Compositio Mathematica 105: 29–41, 1997.

29
c 1997 Kluwer Academic Publishers. Printed in the Netherlands.

On the size of the Shafarevich–Tate group of elliptic

curves over function fields

C. S. RAJAN
School of Mathematics, Tata Institute of Fundamental Research, Homi Bhabha Road,
Mumbai – 400 005, India. e-mail: [email protected]

Received 19 June 1995; accepted in final form 3 August 1996

Abstract. Let E be a nonconstant elliptic curve, over a global field K of positive, odd characterisitc.
Assuming the finiteness of the Shafarevich-Tate group of E , we show that the order of the Shafarevich-
Tate group of E , is given by O(N 1=2+6 log(2)= log(q) ), where N is the conductor of E; q is the cardinality
of the finite field of constants of K , and where the constant in the bound depends only on K . The
method of proof is to work with the geometric analog of the Birch-Swinnerton Dyer conjecture for
the corresponding elliptic surface over the finite field, as formulated by Artin-Tate, and to examine
the geometry of this elliptic surface.
AMS classification: 11G05 11G25, 14G15

Key words: Elliptic curve, Shafarevich–Tate group, conductor.

1. Introduction
The class group of a number field K ‘measures’ the extent of departure of the
ring of integers of K from being a unique factorisation domain. It is known to be
finite and the class number formula of Dirichlet gives an explicit formula for the
order of the class group of K , the class number. It follows from this using partial
summation and the absolute lower bound for the regulator of number fields shown
in ([22]), that the class number of a number field K is bounded by K
(1=2)+ , where


K is the discriminant of K . In the case of number fields, the conductor NK of
the Dedekind zeta function of K is equal to the discriminant of K , and so one has
that the class number is bounded by NK
(1=2)+ .
Let E be an elliptic curve defined over a global field K . The analogue of the
( )
class group is the Shafarevich–Tate group III E=K of E over K . For a place v of
K , let Kv denote the completion of K at v. The Shafarevich–Tate group III E=K ( )
is defined to be
!
Y
III (E=K ) := Ker H (K; E ) ! H (Kv ; E ) ;
1 1
v
where the cohomology groups are the Galois cohomology groups, and v runs over
the places of K . The Shafarevich–Tate group measures the extent to which the

JEFF. PREEPROOF: CRCs Corr. M/c 4: PIPS Nr.:119129 MATHKAP

comp3920.tex; 8/05/1997; 8:16; v.5; p.1
30 C. S. RAJAN

( )
Hasse principle fails for E=K . It is conjectured that III E=K is finite and this is
known to hold for certain classes of elliptic curves. Morever an effective bound for
( )
III E=K would be useful in computing a set of generators for the Mordell–Weil
( )
group E K of K -rational points of E . Our results indicate that one can indeed
obtain a bound analogous to the bound obtained for the class numbers of number
fields, for elliptic curves over global fields of positive characteristic, provided one
assumes the finiteness of III.
Manin ([11]) and Lang ([7]), conjectured bounds for the size of III based on
various arithmetical invariants associated to the elliptic curve. Inspired by these
conjectures, Mai and Murty ([8]) in 1992, predicted the growth of III E=Q as ( )
satisfying,

jIIIj = O(N (1=2)+ );

where N is the conductor of the elliptic curve. In ([9]), they produce an infinite
family of elliptic curves E=Q such that,

jIII(E )j  N (1=4)+ :
Since ([8]) has never been published, I have included their argument in an appendix.
Some of the conjectures used in the argument over number fields, viz., Szpiro’s
conjecture and Lang’s conjecture on lower bounds for the canonical heights of
rational points, are known to hold over global fields of positive characteristic ([13,
5]). It would be thus of interest to verify this conjecture for global fields of positive
characteristic.
However a fundamental problem for function fields, is that the estimate on the
rank of the Mordell–Weil group of an elliptic curve is not as sharp as expected for
=
number fields. Let k be a finite field of odd characteristic, with q pf elements.
Let C denote an irreducible, smooth projective algebraic curve over k with function
( )=
field K , genus g , and Euler characteristic  C 2 2g . Let K 0 be the function

field of C over k . Let E be an elliptic curve defined over K with conductor N . A.
Brumer has shown ([1])
!
r(E ) 6 (
log N ) ( logq (N ) 4g + 4) log2 (q )
( ( ) 4g + 4) + O pq log2(logq (N ) 4g + 4) ;
2 log logq N

