brauer groups, tamagawa measures, and rational points
brauer groups, tamagawa measures, and rational points
brauer groups, tamagawa measures, and rational points
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
BRAUER GROUPS,<br />
TAMAGAWA MEASURES,<br />
AND RATIONAL POINTS<br />
ON ALGEBRAIC VARIETIES<br />
Jörg Jahnel
Jörg Jahnel<br />
Mathematisches Institut, Bunsenstraße 3–5, D–37073 Göttingen, Germany.<br />
E-mail : jahnel@uni-math.gwdg.de<br />
Url : http://www.uni-math.gwdg.de/jahnel<br />
2000 Mathematics Subject Classification. —<br />
11G35, 14F22, 16K50, 11-04, 14G25, 11G50.<br />
Acknowledgements. I wish to acknowledge with gratitude my debt to Y. Tschinkel. Most of the work described in this<br />
book was initiated by his numerous mathematical questions. During the years he spent at Göttingen, he always shared<br />
his ideas in an extraordinarily generous manner.<br />
I further wish to express my deep gratitude to my friend <strong>and</strong> collegue Stephan Elsenhans. He influced this book in<br />
many ways, directly <strong>and</strong> indirectly. It is no exaggeration to say that most of what I know about computer programming,<br />
I learned from him. The experiments which are described in Part C were carried out as a joint work. I am also indebted<br />
to Stephan for proofreading.<br />
Special thanks go to my parents, Regina <strong>and</strong> Alfred Jahnel, for their unflaggingbelief that, despite their incomprehension<br />
about what I do, I must be doing that very well. Finally, I would like to thank all the people who constantly supported<br />
<strong>and</strong> encouraged me over the years. I risk doing someone a disservice by not naming all of them here, but plead<br />
paucity of space. Let me mention particularly Ulrich Stuhler, H.-S. Holdgrün, Thomas Fröhlich, Stefanie Schmidt,<br />
<strong>and</strong> Carsten Thiel.<br />
The computerpart of the workdescribed in this bookwas executed on theSun Fire V20z Serversof the GaußLaboratory<br />
for Scientific Computing at the Göttingen Mathematical Institute. The author is grateful to Y. Tschinkel for the<br />
permission to use these machines as well as to the system administrators for their support.<br />
J. J.<br />
Göttingen, Lower Saxony,<br />
September 24, 2008
BRAUER GROUPS, TAMAGAWA MEASURES,<br />
AND RATIONAL POINTS<br />
ON ALGEBRAIC VARIETIES<br />
Abstract. — We study existence <strong>and</strong> asymptotics of <strong>rational</strong> <strong>points</strong> on algebraic<br />
varieties of Fano <strong>and</strong> intermediate type. In the first part, we study the various<br />
versions of the Brauer group. We explain why a Brauer class may serve as an<br />
obstruction to weak approximation or even to the Hasse principle. In the second<br />
part, we discuss to some extent the concept of a height <strong>and</strong> formulate Manin’s<br />
conjecture on the asymptotics of <strong>rational</strong> <strong>points</strong> on Fano varieties. The final<br />
part describes numerical experiments.<br />
Habilitationsschrift<br />
vorgelegt von<br />
Jörg Jahnel<br />
aus Eisenberg<br />
Göttingen, Sommer 2008
CONTENTS<br />
Table of contents . . ....................................................<br />
iii<br />
Notation <strong>and</strong> fundamental conventions . . .............................. vii<br />
Notation <strong>and</strong> conventions . . ............................................ vii<br />
References <strong>and</strong> Citation . . .............................................. ix<br />
Introduction . . ..........................................................<br />
xi<br />
Part A. The Brauer group . . ............................................ 1<br />
I. The Brauer-Severi variety associated with a central simple algebra . . 3<br />
1. Introductory remarks . . .............................................. 3<br />
2. Non-abelian group cohomology . . ................. . . ............... 5<br />
3. Galois descent . . . ......................... .......................... 8<br />
4. Central simple algebras <strong>and</strong> non-abelian H 1 . . . . . . . . . . . . . . . . . . . . . . . . 22<br />
5. Brauer-Severi varieties <strong>and</strong> non-abelian H 1 . . ........................ 28<br />
6. Central simple algebras <strong>and</strong> Brauer-Severi varieties . . . ............... 35<br />
7. Functoriality . . ........................... ........................... 39<br />
8. The functor of <strong>points</strong> . . .............................................. 53<br />
II. On the Brauer group of a scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61<br />
1. Azumaya algebras . . ........................ ........................ 61<br />
2. The Brauer group . . .................................................. 65<br />
3. The cohomological Brauer group . . .................................. 66<br />
4. The relation to the Brauer group of the function field . . . . . . . . . . . . . . 71<br />
5. The Brauer group <strong>and</strong> the cohomological Brauer group . . . . . . . . . . . . 74<br />
6. Orders in central simple algebras . . ................. ................. 76<br />
7. The Theorem of Ausl<strong>and</strong>er <strong>and</strong> Goldman . . . . . . . . . . . . . . . . . ......... 84
iv<br />
CONTENTS<br />
8. Examples . . .......................................................... 86<br />
III. An application: The Brauer-Manin obstruction . . .................. 99<br />
1. Adelic <strong>points</strong> . . ...................................................... 99<br />
2. The Brauer-Manin obstruction . . .................................... 103<br />
3. Technical lemmata . . ........................ ........................ 106<br />
4. Computing the Brauer-Manin obstruction – The general strategy . . 110<br />
5. The examples of Mordell . . .......................................... 113<br />
6. The “first case” of diagonal cubic surfaces . . .......................... 128<br />
Part B. Heights . . ...................................................... 147<br />
IV. The concept of a height . . ............................................ 149<br />
1. The naive height on the projective space overÉ. . . . . . . . . . . . . . . . . . . . 149<br />
2. Generalization to number fields . . . ................ ................. 151<br />
3. Geometric interpretation . . .......................................... 155<br />
4. Basic arithmetic intersection theory . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . 160<br />
5. An intersection product on singular arithmetic varieties . . . . . . . . . . . . 165<br />
6. The arithmetic Hilbert-Samuel formula . . ............................ 169<br />
7. The adelic Picard group . . ............................................ 173<br />
V. On the distribution of small <strong>points</strong> on abelian <strong>and</strong> toric varieties . . 189<br />
1. Introduction . . ...................................................... 189<br />
2. The dynamics of f . . ................................................ 192<br />
3. Perturbing almost semiample metrics . . .............................. 199<br />
4. Equidistribution . . ......................... ......................... 204<br />
VI. Conjectures on the asymptotics of <strong>points</strong> of bounded height . . .... 211<br />
1. A heuristic . . ........................................................ 211<br />
2. The conjecture of Lang . . ............................................ 214<br />
3. The conjecture of Batyrev <strong>and</strong> Manin . . .............................. 216<br />
4. The conjecture of Manin . . .......................................... 220<br />
5. Peyre’s constant . . .................................................... 224<br />
Part C. Numerical experiments . . ...................................... 241<br />
VII. Points of bounded height on cubic <strong>and</strong> quartic threefolds . . ...... 243<br />
1. Introduction — Manin’s Conjecture . . ................................ 243<br />
2. Computing the Tamagawa number . . ................................ 247
CONTENTS<br />
v<br />
3. On the geometry of diagonal cubic threefolds . . . . . . . . . . . . . . . . . . . . . . 252<br />
4. Accumulating subvarieties . . ........................................ 254<br />
5. Results . . ............................................................ 259<br />
VIII. On the smallest point on a diagonal quartic threefold . . . . . . . . . . 265<br />
1. A computer experiment . . .......................................... 265<br />
2. A negative result . . .................................................. 268<br />
3. The fundamental finiteness property . . .............................. 270<br />
IX. General cubic surfaces . . ............................................ 281<br />
1. Rational <strong>points</strong> on cubic surfaces . . .................................. 281<br />
2. Background . . ........................................................ 284<br />
3. The Galois operation on the 27 lines . . .............................. 287<br />
4. Computation of Peyre’s constant . . .................................. 290<br />
5. Numerical Data . . .................................................... 292<br />
6. A concrete example . . ................................................ 297<br />
X. On the smallest point on a diagonal cubic surface . . ................ 301<br />
1. Introduction . . ...................................................... 301<br />
2. The factors α <strong>and</strong> β . . ....................... . ...................... 305<br />
3. Splitting the Picard group . . .......................................... 306<br />
4. A technical lemma . . ................................................ 310<br />
5. A negative result . . .................................................. 311<br />
6. The fundamental finiteness property . . .............................. 316<br />
XI. The Diophantine Equation x 4 + 2y 4 = z 4 + 4w 4 . . . . . . . . . . . . . . . . . . 329<br />
1. Introduction . . ...................................................... 329<br />
2. Congruences . . ...................................................... 330<br />
3. Naive methods . . .................................................... 332<br />
4. An algorithm to efficiently search for solutions . . .................... 332<br />
5. General formulation of the method . . ................................ 335<br />
6. Improvements I – More congruences . . .............................. 336<br />
7. Improvements II – Adaption to our hardware . . . . . . . . . . . . . . . . . . . . . . 341<br />
8. The solution found . . ................................................ 348<br />
XII. New sums of three cubes . . ........................................ 349<br />
1. Introduction . . ...................................................... 349<br />
2. Elkies’ method . . .......................... .......................... 350<br />
3. Implementation . . .................................................... 350
vi<br />
CONTENTS<br />
4. Results . . ............................................................ 351<br />
Appendix . . .............................................................. 353<br />
1. A script in GAP . . .................................................... 353<br />
2. The list . . ............................................................ 356<br />
Bibliography . . .......................................................... 363<br />
Index . . ................................. ................................. 383
NOTATION AND FUNDAMENTAL<br />
CONVENTIONS<br />
Notation <strong>and</strong> conventions<br />
We follow st<strong>and</strong>ard notation <strong>and</strong> conventions from Algebra, Algebraic Number<br />
Theory, <strong>and</strong> Algebraic Geometry. More precisely:<br />
i) We writeÆ,,É,Ê, <strong>and</strong>for the sets of natural numbers, integers, <strong>rational</strong><br />
numbers, real numbers, <strong>and</strong> complex numbers, respectively.<br />
ii) For a group G <strong>and</strong> elements σ 1 , . . . , σ n ∈ G, we denote the subgroup<br />
generated by σ 1 , . . . , σ n by 〈σ 1 , . . . , σ n 〉 ⊆ G.<br />
If G is abelian then G n ⊆ G is the subgroup consisting of all elements of torsion<br />
dividing n.<br />
iii) If a group G operates on a set M then M G denotes the invariants. We write<br />
M σ instead of M 〈σ〉 .<br />
iv) All rings are assumed to be associative.<br />
v) If R is a ring then R op denotes the opposite ring. I.e., the ring that coincides<br />
with R as an abelian group but in which one has xy = z when one had yx = z<br />
in R.<br />
vi) For R a ring with unit, R ∗ denotes the multiplicative group of invertible<br />
elements in R.<br />
vii) All homomorphisms between rings with unit are supposed to respect the<br />
unit elements.<br />
viii) By a field, we always mean a commutative field. I.e., a commutative ring<br />
with unit every non-zero element of which is invertible. If K is a field then K sep<br />
<strong>and</strong> K denote a fixed separable closure <strong>and</strong> a fixed algebraic closure, respectively.<br />
ix) A ring with unit every non-zero element of which is invertible is called a<br />
skew field.<br />
x) If R is a commutative ring with unit then an R-algebra is always understood<br />
to be a ring homomorphism j : R → A the image of which is contained in the<br />
center of A. An R-algebra j : R → A is denoted simply by A when there seems
viii<br />
NOTATION AND FUNDAMENTAL CONVENTIONS<br />
to be no danger of confusion. An R-algebra being a skew field is also called a<br />
division algebra.<br />
xi) If σ : R → R is an automorphism of R then A σ denotes the R-algebra<br />
R −→ σ<br />
R −→ j<br />
A. If M is an R-module then we put M σ := M ⊗ R R σ . M σ is an<br />
R σ -module as well as an R-module.<br />
xii) All central simple algebras are assumed to be finite dimensional over a<br />
base field.<br />
xiii) For K a number field, we write O K to denote the ring of integers in K.<br />
If ν ∈ Val(K ) is a non-archimedean valuation ν ∈ Val(K ) then the ν-adic<br />
completion of K is denoted by K ν <strong>and</strong> its ring of integers by O Kν .<br />
In the particular case that K =É, we denote by ν p the normalized p-adic<br />
valuation corresponding to a prime number p.<br />
xiv) For R a commutative ring, we denote by SpecR the affine scheme constituted<br />
by its spectrum.<br />
xv) The projective space of relative dimension n over a scheme X will be denoted<br />
by P n X . We omit the subscript when there is no danger of confusion.<br />
xvi) If X is a scheme over a scheme T <strong>and</strong> Y is a T -scheme then we also write<br />
X Y for the fiber product X × T Y. If Y = SpecR is affine then write X R instead<br />
of X SpecR .<br />
xvii) For X a scheme over a scheme T , we denote by X t the fiber of X<br />
over t ∈ T .<br />
If C is a scheme over the integer ring O Kν of the completion of the number<br />
field K with respect to the valuation ν then we write C ν for the special fiber.<br />
In the particular case ν = ν p , we write C p instead of C νp .<br />
If C is a scheme over the integer ring O of a number field K then we use same<br />
the notation, not mentioning the base change to O Kν .<br />
xviii) For R any commutative ring, A a commutative R-algebra, <strong>and</strong> X an<br />
R-scheme, a morphism x : SpecA → X of R-schemes is also called an A-valued<br />
point on X. If A is a field then we also adopt more conventional language <strong>and</strong><br />
speak of a point defined or <strong>rational</strong> over A. The set of all A-valued <strong>points</strong> on X<br />
will be denoted by X (A).<br />
xix) If C is a scheme over a valuation ring O <strong>and</strong> x ∈ C (O) then the reduction<br />
of x is denoted by x.
NOTATION AND FUNDAMENTAL CONVENTIONS<br />
ix<br />
References <strong>and</strong> Citation<br />
When we refer to a definition, proposition, theorem, etc., in the same chapter<br />
then we simply rely on the corresponding numbering within the chapter.<br />
Otherwise, we add the number of the chapter.<br />
For the purpose of citation, the articles <strong>and</strong> books being used are encoded in<br />
the manner specified by the bibliography. In addition, we mostly give the<br />
number of the relevant section <strong>and</strong> subsection or the number of the definition,<br />
proposition, theorem, etc. Normally, we do not mention page numbers.
INTRODUCTION<br />
Here, in the midst of this sad <strong>and</strong> barren l<strong>and</strong>scape of the Greek accomplishments<br />
in arithmetic, suddenly springs up a man with youthful energy:<br />
Diophantus. Where does he come from, where does he go to? Who were his<br />
predecessors, who his successors? We do not know. It is all one big riddle.<br />
He lived in Alex<strong>and</strong>ria. If a conjecture were permitted, I would say he was<br />
not Greek; . . . if his writings were not in Greek, no-one would ever think<br />
that they were an outgrowth of Greek culture . . . .<br />
HERMANN HANKEL (1874, translated by N. Schappacher)<br />
Diophantine equations have a long history. More than two thous<strong>and</strong> years ago,<br />
Diophantus of Alex<strong>and</strong>ria considered, among many others, the equations<br />
x 2 + y 2 = z 2 ,<br />
(∗)<br />
y(6 − y) = x 3 − x ,<br />
<strong>and</strong><br />
y 2 = x 2 + x 4 + x 8 .<br />
In Diophantus’ book “Arithmetica”, we find the formula<br />
(<br />
(p 2 − q 2 )λ, 2pqλ, (p 2 + q 2 )λ ) (†)<br />
which generates infinitely many solutions of (∗). For the second <strong>and</strong><br />
third of the equations mentioned, Diophantus gives particular solutions,<br />
namely (1/36, 1/216) <strong>and</strong> (1/2, 9/16), respectively.<br />
In general, a polynomial equation in several indeterminates where solutions<br />
are sought in integers or <strong>rational</strong> numbers, is called a Diophantine equation,<br />
in honour of Diophantus. Diophantus himself was interested in solutions in<br />
positive integers or positive <strong>rational</strong> numbers. Contrary to the point of view<br />
usually adopted today, he did not accept negative numbers.<br />
It is remarkable that algebro-geometric methods have often been fruitful in order<br />
to underst<strong>and</strong> a Diophantine equation. For example, there is a simple geometric<br />
idea behind formula (†).
xii<br />
INTRODUCTION<br />
Indeed, since the equation is homogeneous, it suffices to look for solutions<br />
of X 2 + Y 2 = 1 in <strong>rational</strong>s. This equation defines the unit circle. For every<br />
t ∈Ê, there is the line “x = −ty + 1” going through the point (1, 0).<br />
An easy calculation shows that the second point where this line meets the<br />
unit circle is given by ( 1 − t<br />
2<br />
2t<br />
)<br />
1 + t 2, . (‡)<br />
1 + t 2<br />
As every point on the unit circle may be connected with (1, 0) by a line,<br />
one sees that the parametrization (‡) yields every <strong>rational</strong> point on the circle<br />
(except for (−1, 0) for which the form of line equation given is not adequate).<br />
Consequently, formula (†) delivers essentially every solution of equation (∗),<br />
a fact which was seemingly not known to the ancient mathematicians. The morphism<br />
P 1 −→ P 2<br />
(p : q) ↦→ ( (p 2 − q 2 ) : 2pq : (p 2 + q 2 ) )<br />
provides a <strong>rational</strong> parametrization of the plane conic C given by the equation<br />
x 2 + y 2 = z 2 in P 2 . Essentially the same method works for every conic in<br />
the plane. Further, it may be extended to several classes of singular curves of<br />
higher degree.<br />
Every Diophantine equation defines an algebraic variety X in an affine or projective<br />
space. There is a one-to-one correspondence between solutions of the<br />
Diophantine equation <strong>and</strong>É-<strong>rational</strong> <strong>points</strong> on X. We will prefer geometric<br />
language to number theoretic throughout this book.<br />
The cases when there is an obvious <strong>rational</strong> parametrization are in some sense<br />
best possible. But even when there is nothing like that, algebraic geometry often<br />
yields a guideline of which behaviour to expect, whether there will be no, a few<br />
or many solutions.<br />
The Kodaira classification distinguishes between Fano varieties, varieties of intermediate<br />
type, <strong>and</strong> varieties of general type (at least under the additional<br />
assumption that X is non-singular). It does not use any specifically arithmetic<br />
information, but only information about X as a complex variety. Nevertheless,<br />
there is overwhelming evidence for a strong connection between the<br />
classification of X according to Kodaira <strong>and</strong> its set of <strong>rational</strong> <strong>points</strong>.<br />
To make a vague statement, on a Fano variety, there are infinitely many <strong>rational</strong><br />
<strong>points</strong> expected while, on a variety of general type, there are only finitely many<br />
<strong>rational</strong> <strong>points</strong> or even none at all. More precisely, there is the conjecture<br />
that, on a Fano variety, there are always infinitely many <strong>rational</strong> <strong>points</strong> after a<br />
suitable finite extension of the ground field. On the other h<strong>and</strong>, for varieties of<br />
general type, there is the conjecture of Lang. It states that there are only finitely
INTRODUCTION<br />
xiii<br />
many <strong>rational</strong> <strong>points</strong> outside the union of all closed subvarieties which are not<br />
of general type.<br />
Another method to analyze a Diophantine equation is given by congruences.<br />
Kurt Hensel provided a more formal framework for this method by his invention<br />
of the p-adic numbers. As one is working over local fields, this might be<br />
called the local method.<br />
Consider, for example, the Diophantine equation<br />
x 3 + 7y 3 + 49z 3 + 2u 3 + 14v 3 + 98w 3 = 0 . (§)<br />
It has no solution inÉ6 except for (0, 0, 0, 0, 0, 0). This may be seen by a repeated<br />
application of an argument modulo 7. A more formal reason is the fact that the<br />
projective algebraic variety defined by (§) has no point defined overÉ7.<br />
One might ask whether or to what extent solvability overÉp for every prime<br />
number p together with solvability in real numbers implies the existence of<br />
a <strong>rational</strong> solution. This question has been very inspiring for research over<br />
many decades. As early as 1785, A.-M. Legendre gave an affirmative answer for<br />
equations of the type<br />
q(x, y, z) = 0<br />
where q is a ternary quadratic form. Legendre’s result was generalized<br />
to quadratic forms in arbitrarily many variables by Hasse <strong>and</strong> Minkowski.<br />
The term “Hasse principle” was coined to describe the phenomenon.<br />
A totally different sort of examples where the Hasse principle is valid is provided<br />
by the circle method originally developed by G. H. Hardy <strong>and</strong> J. E. Littlewood.<br />
The circle method uses tools from complex analysis to study the asymptotics<br />
of the number of <strong>points</strong> of bounded height on complete intersections in a very<br />
high-dimensional projective space. It provides an asymptotic formula <strong>and</strong> an<br />
error term. The main term is of the form<br />
τB n+1−d 1− ... −d r<br />
for a complete intersection of multidegree (d 1 , . . . , d r ) in P n . The reader might<br />
want to consult [Va] for a description of the method <strong>and</strong> references to the<br />
original literature.<br />
The exponent of the main term allows a beautiful algebro-geometric interpretation.<br />
The anticanonical sheaf on a complete intersection of multidegree<br />
(d 1 , . . . , d r ) in P n is precisely O(n + 1 − d 1 − . . . − d r )| X . This means,<br />
when working with an anticanonical height instead of the naive height, the<br />
circle method proves linear growth for theÉ-<strong>rational</strong> <strong>points</strong>.<br />
The coefficient τ of the main term is a product of p-adic densities together with<br />
a factor corresponding to the archimedean valuation.
xiv<br />
INTRODUCTION<br />
Unfortunately, it is necessary to make very restrictive assumptions on the number<br />
of variables in comparison with the degrees of the equations. These assumptions<br />
on the dimension of the ambient projective space are needed in order to<br />
ensure that the provable error term is smaller than the main term. On might,<br />
nevertheless, hope that there is a similar asymptotic under much less restrictive<br />
conditions. This is the origin of Manin’s conjecture.<br />
However, aswasobservedbyJ. Franke, Yu. I. Manin, <strong>and</strong>Y. Tschinkel[F/M/T],<br />
the main term as described above is not compatible with the formation of<br />
direct products. Already on a variety as simple as P 1 × P 1 , the growth of the<br />
number of theÉ-<strong>rational</strong> <strong>points</strong> is actually asymptotically equal to τB log B.<br />
This may be seen by a calculation which is completely elementary.<br />
Thus, in general, the asymptotic formula has to be modified by a log-factor.<br />
Franke, Manin, <strong>and</strong> Tschinkel suggest the factor log rkPic(X )−1 B <strong>and</strong> prove that<br />
this factor makes the asymptotic formula compatible with direct products.<br />
Furthermore, it turns out that the coefficient τ has to be modified<br />
when rkPic(X ) > 1. There appears an additional factor which is today<br />
called α(X ). This factor is defined by a beautiful yet somewhat mysterious<br />
elementary geometric construction.<br />
Another problem is that the Hasse principle does not hold universally. Consider<br />
the following elementary example which was given by C.-E. Lind in 1940.<br />
Lind [Lin] dealt with the Diophantine equation<br />
2u 2 = v 4 − 17w 4<br />
defining an algebraic curve of genus 2. It is obvious that this equation is nontrivially<br />
solvable in reals <strong>and</strong> it is easy to check that it is non-trivially solvable<br />
inÉp for every prime number p.<br />
On the other h<strong>and</strong>, there is no solution in <strong>rational</strong>s except for (0, 0, 0).<br />
Indeed, assume the contrary. Then, there is a solution in integers such<br />
that gcd(u, v, w) = 1. For such a solution, one clearly has ∤u. Since 2 is a<br />
square but not a fourth power modulo 17, we conclude that ( u<br />
17)<br />
= −1. On the<br />
other h<strong>and</strong>, for every odd prime divisor p of u, one has v 4 −17w 4 ≡ 0 (mod p).<br />
This shows ( ) (<br />
17<br />
p = 1. By the low of quadratic reciprocity, p<br />
17)<br />
= 1. Altogether,<br />
( u<br />
17)<br />
= 1 which is a contradiction.<br />
One might argue that this example is not too interesting since, on a curve of<br />
genus g ≥ 2, there can be only finitely manyÉ-<strong>rational</strong> <strong>points</strong>. Thus, it might<br />
happen that there are none of them without any particular reason.<br />
However, several other counterexamples to the Hasse principle had been invented.<br />
Some of them were Fano varieties. For example, Sir P. Swinnerton-<br />
Dyer[S-D62] <strong>and</strong> L.-J. Mordell [Mord] (cf. Chapter III, Section 5) constructed
INTRODUCTION<br />
xv<br />
examples of cubic surfaces violating the Hasse principle. A few years later,<br />
J. W. S. Cassels <strong>and</strong> M. J. T. Guy [Ca/G] as well as A. Bremner [Bre] even found<br />
isolated examples of diagonal cubic surfaces showing that behaviour. Typically,<br />
the proofs were a bit less elementary than Lind’s in that sense that they<br />
required not the quadratic but the cubic or biquadratic reciprocity low.<br />
In the late sixties of the 20th century, Yu. I. Manin [Man] made the remarkable<br />
discovery that all the known counterexamples to the Hasse principle could<br />
be explained in a uniform manner. There was actually a class α ∈ Br(X ) in<br />
the Brauer group of the underlying algebraic variety responsible for the lack<br />
ofÉ-<strong>rational</strong> <strong>points</strong>.<br />
This may be explained as follows. The Brauer group ofÉis relatively complicated.<br />
One has, by virtue of global class field theory,<br />
Br(SpecÉ) = ker ( M<br />
s :<br />
2/−→É/) .<br />
pprimeÉ/⊕ M<br />
σ : K→Ê1<br />
Here, s is just the summation. The summ<strong>and</strong>É/corresponding to the prime<br />
number p is nothing but Br(SpecÉp) while the last summ<strong>and</strong> is Br(SpecÊ).<br />
Let α ∈ Br(X ) be any Brauer class of a variety X overÉ. An adelic point<br />
x = (x ν ) ν∈Val(É) ∈ X (É)<br />
defines restrictions of α to Br(SpecÊ) <strong>and</strong> Br(SpecÉp) for each p. If the<br />
sum of all invariants is different from zero then, according to the computation<br />
of Br(SpecÉ), x may not be approximated byÉ-<strong>rational</strong> <strong>points</strong>.<br />
As α ∈ Br(X ) then “obstructs” x from being approximated by <strong>rational</strong> <strong>points</strong>,<br />
the expression Brauer-Manin obstruction became the general st<strong>and</strong>ard for this<br />
famous observation of Manin.<br />
In the counterexamples to the Hasse principle which were known to Manin in<br />
those days, one typically had a Brauer class the restrictions of which had a totally<br />
degenerate behaviour. For example, on Lind’s curve, there is a Brauer class α<br />
such that its restriction is independent of the choice of the adelic point. α restricts<br />
to zero in Br(SpecÊ) <strong>and</strong> Br(SpecÉp) for p ≠ 17 but non-trivially<br />
to Br(SpecÉ17). This suffices to show that there is noÉ-<strong>rational</strong> point on<br />
that curve.<br />
In general, the Brauer-Manin obstruction defines a subset X (É) Br ⊆ X (É)<br />
consisting of the adelic <strong>points</strong> which are not affected by the obstruction. At least<br />
for cubic surfaces, there is a conjecture of J.-L. Colliot-Thélène stating that<br />
X (É) Br is equal to the set of all adelic <strong>points</strong> which may actually be approximated<br />
byÉ-<strong>rational</strong> <strong>points</strong>.<br />
Thus,X(É) Br = ∅whileX(É) ≠ ∅meansthatX isaprovencounterexample<br />
to the Hasse principle. If X (É) Br X (É) then we have a counterexample to
xvi<br />
INTRODUCTION<br />
weak approximation. If Colliot-Thélène’s conjecture were true then one could<br />
say that all cubic surfaces which are counterexamples to the Hasse principle or<br />
to weak approximation are of this form.<br />
The Brauer group of an algebraic variety X over an algebraically nonclosed<br />
field k admits, according to the Hochschild-Serre spectral sequence,<br />
a canonical filtration in three steps. The first step is given by the image<br />
of Br(Speck) in Br(X ). Second, Br(X )/ Br(Speck), has a subgroup canonically<br />
isomorphic to H 1( Gal(k/k), Pic(X k<br />
) ) . The remaining subquotient is a subgroup<br />
of Br(X k<br />
) Gal(k/k) . It turns out that only the group H 1( Gal(k/k), Pic(X k<br />
) )<br />
is relevant for the Brauer-Manin obstruction.<br />
In the cases where the circle method is applicable, the Noether-Lefschetz Theorem<br />
shows that Pic(X ) =with trivial Galois operation. Consequently,<br />
H 1( Gal(k/k), Pic(X ) ) = 0 which is clearly sufficient for the absence<br />
of the Brauer-Manin obstruction. This coincides perfectly well with the observation<br />
that the circle method always proves equidistribution.<br />
By consequence, in a conjectural generalization of the results proven by the<br />
circle method, one can work with X (É) Br instead of X (É) without making<br />
any change in the proven cases. However, in the cases where weak<br />
approximation fails, this does not give the correct answer as was observed<br />
by D. R. Heath-Brown [H-B92a], in 1992. On a cubic surface such that<br />
H 1( Gal(k/k), Pic(X ) ) =/3<strong>and</strong> a non-trivial Brauer class excludes two<br />
thirds of the adelic <strong>points</strong>, there are nevertheless as many <strong>rational</strong> <strong>points</strong> as<br />
naively expected. Even more, E. Peyre <strong>and</strong> Y. Tschinkel [Pe/T] showed experimentally<br />
that if H 1( Gal(k/k), Pic(X ) ) =/3<strong>and</strong> the Brauer class does not<br />
exclude any adelic point then there are three times more <strong>rational</strong> <strong>points</strong> than expected.<br />
Correspondingly, in E. Peyre’s [Pe95a] definition of the conjectural<br />
constant τ, there appears an additional factor β(X ) := #H 1( Gal(k/k), Pic(X k<br />
) ) .<br />
This book is concerned with Diophantine equations from the theoretical <strong>and</strong><br />
experimental <strong>points</strong> of view. It is divided into three parts. Part A deals with the<br />
concepts of a Brauer group <strong>and</strong> their applications. In the first chapter, we begin<br />
with the simplest particular case <strong>and</strong> recall the Brauer group of a field. As an<br />
abstract group, we have Br(k) = H 2( Gal(k/k), k ∗) . Taking this as a definition,<br />
it is not hard to show that Br(X ) classifies two a priori rather different sorts<br />
of objects. Namely, on one h<strong>and</strong>, central simple algebras over k <strong>and</strong>, on the<br />
other, Brauer-Severi varieties over k.<br />
WerecallindetailthemachineryofGaloisdescent. Asanapplication, weexplain<br />
why both central simple algebras of dimension n 2 splitting over an extension<br />
field L as well as Brauer-Severi varieties of dimension n splitting over L are<br />
classified by H 1( Gal(L/K ), PGL n (L) ) . In Section I.7, there is given a functor
INTRODUCTION<br />
xvii<br />
from central simple algebras to Brauer-Severi varieties which yields the classical<br />
correspondence on objects.<br />
The second chapter considers A. Grothendieck’s generalization of the<br />
Brauer group to the case of an arbitrary scheme. We recall the concept of a<br />
sheaf of Azumaya algebras on a scheme <strong>and</strong> explain how such a sheaf of algebras<br />
gives rise to a class in the étale cohomology Br ′ (X ) := H 2 ét (X,m). This is what<br />
is called the cohomological Brauer group. On the other h<strong>and</strong>, a rather naive<br />
generalization of the definition for fields yields the concept of the Brauer group.<br />
One has Br(X ) ⊆ Br ′ (X ). In general, the two are not equal to each other.<br />
In Section II.7, we give a proof for the Theorem of Ausl<strong>and</strong>er <strong>and</strong> Goldman<br />
stating that Br(X ) = Br ′ (X ) in the case of a smooth surface. This result was<br />
originally shown in [A/G] before the actual invention of schemes. The proof<br />
of Ausl<strong>and</strong>er <strong>and</strong> Goldman was formulated in the language of Brauer <strong>groups</strong> for<br />
commutative rings. However, all the arguments given carry over immediately<br />
to the case of a scheme.<br />
The chapter is closed by computations of Brauer <strong>groups</strong> in particular examples.<br />
In the case of a variety over an algebraically non-closed field, we study the<br />
relationship of Br(X ) with H 1( Gal(k/k), Pic(X k<br />
) ) . We prove Manin’s formula<br />
expressing the latter cohomology group in terms of the Galois operation on<br />
a specific set of divisors. For smooth cubic surfaces, one may work with the<br />
classes given by the 27 lines.<br />
This leads to the result of Sir P. Swinnerton-Dyer [S-D93] that, for a smooth<br />
cubic surface, H 1( Gal(k/k), Pic(X k<br />
) ) is one of the <strong>groups</strong> 0,/2,/3,<br />
(/2) 2 , <strong>and</strong> (/3) 2 . Swinnerton-Dyer’s proof filled the entire article [S-D93]<br />
<strong>and</strong> was later modified by P. K. Corn in his thesis [Cor].<br />
We discovered that Swinnerton-Dyer’s result may be obtained in a manner which<br />
is rather brute force but very simple. The Galois group acting on the 27 lines on<br />
a smooth cubic surface is a subgroup of W (E 6 ). There are only 350 conjugacy<br />
classes of sub<strong>groups</strong> of W (E 6 ). We computed H 1( Gal(k/k), Pic(X k<br />
) ) in each of<br />
these cases using GAP. This took 28 seconds of CPU time.<br />
As an application of Brauer <strong>groups</strong>, the third chapter is concerned with the<br />
Brauer-Manin obstruction. We recall the notion of an adelic point <strong>and</strong> define the<br />
local <strong>and</strong> global evaluation maps. An adelic point x = (x ν ) ν∈Val(É) is “obstructed”<br />
from being approximated by <strong>rational</strong> <strong>points</strong> if the global evaluation map ev gives<br />
a non-zero value ev(α, x) for a certain Brauer class α ∈ Br(X ).<br />
We then describe a strategy on how the Brauer-Manin obstruction may be<br />
explicitly computed in concrete examples. We carry out this strategy for two<br />
special types of cubic surfaces.<br />
The first type is given as follows. Let p 0 ≡ 1 (mod 3) be a prime number<br />
<strong>and</strong> K/Ébe the unique cubic field extension contained in the cyclotomic
xviii<br />
INTRODUCTION<br />
extensionÉ(ζ p0 )/É. Fix the explicit generator θ ∈ K given by<br />
θ := trÉ(ζ p0 )/K(ζ p0 − 1) = −2n + ∑<br />
i∈(∗ p0 ) 3 ζ i p 0<br />
for n := p 0−1<br />
. Then, consider the cubic surface X ⊂ P 3É, given by<br />
6<br />
3 ( )<br />
T 3 (a 1 T 0 + d 1 T 3 )(a 2 T 0 + d 2 T 3 ) = T0 + θ (i) T 1 + (θ (i) ) 2 T 2 .<br />
Here, a 1 , a 2 , d 1 , d 2 ∈. The θ (i) denote the three images of θ under Gal(K/É).<br />
Proposition III.5.3 provides criteria to verify that such a surface is smooth<br />
<strong>and</strong> has p-adic <strong>points</strong> for every prime p. More importantly, the Brauer-Manin<br />
obstruction can be understood completely explicitly. At least for a generic<br />
choice of a 1 , a 2 , d 1 , <strong>and</strong> d 2 , one has that H 1( Gal(k/k), Pic(X k<br />
) ) =/3.<br />
Further, there is a class α ∈ Br(X ) with the following property. For an adelic<br />
point x = (x ν ) ν , the value of ev(α, x) depends only on the component x νp0 .<br />
Write x νp0 =: (t 0 : t 1 : t 2 : t 3 ). Then, one has ev(α, x) = 0 if <strong>and</strong> only if<br />
a 1 t 0 + d 1 t 3<br />
t 3<br />
is a cube in∗ p0<br />
. Note that p 0 ≡ 1 (mod 3) implies that only every third element<br />
of∗ p0<br />
is a cube.<br />
∏<br />
i=1<br />
Observe that the reduction of X modulo p 0 is given by<br />
T 3 (a 1 T 0 + d 1 T 3 )(a 2 T 0 + d 2 T 3 ) = T 3<br />
0 .<br />
This means, there are three planes intersecting in a triple line. NoÉp 0<br />
-<strong>rational</strong><br />
point may reduce to the triple line. Thus, there are three different planes, a<br />
Ép 0<br />
-<strong>rational</strong> point may reduce to. The value of ev(α, x) depends only the plane<br />
to which its component v νp0 is mapped under reduction.<br />
For instance (cf. Example III.5.24), for p 0 = 19, consider the cubic surface X<br />
given by<br />
3 ( )<br />
T 3 (T 0 + T 3 )(12T 0 + T 3 ) = T0 + θ (i) T 1 + (θ (i) ) 2 T 2 .<br />
Then, in19, the cubic equation<br />
∏<br />
i=1<br />
T (1 + T )(12 + T ) − 1 = 0<br />
has the three solutions 12, 15, <strong>and</strong> 17. However, in19, 13/12 = 9, 16/15 = 15,<br />
<strong>and</strong> 18/17 = 10 which are three non-cubes. This shows that X (É) = ∅. It is<br />
easy to check that X (É) ≠ ∅. Therefore, X is an example of a cubic surface<br />
violating the Hasse principle.<br />
We construct a number of similar examples. For instance, Example III.5.24<br />
describes a cubic surface X such that H 1( Gal(k/k), Pic(X k<br />
) ) =/3but the
INTRODUCTION<br />
xix<br />
generating Brauer class does not exclude a single adelic point. One would expect<br />
that X satisfies weak approximation. Recall that, in similar examples, E. Peyre<br />
<strong>and</strong> Y. Tschinkel [Pe/T] showed experimentally that there are three times more<br />
É-<strong>rational</strong> <strong>points</strong> than expected.<br />
The historically first cubic surface which could be proven to be a counterexample<br />
to the Hasse principle was provided by Sir P. Swinnerton-Dyer [S-D62].<br />
We recover Swinnerton-Dyer’s example (cf. Example III.5.27) for p 0 = 7,<br />
d 1 = d 2 = 1, a 1 = 1, <strong>and</strong> a 2 = 2. L. J. Mordell [Mord] generalized Swinnerton-Dyer’s<br />
work by giving a series of examples for p 0 = 7 <strong>and</strong> a series of<br />
examples for p 0 = 13. Yu. I. Manin mentions Mordell’s examples explicitly in<br />
his book [Man]. He explains these counterexamples to the Hasse principle by a<br />
Brauer class. We generalize Mordell’s examples further to the case that p 0 is an<br />
arbitrary prime such that p 0 ≡ 1 (mod 3).<br />
We conclude Chapter III by a section on diagonal cubic surfaces. For these, the<br />
Brauer-Manin obstruction was investigated in the monumental work [CT/K/S]<br />
of J.-L. Colliot-Thélène, D. Kanevsky, <strong>and</strong> J.-J. Sansuc. We present an explicit<br />
computation of the Brauer-Manin obstruction under a congruence condition<br />
which corresponds more or less to the “first case” of [CT/K/S]. Our argumentation<br />
is, however, shorter <strong>and</strong> simpler than the original one. The point is<br />
that we make use of the fact that H 1( Gal(k/k), Pic(X k<br />
) ) may be only 0,/2,<br />
/3, (/2) 2 , or (/3) 2 . Further, the group (/3) 2 appears only once in<br />
a very particular case. Thus, in order to prove H 1( Gal(k/k), Pic(X k<br />
) ) =/3,<br />
it is almost sufficient to construct an element of order three.<br />
The second part of this book is devoted to the various concepts of a height.<br />
In the fourth chapter, we start with the naive height forÉ-<strong>rational</strong> <strong>points</strong> on<br />
projective space. Then, our goal is to give an overview of the theories which<br />
provide natural generalizations of this simple concept.<br />
The very first generalization is the naive height for <strong>points</strong> in projective space<br />
defined over a finite extension ofÉ. Then, following André Weil, we introduce<br />
the concept of a height defined by an ample invertible sheaf. This is a height<br />
function which is defined only up to a bounded summ<strong>and</strong>.<br />
To overcome this difficulty, one has to work with arithmetic varieties <strong>and</strong><br />
metrized invertible sheaves. Arithmetic varieties are schemes projective<br />
over Spec. Actually, this leads to a beautiful geometric interpretation of<br />
the naive height.<br />
Indeed, let X be a projective variety overÉ<strong>and</strong> X a projective model of X<br />
over Spec. Fix a hermitian line bundle L on X . Then, according to the valuative<br />
criterion of properness, everyÉ-<strong>rational</strong> point x on X extends uniquely<br />
to a-valued point x : Spec→X. The height function with respect to L is
xx<br />
INTRODUCTION<br />
then given by<br />
h L<br />
(x) := ̂deg ( x ∗ L ) .<br />
Here, ̂deg denotes the Arakelov degree of a hermitian line bundle over Spec.<br />
It turns out that this coincides exactly with the naive height when one works<br />
with X = P n, L = O(1), <strong>and</strong> the minimum metric which is defined by<br />
∣ ‖l‖ min := min<br />
l ∣∣∣<br />
i=0, ... ,n<br />
∣ .<br />
X i<br />
In general, h L<br />
admits a fundamental finiteness property as soon as L is ample.<br />
This is the starting point of arithmetic intersection theory, a fascinating theory<br />
which we may only touch upon. We recall the main definitions <strong>and</strong> results<br />
of the arithmetic intersection theory due to H. Gillet <strong>and</strong> C. Soulé [G/S 90,<br />
G/S 92, S/A/B/K]. As an application, we consider the concept of a height<br />
introduced by J.-B. Bost, H. Gillet, <strong>and</strong> C. Soulé [B/G/S]. This is a height<br />
function for cycles of arbitrary dimension. For a 0-dimensional cycle, the<br />
Bost-Gillet-Soule height coincides with the usual height function defined by the<br />
hermitian line bundle.<br />
Wecloseour overviewofarithmeticintersectiontheoryby a sectiononthe arithmetic<br />
Hilbert-Samuel formula. We formulate the formula as Theorem IV.6.5.<br />
Asaconsequence, weproveS. Zhangs’s[Zh95b, (5.2)]theorem ofsuccessiveminima.<br />
Arithmetic intersection theory seems to be a very general framework. It is,<br />
however, still not general enough. An obvious problem is that the naive height<br />
on P n is not covered. This problem is of technical nature. To define the intersection<br />
product of two hermitian line bundles (or, more correctly, of the<br />
corresponding arithmetic cycles), Gillet <strong>and</strong> Soulé had to assume that all hermitian<br />
metrics are smooth. Unfortunately, the minimum metric is continuous,<br />
but not smooth.<br />
A theory which is general enough to cover the Bost-Gillet-Soule height as well<br />
as the naive height is provided by S. Zhang’s adelic intersection theory [Zh95a].<br />
We give a definition for a version of the adelic Picard group ˜Pic(X ). Elements of<br />
˜Pic(X ) are called integrably metrized invertible sheaves. Then, we construct<br />
the corresponding adelic intersection product. As an application, we explain<br />
the height function defined by an integrably metrized invertible sheaf.<br />
In Chapter 5, we use adelic intersection theory to give a unified proof for two<br />
equidistribution theorems for <strong>points</strong> of small height. On one h<strong>and</strong>, there is<br />
the equidistribution theorem of L. Szpiro, E. Ullmo <strong>and</strong> S. Zhang [S/U/Z]<br />
for small <strong>points</strong> on abelian varieties. On the other h<strong>and</strong>, the completely parallel<br />
theorem for the naive height on projective space which was established
INTRODUCTION<br />
xxi<br />
by Yuri Bilu [Bilu97]. The main result of Chapter 5 is formulated as Theorem<br />
V.1.14. A. Chambert-Loir’s equidistribution result [C-L] for quasi-split<br />
semi-abelian varieties can easily be deduced from this theorem.<br />
Chapter VI is devoted to some of the most popular conjectures concerning<br />
<strong>rational</strong> <strong>points</strong> on projective algebraic varieties. We discuss Lang’s conjecture,<br />
the conjecture of Batyrev <strong>and</strong> Manin, <strong>and</strong>, most notably, Manin’s conjecture<br />
about the asymptotics of <strong>points</strong> of bounded height on Fano varieties (Conjecture<br />
VI.5.37). A large part of the chapter is concerned with E. Peyre’s Tamagawa<br />
type number τ (X ), the coefficient expected in the asymptotic formula.<br />
We discuss in detail all factors appearing in the definition of τ (X ).<br />
In particular, we give a number of examples for which we explicitly compute<br />
the factor α(X ). We mainly consider smooth cubic surfaces of arithmetic Picard<br />
rank two.<br />
Part C collects several reports on practical experiments. Each chapter is concerned<br />
with a particular problem for a specific variety or family of varieties.<br />
We tried to keep the chapters as self-contained as possible. Each one starts with<br />
an introduction recalling those parts of the general theory described in parts A<br />
<strong>and</strong> B which are necessary to follow that chapter.<br />
In Chapter VII, we describe our investigations regarding the families<br />
“ax 3 = by 3 +z 3 +v 3 +w 3 ”, a, b = 1, . . . , 100, <strong>and</strong> “ax 4 = by 4 +z 4 +v 4 +w 4 ”,<br />
a, b = 1, . . . , 100, of projective algebraic threefolds. We report numerical evidence<br />
for the conjecture of Manin in the refined form due to E. Peyre.<br />
Our experiments included searching for <strong>points</strong>, computing the Tamagawa number,<br />
<strong>and</strong> detecting the accumulating subvarieties. Concerning the programmer’s<br />
efforts, detection of accumulating subvarieties was the most difficult part of<br />
this project. For example, one the cubic threefolds, the following non-obvious<br />
lines have been found. These are the only non-obvious lines we know <strong>and</strong> the<br />
only ones containing a point of height less than 5000.<br />
TABLE 1. Sporadic lines on cubic threefolds<br />
a b Smallest point Point s.t. x = 0<br />
19 18 (1 : 2 : 3 : -3 : -5) (0 : 7 : 1 : -7 : -18)<br />
21 6 (1 : 2 : 3 : -3 : -3) (0 : 9 : 1 : -10 : -15)<br />
22 5 (1 : -1 : 3 : 3 : -3) (0 : 27 : -4 : -60 : 49)<br />
45 18 (1 : 1 : 3 : 3 : -3) (0 : 3 : -1 : 3 : -8)<br />
73 17 (1 : 5 : -2 : 11 : -15) (0 : 27 : -40 : 85 : -96)<br />
We describe all the computations which were done as well some background on<br />
the geometry of cubic <strong>and</strong> quartic threefolds.
xxii<br />
INTRODUCTION<br />
In the eighth chapter, we continue our study of the family of quartic threefolds<br />
by comparing the height m(X ) of the smallest <strong>rational</strong> point with Peyre’s constant.<br />
We prove that there is no constant C such that m(X ) < C<br />
τ<br />
for every<br />
(X )<br />
diagonal quartic threefold, not even for those of type ax 4 = y 4 + z 4 + v 4 + w 4 .<br />
1<br />
We also prove that, at least for diagonal quartic threefolds, the reciprocal<br />
τ (X )<br />
1<br />
behaves like a height function. I.e.,<br />
τ<br />
admits a fundamental finiteness property.<br />
(X )<br />
Chapter IX is devoted to the study of general cubic surfaces. A general cubic<br />
surface is described by twenty coefficients. With current technology, it is impossible<br />
to study all cubic surfaces with coefficients below a reasonable bound.<br />
For that reason, we decided to work with coefficient vectors provided by a<br />
r<strong>and</strong>om number generator. Our first sample consists of 20 000 surfaces with coefficients<br />
r<strong>and</strong>omly chosen in the interval [0, 50]. The second sample consists of<br />
20 000 surfaces with r<strong>and</strong>omly chosen coefficients from the interval [−100, 100].<br />
We verified explicitly that each of the surfaces studied is smooth. Then, we<br />
proved that, for each surface, the full Galois group W (E 6 ) acts on the 27 lines.<br />
This implies that the Picard rank is always equal to 1. Also, the Brauer-Manin obstruction<br />
is not present on any of the surfaces considered. The algorithm to<br />
check that the group acting on the 27 lines is indeed W (E 6 ) is described in detail.<br />
Further, we searched for <strong>rational</strong> <strong>points</strong> <strong>and</strong> computed the Tamagawa number,<br />
thereby giving numerical evidence for Manin’s conjecture.<br />
In the next chapter, we return to the more st<strong>and</strong>ard case of diagonal cubic surfaces.<br />
The experiments areanalogous to those described in Chapters VII<strong>and</strong> VIII<br />
for diagonal cubic <strong>and</strong> quartic threefolds. The theory is, however, more complicated.<br />
The geometric Picard rank is equal to 7 <strong>and</strong>, in the generic case,<br />
there is a Brauer-Manin obstruction to weak approximation excluding precisely<br />
two thirds of the adelic <strong>points</strong>. The factors α(X ) <strong>and</strong> β(X ) appearing in the<br />
definition of Peyre’s constant are not always the same <strong>and</strong> need to be considered.<br />
We demonstrate experimentally the connection of Peyre’s constant with the<br />
height m(X ) of the smallest <strong>rational</strong> point. Under the Generalized Riemann<br />
Hypothesis, we prove that there is no constant C such that m(X ) < C<br />
τ for (X )<br />
every diagonal cubic surface. We also prove that, for diagonal cubic surfaces, the<br />
1<br />
reciprocal<br />
τ behaves like a height function. I.e., 1<br />
(X ) τ<br />
admits a fundamental<br />
(X )<br />
finiteness property.<br />
Chapter XI is concerned with the Diophantine equation<br />
x 4 + 2y 4 = z 4 + 4w 4 . ()<br />
This equation gives an example of a K3 surface X defined overÉ. It is an open<br />
question whether there exists a K3 surface overÉwhich has a finite non-zero<br />
number ofÉ-<strong>rational</strong> <strong>points</strong>.
INTRODUCTION<br />
xxiii<br />
X might be a c<strong>and</strong>idate for a K3 surface with this property. (1 : 0 : 1 : 0)<br />
<strong>and</strong> (1 : 0 : (−1) : 0) are two obvious <strong>rational</strong> <strong>points</strong>. Sir P. Swinnerton-<br />
Dyer [Poo/T, Problem/Question 6.c)] had publicly posed the problem to find a<br />
third <strong>rational</strong> point on X. But no <strong>rational</strong> <strong>points</strong> different from the two obvious<br />
ones had been found in experiments carried out by several people.<br />
We explain our approach to efficiently search forÉ-<strong>rational</strong> <strong>points</strong> on algebraic<br />
varieties defined by a decoupled equation. It is based on hashing, a method from<br />
Computer Science. In the particular case of a surface in P 3É, our algorithm is of<br />
complexity essentially O(B 2 ) for a search bound of B.<br />
In the final implementation, we could work with the search bound B = 10 8 .<br />
We discovered the following solution of the Diophantine equation ().<br />
1 484 801 4 + 2 · 1 203 120 4 = 9 050 910 498 475 648 046899201<br />
1 169 407 4 + 4 · 1 157 520 4 = 9 050 910 498 475 648 046899201<br />
Up to changes of sign, this is the only non-obvious solution of () we know <strong>and</strong><br />
the only non-obvious solution of height less than 10 8 [EJ2, EJ3].<br />
The last chapter is a report on our approach to the three cubes problem. To be<br />
more precise, the problem is whether every <strong>rational</strong> integer n ≢ 4, 5 (mod 9)<br />
can be written as a sum of three integral cubes.<br />
Using an implementation of Elkies’ method, we found, among many others,<br />
the following equations.<br />
156 = 26 577 110 807 569 3 − 18 161 093 358 005 3 − 23 381 515 025 762 3<br />
318 = 1 970 320 861 387 3 + 1 750 553 226 136 3 − 2 352 152 467 181 3<br />
= 30 828 727 881 037 3 + 27 378 037 791 169 3 − 36 796 384 363 814 3<br />
366 = 241 832 223 257 3 + 167 734 571 306 3 − 266 193 616 507 3<br />
420 = 8 859 060 149 051 3 − 2 680 209 928 162 3 − 8 776 520 527 687 3<br />
564 = 53 872 419 107 3 − 1 300 749 634 3 − 53 872 166 335 3<br />
758 = 662 325 744 409 3 + 109 962 567 936 3 − 663 334 553 003 3<br />
= 83 471 297 139 078 3 + 77 308 024 343 011 3 − 101 433 242 878 565 3<br />
789 = 18 918 117 957 926 3 + 4 836 228 687 485 3 − 19 022 888 796 058 3<br />
894 = 19 868 127 639 556 3 + 2 322 626 411 251 3 − 19 878 702 430 997 3<br />
948 = 103 458 528 103 519 3 + 6 604 706 697 037 3 − 103 467 499 687 004 3<br />
For none of the numbers on the left, a decomposition into a sum of three cubes<br />
had been known before.<br />
Göttingen,<br />
Summer 2008
PART A<br />
THE BRAUER GROUP
CHAPTER I<br />
THE BRAUER-SEVERI VARIETY<br />
ASSOCIATED WITH<br />
A CENTRAL SIMPLE ALGEBRA ∗<br />
No attention should be paid to the fact that algebra <strong>and</strong> geometry are<br />
different in appearance.<br />
OMAR KHAYYÁM (1070, translated by A. R. Amir-Moez)<br />
1. Introductory remarks<br />
This chapter is devoted to the correspondence between central simple algebras<br />
<strong>and</strong> Brauer-Severi varieties.<br />
This is classical material. Only Section 7 is exceptional to a certain extent.<br />
There, we will give a functor from central simple algebras to Brauer-Severi<br />
varieties which yields the classical correspondence on objects.<br />
Some history. —<br />
Central simple algebras were studied intensively by many mathematicians at the<br />
end of the 19th <strong>and</strong> in the first half of the 20th century. We refer the reader to<br />
N. Bourbaki [Bou-A, Note historique] for a detailed account on the history of<br />
the subject <strong>and</strong> mention only a few important milestones here. The structure<br />
of central simple algebras (being finite dimensional over a field K) is fairly easy.<br />
They are full matrix rings over division algebras the center of which is equal<br />
to K. This was finally discovered by J. H. Maclagan-Wedderburn in 1907<br />
[MWe08] after several special cases had been treated before. T. Molien [Mol]<br />
had considered the case of-algebras already in 1893 <strong>and</strong> the case ofÊ-algebras<br />
had been investigated by E. Cartan [Car]. J. H. Maclagan-Wedderburn himself<br />
had proven the structure theorem for central simple algebras over finite fields<br />
in 1905 [MWe05, Di05].<br />
(∗) This chapter is a revised version of the article: The Brauer-Severi variety associated with a<br />
central simple algebra, Linear Algebraic Groups <strong>and</strong> Related Structures 52(2000), 1-60.
4 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
In 1929, R. Brauer ([Bra], see also [Deu] <strong>and</strong> [A/N/T]), using E. Noether’s<br />
ideas about crossed products of algebras, found the group structure on the<br />
set of similarity classes of central simple algebras over a field K. He constructed,<br />
in today’s language, an isomorphism to the Galois cohomology<br />
group H 2( Gal(K sep /K ), (K sep ) ∗) . Further, he discussed the structure of this<br />
group in the case of a number field.<br />
Relative versions of central simple algebras over base rings instead of fields were<br />
introduced by G. Azumaya [Az] <strong>and</strong> M. Ausl<strong>and</strong>er <strong>and</strong> O. Goldman [A/G].<br />
The case of an arbitrary base scheme was considered by A. Grothendieck in his<br />
famous Bourbaki talks “Le groupe de Brauer” [GrBrI, GrBrII, GrBrIII].<br />
Brauer-Severi varieties are twisted forms of the projective space. The term<br />
“varieté de Brauer” appeared for the first time in 1944 in the article [Ch44]<br />
of F. Châtelet. Nevertheless, F. Severi had proven already in 1932 that a Brauer-<br />
Severi variety over a field K admitting a K-valued point is necessarily isomorphic<br />
to the projective space.<br />
It should be mentioned that Brauer-Severi varieties are important in a number<br />
of applications. The first <strong>and</strong> definitely most striking one is the proof of the<br />
theorem of Merkurjev-Suslin [Me/S] on the cotorsion of K 2 of a field.<br />
The Merkurjev-Suslin theorem has an interesting further application. It provides<br />
a certain underst<strong>and</strong>ing of the (torsion part of the) Chow <strong>groups</strong> of certain<br />
algebraic varieties. The reader should consult the work of J.-L. Colliot-Thélène<br />
[CT91] for information about that.<br />
As a second kind of application, Brauer-Severi varieties often appear in constructions<br />
of varieties with particular properties. For instance, the famous example<br />
due to M. Artin <strong>and</strong> D. Mumford [A/M] of a threefold which is uni<strong>rational</strong><br />
but not <strong>rational</strong> is a variety fibered over a <strong>rational</strong> surface such that the generic<br />
fiber is a conic without <strong>rational</strong> <strong>points</strong>. For more historical details, especially<br />
on the work of F. Châtelet, we refer the reader to the article of J.-L. Colliot-<br />
Thélène [CT88].<br />
The correspondence between central simple algebras <strong>and</strong> Brauer-Severi varieties<br />
was first observed by E. Witt [Wi35] <strong>and</strong> H. Hasse in the special case of<br />
quaternion algebras <strong>and</strong> plane conics.<br />
Here, we are going to present in detail the most elementary approach to this correspondence.<br />
It is based on non-abelian group cohomology. The main observation<br />
is that central simple algebras of dimension n 2 over a field K as well<br />
as (n − 1)-dimensional Brauer-Severi varieties over K can both be described<br />
by classes in one <strong>and</strong> the same cohomology set H 1( Gal(K sep /K ), PGL n (K sep ) ) .<br />
Note that this approach was promoted by J.-P. Serre in his books [Se62, chap. X,
Sec. 2] NON-ABELIAN GROUP COHOMOLOGY 5<br />
§§5,6] <strong>and</strong> [Se73, Remarque III.1.3.1]. It was, however, known to F. Châtelet before.<br />
A second approach is closer to A. Grothendieck’s style. One can give a direct<br />
description of a functor of <strong>points</strong> P XA : {K-schemes} → {sets} in terms of data<br />
of the central simple algebra A. It is possible to prove representability by a<br />
projective scheme using brute force. This approach is presented explicitly in<br />
the book of I. Kersten [Ke]. For a very detailed account, the reader may consult<br />
the Ph.D. thesis of F. Henningsen [Hen]. We are going to prove that these two<br />
approaches are equivalent. In fact, we will compute the functor of <strong>points</strong> of the<br />
variety given by the first approach <strong>and</strong> show that it is naturally isomorphic to<br />
the functor usually taken as the starting point for the second approach.<br />
There is a third approach which we only mention here. It works via algebraic<br />
<strong>groups</strong> <strong>and</strong> can be used to produce twisted forms not only of the projective<br />
space but of any homogeneous space G/P where G is a semisimple algebraic<br />
group <strong>and</strong> P ⊂ G a parabolic subgroup (see [K/R]).<br />
2. Non-abelian group cohomology<br />
In this section, we recall elementary facts about non-abelian group cohomology.<br />
I.e., cohomology of discrete <strong>groups</strong> with non-abelian coefficients. Non-abelian<br />
Galois cohomology will be the central tool for this chapter.<br />
2.1. Definition. –––– Let G be a finite group.<br />
i) A G-set E is a set equipped with a G-operation from the left.<br />
A morphism of G-sets, a G-morphism for short, is a map i : E → F of G-sets<br />
such that the diagram<br />
G × E ·<br />
E<br />
commutes.<br />
id×i<br />
<br />
G × F<br />
· F<br />
ii) A G-set E carrying a group structure such that g (xy) = g x g y for every g ∈ G<br />
<strong>and</strong> x, y ∈ E is called a G-group. Here, we write g x := g ·x for x ∈ E <strong>and</strong> g ∈ G.<br />
If the group underlying E is abelian then E is called a G-module.<br />
2.2. Definition. –––– Let G be a finite group.<br />
i) For a G-set E, one puts<br />
H 0 (G, E) := E G .<br />
I.e., the zeroth cohomology set of G with coefficients in E is equal to the subset of<br />
G-invariants.<br />
i
6 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
ii) Let A be a G-group.<br />
Then, a cocycle from G to A is a map G → A, g ↦→ a g such that a gh = a g · ga<br />
h<br />
for each g, h ∈ G.<br />
Two cocycles a, a ′ are said to be cohomologous if there exists some b ∈ A such<br />
that a ′ g = b −1 · a g · gb for every g ∈ G.<br />
This is an equivalence relation <strong>and</strong> the quotient set, the first cohomology set of G<br />
with coefficients in A, is denoted by H 1 (G, A). This is a pointed set as the map<br />
g ↦→ e defines a cocycle, the so-called trivial cocycle.<br />
2.3. Remarks. –––– i) If E is a G-group then H 0 (G, E) is a group.<br />
ii) For a cocycle a, a ′ such that a ′ g := b−1·a g · gb for each g ∈ G is a cocycle, too.<br />
iii) H 0 (G, A) <strong>and</strong> H 1 (G, A) are covariant functors in A. If i : A → A ′ is a<br />
morphism of G-sets (a morphism of G-<strong>groups</strong>) then the induced map(s) will be<br />
denoted by i ∗ : H 0 (G, A) → H 0 (G, A ′ ) (<strong>and</strong> i ∗ : H 1 (G, A) → H 1 (G, A ′ )).<br />
iv) If A is abelian then the notions defined above coincide with the usual<br />
group cohomology. One of the possible descriptions for H ∗ (G, A) is the cohomology<br />
of the complex<br />
with differentials<br />
0 −→ A d<br />
−→ Map(G, A)<br />
d<br />
−→ Map(G 2 , A)<br />
dϕ(g 1 , . . . , g n+1 ) := g 1<br />
ϕ(g 2 , . . . , g n+1 )<br />
+<br />
n<br />
∑<br />
j=1<br />
d<br />
−→ . . .<br />
(−1) j ϕ(g 1 , . . . , g j g j+1 , . . . , g n+1 )<br />
+ (−1) n+1 ϕ(g 1 , . . . , g n ) .<br />
2.4. Proposition. –––– Let G be a finite group.<br />
a) Let A ⊆ B be a G-subgroup <strong>and</strong> B/A the set of left cosets. Then, there is a natural<br />
exact sequence of pointed sets<br />
1 → H 0 (G, A) → H 0 (G, B) → H 0 (G, B/A) δ → H 1 (G, A) → H 1 (G, B) .<br />
b) If A ⊆ B is a normal G-subgroup then there is a natural exact sequence of<br />
pointed sets<br />
1 → H 0 (G, A) → H 0 (G, B) → H 0 (G, B/A) δ → H 1 (G, A) → . . .<br />
. . . → H 1 (G, B) → H 1 (G,B/A) .
Sec. 2] NON-ABELIAN GROUP COHOMOLOGY 7<br />
c) If A ⊆ B is a G-module lying in the center of B then there is a natural exact<br />
sequence of pointed sets<br />
1 → H 0 (G, A) → H 0 (G, B) → H 0 (G, B/A) δ → H 1 (G, A) → . . .<br />
. . . → H 1 (G, B) → H 1 (G, B/A) → δ′<br />
H 2 (G, A) .<br />
Here,theabeliangroupH 2 (G, A)isconsidered asapointed setwiththeunitelement.<br />
Proof. The connection homomorphism δ is defined in the following way.<br />
Let x ∈ H 0 (G, B/A). Take a representative x ∈ B for x <strong>and</strong> put a s := x −1 · sx.<br />
This is a cocycle <strong>and</strong> its equivalence class is denoted by δ(x). The definition<br />
of δ(x) is independent of the x chosen.<br />
In the situation of c), the map δ ′ is given as follows. Let x ∈ H 1 (G, B/A).<br />
Choose a cocycle (x g ) g∈G that represents x <strong>and</strong> lift each x g to some x g ∈ B.<br />
Put a(g 1 , g 2 ) := g 1 xg2 · x −1<br />
g 1 g 2 · x g1 . This is a 2-cocycle with values in A <strong>and</strong> its<br />
equivalence class is denoted by δ ′ (x). The definition of δ ′ (x) is independent of<br />
the choices.<br />
Exactness has to be checked at each entry, separately.<br />
2.5. Remark. –––– A sequence (A, a)<br />
to be exact in (B, b) if i(A) = j −1 (c).<br />
i<br />
→ (B, b)<br />
□<br />
j<br />
→ (C, c) of pointed sets is said<br />
2.6. Definition. –––– Let h : G ′ → G be a homomorphism of finite <strong>groups</strong>.<br />
Then, for an arbitrary G-set E, one has a natural pull-back map<br />
h ∗ : H 0 (G, E) → H 0 (G ′ , E).<br />
If E is a G-group then the pull-back map is a group homomorphism.<br />
For an arbitrary G-group A, there is the natural pull-back map<br />
which is a morphism of pointed sets.<br />
h ∗ : H 1 (G, A) → H 1 (G ′ , A)<br />
2.7. Remark. –––– If h is the inclusion of a subgroup then the pull-back<br />
res G′<br />
G := h∗ is usually called the restriction map.<br />
If h : G ′ ֒→ G is the canonical projection to a quotient group then inf G′<br />
G := h ∗<br />
is said to be the inflation map.<br />
The composition of res G′<br />
G<br />
or infG′ G with some extension of the G ′ -set E (the<br />
G ′ -group A) is usually called the restriction, respectively inflation, too.
8 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
2.8. Remark. –––– Non-abelian group cohomology may easily be extended to<br />
the case where G is a profinite group <strong>and</strong> A a discrete G-set (respectively<br />
G-group) on which G operates continuously. Indeed, put<br />
H i (G, A) := lim H i (G/G ′ , A G′ )<br />
−→G<br />
′<br />
fori = 0<strong>and</strong>1. Here, thedirectlimitistakenover theinflationmaps<strong>and</strong>G ′ runs<br />
through the open normal sub<strong>groups</strong> G ′ ⊆ G with finite quotient.<br />
3. Galois descent<br />
3.1. Definition. –––– Let L/K be a finite Galois extension <strong>and</strong> σ ∈ Gal(L/K ).<br />
a) Then, a map i : V 1 → V 2 of L-vector spaces is said to be σ-linear if, for every<br />
v, w ∈ V 1 <strong>and</strong> λ ∈ L, one has<br />
i(v + w) = i(v) + i(w)<br />
<strong>and</strong><br />
i(λv) = σ(λ)i(v) .<br />
b) Let π 1 : X 1 → SpecL <strong>and</strong> π 2 : X 2 → SpecL be L-schemes. We say that<br />
f : X 1 → X 2 is a morphism of L-schemes twisted by σ if the diagram<br />
X 1<br />
f<br />
X 2<br />
π 2<br />
π 1<br />
<br />
SpecL<br />
S(σ)<br />
SpecL<br />
commutes. Here, S(σ): SpecL → SpecL denotes the morphism of affine<br />
schemes induced by σ −1 : L → L.<br />
3.2. Theorem. –––– Let L/K be a finite Galois extension <strong>and</strong> G := Gal(L/K ).<br />
Then,<br />
i) there are the following equivalences of categories,<br />
⎧<br />
⎫<br />
⎨L-vector spaces<br />
⎬<br />
{K-vector spaces} −→ with a G-operation from the left<br />
⎩<br />
⎭ ,<br />
such that every σ ∈ G operates σ-linearly<br />
⎧<br />
⎫<br />
⎨L-algebras<br />
⎬<br />
{K-algebras} −→ with a G-operation from the left<br />
⎩<br />
⎭ ,<br />
such that every σ ∈ G operates σ-linearly
Sec. 3] GALOIS DESCENT 9<br />
⎧ ⎫ ⎧ ⎫<br />
⎨ central simple ⎬ ⎨ central simple algebras over L ⎬<br />
algebras<br />
⎩ ⎭ −→ with a G-operation from the left<br />
⎩<br />
⎭ ,<br />
over K<br />
such that every σ ∈ G operates σ-linearly<br />
⎧<br />
⎫<br />
{ } ⎨ commutative L-algebras<br />
⎬<br />
commutative<br />
−→ with a G-operation from the left<br />
K-algebras ⎩<br />
⎭ ,<br />
such that every σ ∈ G operates σ-linearly<br />
⎧ ⎫ ⎧ ⎫<br />
⎨ commutative ⎬ ⎨ commutative L-algebras with unit ⎬<br />
K-algebras<br />
⎩ ⎭ −→ with a G-operation from the left<br />
⎩<br />
⎭ ,<br />
with unit such that every σ ∈ G operates σ-linearly<br />
A ↦→ A ⊗ K L,<br />
ii) there is the following equivalence of categories,<br />
⎧<br />
⎫<br />
quasi-projective L-schemes<br />
{ } ⎪⎨ with a G-operation from the left ⎪⎬<br />
quasi-projective<br />
−→ by morphisms of K-schemes ,<br />
K-schemes<br />
such that every σ ∈ G operates<br />
⎪⎩<br />
⎪⎭<br />
by a morphism twisted by σ<br />
X ↦→ X × SpecK SpecL .<br />
iii) Let X be a K-scheme <strong>and</strong> r be a natural number. Then there are the following<br />
equivalences of categories,<br />
⎧<br />
⎫<br />
quasi-coherent sheaves M<br />
{ }<br />
quasi-coherent<br />
−→<br />
sheaves on X<br />
{ }<br />
locally free sheaves<br />
−→<br />
of rank r on X<br />
on X × SpecK SpecL<br />
⎪⎨<br />
⎪⎬<br />
together with a system (ι σ ) σ∈G<br />
of isomorphisms ι σ : x ∗ σ M → M ,<br />
satisfying ι ⎪⎩<br />
τ ◦ x ∗ τ (ι σ) = ι στ ⎪⎭<br />
for every σ, τ ∈ G<br />
⎧<br />
⎫<br />
locally free sheaves M of rank r<br />
on X × SpecK SpecL<br />
⎪⎨<br />
⎪⎬<br />
together with a system (ι σ ) σ∈G<br />
of isomorphisms ι σ : x ∗ σ M → M ,<br />
satisfying ι ⎪⎩<br />
τ ◦ x ∗ τ (ι σ) = ι στ ⎪⎭<br />
for every σ, τ ∈ G<br />
F ↦→ M := π ∗ F.<br />
Here the morphisms in the categories are those respecting all the extra structures.<br />
π : X × SpecK SpecL −→ X is the canonical morphism <strong>and</strong><br />
x σ : X × SpecK SpecL −→ X × SpecK SpecL
10 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
denotes the morphism induced by S(σ): SpecL → SpecL.<br />
Proof. In each case we have to prove that the functor given is fully faithful <strong>and</strong><br />
essentially surjective. Full faithfulness is proven in Propositions 3.7, 3.8, <strong>and</strong><br />
3.9, respectively. Propositions 3.3, 3.5, <strong>and</strong> 3.6 show essential surjectivity. □<br />
3.3. Proposition (Galois descent-algebraic version). —–<br />
Let L/K be a finite Galois extension of fields <strong>and</strong> G := Gal(L/K ).<br />
Let W be a vector space (an algebra, a central simple algebra, a commutative<br />
algebra, a commutative algebra with unit, ... ) over L together with an operation<br />
T : G × W → W of G from the left. Assume that T respects all extra structures<br />
<strong>and</strong> that for each σ ∈ G the action of σ is a σ-linear map T σ : W → W .<br />
Then, there is a vector space V (an algebra, a central simple algebra, a commutative<br />
algebra, a commutative algebra with unit, ... ) over K such that there is an<br />
isomorphism<br />
V ⊗ K L −→ b<br />
∼=<br />
W .<br />
Here, V ⊗ K L is equipped with the G-operation induced by the canonical operation<br />
on L. b respects all extra structures <strong>and</strong> the operation of G.<br />
Proof. Define V := W G . This is clearly a K-vector space (a K-algebra, a<br />
commutative K-algebra, a commutative K-algebra with unit). If W is a central<br />
simple algebra over L then V is a central simple algebra over K. The latter can<br />
not be seen directly but follows immediately from the formula W G ⊗ K L = W<br />
which we prove below. For that, let {l 1 , . . . , l n } be a K-basis of L.<br />
The assertion follows from the claim below.<br />
□<br />
3.4. Claim. –––– There exist an index set J <strong>and</strong> a subset {x j | j ∈ J } ⊂ W G such<br />
that {l i x j | i ∈ {1, . . . , n}, j ∈ J } is a K-basis of W .<br />
Proof. By Zorn’s Lemma, there exists a maximal subset {x j | j ∈ J } ⊂ W G<br />
such that {l i x j | i ∈ {1, . . . , n}, j ∈ J } ⊂ W is a system of K-linearly<br />
independent vectors. Assume, that system is not a basis of W . Then,<br />
〈l i x j | i ∈ {1, . . . , n}, j ∈ J 〉 K<br />
is a proper G-invariant L-sub-vector space of W <strong>and</strong> one can choose an element<br />
x ∈ W \ 〈l i x j | i ∈ {1, . . . , n}, j ∈ J 〉 K . For every l ∈ L, the sum<br />
∑ T σ (lx) = ∑ σ(l) · T σ (x)<br />
σ∈G σ∈G
Sec. 3] GALOIS DESCENT 11<br />
is G-invariant. Further, by linear independence of characters, the matrix<br />
⎛<br />
⎞<br />
σ 1 (l 1 ) . . . σ 1 (l n )<br />
⎜<br />
⎝<br />
.<br />
· · · ..<br />
⎟ · · · ⎠<br />
σ n (l 1 ) . . . σ n (l n )<br />
is of maximal rank. In particular, there is some l ∈ L such that the image of<br />
x β := ∑ σ(l) · σ(x)<br />
σ∈G<br />
in W /〈l i x j | i ∈ {1, . . . , n}, j ∈ J 〉 K is not equal to zero. Therefore,<br />
{l i x j | i ∈ {1, . . . , n}, j ∈ J } ∪ {l i x β | i ∈ {1, . . . , n} }<br />
is a K-linearly independent system of vectors contradicting the maximality<br />
of {x j | j ∈ J }.<br />
□<br />
3.5. Proposition (Galois descent-geometric version). —–<br />
Let L/K be a finite Galois extension of fields <strong>and</strong> G := Gal(L/K ).<br />
Further, let Y be a quasi-projective L-scheme together with an operation of G from<br />
the left by twisted morphisms. I.e., such that the diagrams<br />
Y<br />
T σ<br />
Y<br />
SpecL<br />
S(σ)<br />
SpecL<br />
commute where S(σ): SpecL → SpecL is the morphism induced by σ −1 : L → L.<br />
Then, there exists a quasi-projective K-scheme X such that there is an isomorphism<br />
of L-schemes<br />
X × SpecK SpecL −→ f<br />
∼=<br />
Y .<br />
Here, X × SpecK SpecL is equipped with the G-operation induced by the operation<br />
on SpecL <strong>and</strong> f is compatible with the operation of G.<br />
Proof. Affine Case. Let Y = SpecB be an affine scheme. The G-operation<br />
on SpecB corresponds to a G-operation from the right on B such that each<br />
σ ∈ G operates σ −1 -linearly. Define a G-operation from the left on B<br />
by σ · b := bσ −1 . The assertion follows immediately from Proposition 3.3.<br />
GeneralCase. ByLemma3.10, thereexistsanaffineopencovering {Y 1 , . . . , Y n }<br />
of Y by G-invariant schemes. Galois descent yields affine K-schemes<br />
X 1 , . . . , X n such that there are isomorphisms<br />
X i × SpecK SpecL ∼ =<br />
−→ Y i
12 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
<strong>and</strong> affine K-schemes X ij , 1 ≤ i < j ≤ n, such that there are isomorphisms<br />
X ij × SpecK SpecL ∼ =<br />
−→ Y i ∩ Y j .<br />
It is to be shown that every X ij admits canonical, open embeddings into X i<br />
<strong>and</strong> X j .<br />
In each case, on the level of rings we have a homomorphism A⊗ K L −→ B ⊗ K L<br />
<strong>and</strong> an isomorphism B ⊗ K L ∼ =<br />
−→ (A ⊗ K L) f such that their composition is the<br />
localization map. Clearly, f may be assumed to be G-invariant. I.e., we may<br />
suppose f ∈ A. Consequently, B ⊗ K L ∼ = A f ⊗ K L <strong>and</strong>, by consideration of<br />
the G-invariants on both sides, B ∼ = A f .<br />
The cocycle relations are obvious. Hence, we can glue the affine schemes<br />
X 1 , . . . , X n along the affine schemes X ij for 1 ≤ i < j ≤ n to obtain the<br />
scheme X desired. Lemma 3.12.ix) below completes the proof. □<br />
3.6. Proposition (Galois descent for quasi-coherent sheaves). —–<br />
Let L/K be a finite Galois extension of fields <strong>and</strong> G := Gal(L/K ).<br />
Further, let X be a K-scheme, π : X × SpecK SpecL → X the canonical morphism<br />
<strong>and</strong> x σ : X × SpecK SpecL → X × SpecK SpecL be the morphism induced by<br />
S(σ): SpecL → SpecL.<br />
Let M bea quasi-coherentsheaf over X × SpecK SpecL together with a system (ι σ ) σ∈G<br />
of isomorphisms ι σ : x ∗ σM → M which are compatible in the sense that for each<br />
σ, τ ∈ G there is the relation ι τ ◦ x ∗ τ(ι σ ) = ι στ .<br />
Then, there exists a quasi-coherent sheaf F over X such that there is an isomorphism<br />
under which the canonical isomorphism<br />
π ∗ F ∼ =<br />
−→ M<br />
i σ : x ∗ σπ ∗ F = (πx σ ) ∗ F = π ∗ F id<br />
−→ π ∗ F<br />
is identified with ι σ for each σ. I.e., the diagrams<br />
b<br />
x ∗ σπ ∗ F<br />
x∗ σ(b)<br />
x ∗ σM<br />
are commutative.<br />
i σ<br />
<br />
π ∗ F<br />
b<br />
M<br />
Proof. First step. Assume X ∼ = SpecA to be an affine scheme.<br />
Then, M = ˜M for some (A ⊗ K L)-module M. We have<br />
x ∗ σ M =<br />
˜<br />
M ⊗ (A⊗K L) (A ⊗ K L σ−1 ) = M ˜ ⊗ L L σ−1 .<br />
ι σ
Sec. 3] GALOIS DESCENT 13<br />
Hence, x ∗ σM = ˜M σ−1 where M σ−1 coincides with M as an A-module. But its<br />
structure of an L-vector space is given by<br />
l ·M σ −1 m := σ−1 (l) ·M m.<br />
Consequently, the isomorphism ι σ : x ∗ σM → M is induced by an A-module<br />
isomorphism j σ : M → M which is σ −1 -linear. The compatibility relations,<br />
required above, are easily translated into the condition that the maps j σ form a<br />
G-operation on M from the right.<br />
Define a G-operation from the left on M by σ ·m := j σ −1(m). By Galois descent<br />
for vector spaces, theK-vector spaceM G ofG-invariants satisfies M G ⊗ K L ∼ = M.<br />
Thisisalsoanisomorphism ofA-modulesastheG-operationonM iscompatible<br />
with the A-operation. Putting F := ˜M G , we obtain a quasi-coherent sheaf<br />
over X such that π ∗ F ∼ = M .<br />
The commutativity of the diagram is a consequence of Proposition 3.3.<br />
Second step. In general, consider an affine open covering<br />
for X α<br />
∼ = SpecRα .<br />
X = [ α∈I<br />
X α<br />
For every intersection X α1 ∩ X α2 , we consider an affine open covering<br />
{X α1 ,α 2 ,β ∼ = SpecR α1 ,α 2 ,β | β ∈ J α1 ,α 2<br />
}.<br />
By the affine case, for each α ∈ I, we are given an R α -module M α such that<br />
π ∗ ˜M α<br />
∼ = M |Xα × SpecK Spec L. Further, for each triple (α 1 , α 2 , β) with α 1 , α 2 ∈ I<br />
<strong>and</strong> β ∈ J α1 ,α 2<br />
, we have R α1 ,α 2 ,β-modules M α1 ,α 2 ,β satisfying<br />
π ∗ ˜ Mα1 ,α 2 ,β ∼ = M | Xα1 ,α 2 ,β× SpecK SpecL .<br />
The construction of these modules is compatible with restriction to affine subschemes.<br />
Therefore, by Proposition 3.9, we get isomorphisms<br />
i α1 ,α 2 ,β : M α1 ⊗ Rα1 R α1 ,α 2 ,β<br />
∼=<br />
−→ M α2 ⊗ Rα2 R α1 ,α 2 ,β.<br />
It is clear that, for every α 1 , α 2 , α 3 ∈ I <strong>and</strong> every β 1 ∈ J α2 ,α 3<br />
, β 2 ∈ J α3 ,α 1<br />
,<br />
<strong>and</strong> β 3 ∈ J α1 ,α 2<br />
, these isomorphisms are compatible on the triple intersection<br />
X α1 ,α 2 ,β 3<br />
∩ X α3 ,α 1 ,β 2<br />
∩ X α2 ,α 3 ,β 1<br />
. I.e., we can glue the quasi-coherent sheaves ˜M α<br />
along the ˜M α1 ,α 2 ,β to obtain the quasi-coherent sheaf M desired. □<br />
3.7. Proposition (Galois descent for homomorphisms). —–<br />
Let L/K be a finite Galois extension of fields <strong>and</strong> G := Gal(L/K ). Then, it is<br />
equivalent
14 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
i) to give a homomorphism r : V → V ′ of K-vector spaces (of algebras over K,<br />
of central simple algebras over K, of commutative K-algebras, of commutative<br />
K-algebras with unit, ... ),<br />
ii) to give a homomorphism r L : V ⊗ K L → V ′ ⊗ K L of L-vector spaces (of algebras<br />
over L, ofcentralsimplealgebrasover L, ofcommutative L-algebras, ofcommutative<br />
L-algebras with unit, ... ) which is compatible with the G-operations. I.e., such that<br />
for each σ ∈ G the diagram<br />
V ⊗ K L<br />
r L<br />
V ′ ⊗ K L<br />
commutes.<br />
σ<br />
V ⊗ K L<br />
r L<br />
σ<br />
V ′ ⊗ K L<br />
Proof. If r is given then one defines r L := r ⊗ K L. Clearly, if r is a ring<br />
homomorphism then r L is, too.<br />
Conversely, in order to construct r from r L , note that the commutativity of<br />
the diagrams implies that r L is compatible with G-invariants. Further, we<br />
know (V ⊗ K L) G = V <strong>and</strong> (V ′ ⊗ K L) G = V ′ . Thus, we obtain a K-linear<br />
map r : V −→ V ′ . If r L is a ring homomorphism then its restriction r is, too.<br />
It is clear that the two procedures described are inverse to each other.<br />
□<br />
3.8. Proposition (Galois descent for morphisms of schemes). —–<br />
Let L/K be a finite Galois extension of fields <strong>and</strong> G := Gal(L/K ).<br />
Then, it is equivalent<br />
i) to give a morphism of K-schemes f : X → X ′ ,<br />
ii) to give a morphism of L-schemes f L : X × SpecK SpecL → X ′ × SpecK SpecL<br />
which is compatible with the G-operations. I.e., such that for each σ ∈ G the<br />
diagram<br />
X × SpecK SpecL<br />
f L<br />
X ′ × SpecK SpecL<br />
commutes.<br />
σ<br />
X × SpecK SpecL<br />
f L<br />
σ<br />
X ′ × SpecK SpecL<br />
Proof. Let f be given. Then, one defines f L := f × SpecK SpecL.<br />
Conversely, in order to construct f from f L , observe that the question is<br />
local in X ′ <strong>and</strong> X. Thus, we may assume we are given a homomorphism
Sec. 3] GALOIS DESCENT 15<br />
r L : A ′ ⊗ K L −→ A ⊗ K L of L-algebras with unit making the diagrams<br />
A ′ ⊗ K L<br />
r L<br />
A ⊗ K L<br />
σ<br />
A ′ ⊗ K L<br />
r L<br />
σ<br />
A ⊗K L<br />
commute. This is exactly the situation covered by Proposition 3.7.<br />
It is clear that the two processes described are inverse to each other.<br />
□<br />
3.9. Proposition (Galois descent for morphisms of quasi-coherent sheaves). —–<br />
Let L/K be a finite Galois extension of fields <strong>and</strong> G := Gal(L/K ). Further, let X<br />
be a K-scheme <strong>and</strong> π : X × SpecK SpecL → X be the canonical morphism.<br />
Then, it is equivalent<br />
i) to give a morphism r : F → G of coherent sheaves over X,<br />
ii) to give a morphism r L : π ∗ F → π ∗ G of quasi-coherent sheaves over<br />
X × SpecK SpecL which is compatible with the G-operations. I.e., such that for<br />
each σ ∈ G the diagram<br />
commutes.<br />
π ∗ F<br />
x ∗ σ<br />
<br />
π ∗ F<br />
Proof. If r is given then one defines r L := π ∗ r.<br />
r L<br />
r L<br />
π ∗ G<br />
x ∗ σ<br />
<br />
π ∗ G<br />
Conversely, if r L is given then the question to construct r is local in X. Thus, assume<br />
A is a commutative ring with unit <strong>and</strong> M <strong>and</strong> N are A-modules. We are<br />
given a homomorphism r L : M ⊗ K L → N ⊗ K L of A ⊗ K L-modules such that<br />
the diagrams<br />
r L<br />
M ⊗ K L N ⊗ K L<br />
σ<br />
M ⊗ K L<br />
r L<br />
σ<br />
N ⊗K L<br />
commute for each σ ∈ G. We get a morphism r : M → N as M <strong>and</strong> N are the<br />
A-modules of G-invariants on the left <strong>and</strong> right h<strong>and</strong> side, respectively.<br />
The two procedures described above are inverse to each other.<br />
3.10. Lemma. –––– Let L be a field <strong>and</strong> Y be a quasi-projective L-scheme equipped<br />
with an operation of some finite group G acting by morphisms of schemes.<br />
Then, there exists a covering of Y by G-invariant affine open subsets.<br />
Proof. Let y ∈ Y be an arbitrary closed point. Everything which is needed is an<br />
affine open G-invariant subset containing y. For that, we choose an embedding<br />
□
16 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
i : Y ֒→ P N L . By Sublemma 3.11, there exists a hypersurface H y such that<br />
H y ⊃ i(Y ) \ i(Y ) <strong>and</strong> i(σ(y)) ∉ H y for every σ ∈ G. Here, i(Y ) denotes the<br />
closure of i(Y ) in P N L . By construction, the morphism<br />
i| Y \i −1 (H y ) : Y \ i −1 (H y ) −→ P N L \ H y<br />
is a closed embedding. As P N L \ H y is an affine scheme, Y \ i −1 (H y ) must be<br />
affine, too. Hence,<br />
O y := \<br />
σ −1 (Y \ i −1 (H y )) ⊂ Y<br />
σ∈G<br />
is the intersection of finitely many affine open subsets in a quasi-projective <strong>and</strong>,<br />
therefore, separated scheme. Thus, O y is an affine open subset. By construction,<br />
O y is G-invariant <strong>and</strong> contains y.<br />
□<br />
3.11. Sublemma. –––– Let L be a field <strong>and</strong> Z P N L be a Zariski-closed subset.<br />
Further, let p 1 , . . . , p n ∈ P N L be finitely many closed <strong>points</strong> not contained in Z.<br />
Then, there exists a hypersurface H ⊂ P N L which contains Z but does not contain<br />
any of the <strong>points</strong> p 1 , . . . , p n .<br />
Proof. We will give two proofs, an elementary one <strong>and</strong> a more canonical one<br />
which uses cohomology of coherent sheaves.<br />
First proof. Let S := L[X 0 , . . . , X N ] be the homogeneous coordinate ring for<br />
the projective space P N L . S is a graded L-algebra. For d ∈Æ, we will denote by<br />
S d the L-vector space of homogeneous elements of degree d.<br />
We proceed by induction. The case n = 0 is trivial. Thus, assume that the<br />
assertion is true for n − 1 <strong>and</strong> consider n <strong>points</strong> p 1 , . . . , p n . By induction<br />
hypothesis, there exists a homogeneous element s ∈ S, i.e., a hypersurface<br />
H := V (s) of a certain degree d, such that H ⊇ Z <strong>and</strong> p 1 , . . . , p n−1 ∉ H .<br />
We may assume p n ∈ H as, otherwise, the proof would be complete.<br />
Z ∪ {p 1 } ∪ . . . ∪ {p n−1 } is a Zariski closed subset of P N L not containing p n.<br />
Therefore, there exist some d ′ ∈Æ<strong>and</strong> some homogeneous s ′ ∈ S d ′ such that<br />
V (s ′ ) ⊇ Z ∪ {p 1 } ∪ . . . ∪ {p n−1 } but p n ∉ V (s ′ ). Every hypersurface<br />
V (a · s ′d + b · s d ′ ) for a, b ∈ L non-zero contains Z but neither p 1 , . . . , p n−1 ,<br />
nor p n .<br />
Second proof. Tensoring the canonical exact sequence<br />
0 −→ I {p1 , ... ,p n } −→ O X −→ O {p1 , ... ,p n } −→ 0<br />
with the ideal sheaf I Z yields an exact sequence<br />
0 −→ I {p1 , ... ,p n }∪Z −→ I Z −→ O {p1 , ... ,p n } −→ 0
Sec. 3] GALOIS DESCENT 17<br />
of coherent sheaves on P N L .<br />
For each d ∈, we tensor with the invertible sheaf O(d) <strong>and</strong> find a long<br />
cohomology exact sequence<br />
Γ(P N L , I Z(d)) −→ Γ(P N L , O {p 1 , ... ,p n }(d)) −→ H 1 (P N L , I {p 1 , ... ,p n }∪Z(d)) .<br />
By Serre’s vanishing theorem [Ha77, Theorem III.5.2],<br />
for d ≫ 0. Hence, there is a surjection<br />
H 1 (P N L , I {p 1 , ... ,p n }∪Z(d)) = 0<br />
Γ(P N L , I Z (d)) −→ Γ(P N L , O {p1 , ... ,p n }(d)) ∼ = κ(p 1 ) ⊕ · · · ⊕ κ(p n ) .<br />
That means that there exists a global section s of I Z (d) which does not vanish<br />
in any of the <strong>points</strong> p 1 , . . . , p n . The section s defines a hypersurface of degree<br />
d in P N L containing Z but not containing any of the <strong>points</strong> p 1 , . . . , p n . □<br />
3.12. Lemma (A. Grothendieck <strong>and</strong> J. Dieudonné). —–<br />
LetL/K beafinitefield extension<strong>and</strong> X beaK-schemesuchthatX × SpecK SpecL is<br />
i) reduced,<br />
ii) irreducible,<br />
iii) quasi-compact,<br />
iv) locally of finite type,<br />
v) of finite type,<br />
vi) locally Noetherian,<br />
vii) Noetherian,<br />
viii) proper,<br />
ix) quasi-projective,<br />
x) projective,<br />
xi) affine,<br />
or<br />
xii) regular.<br />
Then, X admits the same property.<br />
Proof. Let π : X × SpecK SpecL → X denote the canonical morphism. For iii)<br />
through xi), we may assume L/K to be Galois. Put G := Gal(L/K ).<br />
i) If 0 ≠ s ∈ Γ(U,O X ) were a nilpotent section of the structure sheaf over some<br />
open subset U ⊆ X then π ♯ (s) ∈ Γ(U × SpecK SpecL, O X ×SpecK SpecL) would be a<br />
nilpotent non-zero local section of the structure sheaf of X × SpecK SpecL.
18 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
ii) If X = X 1 ∪ X 2 were a decomposition into two closed proper subschemes<br />
then X × SpecK SpecL = X 1 × SpecK SpecL ∪ X 2 × SpecK SpecL would be, too.<br />
iii) Let {U α | α ∈ A } be an arbitrary affine open covering of X. Then,<br />
{U α × SpecK SpecL | α ∈ A }<br />
is an affine open covering of X × SpecK SpecL. Quasi-compactness guarantees<br />
the existence of a finite sub-covering {U α × SpecK SpecL | α ∈ A 0 }.<br />
Then, {U α | α ∈ A 0 } is a finite affine open covering for X.<br />
iv) We may assume X = SpecA to be affine. Then, by assumption, A ⊗ K L<br />
is a finitely generated L-algebra. Let {b 1 , . . . , b n } be a system of generators<br />
for A ⊗ K L. Decomposing into elementary tensors, if necessary, we may<br />
assume without restriction that all b i = a i ⊗ l i are elementary. Consider the<br />
homomorphism<br />
π : K [X 1 , . . . , X n ] → A<br />
of K-algebras given by X i ↦→ a i . π induces a surjection after tensoring with L.<br />
It is, therefore, surjective as ⊗ K L is a faithful functor.<br />
v) This is just the combination of iii) <strong>and</strong> iv).<br />
vi) Let X α<br />
∼ = SpecAα be an affine open subscheme of X. We have to<br />
show that A α is Noetherian under the hypothesis that A α ⊗ K L is. Assume<br />
I 1 ⊂ I 2 ⊂ I 3 ⊂ . . . is an ascending chain of ideals in A α which does<br />
not stabilize. It induces an ascending chain I 1 ⊗ K L ⊂ I 2 ⊗ K L ⊂ I 3 ⊗ K L ⊂ . . .<br />
of ideals in A α ⊗ K L that does not stabilize, either. This is a contradiction.<br />
vii) This is the combination of iii) <strong>and</strong> vi).<br />
viii) By virtue of v), X is of finite type. We apply the valuative criterion<br />
for properness. Consider a commutative diagram<br />
U<br />
i<br />
T<br />
X<br />
SpecK<br />
where T = SpecR is the spectrum of a valuation ring, U = SpecF is the<br />
spectrum of its quotient field <strong>and</strong> i is the canonical morphism. Taking the base<br />
change to SpecL, we find<br />
U × SpecK SpecL<br />
X × SpecK SpecL<br />
ι<br />
i<br />
<br />
T × SpecK SpecL<br />
SpecL .<br />
Here, the morphism ι in the diagonal is the unique one making the diagram commute.<br />
Note that U × SpecK SpecL = Spec(F ⊗ K L) is no longer the spectrum
Sec. 3] GALOIS DESCENT 19<br />
of a field but rather the union of finitely many spectra of fields. Similarly,<br />
T × SpecK SpecL = Spec(R ⊗ K L) is a finite union of spectra of valuation rings.<br />
Further, there is a natural G-operation on the whole diagram without ι. As ι is<br />
uniquely determined by the condition that it makes the diagram commute, the<br />
diagrams<br />
ι<br />
T × SpecK SpecL X × SpecK SpecL<br />
σ<br />
T × SpecK SpecL<br />
ι<br />
σ<br />
X × SpecK SpecL<br />
are commutative for each σ ∈ G. By Proposition 3.8, ι is the base change of a<br />
morphism T → X.<br />
ix) <strong>and</strong> x) Taking v) <strong>and</strong> viii) into account, everything left to be shown<br />
is the existence an ample invertible sheaf on X. By assumption, there<br />
is an ample invertible sheaf M on X × SpecK SpecL. For σ ∈ G let<br />
x σ : X × SpecK SpecL → X × SpecK SpecL be the morphism of schemes induced by<br />
S(σ): SpecL → SpecL, i.e., by σ −1 on coordinate rings. The invertible sheaves<br />
x ∗ σ M are ample, too. Therefore, M := N υ∈G x ∗ υM is an ample invertible sheaf.<br />
For each σ ∈ G, there are canonical identifications<br />
ι σ : x ∗ σ M = O υ∈G<br />
x ∗ σ x∗ υ M = O υ∈G<br />
x ∗ υσ M<br />
id<br />
−→ O υ∈G<br />
x ∗ υ M = M .<br />
Obviously, these are compatible in the sense that, for each σ, τ ∈ G, there is<br />
the relation ι τ ◦x ∗ τ(ι σ ) = ι στ . By Galois descent for locally free sheaves, there is<br />
an invertible sheaf L ∈ Pic(X ) such that π ∗ L ∼ = M . Lemma 3.13 shows that<br />
L is ample.<br />
xi) We suppose that X × SpecK SpecL ∼ =<br />
−→ SpecB is affine. The isomorphism<br />
defines a G-operation from the right on the L-algebra B such that each σ ∈ G<br />
operates σ −1 -linearly. Define a G-operation from the left on B by σ·b := b·σ −1 .<br />
Proposition 3.3 provides us a K-algebra A such that there is an isomorphism<br />
X × SpecK SpecL ∼ =<br />
∼=<br />
−→ SpecB = Spec(A ⊗ K L) −→ SpecA × SpecK SpecL ,<br />
compatible with the G-operations. Proposition 3.5 implies X ∼ = SpecA.<br />
xii) Let (A, m) be the local ring at a point p ∈ X. By vi), we may assume (A, m)<br />
is Noetherian. We have to show A is regular. Let us give two proofs for that,<br />
the st<strong>and</strong>ard one using Serre’s homological characterization of regularity <strong>and</strong> an<br />
elementary one.<br />
First proof. Assume, by contradiction, (A, m) were not regular. Then, [Mat,<br />
Theorem 19.2] shows gl. dim A = ∞. Further, by [Mat, §19, Lemma 1],
20 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
proj. dim A<br />
A/m = ∞ <strong>and</strong><br />
On the other h<strong>and</strong>, we have<br />
sup {i | Tor A i (A/m, A/m) ≠ 0 } = ∞.<br />
Tor A i (A/m, A/m) ⊗ K L = Tor A i (A/m, A/m) ⊗ A (A ⊗ K L)<br />
= Tor A i (A/m, (A ⊗ K L/m ⊗ K L))<br />
= Tor A⊗ KL<br />
i ((A ⊗ K L/m ⊗ K L), (A ⊗ K L/m ⊗ K L))<br />
= 0<br />
for i > dim A ⊗ K L as A ⊗ K L is regular. This is a contradiction.<br />
Note that we only used that π is faithfully flat.<br />
Second proof. Put d := dim A. The ring A ⊗ K L is not local, in general.<br />
But the quotient (A ⊗ K L)/(m⊗ K L) ∼ = A/m⊗ K L is a direct product of finitely<br />
many fields since L/K is a finite, separable field extension. The quotients<br />
(A ⊗ K L)/(m n ⊗ K L) are Artin rings as they are flat of relative dimension zero<br />
over A/m n . Consequently, they are direct products of Artin local rings,<br />
(A ⊗ K L)/(m n ⊗ K L)<br />
∼=<br />
−→ A (n)<br />
1<br />
× · · · × A (n)<br />
l .<br />
Under this isomorphism, (m ⊗ K L)/(m n ⊗ K L) is mapped to the product of the<br />
maximal ideals m (n)<br />
1<br />
× · · · × m (n)<br />
l as it is a nilpotent ideal <strong>and</strong> the quotient has to<br />
be a direct product of fields.<br />
For each i ∈ {1, . . . , l}, let f i ∈ A ⊗ K L be the element which maps to the<br />
st<strong>and</strong>ard vector e i = (0, . . . , 0, 1, 0, . . . , 0). Then,<br />
A (n)<br />
i<br />
∼= ((A ⊗ K L)/(m n ⊗ K L)) fi<br />
∼ = (A ⊗K L) fi /m n (A ⊗ K L) fi .<br />
By assumption, (A ⊗ K L) fi is a regular local ring <strong>and</strong> m(A ⊗ K L) fi is its maximal<br />
ideal. (A ⊗ K L) fi is of dimension d as it is flat of relative dimension zero<br />
over A. The Hilbert-Samuel function of a regular local ring is st<strong>and</strong>ard. We have<br />
length ( A (n)<br />
i<br />
) = length ((A ⊗K L) fi /m n (A ⊗ K L) fi )<br />
= length ((A ⊗ K L) fi /(m(A ⊗ K L) fi ) n )<br />
( ) n + d − 1<br />
= .<br />
d<br />
Consequently, for the function H A such that H A (n) := dim K (A/m n ), one gets<br />
(<br />
H A (n) = dim L (A ⊗K L)/(m n ⊗ K L) )<br />
( )<br />
= dim L A<br />
(n)<br />
1<br />
× · · · × A (n)<br />
l
Sec. 3] GALOIS DESCENT 21<br />
=<br />
=<br />
l<br />
∑<br />
i=1<br />
l<br />
∑<br />
i=1<br />
( )<br />
dim L A<br />
(n)<br />
i<br />
[(<br />
A<br />
(n)<br />
i<br />
/m (n)<br />
i<br />
) ] ( )<br />
: L · length A<br />
(n)<br />
i .<br />
)<br />
. How-<br />
This shows that H A is a constant multiple of n ↦→ ( n+d−1<br />
d<br />
ever, since H A (1) = 1, there is no ambiguity about constant factors. A is regular.<br />
□<br />
3.13. Lemma. –––– Let L/K be an arbitrary field extension, X a K-scheme of<br />
finite type, π : X × SpecK SpecL → X the canonical morphism <strong>and</strong> L ∈ Pic(X ) an<br />
invertible sheaf. Suppose that the pull-back π ∗ L ∈ Pic(X × SpecK SpecL) is ample.<br />
Then, L is ample.<br />
Proof. We have to prove that, for every coherent sheaf F on X <strong>and</strong> every<br />
closed point x ∈ X, the canonical map px n : Γ(X, F ⊗ L ⊗n ) → F x ⊗ Lx<br />
⊗n is<br />
surjective for n ≫ 0. For this, it is obviously sufficient to prove surjectivity of<br />
But, as is easy to see,<br />
p n x ⊗ K L : Γ(X, F ⊗ L ⊗n ) ⊗ K L → (F x ⊗ L ⊗n<br />
x ) ⊗ K L .<br />
Γ(X, F ⊗ L ⊗n ) ⊗ K L = Γ(X × SpecK SpecL, π ∗ F ⊗ π ∗ L ⊗n )<br />
while (F x ⊗ L n<br />
x ) ⊗ K L = Γ(π −1 (x), π ∗ F ⊗ π ∗ L ⊗n ). Here, π −1 (x) denotes the<br />
fiber of π above x. For n ≫ 0, the map p n x ⊗ K L is surjective as π ∗ L is ample.<br />
□<br />
3.14. Lemma. –––– Let R be a ring, F ≠ 0 a free R-module of finite rank <strong>and</strong> M<br />
an arbitraryR-module. Suppose that M⊗ R F is a locally free R-module of finite rank.<br />
Then, M is locally free of finite rank.<br />
Proof. The R-module M⊗ R F is locally free of finite rank. Therefore, there exists<br />
some affine open covering {SpecR f1 , . . . , SpecR fn } of SpecR such that each<br />
(M ⊗ R F )⊗ R R fi = (M ⊗ R R fi ) ⊗ R F<br />
is a free R fi -module. M ⊗ R R fi = M fi is a direct summ<strong>and</strong>. Therefore, M fi is a<br />
projective R fi -module.<br />
We have a surjection (M ⊗ R F )⊗ R R fi ։ M fi , the kernel K of which is a direct<br />
summ<strong>and</strong> of (M ⊗ R F ) ⊗ R R fi , too. In particular, K is a finitely generated<br />
R fi -module. Hence, M fi is finitely presented <strong>and</strong>, therefore, locally free by<br />
[Mat, Theorem 7.12 <strong>and</strong> Theorem 4.10].<br />
□
22 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
3.15. Remark. –––– Galois descent is a central technique in André Weil’s foundation<br />
of Algebraic Geometry. The Grothendieck school gave a far-reaching<br />
generalization of it, the so-called faithful flat descent. It turns out that Galois<br />
descent is sufficient for our presentation of the relationship between central simple<br />
algebras <strong>and</strong> Brauer-Severi varieties. For that reason, we decided to present<br />
it in detail. Faithful flat descent will play a main role in the next chapter.<br />
4. Central simple algebras <strong>and</strong> non-abelian H 1<br />
We are going to make use of the following well-known facts about central<br />
simple algebras.<br />
4.1. Lemma (J. H. Maclagan-Wedderburn, R. Brauer). —– Let K be a field.<br />
a) Let A be a central simple algebra over K. Then, there exist a skew field D with<br />
center K <strong>and</strong> a natural number n such that A ∼ = M n (D) is isomorphic to the full<br />
algebra of n × n-matrices with entries in D.<br />
b) Let L be a field extension of K <strong>and</strong> A be a central simple algebra over K. Then,<br />
A ⊗ K L is a central simple algebra over L.<br />
c) Assume K to be separably closed. Let D be a skew field being finite dimensional<br />
over K whose center is equal to K. Then, D = K.<br />
Proof. See the st<strong>and</strong>ard literature, for example [Lan93], [Bou-A], or [Ke].<br />
4.2. Remarks. –––– a) Let A be a central simple algebra over a field K.<br />
i) The proof of Lemma 4.1.a) shows that in the presentation A ∼ = M n (D) the<br />
skew field D is unique up to isomorphism of K-algebras <strong>and</strong> the natural number<br />
n is unique.<br />
ii) A ⊗ K K sep is isomorphic to a full matrix algebra over K sep . In particular,<br />
dim K A is a perfect square. The natural number ind(A) := √ dim K (D) is called<br />
the index of A.<br />
b) Let A 1 , A 2 be central simple algebras over a field K. Then A 1 ⊗ K A 2 can be<br />
shown to be a central simple algebra over K. Further, if A is a central simple<br />
algebra over a field K then A ⊗ K A op ∼ = Aut K-Vect (A). I.e., it is isomorphic to a<br />
matrix algebra.<br />
c) Two central simple algebras A 1<br />
∼ = Mn1 (D 1 ), A 2<br />
∼ = Mn2 (D 2 ) over a field K are<br />
said to be similar if the corresponding skew fields D 1 <strong>and</strong> D 2 are isomorphic<br />
as K-algebras. This is an equivalence relation on the set of all isomorphism<br />
classes of central simple algebras over K. The tensor product induces a group<br />
structure on the set of similarity classes of central simple algebras over K, this<br />
is the so-called Brauer group Br(K ) of the field K.<br />
□
Sec. 4] CENTRAL SIMPLE ALGEBRAS AND NON-ABELIAN h 1 23<br />
4.3. Definition. –––– Let K be a field <strong>and</strong> A be a central simple algebra over K.<br />
A field extension L of K admitting the property that A ⊗ K L is isomorphic to<br />
a full matrix algebra is said to be a splitting field for A. In this case, one says that<br />
A splits over L.<br />
4.4. Lemma (Theorem of Skolem-Noether). —–<br />
Let R be a commutative ring with unit. Then, GL n (R) operates on M n (R) by<br />
conjugation,<br />
(g, m) ↦→ gmg −1 .<br />
If R = L is a field then this defines an isomorphism<br />
PGL n (L) := GL n (L)/L ∗ ∼ =<br />
−→ Aut L (M n (L)) .<br />
Proof. One has L = Zent(M n (L)). Therefore, the mapping is well-defined <strong>and</strong><br />
injective.<br />
Surjectivity. Let j : M n (L) → M n (L) be an automorphism. We consider the<br />
algebra<br />
M := M n (L) ⊗ L M n (L) op ( ∼ = M n 2(L)).<br />
M n (L) gets equipped with the structure of a left M-module in two ways.<br />
(A ⊗ B) • 1 C := A · C · B<br />
(A ⊗ B) • 2 C := j(A) · C · B<br />
Two M n 2(L)-modules of the same L-dimension are isomorphic, as the<br />
n 2 -dimensional st<strong>and</strong>ard L-vector space equipped with the canonical operation<br />
of M n 2(L) is the only simple left M n 2(L)-module <strong>and</strong> there are no non-trivial<br />
extensions. Thus, there is an isomorphism h : (M n (L), • 1 ) → (M n (L), • 2 ).<br />
Let us put I := h(E) to be the image of the identity matrix.<br />
M ∈ M n (L) we have<br />
h(M ) = h((E ⊗ M ) • 1 E) = (E ⊗ M ) • 2 h(E) = h(E) · M = I · M.<br />
In particular, I ∈ GL n (L). Therefore,<br />
I · M = h(M ) = h((M ⊗ E) • 1 E) = (M ⊗ E) • 2 h(E) = j(M ) · I<br />
For every<br />
for each M ∈ M n (L) <strong>and</strong> j(M ) = IMI −1 .<br />
□<br />
4.5. Definition. –––– Let n be a natural number.<br />
i) If K is a field then we will denote by Az K n the set of all isomorphism classes<br />
of central simple algebras A of dimension n 2 over K.
24 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
ii) Let L/K be a field extension. Then Azn<br />
L/K will denote the set of all isomorphism<br />
classes of central simple algebras A which are of dimension n 2 over K<br />
<strong>and</strong> split over L. Obviously, Az K n := S Az L/K<br />
L/K<br />
n .<br />
4.6. Theorem (cf. J.-P. Serre: Corps locaux [Se62, chap. X, §5]). —–<br />
Let L/K be a finite Galois extension of fields, G := Gal(L/K ) its Galois group,<br />
<strong>and</strong> n ∈Æ.<br />
Then, there is a natural bijection of pointed sets<br />
a = a L/K<br />
n<br />
: Az L/K<br />
n<br />
∼=<br />
−→ H 1 (G, PGL n (L)) ,<br />
A ↦→ a A .<br />
Proof. Let A be a central simple algebra over K which splits over L,<br />
The diagrams<br />
A ⊗ K L ∼ =<br />
−→ M n (L) .<br />
f<br />
A ⊗ K L f M n (L)<br />
σ<br />
σ<br />
do not commute, in general.<br />
A ⊗ K L f M n (L)<br />
For each σ ∈ G, define a σ ∈ PGL n (L) by putting ( f ◦ σ) = a σ ◦ (σ ◦ f ).<br />
It turns out that<br />
f ◦ στ = ( f ◦ σ) ◦ τ<br />
= a σ ◦ (σ ◦ f ) ◦ τ<br />
= a σ ◦ σ ◦ ( f ◦ τ )<br />
= a σ ◦ σ ◦ (a τ ◦ (τ ◦ f ))<br />
= a σ ◦ σ a τ ◦ (στ ◦ f ) .<br />
I.e., a στ = a σ · σa τ <strong>and</strong> (a σ ) σ∈G is a cocycle.<br />
If one starts with another isomorphism f ′ : A⊗ K L −→ M n (L) then there exists<br />
some b ∈ PGL n (L) such that f = b ◦ f ′ . The equality ( f ◦ σ) = a σ ◦ (σ ◦ f )<br />
implies<br />
f ′ ◦ σ = b −1 ◦ f ◦ σ = b −1 · a σ ◦ (σ ◦ (b ◦ f ′ )) = b −1 · a σ · σb ◦ (σ ◦ f ′ ) .<br />
Thus, the isomorphism f ′ yields a cocycle cohomologous to (a σ ) σ∈G . The<br />
mapping a is well-defined.
Sec. 4] CENTRAL SIMPLE ALGEBRAS AND NON-ABELIAN h 1 25<br />
Injectivity. Assume that A <strong>and</strong> A ′ are chosen in such a way that the construction<br />
above yields one <strong>and</strong> the same cohomology class a A = a A ′ ∈ H 1 (G, PGL n (L)).<br />
After choosing suitable isomorphisms f <strong>and</strong> f ′ , one has the equalities<br />
( f ◦ σ) = a σ ◦ (σ ◦ f ) <strong>and</strong> ( f ′ ◦ σ) = a σ ◦ (σ ◦ f ′ ) in the diagram<br />
A ⊗ K L f f<br />
M n (L) ′<br />
A ′ ⊗ K L<br />
σ<br />
A ⊗ K L f<br />
σ<br />
M n (L)<br />
f ′<br />
σ<br />
A ′ ⊗ K L .<br />
Consequently, f ◦ σ ◦ f −1 ◦ σ −1 = f ′ ◦ σ ◦ f ′−1 ◦ σ −1 <strong>and</strong>, therefore,<br />
f ◦ σ ◦ f −1 ◦ f ′ ◦ σ −1 ◦ f ′−1 = id .<br />
The outer part of the diagram commutes. Taking the G-invariants on both sides<br />
yields A ∼ = A ′ .<br />
Surjectivity. Let a cocycle (a σ ) σ∈G for H 1 (G, PGL n (L)) be given. We define a<br />
new G-operation on M n (L) as follows.<br />
Let σ ∈ G act as<br />
a σ ◦ σ : M n (L)<br />
σ a<br />
−→ M n (L) −→<br />
σ<br />
Mn (L) .<br />
Note that this is a σ-linear mapping. Further, one has<br />
(a σ ◦ σ) ◦ (a τ ◦ τ ) = a σ ◦ σ a τ ◦ στ = a στ ◦ στ .<br />
I.e., we constructed a group operation from the left. Galois descent yields the<br />
desired algebra.<br />
□<br />
4.7. Corollary. –––– Let L/K be a finite Galois extension of fields <strong>and</strong> let n be a<br />
natural number.<br />
a) Let L ′ be a field extension of L such that L ′ /K is Galois, too. Then, the following<br />
diagram of morphisms of pointed sets commutes,<br />
Az L/K<br />
n<br />
a L/K<br />
n<br />
H 1 (Gal (L/K ), PGL n (L))<br />
nat. incl.<br />
inf Gal (L′ /K)<br />
Gal(L/K)<br />
Az L′ /K<br />
n<br />
a L′ /K<br />
n<br />
H 1( Gal(L ′ /K ), PGL n (L ′ ) ) .
26 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
b)LetK ′ beanintermediatefield oftheextensionL/K. Then,thefollowingdiagram<br />
of morphisms of pointed sets commutes,<br />
Az L/K<br />
n<br />
a L/K<br />
n<br />
H 1( Gal(L/K ), PGL n (L) )<br />
⊗ K K ′ <br />
res Gal(L/K′ )<br />
Gal(L/K)<br />
Az L/K ′<br />
n<br />
a L/K ′<br />
n<br />
H 1( Gal(L/K ′ ), PGL n (L) ) .<br />
Proof. These are direct consequences of the construction of the bijections a ∗ n .<br />
□<br />
4.8. Corollary. –––– Let K be a field <strong>and</strong> n be a natural number. Then, there is a<br />
unique natural bijection<br />
a = a K n : Az K n −→ H 1( Gal(K sep /K ), PGL n (K sep ) )<br />
such that a K n | Az<br />
L/K<br />
n<br />
= a L/K<br />
n for each finite Galois extension L/K in K sep .<br />
Proof. In order to get connected to the definition of the cohomology of a<br />
profinite group, only one technical point is to be proven. That is the formula<br />
PGL n (K sep ) Gal(K sep /K ′) = PGL n (K ′ )<br />
for K ⊆ K ′ ⊆ K sep any intermediate field. To see this, observe that the<br />
exact sequence<br />
1 −→ (K sep ) ∗ −→ GL n (K sep ) −→ PGL n (K sep ) −→ 1<br />
induces a long exact sequence<br />
1→(K ′ ) ∗ → GL n (K ′ ) → PGL n (K sep ) Gal(K sep /K ′) → H 1( Gal(K sep /K ′ ), (K sep ) ∗)<br />
in cohomology. Finally, the right entry vanishes by Hilbert’s Theorem 90.<br />
Cf. Lemma 5.11.<br />
□
Sec. 4] CENTRAL SIMPLE ALGEBRAS AND NON-ABELIAN h 1 27<br />
4.9. Proposition. –––– Let K be a field <strong>and</strong> m <strong>and</strong> n be natural numbers.<br />
Then, there is a commutative diagram<br />
Az K n<br />
a K n<br />
H 1( Gal(K sep /K ), PGL n (K sep ) )<br />
A↦→M m (A)<br />
Az K mn<br />
(i n mn) ∗<br />
a K mn<br />
H 1( Gal(K sep /K ), PGL mn (K sep ) ) .<br />
Here, (imn) n ∗ is the map induced by the block-diagonal embedding<br />
imn n : PGL n (K sep ) −→ PGL mn (K sep )<br />
⎛ ⎞<br />
E 0 · · · 0<br />
0 E · · · 0<br />
E ↦→ ⎜<br />
⎝ . .<br />
..<br />
⎟ . . ⎠ .<br />
0 0 · · · E<br />
Proof. Let A ∈ Az K n . By the construction above, a cycle representing<br />
the cohomology class a K n (A) is given as follows. Choose an isomorphism<br />
f : A ⊗ K K sep → M n (K sep ) <strong>and</strong> put a σ := ( f ◦ σ) ◦ (σ ◦ f ) −1 ∈ Aut(M n (K sep ))<br />
for each σ ∈ Gal(K sep /K ).<br />
On the other h<strong>and</strong>, for M m (A) ∈ Az K mn one may choose the isomorphism<br />
M m ( f ): M m (A) ⊗ K K sep = M m (A ⊗ K K sep ) −→ M m (M n (K sep )) ∼ = M mn (K sep ).<br />
For each σ ∈ Gal(K sep /K ), this yields the automorphism ã σ of M m (M n (K sep ))<br />
which operates as a σ on each block. If a σ is given by conjugation with a matrix<br />
A σ then ã σ is given by conjugation with<br />
⎛ ⎞<br />
A σ 0 · · · 0<br />
0 A σ · · · 0<br />
⎜<br />
⎝ . .<br />
..<br />
⎟ . . ⎠ .<br />
0 0 · · · A σ<br />
This is exactly what was to be proven.<br />
□<br />
4.10. Remark. –––– The proposition above shows<br />
Br(K ) ∼ = lim −→n<br />
H 1( Gal(K sep /K ), PGL n (K sep ) ) .
28 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
Further, for each m <strong>and</strong> n, there is a commutative diagram of exact sequences<br />
as follows,<br />
1 (K sep ) ∗ GL n (K sep )<br />
PGL n (K sep )<br />
1<br />
1 (K sep ) ∗ GL mn (K sep ) PGL mn (K sep ) 1 .<br />
We note that (K sep ) ∗ is mapped into the centers of GL n (K sep ) <strong>and</strong> GL mn (K sep ), respectively.<br />
Therefore, there are boundary maps to the second group cohomology<br />
group <strong>and</strong> they are compatible with each other to give a map<br />
lim H<br />
−→ 1( Gal(K sep /K ), PGL n (K sep ) ) −→ H 2( Gal(K sep /K ), (K sep ) ∗) .<br />
n<br />
It is not complicated to show that this map is injective <strong>and</strong> surjective. Cf. Corollary<br />
II.5.3.<br />
j n mn<br />
5. Brauer-Severi varieties <strong>and</strong> non-abelian H 1<br />
5.1. Definition. –––– Let K be a field.<br />
i) A scheme X over K is called a Brauer-Severi variety if there exists a finite,<br />
separable field extension L/K such that X × SpecK SpecL is isomorphic to a<br />
projective space P N L .<br />
ii) A field extension L of K admitting the property that X × SpecK SpecL ∼ = P N L<br />
for some n ∈Æis said to be a splitting field for X. In this case, one says that<br />
X splits over L.<br />
5.2. Proposition. –––– Let X be a Brauer-Severi variety over a field K. Then,<br />
i) X is a variety. I.e., X is reduced <strong>and</strong> irreducible as a scheme.<br />
ii) X is projective <strong>and</strong> regular.<br />
iii) X is geometrically integral.<br />
iv) One has Γ(X, O X ) = K.<br />
v) K is algebraically closed in the function field X (K ).<br />
Proof. We will denote the dimension of X by N .<br />
i) <strong>and</strong> ii) are direct consequences of Lemma 3.12.i),ii), x) <strong>and</strong> xii).<br />
iii) is clear from the definition.<br />
iv) We have<br />
Γ(X, O X ) ⊗ K K sep = Γ(X × SpecK SpecK sep , O X ×SpecK SpecK sep)<br />
= Γ(P N K sep, O P N K sep)<br />
= K sep .
Sec. 5] BRAUER-SEVERI VARIETIES AND NON-ABELIAN h 1 29<br />
Consequently, Γ(X, O X ) = K.<br />
v) Assume, g ∈ K (X ) is a <strong>rational</strong> function which is algebraic over K. We fix<br />
an algebraic closure K of K.<br />
The pull-back g K<br />
is a <strong>rational</strong> function on X × SpecK SpecK ∼ = P N which is<br />
K<br />
algebraic over K. Thus, g K<br />
is a constant function. Consequently, g itself has<br />
definitely no poles. I.e., g is a regular function on X. By iv), we see g ∈ K. □<br />
5.3. Lemma. –––– Let R be a commutative ring with unit.<br />
a) Then, there is an operation of the group GL n (R) on P n−1<br />
R<br />
R-schemes as follows.<br />
by morphisms of<br />
A ∈ GL n (R) gives rise to the morphism given by the graded automorphism<br />
R[X 0 , . . . , X n−1 ] −→ R[X 0 , . . . , X n−1 ]<br />
f (X 0 , . . . , X n−1 ) ↦→ f ((X 0 , X 1 , . . . , X n−1 ) · A t )<br />
of the homogeneous coordinate ring.<br />
b) If R = L is a field then this induces an isomorphism<br />
PGL n (L)<br />
∼=<br />
−→ Aut L-schemes (P n−1<br />
L ) .<br />
Proof. a) is clear. For b), see [Ha77, Example II.7.1.1]. The proof given there<br />
works equally well without the assumption on L to be algebraically closed. □<br />
5.4. Remark. –––– Let S be any commutative R-algebra with unit. Then, on<br />
the set P n−1<br />
R (S) of S-valued <strong>points</strong>, the definition above yields the naive operation<br />
⎛ ⎞<br />
a 11 . . . a 1n<br />
⎜<br />
⎝<br />
.<br />
. ..<br />
⎟ . ⎠<br />
a n1 . . . a nn<br />
⊤<br />
↓<br />
(x 0 : . . . : x n−1 ) ↦→ ( (a 11 x 0 + . . . + a 1n x n−1 ) : . . . : (a n1 x 0 + . . . + a nn x n−1 ) )<br />
of GL n (S).<br />
5.5. Definition. –––– Let r be natural number.<br />
i) If K is a field then we will denote by BS K r the set of all isomorphism classes of<br />
Brauer-Severi varieties X of dimension r over K.<br />
ii) Let L/K be a field extension. Then, BS L/K<br />
r will denote the set of all isomorphism<br />
classes of Brauer-Severi varieties X over K which are of dimension r <strong>and</strong><br />
split over L. Obviously, BS K r := S BS L/K<br />
r .<br />
L/K
30 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
5.6. Theorem (cf. J.-P. Serre: Corps locaux [Se62, chap. X, §6). —–<br />
Let L/K be a finite Galois extension, G := Gal(L/K ) its Galois group, <strong>and</strong> n be<br />
a natural number. Then, there exists a natural bijection of pointed sets<br />
α = α L/K<br />
n−1 : BSL/K n−1<br />
∼=<br />
−→ H 1 (G, PGL n (L)),<br />
X ↦→ α X .<br />
Proof. Let X be a Brauer-Severi variety over K that splits over L,<br />
X × SpecK SpecL ∼ =<br />
−→ P n−1<br />
L .<br />
On X × SpecK SpecL, as well as on P n−1<br />
L , there are operations of G from<br />
the left by morphisms of K-schemes. The action of σ ∈ G is induced by<br />
S(σ): SpecL → SpecL in both cases. Unfortunately, the diagrams<br />
do not commute, in general.<br />
X × SpecK SpecL<br />
σ<br />
X × SpecK SpecL<br />
f<br />
f<br />
f<br />
P n−1<br />
L<br />
σ<br />
P n−1<br />
L<br />
For each σ ∈ G, define α σ ∈ PGL n (L) by putting ( f ◦ σ) = α σ ◦ (σ ◦ f ).<br />
It turns out that<br />
I.e., (α σ ) σ∈G is a cocycle.<br />
f ◦ στ = ( f ◦ σ) ◦ τ<br />
= α σ ◦ (σ ◦ f ) ◦ τ<br />
= α σ ◦ σ ◦ ( f ◦ τ )<br />
= α σ ◦ σ ◦ (α τ ◦ (τ ◦ f ))<br />
= α σ ◦ σ α τ ◦ (στ ◦ f ) .<br />
If one starts with another isomorphism f ′ : X × SpecK SpecL −→ P n−1<br />
L then<br />
there exists some b ∈ PGL n (L) such that f = b ◦ f ′ . The equality<br />
( f ◦ σ) = α σ ◦ (σ ◦ f ) implies<br />
f ′ ◦ σ = b −1 ◦ f ◦ σ = b −1 · α σ ◦ (σ ◦ (b ◦ f ′ )) = b −1 · α σ · σb ◦ (σ ◦ f ′ ) .<br />
Thus, the isomorphism f ′ yields a cocycle cohomologous to (α σ ) σ∈G . By consequence,<br />
the mapping α is well-defined.<br />
Injectivity. Assume X <strong>and</strong> X ′ are chosen such that the same cohomology class<br />
α X = α X ′ ∈ H 1 (G, PGL n (L)) arises. After choosing suitable isomorphisms f<br />
<strong>and</strong> f ′ , one has the formulas ( f ◦σ) = α σ ◦ (σ ◦ f ) <strong>and</strong> ( f ′ ◦σ) = α σ ◦ (σ ◦ f ′ )
Sec. 5] BRAUER-SEVERI VARIETIES AND NON-ABELIAN h 1 31<br />
in the diagram<br />
X × SpecK SpecL<br />
σ<br />
X × SpecK SpecL<br />
f<br />
f<br />
P n−1<br />
L<br />
σ<br />
P n−1<br />
L<br />
f ′<br />
f ′<br />
X ′ × SpecK SpecL<br />
σ<br />
X ′ × SpecK SpecL .<br />
Therefore, f ◦ σ ◦ f −1 ◦ σ −1 = f ′ ◦ σ ◦ f ′−1 ◦ σ −1 <strong>and</strong>, consequently,<br />
f ◦ σ ◦ f −1 ◦ f ′ ◦ σ −1 ◦ f ′−1 = id .<br />
The outer part of the diagram commutes. Galois descent yields X ∼ = X ′ .<br />
Surjectivity. Let a cocycle (α σ ) σ∈G for H 1 (G, PGL n (L)) be given. We define a<br />
G-operation on P n−1<br />
L by letting σ ∈ G operate as<br />
α σ ◦ σ : P n−1<br />
L<br />
σ<br />
−→ P n−1<br />
L<br />
α σ<br />
−→ P<br />
n−1<br />
L .<br />
This is a group operation as (α σ ) σ∈G is a cocycle. The geometric version of<br />
Galois descent yields the desired variety.<br />
□<br />
5.7. Corollary. –––– Let L/K be a finite Galois extension of fields <strong>and</strong> n be a<br />
natural number.<br />
a) Let L ′ be a field extension of L such that L ′ /K is Galois, too. Then, the following<br />
diagram of morphisms of pointed sets commutes.<br />
BS L/K<br />
n−1<br />
α L/K<br />
n−1<br />
H 1( Gal(L/K ), PGL n (L) )<br />
nat. incl.<br />
inf Gal(L′ /K)<br />
Gal(L/K)<br />
BS L′ /K<br />
n−1<br />
α L′ /K<br />
n−1<br />
H 1( Gal(L ′ /K ), PGL n (L ′ ) )<br />
b) LetK ′ beanintermediatefieldoftheextensionL/K. Then,thefollowingdiagram<br />
of morphisms of pointed sets commutes.<br />
BS L/K<br />
n−1<br />
α L/K<br />
n−1<br />
H 1( Gal(L/K ), PGL n (L) )<br />
× SpecK SpecK ′ <br />
BS L/K ′<br />
n−1<br />
α L/K ′<br />
res Gal(L/K′ )<br />
Gal(L/K)<br />
n−1<br />
H 1( Gal(L/K ′ ), PGL n (L) )<br />
Proof. These are direct consequences of the construction of the mappings α ∗ n−1.<br />
□
32 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
5.8. Corollary. –––– Let K be a field <strong>and</strong> n be a natural number. Then, there is a<br />
natural bijection<br />
α = α K n−1 : BSK n−1 −→ H 1( Gal(K sep /K ), PGL n (K sep ) )<br />
such that α K n−1 | BS<br />
L/K<br />
n−1<br />
= α L/K<br />
n−1 for each finite Galois extension L/K in K sep .<br />
Proof. The proof follows the same lines as the proof of Corollary 4.8.<br />
□<br />
5.9. Proposition (F. Severi, cf. J.-P. Serre: [Se62, chap. X, §6, Exc. 1]). —–<br />
Let r ∈Æ. Assume that X is a Brauer-Severi variety of dimension r over a field K.<br />
If X (K ) ≠ ∅ then X ∼ = P r K .<br />
Proof. Let L/K be a finite Galois extension which is a splitting field for X.<br />
Denote its Galois group by G. Choose an isomorphism<br />
X × SpecK SpecL ∼ =<br />
−→ P r L .<br />
Here, one may assume without restriction that the L-valued point<br />
x L ∈ X × SpecK SpecL (L)<br />
induced by the K-valued point x ∈ X (K ) is mapped to (1 : 0 : . . . : 0) ∈ P r L(L).<br />
Therefore, the cohomology class α X ∈ H 1( G, PGL r+1 (L) ) is given by a cocycle<br />
(α σ ) σ∈G such that every α σ admits (1 : 0 : . . . : 0) as a fixed point.<br />
Consequently, α X belongs to the image of H 1( G, F/L ∗) under the natural<br />
homomorphism where<br />
⎧⎛<br />
⎞<br />
⎫<br />
⎪⎨ a 11 . . . a 1,r+1<br />
⎪⎬<br />
⎜<br />
F := ⎝<br />
.<br />
. ..<br />
⎟<br />
. ⎠ ∈ GL r+1 (L)<br />
a 21 = a 31 = . . . = a r+1,1 = 0<br />
⎪⎩<br />
a r+1,1 . . . a r+1,r+1<br />
∣<br />
⎪⎭ .<br />
But it is obvious that<br />
⎧⎛<br />
⎞⎫<br />
1 a 12 . . . a 1,r+1<br />
⎪⎨<br />
F/L ∗ ∼ = F ′ 0 a 22 . . . a 2,r+1<br />
⎪⎬<br />
:= ⎜<br />
⎝<br />
.<br />
. . ..<br />
⎟ ⊆ GL r+1 (L) .<br />
. ⎠<br />
⎪⎩<br />
⎪⎭<br />
0 a r+1,2 . . . a r+1,r+1<br />
Further, the natural homomorphism H 1 (G, F ′ ) → H 1 (G, PGL r+1 (L)) factors<br />
via H 1 (G, GL r+1 (L)). The assertion follows from Lemma 5.11. □<br />
5.10. Remark. –––– One can easily show that even H 1 (G, F ′ ) = 0. Indeed,<br />
F 1 := { (a ij ) 1≤i,j≤r+1 | a ij = 0 for i ≥ 2, i ≠ j;<br />
a 11 = a 22 = · · · = a r+1,r+1 = 1 }
Sec. 5] BRAUER-SEVERI VARIETIES AND NON-ABELIAN h 1 33<br />
is a normal G-subgroup of F ′ . Thus, F ′ admits a filtration by G-sub<strong>groups</strong>,<br />
each normal in the succeeding one, such that all the subquotients occurring are<br />
isomorphic either to GL r (L) or to (L, +).<br />
5.11. Lemma (Theorem Hilbert 90). —– Let L/K be a finite Galois extension,<br />
G := Gal(L/K ) its Galois group, <strong>and</strong> n ∈Æ. Then, H 1 (G, GL n (L)) = 0.<br />
Proof. Let (a σ ) σ∈G be a cocycle with values in GL n (L). We define a G-operation<br />
from the left on L n as follows.<br />
We let σ ∈ G act as<br />
a σ ◦ σ : L n<br />
σ<br />
−→ L n<br />
a σ<br />
−→ L n .<br />
Then, it is clear that a σ ◦ σ is a σ-linear map. Galois descent yields a K-vector<br />
space V such that there is an isomorphism<br />
making the diagrams<br />
V ⊗ K L ∼ =<br />
−→<br />
b<br />
L n<br />
V ⊗ K L<br />
σ<br />
V ⊗ K L<br />
b<br />
b<br />
L n<br />
L n a σ ◦σ<br />
commute. In particular, one has that dim K V = n. The choice of an isomorphism<br />
V ∼ =<br />
−→ K n<br />
provides us with an element b ∈ GL n (L) such that b ◦ σ = a σ ◦ σ ◦ b. I.e.,<br />
a σ = b ◦ σ ◦ b −1 ◦ σ −1 = b ◦ σ (b −1 )<br />
for all σ ∈ G. (a σ ) σ∈G is cohomologous to the trivial cocycle.<br />
□<br />
5.12. Definition. –––– Let K be a field, r ∈Æ, <strong>and</strong> X be a Brauer-Severi variety<br />
of dimension r. Then, a linear subspace of X is a closed subvariety Y ⊂ X such<br />
that<br />
Y × SpecK SpecK sep ⊂ X × SpecK SpecK sep ∼ = P<br />
r<br />
K sep<br />
is a linear subspace of the projective space.<br />
This property is independent of the isomorphism chosen.<br />
5.13. Remark. –––– A K-valued point would be a zero-dimensional linear subspace.<br />
But, except for the trivial case X ∼ = P r K<br />
, there are no K-valued <strong>points</strong>.
34 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
Nevertheless, it may happen that there exist linear subspaces Y X of higher dimension.<br />
They can be investigated by cohomological methods generalizing the<br />
argument given above.<br />
5.14. Proposition (F. Châtelet <strong>and</strong> M. Artin). —– Let K be a field, r, d ∈Æ,<br />
X a Brauer-Severi variety of dimension r, <strong>and</strong> Y a linear subspace of dimension d.<br />
Then, the natural boundary maps send the cohomology classes<br />
α K r (X ) ∈ H 1( Gal(K sep /K ), PGL r+1 (K sep ) )<br />
<strong>and</strong><br />
α K d (Y ) ∈ H 1( Gal(K sep /K ), PGL d+1 (K sep ) )<br />
to the same class in the cohomological Brauer group H 2( Gal(K sep /K ), (K sep ) ∗) .<br />
Proof. Let L be a common splitting field for X <strong>and</strong> Y. We may assume that<br />
L/K is a finite Galois extension. Denote its Galois group by G.<br />
We choose an isomorphism X × SpecK SpecL ∼ =<br />
−→ P r L . We may assume without<br />
restriction that the linear subspace<br />
Y × SpecK SpecL ⊂ X × SpecK SpecL ∼ = P r L<br />
is given by the homogeneous equations X d+1 = · · · = X r = 0. Then, the<br />
cohomology class α L/K<br />
r (X ) ∈ H 1( G, PGL r+1 (L) ) is given by a cocycle (α σ ) σ∈G<br />
such that “X d+1 = · · · = X r = 0” is an invariant subspace for every α σ .<br />
Consequently, α L/K<br />
r (X ) is in the image of H 1( G, F/L ∗) under the natural<br />
homomorphism for<br />
{( )<br />
E1 H<br />
F :=<br />
∈ GL r+1 (L)<br />
0 E 2<br />
∣ E }<br />
1 ∈ GL d+1 (L), E 2 ∈ GL r−d (L),<br />
.<br />
H ∈ M (d+1)×(r−d) (L)<br />
F comes equipped with the homomorphism of G-<strong>groups</strong><br />
p : F −→ GL d+1 (L),<br />
( )<br />
E1 H<br />
↦→ E 1 .<br />
0 E 2<br />
Thus, we obtain a commutative diagram that connects the three central inclusions,<br />
GL d+1 (L)<br />
L ∗<br />
p<br />
L ∗<br />
F<br />
L ∗ GL r+1 (L) .<br />
i
Sec. 6] CENTRAL SIMPLE ALGEBRAS AND BRAUER-SEVERI VARIETIES 35<br />
Therefore, there is the following commutative diagram uniting the boundary<br />
maps,<br />
H 1 (G, PGL d+1 (L)) H 2 (G, L ∗ )<br />
p ∗<br />
<br />
H 1 (G, F/L ∗ )<br />
H 2 (G, L ∗ )<br />
i ∗<br />
<br />
H 1 (G, PGL r+1 (L)) H 2 (G, L ∗ ).<br />
Finally, recall that we have a cohomology class α ∈ H 1 (G, F/L ∗ ) such that<br />
i ∗ (α) = αr<br />
L/K (X ) <strong>and</strong> p ∗ (α) = αd L/K (Y ). □<br />
6. Central simple algebras <strong>and</strong> Brauer-Severi varieties<br />
6.1. Theorem. –––– Let n ∈Æ, K a field <strong>and</strong> A a central simple algebra over K<br />
of dimension n 2 .<br />
a) Then,thereexistsaBrauer-SeverivarietyX A ofdimension (n−1)overK satisfying<br />
condition (∗). (∗) determines X A uniquely up to isomorphism of K-schemes.<br />
(∗)If L is a splitting field for A which is finite <strong>and</strong> Galois over K then L is a<br />
splitting field for X A , too. There is one <strong>and</strong> the same cohomology class<br />
associated with A <strong>and</strong> X A .<br />
a A = α XA ∈ H 1( Gal(L/K ), PGL n (L) )<br />
b) The assignment X : A ↦→ X A admits the following properties.<br />
i) X is compatible with extensions K ′ /K of the base field. I.e.,<br />
X A⊗K K ′ ∼ = XA × SpecK SpecK ′ .<br />
ii) L is a splitting field for A if <strong>and</strong> only if L is a splitting field for X A .<br />
Proof. a) Uniqueness is clear from the results of the preceding sections.<br />
Existence. Choose a finite Galois extension L of K which is a splitting field<br />
for A. Take condition (∗) as a definition for X A . This is independent of the<br />
choice of L by Corollaries 4.7.a) <strong>and</strong> 5.7.a).
