The Moduli Stack of A - Stable Curves: Michele Pernice

The moduli stack of Ar -stable curves
Michele Pernice
arXiv:2302.10877v1 [math.AG] 21 Feb 2023
Abstract
This paper is the first in a series of four papers aiming to describe the (almost integral)
Chow ring of M3 , the moduli stack of stable curves of genus 3. In this paper, we introduce
the moduli stack M frg,n of n-pointed Ar -stable curves and extend some classical results
about Mg,n to M fr , namely the existence of the contraction morphism. Moreover, we
g,n
describe the normalization of the locally closed substack of M fr parametrizing curves
g,n
with Ah -singularities for a fixed h 6 r.
Introduction
The geometry of the moduli spaces of curves has always been the subject of intensive investigations,
because of its manifold implications, for instance in the study of families of curves. One of the
main aspects of this investigation is the intersection theory of these spaces, which can used
to solve either geometric, enumerative or arithmetic problems regarding families of curves. In
his groundbreaking paper [Mum83], Mumford introduced the intersection theory with rational
coefficients for the moduli spaces of stable curves. After almost two decades, Edidin and Graham
introduced in [EG98] the intersection theory of global quotient stacks (therefore in particular for
moduli stacks of stable curves) with integer coefficients.
To date, several computations have been carried out. While the rational Chow ring of Mg ,
the moduli space of smooth curves, is known for 2 6 g 6 9 ([Mum83], [Fab90], [Iza95], [PV13],
[CL21]), the complete description of the rational Chow ring of Mg , the moduli space of stable
curves, has been obtained only for genus 2 by Mumford and for genus 3 by Faber in [Fab90]. As
expected, the integral Chow ring is even harder to compute: the only complete description of the
integral Chow ring of the moduli stack of stable curves is the case of M2 , obtained by Larson in
[Lar19] and subsequently with a different strategy by Di Lorenzo and Vistoli in [DLV21]. It is
also worth mentioning the result of Di Lorenzo, Pernice and Vistoli regarding the integral Chow
ring of M2,1 , see [DLPV21].
The aim of this series of four papers is to describe the Chow ring with Z[1/6]-coefficients of
the moduli stack M3 of stable genus 3 curves. This provides a refinement of the result of Faber
with a completely indipendent method, which has the potential to give a more algorithmic way
to compute these Chow rings. The approach is a generalization of the one used in [DLPV21]: we
introduce an Artin stack, which is called the stack of Ar -stable curves, where we allow curves
with Ar -singularities to appear. The idea is to compute the Chow ring of this newly introduced
2020 Mathematics Subject Classification 14H10 (primary), 14H20 (secondary)

Keywords: Moduli stack of curves, Ar -singularities, Contraction Morphism
Michele Pernice
stack in the genus 3 case and then, using localization sequence, find a description for the Chow
ring of M3 . The stack M
fg,n introduced in [DLPV21] is cointained as an open substack inside our
stack.
Outline of the paper

This is the first paper in the series. It focuses on introducing the moduli stack M fr of Ar -stable
g
curves: this is the new actor for the computation of the Chow ring of M3 . Firstly, we prove
several results which are straightforward generalizations of the stable case. We construct the
contraction morphism for the Ar -stable case, which is fundamental for some of the computations
of the Chow ring of M f7 . Finally, we describe the locally closed substack parametrizing curves
3
with An -singularities for a fixed n. The description will be fundamental in the fourth paper
to understand the ideal of relations in the Chow ring of M f7 coming from the closed substack
3
M 7
f r M3 and, using localization sequence, to eventually get the Chow ring of M3 as a quotient
3
of the Chow ring of M f7 .
3
Specifically, Section 1 is dedicated to introducing Ar -singularity (and Ar -stable curves) and
studying contractions in the setting of Ar -stable curves. In particular, we extend some of the
results cointained in the first two chapter of [Knu83] to our setting, simplifying the proofs where
possible. The main result in this section is the existence and uniqueness of (minimal) contractions.
In Section 2, we define the moduli stack M frg,n of n-pointed Ar -stable curves of genus g and
prove that it is a smooth global quotient stack. We also prove that it shares some features with
the classic stable case, for example the existence of the Hodge bundle. The main theorem of this
section is that the contraction morphism is an open immersion (instead of an isomorphism as in
the stable case).
Section 3 focuses on proving some preliminary results needed for Section 4. It is divided in
two indipendent parts. In the first one, we describe a variant of the stack of finite flat algebras,
namely the stack of finite flat extensions of degree d of finite flat (curvilinear) algebras and prove
that it is an affine bundle over a classifying stack. In the second one, we prove some technical
results regarding blowups and pushouts that are probably well-known to experts, aiming to verify
that the two constructions behave well in families and are inverse of each other in our setting.
Finally, in Section 4 we describe the closed complement of Mg,n inside M frg,n , stratifying it
by complexity of the singularities that can appear in the curves. We use pushouts and blowups
to describe the locally closed substack parametrizing curves with a Ah -singularity and we prove
that it is (up to an explicit finite étale cover) an affine bundle over some moduli stack of curves
(which is also described completely) with stricly lower genus.
1. Ar -stable curves and contractions
Fix a nonnegative integer r.
Definition 1.1. Let k be an algebraically closed field and C/k be a proper reduced connected one-
dimensional scheme over k. We say the C is an Ar -prestable curve if it has at most Ar -singularity,
i.e. for every closed point p ∈ C(k), we have an isomorphism
bC,p ' k[[x, y]]/(y 2 − xh+1 )
O
with 0 6 h 6 r. Moreover, suppose Σ ⊂ C is an effective Cartier divisor whose support is in
the smooth locus of a Ar -prestable curve C. We say that the pair (C, Σ) is Ar -semistable if the
2
following two conditions are verified:

– for every irreducible component Γ of C the integer aΓ := deg ωC (Σ)|Γ is nonnegative;
– there exists at most one irreducible component Γ0 such that aΓ0 = 0; such component is
called the semistable component.
we say that a pair is Ar -stable if aΓ > 0 for every irreducible component Γ of C.
Finally, a n-pointed Ar -stable curve over k is Ar -prestable curve together with n smooth distinct
closed points p1 , . . . , pn such that the pair (C, OC (p1 + · · · + pn )) is Ar -stable, i.e. ωC (p1 + · · · + pn )
is ample. If n = 0, we just say that C is a Ar -stable curve.
Remark 1.2. Notice that a Ar -prestable curve is l.c.i by definition, therefore the dualizing
complex is in fact a line bundle.
Notice that every time we talk about genus, we intend arithmetic genus, unless specified
otherwise.
Remark 1.3. Let C be a connected, reduced, one-dimensional, proper scheme over an algebraically
e → C the
closed field. Let p be a rational point of C which is a Ar -singularity. We denote by b : C
partial normalization at the point p and by Jb the conductor ideal of b. Then a straightforward
computation shows that
(i) if r = 2h, then g(C) = g(C)e + h;
(ii) if r = 2h + 1 and Ce is connected, then g(C) = g(C)e + h + 1,
(iii) if r = 2h + 1 and Ce is not connected, then g(C) = g(C) e + h.
If Ce is not connected, we say that p is a separating point. Furthermore, Noether formula gives us
that b∗ ωC ' ωCe (Jb∨ ).
The rest of this section is dedicated to extend to theory of contractions of n-pointed (A1 -
)stable curves to the case of Ar -stable curves. More generally, we will prove several results about
contractions of Ar -semistable pairs over algebraically closed fields, namely the existence and
unicity of contractions.
The following results are the natural generalizations of the ones in the first section of [Knu83],
where the author proves them in the context of classical stable curves. We follow Knutsen’s
strategy, simplifying where possible.
Definition 1.4. Let (C, Σ) be a Ar -semistable pair. Then a contraction of such a pair is a
morphism c : C → C 0 of Ar -prestable curves such that
(i) if aΓ > 0 for every irreducible component Γ of C, then f is an isomorphism;
(ii) if aΓ = 0 for some irreducible component Γ0 then:
a) c(Γ0 ) is contracted to a closed point p0 and c−1 (p0 ) ' Γ0 (schematically);
b) the restriction
c|CrΓ0 : C r Γ0 −→ C 0 r p0
is an isomorphism.
Remark 1.5. It is important to remark that if (C, Σ) is a Ar -semistable pair and Γ is an

irreducible component of C, then aΓ = 0 implies that g(Γ) = 0 and therefore Γ is a projective
line. Moreover, we can have only three possible scenarios:
3
Michele Pernice
(1) Γ intersects the rest of the curve in one node and Σ ∩ Γ is the divisor associated to a point
in Γ (different from the node);
(2) Γ intersects the rest of the curve in two nodes and Σ ∩ Γ = ∅;
(3) Γ interescts the rest of the curve in one tacnode and Σ ∩ Γ = ∅.
The cases (1) and (2) come from the classical stable case and they were deeply studied by Knutsen
in [Knu83]; these components are denoted respectively rational tails and rational bridges. The
only new components are of type (3), which is the limit of case (2) when the two nodes coincide,
and we will denote such components as rational almost-bridges.
Remark 1.6. Using the previous remark, one can see that if c : C → C 0 is a contraction of a
Ar -semistable pair (C, Σ), then c(Σ) ⊂ C 0 is still an effective Cartier divisor and the morphism
c|Σ : Σ → c(Σ) is an isomorphism. Moreover, it follows easily that (C 0 , c(Σ)) is a Ar -stable pair.
First of all, we want to prove the existence of such contractions. To do so, we need a technical
result, which will be very useful later on. This result is a generalization of Theorem 1.8 of [Knu83]
for Ar -semistable pairs.
Proposition 1.7. Let (C, Σ) be a Ar -semistable pair over an algebraically closed field k. Then
i) H1 (C, ωC (Σ)⊗m ) = 0 for every m > 2,
ii) ωC (Σ)⊗m is globally generated for every m > 2.
iii) ωC (Σ)⊗m is normally generated for m > 4.
Proof. Using duality, i) is equivalent to
⊗(1−m)
H0 (C, ωC (−mΣ)) = H0 (C, ωC (Σ)⊗(1−m) (−Σ)) = 0
for every m > 2. As before, we denote by aΓ the degree of ωC (Σ)|Γ where Γ is an irreducible
component of C. Thus,
deg ωC (Σ)⊗1−m (−Σ) 6 (1 − m)aΓ 6 0

