F. Gherardelli ( E d.

Complex Analysis
Lectures given at a Summer School of the
Centro Internazionale Matematico Estivo (C.I.M.E.),
held in Bressanone (Bolzano), Italy,
June 3-12, 1973
C.I.M.E. Foundation
c/o Dipartimento di Matematica “U. Dini”
Viale Morgagni n. 67/a
50134 Firenze
Italy
[email protected]

ISBN 978-3-642-10963-8 e-ISBN: 978-3-642-10964-5

DOI:10.1007/978-3-642-10964-5
Springer Heidelberg Dordrecht London New York

©Springer-Verlag Berlin Heidelberg 2010

Reprint of the 1st ed. C.I.M.E., Ed. Cremonese, Roma 1974
With kind permission of C.I.M.E.

Printed on acid-free paper

Springer.com
 CENTRO INTERNAZIONALE MATEMATICO ESTIVO
(C.LM.E.)

I Cielo - Bressanone - dal 3 al 12 Giugno 1973

«COMPLEX ANALYSIS»
Coordinatore: Prof. F. Gherardelli

A. ANDREOTTI: Nine lectures on complex analysis Pag. 1

J. J. KOHN: Propagations of singularities for the Cauchy-

Riemann equations » 177

YUM-TONG SID: The mixed case of the direct image theo-

rem and its applications » 281
 CENTRO INTERNAZIONALE MATEMATICO ESTIVO
HC. I. M . E. )

NINE LECTURES ON COMPLES ANALYSIS

ALDO ANDREOTTI

Corso tenuto a Bressanone dal 3 al 12 giugno 1973

 - 3-

A . Andreotti

Preface.
In the spring of 1972 I h ad the opportunity to lecture at Lund
University and more extensively at Amsterdam University and at the
C.I.M.E. session in the. summer of 1973.on some topics of complex
analysis of my choice. The subject has been chosen within the
limited range of my personal knowledge and is intended for a
non excessively specialized audience. We have tried th erefore
not to obscure the ideas, in the attempt to obtain the most gen-
eral statements, with an excess of technical details; for this
reason, for instance, our main attention is devoted to complex
manifolds, and we have recalled basic facts and definitions when
needed. The purpose was not to overcome the listeners with admir-
ation for the preacher but to share with hime the pleasure of in-
specting some beautiful facets of this field. Indeed I was very
grateful .to receive many valuable suggestions; in particular I am
indebted to L. Garding, L. Hormander, F. Oort, A.J.R.M. van de
Ven and e.pecially to P. de Paepe who undertook the heroic
task of writing the notes.
The material deals with the theory of Levi convexity and its
applications, with the duality theorem of Serre and Malgrange,
and with the Hans Lewy problem. The limited time at our disposal
may account for some conciseness that, however we hope, will
turn to the advantage of the reader.
P. de Paepe has corrected several mistakes of mathematics and
presentation; probably only few remained undected.

San Pellegrino al Cassero, September 1973,

Aldo Andreotti.
 -4 -

A. An d r eot t i

CONTENTS
Chapter I. Elementary theory of holomorphic convexity.
1.1. Preliminaries. 1
1 ..2. Hartogs domains. 4
1.3. Open sets of holomorphy. 5
1.4. Levi-convexity. 12
Chapter II.. Psuedoconcave mani folds.
Preliminaries. 19
Meromorphic functions and holomorphic line 22
bundles.
Pseudoconcave manifolds. 27
Analytic and algebraic dependence of 31
meromorphic functions.
Algebraic fields of meromor phic functions. 35
Chapter III. Properly discontinuous pseudoconc~e groups,
the Siegel modular group.
Preliminaries. 40
I
Pseudoconcave properly discontinuous groups. 43
Siegel modular group. 45
Psuedoconcavity of t he modular group. 47
Poincare series. 52
Chapter IV. Projective .imbeddings of pseudoconc ave
manifolds.
Meausre of ~ s e ud o c o n c avi t y . 56
The problem of projective imbe dding of
pseudoconcave manifolds. 59
Solution of the problem for O-concave
manifolds. 63
The case 0 f dimU; X ~ 3. 67
Chapter V. Meromorphic functions on complex s paces.
5.1. Preliminaries. 71
5:2. Psuedoconcavity for c omplex s p aces. 73
5.3. The Poincare problem. 74
5.4. Relative theorems. 82
 -5-

.~ . Andreotti

Chapter VI. E.E. Levi problem.

6.1. Prelimina1rl.es. 86
6.2. E.E. Levi problem. 92
6.3. Proof of Grauert's theorem. 94
6.4. Characterization of projective algebraic
manifolds. Kod ai.r-at s theorem. 103
Chapter VII . Generalizations of the Levi-problem.
7.1. d-open sets of holomorphy. 111
7.2. Proo f 0 f theorem (Grauert) d • 113
7.3. Finiteness theorems. 118
Applications to projective algebraic
manifolds. 119
Chapter VIII. Duality theorems on complex manifolds.
8.1. Preliminaries. 124
v
8.2. Cech homology on complex manifolds. 128
8.3. Duality between cohomology and homology. 13Q
v
8.4. Cech homology and the functor ElCT. 131
Divisors and Riemann-Roch theorem. 138
Chapter IX. The H. Lewy problem.
9.1. Preliminaries. 147
9.2. Maijer-Vietrois sequence. 150
Bochner theorem. 154
Riemann-Hilbert and Cauchy problem. 155
Cauchy problem as a vanishing theorem
for cohomology. 158
9.6. Non-val1di~ of_Poincare lemma for the
complex. {R(S),d S S • 160
Global results. 163
Bibliography. 166
 - 6-

A . Andreotti

Chapter I. Ele!llent ary t he ory o f hol omor ohi c c onvexi t y .

1.1 Prel imin aries.
a) Let 11 be 'ill o p en s et in lC
n I t he Cartesi an pr oduct
of n co pies o f t he co mplex f ield II: I wit h c oor di n at e funct-
ions Zl"",Zn'
A f unction f: .a. - lC is c al l ed l101omorl) hic if for ev ery
poi nt Z it J1. t here ex i ~ts a neighborhood U( zo ) of Zo in
o
.!l on which f admi t s an ab solutely c onverge nt power s er i es ex-
p ans i on

f or every

Here a = nat ur al number s including

a. 0.
1 d.
n
0), aa...
.1 I'"
.I '
,'"""n
Z = zl ,zn

A map f = (f ••• , f ) : .Ii. ~ a;m i s s aid to be holomorphic

. l, m
if e ach co mnonen t f 1 ~ i ~ m, is hol omorphi c .
i,
The c ompos i tion of two hol omor phic maps i s ( wh ere it is de f i ned )
a hol omor ph i c map.
b) We will write Z = x + iy with x, y ]Rn , i = ~.
1 -
Then x = 2(z+z) 1 -
y = 2i(Z-Z) where t he b ar den ot es compl ex
con jugation, and we will wri te dx = 2(1 -
d z+dz)
l
1
, dy = 21 -
(dz-dz) .
For any fu nction f : .0.-11: of cl ass C we h ave

a! n of
ae = ~=l <3 Zj
d Zj +
jh
z.
a= dZj
where J

af
aZ j = .!(of + .!~)
2 ex j 1 oYj

af .!(~. 1 of )
= 2 oXj - I a
a'Z j Yj
 -7 -

A. Andreotti

we define
n af
af = j~l azj
if = ~
j=l 0-
u
Zj
The following theorem est ablishes a cr i t er i on for a function
1 .
ot class C to be a holomorphic function.
l is holomor-
Theorem. A function f : .n - ; of cl ass C
phic iff at every point of n f satisfies the Cauchy-Riemann
equations:
at
-°
=
(1.e.
at = 0, •.. ,
af
= 0, i.e.
all oZn

for 1 1 j ~ n) •

c) Let f be holomorphic in a neighborhood of the closed

polycil1nder
for 1 ~ j ~ n

then for every Z c ~, the interior 0 f P, we h ave the

Cauchy integral formula

From this tormula it follows easily (by ex pansion of the

kernel of the integral in power series) that a continuous func-
tion t: Jl, - a: which is separately holomorphic in each variable
is a holomorphic function (Osgood's lemma). This is even true
it the condition that t is continuous is removed (Hartogs'
theorem) but t his is much more difficult to prove.
 -8 -

A. .Andreotti

d) We recall the following result:

the set of points where a holomorphic function has a
zero of infinite order is open and closed (principle
of analytic continuation) ..
In particular if f . is defined in.n. , if.n iscon-
nected and if f vanishes at some point of..o. of infinite order
then f is identically zero on n .
e) We denote by H(n) the set of all holomorphic functions
in n. It is a vector space over t.
We can provide /I (Jl) with a locally convex topology defined
by the family of seminorms

= sup Ir ]
K

where K is a compact s ubset of .fL. A fundamental system of

neighborhoods of the origin is then given by the sets

V(K, e) = fr € !I(Jl.)

for K compac t in 11 and E. > O. This topology is the topol-

ogy of uniform convergence on compact subsets of n . _
If Kl C KZ e, K c, • • • is a ae quenc e of compact sets such
3
for m = 1,Z, ••• ,
<S'
O
that -in
K c, K 1
m+ and.D:: m-=1
U Ifm
one easily verifies th at the count able s et of seminorms
defines the swne topology, Thus #(Q) isa metrizable space,
one c an t ake for Lns t anc e as a di s t ance the function

Ilf-gll~
d(f,g) = ~
mel
~
~m
r, g eo HUl.).
1 + Ilf-gll K
:n
Since continuous functions satisfying the Cauchy integral
formula are necessarily holomorphic, it fo l lows th at H (U) is
a complete met r i c sp ace (i.e. a Frechet sp ace) and therefore a
Batre space.
We - also note th at bounded sets B c: /I (il) are relatively
compact. This is a consequence of
 - 9-

A. Andreotti

Vitali's t :leorem: If ffv~ 1s a seguence of ho1omorphic

functions on II. s uch th at for every compact set K c. n ~

exists a constant C(K) for which we have

/lfJI K ~ C(K) v = 1,2, •••

then we can extract from ffu~ a sUbseguence ~ful ~

converges uniformly on any c ompact subset of n .

In particu1ar'the unit ball in the norm 1s re1a-

II II Km+1
compact with respect to the norm II II ~ 1.e. H(fL) is a space
of Frechet-Schwartz (e r, [14] ). For more det~ls we
refer to [28], [31] I [431.

1.2 Hartogs domains.

Consider the subset in m3 (coordin ates x I .YI t ,
Z =x + iy)

T = liz) < b I 0 s, t c e5 \J fa < I z I <. b, 0 ~ t <d 5

where o < a <b and 0 < c < d.
Bec ause
fits sh ape we call T a "top hat ll • A top hat
0

(or Hartogs domain) in II:

n
a: "-II:
n-1
, n 2 2, is the set of all =
points z (zl""'z= n ) in a:n for which (zl' (j~~ I z jI2)t) is
.
contained in a to p hat in m3 •
Theorem (1.2.1). (Hartogs). ~ f be holomorphic in the
top hat

T = ! Iz 1 / <. b , j~2 I Zj/2 < c2 S U {a <. I zi <: b ,

2
j ~2 I Z j I 2" d S
n
in II: , n ~ 2.
~ f ext ends holo morphical1y t o the filled up top hat
n 2 2
j;2 I Z j) <. d 5•
 -10-

A. Andreotti

!:!::221. Let the functions ~, p e Ii, be defined by the

Cauchy integral

Jl f(~, z2' ••• ,zn)

S -zl
lsI =b--p
If P is large ~ is well-defined, continuous and holomor-
phic in each variable for I zll b-l/p and j~2 Zj 12 <: d 2 I
Therefore ~ is holomorphic. Moreover ~~ is independent of
p. Set g = ~, then g is defined in T, is holomorphic
there and glT =f because g-t is holomorphic in T, is zero
on ~ I zll < b, ~ I Zjl2 < c
2
S and T is connected. Q..E.D.
2
h: "T
n
Let - be a biholomorphic map onto an open set
II:
"
h(T) of n
11:, (i.e. h is invertible and h and h- l are both
holomorphic). Then any holomorphic function on h(T) extends
holomorphically to "
h(T).
This is a consequence of Hartogs l theorem and the fact that
the composition of two holomorphic maps is a holomorphic map.
It is sometimes called the "disc theorem".
We quote some "s i mpl e consequences of Hartogs l theorem."
Let n L 2 and f holomorphic on the punctured ball
to ~ ~ Iz j/ 2
< r 2
J then f extends holomorphically to the ball

1 rI Zj/2 < r2S. In tact we can put a top hat T in the

punctured ball so th at T covers the origin. I n particular it

follows that a holomorphic function f in n ~ 2 variables
cannot have an "isolated s ingularity" nor an isolat ed zero
(since this would be an isolated singuaarity for l/f).

li3 Open set. of holomorphYT

a) Open sets with a smooth boundary. An open set D.. in
len has a smooth boundary if for every point Z e. 0 n
o
rt _ [L =
we can find a neighborhood U(zo) and a C 6' function
f: U(z ) • lR such that
o
 -11-

A . Andreotti

This amounts to say that by a local diffeomorphism near Zo

n 0 U(zo) can be transformed into an open subset of a half
space. Indeed we can select a set of real C tJ> coordinates in
which + - +(zo) = xl is the first coordinate. Thus il () U(zo)
is an open subset of the halfspace t. xl c, °S.
Let.Q. be an open subset of a: n (= m 2n) with a smooth boundary
and let zo" on. be a boundary point of.n.. Given f € /I (Q.)
we will say that f is holo morphically extendable over Zo if
we can find a neighborhood V(zo) of Zo and a holomorphic
function f G II (n u V( z »
o
such th at

fIn. = f •

Definition. Let.ft be an open subs et of a:n with a smooth

boundary. We say th at..Cl. is an open set of holo monphy if for
every boundary point zed Il. =
o
Ii. -
(l. we c an find a holomorphic
function fell (rl) which cannot be extended holomorphically
over Zoe

Examples.

1. Every open subset 12 c.. a: wi th smooth boundary is an open

set of holomorphy. Indeed for every Zo 6 »s: f = (z-Zo)-l
is not extend~ble over Zoe
2
2. The ball it = Tj~l IZ j 12 <:. b ~ in a: n is an open set of
-1
holomorphy. Indeed f = (zl - b) is not extend able over
(b,O, ••• ,0). Since the unit ary gr o up U(n) a c t s transitively
on a Q. ' by holomorphic tr ansform ations, the a..s s e r t i on follows.

3. The ' circular shell Il. = 1a 2 c, jt. I zjl 2 <. b

2
$ for

°< a < b, if n ~ 2
is not an ODen s et of holo morphy.
2
Indeed for every point Z of the inner bound ;ITy o~l Iz = a
o " J= Jo)2
we can p l ace a top h at T in [ l such th at zo" T.
 -12-

A . Andreotti

b) Open sets with arbitrary boundary.

By a domain we mean an open connected set.
Let /J c. A be two domains in a: n and let S Co #(A) be a
set 0 f holomorphic functions in A. We sa:y th at ~" is an
S - completion 0 t /J. if

/\
i.e. if every f e S extends holomorphically to 4 .
Note that by the principle of analytic continuation the extension
of f to f £ #(A) is unique.
For instance for a top h at T, '"T is an )/(T)-completion of
! (nn :2).

Definition.
Let n. be an open set in a: n • We say th at D.. is an open set
of holomorphY if:
for every domain Il c, fl every '"
- completion ,1 of
b. is cont sd.ned in .a. ."
Remark: " Op en sets of holomorphy with a smooth bound ary are
necessarily open sets of holomorphy in the sense of this general
definition. We will see later th at the converse is also true.
we will refer for the moment to the definition given before for
o pen sets of holomorphy with a smooth boundary as the
"provisorlal definition of open sets of holomorphy".
c) Holomorphlc convexity. characterization of ope n sets
of holomorphy.
An open set SL in a: n is called holomorphically convex if
for every compact
,.. subset K c. n.. the holomorphically convex
envelope K of K in IL, de f i ned by

K= tZ".n I If(z)/ ~ IIfll K for every t « #((1.) 5,

is also c omp ac t , '
 - 13-

A . Andreotti

Theorem (1.3.1). An open set .D. e, a: n is ho10morphical1y

(1)
convex if for every divergent seguence ~ X,;5 c (l. ~
exists an f 4. !lUI.) such th at

sup /f(X,,) I = + '" ("condition DII).

V

!:!:£.2.!. Condition D implies th at .n. is ho1omorphic all y

convex. Indeed, if K is comp act and K is not we can f ind a
divergent sequence [xvS in n. t £xv} C K •
But then for every f ~ /I(U) , I f(X) I ~ 11 f II K c: <P which
contr adicts condition D.
Conversely let n be ho10morphical1y convex. We want to
show th at condition D holds. Of t his f act ~e will give two
proofs.
1st proof: By absurdity; suppose th at there eXists a diver-
gent sequence ~ xvl c. Q. such th at for every t ~ II (ft)
sup If(X)1 < tP. By passing to a subsequence we may assume
'"
~ xv~ contained in a connected component of 12.. Without lo ss of
generality .we may t hus assume .fl connected. Set

A = 1 fe J.I (Q.) I sup /f(X ) I ~ 1 ~ .'

v "
Then
1I(n.) =m~l
'" m A.

Now A is a closed subset 0 f 1-1 (.0.) • Thus by the Baire category

theorem A must contain an interior point. But A is convex
and symmetric (A = - A) thus A must contain a neighborhood of
the origin say

V(K, z) =[f .. 1I(n.)

( for a comp act K c, n. and some l. > 0). We may assume as well
that K has non-empty interior.
(1) By thiS We mean a sequence [x ,,~ with no a ccumulation point
in .('L •
 -14-

Ii Andreotti

Now for every f il 0, f ~ #(Q), rEA, there-

fore ror every f <1 I/(Q)

sup [r(xv)1 ~~ IIrJl K •

v
In particular, replacing r by ~, we get

sup I fI(x)1 ~
-f.
! IIfll\K
-V
i.e"
1
sup !r(x
v
)1 So (f}i !lfll K •
v
This shows that

sup I f(X Il ) J s I)f 1/ K

forevery fE-lim). Hence tXv~CK. This 'contradictstheholo-
Inorphic convexity 0 f Jl •
nd
2proof: Select a sequence 1~ S of compact subsets of
such that

Let ~ XlJ~ be a divergent sequence in..Q. Repl acing ~ K

m
S and
[Xv5 by subsequences we may a ssume th at

X
m E. Km+l • for m = 1,2 , •••

Since X
m 1 K
m = Km we c an find ~ ~ lI(fIJ such that

Choose positive int egers .A- succ essively so th at

m
 - 1 5-

A. Andreotti

..Am
Now f s £ ~ converges uniformly on every Km and thus on
every compact subset of D.. as any such set is contained in
some Km'
1hus f is a holomorphic function in LL. But from the last
inequality we derive that /f(X
m)
I> m.

Therefore sup
v
I f(x \J
) J s + rI' •

Theore. (1.3.2). (Cartan - Thullen). An open set a G mn

is an open set of holomorphY ifi n is holomorphicallY convex.

f£22!. If J1 is holo morphically convex, then condition D

holds, thus clearly Jl is an open set of holomorphy.
Conversely suppose that n. is an open set of holomorphy. We
want to prove that fL is holomorphically c onvex. If t his is not
the case, then there exists a compact s ubset K Co fL such that "
K
is not compact.
Because for each coordinate function we h ave IlZjJl = IIZj/l~,
K ._
~ c;"
K is bounded. Let 2 xvJ C K be a divergent
sequence in ..Q such that x" ~ Zo ~ (la. Let Jl Z 1/ = sup
denote the polycilindrical norm in m n
, and let

p = polycil1ndric al di s t an c e of K and an."

Certainly p > o. If P s 00 then .n. = m n
which is cl early
holomorphically convex. We may assume p c <p. Let K' be the
set of points in f.l whose polycilindrical distance from K is
~1-P. Then K' is a compact subset of 11.. For every n - tuple
4 of non-negative integers an d every f ~ #(Q)
 A . Andreotti

1 a 1I(f!.) ( 1 )
at D f •
therefore for an.y x e K"

For z ~ K we have by the Cauchy formula

an.d therefore

(2) 1
1~1
(~)
From (1) and (2) it follows that the Taylor seri es of f at
1\
a point x ~ K

t1
1 D f ( x ) (z_x)d
at
is majorized by the series
d.
1\ r II K' ) ;ii~ I
and therefore is abs ol ut el y convergent in Q(x) =~ f1z-x 1\ c P/4S.
Now for v sufficiently large Q(x,,) contains the p ointz •
o
Let A be the connected component of Q(x.) Il n. containing x 1J.

)in we set D
et
=
 -17-

A. Andreotti

Then Q(x'll) is an II(U)/tJ -completion of A. But Q(X iI ) ¢ .a-

ss it contains the point Zoe Therefore Jl cannot be an open
set of holomorphy.
n
Remark. Let .n. II:
be an open set in with a smooth b oundary.
If .Q is an open set of holomorphy then by the Cartan-Thullen
theorem J1 is holomorphically convex and therefore s atisfies
condi tion D. Hence .R is an open set 0
.
f ho lomorphy in the pro-
.

visorial sense. For open sets with a s mooth bound ary the pro-
visorial and the general definition of open "set of holomorphy are
eqUivalent.

1 :4 Levi (1) - convexity;

n
a) Let.a. be an open subset of II: and let ~ :.Q ~ lR be
a C ~ function. At a point a e D.. we consider t he Taylor
expansion of f; with obvious notations for the partial d er i v a-
tives, we h ave,

~ (z) '" ~ (a) + r. d. 4(a)(za-act) + [,;:4(a)(~-8.ct) +

+ t~~j34(a)(zd-ad)(z~-a~)

+ tZ 9 --t>ca)(z -8. )(z -8. )

etp . do a: fl (3

+6 l) q(a)(z -a )(z -8.) + o( IIz-a /13)

d.p a a; ~ 13

Because and because ~ is r eal-valued, we must

have ~ ~( a) '" ~ d.-«. a) d

~,
f (a) '" a--4J( a) ;
df3
aci~- k. a) = Cl-
c{p
4(a) •
d-
In particular the quadratic form

= ta<!13-f(a)
(i)
Eugenio Elia Levi, 1 883 - 1917.
 -18-

A. Andreotti

is hermitian; it is c alled the Levi-form of ~ ata.

A biholomorphic change of coordinates near a acts on
L(~)
a
with a linear change of variables

v ~ J(a)v

where J( a) is the Jacobian matriX of t he change of vari ables

a t a.
It follows th at the number of positive and the number of nega-
tive eigenvalues of the Levi-form at a do~s not depend on the
:hoice of local coordinates.

Remark. If (dp)a ~ 0 we can perform a change of coordinates

lln which a is at the origin and in which the new ~ - coordinate
1a
~ -0 Ha)(z -a ) + tL' 0d ,,40 ( a )( z -a Hz -af.l.).
d dd. r-- del. ~y

Then ~ takes the following Taylor exp ansion:

~ (z) = 4(0) + 2 Re ~ + L(Oo(z) + 0«1 z 11 3 ) •

b) Let us assume th at (d~)a ~ 0 and, for simplicity of

not ations th at a is at t he origin. Set

U =~:z E.a./ 4>(z) < ~(o)

Then au =u - U i s s mooth near a 0 = and the r e al tangent
plane to U at the ori gin is given by

This plane contains the (n-l) - dimensional complex plane with

equ ation

This is called t he analyt i c t an gent plane to cU at a and

will be denot ed by Ta(aU).
 -19-

A. Andreotti

Consider the Levi- form 0 f ~ at a restricted to Ta ( au) ,

-..

We obtain in this way a hermitian form in n-l variables and

again we realize that the number of positive and negative eigen-
values is 1ndependen~ of the choice of local holomorphic
coordinates.
Suppose now th at U is defined in a neighborhood of a by
another C t9 function t with ( dy-) a # 0

U =zz€ alt(z)( t(a)S.