36 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
b.i) Let K ′ /K be an arbitrary field extension. We have the obvious diagram of<br />
field extensions<br />
LK ′ <br />
<br />
K ′ <br />
L<br />
K<br />
<br />
<strong>and</strong> the canonical inclusion Gal(LK ′ /K ′ ) ⊆ Gal(L/K ). Under the constructions<br />
given above, one assigns to the central simple algebra A ⊗ K K ′ <strong>and</strong> the<br />
Brauer-Severivariety X × SpecK SpecK ′ the restrictions of the cohomology classes<br />
assigned to A, respectively X. I.e.,<br />
a (A⊗K K ′ ) = res Gal(LK ′ /K ′ )<br />
Gal(L/K )<br />
(a A ),<br />
α (X ×SpecK SpecK ′) = resGal(LK ′ /K ′ )<br />
Gal(L/K )<br />
(α X ).<br />
ii) The two statements are equivalent to<br />
<strong>and</strong><br />
res Gal(LK ′ /K ′ )<br />
Gal(L/K )<br />
(a A ) ∈ H 1( Gal(LK ′ /K ′ ), PGL n (LK ′ ) ) = 0<br />
res Gal(LK ′ /K ′ )<br />
Gal(L/K )<br />
(α XA ) ∈ H 1( Gal(LK ′ /K ′ ), PGL n (LK ′ ) ) = 0,<br />
respectively. As we have a A = α XA , the assertion follows.<br />
□<br />
6.2. Corollary. –––– Let K be a field <strong>and</strong> A be a central simple algebra over K.<br />
Then, K ′ is a splitting field for A if <strong>and</strong> only if X A (K ′ ) ≠ ∅.<br />
Proof. “=⇒” is trivial <strong>and</strong> “⇐=” is an easy consequence of Proposition 5.9. □<br />
6.3. Corollary. –––– i) Let K be a field <strong>and</strong> n ∈Æ. Then, X induces a bijection<br />
X K n : AzK n → BS K n−1 .<br />
ii) Let L/K be a field extension. Then, X induces a bijection<br />
X L/K<br />
n<br />
: Az L/K<br />
n<br />
→ BS L/K<br />
n−1 .<br />
iii) These mappings are compatible with extensions of the base field. I.e., the diagram<br />
Az K n<br />
X K n<br />
BS K n−1<br />
⊗ K K ′ <br />
Az K ′<br />
n<br />
X K ′<br />
n<br />
BS K ′<br />
n−1<br />
× SpecK SpecK ′
Sec. 6] CENTRAL SIMPLE ALGEBRAS AND BRAUER-SEVERI VARIETIES 37<br />
commutes for every field extension K ′ /K.<br />
Proof. i) follows immediately from Theorem 6.1.a). ii) is a consequence of<br />
Theorem 6.1.a) together with Theorem 6.1.b.ii). iii) is simply a reformulation<br />
of Theorem 6.1.b.i).<br />
□<br />
6.4. Remark. –––– It may happen that two Brauer-Severi varieties X 1 , X 2 are<br />
bi<strong>rational</strong>ly equivalent but not isomorphic. S. A. Amitsur [Am55] proved in<br />
this case that the corresponding central simple algebras A 1 <strong>and</strong> A 2 generate the<br />
same cyclic subgroup of Br(K ). It is an open question whether the converse<br />
is true. Interesting partial results have been obtained by P. Roquette [Roq64]<br />
<strong>and</strong> S. L. Tregub [Tr].<br />
6.5. Proposition. –––– Let K be a field, n ∈Æ, <strong>and</strong> A a central simple algebra of<br />
dimension n 2 over K. Then, there is an isomorphism<br />
x A : Aut K (A)<br />
∼=<br />
−→ Aut K-schemes (X A ).<br />
Proof. Choose a splitting field L for A which is finite <strong>and</strong> Galois over K.<br />
Put G := Gal(L/K ). Choose an isomorphism A ⊗ K L −→ f<br />
M n (L). Then, for<br />
each σ ∈ G, there is a commutative diagram<br />
A ⊗ K L<br />
f<br />
M n (L)<br />
σ<br />
A ⊗ K L<br />
f<br />
a σ ◦σ<br />
M n (L) .<br />
The L-linear maps a σ : M n (L) → M n (L) form a cocycle for the cohomology<br />
class a A ∈ H 1 (G, PGL n (L)).<br />
Further, by Galois descent, to give an element of Aut K (A) is equivalent to giving<br />
an element of PGL n (L) which is invariant under a σ ◦ σ for every σ ∈ G.<br />
As α XA = a A , there exists an isomorphism X A × SpecK SpecL −→ f ′<br />
P n−1<br />
L such<br />
that the diagrams<br />
X A × SpecK SpecL<br />
f ′<br />
P n−1<br />
L<br />
σ<br />
X A × SpecK SpecL<br />
f ′<br />
P n−1<br />
L<br />
commute. Therefore, by Galois descent for morphisms of schemes, it is equivalent<br />
to give an element of Aut K-schemes (X A ) or to give an element of PGL n (L)<br />
which is invariant under a σ ◦ σ for every σ ∈ G.<br />
□<br />
a σ ◦σ
38 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
6.6. Proposition (F. Châtelet <strong>and</strong> M. Artin). —– Let K be a field, n, d ∈Æ,<br />
<strong>and</strong> A be a central simple algebra of dimension n 2 over K.<br />
Then, the Brauer-Severi variety X A associated with A admits a linear subspace of<br />
dimension d if <strong>and</strong> only if d ≤ n − 1 <strong>and</strong><br />
d ≡ −1<br />
(mod ind(A)).<br />
Proof. We write A = M m (D) with a skew field D <strong>and</strong> put e := ind(A).<br />
Clearly, n = me.<br />
“=⇒” Let H ⊂ X A be a linear subspace of dimension d. By Proposition 5.14,<br />
we know H ∼ = X A ′ for some central simple algebra A ′ which is similar to A.<br />
I.e.,A ′ ∼ = M k (D)for acertaink ∈Æ. Itfollowsthatdim A ′ = k 2·dim D = k 2·e 2<br />
<strong>and</strong> dim H = dim X A ′ = k · e − 1. This implies the congruence desired.<br />
Further, we have dim X A = n − 1. Consequently,<br />
d = dim H ≤ dim X A = n − 1.<br />
“⇐=”LetL beasplittingfieldforA whichisfinite<strong>and</strong> GaloisoverK. Denoteits<br />
Galois group by G. Further, let k be the natural number such that d = k ·e −1.<br />
By assumption, k · e − 1 = d ≤ n − 1 = m · e − 1, hence k ≤ m.<br />
We consider the cohomology class a D ∈ H 1 (G, PGL e (L)), associated with D.<br />
By Proposition 4.9, one has a A = (i e me ) ∗(a D ) where i e me : PGL e(L) → PGL me (L)<br />
is the block-diagonal embedding. Let (a σ ) σ∈G be a cocycle representing the<br />
cohomology class a D . Then, (i e me (a σ)) σ∈G is a cocycle that represents a A .<br />
We define a G-operation on P me−1<br />
L<br />
by letting σ ∈ G act as<br />
i e me (a σ) ◦ σ : P me−1<br />
L<br />
σ<br />
P me−1<br />
L<br />
i e me(a σ )<br />
This is a group operation since (i e me (a σ)) σ∈G is a cocycle.<br />
P me−1<br />
L .<br />
The geometric version of Galois descent yields the Brauer-Severi variety X A .<br />
Further, the G-operation keeps invariant the linear subspace defined by the<br />
homogeneous equations X ke = X ke+1 = · · · = X me−1 = 0. Thus, Galois<br />
descent may be applied to this subspace, too. It yields a variety Y of dimension<br />
ke − 1 = d.<br />
Galois descent for morphisms of schemes, applied to the canonical embedding,<br />
yields a morphism Y −→ X A . This is a closed immersion as that property<br />
descends under faithful flat base change. Consequently, Y is a linear subvariety<br />
of X A of the dimension desired.<br />
□
Sec. 7] FUNCTORIALITY 39<br />
7. Functoriality<br />
7.1. Remark. –––– The preceding results suggest that X : A ↦→ X A should<br />
somehow be a functor. This is indeed the case. But for that, the construction<br />
of the Brauer-Severi variety associated with a central simple algebra, which was<br />
given above, is not sufficient. The problem is that X A is determined by its<br />
cohomology class only up to isomorphism, not up to unique isomorphism.<br />
Thus, in order to make X a functor, it would still be necessary to make choices.<br />
For that reason, we aim at a more natural description of X A .<br />
7.2. Lemma (A. Grothendieck). —– Let n ∈Æ<strong>and</strong> R a commutative ring<br />
with unit. Then, there is a bijection<br />
{ }<br />
submodules M in R<br />
κ R : P n−1<br />
R (R) −→ G(R) :=<br />
n such that R n /M<br />
is a locally free R-module of rank (n − 1)<br />
subject to the conditions below.<br />
i) κ R is natural in R. I.e., for every ring homomorphism i : R → R ′ , the diagram<br />
commutes.<br />
P n−1<br />
R (R)<br />
P n−1 (i)<br />
<br />
P n−1<br />
R ′ (R ′ )<br />
κ R<br />
κ R ′<br />
G(R)<br />
G(R ′ )<br />
G(i): M ↦→M ⊗ R R ′<br />
ii) For every a ∈ PGL n (R), the canonical operations of a on P n−1<br />
R (R) <strong>and</strong> G(R)<br />
are compatible with κ R . That means that the diagrams<br />
commute.<br />
P n−1<br />
R (R)<br />
a<br />
P n−1<br />
R<br />
(R)<br />
κ R<br />
κ R<br />
G(R)<br />
a<br />
G(R)<br />
Here, a ∈ PGL n (R) acts on a submodule M ⊂ R n by matrix multiplication from<br />
the left. I.e., by M ↦→ a · M for a ∈ GL n (R) a representative of a.<br />
Proof. A submodule M ⊆ R n such that the quotient R n /M is locally free of<br />
rank (n − 1) defines an R-valued point in P n−1<br />
R .<br />
To see this, let first m ⊆ R be an arbitrary maximal ideal. Then, Rm/M n m is a<br />
free R m -module of rank (n − 1). In particular, it is projective. Hence, M m is<br />
a direct summ<strong>and</strong> of Rm n <strong>and</strong>, therefore, projective <strong>and</strong> of finite presentation.<br />
Consequently, by [Mat, Theorem 7.12], M m is a free R m -module of rank one.<br />
As R n /M is R-flat, there is an exact sequence<br />
0 −→ M ⊗ R (R/m) −→ R n /mR n −→ R n /(M + mR n ) −→ 0.
40 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
Thus, the canonical map M/mM → R n /mR n is injective. I.e., mM = mR n ∩M.<br />
In particular, M can not be contained in mR n . Hence, the R m -module M m<br />
which is free of rank one is not contained in mRm.<br />
n Consequently, if<br />
M m = 〈(r 0 , . . . , r n−1 )〉 then there is some α ∈ {0, . . . , n − 1} such that<br />
r α ∈ R m is a unit.<br />
This is equivalent to<br />
R n m/(M m + e 0 · R m + · · · + e α−1 · R m + e α+1 · R m + · · · + e n−1 · R m ) = 0<br />
where e 0 , . . . , e n−1 ∈ R n m denote the st<strong>and</strong>ard elements with an index shift<br />
by (−1). The latter is an open condition on m by [Mat, Theorem 4.10].<br />
Therefore, there exists some f ∈ R\m such that<br />
R n f = M f + e 1 · R f + · · · + e α−1 · R f + e α+1 · R f + · · · + e n · R f .<br />
I.e., such that the (α + 1)-th projection M f → R f is surjective.<br />
As, on both sides, there are locally free modules of rank one, this must<br />
be an isomorphism. Consequently, M f is free <strong>and</strong> there is a generator of<br />
type (r 0 , . . . , r α−1 , 1, r α+1 , . . . , r n−1 ). Taking the entries as homogeneous<br />
coordinates, we get a morphism SpecR f → P n−1<br />
R .<br />
It is easy to see that all these can be glued together to give a morphism<br />
SpecR → P n−1<br />
R of R-schemes.<br />
Conversely, an R-valued point SpecR → P n−1<br />
R defines a quotient module of R n<br />
of rank (n − 1) as follows. Cover SpecR by affine, open subsets<br />
SpecR =<br />
n−1<br />
[<br />
α=0<br />
SpecR α<br />
such that for each α the image i(SpecR α ) is contained in the st<strong>and</strong>ard affine<br />
set “X α ≠ 0”. Then, i| SpecRα can be given in the form<br />
(r 0 : r 1 : . . . : r α−1 : 1 : r α+1 : . . . : r n−1 ) .<br />
But R n α/(r 0 , r 1 , . . . , r α−1 , 1, r α+1 , . . . , r n−1 ) is a free R α -module of rank (n−1)<br />
for trivial reasons.<br />
The compatibility stated as i) follows directly from the construction given.<br />
ii) is clear.<br />
7.3. Definition. –––– Let R be a commutative ring with unit.<br />
i) Let X be an R-scheme. Then, by<br />
P X : {R-schemes} op → {sets}<br />
□
Sec. 7] FUNCTORIALITY 41<br />
we will denote the functor defined by<br />
T ↦→ Mor R-schemes (T , X )<br />
on objects <strong>and</strong> by composition on morphisms.<br />
ii) Let F : {R-schemes} op → {sets} be any functor. If, for some R-scheme X,<br />
∼=<br />
there is an isomorphism ι: F −→ P X then we will say F is represented by X.<br />
Having fixed the isomorphism ι then, by Yoneda’s Lemma, the K-scheme representing<br />
the functor is determined up to unique isomorphism.<br />
7.4. Remarks. –––– a) The functors P X depend on X in a natural manner.<br />
I.e., if p : X → X ′ is a morphism of R-schemes then there is a morphism of<br />
functors p ∗ : P X → P X ′ given by composition with p. Therefore, there is a<br />
covariant functor<br />
P : {R-schemes} −→ Fun ({R-schemes} op , {sets})<br />
described by X ↦→ P X on objects.<br />
b) The functor P X induced by an R-scheme X is exactly what in category theory<br />
is usually called the Hom-functor. Nevertheless, we will follow the st<strong>and</strong>ards in<br />
Algebraic Geometry <strong>and</strong> refer to it as the functor of <strong>points</strong>. Note that P X (T ) is<br />
the set of all T -valued <strong>points</strong> on X.<br />
7.5. Corollary (A. Grothendieck). —– Let n ∈Æ<strong>and</strong> R be a commutative ring<br />
with unit. Then, the functor<br />
P n−1<br />
R : {R-schemes} −→ {sets}<br />
T ↦→ P n−1<br />
R (T ) :=<br />
is represented by the R-scheme P n−1<br />
R .<br />
⎧<br />
⎫<br />
⎨ subsheaves M in OT n such that ⎬<br />
OT n /M is a locally free<br />
⎩<br />
⎭<br />
O T -module of rank (n − 1)<br />
Proof. It is clear that both functors satisfy the sheaf axiom for Zariski coverings.<br />
Therefore, it is sufficient to construct the isomorphism on the full subcategory<br />
of affine R-schemes. This is exactly what is done in Lemma 7.2. □<br />
7.6. Corollary-Definition. –––– Let n ∈Æ, L be a field, <strong>and</strong> F be an L-vector<br />
space of dimension n. Put F := ˜F to be the coherent sheaf associated to F<br />
on SpecL.
42 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
Then, the functor<br />
P(F ): {L-schemes} −→ {sets}<br />
⎧<br />
⎫<br />
⎨ subsheaves M ⊂ π ∗ F such that ⎬<br />
(π : T → SpecL) ↦→ P(F )(π) := π ∗ F/M is a locally free<br />
⎩<br />
⎭<br />
O T -module of rank (n − 1)<br />
is representable by an L-scheme which is isomorphic to P n−1<br />
L . We will denote<br />
that scheme by P(F ) <strong>and</strong> call it the projective space of lines in F.<br />
7.7. Remark. –––– If L ′ is a field containing L then P(F ⊗ L L ′ ) is naturally<br />
isomorphic to the functor P(F )| L ′ -schemes. Therefore, there is a canonical isomorphism<br />
P(F ⊗ L L ′ ∼=<br />
) −→ P(F ) × SpecL SpecL ′<br />
between the representing objects. For each L ′ -scheme U, the diagram of natural<br />
mappings<br />
Mor L ′ -schemes(U, P(F )) {M ⊂ πU ∗ F }<br />
∼=<br />
<br />
Mor L ′ -schemes(U, P(F ⊗ L L ′ )) {M ⊂ πU ∗ F }<br />
commutes.<br />
Consequently, for each L-scheme T , there is a commutative diagram<br />
Mor L-schemes (T , P(F ))<br />
{M ′ ⊂ π ∗ T F }<br />
comp. with projection<br />
<br />
Mor L ′ -schemes(T × SpecL SpecL ′ , P(F ))<br />
pull-back<br />
∼=<br />
<br />
Mor L ′ -schemes(T × SpecL SpecL ′ , P(F ⊗ L L ′ )) {M ⊂ π ∗ T × SpecL Spec L ′F } .<br />
Note that the column on the left is, up to the canonical isomorphism above,<br />
exactly the base extension from SpecL to SpecL ′ .<br />
7.8. Definition. –––– Let L/K be a finite Galois extension of fields <strong>and</strong><br />
G := Gal(L/K ) be its Galois group.<br />
i) By L-Vect K , we will denote the category of all finite dimensional<br />
L-vector spaces. As morphisms, we allow all injections which are σ-linear<br />
for a certain σ ∈ G.<br />
ii) L-Vect K 0 is a subcategory ofL-Vect K which consists of the same class of objects.<br />
In L-Vect K 0 , only the bijections are taken as morphisms.
Sec. 7] FUNCTORIALITY 43<br />
7.9. Definition. –––– Let H be a group. Then, by H we will denote the category<br />
consisting of exactly one object ∗ such that Mor(∗, ∗) = H .<br />
7.10. Definition. –––– Let L/K be a finite Galois extension of fields <strong>and</strong> G be<br />
its Galois group. By Sch L/K , we will denote the category of all L-schemes with<br />
morphisms twisted by any element of G.<br />
We note that Sch L/K has a canonical structure of a fibered category over G.<br />
The pull-back under σ ∈ G is given by X ↦→ X × SpecL SpecL σ−1 .<br />
7.11. Lemma. –––– Let L/K be a finite Galois extension of fields <strong>and</strong> G be its<br />
Galois group.<br />
i) Then, P is a covariant functor from L-Vect K to Sch L/K . Here, a σ-linear<br />
monomorphism i : F → F ′ for σ ∈ G induces a morphism<br />
which is twisted by σ.<br />
i ∗ = P(i): P(F ) −→ P(F ′ )<br />
ii) LetF beanL-vectorspace<strong>and</strong> i : F → F bethemultiplicationwithx ∈ L, x ≠ 0.<br />
Then, i ∗ : P(F ) → P(F ) is equal to the identity morphism.<br />
Proof. i) Let i : F → F ′ be a σ-linear monomorphism. We have to consider<br />
the diagram<br />
P(F ) P(F ′ )<br />
SpecL<br />
S(σ)<br />
SpecL .<br />
By Yoneda’s Lemma, there has to be constructed a natural transformation<br />
i + : P(F ) → P(F ′ )(S(σ) ◦ · ) .<br />
I.e., for each L-scheme π : T → SpecL there is a mapping<br />
i + (π): P(F )(π) → P(F ′ )(S(σ) ◦ π)<br />
to be given such that for each morphism p : π 1 → π 2 of L-schemes the diagram<br />
(†)<br />
P(F )(π 2 )<br />
i + (π 2 )<br />
P(F ′ )(S(σ) ◦ π 2 )<br />
commutes.<br />
P(F )(p)<br />
P(F )(π 1 )<br />
i + (π 1 )<br />
P(F ′ )(S(σ)◦p)<br />
P(F ′ )(S(σ) ◦ π 1 )
44 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
We construct i + (π) as follows. There is the description<br />
P(F )(π) = M ⊂ π ∗ F | π ∗ F/M is locally free of rank (dim F − 1) }<br />
{ }<br />
M ⊂ (S(σ) ◦ π)<br />
=<br />
∗ S(σ −1 ) ∗ F |<br />
(S(σ) ◦ π) ∗ S(σ −1 ) ∗ F/M is locally free of rank (dim F − 1) .<br />
Here, S(σ −1 ) ∗ F = ˜F σ for F σ equal to F as an abelian group <strong>and</strong> equipped<br />
with the scalar multiplication given by l · f := σ(l) f . Hence, i gives rise<br />
to an L-linear monomorphism i : F σ → F ′ <strong>and</strong>, therefore, to a morphism<br />
ĩ : S(σ −1 ) ∗ F → F ′ of sheaves on SpecL.<br />
If M is a subsheaf of (S(σ) ◦ π) ∗ S(σ −1 ) ∗ F such that (S(σ) ◦ π) ∗ S(σ −1 ) ∗ F/M<br />
is locally free of rank (dim F − 1) then ( (S(σ) ◦ π) ∗ (ĩ) ) (M ) is a subsheaf of<br />
(S(σ)◦π) ∗ F ′ such that the quotient is locally free of rank (dim F ′ −1). This gives<br />
rise to a morphism of functors i + : P(F ) −→ P(F ′ )(S(σ) ◦ · ) as desired.<br />
By consequence, we have a morphism i ∗ : P(F ) −→ P(F ′ ) of schemes making<br />
the diagram (†) commute.<br />
ii) is clear from the construction.<br />
7.12. Corollary. –––– Let L/K be a finite Galois extension of fields <strong>and</strong><br />
G := Gal(L/K ) its Galois group.<br />
i) Then, P can be made into a contravariant functor from L-Vect K 0 to Sch L/K .<br />
Here, a σ-linear isomorphism i : F → F ′ for σ ∈ G induces a morphism<br />
which is twisted by σ −1 .<br />
i ∗ : P(F ′ ) → P(F )<br />
ii) LetF beanL-vectorspace<strong>and</strong> i : F → F bethemultiplicationwithx ∈ L, x ≠ 0.<br />
Then, i ∗ : P(F ) → P(F ) is equal the identity morphism.<br />
Proof. For an isomorphism i : F → F ′ , put i ∗ := i −1<br />
∗ . □<br />
7.13. Remark. –––– The morphisms i ∗ can also be constructed directly. Indeed,<br />
the task is to give a natural transformation i + : P(F ′ ) → P(F )(S(σ −1 ) ◦ · ).<br />
There are the descriptions<br />
<strong>and</strong><br />
P(F ′ )(π) = { M ⊂ π ∗ F ′ | π ∗ F ′ /M is locally free of rank (dim F ′ − 1) }<br />
P(F )(S(σ −1 ) ◦ π) =<br />
{ }<br />
M ⊂ π ∗ S(σ −1 ) ∗ F |<br />
π ∗ S(σ −1 ) ∗ .<br />
F/M is locally free of rank (dim F − 1)<br />
If M is a subsheaf of π ∗ F ′ such that the quotient π ∗ F ′ /M is locally free of<br />
rank (dim F ′ − 1) = (dim F − 1) then π ∗ (ĩ) −1 (M ) is a subsheaf of π ∗ S(σ −1 ) ∗ F<br />
□
Sec. 7] FUNCTORIALITY 45<br />
such that the quotient is locally free of rank (dim F − 1). This gives rise to a<br />
morphism of functors i + : P(F ′ ) −→ P(F )(S(σ −1 ) ◦ · ) as desired.<br />
7.14. Remark. –––– If L is a field <strong>and</strong> A is a central simple algebra of dimension<br />
n 2 over L then all the non-zero, simple, left A-modules are isomorphic<br />
to each other. Further, if l is a non-zero, simple, left A-module then each<br />
automorphism of l is given by multiplication with an element of the center<br />
of A.<br />
Hence, every automorphism of l induces the identity map on P(l). In particular,<br />
two arbitrary isomorphisms i 1 , i 2 : l → l ′ between non-zero, simple,<br />
left A-modules induce one <strong>and</strong> the same isomorphism i ∗ 1 = i ∗ 2 : P(l′ ) → P(l).<br />
Thismeans, A determinesP(l)notonlyuptoisomorphism, butuptouniqueisomorphism.<br />
7.15. Definition. –––– Let r ∈Æ<strong>and</strong> L/K be a finite Galois extension.<br />
Denote the Galois group Gal(L/K ) by G.<br />
i) By Mat L/K<br />
r , we will denote the category of all split central simple algebras of<br />
dimension r 2 over L. I.e., of all algebras isomorphic to M r (L). As morphisms,<br />
wetakeallhomomorphismsofK-algebraswhichareσ-linear for acertain σ ∈ G<br />
<strong>and</strong> preserve the unit element.<br />
ii) P L/K<br />
r will denote the subcategory of Sch L/K consisting of all L-schemes isomorphic<br />
to the projective space P r L. As morphisms, P L/K<br />
r allows only the isomorphisms.<br />
7.16. Remark. –––– In Mat K r<br />
two objects are isomorphic.<br />
, every morphism is an isomorphism <strong>and</strong> every<br />
7.17. Proposition. –––– Let n ∈Æ<strong>and</strong> L/K be a finite Galois extension.<br />
Denote the Galois group Gal(L/K ) by G.<br />
: Mat L/K<br />
n → P L/K<br />
n−1 .<br />
is given by A ↦→ P(l) where l ⊂ A is a non-zero, simple,<br />
i) There is an equivalence of categories Ξ L/K<br />
n<br />
On objects, Ξ L/K<br />
n<br />
left A-module. If i : A → A ′ is a morphism in Mat L/K<br />
n<br />
morphism of schemes induced by the canonical homomorphism l → A ′ ⊗ A l.<br />
then Ξ L/K<br />
n<br />
(i) is the<br />
ii) If i : A → A ′ is σ-linear for σ ∈ G then Ξ L/K<br />
n (i) is a morphism twisted by σ.<br />
Proof. If l is a non-zero, simple, left A-module then A ′ ⊗ A l is a non-zero, simple,<br />
left A ′ -module. Both are n-dimensional L-vector spaces. Up to isomorphism,<br />
these modules are unique. Therefore, the morphism of schemes<br />
i l∗ := Ξ L/K<br />
n<br />
(i): Ξn<br />
L/K (A) = P(l) −→ P(A ′ ⊗ A l) = Ξn L/K (A ′ )
46 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
induced by the homomorphism<br />
i l : l → A ′ ⊗ A l,<br />
x ↦→ 1 ⊗ x<br />
is well-defined.<br />
If i : A → A ′ is a σ-linear homomorphism then l → A ′ ⊗ A l is a σ-linear<br />
homomorphism of vector spaces. By Lemma 7.11, the morphism Ξn<br />
L/K (i) is<br />
twisted by σ. This proves ii).<br />
As i is invertible, Ξ L/K<br />
n<br />
(i) is an isomorphism of schemes. Consequently, Ξn<br />
L/K is<br />
is an equivalence of<br />
a functor between the categories described. To prove Ξn<br />
L/K<br />
categories, we have to show it is fully faithful <strong>and</strong> essentially surjective. As in<br />
P L/K<br />
n−1<br />
every two objects are isomorphic, essential surjectivity is clear.<br />
For full faithfulness, it will suffice to prove that<br />
Ξ L/K<br />
n | Aut(A) : Aut(A) −→ Aut(X A )<br />
is an isomorphism of <strong>groups</strong> in the case A = M n (L). For that, note<br />
<strong>and</strong><br />
Thus, the functor Ξ L/K<br />
n<br />
Aut(A) = PGL n (L) ⋊ G<br />
Aut(Ξn<br />
L/K (A)) = Aut K-schemes (P n−1<br />
L ) = PGL n (L) ⋊ G .<br />
induces on Aut(A) a group homomorphism<br />
Ξ: PGL n (L) ⋊ G → PGL n (L) ⋊ G .<br />
By Lemma 7.2.ii), the restriction of Ξ to PGL n (L) is the identity. Statement<br />
ii) shows that the quotient map G → G is the identity, too. Consequently,<br />
Ξn L/K | Aut(A) is an isomorphism. □<br />
7.18. Corollary. –––– The categories Mat L/K<br />
n<br />
<strong>and</strong> P L/K<br />
n−1<br />
are anti-equivalent, too.<br />
There is an equivalence of categories ˇΞ n<br />
L/K : Mat L/K<br />
n → (P L/K<br />
n−1 )op given on objects by<br />
A ↦→ P(l) where l ⊂ A is a non-zero, simple, left A-module.<br />
If i : A → A ′ is σ-linear for σ ∈ G then ˇΞ L/K<br />
n (i) is twisted by σ −1 .<br />
Proof. Put simply ˇΞ n<br />
L/K (i) := (Ξn L/K (i)) −1 . □<br />
7.19. Proposition. –––– Let n ∈Æ, K a field <strong>and</strong> A be a central simple algebra of<br />
dimension n 2 over K. Then, the Brauer-Severi variety X A associated with A may<br />
be described as follows.<br />
Let L be a splitting field for A which is finite <strong>and</strong> Galois over K. By covariant<br />
functoriality, the canonical Gal(L/K )-operation on A ⊗ K L induces an operation
Sec. 7] FUNCTORIALITY 47<br />
on the projective space Ξn<br />
L/K (A ⊗ K L). Here, σ ∈ Gal(L/K ) acts by a morphism<br />
twisted by σ. The geometric version of Galois descent yields the K-scheme X A .<br />
Proof. Choose an isomorphism f : A ⊗ K L → M n (L). Then, there is a cocycle<br />
(a σ ) σ∈G from G to PGL n (L) such that for each σ ∈ G the diagram<br />
A ⊗ K L<br />
f<br />
M n (L)<br />
to the whole situation, we obtain com-<br />
commutes. Applying the functor Ξ L/K<br />
n<br />
mutative diagrams<br />
Ξ L/K<br />
n (A ⊗ K L)<br />
Ξ L/K<br />
n ( f )<br />
<br />
P n−1<br />
L<br />
σ<br />
a σ ◦σ<br />
A ⊗ K L<br />
f<br />
M n (L)<br />
σ <br />
Ξ L/K<br />
n (A ⊗ K L)<br />
a σ ◦σ<br />
P n−1<br />
L<br />
Ξ L/K<br />
n ( f )<br />
such that the vertical arrows are isomorphisms. Galois descent on the lower half<br />
of the diagram yields the description of X A given in Section 6. Galois descent<br />
on the upper half of the diagram yields the description claimed. □<br />
7.20. Corollary. –––– Let L/K be a field extension <strong>and</strong> A be a central simple<br />
algebra over K.<br />
i) Then, there is a canonical isomorphism of L-schemes<br />
ξ L/K<br />
A : X A⊗K L −→ X A × SpecK SpecL .<br />
are com-<br />
ii) For field extensions L ′ /L/K, the isomorphisms ξ L′ /K<br />
A , ξ L′ /L<br />
A , <strong>and</strong> ξ L/K<br />
A<br />
patible. I.e., the diagram<br />
X A⊗K L ′ ξ L′ /K<br />
A<br />
<br />
commutes.<br />
ξ L′ /L<br />
X<br />
A<br />
A × SpecK SpecL ′<br />
<br />
X A⊗K L × SpecL SpecL ′<br />
ξ L/K<br />
A × SpecLSpecL ′<br />
Proof. Let n 2 = dim A. We choose a splitting field L ′′ for A containing L ′ .<br />
Then, X A , X A⊗K L, <strong>and</strong> X A⊗K L ′ are constructed from<br />
Ξ L′′ /K<br />
n (A ⊗ K L ′′ ) = Ξ L′′ /L<br />
n (A ⊗ K L ′′ ) = Ξ L′′ /L ′<br />
n (A ⊗ K L ′′ )<br />
by Galois descent using compatible descent data.<br />
□
48 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
7.21. Definition. –––– Let r ∈Æ<strong>and</strong> K be a field.<br />
i) By Az K r , we will denote the category of all central simple algebras of dimension<br />
r 2 over K. As morphisms, we take the homomorphisms of K-algebras<br />
preserving the unit element. Note, in Az K r every morphism is an isomorphism.<br />
Az /K<br />
r will denote the category of all central simple algebras of dimension r 2<br />
over any field extension of K. As morphisms, we take the K-algebra homomorphisms<br />
which preserve the unit element.<br />
ii) By BS K r , we will denote the category of all Brauer-Severi varieties of dimension<br />
r over K. The isomorphisms of K-schemes are taken as morphisms.<br />
BS /K<br />
r<br />
will denote the category of all Brauer-Severi varieties of dimension r over<br />
any field extension of K. As morphisms, we take the compositions of an<br />
isomorphism of K-schemes with a morphism of type<br />
id × Speca : X × SpecL SpecL ′ −→ X.<br />
Here, a : L → L ′ is a homomorphism of fields.<br />
7.22. Theorem. –––– Let K be a field <strong>and</strong> n ∈Æ.<br />
i) There is an equivalence of categories<br />
X /K<br />
n<br />
satisfying the following condition.<br />
: Az/K n −→ (BS /K<br />
n−1 )op<br />
For each field extension L/K, on isomorphism classes, the functor X /K<br />
n<br />
bijection Xn L : Az L n → BS L n−1 from Corollary 6.3.i).<br />
induces an equiva-<br />
ii) In particular, for each field extension L/K, the functor X /K<br />
n<br />
lence of categories X L n : Az L n → (BS L n−1) op .<br />
Proof. First step. Construction of the functor.<br />
induces the<br />
For A ∈ Ob(Az /K<br />
n ), we put X /K<br />
n (A) := X A . We are going to make use of the<br />
intrinsic description of X A given in Proposition 7.19.<br />
If i : A → A ′ is a morphism in Az n<br />
/K then, by restriction to the centers, i induces<br />
a homomorphism i| Z(A) : Z(A) → Z(A ′ ) of fields extending K. Therefore, there<br />
is a unique factorization<br />
A cZ(A′ )<br />
A<br />
A ⊗ Z(A) Z(A ′ )<br />
j<br />
A ′<br />
of i via the canonical inclusion c Z(A′ )<br />
A .<br />
By Corollary 7.20, one has the canonical isomorphism<br />
ξ Z(A′ )/Z(A)<br />
A : X A⊗Z(A) Z(A ′ ) −→ X A × SpecZ(A) SpecZ(A ′ ).
Sec. 7] FUNCTORIALITY 49<br />
We put X /K<br />
n (c Z(A′ )<br />
A ) to be the morphism of schemes induced from ξ Z(A′ )/Z(A)<br />
A by<br />
projection to the first factor.<br />
In order to construct X /K<br />
n (i) as a functor on the category Az/K n , it remains<br />
to describe it on the full subcategories Az K 1<br />
n for all field extensions K 1 /K.<br />
That means, we are left with the case that Z(A) = Z(A ′ ) <strong>and</strong> i| Z(A) is the identity.<br />
In order to ensure functoriality, the construction has to be made compatibly<br />
with field extensions. I.e., such that the diagram<br />
X A ′ ⊗ K1 K 2<br />
X /K<br />
n (i⊗ K1 K 2 ) <br />
X /K<br />
n (c K 2<br />
A ′ ) <br />
X A ′<br />
X /K<br />
n (i)<br />
X A⊗K1 K 2<br />
X A<br />
X /K<br />
n (c K 2<br />
A )<br />
commutes for every field K 1 containing K, every morphism i : A → A ′ in Az K 1<br />
n ,<br />
<strong>and</strong> every extension K 2 /K 1 of fields containing K.<br />
For this, we choose a splitting field L for A. We may assumeL is finite <strong>and</strong> Galois<br />
over K 1 . As i is automatically an isomorphism, L is a splitting field for A ′ , too.<br />
Thus, weobtainaGaloisinvarianthomomorphismi⊗ K1 L : A⊗ K1 L → A ′ ⊗ K1 L.<br />
Applying the functor Ξ L/K 1<br />
n , one gets a Galois invariant morphism of schemes<br />
X A ′ × SpecK1 SpecL = Ξ L/K 1<br />
n (A ′ ⊗ K1 L)<br />
Ξ L/K 1<br />
Ξ L/K 1<br />
n (i⊗ L M )<br />
n (A ⊗ K1 L) = X A × SpecK1 SpecL .<br />
Galois descent for morphisms of schemes yields the morphism<br />
X /K<br />
n (i): X A ′ → X A<br />
desired. X /K<br />
n (i) is an isomorphism of schemes as i is. Consequently, X /K<br />
n is a<br />
functor between the categories stated.<br />
By construction, X /K<br />
n is essentially surjective.<br />
We also note that, for each field extension K ′ /K, X /K<br />
n induces a functor<br />
X K ′<br />
n : AzK ′<br />
n −→ (BS K ′<br />
n−1 )op .<br />
Second step. Full faithfulness on automorphisms.<br />
Let us deal with the following statement which is a special case of full faithfulness<br />
of X /K<br />
n .<br />
(‡) Let K ′ be a field containing K <strong>and</strong> A be a central simple algebra of dimension<br />
n 2 over K ′ . Then, the functor X K ′<br />
n induces an isomorphism of <strong>groups</strong><br />
x A := X K ′<br />
n | AutK ′ (A) : Aut K ′(A) −→ Aut K ′ -schemes(X A ).
50 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
In the case A = M n (K ′ ), this was proven in Proposition 7.17.<br />
Let now A be a general central simple algebra of dimension n 2 over K ′ . Choose a<br />
splitting field L for A which is finite <strong>and</strong> Galois over K ′ .<br />
Injectivity. Assume a, a ′ ∈ Aut K ′(A) induce one <strong>and</strong> the same morphism<br />
a ∗ = a ′∗ ∈ Aut K ′ -schemes(X A ) .<br />
There are unique homomorphisms a L , a ′ L : A⊗ K L → A⊗ K L of central simple<br />
algebras over L making the diagrams<br />
A ⊗ K L<br />
a L A ⊗K L A ⊗ K L<br />
a ′ L<br />
A ⊗ K L<br />
c L A<br />
A<br />
a<br />
A ,<br />
c L A<br />
commute. Applying the functor X K ′<br />
n , we obtain the commutative diagrams below,<br />
X A⊗K L<br />
(c L A )∗ <br />
a ∗ L<br />
a ∗ =a ′∗<br />
X A⊗K L<br />
(c L A )∗<br />
c L A<br />
A<br />
X A⊗K L<br />
(c L A )∗ <br />
a ′ L ∗<br />
a ′<br />
a ∗ =a ′∗<br />
A<br />
X A⊗K L<br />
X A X A , X A X A .<br />
(cA L)∗ is, up to isomorphism, the canonical morphism X A × SpecK SpecL → X A<br />
from the fiber product to its first factor. Therefore, the morphism<br />
a ∗ ◦ (cA L)∗<br />
= a ′∗ ◦ (cA L)∗ admits a unique factorization into a morphism of<br />
L-schemes <strong>and</strong> (cA) L ∗ . This implies a ∗ L = a L ′ ∗ .<br />
SinceA⊗ K L ∼ = M n (L), wemayconcludea L = a L ′ from this. Theequalitya = a′<br />
follows immediately.<br />
By consequence, the homomorphism x A is injective.<br />
Surjectivity. Let p : X A → X A be an automorphism of the K ′ -scheme X A .<br />
(cA L)∗ is, up to isomorphism, the canonical projection X A × SpecK SpecL → X A<br />
from the fiber product to its first factor. Therefore, the composition<br />
p ◦ (cA L)∗ factors uniquely via (cA L)∗ . I.e., there exists a unique morphism<br />
p L : X A⊗K L → X A⊗K L of L-schemes such that there is a commutative diagram<br />
c L A<br />
(c L A )∗<br />
X A⊗K L<br />
p L<br />
X A⊗K L<br />
(c L A )∗ <br />
p<br />
X A X A .<br />
As A⊗ K L ∼ = M n (L), Proposition 7.17 shows there exists some b ∈ Aut(A⊗ K L)<br />
such that p L = b ∗ .<br />
(c L A )∗
Sec. 7] FUNCTORIALITY 51<br />
For σ ∈ Gal(L/K ′ ), let id × σ : A ⊗ K L → A ⊗ K L be the corresponding<br />
automorphism of A ⊗ K L. Clearly, (id × σ) ◦ cA L = cA L . Therefore, the diagram<br />
X A⊗K L<br />
b ∗ =p L<br />
X A⊗K L<br />
(id×σ) ∗ <br />
X A⊗K L<br />
b ∗ =p L<br />
X A⊗K L<br />
(id×σ) ∗<br />
commutes, too. Hence,<br />
(c L A )∗ <br />
p<br />
X A X A .<br />
(c L A )∗<br />
b ∗ = (id × σ) ∗ ◦ b ∗ ◦ (id × σ −1 ) ∗ = ((id × σ −1 ) ◦ b ◦ (id × σ)) ∗ .<br />
The injectivity of (‡) shows b = (id × σ −1 ) ◦ b ◦ (id × σ). I.e., b is invariant<br />
with respect to the operation of Gal(L/K ′ ).<br />
By Galois descent for homomorphisms, there exists a homomorphism<br />
a : A → A of central simple algebras such that the diagram<br />
A ⊗ K L<br />
b<br />
A ⊗ K L<br />
commutes.<br />
c L A<br />
A<br />
a<br />
By consequence, there is a commutative diagram on the level of Brauer-Severi<br />
varieties as follows,<br />
X A⊗K L<br />
(c L A )∗ <br />
b ∗ =p L<br />
a ∗<br />
A<br />
c L A<br />
X A⊗K L<br />
X A X A .<br />
A direct comparison with the definition of p L shows p ◦ (c L A) ∗ = a ∗ ◦ (c L A) ∗ .<br />
Indeed, both compositions are equal to (c L A )∗ ◦ p L . Since (c L A )∗ is dominant, this<br />
implies p = a ∗ .<br />
The homomorphism x A is surjective, too.<br />
Third step. Isomorphisms.<br />
Let K ′ ⊇ K be a field <strong>and</strong> A <strong>and</strong> A ′ be central simple algebras of dimension n 2<br />
over K ′ such that A ∼ = A ′ . Then, X K ′<br />
n induces a bijection<br />
(c L A )∗<br />
Iso K ′ (A, A ′ ) −→ Iso K ′ -schemes(X A ′, X A ) .<br />
This is an immediate consequence of the results from the previous step.
52 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
Fourth step. Faithfulness.<br />
Assume the homomorphisms i 1 , i 2 : A → A ′ induce the same morphism of<br />
schemes i ∗ 1 = i ∗ 2 : X A ′ → X A.<br />
We first observe that i 1 <strong>and</strong> i 2 give rise to the same homomorphism<br />
i 1 | Z(A) = i 2 | Z(A) : Z(A) → Z(A ′ )<br />
between centers. Indeed, X A <strong>and</strong> X A ′ are Brauer-Severi varieties over Z(A)<br />
<strong>and</strong> Z(A ′ ), respectively. Further, by Proposition 5.2.iv), Γ(X A , O XA ) = Z(A)<br />
<strong>and</strong> Γ(X A ′, O XA ′) = Z(A ′ ). I.e., one can recover the centers of A <strong>and</strong> A ′ from<br />
the Brauer-Severi varieties associated with them. By the construction of X /K<br />
n<br />
on morphisms, the pull-back on global sections<br />
(i ∗ 1 )♯ = (i ∗ 2 )♯ : Z(A) = Γ(X A , O XA ) −→ Γ(X A ′, O XA ′<br />
) = Z(A ′ )<br />
is equal to the homomorphism Z(A) → Z(A ′ ) given by i 1 , respectively i 2 , when<br />
restricted to the centers.<br />
Consequently, both i 1 <strong>and</strong> i 2 can be factorized via the canonical homomorphism<br />
c Z(A′ )<br />
A : A → A ⊗ Z(A) Z(A ′ ) .<br />
Let j 1 , j 2 : A⊗ Z(A) Z(A ′ ) → A ′ bethehomomorphismsofcentralsimplealgebras<br />
over Z(A ′ ) such that j 1 ◦ c Z(A′ )<br />
A = i 1 <strong>and</strong> j 2 ◦ c Z(A′ )<br />
A = i 2 . Both j 1 <strong>and</strong> j 2 are<br />
isomorphisms as they are homomorphisms of central simple algebras over the<br />
same base field.<br />
On the other h<strong>and</strong>, the morphisms<br />
j ∗ 1 , j ∗ 2 : X A ′ −→ X A⊗Z(A) Z(A ′ ) = X A × SpecZ(A) SpecZ(A ′ )<br />
coincide. Indeed, their projections to SpecZ(A ′ ) do, both being the structural<br />
morphism. Further, the projections to X A are equal to i ∗ 1 = i ∗ 2 .<br />
This implies j 1 = j 2 .<br />
Fifth step. Fullness.<br />
LetA, A ′ ∈ Ob (Az /K<br />
n )<strong>and</strong> f : X A ′ → X A beamorphism inthecategoryBS /K<br />
n−1 .<br />
The corresponding map of global sections is a homomorphism Z(A) → Z(A ′ )<br />
of fields. From the definition of the category BS /K<br />
n−1<br />
, we see that f gives rise to<br />
an isomorphism of Z(A ′ )-schemes<br />
such that f = (c Z(A′ )<br />
A ) ∗ ◦ f .<br />
f : X A ′ −→ X A × SpecZ(A) SpecZ(A ′ ) = X A⊗Z(A) Z(A ′ )
Sec. 8] THE FUNCTOR OF POINTS 53<br />
By virtue of Corollary 6.3, the central simple algebras A ′ <strong>and</strong> A ⊗ Z(A) Z(A ′ )<br />
over Z(A ′ ) are isomorphic. In the third step, we showed there is an isomorphism<br />
a : A ⊗ Z(A) Z(A ′ ) −→ A ′<br />
such that f = a ∗ . Consequently, f = (c Z(A′ )<br />
A ) ∗ ◦ a ∗ = (a ◦ c Z(A′ )<br />
A ) ∗ is in the<br />
image of X /K<br />
n on morphisms. X /K<br />
n is full. □<br />
7.23. Corollary. –––– Let K be a field <strong>and</strong> n ∈Æ. Then, there is an equivalence<br />
of categories<br />
ˇX K n : AzK n −→ BS K n−1 .<br />
Proof. Compose X K n with the equivalence of categories ι: BSK n−1 → (BS K n−1 )op<br />
given by the identity on objects <strong>and</strong> by ι(g) := g −1 on morphisms. □<br />
8. The functor of <strong>points</strong><br />
8.1. Remark. –––– This section deals with the contravariant functor of <strong>points</strong><br />
on the category of all K-schemes defined by the Brauer-Severi variety X A .<br />
It turns out that this functor may be described completely explicitly in terms of<br />
the central simple algebra A.<br />
Thus, there is a different method to introduce X A . One could start with the<br />
functor <strong>and</strong> has to prove its representability by a scheme. It seems that this<br />
method is closer to A. Grothendieck’s style in Algebraic Geometry than the<br />
approach presented here. Unfortunately, the proof of representability is not<br />
trivial at all. It is presented in detail in [Hen] or [Ke].<br />
8.2. Definition. –––– Let K be a field, n ∈Æ, <strong>and</strong> A ∈ Ob (Az K n ).<br />
Then, by<br />
I A : {K-schemes} op → {sets} ,<br />
we will denote the functor given by<br />
{ }<br />
sheaves of right ideals J in π<br />
∗<br />
T ↦→<br />
T A such that<br />
πT ∗ A /J is a locally free O T -module of rank n 2 − n<br />
on objects <strong>and</strong> by pull-back on morphisms.<br />
Here, A := Ã is the sheaf of O K-algebras associated with A on SpecK.<br />
π = π T : T → SpecK denotes the structural morphism.
54 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
8.3. Remark. –––– The various functors I A depend on A in a natural manner.<br />
I.e., if i : A → A ′ is a morphism in Az K n then there is a morphism of functors<br />
i ∗ : I A ′ → I A given by the inverse image<br />
J ↦→ π ∗ T (ĩ) −1 (J ) .<br />
In other words, there exists a contravariant functor<br />
I : Az K n −→ Fun ({K-schemes} op , {sets})<br />
given on objects by A ↦→ I A .<br />
8.4. Theorem. –––– Let K be a field <strong>and</strong> n ∈Æ. Then, there is an isomorphism<br />
between the contravariant functors<br />
ι: I −→ P ◦ X K n<br />
I, P ◦ X K n : AzK n −→ Fun ({K-schemes} op , {sets}) .<br />
8.5. Remark. –––– For a fixed central simple algebra A of dimension n 2 over K,<br />
Theorem 8.4 asserts that there is an isomorphism ι A : I A −→ P XA between<br />
the functors<br />
I A , P XA : {K-schemes} op → {sets} ,<br />
described in Definitions 8.2 <strong>and</strong> 7.3.<br />
This means, the T -valued <strong>points</strong> on X A are in natural bijection to the<br />
sheaves J ⊂ π ∗ T A of right ideals such that π∗ T A /J is a locally free O T -<br />
module of rank n 2 − n. Further, if i : A → A ′ is a morphism of n 2 -dimensional<br />
central simple algebras over K then the diagram<br />
I A ′<br />
ι A ′<br />
P XA ′<br />
is commutative.<br />
i ∗ [X K n (i)] ∗<br />
ι A<br />
I A PXA<br />
8.6. Remark. –––– I A is canonically a subfunctor of the Graßmann functor<br />
Grass n2<br />
n 2 −n parametrizing (n 2 − n)-dimensional quotients of the n 2 -dimensional<br />
st<strong>and</strong>ard vector space. Thus, one has an embedding X A ֒→ Grass n2<br />
n 2 −n into<br />
the Graßmann scheme. This is the key observation for a direct proof of the<br />
representability of I A [Hen, Ke].
Sec. 8] THE FUNCTOR OF POINTS 55<br />
8.7. Definition. –––– Let G be a group.<br />
By Sets G , we will denote the category of all mappings M → G where M is an<br />
arbitrary set. As morphisms from f 1 : M 1 → G to f 2 : M 2 → G, we allow all<br />
mappings making the diagram<br />
M 1<br />
M 2<br />
f 2<br />
f 1<br />
<br />
G<br />
· g<br />
commutative for a certain g ∈ G. Here, · g : G → G denotes multiplication<br />
by g from the right.<br />
8.8. Notation. –––– Let L/K be a finite Galois extension. Then, Sch L/K<br />
+ will<br />
denote the full subcategory of Sch L/K (cf. Definition 7.10) consisting of all<br />
non-empty L-schemes.<br />
8.9. Fact-Definition. –––– Let L/K be a finite Galois extension <strong>and</strong> G be its<br />
Galois group.<br />
Then, for each X ∈ Ob (Sch L/K ), the contravariant functor<br />
G<br />
Hom Sch<br />
L/K ( · , X ): Sch L/K<br />
+ −→ {sets}<br />
factors canonically via Sets G . The resulting functor<br />
P L/K<br />
X<br />
will be called the functor of <strong>points</strong> of X.<br />
Proof. Each morphism T → X in Sch L/K<br />
+<br />
: Sch L/K<br />
+ −→ Sets G<br />
is twisted by a unique element σ ∈ G.<br />
□<br />
8.10. Remark. –––– For that, we need the assumption T ≠ ∅. On the other<br />
h<strong>and</strong>, X = ∅ could have been allowed since there are no morphisms at all in<br />
this case.<br />
8.11. Remarks. –––– a) P L/K<br />
X<br />
is closely related to the ordinary functor of<br />
<strong>points</strong> P X introduced in Definition 7.3. For each non-empty L-scheme T ,<br />
one has P X (T ) = [P L/K<br />
X<br />
(T )] −1 (id).<br />
b) There is a covariant functor<br />
given on objects by X ↦→ P L/K<br />
X<br />
.<br />
P L/K : Sch L/K → Fun ((Sch L/K<br />
+ )op , Sets G )
56 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
c) The functor Hom L/K Sch<br />
( · , X ) remembers all information on X as an object<br />
+<br />
of Sch L/K<br />
+ .<br />
The functor of <strong>points</strong><br />
P L/K<br />
X<br />
: Sch L/K<br />
+ −→ Sets G<br />
carries additional information related to the structure of Sch L/K<br />
+ as a fibered category.<br />
8.12. Lemma. –––– Let n ∈Æ, L/K be a finite Galois extension of fields <strong>and</strong><br />
G := Gal(L/K ) its Galois group.<br />
of functors between the com-<br />
i) Then, there is an isomorphism jn<br />
L/K<br />
position<br />
P L/K<br />
n<br />
: Mat L/K<br />
n<br />
Ξ L/K<br />
n<br />
P L/K<br />
n−1<br />
embedding<br />
: P L/K<br />
n<br />
−→ I L/K<br />
n<br />
Sch L/K<br />
+<br />
P L/K Fun ((Sch L/K<br />
+ )op , Sets G )<br />
<strong>and</strong> the functor In<br />
L/K given by<br />
⎛ ⎧<br />
⎫⎞<br />
A ↦→ ⎝T ↦→ G ⎨ sheaves of right ideals J in (S(σ) ◦ π T ) ∗ A ⎬<br />
such that (S(σ) ◦ π<br />
⎩<br />
T ) ∗ A /J<br />
⎠<br />
σ∈G is a locally free O T -module of rank (n 2 ⎭<br />
− n)<br />
on objects <strong>and</strong> by pull-back on morphisms.<br />
Here, A := Ã denotes the sheaf of O L-algebras associated with A on SpecL.<br />
π T : T → SpecL is the structural morphism. The set on the right h<strong>and</strong> side is<br />
equipped with the canonical map to G.<br />
ii) Let L ′ /L be a finite field extension such that L ′ /K is Galois. Then, the isomorphism<br />
j L/K<br />
n is compatible with base extension from L to L ′ .<br />
I.e., for every A ∈ Ob (Mat L/K<br />
n<br />
P L/K<br />
n (A)(T )<br />
) <strong>and</strong> every T ∈ Ob (Sch L/K<br />
+ ), the diagram<br />
Mor Sch<br />
L/K (T , Ξn<br />
L/K (A))<br />
j L/K<br />
n (A)(T )<br />
I L/K<br />
n (A)(T )<br />
. × SpecL SpecL ′ <br />
Mor Sch<br />
L ′ /K (T× SpecL SpecL ′ , Ξ L′ /K<br />
n (A⊗ L L ′ ))<br />
pull-back<br />
j I L′ /K<br />
n (A⊗ L L ′ )(T× SpecL SpecL ′ )<br />
P L′ /L<br />
n (A ⊗ L L ′ )(T × SpecL SpecL ′ )<br />
commutes. j is an abbreviation for j L′ /K<br />
n (A ⊗ L L ′ )(T × SpecL SpecL ′ ).
Sec. 8] THE FUNCTOR OF POINTS 57<br />
Proof. i) For each A ∈ Ob (Mat L/K<br />
n ), both functors satisfy the sheaf axiom for<br />
Zariski coverings. Thus, it will suffice to verify the assertion on affine schemes.<br />
Let R be a commutative L-algebra with unit. By construction, we have<br />
Ξ L/K<br />
n (A) = P(l) for l a non-zero, simple, left A-module. By Corollary-<br />
Definition 7.6, the set of all R-valued <strong>points</strong> on P(l) is in natural bijection<br />
to the set of all submodules M ⊂ l⊗ L R such that the quotient l⊗ L R/M is a<br />
locally free R-module of rank (n − 1).<br />
As an R-module, A⊗ L R is isomorphic to the direct sum of n copies of l⊗ L R.<br />
Under this isomorphism, an R-submodule M ⊆ l ⊗ R determines a right<br />
ideal M ⊆ A ⊗ L R according to the definition M := L n<br />
i=1 M. This gives a<br />
natural bijection between the set of all right ideals in A ⊗ L R <strong>and</strong> the set of<br />
all R-submodules M ⊆ l. Obviously,<br />
(A ⊗ L R)/M ∼ = ((l ⊗ L R)/M ) n ∼ = ((l ⊗L R)/M ) ⊗ R R n<br />
as R-modules. Thus, if (l⊗ L R)/M is locally free of rank (n−1) then (A⊗ L R)/M<br />
is locally free of rank (n 2 − n). Conversely, if (A ⊗ L R)/M is locally free of<br />
rank (n 2 − n) then, by Lemma 3.14, (l⊗ L R)/M is locally free. Clearly, it is of<br />
rank (n − 1).<br />
To give a morphism of L-schemes f : SpecR → Ξn<br />
L/K (A) which is twisted by<br />
some σ ∈ G is equivalent to giving a commutative diagram<br />
SpecR<br />
π SpecR<br />
<br />
SpecL<br />
id<br />
S(σ)<br />
SpecR<br />
S(σ)◦π SpecR<br />
<br />
SpecL<br />
f<br />
<br />
Ξ L/K<br />
n (A)<br />
id SpecL .<br />
π L/K Ξ n (A)<br />
Therefore, f becomes a morphism of L-schemes in the ordinary sense if one<br />
just changes the structural morphism of SpecR into S(σ) ◦ π.<br />
Consequently, the functor of <strong>points</strong> P L/K is isomorphic to the functor given<br />
Ξ L/K<br />
R (A)<br />
in the assertion.<br />
ii) Again, we need the concrete description of Ξ n<br />
./K . We have Ξ L/K<br />
n (A) = P(l)<br />
for l a non-zero, simple, left A-module. Further, Ξ L′ /K<br />
n (A⊗ L L ′ ) = P(l⊗ L L ′ ).<br />
The claim now easily follows from Remark 7.7.<br />
□<br />
8.13. Remark. –––– Lemma 8.12 states, in particular, that for A ∈ Ob (Mat L/K<br />
the ordinary functor of <strong>points</strong> of the L-scheme Ξn<br />
L/K (A) is isomorphic to<br />
{ }<br />
sheaves of right ideals J in π<br />
∗<br />
T ↦→<br />
T A such that<br />
πT ∗ A /J is a locally free O T -module of rank (n 2 .<br />
− n)<br />
n )<br />
I.e., for every L-algebra R, the set of R-valued <strong>points</strong> on Ξ L/K<br />
n (A) is naturally<br />
isomorphic to the set of all right ideals I in A⊗ L R such that A⊗ L R/I is a locally<br />
free R-module of rank (n 2 − n).
58 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
8.14. Proof of Theorem 8.4. –––– For every A ∈ Ob (Az K n ), we have to construct<br />
an isomorphism ι A : I A → P XA of functors. ι A has to be natural in A.<br />
We will proceed in three steps.<br />
First step. The case A ∼ = M n (K ).<br />
We have A⊗ K R ∼ = M n (R). For X A , there is the intrinsic description given in<br />
Proposition 7.19. X A = P(l) where l denotes a non-zero, simple, left A-module.<br />
The assertion is true by Lemma 8.12. We note explicitly that P(l) = Ξ K/K ′<br />
n (A)<br />
for every subfield K ′ ⊂ K such that K/K ′ is finite Galois. In particular, the<br />
isomorphism ι A : I A → P XA is compatible with automorphisms of A which are<br />
only K ′ -linear.<br />
Second step. Reduction to T -valued <strong>points</strong> for affine T .<br />
We return to the general case that A is an arbitrary central simple algebra<br />
over K of dimension n 2 . The functors I A <strong>and</strong> P XA satisfy the sheaf axioms for<br />
Zariski coverings. Hence, it suffices to consider the affine case.<br />
There is to be constructed a natural isomorphism ι A : I A → P XA of functors on<br />
the full subcategory of affine schemes. I.e., for each commutative K-algebra R<br />
with unit, an isomorphism<br />
{ }<br />
right ideals J in A ⊗K R such that A ⊗<br />
I A (R) :=<br />
K R/J<br />
is a locally free R-module of rank n 2 − n<br />
↓ι A (R)<br />
P XA (R) := Mor K-schemes (SpecR, X A ) .<br />
For a homomorphism r : R → R ′ of K-algebras, the corresponding diagram<br />
I A (R)<br />
ι A (R)<br />
<br />
P XA (R)<br />
is supposed to commute.<br />
Third step. Galois descent.<br />
Let L be a splitting field for A.<br />
Put G := Gal(L/K ).<br />
I A (r)<br />
P XA (r)<br />
I A (R ′ )<br />
ι A (R ′ )<br />
P XA (R ′ )<br />
Assume that L/K is finite <strong>and</strong> Galois.<br />
By Theorem 3.2.ii), there is a bijection<br />
{ }<br />
p : SpecR → XA |<br />
P XA (R) =<br />
p morphism of K-schemes<br />
⎧<br />
⎫<br />
⎨ p : SpecR ⊗ K L → X A⊗K L |<br />
⎬<br />
∼= p morphism of L-schemes,<br />
⎩<br />
⎭ , (§)<br />
p compatible with the G-operations on both sides<br />
natural in the K-algebra R.
Sec. 8] THE FUNCTOR OF POINTS 59<br />
As A⊗ K L is isomorphic to the full matrix algebra, we have<br />
X A⊗K L = Ξn<br />
L/K (A ⊗ K L) = P(l)<br />
where l is a non-zero, simple, right A⊗ K L-module. Hence, there is a second<br />
natural bijection<br />
⎧<br />
⎫<br />
⎨ p : SpecR ⊗ K L → X A⊗K L |<br />
⎬<br />
p morphism of L-schemes,<br />
⎩<br />
⎭<br />
p compatible with the G-operations on both sides<br />
⎧<br />
⎫<br />
J ⊂ (A⊗ K L) ⊗ L (R⊗ K L) |<br />
⎪⎨ J right ideal,<br />
⎪⎬<br />
∼= ((A⊗ K L) ⊗ L (R⊗ K L))/J locally free (R⊗ K L)-module, . ()<br />
rk R⊗K L ((A⊗ K L) ⊗ L (R⊗ K L))/J = (n<br />
⎪⎩<br />
2 − n) ,<br />
J invariant with respect to the G-operation<br />
⎪⎭<br />
Indeed, this is exactly the result of the first step when we note that the isomorphism<br />
of functors I A⊗K L → P XA⊗K is compatible with K-linear automorphisms<br />
L<br />
of A⊗ K L.<br />
Finally, there is a natural bijection<br />
⎧<br />
⎫<br />
J ⊂ (A⊗ K L) ⊗ L (R⊗ K L) |<br />
⎪⎨ J right ideal,<br />
⎪⎬<br />
((A⊗ K L) ⊗ L (R⊗ K L))/J locally free (R⊗ K L)-module,<br />
rk R⊗K L ((A⊗ K L) ⊗ L (R⊗ K L))/J = n<br />
⎪⎩<br />
2 − n ,<br />
J invariant with respect to the G-operation<br />
⎪⎭<br />
⎧<br />
⎫<br />
J ⊂ A⊗ K R |<br />
⎪⎨<br />
⎪⎬<br />
J right ideal,<br />
∼=<br />
= I<br />
(A⊗ ⎪⎩ K R)/J locally free R-module, A (R) (‖)<br />
⎪⎭<br />
rk R (A⊗ K R)/J = (n 2 − n)<br />
by Lemma 8.16. I.e., by Galois descent for right ideals.<br />
Indeed, everything is clear except for the statements on the ranks of the quotients.<br />
For that, if (A⊗ K R)/J is a locally free R-module of rank (n 2 − n) then<br />
((A⊗ K L) ⊗ L (R⊗ K L))/J ∼ = ((A⊗ K R)/J ) ⊗ R (R⊗ K L)<br />
is a locally free (R⊗ K L)-module of the same rank. On the other h<strong>and</strong>, if<br />
((A⊗ K R)/J ) ⊗ R (R⊗ K L)<br />
is a locally free (R ⊗ K L)-module of rank (n 2 − n) then it is a locally free<br />
R-module, too. By Lemma 3.14, (A ⊗ K R)/J is locally free as an R-module.<br />
Clearly, it is of rank (n 2 − n).
60 THE BRAUER-SEVERI VARIETY OF A CENTRAL SIMPLE ALGEBRA [Chap. I<br />
We finally note that the construction of ι A given is functorial in A. For that,<br />
one first has to choose a splitting field L A ⊃ K for each A ∈ Ob (Az K n ) in such a<br />
way that L A depends only on the isomorphism class of A in Az K n . If i : A → A ′<br />
is a morphism in Az K n then A <strong>and</strong> A′ are automatically isomorphic. We execute<br />
the constructions of ι A <strong>and</strong> ι A ′ via the splitting field chosen.<br />
The natural bijections (§) <strong>and</strong> (‖) are applications of descent <strong>and</strong>, therefore,<br />
compatible with the morphisms induced by i. For (), apply Lemma 8.12.i).<br />
The proof is complete.<br />
8.15. Remark. –––– ι A is independent of the choice of the splitting field made<br />
in the proof. It is not difficult to deduce this from Lemma 8.12.ii).<br />
8.16. Lemma (Galois descent for right ideals). —–<br />
Let L/K be a finite Galois extension. Denote by G := Gal(L/K ) its Galois group.<br />
Further, let A be a K-algebra <strong>and</strong> I ⊂ A ⊗ K L a right ideal invariant under the<br />
canonical operation of G on A⊗ K L.<br />
Then, there is a unique right ideal I ⊂ A such that I = I ⊗ K L.<br />
Proof. I inherits from A⊗ K L a structure of an L-algebra. Further, I is naturally<br />
equipped with an operation of G by homomorphisms of K-algebras. σ ∈ G acts<br />
by a σ-linear map.<br />
By the algebraic version of Galois descent, there exists a K-algebra I such<br />
that I = I ⊗ K L.<br />
Furthermore, the canonical homomorphism<br />
I = I ⊗ K L ֒→ A⊗ K L<br />
of L-algebras is compatible with the G-operations on both sides. Therefore,<br />
by Galois descent for homomorphisms, we get a homomorphism I → A<br />
of K-algebras. As the functor ⊗ K L is exact <strong>and</strong> faithful, that homomorphism<br />
is necessarily injective.<br />
Consider I ⊂ A as a subring. For every a ∈ A, the multiplication ·a : I → I<br />
from the right is compatible with the operation of G. Hence, it descends to a homomorphism<br />
I → I. That homomorphism is compatible with multiplication<br />
by a from the right on the whole of A. Consequently, I ⊂ A is a right ideal.<br />
Uniqueness is clear.<br />
□<br />
□
CHAPTER II<br />
ON THE BRAUER GROUP OF A SCHEME<br />
1. Azumaya algebras<br />
1.1. Definition. –––– Let X be any scheme.<br />
Then, a sheaf of Azumaya algebras or simply an Azumaya algebra over X is a<br />
locally free sheaf A of locally finite rank over X equipped with the structure of<br />
a sheaf of O X -algebras such that the following condition is fulfilled.<br />
For every closed point x ∈ X, one has that A (x) := A ⊗ X k(x) is a central<br />
simple algebra over the residue field k(x).<br />
1.2. Example. –––– If X = Speck is the spectrum of a field then an Azumaya<br />
algebra over X is nothing but a central simple algebra over k.<br />
1.3. Proposition. –––– Let X be any scheme <strong>and</strong> A be any locally free sheaf of<br />
locally finite rank on X having the structure of a sheaf of algebras.<br />
Then, A is an Azumaya algebra over X if <strong>and</strong> only if the canonical homomorphism<br />
ι A : A ⊗ X A op −→ End(A ) ,<br />
a ⊗ b<br />
is an isomorphism of locally free sheaves.<br />
↦→ (x ↦→ axb)<br />
Here, a, b, x ∈ A (U ) denote sections of A over an arbitrary open subset U of X.<br />
Proof. “=⇒” Let A be an Azumaya algebra. Then, A (x) is a central simple<br />
algebra over k(x) for every closed point x ∈ X. By virtue of Remark I.4.2.b),<br />
this implies that ι A is an isomorphism at every closed point. It is therefore an<br />
isomorphism of locally free sheaves.<br />
“⇐=” Let x ∈ X be a closed point. We need to show that A (x) is a central<br />
simple algebra over k(x).<br />
By assumption, we know that<br />
A (x) ⊗ k(x) [A (x)] op ∼ = End(k(x) n ) ∼ = M n (k(x))
62 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
for n := rk x (A ). Further, M n (k(x)) is a central simple algebra over k(x).<br />
If A (x) were not central, Z(A (x)) = L k(x), then<br />
L = L ⊗ k(x) k(x) ⊆ A (x) ⊗ k(x) [A (x)] op ∼ = Mn (k(x))<br />
would be contained in the center of M n (k(x)). This is a contradiction.<br />
Assume, finally, A (x) were not simple. Then, it would contain a proper twosided<br />
ideal (0) I A (x). But, under this assumption, I ⊗ k(x) [A (x)] op would<br />
be a non-trivial, proper two-sided ideal in A (x) ⊗ k(x) [A (x)] op ∼ = M n (k(x)).<br />
We obtained a contradiction. A is, therefore, a central simple algebra.<br />
1.4. Definition. –––– Let X be any scheme <strong>and</strong> A be a locally free sheaf on X<br />
having the structure of a sheaf of algebras.<br />
a) Then, for a point x ∈ X, we will say that A is Azumaya in x if A (x) is a<br />
central simple algebra over k(x).<br />
b) The set of all <strong>points</strong> x ∈ X such that A is not Azumaya in x will be called<br />
the non-Azumaya locus of A .<br />
1.5. Corollary. –––– Let X be any scheme <strong>and</strong> A be any locally free sheaf on X<br />
having the structure of a sheaf of algebras.<br />
a) Then, the non-Azumaya locus T ⊂ X of A is the support of a Cartier divisor.<br />
In particular, it is a closed set in X.<br />
b) Assume, X is regular <strong>and</strong> connected. Then, the non-Azumaya locus of A is either<br />
the whole of X or a subset pure of codimension one.<br />
Proof. a) The assertion is local in X. We may, therefore, assume that A is a free<br />
sheaf, say, of rank n.<br />
ι A : A ⊗ X A op ∼ = O<br />
n 2<br />
X −→ End(A ) ∼ = OX<br />
n2<br />
is then a morphism of free sheaves which are both of rank n 2 .<br />
Whether ι A ⊗ X k(x) is an isomorphism is equivalent to the non-vanishing of its<br />
determinant in x. In the trivialization, the latter is a section s of the structure<br />
sheaf O X . div s is, by definition, the support of a Cartier divisor.<br />
b) This is an immediate consequence of a). □<br />
1.6. Fact. –––– Let A <strong>and</strong> B be two Azumaya algebras on a scheme X.<br />
Then, their tensor product A ⊗ OX B as locally free sheaves over X, equipped with<br />
the obvious structure of an O X -algebra, is again an Azumaya algebra.<br />
Proof. On sections over an open subset U ⊆ X, the algebra structure is given by<br />
(a ⊗ b)(a ′ ⊗ b ′ ) := aa ′ ⊗ bb ′ .<br />
□
Sec. 1] AZUMAYA ALGEBRAS 63<br />
It is clear that A ⊗ OX B is a locally free O X -module.<br />
The property of being Azumaya may be tested in closed <strong>points</strong>. One has<br />
(A ⊗ OX B)(x) = A ⊗ OX B⊗ OX k(x) = (A ⊗ OX k(x)) ⊗ k(x) (B⊗ OX k(x)) .<br />
On the right h<strong>and</strong> side, the tensor product of two central simple algebras is<br />
again a central simple algebra.<br />
□<br />
1.7. Fact. –––– Let f : X → Y be a morphism of schemes <strong>and</strong> A be an Azumaya<br />
algebra on Y.<br />
Then, the pull-back f ∗ A of A as a locally free sheaf, equipped with the obvious<br />
structure of an O X -algebra, is an Azumaya algebra on X.<br />
Proof. As the assertion is local in both, Y <strong>and</strong> X, we may assume that<br />
f : X = SpecS → Y = SpecR is a morphism of affine schemes <strong>and</strong> that<br />
A is given by an R-algebra A which is free of finite rank as an R-module.<br />
Then, f ∗ A is given by the S-algebra A ⊗ R S. It algebra structure is given,<br />
analogously to the one on the tensor product above, by<br />
(a ⊗ s)(a ′ ⊗ s ′ ) := aa ′ ⊗ ss ′ .<br />
Again, we test the property of being Azumaya in closed <strong>points</strong>.<br />
For that, let x ∈ X <strong>and</strong> put y := f (x). Thus, k(x) is a field extension of k(y).<br />
What we have to show is that A⊗ R S ⊗ S k(x) is a central simple algebra over k(x).<br />
But,<br />
A⊗ R S ⊗ S k(x) = A⊗ R k(x) = A⊗ R k(y)⊗ k(y) k(x)<br />
<strong>and</strong> A⊗ R k(y) is, by assumption, a central simple algebra over k(y).<br />
The assertion follows from Lemma I.4.1.b).<br />
□<br />
1.8. Proposition. –––– Let f : X → Y be a morphism of schemes which is faithfully<br />
flat. Further, let A be a quasi-coherent sheaf on Y, equipped with the structure<br />
of a sheaf of O Y -algebras.<br />
Then, A is an Azumaya algebra on Y if <strong>and</strong> only if f ∗ A is an Azumaya algebra<br />
on X.<br />
Proof. “=⇒” This is Fact 1.7.<br />
“⇐=” The assumption that f ∗ A is an Azumaya algebra on X includes that<br />
f ∗ A is a locally free O X -module, locally of finite rank. We aim first at showing<br />
that this implies A is a locally free O Y -module, locally of finite rank.<br />
For that, let y ∈ Y <strong>and</strong> choose x ∈ X such that f (x) = y. We have a flat local<br />
homomorphism of local rings i : O Y,y → O X,x . By [Mat, Theorem 7.2], O X,x is<br />
faithfully flat over O Y,y .
64 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
Further, ( f ∗ A ) x = A y ⊗ OY,y O X,x is a locally free O X,x -module of finite rank.<br />
Lemma I.3.14 implies that A y is indeed a locally free O Y,y -module <strong>and</strong> of finite<br />
rank. This means, A is a locally free O Y -module, locally of finite rank.<br />
In the rest of the proof, flatness will no more be used.<br />
To verify A is an Azumaya algebra over Y, we will use the criterion provided<br />
by Proposition 1.3. We have f ∗ A = A ⊗ OY O X <strong>and</strong>, therefore,<br />
f ∗ A ⊗ OX [ f ∗ A ] op = (A ⊗ OY O X ) ⊗ OX (O X ⊗ OY [A ] op )<br />
∼= A ⊗ OY [A ] op ⊗ OY O X<br />
= f ∗ (A ⊗ OY [A ] op ) .<br />
Further,<br />
End( f ∗ A ) = f ∗ (End(A ))<br />
<strong>and</strong> the isomorphisms are compatible in such a way that<br />
f ∗ ι A = ι f ∗ A .<br />
Thus, we are given that f ∗ ι A = ι f ∗ A is an isomorphism of locally free sheaves.<br />
We have to prove this implies that the original ι A was an isomorphism.<br />
For that, note that det( f ∗ ι A ) = f # det(ι A ). Thus, we know that f # det(ι A ) is<br />
a nowhere vanishing section of the invertible sheaf<br />
f ∗ [Λ max (A ⊗ OY [A ] op ) ∨ ⊗ Λ max (End(A ))] .<br />
We have to show that det(ι A ) is nowhere vanishing.<br />
Again, for y ∈ Y, we choose x ∈ X such that f (x) = y. Then, we have a local<br />
homomorphism of local rings i : O Y,y → O X,x . As we work in stalks, we may<br />
choose a trivialization <strong>and</strong> write det(ι A ) ∈ O Y,y . Then,<br />
f # det(ι A ) = i(det(ι A )) ∈ O X,x .<br />
If det(ι A ) were vanishing at y then this would mean simply det(ι A ) ∈ m Y,y .<br />
Then, i(det(ι A )) ∈ m X,x . I.e., f # det(ι A ) vanishes at x.<br />
This is a contradiction.<br />
1.9. Proposition. –––– Let f : X → Y be a morphism of schemes which is faithfully<br />
flat <strong>and</strong> quasi-compact.<br />
Then, it is equivalent to give<br />
a) an Azumaya algebra A over Y,<br />
b) an Azumaya algebra B over X together with an isomorphism<br />
Φ: pr ∗ 1 B → pr∗ 2 B<br />
□
Sec. 2] THE BRAUER GROUP 65<br />
of Azumaya algebras on X × Y X which satisfies<br />
on X × Y X × Y X.<br />
pr ∗ 31 (Φ) = pr∗ 32 (Φ) ◦ pr∗ 21 (Φ)<br />
Sketch of Proof. “a) =⇒ b)” Put B := f ∗ A .<br />
“b) =⇒ a)” This is what is called faithful flat descent.<br />
The existence of A as a quasi-coherent O Y -module satisfying B := f ∗ A follows<br />
directly from [K/O74, Theorem II.3.2]. The additional algebra structure descends,<br />
too. This may easily be seen from the construction of the descent module<br />
given in the proof. It is shown in [K/O74, Theorem II.3.4].<br />
A is an Azumaya algebra over Y by virtue of Proposition 1.8.<br />
1.10. Remark. –––– This means that an Azumaya algebra over a scheme X<br />
may be described by local gluing data with respect to the fpqc-topology or any<br />
weaker topology.<br />
Intuitively, B describes the desired Azumaya algebra on each set of a cover.<br />
Φ collects the gluing isomorphisms on intersections of two sets of the cover.<br />
The condition on X × Y X × Y X is a cocycle condition encoding that the gluing<br />
maps be compatible.<br />
We will use this approach for constructing Azumaya algebras from data local in<br />
the étale topology.<br />
□<br />
2. The Brauer group<br />
2.1. Definition. –––– Let X be any scheme.<br />
Two Azumaya algebras A <strong>and</strong> B on X are said to be similar if there exist two<br />
locally free O X -modules E <strong>and</strong> F, both everywhere of positive rank, such that<br />
A ⊗ OX End(E ) ∼ = B ⊗ OX End(E ′ ) .<br />
2.2. Remarks. –––– i) Similarity is an equivalence relation.<br />
For that, the only point which is not entirely obvious is transitivity. To achieve<br />
this, assume<br />
A ⊗ OX End(E ) ∼ = B ⊗ OX End(E ′ ) <strong>and</strong> B ⊗ OX End(E ′′ ) ∼ = C ⊗ OX End(E ′′′ ) .
66 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
Then,<br />
A ⊗ OX End(E ⊗ OX E ′′ ) ∼ = A ⊗ OX End(E ) ⊗ OX End(E ′′ )<br />
∼= B ⊗ OX End(E ′ ) ⊗ OX End(E ′′ )<br />
∼= B ⊗ OX End(E ′′ ) ⊗ OX End(E ′ )<br />
∼= C ⊗ OX End(E ′′′ ) ⊗ OX End(E ′ )<br />
∼= C ⊗ OX End(E ′′′ ⊗ OX E ′′′ ) .<br />
ii) Similarity is compatible with the tensor product of Azumaya algebras <strong>and</strong><br />
with pull-back.<br />
2.3. Definition. –––– Let X be any scheme.<br />
The set of all similarity classes of Azumaya algebras on X is called the Brauer<br />
group of X <strong>and</strong> denoted by Br(X ).<br />
2.4. Remarks. –––– i) The similarity class of an Azumaya algebra A will be<br />
denoted by [A ].<br />
ii) A binary operation on Br(X ) is given by [A ] ∗ [A ′ ] := [A ⊗ OX A ′ ] .<br />
iii) For this binary operation, [End(E )] is the neutral element. Here, E is an<br />
arbitrary locally free sheaf, nowhere of rank zero.<br />
iv) The inverse element of [A ] is given by [A op ]. This shows that Br(X ) is<br />
indeed a group.<br />
3. The cohomological Brauer group<br />
3.1. Lemma. –––– Let (R, m) be a Henselian local ring <strong>and</strong> A be an Azumaya<br />
algebra over SpecR. Assume<br />
A ⊗ OSpecR O SpecR/m<br />
∼ = End(O<br />
n<br />
SpecR/m )<br />
for a certain n ∈Æ. Then, A ∼ = End(O n SpecR ).<br />
Proof. We denote by A the R-algebra corresponding to A . In this notation,<br />
we, therefore, have A ⊗ R R/m ∼ = M n (R/m). On the right h<strong>and</strong> side, choose an<br />
idempotent matrix ǫ of rank one.<br />
Let a ∈ A be such that a := (a mod mA) = ǫ. Then, R[a] is a finite<br />
commutative R-algebra. As R is Henselian, [SGA4, Exp. VIII, 4.1] shows that<br />
R[a] is a direct product of local rings. From R[a]⊗ R k = k[ǫ] ∼ = k ⊕ k, we see<br />
that R[a] ∼ = R/I ⊕ R/J is actually composed of two local R-algebras. In k ⊕ k,<br />
let ǫ correspond to the element (1, 0). Thus, the element e ∈ R[a] mapped<br />
to (1, 0) under the isomorphism R[a] → R/I ⊕ R/J is an idempotent lifting ǫ.
Sec. 3] THE COHOMOLOGICAL BRAUER GROUP 67<br />
Consider the homomorphism of R-algebras<br />
φ: A −→ End R (Ae) ,<br />
a ↦→ (be ↦→ abe) .<br />
Both R-modules are free of the same rank. It is classically known that φ⊗ R k<br />
is an isomorphism. Nakayama’s lemma ensures that φ is an isomorphism itself.<br />
□<br />
3.2. Corollary. –––– Let (R, m) be a strictly Henselian local ring <strong>and</strong> A be an<br />
Azumaya algebra over SpecR.<br />
Then, A ∼ = End(O n SpecR ).<br />
Proof. k = R/m is a separably closed field. By virtue of Remark I.4.2.b), we<br />
know that<br />
A ⊗ OSpecR O SpecR/m<br />
∼ = End(O<br />
n<br />
SpecR/m )<br />
is a full matrix algebra over R/m.<br />
□<br />
3.3. Lemma (Theorem of Skolem-Noether, version over local rings). —–<br />
Let R be a commutative ring with unit. Then, GL n (R) operates on M n (R) by<br />
conjugation,<br />
(g, m) ↦→ gmg −1 .<br />
If R = (R, m) is a local ring then this defines an isomorphism<br />
PGL n (R) := GL n (R)/R ∗ ∼ =<br />
−→ Aut R (M n (R)) .<br />
Proof. (Compare Lemma I.4.4 where the special case R is a field was treated.)<br />
One has R = Zent(M n (R)). Therefore, the mapping is well-defined <strong>and</strong> injective.<br />
Surjectivity. Let j : M n (R) → M n (R) be an automorphism. We consider the<br />
algebra<br />
M := M n (R) ⊗ R M n (R) op ( ∼ = End R-mod (M n (R)) ∼ = M n 2(R)) .<br />
M n (R) gets equipped with the structure of a left M-module in two ways.<br />
(A ⊗ B) • 1 C := A · C · B<br />
(A ⊗ B) • 2 C := j(A) · C · B<br />
We will denote the resulting M-modules by N 1 <strong>and</strong> N 2 , respectively.
68 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
Our first claim is that N 1 is a projective M-module. This means, we have to<br />
show that M n (R) is projective when equipped with its canonical structure as<br />
an End R-mod (M n (R))-module.<br />
For that, consider a morphism i : M n (R) → R of R-modules which is a section<br />
of the canonical inclusion mapping r to r·E. Note such a section exists because<br />
M n (R) is a projective R-module.<br />
Then, the surjection of End R-mod (M n (R))-modules End R-mod (M n (R)) → M n (R),<br />
given by f ↦→ f (1), admits the section<br />
m ↦→ (m ′ ↦→ i(m ′ )m) .<br />
This shows M n (R) is a projective End R-mod (M n (R))-module.<br />
Over the field R/m, it is known that two M n 2(R/m)-modules of the same<br />
R/m-dimension are isomorphic. Thus, there is an isomorphism<br />
As M 1 is projective, the map<br />
M 1<br />
g : M 1 ⊗ R R/m → M 2 ⊗ R R/m .<br />
π<br />
−→ M 1 ⊗ R R/m −→ g<br />
M 2 ⊗ R R/m<br />
may be lifted to a homomorphism h : M 1 → M 2 of End R-mod (M n (R))-modules.<br />
h⊗ R R/m is an isomorphism. Nakayama’s lemma implies that h is an isomorphism<br />
itself.<br />
Let us put I := h(E) to be the image of the identity matrix.<br />
M ∈ M n (R) we have<br />
h(M ) = h((E ⊗ M ) • 1 E) = (E ⊗ M ) • 2 h(E) = h(E) · M = I · M.<br />
In particular, I ∈ GL n (R). Therefore,<br />
I · M = h(M ) = h((M ⊗ E) • 1 E) = (M ⊗ E) • 2 h(E) = j(M ) · I<br />
For every<br />
for each M ∈ M n (R) <strong>and</strong> j(M ) = IMI −1 .<br />
□<br />
3.4. Proposition –––– Let X be any quasi-compact scheme. Then, the set of all isomorphismclassesofrankn<br />
2 AzumayaalgebrasonX isgivenbytheČechcohomology<br />
set Ȟ 1 ét (X, PGL n).<br />
Proof. We know by Proposition 1.9 that Azumaya algebras may be described<br />
by gluing data local in the étale topology.<br />
Let A be any Azumaya algebra over X. For every point x ∈ X, Corollary<br />
3.2 yields that A is isomorphic to matrix algebra after strict Henselization,<br />
A ⊗ OX OX,x sh ∼ = M n (OX,x sh ). To describe the isomorphism, only finitely many
Sec. 3] THE COHOMOLOGICAL BRAUER GROUP 69<br />
data on O sh<br />
X,x are necessary. For this reason, there exists an affine scheme U x,<br />
which is an étale neighbourhood of x, such that A | Ux<br />
∼ = End(O<br />
n<br />
Ux<br />
). By virtue of<br />
quasi-compactness, one may select finitely many of the U i covering the whole<br />
of X.<br />
All these étale neighbourhoods together form gluing data for A since the morphism<br />
U x1<br />
F . . .<br />
F<br />
Uxl → X is faithfully flat <strong>and</strong> quasi-compact.<br />
The conditions for gluing data formulated in Proposition 1.9 are equivalent to<br />
giving a Čech cocycle for Ȟ 1 ét (X, Aut(M n)). Different gluing data describing the<br />
same algebra differ by a coboundary.<br />
The Theorem of Skolem-Noether implies that the sheaf Aut(M n ) is the same<br />
as PGL n .<br />
□<br />
3.5. Remark. –––– On a quasi-compact scheme X, the set of all isomorphism<br />
classes of rank n locally free sheaves is given by the Čech cohomology set<br />
Ȟ 1 ét (X, GL n).<br />
This is proven in the same way as Proposition 3.4.<br />
3.6. Notation. –––– Let X be any scheme.<br />
i) We will denote the cohomology class corresponding to an Azumaya algebra<br />
A over X by c(A ) ∈ Ȟ 1 ét (X, PGL n).<br />
ii) We will denote the cohomology class corresponding to a locally free sheaf E<br />
over X by cl(E ) ∈ Ȟ 1 ét (X, GL n).<br />
iii) Recall that there is a short exact sequence<br />
0 −→m −→ GL n −→ PGL n −→ 0<br />
of sheaves. In cohomology, it induces a long exact sequence of pointed sets<br />
Ȟ 1 ét (X, GL n)<br />
ι<br />
−→ Ȟ 1 ét (X, PGL d<br />
n) −→ H 2 ét (X,m).<br />
3.7. Facts. –––– Let X be a quasi-compact scheme.<br />
i) For a locally free sheaf E on X, one has that<br />
ι(cl(E )) = c(End(E )) .<br />
ii) For two Azumaya algebras A <strong>and</strong> B on X, of ranks n <strong>and</strong> m, respectively,<br />
one has<br />
c(A ⊗ OX B) = c(A )∗c(B) .<br />
Here, the operation ∗ is induced by the canonical mapping<br />
PGL n × PGL m −→ PGL nm
70 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
given by<br />
GL n ×GL m = Aut(n a )×Aut(m a ) −→ Aut(n a ⊗m a ) = Aut(nm<br />
a ) = GL nm .<br />
iii) For α ∈ Ȟ 1 ét (X, PGL n) <strong>and</strong> β ∈ Ȟ 1 ét (X, PGL n), we have<br />
d(α∗β) = d(α) + d(β) .<br />
Proof. The verification of these compatibilities is purely technical <strong>and</strong> hence<br />
omitted.<br />
□<br />
3.8. Proposition. –––– Let X be any quasi-compact scheme.<br />
Then, the boundary maps d induce an injective group homomorphism<br />
Proof. First step. i X is well-defined.<br />
i X : Br(X ) −→ H 2 ét (X,m) .<br />
Suppose we are given two Azumaya algebras A <strong>and</strong> B over X which are<br />
similar. This means, we have two locally free sheaves, E <strong>and</strong> F, such that<br />
A ⊗ OX End(E ) ∼ = B ⊗ OX End(E ′ ). In particular,<br />
c(A ⊗ OX End(E )) = c(B ⊗ OX End(E ′ )) .<br />
(∗)<br />
The compatibilities above yield<br />
hence<br />
c(A ⊗ OX End(E )) = c(A )∗c(End(E )) = c(A )∗ι(cl(E )) ,<br />
d(c(A )) = d(c(A ))+d(ι(cl(E ))) = d(c(A )∗ι(cl(E ))) = d(c(A ⊗ OX End(E ))) .<br />
Completely analogously, one shows that d(c(B)) = d(c(B ⊗ OX End(E ′ ))) .<br />
Formula (∗), therefore, yields the claim.<br />
Second step. i X is a group homomorphism.<br />
We have d(c(A ⊗ OX B)) = d(c(A )∗c(B)) = d(c(A )) + d(c(B)) .<br />
Third step. i X is an injection.<br />
Suppose, we have an Azumaya algebra A such that d(c(A )) = 0. Then, according<br />
to the exactness of the long cohomology sequence, c(A ) = ι(cl(E )) for<br />
a locally free sheaf E on X. Thus, A ∼ = End(E ) <strong>and</strong> [A ] = 0 ∈ Br(X ). □<br />
3.9. Definition. –––– Let X be any scheme. Then, H 2 ét (X,m) is called the<br />
cohomological Brauer group of X. It will be denoted by Br ′ (X ).
Sec. 4] THE RELATION TO THE BRAUER GROUP OF THE FUNCTION FIELD 71<br />
3.10. Remark. –––– i X is, in general, not an isomorphism of <strong>groups</strong>. The first<br />
counterexample to surjectivity has been constructed by A. Grothendieck<br />
in [GrBrII, Remarque 1.11.b].<br />
i X is, however, an isomorphism in a number of particular cases. One such case<br />
are smooth algebraic surfaces. This was essentially known to M. Ausl<strong>and</strong>er<br />
<strong>and</strong> O. Goldman [A/G] before Grothendieck’s invention of the general Brauer<br />
group for schemes, may be even before the actual invention of the concept of<br />
a scheme.<br />
In Section 7, we will present the result of Ausl<strong>and</strong>er <strong>and</strong> Goldman in a more<br />
up-to-date formulation. Although this is not the most general result known<br />
today, we hope it may serve as a good illustration for what Br(X ) <strong>and</strong> Br ′ (X )<br />
are <strong>and</strong> which methods might be used in order to compare them.<br />
4. The relation to the Brauer group of the function field<br />
4.1. –––– In this section, we always assume that X is an integral scheme.<br />
Denote by g : SpecK = η → X the inclusion of the generic point.<br />
4.2. Fact. –––– Let X be an integral scheme which is regular, separated, <strong>and</strong> quasicompact.<br />
Then,<br />
Br(X ) ⊆ Br ′ (X ) ⊆ Br ′ (K )<br />
where K := Q(X ) denotes the function field of X.<br />
Proof. Denote by g : η → X the inclusion of the generic point. We have the<br />
short exact sequence<br />
0 −→m,X −→ g ∗m,K −→ Div X −→ 0<br />
of sheaves in the étale topology. Here, Div X = L<br />
H 1 ét (X, Div X ) = M<br />
It follows, at first, that<br />
codim x=1<br />
codimx=1<br />
H 1 ét ({x},) = M<br />
codimx=1<br />
H 2 ét (X,m) ⊆ H 2 ét (X, g ∗m,K ).<br />
Second, we consider the Leray spectral sequence<br />
i x∗x which implies<br />
Hom(G x ,) = 0 .<br />
E p,q<br />
2<br />
:= H p<br />
ét (X, Rq g ∗m,K ) =⇒ H n ét (SpecK,m,K).<br />
By virtue of Hilbert’s Theorem 90, for the first higher direct image, we<br />
have R 1 g ∗m,K = 0. This implies there is a short exact sequence
72 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
0 −→ H 2 ét (X, g ∗m,K ) −→ H 2 ét (SpecK,m,K) −→ H 0 ét (X, R2 g ∗m,K ) −→ 0 .<br />
(†)<br />
of terms of lower order.<br />
□<br />
4.3. Remark. –––– One may therefore think of Br(X ) as follows. It consists of<br />
those classes of the usual Br(Q(X )) which allow an extension over the whole<br />
of X.<br />
4.4. Lemma. –––– Let X be an integral scheme which is separated <strong>and</strong> quasicompact.<br />
Then, for each q > 0, the sheaf R q g ∗m,K is uniquely l-divisible for every<br />
l ∈prime to the residue characteristics of X.<br />
Proof. The inclusion g : η → X is the inverse limit, in the category of all<br />
X-schemes, of the open embeddings g U : U → X for U ⊆ X affine open.<br />
The assumption that X is separated makes sure that all the transition maps<br />
g U ′ ,U : U ′ ⊆<br />
−→ U are affine. In this situation, from [SGA4, Exp. VII, Corollaire<br />
5.11], we see that<br />
R q g ∗ µ l = lim R q g U∗ −→U<br />
µ l .<br />
As the g U are open embeddings, <strong>and</strong> hence smooth, they are acyclic [SGA4,<br />
Exp. XV, Théorème 2.1]. In particular, for q > 0 <strong>and</strong> l prime to the residue<br />
characteristics of X, we have R q g U∗ µ l = R q g U∗ g U ∗ µ l,X = 0.<br />
Altogether,<br />
R q g ∗ µ l = 0 .<br />
The Kummer sequence implies that the multiplication map<br />
l : R q g ∗m,K → R q g ∗m,K<br />
is an isomorphism for every q > 0.<br />
□<br />
4.5. Proposition. –––– Let X be an integral scheme which is regular, separated, <strong>and</strong><br />
quasi-compact.<br />
a) If char(Q(X )) = p > 0 then H 0 ét (X, R2 g ∗m,K ) is a p-power torsion group.<br />
b) If all residue characteristics of X are equal to zero then H 0 ét (X, R2 g ∗m,K ) = 0.<br />
Proof. H 2 ét (SpecK,m,K) = H 2( Gal(K/K ), K ∗) is a torsion group directly by<br />
its definition. As we have a surjection H 2 ét (SpecK,m,K) → H 0 ét (X, R2 g ∗m,K ),<br />
the latter is a torsion group, too.
Sec. 4] THE RELATION TO THE BRAUER GROUP OF THE FUNCTION FIELD 73<br />
b) For every l > 0, the sheaf R 2 g ∗m,K is uniquely l-divisible. Therefore, the<br />
multiplication map<br />
l : H 0 ét (X, R2 g ∗m,K ) → H 0 ét (X, R2 g ∗m,K )<br />
is an isomorphism. As H 0 ét (X, R2 g ∗m,K ) is a torsion group, this implies<br />
H 0 ét (X, R2 g ∗m,K ) = 0.<br />
a) Here, the same argument still works for l prime to p. This shows<br />
H 0 ét (X, R2 g ∗m,K ) has no prime-to-p torsion.<br />
□<br />
4.6. Corollary. –––– Let X be an integral scheme which is regular, separated, <strong>and</strong><br />
quasi-compact. Assume that the residue characteristics of X are equal to zero.<br />
Then, for g : η → X the inclusion of the generic point, one has<br />
H 2 ét (X, g ∗m,K ) ∼ = H 2 ét (SpecK,m,K ) .<br />
Proof. This follows from the exact sequence (†) together with Proposition 4.5.b).<br />
□<br />
4.7. Remark. –––– The same result is true for schemes of dimension at most<br />
one in any characteristic as is often shown in the literature. See, e.g., [GrBrIII,<br />
formule (2.2)].<br />
4.8. Proposition. –––– Let X be an integral scheme which is regular, separated, <strong>and</strong><br />
quasi-compact. Assume that<br />
i) dim X ≤ 1 or<br />
ii) the residue characteristics of X are equal to zero.<br />
Then, there is a short exact sequence<br />
0 −→ Br ′ (X ) −→ Br ′ (SpecQ(X )) −→ M<br />
Hom(G x ,É/) .<br />
codim x=1<br />
Proof. Consider once more the short exact sequence<br />
0 −→m,X −→ g ∗m,K −→ Div X −→ 0<br />
of sheaves in the étale topology. Since H 1 ét (X, Div X ) = 0, as shown above, we<br />
have the following fragment of the long exact sequence<br />
0 −→ H 2 ét (X,m,X ) −→ H 2 ét (X, g ∗m,K ) −→ H 2 ét (X, Div X )<br />
in cohomology. Here, the entry on the left h<strong>and</strong> side is Br ′ (X ), according to<br />
its very definition. The term in the middle is isomorphic to Br ′ (Q(X )) as was
74 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
shown in Corollary 4.6. For the term on the right h<strong>and</strong> side, we have<br />
H 2 ét (X, Div X ) = M<br />
H 2 ét ({x},)<br />
codim x=1<br />
= M<br />
H 2 (G x ,)<br />
codim x=1<br />
= M<br />
Hom(G x ,É/) .<br />
codim x=1<br />
□<br />
4.9. Corollary. –––– Let X be an integral scheme which is regular, separated, <strong>and</strong><br />
quasi-compact. Assume that<br />
i) dim X ≤ 1 or<br />
ii) the residue characteristics of X are equal to zero.<br />
Then, one has<br />
Br ′ (X ) = \<br />
codimx=1<br />
where the intersection takes place in Br ′ (Q(X )).<br />
Proof. “⊆” is clear.<br />
Br ′ (Spec O X,x )<br />
“⊇” From Proposition 4.8, we see that all obstructions against an extension of<br />
a central simple algebra over Q(X ) to the whole of X are given by the <strong>points</strong><br />
of codimension one. For SpecO X,x instead of X, the same is true. □<br />
4.10. Remark. –––– Under certain hypotheses on X, there is the same formula<br />
for the Brauer group instead of the cohomological Brauer group. Compare Theorem<br />
7.1.i), below.<br />
5. The Brauer group <strong>and</strong> the cohomological Brauer group<br />
5.1. –––– In order to compare the two Brauer <strong>groups</strong> in relatively simple cases,<br />
the following lemma is helpful.<br />
5.2. Lemma. –––– Let X = SpecR for a local ring R <strong>and</strong> γ ∈ Br ′ (X ) an element<br />
in the cohomological Brauer group. Assume there exists a morphism π : Y → X of<br />
schemes which is finite <strong>and</strong> faithfully flat such that γ ∈ ker(Br ′ (X ) → Br ′ (Y )).<br />
Then, one has even γ ∈ Br(X ).<br />
Proof. We have Y = SpecS. Nakayama’s lemma implies that S ∼ = R n is free<br />
as an R-module. As π is finite, it has no higher direct images <strong>and</strong> the Leray<br />
spectral sequence collapses to<br />
H 2 ét (Y,m) = H 2 ét (X, π ∗m) .
Sec. 5] THE BRAUER GROUP AND THE COHOMOLOGICAL BRAUER GROUP 75<br />
We also have the long exact sequence<br />
H 1 ét (X, π ∗m/m) −→ H 2 ét (X,m) −→ H 2 ét (X, π ∗m)<br />
at our disposal. γ vanishes in H 2 ét (X, π ∗m). It belongs, therefore, to the image<br />
of H 1 ét (X, π ∗m/m).<br />
The canonical map S ∗<br />
of sheaves<br />
֒→ Aut(R n ) = GL n (R) induces a homomorphism<br />
π ∗m ֒→ GL n .<br />
Finally, we note that the diagram<br />
H 1 ét (X, π ∗m/m)<br />
H 2 ét (X,m)<br />
Ȟ 1 ét (X, GL n/m)<br />
Ȟ 2 ét (X,m)<br />
commutes. Hence, γ is in the image of Ȟ 1 ét (X, GL n/m) = Ȟ 1 ét (X, PGL n).<br />
□<br />
5.3. Corollary. –––– One has Br(SpecK ) = Br ′ (SpecK ) for every field K.<br />
Proof. Let γ ∈ Br ′ (SpecK ) = H 2( Gal(K/K ), K ∗) . Then, there is a finite Galois<br />
extension L/K such that γ ∈ H 2( Gal(L/K ), L ∗) , already. When restricted<br />
to SpecL, the cohomology class γ vanishes.<br />
□<br />
5.4. Corollary. –––– One has Br(SpecR) = Br ′ (SpecR) = Br(SpecR/m) for<br />
every Henselian local ring (R, m). In particular, Br(SpecR) = 0 for R strictly local.<br />
Proof. The equality Br ′ (SpecR) = Br ′ (Spec R/m) follows directly from a<br />
general result on étale cohomology [SGA4, Exp. VIII, Corollaire 8.6].<br />
Thus, let γ ∈ Br ′ (SpecR). For its image in Br ′ (Spec R/m), choose a splitting<br />
field l which is finite over k := R/m. Let a be a primitive element <strong>and</strong> f ∈ k[X ]<br />
be its minimal polynomial. Then, l ∼ = k[X ]/( f ). Choose a lift F ∈ R[X ] of f .<br />
The local ring (S, n) := R[X ]/(F ) is finite <strong>and</strong> flat over R. S is Henselian<br />
by [SGA4, Exp. VIII, 4.1.ii)]. Under pull-back to SpecS, the cohomology<br />
class γ vanishes since Br ′ (SpecS) = Br ′ (Spec S/n).<br />
□<br />
5.5. Corollary. –––– One has Br(SpecR) = Br ′ (SpecR) for every discrete valuation<br />
ring R.<br />
Proof. Let γ ∈ Br ′ (SpecR) ⊆ Br ′ (SpecQ(R)). We know that γ vanishes after<br />
a finite, separable field extension L/Q(R). Take for S the integral closure of R<br />
in L. Then, γ vanishes in Br ′ (SpecS) ⊆ Br ′ (L) <strong>and</strong> S is finite <strong>and</strong> flat over R.<br />
□
76 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
5.6. Remark. –––– For X an integral scheme which is regular, separated, <strong>and</strong><br />
quasi-compact in characteristic zero, one therefore has<br />
Br ′ (X ) = \<br />
Br(SpecO X,x )<br />
in Br(Q(X )).<br />
codimx=1<br />
6. Orders in central simple algebras<br />
i. Orders over a general scheme. —<br />
6.1. Definition. –––– Let X be an integral scheme <strong>and</strong> A be a central simple<br />
algebra over its quotient field Q(X ).<br />
By an order in A over X, one means an O X -algebra A which is coherent as<br />
an O X -module <strong>and</strong> has A as its stalk at the generic point.<br />
6.2. Remark. –––– Assume the integral scheme X is locally Noetherian.<br />
Then, for every central simple algebra A over Q(X ) there exists an order<br />
over X.<br />
Indeed, an order may be constructed as follows. Choose a set {a 1 , . . . , a l } of<br />
generators of A as a Q(X )-vector space. In the sheaf g ∗ A on X, these generate<br />
a coherent subsheaf G . Then, the sheaf (G : G ) ⊆ g ∗ A which is defined by<br />
(G : G )(U ) := {s ∈ g ∗ A(U ) | G (U )·s ⊆ G (U ) }<br />
for every open U ⊆ X is, by definition, an O X -algebra. Clearly, it is coherent<br />
<strong>and</strong> has A as its stalk at the generic point. It is, therefore, an order.<br />
6.3. Remarks. –––– i) If X = SpecR is the spectrum of a Noetherian, integrally<br />
closed domain then every global section s ∈ Γ(X, A ) ⊆ A of every order A is<br />
integral over R.<br />
ii) On a central simple algebra A over a field K, there is the reduced trace<br />
tr: A → K. It admits the property that<br />
(x, y) ↦→ tr(xy)<br />
is a non-degenerate bilinear form [Re, Theorem (9.26)].<br />
Further, if R ⊆ K is integrally closed <strong>and</strong> s a section of an order over SpecR<br />
then tr(s) ∈ R.
Sec. 6] ORDERS IN CENTRAL SIMPLE ALGEBRAS 77<br />
6.4. Lemma. –––– Let X be an integral scheme which is Noetherian <strong>and</strong> normal.<br />
Then, every order A over X in a central simple algebra A over Q(X ) is contained<br />
in a maximal order.<br />
Proof. Let SpecR ⊆ X be affine open. Then, A | SpecR is generated by finitely<br />
many global sections s 1 , . . . , s l . Clearly, s 1 , . . . , s l generate A over K.<br />
If A ′ ⊃ A | SpecR is any order <strong>and</strong> t any global section then ts 1 , . . . , ts l are<br />
global sections <strong>and</strong> tr(ts 1 ), . . . , tr(ts l ) ∈ R. This list of conditions defines a<br />
finite R-module, i.e. a coherent sheaf à over SpecR containing every order<br />
which contains A.<br />
Since X is covered by finitely many open affine sets, every ascending chain of<br />
orders in A over X stops after finitely many steps.<br />
□<br />
6.5. Lemma. –––– Let X be an integral scheme which is Noetherian <strong>and</strong> normal.<br />
Let A be a maximal order over X in a central simple algebra A over Q(X ).<br />
Then, for every (not necessarily closed) point x ∈ X, A | SpecOX,x is a maximal order<br />
in A over Spec O X,x .<br />
Proof. For X affine, this follows from [Re, Theorem (1.11)].<br />
For the general case, it remains to verify that A | V is maximal for every affine<br />
open subscheme SpecR = V ⊆ X. We shall do this by contradiction.<br />
Assume, there would be a larger order B over V . Then, there exists some<br />
non-zero r ∈ R such that r·B ⊆ A . Extend r·B to a coherent subsheaf E ⊆ A<br />
on X in the obvious way. I.e., a local section s ∈ E (U ) is, by definition, a<br />
section of A such that s| U ∩V ∈ (r·B)(U ∩ V ).<br />
Now, consider the order (E : E ) over X. Its restriction to V is easily understood.<br />
One has (E : E )| V = ((r ·B) : (r ·B)) = (B : B) = B as r<br />
commutes with B <strong>and</strong> B is an order. In particular, A | V ⊂ B has the property<br />
that E | V ·A | V ⊆ E | V . Every local section s ∈ (E ·A )(U ) is, therefore, a local<br />
section of A with the additional property that s| U ∩V ∈ E (U ∩ V ). By construction<br />
of E , this means s ∈ E (U ). Hence, E ·A ⊆ E <strong>and</strong> A ⊆ (E : E ).<br />
Since A is a maximal order, this implies A = (E : E ).<br />
Together with (E : E )| V = B, this shows A | V = B in contradiction to our<br />
assumption, B would be a larger order.<br />
(This is nothing but a geometrization of the proof for [Re, Theorem (1.11)].)<br />
□<br />
6.6. –––– Our interest in maximal orders comes from the following proposition.
78 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
6.7. Proposition. –––– Let X be a Noetherian, normal, integral scheme with<br />
generic point η <strong>and</strong> A be an Azumaya algebra over X.<br />
Then, A is a maximal order in A η over X.<br />
Proof. It is clear from the definition that A is an order.<br />
If it were not maximal then there would be a larger order B. B admits a local<br />
section s ∉ A (U ) over a certain subscheme U. We may therefore assume that<br />
X = SpecR is affine, A = Ã, <strong>and</strong> B = ˜B for B A.<br />
We have A⊗ R A op = M n (R) for a certain n ∈Æ. Then, O := B⊗ R A op M n (R)<br />
would be a larger order. I.e., the order M n (R) would not be maximal.<br />
Choose M ∈ O \M n (R). There is one entry x not in R. Elementary matrix<br />
operations produce a diagonal matrix<br />
⎛ ⎞<br />
x 0 · · · 0<br />
0 x · · · 0<br />
⎜<br />
⎝<br />
.<br />
. . ..<br />
⎟ . ⎠ ∈ O .<br />
0 0 · · · x<br />
This element alone generates the commutative R-algebra R[x] which can not be<br />
finite since R is integrally closed. This is a contradiction.<br />
□<br />
6.8. Remark. –––– The same approach yields that M n (A ) is a maximal order<br />
as soon as A is.<br />
ii. Orders over a discrete valuation ring. —<br />
6.9. Notation. –––– In this subsection, we will consider the following situation.<br />
– (R, m) is a discrete valuation ring. By π, we denote a generator of m.<br />
– K := Q(R) is its quotient field,<br />
– A is a central simple algebra over K.<br />
As usual, ̂R <strong>and</strong> ̂K are the completions of R <strong>and</strong> K.<br />
For x ∈ A, we will denote by N A/K (x) := det(·x : A → A) the (non-reduced)<br />
norm of x.<br />
Finally, we will simply formulate that O is an Azumaya algebra (an order)<br />
over R when Õ is an Azumaya algebra (an order) over SpecR.<br />
6.10. Lemma. –––– Assume R to be complete <strong>and</strong> let D = A be a (finite) division<br />
algebra over K. Then,<br />
x ∈ D is integral over R ⇐⇒ N D/K (x) ∈ R .
Sec. 6] ORDERS IN CENTRAL SIMPLE ALGEBRAS 79<br />
Proof. K (x) is a finite field extension of K. In D, two K (x)-modules K (x)·y 1<br />
<strong>and</strong> K (x)·y 2 are either equal or disjoint. Thus, there is a decomposition<br />
rM<br />
D = K (x) · y i<br />
i=1<br />
of D into a direct sum. The multiplication map ·x : D → D is a block matrix.<br />
Consequently,<br />
N D/K (x) = [N K (x)/K (x)] r .<br />
It is, therefore, sufficient to show the assertion for N K (x)/K instead of N D/K .<br />
As K is complete, there is a unique extension of the discrete valuation from K<br />
to K (x). It is given by<br />
ν K (x) (y) :=<br />
1<br />
[K (x) : K ] ν K (N K (x)/K (y)) .<br />
The elements y of valuation ν K (x) (y) ≥ 0 form a finite R-algebra <strong>and</strong> are,<br />
therefore, integral. If ν K (x) (y) < 0 then a relation y n +a n−1 y n−1 + . . . +a 0 = 0<br />
would contradict the ultrametric triangle inequality.<br />
□<br />
6.11. Proposition. –––– Assume R to be complete. If D = A is a (finite) division<br />
algebra over K then, in D, there is exactly one maximal order over R. This is the set<br />
of all elements which are integral over R.<br />
O = {x ∈ D | N D/K (x) ∈ R}<br />
Proof. The only thing not yet proven is that O is actually an order. For this, in<br />
turn, we still have to verify that O is a ring.<br />
Recall that the valuation on D is given by<br />
ν D (x) :=<br />
1<br />
[D : K ] ν K (N D/K (x)) .<br />
The assertion would follow if we could verify the usual formulas<br />
ν D (xy) = ν D (x) + ν D (y) <strong>and</strong> ν D (x + y) ≥ min{ν D (x), ν D (y)}<br />
in the non-commutative situation, too.<br />
For this, note that det(·xy) = det((·x) ◦ (·y)) = det(·x) det(·y) implies<br />
N D/K (xy) = N D/K (x)N D/K (y) from which the first formula follows.<br />
For the ultrametric triangle inequality, we remark first that it is true in the<br />
case x <strong>and</strong> y commute. The point is, K (x, y) is then a finite field extension<br />
of K. One has ν D (·) = ν K (x,y) (·) <strong>and</strong> the inequality carries over from the<br />
commutative case.
80 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
In general, we have x −1 y ∈ O or y −1 x ∈ O. Assume without restriction<br />
that x −1 y ∈ O. Then, since 1 commutes with x −1 y,<br />
<strong>and</strong><br />
ν D (1 + x −1 y) ≥ 0 = min{ν D (1), ν D (x −1 y)}<br />
ν D (x + y) = ν D (x·(1 + x −1 y)) = ν D (x) + ν D (1 + x −1 y) ≥ ν D (x) .<br />
x −1 y ∈ O is equivalent to ν D (x) = min{ν D (x), ν D (y)}.<br />
□<br />
6.12. Remark. –––– For I ⊂ O the maximal ideal, O/I is a division algebra.<br />
O is Azumaya if <strong>and</strong> only if I = mO.<br />
6.13. Proposition. –––– Assume R to be complete. LetD be a (finite) division<br />
algebra over K <strong>and</strong> consider the Azumaya algebra A = M n (D) for a certainn ∈Æ.<br />
Then,<br />
i) ∆ := M n (O) is a maximal order in A.<br />
ii) In ∆, every left ideal is projective as a ∆-module.<br />
Proof. i) is already proven. Cf. Remark 6.8.<br />
ii) Let I ⊆ ∆ be any left ideal.<br />
We observe first that ∆/m∆ ∼ = M n (O/m) is a matrix algebra over a skew field.<br />
In particular, every ∆/m∆-module is projective as a ∆/m∆-module. Consequently,<br />
every ∆/m∆-module is of cohomological dimension ≤ 1 as a ∆-module.<br />
This is true, in particular, for I/mI.<br />
For every i ≥ 1 <strong>and</strong> every ∆-module M, we have the fragments<br />
Ext i ∆ (I, M ) ·π<br />
−→ Ext i ∆ (I, M )<br />
d<br />
−→ Ext i+1<br />
∆ (I/πI, M )<br />
of the Ext-long exact sequence. Here, as i +1 ≥ 2, the rightmost term vanishes.<br />
This means, E := Ext i ∆<br />
(I, M ) has the property that E/mE = 0.<br />
If M is a finite ∆-module then E is a finite ∆-module <strong>and</strong> hence also a finite R-<br />
module. Nakayama’s lemma implies that Ext i ∆ (I, M ) = E = 0. For the general<br />
case, we note that Ext commutes with filtered direct limits in the second argument.<br />
This yields Ext i ∆ (I, M ) = 0 for i ≥ 1, unconditionally. I is projective.<br />
□<br />
6.14. Corollary. –––– Let R be a complete discrete valuation ring <strong>and</strong> A be an<br />
arbitrary Azumaya algebra over K = Q(R). Then, there exists a maximal order<br />
∆ ⊂ A over R such that every left ideal in ∆ is projective.<br />
Proof. There exists a division algebra D such that A = M n (D).<br />
□
Sec. 6] ORDERS IN CENTRAL SIMPLE ALGEBRAS 81<br />
6.15. Remark. –––– We want to abolish the completeness assumption. It will<br />
turn out that Corollary 6.14 is true for arbitrary discrete valuation rings.<br />
6.16. Example. –––– It is, however, not true that every division algebra over a<br />
non-complete discrete valuation ring has a unique maximal order.<br />
Consider, for example, the quaternion algebraÀ:=É·1 ⊕É·i ⊕É·j ⊕É·k<br />
with i 2 = j 2 = k 2 = −1 <strong>and</strong> ij = −ji = k, etc. This is a division algebra<br />
over K =É. InÉ, consider the discrete valuation ring R =(5) corresponding<br />
to the 5-adic valuation. This leads to ̂É=É5.<br />
We have thatÀ⊗ÉÉ5 ∼ = M 2 (É5) splits since (−1) is a square inÉ5. Lemma 6.18<br />
will show that there is a one-to-one correspondence between maximal orders<br />
inÀover(5) <strong>and</strong> maximal orders in M 2 (É5) over5. In particular, there are<br />
several maximal orders.<br />
6.17. Sublemma. –––– Let V be a finite-dimensional K-vector space <strong>and</strong> put<br />
̂V := V ⊗ K ̂K.<br />
Then, there is a bijection<br />
{R-lattices of full rank in V } −→ { ̂R-lattices of full rank in ̂V } ,<br />
Γ ↦→ i<br />
Γ ⊗ R ̂R ,<br />
Proof. We put n := dim V .<br />
Γ ∩ V ι ← Γ .<br />
“ι◦i = id” This follows from the fact that ̂R is faithfully flat over R.<br />
“i◦ι = id” Choose a basis for V .<br />
Then, an ̂R-lattice of full rank in ̂V is given by n linearly independent vectors<br />
in ̂K n , i.e. by a matrix M ∈ GL n (̂K ).<br />
The lattice is not altered by column operations. I.e., by the interchange of two<br />
columns or by adding λ times a column to another column for λ ∈ ̂R. It is not<br />
changed either by the multiplication of a column with a unit of ̂R.<br />
The goal is to verify that these operations allow to transform M into a matrix<br />
in GL n (K ).<br />
Column operations alone may bring M into upper triangular form. Multiplying<br />
the columns by units, we obtain a matrix as follows,<br />
⎛<br />
⎞<br />
π a 1<br />
0 0 . . . 0<br />
c 21 π a 2<br />
0 . . . 0<br />
c 31 c 32 π a 3<br />
. . . 0<br />
.<br />
⎜<br />
⎝<br />
.<br />
. . . ..<br />
⎟<br />
. ⎠<br />
c n1 c n2 c n3 . . . π a n
82 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
The entries under the diagonal are not yet in K. We transform them inductively,<br />
row-by-row. For the first row, there is nothing left to do. So assume, the<br />
entries in rows 1, . . . , k − 1 are already in K. Write c kj = u kj π e j<br />
with a<br />
unit u kj . The operations allowed may alter c kj by any element from ̂Rπ a k .<br />
Since ̂Rπ e j /̂Rπ a k ∼ = Rπ<br />
e j/Rπ a k, the assertion follows.<br />
□<br />
6.18. Lemma. –––– There is a bijection<br />
{ (Maximal) orders in A } −→ { (Maximal) orders in A ⊗ K ̂K } ,<br />
∆ ↦→ ∆ ⊗ R ̂R .<br />
Proof. In our situation, an order is the same as a lattice of full rank which<br />
inherits a ring structure from the central simple algebra. It is, therefore, to<br />
be shown that the operations described in the sublemma above preserve such<br />
ring structures.<br />
This is obvious for ∆ ↦→ ∆ ⊗ R ̂R.<br />
For the other direction, the map is given by Λ ↦→ Λ ∩ A. We note that<br />
to have a ring structure means to contain the unit element <strong>and</strong> closedness<br />
under multiplication. The intersection clearly preserves the unit element.<br />
Closedness under multiplication carries over to ∆ := Λ ∩A, too. Indeed, by the<br />
sublemma, Λ = ∆⊗ R ̂R. If we had ∆ ⊗ R ∆ ։ I <strong>and</strong> I/∆ ≠ 0 then this would<br />
remain true after the faithful flat base change from R to ̂R.<br />
□<br />
6.19. Proposition. –––– LetR be any discretevaluation ring <strong>and</strong> A bean arbitrary<br />
Azumaya algebra over K = Q(R). Then, there exists a maximal order ∆ ⊂ A<br />
over R such that every left ideal in ∆ is projective.<br />
Proof. If R is complete then this is Corollary 6.14.<br />
In general, we have a maximal order Λ ⊂ A⊗ R ̂R over ̂R such that every left<br />
ideal in Λ is projective. Lemma 6.18 provides us with a maximal order ∆ ⊂ A<br />
such that ∆⊗ R ̂R = Λ. We claim, in ∆, every left ideal I is projective.<br />
Indeed, I⊗ R̂R is an ideal in ∆⊗ R̂R <strong>and</strong> therefore projective. Projectivity descends<br />
under this faithful flat map as always<br />
Ext i R (M,N)⊗ R ̂R = Ext îR (M ⊗ R ̂R, N ⊗ R ̂R) .<br />
□<br />
6.20. Lemma. –––– In ∆, every left ideal is principal.<br />
Proof. Let I ⊆ ∆ be any left ideal. Then, KI is a left ideal in A.<br />
Hence, KI = Ae = K ∆e for some idempotent e. In particular, I <strong>and</strong> ∆e<br />
are free R-modules of the same rank.
Sec. 6] ORDERS IN CENTRAL SIMPLE ALGEBRAS 83<br />
From this, it follows that the ideals I/mI <strong>and</strong> (∆e)/m(∆e) in the central simple<br />
algebra ∆/m∆ are isomorphic.<br />
As I <strong>and</strong> ∆e are both projective, we may lift a pair of mutually inverse isomorphisms<br />
to homomorphisms f : I → ∆e <strong>and</strong> g : ∆e → I. f ◦g <strong>and</strong> g ◦ f are<br />
automorphisms by virtue of Nakayama’s lemma. The assertion follows. □<br />
6.21. Proposition. –––– Let R be a discrete valuation ring, K = Q(R), <strong>and</strong> A be<br />
an Azumaya algebra over K.<br />
Then, all maximal orders in A over R are conjugate to each other.<br />
Proof. Let ∆ <strong>and</strong> ∆ ′ be two such orders. We may assume that, in ∆, every left<br />
ideal is principal. Put<br />
F := {x ∈ A | x∆ ′ ⊆ ∆} .<br />
This is a left ∆-module in A, hence a broken ideal. It follows that F = ∆t for a<br />
suitable t ∈ A.<br />
∆ <strong>and</strong> ∆ ′ are orders. Hence, they are full lattices. This implies, there exists<br />
r ∈ R\{0} such that r∆ ′ ⊆ ∆. We have r ∈ F. Consequently, t ∈ A is an<br />
invertible element.<br />
F is, by definition, a right ∆ ′ -module. This shows<br />
t∆ ′ ⊆ ∆t ,<br />
hence ∆ ′ t −1 ⊆ ∆t. Maximality yields the assertion.<br />
□<br />
6.22. Corollary. –––– Let R be a discrete valuation ring, K = Q(R), A an<br />
Azumaya algebra over K, <strong>and</strong> Λ ⊂ A be a maximal order over R.<br />
Then, in Λ, every left ideal is principal.<br />
Proof. Combine Lemma 6.20 with Proposition 6.21.<br />
□<br />
6.23. Remark. –––– There are analogous results for right ideals.<br />
6.24. Remarks. –––– i) If A allows an extension A which is an Azumaya algebra<br />
over SpecR then A is necessarily a maximal order (Proposition 6.7).<br />
Proposition 6.21 shows that every maximal order is Azumaya in this case.<br />
ii) If A does not allow an extension as an Azumaya algebra then the same is true<br />
for M n (A). Indeed, for Λ a maximal order in A, we have that Λ⊗Λ op → End(Λ)<br />
is not surjective. This remains true when going over from Λ to the full matrix<br />
algebra M n (Λ).
84 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
7. The Theorem of Ausl<strong>and</strong>er <strong>and</strong> Goldman<br />
7.1. Theorem (M. Ausl<strong>and</strong>er <strong>and</strong> O. Goldman). —–<br />
Let X be an integral scheme which is regular, separated, <strong>and</strong> quasi-compact. Assume<br />
that the dimension of X is at most two.<br />
i) Then, one has<br />
Br(X ) = \<br />
codimx=1<br />
where the intersection takes place in Br(Q(X )).<br />
ii) In particular, Br(X ) = Br ′ (X ).<br />
Br(Spec O X,x )<br />
Proof. Recall that, by Corollary 5.5, Br(Spec O X,x ) = Br ′ (Spec O X,x ). Further,<br />
Br(Q(X )) = Br ′ (Q(X )).<br />
By virtue of the general Fact 4.2, we have that Br(X ) ⊆ Br(Q(X ))<br />
<strong>and</strong> Br(Spec O X,x ) ⊆ Br(Q(X )). Thus, the natural maps Br(X ) → Br(Spec O X,x )<br />
are all injective. This shows the “⊆”-part of i).<br />
The same argument is true for the cohomological Brauer group. We obtain that<br />
Br ′ (X ) ⊆ \<br />
Br(Spec O X,x ) . (‡)<br />
codimx=1<br />
This, in turn, makes sure that the second assertion immediately follows<br />
from the first. Indeed, the inclusion (‡) together with i) immediately yields<br />
Br ′ (X ) ⊆ Br(X ). The inclusion the other way round was established in Fact 4.2.<br />
The main part is to prove “⊇”. For that, let α ∈ Br ′ (Q(X )) such that<br />
α ∈ \<br />
Br(Spec O X,x ) .<br />
codim x=1<br />
We choose an Azumaya algebra A over Q(X ) corresponding to the class α.<br />
By Lemma 6.4, there exists a maximal O X -order A in A. We claim, A is a<br />
locally free sheaf.<br />
This may be tested locally. Let x ∈ X be any point on X. We write R := O X,x<br />
<strong>and</strong> denote the maximal ideal in R by m. The maximality property of A implies<br />
that A is equal to its bidual. In particular, A x = A ∨∨<br />
x .<br />
We consider the short exact sequence<br />
0 −→ M −→ F 0 −→ A ∨<br />
x −→ 0<br />
of O X,x -modules with F 0 free. Dualizing yields<br />
0 A ∨∨<br />
x<br />
F ∨ 0<br />
M ∨ . . . .<br />
A
Sec. 7] THE THEOREM OF AUSLANDER AND GOLDMAN 85<br />
M ∨ := Hom R (M,R) is a torsion-free R-module. Thus, M ∨ → M ∨ ⊗ R K<br />
is injective. We see, M ∨ may be embedded into a K-vector space. It, therefore,<br />
allows an embedding into a free R-module F 1 , too,<br />
0 −→ A x −→ F ∨ 0 −→ F 1 −→ Q −→ 0 .<br />
R is of cohomological dimension ≤ 2 [Mat, Theorem 19.2]. To establish<br />
that A x is free, it is sufficient to verify that Tor R i (A x , R/m) = 0 for i = 1<br />
<strong>and</strong> i = 2 [Mat, §19, Lemma 1].<br />
For this, writing Q 1 := ker(F 1 → Q), we see<br />
Tor R 2 (A x, R/m) = Tor R 3 (Q 1, R/m) = 0<br />
<strong>and</strong><br />
Tor R 1 (A x, R/m) = Tor R 2 (Q 1, R/m) = Tor R 3 (Q, R/m) = 0 .<br />
A is a locally free sheaf.<br />
Corollary 1.5 shows that the non-Azumaya locus of A is a closed subset T ⊆ X,<br />
pure of codimension one. We have to show that T = ∅.<br />
Assume the contrary. Then, there is an irreducible divisor D = {x} ⊆ T for x<br />
a codimension one point. By assumption, A allows an extension to Spec O X,x<br />
as an Azumaya algebra B. Furthermore, we may consider A | SpecOX,x . This is a<br />
maximal order in A over Spec O X,x . As maximal orders, B <strong>and</strong> A | SpecOX,x are<br />
conjugate by Proposition 6.21. In particular, A | SpecOX,x is an Azumaya algebra.<br />
Thus, the assumption D = {x} ⊆ T leads to a contradiction.<br />
7.2. Remark. –––– Using more refined methods, it was shown that the two<br />
Brauer <strong>groups</strong> coincide in several more cases.<br />
The most general result in this direction is due to O. Gabber ([Ga, Hoo] or<br />
[K/O80]). It shows that Br(X ) = Br ′ (X ) tors if X = U ∪ V is the union of two<br />
affine schemes such that U ∩ V is again affine.<br />
Quite recently, S. Schröer [Schr] showed Br(X ) = Br ′ (X ) tors for any quasicompact,<br />
separated, geometrically normal algebraic surface.<br />
If one assumes X to be regular <strong>and</strong> separated then the subscript may be<br />
=/2<br />
omitted<br />
in view of Fact 4.2. A. Grothendieck knew that Br ′ (X ) may be non-torsion,<br />
even for an affine normal surface over, in 1965/66, already [GrBrII, Remarque<br />
1.11.b].<br />
D. Edidin, B. Hassett, A. Kresch, <strong>and</strong> A. Vistoli [E/H/K/V] (see also [Bt])<br />
constructed an example of a non-separated surface such that Br ′ (X )<br />
but Br(X ) = 0. X is the union of two affine schemes, regular in codimension<br />
one, but not normal.<br />
□
86 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
8. Examples<br />
i. Some rings <strong>and</strong> fields from number theory. —<br />
8.1. Examples. –––– Let p be a prime <strong>and</strong> K be a finite field extension ofÉp.<br />
i) Then, Br(SpecK ) =É/.<br />
This is shown by local class field theory [Se67, Sec. 1, Theorems 1 <strong>and</strong> 2].<br />
Classically, the isomorphism is given as follows. It turns out that<br />
Br(SpecK ) = H 2( Gal(K nr /K ), (K nr ) ∗)<br />
for K nr the maximal unramified extension. The valuation ν : K nr →induces<br />
H 2( Gal(K nr /K ), (K nr ) ∗) ∼ =<br />
−→ H 2( Gal(K nr /K ),) = H 2 (̂,)<br />
= Hom(̂,É/) =É/.<br />
ii) For O K , the ring of integers, one has Br(Spec O K ) = 0.<br />
Indeed, by the description of Br(SpecK ), the canonical homomorphism<br />
Br ′ (SpecK ) −→ Hom ( Gal(p/p),É/)<br />
from Proposition 4.8 is bijective.<br />
8.2. Examples. –––– i) One has Br(SpecÊ) = 1 2/.<br />
ii) One has Br(Spec) = 0.<br />
8.3. Examples. –––– Let K be a number field.<br />
i) Then,<br />
Br(SpecK ) = ker ( M<br />
s :<br />
pprimeÉ/⊕ M<br />
2/−→É/) .<br />
σ : K→Ê1<br />
Here, s is just the summation.<br />
→É/<br />
This is shown by global class field theory [Ta67, 11.2]. The isomorphism is<br />
induced by the canonical maps<br />
i ν : Br(SpecK ) → Br(SpecK ν )<br />
for ν ∈ Val(K ) composed with the invariant map inv ν : Br(SpecK ν )<br />
(or to 1 2/or 0, respectively) from Example 8.1.i).
Sec. 8] EXAMPLES 87<br />
ii) Denote by r the number of real embeddings of K. Then, for the integer<br />
ring O K , one has<br />
{ (<br />
1<br />
Br(Spec O K ) = 2/) r−1 if r ≥ 1 ,<br />
0 otherwise .<br />
In particular, Br(Spec) = 0.<br />
Here, the canonical homomorphism<br />
Br ′ (SpecK ) −→<br />
M<br />
codimx=1<br />
Hom(G x ,É/)<br />
from Proposition 4.8 is nothing but the projection<br />
ker ( M<br />
M<br />
s :<br />
2/−→É/) −→<br />
ii. Geometric examples. —<br />
pprimeÉ/⊕ M<br />
σ : K→Ê1<br />
p primeÉ/.<br />
8.4. Lemma. –––– Let n ∈Æ<strong>and</strong> X be any scheme such that n is invertible on X.<br />
Then, there is the short exact sequence<br />
0 −→ Pic(X )/n Pic(X )<br />
c 1<br />
−→ H<br />
2<br />
ét (X, µ n) −→ Br ′ (X ) n −→ 0 .<br />
Proof. This follows from the long exact sequence in étale cohomology associated<br />
to the Kummer sequence<br />
0 −→ µ n −→m<br />
n<br />
−→m −→ 0 .<br />
□<br />
8.5. Proposition. –––– Let k ⊆be an algebraically closed field <strong>and</strong> X be a<br />
scheme which is regular <strong>and</strong> proper over k.<br />
Then, Br ′ (X ) = (É/) rkH 2 (X (),)−rkNS(X )<br />
⊕ H 3 (X (),) tors .<br />
Proof. Br ′ (X ) is torsion in view of Fact 4.2.<br />
For the middle term in the exact sequence above, by [SGA4, Exp. XVI, Corollaire<br />
1.6], we have H 2 ét (X, µ n) ∼ = H 2 ét (X, µ n ). Further, the comparison theorem<br />
[SGA4, Exp. XI, Théorème 4.4] shows H 2 ét (X, µ n ) ∼ = H 2 (X (),/n).<br />
On the other h<strong>and</strong>, by the Theorem of Murre <strong>and</strong> Oort [SGA6, Exp. XII,<br />
Corollaires 1.2 et 1.5.a)], the Picard functor is representable by a scheme Pic X/k .<br />
This is a group scheme, hence smooth over k. It is the disjoint union of quasiprojective<br />
k-schemes. Among other properties, one has Pic X/k (k) = Pic(X )<br />
<strong>and</strong> Pic X/k () = Pic(X).<br />
The connected components of Pic X/k are in bijection with NS(X ).<br />
Base change from k todoes not change these components. In particular,<br />
NS(X ) = NS(X).
88 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
Corresponding to this, Pic(X ) is an extension of the Néron-Severi group NS(X )<br />
which is finitely generated by a divisible group,<br />
0 −→ Pic 0 (X ) −→ Pic(X ) −→ NS(X ) −→ 0 .<br />
This shows Pic(X )/n Pic(X ) ∼ = NS(X )/nNS(X ) ∼ = NS(X)/nNS(X).<br />
By consequence, we may assume from now on that k =.<br />
Doing this, we first observe that the homomorphism<br />
Pic(X )/n Pic(X ) −→ H 2 ét (X, µ n)<br />
induced by the Kummer sequence is compatible with the usual first Chern class<br />
homomorphism coming from the exponential sequence. Indeed, there is the<br />
commutative diagram<br />
0<br />
2πi O X<br />
exp O ∗ X<br />
0<br />
exp( 2πi<br />
n ·)<br />
0 µ n O ∗ X<br />
exp( 1 n ·)<br />
(·) n O ∗ X<br />
of exact sequences of sheaves which induces the commutative diagram<br />
Pic(X )<br />
Pic(X )<br />
=<br />
0<br />
= H 1 (X, OX ∗ ) H 2 (X,)<br />
=<br />
= H 1 (X, O ∗ X ) H 2 (X, µ n ) .<br />
By the Lefschetz theorem on (1, 1)-classes [G/H, p. 163], the cokernel of the<br />
first Chern class homomorphism<br />
c 1 : Pic(X ) −→ H 2 (X,)<br />
is always torsion-free. Writing r := rkH 2 (X (),) <strong>and</strong> ρ := rkNS(X ), we<br />
therefore have coker c 1<br />
∼ =r−ρ . Thus, the cokernel of the induced homomorphism<br />
c 1 ⊗/n: Pic(X )/n Pic(X ) → H (X,)⊗/n<br />
2<br />
is isomorphic to (/n) r−ρ .<br />
Further, the universal coefficient theorem [Sp, Chap. 5, Sec. 5, Theorem 10]<br />
yields the short exact sequence<br />
0 → H 2 (X (),) ⊗/n−→ H 2 (X (),/n) −→ H 3 (X (),) n → 0 .<br />
Thus, from Lemma 8.4 we see that there is a surjective canonical homomorphism<br />
Br ′ (X ) n → H 3 (X (),) n the kernel of which is isomorphic to<br />
coker(c 1 ⊗/n) ∼ = (/n) r−ρ .
Sec. 8] EXAMPLES 89<br />
Altogether, there is a short exact sequence<br />
0 −→ H −→ Br ′ (X ) −→ H 3 (X (),) tors −→ 0<br />
where #H n = n r−ρ for each n ∈Æ. A simple application of the structure<br />
theorem for finite abelian <strong>groups</strong> shows that (É/) r−ρ is the only torsion<br />
group having this property.<br />
Finally, there are no non-trivial extensions of a finite torsion group by (É/) r−ρ .<br />
□<br />
8.6. Remark. –––– A generalization to an arbitrary algebraically closed field k<br />
of characteristic zero is not difficult. Indeed, to describe X only finitely many<br />
data are needed. Thus, X ∼ = X 0 × Speck0 Speck for a certain scheme X 0 over a<br />
field k 0 which is finitely generated overÉ. k 0 allows an embedding into.<br />
8.7. Examples. –––– Let k ⊆be an algebraically closed field.<br />
i) Let X be a smooth proper curve over k. Then, Br(X ) = 0.<br />
(Actually, in this case, one has even Br(Q(X )) = 0 by the Theorem of Tsen<br />
[SGA4 1 , Arcata, III, Théorème (2.3)].)<br />
2<br />
ii) Let X be a <strong>rational</strong> surface over k (proper, not necessarily minimal).<br />
Then, Br(X ) = 0.<br />
iii) Let X be a K3 surface over k. Then, Br(X ) = (É/) 22−rkPic(X ) .<br />
iv) Let X be an Enriques surface over k. Then, Br(X ) =/2.<br />
8.8. Fact. –––– Let k ⊆be an algebraically closed field <strong>and</strong> X be a smooth<br />
complete intersection of dimension ≥ 3 in P n k .<br />
Then, Br ′ (X ) = 0.<br />
Proof. The Lefschetz hyperplane theorem [Bot, Corollary of Theorem 1]<br />
implies that H 1 (X (),) = 0 <strong>and</strong> H 2 (X (),) =. From the universalcoefficient<br />
theorem for cohomology [Sp, Chap. 5, Sec. 5, Theorem 3], we<br />
deduce H 2 (X (),) =<strong>and</strong><br />
H 3 (X (),) ∼ = Hom(H 3 (X (),),) .<br />
In particular, H 3 (X (),) is torsion-free.<br />
□<br />
iii. Varieties over a number field or local field. —<br />
8.9. –––– In this subsection, we deal with the case that X is a scheme over an<br />
algebraically non-closed field K. The relevant cases for us are that K is either a<br />
number field or a local field.
