On algebraic curves*)

By George R. Kempfat Baltimore

§ 0. Introduction

One hundred and fifty years ago, Abel established bis famous result on an abelian
sum of an algebraic integral along a linear System ([1], p. 145—211). His theorem to be used
effectively requires a good understanding of the linear Systems on a given algebraic curve.
Abel's optimistic footnote ([1], p. 550) indicates that he might have gotten control of this
last problem shortly before his death.
Galois announced some progress on this question [2]. He mentioned in his letter,
a third memoir on algebraic integrals, which seems to have been lost. This paper was
partly inspired by trying to reconstruct how he could have convinced himself of the
existence of integrals, particularly of the second and third kinds.
The main result about linear Systems, the Riemann-Roch theorem, was originally
justified by potential theoretic methods, which are far removed from the rather algebraic
methods used by Abel and his follower Galois. Twentieth Century algebraic proofs of
this theorem have been given by A. Weil, C. Chevalley and J.-P. Serre ([4], Ch. II).
The proof of the Riemann-Roch theorem, that I give in this paper, uses the morphic
and variational point of view, developed by A. Grothendieck.
The heart of the proof is the systematic study of the obstruction of finding a rational
section of an invertible sheaf ££ on an algebraic curve C, which has a simple pole at
a variable point on the curve. This study gives an a priori proof that Ω€ (χ) JS?®"1 has
vanishing cohomology if 5£ has no non-zero regul r section. Further refinements of the
study yield the Serre duality theorem and the Riemann-Roch theorem for any invertible
sheaf on an algebraic curve.

§ 1. Preliminaries

Let C be a fixed smooth irreducible algebraic curve over an algebraically closed

fieldfc.

*) Partly supported by N. S. F. Grant No. MPS 75-05578

 Kempf, On algebraic curves 41

Let «5? be an invertible $c-module. The objective of this paper is to prove the basic
results about the global sections of <g. The method employed to approach this question
uses systematically the obstruction to fmding global sections of <£; i. e., the cohomology
.
We have a natural flasque resolution of JS?, given by the sequence
(1) 0 -> y -» 9t(&) -> 9(&) -> 0 ,
where 9t(&) is the constant sheaf of rational sections of $£ and &(&) is the quotient
sheaf of principal parts of rational sections modulo regulδr sections. The torsion sheaf
is naturally isomorphic to the direct sum of its stalks at each point.
The sequence (1) can be used t o compute the cohomology groups of Jδf. We have an
exact sequence,
0 -> (C, JSf) -> Rat (Jδf) -> Prin (Δ7) —U tf1 (C, J^) -* 0 ,
(2) II
, JS?)
where Rat(JS?) = r(C, «(Ja?)) and Prin(JS?) = r(C, 0»(JS?)). The groups, ^(C, <£?), always
vanish for / ^ 2.
Let Z> be an effective divisor on C. We have an exact sequence of sheaves,
(3) 0 -> <e -> JSP(Z)) -^ JSf (/))|D -^ 0 .
Taking the cohomology of this sequence, we obtain an exact sequence,
(q>D) 0^r(C9^^r(C,X(D))^r(C9&(D)\D)^H\C9W
because Hv (C, J27 (ί)|D) is zero δs ^?(Z))|D is supported by a finite number of closed
points.
Clearly, ^(JSf) may be identified with union over increasing Z),
Furthermore, Prin (.S?) is (J^T^, JSf (D)|D) and the diagram,

, JSf)
commutes.
I want to make two points based directly on the above remarks. The first is
a) Any cohomology class in H1 (C, JSf) is contained in the image of ΦD for some
effective divisor D.
b) If D = D± -l· D2, where Di and D2 are two effective divisors with disjoint support,
then r(C, &(D)\D) is naturally isomorphic to F(C, ^(OOU) T(C, JS?(/)2)lD2)· Further,
we may identify ΦD with the sum <5Dl 0 ^D2.
We now return for a closer look at the sequence . Recall that
dimkr(C,
42 Kempf, On algebraic curves

Thus, by cpD, we always have

dimk[r(C, &(D))/r(C9
This inequality motivates the following defmition.
7
The invertible sheaf is said to be ordinary if, for all effective divisors D, we have
the equality
dim[r(C, JS?(D))/r(C, JSf)] = degZ).
Ordinariness has a simple cohomological Interpretation.
Lemma 1. 1. The invertible sheaf ^ is ordinary ifand only if Hl(C, JS?) = 0.
Proof. By the sequences φΌ, ^ is ordinary if and only if the connecting homo-
morphisms <5D are zero for all effective D. Clearly, SD is zero if its target Hl(C, <£} is zero.
Thus, the vanishing implies ordinariness.
On the other hand, by point a), all of H1 (C, <£} consists of images of (Vs.
Thus, H1 (C, JS?) is zero if all the Vs are. Hence, H1 (C, £) = 0 if & is ordinary. Q.E.D.
One may reduce the problem of proving that a sheaf is ordinary by using the following
lemma, which can be proved by the obvious induction.
Lemma 1. 2. An invertible sheaf ^f is ordinary if and only if.for any effective divisor E
and any point D on C, we have the equality
dim[r(C, £>(D + E))/r(C, JS?(£))] = l .
The reader should notice that the last criterion (replacing J£(E) by ^} only
involves studying the exact sequence <pD, where Z) is a point of C. The main argument
of this paper consists of studying the global Variation of φΏ with D.

