Journal of Algebra 276 (2004) 280291

www.elsevier.com/locate/jalgebra
Algorithm for computing the moduli space
of pointed Gorenstein curves with Weierstrass
gap sequence 1, 2, . . . , g 2, , 2g 3
Francisco Luiz Rocha Pimentel
Universidade Federal do Cear, Campus do Pici, Bloco 914, CEP 60455-750 Fortaleza, Cear, Brazil
Received 7 April 2003
Communicated by Craig Huneke
Abstract
An algorithm for computing the moduli space of pointed Gorenstein curves with Weierstrass gap
sequence 1, 2, . . . , g 2, , 2g 3, where 3g 5 <2 <4g 8 and g 6, is given. This algorithm
consists in deforming a nonreduced reducible canonical curve by the method of Sthr.
2004 Elsevier Inc. All rights reserved.
Keywords: Moduli spaces; Weierstrass gap sequence; Gorenstein curves
1. Introduction
In [4] Mumford proposes to use Petris analysis of the canonical ideal to obtain the
moduli space of the smooth projective curves. Adapting Munfords idea, Sthr [10]
developed a method to construct the moduli space of pointed Gorenstein curves with
a prescribed Weierstrass gap sequence. The Sthrs method consists in deforming a
canonically embedded curve in the (g1)-dimensional projective space and using Grbner
basis techniques for describing explicitly the moduli space. Applying his method, Sthr
[10] gave a rather explicit construction of the moduli space of pointed curves with a given
symmetric Weierstrass semigroup (that is, the largest gap is equal to 2g 1) by deforming
an reduced irreducible singular curve. Later, Oliveira and Sthr [6] constructed the moduli
space of pointed Gorenstein curves with a quasi-symmetric Weierstrass semigroup (that
is, the largest gap is equal to 2g 2) by deforming a reduced reducible singular curve.
E-mail address: pimentel@mat.ufc.br.
0021-8693/$ see front matter 2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.jalgebra.2003.12.026
F.L.R. Pimentel / Journal of Algebra 276 (2004) 280291 281
In this work we follow this way and algorithmically construct the moduli space of
pointed Gorenstein curves with Weierstrass gap sequence 1, 2, . . . , g 2, , 2g 3, where
3g 5 <2 <4g 8 and g 6, by deforming a nonreduced reducible Gorenstein curve.
In Section 2 we obtain monomial bases for the vector spaces of higher order differentials
on such curves. In Section 3 we study the nonreduced reducible curve that is deformed for
obtaining the moduli space. In Sections 4 and 5 we explicitly construct the moduli space.
Furthermore, in Section 5 we present an algorithm for computing the moduli space and
exemplify it by working out the case of the gap sequence 1, 2, 3, 4, 7, 9.
2. Monomial bases for the spaces of higher order differentials
Let C be a regular, complete, irreducible, nonhyperelliptic curve of genus g dened
over an algebraically closed eld k, and let P be a point of C with Weierstrass gap
sequence l
1
, l
2
, . . . , l
g
. By the RiemannRoch theorem there exist regular differentials

l
1
,
l
2
, . . . ,
l
g
on C whose orders at P are l
1
1, l
2
1, . . . , l
g
1, respectively.
Because we assumed that C is nonhyperelliptic, we can identify it with its image under
the canonical embedding
(
l
1
:
l
2
: :
l
g
) : C P
g1
k
.
Thus C becomes a projective, nondegenerate curve of genus g and degree 2g 2 in P
g1
k
.
Let
n
C
(0) be the vector space of regular differentials of order n on C. In order to apply
the Sthrs construction, we present a monomial basis for
n
C
(0).
Proposition 2.1. Let 1, 2, . . . , g 2, , 2g 3 be the Weierstrass gap sequence of C at P,
where 3g 5 <2 <4g 8 and g 6. There exists a monomial basis for the vector space

n
C
(0), for each n 2, given by the expressions
(1)
i
l
1

a
s

b
s

n2i
2g3
(0 i n 2, 2 s l
g
), with a
s
and b
s
xed, a
s
+ b
s
= s,
a
s
b
s
, and a
s
, b
s
{1, 2, . . . , g 2, , 2g 3}.
(2)
l
i