90 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
It is not true that the homomorphism Br(SpecK ) → Br(X ), induced by the<br />
structural map, is always injective. For this, an additional hypothesis is needed.<br />
8.10. Proposition. –––– Let K be a field <strong>and</strong> π : X → SpecK be any scheme<br />
over K. Assume that<br />
a) either K is a local field <strong>and</strong> X (K ) ≠ ∅<br />
b) or K is a number field <strong>and</strong> ∏ X (K ν ) ≠ ∅.<br />
ν∈Val(K )<br />
Then, the canonical map π ∗ : Br(SpecK ) → Br(X ) is injective.<br />
Proof. a) A K-valued point x : SpecK → X induces a section of π ∗ .<br />
b) Put S := F ν SpecK ν . The assumption says that X has an S-valued point.<br />
Therefore, the natural homomorphism<br />
i : Br(K ) → Br(S) = ∏ Br(K ν )<br />
ν<br />
factors via π ∗ . i is injective by Example 8.3.i).<br />
8.11. Proposition. –––– Let K be a number field or a local field <strong>and</strong><br />
π : X → SpecK be a geometrically integral scheme which is proper over K.<br />
i) Then, there is an exact sequence<br />
0 −→ Pic(X )<br />
i<br />
−→ Pic(X K<br />
) Gal(K/K ) π<br />
−→ Br(SpecK ) −→<br />
∗<br />
π ∗<br />
−→ ker(Br ′ (X ) i′<br />
→ Br ′ (X K<br />
)) −→ H 1( Gal(K/K ), Pic(X K<br />
) ) −→ 0 .<br />
Here, i <strong>and</strong> i ′ are induced by the natural morphism X K<br />
→ X.<br />
ii) If π ∗ : Br(SpecK ) → Br ′ (X ) is injective then there is the short exact sequence<br />
0 → Br(SpecK) → π∗<br />
ker(Br ′ (X ) → i′<br />
Br ′ (X K<br />
)) → H 1( Gal(K/K ), Pic(X K<br />
) ) → 0 .<br />
Further, Pic(X ) ∼ = Pic(X K<br />
) Gal(K/K ) .<br />
Proof. Consider the Hochschild-Serre spectral sequence<br />
in étale cohomology.<br />
E p,q<br />
2<br />
:= H p( Gal(K/K ), H q ét (X K ,m) ) =⇒ H n ét (X,m)<br />
As X is integral <strong>and</strong> proper over K, we have that H 0 ét (X K ,m) = K ∗ . Further,<br />
H 1 ét (X K ,m) = Pic(X K<br />
) <strong>and</strong> H 2 ét (X K ,m) = Br ′ (X K<br />
).<br />
This yields E 0,1<br />
2<br />
= Pic(X K<br />
) Gal(K/K ) <strong>and</strong> E 2,0<br />
2<br />
= Br(SpecK ). Furthermore,<br />
E 1,1<br />
2<br />
= H 1( Gal(K/K ), Pic(X K<br />
) ) <strong>and</strong> E 0,2<br />
2<br />
= Br ′ (X K<br />
) Gal(K/K ) .<br />
The classical Theorem Hilbert 90 shows that E 1,0<br />
2<br />
= H 1( Gal(K/K ), K ∗) = 0.<br />
□
Sec. 8] EXAMPLES 91<br />
In addition, if K is a number field then one has E 3,0<br />
2<br />
= H 3( Gal(K/K ), K ∗) = 0<br />
by [Ta67, Section 11.4]. Otherwise, if K is a local field then, by class field theory<br />
H 3( Gal(L/K ), L ∗) ∼ = H<br />
1 ( Gal(L/K ),) = 0 for every finite extension of K.<br />
Thus, H 3( Gal(K/K ), K ∗) = 0.<br />
Finally, E 1 = H 1 ét (X,m) = Pic(X ) <strong>and</strong> E 2 = H 2 ét (X,m) = Br ′ (X ).<br />
The sequence in i) is nothing but a sequence of lower order terms in the spectral<br />
sequence E.<br />
ii) directly follows from i).<br />
8.12. Corollary. –––– Let K be a number field or local field <strong>and</strong> π : X → SpecK<br />
be a geometrically integral scheme which is proper over K.<br />
i) Then, Br(X ) is a countable group.<br />
ii) Suppose that rk NS(X ) = rkH 2 (X (),) <strong>and</strong> rkH 1 (X (),) = 0. Then,<br />
Br(X )/π ∗ Br(SpecK ) is finite.<br />
Proof. i) Br(X K<br />
) is a countable group by Proposition 8.5. It is therefore sufficient<br />
to verify that ker(Br(X ) → Br(X K<br />
)) is a countable group. We will even show<br />
that ker(Br ′ (X ) → Br ′ (X K<br />
)) is countable.<br />
It is known from Example 8.3.i) that Br(SpecK ) is countable. We are left with<br />
proving H 1( Gal(K/K ), Pic(X K<br />
) ) is a countable group.<br />
There is a short exact sequence of Gal(K/K )-modules<br />
0 −→ Pic 0 (X K<br />
) −→ Pic(X K<br />
) −→ NS(X K<br />
) −→ 0 .<br />
Here, H 1( Gal(K/K ), NS(X K<br />
) ) is finite since NS(X K<br />
) is a finitely generated<br />
abelian group acted upon by a finite quotient of Gal(K/K ).<br />
On the other h<strong>and</strong>, Pic 0 (X K<br />
) is divisible <strong>and</strong> all the <strong>groups</strong> Pic 0 (X K<br />
) n are finite.<br />
The Kummer sequence<br />
0 −→ Pic 0 (X K<br />
) n −→ Pic(X K<br />
)<br />
n<br />
−→ Pic(X K<br />
) −→ 0<br />
induces a surjection H 1( Gal(K/K ), Pic 0 (X K<br />
) n<br />
) → H<br />
1 ( Gal(K/K ), Pic 0 (X K<br />
) ) n .<br />
In particular, H 1( Gal(K/K ), Pic 0 (X K<br />
) ) n is finite <strong>and</strong> H 1( Gal(K/K ), Pic 0 (X K<br />
) )<br />
is countable.<br />
ii) Here, the assumption rk NS(X ) = rkH 2 (X (),) makes sure that Br(X K<br />
)<br />
is finite. It is therefore sufficient to verify that<br />
is finite. Again, we will even show that<br />
ker(Br(X ) → Br(X K<br />
))/π ∗ Br(SpecK )<br />
ker(Br ′ (X ) → Br ′ (X K<br />
))/π ∗ Br(SpecK ) ∼ = H 1( Gal(K/K ), Pic(X K<br />
) )<br />
□
92 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
is finite.<br />
For this, the assumption rkH 1 (X (),) = 0 implies that Pic(X K<br />
) = NS(X K<br />
).<br />
The assertion follows.<br />
□<br />
8.13. Remark. –––– The case that Br ′ (X K<br />
) = 0 is of particular interest. For<br />
example, this happens if X K<br />
is a <strong>rational</strong> surface, a K3 surface, or a smooth<br />
complete intersection of dimension ≥ 3 in P n (cf. Subsection ii).<br />
K<br />
In this case, the short exact sequence in assertion ii) simply means<br />
Br ′ (X )/π ∗ Br(SpecK) ∼ = H 1( Gal(K/K ), Pic(X K<br />
) ) .<br />
8.14. Fact. –––– LetK beanumber field or alocalfield <strong>and</strong>X beasmoothcomplete<br />
intersection of dimension ≥ 3 in P n K .<br />
i) Then, Br(X ) = Br ′ (X ).<br />
ii) The homomorphism π ∗ : Br(SpecK ) → Br(X ) induced by the structural map<br />
is surjective.<br />
iii) If X (K ) ≠ ∅ for K a local field or ∏ X (K ν ) ≠ ∅ for K a number field<br />
ν∈Val(K )<br />
then Br(X ) = Br(SpecK ).<br />
Proof. The Lefschetz hyperplane theorem implies that Pic(X K<br />
) =. Therefore,<br />
H 1( Gal(K/K ), Pic(X K<br />
) ) = H 1( Gal(K/K ),) = 0 .<br />
Since Br ′ (X K<br />
) = 0, Proposition 8.11 shows that π ∗ : Br(SpecK ) → Br ′ (X )<br />
is surjective.<br />
In particular, π ∗ does not factorize via any proper subgroup of Br ′ (X ).<br />
Hence, Br(X ) = Br ′ (X ). This proves i) <strong>and</strong> ii).<br />
iii) follows now from Proposition 8.10.<br />
8.15. –––– In the situation of a curve over a local field, Br(X )/π ∗ Br(SpecK )<br />
has been computed by S. Lichtenbaum. The following result is often referred<br />
to as Lichtenbaum duality.<br />
8.16. Theorem (Lichtenbaum). —– Let p beaprimenumber,K afiniteextension<br />
ofÉp, <strong>and</strong> X be a geometrically integral curve which is proper <strong>and</strong> smooth over K.<br />
Then, the canonical pairing<br />
Br(X ) × X (K ) −→É/,<br />
(α, x) ↦→ inv νp (α| x )<br />
□<br />
induces an isomorphism Br(X )/π ∗ ∼=<br />
Br(SpecK ) −→ Hom(Pic 0 (X ),É/).<br />
Proof. This is [Lic, Corollary 1].<br />
□
Sec. 8] EXAMPLES 93<br />
iv. Manin’s formula. —<br />
8.17. –––– If X K<br />
is a <strong>rational</strong> surface then H 1( Gal(K/K ), Pic(X K<br />
) ) is often<br />
effectively computable using the following result due to Yu. I. Manin.<br />
8.18. Proposition (Manin, [Man,кIV,ÈÖÐÓÒº¿]). —–<br />
Let X be a regular, integral scheme which is of finite type over a field k. Suppose<br />
further, we are given a finite, G := Gal(k sep /k)-invariant set {D i } of divisors<br />
on X := X × Speck Speck sep generating Pic(X ).<br />
Denote by S ⊆ Div(X ) the group generated by {D i } <strong>and</strong> by S 0 ⊆ S the subgroup<br />
of all principal divisors. Finally, let H ⊆ G be a normal subgroup acting trivially<br />
on {D i }.<br />
Assume,<br />
i) Pic(X ) is a free abelian group <strong>and</strong><br />
ii) there is a perfect pairing<br />
Then, there is a canonical isomorphism<br />
H 1 (G, Pic(X ))<br />
Pic(X ) × Pic(X ) −→.<br />
i<br />
−→ Hom((NS ∩ S 0 )/NS 0 ,É/) .<br />
Here, N : S → S denotes the norm map on S as a G/H -module.<br />
Proof. By Lemma 8.19, we have<br />
Ĥ −1 (G/H , Pic(X )) ∼ = (NS ∩ S 0 )/NS 0 .<br />
To this relation, we apply the duality theorem [C/E, Chap. XII, Corollary 6.5].<br />
It shows that<br />
Ĥ 0 (G/H , Hom(Pic(X ),É/)) ∼ = Hom(Ĥ −1 (G/H , Pic(X )),É/)<br />
It remains to construct a canonical isomorphism<br />
= Hom((NS ∩ S 0 )/NS 0 ,É/) .<br />
Ĥ 0 (G/H , Hom(Pic(X ),É/)) ∼ = H 1 (G, Pic(X )) .<br />
For this, since Pic(X ) is free, we have a short exact sequence of G/H -modules<br />
0 −→ Hom(Pic(X ),) −→ Hom(Pic(X ),É) −→ Hom(Pic(X ),É/) −→ 0 .<br />
As Hom(Pic(X ),É) is uniquely divisible, this shows<br />
Ĥ 0 (G/H , Hom(Pic(X ),É/)) ∼ = H 1 (G/H , Hom(Pic(X ),)) .
94 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
We finally apply the perfect pairing on Pic(X ).<br />
□<br />
8.19. Lemma. –––– Let G be a finite group <strong>and</strong> D be a finite G-set. Consider the<br />
free abelian group S over D as a G-module. Then, for every G-submodule S 0 ⊆ S,<br />
there is a canonical isomorphism<br />
Ĥ −1 ∼=<br />
(G, P) −→ (NS ∩<br />
=<br />
S 0 )/NS 0 .<br />
Proof. We first calculate Ĥ −1 (G, S) directly, according to the definition. Observe<br />
that S is a direct sum of G-modules of the form[G/H ] for normal<br />
sub<strong>groups</strong> H ⊆ G. For these, one has<br />
H 0 (G,[G/H ]) ∼<br />
<strong>and</strong> the isomorphism is given by the map[G/H ] →sending a formal<br />
sum to the sum of all coefficients. Further, the homomorphism→[G/H ]<br />
mapping 1 to ∑ g∈G/H g has no kernel. Thus, Ĥ −1 (G, S) = 0.<br />
The short exact sequence<br />
0 −→ S 0 −→ S −→ P −→ 0<br />
of G-modules therefore induces a long exact sequence<br />
0 −→ Ĥ −1 (G, P) −→ Ĥ 0 (G, S 0 ) −→ Ĥ 0 (G, S) .<br />
on Tate cohomology. We may consequently write<br />
Ĥ −1 (G, P) = ker(S G 0 /NS 0 → S G /NS) ∼ = (NS ∩ S 0 )/NS 0 .<br />
□<br />
8.20. Remark. –––– The assumptions are fulfilled if X is a proper surface such<br />
that H 1 (X (),) = 0 <strong>and</strong> H 2 (X, O X ) = 0. For example, X may be a (not<br />
necessarily minimal) <strong>rational</strong> surface or a K3 surface.<br />
The first condition implies H 1 (X (),) = 0 <strong>and</strong> H 2 (X (),) is torsion-free.<br />
In particular, the first Chern class<br />
c 1 : Pic(X) → H 2 (X (),)<br />
is injective. The second condition makes sure it is a surjection. As H 2 (X (),)<br />
is torsion-free, Poincaré duality yields that the intersection pairing is perfect.<br />
8.21. –––– Let us consider the case that X is a smooth cubic surface in a bit<br />
more detail.
Sec. 8] EXAMPLES 95<br />
If is well known that, on a cubic surface X over an algebraically closed field,<br />
there are exactly 27 lines D 1 , . . . , D 27 . Further, Pic(X ) ∼ =7 is generated by<br />
the classes of these lines.<br />
The set L := {D 1 , . . . , D 27 } is equipped with the intersection product<br />
〈 , 〉: L ×L → {−1, 0, 1}. The pair (L , 〈 , 〉) is the same for all smooth<br />
cubic surfaces. It is well known [Man,кIV,ÌÓÖѽººµ] that the group<br />
of permutations of L respecting 〈 , 〉 is isomorphic to W (E 6 ). We fix such<br />
an isomorphism.<br />
Further, we put S ∼ =27 to be the free group over L <strong>and</strong><br />
{ ∣ }<br />
∣∣<br />
S 0 := ∑ a i D i ∑a i 〈D i , D j 〉 = 0 for j = 1, . . . , 27 .<br />
i i<br />
Let X be a smooth cubic surface over a number field K. Then, Gal(K/K )<br />
operates canonically on the set L X of the 27 lines on X K<br />
. Fix a bijection<br />
∼ =<br />
i X : L X −→ L respecting the intersection pairing. This induces a group homomorphism<br />
ι X : Gal(K/K ) → W (E 6 ). We denote its image by G ⊂ W (E 6 ).<br />
As L X is clearly Gal(K/K )-invariant, we may apply Proposition 8.18. It shows<br />
H 1( Gal(K/K ), Pic(X K<br />
) ) = H 1 (G, S/S 0 ) = Hom ( (NS ∩ S 0 )/NS 0 ,É/) .<br />
The right h<strong>and</strong> side depends only on the conjugacy class of the subgroup<br />
G ⊆ W (E 6 ). Actually, it depends only on the decomposition of L<br />
into G-orbits.<br />
8.22. Remark. –––– Almost as a byproduct, we may write down a similar<br />
formula for another arithmetic invariant of the cubic surface X, namely for the<br />
Picard rank. Indeed, Proposition 8.11.i) shows<br />
rkPic(X ) = rkH 0( Gal(K/K ), Pic(X K<br />
) )<br />
<strong>and</strong>, therefore, rk Pic(X ) = rk(S/S 0 ) G = rkN(S/S 0 ) = rk[(NS + S 0 )/S 0 ]. I.e.,<br />
rkPic(X ) = rkNS/(NS ∩ S 0 )<br />
= rkNS − rk(NS ∩ S 0 ) .<br />
Observe that rkNS is the number of Gal(K/K ) orbits into which the<br />
27 lines are decomposed. The group NS ∩ S 0 has to be computed anyway<br />
for H 1( Gal(K/K ), Pic(X K<br />
) ) .<br />
8.23. Explicit computation. –––– There are exactly 350 conjugacy classes of<br />
sub<strong>groups</strong> in W (E 6 ). UsingGAP, we computed the right h<strong>and</strong> side in each case.<br />
The result is the following list.
96 ON THE BRAUER GROUP OF A SCHEME [Chap. II<br />
TABLE 1. H 1 (G,Pic) <strong>and</strong> rk Pic(X ) for smooth cubic surfaces<br />
1 #U = 1 [ ], #H^1 = 1, Rk(Pic) = 7, Orbits [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,<br />
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ]<br />
2 #U = 2 [ 2 ], #H^1 = 1, Rk(Pic) = 6, Orbits [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,<br />
1, 2, 2, 2, 2, 2, 2 ]<br />
3 #U = 2 [ 2 ], #H^1 = 4 [ 2, 2 ], Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2,<br />
2, 2, 2, 2 ]<br />
...........................................<br />
347 #U = 1440 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 12, 15 ]<br />
348 #U = 1920 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 10, 16 ]<br />
349 #U = 25920 [ ], transitive<br />
/2<br />
350 #U = 51840 [ 2 ], transitive<br />
/3<br />
0 for 257 classes,<br />
for 65 classes,<br />
for 16 classes,<br />
(/2) 2 for 11 classes,<br />
The code written in GAP as well as the full list are reproduced in the appendix.<br />
8.24. –––– It turns out that H 1( Gal(K/K ), Pic(X K<br />
) ) is isomorphic to<br />
(/3) 2<br />
for one class.<br />
The computations took approximately 28 seconds of CPU time.<br />
In 25 of the 257 classes, we have H 1( Gal(K/K ), Pic(X K<br />
) ) = 0 for the trivial<br />
reason that the operation of G on the 27 lines is transitive. Among these classes,<br />
there are the general cubic surfaces. I.e., the case that G = W (E 6 ). We will<br />
consider general cubic surfaces in more detail in Chapter IX.<br />
A different case where H 1( Gal(K/K ), Pic(X K<br />
) ) = 0 is that when G = 0.<br />
This means that all 27 lines are defined over K.<br />
The case where H 1( Gal(K/K ), Pic(X K<br />
) ) = (/3) 2 was described already<br />
by Yu. I. Manin [Man,кVI,ËÐ×ØÚº]. It occurs in a situation<br />
when G ∼ =/3splits the 27 lines into nine orbits of three lines each. This is<br />
realized, for example, by the diagonal cubic surface “x 3 + y 3 + z 3 + dw 3 = 0”<br />
in P 3 K when K ( √ 3 d)/K is a cubic Galois extension.<br />
8.25. Remark. –––– That only these five <strong>groups</strong> may appear was known to<br />
Sir P. Swinnerton-Dyer in 1993, already. Swinnerton-Dyer’s proof fills almost<br />
the entire article [S-D93]. A different proof is given in the Ph.D. thesis of<br />
P. K. Corn [Cor]. That proof consists of a mixture of mathematical arguments<br />
with Computer work.<br />
A purely computational approach was, seemingly, considered rather hopeless at<br />
that time (cf. [Cor, Proposition 1.3.11] <strong>and</strong> the remarks before).
Sec. 8] EXAMPLES 97<br />
8.26. Remark. –––– We find a Picard rank of<br />
1 for 137 classes,<br />
2 for 133 classes,<br />
3 for 62 classes,<br />
4 for twelve classes,<br />
5 for four classes,<br />
6 for one class,<br />
7 for one class.<br />
8.27. Remark. –––– In the next chapter, we will use the list given in Table 1<br />
in order to describe the Brauer-Manin obstruction for a large class of diagonal<br />
cubic surfaces.
CHAPTER III<br />
AN APPLICATION:<br />
THE BRAUER-MANIN OBSTRUCTION<br />
I was at first almost frightened when I saw such mathematical force made<br />
to bear upon the subject, <strong>and</strong> then wondered to see that the subject stood it<br />
so well.<br />
MICHAEL FARADAY (1857, in a letter to J. C. Maxwell)<br />
1. Adelic <strong>points</strong><br />
i. The concept of an adelic point. —<br />
1.1. Definition. –––– Let K be a number field <strong>and</strong> X be a scheme over K.<br />
Then, an adelic point on X is a morphism SpecK → X of K-schemes from<br />
the spectrum of the adele ring [Cas67, Sec. 14]. The set of all adelic <strong>points</strong> on X<br />
is denoted by X (K).<br />
1.2. Remark. –––– The scheme SpecK is not Noetherian.<br />
1.3. Remark. –––– K is a subring ofK via the diagonal homomorphism.<br />
Thus, every K-valued point on X induces an adelic point.<br />
1.4. Remarks. –––– i)K is canonically a sub-K-algebra of ∏ ν K ν . Therefore,<br />
every adelic point on X gives rise to a ∏ ν K ν -valued point.<br />
ii) A ∏ ν K ν -valued point induces a K ν -valued point for every ν ∈ Val(K ).<br />
If the scheme X is separated <strong>and</strong> quasi-compact then the ∏ ν K ν -valued <strong>points</strong><br />
are in a canonical bijection with ∏ ν X (K ν ). This is a technical point which we<br />
defer to Lemma 3.2.
100 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
1.5. Lemma. –––– Let K be a number field <strong>and</strong> X be a separated K-scheme.<br />
Then, the sequence x = (x ν ) ν ∈ ∏ ν X (K ν ) determines the corresponding adelic<br />
point, if it exists, uniquely.<br />
Proof. We first observe that the set<br />
[<br />
SpecK ν ⊂ SpecK<br />
ν∈Val(K )<br />
is Zariski dense. Indeed, a base of the open subsets is provided by the sets D( f )<br />
for 0 ≠ f = ( . . . , f ν , . . . ) ∈K. If f ν ≠ 0 then SpecK ν is contained<br />
in D( f ).<br />
Now suppose that two morphisms x 1 , x 2 : SpecK → X induce the<br />
same element in ∏ ν X (K ν ). This means simply that x 1 <strong>and</strong> x 2 coincide<br />
on S ν∈Val(K ) SpecK ν .<br />
In the category of schemes, there exists a universal morphism i : V → SpecK<br />
such that x 1 ◦i = x 2 ◦i. According to [EGA, Chapitre I, Proposition (5.2.5)],<br />
i is a closed embedding.<br />
Therefore, x 1 <strong>and</strong> x 2 coincide on the Zariski closure of S ν∈Val(K ) SpecK ν . This is<br />
the whole of SpecK.<br />
□<br />
1.6. Lemma. –––– Let K be a number field, O K its ring of integers, <strong>and</strong> X be a<br />
separated O K -scheme of finite type. Denote by X its generic fiber.<br />
Then, a ∏ ν K ν -valued point x = (x ν ) ν ∈ ∏ ν X (K ν ) is induced by an adelic point if<br />
<strong>and</strong> only if all but finitely many of the x ν extend to O Kν -valued <strong>points</strong> on X .<br />
Proof. “⇐=” We are given a finite set S of valuations, including all archimedean<br />
ones, <strong>and</strong> morphisms SpecO Kν → X for ν ∉ S <strong>and</strong> SpecK ν → X for ν ∈ S.<br />
To ease notation, we write R ν := K ν for ν ∈ S <strong>and</strong> R ν := O Kν for ν ∉ S.<br />
Thus, we have morphisms SpecR ν → X . We claim that these may be put<br />
together to give a morphism SpecR → X for R := ∏ ν∈Val(K ) R ν . The assertion<br />
then follows as there is a canonical morphism SpecK → SpecR.<br />
For that, we first note that the assertion is true when X = SpecA is an<br />
affine scheme. Indeed, to give a ring homomorphism A → ∏ ν∈Val(K ) R ν is the<br />
same as giving a system of ring homomorphisms {A → R ν } ν∈Val(K ) .<br />
We cover X by finitely many affine schemes X j<br />
∼ = SpecAj , j ∈ {1, . . . , n}.<br />
Given a system of morphisms {SpecR ν → X } ν∈Val(K ) , we may find a decomposition<br />
Val(K ) = S 1 ∪ . . . ∪ S n<br />
into mutually disjoint subsets such that SpecR ν maps to X j if ν ∈ S j . For this,<br />
observe that all the rings R ν are local.
Sec. 1] ADELIC POINTS 101<br />
Now we use the assumption that the X j are affine. For j fixed, the morphisms<br />
SpecR ν → X j for ν ∈ S j give rise to a morphism<br />
Finally, observe that<br />
nG<br />
j=1<br />
since the index set is finite.<br />
Spec ( ∏<br />
ν∈S j<br />
R ν<br />
) → Xj ⊆ X .<br />
Spec ( )<br />
∏ R ν ∼=<br />
( )<br />
Spec ∏ R ν<br />
i∈S j ν∈Val(K )<br />
“=⇒” Cover X by affine schemes X 1 , . . . , X n <strong>and</strong> write<br />
X j = Spec O K [T j1 , . . . , T jlj ]/I j .<br />
The adelic point given induces homomorphisms of O K -algebras<br />
ϕ j : O K [T j1 , . . . , T jlj ]/I j −→ (K ) fj<br />
where ( f 1 , . . . , f n ) = (1). Fix adeles g 1 , . . . , g n such that g 1 f 1 + . . . +g n f n = 1<br />
<strong>and</strong> write<br />
h jk<br />
:= ϕ j (T jk )<br />
where h jk ∈K <strong>and</strong> e jk ≥ 0.<br />
f e jk<br />
j<br />
We let S be the set of all valuations ν ∈ Val(K ) which are either archimedean<br />
or such that not all of the adeles g j <strong>and</strong> h jk are integral at ν. This is a finite set<br />
of valuations.<br />
Assume ν ∉ S. Then, for one adele f i , we certainly have that ‖( f i ) ν ‖ ν ≥ 1.<br />
Since (h jk ) ν is integral, we see that [ϕ j (T jk )] ν ∈ O Kν for every k. ϕ j induces a<br />
morphism SpecO K → X j as claimed.<br />
□<br />
1.7. Definition (Topology on X (K)). —– Let K be a number field <strong>and</strong> X be<br />
a separated K-scheme of finite type.<br />
Then, X allows a model X which is separated <strong>and</strong> of finite type over the integer<br />
ring O K . The sets X (K ν ) carry natural topologies as ν-adic analytic spaces.<br />
For ν non-archimedean we have the subsets X (O Kν ) ⊆ X (K ν ) carrying the<br />
subspace topology.<br />
We equip the set X (K) of all adelic <strong>points</strong> on X with the restricted topological<br />
product topology [Cas67, Sec. 13]. A basis is provided by all direct products<br />
∏<br />
ν<br />
Γ ν<br />
where Γ ν ⊆ X (K ν ) is open for all ν <strong>and</strong> Γ ν = X (O Kν ) for almost all ν.
)É<br />
102 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
1.8. Remarks. –––– a) For two different models X 1 <strong>and</strong> X 2 , the isomorphism<br />
=<br />
(X 1 )É∼ −→ X ∼ =<br />
−→ (X 2<br />
may be extended to an open neighbourhood of the generic fiber. This implies<br />
X 1 (O Kν ) ∼ = X 2 (O Kν ) for all but finitely many valuations ν. Hence, the definition<br />
is independent on the choice of the model. The existence of a model is<br />
provided by Lemma 3.4.a).<br />
b) X (K) is a Hausdorff topological space satisfying the second axiom of countability.<br />
i) If X = A 1 then there is a bijection between A 1 (K ) <strong>and</strong> the adele ringK.<br />
The topology we defined on A 1 (K) coincides with the adele topology onK<br />
defined in [Cas67, Sec. 14].<br />
ii) If X = A 1 \{0} then X (K) corresponds to the idele group. It is, however,<br />
equipped with the restriction of the adele topology, not with the idele topology.<br />
1.9. Lemma. –––– Let K be a number field <strong>and</strong> X be a scheme proper over K.<br />
Then,<br />
X (K) = ∏ X (K ν )<br />
ν∈Val(K )<br />
equipped with the Tychonov topology.<br />
Proof. Lemma 3.4.b) guarantees the existence of a model X which is proper<br />
over O K except for a finite set S of special fibers. The valuative criterion for<br />
properness implies that X (O Kν ) = X (K ν ) for ν ∉ S.<br />
□<br />
ii. Weak approximation <strong>and</strong> the Hasse principle. —<br />
1.10. Definition. –––– Let X be a separated scheme which is of finite type over<br />
a number field K.<br />
a) One says that X satisfies the Hasse principle if the condition X (K) ≠ ∅<br />
implies X (K ) ≠ ∅.<br />
b) X is said to satisfy weak approximation if X (K ) is dense in X (K).<br />
1.11. Theorem. –––– a) The following classes of projective schemes satisfy the<br />
Hasse principle.<br />
i) Smooth projective quadrics,<br />
ii) Brauer-Severi varieties,<br />
iii) smooth projective cubics in P nÉfor n ≥ 9,<br />
iv) smooth complete intersections of two quadrics in P n K<br />
for n ≥ 8.
Sec. 2] THE BRAUER-MANIN OBSTRUCTION 103<br />
b) The following projective schemes satisfy weak approximation.<br />
i) Projective varieties <strong>rational</strong> over K,<br />
for n ≥ 6 satisfy-<br />
ii) smooth complete intersections X of two quadrics in P n K<br />
ing X (K ) ≠ ∅.<br />
1.12. Remark. –––– These lists are not meant to be exhaustive. Further notes<br />
<strong>and</strong> references to the literature may be found in [Sk, Sec. 5.1].<br />
2. The Brauer-Manin obstruction<br />
2.1. –––– Weak approximation <strong>and</strong> even the Hasse principle are not always satisfied.<br />
Isolated counter examples have been known for a long time.<br />
Genus one curves violating the Hasse principle have been constructed by<br />
C.-E. Lind [Lin] as early as 1940 <strong>and</strong> by E. S. Selmer [Sel] in 1951.<br />
The first example of a cubic surface not fulfilling the Hasse principle is due to<br />
Sir P. Swinnerton-Dyer [S-D62]. A series of examples generalizing Swinnerton-Dyer’s<br />
work has been given by L. J. Mordell [Mord]. We will generalize<br />
Mordell’s examples even further in Section 5.<br />
Diagonal cubic surfaces which are counterexamples to the Hasse principle have<br />
been constructed by J. W. S. Cassels <strong>and</strong> M. J. T. Guy [Ca/G] as well as A. Bremner<br />
[Bre]. These investigations were systematized by J.-L. Colliot-Thélène,<br />
D. Kanevsky, <strong>and</strong> J.J. Sansuc [CT/K/S]. We will discuss this topic in more<br />
detail in Section 6.<br />
A singular quartic surface which is a counterexample to the Hasse principle has<br />
been given by V. A. Iskovskikh [Is].<br />
A method to explain all these examples in a unified manner was provided by<br />
Yu. I. Manin in his book on cubic forms [Man]. This is what is nowadays called<br />
the Brauer-Manin obstruction.<br />
2.2. Definition. –––– Let K be a number field <strong>and</strong> X be a scheme separated <strong>and</strong><br />
of finite type over K.<br />
i) Then, for each ν ∈ Val(K ), there is the local evaluation map, given by<br />
ev ν : Br(X ) × X (K ν ) −→É/,<br />
(α, x) ↦→ inv ν (α| x ) .<br />
ii) Further, there is the global evaluation map or Manin map<br />
ev: Br(X ) × X (K) −→É/,<br />
( )<br />
α, (xν ) ν ↦→ ∑ev ν (α, x ν ) .<br />
ν
104 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
2.3. Proposition. –––– Let K be a number field <strong>and</strong> π : X → SpecK be a scheme<br />
which is separated <strong>and</strong> of finite type over K.<br />
Then,<br />
a) The local evaluation map is<br />
i) additive in the first variable <strong>and</strong><br />
ii) continuous in the second variable.<br />
iii) If α = π ∗ a for a ∈ Br(SpecK ) then ev ν (α, ·) = inv ν (a) is constant.<br />
iv) For each α ∈ Br(X ), there exists a finite set S of valuations including all archimedean<br />
ones such that<br />
ev ν (α, x) = 0<br />
whenever ν ∉ S <strong>and</strong> x ∈ X (O Kν ).<br />
b) The Manin map is well-defined. Further, it is<br />
i) additive in the first variable <strong>and</strong><br />
ii) continuous in the second variable.<br />
iii) In the first variable, ev factors via Br(X )/π ∗ Br(SpecK ).<br />
iv) Further,<br />
if x ∈ X (K ) ⊆ X (K).<br />
Proof. a) i) <strong>and</strong> iii) are obvious.<br />
ev(α, x) = 0<br />
ii) Let α ∈ Br(X ) <strong>and</strong> x ∈ X (K ν ). We have to show that ev ν (α, ·) is constant in<br />
a neighbourhood of x in the ν-adic topology.<br />
By iii), we may assume that ev ν (α, x) = 0. Then, there is an Azumaya algebra A<br />
over X such that [A ] = α <strong>and</strong> A | x<br />
∼ = Mn (K ν ). By Corollary II.5.4, this implies<br />
A | SpecO h<br />
XKν ,x<br />
∼ = Mn (O h X Kν ,x)<br />
for the Henselization of the local ring at x ∈ X Kν . To define this isomorphism,<br />
only finitely many data are required. Thus, there exist an étale morphism<br />
f : X ′ → X <strong>and</strong> a K ν -valued point x ′ ∈ X ′ such that f (x ′ ) = x<br />
<strong>and</strong> A | X ′<br />
∼ = Mn (O X ′).<br />
f induces a local isomorphism of ν-adic analytic spaces.<br />
A | y<br />
∼ = Mn (K ν ) for y in a ν-adic neighbourhood of x.<br />
In particular,<br />
iv) By Lemma 3.4.a), there exists a model X of X which is separated <strong>and</strong> of<br />
finite type over O K . We may assume that X is reduced.<br />
Let A be an Azumaya algebra over X such that [A ] = α. By Lemma 3.5,<br />
there is an open subset X ◦ ⊆ X such that A may be extended to an Azumaya<br />
algebra à over X ◦ .
Sec. 2] THE BRAUER-MANIN OBSTRUCTION 105<br />
The closed subset X \X ◦ ⊂ X does not meet the generic fiber. Since X is of<br />
finite type, X \X ◦ is contained in finitely many special fibers. We put S to be<br />
the set of the corresponding valuations together with all archimedean valuations.<br />
We have to prove that ev ν (α, x) = 0 for ν ∉ S <strong>and</strong> x ∈ X (K ν ). The point x<br />
induces a morphism x : Spec O Kν → X of O K -schemes. The assumption ν ∉ S<br />
implies that x actually maps to X ◦ .<br />
Over X ◦ , there is the Azumaya algebra à extending A . The restriction<br />
of [A ] ∈ Br(X ) along x : SpecK ν → X coincides with the restriction<br />
of [Ã ] ∈ Br(X ◦ x<br />
) along SpecK ν −→ X −→ ⊂<br />
X . The latter factorizes<br />
via Br(Spec O Kν ). We note, finally, that Br(Spec O Kν ) = 0 by virtue of Example<br />
8.1.ii).<br />
b) In order to show that ev is well-defined, one has to verify that the sum is<br />
always finite. This, in turn, immediately follows from a.iv).<br />
i) <strong>and</strong> iii) are direct consequences from a.i) <strong>and</strong> a.iii).<br />
ii) This follows from a.ii) together with a.iv).<br />
iv) is a consequence of the description of Br(K ) given in Example II.8.3.i).<br />
2.4. Remark. –––– Let x ∈ X (K) be an adelic point. If there exists a Brauer<br />
class α ∈ Br(X ) such that ev(α, x) ≠ 0 then x can not be approximated by a<br />
sequence of K-valued <strong>points</strong>. The Brauer class α therefore “obstructs” the adelic<br />
point x from being approximated by <strong>rational</strong> <strong>points</strong>. This justifies the name<br />
Brauer-Manin obstruction.<br />
2.5. Notation. –––– Let K be a number field <strong>and</strong> π : X → SpecK be a scheme<br />
separated <strong>and</strong> of finite type over K.<br />
For α ∈ Br(X ), we write<br />
We define<br />
X (K) α := {x ∈ X (K) | ev(α, x) = 0} .<br />
X (K ) Br := \<br />
X (K ) α .<br />
α∈Br(X )<br />
2.6. Remarks. –––– a) According to Proposition 2.3.b.ii), X (K) Br is always a<br />
closed subset of X (K).<br />
b) Proposition 2.3.b.iv) shows X (K ) ⊆ X (K) Br .<br />
2.7. Remark. –––– As shown in Corollary II.8.12.ii), there are many particular<br />
cases in which Br(X )/π ∗ Br(K ) is actually a finite group.<br />
In this case, one might choose a finite system α 1 , . . . , α n ∈ Br(X ) generating<br />
Br(X )/π ∗ Br(SpecK ). This leads to X (K) Br = X (K) α 1 ∩ . . . ∩ X (K) α n<br />
.<br />
□
106 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
By Proposition 2.3.a.iv), only finitely many places of K are involved in the<br />
evaluation of α 1 , . . . , α n .<br />
2.8. Definition. –––– Let K be a number field <strong>and</strong> π : X → SpecK be a scheme<br />
separated <strong>and</strong> of finite type over K.<br />
a) If X (K) Br ≠ X (K) then one says that, on X, there is a Brauer-Manin<br />
obstruction to weak approximation.<br />
b) If X (K) Br = ∅ <strong>and</strong> X (K) ≠ ∅ then one says that, on X, there is a<br />
Brauer-Manin obstruction to <strong>rational</strong> <strong>points</strong> on X or that there is a Brauer-Manin<br />
obstruction to the Hasse principle on X.<br />
2.9. Remark. –––– In the case that X (K ) Br ≠ ∅, many authors use the somewhat<br />
confusing formulation that the Brauer-Manin obstruction would be empty.<br />
2.10. Remark. –––– There is clearly no Brauer-Manin obstruction when<br />
Br(X )/π ∗ Br(SpecK ) = 0, neither to the Hasse principle nor to weak approximation.<br />
We know several such cases from the computations described above. For example,<br />
as indicated in II.8.23, we have Br(X )/π ∗ Br(SpecK ) = 0 for X a general<br />
cubic surface or a smooth cubic surface such that all 27 lines are defined over K.<br />
By Fact II.8.14, we have Br(X )/π ∗ Br(SpecK ) = 0 when X is a smooth complete<br />
intersection of dimension ≥ 3 in P n K .<br />
2.11. Definition. –––– If the statement<br />
X (K ) Br ≠ ∅ =⇒ X (K ) ≠ ∅<br />
is true for a certain class of separated K-schemes of finite type then one says that<br />
the Brauer-Manin obstruction is the only obstruction to the Hasse principle for<br />
that class.<br />
2.12. Conjecture (Colliot-Thélène, cf. [CT/S, Conjecture C]). —–<br />
The Brauer-Manin obstruction is the only obstruction to the Hasse principle for<br />
smooth cubic surfaces.<br />
3. Technical lemmata<br />
3.1. –––– Let (K i ) i∈Æbe a sequence of fields. Then, there is the canonical<br />
morphism of schemes<br />
G<br />
ι:<br />
i −→ Spec<br />
i∈ÆSpecK ( )<br />
∏ i .<br />
i∈ÆK
Sec. 3] TECHNICAL LEMMATA 107<br />
3.2. Lemma. –––– LetX beaschemewhichisseparated<strong>and</strong>quasi-compact. Then,ι<br />
induces a bijection<br />
X ( ∏<br />
i∈ÆK i<br />
) −→ ∏<br />
i∈ÆX(K i ) .<br />
Proof. We first note that the assertion is true when X = SpecA is an<br />
affine scheme. Indeed, to give a ring homomorphism A → ∏ i∈ÆK i is the<br />
same as giving a system of ring homomorphisms (A → K i ) i∈Æ.<br />
Surjectivity. We cover X by finitely many affine schemes X j<br />
∼ = SpecAj ,<br />
j ∈ {1, . . . , n}. Given a system of morphisms (SpecK i → X ) i∈Æ, we may find<br />
a decomposition<br />
Æ=S 1 ∪ . . . ∪ S n<br />
into mutually disjoint subsets such that SpecK i maps to X j if i ∈ S j .<br />
Now we use the assumption that the X j are affine. For j fixed, the morphisms<br />
SpecK i → X j for i ∈ S j give rise to a morphism<br />
Finally, observe that<br />
nG<br />
j=1<br />
since the index set is finite.<br />
Spec ( ∏<br />
i∈S j<br />
K i<br />
) → Xj ⊆ X.<br />
Spec ( ∏<br />
i∈S j<br />
K i<br />
) ∼= Spec<br />
(<br />
∏<br />
i∈ÆK i<br />
)<br />
Injectivity. Suppose two morphisms g 1 , g 2 : Spec(∏ i∈ÆK i ) → X induce the<br />
same morphism on F i∈ÆSpecK i . Proposition (5.2.5) of [EGA, Chapitre I] implies<br />
that g 1 <strong>and</strong> g 2 coincide on the Zariski closure of F i∈ÆSpecK i .<br />
We claim that F i∈ÆSpecK i is actually dense in Spec(∏ i∈ÆK i ). Indeed, a base of<br />
the open subsets is provided by the sets D( f ) for<br />
0 ≠ f = ( . . . , f ν , . . . ) ∈ ∏<br />
i∈ÆK i .<br />
If f ν ≠ 0 then SpecK ν is contained in D( f ). The claim follows.<br />
□<br />
3.3. Remark. –––– The assertion of the lemma is certainly not true in general<br />
for X an arbitrary scheme. Indeed, Yoneda’s lemma would then imply<br />
that ι: F i∈ÆSpecK i → Spec(∏ i∈ÆK i ) is an isomorphism.<br />
ι induces, however, not even a bijection on closed <strong>points</strong>. In fact, there is<br />
the ideal ⊕ i∈ÆK i ⊂ ∏ i∈ÆK i . According to the lemma of Zorn, this ideal is<br />
contained in a maximal ideal which is different from those in F i∈ÆSpecK i .
108 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
3.4. Lemma. –––– Let K be a number field <strong>and</strong> X be a scheme which is separated<br />
<strong>and</strong> of finite type over K.<br />
a) Then, there exists a model X of X. This means that X is a scheme which is<br />
separated <strong>and</strong> of finite type over O K <strong>and</strong> fulfills X × SpecOK SpecK ∼ = X.<br />
b) If X is proper over K then X is proper over Spec(O K ) f for some 0 ≠ f ∈ O K .<br />
Proof. First step. Existence of a model.<br />
X is given by gluing together finitely many affine K-schemes U i = SpecR i .<br />
The intersections U i ∩ U j are affine. Thus, gluing takes place along isomorphisms<br />
ϕ ij : Spec(R i ) gij −→ Spec(R j ) gji .<br />
The K-algebras may be described by generators <strong>and</strong> relations in the form<br />
R i = K [T i1 , . . . , T iji ]/(h i1 , . . . , h iki ). In the polynomials h ik , g ij , only finitely<br />
many denominators are occurring. We, therefore, have R i = R ′ i⊗ OK K for some<br />
finitely generated O K -algebras R ′ i such that g ji ∈ R ′ i .<br />
The ϕ ij induce isomorphisms ϕ # ij : (R′ j ) g ji<br />
⊗ OK K → (R i ′) g ij<br />
⊗ OK K. For some suitable<br />
0 ≠ f ij ∈ O K , the isomorphism ϕ # ij<br />
is the base extension of an isomorphism<br />
(R ′ j ) g ji<br />
⊗ OK (O K ) fij → (R i ′) g ij<br />
⊗ OK (O K ) fij .<br />
Putting R i := (R i) ′ f for f := ∏ f ij , the ϕ ij therefore extend to isomorphisms<br />
i,j<br />
˜ϕ ij : Spec(R i ) gij −→ Spec(R j ) gji .<br />
The cocycle relations “˜ϕ ik = ˜ϕ jk ◦ ˜ϕ ij ” are satisfied after restriction to the<br />
generic fiber. Since the schemes Spec(R i ) gij are of finite type, this implies<br />
that these relations are still fulfilled when the R i are localized only by some<br />
suitable 0 ≠ f ′ ∈ O K .<br />
We constructed gluing data for an O K -scheme X . X is clearly of finite type.<br />
Second step. Separatedness.<br />
Denote by ∆ ⊂ X × SpecOK X the Zariski closure of the diagonal. Then, the<br />
diagonal morphism δ : X → ∆ is an isomorphism on the generic fiber. Since X<br />
<strong>and</strong> ∆ are O K -schemes of finite type, δ is an isomorphism outside finitely many<br />
special fibers. We delete the corresponding special fibers from X .<br />
Third step. Properness.<br />
Assume first that X is reduced. Then, by Chow’s lemma, we have a surjective<br />
map X ′ → X from a projective K-scheme X ′ ⊆ P n K . Let X ′ be the Zariski<br />
closure of X ′ in P n O K<br />
equipped with the induced reduced structure. This is a<br />
model of X ′ which is proper over Spec O K .
Sec. 3] TECHNICAL LEMMATA 109<br />
Let X be any model of X in the sense of the previous steps. Then, there<br />
exists some 0 ≠ f ∈ O K such that the morphism X ′ → X extends to a morphism<br />
of (O K ) f -schemes g : X ′ × SpecOK Spec(O K ) f → X . [EGA, Chapitre II,<br />
Corollaire (5.4.3.ii)] shows that im g is proper over Spec(O K ) f . This implies that<br />
im g ⊆ X × SpecOK Spec(O K ) f is a closed subscheme containing the generic fiber.<br />
In particular, the closure ofX in X × SpecOK Spec(O K ) f is proper over Spec(O K ) f .<br />
Return to the general case. Let X be any model of X in the sense of the<br />
two steps above. We may assume that X is equal to the closure of X.<br />
Then, there exists some 0 ≠ f ∈ O K such that X red × SpecOK Spec(O K ) f is<br />
proper over Spec(O K ) f . By virtue of [EGA, Chapitre II, Corollaire (5.4.6)], this<br />
suffices for X × SpecOK Spec(O K ) f being proper.<br />
□<br />
3.5. Lemma. –––– Let K be a number field, X a reduced scheme which is separated<br />
<strong>and</strong> of finite type over the integer ring O K , <strong>and</strong> A be an Azumaya algebra over the<br />
generic fiber X.<br />
Then, there exist an open subset X ◦ ⊆ X containing X <strong>and</strong> an Azumaya algebra<br />
à over X ◦ which extends A .<br />
Proof. First step. Extending A to a coherent sheaf.<br />
Let j : X → X be the embedding of the generic fiber. Then, j ∗ A is a quasicoherent<br />
sheaf on X . We claim, j ∗ A contains a coherent subsheaf F such that<br />
F | X = A .<br />
Assume that would not be the case. We construct an increasing sequence of<br />
subsheaves of j ∗ A recursively as follows.<br />
Put F 1 := 〈1〉 for the section 1 ∈ Γ(X , j ∗ A ). Having F n already constructed,<br />
we have, by our assumption, F n | X A . We observe that X is a Noetherian<br />
scheme. Therefore, j ∗ A is the union of its coherent subsheaves [Ha77,<br />
Chap. II, Exercise 5.15.a)]. By consequence, there exists a coherent sheaf<br />
G ⊆ j ∗ A such that G | X ⊈ F n | X . We put F n+1 := F n + G .<br />
This means, (F n | X ) n∈Æis a strictly increasing sequence of coherent subsheaves<br />
of A . This is a contradiction.<br />
Second step. Extending A to a locally free sheaf.<br />
We may assume X to be connected. Then,<br />
ψ(x) := dim k(x) F X ⊗ OX k(x)<br />
is constant on the generic fiber X. As ψ is upper semicontinuous, there is<br />
an open subset X ′ ⊆ X containing the generic fiber such that ψ is constant<br />
on X ′ . By reducedness, this suffices for F | X ′ being locally free.<br />
Third step. Extending A to an Azumaya algebra.<br />
j ∗ A carries a natural structure of a sheaf of O X -algebras.
110 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
F ⊂ j ∗ A contains the constant section 1 by the construction given in the<br />
first step. To obtain a sheaf of algebras, we must establish closedness under multiplication.<br />
The image F ′ of the multiplication map F ×F → j ∗ A is a coherent subsheaf.<br />
On the generic fiber, we have F ′ | X ⊆ F | X . This implies Q := (F ′ + F )/F<br />
is a coherent sheaf on X such that Q| X = 0. Upper semicontinuity shows that<br />
Q vanishes on an open neighbourhood X ′′ of the generic fiber. F | X ′ ∩X ′′ is a<br />
locally free sheaf of O X ′ ∩X ′′-algebras.<br />
By Corollary 1.5, the non-Azumaya locus is again a closed subset.<br />
□<br />
4. Computing the Brauer-Manin obstruction – The general strategy<br />
4.1. –––– In this section, we want to illustrate that the Manin pairing ev(α, x)<br />
is effectively computable in certain cases.<br />
We will assume that X is a geometrically integral scheme which is proper over<br />
a number field K. Further, we suppose that Br ′ (X K<br />
) = 0 <strong>and</strong> Br(X ) = Br ′ (X ).<br />
Under these assumptions, Br(X )/π ∗ Br(SpecK ) ∼ = H 1( Gal(K/K ), Pic(X K<br />
) ) .<br />
These assumptions are fulfilled, e.g., for X a smooth proper surface such that<br />
X K<br />
is a (not necessarily minimal) <strong>rational</strong> surface.<br />
We may also suppose X (K) ≠ ∅ as, otherwise, the Brauer-Manin obstruction<br />
would not be of much interest.<br />
4.2. –––– Usually, α ∈ Br(X ) is not given itself, but only its image<br />
α ∈ Br(X )/π ∗ Br(SpecK ) ∼ = H 1( Gal(K/K ), Pic(X K<br />
) ) .<br />
In order to compute ev(α, x) = ∑ ν ev ν (α, x ν ), it is therefore necessary to explicitly<br />
lift α to Br(X ). Note that the ev ν , as opposed to ev itself, do not factor<br />
via Br(X )/π ∗ Br(SpecK).<br />
4.3. Lemma. –––– LetK beanumberfield<strong>and</strong>π: X → SpecK beageometrically<br />
integral scheme which is proper over K.<br />
Then, there is a natural isomorphism<br />
ker(Br ′ (X ) → i′<br />
Br ′ (X K<br />
))/π ∗ ∼=<br />
Br(SpecK ) −→ H 1( Gal(K/K ), Pic(X K<br />
) )
Sec. 4] COMPUTING THE BRAUER-MANIN OBSTRUCTION 111<br />
induced by the commutative diagram<br />
Br(SpecK )<br />
Br(SpecK )<br />
π ∗<br />
0 ker(Br ′ (X ) → i′<br />
Br ′ (X K )) H 2( Gal(K/K ),Q(X K ) ∗) H 2( Gal(K/K ),Div(X K ) )<br />
0 H 1( Gal(K/K ),Pic(X K ) ) H 2( Gal(K/K ),Q(X K ) ∗ /K ∗) H 2( Gal(K/K ),Div(X K ) ) .<br />
Here, the bottom row is part of the long exact sequence of cohomology associated to<br />
the exact sequence<br />
0 −→ Q(X K<br />
) ∗ /K ∗ −→ Div(X K<br />
) −→ Pic(X K<br />
) −→ 0 .<br />
The middle column is part of the long exact cohomology sequence associated to the<br />
exact sequence<br />
0 −→ K ∗ −→ Q(X K<br />
) ∗ −→ Q(X K<br />
) ∗ /K ∗ −→ 0 .<br />
Finally, the middle row is obtained when mapping the short exact sequence from<br />
Proposition II.4.8 to<br />
0 −→ H 2( Gal(Q(X K<br />
)/K ), Q(X K<br />
) ∗) −→ H 2( Gal(Q(X K<br />
)/K ), Q(X K<br />
) ∗) −→ 0<br />
0<br />
<strong>and</strong> taking kernels.<br />
□<br />
4.4. Remark. –––– The isomorphism given in Lemma 4.3 is the same as the<br />
isomorphism provided by the Hochschild-Serre spectral sequence (Proposition<br />
II.8.11). This is asserted in [Lic, Sec. 2].<br />
4.5. –––– The diagram above leads to the following general strategy.<br />
General strategy for computing ev(α, x).<br />
a) Compute the image of α in H 2( Gal(K/K ), Q(X K<br />
) ∗ /K ∗) .<br />
b) Lift that to a cohomology class in H 2( Gal(K/K ), Q(X K<br />
) ∗) .<br />
c) For each ν,<br />
i) restrict to H 2( Gal(K ν /K ν ), Q(X Kν<br />
) ∗) .<br />
ii) In a neighbourhood U of x ν such that Pic(U Kν<br />
) = 0, use the exact sequence<br />
H 2( Gal(K ν /K ν ), Γ(U Kν<br />
, O ∗ U Kν<br />
) ) −→ H 2( Gal(K ν /K ν ), Q(X Kν<br />
) ∗) −→<br />
−→ H 2( Gal(K ν /K ), Div(U Kν<br />
) )
112 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
in order to lift to H 2( Gal(K ν /K ν ), Γ(U Kν<br />
, O ∗ U Kν<br />
) ) .<br />
iii) Apply the evaluation map Γ(U Kν<br />
, O ∗ U Kν<br />
) → K ∗ ν at x ν to get a cohomology<br />
class in H 2( Gal(K ν /K ν ), K ∗ ν)<br />
.<br />
iv) Take its invariant inÉ/.<br />
d) Take the sum of all these invariants.<br />
4.6. Remark. –––– In practice, there are several problems with that strategy.<br />
First of all, our skills in dealing with general Galois cohomology<br />
classes are limited. Anyway, for some suitable finite field extension L/K,<br />
α ∈ H 1( Gal(K/K ), Pic(X K<br />
) ) is the image under inflation of a class in<br />
H 1( Gal(L/K ), Pic(X L ) ) .<br />
4.7. Observations (Inflation from a finite quotient Gal(L/K )). —–<br />
a) There is an exact sequence<br />
0 −→ Q(X L ) ∗ /L ∗ −→ Div(X L ) −→ Pic(X L ) −→ 0<br />
inducing a map H 1( Gal(L/K ), Pic(X L ) ) → H 2( Gal(L/K ), Q(X L ) ∗ /L ∗) .<br />
b) ThehomomorphismH 2( Gal(L/K ), Q(X L ) ∗) → H 2( Gal(L/K ), Q(X L ) ∗ /L ∗)<br />
is not surjective, in general. It is, however, when Gal(L/K ) is cyclic. Indeed, in<br />
this case there is a non-canonical isomorphism<br />
H 3( Gal(L/K ), L ∗) ∼ = H<br />
1 ( Gal(L/K ), L ∗)<br />
<strong>and</strong> the latter group vanishes by Hilbert’s Theorem 90.<br />
c) The restriction in i) goes to H 2( Gal(L w /K ν ), Q(X Lw ) ∗) for w a prime above ν.<br />
In step ii), one can work with U = Spec O XLw ,x ν<br />
. There is the exact sequence<br />
H 2( Gal(L w /K ν ), Γ(U Lw , O ∗ U Lw<br />
) ) −→ H 2( Gal(L w /K ν ), Q(X Lw ) ∗) −→<br />
−→ H 2( Gal(L w /K ), Div(U Lw ) )<br />
such that one may lift to H 2( Gal(L w /K ν ), Γ(U Lw , O ∗ U Lw<br />
) ) .<br />
4.8. Remark. –––– To summarize, assume Gal(L/K ) is cyclic <strong>and</strong> we start with<br />
a class in the image under inflation of H 1( Gal(L/K ), Pic(X L ) ) . Then, all the<br />
cohomology classes obtained according to the general strategy are in the image<br />
of H 2( Gal(L/K ), ·) or H 2( Gal(L w /K ν ), ·),<br />
respectively.<br />
4.9. –––– Thus, we assume that G := Gal(L/K ) is a cyclic group of order n.<br />
In this case, H 2 (G,) ∼ =/n<strong>and</strong> the cup product with a generator<br />
induces an isomorphism Ĥ q (G, A) → Ĥ q+2 (G, A) for all integers q <strong>and</strong>
Sec. 5] THE EXAMPLES OF MORDELL 113<br />
all G-modules. We fix a generator of H 2 (G,) in order to determine these<br />
isomorphisms uniquely.<br />
Note that there is no distinguished generator of H 2 (G,) unless a generator<br />
of G is fixed. Thus, the isomorphisms discussed above are not canonical.<br />
Nevertheless, there is a commutative diagram, analogous to the one in<br />
Lemma 4.3, with H 2 replaced by Ĥ 0 <strong>and</strong> H 1 replaced by Ĥ −1 .<br />
4.10. –––– We therefore have the following plan.<br />
Plan for computing ev(α, x).<br />
One has<br />
Ĥ −1 (G, Pic(X L )) ∼ = [Div 0 (X L ) G ∩ N Div(X L )]/N Div 0 (X L ) .<br />
The map from step a) becomes the canonical map to Div 0 (X L ) G /N Div 0 (X L ).<br />
Step b) is the lift to Q(X L ) G /NQ(X L ) = Q(X K )/NQ(X L ) under f ↦→ div f .<br />
For each ν, steps c.i), ii) <strong>and</strong> iii) amount to the evaluation<br />
Q(X K )/NQ(X L ) −→ K ∗ ν/NL ∗ w ,<br />
f ↦→ [ f (x ν )]<br />
at x ν . Note that this map is well-defined although a representative in Q(X K )<br />
might have a pole at x ν .<br />
Finally, in step c.iv), one has to apply the chosen isomorphism backwards,<br />
K ∗ ν/NL ∗ w = Ĥ 0 (G, L ∗ w ) ∼ =<br />
−→ H 2 (G, L ∗ w ) inv ν<br />
−→É/.<br />
ev(α, x) is the sum of all these local invariants.<br />
5. The examples of Mordell<br />
i. Formulation of the results. —<br />
5.1. –––– In this section, we present a series of examples of cubic surfaces<br />
overÉfor which the Hasse principle fails. Our series generalizes the examples<br />
of Mordell [Mord]. It was observed by Yu. I. Manin [Man,кVI,º½¹º]<br />
himself that the failure of the Hasse principle in Mordell’s examples may be<br />
explained by the Brauer-Manin obstruction.
114 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
5.2. –––– Let p 0 ≡ 1 (mod 3) be a prime number <strong>and</strong> K/Ébe the unique<br />
cubic field extension contained in the cyclotomic extensionÉ(ζ p0 )/É. We fix<br />
the explicit generator θ ∈ K given by<br />
More concretely, we write<br />
where n is given by p 0 = 6n + 1.<br />
θ := trÉ(ζ p0 )/K (ζ p0 − 1) .<br />
θ = −2n + ∑<br />
i∈(∗ p0 ) 3 ζ i p 0<br />
5.3. Proposition. –––– Consider the cubic surface X ⊂ P 3É, given by<br />
T 3 (a 1 T 0 + d 1 T 3 )(a 2 T 0 + d 2 T 3 ) =<br />
3<br />
∏<br />
i=1<br />
(<br />
T0 + θ (i) T 1 + (θ (i) ) 2 T 2<br />
)<br />
.<br />
Here, a 1 , a 2 , d 1 , d 2 ∈. The θ (i) are the images of θ under Gal(K/É).<br />
i) Let A ≡ 1 (mod 3) be the unique integer such that 4p 0 = A 2 + 27B 2 .<br />
Suppose that d 1 d 2 ≠ 0, a 1 ≠ 0, a 2 ≠ 0, a 1 /d 1 ≠ a 2 /d 2 , <strong>and</strong> that<br />
T (a 1 + d 1 T )(a 2 + d 2 T ) + 4p 0−A 2<br />
(p 2 0 −3p 0−A) 2 = 0<br />
has no multiple zeroes. Then, X is smooth.<br />
ii) Assume that p 0 ∤ d 1 d 2 , that gcd(a 1 , d 1 ) <strong>and</strong> gcd(a 2 , d 2 ) contain only prime factors<br />
which decompose in K, <strong>and</strong> that T (a 1 + d 1 T )(a 2 + d 2 T ) − 1 = 0 has at least one<br />
simple zero inp 0<br />
. Then, X (É) ≠ ∅.<br />
iii) Assume that X is smooth. Suppose further that p 0 ∤ d 1 d 2 <strong>and</strong> gcd(d 1 , d 2 ) = 1.<br />
Then, there is a class α ∈ Br(X ) with the following property.<br />
For an adelic point x = (x ν ) ν , the value of ev(α, x) depends only on the component<br />
x νp0 . Write x νp0 =: (t 0 : t 1 : t 2 : t 3 ). Then, one has ev(α, x) = 0 if <strong>and</strong><br />
only if<br />
a 1 t 0 + d 1 t 3<br />
t 3<br />
is a cube in∗ p0<br />
.<br />
5.4. Remarks. –––– i) K/Éis an abelian cubic field extension. It is totally<br />
ramified at p 0 <strong>and</strong> unramified at all other primes. A prime p is completely<br />
decomposed in K if <strong>and</strong> only if p is a cube modulo p 0 .<br />
ii) We have ∏<br />
3 ( )<br />
T0 + θ (i) T 1 + (θ (i) ) 2 T 2 = NK/É(T 0 + θT 1 + θ 2 T 2 ) .<br />
i=1<br />
iii) The reduction of T (a 1 + d 1 T )(a 2 + d 2 T ) + 4p 0−A 2<br />
modulo p<br />
(p0 2−3p 0−A) 2 0 is exactly<br />
T (a 1 + d 1 T )(a 2 + d 2 T ) − 1.
Sec. 5] THE EXAMPLES OF MORDELL 115<br />
5.5. Remark. –––– Let (t 0 : t 1 : t 2 : t 3 ) ∈ X (Ép 0<br />
). Then, for the reduction,<br />
one has (t 0 : t 3 ) = (1 : s) for s a solution of T (a 1 + d 1 T )(a 2 + d 2 T ) − 1 = 0.<br />
On the other h<strong>and</strong>, if s is a simple solution then, by Hensel’s lemma, there<br />
exists (t 0 : t 1 : t 2 : t 3 ) ∈ X (Ép 0<br />
) such that (t 0 : t 3 ) = (1 : s).<br />
The possible values of s are in bijection with the up to three planes a p 0 -adic<br />
point on X may reduce to.<br />
We will show in Fact 5.19 that noÉp 0<br />
-valued point on X reduces to the triple line<br />
“T 0 = T 3 = 0”.<br />
5.6. Observation. –––– As<br />
a 1 t 0 + d 1 t 3<br />
t 3<br />
= a 1 + d 1 s<br />
s<br />
the value of ev(α, x) depends only on the plane to which its component x νp0 is<br />
mapped under reduction.<br />
5.7. Remark. –––– Since s is a solution of T (a 1 + d 1 T )(a 2 + d 2 T ) − 1 = 0, we<br />
have s ≠ 0 <strong>and</strong> a 1 + d 1 s ≠ 0. Hence, a 1+d 1 s<br />
s<br />
is well defined <strong>and</strong> non-zero inp 0<br />
.<br />
ii. Smoothness. —<br />
5.8. Lemma. –––– The minimal polynomial of θ is<br />
T 3 + p 0 T 2 p 0 − 1<br />
+ p 0 T + p3 0 − 3p0 2 − Ap 0<br />
.<br />
3 27<br />
5.9. Remark. –––– This is an Eisenstein polynomial. We see once more that<br />
K/Éis totally ramified at p 0 . Further, ν(θ) = 1 3 for ν the extension of ν p 0<br />
to K.<br />
5.10. Proof of the lemma. ––––∗ p0<br />
is decomposed into three cosets C 1 , C 2 ,<br />
<strong>and</strong> C 3 when factored by (∗ p0<br />
) 3 . We have<br />
θ (i) = −2n + ∑<br />
j∈C i<br />
ζ j p 0<br />
.<br />
As the sum over all p 0 -th roots of unity vanishes, this immediately implies<br />
θ (1) + θ (2) + θ (3) = −p 0 .<br />
Further, multiplying terms <strong>and</strong> adding-up shows that<br />
θ (1) θ (2) + θ (2) θ (3) + θ (3) θ (1) = 12n 2 + 2(−2n) ∑ ζ j p 0<br />
+ p 0 − 1<br />
j∈∗ p0<br />
3<br />
∑<br />
[ ] 2 p0 − 1<br />
= 12 + 2 p 0 − 1<br />
− p 0 − 1<br />
6 3 3<br />
= p 0 · p0 − 1<br />
.<br />
3<br />
,<br />
j∈∗ p0<br />
ζ j p 0
116 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
In order to establish θ (1) θ (2) θ (3) = −p3 0 +3p2 0 +Ap 0<br />
27<br />
, multiplying expressions leads to<br />
θ (1) θ (2) θ (3) = −8n 3 + 4n 2 ∑ ζ j p 0<br />
− 2n p 0 − 1<br />
j∈∗ p0<br />
3<br />
∑ ζ j p 0<br />
+<br />
j∈∗ p0<br />
( )(<br />
+ ∑ ζ j p 0<br />
j∈C 1<br />
( )( )( )<br />
= −8n 3 + ∑ ζ j p 0 ∑ ζ j p 0 ∑ ζ j p 0<br />
.<br />
j∈C 1 j∈C 2 j∈C 3<br />
)(<br />
∑ ζ j p 0<br />
j∈C 2<br />
)<br />
∑ ζ j p 0<br />
j∈C 3<br />
To calculate the remaining triple product is more troublesome as one needs to<br />
know how often a product ζ j 1<br />
p 0<br />
ζ j 2<br />
p 0<br />
ζ j 3<br />
p 0<br />
for j i ∈ C i is equal to one.<br />
Sublemma 5.11 shows that this happens exactly p−1 · p0+1+A times. As the<br />
3 9<br />
expression is invariant under Gal(É(ζ p0 )/É) <strong>and</strong> there are all in all (p−1)3 summ<strong>and</strong>s,<br />
we find<br />
27<br />
(<br />
)(<br />
∑ ζ j p 0<br />
j∈C 1<br />
)(<br />
∑ ζ j p 0<br />
j∈C 2<br />
)<br />
∑ ζ j p 0<br />
= p 0 − 1<br />
(p 0 + 1 + A)<br />
j∈C 3<br />
27<br />
+ 1<br />
27 [(p 0 − 1) 2 − (p 0 + 1 + A)] ∑<br />
j∈∗ p0<br />
ζ j p 0<br />
= 3p 0 − 1 + Ap 0<br />
.<br />
27<br />
Further, −8n 3 = −8( p 0−1<br />
) 3 = −p3 0 +3p2 0 −3p 0+1<br />
. The assertion follows. □<br />
6 27<br />
5.11. Sublemma. –––– Let D be a non-cube in∗ p0<br />
. Then,<br />
#{(x, y) ∈2 p0<br />
| Dx 3 + D 2 y 3 = 1} = p 0 + 1 + A .<br />
Proof. The number of solutions inp 0<br />
of such an equation may be counted<br />
using Jacobi sums. As in [I/R, Chapter 8, §3], we see<br />
#{(x, y) | Dx 3 + D 2 y 3 = 1} =<br />
where χ is a fixed cubic character.<br />
=<br />
2 2<br />
∑ ∑ ∑<br />
i=0 j=0 a+b=1<br />
2<br />
∑<br />
i=0<br />
χ i (a/D)χ j (b/D 2 )<br />
2<br />
∑ χ i (1/D)χ j (1/D 2 )J (χ i , χ j ) .<br />
j=0<br />
The summ<strong>and</strong>s for i = j are the same as in the case D = 1. If i = 0 <strong>and</strong> j ≠ 0<br />
or vice versa then the corresponding summ<strong>and</strong> is 0. Finally, by [I/R, Chapter 8,<br />
§3, Theorem 1.c)], we have J (χ, χ 2 ) = J (χ 2 , χ) = −1. Altogether,<br />
#{(x, y) | Dx 3 + D 2 y 3 = 1} = #{(x, y) | x 3 + y 3 = 1} + 3 .
Sec. 5] THE EXAMPLES OF MORDELL 117<br />
The claim now follows from a theorem of C. F. Gauß [I/R, Chapter 8, §3,<br />
Theorem 2)].<br />
□<br />
5.12. Notation. –––– We will write<br />
F (T 0 , T 1 , T 2 , T 3 ) := T 3 (a 1 T 0 +d 1 T 3 )(a 2 T 0 +d 2 T 3 )−<br />
3<br />
∏<br />
i=1<br />
(<br />
T0 +θ (i) T 1 +(θ (i) ) 2 T 2<br />
)<br />
.<br />
5.13. Proof of Proposition 5.3.i) –––– Since d 1 d 2 ≠ 0, the projection to the first<br />
three coordinates induces a morphism of K-schemes q : X → P 2Éwhich is finite<br />
of degree three.<br />
The conditions ∂F<br />
∂T 1<br />
= 0 <strong>and</strong> ∂F<br />
∂T 2<br />
= 0 for a singular point x depend only on q(x).<br />
Let us first analyze them.<br />
Writing l i := T 0 + θ (i) T 1 + (θ (i) ) 2 T 2 , we get the system of equations<br />
the solution of which is<br />
θ (1) l 2 l 3 + θ (2) l 3 l 1 + θ (3) l 1 l 2 = 0 ,<br />
θ (1)2 l 2 l 3 + θ (2)2 l 3 l 1 + θ (3)2 l 1 l 2 = 0 ,<br />
(l 2 l 3 : l 3 l 1 : l 1 l 2 ) = (θ (2) θ (3) (θ (2) − θ (3) ) : θ (3) θ (1) (θ (3) − θ (1) ) : θ (1) θ (2) (θ (1) − θ (2) )) .<br />
As ̟ : (T 0 : T 1 : T 2 ) → (l 2 l 3 : l 3 l 1 : l 1 l 2 ) is a quadratic transformation, there are<br />
exactly four <strong>points</strong> in P 2 which fulfill that condition. These are the three <strong>points</strong><br />
of indeterminacy of ̟ which we denote by P 1 , P 2 , <strong>and</strong> P 3 <strong>and</strong> a non-trivial<br />
solution P.<br />
P 1 , P 2 , <strong>and</strong> P 3 are given by l i = l j = 0 for a pair of indices i ≠ j. This implies<br />
that these <strong>points</strong> are not contained in the line “T 0 = 0”. The fiber of q over<br />
any of these three <strong>points</strong> is therefore given by T 3 (a 1 + d 1 T 3 )(a 2 + d 2 T 3 ) = 0<br />
which shows that q is unramified. Consequently, the <strong>points</strong> over P 1 , P 2 , <strong>and</strong> P 3<br />
are smooth <strong>points</strong> of X.<br />
It remains to consider the fiber of P. For that point, we find<br />
( )<br />
θ<br />
(1)<br />
(l 1 : l 2 : l 3 ) =<br />
θ (2) − θ : θ (2)<br />
(3) θ (3) − θ : θ (3)<br />
.<br />
(1) θ (1) − θ (2)<br />
The linear system of equations<br />
T 0 + θ (1) T 1 + (θ (1) ) 2 T 2 =<br />
T 0 + θ (2) T 1 + (θ (2) ) 2 T 2 =<br />
T 0 + θ (3) T 1 + (θ (3) ) 2 T 2 =<br />
θ (1)<br />
θ (2) − θ (3) ,<br />
θ (2)<br />
θ (3) − θ (1) ,<br />
θ (3)<br />
θ (1) − θ (2)
118 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
has the obvious solution<br />
(T 0 : T 1 : T 2 )<br />
= (−3θ (1) θ (2) θ (3) : 2[θ (1) θ (2) + θ (2) θ (3) + θ (3) θ (1) ] : [−(θ (1) + θ (2) + θ (3) )])<br />
( p<br />
3<br />
= 0 − 3p0 2 − Ap 0<br />
: 2 )<br />
9 3 (p 0 − 1)p 0 : p 0<br />
(<br />
)<br />
6(p 0 − 1)<br />
= 1 :<br />
p0 2 − 3p 0 − A : 9<br />
p0 2 − 3p .<br />
0 − A<br />
which is unique since the coefficient matrix is V<strong>and</strong>ermonde.<br />
Using Lemma 5.8again, a direct calculation which is conveniently done inmaple<br />
shows that<br />
(<br />
1 + 6(p )<br />
0 − 1) 9<br />
p0 2 − 3p 0 − A θ(i) +<br />
p0 2 − 3p 0 − A (θ(i) ) 2 = − 4p 0 − A 2<br />
(p0 2 − 3p 0 − A) . 2<br />
3<br />
∏<br />
i=1<br />
Therefore, the fiber of q over P is given by<br />
4p 0 − A<br />
T 3 (a 1 + d 1 T 3 )(a 2 + d 2 T 3 ) +<br />
2<br />
(p0 2 − 3p 0 − A) = 0 . 2<br />
By assumption, this equation has no multiple solutions. Therefore, q is unramified<br />
over P. The <strong>points</strong> above P are hence smooth <strong>points</strong> of X. □<br />
5.14. Remark. –––– The fact that X is smooth is not mentioned in the original<br />
literature, neither in [Mord], nor in [Man].<br />
iii. Existence of an adelic point. —<br />
5.15. Proof of Proposition 5.3.ii) –––– We have to show that X (Éν) ≠ ∅ for<br />
every valuation ofÉ.<br />
X (Ê) ≠ ∅ is obvious. For a prime number p, in order to show X (Ép) ≠ ∅, we<br />
use Hensel’s lemma. It is sufficient to verify that the reduction X p has a smooth<br />
p-valued point.<br />
Case 1: p = p 0 .<br />
As ν p0 (θ) > 0, the reduction X p0 is given by T 3 (a 1 T 0 +d 1 T 3 )(a 2 T 0 +d 2 T 3 ) = T0 3.<br />
This is the union of three planes meeting in the line given by T 0 = T 3 = 0. By assumption,<br />
one of them has multiplicity one <strong>and</strong> is defined overp 0<br />
. It contains<br />
p0 2 smooth <strong>points</strong>.<br />
Case 2: p ≠ p 0 , p ∤ d 1 d 2 .<br />
It suffices to show that there is a smoothp-valued point on the intersection X ′<br />
p<br />
of X p with the hyperplane “T 0 = 0”. This curve is given by the equation<br />
∏<br />
d 1 d 2 T 3<br />
3 = θ (1) θ (2) θ (3) 3<br />
i=1<br />
(T 1 + θ (i) T 2 ) .
Sec. 5] THE EXAMPLES OF MORDELL 119<br />
If p ≠ 3 then this equation defines a smooth genus one curve. It has anp-valued<br />
point by Hasse’s bound.<br />
If p = 3 then the projection X ′<br />
p → P 1 given by (T 1 : T 2 : T 3 ) ↦→ (T 1 : T 2 ) is oneto-one<br />
onp-valued <strong>points</strong>. At least one of them is smooth since ∏ 3 i=1 (T +θ(i) )<br />
is a separable polynomial.<br />
Case 3: p ≠ p 0 , p|d 1 d 2 .<br />
X ′ p := X p ∩ “T 0 = 0” is given by 0 = θ (1) θ (2) θ (3) ∏ 3 i=1 (T 1+θ (i) T 2 ). In particular,<br />
x = (0 : 0 : 0 : 1) ∈ X p (p). We may assume that x is singular.<br />
Then, X p is given as Q(T 0 , T 1 , T 2 )T 3 +K (T 0 , T 1 , T 2 ) = 0 for Q a quadratic form<br />
<strong>and</strong> K a cubic form. If Q ≢ 0 then there is anp-<strong>rational</strong> line l through x such<br />
that Q| l ≠ 0. Hence, l meets X p twice in x <strong>and</strong> once in anotherp-valued point<br />
which is smooth.<br />
Otherwise, (F mod p) does not depend on T 3 , i.e., the left h<strong>and</strong> side of the<br />
equation of X vanishes modulo p. This means, one of the factors on the left<br />
h<strong>and</strong> side vanishes modulo p. Say, we have a 1 ≡ d 1 ≡ 0 (mod p).<br />
Then, by assumption, p decomposes completely in K. At such a prime, X p ′ is<br />
the union of three lines which are all defined overp, different from each other,<br />
<strong>and</strong> meeting in one point. We have plenty of smooth <strong>points</strong>.<br />
□<br />
iv. Construction of a Brauer class. —<br />
5.16. –––– We write G := Gal(K/É).<br />
According to the general strategy, described in the section above, an element<br />
of H 1 (G, Pic(X K )) ⊆ H 1( Gal(É/É), Pic(XÉ) ) may be given by a <strong>rational</strong><br />
function f ∈ Q(X ) such that div( f ) ∈ N Div(X K ). Such a function is<br />
f := a 1T 0 + d 1 T 3<br />
T 3<br />
.<br />
Indeed, “a 1 T 0 + d 1 T 3 = 0” defines a triangle, the three lines of which are given<br />
by that equation <strong>and</strong> T 0 + θ (i) T 1 + (θ (i) ) 2 T 2 = 0 for i = 1, 2, or 3, respectively.<br />
Considered as a Weil divisor, this triangle is the norm of the divisor given by<br />
one of the lines.<br />
We write [ f ] for the Brauer class defined by f .<br />
5.17. Remarks. –––– i) One might work as well with [ f ′ ] for f ′ := a 2T 0 +d 2 T 3<br />
T 3<br />
or<br />
even with both, [ f ] <strong>and</strong> [ f ′ ], as Manin does in [Man,кVI,ÈÖÐÓÒº].<br />
However,<br />
[ ]<br />
[ f ]·[ f ′ (a1 T 0 + d 1 T 3 )(a 2 T 0 + d 2 T 3 )<br />
] =<br />
=<br />
Consequently, X (É)<br />
T3<br />
2<br />
[ f ] = X (É)<br />
[ f ′] .<br />
[ ( )]<br />
T0 + θT 1 + θ 2 T 2<br />
N<br />
= 0 .<br />
T 3
=/3<br />
120 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
ii) Under very mild hypotheses on X, one has Br(X )/π ∗ Br(SpecÉ) ∼<br />
<strong>and</strong> [ f ] is a generator. We will discuss this point in Subsection vi.<br />
5.18. Lemma. –––– Let ν be any valuation ofÉdifferent from ν p0 . Then,<br />
for every adelic point x ∈ X (É).<br />
Proof. First step: Elementary cases.<br />
ev ν ([ f ], x) = 0<br />
If ν = ν ∞ then ev ν ([ f ], x) = 0 since #G = 3 while only the values 1 2 <strong>and</strong> 0<br />
are possible.<br />
If ν = ν p <strong>and</strong> p is decomposed in K then every element ofÉ∗ p is a norm.<br />
Therefore, ev ν ([ f ], x) = 0.<br />
Second step: Preparations.<br />
It remains to consider the case that p remains prime in K. We have to show that<br />
a 1 t 0 +d 1 t 3<br />
t 3<br />
is a norm for each point (t 0 : t 1 : t 2 : t 3 ) ∈ X (Ép). An element w ∈É∗ p<br />
is a norm if <strong>and</strong> only if 3|ν p (w).<br />
It might happen that θ is not a unit in K ν . As K ν /Ép is unramified, there exists<br />
t ∈É∗ p such that θ := tθ ∈ K ν is a unit. The surface X ′ given by<br />
T 3 (a 1 T 0 + d 1 T 3 )(a 2 T 0 + d 2 T 3 ) =<br />
3<br />
∏<br />
i=1<br />
(<br />
T0 + θ (i) T 1 + (θ (i) ) 2 T 2<br />
)<br />
is isomorphic to X. The map ι: (t 0 : t 1 : t 2 : t 3 ) ↦→ (t 0 :<br />
isomorphism X → X ′ which leaves the <strong>rational</strong> function a 1T 0 +d 1 T 3<br />
T 3<br />
Hence, we may assume without restriction that θ ∈ K ν is a unit.<br />
Third step: The case θ is a unit.<br />
We may assume t 0 , t 1 , t 2 , t 3 ∈p are coprime.<br />
t 1t<br />
:<br />
t 2<br />
t 2<br />
: t 3 ) is an<br />
unchanged.<br />
We could have ∏ 3 i=1(<br />
t0 +θ (i) t 1 + (θ (i) ) 2 t 2<br />
) = 0 only when t0 = t 1 = t 2 = 0 since<br />
θ generates a cubic field extension ofÉp. Then, t 3 (a 1 t 0 + d 1 t 3 )(a 2 t 0 + d 2 t 3 ) = 0<br />
implies t 3 = 0 which is a contradiction.<br />
Therefore, both sides of the equation are different from zero. In particular, we<br />
automatically have t 3 ≠ 0.<br />
If ν ( t 3 (a 1 t 0 + d 1 t 3 )(a 2 t 0 + d 2 t 3 ) ) = 0 then a 1t 0 +d 1 t 3<br />
t 3<br />
is clearly a norm. Otherwise,<br />
we have<br />
ν ( ∏<br />
3 (t 0 + θ (i) t 1 + (θ (i) ) 2 t 2 ) ) > 0 .<br />
i=1<br />
This implies that ν(t 0 ), ν(t 1 ), ν(t 2 ) > 0. Then, t 3 must be a unit.<br />
From the equation of X, we deduce ν(d 1 d 2 ) > 0. If ν(d 2 ) > 0 then, according<br />
to the assumption, d 1 is a unit. This shows ν(a 1 t 0 + d 1 t 3 ) = 0 from which the<br />
assertion follows.
Sec. 5] THE EXAMPLES OF MORDELL 121<br />
Thus, assume ν(d 1 ) > 0. Then, d 2 is a unit <strong>and</strong>, therefore, ν(a 2 t 0 + d 2 t 3 ) = 0.<br />
Further, we note that 3|ν ( ∏ 3 i=1 (t 0 + θ (i) t 1 + (θ (i) ) 2 t 2 ) ) since the right h<strong>and</strong> side<br />
is a norm. By consequence,<br />
3|ν ( t 3 (a 1 t 0 + d 1 t 3 )(a 2 t 0 + d 2 t 3 ) ) .<br />
Altogether, we see that 3|ν(a 1 t 0 + d 1 t 3 ) <strong>and</strong> 3|ν ( )<br />
a 1 t 0 +d 1 t 3<br />
t 3<br />
. The claim follows.<br />
□<br />
5.19. Fact. –––– i) The reduction X p0 of X at p 0 is given by<br />
T 3 (a 1 T 0 + d 1 T 3 )(a 2 T 0 + d 2 T 3 ) = T 3<br />
0 .<br />
ii) Over the algebraic closure, X p0 is the union of three planes meeting in a triple line<br />
“T 0 = T 3 = 0”.<br />
iii) NoÉp 0<br />
-valued point on X reduces to the triple line.<br />
Proof. i) <strong>and</strong> ii) are clear.<br />
iii) Let (t 0 : t 1 : t 2 : t 3 ) ∈ X (Ép 0<br />
). We may assume t 0 , t 1 , t 2 , t 3 ∈p 0<br />
are coprime.<br />
We write ν for the extension of ν p0 to K. Then ν(t 0 ) ≥ 1 <strong>and</strong> ν(t 3 ) ≥ 1 together<br />
imply ν(t 3 (a 1 t 0 + d 1 t 3 )(a 2 t 0 + d 2 t 3 )) ≥ 3. On the other h<strong>and</strong>,<br />
ν ( ∏<br />
3 (t 0 + θ (i) t 1 + (θ (i) ) 2 t 2 ) )<br />
i=1<br />
is equal to 1 or 2, since t 1 or t 2 is a unit <strong>and</strong> ν(θ (i) ) = 1 3 .<br />
□<br />
5.20. Lemma. –––– Let (t 0 : t 1 : t 2 : t 3 ) ∈ X (Ép 0<br />
).<br />
Then, ev νp0 ([ f ], (t 0 : t 1 : t 2 : t 3 )) = 0 if <strong>and</strong> only if a 1t 0 +d 1 t 3<br />
t 3<br />
is a cube in∗ p0<br />
.<br />
Proof. (p 0 ) = p 3 is totally ramified. InÉp 0<br />
, there is a uniformizer which is<br />
a norm. Further, a p 0 -adic unit u is a norm if <strong>and</strong> only if u := (u mod p 0 ) is a<br />
cube in∗ p0<br />
.<br />
We have that a 1t 0 +d 1 t 3<br />
t 3<br />
is automatically a p 0 -adic unit. Indeed, suppose that<br />
t 0 , t 1 , t 2 , t 3 ∈p 0<br />
are coprime. Modulo p 0 , the equation of X is<br />
T 3 (a 1 T 0 + d 1 T 3 )(a 2 T 0 + d 2 T 3 ) = T 3<br />
0 .<br />
As no point may reduce to the singular line, we see that t 0 ≠ 0. This implies<br />
t 3 ≠ 0 <strong>and</strong> a 1 t 0 + d 1 t 3 ≠ 0 which is the assertion.<br />
□
122 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
v. Examples. —<br />
5.21. Corollary. –––– Let p 0 ≡ 1 (mod 3) be a prime number <strong>and</strong> consider the<br />
cubic surface X ⊂ P 3É, given by<br />
T 3 (a 1 T 0 + d 1 T 3 )(a 2 T 0 + d 2 T 3 ) =<br />
3<br />
∏<br />
i=1<br />
(<br />
T0 + θ (i) T 1 + (θ (i) ) 2 T 2<br />
)<br />
.<br />
Here, a 1 , a 2 , d 1 , d 2 ∈<strong>and</strong> assume that a 1 ≠ 0, a 2 ≠ 0, a 1 /d 1 ≠ a 2 /d 2 , p 0 ∤d 1 d 2 ,<br />
<strong>and</strong> that gcd(a 1 , d 1 ) <strong>and</strong> gcd(a 2 , d 2 ) contain only prime factors which decompose<br />
in K.<br />
i) Assume that<br />
T (a 1 + d 1 T )(a 2 + d 2 T ) − 1 = 0<br />
has three different zeroes t (1) , t (2) , t (3) ∈p 0<br />
.<br />
If a 1+d 1 t (1)<br />
, a 1+d 1 t (2)<br />
, <strong>and</strong> a 1+d 1 t (3)<br />
are non-cubes in∗ t (1) t (2) t (3)<br />
p0<br />
then X (É) = ∅. On X, there<br />
is a Brauer-Manin obstruction to the Hasse principle.<br />
If exactly one of the three expressions is a cube in∗ p0<br />
then, on X, there is a Brauer-<br />
Manin obstruction to weak approximation.<br />
ii) Assume that<br />
T (a 1 + d 1 T )(a 2 + d 2 T ) − 1 = 0<br />
has exactly one zero t ∈p 0<br />
which is simple.<br />
a 1 +d 1 t<br />
t<br />
If is not a cube in∗ p0<br />
then X (É) = ∅. On X, there is a Brauer-Manin<br />
obstruction to the Hasse principle.<br />
5.22. Remark. –––– In the case of three different solutions, it is impossible that<br />
exactly two of the expressions a 1+d 1 t (1)<br />
, a 1+d 1 t (2)<br />
, <strong>and</strong> a 1+d 1 t (3)<br />
are cubes. Indeed, a<br />
t (1) t (2) t (3)<br />
direct calculation shows<br />
a 1 + d 1 t (1) + d 1 t<br />
·a1 (2) + d 1 t<br />
·a1 (3)<br />
= d 3<br />
t (1) t (2) t (3) 1 .<br />
This means, the Brauer-Manin obstruction might exclude no, one, or all three<br />
of the planes, the reduction X p0 consists of, but not exactly two of them.<br />
5.23. Example. –––– Let p 0 ≡ 1 (mod 3) be a prime number <strong>and</strong> assume that<br />
d 1 <strong>and</strong> d 2 are non-cubes in∗ p0<br />
such that d 1 d 2 is a cube, p 0 |a 1 <strong>and</strong> p 0 |a 2 . Further,<br />
suppose a 1 ≠ 0, a 2 ≠ 0, <strong>and</strong> a 1 /d 1 ≠ a 2 /d 2 , as well as that gcd(a 1 , d 1 )<br />
<strong>and</strong> gcd(a 2 , d 2 ) contain only prime factors which decompose in K.<br />
Then, the cubic surface X given by<br />
T 3 (a 1 T 0 + d 1 T 3 )(a 2 T 0 + d 2 T 3 ) =<br />
is a counterexample to the Hasse principle.<br />
3<br />
∏<br />
i=1<br />
(<br />
T0 + θ (i) T 1 + (θ (i) ) 2 T 2<br />
)
Sec. 5] THE EXAMPLES OF MORDELL 123<br />
Indeed, one has<br />
a 1 t 0 + d 1 t 3<br />
t 3<br />
≡ d 1 (mod p 0 )<br />
which is a non-cube. The assumption that d 1 d 2 is a cube is important to<br />
guarantee X (Ép 0<br />
) ≠ ∅.<br />
More concretely, for p 0 = 19, a counterexample to the Hasse principle is<br />
given by<br />
T 3 (19T 0 + 5T 3 )(19T 0 + 4T 3 ) =<br />
3<br />
∏<br />
i=1<br />
(<br />
T0 + θ (i) T 1 + (θ (i) ) 2 T 2<br />
)<br />
= T0 3 − 19T0 2 T 1 + 133T0 2 T 2 + 114T 0 T1<br />
2<br />
− 1 539T 0 T 1 T 2 + 5 054T 0 T2 2 − 209T1<br />
3<br />
+ 3 971T1 2 T 2 − 23 826T 1 T2 2 + 43 681T2 3 .<br />
5.24. Example. –––– For p 0 = 19, consider the cubic surface X given by<br />
3 ( )<br />
T 3 (T 0 + T 3 )(12T 0 + T 3 ) = ∏ T0 + θ (i) T 1 + (θ (i) ) 2 T 2 .<br />
i=1<br />
Then, X (É) ≠ ∅ but X (É) = ∅. On X, there is a Brauer-Manin obstruction<br />
to the Hasse principle.<br />
Indeed, in19, the cubic equation<br />
T (1 + T )(12 + T ) − 1 = 0<br />
has the three solutions 12, 15, <strong>and</strong> 17. However, in19, 13/12 = 9, 16/15 = 15,<br />
<strong>and</strong> 18/17 = 10 which are three non-cubes.<br />
5.25. Example. –––– For p 0 = 19, consider the cubic surface X given by<br />
T 3 (T 0 + T 3 )(6T 0 + T 3 ) =<br />
3<br />
∏<br />
i=1<br />
(<br />
T0 + θ (i) T 1 + (θ (i) ) 2 T 2<br />
)<br />
.<br />
Then, on X, there is a Brauer-Manin obstruction to weak approximation.<br />
Indeed, in19, the cubic equation<br />
T (1 + T )(6 + T ) − 1 = 0<br />
has the three solutions 8, 9, <strong>and</strong> 14. However, in19, 10/9 = 18 is a cube while<br />
9/8 = 13 <strong>and</strong> 15/14 = 16 are non-cubes.<br />
The smallestÉ-<strong>rational</strong> point on X is (14 : 15 : 2 : (−7)). Note that indeed<br />
T 3 /T 0 = −7/14 ≡ 9 (mod 19).
124 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
5.26. Example. –––– For p 0 = 19, consider the cubic surface X given by<br />
3 ( )<br />
T 3 (T 0 + T 3 )(2T 0 + T 3 ) = T0 + θ (i) T 1 + (θ (i) ) 2 T 2 .<br />
Then, on X, there is a Brauer-Manin obstruction to the Hasse principle.<br />
Indeed, in19, the cubic equation<br />
∏<br />
i=1<br />
T (1 + T )(2 + T ) − 1 = 0<br />
has T = 5 as its only solution. The other two solution are conjugate to each<br />
other in19 2. However, in19, 6/5 = 5 is a non-cube.<br />
5.27. Example (Swinnerton-Dyer [S-D62]). —– For p 0 = 7, consider the cubic<br />
surface X given by<br />
T 3 (T 0 + T 3 )(T 0 + 2T 3 ) =<br />
3<br />
∏<br />
i=1<br />
(<br />
T0 + θ (i) T 1 + (θ (i) ) 2 T 2<br />
)<br />
= T0 3 − 7T0 2 T 1 + 21T0 2 T 2 + 14T 0 T1 2 − 77T 0 T 1 T 2<br />
+ 98T 0 T2 2 − 7T1 3 + 49T1 2 T 2 − 98T 1 T2 2 + 49T2 3 .<br />
Then, on X, there is a Brauer-Manin obstruction to the Hasse principle.<br />
Indeed, in7, the cubic equation<br />
T (1 + T )(1 + 2T ) − 1 = 0<br />
has T = 5 as its only solution. However, in7, 6/5 = 4 is not a cube.<br />
5.28. Remark. –––– From each of the examples given, by adding multiplies<br />
of p 0 to the coefficients a 1 , d 1 , a 2 , <strong>and</strong> d 2 , a family of surfaces arises which are<br />
of similar nature. In Example 5.23, some care has to be taken to keep a 1 <strong>and</strong> a 2<br />
different from zero <strong>and</strong> a 1 /d 1 different from a 2 /d 2 . In all examples, gcd(a 1 , d 1 )<br />
<strong>and</strong> gcd(a 2 , d 2 ) need some consideration.<br />
5.29. Remark (Lattice basis reduction). —– The norm form in the p 0 = 19 examples<br />
produces coefficients which are rather large. An equivalent form with<br />
smaller coefficients may be obtained using lattice basis reduction. In its simplest<br />
form, this means the following.<br />
For the rank-2 lattice inÊ3 , generated by the vectors v 1 := (θ (1) , θ (2) , θ (3) ) <strong>and</strong><br />
v 2 := ( (θ (1) ) 2 , (θ (2) ) 2 , (θ (3) ) 2) , in fact {v 1 , v 2 + 7v 1 } is a reduced basis. Therefore,<br />
the substitution T 1 ′ := T 1−7T 2 simplifies the norm form. Actually, we find<br />
3<br />
∏<br />
i=1<br />
(<br />
T0 + θ (i) T 1 + (θ (i) ) 2 T 2<br />
)<br />
=T<br />
3<br />
0 −19T0 2 T 1+114T ′<br />
0 T ′<br />
1 2 +57T 0 T ′<br />
1 T 2−133T 0 T2<br />
2<br />
− 209T ′<br />
1 3 − 418T ′<br />
1 2 T 2 + 1045T ′<br />
1 T 2 2 − 209T2 3 .
Sec. 5] THE EXAMPLES OF MORDELL 125<br />
vi. Calculation of the complete Br(X )/π ∗ Br(SpecÉ). —<br />
5.30. Notation. –––– Let X be a cubic surface in P 3 given by<br />
for six linear forms l 1 , l 2 , l 3 , λ 1 , λ 2 , λ 3 .<br />
l 1 l 2 l 3 = λ 1 λ 2 λ 3<br />
Then, we write L ij for the line on X given by l i = λ j = 0.<br />
The subgroup of Pic(X ) generated by these nine lines will be denoted by P.<br />
Further, we let S be the free abelian group over all lines L ij .<br />
5.31. Facts. –––– i) P ⊂ Pic(X ) is a lattice of rank five <strong>and</strong> discriminant 3.<br />
ii) The kernel S 0 of the canonical homomorphism S → P is generated by<br />
D 1 := L 11 +L 12 +L 13 −L 21 −L 22 −L 23 , D 2 := L 11 +L 12 +L 13 −L 31 −L 32 −L 33 ,<br />
D 3 := L 11 + L 13 − L 22 − L 32 , <strong>and</strong> D 4 := L 11 + L 12 − L 23 − L 33 .<br />
Proof. We have D 1 = div(l 1 /l 2 ), D 2 = div(l 1 /l 3 ), D 3 = div(l 1 /λ 2 ),<br />
<strong>and</strong> D 4 = div(l 1 /λ 3 ). Thus, D 1 , D 2 , D 3 , D 4 ∈ ker(S → P). It is easy to<br />
check that D 1 , D 2 , D 3 , D 4 are linearly independent. This shows that rkP ≤ 5.<br />
On the other h<strong>and</strong>, the intersection matrix of L 11 , L 22 , L 23 , L 32 , <strong>and</strong> L 33 is<br />
⎛<br />
⎞<br />
−1 0 0 0 0<br />
0 −1 1 1 0<br />
⎜ 0 1 −1 0 1<br />
⎟<br />
⎝ 0 1 0 −1 1⎠ .<br />
0 0 1 1 −1<br />
The determinant of this matrix is equal to 3 which is a square-free integer.<br />
Assertion i) is proven.<br />
It remains to verify that D 1 , D 2 , D 3 , <strong>and</strong> D 4 generate ker(S → P). For this, we<br />
have to show that the canonical injection<br />
is actually bijective.<br />
〈L 11 , L 22 , L 23 , L 32 , L 33 〉 −→ S/〈D 1 , D 2 , D 3 , D 4 〉<br />
For surjectivity, observe that, modulo 〈D 1 , D 2 , D 3 , D 4 〉, the remaining generators<br />
are given by L 12 ≡ −L 11 + L 23 + L 33 , L 13 ≡ −L 11 + L 22 + L 32 ,<br />
L 21 ≡ L 11 + L 12 + L 13 − L 22 − L 23 ≡ −L 11 + L 32 + L 33 , <strong>and</strong>, finally,<br />
L 31 ≡ L 11 + L 12 + L 13 − L 32 − L 33 ≡ −L 11 + L 22 + L 23 .<br />
□<br />
5.32. Proposition. –––– Let K/Ébe a Galois extension of degree three. Assume<br />
that λ 1 , λ 2 , λ 3 are defined over K <strong>and</strong> form a Galois orbit <strong>and</strong> that l 1 , l 2 , l 3<br />
are linear forms defined overÉ.<br />
Then, Ĥ −1( Gal(K/É), P ) =/3<strong>and</strong> [l 1 /l 2 ] is a generator.
126 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
Proof. By Lemma II.8.19, we have a canonical isomorphism<br />
Ĥ −1( Gal(K/É), P ) ∼ =<br />
−→ (NS ∩ S 0 )/NS 0 .<br />
Let us calculate NS ∩ S 0 .<br />
D 1 = N (L 11 − L 21 ) <strong>and</strong> D 2 = N (L 11 − L 31 ) are clearly norms. If<br />
aD 3 + bD 4 = (a + b)L 11 + bL 12 + aL 13 − aL 22 − bL 23 − aL 32 − bL 33<br />
is a norm then the coefficients of L 31 , L 32 , <strong>and</strong> L 33<br />
Hence, a = b = 0. Consequently,<br />
must coincide.<br />
NS ∩ S 0 = 〈D 1 , D 2 〉 .<br />
Further, N (D 1 ) = 3D 1 , N (D 2 ) = 3D 2 , <strong>and</strong> N (D 3 ) = N (D 4 ) = D 1 + D 2 .<br />
Hence,<br />
NS 0 = 〈3D 1 , 3D 2 , D 1 + D 2 〉 .<br />
The assertion follows.<br />
□<br />
5.33. –––– Return to the case of a cubic surface given an equation of type<br />
3 ( )<br />
T 3 (a 1 T 0 + d 1 T 3 )(a 2 T 0 + d 2 T 3 ) = T0 + θ (i) T 1 + (θ (i) ) 2 T 2 .<br />
To deduce information on H 1( Gal(É/É), Pic(XÉ) ) from H 1 (G, P) ∼ =/3,<br />
we need some underst<strong>and</strong>ing of the remaining 18 lines.<br />
OverÉ(θ), there are defined at least nine lines. The list of the 350 conjugacy<br />
classes of sub<strong>groups</strong> of W (E 6 ), established using GAP (cf. II.8.23 <strong>and</strong> the appendix),<br />
contains only four classes which fix nine or more lines. The corresponding<br />
extract looks like this.<br />
1 #U = 1 [ ], #H^1 = 1, Rk(Pic) = 7, Orbits [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,<br />
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ]<br />
2 #U = 2 [ 2 ], #H^1 = 1, Rk(Pic) = 6, Orbits [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,<br />
1, 2, 2, 2, 2, 2, 2 ]<br />
7 #U = 3 [ 3 ], #H^1 = 1, Rk(Pic) = 5, Orbits [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 3, 3, 3,<br />
3, 3 ]<br />
24 #U = 6 [ 2 ], #H^1 = 1, Rk(Pic) = 5, Orbits [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 3, 3, 3,<br />
3, 3 ]<br />
Thus, the remaining 18 lines are defined over an extension ofÉ(θ) which may<br />
be of degree 6, 3, 2, or 1.<br />
We see that rkPic(XÉ(θ)) ≥ 5. Equality holds if <strong>and</strong> only if the field of definition<br />
of the 27 lines is of degree 3 or 6 overÉ(θ).<br />
We also observe that, in any case,<br />
∏<br />
i=1<br />
H 1( Gal(É/É(θ)), Pic(XÉ) ) = 0 .
Sec. 5] THE EXAMPLES OF MORDELL 127<br />
5.34. Remark. –––– For every concrete choice of the coefficients, one may<br />
determine the field of definition of the 27 lines by a Gröbner base calculation.<br />
It turned out that it was of degree 6 overÉ(θ) in every example, we tested.<br />
Thus, degree 6 seems to be the generic case.<br />
If the field of definition of the 27 lines is a degree 6 or degree 3 extension ofÉ(θ)<br />
then we may describe H 1( Gal(É/É), Pic(XÉ) ) , completely.<br />
5.35. Proposition. –––– Let p 0 ≡ 1 (mod 3) be a prime number, θ (i) as above,<br />
<strong>and</strong> X ⊂ P 3Ébe the cubic surface given by<br />
T 3 (a 1 T 0 + d 1 T 3 )(a 2 T 0 + d 2 T 3 ) =<br />
3<br />
∏<br />
i=1<br />
(<br />
T0 + θ (i) T 1 + (θ (i) ) 2 T 2<br />
)<br />
for a 1 , a 2 , d 1 , d 2 ∈. Assume that the field of definition of the 27 lines on X is of<br />
degree 3 or 6 overÉ(θ).<br />
Then, Br(X )/π ∗ Br(SpecÉ) ∼ =/3<strong>and</strong> [ a 1T 0 +d 1 T 3<br />
T 3<br />
] is a generator.<br />
Proof. Since H 1( Gal(É/É(θ)), Pic(XÉ) ) = 0, the inflation map<br />
H 1( Gal(É(θ)/É), Pic(XÉ) Gal(É/É(θ) ) ) −→ H 1( Gal(É/É), Pic(XÉ) )<br />
is an isomorphism. Hence, in view of Proposition 5.32, it will suffice to<br />
show that P = Pic(XÉ) Gal(É/É(θ)) .<br />
“⊆” is obvious.<br />
“⊇” Our assumption implies that rkPic(XÉ(θ)) = 5. Further, from Proposition<br />
II.8.11, we see that Pic(XÉ(θ)) is always a subgroup of finite index<br />
in Pic(XÉ) Gal(É/É(θ)) . Hence, rkPic(XÉ) Gal(É/É(θ)) = 5, too.<br />
By Fact 5.31.i), we have rkP = 5. Thus, P ⊆ Pic(XÉ) Gal(É/É(θ)) is a sublattice<br />
of finite index. Since DiscP = 3 is square-free, the claim follows. □<br />
5.36. Example. –––– For p 0 = 19, consider the cubic surface X given by<br />
T 3 (T 0 + T 3 )(7T 0 + T 3 ) =<br />
3<br />
∏<br />
i=1<br />
(<br />
T0 + θ (i) T 1 + (θ (i) ) 2 T 2<br />
)<br />
.<br />
Then, on X, there is no Brauer-Manin obstruction to weak approximation.<br />
Indeed, a Gröbner base calculation shows that the 27 lines on X are defined over<br />
a degree 6 extension ofÉ(θ). Thus, Br(X )/π ∗ Br(SpecÉ) ∼ =/3. A generator<br />
is given by the class α := [ T 0+T 3<br />
T 3<br />
].<br />
However, in19, the cubic equation<br />
T (1 + T )(7 + T ) − 1 = 0
128 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
has the only solution T = 11. Hence, everyÉ19-valued point on X has a<br />
reduction of the form (1 : y : z : 11). For the local evaluation, we find<br />
= 12/11 = 8 ∈19 which is a cube.<br />
T 0 +T 3<br />
T 3<br />
Br<br />
Thus, ev(x, α) = 0 for every adelic point x ∈ X (É). By consequence,<br />
X (É) = X (É). One might expect that X satisfies weak approximation.<br />
5.37. Remarks (Degenerate cases). —– i) If the field of definition of the 27 lines<br />
is quadratic overÉ(θ) then Br(X )/π ∗ Br(SpecÉ) = 0.<br />
In fact, in this case, we have a cyclic group G of order 6 acting on the 27 lines.<br />
There are seven conjugacy classes of cyclic <strong>groups</strong> of order 6 in W (E 6 ). We have<br />
the following list.<br />
26 #U = 6 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 6, 6, 6, 6 ]<br />
27 #U = 6 [ 2, 3 ], #H^1 = 4 [ 2, 2 ], Rk(Pic) = 1, Orbits [ 3, 6, 6, 6, 6 ]<br />
28 #U = 6 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 4, Orbits [ 1, 1, 1, 2, 2, 2, 3, 3, 3, 3, 6 ]<br />
30 #U = 6 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 2, 6, 6, 6 ]<br />
32 #U = 6 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 3, 3, 3, 3, 6, 6 ]<br />
36 #U = 6 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 2, 2, 3, 3, 6, 6 ]<br />
39 #U = 6 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 6, 6, 6, 6 ]<br />
As the element of order two fixes 15 lines, we may not have more than two<br />
orbits of size 6. This implies the claim.<br />
Actually, we are in class No. 32.<br />
ii) If the 27 lines are defined overÉ(θ) then, seemingly, there are two cases.<br />
One might either have rk Pic(X ) = 3. Then, Br(X )/π ∗ Br(SpecÉ) = 0.<br />
Or, otherwise, rk Pic(X ) = 1. In this case, Br(X )/π ∗ Br(SpecÉ) = (/3) 2 .<br />
α := [ a 1T 0 +d 1 T 3<br />
T 3<br />
] is one of the generators.<br />
We do not know of an example overÉin which any of these degenerate<br />
cases occurs.<br />
6. The “first case” of diagonal cubic surfaces<br />
i. The statements. —<br />
6.1. –––– In this section, we show how to compute the Brauer-Manin obstruction<br />
for a large class of diagonal cubic surfaces. More precisely, we are concerned<br />
with smooth diagonal cubic surfaces in P 3É, given by an equation of the form<br />
a 0 T 3<br />
0 + a 1 T 3<br />
1 + a 2 T 3<br />
2 + a 3 T 3<br />
3 = 0<br />
for a 0 , . . . , a 3 ∈\{0}, such that the following additional condition is satisfied.<br />
(∗) There exists a prime number p 0 such that p 0 |a 3 but neither p 3 0|a 3 nor p 0 |a i<br />
for i = 0, 1, 2.
Sec. 6] THE “FIRST CASE” OF DIAGONAL CUBIC SURFACES 129<br />
6.2. –––– X has a model over Specgiven by the same equation. We will<br />
denote that model by X .<br />
On X, there is the smooth genus one curve E given by the equation T 3 = 0.<br />
If p 0 ≠ 3 then assumption (∗) implies that the reduction X p0 is a cone over E p0 .<br />
There is the mapping<br />
red: X (Ép 0<br />
) −→ E p0 (p 0<br />
)<br />
(x 0 : x 1 : x 2 : x 3 ) ↦→ ((x 0 mod p 0 ) : (x 1 mod p 0 ) : (x 2 mod p 0 )) .<br />
6.3. Remark. –––– The Brauer-Manin obstruction on diagonal cubic surfaces<br />
was studied by J.-L.Colliot-Thélène, D. Kanevsky, <strong>and</strong> J.-J.Sansuc in [CT/K/S].<br />
For the case when (∗) is fulfilled, the main result asserts that X (É) Br is exactly<br />
one third of the whole of X (É) in a sense made precise.<br />
More concretely, there is the following theorem.<br />
∅. Sup-<br />
6.4. Theorem (Colliot-Thélène, Kanevsky, <strong>and</strong> Sansuc). —–<br />
Let X ⊂ P 3Ébe →É/<br />
a smooth diagonal cubic surface such that X (É) ≠<br />
pose that (∗) is fulfilled for a certain prime number p 0 .<br />
a) Then, Br(X )/π ∗ Br(SpecÉ) ∼ =/3.<br />
b) The image of the evaluation map<br />
ev: Br(X ) × X (É)<br />
is 1 3/.<br />
c) Choose an adelic point (x ν ) ν ∈ X ().<br />
Let α ∈ Br(X ) be such that its image in Br(X )/π ∗ Br(SpecÉ) is non-trivial.<br />
Assume that α is a normalized modulo π ∗ Br(SpecÉ) in such a way that<br />
ev νp0 (α, x νp0 ) = 0.<br />
i) Assume p 0<br />
→É/<br />
≠ 3.<br />
Then, the local evaluation map<br />
ev νp0 : Br(X ) × X (Ép 0<br />
)<br />
has the following property.<br />
ev νp0 (α, ·) is the composition of red with a surjective group homomorphism<br />
(E p0 (p 0<br />
), x νp0 ) → 1 3/.<br />
In particular, ev νp0 (α, x) depends only on the reduction x ∈ X (p 0<br />
).<br />
Further, we have<br />
#{y ∈ X (p 0<br />
) | ev νp0 (α, y)=0} = #{y ∈ X (p 0<br />
) | ev νp0 (α, y)=1/3}<br />
= #{y ∈ X (p 0<br />
) | ev νp0 (α, y)=2/3} .
130 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
ii) Let p 0<br />
→É/<br />
= 3.<br />
Then, the local evaluation map<br />
ev ν3 : Br(X ) × X (Ép 0<br />
)<br />
has the following properties.<br />
The image of ev ν3 (α, ·): X (É3) →É/is 1 3/.<br />
There exists a positive integer n such that ev ν3 (α, x) depends only on the reduction<br />
x (3n )<br />
∈ X (/3 n).<br />
For every y 0 ∈ X (3), we have<br />
In particular,<br />
#{y ∈ X (/3 n) | y = y 0 , ev ν3 (α, y)=0}<br />
= #{y ∈ X (/3 n) | y = y 0 , ev ν3 (α, y)=1/3}<br />
= #{y ∈ X (/3 n) | y = y 0 , ev ν3 (α, y)=2/3} .<br />
#{y ∈ X (/3 n) | ev ν3 (α, y)=0} = #{y ∈ X (/3 n) | ev ν3 (α, y)=1/3}<br />
= #{y ∈ X (/3 n) | ev ν3 (α, y)=2/3} .<br />
6.5. Corollary. –––– Let X ⊂ P 3Ébe a smooth diagonal cubic surface such<br />
that X (É) ≠ ∅. Suppose that (∗) is fulfilled for a certain prime number p 0 .<br />
a) Then, on X, there is a Brauer-Manin obstruction to weak approximation.<br />
b) There is, however, no Brauer-Manin obstruction to the Hasse principle.<br />
6.6. Remark. –––– Colliot-Thélène, Kanevsky, <strong>and</strong> Sansuc verify the same behaviour<br />
in their “first case” which is a bit more general than assumption (∗).<br />
Another particular case in which the same result is true is when p 0 ∤ a 0 , a 1 , but<br />
p 0 |a 2 , a 3 , p 2 0 ∤a 2, a 3 , <strong>and</strong> a 2 /a 3 is a p 0 -adic cube.<br />
The three authors also show that, in the “Second case”, there exist diagonal cubic<br />
surfaces such that X (É) ≠ ∅ <strong>and</strong> X (É) Br = X (É) or ∅. From Theorem 6.4,<br />
it is clear that this may happen only under rather restrictive conditions. For example,<br />
every prime number dividing one of the coefficients must necessarily<br />
divide a second one.<br />
6.7. Remark. –––– The statement that the three values are equally distributed<br />
was not formulated in [CT/K/S].
Sec. 6] THE “FIRST CASE” OF DIAGONAL CUBIC SURFACES 131<br />
6.8. Remark. –––– In order to prove Br(X )/π ∗ Br(SpecÉ) ∼ =/3, we make<br />
heavy use of the list of all possible quotients Br(X )/π ∗ Br(SpecÉ) established<br />
by help of the computer. Cf. II.8.23 <strong>and</strong>, for the list in full, the appendix.<br />
ii. Calculating Br(X )/π ∗ Br(SpecÉ). —<br />
6.9. Notation. –––– Following Manin [Man], we will denote the lines on X<br />
as follows. We fix third roots of a 0 , . . . , a 3 , once <strong>and</strong> for all.<br />
Then, L k (m, n) is the line given by<br />
a 1 3 i T i + ζ m 3 a 1 3<br />
j<br />
T j = 0 ,<br />
a 1 3<br />
k T k + ζ n 3 a 1 3<br />
3<br />
T 3 = 0 .<br />
Here, k ∈ {0, 1, 2}, i, j ∈ {0, 1, 2}\{k}, i < j, <strong>and</strong> m, n ∈/3.<br />
6.10. –––– In this notation, the 45 triangles on X may easily be described.<br />
i) The planes “a 1 3<br />
k T k + ζ n 3 a 1 3<br />
3<br />
T 3 = 0”, for k ∈ {0, 1, 2}, n ∈/3, define 9 triangles<br />
∆ 1 kn . They consist of L k(0, n), L k (1, n), <strong>and</strong> L k (2, n).<br />
ii) Analogously, the planes “a 1 3 i T i + ζ m 3 a 1 3<br />
j<br />
T j = 0”, for i, j ∈ {0, 1, 2}, i < j,<br />
<strong>and</strong> m ∈/3, define 9 triangles ∆ 2 km . They consist of L k(m, 0), L k (m, 1),<br />
<strong>and</strong> L k (m, 2) for k ∈ {0, 1, 2} \ {i, j}.<br />
iii) Finally, the planes given by a 1 3<br />
0<br />
T 0 + ζ a 3 a 1 3<br />
1<br />
T 1 + ζ b 3 a 1 3<br />
2<br />
T 2 + ζ c 3 a 1 3<br />
3<br />
T 3 = 0 for<br />
a, b, c ∈ {0, 1, 2} define 27 triangles ∆ 3 abc . Those consist of L 0(b − a, c),<br />
L 1 (b, c − a), <strong>and</strong> L 2 (a, c − b).<br />
6.11. Notation. –––– We write L :=É( 3 √<br />
a3 /a 0 , 3 √<br />
a2 /a 0 , 3 √<br />
a1 /a 0 , ζ 3 )/É) <strong>and</strong><br />
K :=É( 3 √<br />
a2 /a 0 , 3 √<br />
a1 /a 0 , ζ 3 ). Then, G := Gal(L/É) is the Galois group<br />
acting on the 27 lines.<br />
We have the subgroup G 1 := Gal(L/K ) which is isomorphic to/3by consequence<br />
of assumption (∗). We choose a generator σ of G 1 .<br />
Finally, we write τ for the involution ζ 3 ↦→ ζ3 2 fixing the three third roots.<br />
Then, 〈σ, τ〉 ∼ = S 3 .<br />
6.12. Proposition. –––– The restriction<br />
is not the zero map.<br />
H 1 (G, Pic(XÉ)) −→ H 1 (〈σ〉, Pic(XÉ))
132 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
Proof. The assumption (∗) guarantees that [É( 3 √<br />
a3 /a 0 , ζ 3 ) :É] = 6. The adjunctions<br />
of 3 √<br />
a1 /a 0 <strong>and</strong> 3 √<br />
a2 /a 0 might or might not lead to further cubic extensions.<br />
Thus, there will be three cases to distinguish, #G may be either 6,<br />
or 18, or 54.<br />
We take for S the free group generated by the 27 lines <strong>and</strong> use the canonical<br />
isomorphisms H 1 (G, Pic(XÉ)) ∼ = Hom((NG S ∩ S 0 )/N G S 0 ,É/) <strong>and</strong><br />
H 1 (〈σ〉, Pic(XÉ)) ∼ = Hom((N 〈σ〉 S ∩ S 0 )/N 〈σ〉 S 0 ,É/). The restriction is then<br />
induced by the norm map N G/〈σ〉 : N 〈σ〉 S → N G S.<br />
We will start with the divisor D := div ( (T 0 + (a 1 /a 0 ) 1 3 T1 )/(T 0 + (a 2 /a 0 ) 1 3 T2 ) )<br />
which is the norm of L 2 (0, 0) − L 1 (0, 0). We have to show that N G/〈σ〉 D is not<br />
the norm N G of a principal divisor.<br />
First case: #G = 54.<br />
This is the generic case. We find that<br />
N G/〈σ〉 D = 6 div ( (a 0 T 3<br />
0 + a 1 T 3<br />
1 )/(a 0 T 3<br />
0 + a 2 T 3<br />
2 ) )<br />
= 6 ∑<br />
(m,n)∈(/3) 2<br />
L 2 (m, n) − 6<br />
∑<br />
(m,n)∈(/3) 2<br />
L 1 (m, n) .<br />
On the other h<strong>and</strong>, the group S 0 of all principal divisors in S is generated by the<br />
pairwise differences of all triangles [Man,кVI,ÄÑѺ].<br />
We have<br />
N G ∆ 1 kn = 18 ∑L k (i, j) ,<br />
(i,j)∈(/3) 2<br />
N G ∆km 2 = 18 ∑L k (i, j) ,<br />
(i,j)∈(/3) 2<br />
N G ∆ 3 abc = 6<br />
∑<br />
(i,j)∈(/3) 2<br />
L 0 (i, j) + 6<br />
∑<br />
(i,j)∈(/3) 2<br />
L 1 (i, j) + 6<br />
<strong>and</strong> see that N G/〈σ〉 D is not generated by these elements.<br />
Second case: #G = 18.<br />
∑<br />
(i,j)∈(/3) 2<br />
L 2 (i, j)<br />
Assume without restriction that a 0 = a 1 <strong>and</strong> a 2 /a 0 is not a cube inÉ∗ .<br />
Then, we find<br />
N G/〈σ〉 D = div ( (T0 3 + T1 3 )6 /(T0 3 + (a 2 /a 0 ) 3 T2 3 )2)<br />
= 6[L 2 (0, 0) + L 2 (0, 1) + L 2 (0, 2)] − 2 ∑L 1 (i, j) .<br />
(i,j)∈(/3) 2<br />
On the other h<strong>and</strong>,<br />
⎧<br />
⎪⎨<br />
N G L k (m, n) =<br />
⎪⎩<br />
2 ∑ L k (i, j) if k ≠ 2 ,<br />
(i,j)∈(/3) 2<br />
6 ∑<br />
j∈/3L k (m, j) if k = 2 .
Sec. 6] THE “FIRST CASE” OF DIAGONAL CUBIC SURFACES 133<br />
Hence,<br />
N G ∆ 1 kn = 6 ∑L k (i, j),<br />
⎧<br />
⎪⎨<br />
N G ∆ 2 km =<br />
⎪⎩<br />
N G ∆ 3 abc = 2<br />
(i,j)∈(/3) 2<br />
6 ∑ L k (i, j) if k ≠ 2 ,<br />
(i,j)∈(/3) 2<br />
18 ∑<br />
j∈/3L k (m, j) if k = 2 ,<br />
∑<br />
(i,j)∈(/3) 2<br />
L 0 (i, j) + 2<br />
∑<br />
(i,j)∈(/3) 2<br />
L 1 (i, j) + 6 ∑ 2 (a, j) .<br />
j∈/3L<br />
Factoring modulo all summ<strong>and</strong>s of type L 2 <strong>and</strong> ignoring the round brackets,<br />
these expressions are 54L 0 , 54L 1 , <strong>and</strong> 18L 0 + 18L 1 . They do not generate<br />
(−18)L 1 .<br />
Third case: #G = 6.<br />
Here, we find that<br />
On the other h<strong>and</strong>,<br />
N G/〈σ〉 D = 2 div ( (T 0 + (a 1 /a 0 ) 1 3 T1 )/(T 0 + (a 2 /a 0 ) 1 3 T2 ) )<br />
= 2 ∑<br />
i∈/3L 2 (0, i) − 2 ∑<br />
i∈/3L 1 (0, i).<br />
N G ∆ 1 kn = 2 ∑<br />
i∈/3L k (0, i) + 2 ∑<br />
i∈/3L k (1, i) + 2 ∑<br />
i∈/3L k (2, i)<br />
N G ∆ 2 km = 6 ∑<br />
i∈/3L k (m, i)<br />
N G ∆ 3 abc = 2 ∑<br />
i∈/3L 0 (b − a, i) + 2 ∑<br />
i∈/3L 1 (b, i) + 2 ∑<br />
i∈/3L 2 (a, i) .<br />
Ignoring the round brackets, those are 18L k for k ∈ {0, 1, 2}<strong>and</strong> 6L 0 +6L 1 +6L 2<br />
which do not generate 6L 2 − 6L 1 .<br />
□<br />
6.13. Corollary. –––– The restriction map<br />
H 1 (G, Pic(XÉ)) −→ H 1 (〈σ, τ〉, Pic(XÉ))<br />
is an isomorphism. The <strong>groups</strong> are isomorphic to/3.<br />
Proof. H 1 (〈σ〉, Pic(XÉ)) is purely 3-torsion. Sir P. Swinnerton Dyer’s list<br />
(cf. II.8.24) shows it is either/3or (/3) 2 . Further, we know that the<br />
restriction H 1 (G, Pic(XÉ)) → H 1 (〈σ〉, Pic(XÉ)) is not the zero map. Thus, the<br />
list implies that H 1 (G, Pic(XÉ)), too, is nothing but/3or (/3) 2 .
134 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
The restriction H 1 (G, Pic(XÉ)) → H 1 (〈σ〉, Pic(XÉ)) factors via<br />
H 1 (G, Pic(XÉ)) → H 1 (〈σ, τ〉, Pic(XÉ)).<br />
Thus, the latter is not the zero map, either. Again, a view on the list makes sure<br />
that H 1 (〈σ, τ〉, Pic(XÉ)) may be only/3or (/3) 2 .<br />
At this point, recall that (/3) 2 occurs only when the group acting nontrivially<br />
on Pic(XÉ) is of order three. Since τ ∈ 〈σ, τ〉 ⊂ G is of order two <strong>and</strong><br />
acting non-trivially, we are not in this case.<br />
The assertion follows.<br />
6.14. Proposition. –––– The restriction map<br />
H 1 (G, Pic(XÉ)) −→ H 1 (〈σ〉, Pic(XÉ)) τ<br />
is an isomorphism. The <strong>groups</strong> are isomorphic to/3.<br />
Proof. It suffices to show that the restriction map<br />
H 1 (〈σ, τ〉, Pic(XÉ)) −→ H 1 (〈σ〉, Pic(XÉ)) τ<br />
is an isomorphism. We know, the group on the left h<strong>and</strong> side is isomorphic<br />
to/3. The group on the right h<strong>and</strong> side is clearly a 3-torsion group.<br />
We consider the inflation-restriction spectral sequence<br />
E p,q<br />
2<br />
:= H p (〈τ〉, H q (〈σ〉, Pic(XÉ))) =⇒ H n (〈σ, τ〉, Pic(XÉ))<br />
<strong>and</strong> obtain the following exact sequence of terms of lower order,<br />
0 −→ H 1 (〈τ〉, Pic(XÉ) σ ) −→ H 1 (〈σ, τ〉, Pic(XÉ)) −→<br />
−→ H 1 (〈σ〉, Pic(XÉ)) τ −→ H 2 (〈τ〉, Pic(XÉ) σ ) .<br />
Since the homomorphism considered is encircled by 2-torsion <strong>groups</strong>, the proof<br />
is complete.<br />
→É/<br />
□<br />
iii. Some observations. —<br />
6.15. Lemma. –––– The image of the Manin map<br />
ev: Br(X ) × X (É) is contained in 1 3/.<br />
□<br />
Proof. By Proposition 2.3.b.i) <strong>and</strong> iii), the Manin map is additive in the first<br />
variable <strong>and</strong> factors via Br(X )/π ∗ Br(K ) ∼ =/3.<br />
□
Sec. 6] THE “FIRST CASE” OF DIAGONAL CUBIC SURFACES 135<br />
6.16. Lemma. –––– Let α ∈ Br(X ) be any Brauer class <strong>and</strong> (C, x 0 ) ⊂ XÉp 0<br />
be a<br />
smooth elliptic curve.<br />
Then, one of the following two statements is true.<br />
i) α| C ∈ Br(C )/π ∗ Br(SpecÉ) is zero. Then, ev νp0 (α, ·) is constant on C (Ép 0<br />
).<br />
ii) α| C ∈ Br(C )/π ∗ Br(SpecÉ) is non-zero.<br />
Then, there is a surjective group homomorphism g : C (Ép 0<br />
) → 1 3/such that<br />
for every x ∈ C (Ép 0<br />
).<br />
ev νp0 (α, x) = g(x) + ev νp0 (α, x 0 )<br />
Proof. i) is clear.<br />
ii) follows directly from Theorem II.8.16.<br />
□<br />
6.17. Lemma. –––– Let p 0 ≠ 3 be a prime number, (C, O) an elliptic curve<br />
overÉp 0<br />
3/<br />
, <strong>and</strong> C a minimal model of C overp 0<br />
. Assume that there are no<br />
Ép 0<br />
-valued <strong>points</strong> on C with singular reduction.<br />
Then, every continuous group homomorphism<br />
g : C (Ép 0<br />
) −→ 1<br />
factors via the reduction map C (Ép 0<br />
) → C (p 0<br />
) modulo p 0 .<br />
Proof. Continuity means that g factors via the reduction C (Ép 0<br />
) → C (/p0)<br />
n<br />
for a certain n ∈Æ. The/p0-valued n <strong>points</strong> on C reducing to the neutral<br />
element form a p 0 -group.<br />
□<br />
6.18. Proposition. –––– Suppose that p 0 ≠ 3. Then, for α ∈ Br(X ), the local<br />
evaluation map ev νp0 (α, ·) factors via<br />
red: X (Ép 0<br />
) −→ E p (p 0<br />
)<br />
(x 0 : x 1 : x 2 : x 3 ) ↦→ ((x 0 mod p 0 ) : (x 1 mod p 0 ) : (x 2 mod p 0 )) .<br />
Proof. Let (x 0 : x 1 : x 2 : x 3 ) ∈ X (Ép 0<br />
). We may assume without restriction that<br />
x 0 , x 1 , x 2 , x 3 ∈p 0<br />
are coprime. Then, assumption (∗) implies that x 0 , x 1 , or x 2<br />
is a unit. Assume, again without restriction, that x 2 is a unit. Then, x 0 <strong>and</strong> x 1<br />
can not both be multiples of p 0 . Assume p 0 ∤x 1 .<br />
Hensel’s lemma ensures that there exists a unique t ∈p 0<br />
such that<br />
t ≡ x 2 (mod p 0 ) <strong>and</strong> (x 0 : x 1 : t : 0) ∈ X (Ép 0<br />
). We claim that<br />
ev νp0 (α, (x 0 : x 1 : x 2 : x 3 )) = ev νp0 (α, (x 0 : x 1 : t : 0)) .<br />
Proof of the claim: Both <strong>points</strong> are contained in the intersection of X with the<br />
hyperplane “x 1 T 0 − x 0 T 1 = 0”. This is a genus one curve C. It is given by the
136 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
equation<br />
[a 0 (x 0 /x 1 ) 3 + a 1 ]T 3<br />
1 + a 2 T 3<br />
2 + a 3 T 3<br />
3 = 0.<br />
The assumptions imply that the first coefficient is not divisible by p 0 .<br />
As p 0 |a 3 , the curve C has bad reduction at p 0 . The equation above defines a<br />
minimal model C ofC overp. Thereare noÉp 0<br />
-valued <strong>points</strong> with singular reduction.<br />
C p0<br />
3/<br />
consists of three lines meeting in the singular point (0 : 0 : 0 : 1).<br />
It may happen that all three lines are defined overp 0<br />
or that one line is defined<br />
overp 0<br />
<strong>and</strong> the other two overp 2 <strong>and</strong> conjugate to each other.<br />
We choose a basepoint x ∈ C (Ép 0<br />
). Then, by Lemma 6.15,<br />
ev νp0 (α, ·)| C (Ép 0<br />
) : C (Ép 0<br />
) −→ 1<br />
differs from a continuous homomorphism of <strong>groups</strong> just by a constant summ<strong>and</strong>.<br />
Lemma 6.17 shows that ev νp0 (α, ·)| C (Ép 0<br />
) factors via the reduction<br />
map C (Ép 0<br />
) → C reg<br />
p 0<br />
(p 0<br />
).<br />
If only one line of C p0 is defined overp 0<br />
then we have #C reg<br />
p 0<br />
(p 0<br />
) = p 0 .<br />
Every group homomorphism to 3/is 1 constant which implies the claim in<br />
this case.<br />
Otherwise, we have #C reg<br />
p 0<br />
(p 0<br />
) = 3p 0 , each line having exactly p 0 smooth <strong>points</strong>.<br />
Let x ′ be the third point of intersection of the line tangent to C in x with C.<br />
Then, the group structure on C (Ép 0<br />
) has the property that P+Q+R = 2·x+x ′<br />
if <strong>and</strong> only if P, Q, <strong>and</strong> R are collinear. In particular, P + Q is the third point<br />
on the line through R <strong>and</strong> the neutral element.<br />
The reductions of three collinear <strong>points</strong> are either belonging to the same line<br />
of C p0 or to three different lines. By consequence, the subgroup of C reg<br />
p 0<br />
(p 0<br />
) of<br />
order p 0 is provided by the line containing the neutral element. The three lines<br />
form its cosets.<br />
As the reductions of (x 0 : x 1 : x 2 : x 3 ) <strong>and</strong> (x 0 : x 1 : t : 0) are on the same line,<br />
this implies the claim.<br />
Now, we apply Theorem II.8.16 <strong>and</strong> Lemma 6.17 to the elliptic curve E given<br />
by T 3 = 0. This shows<br />
ev νp0<br />
(<br />
α, (x0 : x 1 : t : 0) ) = g ( (x 0 mod p 0 ) : (x 1 mod p 0 ) : (t mod p 0 ) )<br />
for a map g : E p0 (p 0<br />
) → 3/. 1 Since (t mod p 0 ) = (x 2 mod p 0 ), the proof<br />
is complete.<br />
□<br />
6.19. Remark. –––– In the case p 0 ≠ 3, the only assertion still to be proven is<br />
that ev νp0 (α, ·)| E(Ép 0<br />
) is non-constant. For that, according to Theorem II.8.16,<br />
it suffices to show that α| E ∈ Br(E)/π ∗ Br(SpecÉ) is non-zero. We will verify<br />
this in the next subsection in Proposition 6.24.
Sec. 6] THE “FIRST CASE” OF DIAGONAL CUBIC SURFACES 137<br />
The next lemma, although true in general, will be relevant for us only in the<br />
case p 0 = 3.<br />
6.20. Lemma. –––– Let p be an arbitrary prime number <strong>and</strong> α ∈ Br(X ) be any<br />
Brauer class. Then, there exists a positive integer n such that ev νp (α, x) depends only<br />
on the reduction x (pn )<br />
∈ X (/p n).<br />
Proof. This follows directly from the continuity of ev νp with respect to the<br />
right argument.<br />
□<br />
iv. Evaluating on the elliptic curve. —<br />
6.21. Notation. –––– For the fields K <strong>and</strong> L as in 6.11, we fix a valuation ν of K<br />
lying above ν p0 <strong>and</strong> an extension w of ν to L.<br />
6.22. Lemma. –––– Fix an isomorphism ι: H 2 (〈σ〉,) ∼ =/3. Under the<br />
periodicity isomorphism<br />
H 1 (〈σ〉, Pic(X Kν<br />
))<br />
∼=<br />
−→ Ĥ −1 (〈σ〉, Pic(X Kν<br />
))<br />
= (N Div(X Lw ) ∩ Div 0 (X Kν ))/N Div 0 (X Lw )<br />
induced by ι, a representative of a generator is given by<br />
Proof. In fact,<br />
f := (T 0 + (a 1 /a 0 ) 1 3 T1 )/(T 0 + (a 2 /a 0 ) 1 3 T2 ) .<br />
div( f ) = L 2 (0, 0) + L 2 (0, 1) + L 2 (0, 2) − L 1 (0, 0) − L 1 (0, 1) − L 2 (0, 2)<br />
= N (L 2 (0, 0) − L 1 (0, 0)) .<br />
Further, the norms of the principal divisors are generated by the pairwise differences<br />
of<br />
N ∆ 1 kn = ∑<br />
i∈/3L k (0, i) + ∑<br />
i∈/3L k (1, i) + ∑<br />
i∈/3L k (2, i) ,<br />
N ∆ 2 km = 3 ∑<br />
i∈/3L k (m, i) ,<br />
N ∆ 3 abc = ∑ 0 (b − a, i) + ∑ 1 (b, i) + ∑ 2 (a, i) .<br />
i∈/3L i∈/3L i∈/3L Ignoring the round brackets, these elements are 9L k for k ∈ {0, 1, 2}<br />
<strong>and</strong> 3L 0 + 3L 1 + 3L 2 . They do not generate 3L 2 − 3L 1 .<br />
□<br />
6.23. Remark. –––– Theperiodicityisomorphism iscompatiblewiththe action<br />
of the involution τ. As f is τ-invariant, it represents a non-zero cohomology<br />
class in H 1 (〈σ〉, Pic(X Kν<br />
)) τ .
138 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
6.24. Proposition. –––– Assume p 0 ≠ 3. Then, the pull-back<br />
is not the zero map.<br />
Br(X )/π ∗ Br(SpecÉ) −→ Br(EÉp 0<br />
)/π ∗ Br(SpecÉp 0<br />
)<br />
Proof. It suffices to show that the restriction<br />
Br(X )/π ∗ Br(SpecÉ) −→ Br(E Kν )/π ∗ Br(SpecK ν )<br />
is not the zero map.<br />
It factors via [Br(X Kν )/π ∗ Br(SpecK ν )] τ ∼ = H 1 (〈σ〉, Pic(X Kν<br />
)) τ for which,<br />
by Lemma 6.22, we have a generator f in explicit form.<br />
It is, therefore, sufficient to show that there exists a point x ∈ E(K ν ) such<br />
that f (x) ∈ K ∗ ν is not in the image of the norm map N : L ∗ w → K ∗ ν.<br />
As L w /K ν is totally ramified, this follows directly from the lemma below.<br />
□<br />
6.25. Lemma. –––– Let p ≠ 3 be a prime number, K be a finite field extension<br />
ofÉp, <strong>and</strong> L/K a totally ramified Galois extension of degree three. Further, let C<br />
be the genus one curve over K given by x 3 + y 3 + z 3 = 0.<br />
Then, the range of the <strong>rational</strong> function f := (x + y)/(x + z) on C (K ) contains<br />
an element which is not a norm under N : L ∗ → K ∗ .<br />
Proof. The residue field k := O K /m K is finite <strong>and</strong> #k ≡ 1 (mod 3). We have<br />
to show that the range of the <strong>rational</strong> function f = (x + y)/(x + z) on C (k)<br />
contains a non-cube in k ∗ .<br />
For this, we write P := ((−1) : 1 : 0) <strong>and</strong> Q := ((−1) : 0 : 1).<br />
Then, div f = 3(P) − 3(Q). Choosing an arbitrary point O ∈ C (k) as the<br />
neutral element fixes a group law on C (k). We have the 3-division point P − Q<br />
on C.<br />
Consider the isogeny<br />
i : C ′ ·3<br />
:= C/〈P − Q〉 −→ C.<br />
i corresponds to the field extension k(C ′ ) = k(C ) ( √ )<br />
3 f /k(C ). We claim that<br />
→<br />
i ( C ′ (k) ) ∪ {P, Q} = {x ∈ C (k) | f (x) ∈ (k ∗ ) 3 } ∪ {P, Q} .<br />
Indeed, togiveapointx ∈ C (k)isequivalenttogivingavaluationν x : k(C )<br />
such that, for the corresponding residue field, one has O x /m x<br />
∼ = k. x belongs to<br />
the image of i(k) if <strong>and</strong> only if this valuation splits completely in k(C ′ ).<br />
Thus, our task is to show that i(k): C ′ (k) → C (k) is not surjective. Note that<br />
C (k) contains at least the nine <strong>points</strong> given by (1 : (−ζ3 i ) : 0) <strong>and</strong> permutations.
