1998. Mixed Motives. Mathematical Surveys and Mono
graphs, vol. 57. Providence, R.I.: American Mathematical Society.
[37] Mazza, Carlo, Vladimir Voevodsky, and Charles Weibel. 2006. Lecture
Notes on Motivic Cohomology. Clay Mathematics Monographs, vol. 2.
Providence, R.I.: American Mathematical Society.
[38] Murre, J. P. 1993a. On a conjectural ﬁltration on the Chow groups of an
algebraic variety. I. The general conjectures and some examples. Indag.
Math. (N.S.), 4(2), 177–188.
[39] Murre, J. P. 1993b. On a conjectural ﬁltration on the Chow groups of an
algebraic variety. II. Veriﬁcation of the conjectures for threefolds which
are the product on a surface and a curve. Indag. Math. (N.S.), 4(2), 189–
201.
[40] Nori, Madhav V. 1993. Algebraic cycles and Hodgetheoretic connectiv
ity. Invent. Math., 111(2), 349–373.
xxiv Preface to the second edition
[41] Park, Jinhyun. 2007. Algebraic cycles and additive dilogarithm. Int.
Math. Res. Not. IMRN, Art. ID rnm067, 19.
[42] Rapoport, M., N. Schappacher, and P. Schneider (eds). 1988. Beilinson’s
Conjectures on Special Values of LFunctions. Perspectives in Mathe
matics, vol. 4. Boston: Academic Press.
[43] R¨ ulling, Kay. 2007. The generalized de RhamWitt complex over a ﬁeld
is a complex of zerocycles. J. Algebraic Geom., 16(1), 109–169.
[44] Soul´ e, Christophe. 1984. Kth´ eorie et z´ eros aux points entiers de fonc
tions zˆ eta. Pp. 437–445 in Proceedings of the International Congress of
Mathematicians, Vol. 1, 2 (Warsaw, 1983). Warsaw: PWN.
[45] Soul´ e, Christophe. 1986. R´ egulateurs. Ast´ erisque, 237–253. Seminar
Bourbaki, vol. 1984/85.
[46] Soul´ e, C., and C. Voisin. 2005. Torsion cohomology classes and algebraic
cycles on complex projective manifolds. Adv. Math., 198(1), 107–127.
[47] Totaro, Burt. 1997. Torsion algebraic cycles and complex cobordism. J.
Amer. Math. Soc., 10(2), 467–493.
[48] Totaro, Burt. 1999. The Chow ring of a classifying space. Pp. 249–281 in
Algebraic KTheory (Seattle, 1997). Proc. Sympos. Pure Math., vol. 67.
Providence, R.I.: American Mathematical Society.
[49] Voevodsky, Vladimir, Andrei Suslin, and Eric Friedlander. 2000. Cy
cles, Transfers, and Motivic Homology Theories. Annals of Mathematics
Studies, vol. 143. Princeton, N.J.: Princeton University Press.
[50] Voisin, Claire. 1994. Transcendental methods in the study of algebraic
cycles. Pp. 153–222 in Algebraic Cycles and Hodge Theory (Torino,
1993). Lecture Notes in Math., no. 1594. Berlin: Springer.
[51] Voisin, Claire. 2004. Remarks on ﬁltrations on Chow groups and the
Bloch conjecture. Ann. Mat. Pura Appl. (4), 183(3), 421–438.
LECTURES ON ALGEBRAIC CYCLES
0
Introduction
The notes which followare based on lectures given at Duke University in April,
1979. They are the fruit of ten years reﬂection on algebraic cycles; that is
formal linear combinations
_
n
i
[V
i
] of subvarieties V
i
of a ﬁxed smooth and
projective variety X, with integer coeﬃcients n
i
.
Classically, X was an algebraic curve (Riemann surface) and the V
i
were
points p
i
. In this context, cycles were a principal object of study for nineteenth
century complex function theory. Two (at least) of their results merit immor
tality: the theorems of Riemann–Roch and Abel–Jacobi.
The reader will recall that to a meromorphic function f on a smooth variety
or complex manifold one can associate a divisor (algebraic cycle of codimen
sion 1) ( f ) by taking the sum of its components of zeros minus its components
of poles, all counted with suitable multiplicity. An algebraic cycle
_
n
i
[V
i
] is
said to be eﬀective if all the n
i
≥ 0. The Riemann–Roch theorem computes the
dimension (D) of the vector space of functions f such that ( f ) + D is eﬀec
tive for a given divisor D. It has been generalized quite considerably in recent
years, but its central role in the study of divisors does not seem to carry over to
cycles of codimension greater than 1. (For example, one can use the Riemann–
Roch theorem to prove rationality of the zeta function of an algebraic curve
over a ﬁnite ﬁeld [9], but the corresponding theorem for varieties of dimension
> 1 lies deeper.)
The other great pillar of function theory on Riemann surfaces, the Abel–
Jacobi theorem, tells when a given divisor D =
_
n
i
(p
i
) is the divisor of a
function. Clearly a necessary condition is that the degree
_
n
i
must be zero,
so we may assume this and write our cycle D =
_
n
i
((p
i
) − (p
0
)) for some
base point p
0
. We denote by Γ(X, Ω
1
X
) the space of all global holomorphic
diﬀerential 1forms on X (where ω ∈ Γ(X, Ω
1
X
) can be written locally as f (z) dz
for z and f (z) holomorphic). Anecessary and suﬃcient condition for Dto equal
4 Lecture 0
( f ) is that
n
i
_
p
i
p
0
ω =
_
γ
ω
for some (topological) 1cycle γ ∈ H
1
(X, Z) and all ω ∈ Γ(X, Ω
1
X
). Even more
is true; writing Γ(X, Ω
1
X
)
∗
= Hom
C
(Γ(X, Ω
1
X
), C) for the periods, the map
A
0
(X)
divisors of degree 0
divisors of functions
→ Γ(X, Ω
1
X
)
∗
/H
1
(X, Z) := J(X)
is an isomorphism from A
0
(X) to the abelian variety J(X), where J(X) is the
jacobian of X.
To extend these results, we may deﬁne for X any smooth projective variety
over a ﬁeld and r ≥ 0 an integer
z
r
(X) = free abelian group generated by irreducible
subvarieties of X of codimension r.
Given A =
_
m
i
A
i
∈ z
r
(X) and B =
_
n
j
B
j
∈ z
s
(X) such that every compo
nent of A
i
∩ B
j
has codimension r + s for all i and j, there is deﬁned a product
cycle A B ∈ z
r+s
(X) obtained by summing over all components of A
i
∩ B
j
with suitable multiplicities. If f : X → Y is a proper map, where dimX = d,
dimY = e, one has f
∗
: z
r
(X) → z
r+e−d
(Y) deﬁned on a single irreducible
codimensionr subvariety V ⊂ X by f
∗
(V) = [k(V) : k( f (V))] f (V). Here
[k(V) : k( f (V))] is the degree of the extension of function ﬁelds. By conven
tion this degree is zero if the extension is not ﬁnite.
Suppose now Γ ∈ z
r
(X × P
1
) and no component of Γ contains either X × ¦0¦
or X × ¦∞¦. The cycle
pr
1∗
_
Γ (X × ((0) − (∞)))
_
∈ z
r
(X), pr
1
: X × P
1
→ X,
is then deﬁned. By deﬁnition, z
r
rat
(X) ⊂ z
r
(X) is the subgroup of cycles of this
form. Replacing P
1
by an arbitary (variable) smooth connected curve C, and
0, ∞by any two points a, b ∈ C one obtains a group z
r
alg
(X) with
z
r
rat
(X) ⊂ z
r
alg
(X) ⊂ z
r
(X).
We write
CH
r
(X) = z
r
(X)/z
r
rat
(X) (the Chow group),
A
r
(X) = z
r
alg
(X)/z
r
rat
(X),
CH
r
(X) = CH
d−r
(X),
A
r
(X) = A
d−r
(X), d = dimX.
Introduction 5
Cycles in z
r
rat
(X) (resp. z
r
alg
(X)) are said to be rationally (resp. algebraically)
equivalent to zero (written x ∼
rat
0 or x ∼
alg
0, or if no ambiguity is possible just
x ∼ 0). As an exercise, the reader might prove that when the ground ﬁeld k is
algebraically closed, the association
_
n
i
(p
i
) ·→
_
n
i
deﬁnes an isomorphism
CH
0
(X)/A
0
(X) Z. A bit more diﬃcult is to show
CH
1
(X)
divisors
divisors of functions
= divisor class group of X Pic(X),
where Pic(X) is the group of isomorphism classes of line bundles on X. Still
harder is to prove that when X has a rational point, A
1
(X) is isomorphic to the
group of kpoints of the Picard variety of X.
The purpose of these notes is to study algebraic cycles, particularly those
of codimension > 1 as it is here that really new and unexpected phenomena
occur. In keeping with the author’s philosophy that good mathematics “opens
out” and involves several branches of the subject, we will consider geometric,
algebraic and arithmetic problems.
The ﬁrst three lectures are geometric in content, studying various aspects
of the Abel–Jacobi construction in codimensions > 1. We work with varieties
X over the complex numbers, and deﬁne (following Griﬃths, Lieberman, and
Weil) compact complex tori (intermediate jacobians) J
r
(X) and Abel–Jacobi
maps
Θ: A
r
(X) → J
r
(X).
When r = d = dimX, J
r
(X) = J
0
(X) = Γ(X, Ω
1
X
)
∗
/H
1
(X, Z) is called the Al
banese variety and the Abel–Jacobi map is deﬁned precisely as with curves.
The map Θ: A
0
(X) → J
0
(X) is surjective, and in the ﬁrst lecture we study its
kernel T(X) when dimX = 2. We sketch an argument of Mumford that T(X)
0 (in fact T(X) is enormous) when P
g
(X) 0, i.e. when X has a nonzero
global holomorphic 2form. We give several examples motivating the conjec
ture that Γ(X, Ω
2
X
) actually controls the structure of T(X) and that in particular
T(X) = 0 ⇐⇒ Γ(X, Ω
2
X
) = 0. This conjecture can be formulated in various
ways. One vague but exciting possibility is that groups like T(X) provide a ge
ometric interpretation of the category of polarized Hodge structures of weight
two, in much the same way that abelian varieties do for weight 1.
The second and third lectures consider curves on threefolds. In Lecture 2
we focus on quartic threefolds, i.e. smooth hypersurfaces X of degree 4 in P
4
,
and verify in that case Θ is an isomorphism. Lecture 3 considers relative alge
braic 1cycles on X = C × P
1
× P
1
, where C is a curve. We deﬁne a relative
intermediate jacobian which turns out to be a noncompact torus isomorphic
6 Lecture 0
to H
1
(C, C
∗
). The machinery of cycle classes constructed in this lecture is re
lated in Lectures 8 and 9 to special values of Hasse–Weil zeta functions. Of
particular importance are the classes in H
1
(C, R) deﬁned by factoring out by
the maximal compact. We show such classes can be deﬁned for any (not nec
essarily relative) cycle on C × P
1
× P
1
.
Lectures 4 through 6 develop the algebraic side of the theory: the cohomol
ogy groups of the Ksheaves H
p
(X, 7
q
), the Gersten–Quillen resolution, and
analogues for singular and ´ etale cohomology theories. Of particular interest
are the cohomological formulae
CH
p
(X) H
p
(X, 7
p
)
and (for X deﬁned over C)
CH
p
(X)/A
p
(X) H
p
(X, 1
p
),
where 1
p
denotes the sheaf for the Zariski topology associated to the presheaf
U ⊂ X → H
p
(U, Z) (singular cohomology). These techniques are used to
prove a theorem of Roitman that A
0
(X)
tors
, the torsion subgroup of the zero
cycles, maps isomorphically to the torsion subgroup of the Albanese variety.
The heart of the proof is a result about the multiplicative structure of the Ga
lois cohomology ring of a function ﬁeld F of transcendence degree d over an
algebraically closed ﬁeld. We prove for prime to char F that the cup product
map
H
1
(F, Z/Z) ⊗ ⊗ H
1
(F, Z/Z)
.,,.
d times
→ H
d
(F, Z/Z)
is surjective. It would be of great interest both algebraically and geometrically
to know if the whole cohomology ring H
∗
(F, Z/Z) were generated by H
1
.
The last three lectures are devoted to arithmetic questions. In Lecture 7 we
consider A
0
(X), where X is a surface over a local or global ﬁeld k. We assume
the base extension of X to the algebraic closure of k is a rational surface. Using
a technique of Manin involving the Brauer group we show by example that
in general A
0
(X) 0. We prove that A
0
(X) is ﬁnite when X has a pencil of
genuszero curves (conic bundle surface). The key idea in the proof is a sort
of generalization of the Eichler norm theorem, describing the image of the
reduced norm map Nrd: A
∗
→ k(t)
∗
when A is a quaternion algebra deﬁned
over a rational ﬁeld in one variable over k.
Finally, Lectures 8 and 9 take up, from a numbertheoretic point of view, the
work of Lecture 3 on relative intermediate jacobians for curves on C×P
1
×P
1
.
We consider the case C = E = elliptic curve and compute explicitly the class
Introduction 7
in H
1
(E, R) associated to the curve
γ
f ,g
=
_
(x, f (x), g(x)) ¦ x ∈ E
_
for f , g rational functions on E. When the zeros and poles of f are points of ﬁ
nite order, we show how to associate to f and g an element in Γ(E, 7
2
) and how
to associate to an element in Γ(E, 7
2
) a relative algebraic 1cycle on E×P
1
×P
1
.
When E has complex multiplication by the ring of integers in an imaginary
quadratic ﬁeld of class number one, we construct an element U ∈ Γ(E
Q
, 7
2
)
such that the image under the Abel–Jacobi map into H
1
(E, R) C of the asso
ciated relative algebraic cycle multiplied by a certain simple constant (involv
ing the conductor of the curve and Gauss sum) is the value of the Hasse–Weil
zeta function of E at s = 2. Conjecturally for E deﬁned over a number ﬁeld
k, the rank of Γ(E
k
, 7
2
) equals the order of the zero of the Hasse–Weil zeta
function of E at s = 0.
I want to thank Duke University for ﬁnancial support, and my auditors at
Duke for their enthusiasm and tenacity in attending eight lectures in ten days.
I also want to acknowledge that much of the function theory of the diloga
rithm which underlines the calculations in the last two lectures was worked
out in collaboration with David Wigner, and that I never would have gotten
the damned constants in (9.12) correct without help from Dick Gross. (I may,
indeed, not have gotten them correct even with help.)
Finally, a pedagogical note. When an idea is already well documented in
the literature and extensive detail would carry us away from the focus of these
notes on algebraic cycles, I have been very sketchy. For example, Quillen’s
work on the foundations of Ktheory are magniﬁcently presented in his own
paper [10]. I only hope the brief outline given here will motivate the reader to
turn to that source. The reader may also ﬁnd the rapid treatment of the Mum
ford argument showing P
g
0 ⇒ T(X) 0 in Lecture 1 to be unsatisfactory.
To remedy this, another demonstration rather diﬀerent in spirit from Mum
ford’s is included as an appendix. Historically, the negative force of this result
led us all to conclude that, except in certain obvious cases such as ruled sur
faces, the structure of zerocycles on a surface was total chaos. I wanted to
devote time to various examples showing that this is not the case. In retro
spect, I have certainly “left undone those things which I ought to have done”
(e.g. Griﬃths’ proof that homological equivalence algebraic equivalence,
and Tate’s proof of the Tate conjecture for abelian varieties). I hope the reader
will spare me.
8 Lecture 0
References for Lecture 0
A list of references appears at the end of each lecture. What follows are refer
ences for foundational questions about cycles, etc.
[1] C. Chevalley et al., Anneaux de Chow et applications, S´ eminaire C. Che
valley, 2
e
ann´ ee, S´ ecr. Math. Paris (1958).
[2] W. L. Chow, On equivalence classes of cycles in an algebraic variety,
Ann. of Math. (2), 64, 450–479 (1956).
[3] A. Grothendieck et al., Th´ eorie des intersections et th´ eor` eme de Riemann–
Roch (SGA 6), Lecture Notes in Math., no. 225, Springer, Berlin (1971).
[4] W. Fulton, Rational equivalence on singular varieties, Inst. Hautes
´
Etudes
Sci. Publ. Math., no. 45 (1975), 147–167.
[5] W. Fulton and R. MacPherson, Intersecting cycles on an algebraic variety,
Aarhus Universitet Preprint Series, no. 14 (1976). [Pp. 179–197 in Real
and complex singularities (Proc. Ninth Nordic Summer School/NAVF Sym
pos. Math., Oslo, 1976), Sijthoﬀ and Noordhoﬀ, Alphen aan den Rijn
(1977).]
[6] P. Samuel, Rational equivalence of arbitrary cycles, Amer. J. Math., 78
(1956), 383–400.
[7] J.P. Serre, Alg` ebre locale. Multiplicit´ es, Lecture Notes in Math., no. 11,
Springer, Berlin (1965).
[8] A. Weil, Foundations of Algebraic Geometry, A.M.S. Colloquium Publi
cations, vol. 29, American Mathematical Society, Providence, R.I. (1946).
[9] A. Weil, Courbes Alg´ ebriques et Vari´ et´ es Ab´ eliennes, Hermann, Paris
(1971).
[10] D. Quillen, Higher algebraic Ktheory. I, pp. 85–147 in Algebraic K
Theory I, Lecture Notes in Math., no. 341, Springer, Berlin (1973).
1
Zerocycles on surfaces
In this ﬁrst section I want to consider the question of zerocycles on an alge
braic surface from a purely geometric point of view. I will consider a number
of explicit examples, and give a heuristic description of a result of Mumford
[4] that P
g
0 implies A
0
(X) is “very large”. In particular A
0
(X) is not an
abelian variety in this case. Finally I will discuss some conjectures motivated
by these ideas.
For purposes of this lecture, algebraic surface X will mean a smooth pro
jective variety of dimension 2 over the complex numbers (or, if you prefer, a
compact complex manifold of dimension 2 admitting a projective embedding).
The space of global holomorphic iforms (i = 1, 2) will be written Γ(X, Ω
i
X
)
(or H
i,0
) and we will write
q = dimΓ(X, Ω
1
X
), P
g
= dimΓ(X, Ω
2
X
).
If γ is a topological 1chain on X, the integral over γ,
_
γ
, is a welldeﬁned
element in the dual Cvector space Γ(X, Ω
1
X
)
∗
. If C is a 2chain,
_
∂C
ω =
_
C
dω (Stokes’)
so
_
∂C
= 0 in Γ(X, Ω
1
X
)
∗
, since global holomorphic forms are closed. Hence
we may map H
1
(X, Z) → Γ(X, Ω
1
X
)
∗
. It follows from the Hodgetheoretic de
composition H
1
(X, C) = H
0,1
⊕ H
1,0
that the quotient Γ(X, Ω
1
X
)
∗
/H
1
(X, Z) is
a compact complex torus, called the Albanese of X and written Alb(X). This
torus admits a polarization satisfying the Riemann bilinear relations and hence
is an abelian variety.
Fixing a base point p
0
∈ X, we deﬁne φ: X → Alb(X) by φ(p) =
_
p
p
0
.
Note that this is well deﬁned: two paths from p
0
to p diﬀer by an element in
H
1
(X, Z). Notice, ﬁnally, that the above discussion is equally valid for smooth
projective varieties X of arbitrary dimension.
10 Lecture 1
Proposition (1.1)
(i) Let φ
n
: X × × X
.,,.
n times
→ Alb(X) be deﬁned by φ
n
(x
1
, . . . , x
n
) = φ(x
1
) +
+ φ(x
n
). Then for n > 0, φ
n
is surjective.
(ii) Let ψ: X → A be a map from X to a complex torus, and assume ψ(p
0
) = 0.
Then there exists a unique homomorphism θ: Alb(X) → A such that the
diagram
X
φ
ψ
Alb(X)
A
θ
commutes.
(iii) The map z
0
(X)
φ
→Alb(X), where (x) ·→ φ(x), factors through the map
φ: CH
0
(X) → Alb(X). The induced map φ: A
0
(X) → Alb(X) is surjec
tive, and independent of the choice of base point p
0
.
Proof These results are more or less well known. The reader who is unfamil
iar with them might try as an exercise to ﬁnd proofs. (Hint: In (i) consider the
question inﬁnitesimally and use the fundamental theorem of calculus to calcu
late the derivative of
_
p
p
0
ω. For (iii), reduce the question to showing that a map
P
1
→complex torus is necessarily constant.)
Example (1.2) (Bloch et al. [2]) Let E and F be elliptic curves (Riemann
surfaces of genus 1). We propose to calculate the Chow group of the quotient
surface X = (F × E)/¦1, σ¦, where σ is a ﬁxedpointfree involution on F × E
obtained by ﬁxing a point η ∈ E of order 2, η 0, and taking σ( f , e) =
(−f , e+η). Let E
·
= E/¦1, η¦. There is a natural map ρ : X → E
·
with all ﬁbres
of ρ F.
Notice
Γ(X, Ω
1
X
) Γ(F × E, Ω
1
F×E
)
¦1,σ¦
_
Γ(E, Ω
1
E
) ⊕ Γ(F, Ω
1
F
)
_
¦1,σ¦
Γ(E, Ω
1
E
) C
since the automorphism f → −f acts by −1 on Γ(F, Ω
1
F
). We conclude Alb(X)
has dimension 1. Since Alb(X) → E
·
and the ﬁbres of ρ are connected, it
follows that Alb(X) E
·
.
Lemma (1.3) Let Y be a smooth quasiprojective variety, n > 0 an integer.
Then A
n
(X) is a divisible group.
Zerocycles on surfaces 11
Proof By deﬁnition A
n
(X) ⊂ CH
n
(X) is generated by images under corre
spondences from jacobians of curves. Since jacobians of curves are divisible
groups, the lemma follows.
Lemma (1.4) Let Y be a smooth projective variety, and let
T(Y) = Ker (A
0
(Y) → Alb(Y)).
Then T(Y) is divisible.
Proof For any abelian group A, let
N
A ⊂ A be the kernel of multiplication by
N. From the divisibility of A
0
(Y) one reduces to showing
N
A
0
(Y)
N
Alb(Y)
for any N. If Y is a curve, the Chow group and Albanese both coincide with the
jacobian so
N
A
0
(Y)
N
Alb(Y). It will suﬃce therefore to assume dimY > 1
and show
N
Alb(W)
N
Alb(Y) for W ⊂ Y a smooth hyperplane section.
As a real torus, Alb(Y) can be identiﬁed with H
1
(Y, R/Z) and
N
Alb(Y) =
H
1
(Y, Z/NZ), so the question becomes the surjectivity of H
1
(W, Z/NZ) →
H
1
(Y, Z/NZ) or the vanishing of H
1
(Y, W; Z/NZ). This group is identiﬁed by
duality with H
2dimY−1
(Y − W, Z/NZ) = (0) because the Stein variety Y −W has
cohomological dimension = dimY.
We now return to our surface X = F × E/¦1, σ¦.
Claim A
0
(X) E
·
Alb(X). That is, T(X) = 0.
Proof We work with the diagram
F × E
π
E
X E
·
.
If z ∈ T(X) we may write
π
∗
z =
r
i
_
(q
i
, p
i
) + (−q
i
, p
i
+ η)
_
,
where
_
r
i
= 0 and
_
2 r
i
p
i
= 0 in E. By Abel’s theorem
2 (q, p) ∼ 2 (q, p + η) on F × E,
(q, p) + (−q, p) ∼ 2 (0, p),
so
2 π
∗
z ∼
2 r
i
[(q
i
, p
i
) + (−q
i
, p
i
)] ∼
4 r
i
(0, p
i
)
= [(0) × E]
_
F ×
4 r
i
(p
i
)
_
∼ 0 .
12 Lecture 1
It follows that 0 ∼ 2π
∗
π
∗
z = 4z, and hence 4T(X) = 0. Since T(X) is divisible,
this implies T(X) = 0.
Example (1.5) (Inose and Mizukami [3]) Let Y : T
5
0
+T
5
1
+T
5
2
+T
5
3
= 0 in P
3
,
and let X = Y/(Z/5Z), where we identify Z/5Z with the group of ﬁfth roots of
1 and we let a ﬁfth root ω act on Y by (t
0
, t
1
, t
2
, t
3
) → (t
0
, ωt
1
, ω
2
t
2
, ω
3
t
3
). This
action is ﬁxed point free, so X is smooth.
Claim A
0
(X) = (0).
Proof Let π: Y → X be the projection. Since A
0
(X) is divisible, it suﬃces
to show 5A
0
(X) = π
∗
π
∗
A
0
(X) = (0). Since π
∗
is surjective, it suﬃces to show
π
∗
π
∗
: A
0
(Y) → A
0
(Y) is the zero map. We ﬁx a ﬁfth root of 1, ω 1, and let
Z/5Z
⊕3
act on Y by
e
∗
i
T
j
=
_
T
j
j i,
ωT
i
i = j,
i = 1, 2, 3.
This gives us a representation Z[Z/5Z
⊕3
] R ⊂ End(A
0
(Y)), which we
denote by x ·→ ¯ x. One checks immediately
(1.5.1) π
∗
π
∗
= 1 + ¯ e
1
¯ e
2
2
¯ e
3
3
+ (¯ e
1
¯ e
2
2
¯ e
3
3
)
2
+ + (¯ e
1
¯ e
2
2
¯ e
3
3
)
4
.
The reader can verify also that the quotient of Y by any of the cyclic groups
(e
1
e
2
e
3
, . . . , e
5
1
e
5
2
e
5
3
) or
(e
i
, e
2
i
, . . . , e
5
i
), i = 1, 2, 3 or
(e
i
e
j
, e
2
i
e
2
j
, . . . , e
5
i
e
5
j
), i j
is rational, leading to
0 =
4
n=0
¯ e
n
1
¯ e
n
2
¯ e
n
3
,
4
n=0
e
−n
i
= 0 =
4
n=0
¯ e
n
i
¯ e
n
j
in R ⊂ End (A
0
(Y)). (1.5.2)
Since A
0
(Y) is divisible, it will suﬃce to show π
∗
π
∗
(1.5.1) is trivial in R⊗Q.
Writing F = Q(ω), the lefthand identities in (1.5.2) show that R ⊗ Q is a
quotient of the semisimple ring F⊗
Q
F⊗
Q
F, so it will suﬃce to show that
π
∗
π
∗
goes to zero under any homomorphism R⊗Q →
¯
Q= algebraic closure of
Q. A homomorphism h: F ⊗ F ⊗ F →
¯
Q amounts to the choice of three non
trivial ﬁfth roots of 1, ω
1
, ω
2
, ω
3
. For h to factor through R ⊗ Q the righthand
identities in (1.5.2) force ω
i
ω
j
1, i j. On the other hand, for the image of
Zerocycles on surfaces 13
π
∗
π
∗
to be nontrivial, one must have ω
1
ω
2
2
ω
3
3
= 1. One checks easily that the
conditions
ω
1
ω
2
ω
3
1,
ω
i
1, ω
i
ω
j
1, ω
1
ω
2
2
ω
3
3
= 1, i j = 1, 2, 3
cannot all hold, so π
∗
π
∗
= 0. This proves the claim.
Remarks The surface in (1.2) is called hyperelliptic; that in (1.5) is a Godeaux
surface. The reader should be warned that the author has exercised his prerog
ative in selecting these examples very carefully. In particular, both examples
have P
g
= 0.
I want now to sketch Mumford’s idea [4] for showing that P
g
> 0 im
plies T(X) is enormous. For another argument, see the appendix to this lec
ture. Choose a base point p
0
∈ X and deﬁne a map ρ
n
from the nth sym
metric product S
n
X = set of unordered ntuples of elements of X to A
0
(X),
ρ
n
: S
n
X → A
0
(X), where ρ
n
(x
1
, . . . , x
n
) =
_
(x
i
) − n(p
0
). Notice if T(X) = 0,
so A
0
(X) Alb(X), the ﬁbres of ρ
n
would be subschemes of S
n
X of codimen
sion ≤ q = dimAlb(X). Mumford proves
Theorem (1.6) Assume P
g
> 0 and let t ∈ S
n
X be a general point. Assume
t ∈ W ⊂ S
n
X is a subscheme on which ρ
n
is constant; that is, ρ
n
(W) = ρ
n
(t).
Then the codimension of W in S
n
X is ≥ n.
Idea of proof Let ω be a nonzero global holomorphic 2form on X. By pull
ing back along the various projections and adding up, one gets a 2form
˜ ω
n
= pr
∗
1
ω + + pr
∗
n
ω
on the cartesian product X
n
= X × × X. The form ˜ ω
n
is invariant under the
action of the symmetric group S
n
on X
n
and so in a certain sense ˜ ω
n
descends
to a form ω
n
on S
n
X = X
n
/S
n
. (This part of the argument requires considerable
care because S
n
X will have singularities.) Quite precisely, if T is a nonsingular
variety parameterizing an eﬀective family of zerocycles of degree n a general
element of which consists of n distinct points with multiplicity 1, there will be a
“pullback” ω
n,T
∈ Γ(T, Ω
2
T
) deﬁned. Note that the singularities of S
n
X occur at
points (x
1
, . . . , x
n
) where two or more of the x
i
coincide. On the complement
(S
n
X)
smooth
of the singular set, ω
n
is well deﬁned. There will be an open set
T
0
⊂ T and a morphism T
0
ψ
→ (S
n
X)
smooth
, and ω
n,T
will be a holomorphic
extension of ψ
∗
ω
n
.
Using the deﬁnition of rational equivalence and the fact that there are no
global holomorphic forms on projective space, Mumford shows that if the
14 Lecture 1
cycles in the family parameterized by T are all rationally equivalent, then
ω
n,T
= 0. One next notices that if t ∈ (S
n
X)
smooth
is general, the twoform ω
n
will give a nondegenerate alternating pairing on the tangent space at t. Shifting
t along W, our given subvariety, if necessary we may assume t is a nonsingular
point of W. Resolving singularities of W then does not change the situation
near t, but does show that our nondegenerate pairing, when restricted to the
tangent space of W at t, gives zero. It follows that dimW ≤
1
2
dimS
n
X = n, so
codim W ≥ n.
Example (1.7) (Fatemi [10]) A surface with P
g
> 0 and an interesting Chow
group is the Fano surface. Let T be a cubic threefold, that is, a smooth hyper
surface of degree 3 in P
4
. Let G be the Grassmann of lines in P
4
, and let S ⊂ G
be the subvariety of lines on T. The subvariety S is known to be a smooth
connected surface. We write Pic(S ) = CH
1
(S ), and Pic
0
(S ) = divisors alge
braically equivalent to zero modulo rational equivalence. Pic
0
(S ) is the group
of closed points of the Picard variety of S , and there is an exact sequence
0 → Pic
0
(S ) → Pic(S ) → NS(S ) → 0,
where NS(S ) = N´ eron–Severi group of S = subgroup of H
2
(S, Z) generated
by classes of divisors.
Intersection of divisors gives bilinear maps
Pic(S )⊗
Z
Pic
0
(S ) → A
0
(S ), (1.7.1)
Pic
0
(S )⊗
Z
Pic
0
(S ) → T(S ). (1.7.2)
(If D
0
is algebraically equivalent to zero and D is any divisor, D D
0
will
have degree 0. The fact that D D
0
∈ T(S ) if both D and D
0
are algebraically
equivalent to zero can be checked by noting that the map
Pic
0
(S ) × Pic
0
(S )
intersect
A
0
(S ) → Alb(S )
is a continuous bilinear map of compact tori, hence trivial (exercise!).)
Claim The intersection maps (1.7.1) and (1.7.2) are surjective.
Proof I will give the proof only for (1.7.1). The minor modiﬁcation needed
to prove (1.7.2) is left for the reader. Necessary facts about cubic 3folds can
be found in Clemens and Griﬃths [11] or Tyurin [12]. Let r, s ∈ S be general
points. It will suﬃce (using divisibility of A
0
(S )) to show that 2(r) − 2(s) ∈
Image(Pic
0
(S )⊗
Z
Pic(S ) →A
0
(S )). The line on T corresponding to s ∈ S will
be denoted
s
, and D
s
will denote the divisor on S
D
s
= Zariski closure of
_
t ∈ S ¦
t
∩
s
∅, t s
_
.
Zerocycles on surfaces 15
In general, s D
s
, and D
s
D
t
has degree 5. Thus, if r, s are general (“general,”
in this context, means the points in question do not lie on a ﬁnite number of
a priori speciﬁed proper closed subvarieties of S ) we can ﬁnd a general t such
that
s
∩
t
∅ and
t
∩
r
∅. Since 2((r) − (s)) = 2((r) − (t)) + 2((t) − (s)),
we may assume
r
∩
s
∅.
Let L be the plane spanned by
r
and
s
in P
4
and let
t
be the third line in
L ∩ T =
r
∪
s
∪
t
.
l
t
1
l
t
2
l
t
3
l
t
4
l
s
l
l
t
b
a
c
r
Through a general point of T there pass 6 lines, and we denote by θ: CH
0
(T) →
CH
0
(S ) the correspondence thus deﬁned. CH
0
(T) Z (any two points on T
can be connected by a chain of lines) so we have
θ(a) = (r) + (t) + (s
1
) + + (s
4
)
∼ θ(b) = (s) + (t) + (r
1
) + + (r
4
)
∼ θ(c) = (r) + (s) + (t
1
) + + (t
4
).
On the other hand
D
r
D
s
= (t) + (t
1
) + + (t
4
),
D
s
D
t
= (r) + (r
1
) + + (r
4
),
D
r
D
t
= (s) + (s
1
) + + (s
4
).
Hence
θ(a) − D
r
D
t
= (r) + (t) − (s),
θ(b) − D
s
D
t
= (s) + (t) − (r),
(D
s
− D
r
) D
t
∼ θ(a) − D
r
D
t
− θ(b) + D
s
D
t
= 2(r) − 2(s).
16 Lecture 1
Philosophy The dual to Γ(S, Ω
2
S
) is the second cohomology group of the
structure sheaf, H
2
(S, O
S
) (Serre duality). The tangent space to Pic
0
(S ) is
H
1
(S, O
S
). In the case of the Fano surface, the cup product map induces an
isomorphism Λ
2
H
1
(S, O
S
) H
2
(S, O
S
). In some mystical sense, the control
exercised by the geometric genus P
g
over the size of T(S ) can be seen in the
following schema:
Pic
0
(S ) ⊗ Pic
0
(S )
tang. space tang. space
intersection
T(S )
bitang. space
H
1
(S, O
S
) ⊗ H
1
(S, O
S
)
cup product
H
2
(S, O
S
) .
The idea of a bitangent space to T(S ) will be discussed further in Lecture 6.
I want to end with a conjecture, motivated in part by these examples. For
any smooth projective variety V, we may deﬁne at least the beginnings of a
ﬁltration on CH
0
(V) by
F
1
CH
0
(V) = A
0
(V) = Ker (deg: CH
0
(V) → Z),
F
2
CH
0
(V) = Ker (F
1
CH
0
(V) → Alb(V)).
Presumably the ﬁltration will in general have n steps where n = dimV. Except
in certain simple cases like abelian varieties or products of curves (Bloch [13]),
I don’t have much feeling for what should come beyond F
2
. We will therefore
“truncate” at F
2
.
Let X be a surface, and let Γ be a cycle on V × X with dimΓ = n = dimV.
The cycle Γ then induces a map
Γ
∗
: CH
0
(V) → CH
0
(X) .
The above ﬁltration being functorial for correspondences, we get also
gr Γ
∗
: gr CH
0
(V) → gr CH
0
(X) Z ⊕ Alb(X) ⊕ T(X).
Conjecture (1.8) The map gr Γ
∗
depends only upon the cohomology class
[Γ] ∈ H
4
(V × X).
I want to discuss some consequences of this conjecture, and to try to make it
more precise. Let Γ(i, j) ∈ H
i
(V) ⊗ H
j
(X) denote the K¨ unneth components of
Γ (since the graded pieces of CH
0
are either divisible or torsion free, we may
as well work with rational cohomology). Because X is a surface, one knows
Zerocycles on surfaces 17
(cf. Kleiman [15], 2A10 and 2.9) that the Γ(i, j) are algebraic. Altering Γ by
constant correspondences, we may suppose Γ(0, 4) = Γ(4, 0) = 0.
Lemma (1.9) Assuming conjecture (1.8), Γ(3, 1)
∗
: gr CH
0
(V) → gr CH
0
(X)
is zero. The map Γ(1, 3)
∗
is zero on F
2
CH
0
(V).
Proof Let q = dimAlb(X). A Poincar´ e divisor D of K¨ unneth type (1,1) on
X × Pic
0
(X) induces via correspondences an isomorphism
D
∗
: H
2q−1
(Pic
0
(X))
H
1
(X).
The inverse of this isomorphism will also be induced by an algebraic corre
spondence (Kleiman [15], 3.11), so we can factor Γ(3, 1)
∗
via an algebraic
correspondence Σ ∈ H
3
(V) ⊗ H
2q−1
(Pic
0
(X)). Such a Σ is represented by an
algebraic cycle on V × Pic
0
(X) of dimension dimV − 1, so on the cycle level
Σ
∗
CH
0
(V) = (0). To study Γ(1, 3)
∗
let j : C → X be a smooth hyperplane
section. The map Jacobian(C) → Alb(X) is split up to isogeny (Poincar´ e re
ducibility). Since H
3
(X) H
2q−1
(Alb(X)) one shows that there exists an alge
braic correspondence ψ ∈ H
1
(V) ⊗ H
1
(C) such that (1 × j
∗
) ψ = Γ(1, 3). Since
F
2
CH
0
(C) = (0), this implies Γ(1, 3)
∗
F
2
CH
0
(V) = (0).
From the point of view of cycles, the interesting correspondences to study
therefore are those of K¨ unneth type (2, 2). Looking at an example like (1.7)
one might be tempted to formulate the fantastically vague:
Metaconjecture (1.10) There is an equivalence of categories between a suit
able category of polarized Hodge structures of weight 2 and a category built
up from the F
2
CH
0
(X).
Such an equivalence might provide a geometric interpretation of the cate
gory of Hodge structures of weight 2 analogous to the abelian variety interpre
tation for weight 1!
I want ﬁnally to record two other consequences of (1.8).
Proposition (1.11) Assume conjecture (1.8) holds, and let X be a surface of
geometric genus 0. Then F
2
CH
0
(X) = (0).
18 Lecture 1
Proof Let ∆ ⊂ X×X be the diagonal. Then ∆
∗
= ∆(2, 2)
∗
= Id on F
2
CH
0
(X).
Since X has geometric genus 0, H
2
(X) is generated by divisors, so ∆(2, 2) =
_
a
i j
D
i
× D
j
where the D
i
are divisors. But a zerocycle can be moved away
from a ﬁnite union of divisors, so ∆(2, 2)
∗
: CH
0
(X) → CH
0
(X) is zero.
Proposition (1.12) (Conjectural theorem of the hypersquare) Assume con
jecture (1.8) holds. Let X be a surface and V = V
1
×V
2
×V
3
, with all varieties
smooth and connected. Let Γ be a codimension2 cycle on V×X and let v
i j
∈ V
i
be points, i = 1, 2, 3 and j = 1, 2. Then
Γ
∗
_
j,k,=1,2
(−1)
j+k+
(v
i j
, v
2k
, v
3
)
_
= 0 in CH
0
(X).
Proof View Γ as a family ¦Γ
t
¦
t∈V
3
of cycles on V
1
× V
2
× X parameterized by
V
3
. The Γ
t
are clearly all homologous, so
Γ
v
31
∗
= Γ
v
32
∗
: F
2
CH
0
(V
1
× V
2
) → F
2
CH
0
(X).
But the cycle
_
j,k=1,2
(−1)
j+k
(v
i j
, v
2k
) is an element of F
2
CH
0
(V
1
× V
2
) (exer
cise!). The desired identity is now straightforward.
For further discussion of this sort of conjecture, the reader can see Bloch [14].
References for Lecture 1
For general facts about algebraic surfaces over C, a good reference is
[1] A. Beauville, Surfaces alg´ ebriques complexes, Ast´ erisque, 59 (1978),
as well as the references cited there.
Example (1.2) is done more generally in
[2] S. Bloch, A. Kas, and D. Lieberman, Zero cycles on surfaces with P
g
= 0,
Compositio Math., 33 (1976), 135–145,
along with other surfaces with P
g
= 0 and Kodaira dimension < 2.
Example (1.5) is due to H. Inose and M. Mizukami. Details and other analo
gous examples are in
[3] H. Inose and M. Mizukami, Rational equivalence of 0cycles on some
surfaces of general type with p
g
= 0, Math. Ann., 244 (1979), no. 3,
205–217.
Zerocycles on surfaces 19
To understand the role of the geometric genus, the reader should see the semi
nal paper of Mumford:
[4] D. Mumford, Rational equivalence of zerocycles on surfaces, J. Math.
Kyoto Univ., 9 (1968), 195–204,
as well as
[5] A. A. Roitman, Γequivalence of zerodimensional cycles (in Russian),
Mat. Sb. (N.S.), 86 (128) (1971), 557–570. [Translation: Math USSRSb.,
15 (1971), 555–567.]
[6] A. A. Roitman, Rational equivalence of zerodimensional cycles (in Rus
sian), Mat. Sb. (N.S.), 89 (131) (1972), 569–585, 671. [Translation: Math.
USSRSb., 18 (1974), 571–588.]
[7] A. Mattuck, Ruled surfaces and the Albanese mapping, Bull. Amer. Math.
Soc., 75 (1969), 776–779.
[8] A. Mattuck, On the symmetric product of a rational surface, Proc. Amer.
Math. Soc., 21 (1969), 683–688.
For a more positive point of view, see
[9] S. Bloch, K
2
of Artinian Qalgebras with application to algebraic cycles,
Comm. Algebra, 3 (1975), 405–428.
For a time it was believed that Mumford’s technique could be applied to pluri
canonical diﬀerentials, and that one could in this way show A
0
(X) was not rep
resentable for any nonruled surface. A number of such statements appeared
in the literature. The argument fails because the trace of a pluricanonical form
under a ﬁnite ramiﬁed map can acquire singularities.
Example (1.7) was worked out by Fatemi.
[10] T. Fatemi, L’equivalence rationelle des z´ ero cycles sur les surfaces al ´ ge
briques complexes a cup product surjectif, These du 3
e
cycle, Universit´ e
de Paris VII (1979).
For more details about the geometry of the Fano surface and the cubic three
fold, see
[11] C. H. Clemens and P. A. Griﬃths, The intermediate Jacobian of the cubic
threefold, Ann. of Math. (2), 95 (1972), 281–356.
[12] A. N. Tyurin, Five lectures on threedimensional varieties (in Russian),
Uspehi Mat. Nauk, 27 (1972), no. 5, (167) 3–50. [Translation: Russian
Math. Surveys, 27 (1972), no. 5, 1–53.]
20 Lecture 1
Some references for the conjectures at the end of Lecture 1 are:
[13] S. Bloch, Some elementary theorems about algebraic cycles on abelian
varieties, Invent. Math., 37 (1976), 215–228.
[14] S. Bloch, An example in the theory of algebraic cycles, pp. 1–29 in Alge
braic KTheory, Lecture Notes in Math., no. 551, Springer, Berlin (1976).
[15] S. Kleiman, Algebraic cycles and the Weil conjectures, pp. 359–386 in
Dix expos´ es sur la cohomologie des sch´ emas, North Holland, Amsterdam
(1968).
Appendix: On an argument of Mumford in the
theory of algebraic cycles
S. Bloch
1
Let X be a smooth projective surface over an algebraically closed ﬁeld k. Let
CH
0
(X) denote the Chow group of zero cycles modulo rational equivalence on
X, and let A
0
(X) ⊂ CH
0
(X) be the subgroup of cycles of degree 0. I will say
that A
0
(X) is ﬁnite dimensional if there exists a complete smooth (but possibly
disconnected) curve C mapping to X such that the map J(C) = Jacobian(C) →
A
0
(X) is surjective. Some years ago, Mumford proved, in the case k = C, that
P
g
(X) > 0 implies A
0
(X) is not ﬁnite dimensional. The purpose of the present
note is to prove an analog of this result applicable in all characteristics. The
role of the geometric genus, which is not a good invariant in characteristic p,
is played by the “transcendental part” of H
2
et
(X, Q
). The present proof also
reveals the inﬂuence of the ﬁnite dimensionality of the Chow hypothesis on
the structure of the “motive” of X.
The idea that one could deduce interesting information about the Chow
group by considering the generic zerocycle was suggested by ColliotTh´ el` ene.
I am indebted to him for letting me steal it.
Lemma (1A.1) Let X be a smooth variety over an algebraically closed ﬁeld
k, and Y any kvariety. Let n ≥ 0 be an integer. Then, writing K = k(Y),
CH
n
(X
K
) lim
−−→
U⊂Y open
CH
n
(X ×
k
U),
where CH
n
equals codimensionn cycles modulo rational equivalence.
Proof For any variety W and any integer m, let W
m
= set of points of codi
1
The author gratefully acknowledges support from the NSF and from the conference on Chow
groups of rational varieties. The present note grew out of conversations with ColliotTh´ el` ene
at the conference.
22 Lecture 1
mension m on W. One has
(X
K
)
m
= lim
←−−
U
(X ×
k
U)
m
.
Since
CH
n
(X
K
) = Coker
_
_
x∈(X
K
)
n−1
K(x)
∗
−→
_
x∈(X
k
)
n
Z
_
,
CH
n
(X × U) = Coker
_
_
y∈(X×U)
n−1
k(x)
∗
−→
_
y∈(X×U)
n
Z
_
,
the desired result is immediate.
Proposition (1A.2) Let X be a smooth projective surface over k and
let Ω ⊃ k be a universal domain in the sense of Weil. Assume A
0
(X
Ω
)
is ﬁnite dimensional. Then there exist onedimensional subschemes
C
·
, C
··
⊂ X and a 2cycle Γ supported on (C
·
× X) ∪ (X × C
··
) such that
some nonzero multiple of the diagonal ∆ on X ×
k
X is rationally equivalent
to Γ.
Proof Let C → X be such that J(C
Ω
) A
0
(X
Ω
), and let C
·
⊂ X be the
image of C. Enlarging k, we may assume C
·
deﬁned over k.
Lemma (1A.3) Let k ⊂ K ⊂ K
·
be extensions of ﬁelds. Then the kernel of
CH
2
(X
K
) → CH
2
(X
K
· ) is torsion.
Proof If [K
·
: K] < ∞this follows from the existence of a norm CH
2
(X
K
· ) →
CH
2
(X
K
). The case K
·
algebraic over K follows by a limit argument. Enlarging
K and K
·
, we may thus assume K algebraically closed. In this case, CH
2
(X
K
· )
is a limit of Chow groups CH
2
(X ×
K
U), where U is a Kvariety of ﬁnite type.
A Kpoint of U gives a section of CH
2
(X) → CH
2
(X ×
K
U) so the lemma
follows.
Proof of proposition (1A.2) Let K be the function ﬁeld of X over k, and ﬁx an
embedding K → Ω. Let P ∈ X(Ω) be the corresponding point. Our hypotheses
imply CH
2
((X − C
·
)
Ω
) = (0), so by Lemma (1A.3), there exists N ≥ 1 such
that N(P) = 0 in CH
2
((X −C
·
)
K
). (We abuse notation by writing P also for the
generic point of X. In other words, P is the image in CH
2
(X
K
) of the diagonal
∆ in X × X.) By Lemma (1A.1), there exists U ⊂ X open ∅ such that N ∆ is
rationally equivalent to zero on (X − C
·
)×
K
U. Let C
··
⊂ X be a subscheme of
codimension 1 containing X − U. The proposition now follows from the exact
Zerocycles on surfaces 23
sequence
_
cycles supported on
(C
·
× X) ∪ (X × C
··
)
_
→ CH
2
(X ×
k
X)
→ CH
2
((X − C
·
)×
k
(X − C
··
)) → 0.
Exercise (1A.4) Generalize the deﬁnition of ﬁnite dimensionality for A
0
(X)
to varieties X of dimension > 2 and prove the analogue of Proposition (1A.2).
Fix now an prime to char p, and deﬁne
H
2
(X)
trans
= H
2
et
(X, Q
)/Image(NS(X) ⊗ Q
−→ H
2
et
(X, Q
)),
where NS(X) is the N´ eron–Severi group. We systematically ignore twisting by
roots of 1. An alternative description of H
2
(X)
trans
is given by
H
2
(X)
trans
= Image(H
2
et
(X, Q
) → H
2
Gal
(
¯
K/K, Q
)).
Note H
2
et
(X), NS(X), and hence also H
2
(X)
trans
are functorial for correspon
dences, which we will take to be elements in CH
2
(X ×
k
X). In particular, the
diagonal induces the identity on H
2
(X)
trans
.
Lemma (1A.5) With notation as in Proposition (1A.2), let Γ be a codimension
2 cycle on X × X supported on (C
·
× X) ∪ (X ×C
··
). Then the correspondence
Γ
∗
: H
2
(X)
trans
→ H
2
(X)
trans
is zero.
Proof It is convenient to work with ´ etale homology, which is deﬁned for any
kscheme Y which can be embedded in a smooth kvariety. Write Γ = Γ
·
+ Γ
··
with Supp Γ
·
⊂ C
·
× X and Supp Γ
··
⊂ X × C
··
. Let i
·
: C
·
× X → X × X,
i
··
: X ×C
··
→ X ×X, and write [Γ
·
] ∈ H
4
(C
·
×X, Q
), [Γ
··
] ∈ H
4
(X ×C
··
, Q
).
For α ∈ H
2
et
(X, Q
), we have
Γ
·
∗
α = pr
2∗
(pr
∗
1
(α) i
·
∗
[Γ
·
]) = pr
2∗
i
·
∗
(i
·∗
pr
∗
1
(α) Γ
·
)
= pr
2∗
i
·
∗
(pr
∗
1
i
·∗
(α) Γ),
with morphisms labeled as indicated:
C
·
× X
i
·
pr
1
X × X
pr
1
C
·
i
·
X.
Note that NS(X) ⊗Q
is selfdual under the cup product pairing on H
2
et
(X, Q
).
Without changing the image of α in H
2
(X)
trans
we may assume, therefore, that
α is perpendicular to NS(X). Since H
2
(C
·
, Q
) is generated by the classes of
components of C
·
, we ﬁnd i
·∗
(α) = 0 in H
2
(C
·
, Q
), so Γ
·
∗
(α) = 0.
24 Lecture 1
It remains to show Γ
··
∗
(α) = 0 where Γ
··
is supported on X × C
··
. We have a
diagram
X × C
··
i
··
pr
2
X × X
pr
2
C
··
i
··
X
and the projection formula gives
Γ
··
∗
α = pr
2∗
(i
··
∗
Γ
··
pr
∗
1
α)
= i
··
∗
pr
2∗
(Γ
··
i
··∗
pr
∗
1
α)
∈ Image(H
2
(C
··
, Q
) → H
2
(X, Q
)) ⊂ NS(X) ⊗ Q
,
so Γ
··
∗
α → 0 in H
2
(X)
trans
.
We have proven:
Theorem (1A.6) Let X be a smooth projective surface over an algebraically
closed ﬁeld k, and let k ⊂ Ωbe a universal domain. If H
2
et
(X, Q
) NS(X)⊗Q
( char k), then A
0
(X
Ω
) is not ﬁnite dimensional.
Exercise (1A.7) Formulate and prove an analogous result for varieties of
dimension > 2.
Question (1A.8) If E is a supersingular elliptic curve over a ﬁeld k of char
acteristic 0, , then H
2
et
(E × E, Q
) = NS(E × E) ⊗ Q
. Is A
0
(E × E) ﬁnite
dimensional?
When k = C, H
2
et
(X) NS(X) ⊗ Q
if and only if P
g
(X) > 0, so we recover
Mumford’s result:
Corollary (1A.9) When k = C, P
g
(X) > 0 implies A
0
(X
C
) is not ﬁnite dimen
sional.
References for Lecture 1 Appendix
D. Mumford, Rational equivalence of 0cycles on surfaces, J. Math. Kyoto
Univ., 9 (1968), 195–204.
A. A. Roitman, Rational equivalence of zerodimensional cycles (in Russian),
Mat. Sb. (N.S.), 89 (131) (1972), 569–585, 671. [Translation: Math. USSRSb.,
18 (1974), 571–588.]
2
Curves on threefolds and intermediate jacobians
The purpose of this lecture is to give some feeling for the geometry of curves
on threefolds over the complex numbers; in particular the link with intermedi
ate jacobians. We will focus on the example of a quartic threefold, a smooth
hypersurface X in P
4
of degree 4 and will show in this case that A
2
(X) is iso
morphic to the intermediate jacobian J
2
(X). (In all honesty the proof will not
be quite general, as we shall suppose for technical simplicity that the curve of
lines on X is smooth. This is false, for example, for the Fermat quartic; see Ten
nison [9]. No one has worked out the details in the general case, which would
involve jacobians of singular curves. It seems likely the ﬁnal result would be
the same.)
The intermediate jacobians to which these cycles are related can be de
scribed as follows. Let X be a smooth projective variety and r > 0 an integer.
The complex cohomology H
2r−1
(X, C) has a Hodge ﬁltration (Deligne [5])
H
2r−1
(X, C) = F
0
H
2r−1
⊃ F
1
⊃ ⊃ F
2r−1
⊃ (0).
This complex vector space also has a Zstructure deﬁned by the image of
H
2r−1
(X, Z) → H
2r−1
(X, C). In particular, it has an Rstructure, so it makes
sense to talk about conjugating an element or a subspace. The key property of
the Hodge ﬁltration is
F
i
⊕ F
2r−i
H
2r−1
(X, C),
F
i
∩ F
2r−i−1
H
i,2r−i−1
,
where
H
i, j
H
j
(X, Ω
i
X
).
26 Lecture 2
In particular, we see from this that
dim
C
F
r
=
1
2
dim
C
H
2r−1
(X, C),
F
r
∩ Image(H
2r−1
(X, Z) → H
2r−1
(X, C)) = (0),
so the quotient
J
r
(X) = H
2r−1
(X, C)/F
r
+ H
2r−1
(X, Z)
is a compact complex torus, called the intermediate jacobian.
If dimX = n, Poincar´ e duality implies
H
2r−1
(X, C)/F
r
(F
n−r+1
H
2n−2r+1
(X, C))
∗
,
where
∗
denotes Clinear dual. It follows that
J
r
(X) (F
n−r+1
H
2n−2r+1
(X, C))
∗
/H
2n−2r+1
(X, Z).
For example, if r = n we ﬁnd
J
n
(X) Γ(X, Ω
1
X
)
∗
/H
1
(X, Z) Alb(X).
The next step is to deﬁne cycle classes in these intermediate jacobians. In
this lecture we restrict ourselves to a concrete intuitive description valid for the
case at hand.
Let X be a smooth projective threefold, and let γ =
_
n
i
γ
i
be an algebraic
1cycle on X. We assume γ is homologous to 0, that is [γ] = 0 in H
4
(X, Z),
and the idea will be to mimic the construction in Lecture 1 of X → Alb(X) by
associating to (p) − (p
0
) the integral
_
p
p
0
∈ Γ(X, Ω
1
X
)
∗
/ H
1
(X, Z).
We choose a 3chain ∆ on X with ∂∆ = γ and consider
_
∆
as an element in
the dual space to the space of 3forms of types (3, 0) + (2, 1). (We are thinking
of J
2
(X) = F
2
H
3
(X, C)
∗
/H
3
(X, Z).) The ﬁrst point to note is that if ω is of type
(3, 0) +(2, 1) and ω is exact, then we can write ω = dη, where η has type (2, 0).
(This is an easy consequence of Hodge theory; see Griﬃths and Harris [7].)
Stokes’ theorem gives
_
∆
ω =
_
γ
η = 0
because there are no (2, 0) forms on a curve. Notice ﬁnally if ∆
·
is another
3chain with ∂∆
·
= ∂∆ = γ, then ∆
·
− ∆ is a 3cycle so
_
∆
·
−
_
∆
∈ Image (H
3
(X, Z) → F
2
H
3
(X, C)
∗
).
Curves on threefolds and intermediate jacobians 27
These remarks suﬃce to deﬁne a cycle class
j(γ) ∈ J
2
(X).
Recall that A
2
(X) is the group of codimension2 cycles algebraically equiv
alent to zero, modulo rational equivalence. Since a cycle algebraically equiva
lent to zero is certainly homologous to zero, one can hope for a map
Θ: A
2
(X) → J
2
(X), Θ(γ) = j(˜ γ),
where ˜ γ is a codimension2 cycle representing γ ∈ A
2
(X).
Lemma (2.1) The above map Θ is well deﬁned. That is, given a family
¦Γ
t
¦
t∈P
1 of codimension2 cycles parameterized by P
1
, j(Γ
t
) ∈ J
2
(X) is inde
pendent of t.
Proof It is possible to show that t ·→ j(Γ
t
) deﬁnes a holomorphic function
P
1
→ J
2
(X). Since P
1
is simply connected, this lifts to a holomorphic map
P
1
→ C
N
, where N = dim J
2
(X). But P
1
has no nonconstant global holomor
phic functions, so any such map is constant.
We want to study the map Θ when X ⊂ P
4
is a smooth hypersurface
of degree 4. The family of lines on X is known to form a connected curve F
(Bloch and Murre [3]). In order to avoid certain technical problems which have
never been carefully considered, we will assume that F is smooth. This is the
case “in general” but not, for example, for X : T
4
0
+ T
4
1
+ T
4
2
+ T
4
3
+ T
4
4
= 0
(Tennison [9]).
Theorem (2.2) Let X be a smooth quartic threefold and assume the curve of
lines F is smooth. Then
Θ: A
2
(X) → J
2
(X)
is an isomorphism.
Remark (2.3) If s ∈ F corresponds to the line
s
⊂ X, one can deﬁne an
incidence correspondence Σ on F essentially by taking the closure of the set
¦(s, t) ∈ F × F − ∆ ¦
s
∩
t
∅¦.
One can then identify J
2
(X), using a beautiful idea of Tyurin [10] (note, how
ever, the criticism of Tyurin’s argument in Bloch and Murre [3]) with general
ized Prym
P
Σ
(F) = Image (Σ
∗
− 1: J(F) → J(F)).
Here J(F) is the jacobian of F and Σ
∗
satisﬁes a certain quadratic relation
Σ
2
∗
+ (q − 2)Σ
∗
− (q − 1) = 0. The standard Prym would correspond to the case
28 Lecture 2
q = 1. For the quartic threefold, q = 24. We shall not pursue these ideas further
here.
As a ﬁrst step toward proving (2.2), we show
Proposition (2.4) Let X be a smooth quartic threefold. Then Θ: A
2
(X) →
J
2
(X) is an isogeny (i.e., Θ is surjective with ﬁnite kernel).
Proof In several lemmas:
Lemma (2.5) Let X be a smooth projective threefold, and let π: Y → X
be obtained by blowing up a nonsingular closed subvariety V ⊂ X. Then
Θ
X
: A
2
(X) → J
2
(X) is an isogeny if and only if Θ
Y
is.
Proof If V is a point, then A
2
(X) A
2
(Y) and J
2
(X) J
2
(Y). If V is a
curve, one ﬁnds (cf. Beauville [1], Clemens and Griﬃths [4]) that A
2
(Y)
A
2
(X) × J(V) and J
2
(Y) J
2
(X) × J(V), where J(V) is the jacobian. In either
case, the assertion is clear since Θ
Y
induces the identity on J(V).
Lemma (2.6) Let X and Y be smooth projective threefolds, and let π: Y X
be a rational map of degree d 0 (i.e. [k(Y) : k(X)] = d). If Θ
Y
is an isogeny,
then so is Θ
X
.
Proof One knows from resolution theory that there exists a map Y
·
→ Y
obtained by a succession of blowings up with nonsingular centers such that in
the diagram
Y
·
π
·
Y
π
X
the arrow π
·
is everywhere deﬁned. We know that Θ
Y
an isogeny implies Θ
Y
·
an isogeny, so we are reduced to the case π: Y → X everywhere deﬁned.
Consider the diagram(commutative for either π
∗
or π
∗
– see Lieberman [12])
A
2
(Y)
Θ
Y
π
∗
J
2
(Y)
π
∗
A
2
(X)
Θ
X
π
∗
J
2
(X).
π
∗
We have π
∗
π
∗
= multiplication by d, and since A
2
(X) and J
2
(X) are divisible
(Lecture 1, (1.3)),
π
∗
: π
∗
(A
2
(X)) A
2
(X), π
∗
: π
∗
(J
2
(X)) J
2
(X).
Curves on threefolds and intermediate jacobians 29
In particular, Θ
X
is surjective. This implies that π
∗
(A
2
(Y)) π
∗
(J
2
(X)), and
hence that this map is an isogeny (its kernel is contained in Ker Θ
Y
). The map
π
∗
: π
∗
(J
2
(X)) J
2
(X) is a map of tori and hence easily seen to be an isogeny,
so the composition Θ
X
◦ π
∗
= π
∗
◦ Θ
Y
: π
∗
(A
2
(X)) → J
2
(X) is an isogeny. It
now follows that Θ
X
is an isogeny.
A rational map π: V W will be called a conic bundle if there exists
an open dense set W
0
⊂ W for the Zariski topology such that π is everywhere
deﬁned on π
−1
(W
0
), and a closed embedding
π
−1
(W
0
)
φ
P
2
W
0
W
0
realizing π
−1
(W
0
) as a family of conic curves over W
0
. If ¯ η → W denotes the
geometric generic point in the sense of algebraic geometry, this is equivalent
to the condition that the ﬁbre V
¯ η
be isomorphic to the projective line P
1
¯ η
.
Lemma (2.7) Let π: V → W be a conic bundle, where V and W are smooth
and projective of dimensions three and two respectively. Assume that the map
A
0
(W) → Alb(W) is an isogeny. Then A
2
(V)
Θ
V
−→ J
2
(V) is an isogeny.
Proof Since the geometric generic ﬁbre of π is P
1
, there exists a map f : W
·
→
W of ﬁnite degree and a diagram
X
ρ
φ
W
·
× P
1
pr
1
V
π
W
·
f
W
where ρ is obtained by a succession of blowings up with nonsingular centers,
and φ is everywhere deﬁned.
We now enumerate a series of exercises for the reader.
Exercise 1 Let T(W
·
) = Ker (A
0
(W
·
) → Alb(W
·
)). Show that f
∗
T(W
·
) = 0.
Exercise 2
T(W
·
)
f
∗
Ker (A
2
(W
·
× P
1
) →J
2
(W
·
× P
1
))
ρ
∗
Ker (A
2
(X) →J
2
(X)) .
30 Lecture 2
Exercise 3 π
∗
◦ f
∗
= φ
∗
◦ ρ
∗
◦ pr
∗
1
: A
0
(W
·
) → A
2
(V). (Hint: Move to general
position.)
Exercise 4 φ
∗
: Ker (A
2
(X) →J
2
(X)) Ker (A
2
(V) →J
2
(V)).
Exercise 5 Prove (2.7).
We return now to the proof of (2.4). There exists a nonempty open subset
X
0
of our quartic threefold X such that for all x ∈ X
0
we have:
(i) The intersection of the tangent hyperplane H
x
to X at x with X, H
x
∩ X,
has an isolated ordinary double point at x and no other singularities.
(ii) There exist no lines ⊂ P
4
supported on V and passing through x. (This is
because the family of lines on X has dimension 1.)
For x ∈ X
0
, let Q
x
⊂ H
x
be the tangent cone to x on H
x
∩X. Settheoretically
Q
x
is the union of all lines ⊂ P
4
which are at least triply tangent to X at x. In
terms of equations, one can choose homogeneous forms T
0
, . . . , T
4
on P
4
such
that x = (1, 0, 0, 0, 0) and H
x
: T
4
= 0. The fact that x is an ordinary double
point on X ∩ H
x
means that the equation for X has the form
T
2
0
q
2
(T
1
, T
2
, T
3
) + T
0
r
3
(T
1
, . . . , T
4
)
+ s
4
(T
1
, . . . , T
4
) + T
4
t
3
(T
0
, . . . , T
4
) ,
where q
2
, r
3
, s
4
, t
3
denote homogeneous polynomials of degrees 2, 3, 4, and
3 respectively. Moreover, q
2
is nondegenerate (i.e. equivalent after change of
coordinates to the quadric T
2
1
+ T
2
T
3
), and Q
x
: T
4
= q
2
(T
1
, T
2
, T
3
) = 0. In
particular, Q
x
is a cone over the smooth rational curve D
x
deﬁned by
D
x
: T
0
= T
4
= q
2
(T
1
, T
2
, T
3
) = 0 .
There is a conic bundle over X, π: D → X deﬁned with π
−1
(x) = D
x
for
v ∈ X
0
.
There is a rational morphism ρ: D → X deﬁned as follows: for x ∈ X
0
and
d ∈ D
x
, let
d
⊂ Q
x
be the line (ruling) passing through d and x. The line
d
meets X in 4 points, at least three of which coincide with x. Deﬁne ρ(d) =
“fourth point of intersection of
d
and X.”
Lemma (2.8) Let x ∈ X
0
and deﬁne C
x
= ρ(D
x
). Then C
x
= Q
x
∩ X is
a reduced, irreducible rational curve of degree 8. If x
·
∈ X
0
, x
·
x, then
C
x
· C
x
.
Proof of lemma Q
x
∩ X = Q
x
∩ (X ∩ H
x
) is the intersection of two distinct
irreducible hypersurfaces in H
x
and hence has pure dimension 1. In particular,
Curves on threefolds and intermediate jacobians 31
¦x¦ ∈ Q
x
∩ X is not an isolated component. It is clear, settheoretically, that
ρ(D
x
) ⊂ Q
x
∩ X and Q
x
∩ X − ¦x¦ ⊂ ρ(D
x
), so we have ρ(D
x
) = Q
x
∩ X.
The degrees of Q
x
and X are 2 and 4 respectively, so degC
x
= 8. Also
C
x
= ρ(D
x
), so C
x
is an irreducible rational curve. To show that the intersection
Q
x
∩X has multiplicity 1, choose two general points p
1
, p
2
∈ C
x
and let
1
,
2
⊂
Q
x
be the corresponding lines. There exists a hyperplane L ⊂ P
4
such that
L ∩ Q
x
=
1
∪
2
. Note that (
1
∪
2
) ∩ X contains p
1
, p
2
with multiplicity one,
because
i
∩ X contains x with multiplicity 3. On the other hand,
(
1
∪
2
) ∩ X = L ∩ Q
x
∩ X = L ∩ C
x
.
If the intersection Q
x
∩ X were not smooth at p
i
, the multiplicity of p
i
on
L ∩ Q
x
∩ X would be > 1.
It remains to show that x
·
∈ X
0
, x
·
x, implies C
x
· C
x
. Note that C
x
⊂ H
x
,
C
x
· ⊂ H
x
· , H
x
H
x
· . If C
x
= C
x
· , we would have C
x
⊂ Q
x
∩ H
x
· = curve of
degree 2. This is impossible (even settheoretically) by reason of degrees.
Returning to the proof of (2.4), let π
0
: D
0
→ X
0
denote the restriction of π
to D
0
= π
−1
(X
0
). The morphism π
0
is a regular morphism, as is ρ
0
: D
0
→ X.
The idea now is to construct a surface S and a conic bundle U over S by a
“bootstrap” technique. Let Γ ⊂ X be a reduced, irreducible curve. We assume
Γ
0
= Γ ∩ X
0
∅, and that for some x ∈ Γ
0
, Γ C
x
. It follows from these
hypotheses that S
0
= π
0−1
(Γ
0
) is an irreducible quasiprojective surface, and
ρ
0
(S
0
) ⊂ X has dimension 2. Note that Γ
0
⊂ ρ
0
(S
0
) (because x ∈ C
x
= ρ(D
x
)),
so ρ
0
(S
0
) ∩ X
0
∅. Choose a projective desingularization and completion S
of S
0
such that ρ
0
¦
S
0 extends to a morphism ψ: S → X. Deﬁne U
0
by the ﬁbre
square
U
0
f
0
h
0
D
0
ρ
0
π
0
X
S
ψ
X
U
0
∅, and we have by composition a regular morphism h
0
: U
0
→ X.
S is generically ﬁbred over Γ with rational ﬁbres, so it is easy to see A
2
(S )
Alb(S ). Also f
0−1
(s) P
1
for s ∈ S a general point. The hypotheses of (2.7)
are thus veriﬁed (taking U = suitable completion of U
0
) and (2.4) will follow
if we show h
0
(U
0
) is dense in X.
Suppose that h
0
(U
0
) is not dense in X. This can only happen if for all points
x ∈ ψ(S ) ∩ X
0
, C
x
⊂ ψ(S ). Take another curve Γ
·
analogous to Γ such that
Γ
·
ψ(S ) but Γ
·
∩ψ(S )∩X
0
∅. Now repeat the above construction replacing
32 Lecture 2
Γ by Γ
·
!! We get ψ
·
: S
·
→ X. Every surface in X is a hypersurface section by
Lefschetz theory, so ψ
·
(S
·
) ∩ ψ(S ) = ﬁnite union of curves. Let E = X
0
∩
ψ
·
(S
·
) ∩ ψ(S ). E is nonempty (it meets Γ
·
) and open in ψ(S ) ∩ ψ
·
(S
·
), hence
E is an inﬁnite set.
Assume now that the triple (U
·
, S
·
, f
·
) is no good either – that is, that
dimh
·
(U
·0
) < 3. Then for every x ∈ E, we must have C
x
⊂ ψ(S ) ∩ ψ
·
(S
·
).
This intersection is only a ﬁnite union of curves, so the inﬁnite set E must
contain points x x
·
with C
x
= C
x
· . This contradicts (2.8), so we conclude
that one of the maps h: U → X or h
·
: U
·
→ X must be of ﬁnite degree. This
completes the proof of (2.4).
We now have for the quartic threefold an isogeny θ
X
: A
2
(X) ≈ J
2
(X), and
we must verify
Theorem (2.2) Θ
X
is an isomorphism.
Proof Let V ⊃ X be a quartic fourfold which is general containing X. Let
C ⊂ S be respectively the curve of lines on X and the threefold of lines on V.
One knows that S and C are connected (for details cf. Bloch and Murre [3], §1)
and we have assumed C is smooth. Writing Gr(1, n) for the Grassmann of lines
in P
n
we have Gr(1, n) is locally a complete intersection (in fact, smooth) of
codimension 2 in Gr(1, n + 1). Since schemetheoretically C = S ∩ Gr(1, 4) ⊂
Gr(1, 5), smoothness of C implies smoothness of S along C. We now resolve
singularities of S , pulling back the line correspondence in S × V, and assume
both C and S smooth:
X
k
V
L(C)
r
2
r
1
P
j
p
2
p
1
L(S )
q
2
q
1
C S
(2.9)
where L(C) and L(S ) are the incidence correspondences
L(?) = ¦(a, b) ¦ a ∈ ?, b ∈
a
= line corresponding to a¦
and
P = L(S ) ×
V
X = ¦(a, b) ¦ a ∈ S, b ∈
a
∩ X¦.
