What is a Sheaf? Exploring Germs of Holomorphic Functions

What is a Sheaf?
Author(s): J. Arthur Seebach, Jr., Linda A. Seebach and Lynn A. Steen

Reviewed work(s):
Source: The American Mathematical Monthly, Vol. 77, No. 7 (Aug. - Sep., 1970), pp. 681-703
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2316199 .
Accessed: 22/11/2012 17:28
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to
The American Mathematical Monthly.
http://www.jstor.org
This content downloaded by the authorized user from 192.168.72.226 on Thu, 22 Nov 2012 17:28:25 PM
All use subject to JSTOR Terms and Conditions
WHAT IS A SHEAF?
J. ARTHUR SEEBACH, JR., LINDA A. SEEBACH, AND LYNN A. STEEN,
St.OlafCollege
Ever since Jean Leray and Henri Cartan in 1950 formallyintroduced the
concept of a sheaf,the various examples and applications of sheaves have come
to play a major role in such diverse fieldsas several complexvariables, algebraic
geometry,and differentialand algebraic topology.Yet nearly all monographs
which use or introduce sheaves assume the sophistication of graduate level
algebraic topology. So it is very difficultforan undergraduateto acquire from
the available literaturea real understandingof sheaves and theirapplications.
It is the purpose of this article to introduce the theoryof sheaves at an ele-
mentary level with the hope that the interestedreader will then be able to
approach any of the standard treatises (e.g., [2], [3], [6], [10], or [11]) with
significantinsight.
Our avenue of approach to the theoryof sheaves will be throughexamples
drawn fromthreemajor areas of mathematics:fromanalysis, the sheafof germs
of holomorphicfunctions;fromalgebra, the sheafof local rings;and fromgeome-
try, the sheaf of differentialforms.We will develop each of these particular
sheaves in considerable detail, for the differentperspectivesthus revealed will
more readily make transparentthe subsequent discussion of the general theory
of sheaves.
1. The sheaf of germs of holomorphicfunctions.A holomorphic(or analytic)
functionon an open subset D of complex n-space C- is definedto be a complex
valued functionon D which has a local power series representationat each
point of D. Osgood's lemma [6, p. 2] asserts that a continuous functionon
DC C- is holomorphicifand only ifit is holomorphicin each variable separately.
A very important property of holomorphic functions is that they are
uniquely determinedby theirbehavior on open sets: iff and g are holomorphic
on a domain D (a domain is a connected open set), and iff equals g on a non-
empty open subset of D, thenf equals g on all of D. To see this,we need only
observe that the largest open subset of D on whichf = g is also closed (relative
to D), since the partial derivativeswhich determinethe power series expansion
are continuous. Since D is connected,this set must be D.
Now if Sz C- we say thatf is holomorphicat z if it is holomorphicon some
neighborhoodof z. The collection A, of functionsholomorphicat z formsan
algebra over the fieldof complex numbersin which the operations of sum and
ProfessorSteenreceivedhis Ph.D. in 1965 underKennethHoffman See-
at M.I.T.; Professor
bach receivedhis Ph.D. in 1968underA. I. Weinzweigat Northwestern;hiswifeLinda Seebach
Steenand Seebach
Professors
is a Ph.D. candidateunderDaniel Zelinsky,also at Northwvestern.
in Topology.
are the authors of the recentlypublished Counterexamnples
The paper printedherewas derivedfroma seriesof lecturespresentedby the authorsat a
weeklyjointcolloquiumofthemathematics staffsofSt. Olafand CarletonColleges,bothlocated
in Northfield,
Minnesota.Editor.
681
682 J. ARTHUR SEEBACH, JR., LINDA A. SEEBACH, AND LYNN A. STEEN [September
product involve intersectionof domain: if f: U->C and g: V-*C, then f+g:

Un V->C and fg: Un V->+C. We let Ih be the ideal in A, consistingof those
functionsin A, whichvanish identicallyon some neighborhoodofZ.
The algebra of germsof holomorphic functionsat z is then definedto be the
quotient ring (algebra) A,/I,, and is denoted by Oz. So a germof a holomorphic
functionis an elementf+I, of O., wheref is holomorphicat z. We will usually
denote thisgermby f]X.Followingthe usual practice,we shall oftenidentify,or
fail to distinguishbetween, two functionswhich belong to the same germ. This
sloppiness is somewhat justifiedby the uniqueness propertystated above, for
two functionswhich belong to the same germ and are definedon the same do-
main D must differby a functionin I,, which means that they agree on some
neighborhoodof z, and thus must agree on D.
We may now define the stalk space (espace etale) of germs of holomorphic
functionsto be the set S = { (z, [f]2)if is holomorphicat z E C-} togetherwith
the natural mappingp fromS to C- definedby p((z, f])) =z. We call p-'(z) the
stalk at the point zEC-; it is simply a copy of O, the algebra of germsof holo-
morphicfunctionsat z. The stalk space S is thus the disjoint union of the stalks.
Intuitively,we shall picture S as a space of interpenetratingsheets lying over
Cn, withp projectingS onto C- (Fig. 1).
S
S
r ~~~~z
FIG. 1
To make this intuitivepicture more precise,we liftthe topologyof C- back

to S, to make S into a topological space. For each open set U in Cn and each
functionf which is holomorphicon U, we defineV(f, U) { (z, [f]2) Iz U }. Each
such V(f, U) is contained in S, and the collectionof all such sets covers S, for
if (z, [fo]I) CS, fomust be holomorphicon some neighborhoodUo of z and(z,
[fo]z)C V(fo,Uo). Furthermore,V(fi, U1)fl V(f2,U2) = V(f, U) where U= {zE Us
19701 WHAT IS A SHEAF? 683
r)U2j [fi]Z= [f2]Z} and f=fi Iu=f2l u. Thus the sets V(f, U) forma basis for a
topology on S, and relative to this topology,the projectionp is a local homeo-
morphism.That is, foreach basis neighborhoodV(f, U) in S, theone-to-onemap
PI vU,U) is a homeomorphismonto U. For if we let pf,udenote PI vU,U), and if N
is an open subset of U, then p7j(N)= V(f, N) which is open in S, while if
V(f, U') C V(f, U), thenpf,u(V(f, U')) U'. The topologyon S is uniquelydeter-
mined by the requirementthat the projection p be a local homeomorphism.
This topological space S, togetherwith the local homeomorphismp which
projects S onto C-, is called the sheafofgermsof holomorphic functionsover the
base space C". As the agriculturalterminologyimplies,we thinkof the sheaf as
a bundle of stalks (Fig. 2), each with a fullhead of germs(or, if you wish,seeds,
or grain).
I D
FIG. 2
We can show that the stalk space S is Hausdorffas follows:points of S may

differeitherbecause theyare on different stalks, or because theyare on different
levels of the same stalk. In the firstcase, the projectionsof two points p, qES
differin C-; so, since C" is Hausdorff,there are disjoint neighborhoodsof p(p)
and p(q) which may be liftedback to S. To be specific,ifp = (z, [f1])and q = (w,
[g]w)wherezOw, then thereexist disjoint open neighborhoodsUz, Uwof z and
w respectivelyand on them holomorphicfunctionsf [f]z and gGE[gbw,respec-
tively,so that V(f, U,) and V(g, U.) are disjoint neighborhoodsofp and q.
The second case is a bit more complex, since it depends on the uniqueness
propertyof holomorphicfunctions.If p = (z, [ff]) and q = (z, [g],) are different
points on the same stalk, then [f]Z0 [g],; so there must exist differentholo-
morphicfunctionsfE [f],and gE [g]2which are both definedon some neighbor-
hood U of z. We claim that V(f, U) and V(g, U) are then disjoint neighbor-
hoods of p and q, forif (w, fh]w)EE V(f, U)fl V(g, U), then wC Uand [f]-= [h]w,
= [g],. But as we observed above, the uniqueness propertyof holomorphic
functionsimplies that the two functionsfand g with the same domain U which
belong to the same germ [h] must be identicalon U. But theyare not identical
on U, since [f],# [g].. So V(f, U)C V(g, U) = 0, and thus S is Hausdorff.
There is still anotherconsequence of the uniqueness propertyof holomorphic

