Notes on Stable Curves

Tabes Bridges
Spring 2014

First Properties
Definition. A stable curve is a complete, reduced, separated scheme C of dimension 1
over C such that

• The singularities of C are at worst ordinary nodes

• |Aut(C)| is finite.

A pointed stable curve is a tuple (C, p1 , ..., pn ) where C is a stable curve, the pi are distinct
points in the smooth locus of C, and the finiteness condition is instead imposed on the group
of automorphisms fixing each pi .

We are concerned with the coarse moduli space of stable curves of genus g, denoted Mg ,
and the moduli space of pointed stable curves of genus g with n markings, denoted Mg,n .
These are irreducible projective compactifications of Mg and Mg,n , the moduli spaces of
smooth (resp. pointed smooth) curves. We have the following:

Theorem. (Hurwitz) If C is a curve of general type, then |Aut(C)| ≤ 84(g − 1).

It follows that every curve of general type is stable. If g = 0, recall that Aut(C) = P GL(2, C),
which has dimension 3. It is therefore necessary to take at least 3 marked points to obtain a
stable rational curve, meaning that M0,3 is a point and M0,4 is the first nontrivial moduli
space of rational curves, and has dimension 1 (being isomorphic to the complement of 3 points
in P1 . In general, dim Mg,n+1 = dim Mg,n + 1 (Note: since smooth loci are dense, statements
about dimension for moduli spaces of smooth curves also apply to the compactified spaces).
For an elliptic curve E, we have an automorphism x 7→ x + x0 for each x0 ∈ E, and
the standard elliptic involution. The special curves with j = 0, 1728 have a few more au-
tomorphisms, but the upshot is that Aut(E) is a finite extension of the group E, hence
1-dimensional, so we must have at least 1 marking to obtain a stable curve. The space M1,1
has dimension 1. By the theorem of Hurwitz, we needn’t worry about stablity for curves of
general type, so the fact that marking a point increases the dimension by 1 and the following
result completely determine the dimensions of these spaces.

Theorem. (Riemann) The moduli space of genus g > 1 curves has dimension 3g − 3.

1
Proof. We count branched covers of P1 of a fixed type. By the Riemann Existence Theorem,
such a cover is determined (modulo a finite amount of monodromy data) by its branch
points, so the dimension of the space of such things with a fixed number b of branch points is
dim M0,b = b − 3. A sufficiently large choice of d forces all branching to be simple, so by the
Riemann-Hurwitz formula, a degree d >> 0 cover by a curve of genus g has b = 2d+2g −2, so
dim M0,b = 2d + 2g − 5. On the other hand, such a cover is determined by a triple (C, L, V ),
where C is a curve of genus g, L is a line bundle on X, and V ⊂ H 0 (L) is a two dimensional
subspace. Since · ⊗OC OC (d) is an isomorphism P ic0 (C) → P icd (C), there is a g-dimensional
space of degree d line bundles. For d >> 0, by Riemann-Roch we have h0 (C) = d − g + 1, and
the Grassmannian of two dimensional subspaces of H 0 (L) has dimension 2(d − g − 1). Thus
the space of triples (C, L, V ) has dimension dim Mg + d + 2(d − g − 1). Thus we conclude

2d + 2g − 5 = dim Mg + g + 2d − 2g − 2 ⇒ dim Mg = 3g − 3.

We now consider the problem of classifying different flavors of stable curves of a given
genus by their topological type. To do this, we associate to a curve C its dual graph G(C),
which is constructed according to the following formula:

• To each irreducible component Ci of C, we associate a vertex vi labeled with an integer

g(Ci ), the geometric genus of Ci .

• To each node νj , we associated an edge ej . Thus, if νj is the intersection Ci ∩ Ck of

two irreducible components, ej connects vi and vk . If, on the other hand, νj is a double
point of a single irreducible component Ci , then ej is a cycle based at vi .

A graph G with vertices

P labeled by nonnegative integers gi is the dual graph of a stable curve
of genus g if b1 (G) + gi = g.

