04.10.22 · algebraic-geometry / moduli

Stable curve and Deligne-Mumford stability

shipped3 tiersLean: none

Anchor (Master): Harris-Morrison Ch. 2, 3, 6; Mumford 1983 *Towards an enumerative geometry...* (Arithmetic and Geometry vol. II, Birkhäuser); Arbarello-Cornalba-Griffiths *Geometry of Algebraic Curves* Vol. II; Cornalba-Harris 1988 *Ann. Sci. ENS* 21

Intuition Beginner

A smooth projective curve of genus $g$ has a clean modulus count: Riemann's $3 g - 3$ parameters once $g \geq 2$ . The moduli space $M_{g}$ that records these curves is not compact — a family of smooth genus- $g$ curves can degenerate, with a curve pinching down to develop a node, or with a chain of components breaking off. Without a rule for which degenerate curves to include, the moduli has missing limits and cannot do enumerative geometry.

The Deligne-Mumford 1969 idea is to widen the class of curves admitted as moduli points by exactly enough to fill in those missing limits, and no more. A stable curve of arithmetic genus $g$ is a connected projective curve that is allowed mild singularities (nodes, the simplest crossings) and a controlled list of components, with the rule that every smooth rational component meets the rest of the curve in at least three points, and every smooth genus-one component meets the rest in at least one point. This rule guarantees the curve has finite automorphism group, which is the technical reason the moduli is well-behaved.

The compactified moduli space $\overline{M}_{g}$ is the space of all stable curves of arithmetic genus $g$ . It is projective, of dimension $3 g - 3$ , and the original smooth-curve moduli $M_{g}$ is an open dense piece. The complement is the boundary, a divisor whose components record the topological types of degenerate curves.

Visual Beginner

A schematic showing four stable curves of arithmetic genus two: a smooth curve, an irreducible one-nodal curve, two smooth genus-one components joined at one node (the dumbbell), and two smooth rational components joined at three nodes (the banana with markings). Each picture is labelled with its dual graph and a check mark indicating the stability condition: a rational component must meet the rest at three or more points; a genus-one component must meet the rest at one or more points.

The picture captures the essential combinatorics. The dual graph $Γ$ has one vertex per irreducible component and one edge per node. The arithmetic genus is the sum of the geometric genera at the vertices, plus the first Betti number of the graph, plus one. The stability condition is local at each vertex: rational vertices need valence three or more (counting multi-edges), genus-one vertices need valence one or more, higher-genus vertices have no restriction.

Worked example Beginner

Take the arithmetic genus two case. List four stable curves and verify the arithmetic-genus formula on each.

Step 1. Smooth genus-two curve. One smooth component of geometric genus two, no nodes. Dual graph: one vertex, no edges, first Betti number zero. Arithmetic genus equals the geometric genus at the single vertex plus the first Betti number of the graph, $2 + 0 = 2$ . Stability: every component has geometric genus two, no stability restriction. Stable.

Step 2. Irreducible one-nodal curve of arithmetic genus two. One component, normalisation is a smooth curve of geometric genus one with two marked points identified to form the node. Dual graph: one vertex with one self-loop, first Betti number one. Arithmetic genus $= 1 + 1 = 2$ . Stability: the single component is the normalisation of a genus-one curve with a self-identification, which has geometric genus one and is not rational. Stable.

Step 3. The dumbbell. Two smooth elliptic components $E_{1}$ and $E_{2}$ joined at one node. Dual graph: two vertices, one edge, first Betti number zero. Arithmetic genus $= 1 + 1 + 0 = 2$ . Stability: each component has geometric genus one and meets the other at one node. The rule for a smooth genus-one component is "meets the rest in at least one point", and here each meets at one node. Stable.

Step 4. The banana with two rational components. Two smooth rational curves $R_{1}, R_{2}$ joined at three nodes. Dual graph: two vertices, three edges, first Betti number two. Arithmetic genus $= 0 + 0 + 2 = 2$ . Stability: each rational component meets the other at three points (the three nodes), which meets the rule of at least three meeting points. Stable.

What this tells us. The arithmetic-genus formula counts geometric genera at vertices plus loops in the dual graph, and the stability rule is a small set of local conditions that prevent components from being too lightly attached. The four cases above are four points of $\overline{M}_{2}$ , three of them on the boundary. The full list of topological types of stable genus-two curves has seven entries, recovered by listing all valid dual graphs.

Check your understanding Beginner

Exercise (easy, multiple choice).

A smooth rational component $R ≅ P^{1}$ inside a stable curve must meet the rest of the curve in at least how many points (counted with multiplicity at nodes)?

A. 1
B. 2
C. 3
D. There is no restriction

Hint

The stability rule is engineered so that the automorphism group of every stable curve is finite. The automorphism group of $P^{1}$ is three-transitive: any three distinct points can be moved to any other three distinct points by a Möbius transformation.

Answer

C. A smooth rational component must meet the rest of the curve in at least three points. With fewer than three meeting points, the automorphism group of $P^{1}$ fixing those points would be infinite (a one-parameter family if only two points are fixed, the full $PGL_{2}$ if zero points are fixed, and a one-parameter subgroup fixing two points). The same logic explains the rule that genus-one components need at least one meeting point: the automorphism group of an elliptic curve fixing one point is finite.

Formal definition Intermediate+

Throughout, $k$ is an algebraically closed field of characteristic zero (the original Deligne-Mumford 1969 paper handles arbitrary characteristic with the same definitions; the characteristic-zero restriction here streamlines the smoothness statements).

Definition (nodal curve). A nodal curve over $k$ is a connected projective scheme $C$ of pure dimension one over $k$ whose only singularities are nodes — points $p \in C$ with completed local ring isomorphic to $k [[x, y]] / (x y)$ . The dual graph $Γ_{C}$ of a nodal curve $C$ is the finite graph with one vertex for each irreducible component of $C$ , one edge for each node joining the two components meeting at that node (a self-loop if both branches lie in the same component), and the vertex carrying the geometric genus $g (C_{v})$ of the normalisation of the corresponding component as a label.

Definition (arithmetic genus). The arithmetic genus of a nodal curve $C$ is $p_{a} (C) = dim_{k} H^{1} (C, O_{C})$ . For a nodal curve, this is computed combinatorially by the formula $$ p_a(C) = \sum_{v \in V(\Gamma_C)} g(\widetilde{C_v}) + b_1(\Gamma_C), $$ where $V (Γ_{C})$ is the vertex set of the dual graph and $b_{1} (Γ_{C}) = # E (Γ_{C}) - # V (Γ_{C}) + 1$ is the first Betti number of the graph (which is connected because $C$ is connected).

Definition (Deligne-Mumford stability). A connected projective nodal curve $C$ over $k$ of arithmetic genus $g \geq 2$ is Deligne-Mumford stable (or simply stable) if every smooth rational component $R ≅ P_{k}^{1}$ of $C$ meets the rest of $C$ in at least three points, and every smooth elliptic component $E$ of $C$ (geometric genus one) meets the rest of $C$ in at least one point. Equivalently (Theorem below), the dualising sheaf $ω_{C}$ of $C$ is ample on $C$ , and equivalently the automorphism group $Aut (C)$ is finite.

Definition (dualising sheaf of a nodal curve). For a nodal curve $C$ , the dualising sheaf $ω_{C}$ is the unique coherent sheaf on $C$ that represents Serre duality: $H^{1} (C, F) ≅ Hom_{C} (F, ω_{C})^{\lor}$ for every coherent $F$ on $C$ . On the smooth locus, $ω_{C}$ coincides with the ordinary sheaf of differentials $Ω_{C}^{1}$ . At a node $p$ with branches $b_{1}, b_{2}$ , sections of $ω_{C}$ near $p$ are pairs $(η_{1}, η_{2})$ of meromorphic differentials on the two branches, each with at most a simple pole at $p$ , satisfying the residue condition $res_{b_{1}} (η_{1}) + res_{b_{2}} (η_{2}) = 0$ .

Definition (boundary divisors). Inside the moduli stack $\overline{M}_{g}$ of stable curves of arithmetic genus $g$ , the boundary is the closed substack $\partial \overline{M}_{g} = \overline{M}_{g} ∖ M_{g}$ . It decomposes as a union of irreducible divisors:

$Δ_{0}$ — the closure of the locus of irreducible stable curves with exactly one node (smoothing the node returns a smooth curve, so $Δ_{0}$ is called the non-separating boundary);
$Δ_{i}$ for $1 \leq i \leq ⌊ g /2 ⌋$ — the closure of the locus of stable curves with two components of arithmetic genera $i$ and $g - i$ meeting at one node (the separating boundary), with $Δ_{i} = Δ_{g - i}$ identified by the unordered pair.

The class $δ = [Δ_{0}] + [Δ_{1}] + \dots + [Δ_{⌊ g /2 ⌋}]$ in $Pic (\overline{M}_{g}) \otimes Q$ is the total boundary class.