n1
l
g
(i =1, . . . , g).
(3)
l
g

nj1

j
l
g
(j =0, 1, . . . , n 3).
(4)
nj

j
l
g
(j =0, 1, . . . , n 2).
Proof. Observing that 2l
g
is not in {1, 2, . . . , g 2, , 2g 3}, the result follows from
[8, Theorem 2.1]. 2
3. A canonical nonreduced reducible curve
Let L = {1, 2, . . . , g 2, , 2g 3}, where 3g 5 < 2 < 4g 8 and g 6. Easily
we verify that N L is an additive subsemigroup of N, i.e., L is the gap sequence of
282 F.L.R. Pimentel / Journal of Algebra 276 (2004) 280291
a numerical semigroup. We will denote by L
2
the set {2, 3, . . . , 3g 5, 2, +2g 3,
4g 6} of all sums of two gaps. For each s L
2
we consider all the partitions of s as sums
of two gaps
s =a
si
+b
si
(i =0, . . . ,
s
),
where a
si
b
si
and a
si
, b
si
L. We put a
s
= a
s0
and b
s
= b
s0
, where b
s0
is the biggest
among the b
si
s. We want to study the subscheme C
0
of P
g1
k
dened by the homogeneous
ideal I
0
of k[W
l
1
, . . . , W
l
g
] generated by the
1
2
(g 2)(g 3) quadratic forms
F
(0)
si
=W
a
si
W
b
si
W
a
s
W
b
s
(s L
2
, i =1, . . . ,
s
). (1)
We shall denote by k[W
l
1
, . . . , W
l
g
]
n
the vector space of all forms of degree n in
W
l
1
, . . . , W
l
g
, by I
0
n
the vector space of the forms of degree n in I
0
, and by
n

k[W
l
1
, . . . , W
l
g
]
n
the vector space generated by the lifting of the basis of
n
(0) obtained
in Proposition 2.1.
To apply some Grbner basis techniques, we dene a total ordering on the additive
semigroup N
g
of the exponents of the monomials in W
l
1
, . . . , W
l
g
as follows.
Denition 3.1. We put (i
1
, . . . , i
g
) (j
1
, . . . , j
g
) the rst non-zero entry of the vector
(

g
k=1
(j
k
i
k
),

g
k=1
l
k
(i
k
j
k
), i
g
j
g
, i
g1
j
g1
, . . . , i
1
j
1
) is positive.
Related to the second term appearing on the denition given above, we dene the
following.
Denition 3.2. Given a monomial M = W
i
1
l
1
. . . W
i
g
l
g
k[W
l
1
, . . . , W
l
g
], we dene the
weight of M as

g
k=1
l
k
i
k
.
Now we are ready to prove the following technical result.
Proposition 3.3.
(i) The monomials in W
l
1
, . . . , W
l
g
of degree n that are not divisible by the products
W
a
si
W
b
si
(s L
2
, i =1, . . . ,
s
) and do not belong to
n
are
W
1
W
2

, W
2
W
2

, . . . , W
2g4
W
2

multiplied by the monomials in W

1
, W

, W
l
g
of degree n 3.
(ii) If M is a monomial listed in item (i), then there exists a monomial M

n
of the
same degree and the same weight but of smaller exponent satisfying a homogeneous
(and isobaric) equation
M M

si
H
(M)
si
F
(0)
si
.
(iii) For each n 2 it is true that I
0
n

n
=0.
F.L.R. Pimentel / Journal of Algebra 276 (2004) 280291 283
Proof. (i) Let M be a monomial which is not divisible by W
a
si
W
b
si
(s L
2
, i =
1, . . . ,
s
). It is easily seen that M is not divisible by W
2
l
if l is different from 1, , and
2g 3. Let M = N.W
a
1
W
b

W
c
2g3
with N free of squares and not divisible by W
1
, W

,
and W
2g3
. Obviously N has degree smaller than 3. If b = 0, then M belongs to
n
. If
b >0, it follows that the possible divisors of N are W
2
, . . . , W
2g4
and N has degree 1.
If b =1, then M is in
n
. Finally, if b =2, we have the cases
W
1
W
2

, W
2
W
2

, . . . , W
2g4
W
2

and they can be multiplied by monomials in W

1
, W

, W
l
g
of degree n 3.
(ii) If 1 j 2g 4 , we have
W
j
W
2

W
j+(g2)
W
g2
W

_
mod I
0
3
_
W
j+(g2)
W
+(g2)(2g3)
W
2g3
_
mod I
0
3
_
W
a
s
W
b
s
W
2g3
_
mod I
0
3
_
,
where s =j +2 (2g 3).
(iii) A necessary condition to I
0
n

n
= 0 is that there exist monomials in
n
with
the same weight. Thus we should only consider the monomials W
l
g

W
nj1

W
j
l
g
and
W
nj

W
j
l
g
with j =0, 1, . . . , n3. But there is no monomial in
n
congruent to W
nj

