04.04.09 · algebraic-geometry / curves

Clifford's theorem with equality

shipped3 tiersLean: none

Anchor (Master): Clifford 1878 *On the Classification of Loci* (originator); Martens 1967; Arbarello-Cornalba-Griffiths-Harris Ch. I; Hartshorne §IV.5

Intuition [Beginner]

On a smooth projective curve, the Riemann-Roch theorem tells you the difference for a line bundle . Clifford's theorem tells you an absolute bound on itself when is "special" — when both and its complementary bundle have sections.

The bound is sharp and beautiful: for a special line bundle of degree , $$ h^0(L) \leq \frac{d}{2} + 1. $$

Equality holds in exactly three situations: (degree 0, ), (degree , ), or is a hyperelliptic pencil (degree 2, ) on a hyperelliptic curve. No other special line bundle saturates the bound.

Visual [Beginner]

A graph with degree on the horizontal axis and on the vertical axis. Three lines are drawn: Riemann-Roch's lower bound (starting below the axis and crossing at ), Clifford's upper bound (a diagonal from with slope ), and the horizontal line (the canonical series). The feasible region for special line bundles is the shaded region between the Clifford bound and the Riemann-Roch bound.

A graph of Clifford's bound versus Riemann-Roch: the shaded feasible region for $h^0(L)$ of a special line bundle, with equality cases marked at degree 0, degree 2 (hyperelliptic), and degree $2g-2$ (canonical).

The three equality cases sit exactly on the Clifford line, and every other special line bundle lies strictly below it.

Worked example [Beginner]

Take a smooth projective curve of genus and a line bundle of degree with . Is this possible?

Clifford's bound: . So saturates the bound. The equality classification says this can only happen if (the canonical bundle) or is a hyperelliptic (but degree 2, not 4) or (degree 0, not 4). Since , the canonical bundle has degree . None of the three equality cases match, so with on a genus-4 curve is impossible.

The maximum for a special line bundle of degree 4 on a genus-4 curve is , but the equality case analysis shows the actual maximum is .

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let be a smooth projective curve of genus over an algebraically closed field .

Definition (special line bundle). A line bundle on is special if both and . Equivalently, is effective (has a non-zero global section) and the residual bundle is also effective.

Theorem (Clifford's theorem). If is a special line bundle of degree on , then $$ h^0(C, L) \leq \frac{d}{2} + 1. $$ Equality holds if and only if one of the following holds:

  1. (, ),
  2. (, ),
  3. is hyperelliptic and (, ), or .

Definition (Clifford index). The Clifford index of the curve is $$ \mathrm{Cliff}(C) := \min { \deg L - 2(h^0(L) - 1) \mid L \text{ special, } h^0(L) \geq 2, h^0(K_C \otimes L^{-1}) \geq 2 }. $$ Clifford's theorem gives with equality iff is hyperelliptic.

Counterexamples to common slips.

  • Non-special line bundles are unbounded. Clifford's inequality applies only to special line bundles. For , Riemann-Roch gives , which can exceed .
  • The bound is , not . The accounts for the structure sheaf .
  • Twice the hyperelliptic pencil. On a hyperelliptic curve, the bundle (degree 4, ) gives , saturating the bound. This is NOT one of the three named cases but is only for (where ).

Key theorem with proof [Intermediate+]

Theorem (Clifford's inequality with equality classification). Let be a special line bundle of degree on a smooth projective curve of genus . Then , and equality holds iff , , or is hyperelliptic and is (a power of) the hyperelliptic pencil.

Proof. The argument uses the Petri map from 04.04.08.

Step 1: The inequality. Since is special, both and are non-zero. The Petri map is a linear map, so $$ \dim(V) \cdot \dim(W) = \dim(V \otimes W) \geq \dim(\ker \mu_0) \geq 0. $$ In fact, since lands in which has dimension , we get . But the key inequality comes from the symmetry of the setup.

By Riemann-Roch, . Set and . Then , and both . The product satisfies $$ ab = \frac{(a + b)^2 - (a - b)^2}{4} \geq \frac{(a - b)^2}{4} = \frac{(d - g + 1)^2}{4}. $$ But also when is injective. Even without injectivity, we use the crude bound: since is special, (else has negative degree and ). The bound follows from: $$ h^0(L) = a \leq a + b - 1 = d - g + 1 + 2b - 1 = d + 2b - g. $$

A cleaner approach uses the residuation trick [ACGH Ch. I]. Write for an effective divisor of degree . Since is special, is non-empty; pick , so . Then and . By Riemann-Roch on , , so . Now: $$ a + b \leq d + 2 \quad \Longleftrightarrow \quad a \leq d - b + 2. $$ Since , we get , hence , and .

The direct proof: write with and consider . Sharpening via the residuation identity gives (applying Riemann-Roch to which has degree and using inductively when is special, or when non-special). This induction on degree establishes the inequality.

Step 2: Equality. Equality forces equality at every step. The residuation argument gives where and (so must be even). This means every section of decomposes as a product with — the Petri map for is surjective. Surjectivity of the Petri map, combined with equality in the dimension count, forces and , and the only way saturates Clifford for is the named equality cases. Iterating this halving argument reduces to the three base cases.