functionsthat can be used to furtherilluminate the sheaf of germs of holo-
morphic functions.The uniqueness property may be roughly interpretedas
saying that the global behavior of a holomorphicfunctionis uniquely deter-
mined by its behavior on any open set. This makes meaningfulthe vague ques-
tion of identifyingthe largest domain to which a given holomorphicfunction
can be extended. In the classical study of analytic functionsthis question led
to the concept of a Riemann surface, or more generally to complex analytic
manifolds.
_~~~
C"
FIG. 3
A manifoldis, essentially,a topological space whichis locally homeomorphic

to complex Euclidean n-space C". To be more precise,we will call a topological
space X locallyEuclidean (ofdimensionn) if everyx zX is contained in an open
set U, which is homeomorphicunder a mapping7r.to some subset of C-, where,
furthermore, the coordinatepatches U. are coherentin the sense that for each
x, yEX, 7r.o7r;,'is a homeomorphismbetween7ry,(U.Cn U,) and 7r.(U.CnU,). The
firsthalf of this definitionguarantees that X is locally like C-, while the second
condition requires that the locally Euclidean patches overlap so as to forma
coherentEuclidean structureon all of X. We shall call each pair (U., 7r$) a local
coordinatesystem,since 7r-1 liftsthe coordinate systemof C- back to U (Fig. 3).
Since the same topological space may be covered by several differentcol-
lectionsof coordinate systems (Us, 7rs),and since we do not wish to distinguish
1970] WHAT IS A SHEAF? 685
between two covers which provide essentiallythe same coordinate structureon

X, we definea manifoldto be a locally Euclidean topological space in which the
collectionof coordinatesystems (U., 7r.)is maximal with respect to the defining
properties for a locally Euclidean space. Since each locally Euclidean space
generates a unique manifold,we shall oftenreferto locally Euclidean spaces as
manifoldseven ifthe collectionof local coordinatesystemsis not maximal. Other
types of manifoldsmay be produced by projectingto real Euclidean space Rn
instead of to C- or by requiringthat the homeomorphisms-xw o wX'1be analytic
or Coo (infinitelydifferentiable);such manifolds are naturally called analytic
manifoldsor CIOmanifolds.
A functionf: X-* Y fromone analytic manifold to another is called holo-
morphicif foreach x and y, 7rn, of o 7r-I is holomorphicon its domain, which is
7r(Uf-1(Uy)). In the special case, where Y= C1, the identity map i: Y-*C'
is used to definethe local coordinatesystems.So a holomorphicfunctionffrom
the analytic manifoldX to C' is characterizedby the propertythat f o 7r-1is
holomorphicon x((Uz).
It should be clear from this description that the sheaf of germs of holo-
morphicfunctionscan be regardedas an analytic manifold,using the projection
p to definethe local coordinatesystems.It is a particularlyimportantmanifold,
since on it we can definewhat is known as the universalholomorphic function.
This is the mapping F: S- C defined by F((z, [fU]))=f(z). F is clearly holo-
morphic since for each local coordinate system (U, p u), we have Fo (p u)-1
=f p(u)wheref:p(U)->Cis holomorphic.
5
D go..C
f
FIG. 4
F is universalin the sense that the behavior of all holomorphicfunctionson

Cnis subsumed in that of F. In particular,wheneverf:D-?>C is holomorphic
(where D is a domain in C-), we can factorf throughthe sheaf S as follows:
thereexists a unique functionf: D->S such that F o ] =f. Clearly] is definedby
1(z) = (z, [f]z), so the associated diagram (Fig. 4) is commutative;Jis continuous
since -,( V(f, U)) = U.
With this structure,we can now describe the domain of holomorphyof a
given holomorphicfunctionf-that is, the largest domain to which f can be
uniquely extended. It is the connected componentE of S which containsJ(D).

(Of coursef(D) is connectedsince it is the continuousimage of a connectedset.)
AlthoughD is not literallya subset of E, it is imbedded by fin E, and thus the
universalholomorphicfunctionF is the extensionto E of the functionf.
2. The sheaf of local rings. Let A be a commutative ring with 1, and S a
multiplicativelyclosed nonemptysubset of A with 0q1S. We constructfromA
and S a ringAs called a ringof quotientsofA, in which the elementsof S have
multiplicativeinverses.On the set
A XS = {(a,s)| aE A,s S}
we definean equivalence relation (a, s)r--(b,t) if and only if there exists rEES
such that (at-bs)r = 0. We also definetwo operations,
(a, s) + (b, t) = (at + bs, st) and (a, s)(b, t) = (ab, st),
whichare compatiblewith the relation.We denote by As the ringof equivalence

classes with the induced operations. As in the ring of integerswith the set of
nonzeroelementsas S, the equivalence class of (a, s) is denoted by a/s; thus we
call S thesetofdenominators.
The 0 of As is 0/s (any s in S), the identityis s/s; and if sES, s'1=1/s.
There is a homomorphisma:A--As defined by a(a) = as/s, which is indepen-
dent of the choice of s. Of course if A is an integraldomain, a is one-to-onebe-
cause Ker a = jai sa=0 forsome sCES} .
If I is an ideal of A, the ideal a(I) in As can be representedby
a(I) = {a/sa I, s E S},
and we shall writeIAs fora(I). This functiona on the set of ideals of A defines
a one-to-onecorrespondencebetween the set of primeideals in As and the set of
primeideals in A whose intersectionwith S is empty [12, p. 223 ].
Since we may describe a prime ideal P in A as one whose complementis
multiplicativelyclosed, we may formthe ring of quotients of A whose set of
denominatorsis the complementof P. We shall denote this ringof quotientsby
Ap, and call it the local ringof A at P. The ringAp has only one maximal ideal,
PAp, since clearly P is the largest prime ideal of A with the propertythat its
intersectionwith the complementof P is empty.
These local ringswill be the stalks forthe sheaf of local ringsand the set of
primeideals of A will formthe base space. This space is called the spectrumof
A, denoted by Spec A, and is topologized by taking as a basis forthe topology
all sets V-= {PESpec AIx BP} wherexCA. Then V,=Spec A, V0=0, and
Vn VV, = V-,,,;thus { VX}IeA is a basis. Since UXeMVU= {P (x)xeM tP } where
(X)XEMis the ideal generatedby the subset M of A, any ideal I of A definesan
open set VI = I PI Id?P }, and every open set U is of this formalthough I is not
uniquely determinedby U. A closed set, then,is a set of primescontainingsome
fixedideal, so that a point P in Spec A is closed if and only if it is a maximal
ideal. For most rings,therefore,the base space is not even T1; however,Spec A
is always To.
It is possible to definea ringof quotients with respect to the complementof
a primeP because it is multiplicativelyclosed, but the complementof a union
of primes is also multiplicativelyclosed. Hence, with each nonemptyopen set
U in Spec A we may associate the ring of quotients Au = { a/s IPEE U=s ErP }
whose set of denominatorsis the complementof the union of all the primes in
U. If U and V are open sets in Spec A such that UC V, we may definerestriction
homomorphisms pu,v:Av-?Au as follows:if a/sGAv, s is not an element of any
prime ideal P in V, and so a fortiorinot an element of any primeideal P in U.
Hence a/s is also an element of Au. We definepu,v(a/s) = a/s, but this map is
not the identity,or even one-to-one,since the equivalence classes which are
used to definethe ringAu are largerthan those used to defineAv. The kernelof
Pu,v consistsof those elementsa/s such that a is a zero divisorwithrespectto an
elementin one of the primeideals of V whichis not in any elementof U.
An importantpropertyof pu,v is the commutativityof the diagram in Fig.
5, where au:A->Au takes a to a/i. Now pu, is uniquely determinedby this
propertyand from this it follows that pu,u is the identity map and that if
UC VC W, then pu,w=pu,v o pv,w.This system,consistingof Spec A, the rings
Au, and maps pu,v:Av->Au when UC V, is called a presheafover Spec A.
Besides the maps pujv correspondingto pairs of open sets forwhich UC V, we
AvyA
A Pu.v A/PV
A ~~ ~~~A
FIG. S FIG. 6
can define maps pP,V:Av--*Apwhen PG V. If V is open and PG V, we define

