Sede Electrónica de MUFACE

ELEMENTARY 3-DESCENT WITH A 3-ISOGENY
arXiv:0903.4963v2 [math.NT] 27 Apr 2009
HENRI
COHEN AND
FABIEN
PAZUKI
Abstra t. In this expository paper, we show how to use in pra ti e 3-des ent with a
3-isogeny to nd an estimate for the rank of an ellipti urve having a rational 3-torsion
subgroup, and we also give a geometri interpretation of these omputations.
Keywords : Ellipti urves, 3-des ent.

Mathemati s Subje t Classi ation :
11G05, 14G05, 14H52.
1. Introdu tion
The aim of this work is to give a very expli it way to estimate the rank of an ellipti
We will suppose that the ellipti urve has a rational 32
3
2
torsion subgroup. It allows us to pi k an ane model of the form y = x + D(ax + b) .
urve over
using
3-des ent.
3-des ent. Then we

D = 1 in se tion 3. We do
After introdu ing the des ent maps, we explain in se tion 2 how to us
show how to ompute prin ipal homogeneous spa es in the ase
the same in se tion 4 for the ase
D 6= 1,
whi h is a bit more te hni al. Se tions 5 and 6
in lude all the results needed for lo al solubility. For the sake of brevity we do not in lude
all the details of the al ulations, they are of ourse availaible upon request.
Q-rank,
7, one nds several examples of families where we nd the
In se tion
and we also give some
appli ations, su h as prime values of ertain ubi forms.

The main strategy here is improving on [4, Se tion 8.4, where this expli it way of
doing des ent is explained for
2-des ent.
The
3-Selmer group has
also been studied in [20,
pre ised by [8. See for example [6, 7, 13 and their referen es for a more general treatment.
We would like to thank J. Cremona for his interest and the anonymous referee for his
advi e.
1.1.
E/k
The Geometri Setting.
We re all a few fa ts about des ent on ellipti urves. Let
an ellipti urve over a number eld
and let
ohomology, we have the short exa t sequen es:
Y
/
E(k)/nE(k)
E(kv )/nE(kv )
v v
n 2 be an integer.
First, using Galois
/ H 1 (k, E)[n]
PPP
PPP
Q
Q
PPP
resv
PPP
v resv
(Y

v
H 1 (k, E[n])
H 1 (kv , E[n])
H 1 (kv , E)[n]
0
/
Henri COHEN and Fabien PAZUKI
n-Selmer group :

Y
(n)
1
Sel (k, E) := Ker : H (k, E[n])
H 1(kv , E)[n] .
We re all the denition of the
TateShafarevi h group :

Y
X(k, E) := Ker H 1 (k, E)
H 1 (kv , E) .
We also re all the denition of the
This leads to the following short exa t sequen e:
0 E(k)/nE(k) Sel(n) (k, E) X(k, E)[n] 0 ,

where one an show that every term is a nite group, so that

(n)

Sel (k, E) = E(k)/nE(k) X(k, E)[n] ,
whi h gives

(n)

Sel
(k,
E)

rk(E/k)
.

n
=

E(k)tors /nE(k)tors X(k, E)[n]
rk(E/k),
Thus, to get the exa t value of the rank

and the
n-torsion
we must ompute the
twist of an obje t X
X over k .
(n)
1
Sel (k, E) H (k, E[n]), if
Re all that a
is an obje t
dened over
we nd a geometri obje t
su h that
dened over
isomorphi to
Sin e
E[n],
n-Selmer
group
part of the TateShafarevi h group.
we an interpret the elements of the
n-Selmer
group as twists of the
idea gives rise to dierent interpretations of the elements of the
that is
Autk (X)
=
obje t X . This
n-Selmer group, as is learly
explained in [6, 7. In the present paper we are going to des ribe expli itly the geometri al
interpretation of those elements that we now re all. First, if
of
E,
the omplete linear system given by
Denition 1.1. A
|n.O|
O denotes the identity element

E Pn1 .
indu es a morphism:
diagram [C S] is a morphism from a torsor C under E to a

variety S . We will say that two diagrams [C1 S1 ] and [C2 S2 ] are isomorphi if
the following diagram is ommutative:
C1
S1
/
C2
/
S2
We will dene a BrauerSeveri diagram [C S] to be a twist of the diagram X =

[E Pn1 ]. In parti ular S is a twist of Pn1 , alled a BrauerSeveri variety.
Following [7, we interpret an element of the
[C S]
S = Pn1 .
diagram
take
su h that the urve
n-Selmer
group of
as a BrauerSeveri
has points everywhere lo ally, hen e one an
3-des ent
n = 3. In this parti ular ase, the BrauerSeveri diagrams

[C P2 ], the urve C being a plane ubi with points
an a tion of E[3] on it by linear automorphisms.
We now spe ialize to the ase
we are looking for are of the type

everywhere lo ally, given with
1.2.
The Arithmeti Setting.
For the proofs of all the results given in this se tion, we
refer to [4, Se tion 8.4, although there are other pointers in the literature. The
3-Selmer
group in this parti ular ase has also been studied in [20, pre ised by [8.
Let
be an ellipti urve dened over
we denote
{O, T, T }.
and having a rational
Let us stress here that the point
itself. It is easy to see that
3-torsion subgroup
that
does not need to be rational
an be given by an ane equation of the type
y 2 = x3 + D(ax + b)2
Q, and the dis riminant of E is
3
we must have b and D nonzero and 4Da 27b 6= 0.
(0, b D), so is rational if and only if D Q 2 .

with
a, b,
and
in
16b3 D 2 (4Da3 27b),

3-torsion point T is equal
equal to
so
The
to
Lemma 1.2. There exists a unique equation of E of the form y 2 = x3 + D(ax + b)2 , where
a, b, and D are in Z, D is a fundamental dis riminant (in luding 1), b > 0, and if we
write b = b1 b33 with b1 ubefree then (a, b3 ) = 1.
Proof. We an write uniquely D = D0 f 2 , where D0 is a fundamental dis riminant and

f Q , so our initial equation an be written y 2 = x3 + D0 (a x + b )2 with a = f a and
b = f b. Changing (x, y) into (x/u2 , y/u3) hanges (a , b ) into (ua , u3 b ), so it is lear
3
that we may assume that a and b are in Z, and if b = b1 b3 and g = (b3 , a), hanging
(x, y) into (xg 2 , yg 3) hanges (a, b) into (a/g, b/g 3), hen e (a, b3 ) into (a/g, b3 /g), so insures
that (a, b3 ) = 1. Finally, hanging (a, b) into (a, b) insures that b > 0. Uniqueness is
immediate and left to the reader.
From now on, we will always assume that the equation of our urve is given satisfying
the onditions of the above lemma (although we will mainly use the fa t that
fundamental dis riminant), and we will denote by
D,
the eld
is a
K = Q( D) of dis riminant
Q if D = 1 and to a quadrati eld otherwise.

E , our aim is to give an estimate for the rank of E , and if possible
3-des ent. We rst re all the denition and main properties of the
whi h will be equal to
Given su h a urve
the rank itself, using
3-des ent
maps.
Denition 1.3. Let E be an ellipti urve dened over Q, hoose an ane model given
by
an equation y 2 = x3 +D(ax+b)2 with D a fundamental dis riminant, and let T = (0, b D)

be a 3-torsion point.
(1)
The 3-des ent map is a map from E(Q) to the subgroup G3 of lasses of elements
of K /K 3 whose norm is a ube (or G3 = Q /Q 3 if D = 1) dened by (O)
= 1,
((0, b)) = 1/(2b) when D = 1, and in general by ((x, y)) = y (ax + b) D.
(2)
(3)
b is dened by a similar equation y 2 = x3 + D(b

b ax+bb)2 , where D
b = 3D ,
The urve E
3
b
a = a, and bb = (27b 4a D)/9
3-des ent map is denoted
b.
p , and the orresponding
b = (0, (27b 4a3 D) 3D/9).

Moreover we have Tb = (0, bb D)
b is dened by
The map from E to E
3

x + 4D((a2 /3)x2 + abx + b2 ) y(x3 4Db(ax + 2b))
(P ) =
,
x2
x3
b if P = O or P = T , and the map b from
for P 6= O and P 6= T , and (P ) = O
b to E is dened in the same way, repla ing the oe ients of E by those of E
b,
E
ex ept that the x- oordinate must be divided by 9 and the y - oordinate by 27.
Proposition 1.4.
(2)
and b are dual 3-isogenies (in parti ular group homomorb, so that b and b are the multipli ation-by-3 maps
phisms) between E and E
b respe tively. The kernel of (over Q) is {O, T }, and that of b is
on E and E
b Tb}.
{O,
b.
The map is a group homomorphism from E(Q) to G3 , and Ker() = Im()
(1)
2.
3-Des ent
with a Rational
We now explain how the use of the

the rank of
3-Isogeny
3-des ent maps and

b gives a pre ise estimate on
b , whi h has the same rank). Before proving
E
(and of the isogenous urve
the main result (Proposition 2.2 below), we need the following pre ise des ription of the
3-torsion points of an ellipti urve (evidently, if an ellipti urve does

a rational 3-torsion subgroup, in other words if it does not have an equation of
y 2 = x3 + D(ax + b)2 , the only rational 3-torsion point is O ).
rational
not have
the form
Lemma 2.1. Let y 2 = x3 + D(ax + b)2 be the equation of an ellipti urve E with rational 3-torsion subgroup, and assume as usual that this equation is written so that D is a
fundamental dis riminant. The rational 3-torsion points of E are the following:
(1) If D = 1, the points O and (0, b).
3
3
(2) If D = 3 and 2(9b + 4a ) = t is the ube of a rational number t 6= 0, the point O
t2
16 4 4a2
3
and the points P su h that x(P ) = + 2 4ab + a +
.
3
t
9
3
(3) Otherwise, only the point O .
Proof.
Let
Q = (x, y)
be a
3-torsion
point. Then we have
x([2]Q) = x(Q) = x(Q),
whi h gives, using the formulas on p. 59 of [19 for the dupli ation law on the ellipti urve
E:
x(3x3 + 4Da2 x2 + 12Dabx + 12Db2 ) = 0.
3
2 2
2
2
3 2 2
Let P (x) = 3x +4Da x +12Dabx+12Db . Note that Disc(P ) = 48D (27b+4Da ) b .
2
2
So either we have x = 0, then y = Db , or we have P (x) = 0 and we obtain after an easy
al ulation:
y2 =
D
(ax + 3b)2 .
3
3-des ent
It is then straightforward to nd the rational solutions, keeping in mind that
is a
fundamental dis riminant.
We an now give the following the exa t analogue of Proposition 8.2.8 of [4, whose proof
we follow verbatim.
Proposition 2.2. Let E be the ellipti urve y 2 = x3 + D(ax + b)2 and Eb the 3-isogenous
urve with equation y 2 = x3 3D(ax + (27b 4a3 D)/9)2 as above, and let and
b be the
orresponding 3-des ent maps. Then
|Im()||Im(b
)| = 3r+ ,
b ), and = 1 if D = 1 or D = 3 and = 0 otherwise.

where r is the rank of E (and of E
Proof.
If
Et
denotes the torsion subgroup of
we have
E(Q)/3E(Q) Et (Q)/3Et (Q) (Z/3Z)r .

G = Et (Q). We know that if G is a nite Abelian group then G/3G is non anoni ally
isomorphi to G[3], in other words to the group of 3-torsion points of G. Thus by Lemma
3
2.1, Et (Q)/3Et (Q) is trivial unless either D = 1, or D = 3 and 2(9b + 4a ) is a ube.
3
Write D,n for the usual Krone ker -symbol, and (a, b) for the ondition that 2(9b + 4a )
Set
is a ube. With this notation we an thus write
|E(Q)/3E(Q)| = 3r+D,1 +D,3 (a,b) .