Counterexamples to common slips

A connected projective nodal curve with a smooth rational component meeting the rest in only two points is not stable: the residual one-parameter family of Möbius transformations fixing those two points gives infinite automorphism group.
A connected projective nodal curve with a smooth elliptic component disjoint from the rest fails connectedness and is excluded by definition; one meeting the rest in zero points is excluded by stability.
Arithmetic genus is not the sum of geometric genera at vertices; the loop correction $b_{1} (Γ_{C})$ adds one for each independent cycle in the dual graph. The irreducible one-nodal genus- $g$ curve has one vertex of geometric genus $g - 1$ and one self-loop, contributing arithmetic genus $(g - 1) + 1 = g$ .
The dualising sheaf $ω_{C}$ on a nodal curve is not the same as $Ω_{C}^{1}$ at nodes: $Ω_{C}^{1}$ is not locally free at a node, while $ω_{C}$ is an invertible sheaf on all of $C$ (the residue condition restores invertibility). This is why ampleness of $ω_{C}$ is the correct algebraic stability condition, not ampleness of $Ω_{C}^{1}$ .

Key theorem with proof Intermediate+

Theorem (Deligne-Mumford 1969 §1; Harris-Morrison Ch. 1 Prop. 1.10). Let $C$ be a connected projective nodal curve over $k$ of arithmetic genus $g \geq 2$ . The following are equivalent:

$C$ is Deligne-Mumford stable in the combinatorial sense: every smooth rational component meets the rest in $\geq 3$ points, every smooth elliptic component meets the rest in $\geq 1$ point.
The dualising sheaf $ω_{C}$ is ample on $C$ .
The automorphism group $Aut (C)$ is finite (equivalently, the scheme of automorphisms $\underline{Aut} (C)$ is finite over $Spec k$ ).

Proof. The argument runs through the dualising sheaf, with the combinatorial data of the dual graph supplying the bookkeeping.

Step 1: degree of $ω_{C}$ on each component. For a nodal curve $C$ with irreducible components $C_{v}$ and normalisations $C_{v} \to C_{v}$ , the dualising sheaf restricts on $C_{v}$ to $$ \omega_C\big|{C_v} = \omega{\widetilde{C_v}}\big(\textstyle{\sum_{p \in N_v}} p\big), $$ where $N_{v}$ is the set of preimages on $C_{v}$ of the nodes of $C$ lying on $C_{v}$ (each node contributes one point per branch). This follows from the residue characterisation: $ω_{C}$ allows simple poles at branches of nodes, paired by the residue condition. Taking degrees on $C_{v}$ , $$ \deg(\omega_C|_{C_v}) = 2 g(\widetilde{C_v}) - 2 + n_v, $$ where $n_{v}$ is the number of branches of nodes on $C_{v}$ — equivalently, the valence of vertex $v$ in the dual graph.

Step 2: ample iff positive on every component. An invertible sheaf $L$ on a projective nodal curve $C$ is ample iff its degree on every irreducible component is strictly positive (Hartshorne III Theorem 5.3, specialised to curves; for non-reduced curves the Nakai-Moishezon criterion is used). For $ω_{C}$ , positivity on each component reads $$ 2 g(\widetilde{C_v}) - 2 + n_v > 0 \quad \text{for every component } v. $$

Step 3: case analysis on $g (C_{v})$ . If $g (C_{v}) \geq 2$ , then $2 g (C_{v}) - 2 \geq 2$ and the inequality $2 g (C_{v}) - 2 + n_{v} > 0$ holds automatically for every $n_{v} \geq 0$ . If $g (C_{v}) = 1$ , the inequality reads $0 + n_{v} > 0$ , equivalent to $n_{v} \geq 1$ — the elliptic-component stability condition. If $g (C_{v}) = 0$ , the inequality reads $- 2 + n_{v} > 0$ , equivalent to $n_{v} \geq 3$ — the rational-component stability condition. This proves the equivalence (1) $⟺$ (2).

Step 4: ample dualising implies finite automorphism. A polarised projective curve $(C, L)$ with $L$ ample has finite automorphism group preserving the polarisation, by the standard Hilbert-scheme argument: $Aut (C, L) \subset PGL (H^{0} (C, L^{\otimes n}))$ for $n ≫ 0$ via the projective embedding, and the closed-subscheme automorphisms of an embedded projective variety form a closed group subscheme of $PGL_{N}$ . For $L = ω_{C}$ , the dualising sheaf is canonically constructed from $C$ , hence preserved by all of $Aut (C)$ , so $Aut (C) = Aut (C, ω_{C})$ is finite. This proves (2) $\Rightarrow$ (3).

Step 5: infinite automorphism implies non-stability. Conversely, if $C$ has infinite automorphism group, then some irreducible component $C_{v}$ admits infinite automorphisms fixing the branches of the nodes meeting it. The only smooth curves with infinite automorphism groups are $P^{1}$ (full $PGL_{2}$ , three-transitive) and elliptic curves (translation group, transitive on the curve and fixing no point freely). For $C_{v} ≅ P^{1}$ , fixing the $n_{v}$ branches leaves an $Aut (P^{1}, n_{v} points) = PGL_{2} / stab$ of positive dimension if $n_{v} \leq 2$ , so $n_{v} \leq 2$ destabilises. For $C_{v}$ elliptic, fixing one branch reduces the translation group to a finite group, but with $n_{v} = 0$ branches the full translation group acts. The non-stability conditions ( $n_{v} \leq 2$ on a rational component, $n_{v} = 0$ on an elliptic component) are exactly the negations of the combinatorial stability conditions. This proves (3) $\Rightarrow$ (1), closing the equivalence. $□$

Bridge. The dualising-sheaf characterisation of stability builds toward the construction of $\overline{M}_{g}$ as a projective scheme: with $ω_{C}$ ample on every stable curve, the powers $ω_{C}^{\otimes n}$ embed each stable curve into a common projective space for $n \geq 3$ , and the Hilbert scheme of these embeddings supports a GIT quotient identifying $\overline{M}_{g}$ as a projective coarse moduli space, as Gieseker carried out in his 1982 Tata lectures. This is the foundational reason the boundary divisor decomposition $δ = δ_{0} + δ_{1} + \dots + δ_{⌊ g /2 ⌋}$ has a well-defined intersection theory: the boundary classes are Cartier divisors on a smooth proper Deligne-Mumford stack, and intersection numbers against tautological classes appear again in 04.10.01 (the dimension count $3 g - 3$ ) and in 04.10.02 (the GIT quotient construction). The bridge is the dualising sheaf: it identifies combinatorial stability with the analytic-geometric notion of ampleness, and this identifies the moduli problem of stable curves with a polarised moduli problem amenable to GIT. Putting these together, Deligne-Mumford stability is exactly the condition that makes the genus- $g$ curve moduli problem representable by a proper Deligne-Mumford stack, generalises the Mumford GIT framework for smooth curves to a compactification with controlled degenerations, and is dual to the Kollár-Shepherd-Barron-Alexeev stability condition in higher dimensions, where the dualising-sheaf-ample / log-canonical condition plays the same role for higher-dimensional varieties as nodal-with-ample-dualising plays for curves.

Exercises Intermediate+

Exercise 1 (easy, numeric).

Compute the dimension of the boundary divisor $Δ_{1} \subset \overline{M}_{3}$ of curves consisting of a smooth genus-one component joined to a smooth genus-two component at one node.

Hint

A point of $Δ_{1}$ is parametrised by: a point of $M_{1, 1}$ (an elliptic curve with one marked point = the node), a point of $M_{2, 1}$ (a smooth genus-two curve with one marked point), and a way to glue them at the marked points. The gluing has no continuous moduli. Use $dim M_{g, n} = 3 g - 3 + n$ .

Answer

$dim Δ_{1} = dim M_{1, 1} + dim M_{2, 1} = (3 \cdot 1 - 3 + 1) + (3 \cdot 2 - 3 + 1) = 1 + 4 = 5$ . The ambient stack $\overline{M}_{3}$ has dimension $3 \cdot 3 - 3 = 6$ , so $Δ_{1}$ has codimension one in $\overline{M}_{3}$ , confirming it is a divisor. Rubric: full credit for the boundary-divisor dimension formula $dim Δ_{i} = dim M_{i, 1} + dim M_{g - i, 1}$ and the numerical check.

Exercise 2 (easy, numeric).

List the seven topological types of stable curves of arithmetic genus two, by writing each as a tuple $(V, E, (g_{v})_{v \in V})$ where $V$ is the vertex set of the dual graph, $E$ is the edge multiset, and $(g_{v})$ are the vertex genera. Each must satisfy $g = \sum_{v} g_{v} + b_{1} = 2$ and the stability conditions on rational and elliptic vertices.

