Skip to main content

Full text of "Elliptic Curves from Sextics"

See other formats


ELLIPTIC CURVES FROM SEXTICS 



MUTSUO OKA 



Abstract. Let Af be the moduli space of sextics with 3 (3,4)-cusps. The quotient 
moduli space Af/G is one-dimensional and consists of two components, Aftorus/G and 
Afgen/G. By quadratic transformations, they are transformed into one-parameter families 
C s and D s of cubic curves respectively. First we study the geometry of Af e /G, e — 
torus, gen and their structure of elliptic fibration. Then we study the Mordell-Wcil 
torsion groups of cubic curves C s over Q and D s over Q(\/— 3) respectively. We show 
that C s has the torsion group Z/3Z for a generic s £ Q an d it also contains subfamilies 
which coincide with the universal families given by Kubert |Ku| with the torsion groups 
Z/6Z, Z/6Z + Z/2Z, Z/9Z or Z/12Z. The cubic curves D s has torsion Z/3Z + Z/3Z 
generically but also Z/3Z + Z/6Z for a subfamily which is parametrized by Q(V~ 3). 



1. Introduction 

Let A3 be the moduli space of sextics with 3 (3,4)-cusps as in ||Q2|| . For brevity, we 
denote A3 by Af. A sextic C is called of a torus type if its defining polynomial / has 
the expression f(x,y) = f 2 (x,y) 3 + f 3 (x,y) 2 for some polynomials / 2 , / 3 of degree 2, 3 
respectively We denote by Aftoms the component of Af which consists of curves of a torus 
type and by Af gen the curves of a general type (=not of a torus type). We denote the dual 
curve of C by C*. Let G = PGL(3, C). The quotient moduli space is by definition the 
quotient space of the moduli space by the action of G. 

In §2, we study the quotient moduli space Af/G. We will show that Af/G is one 
dimensional and it has two components Aftorus/G and Af gen /G which consist of sextics of 
a torus type and sextics of a general type respectively. After giving normal forms of these 
components C s ,s G P 1 (C) and D s ,s G P 1 (C), we show that the family C s contains a 
unique sextic C54 which is self dual (Theor em |2.8| ) and C54 has an involution which is 
associated with the Gauss map (Proposition |2.12|) . 

In section 3, we study the structure of the elliptic fibrations on the components Af £ / G, e = 
torus, gen which are represented by the normal families C s , s G P X (C) and D s , s G P 1 (C). 
Using a quadratic transformation we write these families by smooth cubic curves C s and 
D s which are defined by the following cubic polynomials. 

C s : x 3 - \s{x - l) 2 + sy 2 = 
D s : -8x 3 + 1 + (s + 35)y 2 - 6x 2 + 3x - 6y/^3y - 3. 

— 6\/ r ^3x 2 — 12V— 3xy + (s — 35)xy 




Date: November, 1999, first version. 



1 



We show that C s , s G P 1 (C) (respectively D Sl s G P 1 (C)) has the structure of rational 
elliptic surfaces X 431 (resp. A3333) in the notation of ||Mi-P|| . 

In section 4, we study their torsion subgroups of the Mordell-Weil group of the cubic 
families C s and D s . The family C s is defined over Q and D s is defined over quadratic 
number field Q(V — 3). Both families enjoy beautiful arithmetic properties. We will show 
that the torsion group (C s )t or {Q,) is isomorphic to Z/3Z for a generic s G Q but it has 
subfamilies C V6 ^ U ), C Va ^ r ), C^t) and C Vl2 ( u ), u,r,t,u G Q for which the Mordell-Weil 
torsion group are Z/6Z, Z/6Z + Z/2Z, Z/9Z and Z/12Z respectively. Each of these 
groups is parametrized by a rational function with Q coefficients which is defined over Q 
and this parametrization coincides, up to a linear fractional change of parameter, to the 



universal family given by Kubert in [|Ku|| 



As for (D s ) tor (Q(y/—3)), we show that (D s ) tor (Q(y/^ 3) is generically isomorphic to 
Z/3Z + Z/3Z but it also takes Z/3Z + Z/6Z for a subfamily -Dg 6 (t) parametrized by a 
rational function with coefficients in Q and defined on Q(v / — 3). 

2. Normal forms of the moduli J\f 

We consider the submoduli Af^ 1 ' of the sextics whose cusps are at O := (0, 0), A := (1, 1) 
and B :— (1, — 1). As every sextic in Af can be represented by a curve in Af^ by the 
action of G, we have Af/G £ Af (1) /G^ where is the stabilizer of A" (1) : := {g G 
G;g(Af^) = Af^}. By an easy computation, we see that G^ is the semi-direct product 
of the group Gq and a finite group /C, isomorphic to the permutation group £3 where 
G^ is defined by 

(a l a 2 \ 
a 2 ai G G;a 3 (a2 - a^) ^ 0} 
ai - a 3 «2 a 3 y 

Note that G^ is normal in G^ and (7 G Gq 1 fixes singular points pointwise. The 
isomorphism /C = 1S3 is given by identifying (7 G K, as the permutation of three singular 
locus O, A, B. We will study the normal forms of the quotient moduli Af /G = Af^ /G^\ 

Lemma 2.1. For a given line L := {y = bx} with b 2 — 1 7^ 0, there exists M G Gq such 
that L M is given by x = 0. 

Proof. By an easy computation, the image of L by the action of M~ l , where M is 
as above, is defined by (ai — ba 2 )y + (a 2 — ba\)x = 0. Thus we take ai = ba 2 . Then 
a\ — a 2 = a^ip 2 — \) ^ Q \yy the assumption. □ 

Lemma 2.2. T/ie tangent cone at O is not y ± x = /or C G A/^ 1 ). 

Proof. Assume for example that y — x = is the tangent cone of C at O. The 
intersection multiplicity of the line L\ := {y — x = 0} and C at O is 4 and thus Li ■ C > 7, 
an obvious contradiction to Bezout theorem. □ 

Let AA 2 "* be the subspace of Af^ consisting of curves C G Af^ whose tangent cone at 
O is given by x = 0. Let G^ be the stabilizer of Af^ 2 \ By Lemma |2.1| and Lemma |2.2j , 
we have the isomorphism : 

2 



Corollary 2.3. A/" (1) /G (1) = A/" (2) /G (2) . 



It is easy to see that is generated by the group Gj> 2) := G® n G^ and an element 
r of order two which is defined by r(x,y) = (x, —y). Note that 




Gf = {M=\ 01 OleG^; a x a 3 ^ 0} 
yai-a 3 a 3 ) 

For C G A/"*- 2 -', we associate complex numbers 6(C), c(C) G C which are the directions of 
the tangent cones of C at A, B respectively. This implies that the lines y — 1 — b(C)(x — l) 
and y + 1 = c(C)(x — 1) are the tangent cones of C at A and S respectively. We have 

("2") 

shown that C G A/" torus if and only if 6(C) + c(C) = and otherwise C is of a general type 
and they satisfy c(C) 2 + 3c(C) - 6(C)c(C) + 3 - 36(C) + 6(C) 2 = (§4, @). 
We consider the subspaces: 

A/£L := {C G A/2 US ; 6(C) = 0}, A/JJ> := {C G A® ; 6(C) = c(C) = v^} 
and we put Af (3) := A/SL U N$n. 

