BULLETIN (New Series) OF THE

AMERICAN MATHEMATICAL SOCIETY

Volume 39, Number 4, Pages 455–474
S 0273-0979(02)00952-7
Article electronically published on July 8, 2002

RANKS OF ELLIPTIC CURVES

KARL RUBIN AND ALICE SILVERBERG

Abstract. This paper gives a general survey of ranks of elliptic curves over
the field of rational numbers. The rank is a measure of the size of the set of
rational points. The paper includes discussions of the Birch and Swinnerton-
Dyer Conjecture, the Parity Conjecture, ranks in families of quadratic twists,
and ways to search for elliptic curves of large rank.

Introduction
L. J. Mordell began his famous paper [49] with the words “Mathematicians have
been familiar with very few questions for so long a period with so little accomplished
in the way of general results, as that of finding the rational [points on elliptic
curves].”
The history of elliptic curves is a long one, and exciting applications for elliptic
curves continue to be discovered. Recently, important and useful applications of
elliptic curves have been found to cryptography [29], [48], for factoring large integers
[35], and for primality proving [17], [1], [18]. The mathematical theory of elliptic
curves was crucial in the proof of Fermat’s Last Theorem [76].
It is easy to find the rational points on a line. There is a well-known method
for parametrizing the rational points on a conic C in the plane: namely, if P is a
rational point on C, then every line through P intersects C in P and one other
point, and this gives a bijection between the rational points on C and the slopes
of the rational lines through P , which can be identified with the rational points on
the projective line. Thus, it is easy to find the rational points on a plane curve
defined by a linear or quadratic equation. Increasing the degree of the polynomial,
the next case to consider is that of cubics. This brings us to the case of elliptic
curves.
In this paper we give a survey of ranks of elliptic curves over the field of rational
numbers. The rank of an elliptic curve is a measure of the size of the set of rational
points. In 1901 Henri Poincaré [60] stated that the rank is obviously very important
in the classification of rational cubics. The major open questions about elliptic
curves today, including the Birch and Swinnerton-Dyer Conjecture, have to do
with the rank (see [66]).

Received by the editors January 5, 2002, and, in revised form, February 1, 2002.
2000 Mathematics Subject Classification. Primary 11G05; Secondary 11-02, 14G05, 11G40,
14H52.
The authors thank the NSF (grants DMS-9800881 and DMS-9988869), the Alexander von
Humboldt Foundation, and the Universität Erlangen-Nürnberg. Silverberg also thanks the NSA
(grant MDA904-99-1-0007), MSRI, and AIM.

c 2002 American Mathematical Society

455
456 KARL RUBIN AND ALICE SILVERBERG

We begin by giving the basic definitions about elliptic curves over the field of
rational numbers, including the definition of the rank. We discuss the Birch and
Swinnerton-Dyer Conjecture and the Parity Conjecture, and consider ranks in fam-
ilies of quadratic twists. We give lower bounds for densities of quadratic twists with
a given rank, and in the process consider ranks of elliptic curves over the function
field Q(t). We also discuss some ways to search for elliptic curves of large rank.
The authors thank B. Mazur and J.-P. Serre for helpful comments on an earlier
version of the paper.

1. Elliptic curves over Q

An elliptic curve over the field Q of rational numbers is a curve E defined by a
Weierstraß equation
(1) y 2 = x3 + ax + b
where a, b ∈ Z and
∆ := −16(4a3 + 27b2 ) 6= 0.
The condition that the discriminant ∆ be nonzero is equivalent to the curve
being smooth. It is also equivalent to the cubic x3 + ax + b having 3 different
complex roots.
We can view an elliptic curve E as a curve in projective space P2 , with homoge-
neous equation y 2 z = x3 + axz 2 + bz 3 , and one point at “infinity”, namely (0, 1, 0).
This point ∞ is the point where all vertical lines meet. Write
E(Q) = {rational solutions (x, y) of y 2 = x3 + ax + b} ∪ {∞}.
Basic Problem. Given an elliptic curve E, find all of its rational points E(Q).
Example 1.1. Let E be the elliptic curve y 2 = x3 − x. We obtain three points on
the curve by setting y = 0. It is easy to show that these are the only integer-valued
points on E. It is true, but much more difficult to show, that these are the only
rational points on E, i.e.,
E(Q) = {(0, 0), (1, 0), (−1, 0), ∞}.
This was proved by Fermat using his method of infinite descent (see §§X, XV, XVI
in Chapter II of [75]).
Over the complex numbers, a line intersects an elliptic curve in three points
(counting multiplicity), and if two of these points are rational then so is the third.
One can use this fact to define an addition law on E(Q). Namely, given P, Q ∈
E(Q), draw the line through P and Q. Let R be the third point of intersection of
that line with E, and define P + Q to be the third point of intersection of E with
the (vertical) line through R and ∞.
Figure 1 shows (among other things) the graph of the real-valued points on the
elliptic curve y 2 = x3 − x + 1, and an example of its addition law.
Concretely, if E is the elliptic curve y 2 = x3 + ax + b, and P = (x1 , y1 ), Q =
(x2 , y2 ) ∈ E(Q) with x1 6= x2 , then
 
y2 − y1 2 y −y 
2 1 y1 x2 − y2 x1
P +Q= − x1 − x2 , x3 − .
x2 − x1 x2 − x1 x2 − x1
Theorem 1.2. With the above addition law, E(Q) is a commutative group with ∞
as the identity element.
Under this operation, three collinear points on the curve sum to the identity
 RANKS OF ELLIPTIC CURVES 457

Figure 1. y 2 = x3 − x + 1 and its addition law

element. We note that proving associativity is nontrivial.

The following important result was proved by Mordell using Fermat’s method of
descent.
Theorem 1.3 (Mordell [49]). If E is an elliptic curve over Q, then the commuta-
tive group E(Q) is finitely generated.
Definition 1.4. By Mordell’s theorem we can write
E(Q) ∼ = Zr ⊕ E(Q)tors
where r is a nonnegative integer and E(Q)tors is the subgroup of elements of finite
order in E(Q). This subgroup is called the torsion subgroup of E(Q). The integer
r is called the rank of E and is written rank(E).
Example 1.5. For the curve y 2 = x3 − x of Example 1.1,
E(Q) = E(Q)tors ∼ = Z/2Z × Z/2Z, rank(E) = 0.
In other words, each of the points (0, 0), (1, 0), and (−1, 0) has order 2 in E(Q).

2. Torsion subgroups
The torsion subgroup is “well-understood”. First, there is an effective algorithm
to determine E(Q)tors given E.
Theorem 2.1 (Nagell [54], Lutz [36]). Let E be the elliptic curve y 2 = x3 + ax+ b.
If (x, y) ∈ E(Q)tors and (x, y) 6= ∞, then
(i) x, y ∈ Z,
(ii) either y = 0, or y 2 divides 4a3 + 27b2 .
Second, a deep theorem of Mazur states which finite groups can occur as torsion
subgroups of elliptic curves.
Theorem 2.2 (Mazur [39]). If E is an elliptic curve, then E(Q)tors is one of the
following 15 groups:
(i) Z/nZ, with 1 ≤ n ≤ 10 or n = 12,
(ii) Z/2mZ × Z/2Z, with 1 ≤ m ≤ 4.
Each of the groups in Theorem 2.2 occurs infinitely often as the torsion subgroup
of an elliptic curve over Q.
458 KARL RUBIN AND ALICE SILVERBERG

Example 2.3. Let E be the curve y 2 = x3 − x of Example 1.1. By the Nagell-

Lutz theorem, the nontrivial rational torsion points (x, y) have y ∈ {0, ±1, ±2}.
The only such points are (−1, 0), (0, 0), (1, 0).
Example 2.4 (See Table 3 of [34]). Suppose t ∈ Q and t 6= 0, −1. Let E be
(2) y 2 + (1 − t − t2 )xy + (t2 + t3 )y = x3 + (t2 + t3 )x2 .
Then E is an elliptic curve, (0, 0) ∈ E(Q), and one can check that
7 · (0, 0) = ∞.
By Mazur’s theorem, the subgroup generated by (0, 0) must be all of E(Q)tors , so
E(Q)tors ∼
= Z/7Z.
Conversely, one can show that every elliptic curve over Q with torsion subgroup of
order 7 is isomorphic to a curve of the form (2) for some t ∈ Q.