where r (E ) is the rank of the Mordell–Weil group E (K ). Using this and arguing
as in the case of number fields, provides an upper bound for III, which grows faster
than any power of N , and will not provide the conjectured O (N (1=2)+ ). For details
we refer to the last remark in the Appendix. Thus it does not seem to be possible to
establish the desired bounds on III, arguing completely in analogy with the number
fields.
()
For an abelian group A, let A l denote the l-primary component of A. We have

comp3920.tex; 8/05/1997; 8:16; v.5; p.2

THE SIZE OF SHAFAREVICH – TATE GROUP OF ELLIPTIC CURVES 31

THEOREM 1. Let E be an elliptic curve over K as above. Assume that III l is ()

finite for some prime l and that the j -invariant jE of E is transcendental over k .
Then
jIII(E )j 6 (16q) (C )=2 (212q) pe (C )=2 N pe ((1=2)+(6 log(2)= log(q))
where pe is the inseparable degree of K over k jE .( )
In our proof, we work with the corresponding elliptic surface rather than directly
working with the elliptic curve, and with the analogue of the Birch–Swinnerton
Dyer conjecture for the surface. Instead of the regulator and the canonical height
pairing for the elliptic curve, we have the Neron–Severi group of the surface and
the intersection pairing on the divisors on the surface, which takes integral values.
Morever the product over the periods occuring in the Birch–Swinnerton Dyer
formula, is ‘replaced’ by the term q (X ) , which has an expression in terms of the
cohomology of the surface, and is thus easier to compute.

2. Conjectures of Artin and Tate

Let X be the proper, minimal, regular model in the birational equivalence class of
=
surfaces fibered over C and having generic fiber isomorphic to E . Let X X k k . 
Let  : X ! C denote the projection map of X onto C . X is an elliptic surface
( )
and  has a natural section associated to the zero element of E K .
(
Let P2 X; T )= ( )
det 1 2;l T denote the characteristic polynomial of the
( )
endomorphism 2;l of the etale cohomology groups Het2 X; Ql induced by the
( )
Frobenius morphism on X . Deligne has shown in ([2]), that P2 X; T is a polyno-
Q of l, and that the analogue
mial with rational integral coeffecients, is independent
(
of the Riemann hypothesis holds i.e., P2 X; T )= ( )
1  T with j j q . =
Inspired by the Birch–Swinnerton Dyer conjecture, Tate ([19]) conjectured that
the order of zeros of the L-function should be related to geometry of the cycles
(
on X . In our situation the conjecture predicts that q 1 is a root of P2 X; T with )
( ) ( )
mutiplicity exactly  X , where  X , the base number of X , is the rank of the
( ) ( )
Néron-Severi group NS X of X . NS X is defined to be the image of Pic X ( )
( ) 
in NS X , the Néron-Severi group of X , which is the group of divisors taken
modulo algebraic equivalence. Under our hypothesis on the j -invariant of E , it is
( )
known by Néron’s theorem of the base ([6]), that NS X is a finitely generated
abelian group.
(
Let R X; T )= ( P2 X; T = 1 qT (X ) . By Tate’s conjecture R X; T is a
)( ) ( )
polynomial with rational integral coeffecients which does not vanish when T =
q 1.
In our situation, a theorem of Artin ([20]) applies, and we have that
Br(X ) ' III(E=K );
where Br(X ) is the Brauer group of the surface X . Led by this result Artin and
Tate formulated a geometric analogue of the Birch–Swinnerton Dyer conjecture,

comp3920.tex; 8/05/1997; 8:16; v.5; p.3

32 C. S. RAJAN

describing the leading coefficient of the L-function at the center of the critical
( )
strip, where the role of the Shafarevich–Tate group III E is replaced by that of the
( )
Brauer group Br X of X :

R(X; q 1 ) = jBr (X )jjdet(Di; Dj )j

q (X ) jNS (X ) j2
 tors
(1)

( ) ( )
where Di 16i6 is a base for NS X modulo torsion. The symbol Di ; Dj ( )
denotes the total intersection multiplicity of the divisors Di and Dj . The term
( ) ( )
jdet Di; Dj j is the regulator for the intersection pairing on NS X .
(X ) = (X; OX ) 1 + dim (Pic(X )): (2)

It is known that (X ) is non-negative.