Remark . The common solution of both equations: 6 + c = c 2 + 3c — be + 3 — 36 + 6 2 = is 
(6, c) = (1, —1) and in this case, C degenerates into two non-reduced lines (y 2 — x 2 ) 2 = 
and a conic. 

Lemma 2.4. Assume that C G M^. Then there exists a unique C G and an 

element g G such that C 9 = C' . This implies that 

Ntoms/G = Af torus /G^ ^ = Mt } us) Af gen /G = NgJ n /G^ ^ = Ng2t 

Proof. Assume that C G M'^ ua , 6 + c = 0. Consider an element g G G^\ 

/ 1 

g' 1 = 10 
\l - a 3 a 3/ 

The image is given by y — x + 0:03 — 03 — 6x03 + 603 = 0. Thus we can solve the 
equation a 3 (l — 6) — 1 = in a 3 uniquely as a 3 = 1/(1 — 6) as 6 7^ 1. Thus g G Gq 1 ^ is 
unique if it fixes the singular points pointwise and thus C' is also unique. It is easy to 
see that the stabilizer of ftf}^ us is the cyclic group of order two generated by r, as C' is 
even in y (see the normal form below) and C /T = C' for any C' G A/"/ 3 ^ us . Thus we have 

Consider the case C G Afgel- Then the images of the tangent cones at A, B by the action 
of g are given by y — x + xa 3 — a 3 — bxa 3 + ba 3 = and y + x — xa 3 + a 3 — cxa 3 + ca 3 = 
respectively. Assume that 6(C 9 ) = c(C g ). Then we need to have a 3 (l — 6) — 1 = 
a 3 (— 1 — c) + 1, which has a unique solution in 03, if (*) 6 — c — 2 7^ 0. Assume that 
c 2 + 3c— 6c+3 — 36+6 2 = and 6 — c— 2 = 0. Then we get (6, c) = (1, —1) which is excluded 
as it corresponds to a non-reduced sextic. Thus the condition (*) is always satisfied. Put 
(6', d) := (6(C 9 ), c(C 9 )). They satisfy the equality c' 2 + 3c' - 6V + 3 - 36' + 6 /2 = and 

3 



V = d . Thus we have either V = d = y— 3 or V = d = — y— 3. However in the second 
case, (C 9 ) T belongs to the first case. Thus b' — d — 3 and C 9 G Afgen as desired. □ 

2.1. Normal forms of curves of a torus type. In | p2| , we have shown that a curve in 

Aftorus is defined by a polynomial f(x, y) which is expressed by a sum f2{x, y) 3 + sf 3 (x, y) 2 
where f2^x, y) is a smooth conic passing through O, A, B and f 3 (x, y) = (y 2 — x 2 )(x — 1). 

Proposition 2.5. The direction of the tangent cones at O, A and B are the same with 
the tangent line of the conic f2(x,y) = at these points. 

This is immediate as the multiplicity of f 3 (x,y) 2 at 0,A, B are 4. Assume that C G 
Ntorusi that is, the tangent cones of C at O, A and B are given by x = 0, y — 1 = 
and y + 1 = respectively. Thus the conic f2{x,y) = is also uniquely determined as 
f2{x, y) = y 2 + x 2 — 2x. Therefore Af^rus is one-dimensional and it has the representation 

(2.6) C s : f torus (x,y,s) := f 2 (x,y) 3 + sf 3 (x,y) 2 = 

For s ^ 0, 27, oo, C s is a sextic with three (3,4)-cusps, while C27 obtains a node. If 
g G fixes the tangent lines y ± 1 =0, then g = e or r. As CJ = C s , this implies that 
C\ = C s . Thus C s ^ C t if s ^ t. 

2.2. Normal form of sextics of a general type. For the moduli M ge n of sextics of a 
general type, we start from the expression given in §4.1, ||Q2 |. We may assume b = c = 
\J — 3. Then the parametrization is given by 

f gen (x,y,s) := f (x,y) + sf 3 (x,y) 2 , h(x,y) = (y 2 - x 2 )(x - 1) 

where s is equal to a 06 in ||Q2|| and f is the sextic given by 

(2.7) f (x, y) := y 6 + y 5 {6\/=3 - Qy/^3x) + y 4 (35 - 76x + 38x 2 ) 
+?/ 3 (-24v^3x + 36^3x 2 - ^v^a; 3 ) + y 2 (-94x 2 + 200a; 3 - 103x 4 ) 

+y(24y/^3x 3 - 42 v / ^3x 4 + lSv^a; 5 ) + 64x 3 - 133x 4 + Q8x 5 

Let D s := {f gen (x,y,s) = 0} for each s G C. Observe that D = {fo(x,y) = 0} is a 
sextic with three (3,4)-cusps and of a general type. For the computation of dual curves 
using Maple V, it is better to take the substitution y \— > to make the equation to 

be defined over Q. Summarizing the discussion, we have 

Theorem 2.8. The quotient moduli space Af/G is one dimensional and it has two com- 
ponents. 

(1) The component Aftorus / G has the normal forms C s = {f(x, y, s) = 0} where f(x, y, s) = 
f2(x,y) 3 + sf 3 (x,y) 2 , f 2 (x,y) = y 2 + x 2 - 2x and f 3 (x,y) = {y 2 - x 2 )(x - 1). The curve 
C54 is a unique curve in Af/G which is self-dual. 

(2) The component Af gen / G has the normal form: f ge n(x, y, s) = fo(x, y)+sf 3 (x, y) 2 where 
/ 3 is as above and the sextic fo(x,y) = is contained in Af gen . This component has no 
self-dual curve. 

Proof of Theorem Wfh. We need only prove the assertion for the dual curves. The 
proof will be done by a direct computation of dual curves using the method of §2, |U2[ | 
and the above parametrizations. We use Maple V for the practical computation. Here 

4 



is the recipe of the proof. Let X*,Y*,Z* be the dual coordinates of X,Y,Z and let 
(x*,y*) := (X*/Z*,Y*/Z*) be the dual affine coordinates. 

(1) Compute the defining polynomials of the dual curves C* and D* respectively, us- 
ing the method of Lemma 2.4, ||Q2|| . Put gtorus(x* ,y* , s) and g gen (x* ,y* , s) the defining 
polynomials. 

(2) Let G e (X* ,Y* , Z* , s) be the homogenization of g e (x* ,y* , s), e = torus or gen. 
Compute the discriminant polynomials Ay*G £ which is a homogeneous polynomial in 



X*, Z* of degree 30 (cf. Lemma 2.8, ||01|| ). Recall that the multiplicity in A Y *G e of the 
pencil X* — rjZ* = passing through a singular point is generically given by \i + m — 1 
where /i is the Milnor number and m is the multiplicity of the singularity ( ||Q2|| ). Thus 



the contribution from a (3,4)-cusp is 8. Thus if C* has three (3,4)-cusps, it is necessary 
that Ay*{G) = has three linear factors with multiplicity > 8. 