Sec. 6] THE “FIRST CASE” OF DIAGONAL CUBIC SURFACES 139<br />
The exact sequence of Gal(k/k)-modules<br />
induces a long exact sequence<br />
0 −→ µ 3 −→ C ′ (k) −→ C (k) −→ 0<br />
C ′ (k) −→ C (k) −→ k ∗ /(k ∗ ) 3 −→ 0 .<br />
Observe, by Hasse’s bound, every genus one curve has a point over k. Thus,<br />
there are no non-trivial torsors over C ′ <strong>and</strong> we have H 1( Gal(k/k), C ′ (k) ) = 0.<br />
The assertion follows.<br />
□<br />
v. Completing the proof for p 0 = 3. —<br />
6.26. Observation. –––– We have K ν =É3(ζ 3 ) or K ν =É3(ζ 3 , 3 √<br />
2).<br />
√ 3√<br />
Proof. By construction, K ν =É3(ζ 3 , 3 a, b) for a, b units inÉ∗ 3 . Modulo third<br />
powers, units inÉ∗ √ 3 fall into √ three classes, represented by 1, 2, <strong>and</strong> 4, respectively.<br />
AsÉ3(ζ 3 , 3 2) =É3(ζ 3 , 3 4), the assertion follows.<br />
□<br />
6.27. Notation. –––– For F ∈3, we denote the genus one curve, given by the<br />
equation T 3 = FT 0 on X, by E F . In particular, E 0 = E in our previous notation.<br />
The projection of E F to the first three coordinates is “Ax 3 0 + a 1 x 3 1 + a 2 x 3 2 = 0”<br />
for A := a 0 + F 3 a 3 . The assumptions 3 ∤a 0 <strong>and</strong> 3|a 3 make sure that A is a<br />
3-adic unit.<br />
We write E F for the model of E F over Spec3 given by T 3 = FT 0 on X .<br />
6.28. Proposition. –––– For every F ∈3, the pull-back<br />
is not the zero map.<br />
Br(X )/π ∗ Br(SpecÉ) −→ Br(E FÉ3 )/π∗ Br(SpecÉ3)<br />
Proof. It suffices to show that the restriction<br />
is not the zero map.<br />
Br(X )/π ∗ Br(SpecÉ) −→ Br(E F K ν<br />
)/π ∗ Br(SpecK ν )<br />
It factors via [Br(X Kν )/π ∗ Br(SpecK ν )] τ ∼ = H 1 (〈σ〉, Pic(X Kν<br />
)) τ for which,<br />
by Lemma 6.22, we have a generator [ f ] in explicit form.<br />
It is, therefore, sufficient to show that there exists a point x ∈ E F (K ν ) such that<br />
f (x) ∈ Kν ∗ is not in the image of the norm map N : L ∗ w → Kν. ∗ Since a 2 /a 1 is a<br />
cube in Kν ∗ , this follows directly from the lemma below.<br />
□
140 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
6.29. Lemma. –––– Let A ∈ O Kν be different from zero <strong>and</strong> C be the genus one<br />
curve over K ν given by Ax 3 + y 3 + z 3 = 0.<br />
Then, the range of the <strong>rational</strong> function f := (x + y)/(x + z) on C (K ν ) contains<br />
an element which is not a norm under N : L ∗ w → K ∗ ν.<br />
Proof. First step: Description of the norms.<br />
Let π be a uniformizing element of K ν . Then, according to [Se62, Chap. V,<br />
Proposition 5], there exists a positive integer t such that the homomorphism<br />
N n : (1 + m n L w<br />
)/(1 + m n+1<br />
L w<br />
) −→ ( 1 + (π) n) / ( 1 + (π) n+1)<br />
is a bijection for n < t <strong>and</strong> three-to-one for n = t. In particular, there is a<br />
unit α ∈ O ∗ K ν<br />
such that every element<br />
is not a norm.<br />
For t, there is the formula<br />
w ≡ 1 + απ t (mod π t+1 )<br />
t = ν L(D Lw /K ν<br />
)<br />
2<br />
where D Lw /K ν<br />
denotes the different of L w /K ν .<br />
Second step: Calculation of the different.<br />
Case 1: K ν =É3(ζ 3 , 3 √<br />
2).<br />
Then, L w =É3(ζ 3 , 3 √<br />
2,<br />
3√<br />
3). For the discriminants, one calculates<br />
d Lw /É3 = (3)37 <strong>and</strong> d Kν /É3 = (3)7 . This results in ν L (D Lw /K ν<br />
) = 16 <strong>and</strong> t = 7.<br />
For comparison, ν K (3) = 6.<br />
Case 2: K ν =É3(ζ 3 ).<br />
− 1<br />
Then, L w =É3(ζ 3 , 3 √<br />
3),É3(ζ 3 , 3 √<br />
6), orÉ3(ζ 3 , 3 √<br />
12).<br />
For each possibility the discriminant is the same, d Lw /É3 = (3)11 . Together with<br />
d Kν /É3 = (3), this shows ν L(D Lw /K ν<br />
) = 8 <strong>and</strong> t = 3.<br />
For comparison, ν K (3) = 2.<br />
Third step: Construction of the point.<br />
We claim that, on C, there is a K-valued point (x : y : z) such that x = π t ,<br />
y ≡ −1/α (mod π), <strong>and</strong> z ≡ 1/α (mod π). Then, the assertion follows since<br />
is not a norm.<br />
x + y<br />
x + z ≡ πt − 1/α<br />
π t + 1/α ≡ απt − 1<br />
απ t + 1 ≡ −(1 + απt ) (mod π t+1 )
Sec. 6] THE “FIRST CASE” OF DIAGONAL CUBIC SURFACES 141<br />
To show the existence of the point, we choose y ≡ −1/α (mod π), arbitrarily.<br />
Then, the equation<br />
g(Z) := Z 3 + (y 3 + Aπ 3t ) = 0<br />
has a solution Z ≡ −y (mod π) by Hensel’s lemma since<br />
ν K (g(−y)) = ν K (Aπ 3t ) ≥ 3t ,<br />
ν K (g ′ (−y)) = ν K (3y 2 ) = ν K (3) ,<br />
<strong>and</strong> 3t > 2ν K (3).<br />
□<br />
6.30. Proposition. –––– Assume that a 2 /a 1 ∈É∗ 3 is a non-cube <strong>and</strong> let<br />
t = (t 0 : t 1 : t 2 : t 3 ) ∈ X (3) be a point with t 0 ≠ 0. Choose n ∈Æsuch<br />
that ev νp (α, x) depends only on the reduction x (pn )<br />
∈ X (/p n).<br />
Then, for any F ∈3 such that (F mod 3) = t 3 /t 0 , one has<br />
#{y ∈ E F (/3 n) | y = t, ev ν3 (α, y)=0}<br />
= #{y ∈ E F (/3 n) | y = t, ev ν3 (α, y)=1/3}<br />
= #{y ∈ E F (/3 n) | y = t, ev ν3 (α, y)=2/3} .<br />
Proof. Proposition 6.28 shows that α| E ∈ Br(E FÉ3<br />
FÉ3 )/π∗ Br(SpecÉ3) is different<br />
from zero. Thus, α| E yields a surjective homomorphism of <strong>groups</strong><br />
FÉ3<br />
E F (É3) → 3/. 1 From this, we immediately see<br />
#{y ∈ E F (/3 n) | ev ν3 (α, y) = 0} = #{y ∈ E F (/3 n) | ev ν3 (α, y) = 1/3}<br />
We have to show the same for <strong>points</strong> reducing to t.<br />
= #{y ∈ E F (/3 n) | ev ν3 (α, y) = 2/3} .<br />
A unit u ∈3 is a cube if <strong>and</strong> only if u ≡ ±1 (mod 9). Permuting coordinates<br />
<strong>and</strong> changing a sign, if necessary, we may therefore assume that a 1 ≡ 1 (mod 9)<br />
<strong>and</strong> a 2 ≡ 7 (mod 9).<br />
E F is given by Ax 3 0 + a 1 x 3 1 + a 2 x 3 2 = 0 for A := a 0 + F 3 a 3 . In principle, there<br />
are three cases.<br />
First case: A ≡ ±7 (mod 9).<br />
We may assume A ≡ 7 (mod 9). Then, modulo 3, allÉ3-valued <strong>points</strong> on E F<br />
reduce to (1 : 0 : (−1)). Hence, the assertion is true in this case.<br />
Second case: A ≡ ±4 (mod 9).<br />
This is impossible as 4x 3 0 + 1x 3 1 + 7x 3 2 = 0 allows no solutions inÉ3 3 except<br />
for (0, 0, 0).<br />
Third case: A ≡ ±1 (mod 9).
142 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
Assume without restriction that A ≡ 1 (mod 9). AÉ3-valued point on E F<br />
may reduce either to (1 : (−1) : 0) or to (1 : 1 : 1).<br />
Take (1 : (−a) : 0) ∈ E F (É3), fora thethirdrootofA/a 1 , astheneutralelement.<br />
Then, Lemma 6.31 shows that the reduction map E F (É3) → E F (3) gives rise<br />
to a surjective group homomorphism E F (É3) →/2.<br />
As 2 is prime relative to 3, the asserted equidistribution of ev ν3 (α, ·) holds in<br />
each class, separately.<br />
□<br />
6.31. Lemma. –––– Consider the elliptic curve C overÉ3, given by<br />
Take (1 : (−1) : 0) as its basepoint.<br />
x 3 + y 3 + 7z 3 = 0 .<br />
C has a minimal Weierstraß model such that C 0 (É3), the group of <strong>points</strong> with<br />
non-singular reduction, coincides with the set of the <strong>points</strong> reducing (naively)<br />
to (1 : (−1) : 0). Further, C (É3)/C 0 (É3) ∼ =/2.<br />
Proof. The substitutions<br />
lead to the Weierstraß equation<br />
x ′ := −21z , y ′ := 63<br />
2 (x − y) , z′ := x + y<br />
z ′ y ′2 = x ′3 − 72 3 3<br />
4 z′3 .<br />
If the reduction of (x : y : z) is (1 : 1 : 1) then (x ′ : y ′ : z ′ ) reduces to (0 : 0 : 1).<br />
This is the cusp.<br />
On the other h<strong>and</strong>, consider the case that x = 1, z = 3k, <strong>and</strong>, therefore,<br />
y ≡ −1−7·3 2 k 3 (mod 27). Dividing the substitution formulas by 63, we obtain<br />
x ′ = −k, y ′ ≡ 1+ 7 2 32 k 3 (mod 27), i.e., y ′ ≡ 1 (mod 9), <strong>and</strong>z ′ = −k 3 (mod 3).<br />
In particular, y ′ is always a unit. The reduction of (x ′ : y ′ : z ′ ) is never equal to<br />
the cusp.<br />
It remains to show that C (É3)/C 0 (É3) ∼ =/2<strong>and</strong> that the Weierstraß model<br />
found is minimal. For this, we follow Tate’s algorithm [Ta75].<br />
We have the affine equation<br />
Y 2 = X 3 − 72 3 3<br />
4 .<br />
According to [Ta75, Summary], we are in case 6). The polynomial P is given<br />
by P(T ) = T 3 − 72 . It has a triple zero modulo 3.<br />
4<br />
We observe that 72<br />
≡ 1 (mod 9) <strong>and</strong> writeu for the 3-adic unit such 4 thatu3 = 72<br />
. 4<br />
The substitution X ′ := X − 3u leads to Y 2 = X ′3 + 9uX ′2 + 27u 2 X ′ .
Sec. 6] THE “FIRST CASE” OF DIAGONAL CUBIC SURFACES 143<br />
J. Tate’s equation (8.1) becomes<br />
Y 2 2 = 9X 3 2 + 9uX 2 2 + 3u 2 X 2 .<br />
Here, Y =: 9Y 2 <strong>and</strong> X ′ =: 9X 2 . Equation (9.1) is<br />
3Y 2 3 = 3X 3 2 + 3uX 2 2 + u 2 X 2<br />
for Y 2 =: 3Y 3 . As u 2 is a unit, we see that the Weierstraß model is minimal <strong>and</strong><br />
of Kodaira type III ∗ . [Ta75, table on p. 46] indicates C (É3)/C 0 (É3) ∼ =/2.<br />
□<br />
6.32. Proposition. –––– Let X be any diagonal cubic surface fulfilling (∗).<br />
Choose n ∈Æsuch that ev νp (α, x) depends only on the reduction<br />
x (pn )<br />
∈ X (/p n).<br />
Then, for every t = (t 0 : t 1 : t 2 : t 3 ) ∈ X (3), we have<br />
#{y ∈ X (/3 n) | y = t, ev ν3 (α, y)=0}<br />
= #{y ∈ X (/3 n) | y = t, ev ν3 (α, y)=1/3}<br />
= #{y ∈ X (/3 n) | y = t, ev ν3 (α, y)=2/3} .<br />
Proof. The case a 0 ≡ a 1 ≡ a 2 (mod 9) is treated in Example 6.33.<br />
We may therefore assume that two of the quotients a 0 /a 1 , a 1 /a 2 , <strong>and</strong> a 2 /a 0 ,<br />
say a 1 /a 2 <strong>and</strong> a 2 /a 0 , are non-cubes inÉ∗ 3 . It is impossible that t 0 = t 1 = 0.<br />
Again without restriction, assume t 0 ≠ 0.<br />
We consider the projection<br />
π : X −→ P 1 ,<br />
(x 0 : x 1 : x 2 : x 3 ) ↦→ (x 3 : x 0 ) .<br />
On aÉ3-valued point x reducing to t, the bi<strong>rational</strong> map π is defined.<br />
We have π(x) = (F : 1) for some F ∈3 such that (F mod 3) = t 3 /t 0 .<br />
By Proposition 6.30, the asserted equality is true in every fiber of π. It is<br />
therefore true in general.<br />
□<br />
6.33. Example. –––– Let d ∈<strong>and</strong> consider the cubic surface overÉgiven by<br />
T0 3 + T1 3 + T2 3 + 3dT3 3 = 0 .<br />
=/3<br />
By the results shown, we have<br />
Br(X )/π ∗ Br(SpecÉ) = H 1( Gal(L/K ), Pic(X L ) ) ∼
144 AN APPLICATION: THE BRAUER-MANIN OBSTRUCTION [Chap. III<br />
for L =É(ζ 3 , 3 √<br />
3d) which is equal to the field of definition of the 27 lines<br />
on X. As K =É(ζ 3 ) does not contain any cubic extensions, the Brauer-<br />
Manin obstruction may be described completely explicitly, without relying on<br />
Lichtenbaum’s duality.<br />
In fact,<br />
H 1( Gal(L/É), Pic(X L ) ) ∼ = H<br />
1 ( 〈σ〉, Pic(X L ) ) τ<br />
= [Br(XÉ(ζ 3 ))/π ∗ Br(SpecÉ(ζ 3 ))] τ .<br />
For the latter, we have the explicit generator [ f ] for f := (T 0 +T 1 )/(T 0 +T 2 ).<br />
This means, for x = (x 0 : x 1 : x 2 : x 3 ) ∈ X (É3(ζ 3 )) <strong>and</strong> ν ∈ Val(É(ζ 3 )) the<br />
extension of ν 3 , we have<br />
(<br />
ev ν ([ f ], x) = i ν (t0 + t 1 )/(t 0 + t 2 ) ) .<br />
Here, i ν is the homomorphism<br />
É3(ζ 3 ) ∗ −→É3(ζ 3 ) ∗ /NL ∗ ∼=<br />
w −→ Ĥ 0 (〈σ〉, L ∗ ∼=<br />
w) −→ H 2 (〈σ〉, L ∗ w) inv ν<br />
−→É/<br />
for w the extension of ν to L.<br />
However, we want to considerÉ3-valued <strong>points</strong>, notÉ3(ζ 3 )-valued ones.<br />
Thus, let α ∈ Br(X ) be a Brauer class mapping to [ f ] under restriction.<br />
Then, for x ∈ X (É3),<br />
2 · ev ν3 (α, x) = ev ν ([ f ], x) .<br />
Fortunately, multiplication by 2 is an automorphism of 1 3/.<br />
On X, two kinds ofÉ3-valued <strong>points</strong> may be distinguished.<br />
First kind: x 0 , x 1 , <strong>and</strong> x 2 are units.<br />
Then, x 0 ≡ x 1 ≡ x 2 (mod 3) <strong>and</strong> d ∤ 3. We have<br />
ev ν3 (α, x) = 0 ⇐⇒ x 1 ≡ x 2 (mod 9) .<br />
Second kind: Among x 0 , x 1 , <strong>and</strong> x 2 , there is an element which is a multiple of 3.<br />
Suppose 3|x 0 . Then, x 1 ≡ −x 2 (mod 3). We have<br />
ev ν3 (α, x) = 0 ⇐⇒ 2x 0 + x 1 + x 2 (mod 9) .<br />
6.34. Corollary. –––– If x 3 0 +x 3 1 +x 3 2+3x 3 3 = 0 for x 0 , . . . , x 3 ∈then we have<br />
the following non-trivial congruences.<br />
i) If x 0 , x 1 , <strong>and</strong> x 2 are units then<br />
x 0 ≡ x 1 ≡ x 2 (mod 9) .
Sec. 6] THE “FIRST CASE” OF DIAGONAL CUBIC SURFACES 145<br />
ii) If 3|x 0 then<br />
x 0 ≡ 0 (mod 9) .<br />
Proof. This is the particular case d = 1. Then, L =É(ζ 3 , 3 √<br />
3) is unramified at<br />
every prime different from 3. As ν p<br />
(<br />
(t0 + t 1 )/(t 0 + t 2 ) ) is always divisible by 3,<br />
we have ev νp (α, x) = 0 for every p ≠ 3.<br />
The congruences obtained from ev ν3 (α, x) = 0 imply the congruences asserted.<br />
□<br />
6.35. Remark. –––– Thus, we have shown that, for the equation<br />
T 3<br />
0 + T 3<br />
1 + T 3<br />
2 + 3T 3<br />
3 = 0<br />
overÉ, the restrictions coming from the Brauer-Manin obstruction are exactly<br />
the congruences obtained by R. Heath-Brown [H-B92a] in 1993.<br />
Note that Heath-Brown used rather different methods. His main tool is the law<br />
of cubic reciprocity.
PART B<br />
HEIGHTS
CHAPTER IV<br />
THE CONCEPT OF A HEIGHT<br />
Equations are just the boring part of mathematics. I attempt to see things<br />
in terms of geometry.<br />
STEPHEN HAWKING (A Biography (2005) by Kristine Larsen, p. 43)<br />
1. The naive height on the projective space overÉ<br />
1.1. –––– Heights have been studied by number theorists for a very long time.<br />
A height is a function measuring the size or, more precisely, the arithmetic complexity<br />
of certain objects. These objects are classically solutions of Diophantine<br />
equations or <strong>rational</strong> <strong>points</strong> on an algebraic variety. A height then might answer<br />
the question how many bits one would need in order to store the solution or<br />
the point on a computer.<br />
More recently, starting with G. Faltings’ ideas of heights on moduli spaces, it<br />
became more common to consider heights for more complicated objects such<br />
as cycles.<br />
1.2. Definition. –––– For (x 0 : . . . : x n ) ∈ P n (É), put<br />
Here,<br />
H naive (x 0 : . . . : x n ) := max<br />
i=0,...,n |x′ i| .<br />
(x ′ 0 : . . . : x′ n ) = (x 0 : . . . : x n )<br />
such that all x ′ i are integers <strong>and</strong> gcd(x′ 0 , . . . , x′ n ) = 1.<br />
The function H naive : P nÉ(É) →Êis called the naive height.<br />
1.3. Fact (Fundamental finiteness). —– For every C ∈Ê, there are only finitely<br />
many <strong>points</strong> x ∈ P nÉ(É) such that<br />
H naive (x) < C .<br />
Proof. For eachcomponentof (x 0, ′ . . . , x n), ′ wehave −C < x i ′ < C. Thus, there<br />
are only finitely many choices.<br />
□
150 THE CONCEPT OF A HEIGHT [Chap. IV<br />
1.4. –––– The naive height is probably the simplest function one might think<br />
of which fulfills the fundamental finiteness property. For more general height<br />
functions, the fundamental finiteness property will always be required.<br />
1.5. Remark. –––– Let X ⊂ P nÉbe a subvariety. Then, everyÉ-<strong>rational</strong> point<br />
on X is also aÉ-<strong>rational</strong> point on P n .<br />
We call the restriction of H naive to X (É) the naive height on X.<br />
Is is obvious that the naive height on X fulfills the fundamental finiteness property.<br />
1.6. Notation. –––– i) For a prime number p, denote by ‖ . ‖ p the normalized<br />
p-adic valuation.<br />
I.e., for x ∈É\{0}, let x = ±2 v 2 ·3<br />
v<br />
v pk<br />
3 · . . . · p<br />
k<br />
prime factors. Then, put<br />
‖x‖ p := p −v p<br />
.<br />
Further, ‖0‖ p := 0.<br />
be its decomposition into<br />
ii) We use ‖ . ‖ ∞ as an alternative notation for the usual absolute value,<br />
{ x if x ≥ 0 ,<br />
‖x‖ ∞ :=<br />
−x if x < 0 .<br />
1.7. Fact. –––– For p a prime number or infinity, ‖ . ‖ p is indeed a valuation.<br />
This means, for x, y ∈É,<br />
i) ‖x‖ p ≥ 0,<br />
ii) ‖x‖ p = 0 if <strong>and</strong> only if x = 0,<br />
iii) ‖xy‖ p = ‖x‖ p · ‖y‖ p ,<br />
iv) ‖x + y‖ p ≤ ‖x‖ p + ‖y‖ p .<br />
For p ≠ ∞, one even has that ‖x + y‖ p ≤ max{‖x‖ p , ‖y‖ p }.<br />
1.8. Fact (Product formula). —– For x ∈É\{0}, one has<br />
∏ ‖x‖ p = 1 .<br />
p prime or ∞<br />
1.9. Lemma. –––– Let (x 0 : . . . : x n ) ∈ P nÉ(É). Then,<br />
[<br />
H naive (x 0 : . . . : x n ) =<br />
∏<br />
p prime or ∞<br />
]<br />
max ‖x i‖ p<br />
i=0,...,n<br />
Proof. The product formula implies that the right h<strong>and</strong> side remains unchanged<br />
when (x 0 : . . . : x n ) is replaced by (λx 0 : . . . : λx n ) for λ ≠ 0. Thus, we may<br />
suppose that all x i are integers <strong>and</strong> gcd(x 0 , . . . , x n ) = 1.<br />
.
Sec. 2] GENERALIZATION TO NUMBER FIELDS 151<br />
The assumptions that x 0 , . . . , x n ∈<strong>and</strong> gcd(x 0 , . . . , x n ) = 1 imply<br />
max<br />
i=0,...,n ‖x i‖ p = 1<br />
for every prime number p. Hence, the formula on the right h<strong>and</strong> side may be<br />
simplified to max i=0,...,n |x i |.<br />
This is precisely the assertion.<br />
1.10. Remark. –––– Despite being so primitive, the naive height is actually<br />
sufficient for most applications. For example, in the numerical experiments<br />
described in Part C, we will always work with the naive height.<br />
□<br />
2. Generalization to number fields<br />
i. The definition. —<br />
2.1. –––– Let K be a number field. I.e., K is a finite extension ofÉ. It is<br />
well known from algebraic number theory [Cas67] that there is a set Val(K ) of<br />
normalized valuations ‖ . ‖ ν on K satisfying the following conditions.<br />
a) The functions ‖ . ‖ ν : K →Êare indeed valuations. I.e., for x, y ∈É,<br />
i) ‖x‖ ν ≥ 0,<br />
ii) ‖x‖ ν = 0 if <strong>and</strong> only if x = 0,<br />
iii) ‖xy‖ ν = ‖x‖ ν · ‖y‖ ν ,<br />
iv) ‖x + y‖ ν ≤ ‖x‖ ν + ‖y‖ ν .<br />
b) There is the product formula<br />
∏ ‖x‖ ν = 1<br />
∈Æ<br />
ν∈Val(K )<br />
for every x ∈ K \{0}.<br />
2.2. –––– Further, for L/K a degree d extension of number fields, the sets<br />
Val(K ) <strong>and</strong> Val(L) are compatible in the following sense.<br />
i) For every ‖ . ‖ µ ∈ Val(L), there are a valuation ‖ . ‖ ν ∈ Val(K ) <strong>and</strong> d µ<br />
such that<br />
‖ . ‖ µ | K = ‖ . ‖ d µ<br />
ν .<br />
In this case, it is said that ‖ . ‖ µ is lying above ‖ . ‖ ν .<br />
ii) For every ‖ . ‖ ν ∈ Val(K ), there are only a finite number of valuations<br />
‖ . ‖ µ1 , . . . , ‖ . ‖ µl ∈ Val(L) lying above ‖ . ‖ ν . One has<br />
l<br />
∑<br />
i=1<br />
d µi = d .
152 THE CONCEPT OF A HEIGHT [Chap. IV<br />
This implies<br />
for every x ∈ K.<br />
‖x‖ d ν =<br />
l<br />
∏<br />
i=1<br />
‖x‖ µi<br />
2.3. –––– A valuation is called archimedean if it lies above the valuation ‖ . ‖ ∞<br />
ofÉ. Otherwise, it is called non-archimedean.<br />
If ν is non-archimedean then one has the ultrametric triangle inequality<br />
‖x + y‖ ν ≤ max{‖x‖ ν , ‖y‖ ν } .<br />
2.4. Definition. –––– Let K be a number field of degree d. Then, for<br />
(x 0 : . . . : x n ) ∈ P n (K ), one puts<br />
[ ] 1<br />
H naive (x 0 : . . . : x n ) := ∏ max ‖x d<br />
i‖ ν .<br />
i=0,...,n<br />
ν∈Val(K )<br />
This height function is the number field version of the naive height on P n . It is<br />
usually called the absolute height.<br />
2.5. Lemma. –––– Let K be a number field <strong>and</strong> (x 0 : . . . : x n ) ∈ P n (K ). Further,<br />
let L ⊃ K be a finite extension.<br />
Then, the absolute height H naive (x 0 : . . . : x n ) remains unchanged when<br />
(x 0 : . . . : x n ) is considered as an L-<strong>rational</strong> point.<br />
Proof. Put d ′ := [L : K ]. Then, by the properties of the valuations, we have<br />
[<br />
H naive (x 0 : . . . : x n ) = ∏<br />
ν∈Val(K )<br />
= ∏<br />
ν∈Val(K )<br />
= ∏<br />
µ∈Val(L)<br />
] 1<br />
max ‖x d<br />
i‖ ν<br />
i=0,...,n<br />
[<br />
∏<br />
µ∈Val(L)<br />
µ above ν<br />
[<br />
] 1<br />
max ‖x dd<br />
i‖ ′<br />
µ<br />
i=0,...,n<br />
] 1<br />
max ‖x dd<br />
i‖ ′<br />
µ<br />
i=0,...,n<br />
which is exactly the formula for H naive (x 0 : . . . : x n ) considered as an<br />
L-<strong>rational</strong> point.<br />
□<br />
2.6. Proposition (Northcott). —– Let B, D ∈Ê. Then,<br />
is a finite set.<br />
{x ∈ P n (É) | x ∈ P n (K ) for [K :É] < D <strong>and</strong> H naive (x) < B}<br />
Proof. We may work with the number fields K of a fixed degree d.
Sec. 2] GENERALIZATION TO NUMBER FIELDS 153<br />
For x, we choose homogeneous coordinates such that some coordinate equals 1.<br />
Then, it is clear that, for every valuation ‖ . ‖ ν <strong>and</strong> every index i, we have<br />
max{‖x 0 ‖ ν , . . . , ‖x n ‖ ν } ≥ max{1, ‖x i ‖ ν } .<br />
Multiplying over all ν <strong>and</strong> taking the d-th root therefore shows<br />
Hence, it suffices to verify that the set<br />
H naive (x 0 : . . . : x n ) ≥ H naive (1 : x i ) .<br />
{(1 : x) ∈ P 1 (É) | (1 : x) ∈ P 1 (K ) for [K :É] = d <strong>and</strong> H naive (1 : x) < B}<br />
is finite.<br />
For this, wewriteσ 1 (x), . . . , σ d (x) ∈Éfor theelementarysymmetricfunctions<br />
in the conjugates of x. According to Vieta, x is a zero of the polynomial<br />
F x (T ) := T d − σ 1 (x)T d−1 + σ 2 (x)T d−2 + . . . + (−1) d σ d (x) ∈É[T ] .<br />
Lemma 2.7 shows that<br />
(<br />
H naive 1 : σr (x) ) ( d<br />
)<br />
( d<br />
)<br />
≤ · H naive (1 : x) rd ≤ · B rd .<br />
r r<br />
Thus, by Fact 1.3, we know that for each σ r (x) there are only finitely many possibilities.<br />
Hence, there are only finitely many possibilities for the polynomial F x<br />
<strong>and</strong>, therefore, only finitely many possibilities for x.<br />
This completes the proof.<br />
2.7. Lemma. –––– Let K be number field of degree d which is Galois overÉ.<br />
For x ∈ K, denote by σ 1 (x), . . . , σ d (x) ∈Éthe elementary symmetric functions<br />
in the conjugates of x. Then,<br />
(<br />
H naive 1 : σr (x) ) ( d<br />
)<br />
≤ · H naive (1 : x) rd .<br />
r<br />
Proof. We denote the conjugates of x by x 1 , . . . , x d . Let ν be any valuation<br />
of K. Then,<br />
∥ ∥∥∥∥ν<br />
‖σ r (x)‖ ν =<br />
∥<br />
∑ x i1 · . . . · x ir<br />
1≤i 1
154 THE CONCEPT OF A HEIGHT [Chap. IV<br />
By consequence,<br />
max{1, ‖σ r (x)‖ ν } ≤ C (d, r, ν) · max max{1, ‖x i ‖ ν } r<br />
i<br />
≤ C (d, r, ν) · ∏ max{1, ‖x i ‖ ν } r .<br />
i<br />
Multiplying over all valuations of K yields<br />
(<br />
H naive 1 : σr (x) )d ( d<br />
) d<br />
≤ ·<br />
r<br />
∏ H naive (1 : x i ) rd<br />
i<br />
( d<br />
) d<br />
= · Hnaive (1 : x) rd2<br />
r<br />
since conjugate <strong>points</strong> have the same height. This inequality is equivalent to<br />
the assertion.<br />
□<br />
2.8. –––– For most theoretical investigations <strong>and</strong>, may be, some practical ones,<br />
it is actually better to work with the logarithm of the absolute height.<br />
2.9. Definition. –––– For (x 0 : . . . : x n ) ∈ P n (É), put<br />
h naive (x 0 : . . . : x n ) := log H naive (x 0 : . . . : x n ) .<br />
The height function h naive is called the logarithmic height.<br />
ii. Application: The height defined by an ample invertible sheaf. —<br />
2.10. Definition. –––– Let X be a projective variety over a number field K<br />
<strong>and</strong> L be an ample invertible sheaf on X.<br />
Then, a height function h L induced by L is given as follows. Let m ∈Æbe such<br />
that L ⊗m is very ample. Put<br />
h L (x) := 1 m h (<br />
naive iL ⊗m(x) )<br />
for x ∈ P a closed point. Here, i L ⊗m : P → P N denotes a closed embedding<br />
defined by L ⊗m .<br />
2.11. Lemma. –––– Let X be a projective variety over a number field K <strong>and</strong> L<br />
be an ample invertible sheaf on X. If h (1)<br />
L <strong>and</strong> h(2) L are two height functions induced<br />
by L then there is a constant C such that<br />
for every x ∈ X (É).<br />
| h (1)<br />
L (x) − h(2) L (x)| ≤ C
Sec. 3] GEOMETRIC INTERPRETATION 155<br />
Proof. First, it is clear that the k-uple embedding fulfills<br />
h naive<br />
(<br />
ρk (x) ) = k·h naive (x)<br />
for every k ∈Æ<strong>and</strong> x ∈ P N (É). Hence, if h L is defined using L ⊗m then the<br />
same height function may be defined using L ⊗km .<br />
We may therefore assume that h (1)<br />
L<br />
<strong>and</strong> h(2) L<br />
are defined using the same tensor<br />
L<br />
differ by an automor-<br />
⊗m<br />
power of L . Then, the two embeddings i (1)<br />
L<br />
<strong>and</strong> i (2)<br />
⊗m<br />
phism ι: P N → P N .<br />
We have to verify that<br />
| h naive<br />
(<br />
ι(x)<br />
) − hnaive (x)| ≤ C<br />
for every x ∈ P N (É). ι is explicitly given by<br />
(x 0 : . . . : x N ) ↦→ (x ′ 0 : . . . : x ′ N )<br />
for some a = (a ij ) ij ∈ M N +1 (K ).<br />
:= (a 00 x 0 + . . .+ a 0N x N ) : . . . : (a N 0 x 0 + . . .+ a NN x N )<br />
For all but finitely many ν ∈ Val(K ), we have that ν is non-archimedean<br />
<strong>and</strong> ‖a ij ‖ ν = 1 for all i <strong>and</strong> j. This yields<br />
max<br />
i=0,...,N ‖x′ i‖ w ≤ max ‖x i‖ w<br />
i=0,...,N<br />
for every number field L <strong>and</strong> every w ∈ Val(L) lying above such a valuation.<br />
For every other valuation of K, we find a constant D ν such that<br />
The desired inequality<br />
follows immediately from this.<br />
max<br />
i=0,...,N ‖x′ i‖ w ≤ D d w/ν<br />
ν<br />
h naive<br />
(<br />
ι(x)<br />
) − hnaive (x) ≤<br />
· max<br />
i=0,...,N ‖x i‖ w .<br />
∑<br />
ν∈Val(K )<br />
log D ν<br />
For the inequality the other way round, one may work with ι −1 instead of ι.<br />
□<br />
3. Geometric interpretation<br />
3.1. –––– Heights are more closely related to modern algebraic geometry than<br />
this might seem from the definitions given in the sections above. The geometric<br />
interpretation of the concept of a height is the starting point of arithmetic<br />
intersection theory, a fascinating theory which we may only touch upon here.
156 THE CONCEPT OF A HEIGHT [Chap. IV<br />
The reader is advised to consult the articles [G/S 90, G/S 92] of H. Gillet<br />
<strong>and</strong> C. Soulé, the textbook [S/A/B/K], <strong>and</strong> the references therein. More recent<br />
developments are described by J. I. Burgos Gil, J. Kramer, <strong>and</strong> U. Kühn<br />
in [B/K/K].<br />
The particular case of the arithmetic intersection theory on a curve over a<br />
number field had been developed earlier. The articles of S. Yu. Arakelov [Ara]<br />
<strong>and</strong> G. Faltings [Fa84] present the point of view taken before around 1990 which<br />
is a bit different from today’s.<br />
3.2. –––– We return to the assumption that the ground field is K =É. This is<br />
done mainly in order to ease notation. The theory would work equally well<br />
over an arbitrary number field.<br />
3.3. Definition. –––– An arithmetic variety is an integral scheme which is projective<br />
<strong>and</strong> flat over Spec.<br />
3.4. Definition. –––– A hermitian line bundle on an arithmetic variety X is a<br />
pair (L , ‖ . ‖) consisting of an invertible sheaf L ∈ Pic(X ) <strong>and</strong> a continuous<br />
hermitian metric on the line bundle Lassociated with L on the complex<br />
space X ().<br />
All hermitian line bundles on X form an abelian group which is denoted<br />
by ̂Pic C 0 (X ).<br />
For the group of all smooth (C ∞ ) hermitian line bundles, one writes ̂Pic(X ).<br />
̂Pic(X ) is usually called the arithmetic Picard group.<br />
3.5. Example. –––– The projective space P nover Specis an arithmetic variety.<br />
On P n, there is the tautological invertible sheaf O(1). It is generated by<br />
the global sections X 0 , . . . , X n ∈ Γ(P n, O(1)).<br />
Let us show two explicit hermitian metrics on O(1)making O(1) to a hermitian<br />
line bundle.<br />
i) The Fubini-Study metric is given by<br />
for l ∈ Γ(P n, O(1)).<br />
‖l‖ FS :=<br />
√<br />
(X0<br />
ii) The minimum metric is given by<br />
‖l‖ min := min<br />
l<br />
1<br />
) 2<br />
+ . . . + ( X n<br />
) 2<br />
l<br />
i=0,...,n<br />
∣ l ∣∣∣<br />
∣ .<br />
X i
Sec. 3] GEOMETRIC INTERPRETATION 157<br />
In the case of the Fubini-Study metric, the formula should be interpreted in<br />
such a way that ‖l‖ FS (x) = 0 at all <strong>points</strong> x ∈ P n () where l vanishes. Similarly,<br />
in the definition of ‖l‖ min (x), the minimum is actually to be taken over<br />
all i such that x i ≠ 0.<br />
The Fubini-Study metric is smooth (C ∞ ). The minimum metric is continuous<br />
but not even C 1 .<br />
3.6. Definition (Arakelov degree). —– For a hermitian line bundle (L , ‖ . ‖)<br />
on Spec, define its Arakelov degree by<br />
̂deg (L , ‖ . ‖) := log #(L /sL ) − log ‖s‖<br />
for s ∈ Γ(Spec, L ) a non-zero section.<br />
3.7. Remark. –––– L is associated with a free-module L of rank one. Lis<br />
the-vector space L⊗. Thus, we work with a non-zero element s ∈ L <strong>and</strong><br />
with the norm ‖s ⊗ 1‖.<br />
The definition is independent of the choice of s since<br />
#(L /nsL ) = n·#(L /sL ) ,<br />
therefore log #(L /nsL ) = log #(L /sL ) + log n.<br />
log ‖ns‖ = log ‖s‖ + logn.<br />
On the other h<strong>and</strong>,<br />
3.8. Fact. –––– Let L be a hermitian line bundle on Spec. If there is a section<br />
s ∈ Γ(Spec, L ) of norm less than one then ̂deg L > 0.<br />
Proof. This follows immediately from the definition.<br />
3.9. Fact. –––– For two hermitian line bundles L 1 , L 2 on Spec, one has<br />
̂deg (L 1 ⊗ L 2 ) = ̂deg L 1 + ̂deg L 2 .<br />
Proof. Apply the definition to arbitrary non-zero sections s 1 ∈ Γ(Spec, L 1 ),<br />
s 2 ∈ Γ(Spec, L 2 ), <strong>and</strong> to their tensor product s 1 ⊗s 2 ∈ Γ(Spec, L 1 ⊗L 2 ). □<br />
3.10. –––– AÉ-<strong>rational</strong> point on the projective space is actually a morphism<br />
x : SpecÉ→P nof schemes. The valuative criterion for properness implies<br />
that there is a unique extension to a morphism from Spec,<br />
x<br />
SpecÉ n P<br />
<br />
<br />
∩ <br />
<br />
<br />
x<br />
Spec.<br />
□
158 THE CONCEPT OF A HEIGHT [Chap. IV<br />
3.11. Observation (Arakelov). —– Let x ∈ P n (É) <strong>and</strong> x : Spec→P nits<br />
extension to Spec. Then,<br />
h naive (x) = ̂deg ( x ∗ (O(1), ‖ . ‖ min ) ) .<br />
Proof. Write x in coordinates as (x 0 : . . . : x n ) for x 0 , . . . , x n ∈. Then, the<br />
pull-back of X i ∈ Γ(P n, O(1)) is x i ∈ Γ(Spec, x ∗ (O(1))). Since O(1) is generated<br />
by the global sections X i , we have<br />
x ∗ (O(1)) = 〈x 0 , . . . , x n 〉 = gcd(x 0 , . . . , x n ) ·.<br />
We choose an index i 0 such that x i0 ≠ 0. Using this section, we obtain<br />
̂deg (x ∗ (O(1)), ‖ . ‖) = log # ( x ∗ (O(1))/x i0 x ∗ (O(1)) ) − log ‖x i0 ‖<br />
which is exactly the logarithmic height of x.<br />
= log # ( gcd(x 0 , . . . , x n )/x i0) − log min<br />
∣ = log<br />
x i0<br />
∣gcd(x 0 , . . . , x n ) ∣ + log max<br />
x i ∣∣∣<br />
i=0,...,n<br />
∣x i0<br />
max |x i|<br />
i=0,...,n<br />
= log<br />
gcd(x 0 , . . . , x n )<br />
i=0,...,n<br />
∣ x i0 ∣∣∣<br />
∣ x i<br />
□<br />
3.12. –––– This motivates the following general definition.<br />
Definition. Let X be an arithmetic variety <strong>and</strong> L be a hermitian line bundle<br />
on X .<br />
Then, the height with respect to L of theÉ-<strong>rational</strong> point x ∈ X (É) is given by<br />
h L<br />
(x) = ̂deg ( x ∗ L ) .<br />
Here, x : Spec→X is the extension of x to Spec.<br />
3.13. Examples. –––– i) Let X = P n<strong>and</strong> let L be the invertible sheaf O(1)<br />
equipped with the minimum metric. Then, as shown in 3.11, the height with<br />
respect to L is the naive height.<br />
ii) Let X = P n<strong>and</strong> let L be the invertible sheaf O(1) equipped with the<br />
Fubini-Study metric. Then, the height with respect to L is the l 2 -height which<br />
is given by<br />
√<br />
h l 2(x 0 : . . . : x n ) := log x<br />
0 ′2 + . . . + x′2 n<br />
for<br />
(x ′ 0 : . . . : x′ n ) = (x 0 : . . . : x n )
Sec. 3] GEOMETRIC INTERPRETATION 159<br />
projective coordinates such that all x ′ i are integers <strong>and</strong> gcd(x ′ 0, . . . , x ′ n) = 1.<br />
3.14. Lemma. –––– Let f : X → Y be a morphism of arithmetic varieties <strong>and</strong><br />
L be a hermitian line bundle on Y .<br />
Then, for every x ∈ X (É),<br />
h f ∗ L (x) = h L ( f (x)) .<br />
Proof. Let x be the extension of x to Spec. Then, f ◦x is the extension of f (x)<br />
to Spec. Thus,<br />
h f ∗ L (x) = ̂deg ( x ∗ ( f ∗ L ) ) = ̂deg ( ( f ◦x) ∗ L ) = h L<br />
( f (x)) .<br />
□<br />
3.15. Proposition. –––– Let X be an arithmetic variety.<br />
a) Let ‖ . ‖ 1 <strong>and</strong> ‖ . ‖ 2 be hermitian metrics on one <strong>and</strong> the same invertible sheaf L .<br />
Then, there is a constant C such that<br />
for every x ∈ X (É).<br />
|h (L ,‖ .‖1 )(x) − h (L ,‖ .‖2 )(x)| < C<br />
b) Let L 1 <strong>and</strong> L 2 be two hermitian line bundles. Then, for every x ∈ X (É),<br />
h (L1 ⊗L 2 ) (x) = h L 1<br />
(x) + h L2<br />
(x) .<br />
c) LetL beahermitianlinebundlesuchthattheunderlyinginvertiblesheafisample.<br />
Then, for every C ∈Ê, there are only finitely many <strong>points</strong> x ∈ X (É) such that<br />
h L<br />
(x) < C .<br />
Proof. a) For x the extension of x to Spec, we have<br />
h (L ,‖ .‖1 )(x) − h (L ,‖ .‖2 )(x) = ̂deg ( x ∗ L , x ∗ ‖ . ‖ 1<br />
) − ̂deg<br />
(<br />
x ∗ L , x ∗ ‖ . ‖ 2<br />
)<br />
.<br />
Working with a non-zero section s ∈ Γ(Spec, x ∗ L ), we see that the latter difference<br />
is equal to<br />
log #(x ∗ L /sx ∗ L ) − log ‖s‖ 1 (x) − ( log #(x ∗ L /sx ∗ L ) − log ‖s‖ 2 (x) )<br />
= log ‖ . ‖ 2(x)<br />
‖ . ‖ 1 (x) .<br />
Since X () is compact, there is a positive constant D such that<br />
1<br />
D < ‖ . ‖ 2(x)<br />
‖ . ‖ 1 (x) < D<br />
for every x ∈ X (). This implies the claim.
160 THE CONCEPT OF A HEIGHT [Chap. IV<br />
b) Clearly, for every x ∈ X (É), one has<br />
h (L1 ⊗L 2 ) (x) = ̂deg ( x ∗ (L 1 ⊗L 2 ) ) = ̂deg ( (x ∗ L 1 )⊗(x ∗ L 2 ) )<br />
= ̂deg (x ∗ L 1 ) + ̂deg (x ∗ L 2 ) = h L1<br />
(x) + h L2<br />
(x) .<br />
c) There is some n ∈Æsuch that L ⊗n is very ample. Part b) shows that<br />
it suffices to verify the assertion for L ⊗n . Thus, we may assume that L is<br />
very ample.<br />
Let i : X ֒→ P n be the closed embedding induced by L . Then, L = i ∗ O(1).<br />
It follows from Tietze’s Theorem that there is a hermitian metric on O(1) such<br />
that L = i ∗ O(1). Then, h L<br />
(x) = h O(1)<br />
(i(x)). It, therefore, suffices to show<br />
fundamental finiteness for the height function h O(1)<br />
on P n (É).<br />
Part a) together with Observation 3.11 shows that h O(1)<br />
differs from h naive by a<br />
bounded summ<strong>and</strong>. Fact 1.3 yields the assertion.<br />
□<br />
4. Basic arithmetic intersection theory<br />
4.1. Remark. –––– It should be noticed that there is a strong formal analogy of<br />
the concept of a height on an arithmetic variety to the concept of a degree in<br />
algebraic geometry over a ground field. The only obvious difference is that the<br />
role of the sections of an invertible sheaf is now played by small sections, say, of<br />
norm less than one. Nevertheless, it seems that the height of a point is actually<br />
some sort of arithmetic intersection number.<br />
This is an idea which has been formalized first by S. Yu. Arakelov [Ara] for twodimensional<br />
arithmetic varieties <strong>and</strong> later by H. Gillet <strong>and</strong> C. Soulé [G/S 90]<br />
for arithmetic varieties of arbitrary dimension.<br />
Let us briefly recall the most important definitions <strong>and</strong> some fundamental results<br />
from arithmetic intersection theory.<br />
4.2. Definition (Gillet-Soulé). —– Let X be a regular arithmetic variety<br />
<strong>and</strong> i ∈Æ.<br />
i) Then, an arithmetic cycle of codimension i on X is a pair (Z, g Z ) consisting a<br />
codimensioni cycleZ ontheschemeX together withaGreen’scurrent g Z forZ.<br />
The latter means that g Z ∈ D p−1,p−1 (X ()) is a real current on the complex<br />
manifold X () invariant under the complex conjugation F ∞ : X () → X ()<br />
such that<br />
ω Z := 1<br />
2πi ∂∂g Z + δ Z<br />
is a smooth differential form on X (). Here, δ Z is the current given by<br />
integration along Z() ⊆ X ().
Sec. 4] BASIC ARITHMETIC INTERSECTION THEORY 161<br />
The abelian group of all codimension i arithmetic cycles on X is denoted<br />
by Ẑi (X ).<br />
ii) A codimension i arithmetic cycle on X is said to be linearly equivalent to<br />
zero if it belongs to the subgroup of Ẑi (X ) generated by all arithmetic cycles<br />
of types<br />
(0, ∂u)<br />
for u ∈ D p−2,p−1 (X ()),<br />
(0, ∂v)<br />
for v ∈ D p−1,p−2 (X ()), <strong>and</strong><br />
(div f , − log | f | 2 )<br />
for f a non-zero <strong>rational</strong> function on an integral subscheme of X of codimension<br />
i − 1.<br />
The abelian group of all codimension i arithmetic cycles on X which are<br />
linearly equivalent to zero is denoted by ̂R i (X ).<br />
iii) The i-th arithmetic Chow group of X is defined by<br />
ĈH i (X ) := Ẑi (X )/̂R i (X ) .<br />
=<br />
4.3. Remark. –––– − log | f | 2 is indeed a Green’s current for div f . This fact is<br />
classically known as the Poincaré-Lelong equation [G/H, p. 388].<br />
4.4. Example. –––– For every regular arithmetic variety X , we have<br />
ĈH 0 (X ) = CH 0 (X )<br />
since, on X (), there are no currents of type (−1, −1).<br />
4.5. Example-Definition. –––– The arithmetic degree<br />
is an isomorphism.<br />
̂deg : ĈH1 (Spec) −→Ê<br />
[ (∑<br />
i<br />
a i (p i ), r )] ↦→ 1 2 r + ∑a i log p i<br />
i<br />
Indeed, the homomorphism ̂deg is well-defined since (div n, − log |n| 2 ) is<br />
mapped to zero. It is injective <strong>and</strong> surjective as one has CH 1 (Spec) = 0<br />
for the usual Chow group <strong>and</strong> [D 0,0 (pt)] − =Ê.
162 THE CONCEPT OF A HEIGHT [Chap. IV<br />
4.6. Example. –––– Obviously,<br />
for i ≥ 2.<br />
ĈH i (Spec) = CH i (Spec) = 0<br />
4.7. Definition. –––– Let X be a regular arithmetic variety <strong>and</strong> L ∈ ̂Pic(X )<br />
be a hermitian line bundle. Then, the first arithmetic Chernclass of L is given by<br />
for a non-zero section s of L .<br />
ĉ 1 (L ) := [(div s, − log ‖s‖ 2 )] ∈ ĈH1 (X )<br />
4.8. Lemma. –––– For every regular arithmetic variety X , the first arithmetic<br />
Chern class provides an isomorphism<br />
ĉ 1 : ̂Pic(X ) −→ ĈH1 (X ) .<br />
Proof. It follows form the Poincaré-Lelong equation that the homomorphism<br />
ĉ 1 is well-defined.<br />
To show injectivity, assume [(div s, − log ‖s‖ 2 )] = 0 ∈ ĈH1 (X ) for a non-zero<br />
<strong>rational</strong> section s of a hermitian line bundle L . Since there are no (0, 0)-<br />
currents of types ∂u or ∂v, this means (div s, − log ‖s‖ 2 ) = (div f , − log ‖ f ‖ 2 )<br />
for a <strong>rational</strong> function f .<br />
Hence, s/f is a global section of L without zeroes or poles <strong>and</strong> of constant<br />
norm one. This implies that L represents the neutral element of ̂Pic(X ).<br />
For surjectivity, let any arithmetic 1-cycle (Z, g Z ) be given. The invertible<br />
sheaf O(Z) may be equipped with a smooth hermitian metric. This reduces the<br />
question to the case Z = 0.<br />
This means, we consider an arithmetic 1-cycle (0, g). I.e., g is a (0, 0)-current<br />
such that ω := ∂∂g is a smooth (1, 1)-form. We claim this implies already that<br />
g is smooth.<br />
Indeed, the ∂∂-lemma [G/H, p. 149] yields a smooth function g ′ such<br />
that ω = ∂∂g ′ . Hence, ∂∂(g ′ − g) = 0. This means that the current g ′ − g<br />
vanishes on all test forms of type ∂∂η. By the ∂∂-lemma, these are all d-closed<br />
forms of top degree. By consequence of Poincaré duality for de Rham cohomology,<br />
g ′ − g is the current associated with a constant function.<br />
For the arithmetic 1-cycle (0, g), we therefore have (0, g) = ĉ 1 (O X ) where O X<br />
is the structure sheaf equipped with the smooth hermitian metric given<br />
by ‖ f ‖ := e − g 2 | f |.<br />
□<br />
4.9. Remark. –––– For a hermitian line bundle L ∈ ̂Pic(Spec), its Arakelov<br />
degree as defined in the previous section is the same as ̂deg ( ĉ 1 (L ) ) .
)⊗É<br />
Sec. 4] BASIC ARITHMETIC INTERSECTION THEORY 163<br />
4.10. –––– H. Gillet <strong>and</strong> C. Soulé [G/S 90, Theorem 4.2.3] define an intersection<br />
product<br />
·: ĈHi (X ) × ĈHj (X ) −→ ĈHi+j (X<br />
with the following properties.<br />
i) “·” is commutative <strong>and</strong> associative.<br />
ii) If Y <strong>and</strong> Z are irreducible subvarieties meeting properly, i.e.,<br />
codim(Y ∩ Z) ≥ codim Y + codim Z ,<br />
then<br />
[(Y, g Y )] · [(Z, g Z )] = [Y ·Z, g Y ∗ g Z ] .<br />
Here, Y ·Z is the usual intersection cycle defined by the Tor-formula [Fu84,<br />
Section 20.4]. The ∗-product of two Green’s currents is morally given by<br />
the formula<br />
g Y ∗ g Z := g Y ·δ Z + ω Y ·g Z .<br />
This definition is obviously problematic as g Y ·δ Z is formally a product of<br />
two currents. It may be justified in the expected manner when g Y is a Green’s<br />
form of log type along Y [G/S 90, 2.1.3]. The latter is always fulfilled if Y is a<br />
cycle of codimension one.<br />
4.11. –––– For a morphism f : X → Y of regular arithmetic varieties,<br />
H. Gillet <strong>and</strong> C. Soulé [G/S 90, 4.4.3] define a pull-back homomorphism<br />
It has the properties below.<br />
f ∗ : ĈHi (Y ) −→ ĈHi (X ) .<br />
i) For a smooth hermitian line bundle L ∈ ̂Pic(Y ), one has<br />
ii) For y ∈ ĈHi (Y ) <strong>and</strong> z ∈ ĈHj (Y ),<br />
f ∗ ĉ 1 (L ) = ĉ 1 ( f ∗ L ) .<br />
f ∗ (y·z) = f ∗ y · f ∗ z .<br />
iii) For two morphisms f : X → Y <strong>and</strong> g : Y → Z of regular arithmetic<br />
varieties, the equality<br />
is true.<br />
(g ◦ f ) ∗ = f ∗ ◦ g ∗ : ĈHi (Z ) −→ ĈHi (X )
164 THE CONCEPT OF A HEIGHT [Chap. IV<br />
4.12. –––– For a morphism f : X → Y of regular arithmetic varieties such<br />
that fÉ: XÉ→YÉis smooth, there is a push-forward homomorphism<br />
with the following properties.<br />
i) For [(Z, g Z )] ∈ ĈHi (X ), one has<br />
f ∗ : ĈHi (X ) −→ ĈHi−dimX+dim Y (Y )<br />
f ∗ [(Z, g Z )] = [( f ∗ Z, f ∗ g Z )] .<br />
Here, f ∗ Z is the usual push-forward cycle [Fu84, Section 1.4] <strong>and</strong> f ∗ g Z is the<br />
push-forward of g Z as a current.<br />
ii) There is the projection formula<br />
for x ∈ ĈHi (X ) <strong>and</strong> y ∈ ĈHj (Y ).<br />
f ∗ (x · f ∗ y) = f ∗ x · y<br />
iii) For two morphisms f : X → Y <strong>and</strong> g : Y → Z of regular arithmetic<br />
varieties such that fÉ<strong>and</strong> gÉare smooth, one has that<br />
(g ◦ f ) ∗ = g ∗ ◦ f ∗ : ĈHi (X ) −→ ĈHi−dimX+dimZ (Z ) .<br />
4.13. Remarks. –––– It requires rather hard work to establish all the results<br />
described in 4.10 through 4.12. There are two different sorts of problems.<br />
i) On the scheme-theoretic side, it is already a problem to define the intersection<br />
of two cycles a ∈ CH i (X ) <strong>and</strong> b ∈ CH j (X ) in the ordinary Chow <strong>groups</strong>.<br />
The point is that on a scheme of finite type over, there is no moving lemma<br />
available except for divisors. Thus, in order to define the intersection product<br />
of two cycles, a highly indirect approach must be chosen.<br />
H. Gillet <strong>and</strong> C. Soulé use the fact that the Brown-Gersten-Quillen spectral<br />
sequence degenerates after tensoring withÉ. The Chow <strong>groups</strong> of X appear<br />
as part of the E 2 -terms while the spectral sequence converges versus the algebraic<br />
K-<strong>groups</strong> K ∗ (X ). The product structure on K ∗ (X ) provided by the<br />
tensor product of vector bundles then induces the desired intersection product<br />
·: CH i (X ) × CH j (X ) −→ CH i+j (X )⊗É.<br />
Many more details are given in [S/A/B/K, Chapter I].<br />
Today, there is a second approach to this intersection product which is based<br />
on the resolution of singularities by alterations found by A. J. de Jong [dJo].<br />
This method is rather indirect, too. It yields the same intersection product as<br />
the K-theory approach.
Sec. 5] AN INTERSECTION PRODUCT ON SINGULAR ARITHMETIC VARIETIES 165<br />
One would prefer to have an intersection product<br />
·: CH i (X ) × CH j (X ) −→ CH i+j (X ) .<br />
In other words, one would like to get rid of the tensor product withÉ.<br />
For schemes over, there is no such intersection product with reasonable<br />
properties known up to now.<br />
ii) On the other h<strong>and</strong>, it requires enormous technical efforts to justify the<br />
definitions made on the complex-analytic side. H. Gillet <strong>and</strong> C. Soulé show that<br />
for every arithmetic cycle there is an equivalent one which is of the form (Y, g Y )<br />
for g Y a Green’s form of log type along Y. This is necessary to give the ∗-product<br />
of Green’s currents a precise meaning.<br />
Still, a lot of work has to be done in order to verify all the properties of<br />
the intersection product which are asserted. Here, the hardest point is to<br />
prove associativity.<br />
4.14. –––– Over the years, a lot of progress has been made on this part of<br />
the theory. Several people tried to define the arithmetic intersection product in<br />
a more canonical manner. It turned out that the Gillet-Soulé intersection theory<br />
may be understood as a particular case of a more general framework. From such<br />
a more general point of view, the ∗-product of two Green’s currents appears as<br />
being induced by the cup product on Deligne-Beilinson cohomology. Such an<br />
approach therefore reduces commutativity <strong>and</strong> associativity of the intersection<br />
product into formalities.<br />
The clearest <strong>and</strong> most thorough framework for arithmetic intersection theory<br />
on regular arithmetic varieties seen up to now has been provided by J. I. Burgos<br />
Gil, J. Kramer, <strong>and</strong> U. Kühn [B/K/K].<br />
4.15. Remark. –––– One of the main drawbacks of the arithmetic intersection<br />
theory developed so far is that the push-forward is defined only for morphisms<br />
which are smooth on the generic fiber. Replacing in Definition 4.2 the requirement<br />
that ω Z has to be smooth by a slightly weaker condition, one obtains<br />
a theory which is free of this drawback. Such theories are due to J. I. Burgos<br />
Gil [Bu] <strong>and</strong> A. Moriwaki [Mori]. The reader may find a description<br />
in [B/K/K, Theorem 6.35].<br />
5. An intersection product on singular arithmetic varieties<br />
5.1. –––– For arbitrary, i.e., possibly singular, arithmetic varieties, arithmetic<br />
intersection theory is considerably less far developed. The following theorem is<br />
a consequence from the theory presented so far which is valid in more generality.
166 THE CONCEPT OF A HEIGHT [Chap. IV<br />
5.2. Theorem. –––– a) For every arithmetic variety π : X → Specsuch that its<br />
generic fiber XÉis smooth of dimension d, there is a unique-multilinear map<br />
pGS X : ̂Pic(X ) × . . . × ̂Pic(X ) −→Ê<br />
} {{ }<br />
(d+1) times<br />
satisfying the following conditions.<br />
i) If X is regular then<br />
p X GS (L 1, . . . , L d+1 ) = ̂deg ( π ∗<br />
(ĉ1 (L 1 ) · . . . · ĉ 1 (L d+1 ) ))<br />
for every L 1 , . . . , L d+1 ∈ ̂Pic(X ).<br />
ii) If f : X → Y is a generically finite morphism of arithmetic varieties such that<br />
XÉ<strong>and</strong> YÉare smooth then<br />
p X GS ( f ∗ L 1 , . . . , f ∗ L d+1 ) = deg f · p Y GS (L 1, . . . , L d+1 )<br />
for every L 1 , . . . , L d+1 ∈ ̂Pic(Y ).<br />
b) pGS X is symmetric.<br />
Proof. a) Let π : X → Specbe any arithmetic variety such that its generic<br />
fiber XÉis smooth of dimension d.<br />
Uniqueness: Let r : X ′ → X be a generically finite morphism of arithmetic<br />
varieties such that X ′ is regular. The existence of such a morphism is provided<br />
by the main result of [dJo]. Then, conditions i) <strong>and</strong> ii) imply that, necessarily,<br />
p X GS (L 1, . . . , L d+1 ) := 1<br />
degr ̂deg ( (π◦r) ∗<br />
(ĉ1 (r ∗ L 1 ) · . . . · ĉ 1 (r ∗ L d+1 ) )) .<br />
Existence: We take formula (∗) as the definition for p GS .<br />
One has to show that the definition does not depend on the choice of r.<br />
For this, suppose there are given two generically finite morphisms r ′ : X ′ → X<br />
<strong>and</strong> r ′′ : X ′′ → X such that X ′ <strong>and</strong> X ′′ are regular.<br />
The fiber product X ′ × X X ′′ is generically finite over X . Again by virtue<br />
of [dJo], there is a generically finite morphism X ′′′ → X ′ × X X ′′ such<br />
that X ′′′ is regular. Thus, we are reduced to the case that r ′′ = r ′ ◦ g<br />
for g : X ′′ → X ′ a morphism of regular schemes.<br />
Then, r ′′∗ L i = g ∗ r ′∗ L i for every i. We may now apply the results from<br />
the previous section as g is a morphism of regular arithmetic varieties.<br />
Thus, ĉ 1 (r ′′∗ L i ) = g ∗( ĉ 1 (r ′∗ L i ) ) <strong>and</strong><br />
ĉ 1 (r ′′∗ L 1 ) · . . . · ĉ 1 (r ′′∗ L d+1 ) = g ∗( ĉ 1 (r ′∗ L 1 ) · . . . · ĉ 1 (r ′∗ L d+1 ) ) .<br />
(∗)
Sec. 5] AN INTERSECTION PRODUCT ON SINGULAR ARITHMETIC VARIETIES 167<br />
The projection formula implies<br />
(π◦r ′′ ) ∗<br />
(<br />
g<br />
∗ ( ĉ 1 (r ′∗ L 1 ) · . . . · ĉ 1 (r ′∗ L d+1 ) ))<br />
= (π◦r ′ ◦g) ∗<br />
(<br />
g<br />
∗ ( ĉ 1 (r ′∗ L 1 ) · . . . · ĉ 1 (r ′∗ L d+1 ) ))<br />
= (π◦r ′ ) ∗ g ∗<br />
(<br />
g<br />
∗ ( ĉ 1 (r ′∗ L 1 ) · . . . · ĉ 1 (r ′∗ L d+1 ) ))<br />
= (π◦r ′ ) ∗<br />
[<br />
(g∗ 1) · (ĉ<br />
1 (r ′∗ L 1 ) · . . . · ĉ 1 (r ′∗ L d+1 ) )] .<br />
This yields the claim as g ∗ 1 = deg g.<br />
p X GS is-multilinear by definition.<br />
It fulfills condition i) as one may work with r = id in the case that X is regular.<br />
ii) Choose a generically finite morphism r : X ′ → X such that X ′ is regular.<br />
The definitions of p X GS ( f ∗ L 1 , . . . , f ∗ L d+1 ) <strong>and</strong> p Y GS (L 1, . . . , L d+1 ) using r<br />
<strong>and</strong> f ◦r, respectively, differ only by the degree divided out.<br />
b) This follows from formula (∗) together with the commutativity <strong>and</strong> associativity<br />
of the intersection product of H. Gillet <strong>and</strong> C. Soulé [cf. 4.10.i)]. □<br />
5.3. Definition. –––– Let X be an arithmetic variety <strong>and</strong> L ∈ ̂Pic(X ) be a<br />
hermitian line bundle.<br />
Then, for every cycle Z ∈ Z(XÉ) of dimension i, one defines its height by<br />
h BGS<br />
L<br />
(Z) := pZ GS(L | Z<br />
, . . . , L | } {{ Z<br />
) . }<br />
(i+1) times<br />
Here, Z is the Zariski closure of Z in X equipped with its induced reduced structure.<br />
5.4. Remark. –––– This definition for the height of a cycle of arbitrary<br />
dimension is, in slightly different form, due to J.-B. Bost, H. Gillet,<br />
<strong>and</strong> C. Soulé [B/G/S, Section 3.1.1]. It fulfills a positivity condition <strong>and</strong> a<br />
very strong fundamental finiteness statement.<br />
5.5. Theorem. –––– Let i ∈Æ, X be an arithmetic variety, <strong>and</strong> L ∈ ̂Pic(X ) be<br />
a hermitian line bundle.<br />
a) If the curvature form c 1 (L ) is positive <strong>and</strong> some positive power L ⊗n is generated<br />
by global sections of norm less than one then<br />
for every effective cycle Z on XÉ.<br />
h BGS<br />
L (Z) > 0
168 THE CONCEPT OF A HEIGHT [Chap. IV<br />
b) If L is ample on X then for every B ∈Êthere are only finitely many effective<br />
cycles Z ∈ Z i (XÉ) such that deg L<br />
(Z) < B <strong>and</strong><br />
h BGS<br />
L (Z) < B .<br />
Proof. These are [B/G/S, Proposition 3.2.4 <strong>and</strong> Theorem 3.2.5].<br />
□<br />
5.6. Remark. –––– There are rather few computational results for the Bost-<br />
Gillet-Soulé height h BGS<br />
L for cycles of positive dimension.<br />
One of them is the height of projective space itself. In [B/G/S, Section 3.3], it<br />
is shown that<br />
Z<br />
h (O(1),‖ .‖FS )(P n ) = h (O(1),‖ .‖FS )(P n−1 ) − log ‖x 0 ‖ FS ωFS n .<br />
As the latter integral may easily be computed to 1 2<br />
h (O(1),‖ .‖FS )(P n ) = 1 2<br />
n<br />
∑<br />
k=1<br />
k<br />
∑<br />
m=1<br />
P n ()<br />
n<br />
1<br />
∑<br />
m<br />
m=1<br />
1<br />
m = n + 1<br />
2<br />
n<br />
∑<br />
m=1<br />
, this leads to<br />
1<br />
m − n 2 .<br />
5.7. Definition. –––– Let X be an arithmetic variety, X its generic fiber,<br />
<strong>and</strong> L ∈ ̂Pic(X ) be a hermitian line bundle.<br />
Then, for a closed point x ∈ X, one defines its absolute height by<br />
1 ( )<br />
h L<br />
(x) :=<br />
[L :É] hBGS L (x) .<br />
Here, L is the field of definition of x. (x) denotes the 0-dimensional cycle on X<br />
associated with x.<br />
5.8. Remarks. –––– i) Theorem 5.2.a.ii) implies that this definition is independent<br />
of the choice of L.<br />
ii) Since x ∈ Z 1 (X ) is a one-dimensional cycle, this definition actually does not<br />
involve the arithmetic intersection product. Only the arithmetic degree is used.<br />
For this reason, there is an explicit formula for the dependence of the height on<br />
the choice of the metric. Indeed,<br />
h (L ,‖ .‖ ′ )(x) = h (L ,‖ .‖) (x) − 1<br />
[L :É] ∑<br />
σ : L→log ‖ . ‖′( σ(x) )<br />
‖ . ‖ ( σ(x) ) .<br />
In particular, there is a constant C such that<br />
| h (L ,‖.‖ ′ )(x) − h (L ,‖.‖) (x)| < C<br />
for every x ∈ X. This naturally generalizes Lemma 2.11 <strong>and</strong> Proposition 3.15.
Sec. 6] THE ARITHMETIC HILBERT-SAMUEL FORMULA 169<br />
6. The arithmetic Hilbert-Samuel formula<br />
6.1. Definition. –––– Let X be a Noetherian scheme <strong>and</strong> h: X →Êbe<br />
any function. For i ∈Æ, we define the i-th successive minimum of h by<br />
e i (h) :=<br />
sup<br />
inf<br />
codim X Y=i x∈X \Y<br />
h(x) .<br />
Here, Y runs through all (possibly reducible) closed subsets of X of codimension<br />
i. x runs through the closed <strong>points</strong>.<br />
6.2. Remark. –––– A priori, one has that e i (h) ∈ [−∞, ∞]. Further, directly<br />
from the definition, we see<br />
e 1 (h) ≥ e 2 (h) ≥ . . . ≥ e dim X+1 (h) = inf<br />
x∈X h(x) .<br />
6.3. Lemma. –––– Let X be an arithmetic variety the generic fiber of which is X.<br />
Let L ∈ ̂Pic(X ) <strong>and</strong> suppose that L is ample. Then,<br />
for every i ∈Æ.<br />
e i (h L<br />
) > −∞<br />
Proof. First, we note that it is sufficient to show e dim X+1 (h L<br />
) > −∞.<br />
For this, we may suppose that L is very ample. Then, L is generated by<br />
global sections s 1 , . . . , s k ∈ Γ(X , L ). Further, on L , there exists a hermitian<br />
metric ‖ . ‖ ′ such that c 1 (L , ‖ . ‖ ′ ) is positive. Scale ‖ . ‖ ′ by a positive constant<br />
factor such that ‖s 1 ‖ ′ , . . . , ‖s k ‖ ′ < 1.<br />
Then, by Theorem 5.5.a), h (L ,‖ .‖ ′ )(x) > 0 for every closed point x ∈ X. Remark<br />
5.8.ii) shows that there is a constant C such that<br />
h L<br />
(x) > C<br />
for every x ∈ X. Consequently, e dim X+1 (h L<br />
) > −∞.<br />
□<br />
6.4. Notation. –––– Let X be an arithmetic variety, X its generic fiber,<br />
(L , ‖ . ‖) ∈ ̂Pic(X ), <strong>and</strong> n ∈Æ.<br />
Further, let s ∈ Γ(X, LÉ) ⊗n be any global section. We define its norm by<br />
‖s‖ := sup ‖s‖(x) .<br />
x∈X ()<br />
This induces a norm on theÊ-vector space Γ(X, LÉ)⊗ÉÊ.<br />
⊗n<br />
On Γ(X, LÉ)⊗ÉÊ, ⊗n we consider the normalized Lebesgue measure µ n such<br />
that the unit ball<br />
B n := {s ∈ Γ(X, LÉ)⊗ÉÊ|‖s‖ ⊗n<br />
≤ 1}
É)⊗ÉÊ<br />
170 THE CONCEPT OF A HEIGHT [Chap. IV<br />
is of measure one.<br />
Further, we observe that<br />
Γ(X , L ⊗n ) ⊂ Γ(X, L ⊗n<br />
is a lattice of maximal rank.<br />
6.5. Theorem (Arithmetic Hilbert-Samuel formula). —–<br />
Let X be an arithmetic variety, X its generic fiber, <strong>and</strong> L ∈ ̂Pic(X ).<br />
Putd := dim X. Assume thatX is smooth <strong>and</strong> the curvature formc 1 (L ) is positive.<br />
Then,<br />
− log covol µn Γ(X , L ⊗n ) ∼ pX GS (L , . . . , L ) n d+1 .<br />
(d + 1)!<br />
Proof. When X is regular, this result may be deduced from the arithmetic<br />
Riemann-Roch Theorem proven by H. Gillet <strong>and</strong> C. Soulé [G/S 92].<br />
See also [Fa92]. This, in turn, is a very deep theorem. Its proof combines the<br />
work of several people including, besides H. Gillet <strong>and</strong> C. Soulé, D. Quillen,<br />
J.-M. Bismut, <strong>and</strong> G. Lebeau.<br />
A direct proof for the arithmetic Hilbert-Samuel formula which does not rely<br />
on the arithmetic Riemann-Roch Theorem has been provided by A. Abbes<br />
<strong>and</strong> T. Bouche [A/B]. This proof works for singular arithmetic varieties, too.<br />
□<br />
6.6. Corollary (Theorem of successive minima, cf. [Zh95b, (5.2)]). —–<br />
Let X be an arithmetic variety, X its generic fiber, <strong>and</strong> L ∈ ̂Pic(X ).<br />
Putd := dim X. Assume that X is smooth <strong>and</strong> the curvature formc 1 (L ) is positive.<br />
Then,<br />
e 1 (h L<br />
) ≥ pX GS (L , . . . , L )<br />
(d + 1) deg L<br />
X .<br />
Proof. Write (L , ‖ . ‖) for L .<br />
When one multiplies ‖ . ‖ by a constant factor e −C , the summ<strong>and</strong> C is added<br />
to the height function h L<br />
. Therefore, the term e 1 (h L<br />
) on the left h<strong>and</strong> side is<br />
changed by C, too. On the other h<strong>and</strong>, from Lemma 6.9, we see that the right<br />
h<strong>and</strong> side is changed by the same summ<strong>and</strong>.<br />
Consequently, it will suffice to show that e 1 (h L<br />
) ≥ 0 under the assumption<br />
p X GS (L , . . . , L ) > 0 .<br />
For this, we first observe that the usual Hilbert-Samuel formula yields a constant<br />
D such that<br />
dim Γ(X, L ⊗n É)⊗ÉÊ
Sec. 6] THE ARITHMETIC HILBERT-SAMUEL FORMULA 171<br />
Hence, by the arithmetic Hilbert-Samuel formula,<br />
⊗n dim Γ(X,L<br />
log [2<br />
É)⊗ÉÊ· covol µn Γ(X , L ⊗n )]<br />
= (log 2) dim Γ(X, LÉ)⊗ÉÊ+log ⊗n covol µn Γ(X , L ⊗n )<br />
< D(log 2)n d − 1 2<br />
< 0<br />
p X GS (L , . . . , L )<br />
(d + 1)!<br />
n d+1<br />
for n ≫ 0. Exponentiating, we see<br />
⊗n dim Γ(X,L<br />
2<br />
É)⊗ÉÊ· covol µn Γ(X , L ⊗n ) < 1 = vol B n .<br />
This shows that we may apply Minkowski’s lattice point theorem [B/S,<br />
кII, §4]. The unit ball B n contains a lattice point which is different from<br />
the origin.<br />
In other words, there is a non-zero section s ∈ Γ(X , L ⊗n ) such that ‖s‖ < 1.<br />
We now claim that h L<br />
(x) ≥ 0 for every x ∈ X \div s| X . This clearly suffices for<br />
the asserted inequality.<br />
To verify the claim, let L be the field of definition of the point x <strong>and</strong> x its<br />
Zariski closure in X . Then,<br />
h L<br />
(x) = 1 ( ) 1<br />
[L :É] hBGS L (x) =<br />
[L :É] px GS (L | x) .<br />
The subscheme x might be singular but there is a morphism i : Spec O L → x of<br />
schemes which is generically one-to-one. Theorem 5.2 shows<br />
p x GS(L | x ) = p SpecO L<br />
GS<br />
(i ∗ L ) = ̂deg π ∗ ĉ 1 (i ∗ L )<br />
for π : Spec O L → Specthe structural morphism.<br />
By our assumption, i # s ∈ Γ(Spec O L , i ∗ L ) is a non-zero section. Thus,<br />
ĉ 1 (i ∗ L ) = [(div i # s, − log ‖i # s‖ 2 )] .<br />
Here, div i # s ∈ Z 0 (Spec O L ) is an effective cycle. − log ‖i # s‖ 2 is a function taking<br />
only positive values on the 0-dimensional manifold Hom fields (L,). Hence,<br />
π ∗ ĉ 1 (i ∗ L ) = [(π ∗ div i # s, π ∗ (− log ‖i # s‖ 2 ))] =: [(z, r)]<br />
for an effective cycle z ∈ Z 0 (Spec) <strong>and</strong> r > 0.<br />
It follows directly from Example-Definition 4.5 that ̂deg [(z, r)] > 0.<br />
□
172 THE CONCEPT OF A HEIGHT [Chap. IV<br />
6.7. Remark. –––– Using the same techniques, S. Zhang [Zh95b, (5.2)] also<br />
shows the inequality<br />
p X GS (L , . . . , L )<br />
deg L<br />
X<br />
6.8. Remark. –––– By definition,<br />
≥ e 1 (h L<br />
) + . . . + e d+1 (h L<br />
) .<br />
p X GS (L , . . . , L ) = hBGS (X ) .<br />
The theorem of successive minima therefore provides a comparison of the height<br />
of the variety X itself with the heights of the <strong>points</strong> on it.<br />
In order to simplify formulas, S. Zhang [Zh95b] normalizes the height of a<br />
closed subvariety Z ⊆ X to be<br />
h L<br />
(Z) :=<br />
h BGS<br />
L (Z)<br />
(d + 1) deg L<br />
X .<br />
This normalization has to be used with some care. The problem is that the<br />
absolute height of a closed point is not the same as the normalized height of the<br />
corresponding 0-cycle.<br />
6.9. Lemma. –––– Let X be an arithmetic variety, X its generic fiber, <strong>and</strong><br />
(L , ‖ . ‖) ∈ ̂Pic(X ). Write d := dim X. Then, for every C > 0,<br />
( ) ( )<br />
(L , C ·‖ . ‖), . . . , (L , C ·‖ . ‖) = p<br />
X<br />
GS (L , ‖ . ‖), . . . , (L , ‖ . ‖)<br />
p X GS<br />
− (d + 1)(log C ) deg L<br />
X .<br />
Proof. By [dJo], we may choose a regular arithmetic variety X ′ <strong>and</strong> a generically<br />
finite morphism f : X ′ → X which surjects to X . Theorem 5.2.ii) implies<br />
that the assertion for X <strong>and</strong> L is equivalent to the assertion for X ′ <strong>and</strong> f ∗ L .<br />
We are therefore reduced to the case that X is regular.<br />
Then, by Theorem 5.2.i),<br />
( )<br />
(L , C ·‖ . ‖), . . . , (L , C ·‖ . ‖)<br />
p X GS<br />
= ̂deg π ∗<br />
(ĉ1 (L , C ·‖ . ‖) · . . . · ĉ 1 (L , C ·‖ . ‖) )<br />
= ̂deg π ∗<br />
(<br />
(ĉ1 (L , ‖ . ‖)+[(0, −2 logC)]) · . . . · (ĉ 1 (L , ‖ . ‖)+[(0, −2 logC)]) ) .<br />
Here, π : X → Specdenotes the structural morphism.<br />
Next, we apply the binomial theorem. The description of the intersection<br />
product given in 4.10 immediately shows<br />
[(0, −2 logC)]·[(Z, g Z )] = [(0, −2(logC )δ Z )]
Sec. 7] THE ADELIC PICARD GROUP 173<br />
for (Z, g Z ) an arbitrary arithmetic cycle. In particular,<br />
[(0, −2 logC )]·[(0, −2 logC)] = 0 .<br />
By consequence, we find<br />
( )<br />
(L , C ·‖ . ‖), . . . , (L , C ·‖ . ‖)<br />
p X GS<br />
( )<br />
= pGS<br />
X (L , ‖ . ‖), . . . , (L , ‖ . ‖)<br />
+ (d + 1) ̂deg π ∗<br />
(<br />
[(0, −2 logC )] · ĉ1 (L , ‖ . ‖) · . . . · ĉ 1 (L , ‖ . ‖) ) .<br />
Here, ĉ 1 (L , ‖ . ‖) · . . . · ĉ 1 (L , ‖ . ‖) = [(Z, g Z )] for Z a cycle of dimension one.<br />
The restriction of Z to the generic fiber is a 0-cycle of degree deg L<br />
X. Therefore,<br />
π ∗<br />
(<br />
[(0, −2 logC )] · [(Z, gZ )] ) = π ∗ [(0, −2(logC )δ Z )]<br />
= [(0, −2(logC ) deg L<br />
X )] .<br />
The claim now follows from Example-Definition 4.5.<br />
□<br />
7. The adelic Picard group<br />
7.1. –––– The arithmetic intersection theory discussed so far seems to be<br />
very general. It is, however, still not general enough. An obvious problem<br />
is that the naive height on P n is not covered. The requirement that<br />
(L , ‖ . ‖) = L ∈ ̂Pic(X ) includes that ‖ . ‖ has to be smooth. For the naive<br />
height, we want to include the minimum metric. This means, we need a generalization<br />
to continuous hermitian metrics.<br />
This may be done in a rather ad hoc manner by taking into account certain<br />
limits of smooth hermitian metrics. A way to formalize this as well as the<br />
dependence of the height on the choice of a model is provided by S. Zhang’s<br />
adelic Picard group [Zh95a].<br />
i. The local case. Metrics induced by a model. —<br />
7.2. –––– Let K be an algebraically closed valuation field. The cases we have<br />
in mind are K =Ép for a prime number p <strong>and</strong> K =É∞ =.<br />
We will denote the valuation of x ∈ K by |x|. In the case K =Ép, we assume<br />
| . | is normalized by |p| = 1 . We also write ν(x) := − log |x|.<br />
p<br />
7.3. Definition. –––– Let X be a K-scheme. Then, by a metric on an invertible<br />
sheaf L ∈ Pic(X ), we mean a system of K-norms on the K-vector spaces L (x)<br />
for x ∈ X (K ).
174 THE CONCEPT OF A HEIGHT [Chap. IV<br />
7.4. Remark. –––– If K =then a metric on L is the same as a (possibly<br />
discontinuous) hermitian metric.<br />
7.5. Definition. –––– Assume K to be non-archimedean <strong>and</strong> let O K be the ring<br />
of integers in K. Further, let X be a K-scheme <strong>and</strong> L ∈ Pic(X ).<br />
Then, by a model of (X, L ) we mean a triple (X , ˜L , n) consisting of a natural<br />
number n, a flat projective scheme π : X → O K such that X K<br />
∼ = X, <strong>and</strong> an<br />
invertible sheaf ˜L ∈ Pic(X ) fulfilling ˜L | X<br />
∼ = L ⊗n .<br />
7.6. Example. –––– Assume K to be non-archimedean, let O K be the ring of<br />
integers in K, <strong>and</strong> let X be a K-scheme equipped with an invertible sheaf L .<br />
Then, a model (X , ˜L , n) of (X, L ) induces a metric ‖ . ‖ on L as follows.<br />
x ∈ X (K ) has a unique extension x : Spec O K → X . Then, x ∗˜L is a projective<br />
O K -module of rank 1. Each l ∈ L (x) induces l ⊗n ∈ L ⊗n (x) <strong>and</strong>, therefore, a<br />
<strong>rational</strong> section of x ∗˜L . Put<br />
‖l‖(x) :=<br />
[<br />
inf<br />
{<br />
|a| | a ∈ K, l ∈ a · x ∗˜L }] 1<br />
n<br />
. (†)<br />
7.7. Definition. –––– The metric ‖ . ‖ given by (†) is called the metric on L<br />
induced by the model (X , ˜L , n).<br />
7.8. Remark. –––– Note here that O K is, in general, a non-discrete valuation<br />
ring. In particular, O K will usually be non-Noetherian.<br />
Nevertheless, projectivity includes being of finite type [EGA, Chapitre II, Définition<br />
(5.5.2)]. This means, for the description of X , only a finite number of<br />
elements from O K are needed.<br />
In the particular case K =Ép, the group ν(K ) is isomorphic to (É, +).<br />
Thus, for any finite set {a 1 , . . . , a s } ⊂ , there exists a discrete valuation ring<br />
OÉp<br />
O ⊆ containing a OÉp<br />
1, . . . , a s .<br />
By consequence, X is the base change of some scheme which is projective over<br />
a discrete valuation ring.<br />
7.9. Definition. –––– Let K be an algebraically closed valuation field. Assume<br />
K to be non-archimedean.<br />
Then, a metric ‖ . ‖ on L ∈ Pic(X ) is called continuous, respectively bounded,<br />
if ‖ . ‖ = f · ‖ . ‖ ′ for ‖ . ‖ ′ a metric induced by some model <strong>and</strong> f a function<br />
on X (K ) which is continuous or bounded, respectively.<br />
7.10. Remark. –––– If K =then we adopt the concepts of bounded, continuous,<br />
<strong>and</strong> smooth metrics in their the usual meaning from complex geometry.<br />
Note that smooth metrics are continuous <strong>and</strong> that continuous metrics are automatically<br />
bounded in the case K =.
Sec. 7] THE ADELIC PICARD GROUP 175<br />
ii. The global case. Adelicly metrized invertible sheaves. —<br />
7.11. Definition. –––– Let X be a projective variety overÉ<strong>and</strong> m ∈Æ.<br />
Then, by a model of X over Spec[ 1 ], we mean a scheme X which is projective<br />
m<br />
<strong>and</strong> flat over Spec[ 1 ] such that the generic fiber of X is isomorphic to X.<br />
m<br />
7.12. Definition. –––– Let X be a projective variety overÉ<strong>and</strong> L ∈ Pic(X )<br />
be an invertible sheaf.<br />
a) Then, an adelic metric on L is a system<br />
‖ . ‖ = {‖ . ‖ ν } ν∈Val(É)<br />
of continuous <strong>and</strong> bounded metrics on ∈ ) such that<br />
LÉν<br />
Pic(XÉν<br />
i) for each ν ∈ Val(É), the metric ‖ . ‖ ν is Gal(Éν/Éν)-invariant,<br />
ii) for some m ∈Æ, there exist a model X of X over Spec[ 1 ], an invertible<br />
m<br />
sheaf ˜L ∈ Pic(X ), <strong>and</strong> a natural number n such that<br />
˜L | X<br />
∼ = L<br />
⊗n<br />
<strong>and</strong>, for all prime numbers p ∤m, the metric ‖ . ‖ νp is induced by (X p , ˜L | Xp , n).<br />
b) Aninvertiblesheafequipped withan adelic metriciscalled an adeliclymetrized<br />
invertible sheaf.<br />
All adelicly metrized invertible sheaves on X form an abelian group which will<br />
be denoted by Pic ad (X ).<br />
7.13. Notation. –––– Let X be a model of X over Spec. Then, taking the<br />
induced metric yields two natural homomorphisms<br />
i X : ̂Pic(X ) → Pic ad (X ) ,<br />
a X : ker(Pic(X ) → Pic(X )) ⊗É→Pic ad (X ) .<br />
Further, one has the forgetful homomorphism<br />
v : Pic ad (X ) → Pic(X ) .<br />
7.14. Notation. –––– The models of X together with all bi<strong>rational</strong> morphisms<br />
between them form an inverse system of schemes. This is a filtered inverse<br />
system since, for two models X <strong>and</strong> X ′ , the closure of the diagonal<br />
∆ ⊂ X × SpecÉX ⊂ X × SpecX ′ projects to both of them.<br />
Thus, the arithmetic Picard <strong>groups</strong> ̂Pic(X ) for all models X of X form a<br />
filtered direct system. The injections ̂Pic(X ) ֒→ Pic ad (X ) fit together to yield<br />
an injection<br />
ι X : lim −→ ̂Pic(X ) ֒→ Pic ad (X ) .
176 THE CONCEPT OF A HEIGHT [Chap. IV<br />
Similarly, the usual Picard <strong>groups</strong> Pic(X ) form a filtered direct system, too.<br />
We get a homomorphism<br />
α X : lim −→<br />
ker(Pic(X ) → Pic(X )) ⊗É−→ Pic ad (X ) .<br />
7.15. Definition (Metric on v −1 (L ) ⊆ Pic ad (X )). —–<br />
Let X be a projective variety overÉ. On X, let (L , ‖ . ‖) <strong>and</strong> (L , ‖ . ‖ ′ ) be two<br />
adelicly metrized invertible sheaves with the same underlying sheaf.<br />
Then, the distance between (L , ‖ . ‖) <strong>and</strong> (L , ‖ . ‖ ′ ) is given by<br />
δ ( (L , ‖ . ‖), (L , ‖ . ‖ ′ ) ) ( )<br />
:= ∑ δ ν ‖ . ‖ν , ‖ . ‖ ′ ν<br />
ν∈Val(É)<br />
for<br />
( δ ν ‖ . ‖ν , ‖ . ‖ ν) ′ := sup<br />
∣ log ‖ . ‖′ ν(x)<br />
‖ . ‖ ν (x) ∣ .<br />
x∈X (Éν )<br />
7.16. Lemma. –––– δ is a metric on the set v −1 (L ) of all metrizations of L .<br />
Proof. We have to show that the sum is always finite.<br />
For this, we note first that the metrics ‖ . ‖ ν <strong>and</strong> ‖ . ‖ ′ ν are bounded by definition.<br />
Therefore, each summ<strong>and</strong> is finite.<br />
We may thus ignore a finite set S of primes <strong>and</strong> assume that ‖ . ‖ <strong>and</strong> ‖ . ‖ ′<br />
are given by triples (X , ˜L , n) <strong>and</strong> (X ′ , ˜L ′ , n ′ ), respectively, in the sense of<br />
Definition 7.12.a.ii).<br />
=<br />
The isomorphism XÉ∼<br />
−→ X ∼ =<br />
−→ X ′Émay be extended to an open neighbourhood<br />
of the generic fiber. Therefore, enlarging S if necessary, we have<br />
an isomorphism X ′ ∼ =<br />
−→ X of schemes over Spec\S. Further, the<br />
triple (X , ˜L , n) may be replaced by (X , ˜L ⊗n′ , nn ′ ) without any change of<br />
the induced metric. Thus, without restriction, n = n ′ .<br />
To summarize, we are reduced to the case that<br />
XÉ‖ . ‖ <strong>and</strong> ‖ . ‖ ′ are given<br />
by (X , ˜L , n) <strong>and</strong> (X , ˜L ′ , n). We have an isomorphism<br />
= ˜L | XÉ∼<br />
−→ L ⊗n ∼ =<br />
−→ ˜L ′ |<br />
which may be extended to an open neighbourhood of the generic fiber. Therefore,<br />
in the definition of δ ( (L , ‖ . ‖), (L,‖.‖ ′ ) ) , all the summ<strong>and</strong>s vanish,<br />
except finitely many.<br />
Positivity, symmetry, <strong>and</strong> the triangle inequality are clear.<br />
7.17. Remark. –––– It is convenient to consider two adelicly metrized invertible<br />
sheaves with different underlying sheaves as of distance infinity. Then, the<br />
distance δ is no longer a metric but only a separated écart in the sense of N. Bourbaki<br />
[Bou-T, §1].<br />
□
Sec. 7] THE ADELIC PICARD GROUP 177<br />
7.18. Lemma. –––– Let f : X → Y be a morphism of projective varieties overÉ.<br />
i) Then, the homomorphism f ∗ : Pic ad (Y ) → Pic ad (X ) is continuous with respect<br />
to the metric topology.<br />
ii) Even more,<br />
δ( f ∗ L 1 , f ∗ L 2 ) ≤ δ(L 1 , L 2 )<br />
for arbitrary adelicly metrized invertible sheaves L 1 , L 2 ∈ Pic ad (Y ).<br />
Proof. ii) is obvious. i) follows immediately from ii).<br />
□<br />
iii. The adelic Picard group. —<br />
7.19. Definition. –––– Let X be a regular, projective variety overÉ.<br />
a) We call an adelicly metrized invertible sheaf (L , ‖ . ‖) on X semipositive if<br />
there exist a model X of X over Spec<strong>and</strong> a smooth hermitian line bundle<br />
(L , ‖ . ‖ ∞ ) ∈ ̂Pic(X ) fulfilling<br />
i) (L , ‖ . ‖) = i X (L , ‖ . ‖ ∞ ),<br />
ii) L | Xp<br />
is nef for every prime number p, <strong>and</strong><br />
iii) the curvature form c 1 (L, ‖ . ‖ ∞ ) is non-negative.<br />
b) By the semipositive cone C ≥0 ⊂ X<br />
Picad (X ), we mean the set of all adelicly<br />
metrized invertible sheaves which may be written as positiveÉ-linear combinations<br />
of semipositive elements.<br />
7.20. Definition. –––– Let X be a regular, projective variety overÉ.<br />
We define the adelic Picard group ˜Pic(X ) ⊆ Pic ad (X ) to be the subgroup generated<br />
by C ≥0<br />
X . Here, overlining indicates the topological closure in Picad (X ) with<br />
respect to the metric topology.<br />
If (L , ‖ . ‖) ∈ ˜Pic(X ) then ‖ . ‖ is called an integrable metric on L . (L , ‖ . ‖) is<br />
then said to be an integrably metrized invertible sheaf.<br />
7.21. Remark. –––– Since C ≥0<br />
X<br />
is a surjection.<br />
diff : C ≥0<br />
X<br />
is closed under addition, the difference map<br />
× C ≥0<br />
X −→ ˜Pic(X )<br />
7.22. Definition. –––– Let X be a regular, projective variety overÉ.<br />
We define the integrable topology on ˜Pic(X ) to be the finest topology such that<br />
the difference map diff is continuous.
178 THE CONCEPT OF A HEIGHT [Chap. IV<br />
7.23. Remarks. –––– i) The integrable topology on ˜Pic(X ) is finer than the<br />
metric topology.<br />
ii) We will consider ˜Pic(X ) as being equipped with the integrable topology.<br />
Not with the metric topology which is the same as the topology as a subspace<br />
of Pic ad (X ).<br />
iii) A map i : ˜Pic(X ) → M to any topological space is continuous if <strong>and</strong> only if<br />
i ◦ diff : C ≥0<br />
X<br />
× C ≥0<br />
X<br />
−→ M<br />
is continuous with respect to the metric topology.<br />
7.24. Lemma. –––– Let X be a regular, projective variety overÉ. Then, the<br />
difference map<br />
diff : C ≥0 × X<br />
C ≥0 −→ ˜Pic(X X<br />
)<br />
is open.<br />
Proof. Let U ⊆ C ≥0 × X<br />
C ≥0<br />
X<br />
be any open set. We have to show that<br />
diff −1 (diff(U )) is open. But this is clear since<br />
diff −1 (diff(U )) = [ [ ] [ ]<br />
U + (x, x) ∩ C<br />
≥0 ≥0<br />
X<br />
×CX<br />
x∈Pic ad (X )<br />
<strong>and</strong> addition is continuous with respect to the metric topology.<br />
7.25. Lemma. –––– Let X be a regular, projective variety overÉ. Then, the<br />
topological closures of C ≥0<br />
X<br />
with respect to the metric topology <strong>and</strong> the integrable<br />
topology coincide.<br />
Proof. Let C ≥0<br />
≥0<br />
X<br />
be the closure of CX<br />
with respect to the metric topology <strong>and</strong><br />
C be the closure with respect to the integrable topology. Since the integrable<br />
topology is finer, we have C ⊆ C ≥0<br />
X .<br />
On the other h<strong>and</strong>, let l ∈ C ≥0<br />
X . Then, there is a sequence (l i) i in C ≥0<br />
X<br />
which<br />
converges to l. As the canonical map diff(·, 0): C ≥0 → ˜Pic(X X<br />
) is continuous,<br />
we see that l i → l with respect to the integrable topology. Hence, l ∈ C. □<br />
7.26. Definition. –––– Let X be a smooth, projective variety overÉ<br />
<strong>and</strong> L ∈ Pic(X ).<br />
a) Let p be a prime number.<br />
Then, a metric ‖ . ‖ on will be called semipositive if there is a sequence<br />
LÉp<br />
(‖ . ‖ i ) i≥0 of metrics on L satisfying the following conditions.<br />
i) (‖ . ‖ i ) i≥0 converges uniformly to ‖ . ‖.<br />
ii) Each ‖ . ‖ i is induced by a triple (X × Specp, Spec(p)<br />
pr ∗ 1˜L , n) where X is a<br />
scheme over(p) :=p ∩É, ˜L ∈ Pic(X ), <strong>and</strong> ˜L | Xp<br />
is nef.<br />
□
Sec. 7] THE ADELIC PICARD GROUP 179<br />
b) We will call a metric ‖ . ‖ on Lpositive if ‖ . ‖ is smooth <strong>and</strong> the curvature<br />
form c 1 (L , ‖ . ‖) is positive.<br />
‖ . ‖ is said to be almost semiample if there is a sequence of positive metrics<br />
(‖ . ‖ i ) i≥0 on Lthat converges uniformly to ‖ . ‖.<br />
7.27. Lemma. –––– Let X be a regular, projective variety overÉ. Then, one has<br />
C ≥0<br />
X = {(L , {‖ . ‖ ν} ν∈Val(É)) ∈ Pic ad (X ) |<br />
‖ . ‖ ν semipositive for ν non-archimedean,<br />
Proof. We consider C ≥0<br />
X<br />
“⊆” is clear from Definitions 7.19 <strong>and</strong> 7.26.<br />
‖ . ‖ ν almost semiample for ν archimedean}.<br />
as the closure with respect to the metric topology.<br />
“⊇” Let (L , {‖ . ‖ ν } ν ) ∈ Pic ad (X ) be such that ‖ . ‖ νp is semipositive for every<br />
prime number p <strong>and</strong> ‖ . ‖ ν∞ is almost semiample. The assertion says that, for<br />
every ε > 0, there exists (L , {‖ . ‖ ′ ν} ν ) ∈ C ≥0<br />
X<br />
such that<br />
δ ( (L , {‖ . ‖ ν } ν ), (L , {‖ . ‖ ′ ν} ν<br />
L<br />
) ) < ε .<br />
The archimedean valuation. The hermitian metric ‖ . ‖ ν∞ is almost semiample.<br />
This means that, for every ε > 0, there is a hermitian metric ‖ . ‖ ′ ν ∞<br />
on<br />
which is smooth <strong>and</strong> positively curved such that<br />
δ ν∞<br />
( ‖ . ‖ν∞ , ‖ . ‖ ′ ν ∞<br />
) < ε .<br />
Good primes. By Definition 7.12, we have a finite set S of bad prime numbers,<br />
a model X of X over Spec\S, an invertible sheaf ˜L ∈ Pic(X ), <strong>and</strong> a natural<br />
number n such that<br />
i) ˜L | X<br />
∼ = L ⊗n <strong>and</strong><br />
ii) ‖ . ‖ νp is induced by (X ν , ˜L | Xν , n) for all prime numbers p ∉ S.<br />
Bad primes. Let p ∈ S. According to Definition 7.26.a), for every ε > 0, there<br />
exists an induced metric ‖ . ‖ ′ ν p<br />
such that<br />
δ νp<br />
( ‖ . ‖νp , ‖ . ‖ ′ ν p<br />
) < ε .<br />
More precisely, ‖ . ‖ ′ ν p<br />
is induced by a triple<br />
(X p × Spec(p) Specp, pr ∗ 1˜L (p) , n p )<br />
such that X p is a model of over(p), ˜L (p) ∈ Pic(M XÉ(p) p ), <strong>and</strong> ˜L (p) | Xp<br />
is nef.
180 THE CONCEPT OF A HEIGHT [Chap. IV<br />
Gluing. We claim that X <strong>and</strong> all the X p for p ∈ S may be glued together along<br />
the generic fiber X to a scheme X ′ over Spec. Further, for<br />
N := n ∏ n p ,<br />
p∈S<br />
the tensor powers ˜L ⊗N/n <strong>and</strong> [˜L (p) ] ⊗N/n p<br />
fit together to give an invertible<br />
sheaf ˜L ′ on X .<br />
Then, {‖ . ‖ ′ ν} ν is defined by<br />
(L , {‖ . ‖ ′ ν} ν ) := 1 N i X (˜L ′ , ‖ . ‖ ′ ν ∞<br />
) .<br />
This guarantees that ‖ . ‖ ′ ν p<br />
<strong>and</strong> ‖ . ‖ νp coincide for all p ∉ S. In particular,<br />
˜L ′ | Xp<br />
is automatically nef for these primes. For the bad primes p ∈ S, the<br />
same property follows immediately from the construction.<br />
Gluing may, however, not be done directly as X ⊂ X is not an open set.<br />
Instead, we may extend each X p to a scheme X (p) over Spec[ 1 ] for some<br />
m<br />
m ∈Æwhich is not a multiple of p. We may assume that m is divisible by all<br />
bad primes different from p.<br />
The isomorphism between the generic fibers of X <strong>and</strong> X (p) extends to an open<br />
neighbourhood of X. More precisely, there exists some f p ∈not divisible by<br />
any of the bad primes such that the induced bi<strong>rational</strong> map<br />
is an isomorphism.<br />
X (p) × SpecSpec[ 1<br />
p f p<br />
] −→ X × SpecSpec[ 1 f p<br />
]<br />
Putting f := ∏ p∈S f p <strong>and</strong> U (p) := X (p) × SpecSpec[ 1 ], we have an isomorphism<br />
f<br />
ϕ p : U (p) × SpecSpec[ 1 ] −→ X × p SpecSpec[ 1 ] . f<br />
The definition<br />
ϕ p1 ,p 2<br />
:= ϕ −1<br />
p 2<br />
◦ϕ p1 : U (p1 ) × SpecSpec[ 1 p 1<br />
] −→ U (p2 ) × SpecSpec[ 1 p 2<br />
]<br />
makes the gluing data complete.<br />
The cocycle relations are fulfilled since the corresponding relations are satisfied<br />
on the generic fiber. Observe, U (p) for p ∈ S <strong>and</strong> X are integral schemes.<br />
By consequence, the U (p) <strong>and</strong> X may be glued together to give a-scheme X ′ .<br />
To glue the invertible sheaves together, we need isomorphisms<br />
ι p : ϕ ∗ p˜L ⊗N/n | U −→ [˜L (p) ] ⊗N/n p<br />
| U(p) × SpecSpec[ 1 p ]<br />
extending the isomorphism over X.
Sec. 7] THE ADELIC PICARD GROUP 181<br />
Generally, there is an open neighbourhood O ⊆ U (p) × SpecSpec[ 1 p ] of X<br />
such that ι p | O exists <strong>and</strong> is an isomorphism. Adding to f a suitable set of prime<br />
factors guarantees that O = U (p) × SpecSpec[ 1 p ].<br />
This completes the proof.<br />
7.28. Remark (concerning the notion of a semipositive metric). —–<br />
Condition ii) in Definition 7.26 seems to be somewhat unnatural. One might<br />
want to allow arbitrary models overp or, even more generally, over the integer<br />
ring O L of a finite extension L ofÉp.<br />
It turns out from a st<strong>and</strong>ard descent argument that finite extensions actually<br />
make no difference. Indeed, a metric is supposed to be Gal(Ép/Ép)-invariant.<br />
Thus, it may be shown by descent that for every model over O L there is an<br />
equivalent model overp. Similarly, for every model over the integer ring of a<br />
finite extension ofÉ(p), there is an equivalent model over(p).<br />
The descent needed here may actually be seen as a particular case of faithful<br />
flat descent [K/O74] or as a slight generalization of Proposition I.3.5<br />
to O K -schemes instead of K-schemes.<br />
It seems, however, that allowing the completion leads to a more general notion<br />
of semipositivity. We do not know whether Lemma 7.27 is still true for<br />
that notion.<br />
7.29. Lemma. –––– Let X be a regular, projective variety overÉ.<br />
a) Then, the natural homomorphisms<br />
<strong>and</strong><br />
factorize via ˜Pic(X ).<br />
ι X : lim −→ ̂Pic(X ) → Pic ad (X )<br />
α X : lim −→<br />
ker(Pic(X ) → Pic(X )) ⊗É→Pic ad (X )<br />
b) The subgroup 〈 im ι X , im α X 〉 is dense in ˜Pic(X ).<br />
Proof. a) Let X be any model of X over Spec. We have to show that the<br />
homomorphisms i X <strong>and</strong> a X factorize via ˜Pic(X ).<br />
Since X is a projective scheme, every (L , ‖ . ‖) ∈ ̂Pic(X ) may be written<br />
as (L 1 , ‖ . ‖ 1 )⊗ OX (L 2 , ‖ . ‖ 2 ) ∨ where both oper<strong>and</strong>s are ample invertible sheaves<br />
with positive metrics. Further, we note that ampleness of L i implies that L i | Xp<br />
is ample on Xp<br />
for every p (cf. [EGA, Chapitre II, Corollaire (4.5.10.ii)]).<br />
Hence,<br />
i X (L , ‖ . ‖) = i X (L 1 , ‖ . ‖ 1 ) − i X (L 2 , ‖ . ‖ 2 ) ∈ diff(C ≥0<br />
X , C ≥0<br />
X ) .<br />
□
182 THE CONCEPT OF A HEIGHT [Chap. IV<br />
On the other h<strong>and</strong>, let k ∈Æ<strong>and</strong> L ∈ ker(Pic(X ) → Pic(X )) be given.<br />
We may write L = L 1 ⊗ OX L2 ∨ for two ample invertible sheaves such<br />
that L 1 | X<br />
∼ = L2 | X is k-divisible in Pic(X ). Then,<br />
a X (L ⊗ 1/k) = 1 k ·i X (L 1 , ‖ . ‖) − 1 k ·i X (L 2 , ‖ . ‖) ∈ diff(C ≥0<br />
X , C ≥0<br />
X )<br />
independently of the metric ‖ . ‖ chosen. Note that 1 k ·i X (L i , ‖ . ‖) ∈ Pic ad (X )<br />
since L 1 | X is k-divisible.<br />
b) Since the difference map diff : C ≥0<br />
X<br />
× C ≥0<br />
X<br />
surjective, it suffices to prove that 〈 im ι X , im α X 〉∩C ≥0<br />
X<br />
show that even 〈 im ι X , im α X 〉 ⊇ C ≥0<br />
X .<br />
For this, let (L , ‖ . ‖) ∈ C ≥0<br />
X<br />
. By definition, we have<br />
→ ˜Pic(X ) is continuous <strong>and</strong><br />
≥0<br />
is dense in C . We will<br />
(L , ‖ . ‖) = a 1 i X (L 1 , ‖ . ‖ 1 ) + . . . + a n i X (L n , ‖ . ‖ n )<br />
for a certain model X of X, (L i , ‖ . ‖ i ) ∈ ̂Pic(X ), <strong>and</strong> <strong>rational</strong> numbers a i ≥ 0.<br />
We choose a common denominator k of all the a i . Then,<br />
(L , ‖ . ‖) ⊗k = i X (L , ‖ . ‖ ′ )<br />
for some (L , ‖ . ‖ ′ ) ∈ ̂Pic(X ). In particular, L ⊗k ∼ = L | X .<br />
We may write L = R ⊗k ⊗ T for R ∈ Pic(X ) an arbitrary extension of L<br />
<strong>and</strong> T ∈ ker(Pic(X ) → Pic(X )). Then,<br />
(L , ‖ . ‖) = i X (R, ‖ . ‖ ′ 1 k ) + aX (T ⊗ 1/k) .<br />
X<br />
The assertion is proven.<br />
□<br />
iv. Pull-back <strong>and</strong> intersection product. —<br />
7.30. Lemma. –––– Let f : X → Y be a morphism of projective varieties overÉ.<br />
a) Then, there is a natural homomorphism<br />
f ∗ : ˜Pic(Y ) → ˜Pic(X )<br />
(L , ‖ . ‖) ↦→ ( f ∗ L , f ∗ ‖ . ‖) .<br />
f ∗ admits the property that f ∗( )<br />
CY<br />
≥0 ⊆ C<br />
≥0<br />
X .<br />
b) f ∗ is continuous.<br />
Proof. a) It is clear that if ‖ . ‖ is an adelic metric then f ∗ ‖ . ‖ is an adelic metric.<br />
According to Definition 7.20, for the existence of f ∗ : ˜Pic(Y ) → ˜Pic(X ), it will<br />
be sufficient to show<br />
f ∗( )<br />
CY<br />
≥0 ⊆ C<br />
≥0<br />
X .
Sec. 7] THE ADELIC PICARD GROUP 183<br />
Let us first verify that f ∗ (CY ≥0 ) ⊆ C ≥0<br />
X .<br />
≥0<br />
For this, let (L , ‖ . ‖) ∈ CY .<br />
I.e., (L , ‖ . ‖) = i Y (L , ‖ . ‖ ∞ ) for a smooth hermitian line bundle (L , ‖ . ‖ ∞ )<br />
on a model Y of Y such that L | Yp<br />
is nef for every prime number p <strong>and</strong> the<br />
curvature form c 1 (L, ‖ . ‖ ∞ ) is non-negative.<br />
Fix an arbitrary model X ′ of X. We define X to be the closure of the<br />
graph of f in X ′ ×Y . Then, the projection to the second factor yields a<br />
morphism f : X → Y which extends f . This yields<br />
f ∗ (L , ‖ . ‖) = i X ( f ∗ L , f ∗ ‖ . ‖ ∞ ) .<br />
The pull-back of a nef invertible sheaf is always nef <strong>and</strong> the pull-back of<br />
a smooth non-negatively curved hermitian line bundle is smooth <strong>and</strong> nonnegatively<br />
curved. Hence, we indeed have f ∗ (CY ≥0 ) ⊆ C ≥0<br />
X .<br />
Further, by Lemma 7.18, the homomorphism f ∗ : Pic ad (Y ) → Pic ad (X ) is continuous<br />
with respect to the metric topology. Lemma 7.25 yields the assertion.<br />
b) f ∗ : CY<br />
≥0 → C ≥0<br />
X<br />
is continuous with respect to the metric topology. Thus, we<br />
have a continuous map<br />
C ≥0<br />
Y<br />
× C ≥0<br />
Y<br />
f ∗ ×f<br />
−→ ∗<br />
C ≥0 × X<br />
C ≥0 diff<br />
X<br />
−→ Pic ad (X ) .<br />
The definition of the integrable topology yields the claim.<br />
□<br />
7.31. Theorem. –––– Let X be a regular, projective variety overÉ.<br />
i) Then, there is exactly one continuous-multilinear map, the adelic intersection<br />
product<br />
pZ X : ˜Pic(X ) × . . . × ˜Pic(X )<br />
Ê<br />
−→Ê,<br />
} {{ }<br />
dim X+1 times<br />
such that for every model X of X over Specthe diagram<br />
˜Pic(X ) × . . . × ˜Pic(X<br />
pZ<br />
X<br />
)<br />
commutes.<br />
ii) p X Z<br />
is symmetric.<br />
i dimX+1<br />
X<br />
̂Pic(X ) × . . . × ̂Pic(X )<br />
<br />
Proof. i) By Theorem 5.2.ii), the intersection products pGS X are compatible with<br />
pull-backs under bi<strong>rational</strong> maps. We obtain a-multilinear map p such that<br />
p X GS
Ê<br />
184 THE CONCEPT OF A HEIGHT [Chap. IV<br />
the diagrams<br />
p<br />
im ι X × . . . × im ι X<br />
i dimX+1<br />
X<br />
<br />
p X<br />
̂Pic(X ) × . . . × ̂Pic(X<br />
GS<br />
)<br />
commute.<br />
Every element of im α X has an integral multiple in im ι X . Therefore, p extends<br />
uniquely to a-multilinear map<br />
p : 〈 im ι X , im α X 〉 × . . . × 〈 im ι X , im α X 〉 −→Ê.<br />
By Lemma 7.29.b), 〈 im ι X , im α X 〉 is a dense subset of ˜Pic(X ). This implies that<br />
pZ X is unique.<br />
To show the existence, we need that p is continuous. This may be tested locally.<br />
Let<br />
We write<br />
l = (l 0 , . . . , l dimX ) ∈ 〈 im ι X , im α X 〉 × . . . × 〈 im ι X , im α X 〉<br />
⊆ ˜Pic(X ) × . . . × ˜Pic(X ) .<br />
(l 0 , . . . , l dimX ) = (l (1)<br />
0 , . . . , l(1) dim X ) − (l(2) 0 , . . . , l(2) dimX )<br />
for l (i)<br />
0 , . . . , l(i) ∈ dimX C ≥0<br />
(i)<br />
X<br />
<strong>and</strong> define U<br />
j<br />
in C ≥0<br />
X<br />
for ε := 1. By Lemma 7.24,<br />
to be the ε-neighbourhood of l (i)<br />
j<br />
U l := diff(U (1)<br />
0 , U (2)<br />
(1)<br />
0<br />
) × . . . × diff(U<br />
dim X , U (2)<br />
is an open neighbourhood of (l 0 , . . . , l dimX ).<br />
dimX )<br />
It is therefore sufficient to verify that p is continuous on U l . This is a direct<br />
consequence of Lemma 7.32. Indeed, for every (L , ‖ . ‖) ∈ U (i)<br />
j<br />
the restriction<br />
:= L | X is always the same. Put<br />
L (i)<br />
j<br />
M 0 :=<br />
2 2<br />
∑ ∑<br />
i 1 =1 i 2 =1<br />
. . .<br />
2<br />
∑<br />
i dimX =1<br />
deg ( c 1 (L (i 1)<br />
1<br />
) · . . . · c 1 (L (i dimX )<br />
dim X )) .<br />
Then, p is Lipschitz continuous on U l with respect to the first variable. M 0 is a<br />
Lipschitz constant.<br />
Lipschitz continuity in the other variables may be shown analogously.<br />
ii) This follows immediately from the uniqueness together with the symmetry<br />
of the intersection products pGS X proven in Theorem 5.2.b).<br />
□
Sec. 7] THE ADELIC PICARD GROUP 185<br />
7.32. Lemma. –––– Let π : X → Specbe an arithmetic variety. Suppose its<br />
generic fiber X is regular <strong>and</strong> put d := dim X.<br />
Let L 1 , . . . , L d ∈ ̂Pic(X ). Suppose that L 1 | , . . . , L Xp<br />
d| Xp<br />
are nef for every<br />
prime number p. Further, consider (L , ‖ . ‖) ∈ ̂Pic(X ) such that L | X<br />
∼ = OX .<br />
Then,<br />
|p X GS(<br />
(L , ‖ . ‖), L1 , . . . , L d<br />
) | ≤<br />
≤ δ ( i X (L , ‖ . ‖), 0 ) · deg ( c 1 (L 1 | X ) · . . . · c 1 (L d | X ) )<br />
for 0 ∈ Pic ad (X ) the neutral element.<br />
Proof. The assertion is compatible with pull-back under a generically finite, surjective<br />
morphism. Thus, we may assume without restriction that X is regular.<br />
The asserted inequality then reads<br />
∣<br />
∣̂deg π ∗<br />
(ĉ1 (L , ‖ . ‖) · ĉ 1 (L 1 ) · . . . · ĉ 1 (L d ) )∣ ∣ ≤<br />
To show this, we write first<br />
≤ δ ( i X (L , ‖ . ‖), 0 ) · deg ( c 1 (L 1 | X ) · . . . · c 1 (L d | X ) ) .<br />
ĉ 1 (L , ‖ . ‖) =: [(D, g)] .<br />
Here, D = div s for s the <strong>rational</strong> section of L such that s| X = 1 ∈ Γ(X, O X ).<br />
In particular, D is a divisor supported in the special fibers. Further,<br />
g := − log ‖s‖ 2 .<br />
It is sufficient to treat the cases (0, g) <strong>and</strong> (D, 0).<br />
First Case: D = 0.<br />
We have ĉ 1 (L 1 ) · . . . · ĉ 1 (L d ) = [(Z, g Z )] for Z a cycle of dimension<br />
one. The restriction of Z to the generic fiber is a 0-cycle of degree<br />
deg ( c 1 (L 1 | X ) · . . . · c 1 (L d | X ) ) . Therefore, by the description of the arithmetic<br />
intersection product given in 4.10,<br />
(<br />
π ∗ [(0, g)] · ĉ1 (L 1 ) · . . . · ĉ 1 (L d ) ) (<br />
= π ∗ [(0, g)] · [(Z, gZ )] )<br />
= π ∗ [(0, gδ Z )] .<br />
The absolute value of the arithmetic degree is bounded by<br />
1<br />
2 sup |g(x)|·degZÉ≤ sup |log ‖1‖(x)| · deg ( c 1 (L 1 | X ) · . . . · c 1 (L d | X ) ) .<br />
x∈X ()<br />
x∈X ()<br />
Second Case: g = 0.<br />
We may assume that D is contained in one special fiber, say, in Xp .
186 THE CONCEPT OF A HEIGHT [Chap. IV<br />
Assume first that D = (V ) for an irreducible component V of of maximal<br />
dimension. Then, by the definition of the push-forward <strong>and</strong> the arithmetic<br />
Xp<br />
degree, we have<br />
̂deg π ∗<br />
(<br />
[(D, 0)] · ĉ1 (L 1 ) · . . . · ĉ 1 (L d ) ) = (log p)·deg ( c 1 (L 1 | V )·. . . ·c 1 (L d | V ) ) .<br />
We observe, if D is effective then the expression on the right h<strong>and</strong> side is nonnegative.<br />
Indeed, the intersection number of nef divisors is always non-negative<br />
by a theorem of S. Kleiman [Laz, Example 1.4.16].<br />
By consequence, if<br />
for r ∈Êthen<br />
∣ ̂deg<br />
(<br />
π ∗ [(D, 0)] · ĉ1 (L 1 ) · . . . · ĉ 1 (L d ) )∣ ∣ ≤<br />
) ≤ D ≤ ) (‡)<br />
−r(Xp r(Xp<br />
≤ r(log p) · deg ( c 1 (L 1 | Xp ) · . . . · c 1(L d | Xp ))<br />
= r(log p) · deg ( c 1 (L 1 | X ) · . . . · c 1 (L d | X ) ) (§)<br />
since the fibers are linearly equivalent to each other.<br />
We choose the minimal number r such that property (‡) is fulfilled.<br />
This means, there is some ξ ∈ Xp<br />
such that O X ,ξ (D) = O X ,ξ (±rXp ).<br />
In particular, r = s/t is a <strong>rational</strong> number. We assume the plus sign without<br />
restriction. Then, O X ,ξ (D) ⊗t = O X ,ξ (Xp )⊗s .<br />
The definition of D shows that L = O X (D). By consequence,<br />
L ⊗t<br />
ξ<br />
= O X,ξ (D) ⊗t = O X ,ξ (Xp )⊗s .<br />
Definition 7.7implies that ‖1‖ p (x) = p −r for x ∈ X specializing to ξ. Hence, by<br />
Lemma-Definition 7.15,<br />
δ νp<br />
(<br />
iX (L , ‖ . ‖), 0 ) ≥ |log ‖1‖ p (x)| = r log p .<br />
Formula (§) yields the assertion.<br />
□<br />
7.33. Lemma. –––– Let f : X → Y be a generically finite morphism of projective<br />
varieties overÉ. Then, for every L 1 , . . . , L dim Y+1 ∈ ˜Pic(Y ),<br />
p X Z ( f ∗ L 1 , . . . , f ∗ L dim Y+1 ) = p Y Z (L 1, . . . , L dim Y+1 ) .<br />
Proof. This follows directly from Theorem 5.2.a.ii) together with<br />
Lemma 7.29.b).<br />
□
Sec. 7] THE ADELIC PICARD GROUP 187<br />
7.34. Remark. –––– One might want to enlarge ˜Pic(X ). This is easily possible<br />
by omitting condition ii) in Definition 7.12. This would not destroy any of the<br />
properties proven except for the explicit description of the closure C ≥0<br />
X<br />
given<br />
in Lemma 7.27.<br />
7.35. Remark. –––– A priori, S. Zhang’s adelic Picard group [Zh95a] is somewhat<br />
smaller.<br />
S. Zhang works with a more restrictive notion of convergence. Instead of<br />
the closure with respect to the metric topology, he works with all limits of<br />
sequences ( {‖ . ‖ ν,i } ν<br />
)i<br />
of adelic metrics which are convergent in the following<br />
sense.<br />
(‖ . ‖ ν,i ) i is a constant sequence for all but finitely many valuations ν. It converges<br />
uniformly for the other ν.<br />
The proof of Lemma 7.27 indicates, however, that this should make no difference.<br />
v. Examples. —<br />
7.36. Example. –––– Let K be a number field. Then,<br />
I.e., there is an exact sequence<br />
˜Pic(SpecK ) = ̂Pic(Spec O K ) .<br />
0 −→ Cl K −→ ˜Pic(SpecK ) ˜deg<br />
−→Ê−→ 0 .<br />
Here, the class group Cl K is a discrete subgroup of ˜Pic(SpecK ).<br />
homeomorphism on every cofactor.<br />
˜deg is a<br />
Proof. The models of SpecK are the spectra of the orders of K. All induced<br />
metrics come from the maximal order.<br />
□<br />
7.37. Example. –––– Let V be a semistable projective curve overÉ.<br />
Then, there is an exact sequence<br />
where= L<br />
ν∈Val(É)<br />
0 →→ ˜Pic(V ) → Pic(V ) → 0<br />
C 0 (V (Éν)).<br />
Here, C 0 (V (É∞)) = C(V ()) F ∞<br />
is the space of all continuous F ∞ -invariant<br />
functions on V ().<br />
For p a prime number, C 0 (V (Ép)) is given as follows.<br />
We choose a minimal semistable model V of V over SpecÉp. Let D 1 , . . . , D r<br />
be the irreducible divisors in the special fiber. Further, for each k ∈Æ, we
→Ê<br />
188 THE CONCEPT OF A HEIGHT [Chap. IV<br />
define a countable disjoint covering {U i<br />
(k) } i∈Æof V (Ép) by the requirement<br />
that x, y ∈ V (Ép) are in the same set if they coincide modulo p k .<br />
Then, C 0 (V (Ép)) consists of all Gal(Ép/Ép)-invariant functions V (Ép)<br />
of the type<br />
r<br />
∞<br />
∑ α i ‖1‖ (V ,O(Di )) + ∑ ∑ β ikl χ U<br />
(k) .<br />
i<br />
i=1<br />
l=1 ik<br />
Here, ‖ . ‖ (V ,O(Di )) denotes the induced metric, induced by the model (V , O(D i )).<br />
χ U<br />
(k)<br />
i<br />
is the characteristic function of U (k)<br />
i<br />
. α i <strong>and</strong> β ikl are real numbers.<br />
We require that the inner sum is finite for each l ∈Æ<strong>and</strong> that the outer series<br />
converges uniformly.<br />
vi. Adelic heights. —<br />
7.38. Definition. –––– Let X be a regular, projective variety overÉ<strong>and</strong><br />
L ∈ ˜Pic(X ) be an integrably metrized invertible sheaf.<br />
Then, the absolute height with respect to L of an L-valued point x ∈ X (L)<br />
for L a number field is given by<br />
1<br />
h L<br />
(x) :=<br />
[L :É] ˜deg L | x .<br />
7.39. Remark. –––– Lemma 7.33 shows that this definition is independent of<br />
the choice of L.<br />
7.40. Corollary (Theorem of successive minima). —–<br />
Let X be a regular, projective variety overÉ<strong>and</strong> L ∈ ˜Pic(X ). Assume<br />
a) L ∈ C ≥0<br />
X<br />
or<br />
b) L ∈ ˜Pic(X ) is such that ‖ . ‖ p is semipositive for all p ≠ ∞ <strong>and</strong> ‖ . ‖ ∞ is<br />
almost semiample.<br />
Then,<br />
e 1 (h L<br />
) ≥ pX Z (L , . . . , L )<br />
(dim X + 1) deg L<br />
X ≥ e 1(h L<br />
) + . . . + e dim X+1 (h L<br />
)<br />
.<br />
dim X + 1<br />
Proof. Cf. [Zh95a, Theorem 1.10].<br />
a) From Theorem 7.31, we see that Corollary 6.6, together with Remark 6.7,<br />
immediately implies the assertion for L ∈ C ≥0<br />
X . Further, pX Z <strong>and</strong> ˜deg are continuous.<br />
This shows that the inequalities are true for L ∈ C ≥0<br />
X , too.<br />
b) Lemma 7.27 says that L = (L , ‖ . ‖) ∈ C ≥0<br />
X<br />
is equivalent to the statement<br />
that (L , ‖ . ‖) ∈ ˜Pic(X ) such that ‖ . ‖ p is semipositive for all p ≠ ∞ <strong>and</strong> ‖ . ‖ ∞<br />
is almost semiample.<br />
□
CHAPTER V<br />
ON THE DISTRIBUTION OF SMALL POINTS<br />
ON ABELIAN AND TORIC VARIETIES ∗<br />
1. Introduction<br />
i. Classical results. —<br />
1.1. –––– In 1972, J.-P. Serre proved the following remarkable result.<br />
Theorem (Serre). Let K be an algebraic number field <strong>and</strong> E be an elliptic curve<br />
over K without complex multiplication over K. Then, for almost every prime<br />
number l, the Galois group Gal(K/K ) acts transitively on the l-torsion <strong>points</strong> of E.<br />
1.2. Notation. –––– For x ∈ E(K ), we denote by δ x the Dirac measure associated<br />
to its Galois orbit. I.e., if x ∈ E(F ) for some number field F ⊇ K then<br />
for each ϕ ∈ C 0 (E()).<br />
δ x (ϕ) :=<br />
1<br />
♯ Gal(F/K )<br />
∑<br />
σ : F֒→ϕ(σ(x))<br />
1.3. –––– In this language, Serre’s theorem immediately implies the following<br />
corollary.<br />
Corollary. Let K be an algebraic number field <strong>and</strong> E be an elliptic curve over K<br />
without complex multiplication over K. Further, let (x i ) i∈Æbe a sequence of torsion<br />
<strong>points</strong> in E which does not contain any constant subsequence.<br />
Then, the associated sequence (δ xi ) i∈Æof <strong>measures</strong> on E() converges weakly to the<br />
Haar measure of volume one.<br />
(∗) This chapter is a revised version of the article: On the distribution of small <strong>points</strong> on abelian<br />
<strong>and</strong> toric varieties, Preprint.
190 DISTRIBUTION OF SMALL POINTS [Chap. V<br />
1.4. –––– In 1997, L. Szpiro, E. Ullmo <strong>and</strong> S. Zhang [S/U/Z] showed a generalization<br />
of this corollary from torsion <strong>points</strong> (which are of Néron-Tate height<br />
zero) to <strong>points</strong> of a small positive height. The three authors heavily make use<br />
of the powerful methods of Arakelov geometry as presented in Chapter IV.<br />
Their result is as follows.<br />
i∈Æ<br />
1.5. Theorem (Szpiro, Ullmo, Zhang). —– Let A be an abelian variety over<br />
some number field K <strong>and</strong> (x i ) i∈Æbe a sequence of closed <strong>points</strong> in A. Assume (x i )<br />
converges to the generic point of A in the sense of the Zariski topology. Suppose further<br />
that lim h NT (x i ) = 0.<br />
i→∞<br />
Then, the associated sequence (δ xi ) i∈Æof <strong>measures</strong> on A() converges weakly to the<br />
Haar measure of volume one.<br />
1.6. –––– Surprisingly, there is a completely parallel theorem for the naive<br />
height on projective space which was proven by Yuri Bilu [Bilu97]. In fact,<br />
for P 1 , he proved this result already in 1988 [Bilu88]. It would have been easy<br />
to deduce the general result from this.<br />
1.7. Theorem (Bilu). —– Let (x i ) i∈Æbe a sequence of closed <strong>points</strong> in P nÉ.<br />
Assume that (x i ) i∈Æconverges to the generic point of P n in the sense of the<br />
Zariski topology. Assume further that lim<br />
i→∞<br />
h naive (x i ) = 0.<br />
Then, the associated sequence (δ xi ) i∈Æof <strong>measures</strong> on P n () converges weakly to the<br />
Haar measure of volume one on (S 1 ) n ⊂n ⊂ P n ().<br />
ii. Canonical heights. —<br />
1.8. General situation. –––– Let P be a projective variety overÉequipped<br />
with an ample invertible sheaf L ∈ Pic(P). Further, we assume there is given<br />
a self map f : P → P such that there is some isomorphism Φ: L ⊗d ∼ =<br />
−→ f ∗ L .<br />
1.9. Definition. –––– The canonical height function h f ,L on P is given by<br />
1<br />
h f ,L (x) := lim<br />
n→∞ d h ( n L f (n) (x) )<br />
for x ∈ P a closed point. Here, f (n) means the n-fold iteration of f . h L denotes<br />
the height defined by the invertible sheaf L introduced in Definition IV.2.10.<br />
1.10. Remark. –––– The limit process defining h f ,L is a generalization of the<br />
classical one for the Néron-Tate height [C/S, Chapter VI, §4].<br />
We will see in Example 1.16.ii) that the naive height is a canonical height, too.
Sec. 1] INTRODUCTION 191<br />
1.11. Remark. –––– h L is defined only up to a bounded summ<strong>and</strong>. Nevertheless,<br />
h f ,L is independent of the choice of a particular representative. Indeed, if<br />
|h (1)<br />
L (x) − h(2) L (x)| ≤ C for every x ∈ P then<br />
1 (<br />
∣d n h(1) L f (n) (x) ) − 1 (<br />
d n h(2) L f (n) (x) )∣ ∣∣ ≤ C d . n<br />
iii. An equidistribution result. —<br />
1.12. –––– The goal of the present chapter is to prove a common generalization<br />
of Theorems 1.5 <strong>and</strong> 1.7. We will give an Arakelovian approach to a situation<br />
which covers the case of the Néron-Tate height as well as the naive height.<br />
1.13. Situation. –––– Let P be a regular projective variety overÉcontaining<br />
a group scheme G ⊆ P as an open dense subset. Further, let a morphism<br />
f : P → P be given <strong>and</strong> assume that<br />
i) the unit e is a repelling fixed point. I.e., all eigenvalues of the tangent<br />
map T e f : T e G → T e G are of absolute value strictly bigger than 1.<br />
ii) f | G is a group homomorphism f | G : G → G.<br />
iii) There is some compact, Zariski dense subgroup K ⊆ G() which is both<br />
backward <strong>and</strong> forward invariant under f.<br />
Finally, let L ∈ Pic(P) be ample <strong>and</strong> Φ: L ⊗d ∼ =<br />
−→ f ∗ L be an isomorphism.<br />
This guarantees that we have the canonical height h f ,L .<br />
1.14. Theorem (Equidistribution on P()). —– Let P, f , L , <strong>and</strong> Φ are as described<br />
in 1.13.<br />
Let (x i ) i∈Æbe a sequence of closed <strong>points</strong> in P. Assume that (x i ) i∈Æconverges<br />
to the generic point of P in the sense of the Zariski topology. Further, suppose<br />
that h f ,L (x i ) → 0.<br />
Then, the associated sequence (δ xi ) i of <strong>measures</strong> on P() converges weakly to the<br />
measure τ on P() being the zero measure on P() \ K <strong>and</strong> the Haar measure of<br />
volume one on K.<br />
1.15. Remark. –––– For a general self map f , iterating will lead to fractals.<br />
Unfortunately, not much is known about the corresponding fractal heights.<br />
1.16. Examples. –––– The condition that e is a repelling fixed point is fulfilled,<br />
in particular, if G is commutative <strong>and</strong> f : g ↦→ g l is the homomorphism raising<br />
to the l-th power for l ≥ 2. This includes the two st<strong>and</strong>ard cases. Namely,<br />
i) let P = G = A be an abelian variety <strong>and</strong> L ∈ Pic(A) a symmetric, ample<br />
invertible sheaf. One has<br />
f = [l]: A → A
192 DISTRIBUTION OF SMALL POINTS [Chap. V<br />
<strong>and</strong> puts K = A(). Then, independently of the choice of l, h f ,L is the<br />
Néron-Tate height corresponding to L .<br />
ii) Let P = P n , G =n+1 ∼<br />
m /m =n m <strong>and</strong> L = O(1). We have<br />
f = [l]: (x 0 , . . . , x n ) ↦→ (x l 0 , . . . , xl n )<br />
<strong>and</strong> put K := U (1) n . Here, h f ,L is the naive height.<br />
iii) One may combine abelian varieties <strong>and</strong> projective spaces <strong>and</strong> consider a<br />
split semi-abelian variety A × P n . Here, G = A ×n m , K = A × U (1),<br />
<strong>and</strong> L = S ⊠O(p) for S a symmetric, ample invertible sheaf on A. One may<br />
put f := [l] × [l 2 ].<br />
A. Chambert-Loir’s equidistribution result [C-L] for quasi-split semi-abelian<br />
varieties can easily be deduced from this.<br />
iv) Consider a free abelian group N of finite rank <strong>and</strong> a finite <strong>rational</strong> polyhedral<br />
decomposition {σ α } α of N ⊗Ê. These data define a proper toric variety X.<br />
The multiplication map N → N , n ↦→ ln induces a self-morphism<br />
f = [l]: X → X.<br />
Further, consider a function g : N ⊗Ê→Êwhich is strictly convex in the<br />
sense of [K/K/M/S, Chapter I, Theorem 13]. Then, L := F g is a T -invariant<br />
ample invertible sheaf on X. One has [l] ∗ F g<br />
∼ = Flg = F ⊗l<br />
g .<br />
In this situation, P = X, G = T ∼ =n m is the Zariski dense torus in X, <strong>and</strong><br />
K ⊂ T is its maximal compact subgroup.<br />
1.17. Remark. –––– The latter example clearly contains the case of the naive<br />
height on projective space. Notice that it also contains the blow-up Bl (0:0:0) (P 2 )<br />
of the projective plane in the origin <strong>and</strong> the canonical height defined by the<br />
ample divisor nH − mE for n, m ∈Æ, n > m.<br />
1.18. Remark. –––– For abelian varieties, our proof coincides with that of<br />
Szpiro, Ullmo, <strong>and</strong> Zhang. In the case of projective space, we provide a more<br />
geometric proof for Bilu’s theorem.<br />
2. The dynamics of f<br />
i. Arakelovian interpretation of the canonical height. —<br />
2.1. –––– Let P, f , L , <strong>and</strong> Φ are as described in 1.8.<br />
To underst<strong>and</strong> the canonical height better, one has to use the language of<br />
Arakelov geometry introduced in the previous chapter.
Sec. 2] THE DYNAMICS OF f 193<br />
First, recall that, in Definition 1.9, we may replace h L by any other height<br />
function which differs only by a bounded summ<strong>and</strong>. For example, we may<br />
work with h ′ L (x) := 1 m h l 2 (<br />
iL ⊗m(x) ) .<br />
Further, there is a projective model P of P over Specto which L ⊗m extends.<br />
Indeed, L ⊗m defines a closed embedding i L ⊗m : P → P N <strong>and</strong> one may<br />
put P := i L ⊗m(P).<br />
Then, L := O(1)| P is an extension of L ⊗m to P. We equip L with the<br />
restriction of the Fubini-Study metric. This guarantees that, for every closed<br />
point x ∈ P,<br />
h L ,0 (x) := h L (x) = 1 m h (L ,‖ .‖)(x) .<br />
Here, h (L ,‖ .‖) is the absolute height defined by the smooth hermitian line bundle<br />
(L , ‖ . ‖), cf. Definition IV.5.7.<br />
The recursion is given by<br />
h L ,i+1 (x) := 1 d h L ,i(<br />
f (x)<br />
)<br />
.<br />
Unfortunately, the isomorphism Φ: L ⊗d ∼ =<br />
−→ f ∗ L does not extend to P. If it<br />
would then we could put ‖ . ‖ 0 := ‖ . ‖ <strong>and</strong>, recursively,<br />
This would then lead to<br />
‖ . ‖ i := Φ −1 f ∗ ‖ . ‖ 1/d<br />
i−1 .<br />
h L ,i (x) = 1 m h (L ,‖ .‖ i )(x) .<br />
2.2. –––– To make the plan above precise, one has to work in the more flexible<br />
context of adelic Picard <strong>groups</strong> introduced in Section IV.7.<br />
There is the adelic metric ‖ . ‖ (0)<br />
∼ on L ⊗m given by<br />
(L ⊗m , ‖ . ‖ (0)<br />
∼ ) := i P(L , ‖ . ‖ 0 ) .<br />
Since L is very ample <strong>and</strong> ‖ . ‖ 0 is a restriction of the Fubini-Study metric, we<br />
see that (L ⊗m , ‖ . ‖ ∼ (0))<br />
∈ C ≥0<br />
P .<br />
Further, we define, recursively,<br />
‖ . ‖ ∼ (i) := (Φ⊗m ) −1 f ∗[ ]<br />
‖ . ‖ ∼<br />
(i−1) 1<br />
d<br />
.<br />
Lemma IV.7.30.a) shows (L ⊗m , ‖ . ‖ ∼ (i) ) ∈ C ≥0<br />
P<br />
for every i ∈Æ.<br />
We also observe that the isomorphism Φ does extend to an open neighbourhood<br />
of the generic fiber of P. Therefore, the sequence (‖ . ‖ ∼ (i) ) i is actually constant<br />
at all but finitely many valuations.