3. Ranks
There are no analogues of Theorems 2.1 or 2.2 for ranks:
• there is no known algorithm guaranteed to determine rank(E);
• it is not known exactly which integers can occur as the rank of an elliptic
curve.
For the first question, there are algorithms for computing both upper bounds and
lower bounds for rank(E); with luck and enough work, they might be equal.1 For
the second, it is not even known if the set of ranks of elliptic curves over Q is
bounded.
Table 1 shows, for certain r between 4 and 24, the date of publication (in print
or electronically) of an elliptic curve known to have rank at least r.

Table 1. Rank records

Rank ≥ Year Discoverers
3 1945 Billing [2]
4 1945 Wiman [77]
6 1974 Penney & Pomerance [58]
7 1975 Penney & Pomerance [59]
8 1977 Grunewald & Zimmert [21]
9 1977 Brumer & Kramer [6]
12 1982 Mestre [40]
14 1986 Mestre [41]
15 1992 Mestre [44]
17 1992 Nagao [50]
19 1992 Fermigier [13]
20 1993 Nagao [51]
21 1994 Nagao & Kouya [53]
22 1997 Fermigier [14]
23 1998 Martin & McMillen [37]
24 2000 Martin & McMillen [38]

Note: In the early 1950’s, Néron [56], [57] showed that there exist elliptic curves
with rank ≥ 11, but his proof did not yield examples.

1 See p. 193 of [70].

 RANKS OF ELLIPTIC CURVES 459

The curve in Table 1 with rank at least 24 is

y 2 + xy + y = x3 − 120039822036992245303534619191166796374x +
504224992484910670010801799168082726759443756222911415116.

(Note that this is not exactly in the form (1), but it can be put in that form by a
simple change of variables, at the expense of increasing the size of the coefficients.
See (3) below.) The rank is “at least” 24 because Martin and McMillen exhibited
24 independent points in E(Q), but it has not been proved that the rank is exactly
24.
Many of the ideas for finding elliptic curves of high rank are due to Mestre. See
§9 below.

4. Elliptic curves over arbitrary fields

To fully understand elliptic curves over Q it is helpful to study elliptic curves
over finite fields (see §5) and over function fields (see §8 and §9).
If F is a field, an elliptic curve over F is a nonsingular curve defined by a
generalized Weierstraß equation
(3) y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6

with ai ∈ F . (Compare with (1). If the characteristic of F is not 2, then we can

complete the square in y and change variables to make a1 = a3 = 0; similarly if
the characteristic is not 3, we can suppose that a2 = 0.) Such a curve always has
a single point at infinity in projective space P2 (F ).
Conversely, one can show that every nonsingular plane cubic with coefficients in
F that has a point in P2 (F ) has an equation of the form (3).
As when F = Q, the set E(F ) of F -points (including the point at infinity) is an
abelian group under the geometric composition law described in §1. For example,
the theory of elliptic functions shows that if E is an elliptic curve defined over the
complex numbers C, then there are a lattice L ⊂ C and an analytic group isomor-
phism E(C) ∼ = C/L. Thus the group E(C) is not finitely generated. However, for
certain fields one does have an analogue of Mordell’s Theorem (Theorem 1.3):
Theorem 4.1 (Néron [56]). If K is either Q or a finite field, F is a finitely gen-
erated extension of K, and E is an elliptic curve defined over F , then the group
E(F ) is finitely generated.

5. The Birch and Swinnerton-Dyer Conjecture

Fix an elliptic curve E : y 2 = x3 + ax + b over Q. For every prime number p not
dividing the discriminant ∆ = 16(4a3 + 27b2) of E, we can reduce a and b modulo p
and view E as an elliptic curve over the finite field Fp . Reduction modulo p induces
a group homomorphism
E(Q) −→ E(Fp ).

The idea of Birch and Swinnerton-Dyer was that the larger E(Q) is, the larger
the E(Fp )’s should be “on average” as p varies. The size of E(Q) can be measured
by rank(E), but how can one measure the average size of the E(Fp )’s?
460 KARL RUBIN AND ALICE SILVERBERG

Definition 5.1. For every prime number p not dividing ∆ let

Np = #E(Fp )

= 1 + # 0 ≤ x, y ≤ p − 1 : y 2 ≡ x3 + ax + b (mod p) .
Theorem 5.2 (Hasse [22], [23]). For every prime p not dividing ∆,
√ √
p + 1 − 2 p < Np < p + 1 + 2 p.
To test their idea, in the 1950’s Birch and Swinnerton-Dyer computed
Y Np
(4) πE (X) :=
p
p≤X,p-∆

as X grows, for certain elliptic curves E.

Figure 2 shows the behavior of πEd (X) for X up to about 1.5 × 107 for five
different curves Ed : y 2 = x3 − d2 x (using the first five values of d in Table 2 of §6,
so these curves have ranks 0, 1, 2, 3, and 4). The horizontal axis is log log(X) and
the vertical axis is log(πEd (X)).
From their data Birch and Swinnerton-Dyer [3] were led to conjecture that
(5) πE (X) ∼ C(log(X))rank(E)
as X → ∞ for some constant C depending only on E. (Note that this relation is
consistent with the data in Figure 2 — if the axes were to scale, then the slopes of
the lines would be the ranks of the curves.) The function πE does not behave very
nicely and therefore is difficult to work with. Birch and Swinnerton-Dyer stated a
related conjecture, using the L-function of E in place of πE .

Figure 2. Birch and Swinnerton-Dyer data for y 2 = x3 − d2 x

 RANKS OF ELLIPTIC CURVES 461

Definition 5.3. Define the Hasse-Weil L-function of E, a function of a complex

variable s, by
Y 1 + p − Np p −1 Y
(6) L(E, s) = 1− s
+ 2s
× `p (E, s)−1
p p
p-∆ p|∆

where `p (E, s) is a certain polynomial in p−s with the property that `p (E, 1) 6= 0
(see for example p. 196 of [70]).
It follows from Theorem 5.2 that L(E, s) converges absolutely and uniformly on
compact subsets of the complex half-plane {s ∈ C : Re(s) > 3/2}. The Shimura-
Taniyama Conjecture, recently proved by Breuil, Conrad, Diamond, and Taylor [5]
by extending work of Wiles [76], implies the following long-standing conjecture of
Hasse and Weil.
Theorem 5.4 ([76], [72], [5]). L(E, s) has an analytic continuation to all of C and
satisfies a functional equation
Λ(s) = wE Λ(2 − s)

where wE = ±1 and Λ(s) = N s/2 (2π)−s Γ(s)L(E, s) for some positive integer N
(depending on E).
See for example p. 196 of [70] for a definition of the conductor N of E.
While the Euler product (6) for L(E, s) may not in general converge at s = 1,
purely formally evaluating (6) at s = 1 gives
Y N Y −1
p
(7) L(E, 1) “=” × `p (E, 1) .
p
p-∆ p|∆