(3-1) For the curves of a general type, an easy computation shows that it is not possible 
to get a degeneration into a sextic with 3 (3,4)-cusps by the above reason. 

(3-2) For the curves of a torus type, we can see that s = 54 is the only parameter such 
that C* G Af. Thus it is enough to show that C| 4 = C54. 

(4) The dual curve C| 4 of C54 is defined by the homogeneous polynomial 

G(X*, Y*, Z*) := 128X* 5 Z* + 1376X* 4 Z* 2 - 192X* 3 Y* 2 Z* + 4664X* 3 Z* 3 - 2X* 2 Y* 4 

-1584X* 2 F* 2 Z* 2 + 7090X* 2 Z* 4 + 58X*Y* 4 Z* - 3060X*Y* 2 Z* 3 

+5050X*Z* 5 + Y* 6 + 349Y* 4 Z* 2 - 1725Y* 2 Z* 4 + 1375Z* 6 

We can see that C 54 * is isomorphic to C 54 as (Cg 4 ) A = C 54 where 

'-4/3 -5/3 
.4 10 1 

-5/3 -13/3, 

2.3. Involution r on 654. For a later purpose, we change the coordinates of P 2 so 
that the three cusps of C s are at O z := (0,0,l),O y := (0, 1,0), Ox := (1,0,0). A 
new normal form in the affine space is given by C s : f2{x,y) 3 + sfs(x,y) 2 = where 
f 2 (x,y) := xy — x + y and fs(x, y) := —xy. In particular, C 54 is defined by 

(2.9) f(x, y) = (xy - x + yf + hAx 2 y 2 = 

In this coordinate, C| 4 is defined by 

-28y 3 - 17xV - I7x 2 y 4 - 28x 3 y 3 - 2y 5 + 1788x 3 ?/ + 1788x 2 y - I7y* - 17x 4 

+262xy + I188x 2 y 3 - 1788a;?/ 2 - 262xy 4 + I788xy 3 - 1788x 3 y 2 - 8166a; 2 ?/ 2 + 28a; 3 

+2Q2x A y - 2x 5 y - 2xy 5 + 1 - 17y 2 - 17x 2 + 2a; 5 + 2a; - 2y + x 6 + y 6 = 

It is easy to see that (C^)^ 1 = C 54 where 

/-1/3 7/3 -1/3 N 
A x := 7/3 -1/3 1/3 
\-l/3 1/3 -7/3, 

For a given A G GL(3, C), we denote the automorphism defined by the right multiplication 
of A by if a- Let F(X,Y, Z) be the homogenization of f(x,y). Then the Gauss map 

5 



dualc 54 : C 54 — > Cg 4 is denned by 

dual C54 (X, Y, Z) = (F X (X, Y, Z), F Y (X, Y, Z), F Z (X, Y, Z)) 

where Fx, Fy, Fz are partial derivatives. We define an isomorphism r : C54 — > C54 by the 
composition ipA 1 °d\ialc 54 . Then r is the restriction of the rational mapping: \1> : C 2 — > C 2 , 

(x, y) i-> (ar d , y d ) and 



(2.10) 



— y 3 +4x 2 — a: 2 y 3 +4a: 3 y 2 — 8x 3 y— Ax 2 y 2 — 8xy— Axy 2 — 2xy 3 +109x 2 y+Ay 2 +Ax 

d ' — Ay 3 +x 2 — Ax 2 y 3 +Ax 3 y 2 — 8x 3 y— 109x 2 y 2 — 2xy— Axy 2 — 8xy 3 +Ax 2 y+y 2 +Ax 3 

. _ - Ay 3 +Ax 2 —Ax 2 y 3 +x 3 y 2 — 2x 3 y—Ax 2 y 2 -8xy-109xy 2 — 8xy 3 +Ax 2 y+Ay 2 +x 3 

' — Ay 3 +x 2 — Ax 2 y 3 +Ax 3 y 2 — 8x 3 y— 109x 2 j/ 2 — 2xy— Axy 2 — 8xy 3 +Ax 2 y+y 2 +Ax 3 



Observe that r is defined over Q. C 54 has three flexes of order 2 at F x := (1, —1/4, 1), F 2 :- 
(1/4, —1, 1), F 3 := (4, —4, 1) and r exchanges flexes and cusps: 



(2.11^ 



r{O x ) = F u t{O y ) = F 2 ,t(O z ) = F 3 , 
r(Fi) = O x , t(F 2 ) = Y , t{F 3 ) = O z 

Furthermore we assert that 

Proposition 2.12. The morphism r is an involution on C54. 

Proof. By the definition of r and Lemma |2.13| below, we have (C := C54): 

r o r = (ft^-i o dualc) 2 = (dual^^! o <pa.i) (v^a- 1 dualc) = id 

as A± is a symmetric matrix. □ 
Let C be a given irreducible curve in P 2 defined by a homogeneous polynomial F(X, Y, Z) 
and let B G GL(3, C). Then C B is defined by G(X, Y, Z) := F((X, Y, Z)B~ l ). 

Lemma 2.13. Two curves (C B )* and (C*) tB 1 coincide and the following diagram com- 
mutes. 

q dualg q* 

QB d ual C? njhy 



Proof. The first assertion is the same as Lemma 2, [|0 2|| . The second assertion follows 
from the following equalities. Let (a, b, c) G C. 

dual C B ((p B (a,b,c)) = (G x ((pB(a,b,c)),GY((pB(a,b,c)),G z (<fi B (a,b,c))) 
= (F x (a, b, c), F Y (a, b, c), F y (a, b, c)) t B^ 1 = y>t B -i(dualc?(a, b, c) □ 

In section 5, we will show that r is expressed in a simple form as a cubic curve. 



3. Structure of elliptic fibrations 

We consider the elliptic fibrations corresponding to the above normal forms. For 
this purpose, we first take a linear change of coordinates so that three lines defined 
by f$(x,y) = changes into lines X = 0, Y = and Z = 0. The correspond- 
ing three cusps are now at O z = (0,0,1),CV = (0,1,0), Ox = (1,0,0) in P 2 . Then 
we take the quadratic transformation which is a birational mapping of P 2 defined by 

6 



(X,Y,Z) I— > (YZ, ZX, XY). Geometrically this is the composition of blowing-ups at 
Ox,Oy,Oz and then the blowing down of three lines which are strict transform of 
X,Y, Z = 0. It is easy to see that our sextics are transformed into smooth cubics for 
which X = 0, Y = and Z = are tangent lines of the flex points. Those flexes are the 
image of the (3,4)-cusps. We take a linear change of coordinates so that the flex on Z = 
is moved at O := (0, 1, 0) with the tangent Z = 0. Then the corresponding families are de- 
scribed by the families given by {h t0 rus(x, t/,s)=0;sG P 1 } and {h gen (x, y, s) = 0, s G P 1 } 
where 

C s : h torus (x, y, s) := x 3 - \s{x - l) 2 + sy 2 , 
D s : h gen {x, y, s) := -8x 3 + 1 + (s + 35)y 2 - Qx 2 + 3a; 

— 6V—32/ — 3 V— 3x — 6\^3x 2 — Yl\J— 3xy + (s — 35)xy 

Let tf £ pf, Y, Z, S, T) = h £ (X/Z, Y/Z, S/T)Z 3 T for e = torus, gen. We consider the ellip- 
tic surface associated to the canonical projection 7r : S e — > P 1 where S s is the hypersurface 
in P 1 x P 2 which is defined by H e (X, Y, Z, S, T) = 0. 