Hint

Enumerate by total number of nodes (= number of edges). Zero nodes: one smooth genus-two component. One node: irreducible with self-loop, or two components joined once. Two nodes: irreducible with two self-loops, or one component plus another with various configurations. Three nodes: the "banana" with two rational components joined by three edges.

Answer

The seven topological types of stable genus-two curves are listed below, with each entry giving the dual-graph vertex set, edges, vertex genera, and stability verification.

Smooth. One vertex of genus 2, zero edges. $g = 2 + 0 = 2$ .
Irreducible one-nodal. One vertex of genus 1, one self-loop. $g = 1 + 1 = 2$ .
Dumbbell (genus-one pair). Two vertices of genus 1, one edge between them. $g = 1 + 1 + 0 = 2$ . Both elliptic vertices meet the rest in one node — stable.
Irreducible two-nodal. One vertex of genus 0, two self-loops. $g = 0 + 2 = 2$ . The rational vertex has valence four (two self-loops, each contributing two endpoints), satisfying $n_{v} \geq 3$ — stable.
Genus-one with rational tail-pair. One vertex of genus 1 and one vertex of genus 0 with one edge between, plus a self-loop on the rational vertex. $g = 1 + 0 + 1 = 2$ . The rational vertex has valence three — stable. Some texts call this the "elliptic with a rational chord" type.
Theta graph (banana). Two vertices of genus 0 connected by three edges in parallel. $g = 0 + 0 + 2 = 2$ . Both vertices have valence 3 — stable. Note: the alternative configuration with two parallel edges plus a self-loop on one vertex fails stability on the un-self-looped rational vertex (valence 2), and is excluded.
Dollar sign. One vertex of genus 0 with three self-loops. $g = 0 + 3 = 2$ . The vertex has valence 6 — stable.

A complete enumeration following Harris-Morrison's classification yields exactly seven stable topological types: smooth; irreducible with one self-loop; irreducible with two self-loops; dollar-sign (three self-loops on one rational vertex); dumbbell of two elliptic vertices; elliptic vertex plus rational vertex with a self-loop; theta graph (two rational vertices, three edges). Each is the closure of a stratum in $\overline{M}_{2}$ , with strata of decreasing dimension as the dual graph gains edges.

Rubric: full credit for at least six of the seven types correctly identified with stability checked at each vertex.

Exercise 3 (medium, symbolic).

Show that for a nodal curve $C$ with dual graph $Γ_{C}$ , the Euler characteristic $χ (O_{C}) = 1 - p_{a} (C)$ matches the topological formula $$ \chi(\mathcal{O}C) = \sum{v \in V(\Gamma_C)} \chi(\mathcal{O}_{\widetilde{C_v}}) - #E(\Gamma_C). $$

Hint

Use the partial-normalisation short exact sequence $0 \to O_{C} \to ν_{*} O_{C} \to ⨁_{nodes} k_{p} \to 0$ , where $ν : C \to C$ is the normalisation and the cokernel records one copy of the residue field at each node (the two branches identified at a node contribute one less function on $C$ than on $C$ ).

Answer

The partial-normalisation sequence is $$ 0 \to \mathcal{O}C \to \nu* \mathcal{O}{\widetilde{C}} \to \mathcal{Q} \to 0, $$ where $Q$ is supported at the nodes of $C$ . At each node $p$ with branches $b_{1}, b_{2}$ , the stalk of $\nu* \mathcal{O}{\widetilde{C}} $a t$ p $i s$ \mathcal{O}{\widetilde{C}, b_1} \times \mathcal{O}_{\widetilde{C}, b_2} $, w hi l e t h es t a l k o f$ \mathcal{O}C $a t$ p $i s t h es u b r in g o f p ai r s w i t h$ f(b_1) = f(b_2) $. T h eco k er n e l a t$ p $i so n e - d im e n s i o na l o v er$ k $, g i v e nb y t h e d i f f er e n ceo f v a l u es$ f(b_1) - f(b_2) $. S o$ \mathcal{Q} = \bigoplus{\text{nodes}} k_p $, a s k y scr a p er o f t o t a l l e n g t h e q u a l t o$ #E(\Gamma_C)$.

Taking Euler characteristics in the short exact sequence (which is additive on $χ$ ), $$ \chi(\mathcal{O}C) = \chi(\nu* \mathcal{O}{\widetilde{C}}) - \chi(\mathcal{Q}) = \chi(\mathcal{O}{\widetilde{C}}) - #E(\Gamma_C), $$ using $χ (ν_{*} O_{C}) = χ (O_{C})$ ( $ν$ is finite, so higher direct images vanish) and $χ (Q) = # E (Γ_{C})$ ( $Q$ is a skyscraper). Since $C = ⨆_{v} C_{v}$ is a disjoint union, $χ (O_{C}) = \sum_{v} χ (O_{C_{v}}) = \sum_{v} (1 - g (C_{v}))$ .

Combining with $p_{a} (C) = 1 - χ (O_{C})$ , $$ p_a(C) = 1 - \sum_v (1 - g(\widetilde{C_v})) + #E(\Gamma_C) = \sum_v g(\widetilde{C_v}) + #E(\Gamma_C) - #V(\Gamma_C) + 1 = \sum_v g(\widetilde{C_v}) + b_1(\Gamma_C). $$

The arithmetic-genus formula. Rubric: full credit for the partial-normalisation sequence, the cokernel identification, and the algebraic manipulation to the dual-graph form.

Exercise 4 (medium, symbolic).

Verify that the dualising sheaf $ω_{C}$ has total degree $2 p_{a} (C) - 2$ on a nodal curve $C$ of arithmetic genus $p_{a} (C) \geq 1$ .

Hint

Use $de g (ω_{C} ∣_{C_{v}}) = 2 g (C_{v}) - 2 + n_{v}$ from Step 1 of the key-theorem proof, and sum over components. Use $\sum_{v} n_{v} = 2# E (Γ_{C})$ (each node contributes two branches).

Answer

Sum the per-component degrees: $$ \deg \omega_C = \sum_v \deg(\omega_C|_{C_v}) = \sum_v (2 g(\widetilde{C_v}) - 2 + n_v) = 2 \sum_v g(\widetilde{C_v}) - 2 #V(\Gamma_C) + \sum_v n_v. $$ Each node contributes two branches (one on each meeting component, or two on the same component for a self-loop), so $\sum_{v} n_{v} = 2# E (Γ_{C})$ . Hence $$ \deg \omega_C = 2 \sum_v g(\widetilde{C_v}) - 2 #V(\Gamma_C) + 2 #E(\Gamma_C) = 2 \big(\sum_v g(\widetilde{C_v}) + b_1(\Gamma_C)\big) - 2 = 2 p_a(C) - 2, $$ matching the smooth-curve formula $de g ω_{C} = 2 g - 2$ . This is the foundational reason Riemann-Roch extends from smooth curves to nodal curves with $ω_{C}$ in place of $Ω_{C}^{1}$ . Rubric: full credit for the summation, the $\sum n_{v} = 2# E$ identity, and the arithmetic-genus formula assembly.

Exercise 5 (medium, short-answer).

Explain why $\overline{M}_{1, 1}$ (the coarse moduli space of one-pointed stable curves of arithmetic genus one) is compact, while the naive un-marked $\overline{M}_{1}$ fails to be separated.

Hint

A genus-one curve without a marked point has the full translation group as automorphisms, which is positive-dimensional. A genus-one curve with one marked point has finite automorphism group (the marked point breaks translation). In moduli-stack language, the un-marked moduli has positive-dimensional stabilisers, while the one-pointed moduli has finite stabilisers.

Answer

A smooth elliptic curve $E$ over $C$ has automorphism group equal to its translation group $E$ as a complex Lie group, which is one-dimensional. Without a marked point, the moduli problem assigns to a family $E \to S$ the set of fibrewise iso-classes, but no representable moduli space exists because the automorphism group is too large: the moduli problem is represented by a non-separated Artin stack, not by a Deligne-Mumford stack.

Marking one point $p \in E$ breaks translation: the automorphism group of $(E, p)$ is the subgroup of translations fixing $p$ , which is the identity. (More precisely, $Aut (E, p) = {\pm 1, \pm i, \pm ζ_{6}, \pm ζ_{6}^{2}}$ for the special $j$ -values, otherwise $Aut (E, p) = {\pm 1}$ — finite in all cases.) The stack $\overline{M}_{1, 1}$ is a Deligne-Mumford stack with finite stabilisers, and its coarse moduli space $\overline{M}_{1, 1} ≅ P^{1}$ (the $j$ -line compactified by a single boundary point — the nodal cubic).

The Knudsen 1983 Math. Scand. construction of pointed stable curves is exactly the mechanism that produces the right compact moduli: marking enough points to break positive-dimensional automorphism groups, then compactifying by adding stable pointed degenerations. For $(g, n) = (1, 1)$ this works; for $(g, n) = (1, 0)$ it does not, and the moduli is not separated.