By subtracting constants from <t and i' we may assume that
~(a) .. t( a) .. O. Then either ~ or t can be t a ;:en among a set
of C <!' r eal local coord in at es (c r , 1. 3. a)~ Applying the
Taylor formula with the rest in integral form we r e~ize that in
a neighborhood of a ~ = h'f with haC d' function and
invertible (i.e. h( a) # 0). Since t > 0 where"t.> 0 we
mus t h ave h( a).> O.
Now
a •H =a (h ar + ah.r)
= hoJt + ah.at + Jh.;;>'F +d ah.Y
and therefore

This sho ws th at the signat ure (i.e. the number of positive and
neg ative eigenvalues) of the Levi-form r estrict ed t o t he analytic
t ange nt plane to d U at a i s inde pendent also of the choice of
the defining function ~ f or U near a.
 - 20 -

1I.Andr~o1ti

Proposition (1.4.1). ~ U be an open subset of IC n with a

smooth boundary. At any point a e aU the Levi-fchrm of any
defining function for a U restri cted to the analytic tangent
plane to. oU !i a has a signature which is independent of
local holomorphic coordinates and of the choice of the
defining function.
Let pea) (q(a» be the number of strictly positive
(strictly negative) eigenvalues of L(4')aIT ( d U) . These are
a
biholomorphic invariants of the triple (U, a u, a). Note that
we must have

pea) + q(a) S n - 1 •

As an exercise we can show th at there is an analytic disc of

dimension p

(i.e. the biholomorphic image of the unit ball

DP .. l t c mil I ~ It i I 2 < 1 in;P) such that

'\ ( 0) .. a

Analogously there is an analytic disc t$ : Dq ~ a: n of dimension

q such that
/S(O) = a
6"' (Dq) -1 aJ c=. U
Indeed we can moose coordinates at the origin such that

with
 - 21-

A. Andreotti

.A.
j

where the j's are ) 0 •

Therefore near 0, for a > 0 sufficiently small, if

o (lz 2 J2 + ••• + Izp+ 112<. £. and ~. Zp+2 = ..• • zn = 0

then Hz) 0 • >

. This proves the first statement. The second one is proved with
a similar argument.

c) Theorem (1. 4. 2). (E. E. Levi (36]) • ~ .Q be an open

set of holomorphY with a smooth boundary. Then the Levi-form at
each boundary point restricted to the analytic tangent plane is
~ositive semidefinite.

~. Assume, if possible, that L(<$)oIT (an.) has a negative

o
eigenvalue at the point oe<3D., + being a defining function
for -en with oj> (0) = o. By suitable choice of the holomorphic
aoordinates we may write near 0

n
, (z) = 2 Re zl (1 + j;l

:B':l.rst restrict ~ to nr =lIm zl = 0, z3 = ••• =zn=OS.

There exists e >0 such that for (\ z II <. 2e on the region

lR 3 (l ~ Xl s, 0 ~ , (zl = Xl + iYl)' we have

Therefore: for e SUfficiently s mall, 4> < 0 on the discs

,i.e. D C Q..
r
 -22-

A . Andreotti

Also if £. is sufficiently small,

Zl ,= 0, ~('
z l zzl c e , z3 = ••• = zn = OSC:: Q.
Hence there exists a , 0<:'0<:' e , such that

A =2" ~ c )z21 <. 8, I zl l2 + I z3/ 2 + ••• + Iz)2 c ~5C::

n
a,
and there eXists '1, 0 <: 'l..£. 8, such th at

Let A LJ B = 1J. and let

I zll2 + i Z3/ 2 + ••• + ~ ..

By the disc-the!?rem A is an /I em IA -completion 0 r J. But "4
cont~ns the origin 0 In., thus Q is not an open set of
holomorphy.

It is natural to ask if the above necessary condition for an

open set.Q in a: n with a smooth bound ary to be an open set a:
holomorphy is also sufficient (Levi-problem). The answer is
n
affirmative for open sets in a: but not for open sets on c om-
op:L.ex manifolds. We will r (turn later to this question.

Exercises.

I.' Prove that every convex domain in a: n is a domain of

holomorphy.

2. Suppose that Q. has a smooth bound ary and th at a t a point

~ • aa the Levi-form restricted to the analytic tangent plane
at a to d n. is strictly positive. Prove th at we can choose
lo~al holomorphic coordinates at a such that JQ is locally
elementary convex at a.
 -23-

A. Andreotti

Hint: we can replace the defining function ~ by an increasing

convex func tiont 0 f ~ so th at L(V)a is strictly positive
(for instance take t = e Cf -wit h c » 0, see I 23J, p. 263)
and then use the remark in a).
3. Under the same assumption of the previous exercise, prove
that there is a fundamental system of neighborhoods B(a) of
a which are domains of holomorphy such that B( a) (l Q is an
open set of holomorphy.
Hint: in the above specified local coordinates take for B(a)
any small coordinate ball with center in a, then apply the
first exercise.

The material of this chapter is covered in all standard books

on complex analysis as [ 28], [ 31].
 -24-

A. Andreotti

Chapter II. Pseudoconc ave In ani folds.

2.1 Preliminaries •
• ) Presheaves. A presheaf on a topological space X is a
contravariant functor from the category of open subsets U of
X to the category of abelian groups i.e.
for every U an abelian group 5(U) is given and
for every inclusion of open sets V C U a homomorphism

r UV 5(U) ~ 5(V)
is given such that for every chain of inclusions
Wc.Vc=.U of open subsets of X we have
V U
r UW
r W0 r V =

A presheaf ;) = 5(U) is called a ~ if for every

open set n c:.. X- and every open covering ru. = ~ Ui1 iG I 0 f
the following sequence iS lexact

o ~ 5(.Q)

where e is defined by
(l
= r U (f) f E 5 (0..)
i
and where cS is de fi ned by

Example:
~ = fHomcont (U, D:), r Uv 5, where Homcont (U, lC) denotes
the space of continuous functions on U with values in lC and
where r UV are the natural restriction maps, is a presheaf and
also a sheaf ..
 -25-

A . Andreotti

In a similar way one defines sheaves of rings and also sheaves

of modules over a she af of rings.

b) a ~ ~ over X of abelian gro ups is the data of a

topological space T ,
a .continuous surjective map ~: ~ ~ X
such that
d) tt is a lac al homeomorphism 1. e. every point #
.L
rz::
cJ has
an open neighborhood s = s(f) s uch th at w! s is ~ homeomor-

or
phism of s onto an open subset of X
~) for each p oint x eo X, x = 1/"-l(x) has the structure
of an abelian group in such a way that the map
~ l(
X
1J' (1) ~ t
given by

is continuous ..

Given a stack (r, ir, X) of abelian groups, for every open

set Uc X we can consider the abe l i an group

r( n, 'f) =l B : U 7 Tis continuous, tr 0 B = i d e ntity on

uS
of all "Bections" s over U. If V e, U , t he n atural
restriction map
U
r V: 7 r (V, 'Jl is defined and one
obtains in this way a presh eaf which is also a Bheaf.

Conversely, given a presheaf 3 =t 5(U) U

; r V S one can
as s oc i a t e t o it a e t ack (1', rr, X) as followB.
We set for every x.. X
~ = lim 5(U) , i.e. an element of
x U~X

1i is a cl .a a of equivalence of couples (U, r) with

or
:r. r
(1) the "fibered p r od uc t " '!.><.X is d e f i n ed as the part of
lying above the diagonal b. of Xx X by p ro j e c t i on
1I.f1: 'j.T 7 Xx.X ;

j " ter = (1r ,, 17' ) - 1 (/.1).

 - 2 6-

A . Andr eotti

x e U, f e- S(U) under t he r el ation

(U1' f l) 'U
( U2 ' f 2)

if th ere ex ists U " x, U c "i () U2 s uch t hat

3 3
Ul U2
r fl = r U f2
U
3 3
:r x o!
The equivalence cl ass in or (U; f) is den ot ed by
and it is called the germ of t at x,
We then define cr - _ U C'r'
x~xj x and 11 by T1 (
r;-?
:J x) = x ,
If we t ake on ~ as ~ basis for open sets the s ets of the
form f or all f E S(U), we obt ain, as one verifies,
a stack of abeli an groups ('r, fi", X).

st arting in t his construction with a sheaf, constructing the

corresponding st ac k and t hen the corresponding s heaf of sections
we ge t back the original s he af. We t hus h ave a one~to-one
correspondence between sh e aves of abeli an groups and st acks of
a be lian groups. Although this could ge nerate some confusion it
is cu stomary to "r ep r es ent a s he a! by the associated s t ack (see
for i ns t an c e [25] and r
30] or [18J).
c) Meromorphic functions. Let now X be a complex man i f ol d
and let e- be t he she a! 0 f germs of holomorphic f unctions on
X. For every open set U c X it i s defined by t he s p ace II ( U)
and the natural restriction maps. The sp ac e /I (U) i s a r ing.
Let D(U) be the aubae t of I-I(U) of di vis or s of zero, i. e .
D(U) is the set of those holomorphic functi ons on U v anishing
on some c onn ected c omponent of U. Let cr(U) be the quotient
r ing of //(U) wi t h respect to D (u) i.e. ~(U) i s the set
of quotients ~ with f e- II(U), g e- #(U) - [.J(U) with obvious
i dent ific ations:

! fl
g = gl iff fg ' = f Ig
 -27 -

A . Andreotti

It Vc U 1s an inclusion of open sets, the restriction map

U
r V: 1/ (U) ~ ..r(V) sends #(U) - O(U) into #(V) - 0 (V) and
thus induces a homomorphism of rings
U
r V: t$ (U) -i" e;t (V) ,
We obtain in this way a presheat. The corresponding sheaf;?/( is
called the sheaf of germs of meromorphic functions on X. the ring
':it(X) :: [(X,?11 is called the ring. of meromorphic functions on
on X. Note that ~ (X) E:. ?t(X) but' q?(X) may be actually
amaller than %(X).
Example: Take X:: F (D:), the Riemann sphere. Then #.(X) :: I:
l
thus 4(X) :: a: while /(X) is isomorphic to the field of all
rationali'functions in one variable t, '/t'(X) ::: lC(t).
If X is connected then ./t (X) and 6? (X) are fields.
In the sequel we will always assume that X is a connected
manifold.

2.2. Mermorphic functions and holomorphic line bundles.

a) Holomorphic line bundles. Let X be a complex manifold;
by a holomorphic line bundle on X we mean a triple (F, 71', X)
where F is a complex manifold, TT: F -." X a holomorphic
surjective map such that
i) ff is of maximal rank
ii) for every x e X ,7-1(X) is isomorphic to the complex
field II: in such a was th at
q) the map
F "x F -;> F

given by (u, v) -;> v+v is ho1omorphic

13) the map
D:xF-;.F
given by (A, v)-;:>Av is ho1omorphic.
Given two ho1omorphic line bundles (F, IT, X), (E,,,,-', X) over
X a mOrphism (or bundle map) is a ho10morphic map
f : F -7 E such that
L) ff:: W 0 f
 -28-

A. Andre.atii.

1i) for every x IS X the induced map f n--l(x) ~u:-l(x)'

x:
is a:-linear.
A holomorphic line bundle (F, ff, X) is said to be trivial if
it is isomorphic to the bundle (X ~ a:~ prX ' X).
Every holomorphic line bundle is locally trivial (as it
follows from the implicit function theorem). Therefore there
exist an open covering 'U = U 0 f X and biholomorphic maps
i

such that j -1 (x, y) = x,

fr0'f i On U
i
() U j

we have two trivializations of F and thus

~i .. -1 ( x,
0 ~ j v ) = ( x, gij(X) v)
with
(1) is holomorphic and never zero

(2) On Ui n Uj n Uk we must have gij gjk = gik (consis-

tency con ditions). The collection of f unctions ~ gij ~ are
called t he transition fu nctions of t he bundl e F (rel at ive to
th e l oc al trivi aliz ations 4i).
Conversely, gi ven on an open covering 2( = 1Ui $ a system of
transition f uncti ons (1) s atisfying the consi s t ency c ondi t i ons
(2) one can construct a holomorphic line bundl e wi t h l oc al tri-
vializations on the s ets U having t he given s yat.e ms as a
i
sys t em of tr ansition fUllctions.
Two holomorphic line bundl es gi ve n on t he s mae covering
U = U1 by. tr ansit i on f unctio ns ~ gi jS , f f i j are isomorphic5
i f th ere exist holomorphi c maps
Ai U ~ lC*
i s uch th at

on
 -29-

A. Andreotti

Given a holomorphic line bundle ( f, zr , X) we c an consider

t he s p ace of holomorphic sections
r (Xi F) =l s : X ~ F l 5 = i dX S•
ho LomorphLc , tro s

In terms of local trivializations of F on the covering ~ = Ui

a holomorphic section is given by a collection

of holomorphic functions such th at

(cf. 30 ).

b) Given two holomorphic sections 51 = ~ sl1~ and So = f 80i1

of the bundle F, if 8 is not i dentically ze r o on any open
0
s et we can construct a meromorphic function

on X, given locally by

In this way we e an obtain all meromorphic f unctions of /t'( x)

indeed we have
Proposition (2.2.1). Every mero morphic func tion m ~
complex manifold X is the quotient of two holomorphic sections
of an appropri ate holomorphic line bundle on X.
f!22!. For every point x ~ X we can find a nei ghbor hood
V such th at

m\ V =~ with p, q 6 !I(v) , q Ei )/(V) - D(V).

Since. the ring (7 x is a unique f actorization domain (c r, [28]),
if we take V s Ufficien tly small we may ass ume th ~t the germs
Px and llx of p and q at x are coprime. But if Px and
llx are c oprime and if V is sUfficiently s mall, then also the
germs Py and qy of p and q at any point y € V are
co prime (cf. [29], [48]).
 -30-

A. Andreotti

Let fV(xi)] i€I be a covering of X with such neighborhoods.

Then on V(X
i)

Pi
qi =~j i.e. Piqj = Pjqi

By the Euclid lemma Pi must divide Pj and Pj must divide

Pi' i.e.

Pi = gij Pj with gij a unit in H(V ) (l H(V )

i j

this means th at gij

i)
n V(X j) is holomorphic and never
on V(x
z ero. It follows then th at we also have

qi = gij qj
Moreover on V(X
i)
(l vex j) n V(Xk ) we must have

= 1

This shows th ,t the coll ection ~. gij5 is a s e t of transition

functions of a holo mor ~h ic line bunJle F over X, the collect-
ion 1 Pi! gives CI. holo mor phic s e c t i on "i of F and the
coll ection r qi ~ giv es Do holomorphic s e c tion of F with
So not i de ntically zero on ill1y open set. So
51
m as r equired.
So
c) Let (F, 7r, X) be a holo morphic line b un dl,e on X gi ve n
on a covering U = ~ u~ i"I of X by tr an s it ion functi ons f gil>.

One c an c onsider t "e Itl-th tensor p owo r' of Fit, (~, 'ill' X)
which i s gi v e n by t h e t ran s it ion fun e tions t gij)
1 '2 •
 -31 -

.A. .Al1dreott i

We can then consi der t h e grad ed ring

/!.(X, F) l~ reX, = rl)

of the holomorphic sections of the different te nsor po wers of F
(F
O
=
trivi al bundle). Note th at i f s € t E (X , ~) rex, rl), r
then st £ reX, r~). If X is connec ted, as we always assume,
then J(X. F) is an integral domain and one can consider the
field of quotieits
flex, F) =f..!1
So sl' so£ [(X, rl) - f or some 1, so'; 01.
We have
~ (X, F) c ~(X),
in particul ar <9 (X) = tJ (X, trivial bundle).

Theorem (2.2.2). For every holomorphic liBe bundle F the field

~(X, F) is an algebraically closed sub field of::Jf (X) (X ccnnec t .ed) ,

~. Let h c ;{(X) be al gebraic over t? (X, F) i .e. h

satisfies an equation
v
h
v- l
h k_
+~ + • • • + k v -"' O
si li
where ki E. Q(X, F). Let ki Z t ~th si' ti E rex, F );
i
multiplying the above equation by i~ t i we·.obtain &n ell..u~t1on
~ v-l
dOh + 0 h + • • • + 0v;: 0
1 ,)

where 0i [(X,
£ rl)
for a suitable 1 (1 i~1 1 i) and where =
dO ~ O. After multiplication by d~-1 the above equqtion can
be written as follows:
'"'
(doh) + dl(dOh)
~-1
+ •• • + dO
J-1
dy =O.
At each point x e, X dOh satisfies an equation with holomorphic
coefficients and with the coefficient of the highest power equal
to one . This shows that dOh is meromorphic at x and integral
over B x; since t) x is integrally closed dOh .. r must be
holomorphic at x.
Hence 60 IE reX, rl) and also doh'" r E I'(x, rl), thus

c$(X, F).
 A.Andreotti

a h
thus = ...Q...
a 6 q (X, F).
O

2.3 Pseudoconcave manifolds. A connected complex manifold

X is called pseudoconcave if we can find a non-empty open sub-
s et Y X with the following properties
i) Y is relatively compact in X, Y e c, X.
ii) aY = I-Y is smooth and the Levi form of a Y restricted
to the analytic tangent plane has at least one negative eigen-
value at each point of d y(l) •
In particular for any point Z o € 8Y there is an analytic disc
of dimension ~ 1 which is tangent at Zo to d Y and is con-
:;t ained in Y except for the point Zoe
Examples.
1) Every compact connected manifold is pseudoconcave (take
y = X then ~Y = ~ thus condition ii) is void).
2) Let Z be a compact connected manifold of dimE Z 2 2. Let
fal, ••• , ~} be a finite subset of Z. Then X = Z - r~, ... ,am~
is pseudoconcave (take for Y the complement of a set of dis-
joint coordinate balls centered at the points ~).
3) Not every pseudoconcave manifold is compactibiable (i.e.
isomorphic to an open subset of a compact manifold).
For instance if we take P2(a:) - '{O}~ a:2 - to
3 and i f zIt Zz
are the holomorphic coordinates on a:2 , we can consider the
exterior form

and define a function f to be holomorphic if it satisfi es the

(1)
As usual the defining function for ay is chosen so that
it is < 0 on Y and > 0 outside of Y.
 -33-

A Andreotti

differential equation

df A l' e, = 0
In this way we define a complex structure on ~2
(which - lO~
agrees with the natural one if e =
0). One can show that this
complex structure can be extended to JP2 ( £ ) - [0 S, that if
£.# 0 is small i t provides P2(1I:) -fO~ with a pe eudoconc ave
structure and that it is not compactifiable (cf. section 4.4
and (101).'
Remark: Every holomorphic function on a pseudoconcave mani-
fold is constant.

In fact let. l' be holomorphic and non-constant on X and let

Zo .. 'tsuch that If(z ) I = sup I fl. By the maximum modulus
o y
principle, zo~ () Y. If D is a I-dimensional disc tangent to
3Y at Zo and except Zo contained in Y then 11'1 D has a
maximum on an interior point of D. Thus l' is constant on D
and thus there is an interior point zl of Y such that
!f(zl)l = I fez ) I
o
=
sup 11'1. This is a contradiction.
!
In particular a pseudoconcave manifold X (not reduced to a
single point) cannot be isomorphic to any locally closed sub-
manifold of numeric al space lI:N (oth erwise th ere will be a
polynomial on lI:N inducing on X a non-constant holomorphic
fun ction) •
More generally one can prove the fol lowing
Theorem (2.3.1). For any holomorphic line bundle F ~

pseudoconc ave mani fold X we have

dim~ f(X, F) < &.

~ (2.3.2).
Let F be a holomorphic line bundl e over a
pseudoconcave manifold X. There exists a finite number of points
~I"" ~ !.!:. X and an i nte gerh = h(F) s uch that if
s I<: ['(X, F) vanishes at each point ai of order L h lli!!. s= O.
 -34-

A . Andreotti

f!22!. Let Y be as 1n the definition of pseudoconcave

man1folds. For every point x Q , we can choose a coordinate
polycilinder Px' coordinates P1' 1 .. 1, ••• , n, with center x
and of radius r x such that
1) FI~ 1s tr1vial

11) .s(P~)=~YE:U I
l p1( Y) - Pi ( x ) / = r x , l~i.s.nSc:.Y
where U is the coordinate patch on which P1 are coordin ates.
Th1s 1s poss1ble in view of the pseudoconcavity of Y.
Let Px' be the concentr1c polyci11n~er to P with
-1 x
radius rxe •
We can select a f1n1te number of po1nts ~, ••• , ~ such that
111) u p~ i» ,
1
Let F be given by trans1t1on functions :

f1 j 15 ~ () l5aj ~ 11:-

a nd set

II Ff I = sup sup I f 1 j I .. e tJ •
1,j 15 n P
~ aj

Note th at since -1 we must have ~

i j .. f j 1
f O. Z
Now choose an 1nteger h with h > fJ, for inst ance h =[flJ + 1
where [}J] denotes the integral part of ')J •
Let s ~ reX, F), vanishing a t the points ~ 0 f order Z h.
The section s is given b y holomorphic functi ons si : p~ ~a:.
We- set
M = sup
1

There eXi sts a point z0 f. sO; a ) , for s ome ai ' s uch that
1 0
0

lSi
0
(Zo) I = M
 -35-

A . Andreotti

(indeed S(Pa ) is the Silov boundary of Pa ).

i i

Since zo € Y there eXists a P' containin g zo. Certainly

aj
jo ~ i o and we will have 0

Therefore

By Schwarz's lemma

where

the functions Pi being the coordinates on Hence

H s: II F II H e -h =f -h H •

Since ;t -h < 0 we must have H =0 which implies s ; O.

Proof or theorem (2.3.1).

'!he natural map rex, F) ....
"7" a
k

•
" a
h
i

m.~
which associates to each section s e r (X, F) the Taylor expan-
sion of s up to order h - 1 at each poi nt ~, is an in-
jective map by the previous lemma. The righthand space is a
n+h
finite-dimensional vector space over ~ (dimension ~ k( h)}.

Remark. Let X be pseudoconcave and Y ~ X as in the

definit ion. Then f(X, F) ~ r(y, F) is injective. Using
 - 30-

A . Andr eotti

Hartogs' theorem and t he p s eud oconc avi ty of Y we c an construct

....
an open neighborhood Y of Y such th at th e re striction map
y ~
r Y :r(y, r) ~ fey, F) i s an i s omorphi s m. Now r Y i s a
Y
c omp act map for t he Frechet topology of J (Y, F) and fey . F).
~us the Frec het sp ace f (Y. F) is l oc al ly co mp act and t herefore
f inite-dimensional. This wo uld give a mo r e direct p r oo f of
theorem (2. 3.1). , However t he p revious proof has the mer i t to
gi ve an estimate for the dimension 0 f ( X. F) .r which will be
useful i n the sequel.

2.4.
Analytic and algebraic d epend enc e of mer omorp hi c
functions.
Let X be a connected complex manifold. Let f 1 .... , f k <!: ,reX) .
We s ay th at these meromorphic f unctions are an al yti cally depend -
!!!lit
wherever t hi s i s d efi ned .

In oth2r words f l ••••• f are analyt i cal ly d ependent i f at any

k
point whe r e e ac h one o f t hese f unc t i ons i s holomorphic the
aUp .•.• f k )
Jacobian 1'\ ( ) with resp ect t o a s ys tem zl ' · · •• zn
o zl.···. zn
.J f loc al ho l omor phi c coo r di nat es. h 8.S r ank -< k ,
The mer omo rphi c f unctio ns fl ' • • •• f k are s aid to be alge br ai-
call y depend en t i f th ere ex i sts a non- id entically zero polynomi al
p(x l •• ••• x k ) i n k variabl es and with co mpl ex co effic i ents
s uc h th at
wherever i t i s d e fi ned .