Case I. Structure of S t0 rus P 1 - For simplicity, we use the affine coordinate s = 
S/T of {T ^ 0} C P 1 and denote ^(s) by C s . We see that the singular fibers are 
0, 27, oo. Cqo consists of three lines, isomorphic to J3 in Kodaira's notation, ||Ko| 



C27 obtains a node and this fiber is denoted by I\ in |Ko|| . The fiber Cq is a line with 



multiplicity 3. The surface S torus has three singular points on the fiber C at (X, Y, Z) 
(0, 1/2, 1), (0, —1/2, 1), (0, 1, 0). Each singularity is an ^-singularity. We take minimal 
resolutions at these points. At each point, we need two P 1 as exceptional divisors and 
let p : Storus — ► Storus be the resolution map. The composition n := n o p : S t0 rus — * P 1 
is the corresponding elliptic surface. Now it is easy to see that Co '■= 7r _1 (0) is a singular 
fiber with 7 irreducible components, which is denoted by IV* in |[Ko|| . Here we used the 



following lemma. The elliptic surface S tor us is rational and denoted by X 431 in [ [Mi-P|| . 

Assume that the surface V := {(s,x, y) G C 3 ; f(s, x,y) = 0} has an A 2 singularity 
at the origin where f(s,x,y) := sx + y 3 + sx ■ h(s,x,y) where h(0) = 0. Consider the 
minimal resolution ir : V —>■ V and let 7r _1 (0) = Ei\jE%. It is well-known that Ei-E 2 = 1 
and E 2 = -2 for i = 1,2. 

Lemma 3.1. Consider a linear form £(s,x,y) = as + bx + cy and let L' be the strict 
transform of £ = to V. 

(1 ) Assume that b = c = anda^O. Then (ir*£) = 3L' + 2E X + E 2 and L' ■ E x = 1 and 
V does not intersect with E 2 , under a suitable ordering of E\ and E 2 . 

(2) Assume that abc ^ 0. Then we have (n*£) = V + Ei + E 2 and V ■ Ei — 1 for i = 1,2. 

The proof is immediate from a direct computation. 

Case II. Structure of S gen — > P 1 . Now consider the elliptic surface S gen . Put D s = 
7r _1 (s). The singular fibers are at s = —35, —53 + 3, —53 — 6\/— 3 and s = 00. The 
fiber s = 00 is already ^3 and is smooth on this fiber. On the other hand, S gen has a 
^-singularity on each fiber D s , s = —35, —53 + 6-\/— 3, —53 — 6V — 3. Let p : — > S gen 
be the the minimal resolution map and we consider the composition n := nop : 5 9en — > P 1 



as above. Then using (2) of Lemma EO , we see that tt 
fibers and each of them is J3 in the notation 
and denoted as A3333 in IllVli-P 



S, 



P 1 has four singular 
Ko(|. This elliptic surface is also rational 



gen 



4. Torsion group of C s and D s 

Consider an elliptic curve C defined over a number field K by a Weierstrass short 
normal form: y 2 = h(x), h(x) = x 3 + Ax + 5. The j-invariant is defined by j(C) = 
-1728(4A) 3 /A with A = -16{AA 3 + 27B 2 ). We study the torsion group of the Mordell- 
Weil group of C which we denote by C tor (K) hereafter. 

Recall that a point of order 3 is geometrically a flex point of the complex curve C 
and its locus is defined by F{f) := f x , x f 2 - 2f x>y f x f y + f y , y f x = where f{x,y) is 
the defining polynomial of C ( |Q1|| ). In our case, F(f) = 24xy 2 — 18x 4 — 12x 2 A — 2 A. 
The unit of the group is given by the point at infinity O := (0, 1,0) and the inverse of 
P = (a,/3) G C is given by (a, — (3) and we denote it by —P. For a later purpose, we 
prepare two easy propositions. Consider a line L(P,m) passing through — P defined by 
y = m(x — a) — (3. The x-coordinates of two other intersections with C are the solution 
of q{x) := f(x, m{x — a) — (3)/(x — a) which is a polynomial of degree 2 in x. Let A x q be 
the discriminant of q in x. Note that A x q is a polynomial in m. 

(A) When does a point Q G C exist such that 2Q = P. 

Assume that a K point Q = (xi,yi) satisfies 2Q = P. Geometrically this implies that 
the tangent line TqC passes through —P. 

Proposition 4.1. There exists a K -point Q with 2Q = P if and only if m is a Re- 
solution of A x q(m) = and X\ is the multiple solution of q(x) =0. If P is a flex point, 
A x q(m) = contains a canonical solution which corresponds to the tangent line at P and 
m = —f x (a, (3)/ f y (a, (3) . For any K -solution m with m 7^ —f x (a, (3)/ f y (a, (3) , the order 
of Q is equal to 2 ■ order P. 

(B) When does a point Q G C exist such that 3Q = P. 

Assume that a K-pomt Q = (xi,y%) satisfies 3Q = P. Put Q' := 2Q and put Q' = (x 2 , y 2 ). 
Let TqC be the tangent line at Q. Then TqC intersects C at —Q'. Then — 3Q is the third 
intersection of C and the line L which passes through Q, Q'. Thus three points — P, Q, Q' 
are colinear. Write L as y = m(x — a) — (3. Then X\,x 2 are the solutions of q(x) = 0. 
Thus we have 

(4.2) x 2 = — coeff(g, x)/coeff(g, x 2 ) — x±, y\ — m[x\ — a) — (3 

where coeff(g, x % ) is the coefficient of x % in q(x). Let Lq{x, y) be the linear form defin- 
ing TqC and let R(x) be the resultant of f(x,y) and Lq(x, y) in y. Put Ri(x) : = 
R(— coeff(g, x)/coeff(g, x 2 ) — x). Then by the above consideration, x = x\ is a com- 
mon solution of q(x) = Ri(x) = 0. Let R2{m) be the resultant of q(x) and Ri(x). Note 
that if A x q(m) = 0, L is tangent to C at Q and R^m) = 0. In this case, 2Q = P. 

Proposition 4.3. Assume that there exists a K-point Q with 3Q = P and order Q = 
3 ■ order P and let m be as above. Then R2(m) = and A x q(m) 7^ 0. Moreover X\ is 
given as a common solution of q(x) — R\{x) = 0. 

