04.10.01 · algebraic-geometry / moduli

Moduli of curves

shipped3 tiersLean: partial

Anchor (Master): Harris-Morrison; Mumford-Fogarty-Kirwan *GIT*; Arbarello-Cornalba-Griffiths *Geometry of Algebraic Curves* Vol II

Intuition [Beginner]

The moduli space of curves $M_{g}$ is a remarkable mathematical object: a single space whose points are themselves entire algebraic curves. Each point of $M_{g}$ corresponds to one isomorphism class of smooth projective curves of genus $g$ . So a path in $M_{g}$ is a continuous deformation of curves, a divisor in $M_{g}$ is a family of curves with some special property, and so on.

Riemann introduced the idea in 1857 in Theorie der Abelschen Functionen: he counted the moduli — the parameters needed to specify a curve of genus $g$ — and got the formula

dim M_{g} = 3 g - 3 for g \geq 2.

For $g = 0$ there are no moduli (every genus-0 curve is the projective line). For $g = 1$ there is one modulus (the j-invariant of an elliptic curve). For $g \geq 2$ , Riemann's $3 g - 3$ .

David Mumford in 1965 gave the first rigorous construction of $M_{g}$ as a quasi-projective scheme using Geometric Invariant Theory (GIT) — a Hilbert-style construction: parameterise embedded curves with abundant linear data, then quotient by the group of changes of basis. Deligne-Mumford in 1969 extended to a compactification $\overline{M_{g}}$ allowing curves with mild singularities.

Visual [Beginner]

A space whose points are smooth projective curves of genus $g$ , with paths corresponding to continuous deformations.

Worked example [Beginner]

For $g = 1$ (elliptic curves): the moduli space $M_{1}$ has dimension $3 \cdot 1 - 3 = 0$ via the naive Riemann count, but it actually has dimension 1 — the formula $3 g - 3$ requires $g \geq 2$ .

The moduli space $M_{1}$ is the modular curve: every elliptic curve $E$ over $C$ corresponds to a point in the upper-half plane $H = {τ \in C : Im (τ) > 0}$ modulo the action of the modular group $SL_{2} (Z)$ . Two elliptic curves are isomorphic iff their lattice parameters differ by a Möbius transformation in $SL_{2} (Z)$ .

The quotient $H / SL_{2} (Z)$ is identified with the affine line $A^{1}$ via the j-invariant — a specific rational function on $H$ that descends to a coordinate on the quotient. So $M_{1} ≅ A_{C}^{1}$ , with the j-invariant as the universal coordinate. Each value of $j \in C$ corresponds to one elliptic curve up to isomorphism.

For $g = 2$ : $dim M_{2} = 3 \cdot 2 - 3 = 3$ . The moduli space of genus-2 curves is 3-dimensional, parametrising the position of branch points of the unique 2-to-1 map to $P^{1}$ .

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $g \geq 0$ be an integer.

Coarse moduli space (Mumford 1965). The moduli space of smooth projective genus- $g$ curves $M_{g}$ is the coarse moduli space representing the moduli functor

M_{g} (T) := {flat families of smooth projective genus- g curves over T} / \sim,

where $T$ is a scheme and $\sim$ denotes isomorphism. Mumford 1965 (Geometric Invariant Theory) proves that $M_{g}$ is a quasi-projective scheme over $Z$ of dimension $3 g - 3$ (for $g \geq 2$ ).

Caveat: not a fine moduli space. $M_{g}$ is not a fine moduli space (no universal family on $M_{g}$ representing the functor) because curves can have nonzero (non-identity) automorphisms. The Deligne-Mumford stack $M_{g}^{stack}$ is the right object — a Deligne-Mumford stack representing the moduli functor exactly.

Moduli with marked points. $M_{g, n}$ is the moduli space of smooth projective genus- $g$ curves with $n$ ordered marked points $p_{1}, \dots, p_{n}$ , all distinct. $dim M_{g, n} = 3 g - 3 + n$ for $2 g - 2 + n > 0$ .

Construction via GIT. The construction (Mumford 1965):

(M1) Tri-canonical embedding. For $g \geq 2$ , every smooth projective genus- $g$ curve $C$ admits a closed embedding $C ↪ P^{N}$ via the linear system $∣3 K_{C} ∣$ — tri-canonical sections. This embeds $C$ as a curve of degree $6 (g - 1)$ in $P^{5 g - 6}$ (via Riemann-Roch and asymptotic vanishing).

