Documente Academic
Documente Profesional
Documente Cultură
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.
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.
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,
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,
In se tion
2-des ent.
The
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
and let
Y
/
E(k)/nE(k)
E(kv )/nE(kv )
v v
n 2 be an integer.
/ 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
/
n-Selmer group :
Y
(n)
1
Sel (k, E) := Ker : H (k, E[n])
H 1(kv , E)[n] .
TateShafarevi
h group :
Y
X(k, E) := Ker H 1 (k, E)
H 1 (kv , E) .
(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),
n-torsion
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
su h that
dened over
isomorphi
to
Sin
e
E[n],
n-Selmer
group
n-Selmer
that is
Autk (X)
=
obje
t X . This
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,
Denition 1.1. A
|n.O|
indu es a morphism:
C1
S1
/
C2
/
S2
[C S]
S = Pn1 .
diagram
take
n-Selmer
group of
as a BrauerSeveri
3-des ent
1.2.
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
we denote
{O, T, T }.
3-torsion subgroup
that
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.
a, b,
and
in
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.
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
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
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)
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
3-Isogeny
the main result (Proposition 2.2 below), we need the following pre ise des ription of the
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
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
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+ ,
Proof.
If
Et
we have
we evidently have
Thus
Sin e
b E(Q))][
b E(Q))
b
b
b
|E(Q)/3E(Q)| = [E(Q) : (
(
: ((E(Q)))]
.
3-isogenies and b.
b and
subgroup
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
Tb (E(Q))
2(9b + 4a3 ) is a
(a, b) = 1. Thus,
if and only if
ube, in
b
[E(Q)
: (E(Q))]
b E(Q))
b
b
[(
: ((E(Q)))]
=
.
3 D,3(1(a,b))
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
|Im()|,
and
|Im()||Im(b
)| = 3r+D,1+D,3 ,
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()|.
2-des
ent, we
in the 3-part of
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
Note:
E(Q).
nontrivial
prime
p,
of them having
p-adi
valuation equal to
0,
typi ally
one
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)
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
so
(x, y) E(Q).
Furthermore,
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.
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,
gcd(X, Y, Z) = 1.
We an thus divide by
Geometri
Interpretation:
The plane
ubi
given in (1) of Theorem 3.1 is the equation of a twist
urve
we
of the ellipti
with
repla ed by
us3 .
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
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 .
3-des ent
D 6= 1,
D 6= 1
with a rational
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.
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
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
(1)
3-des ent
3
2
(ax + b) D =
uz then (X,
Y, Z) = (z , x, z) is a solution to the modied
ubi
uX 3 +
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
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
(X, Y, Z)
knowing
and
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
(v(X + Y
(2) By the theorem, the solubility of the equation depends only on the
lass [u] of u,
3
10
(4)
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
4.2.
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
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
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
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()
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 .
of
12
on an ane
3-des
ent map is dened
2
y (ax + b) D = (n d(am + bd ) D)/d3 . Thus
Remark.
q1 ,
with
v = f ,
as laimed.
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
and map
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
S3 (K) G3 ,
3-Selmer
group of
and denoted
S3 (K).
and let
Gi3
be the
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
the norm of
G3 ,
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()| .
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.
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 .
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
[] S3 (K)
hoose a representative
[]
v=
K ,
TG
TS
be the
the subgroup
generated by the elements for whi h we nd that it has a global solution. Thus
RS
of representatives of
S3 (K)/TS
and a set
RG
of representatives of
TS /TG .
(2) Let
primes, write
Cl(K)
will be small).
v that we have found write v = uq3 for some ideal q and some
as above.
(4) For ea
h
[S ] RS ,
where
RS
v = uS
for ea h
[G ] RG ,
where
RG
14
|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
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()|,
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.
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.
suitable powers of
u3
and
and
are
By multiplying
X, Y ,
p-integral.