8 



Actually one can show that i? 2 (m) is divisible by (A x q) 2 . 

4.1. Cubic family associated with sextics of a torus type. We have observed that 
the family C s for s G Q is defined over Q. First, recall that C s is defined by 

(4.4) C s : x 3 - -s(x - l) 2 + sy 2 = 

and the Weierstrass normal form is given by C s : y 2 = x 3 + a(s)x + b(s) where 

(4.5) a(s ) = _^ + l s 3, Ks) = _^ + ^ + _L/ 

Put X := {0, 27, oo}. This corresponds to singular fibers. We have two sections of order 
3: s t— > (j^s 2 , ±|s 2 ). Put Pi := (j^s 2 , \s 2 ). Thus the torsion group is at least Z/3Z. By 
[Ma|| , the possible torsion group which has an element of order 3 is one ofZ/3Z, Z/6Z, 
Z/2Z + Z/6Z, Z/9Z or Z/12Z. The j-invariant of C s is given by 

(4.6) j(C s ) := jtomsis), jtorus(s) := s{s - 24)7(s - 27) 

(1) Assume that (C s ) tor (Q) has an element of order 6, say P2 ■= («2, P2) € C s fl Q 2 . We 
may assume that P 2 + P 2 = -Pi- By Proposition 4~T, this implies that x = a<i must be the 
multiple solution of 

q[x) := s 4 — 36s 3 — 72ms 2 — 6a;s 2 — 6s 2 m 2 + 72m 2 x — 72x 2 = 

As —f x (—Pi)/fy(—Pi) = —s/2, we must have m 7^ — s/2 and thus 

(4.7) A' x q := A x q/(2m + s) = s 3 - 32s 2 - 2ms 2 - 4m 2 s + 8m 3 = 

The curve A' x (q) = is a rational curve and we can parametrize A' x q = as s = (pe(u), 
m = {Pq(u)u where 

(4.8) <p 6 (u) := 32/(1 + 2u)(2u- l) 2 
The point P 2 is parametrized as 

128 -l + 12u 2 512(6u + l) 



^ P ' ^" 3 (2u + 1) 2 (-1 + 2m) 4 ' (-1 + 2uf{2u + l) 2 ' 

where u G Q. We put A 6 := {s = ip 6 (u);u G Q} and S 6 := ^ _1 (E). Note that 
S 6 = {-1/2, 1/2, 5/6, -1/6}. 

(1-2) Assume that we are given s = (f(u) and we consider the case when ( |4.7| ) has three 
rational solutions in m for a fixed s. This is the case if ipe(u) = tp§{v) has two rational 
solutions different from u. This is also equivalent to (C s ) ior (Q) has Z/2Z + Z/2Z as a 
subgroup. The equation is given by the conic 

(4.10) Q : 4u 2 -2u + 4uv -l-2v + 4v 2 = 

By an easy computation, Q is rational and it has a parametrization as follows. 

. . .. -36 + 5r 2 l(r 2 + 24r-36) 

(4.11) M = o9 2 (r):=— r-, u(r) := — - — ; — - 

K J Y W 6(12 + r 2 )' V ; 6 (12 + r 2 ) 

9 



The generators are P 2 of order 6 and R = (7, 0) of order 2 where 

81 (r 4 - 48r 3 + 72r 2 - 432) (12 + r 2 ) 4 
7 '~ ~T (r 2 - 36) 4 r 4 

Put ^6,2 ( r ) : = fe(f2( r )), which is given explicitly as 

^ 62 ( r ) = 27(12 + r 2 )/r 2 (r - 6) 2 (r + 6) 2 

We define a subset y4 6j2 of A 6 by the image </ } 6,2(Q)- Put £6.2 : = < / ? 62(^)- ^ * s gi ven by 
£ 6 , 2 = {0,±2,±6}. 

(2) Assume that there exists a rational point P3 = (0:3, /?3) of order 9 such that 3P3 = P. 
By Proposition |4.3j , this is the case if and only if 

R 3 (m, s) : = 512m 9 + 768m 8 s - 512m 6 s 3 - 1536m 6 s 2 - 192s 4 m 5 
-6144m 5 s 3 - 6528mV + 96s 5 m 4 - 12288m 3 s 4 - 2048m 3 s 5 + 64s 6 m 3 + 480s 6 m 2 
-15360s 5 m 2 - 6144s 6 m + 384s 7 m - 6s 8 m + 56s 8 - 512s 6 - 768s 7 - s 9 = 

has a rational solution. Here R% is P 2 /(A,j.g) 2 (s + 2m)s 4 up to a constant multiplica- 
tion. Again we find that the curve {(m, s) G C 2 ;Ps(m, s) = 0} is rational and we can 
parametrize this curve as s = fg(t), m = ipg(t) where 

f (f\ ._ 1 (-l+9t 2 -3f+3t 3 ) 3 

^■ lZ > ) . / 1\ ._ 1 (-l+9i 2 -3f+3t 3 ) 2 (-t+t 3 +l+7i 2 ) 

^ V9W •- i 6 t 3 (t-l) 3 (t+l) 3 

The generator P 3 = (03,^3) is given by 

1 (l-18t+15t 2 -12f 3 +15f 4 +30t 5 +33t 6 )(9f 2 -l+3f 3 -3f) 4 
° 3 ~~ 768 (t-l) 6 (m) 6 i 6 

/j 1 (l+3t 2 )(9t 2 -l+3f 3 -3t) 6 

^ 3 ~~ 512 (t-l) 5 (m) 7 i 8 

We put A 9 := {ip 9 (t);t E Q} and S 9 := ^(E) = {0,1,-1}. 

(3) Assume that s £ A 6 and (C s ) tor (Q) has an element P 4 = (a^, ^4) 6 C s n Q 2 of order 
12. Then we may assume that P4 + P4 = P 2 . This implies that the tangent line at P 4 
passes through — P 2 . Write this line as y = n(x — a 2 ) — /3 2 . By the same discussion as 
above, the equality r(ni,w) = holds where T is the polynomial defined by 

(4.13) T(u, ni) := -786432m 4 - 98304n lM 3 - 524288m 3 + 393216m 2 - 16384n lM 2 
-3072n 2 w 2 + 131072u + 24576niu + 4096rii + 16384 + 256n 2 + n\ 

and n = n\/(2u + l)(2u — l) 2 . Again we find that T = is a rational curve and we have 
a parametrization: u = u(v) and rii = rti{v) where 

I".") „(„) = -iggg}, „ lM = -i6 '"'-y_-xt- 3 ' 

( 4 -!5) s = ^i 2 (i/) :=WiKf)), y>i2(y) := - fr-i)^^^) 

The generator of the torsion group Z/12Z is P 4 = (a 4 , /3 4 ) where 



ct 4 



1 (i^ 8 -12t/ 7 +24i/ 6 -36^ 5 +42^ 4 +12t/ 3 +36i/-3)(i/ 4 -6t/ 2 -3) 

12 (^-1) 8 (!/+1) 8 (^ 2 +1) 



2 



1 (v A -&v 2 -3) 6 v(v 2 +S) 
1 -~ 2 (i/-l) 7 (i/+l) n (i/ 2 +l) 2 

10 



We put A12 := {ip^iy)]^ G Q}. By definition, A 12 C A 6 . The singular fibers X 12 : = 
</? _1 (E) is given by {0, ±1}. Summarizing the above discussion, we get 

Theorem 4.16. The j-invariant is given by jtorus( s ) — s i s ~ 24) 3 /(s — 27) and the 
Mordell-Weil torsion group of C s is given as follows. 