On the other hand, let us onsider our
by-3 map, we evidently have
we evidently have
Thus
Sin e
b is the multipli ation
b E(Q))][
b E(Q))
b
b
b
|E(Q)/3E(Q)| = [E(Q) : (
(
: ((E(Q)))]
.
Now for any group homomorphism
3-isogenies and b.
b and
subgroup
of nite index in an abelian group
b
(A)
A
A/B
A/B
.
b
b
b
b
b B)
(B)
B + Ker()
(B + Ker())/B
Ker()/(Ker(
)
b
b
[(A)
: (B)]
=
[A : B]
b : Ker()
b B]
[Ker()
b
b
A = E(Q)
and B = (E(Q)). We know that Ker()
b and Tb, and Tb (E(Q)) if and only if D = 3, so (on e
(over Q) has three elements O
b = {O, Tb} if D = 3, and is trivial otherwise. Thus, if D 6= 3 we
again over Q), Ker()
b : Ker()
b B] = 1. Assume now that D = 3, so that the equation of E is
have [Ker()
2
3
2
b an be taken to be y 2 = x3 + (3ax + (9b + 4a3 ))2 . Then
y = x 3(ax + b) , and that of E
b : Ker()
b B] = 1 if Tb (E(Q)), and is equal to 3 otherwise. We know (see for
[Ker()
We are going to use this formula with
instan e [4, Proposition 8.4.4) that
Tb (E(Q))
2(9b + 4a3 ) is a
(a, b) = 1. Thus,
if and only if
other words with the notation introdu ed above, if and only if
ube, in
b
[E(Q)
: (E(Q))]
b E(Q))
b
b
[(
: ((E(Q)))]
=
.
3 D,3(1(a,b))
Putting everything together we obtain
b E(Q))][
b E(Q))
b
b
b
3r+D,1 +D,3 (a,b) = |E(Q)/3E(Q)| = [E(Q) : (
(
: ((E(Q)))]
b E(Q))][
b
b
= [E(Q) : (
E(Q)
: (E(Q))]3D,3((a,b)1) .
b E(Q))
b E(Q))]
b
b
3-des ent map on E(Q) has kernel (
, so [E(Q) : (
=
b
similarly [E(Q)
: (E(Q))] = |Im(b
)|, so nally we obtain
On the other hand the
|Im()|,
and
|Im()||Im(b
)| = 3r+D,1+D,3 ,
proving the proposition.
It follows from this proposition that to ompute the rank it is su ient to ompute the
ardinality of Im() and of Im(b
), whi h we do separately. As in the ase of
annot give an algorithm for this, sin e there is an obstru tion embodied
E , but the
ompute |Im()|.
the TateShafarevi h group of

next se tions is thus to
2-des ent, we
in the 3-part of
method works in many ases. The goal of the
3. The Case
D=1
2
We rst treat the ase D = 1. We hoose the equation of our ellipti urve as y =
x3 + (ax + b)2 with a and b as in Lemma 1.2, and we re all that the fundamental 3
3
is dened by (O) = 1, ((0, b)) = 1/(2b), and
des ent map from E(Q) to Q /Q
((x, y)) = y (ax + b)
Note:
for all other points of
E(Q).
To avoid unne essary wording, when we speak of a solution to a homogeneous
nontrivial
solution, where the variables are not all equal to 0.

k
Similarly, when we speak of a solution to a homogeneous ongruen e modulo p for some
equation we always mean a
prime
p,
p-integral, and at least

min(vp (X), vp (Y ), vp (Z)) = 0.
we always mean a solution where all the variables are
of them having
p-adi
valuation equal to
0,
typi ally
one
Theorem 3.1. Keep the above notation.

(1)
An element u Q /Q 3 belongs to the image of if and only if for some (or any)
representative u Q the homogeneous ubi equation
uX 3 + (1/u)Y 3 + 2bZ 3 2aXY Z = 0
(2)
has an integer (or rational) solution.

More pre isely, for u = 1 it has the solution (X, Y, Z) = (1, 1, 0), for u = 1/(2b) it
has the solution (X, Y, Z) = (0, 1, 1), and if y (ax + b) = uz 3 for some z Q it
has the solution (X, Y, Z) = (z 2 , x, z). Conversely, if (X, Y, Z) is a solution of the
ubi with Z 6= 0 then (x, y) = (XY /Z 2, (uX 3 (1/u)Y 3 )/(2Z 3 )) is a preimage
of u in E(Q), and z = X/Z .
3-des ent
(3)
If the above ubi has a rational solution and if u is the unique positive integer
ubefree representative of u, then u1 u2 | (2b), where u = u21 u2 with the ui squarefree
and oprime, and the solubility of the ubi is equivalent to that of
u1 X 3 + u2 Y 3 + (2b/(u1 u2 ))Z 3 2aXY Z = 0 .
Proof. (1) and (2). The ases u = 1 and u = 1/(2b) ( orresponding to the points O and
T = (0, b) respe tively) being lear, we assume that we are not in these ases. Then by
denition if u belongs to the image of there exists (x, y) E(Q) and z Q su h that
3
2
uz = y (ax + b), and if we set X = z , Y = x, and Z = z then
1
uX 3 + (1/u)Y 3 + 2bZ 3 2aXY Z = (u2 z 6 + 2uz 3 (ax + b) x3 )
u
1
= ((uz 3 + ax + b)2 x3 (ax + b)2 )
u
1
= (y 2 (x3 + (ax + b)2 )) = 0 ,
u
as laimed. Note that sin e z Q we have Z 6= 0. Conversely, let (X, Y, Z) be a solution

2
3
3
3
to our ubi with Z 6= 0. If we set x = XY /Z and y = (uX (1/u)Y )/(2Z ), we have
x3 + (ax + b)2 = (X 3 Y 3 + Z 2 (bZ 2 aXY )2 )/Z 6 , and sin e by the ubi equation we have
2Z(bZ 2 aXY ) = uX 3 + (1/u)Y 3 , it follows that
x3 + (ax + b)2 = ((uX 3 + (1/u)Y 3 )2 4X 3 Y 3 )/(4Z 6 )
so
(x, y) E(Q).
= (uX 3 (1/u)Y 3 )/(4Z 6 ) = y 2 ,
Furthermore,
((x, y)) = y (ax + b) = (uX 3 (1/u)Y 3 )/(2Z 3) (bZ 2 aXY )/Z 2

= (1/(2Z 3 ))(uX 3 (1/u)Y 3 2Z(bZ 2 aXY ))
= (1/(2Z 3 ))(2uX 3) = u(X/Z)3 ,
u up to ubes, hen e (x, y) is indeed a preimage of u, as laimed.

For (3), let u be the (unique) positive integer ubefree representative of u, and write
u = u21 u2 with the ui squarefree and oprime. Repla ing Y by u1 u2 Y in the ubi equation
so is equal to
we obtain the equivalent equation
u21 u2 X 3 + u1 u22 Y 3 + 2bZ 3 2au1 u2 XY Z = 0 .
It is lear that this homogenous ubi has a rational solution if and only if it has an integer
solution, and we may in addition assume that
gcd(X, Y, Z) = 1.
Assume by ontradi tion
u1 u2 2b. Sin e u1 u2 is squarefree, this means that there exists a prime p su h that
p | u1 u2 and p 2b. Sin e ex hanging X and Y in the above equation is equivalent to the
ex hange of u1 and u2 , we may assume that p | u1 , hen e p u2 . Sin e p divides the rst,
se ond and fourth term of the equation it divides the third, and sin e p 2b, we dedu e
2
that p | Z . Thus p divides the rst, third and fourth term, so it divides the se ond, and
3
sin e u1 is squarefree and p u2 we dedu e that p | Y . Thus, p divides the se ond, third,
and fourth term, so it divides the rst, and again sin e u1 is squarefree and p u2 , we
that
dedu e that
u1 u2
p | X,
ontradi ting the assumption
gcd(X, Y, Z) = 1.
We an thus divide by
to obtain the nal equation given in (3).
Geometri Interpretation:
The plane ubi given in (1) of Theorem 3.1 is the equation of a twist
E . We an re over a linear a tion
pi k s Q and onsider the a tion:
urve
we
of the ellipti
by linear automorphisms by doing the following:
(X : Y : Z) (sX : (1/s)Y : Z).
This a tion gives a urve isomorphi to
with
repla ed by
us3 .
This is what was
predi ted in the geometri al interpretation of [6, 7 re alled in the introdu tion.
4. The Case
We now assume spe i ally that
so that
K = Q( D)
is a genuine quadrati
3-torsion subgroup, we must ompute the

b , whose
image of the 3-des ent map both for the urve E , and for a 3-isogenous urve E
b is su h that D
b = 3D . Thus, we always need to onsider urves with D 6= 1. We will
D
denote by the nontrivial element of the Galois group Gal(K/Q), so that ( D) = D ,
and by N = NK/Q the norm from K to Q, both for elements and for ideals. If u K we
3
denote by [u] the lass of u in K /K .
eld. Note that to use
3-des ent
D 6= 1,
D 6= 1
with a rational
Des ription of the Image of .
y 2 = x3 + D(ax + b)2 ,
3
of lasses
and the 3-des ent map is a map from E(Q) to the subgroup G3 of K /K
3
[u] of elements
(O) = 1 and ((x, y)) =
u su h that u (u) = NK/Q (u) Q , dened by
/ E(Q)). The image
y (ax + b) D for all other points of E(Q) (note that T = (0, b D)
of an be des ribed as follows.
4.1.
The equation of our urve is
Theorem 4.1. Keep the above notation.
An element [u] G3 K /K 3 belongs to the image of if and only if for some (or
any) representative u K of the form u = v 2 (v) the homogeneous ubi equation
2b
2v2 X 3 + 2Dv1 Y 3 + 2
Z 3 + 6v1 X 2 Y + 6v2 DXY 2 + 2a(X 2 Z DY 2 Z) = 0
v1 Dv22
has an integer (or a rational) solution, where we write v = v1 + v2 D .

(2) More pre isely, for u = 1 it has the solution (X, Y, Z) = (1, 0, 0), and if y (ax +
b) D = v 2
(v)z 3 for some z Q it has the solution (X, Y, Z) = (z1 , z2 , 1), where
z = z1 + z2 D . Conversely, if (X, Y, Z) is a solution of the ubi with Z 6= 0 then
!
X 2 DY 2
(v(X + Y D)3 )
(x, y) = v (v)
, v (v)
Z2
Z3
is a preimage of u in E(Q) with z = (X + Y D)/Z , where by abuse of notation

we write () = ( + ())/2.
(1)
3-des ent
b = b D , and ignore for the moment all rationality

2
3
questions, the equation of our urve is y = x + (a x + b ) , so if y (a x + b ) = y
3
2
(ax + b) D =
uz then (X,
Y, Z) = (z , x, z) is a solution to the modied ubi uX 3 +
(1/u)Y 3 + 2b DZ 3 2a DXY Z = 0, and onversely if (X, Y, Z) is su h a solution then

(x, y) = (XY /Z 2 , (uX 3 (1/u)Y 3 )/(2Z 3 )) is a point on the urve. So formally there
is no problem. We now must add the ondition that not only (x, y) E(K), but that
(x, y) E(Q).
We rst hoose suitable representatives u. Let for the moment u be any representative
3
2
3
2
3
of [u], so that u (u) = for some Q. This implies that (u / ) (u / ) = 1, hen e
2
3
by a weak form of Hilbert's theorem 90, we have u = v/ (v) for some v K , hen e
u = (u/)3 (v)/v = (u/(v))3v 2 (v). Sin e u is dened up to ubes, we may therefore
2
3
assume that u = v (v) for some v K . Thus u (u) = with = v (v). Multiplying
v by a suitable rational number we an assume that v ZK .
2
Now the ondition x Q means that XY /Z Q, so that Y /Z = (X/Z) for some
Q . The ondition y Q is thus that u3 (3 /u) ()3 Q, where = X/Z .
2
Repla ing u by v (v) and dividing by the rational number v (v) gives the ondition
v3 (3 /(v 3 (v)2 )) ()3 Q. Setting
= v3 and r = (/(v (v)))3 Q , this gives
the
ondition r () Q, so if = s + t D , we have r () = s + t D r(s t D),
hen e the ondition is t(r + 1) = 0, in other words either r = 1, so = , or t = 0.
3
2
2
3
If t = 0 then = v Q, hen e u = v (v) = (/( ())) , so [u] is trivial. It follows
that the ase t = 0 orresponds to the unit element of G3 , whi h we now ex lude.
3
3
The ubi equation is thus u (u ) +
Thus we may assume that = .
3
3
2a D ()+2b D = 0, and sin e = v (v) this gives (v
(v ))/(2 D)+a
()+
b/(v (v)) = 0. Re all that v is given, so write v = v1 + v2 D and = x1 + x2 D. The
Proof.
If we set
a = a D
and
above equation is thus
2v2 x31 + 2Dv1 x32 + 6v1 x21 x2 + 6v2 Dx1 x22 + 2a(x21 Dx22 ) + 2b/(v12 Dv22) = 0 ,
so setting
x1 = X/Z , x2 = Y /Z
with
X, Y , Z
in
we obtain nally
2v2 X 3 + 2Dv1 Y 3 + (2b/(v12 Dv22 ))Z 3 + 6v1 X 2 Y + 6v2 DXY 2 + 2a(X 2 Z DY 2 Z) = 0 .

The formulas for
(X, Y, Z)
knowing
and
and vi e versa are immediately obtained by
repla ing the orresponding quantities in the formula given for the ase
D = 1.
Remarks.
(1) The ubi equation in (1) of Theorem 4.1 an be written
D)3 (v)(X Y D)3 )/ D
+ 2aZ(X + Y D)(X Y D) + 2b/(v (v))Z 3 = 0 .
(v(X + Y
(2) By the theorem, the solubility of the equation depends only on the lass [u] of u,
3
hen e we an hange v into v for any K , or v into vr for any r Q without

hanging the solubility of the equation. This is of ourse lear dire tly.
10
is a group, [u] Im() if and only if [1/u] Im(), so the

v is equivalent to that for v 1 . Furthermore sin e (u) = u1 (v (v))3 ,
[ (u)] Im(), so the solubility for v is equivalent to that for (v).
Geometri Interpretation: The plane ubi given in (1) of Theorem 4.1 is the equation of a twist C of the ellipti urve E . We get a linear a tion by linear automor
phisms by doing the following: we pi k s Q and onsider the a tion:
(3) Sin e the image Im() of

solubility for
(4)
(X : Y : Z) (sX : sY : (1/s2 )Z).
This a tion gives a urve isomorphi to
with
repla ed by
us9 .
The reader will noti e that we have not given an analogous result to (3) of Theorem 3.1,
whi h is essential sin e it is ne essary to he k only a nite number of elements of
G3 .
We
do this in the next subse tion.
Redu tion of Elements of G3 . We begin by the following lemma.