Z by
u1 , u2 ,
and
Lemma 5.1. Assume as above that min(vp (u1), vp (u2 ), vp(u3 ), vp (c)) = 0.
(1)
(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
cXY Z = 0,
c = 0.
c 6= 0.
g = gcd(u1 , u2, u3 , c),
(u1 , u2, u3 , c)
gcd(u1, u2 , u3 , c) = 1.
and repla e
by
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
(b)
(
)
(d)
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
vp (u1 u2 u3 ) 2.
16
Thanks to the above lemma, we
an therefore always assume that our
ubi
is
sin
e if
p-adi
p-redu ed,
solution.
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.
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
(2)
17
3-des
ent
5.3.
3 su h
that
p | (27u1u2 u3 c3 ).
p is
p u1 u2 u3
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.
v3 (c) = 0
p = 3.
at
or that
v3 (c) =
v3 (c) > 0.
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.
k = 1,
k = 2.
k = 2.
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.
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)
Remarks.
v3 (c) > 0
and not
v3 (c) = 1,
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
3.
si :
s2 u 1 u 3
(3) In ase (2), there is no need to sear h among the four possibilities for the signs
3|c
(mod 3).
sin e
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,
of Theorem 3.1.
6.1.
D 6= 1
solubility when
D = 1,
D 6= 1.
We now onsider lo al
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
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 .
ZK
with
nZ
if and only if
is said to be primitive if
n = 1.
v/n
Im().
is a split prime.
vp (2b/(v (v))) 0.
vp (v (v)) = 0,
20
+ 2aZ(X + Y D)(X Y D) ,
with
u3 = 2b/(v (v)),
hen e su h that
vp (u3 ) 0.
6.2.
If
(1)
(2)
(3)
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
Lemma 6.6. 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, 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
Lemma 6.7. 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) = 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
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.
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 .
Lemma 6.11. 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. 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).
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.
family of urves
with equation
y 2 = x3 + (kp)2 ,
where
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
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 ),
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
When
k = 1,
u3 = kp.
u 1 = 1, u 2 = 2,
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
2-adi
k=4
the
2-adi
(0, 1, 2),
k = 1 or
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)
Proof.
(1), (3), (4), and (5) are lear from the lemma by inspe tion. For (2), we use the
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
Lemma 7.3.
(2)
Proof.
We have
2v1 = 1, 2v2 = 1,
and
2b = 6kp,
so
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.
(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),
if and only if
25
3-des ent
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
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.
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)
Proof.
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
Proposition 7.6.
(2)
(3)
Proof.
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)
Proof.
(1)
= 3r+1 .
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=
2,
p = 9511
y 2 = x3 + p2 has analyti
mwrank even at a high sear
h
the urve
limit nds only the one-dimensional subspa e of the (free part of the) MordellWeil group
generated by
has basis
32701984517186448621442294824950874787830128281
456289760665179363242981599270033206574137600
92890043770264171014255964610503972850176417273682124237369198272789821
y=
9746778232027925565271633950191532413151456450450966045051557376000
x=
3-des ent
as above,
28
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,
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.
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)
is globally soluble.
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
is globally soluble.
Proof.
=3
3.
by Elkies's result.
is made
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 ).
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
and
3.
ubes in
hara ter of
(3)
and
modulo
p.
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)
in
Corollary 7.10.
(2)
(3)
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
|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
p = 3889,
C 0
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
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.,
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,
Springer (New York), 193 (2000).
Cohen, H., Number theory. Vol. I. Tools and Diophantine equations. Graduate Texts in Mathemati
s,
Springer (New York), 239 (2007).
Cohen, H., Number theory. Vol. II. Analyti
and modern tools. Graduate Texts in Mathemati
s,
Springer (New York), 239 (2007).
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
Fabien Pazuki, IMJ Universit Paris 7, site Chevaleret, 2, pla
e de Jussieu, 75 251 Paris
Cedex 05, Fran
e