Z/3Z, sGQ-A 6 UA)U£ 
Z/6Z, s = (p 6 {u) G As- A 6j2 U A 12 , ttGQ-Eg 
(C s )t r(Q) = <! Z/6Z + Z/2Z, s = y? 6 , 2 (r) G A 6>2 , r G Q - £ 6 , 2 

Z/9Z, s = y> B (t) e4 t g Q - s 9 

Z/12Z, s = p 12 (i/) G A 12 , i/ G Q - Ei 2 

4.2. Comparison with Kubert family. In ||Ku|| , Kubert gave parametrizations of the 
moduli of elliptic curves defined over Q with given torsion groups which have an element 
of order > 4. His family starts with the normal form: 

(4.17) E(b, c) : y 2 + (1 - c)xy - by = x 3 - bx 2 

We first eliminate the linear term of y and then the coefficient of x 2 . Let K w (b, c) be the 
Weierstrass short normal form, which is obtained in this way. The j-invariant is given by 

_ (l-86c 2 -8c6-4c+166 + 6c 2 + 166 2 -4c 3 + c 4 ) 3 
J( ( ' C '' ~ 6 3 (3c 2 - c - 3c 3 - 86c 2 + b - 20cb + c 4 + 166 2 ) 

For a given elliptic curve E defined over K with Weierstrass normal form E : y 2 = 
x 3 + ax + b and a given k E K, the change of coordinates ihi/P^h 2//& 3 changes 
the normal form into y 2 = x 3 + afc 4 x + 6fc 6 . We denote this operation by ^>k{E). 

1. Elliptic curves with the torsion group Z/6Z. This family is given by a parameter c 
with 6 = c + c 2 . 

2. Elliptic curves with the torsion group Z/6Z + Z/2Z. This family is given by a 
parameter c\ with 6 = c + c 2 and c = (10 — 2ci)/(c 2 — 9). 

3. Elliptic curves with the torsion group Z/9Z. The corresponding parameter is / and 
b = cd,c = fd-f,d = f(f-l) + l. 

4. Elliptic curves with the torsion group Z/12Z. The corresponding parameter is r and 
b = cd,c = fd — f,d = m + r, / = m/ (1 — r) and m = (3r — 3r 2 — l)/(r — 1). 

Proposition 4.18. Our family C^^, C V6 2 ^, C^ft), are equivalent to the respec- 

tive Kubert families. More explicitly, we take the following change of parameters to make 
their j-invariants coincide with those of Kubert and then we take the change of coordinates 
of type to make the Weierstrass short normal forms to be identical with K w (x,y). 

1. ForC w{u) , take u = -(c - l)/2(3c + 1) and k = c 2 {c + l)/(3c + l) 2 . 



2. For a(r) , ta£;e r = -12/(ci - 3) and k = 4(-5 + ci) 2 (ci - l) 2 /(c 2 - 6c x + 21) 2 



y V6(u) 

/( Cl -3)( Cl + 3). 

3. For C M t), take t = -//(/ - 2) and k = f 3 (f - l) 3 /(/ 3 - 3/ 2 + l) 2 . 

4. For C 912{v) , take v = -l/(2r - 1) and k = (r - l)r 4 (-2r + 2r 2 + 1)(-1 + 2r) 2 / 
(6r 4 -12r 3 + 12r 2 -6r + l) 2 . 

We omit the proof as the assertion is immediate from a direct computation. 

11 



4.3. Involution on C54. We consider again the self dual curve C := C54 (see §3). The 
Weierstrass normal form is y 2 = x 3 -98415x+ 11691702. Note that 54 e A 6 -A 12 UA 6i2 U£. 
In fact, 54 = <£>g(l/6) and 54 A12 U Ag, 2 . The j-invariant is 54000 and the torsion 
group C tor (Q) is Z/6Z and the generator is given by P = (—81,4374). Other rational 
points are 2P = (243, -1458), 3P = (162,0),4P = (243, 1458), 5P = (-81,-4374), and 
O = (0, 1, 0) (= the point at infinity). Recall that C has an involution r which is defined 
by ( |2.10p in §3. To distinguish our original sextic and cubic, we put 

C (6) : (xy - x + y f + 5Ax 2 y 2 = 0, C {3) : y 2 = x 3 - 98415a; + 11691702 

The identification $ : C {3) C {6) is given by the rational mapping: 

$( x , y ) = (-2916/(27x - 5103 - y), 2916/(y + 27x - 5103)) 

and the involution on is given by the composition o t o $. After a boring 
computation, i s 

reduced to an extremely simple form in the Weierstrass normal form 
and it is given by r^ 3 \x,y) = (p(x,y), q(x,y)) where 

(4.19) p(x,y):= 81 — - q(x, y) := -19683- 



x - 162 ^ v ' (x- 162) 2 

Note that C has another canonical involution 1 which is an automorphism defined by 
l : (x,y) I— > (x, —y). We can easily check that ot = to r^ 3 \ Note that r^ 3 '(P) = 
2P,t^(2P) = P,t®(3P) = 0,tW(0) = 3P,t (3 \4P) = 5P,r (3 \5P) = 4P. Let r] : C -> 
C be the translation by the 2-torsion element 3P i.e., T](x, y) = (x, y) + (162, 0). It is easy 
to see that is the composition t o 77. That is (x, y) = (x, — y) + (162, 0) where the 
addition is the addition by the group structure of C54. Thus 

Theorem 4.20. The involution r on sextics is equal to the involution on 
which is defined by ( U-1Q ) an d it is also equal to (x,y) t— > (x, —y) + (162,0). 