§ 2. The existence of rational differentials

We begin this section with the globalization of the sequences φΏ. Recall that the
diagonal Δ in C χ C is a closed subvariety, which is locally defined by one equation.
The sheaf, c^c( — A)\^ = GCxC( — A)/0CxC( — 2A), is naturally isomorphic to the sheaf
of differentials, ΩΔ, on the diagonal.
We need to have an explicit choice of this isomorphism. It will suffice to defme
a universal derivation, D: &Δ —* &C^C(~^)\A- Let / be a rational function on Δ. Denote
by ft the composed rational functions, /4: Cx C—^-» C— >A »^, where n{
denotes the projection of C χ C on the i'-th factor. To define the derivation Z), let
/)(/)=/!—/2. Clearly, D(f) is a regul r section of ®cxc( — A)\A at points where / is
regul r. One may check that D is a universal derivation on &Δ. Thus, we have a canonical
isomorphism ρ in a commutative diagram,
 Kempf, On algebraic curves 43

We may use the above isomorphism to make an exact sequence,

(1) 0^0CxC^0CxC(J)--iU f -*-><),

by replacing CxC(A)\A by the sheaf of vector fields, f ~ l . Explicitly, if τ is a parameter
function at a point D of Δ and g is a regul r section of &CxC(A) in a neighborhood of D,
then μ (g) is the vector field, (T! — T 2 )g|j · — — .
οτ
Now, if ? is an invertible sheaf on C, we may tensor (1) by π*^ to obtain a global
form of the sequence (3) of section 1. The result is an exact sequence,
(2) 0-
In fact, the restriction of (2) to the fiber of the projection π2 is the sequence (3) of section 1.
To get the global form of φΌ, we will take the direct images of the sequence (2)
under π 2 . In this way, we find an exact sequence,
0->π 2 ,π*^->π 2+ (π*^(Ζΐ))-^π 2 ,(πί^® Ω® - 1 )->/? 1 π 2 ,π*^^Λ 1 π 2 ,(π*Α ? (ζ1))^0 ,
which ends in 0 because π2 maps Δ isomorphically onto C. The terms of this sequence
need a little Interpretation. Firstly, π2*(πί & (χ) Ω®-1) is naturally isomorphic to
¥®Ω®~1.
A basic result on the constancy of direct image of constant families of invertible
sheaves [cf. [3] III. 1. 4. 15] yields canonical isomorphisms,
Rln2*nf & ^ T/1' (C, JS?) ®k &c .
Using these isomorphisms, we may rewrite the above long exact sequence in the form
0 -» r(C, JSf) ®k (9C -> π2*(π*^(Α)} -> ^ (χ) Ω®' 1
(<P)
-Λ //HC JSf) ® k fl?c -> ^ 2 *^fJS?(J)) -> 0 :
The sequence φ is the globalization of the <pD. The critical part of the sequence is
the connecting homomorphism δ. If D is a point of C, <5D may be identified with
o®<rck(D), where k(D} is the residue field at D. Thus we have the

Scolium. δ vanishes at D ifand only ifdD is zero.

We now come to the first vanishing theorems. For the rest of this paper, we will
assume that any everywhere regul r function on C is a constant, i. e.,
r(C,Oc) = k.
Our first result is about the existence of some rational differentials.

Proposition 2. 1. Lei E be any effective divisor on C of positive degree. Then the

sheaf QC(E) is ordin r y.
Proof. Let & = QC(E). By lemma 1. 2, we must see that <5D is zero for any point D
on C. It suffices to see that the global connecting homomorphism δ is zero. In this case,
δ is a homomorphism from £ ® Qj)-i = Qc(E) ® Ω?""1 &0C(E) to the free sheaf
Hl(C, <£} ®k&c· As there are no non-zero everywhere regul r functions on C, which
vanish along E, δ must be zero. Q.E.D.
44 Kempf, On algebraic curves

Immediately, using Lemma 1. l, we have

Corollary 2. 2. H1 (C, €( )) = 0/or any effective divisor E of positive degree.

