NOTES ON SEVERAL COMPLEX VARIABLES

J. L. Taylor
Department of Mathematics
University of Utah
July 27, 1994
Revised June 9, 1997

Notes from a 199394 graduate course

Revised for a 1996-97 graduate course

J. L. TAYLOR

TABLE OF CONTENTS
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.

Holomorphic Functions
Local Theory
A Little Homological Algebra
Local Theory of Varieties
The Nullstellensatz
Dimension
Completion of Local Rings
Sheaves
Sheaf Cohomology
Coherent Algebraic Sheaves
Dolbeault Cohomology
Coherent Analytic Sheaves
Projective Varieties
Algebraic vs. Analytic Sheaves Serres Theorems
Stein Spaces
Frechet Sheaves Cartans Theorems
The Borel-Weil-Bott Theorem

p.
p.
p.
p.
p.
p.
p.
p.
p.
p.
p.
p.
p.
p.
p.
p.
p.

3
10
15
24
30
46
54
63
71
87
99
106
116
124
136
149
166

NOTES ON SEVERAL COMPLEX VARIABLES

1. Holomorphic Functions
There are a number of possible ways to define what it means for a function defined on a
domain in Cn to be holomorphic. One could simply insist that the function be holomorphic
in each variable separately. Or one could insist the function be continuous (as a function
of several variables) in addition to being holomorphic in each variable separately. The a
priori strongest condition would be to insist that a holomorphic function have a convergent
expansion as a multi-variable power series in a neighborhood of each point of the domain.
The main object of this chapter is to show that these possible definitions are all equivalent.
In what follows, (a, r) will denote the polydisc of radius r = (r1 , r2 , ..., rn ) about
a = (a1 , a2 , ..., an ):
(a, r) = {z = (z1 , z2 , ..., zn ) : |zj aj | < rj

j = 1, 2, ..., n}

and (a,
r) will denote the corresponding closed polydisc. Note that (a, r) is just the

Cartesian product of the open discs (ai , ri ) C and (a,

r) is the Cartesian product of

the closed discs (ai , ri ).

1.1 Proposition (Cauchy Integral Formula). If f is a function which is continuous

in a neighborhood U of the closed polydisc (a,

r) and holomorphic in each variable zi at
each point of U , then

f (z) =

1
2i

n Z

|n an |=rn

|1 a1 |=r1

f (1 , ..., n )d1 ...dn

(1 z1 )...(n zn )

for any z (a, r).

This follows immediately from repeated application of the one variable Cauchy Theorem.
1.2 Proposition (Osgoods Lemma). If a function is continuous in an open set U Cn
and is holomorphic in each variable at each point of U then at each point a U there is
a power series of the form

ci1 ...in (z1 a1 )i1 ...(zn an )in

i1 ,...,in =0

which converges uniformly to f on every compact polydisc centered at a and contained in

U.

Proof. Let (a,

s) be a compact polydisc centered at a and contained in U . Since (a,
s)

is compact in U we may choose numbers ri > si such that the compact polydisc (a, r)
with polyradius r = (r1 , . . . , rn ) is also contained in U . We substitute into the integrand
of 1.1 the series expansion:

X
1
(z1 a1 )i1 ...(zn an )in
=
(1 z1 )...(n zn ) i ,...,i =0 (1 a1 )i1 +1 ...(n an )in +1
1

J. L. TAYLOR

This series converges uniformly in and z for |zi ai | < si < ri = |i ai |. Thus, the
series in the integrand of 1.1 can be integrated termwise and the resulting series in z is

uniformly convergent for z (a,

s). The result is a series expansion of the form
f (z) =

ci1 ...in (z1 a1 )i1 ...(zn an )in

i1 ,...,in =0

uniformly convergent on (a,

s), with

ci1 ...in =

1
2i

n Z

|n zn |=rn

|1 z1 |=r1

f (1 , ..., n )d1 ...dn

(1 a1 )i1 +1 ...(n an )in +1

This completes the proof.

A function f defined on an open set U Cn is called holomorphic at a if it has a power
series expansion as in 1.2, convergent to f in some open polydisc centered at a. If f is
holomorphic at each point of U then it is called holomorophic on U . Thus, Osgoods Lemma
says that a continuous function on U which is holomorphic in each variable separately at
each point of U is holomorphic on U . We will later prove that the continuity hypothesis
is redundant.
1.3 Proposition (Cauchys inequalities). If f is a holomorphic function in a neigh
borhood of the closed polydisc (a,
r) and |f | is bounded by M in this polydisc, then for
each multi-index (i1 , ..., in )
i +...+i

n
1

M (i1 !)...(in !)r1 i1 ...rn in

f
(a)
z1 i1 ...zn in

Proof. If f is expressed as a power series convergent on (a,

r) as in 1.2, then repeated
differentiation yields
i1 +...+in
f (a) = (i1 !)...(in !)ci1 ...in
z1 i1 ...zn in
where, according to the proof of 1.2,

ci1 ...in =

1
2i

n Z

|n zn |=rn

|1 z1 |=r1

f (1 , ..., n )d1 ...dn

(1 a1 )i1 +1 ...(n an )in +1

From the obvious bounds on the integrand of this integral it follows that
|ci1 ...in | M r1 i1 ...rn in
and the Cauchy inequalities follow from this.
In what follows, the notation |U | will be used to denote the volume of a set U Cn .

NOTES ON SEVERAL COMPLEX VARIABLES

=
1.4 Theorem (Jensens inequality). If f is holomorphic in a neighborhood of

(a, r) then
Z
1
log |f (a)|
log |f (z)|dV (z)
||
Proof. We recall the single variable version of Jensens inequality:
1
log |g(a)|
2

log |g(a + ei )|d

and apply it to f considered as a function of only z1 with the other variables fixed at
a2 , ..., an . This yields:
1
log |f (a)|
2

log |f (a1 + 1 ei1 , a2 , ..., an )|d1

We now apply Jensens inequality to the integrand of this expression, where f is considered
as a function of z2 with z1 fixed at aa +1 ei1 and the remaining variables fixed at a3 , ..., an
to obtain:

log |f (a)|

1
2

2 Z

2
0

log |f (a1 + 1 ei1 , a2 + 2 ei2 , a3 , ..., an )|d1 d2

Continuing in this way, we obtain:

log |f (a)|

1
2

n Z

...
0

log |f (a1 + 1 ei1 , a2 + 2 ei2 , ..., an + n ein )|d1 ...dn

Finally, we multiply both sides of this by 1 ...n and integrate with respect to d1 ...dn
over the set {i ri ; i = 1, ..., n} to obtain the inequality of the theorem.
In the next lemma we will write a point z Cn as z = (z 0 , zn ) with z 0 Cn1 . Similarly
we will write a polyradius r as r = (r0 , rn ) so that a polydisc (a, r) can be written as
(a0 , r0 ) (an , rn ).
1.5 Lemma (Hartogs lemma). Let f be holomorphic in (0, r) and let
f (z) =

fk (z 0 )znk

be the power series expansion of f in the variable zn , where the fk are holomorphic in

(0, r0 ). If there is a number c > rn such that this series converges in (0,
c) for each
0
0
0
z (0, r ) then it converges uniformly on any compact subset of (0, r )(0, c). Thus,
f extends to be holomorphic in this larger polydisc.
0 , s0 ) (0, r0 ) and
Proof.. Choose any point a0 (0, r0 ), choose a closed polydisc (a
choose some positive b < rn . Then we may choose an upper bound M > 1 for f on the

J. L. TAYLOR

0 , s0 ) (0,

polydisc (a
b). It follows from Cauchys inequalities that |fk (z 0 )| M bk for
0 , s0 ). Hence,
all z 0 (a
k 1 log |fk (z 0 )| k 1 log M log b log M log b = M0
0 , s0 ) and all k. By the convergence of the above series at (z 0 , c) we also
for all z 0 (a
0 , s0 ) and hence:
have that |fk (z 0 )|ck converges to zero for each z 0 (a
lim sup k 1 log |fk (z 0 )| log c
k

0 , s0 ).
for all z 0 (a
We have that the functions k 1 log |fk (z 0 )| are uniformly bounded and measurable in

(a0 , s0 ) and by Fatous lemma

Z
Z
1
0
0
lim sup
k log |fk (z )|dV (z )
lim sup k 1 log |fk (z 0 )|dV (z 0 )
k

(a0 ,s0 )

(a0 ,s0 )
0 0

|(a , s )| log c
By replacing c by a slightly smaller number if neccessary we may conclude that there is a
k0 such that
Z
k 1 log |fk (z 0 )|dV (z 0 ) |(a0 , s0 )| log c
(a0 ,s0 )

for all k k0 . Now by shrinking s0 to a slightly smaller multiradius t0 and choosing small
enough we can arrange that
(a0 , t0 ) (w0 , t0 + ) (a0 , s0 )
for all w0 (a0 , ), from which we conclude that
Z
k 1 log |fk (z 0 )|dV (z 0 ) |(a0 , t0 )| log c + M0 |(w0 , t0 + ) (w0 , t0 )|
(w0 ,t0 +)

Now if we choose a c0 slightly smaller than c we may choose small enough that the right
hand side of this inequality is less than |(w0 , t0 + )| log c0 . This yields
Z
k 1 log |fk (z 0 )|dV (z 0 ) |(w0 , t0 + )| log c0
(w0 ,t0 +)

and this, in turn, through Jensens inequality, yields

k 1 log |fk (w0 )| log c0
or

|fk (w0 )|ck0 1

for all k k0 and w0 (a0 , ). This implies that our series is uniformly convergent
in (a0 , ) (0, c0 ) and since a0 is arbitrary in (0, r0 ) and c0 is an arbitrary positive
number less than c, we have that our power series serves to extend f to a function which
is continuous on (0, r0 ) (0, c) and certainly holomorphic in each variable. It follows
from Osgoods lemma that this extension of f is, in fact, holomorphic.

NOTES ON SEVERAL COMPLEX VARIABLES

1.6 Theorem (Hartogs theorem). If a complex-valued function is holomorphic in each

variable separately in a domain U Cn then it is holomorphic in U .
Proof. We prove this by induction on the dimension n. The theorem is trivial in one
dimension. Thus, we suppose that n > 1 and that the theorem is true for dimension n 1.

Let a be any point in U and let (a,

r) be a closed polydisc contained in U . As in the
0

0 , r0 )
previous lemma, we write z = (z , zn ) for points of Cn and write (a,
r) = (a
n , rn ).
(a
Consider the sets
n , rn /2) :
Xk = {zn (a

|f (z 0 , zn )| k

0 , r0 )}
z 0 (a

For each k this set is closed since f (z 0 , zn ) is continuous in zn for each fixed z 0 . On the
0 , r0 ) for each
other hand, since f (z 0 , zn ) is also S
continuous in z 0 and, hence bounded on (a
n , rn /2)
zn , we have that (a
k Xk . It follows from the Baire category theorem that
for some k the set Xk contains a neighborhood (bn , ) of some point bn (an , rn /2).
We now know that f is separately holomorphic and uniformly bounded in the polydisc
n , ). It follows from Cauchys inequalities in each variable separately that
(a0 , r0 ) (b
the first order complex partial derivatives are also uniformly bounded on compact subsets
of this polydisc. This, together with the Cauchy-Riemann equations implies that all first
order partial derivatives are bounded on compacta in this polydisc and this implies uniform
continuity on compacta by the mean value theorem. We conclude from Osgoods theorem
that f is holomorphic on (a0 , r0 ) (bn , ) and, in fact, has a power series expansion
about (a0 , bn ) which converges uniformly on compact subsets of this polydisc.
Now choose sn > rn /2 so that (bn , sn ) (an , rn ). Then f (z 0 , zn ) is holomorphic in
zn on (bn , sn ) for each z 0 (a0 , r0 ) and, hence, its power series expansion about (a0 , bn )
converges as a power series in zn on (bn , sn ) for each fixed point z 0 (a0 , r0 ). It follows
from Hartogs lemma that f is actually holomorphic on all of (a0 , r0 ) (bn , sn ). Since
a is in this set and a was an arbitrary element of U , the proof is complete.
We introduce the first order partial differential operators

=
i
=
+i
zj
2 xj
yj
zj
2 xj
yj
Clearly, the Cauchy-Riemann equations mean that a function is holomorphic in each
variable separately in a domain U if and only if its first order partial derivatives exist in U
and zj f = 0 in U for each j. We may write these equations in a more succinct form by
using differential forms. We choose as a basis for the complex one forms in Cn the forms
dzj = dxj + idyj ,

d
zj = dxj idyj ,

j = 1, ..., n

, where
then the differential df of a function f decomposes as f + f
f =

X f
dzj ,
z
j
j

=
f

X f
d
zj

j
j

J. L. TAYLOR

= 0. Thus, a function defined

and the Cauchy-Riemann equations become simply f
in a domain U is holomorphic in U provided its first order partial derivatives exist and
= 0 in U . A much stronger resultis true: A distribution defined in a domain in Cn
f
= 0 in the distribution sense is actually
which satisfies the Cauchy-Riemann equations
a holomorphic function. We will not prove this here, but it follows from appropriate
regularity theorems for elliptic PDEs.
If U is an open subset of Cn then the space of all holomorphic functions on U will be
denoted by H(U ). This is obviously a complex algebra under the operations of pointwise
addition and multiplication of functions. It also has a natural topology - the topology of
uniform convergence on compact subsets of U . We end this section by proving a couple of
important theorems about this topology.
The topology of uniform convergence on compacta is defined as follows. For each compact set K U we define a seminorm || ||K on H(U ) by
||f ||K = sup |f (z)|
zK

Then a neighborhood basis at g H(U ) consists of all sets of the form

{f H(U ) : ||f g||K < }
for K a compact subset of U and > 0. If {Kn } is an increasing sequence of compact
subsets of U with the property that each compact subset of U is contained in some Kn ,
then a basis for this topology is also obtained using just the sets of the above form with
K one of the Kn0 s and one of the numbers m1 for m a positive interger. Thus, there
is a countable basis for the topology at each point and, in fact, H(U ) is a metric space
in this topology. Clearly, a sequence converges in this topology if and only if it converges
uniformly on each compact subset of U . A nice application of Osgoodss Lemma is the
proof that H(U ) is complete in this topology:
1.7 Theorem. If {fn } is a sequence of holomorphic functions on U which is uniformly
convergent on each compact subset of U , then the limit is also holomorphic on U .
Proof. For holomorphic functions of one variable this is a standard result. Its proof is a
simple application of Moreras Theorem. In the several variable case we simply apply this
result in each variable separately (with the other variables fixed) to conclude that the limit
of a sequence of holomorphic functions, converging uniformly on compacta, is holomorphic
in each variable separately. Such a limit is also obviously continuous and so is holomorphic
by Osgoods Lemma (we could appeal to Hartogs Theorem but we really only need the
much weaker Osgoods Lemma).
A topological vector space with a topology defined by a sequence of seminorms, as
above, and which is complete in this topology is called a Frechet space. Thus, H(U ) is a
Frechet space. It is, in fact, a Montel space. The content of this statement is that it is a
Frechet space with the following additional property:

NOTES ON SEVERAL COMPLEX VARIABLES

1.8 Theorem. Every closed bounded subset of H(U ) is compact, where A H(U ) is
bounded if {||f ||K ; f A} is bounded for each compact subset K of U .
Proof. Since H(U ) is a metric space we need only prove that every bounded sequence in
H(U ) has a convergent subsequence. Thus, let {fn } be a bounded sequence in H(U ). By
the previous theorem, we need only show that it has a subsequence that converges uniformly
on compacta to a continuous function - the limit will then automatically be holomorphic
as well. By the Ascoli-Arzela Theorem, a bounded sequence of continuous functions on a
compact set has a uniformly convergent subsequence if it is equicontinuous. It follows from
the Cauchy estimates that {fn } has uniformly bounded partial derivatives on each compact
set. This, together with the mean value theorem implies that {fn } is equicontinuous on
each compact set and, hence, has a uniformly convergent subsequence on each compact
set. We choose a sequence {Km } of compact subsets of U with the property that each
compact subset of U is contained in some Km . We then choose inductively a sequence
of subsequences {fnm,i }i of {fn } with the property that {fnm+1,i }i is a subsequence of
{fnm,i }i and {fnm,i }i is uniformly convergent on Km for each m. The diagonal of the
resulting array of functions converges uniformly on each Km and, hence, on each compact
subset of U . This completes the proof.
of
1.
2.
3.
4.
5.

6.

If U is a domain in C n and F : U C m is a map, then F is called holomorphic if each

its coordinate functions is holomorphic.
1. Problems
Prove that the composition of two holomorphic mappings is holomorphic.
Formulate and prove the identity theorem for holomorphic functions of several variables.
Formulate and prove the maximum modulus theorem for holomorphic functions of several variables.
Formulate and prove Schwarzs lemma for holomorphic functions of several variables.
Prove that a function which is holomorphic in a connected neighborhood of the boundary
of a polydisc in Cn , n > 1, has a unique holomorphic extension through the interior
of the polydisc.
Prove that if a holomorphic function on a domain (a connected open set) is not identically zero, then the set where it vanishes has 2n-dimensional Lebesgue measure zero
(hint: use Jensens inequality).

10

J. L. TAYLOR

2. Local theory
Let X be a topological space and x a point of X. If f and g are functions defined
in neighborgoods U and V of x and if f (y) = g(y) for all y in some third neighborhood
W U V of x, then we say that f and g are equivalent at x. The equivalence class
consisting of all functions equivalent to f at x is called the germ of f at x.
The set of germs of complex valued functions at x is clearly an algebra over the complex
field with the algebra operations defined in the obvious way. In fact, this algebra can be
described as the inductive limit lim F (U ) where F (U ) is the algebra of complex valued

functions on U and the limit is taken over the directed set consisting of neighborhoods of
x. The germs of continuous functions at x obviously form a subalgebra of the germs of all
complex valued functions at x and, in the case where X = Cn , the germs of holomorphic
functions at x form a subalgebra of the germs of C functions at x which, in turn, form a
subalgebra of the germs of continuous functions at X.
We shall denote the algebra of holomorphic functions in a neighborhood U Cn by
H(U ) and the algebra of germs of holomorphic functions at z Cn by Hz or by n Hz in case
it is important to stress the dimension n. We have that Hz = lim H(U ) where the limit is

taken over a system of neighborhoods of z. We also have as an immediate consequence of

the definition of holomorphic function that:
2.1 Proposition. The algebra n H0 may be described as the algebra C{z1 , ..., zn } of convergent power series in n variables.
Another important algebra of germs of functions is the algebra n Oz of germs of regular
functions at z Cn . Here we give Cn the Zariski topology. This is the topology in which
the closed sets are exactly the algebraic subvarieties of Cn . By an algebraic subvariety of
Cn we mean a subset which is the set of common zeroes of some set of complex polynomials
in C[z1 , ..., zn ]. A regular function on a Zariski open set U is a rational function with a
denominator which does not vanish on U . The algebra of regular functions on a Zariski
open set U will be denoted O(U ). The algebra of germs of regular functions at z Cn is
then n Oz = lim O(U ). One easily sees that:

2.2 Proposition. The algebra n Oz is just the ring of fractions of the algebra C[z1 , ..., zn ]
with respect to the multiplicative set consisting of polynomials which do not vanish at z.
By a local ring we will mean a ring with a unique maximal ideal. It is a trivial observation
that:
2.3 Proposition. The algebras n Oz and n Hz are local rings and in each case the maximal
ideal consists of the elements which vanish at z.
The algebras n Oz and n Hz are, in fact, Noetherian rings (every ideal is finitely generated). For n Oz this is a well known and elementary fact from commutative algebra. We
will give the proof here because the main ingredient (Hilberts basis theorem) will also be
needed in the proof that n Hz is Noetherian.
We will use the elementary fact that if M is a finitely generated module over a Noetherian ring A then every submodule and every quotient module of M is also finitely
generated.

NOTES ON SEVERAL COMPLEX VARIABLES

11

2.4 Theorem (Hilbert basis theorem). If A is a Noetherian ring then the polynomial
ring A[x] is also Noetherian.
Proof. Let I be an ideal in A[x] and let J be the ideal of A consisting of all leading coeficients of elements of I. Since A is Noetherian, J has a finite set of generators {a1 , ..., an }.
For each i there is an fi I such that fi = ai xri + gi where gi has degree less than ri .
Let r = maxi ri and let f = axm + g (deg P
g < m) be an element
degree m r.
P of I of
mri
We may choose bi , ..., bn A such that a = i bi Ai . then f i bi fi x
belongs to
I and has degree less than m. By iterating this process we conclude that every element
of I may be written as the sum of an element of the ideal generated by {f1 , ..., fn } and
a polynomial of degree less than r belonging to I. However, the polynomials of degree
less than r form a finitely generated A module and, hence, the submodule consisting of its
intersection with I is also finitely generated. A generating set for this A module, together
with {f1 , ..., fn } provides a set of generators for I as an A[x]-module. This completes the
proof.
The above result and induction show that C[z1 , ..., zn ] is Noetherian. This implies that
Oz is Noetherian as follows: If I is an ideal of Oz and J = I C[z1 , ..., zn ] then J generates
I as an Oz -module but J is finitely generated as a C[z1 , ..., zn ]-module. It follows that I
is finitely generated as an Oz -module. In summary:
2.5 Theorem. The polynomial algebra C[z1 , ..., zn ] and the local algebra n Oz are Noetherian rings.
We now proceed to develop the tools needed to prove that n Hz is Noetherian.
A holomorhic function defined in a neighborhood of 0 is said to be regular of order k in
zn at 0 provided f (0, ..., 0, zn ) has a zero of order k at 0.
2.6 Theorem. If f is holomorphic in a neighborhood U of 0 in Cn and regular of order
k in zn at 0, then there is a polydisc (0, r0 ) (0, rn ) such that for each z 0 (0, r0 ),
as a function of zn , f (z 0 , zn ) has exactly k zeroes in (0, rn ), counting multipicity, and no
zeroes on the boundary of (0, rn ).
rn ) occur at zn = 0.
Proof. Choose rn small enough that the only zeroes of f (0, zn ) on (0,
Set
= inf{|f (0, zn )| : |zn | = rn }
and choose r0 small enough that
|f (z 0 , zn ) f (0, zn )| < whenever z 0 (0, r0 ),

|zn | = rn

Then Rouches theorem implies that for each z 0 (0, r0 ) the functions f (0, zn ) and
f (z 0 , zn ) of zn have the same number of zeroes in the disc (0, rn ). This completes the
proof.
A thin subset of Cn is a set which locally at each point of Cn is contained in the zero
set of a holomorphic function.

12

J. L. TAYLOR

2.7 Theorem (Removable singularity theorem). If f is bounded and holomorphic

in an open set of the form U T where U is open and T is thin, then f has a unique
holomorphic extension to all of U .
Proof. This is a local result and needs only to be proved for a neighborhood of each point
a U . We may assume that T intersected with some neighborhood of a is contained in
the zero set of a holomorphic function g which we may assume is regular in zn of some
order k (otherwise we simply perform a coordinate change). By the proceeding result we
may choose a neighborhood of the form (a, r) = (a0 , r0 ) (an , rn ) and assume that
n , rn ) in at most k points all of which lie
for each z 0 (a0 , r0 ) the set T meets {z 0 } (a
0
in {z } (an , rn ). Then the function
1
h(z , zn ) =
2i

|n an |=rn

f (z 0 , n )
dn
n zn

is holomorphic in all of (a, r) and agrees with f off T by the one variable removable
singularity theorem and the Cauchy integral theorem.
A Weierstrass polynomial of degree k in zn is a polynomial h

n1 H0 [zn ]

of the form

h(z) = znk + a1 (z 0 )znk1 + ... + ak1 (z 0 )zn + ak (z 0 )

where z = (z 0 , zn ) and each ai is a non-unit in

n1 H0 .

2.8 Theorem (Weierstrass preparation theorem). If f n H0 is regular of order k

in zn , then f has a unique factorization as f = uh where h is a Weierstrass polynomial of
degree k in zn and u is a unit in n H0 .
Proof. Choose some representative of f and a polydisc (0, r) in which f (z 0 , zn ) has exactly
k zeroes (none on the boundary) as a function of zn for each z 0 (0, r0 ) as in 2.6. Label
the zeroes b1 (z 0 ), ..., bk (z 0 ). The polynomial we are seeking is
h(z) =

k
Y

(zn bj (z 0 )) = znk + a1 (z 0 )znk1 + ... + ak1 (z 0 )zn + ak (z 0 )

j=1

Now the functions bj (z 0 ) need not even be continuous because of the arbitrary choices
made in labeling the zeroes of f . However, the functions aj (z 0 ) are, in fact, holomorphic.
To see this, note that these functions are the elementary symmetric functions of the bj s
and these, in turn, may be written as polynomials in the power sums sm where
0

sm (z ) =

k
X
j=1

0 m

bj (z )

1
=
2i

Z
m
||=rn

f (z 0 , ) d

f (z 0 , )

These functions and, consequently, the aj (z 0 ) are holomorphic in (0, r0 ). Note that the
bj s all vanish at z 0 = 0 and, thus, so do the aj s. We now have that h is a Weierstrass
polynomial.

NOTES ON SEVERAL COMPLEX VARIABLES

13

The proof will be complete if we can show that u = f /h is holomorphic and nonvanishing in (0, r). For each fixed z 0 (0, r0 ) the function f (z 0 , zn )/h(z 0 , zn ) is holomorphic in zn in (0, rn ) since numerator and denominator have exactly the same zeroes
in this polydisc. Furthermore, h is bounded away from zero on (0, r0 ) (0, rn ). This
and the maximum modulus principal imply that f /h is bounded on (0, r). Since it is
holomorphic in this set except where h vanishs, the removable singularities theorem (2.7)
implies that it extends to be holomorphic and non-vanishing in the entire polydisc. The
uniqueness is clear from the construction. This completes the proof.
2.9 Theorem (Weierstrass division theorem). If h n1 H[zn ] is a Weierstrass polynomial of degree k and f n H, then f can be written uniquely in the form f = gh + q
where g n H and q n1 H[zn ] is a polynomial in zn of degree less than k. Furthermore,
if f is a polynomial in zn then so is g.
Proof. Choose representatives of f and h (still call them f and h) which are defined in a
r) which is chosen small enough that h(z 0 , zn ) has exactly
neighborhood of a polydisc (0,

k zeroes in (0, rn ) as a function of zn for each z 0 (0,

r0 ) with none occuring on
|zn | = rn . Then the function
1
g(z) =
2i

Z
||=rn

f (z 0 , ) d
h(z 0 , ) zn

is holomorphic in (0, r) as is the function q = f gh. The function q may be written as

1
q(z) =
2i
But the function

Z
||=rn

f (z 0 , ) h(z 0 , ) h(z 0 , zn )
d
h(z 0 , )
zn

h(z 0 , ) h(z 0 , zn )
zn

is a polynomial in zn of degree less than k (with coeficients which are functions of ) which
shows that q is a polynomial in zn of degree less than k.
To show that this representation is unique, suppose we have two representations
f = gh + q = g1 h + q1
with q and q1 both polynomials of degree less than k in zn . Then q q1 = h(g1 g)
is a polynomial of degree less than k in zn with at least k zeroes for each fixed value of
z 0 (0, r0 ). This is only possible if it is identically zero.
Now if f itself is a polynomial in zn then the usual division algorithm for polynomials
over a commutative ring gives a representation of f as above with g a polynomial in zn .
The uniqueness says that this must coincide with the representation given above. This
completes the proof.

14

J. L. TAYLOR

2.10 Theorem. The ring n H is a Noetherian ring.

Proof. We proceed by induction on the dimension n. For n = 0 we are talking about a
field, C, which has no non-trivial ideals and is trivially Noetherian. Suppose that n1 H
is Noetherian. Then we know from Hilberts basis theorem that n1 H[zn ] is Noetherian.
Suppose that I is an ideal in n H. Let h be any non-zero element of I. By performing
a coordinate change if necessary, we may assume that h is regular in zn and from the
Weierstrass preparation theorem we may (after multiplying by a unit if neccessary) assume
that h is a Weierstrass polynomial in zn . In other words, h belongs to I n1 H[zn ] which
is an ideal in the Noetherian ring n1 H[zn ] and is, thus, finitely generated by say g1 , ..., gm .
Now by the Weierstrass division theorem any element f I may be written as f = gh + q
with g n H and q I n1 H[zn ]. But this means that h and q both belong to I n1 H[zn ]
and, hence, to the ideal generated by g1 , ..., gm . Therefore f belongs to the ideal generated
by g1 , ..., gm and we conclude that this set of elements generates I. this completes the
proof.
1.

2.
3.
4.

5.

2. Problems
A unique factorization domain is an integral domain in which each element has a unique
(up to units) factorization as a finite product of irreducible factors. Prove that if A is
a unique factorization domain then so is A[x].
Prove that C[z1 , ..., zn ] and n O0 are unique factorization domains.
Prove that n H0 is a unique factorization domain.
Prove Nakayamas lemma: If M is a finitely generated module over a local ring A with
maximal ideal m and if mM = M then M = 0. Hint: Prove that if M has a generating
set with k elements with k > 0 then it also a generating set with k 1 elements.
Prove the implicit function theorem: If f is holomorphic in a neighborhood of a =
f
(a0 , an ), f (a) = 0, and
(a) 6= 0 then there is a polydisc (a, r) = (a0 , r0 )(an , rn )
zn
and a holomorphic map g : (a0 , r0 ) (a, r) such that g(a0 ) = a and, for each
z (a, r), f (z) = 0 if and only if z = g(z 0 ) for some z 0 (a0 , r0 ). Hint: Use the
Weierstrass preparation theorem with k = 1.

NOTES ON SEVERAL COMPLEX VARIABLES

15

3. A Little Homological Algebra

Let A be a commutative algebra over a field and let M and N be A-modules. We will
denote the tensor product of M and N as vector spaces over the field simply by M N .
The tensor product of M and N as A modules is denoted M A N and is the cokernel of
the map M A N M N defined by m a n am n m an. Clearly, M A N
is a covariant functor of each of its arguments for fixed values of the other argument.
The functor M () from the category of vector spaces to itself is exact. That is, if the
sequence
0 N1 N2 N3 0
is exact (the kernel of each map is equal to the image of the preceding map) then so is the
sequence
0 M N1 M N2 M N3 0
On the other hand, the functor M A () from the category of A-modules to itself fails to
be exact. It is right exact (kernel equals image at the right hand and middle stages) as is
easily seen by chasing the diagram
0 M A N1 M A N2 M A N3 0

y
y
y
0

M N1

M N2

M N3

M A N1

M A N2

M A N3

However, it is not generally true that this functor preserves exactness at the left stage.
The tensor product relative to A is an important functor and the fact that it is right
exact but not exact cannot be ignored. This circumstance requires careful analysis and
the development of tools that allow one to deal effectively with the problems it poses.
A similar problem arises with the functor homA . Again, if M and N are modules over A
then the space of linear maps from M to N is denoted hom(M, N ) while the A module of Amodule homomorphisms from M to N is denoted homA (M, N ). It can be described as the
kernel of the map hom(M, N ) hom(AM, N ) defined by {am a(m)(am)}.
The functor hom(M, ) is covariant and exact while the functor hom(, N ) is contravariant
and exact. On the other hand, the functors homA (M, ) and homA (, N ) are not generally
exact. It is true that they are both left exact, as is easily seen from the definition and a
diagram chase like the one above. Here again, the fact that homA (, N ) is not exact leads
to the need to develop tools to deal with the problems that this poses.
The first step in this program is to understand exactly when the two functors in question
are exact.

16

J. L. TAYLOR

3.1 Definition. An A-module M for which M A () is an exact functor from the category
of A-modules to itself is called a flat A-module.
An A-module M for which homA (M, ) is an exact functor is called a projective Amodule.
An A-module for which homA (, M ) is an exact functor is called an injective A-module.
Of course, each of these functors already satisfies two of the three conditions for exactness. Thus, M is flat if and only if for every injection i : N1 N2 the induced morphism
M A N1 M A N2 is an injection . Similarly, M is projective if and only if for every surjection N2 N3 the induced morphism homA (M, N2 ) homA (M, N3 ) is a surjection ( ie.
that every morphism from M to N3 lifts to N2 ). Finally, an A-module M is injective if and
only if for every injection N1 N2 the induced morphism homA (N2 , M ) homA (N1 , M )
is surjective (ie. every morphism from N1 to M extends to N2 ).
3.2 Proposition. Let M be an A-module. Then there are natural isomorphisms
A A M M

and M homA (A, M )

Proof. The map a m am : A M M has kernel which contains the image of the
map
a b m ab m a bm : A A M A M
P
P
In fact the
ai mi = 0, then
ai mi is the image under the latter
P two are equal since, if
map of
1 ai mi . It follows from the definition of A that a m am : A M M
induces an isomorphism from A A M to M .
The case of hom(A, M ) is even easier. Each element m M determines a homomorphism a am : A M and every homomorphism from A to M arises in this way from
a unique element m. In fact, m is just the image of 1 under the homomorphism. This
completes the proof.
A trivial consequence of this proposition is that the functors M A () and homA (M, )
are exact in the case where M = A. This is also clearly true if M is a finite direct sum of
copies of A. It is only slighltly less trivial that this continues to hold if M is an arbitrary
direct sum of copies of A. A module M which is a direct sum of an (arbitrary) set of
copies of A is called a free A-module . Thus, the functors M A () and homA (M, ) are
exact when M is any free A- module. Finally, it is an easy consequence of the definition
of direct summand that if M is a direct summand of a free A-module then M A () is an
exact functor. The same arguments applied to the functor homA (M, ) show that it is also
exact when M is a direct summand of a free A-module. Thus, direct summands of free
modules are both flat and projective.
Since A is an algebra over a field, an A-module M is free if and only if it has the form
M = A X where X is a vector space over the same field. Here, the module action is
given by the action of A on the left factor in the tensor product; that is, a(b x) = ab x.
Note, every A-module M is a quotient of a projective, in fact a free, A-module. In fact
the morphism
: A M M, (a m) = am

NOTES ON SEVERAL COMPLEX VARIABLES

17

expresses M as a quotient of the free A-module A M . If M happens to be projective,

the identity morphism from M to M lifts to A M and embedds M as a direct summand
of the free module A M . In summary, we have proved:
3.3 Proposition.
(a) An A-module is projective if and only if it is a direct summand of a free A-module;
(b) every projective module is flat;
(c) every A module is a quotient of a projective, in fact a free, A-module.
If P0 M is a morphism which expresses M as a quotient of a projective module P0
and if K is the kernel of this morphism then we may express K as a quotient of a projective
module P1 . This yields an exact sequence
P1 P0 M 0
by continuing in this way we construct a projective resolution of M , that is, an exact
sequence of the form:
Pn Pn1 P1 P0 M 0
in which each Pi is projective. Thus, we have proved:
3.4 Proposition. Each A module has a projective resolution.
For technical reasons it is useful to have a construction of projective resolutions which
is functorial, that is, one in which the complex of projectives
Pn Pn1 P1 P0 0
is a functor of the module M . There are many ways of doing this. We briefly describe one,
the Hochschild resolution, which has some nice properties and is used extensively. This is
the resolution
n+1

. . . Fn (M ) n . . . 2 F1 (M ) 1 F0 (M ) M 0
where Fn (M ) is the free A-module (n+1 A) M , (a m) = am, and
n (a0 an m) =

n1
X

(1)i a0 ai ai+1 an m+(1)n a0 an1 an m

i=0

A simple calculation shows that n n+1 = 0 for n 1 and 1 = 0, so that this is a

complex. To show that this complex is exact and, hence, provides a projective resolution
of M we construct a contracting homotopy for it as a sequence of vector spaces by defining
sn : Fn (M ) Fn+1 (M ) by
sn (a0 a1 an m) = 1 a0 a1 an m

18

J. L. TAYLOR

and s1 : M F0 (M ) by
s1 (m) = 1 m
Note that though the maps n are A-module homomorphisms the maps sn are only linear
maps and not A-module homomorphisms. A direct calculation shows that
sn1 n + n+1 sn = 1,

n1

and
s1 + 1 s0 = 1
and this is exactly what is meant by the statement that {sn } is a contracting homotopy
for the above complex. It is immediate that a complex with a contracting homotopy is
exact. We shall write F (M ) for the complex
n+1

. . . Fn (M ) n . . . 2 F1 (M ) 1 F0 (M ) 0
and

F (M ) M 0
for the corresponding resolution of M . To make sense of this notation, just think of
as being a morphism between two complexes, where M is thought of as the complex
whose only non-zero term is the degree zero term which is the module M . Finally, note
that the each of the functors Fn is, by construction, an exact functor from A-modules to
free A-modules and, hence, F is an exact functor from A-modules to complexes of free
A-modules.
To summarize the above discussion, we have
3.5 Proposition. The Hochschild functor functor M F (M ) is an exact functor from
A-modules to complexes of free A-modules and for each M

F (M ) M 0
is a free resolution of M .
We have used the term complex several times in this discussion. Actually two types of
complexes occur and it is time to be more precise. A chain complex C of A-modules is a
sequence of modules and morphisms of the form
n+2

n+1

n1

. . . Cn+1 Cn n Cn1 . . .
If C is a chain complex then its nth homology is
Hn (C) = ker n /im n+1
A cochain complex is just a chain complex with the modules indexed in increasing order
instead of decreasing order. Also, it is traditional to index cochain complexes with superscripts. Thus, a cochain complex C is a sequence of modules and morphisms of the
form
n2

n1

n+1

. . . C n1 C n C n+1 . . .
If C is a cochain complex then its nth cohomology is
H n (C) = ker n /im n1
n
We can now define the functors torA
n and extA .

NOTES ON SEVERAL COMPLEX VARIABLES

19

3.6 Definition. Let M and N be A-modules. Then we define

n
n
torA
n (M, N ) = Hn (M A F (N )) and extA (M, N ) = H (homA (F (M ), N ))

A quick look at the first two terms of the complexes M A F (N ) and homA (F (M ), N )
shows that:
0
3.7 Proposition. torA
0 (M, N ) = M A N and extA (M, N ) = homA (M, N ).

Another simple fact is that the isomorphism n m m n : M N N M

extends to an isomorphism between the complexes M A F (N ) and N A F (M ) and,
A
hence, determines an isomorphism torA
n (M, N ) torn (N, M ).
3.8 Theorem. Let 0 M1 M2 M3 0 be a short exact sequence of A-modules
and let N be any A-module. Then there is a natural long exact sequence
A
A
A
torA
n+1 (M3 , N ) torn (M1 , N ) torn (M2 , N ) torn (M3 , N )

torA
1 (M3 , N ) M1 A N M2 A N M3 A N 0
Again, the same result holds with the roles of M and N reversed.
Proof. If we take the tensor product of the short exact sequence 0 M1 M2 M3 0
with the Hochschild complex F (N ) for N we obtain a short exact sequence of complexes
0 M1 A F (N ) M2 A F (N ) M3 A F (N ) 0
That this is exact follows from the fact that F (N ) is a complex of free A-modules. A simple
diagram chase proves the standard result that every short exact sequence of complexes
induces a long exact sequence of the corresponding homology. This completes the proof.
The same sort of arguments yield analogous results for ext:
3.9 Theorem. Let 0 M1 M2 M3 0 be a short exact sequence of A-modules
and let N be any A-module. Then there are natural long exact sequences
0 homA (M3 , N ) homA (M2 , N ) homA (M1 , N ) ext1A (M3 , N )
extnA (M3 , N ) extnA (M2 , N ) extnA (M1 , N ) extn+1
A (M3 , N )
and
0 homA (N, M1 ) homA (N, M2 ) homA (N, M3 ) ext1A (N, M1 ) . . .
extnA (N, M1 ) extnA (N, M2 ) extnA (N, M3 ) extn+1
A (N, M1 )

20

J. L. TAYLOR

3.10 Theorem. Let M be a module. Then M is flat if and only if tornA (M, N ) = 0 for all
n > 0 and all modules N . Of course, the same statement holds with M and N reversed.
The module M is projective if and only if extnA (M, N ) = 0 for all n > 0 and all modules
N.
The module M is injective if and only if extnA (N, M ) = 0 for all n > 0 and all modules
N.
Proof. That tornA (M, N ) = 0 for all N and all n > 0 if M is flat follows from the fact that
M A () preserves exactness and, in particular, preserves the exactness of the Hochschild
resolution. The reverse implication follows from the long exact sequence for tor. The
proofs of the results for ext are completely analogous.
The following results are trivial consequences of Theorem 3.10 and the existence of the
long exact sequences:
3.11 Theorem. If 0 K P M 0 is a short exact sequence of modules with P
projective and N is any module, then there are natural isomorphisms
A
(M, N ) ' tornA (K, N ) n > 0
torn+1

tor1A (M, N ) ' ker{K A N P A N }

n
extn+1
A (M, N ) ' extA (K, N ) n > 0

ext1A (M, N ) ' coker{homA (P, N ) homA (K, N )}

This result is the basis for an induction argument that shows that tor and ext can be
computed from any projective resolution: Let
Pn Pn1 P1 P0 M 0
be a projective resolution of M and let Kn be the kernel of Pn Pn1 for n > 0, K0
the kernel of P0 M and set K1 = M . Then Theorem 3.11, applied to the short exact
sequences
0 Kn Pn Kn1 0, n 0
yields
tor1 (Kp2 , N ) ' ker{Kp1 N Pn1 N } ' Hp (P N ),

p1

and
torn+1 (Kq , N ) ' torn (Kq+1 , N ) n 0, q 0
By induction, this yields
torp (M, N ) = torp (K1 , N ) ' tor1 (Kp2 , N ) ' Hp (P N )
A similar argument works for ext. This leads to:

NOTES ON SEVERAL COMPLEX VARIABLES

21

3.12 Theorem. Given any projective resolution P M , as above, and any module N
there are natural isomorphisms
tornA (M, N ) ' Hn (P A N ) and extnA (M, N ) ' H n (homA (P, N ))
The result for tor holds if P is just a flat resolution of M .
Finally, we return to the study of the algebras n O and n H with a result which shows
that they have particularly simple free resolutions. First, note that if A is Noetherian then
any finitely generated A-module has a resolution by free finitely generated modules, that
is, a resolution of the form
n+1

n1

. . . Akn n Akn1 . . . 1 Ak0 M 0

Such a resolution is called a chain of syzygies for M . A chain of syzygies, as above, is said
to terminate at the nth stage if the kernel of n1 is free. In this case, the kernel of n1
may be used to replace An resulting in a chain in which the terms beyond the nth one are
all zero.
3.13 Theorem (Hilberts syzygy theorem). If A is the algebra n H or the algebra n O
and M is a finitely generated A-module then any chain of sysygies for M terminates at
the nth stage.
Proof. The proofs are not different in the two cases so we shall just do the case of A = n H.
Let Ij H = n H be the ideal generated by the germs z1 , . . . zj of the first j coordinate
functions. Thus, In is the maximal ideal of H. Also, let Kp Hkp denote the kernel of
the map p of the syzygy.
Our objective is to prove that Kn1 is free. To this end, let f1 , . . . , fs be a minimal set
of generators of Kn1 . Thus, no proper subset of this
P set generates Kn1 . We consider the
s
kn1
map : H H
defined by (g1 , . . . , gs ) =
gi fi . The image of is Kn1 . Thus,
without loss of generality we may replace the original syzygy with one in which n = .
We now show that the kernel Kn of this map is necessarily zero. Note that Kn In Hs
since, otherwise, there exist g1 , gs H such that
X

gi fi = 0

with some gj a unit. Then the set of generators of Kn1 could be reduced by throwing
out the corresponding fj . We shall show that this implies that Kn = In Kn which, by
Nakayamas lemma, shows that Kn = 0, as desired.
The fact that Kn In Hs implies Kn = In Kn follows immediately from the case
k = j = n of the equality
Kp I j H k p = I j Kp

for 1 j p

which we shall prove by induction on j. We need only prove that Kp Ij Hkp Ij Kp since
the reverse containment is clear.

22

J. L. TAYLOR

Suppose that j = 1 and f Kp I1 Hkp for some p 1. Then p (f ) = 0 and f = z1 g

for some g Hkp . Since z1 p (g) = p (f ) = 0, and H is an integral domain, it follows that
p (g) = 0 and that g Kp . Then f = z1 g I1 Kp as desired. Now assume that the above
equality holds for some j and all p j. Suppose that f Kp Ij+1 Hkp for some p j + 1.
This means that p (f ) = 0 and
f = z1 g1 + + zj+1 gj+1
for some g1 , . . . , gj+1 Hkp . From these two facts it follows that
zj+1 p (gj+1 ) = z1 p (g1 ) zj p (gj ) Hkp1
This implies that each monomial in the power series expansion of each component of
zj+1 p (gj+1 ) is divisible by one of the germs z1 , . . . , zj and it follows that the same thing
is true of p (gj+1 ) itself. Hence,
p (gj+1 ) = z1 g10 + + zj gj0
for elements g10 , . . . , gj0 Hkp1 . Since p1 p = 0 this implies that
p (gj+1 ) Kp1 Ij Hkp1
Since p 1 j it follows from the induction hypothesis that p (gj+1 ) Ij Kp1 =
Ij p (Hkp ). In other words,
p (gj+1 ) = z1 p (h1 ) + + zj p (hj )
for elements h1 , . . . , hj Hkp . If we set
hj+1 = gj+1 z1 h1 zj hj Hkp
then p (hj+1 ) = 0 so that hj+1 Kp . Also,
f zj+1 hj+1 = z1 (g1 + zj+1 h1 ) + + zj (gj + zj+1 hj )
and, hence, f zj+1 hj+1 Kp Ij Hkp which is Ij Kp by the induction hypothesis. Hence,
f Ij+1 Kp as required. this completes the induction and the proof.
3.14 Corollary. If A is n H or n O then every finitely generated A module has a free
resolution of the form:
0 Akn Akn1 Ak0 M 0

NOTES ON SEVERAL COMPLEX VARIABLES

23

3.15 Corollary. If A is n H or n O then for every pair of A-modules M and N we have

torpA (M, N ) = 0 = extpA (M, N ) for p > n.

3. Problems
1. Prove that if X is a vector space and hom(A, X) is given the obvious A-module structure,
then hom(A, X) is an injective A-module. Show that every A module is a submodule of
a module of this form. Then prove that a module is injective if and only if it is a direct
summand of a module of the form hom(A, X).
2. Fix C. Find a resolution of the form given in corollary 3.14 for the one dimensional
C[z]-module, C , on which each p C[z] acts as multiplication by the scalar p().
3. If V is any vector space and L end(V ) is any linear transformation, then we can make
V into a C[z]-module VL by letting p C[z] act on V as the linear transformation p(L).
A
Show that torA
1 (C , VL ) = ker ( L) and tor0 (C , VL ) = coker ( L). Thus, L
is invertible if and only if both of these tor groups vanish.
4. Prove a result analogous to the result of problem 3 but with tor replaced by ext.
5. Verify (if you have never done so before) that a short exact sequence of complexes yields
a long exact sequence of homology.
6. Show that each non-zero element of ext1A (M, N ) corresponds to a non-trivial extension
of M by N , that is, to a short exact sequence
0N QM 0
which does not split.

24

J. L. TAYLOR

4. Local Theory of Varieties

A subset V of Cn is a subvariety of Cn if for each point V there is a neighborhood
U of and functions f1 , . . . , fk holomorphic in U such that V U = {z U : f1 (z) =
= fk (z) = 0}. Algebraic subvarieties are defined in the same way except that Zariski
open sets are used and regular functions replace holomorphic functions. However, since
we can always multiply through by the common denominator of a finite set of regular
functions, we can always choose the functions f1 , . . . , fk that define the variety on U to
be polynomials in the case of an algebraic subvariety. If V and W are subvarieties of Cn
and V W then we will also say that V is a subvariety of W . If V is also a closed subset
of W then we will call it a closed subvariety of W .
Warning! In much of the several complex variables literature, subvarieties are required
to be closed. Our terminology is consistent with the algebraic geometry literature.
Note that V is a closed subvariety of an open set U Cn if and only if for each point
U there is a neighborhood U of and functions f1 , . . . , fk holomorphic in U such
that V U = {z U : f1 (z) = = fk (z) = 0}.
It is obvious that finite unions and intersections of closed subvarieties of an open set U
are also subvarieties of U . We may also define the germ of a closed subvariety of U at a
point U . That is, we define two closed subvarieties V and W of U to be equivalent at
if there is a neighborhood U of such that V U = W U . The germ of a closed
subvariety V at is then the equivalence class containing V . We will say V is the germ of
a variety at Cn if V is the germ of some closed subvariety of some neighborhood of .
Given finitely many germs of varieties, V1 , , Vk at , we may choose (by taking intersections, if necessary) a common neighborhood U of in which these germs have representatives. The germ of the intersection of these varieties is then well defined independent
of the choice of U and the representatives of the Vi . We will call this the intersection,
V1 Vk , of the germs V1 , , Vk . The union of finitely many germs of varieties is
defined in like manner, as is the relation .
4.1 Definition. If V is the germ of a holomorphic (algebraic) variety at 0 then id V is
defined to be the ideal of n H consisting of all germs which vanish on V . On the other
hand, if I is an ideal of n H (n O), then loc I is defined to be the germ of the subvariety of
a neighborhood of 0 defined by the vanishing of a finite set of generators for I.
One needs to check that these definitions make sense; that is, are id V and loc I well
defined? However, this is easy to do. The next theorem lists elementary properties of these
ideas that follow directly from the definitions:
4.2 Theorem. The following relationships hold between ideals of n H (n O) and germs of
holomorphic (algebraic) subvarieties at zero.
(a)
(b)
(c)
(d)
(e)
(f)

V1 V2 = id V1 id V2 .
I1 I2 = loc I1 loc I2 .
V = loc id V .
I id loc I but they are not generally equal.
id(V1 V2 ) = (id V1 ) (id V2 ) (id V1 ) (id V2 ).
id(V1 V2 ) (id V1 ) + (id V2 ).

NOTES ON SEVERAL COMPLEX VARIABLES

25

(g) loc(I1 I2 ) = loc(I1 I2 ) = loc(I1 ) loc(I2 ).

(h) loc(I1 + I2 ) = loc(I1 ) loc(I2 ).
A germ V of a variety is called reducible if V = V1 V2 where V1 and V2 are both
subvarieties contained properly in V . If V is not reducible then it is called irreducible.
Another elementary result which follows easily from the above is:
4.3 Theorem. A germ of a subvariety V is irreducible if and only if id V is a prime ideal.
4.4 Theorem. A germ of a subvariety can be uniquely (up to order) written as V1 Vk
where each Vi is an irreducible germ of a subvariety, each Vi is a proper subgerm of V and
no Vi is contained in a distinct Vj for i 6= j.
Proof. Suppose V is a germ of a subvariety and it cannot be written as a finite union of
irreducible germs of subvarieties. Then V itself is not irreducible and so it can be written
as V1 V2 for two germs of subvarieties which are properly contained in V . Then at least
one of V1 and V2 also fails to be a finite union of irreducibles. We assume without loss
of generality that this is V1 and then write V1 as the union of two proper subvarieties.
Continuing in this way will produce an infinite decreasing sequence of subvarieties each
properly contained in the next. This means that the corresponding sequence of ideals of
these subvarieties will form an infinite ascending chain of ideals of our Noetherian local
ring (n H or n O), which is impossible. Thus we have proved by contradiction that every
germ of a subvariety is a finite union of irreducibles. By deleting redundant subvarieties
we may get the other conditions satisfied.
If
V = V1 Vk = V10 Vm0
are two nonredundant ways of writing V as a union of irreducibles then for each i, Vi =
(Vi V10 ) (Vi Vm0 ) which implies that Vi is contained in one of the Vj0 . Likewise,
each Vj0 is contained in one of the Vi s. Since the decompositions are non-redundant we
conclude that each Vi is equal to some Vj and vice-versa. In other words, the decomposition
is unique up to order. This completes the proof.
Of particular interest are non-singular subvarieties. We shall describe these in the
holomorphic case first where the structure is particularly simple.
If U and U 0 are open subsets of Cn then a biholomorphic mapping from U to U 0 is a
holomorphic map with a holomorphic inverse. A holomorphic submanifold of an open set
U is a relatively closed subset V such that for each point V there is a neighborhood
U and a biholomorphic map F : U (0; r) onto some open polydisc in Cn such that
F () = 0 and
F (U V ) = {a (0; r) : zk+1 , . . . , zn = 0}
for some integer k. The integer k is called the dimension of the submanifold at .
Thus, a submanifold is a subvariety that locally, up to biholomorphic equivalence, looks
like a complex linear subspace. A germ of a holomorphic subvariety is called non-singular
if it is the germ of a holomorphic submanifold. In order to characterize holomorphic
submanifolds, we need the complex version of the familiar implicit mapping theorem from
advanced calculus. If U is a domain in Cn and F : U Cm is a holomorphic map

26

J. L. TAYLOR

with
coordinate
functions f1 , . . . , fm then the Jacobian of F is the m n matrix JF (z) =

fi
(z) .
zj
4.5 Theorem (Implicit mapping theorem). Let F be a holomorphic mapping as
above and suppose U and F () = 0. Suppose also that the last m columns of JF ()
form a non-singular m m matrix. Then there is a polydisc
(; r) = (0 ; r0 ) (00 ; r00 ) Cnm Cm
and a holomorphic map G : (0 ; r0 ) (00 ; r00 ) such that G(0 ) = 00 and F (z) = 0 for
z = (z 0 , z 00 ) (; r) if and only if G(z 0 ) = z 00 .
Proof. When m = 1 this is the implicit function theorem which is a simple corollary of the
Weierstrass preparation theorem in the case where the function is regular of degree one in
its last variable. We prove the general case by induction on m. Thus, we assume that the
result is true for m 1 and proceed to prove it for m.
Let JF () = (JF0 (), JF00 ()) be the separation of JF () into its first n m columns and
its last m columns. The hypothesis is that JF00 () is non-singular. By a linear change of
variables in the range space Cm we may assume that JF00 () is the m m identity matrix.
Then, since fm /zn = 1 at , it follows from the implicit function theorem that there is
a polydisc (; r) and a holomorphic mapping
h : (1 , . . . , n1 ; r1 , . . . , rn1 ) (n , rn )
such that h(1 , . . . , n1 ) = n and fm (z) = 0 for z (; r) exactly when zn =
h(z1 , . . . , zn1 ). Then we may define a holomorphic mapping
F 0 : (1 , . . . , n1 ; r1 , . . . , rn1 ) Cm1
0
to be
by defining its coordinate functions f10 , . . . , fm1

fi0 (z1 , . . . , zn1 ) = fi (z1 , . . . , zn1 , h(z1 , . . . , zn1 ))

Then F 0 (1 , . . . , n1 ) = 0 and the Jacobian JF 0 () has the property that its last m 1
columns form an (m1)(m1) identity matrix. It follows from the induction hypothesis
that, after possibly shrinking the polydisc, there is a holomorphic mapping
G0 : (0 ; r0 ) (nm+1 , . . . , n1 ; rnm+1 , . . . , rn1 )
such that G0 (0 ) = (nm+1 , . . . , n1 ) and such that F 0 (z1 , . . . , zn1 ) = 0 for a point
(z1 , . . . , zn1 ) (1 , . . . , n1 ; r1 , . . . , rn1 ) precisely when (znm+1 , . . . , zn1 ) = G0 (z 0 ).
Since F (z) = 0 for z (; r) precisely when zn = h(z1 , . . . , zn1 ) and F 0 (z1 , . . . , zn1 ) =
0, the mapping
G(z 0 ) = (G0 (z 0 ), h(z 0 , G0 (z 0 )))
has the required properties. This completes the proof.