Thus, since there are only a finite number of terms in the second product, we can
hope that the behavior of L(E, s) near s = 1 will reflect the average size of the
Np : the larger the Np are, the faster L(E, s) will tend to 0 as s tends to 1. The
following quantitative version of this statement is part of the conjecture of Birch
and Swinnerton-Dyer.
Conjecture 5.5 (Birch and Swinnerton-Dyer [4]). For every elliptic curve E,
rank(E) = ords=1 L(E, s).
Goldfeld proved the following surprising result, which says in particular that the
“purely
√ formal” connection between πE (X) and L(E, s) in (7) is off by a factor of
2.
Theorem 5.6 (Goldfeld [16]). Suppose that πE (X) ∼ C(log(X))r with constants
C ∈ R+ and r ∈ R. Then r = ords=1 L(E, s) and
L(E, s) √ Y
lim = 2erγ C −1 `p (E, 1)−1
s→1 (s − 1)r
p|∆

where γ is Euler’s constant. In particular, if r = 0 then

√  Y Np Y −1
L(E, 1) = 2 × `p (E, 1) .
p
p-∆ p|∆
462 KARL RUBIN AND ALICE SILVERBERG

The lines log(C)+r log log(X) in Figure 2 were calculated using (5), Theorem 5.6,
and the full Birch and Swinnerton-Dyer Conjecture to determine C and r. (Birch
and Swinnerton-Dyer predicted not only the order of vanishing of L(E, s) at s = 1,
but also the first nonvanishing coefficient of its Taylor expansion about s = 1.)
Definition 5.7. With the Birch and Swinnerton-Dyer Conjecture in mind, call the
order of vanishing of L(E, s) at s = 1 the analytic rank of E and write
rankan (E) := ords=1 L(E, s).
The following theorem, a combination of work of Kolyvagin [31], [32], Gross and
Zagier [20], and others, is the best result to date in the direction of the Birch and
Swinnerton-Dyer Conjecture.
Theorem 5.8 ([31], [32], [20]). (i) rankan (E) = 0 ⇒ rank(E) = 0,
(ii) rankan (E) = 1 ⇒ rank(E) = 1.
Assertion (i) can be rephrased as “L(E, 1) 6= 0 ⇒ E(Q) is finite”. The case
rankan(E) ≥ 2, except for isolated examples, remains completely open.
There are elliptic curves that can be proved to have analytic ranks 0, 1, 2, and
3 (see [20]). There is no elliptic curve that has been proved to have analytic rank
greater than 3.
Example 5.9. If E is the curve y 2 = x3 − x of Example 1.1, then
L(E, 1) = 0.65551438857302995 . . . 6= 0.
Thus Theorem 5.8(i) shows that (as Fermat said) E(Q) is finite.
The sign wE in the functional equation (Theorem 5.4) for L(E, s) determines
the parity of rankan (E):
(
even if wE = +1,
rankan (E) is
odd if wE = −1.
The Birch and Swinnerton-Dyer Conjecture predicts in particular that rank(E) and
rankan(E) have the same parity, so the following is a consequence of the Birch and
Swinnerton-Dyer Conjecture.
Conjecture 5.10 (Parity Conjecture).
(
even if wE = +1,
rank(E) is
odd if wE = −1.
To describe recent progress concerning the Parity Conjecture, we need to intro-
duce the Tate-Shafarevich group and the Selmer group. For definitions see pp. 238–
239 of [8] or Definitions 4.6.8 and 4.8.1 of [66]. The Tate-Shafarevich group XE is
a torsion group that measures the failure of the Hasse Principle for curves that are
principal homogeneous spaces for E.
Conjecture 5.11. (Tate-Shafarevich Conjecture).2 XE is finite.
2 In his article in the proceedings of the 1962 ICM ([8], pp. 239–240), Cassels writes, “Indeed
Tate and Šafarevič have, I believe, independently conjectured that X itself is always finite, al-
though, so far as I know, it has not been completely determined in any individual case.” In a
footnote he adds, “In his lecture Tate denied paternity but adopted the conjecture. In conversation
during the Congress Šafarevič expressed strong doubts.”
 RANKS OF ELLIPTIC CURVES 463

The proof of Theorem 5.8 also proves the following theorem.

Theorem 5.12 ([31], [32], [20]). If rankan (E) ≤ 1, then XE is finite.
There is no example known of an elliptic curve with rankan (E) > 1 for which
XE has been proved to be finite.
Although there is no known general algorithm guaranteed to determine E(Q)
or XE , there are effectively computable groups, known as Selmer groups, which
combine information about both E(Q) and XE . More precisely, for every natural
number m, the m-Selmer group Sm (E) is a finite group of exponent dividing m
that sits in an exact sequence
(8) 0 → E(Q)/mE(Q) → Sm (E) → X[m] → 0
E

where X[m] denotes the kernel of multiplication by m in an abelian group X.

Thus the Tate-Shafarevich group can also be viewed as an obstruction to effectively
computing the rank of E: if p is a prime, then (8) and Theorem 1.3 show that
(9) dimFp Sp (E) = rank(E) + dimFp E(Q)[p] + dimFp X[p].
E
In practice the only way to prove upper bounds for the rank of E has been to
prove upper bounds for #Sm (E). For example, Theorems 5.8 and 5.12 follow from
the statements
(i) rankan (E) ≤ 1 ⇒ rank(E) ≥ rankan (E);
(ii) rankan (E) ≤ 1 ⇒ #Sm (E) ≤ Cmrankan (E) , with a constant C independent of
m.
The first assertion is trivial if rankan (E) = 0 and was proved by Gross and Zagier
[20] when rankan(E) = 1 by constructing a point of infinite order (a Heegner point).
The second assertion uses Kolyvagin’s method of Euler systems and an infinite
family of Heegner points. Combining these two statements with (8) proves that if
rankan(E) ≤ 1, then rank(E) = rankan (E) and # XE ≤ C.
Theorem 5.13 (Nekovář [55]). If XE is finite, then the Parity Conjecture holds
for E.
What Nekovář proved (using recent results of Vatsal [74] and Cornut [10]) is that
if p is a prime not dividing #E(Q)tors , and if E has good ordinary reduction at p
(see Chapters V and VII of [68]), then dimFp Sp (E) and rankan (E) have the same
parity. But if XE is finite, then the Cassels pairing [7] is a nondegenerate skew-
symmetric pairing on XE , and it follows that dimFp XE is even. Hence by (9),
rank(E) and dimFp Sp (E) have the same parity. Every elliptic curve has infinitely
many primes of good ordinary reduction, so the Parity Conjecture for E follows.
It had been proved earlier that if the Tate-Shafarevich Conjecture is true, then
the Parity Conjecture holds for semistable elliptic curves (combining [33] and The-
orem 5.8) and for the curves y 2 = x3 − d2 x (see Monsky’s appendix to [24]).