(M2) Hilbert scheme. The Hilbert scheme $Hilb_{P^{N}}^{P}$ parametrises closed subschemes of $P^{N}$ with given Hilbert polynomial $P$ . For tri-canonically embedded smooth curves, the Hilbert polynomial is $P (m) = 6 (g - 1) m - g + 1$ . Let $H \subset Hilb^{P}$ be the open subscheme of smooth curves.

(M3) PGL action. The group $PGL_{N + 1}$ acts on $H$ by changing the embedding (different choices of basis for tri-canonical sections). Two points of $H$ correspond to the same abstract curve iff they differ by a $PGL_{N + 1}$ action.

(M4) GIT quotient. The moduli space is the GIT quotient $M_{g} = H / / PGL_{N + 1}$ , taken with respect to a suitable linearisation. Mumford's GIT theorem (1965) identifies this quotient as a quasi-projective scheme.

Stable-curve compactification (Deligne-Mumford 1969). The non-compactness of $M_{g}$ comes from families of smooth curves that degenerate (one curve in the family becomes singular). Deligne-Mumford 1969 The irreducibility of the space of curves of given genus (Publ. Math. IHES 36) introduced the moduli space of stable curves $\overline{M_{g}}$ :

A stable curve of genus $g$ is a connected projective curve $C$ such that:

Every singularity of $C$ is a node (i.e., $O_{C, p}$ analytically isomorphic to $k [x, y] / (x y)$ ).
The automorphism group $Aut (C)$ is finite.
Equivalently: $ω_{C}$ is ample on $C$ .

Stability ensures the moduli space is Hausdorff / separated and proper (compact in the sense of algebraic geometry). $\overline{M_{g}}$ is a projective scheme (or proper Deligne-Mumford stack) containing $M_{g}$ as a dense open.

Boundary divisors. $\partial \overline{M_{g}} = \overline{M_{g}} ∖ M_{g}$ has a stratification by boundary strata. The components $Δ_{i}$ for $0 \leq i \leq ⌊ g /2 ⌋$ :

$Δ_{0}$ : irreducible nodal curves of geometric genus $g - 1$ .
$Δ_{i}$ for $1 \leq i \leq g /2$ : reducible curves with two components of genera $i$ and $g - i$ joined at a single node.

The Picard group $Pic (\overline{M_{g}}) \otimes Q$ is generated by the $Δ_{i}$ 's and the Hodge class $λ$ — modulo a few relations (Harer 1983; Arbarello-Cornalba 1987).

Tautological classes. On $\overline{M_{g, n}}$ there are distinguished classes:

$ψ_{i} = c_{1} (L_{i})$ for $i = 1, \dots, n$ , where $L_{i}$ is the cotangent line at the $i$ -th marked point.
$κ_{a} = π_{*} (ψ_{n + 1}^{a + 1})$ for $a \geq 0$ , with $π : \overline{M_{g, n + 1}} \to \overline{M_{g, n}}$ the forgetting map.
$λ_{a} = c_{a} (E)$ where $E$ is the Hodge bundle — the rank- $g$ vector bundle whose fibre over $[C]$ is $H^{0} (C; ω_{C})$ .

These classes generate a subring $R^{*} (\overline{M_{g}}) \subset CH^{*} (\overline{M_{g}})$ — the tautological ring — central in modern moduli theory.

Key theorem with proof [Intermediate+]

Theorem (Mumford 1965). For $g \geq 2$ , the moduli space $M_{g}$ of smooth projective genus- $g$ curves over $C$ exists as a quasi-projective complex variety of dimension $3 g - 3$ .

Proof outline. Step 1 — tri-canonical embedding. For $g \geq 2$ , $ω_{C}$ is ample on a smooth genus- $g$ curve $C$ , so $ω_{C}^{\otimes 3}$ is very ample (Mumford-Fogarty-Kirwan §6). The complete linear system $∣3 K_{C} ∣ = H^{0} (C; ω_{C}^{\otimes 3})$ has dimension $5 g - 5$ (by Riemann-Roch: $dim H^{0} (ω^{\otimes 3}) = 6 (g - 1) + 1 - g = 5 g - 5$ for $g \geq 2$ , since the first cohomology vanishes by Serre's vanishing for sufficiently positive bundles). It defines a closed embedding $C ↪ P^{5 g - 6}$ as a curve of degree $6 (g - 1)$ .

