Algebraic Geometry IID 2013

7 Divisors on curves
For the rest of this course, curve will mean smooth, projective, irreducible curve,
unless explicitly stated to the contrary.
P
A divisor on a curve V is a finite formal sum P∈ V nP P with nP ∈ Z (finite
means that for all but finitely many P , nP = 0). Sometimes the points are put
in brackets (P ) to make the notation clearer. The set of divisors
P on V forms an
abelian group under
P obvious addition, denoted Div(V ). If D = nP P is a divisor,
define deg(D) = nP ∈ Z. The map D 7→ deg(D) is obviously a homomorphism,
whose kernel is denoted Div0 (V ) (divisors of degree 0). We sometimes write vP (D)
for the coefficient nP of P in D.
Let f ∈ k(V )∗ be a nonzero rational function. Define the divisor of f to be
X
div(f ) = (f ) := vP (f )P.
P ∈V

Corol.6.8(ii) says that div(f ) ∈ Div0 (V ). Divisors of the form div(f ) are called
principal divisors, and form a subgroup div(k(V )∗ ) ⊂ Div0 (V ).
If you’re doing Number Fields you will probably notice the similarity between this
and ideal theory for number fields. In particular, we can also define the divisor
class group of V for be the quotient

Cl(V ) = Div(V )/div(k(V )∗ ).

and for D ∈ Div(V ) write [D] for the class of D in Cl(V ). Divisors in the same
divisor class are said to be linearly equivalent, written D ∼ E. So D ∼ E iff
D − E is a principal divisor. If so then deg(D) = deg(E).
Proposition 7.1. Every divisor of degree 0 on P1 is principal.
P
Proof.
P Write the Q divisor as D = a∈k na (a) + n∞ (∞). As deg(D) = 0, n∞ =
− na . Let f = a∈k (t − a)na . Then since (t − a) is a local parameter at a and
a unit at b 6= a,
Pva (f ) = na , and since 1/(t − a) is a local parameter at ∞ for any
a, v∞ (f ) = − na = n∞ .

For a general curve, deg : Div(V ) → Z induces a homomorphism Cl(V ) → Z,

obviously surjective, and (by 7.1) an isomorphism for V = P1 . Later will see this
is a characteristic property of P1 .
Other ways divisors arise:
Hyperplane sections div(L). Let V ⊂ Pn and consider a hyperplane H =
V (L) ⊂ Pn not containing V , some linear form L. Define
X
div(L) = nP P, where if Xi (P ) 6= 0, nP = vP (L/Xi )

Note that this is independent of i, and that the only P occuring in the sum are
P ∈ V ∩ H.
If L0 is another linear form then div(L0 ) − div(L) = div(L0 /L) which is a principal
divisor, so div(L) and div(L0 ) are linearly equivalent and in particular have the

26
same degree, called the degree of V . For an irreducible plane curve V = V (F ),
vP (L/Xi ) is just the multiplicity mP (V, H) (see proof of 5.1) and the degree of V
is just the degree of F .
Likewise, any homogeneous G ∈ k[X] of degree m such that V (G) 6⊃ V determines
a divisor div(G) which is linearly equivalent to m×div(L), and therefore has degree
md.
Special case: V = V (F ) ⊂ P2 , F irreducible of degree n. see that

#V (F ) ∩ V (G)} ≤ mn

i.e. (cf. the special case Prop.5.3):

Theorem 7.2 (Bezout’s Theorem, basic version). Two distinct irreducible plane
curves of degrees m, n intersect in at most mn points.
P
A divisor D = nP P is effective if nP ≥ 0 for all P — notation D ≥ 0. (Some
authors confusingly use the term positive.) Let D be any divisor. Then associated
to D are two important invariants: the first is

L(D) = L(D) = {f ∈ k(V ) | f = 0 or div(f ) + D ≥ 0}

X
= {f ∈ k(V ) | ∀P ∈ V, vP (f ) + nP ≥ 0} if D = nP P.
P

Noting that vP (f + g) ≥ vP (f ) we see that L(D) is a vector space. Its dimension is

written `(D), which is finite. For example, let ∞ = (0 : 1) ∈ V = P1 , D = m(∞).
Writing x = X1 /X0 we see that L(D) is spanned by 1, x, . . . , xm so `(D) = m + 1.
In general we have:

Proposition 7.3. Let D ∈ Div(V ). Then:

(i) deg(D) < 0 =⇒ L(D) = 0.
(ii) deg(D) ≥ 0 =⇒ `(D) ≤ deg(D) + 1.
(iii) For any P ∈ V , `(D) ≤ `(D − P ) + 1.

Proof. (i) If L(D) 6= 0 then for 0 6= f ∈ L(D), div(f ) + D = E ≥ 0. But then

deg(D) = deg(E) ≥ 0 (as deg div(f ) = 0).
(iii) Let n = vP (D). Define α : L(D) → k by α(f ) = (πPn f )(P ). The kernel of this
homomorphism is then L(D − P ) so `(D − P ) ≥ `(D) − 1.
(ii) now follows: if d = deg(D) ≥ 0 we see `(D) ≤ `(D − (d + 1)P ) + d + 1 = d + 1
since deg(D − (d + 1)P ) = 0.