NOTES ON SEVERAL COMPLEX VARIABLES

27

4.6 Inverse mapping theorem. If F is a holomorphic mapping from a neighborhood U

of Cn into Cn and if JF () is non-singular, then, on some possibly smaller neighborhood
U 0 of , F is a biholomorphic mapping to some neighborhood of F ().
Proof. This follows immediately from the implicit mapping theorem applied to the mapping H : Cn U Cn defined by H(z 0 , z 00 ) = F (z 00 ) z 0 .
4.7 Theorem. If F is a holomorphic mapping from a domain U in Cn into Cm and if JF
has constant rank k in U , then for each point U there is a neighborhood U of in which
F is biholomorphically equivalent to the mapping (z1 , . . . , zn ) (z1 , . . . , zk , 0, . . . , 0) from
a neighborhood of zero in Cn to a neighborhood of zero in Cm . Thus, {z U : F (z) = 0}
is a submanifold of dimension n k in U and, for each U , F (U ) is a submanifold of
dimension k in a neighborhood of F ().
Proof. We may assume that and F () are both the origin. After a linear change of coordinates, we may assume that the upper left hand k k submatrix of JF (z) is nonsingular
in a neighborhood U 0 of 0. We then define a new mapping G from U 0 into Cn by
G(z1 , . . . , zn ) = (f1 (z1 , . . . , zn ), . . . , fk (z1 , . . . , zn ), zk+1 , . . . , zn )
Then JG is non-singular in U 0 and thus G is a biholomorphic mapping of one neighborhood
0
0
U 00 of 0 Cn to another. Then F G1 has the form (z1 , . . . , zk , fk+1
, . . . , fm
). Since the
1
00
Jacobian of F G also has rank k throughout U , it follows that the functions fj0 /zi
vanish identically for i > k and, thus, that the functions fj0 are functions of z1 , . . . , zk
alone. If we set
0
0
H(z1 , . . . , zm ) = (z1 , . . . , zk , zk+1 fk+1
(z1 , . . . , zk ), . . . , zm fm
(z1 , . . . , zk ))

then H is a biholomorphic map on some neighborhood of 0 Cm and

H F G1 (z1 , . . . , zn ) = (z1 , . . . , zk , 0, . . . , 0)
on a neighborhood U of 0 Cn , as required. This completes the proof.
Note that for any holomorphic mapping F on a domain U , the set on which JF has
rank less than or equal to k is a subvariety of U since it is defined by the condition that
the determinants of certain submatrices of JF vanish. Let V = {z U : F (z) = 0} and
for each k let Vk = {z : rankJF (z) k} and let j be the largest integer k for which Vk is
a proper subvariety of U . Then U Vj is an open dense set on which rank JF is equal to
j + 1. The set V0 = V (U Vj ) consists of regular points of V and is a submanifold of
dimension n j of U Vj . It would be nice to know that V0 is an open dense subset of V ,
since this would show that most points of a variety are regular. However, is quite possible
for V and Vj to coincide, in which case, our discussion so far tells us nothing about the
regular points of V . In this situation we have made a bad choice of a mapping F to define
our subvariety. We need to be able to describe a variety as the set of common zeroes of a
set of functions which are chosen in a way that gives us much more detailed information
about the local structure of our subvariety at a point. We also need such a description
for the proof of Hilberts nullstellensatz. We obtain such an optimal choice of functions
defining a subvariety in the next section and use it to prove the nullstellensatz and several
other important facts concerning varieties.
We end this section with a brief discussion of the local ring of the germ of a variety.

28

J. L. TAYLOR

4.8 Definition. If V is a subvariety of Cn then a holomorphic (regular) function f on V

is a complex valued function on V with the property that for each point V there is a
neighborhood U of such that f extends to be holomorphic (regular) in U . The algebra
of functions holomorphic (regular) on V will be denoted V H (V O) while the algebra of
germs at V of functions holomorphic (regular) on neigborhoods in V of will be
denoted V H (V O ). The later is called the local ring of V at .
The following is immediate from the definition:
4.9 Theorem. If V is a holomorphic (algebraic) subvariety of Cn and V is its germ at
then V H = n H / id V and this ring is a noetherian local ring. The analogous statement
holds for V O .
4.10 Definition. If V and W are holomorphic (algebraic) subvarieties of Cn and Cm and
F : V W is a mapping then F is called holomorphic (regular) if each of its coordinate functions is a holomorphic (regular) complex valued function on V . A holomorphic
(regular) function with a holomorphic inverse is called biholomorphic (biregular).
Again it is immediate from the definition that:
4.11 Theorem. If V and W are holomorphic (algebraic) subvarieties of Cn and Cm and
F : V W is a mapping and V then F is a holomorphic (regular) mapping on some
neighborhood of if and only if F (g) = g F defines an algebra homomorphism F from
W HF () to V H (W OF () to V O ).
A somewhat deeper result is the following:
4.12 Theorem. If V and W are holomorphic subvarieties of Cn and Cm , V , W
and : W H V H is any algebra homomorphism, then is induced as in Theorem 4.11
by a holomorphic mapping from a neighborhood of in V to a neighborhood of in W .
The analogous result also holds for algebraic subvarieties.
Proof. The algebra homomorphism maps units to units. However, every element f in
either local ring has the property that there is a unique complex number c such that f c
is a non-unit. It follows that also maps non-units to non-units, ie. f is in the maximal
ideal of V H if and only if (f ) is in the maximal ideal of W H .
If w1 , . . . , wm are the germs in W H of the restrictions of the coordinate functions in
m
C to W then (w1 ), . . . , (wm ) are germs in V H and thus are represented by functions
which extend to be holomorphic in some neighborhood of in Cn . Let f1 , . . . , fm be such
holomorphic extensions. Then they are the coordinate functions of a holomorphic map F
from a neighborhood of in Cn into Cm . Since, for each j, (wj wj ()) = (fj )|V wj ()
which belongs to the maximal ideal of W H , it follows that fj () = wj () and, hence, that
F () = . Thus, the mapping F induces an algebra homomorphism F : m H
n H . If

we follow this by restriction to V then we have a homomorphism F : m H

V H . On
the other hand, if we precede by restriction from Cm to W , then it also determines a
homomorphism : m H
V H . Clearly the proof will be complete if we can show that

= F and that F maps some neighborhood of in V into W .

j ) for each j. This implies that the
Note that, by the construction of F , F (wj ) = (w
two homomorphisms agree on polynomials. Since, for each k every element of m H is a

NOTES ON SEVERAL COMPLEX VARIABLES

29

polynomial of degree k plus an element of the k th power of the maximal ideal of m H , and
since each of F and maps the k th power of the maximal ideal of m H to the k th power
) belongs to the k th power
of the maximal ideal of V H , we conclude that F (f ) (f
of the maximal ideal of V H for every f m H and every positive integer k. However,
by Nakayamas lemma the intersection of all powers of the maximal ideal in a Noetherian

local ring is zero. Therefore, F = .

It remains to prove that F maps a neighborhood of in V into W . However, id W
Hence, for every f id W we have
is in the kernel of = F by the definition of .

(f F )|V = F (f ) = 0, ie. f vanishes on F (V ). If we apply this fact to a finite set of

generators of id W we conclude that a suitably small neighborhood of in V is mapped
by F into W .
Two germs of varieties are said to be biholomorphically equivalent if there is a biholomorphic map between suitable representative neighborhoods. In view of the preceding
theorem we have:
4.13 Corollary. Two germs of varieties are biholomorphically equivalent if and only if
their local rings are isomorphic as algebras.

4. Problems
1. Give an example which shows that the implicit function theorem, the inverse mapping
theorem and the Weierstrass preparation theorem fail in the algebraic case.
2. Consider the polynomial on C2 defined by p(z, w) = z 2 w3 . Prove that p is irreducible
in both 2 H and 2 O and, hence, generates a prime ideal in each algebra.
3. Show that in either 2 H or 2 O the ideal generated by the polynomial p in problem 2 is
id V where V = {(z, w) C2 : p(z, w) = 0.
4. With V as in problem 2, show that there are irreducible elements f, g V H (V O) such
that f 2 = g 3 . Conclude that these algebras are not unique factorization domains.
5. With V as in problem 2, let M be the maximal ideal of V H. Show that M is generated
by two of its elements in such a way the the resulting morphism V H2 M has kernel
isomorphic to M as a V H -module. Conclude that the Hilbert syzygy theorem fails to
hold for V H.

30

J. L. TAYLOR

5. The Nullstellensatz
In rings A for which it is true, the Nullstellensatz says that

id loc I = I = {x A : xn I for some n }

for each ideal I. This is true for the rings n H and n O as well as
for the ring of polynomials
C[z1 , . . . , zn ]. In each case, it is trivial from the definition that I id loc I. Also in each
case, the theorem can easily be reduced
to the case of prime ideals. In fact, a primary

ideal is an ideal I whose radical I is prime and, in a Noetherian ring, each ideal I has
a primary decomposition I = m
j=1 Ij . Thus, if we assume the Nullstellensatz for prime
ideals, then we have
id loc I = id(

m
[

m
loc Ij ) = m
j=1 id loc Ij j=1 id loc

p
p
Ij = m
I
=
I
j
j=1

j=1

which is the Nullstellensatz for general ideals since we already have the reverse containment.
Our goal in this section is to prove the Nullstellensatz for the ring n H. However, we
first prove the much easier result that the Nullstellensatz holds for n O. It is easy to see
that the Nullstellensatz for C[z1 , . . . , zn ] implies the Nullstellensatz for n O (this is left as
an exercise) so we shall prove the Nullstellensatz for C[z1 , . . . , zn ]. We first need a couple
of lemmas from commutative algebra.
5.1 Lemma. Let A B C be rings with A Noetherian. Suppose that C is a finitely
generated A-algebra and C is integral over B. Then B is also a finitely generated A
algebra.
Proof. We first show that C is actually a finitely generated B-module. The fact that C is
finitely generated as an A-algebra means that it is also finitely generated as a B-algebra
and, hence, that every element of C is a polynomial in a finite set of generators c1 , . . . , ck
with coeficients in B. However, the fact that C is integral over B means that for each
c C there is an integer nc such that every polynomial in c is equal to one of degree
less than or equal to nc . Thus, the algebra generated over B by c1 is a finitely generated
B-module. An induction argument on the number of generators now shows that C is a
finitely generated B-module as claimed.
Now let x1 , . . . , xl generate C as an A-algebra and y1 , . . . , ym generate C as a B-module.
Then there exist elements bij and bijk in B such that
X
X
xi =
bij yj , yi yj =
bijk yk
j

Let B0 be the algebra generated over A by the bij and the bijk . We have A B0 B and
B0 is Noetherian since A is Noetherian and B0 is a quotient of a polynomial ring over A.
The above equations show that each element of C is a linear combination of the yi with
coeficients from B0 so that C is a finitely generated B0 -module. Since B0 is Noetherian and
B is a submodule of C it follows that B is also a finitely generated B0 -module. Since B0 is
finitely generated as an A-algebra it follows that B is finitely generated as an A-algebra.
This completes the proof.

NOTES ON SEVERAL COMPLEX VARIABLES

31

5.2 Lemma (weak Nullstellensatz). Let k be an algebraically closed field and A a

finitely generated k-algebra. Then for each maximal ideal M of A, the inclusion of k into
A/M is an isomorphism.
Proof. Let x1 , . . . , xn be generators of A/M over k and assume that they are chosen such
that x1 , . . . , xj are algebraically independent over k and each of the others is algebraic
over the field F = k(x1 , . . . , xj ). Thus, A/M is a finite algebraic field extension of F
and, hence, a finitely generated F -module. It follows from the previous lemma applied
to k F A/M that F is a finitely generated k-algebra - that is, there is a finite
set of fractions yi = fi /gi , with fi , gi k[x1 , . . . , xj ] such that every element of F is a
polynomial in the yi . Now if the set {x1 , . . . , xj } is not the empty set then there is an
irreducible polynomial
h k[x1 , . . . , xj ] prime to all the gi s since the gi s have no common
Q
factors with 1 + gi . However, h1 F and so it must be a polynomial in the yi s. This
i

implies that h1 k k[x1 , . . . , xj ] where k is some product of the gi0 s, which is impossible if
h is relatively prime to all the gi0 s in k[x1 , . . . , xj ]. The resulting contradiction shows that
{x1 , . . . , xj } must be empty. But in this case A/M is algebraic over k and, hence, equal
to k since k is algebraically closed.
5.3 Theorem (Nullstellensatz for polynomial algebras). If A = C[z1 , . . . , zn ] then
for each ideal I A we have

I = id loc I
Proof. We may assume I is prime. We know that I id loc I and so we need only prove
the reverse containment. Let f A be any element not in I and let B = A/I and C = Bf ,
the algebra of fractions over B with denominators that are powers of the image of f in
B. Now let M be a maximal ideal of C. Since C is finitely generated over C it follows
from the previous lemma that C/M = C. Then the images of z1 , . . . , zn in C/M are the
coordinates of a point Cn . It is clearly a point in loc I (since the maximal ideal it
determines contains I by construction) and a point at which f () 6= 0 (since f is invertible
in C). Thus, f is not in id loc I and the proof is complete.
The above proof depended heavily on the fact that quotients of C[z1 , . . . , zn ] are finitely
generated algebras over the ground field. We have no such finite generation conditions in
the holomorphic case and must use entirely different methods. The proof of the Nullstellensatz for n H depends on a fairly precise description of the locus of a prime ideal as the
germ of a certain kind of finite branched cover of a neighborhood in Cm . We now proceed
to develop this description.
In what follows we will be making fairly heavy use of field theory. Since this discussion
will not make much sense to someone who doesnt know a certain amount of field theory,
we present below a list of facts from this subject that we will use implicitly or explicitly.
This is presented as a study guide for those who need to brush up on the subject. Proofs
can be found in any algebra book with a good treatment of field theory (eg: Hungerford).
If K F are fields and x F then x is called algebraic over K if it is the root of a
polynomial with coeficients in K. The field F is called an algebraic field extension of K if
each of its elements is algebraic over K.

32

J. L. TAYLOR

F1 Theorem. The following are equivalent for fields K F :

(a) F is an algebraic field extension of K and is finitely generated over K;
(b) F is generated over K by finitely many elements which are algebraic over K;
(c) F is a finite dimensional vector space over K.
F2 Theorem of the primitive element. If K is an infinite field and F is a finitely
generated algebraic extension of K then F is generated by a single element which may be
chosen to be a linear combination of any given set of generators.
The subfield of F generated by x1 , . . . , xn is denoted K(x1 , . . . , xn ).
F3 Theorem. If x F is algebraic over K then x is a root of a unique irreducible
monic polynomial with coefficients in K. If the degree of that polynomial is n, then
{xn1 , . . . , x, 1} is a basis for K(x) as a vector space over K.
The unique irreducible monic polynomial having x as a root is called the minimal
polynomial of the element x.
If p is a polynomial with coeficients in K then a splitting field F for p is an extension
field of K which is generated over K by the roots of p and in which p factors as a product
of linear factors (so that all possible roots of p are included in F ).
F4 Theorem. If p is a polynomial with coeficients in K then there is a splitting field
for p which is unique up to isomorphism. If p is the minimal polynomial of an algebraic
element x over K, then a splitting field for p may be chosen which is a field extension of
K(x).
The Galois group of a field extension K F is the group of automorphisms of F which
leave all elements of K fixed. The extension is called a Galois extension if K is exactly
the set of elements fixed by the Galois group.
F5 Theorem. If F is a splitting field for some polynomial p with coeficients in K and if
K has characteristic zero, then F is a Galois extension of K. Every element of the Galois
group of such an extension is uniquely determined by a permutation of the roots of p. If
p is irreducible then the Galois group acts transitively on the roots.
If p is a polynomial with coeficients in K then the discriminant of p is the product
d=

Y
(xi xj )
i6=j

where x1 , . . . , xn are the roots of p in some splitting field.

F6 Theorem. If p is a polynomial with coeficients in K then its discriminant is a well
defined element of K and it is non-zero if and only if p has no multiple roots.
F7 Theorem. If p is an irreducible polynomial with coeficients in K and if K has characteristic zero then p has no multiple roots and, thus, has non-vanishing discriminant.

NOTES ON SEVERAL COMPLEX VARIABLES

33

F8 Theorem. If p is a polynomial with coeficients in K and if its roots in some splitting

field are x1 , . . . , xn then the discriminant of p is the square of the Vandermonde determinant
n1
x1
n1
x2

n1
xn

xn2
1
xn2
2

xn2
n

...
...
...
...
...

x1
x2

xn

A ring is said to be integrally closed if it is integrally closed in its quotient field. That
is, if whenever a monic polynomial with coeficients in the ring has a root in the quotient
field of the ring then that root actually lies in the ring.
F9 Theorem. A unique factorization domain is integrally closed.
The next result may be less well known than its predecessors and so we shall actually
prove it.
F10 Theorem. If A is a unique factorization domain with quotient field K of characteristic zero and if F = K(x) is the field extension generated by an element x, integral over
A, with minimal polynomial p of degree n, then every element of F which is integral over
xn1
x 1
A belongs to the A submodule of F generated by the elements
, . . . , , , where d is
d
d d
the discriminant of p.
Proof. For an element f (x) K(x), integral over A, we wish to find coeficients a0 , . . . , an1
such that
an1 xn1 + an2 xn2 + + a1 x + a0 = d f (x)
Let x1 = x and let x2 , . . . , xn be the other roots of p in a splitting field for p. We may then
write down a system of n equations, each of which is a copy of the one above but with x
replaced by xj in the j th equation. If we consider this as a system of equations in which
the unknowns are the elements a1 , . . . , an , Kramers rule give as solution
n1
x1
n1
x2

aj = d

n1
xn

xn2
1
xn2
2

n2
xn

...
...
...
...
...

x1
x2

xn

1
1 xn1
1
1 xn1
2



n1
1
xn

xn2
1
xn2
2

n2
xn

...
...
...
...
...

f (x1 )
f (x2 )

f (xn )

...
...
...
...
...

x1
x2

xn

where in the second determinant the f (xi ) replace the j th column of the first determinant.
Now, of course, the first determinant is the Vandermonde determinant which has square
equal to d by F8. Thus, aj is the product of the Vandermonde and the determinant
obtained from the Vandermonde by replacing its j th column with the column formed by
the f (xi ). Clearly this product is left fixed by any permutation of the roots x1 , . . . , xn
since this just amounts to applying the same permutation to the rows in both matrices.
Thus, the elements aj so determined are fixed by the Galois group of the splitting field of
p and belong to K by F5. However, since x and f (x) are integral over A so are all the xi

34

J. L. TAYLOR

and f (xi ), again by F5. It follows that the aj are also integral over A since they lie in the
ring generated by the xi and f (xi ). However, A is integrally closed in its quotient field K
by F9 and, hence, aj A for j = 1, . . . , n. This completes the proof.
We now return to our study of a prime ideal P n H.
Recall that a holomorphic function f in a neighborhood of 0 is called regular in zn if
f (0, . . . , 0, zn ) is not identically zero. In what follows, we consider j H for j n to be the
subring of n H consisting of functions that depend only on the first j variables.
5.4 Definition. An ideal I n H is called regular in the variables zm+1 , . . . , zn if m HI =
0 and for each j {m + 1, . . . , n} there is an element fj j H I which is regular in zj .
5.5 Lemma. For each non-zero ideal I n H there is a choice of a complex linear coordinate system for Cn and an m < n such that I is regular in the variables zm+1 , . . . , zn .
Proof. We can choose a non-zero fn I and then by a suitable linear change of coordinates
arrange that fn is regular in zn . Suppose we have chosen fj+1 , . . . , fn satisfying the
conditions of Definition 5.4. Then either j H I = 0, in which case we are through, or
there is a nonzero fj j H I. The function fj can be made regular in zj by a linear
change of coordinates that involves only the first j coordinates and, hence, does not effect
the regularity of the functions chosen previously. The Lemma follows by induction.
The notion of an ideal being regular in the variables zm+1 , . . . , zn seems to depend on
the ordering of these variables. However, the next lemma shows that it depends only on
the decomposition Cn = Cm Cnm and not on the choice of coordinate systems within
the two factors.
5.6 Lemma. An ideal I n H is regular in the variables zm+1 , . . . , zn if and only if m H
in n H
= n H/I and n H
is an integral algebraic extension of
is isomorphic to its image m H

m H generated by the images of zm+1 , . . . , zn in n H/I.

is an
Proof. Obviously we have that m HI = 0 if and only if the natural map of m H to m H
isomorphism. Thus, we need to prove that the existence of the functions fj in Definition
are all integral over
5.4 is equivalent to the fact that the images of zm+1 , . . . , zn in n H

j of zj in n H is integral over m H then it is a root of a monic

m H. However, if the image z
Lifting these coeficients to m H results in a Weierstrass
polynomial with coeficients in m H.
polynomial in j H I which will serve as the element fj . Now on the other hand, if there
are fj j H I, regular in zj , then we may assume that they are Weierstrass polynomials
by the Weierstrass preparation theorem. Since these polynomials belong to I this implies
for j > m is integral over j1 H.
Furthermore, by the
that the image zj of each zj in n H
Weierstrass division theorem, for any f j H there are elements g j H and r j1 H[zj ]
such that f = fj g + r. On passing to residue classes mod I and remembering that fj I,
is a polynomial in zj over the ring j1 H.
Thus, j H

we conclude that every element of j H

by the element zj . We now have that
is the integral algebraic extension of the ring j1 H

j for
n H is obtained from m H by successive integral algebraic extensions by the elements z
j = m + 1, . . . , n. The theorem of transitivity of integral dependence now implies that

zm+1 , . . . , zn ] where each of the elements zj is integral over m H.

This completes
n H = m H[
the proof.

NOTES ON SEVERAL COMPLEX VARIABLES

35

We can do even better in the case that our ideal, regular in zm+1 , . . . , zn , is a prime
= n H/P is an integral domain and has a field of quotients n M.

ideal P . In this case, n H

' m H have fields of quotients m M
' m M and n M
is an algebraic field extension
Also m H
by the elements zm+1 , . . . , zn .
of m M
5.7 Definition. A prime ideal is called strictly regular in the variables zm+1 , . . . , zn if it
is generated over m M
by the single element zm+1 .
is regular in these variables and n M
5.8 Lemma. Any non-zero prime ideal P can be made strictly regular in some set of
variables zm+1 , . . . , zn by a linear change of coordinates.
Proof. By a linear change of coordinates we can make P regular in some set of variables
is an algebraic field extension of m M
by the
zm+1 , . . . , zn . Then, as noted above, n M
is actually generelements zm+1 , . . . , zn . By the theorem of the primitive element, n M
by a single element which may be chosen to be a linear combination of
ated over m M
zm+1 , . . . , zn . Another linear change of variables effecting only these coordinates can be
used to transform this element into zm+1 . Such a change of variables does not change the
fact that P is regular in the variables zm+1 , . . . , zn by Lemma 5.6. This completes the
proof.
Let P be a prime ideal of n H. We will choose a particularly nice finite set of elements of
P which determine P in a certain fashion although they are not quite a set of generators.
We assume that coordinates have been chosen so that P is strictly regular in the variables
For each j we let pj be the minimal
zm+1 , . . . , zn . Then the zj are all integral over m H.
of m H.
However,
polynomial of zj . A priori, pj has coeficients in the quotient field m M

all roots in a splitting field of pj are also integral over m H by the transitivity of the Galois
group and, hence, the coeficients of pj , being elementary symmetric functions of the roots,
Since the latter algebra is a unique factorization domain and, hence,
are integral over m H.
and ,
integrally closed, we conclude that the coeficients of each pj are actually in m H
hence, may be considered elements of m H. Then each polynomial pj (zj ) is an element of
m H[zj ] n H which belongs to P since it vanishes mod P . The pj (zj ) are some of the
elements of P that we are seeking.
We use F10 to choose the remaining elements. For j = m + 2, . . . n the image zj of zj in

zm+1 ) is integral over m H

and, thus by F10, d zj = sj (
zm+1 ) where d is the
n M = m M(
discriminant of pm+1 and sj is a unique polynomial of degree less than the degree of pm+1
' m H. Then qj = d zj sj (zm+1 ) belongs to m H[zm+1 , zj ] P .
with coeficients in m H
The qj together with the pj are the elements we need to adequately describe P and loc P .
We introduce two sub ideals of P that will play a role in what follows. We let I denote
the ideal generated by pm+1 , . . . , pn , qm+2 , . . . , qn and I 0 denote the ideal generated by
pm+1 , qm+2 , . . . , qn .
Lemma 5.9. With d the discriminant of pm+1 , as above, there is an integer such that
for any f n H there is a polynomial r m H[zm+1 ] with degree less than degree pm+1
such that d f r I. Furthermore,
d P I P

and

d I I 0 I

36

J. L. TAYLOR

Proof. The polynomial pn is a Weierstrass polynomial and so, for any f n H, the Weierstrass division theorem allows us to write f = pn gn + rn for some gn n H and some
rn n1 H[zn ] of degree less than the degree of pn . We now apply the Weierstrass division theorem to each coeficient of rn with the divisor being the Weierstrass polynomial
pn1 m H[zn1 ] n2 H[zn1 ]. If we gather together the terms this yields
f = pn gn + pn1 gn1 + rn1
with gj n H and rn1 n2 H[zn1 , zn ]. By repeating this argument as long as we have
pj s we eventually get
f = pn gn + + pm+1 gm+1 + rm+1
with gj n H and rm+1 m H[zm+1 , . . . , zn ]. Also note that the degree of rj is less than
the degree of pj for j = m + 1, . . . , n.
Now for each j = m + 1, . . . , n we have d zj = qj + sj and so, by the binomial theorem,
dk zjk = hjk qj + skj for each integer k and some hjk
P n H. If we apply this to each power
of each zj appearing in rm+1 and if we choose = (degree pj 1) then we conclude that
d rm+1 = qm+2 hm+2 + + qn hn + r0
for some hj n H and r0 m H[zm+1 ]. Another application of the Weierstrass division
theorem gives us r0 = pm+1 hm+1 + r where hm+1 n H and r m H[zm+1 ] with degree of
r less than the degree of pm+1 . Finally, this gives us

d f =

n
X
j=m+1

d gj pj +

n
X

hj qj + hm+1 pm+1 + r

j=m+2

which implies d f r I as required.

Obviously I P . Now suppose that f P and apply the above conclusion to f . Then
r P and so r = 0 since it has degree less than that of pm+1 which is the polynomial of
lowest degree in P m H[zm+1 ]. Thus, d f I and we conclude that d P I.
Since I 0 I is obvious, to finish the proof we need to prove that d I I 0 or, in other
words, that d pj I 0 for j = m + 1, . . . , n. To show this, we again apply the binomial
theorem to the expression (d zj )k = (qj + sj )k for each power of zj appearing in the
polynomial pj to obtain d pj = hj qj + r0 where r0 m H[zm+1 ]. Another application of
the Weierstrass division theorem gives us r0 = hm+1 pm+1 + r where r m H[zm+1 ] and
degree of r is less than the degree of pm+1 . As before it follows that r P and , Hence,
r = 0 since pm+1 has minimal degree for an element of P m H[zm+1 ]. It follows that
d pj = hj qj + hm+1 pm+1 I 0 as required. This completes the proof.
This has the following as an immediate corollary:
5.10 Corollary. If D is the locus of the ideal in n H generated by d then
(i) loc P loc I loc I 0 ;
(ii) D loc P = D loc I = D loc I 0 .

NOTES ON SEVERAL COMPLEX VARIABLES

37

5.11 Definition. Let V and W be closed holomorphic subvarieties of open subsets of Cn

and Cm , respectively and let : V W be a finite to one, proper holomorphic map.
Then is said to be a finite branched holomorphic cover if there are dense open subsets
W0 W and V0 = 1 (W0 ) V such that W W0 is a subvariety of W and : V0 W0
is a locally biholomorphic mapping.
Let : V W is a finite branched holomorphic cover, as above, and let p be a point of
W0 . Let 1 (p) = {q1 , , qk } and suppose we have chosen for each i a neighborhood Ui
of qi in V0 such that UiS
Uj = for i 6= j. If A is a neighborhood of p with compact closure
1
in W0 , then (A) Ui is a compact subset of V0 . The collection of sets of this form is
closed under finite intersection and so, if they are all non-empty, then there is a point q in
their intersection. Then, necessarily, (q) = p and, hence, q be one of the qi . This is not
possible since q is in the complement of each Ui . It follows that for some choice of A the
set 1 (A) is contained in the disjoint union of the open sets 1 (A) Ui . Now suppose
that p is a point of W0 and the Ui have been chosen so that |Ui is T
a biholomorphic map
onto a neighborhood of p for each i. If A is chosen to be a subset of (Ui ), then 1 (A)
is the disjoint union of the sets 1 (A) Ui and, for each i, is a biholomorphic map of
1 (A)Ui onto A. Thus, each point of W0 has a neighborhood which, under , is covered
by a finite number of biholomorphic copies of itself. A map with this property is a finite
holomorphic covering map. In particular, it is a finite covering map in the topological
sense.
Note that the number of points in the inverse image of a point of W0 is locally constant
on W0 . If W0 is connected then this number is a constant r and, in this case, we say that
: V W is a finite branched holomorphic cover of pure order r and : V0 W0 is a
finite holomorphic covering map of pure order r.
Now suppose W0 is locally connected in W (so that each point of W has a neighborhood
which intersects W0 in a connected set). Also suppose that p W but p is no longer
necessarily in W0 . Suppose {U1 , , Uk } is a pairwise disjoint collection of open subsets
S of
1
V with qi Ui . As above, we may choose a neighborhood A of p so that (A) Ui .
By replacing each Ui with Ui 1 (A), we may assume that
[
1 (A) =
Ui
We may also choose A so that A0 = W0 A is connected. By the above paragraph, 1 (A0 )
is a finite holomorphic cover of A0 of pure order. If we set Ui0 = 1 (A0 )Ui = Ui V0 , then
Ui0 is dense in Ui . It follows that the restriction of to Ui is a finite branched holomorphic
cover of A. The next proposition summarizes the preceding discussion:
5.12 Proposition. If : V W is a finite branched holomorphic cover with W0 and
V0 = 1 (W0 ) as in Definition 5.11, then
(a) : V0 W0 is a finite holomorphic covering map; that is, each point p W0 has
a neighborhood A such that 1 (A) is a finite disjoint union of open sets on each
of which is a biholomorphic map onto A;
(b) if W0 is locally connected in W , p W and q 1 (p), then there are arbitrarily
small neighborhoods U of q and A = (U ) of p such that : U A is a finite
branched holomorphic cover of pure order.

38

J. L. TAYLOR

With : V W as in part (b) of the above proposition and q V , we know that there
are arbitrirly small neighborhoods of q on which is a finite branched holomorphic cover
of pure order. For small enough neighborhoods the order must stabilize at some positive
integer oq (). We call this integer the branching order of at q.
5.13 Lemma. Suppose pm+1 , . . . , pn are monic polynomials with coefficients in m HU for
some open set U Cm and with non-zero discriminants dm+1 , . . . , dn and let V = {z
Cn : pj (zj ) = 0, j = m + 1, . . . n}. Then the projection Cn Cm exhibits V as a finite
branched holomorphic cover of U .
Proof. We let D be the union of the zero sets of the discriminants dj . Then U0 = U D
is an open dense subset of U and we set V0 = 1 (U0 ) where : V U is the restriction
of the projection Cn Cm to V . We need to show that V0 is dense in V , that is proper
and finite to one on V and locally biholomorphic on V0 .
Now let K U be compact and set L = 1 (K). We claim that L is a bounded subset
of Cn . Clearly the j th components of points of L are bounded if j m since then they are
j th components of points of K. For j > m the j th components of points of L are bounded
because they satify the monic polynomial equation pj (zj ) = 0. If we divide this equation
n 1
by zj j , where nj is the order of pj , we may use the resulting equation to estimate |zj | in
terms of the coefficients of pj , on the set where |zj | 1. We conclude that |zj | is bounded
by the maximum of 1 and the sum of the suprema of the absolute values of the coefficients
of pj on K. Thus, the set of j th components of points of L is a bounded set for all j and,
therefore, L is bounded. The set L is also closed in Cn since it is just the set of points
in Cn which map to K under : Cn Cm and at which each pj vanishes. Hence, L
is compact. Thus, is a proper holomorphic map. The fact that each pj has a finite,
non-empty set of zeroes for each fixed value of (z1 , , zm ) shows that it is finite to one
and surjective.
Let a = (a0 , a00 ) be a point of V U Cnm with a0 the corresponding point of U .
Each polynomial pj is monic and, hence, pj (zj ) is regular of some order greater than zero
in the variable zj at a. Given > 0, Theorem 2.6 implies that we may choose a polydisc

0 , r0 ) (a
00 , r00 ) U Cnm , with r00 < , such that for each j and for
(a,
r) = (a
j
r0 ) the roots of pj (z 0 , zj ) all lie in the interior of (a
00 , r00 ) and the number of
z 0 (0,
j
0 , r0 ). This is a kind
these roots is constant, counting multiplicity, as a function of z 0 (a
of continuity of the roots result. It implies in particular that for each point of V there are
arbitrarily nearby points that lie over points of the open dense set U0 . In other words, V0
is dense in V .
Since U0 is exactly the set of z 0 U at which the roots of all the pj are distinct, the
inverse image V0 of this set under is the subset of U Cnm on which each pj vanishes
but its derivative with respect to zj does not vanish. Thus, the map F : Cn Cnm
defined by
F (z1 , , zn ) = (pm+1 (z1 , , zm , zm+1 ), , pn (z1 , , zm , zn ))
has Jacobian JF in which the last n m columns form a diagonal matrix with entries that
do not vanish on V0 . Thus, JF has rank n m in an open set containing V0 . It follows
from the implicit mapping theorem that, for each V0 , there is a neigborhood A of

NOTES ON SEVERAL COMPLEX VARIABLES

39

in Cn , a neighborhood B of () in Cm and a holomorphic map G : B A such that

the points of A where F vanishes (i. e. the points of V0 A ) are exactly the points in the
image of G. It follows that G is a holomorphic inverse for the restriction of to V0 A .
Thus, locally has a holomorphic inverse on V0 . In other words is locally biholomorphic
on V0 and exhibits V as a branched holomorphic cover of U This completes the proof.
Our further study of finite branched holomorphic covers depends on the following technical lemma.
5.14 Lemma. Given positive integers n and r, there exists a finite set {f1 , , fq } of
linear functionals on Cn such that, for any set of r distinct points {z1 , , zr } Cn there
is some i for which the numbers fi (z1 ), , fi (zr ) are distinct.
Proof. We may assume r 2. Choose an integer q > 21 r(r 1)(n 1) and choose a set of
linear functionals on Cn , {f1 , , fq }, such that every subset of this set with n elements is
linearly independent. If the elements fi are interpreted as the rows of a q n matrix, then
this is just the condition that each n n submatrix has non vanishing determinant. Such
a choice is clearly possible since the union of the zero sets in Cnq of these determinants is
a subvariety of dimension less than nq.
Now, given distinct integers j and k between 1 and r, the set of linear functionals f on
n
C for which f (zj zk ) = 0 is a linear subspace of dimension n 1 and, hence, it may
contain at most n 1 of the functionals f1 , , fq since any set of n of these is linearly
independent. There are 12 r(r 1) unordered pairs of distinct points in the set {z1 , , zr }
so there are at most 21 r(r 1)(n 1) integers i for which an equation of the form
fi (zj zk ) = 0,

with j 6= k

can be satisfied. By the choice of q, there must be at least one index i between 1 and q such
that no such equation holds. For this i, the functional fi separates the points z1 , , zr .
5.15 Lemma. Let : V W be a finite branched holomorphic cover where W is a
domain in Cm and let D be a proper subvariety of W , with the property that V0 =
V 1 (D) is dense in V and is locally biholomorphic on V0 . If V1 is a connected
component of V0 , then the closure V1 of V1 in V is a holomorphic subvariety of V .
Proof. The lemma is purely a local statement and so we may assume that W is a polydisc
in Cm and that V is a closed subvariety of some polydisc in Cn .
The set W0 = W D is connected (by Problem 5.2) and V0 has finitely many components, each of which is a finite unbranched holomorphic cover of W0 and one of which
is V1 . Let the points of V1 over a point w W0 be labeled 1 (w), . . . , k (w) and for a
function f holomorphic on V let p be the polynomial in the indeterminant x defined by
p(w, x) =

k
Y

(x f (j (w)))

j=1

For w in a neighborhood of each point of W0 it is possible to choose the labeling of the

j (w) in such a way that these functions are holomorphic, although it may not be possible

40

J. L. TAYLOR

to do this globally. However, since the coefficients of the polynomial p are independent of
the labeling of the roots, they are well defined and holomorphic in all of W0 . In fact, if f is
holomorphic in all of V , it is locally bounded there, which implies that the coefficients of
p are as well. The generalized removable singularities theorem (Theorem 2.7) then implies
that they extend to be holomorphic in all of W . Now we have that p is a polynomial
with coefficients holomorphic in W and with the property that, whenever w W0 , the
roots of p(w, x) are exactly the values assumed by f on the set 1 (w) V1 . In particular,
p((z), z) vanishes on V1 . By continuity, this function also vanishes on V1 .
Since W0 is connected, is a cover of pure order r for some r. We apply the previous
lemma to obtain linear functionals f1 , , fq such that any set of r distinct points in
Cn can be separated by some one of the functionals fi . We then let p1 , , pq be the
polynomials constructed, as above, for the functions f1 , , fq . Each of the functions
pj ((z), fj (z)) vanishes on V1 and so we let V be the subvariety of V on which they all
vanish. We endeavor to prove that V1 = V . To this end, let a1 be a point of V and let
a1 , a2 , , ak be the distinct points of 1 ((a1 )). Note that k r and so there is an i for
which the numbers fi (a1 ), , fi (ak ) are all distinct. These are all roots of the polynomial
pi ((a1 ), x). Then for any w W0 , sufficiently near (a1 ), there must be a root of pi (w, x)
near fi (a1 ). But by the construction of pi this root must be a value assumed by fi on
1 (w) V1 . It follows that for any sequence {ws } W0 , converging to (a1 ) there is a
sequence {zs } V1 with (zs ) = ws and with fi (zs ) converging to fi (a1 ). Since is a
proper map, the sequence {zs } must have a limit point z V1 . Necessarily fi (z) = fi (a1 )
and, hence, z = a1 since fi separates the points a1 , , ak . Thus, a1 V1 and V1 = V .
This completes the proof.
The branch locus of a finite branched holomorphic cover is the set on which the branching order is at least 2.
5.16 Proposition. If : V W is a finite branched holomorphic cover over a domain
W in Cm and k is any positive integer, then the subset of V on which has branching
order at least k is a closed subvariety of V . In particular, the branch locus of is a closed
subvariety of V . Furthermore, the image under of the branch locus is a closed subvariety
of W .
Proof. With V0 and W0 as above and V1 = V0 we choose linear functionals f1 , , fq and
polynomials p1 , , pq as in the proof of the previous lemma. Then for any w W the
roots of the polynomial pj are exactly the values that fj assumes on 1 (w). A root is
(s)
a multiple root of pj of multiplicity k if and only if the polymomials pj vanish there for
s k. It follows that 1 (w) is a point of branching order at least k if and only if the
(s)
functions pj ((z), fj (z)) vanish at for s k and for all j. This proves the first part of
the Proposition. The second part follow from the fact that the set where the branching
order is at least two is the set where all the pj have multiple roots and this is the set
where the discriminants of all the polynomials pj vanish. The discriminants are functions
of w W only and the result follows.
5.17 Theorem. Let P n H be a prime ideal which is strictly regular in the variables
zm+1 , . . . , zn . Then the projection Cn = Cm Cnm Cm exhibits loc P as the germ of

NOTES ON SEVERAL COMPLEX VARIABLES

41

a finite branched holomorphic cover of pure order r over Cm , where r is the order of the
defining polynomial for zm+1 .
Proof. We let the polynomials pm+1 , . . . , pn , qm+2 , . . . , qn be as in Lemma 5.9. Choose a
polydisc centered at 0 on which each of these germs has a representative and on which
each element of a generating set for P (containing the p0 s and q 0 s) also has a representative.
We let = 0 00 be the usual decomposition induced by representing Cn as Cm Cnm .
We replace each pj and each qk by its representative on . We let V be the locus of common
zeroes of the set of representatives on of our generating set for P . Then the germ of
V at 0 is loc P . We let W be the zero set in Cm+1 of pm+1 (zm+1 ). We also let D
be the zero set in of the discriminant d of pm+1 . Since d depends only on the first m
variables, we have D = D0 00 where D0 is the zero set in 0 of d. By Lemma 5.13 the
projection Cm+1 = Cm C Cm induces a holomorphic branched covering of pure order
r, 2 : W 0 . Also, if W0 = W (D W ) then W0 is the set on which this covering is
regular (locally biholomorphic).
Another finite branched holomorphic covering is obtained by applying Lemma 5.13 to
the set
V1 = {z : pm+1 (zm+1 ) = 0, . . . , pn (zn ) = 0}
and the map 1 : V1 0 induced by the canonical projection Cn = Cm Cnm Cm .
This is regular on the set V1 (E V1 ), where E = E 0 00 is the union of the zero sets of
the discriminants of pm+1 , , pn . We obtain the covering map that we are looking for
by restricting 1 to the subvariety V V1 . It remains to show that is a finite branched
holomorphic cover of pure order r.
It is clear that is a finite to one, proper, holomorphic map. The density of the
regular points and pure order r are not yet clear. We will exploit the fact that factors
as = 2 3 where 2 is the finite branched holomorphic cover introduced above and
3 : V W is induced by the canonical projection Cn Cm+1 . Note that 3 is also the
germ of a proper finite to one holomorphic map. We let V0 = V (E V ). We will show
that V0 is a dense open subset of V on which is locally biholomorphic. First note that,
by corollary 5.10, since D E, the germ of V0 at 0 is loc I 0 (E loc I 0 ). Thus, we may
choose small enough that (z 0 , zm+1 , . . . , zn ) is in V0 exactly when z 0
/ E 0 and
pm+1 (z 0 , zm+1 ) = 0
qj (z 0 , zm+1 , zj ) = d(z 0 )zj sj (z 0 , zm+1 ) = 0

and
for j = m + 2, . . . , n.

Since d(z 0 ) 6= 0 when z 0

/ E 0 , the map 3 : V0 W0 is a biholomorphic mapping
with inverse given by the equations zj = d(z 0 )1 sj (z 0 , zm+1 ), j = m + 2, . . . , n. Thus,
: V0 0 E 0 is a locally biholomorphic covering map of degree r and the proof will
be complete if we can show that the closure V0 of V0 in V1 is V .
Now 1 : V1 0 is a finite branched holomorphic cover which is regular on the set
V1 (E V1 ). Thus, 1 is a locally biholomorphic covering map from V1 (E V1 ) to
0 E 0 . Since it is also a locally biholomorphic covering map from V0 to 0 E 0 , it follows
that V0 must be a component of V1 E V1 Now by Lemma 5.15Sthe set V0 is a subvariety
of V1 . Therefore, V0 is also a subvariety of V . Clearly V = V0 (V D) and since V is
irreducible and not contained in D we must have V0 = V . This completes the proof.

42

J. L. TAYLOR

A thin subset W of a subvariety V is a set which is contained in a closed subvariety of

V with dense complement.
5.18 Corollary. For any subvariety V of a domain in Cn , the set of singular points is a
thin set.
Proof. This is a local result. At any point
S p decompose the germ of the variety V into
irreducible germs of subvarieties: V = Vi . Then, in a sufficiently small neighborhood
of p, the singular set of V will be the union of the intersection sets Vi Vj for i 6= j and
the singular sets of the Vi0 s themselves. If the neighborhood of p is chosen small enough,
then Theorem 5.17 implies that each Vi has a representative that is a finite branched
holomorphic cover of a polydisc in Cm for some m. If Vi0 is the regular part of Vi for
the covering map, then Vi0 is dense in Vi and the singular set of Vi is contained in the
complement of Vi0 in Vi which is a closed subvariety. Furthermore, Each set Vi Vj for
i 6= j is a proper closed subvariety of Vi which necessarily meets Vi0 in a proper closed
subvariety of Vi0 . However, Vi0 is locally biholomorphic to a polydisc in Cm and so the
complement of a closed proper subvariety is dense. Since Vi0 is a dense open subset of Vi ,
it follows that the complement of Vi Vj in Vi is also dense in Vi .
The above paragraph shows that the set of singular points of V which lie in Vi is a thin
subset of Vi . It also follos from the above that a thin subset of Vi is also a thin subset of
V . Since the union of finitely many thin sets is clearly thin, we conclude that the set of
all singular points of V is a thin set.

5.19 Theorem (Nullstellensatz). If I is an ideal of n H then id loc I = I.

Proof. To complete the proof we need only prove that for a prime ideal P we have id loc P
P . Suppose f id loc P and suppose coordinates have been chosen so that P is strictly
regular in zm+1 , . . . , zn . By Lemma 5.9 there is a polynomial r with coeficients in m H of
degree less than the degree of the minimal polynomial pm+1 of zm+1 such that d f (z)
r(z 0 , zm+1 ) P . But then r must vanish on V = loc P since f does. However, by
Theorem 5.17, the projection on the first m coordinates exhibits V as the germ of a finite
branched holomorphic cover of Cm of pure degree equal to the degree of pm+1 . If we
choose a sufficiently small polydisc , on which f , r, and pm+1 may all be replaced by
representatives, then this implies that r has as many distinct roots as the degree of pm+1
on an open dense subset of and, unless r = 0, this contradicts the fact that the degree
of r is less than the degree of pm+1 . Hence, d f P , which implies f P since P is prime
and d
/ P . This completes the proof.
The following three theorems represent very useful refinements of the information in
Theorem 5.17 and their proofs include some nice applications of the Nullstellensatz.
5.20 Theorem. If P is a prime ideal that is regular in the variables zm+1 , . . . , zn and
: loc P Ck is the germ of the holomorphic mapping induced by the projection Cn Ck
where m k n then the image of is a germ of a holomorphic subvariety of Ck and
is the germ of a finite branched holomorhic cover of its image.
Proof. Let V = loc P . The projections Cn Ck and Cn Cm induce germs : loc P
Ck and 0 : loc P Cm of holomorphic mappings. By Problem 3, 0 is the germ of a finite

NOTES ON SEVERAL COMPLEX VARIABLES

43

branched holomorphic cover of Cm . With the polynomials pj as in the proof of Theorem

5.17, consider the germ
W = {z Ck : pm+1 (zm+1 ) = = pk (zk ) = 0}
By Theorem 5.13 the projection Ck Cm induces the germ 00 : W Cm of a finite
branched holomorphic cover. Since pj (zj ) P it is clear that (V ) W so that 0 = 00 .
Now choose representatives so that 0 : V U and 00 : W U are finite branched
holomorphic covers of a connected open set U Cm . For a suitable proper subvariety
D U the sets V0 = 01 (U D) and W0 = 001 (U D) are sets on which 0 and 00
are locally biholomorphic covering maps. Since 0 = 00 , it is clear that is a finite,
proper holomorphic mapping and |V0 is a locally biholomorphic covering map. The image
(V0 ) is the union of some connected components of W0 and, hence, its closure in W is a
subvariety by Lemma 5.15. Since V0 is dense in V and is a proper continuous mapping,
it follows that the closure of (V0 ) in W is (V ). Thus, (V ) is a subvariety and is a
finite branched holomorphic cover. This completes the proof.
5.21 Theorem. Let I be an ideal of n H and set V = loc I. Then the following three
conditions on I are equivalent:
(i) For each j = m + 1, . . . , n there is an element fj j H I which is regular in zj ;
(ii) For each irreducible component Vi of V the projection Cn Cm induces the germ
i : Vi i (Vi ) of a finite branched holomorphic cover and i (Vi ) is the germ of a
holomorphic subvariety of Cn ;
(iii) If L is the germ of {z Cn : z1 = = zm = 0}, then L V = {0}.
Proof. We first show that (i) implies (ii). Suppose that I satisfies (i). Then we may
enlarge the sequence {fj } to the extent possible by sequentially choosing non-zero functions
fj j H I for j = m, m 1, . . . as long as j H I 6= 0 and by changing variables in the
first j coordinates at each stage, if necessary, to make fj regular in zj . The result will be
a sequence 0 6= fj j H I for j = k + 1, . . . , n and some k m
Srwith k H I = 0. This
means that I is regular in the variables zk+1 , . . . , zn . Now V = i=1 Vi where the Vi are
the irreducible components of V . If Pi = id Vi , then each Pi is a prime ideal containing
fj for j = k + 1, . . . , n. As above, after a change of variables in the first k coordinates,
we may assume that Pi is regular in some variables zki +1 , . . . , zn with ki k. Then by
the previous theorem the projection Cn Cm induces the germ i : Vi i (Vi ) of a
finite branched holomorphic cover of the germ of a subvariety i (Vi ). Thus, condition (ii)
is satisfied and we have proved that (i) implies (ii).
Now suppose condition (ii) is satisfied. Let V1 , . . . , Vr be the irreducible components
of V , as before. The projection Cn Cm induces the germ i : Vi i (Vi ) of a finite
branched holomorphic cover of the germ of a subvariety i (Vi ) of a neighborhood of zero.
If L = {z Cn : z1 = = zm = 0} then L Vi = 1 (0) which must be the germ of a
finite set containing 0 since i is finite to one. But the germ at 0 of a finite set containing
0 is 0. Thus, (ii) implies (iii).
Now suppose I satisfies condition (iii) for some m < n. This means that the ideal J
generated by I and
at 0 it belongs to
z1 , . . . , zm satisfies loc J = {0}. Since zn vanishes

id loc J which is J by the Nullstellensatz. This implies that zn J for some . In other

44

J. L. TAYLOR

words, zn = fn + z1 g1 + + zm gm for some fn I and g1 , . . . , gm n H. Thus we have

produced an fn I that is regular in the variable zn . The rest of the proof amounts to
iterating this procedure as long as we can. In order to carry out the next step, we must
prove that if I n H satisfies (iii) for some m < n then the ideal I n1 H satisfies (iii)
in Cn1 for the same m. Since we have just showed that I contains an element regular
in zn and we know that (i) implies (ii) we conclude that the projection Cn Cn1 maps
V to the germ Vn1 of a holomorphic subvariety of Cn1 . Condition (iii) for I implies
that L Vn1 = 0. If we set In1 = I n1 H then clearly In1 id Vn1 . But if
f id Vn1 then when f is considered as an element of n H, constant in zn , it vanishes on
V and, by the Nullstellensatz,
some power of it belongs to I and, hence, to In1 . It follows

that In1 id Vn1 I n1 and from this that loc In1 = Vn1 . Thus, In1 satisfies
condition (iii) as desired. Either m = n 1 or we may now conclude as above that there
is fm1 In1 = n1 H I which is regular in zn1 . Clearly we can repeat this procedure
n m times to achieve condition (i). This completes the proof.
We say that a germ : V W of a holomorphic map between two germs of varieties
is finite if 1 (0) = (0).
5.22 Theorem. A germ : V W of a holomorphic mapping between two germs of
holomorphic varieties is finite if and only if for each irreducible component Vi of V , the
image (Vi ) is the germ of a holomorphic subvariety of W and : Vi (Vi ) is a finite
branched holomorphic cover.
Proof. Suppose is finite. We may represent V and W by subvarieties of neighborhoods
in Cn and Cm (m n) in such a way that the projection Cn Cm induces , by Problem
8. Then 1 (0) = L V where L is the germ of {z Cn : z1 = = zm = 0}. Since
is finite we have L V = 0 . It then follows from the previous theorem that for each
irreducible component Vi of V , : Vi (Vi ) is the germ of a finite branched holomorphic
cover of the germ (Vi ) of a subvariety. The reverse implication is obvious and so the proof
is complete.

5. Problems
1. Assuming the Nullstellensatz for C[z1 , . . . , zn ], prove it for n O.
2. Prove that if U Cn is a connected open set and D U is a closed subvariety, then
U D is also connected.
3. Prove that if P n H is a prime ideal which is just regular in the variables zm+1 , . . . , zn ,
then the projection Cn = Cm Cnm Cm still exhibits loc P as a finite branched
holomorphic cover of pure order r over Cm for some r.
4. In a finite branched holomorphic cover : V W prove that each point where the
branching order is one is a regular point of the variety V - that is, a point which has a
neighborhood in V biholomorphic to a polydisc in Cm for some m.
5. Prove that if V is an irreducible germ of a variety at 0 Cn then V has a representative
in some neighborhood of 0 for which the set of regular points is connected.
6. Prove the Nullstellensatz for the local ring V H of the germ of a holomorphic variety.

NOTES ON SEVERAL COMPLEX VARIABLES

45

7. Use Theorem 5.22 to prove that if U is an open subset of Cn and : U Cn is a one

to one holomorphic map then (U ) is open.
8. Prove that if : V W is the germ of a holomorphic map between two varieties, then
we may represent V and W by subvarieties of neighborhoods in Cn and Cm (m n) in
such a way that the projection Cn Cm induces .
9. Use Proposition 5.12(a) to prove that if : V W is a finite branched holomorphic
cover and V 0 is a subset of V such that : V 0 W is also a finite branched holomorphic
cover, then V00 is both open and closed in V0 . Here, V0 = pi1 (W0 ) and V00 = V 0
1 (W0 ) with W0 as in Def. 5.11.

46

J. L. TAYLOR

6. Dimension
We continue with our study of the local properties of regular and holomorphic functions
and algebraic and holomorphic varieties. In this section we will be concerned with a germ
V of an algebraic or holomorphic variety and with the local ring V O or V H. In each case
there are three notions of dimension of the local ring: a topological dimension, a geometric
dimension and a tangential dimension. We show that the first two agree and that they
agree with the third if and only if the variety is regular at the point in question. We shall
use this last result to prove that the singular locus of a variety is a subvariety. Here the
regular locus of a holomorphic variety is the set at which it is locally biholomorphic to
a polydisc in complex Euclidean space and the singular locus is the complement of the
regular locus. The regular and singular locus of an algebraic variety have not yet been
defined and, in fact, the usual definition is that the regular locus is the set where the
second and third of the three dimensions, referred to above, are equal.
We initially take as our definition of the dimension of a holomorphic variety the topological definition:
6.1 Definition. If V is a holomorphic variety then the dimension of V is the dimension
of the complex manifold that is the regular locus of V .
Note that if V then the dimensions of smaller and smaller neighborhoods of in V
eventually stabilize and so it makes sense to talk about the dimension of the germ of V at
. Also note that if V is not irreducible then its regular locus may decompose into germs
of manifolds of different dimensions. In this case, by dimension we mean the maximal
dimension that occurs. Thus, if V = V1 V2 Vk is the decomposition of a germ of
a variety into its irreducible components, then the dimension of V is the maximum of the
dimensions of the Vj .
6.2 Lemma. If V is the germ of a variety in Cn and I = id V is regular in the variables
zm+1 , . . . , zn then dim V = m.
Proof. If V is irreducible then this is an obvious corollary of Theorem 5.17. If V is not
irreducible, let V = V1 . . . Vr be a decomposition of V into irreducibles. By Theorem
5.21, the projection : Cn Cm determines a finite branched holomorphic cover of each
Vi onto a germ of a subvariety of Cm . The dimension of Vi is the same as the dimension of
(Vi ) and this is less than or equal to m for each i. To complete the proof, we must show
that one of these dimensions is equal to m. If not, then, for each i, (Vi ) is a germ of a
proper subvariety of Cm and, hence, there
is a non-zero element gi m H which vanishes
on (Vi ). Then g = g1 g2 . . . gr id V = I and so some power of g is a non-zero element
of I m H which violates the assumption that the ideal I is regular in zm+1 , . . . , zn . This
completes the proof.
6.3 Theorem. If V and W are germs of holomorphic varieties with V W then dim V
dim W and the two are equal exactly when V and W have a common irreducible component
of dimension dim W .
Proof. Since V W , each irreducible component of V is contained in some irreducible
component of W . Thus, the theorem can be reduced to the case where V and W are

NOTES ON SEVERAL COMPLEX VARIABLES

47

irreducible. In this case we can assume that W is a germ of a subvariety of Cn and id W is

regular in variables zm+1 , . . . , zn where m = dim W . This means that there are functions
fj j H id W regular in zj for j = m + 1, . . . , n. Note that these functions also belong
to id V since V W . Now if m H id V = 0 then id V is also regular in the variables
zm+1 , . . . , zn . If not, then we can choose a non-zero fm m H id V and, after a change
of coordinates involving only z1 , . . . , zm , assume that it is regular in zm . Continuing in
this way, we may choose coordinates so that id V is regular in the coordinates zk+1 , . . . , zn
for some k m which by the previous lemma must be dim V . This makes it obvious
that dim V dim W and that if dim V = dim W then both V and W have germs that
are exhibited as finite holomorphic branched covers of the same polydisc by the projection
Cn Cm . If the points that lie above a proper subvariety D of this polydisc are removed
then the remaining regular parts V0 and W0 of V and W are unbranched holomorphic
covers of the same set in Cm . Both are connected since V and W are irreducible. It
follows that V0 and W0 must coincide (Problem 5.9), from which it follows that their
closures V and W coincide.
6.4 Theorem. If V is the germ of a holomorphic subvariety at the origin in Cn then
dim V is the smallest integer k so that there is a linear subspace L of dimension n k such
that L V = 0.
Proof. If there is a n k-dimensional linear subspace L with V L = 0 then choose
coordinates for which L = {z Cn : z1 = = zk = 0}. By Theorem 5.21 there are
elements fj j H id V for j = k + 1, . . . n such that fj is regular in zj . But then after
some linear change of variables in the first k coordinates we may assume that id V is
regular in zm+1 , . . . , zn for some m k. By lemma 6.2 it follows that m is necessarily
dim V . Hence, dim V k.
On the other hand, if k = dim V let V = i Vi be the decomposition of V into irreducibles
and. After a change of variables, it may be assumed that for each i, id Vi is regular in the
variables zni +1 , . . . , zn for some integer ni . Necessarily, ni = dim Vi by lemma 6.2 again. It
follows from Theorem 5.21 that Li Vi = 0 where Li = {z Cn : z1 = = zni = 0}. Since
k = maxi ni it follows that L Vi = 0 for each i where L = {z Cn : z1 = = zk = 0}.
Therefore L is an n k dimensional subspace with L V = 0 and this completes the proof.
A germ of a variety has pure dimension r if each of its irreducible components has
dimension r.
6.5 Theorem. A germ of a variety V in Cn has pure dimension n 1 if and only if id V
is a principal ideal.
Proof. It is easy to see that this is true in general if it is true for irreducible varieties
(Problem 6.1). Thus, let V be an irreducible variety such that the prime ideal P = id V is
generated by a single element f . Then f must be irreducible. After a change of variables,
we may also assume that f is regular in zn and, after multiplying by a unit, we may
assume it is a Weierstrass polynomial in n1 H[zn ]. We also have that P n1 H = 0 since
otherwise there would be an element g = f h n1 H. But for each z 0 Cn1 near the
origin the polynomial f has at least one root and this implies that g(z 0 ) = 0 for all z 0 near
the origin, i. e. that g is the zero germ. Now we have that P is regular in the variable zn
and, by Lemma 6.2, V = loc P has dimension n 1.