194 DISTRIBUTION OF SMALL POINTS [Chap. V<br />
Furthermore, Lemma IV.7.18 implies that<br />
δ ( (L ⊗m , ‖ . ‖ (i)<br />
∼ ), (L ⊗m , ‖ . ‖ (i+1)<br />
∼ )) ≤ 1 d i ·δ( (L ⊗m , ‖ . ‖ (0)<br />
∼ ), (L ⊗m , ‖ . ‖ (1)<br />
∼ )) .<br />
This means, for every ν ∈ Val(É), the sequence (‖ . ‖ (i)<br />
ν ) i of metrics is uniformly<br />
convergent.<br />
The limit ‖ . ‖ ∼ clearly fulfills all the requirements of Definition IV.7.12.<br />
I.e., ‖ . ‖ ∼ is an adelic metric on L ⊗m . Even more,<br />
L := (L ⊗m , ‖ . ‖ ∼ ) ∈ C ≥0<br />
P<br />
⊆ ˜Pic(P) .<br />
By Lemma IV.7.27, ‖ . ‖ νp is semipositive for every prime number p ≠ ∞ <strong>and</strong><br />
‖ . ‖ ν∞ is almost semiample.<br />
Finally, we have<br />
h f ,L = 1 m h L = 1 m h (L ⊗m ,‖ .‖ ∼ ) .<br />
2.3. Lemma. –––– Let P, f , L , <strong>and</strong> Φ are as described in 1.8. Then, f is<br />
automatically finite of degree d dimP .<br />
Proof. P is a projective variety overÉ<strong>and</strong> f : P → P is a morphism such<br />
that L ⊗d ∼ = f ∗ L for some ample L ∈ Pic(P).<br />
In particular, f ∗ L is ample which implies f is quasi-finite. Since f is a projective<br />
morphism, this is sufficient for finiteness. Further, we have deg L<br />
P ≠ 0<br />
<strong>and</strong> deg f ∗ L P = deg L<br />
P = d dimP deg ⊗d L<br />
P.<br />
□<br />
2.4. Lemma. –––– Let P, f , L , <strong>and</strong> Φ are as described in 1.8. Then,<br />
i) p P Z(<br />
(L ⊗m , ‖ . ‖ ∼ ), . . . , (L ⊗m , ‖ . ‖ ∼ ) ) = 0,<br />
ii) e 1 (h L<br />
) = . . . = e dim P+1 (h L<br />
) = 0.<br />
Proof. i) For every i ∈Æ, write<br />
(<br />
P i := pZ<br />
P (L ⊗m , ‖ . ‖ ∼ (i) ), . . . , (L ⊗m , ‖ . ‖ ∼ (i) )) .<br />
Then, since f is of degree d dimP , Lemma 7.33 shows<br />
(<br />
d dimP P i−1 := pZ<br />
P ( f ∗ L ⊗m , f ∗ ‖ . ‖ ∼<br />
(i−1) ), . . . , ( f ∗ L ⊗m , f ∗ ‖ . ‖ ∼ (i−1) ) ) = 0 .<br />
Applying the isomorphism Φ ⊗m <strong>and</strong> dividing each of the (dim P +1) operators<br />
by d yields<br />
P i−1<br />
d<br />
= ( pP Z (L ⊗m , ‖ . ‖ ∼ (i) ), . . . , (L ⊗m , ‖ . ‖ ∼ (i) )) = P i .<br />
(P i ) i is, therefore, a zero sequence.<br />
ii) Lemma IV.6.3 implies that there is a constant C such that<br />
h L ,0 (x) > C
Sec. 2] THE DYNAMICS OF f 195<br />
for everyx ∈ P. Consequently, h L ,i (x) > C/d i <strong>and</strong> h L<br />
(x) ≥ 0 for everyx ∈ P.<br />
This means<br />
e dim P+1 (h L<br />
) ≥ 0 .<br />
Corollary IV.7.40 together with the inequality<br />
e 1 (h L<br />
) ≥ . . . ≥ e dim P+1 (h L<br />
)<br />
implies the claim.<br />
□<br />
2.5. Remark. –––– As the construction of ‖ . ‖ ∼ depends on the iterated pullbacks<br />
under f , one has to underst<strong>and</strong> the dynamics of the system ( f i ) i≥1 . It will<br />
turn out to be sufficient for our purposes to pay attention to the infinite place.<br />
I.e., to the dynamics of that system on P().<br />
ii. Analyzing the dynamics of f . —<br />
2.6. –––– Let us analyze the special situation described in 1.13.<br />
i) G ⊆ P is an open dense subset being a group scheme such that f has a<br />
restriction m := f | G : G → G which is a homomorphism of group schemes.<br />
Therefore, kerm is a group scheme of finite order d dimP .<br />
As f is surjective, im m is a priori Zariski dense. Since m is a homomorphism,<br />
it is surjective, too.<br />
There is a compact subgroup K ⊆ G() which is both forward <strong>and</strong> backward<br />
invariant under m. In particular, all the finite <strong>groups</strong> ker(m◦ . . . ◦m) are<br />
contained in K.<br />
ii) All eigenvalues of the tangent map T e m have absolute value strictly bigger<br />
than 1. Therefore, on G(), there is a left invariant Riemannian metric µ<br />
such that<br />
q := ‖(T e m) −1 ‖ max < 1 .<br />
Indeed, based on the Jordan normal form, one finds easily a hermitian scalar<br />
product on T e G() satisfying the analogous inequality. We take its real part.<br />
By transport of structure, we find a left invariant Riemannian metric µ on G().<br />
We will denote by δ the metric on G() given by the lengths of the<br />
µ-shortest paths.<br />
2.7. Convention. –––– In order to simplify notation, we will often write f<br />
<strong>and</strong> m instead of f<strong>and</strong> mwhen there is no danger of confusion.
196 DISTRIBUTION OF SMALL POINTS [Chap. V<br />
2.8. Lemma. –––– a) m induces a mapping<br />
which is a diffeomorphism.<br />
m : G()/K → G()/K<br />
b) m is exp<strong>and</strong>ing for the Riemannian metric µ on G()/K induced by µ.<br />
More precisely,<br />
for any z 1 , z 2 ⊆ G()/K.<br />
δ ( m(z 1 ), m(z 2 ) ) ≥ 1 q δ(z 1, z 2 )<br />
Proof. a) m is well-defined since K is forward invariant under m.<br />
Backward invariance of K implies that m is even an injection. Further,<br />
as m : G() → G() is a surjection, the same is then true for m.<br />
Finally, the tangent map<br />
T e m : T e G()/K → T e G()/K<br />
is exp<strong>and</strong>ing, too. In particular, T e m is invertible. This yields that m is a<br />
local diffeomorphism.<br />
b) The preimage m −1 (w) of a path w in G()/K is a path, too. All tangent<br />
maps of m are exp<strong>and</strong>ing by a factor ≥ 1 q . Therefore,<br />
l ( m −1 (w) ) ≤ q · l(w) .<br />
This implies the claim.<br />
□<br />
2.9. Corollary. –––– Let G() = G()∪{∞} be the one-point compactification<br />
of G().<br />
Then, for each x ∈ G()\K, the sequence (x i ) i such that x 0 = x <strong>and</strong> x i+1 = m(x i )<br />
converges to ∞. In other words, K is the Julia set of the dynamical system (m i ) i≥1<br />
on G().<br />
Proof. We have to show that for each compact set L ⊆ G() there are only<br />
finitely many i ∈Æsuch that x i ∈ L. Replacing L by KL, if necessary, we may<br />
assume that L is left K-invariant.<br />
This means, we are given a compact set L ⊆ G()/K <strong>and</strong> have to verify that<br />
x i ∈ L for only finitely many values of i.<br />
As m : G()/K → G()/K is exp<strong>and</strong>ing <strong>and</strong> x 0 ≠ [K ], we have<br />
δ(x i , [K ]) → ∞ .<br />
However, L is a compact set. In particular, L is bounded.<br />
□
Sec. 2] THE DYNAMICS OF f 197<br />
2.10. Corollary. –––– The union over all the <strong>groups</strong> ker(m ◦ . . . ◦ m) is dense<br />
in K.<br />
Proof. Otherwise, the topological closure of<br />
K ′ := [ ker(m ◦ . . . ◦ m)<br />
} {{ }<br />
i<br />
i times<br />
would be a compact Lie group properly contained in K. It is obvious that K ′ is<br />
forward <strong>and</strong> backward invariant under f. Therefore, the map induced by f<br />
on the compact homogeneous space K/K ′ would be injective <strong>and</strong> exp<strong>and</strong>ing.<br />
□<br />
2.11. Notation. –––– For N ∈Ê, we will write U N (K ) for the set<br />
{x ∈ G() | δ(x, K ) ≤ N } .<br />
2.12. Proposition –––– Let µ i be the measure induced by the smooth differential<br />
form<br />
c 1 (L , ‖ . ‖ ∞) (i) ∧ . . . ∧ c 1 (L , ‖ . ‖ ∞)<br />
(i)<br />
of type (dim P, dimP) on P(). Then, the sequence (µ i ) i converges weakly to<br />
the measure µ which is the zero measure on P()\K <strong>and</strong> the Haar measure of<br />
volume deg L<br />
P on K.<br />
Proof. First step. Some observations.<br />
Each µ i is of volume deg L<br />
P. Further, the sequence (µ i ) i obeys the recursive low<br />
Second step. µ i | P()\K → 0.<br />
µ i+1 = 1<br />
d dim P m∗ µ i .<br />
For this, let g ∈ C 0 (P() \K) be a continuous function with compact support.<br />
We consider the sequence (g i ) i such that g 0 := g <strong>and</strong> g i+1 := 1 f<br />
d dimP ∗ g i . Here, f ∗ ,<br />
the push-forward of a function, is given by summation over the fibers of f .<br />
By definition, µ i (g) = µ 0 (g i ). Thus, let us show µ 0 (g i ) → 0 for i → ∞.<br />
For this, we note that |g| is bounded by some constant C. This implies |g i | ≤ C<br />
for every i. Further, there is some ε > 0 such that supp(g) ⊆ P()\U ε (K ).<br />
Consequently,<br />
Z<br />
|µ i (g)| = |µ 0 (g i )| ≤ C χ (P()\U ε/q i ) dµ 0 .<br />
Here, χ A denotes the characteristic function of a measurable set A. The integr<strong>and</strong><br />
converges monotonically to χ P()\G() which is of integral zero. Therefore,<br />
the Theorem of Beppo Levi yields µ i (g) → 0.<br />
P()
198 DISTRIBUTION OF SMALL POINTS [Chap. V<br />
Third step. The assertion in general.<br />
Let g ∈ C 0 (G()) be a continuous function with compact support. Again, we<br />
put g 0 := g <strong>and</strong> g i+1 := 1 m<br />
d dimP ∗ g i such that we have µ i (g) = µ 0 (g i ). As the sequence<br />
(g i ) i is uniformly bounded, it suffices to show that it converges pointwise<br />
to the constant<br />
I := 1 Z<br />
g dρ .<br />
volK<br />
We claim that (g i ) i converges even uniformly in every compact set A ⊆ G().<br />
For this, we may assume without loss of generality that A = U N (K ) for<br />
some N ∈Ê. Further, note that<br />
g i (x) =<br />
K<br />
1<br />
♯ kerm i<br />
∑<br />
k∈kerm i g(ky)<br />
for any y such that f (y) = x. Here, we clearly have (m i ) −1 (A) ⊆ U Nq i(K )<br />
<strong>and</strong> Nq i → 0 for i → ∞.<br />
In addition, g is uniformly continuous on A <strong>and</strong> m is exp<strong>and</strong>ing. Thus, for each<br />
ε > 0, there is some i ∈Æsuch that, for two arbitrary <strong>points</strong> x, x ′ ∈ A, one<br />
can find y, y ′ such that f (y) = x, f (y ′ ) = x ′ , <strong>and</strong> δ(y, y ′ ) < ε. Consequently,<br />
max g i | A − min g i | A<br />
tends to zero for i → ∞.<br />
Finally, for every i ∈Æ, I is the mean value of g i on K.<br />
□<br />
iii. Some further observations. —<br />
2.13. Caution. –––– It is not true in general that the curvature current, given by<br />
c 1 (L, ‖ . ‖ ∞ ) := 1 ∂∂ log ‖s‖2<br />
2πi<br />
for a non-zero <strong>rational</strong> section of L, vanishes on P() \K. Nevertheless, one<br />
has at least the following.<br />
2.14. Lemma. –––– The curvature current c 1 (L, ‖ . ‖ ∞ ) has the properties below.<br />
a) The restriction c 1 (L, ‖ . ‖ ∞ )| G() is left <strong>and</strong> right K-invariant.<br />
b) It satisfies the equation f ∗ c 1 (L, ‖ . ‖ ∞ ) = d · c 1 (L, ‖ . ‖ ∞ ).<br />
Proof. a) By construction, c 1 (L, ‖ . ‖ ∞,i ) is invariant under ker(m i).<br />
Thus, c 1 (L, ‖ . ‖ ∞ ), which is the weak limit of that sequence of currents,<br />
is invariant under S i ker(m i). Invariance under any k ∈ K follows since<br />
k i → k implies that the test forms k i ω converge to kω in the Schwartz space.<br />
b) is clear. □
Sec. 3] PERTURBING ALMOST SEMIAMPLE METRICS 199<br />
3. Perturbing almost semiample metrics<br />
3.1. –––– One of the most natural ways to produce a new adelic metric from<br />
an old one is to replace ‖ . ‖ ∞ by ‖ . ‖ ∞ · exp(−g) for some continuous function<br />
g ∈ C (P()). As we want to use Corollary IV.7.40 as a fundamental tool,<br />
it is necessary to underst<strong>and</strong> for which g this metric is almost semiample.<br />
3.2. Notation. –––– In this section we will continue to use the notations of 1.13.<br />
Further, we denote by S the set of all continuous functions g ∈ P() such that<br />
‖ . ‖ ∞ · exp(−g) is almost semiample.<br />
3.3. Lemma. –––– a) One has C ∈ S for every constant C.<br />
b) Let g 1 , g 2 ∈ S <strong>and</strong> 0 < a < 1. Then, ag 1 + (1 − a)g 2 ∈ S.<br />
c) If g ∈ S then 1 d f ∗ g ∈ S.<br />
d) Let g ∈ S be a function such that supp g ⊆ G(). Then, for each k ∈ K, one<br />
has g · k ∈ S for<br />
{ g(kx) if x ∈ G() ,<br />
(g · k)(x) :=<br />
0 otherwise .<br />
e) S is closed under uniform convergence.<br />
Proof. e) is trivial.<br />
a) ‖ . ‖ ∞ is almost semiample. Thus, there is a sequence (‖ . ‖ i ∞ ) i of smooth<br />
hermitian metrics with strictly positive curvature on Lwhich is uniformly<br />
convergent to ‖ . ‖ ∞ . The metrics ‖ . ‖ ∞,i · exp(−C ) have the same curvatures<br />
<strong>and</strong> converge uniformly to ‖ . ‖ ∞ · exp(−C ).<br />
b) Let (‖ . ‖ i ′) i <strong>and</strong> (‖ . ‖ i ′′)<br />
i be sequences uniformly convergent versus<br />
‖ . ‖ ∞ · exp(−g 1 ) <strong>and</strong> ‖ . ‖ ∞ · exp(−g 2 ), respectively. Put<br />
‖ . ‖ i := ‖ . ‖ ′a<br />
i<br />
′′ (1−a)<br />
· ‖ . ‖ i .<br />
These metrics have positive curvature <strong>and</strong>, uniformly,<br />
‖ . ‖ i → ‖ . ‖ ∞ · exp ( − ag 1 − (1 − a)g 2<br />
)<br />
.<br />
c) Put ‖ . ‖ ∞ ′ := ‖ . ‖ ∞ · exp(−g). Then,<br />
(<br />
Φ −1 f ∗ ‖ . ‖ ′ d<br />
1<br />
∞ = Φ −1 f ∗ d<br />
‖ . ‖ 1 ∞ · exp − 1 )<br />
d f ∗ g<br />
(<br />
= ‖ . ‖ ∞· exp − 1 )<br />
d f ∗ g .<br />
Therefore, if (‖ . ‖ ∞,i ′ ) i is a sequence of metrics with positive curvature<br />
which converges uniformly to ‖ . ‖ ∞ ′ then (Φ −1 f ∗ ‖ . ‖ ′ 1 d<br />
∞,i) i converges uniformly<br />
to ‖ . ‖ ∞ · exp(− 1 f ∗ g).<br />
d<br />
d) We denote by<br />
e k : G() → G()
200 DISTRIBUTION OF SMALL POINTS [Chap. V<br />
multiplication by k from the left.<br />
By e), we may assume that k ∈ ker(m j ) for some j ∈Æ. To show<br />
‖ . ‖ ∞ · exp(−g · k)<br />
is almost semiample, it suffices to consider the hermitian metric<br />
‖ . ‖ d j<br />
∞ · exp(−d j g · k)<br />
on L.<br />
⊗d j<br />
Let us consider the restriction to G(), first. Iterated application of Φ induces<br />
an isomorphism<br />
e ∗ k L | ⊗d j ∼ G() = e ∗ k ( f j ) ∗ L| ∼ G() = ( f j ) ∗ L| ∼ ⊗d j<br />
G() = L| G() . (∗)<br />
Here, the isomorphism in the middle is canonical as f j ◦e k = f j .<br />
By construction 2.2, ‖ . ‖ ∞ is the uniform limit of a sequence (‖ . ‖ ∞ (i))<br />
i such that,<br />
for every i ≥ j,<br />
e ∗ k<br />
[ ] ‖ . ‖<br />
(i) d j<br />
∞<br />
under the identification (∗). By consequence,<br />
= [ ]<br />
‖ . ‖ ∞<br />
(i) d j<br />
e ∗ k‖ . ‖ d j<br />
∞ = ‖ . ‖ d j<br />
∞ ,<br />
too, which shows<br />
[ ‖ . ‖<br />
d j<br />
∞ · exp(−d j g) ] = ‖ . ‖ d j<br />
∞ · exp(−d j g · k) .<br />
e ∗ k<br />
As almost semiampleness is preserved under holomorphic pull-back, we see that<br />
‖ . ‖ d j<br />
∞ · exp(−d j g · k) is almost semiample on G().<br />
On the other h<strong>and</strong>, there is some neighbourhood U of P() \G() such<br />
that g| U = 0. This implies −d j g · k = 0 <strong>and</strong>, therefore,<br />
‖ . ‖ d j<br />
∞ · exp(−d j g · k) = ‖ . ‖ d j<br />
∞<br />
on U. In particular, ‖ . ‖ d j<br />
∞ · exp(−d j g · k) is almost semiample on U.<br />
The assertion follows from Proposition 3.4.<br />
□<br />
3.4. Proposition –––– Let M be a projective complex manifold, G ∈ Pic(M ) be<br />
ample, <strong>and</strong> {U 1 , . . . , U n } an open covering. Let ‖ . ‖ 1 , . . . , ‖ . ‖ n be continuous<br />
hermitian metrics on G | U1 , . . . , G | Un such that, for some δ > 0, the<br />
sets D 1,δ , . . . , D n,δ , given by<br />
D i,δ := {x ∈ U i | ‖ . ‖ i,(x) ≤ (1 + δ) · min{‖ . ‖ 1,(x) , . . . , ‖ . ‖ n,(x) }} ⊆ U i
Sec. 3] PERTURBING ALMOST SEMIAMPLE METRICS 201<br />
are compact. Assume, each ‖ . ‖ i is almost semiample.<br />
Then, the continuous hermitian metric ‖ . ‖ := min i ‖ . ‖ i is almost semiample.<br />
Proof. We may assume that G is very ample.<br />
For every i, let (‖ . ‖ ij ) j be a sequence of smooth <strong>and</strong> positively curved metrics<br />
on G | Ui which converges uniformly to ‖ . ‖ i . Put<br />
‖ . ‖ − j<br />
:= min{‖ . ‖ 1j , . . . , ‖ . ‖ nj } .<br />
For j ≫ 0, ‖ . ‖ − j<br />
is a continuous hermitian metric on the whole of G . The sequence<br />
(‖ . ‖ − j ) j converges uniformly to ‖ . ‖.<br />
We choose an embedding i : M ֒→ P N such that i ∗ O(1) ∼ = G . Further, we<br />
extend the smooth metrics ‖ . ‖ ij to smooth metrics ‖ . ‖ ′ ij<br />
on O(1) which are<br />
defined on open subsets W i ⊇ U i of P N .<br />
Under the tautological action PGL n () × P N → P N , each γ ∈ PGL n ()<br />
defines an automorphism e γ : P N → P N such that there is a natural identification<br />
eγ ∗ O(1) ∼ = O(1). We find an open neighbourhood O of e ∈ PGL n () such<br />
that, for every γ ∈ O <strong>and</strong> every i ∈ {1, . . . , n}, the pull-back eγ ∗ ‖ . ‖<br />
ij ′ is<br />
well-defined on D i,δ <strong>and</strong><br />
eγ‖ ∗ . ‖ ′ ij | Ui ∩eγ −1 (W i )<br />
has strictly positive curvature in every point of D i,δ .<br />
We claim that, for each γ ∈ O, the perturbation<br />
‖ . ‖ −,γ<br />
j<br />
:= e ∗ γ(<br />
min{‖ . ‖<br />
′<br />
1j , . . . , ‖ . ‖ ′ nj} ) | M<br />
of ‖ . ‖ − j<br />
has a positive curvature current on each holomorphic curve inside M.<br />
Indeed, this is a local statement. Let x ∈ M. Fix a holomorphic section<br />
0 ≠ s ∈ Γ(U,G) defined in some neighbourhood U of x. Then,<br />
c 1 (G , ‖ . ‖ −,γ<br />
j<br />
)| U = −dd c log ( ‖s‖ −,γ ) 2<br />
j<br />
= dd c( max{− logeγ‖s‖ ∗ ′ 1j 2 , . . . , − loge∗ γ‖s‖ nj} ) ′ 2 .<br />
Observe that the maximum of a finite system of plurisubharmonic functions is<br />
plurisubharmonic, again.<br />
Now use Sobolev’s averaging procedure. For every non-negative smooth function<br />
ϕ ≠ 0 on PGL n () such that supp ϕ ⊆ O, one gets an approximation<br />
/<br />
Z<br />
Z<br />
‖ . ‖ −,ϕ<br />
j<br />
:= ϕ(γ) · ‖ . ‖ −,γ<br />
j<br />
dρ γ ϕ(γ)dρ γ<br />
PGL n ()<br />
PGL n ()<br />
of ‖ . ‖ − j . Here, ρ is the left Haar measure on PGL n().
202 DISTRIBUTION OF SMALL POINTS [Chap. V<br />
Consider a sequence (ϕ k ) k of functions as above which converges weakly to the<br />
delta distribution δ e . Every ‖ . ‖ −,ϕ k<br />
j<br />
is smooth <strong>and</strong> positively curved. The sequence<br />
(‖ . ‖ −,ϕ k) k k<br />
converges uniformly to ‖ . ‖.<br />
3.5. Corollary. –––– If g 1 , g 2 ∈ S then g := max{g 1 , g 2 } ∈ S.<br />
3.6. Corollary. –––– Let l ∈Æ, 0 < a < 1, <strong>and</strong> 0 ≠ s ∈ Γ(P(), L). ⊗l Then,<br />
Proof. Consider the function g on<br />
g := max{a log ‖s‖ l , 0} ∈ S .<br />
U := {x ∈ P() | log ‖s‖(x) >−1}<br />
given by g(x) := a log ‖s‖ l (x). We have to show that the hermitian metric<br />
‖ . ‖ ′ ∞ := ‖ . ‖l ∞ · exp(−g)<br />
on L| ⊗l U is the limit of a uniformly convergent sequence of smooth metrics<br />
with positive curvature.<br />
We put ‖ . ‖ ′ ∞,i := ‖ . ‖l ∞,i/ ( ‖s‖ l ∞,i) a. It is clear that, uniformly,<br />
Further,<br />
‖ . ‖ ′ ∞,i → ‖ . ‖ ′ ∞ .<br />
c 1 (L| U , ‖ . ‖ ′ ∞,i ) = ddc (− log ‖s‖ ′2 ∞,i )<br />
= dd c( − log(‖s‖ l ∞,i )2(1−a))<br />
= (1 − a)l · c 1 (L| U , ‖ . ‖ ∞,i ) > 0 . □<br />
□<br />
3.7. Lemma. –––– Let g ∈ S suchthatsupp g ⊆ G(). Consider the K-invariant<br />
function g given by<br />
{<br />
max g(kx) if x ∈ G() ,<br />
g(x) := k∈K<br />
g(x) otherwise .<br />
Then, g ∈ S, too.<br />
Proof. g is the limit of the uniformly convergent sequence (g i ) i , defined by<br />
g i (x) := max g(kx) .<br />
k∈ker(m i )<br />
□
Sec. 3] PERTURBING ALMOST SEMIAMPLE METRICS 203<br />
3.8. Proposition –––– Let ε > 0 <strong>and</strong> U ⊆ G() an open set containing K.<br />
Then, there is some non-negative K-invariant function 0 ≠ g ∈Ê+·S such that<br />
i) g(e) = 1,<br />
ii) max x∈P() g(x) ≤ 1 + ε, <strong>and</strong><br />
iii) supp g ⊆ U.<br />
Proof. Put D := P \G. For some j ≫ 0, the coherent sheaf L ⊗j ⊗I D has a<br />
section which does not vanish in e ∈ G. I.e., there is a section s ∈ Γ(P, L ⊗j )<br />
vanishing in D but not in e.<br />
Using Corollary 3.6, we see<br />
1<br />
}<br />
˜g C := max{<br />
2j log ‖Cs‖j , 0<br />
{ 1<br />
= max<br />
2j log ‖s‖j + logC }<br />
, 0 ∈ S<br />
2j<br />
for every C > 0. It is clear that, for each C, supp ˜g C ⊆ G() is a compact set.<br />
We put<br />
<strong>and</strong><br />
A := max<br />
x∈K<br />
1<br />
2j log ‖s‖j (x)<br />
1<br />
B := max<br />
x∈P() 2j log ‖s‖j (x) .<br />
Then, we choose C such that log C + A > 0 <strong>and</strong><br />
logC + B<br />
log C + A ≤ 1 + ε .<br />
Further, let g 0 be the K-invariant function associated to ˜g C . I.e.,<br />
{<br />
max ˜g C (kx) if x ∈ G() ,<br />
g 0 := k∈K<br />
g(x) otherwise .<br />
By construction, g 0 (e) > 0 <strong>and</strong><br />
max g 0 (x)/g 0 (e) ≤ 1 + ε .<br />
x∈P()<br />
Further, supp g 0 ⊆ K ·supp ˜g C is compact. Lemma 3.7 shows that g 0 ∈ S.<br />
Finally, we define a sequence (g i ) i of functions on P() by putting, recursively,<br />
g i+1 := 1 d f ∗ g i . Then, one clearly has g i (e) = g 0 (e) > 0 <strong>and</strong><br />
max g i(x)/g i (e) = max g 0(x)/g 0 (e) ≤ 1 + ε<br />
x∈P() x∈P()<br />
for every i ∈Æ. Lemma 3.3.c) implies that g i ∈ S.<br />
As supp g 0 is compact, there is some N ∈Êsuch that supp g 0 ⊆ U N (K ).<br />
This yields, by Lemma 2.8.b), that supp g i ⊆ U Nq i(K ).<br />
Therefore, the function g, given by g(x) := g i (x)/g i (e) for some i ≫ 0, has all<br />
the properties desired.<br />
□
204 DISTRIBUTION OF SMALL POINTS [Chap. V<br />
4. Equidistribution<br />
4.1. Definition. –––– Let X be a projective variety overÉ. Then, a sequence<br />
(x i ) i of closed <strong>points</strong> on X is called generic if no infinite subsequence is<br />
contained in a proper closed subvariety of X.<br />
4.2. Remark. –––– In other words, (x i ) i is generic if it converges to the<br />
generic point with respect to the Zariski topology.<br />
4.3. Definition. –––– Let X be a projective variety overÉ<strong>and</strong> L ∈ ˜Pic(X ).<br />
Suppose that e dim X+1 (h L<br />
) = 0.<br />
Then, a sequence (x i ) i of closed <strong>points</strong> on X is called small if h L<br />
(x i ) → 0.<br />
4.4. Lemma. –––– Let X be a projective variety overÉ<strong>and</strong><br />
L = (L , ‖ . ‖) ∈ ˜Pic(X )<br />
be such that ‖ . ‖ p is semipositive for every prime number p <strong>and</strong> ‖ . ‖ ∞ is almost semiample.<br />
Assume e dim X+1 (L ) = 0.<br />
Then, pZ X (L , . . . , L ) = 0 is equivalent to the existence of a sequence of closed<br />
<strong>points</strong> on X which is generic <strong>and</strong> small.<br />
Proof. “=⇒” There are not more than countably many closed subvarieties<br />
X 1 , X 2 , . . . ⊂ X. Further, Corollary IV.7.40 implies e 1 (h L<br />
) = 0. Therefore,<br />
we may choose a sequence (x i ) i of closed <strong>points</strong> on X such that<br />
x i ∈ X \X 1 \X 2 \ . . . \X i<br />
<strong>and</strong> h L<br />
(x i ) < 1 i . (x i) i is generic <strong>and</strong> small.<br />
“⇐=” Let (x i ) i be a sequence of closed <strong>points</strong> on X which is generic <strong>and</strong> small.<br />
Let Y ⊂ X be a closed subset of codimension one. Then, only finitely many<br />
of the x i are contained in Y. Hence, inf x∈X \Y h L<br />
(x) = 0. As this is true for<br />
every Y, we see e 1 (h L<br />
) = 0. Corollary IV.7.40 yields pZ X (L , . . . , L ) = 0.<br />
□<br />
4.5. Lemma. –––– Let P, f , L = (L ⊗m , ‖ . ‖ ∼ ), <strong>and</strong> Φ be as described<br />
in 1.13. Denote by S the set of all continuous functions g ∈ C (P()) such that<br />
‖ . ‖ ∞· exp(−g) is almost semiample.<br />
Let ϕ ∈ C (P()). Then, for every ε > 0, there exist a function ϕ 1 ∈ C (P())<br />
supported in P()\K <strong>and</strong> a function ϕ 2 ∈Ê+·S such that<br />
‖ϕ − ϕ 1 − ϕ 2 ‖ max < ε .
Sec. 4] EQUIDISTRIBUTION 205<br />
Proof. The set<br />
T := {h ∈ C (K ) | h = ϕ| K for some ϕ ∈Ê+·S}<br />
fulfills all the assumptions of Sublemma 4.6, except for closedness under uniform<br />
convergence.<br />
Indeed, Lemma 3.3.a) <strong>and</strong> b) imply that assumptions i) <strong>and</strong> ii) are satisfied.<br />
Corollary 3.6 is general enough to guarantee assumption iii). Lemma 3.3.d)<br />
yields assumption iv). Finally, Corollary 3.5 implies that assumption v) is fulfilled.<br />
Therefore, there is some ϕ 2 ∈Ê+·S such that |ϕ(x)−ϕ 2 (x)| < ε for every x ∈ K.<br />
2<br />
A decomposition of the unit adapted to a suitable open covering {P()\K, U }<br />
of P() yields some ϕ 1 ∈ C (P()) such that supp ϕ 1 ⊆ P()\K <strong>and</strong><br />
∣ ( ϕ(x) − ϕ 2 (x) ) − ϕ 1 (x) ∣ < ε<br />
for every x ∈ P().<br />
□<br />
4.6. Sublemma. –––– Let L be a compact group <strong>and</strong> T ⊆ C (L) a set of continuous<br />
real-valued functions such that the following conditions are fulfilled.<br />
i) T contains all the constant functions.<br />
ii) T is closed under addition <strong>and</strong> multiplication by positive constants.<br />
iii) For each x ∈ L such that x ≠ e, there is some g ∈ T fulfilling<br />
g(e) > g(x) .<br />
iv) For each x ∈ L <strong>and</strong> g ∈ T , the shift g · x is also in T .<br />
v) If g ∈ T then g + ∈ T for<br />
g + (x) := max {g(x), 0} .<br />
vi) T is closed under uniform convergence.<br />
Then, T = C (L) contains all continuous functions.<br />
Proof. We fix a Haar measure ρ on L. Then, the conditions ii,) iv) <strong>and</strong> vi)<br />
together imply the following.<br />
(†) If h ∈ T <strong>and</strong> g is a non-negative, measurable, <strong>and</strong> bounded function then,<br />
for the convolution, we have g∗h ∈ T .
206 DISTRIBUTION OF SMALL POINTS [Chap. V<br />
Hence, it suffices to show that, for each open set U ⊆ L containing e, there<br />
is some non-negative function g U ∈ T such that g U ≠ 0 <strong>and</strong> supp g U ⊆ U.<br />
For this, by i), ii), <strong>and</strong> v), we only need a function g ∈ T having its maximum<br />
entirely in U.<br />
Assume, for some open U 0 , there would be no such function. Then, for<br />
every g ∈ T having its maximum in e, there is a non-empty set<br />
A g := {y ∈ L\U 0 | g(y) = g(e)} ⊆ L\U 0<br />
where this maximum is taken, too. If<br />
\<br />
g∈T<br />
A g = ∅<br />
then, by compactness, already a finite intersection A g1 ∩ . . . ∩ A gn would<br />
be empty. Then, g 1 + . . . + g n ∈ S had its maximum in U 0 , only.<br />
Consequently,<br />
is a non-empty set. Put<br />
A U0 := \<br />
g∈T<br />
A := {e} ∪ [<br />
A g<br />
U ∋e,U open<br />
By construction, this set has the property below.<br />
A U .<br />
(‡) Every function g ∈ T assuming its maximum in e is necessarily constant<br />
on A. Further, A ⊆ L is the largest set containing e with this property.<br />
The property (‡) implies that A is a closed set.<br />
For y ∈ A <strong>and</strong> g ∈ T , the shift g·y −1 has its maxima in yA. Therefore, A = yA<br />
for every y ∈ A. This means that A ⊆ L is a subgroup.<br />
Fix a Haar measure ρ A on A <strong>and</strong> denote by i : A → L the natural inclusion.<br />
Property (‡) together with i), ii) <strong>and</strong> v) implies that there is a sequence of<br />
functions of T such that the associated distributions converge weakly to i ∗ ρ A .<br />
Then, property (†) guarantees together with vi) that every A-left invariant<br />
continuous function is an element of T .<br />
Now, we use condition iii). Choose some x 0 ∈ A different from e. There exists<br />
some g ∈ T such that g(e) > g(x 0 ). Shifting by an element of A, if necessary,<br />
we may assume that g(e) = max x∈A g(x).<br />
Define another continuous function M on L by<br />
M (x) := − max<br />
a∈A g(ax) .
Sec. 4] EQUIDISTRIBUTION 207<br />
Obviously, M is A-invariant. Thus, M ∈ T . Consequently, ˜g := g + M ∈ T ,<br />
too.<br />
However, ˜g(x) ≤ 0 for every x ∈ L, ˜g(e) = 0, <strong>and</strong> ˜g(x 0 ) < 0 together form a<br />
contradiction to property (‡).<br />
□<br />
4.7. Proposition. –––– Let P, f , L , <strong>and</strong> Φ be as described in 1.13.<br />
Let (x i ) i be a sequence of closed <strong>points</strong> on P which is generic <strong>and</strong> small. Assume, some<br />
non-negative ϕ ∈ C (P()) fulfills either supp ϕ ⊆ P()\K or ϕ ∈Ê+·S.<br />
Then,<br />
lim inf<br />
i→∞<br />
Z<br />
P()<br />
ϕ dδ xi ≥<br />
Z<br />
P()<br />
ϕ dτ .<br />
Here, τ is the zero measure on P() \K <strong>and</strong> the Haar measure of volume one on K.<br />
Proof. For ϕ such that supp ϕ ⊆ P() \K, the right h<strong>and</strong> side is zero. Thus, in<br />
this case, the assertion is clear.<br />
Let ϕ ∈Ê+·S. Then, for any positive λ ∈Ê, let ‖ . ‖ ∼ λ be the adelic metric<br />
on L ⊗m given by ‖ . ‖ λ p := ‖ . ‖ p for the non-archimedean valuations <strong>and</strong><br />
by ‖ . ‖ ∞ λ := ‖ . ‖ ∞ · exp(−λϕ) for the archimedean valuation.<br />
For λ → 0, the adelic metric ‖ . ‖ λ fulfills the assumptions of Corollary IV.7.40.<br />
Further, one clearly has<br />
Z<br />
h (L ⊗m ,‖ .‖ λ )(x i ) = h L<br />
(x i ) + λ ϕ dδ xi .<br />
As (x i ) i is generic <strong>and</strong> h L<br />
(x i ) → 0, we obtain, according to the very definition<br />
of e 1 ,<br />
Z<br />
λ·liminf ϕ dδ xi ≥ e 1 (h (L ⊗m ,‖.‖ λ )) . (§)<br />
On the other h<strong>and</strong>,<br />
=<br />
1<br />
(dim X + 1)c 1 (L ⊗m<br />
i→∞<br />
X ()<br />
É)<br />
dim X pX Z<br />
1<br />
(dim X + 1)m dimX deg L<br />
X pX Z (L , . . . , L )<br />
X ()<br />
(<br />
(L ⊗m , ‖ . ‖ λ ∼ ), . . . , (L ⊗m , ‖ . ‖ λ ∼ ))<br />
1<br />
+<br />
m dim X deg L<br />
X ·<br />
(<br />
(OX , ‖ . ‖ ∞· exp(−λϕ)), (L ⊗m , ‖ . ‖ ∼ ), . . . , (L ⊗m , ‖ . ‖ ∼ ) )<br />
+ O(λ 2 ) .<br />
p X Z
208 DISTRIBUTION OF SMALL POINTS [Chap. V<br />
By virtue of Lemma 4.4, we have p X Z (L , . . . , L ) = 0. Further,<br />
1 (<br />
m dimX deg L<br />
X ·pX Z (OX , ‖ . ‖ ∞· exp(−λϕ)), (L ⊗m , ‖ . ‖ ∼ ), . . . , (L ⊗m , ‖ . ‖ ∼ ) )<br />
= λ lim<br />
i→∞<br />
I i (ϕ)<br />
for<br />
I i (ϕ) := 1 ( )<br />
m dim X pX Z (OX , ‖ . ‖ ∞· exp(−ϕ)), (L ⊗m , ‖ . ‖ ∼ (i) ), . . . , (L ⊗m , ‖ . ‖ ∼ (i) ) .<br />
} {{ }<br />
dim X times<br />
At this level, we can make the arithmetic intersection product explicit <strong>and</strong> find<br />
I i (ϕ) =<br />
Z<br />
P()<br />
ϕc 1 (L , ‖ . ‖ (i)<br />
∞) ∧ . . . ∧ c 1 (L , ‖ . ‖ (i)<br />
∞) .<br />
Corollary IV.7.40 therefore yields in view of Proposition 2.12<br />
Z<br />
e 1 (h (L ⊗m ,‖.‖ λ )) ≥ λ· ϕ dτ . ()<br />
P()<br />
The equations (§) <strong>and</strong> () together imply the assertion.<br />
4.8. Remark. –––– The argument for the case ϕ ∈Ê+·S is due to E. Ullmo.<br />
Cf. [S/U/Z] or [Zh98].<br />
4.9. Theorem (Equidistribution on P()). —– Let P, f , L , <strong>and</strong> Φ be as described<br />
in 1.13.<br />
Then, for each sequence (x i ) i of closed <strong>points</strong> on P which is generic <strong>and</strong> small, the<br />
associated sequence of <strong>measures</strong> (δ xi ) i converges weakly to the measure τ which is the<br />
zero measure on P()\K <strong>and</strong> the Haar measure of volume one on K.<br />
Proof. First step. δ xi | P()\K → 0.<br />
By Proposition 3.8, for every ε 1 , ε 2 > 0, there is some non-negative g ε1 ,ε 2<br />
∈ S<br />
such that supp g ε1 ,ε 2<br />
⊆ U ε1 (K ),<br />
□<br />
<strong>and</strong><br />
Then, by Proposition 4.7,<br />
max<br />
x∈P() g ε 1 ,ε 2<br />
≤ 1 + ε 2 ,<br />
Z<br />
P()<br />
Z<br />
lim inf<br />
i→∞<br />
P()<br />
g ε1 ,ε 2<br />
dτ = 1 .<br />
g ε1 ,ε 2<br />
dδ xi ≥ 1 .
Sec. 4] EQUIDISTRIBUTION 209<br />
This yields<br />
But<br />
Consequently,<br />
Z<br />
lim sup<br />
i→∞<br />
1 − g ε1 ,ε 2<br />
(x) ≥<br />
P()<br />
(1 − g ε1 ,ε 2<br />
) dδ xi ≤ 0 .<br />
{ 1 if x ∈ P()\U ε1 (K ) ,<br />
−ε 2 if x ∈ U ε1 (K ) .<br />
This shows<br />
lim sup<br />
i→∞<br />
[<br />
δxi<br />
(<br />
P()\U ε1 (K ) ) − ε 2 δ xi<br />
(<br />
Uε1 (K ) )] ≤ 0 .<br />
lim sup<br />
i→∞ δ x i<br />
(<br />
P()\U ε1 (K ) ) ≤ ε 2 .<br />
As the latter is true for every ε 2 > 0, we have<br />
lim δ (<br />
x i P()\U ε1 (K ) ) = 0 .<br />
i→∞<br />
Further, this formula is true for each ε 1 > 0. Hence, (δ xi | P()\K ) i converges<br />
weakly to the zero measure.<br />
Second step. The assertion in general.<br />
Let ϕ ∈ C (P()) be an arbitrary continuous function. As we can interchange<br />
the roles of ϕ <strong>and</strong> (−ϕ), it suffices to prove<br />
Z<br />
Z<br />
lim inf ϕ dδ xi ≥ ϕ dτ.<br />
i→∞<br />
P()<br />
By Lemma 4.5, we have continuous functions ϕ 1 <strong>and</strong> ϕ 2 on P() such that<br />
supp ϕ 1 ⊆ P()\K <strong>and</strong> ϕ 2 ∈Ê+·S.<br />
The result of the first step shows<br />
lim<br />
Z<br />
i→∞<br />
P()<br />
i→∞<br />
P()<br />
P()<br />
ϕ 1 dδ xi = 0 .<br />
Further, by Proposition 4.7,<br />
Z<br />
lim inf ϕ 2 dδ xi ≥<br />
Consequently,<br />
Z<br />
lim inf ϕdδ xi ≥<br />
i→∞<br />
P()<br />
Z<br />
P()<br />
The assertion follows since ε > 0 is arbitrary.<br />
Z<br />
P()<br />
ϕ 2 dτ .<br />
ϕdτ − ε .<br />
4.10. Corollary. –––– Let X ⊂ P be a closed subvariety <strong>and</strong> (x i ) i a sequence of<br />
closed <strong>points</strong> on X which is generic <strong>and</strong> small.<br />
□
210 DISTRIBUTION OF SMALL POINTS [Chap. V<br />
Then, the sequence (δ xi | X ()\K) i converges weakly to the zero measure.<br />
Proof. As there are only finitely many i such that x i ∉ G, let us assume that<br />
the whole sequence (x i ) i is contained in G. Then, we choose some K-invariant<br />
metric on G() <strong>and</strong> assume that, for some ε > 0, we would have<br />
lim inf<br />
i→∞ δ x i<br />
(<br />
X ()\U ε (K ) ) = δ > 0 .<br />
We will construct another sequence (y i ) i which is generic <strong>and</strong> small on the<br />
whole of P.<br />
For this, note first that we have<br />
h L<br />
(x) = 1 ( )<br />
h<br />
deg f L f (x) .<br />
Consequently, h L<br />
is invariant under shift by any torsion point<br />
t ∈ K tor =<br />
j∈Æker(m [ j ) .<br />
Further, all the sequences (t·x i ) i fulfill<br />
lim inf<br />
i→∞ δ t·x i<br />
(<br />
P()\U ε (K ) ) = δ .<br />
Finally, K ⊆ G() is assumed to be Zariski dense. Therefore, for each i, the<br />
union S {t·x i } is a Zariski dense subset of P.<br />
t∈K tor<br />
Finally, we make use of the fact there are only countably many proper closed<br />
subvarieties P 0 , P 1 , P 2 , . . . ⊂ P. We can choose a sequence (y i ) i such that, for<br />
every i ∈Æ, one has<br />
y i ∈ [<br />
{t·x i }<br />
<strong>and</strong><br />
t∈K tor ,i≥0<br />
y i ∈ P \P 0 \ . . . \P i .<br />
Then,<br />
δ yi<br />
(<br />
P()\U ε (K ) ) > δ/2<br />
for i ≫ 0 <strong>and</strong> h L<br />
(y i ) → 0. This is a contradiction to Theorem 4.9.<br />
□<br />
4.11. Remark. –––– This corollary shows, in particular, the following.<br />
If X ∩ K = ∅ then<br />
p X Z (L , . . . , L ) > 0 .<br />
In other words, there are no small <strong>and</strong> generic sequences on X.
CHAPTER VI<br />
CONJECTURES ON THE ASYMPTOTICS OF<br />
POINTS OF BOUNDED HEIGHT<br />
I have no satisfaction in formulas unless I feel their numerical magnitude.<br />
WILLIAM THOMSON 1ST BARON KELVIN,<br />
(Life (1943) by Sylvanus Thompson, p. 827)<br />
1. A heuristic<br />
1.1. –––– Let X be a projective variety overÉ. Then, one of the most natural<br />
questions to ask is<br />
(∗) Does X have finitely many or infinitely manyÉ-<strong>rational</strong> <strong>points</strong>?<br />
1.2. –––– If the answer is that #X (É) < ∞ then one might ask for the precise<br />
number. Otherwise, ifX(É) is infinite then one may ask for the asymptotics<br />
of the <strong>rational</strong> <strong>points</strong> with respect to a certain height function H on X (É).<br />
(†) What is the asymptotics of the function N X,H given by<br />
for B → ∞?<br />
N X,H (B) := #{x ∈ X (É) | H(x) < B}<br />
1.3. –––– If X is a hypersurface of degree d in P n then there is a statistical<br />
heuristic for the asymptotics of N X,Hnaive .<br />
1.4. Statistical heuristic. –––– Let X be a hypersurface of degree d in P nÉ.<br />
Then,<br />
for some positive constant C.<br />
N X,Hnaive (B) ∼ C ·B n+1−d<br />
“Proof.” On P n , the total number ofÉ-<strong>rational</strong> <strong>points</strong> of height < B is ∼B n+1 .<br />
Indeed, these <strong>points</strong> may be given in the form<br />
(x 0 : . . . : x n )
212 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
for x i ∈such that x i ∈ [−B, B]. There are ∼ B n+1 such (n + 1)-tuples <strong>and</strong><br />
the probability that gcd(x 0 , . . . , x n ) = 1 is positive.<br />
Further, X is given by a homogeneous form F of degree d. Its range of values<br />
on [−B, B] n+1 is, a priori, [−DB d , DB d ] for a certain positive constant D.<br />
We assume that the values of F are equally distributed in that range. The value<br />
zero is then hit ∼B n+1−d times.<br />
□<br />
1.5. Remark. –––– This heuristic is definitely an interesting guideline about<br />
which behaviour to expect. It is, however, too naive in oder to work well in<br />
all cases.<br />
1.6. Example (Too few <strong>points</strong> – p-adic unsolvability). —–<br />
Consider the cubic fourfold X in P 5Égiven by the equation<br />
x 3 + 7y 3 + 49z 3 + 2u 3 + 14v 3 + 98w 3 = 0 .<br />
Then, X (É7) = ∅ which implies X (É) = ∅.<br />
Note that the statistical heuristic would predict cubic growth for the number<br />
ofÉ-<strong>rational</strong> <strong>points</strong> x on X such that H naive (x) < B.<br />
1.7. Example (Too few <strong>points</strong> – The Brauer-Manin obstruction). —–<br />
Consider the cubic surface X in P 3Égiven by the equation<br />
Here,<br />
w(x + w)(12x + w) =<br />
3<br />
∏<br />
i=1<br />
(<br />
x + θ (i) y + (θ (i) ) 2 z ) .<br />
θ := −38 + ∑<br />
i∈(∗ 19 ) 3 ζ i 19 ∈É(ζ 19 )<br />
<strong>and</strong> θ (i) are the three conjugates of θ inÉ.<br />
As shown in Example III.5.24, X (É) ≠ ∅ but X (É) = ∅. On X, there is a<br />
Brauer-Manin obstruction to the Hasse principle.<br />
In this case, the statistical heuristic would predict linear growth for the number<br />
ofÉ-<strong>rational</strong> <strong>points</strong> x on X such that H naive (x) < B.<br />
1.8. Examples (Too many <strong>points</strong> – Accumulating subvarieties). —–<br />
i) Consider the smooth threefold X ⊂ P 4Égiven by<br />
x 4 + y 4 = z 4 + v 4 + w 4 .<br />
Then, the statistical heuristic predicts linear growth. I.e., there should be ∼CB<br />
<strong>rational</strong> <strong>points</strong> of height
Sec. 1] A HEURISTIC 213<br />
ii) Other examples of interest are provided by the diagonal cubic threefolds<br />
V a,1 ∈ P 4Éof the form<br />
V a,1 : ax 3 = y 3 + z 3 + v 3 + w 3 .<br />
Here, quadratic growth is predicted by the statistical heuristic.<br />
However, V a,1 clearly contains the lines given by v = −w <strong>and</strong><br />
(x : y : z) = (x 0 : y 0 : z 0 )<br />
for (x 0 : y 0 : z 0 ) aÉ-<strong>rational</strong> point on the curve<br />
E a : ax 3 = y 3 + z 3 .<br />
This is a twisted Fermat cubic. One has the <strong>rational</strong> point (0 : 1 : (−1)) ∈ E a (É).<br />
Therefore, E a is an elliptic curve overÉ. The lines described form a cone over<br />
an elliptic curve. There are ∼B 2É-<strong>rational</strong> <strong>points</strong> of height < B alone on<br />
this cone.<br />
Examples of twisted Fermat cubics of rank up to 3 have been known to<br />
Ernst S. Selmer as early as 1951 [Sel]. For example, E 6 (É) is of rank one,<br />
generated by (21 : 17 : 37).<br />
1.9. Remark. –––– These examples grew out of our study of diagonal cubic <strong>and</strong><br />
quartic threefolds which is described in Chapter VII.<br />
1.10. Geometric interpretation. –––– The exponent n + 1 − d appearing in<br />
the statistical heuristic may be positive, zero, or negative. According to this<br />
distinction, there are three cases.<br />
This distinction into three cases coincides perfectly well with the Kodaira classification<br />
of projective varieties into Fano varieties, varieties of intermediate type,<br />
<strong>and</strong> varieties of general type. Indeed, the anticanonical divisor (−K ) on a<br />
degree d hypersurface in P n is exactly O(n + 1 − d).<br />
1.11. Remark. –––– The Kodaira classification is a purely geometric one.<br />
It does not make use of any arithmetic information on the projective variety<br />
considered. Only the complex algebraic variety produced by base change<br />
tois exploited.<br />
It is a very remarkable observation that whether a projective variety X is Fano,<br />
of intermediate type, or of general type seemingly has a lot of influence on the<br />
set ofÉ-<strong>rational</strong> <strong>points</strong> on X.
214 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
1.12. –––– More concretely, expectations are as follows.<br />
First case. X is a variety of general type.<br />
By definition, this means that the canonical invertible sheaf K is ample.<br />
In the case of a hypersurface, this corresponds to the case that n + 1 − d < 0.<br />
Here, the statistical heuristic is rather illogical. It states that for a large height<br />
bound we expect less <strong>points</strong> than for a small one. In the limit for B → ∞, we<br />
have B n+1−d → 0.<br />
One would therefore expect only very fewÉ-<strong>rational</strong> <strong>points</strong> on X. This is<br />
exactly the content of the conjecture of Lang.<br />
Second case. X is a variety of intermediate type.<br />
In the case of a hypersurface, this corresponds to the case that n + 1 − d = 0.<br />
Here, the statistical heuristic states that there should be a constant number of<br />
<strong>points</strong>, independently of the height bound B.<br />
One would therefore expect that there are a fewÉ-<strong>rational</strong> <strong>points</strong> on X. A more<br />
precise statement is given by the conjecture of Batyrev <strong>and</strong> Manin.<br />
Third case. X is a Fano variety.<br />
By definition, this means that the anticanonical invertible sheaf K ∨ is ample.<br />
In the case of a hypersurface, this corresponds to the case that n + 1 − d > 0.<br />
Here, one would expect that there are a lot ofÉ-<strong>rational</strong> <strong>points</strong> on X. A by far<br />
more precise formulation is given by the conjecture of Manin.<br />
2. The conjecture of Lang<br />
2.1. –––– The conjecture of Langdeals with the case of a variety of general type.<br />
There are actually several versions of it [Lan86].<br />
2.2. Conjecture (Lang). —– Let X be a smooth, projective variety of general type<br />
over a number field K. Then,<br />
i) (Weak Lang conjecture)<br />
the set of <strong>rational</strong> <strong>points</strong> X (K ) is not Zariski dense in X.<br />
ii) (Strong Lang conjecture)<br />
Thereis aZariski closed subset Z ⊂ X such that, for any finite field extension L ⊇ K,<br />
one has that X (L)\Z(L) is finite.<br />
2.3. Conjecture (Geometric Lang conjecture). —–<br />
LetX beasmooth, projective variety of generaltype over a field K of characteristic 0.<br />
Then, there is a proper Zariski closed subset Z(X ) ⊂ X, called the Langian exceptional<br />
set, which is the union of all positive dimensional subvarieties which are not<br />
of general type.
Sec. 2] THE CONJECTURE OF LANG 215<br />
2.4. Remark. –––– These conjectures are strongly interrelated. If the geometric<br />
Lang conjecture is true then the weak Lang conjecture implies the strong<br />
Lang conjecture.<br />
2.5. –––– Lang’s conjectures are known to be true when X is a subvariety of<br />
an abelian variety. This was proven by G. Faltings in [Fa91].<br />
Faltings’ 1991 result includes the case that X is a curve of general type. I.e., a<br />
curve of genus g ≥ 2. This particular case of Lang’s conjecture has been popular<br />
for decades as the Mordell conjecture. The Mordell conjecture was proven<br />
by G. Faltings in 1983 [Fa83].<br />
2.6. Examples. –––– If X is a curve then there is no Langian exceptional set Z.<br />
Weak <strong>and</strong> strong Lang conjecture are therefore equivalent.<br />
This is no longer the case for surfaces of general type.<br />
i) For example, the quintic surface given by<br />
x 5 + y 5 + z 5 + w 5 = 0<br />
in P 3Écontains the line “x = −y, z = −w” on which there are infinitely many<br />
É-<strong>rational</strong> <strong>points</strong>.<br />
ii) Another example of a surface of general type containing a projective line is<br />
provided by the Godeaux surface [Bv, Example X.3.4].<br />
iii) Let V 3<br />
a,b be the cubic threefold in P4Égiven by<br />
ax 3 = by 3 + z 3 + v 3 + w 3 .<br />
Then, the moduli space L a,b of the lines onV a,b is a surface of general type [Cl/G,<br />
Lemma 10.13].<br />
If the cubic curve<br />
E a,b : ax 3 + by 3 + z 3 = 0<br />
contains a <strong>rational</strong> point then, on V a,b , there are the lines given by v = −w<br />
<strong>and</strong> (x : y : z) = (x 0 : y 0 : z 0 ) for (x 0 : y 0 : z 0 ) ∈ E a,b (É). We call these lines the<br />
obvious lines on V a,b .<br />
Inother words, ifE a,b (É) ≠ ∅thenthesurfaceL a,b , whichisofgeneraltype, containsacopyoftheelliptic<br />
curveE a,b . ThereareinfinitelymanyÉ-<strong>rational</strong> <strong>points</strong><br />
on E a,b , for example for a = 6 <strong>and</strong> b = 1.<br />
2.7. An experiment. –––– We searched systematically forÉ-<strong>rational</strong> lines on<br />
the cubic threefolds V a,b for a, b = 1, . . . , 100, a ≥ b. Our method is<br />
described in detail in Chapter VII, Section 4. It guarantees that every line which<br />
contains a point of height
216 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
The results may be interpreted as providing numerical evidence for Lang’s conjecture.<br />
Indeed, aÉ-<strong>rational</strong> point on L a,b corresponds to aÉ-<strong>rational</strong> line<br />
on V a,b . Points on the elliptic curves lying on L a,b correspond to the obvious<br />
lines.<br />
In agreement with Lang’s conjecture, only very few non-obvious lines<br />
were found. On all the varieties V a,b for a, b = 1, . . . , 100, a ≥ b together,<br />
there are only 42 non-obvious lines containing a point of height 0, there exists a Zariski open subset X ◦ ⊆ X such that<br />
for B → ∞.<br />
N X ◦ ,H L<br />
(B) = #{x ∈ X ◦ (k) | H L (x) < B} ≪ B r+ε<br />
3.2. –––– Here, H L is the exponential of the height h L defined by L as<br />
introduced in Definition IV.2.10. It is determined only up to a factor which is<br />
bounded above <strong>and</strong> below by positive constants.<br />
The conjecture is consistent with these changes of the height function since B r+ε<br />
<strong>and</strong> (CB) r+ε differ by a constant factor.
Sec. 3] THE CONJECTURE OF BATYREV AND MANIN 217<br />
3.3. Remark. –––– This conjecture was first formulated by V. V. Batyrev <strong>and</strong><br />
Yu. I. Manin in [B/M]. An excellent presentation may be found in the survey<br />
article [Pe02] by E. Peyre.<br />
3.4. Fact (Varieties of general type). —– The conjecture of Batyrev <strong>and</strong> Manin<br />
implies the weak Lang conjecture.<br />
Proof. Indeed, if X is a variety of general type then the canonical invertible<br />
sheaf K itself is ample <strong>and</strong> we may work with L := K .<br />
Then, K + (−1)L = 0 is an effectiveÊ-divisor. Hence, for every ε > 0, the<br />
conjecture of Batyrev <strong>and</strong> Manin yields that<br />
#{x ∈ X ◦ (k) | H L (x) < B} ≪ B −1+ε<br />
for a suitable Zariski open subset X ◦ ⊂ X. In the limit for B → ∞, this shows<br />
that X ◦ (k) is empty.<br />
The set X (k) is therefore not Zariski dense in X.<br />
3.5. –––– Let X be a smooth Fano variety. Then, the anticanonical invertible<br />
sheaf K ∨ is ample <strong>and</strong> we may consider an anticanonical height which is defined<br />
by L := K ∨ .<br />
3.6. Fact (Fano varieties). —– For X a smooth Fano variety, the conjecture of<br />
Batyrev <strong>and</strong> Manin yields that<br />
N X ◦ ,H K ∨(B) = #{x ∈ X ◦ (k) | H K ∨(x) < B} ≪ B 1+ε<br />
□<br />
for a suitable Zariski open subset X ◦ ⊂ X.<br />
Proof. In this situation, K + (−K ) = 0 is an effectiveÊ-divisor.<br />
□<br />
3.7. Remarks. –––– i) In the case of a Fano hypersurface, this assertion fits<br />
perfectly well with the statistical heuristic.<br />
ii) However, for Fano varieties, the conjecture of Manin describes the growth<br />
of N X ◦ ,H K ∨<br />
much more precisely. The most interesting case of the conjecture<br />
of Batyrev <strong>and</strong> Manin is, therefore, that of a variety of intermediate type.<br />
ii. Varieties of intermediate type. —<br />
3.8. –––– Let X be a smooth, projective, minimal surface of Kodaira dimension<br />
0. Fix an ample invertible sheaf L ∈ Pic(X ).<br />
Then, the conjecture of Batyrev <strong>and</strong> Manin states that, for every ε > 0, there<br />
exists a Zariski open subset X ◦ ⊂ X such that<br />
N X ◦ ,H L<br />
(B) = #{x ∈ X ◦ (k) | H L (x) < B} ≪ B ε .
)⊗Ê<br />
218 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
Indeed, 12K is linearly equivalent to zero in any of the four cases of the Kodaira<br />
classification. Hence,<br />
[K ] ∈ NS(X<br />
is the class of an effectiveÊ-divisor on X.<br />
3.9. Example. –––– If X is an abelian variety then the conjecture of Batyrev<br />
<strong>and</strong> Manin is true.<br />
Indeed, if the rank of X (k) is equal to r then<br />
N X ◦ ,H L<br />
(B) ∼ C ·log r/2 B .<br />
3.10. Remark. –––– On the other h<strong>and</strong>, for K3 surfaces, the conjecture of<br />
Batyrev <strong>and</strong> Manin is open.<br />
For special K3 surfaces, most notably for Kummer surfaces such that the associated<br />
abelian surface is a product of two elliptic curves, a particular result has<br />
been obtained by D. McKinnon [McK]. If d is the minimal degree of a <strong>rational</strong><br />
curve on X then McKinnon shows N X ◦ ,H L<br />
(B) ≪ B 2/d for X ◦ the complement<br />
of the union of all <strong>rational</strong> curves on X of degree d.<br />
3.11. Observation. –––– There are examples of K3 surfaces over a number<br />
field k which contain infinitely many <strong>rational</strong> curves defined over k [Bill, Sec. 3].<br />
Let X be such a K3 surface <strong>and</strong> let C 1 , C 2 , . . . be <strong>rational</strong> curves on X.<br />
Denote their degrees by d 1 , d 2 , . . . . Assume that the curves are listed in such a<br />
way that the degrees are in ascending order.<br />
Then, on<br />
X ◦<br />
l := X \C 1 \ . . . \C l−1 ,<br />
there are still ≥ c l B 2/d l<br />
k-<strong>rational</strong> <strong>points</strong> to be expected. Indeed, X ◦<br />
l contains a<br />
non-empty, open subset of C l .<br />
In other words, there is no way to choose a uniform Zariski open subset X ◦ ⊂ X<br />
such that<br />
N X ◦ ,H L<br />
(B) = #{x ∈ X ◦ (k) | H L (x) < B} ≪ B ε<br />
for every ε > 0. In order to fulfill the conjecture of Batyrev <strong>and</strong> Manin, one<br />
therefore has to choose X ◦ in dependence of ε.<br />
3.12. Example. –––– Consider the diagonal quartic surface X given in P 3Éby<br />
the equation<br />
x 4 + 2y 4 = z 4 + 4w 4 .
Sec. 3] THE CONJECTURE OF BATYREV AND MANIN 219<br />
On this K3 surface, there are theÉ-<strong>rational</strong> <strong>points</strong> (1 : 0 : ±1 : 0) <strong>and</strong><br />
(±1 484 801 : ±1 203 120 : ±1 169 407 : ±1 157 520) .<br />
These are the onlyÉ-<strong>rational</strong> <strong>points</strong> onX known <strong>and</strong> the onlyÉ-<strong>rational</strong> <strong>points</strong><br />
of (naive) height less than 10 8 [EJ2, EJ3].<br />
A systematic search forÉ-<strong>rational</strong> <strong>points</strong> on the K3 surface X is described in<br />
Chapter XI.<br />
3.13. Remark. –––– In general, not much is known about the arithmetic<br />
of K3 surfaces. Nevertheless, in 1981, F. Bogomolov formulated a very optimistic<br />
conjecture.<br />
3.14. Conjecture (F. Bogomolov, cf. [Bo/T]). —– Let X be a K3 surface over a<br />
number field k. Then, every k-<strong>rational</strong> point on X lies on <strong>rational</strong> curve C ⊂ X<br />
(defined over the algebraic closure k).<br />
iii. A somewhat different example. —<br />
3.15. Example. –––– Let X be the blow-up of P 2 in nine <strong>points</strong> P 1 , . . . , P 9 in<br />
general position.<br />
Denote by E := E 1 + . . . + E 9 the sum of the corresponding nine exceptional<br />
lines. According to the Nakai-Moizhezon criterion, a divisor aL − bE is<br />
ample if <strong>and</strong> only if a > 3b > 0. Further, we have K = −3L + E.<br />
The conjecture of Batyrev <strong>and</strong> Manin therefore asserts that, for every ε > 0,<br />
there exists a Zariski open subset X ◦ ⊂ X such that<br />
N X ◦ ,H O(aL−bE)<br />
(B) ≪ B 3/a+ε .<br />
On the other h<strong>and</strong>, assuming aL − bE to be very ample for simplicity, the<br />
embedding ι: (X<br />
(<br />
։)P<br />
) 2 ֒→ P N is given by homogeneous forms of degree a.<br />
Hence, H O(aL−bE) ι(x) ≤ c · H<br />
a<br />
naive(x) for some constant c which shows the<br />
lower bound<br />
N X ◦ ,H O(aL−bE)<br />
(B) = Ω(B 3/a )<br />
for any Zariski open subset X ◦ ⊂ X.<br />
3.16. –––– On X, there are infinitely many exceptional curves D such that<br />
D 2 = −1 <strong>and</strong> DK = −1. For every d ∈Æ, there are finitely many of the<br />
type dL − a 1 E 1 − . . . − a 9 E 9 .
220 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
Relative to aL − bE, their degrees are<br />
(dL − a 1 E 1 − . . . − a 9 E 9 )(aL − bE) = da − b(a 1 + . . . + a 9 )<br />
= da − b(3d − 1)<br />
= d(a − 3b) + b .<br />
Since a − 3b > 0, this expression tends to infinity for d → ∞. Only finitely<br />
many of the exceptional curves are of degree < 2 a which is equivalent to ≫B3/a<br />
3<br />
<strong>rational</strong> <strong>points</strong> of height
Sec. 4] THE CONJECTURE OF MANIN 221<br />
4.2. Lemma. –––– Let k ⊆be an algebraically closed field of characteristic 0<br />
<strong>and</strong> X be a smooth, projective variety over k. Assume that X is Fano.<br />
Then, for every prime number l,<br />
i) H 1 ét (X,l) = 0.<br />
ii) The first Chern class induces an isomorphism<br />
∼=<br />
Pic(X −→ H )⊗l<br />
2 ét (X,l(1)) .<br />
Proof. Let first i ∈Æbe arbitrary. By [SGA4, Exp. XVI, Corollaire<br />
1.6], we have, for every k ∈Æ, H i ét (X, µ l k) ∼ = H i ét (X, µ l k). Further,<br />
the comparison theorem [SGA4, Exp. XI, Théorème 4.4] shows<br />
that H i ét (X, µ l k) ∼ = H i (X (),/l k).<br />
On the other h<strong>and</strong>, by the universal coefficient theorem for cohomology [Sp,<br />
Chap. 5, Sec. 5, Theorem 10],<br />
H i (X (),/l k) ∼ = H i (X (),)⊗/l k⊕Tor 1 (H i+1 k<br />
(X (),),/l k)<br />
∼= H i (X (),)⊗/l k⊕H i+1 (X (),) l k .<br />
Here, the transition maps H i+1 (X (),) l k+1 → H i+1 (X (),) l k are given by<br />
multiplication by l. A finite composition of them is the zero map. This implies<br />
H i ét (X,l(1)) ∼ = lim ←−<br />
H i (X (),)⊗/l<br />
= H i (X . (‡)<br />
(),)⊗l<br />
i) Since H 1 (X (), O X ()) = 0, the long exact cohomology sequence associated<br />
to the exponential sequence yields that H 1 (X (),) = 0. Formula (‡) implies<br />
the claim.<br />
ii) Here, we use both, H 1 (X (), O X ()) = 0 <strong>and</strong> H 2 (X (), O X ()) = 0.<br />
The long exact cohomology sequence associated to the exponential sequence<br />
then shows that<br />
c 1 : Pic(X ) −→ H 2 (X (),)<br />
is an isomorphism. Tensoring yields isomorphisms<br />
Pic(X )⊗/l k.<br />
<strong>and</strong> going over to the inverse limit implies the assertion.<br />
k−→ H 2 (X (),)⊗/l<br />
□
222 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
4.3. Remarks. –––– i) The first Chern class in étale cohomology is defined using<br />
the Kummer sequence. Recall that there is the commutative diagram<br />
0<br />
2πi O X<br />
exp O ∗ X<br />
0<br />
exp( 2πi<br />
n ·)<br />
0 µ n O ∗ X<br />
exp( 1 n ·)<br />
(·) n O ∗ X<br />
showing that this agrees with the definition based on the exponential sequence.<br />
ii) The tensor product does in general not commute with inverse limits. However,<br />
for a finitely generated-module A,<br />
k∼ = A⊗l<br />
lim<br />
←− A⊗/l<br />
is the l-adic completion of A [Mat, Theorem 8.7].<br />
=<br />
0<br />
ii. Fixing a particular anticanonical height. —<br />
4.4. Remark. –––– The anticanonical height H K ∨ on X is determined only up<br />
to a certain factor which is bounded above <strong>and</strong> below by positive constants.<br />
To be able to make any statement on the value of the constant τ, we have to fix<br />
a particular height function.<br />
4.5. –––– For this, according to Definition IV.7.38, it is necessary to choose an<br />
adelic metric ‖ . ‖ = {‖ . ‖ ν } ν∈Val(É) on K ∨ such that (K ∨ , ‖ . ‖) ∈ ˜Pic(X ).<br />
For the remainder of this section, we fix such an adelic metric once <strong>and</strong> for all.<br />
4.6. Definition. –––– Put<br />
H K ∨(x) := exph (K ∨ ,‖.‖)(x)<br />
for h (K ∨ ,‖ .‖) theabsoluteheightwithrespecttotheintegrablymetrizedinvertible<br />
sheaf (K ∨ , ‖ . ‖) in the sense of Definition IV.7.38.<br />
We will call this height function the anticanonical height defined by the adelic<br />
metric given in 4.5.<br />
4.7. –––– Most height functions occurring in practice are a lot simpler than<br />
the general theory. For this reason, it is probably wise to recall an elementary<br />
particular case.<br />
Choose<br />
i) a projective embedding ι: X → P NÉsuch that K ∨ ∼ = ι ∗ O(d) for some d ∈Æ,<br />
<strong>and</strong><br />
ii) a continuous hermitian metric ‖ . ‖ ∞ on K.<br />
∨
Sec. 4] THE CONJECTURE OF MANIN 223<br />
Then, the topological closure of ι(X ) in P Nis an arithmetic variety X which<br />
is a model of X over Spec.<br />
Put<br />
H K ∨(x) := exp h (O(d)|X ,‖ .‖ ∞ )(x)<br />
for h (O(d)|X ,‖.‖ ∞ ) the height function with respect to the hermitian line bundle<br />
(O(d)| X , ‖ . ‖ ∞ ) ∈ ̂Pic C 0 (X ) in the sense of Definition IV.3.12.<br />
4.8. –––– This height function corresponds to the adelic metric on K ∨ which<br />
is given by the following construction. (Cf. Example IV.7.6.)<br />
i) ‖ . ‖ ∞ is part of the data given.<br />
ii) For a prime number p, the metric ‖ . ‖ p is given as follows.<br />
Let x ∈Ép. There is a unique extension x : SpecOÉp → X of x. Then, x∗ O(d)<br />
is a projective O K -module of rank 1. Each l ∈ O(d)(x) induces a <strong>rational</strong> section<br />
of x ∗ O(d). Put<br />
‖l‖(x) := inf {| a| | a ∈ K, l ∈ a · x ∗ O(d)}.<br />
4.9. Remarks. –––– i) To define the metric ‖ . ‖ p for a particular prime p, a<br />
model L of X over Spec[ 1 ] for p ∤m is sufficient.<br />
m<br />
ii) For every adelic metric ‖ . ‖ = {‖ . ‖ ν } ν∈Val(É) on K ∨ , there exist a model X<br />
of X over Spec[ 1 ] for a certain m ∈Æ<strong>and</strong> an extension of K ∨ to X such<br />
m<br />
that ‖ . ‖ νp is induced by that model for every p ∤m.<br />
iii. The conjecture. —<br />
4.10. –––– The conjecture of Manin deals with the anticanonical height H K ∨<br />
on a Fano variety.<br />
4.11. Conjecture (Manin). —– Let X be a smooth, projective variety overÉ.<br />
Assume that X is Fano.<br />
Then, there exist a positive integer r, a real number τ, <strong>and</strong> a Zariski open subset<br />
X ◦ ⊆ X such that<br />
for B → ∞.<br />
N X ◦ ,H K ∨(B) = #{x ∈ X ◦ (k) | H K ∨(x) < B} ∼ τB log r B<br />
4.12. Remark. –––– The factor log r B is new in comparison with the statistical<br />
heuristic given in 1.4. It is known for a long time that such a factor is, in general,<br />
necessary. In fact, J. Franke, Yu. I. Manin, <strong>and</strong> Y. Tschinkel [F/M/T] showed<br />
in 1989 that Manin’s conjecture becomes compatible with direct products of<br />
Fano varieties only when a suitable log r B-factor is added.
224 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
4.13. Remark. –––– At least in the case that X is a surface, it is expected<br />
that r = rkPic(X ) − 1. There are, however, counterexamples to this formula<br />
in dimension three [Ba/T].<br />
On the other h<strong>and</strong>, in Chapter VII, we will present numerical evidence<br />
for Manin’s conjecture for diagonal cubic <strong>and</strong> quartic threefolds. For those,<br />
rkPic(X ) = 1 <strong>and</strong> our experiments indicate that r = 0.<br />
4.14. –––– In [Pe95a], E. Peyre refined Manin’s conjecture by providing an<br />
explicit description for the value of the coefficient τ. Peyre’s constant is a<br />
product of several factors which we will subsequently explain.<br />
5. Peyre’s constant<br />
i. The factor α. —<br />
5.1. Definition (Cf. [Pe95a, Définition 2.4]). —–<br />
Let X be a projective algebraic variety overÉ. Choose an isomorphism<br />
ι: Pic(X )/ Pic(X ) tors<br />
∼ =<br />
−→t .<br />
Identify Pic(X )⊗ÊwithÊt according to ι.<br />
Further, let Λ eff (X ) ⊂ Pic(X )⊗Ê=Êt be the cone generated by the effective<br />
divisors. Consider the dual cone Λ ∨ eff (X ) ⊂ (Êt ) ∨ . Then,<br />
α(X ) := t · vol {x ∈ Λ ∨ eff<br />
| 〈x, −K 〉 ≤ 1 } .<br />
Here, 〈 . , . 〉 denotes the tautological pairing 〈 . , . 〉: (Êt ) ∨ ×Êt →Ê. vol is the<br />
ordinary Lebesgue measure on (Êt ) ∨ .<br />
5.2. Remark. –––– In [Pe/T, Definition 2.5], α(X ) is defined by an integral.<br />
An elementary calculation shows that the two definitions are equivalent.<br />
5.3. Example. –––– Suppose Pic(X ) =. Denote by [L] ∈ Pic(X ) the ample<br />
generator. Let δ ∈Æbe such that [−K ] = δ[L]. Then, α(X ) = 1/δ.<br />
In particular, one has α(X ) = 1 for every smooth cubic surface such that<br />
rkPic(X ) = 1. Indeed, (−K ) 2 = 3 is square-free. Therefore, [−K ] ∈ Pic(X ) is<br />
not divisible.
Sec. 5] PEYRE’S CONSTANT 225<br />
5.4. Example. –––– α(P 1 × P 1 ) = 1/4.<br />
Indeed, Pic(P 1 ×P 1 ) =L 1 ⊕L 2 . The effective cone is generated by L 1 <strong>and</strong> L 2 .<br />
In the dual space,<br />
Λ ∨ eff (P1 × P 1 ) = {al 1 + bl 2 | a, b ≥ 0} .<br />
Further, −K = 2L 1 + 2L 2 . The condition 〈x, −K 〉 ≤ 1 is therefore equivalent<br />
to 2a + 2b ≤ 1.<br />
The area of the triangle with vertices (0, 0), (1/2, 0), <strong>and</strong> (0, 1/2) is equal to 1/8.<br />
5.5. Example. –––– Let X be the blow-up of P 2 in aÉ-<strong>rational</strong> point.<br />
Then, α(X ) = 1/6.<br />
Here, Pic(X ) =L ⊕E. The effective cone is generated by E <strong>and</strong> L − E.<br />
In the dual space,<br />
Λ ∨ eff (X ) = {al + be | b ≥ 0, a − b ≥ 0} .<br />
Further, −K = 3L−E. The condition 〈x, −K 〉 ≤ 1 is equivalent to 3a−b ≤ 1.<br />
The area of the triangle with vertices (0, 0), (1/3, 0), <strong>and</strong> (1/2, 1/2) is equal<br />
to 1/12.<br />
5.6. Example. –––– Let X be the blow-up of P 2 in sixÉ-<strong>rational</strong> <strong>points</strong> which<br />
form an orbit under Gal(É/É). Then, α(X ) = 4/3.<br />
Again, Pic(X ) =L⊕E. Here, the effective cone is generated by E, L−1/3E,<br />
<strong>and</strong> 2L − 5/6E. I.e., by E <strong>and</strong> L − 5/12E.<br />
In the dual space,<br />
Λ ∨ eff (X ) = {al + be | b ≥ 0, a − 5/12b ≥ 0} .<br />
Further, −K = 3L − E. The condition 〈x, −K 〉 ≤ 1 is therefore equivalent<br />
to 3a − b ≤ 1.<br />
The area of the triangle with vertices (0, 0), (1/3, 0), <strong>and</strong> (5/3, 4) is equal to 2/3.<br />
5.7. Example. –––– Let X be a smooth cubic surface overÉ. Assume the<br />
orbit lengths of the 27 lines under the Gal(É/É)-operation are [1, 10, 16].<br />
Then, α(X ) = 1.<br />
Note that this is the generic case of a cubic surface containing aÉ-<strong>rational</strong> line.<br />
The computation using GAP discussed in II.8.23 shows that rk Pic(X ) = 2.<br />
Compare the list given in the appendix. We claim that Pic(X ) =K ⊕E<br />
for E theÉ-<strong>rational</strong> line.
∈<br />
226 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
Indeed, K <strong>and</strong> E are linearly independent since K 2 = 3 <strong>and</strong> E 2 = −1. Further,<br />
if aK + bE ∈ Pic(X ) then intersecting with K shows that 3a − b<br />
while intersecting with a line skew to E shows −a ∈. Altogether, a, b ∈.<br />
Write D 1 for the sum of the ten lines meeting E <strong>and</strong> D 2 for the sum of the sixteen<br />
lines skew to E. As D 1 K = −10 <strong>and</strong> D 1 E = 10, we have D 1 = −5K − 5E.<br />
Similarly, D 2 K = −16 <strong>and</strong> D 2 E = 0 imply D 2 = −4K + 4E.<br />
The effective cone is generated by E, D 1 , <strong>and</strong> D 2 . The calculations show that E<br />
<strong>and</strong> −K − E form a simpler system of generators.<br />
In the dual space,<br />
Λ ∨ eff (X ) = {ak + be | b ≥ 0, −a − b ≥ 0} .<br />
Further, the condition 〈x, −K 〉 ≤ 1 is equivalent to −a ≤ 1.<br />
The area of the triangle with vertices (0, 0), (−1, 0), <strong>and</strong> (−1, 1) is equal to 1/2.<br />
5.8. Example. –––– Let X be a smooth cubic surface overÉ. Assume the<br />
orbit lengths of the 27 lines under the Gal(É/É)-operation are [2, 5, 10, 10].<br />
Then, α(X ) = 3/2.<br />
Note that the two lines conjugate to each other are skew. Indeed, if they<br />
were contained in a plane then the third line contained in that plane would be<br />
É-<strong>rational</strong>. This example describes the generic case of a cubic surface containing<br />
two skew lines which are conjugate to each other over a quadratic number field.<br />
Again, the computation usingGAP discussed in II.8.23 shows that rkPic(X ) = 2.<br />
We claim that Pic(X ) =K ⊕E for E := E 1 + E 2 the sum of the two lines<br />
conjugate to each other.<br />
Indeed, the discriminant of the lattice spanned by K <strong>and</strong> E is<br />
−2 −2<br />
∣ −2 3∣ = −10 .<br />
As this number is non-zero, we see that K <strong>and</strong> E are linearly independent.<br />
Since (−10) is a square-free integer, the lattice may not be refined.<br />
Write D 1 for the sum of the five lines meeting E 1 <strong>and</strong> E 2 , D 2 for the sum of<br />
the ten other lines meeting E 1 or E 2 , <strong>and</strong> D 3 for the sum of the ten lines not<br />
meeting E at all. Then, D 1 K = −5 <strong>and</strong> D 1 E = 10 imply that D 1 = −3K −2E.<br />
Similarly, D 2 K = −10 <strong>and</strong> D 2 E = 10 show that D 2 = −4K − E. Finally,<br />
D 3 K = −10, D 3 E = 0, <strong>and</strong> D 3 = −2K + 2E.<br />
The effective cone is generated by E, D 1 , D 2 , <strong>and</strong> D 3 . The calculations yield E<br />
<strong>and</strong> −3K − 2E as a simpler system of generators.<br />
In the dual space,<br />
Λ ∨ eff (X ) = {ak + be | b ≥ 0, −3a − 2b ≥ 0} .
Sec. 5] PEYRE’S CONSTANT 227<br />
Further, the condition 〈x, −K 〉 ≤ 1 is equivalent to −a ≤ 1.<br />
The area of the triangle with vertices (0, 0), (−1, 0), <strong>and</strong> (−1, 3/2) is 3/4.<br />
5.9. Remarks. –––– i) Let X be a smooth cubic surface overÉcontaining a<br />
É-<strong>rational</strong> line. Assume that rk Pic(X ) = 2. Then, α(X ) = 1.<br />
Indeed, the arguments given in Example 5.7 still show that Pic(X ) =K ⊕E.<br />
On the other h<strong>and</strong>, the Gal(É/É)-orbits of the 27 lines might break into<br />
smaller pieces. For example, assume there is an orbit consisting of k of the<br />
ten lines meeting E. This causes another generator D 1 ′ of the effective cone.<br />
However, we have that D 1 ′ K = −k <strong>and</strong> D 1 ′ E = k. Therefore,<br />
is simply a scalar multiple of D 1 .<br />
D ′ 1 = − k 2 K − k 2 E = k 10 D 1<br />
Analogously, an orbit consisting of some of the sixteen lines skew to E leads to<br />
a scalar multiple of D 2 . This ensures that Λ ∨ eff<br />
(X ) is the same as in the generic<br />
case described in Example 5.7.<br />
ii) Similarly, let X be any smooth cubic surface overÉcontaining two skew<br />
lines conjugate over a quadratic number field. Assume that rk Pic(X ) = 2.<br />
Then, α(X ) = 3/2.<br />
5.10. Remark. –––– For smooth cubic surfaces in general, the value of α depends<br />
only on the orbit structure of the 27 lines under the Gal(É/É)-operation.<br />
In particular, there are only 350 cases corresponding to the conjugacy classes of<br />
sub<strong>groups</strong> of W (E 6 ). The examples given above actually cover the lion’s share<br />
of these cases.<br />
In fact, 137 of the 350 conjugacy classes of sub<strong>groups</strong> of W (E 6 ) lead to Picard<br />
rank one. Then, α(X ) = 1 as shown in Example 5.3.<br />
Further, 133 conjugacy classes yield Picard rank two. The argument of Remark<br />
5.9.i) alone covers 98 of them. Eight further conjugacy classes are treated<br />
by Remark 5.9.ii).<br />
This might indicate that α(X ) may be effectively computed for each of the<br />
350 conjugacy classes of sub<strong>groups</strong> of W (E 6 ) which could appear as the Galois<br />
<strong>groups</strong> acting on the 27 lines. This has, however, not yet been done.<br />
In the cases of high Picard rank, the computation of the volume of the resulting<br />
polytope should be the main challenge. For example, László Lovász [Lo]<br />
explains that good algorithms for the exact computation of such volumes are impossible.<br />
His suggestion is to use a Monte-Carlo method instead.<br />
The case of maximal rank has, however, been settled.
228 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
5.11. Example. –––– Let X be a smooth cubic surface over the <strong>rational</strong>s such<br />
that rk Pic(X ) = 7. I.e., such that each of the 27 lines on X is defined overÉ.<br />
Then, α(X ) = 1<br />
120 .<br />
This has been proven in the Ph.D. thesis of U. Derenthal [Der, Theorem 8.3].<br />
ii. The factor β. —<br />
5.12. Definition. –––– Let X be a projective variety overÉ. Then, β(X ) is<br />
defined as<br />
β(X ) := #H 1( Gal(É/É), Pic(XÉ) ) .<br />
5.13. Remarks. –––– We studied H 1( Gal(É/É), Pic(XÉ) ) in Chapter II. Recall<br />
the following facts.<br />
i) If Pic(XÉ) =then β(X ) = 1.<br />
ii) In the particular case that X is a smooth cubic surface, β(X ) depends only<br />
on the conjugacy class of the subgroup G ⊆ W (E 6 ) defined by the operation<br />
of Gal(É/É) on the 27 lines. Actually, β(X ) depends only on the decomposition<br />
of the 27 lines into Galois orbits.<br />
We computed β(X ) for each of the 350 conjugacy classes of sub<strong>groups</strong> of W (E 6 ).<br />
The possible values are β(X ) = 1, 2, 3, 4, <strong>and</strong> 9. More details are given in II.8.23<br />
<strong>and</strong> the list presented in the appendix.<br />
iii. The value of L at 1. —<br />
5.14. Lemma. –––– Let X be a projective algebraic variety overÉ. Denote by<br />
L( . , χ Pic(XÉ)) the Artin L-function of the Gal(É/É)-representation Pic(XÉ)⊗.<br />
Then, for t the Picard rank of X,<br />
is a real number different from zero.<br />
lim<br />
s→1<br />
(s − 1) t L(s, χ Pic(XÉ))<br />
Proof. The Gal(É/É)-representation Pic(XÉ)⊗contains the trivial representation<br />
t times as a direct summ<strong>and</strong>. Therefore, L(s, χ Pic(XÉ)) = ζ(s) t ·L(s, χ P )<br />
<strong>and</strong><br />
lim<br />
s→1<br />
(s − 1) t L(s, χ Pic(XÉ)) = L(1, χ P ) .<br />
Here, ζ denotes the Riemann zeta function <strong>and</strong> P is a Gal(É/É)-representation<br />
which does not contain trivial components. [Mu-y, Corollary 11.5 <strong>and</strong> Corollary<br />
11.4] show that L( . , χ P ) has neither a pole nor a zero at 1. □
Sec. 5] PEYRE’S CONSTANT 229<br />
iv. The p-adic <strong>measures</strong>. —<br />
5.15. Definition. –––– Let p be a prime number <strong>and</strong> x ∈ X (Ép) be arbitrary.<br />
Choose local coordinates t 1 , . . . , t n on X × SpecÉSpecÉp in a neighbourhood<br />
of x. These define a morphism ofÉp-schemes<br />
ι x : U x −→ A nÉp<br />
from a Zariski open neighbourhood U x of x such that ι(x) = (0, . . . , 0) <strong>and</strong> ι is<br />
étale at x.<br />
The tensor field (ι −1 ) [ ∂<br />
]<br />
x,Ép ∗ ∂t 1<br />
∧ . . . ∧ ∂<br />
∂t n<br />
is the restriction to the p-adic <strong>points</strong><br />
of a section of the anticanonical sheaf K ∨Ép .<br />
In a neighbourhood of x, we define the measure ω p by<br />
[<br />
∥ ∂<br />
∥(ι −1 ) x,Ép ∗ ∧ . . . ∧ ∂ ]∥ ∥∥ · ι<br />
∗<br />
∂t 1 ∂t (dt x,Ép 1 · . . . · dt n ) .<br />
n<br />
Here, each dt i denotes a copy of the Haar measure onÉp normalized in the<br />
usual manner.<br />
5.16. Remark. –––– This definition is independent of the choice of the coordinates<br />
t 1 , . . . , t n . Indeed, for t ′ 1, . . . , t ′ n forming another system of local<br />
coordinates near x, we have<br />
dt ′ 1 · . . . · dt ′ n = det ∂(t′ 1 , . . . , t′ n )<br />
∂(t 1 , . . . , t n ) · dt 1 . . . dt n<br />
<strong>and</strong><br />
∂<br />
∂t ′ 1<br />
∧ . . . ∧ ∂<br />
∂t ′ n<br />
= det ∂(t 1, . . . , t n )<br />
∂(t<br />
1 ′, . . . , t′ n ) · ∂<br />
∧ . . . ∧ ∂ .<br />
∂t 1 ∂t n<br />
5.17. Lemma. –––– For a primenumber p, suppose that the metric ‖ . ‖ p is induced<br />
by a model X of X. Then, the measure ω p on X (Ép) is given as follows.<br />
Let a ∈ X (/p k) <strong>and</strong> put<br />
Then,<br />
U (k)<br />
a := {x ∈ X (Ép) | x ≡ a (mod p k ) } .<br />
ω p (U a (k) ) = lim #{ y ∈ X (/p n) | y ≡ a (mod p k ) }<br />
.<br />
n→∞ p n dim X<br />
Proof. This is simply a more concrete reformulation of the abstract definition.<br />
□<br />
5.18. Corollary. –––– Let p be a prime such that X is smooth over p. Then, for<br />
every a ∈ X (/p),<br />
ω p (U a (1) ) = 1<br />
p . dim X
230 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
Proof. This is a consequence of Hensel’s lemma.<br />
□<br />
5.19. Remark. –––– ω p may as well be described in terms of the affine cone<br />
over X. For example, if X is a hypersurface of degree d in P n then one has<br />
(<br />
ω ) p X (Ép) = 1 − p−k<br />
· lim<br />
1 − p #C (dim X+1)n<br />
−1 X (/p<br />
n)/p<br />
n→∞<br />
for k := n + 1 − d. A proof is given in [Pe/T, Corollary 3.5].<br />
5.20. Definition (The measure τ p on X (Ép)). —– Let p be a prime number.<br />
We define the measure τ p on X (Ép) by<br />
τ p := det ( 1 − p −1 Frob p | Pic(XÉ) I p) · ωp .<br />
Here, Pic(XÉ) I p<br />
denotes the fixed module under the inertia group.<br />
5.21. Remark. –––– In the easiest case that Pic(XÉ) =, one has<br />
det ( )<br />
1 − p −1 Frob p | Pic(XÉ) I 1<br />
p = 1 −<br />
p .<br />
v. The real measure. —<br />
5.22. Definition (The measure τ ∞ nÊ<br />
on X (Ê)). —– Let x ∈ X (Ê) be arbitrary.<br />
Choose local coordinates t 1 , . . . , t n on X × SpecÉSpecÊin a neighbourhood<br />
of x. These define a morphism ofÊ-schemes<br />
ι x : U x −→ A<br />
from a Zariski open neighbourhood U x of x such that ι(x) = (0, . . . , 0) <strong>and</strong> ι is<br />
étale at x. Consequently, ι x,Ê: U x (Ê) →Ên is a diffeomorphism near x.<br />
[<br />
The tensor field (ιx,Ê) −1 ∂<br />
]<br />
∗ ∂t 1<br />
∧ . . . ∧ ∂<br />
∂t n<br />
is the restriction to the real <strong>points</strong> of a<br />
section of the anticanonical sheaf K.<br />
∨<br />
In a neighbourhood of x, we define the measure τ ∞ by the differential form<br />
[<br />
∥ ∂<br />
∥(ι −1<br />
x,Ê) ∗ ∧ . . . ∧ ∂ ]∥ ∥∥ · ι<br />
∗<br />
∂t 1 ∂t x,Ê(dt 1 ∧ . . . ∧ dt n ) .<br />
n<br />
5.23. Remark. –––– Again, this definition is independent of the choice of the<br />
coordinates t 1 , . . . , t n . If t ′ 1, . . . , t ′ n form another system of local coordinates<br />
near x then<br />
dt ′ 1 ∧ . . . ∧ dt ′ n = det ∂(t′ 1 , . . . , t′ n )<br />
∂(t 1 , . . . , t n ) · dt 1 ∧ . . . ∧ dt n<br />
<strong>and</strong><br />
∂<br />
∂t ′ 1<br />
∧ . . . ∧ ∂<br />
∂t ′ n<br />
= det ∂(t 1, . . . , t n )<br />
∂(t<br />
1 ′, . . . , t′ n) · ∂<br />
∧ . . . ∧ ∂ .<br />
∂t 1 ∂t n
Sec. 5] PEYRE’S CONSTANT 231<br />
5.24. –––– In the situation that X is a hypersurface, the measure τ ∞ may be<br />
described by far more concretely in terms of the Leray measure.<br />
5.25. Definition. –––– Let f ∈Ê[X 0 , . . . , X n ] be a homogeneous polynomial<br />
such that (grad f )(x 0 , . . . , x n ) ≠ 0 for every (x 0 , . . . , x n ) different from<br />
the origin.<br />
Then, the Leray measure on “ f = 0” is given by the formula<br />
1<br />
ω Leray =<br />
‖ grad f ‖ ω hyp .<br />
Here, ω hyp denotes the usual hypersurface measure.<br />
5.26. Lemma. –––– Let f ∈Ê[X 0 , . . . , X n ] be as in Definition 5.25 <strong>and</strong><br />
U ⊂Ên+1 be an open subset such that ∂ f<br />
∂x 0<br />
does not vanish on U.<br />
Then, on U ∩“ f = 0”, the Leray measure is given up to sign by the differential form<br />
1<br />
| ∂f/∂x 0 | dx 1 ∧ . . . ∧ dx n .<br />
Proof. This is an immediate consequence of Sublemma 5.27 below.<br />
5.27. Sublemma. –––– Let U ⊂Ên+1 be an open subset <strong>and</strong> f : U →Êbe a<br />
smooth function such that ∂ f<br />
∂x 0<br />
does not vanish.<br />
Then,thehypersurfacemeasureon“f = 0”isgivenuptosignbythedifferentialform<br />
‖ grad f ‖<br />
| ∂f/∂x 0 | dx 1 ∧ . . . ∧ dx n .<br />
Proof. It is well known that, having resolved the equation<br />
f (x 0 , . . . , x n ) = 0<br />
by x 0 , the hypersurface measure is given by the form<br />
√<br />
1 + (∂x 0 /∂x 1 ) 2 + . . . + (∂x 0 /∂x n ) 2 dx 1 ∧ . . . ∧ dx n .<br />
For every i = 1, . . . , n, the equation f ( )<br />
x 0 (x 1 , . . . , x n ), x 1 , . . . , x n = 0<br />
immediately yields ∂f ∂x 0<br />
∂x 0 ∂x i<br />
+ ∂f<br />
∂x i<br />
= 0 . In other words,<br />
∂x 0<br />
= − ∂f / ∂f<br />
.<br />
∂x i ∂x i ∂x 0<br />
Altogether, we find the differential form<br />
√<br />
1 +<br />
n<br />
∑<br />
i=1<br />
( ∂f<br />
∂x i<br />
/ ∂f<br />
∂x 0<br />
) 2<br />
dx 1 ∧ . . . ∧ dx n<br />
for the hypersurface measure. This is exactly the assertion.<br />
□<br />
□
232 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
5.28. Notation. –––– Let X ⊆ P nÉbe a projective variety. We denote by CX<br />
the affine cone over X.<br />
A metric ‖ . ‖ on O(1)| X () defines a compact subset<br />
N ‖.‖ := {x = (x 0 , . . . , x n ) ∈ CX (Ê) | | x 0 | ≤ ‖X 0 ‖(x), . . . , | x n | ≤ ‖X n ‖(x)}<br />
of the affine cone which is symmetric to the origin. Note that the conditions<br />
| x i | ≤ ‖X i ‖(x) <strong>and</strong> | x j | ≤ ‖X j ‖(x) are equivalent to each other as long<br />
as x i , x j ≠ 0.<br />
5.29. Examples. –––– i) Consider on O(1)| X () the minimum metric ‖ . ‖ min<br />
from Example IV.3.5. Then,<br />
N ‖ .‖min<br />
= {x = (x 0 , . . . , x n ) ∈ CX (Ê) | | x 0 | ≤ 1, . . . , | x n | ≤ 1}<br />
is a hypercube.<br />
ii) Similarly, for the l 2 -metric,<br />
is the unit ball.<br />
N ‖.‖l 2 = {x = (x 0, . . . , x n ) ∈ CX (Ê) | x 2 0 + . . . + x 2 n ≤ 1}<br />
5.30. Proposition. –––– Let F ∈É[X 0 , . . . , X n ] be a homogeneous polynomial<br />
of degree d <strong>and</strong> X ⊂ P nÉbe the hypersurface defined by the equation F = 0.<br />
According to the adjunction formula, there is a canonical isomorphism<br />
ι: O(n − d + 1)<br />
∼=<br />
−→ K ∨<br />
X .<br />
The metric ‖ . ‖ on K ∨<br />
X induces a metric ι−1 ‖ . ‖ 1<br />
n+d−1 on O(1)|X (). Let<br />
N := N ι −1 ‖ .‖<br />
be the subset defined by ι −1 ‖ . ‖ 1<br />
n+d−1 .<br />
1<br />
n−d+1<br />
⊂ CX (Ê)<br />
Then, in terms of N <strong>and</strong> the Leray measure, one may write<br />
τ ∞ (U ) = n − d + 1 Z<br />
ω Leray<br />
2<br />
for every measurable set U ⊆ X (Ê).<br />
Proof. First step: Explicit description of ι.<br />
CU ∩N<br />
The assertion is local in X. We may therefore assume without restriction that<br />
X 0 ≠ 0 <strong>and</strong> ∂F<br />
∂X 1<br />
≠ 0. Then, t 2 , . . . , t n for t i := X i /X 0 form a local system of<br />
coordinates on X.
Sec. 5] PEYRE’S CONSTANT 233<br />
When putting F = X d 0 f (t 1, . . . , t n ), we find<br />
∂F<br />
= X d−1 ∂f<br />
0 .<br />
∂X 1 ∂t 1<br />
Under the Poincaré residue map [G/H, p. 147], the differential form<br />
∂f<br />
∂t 1<br />
df<br />
f<br />
∧ dt 2 ∧ . . . ∧ dt n = ∂f<br />
∂t 1<br />
dt 1 ∧ . . . ∧ dt n<br />
on P nis mapped to dt 2 ∧ . . . ∧ dt n . Dually, under this correspondence, the<br />
tensor field ∂<br />
∂t 2<br />
∧ . . . ∧ ∂<br />
∂t n<br />
on X () is identified with<br />
f ∂<br />
∧ . . . ∧ ∂ = 1 F ∂<br />
∧ . . . ∧ ∂ .<br />
∂t 1 ∂t n X<br />
∂F 0 ∂t<br />
∂X 1 ∂t n 1<br />
Furthermore, the Euler sequence identifies ∂<br />
∂t 1<br />
∧ . . . ∧ ∂<br />
∂t n<br />
with the global section<br />
X n+1<br />
0 ∈ Γ ( P n , O(n + 1) ) .<br />
Altogether,<br />
ι −1 ( ∂<br />
∂t 2<br />
∧ . . . ∧ ∂<br />
∂t n<br />
)<br />
= 1<br />
∂F<br />
∂X 1<br />
X n 0 .<br />
Second step: The measure τ ∞ of Peyre.<br />
Peyre’s measure τ ∞ on X (Ê) is therefore given by the differential form<br />
1<br />
∥ ∂F ∥ dt 2 ∧ . . . ∧ dt n = | X0 d−1 / ∂F<br />
∂X 1<br />
|·‖X n−d+1<br />
0 ‖dt 2 ∧ . . . ∧ dt n . (§)<br />
∂X 1<br />
X n 0<br />
On the other h<strong>and</strong>, the subset N ⊂ CX (Ê) is given by<br />
| x 0 | ≤ A := ‖X 0 ‖ .<br />
Therefore, according to Fubini’s theorem,<br />
Z<br />
CU ∩N<br />
ω Leray =<br />
=<br />
=<br />
Z<br />
CU<br />
|x 0 |≤A<br />
Z A<br />
−A<br />
1<br />
| ∂F<br />
x n−1<br />
0<br />
x d−1<br />
0<br />
∂X 1<br />
| dx 0 ∧ dx 2 ∧ . . . ∧ dx n<br />
dx 0 ·<br />
Z<br />
CU<br />
x 0 =1<br />
2<br />
n − d + 1 · An−d+1 ·<br />
1<br />
| ∂F<br />
∂X 1<br />
| dx 2 ∧ . . . ∧ dx n<br />
Z<br />
CU<br />
x 0 =1<br />
1<br />
| ∂F<br />
∂X 1<br />
| dx 2 ∧ . . . ∧ dx n .<br />
Note here that CX is an (n −1)-dimensional cone while the integr<strong>and</strong> is homogeneous<br />
of degree −(d − 1).
234 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
Consequently,<br />
n − d + 1<br />
2<br />
Z<br />
CU ∩N<br />
ω Leray =<br />
Z<br />
CU<br />
x 0 =1<br />
‖X 0 ‖ n−d+1<br />
dx<br />
| ∂F 2 ∧ . . . ∧ dx n<br />
∂X 1<br />
|<br />
which, according to formula (§), is exactly equal to τ ∞ (U ).<br />
□<br />
5.31. Remarks. –––– i) This result allows a generalization to complete intersections.<br />
ii) If X is a cubic surface then n − d + 1 = 1.<br />
On the other h<strong>and</strong>, consider the case that X is a hypersurface in P n for n ≥ 4.<br />
Then, Pic(X ) = 1 <strong>and</strong> α(X ) = 1 . We therefore have that<br />
n−d+1<br />
α(X )·τ ∞ (U ) = 1 Z<br />
ω Leray<br />
2<br />
for every measurable set U ⊆ X (Ê).<br />
vi. The Tamagawa measure. —<br />
CU ∩N<br />
5.32. Definition. –––– The Tamagawa measure τ H on the set X (É) of<br />
<strong>points</strong> on X is defined to be the product measure<br />
τ H := ∏ τ ν .<br />
ν∈Val(É)<br />
adelic<br />
5.33. Remark. –––– X is projective. Therefore, one has that<br />
X (É) = ∏ X (Éν) .<br />
ν∈Val(É)<br />
5.34. Lemma. –––– The infinite product<br />
(<br />
∏ τ )<br />
ν X (Éν)<br />
ν∈Val(É)<br />
is absolutely convergent. In particular, the infinite product measure ∏ τ ν is well defined.<br />
ν∈Val(É)<br />
Proof. Absolute convergence may not be destroyed by a finite set of factors.<br />
Thus, assume that almost all of the metrics ‖ . ‖ νp of the adelic metric<br />
‖ . ‖ = {‖ . ‖ ν } ν∈Val(É) are induced by a model X of X. Further, we may<br />
restrict the infinite product to all prime numbers p such that X is smooth<br />
over p.<br />
Corollary 5.18 assures that<br />
(<br />
τ ) p X (Ép) = det ( )<br />
1 − p −1 Frob p | Pic(XÉ) I #X (p)<br />
p · . ()<br />
pdim X
Sec. 5] PEYRE’S CONSTANT 235<br />
Further, smoothness over p implies that<br />
Pic(XÉ) I p<br />
= Pic(XÉ) .<br />
We denote the eigenvalues of Frob p on Pic(XÉ) by λ 1 , . . . , λ r . As every<br />
divisor on is actually defined over a finite field extension, all these are roots<br />
Xp<br />
of unity.<br />
Since Pic(XÉ) is a-module of finite rank, the characteristic polynomial<br />
of Frob p is monic with integral coefficients. This shows, if λ is an eigenvalue<br />
then λ = 1 λ is an eigenvalue of Frob p, too.<br />
Clearly, one has<br />
det ( 1 − p −1 Frob p | Pic(XÉ) I p) = (1 − λ1 p −1 ) · . . . · (1 − λ r p −1 )<br />
= 1 − (λ 1 + . . . + λ r )p −1 + E<br />
where<br />
| E| <<br />
( ( k k<br />
p<br />
2)<br />
−2 + . . . + p<br />
k)<br />
−k<br />
for p > 2k.<br />
< k2<br />
p 2 + k3<br />
p 3 + . . .<br />
< 2k2<br />
p 2<br />
In order to determine #X (p), we use the Lefschetz trace formula in étale cohomology<br />
[SGA4 1 , Rapport, Théorème 3.2]. This yields<br />
2<br />
for every prime l ≠ p.<br />
dimX p<br />
#X (p) =<br />
∑<br />
i=0<br />
(−1) i tr ( )<br />
Frob p | H i ét ,Él) (‖)<br />
(Xp<br />
Here, according to the Weil conjectures proven by P. Deligne [Del, Théorème<br />
(1.6)], every eigenvalue of the Frobenius on H i ét (Xp ,Él) is of absolute<br />
value p i/2 .<br />
The theorem on smooth base change implies a comparison theorem even for<br />
the unequal characteristic case [SGA4, Exp. XVI, Corollaire 2.5 <strong>and</strong> Exp. XV,<br />
Théorème 2.1]. We therefore know that<br />
H i ét (Xp ,Él) ∼ = H i ét (XÉ,Él) .<br />
According to [SGA4, Exp. XVI, Corollaire 1.6 <strong>and</strong> Exp. XI, Théorème 4.4], the<br />
latter is isomorphic to the usual cohomology H i (X (),Él).
236 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
Let C be a constant such that dim H i (X (),Él) ≤ C for every i ∈Æ. Then,<br />
dim H i ét (Xp ,Él) ≤ C<br />
for every i ∈Æ, too. Further, Lemma 4.2 describes the first <strong>and</strong> second<br />
cohomologies of a Fano variety more concretely. We have H 1 ét (Xp ,Él) = 0<br />
while the first Chern class induces an isomorphism<br />
Pic(XÉ)⊗Él<br />
∼= H 2 ét (XÉ,Él(1))<br />
∼= H 2 ét (Xp ,Él(1)) .<br />
This isomorphism is compatible with the operation of Frob p . Consequently,<br />
the Frobenius eigenvalues on H 2 ét (Xp ,Él) are λ 1 p, . . . , λ r p.<br />
2 dim X −1<br />
Poincaré duality shows that Hét (Xp ,Él) = 0. Further, by [Del, (2.4)],<br />
Poincaré duality is compatible with the Frobenius eigenvalues. Therefore, the<br />
eigenvalues of the Frobenius on H 2 dimX−2 ,Él) are<br />
ét<br />
(Xp<br />
λ 1 p dimX−1 , . . . , λ r p dimX−1 .<br />
For the number ofp-<strong>rational</strong> <strong>points</strong>, this yields, according to (‖),<br />
where<br />
X (p) = p dim X + (λ 1 + . . . + λ r )p dim X −1 + Dp dim X<br />
| D| < C [p −3/2 + p −2 + p −5/2 + . . . ] < Cp −3/2 1<br />
1 − p −1/2 < 4Cp−3/2 .<br />
Altogether,<br />
τ p<br />
(<br />
X (Ép) )<br />
= (1 − (λ 1 + . . . + λ r )p −1 + E) · (1 + (λ 1 + . . . + λ r )p −1 + D)<br />
= 1 + E + D + (λ 1 + . . . + λ r ) 2 p −2 + (E − D)(λ 1 + . . . + λ r )p −1 + ED .<br />
For p > 4C 2 , we certainly have<br />
| E| < 2k2<br />
p < 2C 2<br />
2 p 2<br />
Therefore, τ p<br />
(<br />
X (Ép) ) = 1 + X where<br />
= 4C · C<br />
2<br />
4C<br />
<<br />
p3/2·p1/2 p . 3/2<br />
| X | < 8C<br />
p + C 2<br />
3/2 p + 8C 2 2<br />
16C<br />
+ 2 p5/2 p 3<br />
< 8C + C 2<br />
2C + 8C 2<br />
4C 2 + 16C 2<br />
8C 3<br />
p 3/2
Sec. 5] PEYRE’S CONSTANT 237<br />
= 2 + 17<br />
2 C + 2 C<br />
p 3/2 .<br />
Since 3/2 > 1, the infinite product is absolutely convergent.<br />
□<br />
vii. The constant of E. Peyre. —<br />
5.35. Definition (E. Peyre, [Pe/T, Definition 2.4]). —–<br />
Peyre’s constant or Peyre’s Tamagawa type number is defined as<br />
τ (X ) := α(X )·β(X ) · lim<br />
s→1<br />
(s − 1) t L(s, χ Pic(XÉ)) · τ H<br />
(<br />
X (É)<br />
for t = rk Pic(X ).<br />
Br)<br />
5.36. Remark. –––– Here, X (É) denotes the part which is not<br />
affected by the Brauer-Manin obstruction. The Brauer-Manin obstruction was<br />
discussed in detail in Chapter III. The precise definition of X (É) Br is given<br />
in III.2.5. According to Proposition III.2.3.b.ii), X (É) Br is a closed subset<br />
of X (É) <strong>and</strong> therefore measurable.<br />
Br ⊆ X (É)<br />
5.37. –––– Thus, there is the following conjecture which refines Conjecture<br />
4.11.<br />
Conjecture (Manin-Peyre). Let X be a smooth, projective variety overÉ.<br />
AssumethatX isFano. DenotebyH K ∨ theanticanonicalheightfunctionintroduced<br />
in Definition 4.6.<br />
a) If dim X ≤ 2 or X ∈ P n is a complete intersection then there exist a real<br />
number τ <strong>and</strong> a Zariski open subset X ◦ ⊆ X such that<br />
for B → ∞.<br />
N X ◦ ,H K ∨(B) = #{x ∈ X ◦ (k) | H K ∨(x) < B} ∼ τB log rk Pic(X )−1 B<br />
b) Suppose that, for X,thereexist a realnumber τ <strong>and</strong> aZariski opensubset X ◦ ⊆ X<br />
such that<br />
for B → ∞.<br />
N X ◦ ,H K ∨(B) = #{x ∈ X ◦ (k) | H K ∨(x) < B} ∼ τB log rk Pic(X )−1 B<br />
Then, τ = τ (X ) is Peyre’s constant.
238 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
5.38. Remarks (Some motivation). —– i) Morally, the conjecture of Manin in<br />
the refined form due to Peyre states the following. The (conjectural) constant<br />
describing the growth of the number ofÉ-<strong>rational</strong> <strong>points</strong> on X is equal to a<br />
regularized product over the densities of the p-adic <strong>points</strong> on X together with<br />
the density of the real <strong>points</strong>.<br />
That there is such a connection is actually not very surprising. For instance,<br />
a low density of p-adic <strong>points</strong> on X for a particular prime p implies strong<br />
congruence conditions forÉ-<strong>rational</strong> <strong>points</strong>.<br />
ii) More precise results have been obtained by the classical circle method [Bir].<br />
For complete intersections in a very high-dimensional projective space, the<br />
circle method provides an asymptotic formula for the number ofÉ-<strong>rational</strong><br />
<strong>points</strong> on X <strong>and</strong> an error term. The assumptions on the dimension of the<br />
ambient projective space are rather restrictive. They are needed in order to<br />
ensure that the provable error term is smaller than the main term.<br />
The coefficient of the main term is a product of p-adic densities together with a<br />
factor corresponding to the archimedean valuation. Unlike the p-adic densities,<br />
the latter factor does not coincide in general with E. Peyre’s factor τ ∞<br />
(<br />
X (Ê) ) .<br />
Rather, it is an integral over the Leray measure (at least in the case that X is<br />
a hypersurface).<br />
By Proposition 5.30, the factor corresponding to the archimedean valuation<br />
may be rewritten in the form α(X ) · τ ∞<br />
(<br />
X (Ê) ) . Therefore, one sees that the<br />
coefficient of the main term provided by the circle method is equal to<br />
Here,<br />
(<br />
∏τ ) (<br />
p X (Ép) · α(X ) · τ ) (<br />
∞ X (Ê) = α(X ) · τ H X (É) p<br />
τ p<br />
(<br />
X (Ép) ) =<br />
) .<br />
(<br />
1 − 1 )<br />
#{ y ∈ X (/p n) | y ≡ a (mod p k ) }<br />
· lim<br />
.<br />
p n→∞ p n dim X<br />
The (1 − 1 ) are convergence-generating factors.<br />
p<br />
In the cases where the circle method is applicable, one always has Pic(X ) ∼ =.<br />
It seems very natural to use (1 − 1 p )t for t = rk Pic(XÉ) as the convergencegenerating<br />
factors for the general case.<br />
In the definition of Peyre’s constant, this is done <strong>and</strong> more than that. Indeed, the<br />
Gal(É/É)-representation Pic(XÉ) I p ⊗contains the trivial representation<br />
t times. Denote the complementary summ<strong>and</strong> by P. Then, the factors used in<br />
Definition 5.20 may be decomposed as<br />
det ( (<br />
)<br />
1 − p −1 Frob p | Pic(XÉ) I p = 1 − 1 t<br />
· det<br />
p) ( 1 − p −1 Frob p | P ) .
Sec. 5] PEYRE’S CONSTANT 239<br />
The factors det ( 1 − p −1 Frob p | P ) form the Euler product for<br />
1/L(s, χ P )<br />
at s = 1. The value of the function L at 1 is introduced into Definition 5.35<br />
only in order to cancel these factors out.<br />
iii) The definition of the factor α(X ) is somehow more complicated <strong>and</strong>, may<br />
be, more mysterious than those of the other parts. Some motivation for the<br />
appearance of such a factor is given by the circle method.<br />
There is, however, another point which is at least as important. The definition<br />
of α(X ) implies that the conjecture of Manin-Peyre is compatible with direct<br />
products of Fano varieties.<br />
iv) For a smooth cubic surface of the “first case” of Colliot-Thélène, Kanevsky,<br />
<strong>and</strong> Sansuc, the assertion of Theorem III.6.4.c) implies that<br />
(<br />
τ ) H X (É) .<br />
On the other h<strong>and</strong>, β(X ) = 3 such that one might want to simplify Peyre’s<br />
formula to<br />
) . (∗∗)<br />
Br) = 1 3 ·τ H(<br />
X (É)<br />
τ (X ) := α(X ) · lim<br />
s→1<br />
(s − 1) t L(s, χ Pic(XÉ)) · τ H<br />
(<br />
X (É)<br />
Clearly, this is also true in all cases when β(X ) = 1.<br />
Formula (∗∗) is, however, wrong in general. For example, when X is a smooth<br />
cubic surface on which there is a Brauer-Manin obstruction to the Hasse principle<br />
then we need τ (X ) = 0 <strong>and</strong> this is not provided by the simplified formula.<br />
Furthermore, in [Pe/T], E. Peyre <strong>and</strong> Y. Tschinkel reported strong numerical<br />
evidencefor Peyre’sformulaincasessimilar toExampleIII.5.36. I.e., ifβ(X ) = 3<br />
but the Brauer-Manin obstruction does not exclude any adelic point on X then<br />
there are three times moreÉ-<strong>rational</strong> <strong>points</strong> on X than naively expected.<br />
5.39. Remark (Known cases). —– As indicated above, Manin’s conjecture in<br />
the refined form due to Peyre is established for smooth complete intersections<br />
of multidegree d 1 , . . . , d n in the case that the dimension of V is very large<br />
compared to d 1 , . . . , d n [Bir]. This is the classical circle method.<br />
Further, Conjecture 5.37 is proven for projective spaces <strong>and</strong> quadrics. Finally,<br />
there are a number of further particular cases in which Manin’s conjecture<br />
is known to be true. See, e.g., the survey given in [Pe02, Sec. 4].<br />
5.40. Remark (Numerical evidence). —– R. Heath-Brown [H-B92a] as well as<br />
E. Peyre <strong>and</strong> Y. Tschinkel [Pe/T] reported numerical evidence for Conjecture<br />
5.37 for isolated examples of smooth cubic surfaces.
240 CONJECTURES ON POINTS OF BOUNDED HEIGHT [Chap. VI<br />
In Chapter VII, we present numerical evidence for Manin’s conjecture in the case<br />
of the threefolds V e<br />
a,b given by axe = by e + z e + v e + w e in P 4Éfor e = 3 <strong>and</strong> 4.<br />
In Chapter IX, we test Manin’s conjecture numerically for general cubic surfaces.<br />
5.41. Remark. –––– It is not all trivial that the conditionally convergent Euler<br />
product<br />
1<br />
det ( 1 − p −1 Frob p | P )<br />
∏<br />
p<br />
really converges versus L(1, χ P ). The point is that the value L(1, χ P ) is actually<br />
defined by analytic continuation.<br />
5.42. Remark. –––– The idea behind Peyre’s constant is in fact an old one.<br />
The main term of the asymptotic formula provided by the circle method was<br />
used before, for example, by D. R. Heath-Brown in his experimental investigations<br />
on cubic surfaces where weak approximation fails [H-B92a].<br />
Actually, Heath-Brown’s point of view was a little different. He considered<br />
the factors (1 − 1 p )t as convergence-generating factors growing out of the circle<br />
method. The resulting conditionally convergent infinite product was treated<br />
like a definition. The value of the function L at 1 appeared, too, but only as<br />
part of a numerical method to speed up convergence.<br />
E. Peyre’s approach is most likely inspired by considerations like those<br />
in [H-B92a]. It has, however, the potential to work in much more generality.
PART C<br />
NUMERICAL EXPERIMENTS
CHAPTER VII<br />
POINTS OF BOUNDED HEIGHT ON CUBIC<br />
AND QUARTIC THREEFOLDS ∗<br />
. . . , one by one, or all at once.<br />
W. S. GILBERT AND A. SULLIVAN: The Yeomen of the Guard (1888)<br />
1. Introduction — Manin’s Conjecture<br />
i. Summary. —<br />
1.1. –––– For the families ax 3 = by 3 + z 3 + v 3 + w 3 , a, b = 1, . . . , 100, <strong>and</strong><br />
ax 4 = by 4 +z 4 +v 4 +w 4 , a, b = 1, . . . , 100, of projective algebraic threefolds,<br />
we test numerically the conjecture of Manin (in the refined form due to Peyre,<br />
cf. Conjecture VI.5.37) about the asymptotics of <strong>points</strong> of bounded height on<br />
Fano varieties.<br />
This includes searching for <strong>points</strong>, computing the Tamagawa number, <strong>and</strong> detecting<br />
the accumulating subvarieties. The goal of this chapter is describe these<br />
computations as well some background on the geometry of cubic <strong>and</strong> quartic<br />
threefolds.<br />
ii. Manin’s conjecture. —<br />
1.2. –––– Let X be a projective algebraic variety overÉ. We fix an embedding<br />
ι: X → P nÉ. In this situation, recall there is the well-known naive<br />
height H naive : X (É) →Êgiven by<br />
H naive (P) := max<br />
i=0,...,n | x i| .<br />
Here, (x 0 : . . . : x n ) := ι(P) ∈ P n (É) where the projective coordinates are<br />
integers satisfying gcd(x 0 , . . . , x n ) = 1.<br />
(∗) This chapter is a revised <strong>and</strong> slightly extended version of the article: The Asymptotics of<br />
Points of Bounded Height on Diagonal Cubic <strong>and</strong> Quartic Threefolds, in: Algorithmic number<br />
theory, Lecture Notes in Computer Science 4076, Springer, Berlin 2006, 317–332, joint with<br />
A.-S. Elsenhans.
244 POINTS OF BOUNDED HEIGHT ON THREEFOLDS [Chap. VII<br />
1.3. –––– It is of interest to ask for the asymptotics of the number ofÉ-<strong>rational</strong><br />
<strong>points</strong> on X of bounded naive height. This applies particularly to Fano varieties<br />
as those are expected to have many <strong>rational</strong> <strong>points</strong> (at least after a finite extension<br />
of the ground-field).<br />
Simplest examples of Fano varieties are complete intersections in P nÉof a multidegree<br />
(d 1 , . . . , d r ) such that d 1 + . . . + d r ≤ n. In this case, the conjecture<br />
of Manin discussed in general in Chapter VI reads as follows.<br />
1.4. Conjecture. –––– Let X ⊆ P nÉbe a non-singular complete intersection of<br />
multidegree (d 1 , . . . , d r ). Assumedim X ≥ 3<strong>and</strong>k := n+1−d 1 − . . . −d r > 0.<br />
Then, there exists a Zariski open subset X ◦ ⊆ X such that<br />
#{x ∈ X ◦ (É) | H naive (x) k < B} ∼ τB<br />
(∗)<br />
for a constant τ.<br />
1.5. Example. –––– Let X ⊂ P 4Ébe a smooth hypersurface of degree 4.<br />
Conjecture 1.4 predicts ∼ τB <strong>rational</strong> <strong>points</strong> of height < B. However, the<br />
hypersurface x 4 + y 4 = z 4 + v 4 + w 4 contains the line given by x = z, y = v,<br />
<strong>and</strong> w = 0 on which there is quadratic growth, already. This explains the<br />
necessity of the restriction to a Zariski open subset X ◦ ⊆ X.<br />
1.6. Remark. –––– Conjecture 1.4 is proven for P n , linear subspaces, <strong>and</strong><br />
quadrics. Further, it is established [Bir] in the case that the dimension of X<br />
is very large compared to d 1 , . . . , d r . The general version (Conjecture VI.5.37)<br />
is known to be true in a number of further particular cases. A rather complete<br />
list may be found in the survey article [Pe02, Sec. 4].<br />
In this chapter, we present numerical evidence for Conjecture 1.4 in the case of<br />
the varieties X e a,b given by axe = by e + z e + v e + w e in P 4Éfor e = 3 <strong>and</strong> 4.<br />
1.7. Remark. –––– By the Noether-Lefschetz Theorem, the assumptions made<br />
on X imply that Pic(XÉ) ∼ =[Ha70, Corollary IV.3.2]. This is no longer true<br />
in dimension two. See Remark 1.14.ii) for more details.<br />
iii. The constant. —<br />
1.8. –––– Conjecture 1.4 is compatible with results obtained by the classical<br />
circle method (e.g. [Bir]). Motivated by this, E. Peyre provided a description<br />
of the constant τ expected in (∗). We formulated the definition for the general<br />
case in VI.5.35.<br />
In the situation considered here, Pic(XÉ) ∼ =implies that β(X ) = 1 <strong>and</strong> there<br />
is no Brauer-Manin obstruction on X. Peyre’s constant is therefore equal to the
Sec. 1] INTRODUCTION — MANIN’S CONJECTURE 245<br />
Tamagawa-type number<br />
τ (X ) := α(X ) · ∏ (1 − 1 ) ω p p(X (Ép)) .<br />
p∈È∪{∞}<br />
In this formula, α(X ) := 1/k for k ∈Æsuch that O(−K ) ∼ = O(k).<br />
The measure ω p is given in local p-adic analytic coordinates x 1 , . . . , x d by<br />
∂<br />
∥ ∧ . . . ∧<br />
∂ ∥ ∥∥∥p<br />
dx 1 . . . dx d .<br />
∂x 1 ∂x d<br />
Here, each dx i denotes a Haar measure onÉp which is normalized in the<br />
∂<br />
usual manner.<br />
∂x 1<br />
∧ . . . ∧ ∂<br />
∂x d<br />
is a section of O(−K ).<br />
1.9. –––– For p finite, one has a natural model X ⊆ P of X given by the<br />
np<br />
defining equation. This induces the metric ‖ . ‖ p on O(k). It is almost immediate<br />
from the definition that<br />
(<br />
ω ) #X (/p m)<br />
p X (Ép) = lim .<br />
m→∞ p m dim X<br />
1.10. Remark. –––– There is another description of ω p of interest for finite p.<br />
One has<br />
(<br />
ω ) p X (Ép) = 1 − p−k<br />
lim<br />
1 − p #CX (/p m)/p (d+1)m .<br />
−1 m→∞<br />
Here, CX denotes the affine cone over X . For a proof, see [Pe/T, Corollary<br />
3.5].<br />
1.11. –––– The hermitian metric ‖ . ‖ ∞ corresponding the naive height H naive<br />
is given by ‖ . ‖ ∞ := ‖ . ‖ k min on O(−K ) ∼ = O(k). Here, ‖ . ‖ min is the hermitian<br />
metric on O(1) defined by<br />
‖x i ‖ min := inf<br />
j=0,...,n | x i/x j | .<br />
The factor ω ∞<br />
(<br />
X (É∞) ) may then be described as follows.<br />
1.12. Lemma. –––– If X ⊂ P n is a hypersurface defined by the equation f = 0<br />
then<br />
(<br />
α(X ) · ω ) ∞ X (Ê) = 1 Z<br />
ω Leray .<br />
2<br />
The Leray measure ω Leray on<br />
f (x 0 ,...,x n )=0<br />
|x 0 |,...,|x n |≤1<br />
{(x 0 , . . . , x n ) ∈Ên+1 | f (x 0 , . . . , x n ) = 0}
246 POINTS OF BOUNDED HEIGHT ON THREEFOLDS [Chap. VII<br />
1<br />
is related to the usual hypersurface measure by the formula ω Leray =<br />
‖ grad f ‖ ω hyp.<br />
On the other h<strong>and</strong>, one may also write<br />
1<br />
ω Leray =<br />
| ∂ f<br />
∂x i<br />
(x 0 , . . . , x n )| dx 0 ∧ . . . ∧ ̂dx i ∧ . . . ∧ dx n .<br />
Proof. The equivalence of the two descriptions of the Leray measure is shown<br />
in Lemma VI.5.26. The main assertion is proven in Proposition VI.5.30. □<br />
1.13. Remark. –––– As the description of the constant τ given here grew out<br />
of a definition given in canonical terms, it is no surprise that τ is invariant<br />
under scaling.<br />
This can also be seen directly. When we multiply f by a prime number p then<br />
τ p gets multiplied by a factor of p. On the other h<strong>and</strong>, τ ∞ gets multiplied by a<br />
factor of 1/p <strong>and</strong> all the other factors remain unchanged.<br />
1.14. Remark. –––– There are several ways to generalize Conjecture 1.4.<br />
These are presented in detail in Chapter VI. For completeness, let us recall<br />
the following facts.<br />
i) One may consider more general heights corresponding to the tautological<br />
invertible sheaf O(1). This includes to<br />
a) replace the minimum metric by an arbitrary continuous hermitian metric<br />
on O(1). This would affect the domain of integration for the factor at infinity.<br />
b) multiply H naive (x) with a function that depends on the reduction of x modulo<br />
some N ∈Æ. This augments Conjecture 1.4 by an equidistribution statement.<br />
ii) Instead of complete intersections, one may consider arbitrary projective Fano<br />
varieties X. Then, H k naive needs to be replaced by a height defined by the<br />
anticanonical sheaf O(−K X ).<br />
If Pic(XÉ) ̸∼ =then the description of the constant C gets more complicated in<br />
several ways. First, there is an additional factor β := #H 1( Gal(É/É), Pic(XÉ) ) .<br />
Further, instead of (1 − 1 ), one has to use more complicated convergencegenerating<br />
factors. (Cf. Remark VI.5.38.) Finally, the Tamagawa measure has<br />
p<br />
to be taken not of the full variety X (É) but of the subset which is not affected<br />
by the Brauer-Manin obstruction.<br />
If Pic(X ) ̸∼ =already overÉthen the right h<strong>and</strong> side of (∗) has to be replaced<br />
by CB log t B. For the exponent of the log-term, there is the expectation that<br />
t = rk Pic(X ) − 1. There are, however, examples [Ba/T] in dimension three in<br />
which the exponent is larger.<br />
The definition of α is rather complicated, in general. The factor α depends on<br />
the structure of the effective cone in Pic(X ) <strong>and</strong> on the position of (−K X ) in it.<br />
(Cf. Definition VI.5.1).
Sec. 2] COMPUTING THE TAMAGAWA NUMBER 247<br />
2. Computing the Tamagawa number<br />
i. Counting <strong>points</strong> over finite fields. —<br />
2.1. –––– We consider the projective varieties X e a,b<br />
given by<br />
ax e = by e + z e + v e + w e<br />
in P . We assume a, b ≠ 0 (<strong>and</strong> p ∤e) in order to avoid singularities. Observe<br />
that, even for large p, these are at most e 2 varieties up to obviousp-<br />
4p<br />
isomorphism as∗ p consists of no more than e cosets modulo (∗ p ) e .<br />
It follows from the Weil conjectures, proven by P. Deligne [Del, Théorème (8.1)],<br />
that<br />
#X e a,b (p) = p 3 + p 2 + p + 1 + E e a,b<br />
where the error-term E e a,b may be estimated by | Ee a,b | ≤ C e p 3/2 .<br />
Here, C 3 = 10 <strong>and</strong> C 4 = 60 as dim H 3 (X 3 ,Ê) = 10 for every smooth cubic<br />
threefold<strong>and</strong>dim H 3 (X 4 ,Ê) = 60for everysmoothquarticthreefoldinP 4 ().<br />
These dimensions result from the Weak Lefschetz Theorem together with<br />
F. Hirzebruch’s formula [Hi, Satz 2.4] for the Euler characteristic which actually<br />
works in much more generality.<br />
2.2. Remark. –––– Suppose e = 3 <strong>and</strong> p ≡ 2 (mod 3). Then,<br />
#X 3 a,b (p) = #X 1 a,b (p)<br />
as gcd(p − 1, 3) = 1. Similarly, for e = 4 <strong>and</strong> p ≡ 3 (mod 4), one has<br />
gcd(p − 1, 4) = 2 <strong>and</strong><br />
#X 4 a,b (p) = #X 2 a,b (p).<br />
In these cases, the error term vanishes <strong>and</strong><br />
#X e a,b (p) = p 3 + p 2 + p + 1.<br />
In the remaining cases p ≡ 1 (mod 3) for e = 3 <strong>and</strong> p ≡ 1 (mod 4) for e = 4,<br />
our goal is to compute the number ofp-<strong>rational</strong> <strong>points</strong> on X e a,b . As X e a,b ⊆ P4 ,<br />
there would be an obvious O(p 4 )-algorithm. We can do significantly better<br />
than that.<br />
2.3. Definition. –––– Let K be a field <strong>and</strong> let x ∈ K n <strong>and</strong> y ∈ K m be two vectors.<br />
Then, their convolution z := x∗y ∈ K n+m−1 is defined to be z k := ∑ x i y j .<br />
i+j=k
248 POINTS OF BOUNDED HEIGHT ON THREEFOLDS [Chap. VII<br />
2.4. Theorem (FFT convolution). —– Let n = 2 l <strong>and</strong> K be a field which contains<br />
the 2n-th roots of unity. Then, the convolution x ∗ y of two vectors x, y of<br />
length ≤ n can be computed in O(n logn) steps.<br />
Proof. The idea is to apply the Fast Fourier Transform (FFT) [Fo, Satz 20.3].<br />
The connection to the convolution is shown in [Fo, Satz 20.2] or [C/L/R,<br />
Theorem 32.8].<br />
□<br />
Theorem 2.4 is the basis for the following algorithm.<br />
2.5. Algorithm (FFT point counting on X e a,b<br />
). —–<br />
i) Initialize a vector x[0 . . . p] with zeroes.<br />
ii) Let r run from 0 to p − 1 <strong>and</strong> increase x[r e mod p] by 1.<br />
iii) Calculate ỹ := x ∗ x ∗ x by FFT convolution.<br />
iv) Normalize by putting y[i] := ỹ[i] + ỹ[i + p] + ỹ[i + 2p] for each<br />
i ∈ { 0, . . . , p − 1 }.<br />
v) Initialize N as zero.<br />
vi) (Counting <strong>points</strong> with first coordinate ≠ 0)<br />
Let j run from 0 to p − 1 <strong>and</strong> increase N by y[(a − bj 4 ) mod p].<br />
vii) (Counting <strong>points</strong> with first coordinate 0 <strong>and</strong> second coordinate ≠ 0)<br />
Increase N by y[(−b) mod p].<br />
viii) (Counting <strong>points</strong> with first <strong>and</strong> second coordinate 0)<br />
Increase N by (y[0] − 1)/(p − 1).<br />
ix) Return N as the number of allp-valued <strong>points</strong> on X e a,b .<br />
2.6. Remark. –––– For the running-time, step iii) is dominant. Therefore, the<br />
running-time of Algorithm 2.5 is O(p log p).<br />
To count, for fixed e <strong>and</strong> p,p-<strong>rational</strong> <strong>points</strong> on X e a,b<br />
with varying a <strong>and</strong> b,<br />
one needs to execute the first four steps only once. Afterwards, one may<br />
perform steps v) through ix) for all pairs (a, b) of elements from a system<br />
of representatives for∗ p /(∗ p ) e . Note that steps v) through ix) alone are of<br />
complexity O(p).<br />
2.7. –––– We ran this algorithm for all primes p ≤ 10 6 <strong>and</strong> stored the cardinalities<br />
in a file.<br />
2.8. Remark. –––– There is a formula for #X e a,b (p) in terms of Jacobi sums.<br />
A skilful manipulation of these sums should lead to another efficient algorithm<br />
which serves the same purpose as Algorithm 2.5.
Sec. 2] COMPUTING THE TAMAGAWA NUMBER 249<br />
ii. The local factors at finite places. —<br />
2.9. –––– We are interested in the Euler product<br />
τ e a,b,fin := ∏ −<br />
p∈È(1 1 )<br />
p<br />
lim<br />
m→∞<br />
#X e<br />
a,b<br />
(/pm)<br />
.<br />
p 3m<br />
2.10. Lemma. –––– a) (Good reduction)<br />
If p ∤ abe then the sequence (#X e<br />
a,b (/p m)/p 3m ) m∈Æis constant.<br />
b) (Bad reduction)<br />
i) If p divides ab but not e then the sequence (#X e<br />
a,b (/p m)/p 3m ) m∈Æbecomes<br />
stationary as soon as p m divides neither a nor b.<br />
ii) If p = 2 <strong>and</strong> e = 4 then the sequence (#X e<br />
a,b (/p m)/p 3m ) m∈Æbecomes<br />
stationary as soon as 2 m does not divide 8a or 8b.<br />
iii) If p = 3 <strong>and</strong> e = 3 then the sequence (#X e<br />
a,b (/p m)/p 3m ) m∈Æbecomes<br />
stationary as soon as 3 m divides neither 3a nor 3b.<br />
iii. An estimate. —<br />
2.11. Theorem. –––– For every pair (a, b) of integers such that a, b ≠ 0, the<br />
Euler product τ e a,b,fin<br />
is convergent.<br />
Proof. Cf. Lemma VI.5.34 where this is proven in more generality.<br />
Let p be a prime bigger than | a|, | b|, <strong>and</strong> e. Then, the factor at p is<br />
τ p := (1 − 1 p )(1 + p + p2 + p 3 + D p p 3/2 )/p 3<br />
where |D p | ≤ C e for C 3 = 10 <strong>and</strong> C 4 = 60, respectively. As sums are easier to<br />
estimate than products, we take a look at the logarithm,<br />
log τ p = D p<br />
p 3/2 + O(p−5/2 ) .<br />
Taking the logarithm, we consider ∑ p log τ p . In the case e = 3, the sum is<br />
effectively over the primes p = 1 (mod 3). If e = 4 then summation extends<br />
over all primes p = 1 (mod 4). In either case, we take a sum over one-half of
250 POINTS OF BOUNDED HEIGHT ON THREEFOLDS [Chap. VII<br />
all primes. This leads to the following estimate,<br />
[ ]<br />
Ce<br />
∑ | log τ p | ≤ ∑<br />
p≥N<br />
p≥N<br />
p + 3/2 O(p−5/2 ) ∼ C Z ∞<br />
e<br />
2<br />
≤<br />
N<br />
C e<br />
2 logN<br />
1<br />
t 3/2 logt dt<br />
Z ∞<br />
N<br />
t −3/2 dt =<br />
C e<br />
√ N logN<br />
.<br />
2.12. Remark. –––– We are interested in an∣explicit upper bound for<br />
∣∣∣∣ ∣ ∑ log τ p .<br />
p≥10 6<br />
Using Taylor’s formula, one gets<br />
∣ ∣ ∑<br />
D ∣∣∣∣<br />
p<br />
log τ p − ∑ ≤ 10 −8 .<br />
p<br />
p≥10 6 p≥10 3/2 6<br />
Since D p<br />
is zero for p ≡ 3 (mod 4) (or p ≡ 2 (mod 3)), the sum should be compared<br />
with log(ζ K (3/2)). Here, ζ K is the Dedekind zeta function of K =É(i)<br />
p 3/2<br />
orÉ(ζ 3 ), respectively. This yields<br />
∑<br />
p≥10 6<br />
p≡1(mod 4)<br />
√<br />
1<br />
≤ log<br />
p3/2 (1 − 2 −3/2 ) −1/2 · ∏<br />
p≡3(mod 4)<br />
ζÉ(i)(3/2)<br />
(1 − p −3 ) −1/2 · ∏(1 − p −3/2 ) −1<br />
p
Sec. 2] COMPUTING THE TAMAGAWA NUMBER 251<br />
numbers had been precomputed using FFT point counting (Algorithm 2.5).<br />
The algorithm below is based on the fact that the vast majority of the factors<br />
actually do not need to be computed. They are available from a list.<br />
2.15. Algorithm (Compute an approximate value for τ 3 (τ 4 a,b,fin a,b,fin<br />
)). —–<br />
i) Let p run over all prime numbers such that p ≡ 2 (mod 3) (p ≡ 3 (mod 4))<br />
<strong>and</strong> p ≤ N <strong>and</strong> calculate the product of all values of (1 − 1/p 4 ).<br />
ii) Compute the factor corresponding to p = 3 (p = 2) by Lemma 2.10.b).<br />
iii) Let p run over all prime numbers such that p ≡ 1 (mod 3) (p ≡ 1 (mod 4))<br />
<strong>and</strong> p ≤ N . Calculate the product of the factors described below.<br />
If p|ab then the corresponding factor is given by Lemma 2.10.b). Otherwise,<br />
compute the e-th power residue-symbols of a <strong>and</strong> b <strong>and</strong> look up the<br />
precomputed factor for thisp-isomorphism class of varieties in the list.<br />
iv) Multiply the two products from steps i) <strong>and</strong> iii) <strong>and</strong> the factor from step ii)<br />
with each other. Correct the product by taking the bad primes p ≡ 2 (mod 3)<br />
(p ≡ 3 (mod 4)) into consideration.<br />
2.16. Remark. –––– When we meet a bad prime p, we have to count<br />
/p m-valued <strong>points</strong> on X e<br />
a,b<br />
. This is done by an algorithm which is very<br />
similar to Algorithm 2.5.<br />
2.17. –––– We used Algorithm 2.15 to compute the Euler products τ 3 a,b,fin<br />
<strong>and</strong> τ 4 a,b,fin for a, b = 1, . . . , 100. We did all calculations for N = 106 .<br />
Note that step i) had to be done only once for e = 3 <strong>and</strong> once for e = 4.<br />
v. The factor at the infinite place. —<br />
2.18. –––– In order to get an approximation for the integrals, we tried to use<br />
some st<strong>and</strong>ard methods from numerical analysis [Kr, Chapter 9], particularly<br />
the Gauß-Legendre formula. A direct application of this method is known to<br />
work well as long as the integr<strong>and</strong> is smooth enough <strong>and</strong> the dimension of the<br />
domain of integration is not too big.<br />
2.19. –––– For the quartic X 4 a,b<br />
, we have the integral<br />
over<br />
ω ∞<br />
(<br />
X<br />
4<br />
a,b (Ê) ) = 1<br />
4 4 √ a<br />
ZZZZ<br />
R<br />
1<br />
dy dz dv dw<br />
(by 4 + z 4 + v 4 + w 4 )<br />
3/4<br />
R := {(y, z, v, w) ∈Ê4 | |y|, |z|, |v|, |w| ≤ 1 <strong>and</strong> |by 4 + z 4 + v 4 + w 4 | ≤ a} .<br />
The integr<strong>and</strong> is singular in one point. We used a simple substitution to make<br />
it sufficiently smooth for numerical integration.