for every m > 2. Thus the sections of the line bundle are zero restricted on every irreducible
component Γ such that aΓ > 0. Because the restriction of the line bundle to a semistable
component is non-canonically isomorphic to OP1 , which is globally generated, it easily follows
that all the sections must vanish.
Regarding ii), we need to prove that for every closed point p ∈ C, we have that
H1 (C, mp ωC (Σ)⊗m ) = 0
where mp is the ideal defining the point p. It is equivalent by duality to
⊗(1−m)
HomC (mp , ωC (−Σ)) = 0.
If p is singular on C, let us call π : C̃ → C the partial normalization in p of C. We have that
(see Lemma 2.1 of [Cat82])
Hom(mp , OC ) ⊂ π∗ OC̃
therefore
HomOC (mp , ωC (Σ)⊗(1−m) (−Σ)) ⊂ H0 (C̃, π ∗ ωC (Σ)⊗(1−m) (−Σ)).
Thus, it is enough to prove that the right hand side of the previous inclusion is trivial. Notice
that if C̃ is connected, the same argument as before gives us the vanishing of the cohomology
4
group (because π is birational). The same is true if C̃ is disconnected as long as none of the two
connected components is equal to the semistable component.
Suppose now that p is a separating point and p belongs to the semistable component. It is
easy to see that if ωC (Σ)⊗m is not globally generated in p, then it is not globally generated on
every point of the semistable component. Therefore we can reduce ourselves to prove that the
line bundle is globally generated on every smooth point, which is equivalent to the vanishing
H 0 (C, ωC (Σ)⊗m (−Σ + p)) = 0
where p is a smooth point of C. Because m > 2, it is easy to see that the only non-trivial case is
when p belongs to the semistable component.
Suppose p is a smooth point of a semistable component Γ of type (2) or (3) as in Remark 1.5.
We have that
ωC (Σ)⊗m (−Σ + p)|Γ ' ωΓ (D)⊗m (p)
where D is a lenght 2 divisor on Γ. Therefore the restriction of the line bundle to Γ is non-
canonically isomorphic to OP1 (p), which is very ample. Because the curve Γ intersects the rest of
the curve in a lenght 2 divisor, the result follows.
Moreover, suppose p is a smooth point of a semistable component Γ of type (1) as in Remark 1.5.
We have that
ωC (Σ)⊗m (−Σ + p)|Γ ' ωΓ (p + q)⊗m (p − q)
where q is the point where Γ intersects the rest of the curve. Therefore the restriction of the
line bundle to Γ is non-canonically isomorphic to OP1 which is globally generated. Because Γ
intersects the rest of the curve in a point, the result follows.
Finally, we prove that if L := ω(Σ) then L⊗m is normally generated for m > 4. This is a
simplified version of the proof in [Knu83] as the author proved the same statement for m > 3
which requires more work. For simplicity, we write H0 (−) and H1 (−) instead of H0 (C, −) and
H1 (C, −).
Consider the following commutative diagram for k > 1
H0 (L⊗2 ) ⊗ H0 (L⊗m−2 ) ⊗ H 0 (L⊗km ) H0 (L⊗m ) ⊗ H0 (L⊗km )

ϕ1 φ
ϕ2
H0 (L⊗m−2 ) ⊗ H0 (L⊗km+2 ) H0 (L⊗(k+1)m );
we need to prove that φ is surjective, therefore it is enough to prove that ϕ1 and ϕ2 are surjective.
As both L⊗2 and L⊗m−2 are globally generated because m > 4, we can use the Generalized
Lemma of Castelnuovo (see pag 170 of [Knu83]) and reduce to prove that H1 (L⊗(km−2) ) and
H1 (L⊗(km+4−m) ) are zero for k > 1. The vanishing results follow from the fact that km − 2 > 2
and km + 4 − m > 2 for every m > 4 and k > 1.
Remark 1.8. One can also prove that ωC (Σ)⊗m is normally generated for m > 3. In fact, the
same proof of Theorem 1.8 of [Knu83] can be perfectly adapted to our generality.
Corollary 1.9. Let (C, Σ) be a Ar -semistable pair over an algebraically closed field. If we
denote by C 0 the schematic image of the morphism φ4 : C → P(H0 (ωC (Σ)⊗4 )), then the morphism
c : C → C 0 induced by φ4 , is a contraction.
5
Michele Pernice
Proof. If ωC (Σ) is ample, i.e. there are no semistable components, then ωC (Σ)⊗4 is ample and
normally generated, therefore very ample. In this case, the morphism φ4 is a closed embedding
and c is an isomorphism. Suppose Γ is the semistable component. Proposition 1.7 assures us that
φ4 is globally defined and it is easy to see that it contracts Γ to a point. Because ωC (Σ)⊗4 |Γ is
non-canonically isomorphic to OP1 , we have that
c|CrΓ : C r Γ −→ C 0 r c(Γ)
is an isomorphism. We need to prove that C 0 is a Ar -prestable curve and that c−1 (c(Γ)) ' Γ. We
will prove the result only in the case (3) of Remark 1.5, as the other two are classical.
Let p be the tacnode on C where Γ intersects the rest of the curve and q be the image through
c, i.e. q := c(p). We have the morphism of local rings
φ4 : OP(H0 (ωC (Σ)⊗4 )),q −→ OC,p
and the image of this morphism is exactly the local ring OC 0 ,q . If we pass to the completions, we
get the morphism
φ4 : k[[x1 , . . . , xm ]] −→ k[[x, y]]/(y 2 − x4 ).
We identify the target of the morphism with the subalgebra of k[[t]] × k[[t]] where the two
polynomials coincide up to degree 1. Thus because ωC (Σ)⊗4 |Γ ' OP1 , we have that φ4 (xi )|Γ =
xi (p), i.e. the section xi at the point p. As c is birational, it is clear that the image of φ4
is isomorphic to the subalgebra of k[[t]] × k[[t]] with elements of the form (p(t), p(0)) where
p(t) ∈ k[[t]] such that p0 (0) = 0. Therefore the completion of OC 0 ,q is a A2 -singularity, i.e. the
complete local ring of a cusp. In fact, up to isomorphism we have that the morphism c is locally
of the form
k[[x, y]]/(y 2 − x3 ) ,→ k[[x, y]]/(y 2 − x4 )
2 2
x(y+x )
where (x, y) 7→ ( y+x
2 , 2 ). Finally, this implies c−1 (q) ' P1 .
Remark 1.10. Notice that this definition of contraction is not enough to have unicity. For example,
suppose we are in case (3) of Remark 1.5. Corollary 1.9 implies that we have a contraction
c : C −→ C 0
of a pair (C, Σ) with an almost-stable bridge that is contracted to a cusps. We can then find a
map from g : C 0 → C 00 where C 00 is still Ar -prestable, the map g is an isomorphism outside the
cusp of C 0 , the image of the cusp through g is a A4 -singularity and locally on the cusps is defined
as the morphism
g : k[[x, y]]/(y 2 − x5 ) −→ k[[x, y]]/(y 2 − x3 )
such that (x, y) 7→ (x, xy). The composition g ◦ c is still a contraction. The difference is that while
C 0 has the same arithmetic genus of C, the same is not true for C 00 .
The previous remark motivates the following definition.
Definition 1.11. A contraction c : C → C 0 of a Ar -semistable pair (C, Σ) is called minimal if
g(C) = g(C 0 ).
Remark 1.12. It is easy to see that a minimal contraction of a Ar -semistable pair sends rational
tails to smooth points, rational bridges to nodes and almost bridges to cusps. In particular, there
is only one minimal contraction up to isomorphism over an algebraically closed field and the
almost-bridge contraction can be described étale-locally as
k[x, y]/(y 2 − x3 ) −→ k[x, y]/(y 2 − x4 )
6
2 2
x(y+x )
defined by the association (x, y) 7→ ( y+x
2 , 2 ).
The following picture describes the three possible minimal contraction which can appear over
an algebraically closed field.
Figure 1. Contraction morphisms
The first one represents the minimal contraction of a rational bridge, the second one represents
the minimal contraciton of a rational tail and the third one represents the contraction of an
almost bridge.
Before going to analyze unicity of contractions, we introduce the same notion for families.
Definition 1.13. A Ar -prestable curve C of genus g over S is a proper, flat, finitely presented
scheme over S such that for every geometric point s ∈ S the fiber Cs is an Ar -prestable curve of
genus g.
A Ar -semistable pair (C/S, Σ) of genus g over a base scheme S is a Ar -prestable curve of genus
g over S together with an effective Cartier divisor Σ ⊂ C over S such that for every geometric
point s ∈ S the fiber (Cs , Σs ) is a Ar -semistable pair. In the same way, we can define Ar -stable
pairs and Ar -stable curves.
A contraction c : C → C 0 of a Ar -semistable pair (C, Σ) over S is a morphism or Ar -prestable
curves such that for every geometric point s ∈ S the fiber cs : Cs → Cs0 is a contraction. A
contraction over S is called minimal if it is minimal for every geometric fiber.
Remark 1.14. Notice that because we require flatness for C 0 , if a contraction over S is minimal
over a geometric point s ∈ S then it minimal over the whole family.
Remark 1.15. It is important to remark that an effective Cartier divisor Σ ⊂ C over S is a flat,
proper, finitely presented morphism Σ → S such that for every geometric point s ∈ S the fiber
Σs is a Cartier divisor of Cs . The flatness over S is essential, and in fact we can use Corollary
1.5 of [Knu83] to prove that the image of Σ through a contraction c : C → C 0 over S is still flat
over S and satisfy base change. In particular c(Σ) is a Cartier divisor of C 0 over S because of
Remark 1.6.
7
Michele Pernice
Moreover, because Σ restricted to every non-trivial fiber of c is at most of degree 1, this implies
that R1 c∗ O(−Σ) = 0 and in particular Rc∗ O(−Σ) ' O(−c(Σ)) as complexes of quasi-coherent
modules over C 0 .
We are finally ready to state the unicity result for the minimal contractions. Firstly, we have
to prove the Main Lemma from Knutsen’s work in our generality (see Lemma 1.6 of [Knu83]).
Lemma 1.16. Let c : C → C 0 be a minimal contraction of a Ar -semistable pair (C, Σ) over S and
denote by Σ0 the image c(Σ) ⊂ C 0 . Then for every k > 1
(a) there are canonical isomorphisms (commute with base change over S)
c∗ (ωC 0 (Σ0 )⊗k ) ' ωC (Σ)⊗k
and
ωC 0 (Σ0 )⊗k ' c∗ (ωC (Σ)⊗k );
(b) R1 c∗ (ωC (Σ)⊗k ) = 0;
(c) if we denote by π : C → S and π 0 : C 0 → S the two structural morphisms, then
Ri π∗ ωC (Σ)⊗k ' Ri π∗0 ωC 0 (Σ0 )⊗k for i = 0, 1.
Proof. Because ωC (Σ)⊗k restricted to the non-trivial fibers of c is non-canonically isomorphic to

OP1 , both (b) and (c) follows from (a) and Corollary 1.5 of [Knu83]. Notice also that if we have
the canonical isomorphism
c∗ (ωC 0 (Σ0 )⊗k ) ' ωC (Σ)⊗k
we can then pushforward it through c and precomposing it with the counit of the adjunction
c∗ a c∗ . However, if F is a locally free sheaf on C 0 , we have that
F −→ c∗ c∗ F ' F ⊗ c∗ OC
thanks to projection formula. We get that the counit of the adjunction is an isomorphism for
locally free sheaves because the geometric fiber of c are connected and reduced.
Finally, we have reduced to construct the canonical isomorphism
c∗ (ωC 0 (Σ0 )⊗k ) ' ωC (Σ)⊗k ,
and, because tensor and pullback commute, it is enough to construct it for k = 1.
We start by constructing a canonical morphism and then we prove it is an isomorphism. We
claim that
π∗ HomC (c∗ (ωC 0 (Σ0 )), ωC (Σ)) ' OS ;
if it is true, we can choose 1 ∈ H0 (S, OS ) to be our canonical morphism which will be denoted by
ϕc : c∗ (ωC 0 (Σ0 )) → ωC (Σ).
First of all, Grothendieck duality gives us an isomorphism
Rπ∗ R HomC (Lc∗ (ωC 0 (Σ0 ))(−Σ), ωC [1]) ' R HomS (Rπ∗ (Lc∗ (ωC 0 (Σ0 ))(−Σ)), OS ).
Let us study the two complexes. Notice that Lc∗ (ωC 0 (Σ0 )) is isomorphic to ωC 0 (Σ0 ) as a
complex concentrated in degree 0. Therefore the left hand side is a two-term complex concentrated
in [−1, 0] such that