48

J. L. TAYLOR

Conversely, suppose that V is an irreducible germ of a variety of dimension n 1. Then

we may choose coordinates so that P = id V is strictly regular in zm+1 , . . . , zn for some
m < n. However, by Lemma 6.2, m is the dimension of V and, thus, m = n 1. In other
words, P is regular in zn . Let p be the minimal polynomial for zn mod P . Recall that by
5.17, the projection Cn Cn1 induces a finite branched holomorphic cover : V W
where W is some a connected neighborhood of 0 in Cn1 . Furthermore, for z 0 in a dense
open subset W0 of W the points 1 (z 0 ) are exactly the roots of p(z 0 , zn ). Now if f P
then f = pg +r for some g n H and some polynomial r n1 H[zn ] of degree less than the
degree of p. However, since f and p belong to P so does r and, therefore for each z 0 W0
it vanishes at the points of 1 (z 0 ), but there are two many of these and so it vanishes on
W0 and, hence, on W . Thus we have proved that p generates P . This completes the proof.
6.6 Lemma. Suppose X is a compact topological space, D is a domain in Cn containing
0 and f1 , . . . , fk are continuous functions on D X which are holomorphic in z D for
each fixed x X. Let V (x) be the subvariety of D defined by
V (x) = {z D : f1 (z, x) = = fk (z, x)} = 0
Then the dimension of the germ of V (x) at 0 is an upper semicontinuous function of x.
Proof. Let dim V (x) denote the dimension of the germ of V (x) at 0. Since dimension is
integral valued, to show that dim V (x) is upper semicontinuous we must show that for
each x0 X there is a neighborhood of x0 in which dim V (x) dim V (x0 ). Let x0 be such
a point and let m = dim V (x0 ). Then by Theorem 6.4 there is a linear subspace of Cn
of dimension n m such that L V (x0 ) = 0. Choose a neighborhood W of 0 which has
compact P
closure in D. The boundary of W does not meet L V (x0 ). In other words, the
function
|fi (z, x)| does not vanish on the compact set (L W ) {x0 }. It follows that
there is a neighborhood U of x0 such that this function does not vanish on (L W ) U .
Then for x U we have that V (x) L does not meet W and so V (x) L W is either
empty or is a compact subvariety K of W . The maximum modulus theorem, applied to the
restriction to K of each coordinate function in Cn , implies that K is a finite set (Problem
6.2). Thus, the germ of V (x) L at 0 is either empty or 0. In the first case, the germ
of V (x) is empty (dimension -1) and in the second case dim V (x) m = dim V (x0 ) by
Theorem 6.4 again. This completes the proof.
6.7 Theorem. Let V be the germ of a holomorphic variety of pure dimension m and let
f V H be a non unit and a non zero divisor. Then the locus of the ideal generated by f
in V H is a subvariety of V of pure dimension m 1.
Proof. Choose representatives in a neighborhood U of 0 for V , f and holomorphic functions
gi , i = 1, . . . , k, such that
V = {z U : g1 (z) = = gk (z) = 0}
Set W = {z V : f (z) = 0}. Since f is not a unit, W is not empty. Since f is not a zero
divisor, it does not vanish identically on any irreducible component of V and so V and

NOTES ON SEVERAL COMPLEX VARIABLES

49

W have no irreducible components in common. Let be a polydisc centered at 0 small

enough that + U and consider the family of subvarieties V () of given by
V () = (V ) = {z : g1 (z + ) = = gk (z + ) = 0}
for and the family W () given by
W () = (W ) = {z : f (z + ) = g1 (z + ) = = gk (z + ) = 0}
for . Since each V () is the intersection with of a translate of V , if is chosen
small enough, each V () will be a variety of pure dimension m and for an open dense set
of , 0 will be a regular point of V (). For these values of , some nieghborhood of
0 in V () will be biholomorphic to a polydisc. It follows from Theorem 6.5 that, at such
points, W () has pure dimension m 1. Then Lemma 6.6 implies that W = W (0) has
dimension at least m1. However, the only possiblility other than m1 for this dimension
is m and, in this case, V and W would have to have a common irreducible component by
Theorem 6.3. We have already pointed out that this cannot happen since f is not a zero
divisor. This completes the proof.
6.8 Definition. The Krull dimension of a local ring A is the largest integer d for which
there exists a strict chain P0 P1 Pd of prime ideals of A that is, a chain in
which all the containments are proper (A itself is not considered a prime ideal).
This is the second notion of dimension we referred to earlier - the geometric one.
6.9 Theorem. If V is the germ of a holomorphic variety and
the dimension of V is equal to the Krull dimension of V H.

H is its local ring, then

Proof. Note that it follows from the previous theorem that if W 0 W 00 are germs of
irreducible subvarieties of V and dim W 00 dim W 0 2 then there is another irreducible
subvariety W 000 which is properly contained in W 00 and properly contains W 0 . Indeed, there
must be a germ f V H which belongs to id W 0 but not to id W 00 . The zero locus in W 00
of such a function contains W 0 and is a subvariety of W 00 of pure dimension dim W 00 1.
Thus, some irreducible component of this variety properly contains W 0 and is properly
contained in W 00 and will do as our W 000 .
It is clear from the above paragraph that if 0 = Wd Wd1 W0 is a maximal chain of irreducible subvarieties of W , then successive varieties in the chain differ in
dimension by exactly one. It follows that d = dim W0 for such a chain. Clearly d will be
largest possible when W0 is an irreducible subvariety of V of largest dimension. That is,
when W0 is an irreducible component of V with the same dimension as V . This completes
the proof.
We now turn to the third notion of dimension tangential dimension.
6.10 Definition. If V is a germ at 0 of a algebraic or holomorphic variety then a tangent
vector to V is a derivation at 0 that is, linear map t : V H C (t : V O C) such that
t(f g) = f (0)t(g) + g(0)t(f ). The vector space of all tangent vectors is called the tangent
space to V and is denoted T (V ). Its dimension is the tangential dimension of V and is
denoted tdim V .

50

J. L. TAYLOR

6.11 Theorem. The vector space T (V ) is naturally isomorphic to the dual of M/M 2
where M is the maximal ideal of V H (V O).
Proof. If t T (V ) then t(1) = 2t(1) and so t kills constants and is, thus, determined by its
restriction to M . However, if t is any linear functional on V H (V O) which kills constants
and f = 1 + f1 , g = 1 + g1 with f1 , g1 M then
t(f g) = t(f1 ) + t(g1 ) + t(f1 g1 ) = g(0)t(f ) + f (0)t(g) + t(f1 g1 )
from which we conclude that t is a tangent vector if and only if t vanishes on M 2 . Thus,
restriction to M defines an isomorphism between T (V ) and the dual of M/M 2 . This
completes the proof.
It is clear that if Cn is considered either a holomorphic or an algebraic variety then T (Cn )
is naturally isomorphic to Cn where a given point t = (t1 , . . . , tn ) Cn is associated to
P f
the derivation on n H or n O defined by t(f ) =
ti
(0).
zi
If F : V W is a germ of a holomorphic map between varieties, then F induces an
algebra homomorphism F : W H V H (F : W O V O) which, in turn, induces a linear
map dF : T (V ) T (W ) by dF (t)(f ) = t(F (f )).
6.12 Theorem. If V is a germ at 0 of a subvariety of Cn and F : V Cn is the
inclusion, then dF : T (V ) T (Cn ) is injective and its image is {t T (Cn ) : t(g) =
0 whenever g id V }.
Proof. We have that
dF (t)(g) = t(F (g)) = t(g F ) = t(g|V )
and so dF (t) = 0 if and only if t = 0. Furthermore, a derivation on n H of the form dF (t)
clearly vanishes on id V = {g n H : g|V = 0}. Conversely, if s is a derivation on n H which
vanishes on id V , then s determines a well defined linear functional t on V H = n H/ id V
such that t(g|V ) = s(g). Since s is a derivation, t clearly is as well. This completes the
proof for n H. The proof is the same for n O.
In the above theorem, suppose that id V is generated by g1 , . . . , gm . Then a derivation
in T (Cn ) vanishes on id V if and only if it vanishes at each gi . Thus, if we identify T (V )
with its image under dF then
T (V ) = {t Cn :

X gj
ti
(0) = 0
z
i
i

for j = 1, . . . , m}

From this it is immediate that:

6.13 Corollary. If V is a germ of holomorphic (algebraic) subvariety of Cn and g1 , . . . , gm
generate id V then
tdim V = n rank JG (0)
where G : Cn Cm is the holomorphic (algebraic) map with the gj as coordinate functions
and JG is its Jacobian matrix.

NOTES ON SEVERAL COMPLEX VARIABLES

51

6.14 Theorem. A germ V of a holomorphic variety can be represented as a subvariety

of Cn if and only if n tdim V .
Proof. Certainly tdim V n if V is represented as a germ of a subvariety of Cn . On the
other hand if V is a holomorphic subvariety of Cm and n = tdim V then rank JG (0) = mn
where g1 , . . . , gk is a set of generators for id V . It follows from the implicit function theorem
that we may choose holomorphic coordinates for a neighborhood of zero so that the last
m n coordinate functions belong to the set {g1 , . . . , gk }. In other words, we may assume
that zn+1 , . . . , zm id V . This means that V is contained in an n dimensional subspace of
Cm as required.
A germ V of a subvariety of Cn is said to be neatly embedded if n = tdim V . The above
theorem says that every germ of a variety can be neatly embedded.
The following is a form of the implicit function theorem that holds for varieties. It will
play a key role in chapter 15.
6.15 Theorem. If f : V W is a holomorphic mapping between germs of holomorphic
varieties and if df : T (V ) T (W ) is injective, then f is a biholomorphic mapping between
V and a germ of a holomorphic subvariety of W .
Proof. We may assume that V and W are neatly imbedded germs of subvarieties at the
origin in Cm and Cn , respectively. Then f may be regarded as a germ of a holomorphic map
from V into Cn . Furthermore, f extends to a germ of a holomorphic mapping g : Cm Cn
since each of its coordinate functions has a holomorphic extension to a neighborhood of 0
in Cm . Since V and W are neatly imbedded, we have that T (V ) = Cm and T (W ) = Cn
with the natural identifications given by Theorem 6.12. It follows that the linear maps
df : T (V ) T (W ) and dg : Cm Cn agree after this identification and, hence that dg is
injective. The matrix representing dg is the Jacobian matrix Jg (0) and so the usual inverse
mapping theorem implies that g is the germ of a biholomorphic map of a neighborhood of
0 in Cm onto a germ of an m-dimensional submanifold of Cn . Its restriction to V is then
a biholomorphic map of V onto a germ of a subvariety of Cn , as required.
The next theorem is useful for identifying cases where the previous theorem applies.
6.16 Theorem. If f : V Cn is a holomorphic mapping between germs of holomorphic
varieties and if the coordinate functions of f generate the maximal ideal of V H, then
df : T (V ) T (Cn ) is injective.
Proof. This follows from Corollary 6.13 in the following fashion: Assume V is neatly
embedded as a germ at the origin of a subvariety of Cm . Extend f to a map g defined in a
neighborhood in Cm , as in the previous theorem. Let {k1 , , kp } be a set of germs in m H
which generates id V and let k : Cm Cp be the map with these as coordinate functions.
If h = g k : Cm Cn+p , then the set of coordinate functions of h is the union of the
set of coordinate functions of g and those of k and so clearly generates the maximal ideal
of m H. Since the variety (0) has tangential dimension 0, it follows from Corollary 6.13
that rank Jh = m and, hence, that dh is injective. However, k has coordinate functions
which generate id V . Since V is neatly embedded, tdim V = m and so it follows, also by
Corollary 6.13, that rank Jk (0) = 0. Thus, dk vanishes at 0 and we conclude that dg is
injective. Since df and dg agree after the appropriate identification, the proof is complete.

52

J. L. TAYLOR

6.17 Theorem. If V is the germ of a holomorphic variety then dim V tdim V and V
is regular if and only if tdim V = dim V .
Proof. That dim V tdim V clearly follows from Theorem 6.14. A germ of a variety is
regular if and only if it is biholomorphic to the germ of a neighborhood of zero in Cn where
n = dim V . By Theorem 6.14 again, this is equivalent to tdim V = n.
6.18 Theorem. If V is a holomorphic subvariety of a domain in Cn then the singular
locus of V is a holomorphic subvariety of V .
Proof. We dont yet have all of the machinery necessary to prove this. but it fits naturally
into this circle of ideas. Thus, we will present a proof that assumes a result that will be
proved later (Theorem 12.8).
The result we are after is a local result and so we may assume V is a subvariety of
some polydisc and is the union of finitely many irrreducible components Vj . Then the
singular locus of V is the union of the singular loci of the Vj and the sets of intersection
Vi Vj for i 6= j. Thus, it is enough to prove the theorem in the case where V is irreducible.
However, for an irreducible subvariety V of a connected set, the dimension of the germ Vz
of the variety at z V is a constant m. On the other hand, tdim Vz is n rank JG (z)
where JG is the Jacobian of a holomorphic map G whose coordinate functions form a set
of generators of id Vz . By Theorem 12.8, we may choose small enough that there exists
a G such that the coordinate functions of G generate id V at every point z V . The set
where rank JG (z) < k is a subvariety for each k and so the set where tdim Vz > m = dim V
is a subvariety. By the previous theorem, this is the singular locus of V . This completes
the proof.
The dimension of the germ V at of an algebraic variety V is defined to be the Krull
dimension of the corresponding local ring V O . A point of an algebraic variety V is said
to be a regular point if tdim V = dim V . The singular locus of V is the set of singular
points.
Of course, to each algebraic variety there is associated a holomorphic variety V which
is the same point set but with a different topology and a different local ring V H associated
to a point. it makes sense to ask whether or not the germ af an algebraic variety at a
point has the same dimension (or tangential dimension), as the germ of the corresponding
holomorphic variety. It also makes sense to ask if the singular locus of an algebraic variety
is a proper subvariety and whether or not it agrees with the singular locus of V . The rest
of this section is devoted to showing that the answer to all these questions is yes. We will
need the following two lemmas from commutative algebra which we will not prove (see
Atiyah-Macdonald chapter 11):
6.19 Lemma. Let B A be integral domains with B integrally closed and A integral
over B. Then for each prime ideal M of A, M is maximal in A if and only if N = M B is
maximal in B and, in this case, the local rings AM and BN have the same Krull dimension.
6.20 Lemma. The Krull dimension of n O is n.
6.21 Theorem. If V is an irreducible algebraic subvariety of Cn then dim V = dim V
at any point V , where V is the holomorphic subvariety determined by V .

NOTES ON SEVERAL COMPLEX VARIABLES

53

Proof. By proceeding as in section 5, we may choose a coordinate system for Cn such

that id V is regular in the variables zm+1 , . . . , zn , that is, id V C[z1 , . . . , zm ] = 0 and for
j = m + 1, . . . , n there is a pj id V C[z1 , . . . , zj ] which is regular in zj . This implies
that OV = C[z1 , . . . , zn ]/ id V is an integral ring extension of C[z1 , . . . , zm ]. Thus, Lemma
6.19 applies with A = OV and B = C[z1 , . . . , zm ]. It follows from Lemma 6.19 that for
any point V , dim V O = dim m O where is the image of under the projection
Cn Cm . However, dim m O = m by Lemma 6.20. Since the projection Cn Cm
induces a finite branched holomorphic cover V Cm , m is also the dimension of the
holomorphic variety V .
6.22 Theorem. If V is an irreducible algebraic subvariety of Cn and V the corresponding
holomorphic subvariety then tdim V = tdim V at each V . Thus, the singular locus
of V is the same point set as the singular locus of V . Furthermore, the singular locus of
V is a proper algebraic subvariety of V .
Proof. We will prove in the next section that a generating set for the ideal id V n O
is also a generating set for the ideal id V n H (Theorem 7.13). That the tangential
dimensions of the germs V and V are the same follows from Theorem 6.13.
As in the holomorphic case, that the singular locus of V is an algebraic subvariety
follows from a Jacobian argument and the fact that we know from the preceding theorem
that dim V is constant in V for an irreducible algebraic variety. The proof is identical
to that of Theorem 6.18 except that the role played by Theorem 12.8 in Theorem 6.18 is
played by Theorem 10.19 in the algebraic case. The singular locus is a proper subvariety
because it is the same point set as the singular locus of V which is proper.

6. Problems
1. Let V be a germ of a subvariety of Cn . Prove that if id Vi is a principal ideal in n H for
every irreducible component Vi of V , then id V is also a principal ideal.
2. Let V be a subvariety of a domain in Cn . Prove that if the modulus of a holomorphic
function f on V has a local maximum at z V then f is constant on the irreducible
component of V containing z. Use this to prove that a compact subvariety of an open
set in Cn must be finite.
3. Prove that if V is a germ of a variety then tdim V is the minimal number of generators
for the maximal ideal of V H.
4. Prove that if V is a germ of a holomorphic variety and P V H is a prime ideal, then
depth(P ) + height(P ) = dim V , where depth(P ) is the maximal length of a strict chain
of primes with P at the bottom and height(P ) is the maximal length of a strict chain
of primes with P at the top.
5. Prove the first part of Lemma 6.19: If B A are integral domains with A integral over
B and if M A is a prime ideal and N = M B then M is maximal if and only if N
is maximal.

54

J. L. TAYLOR

7. Completion of Local Rings

In this section we begin the study of the passage from an algebraic variety V to the
corresponding holomorphic variety. Locally, at a point of V , this amounts to studying
the relationship between V O and V H . Clearly V O is a subalgebra of V H . Our main
goal in this section is to prove that V H is faithfully flat over V O . We begin with a brief
discussion of this notion.
If A B is a pair consisting of a commutative algebra and a subalgebra containing the
identity, then there is a functor X XB from the category of A-modules to the category
of B-modules defined by
XB = B A X
The situation is particularly nice when this functor is both exact and faithful (X 6= 0
implies XB 6= 0).
7.1 Definition. Let B be a commutative ring and A B a subring containing the
identity. Then B is said to be faithfully flat over A if B is a flat A-module and for each
non-zero A-module X the the B-module XB = B A X is non-zero.
7.2 Lemma. The following are equivalent:
(i)
(ii)
(iii)
(iv)

B is faithfully flat over A;

B is flat over A and x 1 x : X XB is injective for every A-module X;
B is flat over A and IB A = I for every ideal I of A;
B/A is a flat A-module.

Proof. If B is faithfully flat over A and X is an A-module, let K be the kernel of x

1 x : X XB . Then the flatness of B implies that B A K B A X is injective. On
the other hand, the composition of the maps
K B A K B A X
is the zero map. This implies that k 1 k : K B A K is the zero map; but the
image of this map generates B A K over B and, hence, B A K = 0. It follows from (i)
that K = 0. This proves that (i) implies (ii). That (ii) implies (i) is obvious.
Note that I IB A for any ideal I A. Also, if B is flat over A and I A is an
ideal, then applying B A () to the short exact sequence
0 I A A/I 0
and noting that B A I = BI, yields the isomorphism B A A/I ' B/BI. Thus, the
condition IB A = I is equivalent to the injectivity of the map A/I B A A/I. This
proves that (ii) implies (iii). It also proves (iii) implies (ii) since for any A-module X and
any x X the submodule Ax X has the form A/I for an ideal I.
Now for an A-module X the short exact sequence
0 A B B/A 0

NOTES ON SEVERAL COMPLEX VARIABLES

55

leads to a long exact sequence

A
A
torA
1 (A, X) tor1 (B, X) tor1 (B/A, X) A A X B A X

but since torA

1 (A, X) = 0 and A A X = X. This becomes
A
0 torA
1 (B, X) tor1 (B/A, X) X B A X
A
This makes it clear that torA
1 (B/A, X) = 0 for all A-modules X if and only if tor1 (B, X) =
0 and X B A X is injective for all A-modules X. This proves that (ii) is equivalent
to (iv) since an A-module is flat if and only if it has vanishing torA
1 with respect to every
A-module. This completes the proof.

7.3 Theorem. If A B C are algebras with C faithfully flat over A and C faithfully
flat over B then B is faithfully flat over A.
Proof. Suppose X Y is an injective morphism of A-modules and let N be the kernel of
B A X B A X. Then since C is B-flat we have an exact sequence
0 C B N C B (B A X) C B (B A Y )
but by the associativity of tensor product we have that C B (B A X) = C A X and
C B (B A Y ) = C A Y . But since C is A-flat we have that C A X C A Y is
injective and, hence, C A N = 0. But this implies that N = 0 since C is faithfully flat
over A. Thus, we have proved that B is A-flat.
Now suppose that X is an A-module and consider the maps X B A X C A X.
Since the composition is an injection due to the fact that C is faithfully flat over A it
follows that the first map is an injection as well and, hence, that B is faithfully flat over
A. This completes the proof.
Our strategy for proving that V H is faithfully flat over V O will be to inject both of them
into a third algebra the M -adic completion of V O with respect to its maximal ideal M
and to show that this algebra is faithfully flat over both V O and V H. Then the previous
theorem will give us the desired result. To this end, we need to study the completion A of
a local ring with respect to its maximal ideal.
To begin with we need a lemma about graded Noetherian rings. A graded ring is is a ring
A =
n=0 An which is the direct sum of subspaces An in such a way that An Am An+m
for all n, m. The elements of An are said to be homogeneous of degree n. A graded
Noetherian ring is a graded ring which is also Noetherian as a ring.
7.4 Lemma. Let A be a graded Noetherian ring. Then
(i) A0 is a Noetherian ring.
(ii) A is a finitely generated A0 -algebra.
Proof. (i) Put A+ =
n=1 An . Then A+ is an ideal in A and A0 = A/A+ .
(ii) A+ is finitely generated. Let x1 , x2 , . . . , xs be a set of generators of A+ . Without
loss of generality we may assume these generators are homogeneous since, otherwise, we

56

J. L. TAYLOR

may decompose them into homogeneous components. Let di = deg xi , 1 i s. Let B

be the A0 -subalgebra generated by x1 , . . . , xs . We claim that An B, n Z+ . Clearly,
A0 P
B. Assume that n > 0, Am B for m < n and y An . Then y A+ and therefore
s
y = i=1 yi xi where yi Andi . It follows that the induction assumption applies to yi ,
1 i s. This implies that y B which completes the proof.
We use the above result to prove the key ingredient in the study of m-adic completions
of local rings:
7.5 Theorem (Artin, Rees). Let A be a Noetherian local ring with maximal ideal M
and let Y be a finitely generated A-module and X a submodule of Y . Then there exists
m0 Z+ such that
M p+m0 Y X = M p (M m0 Y X)
for all p Z+ .
n

Proof. Put A =
n=0 M . Then A has a natural structure of a graded ring. Let
(a1 , . . . , as ) be a set of elements in M with the property the the images of the ai in M/M 2
generate it. Then for each n, M n is generated as an A-module by the monomials of degree
n in the ai . Thus, we have a natural surjective morphism A[x1 , . . . , xs ]
A determined
by xi1 xin ai1 ain M n which implies that A is a graded Noetherian ring. Let
n

Y =
n=0 M Y . Then Y is a graded A -module. It is clearly generated by Y0 = Y as

an A -module. Since Y is a finitely generated A-module, we conclude that Y is a finitely

generated A -module.
n

In addition, put X =
n=0 (X M Y ) Y . Then

M p (X M n Y ) M p X M n+p Y X M n+p Y
implies that X is an A -submodule of Y . Since A is a Noetherian ring, X is finitely
n

0
generated. There exists m0 Z+ such that m
n=0 (X M Y ) generates X . Then for any
p Z+ ,
X M

p+m0

Y =

m0
X

M p+m0 s (X M s Y ) M p (X M m0 Y ) X M p+m0 Y.

s=0

Therefore, the inclusions are equalities and the proof is complete.

Now let A be a Noetherian local ring with maximal ideal M . Given any A-module X,
we define a topology on X by declaring a neighborhood base for the topology at x X to
consist of the sets x + M n X for n Z+ . This is a uniform topology for the additive group
of X relative to this topology. The
structure of X and so we may define a completion X
completion consists of equivalence classes of Cauchy sequences from X where a sequence
{xk } is Cauchy if for each n Z+ there is a K Z+ such that xk1 xk2 M n X whenever
k1 , k2 > K. In other words, a sequence is Cauchy if and only if it is eventually constant
may also be described
mod M n X for each n Z+ . This description makes it clear that X
as the inverse limit
= lim X/M n X
X

NOTES ON SEVERAL COMPLEX VARIABLES

57

When X = A we obtain a completion A for A itself. It is easy to see that A is also a ring
is a module over A.
In fact, X X
is a
and for each module X over A the completion X

covariant functor from A-modules to A-modules.

This is an exact functor when restricted
to finitely generated modules as is shown below:
7.6 Theorem. Let A be a Noetherian local ring and let
0XY Z0
be an exact sequence of finitely generated A modules. Then the sequence
Y Z 0
0X
is also exact.
Proof. By Artin-Rees we have that there exists an m0 such that
M p+m0 X M p+m0 Y X M p X
This implies that the completion of X with respect to the topology determined by the
filtration {M p X} agrees with that determined by the filtration {X M p Y }. In other
= lim Xp . But we have for each p a short exact
words, if Xp = X/(X M p Y ) then X

sequence
0 Xp Yp Zp 0
where Yp = Y /M p Y and Zp = Z/M p Z. Now limits of inverse sequences preserve left exactness but do not always preserve right exactness. Right exactness is, however, preserved
in the case where the left hand sequence {Xp } is surjective in the sense that each map
Xp+1 Xp is surjective (Problem 7.1) as is true in our situation. It follows that
Y Z 0
0X
is exact, as required.
7.7 Theorem. If A is a Noetherian local ring and X is a finitely generated A-module,
is an isomorphism.
then A A X X
commutes with taking finite direct sums. Thus, since
Proof. It is clear that X X

A A A A is an isomorophism, we conclude that A A F F is an isomorphism

whenever F is a finitely generated free module. Since X is finitely generated, we can find
a short exact sequence
0KF X0
with F free and finitely generated. This yields a diagram
A A K A A F A A X 0

y
y
y
0

58

J. L. TAYLOR

in which the bottom row is exact by Theorem 7.6, the top row is right exact and the

middle vertical map is an isomorphism. A simple diagram chase shows that A A X X

is surjective. However, since A is Noetherian, we also have that K is a finitely generated
is also surjective. It then
A-module and, by what we just proved, the map A A K K

follows from another diagram chase that A A X X is injective. This completes the
proof.
7.8 Theorem. If A is a Noetherian local ring, then A is faithfully flat over A.
Proof. It is easy to see (Problem 7.2) that an A-module Y is flat if and only if whenever
X1 and X2 are finitely generated A-modules and X1 X2 is injective, then Y A X1
Y A X2 is also injective. Since Theorems 7.6 and 7.7 prove that A A () preserves
exactness of short exact sequences of finitely generated A-modules, we conclude that A is
flat as an A-module.
Now suppose that X is a finitely generated A-module. Then the kernel of the map
is E = n M n X. It follows from Artin-Rees applied to E X that M E = E.
X X
Then Nakayamas lemma implies that E = 0. Now by Theorem 7.7 we conclude that the
map X A A X is injective whenever X is finitely generated. But this clearly implies
that this map is injective in general and, hence, that A is faithfully flat over A.
7.9 Theorem. If A is a Noetherian local ring then
= AM
;
(i) the unique maximal ideal of A is M
n for all n Z+ ; and
(ii) M n = A M
M
n+1 is an isomorphism for all n.
(iii) A/M n A/
-adic topology.
(iv) A is complete in the M
Proof. Since M p (A/M ) = 0 for p Z+ we have that A/M is complete in the M -adic
topology. We apply the completion functor to the exact sequence
0 M A A/M 0
and use the fact that this functor is exact to conclude that we have an exact sequence
A A/M 0
0M
is a maximal ideal of A since A/M is a field. We also have that
This implies that M
= A A M = AM
, which implies that M
n = AM
n . Now since A is faithfully flat over
M
n A = M n . This
A, Lemma 7.2 implies that M n A A = M n and we conclude that M
proves (ii).
M
n+1 is surjective follows from the fact that a Cauchy sequence in
That A/M n A/
the M -adic topology is eventually constant modulo M n . That this map is injective follows
from (ii). This completes the proof of (iii).
-adic completion of A
Part (iv) follows immediately from (iii) which shows that the M

is A.
is the only maximal ideal of A.
To
To complete the proof of (i) we must show that M
do this, we need only show that 1 a is a unit in A for every a M . In fact, the inverse
-adic topology
of 1 a for a M is 1 + a + a2 + + an + . . . which converges in the M

of A.

NOTES ON SEVERAL COMPLEX VARIABLES

59

7.10 Theorem. If A is a Noetherian local ring then A is also a Noetherian local ring.
Proof. We have that A is a local ring from the previous theorem. Thus, we need only

P
show that A is Noetherian. The graded ring G(A) =
M n /M n+1 associated to A is a
n=0

finitely generated algebra over the field A/M and is therefore Noetherian by the Hilbert
is also Noetherian since it is isomorphic to G(A) by
basis theorem. It follows that G(A)
If we give I the filtration {M
n I} then G(I)
Theorem 7.9(iii). Suppose I is an ideal of A.
and, as such, it is finitely generated. Let {
embedds as an ideal of G(A)
ai ; i = 1, . . . , n}
be a set of homogeneous generators of G(I), set ri = deg(
ai ) and let ai I M ri be a
representative of a
i for each i. Let J be the ideal in A generated by a1 , . . . , an .
We will prove that J = I. Clearly G(I) = G(J). Suppose u I. Since A is Hausdorff,
p M
p+1 . Then there exist v0i M
pri such that
therePexists p such that u M
p+1

u v0i ai I M
. By continuing this construction we obtain sequences {vji ; j
p+jri and
Z+ , i = 1, . . . , n} such that vji M
u

n X
s
X

p+s+1
vji ai I M

i=1 j=0

P
Since A is complete, the series j=0 vji converges to some vi A for each i and we have
Pn
u = i=1 vi ai . Thus, u J and the proof is complete.
We now return to the study of the algebras V O and V H. Note first that if A is n O
or n H and M is the maximal ideal of A then A/M p is just the quotient of the ring of
polynomials in n variables modulo the ideal consisting of polynomials all of whose terms
are of degree at least p. Thus, the following result is obvious from the definitions:
7.11 Lemma. The algebras n O and n H both have as completion the algebra of formal
power series C[[z1 , . . . , zn ]].
The following technical lemma due to Chevalley is the key to showing that V O and V H
also have the same completion.
7.12 Lemma. Let V be a germ of an algebraic subvariety of Cn . Then there are no
That is, V O
is reduced.
non-zero nilpotent elements of V O.
Proof. We first reduce to the case where V is irreducible. If V is not irreducible and
V = V1 Vk is its irreducible decomposition, then consider the map
V

O iVi O

induced by restricting to each irreducible component. This is an injection and, hence, it

remains an injection if we apply the V O module completion functor to obtain
V

iV O

O
i

as an V O module is the same as its completion as a

However, the completion of Vi O
local ring. Therefore, if the theorem is true for each of the irreducible varieties Vi , then a

60

J. L. TAYLOR

must map to zero in each V O

and, therefore, must be zero. It
nilpotent element of V O
i
follows that the theorem is also true for V .
Thus, we will assume that V is irreducible. By the normalization theorem (the algebraic
version of Lemmas 5.5 and 5.6 on a prime ideal being regular in variables zm+1 , , zn ),
there is an integer m and a subalgebra A V O isomorphic to m O such that V O is integral
over A. If K is the field of fractions of A and L is the field of fractions of V O then
L is generated by V O over K. To see this, let x = a/b L with a, b V O. Then
b satisfies a polynomial equation bp + cp1 bp1 + c1 b + c0 = 0 with c0 6= 0. Thus,
p1
c1
+ cp1 bp2 + + c1 ) K V O is the inverse for b and so x K V O. It follows
0 (b
that we may choose elements q1 , . . . , qm V O which form a basis for L as a vector space
over K. Relative to this basis, L may be represented as an algebra of matrices with entries
from K. We introduce a K-valued bilinear form [, ] on L by
[x, y] = tr(xy)
Since each non-zero element of L has an inverse and since the trace of the identity is m 6= 0,
this form is non-singular. Let d be the determinant of the invertible matrix B = ([qi , qj ]).
Since it belongs to V O, the element qi qj is integral over A and it follows from Problem 7.4
that its trace belongs to A. Thus, the matrix B has entries in A. It follows that d A and
B 1 has entries in d1 A. By applying B 1 to the column vector (qj ), we obtain a vector
(ai ) with entries in d1 V O which forms a dual basis to (qj ) in the sense that [ai , qj ] = ij .
Now suppose x L is integral over V O. Then each qj x is also integral over V O and,
hence, integral over A. It follows from Problem 7.4 that [qj , x] = tr(qj x) K is actually
in A for all j. This implies that when x is expanded in the dual basis (ai ), its coeficients
lie in A and, hence, that x d1 V O. Thus, we have proved that every element of L which
is integral over V O actually lies in d1 V O.
which is a just the ring of formal power series
At this point we pass to the completion A,

by Theorem 7.11 and, hence, is an integral domain. We denote its field of fractions by K
of V O is A A V O
and remark that this is an extension field of K. The completion V O
algebra L
to be the result of passing from L to A A L
by Problem 7.3 . We define a K
and then localizing relative to the multiplicative set consisting of the non-zero elements of
By Theorem 7.8, A is faithfully flat over A and so V O
= A A V O is embedded as a
A.
subalgebra of A A L. We claim that this algebra is, in turn, embedded as a subalgebra
That is, we must show that nothing is killed when we localize. This means we must
of L.
show that am = 0 for 0 6= a A and m A A L implies that m = 0. Since, {qi } forms
P
ci A. Then am = 0
a basis for L over K, we may write
m
=
ai qi /ci with ai A,
Q
implies that acm = 0 where c = ci A. This implies that
X
aai (c/ci ) qi = 0
and, hence, that aai (c/ci ) = 0 for each i. Since A is an integral domain, this implies that
is embedded as
ai = 0 for each i and, hence, that m = O. Thus, we have shown that V O

a subalgebra of L.
as K
K L, that is, as
It is clear from the construction that we may also describe L
This is clearly
the algebra obtained from L by extending its ground field from K to K.

NOTES ON SEVERAL COMPLEX VARIABLES

61

with basis (qj ) as a K

vector space. It has a non-singular bilinear form
an algebra over K
[, ], and dual basis (ai ), determined as above. By Problem 7.5, the algebra A is integrally
closed. Therefore the result of Problem 7.4 applies and we may argue as in the previous
that is integral over V O
is actually in d1 V O.
Now let
paragraph that any element of L
be nilpotent. If a A then x/a is also nilpotent. Since a nilpotent element is,
x VO
In particular, this holds
in particular, integral, it follows as above that x/a d1 V O.
p
when a = y for any non-zero element y of M and any positive integer p and from this we
M
p for each p. Since it follows from Theorems 7.8 and 7.10
conclude that dx M p V O

p
= 0 we conclude that x = 0. This completes the proof.
that p=1 M
When confusion might otherwise result we will denote by V the germ of the holomorphic
variety associated to V . When it is clear which is meant we will simply write V for either
the algebraic or holomorphic variety.
7.13 Theorem. Let V be a germ of an algebraic variety in Cn and V be the corresponding
germ of a holomorphic variety. Then id V = n H id V .

Proof. Let I = id V n O.
If J = n HI then loc J =mV and so it follows from the

Nullstellensatz that id V = J. Thus, if f id V then f J for some positive integer

such
m. If we consider f to be a formal power series then we have an element f n O
m
in the completion n O
= C[z1 , . . . , zn ] of n O. By the
that f belongs to the ideal n OI
exactness property of completion for finitely generated modules (Theorem 7.6), we have
modulo this ideal is just V O.
We conclude that the image of
that the quotient of n O
is nilpotent and, hence, zero by the previous lemma. Thus, f n OI
= n OJ.

f in V O
Since f n H, it follows from Lemma 7.2(iii) that f J = n HI. We conclude that
idV = n H id V as claimed.
7.14 Theorem. If V is a germ of an algebraic variety then

=
H

O.

Proof. Let {p1 , . . . , pr } be a set of generators for id V n O. Then by the previous theorem
we have that it is also a set of generators for id V n H. Thus, we have a commutative
diagram
nO

nH

nO

nH

O 0

y
H 0

with exact rows, where (f1 , . . . , fr ) p1 f1 + + pr fr defines the maps n Or n O and

r
n H n H. On applying the completion functor to this diagram, the first two vertical
maps become isomorphisms and, hence, the third must be one also. This completes the
proof.
= V H.
In view of Theorems 7.3 and 7.8 we have proved
We now have V O V H V O
the main theorem of this chapter:
7.15 Corollary. If V is a germ of an algebraic variety, then V H is faithfully flat over V O.

62

J. L. TAYLOR

7. Problems
1. Prove that an inverse limit of an inverse sequence of short exact sequences is exact
provided the left hand inverse sequence is surjective.
2. Prove that an A-module Y is flat if and only if whenever X1 and X2 are finitely generated
A-modules and X1 X2 is injective, then Y A X1 Y A X2 is also injective.
3. Prove that if A is a Noetherian local ring with maximal ideal M and B is a local ring
which is a finitely generated integral ring extension of A and if N is the maximal ideal
of B then the completion of B in the N -adic topology is the same as its completion as
an A-module with the M -adic topology.
4. Suppose A is an integrally closed integral domain and K is its field of fractions. Prove
that if a matrix with entries in K is integral over A then its trace lies in A.
5. Prove that the formal power series ring C[[z1 , . . . , zn ]] is a unique factorization domain.
Hint: Use induction on the number of variables, Gausss Theorem (A is a UFD implies
A[z] is a UFD), and an extension of the Weierstrass preparation theorem to formal
power series.

NOTES ON SEVERAL COMPLEX VARIABLES

63

8. Sheaves
Sheaf theory provides the formal machinery for passing from local to global solutions for
a wide variety of problems as well as for classifying the obstruction to so doing when local
solutions do not give rise to global solutions. The following is a list of typical examples of
such local to global problems:
1. If X is a compact Hausdorff space and f is a continuous complex valued function on
X which never vanishes, then f locally has a continuous logarithm. Does it have a
logarithm globally? In other words, is there a continuous function g on X such that
f = exp g?
f
= g has a
2. If U is a domain in C and g is a C function on U then the equation
z
solution locally in a neighborhood of each point. Does it have a global solution on U ?
3. If U is a domain in Cn , V U is a holomorphic subvariety and f is holomorphic on V ,
then for each point V there is a holomorphic function defined in a neighborhood
U of in Cn whose restriction to U V agrees with that of f . Is there a holomorphic
function defined on all of U whose restriction to V is f ?
4. If U is a domain in Cn and V U a holomorphic subvariety then V is locally defined
as the set of common zeroes of some set of holomorphic functions. Is there a set of
holomorphic functions defined on all of U so that V is its set of common zeroes?
Generally these problems involve classes of functions continuous, holomorphic, C ,
etc. which make sense on any domain in the underlying space. The notion of sheaf
simply abstracts this idea:
8.1 Definition. Let X be a topological space. We consider the collection of open subsets
of X to be a category where the morphisms are the inclusions U V . Then a presheaf
on X is a contravariant functor from this category to the category of abelian groups. A
morphism between two sheaves on X is a morphism of functors.
Thus, a presheaf S on X assigns to each open set U X an abelian group S(U ) and to
each inclusion of open sets U V a group homomorphism U,V : S(V ) S(U ), called the
restriction map, in such a way that U,U = id for any open set U and U,W = U,V V,W
for any triple U V W .
A morphism : S T between two presheaves on X assigns a morphism U : S(U )
T (U ) to each open set U in a way which commutes with restriction. Unless the context
dictates otherwise, we shall usually drop the subscript from U and write simply .
An example of a presheaf is the assignment to each open subset U X of the algebra
of continuous functions C(U ). The restriction map U,V is just restriction of functions in
this case. The resulting presheaf C is called the presheaf of continuous functions. The
presheave, C , of C -funtions on a C -manifold and the presheaf, H, of holomorphic
functions on Cn are defined in the same way. If Cn is given the Zariski topology, then we
may define on it the presheaf, O, of regular functions. If X is any topological space and
G is a fixed abelian group, then we may define a presheaf called the constant presheaf by
assigning G to each non-empty open set and 0 to the empty set. The first four of the above
examples are actually presheaves of algebras not just of abelian groups. As we shall see,

64

J. L. TAYLOR

the existence of additional structure on the objects S(U ) for a presheaf S is the typical
situation, although the abelian group structure is all that is needed in much of the theory.
If S is a presheaf, then an element s S(U ) will be called a section of S over U . If
U V then the image of a section s S(V ) under the restriction map U,V : S(V ) S(U )
will often be denoted s|U and called the restriction of s to U .
A sheaf is a presheaf which is locally defined in a sense made precise in the following
definition:
8.2 Definition. If S is a presheaf on X then S is called a sheaf if the
S following conditions
are satisfied for each open subset U X and each open cover U = iI Ui of U :
(i) if s S(U ) is a section such that s|Ui = 0 for all i I, then s = 0;
(ii) if {si S(Ui )}iI is a collection of sections such that si |Ui Uj = sj |Ui Uj for all
i, j I, then there is a section s S(U ) such that s|Ui = si for all i I.
Note that since the empty cover is an open cover of the empty set, it follows from 8.2(i)
that S() = 0 if S is a sheaf.
The presheaves of continuous, C , holomorphic and regular functions described earlier
are obviously sheaves. However, the presheaf which assigns a fixed group G to each nonempty open set is not a sheaf unless the underlying space has the property that every open
set is connected. There is, however, a closely related sheaf: the sheaf of locally constant
functions with values in G. In fact, for every presheaf there is an associated sheaf, as we
shall show in Theorem 8.6.
If S is a presheaf on X then the stalk, Sx , of S at x X is the group lim{S(U ) : x U }.

Given a section s S(U ) and an x U , the image of s in Sx is denoted sx and is called

the germ of s at x. The stalks contain the local information in a presheaf. If : S T
is a morphism of presheaves then clearly induces a morphism x : Sx Tx for each x.
The fact that sheaves are presheaves that are locally defined is illustrated by the following
two results:
8.3 Theorem. If S is a sheaf on X and s S(U ) is a section over U , then s = 0 if and
only if sx = 0 for each x U .
Proof. If each sx vanishes then for each x U there is a neighborhood Ux of x such that
the restriction of s to Ux is zero. By 8.2(i), this implies that s = 0.
8.4 Theorem. If : S T is a morphism of sheaves on X, then U is injective for each
open set U if and only if x is injective for each x X. Furthermore, U is an isomorphism
for each open set U if and only if x is an isomorphism for each x X.
Proof. That U injective (surjective) for each open set U implies x is injective (surjective)
for each x X is obvious from the definition of direct limit.
If each x is injective then a section s S(U ) in the kernel of U has vanishing germ
at each x U . By the previous theorem this implies that s = 0. Thus, U is injective.
If x is an isomorphism for each x X then U is injective for each open set U by the
previous paragraph and so to complete the proof we must show it is also surjective. Given
a section t T (U ) there is an sx Sx such that x (sx ) = tx for each x U . But this
means that for each x U there is a neighborhood Ux of x and a section sUx S(Ux )

NOTES ON SEVERAL COMPLEX VARIABLES

65

such that (sUx ) = t|Ux . Then for x, y U , (sUx |Ux Uy sUy |Ux Uy ) = 0. Since Ux Uy
is injective, this implies that sUx |Ux Uy = sUy |Ux Uy and, by 8.2(ii), that there exists a
section s S(U ) such that s|Ux = sUx for each x U . It follows that (s)x = tx for each
x U and, hence that (s) = t. Thus, U is surjective and the proof is complete.
S
If S is a presheaf on X then we may construct a topological space S =
Sx by
xX

choosing as a neighborhood base of a point u Sx all sets of the form {sy : y U }

where U is a neighborhood of x, s S(U ) and sx = u. There is a continuous projection
: S X, defined by (Sx ) = x. A section for over a subset Y X is a continuous
map : X S such that = id. The following theorem has an elementary proof
which is left as an exercise (Problem 8.1).
8.5 Theorem. With S, X, S, and as above,
(a) the projection : S X is a local homeomorphism;
(b) if Y is a subset of X then a function : Y S, with = id, is a section if and
only if for each y Y there is a neighborhood U of y in X and an s S(U ) such
that (x) = sx for each x U .
If S is a presheaf then we may construct another presheaf S as follows: Let S and
) be the
: S X be as constructed above. Then for each open set U X we let S(U
group of sections of : S X over U . There is a morphism of presheaves S S defined
by sending s S(U ) to the section x sx : U S.
8.6 Theorem. If S is a presheaf and S is defined as above, then
(a) S is a sheaf;
(b) S S induces an isomorphism Sx Sx for each x X;
(c) S S is an isomorphism of presheaves if and only if S is a sheaf.
Proof. Part (a) is obvious since continuity is a local condition. Part (c) will follow from
Theorem 8.4 if we can establish part (b). However, if s S(U ) has germ at x which is
sent to zero by Sx Sx then sy = 0 for all y in some neighborhood V U of x. This
implies that s|V = 0, by Theorem 8.3 and, hence, that s has germ 0 at x. Thus, Sx Sx
is injective. That it is surjective follows immediately from Theorem 8.5(b).
We shall call S the sheaf of germs of the presheaf S. It is a simple exercise to see that
the functor S S is the left adjoint functor of the forgetful functor which assigns to a
sheaf its associated presheaf (Problem 8.4).
The preceding two theorems make it clear that the definition of sheaf we have adopted
is equivalent to the definition given below which is the one most often encountered in the
literature:
8.7 Alternate Definition of Sheaf. A sheaf over X is a topological space S with a map
: S X such that:
(i) is a surjective local homeomorphism;
(ii) Sx = 1 (x) has the structure of an abelian group for each x X;
(iii) the group operation is continuous in the topology of S.

66

J. L. TAYLOR

Note that condition (i) above implies that the images of local sections of S form a
neighborhood base for the topology of S and, in view of this, (iii) is equivalent to the
requirement that for each open set U X, the sum of two sections of S over U is again
a section. With the above definition, a morphism : S T of sheaves over X is a
continuous map which commutes with projection and is a group homomorphism Sx Tx
between stalks. The equivalence between the category of sheaves in our previous sense and
the category of sheaves in this sense is given by the constructions S S and S S of
Theorems 8.5 and 8.6. We will normally stick with our original definition, but on occasion
it will be useful to use the fact that it has the above alternate formulation.
One of the things the alternate definition of sheaf allows us to do easily is define the
notion of a section of a sheaf : S X over a subset Y X which is not necessarily
open. A section : Y S is a continuous map such that = id. The group of all
sections over Y will be denoted by (Y, S). We shall have more to say about this later
(Theorem 8.12). Of course, for an open set U X, (U ; S) = S(U ).
An additive category is a category such that for each pair A, B hom(A, B) has an
abelian group structure satisfying a distributive law relative to composition, direct sums
are defined and there is a zero object. If : A B is a morphism in an additive category,
then a kernel ker for is an object with a morphism : ker A such that = 0
and such that any morphism : C A with = 0 factors through . Similarly, a
cokernel coker for is an object with a morphism : B coker such that = 0
and such that any morphism : B D with = 0 factors through . In general, a
morphism need not have a kernel or a cokernel. When they do exist they are unique up
to isomorphism.
An abelian category is an additive category such that every morphism : A B has
both a kernel and a cokernel and the natural map
coim = coker(ker A) im = ker(B coker )
is an isomorphism.
A morphism : S T in the category of presheaves has both a kernel and a cokernel
and these are the obvious presheaves: U ker U and U coker U . It is easy to see
that the presheaves on a given space X form an abelian category. However, we are not
really interested in this category. We are interested in the category of sheaves.
Suppose : S T is a morphism of sheaves. It is easy to see that the kernel of as a
morphism of presheaves (the presheaf U ker U ) is, in fact, a sheaf and is the category
theoretic kernel of as a sheaf morphism (Problem 8.2). However, the analogous statement
is not true in general for the cokernel (Problem 8.3). In other words, U coker U need
not be a sheaf. However, we have the following:
8.8 Theorem. Suppose : S T is a morphism of sheaves. Then the sheaf of germs of
the presheaf U coker U is a cokernel for .
Proof. If C is the presheaf U coker U and C is its sheaf of germs, then the composition
of the presheaf morphisms T C and C C is a sheaf morphism such that = 0.
Any sheaf morphism : T D such that = 0 must factor through T C as a
presheaf morphism but since it is a sheaf morphism it must actually factor through by
Problem 8.4. This completes the proof.

NOTES ON SEVERAL COMPLEX VARIABLES

67

We shall denote the sheaf of germs of the presheaf U coker U by coker . This might
seem ambiguous since one might use the same notation for U coker U itself since it is
the cokernel of as a presheaf. However, we will never do this, since our focus will be on
the category of sheaves.
Note that if U is an open set, then the space of sections (coker )(U ) is not the obvious
candidate, coker U , since U coker U is not a sheaf; however, it is true that we get
the obvious thing at the level of stalks: that is, (coker )x = coker x . This follows from
the fact that a presheaf and its sheaf of germs have the same stalks (Theorem 8.6(b)). It
follows that im = ker(T coker ) is characterized by
(im )(U ) = {t T (U ) : tx im x

for all

x U}

In particular, is an epimorphism (coker = 0) if and only if each x is surjective. We

shall call surjective in this case.
Note that we have defined things in such a way that, for a morphism of sheaves, the
notions of kernel, image, and cokernel as well as epimorphism and monomorphism are
all local, that is, are all defined stalkwise. In particular, a sequence A B C is
exact (im(A B) = ker(B C)) if and only if the corresponding sequence of stalks
Ax Bx Cx is exact for each x X. Using this fact, it is easy to see that the category
of sheaves on X is an abelian category (Problem 8.5).
What we have gained in passing from presheaves to sheaves is that sheaves and their
morphisms are defined locally. What we have lost in passing from presheaves to sheaves
is that the functor which assigns to a sheaf S its group of sections S(U ) over an open set
U is no longer exact. Studying this loss of exactness is the central theme of sheaf theory.
Recall that (U ; ) is the functor which assigns to a sheaf its group of sections over a
set U . For U open, (U ; S) = S(U ). In case U = X, we will sometimes write simply (S)
for (X; S).
8.9 Theorem. For each open set U the functor (U ; ) is left exact.

Proof. Let 0
A
B
C be an exact sequence of sheaves on X. This means

that 0
Ax
Bx
Cx is exact for each x X. Let U X be open. Then
U : A(U ) B(U ) is injective by Theorem 8.4. Now suppose b B(U ) is in the kernel
of U . For each x U there is a neighborhood Vx of x and a section ax A(Vx ) such
that (ax ) = b|Vx . Then (ax )|Vx Vy = (ay )|Vx Vy for each pair x, y U and it follows
from the injectivity of Vx Vy that ax |Vx Vy = ay |Vx Vy . From the definition of a sheaf it
now follows that there exists a A(U ) such that a|Vx = ax for each x and from this that
(a) = b. This completes the proof.
We shall show that the category of sheaves has enough injectives to construct injective
resolutions for each sheaf. From this it follows that we have naturally defined right derived
functors for every left exact functor in particular, for the functors (U ; ). The resulting
functors are those of sheaf cohomology and are the subject of the next section.
We end this section with a discussion of the operations on sheaves that are induced by
a continuous map f : Y X.

68

J. L. TAYLOR

8.10 Definition. Let f : Y X be a continuous map of topological spaces. Then,

(a) if T is a sheaf on Y then the direct image, f T , of T under f is the sheaf on X
defined by U T (f 1 (U ));
(b) if S is a sheaf on X then the inverse image, f 1 (S), of S under f is the sheaf of
germs associated to the presheaf on Y defined by V lim{S(U ) : U open, f (V )

U }.

Part (b) of the above definition is a case where it would be more instructive to use the
alternate definition of sheaf given in 8.7. In fact if : S X is a sheaf in this sense, then
f 1 S is just the pullback of : S X via f : Y X. This is the topological space
f 1 S = {(s, y) S Y : (s) = f (y)}
with projection f 1 S Y given by projection on the second coordinate. This description
of f 1 S makes it clear that it is a sheaf on Y which has stalk at y Y equal to the stalk
of S at f (y). Form this it follows easily that f 1 is an exact functor from sheaves on X
to sheaves on Y .
The direct image functor f from sheaves on Y to sheaves on X is not exact in general.
To see this, consider the map f : Y pt; in this case, f T is the sheaf which assigns to pt
the group (Y ; T ) and we know that is not always exact (Problem 8.6). On the other
hand, an argument like the one in Theorem 8.9 shows that f is alway left exact. Thus, it
is another functor for which we expect to be able to construct right derived functors.
A special case of the inverse image functor is the restriction functor. Here, Y X is a
subset and i : Y X is the inclusion. For a sheaf S on X, i1 S is denoted S|Y and is
called the restriction of S to Y . Using the alternate definition of sheaf, the restriction of
: S X to a subset Y is the space SY = 1 (Y ) with topology and projection SY Y
inherited from S and . The group of global sections of the restricted sheaf, (Y, S|Y ), is
the same as the group (Y ; S) of continuous sections of S over Y . It follows from Theorem
8.9 that
8.11 Corollary. For any subset Y X and any sheaf S on X, (Y ; ) is a left exact
functor.
Thus (Y ; ) is also a functor which we expect to have right derived functors if we can
show that there are enough injective sheaves.
If S is a sheaf on X, Y is a subspace of X and U is an open set containing Y , then
restriction defines a morphism S(U ) = (U ; S) (Y ; S). Thus, restriction defines
a morphism : lim{(U ; S) : Y U } (Y ; S). This is often, but not always an

isomorphism:
8.12 Theorem. The morphism
: lim{(U ; S) : Y U } (Y ; S)

is injective. It is an isomorphism if X is Hausdorff and Y is compact or if X is paracompact

and Y is closed.
Proof. If Y U and s (U ; S) is such that s has image zero in (Y ; S) then sx = 0
for each x Y . This implies that s has restriction zero in an open set V containing Y

NOTES ON SEVERAL COMPLEX VARIABLES

69

and, hence, that s determines the zero element of : lim{(U ; S) : Y U }. Thus, is

injective.
Now suppose s (Y ; S). Then there is an open cover {Ui }iI of X and a family
of sections {si (Ui ; S)}iI such that s|Vi Y = si |Vi Y for each i. If X is Hausdorff
and Y compact or if X is paracompact and Y closed then we may assume that the cover
{Ui } is a locally finite cover of X and, furthermore, that
S there is locally finite collection

of open sets {Vi }iI with Vi Ui for each i and Y iI Vi . Now for each x X set
I(x) = {i I : x Vi } and
W = {x

Vi : six = sjx

for all

i, j I(x)}

Then I(x) is a finite set. For a given x let U be a neighborhood of x which meets Vi for i
in only a finite set, J I. Then the set
Wx = U

Vi

iII(x)

is an open set containing x. In fact, Wx is the set of y U such that I(y) I(x).
Now if x W choose Nx to be a neighborhood of x contained in iI(x) Ui Wx on which
si |Nx = sj |Nx for i, j I(x). Then the conditions for membership in W will also be satisfied
for any point y Nx . In other words, W is an open subset of X. The fact that the si
fit together to define a section s on Y means that Y W . By construction, the si fit
together on W to define a section in (W ; S) which restricts to s on Y .
The functors indroduced in the following definition are also left exact functors on sheaves
(Problem 8.7):
8.13 Definition. If S is a sheaf on X and s (X; S) is a section, then the support of
s is the (necessarily closed) set K = {x X : sx 6= 0}. If Y X is any subset of X
then Y (X; S) is the group of sections s (X; S) with support contained in Y . If is a
family of subsets of X which is closed under finite unions, then (X; S) is the group of
sections s (X; S) with support contained in some member of .
A family as above is called a family of supports. A common situation in which
(X; S) is useful is when X is locally compact and the family of supports is is the
family of compact subsets of X.
8. Problems
1. Prove Theorem 8.5.
2. Prove that if is a morphism of sheaves then the presheaf U ker U is a sheaf and is
a kernel for in the category of sheaves.
3. Give an example of a morphism of presheaves such that the presheaf U coker U
is not a sheaf.
4. Prove that if S is a presheaf then each presheaf morphism S T , where T is a sheaf,
Prove that
factors uniquely through the morphism S S of S to its sheaf of germs S.