Lemma (2.10) P is isomorphic to the blowup of S along C.
Proof One checks easily that P is smooth, p
1
is an isomorphism oﬀ the
Curves on threefolds and intermediate jacobians 33
preimage of C, and p
−1
1
(C) is a Cartier divisor. By a universal property of
blowingsup, this gives a map
P
π
BL
S
(C)
S.
The map π is bijective and birational. Since BL
S
(C) is smooth, π is an isomor
phism.
From the known cohomological structure of blowingsup, we get
(2.11) H
3
(P, Z) H
3
(S, Z) ⊕ H
1
(C, Z).
Lemma (2.12) The factor H
3
(S, Z) ⊂ H
3
(P, Z) maps to zero under the map
p
2∗
: H
3
(P, Z) → H
3
(X, Z).
Proof H
3
(S ) sits in H
3
(P) via p
∗
1
= j
∗
q
∗
1
. Since p
2∗
j
∗
= k
∗
q
2∗
, it suﬃces to
show that q
2∗
q
∗
1
H
3
(S ) = 0. But this group sits in H
3
(V, Z) = (0) by Lefschetz
theory.
Lemma (2.13) The map p
2∗
: H
3
(P, Z) → H
3
(X, Z) is surjective.
Before giving the proof of this, let me show how it implies the theorem.
First, by (2.11) and (2.12) we see the correspondence
r
2∗
r
∗
1
: H
1
(C, Z) → H
3
(X, Z)
is surjective, so the corresponding map on jacobians
J(C) → J
2
(X)
has connected ﬁbres. The diagram
A
0
(C)
J(C)
connected ﬁbres
A
2
(X)
isogeny
Θ
X
J
2
(X)
shows that Θ
X
is an isogeny with connected ﬁbres, hence an isomorphism,
proving (2.2).
For the proof of (2.13), we work with homology rather than cohomology.
Recall that given a Lefschetz pencil, ¦X
t
¦
t∈P
1 , of mdimensional hyperplane
sections of a smooth projective variety V and a base point 0 ∈ P
1
with X = X
0
smooth, there is associated to any choice of a singular ﬁbre X
t
i
and a path
34 Lecture 2
from 0 to t
i
on P
1
a cycle class γ ∈ H
m
(X, Z). The class γ is the base (boundary)
of a sort of cone Γ supported on
_
t∈
X
t
with vertex the singular point on X
t
i
(Wallace [11]). These vanishing cycles are known to generate the kernel of
H
m
(X, Z) → H
m
(V, Z). Now (2.13) follows by taking W = L(S ), Y = P in:
Lemma (2.14) Let V be a smooth projective variety deﬁned over C, dimV =
d + 1, φ: W → V a proper morphism, generically ﬁnite of degree k (with W
irreducible), X → V a smooth hyperplane section, and Y = φ
−1
(X). Then the
image of φ
∗
: H
d
(Y, Z) → H
d
(X, Z) contains the vanishing cycles.
Proof
Step a. We may assume X generic. Indeed, let P
∗
denote the parameter space
for hyperplanes X → V. We have families
X Y
and
P
∗
P
∗
with ﬁbres, respectively, hyperplanes X
t
⊂ V and inverse images Y
t
= φ
−1
(X
t
) ⊂
W, t ∈ P
∗
. For t
0
∈ P
∗
, there exists a neighborhood t
0
∈ U ⊂ P
∗
such that X
U
and Y
U
contract onto X
t
0
and Y
t
0
respectively. For t ∈ U we get a commutative
diagram of specializations
H
d
(Y
t
, Z)
φ
∗
H
d
(Y
t
0
, Z)
φ
∗
H
d
(X
t
, Z) H
d
(X
t
0
, Z).
Assuming X
t
, X
t
0
smooth, the bottomhorizontal arrowis an isomorphism. Thus
to prove the righthand vertical arrow surjective, it suﬃces to show the left
hand one is.
Step b. The sheaf R
d
π
∗
Z is constructible on P
∗
, where π: Y → P
∗
. Let U ⊂ P
∗
be a nonempty Zariski open set such that R
d
π
∗
Z is locally constant on U. Let
∈ P
∗
be a general line corresponding to a Lefschetz pencil ¦X
t
¦ on V.
Let X
0
= X and let X
t
1
, . . . , X
t
n
denote the singular ﬁbres. Choose paths
τ
i
: [0, 1] → such that τ
i
(0) = 0 and τ
i
(1) = t
i
. We may suppose τ
i
(x) ∈ U for
0 ≤ x ≤ 1. To each τ
i
there corresponds a vanishing cycle δ
i
∈ H
d
(X, Z), and
these vanishing cycles generate the group of vanishing cycles. Let us show, for
example, that δ
1
∈ Image(φ
∗
).
Since ¦X
t
¦ is general, we may assume that the singular points of the ﬁbres,
Curves on threefolds and intermediate jacobians 35
p
1
, . . . , p
n
∈ V lie in the open set over which φ is ´ etale. Let N be a neighbor
hood of p
1
in V such that φ
−1
(N) = M
(1)
∪ ∪ M
(k)
, where φ: M
(i)
∼
→ N for
each i. Let ε be close to 1, and let φ
ε
: Y
ε
→ X
ε
be the situation over the point
τ
1
(ε). Since ¦ε−1¦ is small, the vanishing cycle δ
ε
is “close to disappearing” on
X
ε
– that is, it is supported on X
ε
∩N – so it can be lifted to Y
ε
∩M
(i)
→ X
ε
∩N.
But now the fact that τ
1
([0, ε]) ⊆ U gives
H
d
(Y
0
)
=
φ
0∗
H
d
(Y
ε
)
φ
ε∗
H
d
(X
0
)
=
H
d
(X
ε
)
δ
1
δ
ε
.
Since δ
ε
∈ Image(φ
ε
∗
), we get δ
1
∈ Image(φ
0
∗
). This completes the proof of
Lemma (2.14).
Remarks on the literature
Intermediate jacobians were ﬁrst introduced by Weil, and then modiﬁed to vary
holomorphically in a family by Griﬃths. The cycle map is deﬁned in Grif
ﬁths [6] and the ﬁrst application to Fano varieties, the proof of the irrationality
of the cubic threefold, is in Clemens and Griﬃths [4]. The ﬁrst investigation of
the relation between A
2
and J
2
for Fano threefolds is in Murre [8]. I have also
borrowed heavily from Tyurin [10], particularly regarding (2.9) and the proof
of (2.12).
References for Lecture 2
[1] A. Beauville, Vari´ et´ es de Prym et jacobiennes interm´ ediaires, Ann. Sci.
Ecole Norm. Sup. (4), 10 (1977), 304–391.
[2] S. Bloch, An example in the theory of algebraic cycles, pp. 1–29 in Alge
braic KTheory, Lecture Notes in Math., no. 551, Springer, Berlin (1976).
[3] S. Bloch and J. P. Murre, On the Chow groups of certain types of Fano
threefolds, Compositio Math., 39 (1979), 47–105.
[4] C. H. Clemens and P. A. Griﬃths, The intermediate jacobian of the cubic
threefold, Ann. of Math. (2), 95 (1972), 281–356.
36 Lecture 2
[5] P. Deligne, Th´ eorie de Hodge. I, pp. 425–430 in Actes du Congr` es Inter
national des Math´ ematiciens (Nice, 1970), vol. 1, GauthierVillars, Paris
(1971).
[6] P. Griﬃths, On the periods of certain rational integrals. I, II, Ann. of Math.
(2), 90 (1969), 460–495; 90 (1969), 496–541.
[7] P. Griﬃths and J. Harris, Principles of Algebraic Geometry, Wiley, New
York (1978). [Reprinted 1994.]
[8] J. P. Murre, Algebraic equivalence modulo rational equivalence on a cu
bic threefold, Compositio Math., 25 (1972), 161–206.
[9] B. R. Tennison, On the quartic threefold, Proc. London Math. Soc. (3),
29 (1974), 714–734.
[10] A. N. Tyurin, Five lectures on three dimensional varieties, Uspehi Mat.
Nauk, 27 (1972), no. 5 (167), 3–50. [Translation: Russian Math. Surveys,
27 (1972), no. 5, 1–53.]
[11] A. Wallace, Homology Theory on Algebraic Varieties, Pergamon Press,
New York (1958).
[12] D. I. Lieberman, Higher Picard varieties, Amer. J. Math., 90 (1968), 1165–
1199.
3
Curves on threefolds and intermediate jacobians
– the relative case
Let X be a smooth quasiprojective variety over a ﬁeld k, and Y ⊂ X a divisor
with normal crossings. A problem of considerable importance is to develop
a theory of relative algebraic cycles for X relative to Y. This problem arises
geometrically – for example, if X is the resolution of a singular variety X
·
and
Y is the exceptional locus, and one wishes to study cycles on X
·
(compare the
theory of generalized jacobians of Rosenlicht [4] and Serre [5]). It also arises
in Ktheory. Indeed it seems likely that the higher Kgroups of a variety admit a
ﬁltration whose successive quotients can be interpreted up to torsion as relative
cycle groups in much the same way that the γﬁltration on K
0
relates K
0
to the
usual Chow group. (This is a point I hope to take up elsewhere.) Finally, and
rather surprisingly, there are compelling arithmetic reasons to develop such a
theory. An interesting example will be treated at length in Lectures 8 and 9.
One feature of such a theory when k ⊂ C is a theory of cycle classes for
relative codimensionr cycles in the intermediate jacobians J
r
(X, Y) associated
to the relative cohomology group H
2r−1
(X, Y) (see below). J
r
(X, Y) will be a
noncompact complex torus, so one may quotient by the maximal compact
real subtorus to obtain cycle classes in a real or complex vector space. In some
circumstances, one expects these cycle classes to ﬁll out a lattice whose volume
is linked to some value of a zeta function in much the same way the classical
regulator is associated to the value at 1 of the zeta function of a number ﬁeld.
Indeed, the classical case is precisely the case X = P
1
and Y = ¦0, ∞¦. The
relative cycles are points in P
1
−¦0, ∞¦ = G
m
corresponding to units in a given
number ﬁeld.
In this lecture we will work out in some detail the construction of relative
cycle classes of codimension 2 in the intermediate jacobian J
2
(X, Y) for the
particular case X = C × P
1
× P
1
with C a nonsingular projective curve and
Y = (C × P
1
× ¦0, ∞¦) ∪ (C × ¦0, ∞¦ × P
1
) := C × #.
38 Lecture 3
A onedimensional algebraic cycle γ =
_
n
i
γ
i
on C ×P
1
×P
1
will be called
a relative cycle if
(i) all the γ
i
meet C × # properly, and
(ii) if ⊂ # is any of the four lines, γ C × = 0 as a cycle.
These two conditions are suﬃcient if no component of γ meets any of the
“corners” C × ¦i, j¦, for i, j = 0, ∞. However, for the examples it is convenient
to permit γ to meet the corners, but only in a rather special way. Let p ∈ C ×#,
p = (p
0
, p
1
, p
2
), and let M
ε
= D
0
× D
1,ε
× D
2,ε
be a product of closed disks
centered at the p
i
where D
0
is ﬁxed and D
1
and D
2
have radius ε < 1, so p ∈
M
ε
⊂ C × P
1
× P
1
. To ﬁx the ideas we will consider only the hard case where
p
1
, p
2
= 0, ∞. The discussion when p is a smooth point on C × # is analogous
but easier and is left for the reader. Assume p ∈ Supp γ. Condition (ii) above
implies for ε small that γ ∩ M
ε
is trivial in H
2
(M
ε
, ∂M
ε
) H
1
(∂M
ε
). Indeed,
we can assume no component of γ meets ∂D
0
× D
1,ε
× D
2,ε
, so
∂(γ ∩ M
ε
) = n
1
p
0
× ∂D
1
× p
2
+ n
2
p
0
× p
1
× ∂D
2
,
where n
i
is the multiplicity of intersection of γ with C ×
i
(
i
= corresponding
line) at p. By hypothesis both n
i
= 0.
Now ﬁx an ε
0
for which the above discussion is valid and let Γ be a 3chain
on M
ε
0
with ∂ Γ = M
ε
0
∩γ+τ
ε
0
, where τ
ε
0
is supported on D
0
×∂D
1,ε
0
×D
2,ε
0
∪
D
0
× D
1,ε
0
× ∂D
2,ε
0
. Deﬁne Γ
ε
on M
ε
for ε ≤ ε
0
by excision. For γ to be a
relative cycle we require the existence of Γ
ε
0
such that Γ
ε
0
∩ C × # = ¦p¦ and
such that, moreover, writing x
1
, x
2
for the two obvious meromorphic functions
on P
1
× P
1
, we have
(3.1) lim
ε→0
_
τ
ε
dx
1
x
1
∧
dx
2
x
2
= 0.
(The reader should check this is always the case when p is a smooth point of
C × #.)
Example (3.2) Suppose p = (0, 0, 0), and that γ has two irreducible compo
nents through p, one with coeﬃcient +1 and the other with −1, parameterized
locally by
¦(x, a
1
x
r
1
+ hot, b
1
x
s
1
+ hot)¦
¦(x, a
2
x
r
2
+ hot, b
2
x
s
2
+ hot)¦
(hot = higher order terms).
Changing coordinates (the ﬁrst coordinate is immaterial) we can rewrite these
parameterizations
¦(−, x, b
1
a
−s
1
/r
1
1
x
s
1
/r
1
+ hot)¦,
¦(−, x, b
2
a
−s
2
/r
2
2
x
s
2
/r
2
+ hot)¦.
Curves on threefolds – the relative case 39
On D
0
× ∂D
1,ε
× D
2,ε
, a natural chain to take is
τ
·
ε
=
_
(λ(−) + (1 − λ) (−), x, λ(b
1
a
−s
1
/r
1
1
x
s
1
/r
1
+ hot)
+ (1 − λ) (b
2
a
−s
2
/r
2
2
x
s
2
/r
2
+ hot)
¸
¸
¸
¸
¦x¦ = ε, 0 ≤ λ ≤ 1
_
.
Assume now s
1
/r
1
= s
2
/r
2
. One gets formally
_
τ
·
ε
dx
1
x
1
∧
dx
2
x
2
=
2πi
r
1
log
_
b
r
1
1
a
s
1
2
b
r
1
2
a
s
1
1
_
+ o(1).
A similar analysis is valid on D
0
× D
1,ε
× ∂D
2,ε
, and we ﬁnd the conditions
(3.2.1)
r
1
s
1
=
r
2
s
2
, b
r
1
1
a
s
1
2
= b
r
1
2
a
s
1
1
lead to a relative cycle in (3.2). These conditions will really only begin to make
sense in Lecture 8 after we have studied Ktheory and the tame symbol.
In Lecture 2, we deﬁned the cycle map by integration over 3chains. The
process will be similar here, except that we will deal with a family of 3chains
∆
ε
and with a limit of integrals over ∆
ε
, as ε → 0. Let γ =
_
n
i
γ
i
be a
relative algebraic 1cycle on C × P
1
× P
1
. Let p(1), . . . , p(N) be the points
where components of γ meet C × #, and let M
ε
( j) be a neighborhood of p( j)
as above, ε ≤ ε
0
. Taking these neighborhoods to be suﬃciently small we may
assume the disjoint union M
ε
=
M
ε
( j) ⊂ C × P
1
× P
1
. Let M
0
ε
⊂ M
ε
be the
interior.
On each neighborhood M
ε
0
( j) we can deﬁne a 3chain Γ
ε
0
( j) as before with
∂ Γ
ε
0
( j) = γ∩M
ε
0
( j) −τ
ε
0
( j) with τ
ε
0
( j) supported on ∂M
ε
0
( j). We deﬁne Γ
ε
( j)
for ε ≤ ε
0
by excision, so ∂ Γ
ε
( j) = γ ∩ M
ε
( j) − τ
ε
( j), where ¦τ
ε
( j)¦ ⊂ ∂M
ε
( j).
We take Γ
ε
=
Γ
ε
( j), τ
ε
=
τ
ε
( j).
Let γ
ε
be the homology 2chain on C × G
m
× G
m
− M
0
ε
(here γ
ε
is a 2cycle
relative to ∂M
ε
) obtained from γ by excision. The cycle γ
ε
− τ
ε
is a 2cycle on
C × G
m
× G
m
− M
0
ε
. Let D = (G
m
× ¦1¦) ∪ (¦1¦ × G
m
).
Lemma (3.3) There exists a 3chain ∆
ε
0
on C ×G
m
×G
m
− M
0
ε
0
with ∂∆
ε
0
=
γ
ε
0
− τ
ε
0
+ E
ε
0
, where E
ε
0
is supported on C × D.
Proof The inclusion C × G
m
× G
m
− M
0
ε
0
→ C × G
m
× G
m
is easily seen to
be a homotopy equivalence. The K¨ unneth decomposition gives therefore
H
2
(C × G
m
× G
m
− M
0
ε
0
) H
2
(C) ⊗ H
1
(C) ⊗ H
1
(C) ⊗ Z.
Note that
dx
1
x
1
∧
dx
2
x
2
¸
¸
¸
¸
¸
γ
ε
= 0
40 Lecture 3
by reason of type, so
_
γ
ε
−τ
ε
dx
1
x
1
∧
dx
2
x
2
= −
_
τ
ε
dx
1
x
1
∧
dx
2
x
2
→ 0, as ε → 0.
This implies that the class of γ
ε
0
− τ
ε
0
lies in H
2
(C) ⊕ H
1
(C) ⊕ H
1
(C). Since
this subgroup is the image of H
2
(C × D), we can ﬁnd ∆
ε
0
as claimed.
We deﬁne ∆
ε
for ε ≤ ε
0
by considering the chain ∆
ε
0
− Γ
ε
0
and excising
away M
0
ε
. We have
(3.3.1) ∂∆
ε
= γ
ε
− τ
ε
+ E
ε
, Supp E
ε
⊂ C × D.
The next step is to deﬁne the appropriate intermediate jacobian.
Deﬁnition (3.4) (Deligne [2]) A mixed Hodge structure consists of
(a) a free Zmodule of ﬁnite type H
Z
,
(b) a ﬁnite increasing ﬁltration (weight ﬁltration) W on H
Q
= H
Z
⊗
Z
Q,
(c) a ﬁnite decreasing ﬁltration F
∗
on H
C
(Hodge ﬁltration).
These data are required to satisfy the condition that there should exist on
gr
W
H
C
a (necessarily unique) bigradation by subspaces H
pq
such that
(i) gr
n
W
H
C
= ⊕
p+q=n
H
pq
,
(ii) the ﬁltration F
∗
induces on gr
W
H
C
the ﬁltration
gr
W
(F)
p
= ⊕
p
·
≥p
H
p
·
,q
(iii) H
pq
= H
qp
.
Suppose now we are given a mixed Hodge structure H with weights ≤ 2r−1,
that is gr
m
W
H
Q
= 0, for m > 2r − 1. The same prescription as before, namely
J
r
= H
C
/F
r
+ H
Z
,
will in this case give us an abelian complex Lie group; that is, the image of H
Z
in H
C
/F
r
is discrete but not necessarily cocompact.
Example (3.5) Let X be a smooth projective variety, and Y ⊂ X a closed
(not necessarily smooth or even connected) subvariety. In general, for a proper
variety Y, Deligne has deﬁned [1] a mixed Hodge structure with weights ≤ k
on H
k
(Y, C). He also deﬁned [1] a Hodge structure on the relative cohomol
ogy H
k
(X, Y; C) of weight ≤ k compatible with the morphisms H
k−1
(Y, C) →
H
k
(X, Y; C) → H
k
(X, C) in the relative cohomology sequence. We can thus
consider the relative intermediate jacobian
J
r
(X, Y) H
2r−1
(X, Y; C)/F
r
+ H
2r−1
(X, Y; Z).
Curves on threefolds – the relative case 41
More speciﬁc example (3.6) Take X = C × P
1
× P
1
, Y = C × # as above,
and let r = 2. Actually, in this case the relative cohomology has a K¨ unneth
decomposition, and we will focus on H
1
(C) ⊗ H
2
(P
1
× P
1
, #). Combining the
exact sequence of relative cohomology for (P
1
×P
1
, #) with the Mayer–Vietoris
sequence
→ ⊕
4
H
i−1
(P
1
) → ⊕
4
H
i−1
(pt) → H
i
(#) → H
i
(P
1
× P
1
) → ,
one checks that there is an isomorphism of Hodge structures
H
2
(P
1
× P
1
, #) H
0
(pt).
Thus
F
2
(H
1
(C) ⊗ H
2
(P
1
× P
1
, #)) = F
2
H
1
(C) ⊗ H
2
(P
1
× P
1
, #) = (0),
so the intermediate jacobian associated to this piece of H
3
(C × P
1
× P
1
, C × #)
is H
1
(C, C)/H
1
(C, Z) H
1
(C, C
∗
).
We next deﬁne our cycle map
_
relative algebraic
1cycles on C × P
1
× P
1
_
→ H
1
(C, C
∗
).
Given a relative cycle γ we choose a family of 3chains ∆
ε
depending on ε
such that, with notation as before,
∂∆
ε
= γ
ε
− τ
ε
+ E
ε
, Supp E
ε
⊂ C × D.
If η is a closed global C
∞
1form on C, we deﬁne
(3.7) P
γ
(η) = lim
ε→0
−1
4π
2
_
∆
ε
η ∧
dx
1
x
1
∧
dx
2
x
2
.
Lemma (3.7.1)
(i) The limit in (3.7) is deﬁned and < ∞.
(ii) P
γ
(df ) = 0 if f is a C
∞
function on C.
(iii) Diﬀerent choices of ∆
ε
change P
γ
by an integral period of C; that is, P
γ
(η)
changes by
_
α
η for some integral 1cycle α on C independent of η.
Proof (i) It suﬃces to examine the situation in a neighborhood M of the ﬁnite
set of points where Supp γ meets C×#. On M we can write η = df , and Stokes’
42 Lecture 3
theorem gives
_
∆
ε
∩M
d f ∧
dx
1
x
1
∧
dx
2
x
2
=
_
−τ
ε
f
dx
1
x
1
∧
dx
2
x
2
.
_
Note
dx
1
x
1
∧
dx
2
x
2
¸
¸
¸
¸
¸
γ
ε
=
dx
1
x
1
∧
dx
2
x
2
¸
¸
¸
¸
¸
D
= 0.
_
Given our hypotheses about relative cycles, it is easy to show that the right
hand side tends to 0 with ε, which is enough to deduce existence of the limit.
(ii) The argument is precisely the same as for (i), except the integral over
∆
ε
∩ M is replaced by the integral over ∆
ε
.
(iii) Suppose ∆
·
ε
is another family of 3chains such that
∂∆
·
ε
= γ
ε
− τ
·
ε
+ E
·
ε
, E
·
ε
⊂ C × D.
There will exist a 3chain T
ε
supported on the boundary of an M
ε
(notation as
(3.2)) such that ∂T
ε
= τ
ε
− τ
·
ε
. Since H
2
(C × D) → H
2
(C × G
m
× G
m
) there
will exist a 3chain U
ε
supported on C × D such that ∂U
ε
= E
ε
− E
·
ε
. Then
∆
ε
−∆
·
ε
−U
ε
+T
ε
is a 3cycle. Since
1
2πi
dx
j
x
j
represents an integral cohomology
class on G
m
, it follows that
−1
4π
2
_
∆
ε
−∆
·
ε
−U
ε
+T
ε
η ∧
dx
1
x
1
∧
dx
2
x
2
= period of η on C.
Clearly
_
U
ε
η ∧
dx
1
x
1
∧
dx
2
x
2
= 0. Moreover, for ε small, η = df in some neighbor
hood of M
ε
, so
_
T
ε
η ∧
dx
1
x
1
∧
dx
2
x
2
=
_
τ
ε
−τ
·
ε
f
dx
1
x
1
∧
dx
2
x
2
→ 0,
as in (i). We conclude
lim
ε→0
−1
4π
2
_
∆
ε
−∆
·
ε
η ∧
dx
1
x
1
∧
dx
2
x
2
= period of η.
We now have our cycle map
_
algebraic 1cycles on C × P
1
× P
1
relative to C × #
_
γ→P
γ
→ H
1
(C, C
∗
).
Lecture 8 will be devoted to a special example, where the period will be related
to the Hasse–Weil zeta function of the curve C. For this purpose, it will be
important to factor the noncompact torus H
1
(C, C
∗
) by its maximal compact
subgroup – that is, to consider for η a real closed 1form on C
(3.8) ImP
γ
(η) = lim
ε→0
Im
_
−1
4π
2
_
∆
ε
η ∧
dx
1
x
1
∧
dx
2
x
2
_
.
Curves on threefolds – the relative case 43
One virtue of this is that the assignment γ ·→ ImP
γ
extends naturally to
all (not necessarily relative) algebraic 1cycles on C × P
1
× P
1
which meet
C × # properly:
Im P :
_
all algebraic 1cycles on C × P
1
× P
1
meeting C × # properly
_
→ H
1
(C, R).
In order to see this, let γ now denote any such algebraic 1cycle on C×P
1
×P
1
,
and let V
ε
=
k
V
k,ε
⊂ C be a union of small disjoint disks of radius ε about
the C coordinates of points in ¦γ¦ ∩ C × #. Then γ gives a class by excision in
H
2
(C
ε
× G
m
× G
m
, ∂V
ε
× G
m
× G
m
), where C
ε
= C − V
0
ε
.
Writing γ
·
ε
= γ − V
0
ε
× P
1
× P
1
, an analysis of the homology group H
2
(C
ε
×
G
m
×G
m
, ∂V
ε
×G
m
×G
m
) shows that there exists a 3chain ∆
ε
on C
ε
×G
m
×G
m
with ∂∆
ε
= γ
·
ε
+E
ε
+F
ε
, where ¦E
ε
¦ ⊂ (C
ε
×D) and ¦F
ε
¦ ⊂ ∂V
ε
×G
m
×G
m
. As
before, we can construct a continuous family of ∆
ε
depending on ε as ε → 0.
Lemma (3.9) Let η be a closed real 1form on C. Then
Im lim
ε→0
_
∆
ε
η ∧
dx
1
x
1
∧
dx
2
x
2
is independent of the choice of ∆
ε
.
Proof Let W
ε
⊂ C be the connected, simply connected set obtained by lin
early ordering the disks V
k,ε
and connecting them by narrow strips.
W:
Since W is simply connected, we have η¦
W
= df .
Suppose now
¯
∆
ε
is another 3chain with
∂
¯
∆
ε
= γ
·
ε
+
¯
E
ε
+
¯
F
ε
,
¦
¯
E
ε
¦ ⊂ C × D, ¦
¯
F
ε
¦ ⊂ ∂V
ε
× G
m
× G
m
.
Let S = (W×G
m
×G
m
)∪(C×D). One checks easily that H
2
(S ) → H
2
(C×G
m
×
G
m
). Since E
ε
+F
ε
−
¯
E
ε
−
¯
F
ε
is a 2cycle on S which bounds on C×G
m
×G
m
,
44 Lecture 3
there exist 3chains T on W × G
m
× G
m
and U on C × D such that ∂T + ∂U =
∂(∆
ε
−
¯
∆
ε
). In particular ∆
ε
−
¯
∆
ε
− T − U is an integral 3cycle, and
Im
_
∆
ε
−
¯
∆
ε
η ∧
dx
1
x
1
∧
dx
2
x
2
= Im
_
T+U
= Im
_
T
= Im
_
∂T
f
dx
1
x
1
∧
dx
2
x
2
.
Note that
dx
1
x
1
∧
dx
2
x
2
actually represents a class in H
2
(G
m
× G
m
, D; R). Also
∂T =
_
k
F
ε,k
−
¯
F
ε,k
rel (C × D). Thus F
ε,k
−
¯
F
ε,k
is a relative
2cycle on V
k,ε
× G
m
× G
m
, where say V
k,ε
is a closed εdisk about a point
t
k
∈ C. It will suﬃce therefore to show
lim
ε→0
Im
_
F
k,ε
−F
k,ε
( f − f (t
k
))
dx
1
x
1
∧
dx
2
x
2
= 0.
This is straightforward.
References for Lecture 3
[1] P. Deligne, Th´ eorie de Hodge III, Inst. Hautes Etudes Sci. Publ. Math.,
no. 44 (1974), 5–77.
[2] P. Deligne, Poids dans la cohomologie des vari´ et´ es alg´ ebriques, pp. 79–
85 in Proceedings of the International Congress of Mathematicians (Van
couver, B.C., 1974), vol. 1 (1975).
[3] P. Griﬃths, On the periods of certain rational integrals. I, II, Ann. of Math.
(2), 90 (1969), 460–495; 90 (1969), 496–541.
[4] M. Rosenlicht, Generalized Jacobian varieties, Ann. of Math., 59 (1954),
505–530.
[5] J. P. Serre, Groupes algebriques et corps de classes, Hermann, Paris
(1959). [Second edition, 1975; reprinted, 1984.]
4
Ktheoretic and cohomological methods
In this lecture we will discuss brieﬂy some of Quillen’s ideas about Ktheory,
and their extensions to various other “cohomology theories”. Basic references
are Quillen [3] and Bloch and Ogus [6]. The main results are the cohomologi
cal interpretations of the cycle groups
H
p
(X, 7
p
) CH
p
(X),
H
p
(X, 1
p
(Z)) CH
p
(X)/A
p
(X) (X deﬁned over C),
H
p
(X, 1
p
(µ
⊗p
)) CH
p
(X)/CH
p
(X).
(For notations, see below.)
Let ( be a small category. The nerve of (, N( is the simplicial set with
psimplices the set of diagrams
x
0
→ x
1
→ → x
p
with x
i
∈ Ob (, arrows in Morph (.
Face operators arise by omitting objects and composing arrows. Degeneracies
are deﬁned by inserting identity maps. The classifying space for N( is denoted
B(.
An exact category is an additive category A which can be embedded as a
full subcategory of an abelian category 7in such a way that if
0 → A
·
→ A → A
··
→ 0
is exact in 7 and A
·
, A
··
are isomorphic to objects in A, then A is isomorphic
to an object of A. (For an intrinsic characterization, see Quillen [3]). Given a
diagram in
0 → M
·
i
→ M
j
→ M
··
→ 0
which is exact in 7, Quillen refers to i as an admissible monomorphism and
46 Lecture 4
j as an admissible epimorphism. A functor F: A → A
·
on exact categories
is exact if it preserves exact sequences.
The exact category A has a zero object, from which it follows that BA is
contractible (exercise). Quillen’s idea is to build from A a new category QA
in such a way that
π
1
(BQA) k
0
(A)
Z
Ob A
([n] − [n
·
] − [n
··
] ¦ 0 → M
·
→ M → M
··
→ 0 exact in A)
.
He then deﬁnes
K
n
(A) π
n+1
(BQA).
Objects in QA will be the same as objects in A, but a morphism in QA,
M
1
−→
QA
M
2
, is by deﬁnition an isomorphism in A of M
1
with a subquotient
of M
2
, that is a ﬁltration M
·
⊂ M
··
⊂ M
1
with M
2
/M
··
, M
2
/M
·
, and M
··
/M
·
in A and an isomorphism Θ: M
1
M
··
/M
·
. In other words, a morphism is a
diagram in A
M
1
j
M
··
i
M
2
,
where i and j are admissible mono and epimorphisms respectively. For exam
ple, for any M ∈ Ob A, there are two canonical maps 0 → M in QAgiven by
the obvious sub and quotient arrows in A:
0 M.
One deﬁnes in this way a loop 0 M , and the map
K
0
(A) → π
1
(BQA)
sends [M] ·→ 0 M . (A direct proof that this is an isomorphism, or
even a welldeﬁned map might be awkward. The reader is advised to con
sult Quillen [3], where a preliminary discussion of covering spaces for spaces
B( helps to grease the skids.) Note ﬁnally that an exact functor F: A → A
·
induces a functor QA → QA
·
.
When Ais the category of ﬁnitely generated projectives over a ring A (asso
ciated with 1), Quillen gave another deﬁnition of K
∗
(A) K
∗
(A). He deﬁned
an Hspace BGL(A)
+
and a map from the classifying space of the inﬁnite gen
eral linear group ( lim
−−→
BGL
n
(A))
BGL(A) → BGL(A)
+
,
which was acyclic (i.e. induced an isomorphism on homology with coeﬃcients
Ktheoretic and cohomological methods 47
in any local system) and hence (by obstruction theory) was universal for maps
BGL(A) → X, X any Hspace. He deﬁned K
n
(A) = π
n
(BGL(A)
+
), n ≥ 1, and
proved π
n
(BGL(A)
+
) π
n+1
(BQA) (Quillen and Grayson [4]). It is not hard
to show from this
K
1
(A) GL(A)/[GL(A), GL(A)] GL(A)/E(A),
K
2
(A) H
2
(E(A), Z).
Here E(A) is the group of elementary matrices. One knows [1] that E(A) =
[GL(A), GL(A)] and [E(A), E(A)] = E(A). These deﬁnitions agreed, therefore,
with deﬁnitions given earlier by Bass [1] and Milnor [2] for K
1
and K
2
. K
2
can
also be deﬁned to be the center of the universal central extension of E(A):
0 → K
2
(A) → St(A) → E(A) → 1.
Here St(A) denotes the Steinberg group [2], and universal means that any cen
tral extension of E(A) by an abelian group G arises via pushout from a unique
homomorphism K
2
(A) →G.
Here is a compendium of useful facts about K
0
, K
1
, K
2
for a commutative
local ring R:
(i) Projective modules over R are free, so K
0
(R) Z.
(ii) E(R) = SL(R), so K
1
(R) R
∗
, the unit group of R.
(iii) There is a bilinear pairing
R
∗
⊗
Z
R
∗
→ K
2
(R),
sending r
1
⊗ r
2
to the Steinberg symbol ¦r
1
, r
2
¦. This pairing satisﬁes the rela
tions
(a) ¦r, −r¦ = 1, r ∈ R
∗
,
(b) ¦r, 1 − r¦ = 1 (Steinberg relation), r, 1 − r ∈ R
∗
,
(c) ¦r, s¦¦s, r¦ = 1, r, s ∈ R
∗
.
K
2
(R) is generated by Steinberg symbols, and when R is a ﬁeld, the above list
of relations is complete (indeed redundant, as (a) and (c) follow from (b)).
We now merely list some of the main results proved by Quillen about the
spaces BQA for general exact categories A. First, let c be the category of
short exact sequences in A. A sequence E ∈ Ob c can be written
0 → sE → tE → qE → 0.
We may view s, t, q as functors c → A.
c itself is an exact category, a diagram 0 → E
·
→ E → E
·
→ 0 being exact
if and only if the diagrams in Aobtained by applying s, t, and g are exact.
48 Lecture 4
Characteristic sequence theorem (4.1) The functor (s, q) : Qc → QA ×
QAis a homotopy equivalence (i.e. induces a homotopy equivalence BQc →
BQA× BQA).
Corollary (4.2) Let F
·
, F, F
··
: A
1
→ A
2
be exact functors, and suppose
natural transformations F
·
→ F → F
··
are given such that for any object
M
1
of A
1
, 0 → F
·
(M
1
) → F(M
1
) → F
··
(M
1
) → 0 is exact. Then F
∗
=
F
·
∗
+ F
··
∗
: K
∗
(A
1
) → K
∗
(A
2
).
Proof The data given amount to a functor / : A
1
→ c
2
, where c
2
is the
category of exact sequences for A
2
. Now apply (4.1).
Resolution theorem (4.3) Let / be a full subcategory of an exact category
Awhich is closed under extensions and is such that
(i) for any extension 0 → M
·
→ M → M
··
→ 0 in A, if M is in / then M
·
is
in /,
(ii) for any M
··
∈ A, there exists an exact sequence as in (i) with M ∈ Ob /.
Then BQ/ → BQAis a homotopy equivalence.
Corollary (4.4) Let A be a regular ring, A
A
the category of ﬁnitely generated
Amodules, and /
A
the category of ﬁnitely generated projective Amodules.
Then K
∗
(A
A
) K
∗
(/
A
).
Proof Let /
n
⊂ A
A
be the category of all ﬁnitely generated Amodules ad
mitting a projective resolution of length ≤ n, so /
0
= /
A
and A
A
= ∪
n
/
n
.
The hypotheses of (4.3) apply to /
n
⊂ /
n+1
, so
K
∗
(/
A
) = K
∗
(/
0
) K
∗
(/
1
) K
∗
(∪/
n
) U
∗
(A
A
).
Let 7 be an abelian category, J ⊂ 7 a nonempty full abelian subcategory
closed under taking subobjects, quotient objects, and ﬁnite products in 7.
Devissage theorem (4.5) Suppose that every object M of 7 has a ﬁnite ﬁl
tration 0 = M
0
⊂ M
1
⊂ ⊂ M
n
= M such that M
j
/M
j−1
∈ Ob / for all j.
Then K
∗
(J) −→
K
∗
(7).
Localization theorem (4.6) Let J ⊂ 7be a Serre subcategory, and let 7/J
denote the quotient abelian category. Then there is a long exact sequence of
Kgroups → K
n
(J) → K
n
(7) → K
n
(7/J) → K
n−1
(J) → .
As an example, let X be a noetherian scheme of ﬁnite Krull dimension. Let
A
X
be the category of coherent O
X
modules, and let T
i
X
⊂ A
X
denote the full
subcategory of sheaves whose support has codimension ≥ i. Let X
i
denote the
Ktheoretic and cohomological methods 49
set of points of X of codimension i. Then T
i+1
⊂ T
i
is a Serre subcategory, and
the quotient
T
i
/T
i+1
_
x∈X
i
_
n
A
(O
X,x
/m
n
X,x
)
,
where O
X,x
is the local ring at x and m
X,x
⊂ O
X,x
is the maximal ideal. Notice
that an O
X,x
/m
n
X,x
module admits a ﬁnite ﬁltration with successive quotients
O
X,x
/m
X,x
= k(x)modules, so we may apply (4.5) and (4.6) to get (for any R a
ring, K
n
(R) := K
n
(/
R
), where /
R
= f.g. projectives)
(4.7)
→K
n
(T
i+1
) K
n
(T
i
)
_
x∈X
i
K
n
(k(x)) K
n−1
(T
i+1
)
=
→K
n−1
(T
i+2
) K
n−1
(T
i+1
)
_
x∈X
i+1
K
n−1
(k(x)) .
The technique of exact couples shows that the above vertical maps ﬁt to
gether to give a complex
_
x∈X
0
K
n
(k(x)) →
_
x∈X
1
K
n−1
(k(x)) → →
_
x∈X
n
K
0
(k(x)),
which is the complex of E
1
terms of a spectral sequence
(4.8) E
p,q
1
=
_
x∈X
p
K
−p−q
(k(x)) ⇒ K
n
(A
X
).
This construction makes sense for any open set of X, and is functorial for
pullback to open sets, so we obtain a spectral sequence of presheaves for the
Zariski topology on X. To pass to the associated sheaves, some notation is
convenient. For p ∈ X
k
and A an abelian group, let i
p
A denote the sheaf for the
Zariski topology on X obtained by extending by zero the constant sheaf with
value A on the Zariski closure of p, ¦ ¯ p¦, to all of X. For example, if k = dimX
so that p is a closed point, i
p
A is the skyscraper sheaf with stalk A supported
at p. If k = 0 and X is irreducible, p is the generic point and i
p
A is the constant
sheaf A on all of X.
For a noetherian ring or scheme we denote by K
n
(resp. K
·
n
) the Ktheory of
the category of ﬁnitely generated projective modules or vector bundles (resp.
the category of all ﬁnitely generated modules or of coherent sheaves). On the
scheme level, we deﬁne sheaves for the Zariski topology 7
n
and 7
·
n
by sheaﬁ
50 Lecture 4
fying the presheaves
U ⊂
open
X → K
n
(U) or K
·
n
(U).
The sheaf version of (4.8) reads
(4.9) c
p,q
1
=
_
x∈X
p
i
x
K
−p−q
(k(x)) ⇒ 7
·
n
.
The main theorem, conjectured by Gersten [7] and proved by Quillen [3], is
Theorem (4.10) Let X be a regular kscheme, where k is a ﬁeld. Then (4.9)
degenerates at E
2
. In fact, E
p,q
2
= 0 for p 0, so the complex of sheaves
7
n
→
_
x∈X
0
i
x
K
n
(k(x)) →
_
x∈X
1
i
x
K
n−1
(k(x)) → →
_
x∈X
n
i
x
K
0
(k(x)) → 0
is a resolution of the sheaf 7
n
(note 7
n
= 7
·
n
since X is regular).
Since exactness of a sequence of sheaves is a local question, the theorem
can be restated:
Theorem (4.11) Let X = Spec R, where R is a regular local kalgebra which
is a localization of a kalgebra of ﬁnite type. Then the sequence
0 → K
n
(R) →
_
x∈X
0
K
n
(k(x)) → →
_
x∈X
n
K
0
(k(x)) → 0
is exact.
Exactness of (4.11) is easily seen to follow from
Theorem (4.12) Let X = Spec R as above. Then the maps K
∗
(T
i+1
X
) →
K
∗
(T
i
X
) are all zero. For the global case, the corresponding assertion is that
given α ∈ K
∗
(T
i+1
X
) and x ∈ X, there exists an open neighborhood U ÷ x in X
such that α ·→ 0 in K
∗
(T
i
U
).
It turns out that these results can be established in a more general context
which includes certain cohomology theories, as well as Ktheory.
I want to sketch this generalization, as well as the basic idea of the
proof. Let X be a variety over a ﬁeld k, and consider a “cohomology
theory” H
∗
(X) which may be one of the following:
(i) k = C, H
∗
(X) = H
∗
(X, A), ordinary singular cohomology with coeﬃcients
in some abelian group A (e.g. A = Z, Q, R, or C),
(ii) char k = 0 and H
∗
(X) = H
∗
DR
(X), de Rham cohomology [10],
(iii) r prime to char k and H
∗
(X) = H
∗
et
(X, µ
⊗r
n
), ´ etale cohomology with coeﬃ
cients in the rth twist of the sheaf µ
n
of nth roots of 1.
Ktheoretic and cohomological methods 51
The role of µ
⊗r
n
in case (iii) is confusing. For purposes of exposition we will
ﬁx an nth root of 1, ξ ∈ k, and identify µ
⊗r
n
Z/nZ. Then at the end we will
simply state what the result would be, keeping track of the twist by r.
If Z ⊂ X is a closed subvariety of codimension p, there is (in any of the three
cases above) a notion of cohomology with supports in Z, H
∗
Z
(X), ﬁtting into a
long exact sequence
(4.13) → H
r
Z
(X) → H
r
(X) → H
r
(X − Z) → H
r+1
Z
(X) → .
(A topologist would write H
r
Z
(X) = H
r
(X, X − Z).) One of the basic tenets of
Grothendieck duality theory (Verdier [8], [9]) is that if X is smooth over k of
dimension n, then for any of the above theories, if we write
H
r
(Z) := H
2n−r
Z
(X),
then H
r
(Z) is independent of X and covariant functorial for proper maps Z →
Z
·
. In fact, H
∗
(Z) is a Borel–Moore homology theory [9] in the sense that if Z
is itself smooth (but not necessarily complete) of dimension d, then H
r
(Z)
H
2d−r
(Z) (simply take Z = X in the above discussion).
As a good exercise for understanding the various exact sequences, the reader
should construct an exact sequence (for Z
·
⊂ Z ⊂ X closed subschemes, with
X smooth)
(4.14) → H
r
(Z
·
) → H
r
(Z) → H
r
(Z − Z
·
) → H
r−1
(Z
·
) → .
Now write
H
∗
Z
p
(X) = lim
−−→
Z⊂X
codim Z≥p
H
∗
Z
(X)
and deduce from (4.14) long exact sequences (compare with (4.7))
(4.15) →H
r
Z
i+1
(X) H
r
Z
i
(X)
_
x∈X
i
H
r−2i
(k(x)) H
r+1
Z
i+1
(X)
=
→H
r+1
Z
i+2
(X) H
r+1
Z
i+1
(X)
_
x∈X
i+1
H
r−2i−1
(k(x)) .
(Here if x ∈ Z ⊂ X is the generic point of Z, then we deﬁne H
∗
(k(x)) :=
lim
−−→
U⊂Z, open
H
∗
(U).) Precisely as before, one can use the exact couple technique
to construct a spectral sequence
(4.16) E
p,q
1
=
_
x∈X
P
H
q−p
(k(x)) ⇒ H
p+q
(X).
52 Lecture 4
The ﬁltration F
∗
H
∗
(X) deduced from (4.16) is the ﬁltration by codimension of
support; that is, a ∈ F
p
H
∗
(X) if and only if there exists Z ⊂ X a subscheme of
codimension p such that a ·→ 0 in H
∗
(X − Z). Also, the complex of E
1
terms
(4.16) can be sheaﬁﬁed as in (4.10), and one has
Theorem (4.17) The complex of sheaves
_
x∈X
0
i
x
H
p
(k(x)) →
_
x∈X
1
i
x
H
p−1
(k(x)) → →
_
x∈X
p
i
x
H
0
(k(x)) → 0
is a resolution of the Zariski sheaf 1
p
associated to the presheaf U → H
p
(U).
Corollary (4.18) The E
2
term of (4.16) is E
p,q
2
= H
p
Zar
(X, 1
q
). We have
H
p
Zar
(X, 1
q
) = 0 for p > q.
Proof The sheaves in (4.17) have no higher cohomology, so the complex of
abelian groups calculates H
∗
(X, 1
p
).
Remark (4.19) For either the ´ etale or singular cohomology H
∗
(X), one calcu
lates with a ﬁner topology on X than the Zariski topology. One has a morphism
of sites X
ﬁne
π
→ X
Zar
and hence a spectral sequence H
p
Zar
(X, R
q
π
∗
) ⇒ H
p+q
ﬁne
(X).
This spectral sequence coincides with (4.16) from E
2
onward. For a proof
in the de Rham theory, see Bloch and Ogus [6]. The general result is due to
Deligne.
Corollary (4.20) One has
H
p
(X, 7
p
) CH
p
(X);
H
p
(X, 1
p
) CH
p
(X)/A
p
(X) , for X over C, H
∗
(X) = H
∗
(X, Z);
H
p
(X, 1
p
) CH
p
(X)/CH
p
(X),
for X over k =
¯
k, H
∗
(X) = H
∗
et
(X, Z/Z) (, char k) = 1.
Proof One has (because the complexes of global sections of (4.10) and (4.17)
calculate the cohomology)
_
x∈X
p−1
k(x)
∗
→
_
x∈X
p
Z → H
p
(X, 7
p
) → 0, (4.21)
using that K
1
(k(x)) = k(x)
∗
and K
0
(k(x)) = Z, and
_
x∈X
p−1
H
1
(k(x)) →
_
x∈X
p
H
0
(k(x)) → H
p
(X, 1
p
) → 0. (4.22)
Ktheoretic and cohomological methods 53
The idea of the proof for Ktheory is that an irreducible family of codimension
p cycles parameterized by P
1
Γ
π
1
π
2
P
1
X
gives rise to a codimension(p − 1) subvariety π
2
(Γ) and norm
k(Γ)/k(π
2
Γ)
(t) on
π
2
(Γ), where t = π
∗
1
(u) and u is the “standard” function on P
1
(u(0) = 0, u(∞) =
∞). For more details, see Quillen [3].
For the ´ etale theory, H
1
(k(x)) H
1
Gal
(k(x), µ
) (Galois cohomology). This
group can be calculated by the K¨ ummer sequence
0 → µ
→ k(x)
∗
k(x)
∗
→ 0 (k(x) = algebraic closure of k(x))
and the known vanishing of H
1
Gal
(k(x), k(x)
∗
) (Hilbert’s theorem 90). One gets
H
1
(k(x)) k(x)
∗
/k(x)
∗
. Since H
0
(k(x)) = Z/Z, the presentation (4.22) for
the ´ etale theory becomes
_
X
p−1
k(x)
∗
/k(x)
∗
→
_
X
p
Z/Z → H
p
(X, 1
p
) → 0,
whence H
p
(X, 1
p
) CH
p
(X)/CH
p
(X).
Finally, for the singular theory H
∗
(X) = H
∗
(X, Z) one uses the fact that
algebraic and homological equivalence coincide for divisors to establish
CH
1
(Γ)/A
1
(Γ) H
1
Zar
(Γ, 1
1
) ⊂ H
2
sing
(Γ, Z)
for any complete nonsingular Γ. Now for a diagram
Γ
π
1
π
2
C X
as before, where C now is any smooth connected curve and Γ is smooth (us
ing resolution of singularities, we may assume this) one gets a commutative
diagram
H
1
(k(Γ))
norm
_
γ∈Γ
1
Z
norm
H
1
(Γ, 1
1
) 0
_
X
p−1
H
1
(k(x))
_
X
p
Z H
p
(X, 1
p
) 0.
54 Lecture 4
The desired isomorphism follows from this. For more details, the reader should
see Bloch and Ogus [6].
We now sketch a proof of (4.17), indicating at the end how a similar argu
ment proves (4.10). First, Theorem (4.17) can be reformulated in the spirit of
(4.12). Given α ∈ H
r
Z
p
(X) and x ∈ X, it suﬃces to show there exists an open
neighborhood x ∈ U ⊂ X such that α ·→ 0 in H
r
Z
p−1
(U). Now, let n = dimX
and let Z ⊂ X be a subscheme of pure codimension p such that α lies in the
image of H
2n−r
(Z) → H
r
Z
p
(X). It suﬃces to exhibit an open neighborhood U of
x and a closed subscheme Z
·
⊂ U of codimension p − 1 such that Z ∩ U ⊂ Z
·
and such that the map H
2n−r
(Z) → H
2n−r
(Z
·
) is zero.
As a ﬁrst step, replacing X by an open neighborhood of x and Z by its in
tersection with this neighborhood, one can construct a cartesian diagram (by a
variant on Noether normalization)
X ×
S
Z
q
θ
X
π
Z
σ
p
S
with S ⊂ A
n−1
an open subscheme, π smooth with ﬁbre dimension 1, and p
ﬁnite. The pullback θ of π will then also be smooth, so by further localization
one can arrange for the image of Z under the section σ, σ(Z) ⊂ X ×
S
Z, to be
deﬁned by the vanishing of a single function σ(Z) : f = 0. Let Z
·
= q(X×
S
Z) ⊂
X. We have
H
∗
(Z)
σ
∗
−→ H
∗
(X ×
S
Z)
q
∗
−→ H
∗
(Z
·
)
so it suﬃces to show σ
∗
= 0. The best way to think of this is topologically.
s (Z)
X × Z
q* ∆
∆
S
Z
q
Ktheoretic and cohomological methods 55
If ∆ ∈ H
∗
(Z), then σ
∗
(∆) = θ
∗
∆ ∩ [σ(Z)], where θ
∗
: H
∗
(Z) → H
∗+2
(X ×
S
Z)
is topological pullback (which exists even though θ is not proper), ∩ is cap
product, and [σ(Z)] ∈ H
2
(X ×
S
Z) is the divisor class. By hypothesis σ(Z) :
f = 0, so [σ(Z)] = 0 in H
2
(X×
S
Z) and σ
∗
= 0 as claimed. For complete details
of these arguments, the reader is referred to Bloch and Ogus [6].
Remark (4.23) Quillen’s proof of (4.12) diﬀered only in detail from the
above. Given X a regular scheme of ﬁnite type over a ﬁeld k, x ∈ X, and
α ∈ K
∗
(T
i+1
X
), one localizes around x to achieve a diagram
X
·
=
X ×
S
Y
q
θ
X
π
Y
σ
p
S
where Y is a divisor in X, α ∈ Image(K
∗
(T
i
Y
) → K
∗
(T
i+1
X
)), and π smooth with
ﬁbre dimension 1 and p ﬁnite. Localizing further, one can arrange (since θ is
smooth with ﬁbre dimension 1) for σ(Y) : f = 0 in X
·
. Writing S = Spec A,
Y = Spec B, X = Spec C, C
·
= C⊗
A
B, one has K
∗
(T
i
B
)
σ
∗
→
K
∗
(T
i
C
·
)
q
∗
→
K
∗
(T
i
C
)
and one wants to show σ
∗
= 0.
The exact sequence
0 →C
·
. f
C
·
σ
∗
B → 0
θ
∗
is split, and θ
∗
is ﬂat, so for any M ∈ Ob T
i
B
one gets
0 →C
·
⊗
B
M
. f
−→C
·
⊗
B
M → M → 0.
In other words, there is an exact sequence
0 → θ
∗
. f
−→ θ
∗
→ σ
∗
→ 0
of exact functors T
i
B
→ T
i
C
·
. One now applies (4.1)
θ
∗
= θ
∗
+ σ
∗
: K
∗
(T
i
B
) → K
∗
(T
i
C
·
),
that is σ
∗
= 0.
Remark (4.24) For ease of exposition, I ﬁxed an identiﬁcation µ
n
Z/nZ
in working with the ´ etale theory. This can be avoided. Keeping track of the
twisting, the spectral sequence (4.16) becomes
E
p,q
1
=
_
x∈X
p
H
q−p
Gal
_
k(x), µ
⊗m−p
n
_
⇒ H
p+q
et
_
X, µ
⊗m
n
_
,
56 Lecture 4
and the complex (4.17) reads
_
x∈X
0
i
x
H
p
_
k(x), µ
⊗m
n
_
→
_
x∈X
1
i
x
H
p−1
_
k(x), µ
⊗m−1
n
_
→
→
_
x∈X
p
i
x
H
0
_
k(x), µ
⊗m−p
n
_
→ 0.
This complex resolves the sheaf 1
p
( µ
⊗m
n
) associated to the presheaf U ·→
H
p
et
(U, µ
⊗m
n
).
References for Lecture 4
Foundational works on Ktheory:
[1] H. Bass, Algebraic KTheory, Benjamin, New York (1968).
[2] J. Milnor, Introduction to Algebraic KTheory, Annals of Mathematics
Studies, vol. 72, Princeton University Press, Princeton, N.J. (1971).
[3] D. Quillen, Higher algebraic Ktheory. I, in Algebraic KTheory I, Lec
ture Notes in Math., no. 341, Springer, Berlin (1973).
[4] D. Quillen and D. Grayson, Higher algebraic Ktheory. II, pp. 217–240 in
Algebraic KTheory, Lecture Notes in Math., no. 551,
Springer, Berlin (1976).
[5] R. Swan, Algebraic KTheory, Lecture Notes in Math., no. 76, Springer,
Berlin (1968).
Work on Gersten’s conjecture and analogous results for cohomology theories:
[6] S. Bloch and A. Ogus, Gersten’s conjecture and the homology of schemes,
Ann. Sci.
´
Ecole Norm. Sup. (4), 7 (1974), 181–201 (1975).
[6a] S. Bloch, K
2
and algebraic cycles, Ann. of Math. (2), 99 (1974), 349–379.
[7] S. M. Gersten, Some exact sequences in the higher Ktheory of rings,
pp. 211–243 in Algebraic KTheory I, Lecture Notes in Math., no. 341,
Springer, Berlin (1973).
Some expository references for various cohomology theories and Grothen
dieck duality:
[8] J.L. Verdier, A duality theorem in the etale cohomology of
schemes, pp. 184–198 in Proceedings of a conference on local ﬁelds
(Driebergen, 1966), Springer, Berlin (1967).
[9] J.L. Verdier, Dualit´ e dans la cohomologie des espaces localement com
pacts, S´ eminaire Bourbaki, Vol. 9, expos´ e no. 300 (1965), 337–349. [Re
printed Soci´ et´ e Math´ ematique de France, Paris, 1995.]
Ktheoretic and cohomological methods 57
[10] R. Hartshorne, On the de Rham cohomology of algebraic varieties, Inst.
Hautes
´
Etudes Sci. Publ. Math., no. 45 (1975), 5–99.
[11] P. Deligne, Cohomologie ´ etale (SGA4
1
2
), Lecture Notes in Math., no. 569,
Springer, Berlin (1977).
5
Torsion in the Chow group
The purpose of this lecture is to apply the techniques of Lecture 4 in order to
prove
Theorem (5.1) Let X be a smooth projective variety over an algebraically
closed ﬁeld k. Then the map A
0
(X) → Alb(X) induces an isomorphism on
torsion prime to the characteristic of k.
This theorem was ﬁrst proved geometrically by A. A. Roitman. For a diﬀer
ent geometric proof, see Bloch [3].
We have seen already (Lecture 1, proof of Lemma (1.4)) that the map on N
torsion
N
A
0
(X) →
N
Alb(X) is surjective for any N prime to char k. (The proof
given there applies to k = C, but the same discussion is valid quite generally
with singular cohomology replaced by ´ etale.)
Lemma (5.2) Suppose the map
N
A
0
(X) →
N
Alb(X) is injective for any sur
face X. Then it is so for X of arbitrary dimension.
Proof Let dimension X > 2, and suppose we are given a zerocycle
δ =
n
i
(x
i
)
on X such that Nδ is rationally equivalent to zero. By deﬁnition, there exist
irreducible curves C
i
⊂ X and functions f
i
on C
i
such that
_
( f
i
) = Nδ.
By a succession of blowings up of nonsingular points on X, one constructs
X
·
π
→X such that the strict transform C
·
i
of C
i
on X
·
is nonsingular for all i, and
such that moreover no two C
·
i
meet. Let δ
·
=
_
n
i
(X
·
i
) be some lifting of δ to X
·
(i.e. π
∗
δ
·
= δ). Then viewing the f
i
as functions on C
·
i
, we get
_
( f
i
) = Nδ
·
+η,
where π
∗
(η) = 0. Since the exceptional divisor in X
·
is a union of projective
spaces, it is easy to see that there will exist lines
j
⊂ X
·
and functions g
j
on
j
such that π(
j
) = pt,
_
(g
j
) = η, and no more than two
j
meet at a point.
Let D = ∪C
·
i
∪ ∪
j
. Let Y ⊂ X be a hypersurface section of large degree.
60 Lecture 5
Suppose D ⊂ Y but Y is otherwise “general.” I claim that such a Y is nonsin
gular. Let O
X
· (1) be the ample bundle on X
·
and let Y correspond to a section
σ ∈ Γ(X
·
, O
X
· (d)). Let I be the ideal of D ⊂ X
·
. We may assume d > 0
so I(d) is generated by its global sections. Since σ is general, this implies by
a standard Bertini argument that Y is nonsingular oﬀ of D. Further we may
assume Γ(X
·
, I(d)) → Γ(X
·
, (I/I
2
)(d)) is surjective and the sheaf (I/I
2
)(d) is
generated by global sections. Since rk I/I
2
> 1 and this sheaf is locally free
oﬀ the singular points of D, the section of (I/I
2
)(d) induced by σ vanishes at
worst at singular points of D. This implies that Y is only singular (possibly) at
D
sing
. Finally, since a singular point of D consists of two simple branches, it
will have embedding dimension ≤ 2. Since dimY ≥ 2, one sees easily that Y
can be chosen nonsingular.
We may continue this construction inductively, cutting Y by a hypersurface
section, etc., until we get
D ⊂ Z ⊂ X,
where Z ⊂ X is a smooth complete intersection of hypersurfaces with dimZ =
2. Notice the cycle δ
·
is supported on Z, and Nδ
·
is rationally equivalent to 0
(on Z!). One has wellknown isomorphisms (cf. Kleiman [8])
Alb(Z) Alb(X
·
) Alb(X)
and a diagram
δ
·
∈
N
A
0
(Z)
N
Alb(Z)
δ ∈
N
A
0
(X)
N
Alb(X)
Assuming the top arrow injective, we see δ ·→ 0 ⇒ δ = 0. Since δ was arbitary,
this proves the lemma.
Lemma (5.3) To establish injectivity
N
A
0
(X) →
N
Alb(X), it suﬃces to con
sider the case N = = prime.
Proof This is straightforward, using divisibility of A
0
(X) (Lecture 1, (1.3)).
We assume henceforth N = = prime and dimX = 2. We want to prove
A
0
(X) →
Alb(X). We work with the reduction mod of the sheaf 7
2
X
dis
cussed in Lecture 4.
Lemma (5.4) There is an exact sequence (Zariski cohomology)
0 → H
1
(X, 7
2
)/
ν
H
1
(X, 7
2
) → H
1
(X, 7
2
/
ν
7
2
) →
ν A
0
(X) → 0.
Torsion in the Chow group 61
Proof The rows and columns of the following diagram of sheaves are exact
(5.4.1)
0 0 0
0
ν 7
2
X
0
i
x
(
ν 7
2
(k(x)))
a
X
1
i
x
(µ
ν ) 0
0 7
2
x
ν
X
0
i
x
(K
2
(k(x)))
x
ν
X
1
i
x
(k(x)
∗
)
x
ν
X
2
i
x
(Z)
x
ν
0
0 7
2
X
0
i
x
(K
2
(k(x)))
X
1
i
x
(k(x)
∗
)
X
2
i
x
(Z) 0
0 7
2
/
ν
7
X
0
i
x
(K
2
(k(x))/
ν
)
X
1
i
x
(k(x)
∗
/k(x)
∗
ν
)
X
2
i
x
(
ν
Z) 0.
0 0 0 0
This is not quite obvious, but becomes so once one observes that the map
labeled a on the top row is surjective. For example one can tensor the divisor
map
X
0 i
x
(k(x)
∗
)
X
1 i
x
Z with µ
ν and get
_
X
0
i
x
(k(x)
∗
⊗µ
ν )
_
X
1
i
x
µ
ν
_
X
0
i
x
(
ν K
2
(k(x))).
a
The ﬁrst row of (5.4.1) implies
ν 7
2
has cohomological dimension ≤ 1, and
the ﬁrst column now gives the desired sequence, using
ν A
0
(X)
ν CH
0
(X)
ν H
2
(X, 7
2
).
We can, if we like, pass to the limit in (5.4), getting
0 → H
1
(X, 7
2
) ⊗ Q
/Z
→ H
1
(X, 7
2
⊗ Q
/Z
) → A
0
(X)() → 0,
where A
0
(X) () denotes the power torsion subgroup.
Lemma (5.5) H
3
et
(X, Q
/Z
(2)) Alb(X)().
62 Lecture 5
Proof The Albanese and Picard varieties of X are dual, so the e
m
pairing of
Weil (Lang [9]) gives an isomorphism
ν Alb(X) Hom(
ν Pic Var(X), µ
∞).
Passing to the limit over ν
Alb(X)() Hom(T
(Pic Var(X)), µ
∞),
where T
= Tate module. In ´ etale cohomological terms
T
(Pic Var(X)) H
1
et
(X, Z
(1)),
and Poincar´ e duality implies
H
3
et
(X, Q
/Z
(2)) Hom(H
1
(X, Z
(1)), H
4
(X, Q
/Z
(3)))
Hom(H
1
(X, Z
(1)), µ
∞)
since H
4
(X, Q
/Z
(2)) Q
/Z
. The lemma now follows.
Theorem (5.6) There exists an isomorphism
α: H
1
(X, 7
2
⊗ Q
/Z
) H
3
et
(X, Q
/Z
(2))
making the diagram below commute:
H
1
(X, 7
2
⊗ Q
/Z
)
α
A
0
(X)() 0
H
3
et
(X, Q
/Z
(2))
≈
Alb(X)().
0
Note that the righthand vertical arrow above is the cycle map, so Theo
rem (5.1) will follow from (5.6).
We prove (5.6) by constructing isomorphisms for any ν
α
ν
: H
1
(X, 7
2
/
ν
7
2
)
H
3
et
_
X, µ
⊗2
ν
_
.
The ﬁrst step is to interpret H
3
et
as a Zariski cohomology group. Recall from
Lecture 4, the Zariski sheaves 1
q
(µ
⊗
2
ν
) on X were deﬁned to be the sheaves
associated to the presheaves U ·→ H
q
(X, µ
⊗2
ν
). This implies that 1
q
is the qth
right derived functor of the morphism of topoi π: X
et
→ X
Zar
, so there is a
Leray spectral sequence
E
p,q
2
= H
p
Zar
_
X, 1
q
_
µ
⊗2
ν
__
⇒ H
p+q
et
_
X, µ
⊗2
ν
_
.
Torsion in the Chow group 63
In our case 1
q
= 0 for q ≥ 3 because dimX = 2, so locally (say on pieces of
an aﬃne covering) X has cohomological dimension 2. Also we wrote down an
acyclic resolution of 1
q
of length q, so H
p
(X, 1
q
) = 0 for p > q. The above
spectral sequence looks like
0
q
p
0
0 0
0 0
E
1
1,2
from which E
1,2
2
= H
1
Zar
(X, 1
2
(µ
⊗2
ν
)) H
3
et
(X, µ
⊗2
ν
).
Our objective thus becomes to construct an isomorphism
(5.6.1) α
ν
: H
1
(X, 7
2
/
ν
7
2
)
H
1
_
X, 1
2
_
µ
⊗2
ν
__
.
We will actually construct an exact sequence of sheaves
(5.6.2) 0 →C → 7
2
/
ν
7
2
→ 1
2
_
µ
⊗2
ν
_
→ 0,
where C is a constant sheaf. (It may well be that C = 0. This is known when
= 2; Elman and Lam [7].)
Recall Tate’s construction of the Galois symbol [10]
h: K
2
(k(X))/
ν
K
2
(k(X)) → H
2
Gal
_
k(X), µ
⊗2
ν
_
.
We have seen in the proof of (4.20) that H
Gal
(k(X), µ
ν ) k(X)
∗
/k(X)
∗
ν
, so
there is a cup product pairing
k(X)
∗
/k(X)
∗
ν
× k(X)
∗
/k(X)
∗
ν
→ H
2
Gal
_
k(X), µ
⊗2
ν
_
.
Deﬁne h¦ f , g¦ =
¯
f ∪ ¯ g where bar denotes class modulo
ν
th powers. To see
¯
f ∪(1 − f ) = 0, note 1− f is a norm from the ﬁeld L = k(X)( f
−ν
), so (1 − f ) =
cor
L/k(X)
(a) for some a ∈ H
1
Gal
(L, µ
ν ). Using the projection formula
¯
f ∪ (1 − f ) =
¯
f ∪ cor (a) = cor
L/k(X)
(
¯
f ∪ a) =
ν
cor ( f
−ν
∪ a) = 0.
64 Lecture 5
The Galois symbol relates the solutions of 7
2
/
ν
7
2
(5.4.1) and 1
2
(4.17)
as follows:
0 0
K
2
/
ν
K
2
1
2
( µ
⊗2
ν
)
X
0
i
x
(K
2
(k(x))/
ν
K
2
(k(x)))
h
X
0
i
x
_
H
2
Gal
_
k(x), µ
⊗2
ν
__
X
1
i
x
(k(x)
∗
/k(x)
∗
ν
)
X
1
i
x
(k(x)
∗
/k(x)
∗
ν
)
X
2
i
x
(Z/
ν Z)
X
2
i
x
(Z/
ν Z)
0 0.
(5.6.3)
The reader can show as an exercise that the squares of (5.6.3) commute. We
will thus have our exact sequence (5.6.2) and the ball game will be over, once
we show
Theorem (5.7) h: K
2
(k(X)) → H
2
Gal
(k(X), µ
⊗2
ν
) is surjective.
Proof Note that k(X) has Galois cohomological dimension 2, so there is a
diagram
K
2
(k(X))
h
K
2
(k(X))
h
K
2
(k(X))/
2
(k(X))
h
0
H
2
_
k(X), µ
⊗2
ν−1
_
H
2
_
k(X), µ
⊗2
ν
_
H
2
_
k(X), µ
⊗2
_
0.
From this, one reduces to the case ν = 1.
Let f : X → P
1
be a dominant rational map. Let F = k(X), K = k(P
1
) ⊂ F,
and let Y be an aﬃne open subvariety of the generic ﬁbre of f . The subvariety
Y is thus an open curve deﬁned and of ﬁnite type over K. We assume Y is
geometrically reduced and connected (i.e. K is algebraically closed in F).
Let K
·
⊃ K be a ﬁnite extension ﬁeld, Y
·
= Y
K
· , F
·
= F K
·
, π: Y
·
→ Y the
natural ﬁnite morphism. Associated to π there is a covariant trace morphism π
∗
on ´ etale cohomology, which is analogous to the corestriction on Galois coho
Torsion in the Chow group 65
mology. In fact, there is a commutative diagram
H
r
et
_
Y
·
, µ
⊗2
_
π
∗
H
r
Gal
_
F
·
, µ
⊗2
_
cor
H
r
et
_
Y, µ
⊗2
_
H
r
Gal
_
F, µ
⊗2
_
.
(5.7.1)
One way to understand this situation is to notice that the “geometric ﬁbres”
of π (i.e. schemes of the form Y
·
×
Y
Spec A, where A is a strictly Henselian
ring lying over some local ring of Y) are ﬁnite over Spec A and hence strictly
Henselian. This implies that the higher derived functors of π are trivial, so
that H
r
(Y
·
, µ
⊗2
) H
r
(Y, π
∗
(µ
⊗2
)). The map π
∗
comes from a map on sheaves
π
∗
(µ
⊗2
Y
·
) → µ
⊗2
Y
, which on the ﬁbres is simply summing over connected com
ponents. In particular, the map on sheaves is surjective. Since Y has cohomo
logical dimension 2, it follows that π
∗
in (5.7.1) is surjective.
We were free initially to shrink (localize) Y as much as we wanted, so it will
suﬃce to show
Image
_
H
2
_
Y, µ
⊗2
_
→ H
2
_
F, µ
⊗2
__
⊂ Image
_
K
2
(F) → H
2
_
F, µ
⊗2
__
.
There is also a transfer map tr : K
2
(F
·
) → K
2
(F), and the diagram
K
2
(F
·
)
h
tr
H
2
_
F
·
, µ
⊗2
_
cor
K
2
(F)
h
H
2
_
F, µ
⊗2
_
(5.7.2)
is known to commute (Bass and Tate [2]). Together with (5.7.1) and the sur
jectivity of π
∗
, this reduces us to proving
Image
_
H
2
_
Y
·
, µ
⊗2
_
→ H
2
_
F
·
, µ
⊗2
__
⊂ Image
_
K
2
(F
·
) → H
2
_
F
·
, µ
⊗2
__
for some ﬁnite extension K
·
of K.