W
j
l
g
modulo I
0
n
, except for W
nj

W
j
l
g
itself. Furthermore, the monomials that are congruent to
W
l
g

W
nj1

W
j
l
g
modulo I
0
n
are the monomials W
(i+1)(l
g
)
W
nj1+i

W
ji
l
g
, with i j
and (i +1)(l
g
) g 2. The only one in
n
is W
l
g

W
nj1

W
j
l
g
. 2
We will now obtain information about the scheme C
0
. The following proposition tells
us what are the isolated components of C
0
.
Proposition 3.4. The set Z(I
0
) of zeros of the ideal I
0
=F
(0)
si
| s L
2
and i =1, . . . ,
s
,
where the forms F
(0)
si
were dened in (1), consists in the union of the monomial curve
D =
__
a
l
g
l
1
b
l
1
1
: a
l
g
l
2
b
l
2
1
: : a
l
g
l
g
b
l
g
1
_

(a : b) P
1
k
_
with the projective line
E =
_
(0 : : 0 : a : b)

(a : b) P
1
k
_
.
Proof. It is obvious that D Z(I
0
). On the other hand, the numbers + , + l
g
,
and l
g
+ l
g
cannot be written, in a different way, as sums of two gaps. So the line
(0 : : 0 : a : b) with (a : b) P
1
k
is contained in Z(I
0
).
Conversely, let P =(w
1
: : w
l
g
) Z(I
0
). If w
1
=0 we put w
1
=1, w
2
=t and solve
recursively the system F
(0)
si
=0. Easily we nd that w
l
i
=t
l
i
1
for every i. If w
1
=0, then,
284 F.L.R. Pimentel / Journal of Algebra 276 (2004) 280291
observing that 2l
r
=l
i
r
+l
j
r
for i
r
=r and 2 r g 2, it follows that w
l
i
=0 for each
i g 2 and we conclude that Z(I
0
) DE. 2
The Proposition 3.4 shows that the scheme C
0
is a connected curve. A question that
arises naturally is whether C
0
has embedded points. The next proposition answers this
question.
Proposition3.5. The curve C
0
is arithmetically CohenMacaulay, that is, its homogeneous
coordinate ring is CohenMacaulay. In particular, the curve C
0
has only irreducible
components of dimension one.
Proof. We have to prove that the homogeneous ring
R =
k[W
l
1
, . . . , W
l
g
]
I
0
is CohenMacaulay. We know from the Proposition 3.4 that the Krull dimension of R
is 2. Thus it is sufcient showing that W
l
g
, W
l
1
W

form a R-sequence. At rst we

observe that the class of W
l
g
in R is not a zero divisor. In fact, let W
l
g
P I
0
n
. By
Proposition 3.3(ii), we can write P = Q + T with Q I
0
n1
and T
n1
. It follows
that W
l
g
T I
0
n

n
because W
l
g

n1

n
. Thus W
l
g
T =0 which implies that T =0
and P I
0
n1
. Now we suppose that (W
l
1
W

)P I
0
+(W
l
g
). By Proposition 3.3(ii),
we can write P =

s
c
s
W
n3
l
1
W
a
s
W
b
s
+dW
l
g

W
n2

+eW
n1

. Obviously the weights

of the monomials W
n3
l
1
W
a
s
W
b
s
are pairwise different. They are also different from the
weights of the other monomials in (W
l
1
W

)P, so it follows that c

s
=0 for each s. The
terms dW
l
g

W
n2

and eW
n1

have different weights and are not in I

0
+(W
l
g
). It follows
that d = e = 0 and therefore W
l
1
W

is not a zero divisor in R/(W

l
g
). We proved that
W
l
g
, W
l
1
W

form a R-sequence. In particular, the scheme C

0
is of pure dimension one
(cf. [1, Corollary 18.11]). 2
As we will see in the next paragraph, in order to obtain an effective method for the
construction of the moduli space of pointed curves with Weierstrass semigroup having the
set of gaps L ={1, . . . , g 2, , 2g 3} it is necessary that the curve C
0
be a canonical
curve (see Theorem 4.1). Following Schreyer (cf. [9, p. 85]), we dene what we mean by
a canonical curve.
Denition 3.6. A nondegenerated subscheme C P
g1
k
of pure dimension 1 is a canonical
curve of genus g if
O
C
(1)

=
C
, the dualizing sheaf,
and
dim
k
H
0
(C, O
C
) =1 and dim
k
H
0
(C,
C
) =g, the genus of the curve.
F.L.R. Pimentel / Journal of Algebra 276 (2004) 280291 285
We can now prove the main result of this paragraph.
Theorem 3.7. The curve C
0
is a canonical curve.
Proof. We know by Proposition 3.5 that C
0
has pure dimension 1. It is also true
by Proposition 3.3(ii) and (iii) that I
0
n

n
= k[W
l
1
, . . . , W
l
g
]
n
and that the Hilberts
polinomial of the curve C
0
is 2(g 1)n +1 g for each n 2. Thus the curve C
0
has
degree 2g 2 and genus g. On the other hand, putting