252 POINTS OF BOUNDED HEIGHT ON THREEFOLDS [Chap. VII<br />
On the other h<strong>and</strong>, for the cubic X 3 a,b<br />
, we have to consider<br />
ω ∞<br />
(<br />
X<br />
3<br />
a,b (Ê) ) = 1<br />
6 3 √ a<br />
ZZZZ<br />
R<br />
1<br />
dy dz dv dw<br />
(by 3 + z 3 + v 3 + w 3 )<br />
2/3<br />
for R := {(y, z, v, w) ∈Ê4 | |y|, |z|, |v|, |w| ≤ 1 <strong>and</strong> |by 3 +z 3 +v 3 +w 3 | ≤ a} .<br />
The difficulty here is the h<strong>and</strong>ling of the singularity of the integr<strong>and</strong>. It is located<br />
in the zero set of by 3 + z 3 + v 3 + w 3 in R which is a cone over a cubic surface.<br />
Since (by 3 +z 3 +v 3 +w 3 ) −2/3 is a homogeneous function, it is enough to integrate<br />
over the boundary of R. This reduces the problem to several three-dimensional<br />
integrals of functions having a two-dimensional singular locus. If a ≥ b + 3<br />
then R is a cube <strong>and</strong> the boundary of R is easy to describe. We restricted<br />
our attention to this case. We smoothed the singularities by separation of<br />
Puiseux expansions <strong>and</strong> substitutions. The resulting integrals were treated by<br />
the Gauß-Legendre formula [Kr].<br />
3. On the geometry of diagonal cubic threefolds<br />
i. Surfaces on a hypersurface in P 4 . —<br />
3.1. Lemma. –––– Let X ⊂ P 4 be any smooth hypersurface. Then, every (reduced<br />
but possibly singular) surface S ⊂ X is a complete intersection X ∩ H d with a<br />
hypersurface H d ⊂ P 4 .<br />
Proof. By the Noether-Lefschetz Theorem, we have Pic(X ) ∼ =. The surface<br />
S is a Weil divisor on X. Hence, O(S) = O(d) ∈ Pic(X ) for a<br />
certain d > 0. The restriction Γ(P 4 , O(d)) → Γ(X, O(d)) is surjective as<br />
H 1 (P 4 , O X (d − degX)) = 0 [Ha77, Theorem III.5.1.b)].<br />
□<br />
ii. Elliptic Cones. —<br />
3.2. –––– Let X ⊂ P 4 () be the diagonal cubic threefold given by the equation<br />
x 3 + y 3 + z 3 + v 3 + w 3 = 0. Fix ζ ∈such that ζ 3 = 1. Then, for every<br />
point (x 0 : y 0 : z 0 ) on the elliptic curve F : x 3 + y 3 + z 3 = 0, the line given by<br />
(x : y : z) = (x 0 : y 0 : z 0 ) <strong>and</strong> v = −ζw is contained in X. All these lines<br />
together form a cone CF over F the cusp of which is (0 : 0 : 0 : −ζ : 1). C F is<br />
a singular model of a ruled surface over an elliptic curve. This shows, there are<br />
no other <strong>rational</strong> curves contained in C F .
Sec. 3] ON THE GEOMETRY OF DIAGONAL CUBIC THREEFOLDS 253<br />
By permuting coordinates, one finds a total of thirty elliptic cones of that type<br />
within X. The cusps of these cones are usually named Eckardt <strong>points</strong> [Mu-e,<br />
Cl/G]. We call the lines contained in one of these cones the obvious lines lying<br />
on X. It is clear that there are an infinite number of lines on X running through<br />
each of the thirty Eckardt <strong>points</strong> (1 : −1 : 0 : 0 : 0), (1 : 0 : −1 : 0 : 0),<br />
. . . , (0 : 0 : 0 : 1 : −1), (1 : −e 2πi/3 : 0 : 0 : 0), . . . , (0 : 0 : 0 : 1 : −e −2πi/3 ).<br />
3.3. Proposition. (cf. [Mu-e, Lemma 1.18]). —– Let X ⊂ P 4 be the diagonal<br />
cubic threefold given by theequation x 3 +y 3 +z 3 +v 3 +w 3 = 0. Then, through each<br />
pointP ∈ X differentfromthethirtyEckardt<strong>points</strong>therearepreciselysixlinesonX.<br />
Proof. Let P = (x 0 : y 0 : z 0 : v 0 : w 0 ). A line l through<br />
P <strong>and</strong> another point Q = (x : y : z : v : w) is parametrized by<br />
(s : t) ↦→ ((sx 0 + tx) : . . . : (sw 0 + tw)). Comparing coefficients at s 2 t, st 2 ,<br />
<strong>and</strong> t 3 , we see that the condition that l lies on X may be expressed by the three<br />
equations below.<br />
x 2 0x + y 2 0y +z 2 0z + v 2 0v + w 2 0w = 0 (†)<br />
x 0 x 2 + y 0 y 2 + z 0 z 2 + v 0 v 2 + w 0 w 2 = 0 (‡)<br />
x 3 + y 3 + z 3 + v 3 + w 3 = 0 (§)<br />
The first equation means that Q lies on the tangent hyperplane H P at P while<br />
equation (§) just encodes that Q ∈ X. By Lemma 3.6, H P ∩ X is an irreducible<br />
cubic surface.<br />
On the other h<strong>and</strong>, the quadratic form q on the left h<strong>and</strong> side of equation (‡)<br />
is of rank at least 3 as P is not an Eckardt point. Therefore, q is not just the<br />
product of two linear forms. In particular, q| HP ≢ 0.<br />
As H P ∩X is irreducible, Z(q| HP ) <strong>and</strong> H P ∩X do not have a component in common.<br />
By Bezout’s theorem, their intersection in H P is a curve of degree 6. □<br />
3.4. Remark. –––– It may happen that some of the six lines coincide. Actually,<br />
it turns out that a line appears with multiplicity > 1 if <strong>and</strong> only if it is<br />
obvious [Mu-e, Lemma 1.19]. In particular, for a general point P the six lines<br />
through it are different from each other.<br />
Under certain exceptional circumstances it is possible to write down all six<br />
lines explicitly. For example, if P = ( 3 √ −4 : 1 : 1 : 1 : 1) then the line<br />
( 3 √ −4t : (t + s) : (t + is) : (t − s) : (t − is)) through P lies on X. Permuting the<br />
three rightmost coordinates yields all six lines.<br />
3.5. Remark. –––– The following lemma is a special case of Zak’s theorem [Za,<br />
Corollary 1.8] which is more elementary <strong>and</strong> completely sufficient for our purposes.<br />
We present it here for convenience of the reader.
254 POINTS OF BOUNDED HEIGHT ON THREEFOLDS [Chap. VII<br />
3.6. Lemma. –––– Let X ⊂ P 4 be the diagonal cubic threefold given by the equation<br />
x 3 +y 3 +z 3 +v 3 +w 3 = 0. Then, for any hyperplane H ⊂ P 4 , the intersection<br />
H ∩ X is irreducible <strong>and</strong> has at most finitely many singular <strong>points</strong>.<br />
Proof. The intersection of X with a hyperplane H is singular precisely in those<br />
<strong>points</strong> where H is tangent to X.<br />
The tangent hyperplane H at (x 0 : y 0 : z 0 : v 0 : w 0 ) ∈ X is given by<br />
x 2 0 x + y2 0 y + z2 0 z + v2 0 v + w2 0w = 0. From this formula, we see that H is<br />
tangent to X at all <strong>points</strong> of the form (±x 0 : ±y 0 : ±z 0 : ±v 0 : ±w 0 ) which<br />
happen to lie on X <strong>and</strong> at no others. By consequence, H ∩ X admits only a<br />
finite number of singular <strong>points</strong>.<br />
Every irreducible component of H ∩ X is a hypersurface in the projective<br />
3-space H . Two such components would intersect in a curve of singular <strong>points</strong>.<br />
Therefore, H ∩ X is necessarily irreducible.<br />
□<br />
3.7. Remark. –––– For a general point p ∈ X, the intersection of its tangent<br />
hyperplane with X admits exactly one singular point. Indeed, let<br />
p = (x 0 : y 0 : z 0 : v 0 : w 0 ). If (−x 0 : y 0 : z 0 : v 0 : w 0 ) ∈ X then x 0 = 0.<br />
If (−x 0 : −y 0 : z 0 : v 0 : w 0 ) ∈ X then x 0 + ζy 0 = 0 for ζ a third root of<br />
unity or ζ = 1. At this point, up to permutation of coordinates, the list of all<br />
possibilities is already complete. Multi-tangent hyperplanes are caused only by<br />
<strong>points</strong> lying on a finite arrangement of hyperplanes.<br />
4. Accumulating subvarieties<br />
i. The detection ofÉ-<strong>rational</strong> lines on the cubics. —<br />
4.1. –––– On a cubic threefold X 3 a,b<br />
, quadratic growth is predicted for the<br />
number ofÉ-<strong>rational</strong> <strong>points</strong> of bounded height. Lines are the only curves with<br />
such a growth rate.<br />
The moduli space of the lines on a cubic threefold is well-understood. It is a<br />
surface of general type [Cl/G, Lemma 10.13]. Nevertheless, we do not know<br />
of a method to find allÉ-<strong>rational</strong> lines on a given cubic threefold, explicitly.<br />
For that reason, we use the algorithm below which is an ir<strong>rational</strong>ity test for<br />
the six lines through a given point (x 0 : y 0 : z 0 : v 0 : w 0 ) ∈ X 3 a,b (É).<br />
4.2. Algorithm (Test the six lines through a given point for ir<strong>rational</strong>ity). —–<br />
i) Let p run through the primes from 3 to N .<br />
For each p, solve the system of equations (†), (‡), (§) (adapted to X 3 a,b ) in5 p .<br />
If the multiples of (x 0 , y 0 , z 0 , v 0 , w 0 ) are the only solutions then output that there<br />
is noÉ-<strong>rational</strong> line through (x 0 : y 0 : z 0 : v 0 : w 0 ) <strong>and</strong> terminate prematurely.
Sec. 4] ACCUMULATING SUBVARIETIES 255<br />
ii) If the loop comes to its regular end then output that the point is suspicious.<br />
It could possibly lie on aÉ-<strong>rational</strong> line.<br />
4.3. Remark. –––– We use a very naive O(p)-algorithm to solve the system<br />
of equations overp. If, say, x 0 ≠ 0 then it is sufficient to consider quintuples<br />
such that x = 0. We parametrize the projective plane given by (†). Then, we<br />
compute all <strong>points</strong> on the conic given by (†) <strong>and</strong> (‡). For each such point, we<br />
compute the cubic form on the left h<strong>and</strong> side of (§). When a non-trivial solution<br />
is found, we stop immediately.<br />
We carried out the ir<strong>rational</strong>ity test on everyÉ-<strong>rational</strong> point found on any<br />
of the cubics except for the <strong>points</strong> lying on an obvious line. We worked<br />
with N = 600. It turned out that suspicious <strong>points</strong> are rare <strong>and</strong> that, at least in<br />
our sample, each of them actually lies on aÉ-<strong>rational</strong> line.<br />
The lines found. We found only 42 non-obviousÉ-<strong>rational</strong> lines on all of the<br />
cubics X 3 a,b<br />
for 100 ≥ a ≥ b ≥ 1 together. Among them, there are only five<br />
essentially different ones. We present them in the table below. The list might<br />
be enlarged by two, as X21,6 3 <strong>and</strong> X 22,5 3 may be transformed into X 48,21 3 <strong>and</strong> X 40,22 3 ,<br />
respectively, by an automorphism of P 4 . Further, each line has six pairwise<br />
different images under the obvious operation of the group S 3 .<br />
TABLE 1. Sporadic lines on the cubic threefolds<br />
a b Smallest point Point s.t. x = 0<br />
19 18 (1 : 2 : 3 : -3 : -5) (0 : 7 : 1 : -7 : -18)<br />
21 6 (1 : 2 : 3 : -3 : -3) (0 : 9 : 1 : -10 : -15)<br />
22 5 (1 : -1 : 3 : 3 : -3) (0 : 27 : -4 : -60 : 49)<br />
45 18 (1 : 1 : 3 : 3 : -3) (0 : 3 : -1 : 3 : -8)<br />
73 17 (1 : 5 : -2 : 11 : -15) (0 : 27 : -40 : 85 : -96)<br />
4.4. Remark. –––– It is a priori unnecessary to search for accumulating surfaces,<br />
at least if we assume the conjectures of Batyrev/Manin <strong>and</strong> Lang formulated in<br />
Chapter VI.<br />
First of all, only <strong>rational</strong> surfaces are supposed to accumulate that many <strong>rational</strong><br />
<strong>points</strong> that it could be seen through our asymptotics of O(B 2 ). Indeed, a surface<br />
which is abelian or bielliptic may not have more than O(log t B) <strong>points</strong> of<br />
height < B. Non-<strong>rational</strong> ruled surfaces accumulate <strong>points</strong> in curves, anyway.<br />
Further, it is expected (Conjecture VI.3.1) that K3 surfaces, Enriques surfaces,<br />
<strong>and</strong> surfaces of Kodaira dimension one may have no more than O(B ε ) <strong>points</strong><br />
of height < B outside a finite union of <strong>rational</strong> curves. For surfaces of general<br />
type, finally, expectations are even stronger (Conjecture VI.2.2).
256 POINTS OF BOUNDED HEIGHT ON THREEFOLDS [Chap. VII<br />
A <strong>rational</strong> surface S is, up to exceptional curves, the image of a <strong>rational</strong><br />
map ϕ: P 2 X ⊂ P 4 . There is a bi<strong>rational</strong> morphism ε: P → P 2 such<br />
that ϕ := ϕ ◦ ε is a morphism of schemes. ε is given by a sequence of blowingups<br />
[Bv, Theorem II.11]. ϕ is defined by the linear system |dH − E| where<br />
d := deg ϕ, H is a hyperplane section, <strong>and</strong> E is the exceptional divisor. On the<br />
other h<strong>and</strong>, K := K P = −3H + E. Therefore, if d ≥ 3 then<br />
H naive (ϕ(p)) = H dH −E (p) = H 3H −<br />
3<br />
d E(p)d/3 ≥ c · H −K (p) d/3<br />
for p ∉ supp(E). Manin’s conjecture implies there are O((B log t B) 3 d ) = o(B 2 )<br />
<strong>points</strong> of height < B on the Zariski-dense subset ϕ(P \ supp(E)) ⊆ S.<br />
It remains to show that there are no <strong>rational</strong> maps ϕ: P 2 <br />
X of degree<br />
d ≤ 2. Indeed, under this assumption, deg ϕ(P 2 ) ≤ 4. This implies,<br />
by virtue of Lemma 3.1, that ϕ(P 2 ) is necessarily a hyperplane section X ∩ H .<br />
Lemma 3.6 shows that X ∩ H contains only finitely many singular <strong>points</strong>.<br />
It is, however, well known that cubic surfaces in 3-space which are the image<br />
of P 2 under a quadratic map have a singular line [Bv, Corollary IV.8].<br />
ii. The detection ofÉ-<strong>rational</strong> conics on the quartics. —<br />
4.5. –––– On a quartic threefold, linear growth is predicted for the number of<br />
É-<strong>rational</strong> <strong>points</strong> of bounded height. The assumption b > 0 ensures that there<br />
are noÊ-<strong>rational</strong> lines contained in X 4 a,b<br />
. The only other curves with at least<br />
linear growth one could think about are conics.<br />
We do not know of a method to find allÉ-<strong>rational</strong> conics on a given quartic<br />
threefold, explicitly. Worse, we were unable to create an efficient routine to test<br />
whether there is aÉ-<strong>rational</strong> conic through a given point. The resulting system<br />
of equations seems to be too complicated to h<strong>and</strong>le.<br />
Conics through two <strong>points</strong>. A conic Q through (x 0 : y 0 : z 0 : v 0 : w 0 )<br />
<strong>and</strong> (x 1 : y 1 : z 1 : v 1 : w 1 ) may be parametrized in the form<br />
(s : t) ↦→ ((λx 0 s 2 + µx 1 t 2 + xst) : . . . : (λw 0 s 2 + µw 1 t 2 + wst))<br />
for some x, y, z, v, w, λ, µ ∈. The condition that Q is contained in X 4 a,b leads<br />
to a system G of seven equations in x, y, z, v, w, <strong>and</strong> λµ. The phenomenon<br />
that λ <strong>and</strong> µ do not occur individually is explained by the fact that they are not<br />
invariant under the automorphisms of P 1 which fix 0 <strong>and</strong> ∞.<br />
4.6. Algorithm (Test for conic through two <strong>points</strong>). —– i) Let p run through<br />
the primes from 3 to N .
Sec. 4] ACCUMULATING SUBVARIETIES 257<br />
In the exceptional case that G could allow a solution such that p|x, y, z, v, w but<br />
p 2 ∤ λµ, do nothing. Otherwise, solve G in6 p . If (0, 0, 0, 0, 0, 0)is the only solution<br />
then output that there is noÉ-<strong>rational</strong> conic through (x 0 : y 0 : z 0 : v 0 : w 0 )<br />
<strong>and</strong> (x 1 : y 1 : z 1 : v 1 : w 1 ) <strong>and</strong> terminate prematurely.<br />
ii) If the loop comes to its regular end then output that the pair is suspicious.<br />
It could possibly lie on aÉ-<strong>rational</strong> conic.<br />
4.7. –––– To solve the system G in6 p , we use an O(p)-algorithm. Actually,<br />
comparison of coefficients at s 7 t <strong>and</strong> st 7 yields two linear equations in x, y,<br />
z, v, <strong>and</strong> w. We parametrize the projective plane I given by them. Comparison<br />
of coefficients at s 6 t 2 <strong>and</strong> s 2 t 6 leads to a quadric O <strong>and</strong> an equation<br />
λµ = q(x, y, z, v, w)/M with a quadratic form q over<strong>and</strong> an integer<br />
M ≠ 0. The case p|M sends us to the next prime immediately. Otherwise,<br />
we compute all <strong>points</strong> on the conic I ∩ O. For each of them, we<br />
test the three remaining equations. When a non-trivial solution is found, we<br />
stop immediately.<br />
Conics through three <strong>points</strong>. Three <strong>points</strong> P 1 , P 2 , <strong>and</strong> P 3 on X 4 a,b<br />
define a projective<br />
plane P. The <strong>points</strong> together with the two tangent lines P∩ T P1 <strong>and</strong> P∩ T P2<br />
determine a conic Q, uniquely. It is easy to transform this geometric insight<br />
into a formula for a parametrization of Q. We then need a test whether a conic<br />
given in parametrized form is contained in X 4 a,b<br />
. This part is algorithmically<br />
simple but requires the use of multiprecision integers.<br />
Detecting conics. For each quartic X 4 a,b<br />
, we tested every pair ofÉ-<strong>rational</strong><br />
<strong>points</strong> of height < 100 000 for a conic through them. The existence of a<br />
conic through (P, Q) is equivalent to the existence of a conic through (gP, gQ)<br />
for g ∈ (/2) 4 ⋊S 3 ⊆ Aut(X 4 a,b<br />
). This reduces the running time by a factor of<br />
about 96. Further, pairs already known to lie on the same conic were excluded<br />
from the test.<br />
For each pair (P, Q) found suspicious, we tested the triples (P, Q, R) for R running<br />
through theÉ-<strong>rational</strong> <strong>points</strong> of height < 100 000, until a conic was found.<br />
Due to the symmetries, one finds several conics at once. For each conic detected,<br />
all <strong>points</strong> on it were marked as lying on this conic.<br />
Actually, there were a few pairs found suspicious through which no conic could<br />
be found. In any of these cases, it was easy to prove by h<strong>and</strong> that there is actually<br />
noÊ-<strong>rational</strong> conic passing through the two <strong>points</strong>. This means, we detected<br />
every conic which meets at least two of the <strong>rational</strong> <strong>points</strong> of height < 100 000.<br />
Concerning our programming efforts, this was the most complex part of the<br />
entire project.
258 POINTS OF BOUNDED HEIGHT ON THREEFOLDS [Chap. VII<br />
The conics found. Up to symmetry, we found a total of 1 664É-<strong>rational</strong> conics<br />
on all of the quartics X 4 a,b<br />
for 1 ≤ a, b ≤ 100 together.<br />
Among them, 1 538 are contained in a plane of type z = v+w <strong>and</strong> Yx−Xy = 0<br />
for (X, Y, t) a <strong>rational</strong> point on the genus one curve aX 4 − bY 4 = 2t 2 . Further,<br />
there are 93 conics which are slight modifications of the above with y<br />
interchanged with z, v, or w. This is possible if b is a fourth power.<br />
There is a geometric explanation for the occurrence of these conics. The hyperplane<br />
given by z = v + w intersects X 4 a,b<br />
in a surface S with the two singular<br />
<strong>points</strong> (0 : 0 : −1 : e ±2πi/3 : e ∓2πi/3 ). The linear projection π : S P 1 to the<br />
first two coordinates is undefined only in these two <strong>points</strong>. Its fibers are plane<br />
quartics which split into two conics as (v+w) 4 +v 4 +w 4 = 2(v 2 +vw+w 2 ) 2 . After<br />
resolution of singularities, the two conics become disjoint. ˜S is a ruled surface<br />
over a twofold cover of P 1 ramified in the four <strong>points</strong> such that ax 4 − by 4 = 0.<br />
I.e., over a curve of genus one.<br />
In the case a is twice a square, a different sort of conics comes from the equations<br />
v = z + Dy <strong>and</strong> w = Ly when (L, D) is a point on the affine genus three curve<br />
C b : L 4 + b = D 4 . Here, by 4 + z 4 + v 4 + w 4 becomes twice a full square<br />
after the substitutions. This explains why this particular intersection of X 4 a,b<br />
with a plane splits into two conics. We found 28 conics of this type. C b has<br />
aÉ-<strong>rational</strong> point for b = 5, 15, 34, 39, 65, 80, <strong>and</strong> 84. The conics actually<br />
admit aÉ-<strong>rational</strong> point for a = 2, 18, 32, <strong>and</strong> 98.<br />
The remaining five conics are given as follows. For a = 3, 12, 27, or 48<br />
<strong>and</strong> b = 10, intersect with the plane given by v = y + z <strong>and</strong> w = 2y + z.<br />
For a = 17 <strong>and</strong> b = 30, put v = 2x + y <strong>and</strong> w = x + 3y + z.<br />
4.8. Remark. –––– Again, it is not necessary to search for accumulating surfaces.<br />
Here, <strong>rational</strong> maps ϕ: P 2 X ⊂ P 4 such that deg ϕ ≤ 3 need to be<br />
taken into consideration. We claim, such a map is impossible.<br />
If deg ϕ = 3 then we had ϕ: (λ : µ : ν) ↦→ (K 0 (λ, µ, ν) : . . . : K 4 (λ, µ, ν))<br />
where K 0 , . . . , K 4 are cubic forms defined overÉ. K 0 = 0 defines a plane cubic<br />
which has infinitely many real <strong>points</strong>, automatically. As the image of ϕ<br />
is assumed to be contained in X 4 a,b , we have that K 0(λ, µ, ν) = 0 implies<br />
K 1 (λ, µ, ν) = . . . = K 4 (λ, µ, ν) = 0 for λ, µ, ν ∈Ê. By consequence,<br />
K 1 , . . . , K 4 are divisible by K 0 (or by a linear factor of K 0 in the case it is<br />
reducible) <strong>and</strong> ϕ is not of degree three.<br />
For deg ϕ ≤ 2, we had deg ϕ(P 2 ) ≤ 4 such that ϕ(P 2 ) = X ∩ H is a hyperplane<br />
section. Lemma 3.6 shows it has at most finitely many singular <strong>points</strong>.<br />
On the other h<strong>and</strong>, a quartic in P 3 which is the image of a quadratic map from P 2<br />
is a Steiner surface. It is known [Ap, p. 40] to have one, two, or (in generic case)<br />
three singular lines.
Sec. 5] RESULTS 259<br />
5. Results<br />
i. A technology to find solutions of Diophantine equations. —<br />
5.1. –––– In Chapter XI (cf. [EJ2] <strong>and</strong> [EJ3]), we describe a modification<br />
of D. Bernstein’s [Be] method to search efficiently for all solutions of naive<br />
height < B of a Diophantine equation of the particular form<br />
f (x 1 , . . . , x n ) = g(y 1 , . . . , y m ) .<br />
The expected running-time of our algorithm is O(B max{n,m} ). Its basic idea is<br />
as follows.<br />
5.2. Algorithm (Search for solutions of a Diophantine equation). —–<br />
i) (Writing)<br />
Evaluate f on all <strong>points</strong> of the cube {(x 1 , . . . , x n ) ∈n | |x i | < B} of<br />
dimension n. Store the values within a hash table H .<br />
ii) (Reading)<br />
Evaluate g on all <strong>points</strong> of the cube {(y 1 , . . . , y m ) ∈m | |y i | < B}. For each<br />
value, start a search in order to find out whether it occurs in H . When a<br />
coincidence is detected, reconstruct the corresponding values of x 1 , . . . , x n <strong>and</strong><br />
output the solution.<br />
5.3. Remark. –––– In the case of a variety X e a,b<br />
, the running-time is obviously<br />
O(B 3 ). We decided to store the values of z e +v e +w e into the hash table.<br />
Afterwards, we have to look up the values of ax e − by e .<br />
In this form, the algorithm would lead to a program in which almost the entire<br />
running-time is consumed by the writing part. Observe, however, the following<br />
particularity of our method. When we search on up to O(B) threefolds,<br />
differing only by the values of a <strong>and</strong> b, simultaneously, then the running-time<br />
is still O(B 3 ).<br />
5.4. Remark (Running times). —– We worked with B = 5 000 for the cubics<br />
<strong>and</strong> B = 100 000 for the quartics. In either case, we dealt with all threefolds<br />
arising for a, b = 1, . . . , 100, simultaneously.<br />
The by far largest portion of the running time was spent on point search on<br />
the quartics. All in all, this took around 155 days of CPU time. This is<br />
approximately only three times longer than searching on a single threefold<br />
had lasted. Searching for conics was done within 32 days. In comparison with<br />
this, the corresponding computations for the cubics could be done in a negligible<br />
amount of time. The reason for this is simply that for the cubics the search
260 POINTS OF BOUNDED HEIGHT ON THREEFOLDS [Chap. VII<br />
bound was by far lower. A program with integrated line detection took us<br />
approximately ten days.<br />
To compute Peyre’s constants, the precomputation was actually the main part.<br />
Point counting using FFT took 100 hours for the cubics <strong>and</strong> 100 hours for<br />
the quartics. For the final computation, the running-time was a quarter of an<br />
hour for either exponent.<br />
ii. The results for the cubics. —<br />
5.5. –––– We counted allÉ-<strong>rational</strong> <strong>points</strong> of height less than 5 000 on the<br />
threefolds X 3 a,b where a, b = 1, . . . , 100 <strong>and</strong> b ≤ a. Note that X 3 a,b ∼ = X 3 b,a .<br />
Points lying on one of the elliptic cones or on a sporadicÉ-<strong>rational</strong> line<br />
in X a,b were excluded from the count. The smallest number of <strong>points</strong> found is<br />
3 930 278 for (a, b) = (98, 95). The largest numbers of <strong>points</strong> are 332 137 752<br />
for (a, b) = (7, 1) <strong>and</strong> 355 689 300 in the case that a = 1 <strong>and</strong> b = 1.<br />
On the other h<strong>and</strong>, for each threefold X 3 a,b<br />
whereas a, b = 1, . . . , 100<br />
<strong>and</strong> b + 3 ≤ a, we calculated the expected number of <strong>points</strong> <strong>and</strong> the quotients<br />
# { <strong>points</strong> of height < B found } / # { <strong>points</strong> of height < B expected }.<br />
Let us visualize the quotients by two histograms.<br />
250<br />
250<br />
200<br />
200<br />
150<br />
150<br />
100<br />
100<br />
50<br />
50<br />
0 0.88 0.9 0.92 0.94 0.96 0.98 1 1.02 0 0.88 0.9 0.92 0.94 0.96 0.98 1 1.02<br />
FIGURE 1. Distribution of the quotients for B = 1 000 <strong>and</strong> B = 5 000.
Sec. 5] RESULTS 261<br />
The statistical parameters are listed in the table below.<br />
TABLE 2. Parameters of the distribution in the cubic case<br />
B = 1 000 B = 2 000 B = 5 000<br />
mean value 0.98179 0.98854 0.99383<br />
st<strong>and</strong>ard deviation 0.01274 0.00823 0.00455<br />
iii. The results for the quartics. —<br />
5.6. –––– We counted allÉ-<strong>rational</strong> <strong>points</strong> of height less than 100 000 on<br />
the threefolds X 4 a,b<br />
where a, b = 1, . . . , 100. It turns out that on 5 015 of<br />
these varieties, there are noÉ-<strong>rational</strong> <strong>points</strong> occurring at all as the equation<br />
is unsolvable inÉp for some small p. In this situation, Manin’s conjecture is<br />
true, trivially.<br />
For the remaining varieties, the <strong>points</strong> lying on a knownÉ-<strong>rational</strong> conic in X a,b<br />
were excluded from the count. Table 3 shows the quartics sorted by the numbers<br />
of <strong>points</strong> remaining.<br />
TABLE 3. Numbers of <strong>points</strong> of height < 100 000 on the quartics.<br />
a b # <strong>points</strong> # not on conic # expected<br />
(by Manin-Peyre)<br />
29 29 2 2 135<br />
58 87 288 288 272<br />
58 58 290 290 388<br />
87 87<br />
. .<br />
386<br />
.<br />
386<br />
.<br />
357<br />
.<br />
34 1 9938976 5691456 5 673000<br />
17 64 5708664 5708664 5 643000<br />
1 14 7205502 6361638 6 483000<br />
3 1 12657056 7439616 7 526000<br />
We see that the variation of the quotients is higher than in the cubic case.
262 POINTS OF BOUNDED HEIGHT ON THREEFOLDS [Chap. VII<br />
20<br />
15<br />
10<br />
5<br />
0 0.7 0.8 0.9 1 1.1 1.2 1.3<br />
FIGURE 2. Distribution of the quotients for B = 100 000.<br />
7<br />
7<br />
6<br />
6<br />
5<br />
5<br />
4<br />
4<br />
3<br />
3<br />
2<br />
2<br />
1<br />
1<br />
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2<br />
FIGURE 3. Distribution of the quotients for B = 1 000 <strong>and</strong> B = 10 000.<br />
The statistical parameters are listed in the table below.
Sec. 5] RESULTS 263<br />
TABLE 4. Parameters of the distribution in the quartic case<br />
B = 1 000 B = 10 000 B = 100 000<br />
mean value 0.9853 0.9957 0.9982<br />
st<strong>and</strong>ard deviation 0.3159 0.1130 0.0372<br />
iv. Interpretation of the result. —<br />
5.7. –––– The results suggest that Manin’s conjecture should be true for the<br />
two families of threefolds considered. In the cubic case, the st<strong>and</strong>ard deviation<br />
is by far smaller than in the case of the quartics. This, however, is not very<br />
surprising as on a cubic there tend to be much more <strong>rational</strong> <strong>points</strong> than on a<br />
quartic. This makes the sample more reliable.<br />
5.8. Remark. –––– The data we collected might be used to test the sharpening<br />
of the asymptotic formula (∗) suggested by Sir P. Swinnerton-Dyer [S-D05].<br />
5.9. Question. –––– Our calculations seem to indicate that the number of <strong>rational</strong><br />
<strong>points</strong> often approaches its expected value from below. Is that more than<br />
an accidental effect?
CHAPTER VIII<br />
ON THE SMALLEST POINT ON A<br />
DIAGONAL QUARTIC THREEFOLD ∗<br />
In the course of your work, you will from time to time encounter the situation<br />
where the facts <strong>and</strong> the theory do not coincide. In such circumstances,<br />
young gentlemen, it is my earnest advice to respect the facts.<br />
IGOR IVANOVICH SIKORSKY<br />
1. A computer experiment<br />
1.1. –––– Let X ⊆ P nÉbe a Fano variety defined overÉ. If X (Éν) ≠ ∅ for<br />
every ν ∈ Val(É) then it is natural to ask whether X (É) ≠ ∅ (Hasse’s principle).<br />
Further, it would be desirable to have an a-priori upper bound for the height<br />
of the smallestÉ-<strong>rational</strong> point on X as this would allow to effectively decide<br />
whether X (É) ≠ ∅ or not.<br />
When X is a conic, Legendre’s theorem on zeroes of ternary quadratic forms<br />
proves the Hasse principle <strong>and</strong>, moreover, yields an effective bound for the<br />
smallest point. For quadrics of arbitrary dimension, the same is true by an<br />
observation due to J. W. S. Cassels [Cas55]. Further, there is a theorem of<br />
C. L. Siegel [Si73, Satz 1] which provides a generalization to hypersurfaces<br />
defined by norm equations. For more general Fano varieties, no theoretical<br />
upper bound is known for the height of the smallestÉ-<strong>rational</strong> point. Some of<br />
these varieties fail the Hasse principle.<br />
In this chapter, we present some theoretical <strong>and</strong> experimental results concerning<br />
the height of the smallestÉ-<strong>rational</strong> point on quartic hypersurfaces in P 4É.<br />
1.2. –––– There is the conjecture, due to Yu. I. Manin, that the number ofÉ-<strong>rational</strong><br />
<strong>points</strong> of anticanonical height < B on a Fano variety X is asymptotically<br />
equal to τB log rkPic(X )−1 B, for B → ∞. A detailed description of Manin’s<br />
conjecture is given in Chapter VI.<br />
(∗) This chapter is a revised version of the article: On the smallest point on a diagonal quartic<br />
threefold, J. Ramanujan Math. Soc. 22(2007), 189–204, joint with A.-S. Elsenhans.
266 ON THE SMALLEST POINT ON A DIAGONAL QUARTIC THREEFOLD [Chap. VIII<br />
In the particular case of a quartic threefold, the anticanonical height is the<br />
same as the naive height. Further, rkPic(X ) = 1 <strong>and</strong>, finally, the coefficient<br />
τ ∈Êequals the Tamagawa-type number τ (X ) introduced by E. Peyre (Definition<br />
VI.5.35).<br />
Thus, one expects ∼τ (X )B <strong>points</strong> of height < B. Assuming equidistribution,<br />
the height of the smallest point should be ∼ 1<br />
τ<br />
. Being a bit optimistic, this<br />
(X )<br />
might lead to the expectation that m(X ), the height of the smallestÉ-<strong>rational</strong><br />
point on X, is always less than C<br />
τ<br />
for a certain absolute constant C.<br />
(X )<br />
1.3. –––– To test this expectation, we computed the Tamagawa number <strong>and</strong><br />
ascertained the smallestÉ-<strong>rational</strong> point for each of the quartic threefolds<br />
X (−a,b)<br />
4<br />
⊂ P 4Égiven by ax 4 = by 4 + z 4 + v 4 + w 4 for a, b = 1, . . . , 1000.<br />
On 516 820 of these varieties, there are noÉ-<strong>rational</strong> <strong>points</strong> as the equation is<br />
unsolvable inÉp for p = 2, 5, or 29. (Note that, for each prime p different<br />
from 2, 5, or 29, there are p-adic <strong>points</strong> already on the Fermat quartic given<br />
by z 4 + v 4 + w 4 = 0.) On each of the remaining varieties,É-<strong>rational</strong> <strong>points</strong><br />
were found. In other words, there are no counterexamples to the Hasse principle<br />
in this family.<br />
The computation of Tamagawa numbers is explained in Chapter VII, Section 2.<br />
The methods to systematically search for solutions of Diophantine equations<br />
we applied are described in Chapter VII, Section 5.<br />
The results are summarized by the diagrams below.<br />
a, b = 1 . . . 100 a, b = 1 . . . 1000<br />
FIGURE 1. Height of smallest point versus Tamagawa number<br />
It is apparent from the diagrams that the experiment agrees with the expectation<br />
above. The slope of a line tangent to the top right of each of the scatter<br />
plots is indeed near (−1). However, we show in Section 2 that, in general, the<br />
inequality m(X ) < C<br />
τ<br />
does not hold. The following remains a logical possibility.<br />
(X )
Sec. 1] A COMPUTER EXPERIMENT 267<br />
1.4. Question. –––– For every ε > 0, does there exist a constant C (ε) such<br />
that, for each quartic threefold,<br />
m(X ) <<br />
C (ε)<br />
τ (X ) 1+ε ?<br />
1.5. Peyre’s constant. –––– Recall from the previous chapter that, in our case,<br />
E. Peyre’s Tamagawa-type number may be defined as an infinite product<br />
τ (X ) := ∏ τ ν (X ) .<br />
ν∈Val(É)<br />
Indeed, for a smooth quartic threefold X, one has O(−K ) = O(1). In particular,<br />
α(X ) = 1.<br />
For a prime number p, the definition of the local factor may be simplified to<br />
(<br />
τ p (X ) := 1 − 1 )<br />
#X (/p<br />
· m)<br />
lim<br />
p m→∞ p 3m<br />
for X the model in P ngiven by the equation defining X.<br />
τ ∞ (X ) is described in full generality in Definition VI.5.22. In the case of<br />
the diagonal quartic threefold X (a 0, ... ,a 4 )<br />
given by a 0 x 4 0 + . . . + a 4 x 4 4 = 0<br />
(a 0 < 0, a 1 , . . . , a 4 > 0) in P 4É, this yields<br />
τ ∞ (X (a 0, ... ,a 4 ) ) = 1<br />
4 4 √<br />
|a0 |<br />
ZZZZ<br />
where the domain of integration is<br />
R<br />
1<br />
(a 1 y 4 + a 2 z 4 + a 3 v 4 + a 4 w 4 ) 3/4 dy dz dv dw (∗)<br />
R := { (y, z, v, w) ∈ [−1, 1] 4 | | a 1 y 4 + a 2 z 4 + a 3 v 4 + a 4 w 4 | ≤ |a 0 | } .<br />
1.6. –––– In the case of diagonal quartic threefolds, there is an estimate for<br />
1<br />
m(X ) in terms of τ (X ). Namely,<br />
τ<br />
admits a fundamental finiteness property.<br />
(X )<br />
More precisely, in Section 3, we will show the following results.<br />
Theorem. Leta = (a 0 , . . . , a 4 )beavectorsuchthata 0 , . . . , a 4 ∈,a 0 < 0, <strong>and</strong><br />
a 1 , . . . , a 4 > 0. Denote by X a the quartic in P 4Égiven by a 0 x 4 0+ . . . +a 4 x 4 4 = 0.<br />
Then, for each ε > 0 there exists a constant C (ε) > 0 such that<br />
1<br />
τ (X a ) ≥ C (ε) · H( 1<br />
a 0<br />
: . . . :<br />
1<br />
a 4<br />
) 1<br />
4<br />
−ε<br />
.<br />
Corollary (Fundamental Finiteness). For each B > 0, there are only finitely many<br />
quartics X a : a 0 x 4 0 + . . . + a 4 x 4 4 = 0 in P 4Ésuch that a 0 < 0, a 1 , . . . , a 4 > 0,<br />
<strong>and</strong> τ (X a ) > B.
268 ON THE SMALLEST POINT ON A DIAGONAL QUARTIC THREEFOLD [Chap. VIII<br />
Corollary (An inefficient search bound). There exists a monotonically decreasing<br />
function S : (0, ∞) → [0, ∞), the search bound, satisfying the following condition.<br />
Let X a be the quartic threefold given by the equation a 0 x 4 0 + . . . + a 4 x 4 4 = 0.<br />
Assume a 0 < 0, a 1 , . . . , a 4 > 0, <strong>and</strong> X a (É) ≠ ∅. Then, X a admits aÉ-<strong>rational</strong><br />
point of height ≤ S(τ (X a )).<br />
Proof. One may simply put S(t) := max min H(P).<br />
τ (X a )≥t P∈X a (É) X a (É)≠∅<br />
□<br />
In other words, we have m(X a ) ≤ S(τ (X a )) as soon as X a (É) ≠ ∅.<br />
2. A negative result<br />
For a ∈Æ, let X (−a) ⊂ P 4Ébe given by ax 4 = y 4 + z 4 + v 4 + w 4 <strong>and</strong> let<br />
m(X (−a) ) := min { H(x : y : z : v : w) | (x : y : z : v : w) ∈ X (−a) (É)}<br />
be the smallest height of aÉ-<strong>rational</strong> point on X (−a) . We compare m(X (−a) )<br />
with the Tamagawa type number τ (−a) := τ (X (−a) ).<br />
2.1. Notation. –––– For a prime number p <strong>and</strong> integers y, z, . . . , not all of<br />
which are equal to zero, we write gcd p<br />
(y, z, . . . ) for the largest power of p<br />
dividing all of the y, z, . . . .<br />
2.2. Theorem. –––– There is no constant C such that<br />
for all a ∈Æ.<br />
m(X (−a) ) < C<br />
τ (−a)<br />
Proof. The proof consists of several steps.<br />
First step. For a ≥ 4, one has τ (−a)<br />
∞ = 1 4√ a<br />
I where I is an integral independent<br />
of a.<br />
This follows immediately from formula (∗) above.<br />
Second step. For the height of the smallest point, we have m(X (−a) ) ≥ 4 √ a<br />
4 .<br />
|x| ≥ 1 yields y 4 + z 4 + v 4 + w 4 ≥ a <strong>and</strong> max{|y|, |z|, |v|, |w|} ≥ 4 √ a<br />
4 .<br />
Third step. There are two positive constants C 1 <strong>and</strong> C 2 such that, for all a ∈Æ,<br />
C 1 < ∏<br />
τ (−a)<br />
p prime<br />
p>13,p∤a<br />
p < C 2 .
Sec. 2] A NEGATIVE RESULT 269<br />
For a prime p of good reduction, Hensel’s lemma shows<br />
(<br />
τ p (X (−a) ) = 1 − 1 )<br />
· #X (−a) (p)<br />
.<br />
p p 3<br />
Further, for the number of <strong>points</strong> on a non-singular variety over a finite<br />
field, there are excellent estimates provided by the Weil conjectures, proven<br />
by P. Deligne. In our situation, [Del, Théorème (8.1)] may be directly applied.<br />
It shows #X (−a) (p) = p 3 + p 2 + p + 1 + E (−a) with an errorterm<br />
|E (−a) | ≤ 60p 3/2 . Note that dim H 3 (X,Ê) = 60 for every smooth quartic<br />
threefold X in P 4[Hi, Satz 2.4].<br />
Consequently,<br />
1 −<br />
60(1 − 1/p)<br />
p 3/2<br />
− 1 p 4 ≤ τ (−a)<br />
p<br />
≤ 1 +<br />
60(1 − 1/p)<br />
p 3/2 − 1 p 4 .<br />
Here, the left h<strong>and</strong> side is positive for p > 13. The infinite product over all<br />
1 − 60(1−1/p) − 1 (respectively 1 + 60(1−1/p) − 1 ) is convergent.<br />
p 3/2 p 4 p 3/2 p 4<br />
Fourth step. There is a sequence (a i ) i∈Æof<br />
i∈Æ<br />
natural numbers such that<br />
( )<br />
∏ τ (−a i)<br />
p<br />
p prime<br />
p≤13 or p|ai<br />
is unbounded.<br />
Let C ∈Êbe given. We will show that<br />
∏<br />
τ p<br />
(−a)<br />
p prime<br />
p≤13 or p|a<br />
> C<br />
when a := p 1 · . . . · p r is a product of sufficiently many different primes<br />
p i ≡ 3 (mod 4) fulfilling a ≡ 1 (mod M ) for M := 16 · 3 · 5 · 7 ·11·13.<br />
Let p be a prime such that p|a. We made sure that p 4 ∤a. Then, for any<br />
point (x : y : z : v : w) ∈ X (−a) (/p n), the assumption p | y, z, v, w<br />
would imply p 4 |ax 4 <strong>and</strong> p|x. Therefore, gcd p<br />
(y, z, v, w) = 1. Further,<br />
y 4 + z 4 + v 4 + w 4 ≡ 0 (mod p).<br />
As p ≡ 3 (mod 4), the number of solutions of that congruence is the same as<br />
that of y 2 + z 2 + v 2 + w 2 ≡ 0 (mod p). Since p remains prime in[i], this<br />
quadratic form is a direct sum of two norm forms. Its number of zeroes in4 p<br />
is therefore equal to 1 + (p − 1)(p + 1) 2 .<br />
gcd p<br />
(y, z, v, w) = 1 implies that Hensel’s lemma is applicable. It shows<br />
#X (−a) (/p n) = pn · (p − 1)(p + 1) 2 p 3(n−1)<br />
(p − 1)p n−1 = p 3n−2 (p + 1) 2 .
270 ON THE SMALLEST POINT ON A DIAGONAL QUARTIC THREEFOLD [Chap. VIII<br />
Hence, τ (−a)<br />
p<br />
= (1 − 1 p ) (p+1)2<br />
p 2<br />
∏<br />
p prime<br />
p≤13 or p|a<br />
τ (−a)<br />
p =<br />
= 1 + 1 − 1 − 1 . We may consequently write<br />
p p 2 p 3<br />
r (<br />
1 + 1 − 1 − 1 )<br />
·<br />
p i pi<br />
2 p<br />
∏τ (−a)<br />
i<br />
3 p .<br />
∏<br />
i=1<br />
The second product is over p = 2 <strong>and</strong> a finite number of primes of good reduction.<br />
The value of the product depends only on the residue of a modulo<br />
M = 16·3·5·7·11·13by virtue of Lemma VII.2.10.a) <strong>and</strong> b.ii). In particular,<br />
our assumption a ≡ 1 (mod M ) implies<br />
T := ∏<br />
τ p<br />
(−a)<br />
p prime<br />
p≤13<br />
is a constant. It is clear that T > 0 as the equation x 4 = y 4 + z 4 + v 4 + w 4<br />
admits a non-zero solution in/mfor any m > 1.<br />
Itremains to show that there exists a set {p 1 , . . . , p r } of primes p i ≡ 3 (mod 4)<br />
such that p i > 13, r (<br />
1 + 1 − 1 − 1 )<br />
≥ C p i pi<br />
2 pi<br />
3 T , (†)<br />
∏<br />
i=1<br />
<strong>and</strong> p 1 · . . . · p r ≡ 1 (mod M ).<br />
1<br />
Condition (†) is easily satisfied as the series ∑ p≡3 (mod 4) diverges. We find a<br />
p<br />
set {p 1 , . . . , p s } of prime numbers p i > 13 of the form p i ≡ 3 (mod 4) such<br />
− 1 p 3 i<br />
p prime<br />
p≤13<br />
) ≥<br />
C<br />
T . Enlarging {p 1, . . . , p s } makes that product<br />
(<br />
that ∏ s i=1 1 +<br />
1<br />
p i<br />
− 1 pi<br />
2<br />
even bigger. We may therefore assume p 1 · . . . · p s ≡ 3 (mod 4).<br />
The numbers M <strong>and</strong> p 1 · . . . · p s are relatively prime. By Dirichlet’s prime number<br />
theorem, there exists a prime p s+1 , larger than each of the p 1 , . . . , p s , such<br />
that p 1 · . . . · p s+1 ≡ 1 (mod M ). This shows, in particular, p s+1 ≡ 3 (mod 4).<br />
The assertion follows.<br />
Conclusion. The four steps together show that m(X (−a) )·τ (−a) is unbounded.<br />
□<br />
3. The fundamental finiteness property<br />
i. An estimate for the factors at the finite places. —<br />
3.1. Notation. –––– i) For a prime number p <strong>and</strong> an integer x ≠ 0, we<br />
put x (p) := p ν p(x) . Note x (p) = 1/‖x‖ p for the normalized p-adic valuation.<br />
ii) By putting ν(x) := min<br />
ξ∈p<br />
to/p r.<br />
x=(ξ mod p r )<br />
ν(ξ), we carry the p-adic valuation fromp over<br />
Note that any 0 ≠ x ∈/p rhas the form x = ε·p ν(x) where ε ∈ (/p r) ∗ is<br />
a unit. Clearly, ε is unique only in the case ν(x) = 0.
Sec. 3] THE FUNDAMENTAL FINITENESS PROPERTY 271<br />
3.2. Definition. –––– For (a 0 , . . . , a 4 ) ∈5 , r ∈Æ, <strong>and</strong> ν 0 , . . . , ν 4 ≤ r, put<br />
S (r)<br />
ν 0 , ... ,ν 4 ;a 0 , ... ,a 4<br />
:= {(x 0 , . . . ,x 4 ) ∈ (/p r) 5 |<br />
ν(x 0 ) = ν 0 , . . . ,ν(x 4 ) = ν 4 ; a 0 x 4 0 + . . . + a 4 x 4 4 = 0 ∈/p r} .<br />
For the particular case ν 0 = . . . = ν 4 = 0, we will write<br />
I.e.,<br />
Z (r)<br />
a 0 , ... ,a 4<br />
:= S (r)<br />
0, ... ,0;a 0 , ... ,a 4<br />
.<br />
Z (r)<br />
a 0 , ... ,a 4<br />
= {(x 0 , . . . , x 4 ) ∈ [(/p r) ∗ ] 5 | a 0 x 4 0 + . . . + a 4 x 4 4 = 0 ∈/p r} .<br />
We will use the notation z (r)<br />
a 0 , ... ,a 4<br />
:= #Z (r)<br />
a 0 , ... ,a 4<br />
.<br />
3.3. Sublemma. –––– If p k |a 0 , . . . , a 4 <strong>and</strong> r > k then we have<br />
z (r)<br />
a 0 , ... ,a 4<br />
= p 5k · z (r−k)<br />
a 0 /p k , ... ,a 4 /p k .<br />
Proof. Since a 0 x 4 0 + . . . + a 4x 4 4 = pk (a 0 /p k · x 4 0 + . . . + a 4/p k · x 4 4 ), there is a<br />
surjection<br />
ι: Z a (r)<br />
0 , ... ,a 4<br />
−→ Z (r−k) ,<br />
a 0 /p k , ... ,a 4 /p k<br />
given by (x 0 , . . . , x 4 ) ↦→ ((x 0 mod p r−k ), . . . , (x 4 mod p r−k )). The kernel of<br />
the homomorphism of modules underlying ι is (p r−k/pr) 5 . □<br />
3.4. Lemma. –––– Assume gcd p<br />
(a 0 , . . . , a 4 ) = p k . Then, there is an estimate<br />
z (r)<br />
a 0 , ... ,a 4<br />
≤ 8p 4r+k .<br />
Proof. Suppose first that k = 0. This means, one of the coefficients is prime<br />
to p. Without restriction, assume p ∤a 0 .<br />
For any (x 1 , . . . , x 4 ) ∈ (/p r) 4 , there appears an equation of the form<br />
a 0 x 4 0 = c. For p odd, it cannot have more than four solutions in (/p r) ∗<br />
as this group is cyclic. On the other h<strong>and</strong>, in the case p = 2, we have<br />
(/2 r) ∗ ∼ =/2 r−2×/2<strong>and</strong> up to eight solutions are possible.<br />
The general case follows directly from Sublemma 3.3. Indeed, if k < r then<br />
z (r)<br />
a 0 , ... ,a 4<br />
= p 5k · z (r−k)<br />
a 0 /p k , ... ,a 4 /p k ≤ p 5k · 8p 4(r−k) = 8p 4r+k .<br />
On the other h<strong>and</strong>, if k ≥ r then the assertion is completely trivial since<br />
z (r)<br />
a 0 , ... ,a 4<br />
= #Z (r)<br />
a 0 , ... ,a 4<br />
< p 5r ≤ p 4r+k < 8p 4r+k .<br />
□
272 ON THE SMALLEST POINT ON A DIAGONAL QUARTIC THREEFOLD [Chap. VIII<br />
3.5. Remark. –––– The proof shows that in the case p ≠ 2 the same inequality<br />
is true with coefficient 4 instead of 8. If p ≡ 3 (mod 4) then one could even<br />
reduce the coefficient to 2. Unfortunately, these observations do not lead to a<br />
substantial improvement of our final result.<br />
3.6. Lemma. –––– Let r ∈Æ<strong>and</strong> ν 0 , . . . , ν 4 ≤ r. Then,<br />
#S ν (r)<br />
0 , ... ,ν 4 ;a 0 , ... ,a 4<br />
= z(r) · ϕ(p r−ν p 4ν 0a 0 , ... ,p 4ν 0<br />
4a ) · . . . · ϕ(p<br />
r−ν 4)<br />
4<br />
.<br />
ϕ(p r ) 5<br />
Proof. As p 4ν 0 a0 x 4 0 + . . . + p4ν 4 a4 x 4 4 = a 0(p ν 0 x0 ) 4 + . . . + a 4 (p ν 4 x4 ) 4 , we have<br />
a surjection<br />
π : Z (r) −→ S (r)<br />
p 4ν 0a 0 , ... ,p 4ν 4a 4<br />
ν 0 , ... ,ν 4 ;a 0 , ... ,a 4<br />
,<br />
given by (x 0 , . . . , x 4 ) ↦→ (p ν 0 x0 , . . . , p ν 4 x4 ).<br />
For i = 0, . . . , 4, consider the mapping ι:/p r→/pr, x 0 ↦→ p ν i<br />
x 0 .<br />
If ν i = r then ι is the zero map. All ϕ(p r ) = (p − 1)p r−1 units are mapped<br />
to zero. Otherwise, observe that ι is p ν i<br />
: 1 on its image. Further, ν(ι(x)) = ν i<br />
if <strong>and</strong> only if x is a unit. By consequence, π is (K (ν 0)<br />
· . . . · K (ν4) ) : 1 when<br />
we put K (ν) := p ν for ν ≠ r <strong>and</strong> K (r) := (p − 1)p r−1 . Summarizing, we could<br />
have written K (ν) := ϕ(p r )/ϕ(p r−ν ). The assertion follows.<br />
□<br />
3.7. Corollary. –––– Let (a 0 , . . . , a 4 ) ∈ (\{0}) 5 . Then, for the local factor<br />
:= τ p (X (a 0, ... ,a 4 ) ), one has<br />
τ (a 0, ... ,a 4 )<br />
p<br />
τ (a 0, ... ,a 4 )<br />
p<br />
r<br />
r→∞<br />
∑<br />
ν 0 , ... ,ν 4 =0<br />
= lim<br />
Proof. Remark VI.5.19 shows that<br />
τ (a 0, ... ,a 4 )<br />
p<br />
Lemma 3.6 yields the assertion.<br />
z (r) · ϕ(p r−ν 0<br />
) · . . . · ϕ(p r−ν 4<br />
)<br />
p 4ν 0a 0 , ... ,p 4ν 4a 4<br />
.<br />
p 4r · ϕ(p r ) 5<br />
r<br />
#S ν<br />
r→∞<br />
∑<br />
(r)<br />
0 , ... ,ν 4 ;a 0 , ... ,a 4<br />
ν 0 , ... ,ν 4 =0<br />
= lim<br />
p 4r .<br />
3.8. Proposition –––– Let (a 0 , . . . , a 4 ) ∈ (\{0}) 5 . Then, for each ε such<br />
that 0 < ε < 1 , one has<br />
4<br />
(<br />
τ (a 0, ... ,a 4 ) 1<br />
)( 1<br />
) 4 (<br />
p ≤ 8<br />
· a<br />
(p)<br />
1 − 1 1 − 1 0<br />
· . . . · a (p) ) 1−ε (<br />
4<br />
3<br />
a (p) ) ε.<br />
4<br />
p 1−4ε p ε<br />
Proof. By Lemma 3.4,<br />
z (r)<br />
p 4ν 0a 0 , ... ,p 4ν 4a 4<br />
/p 4r ≤ 8 gcd p<br />
(p 4ν 0<br />
a 0 , . . . , p 4ν 4<br />
a 4 )<br />
= 8 gcd ( p 4ν 0<br />
a (p)<br />
0 , . . . , p4ν 4<br />
a (p)<br />
4<br />
)<br />
.<br />
□
Sec. 3] THE FUNDAMENTAL FINITENESS PROPERTY 273<br />
(<br />
(p)) Writing k i := ν p (a i ) = ν p a i , we see<br />
z (r)<br />
p 4ν 0a 0 , ... ,p 4ν 4a 4<br />
/p 4r ≤ 8 gcd(p 4ν 0+k 0<br />
, . . . , p 4ν 4+k 4<br />
)<br />
= 8p min{4ν 0+k 0 , ... ,4ν 4 +k 4 } .<br />
We estimate the minimum by a weighted arithmetic mean with weights 1−ε<br />
, <strong>and</strong> ε,<br />
1−ε<br />
4 , 1−ε<br />
, 1−ε<br />
4 4<br />
min{3ν 0 + k 0 , . . . , 3ν 3 + k 3 }<br />
This shows<br />
≤ 1 − ε<br />
3<br />
· (3ν 0 + k 0 ) + 1 − ε · (3ν 1 + k 1 )<br />
3<br />
+ 1 − ε · (3ν 2 + k 2 ) + ε(3ν 3 + k 3 )<br />
3<br />
= (1 − ε)(ν 0 + ν 1 + ν 2 ) + 3εν 3<br />
+ 1 − ε (k 0 + k 1 + k 2 ) + εk 3 .<br />
3<br />
z (r) /p 4r ≤ 8p (1−ε)(ν 0+ ... +ν 3 )+4εν 4 + 1−ε<br />
p 4ν 0a 0 , ... ,p 4ν 4 (k 0+ ... +k 3 )+εk 4<br />
4a 4<br />
(a<br />
= 8p (1−ε)(ν 0+ ... +ν 3 )+4εν4 (p)<br />
·<br />
0<br />
· . . . · a (p) ) 1−ε (<br />
4<br />
3<br />
a (p) ) ε.<br />
4<br />
We may therefore write<br />
τ (a 0, ... ,a 4 )<br />
p<br />
≤ 8 ( a (p)<br />
0<br />
· . . . · a (p) ) 1−ε (<br />
4<br />
3<br />
a (p)<br />
4<br />
r<br />
r→∞<br />
∑<br />
ν 0 , ... ,ν 4 =0<br />
· lim<br />
) ε<br />
p (1−ε)(ν 0+ ... +ν 3 )+4εν 4 · ϕ(p<br />
r−ν 0) · . . . · ϕ(p<br />
r−ν 4)<br />
.<br />
ϕ(p r ) 5<br />
Here, the term under the limit is precisely the product of four copies of the<br />
finite sum r<br />
p<br />
∑<br />
(1−ε)ν · ϕ(p r−ν r−1<br />
) 1<br />
=<br />
ν=0<br />
ϕ(p r )<br />
∑<br />
ν=0<br />
(p ε ) + p 1<br />
ν p − 1 (p ε ) r<br />
<strong>and</strong> one copy of the finite sum<br />
r<br />
p<br />
∑<br />
4εν · ϕ(p r−ν r−1<br />
)<br />
=<br />
ν=0<br />
ϕ(p r )<br />
∑<br />
ν=0<br />
1<br />
(p 1−4ε ) + p<br />
ν<br />
1<br />
p − 1 (p 1−4ε ) . r<br />
For r → ∞, geometric series do appear while the additional summ<strong>and</strong>s tend<br />
to zero.<br />
□<br />
4 ,<br />
3.9. Remark. –––– Unfortunately, the constants<br />
(<br />
C p (ε) 1<br />
)( 1<br />
) 4<br />
:= 8<br />
1 − 1 1 − 1 p 1−4ε p ε
274 ON THE SMALLEST POINT ON A DIAGONAL QUARTIC THREEFOLD [Chap. VIII<br />
have the property that the product ∏ p C p<br />
(ε) diverges. On the other h<strong>and</strong>, we<br />
have at least that C p<br />
(ε) is bounded for p → ∞, say C p (ε) ≤ C (ε) .<br />
3.10. Lemma. –––– Let C > 1 be any constant. Then, for each ε > 0, one has<br />
∏<br />
p prime<br />
p|x<br />
for a suitable constant c (depending on ε).<br />
C ≤ c · x ε<br />
Proof. This follows directly from [Nat, Theorem 7.2] together with [Nat,<br />
Section 7.1, Exercise 7].<br />
□<br />
3.11. Proposition –––– For each ε > 0, there exists a constant c such that<br />
∏ τ (a 0, ... ,a 4 )<br />
p ≤ c · | a 0 · . . . · a 4 | 4 1 · ∏<br />
p prime<br />
for all (a 0 , . . . , a 4 ) ∈ (\{0}) 5 .<br />
p prime<br />
min ‖a i‖ 1 4 −ε<br />
p<br />
i=0, ... ,4<br />
Proof. As already noticed in the third step of the proof of Theorem 2.2, the<br />
product over all primes of good reduction is bounded by consequence of the<br />
Weil conjectures. It, therefore, remains to show that<br />
∏ τ (a 0, ... ,a 4 )<br />
p ≤ c · |a 0 · . . . · a 4 | 4 1 · ∏<br />
p prime<br />
p|2a 0 ... a 4<br />
p prime<br />
For this, we may assume that ε is small, say ε < 1 4 .<br />
Then, by Proposition 3.8, we have at first<br />
min ‖a i‖ 1 4 −ε<br />
i=0, ... ,4<br />
p .<br />
τ (a 0, ... ,a 4 )<br />
p<br />
≤ C (ε)<br />
p<br />
= C (ε)<br />
p<br />
·<br />
·<br />
(a(p)<br />
0<br />
· . . . · a (p) ) 1<br />
4 − ε5<br />
3<br />
· (a (p)<br />
4 ) 4 5 ε<br />
(a(p)<br />
0<br />
· . . . · a (p) ) 1<br />
3 a(p) 4<br />
− ε5<br />
4<br />
· (a (p)<br />
4 )− 1 4 +ε .<br />
Here, the indices 0, . . . , 4 are interchangeable. Hence, it is even allowed to<br />
write<br />
τ (a 0, ... ,a 4 )<br />
p<br />
≤ C (ε)<br />
p<br />
= C (ε)<br />
p<br />
·<br />
·<br />
(a(p)<br />
0<br />
· . . . · a (p)<br />
4<br />
(a(p)<br />
0<br />
· . . . · a (p)<br />
4<br />
) 1<br />
4<br />
− ε5 · (max<br />
) 1<br />
4<br />
− ε5 · min<br />
i<br />
i<br />
)<br />
a (p) −<br />
1<br />
4<br />
+ε<br />
i<br />
‖a i ‖ 1 4 −ε<br />
p .<br />
Now, we multiply over all prime divisors of 2a 0 · . . . · a 4 . Thereby, on the<br />
right h<strong>and</strong> side, we may twice write the product over all primes since the two
Sec. 3] THE FUNDAMENTAL FINITENESS PROPERTY 275<br />
rightmost factors are equal to one for p ∤a 0 · . . . · a 4 , anyway.<br />
∏ τ (a 0, ... ,a 4 )<br />
p<br />
p prime<br />
p|2a 0 ... a 4<br />
≤ ∏ C p<br />
(ε)<br />
p prime<br />
p|2a 0 ... a 4<br />
(<br />
(p)<br />
· ∏ a<br />
0<br />
· . . . · a (p) ) 1<br />
4<br />
− ε 5<br />
4<br />
· ∏<br />
p prime<br />
= ∏ C p (ε) · | a 0 · . . . · a 4 | 4 1 − ε 5 · ∏<br />
p prime<br />
p|2a 0 ... a 4<br />
p prime<br />
p prime<br />
min ‖a 4<br />
i‖ 1 −ε<br />
p<br />
i=0, ... ,4<br />
min ‖a i‖ 1 4 −ε<br />
p<br />
i=0, ... ,4<br />
when we observe that ∏ p a (p) = | a|. Further, we have C p<br />
(ε)<br />
Lemma 3.10,<br />
∏ C (ε) ≤ c · |2a 0 · . . . · a 4 | ε 5 .<br />
p prime<br />
p|2a 0 ... a 4<br />
We finally estimate 2 ε 5 by a constant. The assertion follows.<br />
≤ C (ε) <strong>and</strong>, by<br />
ii. A bound for the factor at the infinite place. — We want to estimate the integral<br />
τ (a 0, ... ,a 4 )<br />
∞ := τ ∞ (X (a 0, ... ,a 4 ) ) described in 1.5.<br />
3.12. Lemma. –––– There exist two constants C 1 <strong>and</strong> C 2 such that<br />
τ (a 0, ... ,a 4 )<br />
∞<br />
⎧<br />
1<br />
(<br />
⎨ 4√ C1 min(|a 0 |, 2a 4 ) 1/4) , if |a 0 | ≤ a 4 ,<br />
|a0 |a<br />
≤<br />
1·...·a 4<br />
(<br />
1<br />
⎩ 4√ C1 min(|a 0 |, 2a 4 ) 1/4 + C 2 a 1/4<br />
|a0 |a<br />
4<br />
log min(|a 0|,3a 1 )) a 1·...·a 4 4<br />
, otherwise,<br />
for all (a 0 , . . . , a 4 ) ∈Ê5 satisfying a 1 ≥ a 2 ≥ a 3 ≥ a 4 ≥ 1 <strong>and</strong> a 0 ≤ −1.<br />
Proof. A linear substitution shows<br />
τ (a 0, ... ,a 4 )<br />
∞ = 1 4 ·<br />
where<br />
ZZZZ<br />
1<br />
√<br />
| a0 |a 1 · . . . · a 4<br />
4<br />
1<br />
dy dz dv dw<br />
(y 4 + z 4 + v 4 + w 4 )<br />
3/4<br />
R (0)<br />
R (0) := {(y, z, v, w) ∈ [−a 1/4<br />
1<br />
, a 1/4<br />
1<br />
] × · · · × [−a 1/4<br />
4<br />
, a 1/4<br />
4<br />
]<br />
| |y 4 + z 4 + v 4 + w 4 | ≤ |a 0 |} .<br />
The integr<strong>and</strong> is non-negative. We cover R (0) ⊆ R 1 ∪ R 2 by two sets as follows,<br />
R 1 := {(y, z, v, w) ∈Ê4 | y 4 + z 4 + v 4 + w 4 ≤ min(| a 0 |, 2a 4 )} ,<br />
R 2 := {(y, z, v, w) ∈Ê4 | a 4 ≤ y 4 + z 4 + v 4 ≤ min(| a 0 |, 3a 1 ) <strong>and</strong><br />
□<br />
w ∈ [−a 1/4<br />
4<br />
, a 1/4<br />
4<br />
]} ,<br />
<strong>and</strong> estimate. In the case | a 0 | ≤ a 4 , the domain of integration is covered by R 1<br />
alone <strong>and</strong> we may omit R 2 , completely.
276 ON THE SMALLEST POINT ON A DIAGONAL QUARTIC THREEFOLD [Chap. VIII<br />
By homogeneity, we have<br />
ZZZZ<br />
1<br />
(y 4 + z 4 + v 4 + w 4 ) dy dz dv dw = ω 3/4 1 ·<br />
R 1<br />
min(| a 0 |,2a 4 ) 1/4<br />
Z<br />
0<br />
1<br />
r 3 · r3 dr<br />
= ω 1 · min(| a 0 |, 2a 4 ) 1/4<br />
where ω 1 is the three-dimensional hypersurface measure of the l 4 -unit hypersphere<br />
S 1 := {(y, z, v, w) ∈Ê4 | y 4 + z 4 + v 4 + w 4 = 1} .<br />
Further,<br />
ZZZZ<br />
1<br />
dy dz dv dw<br />
(y 4 + z 4 + v 4 + w 4 )<br />
3/4<br />
R 2<br />
ZZZ<br />
≤ 2a 1/4 1<br />
4<br />
dy dz dv<br />
(y 4 + z 4 + v 4 )<br />
3/4<br />
R 3<br />
where<br />
R 3 := {(y, z, v) ∈Ê3 | a 4 ≤ y 4 + z 4 + v 4 ≤ min(| a 0 |, 3a 1 )} .<br />
The latter integral may be treated in much the same way as the one above.<br />
We see<br />
ZZZ<br />
1<br />
(y 4 + z 4 + v 4 ) dy dz dv = ω 3/4 2 ·<br />
R 3<br />
min(|a 0 |,3a 1 ) 1/4<br />
Z<br />
a 1/4<br />
4<br />
1<br />
r 3 · r2 dr<br />
where ω 2 is the usual two-dimensional hypersurface measure of the l 4 -unit sphere<br />
Finally,<br />
min(| a 0 |,3a 1 ) 1/4<br />
Z<br />
a 1/4<br />
4<br />
S 2 := {(y, z, v) ∈Ê3 | y 4 + z 4 + v 4 = 1} .<br />
1<br />
r 3 · r2 dr = log min(| a 0|, 3a 1 ) 1/4<br />
a 1/4<br />
4<br />
= 1 4 log min(| a 0|, 3a 1 )<br />
a 4<br />
. □<br />
3.13. Proposition –––– For every ε > 0, there exists a constant C such that<br />
τ (a 0, ... ,a 4 )<br />
∞<br />
≤ C · | a 0 · . . . · a 4 | − 1 4 +ε · min<br />
i=0, ... ,4 ‖a i‖ 1 4<br />
∞<br />
for each (a 0 , . . . , a 4 ) ∈5 satisfying a 0 < 0 <strong>and</strong> a 1 , . . . , a 4 > 0.<br />
Proof. We assume without restriction that a 1 ≥ . . . ≥ a 4 . There are two cases<br />
to be distinguished.
Sec. 3] THE FUNDAMENTAL FINITENESS PROPERTY 277<br />
First case. | a 0 | ≤ a 4 .<br />
Then, by Lemma 3.12, we have<br />
τ (a 0, ... ,a 4 )<br />
∞ ≤ | a 0 · . . . · a 4 | − 1 4 · C1 min{| a 0 |, 2a 4 } 1 4<br />
= C 1 · | a 0 · . . . · a 4 | − 1 4 · | a0 | 1 4<br />
= C 1 · | a 0 · . . . · a 4 | − 1 4 · min<br />
i=0, ... ,4 ‖a i‖ 1 4<br />
∞ .<br />
Second case. | a 0 | > a 4 .<br />
Here, Lemma 3.12 shows<br />
τ (a 0, ... ,a 4 )<br />
∞ ≤ | a 0 · . . . · a 4 | − 1 4<br />
≤ | a 0 · . . . · a 4 | − 1 4<br />
= | a 0 · . . . · a 4 | − 1 4 · | a4 | 1 4<br />
(<br />
C 1 min{|a 0 |, 2a 4 } 1 4 + C2 a 1/4<br />
(<br />
C 1 (2a 4 ) 1 4 + C2 a 1/4<br />
4<br />
log | a 0|<br />
(<br />
C 1 2 1 4 + C2 log | a 0|<br />
a 4<br />
)<br />
4<br />
log min{|a 0|, 3a 1 }<br />
)<br />
a 4<br />
)<br />
a 4<br />
(<br />
= | a 0 · . . . · a 4 | − 4 1 · min ‖a i‖ 1 4<br />
∞ · C 1 2 4 1 + C2 log | a )<br />
0|<br />
i=0, ... ,4 a 4<br />
≤ | a 0 · . . . · a 4 | − 1 4 · min<br />
i=0, ... ,4 ‖a i‖ 1 4<br />
∞ · (C<br />
1 2 1 4 + C2 log | a 0 · . . . · a 4 | )<br />
≤ | a 0 · . . . · a 4 | − 1 4 · min<br />
i=0, ... ,4 ‖a i‖ 1 4<br />
∞ · (C<br />
3 · | a 0 · . . . · a 4 | ) ε<br />
.<br />
□<br />
iii. The Tamagawa number. —<br />
3.14. Proposition –––– For every ε > 0, there exists a constant C > 0 such that<br />
) 1<br />
1 4<br />
a 4<br />
1<br />
τ ≥ C · H( 1<br />
a 0<br />
: . . . :<br />
(a 0, ... ,a 4 )<br />
| a 0 · . . . · a 4 | ε<br />
for each (a 0 , . . . , a 4 ) ∈5 satisfying a 0 < 0 <strong>and</strong> a 1 , . . . , a 4 > 0.<br />
Proof. By Proposition 3.13, we have<br />
τ (a 0, ... ,a 4 )<br />
∞<br />
On the other h<strong>and</strong>, by Theorem 3.11,<br />
≤ C 1 · | a 0 · . . . · a 4 | − 1 4 + ε 2 · min<br />
i=0, ... ,4 ‖a i‖ 1 4<br />
∞ .<br />
∏ τ (a 0, ... ,a 4 )<br />
p ≤ C 2 · | a 0 · . . . · a 4 | 4 1 · ∏<br />
p prime<br />
p prime<br />
min ‖a i‖ 1 4 − ε 2<br />
i=0, ... ,4<br />
p .
278 ON THE SMALLEST POINT ON A DIAGONAL QUARTIC THREEFOLD [Chap. VIII<br />
It follows that<br />
τ (a 0, ... ,a 4 )<br />
≤ C 3·|a 0 · . . . · a 4 | ε 2 · ∏<br />
<strong>and</strong><br />
p prime<br />
1<br />
τ (a 0, ... ,a 4 )<br />
≥ 1 C 3<br />
·<br />
= 1 C 3<br />
·<br />
= 1 C 3<br />
·<br />
[ ]<br />
min ‖a 4<br />
i‖ 1 p · min ‖a i‖ 1 −<br />
ε<br />
4<br />
∞ · ∏ min ‖a 2<br />
i‖ p<br />
i=0, ... ,4 i=0, ... ,4 i=0, ... ,4<br />
[<br />
∏<br />
p prime<br />
] −<br />
1<br />
min ‖a 4<br />
i‖ p ·<br />
i=0, ... ,4<br />
| a 0 · . . . · a 4 | ε 2 · ∏<br />
p prime<br />
∥ 1 ∏ max ∥ 1 ∥∥ 4<br />
i=0, ... ,4<br />
a i<br />
· max<br />
p<br />
p prime<br />
| a 0 · . . . · a 4 | ε 2 · ∏<br />
p prime<br />
H ( 1<br />
a 0<br />
: . . . :<br />
[<br />
| a 0 · . . . · a 4 | ε 2 · ∏<br />
p prime<br />
p prime<br />
[ ] −<br />
1<br />
min ‖a 4<br />
i‖ ∞<br />
i=0, ... ,4<br />
[<br />
min<br />
‖a ] −<br />
ε<br />
i‖ 2 p<br />
i=0, ... ,4 ∥ 1 ∥ 1 ∥∥ 4<br />
a i<br />
∞<br />
i=0, ... ,4<br />
[<br />
max<br />
i=0, ... ,4 a(p) i<br />
) 1<br />
1 4<br />
a 4<br />
It is obvious that max<br />
i=0, ... ,4 a(p) i ≤ | a (p)<br />
0<br />
· . . . · a (p)<br />
4 | <strong>and</strong><br />
This shows<br />
max<br />
i=0, ... ,4 a(p) i<br />
∏ | a (p)<br />
0<br />
· . . . · a (p)<br />
4 | = | a 0 · . . . · a 4 |.<br />
p prime<br />
] ε<br />
2<br />
] ε<br />
2<br />
.<br />
1<br />
τ ≥ 1 H ( ) 1<br />
1 1<br />
a<br />
·<br />
0<br />
: . . . : 4<br />
a 4 (a 0, ... ,a 4 )<br />
C 3 | a 0 · . . . · a 4 | ε 2 · | a 0 · . . . · a 4 | ε 2<br />
= 1 · H( ) 1<br />
1 1<br />
a 0<br />
: . . . : 4<br />
a 4<br />
. □<br />
C 3 | a 0 · . . . · a 4 | ε<br />
3.15. Lemma. –––– Let (a 0 : . . . : a 4 ) ∈ P 4 (É) be any point such that<br />
a 0 ≠ 0, . . . , a 4<br />
∈<br />
≠ 0. Then,<br />
H(a 0 : . . . : a 4 ) ≤ H( 1 1<br />
a 0<br />
: . . . :<br />
a 4<br />
) 4 .<br />
Proof. First, observe that (a 0 : . . . : a 4 ) ↦→ ( )<br />
1<br />
1<br />
a 0<br />
: . . . :<br />
a 4<br />
is a welldefined<br />
map. Hence, we may assume without restriction that a 0 , . . . , a 4<br />
<strong>and</strong> gcd(a 0 , . . . , a 4 ) = 1. This yields H(a 0 : . . . : a 4 ) = max | a i|.<br />
i=0, ... ,4<br />
On the other h<strong>and</strong>, ( 1 1<br />
a 0<br />
: . . . :<br />
a 4<br />
) = (a 1 a 2 a 3 a 4 : . . . : a 0 a 1 a 2 a 3 ). Consequently,<br />
H ( 1<br />
a 0<br />
: . . . :<br />
1<br />
a 4<br />
) ≤ [ max<br />
i=0,...,4 | a i|] 4 = H(a 0 : . . . : a 4 ) 4 .
Sec. 3] THE FUNDAMENTAL FINITENESS PROPERTY 279<br />
From this, the asserted inequality emerges when the roles of a i <strong>and</strong> 1 a i<br />
are interchanged.<br />
□<br />
3.16. Corollary. –––– Let a 0 , . . . , a 4 ∈such that gcd(a 0 , . . . , a 4 ) = 1. Then,<br />
| a 0 · . . . · a 4 | ≤ H ( 1<br />
a 0<br />
: . . . :<br />
1<br />
a 4<br />
) 20.<br />
Proof. Observe that | a 0 · . . . · a 4 | ≤ max | a i| 5 = H(a 0 : . . . : a 4 ) 5 <strong>and</strong> apply<br />
i=0, ... ,4<br />
Lemma 3.15.<br />
□<br />
3.17. Theorem. –––– For each ε > 0, there exists a constant C (ε) > 0 such that,<br />
for all (a 0 , . . . , a 4 ) ∈5 satisfying a 0 < 0 <strong>and</strong> a 1 , . . . , a 4 > 0,<br />
1<br />
τ ≥ C (ε) · H( ) 1<br />
1 1<br />
(a 0, ... ,a 4 ) a 0<br />
: . . . : 4<br />
−ε<br />
a 4<br />
.<br />
Proof. We may assume that gcd(a 0 , . . . , a 4 ) = 1. Then, by Proposition 3.14,<br />
1<br />
τ ≥ C · H( ) 1<br />
1 1<br />
a 0<br />
: . . . : 4<br />
a 4<br />
.<br />
(a 0, ... ,a 4 )<br />
|a 0 · . . . · a 4 | ε 20<br />
(<br />
Corollary 3.16 yields | a 0 · . . . · a 4 | ε 20 ≤ H 1<br />
)<br />
1 ε.<br />
a 0<br />
: . . . :<br />
a 4<br />
3.18. Corollary (Fundamental finiteness). —– For each B > 0, there are only<br />
finitely many quartics X (a 0, ... ,a 4 ) : a 0 x 4 0 + . . . + a 4 x 4 4 = 0 in P 4Ésuch that a 0 < 0,<br />
a 1 , . . . , a 4 > 0, <strong>and</strong> τ (a 0, ... ,a 4) > B.<br />
Proof. This is an immediate consequence of the comparison to the naive height<br />
established in Theorem 3.17.<br />
□<br />
□
CHAPTER IX<br />
GENERAL CUBIC SURFACES ∗<br />
Anyone who considers arithmetical methods of producing r<strong>and</strong>om digits is,<br />
of course, in a state of sin.<br />
JOHN VON NEUMANN (1951)<br />
1. Rational <strong>points</strong> on cubic surfaces<br />
1.1. –––– The arithmetic of cubic surfaces is a fascinating subject. To a large<br />
extent, it was initiated by the work of Yu. I. Manin, particularly by his fundamental<br />
<strong>and</strong> influential book on Cubic Forms [Man].<br />
In this chapter, we study the distribution of <strong>rational</strong> <strong>points</strong> on general cubic<br />
surfaces overÉ. The main problems are<br />
i) Existence ofÉ-<strong>rational</strong> <strong>points</strong>,<br />
ii) Asymptotics ofÉ-<strong>rational</strong> <strong>points</strong>,<br />
iii) The height of the smallest point.<br />
i. Existence of <strong>rational</strong> <strong>points</strong>. —<br />
1.2. –––– Let X be an algebraic variety defined overÉ. Recall that the<br />
Hasse principle is said to hold for X if<br />
X (É) = ∅ ⇐⇒ ∃ ν ∈ Val(É): X (Éν) = ∅ .<br />
For quadrics in P nÉ, the Hasse principle holds by the famous Theorem of Hasse-<br />
Minkowski.<br />
It is, however, well-known that for smooth cubic surfaces overÉthe Hasse principle<br />
does not hold, in general. In all known examples, this is explained by the<br />
Brauer-Manin obstruction which was discussed in detail in Chapter III.<br />
(∗) This chapter is a revised version of the article: Experiments with general cubic surfaces, to<br />
appear in: The Manin Festschrift, joint with A.-S. Elsenhans.
282 GENERAL CUBIC SURFACES [Chap. IX<br />
ii. Asymptotics of <strong>rational</strong> <strong>points</strong>. —<br />
1.3. –––– On the asymptotics of <strong>rational</strong> <strong>points</strong> of bounded height, there is<br />
Manin’s conjecture which was described in Chapter VI.<br />
1.4. Conjecture (Manin). —– Let X be an arbitrary Fano variety overÉ<strong>and</strong> H<br />
be an anticanonical height on X. Then, there exist a dense, Zariski open subset<br />
X ◦ ⊆ X <strong>and</strong> a constants r <strong>and</strong> τ such that<br />
#{x ∈ X ◦ (É) | H(x) < B} ∼ τB log r B<br />
(∗)<br />
for B → ∞.<br />
In the case that X is a surface, it is expected that r = rkPic(X ) − 1.<br />
1.5. Remark (Known cases). —– Conjecture 1.4 is established for smooth<br />
complete intersections of multidegree d 1 , . . . , d n in the case that the dimension<br />
of X is very large compared to d 1 , . . . , d n [Bir]. Further, it is proven for projective<br />
spaces <strong>and</strong> quadrics. Finally, there are a number of further special cases<br />
in which Manin’s conjecture is known to be true. See, e.g., [Pe02, Sec. 4].<br />
In Chapter VII, numerical evidence for Conjecture 1.4 is presented in the case<br />
of the threefolds X e a,b given by axe = by e + z e + v e + w e in P 4Éfor e = 3 <strong>and</strong> 4.<br />
1.6. Remark (Peyre’s constant). —– Motivated by results obtained by the classical<br />
circle method, E. Peyre refined Manin’s conjecture by a conjectural value for<br />
the leading coefficient τ. We formulated a general description of Peyre’s constant<br />
in Definition VI.5.35.<br />
1.7. –––– We are interested in smooth cubic surfaces. Let us explain<br />
Peyre’s constant when X is the smooth surface in P 3Édefined by a homogeneous<br />
cubic polynomial f ∈[X 0 , X 1 , X 2 , X 3 ].<br />
Consider the anticanonical height which is the same as the naive height. Assume<br />
that rkPic(X ) = 1 <strong>and</strong> H 1( Gal(É/É), Pic(XÉ) ) = 0.<br />
Then, Peyre’s constant is equal to the Tamagawa type number τ given by<br />
where<br />
τ p =<br />
τ := ∏ τ p<br />
p∈È∪{∞}<br />
(<br />
1 − 1 )<br />
#X (/p m)<br />
· lim<br />
p m→∞ p 2m
Sec. 1] RATIONAL POINTS ON CUBIC SURFACES 283<br />
for p finite <strong>and</strong><br />
τ ∞ = 1 2<br />
Z<br />
x∈[−1,1] 4<br />
f (x)=0<br />
1<br />
dS .<br />
‖(grad f )(x)‖ 2<br />
Here, X ⊂ P 3is the integral model of X defined by the polynomial f .<br />
dS denotes the usual hypersurface measure on the cone CX (Ê), considered as a<br />
hypersurface inÊ4 .<br />
1.8. Remarks. –––– i) The assumption that rk Pic(X ) = 1 guarantees that there<br />
are no log-factors in the conjectural asymptotics.<br />
ii) According to Example VI.5.3, the assumption rkPic(X ) = 1 is sufficient<br />
for α(X ) = 1. On the other h<strong>and</strong>, H 1( Gal(É/É), Pic(XÉ) ) = 0 guarantees<br />
that there is no Brauer-Manin obstruction to weak approximation. Further,<br />
β(X ) = 1 <strong>and</strong> both additional factors do not need to be considered.<br />
In general, the definition of Peyre’s constant is more complicated, even for<br />
smooth cubic surfaces.<br />
1.9. Remark. –––– According to Definition VI.5.35, the local factors τ p<br />
are equipped with an additional factor det ( 1 − p −1 Frob p | p) I .<br />
Here, Pic(XÉ) I p<br />
denotes the fixed module under the inertia group. For compensation,<br />
another factor is added to Peyre’s constant, the value L(1, χ P ) of the<br />
Artin-L-function associated to the Gal(É/É)-representation P ⊂ Pic(XÉ)⊗<br />
complementary to the trivial summ<strong>and</strong>.<br />
Inour case, therepresentationP turnsouttobeirreducibleofranksix. Thefinite<br />
quotient of Gal(É/É) acting on P is isomorphic to W (E 6 ) of order 51 840.<br />
We do not even make an attempt to compute the exact value of L. Instead, we<br />
work with the conditionally convergent series described.<br />
iii. The smallest point. —<br />
1.10. –––– It would be desirable to have an a-priori upper bound for the height<br />
of the smallestÉ-<strong>rational</strong> point on X as this would allow to effectively decide<br />
whether X (É) ≠ ∅ or not.<br />
When X is a conic, Legendre’s theorem on zeroes of ternary quadratic forms<br />
yields an effective bound for the smallest point. For quadrics of arbitrary dimension,<br />
the same is true by an observation due to J. W. S. Cassels [Cas55].<br />
Further, there is a theorem of C. L. Siegel [Si73, Satz 1] which provides a generalization<br />
to hypersurfaces defined by norm equations. This certainly includes<br />
some special cubic surfaces but, in general, no theoretical upper bound is known<br />
for the height of the smallestÉ-<strong>rational</strong> point on a cubic surface.
284 GENERAL CUBIC SURFACES [Chap. IX<br />
1.11. Remark. –––– If one had an error term [S-D05] for (∗) uniform over all<br />
cubic surfaces X of Picard rank 1 then this would imply that the height m(X )<br />
of the smallestÉ-<strong>rational</strong> point is always less than<br />
C<br />
τ<br />
for certain constants<br />
(X ) α<br />
α > 1 <strong>and</strong> C > 0.<br />
The investigations on quartic threefolds made in Chapter VIII (cf. [EJ5]) indicate<br />
that one might have even m(X ) < C (ε) for any ε > 0. The same inequality<br />
τ (X ) 1+ε<br />
is suggested by the experiments with diagonal cubic surfaces which we will<br />
describe in the next chapter.<br />
Assuming equidistribution, one would expect that the height of the smallestÉ-<strong>rational</strong><br />
point on X should be even ∼ 1<br />
τ . An inequality of the<br />
(X )<br />
form m(X ) < C<br />
τ<br />
is, however, known to be wrong for diagonal quartic threefolds.<br />
(Cf. Theorem VIII.2.2.) Under the Generalized Riemann Hypothesis,<br />
(X )<br />
Theorem X.5.3 will show that m(X ) < C<br />
τ<br />
is wrong for diagonal cubic surfaces,<br />
(X )<br />
too.<br />
iv. The results. —<br />
1.12. –––– We consider two families of cubic surfaces which are produced by<br />
a r<strong>and</strong>om number generator. For each of these surfaces, we do the following.<br />
i) We verify that the Galois group acting on the 27 lines is equal to W (E 6 ).<br />
This implies that rk Pic(X ) = 1 <strong>and</strong> H 1( Gal(É/É), Pic(XÉ) ) = 0.<br />
ii) We compute E. Peyre’s constant τ (X ).<br />
iii) Up to a certain bound for the anticanonical height, we count allÉ-<strong>rational</strong><br />
<strong>points</strong> on the surface X.<br />
Thereby, we establish the Hasse principle for each of the surfaces considered.<br />
Further, we test numerically the conjecture of Manin, in the refined form due to<br />
E. Peyre, on the asymptotics of <strong>points</strong> of bounded height. Finally, we study the<br />
behaviour of the height of the smallestÉ-<strong>rational</strong> point versus E. Peyre’s constant.<br />
This means, we test the estimates formulated in Remark 1.11.<br />
2. Background<br />
i. The 27 lines. —<br />
2.1. –––– A non-singular cubic surface defined overÉcontains exactly 27 lines.<br />
The symmetries of the configuration of the 27 lines respecting the intersection<br />
pairing are given by the Weyl group W (E 6 ). The order of the group #W (E 6 )<br />
is 51 840.
Sec. 2] BACKGROUND 285<br />
These are st<strong>and</strong>ard facts which may be found in many places in the literature.<br />
We discussed them before in Chapter II, 8.21 ff. Observe the list of the 350 conjugacy<br />
classes of W (E 6 ) generated using GAP which is given in the appendix.<br />
2.2. Remark. –––– The group W (E 6 ) admits the following properties.<br />
i) W (E 6 ) has 3 4·5 = 405 2-Sylow sub<strong>groups</strong>.<br />
ii) W (E 6 ) has 2 5·5 = 160 3-Sylow sub<strong>groups</strong>.<br />
iii) W (E 6 ) has 2 4·3 4 = 1296 5-Sylow sub<strong>groups</strong>.<br />
2.3. Fact. –––– LetX beasmoothcubic surface definedoverÉ<strong>and</strong> let K bethe field<br />
of definition of the 27 lines on X. Then K is a Galois extension ofÉ. The Galois<br />
group Gal(K/É) is a subgroup of W (E 6 ).<br />
Proof. This was shown in Chapter II, 8.21.<br />
□<br />
2.4. Remark. –––– W (E 6 ) contains a subgroup U of index two which is<br />
isomorphic to the simple group of order 25 920. It is of Lie type B 2 (3),<br />
i.e. U ∼ = Ω 5 (3) ⊂ SO 5 (3). In U, the stabilizer of two non-intersecting lines<br />
is the alternating group A 5 .<br />
2.5. Remark. –––– On the other h<strong>and</strong>, the operation of W (E 6 ) on the 27 lines<br />
gives rise to a transitive permutation representation ι: W (E 6 ) → S 27 . It turns<br />
out that the image of ι is contained in the alternating group A 27 . We will call<br />
an element σ ∈ W (E 6 ) even if σ ∈ U <strong>and</strong> odd, otherwise. This should not be<br />
confused with the sign of ι(σ) ∈ S 27 which is always even.<br />
2.6. Remark. –––– The operation of the group W (E 6 ) on the 27 lines satisfies<br />
the properties below.<br />
i) It is transitive.<br />
ii) Fix a line l 1 . Then, the remaining 26 lines split into two classes. There are<br />
10 lines which intersect l 1 <strong>and</strong> 16 lines which do not. The stabilizer of l 1 acts<br />
transitively on each class.<br />
iii) Choose a second line l 2 which does not intersect l 1 . Then, there are precisely<br />
five lines g 1 , . . . , g 5 intersecting both l 1 <strong>and</strong> l 2 . The stabilizer of l 1<br />
<strong>and</strong> l 2 is isomorphic to S 5 . An isomorphism is given by the operation on the<br />
lines g 1 , . . . , g 5 .<br />
In particular, the group order is indeed 27·16·5! = 51 840.<br />
ii. The Brauer-Manin obstruction. —<br />
2.7. –––– For Fanovarieties, allknownobstructionsagainsttheHasseprinciple<br />
are explained by the following observation.
286 GENERAL CUBIC SURFACES [Chap. IX<br />
2.8. Observation (Manin). —– Let π : X → SpecÉbe a non-singular variety<br />
overÉ. Choose an element α ∈ Br(X ) from its Brauer group.<br />
Then, anyÉ-<strong>rational</strong> point x ∈ X (É) gives rise to an adelic point (x ν ) ν ∈ X (AÉ)<br />
satisfying the condition<br />
ev(α, x) :=<br />
∑ inv(α| xν ) = 0 .<br />
ν∈Val(É)<br />
Here, inv: Br(Éν) →É/(respectively inv: Br(Ê) → 1 2/) denotes the canonical<br />
isomorphism.<br />
Proof. This is the Brauer-Manin obstruction which was studied in detail in Chapter<br />
III. The assertion was proven in Proposition III.2.3.b.iv).<br />
□<br />
2.9. –––– Recall the following facts.<br />
i) Proposition III.2.3.a.ii) shows that inv(α| xν ) depends continuously<br />
on x ν ∈ X (Éν). Further, by Proposition III.2.3.b.ii), for every proper variety<br />
X overÉ<strong>and</strong> each α ∈ Br(X ), there exists a finite set S ⊂ Val(É) such that<br />
inv(α| xν ) = 0 for every ν ∉ S <strong>and</strong> x ν ∈ X (Éν).<br />
This implies that the Brauer-Manin obstruction, if present, is an obstruction to<br />
the principle of weak approximation.<br />
ii) For certain varieties X, it happens, however, that the Brauer-Manin obstruction<br />
is not an obstruction to the Hasse principle. The point is that it possibly<br />
excludes only part of the adelic <strong>points</strong>.<br />
This is true, e.g., for the cubic surface given in Example III.5.25. Likewise for<br />
all diagonal cubic surfaces of the “first case” according to the classification<br />
of J.-L. Colliot-Thélène, D. Kanevsky, <strong>and</strong> J.-J. Sansuc [CT/S]. See Theorem<br />
III.6.4.<br />
iii) It is obvious that altering α ∈ Br(X ) by some Brauer class π ∗ ρ for ρ ∈ Br(É)<br />
does not change the obstruction defined by α. By consequence, it is only the<br />
factor group Br(X )/π ∗ Br(É) which is relevant for the Brauer-Manin obstruction.<br />
As was shown in Proposition II.8.11 <strong>and</strong> Remark II.8.13, the latter<br />
is canonically isomorphic to H 1( Gal(É/É), Pic(XÉ) ) . In particular,<br />
if H 1( Gal(É/É), Pic(XÉ) ) = 0 then there is no Brauer-Manin obstruction<br />
on X.<br />
2.10. –––– For a smooth cubic surface X, the geometric Picard group Pic(XÉ)<br />
is generated by the classes of the 27 lines on XÉ. Its first cohomology group<br />
can be described in terms of the Galois action on these lines. Indeed, by<br />
Proposition II.8.18, there is a canonical isomorphism<br />
H 1( Gal(É/É), Pic(XÉ) ) ∼ = Hom<br />
(<br />
(NF ∩ F0 )/NF 0 ,É/) . (†)
Sec. 3] THE GALOIS OPERATION ON THE 27 LINES 287<br />
Here, F ⊂ Div(XÉ) is the group generated by the 27 lines, F 0 ⊂ F denotes<br />
the subgroup of principal divisors, <strong>and</strong> N is the norm map under the operation<br />
of Gal(É/É)/H , H being the stabilizer of F.<br />
2.11. Remark. –––– Consider the particular case when the Galois group acts<br />
transitively on the 27 lines. Then, (†) shows that H 1( Gal(É/É), Pic(XÉ) ) = 0.<br />
In particular, there is no Brauer-Manin obstruction in this case.<br />
It is expected that the Hasse principle holds for all cubic surfaces such that<br />
H 1( Gal(É/É), Pic(XÉ) ) = 0. A statement which is very close to the<br />
Hasse principle was conjectured by J.-L. Colliot-Thélène <strong>and</strong> J.-J. Sansuc [CT/S,<br />
Conjecture C].<br />
3. The Galois operation on the 27 lines<br />
Let X be a smooth cubic surface defined overÉ<strong>and</strong> let K be the field of<br />
definition of the 27 lines on X. By Fact 2.3, K/Éis a Galois extension <strong>and</strong><br />
the Galois group G := Gal(K/É) is a subgroup of W (E 6 ). For general cubic<br />
surfaces, G is actually equal to W (E 6 ). To verify this for particular examples,<br />
the following lemma is useful.<br />
3.1. Lemma. –––– Let H ⊆ W (E 6 ) be a subgroup which acts transitively on the<br />
27 lines <strong>and</strong> contains an element of order five. Then, either H is the subgroup<br />
U ⊂ W (E 6 ) of index two or H = W (E 6 ).<br />
Proof. H ∩ U still acts transitively on the 27 lines <strong>and</strong> still contains an element<br />
of order five. Thus, we may suppose H ⊆ U.<br />
Assume that H U. Denote by k the index of H in U. The natural action<br />
of U on the set of cosets U/H yields a permutation representation i : U → S k .<br />
As U is simple, i is necessarily injective. In particular, since #U ∤ 8!, we see<br />
that k > 8. Let us consider the stabilizer H ′ ⊂ H of one of the lines. As H acts<br />
transitively, it follows that #H ′ = #H = #U . We distinguish two cases.<br />
27<br />
= 960<br />
27·k k<br />
First case: k > 16. Then, k ≥ 20 <strong>and</strong> #H ′ ≤ 48. This implies that the<br />
5-Sylow subgroup is normal in H ′ . Its conjugate by some σ ∈ H therefore depends<br />
only on σ ∈ H /H ′ . By consequence, the number n of 5-Sylow sub<strong>groups</strong><br />
in H is a divisor of #H /#H ′ = 27. Sylow’s congruence n ≡ 1 (mod 5) yields<br />
that n = 1.<br />
Let H 5 ⊂ H be the 5-Sylow subgroup. Then, ι(H 5 ) ⊂ S 27 is generated by a<br />
product of disjoint 5-cycles leaving at least two lines fixed. It is, therefore, not<br />
normal in the transitive group ι(H ). This is a contradiction.<br />
Second case: 9 ≤ k ≤ 16. We have k | 960. On the other h<strong>and</strong>, the assumption<br />
5 | #H implies 5 ∤k. This shows, there are only two possibilities, k = 12
288 GENERAL CUBIC SURFACES [Chap. IX<br />
<strong>and</strong> k = 16. As, in U, there is no subgroup of index eight or less, H ⊂ U<br />
must be a maximal subgroup. In particular, the permutation representation<br />
i : U → S k is primitive.<br />
Primitive permutation representations of degree up to 20 have been classified<br />
already in the late 19th century. It is well known that no group of order 25 920<br />
allows a faithful primitive permutation representation of degree 12 or 16 [Sim,<br />
Table 1].<br />
□<br />
3.2. Remark. –––– The sub<strong>groups</strong> of the simple group U have been completely<br />
classified by L. E. Dickson [Di04] in 1904. There are 114 conjugacy classes of<br />
proper sub<strong>groups</strong> H ≠ {e} in U. It would not be complicated to deduce the<br />
lemma from Dickson’s list.<br />
3.3. –––– Let the smooth cubic surface X be given by a homogeneous equation<br />
f = 0 with integral coefficients. We want to compute the Galois group G.<br />
An affine part of a general line l can be described by four coefficients a, b, c, d<br />
via the parametrization<br />
l: t ↦→ (1 : t : (a + bt) : (c + dt)) .<br />
l is contained in S if <strong>and</strong> only if it intersects S in at least four <strong>points</strong>. This implies<br />
that<br />
f (l(0)) = f (l(∞)) = f (l(1)) = f (l(−1)) = 0<br />
is a system of equations for a, b, c, d which encodes that l is contained in S.<br />
By a Gröbner base calculation in SINGULAR, we compute a univariate polynomial<br />
g of minimal degree belonging to the ideal generated by the equations. If g<br />
is of degree 27 then the splitting field of g is equal to the field K of definition<br />
of the 27 lines on X. We then use van der Waerden’s criterion [Poh/Z, Proposition<br />
2.9.35]. This means, we factor g modulo various primes. If all irreducible<br />
factors are distinct then the degrees of the factors describe the cycle structure of<br />
an element in the Galois group of g.<br />
If the various factorizations are incompatible [Coh, Section 3.5.2] then we get<br />
the information that g is irreducible. The Galois group G then acts transitively<br />
on the 27 lines. Furthermore, if we find an element of order five <strong>and</strong> an element<br />
which is not contained in the index two subgroup then G is actually equal<br />
to W (E 6 ). More precisely, our algorithm works as follows.<br />
3.4. Algorithm (Verifying G = W (E 6 )). —– Given the equation f = 0 of a<br />
smooth cubic surface, this algorithm verifies G = W (E 6 ).
Sec. 3] THE GALOIS OPERATION ON THE 27 LINES 289<br />
i) Compute a univariate polynomial 0 ≠ g ∈[d] of minimal degree such that<br />
g ∈ ( f (l(0)), f (l(∞)), f (l(1)), f (l(−1))) ⊂É[a, b, c, d]<br />
where l: t ↦→ (1 : t : (a + bt) : (c + dt)).<br />
If g is not of degree 27 then terminate with an error message. In this case, the<br />
coordinate system for the lines is not sufficiently general. If we are erroneously<br />
given a singular cubic surface then the algorithm will fail at this point.<br />
ii) Factor g modulo all primes below a given limit. Ignore the primes dividing<br />
the leading coefficient of g.<br />
iii) If one of the factors is multiple then go to the next prime immediately.<br />
Otherwise, check whether the decomposition type corresponds to one of the<br />
cases listed below,<br />
A := {(9, 9, 9)} , B := {(1, 1, 5, 5, 5, 5, 5), (2, 5, 5, 5, 10)} ,<br />
C := {(1, 4, 4, 6, 12), (2, 5, 5, 5, 10), (1, 2, 8, 8, 8)} .<br />
iv) If each of the cases occurred for at least one of the primes then output the<br />
message “The Galois group is equal to W (E 6 ).” <strong>and</strong> terminate.<br />
Otherwise, output “Can not prove that the Galois group is equal to W (E 6 ).”<br />
3.5. Remark. –––– The cases above are functioning as follows.<br />
a) Case B shows that the order of the Galois group is divisible by five.<br />
b) Cases A <strong>and</strong> B together guarantee that g is irreducible. Therefore,<br />
by Lemma 3.1, A <strong>and</strong> B prove that G contains the index two subgroup<br />
U ⊂ W (E 6 ).<br />
c) Case C is a selection of the most frequent odd conjugacy classes in W (E 6 ).<br />
3.6. Remark. –––– One could replace cases B <strong>and</strong> C by their common element<br />
(2, 5, 5, 5, 10). This would lead to a simpler but less efficient algorithm.<br />
3.7. Remark. –––– Actually, a decomposition type as considered in step iii)<br />
does not always represent a single conjugacy class in W (E 6 ). Two elements<br />
ι(σ), ι(σ ′ ) ∈ S 27 might be conjugate in S 27 via a permutation τ ∉ ι(W (E 6 )).<br />
For example, as is easily seen using GAP, the decomposition type (3, 6, 6, 6, 6)<br />
falls into three conjugacy classes two of which are even <strong>and</strong> one is odd (cf. Remark<br />
2.4). However, all the decomposition types searched for in Algorithm 3.4<br />
do represent single conjugacy classes.
290 GENERAL CUBIC SURFACES [Chap. IX<br />
3.8. Remark. –––– Since we expect G = W (E 6 ), we can estimate the probability<br />
of each case by the Chebotarev density theorem. Case A has a probability<br />
of 1 . This is the lowest value among the three cases.<br />
9<br />
3.9. Remark. –––– As we do not use the factors of g explicitly, it is enough<br />
to compute their degrees <strong>and</strong> to check that each of them occurs with multiplicity<br />
one. This means, we only have to compute gcd(g(X ), g ′ (X ))<br />
<strong>and</strong> gcd(g(X ), X pd −X ) inp[X ] for d = 1, 2, . . . , 13 [Coh, Algorithms 3.4.2<br />
<strong>and</strong> 3.4.3].<br />
4. Computation of Peyre’s constant<br />
i. The Euler product. —<br />
4.1. –––– We want to compute the product over all τ p . For a finite place p,<br />
we have<br />
(<br />
τ p = 1 − 1 )<br />
X (/p m)<br />
· lim .<br />
p m→∞ p 2m<br />
If the reduction is smooth then the sequence under the limit is constant<br />
Xp<br />
by virtue of Hensel’s Lemma. Otherwise, it becomes stationary after finitely<br />
many steps.<br />
We approximate the infinite product over all τ p by the finite product taken over<br />
the primes less than 300. Numerical experiments show that the contributions of<br />
larger primes do not lead to a significant change. (Compare the values calculated<br />
for the concrete example in Section 6.)<br />
ii. The factor at the infinite place. —<br />
4.2. –––– We want to compute<br />
τ ∞ = 1 2<br />
where the domain of integration is given by<br />
Z<br />
R<br />
1<br />
‖ grad f ‖ 2<br />
dS<br />
R = {(x, y, z, w) ∈ [−1, 1] 4 | f (x, y, z, w) = 0} .<br />
Here, dS denotes the usual hypersurface measure on R, considered as a hypersurface<br />
inÊ4 . Thus, τ ∞ is given by a three-dimensional integral.<br />
Since f is a homogeneous polynomial, we may reduce to an integral over the<br />
boundary of R which is a two-dimensional domain. In our particular case, we
Sec. 4] COMPUTATION OF PEYRE’S CONSTANT 291<br />
have deg f = 3. Then, a direct computation leads to<br />
Z<br />
Z<br />
τ ∞ =<br />
dA +<br />
‖( ∂ f<br />
‖( ∂ f<br />
Z<br />
+<br />
1<br />
, ∂ f , ∂ f )‖ R 0 ∂y ∂z ∂w 2<br />
1<br />
‖( ∂ f , ∂ f , ∂ f )‖ R 2 ∂x ∂y ∂w 2<br />
where the domains of integration are<br />
Z<br />
dA +<br />
1<br />
, ∂ f , ∂ f )‖ R 1 ∂x ∂z ∂w 2<br />
1<br />
‖( ∂ f , ∂ f , ∂ f )‖ R 3 ∂x ∂y ∂z 2<br />
R i = {(x 0 , x 1 , x 2 , x 3 ) ∈ [−1, 1] 4 | x i = 1 <strong>and</strong> f (x 0 , x 1 , x 2 , x 3 ) = 0} .<br />
dA denotes the two-dimensional hypersurface measure on R i , considered as a<br />
hypersurface inÊ3 .<br />
We therefore have to integrate a smooth function over a compact part of a<br />
smooth two-dimensional submanifold inÊ3 . To do this, we approximate the<br />
domain of integration by a triangular mesh.<br />
4.3. Algorithm (Generating a triangular mesh). —–<br />
Given the equation f = 0 of a smooth surface inÊ3 , this algorithm constructs<br />
a triangular mesh approximating the part of the surface which is contained in a<br />
given cube.<br />
i) We split the cube into eight smaller cubes <strong>and</strong> iterate this procedure a predefined<br />
number of times, recursively. During recursion, we exclude those<br />
cubes which obviously do not intersect the manifold. To do this, we estimate<br />
‖grad f ‖ 2 on each cube.<br />
ii) Then, each resulting cube is split into six simplices.<br />
iii) For each edge of each simplex which intersects the manifold, we compute<br />
an approximation of the point of intersection. We use them as the vertices of<br />
the triangles to be constructed. This leads to a mesh consisting of one or two<br />
triangles per simplex.<br />
4.4. Remark. –––– Similar algorithms are popular in the computer graphics<br />
community.<br />
4.5. –––– The next step is to compute the contribution of each triangle ∆ i to<br />
the integral. For this, we use some adaption of the midpoint rule.<br />
We approximate the integr<strong>and</strong> g by its value g(C i ) at the barycenter C i of<br />
the triangle. Note that this point usually lies outside the surface given by f = 0.<br />
Algorithm 4.3 guarantees only that the three vertices of each facet are contained<br />
in that surface. We map it to the manifold along the flow of the gradient<br />
field grad f .<br />
dA<br />
dA
292 GENERAL CUBIC SURFACES [Chap. IX<br />
The product g(C i )A(∆ i ) seems to be a reasonable approximation of the contribution<br />
of ∆ i to the integral. We correct by an additional factor, the cosine of the<br />
angle between the normal vector of the triangle <strong>and</strong> the gradient vector grad f<br />
at the barycenter C. We calculated the contribution of each triangle by some<br />
adaption of the midpoint rule. The problem is our algorithm makes sure for<br />
each facet that its three vertices are contained in the manifold R i . This means<br />
that its barycenter is usually outside R i . We map it to the manifold along the<br />
flow of the gradient field grad f .<br />
4.6. Remark. –––– In our application, these correctional factors are close to 1<br />
<strong>and</strong> seem to converge versus 1 when the number of recursions is growing.<br />
This is, however, not a priori clear. H. A. Schwarz’s cylindrical surface [Schw]<br />
constitutes a famous example of a sequence of triangulations where the triangles<br />
become arbitrarily small <strong>and</strong> the factors are nevertheless necessary for<br />
correct integration.<br />
We use the method described above to approximate the value of τ ∞ . In Algorithm<br />
4.3, we work with six recursions.<br />
4.7. Remark. –––– Our method of numerical integration is a combination of<br />
st<strong>and</strong>ard algorithms for 2.5-dimensional mesh generation <strong>and</strong> two dimensional<br />
integration which are described in the literature [Hb].<br />
On a triangle, we integrate linear functions correctly. This indicates that the<br />
method should converge of second order. The facts that we work with the<br />
area of a linearized triangle <strong>and</strong> that the barycenters C i are located in a certain<br />
distance from the manifold generate errors of the same order.<br />
5. Numerical Data<br />
i. The computations carried out. —<br />
5.1. –––– A general cubic surface is described by twenty coefficients. With current<br />
technology, it is impossible to study all cubic surfaces with coefficients<br />
below a reasonable bound. For that reason, we decided to work with coefficient<br />
vectors provided by a r<strong>and</strong>om number generator. Our first sample consists<br />
of 20 000 surfaces with coefficients r<strong>and</strong>omly chosen in the interval [0, 50].<br />
The second sample consists of 20 000 surfaces with r<strong>and</strong>omly chosen coefficients<br />
from the interval [−100, 100].<br />
These limits were, of course, chosen somewhat arbitrarily. There is, at least,<br />
some reason not to work with too large limits as this would lead to low values
Sec. 5] NUMERICAL DATA 293<br />
of τ. (The reader might want to compare Theorem X.6.21 where this is rigorously<br />
proven in a different situation.) Low values of τ are undesirable as they<br />
require high search bounds in order to satisfactorily test Manin’s conjecture.<br />
We verified explicitly that each of the surfaces studied is smooth. For this,<br />
we inspected a Gröbner base of the ideal corresponding to the singular locus.<br />
The computations were done in SINGULAR.<br />
Then, using Algorithm 3.4, <strong>and</strong> a prime limit of 1000, we proved that, for each<br />
surface, the full Galois group W (E 6 ) acts on the 27 lines. The largest prime used<br />
was 457. This means that all our examples are general from the Galois point<br />
of view. By consequence, their Picard ranks are equal to 1. Further, according<br />
to Remark 2.11, the Brauer-Manin obstruction is not present on any of the<br />
surfaces considered.<br />
Almost as a byproduct, we verified that no two of the 40 000 surfaces are isomorphic.<br />
Actually, when running part ii) of Algorithm 3.4, we wrote the<br />
decomposition types found into a file. Primes at which the algorithm failed<br />
were labeled by a special marker. A program, written in C, ran in an iterated<br />
loop over all pairs of surfaces <strong>and</strong> looked for a prime at which the decomposition<br />
types differ. The largest prime needed to distinguish two surfaces was 73.<br />
We counted allÉ-<strong>rational</strong> <strong>points</strong> of height less than 250 on the surfaces of the<br />
first sample. It turns out that, on two of these surfaces, there are noÉ-<strong>rational</strong><br />
<strong>points</strong> occurring as the equation is unsolvable inÉp for some small p. In this<br />
situation, Manin’s conjecture is true, trivially.<br />
On each of the remaining surfaces, we found at least oneÉ-<strong>rational</strong> point.<br />
228 examples contained less than ten <strong>points</strong>. On the other h<strong>and</strong>, 1213 examples<br />
contained at least one hundredÉ-<strong>rational</strong> <strong>points</strong>. The largest number of <strong>points</strong><br />
found was 335.<br />
For the second sample, the search bound was 500. Again, on two of these<br />
surfaces, there are noÉ-<strong>rational</strong> <strong>points</strong> occurring as the equation is unsolvable in<br />
a certainÉp. There were 202 examples containing between one <strong>and</strong> nine <strong>points</strong>.<br />
1857 examples contained at least one hundredÉ-<strong>rational</strong> <strong>points</strong>. The largest<br />
number of <strong>points</strong> found was 349.<br />
To find theÉ-<strong>rational</strong> <strong>points</strong>, we used a 2-adic search method which works<br />
as follows. Let a cubic surface X be given. Then, in a first step, we determined<br />
on X all <strong>points</strong> defined over/512(respectively/1024). Then, for each<br />
of the <strong>points</strong> found we checked which of its lifts to P 3 () actually lie on X.<br />
This leads to an O(B 3 )-algorithm which may be efficiently implemented in C.<br />
Furthermore, using the method described in Section 4, we computed an approximation<br />
of Peyre’s constant for each surface.
294 GENERAL CUBIC SURFACES [Chap. IX<br />
5.2. Remark. –––– To search for theÉ-<strong>rational</strong> <strong>points</strong>, there are algorithms<br />
which are asymptotically faster. For example, Elkies’ method which is O(B 2 )<br />
<strong>and</strong> implemented in Magma. A practical comparison shows, however, that<br />
Elkies’ method is not yet faster for our relatively low search bounds.<br />
ii. The density results. —<br />
5.3. –––– For each of the surfaces considered we calculated the quotient<br />
#{ <strong>points</strong> of height < B found } / #{ <strong>points</strong> of height < B expected }.<br />
Let us visualize the distribution of the quotients by some histograms.<br />
2.5<br />
2.5<br />
2<br />
2<br />
1.5<br />
1.5<br />
1<br />
1<br />
0.5<br />
0.5<br />
0<br />
0.5 1 1.5 2 2.5 0<br />
0.5 1 1.5 2 2.5<br />
First sample, B =125 First sample, B =250<br />
FIGURE 1. Distribution of the quotients for the first sample
Sec. 5] NUMERICAL DATA 295<br />
2.5<br />
2.5<br />
2<br />
2<br />
1.5<br />
1.5<br />
1<br />
1<br />
0.5<br />
0.5<br />
0<br />
0.5 1 1.5 2 2.5 0<br />
0.5 1 1.5 2 2.5<br />
Second sample, B =250 Second sample, B =500<br />
FIGURE 2. Distribution of the quotients for the second sample<br />
Some statistical parameters are as follows.<br />
TABLE 1. Parameters of the distribution for the first sample<br />
search bound 125 250<br />
mean 0.99993 0.99887<br />
st<strong>and</strong>ard deviation 0.23558 0.16925<br />
TABLE 2. Parameters of the distribution for the second sample<br />
search bound 250 500<br />
mean 1.00093 0.99943<br />
st<strong>and</strong>ard deviation 0.22527 0.16158<br />
iii. The results for the smallest point. —<br />
5.4. –––– For each of the surfaces in our samples, we determined the height<br />
m(X ) of its smallest point. We visualize the behaviour of m(X ) in the diagrams<br />
below.<br />
At the first glance, it looks very natural to consider the distribution of the values<br />
of m(X ) versus the Tamagawa type number τ (X ). In view of the inequalities<br />
asked for in the introduction, it seems, however, to be better to make a slight<br />
modification <strong>and</strong> plot the product m(X )τ (X ) instead of m(X ) itself.
296 GENERAL CUBIC SURFACES [Chap. IX<br />
m<br />
1000<br />
m<br />
1000<br />
100<br />
100<br />
10<br />
10<br />
1<br />
0.001 0.01 0.1 1 10<br />
First sample<br />
τ<br />
1<br />
0.001 0.01 0.1 1 10<br />
Second sample<br />
τ<br />
FIGURE 3. The smallest height of a <strong>rational</strong> point versus the Tamagawa number<br />
5.5. Conclusion –––– Our experiments suggest that, for general cubic surfaces<br />
X overÉ, the following assertions hold.<br />
i) There are no obstructions to the Hasse principle.<br />
ii) Manin’s conjecture is true in the form refined by E. Peyre.<br />
Further, it is apparent from the diagrams in Figure 3 that the experiment agrees<br />
with the expectation for the heights of the smallest <strong>points</strong> formulated in Remark<br />
1.11. Indeed, for both samples, the slope of a line tangent to the top right<br />
of the scatter plot is near (−1). This indicates that even the strong form of the<br />
estimate should be true. I.e., m(X ) < C (ε)<br />
τ (X ) 1+ε for any ε > 0.<br />
5.6. Running Times –––– The largest portion of the running time was spent on<br />
the calculation of the Euler products. It took 20 days of CPU time to calculate<br />
all 40 000 Euler products for p < 300.<br />
For comparison, we estimated all the integrals, using six recursions, within<br />
36 hours. Further, it took eight days to systematically search for all <strong>points</strong> of<br />
height less than 500 on the surfaces of the second sample. Search for <strong>points</strong> of<br />
height less than 250 on the surfaces of the first sample took only one day.<br />
When running Algorithm 3.4, the lion’s share of the time was used for the<br />
computation of the univariate degree 27 polynomials. This took approximately<br />
seven days of CPU time.<br />
In comparison with that, all other parts were negligible. It took only twelve<br />
minutes to ensure that all 40 000 surfaces are smooth. The C program verifying<br />
that no two of the surfaces are isomorphic to each other ran approximately<br />
80 seconds.
Sec. 6] A CONCRETE EXAMPLE 297<br />
6. A concrete example<br />
i. The Example. —<br />
6.1. Example –––– Let us conclude the article by some results on the particular<br />
cubic surface X given by<br />
x 3 + 2xy 2 + 11y 3 + 3xz 2 + 5y 2 w + 7zw 2 = 0 . (‡)<br />
Example (‡) was not among the surfaces produced by the r<strong>and</strong>om number generator.<br />
Our intention is just to present the output of our algorithms in a specific<br />
(<strong>and</strong> not too artificial) example <strong>and</strong>, most notably, to show the intermediate<br />
results of Algorithm 3.4.<br />
A Gröbner base calculation inMagma shows that X has bad reduction at p = 2,<br />
3, 7, 23, <strong>and</strong> 22 359 013 270 232 677. The idea behind that calculation is the same<br />
as described above for the verification of smoothness. The only difference is<br />
that we consider Gröbner bases overinstead ofÉ.<br />
6.2. The Galois group –––– The first step of Algorithm 3.4 works well on X.<br />
I.e., the polynomial g is indeed of degree 27. Its coefficients become rather large.<br />
The absolutely largest one is that of d 13 . It is equal to 38 300 982 629 255 010.<br />
The leading coefficient of g is 5 3 · 7 12 . We find case A at p = 373. The common<br />
decomposition type (2, 5, 5, 5, 10) of the cases B <strong>and</strong> C occurs at p = 19, 31, 59,<br />
61, 191, 199, <strong>and</strong> 223.<br />
Consequently, X is an explicit example of a smooth cubic surface overÉ<br />
admitting the property that the Galois group which acts on the 27 lines is equal<br />
to W (E 6 ).<br />
6.3. Remark. –––– ThefirstsuchexampleshavebeenconstructedbyT. Ekedahl<br />
[Ek, Theorem 2.1].<br />
6.4. Remark. –––– Our example (‡) is different from Ekedahl’s. Indeed, in<br />
Ekedahl’s examples, the Frobenius Frob 11 acts on the 27 lines as an element<br />
of the conjugacy class C 15 ⊂ W (E 6 ) (in Sir P. Swinnerton-Dyer’s numbering).<br />
In our case, however, the first two steps of Algorithm 3.4 show that Frob 11<br />
yields the decomposition type (1, 1, 1, 1, 1, 2, 4, 4, 4, 4, 4). This corresponds to<br />
the class C 18 [Man,кIV, §9,Ìн]. Note that Ekedahl’s examples, as<br />
well as ours, have good reduction at p = 11.
298 GENERAL CUBIC SURFACES [Chap. IX<br />
ii. The computation of Peyre’s constant. —<br />
6.5. –––– As an approximation of the Euler product, we get<br />
∏ τ p ≈ 0. 729 750 .<br />
p
Sec. 6] A CONCRETE EXAMPLE 299<br />
TABLE 4. Points on X of height ≤ 20<br />
Point Height Point Height<br />
( 0 : 0 : 0 : 1) 1 ( 4 : -6 : -13 : 2) 13<br />
( 0 : 0 : 1 : 0) 1 ( 5 : 5 : 0 : -14) 14<br />
( 1 : 2 : -3 : -2) 3 (10 : -14 : 12 : 11) 14<br />
( 3 : -2 : -3 : 2) 3 ( 0 : 7 : -16 : 7) 16<br />
( 0 : 7 : -6 : -7) 7 (16 : -8 : 3 : -4) 16<br />
( 0 : 4 : -3 : 8) 8 ( 6 : -9 : 16 : 3) 16<br />
( 5 : -5 : 0 : 8) 8 (12 : 7 : -6 : 17) 17<br />
( 2 : -8 : 8 : 7) 8 (14 : -9 : -2 : 17) 17<br />
(10 : -5 : 0 : -1) 10 ( 6 : -3 : -18 : 7) 18<br />
( 0 : 5 : 0 : -11) 11 ( 9 : -6 : -1 : 18) 18<br />
( 8 : 6 : -11 : -8) 12 ( 3 : 6 : -19 : -6) 19<br />
(12 : -6 : -4 : -3) 12 ( 8 : 7 : -4 : 19) 19<br />
( 9 : -12 : 9 : 10) 12<br />
iv. The field of definition of the 27 lines. —<br />
6.7. –––– Having done the Gröbner base calculation in Algorithm 3.4.i),<br />
the 27 lines may be computed at high precision. This allows to find the 45 triangles<br />
on X,explicitly. We calculated a degree 45 resolvent G of the degree 27 polynomial<br />
g the zeroes of which are all the sumsa i1 +a i2 +a i3 for l i1 , l i2 , l i3 representing<br />
three lines which form a triangle. Here, l i : t ↦→ (1 : t : (a i +b i t) : (c i +d i t))<br />
denote parametrizations of the 27 lines. As G ∈[X ], our floating point<br />
calculation is in fact exact.<br />
6.8. Proposition. –––– The unique quadratic subfield in the field K of definition of<br />
the 27 lines on X isÉ(√<br />
−23 · 22 359 013 270 232 677<br />
)<br />
.<br />
Proof. K is unramified at all places of good reduction of X. This leaves us with<br />
only 2 6 − 1 = 63 possibilities for the quadratic subfieldÉ( √ d). To exclude 62<br />
of them is algorithmically easy.<br />
Indeed, for a good prime p, there is a way to compute ( d<br />
p)<br />
without knowledge<br />
of d. We factor the degree 45 resolvent G modulo p. If p divides the<br />
leading coefficient or there are multiple factors then we get no answer. Otherwise,<br />
( d<br />
p) = ±1 depending on whether the decomposition type found is even<br />
or odd in S 45 .<br />
It turns out that it is sufficient to do this for p = 13, 17, 19, 29, 31, <strong>and</strong> 53.<br />
□
300 GENERAL CUBIC SURFACES [Chap. IX<br />
6.9. Proposition. –––– The field extension K/Éis ramified exactly at p = 2, 3, 7,<br />
23, <strong>and</strong> 22 359 013 270 232 677.<br />
Proof. It remains to verify ramification at p = 2, 3, <strong>and</strong> 7. For that, we<br />
computed in Magma the p-adic factorization of g. The decomposition types are<br />
(3, 24) for p = 2 <strong>and</strong> 3 <strong>and</strong> (1, 1, 1, 4, 4, 4, 4, 8) for p = 7.<br />
Let Z p be the decomposition field of p. If p were unramified then Gal(K/Z p )<br />
would be a cyclic group, i.e. Gal(K/Z p ) = 〈σ〉 for some σ ∈ W (E 6 ). On the<br />
other h<strong>and</strong>, on the 27 lines, the orbit structure under the operation of Gal(K/Z p )<br />
isthesameasunder theoperationofGal(Ép/Ép). Thereis, however, noelement<br />
in W (E 6 ) which yields the decomposition type (3, 24) or (1, 1, 1, 4, 4, 4, 4, 8)<br />
[Man,кIV, §9,Ìн].<br />
□<br />
6.10. Remark. –––– This shows that there is no integral model of X which is<br />
smooth over p = 2, 3, 7, 23, or 22 359 013 270 232 677.
CHAPTER X<br />
ON THE SMALLEST POINT ON A<br />
DIAGONAL CUBIC SURFACE ∗<br />
All life is an experiment. The more experiments you make the better.<br />
RALPH WALDO EMERSON (1842)<br />
1. Introduction<br />
1.1. –––– In Chapter VIII, we presented some theoretical <strong>and</strong> experimental<br />
results concerning the height of the smallestÉ-<strong>rational</strong> point on diagonal quartic<br />
threefolds in P 4É. In this chapter, we will consider diagonal cubic surfaces in P 3É.<br />
It will turn out that the analogies to the case treated before are enormous.<br />
However, there are a few <strong>points</strong> indicating that the case of a diagonal cubic<br />
surface is technically more complicated.<br />
An obvious difference is that the Picard rank of a quartic threefold is always<br />
equal to one. By consequence, H 1( Gal(É/É), Pic(XÉ) ) is automatically the<br />
trivial group. Both these observations are wrong for diagonal cubic surfaces.<br />
Hence, the factors α <strong>and</strong> β in the definition of Peyre’s constant (Definition<br />
VI.5.35)will not always be the same <strong>and</strong> need to be considered. Further, one<br />
has to expect that a Brauer-Manin obstruction is present.<br />
1.2. –––– The conjecture of Manin states that the number ofÉ-<strong>rational</strong> <strong>points</strong><br />
of anticanonical height < B on a Fano variety X is asymptotically equal to<br />
τB log rkPic(X )−1 B for B → ∞.<br />
In the particular case of a cubic surface, the anticanonical height is the same<br />
as the naive height. Further, the coefficient τ ∈Êequals the Tamagawa-type<br />
number τ (X ) introduced by E. Peyre.<br />
Thus, one expects at least ∼τ (X )B <strong>points</strong> of height
302 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
this might lead to the expectation that m(X ), the smallest height of aÉ-<strong>rational</strong><br />
point on X, is always less than C<br />
τ<br />
for a certain absolute constant C.<br />
(X )<br />
1.3. –––– To test this expectation, we computed the Tamagawa number <strong>and</strong><br />
ascertained the smallestÉ-<strong>rational</strong> point for each of the cubic surfaces given by<br />
ax 3 + by 3 + 2z 3 + w 3 = 0<br />
for a = 1, . . . , 3000 <strong>and</strong> b = 1, . . . , 300.<br />
Thereby, we restricted our considerations to the cases that<br />
i) a <strong>and</strong> b are odd,<br />
ii) there exists an odd prime p dividing a but not b such that 3 ∤ ν p (a), or<br />
iii) there exists an odd prime p dividing b but not a such that 3 ∤ν p (b).<br />
This guarantees that we are in the “first case” according to the classification of<br />
J.-L. Colliot-Thélène [CT/K/S] <strong>and</strong> his coworkers. In this case, the effect of<br />
the Brauer-Manin obstruction is clear. Precisely two thirds of the adelic <strong>points</strong><br />
are excluded.<br />
The results are summarized by the diagram below.<br />
FIGURE 1. Height of smallest point versus Tamagawa number<br />
It is apparent from the diagram that the experiment agrees with the expectation<br />
above. The slope of a line tangent to the top right of the scatter plot is<br />
indeed near (−1).<br />
However, we will show in Section 5 that, as in the threefold case, the inequality<br />
m(X ) < C<br />
τ<br />
does not hold in general. The following remains a logical possibility.<br />
(X )
Sec. 1] INTRODUCTION 303<br />
1.4. Question. –––– For every ε > 0, does there exist a constant C (ε) such<br />
that, for each cubic surface,<br />
m(X ) <<br />
C (ε)<br />
τ (X ) 1+ε ?<br />
1.5. Peyre’s constant. –––– Recall that, in Definition VI.5.35, E. Peyre’s Tamagawa-type<br />
number was defined as<br />
τ (X ) := α(X )·β(X ) · lim<br />
s→1<br />
(s − 1) t L(s, χ Pic(XÉ)) · τ H<br />
(<br />
X (É)<br />
for t = rk Pic(X ).<br />
The factor β(X ) is simply defined as<br />
β(X ) := #H 1( Gal(É/É), Pic(XÉ) ) .<br />
Br)<br />
α(X ) is given as follows. Let Λ eff (X ) ⊂ Pic(X )⊗Êbe the cone generated by<br />
the effective divisors. Identify Pic(X )⊗ÊwithÊt via a mapping induced by an<br />
∼=<br />
isomorphism Pic(X ) −→t . Consider the dual cone Λ ∨ eff (X ) ⊂ (Êt ) ∨ . Then,<br />
α(X ) := t · vol {x ∈ Λ ∨ eff | 〈x, −K 〉 ≤ 1 } .<br />
L( · , χ Pic(XÉ)) denotes the Artin L-function of the Gal(É/É)-representation<br />
Pic(XÉ) ⊗which contains the trivial representation t times as a direct summ<strong>and</strong>.<br />
Therefore, L(s, χ Pic(XÉ)) = ζ(s) t · L(s, χ P ) <strong>and</strong><br />
lim<br />
s→1<br />
(s − 1) t L(s, χ Pic(XÉ)) = L(1, χ P )<br />
where ζ denotes the Riemann zeta function <strong>and</strong> P is a representation which<br />
does not contain trivial components. [Mu-y, Corollary 11.5 <strong>and</strong> Corollary 11.4]<br />
show that L(s, χ P ) has neither a pole nor a zero at s = 1.<br />
Finally, τ H is the Tamagawa measure on the set X (É) of adelic <strong>points</strong> on X<br />
<strong>and</strong> X (É) Br ⊆ X (É) denotes the part which is not affected by the Brauer-<br />
Manin obstruction.<br />
1.6. The Tamagawa measure. –––– As X is projective, we have<br />
X (É) = ∏ X (Éν).<br />
ν∈Val(É)<br />
τ H is defined to be a product measure τ H := ∏ τ ν .<br />
ν∈Val(É)<br />
For a prime number p, the local measure τ p on X (Ép) is given as follows.<br />
Let X ⊆ P 3be the model of X given by the defining cubic equation.<br />
For a ∈ X (/p k), put<br />
U (k)<br />
a := {x ∈ X (Ép) | x ≡ a (mod p k ) } .
304 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
Then,<br />
τ p (U (k)<br />
a ) := det(1 − p−1 Frob p |Pic(XÉ) I p<br />
)<br />
#{ y ∈ X (/p<br />
· m) | y ≡ a (mod p k ) }<br />
lim<br />
.<br />
m→∞ p m dim X<br />
Here, Pic(XÉ) I p<br />
denotes the fixed module under the inertia group.<br />
τ ∞ is described in Definition VI.5.22. In the case of a cubic surface defined by<br />
the equation f = 0, this yields<br />
τ ∞ (U ) = 1 Z<br />
ω Leray<br />
2<br />
CU<br />
|x 0 |, ... ,|x 3 |≤1<br />
for U ⊂ X (Ê). Here, ω Leray is the Leray measure on the cone CX (Ê) associated<br />
to the equation f = 0.<br />
The Leray measure is related to the usual hypersurface measure by the formula<br />
ω Leray = 1<br />
‖ grad f ‖ ω hyp.<br />
1.7. The results. –––– In the case of diagonal cubic surfaces, there is an estimate<br />
1<br />
for m(X ) in terms of τ (X ). Namely,<br />
τ<br />
admits a fundamental finiteness property.<br />
More precisely, in Section 6, we will show the following<br />
(X )<br />
theorem.<br />
Theorem. Let a = (a 0 , . . . , a 3 ) ∈ (\{0}) 4 be a vector. Denote by X a the cubic<br />
surface in P 3Égiven by a 0 x 3 0 + . . . + a 3 x 3 3 = 0. Then, for each ε > 0 there exists a<br />
constant C (ε) > 0 such that<br />
1<br />
τ (X a ) ≥ C (ε) · H( 1<br />
a 0<br />
: . . . :<br />
1<br />
a 3<br />
) 1<br />
3<br />
−ε<br />
.<br />
Corollary (Fundamental finiteness). For each T > 0, there are only finitely many<br />
diagonal cubic surfaces X a : a 0 x 3 0 + . . . + a 3 x 3 3 = 0 in P 3Ésuch that τ (V a ) > T .<br />
Corollary (An inefficient search bound). There exists a monotonically decreasing<br />
function F : (0, ∞) → [0, ∞), the search bound, satisfying the following condition.<br />
Let X a be the cubic surface given by the equation a 0 x 3 0 + . . . + a 3 x 3 3 = 0.<br />
Assume X a (É) ≠ ∅. Then, X a admits aÉ-<strong>rational</strong> point of height ≤F (τ (X a )).<br />
Proof. One may simply put F (t) := max min H(P).<br />
τ (X a )≥t P∈X a (É) X a (É)≠∅<br />
□<br />
In other words, we have m(X a ) ≤ F (τ (X a )) as soon as X a (É) ≠ ∅.
Sec. 2] THE FACTORS α AND β 305<br />
2. The factors α <strong>and</strong> β<br />
2.1. –––– Recall that on a smooth cubic surface X over an algebraically closed<br />
field, there are exactly 27 lines. For the Picard group, which is isomorphic to7 ,<br />
the classes of these lines form a system of generators.<br />
2.2. Notation. –––– i) ThesetL ofthe27linesisequippedwith theintersection<br />
product 〈 , 〉: L ×L → {−1, 0, 1}. The pair (L , 〈 , 〉) is the same for all<br />
smooth cubic surfaces. It is well known from Chapter II, 8.21 that the group<br />
of permutations of L respecting 〈 , 〉 is isomorphic to W (E 6 ). We fix such<br />
an isomorphism.<br />
Denote by F ⊂ Div(X ) the group generated by the 27 lines <strong>and</strong> by F 0 ⊂ F the<br />
subgroup of principal divisors. Then, F is equipped with an operation of W (E 6 )<br />
such that F 0 is a W (E 6 )-submodule. We have Pic(X ) ∼ = F/F 0 .<br />
ii) If X is a smooth cubic surface overÉthen Gal(É/É) operates canonically<br />
on the set L X of the 27 lines on XÉ. Fix a bijection i X : L X −→ L<br />
∼ =<br />
respecting the intersection pairing. This induces a group homomorphism<br />
ι X : Gal(É/É) → W (E 6 ). We denote its image by G ⊂ W (E 6 ).<br />
2.3. Lemma. –––– There is a constant C such that, for all smooth cubic surfaces X<br />
overÉ,<br />
1 ≤ β(X ) ≤ C.<br />
Proof. By definition, β(X ) = #H 1( Gal(É/É), Pic(XÉ) ) . Using the notation<br />
just introduced, we may write H 1( Gal(É/É), Pic(XÉ) ) = H 1 (G, F/F 0 ).<br />
Note that this cohomology group is always finite. Indeed, since G is a finite<br />
group <strong>and</strong> F/F 0 is a finite[G]-module, the description via the st<strong>and</strong>ard<br />
complex shows it is finitely generated. Further, it is annihilated by #G.<br />
H 1 (G, F/F 0 ) depends only on the subgroup G ⊂ W (E 6 ) occurring. For that,<br />
there are finitely many possibilities. This implies the claim.<br />
□<br />
2.4. Remark. –––– A more precise consideration (cf. Proposition II.8.18) yields<br />
a canonical isomorphism<br />
H 1( Gal(É/É), Pic(XÉ) ) ∼ = Hom<br />
(<br />
(NF ∩ F0 )/NF 0 ,É/) .<br />
Here, N is the norm map under the operation of G.<br />
As an application of this, one may inspect the 350 conjugacy classes of sub<strong>groups</strong><br />
of W (E 6 ) using GAP. See Chapter II, 8.23, or the complete list of the values<br />
of H 1( Gal(É/É), Pic(XÉ) ) given in the appendix. The calculations show that<br />
the lemma is actually true for C = 9.