In ([20]) it is shown that if Tate’s conjecture is assumed to hold for X , then
the subgroup of elements of Br(X ), of order prime to p is finite, and (1) is true
upto a power of p. In ([12]), Milne showed that the conjecture of Tate is equivalent
to assuming the finiteness of Br(X )(l) for some prime l which can also be p.
Morever Milne showed that Tate’s conjecture implies the conjecture of Artin and
Tate. Consequently we have that if III(E )(l) or equivalently Br(X )(l) is finite for
some prime l, then

NS(X )tors j2
jIII(E=K )j = jBr(X )j = R(X; q 1 )q(X ) jjdet(D ; D )j : (3)
i j
( )
Thus to obtain an estimate for jBr X j, we estimate each of the terms in the above
( )
expression for jBr X j, in terms of the conductor N .

3. Estimating the regulator

( )
We need a description of the Néron-Severi group NS X ([15]). Let K 0 be the

function field of C over k . It is known by the theorem of Lang–Néron ([6]), that
under our assumption on E , i.e., the j -invariant of E being transcendental over k ,
( )
that E K 0 is a finitely generated abelian group. Morever the torsion of E K 0 is ( )
generated by at most two elements tj ; j J; jJ j 6 2 with orders nj respectively.
( ) ( )
Let r; r 0 denote the rank of the Mordell–Weil groups E K ; E K 0 respectively.
( )
We have r 6 r 0 . Let si ; 1 6 i 6 r 0 be a set of generators of E K 0 modulo
( )
torsion. Since X is proper, the group E K 0 is canonically identified with the
  () 
group of sections of  : X ! C . Denote by s the image in X of the section of
( ) () () ()
 corresponding to a rational element sE K 0 , and by D s the divisor s 0

on X .
For a place v of C , let Xv denote the fiber over v . Let S be the finite set of places
on C , the ramification locus of the map  : X ! C , where the fiber Xv is singular.

For vS , let mv be the number of irreducible components of the fiber Xv counted

comp3920.tex; 8/05/1997; 8:16; v.5; p.4

THE SIZE OF SHAFAREVICH – TATE GROUP OF ELLIPTIC CURVES 33

 ( )
without multiplicity. Denote by v;i ; 0 6 i 6 mv 1 the irreducible components
()
of the fiber Xv , with the convention that the 0 divisor meets v;0 . Let u0 be a 

point on C outside S . If D1 ; D2 are two divisors on X , denote by D1 ; D2 the ( )
intersection multiplicity, which is a rational integer.
By decomposing divisors into ‘horizontal’ sections and ‘vertical’ fibers, it is
( )
shown in ([15]), that the Néron–Severi group NS X is generated by D s ; 1 6 ( )(
) ( ) ()   (
 6 r0 ; D tj ; j 2 J; 0 ; Xu0 ; v;i ; v 2 S; 1 6 i 6 mv 1 , with the following )
relations for j 2 J :
X
nj D(tj ) ' nj (D(tj ); (0))Xu + 0 nj (v;1 ; : : : v;mv 1)
v;i
Av 1((D(tj ); v;1 ); : : : ; (D(tj ); v;mv 1 ))t ; (4)
where ' denotes algebraic equivalence. Av is the intersection matrix defined by
 ( ) = (  ) (
the fiber Xv , i.e., Av ij )
v;i v;j ; 1 6 i; j 6 mv 1 . It is known that Av is
invertible and negative definite.
 = ( )
Consequently it follows that the rank   X of NS X is given by ( )
X
(X ) = r0 + 2 + (mv 1 ): (5)
v2S
In order to estimate the regulator, we need the following result due to Shioda ([16,
Theorem 3.1]).