Lemma 4.2. Assume that x and y are rational numbers su h that y 2 = x3 + D(ax + b)2 ,
4.2.
y = n/d3 with gcd(m,

d) = gcd(n, d) = 1 and d > 0. Finally, set
and write x = m/d2 and
2
f = gcd(n d(am + bd ) D, n + d(am + bd2 ) D), where the GCD is understood in the
sense of ideals.
(1) There exist integers f and g su h that g is a squarefree integer dividing D , and
f = f gd, where d is the unique ideal su h that d2 = gZK (when D = 1 we have of
ourse g = 1 and d = Z).
3
2
(2) If we write f = f1 q with f1 ubefree then gf1 q
| 2b, and in parti ular, g |
gcd(D, 2b).
(3) If p | f1 then p is split in K/Q, and in parti ular g and f1 are oprime.
Proof. The ase D = 1, whi h we do not need, is left to the reader, so assume that
D 6= 1, so that K is a quadrati eld. We an write uniquely f = F d, where F Z and
d is an integral ideal of K whi h is primitive, in other words whi h is not divisible by any
element of Z other than 1. Evidently d annot be divisible by inert primes; sin e f is
the GCD of two onjugate elements it is stable by onjugation, hen e d annot be divisible
by an ideal p above a split prime p, otherwise it would also be divisible by (p), hen e
2
by p = p (p). Finally sin e p = pZK for a ramied prime p, d annot be divisible by
a ramied prime to a power higher than the rst, so d is equal to a produ t of distin t
Q
ramied primes. Thus we have d =
primes p, and if
pS0 p for some set S0 of ramiedQ
Q
g = pS0 p, where p is the prime number below p, we thus have f = F p|g p and g | D ,
2
2
hen e also f = F g .
2
3
Let us now use our equation. Repla ing x and y by m/d and n/d we obtain the
equation
(n d(am + bd2 ) D)(n + d(am + bd2 ) D) = n2 Dd2 (am + bd2 )2 = m3 .
F 2 g | m3 . Let p | g and p be the prime ideal above p. Sin e the two fa tors
are onjugate, if v 1 is the p-adi valuation of the rst fa tor, it is also that of the se ond.
It follows that
11
3-des ent
This implies both that
v = vp(f)
and that
3vp(m) = 2vp(f),
in other words
3vp (m) = vp(f) = v = vp(f2 )/2 = vp(F 2 g)/2 = vp (F 2 g) = 1 + 2vp (F )

p | g and g is squarefree. We dedu e that vp (F ) 1 (mod 3), and in parti ular
that vp (F ) 1, so p | F , proving that g | F . Thus, we will set F = f g for some f Z.
The same reasoning shows that if p is any inert or ramied prime (dividing f or not) then
3vp (m) = vp(f), so 3 | vp(f).
2 3
3
2
3
3
Sin e f g | m we have g | m and f | (m/g) . Write f = f1 q with f1 ubefree.
For all primes p we have vp (m/g) vp (q) + 2vp (f1 )/3. Sin e 0 vp (f1 ) 2 we have
2vp (f1 )/3 = vp (f1 ), so f1 q 2 | m/g . Note that f2 = f 2 g 3 = f12 g 3 q 6 , so sin e for any inert or
ramied prime we have 3 | vp(f), for su h a prime we have vp (f1 ) = 0, so f1 is only divisible
by split primes. In parti ular it is oprime to D , hen e to g .
2
Sin e (n d(am + bd ) D)/(f g) is an algebrai integer and D is a fundamental dis rim2
3
2
inant, it follows that f g | 2 gcd(n, d(am + bd )), hen e f g = gf1 q | 2 gcd(n, d(am + bd )).
3
2
2
3
2
Sin e 2bd = 2d(am + bd ) 2adm and gf1 q | m we dedu e that 2bd 0 (mod gf1 q ).
3
Sin e d and n are oprime there exist integers u and v su h that un + vd = 1, hen e
2b = 2bvd3 + 2bun, and sin e gf1 q 2 | 2bd3 and gf1q 2 | 2n, we have gf1q 2 | 2b, as laimed.
sin e
Corollary 4.3. Keep the above notation, and let [u]
G3 . There exists an
integral ideal v of K su h that uZK = v (v)q for some ideal q of K , and whi h is su h
that gcd(v, (v)) = 1 and f1 = NK/Q (v) is a ubefree divisor of 2b divisible only by primes
whi h are split in K/Q.
2
Proof.
Im()
The above lemma states that if we set

f = gcd n d(am + bd2 ) D, n + d(am + bd2 ) D
d K su h that f = f1 q 3 gd with f1
2
2
ubefree divisible only by split primes, g | gcd(D, 2b), gZK = d , and gf1 q | 2b. Thus
f = f1 (qd)3 . Set a = (n d(am + bd2 ) D)/f and a+ = (n + d(am + bd2 ) D)/f, so that
gcd(a , a+ ) = 1 and our equation implies a a+ = m3 /f2 (sin e we work with ideals, we
lose some unit information here). Sin e f1 is ubefree and divisible only by split primes,
2
3
2
it is also ubefree as an ideal, and the ondition f | m implies as above that f1 (qd) | m,
2
hen e gf1 q | m (this time in Z), so the equation now reads
there exist integers
f1 , g ,
and
in
and an ideal
a a+ = f1 (m/(gf1 q 2 ))3 ZK .
f is stable by onjugation, it is lear that a+ = (a ) so this equation gives the norm

a . In any ase, write a = f q3 , a+ = f+ q3+ with f ubefree, so that in parti ular
gcd(f , f+ ) = 1, and also f+ = (f ) and q+ = (q ). At the level of ideals our equation
3
2 3
is thus f f+ (q q+ ) = f1 (m/(gf1 q )) ZK . As we already mentioned f1 is also ubefree
in K , and sin e f and f+ are oprime and ubefree, by uniqueness of the de omposition
2
as the produ t of a ubefree ideal with a ube it follows that q q+ = m/(gf1 q )ZK and
f f+ = f1 ZK . In parti ular, f | f1 .
Sin e
of
12
on an ane
3-des ent map is dened
2
y (ax + b) D = (n d(am + bd ) D)/d3 . Thus
Now re all that the

ubes of
point as the lass modulo
((x, y))ZK = fa = f1 (qd)3 f q3 = f1 f (qdq )3 = v2 (v)q31
for some ideal
Remark.
q1 ,
with
v = f ,
as laimed.
It is lear that the ondition
gcd(v, (v)) = 1
v is primitive,
1. Furthermore, sin e
(Cl(K)/Cl(K)3 )[ + 2],
the elements of G killed by
implies that
in other words that the only elements of Z whi h divide it are

v2 (v) = uq3, the ideal lass of v in Cl(K) belongs in fa t to
where for any group
and map
in other words the kernel of
from
to
G, G[]
denotes
Now re all the denition of a 3-virtual unit (or virtual ube ) and 3-Selmer group : an
u K is a virtual ube if uZK = q3 is the ube of an ideal. The group of virtual
element
ubes modulo ubes of elements is alled the

It is lear that
sequen e
S3 (K) G3 ,
3-Selmer
group of
and denoted
S3 (K).
and it is well-known and easy that we have a natural exa t
1 U(K)/U(K)3 S3 (K) Cl(K)[3] 1 .

Let as usual I(K) denote the group of (nonzero) fra tional ideals of K ,
3
subgroup of I(K)/I(K) of lasses of ideals whose norm is a ube.
and let
Gi3
be the
Lemma 4.4. We have a natural exa t sequen e

1 S3 (K) G3 Gi3 Cl(K)/Cl(K)3 1 .
Proof. Consider rst the map i from G3 to Gi3 whi h sends a lass [u] to the lass of
uZK . It is lear that it does send G3 to Gi3 . If [u] is sent to the unit element of Gi3 this
3
means that uZK = q for some ideal q, in other words that [u] S3 (K), giving the kernel.
Consider now the map sending the lass of an ideal modulo ubes to its ideal lass. It
i
3
denes a map from G3 to Cl(K)/Cl(K) . If some ideal a is sent to the unit element of
3
Cl(K)/Cl(K) this means that there exist an ideal q and an element K su h that
a = q3 . Thus, the lass of a modulo ubes of ideals is equal to that of ZK . Furthermore,
is a ube, so that of also (sin e 1 is a ube), hen e does ome from
Gi3 . Finally, let us show that the map is surje tive. Let a be
3
an ideal, representative of an element of Cl(K)/Cl(K) . Then sin e NK/Q (a) = a Q ,

aNK/Q (a) is in the same ideal lass as a, and its norm is evidently equal to (NK/Q (a))3 ,
i
so the lass of a is the image of the lass of aNK/Q (a) in G3 , proving surje tivity and the
the norm of
G3 ,
proving exa tness at
lemma.
Corollary 4.5. If we denote by i = i the 3-des ent map from E to Gi3 we have
|Im()| = |Im(i)||S3 (K) Im()| .
In parti ular, if D < 0, D 6= 3, and 3 h(K) we have |Im()| = |Im(i )|.

Proof.
i indu es an inje tion from G3 /S3 (K) to Gi3 . Thus,

i indu es a bije tion from H/(S3 (K) H) to i(H).
By the above lemma the map
for any subgroup
of
G3
the map
13
3-des ent
Applying this to the nite subgroup Im() gives the formula of the orollary, and the
spe ial ase orresponds to
S3 (K) = 1.
is a 3-virtual unit then N() = () is a ube. More generally, if is

3
2
3
su h that N() = n is a ube we an write = v (v)q with v = and q = 1/n. Thus,
using Corollaries 4.3 and 4.5 and Theorem 4.1 we an ompute |Im()|. Note however
Note that if
that it will be ompletely algorithmi and easy to prove everywhere lo al solubility of our
homogeneous ubi s, so that we will ompute the
3-Selmer group of E .
On the other hand,
for global solubility, either we nd a solution with a reasonable sear h bound, or we are led
to believe that no su h solution exists, oming from a nontrivial element of the
3-part
of
the TateShafarevi h group. Thus, it is reasonable to give an algorithm whi h omputes

both the rank of the
3-Selmer group and a
lower bound for the rank of the urve, so whi h
gives the exa t rank when they oin ide.

(1) For ea h lass
[] S3 (K)
hoose a representative
ubi of Theorem 4.1 has a solution for

group of
[]
v=
K ,
and he k that the
(more on this later), let
for whi h it is everywhere lo ally soluble (ELS), and
TG
TS
be the
the subgroup
generated by the elements for whi h we nd that it has a global solution. Thus
TG T TS S3 (K), where T = S3 (K) Im(). This will allow us to ompute T

exa tly if TS = TG , and otherwise if the sear h bound is su iently large, we suspe t
(but annot prove without further work) that in fa t T = TG and that the elements
of TS /TG orrespond to nontrivial elements of the 3-part of the TateShafarevi h
group. Note that using the fa t that TG and TS are groups, it is not ne essary to
test all lasses [], but in fa t using the fa t that they are even F3 -ve tor spa es it
is su ient to work on bases of these spa es and use linear algebra. Finally, hoose
a set
RS
of representatives of
S3 (K)/TS
and a set
RG
of representatives of
TS /TG .
be the largest positive integer ubefree divisor of 2b divisible only by split

Q
f = 1is pvi i with 1 vi 2. For ea h pi , let pi be one of the
two prime ideals above pi , xed on e and for all. Find all ideals v of the form
Q
v = 1is pixi vi with 1 xi 1 whose ideal lass is a ube (although this seems
s
to be 3 prin ipal ideal tests, it is easy to redu e to only s su h tests using linear
(2) Let
primes, write
algebra, but in pra ti e
Cl(K)
will be small).
v that we have found write v = uq3 for some ideal q and some
element u K , where learly the lass [u] of u in G3 is determined uniquely

modulo multipli ation by an element of S3 (K). For the moment, we hoose any u
(3) For ea h ideal
as above.
(4) For ea h
[S ] RS ,
where
RS
is the system of representatives of S3 (K)/TS omS K , he k whether the ubi of Theorem
puted in (1), and any representative
4.1 is everywhere lo ally soluble (ELS) for
v = uS
(more on this later). If for some

[S ] RS this is the ase, there will be only one, and then v Sel(i), with evident
/ Sel(i). In this latter ase, we do nothing more, otherwise
notation, otherwise v
for ea h
[G ] RG ,
where
RG
in (1), and any representative
is the system of representatives of TS /TG omputed

