04.10.31 · algebraic-geometry / moduli

Severi varieties of nodal plane curves and the Harris irreducibility theorem

shipped3 tiersLean: none

Anchor (Master): Harris 1986 *On the Severi problem* (Invent. Math. 84, 445-461, irreducibility); Severi 1921 *Vorlesungen über algebraische Geometrie* (Anhang F, the original existence/dimension claim); Caporaso-Harris 1998 *Counting plane curves of any genus* (Invent. Math. 131, 345-392, the degree recursion); Diaz-Harris 1988 *Geometry of the Severi variety* (Trans. AMS 309, divisor theory); Arbarello-Cornalba 1981 *Su una proprietà notevole dei morfismi di una curva a moduli* (Boll. UMI, the family-to-moduli map)

Intuition Beginner

Fix a degree $d$ . A plane curve of degree $d$ is the zero locus of a homogeneous polynomial of degree $d$ in three variables. The space of all such polynomials, up to scaling, is itself a projective space: its points are the curves, and its dimension is $N = (2 d + 2) - 1$ . For instance, conics ( $d = 2$ ) form a $5$ -dimensional family, and cubics ( $d = 3$ ) form a $9$ -dimensional family. This is the simplest example of a parameter space — a single space whose points are geometric objects.

Most curves in this family are smooth. But some are singular, and the mildest singularity a curve can have is a node: a point where two smooth branches cross transversally, like the self-crossing of a figure-eight. The Severi variety $V_{d, δ}$ is the set of degree- $d$ plane curves that have exactly $δ$ nodes and no worse singularity. It sits inside the big projective space of all degree- $d$ curves as a special, more rigid subfamily.

Two questions drive the whole subject. How big is this family — what is its dimension? And is it connected in the strongest sense, meaning you can deform any one $δ$ -nodal curve into any other while keeping all $δ$ nodes? The first question has a clean answer: each node you demand costs one dimension, so the family has dimension $\frac{d ( d + 3 )}{2} - δ$ . The second question is the Severi problem, and its answer — yes, the family is irreducible — was settled by Harris in 1986.

Visual Beginner

Picture the giant projective space of all degree- $d$ plane curves as a large region. Inside it, the smooth curves fill up almost everything, an open dense part. The curves that fail to be smooth form a hypersurface, the discriminant, sitting one dimension down. On that hypersurface, the generic point is a curve with a single node. Curves with two nodes form a deeper stratum, curves with three nodes deeper still, and so on. The Severi variety $V_{d, δ}$ is the stratum of curves with exactly $δ$ nodes, and its closure sweeps in the deeper strata where nodes collide into worse singularities.

The picture makes the dimension count visible: each added node digs you one layer deeper into the stratification, so $V_{d, δ}$ has codimension $δ$ , hence dimension $\frac{d ( d + 3 )}{2} - δ$ . The picture also previews irreducibility: a single connected stratum, not a union of separate pieces, even though a priori one might fear that different ways of arranging the nodes give different components.

Worked example Beginner

Count the dimension of the family of nodal cubics, and read off their genus.

Step 1. Take $d = 3$ . A plane cubic is the zero locus of a homogeneous cubic polynomial in three variables. The space of cubic monomials has dimension $(2 3 + 2) = (2 5) = 10$ , so up to scaling the cubics form a projective space of dimension $N = 10 - 1 = 9$ . A general cubic is smooth and has genus $(2 3 - 1) = (2 2) = 1$ , an elliptic curve.

Step 2. Demand one node. Requiring a curve to be singular at some point is one condition, so the family of one-nodal cubics has dimension $9 - 1 = 8$ . The formula $\frac{d ( d + 3 )}{2} - δ$ with $d = 3, δ = 1$ gives $\frac{3 \cdot 6}{2} - 1 = 9 - 1 = 8$ , matching. A one-nodal cubic is the familiar curve with a self-crossing, and resolving the node gives a rational curve, genus $0$ .

Step 3. Read the genus drop. A smooth plane cubic has genus $1$ . Each node, when resolved by separating the two branches, lowers the genus by one. The genus formula is $g = (2 d - 1) - δ = 1 - δ$ . So a one-nodal cubic has geometric genus $1 - 1 = 0$ , a rational curve, exactly as the picture of a self-crossing loop suggests.

Step 4. Push to two nodes. A two-nodal cubic would have $g = 1 - 2 = - 1$ , which is impossible for an actual curve. The deepest a cubic can degenerate while staying reduced is the one-nodal case for an irreducible curve; two nodes force the cubic to break into a conic plus a line, or three lines. This is the first sign that the numbers $d, δ$ cannot be chosen freely: the genus $g = (2 d - 1) - δ$ must stay non-negative, so $δ \leq (2 d - 1)$ .

What this tells us: the dimension formula and the genus formula are two readings of the same bookkeeping. Each node spends one parameter from the family and one unit of genus from the smooth curve, and the family of $δ$ -nodal degree- $d$ curves is non-empty exactly in the range where the genus stays non-negative.

Check your understanding Beginner

Formal definition Intermediate+

Work over an algebraically closed field $k$ of characteristic zero. Let $P^{2} = P_{k}^{2}$ and fix a degree $d \geq 1$ . The complete linear system of degree- $d$ plane curves is the projective space $$ \mathbb{P}^N = |\mathcal{O}{\mathbb{P}^2}(d)| = \mathbb{P}\big(H^0(\mathbb{P}^2, \mathcal{O}{\mathbb{P}^2}(d))\big), \qquad N = \binom{d+2}{2} - 1, $$ whose points are the effective divisors of degree $d$ on $P^{2}$ , that is, the plane curves of degree $d$ .