If D ∼ E, so that D − E = div(g) then L(D) and L(E) are isomorphic by the

map f 7→ f g. So `(D) depends only on the class of D.

8 Differentials
Differentials are a way of doing calculus on varieties, in a coordinate-free way.
K/k field extension. Informally a differential is a finite sum of formal expressions
x dy with x, y ∈ K, subject to the usual rules of calculus. Precisely:

27
Definition The space of Kähler differentials ΩK/k is the quotient M/N where


M = K-vector space generated by symbols δx, x ∈ K
subspace generated by δ(x + y) − δx − δy,
 
N=
δ(xy) − x δy − y δx, δa for x, y ∈ K, a ∈ k.
and define dx = δx + N ∈ ΩK/k . (Think of K as functions, k as constants.)
The map d : K → ΩK/k is the exterior derivative. It is k-linear since if a ∈ k
then d(ax) = a dx
Any k-linear map D : K → U to a K-vector space U satisfying the product rule
D(xy) = xDy +yDx is called a derivation (more precisely, a k-derivation). So d
is a derivation. Another example of a derivation is the formal differentiation map
d/dX : k(X) → k(X). (We make the same definition of K is a ring containing k
and U is a K-module.)
Lemma (/tautology). A map D : K → U is a derivation iff there is a K-linear
map λ : ΩK/k → U such that λ(dx) = D(x) for all x ∈ K.

Proof. If λ is such a K-linear map then obviously D = λ ◦ d is k-linear and

D(xy) = λ(d(xy)) = xλ(dy) + yλ(dx), so D is a derivation. Conversely, given
a derivation D : K → U , write ΩK/k = M/N as in the definition, and define a
K-linear map λ̂ : M → U by δy 7→ D(y) for all y ∈ K. Then as D is a derivation it
follows that λ̂(N ) = 0 so we get a K-linear map λ with the desired properties.

For any derivation (in particular d), if y 6= 0 then Dx = D(y(x/y)) = yD(x/y) +

(x/y)Dy giving the quotient formula D(x/y) = y −2 (yDx − xDy).

P 8.1. (i) If f = g/h ∈ k(X1 , . . . , Xn ) and y = f (x1 , . . . , xn ) ∈ K, then

Lemma
dy = i (∂f /∂Xi )(x1 , . . . , xn ) dxi .
(ii) If K = k(x1 , . . . , xn ) for xi ∈ K then {dxi } spans ΩK/k .

Proof. (i) follows from the rules for d(xy), d(x/y) and k-linearity. (ii) is an imme-
diate consequence.
Theorem 8.2. Let K/k(t) be finite and separable, t transcendental over k. Then
ΩK/k is one-dimensional, spanned by dt.

Proof. First suppose K = k(t). Then by 8.1(ii), ΩK/k is generated by dt so has

dimension ≤ 1. Enough to show it is nonzero. By Lemma-Tautology, enough to
show there is a non-zero derivation K → K, and d/dt is one.
For the general case, write K0 = k(t) so that K = K(α) = k(t, α) by the primitive
element theorem. Let h ∈ K0 [X] be the minimal polynomial of α. Then h0 (α) 6= 0
by separability. By 8.1(ii), ΩK/k is spanned by dt and dα. If for f ∈ K0 [X] we
write Dt f = ∂f /∂t (i.e. apply d/dt to the coefficients of f ), then 8.1(i) gives
0 = d(h(α)) = (Dt h)(α)dt + h0 (α)dα
so ΩK/k is spanned by dt. It therefore is enough to show ΩK/k 6= 0, or equivalently
to write down a none-zero derivation K → K.
Define a derviation D : K0 [X] → K (which is isomorphic to K0 [X]/(h), hence is a
K0 [X]-module) by
(Dt h)(α)
D(f ) = Dt (f ) if f ∈ K0 , D(X) = − , D(X n ) = nαn−1 D(X).
h0 (α)

28
Then D(h) = Dt (h)(α) + h0 (α)D(X) = 0, so for any f ∈ K0 [X], D(f h) =
f (α)D(h) + h(α)D(f ) = 0. So D vanishes on the ideal hK0 [X] ⊂ K0 [X], hence
defines a derivation D̄ : K = K0 [X]/(h) → K, whose restriction to K0 is Dt , hence
is non-zero.
Remark. We have d(xp ) = pxp−1 dx, so if K has characteristic p > 0, then d(xp ) = 0
for all x ∈ K.In what follows I will generally stick to the case of characteristic zero,
but point on when there are issues in the finite characteristic chase.
Our situation; V a curve (smooth, projective & irreducible), K = k(V ). An
element of Ωk(V )/k is called a rational differential on V . As k is fixed I will
usually drop the “/k”. Differentials are usually denoted ω, η, ξ. . . .
P
Let P ∈ V , We say ω ∈ Ωk(V ) is regular at P if it can be expressed as ω = fi dgi
with fi , gi ∈ OV,P . We let ΩV,P or ΩP denote the set of differentials regular at P .
It is obviously an OV,P -module.
Theorem 8.3. ΩV,P is the free OV,P module generated by dπP for any local pa-
rameter πP at P .