G K , he k whether the ubi of Theorem 4.1
14
has a global solution up to a reasonable sear h bound for v = uS R . If this is the

i
ase, on e again there will be only one, and then v Im( ), otherwise we suspe t
/ Im(i ). We let IG be the group generated by the v
(but annot be sure) that v
i
for whi h we are sure, so that IG is a subgroup of Im( ), probably equal to it.
(5) At the end of this pro ess we have omputed the Selmer group ardinality | Sel()| =
| Sel(i )||TS |, and groups |IG | and TG probably equal to Im(i ) and T respe tively,
i
i
but in any ase satisfying TG T TS and IG Im( ) Sel( ), and so
|IG ||TG | | |Im()| | | Sel()|, the unknown quantity being |Im()|. If |IG ||TG | =
| Sel()| then of ourse these quantities are equal to |Im()|. Otherwise, we output
both quantities, and a message saying that we expe t |Im()| to be equal to |IG ||TG |
and that the elements of Sel()/IG TG orrespond to nontrivial elements of the 3part of the TateShafarevi h group.
5. Lo al Solubility of
u1 X 3 + u2 Y 3 + u3 Z 3 cXY Z = 0
There remains to de ide whether or not the ubi s of Theorems 3.1 and 4.1 have a
rational solution. It is unfortunately well-known that there is no algorithm for doing this.
We thus pro eed as follows: we rst he k whether the ubi is everywhere lo ally soluble
(whi h we will abbreviate to ELS). If not, there are no rational solutions. Otherwise, either
there is an obstru tion in the
3-part
of the TateShafarevi h group, or there does exist a
rational solution whi h we an nd using a more or less e ient sear h. If we do nd one,
we are done, otherwise we give up and an only give bounds on
value.
|Im()|,
not its pre ise
Testing ELS is an algorithmi pro ess, but is not ompletely trivial. In this se tion we
give su h an algorithm. Sin e the degree is odd there is no need to look at lo al solubility
in
R.
We treat the following slightly more general problem: de ide solubility in
Qp
of the
equation
D = 1,
For
obtain with
5.1.
F (X, Y, Z) = u1 X 3 + u2 Y 3 + u3 Z 3 cXY Z = 0 .
in other words, for the ubi of Theorem 3.1, we apply the results that we
u3 = 2b/(u1 u2 )
c = 2a.
Redu tion of the Cubi and Bad Primes.
suitable powers of
u3
and
and
are
By multiplying
X, Y ,
p-integral.
Dividing by a suitable power of
Z by
u1 , u2 ,
and
it is lear that without loss of generality we may assume that

we assume that
min(vp (u1 ), vp (u2 ), vp (u3 ), vp (c)) = 0.

We need further redu tions, as follows.
Lemma 5.1. Assume as above that min(vp (u1), vp (u2 ), vp(u3 ), vp (c)) = 0.
(1)
(2)
If min(vp (u1 ), vp (u2), vp (u3 )) > 0 the equation is soluble in Qp .

Assume that vp (c) > 0, so that min(vp (u1 ), vp (u2 ), vp (u3)) = 0. The equation is
equivalent to one where either vp (c) = 0 or
max(vp (u1 ), vp (u2 ), vp (u3 )) 2 .
15
3-des ent
(3)
Assume that vp (c) > 0 and that max(vp (u1 ), vp (u2 ), vp (u3 )) 2 (whi h an be
a hieved by (2)), and without loss of generality order the variables so that 0 =
vp (u1 ) vp (u2 ) vp (u3 ) 2, and let v = (vp (u2 ), vp (u3 )). Then
(a) If v = (1, 2) the equation is not soluble in Qp .
(b) Or the equation is equivalent to one su h that either vp (c) = 0 or vp (u1 u2 u3 )
2 (and min(vp (u1 ), vp (u2), vp (u3 )) = 0 if vp (c) > 0).
u1 X 3 + u2 Y 3 + u3 Z 3
distinguish the ases c 6= 0 and
Thus, given a ompletely general ubi equation of the form
cXY Z = 0,
c = 0.
we use the following pro edure, where we
c 6= 0.
g = gcd(u1 , u2, u3 , c),
Assume rst that

(1) Let
(u1 , u2, u3 , c)
gcd(u1, u2 , u3 , c) = 1.
and repla e
that we may now assume that
by
(u1 /g, u2/g, u3/g, c/g),
so
p | c, do the following.
By dividing u1 , u2 , u3 , and c by suitable powers of p as explained in the lemma,
we redu e to an equation with either vp (c) = 0 or max(vp (u1 ), vp (u2 ), vp (u3 ))
2.
If now vp (c) = 0 or if min(vp (u1 ), vp (u2 ), vp (u3 )) > 0 we do nothing more for
the prime p. Otherwise, reorder the variables ui so that 0 = vp (u1 ) vp (u2 )
vp (u3 ) 2.
If vp (u2 ) = 1 and vp (u3 ) = 2, the equation has no solution.
Otherwise, if ne essary by hanging (u1 , u2 , u3 , c) into
(2) For ea h prime

(a)
(b)
( )
(d)
(u2 /p2 , u3/p2 , pu1, c/p) ,

we may assume that we also have
vp (u1 u2 u3 ) 2.
c = 0.
Let g = gcd(u1 , u2 , u3 ), and repla e (u1 , u2 , u3 ) by (u1 /g, u2 /g, u3 /g), so that we
may now assume that gcd(u1 , u2 , u3 ) = 1.
Repla e u1 , u2 , and u3 by their ubefree part, so that max(vp (u1 ), vp (u2 ), vp (u3 ))
2.
For ea h prime p | u1 u2 u3 , do the following.
(a) Reorder the variables ui so that 0 = vp (u1 ) vp (u2 ) vp (u3 ) 2.
(b) If vp (u2 ) = 1 and vp (u3 ) = 2, the equation has no solution.
( ) Otherwise, if ne essary by hanging (u1 , u2 , u3 ) into
Assume now that

(1)
(2)
(3)
(u2 /p2 , u3 /p2 , pu1 ) ,

we may assume that we also have
vp (u1 u2 u3 ) 2.
This leads to the following denition:
Denition 5.2. We will say that a ubi equation is p-redu ed if

min(vp (u1 ), vp (u2 ), vp (u3 )) = 0
and if for all primes p dividing c (all primes if c = 0) we have vp (u1 u2 u3 ) 2.
16
Thanks to the above lemma, we an therefore always assume that our ubi is
sin e if
min(vp (u1 ), vp (u2 ), vp (u3 )) > 0
the ubi has a
p-adi
p-redu ed,
solution.
Lemma 5.3. Let p be a prime and let u1 X 3 + u2 Y 3 + u3 Z 3 cXY Z = 0 be a ubi with
p-integral oe ients, not ne essarily p-redu ed. If p 6= 3, vp (u1 ) = vp (u2) = vp (u3 ) = 0,
and vp (27u1 u2u3 c3 ) = 0, the ubi is soluble in Qp .
Proof.
Let us look at the singular points of the ubi over Fp . First, a point with Z = 0
u1 X 3 + u2 Y 3 = 0, 3u1 X 2 = 0, and 3u2Y 2 = 0, and sin e we assume p 6= 3 and
is singular if
vp (ui) = 0, this implies X = Y = 0, whi h is not possible. Thus, any singular point has
Z 6= 0, so we may assume that Z = 1. Sin e p 6= 3, the equation has a singular point in Fp
2
2
for Z = 1 if and only if 3u1 X cY = 0, 3u2 Y cX = 0, and 3u3 cXY = 0. If there is
su h a singular point we annot have c = 0, otherwise u3 = 0, in other words vp (u3 ) 1,
2
2
2
4 3
a ontradi tion. Thus Y = 3u1 X /c, X = 3u2 Y /c = 27u1 u2 X /c , hen e either X = 0,
3
3
2
whi h is not possible sin e otherwise X = Y = 0 hen e u3 = 0, or X = c /(27u1 u2 ),
3
3
3
so that 3u3 = cXY = 3u1 X = c /(9u1 u2 ), in other words 27u1 u2 u3 c = 0, whi h is
also ex luded. Thus the ubi is nonsingular over Fp . Sin e it is a urve of genus 1 and
3 2 2 > 0, it follows from the Weil bounds that for every prime p it has a nontrivial
point in Fp . If p is not in the ex luded list this point is ne essary nonsingular, and sin e
we assume p 6= 3 we an perform a Hensel lift to Zp as soon as we know that there is a
solution modulo p, proving the lemma.
5.2.
Lo al Solubility for p | u1u2 u3 , p 6= 3.
Thanks to Lemma 5.1, we may assume that
p-redu ed, and thanks to Lemma 5.3, it is enough to onsider the primes p
vp (u1 u2 u3 ) > 0, vp (27u1 u2 u3 c3 ) > 0, or p = 3. We begin by primes p 6= 3
su h that vp (u1 u2 u3 ) > 0. For su h primes, by symmetry we may assume that vp (u1 ) > 0,
and sin e the ubi is p-redu ed we have min(vp (u2 ), vp (u3 )) = 0, so again by symmetry
we may assume that 0 vp (u1 ) vp (u2 ) vp (u3 ).
our ubi is
su h that
Lemma 5.4. Let p be a prime, assume that our ubi is p-redu ed, and assume that p 6= 3
and vp (u1 u2 u3) > 0, with 0 vp (u1 ) vp (u2 ) vp (u3 ). The ubi is soluble in Qp if and
only if one of the following onditions is satised.
(1) vp (c) = 0.
(2) vp (c) > 0, vp (u1 ) = vp (u2 ) = 0, and the lass of u1 /u2 modulo p is a ube in Fp .
(3) vp (c) > 0, vp (u1 ) = 0, vp (u2 ) = vp (u3 ) = 1, and the lass of u2 /u3 modulo p is a
ube in Fp .
Remarks.
p-redu ed, the above lemma overs all possible ases
p 6= 3 and vp (u1 u2 u3 ) > 0: indeed, if vp (c) > 0 we have ne essarily
vp (u1 u2 u3 ) 2, so up to ordering we have either vp (u1 ) = vp (u2 ) = 0 (and vp (u3 )
2), or vp (u1 ) = 0 and vp (u2 ) = vp (u3) = 1.
It follows from the proof that the ubi is also soluble in ase (1) when p = 3, in
other words if v3 (u1 u2 u3 ) > 0 and v3 (c) = 0, but the assumption p 6= 3 is ne essary
(1) Note that sin e the ubi is

for whi h
(2)
in ases (2) and (3).
17
3-des ent
5.3.
Lo al Solubility for p | (27u1 u2 u3 c3 ), p 6= 3.
a prime dierent from
3 su h
that
p | (27u1u2 u3 c3 ).
p is
p u1 u2 u3
In this se tion, we assume that

We may also assume that
sin e these primes have already been taken are of in the pre eding subse tion.
Lemma 5.5. Let p be a prime, assume that our ubi is p-redu ed, and assume that p 6= 3,
vp (u1 u2 u3 ) = 0, and vp (27u1u2 u3 c3 ) > 0. The ubi is soluble in Qp if and only if u2 /u1
is a ube in Fp .
5.4.
Lo al Solubility for p = 3.
By the remarks made above, when
Finally we onsider lo al solubility at the prime
v3 (c) = 0
p = 3.
we have seen that the ubi is lo ally soluble
3 if v3 (u1 u2 u3 ) > 0. We may therefore assume either that v3 (c) > 0,

v3 (u1u2 u3 ) = 0. In the latter ase the result is immediate:
at
or that
v3 (c) =
Lemma 5.6. If v3 (c) = 0 the ubi has a solution in Q3 .

The nal ase to be treated is the ase
v3 (c) > 0.
In this ase, we need a small strength-
ening of Hensel's lemma, whi h we give in a slightly more general form that we will need
below.
Lemma 5.7. Set P0 = (X0 , Y0 , Z0 ), and let k 1. Assume that v3 (F (P0 )) 2k and that
min(v3 (FX (P0 )), v3 (FY (P0 )), v3 (FZ (P0 ))) = k . Assume that all se ond and third partial
derivatives of F are divisible by 3 at the point P0 , the ondition on the third derivatives
being required only if k = 1. There exists a 3-adi point P = (X, Y, Z) su h that F (P ) = 0
with P P0 (mod 3k ).
F (P ) = u1 X 3 + u2 Y 3 + u3 Z 3 cXY Z we have for instan e FX (P ) = 3u1 X 2
cY Z , and sin e v3 (c) > 0 all the partial derivatives are divisible by 3 at any point, so to
2k
apply the lemma it is enough to nd a point su h that F (P0 ) 0 (mod 3 ) and :
Now for
min(v3 (FX (P0 )), v3 (FY (P0 )), v3 (FZ (P0 ))) = k.
We will mainly use this lemma with
k = 1,
but we will need it also with
In fa t we need a variation of the above lemma for
k = 2.
k = 2.
Lemma 5.8. Let P0 = (X0 , Y0 , Z0 ) be su h that

v3 (F (P0 )) 3 and
min(v3 (FX (P0 )), v3 (FY (P0 )), v3 (FZ (P0 ))) = 2 ,
and assume that all se ond, third, and fourth partial derivatives of F are divisible by 3 at
the point P0 . Assume in addition that for all P1 P0 (mod 3) su h that v3 (F (P1)) 3
we also have min(v3 (FX (P1 )), v3 (FY (P1 )), v3 (FZ (P1 ))) = 2. There exists a 3-adi point
P = (X, Y, Z) su h that F (P ) = 0 with P P0 (mod 3).
Of ourse for our ubi s the fourth partial derivatives vanish.
(X, Y, Z) of
min(v3 (X), v3 (Y ), v3 (Z)) = 0.
Re all that a solution

that
a ongruen e modulo some power of
is always su h
Lemma 5.9. Let p = 3, assume the ubi 3-redu ed, and assume that v3 (c) > 0, so that
v3 (u1u2 u3 ) 2. Reorder the variables so that 0 = v3 (u1 ) v3 (u2 ) v3 (u3 ).
18
(1)
(2)
(3)
(4)
(5)
If v3 (c) 2 and v3 (u1 u2u3 ) = 0 the ubi has a solution in Q3 if and only if
ui uj (mod 9) for some i 6= j .
If v3 (c) 2 and exa tly one of the ui is divisible by 3 (in other words if v3 (u2 ) = 0
and v3 (u3) > 0), the ubi has a solution in Q3 if and only if either u1 u2
(mod 9), or if v3 (u3 ) = 1.
If v3 (c) 2, and two of the ui are divisible by 3 (in other words if v3 (u2 ) =
v3 (u3 ) = 1 sin e the ubi is 3-redu ed), the ubi has a solution in Q3 if and only
if u2 /3 u3 /3 (mod 9).
If v3 (c) = 1 and exa tly one of the ui is divisible by 3 (i.e., v3 (u2 ) = 0 and v3 (u3 ) >
0), the ubi has a solution in Q3 if and only if either u1 u2 (mod 9), or if
there exist s1 and s2 in {1, 1} su h that c s1 u1 + s2 u2 + s1 s2 u3 (mod 9).
If v3 (c) = 1 and two of the ui are divisible by 3 (i.e., v3 (u2 ) = v3 (u3 ) = 1), the
ubi has a solution in Q3 .
v3 (c) = 1 and none of the ui is
3, or equivalently v3 (u1 u2 u3 ) = 0, whi h will be overed by Lemma 5.10 below.
Note that this lemma does not over the ase where
divisible by
Lemma 5.10. Let p = 3 and assume that v3 (c) > 0 and v3 (u1u2 u3 ) = 0.
(1)
(2)
If ui uj (mod 9) for some i 6= j , the ubi has a solution in Q3 .