4.4. Cubic family associated with sextics of a general type. We consider the 
family of elliptic D s curves associated to the moduli of sextics of a general type with 
three (3,4)-cusps. Recall that D s is defined by the equation: 

D s : -8x 3 + 1 + sy 2 + 35y 2 - 6x 2 + 3x - G\^3y - 3^3x 
-Q^f^3x 2 - Yl\T^>xy + (s - 35)xy = 

This family is defined over Q(a/— 3). We change this polynomial into a Weierstrass 
normal form by the usual process killing the coefficient of y and then by killing the 
coefficient of x 2 . A Weierstrass normal forms is given by y 2 = x 3 + a(s)x + b(s) where 

( a(s) := -^(s + 47) (s + 71)(s 2 + 70s + 1657) 
(4.21) I b(s) := ^(s 2 + 70s + 793)(s 4 + 212s 3 + 17502s 2 

[ +648644s + 9038089) 

The singular fibers are s = —35, —53 + 6^—3, —53 — 6\/— 3 and s = 00. Put £ = 
{—35, —53 + ±6^-3, 00}. In this section, we consider the Modell-Weil torsion over the 

12 



quadratic number field Q(\/— 3). First we observe that this family has 8 sections of order 
three ±P 3 i; z = 1, . . . , 4 where P 3 ,j are given by 



(4.22) ^3,1-^3,1^3,1) 

(4.23) P 3 ,2--={x3,2,y 3 ,2 

(4.24) P 3>3 : = (x 3 ,3,y 3 ,3) 5 

(4.25) P 3 ,4 := (0:3,4, 2/3,4) 



x 3 ,i := 5041/48 + 71s/24 + s 2 /48 
2/3,i := 2917/4 + 53s/2 + s 2 /4 

x 3 , 2 := -2209/16 - 47s/8 - s 2 /16 
Vz2 ■= v^3(s 2 + 106s + 2917)(s + 35)/144 

x 3t3 ■ = s 2 /48 + 793/48 + 35s/24 + (s + 35) v /= 3/2 
y 3>3 := (-1 + v^3)(s + 35)(s + 6^=3 + 53)/8 

x 3A ■= s 2 /48 + 793/48 + 35s/24 - (s + 35)v^3/2 
y 34 ;= -(1 + v^3)(s + 53 - 6 v / ^3(s + 35)/8 



Thus they generate a subgroup isomorphic to Z/3Z + Z/3Z. We can take the generators 
^3,1,^3,2 for example. Thus by [ Ke-Mo|| , (D s ) tor (Q(V— 3)) is isomorphic to one of the 
following. 

(a) Z/3Z + Z/3Z, (b) Z/3Z + Z/6Z and (c) Z/6Z + Z/6Z. 
The case (b) is forgotten in the list of [|Kc-Mo|| by an obvious type mistake. By the same 
discussion as in 5.1, there exists P £ D s with order 6 and 2P = P 31 if and only if 

A(s, m) : = s 3 + 85s 2 — 4ms 2 — 568ms + 1555s — 16m 2 s — 1136m 2 

-15465 - 20164m + 64m 3 = 

Fortunately the variety A = is again rational and we can parametrize it as 

(4.26) s = &(*), f 6 (t) :=-(27t 3 - 1304t 2 + 17920t- 71680)/(t-8)(t-16) 2 

(4.27) m = ^(t), ij}{t) := -(-128t 2 + 3t 3 + 1536* - 6144)/(t - 8)(t - 16) 2 

It turns out that the condition for the existence of Q £ D s with 2Q = P3 2 is the same 
with the existence of P, 2P = P 3i i. Assume that s = Then by an easy computation, 

we get P = (£6,1,2/6,1) and Q = (x 6j2 , 2/6,2) where 

._ 1 (-3072t 5 +11796480f 2 +86016f 4 - 1327104t 3 -56623104t+113246208+47f 6 ) 
^e. 1 - — 3 (t-8) 2 (i-16) 4 

-4t 3 (f 2 -24f+192)(7f 2 -144t+768) 

I 1 ®* 1 -~ (t-16) 5 (t-8) 2 

1 (37t 6 -2016j 5 +40704t 4 -294912f 3 - 1179648t 2 +28311552t- 113246208) 



^6,2 q 



4 



3 (t-8) 2 (t-16) 

8 v^3(t- 12) (t- 12-4 V^3) (7t-72+8V^3) (7t-72-8-/=3)t(t- 12+4-/=3) 



^ 6 - 2 :— 7 (i-16) 3 (t-8) 



3 



It is easy to see by a direct computation that 3P = 3Q = (a, 0) where 

2 (t 2 - 48* + 384) (13t 4 - 528t 3 + 8064t 2 - 55296t + 147456) 
a : ~ ~3 (t-8) 2 (t- 16) 4 

and Q — P = P 3j3 . Now we claim that 

C7otm 1. (P s )tor(Q(v /= 3)) = Z/3Z + Z/6Z with generators P 3 , 3 and P. 

13 



In fact, if the torsion is Z/6Z + Z/6Z, there exist three elements of order two. However 
fo(x) := f(x,Q) factorize as (x — a)f 0)0 (x) and their discriminants are given by 

a f 2048f 6 (t- 12) 3 (t 2 -24t +192) 3 (7t 2 - 144t+768) 6 

^xJO ■— (t-8) 9 (t-16) 18 

A*/o,o := 165888(t - 12) 3 (t 2 - 24t + 192) 3 (t - 8) 7 (t - 16) 8 

Consider quartic Q 4 : </(*,«) := 165888(t - 12) (t 2 - 24t + 192) (t - 8) - v 2 = 0. Thus £> s 
has three two torsion elements if and only if the quartic g(t,v) = has Q(V — 3)-point 
(t , vq) with t 7^ 8, 16, 12, 12 ± 4 a/— 3. The proof of Claim is reduces to: 

Assertion 1. There are no such point on Q 4 . 

Proof. By an easy birational change of coordinates, g(t,v) = is equivalent to the 
elliptic curve C := {x 3 + 1/16777216 — y 2 = 0}. We see that C has two element of 
order three, (0, ±1/4096) and three two-torsion (-1/256,0), (1/512 - l/512 v /3 3,0) and 
(1/512 + l/512-/=3, 0). Again by ||Ke-Mo|| , C tor (Q(v^3)) = Z/2Z + Z/6Z. As the rank 



of C is (| JS-Z] |), there are exactly 12 points on C. They correspond to either zeros or 
poles of A x (fo). This implies that the quartic Q 4 has no non-trivial points and thus C 
does not have three 2-torsion points. This completes the proof of the Assertion and thus 
also proves the Claim. □ 
Now we formulate our result as follows. Let A 6 = {s = £e(t)',t E Q(V — 3)} and 
E 6 := £e ^E) is given by S 6 = {8, 16, 0, 12, 12 ± 4 v / ^3, (72 ± 8v^3)/7}. 

Theorem 4.28. The M or dell-Weil torsion of D s is given by 

Z/3Z + Z/3Z s e Q( v / ^3) - A 6 U S 
Z/6Z + Z/3Z s = f 6 (t) G A 6 , te Q( v /Z 3) - S 6 



(D s ) tor (Q(V^3)) = 
The j-invariant is given by 
J(D S ) - 



1 (s + 47) 3 (s + 71) 3 (s 2 + 70s + 1657) s 
64 (s + 35) 3 (s 2 + 106s + 2917) 3 



4.5. Examples. (A) First we consider the case of elliptic curves C s . In the following 
examples, we give only the values of parameter s as the coefficients are fairly big. The 
corresponding Weierstrass normal forms are obtained by ( |4.5| ). 

1. s = 54. The curve C54 with torsion group Z/6Z has been studied in § [4.3| . 

2. Take r = 3, s — y?6,2(3) = 343/9. Then the torsion group is isomorphic to Z/6Z + 
Z/2Z with generators P 2 = (-55223/972,-588245/486) and R = (88837/972,0). The 
j-invariant is given by 7 3 • 127 3 /2 2 • 3 6 • 5 2 . 