So ΩV,P = {f dπP | f ∈ OV,P }. In particular, if πP0 is another local parameter, the

dπP0 = udπP where u ∈ OV,P
∗
is regular an non-zero at P .
(It is not hard to show that ΩP is just the module of differentials ΩOP /k .)
Definition. If ω ∈ Ωk(V ) and P ∈ V , let vP (ω) = vP (f ) where ω = f dπP .

By the last remark this doesn’t depend on the choice of local parameter, and
vP (ω) ≥ 0 iff ω is regular at P .

Proof. Obviously OP dπP ⊂ ΩP . Let f = f (P ) + πP g ∈ OP = k + mP . Then

df = gdπP + πP dg ∈ OP dπP + πP ΩP . Therefore

OP dπP ⊂ ΩP ⊂ OP dπP + πP ΩP

and then applying Nakayama’s Lemma with R = OP , J = mP , M = ΩP ⊃ N =

OP dπP , we get ΩP = OP dπP . The only thing we need to check is that ΩP is
finitely generated. Choose an affine piece V0 ⊂ An of V containing P , so that
k[V0 ] = k[x1 , . . . , xn ] say. If f ∈ OP then f = g(x)/h(x) for polynomials g, h with
g(P ) 6= 0, and then
X h∂g/∂Xi − g∂h/∂Xi
df = (x)dxi
h2
so {xi } generate ΩP .
P
We define the divisor of a non-zero differential ω ∈ Ωk(V ) to be (ω) = P vP (ω)P .
If 0 6= ω 0 ∈ Ωk(V ) then ω 0 = f ω for some f ∈ k(V )∗ , so div(ω 0 ) = div(f ) +
div(ω). Therefore the divisor class of div(ω) doesn’t depend on ω. It is called the
canonical class of V . e write KV for any element of the canonical class, and call
it a canonical divisor. (Note the non-canonical use of the word ”canonical”. . . )
V = P1 . Compute: vP (dt) = 0 if P = a ∈ A1 (since t − a is a local parameter). At
∞, π∞ = t−1 is a local parameter and dt = −t2 d(1/t), so v∞ (dt) = v∞ (t2 ) = −2.
So (dt) = −2(∞) is a canonical divisor.
Lemma 8.4. Let 0 6= ω ∈ Ωk(V )/k . Then vP (ω) = 0 for all but finitely many P .

29
Proof. As vP (f dg) = vP (f ) + vP (dg) and vP (f ) = 0 for all but finitely many P ,
it’s enough to consider ω = dg with k(V )/k(g) finite and separable. Consider
φ = (1 : g) : V → P1 . By the finiteness theorem, there are only finitely many
P ∈ V with g(P ) = ∞ or eP > 1. For all other P , g − g(P ) = φ∗ (t − g(P )) is a
local parameter at P , and therefore by 8.3(ii), vP (dg) = 0.

Define the divisor of ω 6= 0 to be

X
(ω) = vP (ω)P.
P

As any other nonzero ω 0 ∈ Ωk(V )/k is of the form f ω, f ∈ k(V )∗ , the divisors of ω
and ω 0 are linearly equivalent.
Define the canonical class of V to be the class of (ω). Denote by KV any divisor
in the canonical class.
Fix ω ∈ Ωk(V )/k and let KV = (ω). Then f ω is regular iff (f ) + KV ≥ 0, i.e.
∼ Ω(V ),
L(KV ) −→ f 7→ f ω.

In particular, Ω(V ) is finite-dimensional. Major definition:

Definition: g(V ) = dim Ω(V ) = `(KV ) is the genus of V .
Remark. The genus of V depends only on the isomorphism class of V , not on how
V is embedded into projective space (unlike degree).
Ex: V = P1 . We saw KP1 = −2(∞) and therefore g(P1 ) = `(KV ) = 0.
V = V (F ) plane cubic, F = X0 X22 − 3i=1 (X1 − λi X0 ), with λi 6= λj if i 6= j. We
Q
assume ch(k) 6= 2. Then VQis nonsingular (cf. Q1 on example sheet #2). Affine
equation is f (x, y) = y 2 − (x − λi ) = y 2 − g(x) say.
Observe 2y dy = g 0 (x) dx in Ωk(V )/k . Let ω = dx/y.
Claim: vP (ω) = 0 for all P ∈ V .
Assuming this is true then KV = 0, so g(V ) = `(0) = 1 and Ω(V ) = kω. Various
cases:

• P ∈ V0 , y(P ) 6= 0. Then (∂f /∂y)(P ) 6= 0 so x − x(P ) is a local parameter

at P , hence vP (ω) = vP (dx) = vP (d(x − x(P ))) = 0.
• P ∈ V0 , y(P ) = 0, x(P ) = λi . Then (∂f )/(∂x)(P ) = −g 0 (λi ) 6= 0 (simple
root), so y is a local parameter at P . Then vP (ω) = vP (2dy/g 0 (x)) = 0.
• P = P0 = (0 : 0 : 1) point at infinity. Then as vP0 (x) = −2 and vP0 (y) = −3,
have vP0 (dx/y) = (−2 − 1) − (−3) = 0 by 8.5(?) (below).

