The Enriques-Babbage-Petri Theorem on Canonical Curves

Intuition Beginner

Every smooth curve of genus $g\ge 2$ that is not hyperelliptic comes with a god-given way to sit inside a projective space of dimension $g-1$. You build it from the curve's own differential forms: a curve of genus $g$ carries exactly $g$ independent ones, and evaluating all of them at each point of the curve traces out a copy of the curve in ${\mathbb{P}}^{g-1}$. This copy is called the canonical curve.

Once a curve lives in projective space, you can ask which polynomial equations vanish on it. The Enriques-Babbage-Petri theorem gives a clean answer for the canonical curve: in almost every case the equations of lowest degree, the quadrics (degree-two equations), already cut out the curve completely. You never need anything more complicated.

There are exactly two families of exceptions, and they are famous: curves that admit a three-to-one map to a line, and curves of genus six that are smooth plane quintics. For these the quadrics fall short, carving out a larger surface instead of the curve, and you must add degree-three equations to pin the curve down.

Visual Beginner

Picture the canonical curve as a loop sitting inside ${\mathbb{P}}^{g-1}$. Around it, draw the family of quadric hypersurfaces that contain it: each is a single degree-two equation, and together they form a web. In the typical case the web is tight enough that the only points lying on every quadric are the points of the curve itself.

The side panel shows the exceptions: for a trigonal curve or a plane quintic the web of quadrics is too loose and traces out a whole ruled surface, a scroll, that contains the curve as a sub-piece. Only by adding degree-three equations do you recover the curve.

Worked example Beginner

Take a curve of genus $g=4$ that is not hyperelliptic. Its canonical model lives in ${\mathbb{P}}^3$, ordinary three-dimensional projective space. How many quadric equations pass through it?

The space of all degree-two polynomials in four variables has dimension $\left(\genfrac{}{}{0ex}{}{4+1}{2}\right)=10$. The number of these that vanish on the canonical curve is $10$ minus the number of independent values they can take on the curve, and a count using the curve's differentials shows exactly one quadric contains it.

So a genus-4 canonical curve lies on a single quadric surface in ${\mathbb{P}}^3$, plus one cubic surface, and it is the intersection of the two. This matches the theorem: genus 4 curves can be trigonal, and indeed a genus-4 canonical curve always carries a degree-3 map to a line, so a cubic equation is genuinely needed alongside the quadric. The curve is a $(3,3)$ curve on the quadric.

What this shows: the abstract statement about quadrics turns into concrete counting of polynomials, and the answer tells you the precise shape of the equations.

Check your understanding Beginner

Formal definition Intermediate+

Let $C$ be a smooth projective curve of genus $g\ge 2$ over an algebraically closed field $k$, with canonical line bundle $K_C$. By Riemann-Roch 04.04.01, $h^0(C,K_C)=g$ and $\mathrm{deg}{K}_C=2g-2$.

Definition (canonical map and canonical curve). Choose a basis ${\omega }_1,\dots ,{\omega }_g$ of $H^0(C,K_C)$. The canonical map is $$ \phi_K : C \longrightarrow \mathbb{P}^{g-1}, \qquad p \longmapsto [\omega_1(p) : \cdots : \omega_g(p)]. $$ When $C$ is non-hyperelliptic ($g\ge 3$), $K_C$ is very ample and ${\varphi }_K$ is a closed embedding; its image, a nondegenerate curve of degree $2g-2$ in ${\mathbb{P}}^{g-1}$, is the canonical curve. When $C$ is hyperelliptic, ${\varphi }_K$ factors through the degree-2 map to ${\mathbb{P}}^1$ and the image is a rational normal curve, so the canonical model is not an embedding 06.02.03.

Definition (homogeneous ideal and the canonical ring). Let $S=k[x_0,\dots ,x_{g-1}]$ be the homogeneous coordinate ring of ${\mathbb{P}}^{g-1}$ and let $I_C\subset S$ be the homogeneous ideal of ${\varphi }_K(C)$. The canonical ring is $R(K_C)={⨁}_{n\ge 0}{H}^0(C,K_C^{\otimes n})$, and there is a natural graded surjection $S/I_C\twoheadrightarrow R(K_C)$ once $C$ is projectively normal, that is, once each multiplication map ${\mathrm{Sym}}^n{H}^0(K_C)\to H^0(K_C^{\otimes n})$ is surjective.