Let
¯
K be the separable closure of K and write
¯
Y = Y
¯
K
¯
Y has cohomological
dimension 1, and the Hochshild–Serre spectral sequence implies
H
2
_
Y, µ
⊗2
_
H
1
Gal
_
K, H
1
_
¯
Y, µ
⊗2
__
.
The point is (and this explains why we must work with varieties and ´ etale
cohomology rather than the more familiar Galois cohomology of ﬁelds) that
H
1
(
¯
Y, µ
⊗2
) is a ﬁnite group. Thus for some ﬁnite K
·
over K we may suppose
(i) Gal(K/K
·
) acts trivially on H
1
_
¯
Y, µ
⊗2
_
;
(ii) H
1
_
Y
·
, µ
⊗2
_
maps onto H
1
_
¯
Y, µ
⊗2
_
.
66 Lecture 5
By the previous discussion, we can replace Y and Y
·
and assume (i) and (ii)
hold for Y and K. We then have
H
1
(K, µ
) ⊗ H
1
(Y, µ
) H
2
(Y, µ
⊗2
)
res
H
1
(K, µ
) ⊗ H
1
(F, µ
) H
2
(F, µ
⊗2
)
K
2
(F)/K
2
(F)
h
(K
∗
/K
∗
) ⊗ F
∗
/F
∗
(F
∗
/F
∗
) ⊗ (F
∗
/F
∗
)
cup
symbol
(5.7.3)
The reader easily checks that Image(res) ⊂ Image(h) as claimed.
Remark (5.8) The arguments in this lecture can be pushed further in several
directions. For a surface X over an algebraically closed ﬁeld of characteristic
prime to , one can show that the power torsion in K
1
(X), K
1
(X)(), maps
onto H
2
et
(X, Q
/Z
(2)), the map being an isomorphism at least when = 2.
This leads to some detailed conjectures concerning Chern characters linking
Ktheory with ﬁnite coeﬃcients and ´ etale cohomology. I hope to publish a
description of these conjectures and the evidence for them elsewhere.
Concerning (5.7), one can show by a similar argument that for F a func
tion ﬁeld of transcendence degree n over an algebraically closed ﬁeld, the cup
product
H
1
(F, µ
ν )
⊗n
ν
−→H
n
(F, µ
⊗n
ν
)
is surjective.
I wonder whether the whole cohomology algebra ⊕H
r
(F, µ
⊗r
ν
) might not be
generated by H
1
? If true, this would imply (identifying µ Z/Z) that the
Bockstein maps
H
r
(F, Z/Z) → H
r+1
(F, Z/Z)
were all zero. Globally this would imply the existence of a divisor D ⊂ X such
that the composition
H
r
et
(X, Z/Z)
Bockstein
H
r+1
et
(X, Z/Z) → H
r+1
et
(X − D, Z/Z)
was zero. In other words, the image of the Bockstein would lie in the ﬁrst level
of the “coniveau ﬁltration” (Bloch and Ogus [5]). When the ground ﬁeld is
Torsion in the Chow group 67
the complex numbers, ´ etale cohomology can be replaced by ordinary (singu
lar) cohomology. It is known by work of Atiyah and Hirzebruch that torsion
cohomology classes are not always carried by algebraic cycles [1] (unlike the
situation for H
2
(X, Z), whose torsion subgroup is known to come from Chern
classes of divisors). Is it possible, however, that all torsion classes die in the
complement of a divisor?
References for Lecture 5
[1] M. Atiyah and F. Hirzebruch, Analytic cycles on complex manifolds,
Topology, 1 (1962) 25–45.
[2] H. Bass and J. Tate, The Milnor ring of a global ﬁeld, in Algebraic K
Theory II, Lecture Notes in Math., no. 342, Springer, Berlin (1973).
[3] S. Bloch, Torsion algebraic cycles and a theorem of Roitman, Compositio
Math., 39 (1979), 107–127.
[4] S. Bloch, Torsion algebraic cycles, K
2
, and Brauer groups of function
ﬁelds, Bull. Amer. Math. Soc., 80 (1974), 941–945.
[5] S. Bloch and A. Ogus, Gersten’s conjecture and the homology of schemes,
Ann. Sci.
´
Ecole Norm. Sup. (4), 7 (1974), 181–212 (1975).
[6] P. Deligne, Cohomologie ´ etale (SGA4
1
2
), Lecture Notes in Math., no. 569,
Springer, Berlin (1977).
[7] R. Elman and T. Y. Lam, On the quaternion symbol homomorphism
g
F
: k
2
(F) → B(F), in Algebraic KTheory II, Lecture Notes in Math.,
no. 342, Springer, Berlin (1973).
[8] S. Kleiman, Algebraic cycles and the Weil conjectures, in Dix exposes
sur la cohomologie des schemas, North Holland, Amsterdam (1968).
[9] S. Lang, Abelian Varieties, Interscience Publishers (1959). [Reprinted
Springer, 1983.]
[10] J. Tate, Relations between K
2
and Galois cohomology, Invent. Math., 36
(1976), 257–274.
[11] A. A. Roitman, private correspondence.
6
Complements on H
2
(K
2
)
It is possible to study H
2
(X, 7
2
) inﬁnitesimally and analytically, as well as
algebraically. Although one cannot establish at this time any direct relation
between the results obtained and the structure of the Chow group, it is plau
sible that such relations exist. Frequently such formal manipulation suggests
conjectures about cycles which can be tested directly.
Suppose for example X is an algebraic surface deﬁned over C, and let 7
an
2
denote the sheaf for the complex topology on X obtained by sheaﬁfying the
presheaf U → K
2
(Γ(U, O
X
an )). One can derive a version of the Gersten–Quillen
resolution for 7
an
2
. The problem is that the sheaves involved frequently have
nontrivial cohomology so there is no obvious map H
2
(X
an
, 7
an
2
) → CH
2
(X).
There is, however, an obvious map
CH
2
(X) H
2
(X
Zar
, 7
2
) → H
2
(X
an
, 7
an
2
).
The easiest case with which to deal is when P
g
= q = 0. In this case, conjec
turally CH
2
(X) Z.
Theorem (6.1) Let X be a surface as above, and assume P
g
= q = 0. Then
the image of CH
2
(X) → H
2
(X
an
, 7
an
2
) is Z, and CH
2
(X) → Image is the degree
map.
Proof The existence of a dlog map 7
an
2
→ Ω
2
X
an
leads to
CH
2
(X)
d log
deg
H
2
(X
an
, 7
an
2
)
d log
H
2
(X, Ω
2
) H
2
(X
an
, Ω
2
X
an
) C,
so Ker(CH
2
(X) → H
2
(X
an
, 7
an
2
)) ⊂ Ker(deg).
70 Lecture 6
For the other inclusion, let δ be a cycle of degree 0 on X and let C ⊂ X be
a smooth curve with Supp δ ⊂ C. From the exact sequence of analytic sheaves
on C,
0 → C
∗
C
an
→ O
∗
C
an
→ Ω
1
C
an
→ 0,
one gets a class η ∈ H
1
(C
an
, C
∗
) mapping to [δ] ∈ H
1
(C
an
, O
∗
) Pic(C). Let
O
X
an (∞C) denote that sheaf of analytic functions on X
an
meromorphic along
C. One has sequences
0 → O
∗
X
an
→ O
X
an (∞C)
∗
→ j
∗
Z
C
→ 0, j : C → X, (6.1.1)
0 → 7
an
2
→ 7
2
(O
X
an (∞C))
tame
j
∗
O
∗
C
an
→ 0, (6.1.2)
and one has a symbol map
(6.1.1) ⊗
Z
C
∗
→ (6.1.2).
This gives a commutative diagram
η ∈ H
1
(C
an
, Z) ⊗ C
∗
∂⊗1
H
2
(X
an
, O
∗
X
an
) ⊗ C
∗
H
1
(C
an
, C
∗
)
∂
H
2
(X
an
, O
∗
X
an
⊗ C
∗
)
Pic(C) H
1
(C, O
∗
C
an
)
∂
H
2
(X
an
, 7
an
2
).
(6.1.3)
It suﬃces to note nowthat the hypotheses P
g
= q = 0 imply that H
2
(X
an
, O
∗
X
an
)⊗
C
∗
= (0). Indeed, the exponential sequence
0 → Z
X
an → O
X
an → O
∗
X
an
→ 0
gives
H
2
(X, O
X
) → H
2
(X
an
, O
∗
X
an
) → H
3
(X
an
, Z).
Since P
g
= q = 0, the group on the left vanishes and that on the right is
ﬁnite.
The rest of this lecture is devoted to the inﬁnitesimal structure of H
2
(X, 7
2
).
We ﬁx a perfect ground ﬁeld k which in the case of characteristic zero we
assume to be algebraic over Q. (Essentially, when k is not algebraic over the
prime ﬁeld, a cycle on some variety over k must be viewed as the germ of a
Complements on H
2
(K
2
) 71
family of cycles. In the presence of nontrivial derivations of k, the deforma
tions of kcycles will be more complicated.) Let X be a smooth geometrically
connected kvariety. We consider a functor CH
2
X
from the category ( of aug
mented artinian kalgebras to abelian groups deﬁned by
¯
CH
2
X
(A) = Ker(H
2
(X
k
× A, 7
2
)
augmentation
H
2
(X, 7
2
)).
Consider ﬁrst the case char k = 0. The basic local result is
Theorem (6.2) Let R be a local kalgebra, and A an augmented artinian
kalgebra with augmentation ideal m. Put S = R ⊗
k
A and I = R ⊗
k
m, and
deﬁne
K
2
(S, I) = Ker(K
2
(S ) → K
2
(R)),
Ω
1
S,I
= Ker(Ω
1
S
→ Ω
1
R
),
where Ω
1
is the module of absolute K¨ ahler diﬀerentials. The universal deriva
tion d: S → Ω
1
S
induces d: I → Ω
1
S,I
, and we have an isomorphism K
2
(S, I)
Ω
1
S,I
/dI.
The proof is purely algebraic and will not be given in these notes. In terms
of symbols, the map
K
2
(S, I) → Ω
1
S,I
/dI
is given by
(6.2.1) ¦1 + ι, s¦ → log(1 + ι)
ds
s
, ι ∈ I, s ∈ S
∗
.
One has an exact sequence (k
·
= algebraic closure of k in R)
(6.2.2) k
·
⊗
k
m
1⊗d
−→R ⊗
k
Ω
1
A,m
→ Ω
1
S,I
/dI →
_
Ω
1
R
/dR
_
⊗
k
m → 0,
which can be used to analyze the structure of K
2
(S, I).
Thinking of R as the local ring at a variable point on X, we may pass to
Zariski sheaves, getting an exact sequence of sheaves of kvector spaces
(6.2.3)
0 k ⊗
k
dm O
X
⊗
k
Ω
1
A
7
2
X×A,X×m
(Ω
1
X
/dO
X
) ⊗
k
m 0
(Ω
1
X
⊗
k
m/ dO
X
⊗
k
m
·
).
72 Lecture 6
Here m
·
= Ker(d: m → Ω
1
A
). Hodge theory gives an exact sequence when X is
complete:
0 → H
∗
(X, Ω
1
X
) → H
∗
(X, Ω
1
A
/dO
X
) → H
∗+1
(X, O
X
) → 0.
In this case the long exact sequence of cohomology associated to (6.2.3) looks
like
0 → H
2
(X, O
X
)⊗(Ω
1
A
/dm) →
¯
CH
2
X
(A)
→ H
2
(X, Ω
1
X
/dO
X
) ⊗ m H
3
(X, O
X
) ⊗ Ω
1
A
H
3
(X, O
X
) ⊗ m.
1⊗d
(6.2.4)
Remark (6.2.5) (i) When H
3
(X, O
X
) = (0) (e.g. dimX = 2) or when
d: m → Ω
1
A
, one gets
(6.2.6) 0 → H
2
(X, O
X
) ⊗ (Ω
1
A
/dm) →
¯
CH
2
X
(A) → H
2
(X, Ω
1
X
) ⊗ m → 0.
There do exist, however, artinian kalgebras for which d is not injective, so in
general the righthand term is more complicated.
(ii) A functor F: ( → (sets) is said to be prorepresentable if there exists a
complete local kalgebra Λ and an isomorphism of functors
F() Morph
local kalgebra
(Λ, ).
The functor A → H
2
(X, Ω
1
X
) ⊗ m is prorepresented by the completion at
0 of the symmetric algebra on the dual vector space to H
2
(X, Ω
1
X
). When
H
3
(X, O
X
) = (0), (6.2.6) displays
¯
CH
2
X
as a prorepresentable quotient and a
non prorepresentable kernel. If H
2
(X, O
X
) = H
3
(X, O
X
) = (0) we see that
¯
CH
2
X
is itself prorepresentable, a result which should be compared with Lecture 1.
For example, when X is a surface, we ﬁnd
¯
CH
2
X
isomorphic to the formal group
at the origin of the Albanese if and only if H
2
(X, O
X
) = (0).
Example (6.2.7) Let A = k[ε, δ](ε
2
, δ
2
, εδ). Consider deformations of the
trivial divisor
D
ε
∈ Ker(H
1
(X × k[ε], O
∗
) → H
1
(X, O
∗
)) H
1
(X, O
X
),
E
δ
∈ Ker(H
1
(X × k[δ], O
∗
) → H
1
(X, O
∗
)) H
1
(X, O
X
),
Complements on H
2
(K
2
) 73
and consider the product
D
ε
E
δ
∈ Ker(H
2
(X × A, 7
2
) → H
2
(X × k[ε], 7
2
) ⊕ H
2
(X × k[δ], 7
2
))
H
2
(X, O
X
).
If D
ε
and E
δ
are represented by 1cocycles 1 + d
i j
ε and 1 + e
i j
δ respectively,
the intersection is represented by ¦1 + d
i j
ε, 1 + e
jk
δ¦. Viewed as an element in
H
2
(X, O
X
)⊗
k
(Ω
1
A
/dm), we ﬁnd D
ε
E
δ
represented by ¦d
i j
e
jk
¦⊗ε dδ. Identifying
Ω
1
A
/dm k with generator ε dδ, it follows that intersection of cycles is given
by cup product H
1
(O
X
) ⊗ H
1
(O
X
) → H
2
(O
X
). Thus, the non prorepresentable
kernel in (6.2.6) does play a role in the cycle theory.
I want ﬁnally to discuss the work of Stienstra on
¯
CH
2
X
when X is deﬁned
over a perfect ﬁeld k of characteristic p 0. If R is a ring, the big Witt ring of
R, big W(R), is deﬁned additively by
big W(R)
+
= (1 + Rt[[t]])
∗
= group of power series in R
with constant term 1.
The multiplication is determined by functoriality in R together with the re
quirement (1 − at)
−1
∗ (1 − bt)
−1
= (1 − abt)
−1
. When pR = 0, big W(R) splits
into a product of rings W(R),
W(R) =
_
E(a
0
t)E(a
1
t
p
)E(a
2
t
p
2
)
_
,
where
E(t) = exp
⎛
⎜
⎜
⎜
⎜
⎝
t +
t
p
p
+
t
p
2
p
2
+
⎞
⎟
⎟
⎟
⎟
⎠
= Artin–Hasse exponential.
For example, k being perfect one ﬁnds that W(k) is a complete characteristic
zero discrete valuation ring with maximal ideal pW(k) and residue ﬁeld k.
W(R) has a ring endomorphism F induced by the Frobenius r → r
p
on R
and an additive endomorphism V given by P(t) → P(t
p
), P being a power
series in t. One has FV = VF = p. The Frobenius on W(k) is an automorphism
and is usually denoted σ. The relation V(Fx y) = xV(y) holds in W(R), so
W(R) is a module for the Dieudonn´ e ring D = W(k) [F, V]. Relations in D are
FV = VF = p, Fw = w
σ
F, Vw = w
σ−1
V with W ∈ W(k). An important
variant on these ideas arises by deﬁning
big
¯
W(R) = Ker(R[T]
∗
−→
T·→1
R
∗
).
Big
¯
W(R) is a module for big W(R) and inherits the same splitting in character
istic p:
¯
W(R) = ¦E(a
0
t) E(a
n
t
p
n
) ¦ a
i
nilpotent¦.
74 Lecture 6
We view
¯
W as a functor
(artinian kalgebras)
¯
W
−→(left Dmodules).
(6.3.1) Let M be a left Dmodule with the following properties:
(i) M is Vadically complete and separated,
(ii) M has no Vtorsion,
(iii) M/VM is a W(k)module of ﬁnite length.
For example, M = D/D (F
n
− V
m
) or M = k[[V]]. We view M as a right
Dmodule via
mw = wm, mV = Fm, mF = Vm.
The tensor product M⊗
D
¯
W then becomes a functor
M⊗
D
¯
W: (artinian kalgebras) → (abelian groups),
which is prorepresentable, formally smooth (i.e. a surjection A → B of ar
tinian kalgebras induces a surjection M⊗
D
¯
W(A) → M⊗
D
¯
W(B)) and has ﬁnite
dimensional “tangent space”
M⊗
D
¯
W(k[ε]/(ε
2
)) M/VM.
Such a functor is called a (smooth) formal group, and it turns out that any
smooth formal group can be represented as M⊗
D
¯
W for some M.
Examples (6.3.2) (i) M = W(k), F = σ, and V = pσ
−1
. Then
W(k) ⊗
D
¯
W
¯
W/(1 − V)
¯
W
¯
G
m
,
where
¯
G
m
is the formal multiplicative group and the map
¯
W →
¯
G
m
is deﬁned
by
E(a
0
t) E(a
n
t
p
n
) → E(a
0
) E(a
n
).
(ii) M = k[[V]] and F = 0. Then M⊗
D
¯
W
¯
W/V
¯
W
¯
G
a
, the formal
additive group.
As seen in characteristic zero,
¯
CH
2
X
is not in general prorepresentable. This
means that
¯
W and related functors are inadequate for its description. The trick
is to think of
¯
W as being associated to units in a polynomial ring, that is to K
1
.
This suggests considering “formal curves” on K
2
:
¯
CK
2
: (augmented artinian kalgebras) −→(abelian groups)
A −→ Ker (K
2
(A[X])
X·→0
−→K
2
(A)).
Complements on H
2
(K
2
) 75
One can also deﬁne for any kalgebra R, curves of length n on K
2
:
C
n
K
2
(R) = Ker (K
2
(R[X]/X
n+1
) → K
2
(R)),
and curves on K
2
:
CK
2
(R) = lim
←−−
n
C
n
K
2
(R).
There are bilinear pairings
big W(R) ×
¯
CK
2
(A) → K
2
(R⊗
k
A),
(1 − rT)
−1
× α(X) ·→ α(r)
and
CK
2
(R) × big
¯
W(A) → K
2
(R⊗
k
A),
β(x) × (1 − aT)
−1
·→ β(a).
(This makes sense because a is nilpotent.)
Also, when char k = p, there are splittings analogous to splittings on big W
and big
¯
W:
¯
CK
2
(A) =
∞
T
¯
CK
2
(A) (formal typical curves) and CK
2
(R) =
∞
TCK
2
(R) (typical curves). TCK
2
and T
¯
CK
2
are Dmodules, and the above
pairings give maps
Φ: W(R) ⊗
D
T
¯
CK
2
(A) →K
2
(R ⊗
k
A, R ⊗
k
M)
= Ker (K
2
(R ⊗
k
A) → K
2
(R)),
Ψ: TCK
2
(R) ⊗
D
¯
W(A) →K
2
(R ⊗
k
A, R ⊗
k
M).
This suggests that the role of Ω
1
in characteristic p is played by TCK
2
.
Pursuing this analogy, there is a map
d: W(R) → TCK
2
(R)
deﬁned by d(P(t)) = ¦P(t), t ¦, where P(t) is a monic power series and ¦, ¦ is the
Steinberg symbol. (This symbol can be shown to make sense even though t is
not a unit.) Stienstra writes ∂W
− →
(R) for the Dsubmodule generated by d(W(R))
in TCK
2
(R). His main local result is
Theorem (6.3.3) Let R be a regular local kalgebra and A an augmented
artinian local kalgebra. Then the composite
TCK
2
(R) ⊗
D
¯
W(A)
Ψ
→ K
2
(R⊗A, R⊗m) → Coker Φ
76 Lecture 6
is trivial on ∂W
− →
(R) ⊗
D
¯
W(A), and there is an induced exact sequence
0 → W(R) ⊗
D
T
¯
CK
2
(A) → K
2
(R ⊗
k
A, R ⊗
k
m)
→ (TCK
2
(R)/∂W
− →
(R)) ⊗
D
¯
W(A) → 0.
This whole process can be sheaﬁﬁed. One works with presheaves
“lim
←−−
” W
n
(O
X
), “lim
←−−
”(TC
n
K
2
(O
X
)/∂W
− →
(O
X
))
and cohomology groups
H
∗
(X, W) := lim
←−−
n
H
∗
(X, W
n
(O
X
)),
H
∗
(X, TCK
2
/∂W
− →
) := lim
←−−
n
H
∗
(X, TC
n
K
2
(O)/∂W
− →
(O)).
The former was ﬁrst studied by Serre [8]. It is known to satisfy (6.3.1) (i)–(iii),
and the formal groups H
∗
(X, W)⊗
D
¯
W have been studied by Artin and Mazur.
The groups H
∗
(X, TCK
2
/∂W
− →
) may according to Stienstra have Vtorsion and
so not satisfy (6.3.1)(ii). The main geometric consequence of (6.3.3) is
Theorem (6.3.4) Let X be a smooth complete surface over a perfect ﬁeld of
characteristic p > 0. Then there is an exact sequence for any augmented local
artinian kalgebra A
→ H
1
(X, (TCK
2
/∂W
− →
)⊗
D
¯
W(A))
δ
H
2
(X, W)⊗
D
T
¯
CK
2
(A)
→
¯
CH
2
X
(A) → H
2
(X, TCK
2
/∂W
− →
)⊗
D
¯
W(A) → 0.
If H
2
(X, W) has no ptorsion, the map δ is zero.
Remarks (6.3.5) (i) The point of taking X to be a surface is that H
2
is right
exact, so the ⊗
D
can be brought outside the cohomology.
(ii) It seems certain that H
2
(X, TCK
2
/∂W
− →
)/(Vtorsion) is the Dieudonn´ e
module of the Albanese variety of X, although no one has written down a
proof. In the absence of Vtorsion, this would make H
2
(X, TCK
2
/∂W
− →
)⊗
D
¯
W the
formal group at the origin of the Albanese. It is certainly true under mild hy
potheses that H
2
(X, TCK
2
/∂W
− →
) is “the part” of crystalline H
3
(X) with slopes
in [1, 2), but one doesn’t know precisely the correct hypotheses to impose at
this point.
(iii) The moral is that
¯
CH
2
X
is controlled by the Dmodules H
2
(X, W) and
H
2
(X, TCK
2
/∂W
− →
), and these two (playing the role of H
2
(X, O
X
) and H
2
(X, Ω
1
X
)
in characteristic zero) have a tendency to be Dieudonn´ e modules, that is to
satisfy (6.3.1)(i)–(iii).
Complements on H
2
(K
2
) 77
The non prorepresentable part of
¯
CH
2
x
diﬀers from a formal group in that
⊗
D
¯
W is replaced by ⊗
D
T
¯
CK
2
. It would be of interest to study algebraically
functors of the form M⊗
D
T
¯
CK
2
. Can one give an intrinsic characterization of
such a functor? Can one understand the link between the algebraic properties
of these functors and the geometric properties of the cycles?
References for Lecture 6
[1] M. Artin and B. Mazur, Formal groups arising from algebraic varieties,
Ann. Sci.
´
Ecole Norm. Sup. (4), 10 (1977), 87–132.
[2] S. Bloch, K
2
of Artinian Qalgebras, with applications to algebraic cy
cles, Comm. Algebra, 3 (1975), 405–428.
[3] S. Bloch, Algebraic Ktheory and crystalline cohomology, Inst. Hautes
´
Etudes Sci. Publ. Math., no. 47 (1977), 187–268 (1978).
[4] S. Bloch, Some formulas pertaining to the Ktheory of commutative group
schemes, J. Algebra, 53 (1978), 304–326.
[5] S. Bloch, Applications of the dilogarithm function in algebraic Ktheory
and algebraic geometry, pp. 103–114 in Proceedings of the International
Symposium on Algebraic Geometry (Kyoto Univ., Kyoto, 1977), Kinoku
niya Book Store, Tokyo (1978).
[6] J. Stienstra, The formal completion of the second Chow group: a K
theoretic approach, pp. 149–168 in Journ´ ee de G´ eom´ etrie Algebrique de
Rennes (Rennes, 1978), Vol. II, Asterisque, 64 (1979).
[7] J. Stienstra, Deformations of the second Chow group, Thesis,
Utrecht (1978).
[8] J.P. Serre, Sur la topologie des vari´ et´ es algebriques en caract´ eristique p,
pp. 24–53 in Symposium Internacional de Topologia Algebraica, Univer
sidad Nacional Aut´ onoma de Mexico and UNESCO, Mexico City (1958).
7
Diophantine questions
A smooth projective algebraic surface X deﬁned over a ﬁeld k is said to be
rational if its function ﬁeld becomes rational after extension to the algebraic
closure
¯
k of k, that is
¯
k(X)
¯
k(t
1
, t
2
) with t
1
, t
2
independent transcendentals. If
k(X) k(t
1
, t
2
) we say X is krational. The Chow group CH
0
(X) is deﬁned as
usual, either geometrically
CH
0
(X) = Coker
_
_
X
1
k(x)
∗
→
_
X
2
Z
_
or algebraically by H
2
(X, 7
2
). The degree map deg: CH
0
(X) → Z is a bit
tricky. It can be deﬁned either as the composition
CH
0
(X) → CH
0
(
¯
X)
deg
Z, where
¯
X = X ×
Spec k
Spec
¯
k,
or directly by assigning to (x), x ∈ X
2
, deg(x) = [k(x) : k]. Let A
0
(X) ⊂ CH
0
(X)
be the kernel of deg.
Proposition (7.1) If X is krational, then A
0
(X) = (0).
Proof Let f : P
2
k
X be a birational map, deﬁned over k, and let Γ be
its graph. From the geometric deﬁnition of CH
0
one sees easily that a zero
cycle can be moved oﬀ of any given proper closed subscheme of X or P
2
. The
correspondences
CH
0
(X)
Γ
t
∗
−→ CH
0
(P
2
)
Γ
∗
−→ CH
0
(X)
are therefore inverse isomorphisms. One knows CH
0
(P
2
) Z (exercise).
The same argument shows, of course, that A
0
(X) is a kbirational invariant.
To simplify exposition we will tend to assume k is either a local or a global
ﬁeld. By a technique of Manin involving Brauer groups, we will show A
0
(X)
80 Lecture 7
is not in general zero for X a rational surface. We then introduce the N´ eron–
Severi torus, T = T
X
, which is the ktorus with character group the Gal(
¯
k/k)
module
N = NS(
¯
X) = CH
1
(
¯
X).
The rationality of X implies that N is a ﬁnitely generated free abelian group.
By deﬁnition,
T(
¯
k) = Hom
Z
(N,
¯
k
∗
)
with the natural Galois action. This torus has been studied by ColliotTh´ el` ene
and Sansuc [2]. They consider a map (as usual H
∗
(k, M) = H
∗
Gal
(Gal(
¯
k/k), M))
Φ: A
0
(X) → H
1
(k, T(
¯
k))
and show (assuming k to be a local or global ﬁeld) that Image(Φ) is ﬁnite.
Using Ktheory, we construct an exact sequence (
¯
F =
¯
k(X))
(7.2) H
0
(k, T(
¯
k)) −→ H
1
(k, K
2
(
¯
F)/K
2
(
¯
k))
µ
−→ A
0
(X)
Φ
−→ H
1
(k, T(
¯
k)) −→ H
2
(k, K
2
(
¯
F)/K
2
(
¯
k)).
When X has a rational pencil of genuszero curves (X = conic bundle surface)
we will show for k local or global that H
1
(k, K
2
(
¯
F)/K
2
(
¯
k)) is a ﬁnite 2group,
so A
0
(X) is ﬁnite in this case. The heart of the argument is an analog of the
Eichler norm theorem describing the image of the reduced norm
Nrd: A
∗
→ k(t)
∗
,
when A is a quaternion algebra over the rational function ﬁeld k(t) (for k local
or global). As a corollary, we show that if X is a smooth cubic surface (i.e.
X ⊂ P
3
k
is cut out by a homogeneous function of degree 3), then the only
possible inﬁnite torsion in A
0
(X) is 3torsion.
Conjecture (7.3) A
0
(X) is ﬁnite for any rational surface over a local or
global ﬁeld.
Question (7.4) As the reader will see, the structure of A
0
(X) is somehow
linked to the places of k where X has bad reduction. Does the order of
Image(A
0
(X) → A
0
(X
k
ν
))
for the various completions of k have anything to do with the local factors at
bad places in the zeta function of X?
Diophantine questions 81
To convince ourselves there is some interest in the subject, we must show
that A
0
(X) is not in general zero for a rational surface. The idea is due to Manin,
but I want to give a very “motivic” (some might say very pedantic) exposition.
Let ( be the category of ﬁnite separable extension ﬁelds of k. (We avoid the
temptation to work with the larger category of all kalgebras because of possi
ble technical problems.) We consider various covariant functors on (
X: ( → (sets), k
·
→ X(k
·
) (functor of points),
Br : ( → (abelian groups), k
·
→ Br(k
·
) = H
2
(k
·
,
¯
k
∗
),
A
0
(X) ⊂ CH
0
(X) : ( → (abelian groups), k
·
→ A
0
(X
k
· ) ⊂ CH
0
(X
k
· ).
There is a natural morphism of functors X × X → A
0
, where (x
1
, x
2
) ·→
(x
1
) − (x
2
). An element a ∈ H
2
et
(X, G
m
)/H
2
(k,
¯
k
∗
) deﬁnes a morphism of func
tors ˜ a : X × X → Br,
˜ a(x
1
, x
2
) = a(x
1
) a(x
2
)
−1
∈ Br(k
·
), x
1
, x
2
∈ X(k
·
).
Manin’s observation is that there exists a morphism A
0
(X)
ψ
a
Br making
the diagram
A
0
(X)
ψ
a
X × X
˜ a
Br
commute. In particular, if ˜ a
k
: X(k) × X(k) → Br(k) is not zero, it follows that
the group A
0
(X
k
) isn’t either.
To deﬁne ψ
a
, we associate to a zerocycle
_
n
i
(x
i
) deﬁned over k
·
ψ
a
_
n
i
(x
i
)
_
=
_
i
cor
k
·
(x
i
)/k
· (a(x
i
))
n
i
∈ Br(k
·
).
Given a correspondence deﬁned over k
·
Γ
p
1
p
2
P
k
· X
k
·
we have (using the existence of transfer for the Brauer group)
ψ
a
(p
2∗
p
∗
1
((0) − (∞))) = ψ
p
∗
2
(a)
(p
∗
1
((0) − (∞)))
= ψ
p
1∗
p
∗
2
(a)
((0) − (∞)).
82 Lecture 7
But p
1∗
p
∗
2
(a) ∈ Br(P
1
k
·
)/Br(k
·
) = (0), so ψ
a
is trivial on cycles rationally equiv
alent to zero, as desired.
Lemma (7.5) Let X be a rational surface. Then
Br(X)/Br(k) H
1
(Gal(
¯
k/k), Pic(
¯
X)).
Proof The Brauer group of a surface is a birational invariant, so
H
2
et
(
¯
X, G
m
) H
2
et
(P
2
¯
k
, G
m
) = (0)
(Grothendieck [11]). The Hochshild–Serre spectral sequence
E
p,q
2
= H
p
(Gal(
¯
k/k), H
q
et
(
¯
X, G
m
)) ⇒ H
p+q
et
(X, G
m
)
gives rise, therefore, to an exact sequence
Br(k) → Br(X) → H
1
(Gal(
¯
k/k), H
1
(
¯
X, G
m
)) → H
3
(Gal(
¯
k/k),
¯
k
∗
).
The righthand term vanishes for k local or global.
Remarks on calculating H
1
(Gal(
¯
k/k), Pic(
¯
X)) If k ⊂ K is a ﬁnite extension
such that Gal(
¯
k/K) acts trivially on Pic(
¯
X), one checks easily
H
1
(Gal(
¯
k/k), Pic(
¯
X)) H
1
(Gal(K/k), Pic(
¯
X)).
Suppose now F is a ﬁnitely generated free abelian group of divisors on X
K
such that F → Pic(X
K
). Let
L = ¦ f ∈ K(X)
∗
¦ ( f ) ∈ F¦
and consider the exact sequence
0 → L/K
∗
→ F → Pic(X
K
) → 0.
One has H
1
(Gal(K/k), F) = (0) (F is a permutation module) whence an exact
sequence
0 → Br(X)/Br(k)
∂
H
2
(Gal(K/k), L/F
∗
) → H
2
(Gal(K/k), F).
In simple situations, Manin actually writes down 2cochains f
s,t
with val
ues in L ⊂ K(X)
∗
such that the f
s,t
mod(K
∗
) represent the image of Br(X) in
H
2
(Gal(K/k), L/K
∗
). Given a zerocycle
_
n
i
(x
i
) deﬁned over k and supported
in the complement of the support of L, we have ψ
a
(
_
n
i
(x
i
)) represented by
the 2cocycle
_
i
N
K(x
i
)/K
( f
s,t
(x
i
))
n
i
with values in K
∗
. Here a ∈ Br(X)/Br(k) with ∂a represented by f
s,t
.
Diophantine questions 83
When K/k is cyclic, one can use the periodicity of Galois cohomology to
identify
H
2
(Gal(K/k), L/K
∗
) L
Gal(K/k)
/k
∗
N
K(X)/k(X)
(L),
Br(k) ⊃ H
2
(Gal(K/k), K
∗
) k
∗
/N
K/k
(K
∗
).
In this case one can write down f ∈ L
Gal(K/k)
⊂ k(X)
∗
such that
_
n
i
(x
i
) is not
rationally equivalent to 0 if
_
i
N
k(x
i
)/k
f (x
i
)
n
i
NK
∗
⊂ k
∗
.
Example (7.6) Chˆ atelet surfaces are deﬁned in aﬃne coordinates by t
2
1
−at
2
2
=
(x − a
1
)(x − a
2
)(x − a
3
), for a, a
i
∈ k
∗
, a
i
a
j
(Chˆ atelet [1], Manin [5]). In this
case we can take K = k(
√
a). Manin shows that given kpoints p
1
, p
2
on the
surface such that the functions
f
1
=
x − a
1
x − a
3
, f
2
=
x − a
2
x − a
3
are deﬁned and invertible at p
1
and p
2
, if at least one of the ratios
f
i
(p
1
)
f
i
(p
2
)
N
K/k
(K
∗
), i = 1, 2,
then p
1
is not rationally equivalent to p
2
. Take for example
t
2
1
− 5t
2
2
= x(x − 1)(x − 7), k = Q.
For each value of x 0, 1, 7 we obtain a conic curve, the rationality of which
(over Q) is determined by the Hilbert symbol
(5, x(x − 1)(x − 7)).
To obtain rational ﬁbres take x = −1,
(5, (−1)(−2)(−8)) = (5, −1) = 1 (−1 ≡ 2
2
mod 5)
and x = 31,
(5, 31 30 24) ≡ (5, 31 5) = 1
(5 ≡ 6
2
mod 31 and 31 ≡ 1 mod 5; also (5, 5) = (5, −1) = 1). Let p
1
be a
rational point in the curve over x = −1, p
2
a rational point over x = 31:
f
1
(p
1
) =
−1
−8
≡ 2 mod N
K/k
k
∗
,
f
1
(p
2
) =
31
24
≡
31
6
mod N
K/k
k
∗
.
To show p
1
and p
2
not rationally equivalent, we may at this point pass to a
84 Lecture 7
larger ﬁeld, working with the 5adic completions k
5
= Q
5
and K
5
= Q
5
(
√
5).
The ﬁeld K
5
is tamely ramiﬁed over k
5
, so local class ﬁeld theory implies
N
K
5
/k
5
K
∗
5
= 5
Z
(1 + 5O
k
5
)
∗
µ
2
k
5
, µ
k
5
⊂ k
∗
5
roots of 1.
Since 2 is not a square mod 5, we get f
1
(p
1
) N
K
5
/k
5
K
∗
5
. On the other hand,
31
6
≡ 1 mod 5, so
31
6
∈ N
K
5
/k
5
K
∗
5
. We conclude that p
1
is not rationally equiva
lent to p
2
, even over Q
5
.
Our objective now is to construct the fundamental exact sequence (7.2).
Proposition (7.7) Let X be a smooth surface deﬁned over k, and let Y be
obtained from X by blowing up a kpoint. Then Γ(X, 7
2
) = Γ(Y, 7
2
) and
H
1
(X, 7
2
) ⊕ k
∗
= H
1
(Y, 7
2
).
Proof Let F = k(X) and let e be the generic point of the exceptional divisor
E on Y. Consider the diagram
(7.7.1)
0 0
0 k
∗
k(e)
∗
_
E
1
0
Z 0
K
2
(F)
_
y∈Y
1
k(y)
∗
_
Y
2
Z
K
2
(F)
_
x∈X
1
k(x)
∗
_
X
2
Z.
0 0
(As usual, the superscript 0 indicates that the righthand column is in degree 0.)
The middle and righthand columns and the upper row are exact. The middle
and lower rows are complexes whose cohomology is identiﬁed with the coho
mology of 7
2
on Y (resp. X) (cf. Lecture 4). These complexes are covariant
functorial for proper maps. Using the contravariant functoriality of Γ( , 7
2
),
Diophantine questions 85
we get a commutative diagram
(Y, 7
2
)
7
2
(F),
(X, 7
2
)
whence Γ(X, 7
2
) Γ(Y, 7
2
). The assertions about H
1
follow from diagram
chasing on (7.7.1).
Deﬁnition (7.8) A rational surface X deﬁned over k will be said to be split
over an extension ﬁeld k
·
if there exists a diagram
X
·
k
·
p q
X
k
· P
2
k
·
such that the arrows p and q arise from a succession of blowings up of k
·

points.
It is known that there always exists k
·
ﬁnite over k splitting X.
Proposition (7.9) Let X be a krational surface split over k. Then Pic(X)
Pic(
¯
X) and H
1
(X, 7
2
) Pic(X)⊗
Z
k
∗
.
Proof The assertions of the proposition are true for X = P
2
. (For compu
tations of H
p
(P
n
, 7
q
) see Gillet [14], Sherman [15].) One now applies (7.7)
noting that Pic(Y) Pic(X) ⊕ Z, and checking that the extra factor of k
∗
in
H
1
(Y, 7
2
) comes from the map
Pic(Y) ⊗ k
∗
= H
1
(Y, G
m
) ⊗ k
∗
→ H
1
(Y, 7
2
)
deduced from the Steinberg symbol G
m
⊗ k
∗
→ K
2
.
Now let X be a rational surface deﬁned over k, and let k
·
be a Galois exten
sion of k splitting X. Note in particular that X will be k
·
rational, so A
0
(X
k
· ) =
(0) by (7.1). Write X
·
= X
k
· , F
·
= k
·
(X), and N = Pic(X
·
). We get exact
sequences (deﬁning Z)
(7.10) 0 → Z →
_
x
·
∈X
·1
k
·
(x
·
)
∗
→
_
X
·2
0
Z → 0,
0 → K
2
(F
·
)/K
2
(k
·
) → Z → N⊗k
·∗
→ 0.
86 Lecture 7
One has
H
∗
(k
·
/k,
X
·1 k
·
(x
·
)
∗
)
x∈X
1
H
∗
(k
·
/k
x
, k
·
(x)
∗
)
by Shapiro’s lemma, where k
x
= k
·
∩ k(x). Also
Γ(k
·
/k,
X
·2
0
Z)
X
2
0
Z.
It follows fromthe ﬁrst sequence in equation (7.10) and Hilbert 90 that A
0
(X)
H
1
(k
·
/k, Z). Substituting in the second sequence yields
(7.11) Γ(k
·
/k, N⊗k
·
) → H
1
(k
·
/k, K
2
(F
·
)/K
2
(k
·
)) → A
0
(X)
→ H
1
(k
·
/k, N⊗k
·
) → H
2
(k
·
/k, K
2
(F
·
)/K
2
(k
·
)) .
Finally, (7.11) may be identiﬁed with (7.2) by observing that the intersection
pairing on N is perfect, so N⊗k
·∗
Hom(N, k
·∗
) as Galois modules.
Let v be a nonarchimedean place of the ﬁeld k. A surface X deﬁned over k
is said to have good reduction at v if there exists
˜
X → Spec(O
v
) smooth and
projective with
˜
X
k
v
X
k
v
. For k a global ﬁeld, any smooth ksurface will have
good reduction at all but a ﬁnite number of places.
Theorem (7.12) Let X be a rational surface deﬁned over a local or global
ﬁeld k.
(i) If X has good reduction at a nonarchimedean place v of k, then the map
Φ
v
: A
0
(X
k
v
) → H
1
(
¯
k
v
/k
v
, T(
¯
k
v
)) deduced from (7.2) is zero.
(ii) Quite generally, the image of Φ: A
0
(X) → H
1
(
¯
k/k, T(
¯
k)) is ﬁnite.
Proof Local duality (Serre [13]) implies that H
1
(
¯
k/k, T(
¯
k)) is ﬁnite for k a
local ﬁeld. Global duality (Nakayama [12]) says for k global, the kernel of
H
1
(
¯
k/k, T(
¯
k)) →
v
H
1
(
¯
k
v
/k
v
, T(
¯
k
v
)) is ﬁnite. These results show that (i) ⇒
(ii), so it suﬃces to prove (i).
Lemma (7.12.1) Let R be a complete discrete valuation ring with residue
ﬁeld f and quotient ﬁeld k. Let f :
˜
X → Spec R be smooth, projective, geomet
rically irreducible, and assume the closed ﬁbre X
0
is a rational surface split
over f. Then the generic ﬁbre X is rational and split over k.
Proof An fvalued point of X
0
lifts by smoothness to an Rvalued point of
˜
X,
so a morphism X
0
(1) → X
0
obtained by blowing up an fpoint lifts to a diagram
X
0
(1) ⊂
˜
X(1)
blow up an Rpoint
X
0
⊂
˜
X.
Diophantine questions 87
We may thus build a ﬁgure
˜
Y ⊃
blowings up
Y
0
blowings down
˜
X ⊃ X
0
P
2
f
⊂ R
2
R
.
Let E
0
⊂ Y
0
be an exceptional curve of the ﬁrst kind. There exists a unique
family of exceptional curves of the ﬁrst kind
˜
E ⊂
˜
Y lifting E
0
. Indeed, the
normal bundle of E
0
P
1
in Y
0
is O(−1), so both zeroeth and ﬁrst cohomol
ogy groups of the normal bundle vanish, implying uniqueness and existence of
liftings.
We may nowblowdown
˜
E on
˜
Y in much the same way as E
0
on Y
0
. Iterating,
we get a diagram
Y
0
⊂
˜
Y
P
2
f
⊂
˜
Z
where
˜
Z is smooth and projective over Spec R with closed ﬁbre P
2
f
. Since P
2
f
is rigid, it follows that
˜
Z P
2
R
. Passing now to the generic ﬁbres we get X
k
←
Y
k
→ Z
k
P
2
k
, both morphisms being obtained by successive blowings up of
kpoints.
We return now to the proof of (7.12)(i). Let k be the local ﬁeld, o ⊂ k the
ring of integers, and let
˜
X → Spec o
be smooth and projective with special ﬁbre X
0
and generic ﬁbre X. Let k
·
be an
unramiﬁed extension with residue ﬁeld f
·
splitting X
0
. The units o
·∗
in the ring
of integers o
·
⊂ k
·
are cohomologically trivial [13] so N ⊗ o
·∗
is cohomolog
ically trivial (Serre [13], theorem 9, p. 152). (Here N = Pic(X
k
· ) = Pic(X
¯
k
).)
The sequence
0 → N⊗o
·∗
→ N⊗k
·∗
→ N → 0
yields H
1
(k
·
/k, N⊗k
·
) H
1
(k
·
/k, N).
Because N Pic(X
0,f
· ), there is an exact sequence of Gal(k
·
/k) (= Gal(f
·
/f))
modules (F
·
0
= f
·
(X
0
))
0 → F
·∗
0
/f
·∗
→ Div(X
0,f
· ) → N → 0.
88 Lecture 7
Since Div(X
0,f
· ) is a permutation module, we obtain a diagram with exact rows
(F
·
= k
·
(X))
(7.12.2)
A
0
(X)
Φ
H
1
(k
·
/k, N⊗k
·
)
θ(7.2)
H
2
(k
·
/k, K
2
(F
·
)/K
2
(k
·
))
tame
0 H
1
(k
·
/k, N)
∂
H
2
(k
·
/k, F
·∗
0
/f
·∗
).
Here tame: K
2
(F
·
) → F
·∗
0
is the tame symbol associated to the divisor X
0,f
·
on
˜
X ×
Spec o
Spec o
·
. The fact that Φ = 0 now follows from
Lemma (7.12.3) The square in (7.12.2) is commutative up to sign.
Proof A point x
·
∈ X
1
k
·
can be viewed as the generic point of a divisor ¦x
·
¦
on
˜
X × Spec o
·
ﬂat over Spec o
·
with function ﬁeld k
·
(x
·
). The intersection
¦x
·
¦ ∩ X
0,f
· is a divisor on ¦x
·
¦, so we obtain a valuation map
α
x
· : k
·
(x
·
)
∗
→ Div(X
0,f
· ).
Adding up over various x
·
, we obtain a diagram
K
2
(F
·
)
2 _
1 _
_
x
·
∈X
1
k
·
k
·
(x
·
)
∗
α=α
x
·
F
·∗
0
Div(X
0,f
· ).
The maps labeled
1 _
and
2 _
are tame symbols and the diagram commutes up to
a sign. Since Z ⊂
k
·
(x
·
)
∗
(cf. 7.10), we obtain a commutative diagram of
Galois modules:
0 K
2
(F
·
)/K
2
(k
·
)
1 _
Z
α
N ⊗ k
·∗
1⊗valuation
0
0 F
·∗
0
/f
·∗
Div(X
0,f
· ) N 0.
The assertion regarding (7.12.2) is now straightforward.
This completes the proof of (7.12). Returning to the exact sequence (7.2),
we see that ﬁniteness for A
0
(X) for X/k local or global will follow whenever
we can prove H
1
(
¯
k/k, K
2
(
¯
F)/K
2
(
¯
k)) is ﬁnite, for
¯
F =
¯
k(X) and X a rational
surface over k.
Diophantine questions 89
Deﬁnition (7.13) X is a conic bundle surface if there exists a rational pencil
of genuszero curves on X.
Examples (7.13.1) (i) The Chˆ atelet surfaces discussed in (7.6) are conic bun
dle surfaces. More generally any surface
t
2
1
− f (x) t
2
2
= g(x)
is a conic bundle surface.
(ii) A cubic surface (smooth cubic hypersurface in P
3
) which contains a line
deﬁned over k is a conic bundle surface. Indeed, the residual intersection of
X with a pencil of planes through is a pencil of conics.
Before stating our main result, we recall that the Milnor ring [8] k
∗
(k) of a
ﬁeld k is the graded Z/2Zalgebra with generators elements (a) ∈ k
1
(k), for
a ∈ k
∗
, and relations 2(a) = 0, (ab) = (a) + (b), and (a)(1 − a) = 0, for
a 0, 1.
Theorem (7.14) Let X be a conic bundle surface over a ﬁeld k which is
either a local ﬁeld, a global ﬁeld, or a C
i
ﬁeld [13] for i ≤ 3. Then the group
H
1
(
¯
k/k, K
2
(
¯
F)/K
2
(
¯
k)) is a subquotient of k
3
(k).
1
As a consequence of results in Milnor [8] one has
k
3
(k) =
⎧
⎪
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎩
0 k C
2
ﬁeld,
0 k local R,
Z/2Z k = R,
⊕
v real place
Z/2Z k global.
Combining (7.1), (7.13) and the above results of Milnor, one ﬁnds
Corollary (7.14.1) If X is a conic bundle surface over a local or global ﬁeld,
then A
0
(X) is ﬁnite.
Corollary (7.14.2) If X is a cubic surface over a local or global ﬁeld, then
A
0
(X)⊗
Z
Z[
1
3
] is ﬁnite. In particular if X is deﬁned over R, then A
0
(X) is ﬁnite.
Proof of (7.14.2) Let k
·
/k be a ﬁnite Galois extension splitting X. The ex
istence of a transfer A
0
(X
k
· ) → A
0
(X
k
) implies that [k
·
: k]A
0
(X
k
) = (0).
It suﬃces therefore to show the power torsion subgroup A
0
(X)() is ﬁnite
for any prime 3. Let k
⊂ k
·
be the ﬁxed ﬁeld of the Sylow subgroup
of Gal(k
·
/k). Then does not divide [k
: k] so a transfer argument shows
1
Combining their own ideas with arguments given in the text, ColliotTh´ el` ene and Sansuc have
shown H
1
(
¯
k/k, K
2
(
¯
F)/K
2
(
¯
k)) ⊂ K
3
(k) for conic bundle surfaces X, and this cohomology group
is zero for k local or global unless χ(R) = φ for all real places of k. The group is nonzero for
k = R, χ = P
1
× C, where C is the nontrivial conic over R.
90 Lecture 7
that A
0
(X)() → A
0
(X
k
)(). We may assume, therefore, that Gal(k
·
/k) is an
group.
Recall X
¯
k
has 27 lines on it, all of which we may assume (taking k
·
large to
begin with) are deﬁned over k
·
. The group Gal(k
·
/k) will necessarily leave a
line ﬁxed for 3, so X is a conic bundle surface (7.13.1)(ii).
Proof of (7.14) Fix a conic bundle structure on X, that is a rational map
π: X → P
1
k
with ﬁbre of genus 0. Blowing up on X, we may assume π is
everywhere regular. Let K = k(P
1
) and
¯
K =
¯
k(P
1
). Since
¯
K is a C
1
ﬁeld
[13] the generic ﬁbre of π pulled back to
¯
k, X
¯
K
, is isomorphic to P
1
¯
K
. Writ
ing
¯
F =
¯
k(X) =
¯
K(X
¯
K
) we have
(7.14.3) 0 → K
2
(
¯
K) → K
2
(
¯
F) →
_
¯ x∈X ¯
K
¯
K( ¯ x)
∗
N
−→
¯
K
∗
→ 0,
where N is the norm map [7]. Similarly
(7.14.4) K
2
(K) K
2
(
¯
k)⊕
_
0
¯ y∈P
1
¯
k
¯
k
∗
,
where as usual
0
¯
k
∗
= Ker(
¯
k
∗
→
¯
k
∗
).
Let n ⊂ K
∗
be the image of the norm map from
_
_
¯ x∈X ¯
K
¯
K( ¯ x)
∗
_
Gal(
¯
k/k)
=
_
x∈X
k
K(x)
∗
.
From (7.14.3) and (7.14.4) we get an exact sequence
(7.14.5) 0 → H
1
(
¯
k/k, K
2
(
¯
F)/K
2
(
¯
k)) → K
∗
/n
ψ
−→
_
y∈P
1
k
Br(k(y)).
Fix a, b ∈ K
∗
such that X
K
is isomorphic to the conic curve
X
K
: T
2
0
− aT
2
1
− bT
2
2
= 0.
We will work with (a) (b) ∈ k
2
(K), as well as with the quaternion algebra A
over K deﬁned by T
2
= a, U
2
= b, and TU = −UT.
Lemma (7.14.6) Let Nrd: A
∗
→ K
∗
be the reduced norm. Then
Nrd(A
∗
) ⊂ n.
Proof An extension K
·
/K splits A (i.e. A⊗
K
K
·
End(K
·
⊕ K
·
)) if and only
if X
K
has a K
·
point. If α ∈ A, α K, then K(α) splits A, so X
K
has a K(α)
point. Since the reduced norm on A coincides on K(α) ⊂ A with the ﬁeld norm,
we get Nrd(α) ∈ n. Finally, if α ∈ K, Nrd(α) = α
2
∈ n, since X
K
obviously
contains points of degree 2 over K.
Diophantine questions 91
We now can rewrite (7.14.5)
0 H
1
(
¯
k/k, K
2
(
¯
F)/K
2
(
¯
k)) K
∗
/n
ψ
_
y∈P
1
k
Br(k(y))
K
∗
/Nrd(A
∗
).
ψ
·
(7.14.7)
and it will suﬃce to show Ker ψ
·
→ k
3
(k).
Lemma (7.14.8) The map (a)(b) : K
∗
→ k
3
(K), where c ·→ (c)(a)(b),
contains Nrd(A
∗
) in its kernel.
Proof Let β = Nrd(α) ∈ Nrd(A
∗
). We may assume α K
∗
, so K(α) splits A.
This implies (a) (b) → 0 in k
2
(K(α)) (Milnor [9], p. 152). The projection
formula [7] implies
(Nrd(α)) (b)(a) = N
K(α)/K
((α)) (b) (a)
= N
K(α)/K
((α) (b) (a)) = 0.
Lemma (7.14.9) The diagram
K
∗
/Nrd(A
∗
)
ψ
·
(7.14.7)
(a)(b)
_
y∈P
1
k
Br(k(y))
k
3
(K)
tame symbol
_
y∈P
1
k
k
2
(k(y))
Galois symbol
commutes.
Lemma (7.14.10) The maps in the above square satisfy:
(i) “Galois symbol” is injective,
(ii) “tame symbol” has kernel k
3
(k) ⊂ k
3
(K),
(iii) “ (a) (b)” is injective.
Note that these two lemmas suﬃce to complete the proof of (7.14). Note
also their geometric content. Let
_
n
i
(y
i
) be a zerocycle of degree zero on
P
1
k
. Assume the ﬁbres of π: X → P
1
over the y
i
are all smooth. A necessary
condition for there to exist a cycle Z on X such that π
∗
z =
_
n
i
(y
i
) is that n
i
should be even whenever π
−1
(y
i
) is nonsplit (i.e. P
1
k(y
i
)
). If ( f ) =
_
n
i
(y
i
),
92 Lecture 7
(7.14.9) implies
ψ
·
( f ) =
n
i
[π
−1
(y
i
)], where [π
−1
(y
i
)] = class of
Severi–Brauer variety in Br(k(y
i
)).
Thus, in the local or global case when k
3
(k) is ﬁnite, we have divisors of
degree zero and even order at nonsplit places modulo norms of zerocycles
rationally equivalent to on X = ﬁnite group. Finally the reader may wish to
compare this chain of ideas with the Eichler norm theorem describing the im
age of the norm mapping for a divisor algebra over a local or global ﬁeld.
Proof of (7.14.10) Injectivity of the Galois symbol follows from Lam [10],
given our assumptions about k, and the fact that Ker(tame) = k
3
(k) is proven
in [8]. To show injectivity of multiplication by (a) (b), we consider the map
deﬁned in Milnor [8]
k
3
(K) → I
3
/I
4
,
(a)(b)(c) → ((c) + (−1))((b) + (−1))((a) + (−1)),
where I is the augmentation ideal in the Witt ring of K. The above quadratic
form is a Pﬁster form in the terminology of [10]. In particular, this form lies in
I
4
if and only if it is hyperbolic ([10] cor. 3.4, p. 290; thm. X.6.17, p. 367, in
the 2005 book). Thus (a)(b)(c) = 0 implies there exist x
1
, . . . , x
8
∈ K not
all 0 such that
abc x
2
1
− ab x
2
2
− ac x
2
3
− bc x
2
4
+ a x
2
5
+ b x
2
6
+ c x
2
7
− x
2
8
= 0.
Formally, then, we may factor and write
c =
abx
2
2
− ax
2
5
− bx
2
6
+ x
2
8
abx
2
1
− ax
2
3
− bx
2
4
+ x
2
7
.
Notice both numerator and denominator are norms from A
∗
. We may assume
A End(K ⊕ K), else X
K
P
1
K
, X is krational, and the whole discussion
is silly. Thus the denominator in the above expression vanishes if and only if
x
1
= x
3
= x
4
= x
7
= 0. But vanishing of the denominator implies vanishing of
the numerator and hence all the x
i
, a contradiction. Hence neither numerator
nor denominator vanishes and we have written c as a norm from A
∗
.
Proof of (7.14.9)
Lemma (7.14.11) Let X
K
be a conic curve over Spec K with
¯
K/K Galois
splitting X
K
. Let [X
K
] ∈ H
2
(
¯
K/K,
¯
K
∗
) be the class of X
K
as a Severi–Brauer
Diophantine questions 93
variety, and assume [X
K
] 0. Let
¯
F = quotient ﬁeld of X
¯
K
and consider the
exact sequence of Gal(
¯
K/K)modules
0 →
¯
K
∗
→
¯
F
∗
→
_
x∈X ¯
K
Z → Z → 0.
Then [X
K
] = ∂
2
∂
1
(1), where ∂
i
are the boundary maps associated to this exact
sequence.
Proof
_
_
x∈X ¯
K
Z
_
Gal(
¯
K/K)
=
_
x∈X
K
Z.
Since any x ∈ X
K
has even degree over Spec K (this follows from a norm
argument using the fact that X
K
splits over K(x)), we get an exact sequence
0 → Z/2Z → H
2
(
¯
K/K,
¯
K
∗
) → H
2
(
¯
F/F,
¯
F
∗
),
1 ·→ ∂
2
∂
2
(1).
Since [X
K
] → 0 in H
2
(
¯
F/F,
¯
F
∗
), we are done.
Tensoring the sequence in (7.14.11) with K
∗
and using the symbol map, we
obtain a commutative diagram
0
¯
K
∗
⊗
Z
K
∗
¯
F
∗
⊗K
∗
_
¯ x∈X
1
¯
K
K
∗
K
∗
0
0 K
2
(
¯
K) K
2
(
¯
F)
_
¯ x
¯
K( ¯ x)
∗
¯
K
∗
0.
Writing ∂
·
i
for the boundary maps on cohomology associated to the bottom
row, we ﬁnd a commutative triangle
K
∗
[X
k
]⊗
∂
·
0
∂
·
1
H
2
(
¯
K/K,
¯
K
∗
)⊗
Z
K
∗
symbol
H
2
(
¯
K/K, K
2
(
¯
K)).
Now take K = k(P
1
),
¯
K =
¯
k(P
1
) and compose with the tame symbol to get a
94 Lecture 7
commutative square
K
∗
ψ
[X
K
]⊗
_
y∈P
1
k
Br(k(y))
H
2
(
¯
k/k,
¯
K
∗
)⊗K
∗
H
2
(
¯
k/k, K
2
(
¯
K)).
tame
The proof of (7.14.9) now follows from
Lemma (7.14.12) Let k be a ﬁeld of characteristic 2,
¯
k = separable closure
of k. Let K be an extension ﬁeld of transcendence degree 1 over k, and write
¯
K = K
¯
k. Let a, b ∈ K
∗
and write (a)(b) ∈ k
2
(K), (a, b) ∈
2
Br(K). Then
Br(K) H
2
(
¯
k/k,
¯
K
∗
).
Moreover, if y is a place of K over k with residue ﬁeld k(y), the diagram
(7.14.13)
H
2
(k/k, K
∗
)⊗K
∗
K
∗
(a,b)⊗
(a)(b)
H
2
(
¯
k/k, K
2
(
¯
K))
T
1
k
3
(K)
T
2
Br(k(y))
(where T
1
and T
2
are tame symbols) commutes.
Proof Replacing k by its perfect closure, we may assume k(y) separable over
k. Next replacing K by its completion at y we may suppose k(y) ⊂ K and
¯
K =
¯
K
i
, one copy for each place lying over y. Now replacing the Galois
group by the decomposition group for one of the
¯
K
i
we may assume k = k(y),
K = k((π)).
We now have split exact sequences
0 o
∗
K
∗
Z 0, 0 ¯ o
∗
¯
K
∗
Z 0.
π 1
Note that 1 + π¯ o is cohomologically trivial, so
H
2
(
¯
k/k,
¯
K
∗
) H
2
(
¯
k/k,
¯
k
∗
) ⊕ H
2
(
¯
k/k, Z).
Diophantine questions 95
Case 1 For a, b ∈ o, let a
0
, b
0
∈ k
∗
denote the mod π reductions of a, b. In this
case (a, b) = (a
0
, b
0
) ∈ H
2
(
¯
k/k,
¯
k
∗
). It is easy enough to see that going either
way around (7.14.13), f ∈ K
∗
gets taken to ord( f ) (a
0
, b
0
) ∈ Br(k).
Using linearity, it remains only to consider
Case 2 Suppose b = π, a ∈ o
∗
. In this case let G = Gal(
¯
k/k) and let ρ :
G×G → Z be a 2cocycle representing the image of a under the composition
K
∗
→ H
1
(
¯
K/K, µ
2
)
δ
→ H
2
(
¯
K/K, Z). (7.14.14)
a ·→ χ
a
Here δ is the coboundary from the exact sequence
0 → Z → Z → Z/2Z → 0.
Then (a, b) is represented by the cocycle
G × G →
¯
K
∗
, (g
1
, g
2
) ·→ π
ρ(g
1
,g
2
)
.
Indeed, one knows (cf. Serre [13], p. 214) that
(a, π) = (δ χ
a
) π
ρ(,)
.
With reference to (7.14.13), we must show
T
1
¦π
ρ(g
1
,g
2
)
, f ¦ = T
2
((a)(π)( f )).
If F = −π this is clear as both sides are trivial. If, on the other hand, f is a
unit with residue class f
0
, we reduce to showing that (a
0
)( f
0
) ∈ k
2
(k) maps
to the element in Br(k) represented by the cocycle f
ρ(g
1
,g
2
)
0
. This follows as
above with K in (7.14.14) replaced by k and
¯
K replaced by
¯
k. This completes
the proof of (7.14.12) and (7.14.9).
References for Lecture 7
There is a vast literature about points on rational surfaces. A brief list of works
related in one way or another to the point of view of this lecture follows:
[1] F. Chˆ atelet, Points rationnels sur certaines courbes et surfaces cubiques,
Enseignement Math. (2), 5 (1959), 153–170 (1960).
96 Lecture 7
[2] J.L. ColliotTh´ el` ene and J.J. Sansuc, Series of notes on rational varieties
and groups of multiplicative type, C. R. Acad. Sci. Paris Ser. AB, 282
(1976), A1113–A1116; 284 (1977), A967–A970; 284 (1977), A1215–
A1218; 287 (1978), A449–A452.
[3] J.L. ColliotTh´ el` ene and J.J. Sansuc, La R´ equivalence sur les tores,
Ann. Sci.
´
Ecole Norm. Sup. (4), 10 (1977), 175–229.
[4] J.L. ColliotTh´ el` ene and D. Coray, L’´ equivalence rationnelle sur les
points ferm´ es des surfaces rationnelles ﬁbr´ ees en coniques, Compositio
Math., bf 39 (1979), 301–332.
[5] Yu. Manin, Cubic Forms, North Holland, Amsterdam (1974). [Second
edition, 1986.]
[6] Yu. Manin, Le groupe de Brauer–Grothendieck en g´ eom´ etrie diophanti
enne, pp. 401–411 in Actes du Congr` es International Math´ ematiciens
(Nice, 1970), vol. 1, GauthierVillars, Paris (1971).
For a more extensive list, the reader can see the bibliography in [5].
The following are referred to in the text for the arithmetic of symbols in K
theory, and for quadratic forms:
[7] H. Bass and J. Tate, The Milnor ring of a global ﬁeld, pp. 349–446 in
Algebraic KTheory II, Lecture Notes in Math., no. 342, Springer, Berlin
(1973).
[8] J. Milnor, Algebraic Ktheory and quadratic forms, Invent. Math., 9
(1970), 318–344.
[9] J. Milnor, Introduction to Algebraic KTheory, Annals of Mathematics
Studies, vol. 72, Princeton University Press, Princeton, N.J. (1971).
[10] T. Y. Lam, The Algebraic Theory of Quadratic Forms, W. A. Benjamin,
Reading, Mass. (1973). [Revised second printing, 1980. See also Intro
duction to Quadratic Forms over Fields, American Mathematical Society,
Providence, R.I., 2005.]
[11] A. Grothendieck, Le groupe de Brauer I, II, III, pp. 46–188 in Dix expos´ es
sur la cohomologie des sch´ emas, North Holland, Amsterdam (1968).
[12] T. Nakayama, Cohomology of class ﬁeld theory and tensor product mod
ules I, Ann. of Math. (2), 65 (1957), 255–267.
[13] J. P. Serre, Corps Locaux, second edition, Hermann, Paris (1968). [Trans
lation: Local Fields, Springer, New York, 1979.]
[14] H. Gillet, Applications of algebraic Ktheory to intersection theory, The
sis, Harvard (1978).
[15] C. Sherman, Kcohomology of regular schemes, Comm. Algebra, 7
(1979), 999–1027.
8
Relative cycles and zeta functions
We return now to the relative cycle maps described in Lecture 3. Changing
notation, we write
j :
_
relative algebraic
1cycles on C × P
1
× P
1
_
→ H
1
(C, C
∗
),
and
log ¦ j ¦ :
⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
algebraic 1cycles on
C × P
1
× P
1
meeting
C × # properly
⎫
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎭
→ H
1
(C, R)
for the cycle class maps. (Here C is a smooth complete curve over a ﬁeld
k → C.) When C = E is an elliptic curve we will show explicitly how to
write down some interesting relative cycles γ and we will compute log ¦ j ¦ (γ).
In Lecture 9 we consider the case when E/Q has complex multiplication by
the full ring of integers in an imaginary quadratic number ﬁeld. Identifying
H
1
(E, R) C, we show there is a natural relative cycle γ deﬁned over Q such
that
ρ log ¦ j ¦ (γ) = L(2, χ
Gr¨ oss
),
where L(s, χ
Gr¨ oss
) is the Artin–Hasse zeta function of E (given in this case by
a Hecke Lseries, whence the notation) and ρ is a constant built from a Gauss
sum, the conductor of E, and the imaginary part of τ. Since L(s, χ) has an Euler
product converging for Res > 3/2, it will follow that log ¦ j ¦ (γ) 0.
These computations mimic those in an unpublished preprint [2], where I
deﬁne a map from Ktheory
Γ(E, 7
2
) → H
1
(E(C), R) C.
It seems likely that Γ(E, 7
2
) can be identiﬁed with a suitable Chow group for
98 Lecture 8
E ×P
1
×P
1
relative to E ×# (cycles being deﬁned over Q) and that the rank of
this group is the order of vanishing of L(s, χ) at s = 0. Indeed, one may hope
for a similar interplay between relative cycles, Kgroups, and zeta functions
quite generally. I hope this exciting prospect will sustain the reader through
the complicated calculations which follow.
Recall we have an exact sequence
0 Γ(C, 7
2
) K
2
(k(C))
T
_
x∈C
k(x)
∗
.
Our ﬁrst objective will be to associate to an element in Γ(C, 7
2
) a relative
algebraic 1cycle on C×P
1
×P
1
, relative to C×#. This cycle will only be well
deﬁned modulo some equivalence which we shall not describe explicitly.
Consider then an element
¦ f
i
, g
i
¦ ∈ Γ(C, 7
2
) ⊂ K
2
(k(C)). To each pair f , g
of rational functions on C we associate the double graph
γ
f ,g
= ¦(x, f (x), g(x))¦ ⊂ C × P
1
× P
1
.
Let γ
1
=
_
γ
f
i
,g
i
. The name of the game will be to add or subtract “trivial”
curves to γ
1
in such a way that the resulting sum is a relative algebraic 1cycle.
A curve τ ⊂ C ×P
1
×P
1
is “trivial” for our purposes if either τ ⊂ ¦p¦ ×P
1
×P
1
for some p ∈ C, or τ ⊂ (C×P
1
×¦1¦) ∪(C×¦1¦ ×P
1
). One sees from Lecture 3
that log ¦ j ¦ (trivial curve) = 0.
Step 1 (Neutralizing the vertices) It may happen that Supp γ
1
contains a point
(p, i, j), i, j = 0, ∞. Suppose for example one of the components of γ
1
looks
near (p, 0, 0) like
(8.1.1) ¦(x, ax
r
+ hot, bx
s
+ hot)¦, hot = higher order terms,
where x is a local coordinate near p, x(p) = 0. A calculation like (3.2) shows
that this component can be made to look like a relative cycle near (p, 0, 0) by
subtracting the “trivial” curve
(8.1.2) τ =
__
p, a t
r
, b
_
t
1 − t
_
s
__
, t = standard parameter on P
1
.
Note that in addition to the intersection at (p, 0, 0), τ meets C × # with mul
tiplicity r at (p, ∞, (−1)
s
b) and with multiplicity s at (p, a, ∞). We iterate the
above procedure mutatis mutandis for every vertex point (p, i, j) on γ
1
.
Step 2 (Neutralizing other boundary points) Let γ
2
= γ
1
−
_
τ be the cycle
with vertex intersections neutralized. Recall we have on K
2
(k(C)) the tame
Relative cycles and zeta functions 99
symbol at p
T
p
: K
2
(k(C)) → k(p)
∗
,
T
p
¦ f , g¦ = (−1)
ord
p
f ord
p
g
f
ord
p
g
g
ord
p
f
(p).
The tame symbol can be interpreted geometrically as follows: let
W = C × G
m
× ¦0¦ − C × G
m
× ¦∞¦ − C × ¦0¦ × G
m
+ C × ¦∞¦ × G
m
(note we have removed the vertices). For w ∈ Supp W, let ρ(w) ∈ G
m
be either
the second or third coordinate, whichever 0, ∞. For x ∈ C × P
1
× P
1
let
ν(γ
2
, W; x) be the multiplicity of intersection of γ
2
and W at x.
Lemma (8.1.3) With notation as above,
1 = T
p
_
_
i
¦ f
i
, g
i
¦
_
=
_
w∈(¦p¦×P
1
×P
1
)∩Supp W
ρ(w)
ν(γ
2
,W;w)
.
Proof A cycle like (8.1.1) contributes (−1)
rs
a
s
b
−r
to the tame symbol. This
is precisely the contribution of the cycle −τ to the righthand side, where τ is
as in (8.1.2). The bookkeeping for nonvertex points is easier, and is left for
the reader.
Given a, b ∈ G
m
and p ∈ C, we consider the “trivial” cycles
A
a,b,p
=
__
p, t,
(t − a)(t − b)
(t − 1)(t − ab)
_
¸
¸
¸
¸
t ∈ P
1
_
,
B
a,b,p
=
__
p, t,
t − a
t − b
_ ¸
¸
¸
¸
t ∈ P
1
_
.