3. Take t = —3, s = fg(— 3) = 1/216. Then the torsion group is isomorphic to Z/9Z 
and the generator P 3 = (289/559872, -7/419904). The j-invariant is 71 3 -73 3 /2 9 -3 9 -7 3 -17. 

4. Take v = 3, s = y3 12 (3) = —27/80. Then the torsion is isomorphic to Z/12Z with 
generator P 4 = (-2997/25600, -6561/102400). The j-invariant is -ll 3 • 59 3 /2 12 • 3 • 5 3 . 
(B) We consider elliptic curves D s defined over Q(V — 3). The normal form is given by 

5. Take s — 1. Then (-Di)t or (Q(\/— 3)) = Z/3Z + Z/3Z and the generators are 
(>3,i,2/3,i) = (108,756) and (x 3)2 , 2/3,2) = (-144, 756 v /= 3). The j-invariant is 2 15 3 3 /7 3 . 

14 



6. Take t = 4 and s = —299/9. Then the torsion is isomorphic to Z/6Z + Z/3Z. 
The generators can be taken as (x 6>1 ,y 6) i) = (—2351/243,-532/243) and (x 3>3 ,y 3 ^) = 
(8^/^3/9 - 2171/243, -680/81 + 248V^3~/81). The j-invariant is given by 
5 3 • 17 3 • 31 3 • 2203 3 /2 6 • 3 6 • 7 3 • 19 6 . 

4.6. Appendix. Parametrization of rational curves. Parametrizations of a rational 
curves are always possible and there exists even some programs to find a parametrization 
on Maple V. For the detail, see [|Ab-Ba|] and [|vH|1 for example. In our case, it is easy to get 
a parametrization by a direct computation. For a rational curves with degree less than or 
equal four is easy. For other case, we first decrease the degree, using suitable bitational 
maps. We give a brief indication. We remark here that the parametrization is unique up 
to a linear fractional change of the parameter. 

(1) For the parametrization of s 3 — 32s 2 — 2ms 2 — 4m 2 s + 8m 3 = 0, put m = us. 

(2) For the parametrization of 

R 3 (m, s) : = 512m 9 + 768m 8 s - 512m 6 s 3 - 1536m 6 s 2 - 192s 4 m 5 
-6144m 5 s 3 - 6528m 4 s 4 + 96s 5 m 4 - 12288m 3 s 4 - 2048m 3 s 5 + 64s 6 m 3 + 480s 6 m 2 
-15360s 5 m 2 - 6144s 6 m + 384s 7 m - 6s 8 m + 56s 8 - 512s 6 - 768s 7 - s 9 = 

put successively s = S\/m\ and m = 1/mi, then put n\ = n 2 /sl, then s\ = s 2 — 2 and 
n 2 = n i s 2 . This changes degree of our curve to be 6. Then s 2 + s 3 — 4 and n 4 = n 5 + 2 
and ri5 = uqSz. This changes our curve into a quartic. Other computation is easy. 

4.7. Further remark. Professor A. Silverberg kindly communicated us about the paper 
He gave a universal family for Z/3Z + Z/3Z over Q(\/— 3), which is given by 



A(u) : y 2 = x 3 + a (u)x + b (u) where 

a (u) = -27m(8 + u 3 ), b (u) = -54(8 + 20m 3 - u 6 ) 

and the subfamily, given by u = (4 + r 3 )/(3r 2 ), describes elliptic curves with torsion 
Z/6Z + Z/3Z. Again by an easy computation, we can show that by the change of param- 
eter s = —47+ 12m we can identify D s and A(u). Our subfamily for Z/6Z + Z/3Z is also 
the same with that of | R-S ] by the fractional change of parameter: t = 8(r — 2)/(r — 1). 

We would like to thank H. Tokunaga for the valuable discussions and informations 
about elliptic fibrations and also to K. Nakamula and T. Kishi for the information about 
elliptic curves over a number field. I am also gratefull to SIMATH for many computations. 



References 

[Ab-Ba] S. S. Abhyankar and C.L. Bajaj, Automatic parametrization of rational curves and surfaces III: 
Algebraic plane curves. Computer Aided Geometric Design 5 (1988), 309-321. 

[B-K] E. Brieskorn and H. Knorrer, Ebene Algebraische Kurven, Birkhauser (1981), Basel-Boston - 
Stuttgart. 

[D] A. Degtyarev, Alexander polynomial of a curve of degree six, J. Knot Theory and its Ramification, 

Vol. 3, No. 4, 439-454, 1994 
[vH] M. vanHoeij, Rational parametrizations of algebraic curves using a canonical divisor, J. Symbolic 

Computation (1996) 11, 1-19. 
[Ke-Mo] M. A. Kenku and F. Momose, Torsion points on elliptic curves defined over quadratic fields, 

Nagoya Math. J. Vol. 109 (1988), 125-149 



[Ko] K. Kodaira, On compact analytic surfaces II, Ann. of Math. 77 (1963) 563-626 and III, Ann. of 
Math. 78 (1963) 1-40. 

[Ku] D.S. Kubert, Universal bounds on the torsion of elliptic curves, Proc. London Math. Soc. (3) 33 
(1976), 193-237 

[Ma] B. Mazur, Rational isogenics of prime degree, Invent. Math. 44 (1978) 129-162. 

[Mi-P] R. Miranda and U. Pcrsson, On Extremal Rational Elliptic Surfaces, Math. Z. 193, 537-558 (1986) 

[N] M. Namba, Geometry of projective algebraic curves, Decker, New York, 1984 

[01] M. Oka, Flex Curves and their Applications, Geometriae Dedicata, Vol. 75 (1999), 67-100 

[02] M. Oka, Geometry of cuspidal sextics and their dual curves, to appear in Advanced Studies in 

Pure Math. 21, 1999?, Singularities and arrangements, Sapporo- Tokyo 1998. 
[R-S] K. Rubin and A. Silverberg, Mod 6 representations of elliptic curves, 213-220 in Automorphic 
Forms, Automorphic Representations and Arithmetic, Proceedings of Symposia in Pure Mathe- 
matics, vol. 66, Part 1, AMS, 1999. 
[S-Z] U. Schneiders and H. G. Zimmer, The rank of elliptic curves upon quadratic extension, Compu- 
tational number theory (1989), 239-260 
[Si] J. H. Silverman, The Arithmetic of Elliptic Curves, GTM 106, Springer, New- York, 1986. 
[W] R. Walker, Algebraic curves, Dover Publ. Inc., New York, 1949. 

[Z] H. G. Zimmer, Torsion of elliptic curves over cubic and certain biquadratic number fields, Arith- 
metic geometry, 203-220, Comtemp. Math. 174, Amer. Math. Soc. 

Department of Mathematics, Tokyo Metropolitan University 
Minami-Ohsawa, Hachioji-shi Tokyo 192-03, Japan 
E-mail address: oka@comp.metro-u.ac.jp 



16