Tifr13 PDF

Lectures on
The Theory of Functions of

Several Complex Variables
By
B. Malgrange
Tata Institute of Fundamental Research

Bombay
1958
(Reissued 1984)
Lectures on
The Theory of Functions of
Several Complex Variables
By
B. Malgrange
Notes by
Raghavan Narasimhan
Distributed for the

Tata Institute of Fundamental Research
Springer-Verlag
Berlin Heidelberg New York Tokyo
1984
Author
B. Malgrange
Institut Fourier, Universite de Grenoble I
Laboratoire de Mathematiques Pures
38402 Saint-Martin-dHères Cedex
FRANCE
c Tata Institute of Fundamental Research, 1984

ISBN 3-540-12875-1 Springer-Verlag, Berlin. Heidelberg.

New York, Tokyo
ISBN 0-387-12875-1 Springer-Verlag, New York, Heidelberg.
Berlin. Tokyo
No part of this book may be reproduced in any

form by print, microfilm or any other means with-
out written permission from the Tata Institute of
Fundamental Research, Colaba, Bombay 400 005
Printed by M. N. Joshi at The Book Center Limited, Sion East,

Bombay 400 022 and published by the Tata Institute of Fundamental
Research, Bombay.
PRINTED IN INDIA
Contents
I Domains of Holomorphy 1
1 Cauchys formula and elementary consequences 3
2 Reinhardt domains and circular domains 11

1 Reinhardt domains . . . . . . . . . . . . . . . . . . . . 11
2 Domains of convergence of powe series . . . . . . . . . 13
3 Circular domains . . . . . . . . . . . . . . . . . . . . . 16
3 Complex analytic manifolds 19
4 Analytic Continuation 25
5 Envelopes of Holomorphy 29
6 Domains of Holomorphy: Convexity Theory 35
7 Convexity Theory (continued) 41
II Differential Properties of the Cube 53
8 d -cohomology on the cube 55

1 Differential forms . . . . . . . . . . . . . . . . . . . . . 55
2 The operators d and d . . . . . . . . . . . . . . . . . . 56
3 Triviality of d -cohomology on a cube . . . . . . . . . . 57
4 Meromorphic functions . . . . . . . . . . . . . . . . . . 61
v
vi Contents
5 The first Cousin problem . . . . . . . . . . . . . . . . . 62
9 Holomorphic Regular Matrices 65
10 Complementary Results 75
1 Generalization of Grothendiecks theorem . . . . . . . . 75
2 Linear bundles . . . . . . . . . . . . . . . . . . . . . . 76
3 Application to the second Cousin problem . . . . . . . . 78
III Coherent Analytic Sheaves 85
11 Sheaves 87
12 General properties of Coherent Analytic Sheaves 91

1 Analytic Sheaves . . . . . . . . . . . . . . . . . . . . . 91
2 Coherent analytic subsheaves of O m . . . . . . . . . . . 93
3 General coherent analytic sheaves on a... . . . . . . . . . 95
4 Coherent analytic sheaves on subsets of a... . . . . . . . 97
13 Cohomology with coefficients in a sheaf 101

1 Cohomology of a covering . . . . . . . . . . . . . . . . 101
2 Cohomology of the space X . . . . . . . . . . . . . . . . 102
3 The exact cohomology sequence . . . . . . . . . . . . . 104
14 Coherent analytic sheaves on a cube 107

1 The abstract de Rham Theorem . . . . . . . . . . . . . . 107
2 Coherent analytic sheaves on a cube . . . . . . . . . . . 110
15 Stein Manifolds: preliminary results 115

1 Theorems A) and B) for closed polydiscs in C n . . . . . 115
2 Coherent analytic sheaves on an analytic submanifold . . 117
3 Stein Manifolds . . . . . . . . . . . . . . . . . . . . . . 119
16 Coherent analytic sheaves on a Stein manifold 123

2 Topology on H (V, F ) . . . . . . . . . . . . . . . . . . 124
3 Proof of the fundamental theorem . . . . . . . . . . . . 127
Part I
Domains of Holomorphy
1
Chapter 1
Cauchys formula and

elementary consequences
1. Let C n be the space of n complex variables (z1 , . . . , zn ). We write 1

simply z for (z1 , . . . , zn ). Let z j = x j + iy j , j = 1, . . . , n, and let be an
opne set in C n . Suppose that f (z) = f (z1 , . . . , zn ) is a complex-valued
function defined in which is (once) continuously differentiable as a
function of the 2n real variables x1 , y1 ; . . . ; xn , yn .
Set, by definition,
!
f 1 f f
= i
z j 2 x j y j
and !
f 1 f f
= +i .
z j 2 x j y j
f
Definition. f (z) is said to be holomorphic in is = 0, j = 1, . . . , n,
z j
at every point of . (These equations generalize the Cauchy-Riemann
equations to the case of functions of several variables).
A definition which is equivalent to the above, is t the following:

f (z) is said to have a complex derivative at a if, for any b C n ,
3
4 1. Cauchys formula and elementary consequences
complex,
f (a + b) f (a)
lim
0
exists. f (z) is said to be holomorphic in if it has a complex derivative
at every a .
Definition . A polydisc, with centre at the origin is the set K of points

such that
|z1 | 1 , |z2 | 2 , . . . , |zn | n , 1 , 2 , . . . , n > 0.
These inequalities are, for brevity, written
|z| .
2 Let denote the set of points z C n for which
|z1 | = 1 , . . . , |zn | = n .
Consider a function f (z) holomorhpic in a neighbourhood of K and

denote by c j the curve |z j | = j in the z j -plane. Then the following
theorem holds.
Cauchys formula. If z is a point with |z| < (i.e., |z j | < j , j =
1, . . . , n), then
Z Z
1 f (1 , . . . , n )
f (z1 , . . . , zn ) = n
... d1 . . . dn .
(2i) (1 z1 ) . . . (n zn )

R R
(The integral . . . g(1 , . . . , n )d1 . . . dn , is defined, for continu-
R R
ous g, to be c d1 . . . c g(1 , . . . , n )dn and is independent of the order
1 n
in which the repeated integration is performed.)
Proof. Repeated application of the Cauchy formula for holomorphic

functions of one complex variable gives
Z
1 f (1 , z2 , . . . , zn )
f (z1 , . . . , zn ) = d1
2i c1 1 z1
5
Z Z
1 d1 f (1 , 2 , z3 , . . . , zn )
= d2
(2i)2 c1 1 z1 c2 2 z2
Z Z
1 d1 d2
= ... = n
...
(2i) c1 1 z1 c2 2 z2
Z
f (1 , . . . , n )
dn
cn n zn
Z Z
1 f (1 , . . . , n )
= n
d1 . . . dn .
(2i) (1 z1 ) . . . (n zn )

2. If f is a complex valued continuous function on a compact space

K, we define X
|| f ||K = | f (x)|.
xK
P
Definition. A series am (z) of complex valued continuous functions 3
mN n
on a compact space K is said to converge normally if
X
||am ||K < +.
mN
(N is the set of non-negative integers).

If am (z) are functions defined in an open set , we say the series con-
verges normally in if it converges normally on every compact subset
of .
The following are simple consequences of Cauchys formula.
Proposition 1. If f (z) is holomorphic in a neighbourhood U of a poly-
disc K, then
X
j j
f (z1 , . . . , zn ) = a j1 ,..., jn z11 . . . znn for z K. (1)
( j1 ,..., jn )N n
The series converges normally on K.

The proposition is proved simply by applying Cauchys formula to
o o
a polydisc K wiht K K K U (K is the interior of K ).
The coefficients a j = a j1 ,..., jn in (1) are given by

Z Z
1 f (1 , . . . , n )
a j1 ,..., jn = n
... j +1 j +1
d1 . . . dn
(2i) 11 . . . nn

Z 2 Z 2
1 f (1 ei1 , . . . , n ein )
= . . .
(2)n 0 0
j
11 . . . nn
j
ei( j1 1 ++ jn n ) d1 . . . dn (2)
From the expansion of f (z) as a power series (1), it is seen that f (z)
is indefinitely differentiable and that
j1 +...+ jn f (0)
= j1 ! . . . jn !a j1 ,... jn
z1j1 . . . znjn
4 so that the expansion is unique. By differentiation in Cauchys formula,
we have also
Z Z
j1 +...+ jn f (z) j1 ! . . . jn ! f (1 , . . . , n )
j j
= ... d1 . . . dn
z11 . . . zznn (2i)n (1 z1 ) j1 +1 . . . (n zn ) jn +1

(3)
If f (z) is assumed to be holomorphic only in the interior of K, we
can write X
f (z) = aJ zJ
JN n
where Z Z
1 f (1 . . . , n )
aJ = ... d1 . . . dn ,
(2i)n j +1
11
j +1
. . . nn

being a set of points of the form
|z1 | = r1 , . . . , |zn | = rn , 0 < r j < j, j = 1, . . . , n.
The series converges normally in the interior of K.
If f (z) is holomorphic in a neighbourhood of K and M = max
z
| f (z)| = max | f (z)| (the latter equality is easily established using
zK
Cauchys formula), then
M
|aJ |
J
7
i.e.
j j
|a j1 ,..., jn | M/(11 . . . nn ).
This follows at once from the expression (2) for aJ as an integral in

terms of f (z). The inequalities are called the Cauchy inequalities.
Proposition 2. Let { fk (z)} be a sequence of functions holomorphic in

and suppose that { fk (z)} converges, uniformly on every compact subset
of , to a function f (z). Then f (z) is holomorphic in .
Proof. Since fk (z) f (z) uniformly in a neighbourhood of any z , 5

f (z) is continuous in . Also, in a polydisc K about any point of
(lying wholly in ), fk (z) verifies Cauchys formula, and { fk (z)} being
uniformly convergent on K, so does f (z), from which it follows easily
that f (z) is holomorphic in .
Also, using the integral (3) for the derivatives of a holomorphic func-
tion, one proves that the derivatives (opf all orders) of fk (z) converge to
the corresponding derivatives of f (z), uniformly on every compact sub-
set of .
Another interpretation of the Proposition 2 is as follows. If C , H
denote the sets of continuous and holomorphic functions in respec-
tively, C and H are vector spaces over the field of complex num-
bers. We recall that one may topologize C , H by putting on them the
topology of uniform convergence on compact sets, namely if fn C
(or H ), fn 0 if || fn ||K 0 as n for every compact K . A
fundamental system of neighbourhoods of the origin is given by the sets
U (Km , 1/m), where U (K, a)(a > 0) is the set of f C (resp. H ) for
which || f ||K < a, and {Km } is a sequence of compact sets with
o [
o
Km K m+1 , K m = .
m=1
C is an (F )-space, i.e., it has a countable fundamental system of neigh-

bourhood of 0 and is complete.
Proposition 2 may be expressed by saying that H is a closed sub-
space of C .
Proposition 3. Every bounded closed set in H is compact. 6

(A bounded set in any topological vector space is a set such that
to any neighbourhood U of the origin, there exists a > 0 such that
U . C or H , we may say equivalently that a set is bounded
P
if || f ||K < + for every compact K .).
f
Proof. Let K be any compact subset of . Let K be the set of functions

fK , where fK is the restriction of f to K. We prove that K is
equicontinuous from which it follows by means of Ascolis theorem (see
e.g. Bourbaki: Topologie Generale, Chap.X, p.48) that K is relatively
compact in CK .
o
Choose a compact set K so that K K K . K has positive
distance from the boundary of K . Since sup || f ||K < + Cauchys
f
inequlity applied to the derivatives of the f in a suitable polydisc about
an arbitrary point of K, contained in K shows that
f
sup max || ||K = MK < .
f j z j
f f f f
Since = 0, it follows from the definitions of , that ,
z j z j z j x j
f f
, are bounded uniformly on K, for f (actually, || ||K MK ,
y j x j
f
|| ||K MK ). The mean value theorem now shows that K is equicon-
y j
tinuous.
To prove now that is compact, it suffices to prove that any se-
quence { fm }, fm has a limit point in C . We choose a sequence
S
{Km } of compact sets Km with Km K m+1 , m=1 Km = . Since K is
relatively compact for every compact K , we choose, inductively,
subsequences { fm,y } of { fm }, such that each is a subsequence of the pre-
7 ceding, while if fm,+1 = fm ,0 then m > m, and { fm, } converges uni-
S
formly on K . Since K m = , the subsequence { f } of { fm } converges
to a limit in C and Proposition 3 is proved.
9
Suppose given a subset of H such that sup | f (z)| < + for every
f
z . The question arises as to what one can assert about the bounded-
ness of the set . The following result holds.
Proposition 4. There exists an open set , which is dense in
such that (the set of the restrictions to of functions of is a
bounded set in H .)
The proposition and its proof remain valid if H is replaced by C .
We need the following
Theorem of Baire. Let U be an open set Cn and {O}, k = 1, 2, . . . a

T

sequence of open sets U, such that each Ok is dense in U. The Ok
k=1
is dense in U.
Proof. Since O1 is dense in U, given any open ball B U, O1 B
contains an open ball B1 . In the same way, O2 12 B1 ( 21 B1 is the ball
with the same centre as B1 and half its radius) contains an open ball
T
B2 and this process can be continued. Clearly B k , 0 and since
k=1
T
T

B k Bk1 , Bk , 0. Also Bk Ok B, so that Ok B , 0 and
k=1 k=1
T
Ok is dense in U.
k=1
Proof of the proposition.

Let U be an arbitrary open set . Let Ok U be the set of
z U for which there exists at least one f for which | f (z)| > k.
T

Clearly Ok is open. Also since for any z , sup | f (z)| < +, Ok =
f k=1
0. By Baires theorem, at least one Ok is not dense in U. Hence U 8
contains an open set OU contained in the complement of Ok for some
k and the functions of are uniformly bounded for z OU (by k). If
S
we set = OU , is clearly open and dense in . If z , z
U
possesses a neighbourhood on which is uniformly bounded and so
is bounded uniformly on every compact subset of (by the Borel-
Lebesgue lemma).
Chapter 2
Reinhardt domains and

circular domains
1 Reinhardt domains
9
Definition. A Reinhardt domain is an open set Cn such that
(z1 , . . . , zn ) implies (ei1 z1 , . . . , ein zn )
for all real 1 , . . . , n .
Theorem 1. Let C n be a connected Reinhardt domain containing

0 and suppose that f (z) is holomorphic in , Then f (z) can be written
X
f (z) = am zm
mN n
in , the series converging normally in . Such an expansion is unique.

P
Proof. Let be the set of z which have distance > |z|(|z2 = |zi |2 )
from the boundary of . Let be the connected component of
0. Then [
= .
>0
11
12 2. Reinhardt domains and circular domains
For, if z, we can join z to the origin by a path in . This path has a

distance > 0 from the boundary of . If is small enough the path lies
in and so in . In particular, z .
Define, for z
Z Z
1 f (t1 z1 , . . . , tn zn )
g(z1 , . . . , zn ) = . . . dt1 . . . dtn .
(2i)n (t1 1) . . . (tn 1)
|t1 |=1+ |tn |=1+
The integral is defined, for if (z1 , . . . , zn ) , then ((1 + )z1 , . . . , (1 +

)zn ) for the distance between these two points is |z|. Hence, since
is a Reinhardt domain, (t1 z1 , . . . , tn zn ) for all (t1 , . . . , tn ) with |t j | = 1+
. By differentiation under the integral, it is seen that g(z) is holomorphic
10 in . Moreover, if we choose a polydisc K with centre 0 such

that (1 + )K , then for z K, (t1 z1 , . . . , tn zn ) for all |t j | 1 + so
that, by Cauchys formula,

g(z1 , . . . , zn ) = f (z1 , . . . , zn ), z K.
Now, if two holomorphic functions in an open, connected set U co-
incide in some open set in U, then they coincide in the whole of U. (See
the principle of analytic continuation in III). Hence
Z Z
1 f (t1 z1 , . . . , tn zn )
f (z) = n
. . . dt1 . . . dtn
(2i) (t1 1) . . . (tn 1)
|t1 |=1+ |tn |=1+
in . Moreover,
1 1 X 1 1
= . . . mn
(t1 t) . . . (tn 1) t1 . . . tn (m1 ,...,mn )N n
t1m1 tn
the series being normally convergent on |t j | = 1 + , j = 1, 2, . . . , n.

Hence X
f (z1 , . . . , zn ) = m1 ...mn (z)
(m1 ,...,mn )N n
where
Z Z
1 f (t1 z1 , . . . , tn zn )
m1 ...mn (z) = ... dt1 . . . dtn .
(2i)n t1m1 +1 . . . tnmn +1
|t1 |=1+ |tn |=1+
2. Domains of convergence of powe series 13
Exactly as we proved above that g(z) = f (z) in , we prove, using

formula (3) for the derivatives of a holomorphic function, that

1 m1 +mn f (t1 z1 , . . . , tn zn )
m1 ...mn (z) = t j = 0
m1 ! . . . mn ! tm1 . . . tnmn
1
mn
zm 1
1 . . . zn m1 ++mn f (0)
= mn .
m1 ! . . . mn ! zm 1
1 . . . zn
From the integral representation of , it is seen that if z lies in a 11

compact subset K and is small enough
MK
|m1 ...mn (z)|
(1 + )m1 . . . (1 + )mn
where MK depends only on K, so that the series converges normally on
K. The expansion in the whole of follows easily by letting 0.
The uniqueness is obvious.
2 Domains of convergence of powe series

Let X
am zm
mN n
be a given power series. We define its domain of convergence, D, as
follows: zD if the series converges absolutely in a neighbourhood of
z. D is clearly open. We define the set B as the set of z for which there
exists C > 0 such that |am zm | C for all mN n . Clearly D B (interior
of B). But we can prove that D = Bo . This follows from
Abels lemma: If z = (z1 , . . . , zn )B, then (1 z1 , . . . , n zn )D if |1 | <
1, . . . , |n | < 1 and the series converges normally in D.
P
Proof. If |am zm | C, the general term of the series am1 ...mn
mN n
(1 z1 )m1 . . . (n zn )mn is majorized by C|1 |m1 . . . |n |mn and the result fol-
lows.
Consequences of Abels lemma

(a) D = B . This is obvious.
(b) The series represents a holomorphic function in D (by I, Prop 2)
12 (c) D is a Reinhardt domain. More, if (z1 , . . . , zn ) then (1 z1 , . . . ,

n zn )D if |1 | 1, . . . |n | 1,
and so D is a union of polydiscs (all with centre).

We give an example, due to Hartogs, of a domain in C 2 that any
holomorphic function in the domain can be extended to a larger domain.
Let G be the domain consisting of the points z with
|z1 | < , |z2 | < 1
and the points with
|z1 | < 1, 1 < |z2 | < 1.
In the figure we have indicated the variation in |z1 |, |z2 |. By Theorem

1, any function holomorphic in G can be expanded in a power series
in G and, by Abels lemma, the power series represents a holomorphic
function in the open polydisc
|z1 | < 1, |z2 | < 1.
A special case is the following:

Let be an open set in C n (n > 1) and a a point of . Let f (z)
be holomorphic in a. Then f (z) can be continued holomorphically
throughour .
2. Domains of convergence of powe series 15
Consequence (c) of Abels lemma shows that the domain conver-

13 gence of a power series is a Reinhardt domain, but this is not all; not
every Reinhardt domain, which is the union of open polydiscs, is the
domain of convergence of a power series. We can, however, character-
ize the domains of convergence of power series.
Let D Rn be the set consisting of the points (log |z1 |, . . . , log |zn |)
where (z1 , . . . , zn )D i.e. D is the image of D under the mapping :
C n Rn defined by (z1 , . . . , zn ) = (log |z1 |, . . . , log |zn |). Let B be the
image of B under (D and B are the sets of points defined on p.11 asso-
ciated with the power series). If (1 , . . . , n )D then (1 t1 , . . . , n
tn )D if t1 0, . . . , tn 0.
The following fundamental result holds.
Theorem 2. D is a convex set in Rn .

Proof. Since D = B0 , D = B so that it suffices to prove that B is
S
convex. Now B = BC , where BC is the image, under , of the set
C>0
of zC n where |am zm | C for all mN n . Since BC BC , if C < C it is
T
enough to prove that BC is convex. Also BC = BC,m , where BC,m is
mN n
the image of the set of zC n , |am zm | C for a fixed m, under . Thus,
we have only to prove that BC,m is convex. Now B is the image of the

mn
C,m
set of zC n for which am1 ...mn zm 1

1 . . . zn C and so is the set of points
(1 , . . . , n )Rn ( j = log |z j |) at which
log |am1 ...mn | + m1 1 + + mn n log C,
which, being a half space in Rn , is convex. This proves Theorem 2.