pp,v:Av--*Ap,by pp,v(a/s)=a/s, which is possible since a/sCEAv implies sEP.
The map pP,vis the unique map which makes the associated diagram commute
(Fig. 6). As before,it followsfromthe uniqueness of pP,Vthat
PP,W PWu = pP,U if P E W C U.
If U is open in Spec A, and u is any element of Au, we may treat u as a
functionfrom U to the stalk space S='U {ApIPESpec A I by definingu(P)
-pp,u(u) GAp forPE U. If VC U, we call pv,u(u) the restriction of u to V. If
uCAu and vCAv, and if there is an open set WC U() V such that pw,u(u)
-Pw,v(u), we say u and v agreeon theopenset W, forifPE W, then
u(P) = pP,U(u) PP,Wo pMU(u) = PP,Wo pw. V(V) = ppv(v) = v(P).
If uEAu, either the functionu or its image u(U) in S is called a sectionof the
presheaf.The set of sectionsu(U) covers S, forifa/sEAp thens is a denomina-
torin Av, whereV {QIs EfQ}, and a/sEAv is a sectionover V whoseimage
at P is a/s.
The collectionof all these sections u(U) is a basis fora topologyon S, since
the intersectionof two sections is a union of sections. For, suppose uEAu
and vEAv are sectionsover the open sets U and V. If u(U)Cv(V) = 0 thereis
nothingto show. If xEu( U)flv(V), x pP,u(v) =pP,v(v) EAp forsome PE UN V.
Then u a/s, wheres is not an elementof any elementof U, and v = a'/s', where
s' is not an element of any element of V. Since a/s a'/s' in Ap, there exists
t*EP such that t(as'-a's) = 0. Let W-= QItd(Q}. Then a/s= a'/s' in Awu,v
and, since the diagram in Fig. 7 commutes,the section over Wn UN V defined
by a/s=a'/s' is a subset of u(U) and of v(V) and it is a neighborhoodof x.
AV
Awn)unv _ Ap
Au
FIG. 7
In this topology the projection p is a local homeomorphism,for if xEAp

thereis a section v over V throughx forsome V, and the restrictionof p to the
section v(V) is one-to-oneand onto. If UC V is open and xE U, the restriction
of v to U is also a section throughx, and thus u(U), whereu =pu,v(v), is open in
u(V); so p is continuous. If N is an open subset of v(V) it is a union of sections
over open subsets of V and p(N) is the unionof theseopen subsets.
The topologicalspace S, togetherwith the projectionp: S->Spec A, is called
the sheafof local ringsover Spec A. If U is an open subset of Spec A, any con-
tinuous functionf: U-*S such that p o f is the identityon U is called a section
of the sheaf S; the set of all sectionsover the open set U is denoted by r (U, S).
The relationbetween the sectionsof the sheaf S and the sectionsof the presheaf
(that is, the elements uEAu) is rather subtle, for even though each presheaf
section may be thoughtof as a (continuous) functionon U which is a local in-
verse forp, two anomalies may occur. It may be that two differentelements
u = a/s and v = b/tof the ringAu yield the same functionunder the interpreta-
tion outlined above, or there may be functionsin r(U, S) which cannot be de-
rived fromany section uEAu. Thus the interpretationmap fromAu to P(U, S)
need not be eitherone-to-oneor onto, thoughif U is a basis set (that is, one of
the form V. forsome xCA) this map is an isomorphism[5, No. 4, p. 86]. For
this reason, the open sets of the formV, are oftencalled distinguishedopen sets.
If the base space Spec A is not Hausdorff,certainlythe sheaf S cannot be
Hausdorff.But even if Spec A is T2 it may well happen (although it is difficult
to visualize it on Hausdorffpaper) that two distinctsections over U, u and u',

may agree on a properopen subset V of U (Fig. 8). If P is in the closure of V
but not in V, u(P) and u'(P) are distinct points of the sheaf but cannot be
separated because every neighborhoodof P contains points of V on which u
and u' agree.
i(U)
wfo^?- (V) U'(V)

A
g b~~~~~(P);-
{ \
IM
FIG. 8
A special case. In order to provide a more geometricinterpretationof the

sheaf of local ringsas well as a glimpseof the originof the subject, we shall look
at some special rings. Let Q be the fieldof rational numbers,Q[xi, . .. , xnJ
the ring of polynomialsin n variables with rational coefficients, and I an ideal
of Q[x1, * - * , Xnj. Let A be the quotient ringQ[xi, * * * , xnj/I, which we may
write Q[sic, . . . I xn, where xi=xi+I, i-1, *. * , n. If E= {(x1, * * *,xIn)
(Cnp(xI, *... , xn)=0 forall pEI}, i.e., E is the intersectionof the zero sets
of all polynomialsin the ideal I, we may interpretA as a ringof complexvalued
functionson E. For since p1(x) -P2(x) GI, wheneverpi() =p2(9), PI(X) =P2(X)
impliesp1(x) =p2(x) if x=(xl, . . . x) CE.
We may also construct,from the pairing which takes pEA and xEE to
p(x)CEC, a one-to-onecorrespondencebetween the points in E and homomor-
phisms from A to C. If x= (xi, * * - , xn)CE the function which takes p to
p(x) is a homomorphismfromA to C, and conversely,if4.:A- Cis a homomor-
phism, x = (5 (li), * * * , (xC)) E, since if p E I,
p(X) = p@f(tD.. ,
f(tn)) = 4.(p(i 1) i... ,
p(n)) 0( ==
q'OM
Since C is an integral domain, the kernel of the homomorphismdefined by

xeE is a primeideal in A. Conversely,it may be shown that ifP is an element
of Spec A it is the kernelof some homomorphismfromA to C [13, p. 164]. So
we may regard Spec A as a set of equivalence classes of E under the relation
x/_-x'if and only if x and x' are zeros of the same polynomialspEA. Thus if
x/x', p(x) =p(x') forall p in A, so A may also be interpretedas a ringof func-
tions on Spec A.
That an elementp of A not be in a primeP means that p is not in the kernel
of the map A-?A/P, or that p is not zero on any point x in E forwhich the
kernel of the correspondinghomomorphismis contained in P. Thus, the local
ringAp at P, which contains the inversesforall pEIP, is the ringof all rational
functionsdefinedlocally, i.e., in some neighborhoodof P in Spec A. The set of
such functionswhich vanish at P is the only maximal ideal in this ring. The
ring Au of presheafsections over an open set UCSpec A is a ringof functions
definedat every point of U. (If U is a distinguishedopen set, Au is the ringof
all such functions.)The kernelof a restrictionmap pu.v:Av-+Au (where UC V)
is the set of functionswhich vanish at every point of U, though they may not
vanish on all of V. The elementsof Au which are not restrictionsof functions
in AVare inversesoffunctionswhichhave zeros in V U.
The topology of Spec A induces a topology in E, the weakest one in which
the identificationmap fromE to Spec A is continuous. In this topology, the
closure of a point xGE is the set of all points in E which satisfythe same poly-
nomials as x.
We choose a particular ring A to study in more detail. In Q[x, y], let I
- (xy), and
A = Q[x,y]/(xy) Q[X,y],
wherexy=xy+ (xy)= (xy)= 0. Then
E ={(Z,W)EC21iZW } = I(z,0)IzcU}'J{(0ow)I w c},
the union of the complex coordinate axes (planes) in C2.
The ideals (xt)and (y) in A are prime but not maximal, since Q[x, y3/l()
=Q [y] Q [y], which is not a field.The points (z, w) in E whose corresponding
ideal is (x) are those of the form(0, w) where w is transcendentalover Q, and
6imilarly,the points correspondingto the ideal (y) are of the form(z, 0) for
transcendentalz, since the homomorphismof Q[t, y]->C given by X-0, y->w
has kernel (x) if and only if w is transcendental.Further,(x)fn (y) = (0) in A,
any prime ideal contains either (x) or (y), and the set of zero divisors in A is
(x)\J(y).
4: t, 9
The maximalprimeideals in A are kernelsof homomorphisms
-+C for which the image is a field,so that O(X) and +(y) must be algebraic,
and since +(x)q+(y) =G(xy) = 0, one of 4(x) and ?5(y)must be zero, and the other
algebraic. Such points (+(x), 4(y)) in E are either of the form (z, 0) with z
algebraic, or (0, w) with w algebraic. The kernels of the homomorphismsto
(z, 0) and (0, w) are the ideals (p(x), y) and ((x), q(y)) where p is the minimal
polynomialof z (and q of w). These ideals togetherwith (x) and (y) are pre-
cisely the points of Spec A.
The points of ECC2 are divided into equivalence classes correspondingto
the ideals in Spec A: to each of (x) and (y) there correspondsa class of tran-
scendental points,while to each maximal ideal (p(xc),y) (wherep is irreducible)
there correspondsthe set of algebraic points I (z, 0) } where z is a root of p.
In the topology induced on E by Spec A, the closure of a transcendental
point (z, 0) is the z-axis and the closure of an algebraic point (z, 0) is all (z', 0)
where z' is also a root of the minimalpolynomialof z. A basis forthe open sets
in Spec A is the collection of all sets of the form V P{PESpec A I pEP . In
the special case p=
Vt= { P CSpec A I EqPP } (9, p(x)) I p CA is irreducibleor zero,