Algebr aic dependence i mnli es anal yti c dep end en c e . I n f act if

k > n = dima: . X th ere is nothing t o prove. As s ume k.s...n .
Wi t hout 10 5s 0 f gen er alit y we may al s o as s ume th at fl' ••• ,fk _ l
are al ge brai cally i nde pendent. Let p ( x l • ••• • xk ) be a
polynomial j1 0 of mi ni mal degr e e in x s uch t hat
k
p ( f l' .... f k ) = O.
 - 3 7-

A. Andreotti

Differenti atin g this i ~e ntity we get

But UX (f) ~ 0, thus we get a non-trivial line ar r elation

a k
between the differentials d f i in an open dense subset of X.
ibis implies th At dfl /I ••• A dfk ; 0 wherever defined 'on X.
The converse, of this statement (except for k = 1) is not
true in general. For instance the functions
Zs
fs(x) =
e , s =
l,2,3, •• ~ in ~(=) are all al gebraically
:lmdependent while any two of them are analytically dep endent .
The converse is however true for pseudoconc ave manifolds; we
have in fact the following

Theorem (2.4.1). Let S be a pseudoconc ave mani f ol d . !!

f1' ••• ' f k , f E ?(X) are.. analytically dependent then they are
also algebraically dependent (i.e. on pseudoconc ave mani f ol ds
: ~ al y t i c depe. dence =
algebraic dependence).

Proof. ' It is not restrictive to assume th at f l, ••• , f are

k
analytically independent. Otherwise r epl ace f ••• , f by a
l, k
maximal subset of f , ••• , f , f
l k of an a ~ytically independe nt

functions and tWte for f one of the remaining functions.

ibere exists a holomorphic line bundle F on X and hol omorphi c
sections such that

Indeed for each fi t here exists a holomorphic line bundle Fi

and holomorphic sections t(i) with t(i) ~ 0 such xhat

1 0
t(i)
1
= :tIT
t 0
 - 38 -

lJ. . .An d r eotti

k
t(i) Eo r ( X, F)
Taking F = Fl ••• F
k th en So = i1J"l 0
and So J! O.

Moreover (1) t(i)

si = to • • • 1 t~k)€ r (X, F)

and we have

We can choose a covering of Y by coprdinate concentric

polycilinders P ~ p' 1 ~i ~ N, as in the lem ma (2.3.2)
ai ~
such that
i) Fl p i s trivial
a
i
ii) S(15 ) Co Y
ai
11i) u p' .=:> Y
~
1v) at e ach point a t he f unctions f l, •• • , f k are
1
f l - fl(ai) = e, 1 , •• • , f k - fk(~)= ~ k
(1 ) (i)
holomorphic and
can be t aken among a s et of local holomorphic
coordin ates ;
This can be done by s mall tr anslations i n t he co ordinate patches
of the polyc11inders P a .:;J P'a as condi tions i), ii), iii)
i 1
are not effected by these transl ations and as t he s e t of points
where condition iv) c annot be satisfi ed has an ope n dense
complement :
Also t he r e exists a holomorphic l i ne bundle G on X and
holomorphic sections a 0' 1 e r (X • G) wi th 0- 0 _ 0 such that

we may ass ume t h at

v)
GJ Pa is trivi al
i
vi) f is holomorphic at each p oi nt ai •
 -39-

A. Andreotti

As in the lemma (2.3.2) we define

1/~/1 = ek (where ~ =F • •• F (k times»

IIG 1\ = e
c,)

Consider a generic polynomial in k+l variabl es of d e gr ee r

in each one of the variables xl"'" xk and of degree s
in x k +l '
c1 <i d
=£ c... 1
xl ••• x k
k x k+l
k~
~l' ••• , d k' ex k+l

where 0 s; q i So. r for 1 $. i So k and 0 s, d k+l s, e , Let

be the correspondin g homogeneous polynomial.

These polynomials form a vector sp ace W(r, s) over t of
dimension (r+l)k(s+l).
Now note th at

~r
S
1T(sO' sp.'" sk' 0' l)€ f(X, • G ) •

The theorem · will be proved if we s how th at Ker e # fOs,

where E. is the above natural line ar map W(r,s) ~ [(X, ~r.Gs).
Let h be the smallest integer > kr}J + s w. The map which
a ssociates to (1(sO, •••• sk' 6 0-
1
) th e Taylor exp ansion up
0,
to order h-l of t he function P(f l, •••• f k , f) at each point
a gives a line 3 I' map :
i
N
lm e- 8-
where /1(.~ is t he maxi mal i de al of t he loc al r ing

a:zt;ii) •...• S ~i) 5 of convergent po wer s eri es in t he vari ables

S (i). By Le.nma t his map is i n j ec t i ve .
 -40-

A.Andreotti

Now the target apace has a dimension

r
k r Jl + Bwl + l+It)
6=N
k
= N «(krH + :c..lJ + l+k ) ( [krp + awl + 1 + It - 1)
k-I

N«(krp + aw) + 2)1t

So Nkkpkrk + lower order terms in r.

If we select a such that

s+l) Nkltl t

then, if r is sufficiently large, we get

dima; W( r, s) :> dima: Im l!. and there fore Ker £, .;. o.

2.5. Algebraic fields of meromorphic functions.

a) By an algebraic field of transcendence degree d we
mean a finite algebraic extension of the field D:(t ••• , t d)
l,
of all rational functions in d variables. Since the ground
field D: is of characteristic zero, any extension of this kind
is primitive so is of the form ~(tl"'" t d ' Q) with Q
algebraic over D:(t ••• , t d). Let P(t l, ... , t d, t) = 0 be
l,
the minimal equation for t = Q over D:(tl, ••• , t d). Chasing
denominators and dividing off any factor in the variables
t ••• , t d only, we may assume th at P ia a polynomial in all
l,
the variables and th at it is irreducible. If V is the
d
al g ebraic, v :u-iety defined in a: +l where t ••• , t d, t are
l,
are coordin ates by the equ ation
P(t l, ••• , t d, t) = 0 t
then V is an irreducible variety and the field a:(tl, ... ,t d, Q)
is isomorphic to the field of r ational functions on V.
Moreover d = di ma: V.
 -41-

.A• Andreotti

We want to prove the followin~

Theorem (2.5.1). On a pseudoconcave manifold X of complex

dimension n, the field /f(X) of all meromorphic functions is
an algebraic field of transcendence degree d s.. n ,
That the transcendence degree of ?z: (X) cannot exceed dimlf
tollowsalready from the fact th at on pseudoconcave manifolds
algebraic and analytic dependence are the same (theorem (2.4.1».
The remaining part of the theorem is a consequence of

Proposition (2.5.2). ~ X be a pseudoconcave manifold and

III t1' •• " t k E Jt'(X) be algebraically independent. ~
exists an integer v = v(t1' •••• f k) such th at any f E ~(X)
which is algebraically dependent on t l, ••• , f , satisfies a
k
non-trivial equation over 11:( f l, ••• , f k) 0 f degree s V •

~. We tollow the proof of theorem (2.4.1).

First we tind a holomorphic line bun dle F and holomorphic
sections So _.0, sl' •• " sk of F such that

si
f
i =So
-

Secondly we find coordinate polycilinders P ~ pI • 1 ~ i ~ Nt

ai ~
such that
i) F restricted to a neighborhood of is ~ is trivial

ii)

11i) U pI z» Y
~

~ (i)
k
are holomorphic and can be taken among a set of local
holomorphic coordinates.
 -42-.

A. Andrec tti

Thirdly, since the conditions i), ii), iii), iv) remain

valid by small translations of P a within its coordin ate p atch,
i
we may de~ermine for each ~ a small clos ed neighborhood v(~)
so th at no matter how we translate the center a.... of P a on
i
a point of V(a the above four conditions remain valid.
i)
Let Q be the union of the translates of P a just considered
i i
and let us compute II F II with respect to the covering
lQ~ l~i~N .'
Finally, from the proof of theorem (2.2.1) we realize that
there exists a holomorphic line bundle G and two holomorphic
sections ""0" 0,0"1 of G such that f ="-1dt1
and satisfying
the following condition

v) Gl~ is trivial.

We set

and we choose

Then v depends only on f 1 , •• " f but not on f. We also

k
de fine, with respect to the covering l ~~ t

II Gil = e
and we choose the centers of P in V(~) so that at
~
also f is holomorphic, 1 ~ i ~ No'

We can now proceed as in the proof of theorem (2.4.1) and we

realize that if r is sufficiently large then f s atisfi es a
non-trivial equation over 11:( f 1, ••• , f k) of degree ~ v.
Proof of theorem (2.5.1). Let fIt ••• , f k be a maximal set
of algebraically independent meromorphic functions. Let
f e ~(X) be so chosen th at its degree over lI:(f1, ••• , f is
k)
maximal. This is possible by virtue of proposition (2.5.2).
We claim that
 A. Andreotti

Clearly 1I:(f1"'"
k,
f f) C ?( (X). Let h ~ '1«X) , we can
find Q E 5t (X) such th at
.I

lI:(fl"'" fk , r, h) = lI:(f l,· .. , f k , Q).

Then
c( 2 [lI:(fl''''' f k, Q) : 1I:(f1''''' f k) =
11:( fl"'" f k, Q) : 1I:(f1''''' f k, f». [II:( ri , .•• , f k, f) :

1I:(f1"'" f k )1 •

But the second factor of this product equals ct; therefore the
first factor equals 1. This means that h £ a:( f • •• , f f)
l, k,
and thus our contention is or ov e d .

Theorem (2.5.3). Let X be a pseudoconcave manifold and let

l' : X - !N(a:) be a holomornhic map of rank n = d i maf at some
point of X. ~ Im ') is contained in an i rreducible alge-
braic variety Y of the s ame dimensions than X.

~. Let Y be the smallest algebraic subvariety of

IN(a:) containing (X). Certainly Y exists and is i rreduc-
ible, it is defined by the homogeneous prime ideal

~Y = ?p Eo II: [Zo"'" zN] r pOl = °S

where a:[zo"'" ~1 denotes the gr ad e d ring of homo geneous
polynomials on !N~~?'

Let !?. (y) be the field 0 f r ational fu nctions on Y. Any

element f ~ I(Y) is represented as a quoti ent of t wo homo gen-
eous polynomials f = ~q with q t ~Y' If f = ~q = £:
q
then

pq' - p'q (, rf\. This s hows th at f OJ is a well d e f i n e d

mero morphic function on X. we h ave t herefore d e f i ned a,
necessarily inj ective, ho momorphism

t* : R(y) ~ .::{(X).
 -44-

A...An.d.r.e.otti

Now
di mlCY = tr ansende nce d e gre ~ of R (y) _ t .r-anecend enc e
de gr ee 0 f ~( X) s dimi L But di mlC Y i.. dima: I(X ) = 'iim~ by t he
as s ump t i on ab out t he r-ank 0 f the map T".
In part i cu lar every co nnected complex submanifol d of IN(~) is
a pr o j ec tive al gebraic variety (Chow - t heorem).

b) Excercises.
1. Let A be a pu r e- dimens i onal non-sin g ul "~ al ge brai c s ub-
v ariety of lPn(lC). Let ~ = di mlCA . Prove t h J.t A h as a f unda-
ment al system of neighborhoods V(A) in lPn(~) with a smooth
bound ary at which the Levi-form r estricted t o t he analy t i c
tangent pl ane has at least a n e g ~t ive eigenvalues. Let B be a
connected co mplex submanifold of V( A) with di ma:B + a 1 n+l.
Prove th at B is contain ed in an irred uci bl e al ge br ai c subvariety
of lPn(lC) of the s ame dimensi on then B. (e r. (1 51).

2. Prove th at any pse udoconc ave co mplex Lie group is a complex

torus ([ 6]).
3. Let K be the canonical bun dl e o f t he ps eu doc on c ave mani fol d
X ; d e fi ne the "c anonic al dimen s i on of X" as t he tr anscendence
de gr ee of lj (X :, K ) . Prove t h at 0 s, c an di m X So d im~ . Prove
by ex amples t h at any value i n t h at r an ge i s pe r mit t ed .

The proofs given in t his ch apt er are i ns pi r ed by an i dea of

Serre ([46]); t he method o f exposi t ion f ol lo ws ver y clo se l y an
impr oved versi on given by Si eg el ([ 501) f or th e c as e o f a
compact mani f ol d . For th e ps eud oc oncave c as e t hey were gi ven
1'; 'rst i n rl1 and f31.
 -45-

A Andreotti

Chapter III. Properly discontinuous pseudoconcave groups:

the Siegel modular group.

3.1. Preliminaries.
a) The notions developed in the previous chapter can be slightly
generalized with respect both of the notion of manifold and the
notion of pseudoconcavity.
Let us consider first the following situation; X is a complex
connected manifold and r c Aut(X) a group of automorphisms of X.
We will say th at r is properlY discontinuous on X if for every
compact set K c X the set

fr~ flyKnK"~J
is a finite set. In particular taking for K a point f xos we
have that the isotropy group of X
o
t, =fYcC!yx = xoS
o 0
is a finite group.
For any point X £ X there exa st s a
o
rx o
-invariant relatively
comp act neighborhood U( x ) such that
o
II y e tJ(x )
o
~ I y € U(x
o)
for s ome '(f. r , then rt e rx •
o
In fact let us choose a coordin a te p atch around X
o an d let
v(n, for E. > 0 and sufficiently small, d e n ot e the coordinate
b~l with center x and radius S •
o
Set Set:) = lY E rl
, V(e..) n v et) '" ~ •
Choose e. s mall enough s uc h th at V(E.) is r elatively co mpact
o
for c. " [. o ' Then S(E) is f i n it e if ~ < Go. Certainly
S(£.)::/ fx • If for any e, 0 <. e. < Go ' S( ~) ~fx th ere exi sts

a 1e
o
r , 'f ,J.
~
rx wi th -y E:. n
O ~L C ~
S ee,) •
o

o 0
Therefore th ere ex t a t ,s a s e quen c e x - ;> X with -> x
YX By
n o n o•
c ontinui ty Y xo = x
0
and thus 'Y t:: 1" x' a c ontradiction. Th ere-
fore th ere exi sts [ 1 > 0, 0 <' (, 1 < [.0 ' s uc h that S( L) =.lx
o
for 0 <.. s: ~ c 1. I t is t hen enough to t ake

U(x o) = V(ll)·
 -46-

A. Andreotti

It follows then th at the equivalence relation ·

R =2(x, Y) G X »; X I x =J y for some Y ~ I' :5

is closed and therefore t he quotient space Z = x/r is·a
Hausdorff space. Let p : X ~ X/r denote the n atural projection,
then ' l'U(x
o)
is a neighborhood or the orbit rx of X
o
and
p ( r U(x o » = U(xo)/tx • 0
o
If we stipulate th at a function f: U ~ a: for U open in
Z is ho 1 omorphic i f f 0 P is ho 1 omorphic on P - l ( U) we can
extend to this new t ype of spaces the notions considered 'i n the
previous chapter. This type of space is sometimes called a
generalized complex manifold (or V-manifold).
In particular we can talk about the field ';f (Z) of mero morphic
fun ctions on Z and of (locally txrvial) holomorphic line
bundles on Z. Every st ~tement about the generaiized mani f ol d
Z can be restated in terms of X an d the Broup r .
Not e th at i f I' has no fixed p Oi n t s ( l ) X/I h as actually the
struct ure of a complex ma~i fo l d .

b) By a [' -automorphic holomorphic line bundle over X we

we mean a holo:norp hic l ine bun dle (F,", X) over X with a
liftin g of the ac t Lon of r over X to a gr oup of Bund l e maps
of F, i.e. for every Y €. f a bundle map P
y : F ~ F is
given such th at
L) r 0 rr = rr Pi 0

11) if Y =T 2Yl t hen P,. =P, 2 0 Pl'

1
For inst ance if F is th e trivial bundle X)( lI:, we will h ave
P, (x, v) = (-t X , Pr(x)v),
where Py ( x) is a never v anishing holomorp hic fun c:tion. Thus

(1)
By this we mean t h at if "ye.1 a nd oE X are such
1
X

that T X
o
= Xo t hen Y is t he i de ntity.
 - 47-

A. Andreotti

5 p 1 can be identified wi t h a collection of holomorphic

~ 'Y
functions
p y : X ~ a:*
satisfying the consistency relation

Pr2"Y"1(X) = Py/'flX) PYl(X).

A system of functions of this sort is called a system of factors
n
of automorphY. For example if X is an open subset of a: we
obtain a system of factors of automorphy considering the jacobian
determinants

PT = det (ClTX)
U& for every 'T Eo r •

Given two l' -automorphic holomorphic line bundles

(F,11, X,fp~) and · (G,CAl, X,lo'"yS), a bundle map of the first
bundle .i n t o the second will be a bu ndle map ~: F -~ G such that
~0 PT =
~ yO ~ for every y c ['.
Fbr instance, given a r -automorphic line bundle by a s~stem of
factors of automorphy ~ PY (xH, t his will be isomorphic to the
trivial bundle (which corresponds to factors of automorphy ident-
ically equal to 1) if and only if there exists a never vanish-
ing holomorphic function cjl : X ~ a:* such that
PI (x) =~ (x) p (yx)-l(trivial ~actor of automorphy)~l)

Given a (locally trivial) ho'l ouo rphf,c line bundle (f', rt ',Z)
on Z its reciprocal image on X is a f-automorphic line
bundle (F, ft, X, ZPrJ) with the property th at
p (x ) = 1 for every "'( eo f x
r o o

One can verify th at this.property characterizes the reci procal

images on X of locally trivial holomorphic line bundles on Z.

(1) Thus the classes of facto~s of automorphY~P"'Y(X)~ modulo

the eqUivalence relation ~ PT (x>.? "'" l6r (x>! 1f£~ p r -ldy(X) is
a trivial factor of auto morphy, correspond to classes of isomor-
phiBms of ['-automorphic line bundles ~F,"'; X,~ pyH in which
F is the trivial bundle.
 - 48 -

A. An d r e otti

c) Given a r -automorphic line bundle ( F, TT, X, tP rJ.5 we

can consider the sp ace of r -invariant sections fo F:
r (X, F) r = [e (. rex, F) I s(,x) = Py( .x ) sex) for every x ~ x.
Gi ven two [' -invari ant sec tions So _ 0, sl 0r F, th en
sl
- is a [' -invariant meromorphic function on X i.e. an
So element of p
* )t(Z).
Conversely for every [-invariant meromorphic function m p *it(Z)
we can find a r -automorphic holomorphic line bundle and two
sl
[-invariant sections of it: sO; 0, and sl such that m = -.
So
The proof is straightforward and is thus omitted. We can also
repeat the consid~rations of section 2.2 c) in this more general
case.
3.2. psuedoconcave properly discontinuous groups.
a) For practical reasons it is convenient to generalize the
notion of p s e ud oc on c avi t y as follows.
n
Let J1 be an open set in a: , and z to d fl. = .'1 51. a boundary
o -
point. We will say th at Zo is a pseudoconcave boundary point
if we ~ an fin d a complex 2-dimensional line ar sp ace E t hrough
zo:
z = Zo + ~t l + a
2t 2;
~,a2 ~ a: n , lin early independent,
(t l, t variable in a: 2 and a c" function <P on a ne ighbor-
2)
h o od V of Zo i n E, real-valued an d such th at
L) V (151-2, {z Eo V I 4>(z)< ¢(zo) 5
ii) L(q)z <: o.
o
Let now r be a prop erly di s c on t i n uo us gro up of a ut omo rp hi s ms
of a connected complex nanifold X. We will say th at C' is a
pseudoconcave group of automorphis ms if we can fin d a non-empty
open subset Jl C X such th at

L) fL is rel atively co mp ac t in X

ii) for ~very po int Zo 6 .::> l~ the orbit rz contains

o
ei th er an interior p oi nt 0 f ...;'1.. or a p s e udoc o nc av e bound-
ary po i n t o f v~ •
 -49-

.A. Andreotti

Clearly for r =identity we obtain a generali zation of th e

notion of pseudoconcave manifold given i n the pr evi ous chapter.
It is not a di fficult ex er c i s e to c arr y over to this more
general case all the theory developed t here. In p art i cul ar we
obtain t he theorem (3):
be a connected complex
, of X.
cave group of automorphisms
[-invariant meromorphic functions on
of transcendence degr ee d ~ dim~.
Note th at 7( (Xl ::: p * ~( z ) wh ere Z = X/f.
Suppose for instance that there exists a I' -automorphic line
bundle F on X such th at the quotient fie ld ~ ( X , F) of
t he graded ring

j{(X, F/ = k~O ['(X, ~/

has transcendence de gree equal to th at 0 f 1< (X) r (for instance
equal the dimension of X) t hen it follows, since C!
(X, F/' is
algebraically Closed in j( ( X ) L' th at we must have
5i (X, Fl = ~( X)r.
b) The notion of pseudoconcave prop erly discontinuous group
of automorphisms is stable by "commensurability"; two subgroups
I' I' t 2 of Aut (X) are called commensurable if ['I (\ i'2 is
of finite index in rI and r 2'
Proposition (3.2.1). If r 1 and [2 iU'e commensurable and if
one of them is properlY discontinuous and pse udoconcave so i.
the other.
~. Let r 1 be properly discontinuous, then G =
as a subgroup of f 1 is properly discontinuous.
Set r 2 = G~ U ••• o G~ and let K be compact in X. I f the
set lye r 2 I y K () K I- rJ is infinite, th ere exi sts an index
 A. And reotti

i with l~. i s k and infinitely many g'a in G such that

gaiK n K #~. But then for infinitely many g's in G
g(K U aiK) n (K lJ aiK) # ¢, which is a contradiction sinCE
K U aiK is c ompact.

Su"pose th at l' 1 is pseudoconcave. It is enough to show

th at G = ['1 (1 r2 is pseudoconc ave. Let n c. X satisfy the
condi tions i) an d ii) specified above in a) for r = r l'
Set f 1 = ~ G u ••• o ~ G.
Then n, = ail n. u ••• u ail n. veri fies th e same condi t 'ions
for r= G.

3.3 Siegel modular group.

We will ap ply th e abov e c on sider ations to the particul ar case

of the Si egel up per half pl ane and Si egel modular group (r3]).
Let n be an int eger, and l e t ??'( (n , 0:) = Hom(o:n,a: n)
n > 0,
be the s e t of nAn matrices wi th coolllplex entries. The
ge nerali zed upper h alf pl an e i s the set

Hn = ~Z G rt c«, 0:)
t
Z = Z, Im Z > °.
This s et c an " be i de ntifi ed with an open s ubset of o:tn(n+l).
Note t h at H is t he usual P oincar~ u p per h alf plane .
l
Consider t he simplect ic group S (n , lR) Le. t h e s et of linear
n n p
a ut.o mcz-oh t.sms of lR >< lR which le ave invari ant t he exteri or
n n
form i~l dx i " dyi' (x, y) E:. lR x lR •
in matrix no tat i on

Sp(n, lR) = ~ g 00 1f!(2n, lR ) I t gJ g = JS where

J = (-I ° I
0)' '»r
{ (zn , lR) = Hom(lR 2n ,lR 2n i.

We c an l et s p ( n , m) op e r-a t e on t he ge n e rali zed up p e r hal f

plan e as follo Vis
 -si ,

A _ Andreotti

Proposition (3.3.1). This oQeration i s a well J e fi ned

automorphism 0 f Hn •

E!:2£.!. Let Z € Hn and let I denot e the n x 11 identity

matrix. The conditions defining Hn can be reformul at ed as
follows:
(t z , I) J(Z)
I = 0, i(t z, I) J ( Iz) o.
Set
(AZ + B)
(U)
V = (A
C D
B) ( Z)
I = cz + D

Then, since g is symplectic, we must h ave

i ( to, til) J (~) > 0 ,

t vU) > °.
This shows th at V must be non-singular, otherwise f or a vector
w # 0 vw = °
thus it;(t uV tVW)w = 0 which is impossible.
Then we can consider the matrix
-1
(Zl)
I
i (U V
I
)
=
From .the first condition we derive
t
( Zl' I) J(Z1 l) 0 i.e. =
t
z1 = Zl .
From the second we derive
i(t z1 t I) J(Z11) > O i.e. t-
i( Zl - Zl) » 0 .
This shows th lt Zl = (AZ + B)(CZ + D)-l is well de f i ned and
represents a point of Hn• Since the transformation is invert-
ible it gives an aubemorphf.em of Hn•
By the map Z ~ (Z - i1)(Z + i1)-l the generali zed upper
half plane is mapped into the generalized unit disc ZZ < I
which is a bounded domain in II:t n ( n+1 ) and Sp(n, lR) appears
as a group of automorphisms of a bounded- dom ain. Consider the
di screte subgroup Sp(n, ~) of Sp(n, lR) of tho se matrices
with integer entrices. By the above rem ar~ it follows th at
Sp(n, lZ) acts in a properly discontinuous way on H
n
 -52-

A. Andreotti

The transform ations g =+ I are the only tr an sformations which

act as identity on H and a r e cont ained in Sp(n, ~(l). The
n
group Sp( n , lZ) (or more precisely t he group Sp (N, ~) / ~ .:!:IS)
is called the modular group of rank n, For n = 1 we ge t
the usual modul ar group in one v ari able.