For example, consider a series convergent in the domain in C 2 con- 14
sisting of the points |z1 | < 1, |z2 | < e, and the points |z1 | < e, |z2 | < 1. It
has the image shown in Fig. (a) (the set ) in the (|z1 |, |z2 |)-plane and D
contains the set in Fig. (b). Since D is convex, it contains also the
set in Fig. (b) and so the series converges at the points in C 2 which
are mapped by
: z (log |z1 |, log |z2 |)
into the set in Fig. (a).
Fig. (a) Fig. (b)
The converse of Theorem 2 is true; that is if a Reinhardt domain,

which is the union of polydiscs, is such that its image under is convex,
then it is (precisely) the domain of convergence of a power series. We
shall prove this later (in VII, p.44)
3 Circular domains
We consider next the expansion of holomorphic functions in series of
homogeneous polynomials.
Definition . An open set C n is said to be a circular domain if z
15 implies ei z i.e. (ei z1 , . . . , ei zn ) for all real .
Theorem 3. Let be a connected circular domain and let 0. Sup-
pose f (z) to be holomorphic in . Then f (z) can be expanded in a series
of homogeneous polynomials,
X

f (z) = Pk (z) in
k=0
(Pk (z) is homogeneous, of degree k in z1 , . . . , zn ) and the series con-

verges normally in . An expansion of this form is unique.
Proof. Define as in Theorem 1 and consider the integral
Z
1 f (tz1 , . . . , tzn )
dt.
2i t1
|t|=1+
3. Circular domains 17
Exactly as in the proof of Theorem 1, we choose a polydisc K about

the origin such that (1 + )K and then, if zK, (tz1 , . . . , tzn )
for |t| 1 so that, the integral being holomorphic in we have, by
Cauchys formula,
Z
1 f (tz1 , . . . , tzn )
f (z1 , . . . , zn ) = dt.
2i t1
|t|=1+

in K and so, since is connected, in . Since
1X1

1
= ,
t1 t k=0 tk
the series converging normally on |t| = 1 + , we have

X

f (z1 , . . . , zn ) = Pk (z)
k=0
where Z
1 f (tz1 , . . . , tzn )
Pk (z) = dt.
2i tk+1
|t|=1+
As in the proof of Theorem 1, the series converges normally in and, 16

repeating the above argument, it follows that

1 dk f (tz1 , . . . , tzn )
Pk (z) =
k! t=0
1 X k f (0)
= zm1 . . . zm n
n
k! m ++m =k m1 z1 . . . zm n
n 1
1 n
The theorem follows by letting 0.

P

It is easily seen that if f (z) has an expansion Pk (z) which con-
k=0
verges uniformly in a neighbourhood of 0, then Pk (z) has the above
form. This proves Theorem 3.
The above theorem shows that if a function f (z) is holomorphic is
a circular domain (which is connected and contains 0) then it can be
S
holomorphically continued to (t). This is proved in the same way
0t1
as Abels lemma.
Chapter 3
Complex analytic manifolds
Definition . A Hausdorff, topological space V n is called a topological 17

manifold of dimension n (n 0 an integer) if it has followsing property:
every point a V n has a neighbourhood homeomorphic to an open set
Rn .
(Note that a space having this property is not automatically Haus-

dorff.)
V n is said to be countable at infinity, if V n is a countable union of
compact sets.
We next recall, without proof, the following two propositions.
Proposition 1. If V n is connected the following three conditions are

equivalent:
(1) V n is countable at infinity.
(2) V n is paracompact (i.e. any open covering (Ui )iI admits a locally
finite refinement, namelyl there is another open convering (W j ) jJ
each set of which is contained in at least one Ui and such that
any point has a neighbourhood intersecting only a finite number
of the W j )
(3) V n has a countable open base.
19
20 3. Complex analytic manifolds
Proposition 2 (Poincare - Volterra theorem). If V n , W n are two n di-

mensional manifolds and if (1) V n is connected, (2) W n is countable at
infinity, (3) there exists a continuous mapping : V n W n which is
a local homeomorphism (i.e. every a V n has an open neighbourhood
18 which is mapped homeomorphically on an open set of W n ). Then V n is
countable at infinity.
Differentiable manifolds. Let V n be a topological manifold of dimen-

sion n. by an (indefinitely) differentiable or C -structure on V n is meant
a family {Oi }iI of open sets V n which cover V n , and mappings { fi }iI
such that fi maps Oi homeomorphically onto an open subset O i Rn
and such that the mappings fi f j1 and f j fi1 are C -mappings of
f j (Oi O j ), fi (Oi O j ) respectively, onto fi (Oi O j ), f j (Oi O j ), i.e.
if the correspondence
fi (Oi O j ) f j (O j O j )
is C .
(Note that it may happen that one can define more than one differen-
tiable structure on a manifold V n , so that a C -manifold is not a special
topological manifold, but is a topological manifold with an additional
structure).
If {Ok , fk } defines a C -structure on V n , we say that it defines the
same structure as {Oi , fi } if (and only if) the correspondence fi (Oi
Ok ) fk (Oi Ok ) is C . If the intersections Oi O j or Oi Ok
are empty we take it that the condition is satisfied. We make a similar
convention whenever we speak of properties of mappings of Oi O j or
Oi Ok , without stating this explicitly.
If is a mapping of V to W, where V, W are C -manifolds (not
necessarily of the same dimension) with C -structures {Oi , fi }, {O j , f j }
we say that is a differentiable or a C -mapping if f j fi1 is a
19 C -mapping, of fi (Oi 1 (O )). If V and W have the same dimension,
we say that is a diffeomorphism if is a homeomorphism of V onto W
and if and 1 are differentiable.
System of local coordinates
21
Definition . Let V be a C -manifold and let a V. If (X1 , . . . , Xn ) is a

system of real valued functions in an open neighbourhood W of a such
that the mapping : W Rn defined by b (X1 (b), . . . , Xn (b)) for
every b W, is a diffeomorphism, then (X1 , . . . , Xn ) are said to form a
system of local coordinates at a.
If Y1 , . . . , Yn are C -functions in a neighbourhood of a, they form a
system of local coordinates at a if and only if the Jacobian
D(Y1 , . . . , Yn )
D(X1 , . . . , Xn )
does not vanish at a. The proof follows easily from the implicit function
theorem.
Complex analytic manifolds.
Definition. Let , be two open sets in C n , f1 , . . . , fn complex valued
functions in . We say that f = ( f1 , . . . , fn ) is an analytic isomorphism
of onto if the mapping f = ( f1 , . . . , fn ) of C n is a diffeomor-
phism of onto and the functions f1 , . . . , fn are holomorphic in .
The composite of analytic isomorphisms is an analytic isomorphicm.
Moreover, the inverse of an analytic isomorphism is also one. This is
seen as follows:
Let (!f1 , . . . , fn ) be n holomorphic functions of z1 , . . . , zn , and let J = 20
fk
det and fk = fk(1) + i fk(2) , zk = xk + iyk . Then the determinant
zl
A B
C D, where
(1) (2) (1) (2)
f f f f
A = k
, B = k
, C = k
, D = k
xl xl yl yl

A1 B1
is equal to , where
C1 D1
! ! ! !
fk fk fk fk
A1 = , B1 = , C1 = , D1 =
zl zl zl zl
and, by the Cauchy-Riemann equations, this equals |J|2 so that, since

( f1 , . . . , fn ) = f is a diffeomorphism, J , 0. We can now easily compute
z
and show that it is zero which proves the statement.
fk
Definition. Let V be a topological manifold of dimension 2n. We shall
identify R2n with C n . A system {Oi , fi }iI , where {Oi }iI is an open cov-
ering of V and fi are mappings, fi : Oi O e i C n , is said to define
a complex analytic structure if the mapping fi fk1 defines an analytic
isomorphism of f j (Oi O j ) onto fi (Oi O j ), i.e., if the correspondence
fi (Oi O j ) f j (Oi O j )
is an analytic isomorphism for every i, j I. Two systems {Oi , fi },

21 {Ok , fk } define the same complex analytic structure if the correspon-
dence fi (Oi Ok ) fk (Oi Ok ) is an analytic isomorphism for every
i, k.
We call V a complex analytic manifold of (complex) dimension n.
Definition. Let V n , W m be two complex analytic manifolds, with struc-

tures {Oi , fi }, {O j , f j } and a map V n W m . is said to be an analytic
mapping if the mappings f j fi1 are analytic mappings of O j to
C m , i.e., if the component functions are holomorphic in Oi . is called
an analytic isomorphism of V n onto W m if it is an analytic map and is,
moreover, a diffeomorphism.
Local coordinates.
Let V be a complex, analytic manifold. A system of n complex
valued functions (z1 , . . . , zn ) in a neighbourhood of a point a V is
said to form a system of local coordinates at a if there exists an open
set W, a W such that the mapping : W W C n defined by
b (z1 (b), . . . , zn (b)) is an analytic isomorphism n holomophic func-
tions t1 , . . . , tn (i.e., analytic mappings into C 1 ) in a neighbourhood of a
form system of local coordinates if and only if the determinant
" #
tk
J = det , 0.
zl a
23
This follows from the fact that Jacobian of the Rti , Jt j in terms of the
Rzi , Jz j equals |J|2 , 0 and the remark on the inverse of an analytic
isomorphism.
The question arises as to which theorems on holomorphic functions
in an open set C n generalize to holomorphic functions on a complex
analytic manifold V. The following theorems, and their proofs, do. 22
(1) HV is closed in CV (the notation is the obvious one).
(2) A closed bounded set in HV is compact. The diagonal process

fails if V is not countable at infinity, but it is easy to prove that
any ultrafilter in converges, by considering, for any compact
K V, the restrictions to K of the functions in the elements of
the ultrafilter.
P
(3) If HV (or CV ) and | f (a)| < + for every a V, then
f
there exists an open set V V dense in V such that V (the set
of restrictions to V of the functions of ) is a bounded set in HV
(respectively CV ).
This is because Baires theorem (used in I, Prop.4) is true for open

subsets of locally compact spaces or complete metric spaces (and any
manifold is locally compact).
It is also sometimes of interest to apply Baires theorem in HV , but
this necessitates the assumption that V is countable at infinity when HV
is a Frechet space and so a complete metric space.
The principle of analytic continuation.
Let V be a connected, W an arbitrary, complex analytic manifold,
and let f1 , f2 be two analytic mappings of V into W. Let {Oi , i }, {O j , j }
define the structures on V, W respectively.
Let a Oi , f1 (a) = f2 (a) O j . We say that f1 , f2 have the same
derivatives at a if the components of the mappings j f1 1 i and
1
j f2 j have the same derivatives (of all orders) at i (a). This
definition is independent of the Oi , O j containing a, f1 (a) respectively,
and of the systems chosen to define the analytic structures.
The following theorem then holds: 23
Theorem. (1) (The strong principle of analytic continuation).

If f1 and f2 and all their derivatives coincide at a point a V,
then f1 = f2 everywhere on V.
(2) (The weak principle of analytic continuation).

If f1 = f2 in an open set on V, f1 = f2 everywhere on V.
Proof. The weak principle follows at once from the strong principle so
that we have only to prove the latter. Let E be the set of points where
f1 and f2 and all their derivatives coincide. It is clear that E is closed.
Suppose now that b E. Then all the components of the functions
k f1 1 1
l and k f2 l (b Ol , f1 (b) = f2 (b) Ok ) and all
their derivatives coincide at l (b). By expanding in power series about
l (b), it follows that these components and their derivatives coincide
in a neighbourhood of l (b) so that E is open. Since a E and V is
connected, E = V and the theorem follows.
Chapter 4
Analytic Continuation
All the manifolds considered in this and the next lecture are assumed to 24
be connected.
Definition . Let V, W be two complex analytic manifolds of complex

dimension n, a mapping of V W. We say that is a local analytic
isomorphism if every point a V has an open neighbourhood O such
that restricted to O is an analytic isomorphism.
We say that V is spread in W and that spreads V in W if is a local

analytic isomorphism of V in W.
One may define the continuation of a holomorphic function f on V,
to the manifold W in which V is spread by by saying that g is the
continuation of f to W if f = g . If such a g exists, it is unique, for if
g = f = g , then since is a local homemorphism, g = g in an
open set on W, and W being connected, g = g on W.
Example. (i) V is an open set W, is the inclusion map (a) = a

for every a V. The functions in V which can be continued to W
are precisely the restrictions to V of holomorphic functions on W.
(ii) V is a convering space of W and the natural projection. The

functions on V which can be continued to W are those functions
which have the same value at all points which lie over one point
of W.
25
26 4. Analytic Continuation
However, this definition of continuation turns out to be too general

25 to be of use. To have interesting theorems, it is necessary to restrict the
definition, and we introduce therefore the following
Definition . Let V be a complex analytic manifold, a map which

spreads V in C n . (The necessary and sufficient condition that such a
exist is that there are n global functions on V which form a system of
local coordinates at each point of V). Let (V , ) be another such pair,
spreading V in C n . Suppose also that there exists a map : V V
which spreads V in V , such that = (, , are assumed to be
given and fixed). Let f be a holomorphic function on V. We say that f
is the continuation of f from V to V if f is a holomorphic function on
V such that f = f .
Maximal continuation.
Let (V, ) be a pair consisting of the (n dimensional) complex an-
alytic manifold V and a spread of V in C n . Let f be a holomorphic
)
function on V. Suppose that there exists another such pair (V, with
the following properties:
),
(i) f can be continued to (V, i.e., there exists a holomorphic func-
tion f on V and a spread of V in V such that f = f ,
=
.

(ii) If f is a continuation of f to the pair (V , ) and is the spread

of V in V such that f = f , = , then there exists
a spread of V in V such that f = f and such that the
mapping : V V factorises into = . Then we can show
that we have also = and we call (V, , f) a maximal
,
continuation of (V, , f ).
26 To consider the problem of the existence and uniquencess of a maxi-

mal continuation, we shall have to introduce the so-called sheaf of germs
of holomorphic functions.
Let W be a complex analytic manifold (of complex dimension n).
Let a W. Consider the set of all holomorphic functions in open sets
containing the point a. We introduce an equivalence relation in this set
of functions by identifying two functions f , g, if, in a neighbourhood
27
of a, f = g. The equivalence classes are called germs of holomorphic

functions at a. fa will stand for a germ at a. It is clear that fa (a) has an
unambiguous meaing.
Denote by Oa the set of germs at a. The sheaf of germs of holomor-
S
phic functions on W is defined to be O W = O = Oa . A complex
aW
analytic structure can be put on O in the following way.
Let a W and let (a, fa ) Oa fa is defined by a holomorphic
function f in a neighbourhood U of a. Also, for every b U, f defines
S
a germ fb at b. We define (b, fb ) to be a neighbourhood of a. It is
bU
easy to verify that this defines a topology on O.
Proposition 1. O is a Hausdorff space.
Proof. Let (a, fa ) , (b, gb ) be two points of O.
If a , b we choose neighbourhoods Ua , Vb of a, b in W such that fa
S
is determined by f in Ua , gb by g in Vb and Ua Vb = 0. Then (c, fc ),
S cU a
(d, gd ) are disjoint neighbourhoods of (a, fa ), (b, gb ) in O.
dVb
If a = b, then fa , ga . Let U be a connected neighbourhood of a on 27

W such that fa , ga are defined by holomorphic functions f , g in U. Then
S S
the neighbourhoods (c, fc ), (c, gc ) of (a, fa ) (a, ga ) are disjoint, for
cU cU
if (c, fc ) = (c, gc ), then f , g coincide in a neighbourhood of c, and since
U is connected, f = g in U so that fa = ga which is not the case.
Let p be the projection O W defined by p(a, fa ) = a. This is a
mapping of O onto W. It is a local homeormorphism as follows at once
from the definition of the topology on O. It is clear now, how p can be
used to carry over the complex analytic structure from W to O.
Let now V be an n dimensional complex analytic manifold, a
spread of V in C n . Let O be the sheaf of germs of holomorphic func-
tions on C n . Let f be a holomorphic function on V. Let a V and U
a neighbourhood of a such that , restricted to U is an analytic isomor-
phism. Then the holomorphic function f 1 in (U) defines a germ
in O, viz., ( f 1 )(a) . We define a mapping : V O by setting

(a) = ((a), ( f 1 )(a) ) O. is a local analytic isomorphism which
spreads (V, ) in O.
28 4. Analytic Continuation
Let V be the connected component in O of (V), and define to be

the restriction to V of the projection p : O C n and define f(b, gb ) =
gb (b) for (b, gb ) V. Clearly = and f = f . Hence (V, ,
, f)
is a continuation of (V, , f ). We assert that it is a maximal continuation.
Suppose that (V , , f ) is any continuation of (V, , f ). Let be a local
analytic isomorphism V V such that f = f , = . We
28 can apply the above reasoning to (V , , f ) to continue it to (V , , f ).
The point (a) V (a V) is mapped onto ( ((a)), ( f 1 ) ((a)) )
and this = ((a), ( f 1 )(a) ). Hence the images of V and V have a
common point in O and by the definition of V, V we have V = V . It
follows easily that (V, , , f) is maximal.
If now (V, , f ) and (V , , f ) are both maximal continuations of

(V, , f ), : V V, : V V the corresponding spreads of V
in V, V respectively, then there exists a spread of V in V such that
f = f , = ,
= . Using the similar spread 1 of V
in V it is easily shown that is an analytic isomorphism. Thus we have
the following
Theorem . Let V be a complex analytic manifold, a spread of V in

C n . Let f be a holomorphic function on V. Then there exists a maximal
continuation (V, , f) of (V, , f ). If (V,
, , f) and (V , , , f) are
,
two maximal continuations, then there exists an analytic isomorphism
: V V such that f = f , = , = .
Chapter 5
Envelopes of Holomorphy
Let V be a complex analytic manifold of (complex) dimension n and 29

let spread V in C n . Let F = ( fi )iI be a subset of the set of all
holomorphic functions on V. We say that (V, , ( fi )iI ) is continuted
to (V , , , ( fi )iI ) if there exists a complex analytic manifold V , a
spread of V in C n , a system ( fi )iI of holomorphic functions fi on
V and a local isomorphism of V into V such that = and
fi = fi for i I. This process is called simultaneous continuation
of ( fi )iI to (V , ) from (V, ). A maximal continuation of ( fi )iI is a
continuation (V, , , ( fi )iI ) such that if (V , , , ( f )iI ) is any con-
i
tinuation of (V, , ( fi )iI ), then there is a local isomorphism of V V
such that for all i I, f = fi and = . It follows that = .
i

We can prove the existence and uniquencess of a maximal (simulta-
neous) continuation of a given system (V, , ( fi )iI ) in a manner similar
to the proof of the theorem in IV.
Consider an open neighbourhood U of a point a C n . Let (gi )iI
be a family of holomorphic functions in U (indexed by I). Let (gi )iI
be another such family defined in a neighbourhood U of a. Identify
(gi )iI and (gi )iI if there existsaneighbourhood W of a, W U U
such that gi = gi in W for every i I. Denote by (gi )a an equivalence
class of the set of all families (gi )iI of holomorphic functions, such that
all functions of one family are defined in a fixed neighbourhood of a
by the equivalence relation defined by this identification. The set of all
29
30 5. Envelopes of Holomorphy
S
(gi )a is denoted by OI,a . Let OI = OI,a . Then O I is a sheaf. The
aC n
30 topology on O I is defined exactly as before: if (a, (gi )a ) O I and U
is a neighbourhood of a, (gi )iI a family of holomorphic functions in
S
U defining (gi )a , then U = (b, (gi )b ) [(gi )b is the equivalence class
bU
defined by (gi )iI at b] is an open neighbourhood of (a, (gi )a ). Exactly as
in IV, we put on O I a complex analytic structure and define a mapping
from (V, ) to O I : (a) = ((a), (gi 1 )(a) ) and show that this indeed
gives us a maximal continuation. The uniqueness is proved in the same
way as in IV.
The most important case is that in which F consists of all holomor-
phic functions on V. In this case the maximal continuation (V, ,
)
is
called the envelope of holomorphy of (V, ).
Let V be a complex analytic manifold, 1 , 2 two local analytic
isomorphisms of V into C n . Let F = ( fi )iI be a family of holomorphic
functions on V. Let (V 1 , 1 , 1 , f1i ), (V 2 , 2 , 2 , f2i ) be the maximal con-
tinuations of (V, 1 , fi ), (V, 2 , fi ) respectively. The two continuations
are said to be isomorphic if there exists an analytic isomorphism of V 1
onto V 2 such that 2 = 1 and f1i = f2i for all i I.
We have the following
Theorem 1. Let F = ( fi )iI consist of all the holomorphic functions on
the complex analytic manifold V. Let 1 , 2 be two maps which spread
V in C n . Let (V 1 , 1 , 1 ), (V 2 , 2 , 2 ) be the envelopes of holomorphy
of (V, 1 ), (V, 2 ) respectively. Then (V 1 , 1 , 1 ) and (V 2 , 2 , 2 ) are iso-
morphic.
31 Proof. Consider the components of the mapping 2 : V C n . They
are holomorphic functions on V and they can be continued to V 1 and this
gives us a mapping 2 of V 1 to C n , such that 2 = 2 1 . Let J be the
Jacobian of 2 with respect to the local coordinates defined by 1 . Then
J is a holomorphic function on V, and since 2 is a local isomorphism,
J , 0 so that 1/J is holomorphic on V. Hence 1/J (resp. J) has a
g 1 (resp. J1 ) to V 1 . Clearly we have (1/J)
continuation (1/J) g 1 X J1 = 1 on

the image of V by 1 in V1 and since V1 is connected, (1/J) g 1 X J1 = 1
everywhere on V1 so that J1 , 0 throughout V1 . Moreover J1 is the

Jacobian of 2 with respect to the local coordinates defined by 1 so that
31
2 is a local analytic isomorphism, and spreads V 1 in C n . The situation

is explained by the following diagram:
V 7@
~~~ 7@77@@
~~ 77@@
~~ 1 77@@
~~ 2 77 @@@
~~ @@
1
1 ~~ /
~~ Ve1 So SS e2 G@@@2
V
~
~ w w S S
SS2SS GGG@@@
~~~www SSSS GG @@
~www SSS G @
~
~
~ww
SSSS GGG@G@@
w 1 S
S 2
SSS GG@@
~~w 2
SSSSG#
{w~~w )
C1n C2n
Since (V 2 , 2 , 2 ) is maximal, this implies that there exists a local

analytic isomorphism 1 : V 1 V 2 such that 2 = 2 1 and f1i =
f2i 1 . In the same way, we prove that there exists a local analytic
isomorphism 2 : V 2 V 1 such that 1 = 2 2 .
It follows that 2 = 1 2 2 so that 1 2 = IV 2 (the identity 32
mapping of V 2 ) on the image of V under 2 in V 2 and hence, by the
principle of analytic continuation, on all V 2 . Similarly 2 1 = IV 1 on
V 1 , so that 1 is an analytic isomorphism of V 1 onto V 2 . Since 2 =
1 1 and f1i = f2i 1 , this proves the theorem.
We can prove also the following
Theorem 2. Let V, V be complex analytic manifolds, and let ,

spread V, V in C n . Let g spread V in V . (It is not required that =
,
g. Let (V, , (V , , ) be the envelopes of holomorphy of (V, ),
)
(V , ). Then there exists a local analytic isomorphism g of V such that

the following diagram is commutative:

g
V / V

V / V
g
32 5. Envelopes of Holomorphy
Proof. Consider the following diagram:

g
V / V
>> >>
} 00
>> } }} 00
> }} 00
g ~} 00
TTTT / V OO 00
p p V TTTT OOO
pppp TTTT OOO 00
pp pp

TT TTTT OO 000
O O
wpp pp TTTOTOO
T'* n
C1n C2
V is spread in C n by g and by Theorem 1, since V is maximal,

there is a local analytic isomorphism : V C n such that g =
.