70

J. L. TAYLOR

this means that S S is a left adjoint functor for the forgetful functor which regards
a sheaf as just a presheaf.
5. Prove that the category of sheaves is an abelian category.
6. Give an example to show that the functor (X; ) is not necessarily exact.
7. Prove that the functors Y (X, ) and (X, ) of Definition 8.11 are left exact.

NOTES ON SEVERAL COMPLEX VARIABLES

71

9. Sheaf Cohomology
Recall that in any abelian category an object A is called injective if the functor hom(, A)
is exact and is called projective if the functor hom(A, ) is exact. In this section we shall be
concerned with injective objects. The following summarizes the most elementary properties
of injectives (Problem 9.1):
9.1 Theorem. In any abelian category:
(1) An object A is injective if and only if for every monomorphism i : B C, each
morphism : B A extends to a morphism : C A such that = i.
(2) Every monomorphism i : A B, with A injective, splits (has a left inverse).
(3) If 0 A B C 0 is an exact sequence and A is injective, then B is injective
if and only if C is injective.
Our abelian category is said to have enough injectives if for every object A there is an
injective object I and a monomorphism A I. This, and the fact that every morphism
has a cokernel, allows one to construct injective resolutions
0 A I0 I1 In
for each object A. There is another way of thinking about such resolutions which yields
both economy of notation and additional insight and so is worth introducing. We will let
I denote the complex
0 I0 I1 In
and identify A with the complex
0A0
where A appears in degree zero. Then an injective resolution of A may be thought of as a
morphism of complexes
i:AI
where I is a complex of injective objects (zero in negative degrees) and i induces an
isomorphism on cohomology (both complexes have vanishing cohomology in all degrees
except zero where the cohomology is A). A morphism of complexes which induces an
isomorphism of cohomology is called a quasi-isomorphism. Thus, an injective resolution of
an object A is a quasi-isomorphism A I where I is a complex of injectives which vanishes
in negative degrees. Actually, insisting that I vanish in negative degrees is equivalent, for
the purposes of this theory, to insisting that it be bounded on the left that is, vanish for
sufficiently high negative degrees.
A morphism : X Y of complexes is homotopic to zero if there are morphisms
n
n
hn : X n Y n1 such that hn+1 X
+ Yn1 hn = n for each n. Here, X
: X n X n+1
and Yn1 : Y n1 Y n are the differentials in the complexes X and Y . Two morphisms
, : X Y of complexes are said to be homotopic if their difference, is homotopic
to zero. A key result is the following (Problem 9.2):

72

J. L. TAYLOR

9.2 Theoem. If : A B is a morphism in an abelian category and A I and B J

are injective resolutions of A and B, then there is a morphism of complexes
:I J
such that the diagram
A I

y
y
B J
is commutative. Furthermore, any two maps I J with this property are homotopic.
When Theorem 9.2 is applied to two different injective resolutions of A and the identity
morphism from A to A, it implies the following:
9.3 Corollary. If i : A I and j : A J are two injective resolutions of A then there
are morphisms : I J and : J I such that
I

I
x

J A J
is a commutative diagram and both and are homotopic to the identity.
If X = {X n , n } is a complex, then its cohomology is the graded group {H n (X)} where
H n (X) = ker n / im n1 . A morphism : X Y of complexes induces a morphism
H() : H(X) H(Y ) of cohomology. Two morphisms which are homotopic induce the
same morphism of cohomology. Also, if two morphisms and are homotopic and F is
a functor into another abelian category, then F () and F () are also homotopic. These
facts and Theorem 9.2 imply the following:
9.4 Corollary. If : A B is a morphism, i : A I and j : B J are injective
resolutions of A and B and F is a functor from our category to another abelian category,
then the morphisms
: I J of Theorem 9.2 induce a single, well defined morphism
: H(F (I)) H(F (J)). This is an isomorphism if is an isomorphism.
In any reasonable category with enough injectives there is a way of assigning to each
object A a particular injective resolution I(A). In a small category this can be done using
the axiom of choice. We shall show how to do this in the case of the category of sheaves.
In many cases this can be done in such a fashion that A I(A) is a functor ideally an
exact functor. This is nice when it can be done (and it can be for sheaves) but it is not
necessary for the developement of the theory. In any case, assuming that we are working
in a category with enough injectives and with some way of assigning an injective resolution
to an object, we may construct the higher derived functors of a left exact functor F as
follows: the n-th derived functor of A is
Rn F (A) = H n (F (I(A)))
the n-th cohomology of the complex obtained by applying the functor F to the complex of
injectives I(A). Of course, by Corollary 9.4, different choices of resolutions I(A) will yield

NOTES ON SEVERAL COMPLEX VARIABLES

73

isomorphic objects Rn F (A) but, in order that the Rn F be functors, it is important that
a specific way of making the assignment A I(A) be available. That Rn F is a functor
then follows from Corollary 9.4.
The fact that F is left exact means that if A I 0 I 1 . . . is an injective resolution,
then on applying F exactness is preserved at the first two terms; that is,
0 F (A) F (I 0 ) F (I 1 )
is exact. From this it follows that:
9.5 Theorem. If F is a left exact functor from an abelian category with enough injectives
to an abelian category, then there is an isomorphism of functors F R0 F .
If (as is the case for the category of sheaves) A I(A) is an exact functor then an
exact sequence of objects
0ABC0
yields an exact sequence of complexes
0 I(A) I(B) I(C) 0
because the objects in these complexes are injective, this short exact sequence splits and,
hence, remains exact when we apply a left exact functor F . This yields a short exact
sequence of complexes
0 F (I(A)) F (I(B)) F (I(C)) 0
which, in turn, yields a long exact sequence of cohomology
0 F (A) F (B) F (C) R1 F (A) R1 F (B) R1 F (C) R2 F (A)
However, to get this result it is not neccessary to assume that A I(A) is exact or is even
a functor. In fact, given injective resolutions A I(A) and C I(C) one may construct
an injective resolution B J along with morphisms I(A) J and J I(C) such that
0 I(A) J I(C) 0
is an exact sequence of complexes. To do this, we set J n = I n (A) I n (C) and define
morphisms j : B J 0 and n : J n J n+1 as follows: Using the fact that I 0 (A) is
injective, we extend iA : A I 0 (A) to a morphism j1 : B I 0 (A). We let j2 : B I 0 (C)
be the composition of B C with iC : C I 0 (C). Then j = j1 j2 . Clearly j is a
monomorphism of B into J 0 and the diagram
0

iA y

jy

iC y

0 I 0 (A) J 0 I 0 (C) 0

74

J. L. TAYLOR

is commutative. We repeat this argument with iA : A I 0 (A) and iC : C I 0 (C) replaced by coker iA I 1 (A) and coker iC : I 1 (C), respectively, and obtain a commutative
diagram
0 A B C 0

jy
iA y
iC y
0 I 0 (A) J 0 I 0 (C) 0

0
0
0
A

y
y
Cy
0 I 1 (A) J 1 I 1 (C) 0
Continuing in this way, we construct an injective resolution B J of B and morphisms
of complexes I(A) J and J I(C) for which the diagram
0

iA y

jy

iC y

0 I(A) J I(C) 0
is commutative with exact rows. On applying any left exact functor F we conclude, without
the assumption that I() is exact or even a functor, that
9.6 Theorem. If F is a left exact functor from an abelian category with enough injectives
to an abelian category, and if
0ABC0
is a short exact sequence in the first category, then there is a long exact sequence
0 F (A) F (B) F (C) R1 F (A) R1 F (B) R1 F (C) R2 F (A)
In this result, the morphisms Rn F (A) Rn F (B) and Rn F (B) Rn F (C) are just
those induced by A B and B C as in Corollary 9.4; i. e. they are the images of these
morphisms under the functor Rn F . The connecting morphisms Rn F (C) Rn+1 F (A) a
priori depend on the choices made in the construction of J. In fact, they do not depend on
these choices. They are well defined and depend functorially on the short exact sequence
0 A B C 0. To prove this requires some diagram chasing which we choose not
to do here. It is done in any number of books on homological algebra.
Of course, one does not really compute the objects Rn F (A) using injective resolutions.
In practice, one uses the long exact sequence to reduce the computation of Rn F (A) for
complicated objects A to that for simpler objects or one uses Theorem 9.8 below which
often allows one to compute Rn F (A) using much simpler resolutions of A.
An object C is said to be acyclic for the functor F if Rn F (C) = 0 for n > 0.
9.7 Theorem. Let F be as in Theorem 9.6. Then injective objects are F -acyclic.
Proof. If I is injective, then 0 I I 0 is an injective resolution of I. It follows from
Corollary 9.4 that Rn F (I) may be computed by applying F to this resolution and taking
cohomology. Thus, Rn F (I) = 0 for n > 0 and I is acyclic.

NOTES ON SEVERAL COMPLEX VARIABLES

75

9.8 Theorem. Let F be as in Theorem 9.6 and suppose that A J is a resolution of A

by a complex of F -acyclic objects. Then there is an isomorphism Rn F (A) H n (F (J))
for each n.
Proof. Let K n = ker{J n J n+1 } and consider the long exact sequences for {Rn F }
determined by the short exact sequences
0 K p J p K p+1 0,

p0

where K 0 = A. Using the fact that Rn F (J p ) = 0 for n > 0, we conclude from these long
exact sequences that
Rq F (K p ) ' Rq1 F (K p+1 ),
and

p 0, q > 1

R1 F (K n1 ) ' F (K n )/ im{F (J n1 ) F (K n )} ' H n (J)

An induction argument then shows that Rn F (A) = Rn F (K 0 ) ' H n (J), as required.

At this point we leave general homological algebra and return to the study of sheaves.
It is convenient to suppose we have a fixed sheaf of rings R on a space X and then to study
the abelian category consisting of all sheaves of R-modules. A sheaf of R-modules is a
sheaf M on X such that for each open set U X, M(U ) is a R(U )-module in such a way
that the module action R(U ) M(U ) M(U ) commutes with restriction. An abelian
group may be regarded as a module over the ring of integers. Therefore, an ordinary sheaf
of abelian groups is a sheaf of modules over a sheaf of rings if for the sheaf of rings we
choose the sheaf of germs of the constant presheaf U Z.
It is well known that the category of modules over a ring has enough injectives and,
in fact, that there is a functor which assigns to each module a monomorphism into an
injective module. We give a brief description of the construction:
An abelian group is injective if and only if it is divisible (for every element g and every
integer n there is an element h so that nh = g). If G is an abelian group, let F (G) be the free
abelian group generated by G as a set. Thus, an element of F (G) is a function f : G Z
which is zero except at finitely many points and the group operation is pointwise
P addition
of functions. There is a group homomorphism F (G) G defined by f
f (g)g. Let
K(G) be its kernel. Then let Q(G) be the free module over the rationals generated by G
(finitely non-zero funtions from G to Q) and note that Q(G) contains F (G) and, hence,
K(G) as subgroups. Finally, we set D(G) = Q(G)/K(G). Then D(G) is a divisible group
containing a copy of G as a subgroup. Now if R is any ring and M is a module over R
then we consider M as an abelian group and construct D(M ). Then the R-module
homZ (R, D(M ))
is injective (problem 9.3) and there is a monomorphism M homZ (R, D(M )) defined
by m (r rm) where M is regarded as a subgroup of D(M ). Thus, we have defined
a functorial way to assign to each R-module M a monomorphism of M into an injective
module.

76

J. L. TAYLOR

If R happens to be an algebra over a field k, then M homk (R, M ) is a simpler way

of embedding each R-module into an injective R-module.
Now if we have a sheaf of modules M over a sheaf of rings R on X it is a simple matter
x be the
to embed M in an injective sheaf of R-modules. For each stalk Mx of M let M
injective Rx module containing Mx constructed using one of the methods described above.
Then I 0 (M) is the sheaf defined as follows: I 0 (M)(U ) is the R(U ) module consisting of all
x . Note that there is no requirement
functions which assign to each x U an element of M
that these functions be continuous. Clearly M is embedded in I 0 (M) as those functions
which are continuous and have values in the modules Mx .
9.9 Theorem. The sheaf I 0 (M) is an injective object in the category of sheaves of Rmodules.
Proof. Suppose : S T is a monomorphism of sheaves of R modules and : S
I 0 (M) is a morphism of sheaves of R-modules. Then determines a function which
x ) as follows: x (sx ) is the value at
assigns to each x X an element x homRx (Sx , M
0
x of the germ x (sx ) I (M). Then, for an open set U and a section s S(U ) we have
that (s) I 0 (M) is the function on U defined by (s)(x) = x (sx ). By the injectivity of
x the morphism x : Sx M
has an extension to a morphism x : Tx M
for each x
M
(we fix one such extension for each x using the axion of choice). If we set (t)(x) = x (tx )
for U open and t T (U ) then we have defined a morphism : T I 0 (M) which is an
extension of , that is = . Thus, I 0 (M) is injective in the category of sheaves of
R-modules.
9.10 Corollary. If R is a sheaf of rings on X then the category of sheaves of R-modules
on X has enough injectives.
In fact, it is clear that the above construction can be used to define a functor which
assigns to each R-module M an injective resolution M I(M).
Since we have enough injectives, we know that for any left exact functor from the category sheaves of R-modules to an abelian category there are corresponding higher derived
functors for which Theorems 9.5, 9.6, 9.7 and 9.8 hold.
9.11 Definition. If M is a sheaf of R-modules on X, Y X is a subset of X and a
family of closed subsets of X closed under finite union, then we set
(i) H p (Y ; M) = Rp (Y ; M);
(ii) HYp (X; M) = Rp Y (X; M);
(iii) Hp (X; M) = Rp (X; M)
for each p. These groups are called the p th sheaf cohomology group of M on Y , of M on
X with support in Y and of M on X with supports in , repectively.
If f : Y X is a continuous map then there are also higher derived functors, Rp f (),
for the direct image functor, though we wont bother to give them special names. Here,
if R is a sheaf of rings on X then f 1 R is a sheaf of rings on Y and the functor f from
sheaves on Y to sheaves on X maps a sheaf of f 1 R-modules to a sheaf of R-modules.
One of the most common uses of the long exact sequence theorem is to derive a relation
between the cohomology of a sheaf on X and on a closed subset Y X. Thus, let M

NOTES ON SEVERAL COMPLEX VARIABLES

77

be a sheaf of R-modules on X, let i : Y X be the inclusion of a closed subset and let

U = X Y . Consider the sheaf MY = i i1 M on X. This is the sheaf V M(Y V ).
Thus, an intuitive description of the construction of MY is restrict M to Y and then
extend by zero to X. There is an epimorphism from M to MY given by s s|Y V :
M(V ) MY (V ) = M(Y V ). In fact MY is the unique quotient of M whose stalks at
points of Y agree with those of M and whose stalks at points not in Y are zero. We will
call the kernel of this map MU since its intuitive description is also restrict M to U and
then extend by zero to all of X. However, its precise description is quite different from
that of MY . It is the sheaf of germs of the presheaf that sends V to M(V ) if V U and
to zero otherwise. It is the unique subsheaf of M which has stalks equal to those of M at
points of U and zero at points not in U .
Since we have a short exact sequence
0 MU M M Y 0
we have a corresponding long exact sequence of cohomology.
H p (X; MU ) H p (X; M) H p (X; MY ) H p+1 (X; MU )
It remains to interpret the meaning of H p (X; MU ) and H p (X; MY ). These can be interpreted as the pth derived functors, applied to M of the functors M (X; MU ) and
M (X; MY ), repectively. Then it is easy to see that (Problem 9.4)
H p (X; MY ) = H p (Y ; M)
and

H p (X; MU ) = Hp (U ; M)

where is the family of supports in U consisting of subsets of U which are closed in X.

We thus have the following:
9.12 Theorem. Let M be a sheaf of R-modules on X, Y a closed subset of X and
U = X Y . Then there is a long exact sequence:
Hp (U ; M) H p (X; M) H p (Y ; M) Hp+1 (U ; MU )
where is the family of supports consisting of subsets of U which are closed in X.
From Theorem 9.8 we know that we can compute the sheaf cohomology groups from
any resolution by sheaves that are acyclic for the given functor (, Y , ). Thus, we will
devote the remainder of this chapter to describing various classes of acyclic sheaves and
dicussing some examples.
9.13 Definition. Let M be a sheaf of R-modules on X then
(a) M is called flabby if for every open set U X the restriction map (X; M)
(U ; M) is surjective;
(b) M is called soft if for every closed set Y X the restriction map (X; M)
(Y ; M) is surjective;
(c) M is called fine if for every locally finite open cover {Ui }iI of X there is a family
of
P morphisms {i : M M}iI such that i is supported in Ui for each i and
i i = id. The family {i }iI is called a partition of unity for the sheaf M
subordinate to the cover {Ui }.

78

J. L. TAYLOR

The definition of fine sheaf needs some comment. The statement that i is supported
in Ui means
P that there is a closed subset Ki of Ui such that x = 0 for all x X Ki .
The sum i i makes sense as a morphism from M to M because the open cover {Ui }
is locally finite. This and the fact that i is supported in Ui implies
P that, in a sufficiently
small neighborhood of each point, only finitely many terms of i i are non zero and,
hence, the sum makes sense in such a neighborhood; but then these local morphisms fit
together to define a morphism of sheaves globally.
In what follows, the term -acyclic will mean acyclic for the functor (X; ).
9.14 Theorem. If X is any topological space, then in the category of sheaves of Rmodules on X
(i) if the first two terms of a short exact sequence are flabby then so is the third;
(ii) injective flabby -acyclic;
Proof. For U open and M a sheaf of R-modules on X, let RU be the restriction of R to
U followed by extension by zero to all of X. Then we have an inclusion RU R. Thus,
if M is injective then each morphism from RU to M extends to a morphism from R to
M. But a morphism RU M is just a section of M over U and a morphism R M is
just a section of M over X (Problem 9.5). Thus, sections of M over U extend to sections
of M over X. Thus, injective flabby.
Now suppose that A is a flabby sheaf and

0 A B C 0
is an exact sequence of sheaves of R-modules. We wish to prove that

0 (X; A) (X; B) (X; C) 0

is also exact. Of course, we need only prove that : (X; B) (X; C) is surjective. To
this end, we let c be an element of (X; C) and consider the class of pairs (U, b) where U
is an open subset of X, b (U ; B) and (b) = c|U . This class is non-empty since c is
locally in the image of and it is partially ordered under the relation: (U1 , b1 ) < (U2 , b2 )
if U1 U2 and g1 = g2 |U1 . It also has the property that a maximal totally ordered subset
has a maximal element (by taking union). Thus, it follows from Zorns Lemma that there
is a maximal element (U, b) in this class. If U = X we are done. If not then there is
an x X U . We may choose a neighborhood V of x and an element b0 (V ; B)
such that (b0 ) = c|V . We then have (b|U V b0 |U V ) = 0 and so there is an element
a0 (U V ; A) such that (a) = bU V b0 |U V . We use the fact that A is flabby to
extend a0 to a section a of A on all of V . Then b and b0 (a) agree when restricted to
U V and, hence, define a section b0 (U 0 ; B) where U 0 = U V . But then (U 0 , b0 ) is in
our class and is larger than (U, b), contradicting the maximality. This proves the exactness
of the above sequence.
We may now prove part(i) of the Theorem. If A and B are flabby in the above short
exact sequence, then their restrictions to an open set U are also flabby. It follows from the
above paragraph and the fact that A|U is flabby that any section c of C over U is (b) for

NOTES ON SEVERAL COMPLEX VARIABLES

79

some section of B over U . But, since B is flabby, the section b is the restriction to U of a
global section b0 of B. Then (b0 ) provides an extension of c to a global section of C.
Now to prove that a flabby sheaf A is -acyclic we embed it in an injective I and use
the long exact sequence of cohomology for the short exact sequence

0 A I C 0
where is the inclusion and C is its cokernel. Since I is -acyclic we have that
H p (X; A) ' H p1 (X; C)

p>1

and, by what we proved in the first part of the argument, H 1 (X; A) = 0. Also, C is flabby
because A and I are flabby. Thus, H 2 (X; A) ' H 1 (X; C) = 0. By iterating this argument
we conclude that H p (X; A) = 0 for all p > 0. Thus, we have proved that flabby acyclic.
There is a similar result with a similar proof for soft sheaves. It is slightly more complicated and requires that the space be paracompact.
9.15 Theorem. If X is paracompact, then for sheaves of R-modules on X
(i) if the first two terms of a short exact sequence are soft then so is the third;
(ii) flabby soft -acyclic;
(iii) fine soft -acyclic.
Proof. Let U X be open and let M be a sheaf of R-modules on X. Since X is paracompact, By Theorem 8.12, every section of a sheaf on a closed set Y extends to an open set
containing Y and, hence, extends to all of X if the sheaf is flabby. Thus, flabby soft.
Now suppose that A is a soft sheaf and

0 A B C 0
is an exact sequence of sheaves of R-modules. We wish to prove that

0 (X; A) (X; B) (X; C) 0

is also exact. Of course, we need only prove that : (X; B) (X; C) is surjective. To
this end, let c be a section of C over X and note that c is locally in the image of thus, there
is an open cover {Ui } of X, which we may assume locally finite since X is paracompact,
and elements bi (X; Ui ) such that (bi ) = c|Ui . Since paracompact spaces are normal,
we may choose an open cover Vi of X which has the property that Vi Ui . We then
consider the class of pairs (Y, b) where Y is a subset of X which is a union of some of the
sets in the collection Vi , b (Y ; B) and (b) = c|Y . Any set Y from such a pair is closed
due to the fact that the collection {Vi } is locally finite. This class is non-empty since it
contains each pair (Vi , bi ) and it is partially ordered under the relation: (Y1 , b1 ) < (Y2 , b2 )
if Y1 Y2 and g1 = g2 |Y1 . It also has the property that a maximal totally ordered subset
has a maximal element (by taking union). Thus, it follows from Zorns Lemma that there
is a maximal element (Y, b) in this class. If Y = X we are done. If not then there is an

80

J. L. TAYLOR

i so that Vi is not contained in Y . We then have (b|Y Vi bi |Y Vi ) = 0 and so there is

an element a0 (Y Vi ; A) such that (a0 ) = b|Y Vi bi |Y Vi . We use the fact that
A is soft to extend a0 to a section a of A on all of Vi . Then b and bi (a) agree when
restricted to Y Vi and, hence, define a section b0 (Y 0 ; B) where Y 0 = Y Vi . But then
(Y 0 , b0 ) is in our class and is larger than (Y, b), contradicting the maximality. This proves
the exactness of the above sequence. This result may be used to prove part (i) (Problem
9.6). Then to prove that A is acyclic we embed it in an injective I and use the long exact
sequence of cohomology for the short exact sequence

0 A I C 0
where is the inclusion and C is its cokernel. Since I is acyclic we have that
H p (X; A) ' H p1 (X; C)

p>1

and, by what we proved in the first part of the argument, H 1 (X; A) = 0. Also, C is soft
because A and I are soft. Thus, H 2 (X; A) ' H 1 (X; C) = 0. By iterating this argument
we conclude that H p (X; A) = 0 for all p > 0. Thus, we have proved that soft acyclic.
It remains to prove that fine soft. Thus, let Y X be closed and let M be a
fine sheaf on X. If s (Y ; M) then for each x Y the germ of s at x is represented
by a section defined in a neighborhood of x which agrees with s when restricted to that
neighborhood intersected with Y . Thus, we may choose an open cover {Ui } of X and
elements si (Ui ; M) such that s|Ui Y = si |Ui Y for each i - one of these open sets
will be the complement of Y and will have the zero section assigned to it and the others
will be neighborhoods of points of Y . Because X is paracompact, we may assume that
{Ui } is locally finite. Now because M is fine, we may
P choose for each i an endomorphism
i : M M supported in Ui in such a way that i i = id. For each i we interpret i si
to be a section
on all of X by extending it to be zero on the complement of Ui . We then
P
set s0 = i i si (X, M). That this sum makes sense follows from the local finiteness
of the open cover which means that in a neighborhood of any point we are summing only
finitely many non-zero terms. We also have that s0x = sx at each point x Y so that s0 is
an extension of s to all of X. It follows that M is soft. Thus, we have proved that fine
soft and completed the proof of the Theorem.
We conclude this chapter with some examples of acyclic resolutions which show that
certain classical cohomology theories are just examples of sheaf cohomology.
By C 0 we shall mean the sheaf of continuous functions on X. If X is the appropriate
kind of differentiable manifold, we shall denote by C p and C the sheaves of functions
with continuous partial derivatives up to order p and functions with continuous partial
derivatives of all orders, respectively.
9.16 Theorem. Each of C 0 , C p and C is a fine sheaf if X is a paracompact space and,
in the case of C p and C , X has the appropriate differentiable manifold structure.
Proof. Paracompact implies normal which implies that Urysohns Lemma holds. Urysohns
Lemma can be used to constuct continuous partitions of unity subordinate to any locally
finite open cover. A continuous function with support inside a given open set defines, by

NOTES ON SEVERAL COMPLEX VARIABLES

81

multiplication, an endomorphism of C 0 with support in the open set. Thus, a partition of

unity in the algebra of continuous functions on X defines a partition of unity for the sheaf
C 0 in the sense of definition 9.13(c). This proves that C 0 is a fine sheaf. The same result
for C p and C follows from the fact that on a C p or C manifold partitions of unity may
be chosen to be comprised of C p or C functions.
One of the nice things about the notion of fine sheaf is that if a sheaf of rings R is
fine then so is any sheaf of modules over R (Problem 9.7). In particular, any sheaf of
modules over C p , (p = 0, 1, , ) is fine and, hence, -acyclic. In particular, the sheaf
of continuous sections of a vector bundle on a topological space is a fine sheaf as is the
sheaf of C differential forms of degree p on a C -manifold.
Let X be a C -manifold and denote by E P the sheaf of C differential forms of degree
p on X. The de Rham complex of sheaves is the complex
d0

d1

dp1

dp

0 C E 0 E 1 E p
where dp is exterior differentiation and C here stands for the constant sheaf with stalks
C. The Poincare Lemma says that if X is any open ball in Rn then the corresponding
sequence of sections is exact (this is proved by constructing an explicit homotopy between
the identity and zero using integration along lines from the center of the ball). Since a C
manifold looks locally like a ball in Rn it follows that the de Rham complex is exact as
a complex of sheaves. Thus, it defines a resolution C E of the constant sheaf C by a
complex E of fine sheaves. On passing to global sections of E, we obtain the classical de
Rham complex E(X) of differential forms on X. The cohomology of this complex is called
the de Rham cohomology (with coefficients in C) of X. By theorem 9.8 we have:
9.17 Theorem. There is a natural isomorphism between the de Rham cohomology of a
C manifold X and the sheaf cohomology H(X; C) for the constant sheaf C on X.

We end this chapter with a discussion of Cech

cohomology and its relation to sheaf
cohomology. Let S be a sheaf of R-modules on X and let U = {Ui }iI be an open cover of
X. If = (i0 , , ip ) I p+1 is a multi-index then we set U = Ui0 Uip . Then for

each open
on U , for the cover U, is the direct
Q set U X, the module of Cech p-cochains
product I p+1 (U U, S) and is denoted C p (U, S)(U ). In other words, a p-cochain for
U on U is a function f which assigns to each I p+1 an element f () (U U ; S). If
V U then restriction clearly defines a morphism C p (U , S)(U ) C p (U, S)(V ) and, thus,
U C p (U, S)(U ) is a presheaf. In fact, it is clearly a sheaf. We denote it by C p (U, S).
The module of global sections of this sheaf is C p (U, S)(X). This is the classical space of

Cech
cochains for S and the cover U and will be denoted C p (U, S). We next define a
coboundary mapping
p : C p (U, S) C p+1 (U, S)
by
p f () =

p+1
X

(1)j f (j )|U U

j=0

where f C p (U , S)(U ) and j I p+1 is obtained from I p+2 by deleting its j th entry.

82

J. L. TAYLOR

9.18 Theorem. We have p+1 p = 0 so that

0

p1

0 C 0 (U, S) C 1 (U, S) C p (U , S)
is a complex of sheaves and sheaf morphisms.
Proof. For I p+3 let j,k denote the result of deleting both the j th and the k th entries
from . Then
(j )k = j,k if 0 k < j p + 2
and
(j )k = j,k+1

if 0 j < k p + 1

It follows from this that for f Cp (U , S)(U ) we have

p+1 p f () =

p+2
X

(1)j

j=0

"p+1
X

#
(1)k f ((j )k )|Uj U |U U

k=0

(1)j+k f (j,k )|U U +

k<j

X
(1)j+k f (j,k+1 )|U U
kj

=0
due to the fact that, in the middle line above, the second term is equal to the negative of
the first, which is evident from the change of variables j k, k + 1 j applied to the
second term and the observation that j,k = k,j .
The restriction maps S(U ) S(Ui U ) define a sheaf morphism : S C 0 (U , S)
whose composition with 0 is 0. In fact, much more is true:
Theorem 9.19. If U is an open cover of X and S a sheaf on X then the complex
0

p1

0 S C 0 (U , S) C 1 (U, S) C p (U, S)
is an exact sequence of sheaves.
Proof. Fix an x X. Let U be any neighborhood of x which is contained in a member of
U, say U Uk . Now suppose f C p (U, S)(U ) and p f = 0. Define g C p1 (U, S)(U ) by
g() = f ((k, ))
where we set (k, ) = (k, i0 , . . . , ip1 ) I p+1 for = (i0 , . . . , ip1 ) I p . Note that
g C p1 (U, S)(U ) due to the fact that U Uk which implies that Uk U U = U U .
Then for I p+1 ,
p

0 = f ((k, )) = f ()|Uk U U

p
X
j=0

= f ()

p1

g()

(1)j f ((k, j ))|Uk U U

NOTES ON SEVERAL COMPLEX VARIABLES

83

again due to the fact that Uk U U = U U . This shows that, locally, the kernel of p
is the image of p1 for p > 0. The same argument also works for p = 0 with 1 replaced
by . This proves the exactness of the above sequence.

The complex of Theorem 9.19 is called the Cech

complex of sheaves and is denoted
C(U, S). Its complex of global sections, (C(U, S)) = C(U, S), is what is classically called

the Cech
complex and we shall call the global Cech
complex for the sheaf S and the open
cover U.

p (U, S)}p0 of the sheaf S for the cover U is

9.20 Definition. The Cech
cohomology {H

the cohomology of the global Cech

complex C(U, S).
9.21 Theorem. For any sheaf S and any open cover U we have
0 (U, S) ' (X, S); and
(i) H
p (U, S) = 0 for p > 0.
(ii) if S is flabby, then H
Proof. By Theorem 9.19 we have an exact sequence of sheaves
0

0 S C 0 (U, S) C 1 (U, S),

which implies that

0 (X, S) C 0 (U, S) C 1 (U , S),

is exact and this implies statement (i).

The sheaf C p (U, S)of Cech

p-cochains is the direct product of the sheaves (i ) i1
S
where i : U X is the inclusion map. If S is flabby then so is its restriction, i1
S,
to
U

1
and the direct image of a flabby sheaf is flabby by Problem 9.8. It follows that (i ) i S
is flabby for each and, hence, that C p (U, S) is flabby for each p. Thus, the complex in
Theorem 9.19 is an exact sequence of flabby sheaves. Then Theorem 9.14 implies that the
complex obtained from it by applying (X, ) is also exact. Part (ii) follows.
Let S be a sheaf on X, U an open cover of X and S C(U, S) the corresponding

Cech
resolution as in Theorem 9.19. Let S I be an injective resolution of S. Then the
injectivity of the terms in I can be used, as in Theorem 9.2, to inductively construct a
morphism of complexes C(U, S) I for which the diagram
S C(U , S)

y
S
I
is commutative. This morphism is unique up to homotopy and, hence, after we apply
p (U, S) H p (X, S).
(X, ) it determines a well defined morphism H
The open cover U is called a Leray covering for the sheaf S if for each multi-index
the sheaf S is acyclic on U . For example, it follows from the Poincare lemma that the
deRham cohomology of any convex open set in Rn vanishes in positive degrees. From this
and Theorem 9.17 we conclude that the constant sheaf C is acyclic on any convex open
set in Rn . This implies that any open cover of Rn by open convex sets is a Leray cover for
the constant sheaf C.

84

J. L. TAYLOR

9.22 Theorem. If U is a Leray cover for the sheaf S then the natural morphism
p (U, S) H p (X, S)
H
is an isomorphism.
Proof. Choose an embedding S F of S in a flabby sheaf (an injective, for example).
Let G be the cokernel and consider the short exact sequence:
0 S F G 0
Then from the long exact sequence and the fact that S is acyclic on U we conclude that
0 S(U ) F(U ) G(U ) 0
is exact for each multi-index and from this that we have an exact sequence
0 C(U, S) C(U, F) C(U, G) 0

of global Cech
complexes. This, in turn, gives us a long exact sequence of Cech
cohomology,
which, along with the long exact sequence for sheaf cohomology, the induced morphisms

from Cech
to sheaf cohomology discussed above and the fact that F is flabby, gives us the
following commutative diagram with exact rows:
0 (U , S) H
0 (U , F) H
0 (U , G) H
0 (U, S) 0
0 H

y
y
y
y
0 (X, S) (X, F) (X, G) H 1 (X, S) 0
The first three vertical arrows are isomorphisms and, hence, so is the fourth. This establishes our result in the case p = 1. We prove the general case using the following
commutative diagram, which is also part of the the diagram which comes from the long
exact sequences associated to the above short exact sequence:
p (U, G) H
p+1 (U, S) 0
0 H

y
y
0 H p (X, G) H p+1 (X, S) 0
for p > 0. This has exact rows and so if the first vertical morphism is an isomophism so
is the second. Consider the class of all sheaves T such that the open cover U is a Leray
cover for T . From the long exact sequence for sheaf cohomology it follows that if the
first two sheaves in a short exact sequence belong to this class then so does the third. By
hypothesis, S belongs to this class and F does since it is flabby. It follows that G belongs
as well. Thus, if we have proved the theorem for all members of this class and for all

NOTES ON SEVERAL COMPLEX VARIABLES

85

degrees less thatn or equal to p then the above diagram shows that it is true for degree
p + 1 as well. By induction, the proof is complete.

What we have been discussing so far is Cech

cohomology for an open cover and a sheaf.

Cech cohomology for a sheaf without reference to a cover is obtained by passing to a limit
over covers. More precisely, if V is an open cover which is a refinement of the open cover U
then the restriction maps define a morphism of complexes of sheaves C(U, S) C(V, S). By
passing to the limit over the directed set of open covers of X we obtain a complex of sheaves

C(S), which we shall call the limit Cech

complex of sheaves, and a quasi-isomorphism
S C(S). The complex of global sections of C(S) is C(S) = C(S)(X) = lim C(U, S) and

is called the limit global Cech complex. Its cohomology is the Cech cohomology of S on
p (X, S)}. Clearly the morphisms H
p (U, S) H p (X, S) induce a
X and is denoted {H
p (X, S) H p (X, S).
morphism H
9.23 Theorem. If X is paracompact then the natural morphism
p (X, S) H p (X, S)
H
is an isomorphism.
Proof. . We shall prove that S C(S) is a resolution of S by soft sheaves. Since soft
sheaves are acyclic, the result will then follow from Theorem 9.8.
We have already remarked that S C(S) is a quasi-isomorphism in other words,
that the exactness of the sequences in Theorem 9.19 is preserved on passing to the direct
limit since direct limits generally preserve exactness. It remains to prove that each
C p (S) is a soft sheaf. Thus, let Y X be closed and suppose f is a section of C p (S)
over Y . Since X is paracompact and Y is closed, we have that f may be represented by a
section in a neighborhood U of Y . This section may, in turn, be represented by an element
f 0 C p (U, S)(U ) for some open cover U of X. We may choose a locally finite refinement
V of U with the property that each set in V is either contained in U or is contained in
X Y . We define a section g 0 of C p (V, S) on X by setting g 0 equal to the image of f 0
under the refinement map on those multi-indices for which all the correspondings sets in
V are contained in U and setting it equal to zero otherwise. On passing to the image g of

g 0 in the space of limit Cech

p-cochains, we obtain a global section of Cp (S) which has f
as its restriction to Y . Thus, Cp (S) is soft and the proof is complete.

When S is a constant sheaf G. The Cech

cohomology in the sense of this chapter is just

the classical Cech cohomology with coeficients in the group G from the theory of algebraic
topology.
We end this section with an example which shows how to solve one of the local to
global problems posed in chapter 8. This is the problem of finding a global logarithm for a
non-vanishing continuous function on X. We assume X is paracompact. Let C denote the
sheaf of continuous functions on X with addition as group operation and C 1 the sheaf
of invertible (non-vanishing) continuous functions with multiplication as group operation.
Then, due to the fact that a non-vanishing continuous function has a logarithm locally in
a neighborhood of each point, the sequence of sheaves:
exp

0 2iZ C C 1 0

86

J. L. TAYLOR

is exact. Since C is a fine sheaf and hence acyclic, we conclude from the long exact sequence
of cohomology for this short exact sequence that
1 (X, Z)
C 1 (X)/ exp(C(X)) ' H 1 (X, Z) ' H
where, of course, Z stands for the constant sheaf Z and we use the fact that 2iZ ' Z.
Thus, every non-vanishing continuous function on X has a global logarithm if and only if

the first Cech

cohomology of X with integral coeficients vanishes. More generally, there is
a epimorphism f [f ] from the group of non-vanishing continuous function on X to the

1 (X, Z) with the property that f has a global logarithm if and

Cech
cohomology group H
only if [f ] vanishes. This is an elementary but very instructive example of the use of sheaf
theory to analyze a problem involving passing from local to global solutions.

9. Problems
1.
2.
3.
4.

Prove Theorem 9.1.

Prove Theorem 9.2.
Prove that homZ (R, D) is an injective R-module if D is a divisible abelian group.
With MU and MY as defined in the discussion preceding Theorem 9.12, prove that
H p (X; MY ) = H p (Y ; M)
and

5.
6.

7.
8.
9.

H p (X; MU ) = Hp (U ; M)

where is the family of supports in U consisting of subsets of U which are closed in X.

Prove that if U is an open subset of X, M a sheaf of R-modules on X and R|U is the
extension by zero of the restriction of R to U , then hom(RU , M) ' (U ; M).
Finish the proof of Theorem 9.15 by proving that if the first two terms of a short exact
sequence of sheaves are soft then so is the third. Hint: you may use the fact, proved in
9.15, that if A is soft and 0 A B C 0 is a short exact sequence of sheaves on
X then (X; B) (X; C) is surjective.
Prove that if a sheaf of rings is fine then so is every sheaf of modules over this sheaf of
rings.
Prove that if f : Y X is any continuous map between topological spaces then f S is
a flabby sheaf on X if S is a flabby sheaf on Y .
Prove that H 2 (X, Z) is isomorphic to the group, under tensor product, formed by equivalence classes of complex line bundles over X.

NOTES ON SEVERAL COMPLEX VARIABLES

87

10. Coherent Algebraic Sheaves

Up to this point we have only worked with algebraic or holomorphic subvarieties of open
sets in Cn . It is time to define the general notions of algebraic and holomorphic varieties
and to introduce and study coherent sheaves on such objects.
10.1 Definition. A ringed space is a pair (X, R) consisting of a topological space X and
a sheaf of rings R on X. A morphism of ringed spaces is a pair (f, f # ) : (X, R) (Y, S)
consisting of a continuous map f : X Y together with a morphism of sheaves of rings
f # : S f R. The sheaf of rings R is called the structure sheaf of the ringed space
(X, R).
It is easy to see that a morphism (f, f # ) : (X, R) (Y, S) of ringed spaces is an
isomorphism (has a two sided inverse) if and only if f is a homeomorphism and f # is an
isomorphism of sheaves of rings.
If V is an algebraic (holomorphic) subvariety of a domain in Cn then, henceforth, V O
(V H) will denote the sheaf of regular (holomorphic) functions on V . That is, for each open
subset U V we let V O(U ) (V H(U )) be the space of regular (holomorphic) functions on
U.
A subvariety V (algebraic or holomorphic) of an open set (Zariski or Euclidean) in Cn
is a ringed space with the obvious structure sheaf (V O or V H).
10.2 Definition. A holomorphic variety is a ringed space which is locally isomorphic to
a holomorphic subvariety of an open set in Cn and whose topological space is Hausdorff
and second countable. If a holomorphic variety is locally isomorphic, as a ringed space, to
an open set in Cn then it is called a complex manifold.
The definition of algebraic variety is somewhat more complicated due to the need for a
condition which replaces Hausdorff. We begin by defining the notion of affine variety.
10.3 Definition. An affine variety is a ringed space which is isomorphic to an algebraic
subvariety of Cn .
In most of the algebraic geometry literature an affine variety is defined to be an irreducible subvariety of Cn , but this is too restrictive for our purposes.
10.4 Theorem. If V is an affine variety and f a regular function on V , we set Vf = {z
V : f (z) 6= 0}. Then the open subset Vf is also an affine variety.
Proof. Suppose V is an algebraic subvariety of Cn and consider the algebraic subvariety
W of Cn+1 defined by
W = {(z, zn+1 ) Cn+1 : 1 f (z) zn+1 = 0,

z V}

Then the map z (z, f (z)1 ) : Vf W is a morphism of ringed spaces (Problem 10.4)
with inverse (z, zn+1 ) z : W Vf . Thus, Vf is affine since it is isomorphic to W .
An algebraic prevariety is an ringed space which has a finite cover by open sets which
are affine varieties. Note that by Theorem 10.4 each open subset of an affine variety V
is, itself, a finite union of open sets which are affine varieties (sets of the form Vf ). Thus,

88

J. L. TAYLOR

an open subset of a prevariety is a prevariety. A closed subset of an affine variety is also

clearly an affine variety and so a closed subset of a algebraic prevariety is also an algebraic
prevariety.
We need one more condition to define the class of algebraic varieties. This is to eliminate
pathological examples like the following: Consider the space V which is two copies of the
complex plane glued together by the identity map everywhere except at 0. Under the
Zariski topology the topology in which open sets are complements of finite sets this is
a union of two open sets each of which is a copy of C with the Zariski topology. Thus, it is
an algebraic prevariety. However, it is a strange space since it contains two copies of the
origin and just one copy of every other point of the plane. Note that if f, g : C V are
the embeddings of the two copies of C into V then each is a morphism of ringed spaces
but {z C : f (z) = g(z)} = C 0 is not a closed subset of C. Thus, this space V will not
be an algebraic variety if we use the following definition:
10.5 Definition. An algebraic variety V is an algebraic prevariety with the property that
for any algebraic prevariety W and any pair of morphisms f : W V and g : W V it
is true that {w W : f (w) = g(w)} is a closed set.
It turns out that the above condition is equivalent to the condition that the diagonal in
V V is a closed set. However, one has to be careful how one defines the product of two
prevarieties V and W as a pointset it is V W but it does not have the cartesian product
topology. One defines the product of two affine varieties to be the pointset V W with
the Zariski topology determined by the tensor product O(V ) O(W ) of the corresponding
rings of regular functions. This ring can be regarded as a ring of functions on V W via
the map f g ((z, w) f (z)g(w)). Its localizations to Zariski open sets of V W
define a presheaf whose sheaf of germs is the sheaf of regular functions for V W . It is
easy to see that the product in this sense of two affine varieties is an affine variety. Once
the product of affine varieties is defined then one defines the topology on the product of
general algebraic prevarieties by using the cover by products of open affine subvarieties.
We leave the problem of proving that an algebraic prevariety V is an algebraic variety if
and only if the diagonal is closed in V V as a problem (Problem 10.2).
Using the coordinate functions and subtraction in Cn , it is easy to see that affine varieties
are algebraic varieties. In fact, the following theorem shows that there are lots of algebraic
varieties.
10.6 Theorem. If V is an algebraic prevariety with the property that for any two points
x and y in V there is an open affine subvariety of V containing both x and y, then V is
an algebraic variety.
Proof. Let f, g : W V be two morphisms from a prevariety W to V and let Z = {w
W : f (w) = g(w)}. For w W Z choose x = f (w), y = g(w) and let U be an open affine
subvariety of V containing both x and y. Then Q = f 1 (U ) g 1 (U ) is an open set in W
containing w. However, f and g both map Q into the affine variety U and so the subset
of Q on which they agree, Z Q, is closed in Q. Its complement in Q is, thus, an open set
containing w and missing Z and, hence, Z is closed.
We will denote the structure sheaf of any algebraic variety V by V O and of any holomorphic variety V by V H. The corresponding algebras of global sections will be denoted

NOTES ON SEVERAL COMPLEX VARIABLES

89

O(V ) and H(V ).

In what follows we will need a precise description of the global regular functions on an
affine variety. Recall that, by definition, a regular function f on a subvariety V of an open
set in Cn is a function with the property that for each point z V there is a neighborhood
U of z in Cn and a rational function on Cn with denominator non-vanishing on U such
that f agrees with this rational function on U V . However, it is a theorem that regular
functions on a subvariety V of Cn are actually restrictions to V of polynomials on Cn :
Theorem 10.7. If V is a subvariety of Cn and A = C[z1 , . . . , zn ]/ id(V ) is the algebra of
restrictions to V of polynomials on Cn , then
(i) as a topological space, V is the space of maximal ideals of A with its Zariski
topology;
(ii) if f A and Vf = {v V : f (v) 6= 0} then the algebra V O(Vf ) of regular functions
on Vf is the localization Af of A relative to the multiplicative set {f n : n 0};
(iii) in particular, the algebra, O(V ), of regular functions on V is the algebra, A, of
restrictions to V of polynomials on Cn ;
Proof. Each point v of V determines a maximal ideal Mv = {f A : f (v) = 0}.
On the other hand, if M is a maximal ideal of A, then M is generated by a finite set
f1 , , fm /inA. If there is a point v V where the functions fi all vanish, then M Mv
and, hence, M = Mv since M is maximal. Thus, to prove (1) it suffices to show that if
f1 , . . P
. , fm A have no common zeroes on V then there are elements g1 , . . . , gm A such
that
fi gi = 1; in other words, a finite set of functions with no common zero lies in no
maximal ideal.
To see that this condition holds, choose h1 , , hm C[z1 , , zn ] such that hj |V = fj
for j = 1, , n and choose hm+1 , , hk C[z1 , , zn ] to be a set of polynomials for
which V is exactly its set of common zeroes. Then the set h1 , , hk has no common
zeroes on Cn ; in other words,the ideal, I, which it generates has loc(I) = . It follows
from the Nullstellensatz that I = C[z1 , , zn ]. But this implies that the identity is in

P
I and, hence, in I. Thus, the equation
pj hj = 1 has a solution. When we restrict the
hj s to V , the last k m restrict to zero. Thus,P
the restriction to V of p1 , , pm gives
the required solution g1 , , gm to the equation
gi fi = 1. This shows that v Mv is
a bijection from points of V to maximal ideals of A. The topology of V is the relative
topology inherited from Cn . This is the topology in which closed sets are common zero
sets of families of polynomials. Thus, the closed subsets of V are common zero sets of
families of polynomials restricted to V . This is the Zariski topology induced on V by A.
Thus, we have proved part (i).
One direction of (ii) is trivial: Restriction to Vf defines a natural map Af V O(Vf ).
Suppose g/f k Af (g A) determines the zero section of V O over Vf . This means that
g = 0 on Vf which implies that f g = 0 in A which implies that g/f k = 0 in Af . Thus,
Af V O(Vf ) is injective. The surjectivity will be proved after we prove (iii).
We now prove (iii). Certainly the restriction to V of a polynomial on Cn is a regular
function on V and so A O(V ). Thus, we must show that the restriction map is surjective.
Suppose now that h O(V ). We can cover V with a collection {Ui } of open sets such that
f |Ui has the form pi /qi , O(Ui ) where pi and qi are elements of A and qi does not vanish

90

J. L. TAYLOR

on Ui . Since A is Noetherian we may assume the collection {Ui } is finite. Furthermore,

we may assume each set Ui is of the form Vg for some g A since these sets form a base
for the topology. In fact we may actually assume that Ui = Vqi = {v V : qi (V ) 6= 0}
since if this is not true it can be achieved by replacing pi and qi by rpi and rqi for an
appropriate r A. Then the condition that the pi /qi define a global section of V O is that
pi /qi = pj /qj on Vqi Vqj for each pair i, j. This means that (pi qj pj qi ) = 0 on Vqi Vqj
which, means, as in the previous paragraph, that its product with qi qj is the zero element
of A. Thus, if we set p0i = pi qi and qi0 = qi2 for each i, then p0i /qi0 = pi /qi in Aqi for each i
the qj0 have no common
and for each pair i, j the equation p0i qj0 = p0j qi0 holds in A. Since P
zeroes on V , by part (i) we may choose a set {gi } A such that
gi qi0 = 1. Then
X
X
qj
gi p0i = p0j
gi qi0 = p0j
P 0
gi pi A. Then clearly the image of h0 in O(V ) is h. Thus,
or qj0 h0 = p0j where h0 =
A O(V ) is surjective and (iii) is proved.
We showed above that Af V O(Vf ) is injective. To prove the surjectivity in (ii) we
simply apply (iii) to the image of Vf under the map z (z, f (z)1 ) : Vf Cn+1 . As
we showed in Theorem 10.4, this map is an isomorphism of Vf onto a subvariety W of
Cn+1 . The composition of this map with a polynomial on Cn+1 is the restriction to Vf of a
function of the form g/f k , where g is a polynomial on Cn . This shows that Af V O(Vf )
is surjective and finishes the proof of (ii).
There are many operations on sheaves which yield new sheaves. If M and N are sheaves
of modules over a sheaf of rings A, then U hom(M|U , N |U ) is a presheaf which yields
a sheaf homA (M, N ) when we pass to its sheaf of germs. One can define the tensor
product M A N of two sheaves of modules over a sheaf of rings A or the tensor product
M A N of two sheaves of modules over a fixed ring A. These sheaves are defined as
the sheaves of germs of the obvious presheaves. One such construction, localization, is the
key tool of algebraic geometry. We shall describe it in some detail. Suppose that A is a
sheaf of rings on a space X, A = (X; A) and M is an A-module. Then one can localize
M on X by constructing the sheaf A A M . This is the sheaf of germs of the presheaf
U A(U ) A M . Clearly, M A A M is a functor from the category of A-modules
to the category of sheaves of A-modules. While there is no reason to expect this functor
to have nice properties in general, there are many situations where it is very nice indeed.
What properties would we like it to have? We would like to be able to recover M from
A A M as (X; A A M ). It would be nice if M A A M were an exact functor
or, even better, an equivalence of categories with (X; ) as inverse. This is exactly what
happens for the localization functor of algebraic geometry. The version that we describe
below is a special case of a much more general theory.
Suppose V is an affine variety and M is a module over the algebra O(V ). Then we set
= V O O(V ) M
M
As noted above, this is the sheaf of germs of the presheaf U M (U ) = V O(U ) O(V ) M .
Now, a basis of neighborhoods for the topology of V is given by the sets of the form
is determined by the restriction
Vf = {v V : f (v) 6= 0} for f O(V ). Thus, the sheaf M
of the presheaf U M (U ) to sets of the form U = Vf .

NOTES ON SEVERAL COMPLEX VARIABLES

91

10.8 Theorem. Let V be an affine variety, M an O(V )-module and f an element of

O(V ). Then there is a natural isomorphism M (Vf ) Mf , where Mf is the localization
of M relative to the multiplicative set {f k : k 0}.
Proof. The localization, Mf , of M relative to the multiplicative set {f k : k 0} consists of
equivalence classes of expressions of the form m/f k where m M , with the obvious module
operations. Two such expressions m/f k and n/f j are equivalent if f p (f j m f k n) = 0 for
some p 0. Note that this is the same as the localization of M relative to the multiplicative
set consisting of functions in O(V ) which are non-vanishing on Vf . This is due to the fact
that if g O(V ) is non-vanishing on Vf then, by the Nullstellensatz, f is in the radical of
the ideal generated by g in O(V ), from which it follows that f k = hg for some k 0 and
some h O(V ). This implies that the two multiplicative sets {f k : k 0} {g O(V ) :
g(v) 6= 0 v Vf } yield the same localization.
As noted in Theorem 10.7 the ring V O(Vf ) is the localization of O(V ) relative to
k
{f : k 0}. We define the morphism
V

O(Vf ) O(V ) M = M (Vf ) Mf

by
g/f k m gm/f k
Its inverse is the morphism determined by
m/f k 1/f k m : Mf M (Vf )
It is clear that both morphisms are well defined and they are inverses of one another.
10.9 Theorem. Let V be an affine variety. Then
(i) for each open set U V of the form U = Vf , the functor M M (U ) =
V O(U ) O(V ) M is exact; that is, V O(U ) is a flat O(V )-module;
is exact.
(ii) the localization functor M M
Proof. Using the description of M (U ) as V O(U )O(V ) M makes it clear that this functor is
right exact. Thus, we need only prove that if N M is a submodule then N (U ) M (U ) is
injective. To prove this we use the fact, proved in the previous theorem, that M (U ) = Mf
and N (U ) = Nf if U = Vf . Thus, let n/f k represent an element of Nf and suppose that
it determines the zero element of Mf . This means that f p n = 0 for some p. But if this
equation holds in M it also holds in N and, hence, n/f represents the zero element of N
as well. This proves (i); however, (ii) is an immediate consequence of (i) and the fact that
sets of the form Vf form a basis for the topology of V
10.10 Theorem. If V is an affine variety and M is an O(V )-module then
) is
(i) on an open set of the form Vf the natural morphism Mf = M (Vf ) (Vf ; M
an isomorphism;
) is an isomorphism.
(ii) in particular, M (V ; M

92

J. L. TAYLOR

Proof. Statements (i) and (ii) are actually equivalent since, if V is affine, so is each Vf .
Thus, we will just prove (ii).
This is almost the same as the proof of Theorem 10.7 but there are a few differences.
over V . This means that we may
Suppose m M (V ) determines the zero section of M
(Ui ) for each i. Without
cover V with finitely many open sets Ui such that m|Ui = 0 in M
loss of generality we may assume that these sets are of the form Ui = Vqi where qi O(V ).
Then m|Ui = 0 means that for each i there is an integer ni such that qini m = 0. Since the
sets Ui cover V , the collection {qi } has no common zero on V . Since V is affine, Theorem
P ni
(V ). However, this
10.7 (i) implies there exists a set {gi }, such that
gi qi = 1 in M
ni
)
implies that m = 0 since m is killed by qi for every i. We conclude that M (V ) (V ; M
is injective.
). We can cover V with a finite collection {Ui = Vq } of
Now suppose that s (V ; M
i
) of an element mi /q ni , M
(Ui ),
basic open sets such that s|Ui is the image in (Ui ; M
i
where mi M . In fact by relabeling each qini as qi we may assume that s|Ui is the
) of an element of the form mi /qi . Since these elements fit together
image in (Ui ; M
to form a section over V , we may assume (after refining the cover if necessary) that
(mi /qi )|Ui Uj (mj /qj )|Ui Uj = 0. However, Ui Uj = Vqi qj and so this equality means
that there is a positive integer n so that
(qi qj )n (qj mi qi mj ) = 0
in M for each pair i, j. If we simply relabel qin mi by mi and qin+1 by qi , then the fractions
mi /qi dont change but the above equality becomes simply
qj mi qi mj = 0
Since
V is affine and the sets Ui cover V , we may find elements gi
P
gi qi = 1 on V . Then, the equation

O(Vf ) such that

X
X
qj (
gi mi ) = (
gi qi )mj = mj
P
holds in M . It says that mj = qj m in M where m =
gi mi . Then m is an element of M
which restricts to mj /qj on Uj for each j. In other words, s is the image of m under the
).
map M (V ; M
10.11 Definition. A sheaf M of X O-modules on an algebraic variety X is called a quasicoherent sheaf if each point of X is contained in an affine neighborhood V such that M|V
for some O(V )-module M . A quasi-coherent sheaf is called coherent if
is isomorphic to M
for each point of X this can be achieved with a module M which is finitely generated over
O(V ).
Note that the structure sheaf X O of an algebraic variety X is a coherent sheaf, since
on any affine open set V it is the localization to V of the ring O(X). It follows that direct
sums of copies of the structure sheaf are quasi-coherent and finite direct sums are coherent.

