TheoremComplete

Riemann–Roch for Curves

The Riemann–Roch theorem is the central result in the theory of algebraic curves, relating the dimension of the space of meromorphic functions with prescribed poles to the genus of the curve. It underlies virtually everything in curve theory: embeddings, classification, moduli, and special divisors.


Statement

Theorem4.1Riemann–Roch

Let CC be a smooth projective curve of genus gg over an algebraically closed field kk, and let DD be a divisor on CC. Then:

β„“(D)βˆ’β„“(KCβˆ’D)=deg⁑Dβˆ’g+1,\ell(D) - \ell(K_C - D) = \deg D - g + 1,

where β„“(D)=dim⁑kH0(C,O(D))\ell(D) = \dim_k H^0(C, \mathcal{O}(D)) is the dimension of the Riemann–Roch space, and KCK_C is a canonical divisor (deg⁑KC=2gβˆ’2\deg K_C = 2g - 2, β„“(KC)=g\ell(K_C) = g).

RemarkEquivalent formulations

Using Serre duality, β„“(KCβˆ’D)=dim⁑H1(C,O(D))\ell(K_C - D) = \dim H^1(C, \mathcal{O}(D)), so Riemann–Roch equivalently says:

Ο‡(O(D))=h0(O(D))βˆ’h1(O(D))=deg⁑D+1βˆ’g.\chi(\mathcal{O}(D)) = h^0(\mathcal{O}(D)) - h^1(\mathcal{O}(D)) = \deg D + 1 - g.

The Euler characteristic of a line bundle depends only on its degree and the genus, not on the specific divisor class.

For D=0D = 0: β„“(0)βˆ’β„“(KC)=0βˆ’g+1\ell(0) - \ell(K_C) = 0 - g + 1, giving 1βˆ’g=1βˆ’g1 - g = 1 - g βœ“ (with β„“(0)=1\ell(0) = 1 and β„“(KC)=g\ell(K_C) = g).


Basic consequences

Theorem4.2Riemann's inequality (weak form)

For any divisor DD on a curve of genus gg:

β„“(D)β‰₯deg⁑Dβˆ’g+1\ell(D) \geq \deg D - g + 1

(since β„“(KCβˆ’D)β‰₯0\ell(K_C - D) \geq 0). Equality holds when β„“(KCβˆ’D)=0\ell(K_C - D) = 0, which happens whenever deg⁑D>2gβˆ’2\deg D > 2g - 2 (since then deg⁑(KCβˆ’D)<0\deg(K_C - D) < 0, forcing β„“(KCβˆ’D)=0\ell(K_C - D) = 0).

ExampleLarge degree: exact formula

If deg⁑D>2gβˆ’2\deg D > 2g - 2, then β„“(KCβˆ’D)=0\ell(K_C - D) = 0 so:

β„“(D)=deg⁑Dβˆ’g+1.\ell(D) = \deg D - g + 1.

This is a complete, explicit formula. For instance:

  • g=0g = 0: β„“(D)=deg⁑D+1\ell(D) = \deg D + 1 for all deg⁑Dβ‰₯0\deg D \geq 0. So β„“(nβ‹…[P])=n+1\ell(n \cdot [P]) = n + 1, the space of polynomials of degree ≀n\leq n on P1\mathbb{P}^1.
  • g=1g = 1: β„“(D)=deg⁑D\ell(D) = \deg D for deg⁑Dβ‰₯2\deg D \geq 2. In particular, β„“(2P)=2\ell(2P) = 2, β„“(3P)=3\ell(3P) = 3.
  • g=2g = 2: β„“(D)=deg⁑Dβˆ’1\ell(D) = \deg D - 1 for deg⁑Dβ‰₯3\deg D \geq 3.

Applications to genus 0

ExampleRiemann–Roch characterizes β„™ΒΉ

A smooth projective curve CC has genus 00 if and only if Cβ‰…P1C \cong \mathbb{P}^1 (assuming C(k)β‰ βˆ…C(k) \neq \emptyset).

Proof: If g=0g = 0 and P∈C(k)P \in C(k), then β„“(P)=1βˆ’0+1=2\ell(P) = 1 - 0 + 1 = 2 by Riemann–Roch. So there exists a non-constant f∈L(P)f \in L(P), i.e., a rational function with a simple pole at PP and no other poles. This ff gives a morphism Cβ†’P1C \to \mathbb{P}^1 of degree 11, hence an isomorphism.