The open set VgCSpec A correspondsin E to the complementof the w-axis.
Similarly,V, may be regardedas the complementof the z-axis. The complement
of a finiteset of algebraic points correspondingto the maximal ideals
(Xxpl(y)); - *- X
(1 pk-(Y))j (pk',1(9)2 Y)i
.. * (Pn(9) iY)
is Vvl(9)n . . . n
Vpf(x) Since everypolynomialp EA = Q [., yJcan be writtenas
p=-p(x) +P2(y) -c, wherepI(0) = P2(0) =p(0, 0) = C, the pointsin Vpare (z, 0)
wherep1(z)= 0, and (0, w) wherep2(W) = 0.
Geometrically,the ringA = Q [x,y] is the ringof polynomialsdefinedevery-
where on E, the union of the complex axes. At a point (z, w) in E (where either
z or w is 0) correspondingto a prime ideal P, the local ring Ap contains all
rational functionswhose denominatorsdo not vanish at (z, w). The local ring
AX is the local ring at any (0, w) where w is transcendental.Algebraically,we
have the set of denominators S=A - (X) = {Pj P2(Y) #01 where, as above,
p=pI + p2-c. The kernelI of the homomorphism a: A--A is { qEA 1f3pES
such that pq = 0 } If PI+p2-c is a zero divisor,c = 0; if such a point is also in S,
we have P2(y) #0 and p2(0) =0, SO it is an element of (y) - {0 }. Therefore
I=(x). Thus A/I=Q[x, y]/(x)=Q[y] and the image of S in Q[y] is Q[y]
- {0 }; so Ag is isomorphic to the field of rational functionsin y. Similarly
Ay= Q(x).
For the local ringat a maximal prime other than (x?,y), forinstance P= (x1
y2-2), a similar willlead to theringAp= {p/qEQ(y)I
argument (y2-2) does
not divide q }. That is, AP contains inversesforall functionsthat do not vanish
at (0, V/2),while two polynomialsin x and y which agree on an open set con-
taining(0, V/2)are identifiedin Ap. Thus the map a:A-+Ap is neitherone-to-
one nor onto. Now if P = (x, y), the set of denominatorsforAp is exactly the
set of polynomials
pl+p2-C forc # 0. So in thiscase the map a: A-Ap is an
inclusionwhich is not onto.
Finally, we describe the ringsof presheafsections Au, foropen sets U. If U
is the complement of a finiteclosed set of maximal primes (x, pi(y)), * *
(x, Pk(y)), (Pk+i(x), y), . , (pn(x), y) the set of denominators for Au is

{pEA IpEXEUpeuP wvhich }, is the set of all products p'l . . . pcnwith nonzero
constant term. The map from Au->Ap when PECU will have kernel (x) if
P = (x, p(y)) (where p is any irreduciblepolynomialexcept y); (y) if P = (y) or
(y,p(x)) (wherep(x) O x); and (O) ifP = (x,y).
If U= Vx is the complementof the w-axis, the set of denominatorsforAu
is the set of polynomialsyp(y)+Ctn forpEQ [y] and n? O.Thus
Au = {p(x)/xnI p E Q[x]}.
Since A(9)-Q(x), the map p(,),U:AU-*A(y) fromthe ringof sections to the stalk
at (y) is one-to-onebut not onto. For a maximal prime(y,p) E Vg,wherep (x) O x,
p(y,p),u:Au--)A(y,p)is an inclusionsince x does not divide p. If U is the comple-
mentof the closed set (x, pi(y)), * *(, pk(y)), then VgC U, and the re-
strictionmap pvg,u:Au-+Avg has kernel (y).
The ring of sections over Spec A is A, for every element of A vanishes at
some point in Spec A. Thus for every kind of open set U in Spec A, the ele-
ments of the rings Au are the rational functionsdefinedat every point of U.
In particularthe functionsx and x+y are sections on the open set Spec A.
They agree on the properopen subset Vg,for (y) is the kernelof the restriction
map pvg,Spec A, but do not agree on (x, y), whichis in the closureof Vg. Thus the
points x and x+y in A(g,y though distinct, cannot be separated by sections
since any open set containing (x, y) must intersectVg. Thus this sheaf fails to
be Hausdorffboth verticallyas well as horizontally.
forms.Let p EX whereX is a Cc?manifoldover
3. The sheaf of differential
Rn.We denote by C7' the set of all functionsfromX to the real line RI whichare
Cc?in some neighborhoodof p. Clearly Cp1is a vector space over R in which the
sumf+g is definedon the intersectionof the domains off and g.
A tangentto X at p is a linear functiont: Cp ->R such that
t(fg) = t(g)*g(p) + f(p) *t(g) wheneverf, g E Cpi.
The set of all tangentsto X at p formsa vector space over R which we call Xp,
the tangentspace to X at p. If 'y: [0, 1]->X is a Cc?functionsuch that #y(to) =p,
and iffE Cp, then f o y: [0, 1 ]XR. If (f o y)' is the derivative of f o0,
the function *: Cp;-Rl definedby y*(f) (f oy)'(to) is a tangent.To interpret
thisgeometrically,we considera particularlocal coordinatesystem(UUp, 7rp)where
trp:Up-+Rn.If 7 :Rn-->R1 is the projection functiondefinedby 7ri(ti,* * *, tn)
=ti, we write xi=iri o07rp; then 7rp(p)=(xi(p), * * *, x,(p))CRn. If es
= (0, 0, * * *, 1, * * *, O)ERn, we think of the curve y(t) =w7r71(7rp(p)+tei) as
the ith coordinate axis in Up since (d/dt)(rp o y) o = ei. Then the tangent y*
satisfies
d
Y*(f = (fo? y)'(0) = -(f o 7r'o1 0 ,rp
o Y)
dl 0
0 (fo
(f r-1) a(fo r1)) e a(fo r;)
ad1t, atn J ., (p) t p (P)
Thus y*(f) is usually denoted by (a/ox&)f;the tangents (o/Oxi), * * , (O/1x.)

forma basis for the vector space Xp [8, p. 7]. We thinkof X, as an n-dimen-
sional hyperplanetangent to X at p (Fig. 9). If
'Y*~ ai
Ox;
then 'y*(f)is a 'grad f, that is, the derivative off in the directiona (al,
an) ERn.
Xp
FIG. 9
If q X-* Y is a C??mapping of manifoldswe definethe differential of( to be