NOTES ON SEVERAL COMPLEX VARIABLES

93

10.12 Lemma. If V is an affine variety, f O(V ), and S a quasi-coherent sheaf on V ,

then
(i) if s (V ; S) and s|Vf = 0 then there exists a positive integer n so that f n s = 0;
(ii) if t (Vf ; S) then there exist a positive integer n such that f n t is the restriction
to Vf of a section in (V ; S);
(iii) restriction defines a natural isomorphism (V ; S)f (Vf ; S).
i
Proof. We may choose a finite collection of basic open sets {Vgi } such that, S|Vgi = M
for a O(Vgi )-module Mi for each i. If s (V ; S) then for each i we have that si =
i ) may be regarded as as element of Mi by Theorem 10.10. If s|V = 0
s|Vgi (Vgi , M
f
then (si )|Vf Vgi = 0 for each i. Since Vf Vgi = Vf gi this implies that the image of si in
(Mi )f is zero by Theorem 10.8. It follows that f n si = 0 in Mi for some n and each i. We
may choose n independent of i since the open cover {Vgi } is finite. Then f n s is a global
section of S which restricts to zero on each set in this cover and, hence, is the zero section.
This proves part (i).
Now suppose that t (Vf ; S). Then for each i, t|Vf Vgi may be regarded as an element
of (Mi )f using Theorems 10.10 and 10.8 again. Thus, for each i, t|Vf gi is a fraction with
the numerator the restriction of an element ti Mi and the denominator a power of f .
That is, there is an n so that ti Mi and f n t (V, S) agree when restricted to Vf gi .
The integer n may be chosen independent of i. Now on Vgi Vgj = Vgi gj we have two
sections of S which agree on Vf gi gj since they both agree with the restriction of f n t. By
part(i) there is an integer m such that f m (ti tj ) = 0 on Vgi gj . Again, we may choose m
large enough to work for all i, j. But this means that the sections f m ti on Vgi agree on
intersections and, thus, define a global section s of S. Clearly, the restriction of s to Vf is
f n+m t. This completes the proof of (ii).
Part(iii) means exactly that (i) and (ii) hold.
10.13 Theorem. If V is an affine variety, S is a quasi-coherent sheaf on V and M =
(V, S), then there is a natural isomorphism
S
M
of sheaves of V O-modules. Furthermore, S is coherent if and only if M is finitely generated.
Proof. Suppose that Vf is a basic affine open set set for which S|Vf is the localization of
a V O(Vf )-module. Then necessarily it is the localization of S(Vf ) = (Vf ; S) by Theorem
S is defined on such an open set as follows: We have natural
10.10. The map M

isomorphisms M (Vf ) Mf by Theorem 10.10 and Mf = S(V )f S(Vf ) by Theorem

10.12. The composition gives us our isomorphism on sets of the form Vf . It clearly
commutes with restriction from one set of this form to a subset of this form and, since
open sets of this kind, form a basis for the topology of V , this defines an isomorphism of
S.
sheaves of V O-modules from M
If M is finitely generated, then S is coherent by definition since it is then the localization
of a finitely generated module. On the other hand, if S is coherent then it is locally the
localization of a finitely generated module. That is, we may cover V with basic open sets

94

J. L. TAYLOR

Vfi , such that for each i, S|Vfi is the localization of a finitely generated module necessarily
isomorphic to Mfi . In other words, there is a finite set fi O(V ) with no common zeroes
on V and with the property that Mfi is finitely generated for each i. This implies that M
is finitely generated (Problem 10.5). This completes the proof.
is an equivalence of
10.14 Corollary. If V is an affine variety, then the functor M M
categories from the category of O(V )-modules to the category of quasi-coherent sheaves of
V O-modules on V . The functor (V ; ) is its inverse functor.
and S (V ; S)
Proof. By Theorems 10.10 and 10.13, the composition of M M
in either order is naturally isomorphic to the identity. Thus, each is an equivalence of
categories and they are inverses of one another.
Our next major result is that quasi-coherent sheaves on an affine variety are -acyclic.
This will follow easily from Theorem 10.16. But first we need to prove the following
technical lemma:
10.15 Lemma. If A is a Noetherian ring, I an injective A-module and K an ideal of
A, then the submodule J I defined by J = {x I : K n x = 0 for some n} is also
injective.
Proof. To prove that J is injective, it suffices to prove that if N M are A-modules with M
finitely generated, then every morphism N J extends to a morphism M J (Problem
10.6). Thus, suppose that : N J is such a morphism. Then since (N ) J and N
is also finitely generated, we may choose a fixed n such that (K n N ) = K n (N ) = 0. By
Krulls Theorem (Problem 10.7), there is an integer m such that K m M N K n N . Thus,
factors through N N/(K m M N ). Since I is injective, the map N/(K m M N )
J I induced by extends to a morphism : M/K m M I. However, the image of lies
in J since it is, necessarily, killed by K n . Then the composition of with M M/K n M
is the required extension of . This completes the proof.
10.16 Theorem. Let V be an affine variety. If I is an injective module over the ring
O(V ), then I is a flabby sheaf on V .
Proof. We first show that for any f O(V ) the natural map I If is surjective. To this
end, let x/f n be an element of If , with x I and n a non-negative integer. We consider
the morphism f n+1 g f gx : f n+1 O(V ) I. This is well defined since if f n+1 g is zero
then so is f g and, hence, f gx. Since I is injective, this morphism extends to a morphism
: O(V ) I such that (f n+1 g) = f gx. Then f n+1 y = f x if y = (1). However, this
implies that x/f n is the image of y under the localization map I If . Thus, I If is
(U ; I)
is surjective in the case
surjective. This proves that the restriction map (V ; I)
where U V is an open subset of the form Vf .
(U ; I)
is surjective if U is an
To complete the proof we must show that (V ; I)
If
arbitrary open subset of V . Let Y be the subvariety of V which is the support of I.
Y U = then we are through since the only section of I over U is then zero. Suppose
Y U 6= . Then there is an open set of the form Vf U such that Y Vf 6= . If
then by the first paragraph, the restriction of s to Vf is also the restriction
s (U ; I),
has its support in
to Vf of a global section t. Then s = t|U + r where r (U ; I)

NOTES ON SEVERAL COMPLEX VARIABLES

95

Suppose we can show

Z = V Vf which is the zero set of f . That is, r Z (U ; I).
0
Z (U ; I)
is surjective. Then r = r |U for a section r0 Z (V ; I)
and
that Z (V, I)
0
0

s = (t + r )|U for the global section t + r (V ; I).

Now if J = {x I : f n x = 0 for some n}, then J is the subsheaf of I consisting of
sections killed by some power of f and this is exactly the subsheaf of sections with support
= (V ; J)
and Z (U ; I)
= (U ; J).
The support of J is contained
in Z. Thus, Z (V ; I)
in Y Z = Y Vf which is a proper subvariety of Y . Furthermore, J is also an injective
O(V )-module by Lemma 10.15. Thus we have reduced our problem to the same problem
for a different injective module - one with smaller support. Since subvarieties of an affine
variety satisfy the decending chain condition, we may reduce the problem, by induction,
to the case where the support is a single point. However, any sheaf supported on a single
point is flabby (Problem 10.8). This completes the proof.
10.17 Theorem. Let V be an affine algebraic variety. Then H p (V ; S) = 0 for p > 0 and
for all quasi-coherent sheaves S on V .
Proof. We set M = (V ; S) and choose an injective resolution M I of M . On localizing
I
this, and using Corollary 10.14 and Theorem 10.16, we obtain a resolution S = M
of S by flabby sheaves. Thus, we obtain the cohomology of S by taking global sections of
I and then taking cohomology of the resulting complex. However, by Theorem 10.14, we
Thus,
simply get back the resolution M I when we apply global sections to S I.
p
H (V : S) = 0 for all p > 0.
10.18 Definition. On an algebraic variety, a sheaf of submodules (ideals) of the structure
sheaf V O is called an ideal sheaf. If an ideal sheaf is coherent as a sheaf of modules then
it is called a coherent ideal sheaf. Any subvariety Y V determines an ideal sheaf the
sheaf of sections of V O which vanish on Y . This is called the ideal sheaf of Y and is often
denoted IY .
10.19 Theorem. Let X be an algebraic variety and Y X a closed subvariety. Then
the ideal sheaf IY is a coherent ideal sheaf.
Proof. We may cover X with affine open sets V . For each such V , the algebra
the quotient of a polynomial algebra and, hence, Noetherian. Thus, the ideal
I = IY (V ) = {g X O(V ) : g(z) = 0

X O(V

) is

z Y V }

is finitely generated. We claim that IY (Vf ) = If for any basic open set Vf V with
f X O(V ). In fact, an element of IY (Vf ) is a function on Vf of the form g/f n , with
g X O(V ), which vanishes on Y Vf . Then f g X O(V ) vanishes on Y V and, hence,
belongs to I. Then, on Vf we have
g/f n = gf /f n+1 If .
This proves that, on V , the ideal sheaf I is the localization I of the finitely generated ideal
I. Since X may be covered by such sets V , we have proved that the ideal sheaf is coherent.
Obviously, Corollary 10.14 implies that on an affine variety a coherent sheaf of ideals I
is the localization of an ideal, (V ; I), of O(V ) and, conversely, an ideal I O(V ) is the

ideal of global sections of a coherent sheaf of ideals, I.

96

J. L. TAYLOR

10.20 Theorem. Let V be an algebraic variety. Then the following statements are equivalent:
(i) V is affine;
(ii) H p (V ; S) = 0 for p > 0 and for all quasi-coherent sheaves S on V ;
(iii) H 1 (V ; S) = 0 for all coherent S sheaves on V .
Proof. That (i) implies (ii) is Theorem 10.17 and the implication (ii) implies (iii) is trivial;
thus, to complete the proof we must prove that (iii) implies (i).
Assume (iii) holds. The strategy is to prove that V is the space of maximal ideals of the
algebra O(V ) with the Zariski tolology and with its structure sheaf given by localization
of O(V ). Then we will prove that O(V ) is finitely generated and, hence, is the algebra
O(W ) for some affine variety. Necessarily, then, V is isomorphic to the affine variety W
by Theorem 10.7.
For x V let U be an affine open set containing x and set Y = V U . Consider the
exact sequence of sheaves
0 IY {x} IY C{x} 0
where IY {x} and IY are the ideal sheaves in V O of the subvarieties Y {x} and Y ,
respectively and C{x} is the skyscraper sheaf which is C at x and zero at all other points.
Since the ideal sheaf IY {x} is coherent, H 1 (V ; IY {x} ) = 0 and the long exact sequence
of cohomology implies that
0 (V ; IY {x} ) (V ; IY ) (V ; C{x} ) 0
is also exact. This just means that there is a function f O(V ) which vanishes on Y
and does not vanish at x. In other words, every affine neighborhood U of x contains a
neighborhood of x of the form Vf where f is a global section in O(V ). This implies, in
particular, that the functions in O(V ) separate points in V . It also implies that the open
sets Vf form a basis for the topology of V and, hence, that V has the Zariski topology
determined by the algebra O(V ) that is, the topology in which the closed sets are the
common zero sets of finite sets of functions from O(V ). It also implies that we may cover
V with a finite collection of affine open sets {Vfi }ni=1 with fi O(V ) for each i.
We claim that every maximal ideal of the algebra O(V ) has the form id({x}) = {f
O(V ) : f (x) = 0} for some point x of V . In fact if M is a maximal ideal which does
not have this form then there is no point of V at which all functions in M vanish. This
means the the collection on open sets {Vf : f M } covers V and this, in turn, implies
that some finite subcollection covers V by Problem 10.3. In other words, there is a finite
set {fi } M with no common zero on V . Given any such set {fi } consider the sheaf
morphism
n V O V O
P
defined by (g1 , . . . , gn ) i gi fi . This map is surjective, since at each x in V some fi
is non-vanishing and, hence, invertible in some neighborhood. The kernel of this map is
coherent (Problem 10.9) and, thus, has vanishing first cohomology. It follows from the long

NOTES ON SEVERAL COMPLEX VARIABLES

97

exact sequence of cohomology that the map induced on global

P sections is also surjective
and, hence, that we may solve for g1 , . . . , gn O(V ) such that gi fi = 1. This contradicts
the assumption that {fi } is contained in a maximal ideal and establishes the claim.
Since we have proved that the functions in O(V ) separate points of V , we now have
that V is exactly the set of maximal ideas of O(V ) as a point set and as a topological
space if the maximal ideal space is given the Zariski topology.
Next, we show that for any affine open set of the form Vf we have V O(Vf ) ' O(V )f .
Suppose g O(V ) and g|Vf = 0. Then g restricts to zero in each of the open sets Vf Vfi .
Since the Vfi are affine, it follows from Theorem 10.7 that there is an integer n such that
f n g restricts to be zero in Vfi for each i. However, this means that f n g = 0 in O(V ).
Now suppose that h V O(Vf ) and hi is its restriction to Vf Vfi . Then, also by Theorem
10.7, there is an integer n so that, for each i, f n hi = gi for some section gi V O(Vfi ).
Then gi and gj agree on the intersection Vfi Vfj for each pair i, j and so the gi define a
global section g. Clearly h = g/f n on Vf . Thus, V O(Vf ) is the localized algebra O(V )f .
This proves that the structure sheaf V O of V is just the sheaf obtained by localizing the
algebra O(V ).
Next, we show that the algebra O(V ) is finitely generated. Let {Vfi } be an affine open
cover, as above, with fi O(V ) for each i. We know that each algebra V O(Vfi ) = O(V )fi is
finitely generated since it is a quotient of a polynomial algebra by Theorem 10.7 and the fact
that each Vfi is affine. Thus, we may choose an integer k and elements hij O(V ) such that
for each i,P
the set {hij /fik }j generates V O(Vfi ). We may also choose elements gi O(V )
such that
fi gi = 1. Let A be the subalgebra of O(V ) generated by {fi } {gi } {hij }.
Then, for each i we have Afi = V O(Vfi ). Thus, if g O(V ) then we may choose an integer
n and elements pi A such that fin g = pi in O(V ). However, the fact that the equation
P
fi gi = 1 holds in A implies that the set {fi } is contained in no maximal ideal of A and
this implies that
set {fin } is contained in no maximal ideal of A. Hence, we may solve
P the
the equation
gi0 fin = 1 for gi0 A. Then
X
X
g=g
gi0 fin =
gi0 pi A
Thus, A = O(V ) and O(V ) is finitely generated.
Finally, since O(V ) is finitely generated, it is the quotient of a polynomial algebra and,
hence, is isomorphic to O(W ), where W is an affine variety. Now V is the maximal ideal
space of O(V ) with the Zariski topology and the structure sheaf obtained by localization
of O(V ). But W is the maximal ideal space of O(W ) with the Zariski topology and the
structure sheaf obtained by localization of O(W ). Since the two algebras are isomorphic,
V and W are isomorphic as algebraic varieties.

10. Problems
1. Prove that the product V W of two prevarieties and the projection morphisms V :
V W V and V : V W V have the following universal property: For each
prevariety Q and pair of morphisms f : Q V and f : Q W there is a morphism
h : Q V W such that f = V h and g = W h.

98

J. L. TAYLOR

2. Prove that an algebraic prevariety V is an algebraic variety if and only if the diagonal
is closed in V V .
3. Prove that every open cover of an algebraic variety has a finite refinement.
4. Prove that if V is a subvariety of an open set in Cn , W is a subvariety of an open set
in Cm and f : V W is algebraic (has coordinate functions which are regular) then f
is a morphism of ringed spaces.
5. Prove that if V is an affine variety, {fi } a finite set of elements of O(V ) which have no
common zero on V and M is an O(V )-module such that Mfi is finitely generated for
each i, then M is finitely generated.
6. Use a trascendental induction argument to prove that I is injective if and only if for
every singly generated module M and every submodule N M , any morphism N I
has an extension to M .
7. Prove Krulls Theorem: If A is a Noetherian ring, K an ideal of A, M a finitely generated
A-module and N M a submodule, then for each positive integer n there is a positive
integer m such that K m M N K n N .
8. A sheaf supported on a single point is called a skyscraper sheaf. Prove that every
skyscraper sheaf is flabby.
9. Prove that the kernel, image and cokernel of a morphism between quasicoherent (coherent) sheaves on an algebraic variety is also quasi-coherent (coherent).

NOTES ON SEVERAL COMPLEX VARIABLES

99

11. Dolbeault Cohomology

In sheaf theory, a vanishing theorem is a theorem which asserts that cohomology vanishes for some class of sheaves in some range of degrees p (usually p > 0). Vanishing
theorems generally insure the existence of global solutions to certain locally solvable problems. Theorem 10.17, which says that cohomology vanishes in degree greater than zero
for all quasi-coherent sheaves on an affine variety, is an example of a vanishing theorem.
Its proof is fairly elementary. We shall find it much more difficult to prove the analogous
vanishing theorem for sheaves of modules over a sheaf of rings of holomorphic functions.
Ultimately, we want to define the class of coherent analytic sheaves and the class of Stein
spaces and prove that the cohomology of a coherent analytic sheaf on a Stein space vanishes in positive degrees. Stein spaces are the holomorphic analogues of affine varieties.
However, it is considerably more difficult than it was in the algebraic case just to define
and develop the elementary properties of coherent analytic sheaves. This is, in part, due
to the fact that, although the local ring H is Noetherian, for every open set U the ring
H(U ) fails to be Noetherian. We will carry out this development in the next chapter.
In this chapter we prove the one vanishing theorem for holomorphic sheaves that can be
proved without a great deal of work Dolbeaults Theorem. This is a vanishing theorem
for the structure sheaf n H of Cn as a complex manifold and it involves constructing a
resolution of n H by a complex of sheaves of differential forms which is analogous to the
deRham complex.
On a smooth (C ) manifold X, a (complex) differential p form is a smooth section of
the vector bundle which assigns to each x X the space of skew symmetric complex pmultilinear forms on the (complexified)
tangent space of X at x. On an open set U in Rn ,

where we may choose bases

for the tangent space and {dxi } for the cotangent space
xi
which correspond to a basis {xi } for Rn , the typical differential p-form may be written as
=

i1 ...ip dxi1 dxip

where i1 ...ip C (U ). We denote the C -module of differential p-forms on U by E p (U ).

The correspondence U E p (U ) is a sheaf, E p , of C modules. Exterior differentiation
d : E p E p+1 is defined by
d(i1 ...ip dxi1 dxip ) =

X i1 ...ip
i

xi

dxi dxi1 dxip

It follows from the Poincare Lemma that E = {E P , dp } is a complex which provides a fine
resolution C E of the constant sheaf C.
If U is an open set in Cn , then we consider it as an open subset of the real vector
space R2n with basis {x1 , . . . , xn , y1 , . . . , yn }. However, instead of the usual bases for the
complexified tangent and cotangent spaces we use the basis consisting of
dzi = dxi + i dyi ,

d
zi = dxi i dyi ,

i = 1, . . . , n

100

J. L. TAYLOR

for the complexified cotangent space and

= 1/2
i
,
= 1/2
+i
,
zi
xi
yi
zi
xi
yi

i = 1, . . . , n

for the complexified tangent space. Note that these are dual bases to one-another. In terms
of the above basis for the cotangent space, a differential form in E r (U ) may be written as
X
ji ...jp k1 ...kq dzj1 dzjp d
zk1 d
zkq
p+q=r

with coeficients ji ...jp k1 ...kq C (U ). If we let E p,q (U ) denote the differential forms in
E p+q (U ) which are of degree p in the dzi and of degree q in the d
zi , then we have a direct
sum decomposition:
X
E p,q (U ).
E r (U ) =
p+q=r
p,q

Forms in the space E (U ) are said to have bidegree (p, q) and total degree p + q.
When restricted to E p,q , exterior differentiation defines a map
d : E p,q E p+1,q + E p,q+1 .
If we define and to be the operators which, on forms of bidegree (p, q), act as d followed
by projection on E p+1,q and E p,q+1 , respectively, then
: E p,q E p+1,q , : E p,q E p,q+1
and

d = + .

+ 2 . After sorting out terms of different bidegree, it

Note that 0 = d2 = 2 + +
follows that
= 0, 2 = 0.
2 = 0, +
Also, since

d( ) = d + (1)r d,

it follows that
( ) = + (1)r ,

) =
+ (1)r

for E r (U ) and an arbitrary form.

The operator is given explicitly by
j ...j k ...k dzj dzj d
(
zk1 d
zkq )
i
p 1
q
1
p
X ji ...jp k1 ...kq
=
d
zi dzj1 dzjp d
zk1 d
zkq .
zi
i
In particular, for f E 0,0 (U ) = C (U ),
=
f

X f
d
zi
zi
i

= 0 if and only if f is holomorphic

and so the Cauchy-Riemann equations assert that f
in U .

NOTES ON SEVERAL COMPLEX VARIABLES

101

11.1 Definition. For a domain U in Cn the pth Dolbeault complex is the complex

0 E p,0 (U ) E p,1 (U ) E p,n (U ) 0

Its q th cohomology group is called the (p, q)-Dolbeault cohomology of U and is denoted
Hp,q (U ).
Note that the (p, 0)-Dolbeault cohomology is just the space of p-forms in dz1 , . . . , dzn
with holomorphic coeficients. This is called the space of holomorphic p-forms and is denoted Hp (U ). The correspondence U Hp (U ) is clearly a sheaf and will be denoted Hp .
Note that H0 is the sheaf of holomorphic functions H.
Our main objective in this chapter is to prove that Dolbeault cohomology vanishes
for q > 0 if U is a polydisc. Among other things, this will imply that for each p, as a
complex of sheaves, the Dolbeault complex is exact except in degree zero and provides a
fine resolution of the sheaf of holomorphic p-forms. In particular, when p = 0 the Dolbeault
complex provides a fine resolution of the sheaf of holomorphic functions and, hence, can
be used to compute its sheaf cohomology.
The first step is to prove the generalized Cauchy integral theorem:
11.2 Theorem. Let U be an open subset of C bounded by a simple closed rectifiable
and z U , then
curve . If f is a C function in a neighborhood of U
Z
ZZ
1
d
1
f () d d
f (z) =
f ()
+
.
2i
z
2i
z
U

Proof. Note that

f ()
d

f () d d
d f ()
=
d d =
z
z
z
Thus, if r is the boundary of the disc D(z, r) and if r is chosen small enough that this
disc is contained in U , then Stokes Theorem implies that
ZZ
Z
Z
f () d d
d
d
=
f
()

f
()
z
z
z

r
Ur

where Ur = U D(z, r). Now ( z)1 is integrable on any bounded region of the plane,
as is easily seen by integrating its absolute value using polar coordinates centered at z.
Thus,
ZZ
ZZ
f () d d
f () d d
lim
=
.
r0
z
z
Ur

Also,

d
lim
= lim
f ()
r0
z r0
r
The result follows.

Z
0

f (z + reit ) idt = 2if (z)

102

J. L. TAYLOR

11.3 Theorem. If f C (U ) for an open set U C containing a compact set K then

there exists a neighborhood V of K with V U and a g C (V ) such that g/ z = f
in V .
Proof. We modify f so that it is actually C on all of C with compact support in U by
multiplying it by a C function which is one in a neighborhood V of K and has compact
support in U and then extending the resulting function to be zero on the complement of
U . Then the integral
ZZ
d d
g(z) =
f ()
z
C

is defined for all z C and defines a function g C (C). We calulate the derivative g/ z
of g using the change of variables + z:
ZZ

1
d d
g(z) =
f ( + z)
z
2i z

ZZ
ZZ
1
f ( + z) d d
1
f ( + z) d d
=
=
2i
z

2i

ZZ

f () d d
1
=
= f (z)
2i
z
where the last line follows from reversing the change of variables and using the generalized
Cauchy integral theorem on U (recall that f vanishes on the complement of a compact
subset of U and so the line integral in Theorem 10.2 vanishes). Thus, g/ z = f on all of
C. Of course, we modified f on the complement of V and so this equation holds for our
original f only on V , but this is what was to be shown.
If K Cn is compact, we denote by E p,q (K) the space (K; E p,q ) of C forms of
bidegree (p, q) defined in a neighborhood of K. The Dolbeault cohomology Hp.q (K) for K

is then the cohomology of the complex {E p,q (K), }.

is a compact polydisc in Cn then Hp,q ()
= 0 for q > 0
11.4 Dolbeaults Lemma. If
and for each p.
= 0. We must show that
be a form such that
Proof. Suppose q > 0 and let E p,q ()
p,q1

= for some form E

(). Let k be the least integer such that the expression
for involves no conjugate differential d
zj with j > k; that is, can be written in terms
of the conjugate differentials d
z1 , , d
zk and the differentials dz1 , . . . , dzn . We proceed
by induction on k. If k = 0 then = 0 since q > 0 and, hence, there is nothing to prove.
Thus, we assume that k > 0 and the result is true for integers less than k. We write as
= d
zk +
where and involve only the conjugate differentials d
z1 , d
zk1 . Then
= d
+

0 =
zk

NOTES ON SEVERAL COMPLEX VARIABLES

103

d
If j > k then no cancellation can occur between terms of
zk involving d
zj and

terms of involving d
zj . It follows that such terms individually vanish and, hence, the
coeficients of and are holomorphic in the variables zk+1 , , zn . Now it follows from
which
Theorem 11.3 that if f is a coeficient of then f = g/ zk for some g C ()
is also holomorphic in the variables zk+1 , , zn . That the solution g given by Theorem
10.3 is actually C in all the variables and not just in k and that it is holomorphic in
the variables zk+1 , , zn follows from the fact that these things are true of f and the
solution g is given in terms of f by an explicit integral formula which commutes with the
differential operators in question. By replacing each coeficient f of by the corresponding
g as above, we obtain a (p, q 1)-form with the property that
= + d

zk
=
where is a form involving only the conjugate differentials d
z1 , , d
zk1 . Then

involves only the conjugate differentials d

z1 , , d
zk1 and is closed. That is,

=
for
( ) = 0. Thus, by the induction hypothesis, we conclude that
p,q1
p,q1

some E
(), from which it follows that = with = + E
(). This
completes the proof.
be a polydisc. It
Note that it was not really important in the above argument that
was important that it be a Cartesian product that is, a set of the form K1 K2 Kn
for some collection of compact sets Ki C. This is due to the fact that the solution was
obtained by applying Theorem 11.3 in each variable separately while treating the other
variables as parameters.
The following theorem concerns an open polydisc = (, r). Note that we allow
some or all of the radii ri to be infinite, Thus, Cn itself is an open polydisc.
11.5 Theorem. Let be an open polydisc. Then Hp,q () = 0 for q > 0 and for all p.
j j+1
Proof.
of open polydiscs with compact closure such that
S Let j be a sequence
p,q
p,q1

() such
and j j = . If E () and = 0 then we will construct E

j . We first take care of

that = inductively using Dolbeaults Lemma on the sets
the case q > 1 which is different and considerably easier than the case q = 1.
i ),
If q > 1 we inductively construct a sequence of forms {j } such that j E p,q1 (

j = on j and j+1 |j = j . Clearly this will give the desired result, since we can
then define a solution E p,q1 () by |j = j . Suppose the sequence {j } has been
constructed with the above properties for j < k. Then we use Dolbeaults Lemma to find
= in a neighborhood of
k ) such that
k . We then have
E p,q1 (
k1 ) = 0
(
k1 and, since q > 1, we may apply Dolbeaults Lemma again to
in a neighborhood of
p,q2
= k1 in a neighborhood of
k1 . By multiplying
find E
(k1 ) such that
k1 and has compact support in
by a C function which is one in a neighborhood of
k1 , we may assume that is actually
an appropriate slightly larger neighborhood of
p,q1
p,q2

in E
(k ). Then k = E
(k ) gives the required next function in our
sequence since
k = (
)
=

104

J. L. TAYLOR

on k and k = k1 on k1 . This completes the proof in the case q > 1.

In the case q = 1 we use the sequence {j } as before, but this time we inductively
j = on
j ) such that
j and ||j j+1 ||j < 2j
construct a sequence j E p,q1 (
j of the coeficients of the
for each j. Here, ||||j is the sum of the supremum norms on
form . Suppose such a sequence {j } has been constructed for all indices j < k. We use
= in a neighborhood of
k ) such that
k . As
Dolbeaults Lemma to find E p,0 (
before,
k1 ) = 0
(
k1 . This means that k1 has coeficients which are holomorphic
in a neighborhood of
k1 . If we represent these coeficients as convergent power series
in a neighborhood of
k1 , then it is clear that we may choose a form
about the point which is the center of
, with polynomial coeficients, such that || k1 ||k1 < 2k+1 . Then k =
k = in a neighborhood of k and ||k k1 ||k1 < 2k+1 .
has the properties that
Thus, by induction, we may construct the sequence {j } as claimed. Now on a given k
consider the sequence {(j )|k }
j=k . This is a Cauchy sequence in the norm || ||k since
||j+1 j ||k < ||j+1 j ||j < 2j for j k. Furthermore, the terms of this sequence
differ from the first term by forms with holomorphic coeficients (since the difference is killed
Thus, the sequence may be regarded as a fixed form plus a uniformly convergent
by .
sequence of forms with holomorphic coeficients. It follows that this sequence actually
converges in the topology of E p,q1 (k ) the topology in which all derivatives of all
coeficients converge uniformly on compact subsets of k . Now since this is true for each k,
= . This completes
the limit determines a form E p,q1 () which clearly satifies
the proof.
Note that, as with the previous Lemma, it was not really important in the above argument that be a polydisc. It was important that it be a Cartesian product that is, a
set of the form U1 U2 Un for some collection of open sets Ui C.
Clearly, the Dolbeault complex and Dolbeault cohomology can be defined on any complex manifold, since the division of E r into bigraded terms E p,q is independent of the choice
of complex coordinate system (Problem 11.1).
11.6 Corollary. If X is any complex manifold, then for each p there is a natural isomorphism Hp,q (X) H q (X; Hp ) between Dolbeault cohomology on X and sheaf cohomology
of Hp on X. In particular, H0,q (X) is isomorphic to the sheaf cohomology H q (X; H) of
the sheaf of holomorphic functions.
Proof. Since a complex manifold has a neighborhood base consisting of sets which are
biholomorphic to open polydiscs, it follows from Theorem 11.5 that the complex
0 Hp E p,0 E p,1 E p,n 0
is exact. Since each E p,q is a sheaf of C -modules, this sequence is a fine resolution of HP
and, hence, may be used to compute its sheaf cohomology.
11.7 Corollary. If Cn is an open polydisc, then for each p the sheaf cohomology
H q (; Hp ) vanishes for q > 0.

NOTES ON SEVERAL COMPLEX VARIABLES

105

11.8 Corollary. If X is a complex manifold of dimension n, then for each p the sheaf
cohomology H q (X; Hp ) vanishes for q > n.

11. Problems
1. Give a coordinate free definition of the space E p,q .
2. Prove that if is an open polydisc and f H() then a function g H() belongs
to the ideal generated by f in H() if and only if its germ at belongs to the ideal
generated by the germ of f at in H for each f 1 ({0}).
3. Prove Hartogs extension theorem: If K is a compact subset of an open set U Cn ,
U K is connected and n > 1, then each function f which is holomorphic in U K
extends to be holomorphic in U . Hint: Let K 0 be chosen so that K 0 is a compact set
containing K in its interior, K 0 U and U K 0 is connected. Then multiply f by a C
function which is zero in a neighborhood of K and one in a neighborhood of U K 0 .
The resulting function then extends to a C function g in U which agrees with f on
the connected set U K 0 . Now use Corollary 11.7 and Problem 1.5 to show that you
can find a C function h on U which vanishes on U K 0 and is such that g h is
holomorphic in all of U .
4. Prove that H q (U, Hp ) = 0 for q > 0 and for any open subset U C. Hint: use Theorem
11.3 and an approximation argument like that used in the second half of the proof of
Theorem 11.5.
5. Prove that H q (P 1 (C), Hp ) = C p,q , where P 1 (C) is the Riemann sphere. Hint: Cover
the sphere with two open disks which overlap in an annulus. By the previous problem,

this is a Leray cover for the sheaves Hp . Now compute Cech

cohomology for each of the
p
sheaves H and this cover.

106

J. L. TAYLOR

12. Coherent Analytic Sheaves

A coherent algebraic sheaf S on an algebraic variety X is a sheaf of X O modules which,
of a
in an affine neighborhood V of each point, can be realized as the localization M
finitely generated O(V ) module M . Since M is finitely generated, there is a surjection
O(V )n M and, since O(V ) is Noetherian, the kernel of this map is finitely generated
also. Thus, there is an exact sequence
O(V )m O(V )n M 0
If we localize this sequence over V we obtain an exact sequence of sheaves
V

Om V On S|V 0

Thus, a coherent algebraic sheaf is locally the cokernel of a morphism V Om V On

between free finite rank sheaves of X O-modules. The converse is also true by Problem 10.9.
Thus, this condition characterizes coherent algebraic sheaves. We will use the analogous
condition as the definition of coherent analytic sheaves:
12.1 Definition. Let X be a holomorphic variety. An analytic sheaf on X is a sheaf of
X H-modules and a morphism of analytic modules is a morphism of sheaves of X H-modules.
An analytic sheaf S will be called a coherent analytic sheaf if each point of X is contained
in a neighborhood V such that S|V is the cokernel of a morphism V Hm V Hn between
free finite rank analytic sheaves.
There are two reasons why a definition in terms of localization, like that of chapter 10,
would not work well: H(V ) is never Noetherian and, tensor product relative to H(V ) is
not a well behaved operation.
The purpose of this chapter is to show that the sheaves of greatest interest in holomorphic function theory are, in fact, coherent analytic sheaves and to show that the category
of coherent analytic sheaves on X is an abelian category. Later chapters will deal with
vanishing theorems for the cohomology of coherent analytic sheaves and applications to
holomorphic function theory.
If X is a holomorphic variety, note that each morphism of analytic sheaves : X H X H
is given by multiplication by a global holomorphic function f . In fact, f is just (1)
the image of the identity section of X H. Similarly, each morphism of analytic sheaves
: X Hk X Hm is given by a m k matrix with entries which are holomorphic functions
on X.
12.2 definition. An analytic sheaf S is said to be locally finitely generated if for each
point x X there is a neighborhood U of x and finitely many sections s1 , . . . , sk of S such
that the germs of these sections at y generate Sy for every y U .
In other words S is locally finitely generated if for each x X there is a neighborhood
U of x and a surjective morphism U Hk S|U of analytic sheaves. Thus, S is coherent
if it is locally finitely generated in such a way that the resulting morphisms U Hk S|U
have kernels which are also locally finitely generated. It will take some work to show that
this is a reasonable condition. The key result is Okas Theorem, which is a substitute for
the Noetherian property:

NOTES ON SEVERAL COMPLEX VARIABLES

12.3 Okas Theorem. Let U Cn be an open set. Then the kernel of any
morphism U Hm U Hk is locally finitely generated.

107
U H-module

Proof. We prove this by induction on n. It is trivial for n = 0, since C0 is a point and

m
k
U H U H is a linear map between finite dimensional vector spaces over C in this case.
Thus, we assume that n > 0 and the theorem is true in dimension less than n.
To prove the theorem is true for dimension n, we first show that we may reduce the
proof to the case where k = 1. We do this using induction on k. Thus, suppose k > 1 and
the theorem is true in dimension n for all morphisms U Hm U Hj with j < k. Consider
a morphism : U Hm U Hk . We may write = (, ) where : U Hm U Hk1
is followed by the projection (f1 , . . . , fk ) (f1 , . . . , fk1 ) and is followed by the
projection (f1 , . . . , fk ) fk . Now by the induction hypothesis, ker is locally finitely
generated. Hence, if x U then there is a neighborhood V of x, contained in U , a p > 0
and a morphism : V Hp V Hm which maps onto ker . Thus, we have a diagram

Hp

Hm

Hk1

with the top row exact. We also have, by assumption, that the kernel of : V Hp V H is
locally finitely generated. This means that, after shrinking V if necessary, we can find q > 0
and a morphism : V Hq V Hp which maps onto ker . But ker = ker ker =
(ker ) = im . Thus, the kernel of is also locally finitely generated as was to be
shown.
Thus, we have reduced the proof to showing, in dimension n, that the kernel of a
morphism of the form : U Hm U H is locally finitely generated, under the assumption
that the theorem holds for all m and k in dimensions less than n. The strategy of the proof
is to use the Weierstrass theorems to reduce the problem to an analogous one involving
polynomials of a fixed degree in zn and then to apply the induction assumption to the
coeficients of these polynomials.
Given a point x U we must show that ker is finitely generated in some neighborhood
of x. Without loss of generality, we
Pmay assume that the point x is the origin. The map
has the form (g1 , . . . , gm ) =
fi gi for an m-tuple of functions fi n H(U ). By
appropriate choice of coordinates, we may assume that the germ at 0 of each fi is regular
of some degree at 0 and, hence, by the Weierstrass preparation theorem, has the form ui pi
where pi is a Weierstrass polynomial and ui is a unit. We may replace these germs by
their representatives in some neighborhood V of 0 and by choosing V small enough we may
assume the ui are non vanishing in V . Then the map (g1 , . . . , gm ) (u1 g1 , . . . , um gm )
is an automorphism of V Hm which maps the kernel of to the kernel of the morphism
determined by the m-tuple (p1 , . . . , pm ). Thus, without loss of generality,Pwe may replace
the fi s with the pi s and assume that has the form (g1 , . . . , gm ) =
pi gi . Let d be
the maximum of the degrees of the pi s. We may assume that V has the form V 0 V 00 for
open sets V 0 Cn1 and V 00 C. Let Kd denote the sheaf on V 0 defined as follows
Kd (W ) ker : V Hm (W V 00 ) V H(W V 00 )

108

J. L. TAYLOR

is the subspace consisting of m-tuples whose entries are polynomials of degree less than or
equal to d in the variable zn with coeficients in n1 H(W ). This is a sheaf of V 0 H-modules.
We shall show that it is locally finitely generated on V 0 . Now for each neighborhood
W V 0 , the space of m-tuples (q1 , . . . , qm ) where each qi is a polynomial in zn of degree
at most d, with coeficients which are in n1 H(W ) is a free module of rank m(d + 1) over
n1 H(W ). The map, determines a n1 H(W )-module morphism of this free module into
the free n1 H(W )-module of rank 2d + 1 consisting of polynomials of degree at most 2d
in zn with coeficients in n1 H(W ). Furthermore, Kd (W ) is the kernel of this morphism.
In other words, we may regard as determining a morphism of sheaves of V 0 H-modules,
m(d+1)
V 0 H2d+1 and our sheaf Kd is its kernel. By the induction hypothesis, Kd is
V 0H
locally finitely generated as a sheaf of V 0 H-modules. To complete the proof, we will show
that Kd generates ker as a sheaf of V H-modules.
Let be a point of W V 00 . We must show that the stalk of Kd at generates the
stalk of ker at . By performing a translation, we may assume that the point is the
origin. In the process, however, we lose the fact that the polynomials pi are Weierstrass
polynomials at 0. They are, however, still polynomials in zn and, by the Weierstrass
preparation theorem, we may factor p1 as p1 = p01 p001 where the germ at 0 of p01 is a
Weierstrass polynomial and the germ at 0 of p001 is a unit. We set
d01 = deg p01 ,
and note that

d001 = deg p001

d01 + d001 = deg p1 < d.

If h = (h1 , . . . , hm ) n H0m we then use the Weierstrass division theorem on each component of h to write h = p01 h00 + r0 , where r0 , h00 n H0m and r0 has components which are
polynomials in zn of degree less than d01 . If we set h0 = (p001 )1 h then
h = p1 h0 + r0
Also,
0

p1 h = q +

m
X

h0j ej

j=2

where
q=(

pi h0i , 0, . . . , 0)

and ej = (pj , 0, . . . , p1 , 0. . . . , 0)

with p1 occuring in the j th position of ej . Note that each ej belongs to (Kd )0 . Thus, if
we set r = r0 + q, we have
m
X
h=r+
h0j ej
j=2

where ej (Kd )0 and r = (r1 , . . . , rm ) an element of n H0 with r2 , . . . , rm polynomials in

zn of degree less than d01 . Now suppose h belongs to ker . Then so does each ej and,
hence, so does r. This means that
p1 r1 = (p2 r2 + + pm rm )

NOTES ON SEVERAL COMPLEX VARIABLES

109

which implies that p01 p001 r1 = p1 r1 is a polynomial in zn of degree less than d + d01 since this
is true of each term on the right above. However, since p01 is a Weierstrass polynomial, the
Weierstrass division theorem implies that p001 r1 must be a polynomial of degree at most d.
Then the entries of p001 r all have degree at most d and we conclude that p001 r (Kd )0 . Since
p001 is a unit we conclude that r and, thus, h belongs to the submodule of n H0 generated by
Kd . Thus, we have shown that at every point of V the stalk of Kd generates the stalk of the
kernel of . Since, Kd is itself locally finitely generated over V 0 O, the proof is complete.
12.4 Corollary. If U is an open set in Cn and M is a locally finitely generated sheaf of
submodules of U Hk , then M is coherent. In particular, if : U Hm U Hk is a morphism
of analytic sheaves , then ker and im are coherent analytic sheaves on U .
Proof. Since M is locally finitely generated, for each point of U there is a neighborhood
V on which M is the image of a morphism : V Hm V Hk of sheaves of V H-molules.
By Okas Theorem we know that ker is locally finitely generated. Thus, by shrinking V
if necessary, we may assume ker is the image of a morphism : V Hp V Hm . Then, on
V , M is the cokernel of . Thus, M is coherent.
Since the image of a morphism : U Hm U Hk of analytic sheaves is finitely generated
by definition and the kernel is locally finitely generated by Okas Theorem, they are both
coherent by the result of the previous paragraph.
12.5 Corollary. If U is an open set in Cn and M and N are coherent sheaves of submodules of U Hm , then so is M N .
Proof. Every point of U has a neighborhood V on which there are morphisms : V Hp
m
with image M|V and : V Hq V Hm with image N |V . Consider the map :
VH
p+q
V Hm defined by writing V Hp+q as V Hp V Hq and setting (f g) = (f )(g).
VH
The kernel of is coherent by Okas Theorem and, hence, is locally finitely generated.
Furthermore, on V , M N is the image of the kernel of under . Thus, after shrinking
V if neccessary, we may choose a finite set of generators for ker . Then the image of this
set under will generate M N on V . Thus, M N is locally finitely generated. In view
of the previous corollary, this proves that it is coherent.
If U Cn is an open set and I and J are sheaves of ideals of U H then I : J will denote
the sheaf which assigns to the open set V U the ideal I(V ) : J (V ) = {f n H(V ) :
f J (V ) I(V )}.
12.6 Corollary. If U is an open set in Cn and I and J are coherent ideal sheaves, then
so is I : J .
Proof. First, suppose that J is generated by a single element h n H(U ) and I is generated
by the elements g1 , . . . , gk n H(U ). Then consider the map : U Hk+1 U H defined by
(f0 , . . . , fk ) = hf0 g1 f1 gk fk
Then I : J is the image of the kernel of under the projection (f0 , . . . , fk ) f0 :
k+1
U H. The kernel of is locally finitely generated by Okas Theorem and, thus,
UH
so is I : J . It is then coherent by Corollary 12.4.

110

J. L. TAYLOR

For the general case, for any point of U we may choose a neighborhood V in which J
is finitely generated, say by h1 , . . . , hm . Also, on V , I : J is just the intersection of the
sheaves I : Ji , where Ji is the sheaf of ideals on V generated by hi . Thus, it follows from
the previous corollary that I : J is coherent.
We will prove that the ideal sheaf IY of a subvariety Y U of an open subset of Cn is
coherent. The key to the proof is the following lemma:
12.7 Lemma. Let U be an open set in Cn containing the origin. Let f1 , . . . , fp be a set of
holomorphic functions on U with Y as its set of common zeroes and I as the ideal sheaf it
generates. Suppose that I0 is a prime ideal of n H0 , and I = id Y at all points Y Z
where Z is a holomorphic subvariety of Y with Z0 6= Y0 . Then there is a polydisc U
centered at 0 such that I = id Y at all points .
Proof. Since the
functions f1 , . . . , fp determine Y , it follows from the Nullstellensatz that
I0 id Y
0 I0 . However, by assumption I0 is prime and, hence, we have that I0 =
id Y0 = I0 . Let d be a function holomorphic in a polydisc centered at 0 such that
its germ d0 at 0 belongs to id Z0 but not to id Y0 . We may assume (by shrinking if
neccessary) that d vanishes on Z but does not vanish identically on Y . By Corollary
12.6, the sheaf I : n Hd on is locally finitely generated and, hence, we may assume it is
finitely generated by shrinking . Let g1 , . . . , gq n H() be a set of generators for this
sheaf. Then, dgj I() and, in particular, its germ at 0 belongs to I0 . However, this is a
prime ideal and it does not contain the germ of d. Hence, we must have that the germ of
gj at 0 belongs to I0 for every j. That is, the germs at 0 of each gj belong to the ideal at
0 generated by the fi . If this is true at 0, it is true in a neighborhood of 0 and, hence, we
may as well choose small enough that it is true at every point of . This implies that
I : n H0 d = I

on

For any point Y and any germ f id Y , we choose a representative f of f in a

neighborhood V of and note that, if the neighborhood is sufficiently small, then I : n Hf
will be finitely generated on V by Corollary 12.6. The hypothesis that I = id Y at points
of Y Z implies that I : n Hf = n H at points of Y Z. Thus, if h1 , . . . , hm n H(V ) are
generators of I : n Hf on V then the set of common zeroes of the hi must lie in Z V .
The function d vanishes on Z and, hence, by the Nullstellensatz, dr (I : n Hf ) for some
r. This means that (dr f ) I and, since I : n H0 d = I on , it follows that f I .
Therefore, I = id Y at all points of Y and, hence, at all points of .
Finally, we can prove:
12.8 Theorem. If Y is a holomorphic subvariety of an open set U Cn , then its ideal
sheaf IY is coherent.
Proof. Fix a point of Y which we may assume is the origin. We assume at first that the
germ Y0 is irreducible. We may assume (after a change of variables if necessary) that
the ideal id Y0 is strictly regular in the variables zm+1 , . . . , zn (see chapter 5). We choose
a set of generators f1 , . . . , fp for id Y0 which includes the polynomials pj m H0 [zj ] for
m + 1 j n and qj m H0 [zm+1 , zj ] for m + 2 j n of Lemma 5.9. Now, by

NOTES ON SEVERAL COMPLEX VARIABLES

111

Corollary 5.18 the subvariety Y is actually a complex manifold outside a proper subvariety
Z. Furthermore, it can be seen from the proof of Lemma 5.13 and Theorem 5.17 that
the functions pm+1 , qm+2 , . . . , qn can be taken as the last n m coordinate functions of a
coordinate system near any point of Y Z a coordinate system in which Y is expressed
as the set where these last n m coodinates vanish. It follows that the functions f1 , . . . , fp
generate the germ of id Y at such points. It then follows from the previous lemma that
these functions generate (IY ) = id Y at all points of some polydisc containing 0. Thus,
IY is coherent.
If the germ Y0 is not irreducible, then we write Y = Y1 Yq in some neighborhood
of 0, where the varieties Yj have irreducible germs at 0. Then IYj is finitely generated in
some neighborhood of 0 for each i by the previous paragraph. We may choose to be a
neighborhood in which this is true for all j. Then each ideal sheaf IYj is coherent on
and, hence, so is the intersection IYj IYq by Corollary 12.5. But this intersection is
just IY and, thus, the proof is complete.
We are now in a position to prove the strong form of Okas Theorem:
12.9 Okas Theorem on Varieties. Let : X Hm X Hk be a morphism of analytic
sheaves on an analytic variety X. Then ker is coherent.
Proof. This is a local result and so we may assume that X is a subvariety of an open
set U contained in Cn . The morphism is determined by a k m-matrix with entries
which are holomorphic functions on X, and, hence, extend locally to be holomorphic in
neighborhoods in U . Again, since we are proving a local result, we may as well assume that
the entries of this matrix extend to be holomorphic in U . Then extends to a morphism
: U Hm U Hk of analytic sheaves. Furthermore, by the previous result, we know that
the ideal sheaf IX is locally finitely generated and, hence, we may as well assume that it is
finitely generated on U . Then we may represent it as the image of a U H-module morphism
: U Hp U H. Let k : U Hkp U Hk denote the morphism that is just the direct sum
of k copies of . Then consider the morphism
: U Hm U Hkp U Hk

where

) k (g)
(f, g) = (f

Then f U Hm (V ) is the first element of a pair (f, g) ker over a neighborhood V if

) I k (V ) that is, if and only if the restriction of f to X V is zero.
and only if (f
X
Thus, the kernel of on V can be characterized as those functions which are restrictions
to X V of first elements of pairs (f, g) in the kernel of on V . Thus, it is clear that
ker is locally finitely generated if ker is locally finitely generated. But ker is locally
finitely generated by Okas Theorem. Thus, ker is locally finitely generated. However,
knowing this for all such morphisms allows us to prove in exactly the same manner as
in Corollary 12.4 that a locally finitely generated subsheaf of a free analytic sheaf X Hm is
coherent. It follows that ker is coherent.
We used in the preceding proof, that once we know that the kernel of any morphism
m
H
X Hk is locally finitely generated, then we may prove, as in Corollary 12.4, that
X
any locally finitely generated subsheaf of a free finite rank analytic sheaf X Hm is coherent.
In fact, the analogs for varieties of Corollaries 12.4, 12.5, and 12.6 are all true and have
exactly the same proofs. We state this fact as a Corollary to the above theorem.

112

J. L. TAYLOR

12.10 Corollary. If X is a holomorphic variety then

(i) any locally finitely generated analytic subsheaf of X Hm is coherent;
(ii) the image and kernel of any morphism of analytic sheaves : X Hm X Hk are
coherent;
(iii) the intersection of two coherent subsheaves of X Hm is coherent;
(iv) if I and J are two coherent ideal sheaves in X H, then I : J is also coherent.
Once we have Theorem 12.8, it is a simple matter to prove the following:
12.11 Cartans Theorem. If X is any holomorphic variety and Y X is a holomorphic
subvariety, then the ideal sheaf IY X H is coherent.
Proof. Again, this is a local result and so we may assume that X is a subvariety of an open
set U in Cn . Then Y is also a subvariety of U and, as such, its ideal sheaf is a coherent
sheaf of U H-modules by Theorem 12.8. In particular, it is locally finitely generated as an
U H-module and we may as well assume that it is finitely generated on U . But then the
restriction to X of a set of generators of this ideal sheaf will be a set of generators over
X H of its ideal sheaf IY . Thus, IY is locally finitely generated and, hence, coherent by
Corollary 12.10.
The main result remaining to be proved in this chapter is that the kernel, image and
cokernel of a morphism between coherent analytic sheaves are also coherent. The next
three results lead up to this theorem.
12.12 Lemma. If X is a holomorphic variety and : X Hm S is a surjective morphism
of analytic sheaves, then for each morphism of analytic sheaves : X Hk S and each
point x X there is a neighborhood U of x in which lifts to a morphism of analytic
sheaves : U Hk U Hm such that = on U .
Proof. The stalk U Hxk is a free, hence projective, U Hx -module and so the morphism :
k
k
m
U Hx Sx lifts to a morphism : U Hx U Hx such that = . The morphism is
represented by a matrix with entries from U Hx and we may assume that U is chosen small
enough that each of these entries is represented by a holomorphic function on U . The
resulting matrix defines a morphism of analytic sheaves : U Hk U Hm and the identity
= 0 is an identity which holds at x and, hence, in some neighborhood of x since
k
U Hx is finitely generated. Again, we may as well assume that neighborhood is U . This
completes the proof.
12.13 Theorem. If X is a holomorphic variety, then any locally finitely generated analytic
subsheaf M of a coherent analytic sheaf S is also coherent.
Proof. Since S is coherent, each point x of X has a neighborhood U for which there is a
surjective morphism of analytic sheaves : U Hm S|U with a locally finitely generated
kernel. Also, since M is locally finitely generated, we may assume U is chosen small
enough that M|U is finitely generated. Thus, there is a morphism of analytic sheaves
: U Hk S|U which has M|U as image. In order to prove that M is coherent, we
need to prove that ker is locally finitely generated, or, since our chosen point x is quite
general, that ker is finitely generated if U is chosen small enough. Also if U is chosen

NOTES ON SEVERAL COMPLEX VARIABLES

113

small enough, we may lift to a morphism of analytic sheaves : U Hk U Hm such that

= on U by the preceding lemma.
Now the image of is a finitely generated subsheaf of U Hm as is the kernel of . Thus,
both sheaves are coherent, as is their intersection, by Corollary 12.10. Since, ker im is
coherent, it is, after shrinking U if necessary, the image of a morphism of analytic sheaves
: U Hp U Hm . Using Lemma 12.12, we conclude that if U is chosen small enough,
there is a morphism of analytic sheaves : U Hp U Hk such that the following diagram
is commutative:

m
m
S
UH
UH
x
x
x

UH

UH

UH

Now for y U , f U Hyk is in the kernel of if and only if (f ) ker im = im ,

that is, if and only if there exists g U Hyp such that (f ) = (g) = (g). The latter is
equivalent to f (g) ker . Since ker is coherent, we may, after shrinking U again if
necessary, assume that ker is the image of a morphism : U Hq U Hk . Then f belongs
to ker if and only if f im + im . This implies that ker is locally finitely generated
and, hence, that M is coherent.
The following has a proof much like that of Corollary 12.5 (Problem 12.1):
12.14 Corollary. If M and N are coherent subsheaves of a coherent analytic sheaf over
a holomorphic variety X, then M N is also coherent.
12.15 Theorem. If M and N are coherent analytic sheaves over a holomorphic variety
X, then the image, kernel, and cokernel of any morphism of analytic sheaves : M N
are also analytic.
Proof. Since M is locally finitely generated, so is im . Hence, im is coherent by Theorem
12.13.
Fix a point x X. Since N and im are locally finitely generated, there is a neighborhood U of x and surjective morphisms of analytic sheaves : U Hn N |U and
: U Hp im |U . In fact, by choosing U small enough and using Lemma 12.12 to
construct , we may construct the following commutative diagram of morphisms of analytic sheaves:
0 im |U N |U coker |U 0
x
x
x

UH

UH

UH

The top row of this diagram is exact and the vertical maps are all surjective. It evident
from the diagram that the kernel of is 1 (im ) = im . Thus, by definition, coker is
coherent.
It remains to prove that ker is coherent. To this end, we fix x X and choose a
neighborhood U of x small enough that we may find surjective morphisms of analytic

114

J. L. TAYLOR

sheaves : U Hm M and : U Hn N such that ker and ker are finitely generated
and, thus, coherent. We then construct the following commutative diagram after shrinking
U appropriately:

0 ker |U M|U N |U
x
x

UH

UH

UH

UH

UH

Here we use Lemma 12.12 to construct the map . Its image is finitely generated and, hence,
coherent. Then ker im is coherent and, hence, is the image of a morphism as above.
Then is obtained as another application of Lemma 12.12. The map is a morphism
of analytic sheaves which maps onto the kernel of and it exists, for small enough U , by
Corollary 12.10(ii). Now a simple diagram chase shows that (+) : U Hp U Hq M|U
has ker |U as its image. This completes the proof that ker is locally finitely generated
and, hence, coherent.
The above theorem implies that the category of coherent analytic sheaves is an abelian
category.
Finally, we have the following result:
12.16 Theorem. If

0 K M N 0
is an exact sequence of analytic sheaves and if any two of the three are coherent, then the
third is also coherent.
Proof. Two of the three cases have already been proved in the preceding theorem. It
remains to prove that if K and N are coherent then M is coherent. To this end, let x be
a point of X and U a neighborhood of x for which we may find surjective morphisms of
analytic sheaves : U Hk K|U and : U Hn N |U with finitely generated kernels. We
may then use Lemma 12.12 to construct the following commutative diagram with exact
rows:

0 K|U M|U N |U 0
x
x
x

0 U Hk U Hk+n U Hn 0
Here the morphisms on the bottom row are just the canonical injection and projection
associated with writing U Hk+n as U Hk U Hn . The morphism is constructed, for
sufficiently small U , by lifting to a morphism 0 : U Hn M|U with 0 = using
Lemma 12.12, then writing U Hk+n as U Hk U Hn and defining by
(f g) = (f ) + 0 (g).

NOTES ON SEVERAL COMPLEX VARIABLES

115

Now the fact that and are surjective implies that is also surjective. Also, the kernels
of the vertical maps form a short exact sequence
0 ker |U ker |U ker |U 0
in which the first and third terms are locally finitely generated and, hence, coherent. Thus,
we may repeat the above argument for this sequence and conclude that, for sufficiently
small U , ker |U is also finitely generated. It follows that M is coherent and the proof is
complete.