Stable Curves of Genus 2

We will show that such a curve cannot have more than 2 irreducible components. We will
classify dual graphs by considering ordered partitions of the number 2, first of the form
(b1 (G), g1 ) then of the form (b1 (g), g1 , g2 ). The possibilities of the first form are (2, 0), (1, 1),
and (0, 2), and in fact each of these is a valid configuration. The graph

2•

corresponds to a smooth curve of genus 2. This is the dense locus M2 ⊂ M2 of dimension

3 · 2 − 3 = 3. The graph

•1

2
corresponds to an elliptic curve with two points identified. Its normalization is an elliptic
curve with two markings, so the locus of such curves in the moduli space has dimension
dim M1,2 = dim M1,1 + 1 = 2. The graph

•0

corresponds to a rational curve with two pairs of points identified, which normalizes to a
rational curve with 4 markings, so as noted above this locus has dimension 1. Moving on
now to stable curves with two components, we have (not repeating those which differ by
swapping the second and third slots, since interchanging the components describes the same
class of curves) the possible configurations (2, 0, 0), (1, 1, 0), (0, 2, 0), and (0, 1, 1). Since any
rational component must have 3 or more markings in the normalization, the valence of any
vertex labeled 0 must be at least 3. This condition excludes the configuration (0, 2, 0). The
graph
1• •1

corresponds to a pair of smooth elliptic curves meeting at a point. This normalizes to a pair
of elliptic curves with 1 marking each, so the locus of such curves has dimension 2. The
graph
1• •0

corresponds to a rational curve with two points identified and meeting a smooth elliptic curve
at a third point. The normalization of this curve is the disjoint union of a smooth elliptic
curve with a marking and a smooth rational curve with three markings, so the dimension of
this locus is 1. Finally, the graph

0• •0

corresponds to a pair of rational curves meeting at a point, and each additionally has a pair
of points identified. This configuration normalizes to a pair of smooth rational curves, each
with three markings, so this locus is 0 dimensional, i.e. any two such curves are isomorphic.
Indeed, the (2, 0, 0) locus is the unique intersection point of the (1, 1, 0) and (2, 0) loci.

ACGH 1.A-3
The affine curve cut out by y 2 = x3 + 1 has projective closure C cut out by y 2 z = x3 + z 3 .
Intersecting with the hyperplane at infinity z = 0, we get 0 = x3 , so the only added point
is (0 : 1 : 0). Let f = x3 + z 3 − y 2 z. The partial derivatives are fx = 3x2 , fy = −2yz, and

3
fz = 3z 2 − y 2 . Since fz 6= 0 at the point at infinity, C is smooth there. Specializing to A2 ,
we have fx = 3x2 and fy = −2y, but (0, 0) is not on our curve. Now set

p1 , p2 , p2 = (−1, 0), (−ω, 0), (−ω 2 , 0), ω = e2πi/3

q1 , q2 = (0, 1), (0, −1)

r = π −1 (∞)
s1 , s2 = (2, 3), (2, −3).

Exercise. (i) Find the branch points of π : C → P1 and show that g(C) = 1.

Solution. The map is given explicitly by π(x : y : z) = (x : z). Since we are projecting
from the point (0 : 1 : 0), we are projecting along lines through that point, each of which
generically intersects C in two other points, so deg π = 2. To find π −1 (0 : 1), plug in and
get y 2 = 1, so π −1 (0 : 1) = {(0, 1), (0, −1)}. Now consider the point at infinity r = (1 : 0),
which is the image of the point at infinity in C ⊂ P2 since π is the projective closure of an
affine projection map which takes the line at infinity in P2 to the point at infinity on P1 . We
claim that r is a branch point. Consider a loop around the point on the Riemann sphere,
which is given by M eiθ for M sufficiently large as θ ranges from 0 to 2π. Taking the preimage
of (M eiθ : 1), we get y 2 ≈ M 3 e3iθ . Something something r is a branch point. Finally, let
3 3
(a : b) be any point with a, b 6= 0. Then we have y 2 = a +b b
, which has two solutions unless
3 3
a + b = 0, so the other branch points are p1 , p2 , p3 .
We have four branch points, each with simple branching (this is the only possibility since
the map has degree 2), so by Riemann-Hurwitz we have

2 − 2g(C) = 2(2) − 4 = 0 ⇒ g(C) = 1.