[Alternative calculation at infinity: in the affine patch {X2 6= 0}, Q

use coordinates
(z, t) = (X0 /X2 , X1 , X2 ), P = (0, 0). Equation of V becomes z = (t − λi z), and
vP (z) = 3, vP (t) = 1. Therefore dx/y = d(1/t)/(z/t) = −(t3 /z)dt and vP (ω) = 0.]
In particular, this proves that V is not isomorphic to P1 .
Proposition 8.5. (i) Suppose char(k) = 0. Let 0 6= f ∈ k(V ), and assume
vP (f ) 6= 0. Then vP (df ) = vP (f ) − 1.
(ii) Suppose char(k) = p 6= 0, and n = vP (f ). Then vP (df ) ≥ n − 1, with equality
if (p, n) = 1.

30
Proof. Let n = vP (f ), so f = πPn u with u ∈ OP∗ . Write du = g dπP . Then
df = πPn−1 (nu + πP g)dπP . So vP (df ) = (n − 1) + vP (nu + πP g). Both results
follow.
Proposition 8.6. Let V = V (F ) ⊂ P2 be a plane curve (irreducible projective
nonsingular) of degree d ≥ 1. Then KV = (d − 3)H, where H is the divisor of a
hyperplane (i.e. line) section.

Proof. Choose coordinates so that (0 : 1 : 0) ∈/ V . Let x = X1 /X0 , y = X2 /X0

viewed as rational functions on V . Then f (x, y) = 0 where f (X, Y ) = F (1, X, Y )
is the affine equation of V , so (∂f /∂X)(x, y) dx + (∂f /∂Y )(x, y) dy = 0 in ΩV /k .
So let
dx dy
ω= =−
(∂f /∂Y )(x, y) (∂f /∂X)(x, y)
Claim (w) = (d − 3)H with H = hyperplane at infinity.
Let P ∈ V ∩ A2 . As in the previous example, if (∂f )/∂Y )(P ) 6= 0, then x − x(P )
is a local parameter at P and so vP (ω) = vP (1/(∂f )/∂Y )(P )) = 0. Otheriwise,
(∂f )/∂Y )(P ) 6= 0, in which case y − y(P ) is a local parameter and vP (ω) = 0.
It remains to consider points at infinity. Since (0 : 1 : 0) ∈
/ V , any point at infin-
ity is contained in the affine piece {X2 6= 0}, on which V has equation g = 0
with z = X0 /X2 = 1/y, t = X1 /X2 = x/y and g(Z, T ) = F (Z, T, 1) ∈ k[Z, T ].
Let η = dz/(∂g/∂T )(z, t) = −dt/(∂g/∂Z)(z, t). The preceding argument shows
that vP (η) = 0 for any P in this the affine piece {X2 6= 0}. But f (X, Y ) =
Y d g(1/Y, X/Y ) so ∂f /∂X = Y d−1 (∂g/∂V )(1/Y, X/Y ) and so
dy z −2 dz
ω=− = d−1 = z d−3 η
(∂f /∂X)(x, y) y (∂g/∂T )(z, t)
and so if X2 (P ) 6= 0, vP (ω) = (d−3)vP (z)+vP (η) = (d−3)vP (z). Since z = X0 /X2 ,
this means (ω) = (d − 3)div(X0 ) = (d − 3)H.

Mention: topological nature of genus. Curvature.

9 Riemann-Roch
Let C be a (smooth, projective) curve. We have already seen the space L(D) =
{f | (f ) + D ≥ 0}, where D is a divisor on C, and its dimension `(D) = dim L(D).
By definition, `(D) > 0 iff D is linearly equivalent to an effective divisor.
The Riemann-Roch problem is to determine `(D).
Recall (7.3) that `(D) ≤ deg(D) + 1. When V = P1 we have seen that for all D,
`(D) = max(0, deg(D) + 1).
Theorem 9.1 (Riemann-Roch). Let g be the genus of V , and K = KV a canonical
divisor. For any divisor D,
`(D) − `(K − D) = 1 − g + deg(D).

This is a hard theorem, and the proof is beyond the course. The simplest proof
uses sheaf cohomology — see chapter 2 of Serre, Algebraic Groups and Class Fields
for a readable proof, or Hartshorne chapter 5 for a shorter but much fancier one.
We will content outselves to discovering how powerful this result is.

31
Corollary 9.2. deg(K) = 2g − 2.

Proof. Take D = K so that `(D) = `(K) = g and `(K − D) = `(0) = 1.

Corollary 9.3. A plane (smooth, projective) curve of degree d has genus (d −

1)(d − 2)/2.

Proof. By 8.5 K = (d − 3)H and deg(H) = d so deg(K) = (d − 3)d = 2g − 2 =⇒

g = (d2 − 3d + 2)/2.

So d = 1 or 2 =⇒ g = 0 (line or conic, which we already know to be ' P1 ).

For d = 3 we get g = 1, and for plane quartics, g = 3. In particular, no (smooth)
plane curve has genus 2. (There are plenty of curves of genus 2 in P3 however.)
In particular we see that if nonsingular curves V , V 0 of degrees d 6= d0 are isomor-
phic, then {d, d0 } = {1, 2}. (As they must have the same genus, d(d−3) = d0 (d0 −3)
i.e. (d0 − d)(d0 + d − 3) = 0.) The converse is far from true: if d > 2 there are
infinitely many isomorphism classes of plane curves of degree d (we’ll do the case
d = 3 later).