12. Problems
1. Prove Corollary 12.14.
2. Prove that if : X Hk X Hm is a morphism of analytic sheaves then x : X Hxk X Hxm
is surjective if and only if the matrix of holomorphic functions on X defining has rank
m at x.
3. Use the result of the preceding problem to prove that if M is a coherent analytic sheave
on a holomorphic variety X, then Support(M) = {x X : Mx 6= 0} is a holomorphic
subvariety of X.
4. Use the result of the preceding problem to prove that if : M N is a morphism of analytic sheaves between two coherent analytic sheaves, then {x X : x is not injective}
and {x X : x is not surjective} are holomorphic subvarieties of X.

116

J. L. TAYLOR

13. Projective Varieties

Complex projective space of dimension n is the algebraic variety defined in the following
fashion: We consider the point set P n which is Cn+1 {0} modulo the equivalence relation
defined by
(z0 , . . . , zn ) (z0 , . . . , zn ) C {0}
We shall define a topology and a ringed space structure on P n and show that the resulting
ringed space is an algebraic variety.
Let p C[z0 , . . . , zn ] be a homogeneous polynomial of some degree say k. Then p
does not define a function on P n but the relation
p(z0 , . . . , zn ) = k p(z0 , . . . , zn )
means that the zero set of p is a union of equivalence classes and, hence, defines a subset
of P n . Similarly, the set of common zeroes of any set of homogeneous polynomials is a
union of equivalence classes and defines a subset of P n .
13.1 Definition. An algebraic subset of P n is a subset which is the set of common zeroes
of some family of homogeneous polynomials.
Of course, because subvarieties of Cn+1 satisfy the descending chain condition, any
algebraic subset of P n is actually the zero set of a finite set of homogeneous polynomials.
It is easy to see that the collection of algebraic subsets of P n is closed under finite
union and arbitrary intersection. Thus, this collection may be taken as the closed sets in
a topology for P n .
13.2 Definition. The Zariski topology on P n is the topology in which the open sets are
the complements of algebraic sets.
We now associate to each integer k a sheaf O(k) on P n in the following way: Let
: Cn+1 {0} P n be the projection. If U is open in P n then 1 (U ) is open in
Cn+1 {0} and, in fact, is the complement of the set of common zeroes of a finite set of
homogeneous polynomials.
Definition 13.3. If U is an open subset of P n and k is an integer, we define O(k)(U ) to
be the space of regular functions on 1 (U ) which are homogeneous of degree k.
Clearly O(k) forms a sheaf for each k. Furthermore, if f O(j)(U ) and g O(k)(U )
then f g O(j + k)(U ). Thus,
k= O(k) is a sheaf of graded rings. In particular, O(0)
is a sheaf of rings and each O(k) is a sheaf of modules over O(0).
Definition 13.4. We make P n into a ringed space by defining the structure sheaf to be
the sheaf of rings O = O(0).
Consider the open set Ui in P n which is the complement of the algebraic set defined by
the zero set of the i th coordinate function zi . We define a map i : Ui Cn by i = i
where

zi1 zi+1
zn
z0
,...,
,
,...
i (z0 , . . . , zn ) =
zi
zi
zi
zi
The map i is clearly well defined and, in fact:

NOTES ON SEVERAL COMPLEX VARIABLES

117

13.5 Theorem. The map i : Ui Cn is an isomorphism of ringed spaces.

Proof. We may as well assume that i = 0. We first prove that 0 is a homeomorphism.
The map 0 is clearly bijective and so, to prove that it is a homeomorphism, we must show
that a set is closed in Ui if and only its image is closed in Cn .
Every closed subset of Ui is a finite intersection of sets of the form Ui Z(p) where p is
a homogeneous polynomial and Z(p) P n is its zero set. But if z0 6= 0 then
p(z0 , z1 , . . . , zn ) = 0

p(1, z1 /z0 , . . . , zn /z0 ) = 0

Thus, the image of Ui Z(p) under 0 is a subvariety of Cn that is, a closed set. It
follows that the image of every closed subset of Ui is closed in Cn .
The closed subsets of Cn are finite intersections of zero sets of polynomials. Let q be
a polynomial in z1 , . . . , zn of degree k. Then we may define a homogeneous polynomial of
degree k in z0 , . . . , zn by
p(z0 , z1 , . . . , zn ) = z0k q(z1 /z0 , . . . , zn /z0 )
and p is zero exactly at points which 0 maps to zeros of q. That is, Z(p) U0 is the
inverse image under 0 of the zero set of q. It follows that the inverse image under 0 of
any closed set in Cn is a closed set in U0 .
To finish the proof, we must show that 0 induces an isomorphism between the structure
sheaves of U0 and Cn . This amounts to showing that, for each open set V U0 , a complex
valued function on V has the form f for a homogeneous regular function f on 1 (V )
if and only if it has the form g 0 for a regular function g on 0 (V ). In other words,
for each open set W = 0 (V ), we must show that g g 0 is an isomorphism from
the regular functions on W to the homogeneous regular functions on 1 (W ). Since 0
is algebraic and homogeneous it is clear that g g 0 is a ring homomorphism from
O(W ) to regular homogeneous functions on 1 (W ). To see that it is an isomorphism, we
simply note that its inverse is given by f f where f(z1 , . . . , zn ) = f (1, z1 , . . . , zn ). This
completes the proof.
13.6 Theorem. The ringed space P n is an algebraic variety.
Proof. We have that, {Ui } is a cover of P n by open subsets which are isomorphic as ringed
spaces to Cn . Thus, P n is an algebraic prevariety It is also clear that, given any two points
p and q of P n , we may choose our coordinate system in Cn+1 in such a way that one of
the Ui contains both p and q. In other words, given any two points, there is an affine open
subset containing both. By Theorem 10.6 this implies that P n is actually an algebraic
variety.
13.7 Definition. A projective variety is an algebraic variety which is isomorphic to a
closed subvariety of P n for some n.
To any algebraic variety we may associate in a canonical way a holomorphic variety. An
algebraic variety is locally isomorphic, as a ringed space, to an algebraic subvariety of Cn .
There is a canonical way to associate to an algebraic subvariety V of Cn an holomorphic
subvariety V h we simply give V the Euclidean topology instead of the Zariski topology

118

J. L. TAYLOR

and let its structure sheaf be the sheaf of holomorphic functions rather than the sheaf
of regular functions. This means that for any algebraic variety we have a canonical way
of changing the topology and ringed space structure on any affine open subset in such
a way as to make it a holomorphic variety. This is canonical because any two ways of
representing an affine open subset as subvarieties of complex Euclidean space are related by
a biregular map between the two subvarieties which will necessarily be a biholomorphic
map between the associated holomorphic subvarieties. In particular, this implies that
holomorphic space structures defined in this way on affine subsets of an algebraic variety
will agree on intersections and, thus, define a global structure of a holomorphic variety.
One thing that does need to be checked is that the resulting topological space is Hausdorff
(Problem 13.2).
From the above discussion, we conclude that projective space P n may also be considered
as a holomophic variety in fact, as a complex manifold, since the maps i : Ui Cn give
local biholomorphic maps onto Cn . We initially defined regular functions on U P n as
just regular functions on 1 (U ) which are homogeneous of degree zero. It turns out that
the analogous statement is true of holomorphic functions:
13.8 Definition. For each integer k we define a sheaf H(k) on P n with the Euclidean
topology as follows: If U P n is open in the Euclidean topology, then H(k)(U ) is the
space of functions in H( 1 (U )) which are homogeneous of degree k.
13.9 Theorem. The sheaf H(0) is a sheaf of rings canonically isomorphic to the structure
sheaf H of P n and the sheaves H(k) are sheaves of H-modules.
Proof. Clearly H(0) is a sheaf of rings and each H(k) is a sheaf of modules over H(0).
Thus, we need only show that H(0) is canonically isomorphic to H. The isomorphism is
obviously the one which sends f H(U ) to f . It is easy to see, using the definition of
the holomorphic structure on P n , that : Cn 0 P n is holomorphic, so that f f
is an homomorphism (obviously injective) of H to H(0). It only remains to show that this
is surjective. If g is a homogeneous holomorphic function on an open set 1 (U ) then
f ((z)) = g(z) certainly defines a function f on U . The only question is whether or not it
is holomorphic. It suffices to prove this in the case where U Ui for some i and, without
loss of generality, we may assume that i = 0. Then to prove that f is holomorphic, we must
1
n
show that f 1
0 is holomorphic on 0 (U ) C . But f 0 (z1 , . . . , zn ) = g(1, z1 , . . . , zn )
and this is certainly holomorphic.
We now turn to the study of the sheaves O(k) and H(k) introduced above. The sheaves
O(k) are sheaves of O = O(0) modules. If V is an open subset of Ui then it is easy to
see that f O(V ) if and only if zik f O(k)(V ). Thus, f zik f defines an O|Ui -module
isomorphism from O|Ui to O(k)|Ui . In other words, O(k) is locally free of rank one as
a sheaf of O-modules. Exactly the same thing is obviously true of H(k) as a sheaf of H
modules. A sheaf of modules with this property is called an invertible sheaf since such
a sheaf always has an inverse under tensor product relative to the structure sheaf. In
this case, the inverse of O(k)(H(k)) is obviously O(k)(H(k)) in view of the following
theorem:
13.10 Theorem. If j and k are integers then multiplication defines an isomorphism
O(j) O O(k) O(j + k).

NOTES ON SEVERAL COMPLEX VARIABLES

119

The analogous statement is true for the sheaves H(k).

Proof. We multiplication map f g f g defines a morphism of algebraic sheaves from
O(j)O O(k) to O(j +k). The only question is whether or not it is bijective. However, this
map is bijective if and only if it is locally bijective. Thus, it suffices to show that the map
is bijective on Ui for each i. However, on Ui , each O(k) is free of rank one with generator
zik while O(j) O O(k) is free of rank one with generator zij zik . The multiplication map
sends zij zik to zij+k , which is the generator of O(j + k) on Ui . This proves the theorem
in the algebraic case. The proof in the analytic case is the same.
The sheaf of sections of a (finite dimensional) holomorphic vector bundle is a locally free
finite rank sheaf of H- modules and vice-verse. This is due to the fact that a holomorphic
vector bundle is locally trivial and, thus, its sections may locally be identified with holomorphic vector valued functions. If the vector space has dimension m and a basis is chosen,
then the space of holomorphic vector valued functions over an open set U is isomorphic to
H(U )m . Thus, the module of holomorphic sections is locally free and of finite rank. The
converse is just as easy (problem 13.3). Sheaves of modules which are locally free of rank
one (invertible sheaves), such as our sheaves H(k), can be realized as sheaves of sections
of holomorphic line bundles vector bundles with one dimensional fiber. Note that locally
free sheaves of finite rank are, of course, coherent, since coherence is a local property and
locally free finite rank sheaves locally have the form Hm (or Om in the algebraic case).
Our next main objective is to compute the sections and cohomology of the sheaves O(k)
and H(k). It will be convenient to introduce three additional sheaves on P n :
13.11 Definition. Let S denote the sheaf which assigns to an open set U P n the ring
O( 1 (U )) and let T denote the sheaf which assigns to U the ring H( 1 (U )). We also
let T0 be the subsheaf of T spanned by the subsheaves H(k).
Thus, O(k) is the subspace of S consisting of elements of degree k and S =
k= O(k),
while H(k) is the subspace of T consisting of elements of degree k. Note that
k= H(k)
is not T but the subsheaf T0 .

We will compute cohomology using the Cech

complex for the open cover {Ui }. Note that
if = (i0 , . . . , ip ) is a multi-index, then U = Ui0 Uip = Uz , where z = zi0 . . . zip
and Uz is the complement of the algebraic set in P n determined by the vanishing of z .
Thus, U is the subset of Uzi0 ' Cn on which the regular function zip
zi1 . . . zip does not
0
vanish. This implies that U is an affine variety for each multi-index from which it
follows that O(k) is acyclic on U for each and each k and, hence, that {Ui } is a Leray
cover for O(k) for each k. Thus, Lerays Theorem applies and we may, in fact, compute

the sheaf cohomology of O(k) or S using the Cech

complex for {Ui }. We also have that
n
each Ui is biholomorphically equivalent to C , each H(k) is a free H module on each Ui
and each U is a Cartesian product of planes and punctured planes. It follows from our
work on Dolbeault cohomology (see the remark following Theorem 11.5) that H(k) is also
acyclic on each U and, hence, that {Ui } is also a Leray cover for the sheaf H(k). Again,

Lerays theorem implies that we may compute the cohomology of H(k) or T0 using Cech
cohomology for the cover {Ui }.

For what follows, we need to use the definition of Cech

cohomology in which only
alternating cochains are used that is, cochains f for which f (()) = sgn()f () for

120

J. L. TAYLOR

each permutation and each multi-index . This is equivalent to the definition we used

earlier the proof of this may be found in nearly any standard text in which Cech
Theory
is developed (or do Problem 13.1).

We will compute Cech

cohomology for O(k) and H(k) by computing it for S and T0 and
then projecting out the part which is homogeneous of degree k. We begin by expressing
each element of S(U ) and T (U ) in terms of its Fourier coeficients. That is, if f T (U )
for = (i0 , . . . , ip ) then f is a holomorphic function on the set of all (z0 , . . . , zn ) Cn+1
such that zij 6= 0 for j = 0, . . . , p. Such a function has a Laurent series expansion
X

am0 ...mn z0m0 . . . znmn

in which am0 ...mn is allowed to be non-zero only for terms such that either mi 0 or
mi < 0 and i is one of the ij appearing in . For such an expansion to converge on U ,
the coeficients am0 ...mn must decay faster than r|m0 |+...|mn | for every positive number r.
In other words, we may identify T (U ) with the space of functions m am : Z n+1 C
which decay at infinity faster than any geometric series and which are supported on
{m = (m0 , . . . , mn ) : mi 0 if

i
/ {i0 , . . . , ip }}

Of course, S(U ) is the subspace of T (U ) consisting of sums, as above, with coeficients

which are non-vanishing for only finitely many indices.
The spaces of global sections S(P n ) and T (P n ) may be described as above if we simply
use for the empty index. Thus, T (P n ) is the set of sums as obove, which satisfy the
decay condition and have Fourier coeficients which vanish except when all indices mi are
non-negative. Similarly, S(P n ) is the set of sums, as above, with coeficients which are
finitely non-zero and are zero except for all non-negative indices mi .
In S(U ) and T (U ), the elements which are homogeneous of degree k are those whose
Fourier coeficient functions are supported on the set where m0 + m1 + + mn = k.
The coeficient functions of global sections are also supported on the set where all mi 0.
Then, since there are only finitely many tuples (m0 , . . . , mn ) of non-negative integers with
m0 + m1 + + mn = k we conclude that O(k)(P n ) and H(k)(P n ) are finite dimensional,
are equal, and, in fact:
13.12 Theorem. For each integer k, H 0 (P n , O(k)) and H 0 (P n , H(k)) are both equal
to the space of homogeneous polynomials of degree k in C[z0 , . . . , zn ]. In particular,
H 0 (P n , O) = H 0 (P n , H) = C and H 0 (P n , O(k)) = H 0 (P n , H(k)) = 0 for k < 0.
13.13 Theorem. For each integer k we have
(a) H n (P n , O(n 1)) ' C;
(b) H n (P n , O(n k 1)) is the vector space dual of H 0 (P n , O(k)) under the pairing
H 0 (P n , O(k)) H n (P n , O(n k 1)) H n (P n , O(n 1)) ' C
induced by multiplication;
(c) H p (P n , O(k)) = 0 if p is not 0 or n.

NOTES ON SEVERAL COMPLEX VARIABLES

121

Also, the same statements are true with O(k) replaced by H(k).

Proof. An alternating Cech

cochain vanishes on any multi-index with a repeated entry.
Thus, n is the largest value of p for which there are non-zero alternating p-cochains for
{Ui }, from which it follows that H p (P n , S) = 0 for p > n.
Via the Fourier representation, we regard a p-cochain for S and {Ui } as an alternating function a which assigns to each = (i0 , . . . , ip ) a finitely non-zero function
n+1
(m0 , . . . , mn ) a
C with support in the set
m0 ...mn : Z
W = {(m0 , . . . , mn : mi 0

for i
/ {i0 , . . . , in }}.

Now multiplication by the characteristic function of a given subset L of Z n+1 is a projection

operator PL on cochains which commutes with the coboundary operator and has as range

the cochains with support in L. Thus, PL also defines a projection operator on Cech
cohomology as well. In particular, the cohomology of O(q) can be obtained from
P that of S
by applying the projection PLk determined by the set Lk = {(m0 , . . . , mn ) :
mi = k}.
Also, it makes sense to talk about an element of cohomology of S being supported in a set
L that is, H p (P n , S) is supported in L if PL = . With this in mind, the strategy
of the proof will be to prove that, for p > 0, every element of H p (P n , S) is supported in
the set K = {(m0 , . . . , mn ) : mi < 0 i}.
We know that H p (Uj , S) = 0 for p > 0, since Uj is affine and S is quasi-coherent (it
is an infinite direct sum of coherent sheaves). In terms of our Fourier coeficient picture,
this means that, for a fixed j and p > 0, any p-cocycle a = {a
m0 ,...,mn } for S and {Ui } is,
when restricted to Uj , the coboundary of some p 1-cochain b = {bm0 ,...,mn } for S and
{Uj Ui }i . Now generally b will have some non-zero coeficients with negative index mj
and a which does not include j among its entries. This, of course, just means that b
is not generally a cochain for {Ui }. However, if Wj = {m Z n+1 : mj 0} then PWj b
is a cochain for {Ui } which is mapped by the coboundary onto PWj a. This implies that
PWj kills elements of H p (P n , S) for p > 0 and, hence, that H p (P n , S) is supported on the
complement of Wj . Since this is true for every j, we conclude that H p (P n , S) is supported
on K as claimed. This implies that H p (P n , S) = 0 for 0 < p < n since every p-cochain
for p < n is supported on the complement of K. We already know that H p (P n , S) = 0 for
p < 0 and for p > n. This completes the proof of (c) in the algebraic case, since we know
that we can obtain H p (O(k)) as a direct summand of H p (P n , S).
For an n-cochain, the only multi-index we need consider is = (0, 1, . . . , n), since all
others without repeated entries are permutations of this one. For this index the set
W is all of Z n+1 . Thus, an n-cochain for S and {Ui } is just a single coeficient function
a = am0 ,...,mn with no restriction other than that only finitely many coeficients are nonzero. Now we know from the previous paragraph that H n (P n , S) is supported on K.
However, every n- cochain is a cocycle and the space of (n 1) cochains is supported in
the complement of K. Thus, it follows that H n (P n , S) is actually isomorphic to the space
of coeficient functions supported in K. This proves
P (a) since an index (m0 , . . . , mn ) which
mi = n 1 if and only if each mi is
belongs to K, that is mi < 0 i, can satisfy
n1
n
1. Thus, H
(P , O(k)) consists of functions with a single non-vanishing coeficient
corresponding to the index (1, . . . , 1). This proves part (a).

122

J. L. TAYLOR

We have from the previous theorem that H 0 (P n , S) may be regarded as the space of
finitely non-zero coeficient functions which are supported in K+ = {(m0 , . . . , mn ) : mi
0 i} and from the previous paragraph, that H n (P n , S) may be regarded as the space of
finitely non-zeroP
coeficient functions which are supported on K. Clearly reflection through
the hyperplane
mi = (n 1)/2 is a bijection between K+ and K. This pairs basis
elements for H 0 (P n , S) and H n (P n , S). Clearly it pairs functions supported on Lk with
functions supported on Lnk1 and, thus, it pairs basis elements for H 0 (P n , O(k)) and
H n (P n , O(n k 1k)) as require in the theorem. This completes the proof of part (a)
in the algebraic case.
The argument is almost the same in the analytic case, where O(k) is replaced by H(k).
The difference is that we must work with T0 rather than S. However, T0 is just the direct
sum of the H(k) and, hence, also has vanishing cohomology on each Ui . The rest of the
argument is the same if one observes that the projection operators which play such a
prominent role preserve the faster than exponential decay at infinity that defines coeficient
functions in T (U ).
Since we use the same open cover by Zariski open sets to compute both, there is a map
H p (P n , O(k)) H p (P n , H(k)) defined by inclusion. Clearly, the above theorem has as a
consequence:
13.14 Corollary. The map H p (P n , O(k)) H p (P n , H(k)) is an isomorphism for every
p and every k.
We complete this section with some results which show how to make strong use of the
sheaves O(k).
13.15 Definition. If F is a coherent algebraic sheaf on P n then we set F(k) = F O O(k).
The sheaf F(k) is said to be obtained from F through twisting it by O(k).
Note that k F(k) is a sheaf of modules over the sheaf of rings S = k O(k) where the
product hf of h O(k) and f F (m) is h f F(k + m).
Note also, that, since O(k) is locally free of rank one, F(k) is locally isomorphic to
F, but it is not generally globally isomorphic to F because of the twist introduced by
tensoring with O(k).
13.16 Theorem. If F is a coherent algebraic sheaf on P n then for some k 0, the sheaf
F(k) is generated by finitely many of its global sections.
Proof. For each i the open set Ui is affine and, hence, the coherent sheaf F|Ui is the image
under localization of (Ui , F) which is finitely generated. Let {fij }j be a finite generating
set of sections for (Ui , F). Now zi is a global section of O(1) which vanishes exactly on
the complement of Ui . For some integer m, which may be chosen large enough to work
for all i, j, the product zim fij extends to be a global section of F(m) (problem 13.4). The
resulting collection of global sections {zim fij }ij clearly generate F(m) on each Ui and,
hence, on all of P n .
13.17 Theorem. If F is a coherent algebraic sheaf on P n , then F is the quotient of a
sheaf which is a finite direct sum of sheaves of the form O(k)

NOTES ON SEVERAL COMPLEX VARIABLES

123

Proof. By the previous theorem, there exists a non-negative integer m such that F(m)
is generated by finitely many global sections. If there are r of these sections then they
determine a surjection Or F(m). On tensoring this with O(m), we get a surjection
O(m)r F as required.
13.18. If F is a coherent algebraic sheaf on P n then
(a) H p (P n , F) is finite dimensional for each p and vanishes for p > n;
(b) there exists an integer m0 such that H p (P n , F(m)) = 0 for all p > 0 and all
m > m0 .
Proof. That H p (P n , F) vanishes for p > n is just the fact that we can use alternating

Cech
cochains for the cover {Ui } to compute it. Note that this implies that the m0 in part
(b) can be chosen independent of p if it can be chosen depending on p since there are only
finitely many ps to worry about. With these things in mind, the proof is by reduction on
p. We assume both statements are true for p + 1 n + 1 and then prove they are also true
for p. We express F as a quotient of a finite direct sum i O(ki ) so that we have a short
exact sequence
0 K i O(ki ) F 0
Then part(a) for p follows from the long exact sequence for coholomogy associated to
this short exact sequence, the assumption that the theorem is true for p + 1 and Theorem
13.13. Part(b) follows in a similar fashion from Theorem 13.13 and the long exact sequence
associated to the short exact sequence
0 K(m) i O(m + ki ) F(m) 0.
obtained by twisting by the sheaf O(m). This completes the proof.

13. Problems

1. Verify that Cech

cohomology does not change if the Cech
cochains of chapter 9 are

replaced by alternating Cech cochains.

2. Prove that if V is an algebraic variety then the topological space V h constructed by
giving each affine open subset of V its Euclidean topology is Hausdorph. Hint: Use the
condition of Definition 10.5 to prove that the diagonal in V h V h is closed.
3 Prove that if V is a holomorphic variety and M is a locally free finite rank sheaf of
V H-modules, then there is a holomorphic vector bundle such that M is isomorphic to
its sheaf of holomorphic sections.
4. Model the proof of Theorem 10.12(ii) to prove that if F is a coherent sheaf on P n and
f (Ui , F), then for some positive integer m the section zim f (Ui , F(m)) extends
to a global section of F(m).

124

J. L. TAYLOR

14. Algebraic vs. Analytic Sheaves Serres Theorems

In this section we prove that, for all practical purposes, analytic projective varieties
and algebraic projective varieties and their coherent sheaves are the same. These are the
results of Serres famous GAGA paper Geometrie Algebrique et Geometrie Analytique.
Note that an algebraic subvariety V of Cn also has the structure of a holomorphic
subvariety V h of Cn . This structure does not depend on the embedding of V in Cn , since,
if V1 Cn and V2 Cm are algebraic subvarieties which are isomorphic as algebraic
varieties, then there is a biregular map of V1 onto V2 . Such a map is also biholomoprhic
and, hence, is an isomorphism between V1h and V2h . In other words, an affine variety V
has a unique structure of a holomorphic variety V h with the property that any algebraic
embedding of V in the Zariski space Cn is also a holomorphic embedding of V h in the
Euclidean space Cn .
14.1 Theorem. If X is an algebraic variety then there is a unique holomorphic variety
X h which is X as a point set, for which every open subset of X is open in X h and for
which every affine open subset has its natural holomorphic structure.
Proof. The uniqueness is clear, since the condition on affine subvarieties fixes the holomorphic structure locally and that fixes it period. To show existence, we must show that the
natural holomorphic structures on two affine open sets agree on their intersection. This
just amounts to the fact that a biregular map between a Zariski open subset U of an
algebraic subvariety of Cn and a Zariski open subset of an algebraic subvariety of Cm is
also a biholomorphic map. Thus, the topology and ringed space stucture of X h are well
defined and X h is locally isomorphic to a subvariety of Euclidean space. By problem 13.2,
X h is Hausdorff. The topology is second countable because X is covered by finitely many
affine open sets and each of these clearly has a second countable Euclidean topology. This
completes the proof.
Our next task will be to show that a coherent algebraic sheaf M on an algebraic variety
X gives rise to a coherent analytic sheaf Mh on X h . First, if M is any sheaf on X let M0
be the sheaf on X h which is the inverse image of M under the continuous map X h X.
Thus,
M0 (U ) = {M(W ) : U W open in X}.
Note that at each point x X, the stalks Mx and M0x are the same.
14.2 Definition. If M is any sheaf of O-modules on an algebraic variety X, then we
define a corresponding analytic sheaf Mh on X h by Mh = H O0 M0 .
The sheaf Mh = HO0 M0 is the sheaf of germs of the presheaf U H(U )O0 (U ) M0 (U )
and so its stalk at x X is
0
0
Mhx = lim
{H(U ) O0 (U ) M (U ) : x U } = Hx Ox0 Mx = Hx Ox Mx .

The second equality above follows easily from the definitions of direct limit and tensor
product (Problem 14.1).
The next theorem is a direct application of our work on faithful flatness in Chapter 7.

NOTES ON SEVERAL COMPLEX VARIABLES

125

14.3 Theorem. If X is an algebraic variety, then M Mh is an exact functor from the

category of sheaves of O-modules on X to the category of sheaves of H-modules on X h .
Proof. It is clear that M Mh is a functor from sheaves of O-modules to sheaves of
H-modules. We have that Mhx = Hx Ox Mx and, by Corollary 7.15, Hx is faithfully flat
over Ox . Since exactness is a stalkwise property, it follows that M Mh is an exact
functor.
14.4 Theorem. For an algebraic variety X, the functor M Mh takes
(a) Om to Hm for each m;
(b) coherent algebraic sheaves to coherent analytic sheaves;
(c) the ideal sheaf in O of an algebraic subvariety V to the ideal sheaf in H of the
corresponding holomorphic subvariety V h .
Proof. Part (a) is trivial since H(U ) O0 (U ) O0 (U ) = H(U ) for any Euclidean open set U .
If M is a coherent algebraic sheaf on X, then for each affine open subset U X,
there is a finitely generated O(U )-module M such that M|U ' O O(U ) M . Since M
is finitely generated, there is a surjective morphism O(U )m M and, since O(U ) is
Noetherian the kernel of this morphism is also finitely generated. Thus, there is a morphism
O(U )k O(U )m so that the sequence
O(U )k O(U )m M 0
is exact. If we apply the localization functor O O(U ) ( ) to this sequence, we obtain an
exact sequence of sheaves
Ok |U Om |U M|U 0
On applying the functor M Mh to this sequence and then using the previous theorem
and part (a) above, we conclude there is an exact sequence
Hk |U Hm |U Mh |U 0
Thus, by definition, Mh is coherent. This completes the proof of part (b).
If V is an algebraic subvariety of X, then the inclusion 0 IV O yields an inclusion
0 IVh H by Theorem 14.3. The image consists of the sheaf of ideals generated in H
by the image of IV0 . By Theorem 7.13, this is the ideal sheaf of V h in H. This completes
the proof of (c).
Now suppose that i : Y X is the inclusion of a holomorphic subvariety in an holomorphic variety X and M is a sheaf of Y H-modules on Y . Then i M the extension of
M by zero is a sheaf on X which is supported on Y . It is certainly a sheaf of iY Hmodules. However, iY H is the quotient sheaf X H/IY and, hence, i M is also a sheaf of
X H-modules. Exactly the same thing is true for an embedding i : Y X of algebraic
varieties and a sheaf M of Y O-modules the sheaf i M is then a sheaf of X O-modules.

126

J. L. TAYLOR

14.5 Theorem. If i : Y X is an embedding of a holomorphic variety Y in an holomorphic variety X and M is a coherent analytic sheaf on Y , then i M is a coherent analytic
sheaf on X. The analogous statement is true for an embedding of algebraic varieties and
a coherent algebraic sheaf.
Proof. . We prove this in the holomorphic case. The proof in the algebraic case is even
easier.
If M is a coherent analytic sheaf on Y then for each y Y there is a neighborhood
V Y of y and an exact sequence

Hk

Hm M|V 0

If V = U Y for an open set U X, then we may rewrite this in terms of the image of
this sequence under the exact functor i restricted to U :

(iY Hk )|U (iY Hm )|U (i M)|U 0

Now iY H is the cokernel of the inclusion j : IY X H and the ideal sheaf IY is coherent
and, hence, locally finitely generated, by Theorem 12.11. Thus, after possibly shrinking U
if necessary, we may find a morphism : X Hp X H with cokernel equal to iY H. This
and the lifting lemma (Lemma 12.12) allow us to construct the following commutative
diagram

kp
X Hmp |U
X H |U

m
k

y
y
XH

k
j y

|U

XH

jmy

|U

i (Y Hk )|U i (Y Hm )|U i M|U 0

y
y
0
0
with exact rows and columns. From the diagram it is evident that j m : X Hm |U
i M|U is surjective with kernel equal to the sum of the images of m : X Hmp |U X Hm |U
and k : X Hkp |U X Hm |U . Thus, i M is coherent.
This is a very important fact. It will allow us to prove a great many theorems about
coherent sheaves on projective varieties once we know they are true for coherent sheaves
on projective space. To make use of it in our study of the functor M Mh we need the
following:
14.6 Theorem. If i : Y X is an embedding of algebraic varieties and M is a coherent
algebraic sheaf on Y , then (i M)h = i (Mh ).
Proof. There is an inclusion M0 Mh of sheaves of Y O0 -modules which yields a morphism
i (M0 ) i (Mh ) of sheaves of X O0 -moldules. Clearly, i (M0 ) = (i M)0 and so we have

NOTES ON SEVERAL COMPLEX VARIABLES

127

a morphism (i M)0 i (Mh ) of sheaves of X O0 -modules. On tensoring this morphism

with X H we obtain a morphism (i M)h (i (Mh ))h = X H X O0 i (Mh ). Since i (Mh )
is already an X H-module, there is a morphism X H X O0 i (Mh ) i (Mh ) given by the
module action. Thus, we have a morphism
(i M)h i (Mh ).
To complete the argument, we need to show that this is an isomorphism on each stalk.
For this, we only need to consider stalks at points x Y , since the stalks of both sheaves
are zero off Y . At such a point x, this amounts to showing that the natural map
X Hx

X Ox0 Mx Y Hx Y Ox0 Mx

is an isomorphism. The latter sheaf can be written as

(X Hx X Ox0

Ox0 ) X Ox0 Mx

by using the fact that X Hx X Ox0 Y Ox0 ' Y Hx , which is deduced from tensoring X Hx
with the exact sequence 0 IY X O Y O 0 and using flatness and Theorem
14.4(c). The theorem then follows from the associativity of tensor product and the fact
that Y Ox0 X Ox0 Mx ' Mx .
We are now in a position to prove the three main theorems of Serres GAGA paper.
Note that if X is an algebraic variety and M is a sheaf on X then M and M0 have the
same global sections and, in fact, the same cohomology. The latter is due to the fact that a
flabby resolution of M0 will also be a flabby resolution of M when restricted to the Zariski
open sets. Thus, H p (X, M) = H p (X h , M0 ). Furthermore, the map m 1 m : M0
Mh induces a morphism H p (X h , M0 ) H p (X h , Mh ). Combining these facts gives us a
morphism H p (X, M) H p (X h , Mh ). The following is the first of Serres three GAGA
theorems:
14.7 Theorem. If X is a projective algebraic variety and M is a coherent algebraic sheaf
on X, then the natural map H p (X, M) H p (X h , Mh ) is an isomorphism for every p.
Proof. Since X is projective, we may assume that it is embedded as a subvariety of P n . If
i : X P n is the embedding, then i M is a coherent algebraic sheaf on P n by Theorem
14.5. Furthermore, the cohomology groups of M and i M are the same (Problem 14.2).
Likewise, i Mh is a coherent analytic sheaf on P n by Theorem 14.5. and the cohomology
groups of Mh and i Mh are the same. By Theorem 14.6, (i M)h = i (Mh ). It follows
that the theorem is true in general if it is true for coherent algebraic sheaves on P n . Thus,
we may assume that X = P n .
By Theorem 13.17, M is the quotient of a sheaf F which is a finite direct sum of sheaves
of the form O(k). If K is the kernel of F M, then we have a short exact sequence of
coherent algebraic sheaves
0 K F M 0

128

J. L. TAYLOR

If we apply ( )h to this sequence and use Theorems 14.3 and 14.4 then we have an exact
sequence
0 Kh F h Mh 0
of coherent analytic sheaves with F h a finite direct sum of sheaves of the form H(k).
If we apply the morphism of cohomology induced by ( )h to the long exact sequences
of cohomology corresponding to these short exact sequences, we obtain a commutiative
diagram
H p (K) H p (F) H p (M) H p+1 (K) H p+1 (F)

1 y
2 y
3 y
4 y
5 y
H p (Kh ) H p (F h ) H p (Mh ) H p+1 (Kh ) H p+1 (F h )
where, to save space, we have suppressed the P n in each cohomology group and written,
for example, H p (M) rather than H p (P n , M). In this diagram, 2 and 5 are always
isomorphisms by Corollary 13.14 and the fact that F is a direct sum of sheaves O(k).
Suppose that H p+1 (M) H p+1 (Mh ) is a isomorphism for every coherent algebraic sheaf
M. Then 4 is an isomrphism. This implies that 3 is surjective for every M and, hence,
that 1 is surjective. This, along with the fact that 2 and 4 are isomorphisms, implies that
3 is also injective and, thus, is an isomorphim. Thus, the proof will be complete if we can
show that there is a p0 so that H p (M) H p (Mh ) is an isomorphism for all M for p > p0 .

However, H p (M) = 0 for p > n because it can be computed using Cech

cohomology for
the Leray cover {Ui } for M. Using Cartans Theorem B from the next Chapter one can
show that {Ui } is also a Leray cover for Mh , but there is a more elementary way to see
that H p (Mh ) = 0 for large p. We know that P n is a 2n dimensional topological manifold,
from which it follows that every open cover of P n has a refinement in which no 2n + 2 sets
have non-empty intersection. Thus, H p (Mh ) = 0 for p > 2n and H p (M) H p (Mh ) is
trivially an isomorphism for p > 2n. This completes the proof.
The next theorem is the second of Serres GAGA theorems. To prove it, we first need
a lemma. If A is a subring of a ring B and E and F are A-modules, then the functor
( ) A B defines a natural morphism of A-modules
homA (E, F ) homB (E A B, F A B)
which, since the right side is a B-module, induces a B-module morphism
: homA (E, F ) A B homB (E A B, F A B).
14.8 Lemma. The morphism , defined above, is an isomorphism if A is Noetherian, E
is finitely generated over A and B is faithfully flat over A.
Proof. For a fixed module F , consider the two functors T and T 0 from A-modules to
B-modules defined by
T (E) = homA (E, F ) A B

NOTES ON SEVERAL COMPLEX VARIABLES

and

T 0 (E) =

129

homB (E A B, F A B) = homA (E, F A B).

Then : T T 0 is a transformation of functors and we are to show that it is an isomorphism on finitely generated modules. Clearly is an isomorphism if E = A, since T (A)
and T 0 (A) are both equal to F A B in this case. Similarly is an isomorphism if E = An
for some n. Note that T and T 0 are left exact since hom is left exact in its first variable
and B is faithfully flat over A. Since A is Noetherian, for each finitely generated A-module
E we can construct an exact sequence of the form
An Am E 0
On applying T and T 0 , we obtain a diagram
0 T (E) T (Am ) T (An )

y
y
y
0 T 0 (E) T 0 (Am ) T 0 (An )
with exact rows and with the last two vertical maps isomorphisms. It follows that the first
vertical map is also an isomorphism and the proof is complete.
14.9 Theorem. If M and N are two coherent algebraic sheaves on a projective algebraic
variety X, then every morphism of analytic sheaves Mh N h is induced by a morphism
M N of algebraic sheaves.
Proof. Let A denote the sheaf hom(M, N ). This is the sheaf which assigns to an open
U X the O(U )-module consisting of morphisms M|U N |U in the category of sheaves
of U O-modules. Similarly, let B = hom(Mh , N h ) be the analogous sheaf for the sheaves
of H modules Mh and N h . The functor ( )h clearly defines a sheaf morphism A0 B
and, since B is an H-module, this induces a morphism Ah = H A0 B, where, in
this argument, will mean tensor product relative to O. We claim that this morphism
Ah B is an isomorphism. As usual, it suffices to check this for the stalks at each point
of X. The fact that M is coherent and, hence, locally finitely generated implies that each
Ox -module homomorphism from Mx to Nx extends to a morphism from M|U to N |U for
some neighborhood U of x and that a morphism from M|U to N |U which vanishes at x
also vanishes in a neighborhood of x. These two statements, taken together, and their
analogues for coherent analytic sheaves mean that
Ax = hom(Mx , Nx )

and Bx = hom(Mhx , Nxh )

We also have that

Ahx = hom(Mx , Nx ) Hx ,

Mxh = Mx Hx

and

Nxh = Nx Hx

Thus, our claim will be established if we can show that the natural homomorphism
hom(Mx , Nx ) Hx hom(Mx Hx , Nx Hx )

130

J. L. TAYLOR

is bijective. But this follows from the previous lemma since Ox is Noetherian, Mx is finitely
generated and Hx is faithfully flat over Ox by Corollary 7.15.
To finish the proof, we consider the morphisms
H 0 (X, A) H 0 (X h , Ah ) H 0 (X, B).
The first of these morphisms is an isomorphism by Theorem 14.7 provided we can show that
A is coherent. This is done in Problem 14.3. The second morphism is an isomorphism by
the claim proved in the previous paragraph. Thus, the composition H 0 (X, A) H 0 (X, B)
is an isomorphism. This completes the proof of the theorem since a global section of B is
a morphism Mh N h while a global section of A is a morphism M N .
The geometric fiber of a coherent analytic sheaf S at a point x is the Hx module
Sx /Mx Sx , where Mx is the maximal ideal of Hx . The geometric fiber of a coherent algebraic
sheaf is defined analogously.
14.10 Lemma. If S is a coherent analytic sheaf on an holomorphic variety X, x X and
F H 0 (X, S) is a set of sections which generates the geometric fiber of S at x, then F
generates S|U for some neighborhood U of x. The analogous statement is true for coherent
algebraic sheaves.
Proof. It follows from Nakayamas Lemma that if F generates Sx /Mx Sx then it generates
Sx (Problem 14.5). However, by Problem 12.4 the set of y at which F fails to generate Sy
is a closed subvariety. Hence, there is a neighborhood U of x such that F generates S|U .
The third and most difficult of Serres GAGA theorems is the following:
14.11 Theorem. If X is a projective algebraic variety and M is a coherent analytic sheaf
on X h , then there is a coherent algebraic sheaf N on X such that N h ' M. Furthermore,
N is unique up to isomorphism.
Proof. The uniqueness is an immediate consequence of the preceding theorem.
Claim 1. The theorem is true if it is true for X = P n for all n.
If i : X P n is an embedding of X as a subvariety of P n , then i M is a coherent
analytic sheaf on P n . Suppose there is a coherent algebraic sheaf S on P n with S h ' i M.
Then we claim that S is i N for a coherent algebraic sheaf N on X. In fact, if I is the
ideal sheaf of X and f Ix for some x P n , then multiplication by f determines
an endomorphism : S|U S|U for some nieghborhood U of x. The corresponding
morphism h : S h |U S h |U is still multiplication by f and is zero in a neighborhood of
x because S h ' i M and M is a sheaf of X H-modules and the H-module action on i M
factors through the quotient map H X H. But if h vanishes in a neighborhood then
so does by Theorem 14.3. Thus, we have proved that IS = 0. This implies that S is
supported on X and its restriction to X is a sheaf N of X O-modules. It is easy to see
that N is a coherent algebraic sheaf on X (Problem 14.4) and, obviously, i N = S. Now
Theorem 14.6 implies that i N h ' (i N )h = S h ' i M. On restricting to X, this implies
that N h ' M. Thus, the theorem is true for any projective variety if it is true for P n .
We have reduced the proof to the case where X = P n . We will now prove it in this
case by induction on n. It is trivial when n = 0 since P 0 is a point and coherent algebraic

NOTES ON SEVERAL COMPLEX VARIABLES

131

and analytic sheaves are just finite dimensional vector spaces. Thus, we will assume that
n > 0 and that the theorem is true in dimensions less than n.
Note that for a coherent analytic sheaf M on P n we may twist by H(k) to construct a
coherent analytic sheaf M(k) = MH H(k) for each k just as we did for coherent algebraic
sheaves. Note also that N (k)h ' (N h )(k) due to the fact that O(k)h = H(k).
Claim 2. Let E be a hyperplane in P n and A a coherent analytic sheaf on E. Then,
under our induction assumption, H p (E, A(k)) = 0 for large enough k.
We have that E is the subvariety of P n defined by the zero set of a linear functional
on Cn+1 (a homogeneous polynomial of degree one). Then E is a copy of P n1 and so,
by assumption, the theorem is true for X = E. That H p (E, A(k)) = 0 for a coherent
analytic sheaf A on E and large enough k then follows from Theorems 14.7 and 13.18 and
our induction assumption which implies that A = B h for some coherent algebraic sheaf B.
The key to the proof of the Theorem is the next claim:
Claim 3. Under our induction assumption, for every coherent analytic sheaf M on X
there is an integer kM such that for every k > kM the sheaf M(k) is generated over H by
its space of global sections H 0 (P n , M(k)).
Note that if M(k)x is generated by H 0 (P n , M(k)) then M(k + p)x is generated by
H (P n , M(k+p)) for all p > 0. This is due to the fact that M(k+p)x = M(k)x Hx H(p)x
and H(p)x is generated by its global sections if p 0 (since H(p) is the sheaf of holomorphic
sections of a line bundle, one only needs to have one global section which is non-vanishing
at x in order to have the global sections generate H(p)x and the existence of such a section
follows from Theorem 13.12). In view of these remarks, the compactness of P n , and Lemma
14.10, to prove Claim 3 it suffices to prove that for each x P n there is a k for which the
geometric fiber of the module M(k)x is generated by H 0 (P n , M(k)).
Now let E be a hyperplane in P n and let IE be its ideal sheaf in H. We may as well
assume that E is the hyperplane on which the coordinate function z0 vanishes. Consider
the exact sequence
0

0 IE H

EH

Now z0 may be regarded as a section of H(1) and so multiplication by z0 defines a morphism

from H(1) to H. The image of this morphism is the ideal sheaf IE . Thus, we have an
isomorphism H(1) IE . Thus, the above exact sequence becomes
0 H(1) H

EH

Which, on tensoring with M relative to H, yields an exact sequence

0 C M(1) M B 0
Where B = E H H M and C = torH
1 (E H, M). If we tensor this with H(k), we get the
sequence of coherent analytic sheaves
0 C(k) M(k 1) M(k) B(k) 0.

132

J. L. TAYLOR

If we set Lk = ker(M(k) B(k)), then this sequence breaks up into two short exact
sequences of coherent analytic sheaves
0 C(k) M(k 1) Lk 0
and
0 Lk M(k) B(k) 0.
Now we can apply the long exact sequences of cohomology to these two short exact sequences. The relevant parts for us are
H 1 (P n , M(k 1)) H 1 (P n , Lk ) H 2 (P n , C(k))
and
H 1 (P n , Lk ) H 1 (P n , M(k)) H 1 (P n , B(k))
Now B = E H H M and C = torH
1 (E H, M) are coherent sheaves of H-modules but the
action of H factors through E H and, hence, they are actually coherent analytic sheaves
on E (Problem 14.4). It follows that B(k) and C(k) are also coherent analytic sheaves on
E. By our induction assumption this means that they are images under the functor ( )h
of coherent algebraic sheaves on E. It then follows from Theorem 14.7 and Therem 13.18
that B(k) and C(k) have vanishing pth cohomology for p > 0 and k sufficiently large. Thus,
the above sequences imply that for large k we have surjective maps
H 1 (P n , M(k 1)) H 1 (P n , Lk ) and

H 1 (P n , Lk ) H 1 (P n , M(k))

which implies that

dim H 1 (P n , M(k 1)) dim H 1 (P n , Lk ) dim H 1 (P n , M(k)).
At this point we must appeal to the results of Chapter 16 specifically, to the Cartan-Serre
Theorem which states that all cohomology spaces of a coherent analytic sheaf on a compact
holomorphic variety are finite dimensional vector spaces. Thus, the spaces H 1 (P n , M(k))
are finite dimensional and, by the above, the dimension is a non-increasing function of k
for sufficiently large k. This implies that eventually the dimension must become constant
as k increases. Thus, for sufficiently large k we have that H 1 (P n , Lk ) H 1 (P n , M(k))
is a surjective map between finite dimensional vector spaces of the same dimension. This
implies that it is injective as well. If we use this fact on the long exact sequence of
cohomology
H 0 (P n , M(k)) H 0 (P n , B(k)) H 1 (P n , Lk ) H 1 (P n , M(k))
we conclude that H 0 (P n , M(k))
H 0 (P n , B(k)) is surjective for k sufficiently large. We
also know, by the induction hypothesis, that the coherent sheaf B on E is the image under
( )h of a coherent algebraic sheaf on E and, hence, that B(k)x is generated by H 0 (P n , B(k))
if k is sufficiently large (since the analogous thing is true of coherent algebraic sheaves by
Theorem 13.16). We conclude that H 0 (P n , M(k)) generates B(k)x = M(k)x Hx E Hx .

NOTES ON SEVERAL COMPLEX VARIABLES

133

The latter module is just the quotient module M(k)x /(IE )x M(k)x . Since H 0 (P n , M(k))
generates this quotient module of M(k)x , it generates the geometric fiber of M(k)x . This
completes the proof of Claim 3 and puts us in a position to complete the proof of the
Theorem.
By Claim 3 we know that if M is a coherent analytic sheaf then there is an integer k
such that M(k) is generated by H 0 (P n , M(k)). Since H 0 (P n , M(k)) is finite dimensional
by Theorem 16.18, there is a surjection Hp M(k) for some p. If we twist this morphism
by the sheaf H(k), we obtain a surjection Hp (k) M. Now by applying the same
analysis to the kernel of this map, which is also a coherent analytic sheaf, we obtain an
exact sequence of the form

Hq (j) Hp (k) M 0
Now Hq (j) = Oq (j)h and Hp (k) = Op (k)h and so, by Theorem 14.9, the morphism
is induced by a morphism of coherent algebraic sheaves : Oq (j) Op (k). If N is
the cokernel of , then the exact functor ( )h applied to the exact sequence

Oq (j) Op (k) N 0
yields an exact sequence

Hq (j) Hp (k) N h 0.
But this implies that M ' N h which completes the induction and the proof of the theorem.
The results of Theorems 14.7, 14.9 and 14.11 (Serres Theorems 1, 2, and 3) can be
summarized as follows:
14.12 Theorem. If X is a projective algebraic variety, then the functor M Mh is a
cohomology preserving equivalence of categories from the category of coherent algebraic
sheaves on X to the category of coherent analytic sheaves on X h .
The above results have a wide variety of applications. We state some of these below, but
prove only a couple of them. The proofs of the others require knowledge of results from
algebraic geometry which we have not developed here. For a more complete discussion of
applications we refer the reader to Serres paper.
At this point, we will drop the use of the X h notation except in situations in which it
is needed to avoid confusion. If X is an algebraic variety then will generally also use X
to denote the corresponding holomorphic variety that is, we will think of an algebraic
variety as having two ringed space structures one algebraic and one holomorphic. We
will call a holomorphic variety X algebraic if it is the holomorphic variety associated to
some algebraic variety.
Our first application is the following theorem of Chow:
14.13 Corollary. If X is a projective variety, then every holomorphic subvariety of X is
algebraic.
Proof. Let Y be an analytic subvariety of X and consider the ideal sheaf IY . Then the
quotient H/IY is a coherent analytic sheaf on X which is isomorphic to iY H, where

134

J. L. TAYLOR

i : Y X is the inclusion. The support of the sheaf H/IY is clearly the subvariety
Y . Now by Theorem 14.12, there is a coherent algebraic sheaf N with the property that
N h ' H/IY . The support of N is the same pointset as the support of N h due to the fact
that Hx is faithfully flat over Ox for each x. Thus, the support of N is Y . However, the
support of a coherent algebraic sheaf is an algebraic subvariety (see Problem 12.4 and note
that the proof works equally well in the algebraic case). This completes the proof.
One can combine this with another result of Chow (on representing a general algebraic
variety as the image under a regular map of a dense open subset of a projective variety)
to obtain (cf. Serre):
14.13 Corollary. If X is an algebraic variety, then every compact holomorphic subvariety
of X h is algebraic.
With this and a little additional work, one can prove (cf. Serre):
14.14 Corollary. Every holomorphic map from a compact algebraic variety to an algebraic variety is regular.
This has the obvious consequence that:
14.15 Corollary. A compact holomorphic variety has at most one structure of an algebraic variety (up to isomorphism).
The category of algebraic vector bundles on an algebraic variety may be identified with
the category of locally free finite rank sheaves of O modules a vector bundle is identified
with its sheaf of sections. In the same way, the category of holomorphic vector bundles may
be identified with the category of locally free finite rank sheaves of H-modules. Clearly,
the equivalence of categories M Mh of Theorem 14.12 has the property that M is free
of finite rank if and only if Mh is free of finite rank. The corresponding functor on vector
bundles is just the functor which assigns to an algebraic vector bundle : E X over an
algebraic variety the holomorphic bundle h : E h X h obtained by putting the canonical
analytic structure on both total space and base. Thus, we have proved:
14.16 Corollary. If X is a projective algebraic variety then the category of algebraic
vector bundles on X is equivalent to the category of holomorphic vector bundles on X h
under the natural correspondence.
Serres paper contains a more general result of the above type which concerns bundles
with structure groups other than Gln (C). It also contains, as an entirely different kind of
application of the GAGA theorems, a proof of the following conjecture of A. Weil:
14.17 Corollary. If V is a projective non-singular variety defined over an algebraic number field K. Then the complex projective variety determined by an embedding of K in C
has Betti numbers which are independent of the embedding that is chosen.

NOTES ON SEVERAL COMPLEX VARIABLES

135

14. Problems
1. Prove that if M and N are sheaves of modules over a sheaf of rings R, then
(M R N )x = lim
{M(U ) R(U ) N (U ) : x U } = Mx Rx Nx
2.

3.

4.

5.

where the first equality is the definition of (M R N )x .

Prove that if i : Y X is an embedding of Y as a closed subspace of a topological
space X and if S is any sheaf on Y , then H p (X, i S) ' H p (Y, S). Hint: Use a flabby
resolution of S and take i of it.
Prove that if M and N are coherent algebraic (analytic) sheaves on an algebraic (holomorphic) variety X, then the sheaf hom(M, N ) is also a coherent algebraic (analytic)
sheaf. Hint: For coherent algebraic sheaves prove this first when M is Om and then
represent a general M as the cokernel of a morphism Ok Om . The proof is analogous
for coherent analytic sheaves.
Suppose Y is a subvariety of X, I is the ideal sheaf of Y and S is a coherent sheaf on
X which satisfies IS = 0. Prove that S is supported on Y and its restriction to Y is a
coherent sheaf on Y . It doesnt matter whether the sheaves and spaces are algebraic or
analytic.
Use Nakayamas Lemma to prove that if S is a finitely generated module over a Noetherian local ring A, M is the maximal ideal of A and F is a subset of S whose image
in S/M S generates S/M S, then F generates S.

136

J. L. TAYLOR

15. Stein Spaces

Stein spaces play the role in the study of holomorphic varieties and functions that is
played by the affine varieties in the algebraic theory. The most important theorems in the
subject of several complex variables are Cartans Theorems A and B. Cartans Theorem
A, says that a coherent analytic sheaf on a Stein space is generated over H by its global
sections and Cartans Theorem B which says that a coherent analytic sheaf on a Stein
space has vanishing cohomology in degrees greater than zero. In this section we define
the category of Stein spaces and lay the groundwork for proving Cartans theorems. The
proofs of the theorems themselves will be carried out in the next chapter, where we develop
and employ some approximation results for coherent analytic sheaves.
The key result in this section is a vanishing theorem which states that a coherent analytic
sheaf defined in a neighborhood of a compact polydisc has vanishing higher cohomology
on the polydisc. The strategy of the proof is to construct a free resolution for each such
sheaf like the one in the following theorem that is, a terminating chain of syzygys.
15.1 Lemma. Let be an open polydisc in Cn and suppose there is an exact sequence
of sheaves of H-modules on of the form
0 Hpm Hp1 Hp0 S 0
Then H p (, S) = 0 for p > 0 and the sequence of global sections
0 (, Hpm ) (, Hp0 ) (, S) 0
is also exact.
Proof. The exact sequence of sheaves in the hypothesis can be decomposed into a collection
of short exact sequences of coherent analytic sheaves
0 Lk Hpk Lk1 0
where Lk = ker{Hpk Hpk1 } for k > 0, Lm = 0 and L1 = S. For such a sequence, the
long exact sequence of cohomology and the fact that Hpk has vanishing qth cohomology
for q > 0 imply that if Lk is also a sheaf with vanishing qth cohomology for all q > 0 then
Lk1 is as well and the sequence
k

0 (, Lk ) (, Hp ) (, Lk1 ) 0
is exact. The theorem follows from descending induction using this fact, beginning on the
left at k = m.
In a corollary to Hilberts Syzygy Theorem (Corollary 3.14) we proved that every finitely
generated module over the local ring n H has a terminating free finite rank resolution a
terminating syzygy. Thus, given a coherent sheaf S on an open set in Cn , a sequence like
the one in the above lemma can always be constructed at each point. Using the results on
coherence of Chapter 12, we are able to construct such a sequence in a neighborhood of
any point:

NOTES ON SEVERAL COMPLEX VARIABLES

137

15.2 Lemma. If S is a coherent analytic sheaf defined on an open set U in Cn , then for
each point x U there is a neighborhood W of x in U and an exact sequence:
n

0 Hpn |W . . . Hp1 |W Hp0 |W S|W 0

In other words, S locally has a free finite rank resolution of length n.
Proof. Let x be any point of U . By the definition of analytic coherence and the results of
Chapter 12, there is a neighborhood W of x and a surjection 0 : Hp0 |W S|W 0 such
that ker is also coherent. Thus, by shrinking W if necessary, we may express ker as
the image of a morpism 1 : Hp1 |W Hp0 |W . Continuing in this manner, we construct
the sequence as above up to stage n 1, where we have a morphism n1 : Hpn1 |W
Hpn2 |W . By Hilberts Syzygy Theorem (Theorem 3.13), the stalk (ker n1 )x is a free
finite rank Hx -module. We choose germs which form a basis for (ker n1 )x and then
choose representatives in a neighborhood (which we may as well assume is W ) for these
germs. The resulting finite set of sections defines a morphism n : Hpn |W ker n1
which is an isomorphism at x. By Problem 12.4, is an isomorphism in a neighborhood
of x. Therefore, after again shrinking W if necessary, the morphism n completes the
construction of our resolution.
The final step in proving a vanishing theorem for the cohomology of coherent sheaves
on a polydisc will be to piece together the local resolutions given by the above theorem to
obtain such a resolution on the entire polydisc. This is not so easy to do. The construction is based on a factorization lemma (Cartans Lemma) for holomorphic matrix valued
functions. The best proof of this lemma uses the infinite dimensional implicit function
theorem. In the next theorem we prove the version of this that we will use. Before we
can prove this theorem, we need to establish some preliminary facts about Banach space
operators.
Let A : X Y be a surjective bounded linear map between two Banach spaces. By the
open mapping theorem, A is an open map. This implies that there is a constant K with
the property that for each y Y there is an x X such that Ax = y and ||x|| K||y||.
The least such K is the inversion constant for A.
15.3 Lemma. If A : X Y is a surjective continuous linear map between Banach spaces,
then there is a > 0 and a K > 0 such that B is surjective with inversion constant less
than or equal to K whenever B : X Y is a bounded linear map with ||A B|| < .
Proof. Let K0 be the inversion constant for A and choose = (2K0 )1 . For B : X Y
with ||A B|| < and y Y we seek an x such that Bx = y. We choose u0 such that
Au0 = y and ||u0 || K0 ||y||. We now choose inductively a sequence {un } of elements of
X such that
Aun = (A B)un1 , ||un || 2n K0 ||y||
The choice of clearly makes this possible. If xn = u0 + u1 + + un , then the sequence
xn converges in X and
y Bxn = Au0 Bxn = Au1 B(xn u0 )
= Au2 B(xn u0 u1 ) = = Aun+1

138

J. L. TAYLOR

and so it is clear that Bx = y if x = lim xn . It also follows from our estimate on the norms
of the un that ||x|| 2K0 . This establishes the Lemma with K = 2K0 .
If f is a function from an open subset U of a Banach space X to a Banach space Y ,
then the derivative (if it exists) of f at x U is a bounded linear map f 0 (x) : X Y with
the property that
lim ||u||1 ||f (x + u) f (x) f 0 (x)u|| = 0.
u0

15.4 Theorem. Let X and Y be Banach spaces and let f : U Y be a (non-linear)

function from a neighborhood of 0 in X into Y . If f 0 (x) exists at each x U , is a continuous
function of x and is surjective at x = 0, then the image of f contains a neighborhood of
f (0).
Proof. By the previous Lemma and the continuity of f 0 , we may assume that U is small
enough that f 0 (x) is surjective with inversion constant bounded by some constant K for
all x U . We may also assume that U is convex so that if x and x + u lie in U then so
does the line segment joining them. In this case,
Z
0

f (x + u) f (x) f (x)u =

[f 0 (x + tu) f 0 (x)]u dt

Thus, by shrinking U if necessary and using the continuity of f 0 again, we may assume
that
||f (x + u) f (x) f 0 (x)u|| < (2K)1 ||u||
for x, x + u U .
The remainder of the proof is just an application of Newtons method. We choose > 0
so that ||u|| < 2 implies that u U . We will show that for ||y f (0)|| < K 1 we can
solve the equation f (x) = y. We proceed as in Newtons method, using x0 = 0 as our
initial guess. We then choose x1 X so that f 0 (0)x1 = y f (0) and ||x1 || K||y f (0)||.
Note that ||x1 || < . We then inductively choose xn so that xn = xn1 + un where
f 0 (xn1 )un = y f (xn1 ).
and
||un || K||y f (xn1 )||.
Then we have
||y f (xn )|| = ||y f (xn1 ) (f (xn ) f (xn1 ))||
= ||f 0 (xn1 )(xn xn1 ) (f (xn ) f (xn1 ))|| < (2K)1 ||un ||
So that

||un || < 21 ||un1 ||

Now ||u1 || < implies that ||u2 || < 21 and, in general, that ||un || < 2n+1 . This
implies that the sequence {xn } is contained in U and converges to an element x U . Our
estimate above on ||y f (xn )|| shows that f (x) = y. This completes the proof.