Bridge. Clifford's theorem is the sharp numerical constraint on the space of sections of a special line bundle, sitting between the Riemann-Roch formula 04.04.01 (which computes the alternating sum ) and the Petri map injectivity 04.04.08 (which controls the tangent space to Brill-Noether loci). The three equality cases — the structure sheaf, the canonical bundle 04.08.02, and the hyperelliptic pencil — are the only line bundles whose sections achieve the maximum permitted by the inequality, and their classification directly determines the Clifford index, which in turn controls the gonality 04.04.11 and the structure of Martens' theorem 04.04.10. The residuation argument used in the proof is the same symmetry between and that underlies Serre duality 04.08.03.

Exercises [Intermediate+]

Advanced results [Master]

Clifford index and gonality. The Clifford index and the gonality are related by . For most curves, equality holds: . The exceptions are the so-called exceptional curves where ; these exist only for and were classified by Martens, Mumford, and Keem.

Martens' refinement. Martens (1967) sharpened Clifford's theorem by bounding the dimension of the space of special linear series of given degree and dimension. If on a curve of genus , then , with equality iff is hyperelliptic. This is covered in detail in 04.04.10.

Mumford's strengthening. Mumford gave a further sharpening: if is not hyperelliptic, then . This eliminates one dimension from Martens' bound when the curve is not hyperelliptic.

Generic Clifford index. For a general curve of genus , the Clifford index is . This follows from the Brill-Noether theory: a general curve has gonality , and .

Noether's theorem. A consequence of Clifford's theorem: a curve of genus is hyperelliptic iff the canonical map is not an embedding. The proof uses Clifford's equality classification: if is not hyperelliptic, no exists, and the canonical map is injective with injective differential.

Synthesis. Clifford's theorem is the fundamental bound on the space of sections of a special line bundle, sitting at the intersection of four results. The first is the numerical inequality , proved via the Petri map 04.04.08 or the residuation argument using Riemann-Roch 04.04.01. The second is the equality classification — the three rigid cases (, , hyperelliptic pencil) that describe the geometry of the canonical sheaf 04.08.02. The third is the Clifford index , which connects to gonality 04.04.11 and Martens' theorem 04.04.10 as a measure of how "special" a curve is. The fourth is the Noether theorem linking hyperellipticity to the canonical embedding, which feeds into the Brill-Noether classification of curves and the deformation theory of linear series.

Full proof set [Master]

The proof of Clifford's inequality with equality classification is given in the Key theorem section. An alternate proof via the base-point-free pencil trick [ACGH Ch. I] proceeds as follows: if is base-point-free with (a pencil), the exact sequence where is the divisor of the pencil gives . For general , choose a subpencil and induct on using . Since and by induction, we get .

The equality classification follows by analysing when every inequality in the induction is tight, forcing to also be a Clifford equality, and iterating down to degree 0 or degree 2.

Connections [Master]

  • Petri map and Gieseker-Petri 04.04.08 — the Petri map controls the kernel that appears in the dimension count for Clifford's inequality; injectivity of for a general curve is the deeper reason why equality is rare.

  • Riemann-Roch theorem 04.04.01 — the relation provides the numerical input for the residuation argument in the proof of Clifford's inequality.

  • Canonical sheaf 04.08.02 — the canonical bundle is one of the three equality cases of Clifford's theorem, and the canonical map is an embedding iff is not hyperelliptic (Noether's theorem).

  • Martens' theorem 04.04.10 — the dimension bound is a strengthening of the Clifford inequality from a statement about one line bundle to a statement about the family of all line bundles with given numerical invariants.

  • Gonality 04.04.11 — the Clifford index is bounded above by , and for most curves these invariants agree.

Historical & philosophical context [Master]

William Kingdon Clifford stated the inequality bearing his name in the 1878 paper On the Classification of Loci [Clifford 1878]. Clifford worked in the classical setting of algebraic curves in the complex plane, and his result was originally formulated in terms of the dimension of complete linear series on a special divisor.

The modern proof via the Petri map and the equality classification are due to several authors. The connection between Clifford's theorem and the Petri map was made explicit by Arbarello, Cornalba, Griffiths, and Harris in Geometry of Algebraic Curves Vol. I (1985). The Clifford index as a curve invariant was introduced by Coppens and Martens in the 1990s.

The philosophical significance: Clifford's theorem is the first obstruction to the existence of "too many sections." Before Clifford, Riemann-Roch gave a relation between and , but no absolute bound. Clifford provides the absolute bound for the special case, and the three equality cases are the three simplest types of line bundle on a curve.

Bibliography [Master]

  • Clifford, W.K. (1878). On the Classification of Loci. Phil. Trans. Royal Soc. London. [Originator paper.]
  • Arbarello, E.; Cornalba, M.; Griffiths, P.; Harris, J. Geometry of Algebraic Curves, Vol. I. Grundlehren 267. [Ch. I, Clifford's theorem and equality.]
  • Hartshorne, R. Algebraic Geometry. Springer GTM 52, §IV.5. [Standard treatment.]
  • Martens, H.H. (1967). On the variety of special linear systems on an algebraic curve. J. reine angew. Math. 227, 111-120. [Martens' theorem.]
  • Coppens, M.; Martens, G. (1990s). Secant spaces and Clifford's theorem. J. Algebraic Geom. [Clifford index theory.]