I
0
for the sheaf of ideals associated
to I
0
in P
g1
k
, it follows from the fact that the homogeneous coordinate ring of C
0
is
CohenMacaulay that H
1
(P
g1
k
,

I
0
) =0. From the exact sequence of sheaves
0

I
0
O
P
g1
k
O
C
0 0
we obtain that
0 H
0
_
P
g1
k
,

I
0
_
H
0
_
P
g1
k
, O
P
g1
k
_
H
0
_
P
g1
k
, O
C
0
_
0
is exact. Thus it follows from H
0
(P
g1
k
, O
P
g1
k
) = k that H
0
(P
g1
k
, O
C
0 ) = k. Now, it is
sufcient to prove that the homogeneous coordinate ring R of C
0
is a Gorenstein ring. In
fact, in this case the canonical module of R is R(n) for some n (cf. [1, p. 545]). Because

C
0

=

R(n) it follows from the duality theorem of Grothendieck that H
1
(C
0
, O
C
0)

=
H
0
(C
0
,
C
0 ) and consequently dim
k
H
0
(C
0
,

R(n)) =dim
k
H
1
(C
0
, O
C
0 ) =g, thus n =1.
We have just proved that
C
0

= O
C
0 (1) and therefore C
0
is a canonical curve. For
proving that the homogeneous coordinate ring of C
0
is Gorenstein we proceed as follows.
Because W
l
g
, W
l
1
W

is a R-sequence, it follows that R is Gorenstein if and only if

A = R/(W
l
1
W

, W
l
g
) is a Gorenstein ring. We know that dim
k
R
0
= 1, dim
k
R
1
= g,
and dim
k
R
n
=(2n 1)(g 1), for n 2, and therefore the Hilbert series of R is
F(R, t ) =
1 +(g 2)t +(g 2)t
2
+t
3
(1 t )
2
.
Thus, the Hilbert series of A is F(A, t ) =1+(g 2)t +(g 2)t
2
+t
3
and hence A can be
written as the sumof vector spaces A=A
0
A
1
A
2
A
3
, where dim
k
A
0
=dim
k
A
3
=1
and dim
k
A
1
= dim
k
A
2
= g 2. But A has dimension zero, so it is a Gorenstein ring if
and only if A
3
is the annihilator of the maximal ideal of A (cf. [1, Proposition 21.5]). It is
straightforward verifying that the following elements form a k-basis for the A
i
s:
A
0
1;
A
1
W
l
1
, W
l
2
, . . . , W
l
g2
;
A
2
W
2
l
1
, W
l
1
W
l
2
, . . . , W
l
1
W
l
g

, W
l
2
W
l
g2
, W
l
3
W
l
g2
, . . . , W
(g2)
W
l
g2
;
A
3
W
2
l
1
W
l
g

.
286 F.L.R. Pimentel / Journal of Algebra 276 (2004) 280291
Let M = M
0
+ M
1
+ M
2
+ M
3
be an element in the annihilator of (W
l
1
, . . . , W
l
g2
),
the maximal ideal of A. Obviously M
0
= 0. Writing M
1
= a
1
W
l
1
+ + a
g2
W
l
g2
, it
follows from W
l
1
M
1
= 0 that a
1
= = a
l
g

= 0. In the same way, it follows from

W
l
M
1
= 0, with l successively equals to l
2
+l
g2
(l
g
+1), l
3
+l
g2
(l
g
),
etc., that a
l
g
+1
, . . . , a
l
g2
= 0. Now, writing M
2
= b
1
W
2
l
1
+ + b
g2
W
(g2)
W
l
g2
and successively multiplying M
2
by W
l
g

, W
l
g
1
, . . . , W
l
1
, we obtain that b
1
= =
b
l
g

=0. Next, successively multiplying M

2
by W
l
g2
, W
l
g3
, . . . , W
l
g
+1
, we conclude
that b
l
g
+1
= = b
g2
= 0. Obviously W
2
l
1
W
l
g