Definition (Petri's quadrics). For a non-hyperelliptic $C$, the degree-2 part $(I_C{)}_2$ is the kernel of ${\mathrm{Sym}}^2{H}^0(K_C)\to H^0(K_C^{\otimes 2})$. Petri exhibits an explicit basis: fixing a point $p$ and a general $g_k^1$ of minimal degree $k$ (the gonality), one writes quadrics $q_{ij}$ indexed by a tableau of multiplication relations among adapted bases of $H^0(K_C)$, so that $\dim (I_C{)}_2=(\genfrac{}{}{0ex}{}{g-1}{2})-(2g-5)=(\genfrac{}{}{0ex}{}{g-2}{2})$.

Counterexamples to common slips.

• Hyperelliptic curves have no canonical embedding. The canonical map is two-to-one onto a rational normal curve; the quadric-generation statement is vacuous for them and they are excluded from the theorem's hypotheses.
• Quadrics need not cut out the curve scheme-theoretically in the exceptions. For trigonal curves and plane quintics the quadrics cut out a surface of minimal degree, not the curve; the curve is recovered only after adjoining cubics.
• Projective normality is part of the statement, not an assumption. The theorem includes that ${\varphi }_K(C)$ is projectively normal for every non-hyperelliptic $C$ (Max Noether's theorem); one does not get to assume it.
• Generation versus scheme-theoretic intersection. "Generated by quadrics" means the ideal $I_C$ is generated in degree 2; the weaker "cut out by quadrics" allows higher-degree generators inside the saturation. The theorem asserts the stronger ideal-generation statement.

Key theorem with proof Intermediate+

Theorem (Enriques-Babbage-Petri). Let $C$ be a smooth non-hyperelliptic projective curve of genus $g\ge 3$, canonically embedded in ${\mathbb{P}}^{g-1}$. Then ${\varphi }_K(C)$ is projectively normal, and its homogeneous ideal $I_C$ is generated by quadrics, unless $C$ is trigonal (carries a $g_3^1$) or $C$ is a smooth plane quintic ($g=6$). In those two exceptional cases the quadrics in $I_C$ generate the homogeneous ideal of a surface of minimal degree — a rational normal scroll, respectively the Veronese surface $v_2({\mathbb{P}}^2)$ — and $I_C$ is generated by these quadrics together with cubics.

Proof (following Saint-Donat 1973). Projective normality is Max Noether's theorem: the multiplication map ${\mathrm{Sym}}^n{H}^0(K_C)\to H^0(K_C^{\otimes n})$ is surjective for all $n\ge 1$. For $n=1$ this is an identity; the base-point-free pencil trick 04.04.08 applied to a general $g_k^1$ reduces surjectivity for $n=2$ to a cohomology vanishing that holds because $C$ is non-hyperelliptic, and an induction using the same trick handles $n\ge 3$ ^{[Arbarello-Cornalba-Griffiths-Harris Ch. III]}.

Step 1: the quadrics cut out a scroll or the curve. Petri analyses the linear system $(I_C{)}_2$ of quadrics through the canonical curve. Let $Z$ be the scheme cut out by all these quadrics. A dimension count using projective normality gives $\dim (I_C{)}_2=(\genfrac{}{}{0ex}{}{g-2}{2})$, and one shows $Z$ is either $C$ itself or a surface $S\supset C$ of minimal degree $g-2$ in ${\mathbb{P}}^{g-1}$ — a rational normal scroll, or the Veronese $v_2({\mathbb{P}}^2)$ when $g=6$ ^{[Saint-Donat 1973]}.

Step 2: identifying when the scroll appears. The quadrics cut out a two-dimensional $Z=S$ exactly when $C$ lies on such a surface of minimal degree. By the geometry of scrolls, $C\subset S$ forces a pencil of low degree: the ruling of the scroll cuts $C$ in a $g_3^1$ (trigonal case), and the Veronese case forces $C$ to be the image of a plane quintic under the second Veronese map. Conversely a trigonal curve or plane quintic does lie on the corresponding minimal surface.