Conversely, on P1\mathbb{P}^1, deg⁑K=βˆ’2=2gβˆ’2\deg K = -2 = 2g - 2 gives g=0g = 0.


Applications to genus 1

ExampleRiemann–Roch for elliptic curves

Let (E,O)(E, O) be an elliptic curve (g=1g = 1), with P=OP = O.

  • β„“(O)=1\ell(O) = 1: only constants (since deg⁑(Kβˆ’O)=βˆ’1<0\deg(K - O) = -1 < 0... wait, deg⁑K=0\deg K = 0 so deg⁑(Kβˆ’O)=βˆ’1\deg(K - O) = -1, giving β„“(Kβˆ’O)=0\ell(K-O) = 0. Thus β„“(O)=1βˆ’1+1=1\ell(O) = 1 - 1 + 1 = 1).
  • β„“(2O)=2\ell(2O) = 2: there is a function xx with a double pole at OO. This gives Eβ†’P1E \to \mathbb{P}^1 of degree 22 (the hyperelliptic map).
  • β„“(3O)=3\ell(3O) = 3: adding yy (triple pole at OO).
  • β„“(4O)=4\ell(4O) = 4: adding x2x^2.
  • β„“(5O)=5\ell(5O) = 5: adding xyxy.
  • β„“(6O)=6\ell(6O) = 6: adding x3x^3 and y2y^2. But dim⁑=6\dim = 6 with 77 functions {1,x,y,x2,xy,x3,y2}\{1, x, y, x^2, xy, x^3, y^2\}... so there must be a relation:

y2+a1xy+a3y=x3+a2x2+a4x+a6.y^2 + a_1 xy + a_3 y = x^3 + a_2 x^2 + a_4 x + a_6.

This is the Weierstrass equation! Riemann–Roch naturally produces the equation of the curve.


Applications to genus 2

ExampleRiemann–Roch for genus 2

Let CC have genus 22. The canonical divisor satisfies deg⁑K=2\deg K = 2 and β„“(K)=2\ell(K) = 2.

The canonical map ϕK:C→P1\phi_K: C \to \mathbb{P}^1 is a degree-22 morphism (since ℓ(K)=2\ell(K) = 2 gives a 11-dimensional linear system of degree 22). Therefore:

Every genus-22 curve is hyperelliptic.

More explicitly, ∣K∣|K| is a g21g^1_2 (a linear system of dimension 11 and degree 22). Taking a point PP with β„“(P)=1\ell(P) = 1 (which is true for "most" points since β„“(P)β‰₯1βˆ’2+1=0\ell(P) \geq 1 - 2 + 1 = 0, and β„“(P)=1\ell(P) = 1 iff PP is not a base point of ∣K∣|K|):

By Riemann–Roch, β„“(Kβˆ’P)=β„“(P)+gβˆ’1βˆ’deg⁑P=β„“(P)\ell(K - P) = \ell(P) + g - 1 - \deg P = \ell(P)... Let's be more careful. β„“(Kβˆ’P)=β„“(P)+deg⁑(Kβˆ’P)βˆ’g+1βˆ’(β„“(P)βˆ’?)\ell(K-P) = \ell(P) + \deg(K-P) - g + 1 - (\ell(P) - ?). Better to just use RR directly:

β„“(2P)\ell(2P): RR gives β„“(2P)βˆ’β„“(Kβˆ’2P)=2βˆ’2+1=1\ell(2P) - \ell(K - 2P) = 2 - 2 + 1 = 1. If PP is a Weierstrass point, β„“(2P)=2\ell(2P) = 2 (and β„“(Kβˆ’2P)=1\ell(K - 2P) = 1). If not, β„“(2P)=1\ell(2P) = 1 (and β„“(Kβˆ’2P)=0\ell(K - 2P) = 0). There are exactly 66 Weierstrass points (over kΛ‰\bar{k}, in char β‰ 2\neq 2).