is in the annihilator of the maximal

ideal of A and consequently this annihilator is A
3
. So, the ring A is Gorenstein. 2
Remark 3.8. By Proposition 3.4 and Theorem 3.7 we conclude that the component E
appears in the scheme C
0
with multiplicity 2.
4. Deforming the curve C
0
by the method of Sthr
In this section we apply the method of Sthr in order to obtain the moduli space of
pointed nonsingular canonical curves (C, P) where P is a point of C with gap sequence
1, 2, . . . , g 2, , 2g 3, where 3g 5 < 2 < 4g 8 and g 6. At rst, for the reader
convenience, we explain the method. The interested reader can obtain more informations
in [5,6,10].
Let C be a nonsingular, irreducible, nonhyperelliptic curve of genus g, canonically
embedded in P
g1
k
and let I (C) be its ideal. Thus I (C) is the set of polynomials f in the
indeterminates W
l
1
, . . . , W
l
g
satisfying f (
l
1
, . . . ,
l
g
) =0, where
l
1
, . . . ,
l
g
form a P-
Hermitian basis of
1
C
(0) that gives the canonical embedding. So I (C) is the homogeneous
ideal

i=2
I
n
(C), with I
n
(C) is the vector space of n-forms that vanish identically on C.
By Proposition 2.1, the homomorphism
k[W
l
1
, . . . , W
l
g
]
1
C
(0)
induced by the liftings W
l
i

l
i
, for i = 1, . . . , g, is onto for each n (that is, it is true
the theorem of Nether) and the canonically embedded curve C P
g1
k
is arithmetically
CohenMacaulay. Then we have that
dim
k
k[W
l
1
, . . . , W
l
g
]
n
I
n
(C)
=(2n 1)(g 1)
or, equivalently dim
k
I
n
(C) =
_
n+g1
n
_
(2n 1)(g 1) for each n 2. In particular, the
vector space I
2
(C) of quadratic relations has dimension ((g 2)(g 3))/2. The 3g 3
quadratic differentials
a
s

b
s
, s L
2
, appearing in Theorem 2.1 form a P-Hermitian
basis of
2
C
(0). After normalization, we obtain for each s L
2
and each i = 1, . . . ,
s
an equation

a
s
i

b
s
i
=
a
s

b
s
+

r>s
c
sir

a
r

b
r
,
F.L.R. Pimentel / Journal of Algebra 276 (2004) 280291 287
where the c
sir
s are constants in k and r varies on the elements of L
2
that are greater than s.
All the ((g 2)(g 3))/2 forms
F
si
=W
a
si
W
b
si
W
a
s
W
b
s

r>s
c
sir
W
a
r
W
b
r
(2)
vanish identically on C and are linearly independent. Thus, they form a k-basis for
the vector space of the quadratic relations I
2
(C). We also observe that setting
n

k[W
l
1
, . . . , W
l
g
]
n
for the vector space generated by the substitutions
l
i
W
l
i
, for i =
1, . . . , g, in the basis given on Theorem2.1, it follows that
n
I
n
(C) =0 and by counting
the dimensions, we obtain
k[W
l
1
, . . . , W
l
g
]
n
=
n
I
n
(C), for each positive integer n.
The method of Sthr consists in inverting the considerations done above. We assume that
there are given ((g 2)(g 3))/2 quadratic forms F
si
, similar to those in (2), whose
coefcients c
sir
s belong to the eld k, where r varies in the set of elements in L
2
that
are larger than s and the integers a
s
, b
s
, a
si
, b
si
, and
s
are dened as in the outset of
Section 3. We impose conditions on the coefcients c
sir
in order to the intersections of the
quadrics F
si
in P
g1
k
be a canonical curve C having a nonsingular point P =(1 : 0 : : 0)
where the intersection multiplicities of C with the osculating spaces of C at P are
l
1
1, l
2
1, . . . , l
g
1.
In relation to the total ordering we have dened in Section 3, the initial monomial of
F
si
is W
a
si
W
b
si
. When M is any of the cubic monomials listed in Proposition 3.3(ii), we
introduce the form
F
M
=

si
H
(M)
si
F
si
and writing F
M
as sum of non-zero isobaric forms of pairwise different weights, the
isobaric form of lowest weight in F
M
is equal to F
(0)
M
= M M

, and thus the initial

monomial in F
M
is M. Now, we write the forms F
si
and F
M
as a nite sequence
G
1
, G
2
, . . . , G
t
in such a way that the exponents of its initial monomial form a increasing
sequence in the total ordering of N
g
.
Theorem 4.1 [6, Theorem 2.2 and Corollaries]. Let I be the ideal generated by the
((g 2)(g 3))/2 quadratic forms F
si
, s L
2
, i =1, . . . ,
s
. Then
k[W
l
1
, . . . , W
l
g
]
n
=
n
+I
n
, for each n,
and the following statements are equivalent:
(i) the quadratic forms F
si
dene a non-degenerated curve C, of arithmetic genus g and
degree 2g 2 in P
g1
k
;
(ii) for each n it is true that I
n

n
=0;
288 F.L.R. Pimentel / Journal of Algebra 276 (2004) 280291
(iii) the homogeneous polynomials F
si
and F
M
form a Grbner basis of I ;
(iv) for each homogeneous isobaric syzygy of I
0
of degree n, say

si
A
si
F
(0)
si
= 0, the
remainder of

si
A
si
F
si
divided by G
1
, G
2
, . . . , G
t
is zero;
(v) each homogeneous isobaric syzygy of degree n between the isobaric forms F
si
is
induced by a homogeneous syzygy of degree n between the forms F
(0)
si
;
(vi) for n =3 it is true that I
3