If ui 6 uj (mod 9) for i 6= j (whi h implies that u1 u2 u3 1 (mod 9)), the ubi
has a solution in Q3 if and only if there exist suitable signs s1 = 1 and s2 = 1
su h that c s1 u1 + s2 u2 + s1 s2 u3 (mod 27).
Remarks.
(1) We only assume that
v3 (c) > 0
and not
v3 (c) = 1,
although the ase
v3 (c) 2
is
overed by Lemma 5.9: indeed, it is easy to see ase (2) of the above lemma annot
o ur when
v3 (c) 2.
(2) It follows from the proof that ase (1) o urs if and only if there exists a solution
(X, Y, Z) with min(v3 (X), v3 (Y ), v3 (Y )) = 0 but
3.
si :
s2 u 1 u 3
one of the variables divisible by
(3) In ase (2), there is no need to sear h among the four possibilities for the signs
3|c
(mod 3).
sin e
it is easy to see that we must take
s1 u2 u3 (mod 3)
and
We have thus nished to give the lo al solubility of the general ubi equation
F (X, Y, Z) = u1 X 3 + u2 Y 3 + u3 Z 3 cXY Z = 0,
hen e in parti ular of the equation
of Theorem 3.1.
u1 X 3 + u2 Y 3 + (2b/(u1 u2 ))Z 3 2aXY Z = 0

6. Lo al Solubility : The Case
6.1.
D 6= 1
Redu tion of the Cubi , Bad Primes, and Split Primes.
solubility when
D = 1,
D 6= 1.
We now onsider lo al
Although we will do the same type of omputations as in the ase
there are evidently some added ompli ations.
19
3-des ent
It is essential to begin by a redu tion of the ubi equation of Theorem 4.1. Re all that
D = 1 we ould redu e to an equation where u = u21 u2 with the ui squarefree
in the ase
and oprime. We have seen in Se tion 4.2 that the analogous statement for
ideals, so we annot immediately reprodu e what we have done.
Re all that the ubi equation of Theorem 4.1 an be written as
if
v = v1 + v2 D
D 6= 1 involves
F (X, Y, Z) = 0,
where
we have
F (X, Y, Z) = 2v2 X 3 + 2Dv1 Y 3 + (2b/(v12 Dv22 ))Z 3
+ 6v1 X 2 Y + 6v2 DXY 2 + 2a(X 2 Z DY 2 Z)
= (v(X + Y D)3 (v)(X Y D)3 )/ D
+ 2aZ(X + Y D)(X Y D) + (2b/(v (v)))Z 3 ,
and we will use indierently both forms.
Lemma 6.1. If p 6= 3, vp (v (v)) = 0, vp (2b) = 0, and vp (27b 4a3 D) = 0 the above ubi
is soluble in Qp .
Re all that, analogously to ideals, an algebrai integer
ZK
with
nZ
if and only if
Lemma 6.2. Let [u]
is said to be primitive if
n = 1.
v/n
Im().
In the above ubi , we may assume that [u] = [v 2 (v)]

where v is a primitive algebrai integer su h that v (v) is divisible only by split primes. In
parti ular, v and (v) generate
oprime ideals. Furthermore, if D 0 (mod 4) we may
also assume that v = v1 + v2 D with v2 Z.
D = 1, we annot dedu e from this lemma that v (v) | 2b.
3 h(K), we may assume that v (v) | (2b)h(K) ,
and in parti ular all prime numbers dividing v (v) (whi h are ne essarily split) divide 2b,
but this may not be true if 3 | h(K).
Sin e v is now an algebrai integer, we have 2v1 Z and 2v2 Z, so all the oe ients
of the equation are integers, ex ept perhaps for 2b/(v (v)).
Corollary 6.3. Assume as above that v = v1 + v2 D is a primitive algebrai integer

su h that v (v) is divisible only by split primes, and let p be any split prime. There exists
dp Qp su h that d2p = D . The ubi of Theorem 4.1 has a solution in Qp if and only if
the ubi u1 X 3 + u2 Y 3 + u3 Z 3 cXY Z = 0 does, where u1 = v1 + v2 dp , u2 = v1 v2 dp ,
u3 = (2b/v (v))dp , and c = 2adp .
Note that, ontrary to the ase
It is easy to show using Corollary 4.3 that if
Sin e we have given a omplete algorithm to determine the solubility in Qp of an equation

3
3
3
of the type u1 X + u2 Y + u3 Z cXY Z = 0, this solves the problem for D 6= 1 in the
ase where
is a split prime.
Thus we are left with the study of ramied and inert
primes, so thanks to Lemma 6.2, we may assume that
vp (2b/(v (v))) 0.
vp (v (v)) = 0,
so that in parti ular
20
p is a ramied or inert prime, that vp (v (v)) = 0,

F (X, Y, Z) = 0 with
F (X, Y, Z) = (v(X + Y D)3 (v)(X Y D)3 )/ D + u3 Z 3
+ 2aZ(X + Y D)(X Y D) ,
Thus in the sequel we assume that

and re all that our equation is
with
u3 = 2b/(v (v)),
hen e su h that
vp (u3 ) 0.
We begin by inert primes.
p is an inert prime, onsider the eld Kp = Qp ( D),

whi h up to isomorphism is the unique unramied extension of degree 2 of Qp , and whose
D in Fp2 , and is dened on

residue eld is Fp2 , so that we an also onsider the lass of
Fp2 as in hara teristi 0. By abuse of notation, if and are elements of K = Q( D) or

of Kp , we will write (mod p) to mean that the lass of and in the residue eld
is the same. Note that we work in Kp or K for pra ti ality, but that the ubi equation
has oe ients in Q, and that we also look for solutions in Q.
6.2.
The Case of Inert Primes.
If
Before studying the bad primes, we need an auxiliary lemma.
Lemma 6.4. Let p 6= 3 be an inert prime. The following

onditions are equivalent:
(1)
(2)
(3)
There exist X and Y su h that (v)/v ((X + Y D)/(X Y D))3 (mod p) in

the above sense.
The lass of (v)/v is a ube in Fp2 .
2
We have either p 1 (mod 3), or p 2 (mod 3) and v (p 1)/3 1 (mod p).
Lemma 6.5. Let D 6= 1, and assume that the ellipti urve is given by an equation
satisfying the onditions of Lemmas 1.2 and 6.2. Let p be an inert prime number su h that
vp (2b) > 0, vp (v (v)) = 0 and p 6= 3, and if p = 2, assume that vp (2b) 2. The ubi of
Theorem 4.1 is lo ally soluble at p if and only if one of the following onditions is satised.
(1) vp (2a) = 0.
(2) vp (2a) > 0 and the lass of (v)/v modulo p is a ube in Fp2 .
p = 2, assumed to be inert, when v2 (2b) 3. Sin e the
2-redu ed, note that either v2 (a) > 0, in whi h
ase v2 (b) 2 hen e v2 (2b) = 3,
or v2 (a) = 0. Furthermore, we an write v = v1 + v2 D = (w1 + w2 D)/2 with w1 and
w2 in Z su h that w1 w2 (mod 2), and sin e v is primitive, either we have w1 w2 1
(mod 2), or w1 and w2 are even with w1 6 w2 (mod 4).
We now onsider the prime
equation is
satisfying the onditions of Lemmas 1.2 and 6.2, assume that p = 2 is an inert prime,
in other words that D 5 (mod 8), assume that v2 (2b) 3, and write w1 = 2v1 and
w2 = 2v2 .
(1) If w1 2 (mod 4) and w2 0 (mod 4) or w1 0 (mod 4) and w2 2 (mod 4)
the ubi has a solution in Q2 .
(2) If w1 w2 1 (mod 2), the ubi has a solution in Q2 if and only if either
v2 (2b) 4 or v2 (a) > 0.
21
3-des ent
satisfying the onditions of Lemmas 1.2 and 6.2. Let p be an inert prime number su h that
vp (2b) = vp (v (v)) = 0, vp (27b 4Da3 ) > 0, and p 6= 3. The ubi of Theorem 4.1 is
lo ally soluble at p if and only if (v)/v is a ube in Fp2 .
Re all that sin e we assume that
is not split we have
vp (v (v)) = 0.
Lemma 6.8. Let D 6= 1, assume that the ellipti urve is given by an equation satisfying
the onditions of Lemmas 1.2 and 6.2, and assume that p = 3 is an inert prime, i.e., that
D 2 (mod 3). Set u1 = 2v2 , u2 = 2v1 D , and u3 = 2b/(v (v)).
(1) If v3 (2a) = 0 the ubi has a solution in Q3 .
(2) If v3 (2a) 2 the ubi has a solution in Q3 if and only if either v3 (u1 ) 2, v3 (u2 )
2, ui uj (mod 9) for some i 6= j and a suitable sign , or if u3 2(u1 u2 )
(mod 9) for suitable signs .
(3) If v3 (2a) = 1, v3 (2b) > 0, and v3 (u1 u2 ) 1, the ubi has a solution in Q3 if and
only if either v3 (u1 ) 2, v3 (u2 ) 2, or v3 (2a + 2b) = 1.
(4) If v3 (2a) = 1, v3 (2b) > 0, and v3 (u1 u2 ) = 0, the ubi has a solution in Q3 if and
only if either u1 u2 (mod 9), or if 2a + b u1 u2 (mod 9) for suitable signs
.
Lemma 6.9. Let D 6= 1, assume that the ellipti urve is given by an equation satisfying
the onditions of Lemmas 1.2 and 6.2, and assume that p = 3 is an inert prime, i.e., that
D 2 (mod 3). Set u1 = 2v2 , u2 = 2v1 D , and u3 = 2b/(v (v)). Assume that v3 (2a) = 1
and v3 (u1 ) = v3 (u2 ) = v3 (2b) = 0, and set u4 = u3 + 2a, so that we also have v3 (u4 ) = 0.
(1) If ui uj (mod 9) for some i 6= j with i, j = 1, 2, or 4, the ubi has a solution
in Q3 .
(2) Otherwise, the ubi has a solution in Q3 if and only if 2a + u4 = 4a + u3 3D
(mod 27) for a suitable sign .
Similar remarks as those given after Lemma 5.10 apply here.
This on ludes the study of lo al solubility in the ase of inert primes.
The Case of Ramied Primes.