H−1 Rπ∗ R HomC (Lc∗ (ωC 0 (Σ0 ))(−Σ), ωC [1] ' π∗ HomC (c∗ (ωC 0 (Σ0 )), ωC (Σ))
8
while

H0 Rπ∗ R HomC (Lc∗ (ωC 0 (Σ0 ))(−Σ), ωC [1] ' R1 π∗ HomC (c∗ (ωC 0 (Σ0 )), ωC (Σ)).
Regarding the right hand side, first of all we have that
Rc∗ (c∗ (ωC 0 (Σ0 ))(−Σ)) ' ωC 0 (Σ0 ) ⊗L Rc∗ O(−Σ) ' ωC 0
where the first isomorphism is the (derived) projection formula and the second isomorphism
follows from Remark 1.15. Because Rπ∗ ' Rπ∗0 ◦ Rc∗ , one easily get that the left hand side is also
a complex concentrated in [−1, 0] as expected and

H−1 R HomS (Rπ∗ (Lc∗ (ωC 0 (Σ0 ))(−Σ)), OS ) ' OS
while
H0 R HomS (Rπ∗ (Lc∗ (ωC 0 (Σ0 ))(−Σ)), OS ) ' π∗0 ωC 0 .
This proves the claim.
By construction, if c is an isomorphism in a point p ∈ C then ϕc is an isomorphism in p as
well. Clearly, it is enough to prove that ϕc is surjective and in particular we can reduce to prove
it restricting the morphism to the geometric fibers of C → S. This follows from Remark 1.12 and
an étale-local computation.
Proposition 1.17. Let (C, Σ) be a Ar -semistable pair over S. Then there exists a unique minimal
contraction up to unique isomorphism, i.e. if c1 : C → C1 and c2 : C → C2 are two minimal
contractions, then there exists a unique f : C1 → C2 which makes the following diagram
C
c1 c2
f
C1 C2
commute.
Proof. The unicity of f follows from the fact that a contraction is an epimorphism. Let us focus
on the existence part. In particular, we will prove that any minimal contraction is isomorphic to
the one defined in Corollary 1.9.
Consider a minimal contraction c : C → C 0 and recall that (C 0 , ωC 0 (Σ0 )) is a Ar -stable pair,
where Σ0 = c(Σ) (see Remark 1.6). Proposition 1.7 implies that
M
C 0 ' ProjS π∗0 (ωC 0 (Σ0 )⊗4n ) ,→ P(π∗0 ωC 0 (Σ0 )⊗4 )
n∈N
where π0
: C0
→ S is the structural morphism. Moreover, Lemma 1.16 gives us the following
isomorphism
π∗ ωC (Σ)⊗k ' π∗0 ωC 0 (Σ0 )⊗k
which clearly implies that C 0 coincides with the contraction defined in Corollary 1.9, i.e. the
image of the morphism
C −→ P(π∗ ωC (Σ)⊗4 ).
The image of such morphism is a Ar -prestable curve over S because it is flat over S thanks to
Proposition 1.7 and the geometric fibers have been described in Corollary 1.9.
We end up this section with a technical lemma. It helps us dealing with the case when we
have two Ar -semistable pairs that contracts to the same curve.
9
Michele Pernice
Lemma 1.18. Let (C1 , Σ1 ) and (C2 , Σ2 ) be two Ar -semistable pairs over an algebraically closed
field such that (C1 , Σ1 ) is not Ar -stable. Suppose we have the following diagram
g
C1 C2
c1 c2
C0
such that
– c1 , c2 are minimal contractions;
– if Γ1 is the semistable component of C1 , then g|Γ1 : Γ1 → C2 is a closed embedding.
Then g is an isomorphism.
Proof. First of all, notice that by Remark 1.5 we have that C1 and C2 have the same type of
semistable components (Γ1 and Γ2 respectively) and clearly g(Γ1 ) ⊂ Γ2 . Because g|Γ1 is a closed
embedding, then it is clear that g|Γ1 : Γ1 → Γ2 is an isomorphism. In the same way, we have that
g restricted to C1 r Γ1 is an isomorphism with C2 r Γ2 . The statement follows easily from a local
computation.
2. Moduli stack of Ar -stable curves

Let g be an integer with g > 2, r be a nonnegative integer and κ be a base field with characteristic
greater than r + 1.
We denote by M fr the category defined in the following way: an object is a proper flat finitely
g
presented morphism C → S over κ such that every geometric fiber over S is a Ar -stable curve of
genus g. These families are called Ar -stable curves over S. Morphisms are defined in the usual
way. This is clearly a fibered category over the category of schemes over κ.
Fix a positive integer n. In the same way, we can define M fr the fibered category whose
g,n
objects are the datum of Ar -stable curves over S with n distinct sections p1 , . . . , pn such that
every geometric fiber over S is a n-pointed Ar -stable curve. These families are called n-pointed
Ar -stable curves over S. Morphisms are just morphisms of n-pointed curves.
The main resultx of this section are the description of M frg,n as a quotient stack and the
description of the contraction morphism. Firstly, we need to prove two results which are classical
in the case of (A1 -)stable curves.
Proposition 2.1. Let C → S a proper flat finitely presented morphism with n-sections si : S → C
for i = 1, . . . , n. There exists an open subscheme S 0 ⊆ S with the property that a morphism
T → S factors through S 0 if and only if the projection T ×S C → T , with the sections induced by
the si , is a n-pointed Ar -stable curve.
Proof. It is well known that a small deformation of a curve with Ah -singularities for h 6 r still
has Ah -singularities for h 6 r. Hence, after restricting to an open subscheme of S we can assume
that C → S is an Ar -prestable curve. By further restricting S we can assume that the sections
land in the smooth locus of C → S, and are disjoint. Then the result follows from openness of
ampleness for invertible sheaves.
We are ready to prove the theorem. To be precise, both the proofs of the following theorem
and of Proposition 2.1 are an adaptation of Theorem 1.3 and of Proposition 1.2 of [DLPV21] to
the more general case of Ar -stable curves.
10
Theorem 2.2. M fr is a smooth connected algebraic stack of finite type over κ. Furthermore, it
g,n
is a quotient stack: that is, there exists a smooth quasi-projective scheme X with an action of
GLN for some positive N , such that M frg,n ' [X/GLN ].
Proof. It follows from Proposition 1.7 that if (π : C → S, Σ1 , . . . , Σn ) is a n-pointed Ar -stable

curve of genus g, then π∗ ωC/S (Σ1 + · · · + Σn )⊗3 is locally free sheaf of rank N := 5g − 5 + 3n,
and its formation commutes with base change, because of Grothendieck’s base change theorem.
Call X the stack over κ, whose sections over a scheme S consist of a n-pointed Ar -stable
curve as above, and an isomorphism OSN ' π∗ ωC/S (Σ1 + · · · + Σn )⊗m of sheaves of OS -modules.
Since π∗ ωC/S (Σ1 + · · · + Σn )⊗m is very ample , the automorphism group of an object of X is
trivial, and X is equivalent to its functor of isomorphism classes. We proved the line bundle is
very ample for m > 4, but it is still true for m = 3. Clearly, the same prove also gives us the
result for n = 4, except that N increases.
Call H the Hilbert scheme of subschemes of PN −1 with Hilbert polynomial P (t), and D → H
κ
the universal family. Call F the fiber product of n copies of D over S, and C → F the pullback
of D → H to F ; there are n tautological sections s1 , . . . , sn : F → C. Consider the largest open
subscheme F 0 of F such that the restriction C 0 of C, with the restrictions of the n tautological
sections, is a n-pointed Ar -stable curve, as in Proposition 2.1. Call Y ⊆ F 0 the open subscheme
whose points are those for which the corresponding curve is nondegenerate, E → Y the restriction
of the universal family, Σ1 , . . . , Σn ⊆ E the tautological sections. Call OE (1) the restriction
−1
of OPN −1 (1) via the tautological embedding E ⊆ PN Y ; there are two section of the projection
Y
3(2g−2+n)
PicE/Y → Y from the Picard scheme parametrizing invertible sheaves of degree 3(2g − 2 + n),
one defined by OE (1), the other by ωE/Y (Σ1 + · · · + Σn )⊗3 ; let Z ⊆ Y the equalizer of these two
sections, which is a locally closed subscheme of Y .
Then Z is a quasi-projective scheme over κ representing the functor sending a scheme S into
the isomorphism class of tuples consisting of a n-pointed Ar -stable curve π : C → S, together
with an isomorphism of S-schemes
−1
PN
S ' P(π∗ ωC/S (Σ1 + · · · + Σn )⊗3 ) .
There is an obvious functor X → Z, associating with an isomorphism OSN ' π∗ ωC/S (Σ1 + · · · +
Σn )⊗3 its projectivization. It is immediate to check that X → Z is a Gm -torsor; hence it is
representable and affine, and X is a quasi-projective scheme over Spec κ.
On the other hand, there is an obvious morphism X → M fr which forgets the isomorphism
g,n
N ⊗3
OS ' π∗ ωC/S (Σ1 + · · · + Σn ) ; this is immediately seen to be a GLN torsor. We deduce that
Mfr is isomorphic to [X/GLN ]. This shows that is a quotient stack, as in the last statement;
g,n
frg,n is an algebraic stack of finite type over κ.
this implies that M
The fact that Mfrg,n is smooth follows from the fact that deformations of Ar -prestable curves
are unobstructed.
Finally, to check that M frg,n is connected it is enough to check that the open embedding
Mg,n ⊆ M fr has a dense image, since Mg,n is well known to be connected. This is equivalent
g,n
to saying that every n-pointed Ar -stable curve over an algebraically closed extension Ω of κ
has a small deformation that is smooth. Let (C, p1 , . . . , pn ) be a n-pointed Ar -stable curve; the
singularities of C are unobstructed, so we can choose a lifting C → Spec Ω[[t]], with smooth
generic fiber. The points pi lift to sections Spec Ω[[t]] → C, and then the result follows from
Proposition 2.1.
11
Michele Pernice
fr ⊂ M
Remark 2.3. Clearly, we have an open embedding M fs for every r 6 s. Notice that
g,n g,n
r 2g+1
Mg,n = Mg,n for every r > 2g + 1.
f f
We prove that the usual definition of Hodge bundle extends to our setting. As a consequence
fr , which is called Hodge bundle.
we obtain a locally free sheaf Hg of rank g on M g,n
Proposition 2.4. Let π : C → S be a n-pointed Ar -stable of genus g. Then π∗ ωC/S is a locally