Exercise. (ii) Establish linear equivalences

2p1 ∼ 2p2 ∼ 2p3 ∼ q1 + q2 ∼ 2r ∼ s1 + s2

p1 + p2 + p3 ∼ 3r
q1 + s1 ∼ q2 + s2 .

Solution. Each divisor in the first string of equivalences is the pullback of a point on P1 ;
those of the form 2p are the ramification points. For the second equivalence, note that the
line y = 0 determines the divisor p1 + p2 + p3 , while the line z = 0 intersects C uniquely in
the point at infinity, so the order of contact is 3. Since any two hyperplane sections are linear
equivalent, we are done. Finally, the lines y = x + 1 and y = −x − 1 give the equivalence
q1 + s1 + p1 ∼ q2 + s2 + p1 , so we are done. .

Exercise. (iii) Determine OC (p1 + q1 ) and describe the linear system |p1 + q1 |.

4
Solution. As we saw in the previous part, the line through p1 and q1 uniquely intersects C
in a third point, s1 . Now let a, b, c denote the intersection points of another line with C, so
we have
p 1 + q1 + s1 ∼ a + b + c ⇒ p1 + q1 ∼ a + b + c − s1 .
Since a complete linear system consists of effective divisors, we see that |p1 + q1 | consists of
all degree 3 effective divisors which arise from intersecting C with a line in the pencil

lines in P2 which pass through s1 ,



which we can write as a linear combination of two lines through s1 , e.g.

α(x − y + z) + β(x − 2z) = 0,

where (α : β) ∈ P1 is a parameter. Thus OC (p1 + q1 ) = OX (s1 )|C ? .

Exercise. (iv) Find the points t ∈ C such that

t ∼ p1 + q2 − r,

t ∼ 2s1 − r.
Solution. Since the line through p1 and q2 also meets s2 , we are looking for t so that t+r+s2 ∼
p1 + q2 + s2 . The line through r = (0 : 1 : 0) and s2 = (2 : −3 : 1), is 2z − x = 0.
Plugging this into our equation, we have y 2 z = (2z)3 + z 3 = 9z 3 , so we want to solve
y 2 z − 9z 3 = z(y 2 − 9z 2 ) = 0. The solution for z = 0 is r, which we already have. The
solutions for y 2 − 9z 2 = 0 are (2 : −3 : 1) and (2 : 3 : 1), so t = s1 .
Similarly, the line with contact order 2 to s1 is y + z = 2x. The remaining intersection
point is q2 , so we are concerned with t so that t + r + q2 ∼ 2s1 + q2 . The line through r and
q2 is x = 0. Plugging in, we have y 2 z − z 3 = z(y 2 − z 2 ) = 0. The z = 0 solution is r, which
we already have. The solutions for y 2 − z 2 = 0 are (0 : 1 : 1) and (0 : −1 : 1), so t = q1 . .

ACGH 1.A-4
The affine curve cut out by y 3 + x3 = 1 has projective closure E cut out by y 3 + x3 = z 3 .
To compute the number of added points in the projective closure, we intersect with z = 0.
Fixing a reference hyperplane x = 1, we are left with y 3 = −1, so the added points are
(1 : −1 : 0), (1 : −ω : 0), and (1 : −ω 2 : 0) where ω = e2πi/3 . The projection map π : E → P1
given by π(x : y : z) = (x : z) has degree 3 since the point of projection (0 : 1 : 0) does not
lie on E.
Exercise. (i) Find the branch points (in P1 ) and ramification points (in E) of π : E → P1
and show that g(E) = 1.
Solution. The preimage of (0 : 1) is given by {y 3 = 1} = {(0 : 1 : 1), (0 : ω : 1), (0 : ω 2 : 1)}.
The preimage of (1 : 0) is given by {y 3 = −1}; this is the intersection with the line at infinity,
i.e. the three points in the projective closure which we earlier computed. The preimage of
(a : b) is {y 3 = b3 − a3 }, which, over any reasonable field, plainly has cardinality 3 unless