6.3.
satisfying the onditions of Lemmas 1.2 and 6.2. If p is a ramied prime su h that p 6= 3,
the ubi is soluble in Qp .
satisfying the onditions of Lemmas 1.2 and 6.2. Assume that p = 3 is ramied, in other
words that 3 | D , and to simplify notation, set u3 = 2b/(v (v)).
(1) If v3 (2a) = 0, the ubi has a solution if and only if one of the following onditions
is satised:
(a) v3 (2v2 ) > 0.
(b) v3 (2v2 ) = v3 (2a + u3 ) = 0.
22
(2)
If v3 (2a) 2, the ubi has a solution if and only one of the following onditions
is satised:
(a) D 3 (mod 9) and v3 (u3 ) = 0.
(b) D 3 (mod 9), v3 (u3 ) > 0, and v3 (2v2 ) > 0.
( ) D 6 (mod 9) and v3 (2v2 ) 2.
(d) D 6 (mod 9) and v3 (2v2 ) = v3 (u3 ) = 1.
(e) D 6 (mod 9), v3 (2v2 ) = 0, and u3 2v2 (mod 9).
(f ) u3 2v1 D (mod 27).
We will treat the ase where
v3 (2a) = 1
below.
Lemma 6.12. Keep the notation and assumptions of the pre eding lemma, assume now
that v3 (2a) = 1, and set u4 = u3 + 2a. The ubi has a solution if and only one of the
following onditions is satised:
(a) D 3 (mod 9) and v3 (u4 ) = 0.
(b) D 3 (mod 9), v3 (u4 ) > 0, and v3 (2v2 ) > 0.
( ) D 6 (mod 9) and v3 (2v2 ) 2.
(d) D 6 (mod 9) and v3 (2v2 ) = v3 (u4 ) = 1.
(e) D 6 (mod 9), v3 (2v2 ) = 0, and u4 2v2 (mod 9).
(f ) v3 (u3 ) = 1 and there exists s = 1 su h that 2v1 (D/3) s(u3 /3 2a(D/3))
(mod 9) and 2v1 s 6 2a/3 (mod 3).
(g) v3 (u3 ) = 1 and there exists s = 1 su h that 2v1 (D/3) s(u3 /3 2a(D/3))
(mod 9), 2v1 s 2a/3 (mod 3), v3 (2v2 ) = 0, and D 3 (mod 9).
(h) v3 (u3 ) = 1 and there exists s = 1 su h that 2v1 (D/3) s(u3 /3 2a(D/3))
(mod 27), 2v1 s 2a/3 (mod 3), v3 (2v2 ) = 0, and D 6 (mod 9).
(i) v3 (u3 ) = 1 and there exists s = 1 su h that 2v1 s 2a/3 (mod 3), v3 (2v2 ) > 0,
and there exists t {1, 0, 1} and r {1, 0, 1} su h that
2v1 (D/3) s(u3 /3 2a(D/3)) 6v2 (D + 3)t 9(2v1 s 2a/3)st2
3r(D(2v1 + a/3) + 6rDv1 + 6at2 ) (mod 81) .
This nishes the study of lo al solubility in the ase of ramied primes, hen e the se tion
on lo al solubility.
7. Examples
7.1.
The Curves y 2 = x3 + (kp)2 for k = 1, 2, or 4.

Ekp
on k
family of urves
with equation
y 2 = x3 + (kp)2 ,
where
In this se tion we onsider the
is a prime and
k = 1, 2 ,
or
4.
is made so that no other prime apart from 2 divides it. Note that it
2
3
2
is not ne essary to onsider higher powers of 2 sin e the urve y = x + (8kp) is trivially
2
3
2
isomorphi to the urve y = x + (kp) . Furthermore the primes p = 2 and 3 give rise
The restri tion
p = 2 the
0, for (k, p) = (1, 3) and (2, 3) the rank is 1, MordellWeil generators being
(2, 1) and (3, 3) respe tively, and for (k, p) = (4, 3) the rank is again equal to 0, and the
torsion is always of order 3 generated by T = (0, kp), ex ept for (k, p) = (4, 2), for whi h it
to a nite number of urves whi h an be treated individually (spe i ally, for
rank is equal to
23
3-des ent
has order
6 generated by (8, 24)). We therefore assume that p 5, so that in parti ular all
3-torsion generated by T = (0, kp) equal to their full rational
of these urves have rational

torsion subgroup.
We rst ompute the image of . For this, we onsider the ubi equations of Theorem
u1 X 3 + u2 Y 3 + u3 Z 3 = 0, where u1 u2 | 2kp and u3 = 2kp/(u1u2 ),
3.1 (3), in other words

where we re all that
u1 u2
u1 and u2 , it is easy to he k
(1, 1, 2kp), (1, 2, kp), (1, p, 2k), (1, 2p, k), and (2, p, k). The
u = 1) gives an evidently soluble equation, orresponding to
is squarefree. Up to ex hange of
that the only of possibilities are

rst one ( orresponding to
the unit element of the ellipti urve.
u = 2p and u = 4p2 ) is also soluble,

and it orresponds to the two nontrivial rational 3-torsion points on the urve, and
2
2
the other three ( orresponding to u = 2, 4, p, p , 4p, and 2p , are equivalent.
When k = 2, the fth one ( orresponding to u = 4p and u = 2p2 ) is also soluble,
and it again orresponds to the two nontrivial rational 3-torsion points on the urve,
2
the se ond and fourth ( orresponding to u = 2, 4, 2p, and 4p are equivalent, and
2
the third orresponds to u = p and p . Sin e the set (of lasses) of u for whi h the
2
ubi is soluble forms a group and sin e 4p and 2p belong to this group, it follows
2
that [p] and [p ] will be in the group if and only if 4 and 2 are, so in fa t the three
When
k = 1,
the fourth one ( orresponding to
equations are equivalent, although slightly less trivially.
k = 4 the third equation ( orresponding to p and p2 ) is learly soluble

(sin e 2k = 8 is a ube), and this again orresponds to the two rational 3-torsion
points. The other equations orrespond respe tively to u = 2 and 4, u = 2p and
4p2 , and u = 4p and 2p2 , and on e again be ause of the group stru ture all the
Finally, when
equations are in fa t equivalent.

We see that in ea h ase it is su ient to onsider the equation with
hen e
u3 = kp.
u 1 = 1, u 2 = 2,
The result is as follows:
Lemma 7.1. Keep the above assumptions. The equation X 3 + 2Y 3 + kpZ 3 = 0 is ELS if
and only if k 6= 4, either p 2 (mod 3) or 2(p1)/3 1 (mod 3), and kp 6 4 (mod 9).
Proof.
This of ourse immediately follows from the study of the equation that we have
made above. More pre isely, if

so the equation has no
2-adi
k=4
the
2-adi
(0, 1, 2),
k = 1 or
valuations of the oe ients are
solutions by Lemma 5.1. On the other hand, if
k = 2, the 2-adi valuations are (0, 0, 1) and (0, 1, 1) respe tively, and sin e all elements
F2 are ubes, we on lude by Lemma 5.4 that the equation has a solution in Q2 . For
p-adi solubility we also use this lemma, sin e the p-adi valuations are (0, 0, 1), and we
on lude that the equation has a p-adi solution if and only if 2 is a ube in Fp , leading to
the given ondition. Finally, sin e the 3-adi valuations are (0, 0, 0) we use Lemma 5.9 (1),
whi h tells us that the equation has a 3-adi solution if and only if ui uj (mod 9) for
some i 6= j , whi h gives kp 1 or 2 (mod 9), in other words kp 6 4 (mod 9) sin e
3 kp.
of
24
Corollary 7.2. Let p 5 be prime, let k = 1, 2, or 4, and let E be the ellipti urve
y 2 = x3 + (kp)2 .
(1)
(2)
(3)
(4)
(5)
For k = 1, if either p 4 (mod 9) or if p 1 or 7 (mod 9) and 2(p1)/3 6 1

(mod p) then Im() = {1, 2p, 4p2}, and in parti ular |Im()| = 3.
For k = 1 and p 2 (mod 9) we have |Im()| = 9.
For k = 2, if either p 2 (mod 9), or if p 1 or 4 (mod 9) and 2(p1)/3 6 1
(mod p) then Im() = {1, 4p, 2p2}, and in parti ular |Im()| = 3.
For k = 4 we always have Im() = {1, p, p2 }, and in parti ular |Im()| = 3.
In all other ases we have |Im()| = 3 or 9. More pre isely the ubi equation
X 3 + 2Y 3 + kpZ 3 = 0 is ELS, and |Im()| = 9 if and only if it is globally soluble.
Proof.
(1), (3), (4), and (5) are lear from the lemma by inspe tion. For (2), we use the
proposition 3.3 p. 438 of Satg [17.
Remark.
In [15, Rodriguez-Villegas and Zagier have hara terized the primes whi h are
sums of two ubes. If their method ould be extended to primes whi h are of the form
x3 + 2y 3, and also of the form x3 + 4y 3, it would determine Im() in all ases.
We now ompute the image on the dual urve
D = 3 and
b = 3kp. We rst determine lo al solubility of the equation orresponding
= (1 + 3)/2, and for the moment we do not ne essarily assume that k | 4.
that
to
b, whose equation is y 2 = x3 27(kp)2 , so

E
Lemma 7.3.
(2)
(1) Let k be su h that 8 k . The equation orresponding to is lo ally

soluble at the primes 2, 3, and p if and only if p 1 (mod 9), k 4 (mod 9),
and 4 | k .
In parti ular, if k = 1, 2, or 4, the equation orresponding to is ELS if and only
if k = 4 and p 1 (mod 9).
Proof.
We have
2v1 = 1, 2v2 = 1,
and
2b = 6kp,
so
u3 = 2b/(v (v)) = 6kp.
The prime
being inert, by Lemma 6.5, if 4 k the equation is lo ally soluble at 2 if and only if is
a ube in F4 , whi h is not the ase sin e the only ube is 1. On the other hand, if 4 | k
Lemma 6.6 tells us that the equation is lo ally soluble at 2. Let us now look at the prime
p.
p 1 (mod 3) then p is split, so by Corollary 6.3 the equation is lo ally soluble at p

(v1 + v2 dp )X 3 + (v1 v2 dp )Y 3 + 6kpZ 3 = 0 is, and sin e the p-adi valuations
2
are (0, 0, 1), by Lemma 5.4 this is true if and only if (v1 + v2 dp )/(v1 v2 dp ) / =
(p1)/3
1 (mod p), whi h is the ase if and
(mod p) is a ube in Fp , hen e if and only if
only if p 1 (mod 9). If p 2 (mod 3) then p is inert, so by Lemma 6.5 the equation is
2
lo ally soluble at p if and only if is a ube in Fp2 , hen e if and only if p 1 (mod 9),
in other words p 1 (mod 9) sin e we assume p 2 (mod 3). It follows that the lo al
ondition at p is p 1 (mod 9). Finally, let us look at the prime 3. Sin e 2a = 0 we
use Lemma 6.11 (2), whi h tells us that the equation is lo ally soluble at 3 if and only if
6kp 3 (mod 27), or equivalently kp 4 (mod 9), proving (1) sin e p 1 (mod 9),
and (2) follows immediately.

If
if and only if
25
3-des ent
p 1 (mod 3). In this ase we an write p = () with

= (w1 + w2 3)/2 in 12 dierent ways, and it is well-known and easy that up to sign
and ex hange of and () there is exa tly one su h de omposition with 3 | w2 .
Lemma 7.4. Let p 1 (mod 3), let = (w1 +w2 3)/2 be su h that () = p, and let k
be su h that 8 k , 3 k , and p k . The equation orresponding to is lo ally soluble at the
primes 2, 3, and p if and only if either 4 | k or 4 k and 2 | w2 Z, if (w1 /(2k))(p1)/3 1
(mod p), and if either 3 | w2 , or 3 w2 and p k 2 + 3 (mod 9).
Next, we assume that
Proof.
We have
2v1 = w1 , 2v2 = w2 ,
and
u3 = 2b/(v (v)) = 6k .
The prime
being
inert, as above Lemmas 6.5 and 6.6 tell us that the equation is lo ally soluble at 2 if
4 | k , or 4 k and (w1 + w2 3)/2 = 1 in F4 , whi h is equivalent to

2 | w2 . By Corollary 6.3 the equation is lo ally soluble at p if and only if (v1 + v2 dp )X 3 +
(v1 v2 dp )Y 3 + 6kdp Z 3 = 0 is. Sin e v12 v22 d2p = p, we may assume for instan e that
dp is hosen so that vp (v1 v2 dp ) = 1, so in parti ular v2 dp v1 (mod p). The p-adi
and only if either
(0, 1, 0), so by Lemma 5.4 lo al solubility is equivalent
to (v1 + v2 dp )/(6kdp ) being a ube in Fp , and sin e v1 v2 dp (mod p), this means that
v1 /(3kdp ) = w1 /(6kdp ) is a ube in Fp . This is equivalent to (w1 /(6kdp ))2 = w12 /(108k 2)
2
being a ube, hen e to (w1 /(2k)) being a ube, hen e to w1 /(2k) being a ube, leading
to the given ondition. Finally, let us look at the prime 3. Sin e 2a = 0 we use Lemma
6.11 (2), whi h tells us (sin e 3 k , so that v3 (u3 ) = 1) that the equation is lo ally soluble
at 3 if and only if either 3 | w2 , or if 6k 3w1 (mod 27), in other words 2k w1
(mod 9). However sin e w12 + 3w22 = 4p, if 3 w2 we have w12 4p 3 (mod 9), and
2
2
sin e the ondition 2k w1 (mod 9) is equivalent to w1 4k (mod 9) (sin e 3 w1 ) we
2
2
obtain the equivalent ondition 4p 4k + 3 (mod 9), or equivalently p k + 3 (mod 9),
nishing the proof of the lemma.
valuations of the oe ients are thus
Corollary 7.5. Let p 5 be prime, let k = 1, 2, or 4, and let Eb be the ellipti urve
y 2 = x3 27(kp)2 .
(1)
(2)
(3)
For k = 1 or k = 2, if either p 2 (mod 3), or if p 1 (mod 3), p 6 k 2 + 3

) is trivial.
(mod 9), and 2(p1)/3 6 1 (mod p) then Im(b
For k = 4, if p 2 or 5 (mod 9) then Im(b
) is trivial,
Otherwise, |Im(b
)| = 1 or 3 when k = 1, k = 2, or k = 4 and p 8 (mod 9), and
|Im(b
)| = 1, 3, or 9 when k = 4 and p 1 (mod 3).
Proof.
2b = 6kp, 2 is inert, and 3 is ramied, with the notation of Se tion 4.2, we

f1 = 1 if p 2 (mod 3) and f1 = 1 or p if p 1 (mod 3). In the rst ase,
2
the only possible v are 1, , and , while in the se ond ase we have in addition the three
possible (up to sign and onjugation) su h that () = p. It follows that:
Sin e
must have
/ Im(b
) and none of the three possible is in Im(b
) then Im(b
) = {1}, so
|Im(b
)| = 1,
If 6 Im(b
) and one of the three possible (so ne essarily exa tly one) is in Im(b
)
then Im(b
) = {1, , ()}, hen e |Im(b
)| = 3.
If
26
If
Im(b
) and none of the three possible (in this ase they are equivalent) is
= {1, , 2 }, hen e |Im(b
)| = 3.
in Im(b
) then Im(b
)
Im(b
) and one (hen e all) of the three possible
{j , j , j (), 0 j 2}, hen e |Im(b
)| = 9.
If
are in Im(b
) then Im(b
)
Sin e in the two pre eding lemmas we have studied lo al solubility of the equation in all
these ases, we on lude by inspe tion.

