TheoremComplete

The Riemann–Roch Theorem

The Riemann–Roch theorem is the central result connecting the algebra (cohomology) and geometry (divisors, genus) of algebraic curves. It computes the dimension of the space of meromorphic functions with prescribed poles and zeros.


Divisors on curves

Definition3.10Divisor

A divisor on a smooth projective curve CC is a formal sum

D=PCnPP,nPZ, almost all nP=0.D = \sum_{P \in C} n_P \cdot P, \quad n_P \in \mathbb{Z}, \text{ almost all } n_P = 0.

The degree is degD=nP\deg D = \sum n_P. The divisor group Div(C)\operatorname{Div}(C) is the free abelian group on closed points.

For fk(C)f \in k(C)^* a nonzero rational function, the principal divisor is div(f)=PvP(f)P\operatorname{div}(f) = \sum_P v_P(f) \cdot P where vPv_P is the valuation at PP. We always have deg(div(f))=0\deg(\operatorname{div}(f)) = 0.

Two divisors are linearly equivalent (DDD \sim D') if DD=div(f)D - D' = \operatorname{div}(f). The Picard group is Pic(C)=Div(C)/\operatorname{Pic}(C) = \operatorname{Div}(C) / \sim.

Definition3.11Riemann–Roch space

For a divisor DD on CC, the Riemann–Roch space (or linear system) is

L(D)=H0(C,O(D))={fk(C)div(f)+D0}{0}.L(D) = H^0(C, \mathcal{O}(D)) = \{f \in k(C)^* \mid \operatorname{div}(f) + D \geq 0\} \cup \{0\}.

This is the space of rational functions whose poles are "bounded by DD." We write (D)=dimkL(D)\ell(D) = \dim_k L(D).

ExampleRiemann–Roch spaces on ℙ¹

On P1\mathbb{P}^1 with coordinate tt:

  • L(0)=kL(0) = k (constant functions): (0)=1\ell(0) = 1.
  • L(P)=k+ktL(P_\infty) = k + k \cdot t (functions with at most a simple pole at \infty): (P)=2\ell(P_\infty) = 2.
  • L(nP)=k+kt++ktnL(n \cdot P_\infty) = k + kt + \cdots + kt^n: (nP)=n+1\ell(nP_\infty) = n + 1.
  • L(P0)=0L(-P_0) = 0 (functions with a zero at 00 and no poles — only 00): (P0)=0\ell(-P_0) = 0.

Statement of Riemann–Roch

TheoremRiemann–Roch for curves

Let CC be a smooth projective curve of genus gg over kk, and DD a divisor on CC. Then

(D)(KD)=degDg+1\ell(D) - \ell(K - D) = \deg D - g + 1

where KK is a canonical divisor (O(K)=ωC\mathcal{O}(K) = \omega_C, the canonical sheaf) with degK=2g2\deg K = 2g - 2.

In sheaf-theoretic notation:

h0(O(D))h0(ωCO(D))=degDg+1h^0(\mathcal{O}(D)) - h^0(\omega_C \otimes \mathcal{O}(-D)) = \deg D - g + 1

or equivalently (by Serre duality H1(O(D))H0(ωC(D))H^1(\mathcal{O}(D)) \cong H^0(\omega_C(-D))^\vee):

h0(O(D))h1(O(D))=degDg+1=χ(O(D)).h^0(\mathcal{O}(D)) - h^1(\mathcal{O}(D)) = \deg D - g + 1 = \chi(\mathcal{O}(D)).

RemarkThe three forms of Riemann–Roch
  1. Classical (Riemann, 1857): (D)degDg+1\ell(D) \geq \deg D - g + 1 (Riemann's inequality, with (KD)\ell(K-D) as the "defect").
  2. With Serre duality (Roch, 1865): (D)(KD)=degDg+1\ell(D) - \ell(K - D) = \deg D - g + 1.
  3. Euler characteristic: χ(O(D))=degD+1g\chi(\mathcal{O}(D)) = \deg D + 1 - g.

The third form shows that the Euler characteristic depends only on the degree, not on the specific divisor DD.


Applications of Riemann–Roch

ExampleGenus 0: rational curves

If g=0g = 0 and PP is any point on CC: (P)(KP)=10+1=2\ell(P) - \ell(K - P) = 1 - 0 + 1 = 2. Since degK=2\deg K = -2 and deg(KP)=3<0\deg(K - P) = -3 < 0, we have (KP)=0\ell(K - P) = 0, so (P)=2\ell(P) = 2.

This means there exists a nonconstant function ff with a single simple pole at PP — equivalently, f:CP1f : C \to \mathbb{P}^1 is an isomorphism. So every genus-0 curve with a rational point is isomorphic to P1\mathbb{P}^1.