33
Also, V is a maximal continuation for the functions on V induced by g
from those on V , so that, since V is maximal for all functions on V, V
is intermediate between V and V and there is a spread g of V in V
with the properties we require.
This has the following
Corollary . Let V be a complex analytic manifold and an analytic

automorphism of V. Let V be spread in C n by , and (V, ,
)
be the
envelope of holomorphy of (V, ). Then, there exists an analytic auto-
morphism of V such that the following diagram is commutative:
/V
V

V / V

Definition 1. Let V be a complex analytic manifold, a spread of V in

,
C n . Let (V, )
be the envelope of holomorphy of (V, ).
(V, ) is called a domain of holomorphy of is an analytic isomor-
phism of V onto V.
Note that, in the definition above, (V, ) is not called a domain of
holomorphy if V and V are isomorphic, but only if is an isomorphism.
Nevertheless, Theorem 1 shows that if V is a complex analytic manifold,
33
1 , 2 , two mappings which spread V in C n , (V, 1 ) is a domain of holo-

morphy if and only if (V, 2 ) is. For, if (V 1 , 1 , 1 ), (V 2 , 2 , 2 ) are the
envelopes of (V, 1 ), (V, 2 ), then there is an analytic isomorphism of
V1 onto V 2 such that 2 = 1 and 1 is an isomorphism if and only
if 2 is. This justifies the following
Definition 2. Let V be a complex analytic manifold which can be spread 34

in C n . Then V is called a domain of holomorphy, if, for some spread
of V in C n , (V, ) is a domain of holomorphy.
Theorem 1 is, in general, false, if, instead of considering the fam-

ily of all holomorphic functions on V, we take only a subfamily. The
following is a counter-example.
Let be the complex -plane, C the complex z-plane. Spread in C
by the two mappings 1 () = and 2 () = e respectively. Let f () =
e . The maximal continuation of (, 1 , f ) is (C, f1 ) where f1 (z) = ez .
Since is the universal convering surface of C = C (0) under the
projection 2 () (which naturally has the same value at all points lying
over one point on C ) and f as a mapping C coincides with 2 , the
maximal continuation of (, 2 , f ) is (C, f2 ) where f2 (z) = z. But (since
e.g. z has zeros and ez has not) there is no isomorphism of C into itself
taking z to ez .
Chapter 6
Domains of Holomorphy:
Convexity Theory
An important problem is to find necessary and sufficient conditions un- 35

der which a manifold which can be spread in C n is a domain of holo-
morphy. The results that are presented below are due to CartanThullen
[2] (see also [1]).
S (z, r) is the open polydisc with centre z and radius r in C n , i.e., the
set of points z C n with
|zj z j | < r, j = 1, . . . , n.
Definition 1. Let V be a complex manifold spread in C n by . Let z V.

By the polydisc S (z, r) V with centre z and radius r is meant the open
set O containing z (if it exists) such that the restriction of to O is an
analytic isomorphism of onto S ((z), r) C n .
By the distance of z V to the boundary of V, d(z), is meant the
radius of the maximal polydisc S (z, r) V; the distance d(K) of a com-
pact set K to the boundary of V is inf d(z). {d(z), d(K) depend of course,
zK
on }.
Definition. Let K be a compact subset of the complex manifold V, and

let C be a family of holomorphic functions on V. The C -envelope of K,
35
36 6. Domains of Holomorphy: Convexity Theory
K C is the set of z V for which there exists a cz > 0 such that for all
f C , | f (z)| cz || f ||K (|| f ||K = sup | f (x)|, see p.2).
xK
Examples. (a) Let V = C be the complex plane, C the family of poly-

nomials. If K is a compact set C and L is the union of the relatively
36 compact components of the complement of K (relatively compact in C),
then it can be shown that K C = K L. If, on the other hand, we consider
C = C (0) and take K to be an annulus enclosing 0, C as the set of
all holomorphic functions in C , then K C = K.
(b) In V = C 2 if we take K to be the closure of the domain in the
example on p.12, viz., |z1 | 1, |z2 | ; 1 |z1 | 1, |z2 | 1, C to be
the family of all polynomials (or all holomorphic functions) in C 2 , then
K C is the polydisc |z1 | 1, |z2 | 1.
(c) If on the manifold V, C is such that f C implies f p C for
every integer p > 0, then
n o
K C = z V | f (z)| || f ||K for every f C .
It the set of points defined above is denoted K1 , clearly K1 K C . If

p 1/p
z K C , | f (z)| p cz || f ||K , | f (z)| cz || f ||K and letting p , it
follows that z K1 .
Let V be a complex manifold and let spread V in C n . Let f be a
holomorphic function on V and z V. Let U be an open neighbour-
f (z)
hood of z such that |U is an isomorphism. Then is defined to be
zi
( f 1 )((z)) 1
, being the inverse of |U.
zi
Theorem 1. Let V be a complex manifold and suppose that spreads
V in C n . Let C be a family of holomorphic functions on V, stable for
f
derivation, i.e., f C implies C . Suppose also that the canonial
zi
mapping of V to the maximal continuation of V is an isomorphism. If K
is an arbitrary compact subset of V, then the distances of K and K C to
the boundary of V are the same.
37 Proof. Since V is itself maximal for C , it is clearly sufficient to prove

37
the following:
If z K C and 0 < < d(K), then any f C can be continued to
S (z, ), i.e., that the functions f 1 in a neighbourhood of (z), can
be continued to the polydisc S ((z), ) C n .
S
Let L = S (x, ). Then L is the continuous image of the com-
xK
pact space K S (0, ){S (0, ) C n } and so L is compact. Let, for
f C , M( f ) = sup | f (z)|. It follows from the Cauchy inequalities
zL
J!M( f )
that |D J f (z)| (where J = ( j1 , . . . , jn ), D J is the operator
|J|
|J|
, |J| = j1 + + jn and J! = j1 ! . . . jn !). From the defini-
z1j1 . . . znjn
tion of K C and the fact that C is stable for derivation, it follows that if
Cz J!M( f )
z K C , |D J f (z)| . If, therefore,
|J|
X D J f (z)
g(z ) = ((z ) (z))J
JN n J!
for any z K C , the series converges normally in the polydisc S ((z),

) C n and it is clear that it continues f to S ((z), ) for every f C .
The theorem follows.
Theorem 2. Let V be a complex analytic manifold, a spread of V

in C n . Let C be a family of holomorphic functions on V having the
following properties:
(1 ) C is a closed subalgebra of HV , 1 C .
(2 ) If (z) = (z )(z , z ), then there exists a function f C such that

fz , fz ( fa is the germ fa = ( f 1 )(a) ; see IV and V).
(3 ) If K is a compact subset of V and z K C , S (z, r), the maximal 38

(open) polydisc about z in V contains points not in K C .
Then, there exists a function g C such that g cannot be continued

outside (V, ).
Proof. The main step in the proof is to construct a function g C such

that
(a) if (z) = (z ), then gz , gz and it is clearly enough that for
a countable dense set {zm } on V, gzm , gzm if zm is any point
such that (zm ) = (zm ). (The existence of a countable dense set
follows from the Poincare-Volterra theorem);
(b) for a countable dense set {zm } of points on V, let S (zm , rm ) de-
note the maximal open polydisc about zm . Then g(z) has zeros of
arbitrarily large multiplicity in every S (zm , rm ).

The proof then divides into three steps.
Step 1. The existence of the function g implies that (V, ) is the maximal
domain for g.
Proof. Let (V, ,

)
be the maximal domain of g. V consists of pairs
1
((z), (g )(z) ) = ((z), gz ). We show that is one-one. If ((z), gz ) =
((z ), gz ) then (z) = (z ), gz = gz and since, if z , z , gz , gz this
implies that z = z . therefore identifies V with an open subset of V. In
what follows, we assume this identification made, and show that V = V.

In the first place, S (zm , rm ) is the maximal polydisc about zm in V for if it
were not, S (zm , rm ) is relatively compact in V and g , the continuation of
g to V cannot have zeros of unbounded multiplicity in S (zm , rm ). Now
39 suppose that V , V. Since V is connected, there is a point b V so
that b < V, b V. If c is near enough to b it is clear that there is a
polydisc S (c, ) containing b. But if c is a zm , then there is a polydisc
S (zm , ) V, 1 V which is not the case. This completes Step 1.
Step 2. Construction of a funtion f C having property (b). Let S m =

S (zm , rm ) and consider the following sequence of polydiscs:
S 1, S 2, S 1, S 2, S 3, S 1, S 2, S 3, S 4, . . .
P
Denote its p-th term by p . Let K p be a sequence of compact sets V
o S
such that K p K p+1 and K p = V. Now, by property 3 in the
p=1
39
P
hypotheses of Theorem 2, p 1 (K p )C . Hence there is a function h C
P
and a point z(p) p so that |h(z(p) )| > ||h||K p , by example (c) on p.36,
since C is an algebra. If f p (z) = (h(z)/h(z(p) )) and is large enough, f p
P
satisfies f p C , | f p (z)| 2p on K p , f p (z(p) ) = 1 with z(p) p . Let
Q

f = (1 f p ) p . It is easily verified that this product converges in HV
p=1
and that f . 0. Since C is a closed subalgebra of HV , f C . Also
P
f has a zero of order at least p in p and since each S m = S (zm , rm )
P
occurs infinitely often in the sequence { p }, this concludes Step 2.
Step 3. Modification of the function f , such that the resulting function

has properties (a) and (b).
Let C f be the closure of the set of all function f h, h C , where f

is the function constructed in Step 2. Since C is closed, C f C and
trivially, each g C f has property (b).
Let (Xm ) be a countable dense set on V and (Ym ) the set of all points 40
having (Ym ) = (Xm ) (this set is countable by the Poincare-Volterra
theorem). Let O(m, Ym ) be the set of functions h C f such that hXm ,
hYm and (Xm ) = (Ym ).Clearly O(m, Ym ) is open in C f . We prove below
that each O(m, Ym ) is dense in C f . It then follows from Baires theorem
applied to C f (C f is a complete metrizable space since V is countable at
infinity) that O = O(m, Ym ) is dense in C f and if g O, g . 0, g has
properties (a) and (b). Thus to complete Step 3, it remains only to prove
that O(m, Ym ) is dense in C f .
Let k C f , (Ym ) = (Xm ). If kYm , kXm , k O(m, Ym ). If k <
O(m, Ym ), let h C f be so that hXm , hYm (h exists: for if f has this
property one may take h = f while if f Xm = fYm and l is such that lXm ,
lYm (hypothesis (2 )) one may take h = f l). Then, if || is small enough,
k + h defines different germs at Xm and Ym , and is in the closure of the
functions k + h, || small, which are in O(m, Ym ) and k O(m, Ym ).
This proves that O(m, Ym ) is dense in C f and thus completes Step 3 and
with Steps 1, 2 and 3, the proof of Theorem 2 is complete.
Corollary. Under the hypotheses of Theorem 2
(i) (V, ) is a domain of holomorphy.

(ii) (V, ) is the maximal continuation of (V, , C ).
Theorems 1 and 2 show that the following are equivalent:
(a) (V, ) is a domain of holomophy.
(b) If (z) = (z ), z , z there exists f HV so that fz , fz , and, if

K is a compact subset of V, d(K) = d(K HV ).
41 The last condition may be replaced by the apparently weaker condi-

tion that for any z K HV the maximal polydisc about z contains points
not in K HV . Also, it follows from Theorems 1 and 2 that if V is a domain
of holomorphy, then there exists a function g HV which separates
points in the sense that if z , z , and (z) = (z ), then gz , gz , such
that g cannot be continued outside V, i.e., if the family of all functions on
V cannot be continued simultaneously outside V, there is one function
which itself cannot be continued.
Chapter 7
Convexity Theory
(continued)
1. The maximal continuation of a family of holomorphic functions on a 42

manifold V (spread in C n ) was defined in V. Certain analogous concepts
will now be defined.
Let V be a complex manifold spread in C n by and let C be a family
of holomorphic functions on V.
An N-continuation of (V, , C ) is a continuation (V , , , C ) such
that { fi }iI is any subfamily of C , normally convergent in V and fi is the
continuation of fi , then the family { fi }iI converges normally in V . A
maximal N-continuation is now defined in the same way as was maximal
continuation.
The following two concepts are defined similarly.
A maximal U-continuation: the property considered is that of con-
vergence of sequence of functions of C , in HV .
Maximal B-continuation: The property considered is that of bound-
edness of subfamilies in HV .
In the same way as before, one can prove the existence and unique-
ness of maximal N, U and B continuations.
The proof of 6, Theorem 1 on p.36 gives us the following result.
Theorem 1 . Let V be a complex manifold, a spread of V in C n . Let
C HV and suppose that C is stable derivation. If (V, ) is itself the
41
42 7. Convexity Theory (continued)
maximal N continuation (or B or U continuation) of (V, , C ), then

the distances of K and K C to be boundary of V are the same.
43 6 Theorem 2 (p.37) also has an analogue:
Theorem 2 . Let V be a complex manifold, a spread of V in C n . Sup-

pose C has the following properties:
(1 ) f C implies f C for every complex ;
(2 ) If (x) = (y), there exists f C such that fz , fy ;
(3 ) If K V is compact, and z K C , the maximal polydisc about z

contains points not in K C .
,
Finally, let (V, C) be the maximal N, B, or U-continuation
,
of (V, , C ).

Then, is an isomorphism of V onto V.
Proof. As in Step 1, in the proof of 6 Theorem 2, is one-one (into).

We now construct a sequence of functions ( f p ) as follows.
Choose a countable dense set {zm } in V and let S m be the maximal
polydisc about zm in V. Consider the sequence.
S 1, S 2, S 1, S 2, S 3, . . .
P
and denote its p-th term by p . Let {K p } be a sequence of compact
o S

sets so that K p K p+1 , K p = V. Then, by hypothesis (3 ), there
p=1
P
is a point z(p) p , (z(p) < (K p )C ) and (by the definition of (K p )C ) a
function f C so that | f (z(p) )| > 22p || f ||K p . This gives rise to a sequence
of functions { f p } such that || f p ||K p 2p , | f p (z(p) )| > 2 p . The fact that
is onto is proved by reasoning analogous to Step 1 of 6, Theorem 2.
Examples. (a) Let C n be a (univalent) Reinhardt domain and let C

be the family of monomials zJ ( complex). It is possible to find K C
44 very simply. Let K be the set of (t1 z1 , . . . , tn zn ), where (z1 , . . . , zn )
K and |t1 | 1, . . . , |tn | 1. Clearly K K C . Also it is clear that
43
K C is completely characterised by its image in the (|z1 |, . . . , |zn |)-space,

and so in the (1 , . . . , n )-space ( j = log |z j |). As in II, let K be the
image of K , K C the image of K C in the (1 , . . . , n ) space. K has the
following propety: if (1 , . . . , n ) K , then (1 a1 , . . . , n an ) K
if a j 0, j = 1, . . . , n. Also K C is defined by inequalitics j1 1 +
+ jn n log ||zJ ||K , so that K C is clearly convex and so contains the
convex closure of K . Since the convex closure of K is the intersection
of all closed half spaces containing K , it follows easily that K C is the
convex closure of K and this gives us the C -envelope of K.
This leads to a necessary and sufficient condition for a Reinhardt do-
main O, containing 0, to be a domain of holomorphy. If O is a domain of
holomorphy, then by 2 Theorem 1, O is the maximal N-continuation do-
main of the family of all monomials. If O is the maximal N-continuation
domain of the monomials, then by Theorem 1 and 6, Theorem 2, O is
a domain of holomorphy (since the envelope of a compact set with re-
spect to the monomials is trivially larger than that with respect to all
holomorphic functions). By the above results, this is so if and only if
the image O of O in the (1 , . . . , n )-space is convex and such that if
(1 , . . . , n ) O then (1 a1 , . . . , n an ) O when a j 0. By 6,
Theorem 2, it follows that if O is a Reinhardt domain which is the union
of polydiscs, with centre 0, and O is convex, then there is a power se-
ries such that O is precisely the domain of convergence of this power
series, a result which was stated on p.14 (Converse of 2, Theorem 2).
(b) Let O be an open set in C, and let C = HO . If K is a compact 45
subst of O and L is the union of the relatively compact components of
the complement of K(inO), then K HO = K L. It is easy to see from 6,
Theorem 2, that O is a domain of holomophy. In fact, it can be proved
that K HO is compact.
3. Some remarks on domains of holomorphy.
Proposition 1. Let O be a (univalent) domain in C n . The following

three conditions are equivalent:
1) O is a domain of holomorphy;
2) If K is a compact subset of O and z K H (where H = HO ), the

maximal polydisc about z in O contains points not in K H ;
3) If K is a compact subset of O, then K H is compact.
Proof. After 6, Theorems 1 and 2, it is enough to prove that 1) implies

3). Clearly K H is closed in O, since H is an algebra and d(K) =
d(K H ). Also K H is bounded in C n : the function zi is holomorphic in
O and, by definition of K H , |zi | ||zi ||K for every z H H .
Suppose K H were not compact. There is then a sequence {X p } of
X p K H having no limit point in K H . However {X p } has a limit point
X C n since K H is bounded. Now X belongs to the boundary of O
since K H is closed in O, so that X has a distance > 0 from K. But if
O is a domain of holomorphy, this implies that X has a positive distance
from K H which is not the case, and K H is compact.
Definition. Let V be a complex manifold. V is called holomorph-convex

if the HV -envelope of every compact set is compact.
46 Proposition 2. Let V be a manifold spread in C n and suppose that V is

holomorph-convex. Then V is a domain of holomorphy.
,
Let (V, )
be the envelope of (V, ).
The proof of Step 1 in 6 Theorem 2 shows that
(1 ) V is a covering spaces of V under projection .