Definition (node). A reduced plane curve $C$ has a node at a point $p$ when, in suitable local coordinates $(x, y)$ centred at $p$ , the local equation of $C$ is $x y = 0$ up to higher-order terms — two smooth branches meeting transversally. Equivalently, the local ring $O_{C, p}$ has $δ$ -invariant equal to $1$ and the projectivised tangent cone is two distinct points.

Definition (Severi variety). The Severi variety $V_{d, δ} \subset P^{N}$ is the locally closed subscheme parametrising reduced curves $C \subset P^{2}$ of degree $d$ having exactly $δ$ nodes and no other singularities. Its closure $\overline{V}_{d, δ}$ in $P^{N}$ adds the boundary curves where nodes collide (forming cusps, tacnodes, triple points) or where the curve becomes non-reduced.

Definition (expected dimension). The expected dimension of $V_{d, δ}$ is $$ \dim V_{d,\delta} = \frac{d(d+3)}{2} - \delta = N - \delta. $$ The right-hand equality records that imposing one node is, generically, one linear condition on the linear system, so $δ$ nodes cut the codimension down by $δ$ .

Definition (geometric genus). For $C \in V_{d, δ}$ irreducible with normalisation $ν : C \to C$ , the geometric genus of $C$ is the genus of the smooth curve $C$ , $$ g = \binom{d-1}{2} - \delta = \frac{(d-1)(d-2)}{2} - \delta, $$ the arithmetic genus $(2 d - 1)$ of a smooth degree- $d$ plane curve minus one for each node resolved.

Definition (node scheme and the equisingular ideal). Let $C \in V_{d, δ}$ have node set $N = {p_{1}, \dots, p_{δ}}$ , regarded as a reduced length- $δ$ subscheme of $P^{2}$ . The tangent space to $V_{d, δ}$ at $[C]$ is, as a subspace of $T_{[C]} P^{N} = H^{0} (P^{2}, O (d)) / ⟨ f ⟩$ (with $f$ the equation of $C$ ), $$ T_{[C]} V_{d,\delta} = H^0\big(\mathbb{P}^2, \mathcal{O}(d) \otimes \mathcal{I}_N\big) / \langle f \rangle, $$ the curves of degree $d$ passing through all $δ$ nodes. Here $I_{N}$ is the ideal sheaf of the reduced node scheme; the sections of $O (d) \otimes I_{N}$ are the adjoint curves, and $I_{N}$ is the conductor ideal of the partial normalisation $ν$ along the nodes (the largest ideal of $O_{C}$ that is also an ideal of $ν_{*} O_{C}$ ).

Counterexamples to common slips

The expected dimension is not always the actual dimension for special configurations. For a general curve in $V_{d, δ}$ the count $N - δ$ holds, but if the $δ$ nodes are forced into a special position (collinear, or on a low-degree curve) the conditions can fail to be independent and the local dimension can jump. The Severi-Zariski smoothness theorem says the failure does not happen for nodes in general position.
A point of $\overline{V}_{d, δ}$ is not always a $δ$ -nodal curve. The closure adds curves with a cusp (a node degenerated), a tacnode (two nodes collided), or extra nodes appearing as the curve degenerates. These boundary strata are where the irreducibility argument does its work — they are not in $V_{d, δ}$ itself.
Irreducibility of $V_{d, δ}$ is not the statement that a single nodal curve is irreducible as a plane curve. It is the statement that the parameter space $V_{d, δ}$ has one component. A $δ$ -nodal curve may itself be reducible (a union of lines, say), and reducible curves can sit in the closure of $V_{d, δ}$ .
The number $δ$ cannot exceed $(2 d - 1)$ for an irreducible curve, since the genus $g = (2 d - 1) - δ$ must stay non-negative. At $δ = (2 d - 1)$ the curves are rational, and $V_{d, (2 d - 1)}$ is the variety of rational nodal plane curves of degree $d$ .

Key theorem with proof Intermediate+

Theorem (smoothness and dimension of the Severi variety; Severi 1921, rigorous form Zariski 1982). Let $C \in V_{d, δ}$ be a $δ$ -nodal plane curve of degree $d$ . Then $V_{d, δ}$ is smooth at $[C]$ of the expected dimension $dim V_{d, δ} = \frac{d ( d + 3 )}{2} - δ = N - δ$ . Equivalently, the $δ$ conditions of passing through the nodes are linearly independent on $H^{0} (P^{2}, O (d))$ , and the tangent space is $$ T_{[C]} V_{d,\delta} = H^0\big(\mathbb{P}^2, \mathcal{O}(d) \otimes \mathcal{I}_N\big) / \langle f \rangle, \qquad \dim = N - \delta. $$

Proof. The argument runs in three steps: identify the equisingular deformations with sections vanishing at the nodes, compute the dimension of that space by a cohomology vanishing, and identify the cokernel with the obstruction.

Step 1: equisingular deformations are adjoint curves. A first-order deformation of $C$ inside $P^{N}$ is a tangent vector $g \in H^{0} (P^{2}, O (d)) / ⟨ f ⟩$ ; the deformed curve is ${f + ϵ g = 0}$ over $k [ϵ] / ϵ^{2}$ . The deformation preserves the node at $p_{i}$ exactly when $g$ vanishes at $p_{i}$ — to first order, perturbing by a section that does not pass through $p_{i}$ smooths the node, while a section through $p_{i}$ keeps a node nearby (in the analytic-local model $x y + ϵ g = 0$ , the node persists when $g (p_{i}) = 0$ ). So the equisingular tangent directions are $g \in H^{0} (P^{2}, O (d) \otimes I_{N})$ , the adjoint curves passing through all $δ$ nodes.