We have for
¯
W = closure of W, viewed as a cycle,
A
a,b,p
¯
W = (p, a, 0) + (p, b, 0) − (p, 0, 1) − (p, ab, ∞)
− (p, 1, ∞) + (p, ∞, 1),
B
a,b,p
¯
W = (p, a, 0) − (p, b, ∞) −
_
p, 0,
a
b
_
+ (p, ∞, 1).
Starting at these formulae and using (8.1.3), it is straightforward to construct a
cycle
γ
3
= γ
2
+
n
i j
A
a
i j
,b
i j
, p
i
+
m
rs
B
a
rs
,b
rs
, p
r
such that γ
3
(C × #) has support in
(C × ¦1¦ × ¦0, ∞¦) ∪ (C × ¦0, ∞¦ × ¦1¦).
100 Lecture 8
Step 3 (Finishing the job) We may now add or subtract cycles of the form
(p) × (1) × P
1
and (p) × P
1
× (1) to γ
3
getting a γ
4
such that γ
4
C × (i) × P
1
and γ
4
C × P
1
× (i) are eﬀective, i = 0, ∞, but no cycle γ
4
− (p) × P
1
× (1) or
γ
4
−(p) ×(1) ×P
1
has this property. Write pr
1
: C×P
1
×P
1
→C. We have that
pr
1∗
[γ
4
C × ((0) − (∞)) × P
1
],
pr
1∗
[γ
4
C × P
1
× ((0) − (∞))]
are zerocycles rationally equivalent to 0 on C. There exist, therefore, functions
h
1
, h
2
on C such that
γ = γ
5
= γ
4
+ ¦(p, h
1
(p), 1) ¦ p ∈ C¦ + ¦(p, 1, h
2
(p)) ¦ p ∈ C¦
is a relative cycle.
We now want to construct some interesting elements in Γ(C, 7
2
) (and hence
some interesting relative algebraic 1cycles) in the case C = E = elliptic curve.
Suppose E is deﬁned over a ﬁeld k. Let f , g be functions on E, N > 0 an integer,
and assume all zeros and poles of f and g are points of E of order N deﬁned
over k. Consider the diagram
k(E)
∗
⊗k
∗
div⊗1
symbol
_
x∈E
k
∗
Pic(E) ⊗k
∗
0
K
2
(k(E))
tame
_
x∈E
k(x)
∗
.
(8.2)
Clearly tame¦ f , g¦
N
∈
k
∗
and goes to 0 in Pic(E) ⊗k
∗
. There thus exist
functions f
i
∈ k(E)
∗
and constants c
i
∈ k
∗
such that
¦ f , g¦
N
_
¦ f
i
, c
i
¦ ∈ Γ(E, K
2
) .
Notation (8.2.1) Fix an integer N and assume all points on E of order N are
deﬁned over k. If a is such a point, let f
a
denote the function with zero of order
N at a, and pole of order N at 0. Let ρ be the function with poles of order 1
at every nonzero point of order N, and a zero of order N
2
− 1 at the origin.
Deﬁne S
a
∈ Γ(E, K
2
) by
S
a
≡ ¦ ρ, f
a
¦
N
mod Image(k(E)
∗
⊗k
∗
→ K
2
(k(E))) .
S
a
is actually well deﬁned modulo torsion and symbols with both entries con
stant, as one sees by looking at the kernel of div⊗1 in (8.2).
Relative cycles and zeta functions 101
Our objective is now to compute log ¦ j(S
a
)¦. The key step in this program
will be to compute log ¦ j(γ
f ,g
)¦ when γ
f ,g
= ¦(x, f (x), g(x)) ¦ x ∈ E¦ for any two
functions f , g on E. We assume for simplicity that the divisors of f and g have
disjoint support. (The general case follows from this one by a limit argument.)
Slit E along nonintersecting simple closed paths between zeros and poles of
f , removing also εdisks about these points:
S
f
f = 0
f = ∞.
T
f
S
f
We do likewise for the zeros and poles of g, taking care that the slits for f and
g do not intersect. (This is possible because removing a simple arc will not dis
connect the surface.) The resulting slit and punctured surface will be denoted
E
·
. The handles of the barbellshaped aﬀairs are called T
f
, T
g
or T
f ,ε
, T
g,ε
. The
εcircles are written S
f
, S
g
or S
f ,ε
, S
g,ε
.
It will be convenient to use the Jacobi parameterization π: C
∗
→ E
C
∗
/q
Z
, q = e
2πiτ
. We let t be the standard parameter on C
∗
and write t = e
2πiz
,
z = x + iy. Thus y =
−1
2π
log ¦ t ¦ is a function on C
∗
, but x is not.
We normalize the Weierstrass σfunction on E as in Lang [4]
φ(z) = e
−
1
2
ηz
2
+πiz
, σ(z) =
1
2πi
(t − 1)
∞
_
n=1
(1 − q
n
t)(1 − q
n
t
−1
)
(1 − q
n
)
2
.
The given functions f , g on E can be factored in terms of the φfunction
f (z) = c
f
_
φ(z − a
j
)
m
j
, g(z) = c
g
_
φ(z − b
k
)
n
k
, c
f
, c
g
∈ C
∗
,
_
m
j
=
_
n
k
=
_
m
j
a
j
=
_
n
k
b
k
= 0.
For any ﬁxed n we have
_
k
e
2πin
k
(z−a
k
−nτ)
= 1,
so if we write
˜
f =
_
j
(1 − α
j
t)
n
j
, α
j
= e
−2πia
j
,
we get
f (t) = c
f
_
n∈Z
˜
f (tq
n
).
102 Lecture 8
Finally, let C
∗·
denote C
∗
slit and punctured so as to lie over E
·
, π: C
∗·
→ E
·
.
Functions like log f and log g (resp. log f , log
˜
f , log g) can be deﬁned on E
·
(resp. C
∗·
). Also a family of paths
−−−−→
1f (p) or
−−−−→
1g(p) lying on G
m
can be chosen
continuously for p ∈ E
·
.
Deﬁne 3chains
∆ = ∆
ε
= ¦(p,
−−−−→
1f (p), g(p)) ¦ p ∈ E
·
¦,
Φ = Φ
ε
= ¦(p,
−−−−→
1f (p),
−−−−→
1g(p)) ¦ p ∈ T
f
¦.
Bearing in mind orientations, we get
∂∆ = ¦(p,
−−−−→
1f (p), g(p)) ¦ p ∈ S
g
∪ T
f
∪ S
f
¦ = γ
·
f ,g
+ deg,
∂Φ = ¦(p,
−−−−→
1f (p),
−−−−→
1g(p)) ¦ p ∈ ∂T
f
¦
+ ¦(p,
−−−−→
1f (p), g(p)) ¦ p ∈ T
f
¦ + deg,
(where deg = cycle supported on E × G
m
× ¦1¦ ∪ E × ¦1¦ × G
m
), whence
∂(Φ−∆) = γ
·
f ,g
+ chains supported on (S
g
∪S
f
) ×G
m
×G
m
+ degenerate chains.
Our duty becomes the calculation of
lim
ε→0
−1
4π
2
Im
_
Φ
ε
−∆
ε
dx ∧
dx
1
x
1
∧
dx
2
x
2
and
lim
ε→0
−1
4π
2
Im
_
Φ
ε
−∆
ε
dy ∧
dx
1
x
1
∧
dx
2
x
2
.
Lemma (8.3)
lim
ε→0
_
∆
ε
dy ∧
dx
1
x
1
∧
dx
2
x
2
= 2πi
_
y log(1 − t)
¸
¸
¸
¸ n
−
_
n
log(1 − t) dy
_
.
(Here n denotes the inﬁnite “cycle” on C
∗
n =
n∈Z
j,k
m
j
n
k
(α
k
β
−1
j
q
n
), α
k
= e
−2πia
k
, β
j
= e
−2πib
j
,
and
_
n
denotes the integral along paths from points of negative multiplicity to
points of positive multiplicity. A diﬀerent choice of paths does not change the
righthand side.)
Proof of lemma The ﬁrst claim is that
lim
ε→0
_
E
·
ε
dy ∧
dg
g
= 0.
Relative cycles and zeta functions 103
To see this, think of g: E → P
1
. The above integral becomes
lim
ε→0
_
−
i
2
_
_
P
1
ε
g
∗
(dz − d¯ z)
du
u
,
where P
1
ε
denotes P
1
with εballs about 0 and ∞ removed, and u is the stan
dard parameter on P
1
. Note that g
∗
(dz) (resp. g
∗
(d¯ z)) are global holomorphic
(resp. antiholomorphic) diﬀerentials on P
1
, and hence are zero, so the integral
vanishes.
Now write
_
∆
ε
dy ∧
dx
1
x
1
∧
dx
2
x
2
= −
_
E
·
ε
log f dy ∧
dx
2
x
2
.
Note that the diﬀerential can be written
log f dy ∧
dx
2
x
2
= c
f
dy ∧
dx
2
x
2
+ “trace”
C
∗
/E
log
˜
f dy ∧
dx
2
x
2
,
where “trace” means the inﬁnite sum over all deck transformations. The inte
gral over c
f
dy ∧
dx
2
x
2
vanishes by the above, so
_
∆
ε
dy ∧
dx
1
x
1
∧
dx
2
x
2
= −
_
C
∗·
log
˜
f dy ∧
dg
g
= −
_
C
∗·
d(y log
˜
f ) ∧
dg
g
=
_
T
f
∪T
g
∪S
f
∪S
g
log g d(y log
˜
f ).
Winding around a zero of g changes log g by 2πi, whence
_
T
g
log g d(y log
˜
f ) = −2πi
_
(g)
d(y log
˜
f ) = −2πi y log
˜
f
¸
¸
¸
¸
(g)
,
where
_
(g)
means integrals taken over paths from poles to zeros of g. Similar
arguments show
lim
ε→0
_
T
f
log g d(y log
˜
f ) = −2πi
_
(
˜
f )
log g dy.
Standard estimates imply
lim
ε→0
_
S
g
log g d(y log
˜
f ) = 0,
and an easy argument (replacing y by (y − y
0
) + y
0
where y
0
is the value of y at
the center of a circle S ⊂ S
f
) gives
lim
ε→0
_
S
f
log g d(y log
˜
f ) = lim
ε→0
_
S
f
y log g
d
˜
f
f
= 2πi y log g
¸
¸
¸
¸
(
˜
f )
.
104 Lecture 8
Combining these calculations
lim
ε→0
_
∆
ε
dy ∧
dx
1
x
1
∧
dx
2
x
2
= 2πi
_
y log g
¸
¸
¸
¸
(
˜
f )
− y log f
¸
¸
¸
¸
(g)
−
_
(
˜
f )
log g dy
_
.
We have (g) =
_
n∈Z,k
n
k
(β
−1
k
q
n
) and (
˜
f ) =
_
j
m
j
(α
−1
j
), so
lim
ε→0
_
∆
ε
dy ∧
dx
1
x
1
∧
dx
2
x
2
= 2πi
_
j,k
n∈Z
−m
j
n
k
2π
log ¦α
−1
j
¦ log
_
1 − β
k
q
n
α
−1
j
_
+
j,k
n∈Z
m
j
n
k
2π
log ¦q
n
β
−1
k
¦ log
_
1 − α
j
β
−1
k
q
n
_
−
_
n
log(1 − t) dy
_
.
Since for ﬁxed n, k
j
m
j
log
_
− α
j
β
−1
k
q
n
_
= 0,
the above sum can be rewritten (replacing n by −n in the second sum)
lim
ε→0
_
∆
ε
dy ∧
dx
1
x
1
∧
dx
2
x
2
= +2πi
_
y log(1 − t)
¸
¸
¸
¸
n
−
_
n
log(1 − t) dy
_
.
This completes the proof of Lemma (8.3).
Lemma (8.4)
lim
ε→0
_
∆
ε
dx ∧
dx
1
x
1
∧
dx
2
x
2
= −i lim
ε→0
_
∆
ε
dy ∧
dx
1
x
1
∧
dx
2
x
2
.
Proof
_
∆
ε
dz ∧
dx
1
x
1
∧
dx
2
x
2
= −
_
E
·
log f dz ∧
dg
g
= 0 by reason of type.
Lemma (8.5) Let du = either dx or dy. Then
lim
ε→0
Im
_
Φ
ε
du ∧
dx
1
x
1
∧
dx
2
x
2
= −2π Re
_
n
log(1 − t) du.
Proof We have
lim
ε→0
Im
_
Φ
ε
du ∧
dx
1
x
1
∧
dx
2
x
2
= lim
ε→0
Im
_
T
f
log f log g du = −2π Re
_
( f )
log g du.
The reader can easily reinterpret this integral as the desired integral
_
n
on C
∗
.
Relative cycles and zeta functions 105
Theorem (8.6) The periods associated to the cycle
γ = γ
f ,g
= ¦(x, f (x), g(x))¦ ⊂ E × P
1
× P
1
are
ImP
γ
(dx) =
−1
4π
2
_
log ¦ t ¦ arg(1 − t)
¸
¸
¸
¸
n
− Im
_
n
log(1 − t)
dt
t
_
,
ImP
γ
(dy) =
−1
4π
2
log ¦ t ¦ log ¦1 − t ¦
¸
¸
¸
¸
n
,
where P
γ
and ImP
γ
are as deﬁned in Lecture 3.
Proof Simply use
dt
t
= 2πi dz and y =
−log ¦ t ¦
2π
, together with the equations
resulting from Lemmas (8.3)–(8.5):
lim
ε→0
Im
_
Φ
ε
−∆
ε
dy ∧
dx
1
x
1
∧
dx
2
x
2
= −2πy log ¦1 − t ¦
¸
¸
¸
¸
n
,
lim
ε→0
Im
_
Φ
ε
−∆
ε
dx ∧
dx
1
x
1
∧
dx
2
x
2
= Im
_
i
_
2πi y log(1 − t)
¸
¸
¸
¸
n
− 2πi
_
n
log(1 − t) dy
__
,
−2π Re
_
n
log(1 − t) dx = −2πy arg(1 − t)
¸
¸
¸
¸
n
− 2π Re
_
n
log(1 − t) dz.
In order to make things neater, it is convenient to introduce two functions:
D(t) = log ¦ t ¦ arg(1 − t) − Im
_
t
0
log(1 − u)
du
u
,
J(t) = log ¦ t ¦ log ¦1 − t ¦.
We ﬁrst descend these functions by a process of summation to functions on E.
Lemma (8.7) The functions
D
q
(t) =
n∈Z
D(q
n
t),
G
q
(t) =
∞
n=0
J(q
n
t) −
∞
n=1
J(q
n
t
−1
)
+
(log ¦ t ¦)
3
3 log ¦q¦
−
(log ¦ t ¦)
2
2
+
log ¦ t ¦ log ¦q¦
6
are continuous on C
∗
and invariant under t → qt, and hence descend to func
tions on E.
106 Lecture 8
Proof We refer the reader to Bloch [1] for a proof that the function D(t) is
well deﬁned. (At ﬁrst glance it would appear to depend on a choice of path
from 0 to t.) Granting this, one easily checks that
_
∞
n=0
D(tq
n
) converges. We
have
log(1 − t)
dt
t
+ log
_
1 −
1
t
_
d (1/t)
1/t
= log(−t)
dt
t
,
Im
_
t
log(−u)
du
u
= Im
_
1
2
(log(−x))
2
_
+ C + C
·
log ¦x¦,
where C
·
accounts for the ambiguity in the branch of log(−t). Thus
D(t) + D(t
−1
) = −
1
2
Im(log((−t))
2
) + arg(−t) log ¦ t ¦ + C + C
··
log ¦ t ¦
= C + C
··
log ¦ t ¦
with C
··
accounting for the ambiguity in the branches of log(−t) and arg(1−t).
For t < 0, however, D(t) = 0, so for all t we ﬁnd
(8.7.1) D(t) = −D(t
−1
).
From this, one deduces convergence of
_
n<0
D(tq
n
).
The corresponding continuity and invariance properties of G
q
are straight
forward, and are left to the reader.
Lemma (8.8) With notation as above,
ImP
γ
(dx) =
−1
4π
2
D
q
(A), ImP
γ
(dy) =
−1
4π
2
G
q
(A),
where, writing ( f ) =
_
m
j
(a
j
) and (g) =
_
n
k
(b
k
) with a
j
, b
k
∈ E (note
change of notation), A =
_
m
j
n
k
(b
j
− a
k
). (We view D
q
and G
q
as functions
on divisors in the obvious way, for example D
q
(
_
r
i
(p
i
)) =
_
r
i
D
q
(p
i
).)
Proof The assertion for ImP
γ
(dx) is immediate from Theorem (8.6). For
ImP
γ
(dy), note that if
_
n
k
(α
k
) and
_
m
j
(β
j
) are divisors on C
∗
with
_
m
j
=
_
n
k
= 0,
α
k
n
k
=
β
j
m
j
= 1, then the function (log ¦ t ¦)
r
vanishes on the
divisor
_
m
j
n
k
(α
k
β
−1
j
) for r ≤ 3. The reader can now easily check G
q
(A) =
log ¦ t ¦ log ¦1 − t ¦
¸
¸
¸
¸
n
as claimed.
We consider now the complexvalued function
Q:
_
1cycles on
E × P
1
× P
1
_
→ C,
(8.8.1)
Q(γ) = ImP
γ
(dx) − i ImP
γ
(dy) .
Relative cycles and zeta functions 107
Notice that Qdepends on the choice of a holomorphic diﬀerential dz = dx+i dy
on E. We have
Q(γ
f ,g
) =
−1
4π
2
_
D
q
((g) ∗ (g)
−
) − i G
q
((g) ∗ ( f )
−
)
_
,
where with ( f ) =
_
m
j
(a
j
) and (g) =
_
n
k
(b
k
),
g ∗ ( f )
−
:=
m
j
n
k
(b
k
− a
j
).
Recall that given a point of order N on E, which we write as
k+τ
N
, we have
deﬁned a relative algebraic 1cycle associated to the element S k+τ
N
∈ Γ(E, K
2
)
S k+τ
N
=
_
p, f k+τ
N
_
N
_
¦ f
i
, c
i
¦, c
i
∈ C
∗
.
Let this cycle be denoted γk+τ
N
. As noted earlier it is not completely well de
ﬁned. However, Q(γk+τ
N
) is deﬁned. Our main result can be given a purely
analytic formulation as follows: ﬁx E = C/[1, τ] with y = Imτ > 0. Let
f : Z/NZ × Z/NZ → C
be an odd function, that is f (a, b) = −f (−a, −b), and write
ˆ
f (k, ) =
1
N
N−1
a,b=0
f (a, b) e
2πi
_
−ak+b
N
_
.
Theorem (8.9) With notation as above,
N−1
k,=0
ˆ
f (k, )Q(γk+τ
N
) =
−y
2
N
5
4π
3
m,n∈Z
(m,n)(0,0)
f (m, n)
(mτ + n)
2
(m¯ τ + n)
.
Proof As in the ﬁrst part of the proof of (8.3), one shows that Q(γ
f ,c
) = 0 for
c ∈ C
∗
, whence
(8.9.1) Q(γk+τ
N
) =
−N
4π
2
_
D
q
((ρ)
−
∗
( f k+τ
N
)) − i G
q
((ρ)
−
∗
( f k+τ
N
))
_
.
Writing Q
q
=
−1
4π
2
(D
q
− i G
q
), viewed as a function on E, one has
Q(γk+τ
N
) = N
2
Na=0
a0
_
Q
q
(a) − Q
q
_
a +
k + τ
N
_
_
+ N
2
(N
2
− 1)
_
Q
q
_
k + τ
N
_
− Q
q
(0)
_
.
Since Q
q
(0) = 0, this yields
Q(γk+τ
N
) = N
4
Q
q
_
k + τ
N
_
.
108 Lecture 8
Lemma (8.10)
N−1
k,=0
ˆ
f (k, )
_
3
3N
3
−
2
2N
2
+
6N
_
=
i N
4π
3
n∈Z
n0
f (0, n)
n
3
.
Proof We use the wellknown Fourier series
n∈Z
n0
e
2πi nx
n
3
= 4π
3
i
_
x
3
3
−
x
2
2
+
x
6
_
to write
k,
ˆ
f (k, )
_
3
3N
3
−
2
2N
2
+
6N
_
=
−i
4π
3
k,
ˆ
f (k, )
n∈Z
n0
e
2πin/N
n
3
=
−i
4π
3
N
a,b,k,
f (a, b) e
2πi
_
−ak+b
N
_
n0
e
2πin/N
n
3
=
−i N
4π
3
n0
f (0, −n)
n
3
=
i N
4π
3
f (0, n)
n
3
.
It is convenient now to split up the function Q
q
. For this, we view Q
q
, D
q
, G
q
as functions on C
∗
, and write
Q
q
(t) =
−1
4π
2
(D
q
(t) − i G
q
(t))
=
i
4π
2
_
∞
n=0
log ¦tq
n
¦ log(1 − tq
n
) −
∞
n=1
log ¦t
−1
q
n
¦ log(1 − t
−1
q
n
)
_
+
1
4π
2
_
∞
n=0
Im
_
xq
n
x
−1
q
n
log(1 − t)
dt
t
_
+
i
4π
2
_
(log ¦ t ¦)
3
3 log ¦q¦
−
(log ¦ t ¦)
2
2
+
log ¦ t ¦ log ¦q¦
6
_
.
(In what follows ¦xq
n
¦ < 1 for n ≥ 0 and ¦x
−1
q
n
¦ < 1 for n ≥ 1, so there
will be no ambiguity about branches of functions.) Write the three terms in
parentheses as R
·
q
(t), R
··
q
(t), and R
···
q
(t) respectively, so
Q
q
(t) =
i
4π
2
R
·
q
(t) +
1
4π
2
R
··
q
(t) +
i
4π
2
R
···
q
(t).
Relative cycles and zeta functions 109
We have already seen
i
4π
2
N−1
k,=0
ˆ
f (k, ) R
···
q
_
e
2πi
_
k+τ
N
_
_
=
i
4π
2
N−1
k,=0
ˆ
f (k, )
_
4π
2
y
2
3
3N
3
−
4π
2
y
2
2
2N
2
+
4π
2
y
2
6N
_
= i y
2
k,
ˆ
f (k, )
_
3
3N
3
−
2
2N
2
+
6N
_
=
−Ny
2
4π
3
n∈Z
n0
f (0, n)
n
3
.
The formula in (8.9) to be proven now reads
(8.11)
N−1
k,=0
ˆ
f (k, )
_
R
··
q
_
e
2πi
_
k+τ
N
_
_
+ i R
·
_
e
2πi
_
k+τ
N
_
__
=
−y
2
N
π
m,n∈Z
m0
f (m, n)
(mτ + n)
2
(m¯ τ + n)
.
Put
L =
1
N
N−1
k,=0
ˆ
f (k, ) R
·
q
_
e
2πi
_
k+τ
N
_
_
, (8.12)
M =
−i
N
ˆ
f (k, ) R
··
q
_
e
2πi
_
k+τ
N
_
_
.
Proposition (8.13) We have
L =
−y
2π
m∈Z
m0
n∈Z
f (m, n)
m(mτ + n)
2
.
Proof
L =
−2πy
N
N−1
k,=0
ˆ
f (k, )
_
∞
n=0
_
n +
N
_
log
_
1 − e
2πi
_
k+τ
N
+nτ
_
_
−
∞
n−1
_
n −
N
_
log
_
1 − e
2πi
_
−k−τ
N
+nτ
_
__
.
To consolidate the two series, note that (replacing k, by N − k, N − , and n
110 Lecture 8
by n + 1),
−
N−1
k,=0
ˆ
f (k, )
∞
n=1
_
n −
N
_
log
_
1 − e
2πi
_
−k−τ
N
+nτ
_
_
=
N−1
k=0
N
=1
ˆ
f (k, )
∞
n=0
_
n +
N
_
log
_
1 − e
2πi
_
k+τ
N
+nτ
_
_
=
N−1
k,=0
ˆ
f (k, )
∞
n=0
_
n +
N
_
log
_
1 − e
2πi
_
k+τ
N
+nτ
_
_
+
N−1
k=0
ˆ
f (k, N)
∞
n=1
n log
_
1 − e
2πi
_
k
N
+nτ
_
_
−
N−1
k=0
ˆ
f (k, 0)
∞
n=0
n log
_
1 − e
2πi
_
k
N
+nτ
_
_
=
N−1
k,=0
ˆ
f (k, )
∞
n=0
_
n +
N
_
log
_
1 − e
2πi
_
k+τ
N
+nτ
_
_
.
Thus
L =
−4πy
N
N−1
k,=0
ˆ
f (k, )
∞
n=0
_
n +
N
_
log
_
1 − e
2πi
_
k+τ
N
+nτ
_
_
=
−4πy
N
2
k,,a,b
f (a, b) e
2πi
_
−ak+b
N
_
∞
n=0
∞
m=1
_
n +
N
_
e
2πim
_
k+τ
N
+nτ
_
m
.
This can be rewritten
(8.14) L =
4πy
N
2
a,b
f (a, b)
∞
n=0
m=1
m≡a(mod N)
ne
2πi
_
mτ+b
N
_
m
.
Lemma (8.15) For Im x > 0,
∞
n=0
ne
2π i nx
=
−1
4 sin
2
πx
.
Proof of lemma
ne
2πi nx
=
1
2πi
d
dx
_
1
1 − e
2πi x
_
=
e
2πi x
(1 − e
2πi x
)
2
=
1
(e
πi x
− e
−πi x
)
2
=
−1
4 sin
2
π x
.
Relative cycles and zeta functions 111
Returning to the expression for L (8.14), we see
L =
−πy
N
2
a,b
f (a, b)
∞
m=1
m≡a(mod N)
1
msin
2
π(
mτ+b
N
)
.
Now use
π
2
sin
2
πx
=
n∈Z
1
(x + n)
2
to write
L =
−y
πN
2
a,b
f (a, b)
∞
m=1
m≡a(mod N)
n∈Z
1
m(
mτ+b
N
+ n)
2
=
−y
2π
∞
m,n=−∞
m0
f (m, n)
m(mτ + n)
2
.
This proves (8.13).
Proposition (8.16) Let M be as in (8.12). Then
M =
−1
2π
∞
m,n=−∞
m0
f (m, n) Im
_
1
m
2
(mτ + n)
_
.
Proof For ¦ x ¦ < 1, the “natural” branch of
_
x
0
log(1 − t)
dt
t
is −
_
∞
m=1
x
m
m
2
. Thus
M =
i
N
N−1
k,=0
ˆ
f (k, ) Im
_
∞
n=0
m=1
e
2πi
_
k+τ
N
+nτ
_
m
m
2
−
∞
n=1
m=1
e
2πi
_
−k−τ
N
+nτ
_
m
m
2
_
.
As with L, it will be convenient to combine the two sums:
−
N−1
k,=0
ˆ
f (k, )
∞
m=n=1
e
2πi m
_
k+τ
N
+nτ
_
m
2
=
N−1
k=0
N
=1
ˆ
f (k, )
∞
m=1
n=0
e
2πi m
_
k+τ
N
+nτ
_
m
2
=
N−1
k,=0
ˆ
f (k, )
∞
m=1
n=0
e
2πi m
_
k+τ
N
+nτ
_
m
2
−
N−1
k=0
ˆ
f (k, 0)
∞
m=1
e
2πi
mk
N
m
2
,
112 Lecture 8
so
M =
2i
N
N−1
k,=0
ˆ
f (k, ) Im
_
∞
n=0
m=1
e
2πi
_
k+τ
N
+nτ
_
m
m
2
_
−
i
N
N−1
k=0
ˆ
f (k, 0) Im
_
∞
m=1
e
2πi
mk
N
m
2
_
= M
1
+ M
2
.
Lemma (8.17)
M
1
=
1
N
N−1
b=0
∞
m=1
f (m, b)
m
2
−
1
2π
∞
m,n=−∞
m0
f (m, n) Im
_
1
m
2
(mτ + n)
_
.
Proof
M
1
=
2i
N
2
a,b,k,
f (a, b) e
2πi
_
−ak+b
N
_
Im
_
∞
n=0
m=1
e
2πi
_
k+τ
N
+nτ
_
m
m
2
_
.
Note that
_
a,b
f (a, b) Re(e
2πi (
−ak+b
N
)
) = 0 because f (a, b) = −f (−a, −b), and
also
Im
_
ie
2πi (
−ak+b
N
)
(∗)
_
= −Im
_
e
2πi
_
−ak+b
N
_
_
Im(∗) + Re
_
e
2πi
_
−ak+b
N
_
_
Re(∗).
It follows that
M
1
=
2
N
2
a,b,k,
f (a, b) Im
_
i
∞
n=0
m=1
e
2πi
_
(m−a)k
N
+(nN+)
mτ+b
N
_
m
2
_
=
2
N
a,b
f (a, b) Im
_
i
∞
m=1
m≡a(mod N)
∞
n=0
e
2πin
_
mτ+b
N
_
m
2
_
.
Using the identity
1
1 − e
2πix
=
i cot πx + 1
2
,
we can write
M
1
=
1
N
a,b
f (a, b) Im
_
m=1
m≡a(mod N)
−cot π(
mτ+b
N
) + i
m
2
_
=
1
N
∞
m=1
b
f (m, b)
m
2
−
1
N
a,b
f (a, b) Im
_
m=1
m≡a
1
m
2
cot
_
π
_
mτ + b
N
___
.
Relative cycles and zeta functions 113
Note that
cot πx =
1
π
lim
N→∞
N
n=−N
1
x + n
,
and
Im(cot πx) =
1
π
n∈Z
Im
_
1
x + n
_
(the series converging without any special sort of summation), so we get
M
1
=
1
N
∞
m=1
b
f (m, b)
m
2
−
1
πN
a,b
f (a, b)
∞
m=1
m≡a
n∈Z
Im
_
1
m
2
_
mτ+b
N
+ n
_
_
=
1
N
∞
m=1
b
f (m, b)
m
2
−
1
2π
m∈Z
m0
n∈Z
f (m, n) Im
_
1
m
2
(mτ + n)
_
.
This completes the computation of M
1
.
Lemma (8.18) M
2
=
−1
N
b=N−1
m=∞
m=1
b=0
f (m, b)
m
2
.
Proof
M
2
=
−i
N
N−1
k=0
ˆ
f (k, 0) Im
_
∞
m=1
e
2πi
mk
N
m
2
_
=
−i
N
2
k,a,b
f (a, b) e
−2πiak
N
Im
_
∞
m=1
e
2πi
mk
N
m
2
_
=
−1
N
2
k,a,b
f (a, b) Im
_
i
∞
m=1
e
2πi
k
N
(m−a)
m
2
_
=
−1
N
f (m, b)
m
2
.
We return now to the proof of (8.16).
M = M
1
+ M
2
=
−1
2π
m0
n∈Z
f (m, n) Im
_
1
m
2
(mτ + n)
_
as desired.
114 Lecture 8
We can now complete the proof of Theorem (8.9):
N−1
k,=0
ˆ
f (k, )
_
R
··
q
_
e
2πi
_
k+τ
N
_
_
+ i R
·
q
_
e
2πi
_
k+τ
N
_
__
= i N(L + M)
=
−i yN
2π
_
m∈Z
m0
n∈Z
f (m, n)
_
1
m(mτ + n)
2
+
1
y
Im
1
m
2
(mτ + n)
__
.
But
1
m(mτ + n)
2
+
1
y
Im
_
1
m
2
(mτ + n)
_
=
1
m(mτ + n)
2
+
1
2iy
_
1
m
2
(mτ + n)
−
1
m
2
(m¯ τ + n)
_
=
1
m(mτ + n)
2
−
1
m(mτ + n)(m¯ τ + n)
= −2i y
1
(mτ + n)
2
(m¯ τ + n)
,
so
N−1
k,=0
ˆ
f (k, )
_
R
··
q
(e
2πi
k+τ
N
) + iR
·
(e
2πi
k+τ
N
)
_
=
−y
2
N
π
m,n∈Z
m0
f (m, n)
(mτ + n)
2
(m¯ τ + n)
.
This completes the proof of (8.11) and also (8.9).
References for Lecture 8
[1] S. Bloch, Applications of the dilogarithm function in algebraic Ktheory
and algebraic geometry, pp. 103–114 in Proceedings of the International
Symposium on Algebraic Geometry (Kyoto Univ., Kyoto, 1977), Kinoku
niyo Book Store, Tokyo (1978).
[2] S. Bloch, Higher Regulators, Algebraic KTheory, and Zeta Functions
of Elliptic Curves, Lectures given at the University of California, Irvine
(1978). [CRM Monograph Series, no. 11, American Mathematical Soci
ety, Providence, R.I., 2000.]
[3] A. Borel, Cohomologie de SL
n
et valeurs de fonctions zeta aux points
entiers, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4), 4 (1977), 613–636;
errata, 7 (1980), 373.
[4] S. Lang, Elliptic Functions, AddisonWesley, Reading, Mass.
(1973). [Second edition: Springer, New York, 1987.]
9
Relative cycles and zeta functions – continued
In this ﬁnal lecture we will show how Theorem (8.9) leads to a regulator for
mula for the value at s = 2 of the zeta function of an elliptic curve E with
complex multiplication by the ring of integers in an imaginary quadratic ﬁeld
k with class number 1. We begin with some general lemmas, valid without
hypotheses on E. Notation will be as in Lecture 8.
Lemma (9.1) Q(γa+bτ
N
) = −Q(γ−a−bτ
N
).
Proof Using formulas (8.7.1) and (8.9.1), together with the fact that the divi
sor (ρ) is invariant under z → −z, we reduce to proving G
q
(t
−1
) = −G
q
(t). We
have
G
q
(t
−1
) + G
q
(t) =
∞
n=0
J(q
n
t
−1
) −
∞
n=1
J(q
n
t)
+
∞
n=0
J(q
n
t) −
∞
n=1
J(q
n
t
−1
) − (log ¦ t ¦)
2
= J(t
−1
) + J(t) − (log ¦ t ¦)
2
= 0.
Lemma (9.2) Q(γa+b
N
) =
i y
2
N
4
4π
3
(m,n)(0,0)
sin(2π(
an−bm
N
))
(m + nτ)
2
(m + n¯ τ)
.
116 Lecture 9
Proof Let g
a+bτ
(k + τ) =
1
N
e
2πi (
a−bk
N
)
. Then
ˆ g
a+bτ
(m + nτ) =
1
N
2
k,
e
2πi (
a−bk
N
)
e
2πi (
−kn+m
N
)
=
1
N
2
k,
e
2πi (
(a+m)−k(b+n)
N
)
=
_
0 a −m or b −n
1 a = −m or b = −n
= δ
−a−bτ
(m + nτ)
with obvious notation. Writing f
a+bτ
= g
−a−bτ
− g
a+bτ
we get
Q(γa+bτ
N
) =
1
2
ˆ
f
a+bτ
(k + τ)Q(γk+τ
N
) =
−y
2
N
5
8π
3
f
a+bτ
(m + nτ)
(m + nτ)
2
(m + n¯ τ)
=
i y
2
N
4
4π
3
sin
_
2π
(an−bm)
N
_
(m + nτ)
2
(m + n¯ τ)
.
The following description of the “modular behavior” of R
q
seemed worth
including although it will not be used:
Lemma (9.3) Let τ
·
=
ατ+β
γτ+δ
with
_
α γ
β δ
_
∈ SL
2
(Z). Write q
·
= e
2πiτ
·
and let
_
b
a
_
=
_
α γ
β δ
__
b
·
a
·
_
. Then
Q
q
· (γa
·
+b
·
τ
·
N
) = (γ¯ τ + δ)
−1
Q
q
(γa+bτ
N
).
(We write Q
q
and Q
q
· to indicate dependence on q and q
·
.)
Proof Since SL
2
(Z) preserves the symplectic form
_
0 1
−1 0
_
, we ﬁnd that
−a
·
·
+ b
·
k
·
= −a + bk, where
_
k
_
=
_
α γ
β δ
__
·
k
·
_
. Thus
Q
q
· (γa
·
+b
·
τ
·
N
) =
i y
·2
N
4
4π
3
k
·
,
·
sin
_
2π
a
·
·
−b
·
k
·
N
_
(k
·
+
·
τ
·
)
2
(k
·
+
·
¯ τ
·
)
=
i y
·2
N
4
4π
3
k
·
,
·
sin
_
2π
a−bk
N
_
(k
·
+
·
τ
·
)
2
(k
·
+
·
¯ τ
·
)
.
Note that
k
·
+
·
τ
·
= (γτ + δ)
−1
(k
·
δ +
·
β + (k
·
γ +
·
α)τ)
= (γτ + δ)
−1
(k + τ),
Relative cycles and zeta functions – continued 117
whence (using y
·
=
y
(γτ+δ)(γ¯ τ+δ)
)
Q
q
· (γa
·
+b
·
τ
·
N
) =
i y
2
N
4
(γτ + δ)
2
(γ¯ τ + δ)
4π
3
(γτ + δ)
2
(γ¯ τ + δ)
2
k,
sin
_
2π
a−bk
N
_
(k + τ)
2
(k + ¯ τ)
= (γ¯ τ + δ)
−1
Q
q
(γa+bτ
N
).
It is convenient to denote by ( ) the bilinear form
( ) : o/No × o/No → C
∗
,
(a + bτ, k + τ) = e
2πi(
−a+bk
N
)
.
We assume henceforth that o = Z + Zτ is an order in an imaginary quadratic
ﬁeld κ.
Lemma (9.4) (xy, z) = (x, ¯ yz), where ¯ y is the complex conjugate.
Proof Write x = x
1
+x
2
τ, y = y
1
+y
2
τ, z = z
1
+z
2
τ, and suppose τ
2
+Aτ+B = 0,
with x
i
, y
i
, z
i
, A, B ∈ Z. Then
xy = x
1
y
1
− x
2
y
2
B + (x
1
y
2
+ x
2
y
1
− x
2
y
2
A)τ,
(xy, z) = exp
_
2πi
x
2
y
2
Bz
2
− x
1
y
1
z
2
+ x
1
y
2
z
1
+ x
2
y
1
z
1
− x
2
y
2
Az
1
N
_
,
¯ yz = y
1
z
1
+ y
2
z
2
B + y
2
z
1
¯ τ + y
1
z
2
τ
= (y
1
z
1
+ y
2
z
2
B − y
2
z
1
A) + (y
1
z
2
− y
2
z
1
)τ,
(x, ¯ yz) = exp
_
2πi
−x
1
y
1
z
2
+ x
1
y
2
z
1
+ x
2
y
1
z
1
+ x
2
y
2
z
2
B − x
2
y
2
z
1
A
N
_
.
Corollary (9.5) If ζ ∈ o
κ
is a unit (i.e. a root of 1), then Q(γζ(a+bτ)
N
) =
ζ
−1
Q(γa+bτ
N
).
Proof
Q(γζ(a+bτ)
N
) =
i y
2
N
4
4π
3
Im(ζ(a + bτ), k + τ)
(k + τ)
2
(k + ¯ τ)
=
i y
2
N
4
4π
3
Im(a + bτ,
¯
ζ(k + τ))
(k + τ)
2
(k + ¯ τ)
=
i y
2
N
4
4π
3
Im(a + bτ, k
·
+
·
τ)
(k
·
+
·
τ)
2
(k
·
+
·
¯ τ)
= ζ
−1
Q(γa+bτ
N
).
Assume now for simplicity that o is the ring of integers in κ, and κ has class
number 1.
118 Lecture 9
Corollary (9.6) Suppose N = f g, with f , g ∈ o
κ
. Then
¯ g
µ∈o/go
Q(γa+bτ
N
+
f µ
N
) = Q(γa+bτ
f
) .
Proof We have
µ∈o/go
(µ,
¯
f (k + τ)) =
_
g¯ g ¯ g ¦ k + τ,
0 otherwise.
Thus
µ∈o/go
Q(γa+bτ
N
+
f µ
N
) =
k,
i y
2
N
4
4π
3
µ
(µ,
¯
f (k + τ))
Im(a + bτ, k + τ)
(k + τ)
2
(k + ¯ τ)
=
i g¯ gy
2
N
4
4π
3
r,s
Im(a + bτ, ¯ g(r + sτ))
(r + sτ)
2
(r + s¯ τ)¯ g
2
g
=
i y
2
N
4
4¯ gπ
3
r,s
Im(g(a + bτ), r + sτ)
(r + sτ)
2
(r + s¯ τ)
= (¯ g)
−1
Q(γg(a+bτ)
N
)
= (¯ g)
−1
Q(γa+bτ
f
).
Lemma (9.7) Let χ have conductor f ¦ N, n = f g. Then
(i) ˆ χ(x) = 0 unless N ¦
¯
f x.
(ii) For x ∈ (o/No)
∗
, ˆ χ(xy) = ¯ χ( ¯ x)χ(y).
(iii) Let f
1
¦ f , f f
1
, and suppose N ¦
¯
f
1
x. Then ˆ χ(x) = 0.
Proof (i) We have
ˆ χ(x) =
1
N
y∈o/No
χ(y)(x, y) =
1
N
y
·
mod f
χ(y
·
)
y
··
mod g
(x, y
·
)(x, f y
··
)
=
1
N
y
·
mod f
χ(y
·
)(x, y
·
)
y
··
mod g
(x
¯
f , y
··
)
=
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
0 ¯ g x,
g¯ g
N
_
y
·
mod f
χ(y
·
)(x, y
·
) ¯ g ¦ x,
where the y
·
(resp. y
··
) run through representatives in o for the congruence
classes modulo f (resp. g).
(ii) If x ∈ (o/No)
∗
,
ˆ χ(xy) =
1
N
z∈o/No
χ(z)(xy, z) =
1
N
¯ χ(¯ y)
z
χ(z¯ y)(x, ¯ yz) = ¯ χ(¯ y) ˆ χ(x).
Relative cycles and zeta functions – continued 119
(iii) Since conductor χ = f f
1
, we can ﬁnd y ≡ 1 mod
¯
f
1
, y a unit in o/No,
with χ(¯ y) 1. Since N ¦
¯
f
1
x
ˆ χ(x) = ˆ χ(xy) = ¯ χ(¯ y) ˆ χ(x).
Thus ˆ χ(x) = 0.
We now have the tools we need for the regulator formula. Assume as above
that o = Z+Zτ is the ring of integers in an imaginary quadratic ﬁeld κ of class
number 1. We ﬁx an embedding κ → C, an integer N, and a character χ of
(o/No)
∗
which restricts to the given embedding µ
κ
→ C
∗
on the roots of 1. Let
χ
Gr¨ oss
denote the Gr¨ ossencharakter
χ
Gr¨ oss
(p) =
¯
h χ(h) , p = (h), p N.
Let f generate the conductor ideal of χ and write N = f g. From (8.9), (9.5),
(9.6), and (9.7) we get
L(2, χ
Gr¨ oss
) =
4π
3
−y
2
N
5
w∈o/No
ˆ χ(w)Q(γ
w
N
) (9.8)
=
4π
3
−y
2
N
5
x∈(o/
¯
f o)
∗
ˆ χ(x¯ g)Q(γ
x
¯
f
−1 ) =
4π
3
ˆ χ(¯ g)
−y
2
N
5
x∈(o/
¯
f o)
∗
¯ χ( ¯ x)Q(γ
x
¯
f
−1 )
=
4π
3
ˆ χ(¯ g)g
−y
2
N
5
x∈(o/No)
∗
¯ χ( ¯ x)Q(γ
x
N
) =
4π
3
¦ µ
κ
¦ ˆ χ(¯ g)g
−y
2
N
5
x∈(o/No)
∗
/µ
κ
¯ χ( ¯ x)Q(γ
x
N
).
The jinvariant j(E) is known to be real and to generate the Hilbert class
ﬁeld of κ. Since κ has class number 1, j(E) ∈ Q so we can choose a model E
Q
deﬁned over Q. Deuring’s theory associates to E
Q
a Gr¨ ossencharakter χ
Gr¨ oss
of
κ with values in κ
∗
,
χ
Gr¨ oss
(p) = ¯ x χ(x), where (x) = p N.
The character χ takes values in µ
κ
and χ( ¯ x) = ¯ χ(x) (this follows from theorem
10, p. 140 of Lang [4], which implies χ
Gr¨ oss
(¯ p) = χ
Gr¨ oss
(¯ p)), so we can rewrite
(9.8)
(9.9) L(2, χ
Gr¨ oss
) =
4π
3
¦ µ
κ
¦ ˆ χ(¯ g)g
−y
2
N
5
Q
_
(o/No)
∗
/µ
κ
S
x ¯ χ(x)/N
_
.
The ray class group (o/No)
∗
/µ
κ
acts on points of order N by
x
_
y
N
_
=
x
−1
χ(x) y
N
120 Lecture 9
and conjugation acts on these points in the natural way. The cycle
(9.10) U :=
(o/No)
∗
/µ
κ
γ
x ¯ χ(x)/N
is invariant under both these actions, and hence is deﬁned over Q. We get
Theorem (9.11) With notations as above, let χ
Gr¨ oss
be the Gr¨ ossencharakter
associated to E
Q
. Let f be a generator for the conductor ideal and let f g =
N ∈ Z, g ∈ o. Then there exists a cycle U on E × P
1
× P
1
relative to E × # and
deﬁned over Q such that
L(2, χ
Gr¨ oss
) =
4π
3
¦ µ
κ
¦ ˆ χ(¯ g)g
−y
2
N
5
Q(U).
Remarks (9.12) (i) L(s, χ
Gr¨ oss
) is related to the zeta function of E
Q
, ζ
E
Q
(de
ﬁned up to a ﬁnite number of factors by the product over all p ∈ Z such that E
has nondegenerate reduction mod p of the zeta function of the corresponding
curve over F
p
), by
ζ
E
Q
(s) = ζ
Q
(s)ζ
Q
(s − 1)L(s, χ
Gr¨ oss
)
−1
.
(ii) Various renormalizations are possible and perhaps desirable to simplify
the righthand constant in (9.11). Dick Gross suggests that the “corrected” for
mula would give L
·
(0, χ
Gr¨ oss
) ∈ Q(U) Q. Since L
·
(0, χ
Gr¨ oss
) ∈ π
−2
L(2, χ
Gr¨ oss
)
Q, one might take
¯
Q = 2πQ (this amounts to modifying (8.8.1) by 2π). A
standard Gauss sum calculation (cf. for example Lang [4], p. 289) yields
¦ ˆ χ(¯ g)¦ =
_
f
¯
f
N
, whence ¦ ˆ χ(¯ g) ¯ g¦
_
f
¯
f g¯ g
N
= 1.
Also for τ = x + i y ∈ k, we have y
2
∈ Q, so
L
·
(0, χ
Gr¨ oss
) ∈ ζ
¯
Q(U) Q,
where ζ is a root of 1.
References for Lecture 9
[1] S. Bloch, Applications of the dilogarithm function in algebraic Ktheory
and algebraic geometry, pp. 103–114 in Proceedings of the International
Symposium on Algebraic Geometry (Kyoto Univ., Kyoto, 1977), Kinoku
niyo Book Store, Tokyo (1978).
Relative cycles and zeta functions – continued 121
[2] S. Bloch, Higher Regulators, Algebraic KTheory, and Zeta Functions
of Elliptic Curves, Lectures given at the University of California, Irvine
(1978). [CRM Monograph Series, no. 11, American Mathematical Soci
ety, Providence, R.I., 2000.]
[3] A. Borel, Cohomologie de SL
n
et valeurs de fonctions zeta aux points
entiers, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4), 4 (1977), 613–636;
errata, 7 (1980), 373.
[4] S. Lang, Elliptic Functions, AddisonWesley, Reading, Mass.
(1973). [Second edition: Springer, New York, 1987.]
Bibliography
Artin, M. and B. Mazur, Formal groups arising from algebraic varieties, Ann.
Sci.
´
Ecole Norm. Sup. (4), 10 (1977), 87–132.
Atiyah, M. and F. Hirzebruch, Analytic cycles on complex manifolds, Topol
ogy, 1 (1962) 25–45.
Bass, H. Algebraic KTheory, Benjamin, New York (1968).
Bass, H. and J. Tate, The Milnor ring of a global ﬁeld, in Algebraic KTheory
II, Lecture Notes in Math., no. 342, Springer, Berlin (1973).
Beauville, A. Vari´ et´ es de Prym et jacobiennes interm´ ediaires, Ann. Sci. Ecole
Norm. Sup. (4), 10 (1977), 304–391.
———. Surfaces alg´ ebriques complexes, Ast´ erisque, 59 (1978),
Bloch, S. K
2
and algebraic cycles, Ann. of Math. (2), 99 (1974), 349–379.
———. Torsion algebraic cycles, K
2
, and Brauer groups of function ﬁelds,
Bull. Amer. Math. Soc., 80 (1974), 941–945.
———. K
2
of Artinian Qalgebras with application to algebraic cycles, Comm.
Algebra, 3 (1975), 405–428.
———. An example in the theory of algebraic cycles, pp. 1–29 in Algebraic
KTheory, Lecture Notes in Math., no. 551, Springer, Berlin (1976).
———. Some elementary theorems about algebraic cycles on abelian vari
eties, Invent. Math., 37 (1976), 215–228.
———. Algebraic Ktheory and crystalline cohomology, Inst. Hautes
´
Etudes
Sci. Publ. Math., no. 47 (1977), 187–268 (1978).
———. Applications of the dilogarithm function in algebraic Ktheory and
algebraic geometry, pp. 103–114 in Proceedings of the International Sympo
sium on Algebraic Geometry (Kyoto Univ., Kyoto, 1977), Kinokuniya Book
Store, Tokyo (1978).
———. Higher Regulators, Algebraic KTheory, and Zeta Functions of El
liptic Curves, Lectures given at the University of California, Irvine (1978).
124 Bibliography
[CRM Monograph Series, no. 11, American Mathematical Society, Provi
dence, R.I., 2000.]
———. Some formulas pertaining to the Ktheory of commutative group
schemes, J. Algebra, 53 (1978), 304–326.
———. Torsion algebraic cycles and a theorem of Roitman, Compositio
Math., 39 (1979), 107–127.
Bloch, S., A. Kas, and D. Lieberman, Zero cycles on surfaces with P
g
= 0,
Compositio Math., 33 (1976), 135–145,
Bloch, S. and J. P. Murre, On the Chow groups of certain types of Fano three
folds, Compositio Math., 39 (1979), 47–105.
Bloch, S. and A. Ogus, Gersten’s conjecture and the homology of schemes,
Ann. Sci.
´
Ecole Norm. Sup. (4), 7 (1974), 181–201 (1975).
Borel, A. Cohomologie de SL
n
et valeurs de fonctions zeta aux points entiers,
Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4), 4 (1977), 613–636; errata, 7 (1980),
373.
Chˆ atelet, F. Points rationnels sur certaines courbes et surfaces cubiques, En
seignement Math. (2), 5 (1959), 153–170 (1960).
Chevalley, C. et al., Anneaux de Chow et applications, S´ eminaire C. Chevalley,
2
e
ann´ ee, S´ ecr. Math. Paris (1958).
Chow, W. L. On equivalence classes of cycles in an algebraic variety, Ann. of
Math. (2), 64, 450–479 (1956).
Clemens, C. H. and P. A. Griﬃths, The intermediate Jacobian of the cubic
threefold, Ann. of Math. (2), 95 (1972), 281–356.
ColliotTh´ el` ene, J.L. and J.J. Sansuc, Series of notes on rational varieties
and groups of multiplicative type, C. R. Acad. Sci. Paris Ser. AB, 282 (1976),
A1113–A1116; 284 (1977), A967–A970; 284 (1977), A1215–A1218; 287
(1978), A449–A452.
———, La R´ equivalence sur les tores, Ann. Sci.
´
Ecole Norm. Sup. (4), 10
(1977), 175–229.
ColliotTh´ el` ene, J.L. and D. Coray, L’´ equivalence rationnelle sur les
points ferm´ es des surfaces rationnelles ﬁbr´ ees en coniques, Compositio
Math., bf 39 (1979), 301–332.
Deligne, P. Th´ eorie de Hodge. I, pp. 425–430 in Actes du Congr` es Interna
tional des Math´ ematiciens (Nice, 1970), vol. 1, GauthierVillars, Paris (1971).
———. Th´ eorie de Hodge III, Inst. Hautes Etudes Sci. Publ. Math., no. 44
(1974), 5–77.
———. Poids dans la cohomologie des vari´ et´ es alg´ ebriques, pp. 79–85 in Pro
ceedings of the International Congress of Mathematicians (Vancouver, B.C.,
1974), vol. 1 (1975).
Bibliography 125
———. Cohomologie ´ etale (SGA 4
1
2
), Lecture Notes in Math., no. 569,
Springer, Berlin (1977).
Elman, R. and T. Y. Lam, On the quaternion symbol homomorphism
g
F
: k
2
(F) → B(F), in Algebraic KTheory II, Lecture Notes in Math.,
no. 342, Springer, Berlin (1973).
Fatemi, T. L’equivalence rationelle des z´ ero cycles sur les surfaces al ´ gebriques
complexes a cup product surjectif, These du 3
e
cycle, Universit´ e de Paris VII
(1979).
Fulton, W. Rational equivalence on singular varieties, Inst. Hautes
´
Etudes Sci.
Publ. Math., no. 45 (1975), 147–167.
Fulton, W. and R. MacPherson, Intersecting cycles on an algebraic variety,
Aarhus Universitet Preprint Series, no. 14 (1976). [Pp. 179–197 in Real and
complex singularities (Proc. Ninth Nordic Summer School/NAVF Sympos.
Math., Oslo, 1976), Sijthoﬀ and Noordhoﬀ, Alphen aan den Rijn (1977).]
Gersten, S. M. Some exact sequences in the higher Ktheory of rings, pp. 211–
243 in Algebraic KTheory I, Lecture Notes in Math., no. 341, Springer,
Berlin (1973).
Gillet, H. Applications of algebraic Ktheory to intersection theory, Thesis,
Harvard (1978).
Griﬃths, P. On the periods of certain rational integrals. I, II, Ann. of Math. (2),
90 (1969), 460–495; 90 (1969), 496–541.
Griﬃths, P. and J. Harris, Principles of Algebraic Geometry, Wiley, New York
(1978). [Reprinted 1994.]
Grothendieck, A. Le groupe de Brauer I, II, III, pp. 46–188 in Dix expos´ es sur
la cohomologie des sch´ emas, North Holland, Amsterdam (1968).
Grothendieck, A. et al., Th´ eorie des intersections et th´ eor` eme de Riemann–
Roch (SGA 6), Lecture Notes in Math., no. 225, Springer, Berlin (1971).
Hartshorne, R. On the de Rhamcohomology of algebraic varieties, Inst. Hautes
´
Etudes Sci. Publ. Math., no. 45 (1975), 5–99.
Inose, H. and M. Mizukami, Rational equivalence of 0cycles on some surfaces
of general type with p
g
= 0, Math. Ann., 244 (1979), no. 3, 205–217.
Kleiman, S. Algebraic cycles and the Weil conjectures, pp. 359–386 in Dix
expos´ es sur la cohomologie des sch´ emas, North Holland, Amsterdam (1968).
Lam, T. Y. The Algebraic Theory of Quadratic Forms, W. A. Benjamin, Read
ing, Mass. (1973). [Revised second printing, 1980. See also Introduction to
Quadratic Forms over Fields, American Mathematical Society, Providence,
R.I., 2005.]
Lang, S. Abelian Varieties, Interscience Publishers (1959). [Reprinted
Springer, 1983.]
126 Bibliography
———. Elliptic Functions, AddisonWesley, Reading, Mass. (1973). [Second
edition: Springer, New York, 1987.]
Lieberman, D. I. Higher Picard varieties, Amer. J. Math., 90 (1968), 1165–
1199.
Manin, Yu. Le groupe de Brauer–Grothendieck en g´ eom´ etrie diophantienne,
pp. 401–411 in Actes du Congr` es International Math´ ematiciens (Nice, 1970),
vol. 1, GauthierVillars, Paris (1971).
———. Cubic Forms, North Holland, Amsterdam (1974). [Second edition,
1986.]
Mattuck, A. Ruled surfaces and the Albanese mapping, Bull. Amer. Math. Soc.,
75 (1969), 776–779.
———. On the symmetric product of a rational surface, Proc. Amer. Math.
Soc., 21 (1969), 683–688.
Milnor, J. Algebraic Ktheory and quadratic forms, Invent. Math., 9
(1970), 318–344.
———. Introduction to Algebraic KTheory, Annals of Mathematics Studies,
vol. 72, Princeton University Press, Princeton, N.J. (1971).
Mumford, D. Rational equivalence of zerocycles on surfaces, J. Math. Kyoto
Univ., 9 (1968), 195–204,
Murre, J. P. Algebraic equivalence modulo rational equivalence on a cubic
threefold, Compositio Math., 25 (1972), 161–206.
Nakayama, T. Cohomology of class ﬁeld theory and tensor product modules I,
Ann. of Math. (2), 65 (1957), 255–267.
Quillen, D. Higher algebraic Ktheory. I, in Algebraic KTheory I, Lecture
Notes in Math., no. 341, Springer, Berlin (1973).
Quillen, D. and D. Grayson, Higher algebraic Ktheory. II, pp. 217–
240 in Algebraic KTheory, Lecture Notes in Math., no. 551,
Springer, Berlin (1976).
Roitman, A. A. Γequivalence of zerodimensional cycles (in Russian), Mat.
Sb. (N.S.), 86 (128) (1971), 557–570. [Translation: Math USSRSb., 15
(1971), 555–567.]
———. Rational equivalence of zerodimensional cycles (in Russian), Mat.
Sb. (N.S.), 89 (131) (1972), 569–585, 671. [Translation: Math. USSRSb., 18
(1974), 571–588.]
Rosenlicht, M. Generalized Jacobian varieties, Ann. of Math., 59 (1954), 505–
530.
Samuel, P. Rational equivalence of arbitrary cycles, Amer. J. Math., 78 (1956),
383–400.
Bibliography 127
Serre, J.P. Sur la topologie des vari´ et´ es algebriques en caract´ eristique p,
pp. 24–53 in Symposium Internacional de Topologia Algebraica, Universi
dad Nacional Aut´ onoma de Mexico and UNESCO, Mexico City (1958).
———. Groupes algebriques et corps de classes, Hermann, Paris (1959).
[Second edition, 1975; reprinted, 1984.]
———. Alg` ebre locale. Multiplicit´ es, Lecture Notes in Math., no. 11, Springer,
Berlin (1965).
———. Corps Locaux, second edition, Hermann, Paris (1968). [Translation:
Local Fields, Springer, New York, 1979.]
Sherman, C. Kcohomology of regular schemes, Comm. Algebra, 7
(1979), 999–1027.
Stienstra, J. Deformations of the second Chow group, Thesis,
Utrecht (1978).
———. The formal completion of the second Chow group: a Ktheoretic ap
proach, pp. 149–168 in Journ´ ee de G´ eom´ etrie Algebrique de Rennes (Rennes,
1978), Vol. II, Asterisque, 64 (1979).
Swan, R. Algebraic KTheory, Lecture Notes in Math., no. 76, Springer, Berlin
(1968).
Tate, J. Relations between K
2
and Galois cohomology, Invent. Math., 36
(1976), 257–274.
Tennison, B. R. On the quartic threefold, Proc. London Math. Soc. (3), 29
(1974), 714–734.
Tyurin, A. N. Five lectures on threedimensional varieties (in Russian), Us
pehi Mat. Nauk, 27 (1972), no. 5, (167) 3–50. [Translation: Russian Math.
Surveys, 27 (1972), no. 5, 1–53.]
Verdier, J.L. Dualit´ e dans la cohomologie des espaces localement compacts,
S´ eminaire Bourbaki, Vol. 9, expos´ e no. 300 (1965), 337–349. [Reprinted
Soci´ et´ e Math´ ematique de France, Paris, 1995.]
———. A duality theorem in the etale cohomology of
schemes, pp. 184–198 in Proceedings of a conference on local ﬁelds
(Driebergen, 1966), Springer, Berlin (1967).
Wallace, A. Homology Theory on Algebraic Varieties, Pergamon Press, New
York (1958).
Weil, A. Foundations of Algebraic Geometry, A.M.S. Colloquium Publica
tions, vol. 29, American Mathematical Society, Providence, R.I. (1946).
———. Courbes Alg´ ebriques et Vari´ et´ es Ab´ eliennes, Hermann, Paris (1971).
Index
A
1
homotopy, ix
A
r
(X), 4
Abel–Jacobi map, 5, 27
Abel–Jacobi theorem, 3
additive group
formal, 74
admissible monomorphism or epimorphism,
46
Alb(X), Albanese variety of X, 9
Albanese kernel, 11, 59
Albanese variety, 5, 9
algebraic cycle, 3
algebraically equivalent to zero, 5
eﬀective, 3
rationally equivalent to zero, 5
algebraic surface, 9
applications, xx
Artin–Hasse exponential, 73
Artin–Hasse zeta function, 97
artinian kalgebra, 71
bad reduction, 80
Beilinson–Soul´ e conjecture, xi
big Witt ring, 73
Bloch’s conjecture on zerocycles, xii, 17
Bloch’s conjectured ﬁltration on Chow groups,
xii, 16
Bloch–Kato conjecture on special values of
Lfunctions, xix
Bloch–Kato–Milnor conjecture, viii
Bloch–Ogus spectral sequence, 52
Bockstein map, 66
Borel–Moore cohomology theory, 51
Brauer group, 81, 94
CH
r
(X), Chow group, 4
Chˆ atelet surface, 83, 89
characteristic sequence theorem, 48
Chern character, 66
χ
Gr¨ oss
, Gr¨ ossencharakter, 119
Chow group, 4
ﬁnite dimensional, 21
higher, xi
of zerocycles, 79
Chow motive, xi
conductor, 118
conic bundle, 29, 80, 89
coniveau ﬁltration, xvii, 66
corestriction, 63
crystalline cohomology, 76
cubic surface, 89
cycle class map, xv
cycle map, 41, 97
cyclic homology, xvii
degree map, 69, 79
Deuring’s theory, 119
devissage theorem, 48
de Rham cohomology, 50
Dieudonn´ e ring, 73
diophantine questions, xviii
Div(X
0,F
· ), 88
divisor, 3
eﬀective motive, x
Eichler norm theorem, 80, 90
elliptic curve, 100
´ etale cohomology, 50
exact category, 45
Fano surface, 14
ﬁltration on Chow groups, 16
ﬁnitedimensional Chow group, 21
ﬁnitely generated projective modules, category
of, 46
130 Index
formal curve, 74
¯
G
a
, formal additive group, 74
G
m
, multiplicative group, 37
¯
G
m
, formal multiplicative group, 74
Galois symbol, 63, 91
geometric motive, x
Gersten resolution, xv, 50, 69
Godeaux surface, 13
good reduction, 86
Gr¨ ossencharakter, 119
Griﬃths group, xvi
Grothendieck’s standard conjectures, xii
1, Zariski sheaf, 52
Hasse–Weil Lfunction, xv, xix
Hodge ﬁltration, xiv, 25, 40
Hodge structure
polarized of weight two, 5
hyperelliptic surface, 13
incidence correspondence, 27
intermediate jacobian, xiv, 5, 25, 37
isogeny, 28
jacobian, 4
intermediate, xiv, 5, 25, 37
Kcohomology, xv, xvii
inﬁnitesimal structure, 70
Ktheory, 45
Milnor, ix
Ktheory sheaf
analytic, 69
7
an
2
, analytic Ktheory sheaf, 69
K¨ ahler diﬀerential, 71
K¨ unneth projectors, xii
localization theorem, 48
power torsion subgroup, 61
Milnor ring, xvi, 89
Milnor–Bloch–Kato conjecture, xvi
mixed Hodge structure, xiv, 40
mixed motive, xii
motive
Chow, xi
eﬀective, x
geometric, x
mixed, xii
motivic cohomology, ix, xvii
multizeta numbers, xx
multiplicative group, 37
formal, 74
Mumford’s theorem, 13
Murre’s conjecture, xii
N´ eron–Severi group, 14
N´ eron–Severi torus, 80
nerve (of a small category), 45
nilpotence conjecture, xii
Nisnevich topology, ix
noetherian scheme, 48
NS(S ), N´ eron–Severi group of S , 14
period, 4, 105
Pic(S ), Picard variety of S, 14
Picard variety, 5, 14
plus construction, Quillen’s, 46
presheaf, viii
with transfers, viii
prorepresentable functor, 72
Prym variety, 27
quantum ﬁeld theory, xx
quartic threefold, 25
Quillen Ktheory, 45
rational surface, 79
split, 85
regulator, xv, xviii, xix
regulator formula, 120
resolution theorem, 48
Riemann–Roch theorem, 3
Roitman’s theorem, 59
scissors conguence, xviii
singular cohomology, 50
site, 52
Soule’s conjecture on zeta functions, xx
split rational surface, 85
Steinberg group, 47
Steinberg symbol, xvi, 47
symbol map, 70
T(Y), Albanese kernel of Y, 11
tame symbol, 88, 91, 98
theorem of the hypersquare, xiii, 18
torsion in the Chow group, xvii, 59
trace morphism (on ´ etale cohomology), 64
transfer map, 65
“trivial” curve, 98
valuation map, 88
vanishing cycle, 34
weight ﬁltration, xiv, 40
Witt vector cohomology, Serre’s, 76
Zariski sheaf, 52, 62
zeta function, xviii
This page intentionally left blank
Lectures on Algebraic Cycles Second Edition
Spencer Bloch’s 1979 Duke lectures, a milestone in modern mathematics, have been out of print almost since their ﬁrst publication in 1980, yet they have remained inﬂuential and are still the best place to learn the guiding philosophy of algebraic cycles and motives. This edition, now professionally typeset, has a new preface by the author giving his perspective on developments in the ﬁeld over the past 30 years. The theory of algebraic cycles encompasses such central problems in mathematics as the Hodge conjecture and the Bloch–Kato conjecture on special values of zeta functions. The book begins with Mumford’s example showing that the Chow group of zerocycles on an algebraic variety can be inﬁnite dimensional, and explains how Hodge theory and algebraic Ktheory give new insights into this and other phenomena. spencer bloch is R. M. Hutchins Distinguished Service Professor in the Department of Mathematics at the University of Chicago.
B. Shlapentokh Hilbert’s Tenth Problem G. Enguehard Representation Theory of Finite Reductive Groups J. For a complete series listing visit http://www. Garnett and D. Valette Kazhdan’s Property (T) J. Grandis Directed Algebraic Topology G. Schertz Complex Multiplication . Marshall Harmonic Measure P. Mrowka Monopoles and ThreeManifolds B. Gubler Heights in Diophantine Geometry Y.org/uk/series/sSeries. Baker and G. Ionin and M. Bombieri and W. P.asp?code=NMM 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 M.NEW MATHEMATICAL MONOGRAPHS Editorial Board B´ la Bollob´ s e a William Fulton Anatole Katok Frances Kirwan Peter Sarnak Barry Simon Burt Totaro All the titles listed below can be obtained from good booksellers or from Cambridge University Press. S. P. J. Berhanu. Cabanes and M. Bekka. Michler Theory of Finite Simple Groups I A. Hounie An Introduction to Involutive Structures A. Shrikhande Combinatorics of Symmetric Designs S.cambridge. Cordaro and J. E. Kronheimer and T. Neisendorfer Algebraic Methods in Unstable Homotopy Theory M. D. Michler Theory of Finite Simple Groups II R. de la Harpe and A. W¨ stholz Logarithmic Forms and Diophantine Geometry u P. Cohn Free Ideal Rings and Localization in General Rings E.
Lectures on Algebraic Cycles Second Edition SPENCER BLOCH University of Chicago .
org Information on this title: www. New York www. Cambridge CB2 8RU. Delhi. Melbourne. Subject to statutory exception and to the provision of relevant collective licensing agreements. accurate or appropriate.CAMBRIDGE UNIVERSITY PRESS Cambridge. Dubai. . UK Published in the United States of America by Cambridge University Press. and does not guarantee that any content on such websites is. São Paulo. Singapore. or will remain.cambridge. Madrid.cambridge. New York. First published in print format 2010 ISBN13 ISBN13 9780511900310 9780521118422 eBook (EBL) Hardback Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or thirdparty internet websites referred to in this publication. Tokyo Cambridge University Press The Edinburgh Building. Cape Town. no reproduction of any part may take place without the written permission of Cambridge University Press.org/9780521118422 © David Leibowitz 2010 This publication is in copyright.
Contents Preface to the second edition 0 1 Introduction page vii 3 9 21 25 37 45 59 69 79 97 115 123 129 Zerocycles on surfaces Appendix: On an argument of Mumford in the theory of algebraic cycles Curves on threefolds and intermediate jacobians Curves on threefolds – the relative case Ktheoretic and cohomological methods Torsion in the Chow group Complements on H (K2 ) Diophantine questions Relative cycles and zeta functions Relative cycles and zeta functions – continued Bibliography Index 2 2 3 4 5 6 7 8 9 .
.
In the original volume I included a quote from Charlie Chan... It is only intended to illustrate the various ideas mentioned in the text. Looking back over these lectures. .. given at Duke University in 1979. I think now that the question remains very very hard. I can say with some pride that they contain early hints of a number of important themes in modern arithmetic geometry.Preface to the second edition 30 Years later. The remaining sections until the last follow roughly the content of the original book. motives in physics. The last section. thirty years later. To bring them up to date would involve writing several more monographs. Given the amazing deep new ideas introduced into the subject in recent years. represents my recent research.. What I propose instead is to comment fairly brieﬂy on several of the lectures in an attempt to put the reader in touch with what I believe are the most important modern ideas in these areas. the great Chinese detective. who told his bumbling number one son “answer simple. however. which is by no means complete. The section on motives just below is intended as a brief introduction to the modern viewpoint on that subject.” It seemed to me an appropriate comment on the subject of algebraic cycles. Of course. seriously out of date. but question very very hard. At the end of this essay I include a brief bibliography. the ﬂip side of that coin is that they are now. a task best left to mathematicians thirty years younger than me. though the titles have changed slightly to reﬂect my current emphasis. but the answer is perhaps no longer so simple.