Step 2 — Hilbert scheme parameter space. The Hilbert scheme $H$ of all closed subschemes of $P^{5 g - 6}$ with Hilbert polynomial $P (m) = 6 (g - 1) m - g + 1$ exists as a projective scheme (Grothendieck 1961). The locus $H^{\circ} \subset H$ of tri-canonically embedded smooth genus- $g$ curves is an open subscheme.

Step 3 — group action and GIT. The group $PGL (5 g - 5)$ acts on $H^{\circ}$ by changes of basis on the tri-canonical sections. The orbits are exactly the isomorphism classes of curves: two points of $H^{\circ}$ correspond to the same abstract curve iff they differ by a $PGL$ action.

Step 4 — GIT quotient. Mumford's GIT theorem produces the quotient $H^{\circ} / / PGL$ as a quasi-projective scheme. The technical input: smooth curves of $g \geq 2$ have finite automorphism groups, so the GIT quotient is well-defined (no positive-dimensional stabilisers). The result is $M_{g} := H^{\circ} / / PGL$ .

Step 5 — dimension count. $dim H^{\circ} = dim H = (dim of Hilbert scheme of smooth curves)$ . By tangent-space analysis, this equals $3 g - 3 + dim PGL (5 g - 5) = 3 g - 3 + (5 g - 5)^{2} - 1$ . Subtracting the dimension of $PGL$ :

dim M_{g} = dim H^{\circ} - dim PGL (5 g - 5) = 3 g - 3.

This recovers Riemann's 1857 count, now rigorously established. $□$

The construction depends on $g \geq 2$ . For $g = 0, 1$ : $M_{0}$ is a point (no moduli), $M_{1}$ is the affine line $A^{1}$ via the $j$ -invariant.

Bridge. The construction here builds toward later units of the strand, where the same pattern is taken up at higher structure. The defining pattern appears again in those units in a sharpened form, where the local data is glued or quotiented. Putting these together, the foundational insight is that the data of this unit gives the structural signature that the rest of the strand reads off.

Exercises [Intermediate+]

Exercise 2 (easy, proof).

Show that the moduli space $M_{1}$ of smooth projective genus-1 curves over $C$ is identified with $A_{C}^{1}$ via the $j$ -invariant.

Hint

The $j$ -invariant is the rational function on Weierstrass-form coefficients that distinguishes elliptic curves up to isomorphism.

Answer

Every smooth projective genus-1 curve over $C$ admits a Weierstrass equation $y^{2} = x^{3} + a x + b$ (with discriminant $Δ = - 16 (4 a^{3} + 27 b^{2}) \neq = 0$ for smoothness). Two such curves with parameters $(a, b)$ and $(a^{'}, b^{'})$ are isomorphic iff there exists $λ \neq = 0$ with $a^{'} = λ^{4} a$ and $b^{'} = λ^{6} b$ .

The $j$ -invariant is the rational function

j (a, b) := - 1728 \cdot \frac{4 a ^{3}}{4 a ^{3} + 27 b ^{2}} \cdot \frac{1}{?} = 1728 \cdot \frac{4 a ^{3}}{4 a ^{3} + 27 b ^{2}},

(up to convention; the standard normalisation gives $j (Λ) = 1728 \frac{g _{2}^{3}}{g _{2}^{3} - 27 g _{3}^{2}}$ for the lattice version).

The $j$ -invariant is invariant under the rescaling $(a, b) \mapsto (λ^{4} a, λ^{6} b)$ , so it descends to a function on $M_{1} = {(a, b) : 4 a^{3} + 27 b^{2} \neq = 0} / C^{\times}$ . The map $j : M_{1} \to A^{1}$ is a bijection (every value of $j \in C$ corresponds to exactly one isomorphism class), and is in fact an isomorphism of schemes.

So $M_{1} ≅ A_{C}^{1}$ . The j-invariant is the universal coordinate on the moduli space of elliptic curves.

Exercise 3 (medium, conceptual).

Why is $M_{g}$ only a coarse moduli space, not a fine moduli space?

Hint

Curves with non-identity automorphisms.