Step 2: the node conditions are independent. Consider the exact sequence of sheaves on $P^{2}$ obtained from the ideal of the reduced node scheme $N$ , $$ 0 \to \mathcal{O}(d) \otimes \mathcal{I}_N \to \mathcal{O}(d) \to \mathcal{O}_N(d) \to 0, $$ where $O_{N} (d)$ is the length- $δ$ skyscraper supported at the nodes. Taking the long exact sequence in cohomology gives $$ 0 \to H^0(\mathcal{O}(d) \otimes \mathcal{I}_N) \to H^0(\mathcal{O}(d)) \xrightarrow{\mathrm{ev}_N} H^0(\mathcal{O}_N(d)) \to H^1(\mathcal{O}(d) \otimes \mathcal{I}_N) \to 0. $$ The evaluation map $ev_{N}$ is surjective — that is, the $δ$ node conditions are independent — provided $H^{1} (P^{2}, O (d) \otimes I_{N}) = 0$ . For nodes in general position this vanishing holds in the relevant range $δ \leq (2 d - 1)$ : the linear system of adjoint curves $∣ O (d - 3) + nodes ∣$ separates the $δ$ general points because a length- $δ$ general scheme imposes independent conditions on plane curves of degree $d$ once $d$ is large enough relative to $δ$ , which is exactly the regularity guaranteed in the Severi range. Hence $$ \dim H^0(\mathcal{O}(d) \otimes \mathcal{I}_N) = \dim H^0(\mathcal{O}(d)) - \delta = (N+1) - \delta, $$ so the projectivised tangent space has dimension $N - δ$ .

Step 3: smoothness via unobstructedness. The obstruction to extending a first-order equisingular deformation to higher order lies in $H^{1} (P^{2}, O (d) \otimes I_{N})$ . The vanishing established in Step 2 kills the obstruction, so every first-order equisingular deformation integrates: $V_{d, δ}$ is unobstructed at $[C]$ , hence smooth, and its dimension equals the tangent-space dimension $N - δ = \frac{d ( d + 3 )}{2} - δ$ . $□$

Bridge. The smoothness-and-dimension theorem builds toward the irreducibility theorem and the map to the moduli space, and the central insight is that the tangent space to $V_{d, δ}$ is the space of adjoint curves $H^{0} (O (d) \otimes I_{N})$ through the nodes — this is exactly the conductor-ideal description that ties the parameter geometry to the normalisation $C$ . The foundational reason the dimension comes out to $N - δ$ is that nodes in general position impose independent conditions, and this independence is dual to the genus drop $g = (2 d - 1) - δ$ : the same $δ$ that cuts the dimension is the $δ$ that lowers the genus. Putting these together, a general point of $V_{d, δ}$ carries a smooth normalisation of genus $g$ , and the assignment $[C] \mapsto [C]$ generalises the classical plane-curve genus formula into a morphism $V_{d, δ} ⇢ M_{g}$ whose study is the bridge from plane-curve degenerations to abstract moduli. This appears again in 04.10.01 (the moduli space $M_{g}$ ) and 04.04.02 (the Hurwitz formula reading of the genus drop), and the bridge is that the Severi variety is the parameter space realising every genus- $g$ curve as a plane nodal model.

Exercises Intermediate+

Exercise 3 (medium, symbolic).

Derive the geometric genus formula $g = (2 d - 1) - δ$ from the arithmetic genus of a smooth degree- $d$ plane curve and the $δ$ -invariant of a node.

Hint

The arithmetic genus $p_{a} = (2 d - 1)$ is a deformation invariant; for a nodal curve $p_{a} = g + \sum_{p} δ_{p}$ where $δ_{p}$ is the $δ$ -invariant, and a node has $δ_{p} = 1$ .

Answer

A smooth plane curve of degree $d$ has arithmetic genus $p_{a} = (2 d - 1)$ by the adjunction formula on $P^{2}$ : $2 p_{a} - 2 = (K_{P^{2}} + C) \cdot C = (- 3 + d) d$ , so $p_{a} = \frac{d ^{2} - 3 d + 2}{2} = (2 d - 1)$ . Arithmetic genus is constant in a flat family, so a $δ$ -nodal degree- $d$ curve also has $p_{a} = (2 d - 1)$ . For a reduced curve with normalisation $ν : C \to C$ of geometric genus $g$ , the relation $p_{a} = g + \sum_{p} δ_{p} (C)$ holds, where $δ_{p}$ is the local $δ$ -invariant $dim_{k} (ν_{*} O_{C} / O_{C})_{p}$ . A node has $δ_{p} = 1$ (two branches glued at a point: one dimension of difference). Summing over the $δ$ nodes, $(2 d - 1) = g + δ$ , hence $g = (2 d - 1) - δ$ . Rubric: full credit for the adjunction computation of $p_{a}$ , the deformation-invariance of $p_{a}$ , and the $δ$ -invariant accounting $p_{a} = g + δ$ .

Exercise 4 (medium, short-answer).

Explain why imposing a node at a fixed general point $p \in P^{2}$ is one condition on the linear system $∣ O (d) ∣$ , but imposing a node at an unspecified point is also one condition on the family.

Hint

A node at a fixed $p$ means $f (p) = 0$ and $\nabla f (p) = 0$ — three conditions — but the position of $p$ is two free parameters. Net count: $3 - 2 = 1$ .