PROPOSITION 2. Suppose X is a non-isotrivial elliptic surface as above. Then
numerically equivalent divisors are algebraically equivalent. Hence the Néron–
( )
Severi group NS X is torsion-free.
Proof. Since this fact is of basic importance to us, we give a brief outline of

the proof, following Shioda ([16]). Given a divisor D on X numerically equivalent
to 0, by Grothendieck–Riemann–Roch, we have  X; O D ( ( )) = (  )
  X; OX . By
the formulas (11) and (12), and by our assumption that the j -invariant jE of E
(  ( ))
is non-constant, we see that  X; O D > 0. Hence either h0 O D > 0 or ( ( ))
( ( ))
h2 O D > 0.
( ( ))
If h0 O D > 0, then D is linearly equivalent to an effective divisor and
numerically equivalent to 0, which implies that D is algebraically equivalent to 0.
( ( ))
If h2 O D > 0, then by duality K D is linearly equivalent to an effective

divisor D 0 , where K is the canonical divisor of X . Since on an elliptic surface K is
fibral, by our assumption on D , we have D 0 ; ( ) =
0, for any vertical divisor on 

X . This forces D0 and hence D to be algebraically equivalent to a vertical divisor.
Morever, it is easy to see from our assumption on D , and the non-degeneracy of the
intersection pairing of the divisors on any fiber, that D is algebraically equivalent

to 0 on X .
Since the intersection pairing is non-degenerate on the group of divisors modulo
numerical equivalence, this gives us that the Néron–Severi group NS X is torsion- ( )
free.

comp3920.tex; 8/05/1997; 8:16; v.5; p.5

34 C. S. RAJAN

( ) (
This allows us to estimate jNS X tors j2 =jdet Di Dj j. Since NS X is a sub-) ( )
( )
group of NS X it is torsion free. The intersection numbers being rational integers,
the determinant of the intersection matrix is an integer and we know it is non-zero.
Hence we have
jNS(X )tors j2 6 1:
jdet(DiDj )j (6)

4. Estimating R (X; q 1 )
Let Bi (1 6 i 6 4) denote the Betti numbers of the surface X , the dimension of
 Ql ) over Ql . The degree of the polynomial P2 (X; T ) is B2. By the results of
Heti (X;
Artin and Ogg–Shafarevich–Grothendieck ([14]), one has a formula for the Picard
number B2 , the dimension of the space of ‘transcendental cycles’ on X  , in
terms of the exponents of the conductor.
Let M be a l-torsion abelian group which is ‘cofinite’, i.e., is of the form
( )
Ql =Zl s  M0 ; M0 a finite group. Then s is called as the corank of M . It is shown
 
in ([4]), under our hypothesis on the map  : X ! C , that B2  is the corank 
() ( )
r0 l of the l-torsion of Br X ' H 2 X; m . ( G )
Let  be the generic point of C . Arguing using the Leray spectral sequnece for
(
the map  , Artin remarks that the map H 1 C;  X ! H 2 X; Gm has finite ) ( )
kernel and cokernel and hence the coranks are same. For almost all l, then the corank
( )
of H 1 C;  X is given by the formula of Ogg–Shafarevich–Grothendieck as
X
r0(l) = 4g 4 r0 + fv ;
v
where g is the genus of the curve C and fv is the exponent of the conductor N of
the elliptic curve E at v . Hence
X
B2  = 4g 4 r0 + fv : (7)
v2S


Let v denote the minimal discriminant of E at v , and let ordv v denote the (
)


exponent of the minimal discriminant v at v . We have the following important
formula due to Ogg([18]), relating the exponents of the discriminant, the conductor
and the number mv of irreducible components, counted without multiplicity of the
singular fibers

ordv (
v ) = fv + mv 1: (8)

Hence by (5) and (7), we have

X
B2 = 4g 2 + ordv (
v ): (9)
v2S

comp3920.tex; 8/05/1997; 8:16; v.5; p.6

THE SIZE OF SHAFAREVICH – TATE GROUP OF ELLIPTIC CURVES 35

( )( )
On the assumption that Br X l is finite Qfor some prime l, we have by the
results of Deligne and Milne, that R X; T
Q ( )=
 1  T where j j q and( ) =
=
 6 q. Hence R X; q ( 1 )= ( )
 1 w with jw j 1; w 6 1. Morever the = =
( )
degree of R X; T is B2 , which is less than B2 r 2, where r is the rank of
( )
E K , the Mordell–Weil group of E over K . Hence
X
deg R (X; T ) 6 4g 4 r+ ordv (
v )
v2S
Hence
Y
jR(X; q 1 )j = j(1 w )j

6 2 r 24g 4
:2v 2s ordv (
v ) : (10)