Answer

A fine moduli space would carry a universal family $C \to M_{g}$ such that every family $C_{T} \to T$ is the pullback of $C$ via a unique morphism $T \to M_{g}$ .

This fails for $M_{g}$ because: families of curves with the same isomorphism class but different automorphism representatives map to the same point of $M_{g}$ , but the families themselves are not isomorphic.

Concretely: take a smooth curve $C$ with non-identity automorphism $ϕ : C \to C$ . Consider the constant family $C \times Spec k [ϵ] \to Spec k [ϵ]$ vs. the twisted family obtained by gluing two copies of $C \times Spec k [ϵ]^{\circ}$ via $ϕ$ . Both families have constant isomorphism class equal to $[C]$ , hence map to the same point of $M_{g}$ , but are not isomorphic as families.

The right object is the Deligne-Mumford stack $M_{g}^{stack}$ — a stack remembering automorphism data, where $M_{g}^{stack} (T)$ is the category (groupoid) of families over $T$ , not the set of isomorphism classes.

The coarse moduli space $M_{g}$ is the "best approximation by a scheme" to the stack — it has the same points as $M_{g}^{stack}$ , but loses the automorphism information.

For curves with $g \geq 4$ generic, the automorphism group is the identity, and $M_{g}$ approximates the stack well in the generic locus. Special curves (hyperelliptic, with extra automorphisms) form a positive-codimensional locus where the stack/coarse-space distinction matters.

Exercise 4 (medium, proof).

Prove that the Hodge bundle $E \to M_{g}$ is well-defined as a rank- $g$ vector bundle, with fibre over $[C]$ equal to $H^{0} (C; ω_{C})$ .

Hint

Use the relative dualising sheaf and constancy of $h^{0} (ω_{C}) = g$ .

Answer

Let $π : C_{g} \to M_{g}$ be a family of smooth projective curves over the moduli space (or over an étale cover of the moduli stack — strictly speaking, we work on the stack). The relative dualising sheaf $ω_{C / M}$ is a line bundle on $C_{g}$ with fibre over $[C] \in M_{g}$ equal to $ω_{C}$ .

The pushforward $π_{*} ω_{C / M}$ is a coherent sheaf on $M_{g}$ whose fibre at $[C]$ is, by base change, $H^{0} (C; ω_{C}) ≅ k^{g}$ — a $g$ -dimensional vector space.

For $π_{*} ω_{C / M}$ to be a vector bundle (locally free), the dimensions must be constant — which they are: $dim H^{0} (C; ω_{C}) = g$ for every smooth projective genus- $g$ curve $C$ .

Moreover, $H^{1} (C; ω_{C}) ≅ k$ is also constant, so by Grauert's theorem (constancy of fibre dimensions implies the pushforward is locally free), $π_{*} ω_{C / M}$ is a vector bundle of rank $g$ . Call it the Hodge bundle $E$ . $□$

The Hodge bundle is one of the most important sheaves on $M_{g}$ . Its Chern classes $λ_{i} = c_{i} (E)$ are the Hodge classes; they generate, alongside the boundary classes, the rational Picard group of $\overline{M_{g}}$ .

Exercise 5 (medium, numeric).

For $g = 2$ , identify $M_{2}$ with a quotient of an open subset of $A^{6}$ by a group action.

Hint

A genus-2 curve is hyperelliptic, given by $y^{2} = f (x)$ with $f$ of degree 5 or 6.

Answer

Every smooth genus-2 curve over $C$ is hyperelliptic: it admits a unique 2-to-1 morphism to $P^{1}$ , with 6 branch points (by Riemann-Hurwitz for genus 2).

The curve is given by $y^{2} = f (x)$ where $f \in C [x]$ has degree 5 or 6 with distinct roots — equivalently, the locus of (squared) sextics ${(b_{0}, \dots, b_{5}) \in C^{6} : f = b_{0} x^{5} + b_{1} x^{4} + \dots + b_{5} has 5 or 6 distinct roots}$ (with degree-6 case requiring an extra root at $\infty$ ).

Two genus-2 curves are isomorphic iff their branch loci on $P^{1}$ differ by a Möbius transformation. The group $PGL_{2} (C)$ acts on the 6-tuples of branch points by Möbius transformations.