Answer

Fix a point $p$ . For the curve ${f = 0}$ to be singular at $p$ requires $f (p) = 0$ and the two partial derivatives $f_{x} (p) = f_{y} (p) = 0$ (in affine coordinates), three linear conditions on the coefficients of $f$ — and for a general such $f$ the singularity is a node. So "singular at the fixed point $p$ " is three conditions on $∣ O (d) ∣$ . But when the node location is allowed to vary, $p$ ranges over a $2$ -dimensional family. The locus of pairs $(C, p)$ with $C$ singular at $p$ has codimension $3$ in $P^{N} \times P^{2}$ , hence codimension $3 - 2 = 1$ after projecting to $P^{N}$ and forgetting $p$ . Net: demanding one node (at an unspecified point) is one condition, so $dim V_{d, 1} = N - 1$ , and inductively $dim V_{d, δ} = N - δ$ when the nodes stay in general position. Rubric: full credit for the $3$ -conditions-minus- $2$ -parameters count and the conclusion that one node is codimension one.

Exercise 5 (medium, short-answer).

State what the tangent space $T_{[C]} V_{d, δ} = H^{0} (O (d) \otimes I_{N}) / ⟨ f ⟩$ says about which deformations of $C$ preserve all $δ$ nodes, and connect $I_{N}$ to the conductor of the normalisation.

Hint

A deformation $f + ϵ g$ preserves the node at $p_{i}$ when $g$ passes through $p_{i}$ . The conductor is the ideal of $O_{C}$ measuring the gap to $ν_{*} O_{C}$ .

Answer

A first-order deformation ${f + ϵ g = 0}$ of $C$ keeps the node at $p_{i}$ exactly when the perturbing section $g$ vanishes at $p_{i}$ : in the local model $x y + ϵ g = 0$ , the deformed curve still has a node near $p_{i}$ when $g (p_{i}) = 0$ , and smooths it when $g (p_{i}) \neq = 0$ . So the equisingular tangent directions are the sections $g \in H^{0} (O (d) \otimes I_{N})$ passing through all $δ$ nodes — the adjoint curves. The sheaf $I_{N}$ is the ideal of the reduced node scheme, which is precisely the conductor $c = Ann_{O_{C}} (ν_{*} O_{C} / O_{C})$ of the normalisation along the nodes: the largest ideal of $O_{C}$ that is simultaneously an ideal in $ν_{*} O_{C}$ . At a node, the conductor is the maximal ideal of the local ring, of colength $1$ , matching $δ_{p} = 1$ . So adjoint curves through the nodes are exactly the curves whose pullback to $C$ lies in the conductor, which is the classical adjoint condition. Rubric: full credit for the node-preservation reading, the identification of $I_{N}$ with the conductor, and the colength- $1$ local statement.

Exercise 6 (hard, short-answer).

Outline why the cohomology vanishing $H^{1} (P^{2}, O (d) \otimes I_{N}) = 0$ for $δ$ general nodes both gives the expected dimension and proves smoothness of $V_{d, δ}$ .

Hint

The $H^{1}$ controls both the surjectivity of the evaluation map (independence of node conditions, giving the dimension) and the obstruction space (vanishing gives unobstructedness, hence smoothness).

Answer

The ideal sequence $0 \to O (d) \otimes I_{N} \to O (d) \to O_{N} (d) \to 0$ gives the cohomology sequence $$ 0 \to H^0(\mathcal{O}(d) \otimes \mathcal{I}_N) \to H^0(\mathcal{O}(d)) \xrightarrow{\mathrm{ev}_N} H^0(\mathcal{O}_N(d)) \to H^1(\mathcal{O}(d) \otimes \mathcal{I}_N) \to 0, $$ with $dim H^{0} (O_{N} (d)) = δ$ . The vanishing $H^{1} = 0$ forces $ev_{N}$ surjective, so the $δ$ node conditions are independent and $dim H^{0} (O (d) \otimes I_{N}) = (N + 1) - δ$ , giving $dim T_{[C]} V_{d, δ} = N - δ$ — the expected dimension. The same $H^{1}$ is the obstruction space for the equisingular deformation functor: an obstruction to integrating a first-order equisingular deformation to all orders lives in $H^{1} (O (d) \otimes I_{N})$ . Its vanishing means the functor is unobstructed, so $V_{d, δ}$ is smooth at $[C]$ of dimension equal to the tangent dimension $N - δ$ . The two consequences — the dimension count and the smoothness — are thus governed by the single cohomology group, and the regularity of a general length- $δ$ scheme in the Severi range $δ \leq (2 d - 1)$ is what guarantees the vanishing. Rubric: full credit for (a) the ideal sequence and surjectivity giving the dimension, (b) the identification of $H^{1}$ as the obstruction space, and (c) the conclusion that one vanishing gives both statements.

Advanced results Master

Theorem (Harris irreducibility theorem; Harris 1986). Over an algebraically closed field of characteristic zero, for every $d \geq 1$ and every $δ$ with $0 \leq δ \leq (2 d - 1)$ , the Severi variety $V_{d, δ}$ of irreducible plane curves of degree $d$ with exactly $δ$ nodes is irreducible, of dimension $\frac{d ( d + 3 )}{2} - δ$ .

This resolves the Severi problem: Severi 1921 asserted both that $V_{d, δ}$ is non-empty of the expected dimension (correct, and provable by the smoothness theorem above) and that it is irreducible, but Severi's irreducibility argument had a gap. Harris 1986 gave the first complete proof. The strategy is a degeneration to the boundary combined with a monodromy argument. One studies the closure $\overline{V}_{d, δ}$ and the way nodes can be made to collide or split as one moves to the boundary $\overline{V}_{d, δ - 1}$ (curves with one fewer node) or into deeper strata. The key technical input is that, near a general point of the boundary $\overline{V}_{d, δ} \cap \overline{V}_{d, δ - 1}$ , the $δ$ -th node can be smoothed independently of the others, and the family of such smoothings is connected. Iterating, the monodromy group acting on the $δ$ nodes — permuting them as one loops around the discriminant in the parameter space — is shown to be the full symmetric group $S_{δ}$ . Since the nodes can be permuted arbitrarily and individually smoothed, any two $δ$ -nodal curves are connected through $V_{d, δ}$ , so $V_{d, δ}$ has a single component.