§ 3. The finiteness theorem

Let D be an effective divisor on the curve C. As F(C, 0C) = k, dim^C, 0C(D))^1 +deg Z).
Hence, T(C, JS?) is always finite dimensional for any invertible 0c-module Z£ because,
if <£ has a non-zero section, JS? is isomorphic to &C(D} for some effective D.
The cohomology group H1 (C, JSf ) is also finite. We have the
Proposition 3. 1. Let J^ be an invertible sheaf on C. Then, F(C9 J?) awd /^(C, JSf)
are finite dimensional k-vector spaces.
Furthermore, the Euler characteristic of ^ ,
X (Jδf ) = dm* (C, JSf ) - dimfc H1 (C, JSf ) ^^wafa l - g + deg Δ7 ,
= genus of C = aimkHl(C, Oc) and deg J£ = deg D for the divisor D of any rational
section o
Proof. By Corollary 2.2 and the above remarks about F(C9 J5f), we know that there
exists at least one invertible sheaf M such that both cohomology groups of M are finite
dimensional.
Let E be an effective divisor on C. For any invertible yT, the long exact sequence
of cohomology of the sequence 0 —* Jf —* < ^( ) —> JY*(E)\E —* 0 shows that , f "(E) has
finite dimensional cohomology <=> J/* does. Furthermore, we have the equality

By applying the same argument twice, we find that the same results hold for an arbitrary
divisor E.
The structure sheaf &c is isomorphic to Jl(E} for some divisor E. Therefore,
(9C has finite dimensional cohomology. Let Jδf be an arbitrary sheaf. Then, ^^&C(D),
where D is the divisor of any rational section of JSf . Hence ^ has finite dimensional
cohomology and

The proposition is finished by noting that X (&c) =l—g. Q.E. D.

This proposition gives essentially two ways to compute the deg^. A traditional
fact, which can be derived from it, is
Corollary 3. 2. A rational function on C has δs many zeroes δs poles (counted with
multiplicity).
Proof. By Proposition 3. l, the deg0c = 0 = deg(/) for any non-zero rational
function (/) on C. Q.E.D.
Another fact that follows directly from the degree formula is
Corollary 3. 3. Let & be an invertible sheaf of negative degree. Then, J£ does not
admit a regulδr section.
This fact allows us to strengthen the main vanishing result of the last section.
 Kempf\ On algebraic curves 45

Proposition 3. 4. Lei & be an invertible sheaf on C, which has degree strictly larger
than the sheaf ofdifferentials, Qc. Then, & is ordinary and H1 (C, Ji?) is zero.
Proof. This proof is the same s Proposition 2.1, except that, now, δ is zero because,
by Corollary 3.3, the sheaf of positive degree, J^(x)i2®" 1 , possesses no non-zero
homomorphisms into a free sheaf. Q.E.D.

§ 4. The canonical class and residues of differentials

The main result of this section is

Proposition 4. 1. There is a natural isomorphism between Hl(C, c) and the ground

fleldk.
The proof of this proposition will give an explicit description of this isomorphism.
We will begin the proof with

Lemma 4. 2.Assume that Hl(C, J?) is one-dimensional for an invertible sheaf ££ on C.

Also, assume that H1 (C, ^(D)} = Ofor each point D ofC. Then, <£ is isomorphic to the sheaf
of differentials, Qc.
Proof. By assumption, the connecting homomorphisms D are never zero. Thus,
the global connecting homomorphism

must be an isomorphism of invertible sheaves. Hence, & « Qc. Q.E. D.

To apply the lemma, note that there must always exist a sheaf JSf , satisfying the
assumptions of the lemma. In fact, take J5? to be an invertible sheaf of maximum degree,
which has a non-zero cohomology group. Such a sheaf exists because Proposition 3.4 shows
that all sheaves of high degree have zero cohomology and Proposition 3. l shows that
sheaves of low degree have non-zero cohomology. Thus, by the lemma 4.2, Ω€ must
satisfy the assumptions of the lemma.
It remains to give the isomorphism of the proposition. By the proof of the lemma,
we know that the connecting homomorphism for c gives an isomorphism,
δ : Ω€ ® β®' l -SU H1 (C, c )j8> k 0c ·
"
&c-
Passing to global sections, ε gives an isomorphism k -^-> H1 (C, c). This proves the
proposition.
We will determine this isomorphism k -^-» H1 (C, Qc) in terms of principal parts in

Lemma 4.3. Lei D be a point ofC and let τ' be a parameter function at D. The principal
άτ'
pari at D, , represents the cohomology class in Hl(C,Qc} which corresponds to l
τ D
under the isomorphism with k.
Journal f r Mathematik. Band 295
46 Kempf, On algebraic curves

Proof. To determine the cohomology class which corresponds to l, we need to study

the isomorphism, Qc (x) Qf~ l « &c. In our case, the basic exact sequence (2) of section 2 is
0 -> π? c -> π? cOJ) -*-> rc? c (χ) Ω® ~ 1 -> 0 .

suc
We may use the calculation of section 2 to find a local section r of πί c (^) h
X(r) = lin^.
Let τ be the parameter function of Δ at (D, D) corresponding to τ' οη C. Let τ1
and τ2 be the rational functions on C χ C s in section 2. The previous calculation shows
that the local section —-- of nf c(A) goes to άτ ® -r— = l under X.
τί — τ2 οτ
If we specialize the above calculation to the sequence
0 -> c -» flc(D) -^ 0C(D)|D -^ 0 ,
8
:

we find that κη takes to 1. Thus, D is the canonical generator of H1 (C, Qc).