ExampleGenus 1: elliptic curves

Let EE be an elliptic curve (g=1g = 1) with origin OO. By Riemann–Roch, (nO)=n\ell(nO) = n for n1n \geq 1 (since degK=0\deg K = 0 and K0K \sim 0):

| nn | (nO)\ell(nO) | Basis | New function | |-----|------------|-------|-------------| | 1 | 1 | {1}\{1\} | — | | 2 | 2 | {1,x}\{1, x\} | xx has a double pole at OO | | 3 | 3 | {1,x,y}\{1, x, y\} | yy has a triple pole at OO | | 4 | 4 | {1,x,y,x2}\{1, x, y, x^2\} | — | | 5 | 5 | {1,x,y,x2,xy}\{1, x, y, x^2, xy\} | — | | 6 | 6 | {1,x,y,x2,xy,x3}\{1, x, y, x^2, xy, x^3\} | — |

In degree 6, y2y^2 and x3x^3 both have a pole of order 6, so they must be linearly dependent modulo lower-order terms. This gives the Weierstrass equation:

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.

ExampleThe canonical embedding

For a genus-gg curve CC with g2g \geq 2, the canonical divisor KK has degK=2g2\deg K = 2g - 2 and (K)=g\ell(K) = g (by RR: (K)(0)=(2g2)g+1=g1\ell(K) - \ell(0) = (2g-2) - g + 1 = g - 1, so (K)=g\ell(K) = g).

If CC is not hyperelliptic, the canonical linear system K|K| gives an embedding:

ϕK:CPg1.\phi_K : C \hookrightarrow \mathbb{P}^{g-1}.

This is the canonical embedding. For g=3g = 3: CP2C \hookrightarrow \mathbb{P}^2 as a smooth quartic curve. For g=4g = 4: CP3C \hookrightarrow \mathbb{P}^3 as the intersection of a quadric and a cubic.

ExampleVery ampleness criterion

A divisor DD on a genus-gg curve CC gives an embedding CP(D)1C \hookrightarrow \mathbb{P}^{\ell(D)-1} if and only if:

  • (DP)=(D)1\ell(D - P) = \ell(D) - 1 for all PP (separates points from tangent vectors), and
  • (DPQ)=(D)2\ell(D - P - Q) = \ell(D) - 2 for all P,QP, Q (separates points).

By Riemann–Roch: if degD2g+1\deg D \geq 2g + 1, then DD is very ample ((KD+P+Q)=0\ell(K - D + P + Q) = 0 since deg(KD+P+Q)<0\deg(K - D + P + Q) < 0). So every curve of genus gg embeds in Pg+1\mathbb{P}^{g+1} via a divisor of degree 2g+12g + 1.

ExampleHyperelliptic curves

A curve CC of genus g2g \geq 2 is hyperelliptic if there exists a degree-2 map f:CP1f : C \to \mathbb{P}^1 (equivalently, (D)2\ell(D) \geq 2 for some divisor of degree 2).

By Riemann–Roch: (D)(KD)=2g+1=3g\ell(D) - \ell(K - D) = 2 - g + 1 = 3 - g. For g3g \geq 3, (KD)g3+(D)g1\ell(K - D) \geq g - 3 + \ell(D) \geq g - 1, so such a DD imposes nontrivial conditions.

  • g=2g = 2: every genus-2 curve is hyperelliptic (since (K)=2\ell(K) = 2 and degK=2\deg K = 2).
  • g=3g = 3: a curve is hyperelliptic     \iff it is not a smooth plane quartic.
  • g3g \geq 3: the hyperelliptic locus is a proper subvariety of the moduli space Mg\mathcal{M}_g.

Riemann–Roch for surfaces

TheoremRiemann–Roch for surfaces (Noether's formula)

For a smooth projective surface SS and a divisor DD:

χ(OS(D))=χ(OS)+12(D2DKS)\chi(\mathcal{O}_S(D)) = \chi(\mathcal{O}_S) + \frac{1}{2}(D^2 - D \cdot K_S)

where D2D^2 and DKSD \cdot K_S are intersection numbers. The Noether formula for the structure sheaf:

χ(OS)=112(KS2+e(S))\chi(\mathcal{O}_S) = \frac{1}{12}(K_S^2 + e(S))

where e(S)e(S) is the topological Euler characteristic.

ExampleRR on ℙ²

On P2\mathbb{P}^2: K=3HK = -3H (where HH is a line), χ(O)=1\chi(\mathcal{O}) = 1, H2=1H^2 = 1. For D=dHD = dH:

χ(O(d))=1+12(d2+3d)=(d+1)(d+2)2=(d+22).\chi(\mathcal{O}(d)) = 1 + \frac{1}{2}(d^2 + 3d) = \frac{(d+1)(d+2)}{2} = \binom{d+2}{2}.

This matches h0(O(d))=(d+22)h^0(\mathcal{O}(d)) = \binom{d+2}{2} for d0d \geq 0 (and higher cohomology vanishes).

ExampleRR on a K3 surface

A K3 surface SS has KS=0K_S = 0 and χ(OS)=2\chi(\mathcal{O}_S) = 2. For a divisor DD:

χ(OS(D))=2+D22.\chi(\mathcal{O}_S(D)) = 2 + \frac{D^2}{2}.