Theorem (the map to moduli; Arbarello-Cornalba 1981, Harris-Morrison Ch. 3). For $δ \leq (2 d - 1)$ , normalisation defines a rational map $$ \mu_{d,\delta} : V_{d,\delta} \dashrightarrow \mathcal{M}_g, \qquad [C] \mapsto [\widetilde{C}], \quad g = \binom{d-1}{2} - \delta, $$ sending a nodal plane curve to the isomorphism class of its smooth normalisation. For $d$ large relative to $g$ the map is dominant, so a general curve of genus $g$ arises as the normalisation of a $δ$ -nodal plane curve of degree $d$ .

Combined with Harris irreducibility, the dominance of $μ_{d, δ}$ gives an independent proof that $M_{g}$ is irreducible (the classical Severi approach to the irreducibility of the moduli space, predating the Deligne-Mumford reduction-mod- $p$ proof): the image of an irreducible variety under a dominant rational map is irreducible. This is one of the historically important routes to irreducibility of $M_{g}$ , and it is the reason the Severi variety sits inside the moduli-of-curves strand rather than only in plane-curve geometry.

Theorem (Severi degrees and the Caporaso-Harris recursion; Caporaso-Harris 1998). The degree of the closure $\overline{V}_{d, δ}$ in $P^{N}$ under a general $δ$ -dimensional family of linear conditions — equivalently the number $N^{d, δ}$ of $δ$ -nodal degree- $d$ plane curves through $\frac{d ( d + 3 )}{2} - δ$ general points — satisfies a recursion in $d$ and $δ$ obtained by degenerating one of the point conditions onto a line and bookkeeping the tangency and node-on-the-line contributions.

These Severi degrees $N^{d, δ}$ are the enumerative shadow of the Severi variety. For $δ = (2 d - 1)$ (rational curves) they coincide with the genus-zero Gromov-Witten invariants of $P^{2}$ , the numbers $1, 1, 12, 620, 87304, \dots$ for $d = 1, 2, 3, 4, 5, \dots$ , and the Caporaso-Harris recursion is one of the two classical computations of them (the other being the Kontsevich WDVV recursion). The tropical reformulation of these same counts — node-counting via lattice paths in the Newton polygon — is the Mikhalkin correspondence theorem of 04.12.05; the Severi variety is the algebraic-geometric object whose enumerative invariants the tropical count computes.

Synthesis. The Severi variety $V_{d, δ}$ is the foundational parameter space that turns the geometry of plane nodal curves into statements about the moduli of abstract curves, and the central insight is that a single integer $δ$ controls three things at once: the codimension $δ$ in the linear system, the genus drop $g = (2 d - 1) - δ$ , and the number of nodes whose monodromy must be analysed. Putting these together, the smoothness theorem (the conductor-ideal tangent space $H^{0} (O (d) \otimes I_{N})$ and the $H^{1}$ -vanishing) gives the local structure, Harris 1986 gives the global structure (irreducibility via full $S_{δ}$ monodromy), and the rational map $μ_{d, δ}$ to $M_{g}$ exports both into the moduli space — this is exactly the bridge from plane-curve degenerations to abstract curves. The foundational reason the whole picture coheres is that nodes in general position impose independent conditions and can be smoothed independently: the independence gives the dimension, and the independent smoothability gives the monodromy, so the local cohomology computation and the global irreducibility statement are two readings of the same node-by-node analysis. This generalises in two directions that recur across the corpus: enumeratively, into the Severi degrees and the Caporaso-Harris recursion, which is dual to the tropical node-count of 04.12.05; and moduli-theoretically, into the dominance of $μ_{d, δ}$ , which appears again in 04.10.01 as a route to the irreducibility of $M_{g}$ . The Severi variety is thus the classical ancestor of both modern enumerative geometry and the modern theory of moduli of curves.

Full proof set Master

Proposition (one node is one condition). The locus $V_{d, 1} \subset P^{N}$ of one-nodal degree- $d$ curves is irreducible of dimension $N - 1$ , and its closure is the discriminant hypersurface.

Proof. Consider the incidence variety $$ \Sigma = {(C, p) \in \mathbb{P}^N \times \mathbb{P}^2 : C \text{ is singular at } p}. $$ The fibre of the second projection $pr_{2} : Σ \to P^{2}$ over a point $p$ is the linear subspace of $P^{N}$ cut out by the three linear conditions $f (p) = f_{x} (p) = f_{y} (p) = 0$ (in local affine coordinates), hence a $P^{N - 3}$ . Since $pr_{2}$ is a projective bundle over the irreducible base $P^{2}$ , the total space $Σ$ is irreducible of dimension $(N - 3) + 2 = N - 1$ . The first projection $pr_{1} : Σ \to P^{N}$ is generically injective: a general singular curve has a single singular point, which is a node. So the image $\overline{V}_{d, 1} = pr_{1} (Σ)$ is irreducible of dimension $N - 1$ , and the general member has exactly one node. The image is the discriminant hypersurface, the locus of singular curves. $□$

Proposition (the genus drop equals the node count). Let $C$ be an irreducible degree- $d$ plane curve with $δ$ nodes and normalisation $ν : C \to C$ . Then $g (C) = (2 d - 1) - δ$ .