(2 ) Over any point of V lie only finitely many points of V: by defini-

,
tion of V, all holomorphic functions on V have the same value
If X is any point of V,
at all points over one point of V. it follows
that all points lying over X belong to the HV -envelope of X and
since is a local homeomorphism, this set cannot have a limit
point; since V is holomorph-convex, this set must be finite.
(3 ) V is holomorph-convex: this follows from (2 ).

x , y, there is a holomorphic function f on V such
(4 ) If x, y V,

that f (x) , f (y): if (x)
, (y),
this is obvious; if (x)
= (y),
then, since V is a domain of holomorphy, there exists a function g
45
such that gx , gy and, by going to a derivative of g of sufficiently

large order, the existence of f follows.
(1 ), (2 ), (3 ), and (4 ) imply that V is a domain of holomorphy

by a theorem of J-P. Serre [1, Chap. XX] which is a consequence of
Theorem A and B of Oka-Cartan-Serre on Stein manifolds and these
will be proved later.
Exercises
1. Let V be a connected complex manifold spread in C n by . Suppose 47

that there exists a n-parameter family T n of analytic automorphisms of
V, (1 , . . . , n ) where the are real numbers mod 2, such that,
if (z) = (1 (z), . . . , n (z)), then (1 , . . . , n )(z) = (ei1 1 (z), . . .
ein n (z)) for all (V, ) is then called a Reinhardt domain.
Prove the following two results:
)
(a) If (V, )
is the envelope of holomorphy of (V, ), then (V, is a
Reinhardt domain.
(b) If there exists a point z V so that (z) = 0, then every holomor-
P
phic function on V can be expanded in a series aJ (z) J on V,
which converges normally in V. Deduce that is one-one, i.e., V
is univalent.
2. Let O be an open set in C n+1 and (z1 , . . . , zn , w) be a generic point of
O. O is called a Hartogs domain if (z, w) O implies (z, ei w) O for
every real . Let now O be a connected Hartogs domain such that there
is a point (z, 0) O. Then, if f (z, w) is holomorphic in O, prove that
f (z, w) can be expanded in a series of the form
X

f (z, w) = a p (z, w)w p ,
p=0
where the a p (z, w) are holomorphic functions in O which are locallyl

a p
independent of w, i.e., = 0 in O, and such that the series converges
w
normally in O.
47
3. Let O be the following open set in C 2 : 48

3 < Rz < 0, |w| < eRz , Jz arbitrary

0 Rz < 3, e1/Rz < |w| < 1, Jz arbitrary.
(O is a Hartogs domain). Prove that the a p (z, w) of Exercise 2 are in-

dependent of w and deduce that every holomorphic function in O can
be continued to O , the union of O and the set of points (z, w) with
0 < Rz < 3, |2| e1/Rz .
Prove also that the mapping (z, w) (ei/2 z, w) spreads O, but can-
not be continued univalently to O .
Remark. Exercise 1 gives an example of a non-univalent domain whose
envelope is univalent, and Exercise 3 gives an example of a univalent
domain whose envelope is not univalent.
In the following two exercises, D will denote the closed unit disc
|z| 1 in the C-plane.
4. If f (z) is holomorphic in a neighbourhood of D,
Z2
1 ei + z
log | f (z)| log | f (ei )|R d
2 ei z
0
5. Let f p (z), (p = 1, 2, . . .) be holomorphic in neighbourhoods of D and

1 1
suppose that (i) lim log || f p ||D < +, (ii) lim log | f p (z)| M,
p p p p
1
z D. Prove that lim log || f p ||D M.
p p
6. (Hartogs main theorem). Let be an open set in C and let { f p (z)}
be a sequence of holomorphic functions in . The domain of absolute
(normal) convergence of the series
X

f p (z)w p
p=0
P

49 is defined to be the set of points (z0 , w0 ) C 2 such that f p (z)w p
p=0
49
converges absolutely (normally) in a neighbourhood of (z0 , w0 ). Let

R(z)(R(z)) be the greatest number ( 0) such that the set {z O, |w| <

R(z)(R(z)) is contained in the domain of absolute (normal) convergence.
Prove that if R(z) > 0, the R(z) = R(z)
(use Exercise 5).
7. Let {Pn (z1 , z2 )} be a sequence of homogeneous polynomials (Pn is of
degree n) in z1 , z2 . Consider the series
X

Pn (z1 , z2 ).
n=0
The domain of absolute convergence, , is defined as the set of (z(0) 1 ,

(0) (0) (0) P

z2 ) C 2 such tht in a neighbourhood of (z1 , z2 ) the series Pn (z1 , z2 )
converges absolutely.
Prove the following statements.
(a) If (z1 , z2 ) then (tz1 , tz2 ) if 0 < |t| 1.
(b) If is not empty, (0, 0) and the series converges normally

near (0, 0).
(c) The series converges normally in .
(Also due to Hartogs; for (b), use Baires theorem and the maximum
principle; for (c) use Exercise 6).
Bibliography
[1] H. Cartan: Seminaire E.N.S., 1951/52 (especially lectures VII-X). 50
[2] H. Cartan and P. Thullen: Zur Theorie der Singularitaten der

Funktionen mehrerer komplexen Varanderlichen, Regularitats -
und Konvergenzbereiche, Math. Ann., 106 (1932), 617-647.
For the exercises, reference may be made to
[3] F. Hartogs: Zur Theorie dur analytischen Funktionen mehrerer un-

abhangiger Veranderlichen, Math. Ann., 62 (1906), 1-88.
[4] L. Schwartz: Lectures on Complex Analytic Manifolds, Tata Insti-

tute of Fundamental Research, Bombay, 1955.
51
Part II
Differential Properties of the

Cube
53
Chapter 8
d-cohomology on the cube
1 Differential forms
Let O Rn be an open set. The concept of (C )-differential form is 51
r
assumed known. A differential form of degree r in O has a represen-
tation X
r
= ai1 ...ir dxi1 . . . dxir (1)
i1 <...<ir
where is the sign of exterior multiplication. The ai1 ...ir are C - func-
tions. Also we define the partial derivatives of a form (1) by
r X ai ...i
1 r
= dxi1 . . . dxir .
xi i <...<i xi
1 r
r
The differential d of the form (1) is defined by
X
n r
r
d = dxi .
i=1
xi
The operator d has the following properties:

(a) d is a local operator: if = in an open set U, d = d in U.
(b) d is linear on the forms considered as a vector space over the
complex numbers (but not as a module over C -functions).
55
56 8. d -cohomology on the cube
p q p q p q
(c) d( ) = d + (1) p d.
(d) dd = 0.
(e) d is invariant under diffeomorphisms.
52 Of course, d has the property that for functions f (forms of degree
P f
0), d f = dxi is the ordinary differential of f .
xi
This property with (b), (c), and (d) characterize d completely. The
following result, called Poincares theorem holds:
p
Let O be an open ball in Rn (or O = Rn ). Let be a form of degree
p p1
p in O, such that d = 0. Then, there exists a form of degree p 1,
such that
p1 p
d = .
This will not a proved here. See for instance [6].
2 The operators d and d

We now identify C n and R2n , and set z j = x j + ixn+ j , j = 1, . . . , n,
r
where (x1 , . . . , x2n ) are the coordinates in R2n . Then an r-form (form
of degree r) in O can be written uniquely in the form
r
X (p,q)
=
p+q=r
p,q0
where
(p,q) X
= ai1 ...i p j1 ... jq dzi1 . . . dzi p dz j1 . . . dz jq
0<i1 <...<i p n
0< j1 <...< jq n
(p,q)
is said to be of type (p, q). Its degree is, of course, p + q = r. It is
easy to verify that one has, for every form ,
X
n
X
n

d = dz j + dz j .
j=1
z j j=1 z j
3. Triviality of d -cohomology on a cube 57
53 The first sum is denoted by d , the second by d d , d are operators

of degree +1, i.e., an r-form goes into an (r + 1)-form under d , d ; more
precisely d is of type (1,0), i.e., a form of type (p, q) goes into one of
type (p + 1, q), while d is of type (0,1), taking forms of type (p, q) into
forms of type (p, q + 1).
The operators d , d have properties similar to those of d. They are
the following:
(a) d , d are the local operators.
(b) d , d are linear.

p q p q p q
(c) d ( ) = d + (1) p d and similarly for d .
(d) d d = 0, d d + d d = 0, d d = 0 .
(e) d , d are invariant under analytic isomorphisms (but not under

diffeomorphisms). A form is said to be d , (d ) closed if d =
0 (d = 0).
3 Triviality of d-cohomology on a cube

A form is said to be holomorphic if it is of type (p, 0) and hte coeffi-
cient of dzi1 . . . dzi p are holomorphic for all i1 < . . . < i p . A (p, 0) form
is holomorphic if and only if d = 0. In particular, a function f is
holomorphic if and only if d f = 0.
For forms of type (p, q) with q 1, we prove an analogue of Poin-
cares theorem, due to A. Grothendieck.
A clsoed cube in C n is a set in C n defined by inequalities
|Rz j | a j , |J z j | b j , a j, b j > 0.
Theorem 1. Triviality of d -cohomology on a cube. Let K be a closed 54

cube C n . Let be a form of type (p, q), q 1 defined in a neighbour-
hood of K and suppose that d = 0. Then there exists a neighbourhood
U of K and a form of type (p, q 1) in U such that d = in U.
We need a lemma.
Lemma . Let (z, , ) be a complex function defined for z U (U a

neighbourhood of the closed unit square with centre 0 in the z-plane),
O C l , Rm . Suppose that is differentiable in all its
variables, and is holomorhic in 1 , . . . , l ( = (1 , . . . , l )). Then
"
1 (, , )
f (z) = f (z, , ) = d d
2i z
o
is differentiable in all its variables in xO x and is a holomorphic
function in , such that
f o
= (z, , ), z .
z
1
Proof of the lemma: The integral exists since is locally summable.
z
o
Let be a closed square, centre 0, . It is sufficient to prove that
f o
= for z . Let (z) be a C -function which is 1 in and (z) = 0
z
in a neighbourhood of the boundary of . (Such a (z) exists). Now
= 1 + 2 , where 1 = , 2 = (1 ), and we have
f (z) = f1 (z) + f2 (z),
where
" "
1 1 () 1 2 ()
f1 (z) = d d, f2 (z) = d d.
2i z 2i z

f2
55 and it is obvious that f2 is holomorphic in z, = 0, if z and
z
that f2 is holomorphic in . Since 1 (z) = 0 in a neighbourhood of the
boundary of , we can define 1 (z) = 0 outside and write
"
1 1 (, , )
f1 (z, , ) = d d
2i z
"
1 1 (u + z, , )
= du du ,
2i u
3. Triviality of d -cohomology on a cube 59
if we substitute u = z (integrals without limits being over the whole

plane). From this form of the integral, it it clear that f1 (z, , ) is differ-
entiable in all the variables and holomorphic in . Also (writing 1 (u+z)
for 1 (u + z, , ))
"
f1 1 1 (u + z) 1
= du du
z 2i z u
"
1 1 (u + z) 1
= du du
2i u u
"
1 1 (u + z) 1
= lim du du
0 2i |u| u u
" 1
1 (1 (u + z) u )
= lim du du .
0 2i u
|u|
If + ( ) is the positively (negatively) oriented circle |u| = , Riemanns

formula applied to the last integral above gives
Z
f1 1 du
= lim 1 (u + z)
z 0 2i u
1
Z
1 du
= lim 1 (u + z) = 1 (z) = (z),
0 2i u
+
if z0 . This proves the lemma.

Proof of Grothendiecks Theorem: The proof will be given first for
forms of type (0, 1) to bring out the method clearly, and then it will be
given in the general case.
The proof is by induction. Consider the following statement: 56
For all forms of type (0, 1) which are d -closed and in which
the coefficients of dzk+1 , . . . , dzn are all 0, there exists an f such that
d f = .
The statement is trivially true when k = 0 for then = 0 and we
may take f = 0.
Suppose the statement true for all forms with k replaced by k 1.
Suppose that in the coefficients of dzk+1 , . . . , dzn are zero. Then the
coefficients of are holomorphic functions of zk+1 , . . . , zn (for, if =

P
k Pk a j
a j dz j , dzl (l > k) occurs as dzl dz j in d and, since
j=1 j=1
zl
a j
is d -closed = 0 for l > k). Now, by the lemma, there exists a
zl
function g(z1 , . . . , zn ), differentiable in all the variables z1 , . . . , zn and
g
holomorphic in zk+1 , . . . , zn (in a neighbourhood of K) such that =
zk
ak .
The problem is to find an f so that d f = . If we put f1 = f g,
the problem becomes that of finding f1 , so that d f1 = 1 = d g.
Clearly d 1 = 0, and by the construction of g, the cofficient of dzi
in 1 is 0 if l k. By inductive hypothesis there is an f1 such that
d f1 = 1 and the statement is true also for k. Grothendiecks theorem
for forms of type (0, 1) follows on taking k = n.
In the general case, the proof is the same.
Consider the following statement: For any d -closed form of type
(p, q) in which all terms in which dzk+1 , . . . , dzn occur are zero, there
exists a form of type (p, q 1) such that d = .
57 The statement is tirvially true if k = 0. Suppose the statement is true
when k is replaced by k 1, and let be a (p, q)-form (form of type
(p, q)) such that all the terms in which dzk+1 , . . . , dzn occur are zero.
Then, in the same way as above, it is seen that all the coefficients of
are holomorphic functions of zk+1 , . . . , zn . Suppose now that
(p,q1) (p,q)
= dzk + .
By the lemma, there exists a form of type (p, q 1) differentiable in

all the variables, such that its coefficients are holomorphic in zk+1 , . . . , zn
(p,q1)
and such that = . (One has merely to apply the lemma to the
zk
(p,q1)
coefficients of ). As above, the problem reduces to finding a form
1 such that d 1 = 1 = d . Since d 1 = 0 and the terms in
which dzk , . . . , dzn occur are zero by construction of , the existence of
1 follows from inductive hypothesis and the theorem of Grothendieck
follows on taking k = n.
4. Meromorphic functions 61
It may be remarked that the theorem of Grothendieck is true also

for open cubes and polydiscs, but the proof necessitates a limit process,
and since this can be carried out for arbitrary Stein manifolds, these
special cases are not considered here.
4 Meromorphic functions
Let V be a complex analytic manifold, and let aV. Let Oa denote the
ring of germs of holomorphic functions at a. It can be easily verified
that Oa is an integrity domain and we may therefore from the quotient
field ma ma is called the set of germs of meromorphic functions at a. 58
S
Let m = ma . A topology may be introduced on as follows. Let
aV
fa
a V and let = ma ma . Let fa and ga be defined by holomorphic
ga
functions f , g in an open connected neighbourhood of a. For every
fb
point b , mb is defined to be . If f , g are two other holomorphic
gb
f
functions in such that a = ma , then fa ga ga fa = 0 and f gg f = 0
ga
in a neighbourhood of a, and by the principle of analytic continuation,
f f
f g g f = 0 in so that b = b and the above definition is unique. A
gb gb S
neighbourhood of (a, ma ) m is now defined to be (b, mb ), where
b
has the properties mentioned above. In this topology, m is a sheaf over
V.
A meromorphic function is now simply defined to be a section of m
over V, i.e, a continuous map f : V m such that f (a) ma for every
a V.
The weak principle of analytic continuation remains valid when
holomorphic functions are replaced by meromorphic functions. Mero-
morphic functions may also be defined in terms of coverings and local
quotients of holomorphic functions, with certain obvious consistency
conditions.
Principle Parts.
A system of principal parts on V is a section of the quotient m/O

(m being the sheaf of additive groups of germs of meromorphic func-
tions, O the sheaf of additive groups of germs of holomorphic func-
tions.) But the following alternative definition is the one that will be
used in Cousins first problem.
59 A system of principal parts on the complex manifold V consists of an
open covering {i } of V and meromorphic functions fi in i ; such that
fi f j is holomorphic in i j . (The meaning of this last statement
is clear). Two systems {i , fi }, {j , f j } define the same principal parts if
fi f j is holomorphic in i j for every i, j. Here, and in what follows,
properties like the above are assumed fulfilled when the intersections in
question ar empty.
5 The first Cousin problem

The problem is the following: Suppose given a sytem of principal parts
{i , fi } on the complex manifold V. Then does there exist a meromor-
phic function f on V such that f fi is holomorphic in i , i.e., when is
the system of principal parts defined by one function?
This problem may be generalized to
The generalized first Cousin problem.
Let {i } be an open covering of the complex manifold V and suppose
given a family of functions {ci j } such that ci j is holomorhic in i j
and having the following properties:
ci j + c ji = 0 in i j , ci j + c jk + cki = 0 in i j k .
Then, is it possible to find holomorphic functions ci in i such that

ci j = ci c j in i j ?
A solution of this problem leads to a solution of the first cousin
problem, for if we take ci j = fi f j and ci j = ci c j , then fi ci = f j c j
in i j , and if we define f = fi ci in i , it is easy to see that f
solves the first Cousin problem.
60 The first Cousin Problem fo the cube.
The following therorem will now be proved.
5. The first Cousin problem 63
Theorem 2. Let K be a cube in C n and let {i , ci j } be a system such

that {i } is an open covering of K and the ci j have the properties given
above. Then there exists a neighbourhood U of K such that the general-
ized Cousin problem has a solution for the system {U i , ci j }.
Proof. We assume first that {i } is a finite covering
Step 1. There exists a neighbourhood U1 of K and a system {i } of C -

functions i in i U1 such that i j = ci j .
Let {i } be a differentiable partition of unity relative to the convering

{i } of K, i.e., i is C and has compact support contained in i , i 0
P
and i = 1 in a neighbourhood U1 of K. Such a partition of unity
exists.
Consisder the following function i on i U1 .
P
Let z i U1 ; define i (z) = j (z)ci j (z). This sum is mean-
j,1
ingful, for if ci j (z) is not defined, then z < j and so j (z) = 0, and
we define j (z)ci j (z), for such z, to be zero. It is easily seen that i
isdifferentiable in i U1 . Now
X
i j = k (cik c jk ) + j ci j i c ji .
k,i, j
Also
cik c jk = cik + ck j = c ji = ci j .
Hence X
i j = k ci j = ci j in i j U1 .
k
Step 1 is completed.
Step 2. Solution of the generalized first Coursin problem.

In i j U1 we have d 1 d j = d ci j = 0 since ci j is 61
holomorphic. Hence if we define a form (of type (0, 1)) by = d i
in i U1 we have a well-defined form on U1 . Clearly d = 0 and by
Grothendiecks theorem, there is a (0, 0) form , i.e., a function such
that d = in a neighbourhood U U1 of K. If we set ci = i in
i U, d ci = d i d = = 0 so that ci is holomorphic in i
while ci c j = i j = ci j in j j U. This commpletes the proof
of the theorem when {i } is finite.
In the general case let 1 , . . . , p be a finite covering of the cube

K, extracted from {i }. By passing to suitable intersections, we may
assume that each is contained in 1 . . . p while the functions
c1 , . . . , c p are defined everywhere in 1 , . . . , p respectively. Given ,
we define c = ci + ci on i (i = 1, . . . , p) . On i j ,
p
S
ci + ci = c j + c j , for ci c j = ci j = c j ci . Since i , c
i=1
is defined on and it is easily verified that the system { , c } solves
the generalized first Cousin problem.
Chapter 9
Holomorphic Regular
Matrices
In IX and X, we shall prove analogues of the theorems of VIII for holo- 62

morphic functions whose values are regular (invertible) matrices, and
give some applications of these generalizations. We begin by restating
Grothendiecks theorem in a form which can be carried over.
Let be an arbitrary differentiable (0, 1) form in a neighbourhood
of the cube K. The necessary and sufficient condition that there exists
a C -function f in a neighbourhood of K such that f , 0 at any point,
and
f 1 d f =
is that d = 0. (We have only to find g so that d g = and set

f = exp g).
In what follows, the functions or forms considered have values or
coefficients in the space of (m m) complex matrices or in the full linear
group GL(m, C) or regular matrices.
Our aim will be to generalize the above result to the case when
is a (0, 1) form whose coefficients are (m m) complex matrices and f
is replaced by a mapping in GL(m, C). We shall need to generalize the
lemma of VIII.
65
66 9. Holomorphic Regular Matrices
Theorem 1. Let K be a rectangle in the C-plane, L, M compact sets in

C l , C m respectively. Let (z, , ) be a matrix valued function, defined in
a neighbourhood of K L M such that it is a differentiable function of
allits variables and a holomorphic function of in this neighbourhood.
Then there exists a C -function f (z, , ) in a neighbourhood of K L
M with values in GL(M, C) which is differentiable in all its variables, is
holomorphic in and is such that
f
= f .
z
63 We need some lemmas. The first two will not be proved here.
Lemma 1. Let B be a Banach space, L , m open sets in C l , C m respec-

tively and U(, ) a continuous linear operator B B which has an
inverse, for every (, ) L m. Suppose that U(, ) is a C -function
of and and is holomorphic in . Suppose, moreover, that X(, ) is
a differentiable function of , with values in B, which is holomorphic
in .
Then U 1 (, )X(, ) has also these properties.
Lemma 2. Let O be an open set in the plane and H an open set in C h .

Let f (z, ) be a continuous function of the set of all continuous functions
of z in O with values in the space of differentiable functions of in H.
Suppose that the derivatives
k f
, k = 0, 1, 2, . . .
zk
(in the sense of distributions) all exist and have the same properties.
Then f (z, ) is an indefinitely differentiable function in O H.
This is a particular case of a theorem on the regularity in the in-
terior of solutions of elliptic partial differential equations. See, for
example, Lions [5] (also exercise 1). The present situation involves vec-
tor functions with values in the space of differentiable functions, but the
proof remains valid.
The proof of Theorem 1 will be in three parts.

67
64 First Part: Proof of Theorem 1 in the particular case when is

sufficiently near to zero. [The last phrase means the following: if
K , L , M are compact neighbourhoods of K, L, M respectively such
that K L M is contained in the domain of definition of , then
|||| < C(K , L , M ) for a suitable C(|||| will denote, in what follows
max ||i j ||K L M if = (i j ))].
i, j
The lemma that follows is the crucial step in the first part.
Lemma 3. Suppose that (z, , ) satisfies the hypothesis of Theorem

1 and that is sufficiently near zero. Then there exists a function
(z, , ) in a neighbourhood of K L M which is C in z, , and
holomorphic in is such that

+ [, ] = .
z z
([, ] stands, as usual, for ).
Proof. Let (z) be a C -function which is 1 in an open neighbourhood

of K and is zero near the boundary of K . Consider the following integral
equation (writing [, ] for [, ](, , )):
"
1 1
(z, , ) + [, ]
2i z
K
"
1 1
d d = d d
2i z
K
Consider the Banach space B of all continuous functions on K whose

values are m m matrices, with norm |||| (defined as for ) for B.
Let A(, ) denote the operator defined by
"
1
A(, )(z) = ()
2i
K
1
{(, , )() ()(, , )} d d
z
for z K . The integral equation can then be written
(I + A(, )) = X(, )
65 where "
1 1
X(, )(z) = d d.
2i z
K
Now, X(, ) and A(, ) are differentiable in , and holomorphic in

. It is clear from the definition of A(, ) that if is sufficiently near
zero, ||A(, )|| < 1 for (, ) L M . Consequently, I + A(, )

has an inverse for every (, ) L M , and so, by Lemma 1, (I +
A(, ))1 X(, ) is differentiable in , and holomorphic in , and the
integral equation has a solution (z, , ) which has the following prop-
erties:

(1) is a differentiable function of (, ) L M with values in B;
(2) is holomorphic in .