3.4 Pseudoconcavity of the modular group.

Theorem (3.4.1). If n L 2 the modular group of rank n ~

pseudoconcave properly discontinuous group of automorphisms

~ Hn •

E!:221. (q;) Every psoi ti ve de fini te sy:nmetric matrix Yean be

written in a unique way in the follow ing J acobi normal form:

Y = t wo w with D = w= [1..o JWi

·1
j\

For a positive r eal number fJ, l et {/ ~ be the open subset of

~ defined by Z = X + iY (,.D..fJ if

i) where

ii) for i c j

:.11i )

Note that if X + iY to D.IJ- ' th en there ex i sts a constant

c(p) > ° such th at Y) c(}1)I (c(~) = min ( ~, L».
n
t-'
If I-l ~ Po is sifficiently large then .1l ~ is a "fundamental
open set" for the modular group l' = Sp(n, lZ), Le.

(1) If (AZ + B)(CZ + D)-l = Z for e very Z c: H then taking

n,
Z = ~ 11 we get th at necessarily =C = B 0, A = D and moreover
AZ = ZA. Thus A = fJI and therefore tJ =.:!: 1.
 -53-

A . Andreotti

i) rfltJ- = Hn

11) ~Y<frlrI2.r' {\ Cl..}J- ,. f1 is a finite set.

For the proof of t his st atement we refer to (481 and L49].

Notice th at from this statement the proper discontinuity of r
could be ded uc e d by a more direct argument than the general one
usee be fore.
() We fix the parameter once and for all. We
will sas- th at a transformation is a transformation at
infinity if
F=y -1 (Q,...)0 n I-'

contains a divergent sequence of points in Hn (i.e. a sequence

with no accumulation point in Hn ) . In view of the definition
of il ~ this is equivalent to sas- that
sup d (Z) =d.
Z ~ F n
v
Note th at the s ame is true f or F = t (F) = IIp (\ Y( OtJ) •
We set

v(Z) = y-l

A di r ec t c alcul ation s hows t h dt e ach component vi of v(Z) is

-1
of th e form vi = d n u where <ii i s a polynomial in the
i
wij's. Therefore i f Z = X + 1Y describes F, II v(Z) II takes
a rbitr ari ly s mall val ues.
Set r(Z) = II v(Z) 1/-1 so th at ~)f~ r(Z) = 0", and mo r e ov e r

UJ ·
Le mma. = (; ~
If I
y w(Z) where w(Z) = n
v(Z)
v( Z) /I

is a t r an s f o r ma t i on at in fi. nity, then

Il w( Z) II = 1.

in t h e matrix C t h E- l ast ro w and c ol umn are zero.

Ind eed if Z ~ F, t here exi s t s

v
Z G 0 r' s uc h t h at Z ='( -l~Z
 -54-

A. Andreotti

(1)
\I
Z(CZ + D) = AZ + R •
Multiplying (1) on the right by w(Z) and equating the real
p arts we obtain

y cr ~ } = (XCX + XD - AX - B) w(Z).
Lr(~)
When Z describes F. the right hand side represents a vector
u(Z) of bounded norm. Since Z" Qt.L there exists a constant
v
c > 0, independent of Y and x, such that
(YX, x) 2 e IIxl/ 2
where (y, x) = tyx denotes the euclidian scalar product;

J' C[r J) c
Therefore

Ollorr ll/2 s, (Y 0r] l/u(z)lI 1I C ( ; ]]/

r?Z)
lr(z)J tr(z) r(Z)
i.e.'

11 0

( r(Z)
q)11 s, ~II u(Z)/1

which is bounded as Z describes F. This is possible only if

the last column of C is zero.
v
As Z describes F, Z describes Set now
v v_l
v(Z) = Y

and proceed as before. Setting we get

[1]
v V' v
= Y w(Z), r(Z) =

v
and
v
sup
v
r(Z) =o ;
Z e. F
 -:5'5-

A. An dr eotti

t v
Multi flying (1) on the left by w(Z) we get

(0, ••• ,0, r(~» C Y = tw(~) d C X +" *D - AX - B).

Arguing as before we obtain th at the l ast row of C mus t be zero.

As a consequence of this lemma we get that, for a trans form a-

tion Y = (~ ~) at infinity, det (CZ + D) is independent of
the last row and column of Z.

(1) Consider on Hn the function "k ( Z) = det Y where

Z + X + iY. A direct calculation shows th ~t for every
= (CA D»
B
Sp(n,2Z) we h ave
k(yZ) = k(Z) det(CZ + D)-2

Moreover one can pr-ove the following property (c f. ( 491) :

for every Zo e H , sup k( Z) c, d and the maximum is
n
Z c IZ o
att ained at some point Zl" a p o l'Zo'
If we set
f (Z). sUP.
1.,]'
k("!'Z) ,

we obtain a r-invariant function on Hn ..

Notice that the sets

Bc .~Z .. .QjJ-1 +(Z)< c

for any real c, are rel atively compact in H In fact Bc

n•
is contained in the set ~ Z ~ 0,.. J det Y < c] which is contained
in a set of the form ~ Z e at' d n c do. .' I
Choose c)' °
large enough to ensure th at for ,-
T =2Z .. a,.. (Z) > c
J ~ every '"'( Eo l r
with ~ (T) () T ~ ~ is a transformation at infinity. This can
be done since the set 2y€f/YIl,..,,,.n rt ~~ is finite. Let
-y l' ••• , Ys be the set 0 f these trans formations s o that,

for Z E T we have

~ (Z)
1
=val,
sup
••• , s k('(" Z).'
Y
 - 5 6-

A . Andr eotti

I~ fact for y € I" such t~ at y (z) €. Q tJ ei t h er IC ( r( Z» ~ c

=
and thus --( (Z) € T so th at -y 1 i for so me i or else
key (Z» s c and therefore k('!( Z) c; ¢ (Z).
we now take for.a. the set Bc +l ' We W.-U1t to show th ,lt for
any point Zo € aQ. t here exists in r Zo either an int erior
point of.a. or a ps eudoco nc ave bound ary point of fl..
Let p : Hn ~ Hn Ir be the natural pr oj ec t i on on t he quot i ent
;space. Since 4' is'£ -invariant, Q p -1 pea). Therefore a =
point Z € a Q which is no t equivalent to an interior point
o
of il must be on th e "surface" l. ~ = 0+1 S. For such a point
we must have
k(Zo) = k(Yy Zo) for those v's, 1 s.v s, s, for which
k(l' y Zo) = HZ o) =
e-i ..
In particular for t hese Y 's we must have

/det (C Zo +D)} =1, "'{\,oJ = (~ ~).

Moreover det (C Zo + D) is indepe ndent of the l ast row and
column of Zo.
Consider th~ linear space L(Zo) of complex dimension n,
through Zo' defined by the equations
o
Zd/3 • Zd~ for 1 ~ c( S n-l, 1 ~ \3 ~ n-l, Z s ymmetric .
Since on that space /det (C Z + D)/ =1 we mus t have
~IL(Z~) • k(Z) ;
I,)

Therefore in a neighbrohood V of Zo on L(Zo) the set

Q 0 L(Zo) can be described as the set ZZ , V J k(Z) <: c+1S.
The proof of the theorem will be complete if we show th at the
function - log k(Z) is strongly ·plurisubfarmonic in Hn
(and thus the same is true for its restriction to L(Zo»'
TO see this we remark th at by the way det Y transforms und er
the action of Sp(n, m) the form
K(Z)dv = (det y)-n+l)dv ,
(where dv is the euclide an volume element), is invariant.
 -57-

A. Andreotti

Since Sp(n, m) acts transitively on H th at form, up to

n
a positive factor must be the invariant Bergman form (c f. [51]).
Therefore that quadratic hermitian form

-(n+l) a.~ log k(Z)

is the Bergman metric of Hn and thus it is positive definite.

Remark: Let m be an integer. One can consider the subgroup rm

of r =Sp(n, ~ defined by

I' m =~. g £f g ': I (mod m)s ..

For every m, rm and [' are commensurable. Therefore for

every m rm is properly discontinuous and pseudoconcave.

3.5. Poincare" series.

a) Let D be a bounded domain in ~m for "'( e Aut (D) we

set
Py (z) = det (~~~») ..
If r c, Aut (D) is any aubgr-oup of Aut (D) and if 1 6 1Z.
then the set l py<Z)lJ is a system of factors of automorphy for
r and thus it defines a r -automorphic line bundle ~.
For any f holomorphic in D we can consider the
. /
"Poincare
series of weight k ll

13 k( f; z) = L.
y,,1 f(.." z)
I
det ( *f ) k

If 1 is properly discontinuous, if f is bounded in absolute

value and if k ~ 2, then this series converges uniformly on
compac t sets. Indeed the convergence of th ~t series reduced to
the convergence of the series

L det
Y6i:.
on compact Sets.

Let Q ec, D be a polycilinder of radius R and let P be

any polycilinder of radius f<. R cont ai.ncd in Q. Since
det
*f
d z
is ho Lo.noz-phLc we h ave at the center x
0
of P
 -58 -

A. Andreotti

dv = vol yep) ~
z=X
o vol!'(Q) •
Therefore if V is the concentric polycilinder of Q of radius
R-r, for any point y " V we have
2
d.t~ (c = (vol P) -1 • vol Q)"
z=y

Let s be the number of transformations y IE r such that

Y Q () Q ~ e, We have for s If. V

,;,1 :rWj2 z
., c c

s; vor (Q) t. vo11(Q)

c s
vol (Q) vol (D),

since in u any point is covered at most s times.

ye.f
This proves the absolute uniform convergence of the given series
in V. It follows then the absolute uniform convergence of that
series on any comp act subset of D.
I
Now if 9 k ( Z) is a convergent Poincare series of weight k,
it satisfies the functional equ ation
for every r~· L •

In f act k
~k(rz) =6.l:
L f(~1 z) (~)
yz

=6,;
~
r f(oyz) (mfll
d z
)
k
(~
yz)
)
k

= PI ( z ) -k ae]
s;
fUt z ) WzrL )
d z
k

=P I ( z)
-k 9 k(z).
 - 59-

A. Andreotti

This shows that any such Poincar: series repres ents an i nvari ant
section of the -automorphic line bundle F- k•
n -2 r
One is led . therefore to consider the ring JL(D, F ) and its
quotient field Cl (D, F- 2) r •
Disposing of the freedom one has in the choice of the function
f one can show (e r, (191) that the field a
(D, F- 2 ) £ con-
tains m analytically independent mero~orphic functions.

b) We can now apply the above consideration to Hn and to the

modular group as Hn is iSGmorphic to a bounded domain of ~m,
m = tn(n+l).
We then deduce the following conclusion.
l2£ n 1 the field of r -automorphic meromorphic functions on
H coincides with the field ~ (Hn, F-2~r. In particular
n
(q) the transcendence degree of th at field equals tn(n+l);
(d) every r -invariant meromorphic function can be written
as the quotient of two Poincare series of the same weight.

Remark. We have reached the above conclusion using the fact that
~ admits a bounded model. Using instead the usual unbounded
model of Hn one can develop a simular aJ:gument replacing the
factor of automorphy P1 by

0r(Z) = det (CZ + D) Y = (~ ~) ~ Sp(n, ~)

One is then led to the theory of Eisenstein series:

l~r Z
f(r ) aet tCZ+D)k
and reaches a similar conclusion but the proof of convergence of
Etsenstein series is more difficult (c r, 491).r
~e result proved in this chapter for the modular group was extend-
ed by A. Borel ([171) to arithmetic groups acting on irreduc-
ible bounded domains of dimension 2 ~ (cf. section 4.1 example 5).
 - 60-

A. Andreotti

Also one can show as in H. Cartan ([19]) th at the quotient

sp ac e x/r has the at r-uctur-e of an analytic space and is
p seudoconcave if r . is in the sense of section 3.2.
 -01-

A. Andreotti

Chapter IV. Projective imbeddings of bs eudoconc ave manifolds.(l)

4.1. Measure of pseudoconc avity.

a) Let U be an open subset 0 f a: n and ~ : U - lR be a C d
function. We will s as th at ~ is strongly q-pseudoc onvex at
the point Zo e U if the Levi-form

L(~)z =.~
a2 di uu
o dZa.3Z~ Zo c:l e
(where zl' ••• ' zn are local holomor?hic coordinates at zo)
has at least n-q pos i t i ve eigenvalues. In particular a strongly
o-pseudoconvex function is a strongly plurisubharmonic function.
A s we have seen before (l./f) this notion is independent of the
choice of local coordinates, and could also be formul ated as
follows:
there exists a (n-q)-complex-dimensional plane E through
Zo such that ~IE is strongly plurisubharmonic.

Remark. For-practical reqsons it is sometimes convenient to

release the assumption th at 4> is C <i' and require only that
(L) in a neighborhoo:l 0 f Zo there is a finite number 0 f
C d' functions 4>1, ••• ,t} 1 such th at ~ = sup (~l' ••• '~l);
(ii) there exists an (n-q)-dimensional plane E through
Zo such that i E' 1 ~ i ~ 1, is strongly pluri-
subharmonic in a neighborhood of Zo in E.

b) Let X be a connected complex manifold of complex dimension

n. We will sas th at X is q-pseudoconcave if a Cd function
~ : X +lR is given such that
(1)
In this chapter the notions of analytic set, complex space,
normal complex sp ace, Stein space will occasionally be used al-
though our main concern are complex manifolds. For the basic
definitions we refer t o the foll owing Chapter 5.
 - 62-

A . Andeeotti

(i) for every c > Ln]' q, the sets

Bc = f x e X ~ (x) > c I S
are relatively compact in X ;
(ii) there exists a compact subset K of X such that at
every point Zo ~ :i - K ~ is strongly q-pseudcconvex ,

Remarks.
1.' For all c > inf.p, except for a set of measure 0, the
boundary aBc = Bc - Bc is smooth (By Sard's theorem). The
Levi-form restricted to the analytic tangent plane at
Zo G' dBc to aBc' c <: inf </>, has at least n-q-l positive
K
eigenvalues. Th erefore if n-q-l ~ 1 these manifolds are a
special c ase of the pseudoconcave manifolds studied in
chapter 1 .
For this reason when we speak of a q-pseudoconcave manifold X,
if X is not comp act, we will assume that , Q. ~ q So n-2 , hence
n ~ 2. Indeed only the eigenvalues of the Levi-form in the
direction of the level surfaces of ~ do have a geometric
meaning; taking a rapidly increasing convex function of <P the
remainin g eigenvalue can be forced to be positive on .any given
compact set.

2. The notion of q-pseudoconc ave manifold could be generalized

in the sense of the remark made in a), above.

EXamples.

1. Every comp act connected manifold X is q-pseudoconcave for

any q, Take K = X, ~ = 1.
2. LetZ be a connected c ompact co mplex manifold of complex
dimension n z 2. Let ~ xl' ••• , x k ~ be a finite subset 0 f Z,
t hen X = i -t Xl to • • , x k S is O-pseudoconc ave. Indeed if

z\i) , 1 s, j So n , ar e loc al complex coordin ates a t xi' vanish-

J
ing at Xi we can select e. > 0 such th at the c oordin ate balls
 A. Andreotti

B
i
!! f
l
¥ \I z(i)
j:l j
I2 <: Eo ~
) t

are relatively compactkin their coordinate p atch and are dis-

=
joint. Take K Z - U Bi and for ~ a C tP function such
i=l

3. Now generally let Z be compact connected mani f ol d of

complex dimension n 2 2. Let Y be a complex submanifold of
Z with dim~ Y q ~ n-2. =
Then X =
Z - Y is q-pseudoconcave.
Indeed, at every point y € Y we can choose a coordinate
neighborhood U(y) with coordinates U ••• ' znU such that
zl'

U(1) n Y =l Z € U Z~+l = ... = z~ = o~.

Let ~ U = j=~+ll Z~ {2 . Cover Y wi t h a finite number

Ul'· •• , Ul of these neighborhoods and let . Uo = Z - Y and

U
4 o = 1- Let f p i~ b¢ a c cP partition of unit y subord in ate

to the covering ~ UiS O~i~l' Set ~ = t1 Pi

l Ui
'f •

Then 4 > 0 on X and for E../ 0 J> ;, E ~ is rel at ively t

co mpact in X. Moreover ~t e ach poi nt of Y L(q) has
n-iq positive eigenvalues. Therefore if E.. > 0 is s Uf fi ci ent l y
small on X -? 4> ~ c. 5 L(~) h as n-sq positive ei genv alues.
4 . Let H be the gener ali zed upper h ~uf pl ane with n L 2,
n
and let r 'o e the modular group. From <3. 4,0:)) it follows
th ~t th ere is an integer 1 > 0 such th at f or ev ery point
Zo € H the isotropy gr oup I' z h as an order S. 1-
n o
 - 94 -

A . Andreotti

Let pbe a prime number with n > 1. Then the gr oup

r
p
r
= g «I' r g!! I (mod p) cannot h ave fixed point s ( l ) • There-
fore X = Hnll'p is a complex manifold;
From the proof of theorem 3.4.1 it foll ows th at X is
q-pseudoconcave with
q = tn(n-l) = dim1
Hn_1 •
Here the concavity is to be understood in the gener al sense of
remark 2 above.
5. More generally let D be an irreducible symmetric bounded
domain in lC n with n ~ 2. Let .c
be an arithm etic . group 0 f
automorphisms of D without fixed points. A result of A.
Borel shows th at X = D/r is q-pseudoconc ave for some q,
o ~ q ~ n-2 «(
17] ).

4.2 The problem of projective imbe dding of pseudoconc ave

manifolds.
a) We have already remarked th at if X is a pseudoconc ave
mani f ol d and T : X -+ P N(lC) a one-to-one ho1o morphic map of
X onto i (X), then j (X) is contained in an irreduci ble
algebraic variety Y of the same dimens ion than X (theorem
(2.5.3». I f y.1; Y is the normali zation of Y (c r , ch ap t er 5),
then the map ) : X ~ Y f actors through ~. Mor eover Y
.
is again a projective algebraic variety. Thus X is i s omorphi c
to an open subset of the projective algebraic variety Y.
.
(1)
Indeed if A 6'£ has a fixed point, for some 1 ~l
1 p 1
1
• e must have A = I. Now A = I + pB thus
1 1 1 1
( 1) pB + (2 1) p2 B2 + + P 1B 1 = o.
1
Dividing this relation by p we see that e ach entry b i j of B
=
must be divisible by p. Thus A I + p2B1• Arguing in the
2
same .~ we s ee th at each entry of B1 i s devisible by p , thus
A = I + P4B2• And so on. It follows th at we must h ave B = 0
and thus A = I.
 - 65-

A. Andreotti

The problem of projective imbedding of speudoconcave manifolds

can be loosely formulated as follows: to find some useful criter-
ion to ensure that a pseudoconcave manifold X is isomorphic to
an open set of some projective algebraic variety. We will then
say that X admits a projective imbedding.
b) If X is compact such criterion is provided by the theorem
of Kodaira [33J
Theorem (4.2.1).k!l X be a compact complex manifold, the
!QllowlDs ·~e .quivalent conditionsl
A. X admits a projective imbedding.
B. X carries a K8hler metric whose aasociated exterior form
has integral periods.'
C. There eXists over X a holomorphic line bundle F!!... X
whose total space F i8 O-pseudoconcave.
D. There eXist over X a holomorphic line bundle F ~
tM~t the graded ring A(X, F) ; l~ .f (X, separates Fl)
points and gives local coordinates everywhere on X.

Let n =
d1m~; to say that ~(X, F) separates points and
gives local coordinates everywhere on X means the following:
(a-:) given x # y X there exist an integer 1, 1 > 0, and
sO' sl e r(x,~) such that
SO(y) ]
det #0
~(y)

re ) for every x X there exists an integer 1 > 0 and

8 , ••• , s G L(X, Fl) such that
0 n
(f (_l)i si d So /\ ••• A A ••• (I dsn ) x # o.
's l So
In ot~er words the meromorphic function - (or - takes
So sl
different "values" at x and y while the meromorphic functions
sl sri
So , ••• , So (if so(x) # 0, otherwise renumber th e sections s)

provide local coord inates at x.

 - 6-£-

A. Andreotti

We will prove here only the implications D~ A. The other

im~lications will be proved in chapter VI.

E!:22.!.
D ~ A. For any two points x # y X we can find an int eger
1 = l(x" y) and two sections q, (3 Eo r:'(X, F
l) s uch t hat

det
« ( X) . q(y) J # 0
[
i5 (x) (3(y)

Replacing ~ and (3 by convenient linear combin ations, we may

assume th at c{ (y) =(3 (x) = 0 and thus d (x) ~(y) ~ o.
Therefore also

det ~ 0 •

Set
l)
Al = ~ (x,y)-" X x X I Y (<i,~) € £(X, F x

r (X, F l) j C( (x)(3(y) - d (y) (3 (x) = 0

By the above remark for any int eger k> 0 Akl c AI . Now for
each 1 Al is an analytic s ubset of X x X cont aining the
diagonal I:>. of X)( X. If Al ~ '" then for s ome
k> 0 : Akl #
Al• Since a decreasing sequence of analytic
subsets of a compact manifold must be st ation ary th ere exists an
1 o > 0 such th at Al = A. Therefore
o
1
(i) the sections of s: (X, F 0) do sep ar ate any couple of
points x # y i n X. Simil arly by a simpler argument one proves
that 1 0 can be cho sen such th at also
1
(11) the sections 0 f r (X, F 0) h ave no common zer os ,

(iii) they give local coordinates everywh ere on X.

 - 67-

A. Andreotti

1
Let {sO"," sk} be a basis of I (X, F 0) and consider t he
map f : X ~ F (II:)
k
given by

x ~ (sO(x), ••• , sk(x»,

Thi s map is holomorphic by (ii), is one-to-one (by(i» and
biholomorphic (by( iii) ) • There f orer is an isomorphism 0 f X
onto ~(X) which is a projective algebraic manifold by
theorem (2.5.3).
A ~ D. It is enough to prove th at implication for X = Fn(m).
Consider on Fn(m) the line bundle of the hyperplane sections.
this is given by

as we can write

Z
~). It is then
zi

immedi ate to verify th at this is a holomorphic line bundle F.

A section on Fn is given by (zO'"" zn) ~ (zO'"" Zn+l)
where separates
points and gives local coordin ates everywhere.
Remark. If X is not compact the proof of the implic ation
D ~ A bre aks down, al though one h as reason able conditions to
ensure th at condition D is fulfilled. For instance:
if X = H/r is a quotient of a bounded demain H c. m
n
by a properly di s co n t i nuous group (without fixed poi nt s ) then
one can show directly, by means of Poinc are series, that condi-
tion D holds;
 - 6 8-

A. Andreotti

if X· is a complex manifold and on X there exists a

holomorphic line bundle such that F is "positive" and
F 8 K- l "positive and complete", K denoting the canonical
bundle of X.
We refer for these st atements to [13J."
We thus formulate the problem of projective imbedding in . the
followin g mann er:

~ X be g-pseudoconcave manifold with o ~ q S" n-2,

n = dima: X. Assume th at there exists on X a holomorphic line
bundle F such th at (X, F) .fl.. separat es points an d gi v es local
coordinates everywhere on X. Does X admi t a projective
imbedding?