Corollary 9.4. deg(D) > 2g − 2 =⇒ `(D) = 1 − g + deg(D).

Proof. `(K − D) = 0 in this case because deg(K − D) = 2g − 2 − deg(D) < 0.

Curves of genus 1.

Corollary 9.5. Suppose g(V ) = 1. Then KV ∼ 0, and deg(D) > 0 =⇒ `(D) =

deg(D).

Proof. As `(KV ) = g = 1 there exists an effective divisor in the class of KV , which

must therefore be 0 as deg(KV ) = 2g − 2 = 0. Second part follows from 9.4.

Fix P0 ∈ V . The pair (V, P0 ) (or, less correctly, just V itself) is called an elliptic
curve. Traditionally we write E instead of V (actually it is also more common to
use C for curves. . . ).
Let P , Q ∈ E. Then `(P + Q − P0 ) = 1 so there exists a unique effective divisor
of degree 1 (i.e. a point) R such that P + Q − P0 ∼ R. We define:

P +E Q = R

(It would perhaps be more correct, but over-pedantic, to write P +(E,P0 ) Q.)

Theorem 9.6. The operation +E makes E into an abelian group, with identity
element P0 . Moreover the map P 7→ [P − P0 ] ∈ Cl(E) is an isomorphism of groups
between E and Cl0 (E), the groups of divisor classes of degree 0 on E.

Proof. Let β(P ) = [P − P0 ] ∈ Cl0 (E). First show that β is a bijection. Have
β(P ) = β(Q) ⇐⇒ P − P0 ∼ Q − P0 ⇐⇒ P ∼ Q ⇐⇒ P = Q since `(P ) = 1.
So β is injective. Also if D is a divisor of degree 0 then as `(D + P0 ) = 1 there
exists P with D + P0 ∼ P , so [D] = β(P ). Therefore β is a bijection (of sets).
Finally, if P +E Q = R then β(P +E Q) = [R − P0 ] = [P + Q − P0 − P0 ] =
[P − P0 ] + [Q − P0 ] = β(P ) + β(Q). So β transforms +E into addition in Cl0 (E),
and therefore (E, +E ) is a group and β is an isomorphism.

32
We’ll often write 0E for the identity point P0 in the group law. A smooth plane
cubic has genus 1. Let’s look at the special case we considered in the last lecture.

Theorem 9.7. Assume char(k) 6= 2, and let E = V (F ) ⊂ P2 be the nonsingular

plane cubic:
3
Y
F (X0 , X1 , X2 ) = X0 X22 − (X1 − λi X0 ), λi 6= λj if i 6= j.
i=1

Let OE = P0 = (0 : 0 : 1) ∈ E. Then in the group law on E

P +E Q +E R = 0E ⇐⇒ P , Q, R are collinear

(We’ll see soon that any curve of genus 1 is isomorphic to such a plane cubic.)
By collinear here we mean that there is a line L ⊂ P2 for which the line section
on E is the divisor P + Q + R (if P , Q, R are distinct this just means that they
line on L.)

Proof. P +E Q +E R = 0E ⇐⇒ P + Q + R ∼ 3P0 (by definition of the group

law) which holds iff ∃f with (f ) = P + Q + R − 3P0 . As L(3P0 ) = h1, x, yi =
h1, X1 /X0 , X2 /X0 i, this holds iff f = G/X0 for a linear form G with (G) = P +
Q + R.

Before getting on to curves of higher genus, we’ll first obtain the Riemann-
Hurwitz formula.
Let φ : V → W be a finite morphism of curves. Assume char(k) = 0 here. Let
ω = f dt ∈ Ωk(W )/k , k(W )/k(t) finite. Then k(V )/φ∗ (k(t)) is also finite so Ωk(V )/k
is generated by dφ∗ (t). Define

φ∗ (ω) = φ∗ (f ) dφ∗ (t).

Let P ∈ V , Q = φ(P ). We will compare vP (φ∗ (ω) and vQ (ω). Let eP be the
ramification degree of φ at P , and πP , πQ local parameters.

Lemma 9.8. Assume char(k) = 0. Then vP (φ∗ ω) = eP vQ (ω)+e−1. In particular,

vP (φ∗ (dπQ )) = e − 1.

Proof. Write ω = uφnQ dπQ , so that vQ (ω) = vQ (f ) = n ∈ Z. Then vP ∗ φ∗ (ω) =

vP (φ∗ u) + nvP (φ∗ πP ) + vP (dφ∗ πP ) = neP + vP (dφ∗ πP ). Now φ∗ (πQ ) = yπPe for
∗
some y ∈ OV,P with dy = z dπP say and so

d(φ∗ πQ ) = (ey + πP z)πPe−1 dπP

so vP (d(φ∗ πQ )) = e − 1 since char(k) = 0.