From (1) it follows that is a continuous function of K with values

in the space of all differentiable function of (, ) in L M .
1 1 ! 1 1
Now, if g(z) = , () f () d d = g f ( being
z 2i K z
g
convolution). Since = o (o is the Dirac distribution at 0; this is
z
essentially the lemma proved before Grothendiecks theorem), it follows
that

= [, ] +
z z

in O L M (in the sense of distributions). Since the terms on the right
are continuous functions of z (with values in the space of differentiable

functions in L M ), so is and so
z
" # " #
2 2
= , , +
z2 z z zz
69
66 has the same property. By iteration
k
is continuous for k 0,
zk
and, by Lemma 2, is C in a neighbourhood of K L M. This proves
Lemma 3.
Proof of Theorem 1 in the particular case. Let O be an open rectangle

C, K O and let L , m be open neighbourhoods of L, M respectively,
such that O L m is contained in the domain of definition of . We
shall find a differentiable, regular matrix f such that
f f
= f , = f , f (0) = I (unit matrix). (1)
z z
If we put f (tz) = z (t), (t) = z (t) satisfies

(0) = I

d (2)

= z(t)(tz) + z(t)(tz) = (t) A,
dt
if f satisfies (1). By the classical theorems on systems of ordinary equa-
tions of the form (2), a solution (t) of (2) exists, is C in z, , ,
holomorphic in and is unique; thus, f , if it exists, is uniquely given by
f (z) = 2 (1).
Let now be the solution of (2); define f by f (z) = z (1). Then f is

C in O L m and holomorphic as a function of . We shall show
that
(i) f (z) is a regular matrix;
(ii) f satisfies the equation (1).
Proof of (i): satisfies 67
(0) = I,
d
= A.
dt
Let be the unique solution of
(0) = I,
d
= A .
dt
d( )
Then = 0, = (0)(0) = I so that, in particular, f (z) (1) =
dt
I.
Proof of (ii): Let gz (t) = z (t)/z. We shall prove that
gz (t) = tz (t)(tz) = hz (t)
(the second equality is a definition) which implies (ii). Now, for t = 0,

gz (0) = hz (0) = 0, and gz (t) satisfies the equation
dgz
= gz {z(tz) + z(tz)} + z (tz) + z {zt (tz) + zt (tz)} (3)
dt z z

Since + [, ] = , the equation (3) remains valied if gz is
z z
replaced by hz . Since gz (0) = hz (0)(= 0), it follows from the uniqueness
of a solution of an equation of the form (3) that gz (t) hz (t) and this
completes the proof of (ii). This concludes the proof of Theorem 1 in
the particular case.
Corollary. Under the hypothesis of Theorem 1, every point of K has an
open neighbourhood U with a function f (z, , ), C in all its variables
and holomorphic in in a neighbourhood of U L M such that
f
= f in a neighbourhood of U L M.
z
68 Proof of the corollary. It is enough to prove thsi for the point 0 K.

Let
t = (tz, , ), t 0.
It is clearly sufficient to find t > 0 such that for t there is a function ft in
a neighbourhood of K L M with the required properties of regularity
ft z
so that = t ft t (by setting f (z) = ft in a neighbourhood of z = 0).
z t
71
It t is small enough, tt is near zero and by the particular case of

Theorem 1, the matrix ft exists.
Before continuing with the proof of Theorem 1, we shall deduce
from the preceding results the following theorem of H. Cartan, which is
all that will be required in the later theory.
Second Part: Theorem on holomorphic regular matrices.
Theorem 2. Let K be a rectangle in the complex plane and L, M two

compact sets in C l , C m respectively. Let H be the intersection of K with
the line Rz = 0. Let C(z, , ) be a C -function in a neighbourhood of
H L M which is holomorphic
in z and in with
values in GL(m, C).
Let K1 = K {z C Rz 0}, K2 = K {z C Rz 0}. Then there
exist functions C1 (z, , ), C2 (z, , ) in neighbourhoods of K1 L M,
K2 L M satisfying the same regularity conditions as C and such that,
in a neighbourhood of H L M,
C = C1C21 .
Proof. The proof will be given first in the case when C differs little from
the indentity matrix I in a sense which is obvious. Let H be a rectangle
with sides parallel to the coordinate axes in the plane containing H such 69
that H L M is contained in the domain of definition of C. Then logC
is defined (as exp1 (C)) and is near zero if C is near I in H L M.
Let be a C -function in a neighbourhood of K such that (z) = 1 if
Rz , = 0 if Rz ( so chosen that the intersection of K with the
strip |Rz | is contained in H ).
Now define
21 = exp[ log C]
and 1 = C2 in a neighbourhood of H L M. 2 is extended to a
neighbourhood of K2 L M by setting 21 = I for Rz , and 1
to a neighbourhood of K1 L M by setting 1 = I for Rz . Then
we have
C = 1 21 .
1 2
Also, if C is near I, 1 , 2 are near I, while , are near 0. Since
z z
C is holomorphic, we have
2 1
C = ,
z z
and
1 2
11 = 21 = .
z z
Since is near 0 if C is near, I, there exists a C -function f (z, , )
holomorphic in in a neighbourhood of K L M such that
f
f 1 = .
z
If
C1 f 1 1 1
C1 = 1 f 1 , C2 = 2 f 1 , = 1 f 1 f + f
z z z
= 1 f 1 + 1 f 1 = 0,
so that C1 and similarly C2 are holomorphic in z and .

70 Clearly,
C = C1C21
in a neighbourhood of H L M.
To prove Theorem 2 in the general case, we proceed as follows. Let
C be any holomorphic regular matrix in a neighbourhood of H L M.
Then there exists a matrix C holomorphic in z and in a neighbourhood
of K L M (and even in one of C L M, C being the complex
plane) which approximates to C (one has only to approximate the entries
of C), and so, there is a C so that C 1 C is near I. By the particular
case proved above, there exist holomorphic regular matrices C1 , C2 in
neighbourhoods of K1 L M, K2 L M respectively so that
C 1C = C1 C21
and so, we have

C = C1C21 ,
where C1 = C C1 . This concludes the proof of Cartans theorem on
holomorphic regular matrices.
73
Third Part: Proof of Theorem 1 in the general case.

After the corollary to the particular case of Theorem 1, we can divide
K into a finite number of closed rectangles Ki with sides parallel to
the axes of coordinates and obtain functions fi in neighbourhoods of
fi
Ki so that = fi and fi has the required regularity properties. We
z
require a certain function f in the whole of K. It is easy to see that
it is enough to solve the following problem: given two functions f1 ,
f2 in neighbourhoods of two adjacent rectangles K1 , K2 such that in
neighbourhoods of K1 L M, K2 L M, respectively,
f1 f2
= f1 , = f2
z z
f
find a function f in a neighbourhood of (K1 K2 ) L M with = 71
z
f1 f2
f . Now, since f11 = f21 , the function c = f1 f21 is C in a
z z
neighbourhood of H L M (H being the common side of (K1 and K2 )
and holomorphic in z and .) Consequently, by Theorem 2, there exist
matrices c1 , c2 in neighbourhoods of K1 L M, K2 L M holomorphic
in z and , so that c = c1 c1
2 in a neighbourhood of H L M. Then
c = c1 c2 = f1 f2 and c1 f1 = c2 f21 in a neighbourhood of H L M.
1 1 1
If we define f = c1 f11 in a neighbourhood of K1 L M, = c2 f21 in a

f f1
neighbourhood of K2 L M, then = c1
1 = c1
1 f1 = f if z
z z
belongs to a neighbourhood of K1 , and the same equation holds also in
a neighbourhood of K2 L M. This completes the proof of Theorem
1 in the general case.
Chapter 10
Complementary Results
1 Generalization of Grothendiecks theorem

Let be an open set in C n , a (0, 1) form in whose coefficients are 72
m m (differentiable) matrices. We ask for a condition that there exist a
differentiable mapping f of in GL(m, C) such that
f 1 d f = .
P 2 f f ak
If = ak dzk and f 1 d f = , then = ak + f
zl zk zl zl
ak 2 f al
= f al ak + f= = f ak al + f ,
zl zk zl zk
al ak
i.e., + [ak , al ] = 0,
zk zl
P
and if we write [, ] = [ak , al ]dzk dzl , then we can write these
k<l
equations as
d + [, ] = 0.
The following generalization of Grothendiecks theorem provides a con-
verse in the case of a cube.
75
76 10. Complementary Results
Theorem 1. Let K be a cube in C n , a(0, 1) differentiable form in a

neighbourhood of K. Suppose that
d + [, ] = 0.
Then there exists a differentiable, regular matrix f in a neighbourhood

U of K such that, in U,
f 1 d f = .
Proof. We use induction as in Grothendiecks theorem. Consider the
statement:
73 For every form of type (0, 1) such that d + [, ] = 0 and for
which the coefficients of dzk+1 , . . . , dzn are zero, there exists f with
values in GL(m, C) so that f 1 d f = . For k = 0, the statement
is trivially true, since = 0 and we may take f = I. Suppose the
Pk
statement true when k is replaced by k 1. Let = a j dz j . Since
j=1
d + [, ] = 0 and [, ] does not contain dzl , l > k, the a j are holo-
morphic in zk+1 , . . . , zn . By 9, Theorem 1 there is a function g, holo-
g
morphic in zk+1 , . . . , zn such that = g ak . If we set f = f g
zk
then f 1 d f = g1 f 1 d f g + g1 d g and the problem reduces to the
finding of an f such that
f 1 d f = g( g1 d g)g1 =
say. It is easily verified that d + [, ] = 0 and clearlyl the coefficients

of dzl , l k, in are zero, and by inductive hypothesis f exists.
2 Linear bundles
Let V be a topological space, {Oi }iJ an open covering of V. Suppose
that in every Oi O j is defined a continuous function ci j with values in
GL(m, C) such that the set {ci j } satisfies
(i) ci j c ji = I in Oi O j ,
(ii) ci j c jk cki = I in Oi O j Ok .
2. Linear bundles 77
S
Consider the set (Oi C m ). Let (x, y) Oi C m , (x , y ) O j C m .
iI
We identify (x, y) and (x , y ) if x = x in V and y = ci j (x)y. The quotient
is denoted by E.
Definition . E together with the given system {Oi , ci j } is called a linear

bundle over V (with fibre C m ). V is said to be the base of E.
E is a topological space with the following properties: 74
(a) There is a canonical mapping p : E V which is continuous and

onto V.
(b) p1 (a) C m for every a V (topologically and as a vector space

over C).
(c) Every point a V has a neighbourhood O such that p1 (O)

O C m the isomorphism being topological and compatible with
(a) and (b) in an obvious sense.
Suppose E is another linear boundle over V, defined by the system

{O , c }A . Suppose p is the p corresponding to E and that there is a
homeomorphism of E onto E compatible with (a), (b) and (c). Then it
is easily shown that {Oi , ci j }I , {O , c }A are related by a finite number
of applications of the following two operations:
(1 ) Passage to refinements or the converse. {O , c }A is a refinement

of {Oi , ci j } if there is a mapping : A J such that O O()
and c = c()() .
(2 ) The covering {Oi }iJ being the same, one passes to new functions
ci j by defining ci j = ci ci j c1
j where ci is continuous in Oi .
(Moreover, it can be shown easily that if, in defining E and E the

same covering is used, then only one application, namely of operation
(2 ), is necessary). If E and E are related by such a homeomorphism,
we say that they are in the same clases.
The trivial class is defined as the class containing the bundle defined
by taking for the covering, just V. With respect to a covering {Oi }iJ this
class can be defined by taking for the functions ci j the unit matrix I or, 75
generally, any functions ci c1 j where ci is continuous in Oi . A bundle is

trivial if it is in the trivial class.
If V were a differentiable (complex analytic) manifold, then we de-
fine a differentiable (analytic) linear bundle over V in the same way, but
now requiring the ci j to be differentiable (analytic).
The differentiable or analytic class of a differentiable or analytic
bundle can be defined in the obvious way and we speak of differentiable
or analytic equivalence and triviality.
The following important theorem holds.
Theorem 2. Let K be a cube in C n , E an analytic bundle over a neigh-

bourhood of K. Then E is trivial over a neighbourhood of K.
Proof. By dividing K into smaller (closed) cubes Ki with faces parallel

to the coordinate hyperplanes, we obtain holomorphic regular matrices
ci j in neighbourhoods of Ki K j respectivelyl. Also, if the bundle is
defined by {Oi , ci j } and is trivial over O1 and O2 , we may replace O1 , O2
by their union and modify the ci j to obtain an equivalent bundle (if c12 =
c1 c1
2 set O1 = O1 O2 , Oi = Oi if i , 1, 2, ci = ci j if i, j , 1,
c1 j = c1 1
1 c1 j (= c2 c2 j ). Then {Oi , ci j } defines an equivalent bundle).
We have thus only to prove the following result: given two adjacent
cubes K1 , K2 and a holomorphic regular matrix c in a neighbourhood
of their common face, we can write c = c1 c1 2 in a neighbourhood of
K1 K2 , c1 , c2 being holomorphic regular matrices in neighbourhoods
of K1 , K2 respectively. But this follows at once from Cartans theorem
on holomorphic regular matrices (9, Theorem 2).
3 Application to the second Cousin problem
76 Divisor: A divisor can be defined in two ways similar to the two defini-
tions of meromorphic functions and of principal parts.
(a) Let m be the sheaf of multiplicative groups of germs of mero-

morphic functions . 0 on the complex manifold V, H the sheaf
of multiplicative groups of germs of invertible holomorphic func-
tions. A divisor is a section of m /H over V.
3. Application to the second Cousin problem 79
(b) Let {Oi } be an open covering of V and let gi be a meromorphic

function in Oi . The system {Oi , gi } defines a divisor if the func-
tion gi g1
j and its reciprocal are holomorphic in Oi O j . Two
systems {Oi , gi } and {O j , gj } define the same divisor if gi g1
j is
holomorphic and invertible in Oi O j .
The second Cousin problem is as follows:

Given a divisor {Oi , gi } on V, does there exist a meromorphic func-
tion f on V such that f = i gi in Oi , where i and i1 are holomorphic
in Oi , i.e., does there exist one meromorphic function f on V which
defines the same divisor as {Oi , gi }? As in the case of the first Cousin
problem, this problem can be generalized.
Let {Oi } be an open covering of V and let there be given a holomor-
phic invertible function i j in Oi O j such that
i j ji = 1 in Oi O j ,
i j jk ki = 1 in Oi O j Ok .
Then, does there exist a holomorphic invertible function i in Oi such

that
i 1
j = i j in Oi O j ?
It is easily seen that a solution of this problem leads to a solution 77

of the second Cousin problem: for if we define i j = gi g1 j in Oi O j
1
({Oi , gi } is the second Cousin datum) and if i j = i j (also in Oi O j ),
then i1 gi = 1 j g j in Oi O j , and the meromorphic function f defined
1
by f = i gi in Oi solves the second Cousin problem.
Theorem 3. The generalized second Cousin problem is always solvable

for a neighbourhood of a cube.
Proof. Given the system {Oi , i j } in a neighbourhood of the cube K, the

system defines an analytic line bundle over a neighbourhood of K (linear
boundle with fibre C). The solubility of the second Cousin problem is
precisely the triviality of this line bundle over a neighbourhood of K,
and this has been proved in Theorem 2.
Eexercise
1. Let (z) be a C -function with compact support in the plane. Let 78

"
1 z z
f (z) = () d d.
2i z
2 f
Prove that = (z). If is only a distribution with compact
z2
support, prove that this equation holds in the sense of distribu-
2 f
tions. Deduce that if f is a distribution such that 2 is a contin-
z
f f
uous functions, then so are f , , .
z z
2. Let V be a complex analytic manifold, and suppose that the gen-
eralized first Cousin problem is always solvable on V. Prove that
the generalized second Cousin problem is solvable, if it is differ-
entiably solvable (in an obvious sense).
Prove also that on the Riemann sphere, the first Cousin problem
is always solvable, while the second is not.
3. Let K be a compact set in C. Prove that an analytical bundle

differentiably trivial over a neighbourhood of K is analytically
trivial over a neighbourhood of K.
4. Let V be a complex analytic manifold and let {Oi , ci j } define an

analytic bundle over V which is differentiably trivial. Let i be a
C -function in Oi with ci j = i 1 1
j in Oi O j . Let i = i d i .
81
Show that the i define a form of type (0, 1). What relation does
satisfy?
Given a form of type (0, 1) with d + [, ] = 0, show that it
defines a class of analytic bundles over V which is differentiably
79 trivial. When do two such forms and define the same (ana-
lytic) class of analytic bundles?
5. Generalize the results of exercise 4 to the case of a nontrivial class

of differentiable bundles on V.
As an application, prove that if V is a Riemann surface and E
and an arbitrary differentiable bundle on V, there always exists an
analytic bundle which is differentiably equivalent with E (use a
device similar to that used in Step 1 in 8, Theorem 2.).
Bibliography
[1] H. Cartan: Sur les matrices holomorphes de n variables complexes, 80

Journal de Mathematiques, 9e -serie, 19 (1940), 1-26.
[2] H. Cartan: Seminaire E.N.S. 1951/52 (especially lecture XVII by

J. Frenkel).
[3] H. Cartan: Seminaire E.N.S. 1953/54 (especially lecture XVIII by

J-P.Serre).
[4] J. L. Koszul and B. Malgrange: Sur certaines structures fibres com-

plexes (to appear in Archiv f -Mathematik).
[5] J. L. Lions: Lectures on Elliptic partial differential equations, Tata

Institute of Fundamental Research, Bombay, 1957.
[6] G. de Rham: Varietes differentiables, Hermann, Paris, 1955.
83
Part III
Coherent Analytic Sheaves
85
Chapter 11
Sheaves
Definition. Let X and F be topological spaces and a mapping F X 81

such that
(i) is onto;
(ii) is a local homeomorphism.
Then we call F (with the mapping ) a sheaf on X. will be called the

projection of F on X.
A section of the sheaf F (over X) is a continuous mapping s : X
F such that s = identity.
If W is any subset of X, 1 (W) is a sheaf on W in a natural way
A section of the sheaf F over a subset W of X is a section of the
sheaf 1 (W) over W.
We shall sometimes say section for the image of the section s in
F.
If x X, Fx will stand for 1 (x).
Proposition 1. A section over an open set U X is an open set in F .
Proof. Let s : U F be a section over U. Let a s(U) and let O be

an open set in s(U) such that O = O s(U) where O is open in F and
restricted to O is a homeomorphism of O onto an open subset of X.
Since s is continuous, s1 (O) is open in X, i.e., (O) is open in X and
87
88 11. Sheaves
so in (O ). Since is a homeormphism in O , O is an open set in O

and so in F .
82 This implies, in particular, that two sections which coincide at a
point, coincide in a neighbourhood of this point.
Proposition 2. If F is a Hausdorff space, a section over a closed set is

closed.
Proof. Let W be a closed set in X, and let a s(W) (s is the given

section). Let x = (a). Let a be an open neighbourhood of a, homeo-
morphic with its projection. a belongs to the closure of a s(W) in a
and so (a) belongs to the closure of (a s(W)) s(W) = W so
that, since W is closed, (a) W. Suppose next that s(x) = b , a and
let a , b be disjoint neighbourhoods of a, b, homeomorphic with their
projections. If y is near enough to x, then s(y) b and consequently
a does not meet s(W), a contradiction.
Proposition 2 is not true if F is not a Hausdorff space.
Examples of sheaves.
1 ) If V, W are manifolds and W is spread on V by a mapping onto

V, W is a sheaf on V.
2 ) X being a topological space, Y an arbitrary set, the set of all map-

pings X Y give rise to a sheaf, the sheaf of germs of the map-
pings X Y in the following way: two mappings of a neighbour-
hood of x X are identified if they coincide in a neighbourhood
S
of x. The set of equivalence classes at x is Fx and F = Fx .
xX
The topology on F is obtained by a method similar to that used
in the case of the sheaf of germs of holomorphic function in IV.
3 ) In the same way, X, Y, being topological spaces, one defines the

sheaf of germs of continuous mappings of X in Y.
83 4 ) If V, W are complex analytic manifolds, we can define the sheaf of

germs of analytic mappings of V in W. This sheaf is a Hausdorff
space, by reasoning similar to that used in IV when W = C.
89
Diagonal product of two sheaves.

Let (F , ), (G , ) be sheaves on X. The set of ( f, g) F G such
that ( f ) = (g) can be topologized in a natural way. This set then
forms a sheaf on X called the diagonal product FV G of the sheaves F
and G .
Sheaf of groups. Let F be a sheaf on X. F is a sheaf of groups if
1 ) for ever x X, Fx is a group;
2 ) the mapping a a1 of F (a, a1 Fx ) in F is continuous.
3 ) the mapping ( f, g) f g of FV F to F is continuous.

Example . The sheaf of germs of continuous mappings of X in a topo-
logical group is a sheaf of groups.
Sheaf of rings.
Let F be a sheaf on X. F is a sheaf of rings if
1 ) for every x X, Fx is a ring;
2 ) F is a sheaf of (additive, abelian) groups;
3 ) the mapping ( f, g) f g of FV F F is continuous.