6. Quadratic twists
Up until now, we have been considering ranks of arbitrary elliptic curves over
Q. To understand ranks, it is useful to consider special families of elliptic curves.
Quadratic twists give perhaps the simplest such families, since even though their
complex analysis is “constant” (i.e., they are isomorphic over C), their arithmetic
varies.
464 KARL RUBIN AND ALICE SILVERBERG

Table 2. Ranks rd in the family Ed : dy 2 = x3 − x

d rd discovery x-coordinates of independent points

1 0 Fermat (∼1640)
5 1 Billing [2] (1937) 9
34 2 Wiman [78] (1945) 17, 178
1254 3 Wiman [78] (1945) 118 , 22 19
3 , 8
41 24 34 121
29274 4 Wiman [78] (1945) 34 , 17 , 7 , 2
205015206 5 Rogers [61] (2000) 649 1650 326 19234 5783298
323 , 1121 , 323 , 8993 , 2468041
61471349610 6 Rogers [61] (2000) 779 52441 228001 21033 56416 4427538
134 , 31691 , 931 , 10658 , 32761 , 2255

Definition 6.1. If E is given by y 2 = x3 + ax + b with a, b ∈ Q, then the quadratic

twist of E by a nonzero rational number d is the elliptic curve y 2 = x3 + ad2 x + bd3 .
It will be convenient to make the change of variables (x, y) 7→ (dx, d2 y), so that we
can rewrite this curve in the equivalent (isomorphic) form
Ed : dy 2 = x3 + ax + b.
In this section we will study the behavior of rank(Ed ) as d varies. Clearly Edt2 is
isomorphic to Ed for every t ∈ Q× , so we need only consider squarefree integers d.
6.1. The curve y 2 = x3 − x. For the remainder of this section, let E be the curve
y 2 = x3 − x. The family Ed : dy 2 = x3 − x of quadratic twists of E has been
studied extensively. This family is closely connected with the classical congruent
number problem, which asks what integers are the areas of right triangles with
three rational sides. The relationship between this problem and the above family
of quadratic twists is the fact that there is a right triangle with rational sides and
area d if and only if rank(Ed ) > 0 (see for example [30], [63], [73]).
Note that Ed is isomorphic to E−d by the change of variables (x, y) 7→ (−x, y),
so we may restrict to d > 0.
The curve E157 has rank one, but the simplest point of infinite order (see p. 5 of
[30]) is
 
277487787329244632169121 22826630568289716631287654159126420
− , .
609760250665615167250729 476144382506163554005382044222449067
For r ≤ 4, Table 2 gives the smallest |d| for which Ed has rank r. For r = 5, 6
there are probably smaller examples of d than the one listed in the table.
In [78], Wiman doubted whether any other d’s with rank(Ed ) = 4 could be found.
Using the method of proof of Theorem 8.2(vii) below and modern computers, it is
no longer difficult to find such examples. In [79], Wiman pointed out that he knew
of no d for which rank(Ed ) > 4 and said that if such exist, they would be almost
insurmountably difficult to find.
If d is squarefree, an upper bound on the rank of Ed is given by twice the number
of odd prime divisors of d (see [78]), and there is an absolute constant C such that
log |d|
rank(Ed ) ≤ C
log log |d|
for all squarefree d with |d| > 2 (see Exercise 3.4.11 of [66]). This is known as
the “trivial bound” for the rank of Ed . It follows that for Ed to have large rank,
 RANKS OF ELLIPTIC CURVES 465

d must have many prime divisors. For example, the last d in Table 2 has prime
factorization
61471349610 = 2 · 3 · 5 · 11 · 19 · 41 · 43 · 67 · 83.

Theorem 6.2 (Tunnell [73]). If E is y 2 = x3 − x and d is a squarefree positive

integer, then
(n − 2m)2 aΩ
L(Ed , 1) = √
16 d
where
a = 1 if d is odd, a = 2 if d is even,
n = #{(x, y, z) ∈ Z3 : x2 + 2ay 2 + 8z 2 = d/a},
m = Z#{(x, y, z) ∈ Z3 : x2 + 2ay 2 + 32z 2 = d/a},
∞
dx
Ω= √ ≈ 2.6220575542921198 . . ..
x 3−x
1
In particular,
L(Ed , 1) = 0 ⇐⇒ n = 2m.
For example, if d = 1, then m = n = 2 and L(E, 1) = Ω/4 as in Example 5.9.

7. Variation of the rank in families of quadratic twists

Fix for this section an elliptic curve E over Q. We will study how the ranks of
quadratic twists are distributed.
Definition 7.1. Let
S(X) = {squarefree d ∈ Z : |d| ≤ X}.
Define the average rank A(E) to be
P
d∈S(X) rank(Ed )
(10) A(E) = lim
X→∞ #S(X)
if this limit exists, and in general define the upper and lower average ranks, A(E)
and A(E), to be, respectively, the corresponding lim sup and lim inf.
Define
N∗ (X) = #{d ∈ S(X) : rank(Ed ) is ∗}
where ∗ can be any property that makes sense, such as “1”, “≥ 2”, “even”, etc. We
write simply N (X) = N≥0 (X) = #S(X).
Let
N∗ (X)
D∗ (E) = lim
X→∞ N (X)

if this limit exists, and in general let D∗ (E) and D∗ (E) be the corresponding lim sup
and lim inf.
The next theorem, which is well-known (see for example the corollary to Propo-
sition 10 of [62]), describes how the sign in the functional equation of the L-function
changes under quadratic twist. If t is a squarefree√ integer, let χt be the quadratic
Dirichlet character attached to the extension Q( t)/Q. Concretely, χt is the unique
466 KARL RUBIN AND ALICE SILVERBERG

Dirichlet character modulo t (if t ≡ 1 (mod 4)) or 4t (if t ≡ 2, 3 (mod 4)) with the
property that for all odd primes p not dividing t,
(
+1 if t is a square modulo p,
χt (p) =
−1 if t is not a square modulo p.
Recall that the conductor of an elliptic curve is defined on p. 196 of [70].
Theorem 7.2. Suppose that d is a squarefree integer, and let Nd be the conductor
of Ed . If t ≡ 1 (mod 4) is a squarefree integer relatively prime to dN1 , then
wEtd /wEd = χt (−Nd ).
Example 7.3. Let E be the curve y 2 = x3 − x. It follows from Theorem 6.2 that
L(E, 1) 6= 0 and L(E2 , 1) 6= 0, so wE = wE2 = 1. Combining this with Theorem 7.2
and the fact that the conductors of E and E2 are 32 and 64, respectively, one can
show that the sign in the functional equation of L(Ed , 1) for d > 0 is given by
(
+1 if d ≡ 1, 2, or 3 (mod 8),
wEd =
−1 if d ≡ 5, 6, or 7 (mod 8).
(Since d is squarefree, it is not 0 or 4 (mod 8).) In particular, if d > 0 and d ≡ 5, 6,
or 7 (mod 8), then the Parity Conjecture predicts that rank(Ed ) is odd. Elkies [11],
[12] has verified that rank(Ed ) ≥ 1 for all positive squarefree d ≡ 5, 6, or 7 (mod 8)
less than 106 . Note that for d ≡ 5, 6, or 7 (mod 8) one can also use Theorem 6.2
to show that L(Ed , 1) = 0, since n = m = 0 in those cases.
Corollary 7.4. Suppose that the Parity Conjecture holds. Then
Deven (E) = Dodd (E) = 1/2 and A(E) ≥ 1/2.
Proof. This follows from Theorem 7.2 applied to the curves Ed for (positive or
negative) d dividing twice the conductor of E.
Conjecture 7.5 (Goldfeld [15]). A(E) = 1/2.
In other words, Goldfeld’s conjecture predicts that the average rank is as small
as the Parity Conjecture allows. The following conjecture is an easy consequence
of Goldfeld’s conjecture combined with the Parity Conjecture and Corollary 7.4.
Conjecture 7.6 (Density Conjecture). D0 (E) = D1 (E) = 1/2 and D≥2 (E) = 0.
Note that N (X) ∼ 2
ζ(2) X = 12
π 2 X. The Density Conjecture would imply that
6
N0 (X) ∼ N1 (X) ∼ X, N≥2 (X) = o(X).
π2
Given the Birch and Swinnerton-Dyer Conjecture, the Density Conjecture can
be interpreted as saying that the set where the L-function has “extra vanishing”,
that is, the set of d for which the value of rankan (Ed ) is larger than the functional
equation forces it to be, has density zero. For some recent motivation for the
Density Conjecture from this point of view, see §5 of [26].
When r ≥ 2 the Density Conjecture predicts that Nr (X) = o(X), and one
can ask for a more precise description of the rate of growth of Nr (X). Numer-
ical evidence suggests that N≥2,even(X) and N≥3,odd(X) grow roughly like X 3/4
(see Figure 2 of [9] and see [12], respectively). The following conjecture, based
on connections between L-functions and random matrix theory, makes this more
precise.
 RANKS OF ELLIPTIC CURVES 467

Conjecture 7.7 (Conjecture 1 and (7) of [9]). There are constants bE and eE ,
with bE 6= 0, such that
N≥2,even(X)
lim = bE .
X→∞ X 3/4 log(X)eE

See §8 for some lower bounds for N≥r (X).