3
=0;
(vii) for each linear syzygy of I
0
, say

lsi
h
lsi
W
l
F
(0)
si
=0 where the coefcients h
lsi
are
constants, the remainder of

lsi
h
lsi
W
l
F
si
divided by G
1
, G
2
, . . . , G
t
is zero.
In addition, the curve C of item (i) has a unique component passing through the non-
singular point P = (1 : 0 : : 0) whose contact orders with the hyperplanes dened by
the equations W
l
=0 are the integers l 1 with l L ={1, 2, . . . , g 2, , 2g 3}.
Proof. Since dim
k

n
= (2g 2)n + 1 g, we conclude that (i) is equivalent to the
existence of a integer n
0
such that for each n n
0
its true that I
n

n
= 0. Thus, it
follows from W
m
l
g

n

n+m
that (i) and (ii) are equivalent.
Now, let F be a homogeneous polynomial in W
l
1
, . . . , W
l
g
of degree n. Applying the
division algorithm, we get the decomposition F = G + R where G I
n
and R is the
remainder of F divided by G
1
, G
2
, . . . , G
t
which, by Proposition 3.3 belongs to
n
. If
F
n
, then F =R, and if furthermore F I
n
, then F =0 provided that G
1
, G
2
, . . . , G
t
form a Grbner basis of I . If F I
n
, then R =F G I
n
and therefore R =0 provided
that I
n

n
=0. This proves that (ii) and (iii) are equivalents.
Denoting by R the remainder in (iv), it is obvious that R I
n

n
, thus (iv) is a
consequence of (ii). Let

si
A
si
F
(0)
si
=0 be an isobaric syzygy of degree n and weight w
where A
si
for each pair (s, w) is either zero or an isobaric form of degree n2 and weight
ws. Observing that F =

si
A
si
F
si
is a sumof monomials of degree n and weight larger
than w, and denoting by R its remainder divided by G
1
, G
2
, . . . , G
t
, we can write F =

si
B
si
F
si
+R where B
si
for each pair (s, i) is the sum of monomials of degree n2 and
weight larger than ws. Thus R =

si
(A
si
B
si
)F
si
and A
si
B
si
is for each pair (s, i) a
sum of isobaric forms of degree n2 and weight not smaller than ws whose component
of weight ws is A
si
. Therefore when R =0 we obtain the desired syzygy of I . We have
proved that (v) is a consequence of (iv). Obviously (vi) is a consequence of (ii). We know
from Theorem 3.7 that C
0
is a canonical curve, thus it follows from [9, Proposition 3.2]
and from the Petris original argument (cf. [6, Proposition 1.6]) that the syzygy of I
0
are
generated by the linear syzygies and the trivial ones. So (iv) is implied by (vii).
To nish our proof we will prove that (v) implies (ii). We argument by way of
contradiction. Assume that there is an equation H =

si
A
si
F
si
where 0 =H
n
and
each A
si
is a form of degree n 2. Let w be the smallest weight occurring in H. Denoting
by H
w
, respectively A
sir
, the isobaric component of H, respectively A
si
, of weight w,
respectively of weight w s, we obtain the equation H
w
=

si
A
si(ws)
F
(0)
si
. Since C
0
is a canonical curve, in particular, we have that I
n