the linear map dk :X1,-+YO(p)defined as follows: if tEXp and fEEC;,C) then
[d4(t) ](f) = t(fo 'k).If (UUp,r7rp) is a local coordinate system at p EX then xi
=-7r o wr definesa COO-mapfromthe submanifold Up to the manifoldR1. Hence
the differential dxi of xi is a linear transformationfromXp to R (1,).Since R' is
isomorphicto R' forany tSR1, we may considerdxi as an element of the dual
vectorspace X* of Xp. Since dxi (a/ax1) = bij (where 3qi= 0 ifi ij, and 1 ifi =j),
the differentialsdxi, **, dxn form a basis of X* dual to the basis aldrx,
*. , a!ax" of X1.
Whereas each differential dxi is a functionof just one variable, a differential
formin general is a functionof several variables. To be precise,a differential- k-
form0 on Xp is an alternatingk-linearfunctionfromk-tuplesof elementsof Xp
to R1. (6 is k-linearif it is linear in each variable separately, and alternatingif
f(t1 . . . tk) = sgnoa f(to(l) .
. . tk)) where a: { 1 . . . k}- {1 . * k} is a permu-
tation and sgn a' is + 1 if a-is even, and -1 if a' is odd.) We denote by the wedge
productdxi1A . . . Adxi1the unique k-linearformon X, definedon a basis for
the set of k-tuplesby
dxi1A . . . A dx*,
where 8=1 if (i1, , ik) =(ji, , jk) and 0 otherwise. The set of all k-
linear formson X, is an nnkdimensional vector space over R and the nk forms
dxi,A * * * Adxi,, where ije {I1, , n}, forma basis for this vector space.
The set of differential k-formsis a subspace of the set of all k-linearformsand it
has as a basis the k-linearformsdxilA .. . Adxik where ij <i2< ... <ik. We
call such a sequence of indices an increasingk-tuple.Thus the dimensionof the
space ofdifferential k-formsis(t) ifk ?n and 0 ifk>n.
Let Sk denote the set of all increasingk-tuplesof positive integersless than
or equal to n. For xeSk we denote by dx, the k-formdxi,A . . . Adxi,, where
s = (i1, * * *, i*). Now let UCX be an open set and 0 a functionwhichassigns to
each pE U a differentialk-format p, 6(p). Thus, once having chosen a local
coordinate system at p, so the formsdxl, dx2,... dxnare defined,we may ,
write
O(p) = E a,(p)dx,,
*eSk
whereeach a,(p) is the coefficientin the expansion of 0(p) in termsof the basis
{dxs}lsE.k* Thus as may be considered as a functionfrom U to RI. Since each
k-format p, we call 0 a C' k-formon U ifeach functiona. is
0(p) is a differential
Coo; we denote by Qk(U) the real vector space of all CIOk-formson U. If UC V
then there is a linear transformationpU,V:2k(V)- >Qk(U) definedby restriction
of the domains of 0(=2k(V). The collection {Qk(U) } for U open in X together
with the linear transformationspu,vis called the presheafofdifferentialk-forms.
Since every paracompact manifoldhas a CX partitionof unity subordinate
to anyopencovering{ U4m
} [8,p. 85], presheavesoversuchmanifoldssatisfythe
followingspecial property: If U= U U,,where Ua is open in X, and if0ayE3-k(Ua)
are coherent in the sense that the restrictionsto UaCnUp of 0atand Opagree
(whenever Uan Up50) then there exists a unique 0EEQk(U) whose restriction
to each Ua is 0a. Certainlyif {fa} is a C* partitionof unityfor { U,,,}so that
vanishes off Ua, and Ef,, = 1; so we may define 0 to be E2faoa.
.f,: U-?+R', fG,
Any such presheafis called a sheaf, so when X is paracompact, we will call
thesystem{I k(U), pU,V} the sheafofdifferential k-forms. If Qk (p) is the set of
differentialk-formsat p, we may think of Qk (p) as the stalks of the sheaf S:
S = U,,ExEk(p). The projection p: S--*X assigns to each differentialform at p
the point p. The C* k-form0EQk(U) is a section of S, and the collection
{ 0(U) CSI U is openinX } formsa basisforthetopologyon S.
4. Sheaves: General definition.Each of the three previous examples re-

flects a differentfacet of the general concept of a sheaf. To emphasize this
1970] WIIAT IS A SHEAF? 695
Sheafof Germs Sheafof

of Holomorphic SheafofLocal Rinigs Differential
Functions Forms
Germn U]t a/sEAp

StalkSpace S = (XI[fIt) I S_U{AJPI PSpec AI
Projection P: S-+-p: S-+Spec A
Stalk I (z, [f].)JfE-A4 Ap(localring)
Ba-seSpace co {
SpecA IPIPis prime X=z-C'mianifold
ideal in A
PresheafSection als EAu 0 E nL-(U)
Restriction
Homomorphism pu,v:.Av--~Au pu.v:01k(V)-+f14(U)
FiG. 10
varietyand to providea coherentframework forthesubsequentgeneraldefini-

tion,we summarizein Fig, 10 the threesheavesalreadydiscussed.
The sheafof germsof holomorphic functionswas definedto be the stalk
space S togetherwitha topologyand a local homeomorphism p onto Cn. The
sheafof local ringswas definedsimillarly,
thoughin thiscase we also 'Identified
the systemconsistingof the ringsAu and the restriction homomorphisms
pu,v:Av-?-Auas a presheafof rings.In the thirdcase, thesheafof differential
k4forms was simnplythepresheafofdifferentialk-formswhenever thebase space
X was paracompact.The recognition of theequivalenceof thesetwo descrip-
tionsconstitutesthe beginning of sheaftheory,We now introducedefinitions
to formalize
thesetwoapproachesand provethedefinitions equivalent.
DEFINITIONI. Let (X, r) be a topological
space,and let C,be a class ofsimi-
lar mathematicalobjects (e.g., abelian groups,modules,rings).Let F be a
function fromr to C and supposeforeach pair U, Ver forwhichUC: V there
is a map (e,g., a homnomorphism, modulehomomorphism, or isomorphism)
pu,v:F( V)-->F(U) whichpreserves thestructureoftheobjectsof e. Ifpu,vo pv,w
- pu,wwheneverUC VCW, and ifpu,uis theidentity functionon FU,we de-
finethefunction F together withtherestriction
mapspu,vto be a presheaf overX.
The elementsof F(U) ar-ecalled sectionsof F over U. (In thelanguageofcate-
gorytheory,a pres-heaf is a contravariantfunctorfromthe categoryof open
sets and inclusionmaps of X to somiecategory-ofobjectsand niorphismns.) A
sheafis a presheaf
whichsatisfiesthefollowing twocoherence axioms:
1. If I U,,} is a familyofopensetsin X, if U= UU,,,and if th~esections,
tEEF(U) agreeon eatchU,,,(i.e,, ifpu.,u(s) = pu,,u(t)foreacha) , thens-=t.
}
2. If I U,, Itsa familyof opensetsin X, if U = U U,,,and 'if the sections
sa E F( U,,) are coherentin the sense thatthe restrictionsof s,, and sp to UaC\n
U#
agree (i.e., ifpu.u,u,,U(s,,)
= whenever-UaC\U#0 0) thenthereexists
pu.-,u,,u,(s#)
a sections C-F( U) such thatfor each a, pua,u (s) = Sf1.
DEFINITION II. Let (X, r) be a topological space, and let eDbe a class of
similarmathematicalobjects (e.g., abelian groups,modules,rings). A sheafover
X is a triple {S, ir,X}, where S is a topological space and 7r:S-*X is a local
homeomorphism(i.e., a map such that each point pETS has a neighborhood U
on which rIu is a homeomorphism)such that each stalk 7r-'(x) EE , and each
operation is continuous as a function from U,x(7r-N(x)XKr-1(x)) (with the
topology induced fromSXS) to U,r=xr-(x) =S.
We shall call a sheaf of type I a sheaf of sections,and a sheaf of type II a
sheafofgerms.The relationshipbetween these two types of sheaves is precisely
as illustratedby the precedingexamples.
To be specific,suppose F is a presheaf (of sections) over W; we construct
the correspondingsheaf of germs by defininga germ at xCX to be an equiva-
lence class of Ax=UXeuF(U) under the relation s=F(U)-tGF(V) if pw,u(s)
-Pw,(t) for some WC Ur) V. Thus the germ of a section sEF(U) at a point
x E U is the collectionof all sections tC=F( V) which agree with s on some neigh-
borhood V of x. We denote, as usual, the germof s at x by [sJ.,and let the stalk
space S be { (x, [s ].,)Is F(U), where x U }. The topology on S is generated
by neighborhoodsof the form V(s, U) { (x, [s]j xCEU}, so the projection
-r:S->X becomes a local homeomorphism.By interpretingthe sections as
functions,the so-called restrictionmaps Pu,v really are restrictionsand the
topology on S is the strongestrelative to which the sections are continuous.
(The topology on S can also be characterized as the quotient of the topology
on Uue= (UX F(U)) under the equivalence relation induced by , where each
U carries the subspace topology and F(U) is discrete.)
Each stalk 7r-'(x) clearly inheritsthe operations of the F(U) and each such
operation is continuous. For example, if each F(U) is an abelian group under
addition and if [s], and [t]rE7r-'(x), then [s]z+ [t], is definedto be [s+t]x. If
V(s+t, U) is a neighborhoodof [s+t],, then the inverse image of V(s+t, U)
under + contains { (r, q) IrE V(s, U), qE V(t, U), ir(r)=r(q) }, an open set in
UEx(7r-1(x) X7r-1(x)). Thus {S, 7r,X} is indeed a sheaf of germs.
Conversely,suppose {S, 7r,X } is a sheaf of germs (perhaps one constructed
as above fromsome presheaf). If U is open in X, let F(U) be the collection of
continuousfunctionss: U-*F such that iro s is the identityon U. F( U) inherits
the algebraic structurefrom the stalks by pointwise definitions,and the re-
strictionmaps Pu.v are just that-the restrictionof s from V to the subset U.
The firstcoherenceaxiom forsheaves is satisfiedtriviallysince the sections are
functions,and the second is satisfiedsince the F(U) contain all continuous func-
tions from U to F which are inversesof ir.
Now if { F, xr,X} is a sheaf of germs,the sheaf derived fromits (pre)sheaf
of sectionsis canonicallyisomorphicto F. However, if S is a presheafof sections,
the presheafof sections S' associated with the sheaf of germsderived fromS is
generallydifferent fromS, forsince S' is a sheaf,it may have moresections than
S (in order to satisfythe second coherenceaxiom), while some sections which
were distinctin S may be identifiedin S' (because of the firstaxiom). Of course,
ifS is a sheaf,thenS' is naturallyisomorphicto S, so in thissense the two defini-