NOTES ON SEVERAL COMPLEX VARIABLES

139

Roughly speaking, the above theorem says that a certain non-linear problem has a
solution if the linearized version is solvable. In our application of this result, the solvability
of the linearized problem is given by the next lemma. In this lemma and in what follows
we will use the following geometric situation. By an open (compact) box in Cn we will
mean an open (compact) set U which is the cartesian product of intervals one from each
of the 2n real and imaginary coordinate axes. An aligned pair of open (compact) boxes
will be a pair (U1 , U2 ) which is the cartesian product of two ordered sets of intervals which
are identical except in one (real or imaginary) coordinate and in that coordinate the two
intervals are overlapping. The coordinate in which the defining intervals are allowed to be
different will be called the exceptional coordinate. It is clear from the definition that if
(U1 , U2 ) is an aligned pair of boxes, then U1 U2 and U1 U2 are also boxes and they are
obtained from U1 and U2 by taking intersection or union of the two defining intervals in
the exceptional coordinate and leaving the defining intervals in all other coordinates the
same.
15.5 Lemma. Let (U1 , U2 ) be an aligned pair of open boxes in Cn . Then each bounded
holomorphic function f on U1 U2 is the difference f1 f2 of a bounded holomorphic
function f1 on U1 and a bounded holomorphic function f2 on U2 .
Proof. Without loss of generality we may assume that the exceptional coordinate for
(U1 , U2 ) is x1 , so that the pair (U1 , U2 ) has the form
U1 = I1 iK W
U2 = I2 iK W
where I1 and I2 are overlapping open intervals on the line, K is an open interval on the
line and W is an open box in Cn1 .
We choose a bounded C function on R that is one in a neighborhood of I1 I2
and is zero in a neighborhood of I2 I1 . We then consider to be a function defined on
Cn which is constant in all variables except x1 . Then (1 )f extends by zero to be a
bounded C function g1 in U1 while f extends by zero to be a bounded C function g2
in U2 . Furthermore, on U1 U2 , g1 g2 = f . In other words, we have solved our problem
in the class of bounded functions which are C in the variable z1 and holomorphic in
the remaining variables. Now we need to modify this solution to arrive at one which is
holomorphic in z1 as well.
The fact that g1 g2 = f is holomorphic on U1 U2 implies that
g1
g2
=
z1
z1
on U1 U2 and that implies that they define a bounded C function g on U1 U2 which
gi
is
on Ui . Let V be an open set with compact closure in (I1 I2 ) K and let 1 be a
z1
C function of z1 which is 1 on V and which has compact support in (I1 I2 ) K. Set
2 = 1 1 . We then proceed as in the proof of Dolbeaults Lemma. We set D = U1 U2
and
Z
1
i (1 )g(1 , z2 , . . . , zn ) d1 d1
hi (z) =
2i D
1 z1

140

J. L. TAYLOR

and let h = h1 + h2 . As we noted in the proof of Dolbeaults Lemma, (1 z1 )1 is

integrable on any bounded subset of C so that the functions hi are bounded C functions
on U1 U2 . They are also holomorphic in (z2 , . . . , zn ). Since 1 is compactly supported
in D z1 , so that the line integral term in the generalized Cauchy theorem can be made to
h1
vanish, we may conclude as in the proof of Theorem 11.3 that
= 1 g in U1 U2 . This
z1
h1
means that
= g on V W . On the other hand, because 2 vanishes on V the function
z1
h
h2 is holomorphic on V W . Thus,
= g in V W . However, h is independent of the
z1
h
choice of V and and, hence,
= g in all of U1 U2 .
z1
We now have a bounded C function h on U1 U2 which is holomorphic in (z2 , . . . , zn )
and which satisfies
h
gi
=
, in Ui
z1
z1
then fi = gi h is bounded and holomorphic in Ui and f = f1 f2 on the intersection
U1 U2 . This completes the proof.
A Banach algebra is a complex algebra A with a norm which makes A a Banach space
and which is submultiplicative that is, satisfies ||ab|| ||a||||b|| for all a, b A. We
will only deal with Banach algebras with an identity of norm one. The set of invertible
elements in a Banach algebra A form a group which we will sometimes denote A1 . The
submultiplicative property, and the fact that Banach algebras are Banach spaces and,
hence, are complete, allows one to use power series arguments. These immediately yield
the following elementary results from Banach algebra theory:
15.6 Theorem. If A is any banach algebra, then
(a) if a A and ||a|| < 1 then 1 a has an inverse (1 a)1 =

an ;

n=0

(b) A1 is open in A, inversion is continuous in A1 and A1 is a topological group;

X
an
(c) there is a map a exp(a) =
from A to A1 which is a homeomorphism
n!
n=0
from some neighborhood of 0 in A to the neighborhood {b : ||1 b|| < 1} in A1 ;

X
(1 b)n
(d) on {b A : ||1 b|| < 1 the map b log b =
is an inverse for exp;
n
n=1
(e) the subgroup of A1 generated by the image of exp is open and is equal to the
connected component of the identity in A1 .
In what follows, Gln (C) will denote the group of invertible nn complex matrices. This
is the group of invertible elements of the algebra Mn (C) of all n n complex matrices.
The latter is a Banach algebra under the standard matrix norm
||a|| = sup{||ax|| : x Cn , ||x|| 1}

NOTES ON SEVERAL COMPLEX VARIABLES

141

and, hence, with the topology determined by this norm, Gln (C) is a topological group. If U
is a domain in Cn , we will also be concerned with the Banach algebra Hb (U, Mn ) of bounded
Mn -valued functions on U . Here the norm is given by ||f || = sup{||f (x)|| : x U }. The
invertible group of this Banach algebra is the group of bounded holomorphic functions
with values in Gln (C).
15.7 Theorem. If U Cn is the Cartesian product of simply connected open subsets of
C and K is a compact subset of U , then each holomorphic mapping f : U Gln (C) may
be uniformly approximated on K by holomorphic mappings from Cn to Gln (C).
Proof. By the Riemann mapping theorem we may, without loss of generality, assume that
U is a polydisc centered at the origin. We let V be an open polydisc centered at the
origin, containing K and with compact closure in U . Then, on V , f and f 1 are bounded
holomorphic functions with values in Gln (C) that is, f is an element of the invertible
group Hb (U, Gln (C)) of the Banach algebra Hb (V, Mn (C)). Then a curve t ft , t [0, 1]
in Hb (V, Gln (C)) joining f to the constant matrix f (0) may be constructed by setting
ft (z) = f (tz) for each t [0, 1]. Since Gln (C) itself is connected (Problem 15.2), this
proves that Hb (V, Gln (C)) is connected. By Theorem 15.6(e), this implies that f is a
product of elements in the range of the exponential function. Hence, on V ,
f = exp(g1 ) exp(g2 ) . . . exp(gk ) with

g1 , g2 , . . . gn Hb (V, Mn (C)).

Now each gi may be regarded as a matrix with entries which are bounded holomorphic
functions on V . By truncating the power series of each entry of each gi we may approximate
each gi by matrices hi with polynomial entries as closely as we like in the uniform topology
on K. Then
f = exp(h1 ) exp(h2 ) . . . exp(hk )
will be a holomorphic Gln (C)-valued function on all of Cn . Clearly, the hi can be chosen
so that f approximates f arbitrarily closely on K.
The next lemma is the key to the vanishing theorem we are seeking:
15.8 Cartans Lemma. Let (K1 , K2 ) be an aligned pair of compact boxes in Cn . Then
each holomorphic Gln (C)-valued function f defined in a neighborhood of K1 K2 may be
factored as f = f21 f1 where fi is a holomorphic Gln (C)-valued function in a neighborhood
of Ki for i = 1, 2.
Proof. We may construct an aligned pair (U1 , U2 ) of open boxes such that Ki Ui and
f and f 1 are holomorphic and bounded in U1 U2 . Let Ai be the Banach algebra
Hb (Ui , Mn (C)) for i = 1, 2 and let B be the Banach algebra Hb (U1 U2 , Mn (C)) and
consider the non-linear map : A1 A2 B defined by
(g1 , g2 ) = log((exp g2 )1 exp g1 ) on

U1 U2

Now it is easy to see that a function from a Banach algebra to itself which is defined by a
convergent power series is infinitely differentiable. It is also easy to see that the analogues of
the chain rule and the product rule hold for functions between Banach algebras. It follows

142

J. L. TAYLOR

that is infinitely differentiable in a neighborhood of (0, 0). We have that (0, 0) = 0 and
a simple calculation (Problem 15.3) shows that
0 (0, 0)(h1 , h2 ) = h1 h2
By Lemma 15.5, 0 (0, 0) is surjective. By Theorem 15.4, the image of contains a neighborhood of 0 in B. After composing with exp, we conclude that the map
(g1 , g2 ) (exp g2 )1 exp g1
has image which contains a neighborhood of the identity in B. This means that the
theorem is true for f sufficiently close to the identity in B. However, by the previous
theorem, we may approximate f arbitrarily closely on U1 U2 by a Gln (C)-valued function
h which is holomorphic on all of Cn . Then, for an appropriate choice of h we will have
f h1 sufficiently close to the identity in B that we may write f h1 = g21 g1 with gi a
holomorphic Gln (C)-valued function on Ui . Then the desired solution is f = f21 f1 with
f1 = g1 h and f2 = g2 .
The procedure used in the next theorem is sometimes known as amalgamation of syzygies.
15.9 Theorem. Let (K, L) be an aligned pair of compact boxes in Cn . Let S be a
coherent analytic sheaf defined in a neighborhood of K L and suppose that there are
exact sequences of analytic sheaves

2
1
0
0 Hpm m . . .
Hp1
Hp0
S 0

over a neighborhood of K and

m

0 Hqm . . . Hq1 Hq0 S 0

over a neighborhood of L. Then there exists an exact sequence of analytic sheaves
m

0 Hrm . . . Hr1 Hr0 S 0

defined in a neighborhood of K L.
Proof. The proof is by induction on the length m of the syzygies. If m = 0 then we have
a pair of isomorphisms:

0
0 Hp0
S 0

0 Hq0 S 0
With the first defined over a neighborhood of K and the second over a neighborhood of
L. On K L, the composition = 01 0 : Hp0 Hq0 is an isomorphism over a

NOTES ON SEVERAL COMPLEX VARIABLES

143

neighborhood of K L. This implies that p0 and q0 are the same integer k and that
is determined by a holomorphic Glk (C)-valued function in a neighborhood of K L. By
Cartans Lemma, this function may be factored as = 1 where () is a holomorphic
Glk (C)-valued function in a neighborhood of K (L). Then 0 1 = 0 1 in a
neighborhood of K L and so these two morphisms fit together to define an isomorphism
0

0 Hk S 0
over a neighborhood of K L, as required.
For the induction step, we assume that the theorem is true of all pairs of sequences,
as above, of length less than m and we suppose we are given a pair of length m. By
applying the Riemann mapping theorem in each variable we see that K L has arbitrarily
small neighborhoods U which are biholomorphically equivalent to open polydiscs. On a
sufficiently small such neighborhood, Lemma 15.1 implies that the sequences

2
1
0
. . .
(U, Hp1 )
(U, Hp0 )
(U, S) 0

. . . (U, Hq1 ) (U, Hq0 ) (U, S) 0

are exact. We can use the fact that (U, Hp0 ) = Hp0 (U ) is a free H(U )-module to construct
the morphism which makes commutative the diagram

0
Hp0 (U )
(U, S) 0

Hq0 (U ) (U, S) 0
Now is a matrix with entries which are holomorphic functions on U and, as such, it defines
a morphism of analytic sheaves : Hp0 Hq0 over U such that 0 = 0 . A similar
argument shows that we may construct a morphism of analytic sheaves : Hq0 Hp0
over U with 0 = 0 . We now modify each of the sequences so that the free modules
that appear in degree 0 will be identical. Thus, the first sequence is modufied by taking
its direct sum with the exact sequence
id

0 Hq0 Hq0 0
to obtain

1
0
Hp1 Hq0
Hp0 Hq0
S 0

where
1 = 1 id and
0 (f g) = 0 (f ). Similarly, we modify the second sequence by
taking its direct sum (on the other side) with the exact sequence
id

0 Hp0 Hp0 0

144

J. L. TAYLOR

to obtain
1

Hp0 Hq1 Hp0 Hq0 S 0

where 1 = id 1 and 0 (f g) = 0 (g).
Now we define two endomorphisms and of Hp0 Hq0 on U to be the maps which
have the following matrix representations relative to the direct sum decomposition:

1
, =
=
1

1
A calculation shows that 0 =
0,
0 = 0 and = 1 . Thus, is an isomorphism
and we have the following commutative diagram over U

1
0
Hp1 Hq0
Hp0 Hq0

S 0

Hp0 Hq1 Hp0 Hq0 S 0

Now by Cartans Lemma, can be factored over some neighborhood of K L as 1 ,
where is an automorphism of Hp0 Hq0 over some neighborhood of K and is an
automorphism of Hp0 Hq0 over some neighborhood of L. Then, in a neighborhood of
K L we have
0 =
0
which means that there is a single morphism
0 : H p 0 H q 0 S
over a neighborhood of K L such that 0 =
0 on a neighborhood of K and 0 = 0
on a neighborhood of L. If K is the kernel of 0 , then we have exact sequences

2
1
0 Hpm m . . .
Hp1
K 0

over a neighbohood of K and

0 Hqm . . . Hq1 1 K 0
over a neighborhood of L. Where p01 = p1 +q0 , q10 = q1 +p0 , 10 = 1 1 and 10 = 1 .
Since these sequences have length m 1, the induction hypothesis implies that there exists
an exact sequence
m

0 Hrm . . . Hr1 K 0
in a neighborhood of K L. Combining this with the morphism 0 gives us the required
sequence for S over a neighborhood of K L. This completes the proof of the theorem.

NOTES ON SEVERAL COMPLEX VARIABLES

145

15.10 Theorem. If U is an open polydisc in Cn and S is a coherent analytic sheaf on U ,

then for any compact subset K U there is an exact sequence of the form
0 Hpm . . . Hp1 Hp0 S 0
defined in some neighborhood of K. In other words, S has a free, finite rank resolution of
length n in a neighborhood of any compact subset of U .
Proof. By applying the Riemann mapping theorem in each coordinate, we may reduce the
problem to one in which U is an open box in Cn . Also, without loss of generality, we
may assume that K is a compact box contained in U . By Lemma 15.2, we know that
S has a resolution like the one above in a neighborhood of each point of U . If K is a
cartesian product I1 I2n of closed intervals, we partition each of these intervals into
m subintervals of equal length. This results in a partition of K into nm compact boxes.
By choosing m large enough, we may assume that S has a free finite rank resolution of
length n in a neighborhood of each of these boxes. Clearly, adjacent pairs of boxes are
aligned pairs and, hence, we may apply the previous theorem to conclude that there is a
free finite rank resolution in a neighborhood of the union of any two adjacent pairs. By
working from left to right along rows in which only the variable x1 changes we can prove
that there is a free finite rank resolution of length n for S in a neighborhood of the union
of the boxes in any such row. These unions provide a new partition of K into compact
boxes in which adjacent boxes are aligned. Repeating this argument one variable at a
time, we eventually end up with a free finite rank resolution of S in a neighborhood of K
as required.
15.11 Corollary. Let U be an open polydisc in Cn and let S be a coherent analytic sheaf
on U . Then H p (, S) = 0 for p > 0 and for any open polydisc with compact closure in
U.
Proof. This follows immediately from Lemma 15.1 and Theorem 15.10.
We are now in a position to define Stein spaces and to prove a vanishing theorem which
is a weak form of Cartans Theorem B.
15.12 Definition. Let X be a holomorphic variety. Then
(i) if K is a compact subset of X, then the holomorphically convex hull of K in X is
the set
b = {x X : |f (x)| sup |f (y)| f H(X)};
K
yK

b = K;
(ii) a compact subset K of X is said to be holomorphically convex in X if K
b is compact for every compact subset
(iii) X is said to be holomorphically convex if K
K X.
b is always a closed subset of X and so it will be compact if it is contained
Note that K
in a compact subset of X. Note also that if X is a holomorphic variety, U an open subset
of X and K a compact subset of U , then it makes sense to talk about the holomorphically
convex hull of K in X and in U . These are not necessarily the same. The open set

146

J. L. TAYLOR

U is said to be holomorphically convex if the holomorphically convex hull of K in U is

compact for every compact subset K of U that is, if it is holomorphically convex as a
holomorphic variety in its own right. It is easy to see that the intersection of any finite set
of holomorphically convex open subsets of a holomorphic variety is also holomorphically
convex (Problem 15.4).
15.13 Definition. A holomorphic variety X is said to be a Stein space if
(i) X is holomorphically convex;
(ii) for each x X the maximal ideal in Hx is generated by a set of global sections of
H;
(iii) the global sections of H separate points in X.
Note that conditions (ii) and (iii) are automatically inherited by open subsets and so an
open subset of a Stein space is also a Stein space provided it is holomorphically convex. In
particular, Cn is clearly a Stein space (Problem 15.5) and so each of its holomorphically
convex open subsets is also a Stein space. It is also easy to see that a closed subvariety of
a Stein space is a Stein space (Problem 15.6).
15.14 Definition. An open subset W of a holomorphic variety X is said to be an OkaWeil subdomain of X if W has compact closure in X and if there is a holomorphic map
: X Cn which, when restricted to W , is a biholomorphic map of W onto a closed
subvariety of (0, 1).
15.15 Theorem. If X is a Stein Space, K is a compact holomorphically convex subset of
X and U is an open subset of X containing K, then there exist an Oka-Weil subdomain
W of X such that K W W U and every coherent analytic sheaf defined on a
neighborhood of W is acyclic on W .
Proof. We may as well assume that U has compact closure U . If x U we may choose
a function f H(X) such that |f (x)| > 1 = ||f ||K , where ||f ||K denotes the supremum
norm of f on K. Since this inequality will continue to hold in some neighborhood of x and
since U is compact, we may choose a finite set of functions {fi }ki=1 so that
K W1 = {x U : |fi (x)| 1,

i = 1, . . . k} U

Thus, the functions {fi }ki=1 are the coordinate functions of a holomorphic map 0 : X Ck
which maps W1 into (0, 1). In fact, 0 is a proper holomorphic map of W1 into (0, 1)
since the inverse image in W1 of any compact subset of (0, 1) will be closed not only in
W1 , but also in the compact set U .
Now, by (ii) of Definition 15.13, for each point x of W 1 there is a finite set of global
sections of H which vanish at x and generate the maximal ideal of Hx . Without loss of
generality, we may assume that these functions all have modulus less than 1 at each point
of W1 . By Theorems 6.15 and 6.16, the map X Cm with these functions as coordinate
functions is a biholomorphic map of some neighborhood of x onto a closed subvariety of
some neighborhood of f (x). Since W 1 is compact, finitely many such neighborhoods will
cover W 1 . By adjoining all the corresponding functions to the set {fi }ki=1 we obtain a set
{fi }qi=1 of functions which are the coordinate functions of a holomorphic map 1 : X Cq

NOTES ON SEVERAL COMPLEX VARIABLES

147

which is proper on W1 and is also has the property that in a neighborhood of each point
x W 1 it is a biholomorphic map onto a closed subvariety of a neighborhood of f (x).
In particular, 1 is locally one to one. This means that the diagonal in W 1 W 1 has
a neighborhood V in which (x) (y) is non-vanishing except on the diagonal itself.
Another compactness argument and (iii) of Definition 15.13 show that we can find another
finite set of global sections of H (again with modulus less than one at points of W1 )
such that whenever (x, y) W 1 W 1 V there is some function f in this set such that
f (x) 6= f (y). By adjoining this finite set to the set {fi }qi=1 , we obtain a set {fi }m
i=1 of
functions which are the coordinate functions of a holomorphic map : X Cm which is
proper, injective and locally biholomorphic on W1 .
A proper, injective, continuous map from one locally compact space into another has
closed image and is a homeomorphism onto its image (Problem 15.7). Thus, the image Y
of on W1 is a closed subset of (0, 1) and is a holomorphic homeomorphism onto Y .
Furthermore, for each x W1 , maps some neighborhood of x biholomorphically onto an
open set in Y which is a closed holomorphic subvariety of an open set in (0, 1). It follows
that Y is a closed subvariety of (0, 1) and maps W1 biholomorphically onto Y . Thus,
W1 is an Oka-Weil subdomain of X.
For each polyradius r which has all coordinates less than or equal to 1 we set Wr =
1 ((0, r)). Then each Wr is obviously also an Oka-Weil subdomain of X. If r < s means
that each coordinate of r is less than the corresponding coordinate of s, then W r Ws
whenever r < s. Fix an r < 1 such that K Wr . If a coherent sheaf S is defined in
a neighborhood of W r , then we may choose an s > r such that Ws is contained in that
neighborhood. Then S may be considered a coherent analytic sheaf defined on Y (0, s)
and may be extended by zero to a coherent analytic sheaf on all of (0, s). It follows from
Corollary 15.11 that such a sheaf will be acyclic on (0, r) and, hence, on Y (0, r).
Thus, if we let W = Wr , then W is an Oka-Weil subdomain of X which contains K, has
compact closure contained in U and has the property that every coherent analytic sheaf
defined in a neighborhood of W is acyclic on W . This completes the proof.
15.16 Corollary. If X is a Stein Space then X is the union of a sequence {Wn } of OkaWeil subdomains such that W n Wn+1 for each n and each coherent analytic sheaf on
W n is acyclic on Wn .
Proof. Since X is countable at infinity, it is the union of an increasing sequence {Kn }
of compact sets. Suppose we have managed to find Oka-Weil subdomains Wj with the
required acyclic property for coherent sheaves and such that Kj Wj and W j1 Wj
bn is also
for j n. Then Cn = W n1 Kn is compact. Since X is a Stein space, C
compact. By Theorem 15.15, Cn is contained in an Oka-Weil subdomain Wn with the
required acyclic property. Thus, the Corollary is true by induction.
15.17 Corollary. Every holomorphic variety X has a neighborhood basis W with the
property that given any x X and any open set U containing x, there is a W W, with
x W such that W has compact closure W U and every coherent analytic sheaf defined
on a neighborhood of W is acyclic on W .
Proof. Every holomorphic variety has a neighborhood base consisting of closed subvarieties
of open polydiscs. Since a closed subvariety of an open polydisc is a Stein space, an

148

J. L. TAYLOR

application of Theorem 15.15 proves the corollary.

Finally, we have the following important application of Theorem 15.11:
15.18. Suppose Y is a closed subvariety of an open polydisc U and is an open polydisc
with compact closure in U . Then every holomorphic function on Y is the restriction
of a holomorphic function on .
Proof. On U we have an exact sequence of coherent analytic sheaves:
0
IY
UH
YH
0
By Theorem 15.11, the sheaf IY has vanishing pth cohomology on for p > 0. Thus, the
long exact sequence of cohomology implies that H() H( Y ) is surjective.
This is as far as we can go in the direction of proving that all coherent sheaves on a Stein
space are acyclic (Cartans Theorem B), without doing an approximation argument like
that used in the proof of the vanishing theorem for Dolbeault cohomology (Theorem 11.5).
However, to carry out such an argument we need a topology on the space of sections of a
coherent sheaf. Specifically, we need to establish that the space of sections of a coherent
analytic sheaf has a canonical Frechet space topology. This project is left to the next
chapter.

15. Problems
1.
2.
3.
4.

Prove Theorem 15.6.

Prove that Gln (C) is connected.
Calculate the derivative at the point (0, 0) of the function in the proof of Lemma 15.8.
Prove that the intersection of any finite collection of holomorphically convex open subsets of a holomorphic variety is also holomorphically convex. Then prove that a holomorphic variety has a locally finite open cover W with the property that any finite
intersection of sets from W is a Stein space.
5. Prove that Cn is a Stein space.
6. Prove that a closed subvariety of a Stein space is a Stein space.
7. Prove that a proper, injective, continuous map from one locally compact space into
another has closed image and is a homeomorphism onto its image.

NOTES ON SEVERAL COMPLEX VARIABLES

149

16. Fr
echet Sheaves Cartans Theorems
A topological vector space is a vector space with a topology under which the vector
space operations are continuous. A topological vector space is locally convex if it has a
neighborhood base at zero consisting of convex balanced sets, where a set is balanced if
it is closed under multiplication by scalars of modulus one. A topological vector space is
locally convex if and only if its topology is given by a family { } of seminorms, where
a semi-norm on a vector space X is a function from X to the non-negative reals which
satisfies
(x + y) (x) + (y) and (x) = ||(x)
for x, y X and C. The topology determined by a family { } of seminorms is the
topology in which a basis of neighborhoods at zero is given by the collection of all sets of
the form
{x X : (x) < }.
A continuous linear functional on a topological vector space is a linear complex valued
function which is continuous. The space of all continuous linear functionals is called the
dual of X and is denoted X . There are a number of topologies that can be put on X
which make it a locally convex topological vector space. They have different properties
and are used in different circumstances.
A Frechet space is a locally convex topological vector space F which is complete and
which has its topology defined by a sequence {n } of semi-norms. Without loss of generality, the sequence {n } may be chosen to be increasing. Equivalently, a Frechet space is
a topological vector space which is the inverse limit of a sequence of Banach spaces and
bounded linear maps. Equivalently, a Frechet space is a complete locally convex topological vector space with a topology defined by a translation invariant metric. A bounded set
in a Frechet space is a set B with the property that each of the defining semi-norms n is
bounded on B equivalently, B is bounded if for every neighborhood U of 0, there is a
positive number k such that B kU . A Frechet space F is called a Montel space if every
closed bounded subset of F is compact. A separated quotient of a topological vector space
is a quotient by a closed subspace.
We will assume knowledge of the following elementary facts concerning locally convex
topological vector spaces. The proofs can be found in any text on functional analysis or
topological vector space theory.
TVS 1. Closed subspaces, separated quotients and countable direct products of Frechet
spaces are Frechet spaces.
TVS 2. Closed subspaces, separated quotients, and countable direct products of Montel
spaces are Montel spaces.
TVS 3 (Open Mapping Theorem). A surjective continuous linear map from a Frechet
space to a Frechet space is an open map.
TVS 4 (Closed Graph Theorem). A linear map from a Frechet space to a Frechet
space is continuous if and only if its graph is closed.

150

J. L. TAYLOR

TVS 5 (Hahn-Banach Theorem). The strong form says that a real linear functional
on a subspace of a real vector space, dominated by a convex functional, extends to the
whole space with preservation of the dominance. We will need two corollaries of this:
(a) every continuous linear functional on a subspace of a locally convex topological
vector space extends to a continuous linear functional on the whole space;
(b) if B is a closed convex balanced set in a locally convex topological vector space X
and x0 X is a point not in B, then there exists a continuous linear functional f
on X such that |f (x)| 1 for all x B but |f (x0 )| > 1.
TVS 6. Every locally compact topological vector space is finite dimensional.
If U is a domain in Cn then H(U ) is a Frechet space in the topology of uniform convergence on compact sets. In fact, by Theorem 1.8, H(U ) is a Montel space. The same thing
is true for holomorphic functions on a variety:
16.1 Theorem. If U is an open subset of a holomorphic variety X, then H(U ) is a Montel
space in the topology of uniform convergence on compact subsets of U .
Proof. We may express U as the union of an increasing sequence {Kn } of compact subsets,
then the sequence {n } of seminorms defined by
n (f ) = sup{|f (x)| : x Kn }
determine the topology of H(U ). It is not obvious that H(U ) is even complete in this
topology. However, by Corollary 15.18, X has a neighborhood base consisting of sets W
for which W may be imbedded as a closed subvariety of an open polydisc and the
restriction map H() H(W ) is surjective. Since the kernel of this map consists of the
functions which vanish on W , it is closed in H(). Hence, H(W ) is a separated quotient
of a Montel space and is, therefore, Montel. Now for a general open set U , let {Wi } be
a countable open cover of U by sets with the aboveQ
property. Then the map f {f |Wi }
embedds H(U ) as a subspace of the Montel space i H(Wi ). In fact, it is embedded as
the closed subspace {{fi } : (fi )|Wi Wj = (fj )|Wi Wj i, j}. Thus, H(U ) is a Montel space.
Given a coherent analytic sheaf S on a holomorphic variety X, we will define a canonical
Frechet space topology on its space S(U ) of sections over an open set U . The strategy for
doing this is to do it first for small neighborhoods of the kind given by Corollary 15.17
and then to argue that this is enough to determine a Frechet space structure on spaces of
sections over arbitrary open sets.
If x X then there is a neighborhood W of x on which S is the cokernel of a morphism
of analytic sheaves : Hm Hk . By Corollary 15.17, by shrinking W if necessary, we
may assume that W has the property that ker is acyclic on W . From the long exact
sequence of cohomology, we conclude that : H(W )m H(W )k has S(W ) as cokernel,
so that S(W ) is a quotient of the Frechet space Hk (W ). If : H(W )m H(W )k has
closed image then S(W ) is a quotient of Hk (W ) by a closed subspace and, hence, is also
a Frechet space. Unfortunately, it is not trivial to prove that has closed image. This
requires an argument using the Weierstrass theorems which is reminiscent of the proof of
Okas Theorem. First a lemma:

NOTES ON SEVERAL COMPLEX VARIABLES

151

16.2 Lemma. Let U be an open set in Cn ,

(a) let z be in U and let M be a submodule of Hzk . Then the set of all f Hk (U ) such
that fz M is closed in Hk (U );
(b) if : Hm Hk is a morphism of analytic sheaves on U and is an open polydisc
with compact closure in U , then the induced map : Hm () Hk () has closed
image.
Proof. We first show that if part (a) holds for given integers n and k then part (b) holds
for the same n and k and arbitrary m. In fact, by part (a), every element in the closure of
the image of : Hm () Hk () has germ at z in the image of z at every point z .
That is, such an element is a section of the sheaf L = im{ : Hm Hk }. If K is the
kermel of this sheaf morphism, then we have a short exact sequence of sheaves on U

0
K
Hm
L
0.
From the long exact sequence of cohomology and the vanishing theorem of the previous
chapter (Corollary 15.11), we conclude that the corresponding sequence of sections over
is also exact. Thus, : Hm () L() is surjective. Since the closure of the image
of : Hm () Hk () is contained in L(), we conclude that the image and its closure
coincide.
The proof of part (a) will be by induction on n. Part (a) is trivially true in the case
n = 0 and so we assume that n > 0 and that the theorem is true whevever U is an open
subset of Cn1 . Under this assumption, we next prove part (a) when U is an open subset
U of Cn and k = 1.
Thus, let M be an ideal in Hz . We may as well assume that z = 0. After a linear change
of coordinates, if necessary, we may assume that there is a polydisc centered at 0 and
a function h H() which is regular in zn and has germ h0 belonging to M . Then the
Weierstrass preparation theorem implies that h0 is a unit times a Weierstrass polynomial.
Hence, by shrinking if necessary, we may assume that h itself is a Weierstrass polynomial
say of degree k. Then, by the Weierstrass division theorem, every germ f H0 has a
unique representation as f = gh + q where g is the germ of a function holomorphic in a
neighborhood of 0 and q is a polynomial in zn of degree less than k with coeficients which
are germs at 0 of holomorphic functions in the variables z1 , . . . , zn1 . Recall from the proof
of the Weierstrass division theorem, that g is given by an integral formula
Z
1
f (z 0 , ) d
g(z) =
2i ||=rn h(z 0 , )( z)
where z = (z 0 , zn ) (0, r) and r = (r0 , rn ) is chosen so that h(z) has no zeroes on the
part of (0, r) where |zn | = rn . This clearly means that if is chosen small enough, then
g may be chosen so as to have a representative in H() for all f H() and, furthermore,
that this representative will depend linearly and continuously on f . In other words, for
small enough , there is a continuous linear map : H() H() such that f (f )h
is a polynomial of degree less than k in n for all f H().
Now let N be the set of elements of M which are polynomials in zn of degree less than
k. Clearly, N is a module over the ring n1 H0 of germs of holomorphic functions in the

152

J. L. TAYLOR

variables z1 , . . . zn1 and, as such, it may be regarded as a submodule of n1 H0k . Now if f

is in the closure of the set of functions in H(U ) with germs in M at 0, then it follows from
the continuity of that f (f )h is in the closure of the set of functions in H() which
are polynomials of degree less than or equal to k and have germs beonging to N at 0. By
the induction hypothesis, this implies that f (f )h has germ at 0 belonging to N . Since
N M and h M , this shows that f M and completes the induction step in the case
where k = 1.
To complete the induction on n for general k, we use induction on k. We assume that
part (a) is true with k replaced by k 1 and prove that it is also true for k. Thus, let M
be a submodule of H0k and define submodules M0 H0k1 and Mn H0 by
M0 = {f = (f1 , . . . , fk ) M : fk = 0},

Mk = {fk : (f1 , . . . , fk ) M }

Then Mk is isomorphic to M/M0 under the projection map which sends a k-tuple of
functions to its kth element. We choose a polydisc centered at 0 and a finite set of
elements of Hk () whose germs at 0 form a set of generators for M . This set of elements
then determines a morphism of analytic sheaves over
: Hm Hk
with the image of 0 equal to M . Over we define : Hm H to be followed by the
projection of Hk on its last component. Then the image of 0 is Mk . Now, by shrinking
if necessary, we may assume that and are defined in an open polydisc containing
the closure of and that has compact closure. Under our induction assumption, we
know part (a) holds for k = 1. Hence, Part (b) holds for k = 1. This means that
: Hm () H() has closed image. By the open mapping theorem for Frechet spaces,
this implies that is an open mapping onto its image. Now suppose that {fj } is a
sequence in Hk (U ) with (fj )|0 M for every j and suppose that this sequence converges
to f Hk (U ). If gj and g are the last components of fj and f , then gj g and,
since is an open map, there exists a convergent sequence hj h in Hm () such that
gj = (hj ) (Problem 16.1). Then fj (hj ) is a convergent sequence in Hk1 () consisting
of elements whose germs at 0 belong to M0 . Thus, by our induction assumption on k, its
limit f (h) also has germ at 0 belonging to M0 . However, (h)0 M and M0 M .
Hence, f0 M . This completes the induction on k and also the induction on n. Hence,
part(a) and part(b) are both proved in general.
16.3 Theorem. Let X be a holomorphic variety and S a coherent subsheaf of Hk for
some k. Then S(X) is a closed subspace of the Frechet space Hk (X).
Proof. We first note that part(a) of Theorem 16.2 extends to varieties. That is, if fj f
in X Hk (X) and if x X, then Corollary 15.18 and the open mapping theorem imply that
there is a neighborhood W of x which may be identified with a closed subvariety of an open
polydisc Cn and there is a convergent sequence gj Hk () such that gj |W = fj |W
(Problem 16.1). Thus, if the functions fj all belong to some submodule M X Hx , then
the gj will belong to the inverse image N n Hxk of this submodule under the restriction
map n Hxk X Hxk . It follows from Theorem 16.2 that if g is the limit of the sequence

NOTES ON SEVERAL COMPLEX VARIABLES

153

{gj }, then g has germ at x belonging to N and, hence, its restriction f to W has germ at
x belonging to M . Thus, part(a) of Theorem 16.2 holds with U replaced by an arbitrary
holomorphic variety X.
Now Theorem 16.3 is an immediate consequence, since a global section of S is just a
global section of Hk which has germ at x belonging to Sx for each x X.
A Frechet sheaf on a space X is just a sheaf of Frechet spaces on X. Of course, the
restriction maps are required to be morphisms in the category of Frechet spaces that is,
continuous linear maps.
16.4 Theorem. If S is a coherent analytic sheaf on an analytic space X, then there is a
unique structure of a Frechet sheaf on S with the property that if U is any open set and
Hk S is a surjective morphism of analytic sheaves from a free finite rank H module to
S defined over U , then Hk (U ) S(U ) is continuous. Furthermore, S is a Montel sheaf
with this structure.
Proof. By Corollary 15.17, we may choose a neighbohood base U for the topology of X
consisting of sets U which have compact closure and have the property that coherent
sheaves defined in a neighborhood of U are acyclic on U . Furthermore, we may choose U
so that for each U U , the sheaf S is the cokernel of a morphism between free finite rank
sheaves on a neighborhood of U .
Fix U U . Then there is an exact sequence of analytic sheaves

Hm Hk S 0
defined in a neighborhood of U . The fact that every coherent sheaf defined in a neighborhood of U is acyclic on U implies that the induced sequence on sections over U :

Hm (U ) Hk (U ) S(U ) 0
is also exact. Furthermore, by Theorem 16.3, the image of : Hm (U ) Hk (U ) is closed.
Hence, S(U ), as a separated quotient of Hk (U ), inherits a Frechet space topology, in fact,
a Montel space topology. Now suppose that V U is another set in our basis U and
suppose for this set we have an exact sequence

Hp Hq S 0
defined in a neighborhood of V . Then we may construct the following commutative diagram
with exact rows:

Hm (U ) Hk (U ) S(U ) 0

r
y
V,U y
y

Hp (V ) Hq (V ) S(V ) 0
Where rV,U is the restriction map and and are constructed by using the familiar
lifting argument (the projectivity of free modules). The maps and are module homomorphisms and, hence, are given by matrices of holomorphic functions on V through

154

J. L. TAYLOR

vector-matrix multiplication. This clearly implies that they are continuous. If S(U ) has
the topology it inherits from being a quotient of Hk (U ), then Hk (U ) S(U ) is an open
map. Also, the map Hk (V ) S(V ) is continuous if S(V ) is given the quotient topology.
These facts combine to force the map rV,U to be continuous. We draw two conclusions
from this:
(1) The topology on S(U ) is independent of the way in which it is expressed as a
quotient of a free finite rank H(U )-module; and
(2) if V U are two sets in our basis, then the restriction map rV,U : S(U ) S(V )
is continuous.
It is now clear how to define the topology on S(U ) for a general open set U . We cover
U
Q by a countable collection {Wi } of sets from our basis. Then f {f |Wi } : S(U )
i S(Wi ) is an
Qinjective continuous linear map of S(U ) onto a closed subspace
Q of the
Montel space i S(Wi ). The image is closed because it is the subspace of i S(Wi )
consisting of {gi } such that gi = gj on Wi Wj for all i, j. Since a closed subspace of a
Montel space is Montel, this serves to put a Frechet space structure on S(U ) under which
it is actually a Montel space. Note that, by construction, this topology has the property
that for each x U there is a basic neighborhood W U containing x such that the
restriction map S(U ) S(W ) is continuous. Now suppose that V U is another open
set and S(V ) is given a Frechet space topology with this same property. Then we claim
that the restriction map rV,U : S(U ) S(V ) has closed graph and, hence, is continuous.
To see this, let {(fn , rV,U (fn ))} be a sequence in the graph which converges to the point
(f, g). Then fn f and rV,U (fn ) g. Now for each point x of V we can choose a basic
neighborhood W of x such that the restrictions S(U ) S(W ) and S(V ) S(W ) are
both continuous. This clearly implies that rV,U (f )|W = g|W . But since this is true for a
neighborhood of each point of V , we conclude that rV,U (f ) = g and, hence, that the graph
of rV,U is closed as required. We draw two conclusions from this:
(1) The Frechet space topology on S(U ) is uniquely defined by the property that for
each basic open set W U the restriction map S(U ) S(W ) is continuous; and
(2) if V U are two open sets, then the restriction map rV,U : S(U ) S(V ) is
continuous.
Thus, we have proved the existence of a Frechet sheaf structure on S under which it is
a Montel sheaf. Furthermore, it is clear from the construction that a morphism T S
is continuous if and only if it is continuous locally, that is, if and only if T (W ) S(W )
is continuous for a neighborhood W of each point of X. This, and the construction of
the topology on basic open sets shows that the topology has the property that if Hk S
is a surjective morphism over an open set U , then Hk (U ) S(U ) is continuous. The
uniqueness is obvious from the construction.
In view of the above theorem, we may, henceforth, assume that every coherent analytic
sheaf comes equipped with a structure of a Montel sheaf. A morphism : S T between
two Frechet sheaves is said to be continuous if : S(U ) T (U ) is continuous for each
open set U . When is a morphism of coherent analytic sheaves continuous? Always!
16.5 Theorem. A morphism : S T of sheaves of H-modules between two coherent
analytic sheaves is automatically continuous.

NOTES ON SEVERAL COMPLEX VARIABLES

155

Proof. As was pointed out in the proof of the previous theorem, a morphism between
coherent analytic sheaves is continuous if and only if it is continuous locally. If x is a
point of X then we may choose a basic neighborhood W for x of the kind used in the
proof of the previous theorem. Then there are surjective morphisms : Hk S and
: Hm T defined over U which are also surjective on sections over U . Then the usual
lifting argument gives us the in the following commutative diagram

Hk (W ) S(W ) 0

y
y

Hm (W ) T (W ) 0
It follows that is continuous because and are continuous and is open. Thus, is
locally continuous and, hence, continuous.
We are now in a position to prove the main theorem in the subject Cartans Theorem
B. The first step is an approximation theorem:
16.5 Theorem. If X is a Stein space, S a coherent analytic sheaf on X and W X is an
Oka-Weil subdomain, then the space of restrictions to W of global sections of S is dense
in S(W ).
Proof. We first prove that this is true in the case where S is the structure sheaf H. Since
W is an Oka-Weil subdomain, there is a holomorphic map : X Cn such that maps
W biholomorphically onto a closed subvariety of the unit polydisc (0, 1) centered at 0.
If f is a holomorphic function on W and K is a compact subset of W then it follows from
Corollary 15.18 that f has the form g in a neighborhood of K, where g is a holomorphic
function on an open polydisc with closure contained in (0, 1). If {hj } is a sequence of
polynomials converging to g in the topology of uniform convergence on compact subsets
of , then fj = hj defines a sequence of holomorphic functions on X which converge
uniformly to f on K.
In order to prove the theorem for a general coherent analytic sheaf S, we choose a
sequence {Wn } of Oka-Weil subdomains with W = W0 and W n Wn+1 for each n.
We claim that the image of S(Wn ) under restriction is dense in S(Wm ) if m < n. In
fact, we can find a surjective morphism Hk S in a neighborhood of W n . Then both
Hk (Wn ) S(Wn ) and Hk (Wm ) S(Wm ) are surjective. Thus, an element f of S(Wm )
can be lifted to an element g of Hk (Wm ) and this can be expressed as the limit of a
sequence of restrictions of elements hj Hk (Wm ) by the result of the above paragraph.
The image of the sequence {hj } in S(Wn ) will then have the property that its restriction
to S(Wm ) converges to f .
To finish the proof, we choose a translation invariant metric n defining the topology
n
P
of S(Wn ) for each n. Since the metric n may be replaced by the metric
i without
i=0

changing the topology it defines, we may assume without loss of generality that the sequence of metrics is increasing in the sense that m (f ) n (f ) if m < n and f S(Wm ).
Then if > 0 and f S(W ) = S(W0 ), we choose g1 S(W1 ) such that 0 (f g1 ) < /2.

156

J. L. TAYLOR

We then inductively choose a sequence gn S(Wn ) with n1 (gn1 gn ) < 2n . Clearly,

for each m, the sequence {(gn )|Wm : n > m} converges in the metric m to an element
hm S(Wm ). Furthermore, hm = (hn )|Wm for m < n and 0 (f h0 ) < . Hence, the hn
define a global section h S(X) such that 0 (f h) < . This completes the proof.
16.6 Cartans Theorem A. If X is a Stein space and S is a coherent analytic sheaf on
X, then S(X) generates Sx at every point x X.
Proof. Let Mx be the submodule of Sx generated by the global sections of S. Let W be
a neighborhood of x which is an Oka-Weil subdomain and which has the property that
S(W ) generates Sx . We find a surjective morphism : Hk S over W with the property
that : Hk (W ) S(W ) is also surjective. Then the space of sections f Hk (W ) such
that (f )x Mx is closed by Theorem 16.2. It follows from the open mapping theorem
that the set of g S(W ) such that gx Mx is also closed. However, this set contains the
set of restrictions to W of all elements of S(X) which by Theorem 16.5 is dense in S(W ).
It follows that Mx = Sx and the proof is complete.
16.7 Cartans Theorem B. If X is a Stein space and S is a coherent analytic sheaf on
X, then S is acyclic on X.
Proof. With the machinery we have built up, the proof is just like the proof of Theorem
11.5. We use Corollary 15.16 to express X as the union of a sequence of Oka-Weil subdomains {Wn } such that W n Wn+1 and S is acyclic on each Wn . Then, since the Wn s
form a nested sequence, {Wn } is a Leray cover of X. We also have that the space of global
sections of S is dense in S(Wn ) for each n. That S is acyclic on X now follows from a
principle that is quite general and which we state in the next lemma.
16.8 Lemma. Suppose S is a sheaf on a topological space X. If X is the union of an
increasing sequence {Wi } of open subsets such that S is acyclic on each Wi , then
(a) H p (X, S) = 0 for p > 1;
(b) if, in addition, S is a Frechet sheaf and S(X) is dense in S(Wi ) for each i then
H 1 (X, S) = 0.
Proof. Let
0

0 S F 0 F 1 F p
be a flabby resolution of S. Suppose p > 1 and f F p (X) with f = 0. Then we will
prove by induction that there is a sequence {gn } with gn F p1 (Wn ), gn = gn1 on
Wn1 and gn = fn on Wn . Clearly, if we can show this then part(a) of the Lemma will
be established. since such a sequence determines a global section g F p1 (X) such that
g = gn on Wn and, consequently, g = f on all of X.
Suppose we have managed to construct the sequence {gn } for all n m. Because S
is acyclic on each Wn , there exists a section gm+1 F p1 (Wm+1 ) such that gm+1 = f .
However, gm+1 gm may not be zero on Wm . Whatever it is, it is in the kernel of and,
hence, there exists hm+1 F p2 (Wm ) such that hm+1 = gm+1 gm on Wm . Since F p2
is flabby we may assume that hm+1 is actually a section defined on all of X. We then set

NOTES ON SEVERAL COMPLEX VARIABLES

157

gm+1 = gm+1 hm+1 on Wm+1 . Clearly, this serves to extend our sequence {Wn } to
n = m + 1 and completes the induction. This completes the proof of part(a). Note that
this proof does not work when p = 1 since, in this case, there is no F p2 .
We now proceed with the proof of part(b). By Lerays Theorem we may compute
1

H (X, S) using Cech

cohomology for the Leray cover {Wn }. Our argument will involve an
approximation argument Q
for 0-cochains. The space of 0-cochains for S and the cover {Wn }
on an open set U is just n S(Wn U ) and, as a countable product of Frechet spaces, is
itself a Frechet space. For each j, we let j be a translation invariant metric on the Frechet

space of 0-Cech
cochains on Wj . As in the proof of Theorem 16.5, we may assume that the
sequence {j } is increasing in the sense that j (g|Wj ) k (g) if j < k and g is a 0-cochain
on Wk . Suppose f is a 1-cocycle for the cover {Wn }. We inductively construct a sequence
{gj } where gj is a 0-cochain on Wj such that gj = f on Wj and j (gj gj+1 |Wj ) < 2j
for each j. Suppose such a sequence {gj } has been constructed for all indices j < k. We
use the fact that H 1 (Wk1 , S) = 0 to find a 0-cochain t on Wk such that t = f on Wk .
Then,
(t gk1 ) = 0
in Wk1 . This means that t gk1 is the 0-cochain on Wk1 determined by a section r
S(Wk1 ). Using the density hypotheses, we may choose a global section that approximates
this section as closely as we desire. Thus, there is a global 0-cochain s such that s = 0
and k1 (t gk1 s) < 2k+1 . Then gk = t s has the properties that gk = f on Wk
and k1 (gk gk1 ) < 2k+1 . Thus, by induction, we may construct the sequence {gj } as
claimed. Now on a given Wk consider the sequence {gj )|Wk }
j=k . This is a Cauchy sequence
j
in the metric k since k (gj+1 gj ) < j (gj+1 gj ) < 2 for j k. Furthermore, the
terms of this sequence differ from the first term by cocycles (since the difference is killed
by . Thus, the sequence may be regarded as a fixed cochain plus a uniformly convergent
sequence of cocycles. It follows that this sequence actually converges in the topology of
0-cochains on Wk and the limit hk satisfies hk = f |Wk . Furthermore, hk+1 = hk on Wk
and, hence, the hk determine a 0-cochain h on X. Clearly, h = f . Thus, every 1-cochain
is a coboundary and the proof is complete.
Cartans Theorem B has a host of corollaries. We list some of these in the next few
pages. The first five follow immediately using sheaf theory techniques which are, by now,
familiar to us and so we leave their proofs as excercises:
Corollary 16.9. If X is a Stein space, then every surjective morphism S T of coherent
analytic sheaves induces a surjective morphism S(X) T (X) on global sections.
Corollary 16.10. If Y is a closed subvariety of a Stein space X then every holomorphic
function on Y is the restriction of a holomorphic function on X.
Corollary 16.11. If X is a Stein space, an if {fi } is a finite set of holomorphic functions on
X which does not vanish simultaneously
at any point of X then there is a set of holomorphic
P
functions {gi } on X such that
gi fi = 1. In other words, each finitely generated ideal of
H(X) is contained in a maximal ideal of the form Mx = {f H(X) : f (x) = 0} for some
x X.