Rubric: full credit for naming the translation group as the obstruction in $\overline{M}_{1}$ , explaining how marking one point gives finite automorphism group, and identifying $\overline{M}_{1, 1} ≅ P^{1}$ as the resulting compact coarse moduli.

Exercise 6 (medium, numeric).

Compute the dimension of the boundary stratum in $\overline{M}_{4}$ corresponding to a stable curve with dual graph: three vertices, the central vertex of genus 0 connected to each of the two outer vertices (one of genus 1, one of genus 2) by a single edge.

Hint

The central rational vertex has valence 2, which fails the rational-stability condition $n_{v} \geq 3$ . The stratum is therefore empty, or equivalently the dual graph does not arise from a stable curve.

Answer

The stratum has dimension $- \infty$ in the sense that it is empty: the central rational vertex has valence 2, failing the stability condition $n_{v} \geq 3$ for a rational vertex. No stable curve has this dual graph.

To fix the configuration: either the central component must have geometric genus $\geq 1$ (in which case valence 2 is allowed for elliptic, valence 0 allowed for higher), or the rational central component must be replaced by a direct edge between the genus-1 and genus-2 components (a stable curve of two components meeting at one node — this is a point in $Δ_{1}$ of $\overline{M}_{3}$ , not $\overline{M}_{4}$ , since the arithmetic genus would be $1 + 2 + 0 = 3$ , not 4). To get arithmetic genus 4 with a rational central node, the central rational must connect to components of total arithmetic genus 4, with valence $\geq 3$ — e.g., three rational tails attached to a higher-genus core. Rubric: full credit for identifying that the stratum is empty due to the valence-2 rational vertex failing stability.

Exercise 7 (hard, short-answer).

State the Mumford formula $12 λ_{1} = κ_{1} + δ$ in $Pic (\overline{M}_{g}) \otimes Q$ , and explain why it is the moduli analogue of Noether's formula $12 χ (O_{S}) = K_{S}^{2} + e (S)$ for a smooth projective surface $S$ .

Hint

The Mumford formula is obtained by applying Grothendieck-Riemann-Roch to the universal family $π : \overline{C}_{g} \to \overline{M}_{g}$ , with the dualising sheaf $ω_{π}$ as the input class. The Hodge bundle $E = π_{*} ω_{π}$ has first Chern class $λ_{1}$ , and the $κ_{1}$ class is the push-forward $π_{*} c_{1} (ω_{π})^{2}$ .

Answer

Mumford formula. On $\overline{M}_{g}$ for $g \geq 2$ , the rational Picard group $Pic (\overline{M}_{g}) \otimes Q$ contains the Hodge class $λ_{1} = c_{1} (E)$ (first Chern class of the rank- $g$ Hodge bundle $E = π_{*} ω_{π}$ ), the kappa class $κ_{1} = π_{*} (c_{1} (ω_{π})^{2})$ , and the boundary class $δ = [Δ_{0}] + [Δ_{1}] + \dots + [Δ_{⌊ g /2 ⌋}]$ . The Mumford formula is the identity $$ 12 \lambda_1 = \kappa_1 + \delta. $$ The formula is Theorem 5.10 of Mumford's 1983 Towards an enumerative geometry of the moduli space of curves in Arithmetic and Geometry (Birkhäuser).

Analogue to Noether. For a smooth projective surface $S$ , Noether's formula reads $12 χ (O_{S}) = K_{S}^{2} + e (S)$ . Identifications: $χ (O_{S})$ is the holomorphic Euler characteristic, computed by Hirzebruch-Riemann-Roch as $\int_{S} td (T_{S}) = (c_{1}^{2} + c_{2}) /12$ . The Mumford formula is the moduli analogue obtained by viewing $\overline{C}_{g} \to \overline{M}_{g}$ as a family of "surface-like" objects (one-parameter family of curves degenerating along $δ$ ), applying Grothendieck-Riemann-Roch on this family with the dualising sheaf $ω_{π}$ , and matching the resulting identity in $Pic (\overline{M}_{g})_{Q}$ to Noether's surface formula via $λ_{1} \leftrightarrow χ (O)$ , $κ_{1} \leftrightarrow K^{2}$ , $δ \leftrightarrow e$ .

Mumford computed both sides of GRR for the universal family. The right-hand side $π_{*} (td (ω_{π}))$ produces $κ_{1}$ from the second-Chern-class term and $δ$ from the node-correction term (the boundary records where the curve $π^{- 1} (t)$ acquires a node). The left-hand side $ch (π_{!} ω_{π})$ produces $12 λ_{1}$ after twisting by the dualising-sheaf normalisation. The matching is exact, and the Mumford formula is the moduli-stack incarnation of Noether's formula for the universal family.

Rubric: full credit for the Mumford formula as stated, the GRR-on-universal-family derivation strategy, and the explicit Noether analogue identification.

Exercise 8 (hard, short-answer).

Sketch the Gieseker-Mumford 1977 GIT construction of $\overline{M}_{g}$ as a projective coarse moduli space. Identify the linearised reductive-group action, the semistability condition, and the input that makes the GIT quotient agree with the moduli stack $\overline{M}_{g}$ .

Hint

Embed every stable curve in a fixed projective space using a high power of the dualising sheaf: $ω_{C}^{\otimes n}$ for $n \geq 3$ is very ample on every stable curve $C$ (Deligne-Mumford 1969 Theorem 1.2). The Hilbert scheme of such embeddings has a natural action of $PGL (H^{0} (C, ω_{C}^{\otimes n}))$ . GIT quotient by this group gives $\overline{M}_{g}$ .

Answer

Step 1: pluricanonical embedding. For $n \geq 3$ and a stable curve $C$ of arithmetic genus $g \geq 2$ , the line bundle $ω_{C}^{\otimes n}$ has degree $n (2 g - 2)$ and is very ample on every stable curve (Deligne-Mumford 1969 Theorem 1.2; the proof uses ampleness of $ω_{C}$ and a uniform Castelnuovo-Mumford regularity bound). Choose $n$ once and for all (Gieseker took $n = 10$ in his 1982 Tata lectures); then every stable curve embeds in $P^{N}$ for $N = h^{0} (ω_{C}^{\otimes n}) - 1 = n (2 g - 2) - g$ via the complete linear system $∣ ω_{C}^{\otimes n} ∣$ (Riemann-Roch on a nodal curve gives $h^{0} = n (2 g - 2) - g + 1$ for $n \geq 2$ ).

Step 2: Hilbert scheme of pluricanonical embeddings. Let $H = Hilb^{P} (P^{N})$ be the Hilbert scheme of closed subschemes of $P^{N}$ with Hilbert polynomial $P (t) = n (2 g - 2) t - g + 1$ . Inside $H$ there is the locally closed subscheme $H^{stab} \subset H$ of stable curves with their $n$ -canonical embedding. The reductive group $PGL_{N + 1}$ acts on $P^{N}$ , hence on $H$ , and the orbits in $H^{stab}$ are exactly the iso-classes of stable curves.

Step 3: GIT linearisation and semistability. The Plücker embedding of $H$ in a Grassmannian induces a $PGL_{N + 1}$ -linearised ample line bundle $L$ on $H$ . Gieseker 1982 proves that the GIT semistable locus $H^{ss} (L) \subset H$ coincides with the locus of stable curves with their $n$ -canonical embedding, and that strictly semistable points (orbits whose closure meets the unstable locus) do not arise — every semistable point is stable. The Hilbert-Mumford numerical criterion is verified by computing weights of one-parameter subgroups on Hilbert points of pluricanonically embedded stable curves: every destabilising one-parameter subgroup corresponds to a non-stable component of the curve (a rational tail with two attachments, or an elliptic component with no attachment).

Step 4: GIT quotient. The GIT quotient $\overline{M}_{g} = H^{ss} (L) / / PGL_{N + 1}$ is a projective variety of dimension $dim H^{ss} - dim PGL_{N + 1} = (N \cdot (N + 1) + dim Aut) - (N + 1)^{2} + 1 = 3 g - 3$ after the dimension calculation collapses. The quotient agrees with the coarse moduli space of the Deligne-Mumford stack $\overline{M}_{g}$ because both represent the same functor on iso-classes of stable curves.

Step 5: stack vs. coarse moduli. The GIT quotient $\overline{M}_{g}$ is a projective scheme but identifies curves with their automorphisms — it is the coarse moduli, not the stack. The stack $\overline{M}_{g}$ retains the automorphism information at each point (via the quotient $[H^{ss} / PGL_{N + 1}]$ as an Artin stack), and is a Deligne-Mumford stack because stable curves have finite automorphism groups, making stabilisers finite group schemes.

Rubric: full credit for naming the pluricanonical embedding, the Hilbert-scheme setup, the GIT semistability identification with stability, and the stack-vs-coarse-moduli distinction.