free sheaf of rank g on S, and its formation commutes with base change.
Proof. If C is an Ar -stable curve of genus g over a field k, the dimension of H0 (C, ωC/k ) is g; so
the result follows from Grauert’s theorem when S is reduced. But the versal deformation space of
an Ar -stable curve over a field is smooth, so every Ar -stable curve comes, étale-locally, from an
Ar -stable curve over a reduced scheme, and this proves the result.
Finally, we describe the contraction morphism. Luckily, most of the work is done in Section 1.
Theorem 2.5. We have a morphism of algebraic stacks
frg,n+1 −→ Ceg,n
γ:M r
r is the universal curve of M

where Ceg,n frg,n . Furthermore, it is an open immersion and its image is
r,62 r parametrizing n-pointed A -stable curves (C, p , . . . , p ) of genus g
the open locus Ceg,n in Ceg,n r 1 n
and a (non-necessarily smooth) section q such that q is an Ah -singularity for h 6 2.
Proof. Let (C, p1 , . . . , pn+1 ) be a n + 1-pointed Ar -stable curve over S. If we denote by Σ the
Cartier divisor O(p1 + · · · + pn ) on C, it is clear that (C, Σ) is a Ar -semistable pair over S.
Therefore we can construct the (minimal) contraction functor γ thanks to Proposition 1.17 and it
r,62
factors through the open Ceg,n thanks to Corollary 1.9.
We can construct an inverse to this map, namely a functor
r,62 frg,n+1
η : Ceg,n −→ M
r,62
which is a quasi-inverse of γ. For the sake of notation, we denote by C the stack Ceg,n . Consider the
universal family of D → C of C with universal sections (q1 , . . . , qn , qn+1 ). Furthermore, consider
the closed substack ∆ of C parametrizing (C̃, q1 , . . . , qn , qn+1 ) such that qn+1 lands in the singular
locus of the morphism C̃ → S. We can define its stacky structure using the 1-st Fitting ideal of
ΩC̃/S .
We denote by Ei the closed substack qn+1 ∩ qi of D for i = 1, . . . , n and by E0 the closed
substack qn+1 (∆). We have that the Ei ’s are disjoint closed substacks of D of codimension 2 and
we denote by E the union. Finally, we denote by B the blowup of D with center E and by pi
the proper transform of qi for every i = 1, . . . , n + 1. We need to prove that (B, p1 , . . . , pn+1 ) is a
n + 1-pointed Ar -stable curve over C. In fact, we are going to prove that the blowup morphism
B → D is a (minimal) contraction for the semistable pair (B, Σ) where Σ ' O(p1 + · · · + pn ).
We have that B is a proper, finitely presented morphism over C. To prove that B is flat over
C, we notice that we have a factorization of the inclusion of the center of the blowup inside D
qn+1
E C D
where E is a Cartier divisor inside C. If we denoteby IE theideal defining E in D, that we have
that IE⊗m is flat over C and therefore B := ProjD
L ⊗m
m IE is also flat over C.
12
We need to prove that the geometric fibers of B → C are Ar -prestable curves. The stability
condition then follows easily. For simplicity, we will prove the case n = 0 but the same proof
works for any n. Moreover, we will explicitly do the computation in a neighborhood of a geometric
object in C of the form (C̃, q) where q is a cusp because it is the only non-classical case. The same
idea can be applied to the other cases.
As usual, deformation theory tells us that we can consider a smooth neighborhood (Spec A, mq )
e q) in C with A a smooth algebra. This implies that we have an Ar -stable curve
of (C,
qA
C
eA Spec A
with qA a global section such that qA ⊗A A/mq = q. By deformation theory of cusps (see Example
6.2.12 of [TV10]), we know that the completion of the local ring of q in C
eA is of the form
A[[x, y]]/(y 2 − x3 − r1 x − r2 )
where r1 , r2 are part of a system of parameters for A because (Spec A, mq ) is versal. Consider
know the element f := 4r13 + 27r22 ∈ A, which parametrizes the locus when the section qA ends in
a singular point, which is either a node or a cusp. Let q1 : Spec A/f ,→ C eA be the codimension
two closed immersion and denote by CA := Blq1 C. We want to prove that the geometric fiber of
e
CA over mq is an almost-bridge.
Notice that the formation of the blowup does not commute with arbitrary base change.
However, consider the system of parameters (r1 , r2 , . . . , ra ) with a := dim A = 3g − 2. Then if we
denote by J := (r2 , . . . , ra ) we have that
TorA n
1 (A/f , A/J) = 0
for every n. This implies that the blowup commutes with the base change for A/J and therefore
we can reduce to the case when A is a DVR, r := r1 is the uniformizer and the completion of the
local ring of q in C
eA is of the form
A[[x, y]]/(y 2 − x3 − rx).

A standard blowup computation proves that over the special fiber of A we have
– CA ⊗A A/mq is an Ar -prestable curve and if we denote by pA the proper transform of qA ,
we have that (CA ⊗A A/mq , pA ⊗A A/mq ) is an 1-pointed Ar -stable curve of genus g,
– CA → CeA ⊗A A/mq is the contraction of an almost-bridge.
We have finally proved the existence of the two functors γ and η. Proposition 1.17 assures us
that there exists a unique isomorphism of functors γ ◦ η ' id. Regarding the other composition,
the universal property of the blowup gives us that it exists a natural transformation id ⇒ η ◦ γ
and it is enough to prove that it is an isomorphism for every geometric point of M fr
g,n+1 . This
follows from Lemma 1.18
3. Preliminary results
In this section, we prove several preliminary results that are used for the description of the locus
parametrizing curves with Ah -singularities.
Firstly, we introduce the moduli stack parametrizing finite flat extensions of pointed finite flat
curvilinear algebras and describe it as an affine bundle over a classifying stack.
13
Michele Pernice
Secondly, we discuss the behaviour of pushouts and blowups in families and we prove that
the two constructions are one the inverse of the other under some strict conditions, which are
satisfied in our situation.
Most of them are probably well-known to experts, but for lack of a better reference in the
literature, we decided to write down the proofs.
3.1 Extensions of finite curvilinear algebras

We describe the moduli stack of finite flat extensions of curvilinear algebras.
Let m be a positive integer and κ be the base field. We denote by F Hm the finite Hilbert
stack of a point, i.e. the stack whose objects are pairs (A, S) where S is a scheme and A is a
locally free finite OS -algebra of degree m. We know that F Hm is an algebraic stack, which is in
fact a quotient stack of an affine scheme of finite type by a smooth algebraic group. For a more
detailed treatment see Section 96.13 of [Sta22].
Given another positive integer d, we can consider a generalization of the stack F Hm . Given
a morphism X → Y, one can consider the finite Hilbert stack F Hd (X /Y) which parametrizes
commutative diagrams of the form
Z X
f
S Y
where f is finite locally free of degree d. This again can be prove to be algebraic. If Y = Spec κ,
we denote it simply by F Hd (X ). For a more detailed treatment see Section 96.12 of [Sta22].
Finally, we define EHm,d to be the fibered category in groupoids whose objects over a scheme
S are finite locally free extensions of OS -algebras A ,→ B of degree d such that A is a finite locally
free OS -algebra of degree m. Morphisms are defined in the obvious way. Clearly the algebra B is
finite locally free of degree dm.
Proposition 3.1. The stack F Hd (F Hm ) is naturally isomorphic to EHm,d , therefore EHm,d is
an algebraic stack.
Proof. The proof follows from unwinding the definitions.
We want to add the datum of a section of the structural morphism OS ,→ B. This can be
done passing to the universal object of F Hdm .
Let n be a positive integer and let Buniv the universal object of F Hn ; consider Fn :=
SpecF Hn (Buniv ) the generalized spectrum of the universal algebra over F Hn . It parametrizes
pairs (B, q) over S where B ∈ F Hn (S) and q : B → OS is a section of the structural morphism
OS ,→ B.
Definition 3.2. We say that a pointed algebra (A, p) ∈ Fn over an algebraically closed field k
is linear if dimk mp /m2p 6 1, where mp is the maximal ideal associated to the section p. We say
that (A, p) is curvilinear if (A, p) is linear and Spec A = {p}.
We want to study the moduli stack parametrizing curvilinear pointed algebras of lenght n.
This is locally closed inside Fn and we endow it with a stacky structure in the following way.
We consider the closed substack defined by the 1-st Fitting ideal of ΩA|OS in Fn . This locus
parametrizes non-linear pointed algebras . Therefore we can just consider the open complement,
which we denote by Fnlin . We can inductively define closed substacks of Fnlin in the following way:
14
suppose Sh is defined, then we can consider Sh+1 to be the closed substack of Sh defined by the
0-th fitting ideal of ΩSh |F Hn . We set S1 = Fnlin . It is easy to prove that the geometric points of
Sh are pairs (A, p) such that A localized at p has length greater or equal than h. Notice that this
construction stabilizes at h = n and the geometric points of Sh are exactly the curvilinear pointed
algebras. Finally, we denote by Fnc := Sn the last stratum. As it is a locally closed substack of an
algebraic stacks of finite type, it is algebraic of finite type too.
Lemma 3.3. If (B, q) ∈ Fnc (S) for some κ-scheme S, then it exists an étale cover S 0 → S such
that B ⊗S S 0 ' OS 0 [t]/(tn ) and q ⊗S S 0 = q0 ⊗ S 0 where
q0 : κ[t]/(tn ) −→ κ
is defined by the assocation t 7→ 0.
Proof. We are going to prove that Fnc has only one geometric point (up to isomorphism) and its
tangent space is trivial. Thus the thesis follows, by a standard argument in deformation theory.
Suppose then S = Spec k is the spectrum of an algebraically closed field and B is a finite
k-algebra (of degree n). Because (B, q) is linear, we have that dimk (mq /m2q ) 6 1. We can construct
then a surjective morphism
k[[t]] −→ Bmq
0 0
whose kernel is generated by the monomial tn . We have then (B, q) ' (k[t]/(tn ), q0 ) for some
positive integer n0 . Because B is local of length n, we get that n0 = n.
Let us study the tangent space of Fnc at the pointed algebra (k[t]/(tn ), q0 ) where k/κ is a field.
We have that any deformation (Bε , qε ) of the pair (k[t]/(tn ), q0 ) is of the form
k[t, ε]/(p(t, ε))
where p(t, 0) = tn . Because the section is defined by the association t 7→ bε, we have also that
p(bε, ε) = p(0, ε) = 0. It is easy to see that (Bε , qε ) ∈ Fnc only if
p(bε, ε) = p0 (bε, ε) = · · · = p(n−1) (bε, ε) = 0
where the derivatives are done over k[ε]/(ε2 ), thus p(t, ε) = (t − bε)n . The algebra obtained is
clearly isomorphic to trivial one.
Remark 3.4. Notice that we do not need the field k to be algebraically closed to prove that
there is only one pointed curvilinear algebra of lenght n over k. This is related to the fact that
the automorphism group of the unique point is special and therefore the previous statement is
true even Zariski-locally.