4.3 Solution of the probl e m for O-pseudoconc ave .man i f o l ds

(c r , [101 , (121).

'lheorem ( 4.3.1). Let X ~ O- pseudoconc ave man ifold.

~"JP o s e th "lt t here exists on X a holomorphic line b ~md l e F
th a t J2 (X ~ F) ;;.s_ep~ar=-a:::.t=-e;;.;s;;....o-=:=:.;;......;.;;,;;.;;.;..
poi n t s an d ..... gives l ocal co o r din ates
.;..;;.;;;...;;;.:..;;.=;....;:;..;;.;:;.===-;;-.

everywhere on X. admits a proj ective imbe dding.

Proof (following an i dea of H. Gr auert).

( u:) We c hoose Co ~ lP, inf ~ < c i nf d such t h at
X 0 K
=2x .; X I~ ( x) .> Co ~ h as a smooth bo undary . Bec ause

Bc is comp act, we can choose an 1 su fficient ly l arge so t h a t

o
l)
the s e c t i ons of I (X, F h ave no c ommon zeros, sep ar ate p o i n t s
and gi v e loc ;J1 coordin at es e v e r ywhe r e on Bc • This is don e by
o
the same argume nt d e v e l op e d i n the p r evi o us s e cti on .

«(3) Let N+l = dimlC [ (X , F )

1
an d l et 'be a
l).
b asis of I (X, F We consid er t he map

;- :
 A . Andreotti

defined by xl-? (sO(x), •••• sN(X»' This map is holomorphic

and one-to-one. (1) By theorem (2.5.3) ~ (B ) is contained in

Co
an irreducible algebraic variety Z of the same
dimension than X.
(T) Bec ause X is o-pseudoconc ave any analytic subset A
of X contained in X- Bc o must be O-dimensional ;

Assume if possible th at A has an i rreducible component Ao

of dimension ~ 1.' Then d A I
has a maximum a t some point
o
Zo e A ' Since A is of positive dimension, we can find a
o f
one-to-one holomorphic map ~ : D ~ X where
I tl <. 15 such th at
1
= zo' fl (D ) c Ao •
1,
Then f 0 I-' is a subharmonic function on D non-constant and
having a max i mum at t = O. This is a contradiction.

() Let n = dimt X. Consider th e subset of X :

A=~x~XI (~(_l)h 6
ih
dS
i
1\ ••• 1\ dS
ih
A.·•• ,.. dS
i
)x =0
0 n

for all ~ i o' • • • • inS c, ~0I ••• , N '§ •

This is an analytic subset of X c ont ained in Thus X - B
co
i t is a dis c r e t e s e t by (y), thus at mos t count able. Obviously
i extends to a holomorphic map

l' X-A"';> Z

which i -s of r-anx n at; e ac h po i n t . We a gr e e t o d ele t e from A

a l l p o Ln ts over whi ch ') may b e ex ten:ied by a ho Lomo r-p hLc map
o f r an k n,
(1)

We can als o as s ume t ••• t

 -70-

A. Andreotti

We will call this set A again.

(e) If we can show that A is a finite set then one com-

pletes the proof as follows. We can replace 1 by a convenient
multiple of 1 so that the corresponding new map is of rank
Ie.-.>
n at the points of A. Let Z ~ Z be the normalization of Z.
It is knowg that Z* is also a projective algebraic variety
(cr. [52], (531) ~ Since X is normal the new map
~
~: X Z factors through the normalization Z* of Z by a
holomorphic map

N* *
Since T is of rank n everywhere, and since Z being
",*
normal is locally irreducible, I must be one-to-one and
therefore an isomorphism.(l)

(~) We are thus reduced to show that A is a ~inite set.

Let
"
T ~e the closure 0 f the graph T of 1"
,., in X)I. lIN( an.'
The set T must be an analytic set as it is t he irreducible
component containing T of the analytic set
l (x , t) 6 X ><.IN(D:) I si(X) t j - Sj(x) t
i
=0 for o x i,
j ~ N •

Let "'*
~ X-A ~ Z * be the factorization of 1'" t hrou gh the
normalization Z*w
_ Z of Z. Let T "* be t he closure of the
t"'*
graph of l' • This is also an analytic set as it is c ontained
. -1 ..
in the analytic s e t (I J' w) (T) which has th e s ame dimension
~. ~* A* *
n than T. Let ex: : T -7 X, l?> : T -:;> Z be the n atural
projections. Now let a G A and set "'*
., ( a) = t3d -1 ( a). Let
U be a coo r dLn ate b all centered at a and not cont aining any
other point of A.

(1)
is of r an k n everywh ere and gen e ral l y one-to-one as

)( (X) = a:
 -71-

A. Andreotti

Now er· must be one-to-one as :z, is normal; so pI -1 must

-1 ~ (U)
be one-to-one as it is one-to-one on d. (U-a) and since
(3/ -1 · mu",t also be one-to-one. Therefore 11 (3o.-leU) is =
op~ ;bl!)thus a neighborhood in Z· of ~( a) . Since Jl, is
normal, the holomorphic map A = d. \3-1 : a - r· ( a) ~ U extends to a
holollorphic map
We · cla1ll that each component
.A :.fI.., -;> n,
B
N.
of 7' (a) which is not of co-
cimension 1 must be an irreducible component of the singular
set ot Z·. In tact if B contains a non-singular point b e Z·
and if B is of codimension 2. 2, then A must have a non-
vanishing Jacobian at b. Thus A. -1 is defined near a and
)..-l(a) = b i.e. ~. must be a local isomorphism at a.
Hence a tP A.·
•
rv·. .-
(tj) Now we remark that C( is a proper map because Z is
compact. Thus for each a '" A T (a) is a union of finitely
many compact analytic subsets of Z - l (Bc ) =Y of codimension
one and finitely many irreducible componentg of the singular set
of Z· (which must also be contained in Y.) Moreover, from the
a,. ,
previous ariument it follows that if
a,. #.
a
..,1* .
then "( (a,.) (l T (a ) ~
tV. = ..
a
2
are in A and

Z Z
We will prove that A is a finite set if we show that Y con-
tains at most tinitely many irreducible compact analytic subsets
of codimension 1 (here we use that the singular set of Z•
has a finite number of irreducible components).
Let e: > 0 be so small that Co- c: > iff ~ and that 7 extends
biholomorphically to B
co -e = ~ 4 > c o- l:S. Set
.
-. • N*
Ic _ e• Z - -r (Bc _€,).
o 0

Then! is compact in Y and we can, by the isomorphism

co-e..
given by r, transpose the function ~ on Y- YCo-e.• Let us

extend this function by putting it equal to c o- €, on Yc and

o- e.
 -72-

A . Andreotti

call t the cont inuous f unction t hus obt ained. The f unction
-r: Y ~ m, just d e fin ed, h as t he fo llowing properties
i) '( is continuous;
11) Yc ={,/,< c ] cc:. Y for any c z, c
= sup 'f;
0
Y
iii) on Y- yc. - °is cO' and stron gly plurisbh armonic.
E-
o
Then every compact irred ucible anal utic s ubse t of Y of
dimension ~ I must be contained in ~c _ E. by the same
o
,'argumen t given in (y).
Moreover Y is holomorphically convex as it follows from i),
ii), iii) and the solution of Levi problem (cf. ch apt er VI).
It then follo ws by the reduction th eory of holomorphically
convex spaces (cf. [20]) th at th ere is a compact analytic
subset C ~ Y of dimension I at each one of its poi n t s and
such th at any irreducible compact analytic subset of Y of dim-
ension ~l must be cont ained in C (see section 6.3 remarks
? and 4). Since C has finitely many irreducible components
of codimens ion 1 the theorem is pr ov ed .

Remark: only if di mE X =
2 this result can be considered
sEj.tisfactory as the only type of pseudoconc avity available is
O-pseudoconcavity.
4.4. The c ase of di mE X L 3.
If dimE X L. 3in the pr evious theorem the assumption th at
th e ring J1(X, F) sep ar at es poi nt s c an be dropped; one has
in fact the following
Theorem (4.4.1). Let X be a O-pseuQoconcave mani f old with
dim~ X ~ 3. Suppose t here exists on X a holomorphic line
bundle F s uc h th at t he r ing Jf(X, F) gives l ocal coordinates
everywhere on X. Then J( (X, F) do e s also separate points
of X so th at X admi t s a proj ecti ve imbedding.
 -7 3-

A. Andreotti

The proof of this t heorem i s r "ther c omplic ;,t ed.cnJ will be

omitted. However t he interest of t he t heorem r es i de s i n th e
f act th ctt it do es not hol l f or dima: X = 2.
We will indic ate ho w to con struct <;. ccun t er- ex amd, e ( c f. [10])
based on the n ext

Lemma. Let
V be a c onnected comp act manifold. Let W ~c.. U
~einrrr open subsets of V. Let ~ be a c onn ect 2d complex
manifold and let
1/: X7V-W

be an holomorphic map malting X int o a ) -sheet ed non-r ami ned

covering of V- w.II dima: V2.2 and if V i s simple con-
=
nected either A 1, or else X c a.ino t be comp actif'J ed
be isomorphic to an open s e t of a c omp act
(i. e • X .;:,c.;;;an=n;.;;o..:t;.-;;;.;;....:.;;;..;:.;:=:..;;,;;.:.;:;.....:.;;.-=~==~..:;.:~:.:.....:::....::.::.:::.I;:.;;::;::..:.
complex s Dace).
The re ason for the validi ty of this l emm a can be said briefly
'V
as follows. If X woul d co mp actify int o X t hen on e c an s how
th at the holomorphic map 11' extends to a holomorphic map
~ N
rr:X~V.

Since V is simply connected the map tr must be r amifi ed, if

..l > l, somewhere i n V. But the r ami fi c at ion s et mus t be anal-
ytic and contained in U i .e. it is a compact an aly t i c subs et
of a Stein manifold. Therefore the r ~ ific at i on s et cons ists
of a fini te s et of points E.
Since V is s imply co nnect ed, non-singul ar, of dime nsion 2.. 2 ,
also V-E is s imply connected. Therefore _~l(E) is an X
irreducible non-ramifi ed covering of V-E. But t hi s is po s sibl e
only if J- = 1.
Remark. One could also assume th at X is r cvaifi ed ov er V
along an analytic set F Co V-U such th at V-F remains s i mply
connected.
(1)
A Stein manifold is a hol omorphi c convex manifold on which
holomorphic functions separ at e po ints.
 -74-

A. Andreotti

Construction of the example. Let T ~21r be an 8lgebr cu.c =

torus of complex dimension two .L e f'Ln ed as the quoti ent of ~Z
by the group r
of translations gener a t ed by the vector columns
of the matrix

r: ~2J
0 zn zll z12
where Z =
1 z2l zZ2
z21 z22
t
is a point of Siegel half plane HZ' z = Z, Im Z ) O.

LetT: T ~ T the map which associates· to each point of the

2
group T its inverse. This map is involutive, 7 = I, and
has 16 fixed points. Let K = TIT; this is a complex analytic
space of dimension Z with 16 conical, non-d e gene r-at e double
points. A theorem of Kummer shows that K is isomorphic to an
algebraic surface ~o of fourth order in F3(~) with 16
isolated t non-degenerate, conical singulc~ points. Select 16
disjoint small balls U<1 in lP ' 1$ a. ~ 16, centered at the
3
singular points of ~
o
and let U \? 'U.
=q,=1 <i
Consider a surface

¢~ of fourth order in lP3(~) close to .0 and han-singular.

It is known th at ~£.' as a non-singular surface of lP , is

3(lC)
simply connected. Moreover

i) UE. = 4>e..(lU is a stein open subset of

11) 4' -U is di ffeomorphic to et> - U
E. E. 0

Therefore <Pf. -U e. admits a non-ramifi ed 2-sheeted covering, the

torus T where 16 sm8l1 neighborhoods of the 16 fixed points
of,. h ave been r emoved. Let X be t ,'is double c over-i.ng , We
give X the complex structure that makes the natural map
11" : X -<> q, e -UE:. ho1omorphic.
Since 4>€. -U e is O-pseudoconcave X is also O-pseudoconcave.
Moreover if F denotes the ho10morphic line bundle of the hyper-
p1ro1e section of ~
e , J~(4
7 e-Uc I F) gives local coordinates,
 -75-

A. Andreotti

everywhere on ~ -u. The s ame is therefore true for

t
"q (X,
e...
tr *F). But th e preceedin c: Le.nma tell s us th cct X cannot
be c ompa c t.Lf'Led , in p ar ti c ul ar X cannot c,d mi t a proj ective
i mbc:dd 1 n g .

Remarks.

1. We h av e constructed an exam ple of a O- fJseudoconc ave mani f ol d

th ,' ,t canno t be compxc td f'Led t a c t ual l y if we let c- vary D. little
on a small disc A around C, we h ave constructed a f Uinily
U ~ fX tS t '£ t1 of 2-dimensional msnd r oLds , all O-pseudoconc ave

and no one of which can be compactified.

2. Makin g use o f the r emark made a f t e r the lemma one can show
that example 3 of (2. 3) gi ve s al so a non-sc ornpac t.Lfd abl e
c omp l ex s t r uc tur e on lP ( G: )
2
-1 Or.
 -76-

A. Andreotti

Chapter V•. Meromorphic functions on complex spaces.

5.1. Preliminaries
n•
a) Let Jl be an open set in a: A closed subset A c Jl. ::Is
called an analytic set if A is locally the zero set of a
f1l1.1te number of holomorphic functions i.e. for every Zo ~ A
there exi - ts a neighborhood U of Zo in .....11. and
f , .•• , f 0; "l1(U) such that
1 k

On A we define as the sheaf 0A of holomorphic functions,

the sheaf of germs of restrictions to A of holomorphic func-
tions in some open set of Il.. A morphism (or holomorphic map)
from the analytic set ( A, D- ) to the analytic set (B,
A
B) e
is a continuous map 1'-: A ~ B such th at

~ • crB,T(x) c=. & A, x for every x e A.

The notion of isomorphism is then defined and then the notion

of complex space is obtained repl acing in the definition of
complex manifold open sets of a numerical space and isomorphisms
between them by analytic sets and isomorphisms between them.
The "structural she af" will be denoted by &- •
A point x £ X of a c omplex sp ace X is called a simple point
if some neighborhood U of x in X is Lso a.o r phf,c to some
open subset of so me a:n. Let SeX) be t he set of non-simple
points of X, t his is c all ed the singular set of X. This is
an analytic subset 0 f X as one can prove.
A complex space X is cal.Led irreducible if X-SeX) is a
conn ected m c~i fo l d ; its dimension is called t he dim ension of X.
The dimension of E~ arbitr-~y c omplex s pace X is, by definition,
the maximal dimension 0 fits components. If Y c:. X are complex
sp aces t hen dim Y So dim X. If x E X is a point o f the
co mpl ex sp ace X we Jefine dim:x:!. as info dim U( x ) when U( x )
describes a fund wnent al system of neighborhoods o f x in X.
Alw ~ys holds: dim SeX) < dim X.
 - 77 -

A . Andreotti

b) The notion of meromorphic function on a complex space is

defined in the same way as for c omplex manifolds. Given a
complex space X we can then define two rings of meromorphic
functions:
~(X), the ring of meromorphic functions which are quotient of
two global holomorphic functions on X, the second of which is
not a divisor of zero.
5t(X) , the ring of all meromorphic functions (which are locally
the quotient of two holomorphic functions, the second of which'
is not a zero divisor).
If X is irreducible, as we will alWayS assume in the sequel,
~ (X) and J< (X) are fields.
c) A complex space (X, e)
is called normal if for every
x ~ X the ring cr x is integr caly closed in its quotient ring.
A complex space (X.·<9-) is called weakly normal if every con-
tinuous meromorphic function f : U -'>' II: defined on an open
set U G X is holomorphic on U. Given a complex space (X,~X)
there exists a (normal) complex space (X, &' i) and a morphi sm
/'
p : X -) X which is surjective and such that for any normal
complex spa.ce (Y, By) any morphism Y -->X such that the im~
of any irreducible component of Y is not contained in seX)
A

f actors through X

'"
/1 X

y~l p
- X
The s pace (X, C1f ) is uniquely lefined up to isomorphisms and
is cxl. Led the normali z ation of (X, l9).
Analogous 's t at ement hoLds r-ep Luc Lng the word " normal" by "weakly-
normal". We thus obt af n the notion of wea\t··normalization space
of (X, 6 ) . A sp ace X and its we.:"k normalization are homeo-
mor phi c . A no r -na'l sp ace is we''';';:!y normal.
 -7~-

A-Andreotti

Examples.

1. The space x t
= (x, y)~ 0: 2 I x 3 = y2 ~ is neither normal nor
weakly-nermal as the function ~ is continuous on X and inte-
gral over the local ring of X at the origin as it satisfies
2
the equation (l ) _ x = 0 on X.
x
Its normalization and we:ik normalization is the complex space
0: with 0: ~ X defined by the map t ~ (t 2, t 3 ) .

2. The space X = I (x, y) <: 0: 2 I x 2_ y2+y 3 ,= 0"5 is not normal

as ~ is integral over the local ring at the origin but not
2
holomorphic; it satisfies the equation (~) + y - 1 = 0 on X.
p
Its normalization is 0: with 0: -"7 X defined by
t .~ (t( I_t 2 ) , I_t 2). However X is we ak l y normal. In fact if
g is holomorphic and continuous ne ar the origin on X goP
must be holomorphic in the neighborhoods of t = 1, t = -1
and assume the same value ct at t '" 1 and t = -1. One
verifies th .tt any holo",orphic ~nction 0 f this type can lE written
as a series of the form a + I 1 a t r(1_t2)r+s, but this
r+s= rs
is the pull back by p of the holomorphic function near the
ortgin in 0: 2 ex: + ' f a x r y s
r+s=l rs •

3. Simpler examples can be obtained by t acing into consiJerat±n

reducible s paces. So for inst~ce
2
lex, y) E 0: \ xy = 0"5 is not normal b~t weakly normal;
{(x, Y) € 0: 2 1 (x_y2) x = Os is no t weakly normal, its weak
nor mali z at i on i s isomorphic to the pr eceding s et.

As reference one can consult [29], L1,2 ] and [8 ] for the

we ~~ normali zation.

5.2 Pseud oconcavity for cO{fi 11ex sp aces.

The notion of paeud oconc av e mm i f olJ c an b e extended to c ompl ex

s p aces with the fol lo\'l .'.n.:; :J e f i ni t i on .
 -79-

A. Andreotti

The complex space (X, f}) is caL.LeC1 pseudoconc ave if an open

sutset Y # f1 of X is given such that
L) Y CG X ;

' i i ) For every zoc,: oy there exists a fundamental

sequence of neighborhoods in X such that
------..,
(Ut) n Y)u is a neighborhood of Zoe
\J

Here

(U~Y)U =i Z € Uv I I fez) 1~us~~ I f I for all f E. )f(U,,'> •

v II

This is generalization of the notion given for complex manifolds.

in fact if X is such a manifold and Y ec. X has a smooth
bound~y and if V zo';; d Y the Levi-form restricted to the
analytic tangent p:\,ane has one negative eigenvalue, one can
construct, using the disc theorem, a fund amental sequence of
coordinate polycilinders P v centered at Zo such that every
holomorphic function on P v n Y extends to P. Taking
Uv = P v the above condition ii) is then fulfilled.
For a pseudoconcave space one h as the equivalent of
theorem (2.5.1) (c r , [lJ).
theorem (5.2.1). If X is a pseudoconcave sp ace of dimension
n , then the field 1«X) of meromorphic functi ons on X !!
an algebraic field of transcendence degree d ~ n.

We remark th at fr om condition ii) it follows again th ~t every

holomorphic function on a pseudoconcave space X is constant.
Therefore ~ (X) = e,
~
5.3 The Poincare problem.
a) If we drop t he assumption of pseudoconcavity there is no
hope to obtain a st 3tement of the n 3ture of theorem (5.2.1); we
h ave already remarked th at for X =
a:, the functions
z2
e~, e ,... are algebraically independent and. therefore, the
 - 80-

A. Andreotti

transcendence degree of }(a:) is infinite. However if we take

as groun l field 11(X) inst€ ~d of ~ the situ ation is more
hopeful. First of all one h ,s the followin g useful fact

Theorem (5.3.1). If ~ is a normal sp ace, t hen t~(X) ;h§.

algebraically closed in 7( (X).
The proof is the same as the proof given for man d f'oLd s
(theorem (2.2.2». Moreover the previous counter-ex~aple disap-
pears as one has:

Theorem (5.3.2).
(a) I f X is a Stein so ace(l) then & (X) =1l(X).
(b) If X is an open connected subset of a Stein manifold
~ Gt (X) = :r(X).
(1)
A Stein space (or holomorphically complete sp ace) is a
complex sp ace. X (with countable topology) satisfying the
following conditions
(d ) /I(X) separates points Le. if x'; y, x, y" X, there
exists an f <=. H(X) with f(X)'; f(y).
(ii) for any divergent sequence tXi~ c. X there exists an
f E, 6'(X) such that
sup I f(X i ) I = 0".

A space X . in which (ii) is satisfied is called holomorphic~

ally convex. A sheaf "1 of 19--modules on a complex space X

is called coherent (cf. (23J) if for each x ~ X there is a
neighborhood U of x a nd an exact sequence of the form

In particular the sheats e P, p 21, are all coherent.

Kernels, cokernels, im ages, coimages of homomorphisms of
coherent she aves are coherent. For a coherent she ,,! '! on a
Stein sp ace X the space t (X, 'T) generates 'J x over r:9 x
 -81-

A . Andreotti

~. (a) Given h € ?((~) consider the she qf

lJ = f f ~ r3 I fxh x lS ex for x 6 X •

Locally h = ~ with P and q holomorphic and q not a

q
zero divisor. Thus loc ally

!.J =1f e l E: tp ~ 0 (mod q) S

= Ker ~ <9- ~ 6J /q 0 J
Since 19, q e and thus & /qe are coherent, .it follows that
/J is coherent. If X
o
€. X is a point where h is holomorphic
we h ave !J x o = r!3 X Therefore by theorem A there eXists
o
S I: ['(X, j ) with s(x
o)';'
o. But, by the definition of tJ ,
sh =t ~ E (X, c9 ) • Thus h =.is and s _ 0 (since. X is

assumed irreducible), hence h ~ t9 (X).