So $M_{2} = Sym^{6} (P^{1})^{\circ} / PGL_{2}$ , where $Sym^{6} (P^{1})^{\circ}$ is the open locus of 6 distinct points. By dimension count: $dim Sym^{6} (P^{1}) = 6$ and $dim PGL_{2} = 3$ , giving $dim M_{2} = 6 - 3 = 3$ . ✓

The 3-dimensional moduli space of genus-2 curves is realised as a 3-dimensional GIT quotient, matching $3 g - 3 = 3$ .

Exercise 6 (hard, proof).

State the Mumford conjecture on the cohomology of $M_{g}$ and explain its resolution by Madsen-Weiss.

Hint

Stable cohomology of $M_{g}$ as $g \to \infty$ stabilises to $H^{*} (BDiff^{+} (S_{g}^{2}))$ .

Answer

Mumford's conjecture (1983). Let $M_{g}$ be the moduli space of smooth projective genus- $g$ curves. The stable rational cohomology — cohomology in degree $i$ for $g$ sufficiently large compared to $i$ — is freely generated as a polynomial ring by the Mumford-Morita-Miller classes $κ_{1}, κ_{2}, \dots$ :

g \to \infty lim H^{*} (M_{g}; Q) = Q [κ_{1}, κ_{2}, κ_{3}, \dots] .

The Mumford-Morita-Miller classes (or kappa classes) $κ_{a}$ are tautological classes of degree $2 a$ on $M_{g}$ , defined as $κ_{a} := π_{*} (ψ^{a + 1})$ where $π : \overline{M_{g, 1}} \to \overline{M_{g}}$ and $ψ$ is the cotangent line at the marked point.

Madsen-Weiss theorem (2007, Annals of Math. 165). Madsen and Weiss proved Mumford's conjecture using techniques from algebraic topology — cobordism categories and a homotopy-theoretic identification of the stable mapping class group with the infinite loop space of a Thom spectrum:

B Γ_{\infty}^{+} ≃ Ω^{\infty} MTSO (2),

where $Γ_{\infty}$ is the stable mapping class group and $MTSO (2)$ is the Thom spectrum of $- γ_{2}$ over $BSO (2)$ . The cohomology of the right side is exactly $Q [κ_{1}, κ_{2}, \dots]$ , and the Madsen-Weiss group identification translates into Mumford's stable cohomology.

Implications.

The polynomial generators $κ_{a}$ are "all" the stable cohomology classes — no other invariants exist in the stable range.
The proof involves cobordism categories, scanning, and the "group completion" of mapping class group spaces — a deep mix of homotopy theory and geometric topology.
Generalisations: Galatius-Madsen-Tillmann-Weiss extended to higher-dimensional manifold cobordisms; Galatius-Randal-Williams to manifold mapping class groups in higher dimensions.

Modern context. Mumford's conjecture occupied algebraic topology and moduli theory for 25 years (1983–2007). Its resolution opened the modern theory of moduli spaces of manifolds, with applications to classical mapping class groups, characteristic classes of fibre bundles, and the cohomology of automorphism groups of smooth manifolds. $□$

Exercise 7 (hard, conceptual).

State the Witten conjecture and Kontsevich's theorem on intersection numbers of $ψ$ -classes on $\overline{M_{g, n}}$ .

Hint

Generating function for intersection numbers satisfies a KdV hierarchy.

Answer

Witten's conjecture (1991). Define the generating function of intersection numbers of $ψ$ -classes on the moduli of stable curves:

F (t_{0}, t_{1}, t_{2}, \dots) := g, n \geq 0 \sum a_{1}, \dots, a_{n} \sum \frac{t _{a_{1}} \dots t _{a_{n}}}{n !} \int_{\overline{M_{g, n}}} ψ_{1}^{a_{1}} \dots ψ_{n}^{a_{n}} .

This formal power series in infinitely many variables encodes all $ψ$ -class intersection numbers on $\overline{M_{g, n}}$ .

Witten's conjecture (1991, Surveys in Differential Geometry). $F$ satisfies the Korteweg-de Vries hierarchy (KdV) — a system of integrable differential equations from mathematical physics:

\frac{\partial F}{\partial t _{n}} = KdV_{n} [F],

where the $KdV_{n}$ are specific differential operators.