Geometric interpretation: linear systems and maps

Definition4.22Complete linear system

For a divisor DD on CC, the complete linear system ∣D∣|D| is the set of effective divisors linearly equivalent to DD:

∣D∣={Dβ€²β‰₯0∣Dβ€²βˆΌD}β‰…P(L(D))=Pβ„“(D)βˆ’1.|D| = \{D' \geq 0 \mid D' \sim D\} \cong \mathbb{P}(L(D)) = \mathbb{P}^{\ell(D) - 1}.

A gdrg^r_d is a linear system of dimension rr and degree dd (not necessarily complete), i.e., a sub-linear-system VβŠ†βˆ£D∣V \subseteq |D| with dim⁑V=r\dim V = r and deg⁑D=d\deg D = d.

ExampleLinear systems give maps to projective space

If ∣D∣|D| has no base points (i.e., β‹‚Dβ€²βˆˆβˆ£D∣Supp⁑(Dβ€²)=βˆ…\bigcap_{D' \in |D|} \operatorname{Supp}(D') = \emptyset), then ∣D∣|D| defines a morphism:

Ο•D:Cβ†’Pβ„“(D)βˆ’1.\phi_D: C \to \mathbb{P}^{\ell(D) - 1}.

  • g=0g = 0, D=nβ‹…[∞]D = n \cdot [\infty]: Ο•D:P1β†’Pn\phi_D: \mathbb{P}^1 \to \mathbb{P}^n is the Veronese embedding (degree nn, β„“(D)=n+1\ell(D) = n+1).

  • g=1g = 1, D=3OD = 3O: Ο•D:Eβ†ͺP2\phi_D: E \hookrightarrow \mathbb{P}^2 is the Weierstrass embedding as a plane cubic.

  • g=2g = 2, D=KD = K: Ο•K:Cβ†’P1\phi_K: C \to \mathbb{P}^1 is the hyperelliptic double cover.

  • g=3g = 3, non-hyperelliptic, D=KD = K: deg⁑K=4\deg K = 4, β„“(K)=3\ell(K) = 3, so Ο•K:Cβ†ͺP2\phi_K: C \hookrightarrow \mathbb{P}^2 embeds CC as a smooth plane quartic. This is the canonical embedding for non-hyperelliptic genus 3.

  • g=4g = 4, non-hyperelliptic, D=KD = K: deg⁑K=6\deg K = 6, β„“(K)=4\ell(K) = 4, so Ο•K:Cβ†ͺP3\phi_K: C \hookrightarrow \mathbb{P}^3 embeds as a curve of degree 66 in P3\mathbb{P}^3.


Special divisors

Definition4.23Special divisor

A divisor DD is special if β„“(KCβˆ’D)>0\ell(K_C - D) > 0, equivalently h1(O(D))>0h^1(\mathcal{O}(D)) > 0. This means β„“(D)\ell(D) exceeds the "expected" value of deg⁑Dβˆ’g+1\deg D - g + 1.

The index of speciality is i(D)=β„“(KCβˆ’D)=h1(O(D))i(D) = \ell(K_C - D) = h^1(\mathcal{O}(D)), so Riemann–Roch reads:

β„“(D)=deg⁑Dβˆ’g+1+i(D).\ell(D) = \deg D - g + 1 + i(D).

ExampleSpecial divisors on curves
  • D=0D = 0: i(0)=β„“(K)=gi(0) = \ell(K) = g, so β„“(0)=0βˆ’g+1+g=1\ell(0) = 0 - g + 1 + g = 1 βœ“.
  • D=KCD = K_C: i(K)=β„“(0)=1i(K) = \ell(0) = 1, so β„“(K)=(2gβˆ’2)βˆ’g+1+1=g\ell(K) = (2g-2) - g + 1 + 1 = g βœ“.
  • deg⁑D>2gβˆ’2\deg D > 2g - 2: i(D)=0i(D) = 0 (non-special), β„“(D)=deg⁑Dβˆ’g+1\ell(D) = \deg D - g + 1.
  • On a hyperelliptic curve of genus gg, the g21g^1_2 divisor DD (degree 2, β„“(D)=2\ell(D) = 2) satisfies i(D)=2βˆ’2+gβˆ’1+?i(D) = 2 - 2 + g - 1 + ? ... By RR: 2βˆ’i(D)=2βˆ’g+12 - i(D) = 2 - g + 1, so i(D)=gβˆ’1i(D) = g - 1. Very special!