Advanced results Master

Theorem (Deligne-Mumford 1969 Theorem 5.2). The moduli stack $\overline{M}_{g}$ of Deligne-Mumford stable curves of arithmetic genus $g \geq 2$ is a proper smooth Deligne-Mumford stack of dimension $3 g - 3$ over $Spec Z$ . Its coarse moduli space $\overline{M}_{g}$ is a projective variety over $Spec Z$ .

The stack $\overline{M}_{g}$ is the étale local quotient of the Hilbert-scheme stratum by $PGL_{N + 1}$ as in Exercise 8, and the smoothness follows from the deformation theory of stable curves: the tangent space at a stable curve $C$ is $Ext^{1} (Ω_{C}^{1}, O_{C})$ , which by Riemann-Roch on the nodal curve has dimension $3 g - 3$ on the nose, with no obstruction (every first-order deformation extends). Properness is the valuative criterion applied to one-parameter families: every family of stable curves over the generic point of a DVR extends uniquely to the closed point as a stable curve, by the stable reduction theorem.

Theorem (stable reduction theorem; Deligne-Mumford 1969 §2). Let $R$ be a discrete valuation ring with fraction field $K$ and residue field $k$ . Every smooth projective curve $C_{K}$ over $K$ of genus $g \geq 2$ admits, after a finite separable extension $K \to K^{'}$ and base change to the integral closure $R^{'}$ of $R$ in $K^{'}$ , a unique extension to a stable curve $C \to Spec R^{'}$ with generic fibre $C_{K^{'}}$ .

The proof is the deep input behind properness of $\overline{M}_{g}$ . Take a regular model of $C_{K}$ over $R^{'}$ (exists by Lipman's resolution of two-dimensional schemes); blow down all $(- 1)$ -curves and $(- 2)$ -curves in the special fibre via Castelnuovo's criterion and the stability check; the resulting special fibre is a stable curve. The extension of base $K \to K^{'}$ is needed to make the regular model exist and to eliminate wild Galois actions on the special fibre. Stable reduction is the curve-moduli analogue of semistable reduction for one-parameter families in the minimal model program.

Theorem (Knudsen 1983 §1; pointed stable curves). For $g \geq 0$ and $n \geq 0$ with $2 g - 2 + n > 0$ , there is a proper smooth Deligne-Mumford stack $\overline{M}_{g, n}$ parametrising stable $n$ -pointed curves of arithmetic genus $g$ . Its dimension is $3 g - 3 + n$ . The stability condition for $(C, p_{1}, \dots, p_{n})$ is that the marked points are distinct, lie in the smooth locus of $C$ , and that the dualising sheaf twisted by the marked points, $ω_{C} (p_{1} + \dots + p_{n})$ , is ample on $C$ .

Knudsen's 1983 Math. Scand. construction adjoins marked points to the Deligne-Mumford framework, and gives the contraction and stabilisation morphisms $\overline{M}_{g, n + 1} \to \overline{M}_{g, n}$ that forget the $(n + 1)$ -th marked point and contract any newly-unstable components. The genus-zero case $\overline{M}_{0, n}$ requires $n \geq 3$ (the dimension $3 \cdot 0 - 3 + n = n - 3$ is non-negative iff $n \geq 3$ ), and the genus-one case $\overline{M}_{1, n}$ requires $n \geq 1$ . The case $\overline{M}_{1, 1}$ — one-pointed stable curves of arithmetic genus one — is the modular curve $X (1) ≅ P^{1}$ over the $j$ -line, with the boundary point at $j = \infty$ corresponding to the nodal cubic.

Theorem (Mumford 1983 Theorem 5.10). On $\overline{M}_{g}$ for $g \geq 2$ , the identity $$ 12 \lambda_1 = \kappa_1 + \delta $$ holds in $Pic (\overline{M}_{g}) \otimes Q$ , where $λ_{1} = c_{1} (E)$ is the first Hodge class, $\kappa_1 = \pi_(c_1(\omega_\pi)^2) $i s t h e f i r s t k a pp a c l a ss, an d$ \delta$ is the total boundary class.*

The Mumford formula is obtained by applying Grothendieck-Riemann-Roch to the universal family $π : \overline{C}_{g} \to \overline{M}_{g}$ with input class $[O_{\overline{C}_{g}}]$ . The right-hand side is $π_{*} td (T_{π})$ where $T_{π}$ is the relative tangent sheaf (the dual of $ω_{π}$ ), expanded in characteristic classes. The boundary contribution arises because $ω_{π}$ has nodes — the relative tangent sheaf is not locally free at the boundary, and the Todd class picks up the boundary classes $δ_{i}$ as correction terms. The left-hand side is $ch (π_{!} O) = ch (π_{*} O) - ch (R^{1} π_{*} O) = 1 - ch (E^{\lor}) = 1 - g + λ_{1} + O (λ_{1}^{2})$ , using $R^{1} π_{*} O_{π} = E^{\lor}$ (dual of the Hodge bundle) by Serre duality on the fibres. Equating degree-one parts gives $12 λ_{1} = κ_{1} + δ$ .

Theorem (Faber's conjectures; partial proofs). The tautological ring $R^(\overline{\mathcal{M}}_g) \subset H^(\overline{\mathcal{M}}_g, \mathbb{Q}) $i s t h es u b r in g g e n er a t e d b y t h e$ \kappa $- c l a sses,$ \lambda $- c l a sses, an d b o u n d a r y c l a sses$ \delta_i $. F ab er 1999 co nj ec t u r e d t ha t$ R^(\mathcal{M}_g) $i s a G or e n s t e in r in g w i t h soc l e in d e g r ee$ g - 2 $, w i t h e x pl i c i t d im e n s i o n f or m u l a s f or e a c h g r a d e d p i ece, an d t ha tt h e p r o p or t i o na l i t y r e l a t i o n s am o n g m o n o mia l s in$ \kappa $- c l a sses mi r r or t h e in t er sec t i o n - n u mb er r e l a t i o n so n$ \overline{\mathcal{M}}_g$ via Witten's conjecture and the Virasoro constraints.*

Major partial results include Looijenga 1995 (vanishing of $R^{i} (M_{g})$ for $i > g - 2$ ), Graber-Vakil 2005 (the socle statement modulo the perfect-pairing part), Buryak-Shadrin-Spitz-Zvonkine 2014 (parts of the explicit Gorenstein structure), and Pixton's relations (2012) for the tautological ring on $\overline{M}_{g, n}$ . The conjecture remains open in full generality; the tautological-ring framework is the central organising principle of modern enumerative geometry of curve moduli.

Theorem (Cornalba-Harris 1988 Ann. Sci. ENS 21). On $\overline{M}_{g}$ for $g \geq 2$ , the divisor class $aλ - b δ - \sum_{i} c_{i} δ_{i}$ is ample on the moduli space iff a system of explicit inequalities among the coefficients holds — the Cornalba-Harris slope inequalities. The slope of a family of stable curves over a base curve is bounded below by $4 + 12/ g - 4 (\sum c_{i} δ_{i}) / (δ)$ , with equality for certain extremal families.

The slope inequality is the foundational tool for showing that $\overline{M}_{g}$ is of general type for $g \geq 24$ (Harris-Mumford 1982 Invent. Math. 67, refined by Eisenbud-Harris 1987 Invent. Math. 90) and rationally connected / unirational for $g \leq 14$ (Severi for $g \leq 10$ , Sernesi for $g = 12$ , Chang-Ran for $g = 11, 13$ , Verra for $g = 14$ ). The Kodaira dimension of $\overline{M}_{g}$ — geometric or general type — is governed by the slope of effective divisors, computed via Cornalba-Harris.

Theorem (Kollár-Shepherd-Barron 1988; Alexeev 2002; KSBA stability of higher-dimensional stable pairs). Let $(X, Δ)$ be a projective log canonical pair of dimension $n$ with $K_{X} + Δ$ ample. The moduli of KSBA-stable pairs of fixed dimension, volume, and coefficient set is a proper Deligne-Mumford stack with projective coarse moduli space.

The KSBA framework is the higher-dimensional generalisation of Deligne-Mumford stability: nodal curves become semi-log-canonical varieties, the dualising sheaf $ω_{C}$ becomes $ω_{X} (Δ)$ , and the ampleness condition $ω_{C}$ ample becomes $K_{X} + Δ$ ample. Kollár-Shepherd-Barron 1988 Invent. Math. 91 introduced the framework for surfaces; Alexeev 2002 Ann. of Math. 155 proved the existence of moduli for stable abelian and toric pairs; Kollár's 2023 book Families of Varieties of General Type gives the modern treatment with the Cambridge-Springer culmination of the KSBA programme. KSBA moduli appears again in 04.10.13 (K-stability), where K-stability is the Fano analogue of KSBA stability for $K_{X} + Δ$ anti-ample.