Let a be a sheaf of rings and m a sheaf on the same space X m is
called a sheaf of a modules if
1 ) m is a sheaf of abelian groups (additive);
2 ) for every x X, m x is an ax -module; 84
3 ) the mapping (, m) m of aV m m is continuous (it being

assumed that mx is a left ax module).
The following is an important example.
Let {O, ci j } cefine a linear bundle E on the space V (the ci j are con-
tinuous mappings of Oi O j into GL(m, C)). Let p be the projection of
E onto V. A cross-section of the bundle is a continuous map s : V E
such that p s is the identity.
90 11. Sheaves
We can now define the sheaf of germs of sections of the bundle in

the usual way. This sheaf is a sheaf of C -modules, where C is the sheaf
of germs of continuous, complex valued functions in V.
It is of importance to decide when a sheaf of modules over C can be
obtained from a bundle by the method given above.
Suppose that the sheaf F can be so obtained. Then, since every
a V has a neighbourhood U such that p1 (U) U C m (as linear
bundles) locally, the contraction of F to U is isomorphic (as a sheaf) to
CUm where CU is the sheaf of germs of continuous functions in U.
Suppose, conversely, that the sheaf (F , ) has the above property.
Let {Oi } be a covering of V such that 1 (Oi ) COmi . Then Oi O j ,
being an open subset of both Oi and O j , gives rise to an automorphism
of COmi O j as a sheaf of modules over OOi O j . Let e p COmi O j be
the element defined by (0, . . . , 0, 1, 0, . . . , 0) where all but the p-th place
P
contain 0. Then (e p ) = a pq eq , (since is an automorphism of the
85 sheaf COmi O j as a sheaf of modules over COi O j ) and we can define the
matrix ci j = (a pq ). Since is an automorphism, ci j is invertible. It is
easy to see that the bundle defined by {Oi , ci j } gives rise to the sheaf F .
This leads to a one-one correspondence between classes of linear
bundles and sheaves locally isomorphic with C m .
Also there is a one-one correspondence between cross-sections of a
bundle and sections of the sheaf defined by the bundle.
Chapter 12
General properties of
Coherent Analytic Sheaves
1 Analytic Sheaves
Let F be a sheaf on the base space X. The concept of a subsheaf is 86
defined in the obvious way (as a subset of F which is made into a sheaf
by the restriction of the projection to the subset). It is clear that if H ,
G are subsheaves of F , so is H G .
Let now F , G be two sheaves of groups on X with projections f ,
g . Let be a mapping F G . is called a (sheaf ) homomorphism if
(i) is continuous;
(ii) f = g ;
(iii) the restriction of to Fx (= 1

f (x), x X) is a homomorphism of
1
the group Fx in Gx (= g (x)).
There are corresponding definitions of subsheaves of sheaves of al-

gebraic structures and of homomorphisms between such sheaves. The
concepts of a one-one mapping (into), mapping onto, image of a homo-
morphism, kernel of a homomorphism are defined in the obvious way.
91
92 12. General properties of Coherent Analytic Sheaves
A sequence
dp d p+1
. . . F p F p+1 F p+2 . . .
of sheaves F p of groups (or other algebraic structures) and homomor-

phisms d p : F p F p+1 is said to be exact at F p if the kernel of
d p+1 = image of d p ; it is exact if it is exact at F p for all p.
Quotient Sheaves.
Let G be a sheaf of groups on the topological space X, F a subsheaf
87 of G such that for every x X, Gx is a normal subgroup of Gx . Then
there is precisely one sheaf H on X such that H x = Gx /Fx and the
mapping : G H ( x : Gx H x is the natural homomorphism)
S
is a sheaf homomorphism. We have only to set H = (Gx /Fx ) and
xX
put quotient topology on H . It is clear the the conditions on H above
determine uniquely the topology on it.
Analytic Sheaves.
Let V be a complex analytic manifold, and O = O V the sheaf of
germs of holomorphic functions on V.
Definition. An analytic sheaf on V is a sheaf of O-modules.

One can then define (analytic) subsheaves of an analytic sheaf F in
the obvious way. Clearly, the intersection of analytic subsheaves of F
is analytic.
Notations. In what follows, O will denote the ring of holomorphic fun-

tions on the complex manifold V, O the sheaf of germs of holomorphic
functions on V; O m , O m will stand for the m-th powe of O, O respec-
tively.
If F is a sheaf on the space X, and U is a subset of X, F U will
denote the restriction of F to U. FU or (U, F ) will stand for the
sections of F over U.
Examples. 1 ) O m is an analytic sheaf;

2. Coherent analytic subsheaves of O m 93
2 ) Let m be a finitely generated submodule of O m (generated by

h1 , . . . , h p say). For a V, let ma be the submodule of Oam gener-
S
ated over Oa by (h1 )a , . . . , (h p )a . Then m = ma is an analytic
aV
sheaf on V. To prove this, one has only to show that if f is a sec-
tion of O m over an open neighbourhood U of a V, then there is
a neighbourhood U1 U of a such that fa ma implies fb mb
for b U1 .
p
P
Now, fa = (i )a (hi )a , (i )a Oa , and there is an open neigh- 88
i=1
p
P
bourhood U1 of a, and functions f , i in U1 such that f = i hi
i=1
p
P
in U1 , which implies that ( f )b = (i )b (hi )b mb for b U1 .
i=1
3 ) The sheaf of relations between q elements of O m . Let h1 , . . . , hq

q
O m . Let Ra be the submodule of Oa consisting of the q-tuples
((c1 )a , . . . , (cq )a )(ci )a Oa such that (c1 )a (h1 )a ++(cq )a (hq )a = 0.
S
Then it is easily verified that R = R(h1 , . . . , hq ) = Ra is an
aV
analytic sheaf, called the sheaf of relations between h1 , . . . , hq .
2 Coherent analytic subsheaves of O m

Definition . Let F be an analytic subsheaf of O m . F is said to be co-
herent if the following is true: for every a V there is a neighbourhood
U of a, and a finite number of sections of F over U, f1 , . . . , fq , such that
for every b U, Fb is Ob -generated by ( f1 )b , . . . , ( fq )b .
The following important theorem holds, but we shall not prove it

here. The theorem is due to K. Oka 7. For the proof see Cartan [2], [3]
Lecture XV.
Theorem of Oka. Let h1 , . . . , hq O m . Then, the sheaf of relations
R(h1 , . . . , hq ) is a coherent analytic sheaf.
Corollary. Let F , G be coherent analytic subsheaves of O p . Then F

G is a coherent analytic sheaf.
Proof of the Corollary. Let U be a neighbourhood of a V, f1 , . . . , fk ;

g1 , . . . , gm sections of F , G over U such that ( f1 )b , . . . , ( fk )b (resp.
(g1 )b , . . . , (gm )b ) Ob -generate Fb (resp. Gb ) for every b U. Consider
the sheaf R( f1 , . . . , fk , g1 , . . . , gm ) = R on U.
89 It is coherent, by Okas theorem. We define a map of R onto F U
G U , as follows: For x U, let r = ((c1 ) x , . . . , (ck ) x , (c1 ) x , . . . , (cm ) x )
P P
R. i.e., (ci )x ( fi )x + (cj ) x (g j ) x = 0. The image (r) of this point is,
P
by definition, (ci )x ( fi ) x Fx . By the above relation, (r) G x and
so (r) Fx Gx . is clearly a homomorphism of R in F U G U .
P P
Also, if fx = (ci )x ( fi )x = (c j )x (g j ) Fx Gx , then (r) = f x where
r = ((c1 ) x , . . . , (ck )x , (c1 ) x , . . . , (cm )x ) and so is a homomorphism
of R onto F U G U . It follows easily that F U G U is coherent and
the result follows.
We give some examples of non-coherent, analytic sheaves.
(a) Let F be a sheaf of ideals on V, i.e., Fa is an ideal of Oa (so an
Oa -module). Suppose that F is coherent and let, in a neighbour-
hood U of a, f1 , . . . , f p generate Fb over Ob . Then, the necessary
and sufficient condition that Fb = Ob is that at least one fi (b) , 0.
Hence the set of b with Fb , Ob is precisely the set of common
zeros of f1 , . . . , f p . Hence the set of b wiht Fb , Ob is an analytic
subset of V [i.e., locally in V, it is the set of common zeros of a
finite number of holomorphic functions].
The complement of an open ball, S , in C n is not an analytic subset.
Hence if we get Fa = Oa for a S , Fa = 0 for a < S , the analytic
S
sheaf F = Fa is not coherent.
aC n
(b) Let , V be an open subset of the complex manifold V. Let

S
Fa = Oa , a < ; Fa = 0, a < . The sheaf F = Fa is
aV
analytic but not coherent (the definition is violated at a point on
the boundary of ).
Let F be a coherent analytic sheaf on a (connected) complex man-
ifold V. Then Fa , 0 at any point a V unless Fa = 0 for all a V: if
90 Fa = 0 and ( f1 )b , . . . , ( f p )b Ob -generate Fb for b U where U is a con-
nected neighbourhood of a, then f1 , . . . , f p are zero in a neighbourhood
3. General coherent analytic sheaves on a... 95
of a since Fa = 0 and Fb = 0 for b U by the principle of analytic

continuation. Hence the set of a with Fa = 0 is open. It is easily proved
in the same way that this set is closed and the result follows.
3 General coherent analytic sheaves on a complex

analytic manifold
Definition . Let V be a complex analytic manifold and F an analytic
sheaf on V. F is said to be coherent if every a V has an open neigh-

bourhood such that F O p N , where N is a coherent subsheaf
of O p (in the first sense).
For subsheaves of O m , the two definitions of coherence coincide.

It is clear that given a subsheaf of O m , which, locally, is an arbitrary
quotient of an O p , there is a natural homomorphism of O p onto F (
being an open neighbourhood of a given point a V) and F is coherent
in the first definition. Converselyl, suppose that F O m and that to
every a V, there is a neighbourhood such that in , p elements
p
g1 , . . . , g p of O generate F . We have a homomorphism
p
((1 )b , . . . , ( p )b ){ Ob } (1 )b (g1 )b + + ( p )b (g p )b
p
of O onto F , of which the kernel is the sheaf of relations R (g1 , . . . ,
g p ) which is coherent by the theorem of Oka. Hence the condition of
the second definition is fulfilled.
An example of a coherent sheaf which is not a subsheaf of some O m
is the sheaf of germs of sections of an analytic, non-trivial, bundle. This
sheaf is locally isomorphic to an O m .
Proposition 1. Let F be a coherent analytic sheaf. Let f1 , . . . , fq be a 91

finite number of sections of F . Then the sheaf R( f1 , . . . , fq ) of relations
between f1 , . . . , fq is coherent.
Proof. Let a V. There are p sections g1 , . . . , g p over a neighbourhood

of a such that
(i) (g1 )b , . . . , (g p )b Ob -generate Fb for b .

p
(ii) R (g1 , . . . , g p ) is a coherent subsheaf of O .
[This is just a reformulation of the definition of coherence]. We have

Pp
j j
( fi )a = (i )a (g j )a ((i )a Oa ) so that there is an open neighbourhood
j=1
P j
of a such that fi = i g j in . Now suppose b and
q
P
that ((1 )b , . . . , (q )b ) Rb ( f1 , . . . , fq ). Then, we have (i )b ( fi )b = 0,
i=1
P j P P
so that (i )b (i )b (g j )b = 0. This implies that ( (i )b (1i )b , . . . , (i )b
i, j i i
q
Rb (g1 , . . . , g p ). Since by hypothesis R (g1 , . . . , g p ) is coher-
(i )b )
ent, there exist functions hkj , j = 1, . . . p, k = 1, . . . , r and a neighbour-
hood of a such that
X X
r
j j
(i )b (i )b = (k )b (hk )b
i k=1
for b , where (k )b Ob . This implies that

j j
(1 , . . . , q , 1 , . . . , r ) R (i , hk ).
By Okas theorem R (ij , hkj ) is coherent and the systems (1 , . . . , q )

q
R ( f1 , . . . , fq ) form a quotient of R (ij , hkj ) and, this being O ,
R ( f1 , . . . , fq ) is coherent. This proves Proposition 1.
92 Theorem 1. Let G and F be two coherent analytic sheaves. Let :

G F be a homomorphism (as analytic sheaves) of G in F . Then
the kernel, the image, the cokernel and the coimage of are coherent
analytic sheaves.
(The cokernel is F /(G ), the coimage is G /1 (O).)
Proof. 1) The image of . Let g1 , . . . , gq be sections of G over an

open set such that (g1 )a , . . . , (gq )a Oa -generate Ga for a .
Then (g1 ), . . . , (gq ) are sections of F over and they Oa -
generate (G ) for a . Hence on , (G ) is isomorphic with
q
O /N where N = R ((g1 ), . . . , (gq )). The result follows
from Proposition 1 and the definition.
4. Coherent analytic sheaves on subsets of a... 97
2) The kernel of . Let g1 , . . . , gq G generate G . Now, the

sheaf R ((g1 ), . . . , (gq )) is coherent. We define a mapping of
R to the kernel of as follows: if ((c1 )a , . . . , (cq )a ) Ra , map
this point on (c1 )a (g1 )a + + (cq )a (gq )a . This gives us a homo-
morphism of R onto the kernel of (restricted to ) and by 1)
the kernel is coherent.
3) and 4)The cokernel and the coimage of . After 1) and 2) it is

clearly sufficient to prove the following statment:
If G is a coherent analytic sheaf, F a coherent analytic subsheaf
of G , then G /F is coherent. To every a V corresponds an
open set , a with G O p /N where N is coherent.
Let f1 , . . . , fq F generate F . Since F G , there
p
are elements f1 , . . . , fq O which go into f1 , . . . , fq . Let R
be the analytic sheaf on generated by f1 , . . . , f p ; this sheaf is
clearly coherent. One has F (R + N )/N , and consequently 93
q
G /F O /(R + N ). Since obviously R + N is coherent,
the result follows.

4 Coherent analytic sheaves on subsets of a complex

analytic manifold
Let X be a subset of the complex manifold V. An analytic sheaf on X
is defined to be a sheaf of O X -modules (O X is, of course, the restriction
of O to X). The definition of a coherent analytic sheaf F on X is the
p
same as before: if F O X , then F is coherent if to every a X
exist a neighbourhood U in X and elements f1 , . . . , f p FU such that
( f1 )b , . . . , ( f p )b OXb -generate Fb for b U. An analytic sheaf F on X
is coherent, if every a X has an open neighbourhood U X such that
F U O Up /N where N is a coherent analytic subsheaf of O Up . The
result of 3 generalize to the sheaves OXp . The following theorem will be
proved.
Theorem 2. Let X be a compact subset of the complex manifold V, and

F a coherent analytic sheaf on X. Then there is an open set X
(open in V) and a coherent analytic sheaf G on such that G X F .
We begin the proof with a remark, which follows at once from the
fact that a section of sheaf is an open mapping and the definition of
coherence.
Remark . Let F , G be two coherent analytic sheaves on a subset Y of

the complex manifold V, f , g two homomorphisms: F G . Then,
the set of y Y with fy = gy ( fy , gy are the homomorphisms Fy Gy
determined by f , g respectively) is open in Y.
For the proof of Theorem 2, we require two lemmas.
94 Lemma 1. Let F , G be coherent analytic sheaves on a subset Y V.

Let a Y and suppose that there is a homomorphism a : Fa Ga .
Then there is a neighbourhood of a in Y such that a can be continued
to a homomorphism : F G (in an obvious sense).
Proof. Suppose that in a neighbourhood of a, f1 , . . . , fq F , gen-

erate F ; let g1 , . . . , g p G define respectively the germs (g1 )a =
a (( f1 )a ), . . . , (g p )a = a (( f p )a ). The sheaf of relations between f1 , . . . ,
f p is coherent, and so, if is small enough, is generated in by
P P
functions (ki ) and (ki )a ( fi )a = 0 so that we have (ki )a (gi )a = 0.
i i
P P
Let be such that ki fi = 0, ki gi = 0 in . Let b
i
P Pi P
and let (i )b ( fi )b = 0. Then (i )b = (ak )b (ki )b and (i )b (gi )b =
P Pi k P i
(ak )b (ki )b (gi )b = 0. Hence, if b , (i )b ( fi )b = 0 implies
Pk i i
(i )b (gi )b = 0 and the homomorphism on can be defined by
i
b (( fi )b ) = (gi )b for b .
Lemma 2. Let X be a compact set in V, F , G , coherent analytic

sheaves on a neighbourhood of X. Let be a homomorphism F X
GX . Then can be continued to a homomorphism F U G U , U being
a suitable neighbourhood of X.
4. Coherent analytic sheaves on subsets of a... 99
Proof. By Lemma 1, for every a X, a can be extended to a homo-

morphism a : F a G a in a neighbourhood a of a. Now and
a determine the same homomorphism a of Fa Ga . By the remark
before the proof of Lemma 1, we may suppose that = a in a X. 95
Since X is compact, we obtain a finite covering {i } of X and homomor-
phisms i : F i G i such that i = j = in i j X. Again, by
the remark before Lemma 1 and the compactness of X, we may assume
that i = j in Ui U j where Ui is an open set containing i X. It is
clear that there is an open set U X such that i = j in i j U
and i = on i X. Lemma 2 follows.
Proof of Theorem 2. From the definition of coherent analytic sheaves

(as locally isomorphic to quotient of O Xp , O X being the restriction of O V )
it follows that if a X, there is an open set a , a a and a coherent
analytic sheaf G a on a such that G aa X F a X . Since X is compact,
there are finitely many coherent analytic sheaves G 1 , . . . , G r on open
S
r
sets 1 , . . . , r respectively ( i X) such that Gi i X F i X .
i=1
Let this isomorphism be given by a mapping ci : G ii X F i X .
On i j X, there is thus a homomorphism ci j : G ii j X
j
G i j X (where ci j = c1
j ci ). Also cii = identity, ci j c jk cki = identity
on i j k X. By Lemma 2, there is a neighbourhood Ui j of
i j X (with Ui j = U ji ) and a homomorphism
j
i j : G ii j Ui j G i j Ui j .
Also, by the remark before Lemma 1, we may suppose that Ui j is such

that ii = identity on i Uii , i j jk ki = identity on i j k
Ui j U jk Uki ; in particular, i j is an isomorphism of G ii j Ui j onto
G i i j Ui j . Since X is compact, there is a neighbourhood U of X 96
j
such that i j is an isomorphism of G ii j U onto uGi j U , and ii =
identity, i j jk ki = identity (on i U, i j k U respectively).
These sheaves and isomorphism give rise to a coherent analytic sheaf G
on a neighbourhood of X. Also, by the definition of the i j it is clear
that G X F and the proof of Theorem 2 is complete.
If the manifold V is paracompact, X may be replaced by any closed

set in Theorem 2. For the details of proof, see Cartan [4] or Dowker [6].
Chapter 13
Cohomology with coefficients

in a sheaf
1 Cohomology of a covering
Let X be a topological space, F a sheaf of abelian groups on X. Let 97
O = {Oi }iI be an open covering of X. We shall denote by Oi ,...,i p the
set Oi . . . Oi p and, U being an open set in X, by (U, F ) = F U the
sections of F over U. (If U is empty, we set (U, F ) = 0).
Definition. A p-cochain of O is a mapping c of I p+1 such that c j ... j p

(Oi i p , F ) and which is, moreover, alternate, (i.e., c j ... j p = ci ...i p if
( j , . . . , j p ) is a permutation of (i , . . . , i p ) and = 1 according as this
permutation is even or odd).
C p (O, F ) will denote the abelian group of the p-cochains of O,

P p
C (O, F ) = C (O, F ) the direct sum of the C p (O, F ) for p 0.
p0
The coboundary operator p : C p (O, F ) C p+1 (O, F ) is defined
as follows: if c C p ,
p+1
X
p
( c)i ...i p+1 = (1) j ci ...i j ...i p+1
j=0
101
102 13. Cohomology with coefficients in a sheaf
where i , . . . i j . . . i p+1 signifies that the j-th index i j is to be omitted.

The p give rise to a coboundary operator : C (O, F ) (O, F ). It
can be verified that
p p1 = 0 i.e., = 0.
Now, the elements c of C (O, F ) such that c = 0 are precisely those

98 elements with ci1 ci = 0 in Oi i1 (by the definition of ) and so, since
ci = ci1 in Oi Oi1 they correspond to sections over the whole of X,
i.e., to elements of (X, F ). Hence we have the following compelx:
i 1
C (O, F ) C 1 (O, F ) . . .
0 (X, F )
and p p1 = 0. Setting z p (O, F ) = kernel of p , B p (O, F ) = image

of p1 , we define the p-th cohomology group H p (O, F ) of the covering
O with coefficient sheaf F by
H p (O, F ) = Z p (O, F )/B p (O, F ), p > 0

H (O, F ) = (X, F ).
2 Cohomology of the space X

Let O = {Oi }iI , = { }A be two (indexed) coverings of X and
suppose that is a refinement of O, i.e., there is a mapping : A I
such that O() . (We do not consider as given once for all, but
merely require its existence). The mapping of C p (O, F ) in C p (, F )
defined by
(c ) ... p = restriction of c( )...( p ) to ,..., p
induces a mapping of H p (O, F ) H p (, F ) (this is easy to verify).
Proposition 1. does not depend on .
Proof. Let , be two mappings A I such that O() O() .