*Kontsevich's theorem (1992, Comm. Math. Phys.).* Maxim Kontsevich proved Witten's conjecture using matrix integrals — Hermitian matrix models computing intersection numbers via Feynman diagrams. The proof identifies $F$ with a matrix integral satisfying KdV via known integrable-system properties of matrix models.

Connections.

String theory. Witten's conjecture arose from 2-dimensional topological gravity — a string-theory model where intersections on $\overline{M_{g, n}}$ compute "graviton scattering amplitudes."
Matrix models. The proof connects $\overline{M_{g, n}}$ to random matrix theory and the GUE ensemble.
Tautological ring. Witten/Kontsevich is the prototype of a tautological intersection-number computation on $\overline{M_{g, n}}$ .
Generalisations. Mirzakhani 2006 (Fields Medal 2014) gave a topological proof using hyperbolic geometry. Givental 2001 extended to Gromov-Witten generating functions of $P^{1}$ . Faber-Pandharipande and Pixton developed the DR cycles and Pixton's relations extending the theory.

Significance. Witten's conjecture was the first major bridge between algebraic geometry of moduli spaces and integrable systems. It opened the modern theory of Gromov-Witten invariants, quantum cohomology, and mirror symmetry — vast areas of mathematical physics with deep ties to algebraic geometry. $□$

Lean formalization [Intermediate+]

lean_status: partial — Mathlib has algebraic-geometry foundations and stack infrastructure in development; the moduli stack of curves is not yet a named Mathlib construction.

[object Promise]

Advanced results [Master]

Deligne-Mumford compactification (1969). $\overline{M_{g}}$ is a smooth Deligne-Mumford stack, projective over $Z$ . The boundary $\partial \overline{M_{g}}$ is a normal-crossing divisor with components $Δ_{0}, Δ_{1}, \dots, Δ_{⌊ g /2 ⌋}$ . Deligne-Mumford 1969 proved $M_{g}$ is irreducible — a long-conjectured statement now standard.

Tautological ring. The tautological ring $R^{*} (\overline{M_{g, n}}) \subset CH^{*} (\overline{M_{g, n}})$ is generated by the $ψ$ -classes, $κ$ -classes, $λ$ -classes, and boundary classes. Faber-Pandharipande conjectured (1999) explicit generators and relations; Pixton's relations (2012) gave a conjectural complete set of relations, with substantial progress (Janda 2018, Pixton-Pandharipande-Zvonkine).

Mumford-Morita-Miller (kappa) classes. The $κ_{a}$ classes on $\overline{M_{g}}$ generate the stable cohomology by Madsen-Weiss (2007, proving Mumford's 1983 conjecture). For $g \to \infty$ , $H^{*} (M_{g}; Q) = Q [κ_{1}, κ_{2}, \dots]$ .

Witten-Kontsevich theorem. The generating function of $ψ$ -class intersection numbers on $\overline{M_{g, n}}$ satisfies the KdV hierarchy. Kontsevich proved Witten's 1991 conjecture using matrix integrals (1992, Comm. Math. Phys.). The first major bridge between moduli-space intersection theory and integrable systems.

Faber's intersection conjecture. Faber (1999) conjectured that $R^{*} (M_{g})$ is a Gorenstein ring with explicit socle. The conjecture is verified through genus 23 by computational methods (Faber-Pandharipande). Pixton's relations are conjecturally a complete set of relations.

Birational geometry of $\overline{M_{g}}$ . Harris-Mumford 1982 proved $\overline{M_{g}}$ is of general type for $g \geq 24$ . This was a major breakthrough: it shows $\overline{M_{g}}$ for large $g$ has positive Kodaira dimension and is far from rational. Subsequent work (Eisenbud-Harris, Farkas) refined the bounds.

Hyperelliptic locus and Brill-Noether theory. The hyperelliptic locus $H_{g} \subset M_{g}$ has dimension $2 g - 1$ (Riemann-Hurwitz: 6 branch points modulo $PGL_{2}$ + degree-2 morphism data). Brill-Noether theory studies the loci of curves with linear systems of given rank and degree.

Modular forms and $M_{1}$ . $\overline{M_{1, 1}}$ is the modular curve $H / SL_{2} (Z) \cup {\infty}$ , with sections of line bundles $ω^{\otimes k}$ corresponding to modular forms of weight $2 k$ . The Eichler-Shimura theory and Galois representations of modular forms tie $M_{1}$ to number theory (Wiles' proof of Fermat's Last Theorem).