Heath-Brown showed that if E is y 2 = x3 − x and one restricts to twists by
odd integers d, then the density of twists with rank at least r goes to zero at
least exponentially with r. From this one can deduce an upper bound for the
average rank and lower bounds for the densities Dr (E) for small values of r. Let
S odd (X) = {odd squarefree d ∈ Z+ : d ≤ X}, define Aodd (E) as in (10) but with
S odd (X) in place of S(X), and similarly write Dodd
∗ (E) for the corresponding density
restricted to odd d.
Theorem 7.8 (Heath-Brown [24]). Let E be the curve y 2 = x3 − x. Then:
odd
(i) A (E) ≤ 1.2645;
odd
for every r ≥ 0, D≥r (E) ≤ 1.7313 · 2−(r −r)/2 ;
2
(ii)
(iii) D0 (E) > .044;
(iv) if the Parity Conjecture holds, then D1 (E) > .26.
Proof. Assertions (i) and (ii) are Corollaries 4 and 3 of [24], respectively. In fact,
Heath-Brown proves more. As before, we may restrict to d > 0. Let s2 (d) =
dimF2 S2 (Ed ) (recall the 2-Selmer group S2 from §5). Since Ed (Q)tors contains
(Z/2Z)2 , (9) shows that
(11) s2 (d) ≥ rank(Ed ) + 2.
Monsky proved in an appendix to [24] that
(
0 (mod 2) if d ≡ 1 or 3 (mod 8),
(12) s2 (d) ≡
1 (mod 2) if d ≡ 5 or 7 (mod 8).

Let SD∗ (h) (resp., SD∗ (h)) denote the upper (resp., lower) density of d ≡ h
(mod 8) such that s2 (d) − 2 is ∗. Corollary 3 of [24] proves that
SD≥r (h) ≤ 1.7313 · 2−(r −r)/2
2

for every r and every odd h. Taking r = 2 and h = 1 or 3, one finds that SD≤1 (1) ≥
.134 and SD≤1 (3) ≥ .134. But (11) and (12) show that if d ≡ 1 or 3 (mod 8) and
s2 (d) ≤ 3, then s2 (d) = 2 and rank(Ed ) = 0. Thus D0 (E) ≥ .044, which proves (iii).
The proof of (iv) is similar, taking r = 3 and h = 5 or 7. In this case (11) and (12)
show that if d ≡ 5 or 7 (mod 8) and s2 (d) ≤ 4, then s2 (d) = 3 and rank(Ed ) ≤ 1.
If the Parity Conjecture holds, then rank(Ed ) is odd, so rank(Ed ) = 1, and (iv)
follows.

In 1960, Honda stated a controversial conjecture that would imply:

Conjecture 7.9 (Honda [25]). Suppose E is an elliptic curve over Q. Then there
is a constant CE depending only on E such that for all d,
rank(Ed (Q)) ≤ CE .
In other words, for all sufficiently large r, Nr (X) = 0 for all X.
468 KARL RUBIN AND ALICE SILVERBERG

Instead of looking at elliptic curves over Q and twisting by elements of Q, one

could consider an elliptic curve over a field K and twist by elements of K. In [71],
Shafarevich and Tate constructed a family of quadratic twists with unbounded
rank for an elliptic curve over the function field Fq (t), where Fq is the field with q
elements. Their result led many to believe that a similar phenomenon should hold
for elliptic curves over Q, i.e., that Honda’s conjecture should be false.

8. Lower bounds for N≥r (X)

In this section we show how to obtain lower bounds for the number of quadratic
twists with rank at least r, for small values of r.
Definition 8.1. For every t ∈ Q× , there Qisnap unique squarefree integer sf(t) such
×
that t = sf(t)y
Q
2
with y ∈ Q . (If t = ± p is the prime factorization of t, then
sf(t) = ± np odd p.)
The following theorem is a combination of results of Gouvêa and Mazur [19],
Stewart and Top [69], and the authors [65] (see also [46] for related results). (In
this theorem, E is always an elliptic curve over Q, and A(X)  B(X) means that
there exists a constant C depending only on E such that A(X) ≥ CB(X) for all
sufficiently large X.)
Theorem 8.2. (i) N≥1 (X)  X 1/2 .
(ii) If E is y 2 = x3 + ax + b and ab 6= 0, then N≥2 (X)  X 1/7 / log2 (X).
(iii) If E is y 2 = x3 + ax + b, and x3 + ax + b has a nonzero rational root, then
N≥2 (X)  X 1/3 .
(iv) If E is y 2 = x(x − f )(x − c2 f ) or y 2 = x(x − f )(x + 2c2 f ) with c, f ∈ Q, or
E is y 2 = x3 − x, then N≥3 (X)  X 1/6 .
Suppose now that the Parity Conjecture holds for all the quadratic twists of E.
(v) N≥2 (X)  X 1/2 .
(vi) If E is y 2 = x3 + ax + b where x3 + ax + b has three real roots and either the
largest or smallest of these roots is rational, then N≥3 (X)  X 1/3 .
(vii) If E is y 2 = x(x − f )(x − c2 f ) with c, f ∈ Q, or E is y 2 = x3 − x, then
N≥4 (X)  X 1/6 .
Sketch of proof. Write E : y 2 = f (x) with a cubic polynomial f (x) ∈ Q[x]. Suppose
that g(t) ∈ Q[t] is a squarefree polynomial, and consider
Eg(t) : g(t)y 2 = f (x).
This is an elliptic curve defined over the rational function field Q(t). Let r =
rank(Eg(t) ), the rank of the finitely generated abelian group Eg(t) (Q(t)), and let
P1 (t), . . . , Pr (t) be r independent points in Eg(t) (Q(t)).
If t0 ∈ Q is not a root of g(t), nor of the denominators of the coordinates of the
Pi (t), then Eg(t0 ) is a quadratic twist of E defined over Q and P1 (t0 ), . . . , Pr (t0 ) ∈
Eg(t0 ) (Q). Theorem C of [67] shows that these points are independent for all
t0 outside a finite exceptional set Σ. Since Eg(t0 ) ∼ = Esf(g(t0 )) , we conclude that
rank(Esf(g(t0 )) ) ≥ r for all t0 ∈ Q − Σ. In other words, N≥r (X) ≥ #M (X), where
M (X) := {d ∈ S(X) : d = sf(g(t0 )) for some t0 ∈ Q − Σ}.
Results of Gouvêa and Mazur [19], improved by Stewart and Top [69], give a lower
bound for #M (X). (The smaller the degree of g(t), the larger the lower bound.)
 RANKS OF ELLIPTIC CURVES 469