Riemann–Roch for genus gg: a table

ExampleRiemann–Roch dimensions for small genus

For a general curve of genus gg, the following table shows β„“(nP)\ell(nP) for a general point PP:

g=0g = 0 (P1\mathbb{P}^1): β„“(nP)=n+1\ell(nP) = n + 1 for all nβ‰₯0n \geq 0.

g=1g = 1 (elliptic): β„“(0)=1\ell(0) = 1; β„“(nP)=n\ell(nP) = n for nβ‰₯1n \geq 1.

g=2g = 2: β„“(0)=1\ell(0) = 1; β„“(P)=1\ell(P) = 1; β„“(2P)=1\ell(2P) = 1 (general PP); β„“(nP)=nβˆ’1\ell(nP) = n - 1 for nβ‰₯3n \geq 3.

g=3g = 3: β„“(0)=1\ell(0) = 1; β„“(nP)=1\ell(nP) = 1 for n=1,2,3n = 1, 2, 3 (general PP); β„“(nP)=nβˆ’2\ell(nP) = n - 2 for nβ‰₯5n \geq 5; β„“(4P)=2\ell(4P) = 2 (by RR: 4βˆ’3+1=24 - 3 + 1 = 2 since i(4P)=0i(4P) = 0 for general PP on non-hyperelliptic curve).

The Weierstrass gap sequence at PP is the set of n>0n > 0 with β„“(nP)=β„“((nβˆ’1)P)\ell(nP) = \ell((n-1)P), i.e., nn is not a pole order of any function with pole only at PP. By Riemann–Roch, there are exactly gg gaps, contained in {1,2,…,2g}\{1, 2, \ldots, 2g\}.


The Riemann–Roch theorem as an index theorem

RemarkRiemann–Roch as Euler characteristic

Riemann–Roch says Ο‡(L)=deg⁑L+Ο‡(OC)\chi(\mathcal{L}) = \deg \mathcal{L} + \chi(\mathcal{O}_C), where Ο‡(OC)=1βˆ’g\chi(\mathcal{O}_C) = 1 - g. This is a special case of:

  • Hirzebruch–Riemann–Roch on a smooth variety XX: Ο‡(L)=∫Xch⁑(L)β‹…td⁑(TX)\chi(\mathcal{L}) = \int_X \operatorname{ch}(\mathcal{L}) \cdot \operatorname{td}(T_X).
  • Grothendieck–Riemann–Roch for a morphism f:Xβ†’Yf: X \to Y: ch⁑(f!F)=fβˆ—(ch⁑(F)β‹…td⁑(TX/Y))\operatorname{ch}(f_! \mathcal{F}) = f_*(\operatorname{ch}(\mathcal{F}) \cdot \operatorname{td}(T_{X/Y})).
  • Atiyah–Singer index theorem in differential geometry.

For curves, all these reduce to Ο‡(L)=deg⁑L+1βˆ’g\chi(\mathcal{L}) = \deg \mathcal{L} + 1 - g, where ch⁑(L)=1+c1(L)\operatorname{ch}(\mathcal{L}) = 1 + c_1(\mathcal{L}) and td⁑(TC)=1+12c1(TC)=1+(1βˆ’g)[pt]\operatorname{td}(T_C) = 1 + \frac{1}{2}c_1(T_C) = 1 + (1-g)[pt].


Summary

RemarkWhy Riemann–Roch matters

The Riemann–Roch theorem is the single most important result about algebraic curves:

  1. Computes dimensions: tells you exactly how many independent functions/sections exist with given pole constraints.
  2. Classifies curves: low-genus curves are classified via their canonical systems (genus 0 = rational, genus 1 = elliptic, genus 2 = hyperelliptic, ...).
  3. Produces embeddings: determines when a divisor gives an embedding into projective space.
  4. Leads to moduli theory: dimensions of linear systems determine geometry of moduli spaces Mg\mathcal{M}_g.
  5. Generalizes: to Hirzebruch–Riemann–Roch (surfaces), Grothendieck–Riemann–Roch (arbitrary morphisms), and the Atiyah–Singer index theorem.