306 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
2.5. Lemma. –––– There are positive constants C 1 <strong>and</strong> C 2 such that, for all smooth<br />
cubic surfaces X overÉsatisfying X (É) ≠ ∅,<br />
C 1 ≤ α(X ) ≤ C 2 .<br />
Proof. Again, we claim that α(X ) is completely determined by the group<br />
G ⊂ W (E 6 ). Thus, suppose that we do not have the full information available<br />
about what surface X is but are given the group G only.<br />
The assumption X (É) ≠ ∅ makes sure that Pic(X ) ∼ = Pic(XÉ) G [K/T, Remark<br />
3.2.ii)]. We may therefore write Pic(X ) ∼ = (F/F 0 ) G . The effective cone<br />
Λ eff (X ) ⊂ Pic(X )⊗∼ = (F/F0 ) G ⊗is generated by the symmetrizations<br />
of the classes l 1 , . . . , l 27 of the 27 lines in F. In particular, it is determined<br />
by G, completely. Further, we have K = − 1(l 9 1 + . . . + l 27 ). These data are<br />
sufficient to compute α(X ) according to its very definition.<br />
□<br />
2.6. Remark. –––– Here, we do not know the optimal values of C 1 <strong>and</strong> C 2 in<br />
explicit form. α(X ) has not yet been computed in all cases.<br />
2.7. Remark. –––– In the experiment, we work entirely with cubic surfaces of<br />
Picard rank one overÉ. This may easily be seem from Theorem 3.6, below.<br />
Therefore, we always have α(X ) = 1. Further, Theorem III.6.4.a) implies<br />
that β(X ) = 3.<br />
3. Splitting the Picard group<br />
3.1. –––– In the case of the diagonal cubic surface X (a 0,...,a 3 )<br />
⊂ P 3É, given<br />
by a 0 x 3 0 + . . . + a 3x 3 3 = 0 for a 0, . . . , a 3 ∈\{0}, the 27 lines on X (a 0,...,a 3 )<br />
may easily be written down explicitly. Indeed, for each pair (i, j) ∈ (/3) 2 ,<br />
the system<br />
√ √<br />
3 a0 x 0 + ζ3<br />
i 3 a1 x 1 = 0<br />
√ √<br />
3 a2 x 2 + ζ j 3<br />
3 a3 x 3 = 0<br />
of equations defines a line on X (a 0,...,a 3 ) . Decomposing the index set<br />
{0, . . . , 3} differently into two subsets of two elements each yields<br />
all the lines. √In particular, √ we √ see that<br />
)<br />
the 27 lines may be defined<br />
over K =É( ζ3 , 3 a1 /a 0 , 3 a2 /a 0 , 3 a3 /a 0 .
Sec. 3] SPLITTING THE PICARD GROUP 307<br />
3.2. Fact. –––– Let p be a prime number <strong>and</strong> a 0 , . . . , a 3 be integers not divisible<br />
by p. Then,<br />
#X (a 0, ... ,a 3 ) (p) =<br />
⎧<br />
p<br />
⎪⎨<br />
2 + ( 1 + χ 3 (a 0 a 1 a 2 2 a2 3 ) + χ 3(a 2 0 a2 1 a 2a 3 )<br />
+ χ 3 (a 0 a 2 1 a 2a 2 3 ) + χ 3(a 2 0 a 1a 2 2 a 3)<br />
+ χ 3 (a 0 a 2 1<br />
⎪⎩<br />
a2 2 a 3) + χ 3 (a 2 0 a 1a 2 a 2 3 )) p + 1 if p ≡ 1 (mod 3) ,<br />
p 2 + p + 1 if p ≡ 2 (mod 3) .<br />
Here, in the case p ≡ 1 (mod 3), χ 3 :∗ p →denotes a cubic residue character.<br />
Proof. If p ≡ 2 (mod 3) then every residue class modulo p has a<br />
unique cubic root. Therefore, the map X (a 0, ... ,a 3 ) (p) → P 2 (p) given by<br />
(x : y : z : w) ↦→ (x : y : z) is bijective. This shows #X (a 0, ... ,a 3 ) (p) = p 2 +p+1.<br />
Turn to the case p ≡ 1 (mod 3). It is classically known that, on a degree m<br />
diagonal variety, the number ofp-<strong>rational</strong> <strong>points</strong> for p ≡ 1 (mod m) may<br />
be determined using Jacobi sums. The formula given follows immediately<br />
from [I/R, Chapter 10, Theorem 2] together with the well-known relation<br />
g(χ 3<br />
É<br />
)g(χ 2 3 ) = p for cubic Gauß sums.<br />
□<br />
3.3. Lemma. –––– Let a 0 , . . . , a 3<br />
É<br />
∈\{0}. Then, for each prime p such that<br />
p ∤ 3a 0 · . . . · a 3 ,<br />
χ<br />
(a Pic(X 0 ,...,a 3 )<br />
)(Frob p ) = tr ( )⊗)<br />
Frob p | Pic(X (a 0,...,a 3 )<br />
⎧<br />
χ 3 (a 0 a 1 a<br />
⎪⎨<br />
2 2 a2 3 ) + χ 3(a 2 0 a2 1 a 2a 3 )<br />
+ χ 3 (a 0 a 2 1<br />
=<br />
a 2a 2 3 ) + χ 3(a 2 0 a 1a 2 2 a 3)<br />
+ χ 3 (a 0 a 2 1<br />
⎪⎩<br />
a2 2 a 3) + χ 3 (a 2 0 a 1a 2<br />
É ⊗<br />
a 2 3 ) + 1 if p ≡ 1 (mod 3) ,<br />
1 if p ≡ 2 (mod 3) .<br />
Proof. As we have good reduction, the trace of Frob p on Pic(X (a 0,...,a 3 )<br />
)<br />
is the same as that of Frob on Pic(X (a 0,...,a 3 )<br />
) ⊗. Further, for the number<br />
p<br />
of <strong>points</strong> on a non-singular cubic surface over a finite field, the Lefschetz trace<br />
formula can be made completely explicit [Man,кIV,ÌÓÖѺ½]. It shows<br />
#X (a 0, ... ,a 3 ) (p) = p 2 + p · tr ( Frob | Pic(X (a ⊗) 0, ... ,a 3 )<br />
) + 1 .<br />
p<br />
The explicit formulas for the numbers of <strong>points</strong> given in Fact 3.2 therefore yield<br />
the assertion.<br />
□<br />
√<br />
3.4. Notation. –––– For A an integer, denote the fieldÉ(ζ 3 , 3 A) by K.<br />
Further, let G := Gal(K/É), H := Gal(K/É(ζ 3 )), <strong>and</strong> χ: H → 〈ζ 3 〉 ⊂∗<br />
be a primitive character. Then, we write ν K := ind G H<br />
(χ) for the induced character<br />
<strong>and</strong> V K for the corresponding G-representation.
308 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
If K is of degree three overÉ(ζ 3 ) then V K is an irreducible rank two representation<br />
of G ∼ = S 3 . Otherwise, K =É(ζ 3 ). Then, V K ∼ =⊕M splits<br />
into the direct sum of a trivial <strong>and</strong> a non-trivial one-dimensional representation<br />
of H ∼ =/2.<br />
We will freely consider V K as a Gal(É/É)-representation.<br />
3.5. Lemma. –––– Let A be any integer. Then, for a prime p not dividing A,<br />
we have<br />
{<br />
νÉ(ζ 3 , 3√ A) χ3 (A) + χ<br />
(Frob p ) =<br />
3 (A) if p ≡ 1 (mod 3) ,<br />
0 if p ≡ 2 (mod 3) .<br />
Proof. The primitive character is unique up to conjugation by an element of G.<br />
Therefore, the induced character λ is well-defined.<br />
The Kummer pairing allows to make a definite choice for χ as follows. Fix an<br />
embedding σ :É(ζ 3 ) →. Then, put χ(g) := σ ( g( 3 √ A)/<br />
3√ A<br />
)<br />
.<br />
If p ≡ 2 (mod 3) then p remains prime inÉ(ζ 3 ). This means, Frob p acts<br />
non-trivially onÉ(ζ 3 ), i.e. Frob p ∈ G\H . Since H is a normal subgroup in G,<br />
the induced character vanishes on such an element.<br />
For p ≡ 1 (mod 3), we have that (p) splits inÉ(ζ 3 ). Let us write (p) = pp.<br />
The choice of p is equivalent to the choice of a homomorphism ι: 〈ζ 3 〉 →∗ p .<br />
The Frobenius Frob p is determined only up to conjugation, we may choose<br />
Frob p = Frob p ∈ H . Then, directly by the definition of an induced<br />
character, νÉ(ζ 3 , 3√ A) (Frob p ) = χ(Frob p ) + χ(Frob p ). We need to show<br />
that χ(Frob p ) = χ 3 (A) or χ(Frob p ) = χ 3 (A).<br />
For this, by the choice made above, we have<br />
χ(Frob p ) := σ ( Frob p ( 3 √<br />
A)/<br />
3√<br />
A<br />
)<br />
.<br />
After reduction modulo p, we may write<br />
Frob( 3 √<br />
A)/<br />
3√<br />
A = (<br />
3√<br />
A) p / 3 √<br />
A = A<br />
p−1<br />
3 .<br />
√<br />
Therefore, Frob p ( 3 A)/<br />
3√ A = ι −1 (A p−1<br />
3 ) which shows<br />
É<br />
χ(Frob p ) = σ ( ι −1 (A p−1<br />
3 ) ) .<br />
É<br />
That final formula is a definition for a cubic residue character at A. □<br />
3.6. Theorem. –––– Let a 0 , . . ., a 3 ∈\{0}. Then, the Gal(É/É)-representation<br />
Pic(X (a 0,...,a 3 )<br />
) ⊗splits into the direct sum<br />
Pic ( X (a 0,...,a 3<br />
)<br />
) ⊗∼ =⊕V K 1<br />
⊕ V K 2<br />
⊕ V K 3
Sec. 3] SPLITTING THE PICARD GROUP 309<br />
√<br />
where K 1 , K 2 , <strong>and</strong> K 3 denote the fieldsÉ(ζ 3 , 3 a0 a 1 a 2 2 a2 3 ),É(ζ √<br />
√ 3 , 3 a0 a 2 1 a 2a 2 3 ),<br />
<strong>and</strong>É(ζ 3 , 3 a0 a 2 1 a2 2 a É<br />
3), respectively.<br />
Proof. We will show that the representations on both sides have the same character.<br />
For that, by virtue of the Chebotarev density theorem, it suffices to<br />
consider the values at the Frobenii Frob p for p ∤ 3a 0 · . . . · a 3 .<br />
For the representation on the left h<strong>and</strong><br />
É<br />
side, χ<br />
(a Pic(X 0 ,...,a 3 )<br />
)(Frob p ) has been computed<br />
in Lemma 3.3. For the representation on the right h<strong>and</strong> side, Lemma 3.5<br />
shows that exactly the same formula is true.<br />
□<br />
3.7. Corollary. –––– Let a 0 , . . . , a 3 ∈\{0} be integers,<br />
P := Pic ( X (a 0, ... ,a 3<br />
)<br />
) ⊗,<br />
considered as a Gal(É/É)-representation,<br />
√<br />
<strong>and</strong> denote by χ P be the associated<br />
character. Put K 1 :=É(ζ 3 , 3 a0 a 1 a 2 2 a2 3 ), K √<br />
√ 2 :=É(ζ 3 , 3 a0 a 2 1 a 2a 2 3<br />
), <strong>and</strong><br />
K 3 :=É(ζ 3 , 3 a0 a 2 1 a2 2 a 3). Then, for the Artin conductor N χP of χ P , we have<br />
where<br />
N 2 χ P<br />
= D(K 1 ) D(K 2 ) D(K 3 )/(−27) ,<br />
D(K ) :=<br />
{ Disc(K/É) if [K :É(ζ 3 )] = 3 ,<br />
−27 if K =É(ζ 3 ) .<br />
Proof. We have to show N 2 = D(K )/(−3). Assume first that [K :É(ζ<br />
ν K 3 )] = 3.<br />
Then, the conductor-discriminant formula [Ne, Chapter VII, Section (11.9)]<br />
shows Disc(K/É) = NN M N 2 <strong>and</strong> −3 = Disc(É(ζ<br />
ν K 3 )/É) = NN M which<br />
together yield the assertion. In the opposite case, we have V K =⊕M<br />
<strong>and</strong> N ν K = NN M = −3.<br />
□<br />
3.8. Lemma. –––– Let a <strong>and</strong> b be integers different from zero. Then,<br />
Proof. We have, at first,<br />
|Disc (É(ζ 3 , 3 √<br />
ab2 )/É) | ≤ 3 9 a 4 b 4 .<br />
| Disc (É(ζ 3 , 3 √<br />
ab2 )/É) | ≤ | Disc<br />
(É(ζ 3 )/É) |3 · Disc (É( 3 √<br />
ab2 )/É) 2<br />
= 27 · Disc (É( 3 √<br />
ab2 )/É) 2.<br />
Further, by [Mc, Chapter 2, Exercise 41], we know<br />
|Disc (É( 3 √<br />
ab2 )/É) | ≤ 3 3 a 2 b 2 .<br />
This shows<br />
|Disc (É(ζ 3 , 3 √<br />
ab2 )/É) | ≤ 3 9 a 4 b 4 .<br />
□
É<br />
310 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
3.9. Corollary. –––– Let a 0 , . . . , a 3 ∈\{0} be integers,<br />
P := Pic ( X (a 0, ... ,a 3<br />
)<br />
) ⊗,<br />
considered as a Gal(É/É)-representation, <strong>and</strong> χ P be the associated character.<br />
Then, for the Artin conductor N χP of χ P , we have the estimate<br />
| N χP | ≤ 3 12 (a 0 · . . . · a 3 ) 6 .<br />
Proof. Lemma 3.8 shows | D(K i )| ≤ 3 9 (a 0 · . . . · a 3 ) 4 for i = 1, 2, <strong>and</strong> 3.<br />
The assertion follows immediately from this.<br />
□<br />
4. A technical lemma<br />
4.1. Sublemma. –––– a) (Good reduction)<br />
If p ∤ 3a 0 · . . . · a 3 then the sequence ( #X (a 0,...,a 3<br />
n∈Æ<br />
) (/p n)/p2n) n∈Æis constant.<br />
b) (Bad reduction)<br />
i) If p divides a 0· . . . ·a 3 but not 3 then the sequence ( #X (a 0,...,a 3 ) (/p n)/p2n) becomes stationary as soon as p n does not divide any of the coefficients a 0 , . . . , a 3 .<br />
ii) If p = 3 then the sequence ( #X (a 0,...,a 3 ) (/p n)/p2n) n∈Æbecomes stationary as<br />
soon as 3 n does not divide any of the numbers 3a 0 , . . . , 3a 3 .<br />
4.2. Lemma. –––– There are two positive constants C 1 <strong>and</strong> C 2 such that, for<br />
all a 0 , . . ., a 3 ∈\{0},<br />
Proof. Cf. Lemma VI.5.34.<br />
C 1 < ∏<br />
p prime<br />
p∤3a 0·····a 3<br />
τ p<br />
(<br />
X<br />
(a 0 , ... ,a 3 ) (Ép) ) < C 2 .<br />
For a prime p of good reduction, Sublemma 4.1 shows<br />
τ p<br />
(<br />
X<br />
(a 0 , ... ,a 3 ) (Ép) ) = det ( 1 − p −1 Frob p | Pic(XÉ) ) · #X (a 0, ... ,a 3 ) (p)<br />
p 2 .<br />
Further, for the number of <strong>points</strong> on a non-singular cubic surface over a finite<br />
field, the Lefschetz trace formula can be made completely explicit [Man,<br />
кIV,ÌÓÖѺ½]. It shows<br />
#X (a 0, ... ,a 3 ) (p) = p 2 + p · tr ( Frob p | Pic(XÉ) ) + 1 .
Sec. 5] A NEGATIVE RESULT 311<br />
Denoting the eigenvalues of the Frobenius on Pic(XÉ) by λ 1 , . . . , λ 7 , we find<br />
τ p<br />
(<br />
X<br />
(a 0 , ... ,a 3 ) (Ép) )<br />
= (1 − λ 1 p −1 )(1 − λ 2 p −1 ) · . . . · (1 − λ 7 p −1 )<br />
· [1 + (λ 1 + · · · + λ 7 )p −1 + p −2 ]<br />
= (1 − σ 1 p −1 + σ 2 p −2 ∓ . . . − σ 7 p −7 )(1 + σ 1 p −1 + p −2 )<br />
= 1 + (1 − σ 2 1 + σ 2 )p −2 − (σ 1 − σ 1 σ 2 + σ 3 )p −3 ±<br />
± . . . − (σ 5 − σ 1 σ 6 + σ 7 )p −7 + (σ 6 − σ 1 σ 7 )p −8 − σ 7 p −9<br />
where σ i denote the elementary symmetric functions in λ 1 , . . . , λ 7 .<br />
We know |λ i | = 1 for all i. Estimating very roughly, we have |σ j | ≤ ( 7 j ) ≤ 7j<br />
<strong>and</strong> see<br />
1 − 99p −2 − 7·99p −3 − . . . − 7 7·99p −9 ≤ τ p<br />
(<br />
X<br />
(a 0 , ... ,a 3 ) (Ép) ) ≤<br />
≤ 1 + 99p −2 + 7·99p −3 + . . . + 7 7 · 99p −9 .<br />
I.e., 1 − 99p −2 1 < τ (<br />
1−7/p p X<br />
(a 0 , ... ,a 3 ) (Ép) ) < 1 + 99p −2 1<br />
1−7/p<br />
product over all 1 − 99p −2 1<br />
1−7/p<br />
The left h<strong>and</strong> side is positive for p > 13.<br />
(respectively 1 + 99p−2<br />
1<br />
1−7/p<br />
. The infinite<br />
) is convergent.<br />
For the small primes remaining,<br />
we need a better lower bound. For this, note that a cubic<br />
surface over<br />
(<br />
a finite fieldp always has at least onep-<strong>rational</strong> point.<br />
This yields τ p X<br />
(a 0 , ... ,a 3 ) (Ép) ) ≥ (1 − 1/p) 7 /p 2 > 0.<br />
□<br />
4.3. Remark. –––– The correctional factors det ( )<br />
1 − p −1 Frob p | Pic(XÉ) I p<br />
are all positive. Indeed, for a pair of complex conjugate eigenvalues, we have<br />
(1 − λp −1 )(1 − λp −1 ) = |1 − λp −1 | 2 > 0 <strong>and</strong> an eigenvalue of 1 or (−1)<br />
contributes a factor 1 ± p −1 > 0. Consequently, we always have<br />
(<br />
1 − 1 p) 7<br />
< det<br />
(<br />
1 − p −1 Frob p | Pic(XÉ) I p<br />
) <<br />
(<br />
1 + 1 p) 7.<br />
5. A negative result<br />
5.1. –––– For an integer q ≠ 0, let X (q) ⊂ P 3Ébe the cubic surface given by<br />
qx 3 + 4y 3 + 2z 3 + w 3 = 0 <strong>and</strong> let<br />
m(X (q) ) := min { H(x : y : z : w) | (x : y : z : w) ∈ X (q) (É)}<br />
be the smallest height of aÉ-<strong>rational</strong> point on X (q) . We compare m(X (q) ) with<br />
the Tamagawa type number τ (q) := τ (X (q) ).
312 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
5.2. Lemma. –––– There is a constant C with the following property. For each<br />
pair (a, b)of natural numbers satisfying gcd(a, b) = 1, thereare two prime numbers<br />
p 1 , p 2 ≡ a (mod b), p 1 ≠ p 2 , such that p 1 , p 2 < C ·b 5.5 .<br />
Proof. Linnik’s Theorem in the version of R. Heath-Brown [H-B92b] shows<br />
there is one prime with the property stated. To get a second one, the following<br />
simple trick helps.<br />
One of the numbers a <strong>and</strong> a + b is relatively prime to 4b. Assume without<br />
restriction that gcd(a, 4b) = 1. Then, we find p 1 ≡ a (mod 4b) <strong>and</strong><br />
p 2 ≡ a + 2b (mod 4b) such that p 1 , p 2 < ˜C ·(4b) 5.5 .<br />
□<br />
5.3. Theorem. –––– Assume the Generalized Riemann Hypothesis. Then, there is<br />
no constant C such that<br />
m(X (q) ) < C<br />
τ (q)<br />
for all q ∈\{0}.<br />
Proof. We will construct a sequence (q i ) i∈Æof prime numbers such that<br />
q i ≡ 1 (mod 72) <strong>and</strong> m(X (qi) ) τ (q i)<br />
→ ∞ for i → ∞. The proof will consist<br />
of several steps.<br />
Firststep. The prime q i divides exactly one of the four coefficients in the equation<br />
defining X (qi) . In this case, it is known by the work of J.-L. Colliot-Thélène <strong>and</strong><br />
his coworkers [CT/K/S, Proof of Proposition 2] that precisely<br />
τ H<br />
(<br />
X (É)<br />
It is therefore sufficient to verify that<br />
Br) = 1 3 τ H(<br />
X (É)<br />
) .<br />
m(X (qi) ) · α(X (qi) )·β(X (qi) )·lim (s −1) t L(s, χ )· (<br />
(qi)<br />
s→1 Pic(X ∏ τ<br />
É) ν X<br />
(q i ) (Éν) ) → ∞.<br />
ν∈Val(É)<br />
Second step. If q is a prime different from 2 then we have rkPic(X (q) ) = 1.<br />
[Man,кIII,ÍÔÖÒÒº½¾] shows that X (q) is a minimal cubic surface<br />
overÉ. By [Man,кIV,ÌÓÖѺ½], we have Pic(X (q) ) =.<br />
Third step. We have α(X (qi) ) = 1 <strong>and</strong> β(X (qi) ) = 3.<br />
α(X (qi) ) = 1 follows immediately from rk Pic(X (qi) ) = 1. β(X (qi) ) can<br />
be computed by the method indicated in Remark 2.4. We work with a<br />
group G ⊂ W (E 6 ) of order 18 which decomposes the 27 lines into three<br />
orbits of nine lines each. An easy calculation shows<br />
Hom ( (NF ∩ F 0 )/NF 0 ,É/) ∼=/3.<br />
Fourth step. For the height of the smallest point, we have m(X (q) ) ≥ 3 √<br />
q<br />
7 .
Sec. 5] A NEGATIVE RESULT 313<br />
There are no <strong>rational</strong> solutions of the equation 4y 3 + 2z 3 + w 3 = 0 as<br />
this is impossible, 2-adically. √ |x| ≥ 1 yields |4y 3 + 2z 3 + w 3 | ≥ q <strong>and</strong><br />
max{|y|, |z|, |w|} ≥ q 3 . 7<br />
Fifth step. For |q| ≥ 7, one has τ ∞<br />
(<br />
X (q) (Ê) ) = 1<br />
3√<br />
|q|<br />
I where I is independent<br />
of q.<br />
By Lemma 6.15, we have<br />
τ ∞ (X (q) ) = 1<br />
4 3 √<br />
|q|<br />
Z<br />
CX (1, ... ,1) (Ê)<br />
|x 0 |≤ 3√ |q|,|x 1 |≤ 3√ 4,|x 2 |≤ 3√ 2,|x 3 |≤1<br />
ω CX (1, ... ,1) (Ê)<br />
Leray .<br />
Since |x 1 | ≤ 3 √<br />
4, |x2 | ≤ 3 √<br />
2, |x3 | ≤ 1, <strong>and</strong><br />
|x 0 | = 3 √<br />
| x 3 1 + x3 2 + x3 3 | ≤ 3 √| x 1 | 3 + | x 2 | 3 + | x 3 3 | ,<br />
the condition |x 0 | ≤ 3 √<br />
|q| is actually empty.<br />
(<br />
Sixth step. There is a positive constant C such that ∏ τ p X (q) (Ép) ) > C for<br />
p prime<br />
every prime q ≡ 1 (mod 72).<br />
By Lemma 4.2, we have C 1 > 0 such that<br />
∏<br />
p prime<br />
p≠2,3,q<br />
τ p<br />
(<br />
X (q) (Ép) ) > C 1 .<br />
It, therefore, remains to give lower bounds for the factors τ 2<br />
(<br />
X (q) (É2) ) ,<br />
τ 3<br />
(<br />
X (q) (É3) ) , <strong>and</strong> τ q<br />
(<br />
X (q) (Éq) ) .<br />
As 2 ∤q, by virtue of Sublemma 4.1 we have, τ 2<br />
(<br />
X (q) (É2) ) = 1 2 7 · #X (q) (/8)<br />
64<br />
.<br />
Further,<br />
#X (q) (/8) ≥ 1<br />
since q ≡ 1 (mod 8) implies (1 : 0 : 0 : (−1)) ∈ X (q) (/8).<br />
Similarly, τ 3<br />
(<br />
X (q) (É3) ) = ( 2 3 )7 · #X (q) (/9)<br />
81<br />
. Again, q ≡ 1 (mod 9) makes sure<br />
that (1 : 0 : 0 : (−1)) ∈ X (q) (/9) <strong>and</strong> #X (q) (/9) ≥ 1.<br />
For the prime q, we argue a bit differently. First,<br />
det ( 1 − q −1 Frob p | Pic(X É) (q) ) I q ≥ (1 − 1/q) 7 ≥ (72/73) 7 .<br />
Furthermore, the reduction of X (q) modulo q is the cone over the elliptic<br />
curve given by 4y 3 + 2z 3 + w 3 = 0. Therefore, on X (q) there are at<br />
least (q − 2 √ q + 1)(q − 1) smooth <strong>points</strong> defined overq. As Hensel’s lemma
314 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
may be applied to them, we get<br />
#X (q) (/q n)<br />
lim<br />
≥ (q − 2√ q + 1)(q − 1)<br />
(<br />
> 1 − 2 )(<br />
√ 1 − 1 )<br />
n→∞ q 2n q 2 q q<br />
≥ 72 (<br />
1 − √ 2 ).<br />
73 73<br />
Seventh step. There is a sequence (q i ) i∈Æof primes such that q i ≡ 1 (mod 72)<br />
<strong>and</strong> [lim (s − 1)L(s, χ (qi)<br />
s→1 Pic(X<br />
)] → ∞ for i → ∞.<br />
É)<br />
Since rk Pic(X (qi) ) = 1, the representation Pic ( É)<br />
X (q i) ⊗contains exactly one<br />
trivial summ<strong>and</strong>. Hence,<br />
L ( )<br />
s, χ (qi) Pic(X = ζ(s) · L(s, χP (qi))<br />
É)<br />
for a representation P (q i) not containing trivial components. Our goal is, therefore,<br />
to show L(1, χ P (qi)) → ∞ for i → ∞.<br />
DenotebyP i thei-thprimenumber p suchthat p ≡ 1 (mod 3). For eachi ∈Æ,<br />
choose a cubic residue r i modulo P i . Then, define q i to be the second smallest<br />
prime number such that<br />
q i ≡ r j (mod P j )<br />
for all j ∈ {1, . . . , i} <strong>and</strong> q i ≡ 1 (mod 72).<br />
This system of simultaneous congruences is solvable by the Chinese Remainder<br />
Theorem. Its solutions form an arithmetic progression with common difference<br />
72P 1 · . . . · P i . By Chebyshev’s inequalities, we know<br />
72P 1 · . . . · P i ≤ 72e θ(P i) < 72e (2 log2)P i<br />
.<br />
According to our definition, q i is the second smallest prime in this arithmetic<br />
progression. From this, we clearly have that q i > 72P 1 · . . . · P i → ∞<br />
for i → ∞. On the other h<strong>and</strong>, Lemma 5.2 shows<br />
for certain constants C 1 <strong>and</strong> C 2 .<br />
q i ≤ C 1 · (72e (2 log2)P i<br />
) 5.5 = C 2 e (11 log2)P i<br />
Corollary 3.9 gives us an estimate for the Artin conductor of the character χ P (qi).<br />
We see N χP ≤ 3 12 (a<br />
(qi) 0 · . . . · a 3 ) 6 = 3 12 8 6 qi 6 ≤ C 3 e (66 log2)P i<br />
for another constant<br />
C 3 . Consequently,<br />
logN χP (qi )<br />
≤ (66 log2)P i + logC 3 .<br />
We observe that (log N χP (qi) )1/2 ≤ P i for i sufficiently large. We assume from<br />
now on that this inequality is fulfilled.
Sec. 5] A NEGATIVE RESULT 315<br />
Recall that P (q i) is actually the direct sum of representations which are induced<br />
from one-dimensional characters (Theorem 3.6). By consequence, it is known<br />
that the Artin L-function L( · , χ P (qi)) is entire. Since we also assume the<br />
Generalized Riemann Hypothesis, we may apply the estimate of W. Duke [Duk,<br />
Proposition 5]. It shows<br />
Here,<br />
logL(1, χ P (qi)) =<br />
∑<br />
p
316 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
| D(K 3 )| = |Disc (É(ζ 3 , 3 √<br />
qi )/É) |<br />
= Disc (É( 3 √<br />
qi )/É)2<br />
· |N ( Disc (É(ζ 3 , 3 √<br />
qi )/É( 3 √<br />
qi ) )) |<br />
≥ Disc (É( 3 √<br />
qi )/É) 2.<br />
According to [Mc, Chapter 2, Exercise 41], we know<br />
| Disc (É( 3 √<br />
qi )/É) | ≥ 3q<br />
2<br />
i . □<br />
5.4. Remark. –––– Note that the estimate for L(1, χ P (qi)) is the only point where<br />
we use the Generalized Riemann Hypothesis.<br />
6. The fundamental finiteness property<br />
6.1. –––– In this section, we return to the case of a general diagonal cubic<br />
surface X (a 0, ... ,a 3 )<br />
⊂ P 3Égiven by a 0 x 3 0 + . . . + a 3x 3 3 = 0. Our goal is to<br />
establish the estimate for τ (a 0, ... ,a 3 ) := τ (X (a 0, ... ,a 3 ) ) formulated as Theorem 1.7.<br />
For this, in the subsections below, we will give an individual estimate for each<br />
of the factors occurring in the definition of τ (X (a 0, ... ,a 3 ) ).<br />
i. An estimate for the L-factor. —<br />
É<br />
)<br />
6.2. Proposition. –––– For each ε > 0, there exist positive constants c 1 <strong>and</strong> c 2<br />
such that<br />
c 1 · |a 0 · . . . · a 3 | −ε < lim (s − 1) t L ( s, χ<br />
s→1<br />
É<br />
)<br />
(a Pic(X 0 , ... ,a 3 ) < c2 · |a 0 · . . . · a 3 | ε<br />
for all (a 0 , . . . , a 3 ) ∈ (\{0}) 4 . Here, t = rk Pic(X ).<br />
Proof. The Galois representation Pic(X (a 0, ... ,a 3 )<br />
) ⊗contains the trivial representation<br />
t times as a direct summ<strong>and</strong>. Therefore,<br />
L ( )<br />
s, χ (a Pic(X 0 , ... ,a 3 ) = ζ(s)t · L(s, χ P )<br />
É<br />
)<br />
where ζ denotes the Riemann zeta function <strong>and</strong> P is a representation which<br />
does not contain trivial components. All we need to show is<br />
c 1 · |a 0 · . . . · a 3 | −ε < L(1, χ P ) < c 2 · |a 0 · . . . · a 3 | ε .<br />
L( · , χ P ) is the product of at most three factors of the form L( · , λ) where λ is<br />
the non-trivial Dirichlet character ofÉ(ζ 3 )/É<strong>and</strong> at most three factors which<br />
are Artin-L-functions L( · , ν K ) for K a purely cubic field extension ofÉ(ζ 3 )
Sec. 6] THE FUNDAMENTAL FINITENESS PROPERTY 317<br />
as above, sayK =É( ζ3 , 3 √<br />
a0 a 1 a 2 2 a2 3)<br />
. AsL(1, λ)does not depend on a0 , . . . , a 3 ,<br />
at all, it will suffice to show<br />
for each ε > 0.<br />
c 1 (ε) · |a 0 · . . . · a 3 | −ε < L(1, ν K ) < c 2 (ε) · |a 0 · . . . · a 3 | ε<br />
V K is the only irreducible two-dimensional representation of Gal(K/É) ∼ = S 3 .<br />
For that reason, by virtue of [Ne, Chapter VII, Corollary (10.5)], we have<br />
ζ K (s) = ζÉ(s) · L(s, λ) · L(s, ν K ) 2<br />
= ζÉ(ζ 3 )(s) · L(s, ν K ) 2<br />
for a complex variable s. It, therefore, suffices in our particular situation to<br />
estimate the residue res s=1 ζ K (s) of the Dedekind zeta function of K.<br />
An estimate from above has been given by C. L. Siegel. In view of the analytic<br />
class number formula, his [Si69, Satz 1] gives<br />
res ζ K (s) < C (log Disc(K/É)) 5<br />
s=1<br />
≤ C [log(3 9 a 4 0a 4 1a 4 2a 4 3)] 5<br />
= C [4 log |a 0 · . . . · a 3 | + 9 log 3] 5<br />
for a certain constant C. The final term is less than c 2 (ε) · |a 0 · . . . · a 3 | ε for<br />
every ε > 0.<br />
On the other h<strong>and</strong>, H. M. Stark [St, formula (1)] shows<br />
res ζ K (s) > C (ε)·Disc(K/É) −ε/4<br />
s=1<br />
for every ε > 0 which implies res<br />
s=1 ζ K (s) > c 1 (ε)·|a 0 · . . . · a 3 | −ε . □<br />
ii. An estimate for the factors at the finite places. —<br />
6.3. Notation. –––– We will adopt the notation from Chapter VIII, 3.1. More<br />
precisely,<br />
i) For a prime number p <strong>and</strong> an integer x ≠ 0, we put x (p) := p ν p(x) .<br />
Note x (p) = 1/‖x‖ p for the normalized p-adic valuation.<br />
ii) By putting ν(x) := min<br />
ξ∈p<br />
to/p r.<br />
x=(ξ mod p r )<br />
ν(ξ), we carry the p-adic valuation fromp over<br />
Note that any 0 ≠ x ∈/p rhas the form x = ε·p ν(x) where ε ∈ (/p r) ∗ is<br />
a unit. Clearly, ε is unique only in the case ν(x) = 0.
318 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
6.4. Definition. –––– For (a 0 , . . . , a 3 ) ∈4 , r ∈Æ, <strong>and</strong> ν 0 , . . . , ν 3 ≤ r, put<br />
N (r)<br />
ν 0 , ... ,ν 3 ;a 0 , ... ,a 3<br />
:= { (x 0 , . . . ,x 3 ) ∈ (/p r) 4 |<br />
ν(x 0 ) = ν 0 , . . . ,ν(x 3 ) = ν 3 ; a 0 x 3 0 + . . . + a 3 x 3 3 = 0 ∈/p r}.<br />
For the particular case ν 0 = . . . = ν 3 = 0, we will write<br />
I.e.,<br />
Z (r)<br />
a 0 , ... ,a 3<br />
:= N (r)<br />
0, ... ,0;a 0 , ... ,a 3<br />
.<br />
Z (r)<br />
a 0 , ... ,a 3<br />
= { (x 0 , . . . , x 3 ) ∈ [(/p r) ∗ ] 4 | a 0 x 3 0 + . . . + a 3 x 3 3 = 0 ∈/p r}.<br />
We will use the notation z (r)<br />
a 0 , ... ,a 3<br />
:= #Z (r)<br />
a 0 , ... ,a 3<br />
.<br />
6.5. Sublemma. –––– If p k |a 0 , . . . , a 3 <strong>and</strong> r > k then we have<br />
z (r)<br />
a 0 , ... ,a 3<br />
= p 4k · z (r−k)<br />
a 0 /p k , ... ,a 3 /p k .<br />
Proof. Since a 0 x 3 0 + . . . + a 3 x 3 3 = p k (a 0 /p k · x 3 0 + . . . + a 3 /p k · x 3 3 ), there is a<br />
surjection<br />
ι: Z a (r)<br />
0 , ... ,a 3<br />
−→ Z (r−k) ,<br />
a 0 /p k , ... ,a 3 /p k<br />
given by (x 0 , . . . , x 3 ) ↦→ ( (x 0 mod p r−k ), . . . , (x 3 mod p r−k ) ) . The kernel of<br />
the homomorphism of modules underlying ι is (p r−k/pr) 4 . □<br />
6.6. Lemma. –––– Assume gcd p<br />
(a 0 , . . . , a 4 ) = p k . Then, there is an estimate<br />
z (r)<br />
a 0 , ... ,a 4<br />
≤ 3p 3r+k .<br />
Proof. Suppose first that k = 0. This means, one of the coefficients is prime<br />
to p. Without restriction, assume p ∤a 0 .<br />
For any (x 1 , x 2 , x 3 ) ∈ (/p r) 3 , there appears an equation of the form a 0 x 3 0 = c.<br />
It cannot have more than three solutions in (/p r) ∗ . Indeed, for p odd, this<br />
follows directly from the fact that (/p r) ∗ is a cyclic group. On the other<br />
h<strong>and</strong>, in the case p = 2, we have (/2 r) ∗ ∼ =/2 r−2×/2. Again, there<br />
are only up to three solutions possible.<br />
The general case may now easily be deduced from Sublemma 6.5. Indeed, if<br />
k < r then<br />
z (r)<br />
a 0 , ... ,a 3<br />
= p 4k · z (r−k)<br />
a 0 /p k , ... ,a 3 /p k ≤ p 4k · 3p 3(r−k) = 3p 3r+k .<br />
On the other h<strong>and</strong>, if k ≥ r then the assertion is completely trivial since<br />
z (r)<br />
a 0 , ... ,a 3<br />
= #Z (r)<br />
a 0 , ... ,a 3<br />
< p 4r ≤ p 3r+k < 3p 3r+k .<br />
□
Sec. 6] THE FUNDAMENTAL FINITENESS PROPERTY 319<br />
6.7. Remark. –––– The proof shows that in the case p ≡ 2 (mod 3) one could<br />
reduce the coefficient to 1. Unfortunately, this observation does not lead to a<br />
substantial improvement of our final result.<br />
6.8. Lemma. –––– Let r ∈Æ<strong>and</strong> ν 0 , . . . , ν 3 ≤ r. Then,<br />
#N ν (r)<br />
0 , ... ,ν 3 ;a 0 , ... ,a 3<br />
= z(r) · ϕ(p r−ν 0<br />
) · . . . · ϕ(p r−ν 3<br />
)<br />
p 3ν 0a 0 , ... ,p 3ν 3a 3<br />
.<br />
ϕ(p r ) 4<br />
Proof. As p 3ν 0 a0 x 3 0 + . . . + p 3ν 3 a3 x 3 3 = a 0 (p ν 0 x0 ) 3 + . . . + a 3 (p ν 3 x3 ) 3 , we have<br />
a surjection<br />
π : Z (r) −→ N (r)<br />
p 3ν 0a 0 , ... ,p 3ν 3a 3<br />
ν 0 , ... ,ν 3 ;a 0 , ... ,a 3<br />
,<br />
given by (x 0 , . . . , x 3 ) ↦→ (p ν 0 x0 , . . . , p ν 3 x3 ).<br />
For i = 0, . . . , 3, consider the mapping ι:/p r→/pr, x ↦→ p ν i<br />
x.<br />
If ν i = r then ι is the zero map. All ϕ(p r ) = (p − 1)p r−1 units are mapped<br />
to zero. Otherwise, observe that ι is p ν i<br />
: 1 on its image. Further, ν(ι(x)) = ν i<br />
if <strong>and</strong> only if x is a unit. By consequence, π is (K (ν 0)<br />
· . . . · K (ν3) ) : 1 when<br />
we put K (ν) := p ν for ν < r <strong>and</strong> K (r) := (p − 1)p r−1 . Summarizing, we could<br />
have written K (ν) := ϕ(p r )/ϕ(p r−ν ). The assertion follows.<br />
□<br />
6.9. Corollary. –––– Let (a 0 , . . . , a 3 ) ∈ (\{0}) 4 . Then, for the local factor<br />
τ p<br />
(<br />
X<br />
(a 0 , ... ,a 3 ) (Ép) ) , one has<br />
(<br />
τ p X<br />
(a 0 , ... ,a 3 ) (Ép) ) = det ( )<br />
1 − p −1 Frob p | Pic(XÉ) I p<br />
r<br />
r→∞<br />
∑<br />
ν 0 ,...,ν 3 =0<br />
· lim<br />
z (r) · ϕ(p r−ν p 3ν 0a 0 , ... ,p 3ν 0<br />
3a ) · . . . · ϕ(p<br />
r−ν 3)<br />
3<br />
.<br />
p 3r · ϕ(p r ) 4<br />
Proof. By Remark VI.5.19, we have<br />
τ p<br />
(<br />
X<br />
(a 0 , ... ,a 3 ) (Ép) ) = det ( 1 − p −1 Frob p | Pic(XÉ) I p )<br />
Lemma 6.8 yields the assertion.<br />
r<br />
#N ν<br />
r→∞<br />
∑<br />
(r)<br />
0 ,...,ν 3 ;a 0 ,...,a 3<br />
ν 0 ,...,ν 3 =0<br />
· lim<br />
p 3r .<br />
□<br />
6.10. Proposition. –––– Let (a 0 , . . . , a 3 ) ∈ (\{0}) 4 . Then, for each ε such<br />
that 0 < ε < 1 , one has<br />
3<br />
(<br />
τ p X<br />
(a 0 , ... ,a 3 ) (Ép) ) (<br />
≤ 1+ 1 ) (<br />
1<br />
)( 1<br />
) 3·(<br />
(p)<br />
) 1−ε<br />
7·3 ( a<br />
p 1 − 1 1 − 1 0 a(p) 1 a(p) 3<br />
2<br />
a (p) ) ε<br />
3 .<br />
p 1−3ε p ε
320 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
Proof. We use the formula from Corollary 6.9. By Remark 4.3, the first factor<br />
is at most (1 + 1/p) 7 . Further, by Lemma 6.6,<br />
z (r)<br />
p 3ν 0a 0 , ... ,p 3ν 3a 3<br />
/p 3r ≤ 3 gcd p<br />
(p 3ν 0<br />
a 0 , . . . , p 3ν 3<br />
a 3 )<br />
(<br />
(p)) Writing k i := ν p (a i ) = ν p a i , we see<br />
= 3 gcd ( p 3ν 0<br />
a (p)<br />
0 , . . . , p3ν 3<br />
a (p)<br />
3<br />
z (r)<br />
p 3ν 0a 0 , ... ,p 3ν 3a 3<br />
/p 3r ≤ 3 gcd(p 3ν 0+k 0<br />
, . . . , p 3ν 3+k 3<br />
)<br />
= 3p min{3ν 0+k 0 , ... ,3ν 3 +k 3 } .<br />
We estimate the minimum by a weighted arithmetic mean with weights 1−ε<br />
1−ε<br />
, <strong>and</strong> ε,<br />
, 1−ε<br />
3 3<br />
min{3ν 0 + k 0 , . . . , 3ν 3 + k 3 }<br />
This shows<br />
≤ 1 − ε<br />
3<br />
· (3ν 0 + k 0 ) + 1 − ε · (3ν 1 + k 1 )<br />
3<br />
+ 1 − ε · (3ν 2 + k 2 ) + ε(3ν 3 + k 3 )<br />
3<br />
= (1 − ε)(ν 0 + ν 1 + ν 2 ) + 3εν 3<br />
)<br />
.<br />
+ 1 − ε (k 0 + k 1 + k 2 ) + εk 3 .<br />
3<br />
z (r) /p 3r ≤ 3p (1−ε)(ν 0+ν 1 +ν 2 )+3εν 3 + 1−ε<br />
p 3ν 0a 0 , ... ,p 3ν 3 (k 0+k 1 +k 2 )+εk 3<br />
3a 3<br />
(a<br />
= 3p (1−ε)(ν 0+ν 1 +ν 2 )+3εν3 (p)<br />
) 1−ε<br />
· ( 0 a(p) 1 a(p) 3<br />
2<br />
a (p) ) ε.<br />
3<br />
We may therefore write<br />
τ p<br />
(<br />
X<br />
(a 0 , ... ,a 3 ) (Ép) ) ≤<br />
r<br />
r→∞<br />
∑<br />
ν 0 , ... ,ν 3 =0<br />
· lim<br />
(<br />
1 + 1 ) 7 ( · 3 a<br />
(p)<br />
) 1−ε (<br />
0<br />
p<br />
a(p) 1 a(p) 3<br />
2<br />
a (p) ) ε<br />
3<br />
p (1−ε)(ν 0+ν 1 +ν 2 )+3εν 3 · ϕ(p<br />
r−ν 0) · . . . · ϕ(p<br />
r−ν 3)<br />
ϕ(p r ) 4 .<br />
Here, the term under the limit is precisely the product of three copies of the<br />
finite sum r<br />
p<br />
∑<br />
(1−ε)ν · ϕ(p r−ν r−1<br />
) 1<br />
=<br />
ν=0<br />
ϕ(p r )<br />
∑<br />
ν=0<br />
(p ε ) + p 1<br />
ν p − 1 (p ε ) r<br />
<strong>and</strong> one copy of the finite sum<br />
r<br />
p<br />
∑<br />
3εν · ϕ(p r−ν )<br />
ν=0<br />
ϕ(p r )<br />
=<br />
r−1<br />
∑<br />
ν=0<br />
1<br />
(p 1−3ε ) + p<br />
ν<br />
1<br />
p − 1 (p 1−3ε ) . r<br />
3 ,
Sec. 6] THE FUNDAMENTAL FINITENESS PROPERTY 321<br />
For r → ∞, geometric series do appear while the additional summ<strong>and</strong>s tend<br />
to zero.<br />
□<br />
6.11. Remark. –––– Unfortunately, the constants<br />
) 7 (<br />
C p (ε) := · 3<br />
(<br />
1 + 1 p<br />
1<br />
)( 1<br />
) 3<br />
1 − 1 1 − 1 p 1−3ε p ε<br />
diverges. On the other h<strong>and</strong>, we<br />
is bounded for p → ∞, say C p (ε) ≤ C (ε) .<br />
have the property that the product ∏ p C (ε)<br />
p<br />
have at least that C (ε)<br />
p<br />
6.12. Lemma. –––– Let C > 1 be any constant. Then, for each ε > 0, one has<br />
∏<br />
p prime<br />
p|x<br />
for a suitable constant c (depending on ε).<br />
C ≤ c · x ε<br />
Proof. This follows directly from [Nat, Theorem 7.2] together with [Nat,<br />
Section 7.1, Exercise 7].<br />
□<br />
6.13. Proposition. –––– For each ε such that 0 < ε < 1 , there exists a constant c<br />
3<br />
such that<br />
(<br />
∏τ p X<br />
(a 0 , ... ,a 3 ) (Ép) ) ≤ c · |a 0 · . . . · a 3 | 3 1 − ε 8 · ∏<br />
p prime<br />
for all (a 0 , . . . , a 3 ) ∈ (\{0}) 4 .<br />
p prime<br />
min ‖a i‖ 1 3 −ε<br />
p<br />
i=0, ... ,3<br />
Proof. The product over all primes of good reduction is bounded by virtue of<br />
Lemma 4.2.a). It, therefore, remains to show that<br />
∏<br />
p prime<br />
p|3a 0 ... a 3<br />
(<br />
τ p X<br />
(a 0 , ... ,a 3 ) (Ép) ) ≤ c · |a 0 · . . . · a 3 | 3 1 − ε 8 · ∏ min ‖a i‖ 1 3 −ε<br />
i=0, ... ,3<br />
For this, by Proposition 6.10, we have at first<br />
τ p<br />
(<br />
X<br />
(a 0 , ... ,a 3 ) (Ép) ) ≤ C (ε)<br />
p<br />
= C (ε)<br />
p<br />
·<br />
·<br />
p prime<br />
(a(p)<br />
) 1<br />
0 a(p) 1 a(p) 3 − ε4<br />
2<br />
· (a (p)<br />
(a(p)<br />
0 a(p) 1 a(p) 2 a(p) 3<br />
3 ) 3 4 ε<br />
) 1<br />
3<br />
− ε4 · (a (p)<br />
3 )− 1 3 +ε .<br />
Here, the indices 0, . . . , 3 are interchangeable. Hence, it is even allowed to<br />
write<br />
(<br />
τ p X<br />
(a 0 , ... ,a 3 ) (Ép) ) (a<br />
≤ C p<br />
(ε) (p)<br />
) 1<br />
·<br />
0 a(p) 1 a(p) 2 a(p) 3<br />
− ε4<br />
3<br />
· (max )<br />
a (p) −<br />
1<br />
3<br />
+ε<br />
i i<br />
)<br />
= C p<br />
(ε)<br />
1<br />
· 3<br />
− ε4 · min ‖a i ‖ 1 3 −ε<br />
p .<br />
(a(p)<br />
0 a(p) 1 a(p) 2 a(p) 3<br />
Now, we multiply over all prime divisors of a 0 · . . . · a 3 . Thereby, on the<br />
right h<strong>and</strong> side, we may twice write the product over all primes since the two<br />
i<br />
p .
322 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
rightmost factors are equal to one for p ∤ 3a 0 · . . . · a 3 , anyway.<br />
(<br />
τ p X<br />
(a 0 , ... ,a 3 ) (Ép) )<br />
∏<br />
p prime<br />
p|3a 0 ... a 3<br />
≤ ∏ C p<br />
(ε)<br />
p prime<br />
p|3a 0 ... a 3<br />
(<br />
(p)<br />
) 1<br />
· ∏ a<br />
0 a(p) 1 a(p) 2 a(p) 3 − ε 4<br />
3<br />
· ∏<br />
p prime<br />
= ∏ C p (ε) · | a 0 · . . . · a 3 | 1 3 − ε 4 · ∏<br />
p prime<br />
p|3a 0 ... a 3<br />
p prime<br />
p prime<br />
min ‖a i‖ 1 3 −ε<br />
p<br />
i=0, ... ,3<br />
min ‖a 3<br />
i‖ 1 −ε<br />
p<br />
i=0, ... ,3<br />
when we observe that ∏ p a (p) = | a|. Further, we have C p<br />
(ε)<br />
Lemma 6.12,<br />
∏ C (ε) ≤ c · |3a 0 · . . . · a 3 | ε 8 .<br />
p prime<br />
p|3a 0 ... a 3<br />
We finally estimate 3 ε 8 by a constant. The assertion follows.<br />
iii. An estimate for the factor at the infinite place. —<br />
6.14. Corollary. –––– Let a 0 , . . . , a 3 ∈Ê\{0}. Then,<br />
[<br />
ω CX (a 0 , ... ,a 3 ) (Ê) 1<br />
Leray = dx<br />
3| a 0 |x 2 1 ∧ dx 2 ∧ dx 3<br />
].<br />
0<br />
Proof. We apply Lemma VI.5.26 to U =Ê4 <strong>and</strong><br />
f (x 0 , . . . , x 3 ) := a 0 x 3 0 + . . . + a 3 x 3 3 .<br />
≤ C (ε) <strong>and</strong>, by<br />
Note that {(x 0 , . . . , x 3 ) ∈ CX (a 0,...,a 3 ) (Ê) | x 0 = 0} is a zero set according to<br />
the Leray measure as it is for the hypersurface measure.<br />
□<br />
6.15. Lemma. –––– Let a 0 , . . . , a 3 ∈Ê\{0}. Then,<br />
(<br />
τ ∞ X<br />
(a 0 , ... ,a 3 ) (Ê) ) 1<br />
= √<br />
2 3 | a0 · . . . · a 3 |<br />
Z<br />
CX (1, ... ,1) (Ê)<br />
|x 0 |≤ 3√ |a 0 |, ... ,|x 3 |≤ 3√ |a 3 |<br />
ω CX (1, ... ,1) (Ê)<br />
Leray .<br />
(<br />
Proof. According to the definition of τ ∞ X<br />
(a 0 , ... ,a 3 ) (Ê) ) <strong>and</strong> the corollary above,<br />
we need to show<br />
Z<br />
1 1<br />
dx 1 ∧dx 2 ∧ dx 3<br />
6 |a 0 |<br />
x 2<br />
CX (a 0 , ... ,a 3 ) 0<br />
(Ê)<br />
|x 0 |≤1, ... ,|x 3 |≤1<br />
Z<br />
1<br />
1<br />
= √ dX<br />
6 3 | a0 · . . . · a 3 | X 2 1 ∧ dX 2 ∧ dX 3 .<br />
CX (1, ... ,1) 0<br />
(Ê)<br />
|X 0 |≤ 3√ |a 0 |, ... ,|X 3 |≤ 3√ |a 3 |<br />
□
Sec. 6] THE FUNDAMENTAL FINITENESS PROPERTY 323<br />
For that, consider the linear mapping l : CX (a 0, ... ,a 3<br />
√ √ ) (Ê) → CX (1, ... ,1) (Ê) given<br />
by (x 0 , . . . , x 3 ) ↦→ ( 3 a0 x 0 , . . . , 3 a3 x 3 ). Then,<br />
( 1<br />
) 3√<br />
l ∗ a1 a 2 a 3 1<br />
dX<br />
X0<br />
2 1 ∧ dX 2 ∧ dX 3 = dx<br />
x 2 1 ∧ dx 2 ∧ dx 3 .<br />
0<br />
This immediately yields the assertion when we take into consideration that<br />
orientations are chosen in such a way that both integrals are positive. □<br />
a 2/3<br />
0<br />
6.16. Proposition. –––– For real numbers 0 < b 0 ≤ b 1 ≤ b 2 ≤ b 3 , we have<br />
Z<br />
(<br />
ω CX (1, ... ,1) (Ê)<br />
Leray ≤ 64 + 64<br />
3 log 3 + 1 3√<br />
3ω2<br />
)b 0 + 64b 0 log b 1<br />
3<br />
b 0<br />
CX (1, ... ,1)(Ê)<br />
| x 0 |≤b 0 , ... ,|x 3 |≤b 3<br />
where ω 2 is the two-dimensional hypersurface measure on the l 3 -unit sphere<br />
S 2 := { (x 1 , x 2 , x 3 ) ∈Ê3 | | x 1 | 3 + | x 2 | 3 + | x 3 | 3 = 1 } .<br />
Proof. First step. We cover the domain of integration by 25 sets as follows.<br />
We put R 0 := [−b 0 , b 0 ] 4 ∩ CX (1, ... ,1) (Ê). Further, for each σ ∈ S 4 , we set<br />
R σ := { (x 0 , . . . , x 3 ) ∈Ê4 | | x σ(0) | ≤ · · · ≤ |x σ(3) |, | x i | ≤ b i , b 0 ≤ | x σ(3) | }<br />
Second step. One has R R σ<br />
ω CX (1, ... ,1) (Ê)<br />
Leray<br />
∩ CX (1, ... ,1) (Ê) .<br />
≤ R R id<br />
ω CX (1, ... ,1) (Ê)<br />
Leray for every σ ∈ S 4 .<br />
Consider the map i σ :Ê4 →Ê4 given by (x 0 , . . . , x 3 ) ↦→ (x σ(0) , . . . , x σ(3) ).<br />
Since CX (1, ... ,1) (Ê) is defined by a symmetric cubic form, it is invariant under i σ .<br />
We claim that i σ (R σ ) ⊆ R id .<br />
Indeed, let (x 0 , . . . , x 3 ) ∈ R σ . Then, i σ (x 0 , . . . , x 3 ) = (x σ(0) , . . . , x σ(3) )<br />
has the properties | x σ(0) | ≤ . . . ≤ | x σ(3) | <strong>and</strong> b 0 ≤ | x σ(3) |. In order to show<br />
i σ (x 0 , . . . , x 3 ) ∈ R id , all we need to verify is | x σ(i) | ≤ b i for i = 0, . . . , 3.<br />
For this, we use that the b i are sorted. We have | x σ(3) | ≤ b σ(3) ≤ b 3 .<br />
Further, | x σ(2) | ≤ b σ(2) <strong>and</strong> | x σ(2) | ≤ |x σ(3) | ≤ b σ(3) , one of which is at<br />
most equal to b 2 . Similarly, | x σ(1) | ≤ b σ(1) , | x σ(1) | ≤ |x σ(2) | ≤ b σ(2) ,<br />
<strong>and</strong> | x σ(1) | ≤ |x σ(3) | ≤ b σ(3) , the smallest of which is not larger than b 1 .<br />
Finally, | x σ(0) | ≤ b σ(0) , | x σ(0) | ≤ |x σ(1) | ≤ b σ(1) , | x σ(0) | ≤ |x σ(2) | ≤ b σ(2) ,<br />
<strong>and</strong> | x σ(0) | ≤ |x σ(3) | ≤ b σ(3) which shows | x σ(0) | ≤ b 0 .<br />
As x 3 0+ . . . +x 3 3 is a symmetric form, the Leray measure on CX (1, ... ,1) (Ê) is invariant<br />
under the canonical operation of S 4 on CX (1, ... ,1) (Ê) ⊂Ê4 . This means,<br />
we have (i σ ) ∗ ω CX (1, ... ,1) (Ê)<br />
Leray = ω CX (1, ... ,1) (Ê)<br />
Leray for each σ ∈ S 4 .
324 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
Altogether,<br />
Z<br />
ω CX (1, ... ,1) (Ê)<br />
Leray<br />
R σ<br />
≤<br />
Z<br />
i −1 σ (R id )<br />
Z<br />
ω CX (1, ... ,1) (Ê)<br />
Leray =<br />
(i σ ) ∗ ω CX (1, ... ,1) (Ê)<br />
Leray =<br />
R id<br />
Z<br />
R id<br />
ω CX (1, ... ,1) (Ê)<br />
Leray .<br />
Third step. We have R R 0<br />
ω CX (1, ... ,1) (Ê)<br />
Leray ≤<br />
3√ 1 3ω2 b<br />
3 0 .<br />
By virtue of Corollary 6.14, we have<br />
Z<br />
Leray = 1 3<br />
R 0<br />
ω CX (1, ... ,1) (Ê)<br />
= 1 3<br />
Z<br />
1<br />
x 2 3<br />
R 0<br />
ZZZ<br />
π(R 0 )<br />
dx 0 ∧ dx 1 ∧ dx 2<br />
1<br />
(x 3 0 + x3 1 + x3 2 )2/3 dx 0 dx 1 dx 2<br />
where π : CX (1, ... ,1) (Ê) →Ê3 , (x 0 , x 1 , x 2 , x 3 ) ↦→ (x 0 , x 1 , x 2 ), denotes the projection<br />
to the first three coordinates.<br />
We enlarge the domain of integration to<br />
R ′ := { (x 1 , x 2 , x 3 ) ∈Ê3 | | x 0 | 3 + | x 1 | 3 + | x 2 | 3 ≤ 3b 3 0 } .<br />
Then, by homogeneity, we see<br />
ZZZ<br />
R ′ 1<br />
(x 3 0 + x3 1 + x3 2 )2/3 dx 0 dx 1 dx 2 = ω 2 ·<br />
Fourth step. We have R<br />
Observe<br />
R id<br />
ω CX (1, ... ,1) (Ê)<br />
3√<br />
3b0<br />
Z<br />
0<br />
1<br />
r 2 · r2 dr = ω 2 · 3√<br />
3b0 .<br />
Leray ≤ ( 8 3 + 8 9 log 3)b 0 + 8 3 b 0 log b 1<br />
| x 3 | = ∣<br />
√x 3 3 0 + x3 1 + x3 2 ∣ ≤<br />
√| 3 x 0 | 3 + | x 1 | 3 + | x 2 | 3 .<br />
For (x 0 , . . . , x 3 ) ∈ R id , this implies | x 3 | ≤ 3 √<br />
3 | x2 | <strong>and</strong> | x 2 | ≥ b 0 / 3 √<br />
3. We find<br />
Z<br />
Leray = 1 3<br />
R id<br />
ω CX (1, ... ,1) (Ê)<br />
Z<br />
1<br />
x 2 3<br />
R id<br />
≤ 1 Z<br />
1<br />
3 x 2 2<br />
R id<br />
< 1 3<br />
≤ 1 3<br />
Z b 0<br />
dx 0 ∧ dx 1 ∧ dx 2<br />
dx 0 ∧ dx 1 ∧ dx 2<br />
Z<br />
−b 0 |x 1 |∈[| x 0 |,b 1 ]<br />
Z b 0 Z<br />
−b 0 |x 1 |∈[|x 0 |,b 1 ]<br />
Z<br />
|x 2 |≥b 0 / 3√ 3<br />
|x 2 |≥|x 1 |<br />
1<br />
x 2 2<br />
dx 2 dx 1 dx 0<br />
2<br />
√<br />
max{b 0 / 3 3, | x1 |} dx 1 dx 0<br />
b 0<br />
.
Sec. 6] THE FUNDAMENTAL FINITENESS PROPERTY 325<br />
≤ 2 3<br />
⎡<br />
⎢<br />
⎣<br />
Z b 0<br />
Z<br />
−b 0 |x 1 |∈[|x 0 |,b 0 / 3√ 3]<br />
3√<br />
3<br />
dx 1 dx 0 +<br />
b 0<br />
Z b 0<br />
Z<br />
−b 0 |x 1 |∈[b 0 / 3√ 3,b 1 ]<br />
⎤<br />
1<br />
| x 1 | dx ⎥<br />
1 dx 0 ⎦<br />
≤ 2 3 · 4b2 0<br />
3√<br />
3 ·<br />
3√<br />
3<br />
+ 2 Z b 0<br />
2 log<br />
b 0 3<br />
−b 0<br />
3√<br />
3b1<br />
b 0<br />
dx 0<br />
= 8 3 b 0 + 8 3 b 0 log<br />
=<br />
3√<br />
3b1<br />
b 0<br />
( 8<br />
3 + 8 9 log 3 )<br />
b 0 + 8 3 b 0 log b 1<br />
b 0<br />
.<br />
□<br />
6.17. Corollary. –––– For every ε > 0, there exists a constant C such that<br />
τ ∞<br />
(<br />
X<br />
(a 0 , ... ,a 3 ) (Ê) ) ≤ C · | a 0 · . . . · a 3 | − 1 3 +ε · min<br />
i=0, ... ,3 ‖a i‖ 1 3<br />
∞<br />
for each (a 0 , . . . , a 3 ) ∈ (\{0}) 4 .<br />
Proof. We assume without restriction that | a 0 | ≤ . . . ≤ | a 3 |. Then,<br />
Lemma 6.15 <strong>and</strong> Proposition 6.16 together show that, for certain explicit positive<br />
constants C 1 <strong>and</strong> C 2 ,<br />
(<br />
τ ∞ X<br />
(a 0 , ... ,a 3 ) (Ê) ) (<br />
√ )<br />
≤ | a 0 · . . . · a 3 | − 3 1 · C 1 | a 0 | 3 1 + C2 | a 0 | 3 1 log<br />
3<br />
| a 1 |<br />
| a 0 |<br />
= | a 0 · . . . · a 3 | − 1 3 · | a0 | 1 3<br />
(<br />
C 1 + 1 3 C 2 log | a 1|<br />
| a 0 |<br />
≤ | a 0 · . . . · a 3 | − 3 1 · min ‖a i‖ 1 3<br />
∞<br />
i=0, ... ,3<br />
(<br />
· C 1 + 1 )<br />
3 C 2 log | a 0 · . . . · a 3 | .<br />
There is a constant C such that C 1 + 1 C 3 2 log | a 0 · . . . · a 3 | ≤ C | a 0 · . . . · a 3 | ε<br />
for every (a 0 , . . . , a 3 ) ∈ (\{0}) 4 .<br />
□<br />
)<br />
iv. The Tamagawa number. —<br />
6.18. Proposition. –––– For every ε > 0, there exists a constant C > 0 such that<br />
1<br />
τ ≥ C · H( ) 1<br />
1 1<br />
a 0<br />
: . . . : 3<br />
a 3 (a 0, ... ,a 3 )<br />
|a 0 · . . . · a 3 | ε<br />
for each (a 0 , . . . , a 3 ) ∈ (\{0}) 4 .
326 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
Proof. We may assume that ε is small, say ε < 2 . Then, immediately from the<br />
3<br />
definition of τ (a 0, ... ,a 3 ) , we have<br />
τ (a 0, ... ,a 3 )<br />
= α(X (a 0, ... ,a 3 ) )·β(X (a 0, ... ,a 3 ) ) · lim<br />
s→1<br />
(s − 1) t L ( s, χ Pic(X<br />
(a 0 , ... ,a 3 )<br />
≤ α(X (a 0, ... ,a 3 ) )·β(X (a 0, ... ,a 3 ) ) · lim<br />
s→1<br />
(s − 1) t L ( s, χ Pic(X<br />
(a 0 , ... ,a 3 )<br />
É<br />
)<br />
= α(X (a 0, ... ,a 3 ) )·β(X (a 0, ... ,a 3 ) ) · lim<br />
s→1<br />
(s − 1) t L ( s, χ Pic(X<br />
(a 0 , ... ,a 3 )<br />
)<br />
)<br />
É<br />
(<br />
·τ H X<br />
(a 0 , ... ,a 3 ) (É) Br)<br />
)<br />
)<br />
(<br />
·τ H X<br />
(a 0 , ... ,a 3 ) (É) )<br />
)<br />
)<br />
·∏<br />
ν∈Val(É)<br />
τ ν<br />
(<br />
X<br />
(a 0 , ... ,a 3 ) (Éν) ) .<br />
Let us collect estimates for the factors. First, by Proposition 6.2, we have<br />
lim<br />
s→1<br />
(s − 1) t L ( s, χ Pic(X<br />
(a 0 , ... ,a 3 )<br />
) < c1 · | a 0 · . . . · a 3 | ε 16<br />
for a certain constant c 1 . Further, Proposition 6.13 yields<br />
(<br />
∏ τ p X<br />
(a 0 , ... ,a 3 ) (Ép) ) ≤ c 2 · | a 0 · . . . · a 3 | 3 1 − ε<br />
16 · ∏<br />
p prime<br />
Finally, Corollary 6.17 shows<br />
p prime<br />
min ‖a 3<br />
i‖ 1 − ε 2<br />
i=0, ... ,3<br />
τ ∞<br />
(<br />
X<br />
(a 0 , ... ,a 3 ) (Ê) ) ≤ C · | a 0 · . . . · a 3 | − 1 3 + ε 2 · min<br />
i=0, ... ,3 ‖a i‖ 1 3<br />
∞ .<br />
p .<br />
We assert that the three inequalities together imply the following estimate for<br />
Peyre’s constant τ (a 0, ... ,a 3) = τ (X (a 0, ... ,a 3 ) ),<br />
τ (a 0, ... ,a 3 ) ≤ C 3 · | a 0 · . . . · a 3 | ε 2 · ∏<br />
p prime<br />
min ‖a i‖ 1 3 p<br />
i=0, ... ,3<br />
[ ]<br />
· min ‖a i‖ 1 −<br />
ε<br />
3<br />
∞ · ∏ min ‖a 2<br />
i‖ p .<br />
i=0, ... ,3 i=0, ... ,3<br />
p prime<br />
Indeed, this is trivial in the case τ (a 0, ... ,a 3 )<br />
= 0. Otherwise, X (a 0, ... ,a 3 ) has an<br />
adelic point <strong>and</strong> we may estimate the factors α <strong>and</strong> β by constants as shown in<br />
Section 2. By consequence,<br />
1<br />
τ (a 0,...,a 3 )<br />
≥ 1 C 3<br />
·<br />
[<br />
∏<br />
p prime<br />
] −<br />
1<br />
min ‖a 3<br />
i‖ p ·<br />
i=0,...,3<br />
[<br />
] −<br />
1<br />
min ‖a 3<br />
i‖ ∞<br />
i=0,...,3<br />
[<br />
| a 0 · . . . · a 3 | ε 2 · ∏ min ‖a ] −<br />
ε<br />
2<br />
i‖ p<br />
p prime i=0,...,3
Sec. 6] THE FUNDAMENTAL FINITENESS PROPERTY 327<br />
= 1 C 3<br />
·<br />
= 1 C 3<br />
·<br />
∏<br />
max<br />
p prime i=0,...,3<br />
∥ 1 ∥ 1 ∥∥ 3<br />
a i p<br />
[<br />
| a 0 · . . . · a 3 | ε 2 · ∏<br />
p prime<br />
· max<br />
i=0,...,3<br />
∥ 1 ∥ 1 ∥∥ 3<br />
a i<br />
∞<br />
max<br />
i=0,...,3 a(p) i<br />
] ε<br />
2<br />
H ( ) 1<br />
1 1<br />
a 0<br />
: . . . : 3<br />
a 3<br />
[ ] ε .<br />
| a 0 · . . . · a 3 | ε 2<br />
2 · ∏ max<br />
p prime i=0,...,3 a(p) i<br />
It is obvious that max<br />
i=0,...,3 a(p) i ≤ | a (p)<br />
0<br />
· . . . · a (p)<br />
3 | <strong>and</strong><br />
This shows<br />
∏ | a (p)<br />
0<br />
· . . . · a (p)<br />
3 | = | a 0 · . . . · a 3 | .<br />
p prime<br />
1<br />
τ (a 0,...,a 3 ) ≥ 1 C 3<br />
·<br />
H ( ) 1<br />
1 1<br />
a 0<br />
: . . . : 3<br />
a 3<br />
| a 0 · . . . · a 3 | ε 2 · | a 0 · . . . · a 3 | ε 2<br />
= 1 C 3<br />
· H( 1<br />
a 0<br />
: . . . :<br />
1<br />
a 3<br />
) 1<br />
3<br />
| a 0 · . . . · a 3 | ε . □<br />
6.19. Lemma. –––– Let (a 0 : . . . : a 3 ) ∈ P 3 (É) be any point such that<br />
a 0 ≠ 0, . . . , a 3<br />
∈<br />
≠ 0. Then,<br />
H(a 0 : . . . : a 3 ) ≤ H( 1 1<br />
a 0<br />
: . . . :<br />
a 3<br />
) 3 .<br />
Proof. First, observe that (a 0 : . . . : a 3 ) ↦→ ( )<br />
1<br />
1<br />
a 0<br />
: . . . :<br />
a 3<br />
is a welldefined<br />
map. Hence, we may assume without restriction that a 0 , . . . , a 3<br />
<strong>and</strong> gcd(a 0 , . . . , a 3 ) = 1. This yields H(a 0 : . . . : a 3 ) = max |a i|.<br />
i=0,...,3<br />
On the other h<strong>and</strong>, ( 1 1<br />
a 0<br />
: . . . :<br />
a 3<br />
) = (a 1 a 2 a 3 : . . . : a 0 a 1 a 2 ). Consequently,<br />
H ( 1<br />
a 0<br />
: . . . :<br />
1<br />
a 3<br />
) ≤ [ max<br />
i=0,...,3 |a i|] 3 = H(a 0 : . . . : a 3 ) 3 .<br />
From this, the asserted inequality emerges when the roles of a i <strong>and</strong> 1 a i<br />
are interchanged.<br />
□<br />
6.20. Corollary. –––– Let a 0 , . . . , a 3 ∈such that gcd(a 0 , . . . , a 3 ) = 1. Then,<br />
|a 0 · . . . · a 3 | ≤ H ( 1<br />
a 0<br />
: . . . :<br />
1<br />
a 3<br />
) 12.<br />
Proof. Observe that | a 0 · . . . · a 3 | ≤ max | a i| 4 = H(a 0 : . . . : a 3 ) 4 <strong>and</strong> apply<br />
i=0,...,3<br />
Lemma 6.19.<br />
□
328 ON THE SMALLEST POINT ON A DIAGONAL CUBIC SURFACE [Chap. X<br />
6.21. Theorem. –––– For each ε > 0, there exists a constant C (ε) > 0 such that,<br />
for all (a 0 , . . . , a 3 ) ∈ (\{0}) 4 ,<br />
1<br />
τ ≥ C (ε) · H( ) 1<br />
1 1<br />
(a 0,...,a 3 ) a 0<br />
: . . . : 3<br />
−ε<br />
a 3<br />
.<br />
Proof. We may assume that gcd(a 0 , . . . , a 3 ) = 1. Then, by Proposition 6.18,<br />
1<br />
τ ≥ C (ε) · H( ) 1<br />
1 1<br />
a 0<br />
: . . . : 3<br />
a 3<br />
.<br />
(a 0,...,a 3 )<br />
| a 0 · . . . · a 3 | ε 12<br />
(<br />
Corollary 6.20 yields | a 0 · . . . · a 3 | ε 12 ≤ H 1<br />
)<br />
1 ε.<br />
a 0<br />
: . . . :<br />
a 3<br />
6.22. Corollary (Fundamental finiteness). —– For each T > 0, there are only<br />
finitely many diagonal cubic surfaces X (a 0,...,a 3 ) : a 0 x 3 0 + . . . +a 3 x 3 3 = 0 in P 3Ésuch<br />
that τ (a 0,...,a 3) > T .<br />
Proof. This is an immediate consequence of the comparison to the naive height<br />
established in Theorem 6.21.<br />
□<br />
□
CHAPTER XI<br />
THE DIOPHANTINE EQUATION<br />
x 4 + 2y 4 = z 4 + 4w 4∗<br />
Hash, x. There is no definition for this word—nobody knows what hash is.<br />
AMBROSE BIERCE: The Devil’s Dictionary (1906)<br />
1. Introduction<br />
1.1. –––– An algebraic curve C of genus g > 1 admits at most a finite number<br />
ofÉ-<strong>rational</strong> <strong>points</strong>. On the other h<strong>and</strong>, for genus one curves, #C (É) may be<br />
zero, finite non-zero, or infinite. For genus zero curves, one automatically has<br />
#C (É) = ∞ as soon as C (É) ≠ ∅.<br />
1.2. –––– In higher dimensions, there is a conjecture, due to S. Lang, stating<br />
that if X is a variety of general type over a number field then all but finitely many<br />
of its <strong>rational</strong> <strong>points</strong> are contained in the union of closed subvarieties which are<br />
not of general type (cf. Conjecture VI.2.2). On the other h<strong>and</strong>, abelian varieties<br />
(as well as, e.g., elliptic <strong>and</strong> bielliptic surfaces) behave like genus one curves.<br />
I.e., #X (É) may be zero, finite non-zero, or infinite. Finally, <strong>rational</strong> <strong>and</strong> ruled<br />
varieties comport in the same way as curves of genus zero in this respect.<br />
This list does not yet exhaust the classification of algebraic surfaces, to say<br />
nothing of dimension three or higher. In particular, the following problem is<br />
still open.<br />
1.3. Problem. –––– Does there exist a K3 surface X overÉwhich has a finite<br />
non-zero number ofÉ-<strong>rational</strong> <strong>points</strong>? I.e., such that 0 < #X (É) < ∞?<br />
1.4. Remark. –––– This question was posed by Sir P. Swinnerton-Dyer as<br />
Problem/Question 6.a) in the problem session to the workshop [Poo/T]. We are<br />
not able to give an answer to it.<br />
(∗) This chapter collects material from the articles:<br />
The Diophantine equation x 4 + 2y 4 = z 4 + 4w 4 , Math. Comp. 75(2006), 935–940 <strong>and</strong><br />
The Diophantine equation x 4 + 2y 4 = z 4 + 4w 4 — a number of improvements, Preprint,<br />
both joint with A.-S. Elsenhans.
330 THE DIOPHANTINE EQUATION x 4 + 2y 4 = z 4 + 4w 4 [Chap. XI<br />
1.5. –––– One possible c<strong>and</strong>idate for a K3 surface with the property<br />
0 < #X (É) < ∞ is given by the following<br />
Problem. Find a third point on the projective surface X ⊂ P 3 defined by<br />
x 4 + 2y 4 = z 4 + 4w 4 .<br />
1.6. Remarks. –––– i) Problem 1.5 is also due to Sir P. Swinnerton-Dyer<br />
[Poo/T, Problem/Question 6.c)]. It was raised, in particular, during his<br />
talk [S-D04, very end of the article] at the Göttingen Mathematisches Institut<br />
on June 2 nd , 2004.<br />
ii) x 4 +2y 4 = z 4 +4w 4 is a homogeneous quartic equation. It, therefore, defines<br />
a K3 surface X in P 3 . As trivial solutions of the equation, we consider those<br />
corresponding to theÉ-<strong>rational</strong> <strong>points</strong> (1:0:1:0) <strong>and</strong> (1:0:(−1):0).<br />
iii) Our main result is the following theorem which contains an answer to<br />
Problem 1.5.<br />
1.7. Theorem. –––– The diagonal quartic surface X in P 3 given by<br />
x 4 + 2y 4 = z 4 + 4w 4<br />
(∗)<br />
admits precisely tenÉ-<strong>rational</strong> <strong>points</strong> having integral coordinates within the hypercube<br />
|x|, |y|, |z|, |w| < 10 8 .<br />
These are (±1:0:±1:0) <strong>and</strong> (±1 484 801:±1 203 120:±1169407:±1157520).<br />
1.8. Remark. –––– This result does clearly not exclude the possibility that<br />
#X (É) is actually finite. It might indicate, however, that a proof for this<br />
property is deeper than one originally hoped for.<br />
2. Congruences<br />
2.1. –––– It seems natural to first try to underst<strong>and</strong> the congruences<br />
x 4 + 2y 4 ≡ z 4 + 4w 4 (mod p) (†)<br />
modulo some prime number p. For p = 2 <strong>and</strong> 5, one finds that all primitive<br />
solutions insatisfy<br />
a) x <strong>and</strong> z are odd,<br />
b) y <strong>and</strong> w are even,<br />
c) y is divisible by 5.
Sec. 2] CONGRUENCES 331<br />
For other primes, it follows from the Weil conjectures, proven by P. Deligne<br />
[Del], that the number of solutions of the congruence (†) is<br />
#C X (p) = 1 + (p − 1)(p 2 + p + 1 + E) = p 3 + E(p − 1) .<br />
Here, E is an error-term which may be estimated by |E| ≤ 21p.<br />
Indeed, consider the projective variety X overÉdefined by (∗). It has good reduction<br />
at every prime p ≠ 2. Therefore, [Del, Théorème (8.1)] may be applied<br />
to the reduction X p . This yields #X p (p) = p 2 + p + 1 + E <strong>and</strong> |E| ≤ 21p.<br />
We note that dim H 2 (X ,Ê) = 22 for every complex surface X of type K3 [Bv,<br />
p. 98].<br />
2.2. –––– Another question of interest is to count the numbers of solutions<br />
to the congruences x 4 + 2y 4 ≡ c (mod p) <strong>and</strong> z 4 + 4w 4 ≡ c (mod p) for a<br />
certain c ∈.<br />
This means to count thep-<strong>rational</strong> <strong>points</strong> on the affine plane curves C l c <strong>and</strong> C r c<br />
defined overp by x 4 + 2y 4 = c <strong>and</strong> z 4 + 4w 4 = c, respectively. If p ∤ c<br />
<strong>and</strong> p ≠ 2 then these are smooth curves of genus three.<br />
By the work of André Weil [We48, Corollaire 3 du Théorème 13], the numbers<br />
ofp-<strong>rational</strong> <strong>points</strong> on their projectivizations are given by<br />
#C l c (p) = p + 1 + E l <strong>and</strong> #C r c (p) = p + 1 + E r ,<br />
where the error-terms can be bounded by |E l |, |E r | ≤ 6 √ p. There may be up<br />
to fourp-<strong>rational</strong> <strong>points</strong> on the infinite line. For our purposes, it suffices to<br />
notice that both congruences admit a number of solutions which is close to p.<br />
The case p|c, p ≠ 2, is slightly different since it corresponds to the case of a<br />
reducible curve. The congruence x 4 + ky 4 ≡ 0 (mod p) admits only the trivial<br />
solution if (−k) is not a biquadratic residue modulo p. Otherwise, it has exactly<br />
1 + (p − 1) gcd(p − 1, 4) solutions.<br />
Finally, if p = 2 then #C l 0 (2) = #C l 1 (2) = #C r 0 (2) = #C r 1 (2) = 2.<br />
2.3. Remark. –––– The number of solutions of the congruence (†) is<br />
#C X (p) = ∑ #Cc l (p) · #Cc r (p).<br />
c∈p<br />
Hence, the formulas just mentioned yield an elementary estimate for that count.<br />
They show once more that the dominating term is p 3 . The estimate for the<br />
error is, however, less sharp than the one obtained via the more sophisticated<br />
methods in 2.1.
332 THE DIOPHANTINE EQUATION x 4 + 2y 4 = z 4 + 4w 4 [Chap. XI<br />
3. Naive methods<br />
3.1. –––– The most naive method to search for solutions of (∗) is probably<br />
the following. Start with the set<br />
{(x, y, z, w) ∈|0≤x, y, z, w ≤ N }<br />
<strong>and</strong> test the equation for every quadruple.<br />
Obviously this method requires about N 4 steps. It can be accelerated using the<br />
congruence conditions for primitive solutions noticed above.<br />
3.2. –––– A somewhat better method is to start with the set<br />
{x 4 + 2y 4 − 4w 4 | x, y, w ∈, 0 ≤ x, y, w ≤ N }<br />
<strong>and</strong> to search for fourth powers. This set has about N 3 Elements, <strong>and</strong> the<br />
algorithm takes about N 3 steps. Again, it can be sped up by the above congruence<br />
conditions for primitive solutions. We used this approach for a trial-run<br />
with N = 10 4 .<br />
An interesting aspect of this algorithm is the optimization by further congruences.<br />
Suppose x <strong>and</strong> y are fixed, then about one half or three-quarter of the<br />
values for w are no solutions to the congruence modulo a new prime. Following<br />
this way, one can find more congruences for w <strong>and</strong> the size of the set may<br />
be reduced by a constant factor.<br />
4. An algorithm to efficiently search for solutions<br />
i. The basic idea. —<br />
4.1. –––– We need to compute the intersection of two sets<br />
{x 4 + 2y 4 | x, y ∈, 0 ≤ x, y ≤ N } ∩ {z 4 + 4w 4 | z, w ∈, 0 ≤ z, w ≤ N } .<br />
Both have about N 2 elements.<br />
It is a st<strong>and</strong>ard problem in Computer Science to find the intersection of two<br />
sets which both fit into memory. Using the congruence conditions modulo 2<br />
<strong>and</strong> 5, one can reduce the size of the first set by a factor of 20 <strong>and</strong> the size of the<br />
second set by a factor of 4.
Sec. 4] AN ALGORITHM TO EFFICIENTLY SEARCH FOR SOLUTIONS 333<br />
ii. Some details. —<br />
4.2. –––– The two sets described above are too big, at least for our computers<br />
<strong>and</strong> interesting values of N . Therefore, we introduce a prime number p p which<br />
we call the page prime.<br />
Define the sets<br />
<strong>and</strong><br />
L c := {x 4 + 2y 4 | x, y ∈, 0 ≤ x, y ≤ N , x 4 + 2y 4 ≡ c (mod p p )}<br />
R c := {z 4 + 4w 4 | z, w ∈, 0 ≤ z, w ≤ N , z 4 + 4w 4 ≡ c (mod p p )} .<br />
This means, the intersection problem is divided into p p pieces <strong>and</strong> the sets L c<br />
<strong>and</strong> R c fit into the computer’s memory if p p is big enough. We worked with<br />
N = 2.5 · 10 6 <strong>and</strong> chose p p = 30 011.<br />
For every value of c, our program computes L c <strong>and</strong> stores this set in a hash table.<br />
Then, it determines the elements of R c <strong>and</strong> looks them up in the table. Assuming<br />
uniform hashing, the expected running-time of this algorithm is O(N 2 ).<br />
4.3. Remark –––– An important further aspect of this approach is that the<br />
problem may be attacked in parallel on several machines. The calculations<br />
for one particular value of c are independent of the analogous calculations for<br />
another one. Thus, it is possible, say, to let c run from 0 to (p p − 1)/2 on one<br />
machine <strong>and</strong>, at the same time, from (p p + 1)/2 to (p p − 1) on another.<br />
iii. Some more details. —<br />
iii.1. The page prime. —<br />
4.4. –––– For each value of c, it is necessary to find the solutions of the congruences<br />
x 4 +2y 4 ≡ c (mod p p ) <strong>and</strong> z 4 +4w 4 ≡ c (mod p p ) in an efficient manner.<br />
We do this in a rather naive way by letting y (w) run from 0 to p p − 1.<br />
For each value of y (w), we compute x 4 (z 4 ). Then, we extract the fourth root<br />
modulo p p .<br />
Note that the page prime fulfills p p ≡ 3 (mod 4). Hence, the fourth roots of<br />
unity modulo p are just ±1 <strong>and</strong>, therefore, a fourth root modulo p p , if it exists,<br />
is unique up to sign. This makes the algorithm easier to implement.
334 THE DIOPHANTINE EQUATION x 4 + 2y 4 = z 4 + 4w 4 [Chap. XI<br />
4.5. –––– Actually, we do not execute any modular powering operation or even<br />
computation of fourth roots in the lion’s share of the running time. For more<br />
efficiency, all fourth powers <strong>and</strong> all fourth roots modulo p p are computed <strong>and</strong><br />
stored in an array during an initialization step. Thus, the main speed limitation<br />
to find all solutions to a congruence modulo p p is, in fact, the time it takes to<br />
look up values stored in the machine’s main memory.<br />
iii.2. Hashing. —<br />
4.6. –––– We do not compute L c <strong>and</strong> R c directly, because this would require<br />
the use of multi precision integers within the inner loop. Instead, we choose<br />
two other primes, the hash prime p h <strong>and</strong> the control prime p c , which fit into<br />
the 32-bit registers of our computers. All computations are done modulo p h<br />
<strong>and</strong> p c .<br />
More precisely, for each pair (x, y) considered, the expression<br />
(<br />
(x 4 + 2y 4 ) mod p h<br />
)<br />
defines its position in the hash table. In other words, we hash pairs (x, y)<br />
whereas (x, y) ↦→ ( (x 4 + 2y 4 ) mod p h<br />
)<br />
plays the role of the hash function.<br />
For each pair (x, y), we write two entries into the hash table, namely the value<br />
of ( (x 4 + 2y 4 ) mod p c<br />
)<br />
<strong>and</strong> the value of y.<br />
In the main computation, we worked with the numbers p h = 25 000 009 for the<br />
hash prime <strong>and</strong> p c = 400 000 009 for the control prime.<br />
4.7. –––– Note that, when working with a particular value of c, there are<br />
around p p pairs ((x mod p p ), (y mod p p )) which fulfill the required congruence<br />
Therefore, approximately<br />
x 4 + 2y 4 ≡ c (mod p p ) .<br />
p p · (N /2<br />
p p<br />
· N /10<br />
p p<br />
) =<br />
N 2<br />
20p p<br />
values will be written into the table. For our choices,<br />
N 2<br />
20p p<br />
≈ 10 412 849 which<br />
means that the hash table will get approximately 41.7% filled.<br />
Like for many other rules, there is an exception to this one. If c = 0 then approximately<br />
1 + (p p − 1) gcd(p p − 1, 4) pairs ((x mod p p ), (y mod p p )) may<br />
satisfy the congruence<br />
x 4 + 2y 4 ≡ 0 (mod p p ) .<br />
As p p ≡ 3 (mod 4) this is not more than 2p p − 1 <strong>and</strong> the hash table will be<br />
filled not more than about 83.3%.
Sec. 5] GENERAL FORMULATION OF THE METHOD 335<br />
4.8. –––– To resolve collisions within the hash table, we use an open addressing<br />
method. We are not particularly afraid of clustering <strong>and</strong> choose linear probing.<br />
We feel free to use open addressing as, thanks to the Weil conjectures, we<br />
have a priori estimates available for the load factor.<br />
4.9. –––– The program makes frequent use of fourth powers modulo p h <strong>and</strong> p c .<br />
Again, we compute these data in the initialization part of our program <strong>and</strong> store<br />
them in arrays, once <strong>and</strong> for all.<br />
5. General formulation of the method<br />
5.1. –––– The method described in the previous section is actually a systematic<br />
method to search for solutions of a Diophantine equation. It works efficiently<br />
when the equation is of the form<br />
f (x 1 , . . . , x n ) = g(y 1 , . . . , y m ) .<br />
We find all solutions which are contained within the (n + m)-dimensional cube<br />
{(x 1 , . . . , x n , y 1 , . . . , y m ) ∈n+m | | x i |, |y i | ≤ B} .<br />
The expected running-time of the algorithm is O(B max{n,m} ).<br />
5.2. –––– The basic idea may be formulated as follows.<br />
Algorithm H.<br />
i) Evaluate f on all <strong>points</strong> of the n-dimensional cube<br />
Store the values within a set L.<br />
ii) Evaluate g on all <strong>points</strong> of the cube<br />
{(x 1 , . . . , x n ) ∈n | |x i | ≤ B} .<br />
{(y 1 , . . . , y m ) ∈m | |y i | ≤ B}<br />
of dimension m. For each value start a search in order to find out whether<br />
it occurs in L. When a coincidence is detected, reconstruct the corresponding<br />
values of x 1 , . . . , x n <strong>and</strong> output the solution.<br />
5.3. Remarks. –––– a) In fact, we are interested in the very particular Diophantine<br />
equation<br />
x 4 + 2y 4 = z 4 + 4w 4<br />
which was suggested by Sir Peter Swinnerton-Dyer.
336 THE DIOPHANTINE EQUATION x 4 + 2y 4 = z 4 + 4w 4 [Chap. XI<br />
b.i) In the form stated above, the main disadvantage of Algorithm H is that it<br />
requires an enormous amount of memory. Actually, the set L is too big to be<br />
stored in the main memory even of our biggest computers, already when the<br />
value of B is only moderately large.<br />
For that reason, we introduced the idea of paging. We choose a page prime p p <strong>and</strong><br />
work with the sets L r := {s ∈ L | s ≡ r (mod p p )} for r = 0, . . . , p p − 1, separately.<br />
At the cost of some more time spent on initializations, this yields<br />
a reduction of the memory space required by a factor of 1 p p<br />
.<br />
ii) The sets L r were implemented in the form of a hash table with open addressing.<br />
iii) It is possible to achieve a further reduction of the running-time <strong>and</strong> the memory<br />
required by making use of some obvious congruence conditions modulo 2<br />
<strong>and</strong> 5.<br />
5.4. –––– The goal of the remainder of this chapter is to describe an improved<br />
implementation of Algorithm H which we used in order to find<br />
all solutions of x 4 + 2y 4 = z 4 + 4w 4 contained within the hypercube<br />
{(x, y, z, w) ∈4 | |x|, |y|, |z|, |w| ≤ 10 8 }.<br />
6. Improvements I – More congruences<br />
6.1. –––– The most obvious way to further reduce the size of the sets L r <strong>and</strong><br />
to increase the speed of Algorithm H is to find further congruence conditions<br />
for solutions <strong>and</strong> evaluate f <strong>and</strong> g only on <strong>points</strong> satisfying these conditions.<br />
As the equation, we are interested in, is homogeneous, it is sufficient to restrict<br />
consideration to primitive solutions.<br />
6.2. –––– It should be noticed, however, that this idea is subject to strict limitations.<br />
If we were using the most naive O(B n+m )-algorithm then, for more or less<br />
every l ∈Æ, the congruence f (x 1 , . . . , x n ) ≡ g(y 1 , . . . , y m ) (mod l) caused a<br />
reduction of the number of (n + m)-tuples to be checked. For Algorithm H,<br />
however, the situation is by far less fortunate.<br />
One may gain something only if there are residue classes (r mod l) which are<br />
represented by f , but not by g, or vice versa. Values, the residue class of which<br />
is not represented by g, do not need to be stored into L r . Values, the residue<br />
class of which is not represented by f , do not need to be searched for.<br />
Unfortunately, if l is prime <strong>and</strong> not very small then the Weil conjectures ensure<br />
that all residue classes modulo l are represented by both f <strong>and</strong> g. In this case, the<br />
idea fails completely. The same is, however, not true for prime powers l = p k .<br />
Hensel’s Lemma does not work when all partial derivatives ∂f<br />
∂x i<br />
(x 1 , . . . , x n ),
Sec. 6] IMPROVEMENTS I – MORE CONGRUENCES 337<br />
respectively ∂g<br />
∂y i<br />
(y 1 , . . . , y m ), are divisible by p. This makes it possible that<br />
certain residue classes (r mod p k ) are not representable although (r mod p) is.<br />
i. The prime 5. Congruences modulo 625. —<br />
6.3. –––– In the algorithm described in the previous chapter, we made use<br />
of the fact that y is always divisible by 5. However, at this point, one can<br />
do a lot better. When one takes into consideration that a 4 ≡ 1 (mod 5) for<br />
every a ∈not divisible by 5, a systematic inspection shows that there are<br />
actually two cases.<br />
Either, 5|w. Then, 5∤x <strong>and</strong> 5∤z. Or, otherwise, 5|x. Then, 5∤z <strong>and</strong> 5∤w.<br />
Note that, in the latter case, one indeed has z 4 + 4w 4 ≡ 1 + 4 ≡ 0 (mod 5).<br />
6.4. –––– The Case 5|w. We call this case “N” <strong>and</strong> use the letter N at a<br />
prominent position in the naming of the relevant files of source code. N st<strong>and</strong>s<br />
for “normal”. To consider this case as the ordinary one is justified by the fact<br />
that all primitive solutions known actually belong to it. Note, however, that<br />
we have no theoretical reason to believe that this case should in whatever sense<br />
be better than the other one.<br />
In case N, we rearrange the equation to f N (x, z) = g N (y, w) where<br />
f N (x, z) := x 4 − z 4 <strong>and</strong> g N (y, w) := 4w 4 − 2y 4 .<br />
As y <strong>and</strong>w are both divisible by 5, we get g N (y, w) = 4w 4 −2y 4 ≡ 0 (mod 625).<br />
Consequently, f N (x, z) ≡ 0 (mod 625).<br />
This yields an enormous reduction of the set L r . To see this, recall 5∤x <strong>and</strong> 5∤z.<br />
That means, for x, there are precisely ϕ(625) possibilities in/625. Further,<br />
for each such value, the congruence z 4 ≡ x 4 (mod 625) may not have<br />
more than four solutions. All in all, there are 4 · ϕ(625) = 2 000 possible pairs<br />
(x, z) ∈ (/625) 2.<br />
Further, these pairs are very easy to find, computationally. The fourth<br />
roots of unity modulo 625 are ±1 <strong>and</strong> ±182. For each x ∈/625∗ , put<br />
z := (±x mod 625) <strong>and</strong> z := (±182x mod 625).<br />
We store the values of f N into the set L r . Only 2 000 out of 625 2 values (0.512%)<br />
need to be computed <strong>and</strong> stored. Then, each value of g N is looked up in L r .<br />
Here, as y <strong>and</strong> w are both divisible by 5, only one value out of 25 (4%) needs to<br />
be computed <strong>and</strong> searched for.<br />
6.5. –––– The Case 5|x. We call this case “S” <strong>and</strong> use the letter S at a prominent<br />
position in the naming of the relevant files of source code. S st<strong>and</strong>s for<br />
“Sonderfall” which means “exceptional case”. It is not known whether there<br />
exists a solution belonging to case S.
338 THE DIOPHANTINE EQUATION x 4 + 2y 4 = z 4 + 4w 4 [Chap. XI<br />
Here, we simply interchange both sides of the equation. Define<br />
f S (z, w) := z 4 + 4w 4 <strong>and</strong> g S (x, y) := x 4 + 2y 4 .<br />
As x <strong>and</strong> y are divisible by 5, we get x 4 + 2y 4 ≡ 0 (mod 625) <strong>and</strong>, therefore,<br />
z 4 + 4w 4 ≡ 0 (mod 625).<br />
Again, this congruence allows only 4 · ϕ(625) = 2 000 solutions<br />
(z, w) ∈ (/625) 2<br />
<strong>and</strong> these pairs are easily computable, too. The fourth roots of (−4)<br />
in/625are ±181 <strong>and</strong> ±183. For each x ∈/625∗ , one has to consider<br />
z := (±181x mod 625) <strong>and</strong> z := (±183x mod 625).<br />
We store the values of f S into the set L r . Then, we search through L r for the<br />
values of g S . As above, only 2 000 out of 625 2 values need to be computed<br />
<strong>and</strong> stored <strong>and</strong> one value out of 25 needs to be computed <strong>and</strong> searched for.<br />
ii. The prime 2. —<br />
6.6. –––– Any primitive solution is of the form that x <strong>and</strong> z are odd while y<br />
<strong>and</strong> w are even.<br />
6.7. –––– In case S, there is no way to do better than that as both f S <strong>and</strong> g S<br />
represent (r mod 2 k ) for k ≥ 4 if <strong>and</strong> only if r ≡ 1 (mod 16).<br />
IncaseN, thesituationissomewhatbetter. g N (y, w) = 4w 4 −2y 4 isalwaysdivisible<br />
by 32 while f N (x, z) = x 4 −z 4 ≡ 0 (mod 32), as may be seen by inspecting<br />
the fourth roots of unity modulo 32, implies the condition x ≡ ±z (mod 8).<br />
This may be used to halve the size of L r .<br />
iii. The prime 3. —<br />
6.8. –––– Looking for further congruence conditions, a primitive solution<br />
must necessarily satisfy, we did not find any reason to distinguish more cases.<br />
But there are a few more congruences which we used in order to reduce the size<br />
of the sets L r .<br />
To explain them, let us first note two theorems on binary quadratic forms.<br />
They may both be easily deduced from [H/W, Theorems 246 <strong>and</strong> 247].<br />
6.9. Theorem. –––– The quadratic formsq 1 (a, b) := a 2 +b 2 , q 2 (a, b) := a 2 −2b 2 ,<br />
<strong>and</strong> q 3 (a, b) := a 2 + 2b 2 admit the property below.<br />
Suppose n 0 := q i (a 0 , b 0 ) is divisible by a prime p which is not represented by q i .<br />
Then, p|a 0 <strong>and</strong> p|b 0 .
Sec. 6] IMPROVEMENTS I – MORE CONGRUENCES 339<br />
6.10. Theorem. –––– A prime number p is representedbyq 1 , q 2 , or q 3 , respectively,<br />
if <strong>and</strong> only if (0 mod p) is represented in a non-trivial way. In particular,<br />
i) p is represented by q 1 if <strong>and</strong> only if p = 2 or p ≡ 1 (mod 4).<br />
ii) p is represented by q 2 if <strong>and</strong> only if p = 2 or ( 2<br />
p) = 1. The latter means<br />
p ≡ 1, 7 (mod 8).<br />
iii) p is represented by q 3 if <strong>and</strong> only if p = 2 or ( −2<br />
p<br />
to p ≡ 1, 3 (mod 8).<br />
6.11. Remark –––– There is the obvious asymptotic estimate<br />
Further,<br />
♯{q i (a, b) | a, b ∈, q i (a, b) ∈È, q i (a, b) ≤ n} ∼<br />
) = 1. The latter is equivalent<br />
n<br />
2 logn .<br />
n<br />
♯{q i (a, b) | a, b ∈, |q i (a, b)| ≤ n} ∼ C i √<br />
logn<br />
where C 1 , C 2 , <strong>and</strong> C 3 are constants which can be expressed explicitly by Euler<br />
products. (For q 1 , this is worked out in [Brü, Satz (1.8.2)]. For the other<br />
forms, J. Brüdern’s argument works in the same way without essential changes.)<br />
6.12. Congruences modulo 81. ––––<br />
In case N,<br />
g N (y, w) = (2w 2 ) 2 − 2(y 2 ) 2 = q 2 (2w 2 , y 2 )<br />
where q 2 does not represent the prime 3. Therefore, if 3|g N (y, w) then 3|2w 2<br />
<strong>and</strong> 3|y 2 which implies y <strong>and</strong> w are both divisible by 3. By consequence, if<br />
3|g N (y, w) then, automatically, 81|g N (y, w).<br />
If 3| f N (x, z) but 81∤ f N (x, z) then f N (x, z) does not need to be stored into L r .<br />
Further, if 3|x <strong>and</strong> 3|z then f N (x, z) does not need to be stored, either, as it<br />
cannot lead to a primitive solution. This reduces the size of the set L r by a<br />
factor of 1 + 4 · 1<br />
9 3 (1 − 1 ) = 131 ≈ 53.9%.<br />
3 81 243<br />
In case S, the situation is the other way round.<br />
f S (z, w) = (z 2 ) 2 + (2w 2 ) 2 = q 1 (z 2 , 2w 2 )<br />
<strong>and</strong> q 1 does not represent the prime 3. Therefore, if 3| f S (z, w) then 3|z 2 <strong>and</strong><br />
3|2w 2 which implies that z <strong>and</strong> w are both divisible by 3 <strong>and</strong> 81| f S (z, w).<br />
We use this in order to reduce the time spent on reading. If 3|g S (x, y) but<br />
81∤g S (x, y) or if 3|x <strong>and</strong> 3|y then g S (x, y) does not need to be searched for.<br />
Although modular operations are not at all fast, the reduction of the number of<br />
attempts to read by 53.9% is highly noticeable.
340 THE DIOPHANTINE EQUATION x 4 + 2y 4 = z 4 + 4w 4 [Chap. XI<br />
iv. Some more hypothetical improvements. —<br />
6.13. –––– i) In the argument for case N given above, p = 3 might be replaced<br />
by any other prime p ≡ 3, 5 (mod 8).<br />
In case S, the same argument as above works for every prime p ≡ 3 (mod 8).<br />
For primes p ≡ 5 (mod 8), the strategy could be reversed. q 3 is a binary<br />
quadratic form which represents (0 mod p) only in the trivial manner. Therefore,<br />
if p|g S (x, y) then p|x <strong>and</strong> p|y. It is unnecessary to store f S (z, w) if p|z<br />
<strong>and</strong> p|w or if p| f S (z, w) but p 4 ∤ f S (z, w).<br />
i ′ ) Each argument mentioned may be extended to some primes p ≡ 1 (mod 8).<br />
For example, in case N, what is actually needed is that 2 is not a fourth power<br />
modulo p. This is true, e.g., for p = 17, 41, <strong>and</strong> 97, but not for p = 73 <strong>and</strong> 89.<br />
ii) f N <strong>and</strong> f S do not represent the residue classes of 6, 7, 10, <strong>and</strong> 11 modulo 17.<br />
g N <strong>and</strong> (−g S ) do not represent 1, 3, <strong>and</strong> 9 modulo 13. This could be used to<br />
reduce the load for writing as well as reading.<br />
6.14. Remarks. –––– a) We did not implement these improvements as it seems<br />
the gains would be marginal or the cost of additional computations would<br />
even dominate the effect. It is, however, foreseeable that these congruences will<br />
eventually become valuable when the speed of the CPU’s available will continue<br />
to grow faster than the speed of memory. Observe that alone the congruences<br />
noticed in a) could reduce the amount of data to be stored into L to a size<br />
asymptotically less than εB 2 for any ε > 0.<br />
b) For every prime p different from 2, 5, 13, <strong>and</strong> 17, the quartic forms f N , g N ,<br />
f S , <strong>and</strong> g S represent all residue classes modulo p. This means, ii) may not be<br />
carried over to any further primes.<br />
This can be seen as follows. Let b be equal to f N , f S , g N , or g S . (0 mod p) is<br />
represented by b, trivially. Otherwise, b(x, y) = r defines an affine curve<br />
C r of genus three with at most four <strong>points</strong> on the infinite line. The Weil<br />
conjectures [We48, Corollaire 3 du Théorème 13] imply that [(p+1−6 √ p)−4]<br />
is a lower bound for the number ofp-<strong>rational</strong> <strong>points</strong> on C r . This is a positive<br />
number as soon as p ≥ 43. In this case, every residue class (r mod p) is<br />
represented, at least, once.<br />
For the remaining primes up to p = 41, an experiment shows that all residue<br />
classes modulo p are represented by f N , f S , g N , as well as g S .
Sec. 7] IMPROVEMENTS II – ADAPTION TO OUR HARDWARE 341<br />
7. Improvements II – Adaption to our hardware<br />
i. A 64 bit based implementation of the algorithm. —<br />
7.1. –––– We migrated the implementation of Algorithm H from a 32 bit<br />
processor to a 64 bit processor. This means, the new hardware supports addition<br />
<strong>and</strong> multiplication of 64 bit integers. Even more, every operation on (unsigned)<br />
integers is automatically modulo 2 64 .<br />
From this, various optimizations of the implementation described in Section 4<br />
are almost compelling. The basic idea is that 64 bits should be enough to<br />
define hash value <strong>and</strong> control value by selection of bits instead of using (notoriously<br />
slow) modular operations. Hash value <strong>and</strong> control value are two integers<br />
significantly less than 2 32 which should be independent on each other.<br />
Note, however, that the congruence conditions modulo 2 imposed imply that<br />
x 4 ≡ z 4 ≡ 1 (mod 16) <strong>and</strong> 2y 4 ≡ 4w 4 ≡ 0 (mod 16). This means, the four<br />
least significant bits of f <strong>and</strong> g may not be used as they are always the same.<br />
7.2. –––– The description of the algorithm below is based on case S, case N<br />
being completely analogous.<br />
Algorithm H64.<br />
I. Initialization. Fix B := 10 8 . Initialize a hash table of 2 27 = 134 217 728<br />
integers, each being 32 bit long. Fix the page prime p p := 200 003.<br />
Further, define two functions, the hash function h <strong>and</strong> the control function c,<br />
which map 64 bit integers to 27 bit integers <strong>and</strong> 31 bit integers, respectively, by<br />
selecting certain bits. Do not use any of the bits twice to ensure h <strong>and</strong> c are<br />
independent on each other <strong>and</strong> do not use the four least significant bits.<br />
II) Loop. Let r run from 0 to p p − 1 <strong>and</strong> execute steps A. <strong>and</strong> B. for each r.<br />
A. Writing. Build up the hash table, which is meant to encode the set L r ,<br />
as follows.<br />
a) Find all pairs (z, w) of non-negative integers less than or equal to B which<br />
satisfy z 4 + 4w 4 ≡ r (mod p p ) <strong>and</strong> all the congruence-conditions for primitive<br />
solutions, listed above. (Make systematic use of the Chinese remainder theorem.)<br />
b) Execute steps i) <strong>and</strong> ii) below for each such pair.<br />
i) Evaluate f S (z, w) := (z 4 + 4w 4 mod 2 64 ).<br />
ii) Use the hash value h( f S (z, w)) <strong>and</strong> linear probing to find a free place in the<br />
hash table <strong>and</strong> store the control value c( f S (z, w)) there.
342 THE DIOPHANTINE EQUATION x 4 + 2y 4 = z 4 + 4w 4 [Chap. XI<br />
B. Reading. Search within the hash table, as follows.<br />
a) Find all pairs (x, y) of non-negative integers less than or equal to B which satisfy<br />
x 4 + 2y 4 ≡ r (mod p p ) <strong>and</strong> all the congruence conditions for primitive solutions,<br />
listed above. (Make systematic use of the Chinese remainder theorem.)<br />
b) Execute steps i) <strong>and</strong> ii) below for each such pair.<br />
i) Evaluate g S (x, y) := (x 4 + 2y 4 mod 2 64 ) on all <strong>points</strong> found in step a).<br />
ii) Search for the control value c(g S (x, y)) in the hash table, starting at the<br />
hash value h(g S (x, y)) <strong>and</strong> using linear probing, until a free position is found.<br />
Report all hits <strong>and</strong> the corresponding values of x <strong>and</strong> y.<br />
7.3. Remarks (Some details of the implementation). —– i) The fourth powers<br />
<strong>and</strong> fourth roots modulo p p are computed during the initialization part of the<br />
program <strong>and</strong> stored into arrays because arithmetic modulo p p is slower than<br />
memory access.<br />
ii) The control value is limited to 31 bits as it is implemented as a signed integer.<br />
We use the value (−1) as a marker for an unoccupied place in the hash table.<br />
iii) In contrast to our previous programs, we do not precompute large tables of<br />
fourth powers modulo 2 64 because an access to these tables is slower than the<br />
execution of two multiplications in a row (at least on our computer).<br />
iv) It is the impact of the congruences modulo 625, 8, <strong>and</strong> 81, described above,<br />
that the set of pairs (y, w) [(x, y)] to be read is significantly bigger than the<br />
set of pairs (x, z) [(z, w)] to be written. They differ actually by a factor of<br />
625 2<br />
· 243<br />
2000·25 112<br />
6252<br />
· 2 ≈ 33.901 in case N <strong>and</strong> · 112<br />
2 000·25 243<br />
≈ 3.601 in case S.<br />
As a consequence of this, only a small part of the running-time is spent on writing.<br />
The lion’s share is spent on unsuccessful searches within L.<br />
7.4. Remarks (Post-processing). —– i) Most of the hits found in the hash table<br />
actually do not correspond to solutions of the Diophantine equation. Hits indicate<br />
only a similarity of bit-patterns. Thus, for each pair of x <strong>and</strong> y reported, one<br />
needs to check whether a suitable pair of z <strong>and</strong> w does exist. We do this by recomputing<br />
z 4 +4w 4 for all z <strong>and</strong> w which fulfill the given congruence conditions<br />
modulo p p <strong>and</strong> powers of the small primes.<br />
Although this method is entirely primitive, only about 3% of the total runningtime<br />
is actually spent on post-processing. One reason for this is that postprocessing<br />
is not called very often, on average only once on about five pages.<br />
For those pages, the writing part of the algorithm needs to be recapitulated.<br />
This is, however, not time-critical as only a small part of the running-time is<br />
spent on writing, anyway.<br />
ii) An interesting alternative for post-processing would be to apply the theory of<br />
binary quadratic forms. The obvious strategy is to factorize x 4 +2y 4 completely
Sec. 7] IMPROVEMENTS II – ADAPTION TO OUR HARDWARE 343<br />
into prime powers <strong>and</strong> to deduce from the decomposition all pairs (a, b) such<br />
that a 2 + b 2 = x 4 + 2y 4 . Then, one may check whether for one of them both<br />
a <strong>and</strong> b are perfect squares.<br />
2<br />
7.5. Remark. –––– The migration to a more bit-based implementation led to<br />
an increase of the speed of our programs by a factor of approximately 1.35.<br />
ii. Adaption to the memory architecture of our computer. —<br />
7.6. –––– The factor of 1.35 is less than what we actually hoped for. For that<br />
reason, we made various tests in order to find out what the limiting bottleneck<br />
of our program is. It turned out that the major slowdown is the access of the<br />
processor to main memory.<br />
Our programs are, in fact, doing only two things, integer arithmetic <strong>and</strong> memory<br />
access. The integer execution units of modern processors are highly optimized<br />
circuits <strong>and</strong> several of them work in parallel inside one processor. They<br />
work a lot faster than main memory does. In order to reach a further improvement,<br />
it will therefore be necessary to take the architecture of memory into<br />
closer consideration.<br />
7.7. The Situation. –––– Computer designers try to bridge the gap between the<br />
fast processor <strong>and</strong> the slow memory by building a memory hierarchy which<br />
consists of several cache levels.<br />
The cache is a very small <strong>and</strong> fast memory inside the processor. The first<br />
cache level, called L1 cache, of our processor consists of a data cache <strong>and</strong> an<br />
instruction cache. Both are 64 kbyte in size. The cache manager stores the most<br />
recently used data into the cache in order to make sure a second access to them<br />
will be fast.<br />
If the cache manager does not find necessary data within the L1 cache then the<br />
processor is forced to wait. In order to deliver data, the cache management first<br />
checks the L2 cache which is 1024 kbyte large. It consists of 16384 lines of<br />
64 byte, each.<br />
7.8. Our Program. –––– Our program fits into the instruction cache, completely.<br />
Therefore, no problem should arise from this.<br />
When we consider the data cache, however, the situation is entirely different.<br />
The cache manager stores the 1024 most recently used memory lines, each being<br />
64 byte long, within the L1 data cache.<br />
This strategy is definitely good for many applications. It guarantees main<br />
memory may be scanned at a high speed. On the other h<strong>and</strong>, for our application,<br />
it fails completely. The reason is that access to our 500 Mbyte hash table is
344 THE DIOPHANTINE EQUATION x 4 + 2y 4 = z 4 + 4w 4 [Chap. XI<br />
completely r<strong>and</strong>om. An access directly to the L1 cache happens in by far less<br />
than 0.1% of the cases. In all other cases, the processor has to wait.<br />
Even worse, it is clear that in most cases we do not even access the L2 cache.<br />
This means, the cache manager needs to access main memory in order to transfer<br />
the corresponding memory line of 64 byte into the L1 cache. After this, the<br />
processor may use the data. In the case that there is no free line available<br />
within the L1 cache, the cache manager must restore old data back to main<br />
memory, first. This process takes us 60 nanoseconds, at least, which seems to<br />
be short, but the processor could execute more than 100 integer instructions<br />
during the same time.<br />
The philosophy for further optimization must, therefore, be to adapt the programs<br />
as much as possible to our hardware, first of all to the sizes of the L1 <strong>and</strong><br />
L2 caches.<br />
7.9. Programmer’s position. –––– Unfortunately, the whole memory hierarchy<br />
is invisible from the point of view of a higher programming language, such asC,<br />
since such languages are designed for being machine-independent. Further, the<br />
hardware executes the cache management in an automatic manner. This means,<br />
even by programming in assembly, one cannot control the cache completely<br />
although some new assembly instructions such as prefetch allow certain direct<br />
manipulations.<br />
7.10. A way out. –––– A practical way, nonetheless to gain some influence on<br />
thememoryhierarchy, istorearrangethealgorithm inanapparentlynonsensical<br />
manner, thereby making memory access less chaotic. One may then hope that<br />
the automatic management of the cache, when confronted with the modified<br />
algorithm, is able to react more properly. This should allow the program to<br />
run faster.<br />
iii. Our first trial. —<br />
7.11. –––– Our first idea for this was to work with two arrays instead of one.<br />
Algorithm M.<br />
i) Store the values of f into an array <strong>and</strong> the values of g into a another one.<br />
Write successively calculated values into successive positions. It is clear that this<br />
part of the algorithm is not troublesome as it involves a linear memory access<br />
which is perfectly supported by the memory management.<br />
ii) Then, use Quicksort in order to sort both arrays. In addition to being<br />
fast, Quicksort is known to have a good memory locality when large arrays<br />
are sorted.
Sec. 7] IMPROVEMENTS II – ADAPTION TO OUR HARDWARE 345<br />
iii) In a final step, search for matches by going linearly through both arrays as<br />
in Mergesort.<br />
7.12. Remark –––– Unfortunately, the idea behind Algorithm M is too simple<br />
to give it any chance of being superior to the previous algorithms. However, it is<br />
a worthwhile experiment. Indeed, our implementation of Algorithm M causes<br />
at least 30 times more memory transfer compared with the previous programs<br />
but, actually, it is only three times slower. This indicates that our approach<br />
is reasonable.<br />
iv. Hashing with partial presorting. —<br />
7.13. –––– Our final algorithm is a combination of sorting <strong>and</strong> hashing. An important<br />
aspect of it is that the sorting step has to be considerably faster than the<br />
Quicksort algorithm. For that reason, we adopted some ideas from linear-time<br />
sorting algorithms such as Radix sort or Bucket sort.<br />
7.14. –––– The algorithm works as follows. Again, the description is based on<br />
case S, case N being analogous.<br />
Algorithm H64B.<br />
I. Initialization. Fix B := 10 8 . Initialize a hash table H of 2 27 = 134 217 728<br />
integers, each being 32 bit long. Fix the page prime p p := 200 003.<br />
In addition, initialize 1024 auxiliary arrays A i each of which may contain<br />
2 17 = 131 072 long (64 bit) integers.<br />
Further, define two functions, the hash function h <strong>and</strong> the control function c,<br />
which map 64 bit integers to 27 bit integers <strong>and</strong> 31 bit integers, respectively, by<br />
selecting certain bits. Do not use any of the bits twice to ensure h <strong>and</strong> c are<br />
independent on each other <strong>and</strong> do not use the four least significant bits.<br />
Finally, let h (10) denote the function mapping 64 bit integers to integers<br />
within [0, 1023] which is given by the ten most significant bits of h. In other<br />
words, for every x, h (10) (x) is the same as h(x) shifted to the right by 17 bits.<br />
II) Outer Loop. Let r run from 0 to p p − 1 <strong>and</strong> execute A. <strong>and</strong> B. for each r.<br />
A. Writing. Build up the hash table, which is meant to encode the set L r ,<br />
as follows.<br />
a) Preparation. Find all pairs (z, w) of non-negative integers less than or equal<br />
to B which satisfy z 4 + 4w 4 ≡ r (mod p p ) <strong>and</strong> all the congruence-conditions<br />
for primitive solutions, listed above. (Make systematic use of the Chinese<br />
remainder theorem.)<br />
b) Inner Loop. Execute steps i) – iii) below for each such pair.
346 THE DIOPHANTINE EQUATION x 4 + 2y 4 = z 4 + 4w 4 [Chap. XI<br />
i) Evaluate f S (z, w) := (z 4 + 4w 4 mod 2 64 ).<br />
ii) Do not store f S (z, w) into the hash table, immediately. Put<br />
first.<br />
i := h (10) ( f S (z, w)),<br />
iii) Add f S (z, w) to the auxiliary array A i . Maintain A i as an unordered list.<br />
I.e., always write to the lowest unoccupied address.<br />
If there is no space left in A i then output an error message <strong>and</strong> abort the algorithm.<br />
c) Storing. Let i run from 0 to 1023. For each i let j run through the addresses<br />
occupied in A i .<br />
For fixed i <strong>and</strong> j, extract from the 64 bit integer A i [j] the 27 bit hash value<br />
h(A i [j]) <strong>and</strong> the 31 bit control value c(A i [j]).<br />
Usethehashvalueh(A i [j])<strong>and</strong>linear probingtofindafreeplaceinthehashtable<br />
<strong>and</strong> store the control value c(A i [j]) there.<br />
d) Clearing up. Clear the auxiliary arrays A i for all i ∈ [0, 1023] to make them<br />
available for reuse.<br />
B. Reading. Search within the hash table, as follows.<br />
a) Preparation. Find all pairs (x, y) of non-negative integers less than or equal<br />
to B which satisfy x 4 + 2y 4 ≡ r (mod p p ) <strong>and</strong> all the congruence conditions<br />
for primitive solutions, listed above. (Make systematic use of the Chinese<br />
remainder theorem.)<br />
b) Inner Loop. Execute steps i) – iii) below for each such pair.<br />
i) Evaluate g S (x, y) := (x 4 + 2y 4 mod 2 64 ).<br />
ii) Do not look up g S (x, y) in the hash table, immediately. Put<br />
first.<br />
i := h (10) (g S (x, y)),<br />
iii) Add g S (x, y) to the auxiliary array A i . Maintain A i as an unordered list.<br />
I.e., always write to the lowest unoccupied address.<br />
If there is no space left in A i then call d[i] <strong>and</strong> add g S (x, y) to A i , afterwards.<br />
c) Searching. Clearing all buffers. Let i run from 0 to 1023. For each i, call d[i].<br />
When this is finished, go to the next iteration of the outer loop.
Sec. 7] IMPROVEMENTS II – ADAPTION TO OUR HARDWARE 347<br />
Subroutine d[i]) Clearing a buffer. Let j run through the addresses occupied<br />
in A i . For fixed j, search for the control value c(A i [j]) within the hash table H ,<br />
starting at the hash value h(A i [j]) <strong>and</strong> using linear probing, until a free place<br />
is found. Report all hits <strong>and</strong> the corresponding values of x <strong>and</strong> y.<br />
Having done this, declare A i to be empty.<br />
7.15. Remark –––– The auxiliary arrays A i play the role of a buffer. Thus, one<br />
could say that we introduced some buffering into the management of the hash table<br />
H . However, this description misses the point.<br />
What is more important is that the values of f S to be stored into L r are partially<br />
sorted according to the 10 most significant bits of h( f S (z, w)) by putting them<br />
into the auxiliary arrays A i . When the hash table is then built up, the records<br />
arrive almost in order. The same is true for reading.<br />
What we actually did is, therefore, to introduce some partial presorting into the<br />
management of the hash table.<br />
7.16. Remark –––– It is our experience that each auxiliary array carries more<br />
or less the same load. In particular, in step II.A.b.iii), when the buffers are filled<br />
up for writing, a buffer overflow should never occur. For this reason, we feel<br />
free to treat this possibility as a fatal error.<br />
v. Running time. —<br />
7.17. –––– Algorithm H64B uses about three times more memory than our<br />
previous algorithms but our implementation runs almost three times as fast.<br />
It was this factor which made it possible to attack the bound B = 10 8 in a<br />
reasonable amount of time.<br />
The final version of our programs took almost exactly 100 days of CPU time<br />
on an AMD Opteron 248 processor. This time is composed almost equally<br />
of 50 days for case N <strong>and</strong> 50 days for case S. The main computation was<br />
executed in parallel on two machines in February <strong>and</strong> March, 2005.<br />
7.18. Why is this algorithm faster? –––– To answer this question, one has to<br />
look at the impact of the cache. For the old program, the cache memory was<br />
mostly useless. For the new program, the situation is completely different.<br />
When the auxiliary arrays are filled in step II.A.b.iii) <strong>and</strong> II.B.b.iii), access to<br />
these arrays is linear. There are only 1024 of them which is exactly the number<br />
of lines in the L1 cache. When an access does not hit into that innermost cache<br />
then the corresponding memory line is moved to it <strong>and</strong> the next seven accesses<br />
to the same auxiliary array are accesses to that line. Altogether, seven of eight<br />
memory accesses hit into the L1 cache.
348 THE DIOPHANTINE EQUATION x 4 + 2y 4 = z 4 + 4w 4 [Chap. XI<br />
When an auxiliary array is emptied in step II.A.c) or II.B.d[i]), the situation<br />
is similar. There are a high number of accesses to a very short segment of the<br />
hash table. This segment fits completely into the L2 cache. It has to be moved<br />
into that cache, once. Then, it can be used many times. Again, access to the<br />
auxiliary array is linear <strong>and</strong> a hit into the L1 cache occurs in seven of eight cases.<br />
All in all, for Algorithm H64B, most memory accesses are hits into the cache.<br />
This means, at the cost of some more data transfer altogether, we achieved that<br />
main memory may be mostly used at the speed of the cache.<br />
8. The solution found<br />
8.1. –––– Ironically, the premature version of our algorithm as described in<br />
Section 4 already solved Problem 1.5.<br />
The improvements made it possible to shift the search bound to 10 8 which<br />
was far beyond our original expectations. However, they did not produce any<br />
new solution.<br />
Thus, let us conclude this chapter by some comments on the simpler approach.<br />
8.2. –––– Test versions of the program were written in Delphi.<br />
The definitive version was written in C. It took about 130 hours of CPU time<br />
on a 3.00 GHz Pentium 4 processor with 512 kbyte cache memory. The main<br />
computation was executed in parallel on two machines during the very first<br />
days of December, 2004.<br />
8.3. –––– Instead of looking for solutions of x 4 + 2y 4 = z 4 + 4w 4 , the algorithm<br />
searches, in fact, for solutions to the corresponding simultaneous congruences<br />
modulo p p <strong>and</strong> p c which, in addition, fulfill that ( (x 4 + 2y 4 ) mod p h<br />
)<br />
<strong>and</strong> ( (z 4 + 4w 4 ) mod p h<br />
)<br />
are “almost equal”.<br />
To this modified problem, we found approximately 3800 solutions such that<br />
(y, w) ≠ (0, 0). These congruence solutions were checked by an exact computation<br />
using O. Forster’s [Fo] Pascal-style multi-precision interpreter language<br />
ARIBAS.<br />
8.4. –––– Among the congruence solutions, exact equality occurred only once.<br />
This solution is as follows.<br />
==> 1484801**4 + 2 * 1203120**4.<br />
-: 90509_10498_47564_80468_99201<br />
==> 1169407**4 + 4 * 1157520**4.<br />
-: 90509_10498_47564_80468_99201
CHAPTER XII<br />
NEW SUMS OF THREE CUBES ∗<br />
There are three cubes (levels) in total, each of which are connected by a<br />
single door. Players have the goal of moving the character from room to<br />
room, cube to cube in an attempt to find the final exit door of all three cubes.<br />
If this door is found the character will appear to leave the cube, walk across<br />
the table surface <strong>and</strong> vanish. The game then begins again.<br />
JULIAN OLIVER: levelHead, A 3D Spatial Memory Game (2007)<br />
1. Introduction<br />
It is a long st<strong>and</strong>ing problem as to whether every <strong>rational</strong> integer<br />
n ≢ 4, 5 (mod 9) can be written as a sum of three integral cubes. According<br />
to the web page [Be-list] of Daniel Bernstein, the first attacks by computer<br />
were carried out as early as 1955.<br />
Nevertheless, for example for n = 3, there is still no solution known apart from<br />
the obvious ones: (1, 1, 1), (4, 4, −5), (4, −5, 4), <strong>and</strong> (−5, 4, 4). For n = 30, the<br />
first solution was found by N. Elkies <strong>and</strong> his coworkers in 2000 [El]. It is interesting<br />
to note that, in 1992, D. R. Heath-Brown [H-B92a] had made a prediction<br />
on the density of the solutions for n = 30 without knowing any solution explicitly.<br />
Over the years, a number of algorithms have been developed in order to attack<br />
the general problem. An excellent overview concerning the various approaches<br />
invented up to around 2000 has been given in [B/P/T/Y], published recently.<br />
The historically first algorithm which has a complexity of O(B 1+ε ) for a search<br />
bound of B is the method of R. Heath-Brown [H/L/R]. The best algorithm<br />
presently known is Elkies’ method described in [El].<br />
(∗) This chapter is a revised version of the article: New sums of three cubes, to appear in:<br />
Math. Comp., joint with A.-S. Elsenhans.
350 NEW SUMS OF THREE CUBES [Chap. XII<br />
2. Elkies’ method<br />
This<br />
√<br />
algorithm is geometric<br />
√<br />
in nature. The idea is to cover the curve<br />
Y = 3 1 − X 3<br />
, X ∈ [0, 1/ 3 2], by very small parallelograms which we call flagstones.<br />
The algorithm finds all <strong>rational</strong> <strong>points</strong> of the particular form (x/z, y/z)<br />
which are contained in one of the flagstones for z ∈Æup to a given bound B.<br />
(This means we are searching for triples such that x 3 + y 3 − z 3 will be small.)<br />
For each flagstone, this is equivalent to the detection of all <strong>points</strong> of the st<strong>and</strong>ard<br />
lattice3 that are contained in a certain pyramid. The problem is that, viewed<br />
in the st<strong>and</strong>ard coordinates, this pyramid has an enormous height in comparison<br />
with the two other dimensions. Thus, it has an extremely sharp apex. Searching<br />
naively for lattice <strong>points</strong> in such a pyramid would be highly inefficient.<br />
The idea to overcome this difficulty is to work in coordinates more<br />
adapted to this pyramid. The drawback in this case is that the basis<br />
{ (1, 0, 0), (0, 1, 0), (0, 0, 1) } of the st<strong>and</strong>ard lattice is then far from being<br />
reduced in whatever sense. One needs to apply lattice basis reduction. Having<br />
done that, searching for lattice <strong>points</strong> within the pyramid is essentially<br />
equivalent to a search for small <strong>points</strong> of the lattice. For this, one may use the<br />
well-known algorithm of Fincke-Pohst [F/P].<br />
The size of the flagstones is somewhat arbitrary. Smaller flagstones mean that<br />
more time is required for lattice basis reductions. Larger ones lead to more time<br />
being spent on the algorithm of Fincke-Pohst. The optimum size depends on<br />
details of the implementation.<br />
3. Implementation<br />
Our implementation of Elkies’ method is written in C <strong>and</strong> C++. We took care<br />
that only initialization parts of the code were written inC++ or made use of the<br />
multi precision floats of GMP.<br />
The time-critical parts were written in plain C making some use of the instruction<br />
asm. It turned out that, for most of the computations, 128-bit fixed-point<br />
arithmetic was sufficiently precise. We realized the 128-bit fixed-point numbers<br />
as arrays consisting of two long ints. The arithmetic of the fixed-point<br />
numbers was implemented in such a way that all loops (of length two) were<br />
manually unrolled.<br />
For lattice basis reduction, we implemented a version of LLL for three-dimensional<br />
lattices. It turned out that adjacent flagstones led to similar reduced bases.<br />
That is why enormous savings could be achieved by doing LLL incrementally.