Let Gn := Aut κ[t]/(tn ), q0 be the automorphism group of the standard pointed curvilinear
algebra. One can describe G as the semi-direct product of Gm and a group U which is isomorphic
to an affine space of dimension n − 2.
Corollary 3.5. We have that Fnc is isomorphic to BGn , the classifying stack of the group Gn .
c
We denote by Em,d c . We get the morphism of algebraic
the fiber product EHm,d ×F Hdm Fdm
stacks
c c
Em,d −→ Fdm
defined by the association (S, A ,→ B → OS ) 7→ (S, B → OS ). Notice that the morphism is
faithful, therefore representable by algebraic spaces. Clearly the composite A → OS is still a
section.
15
Michele Pernice
c .
We now study the stack Em,d
c (S) for some scheme k-scheme S, then it exists an étale cover
Lemma 3.6. If (A ,→ B, q) ∈ Em,d
0
π : S → S such that

π ∗ A ,→ B, q ' φd : OS 0 [t]/(tm ) ,→ OS 0 [t]/(tdm ), q0 ⊗S S 0
where φd (t) = td p(t) with p(0) ∈ OS×0 .
Proof. First of all, an easy computation shows that any finite flat extension of pointed algebras
of degree d
OS [t]/(tm ) ,→ OS [t]/(tdm )
is of the form φd for any scheme S. Therefore, it is enough to prove that if A ,→ B is finite flat of
degree d, then B curvilinear implies A curvilinear. In analogy with the proof of Lemma 3.3, we
prove the statement for S = Spec k and then for S = Spec k[ε]/(ε2 ) with k algebraically closed
field.
Firstly, suppose S = Spec k with k algebraically closed. By Lemma 3.3, we know that
(B, q) ' (k[t]/(tdm ), q0 ). We need to prove now that also A is curvilinear. Clearly A is local
because of the going up property of flatness and we denote by mA its maximal ideal.
If we tensor A ,→ B by A/mA , we get k ,→ k[t]/mA k[t]. However flatness implies that
the extension k ,→ k[t]/(mA k[t]) has degree d, therefore because k[t] is a PID, it is clear that
mA k[t] ⊂ mdq and the morphism mA /m2A → mdq /md+1 q is surjective. Let a ∈ mA be an element
whose image is td modulo td+1 . If we now consider A/a ,→ B/aB ' k[t]/(td ), by flatness it still
has length d, but this implies A/(a) ' k or equivalently mA = (a). Therefore A is curvilinear too.
Suppose now S = Spec k[ε]. We know that, given a morphism of schemes X → Y , we have
the exact sequence
Def X→Y −→ Def X ⊕ Def Y −→ Ext1OX (Lf ∗ NLY , OX )
where NLY is the naive cotangent complex of Y . We want to prove that if X → Y is the spectrum
of the extension A ,→ B, then the morphism
Def X −→ Ext1OX (Lf ∗ NLY , OX )
is injective. This implies the thesis because Y is the spectrum of a pointed curvilinear algebra,
therefore Def Y = 0.
We can describe the morphism
Def X −→ Ext1OX (Lf ∗ NLY , OX )
explicitly using the Schlessinger’s functors T i . More precisely, it can be described as a morphism
T 1 (A/k, A) → T 1 (A/k, B);
an easy computation shows that T 1 (A/k, A) ' k[t]/(tm−1 ) and T 1 (A/k, B) ' k[t]/(td(m−1) ).
Through these identifications, the morphism
T 1 (A/k, A) → T 1 (A/k, B)
is exactly the morphism φd defined earlier, i.e. it is defined by the association t 7→ td p(t) with
p(0) 6= 0. The injectivity is then straightforward.
16
Now we are ready to describe the morphism Em,d → Fmd . Let us define A0 := κ[t]/(tm ) and
B0 := κ[t]/(tmd ). Let Em,d be the category fibered in groupoids whose objects are

S, (A ,→ B, q), φA : (A, q) ' (A0 ⊗ S, q0 ⊗ S), φB : (B, q) ' (B0 ⊗ S, q0 ⊗ S)
where (A ,→ B, q) ∈ Em,d (S). The morphism are defined in the obvious way. It is easy to see
that Em,d is in fact fibered in sets. As before, we set Gm := Aut(A0 , q0 ) and Gdm := Aut(B0 , q0 ).
Clearly we have a right action of Gdm and a left action of Gm on Em,d .
Proposition 3.7. We have the follow isomorphism of fibered categories

Em,d ' [Em,d /Gm × Gmd ]
and through this identification the morphism Em,d → Fdm is just the projection to the classifying
space BGmd .
Proof. It follows from Lemma 3.6.
Lemma 3.6 also tells us how to describe Em,d : the map φd is completely determined by
p(t) ∈ OS [t]/(td(m−1) ) with p(0) ∈ OS× . Therefore, we have a morphism
(A1 r 0) × Ad(m−1)−1 −→ Em,d
which is easy to see that it is an isomorphism. Consider now the subscheme of Em,d defined as
V := {f ∈ (A1 r 0) × Ad(m−1)−1 | a0 = 1, akd (f ) = 0 for k = 1, . . . , m − 2}
where al (f ) is the coefficient of tl of the polynomial f (t). Clearly V is an affine space of dimension
(m − 1)(d − 1).
Lemma 3.8. In the situation above, we have the isomorphism

V ' [Em,d /Gm ],
in particular the morphism Em,d → Fmd is an affine bundle of dimension (m − 1)(d − 1).
Proof. Consider the morphism

Gm × V −→ Em,d
which is just the restriction of the action of Gm to V . A straightforward computation shows that
it is an isomorphism. The statement follows.
3.2 Pushout and blowups in families

In this subsection, we discussion two well-known constructions: pushout and blowup. Specifically,
we study these two constructions in families and give conditions to get flatness and compatibility
with base change. Moreover, we study when the two constructions are one the inverse of the other.
Lemma 3.9. Let S be a scheme. Consider three schemes X,Y and Y 0 which are flat over S.
Suppose we F are given Y ,→ X a closed immersion and Y → Y 0 a finite flat morphism. Then the
0
pushout Y Y X exists in the category of schemes, it is flat over S and it commutes with base
change.
Furthermore, if X and Y 0 are proper and finitely presented scheme over S, the same is true
0
F
for Y Y X.
17
Michele Pernice
Proof. The existence of the pushout follows from F Proposition 37.65.3 of [Sta22]. In fact, F the
0 0
proposition tells us that the morphism Y → Y Y X is a closed immersion and X → Y Y X 0
is integral, in particular affine. It is easy to prove thatF it is in fact finite and surjective because
0 0
Y → Y is finite and surjective. Let us prove that Y Y X → S is flat. Because flatness is local
condition and all morphisms are affine, we can reduce to a statement of commutative algebra.
Suppose we are given R a commutative ring and A,B two flat algebras. Let I be an ideal of
A such that A/I is R-flat and B ,→ A/I be a finite flat extension. Then B ×A/I A is R-flat. To
prove this, we complete the fiber square with the quotients
0 B ×A/I A A Q0 0
0 B A/I Q 0
and we notice that Q0 → Q is an isomorphism. Because the extension B ,→ A/I is flat, then Q
(thus Q0 ) is R-flat and the R-flatness of A and Q0 implies the flatness of B ×A/I A.
Suppose now we have a morphism T → S. This induces a natural morphism
G G
φT : YT0 XT → (Y 0 X)T
YT Y
where by (−)T we denote the base change (−) ×S T . Because being an isomorphism is a local
condition, we can reduce to the following commutative algebra’s statement. Suppose we have a
morphism R → R; e we can consider the same morphism of exact sequences of R-modules as above
and tensor it with R.
e We denote by (−)g the tensor product (−) ⊗R R.e The flatness of Q implies
that the commutative diagram
0 ×A/I A
B^ A
e f0
Q 0
0 B
e A/I
g Q
e 0
is still a morphism of exact sequences. Because also B e ×g A e→Q

e is the kernel of A f0 , we get
A/I
that φT is an isomorphism.
Finally, using the fact that the pushout is compatible with base change, we reduce to the case
S = Spec R0 with R0 of finite type over κ. Thus we can use Proposition 37.65.5 of [Sta22] to
prove that the pushout (in the situation when Y 0 → Y is flat) preserves the property of being
proper and finitely presented.
The second construction we want to analyze is the blow-up.
Lemma 3.10. Let X/S be a flat, proper and finitely presented morphism of schemes and let I be
an ideal of X such that OX /I m is flat over S for every m > 0. Denote by BlI X → X the blowup
morphism, then BlI X → S is flat, proper and finitely presented and its formations commutes
with base change.
Proof. The flatness of OX /I m implies the flatness of I m , therefore ⊕m>0 I m is clearly flat over
S. The universal property of the blowup implies that it is enough to check that the formation
of the blowup commutes with base change when we restrict to the fiber of a closed point s ∈ S.
Therefore it is enough to prove that the inclusion ms I m ⊂ ms ∩ I m is an equality for every m > 0
18
with ms the ideal associated to the closed point s. The lack of surjectivity of that inclusion
is encoded in Tor1S (OX /I m , OS /mS ) which is trivial due to the S-flatness of OX /I m . The rest
follows from classical results.
Remark 3.11. Notice that the flatness of the blowup follows from the flatness of I m for every m
whereas we need the flatness of OX /I m to have the compatibility with base change.
Lemma 3.12. Let R be a ring and A ,→ B be an extension of R-algebras. Suppose I ⊂ A is an
ideal of A such that I = IB. Then the following cocartesian diagram
A B
A/I B/I
is a cartesian diagram of R-algebras. Furthermore, suppose we have a cartesian diagram of
R-algebras
e ×B/I B
A := A B
A
e B/I
then the morphism A → A e is surjective, its kernel coincides (as an R-module) with the ideal I
and the diagram is cocartesian.
Proof. It follows from a straightforward computation in commutative algebra.
Finally, we prove that the two constructions are one the inverse of the other.
Proposition 3.13. Suppose we are given a diagram
D
e X
e
D
e → D is a finite flat morphism and
of proper, flat, finitely presented schemes over S such that D
D ,→ X is a closed immersion of an effective Cartier divisor. Consider the pushout X := X
e e e qe D
D
as in Lemma 3.9 and denote by ID the ideal associated with the closed immersion D ,→ X. Then
the pair (X, ID ) over S verifies the hypothesis of Lemma 3.10. Furthermore, if we denote by
(X, D) the blowup of the pair (X, D), there exists a unique isomorphism (X, e D)e ' (X, D) of
pairs over (X, D).
Proof. Consider the pushout diagram over S
D
e X
e
D X;
because every morphism is finite and flatness is a local condition, we can restrict ourself to the
n = I n for every n > 1.
affine case and Lemma 3.12 assures us that ID = IDe , and in particular ID
D
e
Because IDe is a flat Cartier divisor, the same is true for its powers.
19
Michele Pernice
Regarding the second part of the statement, we know that the unicity and existence of the
morphism
(X, D) −→ (X,
e D)
e
are assured by the universal property of the blowup. As being an isomorphism is a local property,
we can reduce again to the affine case (all the morphisms involved are finite). The statement
follows from the following remark.
Remark 3.14. Let A ,→ B be an extension of algebras and I be an ideal of A such that I = IB
and I is free of rank 1 as a B-module. Therefore we can describe the Rees algebra as follows
M
RA (I) := I n = A ⊕ tB[t]
n>0
because I is free of rank 1 over B. It is immediate to see that the morphism Spec B → ProjA (RA (I))
is an isomorphism over Spec A.
Corollary 3.15. Let b : X → Y be a finite birational morphism and let Jb be the conductor
ideal. Suppose that Jb is a line bundle as a OX -module. If we denote by J ,→ Y the closed
subscheme associated to Jb , then there exists a canonical isomorphism
X ' BlJ Y
over Y .
Proof. It follows from the previous remark.
Proposition 3.16. Let D ,→ X be a closed immersion of proper, flat, finitely presented schemes
n is S-flat for every n > 1, consider the blowup b : X
over S such that the ideal ID e := BlD X → X
and denote by De the proper transform of D. Suppose that D e → D is finite flat (in particular
−1
the morphism b is finite birational) and that b ID = IDe . Then it exists a unique isomorphism
Xe q e D → X, which makes everything commutes.
D
Proof. As in the previous proposition, the existence and unicity of the morphism are conseguences
of the universal property of the pushout. Therefore as all morphisms are finite we can restrict to
the affine case. We can therefore use Lemma 3.12 and conclude.
4. The substack of Ar -singularity