Theorem 9.9 (Riemann-Hurwitz formula). Let φ : V → W be a finite morphism

of curves in characteristic zero. Let n = deg(φ). Then
X
2g(V ) − 2 = n(2g(W ) − 2) + (eP − 1).
P ∈V

33
Proof. Let 0 6= ω ∈ Ωk(W )/k . Then
X
2g(V ) − 2 = deg div(φ∗ ω) = vP (φ∗ ω)
P ∈V
X X
= vP (φ∗ ω)
Q∈W P 7→Q
X X
= (eP vQ (ω) + eP − 1)
Q∈W P 7→Q
X X 
= nvQ (ω) + (eP − 1)
Q∈W P 7→Q
X
= n deg div(ω) + (eP − 1)
P ∈V

Remark. (not from lectures) In characteristic p, things change a bit:

• We must assume that k(V )/k(W ) is separable (otherwise φ∗ : Ωk(W )/k →

Ωk(V )/k is identically zero).
• Assuming separability, let δP = vP (φ∗ dπQ ). The proof of the lemma shows
that δP = eP − 1 if p - eP , and is ≥ eP if p | eP . One says that φ is wildly
ramified at P if p | eP , tamely ramified otherwise.
• The Riemann-Hurwitz formula for a finite separable morphism φ : V → W
(in any characteristic) is then:
X
2g(V ) − 2 = n(2g(W ) − 2) + δP .
P ∈V

Examples Say π : V →PP1 has degree 2. Then eP = 1 or 2. R-H formula

=⇒ 2g − 2 = 2(0 − 2) + (eP − 1), i.e.
n
g = − 1, n = #{P ∈ V | eP = 2} = 2g + 2
2
(thus n is the number of ramificiation points of π). Specifically:

g = 0 =⇒ n = 2.
g = 1 =⇒ n = 4. In fact, if V = E has Legendre equation y 2 = x(x−1)(x−
λ) and P0 is the point at infinity then π = φ2P0 = (1 : x) : E → P1 has degree
2 and is ramified precisely at {P0 , (0, 0), (1, 0), (λ, 0)} (the points of order
dividing 2 in the group of points of E), and π(P ) = π(Q) ⇐⇒ P = ±E Q.

Now consider g > 1.

Definition A curve V of genus g > 1 is hyperelliptic is there exists π : V → P1
of degree 2. If so, then consider D = π ∗ (∞). Have 1, π ∗ (X1 /X0 ) ∈ L(D) and so
`(D) ≥ 2. Moreover if `(D) = 3 then D = P + Q say and `(P ) = 2, hence V = P1
which is impossible. So `(D) = 2.
Theorem 9.10. (i) Let g(V ) > 1. If there exists a divisor D ≥ 0 of degree 2
on V with `(D) = 2 then π = φD : V → P1 has degree 2, π ∗ (∞) = D and V is
hyperelliptic.
(ii) Every curve of genus 2 is hyperelliptic.

34
Proof. (i) Say D = P + Q and π = φD = (1 : x) : V → P1 where L(D) = h1, xi.
Then (x) = D0 − D, some D0 = P 0 + Q0 ≥ 0. We must have {P, Q} ∩ {P 0 , Q0 } = ∅
since if say Q = Q0 then (x) = P 0 − P so `(P ) = 2 and V ' P1 .
Therefore vP (x) = −1 = vQ (x) if P 6= Q, or vP (x) = −2 if P = Q. In either case,
π ∗ (∞) = P + Q.
(ii) If g = 2 then `(K) = 2 = deg(K).

We can write hyperelliptic curves explicitly as follows. Suppose π : V → P1 , D =

π ∗ (∞), L(D) = h1, xi. Then deg(π) = 2 =⇒ k(V )/k(x) is an extension of degree
2, so (as we are assuming char(k) 6= 2!) k(V ) = k(x, y) where y 2 = r(x) ∈ k(x),
r(x) not a square.QAs k[x] is a UFD, we can write r(x) = h(x)(p(x)/q(x))2 for
p, q, h ∈ k[x], h = mi=1 (x − λi ) squarefree.
Then V is (by Theorem 2.5) birational to the plane curve V 0 with affine equation
f (x, y) = y 2 − h(x). The affine part V 0 ∩ A2 is smooth, since if P = (xP , yP ) ∈
V 0 ∩ A2 then if (∂f /∂y)(P ) = 2yP = 0, we have h(x) = −f (P ) = 0. But
(∂f /∂x)(P ) = −h0 (xP ) 6= 0 and h is squarefree. The intersection V 0 ∩ {X0 = 0}
is one point (0 : 0 : 1) which in fact is singular. In any case, we get a birational
morphism
(1 : x : y) : V → V 0 ⊂ P2
and a rational map
(X0 : X1 ) : V 0 −
−→ P1
whose composite is π = φD : V → P1 , and therefore π is ramified over x =
λ1 , . . . , λm and possibly also infinity. Therefore since the number of ramification
points is 2g + 2 which is even, either

• m = 2g + 2 is even, π is unramified over ∞; or

• m = 2g + 1 is odd, π is ramified over ∞.