Proof. By adjunction on $P^{2}$ , a smooth degree- $d$ curve has genus $p_{a} = (2 d - 1)$ , and arithmetic genus is preserved in flat families, so $p_{a} (C) = (2 d - 1)$ for the nodal $C$ as well. The normalisation sequence $0 \to O_{C} \to ν_{*} O_{C} \to S \to 0$ has $S$ a skyscraper supported at the singular points, with stalk length $δ_{p}$ (the local $δ$ -invariant) at $p$ . Taking Euler characteristics, $χ (O_{C}) = χ (O_{C}) + \sum_{p} δ_{p}$ , that is, $1 - g = (1 - p_{a}) + \sum_{p} δ_{p}$ , so $p_{a} = g + \sum_{p} δ_{p}$ . A node has $δ_{p} = 1$ : the normalisation separates the two branches, and locally $ν_{*} O_{C} / O_{C}$ is one-dimensional (the difference between two independent branch-values and the single glued value). Summing over the $δ$ nodes gives $p_{a} = g + δ$ , hence $g = (2 d - 1) - δ$ . $□$

Proposition (tangent space to the equisingular family). At a $δ$ -nodal curve $C \in V_{d, δ}$ with node set $N$ , the Zariski tangent space is $T_{[C]} V_{d, δ} = H^{0} (P^{2}, O (d) \otimes I_{N}) / ⟨ f ⟩$ .

Proof. A tangent vector to $P^{N}$ at $[C]$ is a class $g \in H^{0} (O (d)) / ⟨ f ⟩$ ; it deforms $C$ to $C_{ϵ} = {f + ϵ g = 0}$ over $k [ϵ]$ . Work analytically-locally at a node $p_{i}$ , where $f = x y + (higher order)$ . The deformed equation $x y + ϵ g (x, y) = 0$ defines a family of curves over $Spec k [ϵ]$ ; its singular locus is where $f + ϵ g$ and both partials vanish. To first order in $ϵ$ , the deformed curve retains an ordinary node near $p_{i}$ exactly when $g (p_{i}) = 0$ , and smooths the node (the equation becomes $x y = - ϵ g (p_{i}) + \dots$ , a smooth hyperbola, when $g (p_{i}) \neq = 0$ ). Therefore the deformation is equisingular at $p_{i}$ — node preserved to first order — precisely when $g$ passes through $p_{i}$ . Imposing this at all $δ$ nodes, the equisingular tangent directions are $g \in H^{0} (O (d) \otimes I_{N})$ , where $I_{N}$ is the ideal sheaf of the reduced node scheme. Quotienting by $⟨ f ⟩$ (the deformation that merely rescales $f$ ) gives the stated tangent space. $□$

Proposition (full monodromy on two nodes for conics-plus, the model case). For the universal family of $δ$ -nodal curves, the monodromy on the $δ$ nodes contains every transposition, hence is all of $S_{δ}$ , whenever any two nodes can be brought together to a tacnode and split back symmetrically.

Proof. The monodromy representation is the action of $π_{1} (V_{d, δ})$ on the set of $δ$ nodes of a fixed reference curve, obtained by tracking how the nodes permute as the curve traverses a loop. To produce a transposition swapping nodes $p_{i}$ and $p_{j}$ , degenerate within $\overline{V}_{d, δ}$ so that $p_{i}$ and $p_{j}$ collide into a single tacnode (an $A_{3}$ singularity, two nodes coming together), then re-smooth the tacnode back into two nodes. The local versal deformation of a tacnode contains a loop around its discriminant that exchanges the two nearby nodes — this is the standard local monodromy of an $A_{3}$ singularity, where the two vanishing cycles are interchanged. Lifting this local loop to a loop in $V_{d, δ}$ (possible because the tacnodal stratum lies in $\overline{V}_{d, δ}$ and the other $δ - 2$ nodes are carried along unchanged) produces the transposition $(p_{i} p_{j})$ . Since the nodes are interchangeable — any pair can be collided into a tacnode by Harris's connectivity of the boundary strata — every transposition is realised, and the transpositions generate $S_{δ}$ . Hence the monodromy is the full symmetric group, and $V_{d, δ}$ is irreducible: the components of $V_{d, δ}$ are in bijection with the orbits of the monodromy on the set of labellings of the nodes, and full $S_{δ}$ -monodromy means a single orbit, hence one component. $□$

Theorem (Harris irreducibility), proof strategy. The complete proof in Harris 1986 ^{[Harris 1986]} combines the local monodromy of the tacnode (above) with a global induction on $δ$ . One shows $\overline{V}_{d, δ}$ meets $\overline{V}_{d, δ - 1}$ along the locus where one node is smoothed, that this boundary locus is irreducible by the inductive hypothesis, and that the general such boundary point admits the tacnode-splitting that realises an arbitrary transposition. The base case $δ = 1$ is the irreducibility of the discriminant (proved above). The induction propagates irreducibility up the Severi stratification, and the full $S_{δ}$ monodromy at each stage closes the argument. $□$