5
a3 = b3 , so π is branched over (1 : 1), (1 : ω), and (1 : ω 2 ). The corresponding ramification
points are (1 : 0 : 1), (1 : 0 : ω), (1 : 0 : ω 2 ), each of which has double ramification, so
χ(C) = 3 ·χ(P1 ) −2 · 3 = 0 ⇒ g(C) = 1.
| {z }
=2
.
Exercise. (ii) For any two points p, q ∈ E find the complete linear series |p + q|.
Solution. The line connecting p and q intersects E in a unique third point r, so p + q + r ∼
a + b + c for any other triple a, b, c ∈ E lying on a line in P2 . Thus p + q ∼ a + b + c − r for
any such a, b, c. Since |p + q| consists of effective divisors, we observe that one of a, b, c must
be r to lie in the linear series, so it arises from the pencil of lines in P2 which pass through
r. .
Exercise. (iii) Find a map η : E → P1 of degree two such that η(1 : 0 : 1) = η(0 : 1 : 1), and
determine the branch points of η.
Solution. Since a general line in P2 intersects E in three points, it clearly suffices to connect
the two points of interest with a line, and project from the third point of intersection. The
line through (0 : 1 : 1) and (1 : 0 : 1) is x + y = z; plugging in to the equation for E, we get
x2 y + y 2 x = xy(x + y) = 0,
so the third intersection point is (1 : −1 : 0), thus we define η to be projection from p = (1 :
−1 : 0) ∈ E. In coordinates, this is (x : y : z) 7→ (x + y : z)? .
Exercise. (iv) Using A-3(i) and (ii) show that E ∼ = C and find an explicit isomorphism.
Solution. Maybe η has the same branch points as the projection map in A-3? It would
make sense that there is no room for ambiguity in the monodromy since there are just two
sheets. .

ACGH 1.A-5
The affine curve cut out by y 2 = x6 − 1 has projective closure C̃ given by y 2 z 4 = x6 − z 6 .
Intersecting with the hyperplane at infinity z = 0 we get 0 = x6 , so we add one point.
Set f = x6 − z 6 − y 2 z 4 , so we have √ partial
√ derivatives
√ √ fx = 6x5 , fy = −2yz 4 , and fz =
6z 5 − 4y 2 z 3 = 2z 3 (3z 2 − 2y 2 ) = 2z 3 ( 3z − 2y)( 3z + 2y). Passing to A2 to analyze most
of the curve, we see that the only common zero of the dehomogenized fx and fy is (0, 0) 6∈ C̃.
On the other hand, the point at infinity (0 : 1 : 0) is a common zero for all derivatives. The
vanishing of fy |A2 = −2y is to first order, so we conclude that this point is an ordinary node.
Since the projective closure added one point, we conclude that the normalization adds two
points to make C̃ into a compact Riemann surface C. Let π : C → P1 denote projection to
the x-axis, i.e. projection from the node (0 : 1 : 0), so this map has degree 4? Set
rα = (ξ α , 0), ξ = eπi/3
p + q = π −1 (∞)
s1 , s2 = (0, i), (0, −i).

6
 P5
Exercise. (i) Show that p + q ∼ 2rα and α=0 rα ∼ 3p + 3q.

Solution. p + q is the pull back of a point. We claim that 2rα is as well. The formula for
this projection is (x : y : z) 7→ (x : z), so the image of rα is (ξ α : 1). By my estimation the
correct equivalence seems to be p + q ∼ 4rα . .