Sec. 4] RESULTS 351<br />
We start theLLL computation for the next flagstone with a reduced basis of the<br />
previous one <strong>and</strong> not with a naive basis.<br />
Another substantial improvement was realized in the Fincke-Pohst part.<br />
Here, one has to compute many adjacent values of the same cubic polynomial<br />
in three variables. An implementation of a difference scheme reduced most<br />
of these computations to a few additions of values obtained before.<br />
Some details. — We searched systematically for solutions of x 3 + y 3 + z 3 = n<br />
where the positive integer n < 1000 is neither a cube nor twice a cube <strong>and</strong><br />
| x|, |y|, |z| < 10 14 . The length of the flagstones was chosen dynamically. It was<br />
around 8.4 · 10 −12 near x = 0 <strong>and</strong> around 6.6 · 10 −14 near x = 1/ 3 √<br />
2. The area<br />
of the flagstones was essentially constant at a value near 1.7 · 10 −40 . This led to<br />
a total number of a bit more than 10 13 flagstones to deal with.<br />
We chose the widths of the flagstones such that all <strong>points</strong> in a horizontal<br />
distance of < 10 −30 from the curve are contained in one of the flagstones.<br />
This should make sure that all solutions of heights between 10 11 <strong>and</strong> 10 14 are<br />
certainly found. Indeed, if we arrange variables such that | x| ≤ |y| ≤ |z|<br />
then the point (| x/z|, |y/z|) is at a horizontal distance from the curve of,<br />
√<br />
in<br />
ds<br />
first order approximation, 1/3<br />
|<br />
ds s=(1−X 3 ) · n/|z| 3 . Since X := | x/z| < 1/ 3 2,<br />
the derivative is always less than 0.53.<br />
The whole search took around ten months of CPU time. Only 14% of that<br />
time was spent on lattice basis reductions. The lion’s share was spent searching<br />
for small lattice <strong>points</strong>. I.e., on our implementation of the algorithm of Fincke-<br />
Pohst.<br />
4. Results<br />
In comparison with the computations of [B/P/T/Y] <strong>and</strong> the lists, dating back<br />
to 2001 <strong>and</strong> published in [Be-list], 3519 new solutions have been found.<br />
Among them, there are the following. For each of the nine numbers on the<br />
left, no solution had been given before, neither in D. Bernstein’s lists, nor<br />
in [B/P/T/Y].<br />
156 = 26 577 110 807 569 3 − 18 161 093 358 005 3 − 23 381 515 025 762 3<br />
318 = 1 970 320 861 387 3 + 1 750 553 226 136 3 − 2 352 152 467 181 3<br />
= 30 828 727 881 037 3 + 27 378 037 791 169 3 − 36 796 384 363 814 3<br />
366 = 241 832 223 257 3 + 167 734 571 306 3 − 266 193 616 507 3<br />
420 = 8 859 060 149 051 3 − 2 680 209 928 162 3 − 8 776 520 527 687 3
352 NEW SUMS OF THREE CUBES [Chap. XII<br />
564 = 53 872 419 107 3 − 1 300 749 634 3 − 53 872 166 335 3<br />
758 = 662 325 744 409 3 + 109 962 567 936 3 − 663 334 553 003 3<br />
= 83 471 297 139 078 3 + 77 308 024 343 011 3 − 101 433 242 878 565 3<br />
789 = 18 918 117 957 926 3 + 4 836 228 687 485 3 − 19 022 888 796 058 3<br />
894 = 19 868 127 639 556 3 + 2 322 626 411 251 3 − 19 878 702 430 997 3<br />
948 = 103 458 528 103 519 3 + 6 604 706 697 037 3 − 103 467 499 687 004 3<br />
For 13 values of n, for which exactly one solution was known, we found a<br />
second one. Among those, there is n = 30. The second solution for n = 30 is<br />
30 = 3 982 933 876 681 3 − 636 600 549 515 3 − 3 977 505 554 546 3 .<br />
A second <strong>and</strong> a third solution for n = 75 are<br />
75 = 2 576 191 140 760 3 + 1 217 343 443 218 3 − 2 663 786 047 493 3<br />
= 59 897 299 698 355 3 − 47 258 398 396 091 3 − 47 819 328 945 509 3 .<br />
On the other h<strong>and</strong>, the highest numbers of solutions found are 93 for n = 792<br />
<strong>and</strong> 85 for n = 720.<br />
This fits well with some speculations made in [H-B92a]. D. R. Heath-Brown<br />
predicts ∼A(n) logB essentially different solutions of height
APPENDIX<br />
Diophantus, with this name which is frequent in Greece, was a true Greek,<br />
disciple of Greek science, if one who towers high above his contemporaries.<br />
He was Greek in what he accomplished, as well as in what he was not able<br />
to accomplish. But we must not forget that Greek science, as it conquered<br />
the East from Alex<strong>and</strong>ria . . . , brought new ideas back home from these<br />
campaigns, that Greek mathematics as such has ceased to pick up whatever<br />
it found worth picking up here <strong>and</strong> there.<br />
MORITZ CANTOR (1907, translated by N. Schappacher)<br />
1. A script in GAP<br />
This is the script in GAP which was used to compute H 1( G, Pic(XÉ) )<br />
<strong>and</strong> rkPic(X ) for all <strong>groups</strong> which may operate on the 27 lines upon a smooth<br />
cubic surface. The mathematical background is explained in Section 8, Subsection<br />
iv of Chapter II.<br />
TABLE 1. A GAP script<br />
# A script in GAP to compute H^1(G, Pic) = (NS \cap S_0) / NS_0 <strong>and</strong> the Picard rank for all<br />
# Galois <strong>groups</strong> of a cubic surface.<br />
# Here, S is the free abelian group on the 27 lines, S_0 \subset S the subgroup of all<br />
# principal divisors, <strong>and</strong> N: S --> S the norm map.<br />
# We compute the intersection matrix of the 27 lines.<br />
# g must be a faithful representation of W(E_6) in S_27.<br />
SchnittMatrixBerechnen := function (g)<br />
local i, j, mat, len, tmp;<br />
mat := NullMat(27, 27);<br />
for i in [1..27] do<br />
for j in [1..27] do<br />
if (i = j) then<br />
mat[i][j] := -1;<br />
else<br />
tmp := [i, j]; Sort(tmp);<br />
len := Size(Orbit(g, tmp, OnSets));<br />
# Operation of W(E_6) on unordered pairs of lines.<br />
# A pair of skew lines has orbit length 27*8.<br />
# A pair of intersecting lines has orbit length 27*5.<br />
if (len = 27*5) then<br />
mat[i][j] := 1;<br />
fi;<br />
fi;<br />
od;<br />
od;<br />
return mat;
354 APPENDIX<br />
end;<br />
# Compute the 27x27-matrix representing the norm map N: S --> S.<br />
NBerechnen := function (orb, u)<br />
local N, akt, j, k, l;<br />
# The matrix N.<br />
N := NullMat(27, 27);<br />
for j in [1..Size(orb)] do<br />
akt := orb[j];<br />
for k in [1..Size(akt)] do<br />
for l in [1..Size(akt)] do<br />
N[akt[k]][akt[l]] := Size(u) / Size(akt);<br />
od;<br />
od;<br />
od;<br />
return N;<br />
end;<br />
# Build up a matrix containing a minimal system of generators of NS in its columns.<br />
IBerechnen := function (orb, u)<br />
local my_I, j, k, akt;<br />
# The matrix I.<br />
my_I := NullMat(27, Size(orb));<br />
for j in [1..Size(orb)] do<br />
akt := orb[j];<br />
for k in [1..Size(akt)] do<br />
# The line akt[k] lies in orbit akt numbered j.<br />
my_I[akt[k]][j] := Size(u) / Size(akt);<br />
od;<br />
od;<br />
return my_I;<br />
end;<br />
# A separate routine in order to make a lattice base from the column vectors of a matrix.<br />
NormMat := function (m)<br />
return TransposedMat(HermiteNormalFormIntegerMat(BaseIntMat(TransposedMat(m))));<br />
end;<br />
# The discriminant of a lattice. The lattice needs not be maximal.<br />
# The lattice base is supposed to be given in the column vectors of m.<br />
# It is important that m is a lattice base. The function will return 0 for a linearly<br />
# dependent system.<br />
Disc := function (m)<br />
return Determinant(TransposedMat(m) * m);<br />
end;<br />
# We compute H^1 of the Picard group <strong>and</strong> the arithmetic Picard rank using Manin’s formulas.<br />
h1_pic := function (schnitt_matrix, u)<br />
local orb, I_mat, AI, N, K, linke_seite, rechte_seite, disc_links,<br />
disc_rechts, index, rang, mul, scal_links, schnitt;<br />
# Everything depends only on the combinatorial structure of the orbits.<br />
orb := Orbits(u, [1..27]);<br />
I_mat := IBerechnen(orb, u); # I_mat represents NS.<br />
AI := schnitt_matrix * I_mat;<br />
N := NBerechnen(orb, u); # N represents the norm map.<br />
# Normalization: We choose all integral vectors in the kernel.
APPENDIX 355<br />
# Computation of NS \cap S_0.<br />
K := NullspaceIntMat(TransposedMat(AI));<br />
# K collects the relations among the generators of NS modulo S_0.<br />
linke_seite := NormMat(I_mat * TransposedMat(K));<br />
# linke_seite represents the lattice NS \cap S_0.<br />
K := NullspaceIntMat(TransposedMat(schnitt_matrix));<br />
# The rows of K are generators of S_0.<br />
rechte_seite := NormMat(N * TransposedMat(K));<br />
# The columns of N * TransposedMat(K) are generators of NS_0.<br />
# The discriminats.<br />
disc_links := Disc(linke_seite);<br />
disc_rechts := Disc(rechte_seite);<br />
index := RootInt(disc_rechts / disc_links);<br />
rang := Size(orb) - Size(TransposedMat(rechte_seite));<br />
# The order of H^1(G, Pic).<br />
# The arithmetic Picard rank.<br />
AppendTo("h1pic.txt"," #H^1 = ", index);<br />
if index > 3 then<br />
# We have to compute the primary decomposition.<br />
# This code suffices since index may be only 1, 2, 3, 4, or 9.<br />
if index = 4 then mul := 2; else mul := 3; fi;<br />
scal_links := mul * linke_seite;<br />
schnitt := TransposedMat(BaseIntersectionIntMats(TransposedMat(rechte_seite),<br />
TransposedMat(scal_links)));<br />
if Disc(schnitt) = Disc(scal_links) then<br />
AppendTo("h1pic.txt"," [ ", mul, ", ", mul, " ]");<br />
else<br />
AppendTo("h1pic.txt"," More complicated than the direct sum of Z/", mul, "Z");<br />
fi;<br />
fi;<br />
AppendTo("h1pic.txt", ", Rk(Pic) = ", rang, ",");<br />
end;<br />
# Adds the orbit structure of the lines to the list.<br />
bahnstruktur := function (u)<br />
local ol;<br />
ol := ShallowCopy(OrbitLengths(u, [1..27]));<br />
Sort(ol);<br />
AppendTo("h1pic.txt", " Orbits ", ol);<br />
end;<br />
# Our group taken from the library.<br />
we6 := TransitiveGroup(27, 1161);<br />
schnitt_matrix := SchnittMatrixBerechnen(we6);<br />
# The sub<strong>groups</strong>.<br />
ugv := ConjugacyClassesSub<strong>groups</strong>(we6);;<br />
# We compute H^1(G, Pic) for all Galois <strong>groups</strong>.<br />
for i in [1..Size(ugv)] do<br />
u := Representative(ugv[i]);<br />
AppendTo("h1pic.txt", i, " #U = ", Size(u), " ");<br />
AppendTo("h1pic.txt", AbelianInvariants(u), ",");<br />
if IsTransitive(u) <strong>and</strong> (Size(u) > 1) then<br />
AppendTo("h1pic.txt", " transitive");<br />
else<br />
h1_pic(schnitt_matrix, u);<br />
bahnstruktur(u);<br />
fi;<br />
AppendTo("h1pic.txt", "\n");<br />
od;
356 APPENDIX<br />
2. The list<br />
This is the list produced by theGAP script. For each subgroup of W (E 6 ), we give<br />
its abelian quotient, the corresponding values of H 1( G, Pic(XÉ) ) <strong>and</strong> rkPic(X ),<br />
<strong>and</strong> the orbit structure of the 27 lines.<br />
We make use of this list in several ways. For our presentation of the Brauer-<br />
Manin obstruction in the case of diagonal cubic surfaces given in Section III.6,<br />
the list is of fundamental importance. On the other h<strong>and</strong>, we also use it in the<br />
arguments given in III.5.33 <strong>and</strong> Remarks III.5.37.<br />
It is further applied in Examples VI.5.7 <strong>and</strong> VI.5.8. Here, the goal is simply to<br />
show that one has Picard rank 2 in the situations considered.<br />
Finally, we give reference to the list in Remarks VI.5.13.ii) <strong>and</strong> X.2.4 in order<br />
to prove that the factor β is bounded from above by 9.<br />
TABLE 2. H 1 (G, Pic) <strong>and</strong> rk Pic(X ) for smooth cubic surfaces<br />
1 #U = 1 [ ], #H^1 = 1, Rk(Pic) = 7, Orbits [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,<br />
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ]<br />
2 #U = 2 [ 2 ], #H^1 = 1, Rk(Pic) = 6, Orbits [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,<br />
1, 2, 2, 2, 2, 2, 2 ]<br />
3 #U = 2 [ 2 ], #H^1 = 4 [ 2, 2 ], Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2,<br />
2, 2, 2, 2 ]<br />
4 #U = 2 [ 2 ], #H^1 = 1, Rk(Pic) = 5, Orbits [ 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2,<br />
2, 2, 2 ]<br />
5 #U = 2 [ 2 ], #H^1 = 1, Rk(Pic) = 4, Orbits [ 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,<br />
2, 2 ]<br />
6 #U = 3 [ 3 ], #H^1 = 9 [ 3, 3 ], Rk(Pic) = 1, Orbits [ 3, 3, 3, 3, 3, 3, 3, 3, 3 ]<br />
7 #U = 3 [ 3 ], #H^1 = 1, Rk(Pic) = 5, Orbits [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 3, 3, 3,<br />
3, 3 ]<br />
8 #U = 3 [ 3 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 3, 3, 3, 3, 3, 3, 3, 3, 3 ]<br />
9 #U = 4 [ 4 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 4, 4, 4, 4, 4, 4 ]<br />
10 #U = 4 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 5, Orbits [ 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2,<br />
2, 2, 4 ]<br />
11 #U = 4 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 4, Orbits [ 1, 1, 1, 1, 1, 2, 2, 2, 4, 4, 4, 4 ]<br />
12 #U = 4 [ 2, 2 ], #H^1 = 4 [ 2, 2 ], Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 2, 2, 4, 4, 4, 4 ]<br />
13 #U = 4 [ 4 ], #H^1 = 4 [ 2, 2 ], Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 4, 4, 4, 4 ]<br />
14 #U = 4 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 2, 2, 4, 4, 4, 4 ]<br />
15 #U = 4 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 4, Orbits [ 1, 1, 1, 2, 2, 2, 2, 2, 2, 4, 4, 4 ]<br />
16 #U = 4 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 2, 2, 2, 2, 4, 4, 4 ]<br />
17 #U = 4 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 4, 4, 4, 4, 4 ]<br />
18 #U = 4 [ 4 ], #H^1 = 1, Rk(Pic) = 4, Orbits [ 1, 1, 1, 1, 1, 2, 4, 4, 4, 4, 4 ]<br />
19 #U = 4 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 2, 2, 2, 4, 4, 4, 4 ]<br />
20 #U = 4 [ 4 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 2, 4, 4, 4, 4, 4 ]<br />
21 #U = 4 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 4, 4, 4, 4 ]<br />
22 #U = 4 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 4, Orbits [ 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 4, 4 ]<br />
23 #U = 5 [ 5 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 5, 5, 5, 5, 5 ]<br />
24 #U = 6 [ 2 ], #H^1 = 1, Rk(Pic) = 5, Orbits [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 3, 3, 3,<br />
3, 3 ]<br />
25 #U = 6 [ 2 ], #H^1 = 4 [ 2, 2 ], Rk(Pic) = 1, Orbits [ 3, 3, 3, 6, 6, 6 ]<br />
26 #U = 6 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 6, 6, 6, 6 ]<br />
27 #U = 6 [ 2, 3 ], #H^1 = 4 [ 2, 2 ], Rk(Pic) = 1, Orbits [ 3, 6, 6, 6, 6 ]<br />
28 #U = 6 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 4, Orbits [ 1, 1, 1, 2, 2, 2, 3, 3, 3, 3, 6 ]<br />
29 #U = 6 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 3, 3, 6, 6, 6 ]<br />
30 #U = 6 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 2, 6, 6, 6 ]<br />
31 #U = 6 [ 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 2, 6, 6, 6 ]<br />
32 #U = 6 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 3, 3, 3, 3, 6, 6 ]
APPENDIX 357<br />
33 #U = 6 [ 2 ], #H^1 = 1, Rk(Pic) = 4, Orbits [ 1, 1, 1, 2, 2, 2, 3, 3, 3, 3, 6 ]<br />
34 #U = 6 [ 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 3, 3, 3, 3, 3, 3, 3, 6 ]<br />
35 #U = 6 [ 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 2, 2, 3, 3, 6, 6 ]<br />
36 #U = 6 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 2, 2, 3, 3, 6, 6 ]<br />
37 #U = 6 [ 2 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 3, 3, 3, 6, 6, 6 ]<br />
38 #U = 6 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 3, 3, 6, 6, 6 ]<br />
39 #U = 6 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 6, 6, 6, 6 ]<br />
40 #U = 8 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 8, 8, 8 ]<br />
41 #U = 8 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 2, 2, 8, 8 ]<br />
42 #U = 8 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 3, Orbits [ 1, 1, 1, 4, 4, 4, 4, 4, 4 ]<br />
43 #U = 8 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 2, 2, 8, 8 ]<br />
44 #U = 8 [ 2, 2 ], #H^1 = 4 [ 2, 2 ], Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 4, 4, 4, 4 ]<br />
45 #U = 8 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 4, 4, 8, 8 ]<br />
46 #U = 8 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 4, Orbits [ 1, 1, 1, 2, 2, 2, 2, 2, 2, 4, 4, 4 ]<br />
47 #U = 8 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 2, 2, 2, 4, 4, 8 ]<br />
48 #U = 8 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 3, Orbits [ 1, 1, 1, 4, 4, 4, 4, 4, 4 ]<br />
49 #U = 8 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 4, Orbits [ 1, 1, 1, 1, 1, 2, 4, 4, 4, 4, 4 ]<br />
50 #U = 8 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 8, 8 ]<br />
51 #U = 8 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 8, 8 ]<br />
52 #U = 8 [ 2, 4 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 8, 8 ]<br />
53 #U = 8 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 4, 4, 4, 4, 4 ]<br />
54 #U = 8 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 4, 4, 4, 4, 8 ]<br />
55 #U = 8 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 8, 8 ]<br />
56 #U = 8 [ 2, 4 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 8, 8 ]<br />
57 #U = 8 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 4, 8, 8 ]<br />
58 #U = 8 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 2, 4, 4, 4, 8 ]<br />
59 #U = 8 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 2, 4, 4, 4, 8 ]<br />
60 #U = 8 [ 2, 4 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 4, 4, 8 ]<br />
61 #U = 8 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 2, 4, 4, 4, 4, 4 ]<br />
62 #U = 8 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 8, 8 ]<br />
63 #U = 8 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 4, 8, 8 ]<br />
64 #U = 8 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 4, 4, 8 ]<br />
65 #U = 8 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 4, 4, 8 ]<br />
66 #U = 8 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 8, 8 ]<br />
67 #U = 8 [ 8 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 8, 8 ]<br />
68 #U = 8 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 2, 4, 4, 4, 8 ]<br />
69 #U = 8 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 4, 4, 8 ]<br />
70 #U = 9 [ 3, 3 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 3, 3, 3, 3, 3, 3, 9 ]<br />
71 #U = 9 [ 3, 3 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 9, 9, 9 ]<br />
72 #U = 9 [ 3, 3 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 3, 3, 3, 9, 9 ]<br />
73 #U = 9 [ 9 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 9, 9 ]<br />
74 #U = 10 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 2, 5, 5, 5, 10 ]<br />
75 #U = 10 [ 2, 5 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 2, 5, 5, 5, 10 ]<br />
76 #U = 10 [ 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 5, 5, 5, 5, 5 ]<br />
77 #U = 12 [ 3 ], #H^1 = 1, Rk(Pic) = 4, Orbits [ 1, 1, 1, 1, 1, 4, 4, 4, 4, 6 ]<br />
78 #U = 12 [ 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 6, 6, 12 ]<br />
79 #U = 12 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 4, Orbits [ 1, 1, 1, 2, 2, 2, 3, 3, 3, 3, 6 ]<br />
80 #U = 12 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 2, 6, 6, 6 ]<br />
81 #U = 12 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 3, 3, 6, 12 ]<br />
82 #U = 12 [ 2, 2 ], #H^1 = 4 [ 2, 2 ], Rk(Pic) = 1, Orbits [ 3, 6, 6, 6, 6 ]<br />
83 #U = 12 [ 2, 2, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 2, 4, 6, 12 ]<br />
84 #U = 12 [ 3, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 4, 6, 12 ]<br />
85 #U = 12 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 2, 4, 6, 12 ]<br />
86 #U = 12 [ 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 4, 6, 12 ]<br />
87 #U = 12 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 3, 3, 4, 6, 6 ]<br />
88 #U = 12 [ 2, 2, 3 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 3, 3, 4, 6, 6 ]<br />
89 #U = 12 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 2, 2, 3, 3, 6, 6 ]<br />
90 #U = 12 [ 3, 4 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 12, 12 ]<br />
91 #U = 12 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 6, 6, 12 ]<br />
92 #U = 12 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 6, 6, 12 ]<br />
93 #U = 12 [ 2, 2, 3 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 6, 6, 12 ]<br />
94 #U = 12 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 3, 3, 6, 6, 6 ]<br />
95 #U = 12 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 6, 6, 12 ]<br />
96 #U = 12 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 2, 4, 6, 12 ]<br />
97 #U = 12 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 3, 3, 4, 6, 6 ]
358 APPENDIX<br />
98 #U = 12 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 6, 6, 6, 6 ]<br />
99 #U = 16 [ 2, 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 2, 2, 16 ]<br />
100 #U = 16 [ 2, 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 3, Orbits [ 1, 1, 1, 4, 4, 4, 4, 4, 4 ]<br />
101 #U = 16 [ 2, 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 16 ]<br />
102 #U = 16 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 8, 8, 8 ]<br />
103 #U = 16 [ 2, 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 8, 8 ]<br />
104 #U = 16 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 4, 4, 8, 8 ]<br />
105 #U = 16 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 16 ]<br />
106 #U = 16 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 4, 4, 8, 8 ]<br />
107 #U = 16 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 16 ]<br />
108 #U = 16 [ 4, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 16 ]<br />
109 #U = 16 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 8, 8 ]<br />
110 #U = 16 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 2, 4, 8, 8 ]<br />
111 #U = 16 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 16 ]<br />
112 #U = 16 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 8, 8 ]<br />
113 #U = 16 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 16 ]<br />
114 #U = 16 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 8, 8 ]<br />
115 #U = 16 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 16 ]<br />
116 #U = 16 [ 2, 2, 4 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 8, 8 ]<br />
117 #U = 16 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 16 ]<br />
118 #U = 16 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
119 #U = 16 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 8, 8 ]<br />
120 #U = 16 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 16 ]<br />
121 #U = 16 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 4, 4, 8, 8 ]<br />
122 #U = 16 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
123 #U = 16 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 4, 4, 8 ]<br />
124 #U = 16 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 8, 8 ]<br />
125 #U = 16 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 2, 4, 4, 4, 8 ]<br />
126 #U = 16 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 8, 8 ]<br />
127 #U = 16 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 8, 8 ]<br />
128 #U = 18 [ 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 3, 3, 3, 3, 3, 3, 9 ]<br />
129 #U = 18 [ 2, 3 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 3, 3, 3, 9, 9 ]<br />
130 #U = 18 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 3, 3, 3, 3, 3, 3, 9 ]<br />
131 #U = 18 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 3, 3, 18 ]<br />
132 #U = 18 [ 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 3, 3, 18 ]<br />
133 #U = 18 [ 2, 3 ], #H^1 = 4 [ 2, 2 ], Rk(Pic) = 1, Orbits [ 6, 6, 6, 9 ]<br />
134 #U = 18 [ 2, 3, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 3, 6, 6, 9 ]<br />
135 #U = 18 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 3, 6, 6, 9 ]<br />
136 #U = 18 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
137 #U = 18 [ 2 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 9, 9, 9 ]<br />
138 #U = 18 [ 2 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 3, 6, 9, 9 ]<br />
139 #U = 18 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 6, 18 ]<br />
140 #U = 18 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 3, 6, 6, 9 ]<br />
141 #U = 18 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 6, 6, 6, 9 ]<br />
142 #U = 18 [ 2, 3, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 6, 18 ]<br />
143 #U = 18 [ 2, 3 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 3, 6, 9, 9 ]<br />
144 #U = 18 [ 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 9, 9 ]<br />
145 #U = 18 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
146 #U = 20 [ 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 2, 5, 10, 10 ]<br />
147 #U = 20 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 2, 5, 5, 5, 10 ]<br />
148 #U = 20 [ 4 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 5, 5, 5, 10 ]<br />
149 #U = 24 [ 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
150 #U = 24 [ 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 2, 6, 8, 8 ]<br />
151 #U = 24 [ 2, 3 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 2, 6, 8, 8 ]<br />
152 #U = 24 [ 2 ], #H^1 = 1, Rk(Pic) = 4, Orbits [ 1, 1, 1, 1, 1, 4, 4, 4, 4, 6 ]<br />
153 #U = 24 [ 2, 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
154 #U = 24 [ 3 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 8, 8, 8 ]<br />
155 #U = 24 [ 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
156 #U = 24 [ 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 4, 4, 6, 8 ]<br />
157 #U = 24 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 6, 6, 12 ]<br />
158 #U = 24 [ 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 12, 12 ]<br />
159 #U = 24 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 6, 6, 12 ]<br />
160 #U = 24 [ 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 6, 8, 8 ]<br />
161 #U = 24 [ 2, 3 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 12, 12 ]<br />
162 #U = 24 [ 2, 3 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 12, 12 ]
APPENDIX 359<br />
163 #U = 24 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 6, 6, 12 ]<br />
164 #U = 24 [ 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 12, 12 ]<br />
165 #U = 24 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 4, 4, 6, 8 ]<br />
166 #U = 24 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 6, 8, 8 ]<br />
167 #U = 24 [ 2, 2, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 4, 6, 12 ]<br />
168 #U = 24 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 3, 3, 4, 6, 6 ]<br />
169 #U = 24 [ 8 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
170 #U = 24 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 2, 4, 6, 12 ]<br />
171 #U = 24 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 4, 6, 12 ]<br />
172 #U = 24 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 12, 12 ]<br />
173 #U = 24 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 4, 6, 12 ]<br />
174 #U = 24 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 4, 6, 12 ]<br />
175 #U = 24 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 4, 6, 12 ]<br />
176 #U = 24 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 6, 6, 12 ]<br />
177 #U = 27 [ 3, 3 ], transitive<br />
178 #U = 27 [ 3, 3, 3 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 9, 9, 9 ]<br />
179 #U = 27 [ 3, 3 ], transitive<br />
180 #U = 32 [ 2, 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 8, 8, 8 ]<br />
181 #U = 32 [ 2, 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 2, 2, 4, 16 ]<br />
182 #U = 32 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 8, 8, 8 ]<br />
183 #U = 32 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 16 ]<br />
184 #U = 32 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
185 #U = 32 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 16 ]<br />
186 #U = 32 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
187 #U = 32 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 4, 4, 8, 8 ]<br />
188 #U = 32 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 16 ]<br />
189 #U = 32 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 16 ]<br />
190 #U = 32 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
191 #U = 32 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
192 #U = 32 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 16 ]<br />
193 #U = 32 [ 2, 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 8, 8 ]<br />
194 #U = 32 [ 2, 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 16 ]<br />
195 #U = 36 [ 4 ], #H^1 = 4 [ 2, 2 ], Rk(Pic) = 1, Orbits [ 6, 6, 6, 9 ]<br />
196 #U = 36 [ 2, 2 ], #H^1 = 4 [ 2, 2 ], Rk(Pic) = 1, Orbits [ 6, 6, 6, 9 ]<br />
197 #U = 36 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 3, 3, 3, 3, 3, 3, 9 ]<br />
198 #U = 36 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 3, 3, 18 ]<br />
199 #U = 36 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 3, 6, 6, 9 ]<br />
200 #U = 36 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 3, 6, 6, 9 ]<br />
201 #U = 36 [ 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 6, 6, 6, 9 ]<br />
202 #U = 36 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 6, 6, 6, 9 ]<br />
203 #U = 36 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 6, 9, 12 ]<br />
204 #U = 36 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 6, 18 ]<br />
205 #U = 36 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 6, 18 ]<br />
206 #U = 36 [ 2, 2, 3 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 6, 9, 12 ]<br />
207 #U = 36 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 6, 18 ]<br />
208 #U = 36 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 3, 6, 6, 9 ]<br />
209 #U = 36 [ 2, 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 6, 18 ]<br />
210 #U = 36 [ 2, 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 6, 18 ]<br />
211 #U = 36 [ 2, 2 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 3, 6, 9, 9 ]<br />
212 #U = 36 [ 2, 2, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 3, 6, 6, 9 ]<br />
213 #U = 36 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
214 #U = 40 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 2, 5, 10, 10 ]<br />
215 #U = 48 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 2, 6, 16 ]<br />
216 #U = 48 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
217 #U = 48 [ 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 6, 16 ]<br />
218 #U = 48 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 2, 2, 6, 8, 8 ]<br />
219 #U = 48 [ 2, 2, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 2, 6, 16 ]<br />
220 #U = 48 [ 3, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 6, 16 ]<br />
221 #U = 48 [ 2, 2, 3 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 4, 6, 8, 8 ]<br />
222 #U = 48 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 4, 6, 8, 8 ]<br />
223 #U = 48 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 12, 12 ]<br />
224 #U = 48 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 2, 2, 4, 4, 6, 8 ]<br />
225 #U = 48 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 2, 6, 8, 8 ]<br />
226 #U = 48 [ 2, 2, 3 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 12, 12 ]<br />
227 #U = 48 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 12, 12 ]
360 APPENDIX<br />
228 #U = 48 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 12, 12 ]<br />
229 #U = 48 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 6, 6, 12 ]<br />
230 #U = 48 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 2, 6, 16 ]<br />
231 #U = 48 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 4, 6, 8, 8 ]<br />
232 #U = 48 [ 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
233 #U = 48 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 12, 12 ]<br />
234 #U = 48 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 12, 12 ]<br />
235 #U = 48 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
236 #U = 48 [ 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
237 #U = 48 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
238 #U = 48 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 4, 6, 12 ]<br />
239 #U = 54 [ 2 ], transitive<br />
240 #U = 54 [ 2 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 9, 9, 9 ]<br />
241 #U = 54 [ 2, 3 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 9, 9, 9 ]<br />
242 #U = 54 [ 2, 3, 3 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 9, 9, 9 ]<br />
243 #U = 54 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
244 #U = 54 [ 2, 3, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
245 #U = 54 [ 2, 3 ], transitive<br />
246 #U = 54 [ 2, 3 ], transitive<br />
247 #U = 60 [ ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 6, 6, 15 ]<br />
248 #U = 60 [ ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 5, 5, 5, 10 ]<br />
249 #U = 64 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 8, 8, 8 ]<br />
250 #U = 64 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
251 #U = 64 [ 2, 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 4, 4, 16 ]<br />
252 #U = 64 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
253 #U = 64 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
254 #U = 64 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
255 #U = 64 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
256 #U = 72 [ 2, 2 ], #H^1 = 4 [ 2, 2 ], Rk(Pic) = 1, Orbits [ 6, 6, 6, 9 ]<br />
257 #U = 72 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 6, 9, 12 ]<br />
258 #U = 72 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 6, 9, 12 ]<br />
259 #U = 72 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 6, 6, 6, 9 ]<br />
260 #U = 72 [ 3, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
261 #U = 72 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 6, 9, 12 ]<br />
262 #U = 72 [ 2, 4 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 6, 9, 12 ]<br />
263 #U = 72 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 3, 3, 6, 6, 9 ]<br />
264 #U = 72 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 6, 18 ]<br />
265 #U = 80 [ 5 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 10, 16 ]<br />
266 #U = 81 [ 3, 3 ], transitive<br />
267 #U = 96 [ 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
268 #U = 96 [ 3 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 8, 8, 8 ]<br />
269 #U = 96 [ 2, 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
270 #U = 96 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 2, Orbits [ 1, 4, 6, 8, 8 ]<br />
271 #U = 96 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 2, 6, 16 ]<br />
272 #U = 96 [ 4 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
273 #U = 96 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
274 #U = 96 [ 2, 2, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 6, 16 ]<br />
275 #U = 96 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 6, 16 ]<br />
276 #U = 96 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 6, 16 ]<br />
277 #U = 96 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 6, 16 ]<br />
278 #U = 96 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 6, 16 ]<br />
279 #U = 96 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 3, 12, 12 ]<br />
280 #U = 108 [ 2, 2 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 9, 9, 9 ]<br />
281 #U = 108 [ 2, 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
282 #U = 108 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
283 #U = 108 [ 2, 2, 3 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 9, 9, 9 ]<br />
284 #U = 108 [ 2, 2 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 9, 9, 9 ]<br />
285 #U = 108 [ 4 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
286 #U = 108 [ 4 ], transitive<br />
287 #U = 108 [ 3, 4 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
288 #U = 108 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
289 #U = 108 [ 2, 2 ], transitive<br />
290 #U = 108 [ 2, 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
291 #U = 120 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 6, 6, 15 ]<br />
292 #U = 120 [ 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 12, 15 ]
APPENDIX 361<br />
293 #U = 120 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 2, 5, 10, 10 ]<br />
294 #U = 120 [ 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 5, 5, 5, 10 ]<br />
295 #U = 120 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 2, 5, 10, 10 ]<br />
296 #U = 120 [ 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 12, 15 ]<br />
297 #U = 128 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
298 #U = 144 [ 2, 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 6, 9, 12 ]<br />
299 #U = 144 [ 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
300 #U = 160 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 10, 16 ]<br />
301 #U = 162 [ 2, 3 ], transitive<br />
302 #U = 162 [ 2, 3 ], transitive<br />
303 #U = 162 [ 2 ], transitive<br />
304 #U = 192 [ 2 ], #H^1 = 1, Rk(Pic) = 3, Orbits [ 1, 1, 1, 8, 8, 8 ]<br />
305 #U = 192 [ 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
306 #U = 192 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
307 #U = 192 [ 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
308 #U = 192 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
309 #U = 192 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
310 #U = 192 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
311 #U = 192 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 4, 6, 16 ]<br />
312 #U = 216 [ 2, 2, 2 ], #H^1 = 3, Rk(Pic) = 1, Orbits [ 9, 9, 9 ]<br />
313 #U = 216 [ 2, 2 ], transitive<br />
314 #U = 216 [ 2, 4 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
315 #U = 216 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
316 #U = 216 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
317 #U = 216 [ 2, 2 ], transitive<br />
318 #U = 216 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
319 #U = 216 [ 8 ], transitive<br />
320 #U = 216 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
321 #U = 216 [ 2, 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
322 #U = 240 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 2, 5, 10, 10 ]<br />
323 #U = 240 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 12, 15 ]<br />
324 #U = 288 [ 3, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
325 #U = 320 [ 4 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 10, 16 ]<br />
326 #U = 324 [ 3 ], transitive<br />
327 #U = 324 [ 2, 2 ], transitive<br />
328 #U = 360 [ ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 6, 6, 15 ]<br />
329 #U = 384 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
330 #U = 384 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 2, 8, 16 ]<br />
331 #U = 432 [ 2, 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 9, 18 ]<br />
332 #U = 432 [ 2, 2 ], transitive<br />
333 #U = 576 [ 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
334 #U = 576 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
335 #U = 576 [ 2, 3 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
336 #U = 648 [ 2 ], transitive<br />
337 #U = 648 [ 2 ], transitive<br />
338 #U = 648 [ 2, 3 ], transitive<br />
339 #U = 648 [ 3 ], transitive<br />
340 #U = 720 [ 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 12, 15 ]<br />
341 #U = 720 [ 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 12, 15 ]<br />
342 #U = 720 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 6, 6, 15 ]<br />
343 #U = 960 [ ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 10, 16 ]<br />
344 #U = 1152 [ 2, 2 ], #H^1 = 1, Rk(Pic) = 1, Orbits [ 3, 24 ]<br />
345 #U = 1296 [ 2 ], transitive<br />
346 #U = 1296 [ 2, 2 ], transitive<br />
347 #U = 1440 [ 2, 2 ], #H^1 = 2, Rk(Pic) = 1, Orbits [ 12, 15 ]<br />
348 #U = 1920 [ 2 ], #H^1 = 1, Rk(Pic) = 2, Orbits [ 1, 10, 16 ]<br />
349 #U = 25920 [ ], transitive<br />
350 #U = 51840 [ 2 ], transitive
BIBLIOGRAPHY<br />
[A/B]<br />
[A/V]<br />
[Al]<br />
[AMD]<br />
[Am55]<br />
Abbes, A. et Bouche, T.: Théorème de Hilbert-Samuel “arithmétique”,<br />
Ann. Inst. Fourier 45(1995), 375–401<br />
Abramovich, D. <strong>and</strong> Voloch, J. F.: Lang’s conjectures, fibered powers,<br />
<strong>and</strong> uniformity, New York J. Math. 2(1996), 20–34<br />
Albert, A.: Structure of Algebras, Revised Printing, Amer. Math.<br />
Soc. Coll. Pub., Vol. XXIV, Amer.Math. Soc., Providence, R.I. 1961<br />
Software Optimization Guide for AMD Athlon TM 64 <strong>and</strong> AMD<br />
Opteron TM Processors, Rev. 3.04, AMD, Sunnyvale (CA) 2004<br />
Amitsur, S. A.: Generic splitting fields of central simple algebras,<br />
Annals of Math. 62(1955), 8–43<br />
[Am72] Amitsur, S. A.: On central division algebras, Israel J. Math. 12<br />
(1972), 408–420<br />
[Am81]<br />
[Ap]<br />
[Ara]<br />
Amitsur, S. A.: Generic splitting fields, in: van Oystaeyen, F. <strong>and</strong><br />
Verschoren, A. (Eds.): Brauer <strong>groups</strong> in ring theory <strong>and</strong> algebraic<br />
geometry, Proceedings of a conference held at Antwerp 1981, Lecture<br />
Notes Math. 917, Springer, Berlin, Heidelberg, New York<br />
ÖÐÓڸ˺ºÌÓÖÔÖ×ÕÒÚÞÓÖÓÚÒÖѹ<br />
1982, 1–24<br />
´½µ¸½½ß½½¾¸ ØÕ×ÓÔÓÚÖÒÓ×ظÁÞÚººÆÙËËËʸËÖºÅغ¿<br />
Apéry, F.: Models of the real projective plane, Vieweg, Braunschweig<br />
1987<br />
English translation: Arakelov, S. Yu.: Intersection theory of divisors<br />
on an arithmetic surface, Math. USSR Izv. 8(1974), 1167–1180
364 BIBLIOGRAPHY<br />
[ArE27a]<br />
[ArE27b]<br />
[A/N/T]<br />
Artin, E.: Über einen Satz von J. H. Maclagan-Wedderburn, Abh.<br />
Math. Sem. Univ. Hamburg 5(1927), 245–250<br />
Artin, E.: Zur Theorie der hyperkomplexen Zahlen, Abh. Math.<br />
Sem. Univ. Hamburg 5(1927), 251–260<br />
Artin, E., Nesbitt, C. J., <strong>and</strong> Thrall, R.: Rings with minimum<br />
condition, Univ. of Michigan Press, Ann Arbor 1948<br />
[ArM81a] Artin, M.: Left Ideals in Maximal Orders, in: van Oystaeyen, F.<br />
<strong>and</strong> Verschoren, A. (Eds.): Brauer <strong>groups</strong> in ring theory <strong>and</strong> algebraic<br />
geometry, Proceedings of a conference held at Antwerp<br />
1981, Lecture Notes Math. 917, Springer, Berlin, Heidelberg, New<br />
York 1982, 182–193<br />
[ArM81b]<br />
[A/M]<br />
[A/G]<br />
Artin, M. (Notes by Verschoren, A.): Brauer-Severi varieties, in:<br />
van Oystaeyen, F. <strong>and</strong> Verschoren, A. (Eds.): Brauer <strong>groups</strong> in<br />
ring theory <strong>and</strong> algebraic geometry, Proceedings of a conference<br />
held at Antwerp 1981, Lecture Notes Math. 917, Springer, Berlin,<br />
Heidelberg, New York 1982, 194–210<br />
Artin, M. <strong>and</strong> Mumford, D.: Some elementary examples of uni<strong>rational</strong><br />
varieties which are not <strong>rational</strong>, Proc. London Math. Soc. 25<br />
(1972), 75–95<br />
Ausl<strong>and</strong>er, M. <strong>and</strong> Goldman, O.: The Brauer group of a commutative<br />
ring, Trans. Amer. Math. Soc. 97(1960), 367–409<br />
[Az] Azumaya, G.: On maximally central algebras, Nagoya Math. J. 2<br />
(1951), 119–150<br />
[B/H/P/V] Barth, W., Hulek, K., Peters, C., <strong>and</strong> Van de Ven, A.: Compact<br />
complex surfaces, Second edition, Springer, Berlin 2004<br />
[B/M]<br />
[Ba/T]<br />
[Bv]<br />
Batyrev, V. V. <strong>and</strong> Manin, Yu. I.: Sur le nombre des <strong>points</strong> rationnels<br />
de hauteur borné des variétés algébriques, Math. Ann. 286<br />
(1990), 27–43<br />
Batyrev, V. V. <strong>and</strong> Tschinkel, Y.: Rational <strong>points</strong> on some Fano<br />
cubic bundles, C. R. Acad. Sci. Paris 323(1996), 41–46<br />
Beauville, A.: Complex algebraic surfaces, LMS Lecture Note Series<br />
68, Cambridge University Press, Cambridge 1983
BIBLIOGRAPHY 365<br />
[B/P/T/Y] Beck, M., Pine, E., Tarrant, W., <strong>and</strong> Yarbrough Jensen, K.: New<br />
integer representations as the sum of three cubes, Math. Comp. 76<br />
(2007), 1683–1690<br />
[B-gh]<br />
van den Bergh, M.: The Brauer-Severi scheme of the trace ring of<br />
generic matrices, in: Perspectives in ring theory (Antwerp 1987),<br />
NATO Adv. Sci. Inst. Ser. C: Math. Phys. Sci. 233, Kluwer Acad.<br />
Publ., Dordrecht 1988, 333–338<br />
[Be] Bernstein, D. J.: Enumerating solutions to p(a) + q(b) =<br />
r(c) + s(d), Math. Comp. 70(2001), 389–394<br />
[Be-list]<br />
Bernstein, D. J.: threecubes, Web page available at the URL:<br />
http://cr.yp.to/threecubes.html<br />
[Bt] Bertuccioni, I.: Brauer <strong>groups</strong> <strong>and</strong> cohomology, Arch. Math. 84<br />
ÐÓÖÓÚ׸ººÐÙ¸º℄ÊÚÒÓÑÖÒÓÖ×ÔÖ¹<br />
(2005), 406–411<br />
ÐÒÐÖÕ×Õ×ÐÐÞÒÕÒÓÑÙÖÙÙ¸<br />
[BB] Białynicki-Birula, A.: A note on deformations of Severi-Brauer<br />
varieties<br />
¹¾¸½¾ Î׬ºÆÚÙËËʸËÖº¬Þº¹ÅغÆÚÙ½´½µ¸<br />
<strong>and</strong> relatively minimal models of fields of genus 0, Bull.<br />
Acad. Polon. Sci. Sér. Sci. Math. Astronom. Phys. 18(1970), 175–176<br />
[Bilu88]<br />
[Bilu97]<br />
Bilu, Yu.: Limit distribution of small <strong>points</strong> on algebraic tori, Duke<br />
Math. J. 89(1997), 465-476<br />
[Bill] Billard, H.: Propriétés arithmétiques d’une famille de surfaces K3,<br />
Compositio Math. 108(1997), 247–275<br />
[Bir] Birch, B. J.: Forms in many variables, Proc. Roy. Soc. Ser. A 265<br />
(1961/1962), 245–263<br />
[Bl-ard]<br />
[Bl-et]<br />
[Bo/T]<br />
Blanchard, A.: Les corps non commutatifs, Presses Univ. de France,<br />
Vendôme 1972<br />
Blanchet, A.: Function fields of generalized Brauer-Severi varieties,<br />
Comm. in Algebra 19(1991), 97–118<br />
Bogomolov, F. <strong>and</strong> Tschinkel, Y.: Rational curves <strong>and</strong> <strong>points</strong> on<br />
K3 surfaces, Amer. J. Math. 127(2005), 825–835
ÓÖÚÕ¸ºÁºÖÚÕ¸ÁºÊºÌÓÖÕ×иÌÖØ ÞÒ¸ÆÙ¸ÅÓ×Ú½<br />
366 BIBLIOGRAPHY<br />
[B/S]<br />
English translation: Borevich, Z. I. <strong>and</strong> Shafarevich, I. R.: Number<br />
theory, Academic Press, New York-London 1966<br />
[B/G/S]<br />
Bost, J.-B., Gillet, H., Soulé, C.: Heights of projective varieties <strong>and</strong><br />
positive Green forms, J. Amer. Math. Soc. 7(1994), 903-1027<br />
[Bot] Bott, R.: On a theorem of Lefschetz, Michigan Math. J. 6(1959),<br />
211–216<br />
[Bou-T]<br />
[Bou-A]<br />
Bourbaki, N.: Éléments de Mathématique, Livre III: Topologie<br />
générale, Chapitre IX, Hermann, Paris 1948<br />
Bourbaki, N.: Éléments de Mathématique, Livre II: Algèbre,<br />
Chapitre VIII, Hermann, Paris 1958<br />
[Bra] Brauer, R.: Über Systeme hyperkomplexer Zahlen, Math. Z. 30<br />
(1929), 79–107<br />
[Bre]<br />
[Brü]<br />
[Bu]<br />
[B/K/K]<br />
[Car]<br />
[C/E]<br />
[Cas55]<br />
[Cas67]<br />
Bremner, A.: Some cubic surfaces with no <strong>rational</strong> <strong>points</strong>, Math.<br />
Proc. Cambridge Philos. Soc. 84(1978), 219–223<br />
Brüdern, J.: Einführung in die analytische Zahlentheorie, Springer,<br />
Berlin 1995<br />
Burgos Gil, J. I.: Arithmetic Chow rings, Ph.D. thesis, Barcelona<br />
1994<br />
Burgos Gil, J. I., Kramer, J., <strong>and</strong> Kühn, U.: Cohomological arithmetic<br />
Chow rings, J. Inst. Math. Jussieu 6(2007), 1–172<br />
Cartan, E.: Œvres complètes, Partie II, Volume I: Nombres complexes,<br />
Gauthier-Villars, 107–246<br />
Cartan, H. <strong>and</strong> Eilenberg, S.: Homological Algebra, Princeton<br />
Univ. Press, Princeton 1956<br />
Cassels, J. W. S.: Bounds for the least solutions of homogeneous<br />
quadratic equations, Proc. CambridgePhilos. Soc. 51(1955), 262–264<br />
Cassels, J. W. S.: Global fields, in: Algebraic number theory, Edited<br />
by J. W. S. Cassels <strong>and</strong> A. Fröhlich, Academic Press <strong>and</strong> Thompson<br />
Book Co., London <strong>and</strong> Washington 1967
BIBLIOGRAPHY 367<br />
[Ca/G]<br />
[C-L]<br />
[Cha]<br />
Cassels, J. W. S. <strong>and</strong> Guy, M. J. T.: On the Hasse principle for<br />
cubic surfaces, Mathematika 13(1966), 111–120.<br />
Chambert-Loir, A.: Points de petite hauteur sur les variétés semiabéliennes,<br />
Ann. Sci. École Norm. Sup. 33(2000), 789–821<br />
Chase, S. U.: Two remarks on central simple algebras, Comm. in<br />
Algebra 12(1984), 2279–2289<br />
[Ch44] Châtelet, F.: Variations sur un thème de H. Poincaré, Annales E.<br />
N. S. 61(1944), 249–300<br />
[Ch54/55]<br />
[Cl/G]<br />
[Coh]<br />
[CT83]<br />
[CT88]<br />
[CT91]<br />
[CT/K/S]<br />
[CT/S]<br />
[CT/SD]<br />
Châtelet, F.: Géométrie diophantienne et théorie des algèbres,<br />
Séminaire Dubreil, 1954–55, exposé 17<br />
Clemens, C. H. <strong>and</strong> Griffiths, P. A.: The intermediate Jacobian of<br />
the cubic threefold, Annals of Math. 95(1972), 281–356<br />
Cohen, H.: A course in computational algebraic number theory,<br />
Springer, Berlin, Heidelberg 1993<br />
Colliot-Thélène, J.-L.: Hilbert’s theorem 90 for K 2 , with application<br />
to the Chow group of <strong>rational</strong> surfaces, Invent. Math. 35<br />
(1983), 1–20<br />
Colliot-Thélène, J.-L.: Les gr<strong>and</strong>s thèmes de François Châtelet,<br />
L’Enseignement Mathématique 34(1988), 387–405<br />
Colliot-Thélène, J.-L.: Cycles algébriques de torsion et K-théorie<br />
algébrique, in: Arithmetic Algebraic Geometry (Trento 1991), Lecture<br />
Notes Math. 1553, Springer, Berlin, Heidelberg 1993, 1–49<br />
Colliot-Thélène, J.-L., Kanevsky, D., <strong>and</strong> Sansuc, J.-J.: Arithmétique<br />
des surfaces cubiques diagonales, in: Diophantine approximation<br />
<strong>and</strong> transcendence theory (Bonn 1985), Lecture Notes in<br />
Math. 1290, Springer, Berlin 1987, 1–108<br />
Colliot-Thélène, J.-L. <strong>and</strong> Sansuc, J.-J.: On the Chow <strong>groups</strong> of<br />
certain <strong>rational</strong> surfaces: A sequel to a paper of S. Bloch, Duke<br />
Math. J. 48(1981), 421–447<br />
Colliot-Thélène, J.-L. <strong>and</strong> Swinnerton-Dyer, Sir P.: Hasse principle<br />
<strong>and</strong> weak approximation for pencils of Severi-Brauer <strong>and</strong> similar<br />
varieties, J. für die Reine und Angew. Math. 453(1994), 49–112
368 BIBLIOGRAPHY<br />
[C/L/R]<br />
[Cor]<br />
[C/S]<br />
Cormen, T., Leiserson, C., <strong>and</strong> Rivest, R.: Introduction to algorithms,<br />
MIT Press <strong>and</strong> McGraw-Hill, Cambridge <strong>and</strong> New York<br />
1990<br />
Corn, P. K.: Del Pezzo surfaces <strong>and</strong> the Brauer-Manin obstruction,<br />
Ph.D. thesis, Harvard 2005<br />
Cornell, G. <strong>and</strong> Silverman, J. H.: Arithmetic geometry, Papers<br />
from the conference held at the University of Connecticut,<br />
Springer, New York 1986<br />
[Del] Deligne, P.: La conjecture de Weil I, Publ. Math. IHES 43(1974),<br />
273–307<br />
[Der]<br />
[Deu]<br />
[Di04]<br />
[Di05]<br />
[Dr]<br />
[Duk]<br />
Derenthal, U.: Geometry of universal torsors, Ph.D. thesis, Göttingen<br />
2006<br />
Deuring, M.: Algebren, Springer, Ergebnisse der Math. und ihrer<br />
Grenzgebiete, Berlin 1935<br />
Dickson, L. E.: Determination of all the sub<strong>groups</strong> of the known<br />
simple group of order 25 920, Trans. Amer. Math. Soc. 5(1904), 126–<br />
166<br />
Dickson, L. E.: On finite algebras, Nachrichten von der Königlichen<br />
Gesellschaft der Wissenschaften zu Göttingen, Math.-Phys. Klasse,<br />
1905, 358–393<br />
Draxl, P. K.: Skew fields, London Math. Soc. Lecture Notes Series<br />
81, Cambridge University Press, Cambridge 1983<br />
Duke, W.: Extreme values of Artin L-functions <strong>and</strong> class numbers,<br />
Compositio Math. 136(2003), 103–115<br />
[E/H/K/V] Edidin, D., Hassett, B., Kresch, A., <strong>and</strong> Vistoli, A.: Brauer <strong>groups</strong><br />
<strong>and</strong> quotient stacks, Amer. J. Math. 123(2001), 761–777<br />
[Ek]<br />
[EGA]<br />
Ekedahl, T.: An effective version of Hilbert’s irreducibility theorem,<br />
in: Séminaire de Théorie des Nombres, Paris 1988–1989,<br />
Progr. Math. 91, Birkhäuser, Boston 1990, 241–249<br />
Grothendieck, A. et Dieudonné, J.: Éléments de Géométrie Algébrique,<br />
Publ. Math. IHES 4, 8, 11, 17, 20, 24, 28, 32(1960–68)
BIBLIOGRAPHY 369<br />
[El]<br />
[EJ1]<br />
[EJ2]<br />
[EJ3]<br />
[EJ4]<br />
[EJ5]<br />
[EJ6]<br />
[EJ7]<br />
[EJ8]<br />
[EJ9]<br />
[Fa83]<br />
Elkies, N. D.: Rational <strong>points</strong> near curves <strong>and</strong> small nonzero<br />
|x 3 − y 2 | via lattice reduction, in: Algorithmic number theory<br />
(Leiden 2000), Lecture Notes in Computer Science 1838, Springer,<br />
Berlin 2000, 33–63<br />
Elsenhans, A.-S. <strong>and</strong> Jahnel, J.: The Fibonacci sequence modulo<br />
p 2 —An investigation by computer for p < 10 14 , Preprint<br />
Elsenhans, A.-S. <strong>and</strong> Jahnel, J.: The Diophantine equation<br />
x 4 + 2y 4 = z 4 + 4w 4 , Math. Comp. 75(2006), 935–940<br />
Elsenhans, A.-S. <strong>and</strong> Jahnel, J.: The Diophantine equation<br />
x 4 + 2y 4 = z 4 + 4w 4 — a number of improvements, Preprint<br />
Elsenhans, A.-S. <strong>and</strong> Jahnel, J.: The Asymptotics of Points of<br />
Bounded Height on Diagonal Cubic <strong>and</strong> Quartic Threefolds, in:<br />
Algorithmic number theory, Lecture Notes in Computer Science<br />
4076, Springer, Berlin 2006, 317–332<br />
Elsenhans, A.-S. <strong>and</strong> Jahnel, J.: On the smallest point on a diagonal<br />
quartic threefold, J. Ramanujan Math. Soc. 22(2007), 189–204<br />
Elsenhans, A.-S. <strong>and</strong> Jahnel, J.: Experiments with general cubic<br />
surfaces, to appear in: The Manin Festschrift<br />
Elsenhans, A.-S. <strong>and</strong> Jahnel, J.: On the smallest point on a diagonal<br />
cubic surface, Preprint<br />
Elsenhans, A.-S. <strong>and</strong> Jahnel, J.: New sums of three cubes, to appear<br />
in: Math. Comp.<br />
Elsenhans, A.-S. <strong>and</strong> Jahnel, J.: List of solutions of x 3 +y 3 +z 3 = n<br />
for n < 1 000 neither a cube nor twice a cube, Web page available at<br />
the URL: http://www.uni-math.gwdg.de/jahnel/Arbeiten/<br />
Liste/threecubes_20070419.txt<br />
Faltings, G.: Endlichkeitssätze für abelsche Varietäten über Zahlkörpern,<br />
Invent. Math. 73(1983), 349–366<br />
[Fa84] Faltings, G.: Calculus on arithmetic surfaces, Annals of Math. 119<br />
(1984), 387–424<br />
[Fa91]<br />
Faltings, G.: The general case of S. Lang’s conjecture, in: Barsotti<br />
Symposium inAlgebraicGeometry(AbanoTerme1991), Perspect.<br />
Math. 15, Academic Press, San Diego 1994, 175–182
370 BIBLIOGRAPHY<br />
[Fa92]<br />
[F/P]<br />
[Fo]<br />
[F/M/T]<br />
[Fu84]<br />
[Ga]<br />
[G/T]<br />
[G/S 90]<br />
[G/S 92]<br />
[Gi]<br />
[G/H]<br />
[GrBrI]<br />
[GrBrII]<br />
Faltings, G.: Lectures on the arithmetic Riemann-Roch theorem,<br />
Annals of Mathematics Studies 127, Princeton University Press,<br />
Princeton 1992<br />
Fincke, U. <strong>and</strong> Pohst, M.: Improved methods for calculating vectors<br />
of short length in a lattice, including a complexity analysis,<br />
Math. Comp. 44(1985), 463–471<br />
Forster, O.: Algorithmische Zahlentheorie, Vieweg, Braunschweig<br />
1996<br />
Franke, J., Manin, Yu. I., <strong>and</strong> Tschinkel, Y.: Rational <strong>points</strong> of<br />
bounded height on Fano varieties, Invent. Math. 95(1989), 421–435<br />
Fulton, W.: Intersection theory, Springer, Ergebnisse der Math.<br />
und ihrer Grenzgebiete, 3. Folge, B<strong>and</strong> 2, Berlin 1984<br />
Gabber, O.: Some theorems on Azumaya algebras, in: The Brauer<br />
group (Les Plans-sur-Bex 1980), Lecture Notes Math. 844, Springer,<br />
Berlin, Heidelberg, New York 1981, 129–209<br />
Gatsinzi, J.-B. <strong>and</strong> Tignol, J.-P.: Involutions <strong>and</strong> Brauer-Severi varieties,<br />
Indag. Math. 7(1996), 149–160<br />
Gillet, H., Soulé, C.: Arithmetic intersection theory, Publ. Math.<br />
IHES 72(1990), 93-174<br />
Gillet, H., Soulé, C.: An arithmetic Riemann-Roch theorem, Invent.<br />
Math. 110(1992), 473–543<br />
Giraud, J.: Cohomologie non abélienne, Springer,Grundlehren der<br />
math. Wissenschaften 179, Berlin, Heidelberg, New York 1971<br />
Griffiths, P. <strong>and</strong> Harris, J.: Principles of algebraic geometry, John<br />
Wiley & Sons, New York 1978<br />
Grothendieck, A.: Le groupe de Brauer, I: Algèbres d’Azumaya et<br />
interprétations diverses, Séminaire Bourbaki, 17 e année 1964/65,<br />
n o 290<br />
Grothendieck, A.: Le groupe de Brauer, II: Théorie cohomologique,<br />
Séminaire Bourbaki, 18 e année 1965/66, n o 297
BIBLIOGRAPHY 371<br />
[GrBrIII]<br />
[Hab]<br />
[Hai]<br />
[Hb]<br />
[H/W]<br />
[Ha70]<br />
[Ha77]<br />
[H-B92a]<br />
[H-B92b]<br />
[H/L/R]<br />
[H/P]<br />
[Hen]<br />
Grothendieck, A.: Le groupe de Brauer, III: Exemples et compléments,<br />
in: Grothendieck, A.: Dix exposés sur la Cohomologie<br />
des schémas, North-Holl<strong>and</strong>, Amsterdam <strong>and</strong> Masson, Paris 1968,<br />
88–188<br />
Haberl<strong>and</strong>, K.: Galois cohomology of algebraic number fields,<br />
With two appendices by H. Koch <strong>and</strong> Th. Zink, VEB Deutscher<br />
Verlag der Wissenschaften, Berlin 1978<br />
Haile, D. E.: On central simple algebras of given exponent, J. of<br />
Algebra 57(1979), 449–465<br />
H<strong>and</strong>book of numerical analysis, edited by P. G. Ciarlet <strong>and</strong><br />
J. L. Lions, Vols. II–IV, North-Holl<strong>and</strong> Publishing Co., Amsterdam<br />
1991–1996<br />
Hardy, G. H. <strong>and</strong> Wright, E. M.: An introduction to the theory of<br />
numbers, Fifth edition, Oxford University Press, New York 1979<br />
Hartshorne, R.: Ample subvarieties of algebraic varieties, Lecture<br />
Notes Math. 156, Springer, Berlin-New York 1970<br />
Hartshorne, R.: Algebraic Geometry, Graduate Texts in Mathematics<br />
52, Springer, New York 1977<br />
Heath-Brown, D. R.: The density of zeros of forms for which<br />
weak approximation fails, Math. Comp. 59(1992), 613–623<br />
Heath-Brown, D. R.: Zero-free regions for Dirichlet L-functions,<br />
<strong>and</strong> the least prime in an arithmetic progression, Proc. London<br />
Math. Soc. 64(1992), 265–338<br />
Heath-Brown, D. R., Lioen, W. M., <strong>and</strong> te Riele, H. J. J.: On<br />
solving the diophantine equation x 3 + y 3 + z 3 = k on a vector<br />
computer, Math. Comp. 61(1993), 235–244<br />
Hennessy, J. L. <strong>and</strong> Patterson, D. A.: Computer Architecture:<br />
A Quantitative Approach, 2 nd ed., Morgan Kaufmann, San Mateo<br />
(CA) 1996<br />
Henningsen, F.: Brauer-Severi-Varietäten und Normrelationen<br />
von Symbolalgebren, Doktorarbeit (Ph. D. thesis), available in electronic<br />
form at http://www.biblio.tu-bs.de/ediss/data/<br />
20000713a/20000713a.pdf
372 BIBLIOGRAPHY<br />
[Herm]<br />
[Hers]<br />
[Heu]<br />
[Hi]<br />
[H/W]<br />
[H/J/S]<br />
[Hoo]<br />
[I/R]<br />
[Is]<br />
[Jac]<br />
[J1]<br />
Hermann, G.: Die Frage der endlich vielen Schritte in der Theorie<br />
der Polynomideale, Math. Ann. 95(1926), 736–788<br />
Herstein, I. N.: Noncommutative rings, The Math. Association of<br />
America (Inc.), distributed by John Wiley <strong>and</strong> Sons, Menasha, WI<br />
1968<br />
Heuser, A.: Über den Funktionenkörper der Normfläche einer<br />
zentral einfachen Algebra, J. für die Reine und Angew. Math. 301<br />
(1978), 105–113<br />
Hirzebruch, F.: Der Satz von Riemann-Roch in Faisceau-theoretischer<br />
Formulierung: Einige Anwendungen und offene Fragen,<br />
Proc. Int. Cong. of Mathematicians, Amsterdam 1954, vol. III,<br />
457–473, Erven P. Noordhoff N.V. <strong>and</strong> North-Holl<strong>and</strong> Publishing<br />
Co., Groningen <strong>and</strong> Amsterdam 1956<br />
Hoffman, J. W. <strong>and</strong> Weintraub, S. H.: Four dimensional symplectic<br />
geometry over the field with three elements <strong>and</strong> a moduli space of<br />
abelian surfaces, Note Mat. 20(2000/01), 111–157<br />
Hoffmann, N., Jahnel, J., <strong>and</strong> Stuhler, U.: Generalized vector bundles<br />
on curves, J. für die Reine und Angew. Math. 495(1998), 35-60<br />
Hoobler, R. T.: A cohomological interpretation of Brauer <strong>groups</strong><br />
Á×ÓÚ׸κºÃÓÒØÖÔÖÑÖÔÖÒÔÙÀ××Ð××¹<br />
of rings, Pacific J. Math. 86(1980), 89–92<br />
ØÑÝÚÙÚÖØÕÒÝÓÖÑÓØÔØÔÖÑÒÒݸÅع ÑØÕ×Ñؽ¼´½½µ¸¾¿ß¾<br />
Irel<strong>and</strong>, K. F. <strong>and</strong> Rosen, M. I.: A classical introduction to modern<br />
number theory, Graduate Texts in Mathematics 84, Springer, New<br />
York <strong>and</strong> Berlin 1982<br />
English translation: Iskovskih, V. A.: A counterexample to the<br />
Hasseprinciplefor systemsoftwoquadraticformsinfivevariables,<br />
Mathematical Notes 10(1971), 575–577<br />
Jacobson, N.: Structure of rings, Revised edition, Amer. Math.<br />
Soc. Coll. Publ. 37, Amer. Math. Soc., Providence, R.I. 1964<br />
Jahnel, J.: Line bundles on arithmetic surfaces <strong>and</strong> intersection<br />
theory, manuscripta math. 91(1996), 103-119
BIBLIOGRAPHY 373<br />
[J2]<br />
[J3]<br />
[J4]<br />
[J5]<br />
[dJo]<br />
Jahnel, J.: Heights for line bundles on arithmetic varieties, manuscripta<br />
math. 96(1998), 421-442<br />
Jahnel, J.: A height function on the Picard group of singular<br />
Arakelov varieties, in: Algebraic K-Theory <strong>and</strong> Its Applications,<br />
Proceedings of the Workshop <strong>and</strong> Symposium held at ICTP<br />
Trieste, September 1997, edited by H. Bass, A. O. Kuku <strong>and</strong><br />
C. Pedrini, World Scientific, Singapore 1999, 410-436<br />
Jahnel, J.: The Brauer-Severi variety associated with a central<br />
simple algebra, Linear Algebraic Groups <strong>and</strong> Related Structures 52<br />
(2000), 1-60<br />
Jahnel, J.: On the distribution of small <strong>points</strong> on abelian <strong>and</strong> toric<br />
varieties, Preprint<br />
de Jong, A. J.: Smoothness, semi-stability <strong>and</strong> alterations, Publ.<br />
Math. IHES 83(1996), 51–93<br />
[K/K/M/S] Kempf, G., Knudsen, F. F., Mumford, D., <strong>and</strong> Saint-Donat, B.:<br />
Toroidal embeddings I, Lecture Notes in Mathematics 339,<br />
Springer, Berlin-New York 1973<br />
[Ke]<br />
[K/R]<br />
[K/O74]<br />
[K/O80]<br />
[K/T]<br />
Kersten, I.: Brauergruppen von Körpern, Vieweg, Braunschweig<br />
1990<br />
Kersten, I. <strong>and</strong>Rehmann, U.: Genericsplittingofreductive<strong>groups</strong>,<br />
Tôhoku Math. J. 46(1994), 35–70<br />
Knus, M. A. <strong>and</strong> Ojanguren, M.: Théorie de la descente et algèbres<br />
d’Azumaya, Springer, Lecture Notes Math. 389, Berlin, Heidelberg,<br />
New York 1974<br />
Knus, M. A. <strong>and</strong> Ojanguren, M.: Cohomologie étale et groupe<br />
de Brauer, in: The Brauer group (Les Plans-sur-Bex 1980), Lecture<br />
Notes Math. 844, Springer, Berlin, Heidelberg, New York 1981,<br />
210–228<br />
Kresch, A. <strong>and</strong> Tschinkel, Y.: Effectivity of Brauer-Manin obstructions,<br />
Preprint<br />
[Kr] Kress, R.: Numerical analysis, Graduate Texts in Mathematics 181,<br />
Springer, New York 1998
374 BIBLIOGRAPHY<br />
[Lan86]<br />
[Lan93]<br />
[Laz]<br />
[Le]<br />
[Lic]<br />
[Lin]<br />
[L/O]<br />
[Lo]<br />
[MWe05]<br />
[MWe08]<br />
[Man]<br />
[Mc]<br />
[Mat]<br />
Lang, S.: Hyperbolic <strong>and</strong> Diophantine analysis, Bull. Amer. Math.<br />
Soc. 14(1986), 159–205<br />
Lang, S.: Algebra, Third Edition, Addison-Wesley, Reading, MA<br />
1993<br />
Lazarsfeld, R.: Positivity in algebraic geometry I, Ergebnisse<br />
der Mathematik und ihrer Grenzgebiete (3. Folge) 48, Springer,<br />
Berlin 2004<br />
Lelong, P.: Fonctions plurisousharmoniques et formes différentielles<br />
positives, Gordon & Breach, Paris 1968<br />
Lichtenbaum, S.: Duality theorems for curves over p-adic fields,<br />
Invent. Math. 7(1969), 120–136<br />
Lind, C.-E.: Untersuchungen über die <strong>rational</strong>en Punkte der<br />
ebenen kubischen Kurven vom Geschlecht Eins, Doktorarbeit<br />
(Ph.D. thesis), Uppsala 1940<br />
Lorenz, F. <strong>and</strong> Opolka, H.: Einfache Algebren und projektive<br />
Darstellungen über Zahlkörpern, Math. Z. 162(1978), 175–182<br />
Lovász, L.: How to compute the volume?, Jahresbericht der DMV,<br />
Jubiläumstagung 100 Jahre DMV (Bremen 1990), 138–151<br />
ÅÒÒ¸ºÁºÃÙÕ×ÓÖÑÝÐÖ¸ÓÑØÖ¸Ö¹<br />
Maclagan-Wedderburn, J. H.: A theorem on finite algebras, Trans.<br />
ÑظÆÙ¸ÅÓ×Ú½¾<br />
Amer. Math. Soc. 6(1905), 349–352<br />
Maclagan-Wedderburn, J. H.: On hypercomplex numbers, Proc.<br />
London Math. Soc., 2nd Series, 6(1908), 77–118<br />
English translation: Manin, Yu. I.: Cubic forms, algebra, geometry,<br />
arithmetic, North-Holl<strong>and</strong> Publishing Co. <strong>and</strong> American Elsevier<br />
Publishing Co., Amsterdam-London <strong>and</strong> New York 1974<br />
Marcus, D. A.: Number fields, Springer, New York-Heidelberg<br />
1977<br />
Matsumura, H.: Commutative ring theory, Cambridge Studies in<br />
Advanced Mathematics 8, Cambridge University Press, Cambridge<br />
1986
BIBLIOGRAPHY 375<br />
[McK]<br />
McKinnon, D.: Counting <strong>rational</strong> <strong>points</strong> on K3 surfaces, J. Number<br />
Theory 84(2000), 49–62<br />
[Me-a]<br />
ÅÖÙÖÚ¸ºËºÇ×ØÖÓÒÖÙÔÔÝÖÙ«ÖÔÓиÁÞÚº<br />
O’Meara, O. T.: Introduction to quadratic forms, Springer, Berlin,<br />
ºÆÙËËËÊËÖºÅغ´½µ¸¾ß¸<br />
New York 1963<br />
[Me83] Merkurjev, A. S.: Brauer <strong>groups</strong> of fields, Comm. in Algebra 11<br />
(1983), 2611–2624<br />
[Me85]<br />
Englishtranslation: Merkurjev, A. S.: StructureoftheBrauer group<br />
[Me93]<br />
[Me/S]<br />
ÅÖÙÖÚ¸ºËºÑÒÙØÝØÓÕÑÒÓÓÓÖÞËÚÖ¹ ÖÙ«Ö¸Î×ØÒ˺¹ÈØÖÙÖºÍÒÚºÅغź×ØÖÓ¹ ÒÓѺ½¿¹¾´½¿µ¸½ß¿<br />
of fields, Math. USSR-Izv. 27(1986), 141–157<br />
ÅÖÙÖÚ¸ºËºËÙ×ÐÒ¸ººK¹ÓÓÑÓÐÓÑÒÓÓ¹ ÓÖÞËÚÖ¹ÖÙ«ÖÓÑÓÑÓÖÞÑÒÓÖÑÒÒÓÓÚݹ ÕظÁÞÚººÆÙËËËÊËÖºÅغ´½¾µ¸½¼½½ß½¼¸<br />
English translation: Merkurjev, A. S.: Closed <strong>points</strong> of Brauer-<br />
Severi varieties, Vestnik St.PetersburgUniv.Math. 26-2(1993), 44–46<br />
English translation: Merkurjev, A. S. <strong>and</strong> Suslin, A. A.: K-<br />
cohomology of Severi-Brauer varieties <strong>and</strong> the norm residue homomorphism,<br />
Math. USSR-Izv. 21(1983), 307–340<br />
[deM/F]<br />
[deM/I]<br />
[Mi]<br />
[Mol]<br />
[Mord]<br />
de Meyer, F. <strong>and</strong> Ford, T. J.: On the Brauer group of surfaces <strong>and</strong><br />
subrings ofk[x, y], in: van Oystaeyen, F. <strong>and</strong> Verschoren, A. (Eds.):<br />
Brauer <strong>groups</strong> in ring theory <strong>and</strong> algebraic geometry, Proceedings<br />
of a conference held at Antwerp 1981, Lecture Notes Math. 917,<br />
Springer, Berlin, Heidelberg, New York 1982, 211–221<br />
de Meyer, F. <strong>and</strong> Ingraham, E.: Separable algebras over commutative<br />
rings, Lecture Notes Math. 181, Springer, Berlin, Heidelberg,<br />
New York 1971<br />
Milne, J. S.: Étale Cohomology, Princeton University Press, Princeton<br />
1980<br />
Molien, T.: Über Systeme höherer complexer Zahlen, Math. Ann.<br />
41(1893), 83–156<br />
Mordell, L. J.: On the conjecture for the <strong>rational</strong> <strong>points</strong> on a cubic<br />
surface, J. London Math. Soc. 40(1965), 149–158
376 BIBLIOGRAPHY<br />
[Mori]<br />
[Mu-e]<br />
[Mu-y]<br />
[Nak]<br />
[Nat]<br />
[Ne]<br />
[Noe]<br />
[Or]<br />
[Or/S]<br />
[Or/V]<br />
[Pe95a]<br />
[Pe95b]<br />
Moriwaki, A.: Intersection pairing for arithmetic cycles with degenerate<br />
Green currents, E-print AG/9803054<br />
Murre, J. P.: Algebraic equivalence modulo <strong>rational</strong> equivalence<br />
on a cubic threefold, Compositio Math. 25(1972), 161–206<br />
Murty, M. R.: Applications of symmetric power L-functions, in:<br />
Lectures on automorphic L-functions, Fields Inst. Monogr. 20,<br />
Amer. Math. Soc., Providence 2004, 203–283<br />
Nakayama, T.: Divisionsalgebren über diskret bewerteten perfekten<br />
Körpern, J. für die Reine und Angew. Math. 178(1937), 11–13<br />
Nathanson, M. B.: Elementary methods in number theory, Graduate<br />
Texts in Mathematics 195, Springer, New York 2000<br />
Neukirch, J.: Algebraic number theory, Grundlehren der Mathematischen<br />
Wissenschaften 322, Springer, Berlin 1999<br />
Noether, M.: Rationale Ausführung der Operationen in der Theorie<br />
der algebraischen Funktionen, Math. Ann. 23(1884), 311–358<br />
Orzech, M.: Brauer Groups <strong>and</strong> Class Groups for a Krull Domain,<br />
in: van Oystaeyen, F. <strong>and</strong> Verschoren, A. (Eds.): Brauer <strong>groups</strong> in<br />
ring theory <strong>and</strong> algebraic geometry, Proceedings of a conference<br />
held at Antwerp 1981, Lecture Notes Math. 917, Springer, Berlin,<br />
Heidelberg, New York 1982, 66–90<br />
Orzech, M. <strong>and</strong> Small, C.: The Brauer group of commutative rings,<br />
Lecture Notes in Pure <strong>and</strong> Applied Mathematics 11, Marcel Dekker,<br />
New York 1975<br />
Orzech, M. <strong>and</strong> Verschoren, A.: Some Remarks on Brauer Groups<br />
of Krull Domains, in: van Oystaeyen, F. <strong>and</strong> Verschoren, A. (Eds.):<br />
Brauer <strong>groups</strong> in ring theory <strong>and</strong> algebraic geometry, Proceedings<br />
of a conference held at Antwerp 1981, Lecture Notes Math. 917,<br />
Springer, Berlin, Heidelberg, New York 1982, 91–95<br />
Peyre, E.: Hauteurs et mesures de Tamagawa sur les variétés de<br />
Fano, Duke Math. J. 79(1995), 101–218<br />
Peyre, E.: Products of Severi-Brauer varieties <strong>and</strong> Galois cohomology,<br />
in: Proc. Symp. Pure Math. 58, Part 2: K-theory <strong>and</strong> algebraic<br />
geometry: connections with quadratic forms <strong>and</strong> division algebras
BIBLIOGRAPHY 377<br />
(Santa Barbara, CA 1992), Amer. Math. Soc., Providence, R.I. 1995,<br />
369–401<br />
[Pe02]<br />
[Pe/T]<br />
[Pi]<br />
[Poh/Z]<br />
[Poi]<br />
[Poo/T]<br />
[Qu]<br />
[Re]<br />
[Roq63]<br />
[Roq64]<br />
[Rou]<br />
Peyre, E.: Points de hauteur bornée et géométrie des variétés<br />
(d’après Y. Manin et al.), Séminaire Bourbaki 2000/2001,Astérisque<br />
282(2002), 323–344<br />
Peyre, E. <strong>and</strong> Tschinkel, Y.: Tamagawa numbers of diagonal cubic<br />
surfaces, numerical evidence, Math. Comp. 70(2001), 367–387<br />
Pierce, R. S.: Associative Algebras, Graduate Texts in Mathematics<br />
88, Springer, New York 1982<br />
Pohst, M. <strong>and</strong> Zassenhaus, H.: Algorithmic algebraic number theory,<br />
Encyclopedia of Mathematics <strong>and</strong> its Applications 30, Cambridge<br />
University Press, Cambridge 1989<br />
Poincaré, H.: Sur les propriétés arithmétiques des courbes algébriques,<br />
J. Math. Pures et Appl. 5 e série, 7(1901), 161–234<br />
Poonen, B. <strong>and</strong> Tschinkel, Y. (eds.): Arithmetic of higherdimensional<br />
algebraic varieties, Proceedings of the Workshop on<br />
Rational <strong>and</strong> Integral Points of Higher-Dimensional Varieties held<br />
in Palo Alto, CA, December 11–20, 2002, Birkhäuser, Progress in<br />
Mathematics 226, Boston 2004<br />
Quillen, D.: Higher algebraic K-theory I, in: Algebraic K-theory<br />
I, Higher K-theories, Lecture Notes Math. 341, Springer, Berlin,<br />
Heidelberg, New York 1973, 85–147<br />
Reiner, I.: Maximal orders, Academic Press, London-New York<br />
1975<br />
Roquette, P.: On the Galois cohomology of the projective linear<br />
group <strong>and</strong> its applications to the construction of generic splitting<br />
fields of algebras, Math. Ann. 150(1963), 411–439<br />
Roquette, P.: Isomorphisms of generic splitting fields of simple<br />
algebras, J. für die Reine und Angew. Math. 214/215(1964), 207–226<br />
Roux, B.: Sur le groupe de Brauer d’un corps local à corps résiduel<br />
imparfait, in: Groupe d’Etude d’Analyse Ultramétriuqe, Publ.<br />
Math. de l’Univ. de Paris VII 29(1985/86), 85–97
378 BIBLIOGRAPHY<br />
[Sal80]<br />
[Sal81]<br />
[Sat]<br />
[Scha]<br />
[Sch/St]<br />
[Schr]<br />
[Schw]<br />
Saltman, P. J.: Division algebras over discrete valued fields, Comm.<br />
in Algebra 8(1980), 1749–1774<br />
Saltman, P. J.: The Brauer group is torsion, Proc. Amer. Math. Soc.<br />
81(1981), 385–387<br />
Satake, I.: On the structure of Brauer group of a discretely-valued<br />
complete field, Sci. Papers Coll. Gen. Ed. Univ. Tokyo 1(1951), 1–10<br />
Scharlau, W.: Quadratic <strong>and</strong> Hermitian forms, Springer, Grundlehren<br />
der math. Wissenschaften 270, Berlin, Heidelberg, New<br />
York 1985<br />
Schönhage, A. <strong>and</strong> Strassen, V.: Schnelle Multiplikation großer<br />
Zahlen, Computing 7(1971), 281–292<br />
Schröer, S.: There are enough Azumaya algebras on surfaces, Math.<br />
Ann. 321(2001), 439–454<br />
Schwarz, H. A.: Sur une définition erronée de l’aire d’une surface<br />
courbe, Communication faite à M. Charles Hermite, 1881–<br />
82, in: Gesammelte Mathematische Abh<strong>and</strong>lungen, Zweiter B<strong>and</strong>,<br />
Springer, Berlin 1890, 309–311<br />
[Sed] Sedgewick, R.: Algorithms, Addison-Wesley, Reading 1983<br />
[See97]<br />
[See99]<br />
[Sel]<br />
[Se67]<br />
[SGA3]<br />
Seelinger, G.: A description of the Brauer-Severi scheme of trace<br />
rings of generic matrices, J. of Algebra 184(1996), 852–880<br />
Seelinger, G.: Brauer-Severi schemes of finitely generated algebras,<br />
Israel J. Math. 111(1999), 321–337<br />
Selmer, E. S.: The Diophantine equation ax 3 +by 3 +cz 3 = 0, Acta<br />
Math. 85(1951), 203–362<br />
Serre, J.-P.: Local class field theory, in: Algebraic number theory,<br />
Edited by J. W. S. Cassels <strong>and</strong> A. Fröhlich, Academic Press <strong>and</strong><br />
Thompson Book Co., London <strong>and</strong> Washington 1967<br />
Demazure, A. et Grothendieck, A.: Schemas en groupes, Séminaire<br />
de Géométrie Algébrique du Bois Marie 1962–64 (SGA3), Lecture<br />
Notes Math. 151, 152, 153, Springer, Berlin, Heidelberg, New York<br />
1970
BIBLIOGRAPHY 379<br />
[SGA4]<br />
[SGA4 1 2 ]<br />
[SGA6]<br />
Artin, M., Grothendieck, A. et Verdier, J.-L. (avec la collaboration<br />
de Deligne, P. et Saint-Donat, B.): Théorie des topos et cohomologie<br />
étale des schémas, Séminaire de Géométrie Algébrique du Bois<br />
Marie 1963–1964 (SGA4), Lecture Notes in Math. 305, Springer,<br />
Berlin, Heidelberg, New York 1973<br />
Deligne, P. (avec la collaboration de Boutot, J. F., Grothendieck,<br />
A., Illusie, L. et Verdier, J. L.): Cohomologie Étale, Séminaire<br />
de Géométrie Algébrique du Bois Marie (SGA4 1 ), Lecture Notes<br />
2<br />
Math. 569, Springer, Berlin, Heidelberg, New York 1977<br />
Berthelot, P., Grothendieck, A. et Illusie, L. (Avec la collaboration<br />
de Ferr<strong>and</strong>, D., Jouanolou, J. P., Jussila, O., Kleiman, S., Raynaud,<br />
M. et Serre, J. P.): Théorie des intersections et théorème de<br />
Riemann-Roch, Séminaire de Géométrie Algébrique du Bois Marie<br />
1966-67 (SGA6), Lecture Notes in Math. 225, Springer, Berlin-New<br />
York 1971<br />
[Se62] Serre, J.-P.: Corps locaux, Hermann, Paris 1962<br />
[Se72]<br />
[Se73]<br />
[Sev]<br />
[Si69]<br />
[Si73]<br />
[Sim]<br />
[Sk]<br />
Serre, J.-P.: Propriétés galoisiennes des <strong>points</strong> d’ordre fini des<br />
courbes elliptiques, Invent. Math. 15(1972), 259-331<br />
Serre, J.-P.: Cohomologie Galoisienne, Lecture Notes in Mathematics<br />
5, Quatrième Edition, Springer, Berlin 1973<br />
Severi, F.: Un nuovo campo di richerche nella geometria sopra una<br />
superficie e sopra una varietà algebrica, Mem. della Acc. R. d’Italia<br />
3(1932), 1–52 = Opere matematiche, vol. terzo, Acc. Naz. Lincei,<br />
Roma, 541–586<br />
Siegel, C. L.: Abschätzung von Einheiten, Nachr. Akad. Wiss. Göttingen,<br />
Math.-Phys. Kl. II 1969(1969), 71–86<br />
Siegel, C. L.: Normen algebraischer Zahlen, Nachr. Akad. Wiss.<br />
Göttingen, Math.-Phys. Kl. II 1973(1973), 197–215<br />
Sims, C. C.: Computational methods in the study of permutation<br />
<strong>groups</strong>, in: Computational Problems in Abstract Algebra (Proc.<br />
Conf., Oxford 1967), Pergamon, Oxford 1970, 169–183<br />
Skorobogatov, A.: Torsors <strong>and</strong> <strong>rational</strong> <strong>points</strong>, Cambridge Tracts<br />
in Mathematics 144, Cambridge University Press, Cambridge 2001
380 BIBLIOGRAPHY<br />
[Sm]<br />
Smart, N. P.: The algorithmic resolution of Diophantine equations,<br />
London Mathematical Society Student Texts 41, Cambridge<br />
University Press, Cambridge 1998<br />
[S/A/B/K] Soulé, C., Abramovich, D., Burnol, J.-F., <strong>and</strong> Kramer, J.: Lectures<br />
on Arakelov geometry, Cambridge Studies in Advanced Mathematics<br />
8, Cambridge University Press, Cambridge 1992<br />
[Sp]<br />
[St]<br />
[Su]<br />
[Sw]<br />
[S-D62]<br />
[S-D93]<br />
[S-D04]<br />
[S-D05]<br />
[Sze]<br />
[Szp]<br />
ËÙ×ÐÒ¸ººÃÚØÖÒÓÒÒÝÓÑÓÑÓÖÞÑÐÔÓÐÙÒ¹<br />
Spanier, E. H.: Algebraic topology, McGraw-Hill, New York-Toronto-London<br />
1966<br />
ÒÓÒ¸ÓкºÆÙËËËʾ´½¾µ¸¾¾ß¾¸<br />
Stark, H. M.: Some effective cases of the Brauer-Siegel theorem,<br />
Invent. Math. 23(1974), 135–152<br />
English translation: Suslin, A. A.: The quaternion homomorphism<br />
for the function field on a conic, Soviet Math. Dokl. 26(1982), 72–77<br />
Swiȩcicka, J.: Small deformations of Brauer-Severi schemes, Bull.<br />
Acad. Polon. Sci. Sér. Sci. Math. Astronom. Phys. 26(1978), 591–593<br />
Swinnerton-Dyer, Sir P.: Two special cubic surfaces, Mathematika<br />
9(1962), 54–56<br />
Swinnerton-Dyer, Sir P.: The Brauer group of cubic surfaces, Math.<br />
Proc. Cambridge Philos. Soc. 113(1993), 449–460<br />
Swinnerton-Dyer, Sir P.: Rational <strong>points</strong> on fibered surfaces, in:<br />
Tschinkel, Y. (ed.): Mathematisches Institut, Seminars 2004, Universitätsverlag,<br />
Göttingen 2004, 103–109<br />
Swinnerton-Dyer, Sir P.: Counting <strong>points</strong> on cubic surfaces II,<br />
in: Geometric methods in algebra <strong>and</strong> number theory, Progr.<br />
Math. 235, Birkhäuser, Boston 2005, 303–309<br />
Szeto, G.: Splitting Rings for Azumaya Quaternion Algebras, in:<br />
van Oystaeyen, F. <strong>and</strong> Verschoren, A. (Eds.): Brauer <strong>groups</strong> in<br />
ring theory <strong>and</strong> algebraic geometry, Proceedings of a conference<br />
held at Antwerp 1981, Lecture Notes Math. 917, Springer, Berlin,<br />
Heidelberg, New York 1982, 118–125<br />
Szpiro, L.: Sur les propriétés numériques du dualisant relatif<br />
d’une surface arithmétique, The Grothendieck Festschrift, vol. III,<br />
Birkhäuser, Boston 1990
BIBLIOGRAPHY 381<br />
[S/U/Z]<br />
[Ta67]<br />
[Ta75]<br />
[Ti81]<br />
[Ti87]<br />
[Tr]<br />
[Tu]<br />
[Ul]<br />
[Va]<br />
[Ve]<br />
Szpiro, L., Ullmo, E., Zhang, S.: Equirépartition des petits <strong>points</strong>,<br />
Invent. Math. 127(1997), 337-347<br />
Tate, J.: Global class field theory, in: Algebraic number theory,<br />
Edited by J. W. S. Cassels <strong>and</strong> A. Fröhlich, Academic Press <strong>and</strong><br />
Thompson Book Co., London <strong>and</strong> Washington 1967<br />
Tate, J.: Algorithm for determining the type of a singular fiber in an<br />
elliptic pencil, Modular functions of one variable IV (Proc. Internat.<br />
Summer School, Antwerp 1972), Lecture Notes in Math. 476,<br />
Springer, Berlin 1975, 33–52<br />
Tignol, J.-P.: Sur les decompositions des algèbres à divisions en<br />
produit tensoriel d’algèbres cycliques, in: van Oystaeyen, F. <strong>and</strong><br />
Verschoren, A. (Eds.): Brauer <strong>groups</strong> in ring theory <strong>and</strong> algebraic<br />
geometry, Proceedings of a conference held at Antwerp 1981, Lecture<br />
Notes Math. 917, Springer, Berlin, Heidelberg, New York<br />
ÌÖٸ˺ĺÇÖÓÒÐÒÓ«ÚÚÐÒØÒÓ×ØÑÒÓÓÓ¹ ÖÞÖÙ«Ö¹ËÚÖ¸Í×ÔÅغÆÙ¹´¾¾µ´½½µ¸<br />
1982, 126–145<br />
¾½ß¾½<br />
Tignol, J.-P.: On the corestriction of central simple algebras, Math.<br />
Z. 194(1987), 267–274<br />
English translation: Tregub, S. L.: On the bi<strong>rational</strong> equivalence<br />
of Brauer-Severi varieties, Russian Math. Surveys 46-6(1991), 229<br />
Turner, S.: The zeta function of a bi<strong>rational</strong> Severi-Brauer scheme,<br />
Bol. Soc. Brasil. Mat. 10(1979), 25–50<br />
Ullmo, E.: Positivité et discrétion des <strong>points</strong> algébriques des<br />
courbes, Annals of Math. 147(1998), 167-179<br />
Vaughan, R. C.:The Hardy-Littlewood method, Cambridge Tracts<br />
in Mathematics 80, Cambridge University Press, Cambridge 1981<br />
Verschoren, A.: A check list on Brauer <strong>groups</strong>, in: van Oystaeyen,<br />
F. <strong>and</strong> Verschoren, A. (Eds.): Brauer <strong>groups</strong> in ring theory<br />
<strong>and</strong> algebraic geometry, Proceedings of a conference held at<br />
Antwerp 1981, Lecture Notes Math. 917, Springer, Berlin, Heidelberg,<br />
New York 1982, 279–300<br />
[Wan] Wang, S.: On the commutator group of a simple algebra, Amer. J.<br />
Math. 72(1950), 323–334
382 BIBLIOGRAPHY<br />
[War]<br />
[We48]<br />
Warner, F. W.: Foundations of differentiable manifolds <strong>and</strong> Lie<br />
<strong>groups</strong>, Scott, Foresman <strong>and</strong> Co., Glenview-London 1971<br />
Weil, A.: Sur les courbes algébriques et les variétés qui s’en déduisent,<br />
Actualités Sci. Ind. 1041, Hermann et Cie., Paris 1948<br />
[We56] Weil, A.: The field of definition of a variety, Amer. J. of Math. 78<br />
(1956), 509–524<br />
[Wi35]<br />
ÒÕÚ׸κÁºÈÖÒÔÀ××ÐÖÙÔÔÖÙ«ÖÔÓÐ<br />
Witt, E.: Über ein Gegenbeispiel zum Normensatz, Math. Z. 39<br />
(1935), 462–467<br />
[Wi37]<br />
¾¼´½µ¸¿¼ß¿ ÙÒÒÔÖÓÞÚÒÑÒÓÓÓÖÞËÚÖ¹ÖÙ«Ö ÔÖÓØÚÒÝÚÖ¸ÌÖÙÝÅغÁÒ×غÑÒËØÐÓÚ<br />
Witt, E.: Schiefkörper über diskret bewerteten Körpern, J. für die<br />
Reine und Angew. Math. 176(1937), 153–156<br />
[Ya]<br />
[Yu] Yuan, S.: On the Brauer <strong>groups</strong> of local fields, Annals of Math. 82<br />
(1965), 434–444<br />
[Za]<br />
Zak, F. L.: Tangents <strong>and</strong> secants of algebraic varieties, AMS Translations<br />
of Mathematical Monographs 127, Amer. Math. Soc., Providence<br />
1993<br />
[Zh95a] Zhang, S.: Small <strong>points</strong> <strong>and</strong> adelic metrics, J. Alg. Geom. 4(1995),<br />
281-300<br />
[Zh95b]<br />
[Zh98]<br />
Zhang, S.: Positive line bundles on arithmetic varieties, J. Amer.<br />
Math. Soc. 8(1995), 187-221<br />
Zhang, S.: Equidistribution of small <strong>points</strong> on abelian varieties,<br />
Annals of Math. 147(1998), 159-165
INDEX<br />
A<br />
Abbes, A. . . . . . . . . . . . . . . . . . . . . . . . . . . . 170<br />
abelian surface . . . . . . . . . . . . . . . . . . 218, 255<br />
abelian variety . . . . . . . . . . . . . . . . . . 218, 329<br />
accumulating subvariety xxi, 212, 254f, 258<br />
accumulating surface. . . . . . . . . . . . .255, 258<br />
adelic intersection product . . . . . . . . xx, 183<br />
adelic intersection theory. . . .xx, 183 – 188<br />
adelic metric.175, 182, 187, 193f, 199, 207,<br />
222f<br />
adelic Picard group . xx, 173, 177, 187, 193<br />
adelic point . . xv – xix, xxii, 99ff, 105, 118,<br />
128, 234, 239, 286, 302 f, 326<br />
adelic topology . . . . . . . . . . . . . . . . . . . . . . 101<br />
adelicly metrized invertible sheaf . . . . 175ff<br />
adjunction formula . . . . . . . . . . . . . . . . . . 232<br />
affine cone . . . . . . . . . . . . . . . . . . . . . . 232, 283<br />
algebraic surface . . . . . . . . . . . . . . . . . . . 71, 85<br />
algebraic surfaces<br />
classification of . . . . . . . . . . . . . . . . . . . . 329<br />
Algorithm<br />
Elkies’ method . . . . . . . . . . xxiii, 294, 349<br />
FFT point counting. . . . . . .248, 251, 260<br />
for numerical integration.251, 291f, 298<br />
obvious . . . . . . . . . . . . . . . . . . . . . . . . . . . 247<br />
of Fincke-Pohst. . . . . . . . . . . . . . . . . . .350 f<br />
of Tate. . . . . . . . . . . . . . . . . . . . . . . . . . . .142<br />
to compute approximate value for Peyre’s<br />
constant . . . . . . . . . . . . . . . 251, 290, 293<br />
to detect conics on quartic threefold 257<br />
to detect lines on cubic threefold . . 255,<br />
260<br />
to generate triangular mesh. . . . . . . . .291<br />
to search for solutions of Diophantine<br />
equation.xxi, xxiii, 259, 284, 294, 296,<br />
332 – 348<br />
naive . . . . . . . . . . . . . . . . . . . . . . . . . . . 332<br />
to solve system of equations over finite<br />
field. . . . . . . . . . . . . . . . . . . . . . . .255, 257<br />
to test for conic through two <strong>points</strong>.256<br />
to test lines for ir<strong>rational</strong>ity . . . . . . . 254f<br />
to test whetherconic is contained in quartic<br />
threefold. . . . . . . . . . . . . . . . . . . . .257<br />
to verify Galois group acting on 27 lines<br />
xxii, 288f, 293, 296f, 299<br />
2-adic Hensel lift . . . . . . . . . . . . . . . . . . 293<br />
algorithms, for computation of volume 227<br />
alteration. . . . . . . . . . . . . . . . . . . . . . . . . . . .164<br />
AMD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347<br />
Amir-Moez, A. R.. . . . . . . . . . . . . . . . . . . . . .3<br />
Amitsur, S. A. . . . . . . . . . . . . . . . . . . . . . . . . 37<br />
apex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350<br />
Arakelov, S. Yu. . . . . . . . . . . . . . . . . . 156, 160<br />
Arakelov degree . . . . . . . . . . . . . xx, 157, 162<br />
Arakelov geometry . . . . . . . . . . . . . . 190, 192<br />
archimedean valuation . . . . . . . . . . . . . . . 152<br />
ARIBAS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348<br />
arithmetic Chern class. . . . . . . . . . . . . . . .162<br />
arithmetic Chow group. . . . . . . . .161 – 164<br />
arithmetic cycle . . . . . . . . xx, 160f, 165, 173
384 INDEX<br />
linearly equivalent to zero. . . . . . . . . .161<br />
arithmetic degree . . . . . . . . . . 161, 168, 185f<br />
arithmetic Hilbert-Samuel formula . . . . . xx<br />
arithmetic intersection product . . 163, 165,<br />
168, 185, 208<br />
arithmetic intersection theory . . . . . xx, 155<br />
arithmetic Picard group . . . . . . . . . . . . . . 156<br />
arithmetic variety . . . . . . . . . xix, 156, 158 ff,<br />
162 – 167, 169f, 172, 185<br />
singular . . . . . . . . . . . . . . . . . . . . . 165 – 168<br />
Artin, M. . . . . . . . . . . . . . . . . . . . . . . . 4, 34, 38<br />
Artin conductor. . . . . . . . . . . . . . . .309f, 314<br />
Artin L-function. . . . . .228, 303, 314ff, 326<br />
assembly. . . . . . . . . . . . . . . . . . . . . . . .344, 350<br />
Ausl<strong>and</strong>er-Goldman, Theorem of xvii, 71,<br />
84<br />
automorphisms of P n−1 . . . . . . . . . . . 29, 155<br />
Azumaya algebra . . 61 – 70, 78, 80, 82f, 85,<br />
104, 109<br />
B<br />
bad<br />
prime . . . . . . . . . . . . . . . . . . . . . . . 179f, 251<br />
reduction . . . . . . . . . . . . . . . . . . . . .136, 297<br />
barycenter . . . . . . . . . . . . . . . . . . . . . . . . . . 292<br />
Batyrev, V. V. . . . . . . . . . . . . . . . . . . . . . . . .217<br />
Batyrev <strong>and</strong> Manin, Conjecture ofxxi, 214,<br />
216 – 219, 255<br />
Bernstein, D. . . . . . . . . . . . . . . 259, 349, 351f<br />
Bezout’s theorem . . . . . . . . . . . . . . . . . . . . 253<br />
bielliptic surface. . . . . . . . . . . . . . . . .255, 329<br />
Bierce, A. . . . . . . . . . . . . . . . . . . . . . . . . . . . 329<br />
Bilu, Yu.. . . . . . . . . . . . . . . . . . . .xxi, 190, 192<br />
equidistribution theorem ofxxi, 190, 192<br />
biquadratic reciprocity low . . . . . . . . . . . . xv<br />
Bismut, J.-M. . . . . . . . . . . . . . . . . . . . . . . . . 170<br />
Bogomolov, F. . . . . . . . . . . . . . . . . . . . . . . . 219<br />
Conjecture of . . . . . . . . . . . . . . . . . . . . . 219<br />
Bost, J.-B. . . . . . . . . . . . . . . . . . . . . . . . . xx, 167<br />
Bouche, T.. . . . . . . . . . . . . . . . . . . . . . . . . . .170<br />
Bourbaki, N.. . . . . . . . . . . . . . . . . . . . .3 f, 176<br />
Brauer, R. . . . . . . . . . . . . . . . . . . . . . . . . . . 4, 22<br />
Brauer group . . . . . . .xv ff, 22, 66, 71 f, 74 ff,<br />
84 – 97, 286<br />
cohomological. . . . .xvii, 70 – 76, 84 – 97<br />
of cubic surface. . . . . . . . . . . . . . . . . . . . .96<br />
Brauer-Manin obstruction . . . . . . . . . . . xv ff,<br />
xix, xxii, 97, 103, 105 f, 110, 113, 122,<br />
128f, 144f, 212, 237, 239, 244, 246,<br />
281, 286f, 293, 301ff<br />
is the only obstruction . . . . . . . . . . . . . 106<br />
to <strong>rational</strong> <strong>points</strong> . . . . . . . . . . . . . . . . . . 106<br />
to the Hasse principle . . . xviii, 106, 113,<br />
122ff, 130, 212, 239, 286<br />
to weak approximation. .xxii, 106, 122f,<br />
127, 130, 283, 286<br />
Brauer-Severi variety. . . . .xvi f, 3 f, 22, 28 f,<br />
32 – 39, 46, 48, 51 ff, 102<br />
that splits . . . . . . . . . . . . . . . . . . . . xvi, 28ff<br />
Bremner, A.. . . . . . . . . . . . . . . . . . . . . .xv, 103<br />
Brown-Gersten-Quillen spectral sequence<br />
164<br />
Bucket sort. . . . . . . . . . . . . . . . . . . . . . . . . .345<br />
buffer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346f<br />
Burgos Gil, J. I. . . . . . . . . . . . . . . . . . 156, 165<br />
C<br />
C . . . . . . . . . . . . . . . . . . . . . . 296, 344, 348, 350<br />
C++ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350<br />
Cantor, M. . . . . . . . . . . . . . . . . . . . . . . . . . . 353<br />
Cartan, E. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3<br />
Cassels, J. W. S. . . . . . . . . . .xv, 103, 265, 283<br />
Čech cocycle . . . . . . . . . . . . . . . . . . . . . . . . . 69<br />
Čech cohomology. . . . . . . . . . . . . . . . . . .68ff<br />
center. . . . . . . . . . .3, 7, 22, 28, 45, 48, 52, 62<br />
central simple algebra . . . . . . . . . . . . . . . xvi f,<br />
3 ff, 10, 22 f, 35 – 39, 45 – 54, 58, 61ff,<br />
74, 76, 78, 82 f<br />
index of. . . . . . . . . . . . . . . . . . . . . . . . . . . .22<br />
that splits . . . . . . . . . . . . . . xvi, 23f, 45, 81<br />
Chambert-Loir, A. . . . . . . . . . . . . . . . xxi, 192<br />
equidistribution theorem of . . . . xxi, 192<br />
Châtelet, F. . . . . . . . . . . . . . . . . . . . . 4f, 34, 38<br />
Chebotarev density theorem .290, 309, 315<br />
Chebyshev’s inequalities. . . . . . . . . . . . . .314<br />
Chinese remainder theorem. . . .341f, 345f<br />
circle method . . . . . xiii, xvi, 238ff, 244, 282<br />
class number formula . . . . . . . . . . . . . . . . 317<br />
classification of algebraic surfaces . . . . . 329<br />
clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
INDEX 385<br />
cocycle . . . . . . . . . . . 6f, 24f, 30 – 34, 37 f, 47<br />
cohomologous to another . . . . . 6, 24, 30<br />
trivial . . . . . . . . . . . . . . . . . . . . . . . . . . . 6, 33<br />
cocycle relations . . . . . . . . . . . . . 12, 108, 180<br />
local in étale topology. . . . . . . . . . . . . . .65<br />
cohomological Brauer group . xvii, 70 – 76,<br />
84 – 97<br />
Colliot-Thélène, J.-L..xv f, xix, 4, 103, 106,<br />
129f, 239, 286f, 302, 312<br />
Conjecture of . . . . . . . . . . . . . . . . . . . . . 106<br />
Colliot-Thélène <strong>and</strong> Sansuc, Conjecture of<br />
287<br />
Colliot-Thélène, Kanevsky, <strong>and</strong> Sansuc,<br />
Theorem of. . . . . . . . . . . . . . . . . . . .129 f<br />
collision. . . . . . . . . . . . . . . . . . . . . . . . . . . . .335<br />
computer graphics . . . . . . . . . . . . . . . . . . . 291<br />
conductor-discriminant formula . . . . . . 309<br />
congruencexiii, xix, 38, 144 f, 238, 269, 287,<br />
314, 330 – 334, 336 – 342, 345f, 348<br />
conic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256ff<br />
Conjecture<br />
of Batyrev <strong>and</strong> Manin . . . . . . . . . xxi, 214,<br />
216 – 219, 255<br />
of Bogomolov . . . . . . . . . . . . . . . . . . . . . 219<br />
of Colliot-Thélène . . . . . . . . . . . . . . . . . 106<br />
of Colliot-Thélène <strong>and</strong> Sansuc . . . . . . 287<br />
of Lang . . . . . . . . . xii, xxi, 214ff, 255, 329<br />
geometric. . . . . . . . . . . . . . . . . . . . . . .214<br />
strong . . . . . . . . . . . . . . . . . . . . . . . . . . 214<br />
weak. . . . . . . . . . . . . . . . . . . . . . .214, 217<br />
of Manin . . . . . . xiv, xxi, 214, 223f, 239 f,<br />
243f, 246, 256, 261, 263, 265, 282, 284,<br />
293, 296, 301<br />
of Manin-Peyre . . . . . . . . . . . 237, 239, 261<br />
of Mordell . . . . . . . . . . . . . . . . . . . . . . . . 215<br />
control function. . . . . . . . . . . . . . . . .341, 345<br />
convolution. . . . . . . . . . . . . . . . . . . .205, 247 f<br />
Corn, P. K. . . . . . . . . . . . . . . . . . . . . . . xvii, 96<br />
cubic reciprocity low . . . . . . . . . . . . . xv, 145<br />
cubic surface. . . . . . . . . . . . . . .xv – xix, xxi f,<br />
94 – 97, 103, 106, 113f, 122, 125 – 130,<br />
143, 224 – 228, 234, 239 f, 252f, 256,<br />
281 – 328, 353<br />
diagonal . xxii, 103, 128ff, 143, 301 – 328<br />
general . . . . . . . . xxii, 106, 240, 281 – 300<br />
minimal . . . . . . . . . . . . . . . . . . . . . . . . . . 312<br />
cubic threefold . . . . xxi, 215, 240, 243 – 263<br />
diagonal . . . . . . . . . . . . . . . . . xxi f, 213, 224<br />
curvature . . . . . . . . . . . . . . . . . . . . . . 199, 201f<br />
current . . . . . . . . . . . . . . . . . . . . . . . 198, 201<br />
form . . . . . . . . . . . .167, 170, 177, 179, 183<br />
cusp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252<br />
D<br />
de Jong, A. J. . . . . . . . . . . . . . . . . . . . . . . . . 164<br />
decomposition type . . . . . . . . . . . . . . . . . . 289<br />
decoupled equation. . . . . . . . .xxiii, 259, 335<br />
Dedekind zeta function . . . . . . . . . . 250, 317<br />
Deligne, P. . . . . . . . . . . . . . 165, 235, 247, 269<br />
Deligne-Beilinson cohomology. . . . . . . .165<br />
∂∂-lemma. . . . . . . . . . . . . . . . . . . . . . . . . . .162<br />
Derenthal, U.. . . . . . . . . . . . . . . . . . . . . . . .228<br />
descent. . .xvi, 8, 10 – 15, 19, 22, 25, 31, 33,<br />
37 f, 47, 49, 51, 58 ff, 64f, 181<br />
Dickson, L. E. . . . . . . . . . . . . . . . . . . . . . . . 288<br />
’s list. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .288<br />
Dieudonné, J.. . . . . . . . . . . . . . . . . . . . . . . . .17<br />
difference scheme . . . . . . . . . . . . . . . . . . . . 351<br />
Diophantine equation . . . . . . . . . . . xi – xiv,<br />
xvi, xxi ff, 149, 259, 266, 284, 294, 296,<br />
329, 332 – 348<br />
decoupled . . . . . . . . . . . . . . . xxiii, 259, 335<br />
Diophantus. . . . . . . . . . . . . . . . . . . . . . .xi, 353<br />
Dirichlet character . . . . . . . . . . . . . . . . . . . 316<br />
Dirichlet’s prime number theorem . . . . 270<br />
distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176<br />
division algebra . . . . . . . . . . . . . . . . . . 78 – 81<br />
Duke, W.. . . . . . . . . . . . . . . . . . . . . . . . . . . .315<br />
dynamical system . . . . . . . . . . . . . . . . . . . . 195<br />
E<br />
Eckardt point. . . . . . . . . . . . . . . . . . . . . . . .253<br />
Edidin, D. . . . . . . . . . . . . . . . . . . . . . . . . . . . .85<br />
effective cone . . . . . . . . . . . . . . . . . . 224 – 227<br />
Eisenstein polynomial. . . . . . . . . . . . . . . .115<br />
Ekedahl, T. . . . . . . . . . . . . . . . . . . . . . . . . . . 297<br />
elementary symmetric functions. .153, 311<br />
Elkies, N. . . . . . . . . . . . . . . . . . . . . . . . . . . . 349<br />
’ method . . . . . . . . . . . . . . . xxiii, 294, 349f
386 INDEX<br />
elliptic cone . . . . . . . . . . . . . . . . . . . . . . . . . 252<br />
elliptic curve . . . . . . . . . . . . . . . . . . . . . . . . 137<br />
elliptic surface . . . . . . . . . . . . . . . . . . . . . . . 329<br />
Elsenhans, A.-S. . . .243, 265, 281, 301, 329,<br />
349<br />
Emerson, R. W.. . . . . . . . . . . . . . . . . . . . . .301<br />
Enriques surface . . . . . . . . . . . . . . . . . . . . . . 89<br />
equidistribution . . . xvi, xx, 142, 190f, 208,<br />
246, 266, 284, 301<br />
theorem of Bilu . . . . . . . . . . . xxi, 190, 192<br />
theorem of Chambert-Loir . . . . . xxi, 192<br />
theorem of Szpiro, Ullmo, <strong>and</strong> Zhangxx,<br />
190<br />
étale cohomology . . . . . . . . . . . . 75, 222, 235<br />
étale morphism . . . . . . . . . . . . . . . . . . . . .229f<br />
étale neighbourhood. . . . . . . . . . . .69, 87, 90<br />
étale topology . . . . . . . . . . . . . . 65, 68, 71, 73<br />
Euler product. . . . . . . . .249 ff, 290, 296, 298<br />
Euler sequence. . . . . . . . . . . . . . . . . . . . . . .233<br />
even . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285<br />
exp<strong>and</strong>ing map . . . . . . . . . . . . . . . . . . . . . . 196<br />
exponential sequence . . . . . . . . . . . . . 88, 222<br />
F<br />
factor α . . . . xiv, xxi, 224 – 228, 234, 238f,<br />
245f, 283, 301, 306, 326<br />
factor β . . .xvi, 228, 244, 283, 301, 305, 326<br />
Faltings, G. . . . . . . . . . . . . . . . . .149, 156, 215<br />
Fano variety . . . . . . . . . . . . . . . . . xii, xiv, xxi,<br />
213f, 217, 220f, 223, 236 f, 239, 243f,<br />
246, 265, 282, 285, 301<br />
Faraday, M.. . . . . . . . . . . . . . . . . . . . . . . . . . .99<br />
Fermat cubic . . . . . . . . . . . . . . . . . . . . . . . . 213<br />
FFT convolution . . . . . . . . . . . . . . . . . . . . 248<br />
FFT point counting . . . . . . . . . . . . . 251, 260<br />
Fincke-Pohst, algorithm of . . . . . . . . . . 350f<br />
first arithmetic Chern class . . . . . . . . . . . 162<br />
first cohomology set . . . . . . . . . . . . . . . . . . . 6<br />
flagstone . . . . . . . . . . . . . . . . . . . . . . . . . . . 350f<br />
Forster, O. . . . . . . . . . . . . . . . . . . . . . . . . . . 348<br />
fpqc-topology . . . . . . . . . . . . . . . . . . . . . . . . 65<br />
Franke, J. . . . . . . . . . . . . . . . . . . . . . . . xiv, 223<br />
Frobenius eigenvalues. . . . . . . . . . . .236, 311<br />
Fubini-Study metric. . . . . . . . . . . .156ff, 193<br />
functor of <strong>points</strong> . . . . . . . . . . . 5, 41, 53, 55 ff<br />
functor represented by a scheme . . . . . . . 41<br />
fundamental finiteness . xx, xxii, 149 f, 160,<br />
167, 194, 267, 270, 279, 304, 316, 328<br />
G<br />
G-group . . . . . . . . . . . . . . . . . . . . . . . . 5 – 8, 34<br />
G-morphism . . . . . . . . . . . . . . . . . . . . . . . . . . 5<br />
G-set . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 – 8, 94<br />
Gabber, O. . . . . . . . . . . . . . . . . . . . . . . . . . . . 85<br />
GAP xvii, 95f, 126, 225f, 285, 289, 305, 353,<br />
356<br />
Gauß-Legendre formula . . . . . . . . . . . . . 251f<br />
Gauß sum . . . . . . . . . . . . . . . . . . . . . . . . . . . 307<br />
Generalized Riemann Hypothesis284, 312,<br />
315f<br />
genus one curve . 119, 129, 135, 138ff, 258,<br />
329<br />
Gilbert, W. S.. . . . . . . . . . . . . . . . . . . . . . . .243<br />
Gillet, H.. . . . .xx, 156, 160, 163ff, 167, 170<br />
global class field theory . . . . . . . . . xv, 86, 91<br />
global evaluation map. .see also Manin map<br />
gluing data . . . . . . . . . . . . . . 65, 68f, 108, 180<br />
GMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350<br />
Godeaux surface . . . . . . . . . . . . . . . . . . . . . 215<br />
good<br />
prime . . . . . . . . . . . . . . . . . . . . . . . . 179, 299<br />
reduction . 269f, 274, 297, 299, 307, 310,<br />
321, 331<br />
Gröbner base . . . . . . 127, 288, 293, 297, 299<br />
Graßmann<br />
functor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54<br />
scheme. . . . . . . . . . . . . . . . . . . . . . . . . . . . .54<br />
Green’s<br />
current . . . . . . . . . . . . . . . . . 160f, 163, 165<br />
form of log type . . . . . . . . . . . . . . 163, 165<br />
Grothendieck, A. . . xvii, 4f, 17, 22, 39, 41,<br />
53, 71, 85<br />
Guy, M. J. T. . . . . . . . . . . . . . . . . . . . . . xv, 103<br />
H<br />
Hankel, H. . . . . . . . . . . . . . . . . . . . . . . . . . . . xi<br />
hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341<br />
Hardy, G. H. . . . . . . . . . . . . . . . . . . . . . . . . xiii<br />
hash function . . . . . . . . . . . . . . .334, 341, 345<br />
hash table . . . . . . . . . . . . . . 259, 335, 345, 347<br />
hashing. . . . . . . . . . . . . . . . . . . .xxiii, 334, 345
INDEX 387<br />
uniform . . . . . . . . . . . . . . . . . . . . . . . . . . 333<br />
Hasse, H. . . . . . . . . . . . . . . . . . . . . . xiii, 4, 265<br />
’s bound . . . . . . . . . . . . . . . . . . . . . . 119, 139<br />
Hasse principle. . .xiii – xvi, xix, 102f, 106,<br />
113, 265f, 281, 284f, 287<br />
Brauer-Manin obstruction to . xviii, 106,<br />
113, 122ff, 130, 212, 239, 286<br />
counterexample to . . . . . xviii, 103, 122ff<br />
obstruction to . . . . . . . . . . . . . . . . . . . . .296<br />
Hassett, B.. . . . . . . . . . . . . . . . . . . . . . . . . . . .85<br />
Hawking, S. . . . . . . . . . . . . . . . . . . . . . . . . . 149<br />
Heath-Brown, D. R. . . . xvi, 145, 239f, 312,<br />
349, 352<br />
’s congruences . . . . . . . . . . . . . . . . . . . . . 145<br />
height<br />
absolute 152, 154, 168, 172, 188, 193, 222<br />
adelic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188<br />
anticanonical . . xiii, 217, 222, 237, 265 f,<br />
282, 284, 301<br />
Bost-Gillet-Soulé . . . . . . . . . . . . . . xx, 167 f<br />
canonical. .190, 192, 217, 222, 237, 265 f,<br />
282, 284, 301<br />
canonical . . . . . . . . . . . . . . . . . . . . . . . . . xiii<br />
defined by adelic metric . . . . . . . . . . . . 222<br />
defined by an invertible sheaf. . .xix, 154<br />
l 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .158<br />
logarithmic . . . . . . . . . . . . . . . . . . . 154, 158<br />
Néron-Tate . . . . . . . . . . . . . . . . . . . . . . 190 f<br />
naive . . . . xiii, xix f, 149 – 152, 158, 173,<br />
190ff, 244ff, 266, 279, 282, 301, 328<br />
normalized. . . . . . . . . . . . . . . . . . . . . . . .172<br />
of smallest point . . . . . . . . xxii, 281, 295 f<br />
with respect to hermitian line bundlexix,<br />
158<br />
Hensel, K. . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii<br />
’s lemma . . . .xiii, 66ff, 75, 104, 115, 118,<br />
135, 141, 230, 269, 290, 313, 336<br />
hermitian line bundle .xx, 156 – 159, 162 f,<br />
167f, 183, 223<br />
continuous. . . . . . . . . . . . . . . . . . . . . . . .157<br />
smooth . . . . . . . . . . . . . . . . . . . . . . . . . . . 157<br />
hermitian metric . . xx, 156, 159f, 174, 200,<br />
202, 245<br />
almost semiample . . 179, 188, 194, 199ff,<br />
204<br />
bounded . . . . . . . . . . . . . . . . . . . . . . . . . . 174<br />
continuous . . . . 156, 173f, 200f, 222, 246<br />
positive . . . . . . . . . . . . . . . . . . . . . . . . . . . 179<br />
smooth . . . . . . . . . . . . . . . . . 162, 173f, 199<br />
heuristic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211<br />
Hirzebruch, F. . . . . . . . . . . . . . . . . . . . . . . . 247<br />
Hochschild-Serre spectral sequence xvi, 90,<br />
111<br />
hypersurface measure. . . . . . . . . . .231, 322f<br />
I<br />
index<br />
of central simple algebra . . . . . . . . . . . . 22<br />
induced<br />
character. . . . . . . . . . . . . . . . . . . . . . . . .307f<br />
representation. . . . . . . . . . . . . . . . . . . .307f<br />
inertia group . . . . . . . . . . . . . . . . . . . . . . . . 283<br />
inflation . . . . . . . . . . . . . . . . 7 f, 112, 127, 134<br />
integrable metric. . . . . . . . . . . . . . . . . . . . .177<br />
integrable topology. . . . . . . . . . . . .177f, 183<br />
integrably metrized invertible sheaf. . . .xx,<br />
177, 188<br />
intersection of two sets . . . . . . . . . . . . . . .332<br />
invariant map. . . . . . . . . . . . . . . . . . . . . . . . .86<br />
ir<strong>rational</strong>ity test for lines . . . . . . . . . . . . 254f<br />
isomorphism test<br />
J<br />
for cubic surfaces . . . . . . . . . . . . . . . . . . 296<br />
Jacobi sum . . . . . . . . . . . . . . . . . 116, 248, 307<br />
Julia set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196<br />
K<br />
K3 surface . . . . xxii f, 89, 92, 94, 218ff, 255,<br />
329f<br />
Kanevsky, D.. . . . . .xix, 103, 129f, 239, 286<br />
Kelvin, W. Thomson 1st Baron . . . . . . . 211<br />
Kersten, I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5<br />
Khayyám, O. . . . . . . . . . . . . . . . . . . . . . . . . . . 3<br />
Kleiman, S.. . . . . . . . . . . . . . . . . . . . . . . . . .186<br />
Kodaira classification . . . . . . . . . . . . 213, 255<br />
Kodaira dimension. . . . . . . . . . . . . . . . . . .255<br />
Kramer, J. . . . . . . . . . . . . . . . . . . . . . . 156, 165<br />
Kresch, A.. . . . . . . . . . . . . . . . . . . . . . . . . . . .85
388 INDEX<br />
Kühn, U. . . . . . . . . . . . . . . . . . . . . . . . 156, 165<br />
Kummer pairing . . . . . . . . . . . . . . . . . . . . . 308<br />
Kummer sequence . . . . . . . . 72, 87f, 91, 222<br />
Kummer surface . . . . . . . . . . . . . . . . . . . . . 218<br />
L<br />
L1 cache . . . . . . . . . . . . . . . . . . . . . .343f, 347f<br />
L2 cache . . . . . . . . . . . . . . . . . . . . . . . 343f, 348<br />
l 3 -unit sphere . . . . . . . . . . . . . . . . . . . . . . . . 323<br />
Lang, S. . . . . . . . xii, xxi, 214 – 217, 255, 329<br />
’s conjecture. . . . .xii, xxi, 214ff, 255, 329<br />
geometric. . . . . . . . . . . . . . . . . . . . . . .214<br />
strong . . . . . . . . . . . . . . . . . . . . . . . . . . 214<br />
weak. . . . . . . . . . . . . . . . . . . . . . .214, 217<br />
Langian exceptional set. . . . . . . . . . . . . .214f<br />
Larsen, K. . . . . . . . . . . . . . . . . . . . . . . . . . . . 149<br />
lattice basis reduction. . . . . . . . . . .124, 350f<br />
Lebeau, G. . . . . . . . . . . . . . . . . . . . . . . . . . . 170<br />
Lebesgue measure . . . . . . . . . . . . . . . 169, 224<br />
Lefschetz hyperplane theorem . . . . . . . . . 89<br />
Lefschetz theorem on (1, 1)-classes. . . . . .88<br />
Lefschetz trace formula. . . . . .235, 307, 310<br />
Legendre, A.-M.. . . . . . . . . . . . .xiii, 265, 283<br />
’s Theorem . . . . . . . . . . . . . . . xiii, 265, 283<br />
Lemma<br />
of Grothendieck . . . . . . . . . . . . . . . . . . . .39<br />
of Grothendieck <strong>and</strong> Dieudonné. . . . .17<br />
of Hensel . . . xiii, 66 ff, 75, 104, 115, 118,<br />
135, 141, 230, 269, 290, 313, 336<br />
of Wedderburn <strong>and</strong> Brauer . . . . . . . . . . 22<br />
of Yoneda. . . . . . . . . . . . . . . . . . . . . . . . .107<br />
of Zorn. . . . . . . . . . . . . . . . . . . . . . . .10, 107<br />
Leray measure . . 231f, 238, 245f, 304, 322f<br />
Lichtenbaum, S. . . . . . . . . . . . . . . . . . . . . . . 92<br />
Theorem of . . . . . . . . . . . . . . . . . . . . . . . . 92<br />
Lichtenbaum duality. . . . . . . . . . . . . .92, 144<br />
Lind, C.-E. . . . . . . . . . . . . . . . . . . . . . xiv f, 103<br />
line . . . xii, xviii, 115, 117 ff, 121, 125f, 128,<br />
131, 136, 212f, 215f, 219, 225ff, 244,<br />
252 – 258, 260, 285, 287f, 299<br />
non-obvious. . . . . . . . . .xxi, 216, 255, 260<br />
obvious . . . . . . . . . . . . . xxi, 215 f, 253, 255<br />
sporadic. . . . . . . . . . . . . .xxi, 216, 255, 260<br />
–s, 27 on cubic surface . . see also 27 lines<br />
on cubic surface<br />
linear probing . . . . . . . . . . . . . . . . . . . . . . . 335<br />
linear subspace. . . . . . . . . . . . . . .33 f, 38, 244<br />
Linnik’s Theorem. . . . . . . . . . . . . . . . . . . .312<br />
Littlewood, J. E. . . . . . . . . . . . . . . . . . . . . . xiii<br />
LLL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350f<br />
local evaluation map . . . . . . 103f, 129f, 135<br />
local measure . . . . . . . . . . . . . . . . . . . . . . . . 230<br />
log-factor. . . . . . . . . . . . . . . . . . . . . . . .xiv, 223<br />
Lovasz, L. . . . . . . . . . . . . . . . . . . . . . . . . . . . 227<br />
M<br />
Maclagan-Wedderburn, J. H.. . . . . . . . .3, 22<br />
Magma .. . . . . . . . . . . . . . . . . . . . . . . . . 294, 297<br />
Manin, Yu. I. . . . xiv f, xix, 93, 96, 103, 113,<br />
217, 223, 281<br />
’s conjecture . . .xiv, xxi, 214, 223f, 239f,<br />
243f, 246, 256, 261, 263, 265, 282, 284,<br />
293, 296, 301<br />
’s formula . . . . . . . . . . . . . . . . . . . . . . . . . . 93<br />
Manin map . . . . . . . . . . . xvii, 103f, 110, 134<br />
Manin-Peyre, Conjecture of. .237, 239, 261<br />
maple.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .118<br />
Maxwell, J. C. . . . . . . . . . . . . . . . . . . . . . . . . 99<br />
McKinnon, D.. . . . . . . . . . . . . . . . . . . . . . .218<br />
memory architecture. . . . . . . . . . . . . . . . .343<br />
Mergesort . . . . . . . . . . . . . . . . . . . . . . . . . . . 345<br />
Merkurjev, A. S. . . . . . . . . . . . . . . . . . . . . . . . 4<br />
Merkurjev-Suslin<br />
Theorem of . . . . . . . . . . . . . . . . . . . . . . . . . 4<br />
mesh generation<br />
2.5-dimensional. . . . . . . . . . . . . . . . . . . .292<br />
metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173<br />
bounded . . . . . . . . . . . . . . . . . . . . . . . . . . 174<br />
continuous. . . . . . . . . . . . . . . . . . . . . . . .174<br />
induced by a model . . . . . . . . . . . . . . . . 174<br />
minimum metric. . . . . . .xx, 156ff, 173, 246<br />
Minkowski, H. . . . . . . . . . . . . . . . . . . . . . . xiii<br />
’s lattice point theorem. . . . . . . . . . . . .171<br />
model. . . . . . . . . . .xix, 101f, 104, 108 f, 129,<br />
135f, 139, 142 f, 173ff, 177, 179, 181ff,<br />
187f, 193, 220, 223, 229, 234, 245, 252,<br />
267, 283, 300, 303<br />
modular operation . . . . . . . . . . . . . . . . . . . 341
INDEX 389<br />
Molien, T. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3<br />
Monte-Carlo method. . . . . . . . . . . . . . . . .227<br />
Mordell, L. J. . . . . . . . xiv, xix, 103, 113, 215<br />
’s conjecture. . . . . . . . . . . . . . . . . . . . . . .215<br />
’s examples of cubic surfaces . . . xiv, 103,<br />
113 – 128<br />
Moriwaki, A.. . . . . . . . . . . . . . . . . . . . . . . .165<br />
morphism twisted by σ. . . . . . . . . . . . . . . . .8<br />
moving lemma. . . . . . . . . . . . . . . . . . . . . . .164<br />
Mumford, D. . . . . . . . . . . . . . . . . . . . . . . . . . . 4<br />
N<br />
Néron-Severi group. . . . . . . . . . . . . . . .88, 91<br />
nef . . . . . . . . . . . . . . . . . . 177 – 180, 183, 185 f<br />
Noether, E.. . . . . . . . . . . . . . . . . . . . . . . . . . . .4<br />
Noether-Lefschetz Theorem . xvi, 244, 252<br />
non-archimedean valuation . . . . . . . . . . . 152<br />
non-Azumaya locus . . . . . . . . . . . 62, 85, 110<br />
non-obvious line . . . . . . . . xxi, 216, 255, 260<br />
norm<br />
of global section . . . . . . . . . . . . . . . . . . . 169<br />
norm variety . . . . . . . . . . . . . . . . . . . . . . . . 265<br />
normalized valuation. . . . . . . . . . . . . . . . .150<br />
Northcott’s theorem . . . . . . . . . . . . . . . . . 152<br />
numerical integration . . . . . . 251, 291f, 298<br />
O<br />
obvious line . . . . . . . . . . . xxi, 215f, 253, 255<br />
odd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285<br />
Oliver, J. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349<br />
open addressing. . . . . . . . . . . . . . . . . . . . . .335<br />
Opteron processor . . . . . . . . . . . . . . . . . . . 347<br />
order. . . . . . . . . . . . . . . . . . .76 – 79, 82ff, 187<br />
maximal . . . . . . . . . . . . . . . 77 – 83, 85, 187<br />
P<br />
p-adic measure . . . . . . . . . . . . . . . . . . . . . 229 f<br />
p-adic unsolvability . . . . . . . . . . . . . . . . . . 212<br />
p-adic valuation . . . . . . . . . . . . . . . . . 270, 317<br />
p-adic numbers . . . . . . . . . . . . . . . . . . . . . . xiii<br />
page prime . . . . . . . . . . . . . 333, 336, 341, 345<br />
paging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336<br />
parametrization . . . . . . . . . . . . . . . . . . . . . . xii<br />
Pari................................250<br />
partial presorting . . . . . . . . . . . . . . . . . . . . 347<br />
Pentium 4 processor . . . . . . . . . . . . . . . . . 348<br />
periodicity isomorphism, for cohomology<br />
of cyclic group . . . . . . . . . . . . . . . . . . 137<br />
permutation representation. . . . . . . . . . .285<br />
primitive. . . . . . . . . . . . . . . . . . . . . . . . . .288<br />
Peyre, E. . xvi, xix, xxi, 217, 224, 237 – 240,<br />
244, 266f, 282, 284, 296, 298, 301, 303<br />
Peyre’s constant . . . . . . . . . . . . . . . . xxi f, 224,<br />
237 – 240, 243ff, 247, 260, 266ff, 277,<br />
282ff, 295f, 298, 301 f, 311, 325f<br />
Peyre’s Tamagawa type number . . . . see also<br />
Peyre’s constant<br />
Picard functor . . . . . . . . . . . . . . . . . . . . . . . . 87<br />
Picard group. . . . . . . . . . . . . . . . . . . . . . . . . .91<br />
Picard rank<br />
of cubic surface. . . . . . . . . . . . . . . . . . . . .97<br />
Picard scheme . . . . . . . . . . . . . . . . . . . . . . . . 87<br />
Poincaré residue map. . . . . . . . . . . . . . . . .233<br />
Poincaré-Lelong equation. . . . . . . . . . . . .162<br />
post-processing . . . . . . . . . . . . . . . . . . . . . . 342<br />
presorting, partial. . . . . . . . . . . . . . . . . . . .345<br />
prime number theorem<br />
Dirichlet’s. . . . . . . . . . . . . . . . . . . . . . . . .270<br />
principal divisor<br />
norm of. . . . . . . . . . . . . . . . . . . . . . . . . . .137<br />
product formula . . . . . . . . . . . . . . . . . . . . . 150<br />
projective plane. . . . . . . . . . . . . . . . . . . . . .257<br />
projective space of lines. . . . . . . . . . . . . . . .42<br />
Proposition<br />
of Châtelet-Artin. . . . . . . . . . . . . . . .34, 38<br />
of Severi . . . . . . . . . . . . . . . . . . . . . . . . . . . 32<br />
Puiseux expansion . . . . . . . . . . . . . . . . . . . 252<br />
pull-back<br />
of Azumaya algebra. . . . . . . . . . . . . . . . .63<br />
on adelic Picard group . . . . . . . . . . . . . 182<br />
on arithmetic Chow group . . . . . . . . 163f<br />
on group cohomology . . . . . . . . . . . . . . . 7<br />
pure cubic field<br />
discriminant of. . . . . . . . . . . . . . . .309, 316<br />
push-forward<br />
on arithmetic Chow group . . . . . . . . 164f<br />
pyramid. . . . . . . . . . . . . . . . . . . . . . . . . . . . .350<br />
Q<br />
quadratic form . . . . . . . . . . . . . . . . 338ff, 342
390 INDEX<br />
quadratic reciprocity low . . . . . . . . . . . . . xiv<br />
quadric. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .257<br />
quartic surface<br />
diagonal . . . . . . . . . . . . . . . . . . . . . . . . . . 218<br />
quartic threefold.xxi f, 240, 243 – 279, 284,<br />
301<br />
diagonal . . . . . . . . . . . . xxi f, 213, 224, 284<br />
Quicksort . . . . . . . . . . . . . . . . . . . . . . . . . .344f<br />
Quillen, D.. . . . . . . . . . . . . . . . . . . . . . . . . .170<br />
R<br />
Radix sort . . . . . . . . . . . . . . . . . . . . . . . . . . .345<br />
r<strong>and</strong>om number generator. . .xxii, 284, 292<br />
<strong>rational</strong> <strong>points</strong><br />
Brauer-Manin obstruction to . . . . . . . 106<br />
<strong>rational</strong> surface . . . . . . . . . . 4, 89, 92ff, 255f<br />
<strong>rational</strong> variety . . . . . . . . . . . . . . . . . . . . . . 329<br />
Reading . . . . . . . . . . . . . 259, 339f, 342, 346 ff<br />
reciprocity low<br />
biquadratic. . . . . . . . . . . . . . . . . . . . . . . . .xv<br />
cubic . . . . . . . . . . . . . . . . . . . . . . . . . . xv, 145<br />
quadratic . . . . . . . . . . . . . . . . . . . . . . . . . . xiv<br />
reduced trace . . . . . . . . . . . . . . . . . . . . . . . . . 76<br />
representability, of a functor . . . . . . . . . . . 53<br />
restriction . . . . . . . . . . . . . . . . . . . . . 7, 36, 134<br />
Riemann zeta function . . . . . . . . . . . . . . . 316<br />
ruled surface. . . . . . . . . . . . . . . .252, 255, 258<br />
ruled variety. . . . . . . . . . . . . . . . . . . . . . . . .329<br />
S<br />
Sansuc, J.-J. . . . . . . xix, 103, 129f, 239, 286f<br />
Schappacher, N. . . . . . . . . . . . . . . . . . . xi, 353<br />
Schröer, S.. . . . . . . . . . . . . . . . . . . . . . . . . . . .85<br />
Schwarz, H. A. . . . . . . . . . . . . . . . . . . . . . . 292<br />
’s cylindrical surface . . . . . . . . . . . . . . . 292<br />
search bound . . . xxiii, 260, 268, 284, 293 ff,<br />
304, 348f<br />
searching for <strong>rational</strong> <strong>points</strong> . . . . . . . see also<br />
searching for solutions of Diophantine<br />
equation<br />
searching for solutions of Diophantine<br />
equation.xxi, xxiii, 259, 284, 294, 296,<br />
332 – 348<br />
naive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332<br />
selection of bits . . . . . . . . . . . . . . . . . . . . . . 341<br />
semi-abelian variety<br />
split . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192<br />
semipositive. . . . . .177ff, 181, 188, 194, 204<br />
semipositive cone . . .177ff, 181 – 184, 187f<br />
sequence<br />
generic. . . . . . . . . . . . . . . . . . . . . . . . . . . .204<br />
small . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204<br />
Serre, J.-P. . . . . . . . . . . . . . . . . . . . . . . . . . 4, 189<br />
’s homological characterization of regularity<br />
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 19<br />
’s vanishing theorem . . . . . . . . . . . . . . . . 17<br />
Theorem of . . . . . . . . . . . . . . . . 24, 30, 189<br />
Severi, F. . . . . . . . . . . . . . . . . . . . . . . . . . . . 4, 32<br />
sheaf of Azumaya algebras . . . . . . . . xvii, 61<br />
Siegel, C. L. . . . . . . . . . . . . . . . . 265, 283, 317<br />
’s estimate. . . . . . . . . . . . . . . . . . . . . . . . .317<br />
σ-linear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8<br />
Sikorsky, I. I. . . . . . . . . . . . . . . . . . . . . . . . . 265<br />
similarity<br />
of Azumaya algebras . . . . . . . . . . . 65 f, 70<br />
of central simple algebras . . . . . . 4, 22, 38<br />
simple group . . . . . . . . . . . . . . . . . . . . . . . . 285<br />
SINGULAR.. . . . . . . . . . . . . . . . . . . . . .288, 293<br />
64 bit processor. . . . . . . . . . . . . . . . . . . . . .341<br />
Skolem-Noether, Theorem of . . . 23, 67, 69<br />
smallest point . . . . . . . . . . . . . . . . . . . 283, 295<br />
solutions of Diophantine equation<br />
algorithm to search for. . .xxi, xxiii, 259,<br />
284, 294, 296, 332 – 348<br />
naive . . . . . . . . . . . . . . . . . . . . . . . . . . . 332<br />
sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345<br />
Soulé, C. . . . . . xx, 156, 160, 163ff, 167, 170<br />
splitting field . . 23, 28, 32, 34ff, 38, 46f, 49,<br />
58, 60, 75, 288<br />
sporadic line. . . . . . . . . . . .xxi, 216, 255, 260<br />
∗-product of Green’s currents. . . . .163, 165<br />
Stark, H. M.. . . . . . . . . . . . . . . . . . . . . . . . .317<br />
statistical parameters . . . . . . . . . . . 261ff, 295<br />
Steiner surface . . . . . . . . . . . . . . . . . . . . . . . 258<br />
successive differences . . . . . . . . . . . . . . . . . 298<br />
successive minima.xx, 169f, 172, 188, 194f<br />
Sullivan, A.. . . . . . . . . . . . . . . . . . . . . . . . . .243<br />
surface . . . . . . . . . . . . . . . . . . . 85, 94, 224, 252<br />
abelian . . . . . . . . . . . . . . . . . . . . . . . 218, 255
INDEX 391<br />
algebraic. . . . . . . . . . . . . . . . . . . . . . . .71, 85<br />
bielliptic. . . . . . . . . . . . . . . . . . . . . .255, 329<br />
cubic. . .xv – xix, xxi f, 94 – 97, 103, 106,<br />
113f, 122, 125 – 130, 143, 224 – 228,<br />
234, 239f, 252f, 256, 281 – 328, 353<br />
diagonal . . . . . . . . xxii, 103, 128 ff, 143,<br />
301 – 328<br />
general . . . . . . xxii, 106, 240, 281 – 300<br />
elliptic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329<br />
Enriques . . . . . . . . . . . . . . . . . . . . . . . . . . . 89<br />
Godeaux . . . . . . . . . . . . . . . . . . . . . . . . . .215<br />
in P 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xxiii<br />
K3 . . . xxii f, 89, 92, 94, 218ff, 255, 329ff<br />
Kummer . . . . . . . . . . . . . . . . . . . . . . . . . . 218<br />
non-minimal . . . . . . . . . . . . . . . . . . . . . . 220<br />
non-separated . . . . . . . . . . . . . . . . . . . . . . 85<br />
of general type. . . . . . . . . . . . . . .215, 254 f<br />
of Kodaira dimension one . . . . . . . . . . 255<br />
of Kodaira dimension zero . . . . . . . . . 217<br />
<strong>rational</strong> . . . . . . . . . . . . . . . 4, 89, 92ff, 255 f<br />
ruled. . . . . . . . . . . . . . . . . . . . .252, 255, 258<br />
Steiner. . . . . . . . . . . . . . . . . . . . . . . . . . . .258<br />
surfaces<br />
classification of . . . . . . . . . . . . . . . . . . . . 329<br />
Suslin, A. A. . . . . . . . . . . . . . . . . . . . . . . . . . . .4<br />
suspicious<br />
pair. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .257<br />
point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255<br />
Swinnerton-Dyer, Sir P. xiv, xvii, xix, xxiii,<br />
96, 103, 124, 133, 263, 297, 329f, 335<br />
’s list. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .133<br />
Sylow subgroup . . . . . . . . . . . . . . . . . 285, 287<br />
Szpiro, L. . . . . . . . . . . . . . . . . . . . xx, 190, 192<br />
Szpiro, Ullmo, <strong>and</strong> Zhang, equidistribution<br />
theorem of . . . . . . . . . . . . . . . . . . xx, 190<br />
T<br />
Tamagawa measure. . . . . . . . . .234, 246, 303<br />
Tamagawa number see also Peyre’s constant<br />
Tate, J. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143<br />
’s algorithm . . . . . . . . . . . . . . . . . . . . . . . 142<br />
Tate cohomology . . . . . . . . . . . . . . . . . . . . . 94<br />
tensor field . . . . . . . . . . . . . . . . . . . . . . . . . . 230<br />
test for conic through two <strong>points</strong> . . . . . 256<br />
Theorem<br />
Hilbert 90 . . . . . . . . . . . 26, 33, 71, 90, 112<br />
of Ausl<strong>and</strong>er-Goldman. . . . . .xvii, 71, 84<br />
of Bezout . . . . . . . . . . . . . . . . . . . . . . . . . 253<br />
of Bilu . . . . . . . . . . . . . . . . . . . xxi, 190, 192<br />
of Chambert-Loir . . . . . . . . . . . . . xxi, 192<br />
of Colliot-Thélène, Kanevsky, <strong>and</strong> Sansuc<br />
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 129f<br />
of Hasse-Minkowski . . . . . . . . . . . . . . . 281<br />
of Legendre. . . . . . . . . . . . . . .xiii, 265, 283<br />
of Lichtenbaum . . . . . . . . . . . . . . . . . . . . 92<br />
of Linnik . . . . . . . . . . . . . . . . . . . . . . . . . 312<br />
of Merkurjev-Suslin . . . . . . . . . . . . . . . . . . 4<br />
of Noether-Lefschetz . . . . . . . . . . 244, 252<br />
of Northcott . . . . . . . . . . . . . . . . . . . . . . 152<br />
of Serre . . . . . . . . . . . . . . . . . . . . 24, 30, 189<br />
of Skolem-Noether. . . . . . . . . . .23, 67, 69<br />
of successive minima . . . . . . . . . . 170, 188<br />
of Szpiro, Ullmo, <strong>and</strong> Zhang. . . .xx, 190<br />
of Tietze . . . . . . . . . . . . . . . . . . . . . . . . . . 160<br />
of Tsen. . . . . . . . . . . . . . . . . . . . . . . . . . . . .89<br />
of Zak . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253<br />
Weak Lefschetz. . . . . . . . . . . . . . . . . . . .247<br />
Thompson, S. . . . . . . . . . . . . . . . . . . . . . . . 211<br />
threefold . . . . . . . . . . . . . . . . . . . 212, 282, 302<br />
cubic . . . . . . . . . . .xxi, 215, 240, 243 – 263<br />
diagonal. . . . . . . . . . . . . . .xxi f, 213, 224<br />
quartic. . . .xxi f, 240, 243 – 279, 284, 301<br />
diagonal . . . . . . . . . . xxi f, 213, 224, 284<br />
Tietze’s Theorem . . . . . . . . . . . . . . . . . . . . 160<br />
Tor-formula . . . . . . . . . . . . . . . . . . . . . . . . . 163<br />
toric variety . . . . . . . . . . . . . . . . . . . . . . . . . 192<br />
Tregub, S. L.. . . . . . . . . . . . . . . . . . . . . . . . . .37<br />
triangular mesh . . . . . . . . . . . . . . . . . . . . . . 291<br />
Tschinkel, Y. . . . . . . . .xiv, xvi, xix, 223, 239<br />
Tsen’s theorem. . . . . . . . . . . . . . . . . . . . . . . .89<br />
27 lines on cubic surfacexvii, xxii, 95f, 106,<br />
126ff, 131f, 144, 225 – 228, 284 – 288,<br />
293, 297, 299 f, 305f, 312, 353<br />
2.5-dimensional mesh generation. . . . . .292<br />
Tychonov topology . . . . . . . . . . . . . . . . . . 102<br />
U<br />
Ullmo, E. . . . . . . . . . . . . . . . xx, 190, 192, 208
392 INDEX<br />
V<br />
valuation<br />
archimedean . . . . . . . . . . . . . . . . . . . . . . 152<br />
lying above another. . . . . . . . . . . . . . . .151<br />
non-archimedean . . . . . . . . . . . . . . . . . . 152<br />
van der Waerden’s criterion . . . . . . . . . . . 288<br />
variety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii<br />
Fano . . . . . . . . . . . . . . . . . . . . . . xii, xiv, xxi,<br />
213f, 217, 220f, 223, 236 f, 239, 243f,<br />
246, 265, 282, 285, 301<br />
of general type xii f, 213 ff, 217, 254 f, 329<br />
of intermediate type . . . xii, 26, 31, 213f,<br />
217, 297<br />
Vistoli, A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85<br />
von Neumann, J. . . . . . . . . . . . . . . . . . . . . 281<br />
W<br />
weak approximation. . .xvi, xix, xxii, 102f,<br />
106, 128, 240<br />
Brauer-Manin obstruction to . . xxii, 106,<br />
122f, 127, 130, 283, 286<br />
counterexample to . . . . . . . . . . . . . . . . . xvi<br />
Weak Lefschetz Theorem. . . . . . . . . . . . .247<br />
Weil, A. . . . . . . . . . . . . . . . . . . . . . xix, 22, 331<br />
Weil conjectures . . . 235, 247, 269, 274, 331<br />
for curves . . . . . . . . . . . . . . . . . . . . 331, 340<br />
Witt, E. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4<br />
Writing . . . . . . . . . . . . . 259, 340ff, 345, 347f<br />
Y<br />
Yoneda’s lemma. . . . . . . . . . . . . . . . . . . . . .107<br />
Z<br />
Zak’s theorem . . . . . . . . . . . . . . . . . . . . . . . 253<br />
zeroth cohomology set . . . . . . . . . . . . . . . . . 5<br />
Zhang, S. . . . . . . . . . . xx, 172f, 187, 190, 192<br />
Zorn’s lemma . . . . . . . . . . . . . . . . . . . . 10, 107