Synthesis. The Deligne-Mumford stability condition is the foundational reason the moduli of curves admits a compactification with the right algebro-geometric properties, and the central insight is that combinatorial stability (rational components meet the rest in $\geq 3$ points; elliptic components meet the rest in $\geq 1$ point) coincides exactly with the algebraic condition that the dualising sheaf is ample, which in turn coincides with finiteness of the automorphism group. Three apparently distinct criteria — a graph-combinatorial one, an algebraic one on a line bundle, and a representation-theoretic one on a finite group scheme — identify the same class of nodal curves. Putting these together, the moduli stack $\overline{M}_{g}$ is the proper smooth Deligne-Mumford stack of dimension $3 g - 3$ that compactifies $M_{g}$ , and the bridge is the dualising sheaf: combinatorial stability is exactly ampleness of $ω_{C}$ , ampleness of $ω_{C}$ gives the GIT framework for constructing $\overline{M}_{g}$ as a projective coarse moduli, and finiteness of $Aut (C)$ makes the stack Deligne-Mumford rather than merely Artin.

The construction generalises in two directions. Adding marked points via Knudsen 1983 gives $\overline{M}_{g, n}$ , the moduli of pointed stable curves, with the stability condition becoming ampleness of $ω_{C} (p_{1} + \dots + p_{n})$ and the dimension $3 g - 3 + n$ . This is exactly the framework needed for genus-zero moduli ( $\overline{M}_{0, n}$ for $n \geq 3$ , the configuration spaces of points on $P^{1}$ modulo $PGL_{2}$ ) and for one-pointed genus-one ( $\overline{M}_{1, 1} ≅ P^{1}$ over the $j$ -line). To higher dimensions, the KSBA programme of Kollár-Shepherd-Barron 1988 and Alexeev 2002 generalises: nodal curves become semi-log-canonical varieties, the dualising sheaf $ω_{C}$ becomes the log-canonical bundle $ω_{X} (Δ)$ , and ampleness becomes $K_{X} + Δ$ ample. This bridge appears again in 04.10.01 (the foundational dimension count $3 g - 3$ for smooth moduli is now the deformation-theoretic dimension at a stable curve), in 04.10.02 (the GIT construction of $\overline{M}_{g}$ as a projective coarse moduli is the surface-Hilbert-scheme adaptation of GIT to pluricanonically embedded stable curves), and in 04.10.13 (K-stability is the Fano analogue of KSBA stability with $K_{X} + Δ$ anti-ample, identifying algebraic stability with existence of Kähler-Einstein metrics via Yau-Tian-Donaldson).

The Mumford formula $12 λ_{1} = κ_{1} + δ$ is the moduli analogue of Noether's formula $12 χ (O_{S}) = K_{S}^{2} + e (S)$ for a surface, and is dual to the Hirzebruch-Riemann-Roch identity for the universal family. The synthesis is structural: every classical enumerative computation on the moduli of curves — Witten's conjecture and the Kontsevich theorem on intersections of $ψ$ -classes, the Eisenbud-Harris-Mumford computation of the Kodaira dimension of $\overline{M}_{g}$ , the Cornalba-Harris slope inequality, the Faber-conjecture programme on the tautological ring — runs through the same dualising-sheaf framework. Deligne-Mumford stability identifies which curves to include in the moduli, the dualising sheaf gives the ample polarisation that makes the moduli a projective scheme, the tautological classes $λ, κ, δ, ψ$ generate the cohomology ring on which intersection-theoretic answers live, and the Mumford formula couples them via GRR on the universal family. The bridge is universal: from curves to higher-dimensional pairs, from stack to coarse moduli, from combinatorial valence to algebraic ampleness.

Full proof set Master

Proposition (arithmetic-genus formula). Let $C$ be a connected projective nodal curve over $k$ with dual graph $Γ_{C}$ . Then $$ p_a(C) = \sum_{v \in V(\Gamma_C)} g(\widetilde{C_v}) + b_1(\Gamma_C). $$

Proof. The partial-normalisation map $ν : C \to C$ is finite and birational, with $C = ⨆_{v} C_{v}$ a disjoint union of smooth projective curves indexed by the irreducible components of $C$ . The short exact sequence of sheaves on $C$ , $$ 0 \to \mathcal{O}C \to \nu* \mathcal{O}{\widetilde{C}} \to \bigoplus{p \in \mathrm{Sing}(C)} k_p \to 0, $$ has cokernel a skyscraper of length $# E (Γ_{C})$ supported at the nodes, with one copy of $k$ per node recording the difference of values on the two branches. Taking Euler characteristics (additive on short exact sequences of coherent sheaves), using $χ (ν_{*} O_{C}) = χ (O_{C})$ (since $ν$ is finite, $R^{i} ν_{*} = 0$ for $i > 0$ ), and using $χ (O_{C}) = \sum_{v} χ (O_{C_{v}}) = \sum_{v} (1 - g (C_{v}))$ , $$ \chi(\mathcal{O}_C) = \sum_v (1 - g(\widetilde{C_v})) - #E(\Gamma_C) = #V(\Gamma_C) - \sum_v g(\widetilde{C_v}) - #E(\Gamma_C). $$ Since $p_{a} (C) = 1 - χ (O_{C})$ for a connected curve, $$ p_a(C) = 1 - #V(\Gamma_C) + \sum_v g(\widetilde{C_v}) + #E(\Gamma_C) = \sum_v g(\widetilde{C_v}) + b_1(\Gamma_C), $$ where $b_{1} (Γ_{C}) = # E (Γ_{C}) - # V (Γ_{C}) + 1$ is the first Betti number of the connected dual graph. $□$

Proposition (degree of the dualising sheaf). On a connected projective nodal curve $C$ with dual graph $Γ_{C}$ , $$ \deg \omega_C = 2 p_a(C) - 2. $$

Proof. Sum the per-component degree $de g (ω_{C} ∣_{C_{v}}) = 2 g (C_{v}) - 2 + n_{v}$ (where $n_{v}$ is the valence of vertex $v$ , equal to the number of node-branches lying on $C_{v}$ ) over all components: $$ \deg \omega_C = \sum_v (2 g(\widetilde{C_v}) - 2 + n_v) = 2 \sum_v g(\widetilde{C_v}) - 2 #V(\Gamma_C) + 2 #E(\Gamma_C), $$ using $\sum_{v} n_{v} = 2# E (Γ_{C})$ (each node contributes two branches). Substituting $b_{1} (Γ_{C}) = # E (Γ_{C}) - # V (Γ_{C}) + 1$ , $$ \deg \omega_C = 2 \sum_v g(\widetilde{C_v}) + 2 b_1(\Gamma_C) - 2 = 2 p_a(C) - 2. \qquad \square $$

Proposition (equivalence of stability characterisations). For a connected projective nodal curve $C$ of arithmetic genus $g \geq 2$ , the following are equivalent: (i) combinatorial Deligne-Mumford stability; (ii) ampleness of $ω_{C}$ ; (iii) finiteness of $Aut (C)$ .

Proof. Given in the Key theorem section. The argument runs (i) $⟺$ (ii) via per-component degree positivity of $ω_{C}$ , using the formula $de g (ω_{C} ∣_{C_{v}}) = 2 g (C_{v}) - 2 + n_{v}$ and the case analysis on $g (C_{v}) \in {0, 1, \geq 2}$ ; (ii) $\Rightarrow$ (iii) via the Hilbert-scheme argument that $Aut (C, ω_{C}) \subset PGL (H^{0} (ω_{C}^{\otimes n}))$ for $n ≫ 0$ is a closed subgroup-scheme of the algebraic group $PGL_{N}$ , with $Aut (C) = Aut (C, ω_{C})$ by canonicity of the dualising sheaf; (iii) $\Rightarrow$ (i) by classifying positive-dimensional automorphism groups of nodal-curve components — $P^{1}$ with $\leq 2$ fixed branches has $PGL_{2}$ -stabiliser of positive dimension, elliptic curve with $0$ fixed branches has full translation group of positive dimension. $□$

Proposition (irreducibility of $M_{g}$ ; Deligne-Mumford 1969 Theorem 5.1). The moduli stack $M_{g}$ of smooth projective curves of genus $g \geq 2$ is geometrically irreducible.

Proof sketch (full proof in Deligne-Mumford 1969 §5). Reduce to characteristic zero via the smoothness of $\overline{M}_{g} \to Spec Z$ . In characteristic zero, use the Hurwitz space of degree- $d$ covers $C \to P^{1}$ with simple branching — the Hurwitz space is irreducible (Clebsch 1873, completed by Severi and modernised by Hurwitz himself), and the forgetful map Hurwitz space $\to M_{g}$ is surjective on dimensions for $d \geq 2 g - 1$ . Irreducibility of the source plus surjectivity of the map gives irreducibility of the target. The Deligne-Mumford 1969 contribution was to extend this to mixed characteristic by reduction modulo $p$ using the stack $\overline{M}_{g}$ — the same proof carries through because the boundary $δ$ has codimension one and does not disconnect the moduli. $□$

Proposition (properness via stable reduction). The stack $\overline{M}_{g}$ satisfies the valuative criterion of properness over $Spec Z$ .