5. Estimating q (X )
Let OX denote the structure sheaf of X . It is known for an elliptic surface, the
= ( (  )) 
irregularity q dim Pic X of X is the genus g of the curve C . By Weil’s theory
of Jacobians one has that B1 = =
2q 2g . By duality B3 g. Hence from (9),
2P =
it follows that the topological Euler characteristic top X i ( )= ( )
i 1 Bi is given
by the following interesting formula
X
top (X ) = ordv (
v ): (11)
v2S
By semicontinuity we have  X; OX ( )= ( )
 X; OX . Morever it is known that
( )
canonical divisor is fibral, and is of the form   T , where T is a divisor on
=
C . This gives us K 2 0, where K is the canonical bundle of X . Hence by
Grothendieck–Riemann–Roch formula, we have

(X )top = 12(X; OX ) (12)

We have
(X ) = (X; OX ) 1 + dim (Pic(X ))
!
1 X
= 12 v2S
ordv v
+g 1 (13)

by (2), (11) and (12).

Remark. It is expected that a decomposition as above for  X , as a sum of ( )
local contributions from the ramified primes, should hold in general for any motive.

comp3920.tex; 8/05/1997; 8:16; v.5; p.7

36 C. S. RAJAN

( )
However it is interesting to note that even though  X is an integer, the terms
corresponding to the individual ramified places need not be integral. A different
( )
formula for  X has been obtained by Milne ([12]).
We would also like to recall a conjecture of Szpiro’s, and which is known to
hold over function fields of arbitrary characteristic ([13]),
( )
X X
ordv (
v ) 6 6pe fv + 2g 2 ; (14)
v v
where pe is the inseparable degree of K over k (jE ). Hence we get by (13) and
(14), an estimate for the ‘period’,

q(X ) = (qpv )1=2 q(pe +1)(g 1)

ef

6 q (pe +1)(C )=2 N pe =2 : (15)

We can also express the estimate we obtained for jR(X; q 1 )j in terms of the
conductor as

jR(X; q 1 )j 6 2 r 24g 4 26pe ( f + 2g

v v 2 )
Now N =Q f f
v qvv > q v , where qv is the P
cardinality of the residue field at vS
and we have qv > q . So log N =log q > v2S fv and 2fv 6 2log(N )=log(q)
( ) () =
N log(2)=log(q) . Thus
jR(X; q 1 )j 6 2 r (4 (C ) 2 6pe  (C ) )N 6pe (log(2)= log(q) : (16)

6.
( )( )
Under the assumption that Br X l is finite for some prime l, and that the j -
invariant of E is transcendental over k , we get from (3), (6), (15) and (16),
jIII(E=K )j
= jBr(X )j
6 2 r (16q) (C )=2 (212q) pe (C )=2 N pe ((1=2) + (6 log(2)= log(q)):

Remark. It is interesting to notice the terms which contribute to N 1=2 in the

function field and the number field. Over number fields it is the infinite period
1 which satisfies the relation j1 j > H 1=12 , where H is the height of the
elliptic curve which seems to be contributing to the N 1=2 term. In the function
fields it is the term q (X ) which contributes. One has an ‘analogous’ relationship
( ) = ( )
 X top 12 X; OX . This is in agreement with the belief that the archimedean
places are to be considered as ‘ramified’ places for an elliptic curve, and the formula

comp3920.tex; 8/05/1997; 8:16; v.5; p.8

THE SIZE OF SHAFAREVICH – TATE GROUP OF ELLIPTIC CURVES 37

(13), expressing  (X ) as a sum of contributions over the ramified primes of the

elliptic curve.

Appendix?
Let E be an elliptic curve over Q of conductor N and minimal discriminant .

Let III denote the Shafarevich–Tate group of E . Manin ([11]) and Lang ([7]) have
suggested that the Birch–Swinnerton Dyer conjecture can be used to give upper
bounds for III. Using methods of analytic number theory Mai and Murty ([8]),
conjectured the following
CONJECTURE 1. For any  > 0,

jIIIj = O(N (1=2)+ );

where the implicit constant depends only on .
In fact they make the stronger
CONJECTURE 2. There is an absolute constant c such that
  
jIIIj = O H (1=12 exp c logloglogNN ;
where the implied constant is absolute and where H is a naive height of E defined
as follows
= + +
Let y 2 x3 ax b be an equation for E over Q. Define