Step 3: Petri's relations supply the generators. In the generic case $Z=C$, Petri writes the ideal explicitly. The quadrics $q_{ij}$ satisfy relations of the form $x_a{q}_{ij}-x_b{q}_{i^{\prime }{j}^{\prime }}=(\text{cubic terms})$; when $C$ is neither trigonal nor a plane quintic the cubic terms are themselves combinations of the $x_c{q}_{de}$, so $I_C$ is generated in degree 2. In the two exceptional cases the cubic terms include a genuinely new cubic $\Gamma$ not in the ideal generated by the quadrics, and $I_C=(q_{ij},\Gamma )$ needs that cubic. This is Petri's analysis of the multiplicative structure of the canonical ring. $\square$

Bridge. This theorem builds toward the homological study of canonical curves and appears again in Green's conjecture, and the foundational reason it works is that the canonical bundle is the one line bundle whose sections multiply to fill the whole canonical ring — projective normality is exactly this multiplicative saturation, and the central insight is that the failure of quadric generation is detected by, and only by, the presence of a low-degree pencil. This is exactly the bridge from the extrinsic picture (equations in ${\mathbb{P}}^{g-1}$) to the intrinsic Brill-Noether picture (the gonality and the Clifford index 04.04.09): the trigonal and plane-quintic exceptions are precisely the curves of Clifford index 1, which is dual to saying their canonical syzygies degenerate one step early. The quadric-generation statement generalises to Green's conjecture, where the degree at which generation or higher syzygies first fail is controlled by the Clifford index, putting these together into a single numerical invariant that governs the whole free resolution of the canonical ring.

Exercises Intermediate+

Advanced results Master

Petri's explicit relations and the cubic. Petri's original 1922 analysis writes adapted bases of $H^0(K_C)$ using the gonality pencil $g_k^1$. With coordinates split as $x_0,x_1$ (pulled back from the pencil) and $y_1,\dots ,y_{g-2}$, the quadrics take the shape $q_{ij}=y_i{y}_j-(\text{linear in }y)-(\text{terms in }x)$. The associativity of multiplication in the canonical ring forces relations $R_{ij}$ among the $q_{ij}$, and each $R_{ij}$ has the form $x_0\cdot (\cdots )-x_1\cdot (\cdots )+f_{ij}$ where $f_{ij}$ is a cubic. For a curve that is neither trigonal nor a plane quintic, every $f_{ij}$ lies in the ideal generated by the quadrics, so no new generator appears and $I_C=(q_{ij})$. Exactly in the trigonal and plane-quintic cases one $f_{ij}$ escapes, producing the irreducible cubic generator. This is the precise mechanism behind the dichotomy.

Surfaces of minimal degree and the two exceptions. A nondegenerate surface in ${\mathbb{P}}^{g-1}$ has degree at least $g-2$; equality holds only for rational normal scrolls and, when $g=6$, the Veronese surface $v_2({\mathbb{P}}^2)\subset {\mathbb{P}}^5$ (Del Pezzo-Bertini, also driving the extremal-curve structure of Castelnuovo's bound 04.04.04). When the quadrics through $C$ fail to cut out $C$, they cut out one of these minimal surfaces. The scroll case is trigonal: its ruling induces the $g_3^1$. The Veronese case is the plane quintic: a smooth plane quintic has genus $6$, its canonical bundle is ${\mathcal{O}}_{{\mathbb{P}}^2}(2){|}_C$, and the canonical embedding factors through $v_2({\mathbb{P}}^2)$. These are the only two ways the quadric-generation can break.

Green's conjecture as the generalisation. The quadric-generation statement is the first instalment of a complete homological story. Encode the minimal free resolution of the canonical ring $R(K_C)$ over $S$ by its Koszul cohomology groups $K_{p,q}(C,K_C)$. Quadric generation is the vanishing $K_{1,2}=0$ failing exactly on the exceptional curves; Green's conjecture ^{[Green 1984]} asserts the sharp general statement: $K_{p,1}(C,K_C)=0$ for $p<\mathrm{Cliff}(C)$ and $K_{p,1}\ne 0$ for $p=\mathrm{Cliff}(C)$, where $\mathrm{Cliff}(C)$ is the Clifford index 04.04.09. The Enriques-Babbage-Petri theorem is the case $\mathrm{Cliff}(C)=1$ (gonality $3$ or plane quintic) versus $\mathrm{Cliff}(C)\ge 2$.

Voisin's theorem. Green's conjecture is open in general but is a theorem for the generic curve of each gonality: Voisin ^{[Voisin 2002/2005]} proved the generic Green conjecture by a delicate analysis of Koszul cohomology on the K3 surfaces containing the curve, in even genus first and then odd genus. This is the modern frontier; the Enriques-Babbage-Petri theorem is its classical $p=1$ shadow, and the K3-surface technique echoes Lazarsfeld's degeneration-free proof of Petri injectivity 04.04.16.