n
=0 and we conclude that H
w
=0,
an absurd.
Finally, by dehomogenizing W
l
1
1 we can see that the hypersurfaces given by
the polynomials F
s1
where s = 1 + l
j
(j = 3, . . . , g) intersect transversally at the point
F.L.R. Pimentel / Journal of Algebra 276 (2004) 280291 289
P = (1 : 0 : : 0). So there is a unique component of C passing through P, and it
is nonsingular at P with tangent line given by t (1 : t : 0 : : 0). If (x
l
1
: x
l
2
:
: x
l
g
), where x
l
1
= 1, x
l
2
= t , and x
l
j
t
2
kJt K (j = 3, . . . , g), is a parameterization
of this component at P, its coefcients can be computed recursively from the equations
F
1+l
(1, t , x
l
3
, . . . , x
l
g
) =x
l
+ =0. We obtain x
l
j
=t
l1
+ sum of higher order terms.
Thus the intersection multiplicity of the curve C and the hyperplanes W
l
=0 at P is equal
to l 1 for each l L. 2
5. The moduli space and an algorithm for computing it
Now we search for the moduli space of projective, nonsingular, irreducible, pointed
curves (C, P) where P is a point of C with the prescribed gap sequence 1, 2, . . . , g 2,
, 2g 3, where 3g 5 <2 <4g 8 and g 6. In order to do this, we repeat the Sthrs
construction, applying it to our case. Our use of this method is justied by the following
result.
Theorem 5.1 [6, Theorem 3.2]. The isomorphism classes of the projective, nonsingular,
irreducible, pointed curves with Weierstrass gap sequence 1, 2, . . . , g 2, , 2g3, where
3g 5 < 2 < 4g 8 and g 6 correspond bijectively to the orbits of the equivariant
G
m
(k)-action (z, c
sir
) z
rs
c
sir
on the algebraic set of the vectors of constants c
sir
normalized by [10, Proposition 3.1], and satisfying the isobaric polynomial equations (vii)
of Theorem 4.1 and the Jacobian criterion.
In the case that our semigroup is a Weierstrass semigroup we obtain the following
corollary.
Corollary 5.2 [6, Corollary 3.3]. The isomorphism of the normalized, nonsingular,
irreducible, pointed curves (C, P) are exactly of the form
(z
l
1
: : z
l
g
)
_
z
l
1
z
l
1
: : z
l
g
z
l
g
_
,
where z G
m
(k) with z
rs
=1 wherever c
sir
=0 for some i.
We do not have any proof that our semigroups are Weierstrass semigroups for all g
and or, in other words, that the open subset of the smooth curves in the moduli space
of Theorem 5.1 is not empty for some g and . As it is apparent, our procedure for
calculating these moduli spaces does not control the smoothness of the deformations and,
in our method, the Jacobian criterion is ineffective for a priori reasonings. However, it is
worth noting that the explicit description of the moduli space that the method gives to us is
a very effective tool for obtaining smooth curves that realize the semigroups as Weierstrass
semigroups, in the case they exist. In fact, because the nonsingular curves correspond to an
open subset in the moduli variety, if we choose an aleatory point in the moduli it should
correspond to a nonsingular curve for almost every choice. In the forthcoming paper [7],
G. Oliveira and the author follow these ideas to show that some numerical semigroups are
Weierstrass semigroups.
290 F.L.R. Pimentel / Journal of Algebra 276 (2004) 280291
Remark 5.3. Now we present the algorithm for computing the moduli space:
Input: The parameters g and , where 3g 5 <2 <4g 8.
Output: A nite set of equations for the coefcients c
sir
.
(1) Compute the linear syzygies of I
0
.
(2) Normalize g(g 1)/2 coefcients of the F
si
.
(3) For each linear syzygy of I
0
obtained in (1), say