H = max(jaj3 ; jbj2 ):
Let LE (s) be the L-series associated to E . Define

SE (1 + it) = arg LE (1 + it);

where the argument is obtained by continuous variation along the straight lines
+ +
joining 2; 2 it; 12 it starting with value zero, provided t is not the ordinate of a
zero. Define SE t ()= (+)
lim!0SE t  , if t is the ordinate of a zero of LE s . In ()
analogy with a conjecture of Montgomery, we make
CONJECTURE 3. For jtj > 2
 log Nt
1=2 !
SE (t) = O log log Nt
? I would like to thank Ram Murty for allowing me to include the results of ([8]) here. R. Murty
informs me that these conjectures were presented at the Newton Institute in May 1993, but were never
published.

comp3920.tex; 8/05/1997; 8:16; v.5; p.9

38 C. S. RAJAN

By arguments as in ([21], page 354) and by using the Phragmen–Lindelof

theorem as in ([10]), it can be seen that
THEOREM. Assuming Conjecture 3, we have
 log Nt
1=2 !
jLE (1 + it)j 6 exp A log log Nt
(17)

for some absolute constant A.

COROLLARY. If r = ords=1LE (s) and Conjecture 3 is true, then
 log N
1=2 !
r=O log log N
; (18)

where the implied constant is absolute

Proof. Let C be a circle of radius 1= log log N centred at 1. By Jensen’s theorem,
we have
r6 log maxz"C jLE (z )j
 log N
1=2 !
=O log log N
by the above theorem.
We recall now a conjecture of Lang on lower bounds for the canonical height
of rational points on E . Let h denote the canonical Neron–Tate height on E .
CONJECTURE (Lang). For any non-torsion rational point P 2 E (Q),
h(P )  log j
j
for some positive constant , which does not depend on E .
This conjecture has been proved by Silverman in the case that E has integral
j -invariant.
We also recall a conjecture of Szpiro and Hall relating the height and the
conductor of the elliptic curve E .
CONJECTURE (Szpiro, Hall).
H  N 6+ ;
where the constant depends on .
THEOREM. Let E be a modular elliptic curve over Q satisfying the Birch–
Swinnerton Dyer conjecture, Conjecture 3, and the conjecture of Lang stated
above. Then there exists an absolute constant A such that
 
jIIIj  H 1=12 exp A logloglogNN :

comp3920.tex; 8/05/1997; 8:16; v.5; p.10

THE SIZE OF SHAFAREVICH – TATE GROUP OF ELLIPTIC CURVES 39

Proof. Let r be the order of the zero of LE (s) at s = 1. Birch and Swinnerton–
Dyer conjecture that

L(r) (1) = jIIIjR  Y  ;

r! 2r jE (Q)tors j2
1 p
p 0

where p are certain integers, R is the regulator of E and 1 is a suitably chosen

infinite period of E . Applying Cauchy’s theorem to a circle of radius 1= log log N
centred at 1, and by the estimate (17), we have
!
L(r) (1)  exp c  log N 1=2 (log log N )r+2 ;
r! log log N

where c is an absolute constant. Mazur has shown that jE Q ( )tors j 6 16. The p are
integers. For the real period Lang has shown that
1=12
j1j  H :
From Lang’s conjecture stated above, it follows using Hermite’s theorem (see [7]),
that
p
R  ( 3=2)r (log j
j)r :
2

p
This lower bound for the volume of a lattice is sharp, and neither 3=2 nor the
power r 2 can be improved upon. Thus we have

 log N
1=2 !
1=12
jIIIj  H exp c log log N
p
(2 log log N )r+2 (2= 3)r (log j
j) r : 2

As r = O((log N= log log N )1=2 ), we have

 
jIIIj  H 1=12 exp A logloglogNN
for some absolute constant A.
COROLLARY. If we further assume Szpiro’s conjecture, then

jIIIj  N (1=2)+ :
Conversely, we remark that if the above upper bound for III fails to hold, then our
arguments indicate that there should exist elliptic curves of large rank.

comp3920.tex; 8/05/1997; 8:16; v.5; p.11

40 C. S. RAJAN

p
pRemark. In view of the fact that (2= 3)r
2
occurs in the estimate for III, and
2= 3 > 1, it is of extreme importance to have the square root in the estimate (18)
(
for the rank r . An upper bound for r of the form O log N= log log N , will give )
an upper bound for III, which grows faster than any power of N .