Connections Master

Moduli of curves $M_{g}$ 04.10.01. The normalisation map $μ_{d, δ} : V_{d, δ} ⇢ M_{g}$ sends a $δ$ -nodal degree- $d$ plane curve to its smooth model of genus $g = (2 d - 1) - δ$ . For $d$ large the map is dominant, so every general genus- $g$ curve is a plane nodal model, and the irreducibility of $V_{d, δ}$ (Harris 1986) gives the classical Severi route to the irreducibility of $M_{g}$ — the image of an irreducible variety under a dominant rational map is irreducible. This is the structural reason the Severi variety belongs to the moduli strand.
Hurwitz formula / Riemann-Hurwitz 04.04.02. The genus drop $g = (2 d - 1) - δ$ is the plane-curve analogue of the ramification bookkeeping in the Hurwitz formula: projecting a degree- $d$ plane curve from a general point to $P^{1}$ realises it as a branched cover, and the Plücker / Riemann-Hurwitz count of branch points reproduces the same genus through the dual relation between nodes, the degree, and the ramification. The conductor / adjoint ideal that governs the Severi tangent space is the curve-on-surface counterpart of the ramification divisor.
Hilbert scheme 04.10.05. The Severi variety is a locally closed subscheme of the Hilbert scheme of plane curves $Hilb^{P} (P^{2}) = P^{N}$ (the Hilbert scheme of degree- $d$ divisors is the complete linear system). The equisingular deformation theory that gives the smooth $N - δ$ -dimensional structure is a sub-functor of the Hilbert-scheme deformation theory cut out by the node-preservation conditions, and the closure $\overline{V}_{d, δ}$ is a constructible subset of the Hilbert scheme whose boundary strata are where the Harris monodromy argument operates.
Intersection pairing on a surface 04.05.06. Adjunction on $P^{2}$ — $2 p_{a} - 2 = (K_{P^{2}} + C) \cdot C$ with the intersection pairing on the surface — produces the arithmetic genus $(2 d - 1)$ that the node count corrects to the geometric genus. The whole Severi dimension and genus bookkeeping rests on the self-intersection $C^{2} = d^{2}$ and the canonical class $K_{P^{2}} = - 3 H$ , so the intersection theory of the ambient surface is the foundational input.
Mikhalkin correspondence / Severi degrees 04.12.05. The Severi degrees $N^{d, δ}$ — the number of $δ$ -nodal degree- $d$ plane curves through $\frac{d ( d + 3 )}{2} - δ$ general points — are the enumerative invariants of the Severi variety, computed classically by the Caporaso-Harris recursion and tropically by the Mikhalkin lattice-path count. The Severi variety is the algebraic parameter space whose degree these node-counts measure; the tropical unit computes the same numbers combinatorially, so the two units are the geometric and combinatorial faces of node-counting.

Historical & philosophical context Master

The study of plane nodal curves as a parameter space goes back to the Italian school and in particular to Francesco Severi, whose 1921 Vorlesungen über algebraische Geometrie ^{[Severi 1921]} introduced what is now called the Severi variety in Appendix F. Severi recognised that the curves of degree $d$ with $δ$ nodes form a family of dimension $\frac{d ( d + 3 )}{2} - δ$ — one parameter spent per node — and that the construction realises curves of every genus $g = (2 d - 1) - δ$ as plane models. Severi asserted both that this family is non-empty of the expected dimension and that it is irreducible. The dimension and non-emptiness claims were essentially correct and were later given fully rigorous proofs (the smoothness and independence-of-node-conditions statement, in the modern conductor-ideal form, is due to Zariski 1982 ^{[Zariski 1982]}). The irreducibility claim, however, rested on an argument that did not hold up: Severi's reasoning that any two $δ$ -nodal curves could be connected had a gap in the analysis of how nodes degenerate, and the question — the Severi problem — remained open for over sixty years.

Joe Harris's 1986 paper On the Severi problem ^{[Harris 1986]} (Invent. Math. 84, 445-461) supplied the missing argument and proved that $V_{d, δ}$ is irreducible for all $d$ and all $δ \leq (2 d - 1)$ . Harris's proof is a degeneration-and-monodromy argument: by pushing curves to the boundary of the Severi variety and analysing the local monodromy of colliding nodes (the tacnode splitting, an $A_{3}$ singularity whose versal deformation interchanges two vanishing cycles), he showed that the monodromy group on the $δ$ nodes is the full symmetric group $S_{δ}$ , which forces the variety to have a single component. This monodromy viewpoint — that irreducibility of a parameter space is equivalent to transitivity of the monodromy on the marked structure — became a paradigm, and it is the same circle of ideas Harris had used with Mumford and others in the study of the moduli space of curves.

The Severi variety sits at the confluence of two traditions. On the moduli side, the normalisation map $V_{d, δ} ⇢ M_{g}$ was studied by Enrico Arbarello and Maurizio Cornalba (1981) ^{[Arbarello-Cornalba 1981]} and is the classical route to the irreducibility of $M_{g}$ , predating and complementing the reduction-mod- $p$ proof of Deligne and Mumford. On the enumerative side, the degrees of the Severi varieties — the counts of nodal curves through the right number of general points — were computed by Lucia Caporaso and Joe Harris in 1998 ^{[Caporaso-Harris 1998]} (Invent. Math. 131, 345-392) by a recursion that degenerates point conditions onto a line, and these Severi degrees include the genus-zero Gromov-Witten invariants of $P^{2}$ as the rational-curve case. The same numbers were later recovered tropically by Grigory Mikhalkin's correspondence theorem, where node-counting becomes a lattice-path count in the Newton polygon, making the Severi variety the algebraic origin of a now-flourishing tropical enumerative geometry. The Diaz-Harris 1988 study ^{[Diaz-Harris 1988]} of the divisor theory of the Severi variety extended the picture to the geometry of its boundary, completing the parameter-space portrait Severi had begun.