158

J. L. TAYLOR

If K is compact, then by H(K) we mean the algebra of functions holomorphic in a

neighborhood of K, in other words, the sections of the sheaf H on K.
16.12 Corollary. Let K be a compact holomorphically convex subset of a Stein space
X. Then every coherent analytic sheaf defined in a neighborhood of K is acyclic on K and
is generated as a sheaf of H modules by a finite set of sections over K;
16.13 Corollary. Let K be a compact holomorphically convex subset of a Stein space
X. Then every maximal ideal of H(K) is of the form Mx = {f H(X) : f (x) = 0} for
some x K.
Let H denote the sheaf of invertible holomorphic functions under multiplication. In
view of the short exact sequence
f exp(2if )

0 Z H H 0
we conclude that:
16.14 Corollary. If X is a Stein space then
(a) H 1 (X, Z) ' H (X)/ exp(H(X));
(b) H 2 (X, Z) ' H 1 (X, H );
Using Corollary 16.14 as in problem 9.9, one can now show that
16.15 Corollary. If X is a Stein space then the group of isomorphism classes of holomorphic line bundles is isomorphic to the group of isomorphism classes of continuous line
bundles which is isomorphic to H 2 (X, Z).
Now suppose X is a connected complex manifold. The rings H(U ) for U a connected
open subset of X are all integral domains. Let M denote the sheaf on X generated by the
presheaf which assigns to each connected open set U the quotient field of H(U ). Then M
is called the sheaf of meromorphic functions on U . We let M be the sheaf of non-zero
elements of M under multiplication. Clearly, H M . The quotient sheaf is denoted D
and called the sheaf of divisors on X.
The long exact sequence of sections associated to the short exact sequence
0 H M D 0
combined with Corollary 16.14 yields an exact sequence
(X, M ) (X, D) H 1 (X, H ) H 1 (X, M )
An element of (X, D) is called a divisor and its image in H 1 (X, H ) = H 2 (X, Z) is
called its Chern class. Now by Corollary 16.15 an element of H 1 (X, H ) corresponds to a
holomorphic line bundle L on X and L corresponds to the zero element of H 1 (X, H ) if
and only if it is the trivial line bundle that is, if and only if it has a section that is nowhere
vanishing. However, if L just has a section that is not identically zero, then it is easy to
see that the class in H 1 (X, H ) corresponding to L is sent to 0 by the map H 1 (X, H )
H 1 (X, M ) (problem 16.7). However, the holomorphic sections of a holommorphic line
bundle form a coherent analytic sheaf. By Cartans Theorem A, every coherent sheaf on
a Stein space is generated by its global sections. We conclude that every holomorphic line
bundle on X has a section which is not identically zero. Hence, H 1 (X, H ) H 1 (X, M )
is the zero map, and we conclude from the above exact sequence and Corollary 16.14 that:

NOTES ON SEVERAL COMPLEX VARIABLES

159

16.16 Corollary. If X is a Stein manifold, then there is an exact sequence

0 (X, H ) (X, M ) (X, D) H 2 (X, Z) 0
Thus, every element of H 2 (X, Z) is the Chern class of some divisor and a divisor has Chern
class zero if and only if it is the divisor of a global meromorphic function.
We now turn to the final topic of this section cohomology of coherent sheaves on
compact holomorphic varieties. We cannot expect a compact variety to be a Stein space,
since the only global holomorphic functions on such a variety are constants and so the
conclusion of Theorem A does not hold. However, we will prove that every coherent
analytic sheaf on a compact holomorphic variety has finite dimensional cohomology. The
proof depends on a theorem of Schwartz on compact perturbations of surjective bounded
linear maps between Frechet spaces.
A continuous linear map : X Y between two topological vector spaces is said to be
compact if there exists a neighborhood U of 0 in X such that (U ) has compact closure in
Y.
16.17 Theorem. If S is a coherent analytic sheaf on a Stein space X and U is an open
set with compact closure in X, then the restriction map S(X) S(U ) is a compact map.
Proof. We first note that this is true if S is the structure sheaf H. In fact, {f H(X) :
supU |f (x)| < 1} is a neighborhood of zero in H(X) and its image in H(U ) is bounded
and, hence, has compact closure since H(U ) is a Montel space.
Now suppose S is any coherent sheaf on X. We choose an Oka-Weil subdomain W such
that U W and such that for some k there is a surjective morphism Hk S defined over
W . The corresponding map on global sections is also surjective by Cartans Theorem B.
Thus, we have the following commutative diagram:
Hk (W ) S(W ) 0

y
y
Hk (U ) S(U ) 0
There is an neighborhood of zero in Hk (W ) whose image in Hk (U ) has compact closure
by the result of the previous paragraph. The image of this neighborhood in S(W ) is a
neighborhood of zero in S(W ) by the open mapping theorem and its image in S(U ) will
have compact closure by the commutativity of the diagram. Thus, S(W ) S(U ) is a
compact map. However, S(X) S(U ) is the composition of this map with the restriction
map S(X) S(W ). Since, the composition of a continuous linear map with a compact
map is clearly compact, the proof is complete.
We now prove the Cartan- Serre theorem assuming Schwartzs Theorem. We will end
the Chapter with a proof of Schwartzs Theorem.
16.18 Cartan-Serre Theorem. If X is a compact holomorphic variety and S is a coherent analytic sheaf, then H p (X, S) is finite dimensional for all p.
Proof. We choose a finite open cover W = {Wi } of X consisting of sets which are Stein
spaces. Finite intersections of sets in this cover are also Stein spaces (Problem 15.4). We

160

J. L. TAYLOR

then choose another such cover U = {Uj } which is a refinement of the first cover and,
in fact, has the property that for each j there is an integer (j) such that U j W(j) .
Then for each multi-index = (j0 , . . . , jp ) the multi-index () = ((j0 ), . . . , (jp )) has the
property that U W() so that the restriction map S(W() ) S(U ) is a compact
map. It follows that the refinement morphism : C p (W) C p (U) from the space of

Cech
p-cochains for W to the space of Cech
p-cochains for U is a compact map between
Frechet spaces. Since both covers W and U are Leray covers for S the map induces an

isomorphism of cohomology. Hence, if Z p (W) and Z p (U) are the spaces of Cech
p-cocycles
for W and U, then the map
f g p1 (f ) + (g) : C p1 (W) Z p (W) Z p (W)
is surjective. Since is compact, it follows from Schwartzs theorem (Theorem 16.20)
that p1 has closed image and finite dimensional cokernel. Hence, H p (X, S) is finite
dimensional.
It remains to prove Schwartzs theorem. We first prove a dual version of Schwartzs
theorem from which the theorem itself will follow:
16.19 Theorem. Let X and Y be locally convex topological vector spaces and let A :
X Y be a continuous linear map which has closed image and is a topological isomorphism
onto its image, Let C : X Y be a compact continuous linear map. Then B = A + C
has finite dimensional kernel K, closed image I and the induced map B : X/K I is a
topological isomorphism.
Proof. A and C agree on the kernel K of B. Thus, A|K is a topological isomorphism of
K onto a subspace of Y , but it is also a compact map. This implies that the image of K
in Y under A has a neighborhood of zero with compact closure. However, a topological
vector space which is locally compact is necessarily finite dimensional. It follows that K
is finite dimensional.
Now there is a closed subspace L X which is complementary to K in X. This follows
from the Hahn-Banach theorem. In fact, if {xi } is a basis for the vector space K, then the
Hahn-Banach theorem implies that we can find for each i a continuous linear functional
fi on X with fi (xj ) = ij . The intersection of the kernels of the fi will then be a closed
complement for K. If L is such a complement, then A|L and C|L are continuous linear maps
of L into Y , the first a topological isomorphism onto its image and the second a compact
map. Thus, we are back in our original situation except now we have that A|L + C|L is
injective. Thus, to complete the proof it is enough to prove the theorem in the case where
A + B is injective.
We need to prove that B = A+C has closed image and is a topological isomorphism onto
its image. Thus, let x be a net in X and suppose that B(x ) converges to y Y . Since
C is compact, there is a continuous seminorm on X such that U = {x X : (x) < 1}
is a neighborhood which C maps to a set with compact closure in Y . Suppose {(x )} is
bounded, say by M . Then C(x ) is contained in the set (M +1)C(U ), which has compact
closure in Y . Thus, C(x ) has a cluster point z Y and A(x ) = B(x ) C(x ) has
y z as a cluster point. Since the image of A is closed, we have that y z = Ax for a

NOTES ON SEVERAL COMPLEX VARIABLES

161

unique x X. Since A is a topological isomorphism onto its image, we conclude that the
original net {x } has x as a cluster point and Bx = y since B is continuous. Thus, in the
case where {(x )} is bounded, we have that {x } has a cluster point x and B(x) = y.
Now suppose that {(x )} is unbounded, then, after possibly modifying {x } to eliminate terms where (x ) = 0, we may consider the net {x0 } defined by
x0 =

x
(x )

This net has the property that B(x0 ) converges to zero and {(x0 )} is bounded, in fact
(x0 ) = 1 for every . Thus, we are back in the previous case except that y has been
replaced by zero. We conclude that {x0 } has a cluster point x and B(x) = 0. However,
this is only possible if x = 0, since we are assuming that B is injective, and zero cannot
be a cluster point of a net of elements x0 with (x0 ) = 1. Thus, the first case was the
only one possible and we conclude that every net {x } X for which B(x ) converges in
Y has a cluster point in X. This is enough to imply that the image of B is closed and the
inverse of B on that image is continuous. This completes the proof.
It turns out that the previous theorem is the dual of the one we want. In order to prove
Schwartzs theorem we must first define a topology on the dual of a Frechet space and
prove an important theorem of Mackey and Arens concerning duality. There are many
ways to topologize the dual of a locally convex topological vector space. The useful ways
are of the following type:
A saturated family of bounded subsets of a topological vector space X is a family which
is closed under subsets, multiplication by scalars, finite unions and closed convex balanced
hulls. Such a family is said to cover X if X is the union of the sets in the family.
16.20 Definition. If X is a locally convex topological vector space and is a saturated
family of bounded subsets of X which covers X, then X will denote the space of continuous
linear functionals on X with the topology of uniform convergence on sets in .
Clearly X is a locally convex topological vector space. A family of seminorms defining
the topology is the family of all seminorms of the form K where K is a set in and
K (f ) = sup{|f (x)| : x K}. The family of sets we want to use is the family c of all sets
with compact closure. This is not always a saturated family since it is not always true
that the closed convex, balanced hull of a compact set is compact. However, for Frechet
spaces we have:
16.21 Theorem. If X is a Frechet space, then the closed convex, balanced hull of every
compact subset of X is also compact.
Proof. If D is the closed unit disc in C and K X is compact, then D K is the image of
the compact set D K under the scalar multiplication map and is, hence, compact. The
closed convex hull of D K will be a closed convex balanced set containing K and, hence,
to prove the Theorem it suffices to prove that the closed convex hull of a compact set in a
Frechet space is compact.
Recall that a subset of a complete metric space is compact if and only if it is closed and
totally bounded. A subset S of a metric space is totally bounded if for each > 0 there is
a finite set F in S so that each point of S is within of some point of F .

162

J. L. TAYLOR

Choose a translation invariant metric defining the topology of X. If K is a compact

subset of the Frechet space X, then K is totally bounded. Hence, given > 0 there exists
a finite set of points {yi }ni=1 K such that each x K is within /2 of some yi . Then the
convex hull L of the set {yi } is the image of the map
(s1 , , sn )

si yi : S X

P
where S is the simplex {(s1 , , sn ) (R+ )n :
si = 1}. Thus, L is compact. Furthermore, every element of the convex hull of K is within /2 of a point of L. Since L is
compact, it is also totally bounded and we may find a finite set of points {xj } such that
every point of L is within /2 of some xj . It follows that every point of the convex hull
of K is within of some xj . Thus, we have proved that the convex hull of K is totally
bounded. The closure of a totally bounded set is clearly totally bounded as well. Thus, the
closed convex hull of a compact subset of a Frechet space is totally bounded and, hence,
compact.
The above theorem implies that the family c of sets with compact closure in a Frechet
space is a saturated family. The topology on X which it determines is the topology of
uniform convergence on compact subsets of X. The space X with this topology will be
denoted Xc .
Our final preliminary result before proving Schwartzs Theorem is the following theorem
of Mackey-Arens. Note that, for any saturated family covering X, each element x X
determines a continuous linear functional on X by f f (x) : X C. Thus, X embedds
in the second dual (X ) . In the case where the sets in are precompact, every continuous
linear functional on X has this form.
16.22 Mackey-Arens Theorem. If X is a locally convex topological vector space and
is a saturated family of precompact subsets of X which covers X, then every continuous
linear functional on X is determined by an element of X. Thus, X = (X ) .
Proof. If is a continuous linear functional on X then there is a neighborhood V of zero
in X such that |(f )| < 1 for all f V . We may assume that the neighborhood V has
the form
V = {f Xc : sup |f (x)| < 1}
xK

for some K . Without loss of generality we may assume K is compact convex and
bounded since is saturated and consists of precompact sets. We regard X as embedded
in (X ) and we give the later space the weak-* topology that is, the topology of pointwise
convergence as functions on X . This may also be described as the topology of uniform
convergence on the family of sets which are convex balanced hulls of finite sets. On the
image of X in (Xc ) the topology is weaker than the original topology on X and, hence,
K is also compact in this topology. If we knew the result of the theorem in the case of the
topology, then we could conclude that every continuous linear functional on (Xc ) with
the topology is given by an element of Xc . Let us assume this for the moment. If is
not an element of K, then it follows from the Hahn-Banach theorem (TVS 5(b)) that there
exists a continuous linear functional f on (Xc ) such that |f ()| > 1 and |f (x)| < r < 1

NOTES ON SEVERAL COMPLEX VARIABLES

163

for all x K. However, under our assumption about the topology, we must have that
f Xc and it then follows that f V . This is a contradiction, since |f ()| = |(f )| < 1
if f V . We conclude that does belong to K and, in particular, belongs to X.
It remains to prove the theorem in the case where is the family generated by finite
sets. In this case, let and V and K be as above. Since, = we have that K is the
convex balanced hull of a finite set {xi }ni=1 . Then the set of f X such that f (xi ) = 0
for i = 1, . . . , n is just k k 1 V and so it follows that vanishes on this set. It then follows
from elementary linear algebra that must be a linear combination of the xi and, hence,
must belong to X.
16.23 Theorem (Schwartz). Let X and Y be Frechet spaces and suppose that A : X
Y is a surjective continuous linear map and C : X Y is a compact continuous linear
map. Then B = A + C has closed image and finite dimensional cokernel.
Proof. We consider the duals Xc and Yc of X and Y in the topology of uniform convergence
on compact subsets. Then the continuous linear map A : X Y has a dual A : Yc Xc
defined by
A (f )(x) = f (A(x))
Since A is continuous, it maps compact sets to compact sets, from which it follows that
A is continuous. Similarly, C is continuous.
Claim 1. The linear map C : Yc Xc is compact.
Since C is compact, there is a zero neighborhood U in X such that C(U ) has compact
closure K in Y . Then the zero neighborhood VK = {f Yc : supxK |f (x)| < 1} has
the property that C (VK ) is a family of continuous functions which is uniformly bounded
by one in modulus on U . If L is any compact subset of X then C (VK ) is uniformly
bounded on L, since L kU for some k > 0, and C (VK ) is equicontinuous on L, since
|f (x)f (x0 )| < if xx0 U and f C (VK ). It follows from the Ascoli-Arzela Theorem
that C (VK ) has compact closure in the space of all functions on X in the topology of
uniform convergence on compact subsets of X. However, the space of continuous linear
functionals is clearly closed in this topology since X is a metric space. Thus, C (VK ) has
compact closure in Xc and C is a compact operator.
Claim 2. The linear map A : Yc Xc has closed image and is a topological isomorphism
onto its image.
Let {f } be a net in Yc such that A (f ) converges in Xc to g. Let K be a compact
subset of Y . The fact that A is an open map implies that K is the image under A of a
compact subset L of X (Problem 16.8). We have that A (f ) = f A converges to g
uniformly on L and this implies that f converges uniformly on K. Since this is true for
every compact set K Y , the net {f ) converges in the topology of Yc to an element f .
Clearly A (f ) = g. Hence, the image of A is closed and, on the image, the inverse map
is continuous. This establishes Claim 2.
We now have the hypotheses of Theorem 16.19 satisfied for the pair of operators A and
C . We conclude that B = A + C has finite dimensional kernel Z and closed image and
induces a toplogical isomorphism from Yc /Z to its image in Xc . Now the map Yc /Z Xc

164

J. L. TAYLOR

induced by B is the dual of the continuous linear map B 0 obtained by composing B with
the map Y Z defined by y {f f (y)} f Z, y Y . Thus, to finish the proof
we need to show that B 0 is surjective. In other words, the proof will be complete if we can
prove:
Claim 3. If B : X Y is a bounded linear map between Frechet spaces and B : Yc
Xc is injective, has closed image and is a topological isomorphism onto its image, then B
is surjective.
The map B induces a continuous linear map X/ker B Y which has as dual the
map B as a map from Yc to {f Xc : f (x) = 0 x ker B}. This map has closed
image and so if it is not surjective, then the Hahn-Banach Theorem implies that there is a
continuous linear functional F on X which vanishes identically on B (Yc ) and does not
vanish identically on {f X : f (x) = 0 x ker B}. However, by the Mackey-Arens
Theorem (Theorem 16.22), the functional F has to have the form F (f ) = f (x) for some
x X. In other words, there is an x X such that B (g)(x) = g(B(x)) for all f Yc
but g(x) 6= 0 for some f Xc which vanishes on ker B. This is impossible, since such an
x would necessarily be in ker B. Thus, B is surjective as a map from Yc to the dual of
X/ ker B. It follows that we may assume without loss of generality that B is injective and
B is surjective.
Thus, suppose that {xn } is a sequence in x such that {B(xn )} converges to y Y .
Then the set S = {B(xn )} {y} is compact. Hence, the set
U = {f Y : |f (y)| < 1

y S} {f Y : |f (B(xn ))| < 1

n}

is a neighborhood of zero in Yc and, hence, its image B (U ) is a neighborhood of zero in

in Xc . This implies that there is a compact set K X and a > 0 such that
{f X : |f (x)| <

x K} B (U ) = {f Xc : |f (xn )| < 1

x K}

or
{f X : |f (x)| < 1

x 1 K} B (U ) = {f Xc : |f (xn )| < 1

x K}

From the Hahn-Banach theorem (TVS 5(b)), it now follows that the sequence {xn } lies
in the compact convex balanced hull of 1 K and, hence, lies in a compact subset of X.
Then it has a cluster point x and, clearly, B(x) = y. Thus, B is surjective. This completes
the proof of claim 3 and the proof of the Theorem.

16. Problems
1. Prove that a bounded linear map : X Y between two Frechet spaces is an open
map if and only if every convergent sequence in Y lifts under to a convergent sequence
in X.
2. Prove Corollary 16.9.

NOTES ON SEVERAL COMPLEX VARIABLES

3.
4.
5.
6.
7.

165

Prove Corollary 16.10.

Prove Corollary 16.11.
Prove Corollary 16.12.
Prove Corollary 16.13.
Prove that if the line bundle corresponding to an element of H 1 (X, H ) has a non-zero
section then the element has image zero in H 1 (X, M ).
8. Prove that if X Y is a surjective continuous linear map between Frechet spaces, then
each compact subset of Y is the image of a compact subset of X, Hint: Use the fact
that in a complete metric space the compact sets are the closed totally bounded sets.

166

J. L. TAYLOR

17. The Borel-Weil-Bott Theorem

In this chapter we give an application in group representation theory of some of the
machinery we have developed in previous chapters. The application is a proof of the
Borel-Weil-Bott Theorem. This theorem characterizes the cohomology of equivariant holomorphic line bundles on the flag manifold in terms of the irreducible finite dimensional
representations of the corresponding semisimple Lie algebra. The proof we give here is due
to Milicic. It is not terribly difficult but it does require detailed knowledge of the structure
theory for complex semisimple Lie algebras and their finite dimensional representations.
We begin with a quick review of this material.
Let g be a complex Lie algebra. Associated to g is a unique connected, simply connected
complex Lie group G having g as its Lie algebra. A complex Lie group is a complex
manifold and, hence, we may ask that a homomorphism of it into another complex Lie
group be holomorphic rather than just continuous and, in particular, we may ask that
finite dimensional representations be holomorphic. Any finite dimensional complex linear
representation of g gives rise to a holomorphic representation of G through exponentiation.
Conversely, any holomorphic finite dimensional representation of G gives rise to a complex
linear finite dimensional representation of g through differentiation. The adjoint action of
g on itself, denoted by ad, is such a representation of g. That is, for , g, ad () = [, ].
The corresponding representation of the group G will be denoted by Ad. It is a holomorphic
representation of G as a group of Lie algebra automorphisms of g.
The Killing form for a Lie algebra g is the symmetric bilinear form
h, i = tr(ad ad )
A Lie algebra is said to be semisimple if it is a direct product of simple Lie algebras that
is, non-abelian Lie algebras with no non-trivial ideals. A Lie algebra is semisimple if and
only if its Killing form is non-singular.
17.1 Definition. If g is a semisimple Lie algebra, then
(a) a Borel subalgebra b of g is a maximal solvable subalgebra of g;
(b) a Cartan subalgebra h is a maximal abelian subalgebra of g consisting of elements
which act semisimply on g.
A semisimple Lie algebra g has a Borel subalgebra b and every Borel subalgebra b
contains a Cartan subalgebra h. Furthermore, if n = [b, b], then n is a nilpotent ideal of b
and, as vector spaces, b = h n.
If h is a Cartan subalgebra, then its elements act semisimply on g, which implies that
g decomposes as a direct sum of subspaces g where h and
g = { g : ad () = ()

h}.

Since h is maximal abelian, the space g0 is h itself. The non-zero elements h for
which g is non-empty are called roots. We denote the set of all roots by . Thus,
X
g=h
g .

NOTES ON SEVERAL COMPLEX VARIABLES

167

A simple calculation shows that

[g , g ] g+ .
One can also show that if and are roots then [g , g ] 6= 0 if and only if + is either zero
or is a root. It turns out that for each root the spaces g are one dimensional. It follows
that [g , g ] = g+ if + is a root and that [g , g ] is a one-dimensional subspace of
h. In fact, g , g and [g , g ] span a three dimensional Lie algebra isomorphic to sl2 (C).
Proofs of these facts can be found in any elementary treatment of Lie algebra theory.
The fact that nilpotent matrices have trace zero implies that
hg , g i = 0

if

6= .

Since, h = g0 , this implies that h is othogonal to each g for a root. This in turn implies
that the Killing form remains non-singular when restricted to h. Thus, the Killing form
induces an isomorphism between h and its dual h . That is, to each h there is a unique
h such that () = h, i. Then h, i = h , i defines a non-singular bilinear form
on h which we shall also call the Killing form. If we let h0 be the real subspace of h
spanned by the set of roots, then one can prove that h0 is a real form of h i. e. is a
real subspace with the property that h = h0 ih0 . Furthermore, the restriction of the
Killing form to h0 is positive definite. Again, for proofs we refer the reader to any standard
treatment of Lie algebra theory.
A positive root system for a Cartan h is a subset + such that: (i) it is closed
under addition in the sense that if , + , then + + provided + is a root;
and (ii) for every root , exactly one of , belongs to + . Such a system may
be constructed by choosing a real hyperplane through 0 in h which does not meet and
then letting + be the set of all roots on one side of this hyperplane. If + is a positive
root system, then so is its complement in and this is usually denoted . A system of
positive roots determines two Borel subalgebras b+ and b by
X
X
b+ = h
g , b = h
g ,
+

It is easy to see that if b is any Borel subalgebra of g and h is a Cartan contained in b

then b may be represented as the algebra b+ if the positive root system + is chosen to
be the set of roots whose root spaces are contained in b. One may also represent b as b
by choosing the positive root system to be the complement of this set.
A key result in the theory of semisimple Lie groups is the following theorem which we
will not prove:
17.2 Theorem. Under Ad the group G acts transitively on the set of all Borel subalgebras
and on the set of all Cartan subalgebras.
Let B be the normalizer in G of the Borel subalgebra b that is,
B = {g G : Adg (b) b}.
Then B is the subgroup of G corresponding to the subalgebra b under the Lie correspondence. A subgroup of this form is called a Borel subgroup of G. Note that B is the
stabilizer of b under the action of G on the set of Borel subalgebras. Thus, the set X
of Borel subalgebras of g is in one to one correspondence with the quotient space G/B.
Another key result from Lie theory which we will not prove is the following:

168

J. L. TAYLOR

17.3 Theorem. If B is a Borel subgroup of a complex semisimple Lie group G, then G/B
is a compact complex manifold.
Thus, the set X of Borel subalgebras of g may be given the structure of a compact
complex manifold through its identification with G/B. We call this the flag manifold
of G. Note also that G acts on X through left multiplication if X is represented as
the coset space G/B or, equivalently, through Ad if X is represented as the set of Borel
subalgebras of g. This action is, of course, transitive and is holomorphic in the sense that
g x gx : G X X is a holomorphic map.
We are interested in studying the finite dimensional representations of g. Let (, V ) be
such a representation. By Lie theory, exponentiates to a holomorphic representation of
the Lie group G on V which we also denote by . If h is a Cartan subalgebra of g, and
H is the connected Lie subgroup of G corresponding to h, then H is a connected complex
abelian Lie group; hence, it must be a complex torus that is, a product of copies of
the punctured plane C . The corresponding product of unit circles is a maximal compact
subgroup T of H. The complex vector space V may be given a T -invariant inner product
(by integrating any inner product over the compact group T ). Then (, V ) is a finite
dimensional unitary representation of T . It follows that V may be written as a direct sum
of subspaces on which the representation of T is irreducible and each of these subspaces
consists of eigenvectors for the operators (t) for t T . It follows that they are also
eigenvectors for the operators () for h. In other words,
17.4 Theorem. The restriction of the representation to h is completely reducible.
From this it follows that V is a direct sum of subspaces V where h and
V = {v V : ()v = ()v

h}

The elements h for which V 6= 0 are called the weights of the representation .
Since the representation is finite dimensional, there can be only finitely many of them.
How does the rest of the Lie algebra g act on V ? We get an idea by using the root space
decomposition. Thus if and g , then a calculation shows that ()V V+ .
Now, if the representation is irreducible, then for any weight the weight space V must
generate V . It follows that one must be able to obtain all the weights from a given one by
successively adding roots.
17.5 Definition. For a finite dimensional representation (, V ), a Cartan h, a positive
root system + and weights and , we say < if is a sum of positive roots. A
weight is called a highest weight if it is maximal relative to this relation. If a weight has
the property that, for each root , either + or is not a weight for (, V ), then
will be called an extremal weight.
If is a highest weight, then is an extremal weight since + fails to be a weight
for every positive root . We also have (g )V = 0 for every + since V+ = 0. It
follows from this and the commutation relations in g that the g-invariant subspace of V
generated by any non-zero vector v V is contained in the span of v and the spaces V
for < . This implies the following theorem:

NOTES ON SEVERAL COMPLEX VARIABLES

169

17.6 Theorem. For an irreducible finite dimensional representation (, V ) of g, a Cartan

subalgebra h and a system of positive roots + , there is a unique highest weight h .
Furthermore, V is one dimensional.
Note that this also implies that, for an irreducible finite dimensional representation
there is exactly one extremal weight corresponding to each positive root system.
As before, let h0 be the real subspace of h spanned by the roots. Equivalently, h0 is the
real subspace of h consisting of those weights for which h, i is real for every + .
Each + defines a hyperplane { h0 : h, i = 0} in h0 . The complement in h0
of the union of these hyperplanes has finitely many components and these are open sets
called Weyl chambers. That is, a Weyl chamber is a non-empty subset of h0 of the form
{ h0 : ()h, i > 0 + }, where is a function from + to {1, 1} (not every
such function defines a non-empty set). The positive Weyl chamber is { h0 : h, i >
0 + }.
For + , the operator s on h0 defined by
s () = 2

h, i

h, i

is reflection through the hyperplane determined by . The Weyl group W for g is the
group of transformations of h0 generated by the set of reflections s .
It is clear that the Weyl chambers are in one to one correspondence with the possible
choices of positive root systems. That is, for each Weyl chamber there is exactly one
positive root system for which it is the positive Weyl chamber. It is also easy to prove that
the Weyl group acts transitively on the set of Weyl chambers.
We define a norm in h0 by setting
|||| =

p
h, i.

It turns out, though we shall not prove it here, that each element of the Weyl group arises
from a transformation of h of the form Adg , where g G is in the normalizer of h. In
fact W is isomorphic to the normalizer of h in G mod the centralizer of h in G. It follows
from this and the G invariance of the Killing form, that the Weyl group acts as a group of
isometries relative to the above norm.
We set
1 X
=

2
+

This element plays a special role throughout the theory of semisimple Lie algebras.
17.7 Theorem. Let (, V ) be a finite dimensional irreducible representation of g, h g
a Cartan subalgebra, + a positive root system, h the set of weights for (, V ) and
the highest weight. Then
2h, i
(a)
Z , ;
h, i
(b) is closed under the action of the Weyl group;

170

J. L. TAYLOR

(c) if + then h, i 0 and h, i > 0;

(d) |||| |||| ;
(e) is extremal if and only if |||| = |||| and the Weyl group acts transitively
on the set of extremal weights;
(f) || + || < || + || for every distinct from .
Proof. For a given root , consider the subalgebra l spanned by g , g and [g , g ].
This is a copy of sl2 (C). If , the spaces Vn , for n an integer, span a subspace of V
which is a finite dimensional representation of l . However, we know that the weights for
a finite dimensional irreducible representation of sl2 (C) are symmetric about the origin.
Since every finite dimensional representation decomposes as a direct sum of irreducibles,
the weights for any finite dimensional representation of sl2 (C) are symmetric about the
origin. A calculation shows that an element of the Cartan subalgebra [g , g ] of our
copy of sl2 (C) acts on each space V as a fixed scalar multiple of h, i. It follows from
symmetry about the origin that h m, i = h, i for some integer m and, from this,
that
2h, i
= m.
h, i
This completes the proof of part (a). It also proves part (b) since it implies that the weight
m is s (), from which it follows that the set of weights is closed under the action
of the Weyl group. The first inequality of part (c) also follows from this argument applied
to the case where = , since then only non-negative integers n yield non-zero subspaces
Vn . Thus, the integer m above must be non-negative in this case, which implies that
h, i 0. The proof of the second inequality in part (c) is more complicated. It requires
a development of the properties of systems of simple roots. We leave this development and
the resulting proof of the fact that is in the positive chamber to the exercises (Problems
4 9).
If is any weight in , it has the form = where is a sum of positive roots. If
also satisfies h, i 0 for all + , then we have
h, i = h, i h, i h, i h, i
Thus, |||| |||| for every weight in the closure of the positive chamber. However, this
implies |||| |||| for all since every such may be brought into the closure of the
positive chamber by applying a Weyl group transformation and Weyl group tranformations
are isometries. This completes the proof of part (d);
Suppose that is not an extremal weight. Then there is a root such that +
and are both roots. Then
||||2 + ||||2 =

1
(|| + ||2 + || ||2 ) ||||2
2

by part (d) and this implies that |||| < ||||. Thus, only the extremal weights can have
norm equal to ||||. The proof of part (e) will be complete if we can show that the Weyl
group acts transitively on the set of extremal weights, since this will imply that they all

NOTES ON SEVERAL COMPLEX VARIABLES

171

have norm equal to that of . Using a Weyl group transformation, we can move any
extremal weight into the closure of the positive chamber and it will still be an extremal
weight. Thus, we will have completed the proof of (e) if we show that is the only extremal
root in the closure of the positive chamber. Thus, suppose that is extremal and lies in
the closure of the positive chamber. Then for each + we have h, i 0 and either
+ or is not a root. However, the sl2 (C) argument of the first paragraph shows
that it must be + that fails to be a root if h, i 0. Hence, is the highest root
and the proof of part(e) is complete.
If is any weight in then
h + , + i = h, i + 2h, i + h, i h, i + 2h, i + h, i
= h + , + i 2h , i < h + , + i
by part (c) and the fact that is a sum of positive roots. This proves part (f).
The elements of h that satisfy the condition in part(a) of the above theorem are called
integral weights.
17.8 Definition. We say that a weight h0 is dominant relative to a system of positive
roots if h, + i > 0 for every positive root . Thus, a weight is dominant if and only
if + belongs to the positive Weyl chamber.
We know that for every finite dimensional irreducible representation there is a unique
highest weight and it is easy to see from part(c) of Theorem 17.7 that highest weights
are dominant. In fact, using the theory of Verma modules one can prove that the finite
dimensional irreducible representations of a complex semisimple Lie algebra are classified
by their highest weights. We wont prove it here, but the theorem that does this is the
following:
17.9 Theorem. Let h g be a Cartan subalgebra and + h a system of positive
roots. Then each dominant integral weight is the highest weight for a unique (up to
isomorphism) finite dimensional irreducible representation of g.
We now introduce the objects of study in the Borel-Weil-Bott Theorem the Gequivariant holomorphic line bundles. Each of these is constructed from an integral weight
in h by an induction process which we shall describe below. In what follows, H will denote
the sheaf of holomorphic functions on either G or X. If it is not clear from the context
which is meant, we shall use the notation G H or X H. We also fix, for the remainder of the
discussion, a Cartan subalgebra h, a system of positive roots + and the Borel subalgebra
b = b spanned by h and the negative root spaces. The corresponding Borel subgroup of
G will be denoted B. We represent the flag manifold X as G/B.
Let (, W ) be a finite dimensional holomorphic representation of B equivalently, a
finite dimensional complex linear representation of b. The space G B W is constructed
from G W by identifying points which lie in the same orbit of the B-action described
by b (g w) gb1 (b)w. Projection on the first factor of G W , followed by the
projection : G G/B, induces a well defined holomorphic projection GB W X. The
inverse image of the typical point under this projection is gBB W , which has a well defined

172

J. L. TAYLOR

vector space structure under which it is isomorphic to W . In other words, G B W is a

holomorphic vector bundle over X with fiber W . Furthermore, G acts on G B W through
action on the left in the first factor. By a G-equivariant holomorphic vector bundle over X
we mean a complex manifold M with a holomorphic G action G M M , a holomorphic
projection : M X which commutes with the G actions on M and X and a complex
vector space structure on Mx = 1 (x) for each x X such that m gm : Mx Mgx is
linear for each g G and x X. In this sense G B W is a G-equivariant holomorphic
vector bundle over X.
17.10 Definition. Let (, W ) be a finite dimensional holomorphic representation of a
Borel subgroup B G, then we will call G B W the induced bundle over X determined
by (, W ) and denote it by I(). The sheaf of holomorphic sections of this vector bundle
will be denoted I().
Note that, from the construction of I(), we may characterize the sheaf I() in the
following way:
I()(U ) = {f H(1 (U ), W ) : f (gb1 ) = (b)f (g)

b B}

where : G G/B = X is the projection and H(1 (U ), W ) denotes the space of

holomorphic W valued functions on 1 (U ).
If (, V ) is a finite dimensional holomorphic representation of G then we may speak
of the trivial G equivariant holomorphic vector bundle over X with fiber V . This is the
complex manifold X V with projection X V X just the projection on the first factor
and G action given by g (x v) gx (g)v. Note that if we consider the restriction
of to B, then we have a representation of B and we may consider its induced bundle
I() = X B V . In fact the map
g v (g) (g)v : G V X V
induces a G-equivariant vector bundle isomorphism G B V X V . Thus, the trivial
G-equivariant bundle X V and the induced bundle I() are isomorphic. This is part of
what is to be shown in the next theorem.
17.11 Theorem. The induction functor (, W ) I() is an equivalence of categories
from the category of finite dimensional holomorphic representations of B to the category of
G-equivariant holomorphic vector bundles over X. Furthermore, the bundle I() is trivial
if and only if the representation is the restriction to B of a holomorphic representation
of G.
Proof. Let x0 = B/B be the point of X corresponding to B. Then B is the stabilizer
of x0 . If M X is any G-equivariant vector bundle over X, then the fiber Mx0 over
x0 is invariant under B and, hence, is a finite dimensional representation of B. The
resulting correspondence M Mx0 is a functor from G-equivariant vector bundles over X
to finite dimensional holomorphic representations of B. It is clear from the definitions that
(, W ) I() followed by M Mx0 is isomorphic to the identity. On the other hand,
if M is a G-equivariant holomorphic vector bundle, then we define a map G Mx0 M

NOTES ON SEVERAL COMPLEX VARIABLES

173

by g m gm. This clearly induces a map G B Mx0 M which is an isomorphism of

G-equivariant vector bundles. Thus, the composition M Mx0 followed by (, W ) I()
is also isomorphic to the identity. This completes the proof of the first statement of the
theorem. We have already proved the second statement in one direction. We leave the
other direction as as exercise (Problem 17.1).
Note that, for any finite dimensional holomorphic representation (, W ) of B, the sheaf
of holomorphic sections I() of I() is a coherent analytic sheaf. It follows from the CartanSerre Theorem (Theorem 16.18) that the cohomology H p (X, I()) is finite dimensional for
each p. Because of the G-equivariance, G acts in a holomorphic fashion on H 0 (X, I()).
It is also true, but less obvious, that G acts in a holomorphic fashion on each cohomology
space H p (X, I()) (Problem 17.2). Of course, it is not at all clear at this point which of
these cohomology groups, if any, are non-vanishing.
We now turn to the special case of induction of one dimensional representations of B.
Each integral h determines a character e on H by e (exp()) = exp(()). We
can extend the character e on H to a character (which we also call e ) on B by letting
it be the identity on N . Thus, e defines a holomorphic representation of B on the one
dimensional vector space C. Each one dimensional holomorphic representation of B arises
in this way. If we apply the induction functor, this yields a G-equivariant holomorphic
line bundle I(e ). We denote the sheaf of holomorphic sections of this line bundle by H .
Thus, if U is an open subset of X and : G G/B = X is the projection, then
H (U ) = {f H(1 (U )) : f (gb1 ) = e (b)f (g)}
This raises several questions:
(1) For which integral weights and for which integers p is H p (X, H ) non-zero?
(2) For which integral weights and for which integers p is H (X) an irreducible
representation of g?
(3) Does every irreducible finite dimensional representation of g arise this way?
The first step in answering these questions is to note a relationship between the sheaves
H and the Serre sheaves H(k) on projective space. This is done in Theorem 17.13 below.
First, we prove a technical lemma.
17.12 Lemma. If g is a semisimple Lie algebra, then g has a faithful finite dimensional
irreducible representation.
Proof. If g is simple (is non-abelian with no non-trivial ideals) then the adjoint representation of g on itself is faithful, since the kernel of g adg must be 0, and irreducible, since
an invariant subspace is an ideal.
Q
Now if g is not simple, we express it as a product
gi of simple Lie algebras gi and
find a faithful irreducible representation (i , Vi ) for each Q
of them. Then the tensor product
i i will be a faithful irreducible representation of g = i gi on i Vi .
17.13 Theorem. Let (, V ) be a faithful finite dimensional irreducible representation of
g with highest weight and let P (V ) be the projective space of the dual V of V , then
there is a closed holomorphic embedding : X P (V ) such that H = 1 H(1).

174

J. L. TAYLOR

Proof. Since is a highest weight, the weight space V is a complement to the span of the
subspaces (g )W for . Thus, if W is the space of all elements of V which vanish
on this span, then there is a non-zero element w W . Let ( , V ) denote the dual of
the representation (, V ). Clearly, ( , V ) is also a faithful, irreducible representation.
For the corresponding representation of G, we have
( (g)w)(v) = w((g 1 )v) w V
It follows that W is the lowest weight space of ( , V ) and has weight . Also
(g )W = 0 if . On the other hand, (g )W 6= 0 for each + . For
if g and g both annihilate W , then it follows from the commutation relations among
the root spaces and from the fact that W generates V , that g and g both annihilate all of V . This P
is impossible since is a faithful representation. Thus, the Borel
subalgebra b = h +
g is the stabilizer of W in g. That is,

b = { g : ()W W }.
It follows from this that the corresponding Borel subgroup B is the stabilizer of the one
dimensional subspace W under the action of G on V . Thus, under the action of G on
the complex projective space P (V ) of V induced by , the point p0 P (V ) determined
by W has B as stabilizer. Hence, the orbit of p0 under this action of G is compact, hence
closed, and is a copy of X = G/B. Since the action of G on P (V ) is holomorphic, it
follows from the complex implicit function theorem that the orbit is a compact, complex
submanifold of P (V ). Thus, we have a closed holomorphic embedding : X P (V ).
Now a typical section of the sheaf H(1) on an open subset of P (V ) is determined by
a homogeneous function of degree one on the corresponding open subset of V that is,
by a linear functional on V , i. e. a vector v V . If U is an open subset of X, then the
pullback of this section to U may be identified with the function f on 1 (U ) G defined
by g (g)w(v), where w is a non-zero element of W . This function satifies
f (gb1 ) = (gb1 )w(v) = e (b) (g)w(v) = e (b)f (g)
and, hence, determines a section of H . It follows that 1 H(1) = H .
The above result shows why we choose to depart from the standard convention in the
theory of group representations which assigns to a Borel b the system of positive roots for
which b = b+ . Instead, we assign to b the system of positive roots for which b = b . With
this convention, the of the above theorem is a highest weight rather than a lowest weight.
This is a better convention because then highest weights induce positive line bundles rather
than negative line bundles. This is the convention we will use throughout this section.
As an immediate corollary of the above theorem, we have:
17.14 Corollary. The flag manifold X is a non-singular projective variety.
The Casimir operator is a particular element of the center Z(g) of the enveloping algebra
U (g). It may be described as
X
X
X
X
(2 + [ , ])
( + ) =
i2 +
=
i2 +
+

NOTES ON SEVERAL COMPLEX VARIABLES

175

where the i form a self dual basis for h relative to the Killing form, so that hi , j i = ij .
The elements g are chosen so that h , i = 1. For each the element =
[ , ] belongs to h and has the property that ( ) = h, i for each h , where h, i
also stands for the form on h which is dual to the Killing form. Thus, we have
X
X
X
=
i2 +
+ 2
.
+

Let h be one half the sum of the positive roots.

17.15 Theorem. On a finite dimensional irreducible representation (, V ) of highest
weight , the Casimir operator acts as the scalar h, + 2i.
Proof. Since is irreducible, () must be a scalar times the identity operator. We
calculate the scalar by applying () to a highest weight vector. Thus, let v be a highest
weight vector in V , then ( )v = 0 for each + . It follows that
X
X
()v =
(i )2 +
( )
+

P
From the definition of the dual form on h , it follows that
(i )2 = h, i. We also have
( ) = h, i. Thus,
X
()v = h, i +
h, i = h, i + h, 2i = h, + 2i
+

This completes the proof.

17.16 Theorem. On the sheaf H , the Casimir acts as the scalar h, + 2i.
Proof. The action ` of U (g) on H is the infinitesimal form of the action ` of G on the
bundle I(e ). Thus, let f be a local section of H defined in a neighborhood any point
x X. Let B be Borel subgroup which is the stabilizer of x. Then we may represent X as
G/B and x as the identity coset of B. Choose a Cartan h b and a positive root system
such that the corresponding Borel subalgebra b is b . Then for b B
`( )f (b) =

d
d
`( )f (exp(t )b)|t=0 = e (b1 exp(t ))|t=0 `( )f (e) = 0
dt
dt

and
`()f (b) =

d
d
f (exp(t)b)|t=0 = e (b1 exp(t))|t=0 f (e)
dt
dt
1
= ()e (b )f (e) = ()f (b)

for h. That `()f (b) = h, + 2if (b) now follows as in the previous theorem from
the identity
X
X
X
.
+ 2
=
i2 +
+

176

J. L. TAYLOR

This says that the sections `()f and h, +2if of the line bundle corresponding to H()
agree at the point x. However, since x was arbitrary, this completes the proof.
Note that

h, + 2i = h + , + i h, i = || + ||2 ||||2

If is a weight of the form = w for some Weyl group element w W , then

|| + ||2 ||||2 = ||w||2 ||||2 = 0
since the Weyl group is a group of isometries of h . Thus, the Casimir acts as the zero
operator on H for all {w : w W }.
We can now begin the proof of the Borel-Weil-Bott theorem. The proof described here
is due to Milicic.
17.17 Lemma. Let (, V ) be a finite dimensional irreducible representation of g with
highest weight . Let w be an element of W and set = w . Then, for all p,
(a) H p (X, H ) is a trivial g module; and
(b) as g modules, H p (X, Hw+ ) ' H p (X, H ) V.
Proof. By the remarks preceding this theorem, we know that the Casimir acts on H as
the zero operator. However, the only finite dimensional representations with this property
are the trivial representations (this follows from Theorem 17.15, which implies that any
highest weight of such a representation is the zero weight). This proves part(a).
We consider the representation corresponding to (, V ) of G on V and denote by its
restriction to B. Then the induced bundle I() is trivial by problem 17.1 and, in fact,
is just the trivial G-equivariant bundle X V . Thus, its sheaf of sections I() is just
H V . On the other hand, we may twist by tensoring it with the one dimensional
representation determined by the character e to obtain a representation of B. Then
the corresponding G-equivariant bundle is I( ) = I(e ) I(). The corresponding sheaf
of sections is I( ) = H V . Here the g action is the tensor product of the natural action
on H with the action on V given by the representation . It follows that, as g-modules,
H p (X, I( )) = H p (X, H ) V.
We now have that, as a g-module, H p (X, I( )) is the tensor product of a trivial g-module,
H p (X, H ) and the module (, V ). In particular, this implies that the Casimir acts on
H p (X, I( )) as the scalar h, + 2i by Theorem 17.15.
We next construct a filtration {V p } of V by B-submodules. We set V 0 = V . Now the
highest weight space of V is V and this space is one dimensional. The sum of the weight
spaces V for 6= is a B-submodule V 1 of codimension one in V . It also must have
a highest weight and a highest weight space V . The sum of the weight spaces in V 1
other than V together with any codimension one subspace of V yields a B-submodule V 2
of codimension one in V 1 . Proceeding in this way, one constructs a decreasing filtration
{V p } of V by B-submodules such that V p /V p+1 is a one dimensional B-module for each
p. Since, on a one dimensional B-module, B necessarily acts by a character e for some
p
h , we have a sequence { p } of elements of h such that B acts as e on V p /V p+1 .

NOTES ON SEVERAL COMPLEX VARIABLES

177

The elements p are necessarily weights which occur in the representation V and, hence,
are weights dominated by . When we tensor by the one dimensional representation of
B with character e , we still have the same filtration but each weight p is changed to
p = p + .
It follows from Theorem 17.11 that the sheaf I( ) is filtered by a sequence of subsheaves
{I( p )}, where p is restricted to V p and the subquotient I( p )/I( p+1 ) is isomorphic
p
p
p
to Hp . By Theorem
Q 17.16, pthe pCasimir acts on H as the scalar h , + 2i. From
this it follows that p (h , +2i) acts as zero on I( ). This and Problem 17.3 then
imply that I( ) decomposes as a direct sum of subsheaves which are the - eigenspaces
for the eigenvalues hp , p + 2i).
Note that
p + = p + + = p + w = w(w1 p + )
so that each p + is of the form w( +), where is a weight for the representation (, V ).
It follows from Theorem 17.7(f) that either ||p + || < || + || or p + = w( + ).
In the latter case, p + = w + w + = w + + , so that p = w + .
Since h, + 2i = || + ||2 ||||2 for any , this implies that the only p for which
hp , p + 2i = h, + 2i is p = w + . From this it follows that the summand of I( )
on which h, + 2i vanishes must be a copy of Hw+ since this sheaf appears with
multiplicity one as a subquotient of I( ). Then
H p (X, I( )) = H p (X, Hw+ ) H p (X, J )
Where J is a summand of I( ) on which h, + 2i is non-vanishing. However,
we proved above that the Casimir acts as the scalar h, + 2i on the left side of this
equality. It acts as this same scalar on the first term on the right side but it acts on the
second term as an operator with eigenvalues all distinct from h, + 2i. It follows that
H p (X, J ) = 0 for all p and
H p (X, H ) V = H p (X, I( )) = H p (X, Hw+ )
for all p. This completes the proof.
The above is a very strong result. In particular, it implies the following:
17.18 Corollary. The set of integers p such that H p (X, H ) is non-vanishing is constant
as + varies over the integral weights in a given Weyl chamber.
Proof. If + belongs to the Weyl chamber which is the image of the positive chamber
under w W , then there is a dominant root such that + = w( + ). Thus,
= w + where = w . It then follows from the previous lemma that H p (X, H )
is non-vanishing if and only if H p (X, H ) is non-vanishing. Since, depends only on the
chamber determined by w, the proof is complete.
The Borel-Weil Theorem follows easily from the Lemma 17.17 and the above corollary
applied to the case where w is the identity, so that w + = .

178

J. L. TAYLOR

17.19 Borel-Weil Theorem. If is a dominant integral weight, then as g-modules

(a) H 0 (X, H ) is the irreducible finite dimensional representation with highest weight
; and
(b) H p (X, H ) = 0 for p 6= 0.
Proof. Let (, V ) be the finite dimensional irreducible of highest weight . By Lemma
17.17 with w = id, we have
H p (X, H ) = H p (X, H) V
This immediately implies part(a) since H 0 (X, H) is the space C of constants by virtue
of the fact that X is compact and connected. Also, the set of p for which H p (X, H ) is
non-vanishing is independent of as long as is dominant. Thus, part(b) will be proved
if we can show that there exists a dominant weight such that H p (X, H ) = 0 for all
p > 0. However, if V is a faithful finite dimensional irreducible g-module, then Theorem
17.13 implies that the pullback of H(1) under the embedding : X P (V ) is H , where
is the highest weight of V . It follows that the pullback of H(k) = k H(1) is Hk . Thus,
the cohomology of Hk on X is the cohomology of H(X) H(k) on P (V ), where H(X)
is the structure sheaf of the subvariety (X) extended by zero to P (V ). Since H(X) is
coherent, H(X) H(k) is acyclic for sufficiently large k by Theorems 13.18 and 14.12. It
follows that Hk is acyclic on X for sufficiently large k. This completes the proof.
The Borel-Weil-Bott Theorem, computes H p (X, H ) for all integral not just the
dominant ones. The next step in accomplishing this is to consider the case of singular
that is, for which + is not in a Weyl chamber but lies in a wall.
17.20 Theorem. If is an integral weight for which + lies in a wall, then
H p (X, H ) = 0
for all p.
Proof. For this argument we need to know something about the entire center Z(g) of the
enveloping algebra U (g), not just the Casimir element. We begin with a brief review of
the relevant information about Z(g).
Modulo the right ideal generated by b+ each element of Z(g) is equivalent to an element
in U (h). This correspondence defines a homomorphism : Z(g) U (h) called the HarishChandra homomorphism. Since h is abelian, the algebra U (h) is may be regarded as the
algebra of polynomial functions on the complex vector space h . The Harish-Chandra
homomorphism is an injective homomorphism with image equal to the set of polynomials
f such that f ( ) is invariant under the action of the Weyl group. It follows that every
complex homomorphism of Z(g) has the form = where h and (f ) = f () for
each polynomial f on h . Elements 1 and 2 determine the same complex homomorphism
(i. e. 1 = 2 ) if and only if 1 + and 2 + belong to the same Weyl group orbit.
Thus, the set of complex homomorphisms of Z(g) is in one to one correspondence with the
set of Weyl group orbits. If V is a g-module for which Z(g) acts as scalars, then it acts via
a complex homomorphism . In this case, V is said to have infinitesimal character .

NOTES ON SEVERAL COMPLEX VARIABLES

179

Now for the proof of the Theorem. If + belongs to a wall, then its Weyl group orbit
consists entirely of elements which belong to walls. If H p (X, H ) 6= 0 for some p, then
it will be a g-module with infinitesimal character by Theorem 17.16. However, there
is no finite dimensional representation with infinitesimal character since every finite
dimensional irreducible representation has infinitesimal character for dominant and
the Weyl group orbit of a dominant weight does not meet any walls.
We need one more lemma before proving the Borel-Weil-Bott Theorem:
17.21 Lemma. Let (, V ) be a finite dimensional irreducible representation of g and let
be its set of weights. Let + be a system of positive roots and let be an element
of + . If h0 satisfies h, i = 0 and h, i > 0 for + distinct from , then the
maximal value of || + || for is achieved at exactly two points, = and = s (),
where is the highest weight in .
Proof. Since ||||2 = h, i is a convex function of in h0 , the maximum clearly can occur
only at weights + for which is an extremal weight in . Given two extremal weights
and , we have |||| = |||| by Theorem 17.7 and, hence,
|| + ||2 || + ||2 = 2h, i
If is the highest weight in then is a sum of positive roots and so || + ||2 || +
||2 > 0 except in the case where involves only the root , i. e. has the form n for
some n. The only extremal weight of this form is s ().
Two Weyl chambers are adjacent if their closures have a wall in common. The reflection
through that wall will then interchange the two chambers. If is a positive root defining
the wall (that is, the wall is { h0 : h, i = 0}), then h, i and hs (), i will have
opposite signs, while h, i and hs (), i will have the same sign for other positive roots
. Thus, for in one of the two chambers there will be exactly one more negative number
in the set {h, i : + }. The distance from a Weyl chamber (or one of its elements)
to the positive chamber is the minimal number of such wall crossings needed to pass from
the positive chamber to the given chamber. Thus, it is the number of negative numbers in
the set {h, i : + } for in the chamber. The length of a Weyl group element w is
the distance from w to the positive chamber.
17.22 Borel-Weil-Bott Theorem. Let be an integral weight. Then
(a) if + = w(+) for a dominant weight and w W of length d, then H d (X, H )
is isomorphic to the irreducible g module of highest weight ;
(b) H p (X, H ) vanishes in all other cases.
Proof. We already know from Theorem 17.20 that H p (X, H ) = 0 if + lies in a wall.
Thus, we may assume that + lies in a Weyl chamber i. e. that = w + for a
dominant integral weight , where = w . By Theorem 17.17 we have
H p (X, H ) = H p (X, H ) V
where V is the irreducible finite dimensional representation of highest weight . Thus, the
theorem is true for a given with + in the chamber determined by w if and only if it

180

J. L. TAYLOR

is true for the weight = w that is, if and only if H p (X, H ) = 0 unless p is the
length of w, in which case H p (X, H ) = C. Of course, this means that the theorem is true
for all weights with + in a given chamber if and only if it is true for one such weight.
We prove the theorem by induction on the distance d from our chamber to the positive
chamber. The case d = 0 is the Borel-Weil theorem. Thus, we suppose the theorem is true
for Weyl chambers at distance d 1 from the positive chamber and consider a chamber C
at a distance d, obtained by applying w of length d to the positive chamber. We choose
an integral weight which is in the positive chamber and is the highest weight of an
irreducible finite dimensional representation (, W ). Then w is in the our chamber C.
Now C is adjacent to a chamber C 0 at distance d 1 from the positive chamber. Let
be the positive root defining the wall separating the two chambers. Then w0 = S w is
the Weyl group element of length d 1 which maps the positive chamber to C 0 . We set
= w0 () C 0 .
As in the proof of Lemma 17.17, let be the representation restricted to the Borel
B and let be the tensor product of with the one dimensional representation of B
determined by an integral weight . Here we choose to be a weight that satisfies
h + , i = 0,

h + , i > 0 for

+ , 6=

For example, = w + w0 has this property. We then consider the induced bundle
I( ) and its sheaf of sections I( ). As in the proof of Lemma 17.17, I( ) = I(e ) I()
as G-equivariant bundles, I( ) = I(e ) W as sheaves of g-modules, and for each p
H p (X, I( )) = H p (X, H ) W
as g-modules. Since, + is in a wall, it follows from Theorem 17.20 that
H p (X, H ) = 0 and, hence,

H p (X, I( )) = 0

for all p.
Also, as in the proof of Lemma 17.17, the sheaf H decomposes into eigenspaces for the
action of the Casimir and the possible eigenvalues are of the form h + , + + 2i =
|| + + ||2 ||||2 where is a weight of the representation (, W ). Now we apply
Lemma 17.21 in the case where the positive root system is the one for which + is in
the positive chamber and is + . It implies that the maximal value for the expression
|| + + ||2 ||||2 , as ranges over the weights of , is achieved only for = and
= s ().
We have
h, i
s () = 2

h, i
and, since h, i > 0, this implies that s () < in our ordering of integral weights. We
may define a B-submodule W 0 of W to be the span of all weight spaces for weights < .
If W 00 is the quotient W/W 0 , then we have a short exact sequence of B-modules
0 W 0 W W 00 0

NOTES ON SEVERAL COMPLEX VARIABLES

181

with W 0 containing the weight space for weight s () and W 00 containing the weight
space for weight . If we tensor this sequence with the one dimensional representation
with character e and then induce, we are led to a corresponding short exact sequence of
sheaves of g-modules:
0 A I( ) B 0
If we then apply the projection onto the -eigenspace for eigenvalue t = || + + ||2
||||2 = ||s () + + ||2 ||||2 , we obtain a short exact sequence of sheaves:
0 At I( )t Bt 0
From the construction, it is clear that Bt = H( + ) and At = H(s () + ). Since I( )
has vanishing cohomology in all degrees, the same thing is true of its direct summand
I( )t . From the long exact sequence of cohomology, we conclude that
H p+1 (X, s () + ) ' H p (X, + )
for all p. Since + is in the wall determined by , it is fixed by s and so, since is in
the chamber C 0 , it follows that + + is also in C 0 . Then
s () + + = s ( + + )
is the corresponding element of the chamber C. Since we have assumed the theorem true
for all + in C 0 and, hence, for + = + + , the above identity for cohomology
shows that the theorem is also true when + is the element s ( + + ) of the chamber
C. As noted above, the theorem is true for all + in a chamber if it is true for one. This
completes the proof.

17. Problems
1. Let (, W ) be a finite dimensional holomorphic representation of a Borel subgroup B.
Prove that the induced bundle I() on X is a trivial G-equivariant vector bundle if and
only if is the restriction to B of a holomorphic representation of G.
2. Prove that if (, W ) is a finite dimensional holomorphic representation of B then there
is a holomorphic action of G on H p (X, I()) for each p.
3. Suppose that A is a linear operator on a vector space V (not necessarily
finite dimenQ
sional) and suppose that 1 , . . . , n are distinct scalars such that i (A i ) = 0. Then
prove that V decomposes as a direct sum of eigenspaces of A with eigenvalues 1 , . . . , n .
2h, i
2h, i
4. Let and be roots. Use the fact that
and
are integers (Theorem
h, i
h, i
2h, i
17.7) to prove that if |||| |||| then
can only have the values 0, 1, or 1.
h, i
5. Use the result of Problem 4 to prove that if and are non-proportional roots and
h, i > 0 then is a root.

182

J. L. TAYLOR

6. Let + be a system of positive roots. A simple element of + is one which is not a

sum of two elements of + . Use the result of Problem 5 to prove that if and are
distinct simple elements of + then h, i < 0.
7. Prove that the set {i } of simple elements of + is a basis for the real vector space
spanned by the roots. Furthermore, each element of + has an expansion in this basis
with coeficients which are all positive. This basis is called the system of simple roots
generating + .
8. Let {i } be the system of simple roots generating + as in Problem 7. Choose a basis
{i0 } for h bi-orthogonal to {i } relative to the Killing form ( i. e. choose {i0 } so that
hi , j0 i = ij ). Prove that a weight is in the positive Weyl chamber if and only if its
expansion in terms of the basis {i0 } has all positive coeficients.
9. Prove that h, i > 0 for every positive root .

NOTES ON SEVERAL COMPLEX VARIABLES

183

References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.

Atiyah, M. F. and Macdonald, I. G., Introduction to Commutative Algebra, Addison-Wesley, 1969.

Bourbaki, N., Groupes et alg`
ebras de Lie, Masson, 1975.
Bredon, G. E., Sheaf Theory, McGrawHill, 1967.
Cartan, H. and Serre, J. P., Un th
eor`
em de finitude concernant les vari
eti
es analytiques compactes,
C. R. 237 (1953), 128130.
Gunning, R. C., Introduction to Holomorphic Functions of Several Complex Variables, Wadsworth
and Brooks/Cole, 1990.
Gunning, R. C. and Rossi, H., Analytic Functions of Several Complex Variables, Prentice-Hall, 1965.
Hartshorne, R., Algebraic Geometry, SpringerVerlag, 1977.
Humphreys, J. E., Introduction to Lie Algebras and Representation Theory, SpringerVerlag, 1970.
Hungerford, T. W., Algebra, Holt, Rinehart and Winston, 1974.
Kashiwara, M., Schapira, P., Sheaves on Manifolds, Springer Verlag, 1990.
Mili
ci
c, D., A proof of the BorelWeil theorem, unpublished notes.
Rudin, W., Functional Analysis, McGraw Hill, 1973.
Serre, J. P., Geom
etrie alg
ebrique et g
eom
etrie analytique, Ann. Inst. Fourier 6 (1956), 142.
Serre, J. P., Faisceaux alg
ebriques coh
erents, Ann. of Math. 61 (1955), 197278.
Schaefer, H. H., Topological Vector Spaces, Macmillan, 1966.
Zariski, O and Samuel, P., Commutative Algebra, Van Nostrand, 1960.