Suppose that A is totally ordered. For p = 0, the result is obvious since
2. Cohomology of the space X 103
H (O, F ) = (X, F ) for every O. If p 1, we define a mapping k

(homotopy operator): C p+1 (O, F ) C p (, F ) by
p
X
(kc) ... p = (1) j c( )...( j )( j )...( p ) ,
j=0
if < 1 < . . . < p in the total order of A and then define kc uniquely 99
to be an (alternate) cochain. If the cochains corresponding to the maps
, are c , c in C p (, F ) respectively, it can be verified that
(k + k)c = c c .
Consequently, if c is a cocycle (i.e., c = 0), then c c is a coboundary,
c c = (kc) and the mappings and induce the same homomor-
phism of H p (O, F ) H p (, F ) . This proves Proposition 1.
This homomorphism is denoted by (O, ). It satisfies certain ob-
vious transitivity properties (as a functions of O, ).
If p = 0, is always an isomorphism as observed above.
Proposition 2. If p = 1, is a monomorphism.
We have to show that if : A I is such that O() and
c = 0 in H 1 (, F ), then c = 0 in H 1 (O, F ). Let c be a cochain with
c()() = in . For every i and x Oi if x , set ci (x) =
(x)+c()i (x). If x is also in , then (x)+c()i (x) = (x)+c()i (x)
since a = c()() = c()i c()i . Hence this defines a section ci
on Oi and clearly ci j = ci c j in Oi j , which proves the proposition.
The homomorphism (O, ) defined above depends only on the cov-
erings O, . If O, are refinements of one another, (O, ) is an iso-
morphism. Hence we identify all coverings which are two by two refine-
ments of one another, and consider the class of all indexed coverings
modulo this identification. It is clear that this quotient can be put in
one-one correspondence with a subclass of the power set of X and so is
a set. It is clearly a directed set and we have a directed system
{H p (O, F ), (O, )}O
for every p. The direct limit of this system is called the p-th cohomology 100
group of X with coefficient sheaf F and is denoted H p (X, F ). It is
obvious that H (X, F ) = (X, F ).
3 The exact cohomology sequence

Let
i
0F H 0
G
be an exact sequence of sheaves. This evidently gives rise to an exact
sequence
0 C (O, F ) C (O, G ) C (O, H )
but the last mapping is not in general onto. If we denote by Ca (O, H )
the image C (O, G ) (group of cochanes ascensionelles) then we ob-
tain the exact sequence
0 C (O, F ) C (O, G ) Ca (O, H ) 0.

p p p
It Za is the group of cochains c in Ca (O, H ) with c = 0 and Ba is the
p1 p
group of cochains c, c Ca (O, H ), we define the group Ha (O, H )
by
p p p
Ha (O, H ) = Za /Ba .
p
We now define a mapping d : Ha (O, H ) H p+1 (O, F ) in the fol-
p
lowing way. Let h Za and let hi ...i p = (gi ...i p ) where g C p (C , G );
also, since clearly and commute, {(g )i ...i p+1 } = (h)i ...i p+1 = 0 since
h Zap (g )i ,...i p+1 being a section over Oi ...i p+1 which goes to 0 under
, g C p+1 (O, F ) (since F is the kernel of ). It is easy to see that
101 the class of g in H p+1 (O, F ) remains unchanged if g is replaced by
another cochain g with g = h and if h is replaced bya cohomologous
cocycle. This defines d .
It is clear that i, induce homomorphisms
i
i , : H p (O, F )
H p (O, G ) H p (O, H )
and it can be verified that the following sequence is exact:
i d i
H (O, G ) Ha (O, H ) H 1 (O, F )
0 H (O, F ) ...
d i p d i
H p (O, G ) Ha (O, H ) H p+1 (O, F )
. . . H p (O, F ) ...
3. The exact cohomology sequence 105
p
We can now define the groups Ha (X, H ) by taking direct limits as the
covering becomes finer, as for the groups H p (X, H ). Also there is a
canonical mapping Hap (OH ) H p (O, H ) and so a canonical homo-
morphism Hap (X, H ) H p (X, H ). Since the operation of taking direct
limits commutes with exact sequences we obtain the exact sequence
d i p d i
H p (X, G ) Ha (X, H ) H p+1 (X, F )
. . . H p (X, F ) ...
p
It is of interest to decide when Ha (X, H ) = H p (X, H ). This is so in
the case when X is paracompact (i.e., a Hausdorff space in which every
covering admits a locally finite refinement).
Theorem. If X is paracompact and

i
0F H 0
G
an exact sequence of sheaves on X, then the canonical homomorphism
Hap (X, H ) H p (X, H )
is an isomorphism.
The theorem follows at once from the following
Lemma. If O = {Oi }iI is a covering of X and c C p (O, H ), there ex-

ists a covering = { }A and a mapping : A I with O() 102
such that the induced homomorphism : C p (O, H ) C p (, H )
p
take c to a cochain (c) Ca (, H ).
Proof of the lemma: Since X is paracompact, we may suppose O lo-

cally finite. Since X is normal, (see Dieudonne [5]) there is an open
covering {Oi }iI such that Oi Oi . For every x X, we choose an open
neighbourhood x of x such that
(i) x Oi (respectively Oi ) implies x Oi (respectively Oi ).
(ii) x Oi , 0 implies x Oi .
(iii) If x Oi ...i p , there is a section S of G over x such that (s) =

ci ...i p on x .
Since O is locally finite, it follows from the definition of quotient

sheaf that (iii) can be fulfilled; and (i) and (ii) are then ensured if we
choose the x small enough. This gives us a covering { x } xX = of
X; we choose a mapping : X I such that x O(x) . It is then easy
to verify that and have the property stated in the lemma.

Chapter 14
Coherent analytic sheaves on

a cube
1 The abstract de Rham Theorem

Let X be a paracompact topological space, F a sheaf of abelian groups 103
on X, and suppose that
i d d1 dk1 dk
G G1 . . . Gk . . .
0F
is an exact sequence of sheaves on X and that H p (X, G k ) = 0 for p 1,

k 0. Consider the sequence
dk
i d dk1
0 (X, F )
(X, G ) . . . (X, G k ) . . .
with the induced homomorphisms dk (this is not in general exact).

Then
H k (X, F ) kernel dk / image dk1

for k 1.
Proof. Consider the exact sequence

dk1 dk
G k1 G k G k+1
107
108 14. Coherent analytic sheaves on a cube
and let H k = kernel dk = image dk1 . Then we have an exact sequence
0 H k G k H k+1 0
and X being paracompact, we obtain, for q > 0, the exact sequence
H q (X, G k ) H q (X, H k+1 ) H q+1 (X, H k ) H q+1 (X, G k )
and since H p (X, G k ) = 0, if p 1,
H q (X, H k+1 ) H q+1 (X, Hk )
for q 1. By iteration
H p (X, F ) H p1 (X, H 1 ) . . . H 1 (X, H p1 ). (1)
Also we have the exact sequence
0H p1 G p1 H p 0
104 and the induced exact sequence
H (X, G p1 ) H (X, H p ) H 1 (X, H p1 ) H 1 (X, G p1 ).
Since the last term is 0 by hypothesis,
H 1 (X, H p1 ) H (X, H p )/ image H (X, G p1 ).
It is easy to see that H (X, H p ) kernel dk , while image H (X, G p1 ) =

image dk1 and the result follows from (1).
Applications.
a) de Rhams Theorem. Let X = V be a paracompact differentiable

manifold and E p the sheaf of germs of differentiable p-forms on
V. Then we have a sequence
i d d d d
E
0 C Ep
E p+1

1. The abstract de Rham Theorem 109
(C is a constant sheaf, C being the group of complex numbers).

This sequence is exact by the local form of Poincares theorem
(VIII). By the method given in Step 1 of the solution of the gen-
eralized first Cousin problem (VIII) it is seen that H k (V, E p ) = 0
if k 1, p 0. Hence, the above theorem shows that
H p (V, C)
(group of closed p-forms)/(the differentials of (p 1) forms), i.e.,

H p (V, C) is the same as the p-dimensional d-cohomology of V.
This is the theorem of de Rham.
b) Dolbeaults theorem. Let V: a paracompact complex manifold

and E o,p the sheaf of germs of forms of type (0, p) on V. Then,
the local form of Grothendiecks theorem shows that the sequence
i d d
E o,o E 0,1 . . .
0O
is exact (O is the sheaf of germs of holomorphic functions on V).

Again, by the method of the generalized first Cousin problem, we 105
prove that
H k (V, E o,p ) = 0, p 0, k 1,
and we obtain H p (V, O) (d -closed (0, p) forms)/(d -differenti-
als of (0, p 1) forms), i.e., H p (V, O) is the same as the p - di-
mensional d -cohomology of the (0, r) forms, r = 0, 1, . . . on V.
Similar reasoning proves that the p-th cohomology of V with co-
efficients in the sheaf of germs of holomorphic (q, 0)-forms is the
p-th d -cohomology of the (q, r)-forms, r = 0, 1, . . . on V. This
is a particular case of Dolbeaults theorem.
c) Let K be a cube imbedded in C n , i.e., K is a subset of a fixed C n ,

consisting of the points z C n with
|Rzi | ai |Izi | bi , ai , bi 0.
Consider the sheaf O = O K (in the sense of XII, i.e., the restric-
tion to K of the sheaf of germs of holomorphic functions in C n ).
We define the sheaves E o,p in the same way as the restriction to

K of the sheaf of germs of (0, p)-forms in C n . As before we have
the exact sequence
i d d
E o,o E o,1 . . .
0 O
and H p (K, O) (d -closed (0, p)-forms)/(d -differentials of (0,

p 1)-forms). Grothendiecks theorem shows that this quotient
is zero. (The theorem was proved only when ai , bi > 0, but it is
immediate that the theorem holds also when some of the ai , bi are
0). This proves the following
Theorem 1. If K is a closed cube in C n , H p (K, O) = 0 for p 1.
106 Corollary. If p, q 1, H p (K, O q ) = 0.
(One has only to apply Theorem 1 to the components of an element
c of Z p (K, O q ) to see that c is a coboundary).
It is instructive to compare the above proof with the solution of the
generalized first Cousin problem for a cube which implies that H 1 (K,
O) = 0. (This is exactly the generalization of that proof to the more
general setting here).
2 Coherent analytic sheaves on a cube

Let K be a cube in C n and F a coherent analytic sheaf on K. Funda-
mental Theorem. (Oka-Cartan-Serre).
A) F is (globally) a quotient of a sheaf O N (This can also be
expressed by saying that there are N (global) sections f1 , . . . , fN of F
which Ob -generate Fb for every b K.)
B) For p 1, H p (K, F ) = 0.
We introduce the following statement.
A ) F is (globally) a quotient of a coherent analytic sheaf locally
isomorphic to O N .
The proof of the fundamental theorem now divides into two parts,
the proof that A ) implies A) and B) for a cube and the proof of A ) for
a cube.
2. Coherent analytic sheaves on a cube 111
Step 3. The truth of A ) for every F implies the truth of A) and B) for
every F .
(i) Suppose A ) true for all F . Then F is a quotient of a sheaf G
locally isomorphic to O N . Hence G defines a class of analytic
bundles over a neighbourhood of K (end of XI; the results proved
there relate to topological bundles, but they remain valid with ob-
vious modifications for analytic bundles). By Theorem 2 of X, 107
this class is the trivial class (on some neighbourhood of K) and so
G is O N and A) is proved.
(ii) According to A), we have an exact sequence
O N1 F 0
and if G 1 is the kernel of this mapping, G 1 is coherent analytic by

Theorem 1 of XII, and we have the exact sequence
0 G 1 O N1 F 0.
Since A) is supposed to hold for all coherent analytic sheaves, we

obtain exact sequences
0 G 2 O N2 G 1 0
............................
0 G k O Nk G k1 0
............................
This leads to the exact sequence
H p (K, O Nk ) H p (K, G k1 ) H p+1 (K, G k ) H p+1 (K, O Nk ).
If p 1, the first and last terms are zero by Theorem 1 above and
consequently,
H p (K, G k1 ) H p+1 (K, G k ) (2)
Now, there is an integer m such that every covering of K has a

refinement in which the intersection of any m + 1 sets (of the re-
finement) is empty. [This is seen for example by subdividing K
into smaller cubes by hyperplanes parallel to the coordinate hy-

perplanes. The stetement is, however, essentially of dimension 108
theoretic character; classical dimension theory shows that te best
possible m is 2n + 1 and that this is not special to cubes. The
above property is the starting point of the more modern theory of
dimension]. Hence H m+p (K, H ) = 0 for every sheaf H over K
and p 1. By iterating (2), we obtain
H p (K, F ) H p+1 (K, G 1 ) . . . H m+p (K, G m ) = 0
and B) is proved.
Step 4. Proof of A ) for a cube. The proof will be by induction on

the real dimension of the cube K. If K has dimension 0, A ) is just the
definition of a coherent analytic sheaf. Suppose A ) true for all cubes K
of real dimension p and all coherent analytic sheaves F on K . Then
A) and B) are also true for K and F .
Let now K be a cube of real dimension p + 1. We find a coordinate

hyperplane such that the intersection of K with any hyperplane paral-
lel to it is of dimension p (if it is non-empty). The restriction of F to
each such intersection is coherent analytic, and by inductive hypothesis,
a quotient of O N . The extension theorem of XII shows that there is
a neighbourhood of each intersection in which F induces a coherect
analytic sheaf which is isomorphic to a quotient of same O N . By the
Borel-Lebesgue lemma, it follows that we have only to prove the fol-
lowing result:
Given two adjacent cubes K1 , K2 of dimension p + 1 such that P =
K1 K2 is of dimension p, and a coherent analytic sheaf F on K1 K2
such that F is a quotient of a sheaf O N1 on K1 , and a quotient of a sheaf
O N2 on K2 , then F is a quotient of a sheaf locally isomorphic to O N1 +N2
on K1 K2 .

f1

109 Let ( f ) = ... , f1 , . . . , fN1 being sections of F over K1 which Oa -

fN1
2. Coherent analytic sheaves on a cube 113

g1

generate Fa for a K1 , (g) = ... , g1 , . . . , gN2 being sections of F

gN
over K2 which Ob -generate Fb for b K2 . It is easily seen that it is
enough to find a holomorphic regular matrix c on P such that
! !
f 0
c = .
0 g
To find c, we use the following lemma, which is of interest by itself.
Lemma. If is a section of F over P, there are N1 holomorphic func-

tions 1 , . . . , N1 on P such that
= 1 f1 + + N1 F N1 .
Proof of the lemma: Consider the exact sequence
0 G O N1 F 0
(these being sheaves on P = K1 K2 ; G is the kernel of the homomor-

phism O N1 F ). This gives us the exact sequence
H (P, O N1 ) H (P, F ) H 1 (P, G )
and by inductive hypothesis, H 1 (P, G ) = 0 since P has dimensions p.

Hence the mapping
H (P, O N1 ) H (P, F )
is onto and the lemma follows.

Construction of c. The lemma proves that there is a matrix 1 of N1
columns and N2 rows such that

f1 g1

1 ... = ...

fN1 gN2
i.e., 1 ( f ) = (g). In the same way there is a matrix 2 of N2 columns 110

and N1 rows such that
2 (g) = ( f ).
If we set ! !
I 2 I 0
c=
0 I 1 I
! ! ! !
I 0 f f I 2
it is clear that c is regular. Also takes to ,
1 I 0 g 0 I
! !
f 0
takes to . This concludes Step 2 and with it the proof of the
g g
fundamental theorem.
Remark . All reference to bundles can be avoided if after the construc-

tion of c, one applies Cartans theorem on holomorphic, regular matrices
to prove directly A) and B) without introducing A ).
The statement A ) is true for a much wider class of sets than is A).
Actually the problem of classifying the compact sets for which A ) is
true is still open (even in C n ).
Chapter 15
Stein Manifolds: preliminary

results
1 Theorems A) and B) for closed polydiscs in C n

Let V be a complex analytic manifold, K a compact subset of V. We 111
say that Theorems A) and B) are true for K if A) every coherent analytic
sheaf F on K is a quotient of a sheaf O N and B)H p (K, F ) = 0 for
p 1.
Proposition 1. Theorems A) and B) are true for (closed) polydiscs in

Cn.
Proof. Let P be the given polydisc. P has a fundamental system of

neighbourhoods each of which is analytically isomorphic to a closed
cube: P has a fundamental system of neighbourhoods which are open
Q Q
polydiscs ; , being a product of open discs, is isomorphic to an open
Q Q
cube 1 and the image of P in 1 is contained in a closed cube con-
Q Q
tained in 1 whose inverse image in is a neighbourhood isomorphic
to a closed cube. Since Theorems A) and B) are true for closed cubes,
it follows that Theorems A) and B) are ture for a fundamental system of
neighbourhoods of P and after the extension theorem of XII it is easily
seen that A) is true for P, and B) follows from
115
116 15. Stein Manifolds: preliminary results
Lemma 1. Let X be a paracompact topological space, F a sheaf of

abelian groups on X. Let Y be a closed set in X with a fundamental
system of closed neighbourhoods L. Then H p (Y, F ) is the direct limit of
H p (LF ) as L shrinks to Y.
Proof. For p = 0, this follows from the fact that every section over Y
can be extended to a section over an open neighbourhood of Y and the
fact that the set of points at which two sections coincide is open. For
p > 0, we construct an exact sequence
i d d1 dk1 dk
G G 1 . . . G k . . .
0F (1)
112 such that H p (E, G 1 ) = 0 if p > 0, l 0 for all subsets E of X. To do
this it is clearly sufficient to construct an exact sequence
0 F G with H p (E, G ) = 0 for p > 0
(for the construction can be repeated with the quotient G /F and the
process continued). We define G to be the sheaf of germs of all map-
pings f : X F with f (x) Fx for x X. Then clearly we have an
exact sequence
0F G
and H p (E, G ) = 0 for p > 0 by the theorem in the appendix
Having constructed the exact sequence (1) we consider the associ-
ated sequence
dk
i d dk1
0 (E, F )
(E, G ) . . . (E, G k ) . . .
and by the abstract de Rham theorem, we have
H p (E, F ) kernel dp / image dp1 (p 1). (2)

As in the case when p = 0, as L shrinks to Y, kernel dk , image dk1
(with E replaced by L) have as their direct limit, the kernel and image
of the mappings
(Y, G k ) (Y, G k+1 ) and (Y, G k1 ) (Y, G k )
respectively and an application of (2) with E = Y establishes the lemma.

2. Coherent analytic sheaves on an analytic submanifold 117
2 Coherent analytic sheaves on an analytic subman-

ifold
Let X be a topological space, Y a closed subset of X, F a sheaf of abelian
groups on Y. We define a sheaf F on X by setting Fa = Fa if a Y, =
the group 0 if a < Y. Clearly, this defines a sheaf on X. Then one has 113
Proposition 2. For p = 0, 1, . . .
).
H p (Y, F ) H p (X, F
Proof. If O = {Oi }iI is an open covering of X, Oi = Oi Y, then

{Oi } = O is an open covering of Y and clearly
) H p (O F ).
H p (O, F
Also given an open covering O = {Oi } of Y, if Oi is such that Oi Y =

Oi , then the open covering {Oi , X Y} of X gives rise to O in the above
way and Proposition 2 follows.
Definition . Let V n be a complex analytic manifold of complex dimen-

sion n. Let W m be a closed subset of V n . W m is called an analytic
submanifold of dimension m, if, for a W m , the local coordinates
(z1 , . . . , zn ) at a on V n (in an open set U V n ) can be so chosen that
W m U = {z U zm+1 = . . . = zn = 0}.
An application of the implicit function theorem shows that if W is

a complex analytic manifold of dimension m and i an analytic one-one
mapping of W into V n , i(W) is an analytic submanifold of V n if and only
if
1) i is proper: the inverse image of a compact subset of V n is a com-

pact subset of W;
2) i has rank m (i.e., the Jacobian matrix of i has rank m at every

point of W).
Let V be a complex manifold and W a submanifold of V. Let X V

and Y = X W. We shall denote by V O, W O the sheaves of germs of
holomorphic functions on V and W respectively, considered as complex
manifolds in their own rights.
114 Let F be a coherent W O-analytic sheaf on Y, and F the sheaf which

continues F to X by 0 outside Y. Then F a has a structure of V O-
analytic sheaf on X: if a < Y, Fa is an V Oa -module (Fa being 0); if
a Y, fa Fa = F and ha V Oa, ha fa is defined to be haW fa , where
haW is the restriction of ha to W.
Let I(W) be the subsheaf of V O consisting of those germs which
vanish on W. Then we have the isomorphism
V O/I(W)

W O;
by a theorem of Cartan [2], [3, lecture XVI], I(W) is coherent.

is a
Proposition 3. If F is a coherent W O-analytic sheaf on Y, then F
coherent V O-analytic sheaf on X.
Proof. If F is the sheaf W OY (restriction of WO to Y), then, by what

we observed above,
V O X /I(W)X W O
and so O is a coherent V O-analytic sheaf.
In the general case, let F be a sheaf on Y, a Y, an open neigh-
bourhood of a in Y such that
N
F WO /R,
where R is a coherent analytic subsheaf of W O N .

Let be chosen so small that there are N1 sets of N holomorphic
functions on W which W Ob -generate Rb at every point of . If is
again sufficiently small, there is a neighbourhood of a in V such that
Y = and these functions are restrictions of holomorphic func-
N
115 tions in to . Let R be the subsheaf of V O generated by these
N N
N1 elements of V O . Then R is a coherent analytic subsheaf of V O ,
while clearly F X V O N N
X /R X + I (W) X and the proposition fol-
lows.
3. Stein Manifolds 119
Q
Proposition 4. Let P be a closed polydisc in C n , an open polydisc
Q
P. Let W be an analytic manifold which is a submanifold of . Then,
Theorems A) and B) are true for W P (considered as subset of W).
This follows at once from Propositions 1, 2 and 3.
3 Stein Manifolds
Definition . A complex analytic manifold V of dimension n which is
countable at infinity is said to be a Stein manifold if
() V is holomorph-convex (VII, 3);
() for any two points a , b on V, there exists a holomorphic f on V,

such that f (a) , f (b).
() if a V, there are n functions holomorphic in V which form a

system of local coordinates at a.
Examples of Stein manifolds.