There is an evident functor Smk → Cork which is the identity on objects. Presheaves on Cork are referred to as presheaves with transfers. 2006. In particular. Let k be a ﬁeld. . Voevodsky et al. 2000) – of a triangulated category of mixed motives. A presheaf with transfers F is called homotopy invariant if. one deﬁnes Ztr ( n Gm ) by taking Xi = A1 − {0} and xi = 1. An important elaboration on this idea yields for pointed objects xi ∈ Xi Ztr ((X1 .viii Preface to the second edition Motives Much of the recent work in this area is centered around motives and the construction – in fact various constructions. An Avalued presheaf F on Cork induces a presheaf FSmk on Smk .1) ∆n := Spec k[t0 . 0]) C∗ (F) := U → · · · → F(U × ∆n ) → · · · → F(U × ∆0 ) Here (0. X). For X ∈ Ob(Smk ) one has the representable sheaf Ztr (X) deﬁned by Ztr (X)(U) = HomCork (U. Levine (1998). to lift a presheaf G from Smk to Cork one needs a structure of trace maps or transfers f∗ : G(Z) → G(X) for Z/X ﬁnite. with obvious notation. Intuitively. and Voevodsky (Mazza et al. Morphisms Z : X → Y are ﬁnite linear combinations of correspondences Z = ni Zi where Zi ⊂ X × Y is closed and the projection πi : Zi → X is ﬁnite and surjective. I will sketch Voevodsky’s construction as it also plays a central role in his proof of the Bloch–Kato–Milnor conjecture discussed in Lecture 5. . A presheaf on a category C with values in an abelian category A is simply a contravariant functor F : Cop → A. Then I will discuss various lectures from the original monograph. tn ]/( ti − 1) . The category Cork is an additive category with objects smooth kvarieties: Ob(Cork ) = Ob(Smk ). . due to Hanamura (1995). Intuitively. The complex of chains C∗ (F) on a 0 1 presheaf with transfers F is the presheaf of complexes (placed in cohomological degrees [−∞. we may think of Zi as a map X → Symn Y associating to x ∈ X the ﬁbre fi−1 (x) viewed as a zerocycle on Y. xi )) := Coker Ztr (X1 × · · · Xi × · · · Xn ) → Ztr ( Xi ) . x1 ) ∧ · · · ∧ (Xn . i∗ = i∗ : F(U × A1 ) → F(U). .
Z(p) ⊗ Z(q) → Z(p + q) so one gets a product structure on motivic cohomology. Voevodsky employs the Nisnevich topology. More precisely. the above complex is viewed as a complex of presheaves on Smk and then localized for the Zariski topology.2) · · · → Ztr (U ×X U) → Ztr (U) → Z(X) → 0. Z(n)) Milnor Kn (k). and Ztr (X) ⊗ Ztr (Y ) = Ztr (X × Y ). (K∗ the ideal generated by quadratic relations a ⊗ (1 − a) for a ∈ k − {0. Another important sign that this is the right theory is the link with Milnor Milnor (k) is deﬁned as the quotient of the tensor algebra on k × by Ktheory. p > 0. . q ≥ 0. The fact that the Zariski topology suﬃces to deﬁne motivic cohomology is somewhat surprising because a Zariski open cover π : U → X does not yield a resolution of Zariski sheaves (0. Z(q)). Maps f0 . A1 homotopy is an equivalence relation. Z(1) Gm [−1]. so the homology presheaves Hn (C∗ (F)) are homotopy invariant. In low degrees one has 0 HM (X. Z(0)) = Z[π0 (X)]. The two restrictions i∗ . p HM (X. One has a notion of tensor product for presheaves on the category Cork . Motivic cohomology is then deﬁned (for q ≥ 0) as the hypercohomology of this complex of Zariski sheaves: p p HM (X.Preface to the second edition ix is the algebrogeometric nsimplex. To remedy this.) Theorem n HM (Spec k. 1}. Z(0)) = (0). f1 : X → Y in Cork are A1 homotopic if there exists H : X×A1 → Y in Cork such that f j = i∗ H. In particular. A morphism π : U → X is a Nisnevich cover if for any ﬁeld K/k one has U(K) X(K). i∗ : C∗ (F)(U × A1 ) → C∗ (F)(U) 0 1 are shown to be homotopic. The boundary maps in the complex are the usual alternating sums of restrictions to the faces ∆n−1 → ∆n deﬁned by setting ti = 0. Z(q)) := HZar (X. Gm )[−q]. and A1 j homotopic maps induce homotopic maps f0∗ Voevodsky deﬁnes Z(q) := C∗ Ztr ( q f1∗ : C∗ Ztr (X) → C∗ Ztr (Y).
Projective bundle theorem For V/X a rank n + 1 vector bundle M(P(V)) n i=0 M(X)(i)[2i]. and one deﬁnes eﬀ DMNis (k) := D− ShNis (Cork )[W −1 ]. N(q)). One has the following properties: MayerVietoris M(U ∩ V) → M(U) ⊕ M(V) → M(X) → M(U ∩ V)[1] is a distinguished triangle. the motive associated to a smooth kvariety X is deﬁned by (0. Write M(q) := M ⊗ Z(q). M(V) The category of (not necessarily eﬀective) motives is obtained by inverting the functor M → M(1) in DM eﬀ . Then Hom(M. Finally. . one formally inverts all morphisms with cones in W. ¨ Kunneth M(X × Y ) = M(X) ⊗ M(Y ).x Preface to the second edition To see that (0. Vector bundle theorem M(X) for V/X a vector bundle. One considers the smallest thick subcategory W containing all cones of Ztr (X × A1 ) → Ztr (X). one uses the fact that any ﬁnite cover of a Hensel local ring is a product of local rings.3) eﬀ M(X) := Ztr (X) ∈ DMNis (k).2) becomes exact when localized for the Nisnevich topology. The actual triangulated category of eﬀective motives over k is a quotient category of the derived category D− (ShNis (Cork )) of boundedbelow complexes of Nisnevich sheaves on Cork . Cancellation Assume varieties over k admit a resolution of singularities. eﬀ eﬀ The category of geometric motives DMgeo (k) is the thick subcategory in DMNis (k) generated by the M(X). N) Hom(M(q). Said another way.
the usual Chow group CHq (X) = CHq (X. p.9) p H M (X. m) with X smooth projective over k. Y are smooth projective. M(Y )) CHdimX (X × Y ).Preface to the second edition Chow motives If X. 0).6) p H M (X. we have (0. In fact. Voevodsky proves that for X smooth over a perfect ﬁeld k one has (0. •)[−2q]). n) to be the free abelian group of algebraic cycles on X × ∆n which are in good position with respect to all faces X × ∆m → X × ∆n . eﬀ It follows from (0.8) CHq (X. m). •)). their conjecture was formulated in terms of the γﬁltration in Ktheory. 2) → Zq (X.e. •)[−2q] ⊗ Q has cohomological support in degrees [0. 1) → Zq (X. The complex Zq (X. 2q − p) ⊗ Q grq K p (X) ⊗ Q. Let ∆• be the cosimplicial variety as in (0.) The higher Chow groups are deﬁned by (0. and this relationship lies at the heart of modern cycle theory. 2q]. q. The morphisms are given by (0. 2q − p) = H p (Zq (X. Z(q)[p]). n) := H −n (Zq (X. Deﬁne Zq (X. There are a number of ways to formulate things. n)) := q ◦ CHdimX+n−m (X × Y ) ◦ p. Z(q)) ⊗ Q CHq (X.1) above. Motivic cohomology For X/k smooth. Beilinson and Soul´ conjecture that the shifted chain complex e Zq (X. but one has the further identiﬁcation (0. motivic cohomology is closely related to algebraic cycles. xi The category of Chow motives over a ﬁeld k has as objects triples (X.10) p H M (X. p ∈ CHdimX (X × X)Q a projector (i. Z(q)) CHq (X. Actually. Z(q)) eﬀ HomDMNis (k) (Ztr (X).5) Hom((X. For example.7) · · · → Zq (X. p ◦ p = p) and m ∈ Z. I will use higher Chow groups because they relate most naturally to arithmetic questions. (Here Zq (X. n) is placed in cohomological degree −n.4) Hom(M(X). •) is deﬁned by taking alternating sums of pullbacks in the usual way: (0. p. γ . then (0. (Y.4) and the existence of projectors in DMNis (k) that the cateeﬀ gory of Chow motives embeds in DMNis (k). 0) → 0.
xii
Preface to the second edition
Lecture 1: Zerocycles
The two most important ideas here are ﬁrstly the conjecture that surfaces with geometric genus zero (pg = 0) should have Chow group of zerocycles representable. For S such a surface over C we expect an exact sequence − 0 → Alb(S ) → CH0 (S ) −→ Z → 0. The group T (S ) deﬁned in Lemma 1.4 is conjectured to be zero in this case. Secondly, for any smooth projective variety X, the Chow group of zerocycles CH0 (X) is conjectured (1.8) to carry a descending ﬁltration F ∗ CH0 (X) which is functorial for correspondences such that
p p the map grF CH0 (X) − → grF CH0 (Y ) induced by an algebraic cycle Λ ∈ CHdimY (X × Y ) depends only on the class of Λ in cohomology. Indeed, one may conjecture the existence of such a ﬁltration on CHq (X) for any q. These conjectures remain unproven, but a very beautiful general picture, based on the yoga of mixed motives, has been elaborated by A. Beilinson. Interested readers should consult the important article by Jannsen (1994) and the literature cited there. Let me sketch brieﬂy (following Jannsen) the modern viewpoint. It is convenient to dualize the deﬁnition of M(X) (0.3). Assume X smooth, projective of dimension d. Deﬁne (Hom means the internal Hom in DM) Λ deg
M(X)∗ = HomDM (M(X), Z(0)). The formula for the Chow group becomes
2p CH p (X) = H M (X, Z(p)) = HomDM (Z(0), M(X)∗ (p)[2p]).
One of Grothendieck’s standard conjectures about algebraic cycles is that there exist K¨ nneth projectors u πi ∈ CHd (X × X)Q /{homological equivalence} inducing the natural projections H ∗ (X) → H i (X) on cohomology. If we assume further that the ideal {homological equiv.}/{rational equiv.} ⊂ CHd (X × X)Q is nilpotent (nilpotence conjecture), then the pri lift (noncanonically) to projectors pri,rat ∈ CHd (X × X)Q and we may use (0.5) to decompose M(X)∗ ⊗ Q = i i h [−i] noncanonically as a direct sum of Chow motives. This idea is due to J. Murre (1993a). The hope is that DM admits a tstructure such that (0.11) H i (M(X)∗ ⊗ Q) = hi (X).
Preface to the second edition The resulting spectral sequence
xiii
p,q E2 = HomDM (Z(0), H q (M(X)∗ ( j)[p]) ⇒ HomDM (Z(0), M ∗ (X)( j)[p+q])
would yield ﬁltrations on the Chow groups ⊗Q with F ν CH j (X)Q grν CH j (X)Q F
2 j−ν i=0
Ext2 j−i (Q(0), hi (X)( j)), DM
Extν (Q(0), H 2 j−ν (X)( j)). DM
Murre suggests a natural strengthening of his conjectures, based on the idea that one should be able for i ≤ d, to ﬁnd representatives for πi supported on Xi × X ⊂ X × X, where Xi ⊂ X is a general plane section of dimension i. For example, π0 = {x} × X for a point x. Clearly this would imply πi CH j (X) = 0 for i < j, and since the conjectures imply F ν CH j (X) = we could conclude further that F ν CH j (X) = (0), ν > j.
2 j−ν i=0
πi CH j (X),
Suppose, for example, that dim X = 2. We would get a 3step ﬁltration on the zerocycles: CH0 (X) = F 0 ⊃ F 1 ⊃ F 2 ⊃ (0), with gr0 CH0 (X)Q = HomDM (Q(0), h4 (X)(2)), F gr1 = Ext1 (Q(0), h3 (X)(2)), F DM gr2 = Ext2 (Q(0), h2 (X)(2)). F DM This ﬁts perfectly with the ideas in Lecture 1. Indeed, Murre has computed gr0 and gr1 and he ﬁnds exactly the degree and the Albanese. Of course, gr2 is more problematical, but note that the condition discussed in the text that H 2 (X, Q (1)) should be generated by divisors (which is equivalent to pg = 0 in characteristic zero) means h2 (X)(2) ⊕Q(1). (This is obvious for motives modulo homological equivalence. Assuming the nilpotence conjecture, it holds for Chow motives as well.) In this case, gr2 can be computed for X = P2 when F it is clearly (0). The conjectural “theorem of the hypersquare” (Proposition 1.12) can be understood in motivic terms (Jannsen 1994, conj. 3.12) using the fact that hn (X0 × · · · Xi · · · × Xn ). hn (X0 × · · · × Xn ) is a direct summand of Clearly wrongheaded, however, is Metaconjecture 1.10, which stated that 2 F CH0 (X) is controlled by the polarized Hodge structure associated to H 2 (X). Indeed, Ext2 = (0) in the category of Hodge structures. One may try (compare Carlson and Hain 1989) to look at Exts in some category of variations of Hodge
xiv
Preface to the second edition
structure. In the absence of parameters supporting such variations (i.e. for X over a number ﬁeld), however, the Ext2 term should vanish and we should have F 2 CH2 (X) = (0).
Lectures 2 and 3: Intermediate jacobians
The modern point of view about intermediate jacobians is to view them as Ext1 (Z(0), H) where H is a suitable Hodge structure, and the Ext group is taken in the abelian category of mixed Hodge structures (Carlson 1987). In the classic situation, H = H 2r−1 (X, Z(r)) where X is a smooth projective variety. Note in this case that H has weight −1. An extension 0 → H → E → Z(0) → 0 would yield a mixed Hodge structure E with weights 0, −1 and Hodge ﬁltration EC = F −1 EC ⊃ F 0 EC · · · . Let f, w ∈ EC be liftings of 1 ∈ Z(0) splitting the weight and Hodge ﬁltrations respectively. The diﬀerence between them w − f gives a welldeﬁned class in J = HC /(F 0 HC + HZ ) which is the intermediate jacobian. To deﬁne the class of a codimensionr cycle Z = ni Zi , let Z be the support of the cycle. We 2r have a cycle class with supports [Z] : Z → HZ (X, Z(r)) and a diagram (0.12)
2r−1 HZ (X, Z(r))
H 2r−1 (X, Z(r))
H 2r−1 (X − Z, Z(r))
2r HZ (X, Z(r)) Z
H 2r (X, Z(r))
0
Z
H 2r (X, Z(r))
2r−1 The ﬁrst group HZ (X, Z(r)) is zero by purity, and vanishing of the lower righthand arrow will hold if Z is homologous to 0. Assuming this, we get the desired extension of Hodge structures. (I believe this construction is due to Deligne, though I do not have a good reference.) For H any mixed Hodge structure with weights < 0 one has an analogous construction. Note, however, the resulting abelian Lie group J need not be C× S 1 × R. In Lecture 3, the compact. For example, Ext1 (Z(0), Z(1)) focus is on the case 1 3 H = H 1 (C, Z(2)) ⊗ Hc (Gm , Z)⊗2 ⊂ Hc (C × (Gm )2 , Z(2)).
one considers Zariski sheaves q q Kq (resp. From the point of view of Lecture 3. CH p (X)/{algebraic equivalence}. C× (1)). One obtains resolutions of these sheaves which enable one to identify. resp. HBetti . prop. It is an Rvector space. and then much more deﬁnitively by Beilinson (1985). More generally one can associate to any mixed Hodge structure a nilpotent matrix γ (Cattani et al. see Bloch (2000). for example. xv At the time I was very much inspired by the work of Borel (1977) on regulators for higher Kgroups of number ﬁelds. Het ) associated to the presheaf U → Kq (U) of algebraic q q Kgroups (resp. For details of this construction. Z/nZ)). The quotient of this Ext group by its maximal compact subgroup is the corresponding Ext in the category of RHodge structures. Other constructions are given in Bloch (1986b) and Goncharov (2005). X × ∂∆2q−p . Z(q)). resp. 1986. which is the obstruction to a real splitting of the ﬁltration by weights.. 2000). 2. Brieﬂy. if the curve C in Lecture 3 is allowed to degenerate. H) = HC /(HZ +F 0 HC ). K p ) p H p (X. I describe the “Quillen trick” and use it ´ to construct the Gersten resolution for Ksheaves and also Betti and etale cohomology sheaves for smooth varieties. Z(q)) (0. and that these regulators could be related to values of Hasse–Weil Lfunctions. for example.20).Preface to the second edition Here H has weight −3. U → Het (U. the regulator can be thought of as a relative cycle class map p H M (X. and Ext1 (Z(0).13) H p (X. an important tool in the study of algebraic cycles. These invariants arise. I think it is fair to say that the resulting Kcohomology and the parallel ´ constructions for Betti and etale cohomology have had important technical applications but have not been the breakthrough one had hoped for at the time. 2q− p) → Ext1 (Z(0). so H 1 (C. C)/H 1 (C. This was done in a very limited and ad hoc way in Bloch (1980. HBetti ) CH p (X). (0. Z(2)) H 1 (C. . H) MHS H 1 (C. I believed that similar regulators could be deﬁned for arithmetic algebraic varieties more generally.9) CHq (X. MHS Here H = H 2q−1 (X × ∆2q−p . U → HBetti (U. Lecture 4: Cohomological methods This chapter contains basic information about algebraic Ktheory. Z). Z) is itself a mixed Hodge structure.
It follows in such cases that a is not surjective. The proof is now complete (for an outline with references. f 0. For simplicity I assume µ ⊂ k so there is no need to distinguish powers of µ . 1. and let j : Y − X → Y be the open immersion. i∗ ∂ . Consider the exact sequence → − H p (Y. Z/ ) → H p+1 (Y. is a proof that Kn (F) → H n (F. H 2 ) → HBetti (X) a b c Using (0. One nice application of the Betti theory (see reference [6] at the end of Lecture 4) was to falsify a longstanding conjecture about diﬀerential forms of the second kind p. indeed Coker(a) is inﬁnitely generated. But Γ(X. For some years Voevodsky has been working on a very diﬃcult program.14) with the Griﬃths group of cycles homologous to zero modulo algebraic equivalence.14) 3 − − − 4 HBetti (X) → Γ(X. unlike the case of curves and surfaces. a group which in some cases is known not to be ﬁnitely generated (Clemens 1983). Rost. H 3 ) → H 2 (X. Finiteness results. that is meromorphic forms which at every point diﬀer from an algebraic form which is regular at the point by an exact algebraic form. j! Z/ ). µ∗ ) is an isomorphism. H 3 ) is precisely the space of meromorphic 3forms of the second kind. The conjecture in question states that the natural map to Galois cohomology M K∗ (F)/ → H ∗ (F. The spectral sequence E2 = HZar (X. Weibel). which is explained in LecM ture 5.13) one can identify Ker(c) in (0. diﬀerential forms of the second kind do not necessarily come from global cohomology classes in dimensions ≥ 3. for example. My own contribution to this. Lecture 5: The conjecture of Milnor–Bloch–Kato M Let F be a ﬁeld and a prime with 1/ ∈ K. but there is still no uniﬁed treatment and the arguments use sophisticated techniques in algebraic homotopy theory which I have not understood. it turns out to be diﬃcult to interpret the resulting cohomology. HBetti ) → p+q HBetti (X. Let i : X → Y be a closed immersion of varieties over a ﬁeld k of characteristic prime to . µ⊗n ) is surjective when F has cohomological dimension n. the result can be formulated as follows.xvi Preface to the second edition Despite the Gersten resolution. The Milnor ring K∗ (F)/ is gener× × ated by F /F with relations given by Steinberg symbols f ⊗ (1 − f ). Z/ ) − H p (X. are totally lacking. to prove the conjecture in complete generality.q q p on varieties of dimension ≥ 3. see the webpage of C. Geometrically. using his own motivic theory and results of M. Z) leads to an exact sequence (0. Thus.
9). if A is a nonreduced ring. The situation should be compared with that for abelian varieties A where one has Tate modules T (A) for all and these calculate H1 (A. we do not have contravariant cohomology functors deﬁned on singular schemes (e. on nonreduced schemes) with appropriate properties. •)[−2r]. It is fair to say that we still do not have an adequate notion of motivic cohomology. the cycle complexes Zr (X. For example. I include in the bibliography a couple of relevant papers (Soul´ e and Voisin 2005. An important step has been the work of Goodwillie (1986) computing the Ktheory of nilpotent ideals in characteristic zero in terms of cyclic homology. com´ pute the adic etale cohomology for all prime to the characteristic. and in this lecture we examine what can be learned from the inﬁnitesimal structure of these groups. the reader might work out how this is equivalent to the conjecture for F = k(X). the Kcohomology groups HZar (X. Z ) for prime to the characteristic.Preface to the second edition xvii Given a cohomology class c ∈ H p (X) and a smooth point x ∈ X. p Curiously. Let X be a smooth. Lecture 6: Inﬁnitesimal methods in motivic cohomology The inﬁnitesimal methods developed here were used also in my work on de Rham–Witt cohomology (Bloch 1977). assuming r ≥ dim X. the reader could consult the two rather experimental papers Bloch and . 2000). Z(1)). j! Z/ ). quasiprojective scheme over an algebraically closed ﬁeld k. there exists a Zariski open neighborhood Y ⊃ U x such that 0 = ∂(c) U ∈ H p+1 (U. Z(1)) = A× 1 A[t]× = HM (Spec A[t]. The subject of torsion in the Chow group seems to be important from many points of view. Kq ) discussed in Lecture 4 do have the correct functoriality properties. To understand what motivic cohomology of an inﬁnitesimal thickening might mean. Z/mZ(r)) 2r−n Het (X. Another exercise is to formulate a version of coniveau ﬁltration as described on page 52 for the group H p+1 (Y. Said another way. As an exercise. That is. The whole picture of motivic cohomology with ﬁnite coeﬃcients is now quite beautiful (Voevodsky et al.g. The notion of A1 homotopy invariance which is essential in Voevodsky’s work is not what one wants. in equation (0. Let r ≥ dim X. then typically 1 HM (Spec A. Then 2r−n HM (X. Z/mZ(r)). j! Z/ ) in such a way that the image of ∂ lies in F 1 . Totaro 1997). and let m ≥ 2 be relatively prime to the characteristic.
More deﬁnitive results have been obtained in Krishna and Levine (2008). Z(q)) mean? (Here ε2 = 0 and the map sends ε → 0. but the limit as a → b is Ga given by degreezero line stant Gm for a bundles on P1 with double order trivialization at a = b. that does not happen. For example. b. that is isomorphism classes of line bundles on P1 with trivializations at a. perhaps. consider the Picard scheme of (P1 . this is another one of those limiting phenomena where the radius of the hyperbolic disk is allowed to go to inﬁnity and lengths are scaled so in the limit one gets Euclidean geometry. and it seems likely that there is a similar relation between T H M and Euclidean volumes. Soul´ 1986. It would be nice to have a rigorous description of how this works. The reader who wants to work on diophantine questions regarding zerocycles and Chow groups should consult ColliotTh´ l` ne (1995). 1988. one should probably not think of T H M as a tangent space in the usual sense. regulators. what should the “tangent space” p T HM (X. {a. Lectures 8 and 9: Regulators and values of Lfunctions The whole subject of motivic cohomology.) To begin with. Park (2007). As far as I can tell. Lecture 7: Diophantine questions At the time of these lectures. b}). for ex3 ample for HM (k. It can be nontrivial in situations where the motivic cohomology itself is rigid. and R¨ lling (2007). Intuitively. and I have not thought further in this direction. Bloch and Kato 1990).xviii Preface to the second edition Esnault (1996. The regulator for usual motivic cohomology is closely related to volumes in hyperbolic space (Goncharov 1999). but we now understand much better what should be true (Rapoport et al. For e . and the references given in these papers. I had expected that the Chow group of zerocycles on a rational surface would relate in some way to the zeta function of the surface. ee Esnault (2003). One important area of open questions about these groups concerns regulators and relations with Euclidean scissors congruence groups. and values of Lfunctions remains to a large extent conjectural. to think of a nonsemistable moduli functor where jumps can occur at the boundary. Z(2)) with k a number ﬁeld. 2003). The degreezero part is conb. Z(q)) → H p (X. u Assuming one has a good deﬁnition of motivic cohomology. Better. Z(q)) := Ker H p (X × Spec k[ε].
Q(q)) + for the Betti cohomology of X(C). To understand their idea. → Ext1 f (Q(0). Write MB = HB (X. where HDR is + de Rham cohomology. the reader can consult Goncharov (2005) and Bloch (1986b). The subscript f refers to behavior at ﬁnite primes. The various determinants have Qstructures (though α. and write M ∗ (1) := h2d−p (X)(d − q + 1). •)[−2q] ⊗ Zd+1−q (X.15) one gets a trivialization over R of the tensor product of determinant lines (0. L .15) 0 → HomDM (Q(0). They show that the integral conjecture in Bloch and Kato (1990) is equivalent to this ratio being given by L∗ (M.16) det(R HomDM. the ﬁrst nonvanishing term in the Taylor expansion of L(M. Concerning the basic conjecture. As a consequence of (0. f (Q(0). In fact. does not).) Consider a motive M = h p (X)(q). Q(q))/F 0 . so one may examine in (0. using Galois and adic cohomology. let X be a smooth. okay. Let t M := HDR (X. Z(q)) and R ΓM (X. •)[−2(d + 1 − q)] which one should think of as concrete realizations of motivic cohomology R ΓM (X. M) ⊗ R → Ker α → (Ext1 f (Q(0). Well. there is a lot here we do not understand. Fontaine and PerrinRiou 1994). 0). itself. s) where L(M. M ∗ (1)) ⊗ R)∗ → 0. Z(d + 1 − q)). There is a natural map α : MB ⊗ R → t M ⊗ R. M) ⊗ R → Coker α → (HomDM (Q. f (Q(0). M ∗ (1)))R ⊗ det(α)−1 R.Preface to the second edition xix constructions of the regulator. M ∗ (1)) ⊗ R)∗ DM. Let MB ⊂ MB be the subspace invariant p ∗ under the action of conjugation. (In what follows I gloss over many intractable conjectures. •)[−2q] and Zd+1−q (X. M))R ⊗ det(R HomDM. the authors actually get a Zstructure on the left.16) the ratio of the real trivialization and the rational structure. DM. I am especially attracted to the formulation given by Fontaine and PerrinRiou (Fontaine 1992. and (assuming certain conjectures) Fontaine and PerrinRiou construct an exact sequence of motivic cohomology (0.11). s) is the Hasse–Weil Lfunction associated to M. •)[−2(d + 1 − q)] → Z1 (Spec Q. Here Hom and Ext are taken in the triangulated category of Voevodsky motives over Q. •)[−2d].15) working directly with the cycle complexes Zq (X. projective variety of dimension d over Q. but my thought is that one might redo (0. where h p (X) is a Chow motive p as in (0. The resulting determinant metrics could perhaps be deduced (or at least interpreted) directly from the intersection–projection map (well deﬁned in the derived category) Zq (X.
The result should be some kind of metric or related structure on the determinant of the cycle complex. Z(p)) via the realization map from motivic cohomology to Betti cohomology. a class in HM (XΓ . For example. in computations of scattering amplitudes.e. then HBetti (XΓ . Of course. which says in this e context that for X regular and proper of dimension d over Spec Z. the “wheel with n spokes graph” WS (n) gives rise to a hypersurface XWS (n) of dimension 2n − 2. Q(p)) = −ord s=d−p ζX (s). The Hodge conjecture would say that such a class comes from an algebraic cycle. Q) would necessarily be a Hodge class. I would say it is high time we start looking for applications. Essentially. Dirk Kreimer has been teaching me about Feynman amplitudes and perturbative calculations in quantum ﬁeld theory. so any Tate class. and the Feynman amplitude is a period of this hypersurface relative to a reference symplex. Belkale and Brosnan 2003). Coda: Motives in physics The subjects of algebraic cycles and motives have enjoyed a tremendous theoretical development over the past 30 years. If indeed these are multizeta numbers it should be the case that the cohomology n of XΓ has a big Tate piece. Broadhurst and Kreimer 1997. the XΓ are highly singular.e. The primitive cohomology in the middle dimension for this graph is computed (Bloch et al.e. the Kirchhoﬀ polynomial (Bloch et al. any map of Hodge structures n Z(−p) → HBetti (XΓ . There are all kinds of possibilities for interesting cohomology classes coming via realization from motivic cohomology. Q) would have pure weight n. 2006) to be Q(−2) (independent of n). i. 2p i. but cf. Unfortunately (or perhaps fortunately). •)[−2p]) = i (−1)i dimQ HM (X. •)[−2p] should be deﬁned and we should have χ(Z p (X. n = 2p. The periods in question are associated to graphs. Q(2)). 2006) of a graph Γ deﬁnes a hypersurface XΓ in projective space. the Euler characteristic of Z p (X.xx Preface to the second edition Hanamura suggested this approach to understanding heights and biextensions. An appropriate generalization of 2n−2 the Hodge conjecture would suggest a class in HM (XWS (n) . At the risk of scandalizing the reader. They have a strong tendency to be multizeta numbers (Bloch et al. 2006. It will be necessary to take the cone over the regulator map in some fashion. i. for example. The problem of computing such motivic cohomology groups can be attacked via the . One knows if XΓ were smooth. These are periods that arise. the negative of the order of zero or pole at s = d − p of the zeta function of X. This would ﬁt well with a conjecture of Soul´ (1984). as it stands it is purely algebraic. so all one can say are that the weights of H n are ≤ n.
vol. 2036–2070. Sci. 1984). Math. Varı chenko. 2003. 303–314. [8] Bloch. 1990. Algebraic Geom. [6] Bloch. ´ Inst. 537–568. 30. Providence. 135–172 in The Grothendieck Festschrift.. Goncharov. 61(3). [5] Bloch.. in Math. mixed Hodge structures and motivic cohomology of pairs of triangles on the plane. Higher Regulators. A. 1978). 58. 187–268 (1978). S. Pp. Schechtman. 1985. Contemp. Helsinki: Acad. 1980. and . N. 65–79 in The Lefschetz Centennial Conference. A. A.I. [10] Bloch.. Boston: Birkh¨ user Boston. Adv. Ann. Spencer. [7] Bloch. Spencer. 2000. J. 86. J. Pp..: American Mathematical Society. and H´ l` ne Esnault. Hautes Etudes Sci. motives. 1977. A. knowledge of motives should provide a strong organizing force to study complex physical phenomena. I. Spencer. Algebraic cycles and the Be˘linson conjectures.. The coniveau ﬁltration and ee nondivisibility for algebraic cycles. Math. Algebraic Ktheory and zeta functions of elliptic curves. Spencer. 1986a.. 1986b. Prakash. 3(3). but what has so far proved more powerful is to use classical algebrogeometric techniques to study the geometry of rank stratiﬁcations associated to a homomorphism of vector bundles u : E → F over projective space. Ideally. Math. Publ. Spencer J. A. 147–188. Algebraic cycles and higher Ktheory. V. 116(1). The moving lemma for higher Chow groups. R. [2] Be˘linson. and a conjecture of Kontsevich.. S. ı Pp. Matroids. Fennica. Vol. ı Soviet Math. 511–515 in Proceedings of the International Congress of Mathematicians (Helsinki.. References for preface [1] Be˘linson. 1996. Higher regulators and values of Lfunctions. Progr. V. vol. J. Math. Part I (Mexico City. [9] Bloch. and Patrick Brosnan.Preface to the second edition xxi combinatorics of the graph. B. a [3] Belkale. 304(2). Aomoto dilogarithms. More sophisticated methods involving monodromy and limiting mixed motives may give information about Landau singularities and unitarity of the S matrix. 1994. Algebraic KTheory. 267–304. and A. Algebraic Ktheory and crystalline cohomology. [4] Bloch. Even simple motivic invariants like weight and Hodge level should help physicists understand the periods arising in their computations.. Duke Math.
(4). Sci. Math. Ann. Inst. Armand. 1987). Sup.I. James A. Sympos. James A. I. Math. Ast´ risque. Homological equivalence. 1991/92. septembre 93). is not ﬁnitely generated. Pp. Providence. 4. 19–38 (1984). (2). 11. of Math. ColliotTh´ l` ne. 1989. The geometry of the extension class of a mixed Hodge structure. H´ l` ne. 1995. Providence. Sci. Varieties over a ﬁnite ﬁeld with trivial Chow ee group of 0cycles have a rational point. vol. 1985 (Brunswick.. 181–225. Pisa Cl. JeanLouis. Publ.xxii Preface to the second edition Zeta Functions of Elliptic Curves. Ann. 2003. Sup. 36(3). 187–191. Spencer. Phys. Math. 151(1). Sci. 751.I. Kreimer. Extensions of variations of mixed Hodge structure. and D. and Alexander B. 1987. and H´ l` ne Esnault. Phys. Clemens. Actes du Colloque de e Th´ orie de Hodge (Luminy. Goncharov. no. CRM Monograph Series. An additive version of higher ee ´ Chow groups. 123(3). 1992. Eduardo. and Richard M. Association of multiple zeta values with positive knots via Feynman diagrams up to 9 loops. exp. 1990. 333–400 in The Grothendieck Festschrift.. Ann. Esnault. 613– 636. e Deligne. Pure Math. e Ast´ risque. Invent. Ann. J. On motives ee associated to graph polynomials. a Bloch.: American Mathematical Society. Bloch. L’arithm´ tique des z´ rocycles (exee e e pos´ aux Journ´ es arithm´ tiques de Bordeaux. Fontaine. 1994. H´ l` ne Esnault. R. Cohomologie de SLn et valeurs de fonctions zeta aux points entiers. vol. Sup. 267(1).. 1977.: American Mathematical Society. 557–605. 457–535.. R. 1997. 4(4). Aroldo Kaplan. Mixed Tate motives. Progr. Hain. 463–477. Journal e e e de th´ orie des nombres de Bordeaux. Bowdoin. 2006. and Kazuya Kato.. e Cattani. 39–65. Vol. Hautes Etudes Sci. Carlson. 51–73. 1985). and Igor Kˇ´z. and Dirk Kreimer. of Math. D. Comm. Groupes fondamen´ taux motiviques de Tate mixte. Maine. Pierre. Ecole Norm. Pp. 199–222 in Algebraic Geometry. 2003. 2005. Spencer. S´ minaire Bourbaki. Carlson. Math. Bloch. Boston: Birkh¨ user Boston. Scuola Norm. 38(1). Proc. Spencer. 1983. Ann. 46. (4). e e [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] . 9. Spencer. 205–249. Bloch. (4). rı ˇ (2).. Lett. Broadhurst. Degeneration of Hodge structures. 1986. 86. modulo algebraic ´ equivalence. JeanMarc. B. Herbert. Valeurs sp´ ciales des fonctions L des motifs. and Wilfried Schmid. Ecole Norm. vol. 140(3).. 403–412. 393(3–4). Borel. Lfunctions and Tamagawa numbers of motives. vol. 1–56.
349–373. Lett. vol.: American Mathematical Society. On a conjectural ﬁltration on the Chow groups of an algebraic variety.. III. 811–821. Math. 319–349. Madhav V. 1991). Res.. 1996. 6(1). Masaki. 347–402. 1999. 189– 201. J. 2008. Motivic sheaves and ﬁltrations on Chow groups. Providence. R. 2000. 1994. Veriﬁcation of the conjectures for threefolds which are the product on a surface and a curve. 1. Mixed motives and algebraic cycles. Ann. 1998.. [34] Kahn.. Pure Math. Providence. Volumes of hyperbolic manifolds and mixed Tate motives. 61–82. J. Soc. 142(2). Eric M. Indag. Sympos. vol. I. (2). . Res. 4(2). 2. 124(2). 12(2). P. Thomas G. 4(2). 75–140. [36] Levine. 1986. Reine Angew. Relative algebraic Ktheory and cyclic homology. Alexander. Motivic complexes of Suslin and Voevodsky. Indag. 1991). Math. vol. [26] Friedlander.. Berlin: Springer. 569–618. P. Math. 1993b. 2005. Regulators. J. II. Lett... Berlin: Springer.I. [35] Krishna. Math. J. 599–706 in Motives (Seattle. Doc. 1(17). Math. Mixed Motives. Vol. JeanMarc.: American Mathematical Society. [27] Goncharov. 1081–1104 in Handbook of KTheory. 395–416. 1995. Providence. 2006. Pure Math. 295–349 in Handbook of KTheory. Lecture Notes on Motivic Cohomology. Math. Masaki. 57. Mixed motives and algebraic cycles. Alexander B. Mathematical Surveys and Monographs... 1. (N. (N. Pp. Applications of weighttwo motivic cohomology. 55.). Algebraic cycles and Hodgetheoretic connectivity. vol. [28] Goncharov.I. Additive higher Chow groups of schemes. 1994. Sympos. [30] Hanamura. R.S. Marc. Carlo. [31] Hanamura. [32] Hanamura. 111(2). Homological and cohomological motives of algebraic varieties. of Math. 1999. Amalendu. [38] Murre. Clay Mathematics Monographs. and Charles Weibel. 2(6). 245–302 in Motives (Seattle. Math. [39] Murre. Proc.: American Mathematical Society. 2. 2005. Pp. Math. Invent. [29] Goodwillie.I. The general conjectures and some examples. R. Uwe. [37] Mazza. Vladimir Voevodsky. 1993a.: American Mathematical Society. 55.). 177–188. I.I. and Bernadette PerrinRiou. Amer. [33] Jannsen. Autour des conjectures de Bloch et Kato: cohomologie galoisienne et valeurs de fonctions L. Masaki. 619. Vol. Bruno. Math. Proc.S. Pp. Pp. [40] Nori.Preface to the second edition xxiii [25] Fontaine. and Marc Levine. On a conjectural ﬁltration on the Chow groups of an algebraic variety. R. 2. Providence. Invent. 1993.
M. 2007. Warsaw: PWN. 107–127. Pure Math. and P. Int. Pp. N. Christophe. Algebraic Geom. Math. Mat. [49] Voevodsky.: Princeton University Press. Ast´ risque. The Chow ring of a classifying space. Burt. 16(1). 2007. Kay. J. The generalized de RhamWitt complex over a ﬁeld u is a complex of zerocycles. and Eric Friedlander.. vol. [46] Soul´ . ID rnm067.I. 198(1). 2 (Warsaw. 2000. IMRN. Andrei Suslin. 10(2). Adv. Transfers. 2004. and C. Vladimir. 109–169. Princeton. 421–438. Beilinson’s Conjectures on Special Values of LFunctions. 1594. [51] Voisin..J. R´ gulateurs. Vol. 67. Sympos. Kth´ orie et z´ ros aux points entiers de fonce e e tions zˆ ta. Seminar e e e Bourbaki.xxiv Preface to the second edition [41] Park. vol. Math. 1994. 237–253. Transcendental methods in the study of algebraic cycles.: American Mathematical Society. Claire. 143. 1984/85. 4. Annals of Mathematics Studies. Perspectives in Mathematics. Art.. [47] Totaro. Claire. N. 1983). 1997. vol. no. 1984.. C. Algebraic cycles and additive dilogarithm. Not. 467–493. and Motivic Homology Theories. [44] Soul´ . 19. Proc.. Pp. Lecture Notes in Math. Jinhyun. Torsion cohomology classes and algebraic e cycles on complex projective manifolds. Amer. 1986. [45] Soul´ . Ann. Res. Burt.. [43] R¨ lling.. 1993). R. Pp. 1999. 249–281 in Algebraic KTheory (Seattle. [48] Totaro. 183(3). [50] Voisin. J. 1997). Torsion algebraic cycles and complex cobordism. Boston: Academic Press. Christophe. Providence. Voisin. Schappacher. Cycles. Schneider (eds). Remarks on ﬁltrations on Chow groups and the Bloch conjecture. Soc. . Berlin: Springer. 153–222 in Algebraic Cycles and Hodge Theory (Torino. 1988. Math. [42] Rapoport. Pura Appl. (4). vol. 437–445 in Proceedings of the International Congress of e Mathematicians. 2005. 1.
LECTURES ON ALGEBRAIC CYCLES .
.
An algebraic cycle ni [Vi ] is said to be eﬀective if all the ni ≥ 0. X was an algebraic curve (Riemann surface) and the Vi were points pi . but the corresponding theorem for varieties of dimension > 1 lies deeper. They are the fruit of ten years reﬂection on algebraic cycles. one can use the Riemann– Roch theorem to prove rationality of the zeta function of an algebraic curve over a ﬁnite ﬁeld [9]. that is formal linear combinations ni [Vi ] of subvarieties Vi of a ﬁxed smooth and projective variety X. with integer coeﬃcients ni . Classically.0 Introduction The notes which follow are based on lectures given at Duke University in April. (For example.) The other great pillar of function theory on Riemann surfaces. We denote by Γ(X. The Riemann–Roch theorem computes the dimension (D) of the vector space of functions f such that ( f ) + D is eﬀective for a given divisor D. 1979. Ω1 ) the space of all global holomorphic X diﬀerential 1forms on X (where ω ∈ Γ(X. but its central role in the study of divisors does not seem to carry over to cycles of codimension greater than 1. the Abel– Jacobi theorem. so we may assume this and write our cycle D = ni ((pi ) − (p0 )) for some base point p0 . In this context. The reader will recall that to a meromorphic function f on a smooth variety or complex manifold one can associate a divisor (algebraic cycle of codimension 1) ( f ) by taking the sum of its components of zeros minus its components of poles. A necessary and suﬃcient condition for D to equal . Two (at least) of their results merit immortality: the theorems of Riemann–Roch and Abel–Jacobi. all counted with suitable multiplicity. tells when a given divisor D = ni (pi ) is the divisor of a function. Ω1 ) can be written locally as f (z) dz X for z and f (z) holomorphic). Clearly a necessary condition is that the degree ni must be zero. cycles were a principal object of study for nineteenthcentury complex function theory. It has been generalized quite considerably in recent years.
Here [k(V ) : k( f (V ))] is the degree of the extension of function ﬁelds. Z) := J(X) X divisors of functions is an isomorphism from A0 (X) to the abelian variety J(X). the map X X A0 (X) divisors of degree 0 → Γ(X. Even more X is true. one has f∗ : zr (X) → zr+e−d (Y ) deﬁned on a single irreducible codimensionr subvariety V ⊂ X by f∗ (V ) = [k(V ) : k( f (V ))] · f (V ). C) for the periods. Suppose now Γ ∈ zr (X × P1 ) and no component of Γ contains either X × {0} or X × {∞}. If f : X → Y is a proper map. zr (X)/zr (X). rat alg CHr (X) = CHd−r (X). Ω1 ). .4 ( f ) is that Lecture 0 pi ni p0 ω = ω γ for some (topological) 1cycle γ ∈ H1 (X. we may deﬁne for X any smooth projective variety over a ﬁeld and r ≥ 0 an integer zr (X) = free abelian group generated by irreducible subvarieties of X of codimension r. Replacing P1 by an arbitary (variable) smooth connected curve C. ∞ by any two points a. d = dim X. and 0. rat alg We write CHr (X) = zr (X)/zr (X) rat A (X) = r (the Chow group). Ω1 )∗/H1 (X. writing Γ(X. zr (X) ⊂ zr (X) is the subgroup of cycles of this rat form. pr1 : X × P1 → X. where J(X) is the jacobian of X. The cycle pr1∗ Γ · (X × ((0) − (∞))) ∈ zr (X). Z) and all ω ∈ Γ(X. Ar (X) = Ad−r (X). b ∈ C one obtains a group zr (X) with alg zr (X) ⊂ zr (X) ⊂ zr (X). dim Y = e. By convention this degree is zero if the extension is not ﬁnite. To extend these results. where dim X = d. Ω1 ). there is deﬁned a product cycle A · B ∈ zr+s (X) obtained by summing over all components of Ai ∩ B j with suitable multiplicities. Ω1 )∗ = HomC (Γ(X. is then deﬁned. Given A = mi Ai ∈ zr (X) and B = n j B j ∈ z s (X) such that every component of Ai ∩ B j has codimension r + s for all i and j. By deﬁnition.
The second and third lectures consider curves on threefolds. J r (X) = J0 (X) = Γ(X. in much the same way that abelian varieties do for weight 1. and in the ﬁrst lecture we study its kernel T (X) when dim X = 2. A bit more diﬃcult is to show CH1 (X) divisors = divisor class group of X divisors of functions Pic(X). rat alg where Pic(X) is the group of isomorphism classes of line bundles on X. i. i. In Lecture 2 we focus on quartic threefolds. the association ni (pi ) → ni deﬁnes an isomorphism CH0 (X)/A0 (X) Z. Lieberman. where C is a curve. zr (X)) are said to be rationally (resp. This conjecture can be formulated in various X ways. The map Θ : A0 (X) → J0 (X) is surjective.e. In keeping with the author’s philosophy that good mathematics “opens out” and involves several branches of the subject. Still harder is to prove that when X has a rational point. We work with varieties X over the complex numbers. the reader might prove that when the ground ﬁeld k is algebraically closed. and deﬁne (following Griﬃths.e. studying various aspects of the Abel–Jacobi construction in codimensions > 1. algebraic and arithmetic problems. Lecture 3 considers relative algebraic 1cycles on X = C × P1 × P1 . As an exercise. We give several examples motivating the conjecture that Γ(X. and verify in that case Θ is an isomorphism. we will consider geometric. Ω1 )∗ /H1 (X. Z) is called the AlX banese variety and the Abel–Jacobi map is deﬁned precisely as with curves. One vague but exciting possibility is that groups like T (X) provide a geometric interpretation of the category of polarized Hodge structures of weight two. when X has a nonzero 0 (in fact T (X) is enormous) when Pg (X) global holomorphic 2form. or if no ambiguity is possible just x ∼ 0). smooth hypersurfaces X of degree 4 in P4 . When r = d = dim X. We deﬁne a relative intermediate jacobian which turns out to be a noncompact torus isomorphic . Ω2 ) actually controls the structure of T (X) and that in particular X T (X) = 0 ⇐⇒ Γ(X. The ﬁrst three lectures are geometric in content. algebraically) rat alg equivalent to zero (written x ∼ 0 or x ∼ 0. particularly those of codimension > 1 as it is here that really new and unexpected phenomena occur. and Weil) compact complex tori (intermediate jacobians) J r (X) and Abel–Jacobi maps Θ : Ar (X) → J r (X). Ω2 ) = 0. We sketch an argument of Mumford that T (X) 0. A1 (X) is isomorphic to the group of kpoints of the Picard variety of X. The purpose of these notes is to study algebraic cycles.Introduction 5 Cycles in zr (X) (resp.
the Gersten–Quillen resolution. We consider the case C = E = elliptic curve and compute explicitly the class . Using a technique of Manin involving the Brauer group we show by example that 0. where X is a surface over a local or global ﬁeld k. It would be of great interest both algebraically and geometrically to know if the whole cohomology ring H ∗ (F. H p ). The last three lectures are devoted to arithmetic questions. Z/ Z) ⊗ · · · ⊗ H 1 (F.6 Lecture 0 to H 1 (C. In Lecture 7 we consider A0 (X). from a numbertheoretic point of view. the work of Lecture 3 on relative intermediate jacobians for curves on C × P1 × P1 . The heart of the proof is a result about the multiplicative structure of the Galois cohomology ring of a function ﬁeld F of transcendence degree d over an algebraically closed ﬁeld. Lectures 8 and 9 take up. C∗ ). Z/ Z) were generated by H 1 . Of particular interest are the cohomological formulae CH p (X) and (for X deﬁned over C) CH p (X)/A p (X) H p (X. H p (X. Finally. Z) (singular cohomology). maps isomorphically to the torsion subgroup of the Albanese variety. We assume the base extension of X to the algebraic closure of k is a rational surface. Kq ). Z/ Z) d times is surjective. R) deﬁned by factoring out by the maximal compact. and ´ analogues for singular and etale cohomology theories. We prove for prime to char F that the cup product map H 1 (F. K p ) where H p denotes the sheaf for the Zariski topology associated to the presheaf U ⊂ X → H p (U. These techniques are used to prove a theorem of Roitman that A0 (X)tors . We show such classes can be deﬁned for any (not necessarily relative) cycle on C × P1 × P1 . describing the image of the reduced norm map Nrd : A∗ → k(t)∗ when A is a quaternion algebra deﬁned over a rational ﬁeld in one variable over k. Z/ Z) → H d (F. We prove that A0 (X) is ﬁnite when X has a pencil of in general A0 (X) genuszero curves (conic bundle surface). Of particular importance are the classes in H 1 (C. The key idea in the proof is a sort of generalization of the Eichler norm theorem. The machinery of cycle classes constructed in this lecture is related in Lectures 8 and 9 to special values of Hasse–Weil zeta functions. the torsion subgroup of the zerocycles. Lectures 4 through 6 develop the algebraic side of the theory: the cohomology groups of the Ksheaves H p (X.
except in certain obvious cases such as ruled surfaces. Griﬃths’ proof that homological equivalence algebraic equivalence. To remedy this. K2 ) equals the order of the zero of the Hasse–Weil zeta function of E at s = 0.Introduction in H 1 (E. K2 ) a relative algebraic 1cycle on E×P1 ×P1 . R) C of the associated relative algebraic cycle multiplied by a certain simple constant (involving the conductor of the curve and Gauss sum) is the value of the Hasse–Weil zeta function of E at s = 2.g. In retrospect. Quillen’s work on the foundations of Ktheory are magniﬁcently presented in his own paper [10]. and my auditors at Duke for their enthusiasm and tenacity in attending eight lectures in ten days. Conjecturally for E deﬁned over a number ﬁeld k. a pedagogical note. I also want to acknowledge that much of the function theory of the dilogarithm which underlines the calculations in the last two lectures was worked out in collaboration with David Wigner. we construct an element U ∈ Γ(EQ . we show how to associate to f and g an element in Γ(E. g(x))  x ∈ E 7 for f. I only hope the brief outline given here will motivate the reader to turn to that source. I have certainly “left undone those things which I ought to have done” (e. g rational functions on E. the rank of Γ(Ek . For example. not have gotten them correct even with help. the structure of zerocycles on a surface was total chaos. I want to thank Duke University for ﬁnancial support. Historically. K2 ) such that the image under the Abel–Jacobi map into H 1 (E. f (x). indeed. When E has complex multiplication by the ring of integers in an imaginary quadratic ﬁeld of class number one.) Finally. The reader may also ﬁnd the rapid treatment of the Mumford argument showing Pg 0 ⇒ T (X) 0 in Lecture 1 to be unsatisfactory. and that I never would have gotten the damned constants in (9. R) associated to the curve γ f. I wanted to devote time to various examples showing that this is not the case. I have been very sketchy. the negative force of this result led us all to conclude that. I hope the reader will spare me. another demonstration rather diﬀerent in spirit from Mumford’s is included as an appendix. When the zeros and poles of f are points of ﬁnite order. and Tate’s proof of the Tate conjecture for abelian varieties). K2 ) and how to associate to an element in Γ(E. (I may. . When an idea is already well documented in the literature and extensive detail would carry us away from the focus of these notes on algebraic cycles.g = (x.12) correct without help from Dick Gross.
] [6] P. Grothendieck et al. [3] A. Paris e ee e (1971). MacPherson. R. Providence. no. 64. Oslo. Berlin (1973). On equivalence classes of cycles in an algebraic variety... [8] A. Lecture Notes in Math. [9] A.S. Hautes Etudes Sci. 11. 225. Quillen. .M. Alg` bre locale. 179–197 in Real and complex singularities (Proc. A.8 Lecture 0 References for Lecture 0 A list of references appears at the end of each lecture. Higher algebraic Ktheory. Weil. Springer.. Chee valley. [Pp. 2e ann´ e.I.. [5] W. S´ cr. Sijthoﬀ and Noordhoﬀ. Anneaux de Chow et applications. Th´ orie des intersections et th´ or` me de Riemann– e e e Roch (SGA 6). 341.P. Aarhus Universitet Preprint Series. no. e e [2] W. e e Springer. S´ minaire C. ´ [4] W. Math. Rational equivalence on singular varieties. no. Weil. Springer. What follows are references for foundational questions about cycles. 1976). Multiplicit´ s. Intersecting cycles on an algebraic variety. Chow. American Mathematical Society.. Ninth Nordic Summer School/NAVF Sympos. vol. 29. [7] J. Paris (1958). 85–147 in Algebraic KTheory I. Fulton. Foundations of Algebraic Geometry. Math. no. Math. pp. Fulton and R. 450–479 (1956). 147–167. Berlin (1965). etc. 14 (1976). Serre. Rational equivalence of arbitrary cycles. [1] C. Lecture Notes in Math. of Math. Hermann. 383–400. [10] D. Berlin (1971). (1946). Colloquium Publications. Alphen aan den Rijn (1977). Inst.. (2). Amer. Samuel. Math. Ann. Chevalley et al. Courbes Alg´ briques et Vari´ t´ s Ab´ liennes. Lecture Notes in Math. no. L.. J. 45 (1975). I.. 78 (1956). Publ.
If C is a 2chain.0 ) and we will write q = dim Γ(X. X If γ is a topological 1chain on X. algebraic surface X will mean a smooth projective variety of dimension 2 over the complex numbers (or. I will consider a number of explicit examples.0 that the quotient Γ(X. ﬁnally. Z) → Γ(X. γ . ΩiX ) (or H i. . In particular A0 (X) is not an [4] that Pg abelian variety in this case. This torus admits a polarization satisfying the Riemann bilinear relations and hence is an abelian variety. 2) will be written Γ(X. Hence X we may map H1 (X. if you prefer.1 Zerocycles on surfaces In this ﬁrst section I want to consider the question of zerocycles on an algebraic surface from a purely geometric point of view. Ω1 )∗ . C) = H 0. Z) is X a compact complex torus. Ω1 ). The space of global holomorphic iforms (i = 1. X Pg = dim Γ(X. Ω1 )∗ . X ∂C ω = C dω (Stokes’) so ∂C = 0 in Γ(X.1 ⊕ H 1. and give a heuristic description of a result of Mumford 0 implies A0 (X) is “very large”. called the Albanese of X and written Alb(X). since global holomorphic forms are closed. Z). 0 Note that this is well deﬁned: two paths from p0 to p diﬀer by an element in H1 (X. Ω1 )∗ /H1 (X. we deﬁne φ : X → Alb(X) by φ(p) = p . Notice. Finally I will discuss some conjectures motivated by these ideas. It follows from the Hodgetheoretic deX composition H 1 (X. that the above discussion is equally valid for smooth projective varieties X of arbitrary dimension. Ω1 )∗ . the integral over γ. is a welldeﬁned element in the dual Cvector space Γ(X. a compact complex manifold of dimension 2 admitting a projective embedding). Ω2 ). p Fixing a base point p0 ∈ X. For purposes of this lecture.
e) = (− f.3) Let Y be a smooth quasiprojective variety. . (Hint: In (i) consider the question inﬁnitesimally and use the fundamental theorem of calculus to calcup late the derivative of p ω. and taking σ( f. Let E = E/{1. n > 0 an integer.σ} F×E Γ(E. [2]) Let E and F be elliptic curves (Riemann surfaces of genus 1). Then An (X) is a divisible group. Ω1 ) E F Γ(E. We conclude Alb(X) F has dimension 1. η}. φn is surjective. Ω1 ){1. Lemma (1. e + η). and assume ψ(p0 ) = 0. (ii) Let ψ : X → A be a map from X to a complex torus. reduce the question to showing that a map 0 P1 → complex torus is necessarily constant. where σ is a ﬁxedpointfree involution on F × E obtained by ﬁxing a point η ∈ E of order 2. η 0. For (iii). We propose to calculate the Chow group of the quotient surface X = (F × E)/{1. Proof These results are more or less well known. φ (iii) The map z0 (X) → Alb(X). where (x) → φ(x).2) (Bloch et al.σ} since the automorphism f → − f acts by −1 on Γ(F. it follows that Alb(X) E . Ω1 ) ⊕ Γ(F. Since Alb(X) → E and the ﬁbres of ρ are connected.10 Proposition (1. . . Notice Γ(X. . xn ) = φ(x1 ) + · · · + φ(xn ). factors through the map φ : CH0 (X) → Alb(X). The reader who is unfamiliar with them might try as an exercise to ﬁnd proofs. Then for n 0. Then there exists a unique homomorphism θ : Alb(X) → A such that the diagram X ψ φ n times Alb(X) θ A commutes. .) Example (1. There is a natural map ρ : X → E with all ﬁbres of ρ F. and independent of the choice of base point p0 .1) Lecture 1 (i) Let φn : X × · · · × X → Alb(X) be deﬁned by φn (x1 . Ω1 ) X Γ(F × E. Ω1 ) E C {1. Ω1 ). The induced map φ : A0 (X) → Alb(X) is surjective. σ}.
We work with the diagram F×E π E X If z ∈ T (X) we may write π∗ z = where ri = 0 and E. We now return to our surface X = F × E/{1. so the question becomes the surjectivity of H1 (W. p + η) (q. T (X) = 0. Z/NZ) → H1 (Y. It will suﬃce therefore to assume dim Y > 1 and show N Alb(W ) N Alb(Y) for W ⊂ Y a smooth hyperplane section. pi + η) . let N A ⊂ A be the kernel of multiplication by N. Then T (Y ) is divisible. By Abel’s theorem 2 (q. p) ∼ 2 (0. Z/NZ) or the vanishing of H1 (Y. pi ) + (−qi . σ}. W. p). the lemma follows. Z/NZ).Zerocycles on surfaces 11 Proof By deﬁnition An (X) ⊂ CHn (X) is generated by images under correspondences from jacobians of curves. pi )] ∼ ∼0. Claim Proof A0 (X) E Alb(X). As a real torus. Z/NZ) = (0) because the Stein variety Y − W has cohomological dimension = dim Y. so 2 π∗ z ∼ 2 ri [(qi . Z/NZ). Since jacobians of curves are divisible groups. and let T (Y ) = Ker (A0 (Y ) → Alb(Y )). the Chow group and Albanese both coincide with the jacobian so N A0 (Y ) N Alb(Y ).4) Let Y be a smooth projective variety. From the divisibility of A0 (Y ) one reduces to showing N A0 (Y ) N Alb(Y ) for any N. That is. pi ) + (−qi . Lemma (1. This group is identiﬁed by duality with H 2dim Y−1 (Y − W. p) + (−q. R/Z) and N Alb(Y) = H1 (Y. ri (qi . 4 ri (0. Proof For any abelian group A. p) ∼ 2 (q. on F × E. If Y is a curve. 2 ri pi = 0 in E. Alb(Y) can be identiﬁed with H1 (Y. pi ) 4 ri (pi ) = [(0) × E] · F × .
5) (Inose and Mizukami [3]) Let Y : T 0 +T 1 +T 2 +T 3 = 0 in P3 . it suﬃces to show π∗ π∗ : A0 (Y ) → A0 (Y ) is the zero map.1) ¯ ¯2 ¯3 e ¯2 ¯3 e ¯2 ¯3 π∗ π∗ = 1 + e1 e2 e3 + (¯ 1 e2 e3 )2 + · · · + (¯ 1 e2 e3 )4 .5. i = 1. Claim A0 (X) = (0). i j. ω3 . . One checks immediately ¯ (1. Since π∗ is surjective. 3 i i ei e j . the lefthand identities in (1. . so X is smooth. so it will suﬃce to show that ¯ π∗ π∗ goes to zero under any homomorphism R ⊗ Q → Q = algebraic closure of ¯ amounts to the choice of three nonQ.1) is trivial in R⊗Q. leading to 4 0= n=0 4 en en en . . ω3 t3 ). and let Z/5Z⊕3 act on Y by e∗ T j = i Tj ωT i j i. i = j. where we identify Z/5Z with the group of ﬁfth roots of 1 and we let a ﬁfth root ω act on Y by (t0 . ωt1 . . and let X = Y/(Z/5Z). We ﬁx a ﬁfth root of 1. . . A homomorphism h : F ⊗ F ⊗ F → Q trivial ﬁfth roots of 1. ω 1. and hence 4T (X) = 0. . t2 . On the other hand. . R ⊂ End(A0 (Y )). t1 . .5. ω2 t2 . Writing F = Q(ω).5. which we This gives us a representation Z [Z/5Z⊕3 ] denote by x → x. Since T (X) is divisible. e5 . For h to factor through R ⊗ Q the righthand identities in (1.2) force ωi ω j 1. i i j i j is rational. Proof Let π : Y → X be the projection. The reader can verify also that the quotient of Y by any of the cyclic groups e1 e2 e3 .2) show that R ⊗ Q is a quotient of the semisimple ring F⊗Q F⊗Q F. ¯1 ¯2 ¯3 4 (1. e5 e5 . e2 e2 . 2. . 5 5 5 5 Example (1.2) n=0 e−n = 0 = i en en ¯i ¯ j n=0 in R ⊂ End (A0 (Y )). 3. 2. this implies T (X) = 0. ω2 .5. Since A0 (Y ) is divisible. . t3 ) → (t0 . for the image of . Since A0 (X) is divisible. ω1 . e2 . . This action is ﬁxed point free.12 Lecture 1 It follows that 0 ∼ 2π∗ π∗ z = 4z. i = 1.5. e5 e5 e5 1 2 3 or or j ei . it will suﬃce to show π∗ π∗ (1. it suﬃces to show 5A0 (X) = π∗ π∗ A0 (X) = (0).
1. see the appendix to this lecture.) Quite precisely. . The form ωn is invariant under the ˜ n ˜ action of the symmetric group S n on X and so in a certain sense ωn descends to a form ωn on S n X = X n /S n . There will be an open set T 0 ⊂ T and a morphism T 0 → (S n X)smooth . For another argument. ρn : S n X → A0 (X). Remarks The surface in (1.T ∈ Γ(T. and ωn. The reader should be warned that the author has exercised his prerogative in selecting these examples very carefully. (This part of the argument requires considerable care because S n X will have singularities. where ρn (x1 . ω1 ω2 ω3 = 1. One checks easily that the 2 3 conditions ω 1 ω2 ω3 ωi 1. Notice if T (X) = 0. I want now to sketch Mumford’s idea [4] for showing that Pg > 0 implies T (X) is enormous. Then the codimension of W in S n X is ≥ n. Ω2 ) deﬁned. Note that the singularities of S n X occur at T points (x1 . xn ) = (xi ) − n(p0 ). ωi ω j 1. one gets a 2form ωn = pr∗ ω + · · · + pr∗ ω ˜ n 1 on the cartesian product X n = X × · · · × X. . Mumford proves Theorem (1. ωn is well deﬁned.2) is called hyperelliptic. . 2. the ﬁbres of ρn would be subschemes of S n X of codimension ≤ q = dim Alb(X). By pulling back along the various projections and adding up. 2 3 i j = 1. so A0 (X) Alb(X). 3 cannot all hold. Assume t ∈ W ⊂ S n X is a subscheme on which ρn is constant. that is. . . In particular. . so π∗ π∗ = 0. Using the deﬁnition of rational equivalence and the fact that there are no global holomorphic forms on projective space. ρn (W ) = ρn (t). .T will be a holomorphic extension of ψ∗ ωn . one must have ω1 ω2 ω3 = 1. This proves the claim.6) Assume Pg > 0 and let t ∈ S n X be a general point.Zerocycles on surfaces 13 π∗ π∗ to be nontrivial. if T is a nonsingular variety parameterizing an eﬀective family of zerocycles of degree n a general element of which consists of n distinct points with multiplicity 1.5) is a Godeaux surface. both examples have Pg = 0. that in (1. Idea of proof Let ω be a nonzero global holomorphic 2form on X. Mumford shows that if the ψ . Choose a base point p0 ∈ X and deﬁne a map ρn from the nth symmetric product S n X = set of unordered ntuples of elements of X to A0 (X). there will be a “pullback” ωn. . xn ) where two or more of the xi coincide. On the complement (S n X)smooth of the singular set.
7.T = 0. Pic0 (S ) is the group of closed points of the Picard variety of S . then ωn.) Claim The intersection maps (1. s ∈ S be general points. a smooth hypersurface of degree 3 in P4 . Z) generated e by classes of divisors. (If D0 is algebraically equivalent to zero and D is any divisor. It will suﬃce (using divisibility of A0 (S )) to show that 2(r) − 2(s) ∈ Image(Pic0 (S )⊗Z Pic(S ) →A0 (S )). and let S ⊂ G be the subvariety of lines on T . It follows that dim W ≤ 1 dim S n X = n. D · D0 will have degree 0. The line on T corresponding to s ∈ S will be denoted s . if necessary we may assume t is a nonsingular point of W. One next notices that if t ∈ (S n X)smooth is general. and there is an exact sequence 0 → Pic0 (S ) → Pic(S ) → NS(S ) → 0. and Pic0 (S ) = divisors algebraically equivalent to zero modulo rational equivalence.7. Let r. The minor modiﬁcation needed to prove (1. the twoform ωn will give a nondegenerate alternating pairing on the tangent space at t. where NS(S ) = N´ ron–Severi group of S = subgroup of H 2 (S .1) and (1. Pic0 (S )⊗Z Pic0 (S ) → T (S ). but does show that our nondegenerate pairing. Shifting t along W. The subvariety S is known to be a smooth connected surface. Let G be the Grassmann of lines in P4 . so 2 codim W ≥ n. Necessary facts about cubic 3folds can be found in Clemens and Griﬃths [11] or Tyurin [12].7) (Fatemi [10]) A surface with Pg > 0 and an interesting Chow group is the Fano surface. Intersection of divisors gives bilinear maps (1.14 Lecture 1 cycles in the family parameterized by T are all rationally equivalent. that is. .2) are surjective.7. Let T be a cubic threefold. our given subvariety.1) (1. Proof I will give the proof only for (1.7. hence trivial (exercise!). The fact that D · D0 ∈ T (S ) if both D and D0 are algebraically equivalent to zero can be checked by noting that the map Pic0 (S ) × Pic0 (S ) intersect A0 (S ) → Alb(S ) is a continuous bilinear map of compact tori. We write Pic(S ) = CH1 (S ).2) is left for the reader. Example (1. when restricted to the tangent space of W at t.1). Resolving singularities of W then does not change the situation near t. t s. and D s will denote the divisor on S D s = Zariski closure of t ∈ S  t ∩ s ∅.7.7.2) Pic(S )⊗Z Pic0 (S ) → A0 (S ). gives zero.
” in this context. if r. s are general (“general. and D s · Dt has degree 5. s D s . Dr · Dt = (s) + (s1 ) + · · · + (s4 ). On the other hand Dr · D s = (t) + (t1 ) + · · · + (t4 ). l t1 c lt2 lt3 ls lt b lr a l t4 Through a general point of T there pass 6 lines. Since 2((r) − (s)) = 2((r) − (t)) + 2((t) − (s)). θ(b) − D s · Dt = (s) + (t) − (r). Thus. Let L be the plane spanned by r and s in P4 and let t be the third line in L ∩ T = r ∪ s ∪ t. .Zerocycles on surfaces 15 In general. CH0 (T ) Z (any two points on T can be connected by a chain of lines) so we have θ(a) = (r) + (t) + (s1 ) + · · · + (s4 ) ∼ θ(b) = (s) + (t) + (r1 ) + · · · + (r4 ) ∼ θ(c) = (r) + (s) + (t1 ) + · · · + (t4 ). D s · Dt = (r) + (r1 ) + · · · + (r4 ). Hence θ(a) − Dr · Dt = (r) + (t) − (s). and we denote by θ : CH0 (T ) → CH0 (S ) the correspondence thus deﬁned. means the points in question do not lie on a ﬁnite number of a priori speciﬁed proper closed subvarieties of S ) we can ﬁnd a general t such that s ∩ t ∅ and t ∩ r ∅. we may assume r ∩ s ∅. (D s − Dr ) · Dt ∼ θ(a) − Dr · Dt − θ(b) + D s · Dt = 2(r) − 2(s).
Z ⊕ Alb(X) ⊕ T (X). H 2 (S . space H 1 (S .16 Lecture 1 Philosophy The dual to Γ(S . F 2 CH0 (V ) = Ker (F 1 CH0 (V ) → Alb(V )). we get also gr Γ∗ : gr CH0 (V ) → gr CH0 (X) Conjecture (1.8) [Γ] ∈ H 4 (V × X). I want to end with a conjecture. one knows . Because X is a surface. The tangent space to Pic0 (S ) is H 1 (S . The above ﬁltration being functorial for correspondences. space tang. Let Γ(i. the control exercised by the geometric genus Pg over the size of T (S ) can be seen in the following schema: Pic0 (S ) ⊗ Pic0 (S ) tang. OS ) H 2 (S . Ω2 ) is the second cohomology group of the S structure sheaf. OS ) . motivated in part by these examples. Except in certain simple cases like abelian varieties or products of curves (Bloch [13]). the cup product map induces an isomorphism Λ2 H 1 (S . We will therefore “truncate” at F 2 . OS ). and let Γ be a cycle on V × X with dim Γ = n = dim V. we may as well work with rational cohomology). j) ∈ H i (V ) ⊗ H j (X) denote the K¨ nneth components of Γ (since the graded pieces of CH0 are either divisible or torsion free. The cycle Γ then induces a map Γ∗ : CH0 (V ) → CH0 (X) . The map gr Γ∗ depends only upon the cohomology class I want to discuss some consequences of this conjecture. OS ) cup product H 2 (S . space intersection T (S ) bitang. For any smooth projective variety V. The idea of a bitangent space to T (S ) will be discussed further in Lecture 6. and to try to make it u more precise. OS ). In the case of the Fano surface. I don’t have much feeling for what should come beyond F 2 . OS ) (Serre duality). Presumably the ﬁltration will in general have n steps where n = dim V. Let X be a surface. OS ) ⊗ H 1 (S . In some mystical sense. we may deﬁne at least the beginnings of a ﬁltration on CH0 (V ) by F 1 CH0 (V ) = A0 (V ) = Ker (deg : CH0 (V ) → Z).