If DD is effective with D2=2d2D^2 = 2d - 2 (e.g., a genus-dd curve), then χ=d+1\chi = d + 1. For a smooth curve CDC \in |D| of genus g=dg = d, Riemann–Roch on SS gives h0(OS(C))d+1h^0(\mathcal{O}_S(C)) \geq d + 1 — a (d+1)(d+1)-dimensional linear system of curves.


Hirzebruch–Riemann–Roch and Grothendieck–Riemann–Roch

TheoremHirzebruch–Riemann–Roch

For a smooth projective variety XX of dimension nn and a vector bundle E\mathcal{E}:

χ(X,E)=Xch(E)td(TX)\chi(X, \mathcal{E}) = \int_X \operatorname{ch}(\mathcal{E}) \cdot \operatorname{td}(T_X)

where ch(E)=rk(E)+c1(E)+12(c122c2)+\operatorname{ch}(\mathcal{E}) = \operatorname{rk}(\mathcal{E}) + c_1(\mathcal{E}) + \frac{1}{2}(c_1^2 - 2c_2) + \cdots is the Chern character and td(TX)=1+12c1+112(c12+c2)+\operatorname{td}(T_X) = 1 + \frac{1}{2}c_1 + \frac{1}{12}(c_1^2 + c_2) + \cdots is the Todd class.

ExampleHRR recovers classical RR for curves

For a curve CC and a line bundle L\mathcal{L} of degree dd:

  • ch(L)=1+c1(L)=1+d[pt]\operatorname{ch}(\mathcal{L}) = 1 + c_1(\mathcal{L}) = 1 + d \cdot [\mathrm{pt}].
  • td(TC)=1+12c1(TC)=1+(1g)[pt]\operatorname{td}(T_C) = 1 + \frac{1}{2}c_1(T_C) = 1 + (1 - g)[\mathrm{pt}].

χ(L)=C(1+d[pt])(1+(1g)[pt])=d+1g.\chi(\mathcal{L}) = \int_C (1 + d[\mathrm{pt}])(1 + (1-g)[\mathrm{pt}]) = d + 1 - g. \quad \checkmark

TheoremGrothendieck–Riemann–Roch (GRR)

For a proper morphism f:XYf: X \to Y and a coherent sheaf F\mathcal{F} on XX:

ch(f!F)=f(ch(F)td(TX/Y))\operatorname{ch}(f_! \mathcal{F}) = f_*(\operatorname{ch}(\mathcal{F}) \cdot \operatorname{td}(T_{X/Y}))

in the Chow ring A(Y)QA^*(Y)_\mathbb{Q}, where f!F=(1)i[RifF]f_! \mathcal{F} = \sum (-1)^i [R^i f_* \mathcal{F}] in K0(Y)K_0(Y).

GRR specializes to HRR when Y=SpeckY = \operatorname{Spec} k (a point), and to the classical Riemann–Roch when XX is a curve.

ExampleGRR for a family of curves

For a flat family f:CBf: \mathcal{C} \to B of genus-gg curves, GRR computes the Chern classes of the Hodge bundle E=fωC/B\mathbb{E} = f_*\omega_{\mathcal{C}/B} (a rank-gg vector bundle on BB):

c1(E)=f(c1(ωC/B)22+c1(ωC/B)c1(Ωf1)12).c_1(\mathbb{E}) = f_*\left(\frac{c_1(\omega_{\mathcal{C}/B})^2}{2} + \frac{c_1(\omega_{\mathcal{C}/B})c_1(\Omega^1_f)}{12}\right).

This leads to the Mumford formula and is foundational for intersection theory on the moduli space Mg\overline{\mathcal{M}}_g.


Summary

RemarkRiemann–Roch across dimensions
DimensionFormulaName
Curve (n=1n=1)χ(L)=degL+1g\chi(\mathcal{L}) = \deg \mathcal{L} + 1 - gRiemann–Roch
Surface (n=2n=2)χ(L)=χ(O)+12(D2DK)\chi(\mathcal{L}) = \chi(\mathcal{O}) + \frac{1}{2}(D^2 - D \cdot K)Noether
Any nnχ(E)=ch(E)td(TX)\chi(\mathcal{E}) = \int \operatorname{ch}(\mathcal{E}) \cdot \operatorname{td}(T_X)Hirzebruch–RR
Relativech(f!F)=f(chtd)\operatorname{ch}(f_!\mathcal{F}) = f_*(\operatorname{ch} \cdot \operatorname{td})Grothendieck–RR
Arithmeticch^(f!F)=f(ch^td^)\widehat{\operatorname{ch}}(f_!\overline{\mathcal{F}}) = f_*(\widehat{\operatorname{ch}} \cdot \widehat{\operatorname{td}})Arithmetic RR (Faltings, Gillet–Soulé)