This proves the lemma. Q.E.D.
We will use the fundamental isomorphism to define the residue of a principle part
of a differential. Recall from section l, we have an exact sequence,
Rat ( c) -» Prin( c) -» H1 (C, c) -> 0.
1
Using the isomorphism /i (C, c) —^ /:, we can define by composition a /c-linear
function, Res: Prin(i2c) —> k, which vanishes on and only on the principal parts of rational
differentials.
To compute Res, it will suffice to check the

Lemma 4.4. Let ω be the principal part of a rational differential at a point D of C.

Let t be a parameter function at D. Then, the residue, Res (ω), is the coefficient of— dt in
the Taylor series development of(D at D.

Proof. By lemma 4.3, we know that the result is true when co = (constant)—.
On the other hand, by Proposition 2. l, we know that there exist rational differentials on C
which are regul r except for poles at D of order 2, 3, etc. As the residue of the principal
parts ofthese differentials vanish, the lemma follows from the additivity of Res. Q.E.D.

§ 5. Serre duality and the Riemann-Roch theorem

The most decisive connection between the sections of invertible sheaves and
cohomology groups is given by the Serre duality theorem,
 Kempf, On algebraic curves 47

Proposition 5. 1. For any invertible sheaf ££ on C, the cup product pairing

F(C, Qc ® ^f®- 1 ) χ Hl(C, &) -> Hl(C, c )^&,

is a perfect pairing.
Proof. Consider the global connecting homomorphism of JS?,
δ :&
This induces a /r-linear mapping,

s the composition
(a® l ) o δ: ja?
is given by a section of Qc ® J^®" 1 for each ^-linear function α on Hl(C, J^).
The first step in the proof is
Lemma 5. 2. η is surjective and is compatible with the cup product together with the
canonical isomorphism with k.
Proof of lemma. Let α be an element of T(C, Qc ® J*?®"1). Multiplication by α gives
us a commutative diagram,

, c)
are 6
where δ(<£) and ^( c) th connecting homomorphisms for J^ and JSf = c respectively.
Directly from this diagram, one may check that η(€ o ((J α)) = α, where c is the canonical
isomorphism T/1 (C, c) « λ:. This proves the lemma.
To finish the proof of the proposition, it is enough to show
Lemma 5. 3. η is injective.
Proof. Let β be an element of Ker(?/). We will specialize δ at any point D on C.
By definition of η, the composition,

iszero s β e Ker(>/).
Let £ > ! , · · · > D n be « distinct points of C. By point b) of section l, <5Dl © · · · ® Dn
have the same image in Hl(C,<£} s does the connecting homomorphism Dl + ...+Dn.
Thus, β vanishes on the image of Dl + . . . + D . Hence β is induced by a /r-linear function on

If n + deg. ? >deg c, we know by Proposition 3.4 that the last cohomology group
is zero. Therefore, = 0; i.e., η is injective. This proves the lemma and the proposition.
We are now able to conclude
Theorem 5. 4 (Riemann-Roch). For any invertible 5£,
dimkr(C, JT)-di
48 Kempf, On algebraic curves

Proof. By Proposition 5. l,
dimkr(C, Qc (x) £>®-l) = dimkHi(C, &).
Hence, the left-hand side of the equation in the theorem is the Euler characteristic X(J5f).
The theorem follows from the degree formula of Proposition 3. l. Q.E.D.
The last corollary I leave to the reader.
Corollary 5. 5. dimk (C, ίc) = g and deg (ίc) = 2g - 2.

References

[1] N. H. Abel, Oeuvres completes, Tome Premier, Christiania 1881.

[2] E. Galois, Lettre de Galois δ M. Auguste Chevalier, Journal de Mathematiques pures et appliquees 11 (1846),
408—416.
[3] A. Grothendieck, Elements de Geometrie Algebrique. III, Publications Mathematiques 11, IHES, 1961.
[4] J.-P. Serre, Groupes algebriques et corps de classes, Paris 1959.

Department of Mathematics, The Johns Hopkins University, Baltimore, Maryland 21218, USA

Eingegangen 14. September 1976