4) = Γ(4. Γ(3. Proposition (1. 1)∗ : gr CH0 (V ) → gr CH0 (X) is zero. To study Γ(1. Lemma (1. and let X be a surface of geometric genus 0.8). 3). the interesting correspondences to study therefore are those of K¨ nneth type (2. 3)∗ is zero on F 2 CH0 (V ).Zerocycles on surfaces 17 (cf.10) There is an equivalence of categories between a suitable category of polarized Hodge structures of weight 2 and a category built up from the F 2 CH0 (X). 0) = 0. we may suppose Γ(0. this implies Γ(1. The map Γ(1. The map Jacobian(C) → Alb(X) is split up to isogeny (Poincar´ ree ducibility). 1)∗ via an algebraic correspondence Σ ∈ H 3 (V ) ⊗ H 2q−1 (Pic0 (X)). 2A10 and 2. The inverse of this isomorphism will also be induced by an algebraic correspondence (Kleiman [15]. Altering Γ by constant correspondences. Looking at an example like (1. so on the cycle level Σ∗ CH0 (V ) = (0). Kleiman [15].11) Assume conjecture (1.11). 3. 3)∗ F 2 CH0 (V ) = (0). Such a Σ is represented by an algebraic cycle on V × Pic0 (X) of dimension dim V − 1.9) Assuming conjecture (1. 3)∗ let j : C → X be a smooth hyperplane section. Since F 2 CH0 (C) = (0).7) u one might be tempted to formulate the fantastically vague: Metaconjecture (1. Then F 2 CH0 (X) = (0). Such an equivalence might provide a geometric interpretation of the category of Hodge structures of weight 2 analogous to the abelian variety interpretation for weight 1! I want ﬁnally to record two other consequences of (1.1) on e u X × Pic0 (X) induces via correspondences an isomorphism D∗ : H 2q−1 (Pic0 (X)) H 1 (X).8) holds. so we can factor Γ(3. 2). j) are algebraic. Proof Let q = dim Alb(X). A Poincar´ divisor D of K¨ nneth type (1. From the point of view of cycles.9) that the Γ(i. .8). Since H 3 (X) H 2q−1 (Alb(X)) one shows that there exists an algebraic correspondence ψ ∈ H 1 (V ) ⊗ H 1 (C) such that (1 × j∗ ) ψ = Γ(1.
e e as well as the references cited there. Compositio Math. Zero cycles on surfaces with Pg = 0. Ast´ risque. no. 2)∗ : CH0 (X) → CH0 (X) is zero. 3. 2) = ai j Di × D j where the Di are divisors. Mizukami.5) is due to H. and D. Ann.2 (−1) j+k (vi j . Let Γ be a codimension2 cycle on V ×X and let vi j ∈ Vi be points. 2.2 (−1) j+k+ (vi j . Beauville. =1. Lieberman. a good reference is [1] A. Let X be a surface and V = V1 × V2 × V3 .k. But a zerocycle can be moved away from a ﬁnite union of divisors. with all varieties smooth and connected. Rational equivalence of 0cycles on some surfaces of general type with pg = 0. 205–217. The Γt are clearly all homologous. Then Γ∗ j. Proposition (1. Math. Example (1. Then ∆∗ = ∆(2.. For further discussion of this sort of conjecture. 3 and j = 1.2) is done more generally in [2] S. v2k . Example (1. so Γv31 ∗ = Γv32 ∗ : F 2 CH0 (V1 × V2 ) → F 2 CH0 (X).18 Lecture 1 Proof Let ∆ ⊂ X × X be the diagonal.8) holds. v3 ) = 0 in CH0 (X). v2k ) is an element of F 2 CH0 (V1 × V2 ) (exercise!). A. 2. Surfaces alg´ briques complexes. Details and other analogous examples are in [3] H. so ∆(2. Mizukami. 244 (1979). Bloch. the reader can see Bloch [14]. along with other surfaces with Pg = 0 and Kodaira dimension < 2. Proof View Γ as a family {Γt }t∈V3 of cycles on V1 × V2 × X parameterized by V3 . But the cycle j..12) (Conjectural theorem of the hypersquare) Assume conjecture (1. 33 (1976). i = 1. Kas. 2)∗ = Id on F 2 CH0 (X). 59 (1978). Since X has geometric genus 0. so ∆(2. Inose and M. References for Lecture 1 For general facts about algebraic surfaces over C. Inose and M. The desired identity is now straightforward. 135–145. . H 2 (X) is generated by divisors.k=1.
USSRSb. no. Ann. 15 (1971). [10] T.S.] [7] A. 571–588. Five lectures on threedimensional varieties (in Russian). 195–204. Amer. 86 (128) (1971).. Algebra. For a time it was believed that Mumford’s technique could be applied to pluricanonical diﬀerentials. On the symmetric product of a rational surface. Math.. 569–585. the reader should see the seminal paper of Mumford: [4] D. 95 (1972). 776–779. Ruled surfaces and the Albanese mapping. Γequivalence of zerodimensional cycles (in Russian). Kyoto Univ. A. Uspehi Mat. Sb. 671. [8] A.7) was worked out by Fatemi. H. Sb. 18 (1974). Universit´ de Paris VII (1979). Example (1. Griﬃths. K2 of Artinian Qalgebras with application to algebraic cycles. 557–570. 281–356. 555–567. 405–428.. [Translation: Russian Math. [12] A. [Translation: Math. Mat. 75 (1969). Roitman. Clemens and P. The intermediate Jacobian of the cubic threefold. Amer. L’equivalence rationelle des z´ ro cycles sur les surfaces al´ ee g e e briques complexes a cup product surjectif. These du 3 cycle. and that one could in this way show A0 (X) was not representable for any nonruled surface. [Translation: Math USSRSb. 5. A number of such statements appeared in the literature.Zerocycles on surfaces 19 To understand the role of the geometric genus. Fatemi. no. 89 (131) (1972). J. (167) 3–50.). 21 (1969).] [6] A. Soc.). Math. Surveys. 5. Bull. 27 (1972). The argument fails because the trace of a pluricanonical form under a ﬁnite ramiﬁed map can acquire singularities. Roitman. 27 (1972).. Mattuck. as well as [5] A.S. (N. A. Soc.] . Rational equivalence of zerocycles on surfaces. of Math. Mattuck. A. (2). Mat.. N. see [9] S. Mumford. 683–688. Nauk. Comm. Tyurin. see [11] C. Proc. Rational equivalence of zerodimensional cycles (in Russian). For more details about the geometry of the Fano surface and the cubic threefold. 1–53. Bloch. For a more positive point of view. 9 (1968). Math. (N. 3 (1975).
20 Lecture 1 Some references for the conjectures at the end of Lecture 1 are: [13] S. Some elementary theorems about algebraic cycles on abelian varieties. pp. Bloch. . Amsterdam e e (1968). Bloch. Kleiman. Invent. 215–228. An example in the theory of algebraic cycles. no. 37 (1976). 1–29 in Algebraic KTheory. pp. Springer. [15] S. Math. North Holland. 359–386 in Dix expos´ s sur la cohomologie des sch´ mas. 551. Lecture Notes in Math. [14] S.. Berlin (1976). Algebraic cycles and the Weil conjectures..
Appendix: On an argument of Mumford in the theory of algebraic cycles
S. Bloch
1
Let X be a smooth projective surface over an algebraically closed ﬁeld k. Let CH0 (X) denote the Chow group of zero cycles modulo rational equivalence on X, and let A0 (X) ⊂ CH0 (X) be the subgroup of cycles of degree 0. I will say that A0 (X) is ﬁnite dimensional if there exists a complete smooth (but possibly disconnected) curve C mapping to X such that the map J(C) = Jacobian(C) → A0 (X) is surjective. Some years ago, Mumford proved, in the case k = C, that Pg (X) > 0 implies A0 (X) is not ﬁnite dimensional. The purpose of the present note is to prove an analog of this result applicable in all characteristics. The role of the geometric genus, which is not a good invariant in characteristic p, 2 is played by the “transcendental part” of Het (X, Q ). The present proof also reveals the inﬂuence of the ﬁnite dimensionality of the Chow hypothesis on the structure of the “motive” of X. The idea that one could deduce interesting information about the Chow group by considering the generic zerocycle was suggested by ColliotTh´ l` ne. ee I am indebted to him for letting me steal it. Lemma (1A.1) Let X be a smooth variety over an algebraically closed ﬁeld k, and Y any kvariety. Let n ≥ 0 be an integer. Then, writing K = k(Y ), CHn (XK ) lim CHn (X ×k U), − − →
U⊂Y open
where CHn equals codimensionn cycles modulo rational equivalence. Proof
1
For any variety W and any integer m, let W m = set of points of codi
The author gratefully acknowledges support from the NSF and from the conference on Chow groups of rational varieties. The present note grew out of conversations with ColliotTh´ l` ne ee at the conference.
22 mension m on W. One has
Lecture 1
(XK )m = lim(X ×k U)m . ← − −
U
Since CHn (XK ) = Coker CHn (X × U) = Coker the desired result is immediate. Proposition (1A.2) Let X be a smooth projective surface over k and let Ω ⊃ k be a universal domain in the sense of Weil. Assume A0 (XΩ ) is ﬁnite dimensional. Then there exist onedimensional subschemes C , C ⊂ X and a 2cycle Γ supported on (C × X) ∪ (X × C ) such that some nonzero multiple of the diagonal ∆ on X ×k X is rationally equivalent to Γ. Proof Let C → X be such that J(CΩ ) A0 (XΩ ), and let C ⊂ X be the image of C. Enlarging k, we may assume C deﬁned over k. Lemma (1A.3) Let k ⊂ K ⊂ K be extensions of ﬁelds. Then the kernel of CH2 (XK ) → CH2 (XK ) is torsion. Proof If [K : K] < ∞ this follows from the existence of a norm CH2 (XK ) → CH2 (XK ). The case K algebraic over K follows by a limit argument. Enlarging K and K , we may thus assume K algebraically closed. In this case, CH2 (XK ) is a limit of Chow groups CH2 (X ×K U), where U is a Kvariety of ﬁnite type. A Kpoint of U gives a section of CH2 (X) → CH2 (X ×K U) so the lemma follows. Proof of proposition (1A.2) Let K be the function ﬁeld of X over k, and ﬁx an embedding K → Ω. Let P ∈ X(Ω) be the corresponding point. Our hypotheses imply CH2 ((X − C )Ω ) = (0), so by Lemma (1A.3), there exists N ≥ 1 such that N(P) = 0 in CH2 ((X − C )K ). (We abuse notation by writing P also for the generic point of X. In other words, P is the image in CH2 (XK ) of the diagonal ∆ in X × X.) By Lemma (1A.1), there exists U ⊂ X open ∅ such that N · ∆ is rationally equivalent to zero on (X − C )×K U. Let C ⊂ X be a subscheme of codimension 1 containing X − U. The proposition now follows from the exact K(x)∗ −→
x∈(XK )n−1 x∈(Xk )n
Z , Z ,
y∈(X×U)n
k(x)∗ −→
y∈(X×U)n−1
Zerocycles on surfaces sequence cycles supported on → CH2 (X ×k X) (C × X) ∪ (X × C ) → CH2 ((X − C )×k (X − C )) → 0.
23
Exercise (1A.4) Generalize the deﬁnition of ﬁnite dimensionality for A0 (X) to varieties X of dimension > 2 and prove the analogue of Proposition (1A.2). Fix now an prime to char p, and deﬁne
2 2 H 2 (X)trans = Het (X, Q )/Image(NS(X) ⊗ Q −→ Het (X, Q )),
where NS(X) is the N´ ron–Severi group. We systematically ignore twisting by e roots of 1. An alternative description of H 2 (X)trans is given by
2 2 ¯ H 2 (X)trans = Image(Het (X, Q ) → HGal (K/K, Q )). 2 Note Het (X), NS(X), and hence also H 2 (X)trans are functorial for correspondences, which we will take to be elements in CH2 (X ×k X). In particular, the diagonal induces the identity on H 2 (X)trans .
Lemma (1A.5) With notation as in Proposition (1A.2), let Γ be a codimension2 cycle on X × X supported on (C × X) ∪ (X × C ). Then the correspondence Γ∗ : H 2 (X)trans → H 2 (X)trans is zero. ´ Proof It is convenient to work with etale homology, which is deﬁned for any kscheme Y which can be embedded in a smooth kvariety. Write Γ = Γ + Γ with Supp Γ ⊂ C × X and Supp Γ ⊂ X × C . Let i : C × X → X × X, i : X × C → X × X, and write [Γ ] ∈ H4 (C × X, Q ), [Γ ] ∈ H4 (X × C , Q ). 2 For α ∈ Het (X, Q ), we have Γ∗ α = pr2∗ (pr∗ (α) · i∗ [Γ ]) = pr2∗ i∗ (i ∗ pr∗ (α) · Γ ) 1 1 = pr2∗ i∗ (pr∗ i ∗ (α) · Γ), 1 with morphisms labeled as indicated: C ×X
pr1 i
X×X
pr1
C
i
X.
2 Note that NS(X) ⊗ Q is selfdual under the cup product pairing on Het (X, Q ). Without changing the image of α in H 2 (X)trans we may assume, therefore, that α is perpendicular to NS(X). Since H2 (C , Q ) is generated by the classes of components of C , we ﬁnd i ∗ (α) = 0 in H 2 (C , Q ), so Γ∗ (α) = 0.
24
Lecture 1
It remains to show Γ∗ (α) = 0 where Γ is supported on X × C . We have a diagram X ×C
pr2 i
X×X
pr2
C and the projection formula gives Γ∗ α = pr2∗ (i∗ Γ · pr∗ α) 1 = i∗ pr2∗ (Γ · i ∗ pr∗ α) 1
i
X
∈ Image(H2 (C , Q ) → H 2 (X, Q )) ⊂ NS(X) ⊗ Q , so Γ∗ α → 0 in H 2 (X)trans . We have proven: Theorem (1A.6) Let X be a smooth projective surface over an algebraically 2 closed ﬁeld k, and let k ⊂ Ω be a universal domain. If Het (X, Q ) NS(X)⊗Q ( char k), then A0 (XΩ ) is not ﬁnite dimensional. Exercise (1A.7) dimension > 2. Formulate and prove an analogous result for varieties of
Question (1A.8) If E is a supersingular elliptic curve over a ﬁeld k of char2 acteristic 0, , then Het (E × E, Q ) = NS(E × E) ⊗ Q . Is A0 (E × E) ﬁnite dimensional?
2 When k = C, Het (X) Mumford’s result:
NS(X) ⊗ Q if and only if Pg (X) > 0, so we recover
Corollary (1A.9) When k = C, Pg (X) > 0 implies A0 (XC ) is not ﬁnite dimensional.
References for Lecture 1 Appendix
D. Mumford, Rational equivalence of 0cycles on surfaces, J. Math. Kyoto Univ., 9 (1968), 195–204. A. A. Roitman, Rational equivalence of zerodimensional cycles (in Russian), Mat. Sb. (N.S.), 89 (131) (1972), 569–585, 671. [Translation: Math. USSRSb., 18 (1974), 571–588.]
) The intermediate jacobians to which these cycles are related can be described as follows. H i. C). C). C) = F 0 H 2r−1 ⊃ F 1 ⊃ · · · ⊃ F 2r−1 ⊃ (0). This complex vector space also has a Zstructure deﬁned by the image of H 2r−1 (X. Let X be a smooth projective variety and r > 0 an integer. see Tennison [9]. H 2r−1 (X.2r−i−1 . it has an Rstructure. No one has worked out the details in the general case. C) has a Hodge ﬁltration (Deligne [5]) H 2r−1 (X. It seems likely the ﬁnal result would be the same. The key property of the Hodge ﬁltration is F i ⊕ F 2r−i F i ∩ F 2r−i−1 where H i. In particular. a smooth hypersurface X in P4 of degree 4 and will show in this case that A2 (X) is isomorphic to the intermediate jacobian J 2 (X). ΩiX ). This is false. Z) → H 2r−1 (X. in particular the link with intermediate jacobians. for the Fermat quartic. which would involve jacobians of singular curves. We will focus on the example of a quartic threefold. The complex cohomology H 2r−1 (X.2 Curves on threefolds and intermediate jacobians The purpose of this lecture is to give some feeling for the geometry of curves on threefolds over the complex numbers. j H j (X. as we shall suppose for technical simplicity that the curve of lines on X is smooth. . (In all honesty the proof will not be quite general. for example. so it makes sense to talk about conjugating an element or a subspace.
Z). Let X be a smooth projective threefold. C))∗ /H2n−2r+1 (X.) The ﬁrst point to note is that if ω is of type (3. C))∗ . F r ∩ Image(H 2r−1 (X. (This is an easy consequence of Hodge theory. If dim X = n. X We choose a 3chain ∆ on X with ∂∆ = γ and consider ∆ as an element in the dual space to the space of 3forms of types (3. C). The next step is to deﬁne cycle classes in these intermediate jacobians. C)/F r + H 2r−1 (X. Poincar´ duality implies e H 2r−1 (X. we see from this that dimC F r = 1 2 dimC H 2r−1 (X. It follows that J r (X) (F n−r+1 H 2n−2r+1 (X. Z). if r = n we ﬁnd J n (X) Γ(X. where ∗ denotes Clinear dual. (We are thinking of J 2 (X) = F 2 H 3 (X. C)∗ /H3 (X. then we can write ω = dη. and the idea will be to mimic the construction in Lecture 1 of X → Alb(X) by associating to (p) − (p0 ) the integral p p0 ∈ Γ(X. C)) = (0). Z) → F 2 H 3 (X. We assume γ is homologous to 0. Z) X Alb(X). where η has type (2. . 0) forms on a curve. Notice ﬁnally if ∆ is another 3chain with ∂∆ = ∂∆ = γ. that is [γ] = 0 in H 4 (X. called the intermediate jacobian. and let γ = ni γi be an algebraic 1cycle on X. see Griﬃths and Harris [7]. For example. Z) → H 2r−1 (X. 1). then ∆ − ∆ is a 3cycle so ∆ − ∆ ∈ Image (H3 (X. 0). C)/F r (F n−r+1 H 2n−2r+1 (X. In this lecture we restrict ourselves to a concrete intuitive description valid for the case at hand. Z). 0) + (2. Z) is a compact complex torus. 0) + (2. so the quotient J r (X) = H 2r−1 (X.) Stokes’ theorem gives ∆ ω= γ η=0 because there are no (2.26 Lecture 2 In particular. Ω1 )∗ /H1 (X. Z). 1) and ω is exact. C)∗ ). Ω1 )∗ / H1 (X.
Since a cycle algebraically equivalent to zero is certainly homologous to zero. Here J(F) is the jacobian of F and Σ∗ satisﬁes a certain quadratic relation Σ2 + (q − 2)Σ∗ − (q − 1) = 0. 27 Recall that A2 (X) is the group of codimension2 cycles algebraically equivalent to zero. t) ∈ F × F − ∆  s ∩ t ∅}. one can hope for a map Θ : A2 (X) → J 2 (X). given a family {Γt }t∈P1 of codimension2 cycles parameterized by P1 . Then Θ : A2 (X) → J 2 (X) is an isomorphism. ˜ Lemma (2. In order to avoid certain technical problems which have never been carefully considered. γ where γ is a codimension2 cycle representing γ ∈ A2 (X).3) If s ∈ F corresponds to the line s ⊂ X. This is the 4 4 4 4 4 case “in general” but not. The family of lines on X is known to form a connected curve F (Bloch and Murre [3]). so any such map is constant. Theorem (2. Θ(γ) = j(˜ ). using a beautiful idea of Tyurin [10] (note. however. where N = dim J 2 (X). But P1 has no nonconstant global holomorphic functions. Since P1 is simply connected. That is. we will assume that F is smooth. modulo rational equivalence. this lifts to a holomorphic map P1 → CN . Proof It is possible to show that t → j(Γt ) deﬁnes a holomorphic function P1 → J 2 (X). for X : T 0 + T 1 + T 2 + T 3 + T 4 = 0 (Tennison [9]). j(Γt ) ∈ J 2 (X) is independent of t. one can deﬁne an incidence correspondence Σ on F essentially by taking the closure of the set {(s.1) The above map Θ is well deﬁned. We want to study the map Θ when X ⊂ P4 is a smooth hypersurface of degree 4.Curves on threefolds and intermediate jacobians These remarks suﬃce to deﬁne a cycle class j(γ) ∈ J 2 (X). Remark (2. One can then identify J 2 (X). for example. the criticism of Tyurin’s argument in Bloch and Murre [3]) with generalized Prym PΣ (F) = Image (Σ∗ − 1 : J(F) → J(F)). The standard Prym would correspond to the case ∗ .2) Let X be a smooth quartic threefold and assume the curve of lines F is smooth.
so we are reduced to the case π : Y → X everywhere deﬁned.4) Let X be a smooth quartic threefold.6) Let X and Y be smooth projective threefolds. Consider the diagram (commutative for either π∗ or π∗ – see Lieberman [12]) A2 (Y) π∗ π∗ ΘX ΘY J 2 (Y) π∗ π∗ A2 (X) J 2 (X). As a ﬁrst step toward proving (2. where J(V) is the jacobian. Proof One knows from resolution theory that there exists a map Y → Y obtained by a succession of blowings up with nonsingular centers such that in the diagram Y π Y π X the arrow π is everywhere deﬁned. Clemens and Griﬃths [4]) that A2 (Y) A2 (X) × J(V) and J 2 (Y) J 2 (X) × J(V). π∗ : π∗ (A2 (X)) A2 (X). Then ΘX : A2 (X) → J 2 (X) is an isogeny if and only if ΘY is.e. we show Proposition (2. . We have π∗ π∗ = multiplication by d. then so is ΘX . Θ is surjective with ﬁnite kernel). and let π : Y → X be obtained by blowing up a nonsingular closed subvariety V ⊂ X. If ΘY is an isogeny.. Beauville [1].28 Lecture 2 q = 1. If V is a Proof If V is a point.2). In either case. We shall not pursue these ideas further here. Then Θ : A2 (X) → J 2 (X) is an isogeny (i. and since A2 (X) and J 2 (X) are divisible (Lecture 1.3)). Lemma (2. and let π : Y X be a rational map of degree d 0 (i. then A2 (X) curve. [k(Y) : k(X)] = d). q = 24. We know that ΘY an isogeny implies ΘY an isogeny.5) Let X be a smooth projective threefold. one ﬁnds (cf. Proof In several lemmas: Lemma (2. For the quartic threefold. (1. A2 (Y) and J 2 (X) J 2 (Y).e. the assertion is clear since ΘY induces the identity on J(V). π∗ : π∗ (J 2 (X)) J 2 (X).
and a closed embedding π−1 (W 0 ) φ P2 0 W W0 realizing π−1 (W 0 ) as a family of conic curves over W 0 . It now follows that ΘX is an isogeny. Exercise 1 Let T (W ) = Ker (A0 (W ) → Alb(W )). This implies that π∗ (A2 (Y)) π∗ (J 2 (X)). Exercise 2 T (W ) f∗ ρ∗ Ker (A2 (W × P1 ) → J 2 (W × P1 )) Ker (A2 (X) → J 2 (X)) . Proof Since the geometric generic ﬁbre of π is P1 . Show that f∗ T (W ) = 0. The map J 2 (X) is a map of tori and hence easily seen to be an isogeny. there exists a map f : W → W of ﬁnite degree and a diagram X ρ φ ΘV W × P1 pr1 f V π W W where ρ is obtained by a succession of blowings up with nonsingular centers. If η → W denotes the ¯ geometric generic point in the sense of algebraic geometry. We now enumerate a series of exercises for the reader. .Curves on threefolds and intermediate jacobians 29 In particular. and φ is everywhere deﬁned. Assume that the map A0 (W) → Alb(W) is an isogeny. A rational map π : V W will be called a conic bundle if there exists 0 an open dense set W ⊂ W for the Zariski topology such that π is everywhere deﬁned on π−1 (W 0 ). Then A2 (V) −→ J 2 (V) is an isogeny. π∗ : π∗ (J 2 (X)) so the composition ΘX ◦ π∗ = π∗ ◦ ΘY : π∗ (A2 (X)) → J 2 (X) is an isogeny. where V and W are smooth and projective of dimensions three and two respectively. ΘX is surjective.7) Let π : V → W be a conic bundle. ¯ η ¯ Lemma (2. this is equivalent to the condition that the ﬁbre Vη be isomorphic to the projective line P1 . and hence that this map is an isogeny (its kernel is contained in Ker ΘY ).
0) and H x : T 4 = 0. The line d meets X in 4 points. . .4). T 3 ) = 0. H x ∩ X. at least three of which coincide with x. one can choose homogeneous forms T 0 . π : D → X deﬁned with π−1 (x) = D x for v ∈ X 0. . 0. . T 4 ) . and 3 respectively. and Q x : T 4 = q2 (T 1 . equivalent after change of 2 coordinates to the quadric T 1 + T 2 T 3 ). . . T 4 ) + T 4 · t3 (T 0 . let d ⊂ Q x be the line (ruling) passing through d and x. We return now to the proof of (2. Deﬁne ρ(d) = “fourth point of intersection of d and X. There is a rational morphism ρ : D → X deﬁned as follows: for x ∈ X 0 and d ∈ D x . T 2 . has an isolated ordinary double point at x and no other singularities.) For x ∈ X 0 . T 2 .e. then a reduced. Proof of lemma Q x ∩ X = Q x ∩ (X ∩ H x ) is the intersection of two distinct irreducible hypersurfaces in H x and hence has pure dimension 1.7). (ii) There exist no lines ⊂ P4 supported on V and passing through x. In particular. T 3 ) = 0 . There exists a nonempty open subset X 0 of our quartic threefold X such that for all x ∈ X 0 we have: (i) The intersection of the tangent hyperplane H x to X at x with X. irreducible rational curve of degree 8. . + s4 (T 1 . . q2 is nondegenerate (i. .8) Let x ∈ X 0 and deﬁne C x = ρ(D x ). T 4 on P4 such that x = (1. 3. Then C x = Q x ∩ X is x. r3 . where q2 .30 Lecture 2 Exercise 3 π∗ ◦ f∗ = φ∗ ◦ ρ∗ ◦ pr∗ : A0 (W ) → A2 (V). . . There is a conic bundle over X. . T 4 ) Ker (A2 (V) → J 2 (V)). (Hint: Move to general 1 position.) Exercise 4 φ∗ : Ker (A2 (X) → J 2 (X)) Exercise 5 Prove (2. In particular. If x ∈ X 0 . 4. (This is because the family of lines on X has dimension 1.” Lemma (2. T 3 ) + T 0 · r3 (T 1 . . Q x is a cone over the smooth rational curve D x deﬁned by D x : T 0 = T 4 = q2 (T 1 . Moreover. let Q x ⊂ H x be the tangent cone to x on H x ∩ X. . The fact that x is an ordinary double point on X ∩ H x means that the equation for X has the form 2 T 0 · q2 (T 1 . 0. . T 2 . . x Cx Cx. . 0. Settheoretically Q x is the union of all lines ⊂ P4 which are at least triply tangent to X at x. t3 denote homogeneous polynomials of degrees 2. s4 . In terms of equations.
as is ρ0 : D0 → X. so we have ρ(D x ) = Q x ∩ X. x C x ⊂ H x . Note that ( 1 ∪ 2 ) ∩ X contains p1 . so it is easy to see A2 (S ) Alb(S ). x. p2 with multiplicity one. We assume ∅. Also C x = ρ(D x ). It follows from these Γ0 = Γ ∩ X 0 hypotheses that S 0 = π0−1 (Γ0 ) is an irreducible quasiprojective surface. The hypotheses of (2. Take another curve Γ analogous to Γ such that Γ ψ(S ) but Γ ∩ψ(S )∩ X 0 ∅. There exists a hyperplane L ⊂ P4 such that L ∩ Q x = 1 ∪ 2 . Note that C x ⊂ H x .4). Deﬁne U 0 by the ﬁbre square h0 U0 f0 ψ D0 π0 ρ0 X S U 0 X ∅. p2 ∈ C x and let 1 . The degrees of Q x and X are 2 and 4 respectively. C x ⊂ ψ(S ). that ρ(D x ) ⊂ Q x ∩ X and Q x ∩ X − {x} ⊂ ρ(D x ). Now repeat the above construction replacing . irreducible curve. implies C x C x . we would have C x ⊂ Q x ∩ H x = curve of degree 2.Curves on threefolds and intermediate jacobians 31 {x} ∈ Q x ∩ X is not an isolated component. 2 ⊂ Q x be the corresponding lines. If C x = C x . ( 1 ∪ 2) ∩ X = L ∩ Qx ∩ X = L ∩ C x . S is generically ﬁbred over Γ with rational ﬁbres. If the intersection Q x ∩ X were not smooth at pi . The morphism π0 is a regular morphism. Let Γ ⊂ X be a reduced. On the other hand.4) will follow if we show h0 (U 0 ) is dense in X. and we have by composition a regular morphism h0 : U 0 → X. Returning to the proof of (2. Note that Γ0 ⊂ ρ0 (S 0 ) (because x ∈ C x = ρ(D x )). settheoretically.7) are thus veriﬁed (taking U = suitable completion of U 0 ) and (2. so ρ0 (S 0 ) ∩ X 0 ∅. Also f 0−1 (s) P1 for s ∈ S a general point. Γ C x . because i ∩ X contains x with multiplicity 3. It is clear. This can only happen if for all points x ∈ ψ(S ) ∩ X 0 . so C x is an irreducible rational curve. let π0 : D0 → X 0 denote the restriction of π to D0 = π−1 (X 0 ). the multiplicity of pi on L ∩ Q x ∩ X would be > 1. The idea now is to construct a surface S and a conic bundle U over S by a “bootstrap” technique. and that for some x ∈ Γ0 . H x H x . and ρ0 (S 0 ) ⊂ X has dimension 2. choose two general points p1 . This is impossible (even settheoretically) by reason of degrees. so deg C x = 8. To show that the intersection Q x ∩X has multiplicity 1. Suppose that h0 (U 0 ) is not dense in X. Choose a projective desingularization and completion S of S 0 such that ρ0 S 0 extends to a morphism ψ : S → X. It remains to show that x ∈ X 0 .
b ∈ Lemma (2. f ) is no good either – that is.32 Lecture 2 Γ by Γ !! We get ψ : S → X. hence E is an inﬁnite set. Bloch and Murre [3]. so the inﬁnite set E must contain points x x with C x = C x . b)  a ∈ ?. and we must verify Theorem (2. so ψ (S ) ∩ ψ(S ) = ﬁnite union of curves. This intersection is only a ﬁnite union of curves. we must have C x ⊂ ψ(S ) ∩ ψ (S ). smooth) of codimension 2 in Gr(1.8). Let C ⊂ S be respectively the curve of lines on X and the threefold of lines on V.4). 4) ⊂ Gr(1. This completes the proof of (2.9) r2 X p2 k V q2 L(C) r1 P p1 j q1 L(S ) C S where L(C) and L(S ) are the incidence correspondences L(?) = {(a.2) ΘX is an isomorphism. so we conclude that one of the maps h : U → X or h : U → X must be of ﬁnite degree. p1 is an isomorphism oﬀ the . Then for every x ∈ E. We now resolve singularities of S . One checks easily that P is smooth.10) Proof a a = line corresponding to a} ∩ X}. b ∈ and P = L(S ) ×V X = {(a. 5). and assume both C and S smooth: (2. We now have for the quartic threefold an isogeny θX : A2 (X) ≈ J 2 (X). Since schemetheoretically C = S ∩ Gr(1. smoothness of C implies smoothness of S along C. P is isomorphic to the blowup of S along C. This contradicts (2. Let E = X 0 ∩ ψ (S ) ∩ ψ(S ). One knows that S and C are connected (for details cf. Proof Let V ⊃ X be a quartic fourfold which is general containing X. E is nonempty (it meets Γ ) and open in ψ(S ) ∩ ψ (S ). b)  a ∈ S . §1) and we have assumed C is smooth. n + 1). Assume now that the triple (U . Every surface in X is a hypersurface section by Lefschetz theory. Writing Gr(1. n) for the Grassmann of lines in Pn we have Gr(1. that dim h (U 0 ) < 3. pulling back the line correspondence in S × V. S . n) is locally a complete intersection (in fact.
2). Lemma (2. Since BLS (C) is smooth. this gives a map P π BLS (C) S. Z). Z) = (0) by Lefschetz 1 theory. Z) maps to zero under the map p2∗ : H 3 (P. by (2. of mdimensional hyperplane sections of a smooth projective variety V and a base point 0 ∈ P1 with X = X0 smooth. we work with homology rather than cohomology.13). First.13) The map p2∗ : H 3 (P. Since p2∗ j∗ = k∗ q2∗ . Z) is surjective. Recall that given a Lefschetz pencil. Lemma (2. The diagram A0 (C) J(C) connected ﬁbres A2 (X) ΘX isogeny J 2 (X) shows that ΘX is an isogeny with connected ﬁbres. Z). Z) → H 3 (X. Z) → H 3 (X. From the known cohomological structure of blowingsup. Proof H 3 (S ) sits in H 3 (P) via p∗ = j∗ q∗ . Z) ⊕ H 1 (C. there is associated to any choice of a singular ﬁbre Xti and a path . By a universal property of 1 blowingsup.12) we see the correspondence ∗ r2∗ r1 : H 1 (C. Z) H 3 (S . π is an isomorphism. The map π is bijective and birational. so the corresponding map on jacobians J(C) → J 2 (X) has connected ﬁbres. Z) → H 3 (X. let me show how it implies the theorem. it suﬃces to 1 1 show that q2∗ q∗ H 3 (S ) = 0.Curves on threefolds and intermediate jacobians 33 preimage of C. {Xt }t∈P1 .12) The factor H 3 (S . Z) is surjective. For the proof of (2. Before giving the proof of this.11) and (2. But this group sits in H 3 (V. proving (2.11) H 3 (P. hence an isomorphism. Z) ⊂ H 3 (P. we get (2. and p−1 (C) is a Cartier divisor.
These vanishing cycles are known to generate the kernel of Hm (X. We have families X and P∗ P∗ Y with ﬁbres. . respectively. . Z) → Hm (V. there exists a neighborhood t0 ∈ U ⊂ P∗ such that XU and YU contract onto Xt0 and Yt0 respectively. Z) → Hd (X. Then the image of φ∗ : Hd (Y. The class γ is the base (boundary) of a sort of cone Γ supported on t∈ Xt with vertex the singular point on Xti (Wallace [11]). Xtn denote the singular ﬁbres. Z). Xt0 smooth. Now (2.14) Let V be a smooth projective variety deﬁned over C. Assuming Xt . generically ﬁnite of degree k (with W irreducible). for example. Thus to prove the righthand vertical arrow surjective. Since {Xt } is general. For t0 ∈ P∗ . Let ∈ P∗ be a general line corresponding to a Lefschetz pencil {Xt } on V. Z) φ∗ Hd (Xt . φ : W → V a proper morphism. Let X0 = X and let Xt1 . we may assume that the singular points of the ﬁbres. . Z) contains the vanishing cycles. 1] → such that τi (0) = 0 and τi (1) = ti . where π : Y → P∗ . . Z). Z) Hd (Xt0 . Indeed. let P∗ denote the parameter space for hyperplanes X → V. To each τi there corresponds a vanishing cycle δi ∈ Hd (X. Z). Step b. and Y = φ−1 (X). dim V = d + 1. t ∈ P∗ . and these vanishing cycles generate the group of vanishing cycles. that δ1 ∈ Image(φ∗ ).13) follows by taking W = L(S ). We may assume X generic. .34 Lecture 2 from 0 to ti on P1 a cycle class γ ∈ Hm (X. Z) φ∗ Hd (Yt0 . Z). X → V a smooth hyperplane section. Let U ⊂ P∗ be a nonempty Zariski open set such that Rd π∗ Z is locally constant on U. Let us show. Choose paths τi : [0. We may suppose τi (x) ∈ U for 0 ≤ x ≤ 1. For t ∈ U we get a commutative diagram of specializations Hd (Yt . the bottom horizontal arrow is an isomorphism. Y = P in: Lemma (2. it suﬃces to show the lefthand one is. The sheaf Rd π∗ Z is constructible on P∗ . Proof Step a. hyperplanes Xt ⊂ V and inverse images Yt = φ−1 (Xt ) ⊂ W.
Murre. 95 (1972).9) and the proof of (2. The ﬁrst investigation of the relation between A2 and J 2 for Fano threefolds is in Murre [8]. .14).12). Let ε be close to 1. pn ∈ V lie in the open set over which φ is etale. . Griﬃths. Remarks on the literature Intermediate jacobians were ﬁrst introduced by Weil. H. [4] C. On the Chow groups of certain types of Fano threefolds. the proof of the irrationality of the cubic threefold. This completes the proof of Lemma (2. 10 (1977). Compositio Math. is in Clemens and Griﬃths [4]. ε]) ⊆ U gives Hd (Y0 ) φ0∗ Hd (Yε ) φε∗ Hd (X0 ) Hd (Xε ) δ1 δε . 1–29 in Algebraic KTheory. 47–105. Ann. . 551. References for Lecture 2 [1] A. The cycle map is deﬁned in Grifﬁths [6] and the ﬁrst application to Fano varieties. Berlin (1976).Curves on threefolds and intermediate jacobians 35 ´ p1 . (2). the vanishing cycle δε is “close to disappearing” on Xε – that is. P. But now the fact that τ1 ([0. A. Clemens and P. pp. . Ann. it is supported on Xε ∩ N – so it can be lifted to Yε ∩ M (i) → Xε ∩ N.. Sci. no. Beauville. Bloch. Vari´ t´ s de Prym et jacobiennes interm´ diaires. Since δε ∈ Image(φε∗ ). An example in the theory of algebraic cycles. and let φε : Yε → Xε be the situation over the point τ1 (ε). . ee e Ecole Norm. [2] S. Lecture Notes in Math. where φ : M (i) → N for each i. (4). of Math. I have also borrowed heavily from Tyurin [10]. Bloch and J. and then modiﬁed to vary holomorphically in a family by Griﬃths. we get δ1 ∈ Image(φ0∗ ). 281–356. Springer. [3] S. 39 (1979). The intermediate jacobian of the cubic threefold. 304–391. particularly regarding (2. Let N be a neighbor∼ hood of p1 in V such that φ−1 (N) = M (1) ∪ · · · ∪ M (k) .. Since ε − 1 is small. Sup.
On the quartic threefold. 1970). of Math. Wiley. Murre. London Math. .. I. R. Deligne. 29 (1974). II. 161–206. New York (1958). Griﬃths and J. [Reprinted 1994. [10] A. I. 90 (1969). Wallace. Amer.36 Lecture 2 [5] P. Compositio Math. 496–541. I. Tyurin. GauthierVillars. Th´ orie de Hodge. 1–53.] [11] A. On the periods of certain rational integrals. Homology Theory on Algebraic Varieties. Proc. 90 (1968). 27 (1972).. 425–430 in Actes du Congr` s Intere e national des Math´ maticiens (Nice.] [8] J. New York (1978). 3–50. 27 (1972). Principles of Algebraic Geometry. 714–734. [12] D. N. 5. Algebraic equivalence modulo rational equivalence on a cubic threefold. (2). P. Five lectures on three dimensional varieties. no. Pergamon Press. [Translation: Russian Math. Nauk. Higher Picard varieties. Griﬃths. 90 (1969). 5 (167). Soc. vol. pp. [9] B. (3). Surveys. Harris. Lieberman. Uspehi Mat. J. 1. 460–495. [6] P. 25 (1972). no. Paris e (1971). Math. 1165– 1199. Tennison. [7] P. Ann.
and one wishes to study cycles on X (compare the theory of generalized jacobians of Rosenlicht [4] and Serre [5]). Y) (see below). and Y ⊂ X a divisor with normal crossings. the classical case is precisely the case X = P1 and Y = {0. This problem arises geometrically – for example. An interesting example will be treated at length in Lectures 8 and 9. so one may quotient by the maximal compact real subtorus to obtain cycle classes in a real or complex vector space. The relative cycles are points in P1 − {0. ∞}. ∞} = Gm corresponding to units in a given number ﬁeld.3 Curves on threefolds and intermediate jacobians – the relative case Let X be a smooth quasiprojective variety over a ﬁeld k. A problem of considerable importance is to develop a theory of relative algebraic cycles for X relative to Y. Indeed it seems likely that the higher Kgroups of a variety admit a ﬁltration whose successive quotients can be interpreted up to torsion as relative cycle groups in much the same way that the γﬁltration on K0 relates K0 to the usual Chow group. one expects these cycle classes to ﬁll out a lattice whose volume is linked to some value of a zeta function in much the same way the classical regulator is associated to the value at 1 of the zeta function of a number ﬁeld. ∞}) ∪ (C × {0. (This is a point I hope to take up elsewhere. Y) for the particular case X = C × P1 × P1 with C a nonsingular projective curve and Y = (C × P1 × {0. Indeed. and rather surprisingly. ∞} × P1 ) := C × #. Y) associated to the relative cohomology group H 2r−1 (X. there are compelling arithmetic reasons to develop such a theory. if X is the resolution of a singular variety X and Y is the exceptional locus. J r (X. One feature of such a theory when k ⊂ C is a theory of cycle classes for relative codimensionr cycles in the intermediate jacobians J r (X. Y) will be a noncompact complex torus.) Finally. In this lecture we will work out in some detail the construction of relative cycle classes of codimension 2 in the intermediate jacobian J 2 (X. In some circumstances. . It also arises in Ktheory.
x1 x2 (The reader should check this is always the case when p is a smooth point of C × #. p1 . j = 0.ε be a product of closed disks 1.ε × D2. and that γ has two irreducible components through p.ε . p2 ). ∂Mε ) H1 (∂Mε ). 2 . and (ii) if ⊂ # is any of the four lines.ε0 ∪ D0 × D1. Now ﬁx an ε0 for which the above discussion is valid and let Γ be a 3chain on Mε0 with ∂ Γ = Mε0 ∩ γ + τε0 . b2 x s2 + hot)} (hot = higher order terms). so ∂(γ ∩ Mε ) n1 p0 × ∂D1 × p2 + n2 p0 × p1 × ∂D2 . p = (p0 .1) ε→0 lim τε dx1 dx2 ∧ = 0. x2 for the two obvious meromorphic functions on P1 × P1 . By hypothesis both ni = 0. and let Mε = D0 × D1. we have (3. a1 xr1 + hot. Assume p ∈ Supp γ.) Example (3. one with coeﬃcient +1 and the other with −1. b2 a−s2 /r2 x s2 /r2 + hot)}. where ni is the multiplicity of intersection of γ with C × i ( i = corresponding line) at p. where τε0 is supported on D0 × ∂D1. but only in a rather special way. j}. Changing coordinates (the ﬁrst coordinate is immaterial) we can rewrite these parameterizations {(−. x. ∞. 0).ε × D2. for i. Condition (ii) above implies for ε small that γ ∩ Mε is trivial in H2 (Mε .ε0 × ∂D2. so p ∈ centered at the pi where D0 is ﬁxed and D1 and D2 have radius ε Mε ⊂ C × P1 × P1 . These two conditions are suﬃcient if no component of γ meets any of the “corners” C × {i. we can assume no component of γ meets ∂D0 × D1. b1 a−s1 /r1 x s1 /r1 + hot)}. Indeed. ∞. p2 = 0. To ﬁx the ideas we will consider only the hard case where p1 . 1 {(−. writing x1 . The discussion when p is a smooth point on C × # is analogous but easier and is left for the reader.38 Lecture 3 ni γi on C × P1 × P1 will be called A onedimensional algebraic cycle γ = a relative cycle if (i) all the γi meet C × # properly. Deﬁne Γε on Mε for ε ≤ ε0 by excision. a2 xr2 + hot.2) Suppose p = (0. parameterized locally by {(x. For γ to be a relative cycle we require the existence of Γε0 such that Γε0 ∩ C × # = {p} and such that. Let p ∈ C × #. for the examples it is convenient to permit γ to meet the corners. b1 x s1 + hot)} {(x.ε0 . 0. However.ε0 × D2. γ · C × = 0 as a cycle. moreover. x.
3) There exists a 3chain ∆ε0 on C × Gm × Gm − Mε0 with ∂∆ε0 = γε0 − τε0 + Eε0 .1) s s br1 a21 = br1 a11 1 2 lead to a relative cycle in (3. One gets formally br1 a s1 2πi dx1 dx2 ∧ = log 11 21 + o(1).ε × ∂D2. In Lecture 2. 2 Assume now s1 /r1 = s2 /r2 . .Curves on threefolds – the relative case On D0 × ∂D1. On each neighborhood Mε0 ( j) we can deﬁne a 3chain Γε0 ( j) as before with ∂ Γε0 ( j) = γ ∩ Mε0 ( j) − τε0 ( j) with τε0 ( j) supported on ∂Mε0 ( j). λ(b1 a−s1 /r1 x s1 /r1 + hot) 1 + (1 − λ) (b2 a−s2 /r2 x s2 /r2 + hot) x = ε. These conditions will really only begin to make sense in Lecture 8 after we have studied Ktheory and the tame symbol. The K¨ nneth decomposition gives therefore u 0 H2 (C × Gm × Gm − Mε0 ) H2 (C) ⊗ H1 (C) ⊗ H1 (C) ⊗ Z. . Note that dx1 dx2 ∧ x1 x2 γε =0 . where Eε0 is supported on C × D. and let Mε ( j) be a neighborhood of p( j) as above. 0 ≤ λ ≤ 1 . where τε ( j) ⊂ ∂Mε ( j). a natural chain to take is τε = (λ(−) + (1 − λ) (−). p(N) be the points where components of γ meet C × #. as ε → 0. we deﬁned the cycle map by integration over 3chains. 0 Let γε be the homology 2chain on C × Gm × Gm − Mε (here γε is a 2cycle relative to ∂Mε ) obtained from γ by excision. so ∂ Γε ( j) = γ ∩ Mε ( j) − τε ( j). except that we will deal with a family of 3chains ni γi be a ∆ε and with a limit of integrals over ∆ε .ε × D2. Let Mε ⊂ Mε be the interior.ε . Taking these neighborhoods to be suﬃciently small we may 0 assume the disjoint union Mε = Mε ( j) ⊂ C × P1 × P1 . Let D = (Gm × {1}) ∪ ({1} × Gm ). x. τε = τε ( j). 0 Proof The inclusion C × Gm × Gm − Mε0 → C × Gm × Gm is easily seen to be a homotopy equivalence. We deﬁne Γε ( j) for ε ≤ ε0 by excision. and we ﬁnd the conditions (3. The process will be similar here. .ε . Let p(1). We take Γε = Γε ( j).2). The cycle γε − τε is a 2cycle on 0 C × Gm × Gm − Mε . s1 s2 39 τε A similar analysis is valid on D0 × D1.2. s x1 x2 r1 br a1 2 r1 r2 = . Let γ = 1 1 relative algebraic 1cycle on C × P × P . ε ≤ ε0 . . 0 Lemma (3.
Suppose now we are given a mixed Hodge structure H with weights ≤ 2r−1. These data are required to satisfy the condition that there should exist on grW HC a (necessarily unique) bigradation by subspaces H pq such that n (i) grW HC = ⊕ p+q=n H pq .1) ∂∆ε = γε − τε + Eε . will in this case give us an abelian complex Lie group. The same prescription as before. we can ﬁnd ∆ε0 as claimed. Y. Deligne has deﬁned [1] a mixed Hodge structure with weights ≤ k on H k (Y. (c) a ﬁnite decreasing ﬁltration F ∗ on HC (Hodge ﬁltration). C) → H k (X. m that is grW HQ = 0. and Y ⊂ X a closed (not necessarily smooth or even connected) subvariety. for m > 2r − 1. Y. so γε −τε Lecture 3 dx1 dx2 ∧ =− x1 x2 τε dx1 dx2 ∧ → 0. Y. (ii) the ﬁltration F ∗ induces on grW HC the ﬁltration grW (F) p = ⊕ p ≥p H p .3. that is. We have (3. He also deﬁned [1] a Hodge structure on the relative cohomology H k (X. C)/F r + H 2r−1 (X. We deﬁne ∆ε for ε ≤ ε0 by considering the chain ∆ε0 − Γε0 and excising 0 away Mε . This implies that the class of γε0 − τε0 lies in H2 (C) ⊕ H1 (C) ⊕ H1 (C). (b) a ﬁnite increasing ﬁltration (weight ﬁltration) W on HQ = HZ ⊗Z Q. Example (3.5) Let X be a smooth projective variety. Y. the image of HZ in HC /F r is discrete but not necessarily cocompact. Y) H 2r−1 (X. In general. C) in the relative cohomology sequence. Deﬁnition (3. C) of weight ≤ k compatible with the morphisms H k−1 (Y. We can thus consider the relative intermediate jacobian J r (X. for a proper variety Y. C). . namely J r = HC /F r + HZ . Supp Eε ⊂ C × D. x1 x2 as ε → 0.q (iii) H pq = H qp . C) → H k (X. Z).40 by reason of type.4) (Deligne [2]) A mixed Hodge structure consists of (a) a free Zmodule of ﬁnite type HZ . Since this subgroup is the image of H2 (C × D). The next step is to deﬁne the appropriate intermediate jacobian.
Curves on threefolds – the relative case
41
More speciﬁc example (3.6) Take X = C × P1 × P1 , Y = C × # as above, and let r = 2. Actually, in this case the relative cohomology has a K¨ nneth u decomposition, and we will focus on H 1 (C) ⊗ H 2 (P1 × P1 , #). Combining the exact sequence of relative cohomology for (P1 ×P1 , #) with the Mayer–Vietoris sequence · · · → ⊕4 H i−1 (P1 ) → ⊕4 H i−1 (pt) → H i (#) → H i (P1 × P1 ) → · · · , one checks that there is an isomorphism of Hodge structures H 2 (P1 × P1 , #) Thus F 2 (H 1 (C) ⊗ H 2 (P1 × P1 , #)) = F 2 H 1 (C) ⊗ H 2 (P1 × P1 , #) = (0), so the intermediate jacobian associated to this piece of H 3 (C × P1 × P1 , C × #) is H 1 (C, C)/H 1 (C, Z) H 1 (C, C∗ ). We next deﬁne our cycle map relative algebraic 1cycles on C × P1 × P1 → H 1 (C, C∗ ). H 0 (pt).
Given a relative cycle γ we choose a family of 3chains ∆ε depending on ε such that, with notation as before, ∂∆ε = γε − τε + Eε , Supp Eε ⊂ C × D.
If η is a closed global C ∞ 1form on C, we deﬁne (3.7) Lemma (3.7.1) (i) The limit in (3.7) is deﬁned and < ∞. (ii) Pγ (df ) = 0 if f is a C ∞ function on C. (iii) Diﬀerent choices of ∆ε change Pγ by an integral period of C; that is, Pγ (η) changes by α η for some integral 1cycle α on C independent of η. Proof (i) It suﬃces to examine the situation in a neighborhood M of the ﬁnite set of points where Supp γ meets C ×#. On M we can write η = df , and Stokes’ Pγ (η) = lim
ε→0
−1 4π2
∆ε
η∧
dx1 dx2 ∧ . x1 x2
42 theorem gives
∆ε ∩M
Lecture 3
df ∧
dx1 dx2 ∧ = x1 x2 =
f
−τε
dx1 dx2 ∧ . x1 x2 = 0.
D
Note
dx1 dx2 ∧ x1 x2
γε
dx1 dx2 ∧ x1 x2
Given our hypotheses about relative cycles, it is easy to show that the righthand side tends to 0 with ε, which is enough to deduce existence of the limit. (ii) The argument is precisely the same as for (i), except the integral over ∆ε ∩ M is replaced by the integral over ∆ε . (iii) Suppose ∆ε is another family of 3chains such that ∂∆ε = γε − τε + Eε , Eε ⊂ C × D.
There will exist a 3chain T ε supported on the boundary of an Mε (notation as (3.2)) such that ∂T ε = τε − τε . Since H2 (C × D) → H2 (C × Gm × Gm ) there will exist a 3chain Uε supported on C × D such that ∂Uε = Eε − Eε . Then 1 dx ∆ε − ∆ε − Uε + T ε is a 3cycle. Since 2πi x j j represents an integral cohomology class on Gm , it follows that −1 4π2 Clearly U η ∧ ε hood of Mε , so
dx1 x1 ∆ε −∆ε −Uε +T ε
η∧
dx1 dx2 ∧ = period of η on C. x1 x2
∧
dx2 x2
= 0. Moreover, for ε small, η = df in some neighbordx1 dx2 ∧ = x1 x2 dx1 dx2 ∧ → 0, x1 x2
η∧
Tε
f
τε −τε
as in (i). We conclude
ε→0
lim
−1 4π2
∆ε −∆ε
η∧
dx1 dx2 ∧ = period of η. x1 x2
We now have our cycle map algebraic 1cycles on C × P1 × P1 → H 1 (C, C∗ ). relative to C × # γ→P
γ
Lecture 8 will be devoted to a special example, where the period will be related to the Hasse–Weil zeta function of the curve C. For this purpose, it will be important to factor the noncompact torus H 1 (C, C∗ ) by its maximal compact subgroup – that is, to consider for η a real closed 1form on C (3.8) Im Pγ (η) = lim Im
ε→0
−1 4π2
∆ε
η∧
dx1 dx2 . ∧ x1 x2
Curves on threefolds – the relative case
43
One virtue of this is that the assignment γ → Im Pγ extends naturally to all (not necessarily relative) algebraic 1cycles on C × P1 × P1 which meet C × # properly: Im P : all algebraic 1cycles on C × P1 × P1 meeting C × # properly → H 1 (C, R).
In order to see this, let γ now denote any such algebraic 1cycle on C × P1 × P1 , and let Vε = k Vk,ε ⊂ C be a union of small disjoint disks of radius ε about the C coordinates of points in γ ∩ C × #. Then γ gives a class by excision in 0 H2 (Cε × Gm × Gm , ∂Vε × Gm × Gm ), where Cε = C − Vε . 0 1 1 Writing γε = γ − Vε × P × P , an analysis of the homology group H2 (Cε × Gm ×Gm , ∂Vε ×Gm ×Gm ) shows that there exists a 3chain ∆ε on Cε ×Gm ×Gm with ∂∆ε = γε + Eε + Fε , where Eε  ⊂ (Cε × D) and Fε  ⊂ ∂Vε × Gm × Gm . As before, we can construct a continuous family of ∆ε depending on ε as ε → 0. Lemma (3.9) Let η be a closed real 1form on C. Then Im lim
ε→0
∆ε
η∧
dx1 dx2 ∧ x1 x2
is independent of the choice of ∆ε . Proof Let Wε ⊂ C be the connected, simply connected set obtained by linearly ordering the disks Vk,ε and connecting them by narrow strips.
W:
Since W is simply connected, we have ηW = df . ¯ Suppose now ∆ε is another 3chain with ¯ ¯ ¯ ∂∆ε = γε + Eε + Fε , ¯ Eε  ⊂ C × D, ¯ Fε  ⊂ ∂Vε × Gm × Gm .
Let S = (W ×Gm ×Gm )∪(C×D). One checks easily that H2 (S ) → H2 (C×Gm × ¯ ¯ Gm ). Since Eε + Fε − Eε − Fε is a 2cycle on S which bounds on C × Gm × Gm ,
Generalized Jacobian varieties. 1974).ε is a closed εdisk about a point tk ∈ C. P. reprinted. Math. Deligne.ε −F k. Groupes algebriques et corps de classes. 59 (1954). 79– ee e 85 in Proceedings of the International Congress of Mathematicians (Vancouver.C. Ann. [4] M. 460–495. 5–77. Serre. e no. Griﬃths.44 Lecture 3 there exist 3chains T on W × Gm × Gm and U on C × D such that ∂T + ∂U = ¯ ¯ ∂(∆ε − ∆ε ). x1 x2 Note that ∧ actually represents a class in H (Gm × Gm . Thus Fε. References for Lecture 3 [1] P. II. Inst..ε ( f − f (tk )) dx1 dx2 ∧ = 0. 90 (1969).k is a relative ∂T = k 2cycle on Vk.. [5] J. 496–541. 1984.ε × Gm × Gm .] .k rel (C × D). Paris (1959). x1 x2 This is straightforward. pp. 505–530.k − Fε. [3] P. 1 (1975). D. 1975. and Im ¯ ∆ε −∆ε η∧ dx1 x1 dx1 dx2 ∧ = Im x1 x2 dx2 x2 T +U = Im T = Im 2 f ∂T dx1 dx2 ∧ . Th´ orie de Hodge III. I. 90 (1969). In particular ∆ε − ∆ε − T − U is an integral 3cycle. On the periods of certain rational integrals. Poids dans la cohomologie des vari´ t´ s alg´ briques. Also ¯ ¯ Fε. [Second edition. R). Ann. of Math. Rosenlicht. It will suﬃce therefore to show ε→0 lim Im Fk. of Math.k − Fε. (2). vol. Hermann. Hautes Etudes Sci. [2] P. Deligne.. 44 (1974). B. Publ. where say Vk.
H (µ )) ⊗p (X deﬁned over C).) Let C be a small category. K p ) p p CH p (X). see below. The nerve of C. Basic references are Quillen [3] and Bloch and Ogus [6]. A are isomorphic to objects in M. CH (X)/ CH (X).4 Ktheoretic and cohomological methods In this lecture we will discuss brieﬂy some of Quillen’s ideas about Ktheory. Degeneracies are deﬁned by inserting identity maps. The classifying space for NC is denoted BC. Given a diagram in 0→M →M→M →0 which is exact in A. arrows in Morph C. and their extensions to various other “cohomology theories”. Face operators arise by omitting objects and composing arrows. H p (Z)) H (X. The main results are the cohomological interpretations of the cycle groups H p (X. then A is isomorphic to an object of M. (For notations. NC is the simplicial set with psimplices the set of diagrams x0 → x1 → · · · → x p with xi ∈ Ob C. see Quillen [3]). An exact category is an additive category M which can be embedded as a full subcategory of an abelian category A in such a way that if 0→A →A→A →0 is exact in A and A . Quillen refers to i as an admissible monomorphism and i j . CH p (X)/A p (X) p p H p (X. (For an intrinsic characterization.
a morphism is a diagram in M M1 j QM M i M2 . from which it follows that BM is contractible (exercise). or even a welldeﬁned map might be awkward.e. He deﬁned an Hspace BGL(A)+ and a map from the classifying space of the inﬁnite general linear group ( lim BGLn (A)) − − → BGL(A) → BGL(A)+ . induced an isomorphism on homology with coeﬃcients . and the map K0 (M) → π1 (BQM) sends [M] → 0 M . In other words.46 Lecture 4 j as an admissible epimorphism. The reader is advised to consult Quillen [3]. Quillen’s idea is to build from M a new category QM in such a way that π1 (BQM) k0 (M) ZOb M . When M is the category of ﬁnitely generated projectives over a ring A (associated with 1). for any M ∈ Ob M. which was acyclic (i. [n] − [n ] − [n ]  0 → M → M → M → 0 exact in M He then deﬁnes Kn (M) πn+1 (BQM). Objects in QM will be the same as objects in M. Quillen gave another deﬁnition of K∗ (M) K∗ (A). M1 −→ M2 . M . For example. where i and j are admissible mono and epimorphisms respectively. M2 /M . (A direct proof that this is an isomorphism. that is a ﬁltration M ⊂ M ⊂ M1 with M2 /M . where a preliminary discussion of covering spaces for spaces BC helps to grease the skids. The exact category M has a zero object.) Note ﬁnally that an exact functor F : M → M induces a functor QM → QM . A functor F : M → M on exact categories is exact if it preserves exact sequences. but a morphism in QM. there are two canonical maps 0 → M in QM given by the obvious sub and quotient arrows in M: 0 One deﬁnes in this way a loop 0 M. is by deﬁnition an isomorphism in M of M1 with a subquotient of M2 . and M /M in M and an isomorphism Θ : M1 M /M .
s ∈ R∗ . First. n ≥ 1. the above list of relations is complete (indeed redundant. . Z). and universal means that any central extension of E(A) by an abelian group G arises via pushout from a unique homomorphism K2 (A) → G. Here St(A) denotes the Steinberg group [2]. therefore. K2 for a commutative local ring R: (i) Projective modules over R are free. Here E(A) is the group of elementary matrices. r} = 1. and proved πn (BGL(A)+ ) πn+1 (BQM) (Quillen and Grayson [4]). (c) {r. E(A)] = E(A). (iii) There is a bilinear pairing R∗ ⊗Z R∗ → K2 (R). as (a) and (c) follow from (b)). This pairing satisﬁes the relations (a) {r. s}{s. X any Hspace. q as functors E → M. K1 . Here is a compendium of useful facts about K0 .Ktheoretic and cohomological methods 47 in any local system) and hence (by obstruction theory) was universal for maps BGL(A) → X. t. t. so K1 (R) R∗ . a diagram 0 → E → E → E → 0 being exact if and only if the diagrams in M obtained by applying s. −r} = 1. sending r1 ⊗ r2 to the Steinberg symbol {r1 . We may view s. K2 can also be deﬁned to be the center of the universal central extension of E(A): 0 → K2 (A) → St(A) → E(A) → 1. r. r. so K0 (R) Z. and g are exact. One knows [1] that E(A) = [GL(A). and when R is a ﬁeld. GL(A)] H2 (E(A). These deﬁnitions agreed. 1 − r} = 1 (Steinberg relation). GL(A)] and [E(A). K2 (R) is generated by Steinberg symbols. E itself is an exact category. the unit group of R. 1 − r ∈ R∗ . It is not hard to show from this K1 (A) K2 (A) GL(A)/[GL(A). He deﬁned Kn (A) = πn (BGL(A)+ ). (ii) E(R) = SL(R). We now merely list some of the main results proved by Quillen about the spaces BQM for general exact categories M. r ∈ R∗ . (b) {r. GL(A)/E(A). with deﬁnitions given earlier by Bass [1] and Milnor [2] for K1 and K2 . r2 }. A sequence E ∈ Ob E can be written 0 → sE → tE → qE → 0. let E be the category of short exact sequences in M.
Let A be an abelian category. Then K∗ (B) −→ K∗ (A). Proof Let Pn ⊂ MA be the category of all ﬁnitely generated Amodules admitting a projective resolution of length ≤ n. q) : QE → QM × QM is a homotopy equivalence (i. let X be a noetherian scheme of ﬁnite Krull dimension. induces a homotopy equivalence BQE → BQM × BQM). so K∗ (PA ) = K∗ (P0 ) K∗ (P1 ) ··· K∗ (∪Pn ) U∗ (MA ).3) apply to Pn ⊂ Pn+1 . Resolution theorem (4. Proof The data given amount to a functor F : M1 → E2 . quotient objects. and ﬁnite products in A. Let X i denote the . Then BQP → BQM is a homotopy equivalence. F : M1 → M2 be exact functors.2) Let F . so P0 = PA and MA = ∪n Pn .3) Let P be a full subcategory of an exact category M which is closed under extensions and is such that (i) for any extension 0 → M → M → M → 0 in M. and suppose natural transformations F → F → F are given such that for any object M1 of M1 . Corollary (4. (ii) for any M ∈ M. As an example. The hypotheses of (4.6) Let B ⊂ A be a Serre subcategory. B ⊂ A a nonempty full abelian subcategory closed under taking subobjects. and let T X ⊂ MX denote the full subcategory of sheaves whose support has codimension ≥ i. if M is in P then M is in P. Devissage theorem (4. Then K∗ (MA ) K∗ (PA ). MA the category of ﬁnitely generated Amodules.48 Lecture 4 Characteristic sequence theorem (4. Now apply (4.4) Let A be a regular ring.e. F. where E2 is the category of exact sequences for M2 . Then F∗ = F∗ + F∗ : K∗ (M1 ) → K∗ (M2 ). and PA the category of ﬁnitely generated projective Amodules.5) Suppose that every object M of A has a ﬁnite ﬁltration 0 = M0 ⊂ M1 ⊂ · · · ⊂ Mn = M such that M j /M j−1 ∈ Ob P for all j. Localization theorem (4. Corollary (4. Then there is a long exact sequence of Kgroups · · · → Kn (B) → Kn (A) → Kn (A/B) → Kn−1 (B) → · · · . and let A/B denote the quotient abelian category.1) The functor (s. there exists an exact sequence as in (i) with M ∈ Ob P.1). Let i MX be the category of coherent OX modules. 0 → F (M1 ) → F(M1 ) → F (M1 ) → 0 is exact.
For a noetherian ring or scheme we denote by Kn (resp. If k = 0 and X is irreducible. which is the complex of E1 terms of a spectral sequence (4.7) · · · → Kn (T i+1 ) Kn (T i ) x∈X i = Kn (k(x)) Kn−1 (T i+1 ) ··· · · · → Kn−1 (T i+2 ) Kn−1 (T i+1 ) Kn−1 (k(x)) x∈X i+1 ··· . To pass to the associated sheaves.x OX.x where OX.x /mn ) . if k = dim X ¯ so that p is a closed point.g. let i p A denote the sheaf for the Zariski topology on X obtained by extending by zero the constant sheaf with value A on the Zariski closure of p.x = k(x)modules. Kn (R) := Kn (PR ).x /mn module admits a ﬁnite ﬁltration with successive quotients X. and the quotient T i /T i+1 x∈X i n M(OX.6) to get (for any R a ring. we deﬁne sheaves for the Zariski topology Kn and Kn by sheaﬁ .x /mX. and is functorial for pullback to open sets. { p}. p is the generic point and i p A is the constant sheaf A on all of X.x is the local ring at x and mX. On the scheme level. where PR = f. i p A is the skyscraper sheaf with stalk A supported at p. some notation is convenient. so we may apply (4. the category of all ﬁnitely generated modules or of coherent sheaves). This construction makes sense for any open set of X. The technique of exact couples shows that the above vertical maps ﬁt together to give a complex Kn (k(x)) → x∈X 0 x∈X 1 Kn−1 (k(x)) → · · · → x∈X n K0 (k(x)). X. projectives) (4. Notice that an OX. For example. so we obtain a spectral sequence of presheaves for the Zariski topology on X. Kn ) the Ktheory of the category of ﬁnitely generated projective modules or vector bundles (resp.Ktheoretic and cohomological methods 49 set of points of X of codimension i.q E1 = x∈X p K−p−q (k(x)) ⇒ Kn (MX ). For p ∈ X k and A an abelian group.x is the maximal ideal.8) p. to all of X. Then T i+1 ⊂ T i is a Serre subcategory.x ⊂ OX.5) and (4.
de Rham cohomology [10]. Q.11) Let X = Spec R. In fact. and consider a “cohomology theory” H ∗ (X) which may be one of the following: (i) k = C.q degenerates at E2 . etale cohomology with coeﬃn cients in the rth twist of the sheaf µn of nth roots of 1. or C). It turns out that these results can be established in a more general context which includes certain cohomology theories.12) Let X = Spec R as above.9) p. conjectured by Gersten [7] and proved by Quillen [3]. Then the sequence 0 → Kn (R) → x∈X 0 Kn (k(x)) → · · · → x∈X n K0 (k(x)) → 0 is exact. the corresponding assertion is that i+1 given α ∈ K∗ (T X ) and x ∈ X. I want to sketch this generalization. is Theorem (4. For the global case. R. ∗ ∗ ´ (iii) r prime to char k and H (X) = Het (X.11) is easily seen to follow from i+1 Theorem (4.9) p. .10) Let X be a regular kscheme. as well as Ktheory. H ∗ (X) = H ∗ (X. Then (4.8) reads (4. Let X be a variety over a ﬁeld k. µ⊗r ). where R is a regular local kalgebra which is a localization of a kalgebra of ﬁnite type. where k is a ﬁeld. Exactness of (4. The main theorem.g. E2 = 0 for p 0. there exists an open neighborhood U x in X i such that α → 0 in K∗ (T U ). ordinary singular cohomology with coeﬃcients in some abelian group A (e.50 fying the presheaves Lecture 4 U ⊂ X → Kn (U) or open Kn (U). Since exactness of a sequence of sheaves is a local question. Then the maps K∗ (T X ) → i K∗ (T X ) are all zero. so the complex of sheaves Kn → x∈X 0 i x Kn (k(x)) → x∈X 1 i x Kn−1 (k(x)) → · · · → x∈X n i x K0 (k(x)) → 0 is a resolution of the sheaf Kn (note Kn = Kn since X is regular). A = Z. as well as the basic idea of the proof. the theorem can be restated: Theorem (4. ∗ (ii) char k = 0 and H ∗ (X) = HDR (X). The sheaf version of (4.q E1 = x∈X p i x K−p−q (k(x)) ⇒ Kn . A).
r (A topologist would write HZ (X) = H r (X. then Hr (Z) H 2d−r (Z) (simply take Z = X in the above discussion). (Here if x ∈ Z ⊂ X is the generic point of Z. and deduce from (4. one can use the exact couple technique lim − U⊂Z. As a good exercise for understanding the various exact sequences.) Precisely as before. then we deﬁne H ∗ (k(x)) := H ∗ (U). For purposes of exposition we will n ﬁx an nth root of 1. . if we write 2n−r Hr (Z) := HZ (X). then for any of the above theories. keeping track of the twist by r. there is (in any of the three ∗ cases above) a notion of cohomology with supports in Z. In fact. and identify µ⊗r Z/nZ.13) r r+1 · · · → HZ (X) → H r (X) → H r (X − Z) → HZ (X) → · · · . [9]) is that if X is smooth over k of dimension n. X − Z).q E1 = x∈X P H q−p (k(x)) ⇒ H p+q (X). the reader should construct an exact sequence (for Z ⊂ Z ⊂ X closed subschemes. then Hr (Z) is independent of X and covariant functorial for proper maps Z → Z . Then at the end we will n simply state what the result would be.Ktheoretic and cohomological methods 51 The role of µ⊗r in case (iii) is confusing. If Z ⊂ X is a closed subvariety of codimension p. HZ (X). with X smooth) (4. ξ ∈ k.14) long exact sequences (compare with (4.) One of the basic tenets of Grothendieck duality theory (Verdier [8]. open − → to construct a spectral sequence (4.7)) (4.16) p.14) Now write ∗ HZ p (X) = Z⊂X codim Z≥p ∗ lim HZ (X) − − → · · · → Hr (Z ) → Hr (Z) → Hr (Z − Z ) → Hr−1 (Z ) → · · · . ﬁtting into a long exact sequence (4. H∗ (Z) is a Borel–Moore homology theory [9] in the sense that if Z is itself smooth (but not necessarily complete) of dimension d.15) r · · · → HZ i+1 (X) r HZ i (X) x∈X i = r+1 · · · → HZ i+2 (X) r+1 HZ i+1 (X) H r−2i (k(x)) r+1 HZ i+1 (X) ··· H r−2i−1 (k(x)) x∈X i+1 ··· .
p. Proof The sheaves in (4.10). one calculates with a ﬁner topology on X than the Zariski topology. H p ). One has a morphism π p+q p of sites Xﬁne → XZar and hence a spectral sequence HZar (X. H ∗ (X) = H ∗ (X. Z). CH p (X)/A p (X) . H q ). that is. The general result is due to Deligne. H p ) H p (X. H p ) → 0. Corollary (4. For a proof in the de Rham theory. H ∗ (X) = Het (X. a ∈ F p H ∗ (X) if and only if there exists Z ⊂ X a subscheme of codimension p such that a → 0 in H ∗ (X − Z). char k) = 1.18) The E2 term of (4. Proof One has (because the complexes of global sections of (4.16) can be sheaﬁﬁed as in (4. .16) from E2 onward.21) x∈X p−1 k(x)∗ → x∈X p Z → H p (X. H p ) One has CH p (X). using that K1 (k(x)) = k(x)∗ and K0 (k(x)) = Z.16) is E2 = HZar (X.17) calculate the cohomology) (4.52 Lecture 4 The ﬁltration F ∗ H ∗ (X) deduced from (4. CH p (X)/ CH p (X). This spectral sequence coincides with (4. and (4.17) The complex of sheaves i x H p−1 (k(x)) → · · · → x∈X 1 x∈X p i x H p (k(x)) → x∈X 0 i x H 0 (k(x)) → 0 is a resolution of the Zariski sheaf H p associated to the presheaf U → H p (U).22) x∈X p−1 H 1 (k(x)) → x∈X p H 0 (k(x)) → H p (X.10) and (4. K p ) → 0. the complex of E1 terms (4.16) is the ﬁltration by codimension of support.q p Corollary (4. Also.19) For either the etale or singular cohomology H ∗ (X).20) H p (X. We have p HZar (X. so the complex of abelian groups calculates H ∗ (X. Z/ Z) ( . K p ) H p (X.17) have no higher cohomology. ´ Remark (4. see Bloch and Ogus [6]. R q π∗ ) ⇒ Hﬁne (X). and one has Theorem (4. ∗ ¯ for X over k = k. H q ) = 0 for p > q. for X over C.