Synthesis. Putting these together, the Enriques-Babbage-Petri theorem is one statement read at three depths, and the central insight is that the equations of a canonical curve are governed entirely by a single Brill-Noether invariant. The foundational reason quadrics suffice is projective normality — the multiplicative saturation of the canonical ring — and the only obstruction is a low-degree pencil, so the failure locus is exactly the curves of Clifford index $1$. This is dual to the syzygy picture: quadric generation is the vanishing of the first Koszul group $K_{1,2}$, and Green's conjecture generalises the dichotomy into a sharp threshold $p=\mathrm{Cliff}(C)$ for the linear strand of the resolution, so the classical theorem and the modern conjecture are the same statement at $p=1$ and at $p=\mathrm{Cliff}(C)$ respectively. The bridge is the Clifford index, which simultaneously measures the gonality-type pencils that break quadric generation 04.04.09, the extremal-surface geometry of the scroll 04.04.04, and the deformation-theoretic Petri injectivity that makes the canonical ring as rigid as possible 04.04.08. Voisin's resolution of the generic case via K3 surfaces is exactly the degeneration-free strategy that resolved the Petri conjecture, closing the loop between the equations of the curve and the cohomology of the surfaces that contain it.

Full proof set Master

The Key theorem section proves the main dichotomy following Saint-Donat. The following proposition isolates the projective-normality input — Max Noether's theorem — for the canonical curve, since it is the load-bearing first step.

Proposition (Max Noether: projective normality of the canonical curve). Let $C$ be a smooth non-hyperelliptic curve of genus $g\ge 3$. Then for every $n\ge 1$ the multiplication map $$ \mu_n : \mathrm{Sym}^n H^0(C, K_C) \longrightarrow H^0(C, K_C^{\otimes n}) $$ is surjective; equivalently the canonical curve is projectively normal.

Proof. For $n=1$ the map is the identity. Take $n=2$. Choose a general pencil, that is a base-point-free $L=g_k^1$ of degree equal to the gonality $k$, so $h^0(L)=2$. The base-point-free pencil trick 04.04.08 applied to $L$ gives an exact sequence $$ 0 \to H^0(K_C \otimes L^{-1}) \to H^0(L) \otimes H^0(K_C) \xrightarrow{\ \mu\ } H^0(K_C \otimes L), $$ and chasing the analogous sequence for $K_C$ in place of $L$, the surjectivity of ${\mu }_2$ is equivalent to the vanishing $H^1(C,K_C\otimes L^{-1})=0$ for a suitable factorisation. Since $C$ is non-hyperelliptic, $K_C\otimes L^{-1}$ has $h^1=h^0(L\otimes K_C^{-1}\otimes K_C)=h^0(L\otimes \mathcal{O})$ controlled by Clifford's theorem 04.04.09: the strict Clifford inequality for non-hyperelliptic curves forces the relevant cohomology to vanish, giving surjectivity of ${\mu }_2$. For $n\ge 3$, factor ${\mu }_n$ through ${\mu }_2\otimes \mathrm{id}$ and $H^0(K_C^{\otimes 2})\otimes H^0(K_C)\to H^0(K_C^{\otimes 3})$; the latter is surjective because $K_C^{\otimes 2}$ is nonspecial of degree $4g-4\ge 2g+1$ and so its multiplication against the very ample $K_C$ is surjective by the standard projective-normality criterion for high-degree bundles. Induction on $n$ completes the argument. $\square$

The full Petri analysis of the cubic relations $R_{ij}$ — including the verification that exactly one cubic survives in the trigonal and plane-quintic cases — is carried out in Saint-Donat's paper ^{[Saint-Donat 1973]} and reproduced in ACGH Ch. III ^{[Arbarello-Cornalba-Griffiths-Harris Ch. III]}.

Connections Master

• Petri map and Gieseker-Petri 04.04.08 — the base-point-free pencil trick that proves projective normality here is the same cohomological device that proves injectivity of the Petri map ${\mu }_0$; both are Petri's, and both turn questions about a curve's linear series into the multiplicative structure of the canonical ring. The Gieseker-Petri injectivity is the deformation-theoretic counterpart of the quadric-generation rigidity stated in this unit.

• Clifford's theorem 04.04.09 — the dichotomy is governed by the Clifford index: trigonal curves and plane quintics are exactly the non-hyperelliptic curves of Clifford index $1$, and the strict Clifford inequality is the cohomological input that makes the quadrics generate for index $\ge 2$. Green's conjecture promotes Clifford's invariant from a bound on $h^0$ to the precise threshold controlling the canonical syzygies.