For example, when g(t) = f (t), rank(Eg(t) ) = 1 (we can take P1 (t) = (t, 1)).
The above argument, applied to this example, was used by Gouvêa and Mazur [19]
to prove a slightly weaker form of (i). (As stated here, (i) uses improved bounds of
Stewart and Top.) Assertion (ii) was proved by Stewart and Top using a polynomial
g(t) of degree 14 constructed by Mestre [45]. Assertions (iii) and (iv) are proved in
[65] by finding ways to construct suitable polynomials g(t).
For example, if E is y 2 = x3 − x, and g(t) = 6(t3 − 33t2 − 33t + 1), then
rank(Eg(t) ) = 1, rank(Eg(t2 ) ) = 2, rank(Eg(t4 ) ) = 3.
Three independent points of infinite order on Eg(t4 ) are
 4   4 
t − 6t2 + 1 2 t + 6t2 + 1 2
P1 (t) = − , , P2 (t) = − , ,
3(t2 + 1)2 9(t2 + 1)3 3(t2 − 1)2 9(t2 − 1)3
 4 
t +1 1
P3 (t) = , .
6t2 36t3
Let
M 0 (X) = {d ∈ M (X) : wEd = (−1)r+1 }.
If d ∈ M 0 (X), then rank(Ed ) ≥ r. But assuming the Parity Conjecture, rank(Ed ) ≡
r + 1 (mod 2), so rank(Ed ) ≥ r + 1 and N≥r+1 (X) ≥ #M 0 (X). Under additional
conditions on g(t), one can obtain a lower bound for #M 0 (X). This idea was used
by Gouvêa and Mazur [19] to prove a slightly weaker version of (v). Applying this
idea to some of the polynomials used to prove (i), (iii), and (iv) gives (v), (vi), and
(vii).
See [65] for additional families of curves for which the conclusions of (iv) and
(vii) of Theorem 8.2 hold.
Remark 8.3. Conjectures 7.6 and 7.7 and the numerical evidence in [9] and [12]
suggest that N≥r (X) should grow roughly like X, X, X 3/4 , and X 3/4 for r = 0, 1,
2, and 3, respectively. The lower bounds of Theorem 8.2 are consistent with, but
weaker than, these predictions.

9. Looking for large ranks

The standard method for finding elliptic curves of large rank is due to Mestre
[41], [44]. We describe it here briefly.
Suppose E (t) is an elliptic curve over Q(t) with r independent points. (See
Table 3 for examples with large rank.) As in the proof of Theorem 8.2, specializing
gives, for all but finitely many rational numbers t0 , elliptic curves E (t0 ) over Q of
rank at least r. One would now like to search among these specializations for some
that have even larger rank.

Table 3. Rank records over Q(t)

Rank over Q(t) ≥ Year Discoverers

11 1991 Mestre [42]
12 1991 Mestre [43]
13 1994 Nagao [52]
14 2000–1 Mestre [47], Kihara [28]
470 KARL RUBIN AND ALICE SILVERBERG

To do this, choose a pair of parameters n, m ∈ Z+ . For positive integers t0 ≤ n,

compute the Birch and Swinnerton-Dyer product πE (t0 ) (m) defined by (4). The
Birch and Swinnerton-Dyer philosophy says that those t0 for which this value is
relatively large are good candidates for having “extra” rank, and one searches for
points on those curves. Hopefully one finds (several) new points, independent of the
r specialized points. Modifications of this program (and more and more computing
power) led to all the examples in Table 1 with rank at least 15.
We now describe a method for finding curves of large rank in a fixed family of
quadratic twists, in the spirit of Theorem 8.2.
Fix an elliptic curve E : y 2 = f (x) over Q, and let Ed denote its quadratic twist
dy = f (x) for d ∈ Q× , as in §6. In [19] (see the proof of Theorem 8.2), Gouvêa
2

and Mazur count how many d’s occur as sf(f (t)) for some rational t. Instead, we
will count how often each d occurs.
Definition 9.1. If t ∈ Q, define the height of t
h(t) = max{log |u|, log |v|}
where t = u/v with relatively prime integers u, v.
For B > 0 let
M (d, B) = #{t ∈ Q : h(t) < B, sf(f (t)) = d}.
The next proposition follows easily from basic facts about heights on elliptic
curves (see for example the proposition in §2 of [80]).
Proposition 9.2. For every squarefree integer d,
M (d, B)
lim
B→∞ B rank(Ed )/2
exists and is positive.
In particular if rank(Ed ) > rank(Ed0 ), then for all sufficiently large B we have
M (d, B) > M (d0 , B).
This suggests a computational method for searching for curves Ed with large
rank:
• Let t run through all rational numbers with h(t) < B and make a table of the
values M (d, B).
• Pick out those d for which M (d, B) is large, and compute rank(Ed ).
Rogers [61] implemented this method for the curve E : y 2 = x3 − x and found the
large examples in Table 2: rank(E205015206 ) = 5, rank(E61471349610 ) = 6.
Proposition 9.2 also suggests a method for testing the entire family of curves
Ed at once for curves of large rank. Although the method works generally [64], to
illustrate it we restrict to the curve y 2 = x3 − x.
Define
X
S(j, k) = |sf(x3 − x)|−k h(x)−j .
x∈Q−{0,±1}

If a, b, c, d ∈ Z+ , let ωa,b,c,d ∈ Z2 be a shortest nonzero vector in the lattice

{(u, v) ∈ Z2 : a2 | u, b2 | v, c2 | (u + v), d2 | (u − v)}
 RANKS OF ELLIPTIC CURVES 471

and define
∞ 0
X (abcd)2k
Q(j, k) =
kωa,b,c,dk4k h(ωa,b,c,d )j
a,b,c,d=1

where the sum is over a, b, c, d such that, if ωa,b,c,d = (u, v), then u and v are
relatively prime and uv(u + v)(u − v) 6= 0.
Theorem 9.3 ([64]). If j is a positive real number, then the following are equiva-
lent:
(i) rank(Ed ) < 2j for every d ∈ Z+ ,
(ii) S(j, k) converges for some k ≥ 1,
(iii) S(j, k) converges for every k ≥ 1,
(iv) Q(j, k) converges for some k ≥ 1,
(v) Q(j, k) converges for every k ≥ 1.
p
Idea of proof. If x ∈ Q − {0, ±1} and d = sf(x3 − x), then (x, ± (x3 − x)/d) ∈
Ed (Q). Using this we can rewrite
1 X X
S(j, k) = |d|−k h(x(P ))−j ,
2
d squarefree P ∈Ed (Q)

P d ) ≥ 2j for some d, then it

where x(P ) denotes the x-coordinate of P . If rank(E
follows from Proposition 9.2 that the inner sum P ∈Ed (Q) h(x(P ))−j diverges, so
S(j, k) diverges.
But if rank(Ed ) < 2j for every d, then j > 1 (see Table 2), and one can show
that
X
h(x(P ))−j  log(|d|)−j ,
P ∈Ed (Q)

so S(j, k) converges. It follows that (i), (ii), and (iii) are equivalent.
Further, one can compare Q(j, k) and S(j, k) directly to show that
Q(j, k) converges ⇐⇒ S(j, k) converges
so (ii) is equivalent to (iv) and (iii) is equivalent to (v).
By Theorem 9.3, unboundedness of ranks in the family of quadratic twists of E
is equivalent to the divergence of S(j, k) (or Q(j, k)) for all j > 0 and k ≥ 1. Our
experimental evidence indicates that such divergence would be very slow.