(b) 1 Let US .tssume first th at X is an open set of ~
(a particular Stein manifold). Given h e}{ (X) and 1enoting
by 71( t he s he af 0 f germs 0 f meromorphic func tions on
n
11:, h
defines a section h : X -? J'J( of the s t a c .ced a p ac e 7?( over
the open set X.
J1f .
A
Let X be t he connect ed component of h eX) in Because
n
the n atural pr o j e c t i on .".: ?J1 -i' a: is a local homeo morphism,
X a c qui r e s a n atur ~ compl ex structure of an n-dimensional
complex mani f ol d . This l!l u.nifold is Hausdorff as th e st ac ked
A

sp ac e ??( is Huu sdor ff", By th e very meanin g of X, t h er-e

for every X .c X (t h eorem A of H. Cart an). Every complex

N
subsp ac e of the numberical s p ace a: is a S t e i n sp .c e , Every
Stein s p ace (0 f finite d Lmc a s Lon ) a :lmi t s ;l. p r op e r o n e-to-one
holo mor p hi c me\:) int o a nume r Lcal, a p . .c e a:N • For a Stein
man i f o l d t h .r t m .p c an o e chosen to be an Lso n o r-ohLsm onto
i t s t ma ge ,
 - 82-

A . Andreotti

exists a meromorphic f unction "h on" such th at

X
h I heX) =n-*h; hx
;0.

at each point of X the germ of is the

germ represented by i.
A A
Now X is the l~gest domain s pread over on which h ~n
can be continued. Therefore -log & ( ~ ) , 6(x) polycilindrical
A ~

distance of x from the boundary of X is a plurisubharmonic

function (c r, [31l).
Hence X" is a stein manifold. We can then apply to "h the
,,1\
argument developed in (a) and write h =:e.
;0.
/\

A
with p, q
q

" a"
holomorphic on X and Then "q heX) =rr * h and since
tr*is an isomorphism between X and heX) we get the desired
result.
(b)2 Let u& now consi der the general case, in which X is
an open subset of a Stein manifold Y. By the imbedding theorem
for Stein manifolds (cf. r281, [311) we may a3sume th at Y
is a submanifold of s ome numerical sp ace ~N. By a theorem of
Coquier-Grauert ( [ 23]) ~ o ne can find a connected neighborhood
U of Y in a: N and a holomorphic retr:<ction
i : U -,> Y, ley) = y for al l y € Y. Consider the mero morphic
I *h
f unc t a" on ". on ....., -lex). BY th e s peca' a;
~' c ase (b)l we can
find two ho Lo mor-c hf,c fu nctions p, q wi t h q ~ 0 on 1" -leX)
such th at q T * h = p. Since q ~ 0 there exists a mul t i
d (J
index d.. € :JNl s uch th~t D q IX ~ 0 while D q\X :: 0 i f
t~.1 <. I d. I • Then d
D (q T * h) = Dd p. Restricting to X we get
d d d
D al X h = D P IX wi th D q X~ 0
as we wanted.
The previous theorem gives the solution, in some particular
c ases o f the so-c all ed Poincare problem: when, for a compl ex
~ X, do we h ::tve <9 (X) = J( X ) ? Of course t his probl em
is not always solvable, for inst ance for comp act projective
irreducible al.g ebr-at,c v arieti es we have iR (X) = ~ but
 - 83-

A. Andreotti

}f(X) # ~ if the variety is not a point. One could hope th at

O(X) = 1«X) if /I(X) separates points of X, but this in
general is still an open question.

b) For a complex sp ace X which is not "holomorphically

s ep ar-a t e d ! ' one is L e u d n a t u r - a l . L y to consider the equivalence
rel at Lon tR on X

iff f(X) = f(y) for a ll l' <E #(X).

Let y = X/rte be the topological quoti ent space, and let

p : X 7 Y be the natural projection. A set U c Y is open
iff p-l(U) is open in X, thus Y is a Hausdorff sp ace. We
define for U open in Y
Cy(U) ={f: U-7~ I r continuous, P*f p-l(U)_>~
holomorphic •

We give in t his was to Y the structure of a ringed sp ace (a

.13pace with a she af of rings of germs of continuous functions)
and the problem a.ri s e s to see when (Y, e) is a complex sp ace.
If t his is the c a s e, then (y,(9) is a holo morphically aep ar at ec
sp ace and p : X 7 Y is a holomorphic surj ection. Moreover one
em th en consider 7( (X) as an exte n sion of the field p * '7t(Y)
which c ontains (]i. (X) and may be equal to ~ (X) if on Y
the Poincar~ problem is solvable, as one o b~ously has
~(X) = p*~(Y).
To decide when (Y,<9) has the structure of c' complex sp ace we
h ve the follovdn g sufficient criterion (of topological n ature):

Proposition (5.3.3). Let X be an irreducible weakly-normal

comnlex space If
(i) Y = X/(j( is loc ally comp clct,

(ii) p: l -;> Y is semiproper, i.e. for every comp _Jct set

K c. Y there exi sts a comp "ct set Ie' c. X s uch that
p(K') = K.
Then (Y, e,l) is a wealdy norlnalcomnle x sp .,ce.
 - 84 -

A. An d reotti

Ex ample. Let X be a we akly-normal, h o'Lomo rph Lc al. Ly c onvex

sp ace. Th e n p is a prop er map. Ln d e ed i f K c: Y is c omp act,
p-l (K) must b e comp -vc t , o th e rw t se we C ,1Il s ele c t it di ve r gent
sequenc e ~ xiS c p-l(K) an d thus f i..nd an f e !I( X) wi t h

~p If(X =d'. But sup I fJ = sup p.f where p.f is the

i)/
p-l(K) K
function on Y = X/~ whose pull b a 'c by p is f. Since
p • f is continuous this is impossible.
Moreover Y is Loc al. L y comp sc t , Let y € Y and x €. X
such th at p(x) = y. Since X is holomorphic ~lly convex. we
can find a finite s e t 0 f functions f '" f in fI (X) such
l,. k
th at the set

has a comp act closure K' . Then p(U) is a nei ghborhood of y

and is contain ed in p(K') which i s comp act.
We s atisfy thus the c onditions st ~ted by the pr e vious proposition
and therefore (Y. e) is an analytic ' sp ace whi c h is also weakly-
normal.
But on Y holomorphic f unctions separate points and holo morphic
convexity is i nhe r i te d by Y from X. Thus (Y .0 -) is a weakly
norm al Stein sp ace. In t hi s case we t hus h ave

Proof of proposition (5. 3.3).

(a: ) First we remark th at. given ~ c omp act set KG X we can
find fini tely many ho Lomoro hLc functions f •• " f in
l, k
Jt(X) s uch th at:

iff
 - 85-

A. Andreotti

Indeed the equivalence rel ation ,12 is r-epr-e e e n t ed in X" X

by the analytic set

f[ = f ~1/I(X) ~ (xl' X2 ) f. X- X \ f(x l) - f(X 2 ) = 0~ •

Locally dC can be given by finitely many of its defining

equ ations. Since K ~ K is compact finitely many of these
defining equations suffice to define rJ2 n K '" K.

(~) Sec ondly we show the follow ing fact:

Let X, Y be complex spaces and let ¢: X ~ Y be holo morphi¢
Let g: Y ~ ~ be continuous. Then
(i) if g is mer-omoz-phf,c so is ¢ *g
(11) if <t is semiproper and surjective then if <t * g is
meromorphic so is g.
This fact is proved making use of the following remark.
A continuous function h: X -;>~, .le f i n ed on a complex space
X is meromorphic iff the graph of h is an analytic subset of
X ~ 11:.
This is a p ),rticuLtr case of a general th eor-em (c r, [111,
prop. 3.12) •
In this instance if X is weakly normal there is ~most nothing
to prove; in general a sp ace and its weak norm ;>lization h ave
the "same" meromorphic functions. Making use of this remark,
st atement (L) then follows from the fact th at the graph of 4> * g
equals (~.id)-l (graph of g) while QXid: X x II: -"'Y x II: is
holomorphic an d the graph 0 f g is analytic.
Denote by rN and J' the graphs 0 f ~
*g and g respectively.
We h ave a commutative di agram
... X
r --» I'
1;r 1 fT

X~y

where rr and rr are the n atural projections and where

X. = (1/>)( id) 1 '::' •
1
 - 86-

A. Andreotti

Since ff. and tt are homeomorphisms 7\. is semiproper bec ause

4 is semiproper. Horeover r is closed in y" It. Therefore
the composed map

Y " It
is semipJoper and holomor-phic. By assumption d> * g is meromorphf.c
so th,t r is analytic.
A theorem of Kuhlmann (cf. L40J, se also [111, a generaliza-
tion of a theorem of Remmert l45) stutes th at the image of a
holomorphic semipro per map between complex spaces is an analytic
set. By virtue of th at theorem Em X.
= f mus 't be analytic and
therefore g must be meromorphic.
(Y) Let b ~ Y and let V be a neighborhood of b with a
compact closure ,.
Choose K G X compact such th at p(K) = V. Select
f 1, ••• , f k '"' II (X) such that

Let gi: Y -;> Ie be such that p * gi = fi for 1 s; i s, k, Con-

sider the map
x.:
defined by X(X) = (gl(X)" •• ' gk(X». Let L = (V). Then,
being continuous, L must be compact as V is compact. Since
X is gijective X must be a homeomorphism of V onto L.
Therefore Z = X(V) i6 open in L and there exa s cs an oo en set
fl c. le
k (open in lek ) such that Z = L /) fl. In particular Z
is closed in n .
=
Set W p-1(V) and consider the composite map
 - 87-

A . Andreotti

This is given by x ~(fl(x), ••• , fk(x», therefore J is a

holomorphic map which is also semiproper as p is semiproper.
By the Kuhlmann theorem ~ (w) = Z is an analytic set.
" " be the weak-normalization of Z. We get a
Let (Z,8)
bijective map .p: V .., z. " (7).
We claim that ljr is an isomorphism of (V, e V) onto (Z, '\

This follows directly by application of statement ($) and the

fact that on a weakly normal space continuous meromorphic
functions are holomorphic.
5.4. Relative theorems.
The previous considerations lead us naturally to the following
situation.
Let X, Y be irreducible complex spaces and let p X -";> Y
a holomorphic surjection. This gives an injective map
p* : '~(Y) ~ "Jf (X)
and we want to investigate when /(X) considered as an
extension of p*7f(y) is an algebraic field.
For instance Y could be the separation space of X and in
favorable conditions we would have IX' (y) =
G; (Y) ~ ci(X).
In this connection one can prove the following:
Theorem (5.4.1). I f p: X-:>Y is a proper holomorphic map then
IY(X) is an algebraic field over 'p *'If( Y), 0 f transcendence
degree ~ dim t X - dim t Y.
Example. Let X be a weakly-normal holomorphically convex space.
Let Y be the separation space of X. Since Y is a Stein
sp ace, If (y) = ~(Y) and p* ~(Y) ~~(X). We t hus obt ain the
following result as p : X - 7 Y is proper:
For a weakly-normal holomorphically convex sp ace X the field
It (X) is an algebraic field over q (X) of transc endence degree
~ dim X - ~imt Y, where Y is the senaration s u ace of X.
t
If in theorem ( 5.4.1) Y is a point we ge t back t heorem (5.2.1)
for a compact sp ace. One is le ad to conj ecture an analo gue of
theorem (5. 4.1) also for a pseudoconc ave s itu ation. This actu-
ally is t he case. For t his purpose we introduce th e following
definition.
 -8 8-

A. Andreotti

Let X, Y be irreducible complex sp aces and let f .: X -? Y

be a holomorohic surjection. We s~ th "t f is a pse udoconcave
map if an open, non-empty, subset f2 c, X is given with the
following properties
(i) f:n-'lY is a proper map
(ii) for every ye. Y every irreducible component of f-l(y)
intersects a
(iii) for every point Zo c; d Q. =Q- Cl.. there exist two
fundamental sequences of neighborhoods ~ U), 1 Vv ~ with
VtJ Co u.,; such th at

for every ye. f(U .. ) we h ave

~
(UI/ n il f (y))U nf-l(y) ::> V
v
n f
-1
(y) •
v

Theorem (5.4.2). ~ X, Y be irreducible complex sp aces and

let X be normal.
Let p: X -+ Y be a pseudoconcave map. ~ J( (X) ~
algebraic field over p. }«( Y) of transcendence degree
~ dim; X - dim; Y.

Remark. For the validity of this theorem actually one needs only
to suppose th at (a) p is pseudoconcave ·over some open set
A. c Y with A" fJ and such that p-l(A) is normal,
(13) for every compact subset K c:: Y we can find a compact set
K' c. X such th ::.t for every s « K, every irreducible component
of p-l(y) intersects K'.
The proof of this theorem is rather complic ated. It can be
found in [111 with all the material_considered in this chapter.
Remarks.
1. Theorem (5.4.2) is more general then theorem (5.4.1).
For example the family V = 1Xt f , where LJ =~. t e a: I \ t I < IS
t '-ll
eonstructed in 4.4, remark 1, gives an example of a pseudo-
 - 89-

A. Andreotti

co ncave map in which no on e of t he fib ers Xt can be

comp actified.
2.The assumption of pseudoconc avity i n t heorem ( 5.4 . 2)
c annot be relaxed. He r e is a co un t er ex .anpl,e du e to Kod aira
and Kas ( c f . [111)
Take count abl y many copies A of the unit disc /J. in IC, and
k
count abl y many copies •
II: k of a: •
Consider the sets

M = ku.2Z !.I k "- a: k•

On M-S co nsider t he cyclic gro up f gener ated by the

au t omorphis m

where t
k
~ .d
k'
w
k
* .
eICk Define the ac t i on o f r on S to be
t he i de ntity . One veri fi es t h ct X = Hie is a t wo-dimensional
manf fol d. The :IL1ni fold X looks like A " a: * in whic h [ 0 ~ x IC *
is repl aced by S (c f. ( 111 ) . We h ave a natur al holomo rphi c
map p : X -;> .1 an d one h ·.s
!I (X ) = p * 1-1 ( 6 ) •

Indeed, let f " /I(X) , then f J Ak ;<. IC • is given by a

k

holomorphic function fk = n=-

s; d a.
~K, n
(t) wn • in the Laurent

expansion of f
k ~,n(t) £ H(4). Now

-1
k l (t, w) = f k(t, t
f + w), fo r t ~ 0, t here f ore

= 0. (t ) t
- nk . This means th at i f n ~ 0 has
0 ,n ~ ,n
 - 90-

A. Andreotti

a zero of infinite order at t = 0, t h us u,n = 0 i f n -F. 0

a...

an d th erefo re in the exp ansion of f onl y t he t erm i nd epend ent

of w app e ars i.e. f <'.I/(.6).
It follows then th ).t 6{ ( X) =p*&( Ll) and th at /J is t he
reduction s p ace of X.
On X c onsider the functions given on !J 0 " a:~ by
2
f = w ; g = r tn
n e2Z
w
n

One verifies easily th ,;.t t hese de fi n e meromorphic f unctions on X.

ni 7
Set t = e with lm » 0 since I t\ <l. Set also
2ni z
w= e I then g is tr iJIlsformed into the Jacobi theta series
2
e( l;z) = ~ effi(n +2nz)
n "2Z
For any v alue of 7 B( i,Z) h as an 1nl£inite number of zer os of
the form zo(j) + 1 + k J' ; 1, k G 21a

We show th .:;.t f and g are al gebrai c all y independent over

~ (X) . If, on the c ontrary, f and g were al gebraic ;.11y
dependent over ct
(X) th ere would exist an irreducible polynomial
in X, Y, F(t;X,Y) '10 with coefficients th ,~t can be assumed
to be holomorphic in t 6 IJ, such that
F(t; f, g) ;: O.
Now for so me t = t 0 -F. 0 F(t o' X, 0) F 0, but this equation
f(t o' X, 0 ) = 0 should be sat i s fi ed for all v alues
2 Ri z 0 t 02k k € 2Z • This is impossible since a
X= e
polynomial in one v ari able h .s only f initely many z er os . The
conc lusion is th at the tr ansendence d egr e e of J« X) over 4) ( X )
is 2 2, whil e i f theorem (5. 4.2) is holdin g fo r X, one
should h ave a transcendence degree ~ 1.
 - 91-

A An ilr e otti

Chapter VI. E. E. Levi problem.

6.1 Preliminaries.

a) eech cohomology with values in a she a f , Let (f be a

sheaf of abelian groups over a topologic al sp ace X. Let
l{ = Ui i ~I be an open covering of Xi we set

for q ~ 0

where (iO'···' i q) runs over all (q+l)-tupl es extracted from

1 ••• 1 = Ui
I and where we h ave set U (\ U· () ••• I1U
1 i q
0 q 0 1

Thus an element f l, Cq(Z( ,--:1) is given by a coll ect1on

....<. t 1 ••• 1 ')

1 where fi 1 r, nUi l ' "1) •
0 q 0·" q 0'·' q

Obviously' CqCU, j- ) 1s an abeli an gr oup . We :fe ri n e

S : Cq (U , 1-) - ;> Cq +l ( ZI , '1 )

q
by the formula

q+l U "
i • • .1 j ••• 1 +1
= ~ (-l)j r O q
j=O U
i o · • .1 q +1

where

is the n ::.tural restriction homomorphism.

One verifies th at S q +1 0 ~q = 0, th erefore the s eque n c e
of abe11 an .gr-o upa and maps
o . <S o 1 1 '<r-f b1 b2
C ( 11, "] ) - 7 C ( L-(, ,r) -? - >

1s a cochain comp1es. Its cohomology in d imension q will be

d e n o t e d by Hq( 2(, "'1).
 -92-

A. Andreotti

Note th at, s ince is a sheaf,

HO( 2{, T) = Ker o O =l'(X,Y)

so that the above complex c an be c omp l e t ed wi t h the augment ation
map
° "? l' (X, '1) !.,.~ CO ( V , "J- )

The groups Hq( Z(,~) are c al l ed t h e eech cohomology groups

of the covering U with v alues i n "?f •

If V = 2" Vj S j~ J is anoth er covering of X by open ,s e t s

which is a refinement of U = ~ Ui~ i G-I then one can define a

refinement function J: J -;> I with the prop erty V j c. U-r(' j)'

By means of J one defines a homomorphism of comp lexes

,T* : .~( 2!, 1) -;1>C*( z.J-, '1)

by sett1~g Uj"{jo) •••1(j )

( i f) = r q
jO ••• j Vj j
q 0·· ' q

This defines a homomorphism of cohomolo gy groups

J ~ ..:
v
H* (?(, :J) -,) H* ( V, '1).
n«
One verifi es th at :r '2.-'-- d e oe nd s only on t he coverings 2{ , 2, Q-
b ut not on th e c h o ice of the refinemen t f uncti on. One c an t hen
de fi n e

Hq(X , '1-) = lim Hq( -z( ,c-y:: ) I q ~ 0,

-.;>
7.{
th e d irect limit t akc n o v er al l coverings U
.
of X. Thi s is
v
c all ed th e q-th eec h c oh omol ogy gr o up o f X with val ues in "5-
b) Homomo r phi sms o f sh e ~ves. Given two s h e :.v e s of abelian
gr o up s ( ?f, Jr, X) , ( 5', {~" X) over X a ho momorphism 0 f 7J
into ~ is a c ontinuous map d -:; -> ,~ s uch th at

L) u; o ¢ tr
ii) f or ev e ry x <c X t h e ind lI C ed n ap t x
is a
gr oup h omomorp hi sm .
 -93-

A ....Andr.eolli

If ~ is injective ""1 can be considered as a subsheaf of '!f

and one can, then clefine a quotient she af J I~, as the quotient
space of :1 by the equivalence relation

~ N ~ iff ~-~ t; P('J-x) ,

the topology being the quotient topology. One h~ thus

(~/'l-)x = fxl7- x '
A sequence of sheavesand homomorphisms

(1) °-> T::" f ~> 1/ -"? °

is called exact if for each x ~ X the sequence
e:
°~ "T'
J x
X
-> 1fx
(3 x
-:> Jlx-7> °
is an exact sequence. In particul :cr if 1 is a subsheaf Of';
the sequence
° -;> ~~ !i: -;> fl7 -;> 0 is exact.
Analogously one defines the notion of long exact sequence of
sheaves.

Given a sheaf homomorphism 1: j: .., 'j.., this induces a homo-

morphism of cohomology groups

t> It: Hq(X, T) -> Hq(X,,lJ •

One has the following theorem of Serre:

Given an exact sequence (1) of sheaves over a paracompact

space X then one has an exact cohomology sequence

°-> H°(X, T> trr' d. * HO(X'J)

(3*
~ HO(X,J()
6
-:>

1 ""I' d:.
..." H (X, J) -?'
* Hl(X,J' )
(3*
---;> Hl(X,J!> .~

,.,>H 2 (X, err.

j ) <:t *

where 8 is a "connecting homomorphism" which is defined in

the usual woy.
 - 94-

A. A nd r eotti

y
Cech cohomolo gy i s part i c ularl y useful if the topological
sp ace X. a dmi t s a s yst em of coverings ? ( d cofinal to all
coverings 0 f X. such th a t f or an y
U
i O " .i q
Hr(U i i' T ) =0 for al » > O.
o q

One h 9s in fact the fo l loWing important th eorem of Leray:

Let T be a she af of abelLm gr o up s on the paracomp act

space X. Let 1..( be an open covering of X such th at

Hr(U . i , ~) =0 for all iO, ••• ,i q, ~ q and all r > O.

J. O• •• q

Then the n atur al homomorphism

H ('2(, 7' ) -> Hq(X, "1' )

q
is an i s omo r p hi sm f o r all q ~ O.

Example.
Let X. be a c omp Lex manifold and "2{ = i Ui S be an open

covering o f X by open s e t of ho l omo rphy . Th en any U

i ••• i
O q
is al s o an open set of holomorphy. A theorem of H. Cart an and
J. P. Ser r e ( IITh e o r em BII) st~~~ ~h a:t: for any coheren t she af ":F'
on an open set U of holomorp hy (or mo r e general on a Stein
space) one has Hr(U, T) =0 if r > o. Thus coverings by
open set of holomorphy are Ler ay-coverings.

c) Ac yc l i c resolutions. Let vr be a she af of abelian gr01H!s

on the paracomp act sp ace X. By a r esol u t ion of T we mean
an exact se qu ence of she aves
d
( 2) o -) T~1o
dO
--::> l T ~ .r-J- 2 - '7
2

the resolution is called a c y cli c i f

Hq(X, 1-i ) =0 for all CJ. ) 0 and all i 2. O.

 - 95-

A. Andreotti

De Rham theorem. II (2) is an acyclic resolution of ".:r ~

Gech cohomology of X with values in "}' is n aturally isomorphic
to t h e coholomogy of the complex

[(X, '10 )
i.e.
Hq (X, ar'
j-) 'lJ
'1'
Ker i (X, .....:ag)
__--.. :i.:.;;::-

- Imtf(X, 7'q-1 )
Examples.

1. A sheaf of a b e l ian gro ups lJ- is called flabby if for every

open set Uc X the restriction map .f(X,JeJ -7 r (U, is
surjective. For a flabby sheaf j;:... one has that

Hq(X,X) =0 for all q> O. Every sh eaf ~ of a be li an groups

can be considered as a subshe af of the flabby she af t- 0 = sheaf
of germs of arbitrary (not ne&essarily continuous) sections of
'J-. Making use of this fact on e re alizes th at any ah e a f C?j-
ad mits a flabby (thus a c yc l i c ) resolution

2. A she af;:' is call ed ~ if for every closed subset

A c, X the restriction map r (X,;:') -'> l' (A' 5~) is s ur j e c t i v e .
For :~ soft sheaf ,X one h as again Hq(X,j.) = 0 f o r all q > O.
Therefore soft resolutions of a she af st are acyclic.

Let ·X be a different~able manifold ef dimension n. Let

denote the she ~f of germs of real-valued differe nti able C~
~ r L' r +l
forms of t y p e r (r = 0,1, ••• ) an d l et d : L/t """ .rr be the
she df homomorphism induced by exterior di f f e r en ti dt i on . The
kernel 0 f the h omomo rphism d: "f 0 ~ j( 1 is the she af 0 f germs
of c onstant fun c tions IR Since for e ach r t he s h t af uf r
i s a soft s he af we obt ain a se q uence
 -9.6-

A. Andreotti

One has d. d=O. Moreover the sequence is an exadt sequence

(~oiricaf~ lemma): We h ave thus obtained a soft resolution of the
sheaf 1& The De Rham theorem tells us that the cohomology group
H* (X, lR) can be computed as the cohomology of the complex
f Jt..* (X), d} of global dif f er en t i al flrms.
4. Let X be a a complex manifold of complex dimension n.
Let JI.. r,s denote the she af of germs of complex-valued C'" dif-
ferenti able forms of type (1', s) i.e. of t hose forms ~ that
can be written in a system of local holomorphic coord in ates as

Let j be the operator induced by exterior di f fe r en t iat i on with

respect t o antiholomorphic coordin at es. We obt cun in t his way
a sequence

0-;> ,-,'L r J; J(r , O s, J/.r,l ~.>

in which JLr = Ker t. Jl.r,o-- "~r,l~ is the she af of germs of holo-
morphic forms of degree r. The sequence is a complex as ad = °
and exact (Dolbe ault lemma). We obtain in t his way f or any
I' ~ ° a soft resolution of the she af J7. r.
In p art i cul ar for I' = °
)1.0 .= tJ and we ge t a s o f t resolution
of t he structure she af {} of X. We h ave thus

= Ker [ f eX, J O, k ) ~r(x,JlO, k+IH

for any k 2. 0.
ImtL'()!:.,.jt°,k-l).i [:(X,tAO,k ) 5

Reference s: [25J, [30].