Proof sketch (full proof in Deligne-Mumford 1969 §2). Given a DVR $R$ with fraction field $K$ , residue field $k$ , and a stable curve $C_{K}$ over $K$ , the stable reduction theorem extends $C_{K}$ uniquely (after a finite separable base extension of $K$ ) to a stable curve $C$ over $R^{'}$ with $C ∣_{K} = C_{K}$ . Existence: take a regular model $C$ over $R^{'}$ (exists by Lipman's resolution of two-dimensional excellent schemes), then contract all $(- 1)$ -curves and any rational $(- 2)$ -curves in the special fibre via Castelnuovo's criterion until the special fibre has no contractible curves. The resulting special fibre is a stable curve, and uniqueness follows from the rigidity of stable curves under further blow-downs. The finite base extension is needed to eliminate Galois twists; in characteristic zero this is automatic, in mixed characteristic it eliminates wild Galois actions. $□$

Proposition (Mumford formula), stated without proof here — full proof in Mumford 1983 Arithmetic and Geometry vol. II §5 ^{[source pending]}. Apply Grothendieck-Riemann-Roch to the universal family $π : \overline{C}_{g} \to \overline{M}_{g}$ with input class $[O_{\overline{C}_{g}}]$ . The left side expands as $ch (π_{!} O) = ch (O_{\overline{M}_{g}}) - ch (E^{\lor}) = 1 - g + λ_{1} - \frac{1}{2} (λ_{1}^{2} - 2 λ_{2}) + \dots$ where $E = π_{*} ω_{π}$ is the Hodge bundle and $λ_{i} = c_{i} (E)$ . The right side expands as $π_{*} (td (ω_{π}^{\lor}))$ with the relative tangent class $T_{π} = ω_{π}^{\lor}$ ; the Todd expansion produces $κ_{1}$ from the $c_{1}^{2}$ -term, $κ_{0} = 2 g - 2$ from the $c_{1}$ -term, and the boundary classes $δ_{i}$ from the node-correction terms (the relative tangent sheaf is not locally free at nodes, requiring a correction). Equating degree-one parts of both sides gives $12 λ_{1} = κ_{1} + δ$ .

Proposition (Cornalba-Harris slope inequality), stated without proof here — full proof in Cornalba-Harris 1988 Ann. Sci. ENS 21 ^{[source pending]}. For a one-parameter family $f : X \to B$ of stable curves of genus $g$ over a smooth projective curve $B$ , the slope $s (f) = (12 de g f_{*} ω_{f} - de g δ_{f}) / de g δ_{f}$ satisfies $s (f) \geq (8 g + 4) / (g + 1)$ for non-isotrivial families, with equality for the Cornalba-Harris extremal family. The proof runs via a careful study of effective divisor classes on $\overline{M}_{g}$ pulled back to $B$ .

Connections Master

Moduli of curves 04.10.01. Deligne-Mumford stability is exactly the compactification of $M_{g}$ , with $\overline{M}_{g}$ adding stable curves on the boundary. The dimension $3 g - 3$ is shared by $M_{g}$ and $\overline{M}_{g}$ ; the smooth locus is open and dense, the boundary is a divisor decomposing as $δ = δ_{0} + δ_{1} + \dots + δ_{⌊ g /2 ⌋}$ . The compactification appears again in 04.10.01 as the proper Deligne-Mumford stack underlying the moduli problem.
Geometric invariant theory 04.10.02. The GIT construction of $\overline{M}_{g}$ as a projective coarse moduli space (Mumford-Gieseker 1965-1982) takes pluricanonically embedded stable curves and quotients by $PGL_{N + 1}$ . The semistability condition in GIT coincides with Deligne-Mumford stability of the underlying curve, with the dualising sheaf supplying the equivariant ample polarisation. The bridge to GIT is the dualising-sheaf-ample characterisation of stability.
Canonical sheaf 04.08.02. The dualising sheaf $ω_{C}$ of a smooth curve is the canonical sheaf $Ω_{C}^{1}$ . On a nodal curve, $ω_{C}$ extends $Ω_{C}^{1}$ to allow simple poles at branches of nodes with residues summing to zero, restoring invertibility. This extension is exactly what makes the algebraic stability condition (ampleness of $ω_{C}$ ) coincide with combinatorial stability.
Riemann-Roch for curves 04.04.01. Riemann-Roch on a nodal curve takes the same form as on a smooth curve, with $ω_{C}$ in place of $Ω_{C}^{1}$ . The arithmetic genus formula $p_{a} (C) = \sum_{v} g (C_{v}) + b_{1} (Γ_{C})$ is the combinatorial input that makes Riemann-Roch on the dual graph match Riemann-Roch on each normalised component, with the loop correction coming from the partial-normalisation sequence.
Hilbert scheme 04.10.05. The Gieseker-Mumford GIT construction of $\overline{M}_{g}$ uses the Hilbert scheme of pluricanonically embedded subschemes of $P^{N}$ with the Hilbert polynomial of a stable curve. The Hilbert scheme is the parameter space; the GIT quotient by $PGL_{N + 1}$ extracts the moduli.
K-stability and Yau-Tian-Donaldson 04.10.13. KSBA stability of higher-dimensional pairs (Kollár-Shepherd-Barron 1988; Alexeev 2002) is the generalisation of Deligne-Mumford stability from curves to higher dimensions, with the dualising sheaf condition $ω_{C}$ ample becoming $K_{X} + Δ$ ample. K-stability of Fano varieties is the dual notion, with $K_{X} + Δ$ anti-ample, identifying algebraic stability with existence of Kähler-Einstein metrics via the Yau-Tian-Donaldson correspondence.
Gieseker stability and moduli of sheaves 04.10.11. Gieseker's 1977 stability for sheaves on a projective variety is the higher-rank analogue of Deligne-Mumford stability of curves, with the Hilbert polynomial of a coherent sheaf playing the role of the arithmetic genus and the slope condition replacing the combinatorial valence condition. The GIT construction of the moduli of stable sheaves parallels the GIT construction of $\overline{M}_{g}$ via pluricanonical embedding.
Adjunction formula on a surface 04.05.07. The Mumford formula $12 λ_{1} = κ_{1} + δ$ on $\overline{M}_{g}$ is the moduli analogue of Noether's formula $12 χ (O_{S}) = K_{S}^{2} + e (S)$ on a smooth projective surface, with the Hodge class playing the role of the Euler characteristic, the kappa class playing the role of the canonical square, and the boundary class playing the role of the topological Euler characteristic of the surface.

Historical & philosophical context Master

Riemann counted the moduli of smooth genus- $g$ curves in Theorie der Abelschen Functionen (Crelle 54, 1857) ^{[source pending]}, obtaining the dimension $3 g - 3$ for $g \geq 2$ as the parameter count for the Abelian integrals on a generic genus- $g$ Riemann surface. The compactification of the moduli, with degenerate curves admitted, was implicit in nineteenth-century work on the parameter space of plane curves with prescribed singularities (Cayley, Plücker, Severi), and made explicit only with the rise of scheme-theoretic algebraic geometry in the mid-twentieth century.

David Mumford's 1965 Geometric Invariant Theory (Springer Ergebnisse 34, with later editions by Mumford-Fogarty-Kirwan) ^{[source pending]} gave the first rigorous construction of the smooth-curve moduli $M_{g}$ as a quasi-projective variety via GIT, using pluricanonically embedded curves in projective space and the quotient by $PGL_{N + 1}$ . The compactification was opened up by Pierre Deligne and David Mumford in The irreducibility of the space of curves of given genus (Inst. Hautes Études Sci. Publ. Math. 36, 1969, 75-109) ^{[source pending]}, where they introduced the notion of a stable curve, proved the equivalence with ampleness of the dualising sheaf, constructed the moduli stack $\overline{M}_{g}$ as a proper smooth Deligne-Mumford stack, and used the new compactification to deduce irreducibility of $M_{g}$ in arbitrary characteristic by reduction modulo $p$ . The Deligne-Mumford paper also introduced what is now called a Deligne-Mumford stack as the appropriate categorical setting for moduli problems with finite automorphism groups.