fr which parametrizes Ar -stable curves with at
In this section we describe the closed substack of M g
least a singularity of type Ah with h > 2. We do so by stratifying this closed substack considering
singularities of type Ah with a fixed h greater or equal than 2. We want to remark that all
the following proofs can be upgraded to the pointed case with close to zero effort. However, we
decided to treat the non-pointed case for the sake of exposition. From now on, n is a positive
integer unrelated with the number of sections.
Let g > 2 and r > 1 be two integers and κ be the base field of characteristic either 0 or greater
than r. We recall the sequence of open subset (see Remark 2.3)
f0 ⊂ M
M f1 ⊂ · · · ⊂ M
fr
g g g
fr r M
and we define Ae>n := M f−1 := ∅.
fn−1 for n = 0, . . . , r + 1 setting M
g g g
We now introduce an alternative to A>n which is easier to describe. Suppose n is a positive
e
integer smaller or equal than r and let A>n be the substack of the universal curve Cegr of Mfr
g
20
parametrizing pairs (C/S, p) where p is a section whose geometric fibers over S are Ar -singularities
for r > n. We give to A>n the structure of closed substack of Cegr inductively on n. Clearly if
n = 0 we have A>0 = Cegr . To define A>1 , we need to find the stack-theoretic structure of the
singular locus of the natural morphism Cegr → M frg . This is standard and it can be done by taking
fr
fr . We have that A>1 → Mg is now finite and it is
the zero locus of the 1-st Fitting ideal of ΩCer |M
g g
unramified over the nodes, while it ramifies over the more complicated singularities. Therefore,
we can denote by A>2 the substack of A>1 defined by the 0-th Fitting ideal of ΩA |M fr . A
>1 g
local computation shows us that A>2 → M fr is unramified over the locus of A2 -singularities and
g
ramified elsewhere. Inductively, we can iterate this procedure considering the 0-th Fitting ideal of
ΩA
>n−1
fr to define A>n .
|Mg
A local computation shows that the geometric points of A>n are exactly the pairs (C, p) such
that p is an An0 -singularity for n 6 n0 6 r.
Let us define An := A>n r A>n+1 for n = 0, . . . , r − 1. We have a stratification of A>2
Ar = A>r ⊂ A>r−1 ⊂ · · · ⊂ A>2
where the An ’s are the associated locally closed strata for n = 2, . . . , r.
The first reason we choose to work with A>n instead of Ae>n is the smoothness of the locally
closed substack An of Cegr .
Proposition 4.1. The stack An is smooth.
Proof. We can adapt the proof of Proposition 1.6 of [DLPV21] perfectly. The only thing to
point out is that the étale model induced by the deformation theory of the pair (C, p) would
be y 2 = xn + an−2 xn−2 + · · · + a1 x + a0 , thus the restriction to An is described by the equation
an−2 = · · · = a1 = a0 . The smoothness of M frg implies the statement.
Before going into details for the odd and even case, we describe a way of desingularize a
An -singularity.
Lemma 4.2. Let (C, p) ∈ An (S), then the ideal Ip associated to the section p verifies the hypothesis
of Lemma 3.10. If we denote by b : C e → C the blowup morphism and by D the preimage b−1 (p),
then D is finite flat of degree 2 over S.
– If n = 1, D is a Cartier divisor of C
e étale of degree 2 over S.
– If n > 2, the 0-th Fitting ideal of ΩD|S define a section q of D ⊂ C
e → S such that C
e is an
Ar -prestable curve and q is an An−2 -singularity of C.
e
Proof. Suppose that the statement is true when S is reduced. Because An is smooth, we know
that up to an étale cover of S, our object (C, p) is the pullback of an object over a smooth scheme,
therefore reduced. All the properties in Lemma 4.2 are stable by base change and satisfies étale
descent, therefore we have the statement for any S.
Assume that S is reduced. To prove the first part of the statement, it is enough to prove
that the geometric fiber OXs /Ism has constant length over every point s ∈ S. This follows from a
computation with the complete ring in ps . Regarding the rest of the statement, we can restrict to
the geometric fibers over S and reduce to the case S = Spec k where k is an algebraically closed
field over κ. The statement follows from a standard blowup computation.
21
Michele Pernice
Let us start with (C, p) ∈ An (S). We can construct a (finite) series of blowups which
desingularize the family C in the section p.
Suppose n is even. If we apply Lemma 4.2 iteratively, we get the successive sequence of
blowups over S
bm bm−1 b1
C
em C
em−1 ... C
e0 := C
with sections qh : S → Ceh where m := n/2, the morphism bh : C em := Blq C

h−1 h−1
e →C eh−1 is the
blowup of Ceh−1 with center qh−1 (q0 := p) and qh is the section of C
eh over qh−1 as in Lemma 4.2.
We have that Cm is an Ar -prestable curve of genus g − m and qm is a smooth section.
e
On the contrary, if n := 2m − 1 is odd, the same sequence of blowups gives us a curve C em
which has arithmetic genus either g − m or g − m + 1 depending on whether the geometric fibers
of C
em are connected or not and an étale Cartier divisor D of degree 2 over S.
Definition 4.3. Let (C, p) be an object of An (S) with n = 2m or n = 2m − 1 for S any

scheme. The composition of blowups bm ◦ bm−1 ◦ · · · ◦ b1 described above is called the relative
An -desingularization and it is denoted as bC,p . By abuse of notation, we refer to the source of the
relative An -desingularization as relative An -desingularization. We also denote by Jb the conductor
ideal associated to it.
We say that an object (C/S, p) of An is a separating An -singularity if the geometric fibers
over S of the relative An -desingularization are not connected.
Remark 4.4. By construction, the relative An -desingularization is compatible with base change.
In fact, it can be described as a single blowup of C in the closed subscheme defined by the
conductor ideal of the desingularization.
Lemma 4.5. Let (C, p) be an object of An (S) with n = 2m or n = 2m − 1 and let bC,p : C e→C
the relative An -desingularization. We have that Jb is flat over S and its formation is compatible
with base change over S. Furthermore, we have that Jb = Ibm−1 (p) as an ideal of C,
e where Ib−1 (p) is
the ideal associated to the preimage of p through b.
Proof. The flatness and compatibility with base change are standard. If S is the spectrum of an
algebraically closed field, we know that the equality holds. We can consider the diagram
f∗ Ibm−1 (p)
0 OC f∗ OCe Q 0
and show that the composite morphism f∗ Ibm−1 (p) → Q is the zero map, restricting to the geometric
fibers over S (in fact they are both finite and flat over S). Therefore Ibm−1 (p) can be seen also as
an ideal of OC . Because the conductor ideal is the largest ideal of OC which is also an ideal of
OCe , we get an inclusion Ibm−1 (p) ⊂ Jf whose surjectivity can be checked on the geometric fibers
over S.
Remark 4.6. The stability condition for C e can be described using the Noether formula (see
Proposition 1.2 of [Cat82]). We have that ωC/S is ample if and only if ωC/S ∨
e (Jb ) is ample.
Lastly, we prove that the stack parametrizing separating An -singularities for a fixed positive
integer n is closed inside Cegr .
22
Lemma 4.7. Let C → Spec R a family of Ar -prestable curves over a DVR and denote by K the
function field of R. Suppose there exists a generic section sK of the morphism such that its image
is a separating Ar0 -singularity (with r0 6 r). Then the section sR (which is the closure of sK ) is
still a separating Ar0 -singularity.
Proof. Because sK is a separating Ar0 -singularity, then r0 is necessarily odd. Furthemore, we
have that the special fiber sk := sR ⊗R k is an Ar1 -singularity with r1 > r0 .
Let us call mR the ideal associated with the section sR . Because C/ Spec R is Ar -prestable,
we can compute dimL (OC /IR h ⊗ L) when L is the algebraic closure of either the function field
R
K or the residue field k. We have that
dimL L[[x, y]]/(y 2 − xn , mh ) = 2h − 1
for every h > 1 and every n > 2, where m = (x, y).
Therefore, dimL (OC /IR h ⊗ L) is constant on the geometric fibers over Spec R and we get that
R
h
OC /IR is R-flat because R is reduced. Consider now C e1 := BlI C, which is still R-flat, proper
R
and finitely presented thanks to Lemma 3.10 and commutes with base change. We denote by
b:C e1 → C the blowup morphism. We have that a local computations shows that if r0 > 3 we
have b−1 (sR )red = Spec R and thus it defines a section q1 of Ce1 which is a Ar −2 -singularity at the
0
generic fiber and a Ar1 −2 -singularity at the special fiber. We can therefore iterate this procedure
er /2 → Spec R which is a flat proper finitely presented morphism whose
r0 /2 times until we get C 0
generic fiber is not geometrically connected. Therefore the special fiber is not geometrically
connected as it is geometrically reduced. This clearly implies that r1 = r0 and that sR is a
separating section.
Description of An for the even case