10 Projective embeddings
Let V ⊂ Pn be a curve of degree d, not contained in any hyperplane. Then
D = (X0 ) is an effective divisor of degree d. A given curve V can occur in
projective space in different ways (for example, a curve of genus 0 is isomorphic
to P1 , but also to a conic in P2 , which has degree 2, and to a twisted cubic in P3 ,
etc.) For a fixed curve V , we can ask: as we consider all ways of embedding V
into projective space (or varying dimension) what such divisors D can arise?
P
If F = λi Xi 6= 0 is any linear form, then (F ) ∼ D and F/X0 ∈ L(D). So have
X
β : {linear forms F = λi Xi } ,−→ L(D), F 7→ F/X0 .

(Injective linear map, since V doesn’t lie on a hyperplane.)

2 observations: let P , Q be distinct points of V , not lying on {X0 = 0}. (We can
always change coordinates so that this holds; this amounts to replacing D by a
linearly equivalent divisor).

(1) There exist linear forms F , G with F (P ) 6= 0 and G(P ) = 0 6= G(Q). So

β(F ) ∈ L(D) \ L(D − P ) and β(G) ∈ L(D − P ) \ L(D − P − Q). Therefore
`(D − P − Q) ≤ `(D) − 2, and so by 7.3(iii), `(D − P − Q) = `(D) − 2.

35
 (2) As P is a smooth point, it has a tangent line L = TPproj . There exists a linear
form F with F (P ) = 0 but not vanishing identically on L. Therefore the
multiplicity of P in (F ) is exactly 1, hence β(F ) ∈ L(D − P ) \ L(D − 2P )

Se we deduce that D satisfies:

(∗) For every P , Q ∈ V (not necessarily distinct), `(D − P − Q) = `(D) − 2.

Now start with a curve V and a divisor D with `(D) = n + 1 ≥ 2. Pick a basis
{f0 , . . . , fn } for L(D). It defines a morphism

φD = (f0 : f1 : · · · : fn ) : V → Pn .

We say φD is an embedding if φD is an isomorphism between V and a (necessarily

smooth, irreducible) curve in Pn .
Note that choosing another basis changes φD by a linear transformation of Pn .
Also, if D0 = D − (g) is an equivalent divisor, then {gfi } is a basis for L(D0 ),
hence φD = φD0 depends only on the equivalence class of D.
Theorem 10.1 (Embedding criterion). φD is an embedding iff (∗) holds.

The above discussion shows that condition (∗) is necessary. The meat of the
theorem is therefore that it is a sufficient condition.
I won’t prove the theorem here — see for example Proposition 6.56 in Hulek
(although he finesses some of the difficulties by defining “embedding” in a slightly
different way). I will show however that (∗) implies that φD is injective. Let P ,
Q ∈ V be distinct points. There exist functions p, q ∈ k(V ) with vP (p) = vQ (q) =
1, vP (q) = vQ (p) = 0 (take ratios of suitable linear forms on the projective space
containing V ). Replacing D with D + (pa q b ) for suitable a, b ∈ Z, we may assume
vP (D) = vQ (D) = 0. We have `(D − P − Q) = `(D) − 2, by 7.3(iii) we have
`(D −P ) = `(D)−1 as well. Choose a basis {fi } for L(D) such that {f0 , . . . , fm−2 }
spans L(D − P − Q) and {f0 , . . . , fm−1 } spans L(D − P ). Then all fi are regular
at P and Q and fm−1 (P ) = 0 6= fm (P ), fm−1 (Q) 6= 0. Therefore φD (P ) 6= φD (Q).
This shows that if (∗) holds, then φD is injective. The idea of the rest of the
proof is: by general theory, the image φD (V ) is a possibly singular curve V 0 ⊂ Pm .
The condition with P = Q is then used to show that V 0 is smooth and that
∼ V 0.
k(V ) = k(V 0 ), which then implies that φ : V −→
Corollary 10.2. If deg(D) > 2g then φD is an embedding.

Proof. Apply Riemann-Roch: as deg(D) > deg(D − P − Q) > 2g − 2 = deg(K),

we have `(KD ) = `(K( D − P − Q)) = 0 and so

`(D) = 1 − g + deg(D), `(D − P − Q) = 1 − g + deg(D − P − Q) = `(D) − 2.

Examples:
First consider the case g = 0. Then deg(D) = n > 0 implies `(D) = n + 1 and
D ∼ nP for any P ∈ V . Therefore φD is always an embedding. Taking V = P1
and D = n(∞) we get L(D) = k ⊕ k.x ⊕ · · · ⊕ k.xn , hence

φn(∞) = (1 : x : · · · : xn ) : P1 → Pn

36
is the n-tuple embedding.
Next consider g = 1. Corollary says that if deg(D) ≥ 3 then φD is an embedding.
Pick P0 ∈ V and consider the case D = 3P0 . As `(nP ) = n by Riemann-Roch, we
have:

L(P0 ) = k L(2P0 ) = span{1, x} for some x with vP0 (x) = −2

L(3P0 ) = span{1, x, y} for some y with vP0 (x) = −3

Then L(4P0 ) = L(3P0 ) ⊕ k.x2 and L(5P0 ) = L(4P0 ) ⊕ k.xy, and x3 , y 2 both
have vP0 = −6, hence lie in L(6P0 ) \ L(5P0 ). Therefore there must be a linear
dependence between 1, x, x2 , x3 , y, xy, y 2 in which the coefficients of x3 and y 2 are
nonzero. Replacing y by cy for suitable c 6= 0 this takes the form

y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6

for suitable ai ∈ k.