Exercise. (ii) The line through r0 , r2 , r4 is y = 0. Intersecting this line with C, we get 0 =
x6 − 1, so the other intersection points are r1 , r3 , r5 . Since an arbitrary divisor D ∼ {y = 0}
implies r0 + r2 + r4 ∼ D − r1 − r3 − r5 . To determine the complete linear system, we need to
find all such effective configurations, so we need D to be supported at r1 , r3 , r5 .

Exercise. Resolve the singularity of C̃ explicitly.

Solution. The projective curve y 2 z 4 = x6 − z 6 has a singularity at (0 : 1 : 0). Passing to the

affine patch y = 1, we get the equation x6 − z 6 − z 4 = 0; since the lowest order homogeneous
component is z 4 , the tangent cone at this point is a quadruple line z 4 = 0. The blowup  of
2 z
A in the origin is covered by affine patches U1 and U2 with coordinate rings k x, x and
k xz , z . On U1 , the equation can be written as

z 6 z 4 z 6 z 4
0 = x6 − x6 − x4 = x4 (x2 − x2





x x x
− x
),

While on U2 , the equation can be written as

6 x 6
0 = z 6 xz − z 6 − z 4 = z 4 (z 2 − z 2 − 1).


z

We extract equations for the blown up curve: on U1 with coordinates (u, v), we have f (u, v) =
u2 − u2 v 6 − v 4 = 0, and on U2 with coordinates (s, t), we have g(s, t) = s2 t6 − s2 − 1 = 0. We
have partials
fu = 2u(1 − v 6 ) ; fv = −2v 3 (3u2 v 2 + 2)
gs = 2s(t6 − 1) ; gt = 6s2 t5 .
Now V (g) is clearly smooth, but V (f ) has a singularity at the origin with tangent cone
defined by u2 = 0. Now we will blow up V (f ). We will recycle the terminology of U1 , U2 .
The equation for f on U1 can be written as

6
4 
6
4 
0 = u2 − u2 v 6 − v 4 = u2 − u8 uv − u4 uv = u2 1 − u6 uv − u2 uv

and on U2
 
2 2 6 4 2 u 2 u 2 8 4 2 u 2 u 2 6 2





0=u −u v −v =v v
− v
v −v =v v
− v
v −v .

We again extract equations; on U1 with coordinates (x, y), we have h(x, y) = 1−x6 y 6 −x2 y 4 =
0, and on U2 with coordinates (a, b) we have i(a, b) = a2 − a2 b6 − b2 = 0. We compute partials

hx = 2xy 4 (3x4 y 2 − 1) ; hy = 2x2 y 3 (3x4 y 2 − 2)

ia = 2a(1 − b6 ) ; ib = −2b(3a2 b4 − 1)

7
V (h) looks completely intractable algebraically. The real points are smooth, so I’m going to
assume it’s smooth. V (i) has tangent cone a2 − b2 = (a − b)(a + b). We blow up V (i): on
U1 , we write
a2 − a2 b6 = a2 − a2 ( ab )6 − a2 ( ab )2 = a2 (1 − a6 ( ab )6 − ( ab )2 ).
On U2 , we write

a2 − a2 b6 = ( ab )2 b2 − ( ab )2 b8 − b2 = b2 (( ab )2 − ( ab )2 b6 − 1).

On U1 , the equation of the strict transform is F = 1−x6 y 6 −y 2 = 0, while on U2 the equation

is G = z 2 − z 2 w6 − 1. We have partials

Fx = −6x5 y 6 ; Fy = −y(6x6 y + 2)

Gz = 2z(1 − w6 ) ; Gw = −6z 2 w5 .
Fx vanishes on the axes. F (0, y) = 1 − y 2 , so the possible singular points are (0, ±1). But
these do not kill Fy . F (x, 0) = 1, so there are no possible singular points. G is smooth more
or less by inspection, so we have fully resolved the singularity. .