We an say a little more:
Proposition 7.6.
(2)
(3)
(1) Assume that k = 1 and p 4 (mod 9), or that k = 2 and p 7

(mod 9), and write p = m2 + 3n2 , where m and n are
integers whi h are unique
up to sign. The equation orresponding to = m + n 3 (i.e., with v1 = m and
v2 = n) is ELS, and |Im(b
)| = 3 if and only if it is globally soluble.
Assume that k = 4 and p 8 (mod 9). The equation orresponding to is ELS,
and |Im(b
)| = 3 if and only if it is globally soluble.
Assume that k = 4 and p 4 or 7 (mod 9), and write 4p = m2 + 27n2 , where
m
and n are unique up to sign. The equation orresponding to = (m + 3n 3)/2

(i.e., with v1 = m/2 and v2 = 3n/2) is ELS, and |Im(b
)| = 3 if and only if it is
globally soluble.
Proof.
Apply the same method as above.
We will see below that it follows from BSD that these equations (of Proposition 7.6)
should in fa t always be globally soluble.
Corollary 7.7.
(2)
(3)
If p 5 (mod 9), or p 1 or 7 (mod 9) and 2(p1)/3 6 1 (mod p),

the ellipti urve Ep with equation y 2 = x3 + p2 has rank 0. If p 2 (mod 9), it
has rank 1. Otherwise, if p 4 or 8 (mod 9) it has rank 0 or 1, and if p 1 or 7
(mod 9) it has rank 0, 1, or 2.
If p 2 (mod 9), or p 1 or 4 (mod 9) and 2(p1)/3 6 1 (mod p), the ellipti
urve E2p with equation y 2 = x3 + 4p2 has rank 0. Otherwise, if p 5, 7, or 8
(mod 9) it has rank 0 or 1, and if p 1 or 4 (mod 9) it has rank 0, 1, or 2.
If p 2 or 5 (mod 9) the ellipti urve E4p with equation y 2 = x3 + 16p2 has rank
0. Otherwise, if p 4, 7, or 8 (mod 9) it has rank 0 or 1, and if p 1 (mod 9) it
has rank 0, 1, or 2.
Proof.
(1)
Clear, sin e |Im()||Im(b

)|
= 3r+1 .
(No need of BSD here.)
k = 1 for p = 61,
79, 113, 131, 149, 151, 163, 293, et ..., with k = 2 for p = 29, 83, 137, 139, 173, 181, 199,
et ..., and with k = 4 for p = 41, 59, 101, 131, 137, et ... for whi h mwrank, at least in its
basi version, is not able to determine the rank using 2-des ent.
This orollary allows us to determine the rank exa tly for instan e with
Remarks.
(1) We an use the parity onje ture in this ontext, see for example [10 and [11,
in other words the analyti rank has the same parity as the algebrai rank, then
27
3-des ent
0 or 1 then it is always
1, while when the rank is known to be equal to 0, 1, or 2 then it is always
equal to 0 or 2, and both ases o ur. This has been proved in ertain ases: as
already mentioned, by Satg for k = 1 and p 2 (mod 9), and in an unpublished
work Elkies has shown that for k = 4 and p 4 or 7 (mod 9) the rank is indeed
equal to 1.
The ase k = 4 orresponds to primes whi h are sums of two ubes, so by [15 one
knows that when p 1 (mod 9) the rank is equal to 0 or 2, and exa tly for whi h
primes it is equal to 2. It is possible that either their method or Elkies' an be
extended to the ases k = 1 and k = 2.
The result for k = 4 an also be proved, less naturally, using 2-des ent; see Theorem
whenever in the above the rank is known to be equal to
equal to
(2)
(3)
6.4.17 of [4.
When the ubi s are ELS, we may of ourse try to look for a global solution by sear h. A
very e ient way of looking for rational points on a homogeneous ubi has been des ribed
by N. Elkies in [12, see also an unpublished preprint of J. Cremona on the subje t.
It
has been implemented by several people. Using a slightly modied implementation due
to M. Watkins, we an for instan e nd that a generator P = (x, y) of the MordellWeil
2
3
2
group of y = x + p for p = 1759, whi h has rank 1, is given by
242479559514608433100075350499874221113923535
3063551062176562878606796987394973602467684
8643240396318605197724619647046515784779281219388876514209037894857
y=
5362134274928159502186511847328850266140274118035321166956948248
This generator is not found by mwrank even at a high sear h limit. On the other hand
x=
it ould ertainly be found using the Heegner point method.

For a more ompli ated example, for
rank
2,
p = 9511
y 2 = x3 + p2 has analyti
mwrank even at a high sear h
the urve
so the Heegner point method is not appli able, and
limit nds only the one-dimensional subspa e of the (free part of the) MordellWeil group
P1 = (210, 9011). Using our

(P1 , P2 ) with P2 = (x, y), where
generated by
has basis
implementation, we nd that the full free part
32701984517186448621442294824950874787830128281
456289760665179363242981599270033206574137600
92890043770264171014255964610503972850176417273682124237369198272789821
y=
9746778232027925565271633950191532413151456450450966045051557376000
x=
In the following tables, we summarize what is proved (either using
3-des ent
as above,
by Satg, in Elkies' unpublished work, or Rodriguez-VillegasZagier's work), what is a

onsequen e of BSD, and what remains to be done. The tables are oded as follows. In the
rst olumn we indi ate the residue of p modulo 9, and if relevant, in the se ond olumn

2
(p1)/3
we indi ate the ubi hara ter
of 2 modulo p, 1 meaning that 2
1 (mod p),
p 3
2
(p1)/3
6 1 (mod p). In the third, fourth, and fth olumn
and , meaning of ourse 2
28
we give |Im()|, |Im(b

)|, and the rank of the urve respe tively, and when two values are
given, both o ur. In the last olumn, we give a pair of symbols (A,B), orresponding to
(|Im()|, |Im(b
)|),
where P means proved, BSD means proved under BSD, S means Satg,
ELK means Elkies, RVZ means Rodriguez-VillegasZagier, and U means unknown.
p
1
1
2
4
5
7
7
8
mod 9
2
p 3
1
, 2
1
, 2
|Im()|
9 or 3
3
9
3
3
9 or 3
3
9
Curves
p
1
1
2
4
4
5
7
8
mod 9
2
p 3
1
, 2
1
, 2
p
1
2
4
5
7
8
mod 9
2
p 3
2
0
1
1
0
2
0
1
or 0
proved
(U,U)
(P,P)
(S,P)
(P,BSD)
(P,P)
or 0
(U,U)
(P,P)
(BSD,P)
|Im(b
)|
3 or 1
1
1
3 or 1
1
1
3
1
rank
2
0
0
2
0
1
1
1
or 0
proved
(U,U)
(P,P)
(P,P)
or 0
(U,U)
(P,P)
(BSD,P)
(P,BSD)
(BSD,P)
y 2 = x3 + 4p2 , p 5
|Im()|
3
3
3
3
3
3
Curves
rank
y 2 = x3 + p2 , p 5
|Im()|
9 or 3
3
3
9 or 3
3
9
3
9
Curves
|Im(b
)|
3 or 1
1
1
3
1
3 or 1
1
1
|Im(b
)|
9 or 1
1
3
1
3
3
rank
2
0
1
0
1
1
or 0
proved
(P,RVZ)
(P,P)
(P,ELK)
(P,P)
(P,ELK)
(P,BSD)
y 2 = x3 + 16p2 , p 5
An immediate orollary of the above tables and of Corollary 7.2 and Proposition 7.6 is
the following:
Corollary 7.8.
Assume BSD. If p 2 or 8 (mod 9) there exist x and y in Q su h

that p = x + 2y 3, and if p 5 or 8 (mod 9) there exist x and y in Q su h that
p = x3 + 4y 3.
(1)
29
3-des ent
(2)
Assume that either k = 1 and p 4 (mod 9) or that k = 2 and p 7 (mod 9), and
write p = m2 + 3n2 . If BSD is true the equation
nX 3 3mY 3 + 3kZ 3 + 3mX 2 Y 9nXY 2 = 0
(3)
is globally soluble.
Assume that p 8 (mod 9). If BSD is true the equation
X 3 + 3Y 3 + 24pZ 3 3X 2 Y 9XY 2 = 0
(4)
Assume that p 4 or 7 (mod 9), and write 4p = m2 + 27n2 . Without any assumption the equation
nX 3 mY 3 + 8Z 3 + mX 2 Y 9nXY 2 = 0
Proof.
Clear, sin e these orrespond respe tively to |Im()|
under BSD for (2) and (3), and to |Im(b

)|
7.2.
=3
The Curves y 2 = x3 + (kp)2 for k = 3 or 9.
so that no other prime apart from

powers of
3.
Furthermore the primes
= 9 under BSD, |Im(b

)| = 3
by Elkies's result.
On e again the restri tion on
is made
divides it, and it is not ne essary to onsider higher
p = 2 and p = 3 give rise
to a nite number of urves
whi h an be treated individually (spe i ally, the rank is zero unless
(k, p) = (3, 2),
in
whi h ase it has rank 1, a MordellWeil generator being (3, 3), and the torsion is of order
3 generated by T = (0, kp), unless (k, p) = (9, 3), in whi h ase it has order 6 generated by
(18, 81)). We therefore assume that p 5.
As before, we rst ompute the image of . For this, we onsider the ubi equations
3
3
3
of Theorem 3.1 (3), in other words u1 X + u2 Y + u3 Z
= 0, where u1 u2 | 2kp and
u3 = 2kp/(u1u2 ), where we re all that u1 u2 is squarefree. Up to ex hange of u1 and
u2 , it is easy to he k that the only of possibilities are (1, 1, 2kp), (1, 2, kp), (1, 3, 2kp/3),
(1, 6, kp/3), (1, p, 2k), (1, 2p, k), (1, 3p, 2k/3), (1, 6p, k/3), (2, 3, kp/3), (2, p, k), (2, 3p, k/3),
(3, p, 2k/3), (3, 2p, k/3), and (6, p, k/3). The rst one ( orresponding to u = 1) gives an
evidently soluble equation, orresponding to the unit element of the ellipti urve.
Consider rst k = 3. We obtain the equations (1, 2, 3p), (1, 3, 2p), (1, 6, p), (1, p, 6),
(1, 2p, 3), (1, 3p, 2), (1, 6p, 1), (2, 3, p), (2, p, 3), (2, 3p, 1), (3, p, 2), (3, 2p, 1), and (6, p, 1).
2
The equation (1, 6p, 1) ( orresponding to u = 6p and u = (6p) ) orresponds to the two
3-torsion points of the urve. Apart from that, up to permutation of the ui we have to
2
2
study solubility for (1, 2, 3p) ( orresponding to u = 2, 4, 3p, 9p , 12p, and 18p ), (1, 3, 2p)
2
2
( orresponding to u = 3, 9, 2p, 4p , 18p, and 12p ), (1, 6, p) ( orresponding to u = 6, 36,
p, p2 , 36p, and 6p2 ), (2, 3, p) ( orresponding to u = 12, 18, 4p, 2p2 , 9p, and 3p2 ).
Lemma 7.9. We have two ases :

(1)
If p 2 (mod 3) all the above ubi s are ELS, giving a total of 27 ELS ubi s.
30
(2)
If p 1 (mod 3), then either both 2 and 3 are ubes in Fp , in whi h ase on e
again all the above ubi s are ELS for a total of 27, or either 2 or 3 or both are
non ubes, in whi h ase only 9 equations are ELS.
Proof.
Using the lemmas that we have proved it is immediate to show that all the
equations are soluble at
and
3.
p, and by Lemma 5.4 the

2, 3, 6, and 3/2 respe tively are
The only problem is at
equations are also soluble at p if and only if the lasses of