Let n := 2m be an even number with m > 1. Firstly, we study the A2m -singularity locally, and
then we try to describe everything in families of projective curves.
Let (C, p) be a 1-dimensional reduced 1-pointed scheme of finite type over an algebraically
closed field where p is an A2m -singularity. Consider now the partial normalization in p which
gives us a finite birational morphism
b:Ce −→ C
which is infact an homeomorphism. This implies that the only think we need to understand is
how the structural sheaf changes through b. We have the standard exact sequence
0 → OC → OCe → Q → 0
where Q is a coherent sheaf on C with support on the point p. Consider now the conductor ideal
Jb of the morphism b, which is both an ideal of OC and of OCe . Consider the morphism of exact
sequences
0 OC OCe Q 0
0 OC /Jb OCe /Jb Q 0,

it is easy to see that the vertical morphism on the right is an isomorphism. Moreover, a local
computation shows that Jb = m2m q as an ideal of OCe . Therefore Lemma 3.12 and Lemma 4.5
imply that to construct an A2m -singularity we need the partial normalization C,
e the section q
2m
and a subalgebra of OCe /mq . Notice that not every subalgebra works.
23
Michele Pernice
Remark 4.8. First of all, a local computation shows that the extension OC /Jb ,→ OCe /m2m q is
2m
finite flat of degree 2. Luckily, the converse is also true: if B = k[[t]] and I = (t ), then the
subalgebras C such that C ,→ B/I is finite flat of degree 2 are exactly the ones whose pullback
through the projection B → B/I is an A2m -singularity. This should serve as a motivation for the
alternative description we are going to prove for A2m .
The idea now is to prove that the same exact picture works for families of curves. The result
holds in a greater generality, but we describe directly the case of families of Ar -prestable curves.
We want to construct an algebraic stack whose objects are triplets (C/S,e q, A) where (C/S,
e q)
is a Ar -prestable 1-pointed curve of genus g − m with some stability condition (see Remark 4.6)
and A ⊂ OCe /Iq2m is a finite flat extension of degree 2 of flat OS -algebras, where Iq is the ideal
sheaf associated to the section q.
Firstly, we introduce the stack Mfr parametrizing Ar -prestable 1-pointed curves (C, e q) such
h,[l]
that ωCe (lq) is ample. This is not difficult to describe, in fact we have a natural inclusion
fr
M fr
h,[l+1] ⊂ Mh,[l]
which is an equality for l > 2 if h > 1 and for l > 3 if h = 0. We have that the only curves that
fr r M
live in M fh,1 are curves that have one (irreducible) tail of genus 0 and the section lands
h,[l]
on the tail. We have the following result.
Proposition 4.9. In the situation above, if h > 1 we have an isomorphism
Mfr ' M fr × [A1 /Gm ]
h,[l] h,1
for l > 2. If h = 0, we have an isomorphism

M
f0,[l] ' B(Gm o Ga )
for l > 3.
Proof. The case h = 0 is straightforward. Clearly for h > 1 it is enough to construct the
isomorphism for l = 2.
In the proof of Theorem 2.9 of [DLPV21], they proved the description for r = 2, but the proof
can be generalized easily for any r. We give an alternative proof. First of all, it is easy to see that
Mfr is smooth using deformation theory.
h,[l]
r the universal
We follow the construction introduced in Section 2.3 of [DLPV21]: consider Ceh,1
fr and define Dh,1 to be the blowup of Cer × [A1 /Gm ] in the center M
curve of M fr × BGm ,
h,1 h,1 h,1
r r
where Mh,1 ,→ Ch,1 is the universal section. If we denote by q the proper transform of the closed
f e
substack
fr × [A1 /Gm ] ,→ Cer × [A1 /Gm ]
M h,1 h,1
we get that (Dh,1 fr × [A1 /Gm ], q) define a morphism

→M h,1
fr × [A1 /Gm ] −→ M
ϕ:M fr
h,1 h,[l]
fr .
which by construction is birational, as it is an isomorphism restricted to M h,1
r r 1
We define a morphism φ := (φ1 , φ2 ) : Mh,[l] → Mh,1 × [A /Gm ] and prove that it is a
f f
quasi-inverse of ϕ.
Let (C̃, q) be an object of Mfr and let Σ be the Cartier divisor O(q) associated to the
h,[l]
section q. Then (C̃, Σ) is a Ar -semistable pair, therefore we can apply Proposition 1.17 to get the
24
contraction c : C̃ → C. We define the image of (C̃, q) through φ1 to be the 1-pointed Ar -stable

curve (C, p) where p := c(q).
We denote by ∆0 the substack of Mfr parametrizing curves with a separating node such that
h,[l]
one of the two connected components of the normalization is of genus 0. This is closed because of
fr . Therefore we can
Lemma 4.7. It is easy to see that it is in fact a Cartier divisor inside M h,[l]
fr → [A1 /Gm ] as the morphism associated to the effective Cartier divisor ∆0 .
define φ2 : M h,[l]
Proposition 1.17 implies that there exists an isomorphism of functors φ ◦ ϕ ' id. The universal
property of the blowup implies that we have a natural transformation id ⇒ ϕ ◦ φ and we can check
that it is an isomorphism on the geometric points of Mfr × [A1 /Gm ]. Lemma 1.18 concludes the
h,1
proof.
Finally, we are ready to describe An . In Section 3.1, we define the stack Fnc which parametrizes
c
pointed finite flat curvilinear algebras of length n and Em,d which parametrizes finite flat curvilinear
extensions of degree d of pointed finite flat curvilinear algebras of degree m. We proved that the
natural morphism
c c
Em,d −→ Fmd
defined by the association (A ,→ B) 7→ B is an affine bundle. See Lemma 3.8.
Let
c c
Em,2 −→ F2m
as above and consider the morphism of stacks
fr
M c
g−m,[2m] −→ F2m
e q) 7→ O e /Iq2m where Iq is the ideal defined by the section q.

defined by the association (C, C
0
We denote by A2m the fiber product Em,2 c ×F2m Mfr 0
g−m,[2m] . By definition, an object of A2m
c
e q, A ⊂ O e /Iq2m ) where A ⊂ O e /Iq2m is a finite flat extension of algebras

over S is of the form (C, C C
e1 /S, q1 , A1 ⊂ O e /I 2 m) and (π2 : C
of degree 2. Given two objects (π1 : C e2 /S, q2 , A2 ⊂ O e /I 2 m),
C1 q1 C q2
e1 , q1 ) → (C
a morphism over S between them is a pair (f, α) where f : (C e2 , q2 ) is a morphism in
r
Mg−m,[2m] (S) while α : A2 → A1 is an isomorphism of finite flat algebras over S such that the
f
diagram
α
A2 A1
π2,∗ (OCe2 /Iq2m

2
) π1,∗ (OCe1 /Iq2m
1
)
is commutative.
We want to construct a morphism from A02m to A2m . Let S be a scheme and (C,
e q, A ⊂ O e /I 2m )
C q
be an object of A02m (S). We consider the diagram
SpecOS (OCe /Iq2m ) C

e
2:1
SpecOS (A)
25
Michele Pernice
and complete it with the pushout (see Lemma 3.9), which has a morphism over S:
SpecOS (OCe /Iq2m ) C

e
2:1
SpecOS (A) C
S.
e and C share the same topological space, whereas the structural sheaf OC of the
In this case, C
pushout is the fiber product
OC := OCe ×OCe /Iq2m A OCe
A OCe /Iq2m ;
we define Ip := Iq |OC which induces a section p : S → C of C → S.

Lemma 4.10. In the situation above, C/S is an Ar -stable curve of genus g and ps is a A2m -
singularity for every geometric point s ∈ S. Furthermore, the formation of the pushout commutes
with base change over S.
Proof. The fact that C/S is flat, proper and finitely presented is a consequence of Lemma 3.9.
The same is true for the compatibility with base change over S. We only need to check that Cs is
an Ar -stable curve for every geometric point s ∈ S. Therefore we can assume S = Spec k with k
an algebraically closed field over κ. Connectedness is trivial as the topological space is the same.
We need to prove that p is an A2m -singularity. Consider the cartesian diagram of local rings
OC,p := OC,q
e ×OC,p 2m A
e /mq
OC,q
e
A OC,q 2m
e /mq
and pass to the completion with respect to mp , the maximal ideal of OC,p . Because the extensions
are finite, we get the following cartesian diagram of rings
B k[[t]]
φ2
k[[t]]/(tm ) k[[t]]/(t2m );
using the description of φ2 as in Lemma 3.6, it is easy to see that it is defined by the association
t 7→ t2 up to an isomorphism of k[[t]]/(t2m ). This concludes the proof.
Therefore we have constructed a morphism of algebraic stacks
F : A02m −→ A2m
defined on objects by the association
G
e q, A ⊂ O e ) 7→ (C
(C, e Spec A, p)
C
Spec(OCe /Iq2m )
26
and on morphisms in the natural way using the universal property of the pushout.
To construct the inverse, we use the relative An -desingularization which is compatible with
base change. Lemma 4.5 implies that we can define a functor G : A2m → A02m on objects
e q, OC /Jb ⊂ O e /Iq2m )
(C/S, p) 7→ (C, C
where b : C e → C is the relative A2m -desingularization, Jb is the conductor ideal relative to b

and q is the smooth section of C e → S defined as the vanishing locus of the 0-th Fitting ideal of
Ωb−1 (p)|S . It is defined on morphisms in the obvious way.
Proposition 4.11. The two morphisms F and G are quasi-inverse of each other.
Proof. The statement is a conseguence of Proposition 3.13 and Proposition 3.16.

fr
Corollary 4.12. A2m is an affine bundle of dimension (m − 1) over the stack M g−m,[2m] for
m > 1.
Proof. Because A02m is constructed as the fiber product Em,2

c ×F2m
c fr
M g−m,[2m] , the statement
follows from Lemma 3.8.
4.1 Description of An for the odd case

Let n := 2m − 1 and (C, p) be a 1-dimensional reduced 1-pointed scheme of finite type over an
algebraically closed field and p is an A2m−1 -singularity. Consider now the partial normalization
in p which gives us a finite birational morphism
e −→ C
b:C
which is not an homeomorphism, as we know that b−1 (p) is a reduced divisor of C e of length 2.
We can use Lemma 3.12 to prove that the extension OC ,→ b∗ OCe can be constructed pulling
back a subalgebra of b∗ OCe /Jb through the quotient b∗ OCe → b∗ OCe /Jb , as in the even case. We
can describe the subalgebra in the following way.
Consider the divisor b−1 (p) which is the disjoint union of two closed points, namely q1 and q2 .
Then the composition
OC /Jb f∗ OCe /Jb = Omq1 ⊕ Omq2 Omqi
is an isomorphism for i = 1, 2, where Omqi is the structure sheaf of the support of the Cartier
divisor mqi for i = 1, 2 and the second map is just the projection. Therefore the subalgebra OC /Jb
of f∗ OCe /Jb is determined by an isomorphism between Omq1 and Omq2 . Recall that Remark 1.3
implies that we can have two different situation: either C e is connected and its genus is g − m
or it has two connected components of total genus g − m + 1 and the two points lie in different
components.
Definition 4.13. Let 0 6 i 6 (g − m + 1)/2. We define Ai2m−1 to be the substack of A2m−1

parametrizing 1-pointed curves (C/S, p) such that the geometric fibers over S of the relative
A2m−1 -desingularization are the disjoint union of two curves of genus i and g − m − i + 1.
Furthermore, we define Ans2m−1 the substack of A2m−1 parametrizing curves (C/S, p) such that
the geometric fibers of the relative A2m−1 -desingularization are connected of genus g − m.
Proposition 4.14. The algebraic stack A2m−1 is the disjoint union of Ans i
2m−1 and A2m−1 for
every 0 6 i 6 (g − m + 1)/2.
27
Michele Pernice
Proof. Let (C/S, p) be an object of A2m−1 and consider the relative