tions of a sheaf are essentiallyequivalent.
A common and convenient alternative to the constructionof equivalence
classes is the use of direct limits. If F is a presheafof sections over X, and if
xE X, the restrictionof F to the neighborhoodsof x formsa directedsystem
({F(U)}zeu, {pu,v}xeucv) since Pu,vopv,w-pu,w, whenever xCUCVCW.
A few elementsof such a system may be representedby the commutativedia-
gram of Fig. 11.
Pu.wi
F(WI) moF(U)
P V1.WJ ~ F(Vi) PU,V
.
-_ F(W2) _ F(V1nV2)
PV2,W2 Pv1-V2,V2
FIG. 11
The direct limit of this system is, roughlyspeaking, the firstobject which
can appear to the right of the diagram. Specifically, an object F. together
with maps pu:F(U)-*Fx for each F(U) is called a directlimit of the system
({ F(U) }xeu, {Ipu,v}xucv) provided that
(i) whenever UC V the diagram in Fig. 12 commutes,
(ii) F. is universal with respect to property (i)-that is, if (Gx, {Io,}) also
satisfiesproperty (i), then there exists a unique map q: F.--G. such that for
each u: F( U)>Gx, 0u=l o pu.
Pv
F(V) - s F.
F(U)
FIG. 12
Condition (i) makes explicit the idea of "appearing on the rightof the dia-
gram" while condition (ii) asserts that F- is the firstsuch object. It follows
triviallyfromthese conditions that the direct limit is unique (up to isomor-
phism), so we denote it by LimxeuF(U).
Consider now the stalk of germs at x derived from the presheaf F. If Pu
denotes the map from F(U) to the stalk 7r-'(x) defined by pu(s)=(x, [s]x)
(wherexE U), then 7r-'(x) ='zXeuF(U) since wheneverxE UC V, pv Pu 0 pu,v
and 7r-'(x) is universal with respect to that property.To prove this last asser-
tion we assume that (G, {au}) is another direct limit and observe that if both
EF(U1) and t2E F( U2) are in thesame germ[s]., thenaOU,(ti) = aU2(t2), forby
t1e
the definitionof [s],, thereexists some VC Uln U2such that pv,u1(t) =pv,u2(t2).
Then by the commutativityof the diagram in Fig. 13, we have
aU1(t1) = av O Pv,U1(tl) = O.V 0 PV,U2(t2) = SU2(t2)
F(UM) PPe
F(V)
F(U2) -(
Pu2
FIG. 13
Thus each element [s]r of ir-'(x) is mapped to some point of G by the appropri-
ate au o pul: we call thispoint -(x, [s],) and therebydefinethe required unique
map from7r-'(x) to G.
So we may summarize all three approaches in one sweeping generalization:
foreach xEX, the stalk over x of the sheaf of germsis the direct limit of the
restrictionof the presheafto the neighborhoodsof x.
We close this section with a fourthcharacterizationof sheaves, this one also
based on a universal property,and illustrated to some extent in the previous
examples. Suppose 53is a class of functionsdefinedon open subsets of a topologi-
cal space X. If F(U) is the collection of all f E5 whose domain is U, and if
pu,vis the restrictionmap (i.e., puyv(f)=f Iu whenever UC V and f F(V)), the
collection { F( U), Pu.v} is a presheaf.If this presheafis a sheaf S (as it will be if
f is the set of holomorphicfunctionson open subsets of Cn,or the set of differ-
ential formson open subsets of a paracompact manifoldX) then we can definea
universal continuous function1: S-*R of type ff so that each f EF(U) factors
uniquely throughthe sheaf S: that is, there exists a unique 7: U-S such that
J=(Pof.
In general,any object S and map Ib:S-*R with the propertythat each func-
tion f: U-*R factorsuniquely throughS via 4Dis called universal with respect
to the characterizingpropertiesof the functionsf. The pair (S, b) is uniquely
determined(up to isomorphism)by this property.Thus, forinstance,the sheaf
of holomorphicfunctionsis characterizedby being the unique universal object
forthe familyof holomorphicfunctions.
J
Since I-'(V(f, U)) = U, is continuous. Thus sheaves transforma compli-
cated propertyof functions,such as analyticity,into the simplerone of continu-
ity,forthe topologyon the sheaf is chosen preciselyso that a continuoussection
on U (i.e., an element of F(U)) corresponds to one of the specialized (e.g.,
analytic, differentiable)functionsof F(U).
19701 WHAT IS A SHEAF? 699
5. History and applications. Sheaf theoryis a particularlyeffectivetool in

those areas which ask for global solutions to problems whose hypotheses are
local. Among the early papers which introducedthe ideas, though not the lan-
guage, of sheaf theory,many were concerned with the Cousin problems from
the theoryof functionsof several complex variables; the first(or additive) and
second Cousin problemsask respectivelyabout the existenceof a meromorphic
functionwith specifiedpoles and the existence of a holomorphicfunctionwith
specified zeros. Henri Cartan and Kiyoshi Oka independently solved these
problems,workingin the ringof germs of holomorphicfunctionsintroducedin
our firstexample, where the operations take into account the domains of the
functions.Oka [1950] cites Cartan [1940] as the source of the notion of "id6al
holomorphede domaines indetermines"in this ring, and both Oka [1951] and
Cartan [1944] referto the article of W. Ruckert [1933] which took the concept
of ideal frompolynomial ringsand interpretedit in the ring of functionson a
fixeddomain. Cartan [1944] carried on the investigationof the sheaf of germs,
still in the earlierterminology,clarifyingthe relationsamong the problemswith-
out achieving solutions.
Independently,Oka in 1948 wrote a paper [19501 (seventh in a series pub-
lished from 1936 to 1953 and collected in a single volume [1961]) which de-
veloped the same material in a more complete form,and carriedit throughto a
solution of the firstCousin problem. Building on Oka's paper, Cartan was able
to solve the second problem as well as to simplifyOka's solution to the first,
and his paper [1950] and Oka's were published together.A footnoteacknowl-
edges Oka's solution of the second problem in the meantime [1951].
The 1950 Cartan paper for the firsttime phrases the questions in the sheaf
theoretictermswhich had been developed in the Seminaire Cartan in 1948-49.
An analytic sheaf, that is a sheaf of modules over the sheaf of germs of holo-
morphicfunctions,is called coherentover an open set U if forevery x G U there
is an open set U. such that the sections over U. generate the stalk at y forall y
in a sufficiently small neighborhoodofx. Iffi, * * * ,fkare functionsholomorphic
on a domain D, we may definethe sheaf R of relationsamong the fi by taking
the sections Ru of R over U open in D to be the set of k-tuplesof holomorphic
functions(ga, * * *, gk) forwhichEJ=fJgi=0 on U. In this vocabulary the first
Cousin problem is to show that R is coherent,while the second problem simi-
larly asks whetherthe sheaf over an analytic variety is coherent,where a vari-
ety is the set of common zeros of a set of holomorphicfunctions,and the ideal
of sections over an open set is the ideal of functionson the varietywhichvanish
on that open set.
Cartan borrowed the term "faisceau" (sheaf) from Leray [1945; 1946].
Leray's concept was closer to that of a "presheaf."Cartan [19531 attributesthe
topological definitionto an exposition by Lazard in the Seminaire Cartan
[1950]. In each case, the key concept was that of a system of local coefficients.
Studying sets of invariants for an object (base space) by investigatingwhat
functionscan be definedfromit to some convenient object called a set of co-
as is done in cohomology,leads verynaturallyto a sheaf of coefficients