• Castelnuovo's genus bound 04.04.04 — the surfaces of minimal degree that the exceptional quadrics cut out, the rational normal scrolls and the Veronese, are exactly the minimal-degree surfaces appearing in Castelnuovo's extremal-curve structure theorem; the Del Pezzo-Bertini classification is shared input, tying the equations of canonical curves to the genus ceiling for space curves.

• Riemann-Roch for curves 04.04.01 — every dimension count here, from $h^0(K_C)=g$ fixing the ambient ${\mathbb{P}}^{g-1}$ to $h^0(K_C^{\otimes 2})=3g-3$ fixing the number of quadrics, is a Riemann-Roch computation; the quadric count $\left(\genfrac{}{}{0ex}{}{g-2}{2}\right)$ is a direct Riemann-Roch consequence of projective normality.

• Canonical embedding via Riemann's existence theorem 06.02.03 — the very-ampleness of $K_C$ on a non-hyperelliptic curve, which makes ${\varphi }_K$ an embedding in the first place, is the canonical-embedding theorem; this unit asks what equations the resulting embedded curve satisfies.

Historical & philosophical context Master

The structure of the equations defining a canonical curve was pursued by the Italian school. Federigo Enriques announced in 1919 that the canonical curve is, outside special cases, the intersection of the quadrics through it ^{[Enriques 1919]}, but his argument had a gap in handling the exceptional families. Karl Petri, in his 1922 paper Über die invariante Darstellung algebraischer Funktionen (Math. Ann. 88, 242-289) ^{[Petri 1922]}, gave the first thorough treatment, writing down the explicit quadrics and the cubic relations among them and pinpointing the trigonal and plane-quintic exceptions through his analysis of the multiplicative structure of the canonical ring — the same circle of ideas that produced his celebrated Petri map.

Dennis Babbage in 1939 ^{[Babbage 1939]} returned to Enriques' argument and supplied the correction handling the scroll cases, which is why the result carries all three names. The modern, fully scheme-theoretic proof is due to Bernard Saint-Donat (1973) ^{[Saint-Donat 1973]}, whose paper On Petri's analysis of the linear system of quadrics through a canonical curve (Math. Ann. 206) recast Petri's classical computation in the language of projective normality and minimal-degree surfaces, and it is Saint-Donat's exposition that underlies the account in Arbarello-Cornalba-Griffiths-Harris.

The philosophical arc of the subject is the passage from equations to syzygies. Mark Green's 1984 Koszul-cohomology conjecture ^{[Green 1984]} reframed quadric generation as the bottom rung of a ladder of higher syzygies all governed by the Clifford index, and Claire Voisin's proof of the generic case ^{[Voisin 2002/2005]} via K3 surfaces realised that vision for the general curve. What began as a nineteenth-century question about the polynomials through a curve became, a century later, a statement about the homological complexity of its canonical ring — the equations were only the first chapter.

Bibliography Master

• Petri, K. (1922). Über die invariante Darstellung algebraischer Funktionen mittels Thetafunktionen. Mathematische Annalen 88, 242-289. [Explicit quadric and cubic relations; the exceptions.]
• Enriques, F. (1919). Sulle curve canoniche di genere $p$ nello spazio a $p-1$ dimensioni. Rend. Accad. Sci. Bologna 23. [First announcement of quadric generation.]
• Babbage, D. W. (1939). A note on the quadrics through a canonical curve. Journal of the London Mathematical Society 14, 310-315. [Correction of the Enriques gap.]
• Saint-Donat, B. (1973). On Petri's analysis of the linear system of quadrics through a canonical curve. Mathematische Annalen 206, 157-175. [Modern proof.]
• Green, M. (1984). Koszul cohomology and the geometry of projective varieties. Journal of Differential Geometry 19, 125-171. [Green's conjecture.]
• Voisin, C. (2002, 2005). Green's generic syzygy conjecture for curves of even/odd genus lying on a K3 surface. J. Eur. Math. Soc. 4, 363-404; Compositio Math. 141, 1163-1190. [Generic Green conjecture.]
• Arbarello, E.; Cornalba, M.; Griffiths, P.; Harris, J. (1985). Geometry of Algebraic Curves, Vol. I. Grundlehren 267, Springer. [Ch. III §3: quadrics through a canonical curve.]