 -9 7-

A. An d r e otti

6.2 E.E. Levi problem.

a) Let X be ~ n-dimension al complex manifold. To
clarify the exposition we will restrict our ut t ent i on to open
rel atively compact subsets D of X with a smooth bound~y.
We can thus assume th at D is de f i ned as follows. We h ave
g1 ven a C (1 func tion ~: X -) m wi th d ~ ;. 0 on a D and such
I
th at D =~x 6 X b(x)< o~(l).

As for the case X = tn we de f i ne D to be an open set of

holomorphy if for every zo" a D there exists an f co /I (D)
whi ch is not holomorphically extendable over zoo In chapter
I we h ave also proved th at if D is .S\Il op en set 0 f ho Lomo rphy t
then (*) the Levi-form at each point of 0 D restricted to
the analytic tangent plane to () D has no negative eigenvalues
(theorem (1.4.2 )). The Levi problem consists in asking if an
open set D with a smooth boundary and relatively compact in X
which verifi es the 'condition (*) is a domain of holomor9hy.
Vfuen no additional cond itions are pu t on the n ature of the
m ~ifold X the answer to t his question is negative as the

followin g example of Grauert shows.

Example. Let n 2 2; consider t he real torus

X=m
2n
/ 7Jn
and let 1": lIfn -7 X be the n atural projection. Let

be the unit cube in m 2n t so th at 1'(Q) = X.

Consider the following set
where 0 <. L <. t

(I)
It is easy to verify th at if D has a s mooth bound ary aD
this can be also defined by a global equation.
 - 9 R-

A. Andreotti

and let D =i(Q £). Then D is an open subset of X, rel atively

compact (as X is compact) with a smooth boundary given by
-rO xl =! E. 5 ).
: lR 2n
n
Consider now an lR -linear surjective map 1 II: ~
n
so th c,t lR 2n inherits through )" the complex structure of II: •
Since ~2n is a group of tr~slations, these are holomorphic
for every J-. we consider. Th erefore X, and thus D, is
provided with a complex structure. Moreover D in any of these
structures is Levi-fl at ~. for every

Zo € aD is identically zero. In fact this is a local property

since 7 is a local isomorphism it is enough to verify this
condition on r -l( d D) = 1s Xl = ! £ S. But now we c ant ak e for q,
the function 4> = xl :; E. whic h has i dentically z er o Levi-form.

The set Y = r ( ~ xl = On is a real compact s ubmanifold of D

1 -1
of real dimension 2n-l. Let /- be given by

x = (A, A) z el;n, A = ( a )
ij hi ~2n
ls.jm

In particular xl = 2 Re

fXl = Os is the union o f a cone - parameter f amily of co mplex

linear subsp aces

t " lR.

t
Let c = (0, c 2,···, c 2n) b e a r e .J. v ector in f Xl = 0 J s uch
th ~t the numbers c , c are line arl y independent over the
2,.·. 2n
r ation als. Since t h e mat rix 1\. h as t o s 'J.tisfy th e on l y cond i t i on
de t ( A, A) .#. 0, we c an ch oose A so th at

c f f ~l z l +••• + ~nzn = o}. Then fo r ev ery

t ~ lR 'I ( ICn~l) mus t be everywh e r e de nse in Y. Moreover

 A . Andreotti

y = tam
U
( ICni
1)
•

We can now show that fi'(D) = IC. Indeed, let f '" I/(D) and let
z0 be so chosen in Y so th "t !f(zo)/ =maJC If I. There exists
a t0 ~m such that zo'" T(a:n-lt ).
Y
0

By the maximum principle f must be constant on ., (ICn - l) as

to
for must be constant on a:ni l • But ., (a:ni l)
being dense in
o 0
I, f must be constant on Y bec ause f is continuous. The
set of zeros of a non-constant holomorphicfunction on a complex
connected manifold is of real dimension ~2o-2. Therefore f
must be constant on D as Y has real dimension = 20-1.
Since R(D) = II: every ho1omorphic function is estendable at
each boundary point of D and therefore D cannot be an open
set of holomorphy.

b) If we wish to obtain an affirm ative answer to the Levi-

problem, one need to reinforce the Levi condition (*). For
instance one could assume th at the Levi form restricted to the
analytic t 'illgent pl ane is nowhere degenerate on JD. Then we
have the following
Theorem (6.2.1). (H. Grauert, [26]). If D satisfies cond1t-
~ (*) and if the Leviform restricted to the analytic tangent
~T l an e to aD is non-degenerate (thus is positive definite
everywhere on JD) then D is an open set of holomorphY.
When the cond itions of this theorem are s .t i.sfied we say th at D
has a Gt ro ngly Levi-comvex boundary.
The proof of this theorem will occupy the next section. We
divide the proof in several steps.

6.3 Proof of Grauert's t heorem.

Step 1. A criterion for finite ness for cohomology.
Let X be a complex manifold and l et T be the ah e af of germs
of hoLomor-oh i,c sections of s o:ne hoLomo r-ohi,c VEct or bundle F.
 -100-

A. An dreotti

For every coordinate p atch U where Flu is trivial, we h ave

r(U, ~) ~ h(U)r where r is the fiber dimension of F. Now
leU), and . therefore #(U)r has the structure of a Frechet space.
If can transpose t his structure on r(U,~). Banach's open
mapping theorem ensures us th at this structure is independent of
t he trivialization chosen for F over U.
If' V is cIlpen and V e. U the natural restriction map
rU r(U,Oj) ~ (V,'f)
V
is continuous, and, if' V c.."- U, then r UV is a compact map by
Vitali's theorem. Similar considerations could be repeated for
any coherent sheaf' even if it is not locally free. Since for
the sake of the proof we need only to consider the locally free
case the corresponding argument is omitted.
Let -zo= l Wi! iE: I be a countable covering 0 f X by coordin-
ate patches, then the ~ech cochain groups

a s a countable product of Frechet spaces have the structure of

a Frechet space. Since the restriction maps are continuous,
the coboundary operator 6 is continuous.
Therefore in the complex

COCkl', 1") ~ Cl(tJ/j) ~ C2 (ut; 'Y ) i

the spaces are Frechet spaces and the maps are continuous Le.
it is a "topological complex of Freehet spaces".
Lemma (6.3.1). ~ A ,B be open sets of X such that
i) B e,c- A

ii) r~.: Hq(A,~) ~ Hq(B,~) is surjective

, •.,., ~ dimE Hq( B,<J ) < cP.

 -101--

A. Andr eotti

~.

(eC) Choose a count able covering U:: 1U~ :!eIN of A by open

sets of ho10morphy cont ained in coordin ate p atches on which F

is trivial. We may assume th at B!l U
i
# '1 for only finitely
many Ui's, say BnU i #,,'1 for 1 s, i So k,
In e ach of the open sets U
i,
1 i k, we can take an open
set U such that
i
k
B Co U
i::1

We now choose a countable covering Z/:: ~ Vj ~ j ~J by open sets

of jo1omorphy of the set B and such th at zr is a refinement
of the covering lUi n B i'i.isk'

For each Vj let i ( j) be such that Vj C U~( j)'

(~) We now consider t he cochain groups

Cq(Z(,,,;,):: n r<U i o"' i q ,"J) for A

q
C ( V-, s) :: n rrt j j ,"5) for B, _
0'" q

and the restriction map T* which is

defined as follows:
(J * f)j j:: f~(jO) ••• ~(jq) I Vj j
for
0'" q
0'" q
If we consider the Frechet structure on the cochain groups, then
*
'3" is a comp act map since Vj c
A
Uj(j)Cc. Ur(j) and since

th ere is only a finite number of groups [(U T(jo) ••• • (jq)' ~).

(y) Consider the map

<.J : zq( z{, '1) ~ Cq-1 (V , T ) -;> zq( Z'"-" y)

defined by
G) (<( ~{3 ) *
:: J<L + ~ .'3,

where Z denotes t he s p ace of cocyc1es. By the Leray t h eorem

 -102-

A. Andreotti

we have Hq('Z{, "'F)

= Hq(A, ';) and Hq(~ 'T) = Hq(B,':). By
the assumption the map r AB induced by '1'* among these groups
is sur jective; hence tV is. Since J * is compact, ~ =iX-l"*
has a closed image of finite codimension by a theorem of
L. Schwartz. (1) Hence 'dima: Hq(B, '3=') <. ~.

step 2. ti!Yer-Vietoris sequence. ~ X be a paracompact

topological space and let Xl' X2 be two open subsets of X
such that X .. Xl X2• ~ or
be a she af of abelian groups
sa X. One has the following exact sequence

0..". HO(X, Of) ~ HO(Xl' 'f) & HO(X

2
, :1) ~ HO(X l n x2 '''F )-;>

~ Hl(X, :n -7 H1(Xl' "f) & Hl(X 2, or )....,. Hl(X l (I X2!~) ~ •••

~. Choose· a flabby resolution of ~ on X:

o -;>':f-> eO -> e,1_> •••
One has the following exact sequence for every q ~ 0:
0 ..... rex, Bq) ~ f(Xl' e;q) & f(x 2 , e q) ~ r(X l (1 x2 ' e q ) -.;> 0

where ,,( a) = '\

1
(3 (a & b) = a/x X
1° 2
- bl x1ox2
Taking the firect sum over all q ~ °
we obtain an exact sequence
of complexes. Writing down the corresponding exact cohomology
sequence we get the desired result.

Step 3. The bumps lemma .

~ (6.3.2). ~ B be an open rel atively compact subset of
the complex manifold X with a smooth, strongly Levi-convex
bOundarY.
(1)
Let E, F be Frechet spaces, let u: E ~ F be a continu-
ous linear surjection and let k: E ~ F be a compact linear
map. Then u+k has a closed image of finite codimenBio~,
(c r , [28J. )
 -.103-

A. An dr eotti

Then there exist arbitrarily fine finite coverings 2( = ~ uj h i5.t

2! dB and for each U a sequence of open sets B , j
o s: j S. t, with smooth strongly Levi-complex boundaries such tha~

111) B - B c.c. U .!2!: 1 ~ i~t

1 i_l i

1v) Hr ( U () Bj,"F) =
0 for an;}' 1 and j, any coherent
1
~'"Y and for all r > o •
~. Without lo ss of generality, we may assume th at on a
neighborhood U of B there is a C tP function ~: U -> m,
such that B () U =It , u I~ (x) < 05, (dq)x ~ 0 for every

X E u, L(+)x) 0 for every x G U. The covering

U= ~U.} lSi~t will be chosen with SUfficiently small coordinate

balls such that U 0 B in the coordin ates .o f
U c:. U U 1s
1 1 i
elementary · convex and thus a domain of holomorphy.
We now select C<P functions p i ~ 0 on X, 1 5. 1!.. t, with
,w upp Pi
i
and such that ~ t?i (x)
Co C U >0 for all x €. a B.
Choose €.l, ••• ,E:. successively with E ;> 0 and SUfficiently
t 1
small so th at the functions

4>1 =~ -tlP l, 4>2 = Q- cl f'1 - t.2 P2 ' · · · ' 4>t = 4>-E l Pl - · · · - Ct t' t
have all their Levi forms L( ~ i)X > 0 for all x € U.

Define

B
i
=B l x U G U I <t i (x) < 01, for 1 c i s t ,

BO B. =
If the Gi are chosen SUfficiently small then

~Bi will be smooth for 1 s: i s, t ,

B (I U j will be convex in the coordinates 0 f Uj and thus
i
an 0Den set of holomorphy.
 -104-

A. Andreotti

we then have
(i) B = BO co Bl c, • • • Co Bt since I} =<P 0 2. tJi 2 ···~.pt·
(!i) B c.c; Bt since "'0 > ~t on J,B
O
(11i) Bi-B c<:... U has support csz: Ui
i_l i since 4 i-f_l =€iPi
(iv) Hr (Bi (\ Uj , "F) = 0 for r ~ 0, all i,j and all
coherent sheaves ~ since Bi () Uj is an open set of
holomorphy, by the theorem of H. Cartan and J.P. Serre.

step 4.
~ (6.3.3).
~ B a relatively compact open subset of the
complex manifold X with smooth strongly Levi-convex boundary.'
There exists an open set A ~ X such that
(i) B "=e.- A

(11) for any coherent sheaf ~ 2!!. X and any r? 0, !h!.

restriction map Hr(A, ~) Hr(B,~) is surjective.
£.!:2.2.!.
Applying the previous lemma, taking A = Bt, it
SUffices to ahow that for r > 0 the restriction maps
Hr(Bi,;t) Hr(Bi_l,:F)
are surjective for each k, 1 s; i So t.
Write B = Bi_l U(U 0 B and a~ply the Mayer-Vi etoris
i i i)
sequence:

Since Hr(Ui ~ Bi'~) = 0 = Hr(B i_l

n Ui, ~), the conclusion
follows from the ex actness of the sequence.
As a corollary of the above considerations we get the following
Theorem (6.3.4) Let B e, c; X be open with strongly Levi-conve>
boundary. Then for any locally free s heaf "1 2!! X we have
dima: Hr(B, 'Y ) < & for every r » O.
 - °105-

A. An d r e otti

Note. This theorem is true for ~y c oh e r e n t sh e af as lemma

( 6.3.1) c an b e carri ed o v er t o t his mor e g ener al c ~we wi t h the
s ame p r oo f .

step 5.
Solution of the Levi problem. It is e n o u ch to s ho w th ;;. t gi v en
a divergent sequence 2x li\ "- D we can find f '" H(D) such th at

sup f(x) = ~ ( c ond i t i on D of (1.3.1».

v v
Since D cc. X it is not restrictive to ass ume th at the sequence
x converges to a boundary point zo ':; ~D.

Choose a coordinate p atch U around z with coordin ates

o
zl, ••• zn' z ero at zo' such that

DO U =~z" U 19(Z)< 0 5

where et> U -? lR is a C <Y function with d4 ~ 0 in U and

L(~) > 0 in U. We can write

~ (z) = Re f(z) + L(4)zo (z ) + O( II zll 3.)

wi th t 6 If (U) •
It U is au fticiently smdll, then 1f = 0 s(\ D = fj. Choose a
C .,. function p in U with p 2. 0, a upp p <: e, U, 0 ( 0 ) > 0,
and set d:> 1 =~ -!.P. TaIte ~ ;> 0 s o small th at d~ l ~ 0 on U
and L(h) > 0 on U.
Set
A .. D u ~ X € U ¢l (x) -c OS.
T~en A has a smootq strongly Levi-convex bound ary an d t is
holomorphic on A n U. In particul n- by theorem (6.3.4)
d1ma: H
l(A,
9) =r < oP •

Write A = D U (A () U) and apply the Mayer-Vietoris sequence

for the sheE!-f () :

o -7 HO(A, L9 ) -;> HO(D, 6) ~ HO(A (\ u, e ) -? HO(D () U, f) ) §..

~ Hl(A, & ) -r
 -101;-

A. Andreotti

Consider the sequence of functions ;lJ for jJ = 1, 2, ••• in

D n U. These are ho1omorphic and linearly independent over ~.
1 ~.
Since dim~ H (A,\.7) = r we can find constants c l"'" c r +1 <= I:
not all zero such th at
1 1
b(c 1 1 + ••• + c r +1 fr+l ) = O.
Therefore there must exist holomorphic functions
h <!!! II(D) and h .,; hI(A U~ such th at
d AO U

Hence m is a meromorphic function in A, holomorphic in D

with principal part at Zo
r+1 1
c i t1 t
For the function
g =hn we thus h ave
I g(x)/ = 0". Q..E.D

Remarks.
1. Let us call a manifold X O-nseudoconvex if there exist a
C <f> function 4 X 7 m and a compact set K <... X such that
(i) for every c eo m the set
B
c
=~ X ., X I ¢ (x) < cS
is relatively compact in X
(ii) the Levi form L(t)x of 4> at every point x ~ X - K
is positive definite.
A O-pseudoconvex manifold for which we can take K empty is
also called O-comp1ete.
It can be shown th c~ theorem (6.3.4) extends to O-pseudoconvex
manifolds as follows:
Theorem (6.3.5). For a O-pseudoconvex m;.<nifold X, and any
coherent she af "'t ~ X one has
dimll: Hr(X,"f) " 0" for all r > O.
If moreover X is O-complete then
for all r > O. (d. (2 ] ) .
 -101-

A . Andreotti

The proof is obtained from theorem (6.3.4) by the Mittag-Leffler

procedure making use of the fact that for s;p ~ c <. c the .<;
l 2
restriction map
HO(B c ,'1)
-? HO(B
c ' 'r)
2 2
has a dense image.
2. Let D be an open relatively compact subset of ~n with a
smooth boundary and such that there eXists a C ~ function
4> : a: n ~ JB wi th the properties
D =t Z € cf I ~ (z) < °S,
d4> oJ °on a n, L(~) LOon some neighborhood of ;}D. Then D
is an open set of holomorphy.
Indeed, replacing <p by an increasing convex function of d we
may assume that there exists on D a function ~ : D ~ lR with
t ~ < constant3- cc. D and with L(e» ~ outside a compact set. °
Then if c is a large ~ositive constant the function
t
V- = ~ + C 1zil2 has the property {'f' < constants c.c. D,
L('Y') > °
on D. It follows then that D is an increasing tudon
of open sets of holomorphy and thus a set of holomorphy.
One could also apply the previous theorem of vanishing 'f or
cohomology to derive the f~t that D is an open set of holomorphy.
3. Let D =f x e X I 4> (x) < Os be a strongly Levi-convex
relatively compact subset of the complex manifold X. We have
proved that for any coherent sheaf Hl(D,jf) is finite-dimen-
sional. If L > °
is sUffiaiently small then
=
DE; 1: x e xl
¢J (x) < - e f has the same property. Moreover one
can show that the restriction map
Hl(D, '1') -.> Hl(D ,"J')
e
is an isomorphism, ( 2). It follows in particular that
#(D) = HO(D,e) separates points on D-De•
In fact let us consider the sheaf ~ of germs of holomorphic
functions vanishing at the points x, y E D-Dg• :J is a
coherent sheaf and we have the exact sequence
II: ,~ 0.
y
 - 10 8 -

A. Andreotti

From t his we Qeduce the commut ative diagram with exact rows:
t3
HO(D, $-) ~ II:
x
a II:
y
-;> ~(D, ~ )
t Y.j,\
'"
HO(D ,(}) ~
° ..,-> Hl ( D , ; ~)
Since I is an isomorphism, 1m 13 = ° thus c£ . i s s urjective.
4. Let D be as in 3. Consider the analytic set
C = ~ (x, y) .; D }<. D f(x) = f(y) fo r every 1' '' # ( D ) •

Cl early C .::> the diagonal A 0 l' D " D. Consider the closure

S =a-A of C-~. This is an analytic set. Therefore 5 n 6
is an analytic set. Any compact irreducible an alyt i c subset of
.1 of di men sion ~ 1 mus t be cont ained in S 11 11. Moreover
since N (D ) sep ara t es po i nts in D-D,s t he s et S n 11 mus t be
comp ast.
Therefore:
!! D i s a strongly Levi-convex rel atively comp act open subset
of a complex manifold X t here exi - ts in D a compact analytic
subset A of dimens i on.L l at e ac h po i n t such th ,;.t an y i rreduc-
ible comp act an alytic s ubs et of D o f dimension ~ 1 i s containec
i n A.
Thi s argum ent c :~ be c arri ed throu gh for c omplex s p aces.
6.4. Char ac teriz ati on o f pr o j ecti ve al gebrai c man i fol ds ,
Kodaira l s t he orem ( 0 : J) •
As an applic at i on o f th e p revious c onsider at i ons we gi ve here
a oroo f of t heorem (4. 2 ~l) o f Kod at r- a ,
We have alre ady given t he pro o f of t he i mpli c ati ons A < = 7 D.
We will give here t he »r-oo f 0 f t he impli c at i ons A = ? B = C = ) D

A = ~ B. It is enou ;;h to pr ove t hi s for X = Pn(II: )· Let

zO' . . • , Z n be homogeneous c o o r-a Ln a t e s in j'n (lI: ) and l et

U
i = zi -F ° wh ere we as s um e .>..8 c oo r d i.nrt es

Zo zi_2 zi _l zn
Yl = - t···, Yi - l = , Yi = ~, Yi+l = zi+l t ••• t Yn = -
zi zi a, zi zi
 -109-

A. Andr e otti

Consider the hernd.tian form dSl' =)" J.~ log g ziZi where

A 4: m and ). > O. Then dS). =J.. o.J log (1 + j=lYij) on U

i
and by direct calculation we get

~dS] = (1 +~Yij)-2t(1 +~Yij) rdYjdYj - (ZYjdYj)("iYjdYj)S

2. (1 +iYi;-2 ~dYjdYj.

Thus dsr defines a hernd.tian metric on Pn(t). Its exterior

form is given by the 2-form
i..l
w), = i3.,~
~ log i
(J'
n -
zi zi ="2 a~
d
n -)
log (1 + ~ YjY j •
o 0
By the w~ it is constructed d UJ),. = aw). = 0 thus d w",- =0 and
therefore the hermitian metric dsi is a Kahler metric (c r, (511).
Now H lZ) = lZg where g is a projective line !l(E)
2(J>n(t),
in Fn(E). Taking for g the projective line

g =~z2 = = zn = Os we get «J .~ l g =J.d

~
d-
2 where Y
(l+yy)

Therefore fg 4)). =;. Jg

l • But W g wl J > 0 thus one can choose
A to make t h at p e r i od an int eger (1)

B => C.
(~) The assumption B can b e re pl ac ed by the ass umption
th at t h ere exi sts on X a Kahler metric whos e ext erior form has
r ational period s as we c an al way s multiply th e me t r i c by a p os i -
tive integer to make the periods of t he corresponling exterior
form integr al.
Singular homo logy or cohomology b ased on d i f f e ren t iable sing-
o v
u1 ar simplic es will be d e no t ed by a s uffix "s", Cech c ohomology
by the suffix "v" and th e cohomolo gy o f the d e Rham complex of
di fferenti a b ] e forms b y the suffi x "dR".
(1)
 -119-

A. Andreotti

We then have the following commut ative diagram of isomorphisms

and inclusions:

2 2(X,
sH (X, ~) SH lC)
d.1'S d 1'S

JI2(X, ~) k
<---> JI 2 ( X, lC)
1"'5
JtI2 ( X, t)
d

If l' 2"" is a basis of sH 2 ( X, ~) modulo torsion, a

cohomology plass of sH 2(X, lC) is caracterized, naP', by its

values on -y 1'1 2 , ,, , and i t is in the image of j if its

values on y l' '( 2"'" are rational numbers. A direct calcula-

tion shows that; the isomorphism ~'oct'o T is induced by the map

which associates to a differential 2-form f> the singular
cochain

It follows that if a closed 2-form co has rational periods i.e.

fYi"'i: ~
then it is represented in eech cohomology by a 2-cocycle 1Cijk~
with c t: ~ (modulo coboundari es with values in t).
i jk
In this argUll8nt a: can be replaced by lR everywhere.

(~) Define the sheaf ~ of germs of complex-valued plurihar-

monic functions on X by the exact sequence of sheavea
o~ II: ~ e It $' ~ 11 -?> 0
where o.(c) =c It c and f.,(f Et g) =f - g.
Let ..{ r,s denote the sheaf of germs of C fP complex-valued forms
of type (r, 8). Set for U open in X r (U,J1 r,s) =J/. r,8(U).
we n ave the following
 -J1.l-

A. Andreotti

~ (6.4.1) !!ll U be a contractible domain of holomorphY

~ ~n. One has the exact seguence

° ...;.f(u,1-() . . . . j{0,O(U) ~ j{l,I(U) ~Jl,2(U) • A2 , 1 ( U) ,

!!l!!:.! d = ;) +;i is the exterior differentiation.

f!22!. Obviously the composition of any two consecutive maps in

the sequence is zero. Since U is a domain of holomorphy the
resolution

°-> a: 1"\
-T .I~
°d nl
~ .1<:..
d () 2 d
~..i~ -"7

is acyclic, ..Qi denoting the sheaf of germs of holomorphic

i-forms. Since U is contractible the complex of sections on U

is exact.
Let fO,O (;"q O,O(U) with <r SfO,O = 0. Then afO'O = fo,l
satisfies a fo,l = ° i.e. fO,l is antiholomorphic. Moreover

dfO,1 = ° thus fo,l = dl'o with gO,O ,,·.n>CW :l.e ~ anti-

holomorphic in U. Hence ;itO,O =JgO,O, thus

a(fo,O-gO,o) =° i.e. fO,O _gO,O = hO,O is holomorphic in U.