efficients,
since the presheafstructureallows coefficientsto be assigned locally, that is,
to each open subset of the base space. Formally, the principal constructionof
cohomologywith coefficientsin a (pre)sheaf followsthe Cech constructionof
cohomologywith fixedcoefficients.
Let X be a topological space and S a sheaf over X, say of abelian groups.
For any open cover cl of X, a q-cochain,q being a nonnegative integer,is an
alternatingfunctionwhich assigns to every q+1-tuple of sets in the cover ql
a section over the intersectionof these sets (the zero section if the intersection
is empty). Cq(cU,S) denotes the group of q-cochains. For each q a coboundary
operatoraq, 5': Cq(cIL,S) &Cq+1(cL,S) is definedby
q+1
...** = E (-1)if(Uio, ... * Oij, ... * Uiq+l))
Wf(UiO, U,+)
j=o
wherethe caret over Ui, means that Uij,is to be omittedfromthe argumentsof

f, and each of the sections on the rightis to be interpretedas restrictedto the
intersectionof all the Ui,. By convention we write a for all Aq*Since f is al.
ternating, 5= 0, so the image 5(C01-1(c, S)), whose elements are called co-
boundaries,is contained not merely in C0(cu, S), but in the set of cocycles
Zq(cU,S), the kernelof
B:CQ(Culs) CQJ
0r1 (Cul5).
The qth cohomology groupHQ(%l,S) of the cover cUwith coefficientsin S is the
quotient group ZQ(ul, S)/1(Cq-lQ(U, S)). Although the construction of the
cohomologygroup uses only the presheafof sections of S, the sheaf property
allows us to interpretHo(cL,S). In orderfor0-cochainto be a cocycle,
af(Uo, U1) = PU0U1,U1j(f(U1))
- PuOnu1.uo(f(UO))
must be zero, and in any sheaf, a collection of sections so related defines a

unique global section. Thus HO(cl, S) is Sx-independent of the cover %L.For
all q, if Al is a coveringwhich refinesa coveringVU,the restrictionmaps can be
used to definea canonical map Hq(U, S)--*Hq(9, S). The direct limit,over all
coveringscl of X, of the groupsHQ(cL,S) with these maps, is the qth cohomology
groupofX withcoefficients in the sheafS and is denoted by H2(X, S).
One propertyfrequentlytaken as axiomatic forcohomology theoriesholds
also for this one. If 0--F.4GG4H---0 is an exact sequence of sheaves over X,
thereis a long exact sequenceof cohomology
O ?(X,F) -> HO(X, G) -* HO(X, H) -* H'(X, F) *.*.
-> Hq-l(X, H) -> HI(X, F) -> Hq(X, G) -> Hq(X, H) -> Hq+I(X, F) *
[2, p. 28]. As usual, a pair of homomorphismsAJ-4BC is exact at B if Im f
- Ker g, so the exactness of 0--F-4GAH--*O means that F is a subsheaf of G
(this requiresf to be an open mapping), g is onto, and H is isomorphicto the
quotient sheaf G/F. The existence of the long exact sequence is the major rea-
son forthe usefulnessof cohomology,forthe 0-dimensionalgroups which begin
the sequence are the groups of global sections,while the highergroups by their
constructionreflectthe local propertiesof X.
For instance,if we take F to be the sheaf of germsof holomorphicfunctions
on a complex manifold X, and G to be the sheaf of germs of meromorphic
functions,then F is a subsheaf of G, and the global sections of the quotient
sheaf G/F can be interpretedas the data of the firstCousin problem,since each
section describes the behavior of a functionnear its poles [9, p. 161]. Thus,
since this problemasks whetherthereexists a functionmeromorphicon X with
such poles, the firstCousin problem may be interpretedas asking whetherthe
last map in the sequence 0-?Fx--->Gx-->(G/F)xis onto. This sequence is the
beginningof the long exact cohomologysequence, and the next group in that
sequence is H' (X, F). The Cartan-Oka result is that Hq(X, F) =0 for all q ?1
if X is a Stein manifold,a class of manifoldswith "sufficientlymany" holo-
morphic functions,which includes all Riemann surfaces which are connected
and non-compact. In addition to proving this result, Cartan [1953] and Serre
[1953] give otherapplications of the fundamentaltheoremsfora Stein manifold
X:
THEOREM A. For everycoherentanalytic sheaf S overX, HO(X, S), whichis
themoduleofglobalcrosssectionsSx, generatesthestalkS=foreveryx GX.
analyticsheafS overX and q ? 1,
THEOREM B. For everycoherent
Hq(X, S) = 0.
PropertiesA and B characterizeStein manifolds.

The proceedingsof the 1954 AMS summer institute [1956] illustrate that
by then the basic concepts of sheaf theory had been clarifiedapart fromthe
original example, and the bibliographiesindicate that applications had begun
to diversify,particularlyinto algebraic geometry.For example, Kodaira and
Spencer showed the equivalence of several differentdefinitionsof the arithmetic
genus of an algebraic variety and provided a classificationof complex line
bundles [1953]. Hirzebruch gave a sheaf-theoreticstatement and proof of the
Riemann-Roch theorem [1953; 1956], and Weil of the deRham theorem [1952].
One seminar at the 1954 institutewas based on an early version of Serre's
major article FAC [1955], the firstentirely algebraic development of sheaf
theory. The applications to complex variables had frequentlymade use of
complex integration,and this tool was not available in abstract algebraic
geometry. "Faisceaux algebriques coherents" are coherent sheaves which are
sheaves of modules over the sheaf of local ringson an algebraic variety. Serre
showed that if the base field is the field of complex numbers, the theory of
algebraic coherent sheaves is isomorphic to the theory of analytic coherent
sheaves. Going furtherin the directionof an algebraic treatment,Grothendieck
dealt with sheaves in the context of cohomologyin an abelian category [1957].
The publication of Godement's book [1958] signals the appearance of sheaf