An -desingularization
e q) → (C, p)
b : (C,
over S. Because C e → S is flat, proper, finitely presented and the fibers over S are geometrically
reduced, we have that the number of connected components of the geometric fibers of C e over S is
ns
locally constant. See Proposition 15.5.7 of [Gro66]. Therefore we have that A2m−1 is open and
closed inside A2m−1 . Furthermore, we have that the objects of the complement As2m−1 of Ans 2m−1
are pairs (C/S, p) such that the geometric fibers of C e over S have two connected components.
Suppose that S = Spec R with R a strictly henselian ring over κ. We know that b−1 (p) is étale of
degree 2 over R, thus it is the disjoint union of two copies of R. We denote by q one of the two
sections of Ce → Spec R and we denote by Cs0 the connected component of the fiber C es which
contains qs for every point s ∈ Spec R. Proposition 15.6.5 and Proposition 15.6.8 in [Gro66] imply
that the set-theoretic union
[
C0 := Cs0
s∈S
is a closed and open subscheme of C, e therefore the morphism C0 → S is still proper, flat and
finitely presented. In particular, the arithmetic genus of the fibers is locally constant. It is easy to
see that in particular As2m−1 is the disjoint union of Ai2m−1 for 0 6 i 6 (g − m + 1)/2.
We start by describing Ans fr

2m−1 . Let Mh,2[l] be a fibered category in groupoid over the category
of schemes whose objects are of the form (C/S,
e e → S is a flat, proper, finitely
q1 , q2 ) where C
presented morphism of scheme, q1 and q2 are smooth sections, Cs is an Ar -prestable curve of genus
e
h for every geometric point s ∈ S and ωCe (lq1 + lq2 ) is relatively ample over S. The morphisms
are defined in the obvious way.
Proposition 4.15. We have an isomorphism of fibered category
fr
M fr 1 1
h,2[l] ' Mh,2 × [A /Gm ] × [A /Gm ]
for l > 2 and h > 1. Furthermore, we have

f0,2[l] ' BGm × [A1 /Gm ]
M
for l > 3.
Proof. This follows easily from Proposition 4.9.
fr
We want to construct an algebraic stack with a morphism over M h,2[m] whose fibers parametrize
the isomorphisms between the two finite flat S-algebras Omq1 and Omq2 .
Remark 4.16. Recall that we have a smooth stack Em,d c which parametrizes finite flat extensions
A ,→ B of degree d with B curvilinear of length m. If d = 1, the stack Em,1 c parametrizes
isomorphisms of finite flat algebras of length m. We also have a map
c c c
Em,1 −→ Fm × Fm
defined on objects by the association (A ,→ B) 7→ (A, B), which is a Gm × Gm−2
a -torsor. See
Section 3.1 for a more detailed discussion.
Consider the morphism

fr
M c c
g−m,2[m] −→ Fm × Fm
28
defined by the association

e q1 , q2 ) 7→ (O e /I m , O e /I m );
(C, C q1 C q2
ns be the fiber product M fr c
and let I2m−1 h,2[m] ×(Fm
c ×F c ) Em,1 . It parametrizes objects (C, q1 , q2 , φ)
m
fr
such that (C, q1 , q2 ) ∈ M h,2[m] and φ is an isomorphism between Omq1 and Omq2 as OS -algebras
which commutes with the sections.
We can construct a morphism
ns
I2m−1 −→ Ans
2m−1
e q1 , q2 , φ) ∈ I ns (S), then we have the diagram
in the following way: let (C, 2m−1
F
SpecS (Omq1 ) SpecS (Omq2 ) C
e
(id,φ)
SpecS (Omq1 )
where the morphism (id, φ) is étale of degree 2. We denote by C the pushout of the diagram and
p the image of q. We send (C,e q1 , q2 , φ) to (C, q). Notice that because both q1 and q2 are smooth
sections, we have that Omqi is the scheme-theoretic support of a Cartier divisor of C, e therefore it
is flat for i = 1, 2. Lemma 3.9 assures us that this construction is functorial and commutes with
base change.
Proposition 4.17. The pushout functor
F ns : I2m−1
ns
−→ Ans
2m−1
is representable finite étale of degree 2.

Proof. It is a direct conseguence of both Proposition 3.13 and Proposition 3.16. In fact the
two propositions assure us that the object (C, p) is uniquely determined by the relative A2m−1 -
desingularization b : C e → C, the fiber b−1 (p) and an automorphism of the m-thickening of b−1 (p)
which restricted to the geometric fibers over S acts exchanging the two points of b−1 (p). Because
b−1 (p) is finite étale of degree 2, it is clear that F ns is a finite étale morphism of degree 2.
Let 0 6 i 6 (g − m + 1)/2. In the same way, we can define morphisms
M fg−i−m+1,[m] −→ F c × F c
fi,[m] × M
m m
defined by the association

(C e2 , q2 ) →
e1 , q1 ), (C 7 (OCf1 /Iqm1 , OCe2 /Iqm2 )
i
and we denote by I2m−1 the fiber product
c
(M
fi,[m] × M
fg−m−i+1,[m] ) ×(F c ×F c ) Em,1
m m
.
Similarly to the previous case, we can construct a functor
F i : I2m−1
i
−→ Ai2m−i
using the pushout construction. Again, we have the following result.
Proposition 4.18. The functor F i is an isomorphism for i 6= (g − m + 1)/2 whereas is finite
étale of degree 2 for i = (g − m + 1)/2.
Proof. The proof is exactly the same as Proposition 4.17.
29
Michele Pernice
References
[Cat82] F. Catanese, Pluricanonical Gorenstein Curves, Enumerative Geometry and Classical Algebraic Geometry
(1982), 51–95.
[CL21] S. Canning and H. Larson, The Chow rings of the moduli spaces of curves of genus 7, 8, and 9 (2021).
[DLPV21] A. Di Lorenzo, M. Pernice, and A. Vistoli, Stable cuspidal curves and the integral Chow ring of M 2,1 ,
arXiv, 2021.
[DLV21] A. Di Lorenzo and A. Vistoli, Polarized twisted conics and moduli of stable curves of genus two (2021).
[EG98] D. Edidin and W. Graham, Equivariant intersection theory, Inv. math. 131, 595-634 (1998).
[Fab90] C. Faber, Chow Rings of Moduli Spaces of Curves I: The Chow Ring of M3 , Annals of Mathematics
132 (1990), 331–419.
[Gro66] A. Grothendieck, Éléments de géométrie algébrique : IV. Étude locale des schémas et des morphismes
de schémas, Troisième partie, Publications Mathématiques de l’IHÉS 28 (1966), 5–255.
[Iza95] E. Izadi, The Chow Ring of the Moduli Space of Curves of Genus 5 (1995), 267–303.
[Knu83] F. F. Knudsen, The projectivity of the moduli space of stable curves, II: The stacks Mg,n , Mathematica
Scandinavica 52 (1983), 161–199.
[Lar19] E. Larson, The integral Chow ring of M 2 , arXiv:1904.08081 (2019).
[Mum83] D. Mumford, Towards an Enumerative Geometry of the Moduli Space of Curves, Artin M., Tate J.
(eds) Arithmetic and Geometry. Progress in Mathematics, vol 36. Birkhäuser, Boston, MA (1983).
[PV13] N. Penev and R. Vakil, The Chow ring of the moduli space of curves of genus 6, Algebraic Geometry 2
(2013).
[Sta22] The Stacks project authors, The stacks project, 2022.
[TV10] M. Talpo and A. Vistoli, Deformation theory from the point of view of fibered categories (2010).
Michele Pernice m.pernice(at)kth.se

KTH, Room 1642, Lindstedtsvägen 25, 114 28, Stockholm
30

The Moduli Stack of A - Stable Curves: Michele Pernice

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Moduli Stack of A - Stable Curves: Michele Pernice

Uploaded by

Copyright:

Available Formats

The moduli stack of Ar -stable curves

2020 Mathematics Subject Classification 14H10 (primary), 14H20 (secondary)

Outline of the paper

1. Ar -stable curves and contractions

Fix a nonnegative integer r.

following two conditions are verified:

Remark 1.5. It is important to remark that if (C, Σ) is a Ar -semistable pair and Γ is an

H0 (L⊗2 ) ⊗ H0 (L⊗m−2 ) ⊗ H 0 (L⊗km ) H0 (L⊗m ) ⊗ H0 (L⊗km )

Figure 1. Contraction morphisms

Proof. Because ωC (Σ)⊗k restricted to the non-trivial fibers of c is non-canonically isomorphic to

2. Moduli stack of Ar -stable curves

Proof. It follows from Proposition 1.7 that if (π : C → S, Σ1 , . . . , Σn ) is a n-pointed Ar -stable

Proposition 2.4. Let π : C → S be a n-pointed Ar -stable of genus g. Then π∗ ωC/S is a locally

r is the universal curve of M

A[[x, y]]/(y 2 − x3 − rx).

3.1 Extensions of finite curvilinear algebras

where φd (t) = td p(t) with p(0) ∈ OS×0 .

Proposition 3.7. We have the follow isomorphism of fibered categories

Proof. It follows from Lemma 3.6.

Lemma 3.8. In the situation above, we have the isomorphism

Proof. Consider the morphism

3.2 Pushout and blowups in families

is still a morphism of exact sequences. Because also B e ×g A e→Q

4. The substack of Ar -singularity

Proposition 4.1. The stack An is smooth.

with sections qh : S → Ceh where m := n/2, the morphism bh : C em := Blq C

Definition 4.3. Let (C, p) be an object of An (S) with n = 2m or n = 2m − 1 for S any

Description of An for the even case

0 OC /Jb OCe /Jb Q 0,

for l > 2. If h = 0, we have an isomorphism

we get that (Dh,1 fr × [A1 /Gm ], q) define a morphism

contraction c : C̃ → C. We define the image of (C̃, q) through φ1 to be the 1-pointed Ar -stable

e q) 7→ O e /Iq2m where Iq is the ideal defined by the section q.

e q, A ⊂ O e /Iq2m ) where A ⊂ O e /Iq2m is a finite flat extension of algebras

π2,∗ (OCe2 /Iq2m

SpecOS (OCe /Iq2m ) C

SpecOS (OCe /Iq2m ) C

we define Ip := Iq |OC which induces a section p : S → C of C → S.

where b : C e → C is the relative A2m -desingularization, Jb is the conductor ideal relative to b

Proof. The statement is a conseguence of Proposition 3.13 and Proposition 3.16.

Proof. Because A02m is constructed as the fiber product Em,2

4.1 Description of An for the odd case

Definition 4.13. Let 0 6 i 6 (g − m + 1)/2. We define Ai2m−1 to be the substack of A2m−1

Proof. Let (C/S, p) be an object of A2m−1 and consider the relative

We start by describing Ans fr

for l > 2 and h > 1. Furthermore, we have

Consider the morphism

defined by the association

is representable finite étale of degree 2.

defined by the association

Michele Pernice m.pernice(at)kth.se

You might also like