1. Univalent domains of holomorphy in C n (see VII, Prop. 1)
2. Any open connected Riemann surface. [A Riemann surface is

countable at infinity by Rados theorem; (), () and () follow
from Runges theorem. For the details of proof, see H. Behnke
and K. Stein: Entwicklung analytischer Funktionen auf Riemann-
schen Flachen, Math. Ann., 120 (1948), 430-461, and B. Mal-
grange: Existence at approximation des solutions des e quations
aux dexivees partielles et des e quations de convolution Thèse,
Paris, 1956 (Chap.III, 4)].
3. Analytic submanifolds of C n . In particular, algebraic varieties 116

over C which have no singularities.
Lemma 2 (on Stein manifolds). Let V be a Stein manifold, K a compact
subset of V such that K = K (K is the HV -envelope of K; see VI). Then
K has a fundamental system {L} of compact neighbourhoods L having
the following properties:
To every L correspond an open set L, a closed polydisc P C N

Q
an open polydisc P and a finite number N of holomorphic functions
f1 , . . . , fN on V such that the restrictions to of the fi realize as an
Q
analytic submanifold of and such that (L) = P () where is the
mapping ( f1 , . . . , fN ) of V in C N .
Proof. Let be a relatively compact neighbourhood of K, and let F be

the boundary of . For every a F, there is an f such that | f (a)| > 1,
|| f ||K < 1. Since F is compact and the set of a with | f (a)| > 1 is open,
there are a finite number, f1 , . . . , fN of holomorphic functions such that
|| fi ||K < 1 while max | fi (a)| > 1 for a F. Let be the set of a
i
with | fi (a)| < 1 for i = 1, . . . , N . Also, the set of a with | fi (a)| ,
< < 1 is compact since is compact and the closure of this set

does not intersect F. This shows that the mapping of in C N defined
by ( f1 , . . . , fN ) is proper.
Set = . By adjoining a finite number of functions fN +1 , . . . , fN
to f1 , . . . , fN we can ensure that points of are separated by the map-
ping = ( f1 , . . . , fN ) and such that is of maximal rank {this follows
from properties (), () of Stein manifolds and the compactness of }.
117 If 1 > > and || fN +1 ||K < A, . . . , || fN ||K < A, we thke P to be
the polydisc |zi | , i N , |zi | A, i N + 1 in C N and L to be
the inverse image in of P () under the mapping . Since K has a
fundamental system of relatively compact neighbourhoods , Lemma 2
is proved.
Theorem . Let V be a Stein manifold, K a compact set V such that

Then
K = K.
1) Theorems A) and B) are true for K.
2) Every holomorphic function on K can be approximated, uniformly

on K by holomorphic funtions on V.
Proof. 1) follows from Proposition 4 and Lemmas 1 and 2. To prove

2), let g be a holomorphic function on K, and L a neighbourhood of K
having the properties of Lemma 2, such that g is holomorphic on L.
3. Stein Manifolds 121
Now, by Proposition 4, we have the exact sequence
CN O P V OL 0
Q
(L is considered as a subset of ), and if I is the kernel of the first
mapping, the sequence
0 I C N O P V O L 0
is exact. The associated exact cohomology sequence
H (P, C N O P ) H (P, V O L ) H 1 (O, I)
shows, since H 1 (P, I ) = 0 by Theorem B) for a polydisc, that every

element of H (P, V O L ) is the image of an element of H (P, C N O P ) and
g is the restriction to L of a holomorphic function on P; hence g can be
expanded in a power series in the f1 , . . . , fN which converges uniformly

on K L. Since the partial sums of this power series, being polynomials 118
in f1 , . . . , fN are holomorphic on V, the theorem is proved.
Appendix
Theorem . Let X be a topological space, F a sheaf of abelian groups

on X which is such that any section of F over an open set of X can be
extended to a section of F over X. Then, for any open covering O =
{Oi }iI of X, H p (O, F ) = 0 for p > 0, and in particular H p (X, F ) = 0
for p > 0.
Proof. The proof is by induction on p.
a) p = 1. Let c be a 1-cocycle of the covering O = {Oi }iI . Suppose

J is a subset of the indexing set I such that there is a 0-cochain
of O with i j = ci j for i, j J. Let I, < J. We
define a 0-cochain as follows: i = i if i , , = i + ci on
O Oi , i J. Then on O Oi O j , i + ci = j + c j since
i j = ci j = c j ci (c being alternate). Hence is defined
S
uniquely on (O Oi ). By hypothesis, can be extended to
iJ
a section of F over O and the cochain is defined completely.

Also i j = ci j if i, j J {}. It is clear that J is non-empty
(since c is alternate) and the theorem for p = 1 follows by an
application of Zorns lemma.
b) p > 1. Suppose the theorem true with p replaced by p 1 for

all spaces X, all coverings O of X and all (p 1)-cocycles of O.
Let c be a p-cocycle and let J be a subset of I such that there is
a (p 1)-cochain with ()i ...i p for i , . . . , i p J. Let I,
< J.
119 For every i , . . . , i p2 J, we determine a section i ...i p2 of F

over Oi ...i p2 such that
p1
X
ci ...i p1 = (1)k i ...i ...i + (1) p i ...i p1
k p1
k=0
over Oi ...i p1 . This is possible: it is easily seen, from the definition of

, that the (p 1)-cochain c defined by
ci ...i p1 = ci ...i p1 + (1) p1 i ...i p1

S
is a cocycle of the covering {O Oi }iJ of the space Y = (O Oi )
iJ
(since c is a cocycle) and the existence of the i ...i p2 follows from
inductive hypothesis.
We new define a (p 1)-cochain 1 as follows: if i , . . . , i p2 J,
(1 )i ...i p2 = i ...i p2 ; 1 is defined by the condition that it is alternate
for other p-tuples of indices of J {} which contain and (1 )t = t if
t J p . 1 has the property that
(1 ) j ... j p = c j ... j p for j , . . . , j p J {}
while 1 = on J p . If we partially order the pairs (J, ) by setting

(J, ) < (J , ) if J J and = on J p , the theorem follows by an
application of Zorns lemma.
Chapter 16
Coherent analytic sheaves on

a Stein manifold
1. We shall prove here the fundamental theorem of Oka-Cartan-Serre 120

on Stain manifolds.
Fundamental Theorem. Let V be a Stein manifold and F a coherent
analytic sheaf on V. Then
A) For every a V, H (V, F ) Oa -generates Fa .
B) For p 1, H p (V, F ) = 0.
(It is clear that for compact subsets of V, Theorem A) as formulated
in XV is equivalent to the theorem as formulated above).
The following two results will be required, the first will not be
proved here. For the proof, see Cartan [1].
Theorem 1. Let V be a complex analytic manifold and let a V. Let M
p
be a submodule of Oa {as an Oa -module} and let f = ( f1 , . . . , f p ) O p
(O = HV is the space of all holomorphic funtions on V). Suppose that f
is the limit in O p of functions fi O p such that ( fi )a m. Then fa m.
Lemma . Let K be a compact subset of the Stein manifold V such that
K = K (K is the HV -envelope of K) and F a coherent analytic sheaf
on K. Let f1 , . . . , fm H (K, F ) and suppose that for every a K,
123
124 16. Coherent analytic sheaves on a Stein manifold
f1 , . . . , fm Oa -generate Fa . Then f1 , . . . , fm H (K, O)-generate H (K,

F ).
This lemma is proved, using Theorems A) and B) for K in exactly

the same way as was the lemma in the proof of Theorems A) and B) for
a cube in XIV.
2 Topology on H (V, F )

121 Let {K p } be a sequence of compact subsets of V such that K p K p+1 ,
S

K p = V and K p = K p (such a sequence exists since V is eountable at
1
infinity and, for any compact set K, (K) = K).

For an integer N 1, we introduce a norm in H (K p , O N ) by setting
the norm of f = ( f1 , . . . , fN ) H (K p , O N ) to the equal to the greatest
of the suprema of | f1 |, . . . , | fN | on K p . We then introduce a seminorm
|| . . . || p on H (K p , F ) as follows: by Theorem A) for K p , F K p O N /R
and || . . . || p is defined to be the quotient seminorm of the norm on O N .
It is easy to verify that two isomorphisms F K p O N1 /R 1 O N2 /R 2
give rise to equivalent seminorms. Also, for every p, there is a canonical
mapping H (V, F ) H (K p , F ) (namely, restriction to K p ). On H (V, F )
we put the weakest topology for which these mappings are continuous
in these seminorms (which may also be described by saying that f
H (V, F ) tends to zero if || f || p 0 for every p). Also, it is easily seen
that the topology induced by || . . . || p+1 on H (K p , F ) is finer than that
given by || . . . || p .
The next results will show that H (V, F ) is a Frechet space. One
has only to show that it is Hausdorff and complete.
(a) If f p+1 H (K p+1 , F ) and || f p+1 || p+1 = 0, then the restriction
of f p+1 to K p is zero. (As a consequence, the topology of H (V, F ) is
Hausdorff).
Proof. If 1 , . . . , N p+1 Oa -generate Fa for a K p+1 (Theorem A) for

2. Topology on H (V, F ) 125
K p+1 ) and
N
X p+1
f p+1 = ci i (lemma on p.120)

i=1
(the ci are holomorphic functions on K p+1 ) it follows from the definition 122
of the seminorm k . . . k p+1 and the fact that k f p+1 k p+1 = 0 that, given
> 0 there are holomorphic functions c1 , . . . , cN p+1 on K p+1 such that
N
X p+1
f p+1 = ci i
i=1
and sup |ci (a)| < . If i = ci ci , then (c1 , . . . , cN p+1 ) is uniformly

i,aK p+1
approximated on K p+1 by (1 , . . . , N p+1 ) where (1 , . . . , N p+1 ) is an ele-
ment of the sheaf R of relations between 1 , . . . , N p+1 . It follows from

Theorem 1 (stated on page 120) that on K p+1 K p (c1 , . . . , cN p+1 ) R
and (a) is proved.
(b) If f1 , . . . , f, . . . is a sequence of elements of H (K p+1 , F ) such
that
X
|| fk || p+1 < +,
k=1
P
N
then the sequence { fk } has a limit point in H (K p , F ). The restric-
k=1
tions to K p1 of two such limit points coincide.
Proof. Let 1 , . . . , N p+1 Oa -generate Fa for a K p+1 and let

N
X p+1
fk = c(k)
i i .
i=1
P
P
Then, since || fk || p+1 < +, the c(k)
i can be so chosen that max
k=1 k i
||c(k) < + (by the definition of || . . . || p+1 , one may, for exam-
i ||K p+1
ple, take the c(k) (k)
i such that ||ci ||K p+1 < 2|| fk || p+1 ). Then, for each i,
P (k)
k ci converges to a holomorphic function ci on K p , and it is clear that
P
N NP
p+1
|| fk ci i || p 0 as N . This proves the existence of the
k=1 i=1
limit point. The uniqueness on K p1 follows at once from a).
123 (c) H (V, F ) is a Frechet space.
Given a Cauchy sequence {sk }, ||sk sl || p 0 as k, l for every
p. If we choose a sequence {nk } of integers such that ||sm snk || p < 1/2k
P
for p k and m nk (with nk+1 > bk ) then ||snk+1 snk || p < + for
k
every p. It follows at once from (b) that {sk } has a limit in H (V, F )
which is unique since H (V, F ) is Hausdorff.
(d) (Approximation property). Given f p H (K p , F ) and > 0
there is a section f H (V, F ) such that || f p f || p < .
Proof. If 1 , . . . , N p+1 Oa -generate Fa for a K p+1 , then their restric-

tions to K p clearly Oa -generate Fa for a K p . Hence, by the lemma,
N
X p+1
fp = ci i
i=1
where the ci are holomorphic on K p . By the theorem of XV, the ci

can be approximated uniformly on K p by holomorphic functions on V.
This shows that f p can be approximated on K p (in || . . . || p ) by a sec-
tion f p+1 H (K p+1 , F ). Approximating f p+1 on K p+1 by f p+2
H (K p+2 , F ) in || . . . || p+1 and continuing this process, we construct a se-
quence f p+1 , f p+2 , . . . such that || f p+k+1 f p+k || p+k k . If the k are small
P

enough, it is seen that || f p+m+1 f p+m || p+m < + (since || . . . ||m+1 is
m=k
P

finer than || . . . ||m on Km ) and so f p+k = f p+k + ( f p+m+1 f p+m ) is
k

defined uniquely in K p+k2 . It is clear that f p+k+1 = f p+k on K p+k2 and

so there is an f H (V, F ) with f = f p+k on K p+k2 . If the k are small
enough, f approximates to f p in || . . . || p .
124 [If we say that a sequence {sm }, where sm H (Km , F ), converges
to s H (V, F ) if ||s sm ||m 0 as m , the above proof may be
rephrased by saying simply that f p+m f as m .].
3. Proof of the fundamental theorem 127
3 Proof of the fundamental theorem
Proof of Theorem A):

Let a V and suppose that a K p . Let 1 , . . . , N p Ob -generate Fb
for b K p . Theorem A) asserts that the set of the N p -tuples (a1 , . . . ,
P
aN p ), where ai Oa , such that i belongs to the submodule of Fa
N
generated by H (V, F ) is Oa p . Such N p -tuples from a submodule m
N
(over Oa ) of Oa p and after Theorem 1, we have only to prove that on a
certain fixed open nieghbourhood U of a, every N p -tuple of holomorphic
functions on U is the uniform limit of N p -tuples (b1 , . . . , bN p ) such that
N
Pp
bi i induces at a an element of Fa . But this follows at once from the
1
approximation property.
Proof of Theorem B): We prove first that H p (V, F ) = 0 if p > 1. Let
be a p-cocycle of V. On Km we have = m , m a cochain of Km ,
by Theorem B) for Km (m = 1, 2, . . .). Also, on Km , (m+1 m ) = 0,
so that m+1 m = m where is a (p 2)-cochain of K . By the
m m
definition of cochain, we may suppose that m is the restriction to K
m
of a (p 2)-cochain m of V, and so m = m+1 m on Km , while
(m+1 m ) = on Km+1 . It is clear that by repeating this process
with m = 1, 2, . . . we obtain a (p 1)-cochain of V such that =
and so H p (V, F ) = 0.
Finally, we turn to the proof that H 1 (V, F ) = 0. Let be a 1-
cocycle of V and let = on K p , where p is a 0-cochain of K p .
Again, p+1 p H (K p , F ), i.e., is a cocycle.
By the approximation property, there is a cocycle cp+1 H (V, F ) 125
such that ||cp+1 + p+1 p || p p , where p can be chosen arbitrarily
small. It is clear that we find thus a cochain p of K p with p = on
K p and || p+1 p || p p for every p 1.
If we say that a sequence of cochains, { p }, where p is a cochain of
K p , tends to a cochain of V if p H (K p , F ) and || p || p 0
as p , a repetition of the proof of the approximation property (with
trivial modifications) shows that the sequence { p } defined above tends
to a cochain of V if the p are small enough and it is clear that = .
This proves Theorem B) for p = 1 and the proof of the fundamental

theorem is complete.
Bibliography
[1] H. Cartan: Ideaux de fonctions analytiquesde n variables com- 126

plexes, Annales de lE.N.S. 3e serie, 61 (1944), 149-197.
[2] H. Cartan: Ideaux et modules de fonctions analytiques de variables

complexes, Bull. Soc. Math. de France, 78 (1950) 29-64.
[3] H. Cartan: Seminaire E.N.S. 1951/52 (especially Lectures XV-

XIX).
[4] H. Cartan: Varietes analytiques reelles et varietes analytiques com-

plexes, Bull. Soc. Math. de Frace 85 (1957), 77-99.
[5] J. Dieudonne: Une generalisation des espaces compacts, Jour.

Math. Pure et Appl. (9) 23 (1944), 65-76.
[6] C. H. Dowker: Lectures on Sheaf Theory, Tata Institute of Funda-

mental Research, Bombay, 1956.
[7] K. Oka: Sur les fonctions analytiques des plusieurs variables VII.
Sur quelques notions arithmetiques, Bull. Soc. Math. de France,
78 (1950), 1-27.
[8] J-P. Serre: Faisceaux algebriques coherents, Annals of Math, 61

(1955), 197-278.
Supplementary References
In the papers listed below, the reader will find applications of the the- 127
orems proved in these lectures and several important results that could
129
130 BIBLIOGRAPHY
not be treated here. For further references, see the Scientific report on
the second summer Institute: Several Complex Variables, by W. T. Mar-
tin, S. S. Chern and O. Zariski, Bull. Amer. Math. Soc., 62 (1956),
79-141.
Bibliography

[1] H. J. Bremermann - Uber
die Aquivalenz der pseudo-konvexen
Gebiete und der Holomorphiegebiete in Raum von n komplexen
Varanderlichen, Math. Ann. 128 (1954), 63-91.
[2] H. Cartan - Seminaire E.N.S. 1951/52 (Lecture XX).
[3] H. Cartan - Seminaire E.N.S. 1953/54 (Lecture XVII).
[4] H. Cartan - Espaces fibres analytiques, vol. consacre an Sympo-

sium internat. de Mexico de 1956.
[5] J. Frenkel - Cohomologie non abelienne et espaces fibres. Bull.

Soc. Math. de France, 85 (1957), 135-220.
[6] H. Grauert - Charakterisierung der holomorph vollstandigen kom-

plexen Raume, Math. Ann. 129 (1955), 233-259.
[7] H. Grauert - Approximationssatze fur holomorphe Funktionen mit

Werten in komplexen Raumen, Math. Ann. 133 (1957), 139-159.
[8] H. Grauert - Holomorphe Funktionen mit Werten in komplexen

Lieschen Gruppen, Math. Ann. 133 (1957) 450-472.
[9] H. Grauert - Analytische Faserungen u ber holomorph-

vollstandigen Rauman, To appear in Math. Ann. 1958.
[10] F. Hirzebruch - Neue topologische Methoden in der algebraischen 128

Geometrie, Erg. d. Math. Springer, 1956.
131
132 BIBLIOGRAPHY
[11] F. Norguet - Sur les domaines dholomorphie des fonctions uni-

formes de plusieurs variables complexes. (Passage du local au
global.) Bull. Soc. Math. de France, 82 (1954), 137-159.
[12] K. Oka - Sur les fonctions analytiques de plusieurs variables. VI

Domaines pseudoconvexes, Tohoku Math. J. 49 (1942), 15-52.
[13] K. Oka - Sur les fonctions analytiques de plusieurs variables.

IX Domaines fini sans point critique interieur, Jap. J. Math. 23
(1953/54), 97-155.
[14] R. Remmart - Sur les espaces analytiques holomorphiquementse

parables et holomorphiquement convexes, C.R.Acad. Sci. Paris,
243 (1956), 118-121.
[15] J-P. Serre - Quelques problèmes globaux relatifs aux varietesde

Stein, Colloque de Bruxelles sur les fonctions de plusieurs vari-
ables, (1953), 57-68.

[16] K. Stein - Uberlagerung holomorph-vollstandiger komplexer
Raume, Arch. der Math. 7 (1956), 354-361.

Tifr13 PDF

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Tifr13 PDF

Hochgeladen von

Copyright:

Verfügbare Formate

Lectures on

The Theory of Functions of

Tata Institute of Fundamental Research

Distributed for the

c Tata Institute of Fundamental Research, 1984

ISBN 3-540-12875-1 Springer-Verlag, Berlin. Heidelberg.

No part of this book may be reproduced in any

Printed by M. N. Joshi at The Book Center Limited, Sion East,

1 Cauchys formula and elementary consequences 3

2 Reinhardt domains and circular domains 11

3 Complex analytic manifolds 19

6 Domains of Holomorphy: Convexity Theory 35

7 Convexity Theory (continued) 41

II Differential Properties of the Cube 53

8 d -cohomology on the cube 55

5 The first Cousin problem . . . . . . . . . . . . . . . . . 62

9 Holomorphic Regular Matrices 65

III Coherent Analytic Sheaves 85

12 General properties of Coherent Analytic Sheaves 91

13 Cohomology with coefficients in a sheaf 101

14 Coherent analytic sheaves on a cube 107

15 Stein Manifolds: preliminary results 115

16 Coherent analytic sheaves on a Stein manifold 123

Cauchys formula and

1. Let C n be the space of n complex variables (z1 , . . . , zn ). We write 1

A definition which is equivalent to the above, is t the following:

Definition . A polydisc, with centre at the origin is the set K of points

|z1 | 1 , |z2 | 2 , . . . , |zn | n , 1 , 2 , . . . , n > 0.

These inequalities are, for brevity, written

2 Let denote the set of points z C n for which

Consider a function f (z) holomorhpic in a neighbourhood of K and

Proof. Repeated application of the Cauchy formula for holomorphic

2. If f is a complex valued continuous function on a compact space

(N is the set of non-negative integers).

The series converges normally on K.

The coefficients a j = a j1 ,..., jn in (1) are given by

This follows at once from the expression (2) for aJ as an integral in

Proposition 2. Let { fk (z)} be a sequence of functions holomorphic in

Proof. Since fk (z) f (z) uniformly in a neighbourhood of any z , 5

C is an (F )-space, i.e., it has a countable fundamental system of neigh-

Proposition 3. Every bounded closed set in H is compact. 6

Proof. Let K be any compact subset of . Let K be the set of functions

Theorem of Baire. Let U be an open set Cn and {O}, k = 1, 2, . . . a

Proof of the proposition.

Reinhardt domains and

(z1 , . . . , zn ) implies (ei1 z1 , . . . , ein zn )

for all real 1 , . . . , n .

Theorem 1. Let C n be a connected Reinhardt domain containing

in , the series converging normally in . Such an expansion is unique.

For, if z, we can join z to the origin by a path in . This path has a

The integral is defined, for if (z1 , . . . , zn ) , then ((1 + )z1 , . . . , (1 +

the series being normally convergent on |t j | = 1 + , j = 1, 2, . . . , n.

Exactly as we proved above that g(z) = f (z) in , we prove, using

From the integral representation of , it is seen that if z lies in a 11

2 Domains of convergence of powe series

Consequences of Abels lemma

(a) D = B . This is obvious.

(b) The series represents a holomorphic function in D (by I, Prop 2)

12 (c) D is a Reinhardt domain. More, if (z1 , . . . , zn ) then (1 z1 , . . . ,

and so D is a union of polydiscs (all with centre).

|z1 | < , |z2 | < 1

and the points with

|z1 | < 1, 1 < |z2 | < 1.