H 1 ) 0 H 1 (k(x)) X p−1 Xp Z H p (X. H p ) CH p (X)/ CH p (X). For more details. H p ) → 0. H 1 ) ⊂ Hsing (Γ. H 1 (k(x)) group can be calculated by the K¨ mmer sequence u 0 → µ → k(x) ∗ k(x) → 0 ∗ (k(x) = algebraic closure of k(x)) ∗ 1 and the known vanishing of HGal (k(x). This For the etale theory. Now for a diagram Γ π1 π2 C X as before. Z) one uses the fact that algebraic and homological equivalence coincide for divisors to establish CH1 (Γ)/A1 (Γ) 1 2 HZar (Γ. k(x) ) (Hilbert’s theorem 90). . u(∞) = 1 ∞). Since H 0 (k(x)) = Z/ Z. the presentation (4. see Quillen [3]. One gets H 1 (k(x)) k(x)∗ /k(x)∗ . where C now is any smooth connected curve and Γ is smooth (using resolution of singularities. H p ) 0. where t = π∗ (u) and u is the “standard” function on P1 (u(0) = 0. we may assume this) one gets a commutative diagram H 1 (k(Γ)) γ∈Γ1 norm norm Z H 1 (Γ. Finally. for the singular theory H ∗ (X) = H ∗ (X. 1 ´ HGal (k(x).Ktheoretic and cohomological methods 53 The idea of the proof for Ktheory is that an irreducible family of codimensionp cycles parameterized by P1 Γ π1 π2 P1 X gives rise to a codimension(p − 1) subvariety π2 (Γ) and normk(Γ)/k(π2 Γ) (t) on π2 (Γ). whence H p (X. µ ) (Galois cohomology).22) for ´ the etale theory becomes k(x)∗ /k(x)∗ → X p−1 Xp Z/ Z → H p (X. Z) for any complete nonsingular Γ.
Theorem (4.12). and p ﬁnite. It suﬃces to exhibit an open neighborhood U of x and a closed subscheme Z ⊂ U of codimension p − 1 such that Z ∩ U ⊂ Z and such that the map H2n−r (Z) → H2n−r (Z ) is zero. The best way to think of this is topologically. The pullback θ of π will then also be smooth. Let Z = q(X×S Z) ⊂ X. replacing X by an open neighborhood of x and Z by its intersection with this neighborhood. We have H∗ (Z) −→ H∗ (X ×S Z) −→ H∗ (Z ) so it suﬃces to show σ∗ = 0. Now. We now sketch a proof of (4.10). σ∗ q∗ q*∆ s (Z) X×Z S q ∆ Z . σ(Z) ⊂ X ×S Z. the reader should see Bloch and Ogus [6]. to be deﬁned by the vanishing of a single function σ(Z) : f = 0.54 Lecture 4 The desired isomorphism follows from this. indicating at the end how a similar argument proves (4.17).17) can be reformulated in the spirit of r (4. Given α ∈ HZ p (X) and x ∈ X. For more details. As a ﬁrst step. it suﬃces to show there exists an open r neighborhood x ∈ U ⊂ X such that α → 0 in HZ p−1 (U). First. let n = dim X and let Z ⊂ X be a subscheme of pure codimension p such that α lies in the r image of H2n−r (Z) → HZ p (X). one can construct a cartesian diagram (by a variant on Noether normalization) X ×S Z σ θ q X π Z p S with S ⊂ An−1 an open subscheme. π smooth with ﬁbre dimension 1. so by further localization one can arrange for the image of Z under the section σ.
so [σ(Z)] = 0 in H 2 (X ×S Z) and σ∗ = 0 as claimed.q E1 = x∈X p q−p p+q HGal k(x). one has K∗ (T B ) → K∗ (TC ) → K∗ (TC ) and one wants to show σ∗ = 0. x ∈ X. and [σ(Z)] ∈ H 2 (X ×S Z) is the divisor class. By hypothesis σ(Z) : f = 0. so for any M ∈ Ob T B one gets 0 → C ⊗B M −→ C ⊗B M → M → 0. where θ∗ : H∗ (Z) → H∗+2 (X ×S Z) is topological pullback (which exists even though θ is not proper). Given X a regular scheme of ﬁnite type over a ﬁeld k. Remark (4. the reader is referred to Bloch and Ogus [6]. there is an exact sequence 0 → θ∗ −→ θ∗ → σ∗ → 0 i i of exact functors T B → TC .f σ∗ C θ∗ B→0 i is split.f . q∗ i σ∗ i i Y = Spec B. . Remark (4. Localizing further.1) i i θ∗ = θ∗ + σ∗ : K∗ (T B ) → K∗ (TC ). Writing S = Spec A. then σ∗ (∆) = θ∗ ∆ ∩ [σ(Z)]. one localizes around x to achieve a diagram X = X ×S Y σ θ q X π Y p S i i+1 where Y is a divisor in X. and θ∗ is ﬂat.12) diﬀered only in detail from the above. α ∈ Image(K∗ (T Y ) → K∗ (T X )).16) becomes p. and i+1 α ∈ K∗ (T X ). In other words. n n . This can be avoided. the spectral sequence (4. and π smooth with ﬁbre dimension 1 and p ﬁnite. Keeping track of the twisting.f that is σ∗ = 0. ∩ is cap product. One now applies (4.23) Quillen’s proof of (4.Ktheoretic and cohomological methods 55 If ∆ ∈ H∗ (Z). one can arrange (since θ is smooth with ﬁbre dimension 1) for σ(Y) : f = 0 in X . I ﬁxed an identiﬁcation µn Z/nZ ´ in working with the etale theory. µ⊗m . X = Spec C.24) For ease of exposition. µ⊗m−p ⇒ Het X. C = C⊗A B. The exact sequence 0→C . For complete details of these arguments.
Annals of Mathematics Studies. Verdier. 7 (1974). Lecture Notes in Math. [3] D. 341. n This complex resolves the sheaf H p ( µ⊗m ) associated to the presheaf U → n p Het (U. in Algebraic KTheory I. [9] J. Dualit´ dans la cohomologie des espaces localement come pacts. pp. (1971).L.17) reads i x H p k(x). [4] D. Bass.56 and the complex (4. Quillen. Sup... Higher algebraic Ktheory. n References for Lecture 4 Foundational works on Ktheory: [1] H. Princeton. S´ minaire Bourbaki. Springer. A duality theorem in the etale cohomology of schemes. Ann. Some exact sequences in the higher Ktheory of rings. µ⊗m−p → 0. Paris. Milnor. Berlin (1976). Work on Gersten’s conjecture and analogous results for cohomology theories: [6] S.J. Berlin (1973). Vol. Berlin (1973). Bloch and A. ´ Ann. Some expository references for various cohomology theories and Grothendieck duality: [8] J. Springer. 349–379. Springer. [Ree e printed Soci´ t´ Math´ matique de France. Algebraic KTheory.. [7] S. no. (2). N. Verdier. II. of Math. Gersten. 184–198 in Proceedings of a conference on local ﬁelds (Driebergen. 76. Algebraic KTheory. Berlin (1968). Gersten’s conjecture and the homology of schemes. Quillen and D.. 181–201 (1975). Princeton University Press. M. 217–240 in Algebraic KTheory. vol. pp. Springer. New York (1968). 337–349. Ogus. [6a] S. Sci.] ee e . Grayson.L. Lecture Notes in Math. 300 (1965). Swan. µ⊗m−1 → · · · n → x∈X p i x H 0 k(x). [2] J. µ⊗m ). expos´ no. Lecture Notes in Math. 341. 211–243 in Algebraic KTheory I. Introduction to Algebraic KTheory. Lecture Notes in Math. Higher algebraic Ktheory. no. 1995. Springer. Ecole Norm. [5] R. no. Bloch. 551. (4). Benjamin. pp. 99 (1974). µ⊗m → n x∈X 0 x∈X 1 Lecture 4 i x H p−1 k(x). K2 and algebraic cycles. 1966). Berlin (1967). I. 9. no. 72.
´ 2 Springer. Berlin (1977). no. no. ´ Hautes Etudes Sci. On the de Rham cohomology of algebraic varieties. Publ.. 569. 45 (1975).Ktheoretic and cohomological methods 57 [10] R. . Deligne. Cohomologie etale (SGA 4 1 ). 5–99.. Inst. Math. Lecture Notes in Math. Hartshorne. [11] P.
.
For a diﬀerent geometric proof. . one constructs π X →X such that the strict transform Ci of Ci on X is nonsingular for all i. (The proof given there applies to k = C.2) Suppose the map N A0 (X) → N Alb(X) is injective for any surface X. Then the map A0 (X) → Alb(X) induces an isomorphism on torsion prime to the characteristic of k. and suppose we are given a zerocycle δ= ni (xi ) on X such that Nδ is rationally equivalent to zero. π∗ δ = δ).e.5 Torsion in the Chow group The purpose of this lecture is to apply the techniques of Lecture 4 in order to prove Theorem (5. Let Y ⊂ X be a hypersurface section of large degree. Then viewing the fi as functions on Ci .4)) that the map on Ntorsion N A0 (X) → N Alb(X) is surjective for any N prime to char k. we get ( fi ) = Nδ + η. (g j ) = η. it is easy to see that there will exist lines j ⊂ X and functions g j on j such that π( j ) = pt.1) Let X be a smooth projective variety over an algebraically closed ﬁeld k. We have seen already (Lecture 1. proof of Lemma (1. Then it is so for X of arbitrary dimension. A. Since the exceptional divisor in X is a union of projective spaces. but the same discussion is valid quite generally ´ with singular cohomology replaced by etale. Proof Let dimension X > 2. By a succession of blowings up of nonsingular points on X. By deﬁnition. and no more than two j meet at a point. This theorem was ﬁrst proved geometrically by A. where π∗ (η) = 0. and such that moreover no two Ci meet. there exist irreducible curves Ci ⊂ X and functions fi on Ci such that ( fi ) = Nδ. Let δ = ni (Xi ) be some lifting of δ to X (i.) Lemma (5. Roitman. see Bloch [3]. Let D = ∪Ci ∪ ∪ j .
Since rk I/I 2 > 1 and this sheaf is locally free oﬀ the singular points of D. (I/I 2 )(d)) is surjective and the sheaf (I/I 2 )(d) is generated by global sections. K2 )/ ν H 1 (X. We may continue this construction inductively. Let I be the ideal of D ⊂ X . this implies by a standard Bertini argument that Y is nonsingular oﬀ of D. Kleiman [8]) Alb(Z) and a diagram δ ∈ N A0 Alb(X ) Alb(X) (Z) N Alb(Z) δ ∈ N A0 (X) N Alb(X) Assuming the top arrow injective. etc. we see δ → 0 ⇒ δ = 0. since a singular point of D consists of two simple branches. Since δ was arbitary. the section of (I/I 2 )(d) induced by σ vanishes at worst at singular points of D. Since σ is general. We assume henceforth N = = prime and dim X = 2. I(d)) → Γ(X .3) To establish injectivity N A0 (X) → N Alb(X). . K2 / ν K2 ) → ν A0 (X) → 0. it suﬃces to consider the case N = = prime. Let OX (1) be the ample bundle on X and let Y correspond to a section 0 σ ∈ Γ(X . Finally. OX (d)). Lemma (5. until we get D ⊂ Z ⊂ X. Lemma (5. and Nδ is rationally equivalent to 0 (on Z!). cutting Y by a hypersurface section. This implies that Y is only singular (possibly) at Dsing . One has wellknown isomorphisms (cf. Since dim Y ≥ 2. We work with the reduction mod of the sheaf K2X discussed in Lecture 4. one sees easily that Y can be chosen nonsingular.4) There is an exact sequence (Zariski cohomology) 0 → H 1 (X. this proves the lemma.” I claim that such a Y is nonsingular. (1. We may assume d so I(d) is generated by its global sections. Notice the cycle δ is supported on Z. We want to prove A0 (X) → Alb(X). K2 ) → H 1 (X.3)).. it will have embedding dimension ≤ 2. Further we may assume Γ(X .60 Lecture 5 Suppose D ⊂ Y but Y is otherwise “general. Proof This is straightforward. using divisibility of A0 (X) (Lecture 1. where Z ⊂ X is a smooth complete intersection of hypersurfaces with dim Z = 2.
0 0 0 0 This is not quite obvious. getting 0 → H 1 (X. Q /Z (2)) Alb(X)( ).Torsion in the Chow group Proof (5.4. if we like. and ν CH0 (X) the ﬁrst column now gives the desired sequence. Lemma (5. K2 ) ⊗ Q /Z → H 1 (X.1) implies ν K2 has cohomological dimension ≤ 1. using ν A0 (X) 2 ν H (X. K2 ⊗ Q /Z ) → A0 (X)( ) → 0.5) 3 Het (X. but becomes so once one observes that the map labeled a on the top row is surjective. For example one can tensor the divisor map X 0 i x (k(x)∗ ) X 1 i x Z with µ ν and get i x (k(x)∗ ⊗µ ν ) X0 X1 ix µ ν a i x ( ν K2 (k(x))). pass to the limit in (5. We can. where A0 (X) ( ) denotes the power torsion subgroup.1) 0 0 0 61 The rows and columns of the following diagram of sheaves are exact 0 ν K2 X0 i x ( ν K2 (k(x))) a X1 i x (µ ν ) 0 0 K2 x ν i x (K2 (k(x))) X0 x ν X1 i x (k(x)∗ ) X2 x ν i x (Z) x ν 0 0 K2 i x (K2 (k(x))) X0 X1 i x (k(x)∗ ) X2 i x (Z) 0 0 K2 / ν K i x (K2 (k(x))/ ν ) X0 X1 i x (k(x)∗ /k(x)∗ ) X2 ν i x ( ν Z) 0. K2 ). .4. X0 The ﬁrst row of (5.4).
´ where T = Tate module. Q /Z (3))) Hom(H 1 (X. µ⊗2 ). µ ∞ ) since H 4 (X.1) will follow from (5. In etale cohomological terms T (Pic Var(X)) and Poincar´ duality implies e 3 Het (X. Passing to the limit over ν Alb(X)( ) Hom(T (Pic Var(X)). There exists an isomorphism α : H 1 (X. The lemma now follows. µ⊗2 . µ⊗2 .q p E2 = HZar X. µ ∞ ). K2 / ν K2 ) 3 Het X. ν . Q /Z (2)) making the diagram below commute: H 1 (X. Z (1)). We prove (5. Z (1)). the Zariski sheaves H q (µ⊗ν ) on X were deﬁned to be the sheaves associated to the presheaves U → H q (X. K2 ⊗ Q /Z ) 3 Het (X.6) Q /Z . ν 3 The ﬁrst step is to interpret Het as a Zariski cohomology group. Z (1)). so the em pairing of Weil (Lang [9]) gives an isomorphism ν Alb(X) Hom( ν Pic Var(X). µ ∞ ). K2 ⊗ Q /Z ) α 3 Het (X.6).6) by constructing isomorphisms for any ν αν : H 1 (X. Q /Z (2)) A0 (X)( ) 0 ≈ Alb(X)( ). so there is a Leray spectral sequence p. Q /Z (2)) Theorem (5. Recall from 2 Lecture 4. so Theorem (5. H q µ⊗2 ν p+q ⇒ Het X.62 Lecture 5 Proof The Albanese and Picard varieties of X are dual. Hom(H 1 (X. This implies that H q is the qth ν right derived functor of the morphism of topoi π : Xet → XZar . Q /Z (2)) 1 Het (X. 0 Note that the righthand vertical arrow above is the cycle map. H 4 (X.
1) αν : H 1 (X. The above spectral sequence looks like q 0 1. µ ν ). so H p (X.20) that HGal (k(X).6. g} = f¯ ∪ g where bar denotes class modulo ν th powers. Using the projection formula f¯ ∪ (1 − f ) = f¯ ∪ cor (a) = corL/k(X) ( f¯ ∪ a) = ν cor ( f −ν ∪ a) = 0.2 E1 0 0 0 0 0 p 1. µ⊗2 . . ν We have seen in the proof of (4. ν ν Our objective thus becomes to construct an isomorphism (5. µ⊗2 .6. ν where C is a constant sheaf.2 1 3 from which E2 = HZar (X. ν Deﬁne h{ f. Also we wrote down an acyclic resolution of H q of length q.Torsion in the Chow group 63 In our case H q = 0 for q ≥ 3 because dim X = 2.2) 0 → C → K2 / ν K2 → H 2 µ⊗2 → 0. H q ) = 0 for p > q. This is known when = 2. so (1 − f ) = 1 corL/k(X) (a) for some a ∈ HGal (L. µ⊗2 ). (It may well be that C = 0. so ν 2 k(X)∗ /k(X)∗ × k(X)∗ /k(X)∗ → HGal k(X). ν We will actually construct an exact sequence of sheaves (5. To see ¯ −ν f¯ ∪ (1 − f ) = 0.) Recall Tate’s construction of the Galois symbol [10] 2 h : K2 (k(X))/ ν K2 (k(X)) → HGal k(X). note 1 − f is a norm from the ﬁeld L = k(X)( f ). K2 / ν K2 ) H 1 X. Elman and Lam [7]. µ ν ) there is a cup product pairing ν ν k(X)∗ /k(X)∗ . H 2 (µ⊗2 )) Het (X. so locally (say on pieces of an aﬃne covering) X has cohomological dimension 2. H 2 µ⊗2 .
Let f : X → P1 be a dominant rational map.3) i x (k(x)∗ /k(x)∗ ) X1 X1 ν i x (k(x)∗ /k(x)∗ ) ν i x (Z/ ν Z) X2 X2 i x (Z/ ν Z) 0 0.1) and H 2 (4. We assume Y is geometrically reduced and connected (i. We will thus have our exact sequence (5.3) commute. once we show Theorem (5. K = k(P1 ) ⊂ F. π : Y → Y the natural ﬁnite morphism. µ⊗2 ν H 2 k(X). K is algebraically closed in F). Associated to π there is a covariant trace morphism π∗ ´ on etale cohomology.2) and the ball game will be over.7) 2 h : K2 (k(X)) → HGal (k(X). The reader can show as an exercise that the squares of (5. Let K ⊃ K be a ﬁnite extension ﬁeld. which is analogous to the corestriction on Galois coho . Let F = k(X).17) as follows: 0 K2 / ν K2 0 H 2 ( µ⊗2 ) ν i x (K2 (k(x))/ ν K2 (k(x))) X0 h X0 2 i x HGal k(x).6. The subvariety Y is thus an open curve deﬁned and of ﬁnite type over K. one reduces to the case ν = 1. µ⊗2 0. and let Y be an aﬃne open subvariety of the generic ﬁbre of f . µ⊗2 ) is surjective.64 Lecture 5 The Galois symbol relates the solutions of K2 / ν K2 (5. ν Proof Note that k(X) has Galois cohomological dimension 2. F = F · K .e. Y = YK . From this.6. so there is a diagram K2 (k(X)) h K2 (k(X)) h K2 (k(X))/ 2 (k(X)) h 0 H 2 k(X). µ⊗2 ν−1 H 2 k(X).6.4. µ⊗2 ν (5.
Together with (5. schemes of the form Y ×Y Spec A. so it will suﬃce to show Image H 2 Y. Since Y has cohomological dimension 2.Torsion in the Chow group mology. µ⊗2 → H 2 F. Thus for some ﬁnite K over K we may suppose ¯ (i) Gal(K/K ) acts trivially on H 1 Y. ¯ ¯ Let K be the separable closure of K and write Y = YK · Y has cohomological ¯ ¯ dimension 1. . µ⊗2 ) is a ﬁnite group. µ⊗2 . which on the ﬁbres is simply summing over connected comY Y ponents. In fact. µ⊗2 ) H r (Y. There is also a transfer map tr : K2 (F ) → K2 (F). µ⊗2 One way to understand this situation is to notice that the “geometric ﬁbres” of π (i. µ⊗2 . µ⊗2 cor K2 (F) h H 2 F. µ⊗2 is known to commute (Bass and Tate [2]). π∗ (µ⊗2 )). The map π∗ comes from a map on sheaves π∗ (µ⊗2 ) → µ⊗2 . 65 (5. this reduces us to proving Image H 2 Y . it follows that π∗ in (5. This implies that the higher derived functors of π are trivial.e. so that H r (Y . µ maps onto H Y.1) and the surjectivity of π∗ . there is a commutative diagram r Het Y . µ . µ⊗2 for some ﬁnite extension K of K. where A is a strictly Henselian ring lying over some local ring of Y) are ﬁnite over Spec A and hence strictly Henselian. µ⊗2 cor r HGal F. µ⊗2 .7. µ⊗2 r HGal F . µ⊗2 1 ¯ HGal K. µ⊗2 ⊂ Image K2 (F ) → H 2 F . µ⊗2 . In particular.1) π∗ r Het Y. µ⊗2 ⊂ Image K2 (F) → H 2 F. ⊗2 ⊗2 1 1 ¯ (ii) H Y .7. and the diagram K2 (F ) (5. and the Hochshild–Serre spectral sequence implies H 2 Y.7.1) is surjective. We were free initially to shrink (localize) Y as much as we wanted.7. H 1 Y.2) tr h H 2 F . µ⊗2 → H 2 F . the map on sheaves is surjective. ´ The point is (and this explains why we must work with varieties and etale cohomology rather than the more familiar Galois cohomology of ﬁelds) that ¯ H 1 (Y.
We then have (5. µ ) ⊗ H 1 (F. Z/ Z) were all zero. K1 (X)( ). Concerning (5. µ⊗n ) ν is surjective. µ ) H 2 (Y.7). For a surface X over an algebraically closed ﬁeld of characteristic prime to . the cup product H 1 (F. µ ) ⊗ H 1 (Y. maps 2 onto Het (X. we can replace Y and Y and assume (i) and (ii) hold for Y and K. one can show by a similar argument that for F a function ﬁeld of transcendence degree n over an algebraically closed ﬁeld. Remark (5. µ ) H 2 (F. I wonder whether the whole cohomology algebra ⊕H r (F. Z/ Z) Bockstein r+1 r+1 Het (X.3) H 1 (K. the image of the Bockstein would lie in the ﬁrst level of the “coniveau ﬁltration” (Bloch and Ogus [5]). this would imply (identifying µ Bockstein maps H r (F. In other words.8) The arguments in this lecture can be pushed further in several directions. Globally this would imply the existence of a divisor D ⊂ X such that the composition r Het (X. Z/ Z) → H r+1 (F. µ⊗2 ) h cup K2 (F)/ K2 (F) symbol (K /K ) ⊗ F /F ∗ ∗ ∗ ∗ (F /F ) ⊗ (F /F ) ∗ ∗ ∗ ∗ The reader easily checks that Image(res) ⊂ Image(h) as claimed.7. µ⊗2 ) res H 1 (K. When the ground ﬁeld is . Z/ Z) → Het (X − D. Q /Z (2)). the map being an isomorphism at least when = 2. This leads to some detailed conjectures concerning Chern characters linking ´ Ktheory with ﬁnite coeﬃcients and etale cohomology. I hope to publish a description of these conjectures and the evidence for them elsewhere. Z/ Z) ν was zero. µ⊗r ) might not be ν Z/ Z) that the generated by H 1 ? If true. one can show that the power torsion in K1 (X).66 Lecture 5 By the previous discussion. µ ν )⊗n −→ H n (F.
no. Hirzebruch. 342. Elman and T. 36 (1976). (4). Topology. Lam. 941–945. North Holland. 39 (1979). Deligne. [2] H. Roitman. Berlin (1973).. Amsterdam (1968). that all torsion classes die in the complement of a divisor? References for Lecture 5 [1] M. Amer. Bull. private correspondence. [8] S. Lang. 7 (1974). however. Tate. Ecole Norm. Analytic cycles on complex manifolds. Springer. [5] S. ´ Ann. in Dix exposes sur la cohomologie des schemas. Atiyah and F. Relations between K2 and Galois cohomology. K2 . 80 (1974).. etale cohomology can be replaced by ordinary (singular) cohomology. Algebraic cycles and the Weil conjectures. Soc.. Springer. no. [9] S. [11] A.. Interscience Publishers (1959). ´ 2 Springer. Lecture Notes in Math. Lecture Notes in Math. 181–212 (1975). Sup.] [10] J. and Brauer groups of function ﬁelds. Torsion algebraic cycles. no. Kleiman. [3] S. . Torsion algebraic cycles and a theorem of Roitman. On the quaternion symbol homomorphism gF : k2 (F) → B(F).Torsion in the Chow group 67 ´ the complex numbers.. Z). Math. A. Ogus. Berlin (1977). Lecture Notes in Math. Invent.. It is known by work of Atiyah and Hirzebruch that torsion cohomology classes are not always carried by algebraic cycles [1] (unlike the situation for H 2 (X. Bloch and A. 1 (1962) 25–45. Gersten’s conjecture and the homology of schemes. Y. The Milnor ring of a global ﬁeld. 569. Berlin (1973). in Algebraic KTheory II. [6] P. in Algebraic KTheory II. 107–127. [7] R. Bloch. [4] S. Sci. [Reprinted Springer. whose torsion subgroup is known to come from Chern classes of divisors). Abelian Varieties. Cohomologie etale (SGA 4 1 ). Math. 257–274. Tate. Compositio Math. Is it possible. 1983. Bass and J. 342. Bloch.
.
Ω2 an ) X C. One can derive a version of the Gersten–Quillen an resolution for K2 . In this case. an obvious map CH2 (X) an H 2 (X Zar . and CH2 (X) → Image is the degree map. Although one cannot establish at this time any direct relation between the results obtained and the structure of the Chow group. K2 )) ⊂ Ker(deg). K2 ) → H 2 (X an . OX an )). The easiest case with which to deal is when Pg = q = 0. K2 ). Theorem (6. Ω2 ) H 2 (X an . K2 ) is Z. The problem is that the sheaves involved frequently have an nontrivial cohomology so there is no obvious map H 2 (X an . and let K2 denote the sheaf for the complex topology on X obtained by sheaﬁfying the presheaf U → K2 (Γ(U. . and assume Pg = q = 0. Proof an The existence of a dlog map K2 → Ω2 an leads to X CH2 (X) d log deg an H 2 (X an . Frequently such formal manipulation suggests conjectures about cycles which can be tested directly. K2 ) → CH2 (X). K2 ) d log H 2 (X. however.1) Let X be a surface as above. K2 ) inﬁnitesimally and analytically. Then an the image of CH2 (X) → H 2 (X an .6 Complements on H 2 (K2 ) It is possible to study H 2 (X. an Suppose for example X is an algebraic surface deﬁned over C. conjecturally CH2 (X) Z. an so Ker(CH2 (X) → H 2 (X an . it is plausible that such relations exist. There is. as well as algebraically.
O∗ ) Pic(C). X Since Pg = q = 0.1.1. a cycle on some variety over k must be viewed as the germ of a . (Essentially. Z).1. K2 ). K2 ). X an 0 → K2 → K2 (OX an (∞C)) tame j : C → X. O∗ an ) ⊗ C∗ X (6. This gives a commutative diagram η ∈ H 1 (C an . It suﬃces to note now that the hypotheses Pg = q = 0 imply that H 2 (X an . We ﬁx a perfect ground ﬁeld k which in the case of characteristic zero we assume to be algebraic over Q. OC an ) ∂ an H 2 (X an . O∗ an ⊗ C∗ ) X Pic(C) ∗ H 1 (C. one gets a class η ∈ H 1 (C an . OX ) → H 2 (X an .2) 0 → O∗ an → OX an (∞C)∗ → j∗ ZC → 0.1) ⊗Z C∗ → (6. let δ be a cycle of degree 0 on X and let C ⊂ X be a smooth curve with Supp δ ⊂ C. Indeed.1.1) (6. when k is not algebraic over the prime ﬁeld. the group on the left vanishes and that on the right is ﬁnite. Let OX an (∞C) denote that sheaf of analytic functions on X an meromorphic along C. ∗ ∗ 1 0 → CC an → OC an → ΩC an → 0. One has sequences (6. the exponential sequence 0 → ZX an → OX an → O∗ an → 0 X gives H 2 (X. From the exact sequence of analytic sheaves on C. ∗ j∗ OC an → 0.70 Lecture 6 For the other inclusion. O∗ an ) → H 3 (X an . The rest of this lecture is devoted to the inﬁnitesimal structure of H 2 (X. and one has a symbol map (6.2). C∗ ) ∂ H 2 (X an .1.3) H 1 (C an . Z) ⊗ C∗ ∂⊗1 H 2 (X an . C∗ ) mapping to [δ] ∈ H 1 (C an . O∗ an )⊗ X C∗ = (0).
One has an exact sequence (k = algebraic closure of k in R) (6.2. I) = Ker(K2 (S ) → K2 (R)). I). s} → log(1 + ι) ds .2.I . The basic local result is Theorem (6. s ∈ S ∗ . In terms of symbols. Put S = R ⊗k A and I = R ⊗k m. X .Complements on H 2 (K2 ) 71 family of cycles.1) {1 + ι.I /dI S is given by (6.X×m (Ω1 /dOX ) ⊗k m X 0 (Ω1 ⊗k m / dOX ⊗k m ). and deﬁne K2 (S . S The proof is purely algebraic and will not be given in these notes. getting an exact sequence of sheaves of kvector spaces (6. I) → Ω1 . we may pass to Zariski sheaves.) Let X be a smooth geometrically connected kvariety. K2 )). and we have an isomorphism K2 (S . s ι ∈ I. A.2) k ⊗k m −→ R ⊗k Ω1 → Ω1 . Consider ﬁrst the case char k = 0. S S R a where Ω1 is the module of absolute K¨ hler diﬀerentials.2. Ω1 .3) 0 k ⊗k dm OX ⊗k Ω1 A K2X×A. In the presence of nontrivial derivations of k. Thinking of R as the local ring at a variable point on X.m S R 1⊗d which can be used to analyze the structure of K2 (S . The universal derivation d : S → Ω1 induces d : I → Ω1 .I = Ker(Ω1 → Ω1 ).2) Let R be a local kalgebra.I /dI. the deformations of kcycles will be more complicated. I) S S Ω1 .I /dI → Ω1 /dR ⊗k m → 0. and A an augmented artinian kalgebra with augmentation ideal m. K2 ) 2 augmentation H 2 (X. the map K2 (S . We consider a functor CH2 from the category C of augX mented artinian kalgebras to abelian groups deﬁned by CHX (A) = Ker(H 2 (Xk × A.
O∗ ) → H 1 (X. δ](ε2 . Ω1 /dOX ) → H ∗+1 (X. Example (6. A X 2 There do exist. OX ) = H 3 (X. OX ) = (0). OX ) = (0) (e. If H 2 (X. Ω1 ) ⊗ m is prorepresented by the completion at X 0 of the symmetric algebra on the dual vector space to H 2 (X. O∗ )) Eδ ∈ Ker(H 1 (X × k[δ]. Consider deformations of the H 1 (X. 2 2 Dε ∈ Ker(H 1 (X × k[ε]. Hodge theory gives an exact sequence when X is A complete: 0 → H ∗ (X. Ω1 ) ⊗ m → 0.2. OX ).2.2. (6. OX ) = (0) we see that CHX is itself prorepresentable. εδ). When X H 3 (X. when X is a surface. ·). one gets A (6.2. we ﬁnd CHX isomorphic to the formal group at the origin of the Albanese if and only if H 2 (X.6) displays CHX as a prorepresentable quotient and a non prorepresentable kernel. O∗ ) → H 1 (X. Ω1 ). a result which should be compared with Lecture 1. OX ).72 Lecture 6 Here m = Ker(d : m → Ω1 ).6) 0 → H 2 (X. The functor A → H 2 (X.2. 2 For example.7) trivial divisor Let A = k[ε. so in general the righthand term is more complicated. however. OX ) = (0).5) (i) When H 3 (X. Ω1 ) → H ∗ (X. OX ) ⊗ m. O∗ )) . (ii) A functor F : C → (sets) is said to be prorepresentable if there exists a complete local kalgebra Λ and an isomorphism of functors F(·) Morphlocal kalgebra (Λ. X A In this case the long exact sequence of cohomology associated to (6. Ω1 /dOX ) ⊗ m X (6.g. OX ) ⊗ Ω1 A H 3 (X. artinian kalgebras for which d is not injective. OX )⊗(Ω1 /dm) → CHX (A) A → H 2 (X.4) 1⊗d 2 H 3 (X. OX ) → 0. dim X = 2) or when d : m → Ω1 . δ2 .2. OX ) ⊗ (Ω1 /dm) → CHX (A) → H 2 (X.3) looks like 0 → H 2 (X. Remark (6. H 1 (X.
the intersection is represented by {1 + di j ε. W(R) has a ring endomorphism F induced by the Frobenius r → r p on R and an additive endomorphism V given by P(t) → P(t p ). Viewed as an element in H 2 (X. OX )⊗k(Ω1 /dm). Vw = w σ−1 V with W ∈ W(k). The Frobenius on W(k) is an automorphism and is usually denoted σ. T →1 Big W(R) is a module for big W(R) and inherits the same splitting in characteristic p: W(R) = {E(a0 t) · · · E(an t p )  ai nilpotent}. Identifying A Ω1 /dm k with generator ε dδ. n . I want ﬁnally to discuss the work of Stienstra on CHX when X is deﬁned over a perfect ﬁeld k of characteristic p 0. big W(R). we ﬁnd Dε ·Eδ represented by {di j e jk }⊗ε dδ. W(R) = E(a0 t)E(a1 t p )E(a2 t p ) · · · . K2 )) 73 H 2 (X. K2 ) → H 2 (X × k[ε]. is deﬁned additively by big W(R)+ = (1 + Rt[[t]])∗ = group of power series in R with constant term 1. Fw = w σ F.Complements on H 2 (K2 ) and consider the product Dε · Eδ ∈ Ker(H 2 (X × A. One has FV = V F = p.6) does play a role in the cycle theory. the non prorepresentable kernel in (6. K2 ) ⊕ H 2 (X × k[δ]. the big Witt ring of R. Relations in D are e FV = V F = p. ⎠ ⎝ p p 2 2 For example. it follows that intersection of cycles is given A by cup product H 1 (OX ) ⊗ H 1 (OX ) → H 2 (OX ). k being perfect one ﬁnds that W(k) is a complete characteristic zero discrete valuation ring with maximal ideal pW(k) and residue ﬁeld k. OX ). If Dε and Eδ are represented by 1cocycles 1 + di j ε and 1 + ei j δ respectively. Thus. An important variant on these ideas arises by deﬁning big W(R) = Ker(R[T ]∗ −→ R∗ ). where ⎞ ⎛ 2 ⎟ ⎜ tp tp ⎟ ⎜ ⎟ ⎜ E(t) = exp ⎜t + + 2 + · · · ⎟ = Artin–Hasse exponential.2. so W(R) is a module for the Dieudonn´ ring D = W(k) [F. When pR = 0. The multiplication is determined by functoriality in R together with the requirement (1 − at)−1 ∗ (1 − bt)−1 = (1 − abt)−1 . big W(R) splits into a product of rings W(R). If R is a ring. 1 + e jk δ}. The relation V(F x · y) = xV(y) holds in W(R). P being a power series in t. V].
Then M⊗D W additive group.3. (iii) M/V M is a W(k)module of ﬁnite length. Examples (6. F = σ. and V = pσ−1 .2) (i) M = W(k). where Gm is the formal multiplicative group and the map W → Gm is deﬁned by E(a0 t) · · · E(an t p ) → E(a0 ) · · · E(an ). This suggests considering “formal curves” on K2 : CK2 : (augmented artinian kalgebras) −→ (abelian groups) A −→ Ker (K2 (A[X]) −→ K2 (A)). 2 n W/V W Ga . a surjection A → B of artinian kalgebras induces a surjection M⊗D W(A) → M⊗D W(B)) and has ﬁnitedimensional “tangent space” M⊗D W(k[ε]/(ε2 )) M/V M. (ii) M has no Vtorsion. (6. M = D/D · (F n − V m ) or M = k[[V]].3. CHX is not in general prorepresentable. mV = Fm.74 We view W as a functor Lecture 6 (artinian kalgebras) −→ (left Dmodules). that is to K1 . formally smooth (i. which is prorepresentable. X→0 .1) Let M be a left Dmodule with the following properties: W (i) M is Vadically complete and separated. The tensor product M⊗D W then becomes a functor M⊗D W : (artinian kalgebras) → (abelian groups). For example. We view M as a right Dmodule via mw = wm. the formal As seen in characteristic zero. Such a functor is called a (smooth) formal group.e. The trick is to think of W as being associated to units in a polynomial ring. and it turns out that any smooth formal group can be represented as M⊗D W for some M. This means that W and related functors are inadequate for its description. (ii) M = k[[V]] and F = 0. Then W(k) ⊗D W W/(1 − V)W Gm . mF = Vm.
curves of length n on K2 : Cn K2 (R) = Ker (K2 (R[X]/X n+1 ) → K2 (R)). (1 − rT )−1 × α(X) → α(r) and CK 2 (R) × big W(A) → K2 (R⊗k A).) Also. Pursuing this analogy. R ⊗k M) = Ker (K2 (R ⊗k A) → K2 (R)). Ψ : TCK 2 (R) ⊗D W(A) → K2 (R ⊗k A.Complements on H 2 (K2 ) One can also deﬁne for any kalgebra R. His main local result is Theorem (6.3. R⊗m) → Coker Φ Ψ . and curves on K2 : CK 2 (R) = lim Cn K2 (R).) Stienstra writes ∂W (R) for the Dsubmodule generated by d(W(R)) → − in TCK 2 (R). R ⊗k M). (This makes sense because a is nilpotent. there are splittings analogous to splittings on big W and big W: CK2 (A) = ∞ T CK2 (A) (formal typical curves) and CK 2 (R) = ∞ TCK 2 (R) (typical curves). t }. This suggests that the role of Ω1 in characteristic p is played by TCK 2 . ·} is the Steinberg symbol. when char k = p.3) Let R be a regular local kalgebra and A an augmented artinian local kalgebra. β(x) × (1 − aT )−1 → β(a). ← − − n 75 There are bilinear pairings big W(R) × CK2 (A) → K2 (R⊗k A). where P(t) is a monic power series and {·. TCK 2 and T CK2 are Dmodules. there is a map d : W(R) → TCK 2 (R) deﬁned by d(P(t)) = {P(t). and the above pairings give maps Φ : W(R) ⊗D T CK2 (A) → K2 (R ⊗k A. (This symbol can be shown to make sense even though t is not a unit. Then the composite TCK 2 (R) ⊗D W(A) → K2 (R⊗A.
It is known to satisfy (6. that is to e satisfy (6. (TCK 2 /∂W )⊗D W(A)) → − 2 δ H 2 (X. Ω1 ) X → − in characteristic zero) have a tendency to be Dieudonn´ modules. and the formal groups H ∗ (X. TC n K2 (O)/∂W (O)). although no one has written down a proof. The groups H ∗ (X. OX ) and H 2 (X. TCK 2 /∂W )/(Vtorsion) is the Dieudonn´ → − module of the Albanese variety of X. → − If H 2 (X. TCK2 /∂W )⊗D W(A) → 0. W) has no ptorsion. ← − − and cohomology groups “lim”(TCn K2 (OX )/∂W (OX )) → − ← − − H ∗ (X. TCK 2 /∂W ) may according to Stienstra have Vtorsion and → − so not satisfy (6. but one doesn’t know precisely the correct hypotheses to impose at this point. and these two (playing the role of H 2 (X. e (ii) It seems certain that H 2 (X. this would make H 2 (X. The main geometric consequence of (6.5) (i) The point of taking X to be a surface is that H 2 is right exact.3. It is certainly true under mild hypotheses that H 2 (X. 2). .1) (i)–(iii).1)(ii).3. 2 (iii) The moral is that CHX is controlled by the Dmodules H 2 (X. Then there is an exact sequence for any augmented local artinian kalgebra A · · · → H 1 (X. W) and H 2 (X. Wn (OX )). → − This whole process can be sheaﬁﬁed. In the absence of Vtorsion.3.3) is Theorem (6.4) Let X be a smooth complete surface over a perfect ﬁeld of characteristic p > 0.3.3. and there is an induced exact sequence → − 0 → W(R) ⊗D T CK2 (A) → K2 (R ⊗k A. TCK 2 /∂W ) := lim H ∗ (X. One works with presheaves “lim” Wn (OX ).76 Lecture 6 is trivial on ∂W (R) ⊗D W(A). ← − − n H ∗ (X. TCK 2 /∂W )⊗D W the → − formal group at the origin of the Albanese. W) := lim H ∗ (X.1)(i)–(iii). → − → − ← − − n The former was ﬁrst studied by Serre [8]. the map δ is zero. so the ⊗D can be brought outside the cohomology. R ⊗k m) → (TCK 2 (R)/∂W (R)) ⊗D W(A) → 0. Remarks (6. TCK 2 /∂W ). W)⊗D T CK2 (A) → CHX (A) → H 2 (X. W)⊗D W have been studied by Artin and Mazur.3. TCK 2 /∂W ) is “the part” of crystalline H 3 (X) with slopes → − in [1.
with applications to algebraic cycles. Artin and B. 53 (1978). Applications of the dilogarithm function in algebraic Ktheory and algebraic geometry. 1977). Asterisque. The formal completion of the second Chow group: a Ktheoretic approach. II.P. Some formulas pertaining to the Ktheory of commutative group schemes. 10 (1977). 187–268 (1978). Sup. Can one give an intrinsic characterization of such a functor? Can one understand the link between the algebraic properties of these functors and the geometric properties of the cycles? References for Lecture 6 [1] M. Algebraic Ktheory and crystalline cohomology. J. Deformations of the second Chow group. Utrecht (1978). [5] S. Mazur. (4). Stienstra. Vol. 3 (1975). pp. [4] S. ee e pp.. Ecole Norm. Kinokuniya Book Store. [8] J. 103–114 in Proceedings of the International Symposium on Algebraic Geometry (Kyoto Univ. Publ. [3] S. Kyoto. Serre.. Formal groups arising from algebraic varieties. 64 (1979). K2 of Artinian Qalgebras. Sur la topologie des vari´ t´ s algebriques en caract´ ristique p.Complements on H 2 (K2 ) 2 CH x 77 diﬀers from a formal group in that The non prorepresentable part of ⊗D W is replaced by ⊗D T CK2 . Sci. 149–168 in Journ´ e de G´ om´ trie Algebrique de e e e Rennes (Rennes. 304–326. Comm. Hautes ´ Etudes Sci. o . Bloch. [7] J. no. 47 (1977). 87–132. Universidad Nacional Aut´ noma de Mexico and UNESCO. Math. ´ Ann. Bloch. Tokyo (1978). 24–53 in Symposium Internacional de Topologia Algebraica. Bloch. Stienstra. It would be of interest to study algebraically functors of the form M⊗D T CK2 . Thesis. [6] J. Mexico City (1958). pp. 1978). Algebra. Bloch. [2] S. Algebra. Inst. 405–428.
.
then A0 (X) = (0). and let Γ be k its graph. we will show A0 (X) . Proof Let f : P2 X be a birational map. either geometrically CH0 (X) = Coker X1 k(x)∗ → X2 Z or algebraically by H 2 (X. By a technique of Manin involving Brauer groups. Proposition (7.7 Diophantine questions A smooth projective algebraic surface X deﬁned over a ﬁeld k is said to be rational if its function ﬁeld becomes rational after extension to the algebraic ¯ ¯ ¯ closure k of k. One knows CH0 (P2 ) Z (exercise). deg(x) = [k(x) : k]. The Chow group CH0 (X) is deﬁned as usual. that is k(X) k(t1 . t2 independent transcendentals. Γt∗ Γ∗ The same argument shows. ¯ ¯ where X = X ×Spec k Spec k. If k(X) k(t1 . The degree map deg : CH0 (X) → Z is a bit tricky. t2 ) with t1 . that A0 (X) is a kbirational invariant. t2 ) we say X is krational. To simplify exposition we will tend to assume k is either a local or a global ﬁeld. It can be deﬁned either as the composition ¯ CH0 (X) → CH0 (X) deg Z. From the geometric deﬁnition of CH0 one sees easily that a zerocycle can be moved oﬀ of any given proper closed subscheme of X or P2 . Let A0 (X) ⊂ CH0 (X) be the kernel of deg. of course.1) If X is krational. or directly by assigning to (x). deﬁned over k. The correspondences CH0 (X) −→ CH0 (P2 ) −→ CH0 (X) are therefore inverse isomorphisms. K2 ). x ∈ X 2 .
This torus has been studied by ColliotTh´ l` ne ee ∗ ¯ and Sansuc [2]. the structure of A0 (X) is somehow linked to the places of k where X has bad reduction. M)) ¯ Φ : A0 (X) → H 1 (k. so A0 (X) is ﬁnite in this case. By deﬁnition. we show that if X is a smooth cubic surface (i. when A is a quaternion algebra over the rational function ﬁeld k(t) (for k local or global). T (k)) and show (assuming k to be a local or global ﬁeld) that Image(Φ) is ﬁnite. A0 (X) is ﬁnite for any rational surface over a local or Question (7. Does the order of Image(A0 (X) → A0 (Xkν )) for the various completions of k have anything to do with the local factors at bad places in the zeta function of X? . We then introduce the N´ ron– e ¯ Severi torus. we construct an exact sequence (F = k(X)) (7. The rationality of X implies that N is a ﬁnitely generated free abelian group. The heart of the argument is an analog of the Eichler norm theorem describing the image of the reduced norm Nrd : A∗ → k(t)∗ . They consider a map (as usual H ∗ (k.2) ¯ ¯ ¯ H 0 (k. Conjecture (7. K2 (F)/K2 (k)). ¯ ¯ T (k) = HomZ (N.80 Lecture 7 is not in general zero for X a rational surface. then the only k possible inﬁnite torsion in A0 (X) is 3torsion. T (k)) −→ H 1 (k. K2 (F)/K2 (k)) is a ﬁnite 2group. T (k)) −→ H 2 (k. ¯ ¯ Using Ktheory.4) As the reader will see. µ When X has a rational pencil of genuszero curves (X = conic bundle surface) ¯ ¯ we will show for k local or global that H 1 (k. X ⊂ P3 is cut out by a homogeneous function of degree 3). K2 (F)/K2 (k)) −→ A0 (X) Φ ¯ ¯ ¯ −→ H 1 (k. which is the ktorus with character group the Gal(k/k)module ¯ ¯ N = NS(X) = CH1 (X). M) = HGal (Gal(k/k). T = T X .3) global ﬁeld. As a corollary. k∗ ) with the natural Galois action.e.
2 . ˜ x1 . (functor of points). k∗ ) deﬁnes a morphism of functors a : X × X → Br. A0 (X) ⊂ CH0 (X) : C → (abelian groups). we must show that A0 (X) is not in general zero for a rational surface. x2 ) = a(x1 ) · a(x2 )−1 ∈ Br(k ). The idea is due to Manin. An element a ∈ Het (X. k → A0 (Xk ) ⊂ CH0 (Xk ). Gm )/H 2 (k. x2 ∈ X(k ). In particular. To deﬁne ψa . k∗ ). Given a correspondence deﬁned over k Γ p1 p2 Pk Xk we have (using the existence of transfer for the Brauer group) ψa (p2∗ p∗ ((0) − (∞))) = ψ p∗ (a) (p∗ ((0) − (∞))) 1 1 2 = ψ p1∗ p∗ (a) ((0) − (∞)). where (x1 . x2 ) → 2 ¯ (x1 ) − (x2 ). if ak : X(k) × X(k) → Br(k) is not zero. we associate to a zerocycle ni (xi ) deﬁned over k ψa ni (xi ) = i cor k (xi )/k (a(xi ))ni ∈ Br(k ). There is a natural morphism of functors X × X → A0 . Let C be the category of ﬁnite separable extension ﬁelds of k. it follows that ˜ the group A0 (Xk ) isn’t either.Diophantine questions 81 To convince ourselves there is some interest in the subject.) We consider various covariant functors on C X : C → (sets). ˜ a(x1 . (We avoid the temptation to work with the larger category of all kalgebras because of possible technical problems. but I want to give a very “motivic” (some might say very pedantic) exposition. k → X(k ) Br : C → (abelian groups). ψa Manin’s observation is that there exists a morphism A0 (X) the diagram A0 (X) ψa Br making X×X a ˜ Br commute. ¯ k → Br(k ) = H 2 (k .
t with values in L ⊂ K(X)∗ such that the f s. therefore. Gm )) ⇒ Het (X. Given a zerocycle ni (xi ) deﬁned over k and supported in the complement of the support of L. L/F ∗ ) → H 2 (Gal(K/k). The Brauer group of a surface is a birational invariant. Gm ) 2 Het (P2 . ¯ ¯ Remarks on calculating H1 (Gal( k/k). to an exact sequence ¯ ¯ ¯ ¯ Br(k) → Br(X) → H 1 (Gal(k/k). Then Br(X)/Br(k) Proof ¯ ¯ H 1 (Gal(k/k).5) Let X be a rational surface. Gm ) gives rise.82 Lecture 7 But p1∗ p∗ (a) ∈ Br(P1 )/Br(k ) = (0). H 1 (X. as desired. one checks easily ¯ ¯ H 1 (Gal(k/k). Gm ) = (0) ¯ k (Grothendieck [11]). The righthand term vanishes for k local or global. Here a ∈ Br(X)/Br(k) with ∂a represented by f s. One has H 1 (Gal(K/k). L/K ∗ ). k∗ ).t . Pic(X)). ∗ . Suppose now F is a ﬁnitely generated free abelian group of divisors on XK such that F → Pic(XK ). The Hochshild–Serre spectral sequence p. Gm )) → H 3 (Gal(k/k).q q ¯ p+q ¯ E2 = H p (Gal(k/k).t (xi ))ni i with values in K . so 2 ¯ Het (X. Let L = { f ∈ K(X)∗  ( f ) ∈ F} and consider the exact sequence 0 → L/K ∗ → F → Pic(XK ) → 0. Het (X. In simple situations. F) = (0) (F is a permutation module) whence an exact sequence 0 → Br(X)/Br(k) ∂ H 2 (Gal(K/k).t mod(K ∗ ) represent the image of Br(X) in H 2 (Gal(K/k). Manin actually writes down 2cochains f s. Lemma (7. Pic( X)) If k ⊂ K is a ﬁnite extension ¯ ¯ such that Gal(k/K) acts trivially on Pic(X). Pic(X)). we have ψa ( ni (xi )) represented by the 2cocycle NK(xi )/K ( f s. F). Pic(X)) ¯ H 1 (Gal(K/k). so ψa is trivial on cycles rationally equiv2 k alent to zero.
In this √ case we can take K = k( a). 1. 2. Manin shows that given kpoints p1 . k∗ /NK/k (K ∗ ). (5. f1 (p2 ) = 24 6 To show p1 and p2 not rationally equivalent.Diophantine questions 83 When K/k is cyclic. then p1 is not rationally equivalent to p2 . if at least one of the ratios fi (p1 ) fi (p2 ) NK/k (K ∗ ). Manin [5]). x(x − 1)(x − 7)). ni (xi ) is not In this case one can write down f ∈ LGal(K/k) ⊂ k(X)∗ such that rationally equivalent to 0 if Nk(xi )/k f (xi )ni i NK ∗ ⊂ k∗ . To obtain rational ﬁbres take x = −1. 31 · 5) = 1 (5 ≡ 62 mod 31 and 31 ≡ 1 mod 5. also (5. ai a j (Chˆ telet [1]. Take for example 2 2 t1 − 5t2 = x(x − 1)(x − 7). p2 a rational point over x = 31: −1 ≡ 2 mod NK/k k∗ . k = Q. (−1)(−2)(−8)) = (5. −1) = 1 (−1 ≡ 22 mod 5) and x = 31. 7 we obtain a conic curve. f2 = x − a3 x − a3 are deﬁned and invertible at p1 and p2 . 31 · 30 · 24) ≡ (5. ai ∈ k∗ . For each value of x 0. −8 31 31 ≡ mod NK/k k∗ . K ∗ ) LGal(K/k) /k∗ NK(X)/k(X) (L). Let p1 be a rational point in the curve over x = −1.6) Chˆ telet surfaces are deﬁned in aﬃne coordinates by t1 −at2 = a a (x − a1 )(x − a2 )(x − a3 ). i = 1. p2 on the surface such that the functions x − a1 x − a2 f1 = . 2 2 Example (7. the rationality of which (over Q) is determined by the Hilbert symbol (5. we may at this point pass to a f1 (p1 ) = . for a. one can use the periodicity of Galois cohomology to identify H 2 (Gal(K/k). (5. −1) = 1). 5) = (5. L/K ∗ ) Br(k) ⊃ H 2 (Gal(K/k).
Consider the diagram 0 0 0 k∗ k(e)∗ E1 0 Z 0 (7. Proposition (7. so 6 ∈ NK5 /k5 K5 . K2 ) = Γ(Y. 0 0 (As usual. K2 ) ⊕ k∗ = H 1 (Y. and let Y be obtained from X by blowing up a kpoint. These complexes are covariant functorial for proper maps. The ﬁeld K5 is tamely ramiﬁed over k5 . . Then Γ(X. We conclude that p1 is not rationally equivalent to p2 . On the other hand. the superscript 0 indicates that the righthand column is in degree 0.7. 31 31 ∗ 6 ≡ 1 mod 5. working with the 5adic completions k5 = Q5 and K5 = Q5 ( 5). Using the contravariant functoriality of Γ( · . K2 ). X) (cf. we get f1 (p1 ) NK5 /k5 K5 . The middle and lower rows are complexes whose cohomology is identiﬁed with the cohomology of K2 on Y (resp. even over Q5 .1) K2 (F) y∈Y 1 k(y)∗ Y2 Z K2 (F) x∈X 1 k(x)∗ X2 Z. Lecture 4).84 Lecture 7 √ larger ﬁeld. K2 ) and H 1 (X.2). Our objective now is to construct the fundamental exact sequence (7. ∗ Since 2 is not a square mod 5. k ∗ µk5 ⊂ k5 roots of 1.7) Let X be a smooth surface deﬁned over k. Proof Let F = k(X) and let e be the generic point of the exceptional divisor E on Y.) The middle and righthand columns and the upper row are exact. so local class ﬁeld theory implies ∗ NK5 /k5 K5 = 5Z · (1 + 5Ok5 )∗ · µ25 . K2 ).
It is known that there always exists k ﬁnite over k splitting X. The assertions about H 1 follow from diagram chasing on (7. F = k (X). Gm ) ⊗ k∗ → H 1 (Y. Then Pic(X) ¯ Pic(X) and H 1 (X. Proposition (7.7) noting that Pic(Y) Pic(X) ⊕ Z.1).10) 0→Z→ x ∈X 1 k (x )∗ → X2 0 Z → 0. Deﬁnition (7. Proof The assertions of the proposition are true for X = P2 .9) Let X be a krational surface split over k. K2 ) Γ(Y. K2 ) 85 whence Γ(X.7. so A0 (Xk ) = (0) by (7. and let k be a Galois extension of k splitting X. and checking that the extra factor of k∗ in 1 H (Y. K2 ) Pic(X)⊗Z k∗ .8) A rational surface X deﬁned over k will be said to be split over an extension ﬁeld k if there exists a diagram Xk p q Xk P2 k such that the arrows p and q arise from a succession of blowings up of k points. (X. We get exact sequences (deﬁning Z) (7. K2 ) comes from the map Pic(Y) ⊗ k∗ = H 1 (Y. K2 ) K2 (F). Kq ) see Gillet [14].) One now applies (7. K2 ). . Write X = Xk . K2 ) deduced from the Steinberg symbol Gm ⊗ k∗ → K2 .Diophantine questions we get a commutative diagram (Y. Now let X be a rational surface deﬁned over k. Note in particular that X will be k rational.1). and N = Pic(X ). Sherman [15]. 0 → K2 (F )/K2 (k ) → Z → N⊗k ∗ → 0. (For computations of H p (Pn .
so N⊗k ∗ Hom(N. A surface X deﬁned over k ˜ is said to have good reduction at v if there exists X → Spec(Ov ) smooth and ˜ kv Xkv .10) and Hilbert 90 that A0 (X) H 1 (k /k. any smooth ksurface will have projective with X good reduction at all but a ﬁnite number of places.12) ﬁeld k. X2 0 Z) X2 0 Z. These results show that (i) ⇒ (ii). For k a global ﬁeld. Let X be a rational surface deﬁned over a local or global (i) If X has good reduction at a nonarchimedean place v of k. ˜ Proof An fvalued point of X0 lifts by smoothness to an Rvalued point of X.11) Γ(k /k. N⊗k ) → H 2 (k /k. k ∗ ) as Galois modules. Let v be a nonarchimedean place of the ﬁeld k. Z).86 One has H ∗ (k /k. X1 Lecture 7 k (x )∗ ) x∈X 1 H ∗ (k /k x .2) is zero. Also Γ(k /k.11) may be identiﬁed with (7.12. the kernel of ¯ ¯ ¯ ¯ H 1 (k/k. ¯ ¯ Proof Local duality (Serre [13]) implies that H 1 (k/k. so it suﬃces to prove (i).2) by observing that the intersection pairing on N is perfect. the image of Φ : A0 (X) → H 1 (k/k. k (x)∗ ) by Shapiro’s lemma. projective. Lemma (7. Let f : X → Spec R be smooth. T (k)) → v H 1 (kv /kv . T (kv )) deduced from (7.1) Let R be a complete discrete valuation ring with residue ˜ ﬁeld f and quotient ﬁeld k. T (k)) is ﬁnite for k a local ﬁeld. K2 (F )/K2 (k )) → A0 (X) → H 1 (k /k. ¯ ¯ (ii) Quite generally. (7. then the map ¯ ¯ Φv : A0 (Xkv ) → H 1 (kv /kv . Theorem (7. Finally. so a morphism X0 (1) → X0 obtained by blowing up an fpoint lifts to a diagram ˜ X0 (1) ⊂ X(1) blow up an Rpoint X0 ⊂ ˜ X. T (k)) is ﬁnite. geometrically irreducible. N⊗k ) → H 1 (k /k. K2 (F )/K2 (k )) . where k x = k ∩ k(x). Then the generic ﬁbre X is rational and split over k. . It follows from the ﬁrst sequence in equation (7. Global duality (Nakayama [12]) says for k global. T (kv )) is ﬁnite. Substituting in the second sequence yields (7. and assume the closed ﬁbre X0 is a rational surface split over f.
implying uniqueness and existence of liftings. Indeed.f ). Let k be the local ﬁeld.12)(i). theorem 9. 152). There exists a unique ˜ ˜ family of exceptional curves of the ﬁrst kind E ⊂ Y lifting E0 . both morphisms being obtained by successive blowings up of k kpoints. (Here N = Pic(Xk ) = Pic(Xk ). so both zeroeth and ﬁrst cohomology groups of the normal bundle vanish. it follows that Z P2 . N⊗k ) H 1 (k /k. The units o ∗ in the ring of integers o ⊂ k are cohomologically trivial [13] so N ⊗ o ∗ is cohomologically trivial (Serre [13]. p. Let k be an unramiﬁed extension with residue ﬁeld f splitting X0 . ˜ ˜ We may now blow down E on Y in much the same way as E0 on Y0 . R Let E0 ⊂ Y0 be an exceptional curve of the ﬁrst kind. the 1 normal bundle of E0 P in Y0 is O(−1). Iterating. there is an exact sequence of Gal(k /k) (= Gal(f /f)) modules (F0 = f (X0 )) 0 → F0∗ /f ∗ → Div(X0. and let ˜ X → Spec o be smooth and projective with special ﬁbre X0 and generic ﬁbre X. Because N Pic(X0. . We return now to the proof of (7. Since P2 f f ˜ is rigid.f ) → N → 0.Diophantine questions We may thus build a ﬁgure ˜ Y blowings up 87 ⊃ Y0 blowings down ˜ X ⊃ X0 P2 f ⊂ R2 . we get a diagram ˜ Y0 ⊂ Y ˜ P2 ⊂ Z f ˜ where Z is smooth and projective over Spec R with closed ﬁbre P2 . o ⊂ k the ring of integers. Passing now to the generic ﬁbres we get Xk ← R Yk → Zk P2 .) ¯ The sequence 0 → N⊗o ∗ → N⊗k ∗ → N → 0 yields H 1 (k /k. N).
7.2) 0 H 1 (k /k. 1 Proof A point x ∈ Xk can be viewed as the generic point of a divisor {x } ˜ on X × Spec o ﬂat over Spec o with function ﬁeld k (x ).3) The square in (7.f ).f ˜ on X ×Spec o Spec o .2).12. Since Z ⊂ k (x )∗ (cf. for F = k(X) and X a rational surface over k. K2 (F)/K2 (k)) is ﬁnite. Here tame : K2 (F ) → F0∗ is the tame symbol associated to the divisor X0. we obtain a commutative diagram of Galois modules: 0 K2 (F )/K2 (k ) 1 Z α N⊗k∗ 1⊗valuation 0 0 F0∗ /f ∗ Div(X0.2) is commutative up to sign.10).12. we obtain a diagram with exact rows (F = k (X)) A0 (X) (7.f ) N 0. N⊗k ) θ(7. we obtain a diagram K2 (F ) 1 2 k (x )∗ x 1 ∈Xk α=α x F0∗ Div(X0. The maps labeled 1 and 2 are tame symbols and the diagram commutes up to a sign. Returning to the exact sequence (7.88 Lecture 7 Since Div(X0.12.2) is now straightforward.12. so we obtain a valuation map α x : k (x )∗ → Div(X0. K2 (F )/K2 (k )) tame H 2 (k /k.f is a divisor on {x }. Adding up over various x . N) ∂ Φ H 1 (k /k. The fact that Φ = 0 now follows from Lemma (7. F0∗ /f ∗ ). The assertion regarding (7. The intersection {x } ∩ X0.2) H 2 (k /k. we see that ﬁniteness for A0 (X) for X/k local or global will follow whenever ¯ ¯ ¯ ¯ ¯ we can prove H 1 (k/k.12).f ). This completes the proof of (7.f ) is a permutation module. .
where C is the nontrivial conic over R. then A0 (X) is ﬁnite. for a ∈ k∗ .1) If X is a conic bundle surface over a local or global ﬁeld. we recall that the Milnor ring [8] k∗ (k) of a ﬁeld k is the graded Z/2Zalgebra with generators elements (a) ∈ k1 (k). χ = P1 × C.6) are conic buna dle surfaces. Let k ⊂ k be the ﬁxed ﬁeld of the Sylow subgroup of Gal(k /k). ColliotTh´ l` ne and Sansuc have ee ¯ ¯ ¯ shown H 1 (k/k. The group is nonzero for k = R. K2 (F)/K2 (k)) ⊂ K3 (k) for conic bundle surfaces X. It suﬃces therefore to show the power torsion subgroup A0 (X)( ) is ﬁnite for any prime 3. (ab) = (a) + (b). More generally any surface 2 2 t1 − f (x) t2 = g(x) is a conic bundle surface. In particular if X is deﬁned over R. Indeed. 3 Proof of (7. for a 0.13. and this cohomology group is zero for k local or global unless χ(R) = φ for all real places of k.14) Let X be a conic bundle surface over a ﬁeld k which is either a local ﬁeld. the residual intersection of X with a pencil of planes through is a pencil of conics. Then the group ¯ ¯ ¯ H 1 (k/k.14.1 As a consequence of results in Milnor [8] one has ⎧ ⎪ 0 k C2 ﬁeld. Examples (7. 1. (7. Then does not divide [k : k] so a transfer argument shows 1 Combining their own ideas with arguments given in the text. The existence of a transfer A0 (Xk ) → A0 (Xk ) implies that [k : k]A0 (Xk ) = (0). Theorem (7. and relations 2 (a) = 0.1) (i) The Chˆ telet surfaces discussed in (7. v real place Combining (7. or a Ci ﬁeld [13] for i ≤ 3. (ii) A cubic surface (smooth cubic hypersurface in P3 ) which contains a line deﬁned over k is a conic bundle surface.2) Let k /k be a ﬁnite Galois extension splitting X. then A0 (X) is ﬁnite. ⎪ ⎪ ⎪ ⎩ ⊕ Z/2Z k global. one ﬁnds Corollary (7.Diophantine questions 89 Deﬁnition (7. a global ﬁeld. . ⎨ k3 (k) = ⎪ ⎪ Z/2Z ⎪ k = R. Before stating our main result.1). then A0 (X)⊗Z Z[ 1 ] is ﬁnite. K2 (F)/K2 (k)) is a subquotient of k3 (k). ⎪ ⎪ ⎪ ⎪ 0 ⎪ k local R.13) X is a conic bundle surface if there exists a rational pencil of genuszero curves on X.13) and the above results of Milnor. Corollary (7. and (a) (1 − a) = 0.14.2) If X is a cubic surface over a local or global ﬁeld.14.
therefore. we may assume π is k ¯ ¯ ¯ everywhere regular.14.14. so X is a conic bundle surface (7. ¯ ¯ ¯ where as usual 0 k∗ = Ker( k∗ → k∗ ). ∗ Let n ⊂ K be the image of the norm map from ¯ ¯ K( x)∗ x∈XK ¯ ¯ ¯ Gal(k/k) = x∈Xk K(x)∗ . that Gal(k /k) is an group. Lemma (7. Then End(K ⊕ K )) if and only Proof An extension K /K splits A (i.4) we get an exact sequence (7. XK . Fix a. Proof of (7. From (7. Let Nrd : A∗ → K ∗ be the reduced norm. Recall Xk has 27 lines on it.14. If α ∈ A.14.e.3) and (7. We may assume. Blowing up on X. Similarly (7. and T U = −UT .5) ¯ ¯ ¯ 0 → H 1 (k/k. b ∈ K ∗ such that XK is isomorphic to the conic curve 2 2 2 XK : T 0 − aT 1 − bT 2 = 0. The group Gal(k /k) will necessarily leave a line ﬁxed for 3.14. since XK obviously contains points of degree 2 over K. Since K is a C1 ﬁeld ¯ ¯ [13] the generic ﬁbre of π pulled back to k.3) ¯ ¯ 0 → K2 (K) → K2 (F) → x∈XK ¯ ¯ ¯ ¯ N ¯ K( x)∗ −→ K ∗ → 0. so XK has a K(α)point.1)(ii). WritK ¯ ¯ ¯ ¯ ing F = k(X) = K(XK ) we have (7. then K(α) splits A.4) K2 (K) ¯ K2 (k)⊕ y∈P1 ¯ k ¯ 0 ∗ ¯ k. if α ∈ K. . we get Nrd(α) ∈ n. Nrd(α) = α2 ∈ n. where N is the norm map [7].90 Lecture 7 that A0 (X)( ) → A0 (Xk )( ). Let K = k(P1 ) and K = k(P1 ). K2 (F)/K2 (k)) → K ∗ /n −→ y∈P1 k ψ Br(k(y)). that is a rational map π : X → P1 with ﬁbre of genus 0.6) Nrd(A∗ ) ⊂ n.13. α K. as well as with the quaternion algebra A over K deﬁned by T 2 = a. is isomorphic to P1¯ . all of which we may assume (taking k large to ¯ begin with) are deﬁned over k . A⊗K K if XK has a K point.14. Finally. We will work with (a) · (b) ∈ k2 (K). U 2 = b.14) Fix a conic bundle structure on X. Since the reduced norm on A coincides on K(α) ⊂ A with the ﬁeld norm.
14. (ii) “tame symbol” has kernel k3 (k) ⊂ k3 (K). This implies (a) · (b) → 0 in k2 (K(α)) (Milnor [9]. so K(α) splits A. A necessary k condition for there to exist a cycle Z on X such that π∗ z = ni (yi ) is that ni should be even whenever π−1 (yi ) is nonsplit (i.14.e. Proof Let β = Nrd(α) ∈ Nrd(A∗ ).10) k2 (k(y)) The maps in the above square satisfy: (i) “Galois symbol” is injective. and it will suﬃce to show Ker ψ → k3 (k).14). where c → (c)· (a)· (b). p. (iii) “· (a) · (b)” is injective. K2 (F)/K2 (k)) K ∗ /n ψ y∈P1 k Br(k(y)) K ∗ /Nrd(A∗ ). We may assume α K ∗ . If ( f ) = ni (yi ). Lemma (7. Lemma (7.7) ψ 91 ¯ ¯ ¯ H 1 (k/k.8) The map (a)· (b) : K ∗ → k3 (K).5) 0 (7. Let ni (yi ) be a zerocycle of degree zero on P1 .14. Lemma (7. Assume the ﬁbres of π : X → P1 over the yi are all smooth.Diophantine questions We now can rewrite (7. The projection formula [7] implies (Nrd(α)) · (b) (a) = NK(α)/K ( (α)) · (b) · (a) = NK(α)/K ( (α) · (b) · (a)) = 0. P1 i ) ).7) y∈P1 k Galois symbol Br(k(y)) k3 (K) commutes. Note also their geometric content. 152). Note that these two lemmas suﬃce to complete the proof of (7. contains Nrd(A∗ ) in its kernel.9) The diagram K ∗ /Nrd(A∗ ) · (a)· (b) tame symbol y∈P1 k ψ (7.14.14.14. k(y .