Synthesis. This construction generalises the pattern fixed in 04.04.01 (riemann-roch theorem for curves), with the symmetric data replaced by its skew or twisted analogue. Read in the opposite direction, the construction is dual to the metric story: complements and orthogonality are taken with respect to the bilinear datum of this unit, not a metric, and the resulting category of subobjects is the one the rest of the strand classifies. The central insight is that this datum identifies algebra with geometry: functions become vector fields, subspaces become quotients, and invariants become cohomology classes — and that identification is the engine driving every theorem downstream.

Full proof set [Master]

Mumford's GIT construction is sketched in the formal-definition section; the full proof occupies Mumford-Fogarty-Kirwan §6. Deligne-Mumford 1969 (Publ. Math. IHES 36) constructs $\overline{M_{g}}$ . Tautological-ring relations: Faber-Pandharipande, Pixton 2012, Janda 2018. Mumford conjecture: Madsen-Weiss 2007 (Annals 165). Witten-Kontsevich: Kontsevich 1992 (CMP).

Connections [Master]

Riemann-Roch theorem for curves 04.04.01 — the dimension count $3 g - 3$ comes via deformation theory and Riemann-Roch.
Coherent sheaf 04.06.02 — the Hodge bundle and tautological classes are coherent sheaves on $\overline{M_{g}}$ .
Sheaf cohomology 04.03.01 — cohomology of curves and their moduli is governed by sheaf-cohomology techniques.
Geometric invariant theory 04.10.02 — Mumford's GIT is the construction tool for $M_{g}$ .
Canonical sheaf 04.08.02 — tri-canonical embedding is the key step.
Hodge decomposition 04.09.01 — the Hodge bundle on $M_{g}$ extracts the holomorphic forms.
Riemann surface 06.03.01 — points of $M_{g}$ are isomorphism classes of compact Riemann surfaces of genus $g$ .
Theta function 06.06.05 — theta-divisors on Jacobians of curves trace out divisors on $M_{g}$ .

Historical & philosophical context [Master]

Bernhard Riemann's 1857 Theorie der Abelschen Functionen (J. reine angew. Math. 54) introduced the term moduli (singular Modul) for the parameters needed to describe a Riemann surface. Riemann argued: a Riemann surface of genus $g$ is described by $3 g - 3$ parameters (for $g \geq 2$ ) — his celebrated moduli formula. The argument was heuristic, using the Dirichlet principle and dimension counts on spaces of meromorphic differentials, but the answer was correct.

Riemann's count was rigorised in stages. Felix Klein and Henri Poincaré in the 1880s developed the analytic theory: the moduli space is identified with Teichmüller space modulo the mapping class group — Teichmüller space is the simply connected upper half-space, and $M_{g} = Teich_{g} / Mod_{g}$ where $Mod_{g}$ is the mapping class group of a genus- $g$ surface.

The algebraic-geometric construction came in 1965 with David Mumford's Geometric Invariant Theory (Springer 1965). Mumford realised $M_{g}$ as a quasi-projective scheme via the GIT quotient of the Hilbert scheme of tri-canonically embedded curves by $PGL$ . This was the first rigorous algebraic construction; previously $M_{g}$ was known to exist only as an analytic space (via Teichmüller theory).

The compactification problem — completing $M_{g}$ to a proper space — was solved by Pierre Deligne and David Mumford in 1969 (The irreducibility of the space of curves of given genus, Publ. Math. IHES 36). Their key idea: allow stable curves with at-worst-nodal singularities and finite automorphism groups. The result $\overline{M_{g}}$ is a projective scheme (or Deligne-Mumford stack), proving Mumford's conjectural compactification works. Deligne-Mumford simultaneously proved $M_{g}$ is irreducible — a long-conjectured statement.

The 1980s-90s saw an explosion of moduli-space activity:

Mumford's tautological ring (1983, Towards an Enumerative Geometry of the Moduli Space of Curves): introduced the $κ$ -classes and the tautological subring of $CH^{*} (\overline{M_{g}})$ .
Harris-Mumford (1982, Invent. Math.): proved $\overline{M_{g}}$ is of general type for $g \geq 24$ , a major birational-geometry breakthrough.
Witten's conjecture (1991, Surveys in Differential Geometry): proposed that intersection numbers of $ψ$ -classes on $\overline{M_{g, n}}$ satisfy KdV.
Kontsevich (1992, Comm. Math. Phys.): proved Witten via matrix integrals, opening the Gromov-Witten / quantum cohomology era.