David Gieseker's 1982 Tata Institute lectures Lectures on Moduli of Curves (LNS 69) ^{[source pending]} worked out the GIT details for the compactified moduli, identifying the GIT semistable locus with the locus of pluricanonically embedded stable curves and showing that the GIT quotient is the projective coarse moduli $\overline{M}_{g}$ . Finn Knudsen's 1983 papers The projectivity of the moduli space of stable curves II, III (Math. Scand. 52) ^{[source pending]} introduced pointed stable curves $\overline{M}_{g, n}$ with the contraction and stabilisation morphisms, completing the modern framework. Mumford's Towards an enumerative geometry of the moduli space of curves in Arithmetic and Geometry vol. II (Birkhäuser 1983) ^{[source pending]} introduced the tautological classes $λ$ and $κ$ , proved the Mumford formula $12 λ_{1} = κ_{1} + δ$ via Grothendieck-Riemann-Roch on the universal family, and launched the enumerative-geometry programme on $\overline{M}_{g}$ that culminated in Witten's conjecture (Kontsevich theorem, 1992), Faber's conjectures on the tautological ring (1999, partial), and the Eisenbud-Harris-Mumford classification of the Kodaira dimension of $\overline{M}_{g}$ .

Maurizio Cornalba and Joe Harris's Divisor classes associated to families of stable varieties (Ann. Sci. ENS 21, 1988, 455-475) ^{[source pending]} gave the slope inequality on $\overline{M}_{g}$ and identified the ample cone, which together with Eisenbud-Harris 1987 Invent. Math. 90 led to the Harris-Mumford 1982 Invent. Math. 67 theorem that $\overline{M}_{g}$ is of general type for $g \geq 24$ . The higher-dimensional generalisation — KSBA stability of log canonical pairs $(X, Δ)$ with $K_{X} + Δ$ ample — was introduced by Janos Kollár and Nicholas Shepherd-Barron in Threefolds and deformations of surface singularities (Invent. Math. 91, 1988) and developed by Valery Alexeev in Complete moduli in the presence of semiabelian group action (Ann. of Math. 155, 2002, 611-708) ^{[source pending]}. The KSBA programme is the higher-dimensional analogue of Deligne-Mumford, with semi-log-canonical varieties replacing nodal curves and $K_{X} + Δ$ ample replacing $ω_{C}$ ample, and is the framework for the modern moduli theory of varieties of general type.

Bibliography Master

@article{DeligneMumford1969,
  author    = {Deligne, Pierre and Mumford, David},
  title     = {The irreducibility of the space of curves of given genus},
  journal   = {Publications Math{\'e}matiques de l'IH{\'E}S},
  volume    = {36},
  year      = {1969},
  pages     = {75--109}
}

@book{HarrisMorrison,
  author    = {Harris, Joe and Morrison, Ian},
  title     = {Moduli of Curves},
  publisher = {Springer-Verlag},
  series    = {Graduate Texts in Mathematics},
  volume    = {187},
  year      = {1998}
}

@article{KnudsenII,
  author    = {Knudsen, Finn F.},
  title     = {The projectivity of the moduli space of stable curves II: the stacks $M_{g,n}$},
  journal   = {Mathematica Scandinavica},
  volume    = {52},
  year      = {1983},
  pages     = {161--199}
}

@article{KnudsenIII,
  author    = {Knudsen, Finn F.},
  title     = {The projectivity of the moduli space of stable curves III: the line bundles on $M_{g,n}$, and a proof of the projectivity of $\overline{M}_{g,n}$ in characteristic 0},
  journal   = {Mathematica Scandinavica},
  volume    = {52},
  year      = {1983},
  pages     = {200--212}
}

@book{MumfordGIT,
  author    = {Mumford, David and Fogarty, John and Kirwan, Frances},
  title     = {Geometric Invariant Theory},
  publisher = {Springer-Verlag},
  series    = {Ergebnisse der Mathematik und ihrer Grenzgebiete},
  volume    = {34},
  edition   = {3},
  year      = {1994}
}

@incollection{Mumford1983,
  author    = {Mumford, David},
  title     = {Towards an enumerative geometry of the moduli space of curves},
  booktitle = {Arithmetic and Geometry, Vol. II},
  editor    = {Artin, Michael and Tate, John},
  publisher = {Birkh{\"a}user},
  series    = {Progress in Mathematics},
  volume    = {36},
  year      = {1983},
  pages     = {271--328}
}

@book{Gieseker1982,
  author    = {Gieseker, David},
  title     = {Lectures on Moduli of Curves},
  publisher = {Tata Institute of Fundamental Research / Springer-Verlag},
  series    = {Tata Institute Lecture Notes},
  volume    = {69},
  year      = {1982}
}

@article{CornalbaHarris1988,
  author    = {Cornalba, Maurizio and Harris, Joe},
  title     = {Divisor classes associated to families of stable varieties, with applications to the moduli space of curves},
  journal   = {Annales Scientifiques de l'{\'E}cole Normale Sup{\'e}rieure (4)},
  volume    = {21},
  year      = {1988},
  pages     = {455--475}
}

@article{KollárShepherdBarron1988,
  author    = {Koll{\'a}r, J{\'a}nos and Shepherd-Barron, Nicholas I.},
  title     = {Threefolds and deformations of surface singularities},
  journal   = {Inventiones Mathematicae},
  volume    = {91},
  year      = {1988},
  pages     = {299--338}
}

@article{Alexeev2002,
  author    = {Alexeev, Valery},
  title     = {Complete moduli in the presence of semiabelian group action},
  journal   = {Annals of Mathematics (2)},
  volume    = {155},
  year      = {2002},
  pages     = {611--708}
}

@article{HarrisMumford1982,
  author    = {Harris, Joe and Mumford, David},
  title     = {On the Kodaira dimension of the moduli space of curves},
  journal   = {Inventiones Mathematicae},
  volume    = {67},
  year      = {1982},
  pages     = {23--88}
}

@article{Riemann1857,
  author    = {Riemann, Bernhard},
  title     = {Theorie der Abel'schen Functionen},
  journal   = {Journal f{\"u}r die reine und angewandte Mathematik (Crelle)},
  volume    = {54},
  year      = {1857},
  pages     = {115--155}
}

@article{Kontsevich1992,
  author    = {Kontsevich, Maxim},
  title     = {Intersection theory on the moduli space of curves and the matrix Airy function},
  journal   = {Communications in Mathematical Physics},
  volume    = {147},
  year      = {1992},
  pages     = {1--23}
}

@article{Faber1999,
  author    = {Faber, Carel},
  title     = {A conjectural description of the tautological ring of the moduli space of curves},
  booktitle = {Moduli of Curves and Abelian Varieties},
  publisher = {Vieweg},
  series    = {Aspects of Mathematics},
  volume    = {E33},
  year      = {1999},
  pages     = {109--129}
}

Prerequisites

04.10.01
04.04.01
04.08.02
04.10.02

Tier anchors

beginner: Harris-Morrison *Moduli of Curves* Ch. 1 introduction; Vakil notes on stable curves
intermediate: Harris-Morrison *Moduli of Curves* Ch. 1, 4; Deligne-Mumford 1969 *Publ. Math. IHES* 36
master: Harris-Morrison Ch. 2, 3, 6; Mumford 1983 *Towards an enumerative geometry...* (Arithmetic and Geometry vol. II, Birkhäuser); Arbarello-Cornalba-Griffiths *Geometry of Algebraic Curves* Vol. II; Cornalba-Harris 1988 *Ann. Sci. ENS* 21

References

Deligne-Mumford — The irreducibility of the space of curves of given genus · Publ. Math. IHES 36 (1969), 75-109; the originator paper for stable curves, the moduli stack $\overline{\mathcal{M}}_g$, and the dualising-sheaf characterisation
Harris-Morrison — Moduli of Curves · Springer GTM 187, 1998; Ch. 1 (definitions, boundary, 7 topological types in genus 2) and Ch. 4 (tautological classes, Mumford formula)
Knudsen — The projectivity of the moduli space of stable curves II, III · Math. Scand. 52 (1983), 161-199 and 200-212; pointed stable curves $\overline{M}_{g,n}$, contraction and stabilisation morphisms
Mumford — Geometric Invariant Theory · Springer Ergebnisse 34, 1965 (3rd ed. with Fogarty-Kirwan 1994); GIT framework underlying the construction of $\overline{M}_g$ as a projective coarse moduli space
Mumford — Towards an enumerative geometry of the moduli space of curves · Arithmetic and Geometry, vol. II, Birkhäuser 1983, 271-328; the $\lambda$- and $\kappa$-classes, the Mumford formula $12\lambda = \kappa_1 + \delta$, and the tautological ring
Gieseker — Lectures on moduli of curves · Tata Institute Lecture Notes 69, 1982; the GIT construction of $\overline{M}_g$ via Hilbert points of pluricanonical embeddings
Cornalba-Harris — Divisor classes associated to families of stable varieties · Ann. Sci. ENS (4) 21 (1988), 455-475; the slope inequality and ample divisor classes on $\overline{M}_g$
Kollár-Shepherd-Barron — Threefolds and deformations of surface singularities · Invent. Math. 91 (1988), 299-338; together with Alexeev *Ann. of Math.* 155 (2002), 611-708 — moduli of stable pairs in higher dimension

Estimated time

beginner: 18m
intermediate: 45m
master: 80m