References
[1] A. O. L. Atkin, F. Morain, Elliptic curves and primality proving, Math. Comp. 61 (1993),
no. 203, 29–68. MR 93m:11136
[2] G. Billing, Beiträge zur arithmetischen Theorie der ebenen kubischen Kurven vom Ge-
schlechte eins, Nova Acta Reg. Soc. Sc. Upsaliensis (4) 11 (1937), No. 1.
[3] B. Birch, H. P. F. Swinnerton-Dyer, Notes on elliptic curves. I, J. Reine Angew. Math. 212
(1963), 7–25. MR 26:3669
[4] , Notes on elliptic curves. II, J. Reine Angew. Math. 218 (1965), 79–108. MR 31:3419
[5] C. Breuil, B. Conrad, F. Diamond, R. Taylor, On the modularity of elliptic curves over Q:
wild 3-adic exercises, J. Amer. Math. Soc. 14 (2001), no. 4, 843–939. MR 2002d:11058
[6] A. Brumer, K. Kramer, The rank of elliptic curves, Duke Math. J. 44 (1977), no. 4, 715–743.
MR 56:15658
[7] J. W. S. Cassels, Arithmetic on curves of genus 1. IV. Proof of the Hauptvermutung, J. Reine
Angew. Math. 211 (1962), 95–112. MR 29:1214
472 KARL RUBIN AND ALICE SILVERBERG

[8] , Arithmetic on an elliptic curve, in Proc. Internat. Congr. Mathematicians (Stock-

holm, 1962), Inst. Mittag-Leffler (1963), 234–246. MR 31:167
[9] J. B. Conrey, J. P. Keating, M. O. Rubinstein, N. C. Snaith, On the frequency of vanishing
of quadratic twists of modular L-functions, preprint.
[10] C. Cornut, Mazur’s conjecture on higher Heegner points, Invent. Math. 148 (2002), 495–523.
[11] N. Elkies, Heegner point computations, in Algorithmic Number Theory (ANTS-1), Lect.
Notes in Comp. Sci. 877, Springer-Verlag, Berlin (1994), 122–133. MR 96f:11080
[12] , http://www.math.harvard.edu/∼elkies/compnt.html.
[13] S. Fermigier, Un exemple de courbe elliptique définie sur Q de rang ≥ 19, C. R. Acad. Sci.
Paris Sér. I Math. 315 (1992), no. 6, 719–722. MR 93i:11067
[14] , Une courbe elliptique définie sur Q de rang ≥ 22, Acta Arith. 82 (1997), no. 4,
359–363. MR 98j:11041
[15] D. Goldfeld, Conjectures on elliptic curves over quadratic fields, in Number theory, Car-
bondale 1979 (Proc. Southern Illinois Conf., Southern Illinois Univ., Carbondale, Ill., 1979),
M. B. Nathanson, ed., Lect. Notes in Math. 751, Springer-Verlag, Berlin (1979), 108–118.
MR 81i:12014
[16] , Sur les produits partiels eulériens attachés aux courbes elliptiques, C. R. Acad. Sci.
Paris Sér. I Math. 294 (1982), no. 14, 471–474. MR 84d:14031
[17] S. Goldwasser, J. Kilian, Almost all primes can be quickly certified, in Proc. 18th STOC
(Berkeley, May 28–30, 1986), ACM, New York, 1986, 316–329.
[18] , Primality testing using elliptic curves, J. ACM 46 (1999), no. 4, 450–472. MR
2002e:11182
[19] F. Gouvêa, B. Mazur, The square-free sieve and the rank of elliptic curves, J. Amer. Math.
Soc. 4 (1991), no. 1, 1–23. MR 92b:11039
[20] B. H. Gross, D. B. Zagier, Heegner points and derivatives of L-series, Invent. Math. 84
(1986), no. 2, 225–320. MR 87j:11057
[21] F. J. Grunewald, R. Zimmert, Über einige rationale elliptische Kurven mit freiem Rang ≥ 8,
J. Reine Angew. Math. 296 (1977), 100–107. MR 57:6028
[22] H. Hasse, Beweis des Analogons der Riemannschen Vermutung für die Artinschen und F.
K. Schmidtschen Kongruenzzetafunktionen in gewissen eliptischen Fällen. Vorläufige Mit-
teilung, Nachr. Ges. Wiss. Göttingen I, Math.-phys. Kl. Fachgr. I Math. Nr. 42 (1933), 253–
262 (# 38 in H. Hasse, Mathematische Abhandlungen, Band 2, Walter de Gruyter, Berlin-New
York, 1975). MR 57:5648b
[23] , Abstrakte Begründung der komplexen Multiplikation und Riemannsche Vermutung
in Funktionenkörpen, Abh. Math. Sem. Univ. Hamburg. 10 (1934), 325–348 (# 40 in Helmut
Hasse Mathematische Abhandlungen, Band 2, Walter de Gruyter, Berlin-New York, 1975).
MR 57:5648b
[24] D. R. Heath-Brown, The size of Selmer groups for the congruent number problem. II, Invent.
Math. 118 (1994), no. 2, 331–370. MR 95h:11064
[25] T. Honda, Isogenies, rational points and section points of group varieties, Japan. J. Math.,
30 (1960), 84–101. MR 27:5762
[26] N. M. Katz, P. Sarnak, Zeroes of zeta functions and symmetry, Bull. Amer. Math. Soc. 36
(1999), no. 1, 1–26. MR 2000f:11114
[27] S. Kihara, On an infinite family of elliptic curves with rank ≥ 14 over Q, Proc. Japan Acad.
Ser. A Math. Sci. 73 (1997), no. 2, 32. MR 98d:11059
[28] , On an elliptic curve over Q(t) of rank ≥ 14, Proc. Japan Acad. Ser. A Math. Sci.
77 (2001), no. 4, 50–51. MR 2002a:11057
[29] N. Koblitz, Elliptic curve cryptosystems, Math. Comp. 48 (1987), no. 177, 203–209. MR
88b:94017
[30] , Introduction to elliptic curves and modular forms, 2nd edition, Graduate Texts in
Mathematics 97, Springer-Verlag, New York, 1993. MR 94a:11078
[31] V. A. Kolyvagin, Finiteness of E(Q) and X(E, Q) for a subclass of Weil curves, Izv. Akad.
Nauk SSSR Ser. Mat. 52 (1988), no. 3, 522–540, 670–671 (= Math. USSR – Izvestija 32
(1989), no. 3, 523–541). MR 89m:11056
[32] , Euler systems, in The Grothendieck Festschrift (Vol. II), P. Cartier et al., eds., Prog.
in Math. 87, Birkhäuser, Boston (1990), 435–483. MR 92g:11109
[33] K. Kramer, Arithmetic of elliptic curves upon quadratic extension, Trans. Amer. Math. Soc.
264 (1981), no. 1, 121–135. MR 82g:14028
 RANKS OF ELLIPTIC CURVES 473