Bibliography Master

@article{Harris1986Severi,
  author  = {Harris, Joe},
  title   = {On the {S}everi problem},
  journal = {Inventiones Mathematicae},
  volume  = {84},
  number  = {3},
  year    = {1986},
  pages   = {445--461}
}

@book{Severi1921,
  author    = {Severi, Francesco},
  title     = {Vorlesungen \"uber algebraische {G}eometrie},
  publisher = {B. G. Teubner, Leipzig},
  year      = {1921},
  note      = {Anhang F: families of plane curves with nodes}
}

@article{CaporasoHarris1998,
  author  = {Caporaso, Lucia and Harris, Joe},
  title   = {Counting plane curves of any genus},
  journal = {Inventiones Mathematicae},
  volume  = {131},
  number  = {2},
  year    = {1998},
  pages   = {345--392}
}

@article{DiazHarris1988,
  author  = {Diaz, Steven and Harris, Joe},
  title   = {Geometry of the {S}everi variety},
  journal = {Transactions of the American Mathematical Society},
  volume  = {309},
  number  = {1},
  year    = {1988},
  pages   = {1--34}
}

@article{ArbarelloCornalba1981,
  author  = {Arbarello, Enrico and Cornalba, Maurizio},
  title   = {Su una propriet\`a notevole dei morfismi di una curva a moduli},
  journal = {Bollettino della Unione Matematica Italiana D},
  volume  = {1},
  number  = {6},
  year    = {1981}
}

@book{ACGH1985,
  author    = {Arbarello, Enrico and Cornalba, Maurizio and Griffiths, Phillip A. and Harris, Joseph},
  title     = {Geometry of Algebraic Curves, Volume I},
  publisher = {Springer-Verlag},
  series    = {Grundlehren der mathematischen Wissenschaften},
  volume    = {267},
  year      = {1985}
}

@book{HarrisMorrison1998,
  author    = {Harris, Joe and Morrison, Ian},
  title     = {Moduli of Curves},
  publisher = {Springer-Verlag},
  series    = {Graduate Texts in Mathematics},
  volume    = {187},
  year      = {1998}
}

@article{Zariski1982,
  author  = {Zariski, Oscar},
  title   = {Dimension-theoretic characterization of maximal irreducible algebraic systems of plane nodal curves},
  journal = {American Journal of Mathematics},
  volume  = {104},
  number  = {1},
  year    = {1982},
  pages   = {209--226}
}

@article{Mikhalkin2005,
  author  = {Mikhalkin, Grigory},
  title   = {Enumerative tropical algebraic geometry in $\mathbb{R}^2$},
  journal = {Journal of the American Mathematical Society},
  volume  = {18},
  number  = {2},
  year    = {2005},
  pages   = {313--377}
}

Prerequisites

04.10.01
04.04.02
04.10.05
04.05.06
04.12.05

Tier anchors

beginner: Harris-Morrison *Moduli of Curves* (Springer GTM 187, 1998) Ch. 1, §A-B informal opening — the space of plane curves of fixed degree as a projective space, and the condition of acquiring a node as a single equation cutting down the dimension by one
intermediate: Harris-Morrison *Moduli of Curves* (Springer GTM 187, 1998) Ch. 1 (parameter spaces, the Severi variety example) and Ch. 3 (the map to $\mathcal{M}_g$); Arbarello-Cornalba-Griffiths-Harris *Geometry of Algebraic Curves* Vol. I Ch. I, App. A (families of plane nodal curves and the tangent-space computation via the conductor)
master: Harris 1986 *On the Severi problem* (Invent. Math. 84, 445-461, irreducibility); Severi 1921 *Vorlesungen über algebraische Geometrie* (Anhang F, the original existence/dimension claim); Caporaso-Harris 1998 *Counting plane curves of any genus* (Invent. Math. 131, 345-392, the degree recursion); Diaz-Harris 1988 *Geometry of the Severi variety* (Trans. AMS 309, divisor theory); Arbarello-Cornalba 1981 *Su una proprietà notevole dei morfismi di una curva a moduli* (Boll. UMI, the family-to-moduli map)

References

Harris 1986 — On the Severi problem · Invent. Math. 84, 445-461; the proof that the Severi variety $V_{d,\delta}$ is irreducible, via degeneration to the boundary and the resulting full monodromy on the nodes
Severi 1921 — Vorlesungen über algebraische Geometrie · Anhang F (Appendix F); the original definition of the variety of nodal plane curves, the dimension count $d(d+3)/2 - \delta$, and the (incomplete) irreducibility claim
Caporaso-Harris 1998 — Counting plane curves of any genus · Invent. Math. 131, 345-392; the recursion for the Severi degrees $N^{d,\delta}$ (the enumerative invariants), with tangency conditions along a line
Diaz-Harris 1988 — Geometry of the Severi variety · Trans. Amer. Math. Soc. 309, 1-34; the Picard group and divisor classes of the Severi variety, and degenerations to cuspidal and tacnodal loci
Arbarello-Cornalba 1981 — Su una proprietà notevole dei morfismi di una curva algebrica a moduli · Boll. Un. Mat. Ital. D (6) 1; the rational map from a family of nodal curves to $\mathcal{M}_g$ and the dominance / generic-smoothness statements
Arbarello-Cornalba-Griffiths-Harris — Geometry of Algebraic Curves Volume I · Springer Grundlehren 267, 1985; Ch. I and App. A (the tangent space to the Severi variety as $H^0(C, \mathcal{O}_C(d) \otimes \mathcal{I}_N)$, the adjoint / conductor ideal of the nodes)
Harris-Morrison — Moduli of Curves · Springer GTM 187, 1998; Ch. 1 (parameter spaces and the Severi-variety example) and Ch. 3 (the map to $\mathcal{M}_g$ and the geometry of the boundary)
Zariski 1982 — Dimension-theoretic characterization of maximal irreducible algebraic systems of plane nodal curves · Amer. J. Math. 104, 209-226; the smoothness and expected-dimension theorem for $V_{d,\delta}$ via the conductor and the regularity of the node-imposing conditions

Estimated time

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