theoryas an independentdiscipline. The bibliographywhich followslists both
recent treatmentsof sheaf theoryand books on other subjects which make use
of sheaves.
HistoricalBibliography
1933. W. Riuckert, Zum Eliminationsproblem derPotenzreihenideale, Mathematische Annalen,
107 (1933) 259-281.
1940. HenriCartan,Sur les matricesholomorphes de n variablescomplexes, Journalde Math-
ematiques PuresetAppliquees,19 (1940) 1-26.
1944. , Ideaux de fonctionsanalytiquesde n variablescomplexes,Annalesde l'Ecole
Normale,3e sIrie, 61 (1944) 149-197.
1945. Jean Leray,Sur la formedes espaces topologiques... (3 articles),Journalde Math-
ematiques PuresetAppliquees, 24 (1945) 95-248.
1946. , L'anneau d'homologie d'une representation, C. R. Acad. Sci.,222 (1946) 1366-
1368.
1950. KiyoshiOka, Sur quelquesnotionsarithmetiques, Bulletinde la SocietdMathematique de
France,78 (1950) 1-27.
1950. Henri Cartan, Id6aux et modulesde fonctionsanalytiquesde variablescomplexes,
Bulletinde la SocieteMathematique de France,78 (1950) 29-64.
1950. SeminaireCartan.1950-51,vol. 1, Benjamin,New York,1967.
1951. KiyoshiOka, Lemme fondamental, Journalof theMathematical Societyof Japan, 3
(1951) 204-214,259-278.
1952. AndreWeil,Sur les theor6mes de deRham,Commentarii Mathematici Helvetici,(1952)
119-146.
1953. Henri Cartan,Variet6sanalytiquescomplexeset cohomologie,Colloquede Bruxelles,
(1953) 41-55.
1953. K. Kodairaand D. C. Spencer,On Arithmetic GeneraofAlgebraicVarieties,Proceedings
oftheNationalAcademy ofSciencesUSA, 39 (1953)641-649.
1953. and , Groups of complexline bundles over compact Kahler varieties,
Proceedings oftheNationalAcademyofSciencesUSA, 39, (1953) 868-877.
1953. Jean-Pierre Serre,Quelquesproblemes globauxrelatifs aux varietesde Stein,Colloquede
Bruxelles,(1953) 57-68.
1954. F. Hirzebruch,Arithmeticgeneraand the theoremof Riemnann-Roch for algebraic
varieties,Proceedings oftheNationalAcademyofSciencesUSA, 40 (1954) 110-114.
1955. Jean-Pierre Serre,Faisceaux alg6briquescoherents, Annals of Mathematics, 61 (1955)
197-278.
1956. F. Hirzebruch, Methods
Topological in Algebraic Geometry,3rd ed.,Springer, Berlin,1966.
1956. Shiing-ShenChern,ScientificReportof the Second SummerInstitute;II, Complex
Manifolds, Bulletin,AmericanMathematical Society, 62 (1956) 102-117.
1956. Oscar Zariski,Scientific Reportofthe SecondSummerInstitute;III, AlgebraicSheaf
Theory,Bulletin,AmericanMathematical Society,62 (1956) 117-141.
1957. AlexandreGrothendieck, Sur quelques pointsd'algebrehomologique,T6hokuMathe-
maticalJournalSer 2, 9 (1957) 119-221.
1958. RogerGodement,TopologieAlgebrique et Theoriedes Faisceaux,Hermann,Paris, 1958.
1961. KiyoshiOka, Sur les FonctionsAnalytiquesde plusieursvariables,Iwanami Shoten,
Tokyo,1961.
GeneralBibliography
1. Lipman Bers,Introduction
to SeveralComplexVariables,CourantInstitute,New York,
1964.
2. Glen Bredon,SheafTheory,McGraw-Hill,New York, 1967.
1970] ON THE FOUNDATIONS OF SET THEORY 703
3. C. H. Dowker,Lectureson SheafTheory,Tata Instituteof FundamentalResearch,Bom-

bay, 1956.
4. J. W. Gray,Sheaveswithvaluesin a category, Topology,3 (1965) 1-18.
5. AlexandreGrothendieck and JeanDieudonnen, Dl6mentsde GeometrieAlgebrique,Publica-
tionsMath5matiques de l'Institutdes Hautes etudes Scientifiques,
Nos.4, 7, 11,17,20,24,28,32.
6. Robert Gunningand Hugo Rossi, AnalyticFunctionsof Several Complex Variables,
PrenticeHall, EnglewoodCliffs, N. J., 1965.
7. M. Herve,SeveralComplexVariables,OxfordUniv. Press,1963.
8. Noel J. Hicks,Noteson Differential Geometry, Van Nostrand,Princeton, N. J., 1965.
9. Lars Hormander, ComplexAnalysisin SeveralVariables,Van Nostrand,Princeton,N. J.,
1966.
10. Ian Macdonald,AlgebraicGeometry,Introduction to Schemes,Benjamin,New York,
1968.
11. RichardSwan,The TheoryofSheaves,University ofChicagoPress,Chicago,1964.
12. 0. Zariskiand P. Samuel,Commutative AlgebraI, Van Nostrand,1958.
13. and , Commutative AlgebraII, Van Nostrand,1960.
ON THE FOUNDATIONS OF SET THEORY

ofColorado
J. D. MONK, University
I want to discuss here the relevance to mathematicians,as teachers and
researchers,of some of the recentdiscoveriesabout axiomatic set theory.Most
readers have heard of these advances, which began just a few years ago with
Cohen's work. The results are certainlyintellectuallyamazing to all of us. I
think they may even give rise to certain changes in our teaching and research,
and the purpose of this paper is to describe some possibilitiesalong these lines.
To set the stage and fixthe ideas I shall firstdescribe a fewof these discoveries
in a fairlyprecise way. Then, in the nonexact portion of the paper, I shall dis-
cuss some possible changes in teaching and research,and also some philosophi-
cal views which are affectedby these discoveries.
1. A survey of results. A much more comprehensive(and more technical)
survey can be found in Mathias [7]. Here I state just a very few results,but I
wish to emphasize that the nonmathematical arguments of the next section
apply in some formto virtuallyall of the resultsdescribedin [7]. I assume that
the reader has a modest acquaintance with the idea of a language and a meta-
language, and with the precise notions of a (first-order)sentence, a (formal)
proof,and a theorem.In this section I work in a metalanguage and talk about
the language of mathematics. I leave the metalanguage unspecifiedin detail;
to begin with I assume that it is rather weak, with just enough machineryto
Prof.MonkreceivedhisBerkeleyPh.D. in 1961underAlfredTarski.Aftera post-doctoral year
at Berkeley,he cameto his presentpostat Colorado.His mainresearchis in algebraiclogic,and
he has publishedthebooks,IntrodtctiontoSet Theory(McGraw-Hill1969) and (withL. Henkin
and A. Tarski) Cylindric
Algebras,Part I (NorthHolland,forthcoming).Editor.

What is a Sheaf? Exploring Germs of Holomorphic Functions

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

What is a Sheaf? Exploring Germs of Holomorphic Functions

Uploaded by

Copyright:

Available Formats

What is a Sheaf?

Author(s): J. Arthur Seebach, Jr., Linda A. Seebach and Lynn A. Steen

product involve intersectionof domain: if f: U->C and g: V-*C, then f+g:

To make this intuitivepicture more precise,we liftthe topologyof C- back

We can show that the stalk space S is Hausdorffas follows:points of S may

There is still anotherconsequence of the uniqueness propertyof holomorphic

A manifoldis, essentially,a topological space whichis locally homeomorphic

between two covers which provide essentiallythe same coordinate structureon

F is universalin the sense that the behavior of all holomorphicfunctionson

uniquely extended. It is the connected componentE of S which containsJ(D).

whichare compatiblewith the relation.We denote by As the ringof equivalence

can define maps pP,V:Av--*Apwhen PG V. If V is open and PG V, we define

In this topology the projection p is a local homeomorphism,for if xEAp

to visualize it on Hausdorffpaper) that two distinctsections over U, u and u',

wfo^?- (V) U'(V)

A special case. In order to provide a more geometricinterpretationof the

Since C is an integral domain, the kernel of the homomorphismdefined by

Vt= { P CSpec A I EqPP } (9, p(x)) I p CA is irreducibleor zero,

(x, Pk(y)), (Pk+i(x), y), . , (pn(x), y) the set of denominators for Au is

Thus y*(f) is usually denoted by (a/ox&)f;the tangents (o/Oxi), * * , (O/1x.)

If q X-* Y is a C??mapping of manifoldswe definethe differential of( to be

the set of k-tuplesby

4. Sheaves: General definition.Each of the three previous examples re-

Sheafof Germs Sheafof

Germn U]t a/sEAp

varietyand to providea coherentframework forthesubsequentgeneraldefini-

ifS is a sheaf,thenS' is naturallyisomorphicto S, so in thissense the two defini-

P V1.WJ ~ F(Vi) PU,V

5. History and applications. Sheaf theoryis a particularlyeffectivetool in

as is done in cohomology,leads verynaturallyto a sheaf of coefficients

wherethe caret over Ui, means that Uij,is to be omittedfromthe argumentsof

must be zero, and in any sheaf, a collection of sections so related defines a

PropertiesA and B characterizeStein manifolds.

The publication of Godement's book [1958] signals the appearance of sheaf

3. C. H. Dowker,Lectureson SheafTheory,Tata Instituteof FundamentalResearch,Bom-

ON THE FOUNDATIONS OF SET THEORY

You might also like

Thus y(f) is usually denoted by (a/ox&)f;the tangents (o/Oxi), * , (O/1x.)