In the 2000s-10s:

Mirzakhani's hyperbolic-geometry proof of Witten-Kontsevich (2006, Fields Medal 2014).
Madsen-Weiss (2007, Annals of Math. 165): proved Mumford's 1983 conjecture on stable cohomology.
Pixton's relations (2012): conjectural complete set of relations for the tautological ring.
Galatius-Randal-Williams and the moduli of manifolds program: extended moduli theory to higher-dimensional manifolds.

Today, moduli spaces of curves sit at the intersection of algebraic geometry, mathematical physics, and topology. They appear in:

String theory (worldsheet path integrals over $\overline{M_{g, n}}$ ).
Mirror symmetry (Gromov-Witten invariants of Calabi-Yau threefolds vs. period integrals).
Number theory (modular forms, Galois representations, the Wiles-Taylor proof of Fermat).
Topology (stable cohomology, mapping class groups, Madsen-Weiss).
Integrable systems (Witten-Kontsevich, KdV hierarchy).

Riemann's 1857 prescient $3 g - 3$ remains the fundamental invariant. The modern theory has uncovered far richer structures, but the basic count came first.

Bibliography [Master]

Riemann, B., Theorie der Abelschen Functionen, J. reine angew. Math. 54 (1857), 115–155.
Mumford, D., Geometric Invariant Theory, Springer-Verlag 1965; 3rd ed. with Fogarty & Kirwan 1994.
Deligne, P. & Mumford, D., The irreducibility of the space of curves of given genus, Publ. Math. IHES 36 (1969), 75–109.
Mumford, D., Towards an enumerative geometry of the moduli space of curves, in Arithmetic and Geometry II, Birkhäuser 1983, 271–328.
Harris, J. & Morrison, I., Moduli of Curves, Springer GTM 187, 1998.
Arbarello, E., Cornalba, M., Griffiths, P. & Harris, J., Geometry of Algebraic Curves I, II, Springer Grundlehren 267 (1985), 268 (2011).
Harris, J. & Mumford, D., On the Kodaira dimension of the moduli space of curves, Invent. Math. 67 (1982), 23–86.
Witten, E., Two-dimensional gravity and intersection theory on moduli space, Surveys in Differential Geometry 1 (1991), 243–310.
Kontsevich, M., Intersection theory on the moduli space of curves and the matrix Airy function, Comm. Math. Phys. 147 (1992), 1–23.
Madsen, I. & Weiss, M., The stable moduli space of Riemann surfaces: Mumford's conjecture, Annals of Math. 165 (2007), 843–941.
Mirzakhani, M., Simple geodesics and Weil-Petersson volumes of moduli spaces of bordered Riemann surfaces, Invent. Math. 167 (2007), 179–222.
Pandharipande, R., Pixton, A. & Zvonkine, D., Relations on $\overline{M}_{g}$ via 3-spin structures, J. Amer. Math. Soc. 28 (2015), 279–309.

Prerequisites

04.04.01
04.06.02

Tier anchors

beginner: Strogatz-style: a space whose points are curves
intermediate: Harris-Morrison *Moduli of Curves*; Mumford-Suominen 1972; Vakil notes
master: Harris-Morrison; Mumford-Fogarty-Kirwan *GIT*; Arbarello-Cornalba-Griffiths *Geometry of Algebraic Curves* Vol II

References

TODO_REF
Mumford — Geometric Invariant Theory · Springer 1965, 3rd ed. Mumford-Fogarty-Kirwan 1994
TODO_REF
Harris-Morrison — Moduli of Curves · Springer GTM 187, 1998
TODO_REF
Deligne-Mumford — The irreducibility of the space of curves of given genus · Publ. Math. IHES 36 (1969), 75–109
TODO_REF
Riemann — Theorie der Abelschen Functionen · 1857, the original 3g-3 dimension count

Lean module

Codex.AlgebraicGeometry.Moduli.ModuliOfCurves

Reviewer

TBD

Estimated time

beginner: 12m
intermediate: 40m
master: 70m