Theorem 10.3. Let E, P0 ) be an elliptic curve. Then ∃a1 , a2 , a3 , a4 , a6 ∈ k and

an isomorphism E −→ ∼ V = V (F ) ⊂ P2 where V is a smooth cubic with affine
defining polynomial

f (x, y) = F (1, x, y) = y 2 + a1 xy + a3 y − (x3 + a2 x2 + a4 x + a6 ) (W)

and P0 7→ (0 : 0 : 1). Moreover if char(k) 6= 2 coordinates may be chosen to that in

addition, a1 = a3 = 0 and

f (x, y) = y 2 − x(x − 1)(x − λ), λ ∈ k, λ ∈

/ {0, 1} (L)

The cubic (W) is called a (generalised) Weierstrass equation for E, and the
form (L) is Legendre normal form. The indices are written in such a way that
is the variables x, y are assigned weight 2,3 and ai is assigned weight i then each
term in f has weight 6.

Proof. From the above, φ3P0 : V → P2 is an embedding, and its image lies in V (F )
for some F as in (W). As V is a curve of genus 1 this can only happen if the image
equals V (F ) and if V (F ) is nonsingular.
If char(k) 6= 2 then by completing the square,
3
 a1 a3 2 Y
y+ x+ = (cubic)(x) = (x − λi )
2 3 i−1

and λi 6= λj as V is smooth. Writing

x − λ1 y + a1 x/2 + a3 /3 λ3 − λ1
x0 = , y0 = , λ= 6= 0, 1, ∞
λ2 − λ1 (λ2 − λ1 )3/2 λ2 − λ1

gives (y 0 )2 = x0 (x0 − 1)(x0 − λ).

Consider now Legendre normal form with char(k) 6= 2. Then if P = (1 : a : b) =

(a, b) ∈ E, the P 0 = (a, −b) ∈ E also, and the line x = a cuts out the divisor
P + P 0 + P0 . In other words, P 0 = −P in the group law.

37
For n ∈ Z, write [n]P for n times P in the group law. Then [2]P = 0E iff P = −P ,
so we see that in the Legendre model,
{P ∈ E | [2]P = 0E } = {0E , (0, 0), (1, 0), (λ, 0)
which is therefore isomorphic to Z/2Z × Z/2Z.
What about [3]P = 0E ? This holds iff the tangent at P has 3-fold intersection
with E at P , i.e. iff P is a point of inflection. Using the Hessian one can show
that if char(k) 6= 3 then there are exactly 9 points of inflection on E (P0 being one
of them) and so
{P ∈ E | [3]P = 0E } ' Z/3Z × Z/3Z if char(k) 6= 3
More generally one can show that
{P ∈ E | [n]P = 0E } ' Z/nZ × Z/nZ if char(k) - n
Before leaving curves of genus 1 let’s just explain what happens when k = C.
Consider a pair τ1 , τ2 ∈ C of complex numbers, linearly independent over R. Let
Λ = Zτ1 + Zτ2 ⊂ C. Theory of elliptic functions (see Riemann surfaces course)
tells us that there is a meromorphic function ℘(z), holomorphic on C apart from
double poles at every z ∈ Λ, such that ℘(z + λ) = ℘(z) for all λ ∈ Λ. Moreover
℘(z) satisfies the differential equation
℘0 (z)2 = 4℘(z)3 = g2 ℘(z) − g3 , certain g2 , g3 ∈ C.
The functions ℘, ℘0 are therefore meromorphic functions on the Riemann surface
T = C/Λ, and one shows that the map
(
(1 : ℘(z) : ℘0 (z)/2) if z ∈ C \ Λ
z 7→
(0 : 0 : 1) if z ∈ Λ

is then a bijection between T and a smooth plane cubic curve in P2C . Now T has an
obvious group structure (as a quatient group of C) and this map is an isomorphism
of groups (for the group law on the cubic we have defined earlier).
Finally notice that there is an isomorphism
R/Z × R/Z −→ ∼ T, (x , x ) 7→ x τ + x τ mod Λ
1 2 1 1 2 2

and so the subgroup of elements of order dividing n in T is isomorphic to Z/nZ ×

Z.nZ.
Finally consider a curve V of genus g ≥ 2. If V is hyperelliptic, then we have
already seen a fairly precise description of V .
If not, we have in any case `(K) = g ≥ 2. Consider the morphism φK : V → Pg−1
given by a canonical divisor K.
Theorem 10.4. Suppose V is not hyperelliptic. Then φK : V → Pg−1 is an em-
bedding.

Proof. Suppose φK is not an embedding. Then by the theorem, there exist P and
Q with `(K − P − Q) ≥ g − 1. Apply Riemann-Roch to D = P + Q. We get
`(D) = `(K − D) + 1 − g + deg(D) ≥ 2. So as g 6= 0, `(D) = 2, say L(D) = k ⊕ k.x
with (x) = −P − Q + D0 . Then φD : V → P1 satisfies φ∗D (∞) = D, so φD has
degree 2, i.e. V is hyperelliptic.

38