14. we have divisors of degree zero and even order at nonsplit places modulo norms of zerocycles rationally equivalent to on X = ﬁnite group.6. . Notice both numerator and denominator are norms from A∗ . else XK K is silly. then. Hence neither numerator nor denominator vanishes and we have written c as a norm from A∗ .14. Thus the denominator in the above expression vanishes if and only if x1 = x3 = x4 = x7 = 0. in the local or global case when k3 (k) is ﬁnite.4. 2 3 4 7 5 6 Formally.11) Let XK be a conic curve over Spec K with K/K Galois 2 ¯ ∗ ¯ splitting XK .17. Proof of (7. we may factor and write c= 2 abx2 − ax2 − bx2 + x8 2 6 5 2 abx2 − ax2 − bx2 + x7 1 4 3 . in the 2005 book). K ) be the class of XK as a Severi–Brauer .92 (7. 3. Proof of (7. . we consider the map deﬁned in Milnor [8] k3 (K) → I 3 /I 4 .9) implies ψ (f) = Lecture 7 ni [π−1 (yi )]. this form lies in I 4 if and only if it is hyperbolic ([10] cor.14. 367. In particular. given our assumptions about k. Thus (a) (b) (c) = 0 implies there exist x1 . a contradiction. thm.14. Thus. We may assume P1 . and the whole discussion A End(K ⊕ K).10) Injectivity of the Galois symbol follows from Lam [10]. p. where [π−1 (yi )] = class of Severi–Brauer variety in Br(k(yi )). The above quadratic form is a Pﬁster form in the terminology of [10]. (a) (b) (c) → ( c + −1 )( b + −1 )( a + −1 ). and the fact that Ker(tame) = k3 (k) is proven in [8]. But vanishing of the denominator implies vanishing of the numerator and hence all the xi .9) ¯ Lemma (7. x8 ∈ K not all 0 such that 2 2 abc x1 − ab x2 − ac x2 − bc x2 + a x2 + b x2 + c x2 − x8 = 0. Let [XK ] ∈ H (K/K. . X. To show injectivity of multiplication by (a) · (b). p. X is krational. where I is the augmentation ideal in the Witt ring of K. 290. . Finally the reader may wish to compare this chain of ideas with the Eichler norm theorem describing the image of the norm mapping for a divisor algebra over a local or global ﬁeld.
K2 (K)). ¯ ¯ Now take K = k(P1 ). Tensoring the sequence in (7. F ∗ ). we get an exact sequence ¯ ¯ ¯ ¯ 0 → Z/2Z → H 2 (K/K.14. and assume [XK ] 0. K ∗ ) → H 2 (F/F. Proof Z x∈XK ¯ ¯ Gal(K/K) = x∈XK Z. ¯ ¯ Since [XK ] → 0 in H 2 (F/F.Diophantine questions 93 ¯ variety. we ﬁnd a commutative triangle K∗ ∂0 ·∂1 [Xk ]⊗ · ¯ ¯ H 2 (K/K. K = k(P1 ) and compose with the tame symbol to get a . K ∗ )⊗Z K ∗ symbol ¯ ¯ H 2 (K/K. we are done. we obtain a commutative diagram 0 ¯ K ∗ ⊗Z K ∗ ¯ F ∗ ⊗K ∗ x∈XK ¯ 1 ¯ K∗ K∗ 0 0 ¯ K2 (K) ¯ K2 (F) x ¯ ¯ ¯ K( x)∗ ¯ K∗ 0.11) with K ∗ and using the symbol map. Then [XK ] = ∂2 ∂1 (1). where ∂i are the boundary maps associated to this exact sequence. Let F = quotient ﬁeld of XK and consider the ¯ ¯ exact sequence of Gal(K/K)modules ¯ ¯ 0 → K∗ → F∗ → x∈XK ¯ Z → Z → 0. Since any x ∈ XK has even degree over Spec K (this follows from a norm argument using the fact that XK splits over K(x)). Writing ∂i for the boundary maps on cohomology associated to the bottom row. F ∗ ). 1 → ∂2 ∂2 (1).
14. K2 (K)) T1 T2 k3 (K) Br(k(y)) (where T 1 and T 2 are tame symbols) commutes.12) Let k be a ﬁeld of characteristic 2. Z). K ∗ )⊗K ∗ The proof of (7. Note that 1 + π¯ is cohomologically trivial. so o . b ∈ K ∗ and write (a) (b) ∈ k2 (K).14. k = separable closure of k. K ∗ ). Then Br(K) ¯ ¯ H 2 (k/k. one copy for each place lying over y. k∗ ) ⊕ H 2 (k/k. and write ¯ ¯ K = K k. ¯ Lemma (7. Let a. if y is a place of K over k with residue ﬁeld k(y). Proof Replacing k by its perfect closure. K2 (K)). Moreover.9) now follows from ¯ ¯ H 2 (k/k. K ∗ )⊗K ∗ (a.b)⊗ · K∗ · (a)· (b) ¯ ¯ H 2 (k/k. K ∗ ) ¯ ¯ ¯ H 2 (k/k. 0 ¯ o∗ ¯ K∗ Z 0.13) H 2 (k/k. b) ∈ 2 Br(K). (a. Next replacing K by its completion at y we may suppose k(y) ⊂ K and ¯ ¯ Ki .94 commutative square K∗ [XK ]⊗ · Lecture 7 ψ y∈P1 k Br(k(y)) tame ¯ ¯ H 2 (k/k. Now replacing the Galois K = ¯ group by the decomposition group for one of the Ki we may assume k = k(y). K = k((π)). the diagram (7. we may assume k(y) separable over k. We now have split exact sequences 0 o∗ K∗ π Z 1 ¯ ¯ H 2 (k/k.14. Let K be an extension ﬁeld of transcendence degree 1 over k. 0.
14) ¯ ¯ K ∗ → H 1 (K/K.14.·) . δ Indeed. it remains only to consider ¯ Case 2 Suppose b = π. b. f } = T 2 ( (a) (π) ( f )).g2 ) . a ∈ o∗ . Using linearity. Z). .14. With reference to (7. b0 ∈ k∗ denote the mod π reductions of a.13). In this case let G = Gal(k/k) and let ρ : G × G → Z be a 2cocycle representing the image of a under the composition (7. let a0 . b0 ) ∈ H 2 (k/k. f ∈ K ∗ gets taken to ord( f ) · (a0 . This completes the proof of (7. If. 214) that (a. A brief list of works related in one way or another to the point of view of this lecture follows: [1] F.9). 5 (1959).13). b) = (a0 . b ∈ o. p. f is a unit with residue class f0 . References for Lecture 7 There is a vast literature about points on rational surfaces. we reduce to showing that (a0 ) ( f0 ) ∈ k2 (k) maps to the element in Br(k) represented by the cocycle f0ρ(g1 . µ2 ) → H 2 (K/K. Serre [13]. Points rationnels sur certaines courbes et surfaces cubiques. b) is represented by the cocycle ¯ G × G → K∗. on the other hand. g2 ) → π ρ(g1 .14.Diophantine questions 95 Case 1 For a.12) and (7. b0 ) ∈ Br(k). (2). Chˆ telet. k∗ ).14. (g1 . 153–170 (1960). a → χa Here δ is the coboundary from the exact sequence 0 → Z → Z → Z/2Z → 0. π) = (δ χa ) · π ρ(·. In this ¯ ¯ case (a.14. we must show T 1 {π ρ(g1 . one knows (cf. Then (a.14. It is easy enough to see that going either way around (7. a Enseignement Math.g2 ) .14) replaced by k and K replaced by k. If F = −π this is clear as both sides are trivial.g2 ) . This follows as ¯ ¯ above with K in (7.
L. Y. ColliotTh´ l` ne and D. bf 39 (1979). [Second edition. Coray. vol. (1971). Sup. (4). Ann. [13] J. Springer. Manin. Amsterdam (1974).J. 301–332. Milnor.L. [3] J.] [6] Yu. 401–411 in Actes du Congr` s International Math´ maticiens e e (Nice. Berlin (1973). and for quadratic forms: [7] H. For a more extensive list.I. The Algebraic Theory of Quadratic Forms. 65 (1957).L. Le groupe de Brauer I. second edition. 1. N. W. 284 (1977). pp. Hermann. 349–446 in Algebraic KTheory II. Applications of algebraic Ktheory to intersection theory. New York. Lecture Notes in Math. L’´ quivalence rationnelle sur les ee e points ferm´ s des surfaces rationnelles ﬁbr´ es en coniques. Grothendieck. ee e ´ Ann. The Milnor ring of a global ﬁeld. Corps Locaux. Algebraic Ktheory and quadratic forms. 1979. Harvard (1978). the reader can see the bibliography in [5]. See also Introduction to Quadratic Forms over Fields. 72. Ecole Norm. R. Sci. Mass. Nakayama. P. [15] C. 318–344. Cubic Forms. (2). e [12] T. [Revised second printing. Sherman. Benjamin. pp. Sansuc. Providence. Paris (1971). [5] Yu. 2005. pp. Bass and J. [9] J.. Tate.96 Lecture 7 [2] J. Kcohomology of regular schemes. Comm. Introduction to Algebraic KTheory. A967–A970.] [11] A. C. A449–A452. Le groupe de Brauer–Grothendieck en g´ om´ trie diophantie e enne. Cohomology of class ﬁeld theory and tensor product modules I. 255–267. R. II. [10] T. The following are referred to in the text for the arithmetic of symbols in Ktheory. Serre. La R´ quivalence sur les tores. Princeton University Press. North Holland. American Mathematical Society. AB. Annals of Mathematics Studies. [4] J. Acad. 10 (1977).. vol. 1986.] [14] H. 175–229. Math. 999–1027. III. Gillet.. 1980. Compositio e e Math. 342. Amsterdam (1968). Reading. Algebra.. A1113–A1116. no.J. of Math. 1970). Manin. [Translation: Local Fields. Lam. Paris (1968). Springer. Sci. . (1973). A1215– A1218. Series of notes on rational varieties ee and groups of multiplicative type. Invent. A. ColliotTh´ l` ne and J. 284 (1977). Sansuc. 282 (1976). 9 (1970). 287 (1978). [8] J. North Holland. GauthierVillars. 46–188 in Dix expos´ s e sur la cohomologie des sch´ mas. Paris Ser. Thesis.J. Princeton. Milnor. 7 (1979). ColliotTh´ l` ne and J.
In Lecture 9 we consider the case when E/Q has complex multiplication by the full ring of integers in an imaginary quadratic number ﬁeld. the conductor of E.8 Relative cycles and zeta functions We return now to the relative cycle maps described in Lecture 3. (Here C is a smooth complete curve over a ﬁeld k → C. R) C. C∗ ). where I deﬁne a map from Ktheory Γ(E. χGr¨ ss ). R) ⎪ ⎪ ⎪ ⎭ relative algebraic 1cycles on C × P1 × P1 → H 1 (C. χ) has an Euler product converging for Res > 3/2. o where L(s. it will follow that log  j  (γ) 0. whence the notation) and ρ is a constant built from a Gauss sum. Since L(s. K2 ) → H 1 (E(C). and the imaginary part of τ. χGr¨ ss ) is the Artin–Hasse zeta function of E (given in this case by a Hecke Lseries.) When C = E is an elliptic curve we will show explicitly how to write down some interesting relative cycles γ and we will compute log  j  (γ). we show there is a natural relative cycle γ deﬁned over Q such that o ρ · log  j  (γ) = L(2. we write j: and ⎧ ⎪ algebraic 1cycles on ⎪ ⎪ ⎪ ⎨ log  j  : ⎪ C × P1 × P1 meeting ⎪ ⎪ ⎪ ⎩ C × # properly ⎫ ⎪ ⎪ ⎪ ⎪ ⎬ 1 ⎪ → H (C. Changing notation. K2 ) can be identiﬁed with a suitable Chow group for . Identifying H 1 (E. It seems likely that Γ(E. for the cycle class maps. These computations mimic those in an unpublished preprint [2]. R) C.
0. Our ﬁrst objective will be to associate to an element in Γ(C. gi } ∈ Γ(C. i. relative to C × #. This cycle will only be well deﬁned modulo some equivalence which we shall not describe explicitly. K2 ) K2 (k(C)) T x∈C k(x)∗ . We iterate the above procedure mutatis mutandis for every vertex point (p. Recall we have an exact sequence 0 Γ(C. i. or τ ⊂ (C × P1 × {1}) ∪ (C × {1} × P1 ). χ) at s = 0. Recall we have on K2 (k(C)) the tame . a tr . j). hot = higher order terms. 0) like (8.g = {(x. K2 ) ⊂ K2 (k(C)). τ meets C × # with multiplicity r at (p. Let γ1 = γ fi . i. Kgroups.1. where x is a local coordinate near p. K2 ) a relative algebraic 1cycle on C × P1 × P1 . 0) by subtracting the “trivial” curve (8. bx s + hot)}. Note that in addition to the intersection at (p. and zeta functions quite generally. axr + hot. ∞). a. Indeed. b t 1−t s .2) shows that this component can be made to look like a relative cycle near (p. (−1) s b) and with multiplicity s at (p. one may hope for a similar interplay between relative cycles. A curve τ ⊂ C × P1 × P1 is “trivial” for our purposes if either τ ⊂ {p} × P1 × P1 for some p ∈ C.2) τ= p.gi . ∞. The name of the game will be to add or subtract “trivial” curves to γ1 in such a way that the resulting sum is a relative algebraic 1cycle. 0). Consider then an element { fi . 0.98 Lecture 8 E × P1 × P1 relative to E × # (cycles being deﬁned over Q) and that the rank of this group is the order of vanishing of L(s.1) {(x.1. g(x))} ⊂ C × P1 × P1 . j = 0. ∞. x(p) = 0. g of rational functions on C we associate the double graph γ f. f (x). One sees from Lecture 3 that log  j  (trivial curve) = 0. 0. Step 2 (Neutralizing other boundary points) Let γ2 = γ1 − τ be the cycle with vertex intersections neutralized. t = standard parameter on P1 . Step 1 (Neutralizing the vertices) It may happen that Supp γ1 contains a point (p. To each pair f. A calculation like (3. Suppose for example one of the components of γ1 looks near (p. I hope this exciting prospect will sustain the reader through the complicated calculations which follow. j) on γ1 .
∞} × {1}). (t − a)(t − b) (t − 1)(t − ab) t−a t−b t ∈ P1 . T p { f. ∞}) ∪ (C × {0.pr such that γ3 · (C × #) has support in (C × {1} × {0. let ρ(w) ∈ Gm be either the second or third coordinate. a. For w ∈ Supp W. Lemma (8. + (p. ∞.W.b. w∈({p}×P1 ×P1 )∩Supp W Proof A cycle like (8.p · W = (p. 1) − (p.brs .1) contributes (−1)rs a s b−r to the tame symbol. 0) + (p. ¯ Aa. gord p f 99 The tame symbol can be interpreted geometrically as follows: let W = C × Gm × {0} − C × Gm × {∞} − C × {0} × Gm + C × {∞} × Gm (note we have removed the vertices).3). ∞) − (p. viewed as a cycle.p = p. ∞) + (p.w) . 1.b. ab. b ∈ Gm and p ∈ C. where τ is as in (8.1. we consider the “trivial” cycles Aa. it is straightforward to construct a cycle γ3 = γ2 + ni j Aai j .p · W = (p. ¯ We have for W = closure of W. gi } = ρ(w)ν(γ2 . whichever 0. 1 = Tp i { f i . b. ∞. ∞) − p. W.1. t. 0. 1). g} = (−1)ord p f · ord p g f ord p g (p). t ∈ P1 . b Starting at these formulae and using (8. 0. and is left for the reader. 0) − (p.b. a. p.Relative cycles and zeta functions symbol at p T p : K2 (k(C)) → k(p)∗ .pi + mrs Bars .p = Ba. 1). This is precisely the contribution of the cycle −τ to the righthand side. x) be the multiplicity of intersection of γ2 and W at x.b.bi j . a ¯ Ba. b. ∞. The bookkeeping for nonvertex points is easier.1.3) With notation as above. 0) − (p.1. t. . Given a. For x ∈ C × P1 × P1 let ν(γ2 .2).
1. 1)  p ∈ C} + {(p. Notation (8. therefore. There exist. h2 (p))  p ∈ C} is a relative cycle. pr1∗ [γ4 · C × P1 × ((0) − (∞))] are zerocycles rationally equivalent to 0 on C. and a zero of order N 2 − 1 at the origin. i = 0.100 Lecture 8 Step 3 (Finishing the job) We may now add or subtract cycles of the form (p) × (1) × P1 and (p) × P1 × (1) to γ3 getting a γ4 such that γ4 · C × (i) × P1 and γ4 · C × P1 × (i) are eﬀective. Deﬁne S a ∈ Γ(E. and assume all zeros and poles of f and g are points of E of order N deﬁned over k. K2 ) . S a is actually well deﬁned modulo torsion and symbols with both entries constant.2) K2 (k(E)) tame x∈E k(x)∗ . k∗ and goes to 0 in Pic(E) ⊗ k∗ . . N > 0 an integer. g}N { fi . but no cycle γ4 − (p) × P1 × (1) or γ4 − (p) × (1) × P1 has this property. Write pr1 : C × P1 × P1 → C. Suppose E is deﬁned over a ﬁeld k. Consider the diagram k(E)∗ ⊗k∗ symbol div⊗1 x∈E k∗ Pic(E) ⊗k∗ 0 (8. functions h1 .2). K2 ) by S a ≡ { ρ. We have that pr1∗ [γ4 · C × ((0) − (∞)) × P1 ]. as one sees by looking at the kernel of div⊗1 in (8.1) Fix an integer N and assume all points on E of order N are deﬁned over k.2. g}N ∈ ∗ functions fi ∈ k(E) and constants ci ∈ k∗ such that { f. fa }N mod Image(k(E)∗ ⊗k∗ → K2 (k(E))) . K2 ) (and hence some interesting relative algebraic 1cycles) in the case C = E = elliptic curve. Let ρ be the function with poles of order 1 at every nonzero point of order N. let fa denote the function with zero of order N at a. ∞. and pole of order N at 0. g be functions on E. We now want to construct some interesting elements in Γ(C. h2 on C such that γ = γ5 = γ4 + {(p. Let f. ci } ∈ Γ(E. If a is such a point. h1 (p). There thus exist Clearly tame{ f.
T g or T f. We let t be the standard parameter on C∗ and write t = e2πiz . nk = m ja j = so if we write f˜ = j (1 − αj t)n j . The handles of the barbellshaped aﬀairs are called T f . The εcircles are written S f . nk bk = 0. We assume for simplicity that the divisors of f and g have disjoint support. g(z) = cg mj = For any ﬁxed n we have e2πink (z−ak −nτ) = 1. g on E. z = x + iy. taking care that the slits for f and g do not intersect.g ) when γ f. .Relative cycles and zeta functions 101 Our objective is now to compute log  j(S a ). (This is possible because removing a simple arc will not disconnect the surface. g(x))  x ∈ E} for any two functions f. ∞ The given functions f.ε . The key step in this program will be to compute log  j(γ f. 2πi (1 − qn )2 n=1 cf . S g or S f.ε .) Slit E along nonintersecting simple closed paths between zeros and poles of f .) The resulting slit and punctured surface will be denoted E . k φ(z − bk )nk . It will be convenient to use the Jacobi parameterization π : C∗ → E ∗ Z C /q . (The general case follows from this one by a limit argument. T g. We do likewise for the zeros and poles of g. 1 2 σ(z) = (1 − qn t)(1 − qn t−1 ) 1 (t − 1) . 2π We normalize the Weierstrass σfunction on E as in Lang [4] φ(z) = e− 2 ηz +πiz . Thus y = −1 log  t  is a function on C∗ . g on E can be factored in terms of the φfunction f (z) = cf φ(z − a j )m j .g = {(x. S g. αj = e−2πia j . f (x).ε . but x is not. removing also εdisks about these points: Sf Tf f=0 Sf f = ∞. we get f (t) = cf n∈Z f˜(tqn ). q = e2πiτ . cg ∈ C∗ .ε .
1 f (p). − − → −−→ −− −− ∂Φ = {(p. and n denotes the integral along paths from points of negative multiplicity to points of positive multiplicity. whence ∂(Φ − ∆) = γ f. Deﬁne 3chains −−→ −− ∆ = ∆ε = {(p. Bearing in mind orientations. we get −−→ −− ∂∆ = {(p.) Proof of lemma The ﬁrst claim is that ε→0 lim dy ∧ Eε dg = 0. Our duty becomes the calculation of ε→0 lim −1 Im 4π2 Φε −∆ε dx ∧ dx1 dx2 ∧ x1 x2 and ε→0 lim −1 Im 4π2 Φε −∆ε dy ∧ dx1 dx2 ∧ . log f . 1 f (p). − − → −−→ −− −− Φ = Φε = {(p. Functions like log f and log g (resp. (Here n denotes the inﬁnite “cycle” on C∗ n= n∈Z j. π : C∗ → E . C∗ ).k αk = e−2πiak . j n − n log(1 − t) dy .3) ε→0 lim ∆ε dy ∧ dx1 dx2 ∧ = 2πi y log(1 − t) x1 x2 m j nk (αk β−1 qn ).102 Lecture 8 Finally. 1 f (p). let C∗ denote C∗ slit and punctured so as to lie over E . β j = e−2πib j . 1g(p))  p ∈ T f }. g(p))  p ∈ E }. x1 x2 Lemma (8. 1g(p))  p ∈ ∂T f } −−→ −− + {(p. 1 f (p).g + chains supported on (S g ∪ S f ) × Gm × Gm + degenerate chains. A diﬀerent choice of paths does not change the righthand side. (where deg = cycle supported on E × Gm × {1} ∪ E × {1} × Gm ). 1 f (p). g(p))  p ∈ T f } + deg.g + deg. g(p))  p ∈ S g ∪ T f ∪ S f } = γ f. Also a family of paths 1 f (p) or 1g(p) lying on Gm can be chosen continuously for p ∈ E . g . log f˜. log g) can be deﬁned on E − − → −−→ −− −− (resp.
( f˜) Standard estimates imply ε→0 lim log g d(y log f˜) = 0. x2 Note that the diﬀerential can be written log f dy ∧ dx2 dx2 dx2 = cf dy ∧ + “trace”C∗ /E log f˜ dy ∧ . The integral over cf dy ∧ dx22 vanishes by the above. Winding around a zero of g changes log g by 2πi. think of g : E → P1 . x2 x2 x2 where “trace” means the inﬁnite sum over all deck transformations. antiholomorphic) diﬀerentials on P1 . g∗ (d¯)) are global holomorphic (resp. so x dy ∧ dx1 dx2 ∧ =− x1 x2 =− C∗ ∆ε dg log f˜ dy ∧ g dg = d(y log f˜) ∧ g C∗ T f ∪T g ∪S f ∪S g log g d(y log f˜). The above integral becomes ε→0 103 lim − i 2 P1 ε g∗ (dz − d¯) z du . and hence are zero. . whence log g d(y log f˜) = −2πi Tg (g) d(y log f˜) = −2πi y log f˜ (g) . Similar arguments show ε→0 lim log g d(y log f˜) = −2πi Tf log g dy. Sg and an easy argument (replacing y by (y − y0 ) + y0 where y0 is the value of y at the center of a circle S ⊂ S f ) gives ε→0 lim log g d(y log f˜) = lim Sf ε→0 y log g Sf d f˜ = 2πi y log g f ( f˜) . and u is the stanε z dard parameter on P1 . so the integral vanishes. Note that g∗ (dz) (resp. u where P1 denotes P1 with εballs about 0 and ∞ removed. Now write ∆ε dy ∧ dx1 dx2 ∧ =− x1 x2 log f dy ∧ Eε dx2 . where (g) means integrals taken over paths from poles to zeros of g.Relative cycles and zeta functions To see this.
k j the above sum can be rewritten (replacing n by −n in the second sum) ε→0 lim ∆ε dy ∧ dx1 dx2 ∧ = +2πi y log(1 − t) − n x1 x2 log(1 − t) dy .3). Then ε→0 lim Im Φε du ∧ dx1 dx2 ∧ = −2π Re x1 x2 log(1 − t) du. so j lim ∆ε dy ∧ = 2πi j. Lemma (8.104 Combining these calculations ε→0 Lecture 8 lim ∆ε dy ∧ dx1 dx2 ∧ = 2πi y log g x1 x2 n∈Z. n + j.4) ε→0 lim ∆ε dx1 x1 dx ∧ ∧ dx1 dx2 ∧ = −i lim ε→0 x1 x2 =− E ∆ε dy ∧ dx1 dx2 ∧ . k m j log − α j β−1 qn = 0. x1 x2 Proof ∆ε dz ∧ dx2 x2 log f dz ∧ dg g = 0 by reason of type.k n∈Z dx1 dx2 ∧ x1 x2 −m j nk log α−1  log 1 − βk qn α−1 j j 2π m j nk log qn β−1  log 1 − α j β−1 qn − k k 2π log(1 − t) dy . on C∗ . The reader can easily reinterpret this integral as the desired integral n . n Proof ε→0 We have du ∧ dx1 dx2 ∧ x1 x2 = lim Im ε→0 Tf lim Im Φε log f log g du = −2π Re (f) log g du.k n∈Z Since for ﬁxed n.5) Let du = either dx or dy. ( f˜) We have (g) = ε→0 nk (β−1 qn ) and ( f˜) = k m j (α−1 ).k ( f˜) j − y log f (g) − log g dy . n This completes the proof of Lemma (8. Lemma (8.
and hence descend to functions on E. f (x). .5): ε→0 − log  t  2π . n x1 x2 ε→0 lim Im Φε −∆ε dx ∧ dx1 dx2 ∧ x1 x2 = Im i 2πi y log(1 − t) − 2πi n n log(1 − t) dy . ∞ Gq (t) = n=0 J(qn t) − n=1 J(qn t−1 ) (log  t )3 (log  t )2 log  t  log q − + + 3 log q 2 6 are continuous on C∗ and invariant under t → qt.3)–(8. t together with the equations lim Im Φε −∆ε dy ∧ dx1 dx2 ∧ = −2πy log 1 − t  . it is convenient to introduce two functions: D(t) = log  t  arg(1 − t) − Im 0 t log(1 − u) du .6) The periods associated to the cycle γ = γ f. −2π Re log(1 − t) dx = −2πy arg(1 − t) − 2π Re log(1 − t) dz. n n n In order to make things neater. g(x))} ⊂ E × P1 × P1 are −1 log  t  arg(1 − t) − Im n 4π2 −1 Im Pγ (dy) = 2 log  t  log 1 − t  . Im Pγ (dx) = Proof Simply use dt = 2πi dz and y = t resulting from Lemmas (8. n 4π where Pγ and Im Pγ are as deﬁned in Lecture 3.g = {(x. 105 log(1 − t) n dt . u J(t) = log  t  log 1 − t . Lemma (8. We ﬁrst descend these functions by a process of summation to functions on E.7) The functions Dq (t) = n∈Z ∞ D(qn t).Relative cycles and zeta functions Theorem (8.
1cycles on E × P1 × P1 → C. A = m j nk (b j − ak ). then the function (log  t )r vanishes on the nk = 0.) Proof The assertion for Im Pγ (dx) is immediate from Theorem (8. note that if nk (αk ) and m j (β j ) are divisors on C∗ with m j = β j m j = 1. and are left to the reader.1) D(t) = −D(t−1 ). so for all t we ﬁnd (8.106 Lecture 8 Proof We refer the reader to Bloch [1] for a proof that the function D(t) is well deﬁned. .8.7. For Im Pγ (dy). We n=0 have 1 d (1/t) dt dt + log 1 − = log(−t) . D(t) = 0.1) Q(γ) = Im Pγ (dx) − i Im Pγ (dy) . one easily checks that ∞ D(tqn ) converges. (We view Dq and Gq as functions on divisors in the obvious way. however. one deduces convergence of n<0 D(tqn ). αk nk = −1 divisor m j nk (αk β j ) for r ≤ 3. 4π2 Im Pγ (dy) = −1 Gq (A). For t < 0. From this. Im Pγ (dx) = −1 Dq (A). t t 1/t t t 1 du = Im (log(−x))2 + C + C log x. bk ∈ E (note where. log(−u) u 2 log(1 − t) Im where C accounts for the ambiguity in the branch of log(−t). n We consider now the complexvalued function Q: (8. The corresponding continuity and invariance properties of Gq are straightforward.) Granting this. Lemma (8. (At ﬁrst glance it would appear to depend on a choice of path from 0 to t. for example Dq ( ri (pi )) = ri Dq (pi ). 4π2 nk (bk ) with a j . writing ( f ) = m j (a j ) and (g) = change of notation). The reader can now easily check Gq (A) = log  t  log 1 − t  as claimed.8) With notation as above.6). Thus 1 D(t) + D(t−1 ) = − Im(log((−t))2 ) + arg(−t) log  t  + C + C log  t  2 = C + C log  t  with C accounting for the ambiguity in the branches of log(−t) and arg(1 − t).
(mτ + n)2 (m¯ + n) τ m. b) = − f (−a.3). viewed as a function on E. ci }. −b). f k+ τ N N N · { fi . −y2 N 5 fˆ(k. we have N deﬁned a relative algebraic 1cycle associated to the element S k+ τ ∈ Γ(E. We have −1 Q(γ f. one shows that Q(γ f.1) Writing Qq = Q(γ k+ τ ) = N −N Dq ((ρ)− ( f k+ τ )) − i Gq ((ρ)− ( f k+ τ )) .0) Proof As in the ﬁrst part of the proof of (8. ci ∈ C∗ . which we write as k+ τ . whence (8. this yields Q(γ k+ τ ) = N 4 Qq N k+ τ . Let this cycle be denoted γ k+ τ . ∗ ∗ N N 4π2 −1 (Dq 4π2 − i Gq ). Recall that given a point of order N on E. Q(γ k+ τ ) is deﬁned.9) N−1 k. b) e2πi a.c ) = 0 for c ∈ C∗ . K2 ) N S k+ τ = p. )Q(γ k+ τ ) = N 4π3 f (m. As noted earlier it is not completely well deN ﬁned. Our main result can be given a purely N analytic formulation as follows: ﬁx E = C/[1.n∈Z (m. N . ) = N Theorem (8.g ) = 2 Dq ((g) ∗ (g)− ) − i Gq ((g) ∗ ( f )− ) . and write 1 fˆ(k.b=0 −ak+b N . n) .Relative cycles and zeta functions 107 Notice that Q depends on the choice of a holomorphic diﬀerential dz = dx+i dy on E.n) (0. τ] with y = Im τ > 0. With notation as above. Let f : Z/NZ × Z/NZ → C be an odd function.9. one has Qq (a) − Qq a + k+ τ N + N 2 (N 2 − 1) Qq k+ τ − Qq (0) . that is f (a. 4π where with ( f ) = m j (a j ) and (g) = nk (bk ). However. g ∗ ( f )− := m j nk (bk − a j ). =0 N−1 f (a. N Q(γ k+ τ ) = N 2 N Na=0 a 0 Since Qq (0) = 0.
n) . n∈Z n 0 −i 4π3 N f (a. we view Qq . b) e2πi a. n 0 e2πi n/N n3 −i N = 4π3 n 0 iN f (0.k. so Qq (t) = i 1 i R (t) + 2 Rq (t) + 2 Rq (t). n3 Proof We use the wellknown Fourier series x3 x 2 x e2πi nx − + = 4π3 i 3 2 6 n3 n∈Z n 0 to write fˆ(k. and write Qq (t) = −1 (Dq (t) − i Gq (t)) 4π2 ∞ i log tqn  log(1 − tqn ) − = 2 4π n=0 + + 1 4π2 ∞ xqn ∞ n=1 log t−1 qn  log(1 − t−1 qn ) Im n=0 x−1 qn log(1 − t) dt t i (log  t )3 (log  t )2 log  t  log q − + .) Write the three terms in parentheses as Rq (t). −n) = 3 3 n 4π f (0. 2 6 4π2 3 log q (In what follows xqn  < 1 for n ≥ 0 and x−1 qn  < 1 for n ≥ 1. Rq (t). 4π2 q 4π 4π . and Rq (t) respectively. n3 It is convenient now to split up the function Qq . so there will be no ambiguity about branches of functions.10) N−1 Lecture 8 fˆ(k. ) k. For this. =0 3 3N 3 − 2 2N 2 + 6N = iN 4π3 n∈Z n 0 f (0. Gq as functions on C∗ . 3 3N 3 = = − 2 2N 2 + 6N e2πi n/N n3 −ak+b N −i 4π3 fˆ(k.b. ) k. ) k.108 Lemma (8. n) . Dq .
) Rq e2πi k. n) . N−1 ∞ fˆ(k. n3 = i y2 k. fˆ(k. ) =0 n=0 n+ N log 1 − e2πi ∞ k+ τ N +nτ − n−1 n− N log 1 − e2πi −k− τ N +nτ . =0 fˆ(k. .9) to be proven now reads N−1 (8. =0 4π2 y2 fˆ(k. =0 k+ τ N . ) 3N 3 − 2N 2 = −Ny2 4π3 n∈Z n 0 The formula in (8. ) 3N 3 3 3 − 2 4π2 y2 2N 2 + 6N 2 + 4π2 y2 6N f (0. To consolidate the two series. ) Rq e2πi k.12) L= M= 1 N −i N N−1 fˆ(k. ) Rq e2πi k+ τ N Proposition (8. and n . by N − k. (mτ + n)2 (m¯ + n) τ Put (8. m(mτ + n)2 m∈Z m 0 n∈Z Proof L= −2πy N k. n) .13) We have L= −y 2π f (m. ) Rq e2πi k+ τ N + i R e2πi = k+ τ N −y2 N π m. =0 k+ τ N = i 4π2 N−1 k. n) .11) k. N − .Relative cycles and zeta functions We have already seen i 4π2 N−1 109 fˆ(k. note that (replacing k. fˆ(k.n∈Z m 0 f (m.
) =0 n=0 n+ N log 1 − e2πi ∞ ∞ k+ τ N +nτ f (a. ) n=0 N−1 n+ ∞ N log 1 − e2πi k N +nτ k+ τ N +nτ + k=0 N−1 fˆ(k. Thus L= = −4πy N k. 4 sin2 πx Proof of lemma ne2πi nx = 1 e2πi x 1 d = 2πi x 2πi dx 1 − e (1 − e2πi x )2 1 −1 = πi x = .14) 4πy L= 2 N ∞ f (a.b n=0 m=1 m≡a(mod N) ne2πi m mτ+b N .b −ak+b N n+ n=0 m=1 e2πim N k+ τ N +nτ m . =0 fˆ(k. ) n=0 n+ N log 1 − e2πi k+ τ N +nτ . N) n=1 ∞ n log 1 − e2πi n log 1 − e2πi n=0 − k=0 N−1 fˆ(k. ∞ ne2π i nx = n=0 −1 . This can be rewritten (8. =0 fˆ(k. N−1 ∞ Lecture 8 − k. b) a.110 by n + 1). ) N ∞ log 1 − e2πi n+ −k− τ N +nτ = k=0 =1 N−1 n=0 ∞ N log 1 − e2πi k+ τ N +nτ = k. 0) ∞ k N +nτ = k.a. ) n=1 N−1 N n− fˆ(k.15) For Im x > 0. b) e2πi . −4πy N 2 k. −πi x )2 (e − e 4 sin2 π x . N−1 ∞ fˆ(k. =0 fˆ(k. Lemma (8.
As with L. m2 (mτ + n) x 0 Proof For  x  < 1. n) Im m. m2 m=1 ∞ mk . =0 n=0 m=1 e2πi k+ τ N +nτ m2 − n=1 m=1 e2πi m m2 .13). ) m=n=1 N−1 N e2πi m ∞ k+ τ N +nτ m2 e2πi m k+ τ N +nτ = k=0 =1 N−1 fˆ(k. ) Im k. −k− τ N +nτ Thus M fˆ(k.n=−∞ m 0 1 .14). m(mτ + n)2 M= f (m. b) a. =0 fˆ(k. n) .Relative cycles and zeta functions Returning to the expression for L (8. 0) e2πi N . b) a.b ∞ m=1 m≡a(mod N) 1 mτ+b n∈Z m( N + n)2 m. ) m=1 n=0 m2 k+ τ N +nτ ∞ = k. =0 fˆ(k. Then −1 2π ∞ n∈Z 1 (x + n)2 −y πN 2 −y 2π ∞ f (a.n=−∞ m 0 f (m.b m=1 m≡a(mod N) 1 .16) Let M be as in (8. the “natural” branch of = i N N−1 ∞ log(1 − t) dt is − t m ∞ ∞ xm m=1 m2 .12). we see L= −πy N2 ∞ 111 f (a. m sin2 π( mτ+b ) N Now use π2 = sin2 πx to write L= = This proves (8. Proposition (8. it will be convenient to combine the two sums: N−1 ∞ − k. ) m=1 n=0 e2πi m N−1 m2 − k=0 fˆ(k.
17) M1 = Proof 2i M1 = 2 N Note that also a. b) Im i a.b m=1 m≡a f (m. b) Im i a. N−1 ∞ Lecture 8 fˆ(k. 2 M1 = 2 N = Using the identity 2 N ∞ f (a. 0) Im k=0 e2πi N m2 m=1 ∞ mk = M1 + M2 .k. and −ak+b N Im ie2πi ( It follows that −ak+b N (∗) = −Im e2πi Im(∗) + Re e2πi −ak+b N Re(∗). n) Im m. = 2πix 2 1−e we can write M1 = = 1 N 1 N f (a. b) Re(e2πi ( ) ) ) = 0 because f (a. b) = − f (−a. b) Im a.b 1 N N−1 ∞ b=0 m=1 f (m.k. ) Im =0 n=0 m=1 e2πi k+ τ N +nτ m m2 − i N N−1 fˆ(k. b) e a. n=0 m=1 e2πi ∞ (m−a)k N +(nN+ ) mτ+b N m2 e2πin N m2 mτ+b ∞ f (a. −ak+b N 2πi −ak+b N ∞ Im n=0 m=1 e2πi k+ τ N +nτ m m2 . Lemma (8.112 so M= 2i N k. . f (a. b) 1 − N m2 mτ + b 1 cot π 2 N m .b. i cot πx + 1 1 .b m=1 n=0 m≡a(mod N) . b) Im a.b. m2 (mτ + n) f (a. b) 1 − 2π m2 ∞ f (m.n=−∞ m 0 1 . −b).b ∞ m=1 b m=1 m≡a(mod N) − cot π( mτ+b ) + i N m2 f (a.
m2 fˆ(k. b) a. Lemma (8. b) .a.18) Proof −i M2 = N −i = 2 N = = N−1 −1 M2 = N b=N−1 m=∞ m=1 b=0 f (m.Relative cycles and zeta functions Note that cot πx = and Im(cot πx) = 1 π Im n∈Z 113 1 1 lim . ∞ m=1 b f (m. b) Im i 2 N k. m2 We return now to the proof of (8.16). n) Im 0 n∈Z 1 m2 (mτ + n) . n) Im m∈Z m 0 n∈Z m2 (mτ This completes the computation of M1 . b) e k.b m=1 m≡a Im n∈Z 1 m2 mτ+b N 1 + n) +n . −1 2π m f (m. b) . so we get M1 = = 1 N 1 N ∞ m=1 b f (m. M = M1 + M2 = as desired.a.b −2πiak N Im ∞ e2πi N m2 m=1 k mk −1 e2πi N (m−a) f (a. 0) Im k=0 e2πi N m2 m=1 ∞ ∞ mk f (a.b m2 m=1 −1 N f (m. b) 1 − 2 2π m ∞ f (a. π N→∞ n=−N x + n N 1 x+n (the series converging without any special sort of summation). b) 1 − πN m2 f (m.
Sup.I. Cohomologie de SLn et valeurs de fonctions zeta aux points entiers.n∈Z m 0 f (m. American Mathematical Society.11) and also (8. [2] S. (4). =0 k+ τ k+ τ −y2 N fˆ(k. 373. 1977). 7 (1980). 2 y m(mτ + n) m (mτ + n) 1 1 1 + Im 2 2 y m(mτ + n) m (mτ + n) 1 1 1 1 − 2 + = 2 2 (mτ + n) 2iy m m(mτ + n) m (m¯ + n) τ 1 1 = − τ m(mτ + n)2 m(mτ + n)(m¯ + n) 1 = −2i y .9). Algebraic KTheory. Kyoto. Mass. 11. =0 k+ τ N + i Rq e2πi f (m. ) Rq (e2πi N ) + iR (e2πi N ) = π m. (mτ + n)2 (m¯ + n) τ so N−1 k.] [3] A.. Applications of the dilogarithm function in algebraic Ktheory and algebraic geometry. Tokyo (1978). pp.9): N−1 fˆ(k. n) . Kinokuniyo Book Store. 1987. Irvine (1978). [Second edition: Springer. Scuola Norm. n) k+ τ N = i N(L + M) = But −i yN 2π m∈Z m 0 n∈Z 1 1 1 + Im 2 . (mτ + n)2 (m¯ + n) τ This completes the proof of (8. 4 (1977). ) Rq e2πi k. Elliptic Functions. Reading.114 Lecture 8 We can now complete the proof of Theorem (8. (1973). 103–114 in Proceedings of the International Symposium on Algebraic Geometry (Kyoto Univ. Sci. Lectures given at the University of California. Pisa Cl. AddisonWesley. Ann. References for Lecture 8 [1] S. [4] S. errata. Bloch. Lang.] . [CRM Monograph Series. Providence. 613–636. Bloch. R. Borel.. 2000. Higher Regulators. no. New York. and Zeta Functions of Elliptic Curves.
9) leads to a regulator formula for the value at s = 2 of the zeta function of an elliptic curve E with complex multiplication by the ring of integers in an imaginary quadratic ﬁeld k with class number 1.n) (0. Notation will be as in Lecture 8.9 Relative cycles and zeta functions – continued In this ﬁnal lecture we will show how Theorem (8. together with the fact that the divisor (ρ) is invariant under z → −z.7.9.1) and (8. Lemma (9.1).2) Q(γ a+b ) = N i y2 N 4 4π3 sin(2π( an−bm )) N (m. valid without hypotheses on E. We have Gq (t−1 ) + Gq (t) = ∞ n=0 J(qn t−1 ) − ∞ ∞ J(qn t) n=1 ∞ n + n=0 J(q t) − n=1 2 J(qn t−1 ) − (log  t )2 = J(t−1 ) + J(t) − (log  t ) = 0. We begin with some general lemmas. Lemma (9.0) (m + nτ)2 (m + n¯ ) τ .1) Q(γ a+bτ ) = −Q(γ −a−bτ ). N N Proof Using formulas (8. we reduce to proving Gq (t−1 ) = −Gq (t). .
Then β δ a a ατ+β γτ+δ with αγ β δ ∈ SL2 (Z). N N (We write Qq and Qq to indicate dependence on q and q . . where k = β δ k . (k + τ )2 (k + τ ) ¯ . i y2 N 4 = 4π3 The following description of the “modular behavior” of Rq seemed worth including although it will not be used: Lemma (9. we ﬁnd that i y 2N4 4π3 i y 2N4 4π3 sin 2π a k. (a+m)−k(b+n) ) N a −m a = −m or or b −n b = −n = δ−a−bτ (m + nτ) with obvious notation. Ne Then a −bk N ) ga+bτ (m + nτ) = ˆ 1 N2 e2πi ( k. k + τ = (γτ + δ)−1 (k δ + β + (k γ + α)τ) = (γτ + δ)−1 (k + τ). Write q = e2πiτ and let τ Qq (γ a +b τ ) = (γ¯ + δ)−1 Qq (γ a+bτ ).116 Proof Let ga+bτ (k + τ) = Lecture 9 −bk 1 2πi ( a N ) . e2πi ( −kn+ m ) N 1 = 2 N = 0 1 e2πi ( k.) Proof Since SL2 (Z) preserves the symplectic form αγ −a + b k = −a + bk. −b k N (k + τ )2 (k + τ ) ¯ −bk sin 2π a N = Note that k. Writing fa+bτ = g−a−bτ − ga+bτ we get Q(γ a+bτ ) = N 1 2 −y2 N 5 fa+bτ (m + nτ) fˆ (k + τ)Q(γ k+ τ ) = a+bτ N 8π3 (m + nτ)2 (m + n¯ ) τ sin 2π (an−bm) N (m + nτ)2 (m + n¯ ) τ . Thus Qq (γ a +b τ ) = N 0 1 −1 0 .3) Let τ = b = α γ b .
ζ(k = 4π3 (k + τ)2 (k + τ) ¯ 2 4 iy N Im a + bτ. A. N ¯ yz = y1 z1 + y2 z2 B + y2 z1 τ + y1 z2 τ ¯ = (y1 z1 + y2 z2 B − y2 z1 A) + (y1 z2 − y2 z1 )τ. N Corollary (9.5) ζ Q(γ a+bτ ). Then xy = x1 y1 − x2 y2 B + (x1 y2 + x2 y1 − x2 y2 A)τ. where y is the complex conjugate. Lemma (9. −a +bk ) N a + bτ. ¯ ¯ Proof Write x = x1 +x2 τ. B ∈ Z. and suppose τ2 +Aτ+B = 0. We assume henceforth that o = Z + Zτ is an order in an imaginary quadratic ﬁeld κ. yi . −x1 y1 z2 + x1 y2 z1 + x2 y1 z1 + x2 y2 z2 B − x2 y2 z1 A x. z = z1 +z2 τ. x2 y2 Bz2 − x1 y1 z2 + x1 y2 z1 + x2 y1 z1 − x2 y2 Az1 xy. k + τ = 4π3 (k + τ)2 (k + τ) ¯ −1 = ζ Q(γ a+bτ ).4) xy. z = x. k + τ = e2πi( . a root of 1). then Q(γ ζ(a+bτ) ) = N −1 Proof Q(γ ζ(a+bτ) ) = N i y2 N 4 Im ζ(a + bτ). zi . τ N −bk sin 2π a N k. z = exp 2πi . (k + τ)2 (k + τ) ¯ It is convenient to denote by the bilinear form : o/No × o/No → C∗ . N If ζ ∈ oκ is a unit (i. yz = exp 2πi ¯ .e. and κ has class number 1. . N Assume now for simplicity that o is the ring of integers in κ. with xi .Relative cycles and zeta functions – continued whence (using y = Qq (γ a +b τ N 117 y (γτ+δ)(γ¯ +δ) ) τ i y2 N 4 (γτ + δ)2 (γ¯ + δ) τ )= 4π3 (γτ + δ)2 (γ¯ + δ)2 τ = (γ¯ + δ)−1 Qq (γ a+bτ ). y = y1 +y2 τ. yz . k + τ 4π3 (k + τ)2 (k + τ) ¯ 2 4 ¯ + τ) iy N Im a + bτ.
Then χ(x) = 0. y N y mod f y ⎧ ⎪ 0 ⎪ ⎪ ⎨ g = ⎪ g¯ ⎪ N ⎪ ⎩ y mod f χ(y ) x. and suppose N  f¯ x. g f Lemma (9. y ) run through representatives in o for the congruence classes modulo f (resp. ˆ 1 Proof (i) We have χ(x) = ˆ = 1 1 χ(y) x. f y 1 χ(y ) x. i y2 N 4 4π3 2 4 µ Im a + bτ. g ∈ oκ . ¯ otherwise. y mod g x. Then g ¯ µ∈o/go Q(γ a+bτ + f µ ) = Q(γ a+bτ ) . z = χ(¯ ) N z∈o/No N χ(z¯ ) x.6) Lecture 9 Suppose N = f g. χ(xy) = χ( x)χ(y). f¯(k + τ) (k + τ)2 (k + τ) ¯ Im a + bτ. with f. ¯ g  x. Then (i) χ(x) = 0 unless N  f¯x.7) Let χ have conductor f  N. (ii) If x ∈ (o/No)∗ . χ(xy) = ˆ 1 1 ¯ y χ(z) xy. g(r + sτ) ¯ (r + sτ)2 (r + s¯ )¯ 2 g τg = i g¯ y N g 4π3 r. yz = χ(¯ )χ(x). ¯ where the y (resp. g). r + sτ (r + sτ)2 (r + s¯ ) τ N = (¯ )−1 Q(γ g(a+bτ) ) g = (¯ )−1 Q(γ a+bτ ). y mod g x.118 Corollary (9. y ¯ ¯ y ˆ z .s Im g(a + bτ). n = f g. f¯(k + τ) = µ∈o/go g¯ g 0 g  k + τ. y g x. k + τ µ. Thus Q(γ a+bτ + f µ ) = N N µ∈o/go k. ˆ (ii) For x ∈ (o/No)∗ . N N f Proof We have µ.s i y2 N 4 = 4¯ π3 g r. f f1 . y = χ(y ) N y∈o/No N y mod f y x f¯. ˆ ¯ ¯ (iii) Let f1  f .
we can ﬁnd y ≡ 1 mod f¯ . (9. ¯ ¯ The character χ takes values in µκ and χ( x) = χ(x) (this follows from theorem o o p p 10. 1 with χ(¯ ) 1. χGr¨ ss ) = 4π3  µκ  χ(¯ )g ˆ g Q 2N5 −y S x χ(x)/N . p = (h). o ¯ χGr¨ ss (p) = x χ(x).5). where (x) = p N.8) (9. ¯ (o/No)∗ /µκ The ray class group (o/No)∗ /µκ acts on points of order N by x· y x−1 χ(x) y = N N . and (9. (9. Since κ has class number 1.7) we get o (9. an integer N. so we can rewrite (9.Relative cycles and zeta functions – continued 119 (iii) Since conductor χ = f f1 . Since N  f¯ · x y 1 χ(x) = χ(xy) = χ(¯ )χ(x). ˆ ˆ ¯ y ˆ Thus χ(x) = 0. ¯ ¯ 2N5 2N5 −y −y x∈(o/No)∗ x∈(o/No)∗ /µ κ The jinvariant j(E) is known to be real and to generate the Hilbert class ﬁeld of κ. y a unit in o/No.6). Assume as above that o = Z + Zτ is the ring of integers in an imaginary quadratic ﬁeld κ of class number 1.9) o L(2. ˆ We now have the tools we need for the regulator formula. and a character χ of (o/No)∗ which restricts to the given embedding µκ → C∗ on the roots of 1. j(E) ∈ Q so we can choose a model EQ o deﬁned over Q. From (8. which implies χGr¨ ss (¯ ) = χGr¨ ss (¯ )). Deuring’s theory associates to EQ a Gr¨ ssencharakter χGr¨ ss of o ∗ κ with values in κ . Let f generate the conductor ideal of χ and write N = f g.8) L(2. We ﬁx an embedding κ → C. 140 of Lang [4]. Let o o χGr¨ ss denote the Gr¨ ssencharakter o ¯ χGr¨ ss (p) = h χ(h) .9). p N. χGr¨ ss ) = 4π3 w χ(w)Q(γ N ) ˆ −y2 N 5 w∈o/No χ(x¯ )Q(γ x f¯−1 ) = ˆ g 4π3 χ(¯ ) ˆ g 2N5 −y χ( x)Q(γ x f¯−1 ) ¯ ¯ x∈(o/ f¯o)∗ = = 4π3 −y2 N 5 x∈(o/ f¯o)∗ ˆ g ˆ g 4π3  µκ  χ(¯ )g 4π3 χ(¯ )g x x χ( x)Q(γ N ) = ¯ ¯ χ( x)Q(γ N ). p.
Dick Gross suggests that the “corrected” foro o o mula would give L (0. We get o o Theorem (9. χGr¨ ss )−1 .12) (i) L(s. The cycle (9. (ii) Various renormalizations are possible and perhaps desirable to simplify the righthand constant in (9. References for Lecture 9 [1] S.11). N Also for τ = x + i y ∈ k. 2N5 −y o Remarks (9. Tokyo (1978). and hence is deﬁned over Q.8. ζEQ (deﬁned up to a ﬁnite number of factors by the product over all p ∈ Z such that E has nondegenerate reduction mod p of the zeta function of the corresponding curve over F p ). Bloch. χGr¨ ss ) is related to the zeta function of EQ . χGr¨ ss ) = ˆ g 4π3  µκ  χ(¯ )g Q(U). A Q. Kyoto. 289) yields  χ(¯ ) = ˆ g f f¯ . χGr¨ ss ) · ¯ = 2πQ (this amounts to modifying (8. 1977). by o ζEQ (s) = ζQ (s)ζQ (s − 1)L(s.10) U := (o/No)∗ /µκ γ x χ(x)/N ¯ is invariant under both these actions. χGr¨ ss ) ∈ Q(U) · Q.120 Lecture 9 and conjugation acts on these points in the natural way. . where ζ is a root of 1. pp. Let f be a generator for the conductor ideal and let f g = N ∈ Z. N whence  χ(¯ ) · g ˆ g ¯ f f¯g¯ g = 1. for example Lang [4]. χGr¨ ss ) ∈ ζ · Q(U) · Q. Since L (0. g ∈ o. let χGr¨ ss be the Gr¨ ssencharakter associated to EQ . so o ¯ L (0. one might take Q standard Gauss sum calculation (cf. 103–114 in Proceedings of the International Symposium on Algebraic Geometry (Kyoto Univ. we have y2 ∈ Q. Applications of the dilogarithm function in algebraic Ktheory and algebraic geometry.1) by 2π). χGr¨ ss ) ∈ π−2 L(2. p.. Then there exists a cycle U on E × P1 × P1 relative to E × # and deﬁned over Q such that o L(2.11) With notations as above. Kinokuniyo Book Store.
] [3] A. Higher Regulators.I. Lectures given at the University of California. 4 (1977). Cohomologie de SLn et valeurs de fonctions zeta aux points entiers.] . R. errata. [CRM Monograph Series. (1973). [Second edition: Springer. [4] S. 7 (1980). 11. Ann. Sci. Algebraic KTheory. Elliptic Functions. Bloch. no. Scuola Norm. 613–636. New York. Sup. (4). and Zeta Functions of Elliptic Curves. Borel.. Lang. 1987. Reading. 2000. 373. Pisa Cl. Providence.Relative cycles and zeta functions – continued 121 [2] S. Mass. AddisonWesley. Irvine (1978). American Mathematical Society.
.
Surfaces alg´ briques complexes. Springer.. K2 . Bull. e e Bloch. 349–379. 47 (1977). pp. ———. Irvine (1978). Tokyo (1978). M. Comm. Math. Mazur. 187–268 (1978). Berlin (1976). Lectures given at the University of California. 304–391. 3 (1975). ´ Sci. Algebra. ———. no. no. 215–228. Vari´ t´ s de Prym et jacobiennes interm´ diaires. Ast´ risque. Hirzebruch. K2 and algebraic cycles. Lecture Notes in Math. no. Formal groups arising from algebraic varieties. Lecture Notes in Math. Benjamin. 37 (1976). Ann. Topology. Berlin (1973). Math. New York (1968). and Zeta Functions of Elliptic Curves. . (2). Tate. of Math. A. 10 (1977). Algebraic KTheory. Bass. Ecole Norm. Hautes Etudes Sci. H. ´ ———. Ann. Publ. S..Bibliography Artin. Algebraic Ktheory and crystalline cohomology. 87–132. Applications of the dilogarithm function in algebraic Ktheory and algebraic geometry. Inst. Algebraic KTheory.. 99 (1974). ———. Kyoto. H. 10 (1977). 1–29 in Algebraic KTheory. Atiyah. Higher Regulators. 59 (1978). Analytic cycles on complex manifolds. and B. 103–114 in Proceedings of the International Symposium on Algebraic Geometry (Kyoto Univ. Torsion algebraic cycles.. 405–428. Kinokuniya Book Store. Math. ———. and F. Bass. Soc. 1977). Beauville. Sup. in Algebraic KTheory II. (4). M.. Invent. 342. 80 (1974).. K2 of Artinian Qalgebras with application to algebraic cycles. (4). An example in the theory of algebraic cycles. 551. 941–945. Springer. and J. Some elementary theorems about algebraic cycles on abelian varieties. ———. Amer. The Milnor ring of a global ﬁeld. pp. ———. Sci. and Brauer groups of function ﬁelds. ———. Sup. 1 (1962) 25–45. Ann. Ecole ee e Norm.
Acad. ´ Ann. On equivalence classes of cycles in an algebraic variety. Chevalley. Sup. Deligne. A. J. Compositio Math. ´ ———. A449–A452. Math. H.. ColliotTh´ l` ne. ColliotTh´ l` ne. La R´ quivalence sur les tores. pp.. e ———. bf 39 (1979). Paris (1971). and J. Sup. Providence. Sci. (2). . Clemens. 1970). Publ. 4 (1977). Inst. P. Compositio Math. Th´ orie de Hodge III. Bloch. and A. J. 287 (1978). American Mathematical Society. e e Chow. 7 (1974). 7 (1980). Bloch. 11. (2). 44 e (1974). e e 2 ann´ e. Sci. pp. 39 (1979). Series of notes on rational varieties ee and groups of multiplicative type. 33 (1976). (4). 282 (1976). A.L. 301–332. 1. I. C. L. Paris (1958). et al. 613–636. Ann. Griﬃths. 39 (1979). A967–A970. P.. vol. C. Sci. Zero cycles on surfaces with Pg = 0. Murre. Algebra. errata. S. Ann. 425–430 in Actes du Congr` s Internae e tional des Math´ maticiens (Nice. Lieberman. Chˆ telet.. Gersten’s conjecture and the homology of schemes. J. W. Th´ orie de Hodge. and D. 304–326. S. 1974). Ena seignement Math. Math. Some formulas pertaining to the Ktheory of commutative group schemes. no. Hautes Etudes Sci.] ———. 107–127.. ———. 64. Sansuc. R. S´ cr. Ecole Norm. Anneaux de Chow et applications. 181–201 (1975). 53 (1978). Compositio e e Math. of Math. 47–105. 175–229. C. Bloch. 281–356. S. Torsion algebraic cycles and a theorem of Roitman.J. Chevalley. Ogus. S´ minaire C. Coray. (2). AB. Sup. F. Points rationnels sur certaines courbes et surfaces cubiques.124 Bibliography [CRM Monograph Series. Cohomologie de SLn et valeurs de fonctions zeta aux points entiers.I. and D. ———. 5–77. (4).L. 2000. A. Sci. 284 (1977). Pisa Cl. and J. Compositio Math. 373. and P. R. A1113–A1116. 1 (1975). On the Chow groups of certain types of Fano threefolds. B.. Ann. Poids dans la cohomologie des vari´ t´ s alg´ briques. (4). 135–145. 95 (1972). Borel. 284 (1977). 153–170 (1960). Scuola Norm. 79–85 in Proee e ceedings of the International Congress of Mathematicians (Vancouver. no. vol. 5 (1959). GauthierVillars. The intermediate Jacobian of the cubic threefold.C. Paris Ser. A1215–A1218. Ecole Norm. of Math. L’´ quivalence rationnelle sur les ee e points ferm´ s des surfaces rationnelles ﬁbr´ es en coniques.. Ann.. 10 e (1977).. 450–479 (1956). Kas.
R. Springer.] Lang. II. Springer. Mass. R. 211– 243 in Algebraic KTheory I. III. 1980. Providence. ´ 2 Springer. S. 205–217. Interscience Publishers (1959). Ninth Nordic Summer School/NAVF Sympos. A. and M. Sijthoﬀ and Noordhoﬀ. [Pp. A. Publ. Y.. no. See also Introduction to Quadratic Forms over Fields. Thesis. Rational equivalence of 0cycles on some surfaces of general type with pg = 0. W. 46–188 in Dix expos´ s sur e la cohomologie des sch´ mas. Berlin (1977). Aarhus Universitet Preprint Series. Lecture Notes in Math. Gillet. Ann.] . Hartshorne. no. 90 (1969). 496–541. 147–167. in Algebraic KTheory II. These du 3 cycle. Lam. Universit´ de Paris VII (1979). Kleiman. 14 (1976). no. 359–386 in Dix expos´ s sur la cohomologie des sch´ mas. North Holland. W. 5–99. Oslo. pp. Inose. of Math... Harvard (1978). [Reprinted Springer. (1973). American Mathematical Society. H. MacPherson. M. Some exact sequences in the higher Ktheory of rings. Springer.. 342. Lecture Notes in Math. 569. 45 (1975). et al. R. e e Lam. On the de Rham cohomology of algebraic varieties.Bibliography 125 ———. Ann. Fulton. (2). Cohomologie etale (SGA 4 1 ).. and R. [Revised second printing. Publ. Th´ orie des intersections et th´ or` me de Riemann– e e e Roch (SGA 6). no. Wiley. Elman. Alphen aan den Rijn (1977).] Grothendieck. Lecture Notes in Math. Le groupe de Brauer I. [Reprinted 1994. New York (1978). 341. II.. and T. Inst. 45 (1975). ´ Fulton. 3. Reading. Y. T. no. The Algebraic Theory of Quadratic Forms. T. P. Math.. Applications of algebraic Ktheory to intersection theory. W. and J. e Grothendieck... Algebraic cycles and the Weil conjectures. L’equivalence rationelle des z´ ro cycles sur les surfaces al´ ebriques e g e e complexes a cup product surjectif. Berlin (1973). Hautes Etudes Sci. Lecture Notes in Math. H. Amsterdam (1968). I. Griﬃths. Berlin (1973). 1983. 90 (1969). no. Hautes ´ Etudes Sci. pp. On the periods of certain rational integrals. 244 (1979). Abelian Varieties.I. Fatemi. North Holland. Mizukami. 2005. S. Amsterdam (1968). Principles of Algebraic Geometry.] Gersten. On the quaternion symbol homomorphism gF : k2 (F) → B(F). Griﬃths. Benjamin. Harris. 1976). 225.. P. Inst. Intersecting cycles on an algebraic variety. Rational equivalence on singular varieties. Math. 460–495. A. 179–197 in Real and complex singularities (Proc. S. Math. Math. Berlin (1971). no. pp. no.
Quillen. P. Compositio Math. Grayson. Soc. 1987. Soc.S. Math. Mat. Berlin (1976). (1971). D. 557–570.126 Bibliography ———. ———. no. Ann. 1986. 78 (1956). 1165– 1199. 86 (128) (1971). D. A. A. Math. J. Sb. J. Rational equivalence of zerocycles on surfaces. Math. Springer. Invent. 161–206. USSRSb. Reading. Γequivalence of zerodimensional cycles (in Russian). (2). Amer. (N. 341. Rational equivalence of zerodimensional cycles (in Russian). Bull. New York. 255–267. ———. Princeton. 9 (1968). 59 (1954). Roitman. Higher Picard varieties. 75 (1969). (1973). 671. Ann. 555–567.. 776–779. Samuel. 90 (1968).). Introduction to Algebraic KTheory.. 1970).. I. 217– 240 in Algebraic KTheory. 89 (131) (1972). Mass. Proc. Lecture Notes in Math. 21 (1969).. Higher algebraic Ktheory. Springer.S.. Princeton University Press.. Quillen. Le groupe de Brauer–Grothendieck en g´ om´ trie diophantienne. ———. J. pp. 569–585. 25 (1972). [Translation: Math. no. Amer. Sb. GauthierVillars. AddisonWesley. Kyoto Univ. 551. 15 (1971). e e pp. N.. D. 571–588. 383–400. Lecture Notes in Math.] Lieberman. J. A.). Manin. of Math. M. Algebraic equivalence modulo rational equivalence on a cubic threefold. vol. 9 (1970).] Rosenlicht. in Algebraic KTheory I.] ———. . Math. Berlin (1973). On the symmetric product of a rational surface. II.] Mattuck. Cubic Forms. [Second edition.. [Second edition: Springer. Generalized Jacobian varieties. Murre. Annals of Mathematics Studies. Higher algebraic Ktheory. and D. Math. Mat. (N. 401–411 in Actes du Congr` s International Math´ maticiens (Nice. Ruled surfaces and the Albanese mapping.. 683–688. Amer. Amsterdam (1974). Nakayama. Mumford. T. 195–204.. I. 18 (1974). Amer.. J. Cohomology of class ﬁeld theory and tensor product modules I. [Translation: Math USSRSb. Yu. Paris (1971). 1.. 318–344. of Math. Elliptic Functions. Math. D. Milnor. 505– 530. P.J. Algebraic Ktheory and quadratic forms. e e vol. Rational equivalence of arbitrary cycles. North Holland. 65 (1957). 72.
Hermann. 149–168 in Journ´ e de G´ om´ trie Algebrique de Rennes (Rennes. 1979. 7 (1979). Hermann.P. Berlin (1968). Pergamon Press. Dualit´ dans la cohomologie des espaces localement compacts. Tennison. pp. 257–274. 27 (1972). no. Proc. Wallace. Utrecht (1978). Multiplicit´ s.] Verdier. Hermann. 36 (1976). Deformations of the second Chow group. Homology Theory on Algebraic Varieties. o ———. Foundations of Algebraic Geometry.L. 1984. J. second edition. expos´ no. [Translation: Local Fields. Groupes algebriques et corps de classes. Asterisque. Algebraic KTheory.I. [Second edition. no. Tyurin. Colloquium Publications. no. Paris (1959).. 29 (1974).. A. 1975. [Translation: Russian Math. 184–198 in Proceedings of a conference on local ﬁelds (Driebergen. (1946). Soc. vol. Universidad Nacional Aut´ noma de Mexico and UNESCO. Five lectures on threedimensional varieties (in Russian). C. ee e pp. Springer. 27 (1972). Lecture Notes in Math. Corps Locaux. N. Providence. Tate. 11. On the quartic threefold. Algebra. e ee e . R. II. no. [Reprinted e e Soci´ t´ Math´ matique de France. 1995. J. Invent. Courbes Alg´ briques et Vari´ t´ s Ab´ liennes. e e Berlin (1965). (3). Springer. London Math. American Mathematical Society. Berlin (1967). Springer. reprinted. Mexico City (1958). Surveys. 24–53 in Symposium Internacional de Topologia Algebraica. 999–1027. pp.] ee e ———. 5. New York.S. Uspehi Mat. Alg` bre locale. ———.] Sherman. J. Thesis. The formal completion of the second Chow group: a Ktheoretic approach. Math. 1–53. ———. Paris (1971). Paris (1968). J. Weil. Springer. Relations between K2 and Galois cohomology. 337–349. 300 (1965). A.] ———. New York (1958). 29.Bibliography 127 Serre. A duality theorem in the etale cohomology of schemes. e S´ minaire Bourbaki. B. (167) 3–50. Sur la topologie des vari´ t´ s algebriques en caract´ ristique p. 9. ———. A. Vol. Nauk. Vol. R. Stienstra. A. Swan. Lecture Notes in Math. Comm. Paris. 64 (1979). 5.. e e e 1978).M. 76. R. 714–734. 1966). Kcohomology of regular schemes.
.
3 algebraically equivalent to zero. 5 algebraic surface. 80 Beilinson–Soul´ conjecture. 73 Artin–Hasse zeta function. 51 Brauer group. viii Bloch–Ogus spectral sequence. category of. 46 . 3 additive group formal. 11. 79 Chow motive. 63 crystalline cohomology.F ). 16 Bloch–Kato conjecture on special values of Lfunctions. 21 ﬁnitely generated projective modules. xix Bloch–Kato–Milnor conjecture. 81. 119 devissage theorem. 88 divisor. 97 artinian kalgebra. xii. 118 conic bundle. xviii Div(X0. 48 Chern character. 100 ´ etale cohomology. 80. 45 Fano surface. 66 corestriction. 73 e diophantine questions. Albanese variety of X. 71 bad reduction. x Eichler norm theorem. 5. 9 Albanese kernel. 76 cubic surface. 66 o o χGr¨ ss . 50 exact category. 97 cyclic homology. 69. Chow group. xi e big Witt ring. Gr¨ ssencharakter. 59 Albanese variety. 16 ﬁnitedimensional Chow group. 27 Abel–Jacobi theorem. 89 cycle class map. 21 higher. 17 Bloch’s conjectured ﬁltration on Chow groups. 29. ix Ar (X). xvii degree map. 4 Chˆ telet surface. 9 applications. 41. 119 Chow group. 73 Bloch’s conjecture on zerocycles. 94 CHr (X). 3 rationally equivalent to zero. 74 admissible monomorphism or epimorphism. 89 coniveau ﬁltration. 14 ﬁltration on Chow groups. 80. xi conductor. 3 eﬀective motive. xvii. xii. 48 de Rham cohomology. 52 Bockstein map. 89 a characteristic sequence theorem. 4 ﬁnite dimensional.Index A1 homotopy. xi of zerocycles. 50 Dieudonn´ ring. 90 elliptic curve. xv cycle map. 46 Alb(X). xx Artin–Hasse exponential. 5 eﬀective. 79 Deuring’s theory. 4 Abel–Jacobi map. 5. 66 Borel–Moore cohomology theory. 9 algebraic cycle. 83.
formal additive group. 14 e period. 25. 105 Pic(S ). 70 T (Y). 25. xii u localization theorem. 61 Milnor ring. 98 valuation map. 37 Gm . 13 good reduction. xii Index N´ ron–Severi group. 120 resolution theorem. xv. xvii. 3 Roitman’s theorem. multiplicative group. Zariski sheaf. xiii. N´ ron–Severi group of S . ix Ktheory sheaf analytic. 34 weight ﬁltration. x mixed. xviii. 37 Kcohomology. 5. 119 o Griﬃths group. 91 geometric motive. xiv. 65 “trivial” curve. 74 Ga . 5. xv. 37 formal. xviii singular cohomology. 74 Galois symbol. 52. xiv. 4 intermediate. 5 hyperelliptic surface. Serre’s. 80 e nerve (of a small category). 45 rational surface. xiv. 52 Hasse–Weil Lfunction. xiv. xvii inﬁnitesimal structure. xvi Grothendieck’s standard conjectures. viii with transfers. xvi. 70 Ktheory. 4. 89 Milnor–Bloch–Kato conjecture. 25. 64 transfer map. 47 symbol map. xx quartic threefold. 45 Milnor. xviii . 88 vanishing cycle. xii motive Chow. xix Hodge ﬁltration. xii motivic cohomology. analytic Ktheory sheaf. Albanese kernel of Y. 27 intermediate jacobian. xx multiplicative group. 40 Hodge structure polarized of weight two. 74 Gm . x geometric. 72 Prym variety. 85 regulator. 18 torsion in the Chow group. xiv. xii Nisnevich topology. 46 presheaf. Quillen’s. 48 Riemann–Roch theorem. 14 plus construction. xv. 48 NS(S ). 59 scissors conguence. 76 Zariski sheaf. 13 incidence correspondence. 79 split. xii H. 14 Picard variety. 63. 27 quantum ﬁeld theory. xi eﬀective. ix. 13 Murre’s conjecture. 28 jacobian. 69 Godeaux surface. 59 ´ trace morphism (on etale cohomology). viii prorepresentable functor. 62 zeta function. 45 nilpotence conjecture. 5. 48 power torsion subgroup. xvi. 71 a K¨ nneth projectors. 69 K¨ hler diﬀerential. 25 Quillen Ktheory. 52 Soule’s conjecture on zeta functions. 11 tame symbol. Picard variety of S. 37 isogeny. 98 theorem of the hypersquare. 50 site. 86 Gr¨ ssencharakter. 40 mixed motive. 69 an K2 . 74 Mumford’s theorem. xx split rational surface. 47 Steinberg symbol. x Gersten resolution. xvi mixed Hodge structure. 88. 14 e N´ ron–Severi torus.130 formal curve. xvii multizeta numbers. 50. 85 Steinberg group. ix noetherian scheme. formal multiplicative group. xix regulator formula. 40 Witt vector cohomology. 91. xv.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.