Acknowledgement
I would like to express my sincere thanks to M. Ram Murty, for suggesting this
problem and for many useful discussions. This work was done when the author
was at McGill University, Montreal, for the year 1994–95. I would like to express
my gratitude to McGill University and my colleagues at Montreal for their warm
hospitality and support.
After this paper was written, I received a preprint ‘Bounds for the order of
the Tate–Shafarevich group’ by D. Goldfeld and L. Szpiro, which also provides
estimates for III. However since their arguments are similar to that of number fields,
their methods give upper bounds for III, which seem to grow faster than any power
of the conductor N , and do not give the expected upper bound O N 1=2+ . ( )
References
1. Brumer, A.: The average rank of elliptic curves I, Inv. Math. 109 (1992) 445–472.
2. Deligne. P.: La conjecture de Weil I, Publ. Math. IHES 43 (1974) 273–307.
3. Goldfeld, D. and Szpiro, L.: Bounds for the order of the Tate–Shafarevich group, Comp. Math.
97 (1995) 71–87.
4. Grothendieck, A.: La groupe de Brauer II, Dix Exposés sur la cohomologie des Schemas, eds.
A. Grothendieck and N. H. Kuiper, North-Holland, Amsterdam, 1968.
5. Hindry, M. and Silverman, J. S.: The canonical height and integral points on elliptic curves, Inv.
Math. 93 (1988) 419–450.
6. Lang, S.: Fundamentals of Diophantine Geometry, Springer-Verlag, New York, (1983).
7. Lang, S.: Conjectured Diophantine estimates on elliptic curves, Arithmetic and Geometry–Papers
dedicated to I. R. Shafarevich on his sixtieth birthday, Vol. I, eds. M. Artin and J. Tate, Birkhauser.
8. Mai, L. and Murty, M. R.: On the Shafarevich–Tate group, unpublished.
9. Mai, L. and Murty, M. R.: A note on quadratic twists of elliptic curves, CRM Proceedings and
Lecture Notes, Vol. 4, eds. Kisilevsky and Murty, 1994.
10. Mai, L. and Murty, M. R.: The Phragmen–Lindelof theorem and modular elliptic curves, Con-
temporary Math., vol. 166, 1994.
11. Manin, Yu.: Cyclotomic fields and modular curves, Russian Math Surveys 26 (1971) no.6, 7–78.
12. Milne, J. S.: On a conjecture of Artin and Tate, Ann of Math. 102 (1975) 517–533.
13. Pesenti, J. and Szpiro, L.: Discriminant et conducteur des courbes elliptiques non semi-stable,
preprint.
14. Raynaud, M.: Characteristique d’Euler–Poincaré et cohomologie des variétes abeliennés, Dix
Expos’es sur la cohomologie des schemas, eds. A. Grothendieck and N. H. Kuiper, North-
Holland, Amsterdam, 1968.
15. Shioda, T.: On elliptic modular surfaces, Journal of the Math. Soc. of Japan 24 (1972) 20–59.
16. Shioda, T.: On the Mordell–Weil lattices, Comment. Math. Univ. St. Pauli 39 (1990) 211–240.
17. Silverman, J. H.: The arithmetic of elliptic curves, GTM 106, Springer-Verlag, New York, 1986.
18. Silverman, J. H.: Advanced topics in the arithmetic of elliptic curves, GTM 151, Springer-Verlag,
New York, 1994.
19. Tate, J.: Algebraic cycles and poles of zeta-functions, Arithmetical Algebraic Geometry, Harper
and Row, New York (1965).

comp3920.tex; 8/05/1997; 8:16; v.5; p.12

THE SIZE OF SHAFAREVICH – TATE GROUP OF ELLIPTIC CURVES 41

20. Tate, J.: On a conjecture of Birch and Swinnerton–Dyer and a geometric analogue, Dix Exposés
sur la Cohomologie des Schemas, eds. A. Grothendieck and N. H. Kuiper, North-Holland,
Amsterdam, 1968.
21. Titchmarsh, E. C.: The theory of the Riemann zeta function, revised by D. R. Heath-Brown,
Oxford, 1986.
22. Zimmert, R.: Ideale kleiner Norm in Idealklassen und eine Regulatorabschatzung, Inv. Math. 62,
367–380, (1980).

comp3920.tex; 8/05/1997; 8:16; v.5; p.13