[34] D. S. Kubert, Universal bounds on the torsion of elliptic curves, Proc. London Math. Soc.
33 (1976), no. 2, 193–237. MR 55:7910
[35] H. W. Lenstra, Jr., Factoring integers with elliptic curves, Ann. of Math. 126 (1987), no. 3,
649–673. MR 89g:11125
[36] E. Lutz, Sur l’equation y 2 = x3 − Ax − B dans les corps p-adic, J. Reine Angew. Math. 177
(1937), 238–247.
[37] R. Martin, W. McMillen, posting to Number Theory server, March 16, 1998.
[38] , posting to Number Theory server, May 2, 2000.
[39] B. Mazur, Modular curves and the Eisenstein ideal, Publ. math. IHES 47 (1977), 33–186.
MR 80c:14015
[40] J.-F. Mestre, Construction d’une courbe elliptique de rang ≥ 12, C. R. Acad. Sci. Paris Sér.
I Math. 295 (1982), no. 12, 643–644. MR 84b:14019
[41] , Formules explicites et minorations de conducteurs de variétés algébriques, Comp.
Math. 58 (1986), no. 2, 209–232. MR 87j:11059
[42] , Courbes elliptiques de rang ≥ 11 sur Q(t), C. R. Acad. Sci. Paris Sér. I Math. 313
(1991), no. 3, 139–142. MR 92j:11052
[43] , Courbes elliptiques de rang ≥ 12 sur Q(t), C. R. Acad. Sci. Paris Sér. I Math. 313
(1991), no. 4, 171–174. MR 92m:11052
[44] , Un exemple de courbe elliptique sur Q de rang ≥ 15, C. R. Acad. Sci. Paris Sér. I
Math. 314 (1992), no. 6, 453–455. MR 93b:11071
[45] , Rang de courbes elliptiques d’invariant donné, C. R. Acad. Sci. Paris Sér. I Math.
314 (1992), no. 12, 919–922. MR 93e:11075
[46] , Rang de certaines familles de courbes elliptiques d’invariant donné, C. R. Acad. Sci.
Paris Sér. I Math. 327 (1998), no. 8, 763–764. MR 99j:11063
[47] , Berkeley Number Theory Seminar, September 15, 2000.
[48] V. S. Miller, Use of elliptic curves in cryptography, in Advances in cryptology—CRYPTO
’85, Lect. Notes in Comp. Sci. 218, Springer-Verlag, Berlin (1986), 417–426. MR 88b:68040
[49] L. J. Mordell, On the rational solutions of the indeterminate equations of the third and fourth
degrees, Proc. Cambridge Philos. Soc. 21 (1922), 179–192.
[50] K. Nagao, Examples of elliptic curves over Q with rank ≥ 17, Proc. Japan Acad. Ser. A
Math. Sci. 68 (1992), no. 9, 287–289. MR 93m:11046
[51] , An example of elliptic curve over Q with rank ≥ 20, Proc. Japan Acad. Ser. A Math.
Sci. 69 (1993), no. 8, 291–293. MR 95a:11052
[52] , An example of elliptic curve over Q(T ) with rank ≥ 13, Proc. Japan Acad. Ser. A
Math. Sci. 70 (1994), no. 5, 152–153. MR 95e:11064
[53] K. Nagao, T. Kouya, An example of elliptic curve over Q with rank ≥ 21, Proc. Japan Acad.
Ser. A Math. Sci. 70 (1994), no. 4, 104–105. MR 95e:11063
[54] T. Nagell, Solution de quelque problèmes dans la théorie arithmétique des cubiques planes
du premier genre, Wid. Akad. Skrifter Oslo I (1935), No. 1, 1–25.
[55] J. Nekovář, On the parity of ranks of Selmer groups. II, C. R. Acad. Sci. Paris Sér. I Math.
332 (2001), no. 2, 99–104. MR 2002e:11060
[56] A. Néron, Problèmes arithmétiques et géométriques rattachés à la notion de rang d’une
courbe algébrique dans un corps, Bull. Soc. Math. France 80 (1952), 101–166. MR 15:151a
[57] , Propriétés arithmétiques de certaines familles de courbes algébriques, Proceedings
of the International Congress of Mathematicians, 1954, Amsterdam, vol. III, pp. 481–488,
Erven P. Noordhoff N.V., Groningen; North-Holland Publishing Co., Amsterdam, 1956. MR
19:321b
[58] D. E. Penney, C. Pomerance, A search for elliptic curves with large rank, Math. Comp. 28
(1974), 851–853. MR 51:12861
[59] , Three elliptic curves with rank at least seven, Math. Comp. 29 (1975), 965–967. MR
51:12862
[60] H. Poincaré, Sur les propriétés arithmétiques des courbes algébriques, J. Math. Pures Appl.,
Ser. 5, vol. 7 (1901), 161–233.
[61] N. Rogers, Rank computations for the congruent number elliptic curves, Exper. Math. 9
(2000), no. 4, 591–594. MR 2001k:11104
[62] D. E. Rohrlich, Galois theory, elliptic curves, and root numbers, Comp. Math. 100 (1996),
no. 3, 311–349. MR 97m:11075
474 KARL RUBIN AND ALICE SILVERBERG

[63] K. Rubin, Right triangles and elliptic curves, to appear in Bay Area Math Adventures, D. F.
Hayes, ed., MAA.
[64] K. Rubin, A. Silverberg, Ranks of elliptic curves in families of quadratic twists, Exper. Math.
9 (2000), no. 4, 583–590. MR 2001k:11105
[65] , Rank frequencies for quadratic twists of elliptic curves, Exper. Math. 10 (2001), no.
4, 559–569.
[66] A. Silverberg, Open questions in arithmetic algebraic geometry, in Arithmetic Algebraic Ge-
ometry (Park City, UT, 1999), IAS/Park City Mathematics Series 9, AMS, Providence, RI
(2001), 85–142.
[67] J. H. Silverman, Heights and the specialization map for families of abelian varieties, J. Reine
Angew. Math. 342 (1983), 197–211. MR 84k:14033
[68] , The arithmetic of elliptic curves, Graduate Texts in Mathematics 106, Springer-
Verlag, New York, 1986. MR 87g:11070
[69] C. L. Stewart, J. Top, On ranks of twists of elliptic curves and power-free values of binary
forms, J. Amer. Math. Soc. 8 (1995), no. 4, 943–973. MR 95m:11055
[70] J. T. Tate, The arithmetic of elliptic curves, Invent. Math. 23 (1974), 179–206. MR 54:7380
[71] J. T. Tate, I. R. Safarevic, The rank of elliptic curves, Dokl. Akad. Nauk SSSR 175 (1967),
no. 4, 770–773 (= Soviet Math. Dokl. 8 (1967), no. 4, 917–920). MR 38:5790
[72] R. Taylor, A. Wiles, Ring-theoretic properties of certain Hecke algebras, Ann. of Math. 141
(1995), no. 3, 553–572. MR 96d:11072
[73] J. B. Tunnell, A classical Diophantine problem and modular forms of weight 3/2, Invent.
Math. 72 (1983), no. 2, 323–334. MR 85d:11046
[74] V. Vatsal, Special values of anticyclotomic L-functions, preprint.
[75] A. Weil, Number theory, an approach through history, Birkhäuser, Boston, 1984. MR
85c:01004
[76] A. Wiles, Modular elliptic curves and Fermat’s last theorem, Ann. of Math. 141 (1995), no.
3, 443–551. MR 96d:11071
[77] A. Wiman, Über den Rang von Kurven y 2 = x(x + a)(x + b), Acta Math. 76 (1945), 225–251.
MR 7:70g
[78] , Über rationale Punkte auf Kurven y 2 = x(x2 − c2 ), Acta Math. 77 (1945), 281–320.
MR 7:323b
[79] , Über rationale Punkte auf Kurven dritter Ordnung vom Geschlechte Eins, Acta
Math. 80 (1948), 223–257. MR 10:472c
[80] D. Zagier, The Birch-Swinnerton-Dyer conjecture from a naive point of view, in Arithmetic
Algebraic Geometry, G. van der Geer et al., eds., Prog. in Math. 89, Birkhäuser, Boston
(1991), 377–390. MR 92c:11063

Department of Mathematics, Stanford University, Stanford, California 94305

E-mail address: [email protected]
URL: http://www.math.stanford.edu/~rubin/

Department of Mathematics, Ohio State University, Columbus, Ohio 43210

E-mail address: [email protected]
URL: http://www.math.ohio-state.edu/~silver/