lsi
h
lsi
W
l
F
(0)
si
=0, divide

lsi
h
lsi
W
l
F
si
by G
1
, G
2
, . . . , G
t
and output the remainders.
Comments: Imposing that the coefcients of the output are zero, we obtain the
set of equations describing the moduli space. The correctness follows from
Theorem4.1(vii) and [10, Proposition 3.1]. There are several algorithms for doing
steps 1 and 3. It is worth noting that the choices made in step 2 affect the output.
To conclude our work we illustrate the method with an example. We choose L ={1, 2,
3, 4, 7, 9}, where g =6 and =7. The ideal I (C) is generated by ((g 2)(g 3))/2 =6
quadratic forms F
s
=F
s1
where s {4, 5, 6, 8, 10, 11}. Using Theorem 5.1, we normalise
the following (g(g 1))/2 =15 coefcients: the coefcient c
11,1,14
in F
11
, the coefcients
c
10,1,11
and c
10,1,12
in F
10
, the coefcients c
8,1,9
and c
8,1,14
in F
8
, the coefcients
c
5,1,6
, c
5,1,7
, c
5,1,8
, c
5,1,9
, c
5,1,10
, c
5,1,11
, c
5,1,12
, and c
5,1,13
in F
5
, and the coefcients
c
4,1,6
and c
4,1,9
in F
4
. By the item (vii) of Theorem 4.1, the remainders of
W
3
F
11,1
W
4
F
10,1
+W
9
F
5,1
, W
2
F
11,1
W
3
F
10,1
+W
7
F
6,1
+W
9
F
4,1
,
W
1
F
11,1
W
2
F
10,1
+W
7
F
5,1
, W
2
F
6,1
W
3
F
5,1
+W
4
F
4,1
,
W
1
F
6,1
W
2
F
5,1
+W
3
F
4,1
divided by the Grbner basis composed by the F
s
s and the F
M
s are zero. After some
computations we can translate these conditions by saying that the forms F
s
depend on 12
constants a
1
, a
4
, a
6
, a
7
, a
10
, a
12
, b
2
, b
3
, b
4
, b
5
, b
8
and b
10
, in the following way:
F
4,1
=W
2
2
W
1
W
3
a
1
W
1
W
4
a
4
W
1
W
7
a
6
W
1
W
9
a
7
W
2
W
9
a
10
W
2
7
a
12
W
7
W
9
,
F
5,1
=W
2
W
3
W
1
W
4
,
F
6,1
=W
2
3
W
2
W
4
+a
1
W
3
W
4
+a
4
W
1
W
9
+a
6
W
3
W
9
+a
7
W
4
W
9
+a
10
W
7
W
9
+a
12
W
2
9
,
F
8,1
=W
2
4
W
1
W
7
b
2
W
1
W
9
b
3
W
2
W
9
b
4
W
3
W
9
b
5
W
4
W
9
b
8
W
7
W
9
b
10
W
2
9
,
F
10,1
=W
3
W
7
W
1
W
9
,
F
11,1
=W
4
W
7
W
2
W
9
.
F.L.R. Pimentel / Journal of Algebra 276 (2004) 280291 291
So, the moduli space of pointed nonsingular irreducible curves with Weierstrass gap
sequence 1, 2, 3, 4, 7, 9 is an open subvariety, given by the Jacobian criterion, of the
projective space of dimension 11. Since g = 6, our numerical semigroup is represented
as a Weierstrass semigroup of a nonsingular curve (cf. [3]). Therefore, the open subvariety
is not empty and consequently has dimension 11. It is worth recalling that the expected
dimension in this case is (3g 2) weight of the semigroup = 16 5 = 11, and the
semigroup N =N{1, 2, 3, 4, 7, 9} is dimensionally proper (cf. [2, p. 496]). It is apparent
that what we constructed was a compactication of the moduli variety. An interesting
question is knowing whether in the procedure of compactifying the moduli space of pointed
nonsingular irreducible curves we introduced points that do not correspond to canonical
curves. In [6] it is given a criterion to decide this question in the case of quasi-symmetric
semigroups (the existence of certain cubic). In our case we can apply the methods used in
Section 3 to study the curve C
0
. After some calculations we can verify that all the curves
dened by the equations F
s
s are canonical curves. We can say that we obtained the moduli
space of pointed canonical curves (C, P) with P a nonsingular point of C whose osculating
spaces of C at P pass through P with intersection multiplicity equal to 0, 1, 2, 3, 6, 8 and
nite automorphismgroup. Observe that this construction excludes the curve C
0
(improper
point in the moduli variety) that has G
m
(k) as its group of automorphisms.
Acknowledgments
I express my great gratitude to Professor Karl-Otto Sthr from whom I learned this
subject. I also thank Gilvan Oliveira for helpful comments.
References
[1] D. Eisenbud, Commutative Algebra with a view toward Algebraic Geometry, in: Grad. Texts in Math.,
vol. 150, Springer-Verlag, Berlin, 1995.
[2] D. Eisenbud, H. Harris, Existence, decomposition, and limits of certain Weierstrass points, Invent. Math. 87
(1987) 495515.
[3] J. Komeda, On the existence of Weierstrass gaps sequences on curves of genus 8, J. Pure Appl. Algebra 97
(1994) 5171.
[4] D. Mumford, Curves and their Jacobians, Univ. of Michigan Press, Ann Arbor, 1975.
[5] G. Oliveira, K.-O. Sthr, Gorenstein curves with quasi-symmetric Weierstrass semigroups, Geom.
Dedicata 67 (1997) 4563.
[6] G. Oliveira, K.-O. Sthr, Moduli spaces of curves with quasi-symmetric Weierstrass gap sequences, Geom.
Dedicata 67 (1997) 6582.
[7] F. Pimentel, G. Oliveira, The semigroups 6, 8, 9, 10 and 6, 8, 10, 11 are Weierstrass semigroups, preprint.
[8] F. Pimentel, Intersection divisors of a canonically embedded curve with its osculating spaces, Geom.
Dedicata 85 (2001) 125134.
[9] F.O. Schreyer, A standard basis approach to syzygies of canonical curves, J. Reine Angew. Math. 421 (1991)
83123.
[10] K.-O. Sthr, On the moduli spaces of Gorenstein curves with symmetric Weierstrass semigroups, J. Reine
Angew. Math. 441 (1993) 189213.