Fp , and if p 2 (mod 3) this is trivially true.
Assume now that p 1 (mod 3). We onsider four ases, a ording to the ubi residue
ubes in
hara ter of
(3)
and
modulo
p.
Fp , then either their produ t or their quotient is a ube,

so we dedu e that either (1, 6, p) or (2, 3, p) (but not both) is lo ally soluble, giving
a total of 6 + 3 = 9 possible values of u.
If 2 or 3 is a ube in Fp but not both, then 6 and 3/2 annot be ubes, so we dedu e
that either (1, 2, 3p) or (1, 3, 2p) is lo ally soluble, giving again 9 possible values of
u.
If both 2 and 3 are ubes in Fp then all the equations are lo ally soluble, giving a
total of 24 + 3 = 27 possible values of u, proving the lemma.
(1) If
(2)
2 and 3 are non ubes
in
Corollary 7.10.
(2)
(3)
(1) If p 2 (mod 3) then |Im()| = 3, 9, or 27 and assuming BSD

we have |Im()| = 3 or 27 and the following are equivalent:
(a) |Im()| = 27,
(b) rk(E) = 2,
3
3
( ) There exist x and y in Q su h that p = x + 6y ,
3
3
(d) There exist x and y in Q su h that p = 2x + 3y ,
3
3
(e) There exist x and y in Q su h that p = 4x + 12y ,
3
3
(f ) There exist x and y in Q su h that p = 9x + 18y .
If p 1 (mod 3) and 2 and 3 are not both ubes in Fp then |Im()| = 3 or 9, and
assuming BSD we have |Im()| = 9 and rk(E) = 1, and furthermore for (a, b) =
(1, 6), (2, 3), (4, 12), or (9, 18), there exist x and y in Q su h that p = ax3 + by 3 if
and only if b/a is a ube in Fp .
If p 1 (mod 3) and 2 and 3 are both ubes in Fp then either C0 is globally
soluble, in whi h ase the six onditions of (1) are again equivalent ex ept that
the se ond must be repla ed by rk(E) = 3, or C0 is not globally soluble, in whi h
ase |Im()| = 9 and rk(E) = 1, and exa tly one of the equations p = x3 + 6y 3,
p = 2x3 + 3y 3, p = 4x3 + 12y 3, or p = 9x3 + 18y 3 has a solution with x and y in Q.
Proof.
The assertions independent of BSD follow immediately from the above lemma.
On the other hand, we have that if either p 2 (mod 3) or p 1 (mod 3) and 2 and 3 are
not both ubes in Fp we have |Im(b

)| = 1, so that 3rk(E)+1 = |Im()|. If p 2 (mod 3)
the root number of E is equal to +1, so assuming BSD the rank of E is even, so we have
|Im()| = 3 or 27, and the result learly follows in this ase. If p 1 (mod 3) the root
number of E is equal to 1, so assuming BSD the rank of E is odd, so we have |Im()| = 9
3-des ent
31
2 and 3 are not both ubes, so the result also follows. Finally, if p 1 (mod 3) and 2
3 are both ubes, then |Im(b
)| = 1 or 3, and it is equal to 3 if and only if C0 (whi h is
ELS) is globally soluble. Assuming BSD the rank of E is again odd, so we have two ases:
If C0 is not globally soluble we again have |Im(b
)| = 1, so we must have |Im()| = 9
and rk(E) = 1.
If C0 is globally soluble we have |Im(b
)| = 3, so we must have either |Im()| = 3 and
rk(E) = 1, or |Im()| = 27 and rk(E) = 3.
when
and
p for whi h
mwrank program of Cremona tells us that
to 1, and on the other hand (X, Y, Z) = (91, 211, 19) is a solution of
so |Im()| 9, hen e by the above if C0 was globally soluble we would
Note that although

it is not is
p = 3889,
the rank is equal
C 0
is ELS, it is not always globally soluble: the smallest
for whi h the
2-rank
based
(2, 3, p) ubi
have rk(E) = 3, whi h is not the ase.
Consider now k = 9. We obtain the equations (1, 2, 9p), (1, 3, 6p), (1, 6, 3p), (1, p, 18),
(1, 2p, 9), (1, 3p, 6), (1, 6p, 3), (2, 3, 3p), (2, p, 9), (2, 3p, 3), (3, p, 6), (3, 2p, 3), and (6, p, 3).
2
The equation (3, 2p, 3) ( orresponding to u = 18p and u = 12p ) orresponds to the two 3torsion points of the urve. Apart from that, up to permutation of the ui we have to study
solubility for (1, 2, 9p) ( orresponding to u = 2 and 4), (1, 3, 6p) ( orresponding to u = 3,
9, 6p, and 36p2 ), (1, 6, 3p) ( orresponding to u = 6, 36, 3p, 9p2 ), (1, 9, 2p) ( orresponding
2
2
to u = 2p and 4p ), (1, 18, p) ( orresponding to u = p and p ), (2, 3, 3p) ( orresponding
2
2
to u = 12, 18, 12p, and 18p ), (2, 9, p) ( orresponding to u = 4p and 2p ), and (3, 6, p)
2
2
( orresponding to u = 9p, 3p , 36p, and 6p ).
On e again, the equations are all lo ally soluble at 2. However, they are not all lo ally
soluble at 3 (in fa t (3, 6, p) is never lo ally soluble at 3), and using on e again the lo al
the
solubility results that we have proved, we obtain the following lemma:
Lemma 7.11.
(2)
If p 2 (mod 3), exa tly two of the 7 above equations are ELS,
giving always a total of 9 values of u.
If p 1 (mod 3) and 3/2, 3, or 6 are ubes respe tively for p 1, 4, or 7 (mod 9),
on e again exa tly two of the 7 above equations are ELS, giving always a total of 9
values of u. Otherwise, none are ELS, giving a total of 3 values of u ( orresponding
to the 3-torsion points).
(1)
Proof. The seventh equation is not lo ally soluble at 3, and the six others are ELS if and
only if p 4 (mod 9) and 3 is a ube for (1, 3, 6p) and (1, 9, 2p), p 2 (mod 9) and 6
is a ube for (1, 6, 3p) and (2, 9, p), or p 1 (mod 9) and 3/2 is a ube for (1, 18, p) and
(2, 3, 3p), and the result follows.
Bibliography
[1
Cassels, J. W. S.,
Arithmeti on urves of genus 1. I. On a onje ture of Selmer. J. Reine Angew.
Math., 202 (1959), 5299.
32
[2
[3
[4
[5
[6
[7
[8
[9
[10
[11
[12
[13
[14
[15
[16
[17
[18
[19
[20
Cohen, H., A ourse in omputational algebrai number theory. Graduate Texts in Mathemati s,
Springer (New York), 138 (1993).
Cohen, H., Advan ed topi s in omputational number theory. Graduate Texts in Mathemati s,
Cohen, H., Number theory. Vol. I. Tools and Diophantine equations. Graduate Texts in Mathemati s,
Cohen, H., Number theory. Vol. II. Analyti and modern tools. Graduate Texts in Mathemati s,
Cremona, J. E. and Fisher, T. A. and O'Neil, C. and Simon, D. and Stoll, M., Expli it
n-des ent on ellipti urves. I. Algebra. J. reine angew. Math., 615 (2008), 121-155.
Cremona, J. E. and Fisher, T. A. and O'Neil, C. and Simon, D. and Stoll, M., Expli it
n-des ent on ellipti urves. II. Geometry. J. reine angew. Math., (a epted in 2008).
DeLong, M., A formula for the Selmer group of a rational three-isogeny. A ta Arithmeti a, 105
(2002), 119131.
Djabri, Z. and S haefer, E. F. and Smart, N. P., Computing the p-Selmer group of an ellipti
urve. Trans. Amer. Math. So ., 352 (2000), 55835597.
Dok hitser, T. and Dok hitser V., Parity of ranks for ellipti urves with a y li isogeny. J.
Number Theory, 128 (2008), 662-679.
Dok hitser, T. and Dok hitser V., On the Bir h-Swinnerton-Dyer quotients modulo squares. to
appear in Annals of Math.
3
2
Elkies, N. D., Rational points near urves and small nonzero |x y | via latti e redu tion. Algorithmi number theory (Leiden), Le ture Notes in Comput. S i. 1838 (2000), 3363.
Fisher, T., Finding rational points on ellipti urves using 6-des ent and 12-des ent. J. Algebra,
320 (2008), 853884.
Lemmermeyer, F., Re ipro ity laws. Springer Monographs in Mathemati s, From Euler to Eisenstein
1838 (2000).
Rodrguez Villegas, F. and Zagier, D., Whi h primes are sums of two ubes?. Number theory
(Halifax, NS), CMS Conf. Pro . 15 (1994), 295306.
Satg, P., Groupes de Selmer et orps ubiques. J. Number Theory 23 (1986), 294317.
Satg, P., Un analogue du al ul de Heegner. Invent. Math. 87 (1987), 425439.
S haefer, E. F. and Stoll, M., How to do a p-des ent on an ellipti urve. Trans. Amer. Math.
So . 356 (2004), 12091231 (ele troni ).
Silverman, J. H., The arithmeti of ellipti urves. Graduate Texts in Mathemati s, Springer-Verlag
(New York) 106 (1992).
Top, J., Des ent by 3-isogeny and 3-rank of quadrati elds. Advan es in number theory (Kingston,
ON), Oxford Univ. Press 106 (1991).
Henri Cohen, IMB Universit Bordeaux I, 351 Cours de la Libration, 33405 Talen e,
Fran e
E-mail address : ohenmath.u-bordeaux1.fr

URL: http://www.math.u-bordeaux.fr/~ ohen/
Fabien Pazuki, IMJ Universit Paris 7, site Chevaleret, 2, pla e de Jussieu, 75 251 Paris
Cedex 05, Fran e
E-mail address : pazukimath.jussieu.fr

URL: http://www.math.jussieu.fr/~pazuki

Sede Electrónica de MUFACE

Încărcat de

Informații document

Descriere originală:

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Sede Electrónica de MUFACE

Încărcat de

Drepturi de autor:

Formate disponibile

ELEMENTARY 3-DESCENT WITH A 3-ISOGENY

arXiv:0903.4963v2 [math.NT] 27 Apr 2009

Keywords : Ellipti urves, 3-des ent.

11G05, 14G05, 14H52.

3-des ent. Then we

whi h is a bit more te hni al. Se tions 5 and 6

7, one nds several examples of families where we nd the

and we also give some

appli ations, su h as prime values of ertain ubi forms.

3-Selmer group has

also been studied in [20,

The Geometri Setting.

We re all a few fa ts about des ent on ellipti urves. Let

an ellipti urve over a number eld

ohomology, we have the short exa t sequen es:

First, using Galois

Henri COHEN and Fabien PAZUKI

We re all the denition of the

We also re all the denition of the

This leads to the following short exa t sequen e:

0 E(k)/nE(k) Sel(n) (k, E) X(k, E)[n] 0 ,

Thus, to get the exa t value of the rank

we must ompute the

we nd a geometri obje t

part of the TateShafarevi h group.

we an interpret the elements of the

group as twists of the

idea gives rise to dierent interpretations of the elements of the

n-Selmer group, as is learly

the omplete linear system given by

O denotes the identity element

diagram [C S] is a morphism from a torsor C under E to a

We will dene a BrauerSeveri diagram [C S] to be a twist of the diagram X =

su h that the urve

has points everywhere lo ally, hen e one an

n = 3. In this parti ular ase, the BrauerSeveri diagrams

We now spe ialize to the ase

we are looking for are of the type

The Arithmeti Setting.

For the proofs of all the results given in this se tion, we

be an ellipti urve dened over

and having a rational

Let us stress here that the point

itself. It is easy to see that

does not need to be rational

an be given by an ane equation of the type

(0, b D), so is rational if and only if D Q 2 .

16b3 D 2 (4Da3 27b),

Proof. We an write uniquely D = D0 f 2 , where D0 is a fundamental dis riminant and

Q if D = 1 and to a quadrati eld otherwise.

whi h will be equal to

an equation y 2 = x3 +D(ax+b)2 with D a fundamental dis riminant, and let T = (0, b D)

Henri COHEN and Fabien PAZUKI

b is dened by a similar equation y 2 = x3 + D(b

b = (0, (27b 4a3 D) 3D/9).

We now explain how the use of the

3-des ent maps and

(and of the isogenous urve

3-torsion points of an ellipti urve (evidently, if an ellipti urve does

point. Then we have

x([2]Q) = x(Q) = x(Q),

fundamental dis riminant.

b ), and = 1 if D = 1 or D = 3 and = 0 otherwise.

7, one nds several examples of families where we nd the

an ellipti urve over a number eld

We re all the denition of the

We also re all the denition of the

we nd a geometri obje t

part of the TateShafarevi h group.

idea gives rise to dierent interpretations of the elements of the

We will dene a BrauerSeveri diagram [C S] to be a twist of the diagram X =

n = 3. In this parti ular ase, the BrauerSeveri diagrams

be an ellipti urve dened over

an be given by an ane equation of the type

Q if D = 1 and to a quadrati eld otherwise.

b is dened by a similar equation y 2 = x3 + D(b

of nite index in an abelian group

the TateShafarevi h group of

to obtain the nal equation given in (3).

eld. Note that to use