We have then fO,o = gO,O + hO'O, a sum ofm .antiholomorphic

and a holomorphic function on U. Hence fO'O,;; r (U ,ff). This
proves exactness at J? O' O( U) .

Let i , l ..: fl l , l ( U) with di,l = ° Le. Jol,l = °=5i,1

Now ~gl,l = 0, thus there exists a (1,0) form hl,O on U

such th at gl,l = jhl,O, because U is a domain of holomorphy.

Now ;)- ~hl,O = ° thus dhl,O = h 2,o is holomorphic and closed.

Hence h 2,o = dkl,o with kl,o holomorphic in U. Thus

a h l , o" ;;>k l , o and 3 ( h l ' O_k l , o ) = 0. Consequently

hl,O _kl,O:)k0'O tor some C '" tunction on U. Now

gl,l = )hl,o = 7(h l,0_k1,0) = 3 Jko,0. This s ho ws exactness at

--1'l,l(U).
 -11 2-

A. An d r eotti

0') Let
.
i» =i ~ g
", 7:
dz
d.
1\ dz",
r
be the exterior form of the
Kahler metric. Let "2{ = ~ Ui~ be an open covering of X by
coordinate balls. Since dw = 0 by the p r evio us lemma we can
write
(,<J Iu1 = ~;W1

Since (g~i) is hermitian, w is real, so that ".l lu = ~ a Wi '

and therefore 6J/ u = t ~a(Wi - Wi). We can thus assume Wi to

1
be purely imaginary.
Then on u1 fI Uj

Wi - Wj .. 21 Im Pij =Pi j - Pij ,

where Pij : U 0 Uj ~ £ is holomorphic an d determined up to

i
addition by a real constant.
On U (\ Uj (l Uk we must h ave
1

Pij + P jk + Pki = ci j k ~ lR.

Moreover c i j k - c l j k + c l1k - C
l1 j =0 on Ui (1 Uj (\ Uk () Ul•

2
Hence ! C1jk~ represents an element of ~ ( 2{, lR).
v
Now the cocycle Z. Cijk~ is the eech represent~tive of the
de Rham cohomology class represented by 4) (up to a sign).
Indeed we have wlu .. d(iw
i)
,
i

~.i - aWj = -aP i j .. -dPij' -(Pij + Pjk + Pki) =- c i j k'

which is the string of homomorphisms th at make explicit the is-

omorphism of the de Rham theorem:
By the assumption that C<) has integral periods, by (d.) we
derive that we can assume th at the real numbers ci jk are
integers:
(1) P1 j + P
jk + P ki =c i j k € ~
 -1l~-

A. Andreotti

__ e 2 1]"i P i j
(~) Let gij so th at stj is
holomorphic. ~cause of (1) we get

gij gjk = gik on Ui I' Uj () Uk •

Therefore the collection J stj"5 is a set 0 f transition functions

n
for a ho l.omor-phf,c line bundle F ~ X.
Now
-2Tl1W
where k = e i
i

-1
Let Vi =n (U '2::. Ui J4 II: and let z be the base coordinate
i)
and ~ i the fiber coordinate. One verifies that the func t { "n

j.. (z , e, ) = k i (z ) ISil2
is a well-defined function on F as on

Indeed j.'< z , E) defined a hermitian metric on the fibers of F.

Consider the tube of radius r around t he O-section of F:
Br = Z(Z''S) c F I ;x.(z,;) < rS.

For every r ? 0 B is relatively comp act as the base space

r
X is comp act. Moreover the Levi-form of log f ez , restricted s)
to the analytic tangent sp ace to 3 Br reduces to

J.'3 log f ez , t; ) = ,j . } log k

i
=2 1T L g _ dz dZ .s'
~3 d..

as the analytic tqngent p Lane to dB

r
is given by
,) log k
i 1
;) z« d ZQ + t: d~ = 0 , we can take as
i

coordin at es -alo n g th ,.t plane. He -roe the Levi-form re stricted to

th e analytic t angent plane to B is strictly p os i t i v e definite:
r
fhis s hows th "t t he c omplex space F is
O-pseudoconvex. If we
cons ider Lns t e ad of t h e bundle F t he bundle F- 1 which is
 -114-

A . Andreotti

given by transition functions [g~~f, then b, a simple verifica-

tion we see that the space F-1 is O-pseudoconcave.

C , _> D.

Let J be a sheaf of ideals on X, for instance the sheaf of

germs of holomorphic runctions vanishing on some finite set
SeX. Consider on F the sheaf '}' = n*J • ei e F' In

the previous instance, this is the sheaf of germs of holomorphic

functions on F vanishing on the finite set of fibers n -1(S).
Every t c. r (Vi' ':f) has a power series ex pansion

l' -
""~
- d.=O
Set

so

This gives a filtr ation of [(Vi,t) which is independent or the

choice 0 l' the fiber coo rf Ln at;e ~ i'
Simil arly one filters the groups r ( Vi i , "1) t therefore
o q
v
we get a filtr~tion of the Cech complex c- (V", °1 )

Which is compatible with the coboundary operator.

1)
",:.

If ~ ~ Cq(V; and ;. --~ et ~

=0
id.)
i o " .i q
(z)

one verifies th at, for given a ,

where F- denotes the she ~f of germs of holomorphic sections of

-vv
the bundle F-~ Therefore we get a (split) ex act sequence
(2) 0"7' C~+l u l'/1) ~ C~<Z;z.,'}) -:" CqU{, J ill !-k)-7' 0
 -.U 5 ~

A. And r e o tti

* _.-
Since the filtration of C (1 "', ';J-) is compactible with the
coboundary operator, we obtain a filtr ation of cohomology
H6< V,J) = Hq< U, ~J) ..:> Hi< 2!, :f) :-> H~< r;, r) .::> •••
while from (2) we get

Therefore
(3) s
k=O

q
From theorem (6.3.5), i f s > 0, H <Zf, J ) = Hq(F, T ) is
finite-dimensional as 2Y is a Leray covering of F. Thus if
q > 0 in (3) the left hand side is finite-dimensional. The
same must be true for the right hand side of (3), t her efore

.!2!: q.>O ther'l'!' eXist ko » 0 such th at if

k >.. k
O'
Hq(X, 0 fll J-k) = O.

Now we apply this result to the exact sequence

o 7 tJ fll F-
1/11
k ~ F-k -? Q
1./1F
~ 0

where ~ is the s he af of germs of holomorphic functions vanish-

ing at two di s tinc t points or vanishing of second order at a
given point. Writing down the corresponding cohomology sequence
one realizes th at the ring J« X,
F-
l)
k~ F- k)= reX,
separates points and gives local coordin ates everywhere on X.
Remark. One could apply theorem (6.3.4) instead of theorem
(6.3.5) replacing in the argument F by a tube Br of radius
r around the O-section of F t thU8 avoiding the use of the
Mittag-Leffler argwnent to go from the first to the second of
these theorems.
The extension to complex spaces is given by Brauert in [ 27J •
 -116-

A. Andreotti

Chapter VII. Generalizations of the Levi-problem.

7.1. d-open sets of holomornhy.
Let X be a complex manifold, let D be an open relatively
compact subset of X with a smooth boundary and let d be an
integer, d ~ o. Let ~ be a locally free shea! on X, for
instance cr =
t9 or ~ = o,d. We repeat the considerations
of the previous chapter replacing the ~pace #(D) = HO(D,b') by
the cohomology group Hd(D,~).
Let Zo IS cD, an element E, G Hd(D,:r) is said to be
extendable over Zo if there exists a neighborhood V of Zo
in X and an element ~ ~ Hd(D U V, "5) such that D =
Rem ark & if d > °
the necessary and sufficient condition for
an element ~ E Hd(D, to be extendable over ZoE: JD ~
that there eXists a neighborhood W 2! Zo ~ X such that
;)WnD .. o,
In fact, if c; is extendable over Zo' we can find a neighbor-
hood W of Zo in X which is an open set of holomorphy with
W e V, Then ~ I W=0 I
thus ~ WilD = 0.
Conversely suppose there txists a neighborhood W of Zo such
that ~ \ WilD :: o. We may assume that W is an open set of
holomorphy. By the Mayer-Vietoris sequence we get the exact
sequence :
Hd(D U W, '5) -7 Hd(D, '1) ~ Hd(W, r) ~ Hd(D () w, '3').
Since Hd(W,~) = 0, because d > 0, and since the im age of
~ in Hd(D fl W,"}') is zero, it follows that there eXists
~ € Hd(D ('\ W, 'J') such that ~ D I = ~.
We say that D is a d-open set of holomornhy with resp ect to
the she a! 'J"'
if for every Zo ~ ~D we can find S E Hd(D, "f)
which is not extendable over Zoe
 - 11 7 -

A . Andr eotti

For d =
0 we re ali zed ( The or em (1. 4. 2) o f E. E. Levi) th at
the bound ary of D has t o have a par-t Lcu'l ar shap e . In vi ew of
th at result we ar e Le a .i t o lDnsider t he fo l l owi ng s i tu at ion:
Let U be an open set in a: n and let of : U "7' m be a C ov
function on U. We set
fl- = ~.x € U ) f (x) < 0 5 I

and we assume t Lt on S = 2x E U I ~ (x ) = 0 S
d ~ oJ 0 t so th at
S is a smooth hyp ersurface, which consti tut es th e boun dary of
fl- in U.
Let us consider the Levi-form of ~ restricted t o th e analytic
tangent plane to S at every poi nt Zo G S, and let us 3.Ssume
that at a point Zo e S i t has p pos i tive and q neg at ive
eigenv alues (p+q .s n-l) • Then one can prove the following
Theorem (7.1.1 ). There exists a fund amental se quence of neigh-
borhoods U~ of Zo ~ U s uch th at, for qny locally free
~ ur
on U we h ave
s > n-p-l
=0 or
{ 0< s < q.

These neighborhoods c an be chosen t o be ope n s ets of hol omorphY.

(c f. [2]). The proo f 0 f t his theorem is r ather tedious and
will be omitted. As a coroll ary we get t he an alo gue of E. E.
Levi's th eorem:
Theorem ( 7.1.2). (E.E. Levi). If D is a d-open set of
holomorphy for a loc ally free sheaf '1 t hen at any point
Zo Eo en
The number of negat ive eigenvalues of L( ~) ZIT ~ <d
o Zo
(*)

\
the number of positive eigenvalues of L(~)z IT is ~ n-d-l.
o Zo
 -118-

A. Andreotti

One can then formulate the analogue of Levi-problem for

Hd(D, '1). The example of Grauert shows that condition (*) is
not sufficient to 'ens ur e th at it is solvable as in the eXdmple
any I/X>int 1. " OlD has a fundamental sequence of neighborhoods
o
U", such that D (l V'Ii 1s an open set of holomorphy. We have
thus to reinforce condition (*) assuming for instance that
L( ~) TI
zo zo
for every zo e:: ~ D is non-degenerate. An open set

D of this sort will be called strictly Levi d-coBvex. Then one

has the following
Theorem (7.1.3). (Grauert)d' If D is strictly Levi d-convex
~ D is a d-open set of holomorphY for any locally free
~ l' 2.I! X.
7.2. Proof of theorem (Grauert)d' (e r, (81). The proof
follows the same pattern given for d O. It is based on the=
following remarks.
(ct.) As a substitute for theorem B of H. Cartan and J.P.
Serre, we need only one h,uf of the vanishing theorem (7.1.1)
S
H ( 0U7'v, .-) =0,
'S- if s >d

since (n-p-l) = n-(n-d-l)-l =

d. Here :;: is any locally free
sheaf (although this part of the vanishing theorem holds for
any coherent she af Y. on X). Moreo\'er one has to realize
that this st atement is stable by small defoDBations of the
boundary, precisely
given any sufficiently small neighborhood Uv of zo ~ ~D in X
which is an open set of holomorphy and given any C ~ function
p : U -;> lR with p 2. 0, p(zo) > 0, supp p c..<- Uv ' we can find
Co >0 such th 21.t if we set 4'1 = t -£pwith

0<: t.. <. t.o' U)",) =tX € U,J4 l ( x ) c: O}

then we still have

for s> d •
 -'119-

A. Andreotti

(~) This enables us to repeat the proof of the bumps lemma

with condition iv) replaced by
r
i v) H ( u1 o B j , '1) = 0 tor any i, j, any locally free

sheat -r , and any r <: do'

Then the analogue of lemma (6.3.3) and the criterion of finite-

ness gives
Theor!! (7.2.1). ~ B open relativelY compact in X with a
smooth boundary on which the Levi-torm restrict!d to the analytic
taP,gent space has at least n-d-l positive eigenvalues. Then
tor any locally tree (or coherent) sheaf "f 2!! X we have

dima: Hr(B, 'r') <: & if r » d.

(1) We now have to construct for every z (, ~D a sequence
d 0
ot coho.ology classes .c:;ve. H (IV {l D,5') defined in a fixed
region • n D where W is a neighborhood of Zo in X, such
that
i) they are linearly independent over a:
i1) every non-tirvial linear combination ot them is not
extendable over Zoe
We postpone the proof of this fact to the end. We remark that
to achieve this purpose we will make use 0 f the assumptions
that ~ is locally free and that the Levi form on the analytic
tangent plane at the boundary of D is non-degenerate.
(8) Let us suppose for the moment th at point (,) has been
settled. Then the proot proceeds as in the case d = O.
Let A be constructed as in the solution of (Levi-problem)o
making a b uap on D to swallow the point Zo E: cD ,
A =D U (A () U).
We then apply Mayer-Vietoris
-tHd(A, "f) -:l> Hd(D,"J') • Hd(A (l U, "f) -? Hd(D o u, ':F) ~
.,Hd+l(A, "5) -t e.'"
 -12.0 -

We consider the classes C, '"- Hd(D n V, 5') constructed in (')').

Let r = dim~ Hd+l (A,~)
vr-
(which is finite by theorem (7.2.1».
We can find constants cl' ••• ' c r+l'" a:, not all zero. such that

6(cl~ + ••• + c r+l Sr+l) = o.

Therefore there must exist cohomology classes hn ..
d
H (D, s).
and h € Hd(A n u,'T) such that
M U
r+l
hn = i~ c i 8,i + hA(JU on D () U.
We set g =hn € Hd(D,~), then in any sufficiently small
neighborhood W of Zo which is an open set of holomorphy
hAoulw = 0 thus

is not extendable over Zo because of property ii) of (~).

(t) It remains to prove statement (1) :

Since rris locally free, in a small neighborhood W of Zo
in X, 'J= -::.
f} p for some p. Thus it is enough to prove the
statement for the sheaf 'J' = e.
The proof can be based on
the following
Lemma (7.2.2). Consider in ~n _ to} (n ~ 2) the differential
forms of type (0, n-l)
a. +1
i:(-l)j Zj j
It'd. = - - - - - - - - - - - - - - - - - - -

for ~ ~ ~n. T~ese forms are a-closed th erefore for any open
2.tl U in a:n_[o} they represent cohomology classes
Hn-l(U, tt). If U contains the closed half sphere

!. = {z Eo a;n \ ~ IZ j 12 =E, , Re zl ~0~

j=l
Then th e cl asses t& ~ Hn-l( U, l'-'-
7 ) are linearly inlenendent
(c f. [5]).
 -121-

A. Andreotti

~. The fact that the forms lfd are ~-closed can be

verified by direct computation.
Let '1 .. L cd f cl be a finite linear combination of those
forms with coefficients cd E E. Let us assume that Yf =a r'
for some C (9 form fJ of type (0, n_2)CL on U.
If ~. dZlA ••• Adz then for any holomorphic function f on
n,
En we have the generalized Cauchy formula:

leLl
SJf rJ .. ~2"i~;
w A n-
z c a..:..:..! (0)
a: az a
Let rJ > 0 be so small that the part

Sd II fl I Zj 1
2
:z f., Re zl Z - rJ of 5 S is contained in
U. We have

S6 5 flA)/I"l ZlSd f ft.:JA af"=sa'f d(fwflJ.!) =i)S! tio»,

Thus

~2ntt
n- I Idol!
L ca. CJdZd. (0).. d s! f IN 1\ jJ
f
+ S S f w II 1.
d -d

Let A=~ ~/ZjI2",e., +SSO[Rez l <::: -d/2J and

j=1

B =lj=1
~ I z j l2< d'/4}.

n
Then A and B are Runge domains in lI: • Moreover
Re( zl + d/2) is < 0 on A and >0 on B. Therefore A Ii B
is ~so a Runge domain in En. Let K be ~ oompact subset of
B containing the origin in the interior aQd let g be any hol-
omorphic function on B. We can find a sequence '£ r \J l of
holomorphic funct ions in En such th at

f \J '7 g uniformly on K, f ~..,. 0 uniformly on JS d U (S- Sd)

as K c B and a Sd v (S - S6) = (S - Sd) c tc, Therefore for

any g we mus t h ave

 - l.22 -

A. A n dreotti

lei I
c
d.
2...-...J!a
a.
(0) = o. Hence all CiS
<L
are zero.
az
Let us now cunsider the case d = n-l. By suitable choice of
local coordin ates at zo' oD near Zo is given by a local
~quation h = 0 and we may assume th at Zo is at the origin
and that

h = Re zl + [aijzi Zj + O( II z II 3)

where 1 ~jZi Zj »0 and where h » 0 corresponds to the side

of D. If' E. »0 is sUfficiently small then

{L I z j l 2 = cJ"5 {h > o] for any cI with 0 < cf <of.., contains

the closed half sphere ~~ I Zj 1
2
= a'S (\ f Re zl > O. By the

lemma above, in the coordinate ball W of radius EJ centered

at the origin the forms 1Va represent linearly independent
cohomology classes of Hd(W, 19'), not extendable over z00

In the general case d ~ n-l the local equation of aD near

Zo can be written as h =0 with

\z.1 2
h = Re Z
I + LLI -L
+ 2Re (z
. 1 ~~ aj-z
j)
+ dtl A
~ j
\z
jl2
-

where

Aj>O,lJj>O;

is positive definite and where the side of D corresponds to

the side h > O.
If W is a SUfficiently small polycilindrical neighborhood of
Zo = 0 in those coordinates, then

W (l t zl :& ••• :& zd+l = 01 (l f5 = f zo5 •

 -123-

A. Andreott i

Consider the forms of the lemma in t h e coordinates

Zl'···' zd+l'
all forms defined in W. These have singularities outside of
D. Moreover their classes again satiRfy property 11) of ("Y)
as this is true for their re.trictions to the s )ace
f zd+2 = ••• • zn = o} •
This achieves the proof of statement (1).

7.3 Finiteness theorems. In theorem (7.2.1) we emphasized only

the positive eigenvalues of the Levi form. With the same
procedure one can prove the following more precise theorem,
«t, [2] r
Theorem (7.3.1). k!i D be a relatively compact open subset
of the complex manifold X with a 8/Dooth. strictly d-Lert-convex
boundary. Then for any locally free sheaf '3=' 2A X we have
dim; H
r(D,1=') < d' i f r ~ d.

The group Hd(D,~) not considered in the ~heorem, is actually

infinite - dimensional and moreover has a natural structure of
a Frechet space. Theorem (7.3.1) can be considered as the
"intersection" of the following general theorams of finiteness.
A complex manifold X is called g~pseudoconvex if there
exists on X a C d' function + r X ~ :m and a compact set
K such that
i) the sets
Bc • f X" X I ~ (x) c C1
are relatively compact in X for every c" m
11) on X-K the Levi-form L(.) of p has at least
n-q positive eigenvalues.
If we can choose K. ~ the manifold X is called g-complete.

Theor.. (7.3.2). l! X is g-pseudoconvex then for any coherent

~ 0;' ~ X we have
di.
E H
r(X,1f) < ~ if r > q.
Moreover if X is g-complete then for r > q Hr(X, 'j") • O.
 -124-

A. Andreotti

Analogously we h ave

Theorem C7.3.3). 11 X is g-pseudoconcave, for any locally

tree sheaf "5" 2!! X we have

it r < n-q-l
Due to the maximum principle there cannot be an analogue ot
"q-complete" manitolds tor the pseudoconcave case.
Theorem C7.3.3) is only true for locally tree sheaves. Actually
the real analogue of theorem C7. 3.2) is that "for a g-pseudocon-
cave manifold and any coherent sheat 'S" sm X we have

.!2!: r » q+l ,
~ Hr k
denotes cohomology with compact supports. II
Theorem C7.3.3) is obtained from the one quoted above by duality
and thus, if st ated for any coherent sheaf, the number n dime X =
has to be replaced by the depth of the sheaf "J" •
As reference see [2], [7].
7.4. Applications to projective algebraic manifolds (c r , [9]).
a) We revert to the case d =
0 to begin with.
+
Let D be a complex manifold. Let COCD) denot e th e sp ace
ot positive O-dimensional cycles i.e. the tree abeli an monoid
generated by the points of D. We h ave

Pi' D} =D CI ) o
fl D D
( 2)v

wher~ DCk) denot es t he k-told symmetric product of D. For

each k L I DCk) i s the quotient of the cartesi an product
Dll' ••• Jt D··Ck- t i mes ) by the act ion of the s ymmetric group on k
l etters permuti ng t he f actors of this c -ll't e s i an product in all
pos s i bl e ways. As s uch D( k ) , and t herefore C~ CD ) . carries
a nat ur al struc ture o f ~ c ompl ex sp ace.
 -12 5-

A . Andreotti

Moreover we h ave a natural map

given by

for

Note that the image of Po is a set of ho1omorphic functions

on C~(D) which are additive with respect to the monoid
+
structure of Co(D).

Proposition (7. 1+. 1 ) . The sp ace C~(o) is ho1omorphical1y convex

(holomorphical1y complete) iff D is ho1o morphical1Y convex
(ho1omorphical1y compl ete).
~. If C~(D) is ho1omorp hic al1y convex (complete) t hen 0
as a connected co mponent of C~( D) has the same pr oper t y .
+
If D is ho1omorphic ally convex then also CO(D) is. In
+
fact let t Cv $ C COCO) be a divergent sequence. Let for every
c = ~ niPi deg(c) = In Then deg is a ho1omorphic function on
i•
C~(O) tellin g us on which s ymmetric pr oduc t O(k) the considered
cycles st ay. If deg (c y ) ~ & t here is nothing to prove. If
deg (c v ) s: k , then we can extrac t a divergent seqUer.c e ~ c '5
cont ained in a given product 0 ( 1 ) (with 1 S k). \I

Now Dl = D~Ox ••• xD (l-times) is ho1omorphical1y convex. Let

rr: 0 1 ~ D(l) be the natural pr o j ec t i on and l et ~ 1 be the
l
s ymmetric -gr oup acting on D • For e ac h c ~, let s 'll' .;; D1
be such th at 1J'(s li') = c"" The se quence { s" , $ must be diver -
1
gent in 0 • There fore there exists a ho1o morphic fu nction
g '" H(Ol) such th at su., I g( s ,) I = ry . Consi de r the
y v
pol ynomi al in t
11" l' 1'-1
(~ll (t - g(y x» = t . + ~( x) t + ••• + ~!(x ).

The co efficients ai ( x ) are holomorphic fun ctio ns which are

~l-inv ari antt hu s for e ach i t her e eXi s ts a hol o ~o rp hi c
function f € 17'( 0 ( 1 » with l7" * f = ,,\ . I S. i s. 1!. On t he
i i
 '- lZ6~

A. Andneatti

sequence sv, at least one of these functions ~, mus t h ~ve

unbounded acsolute value. Therefore sup !fi(sv')! =~.
If D i~ Stein then also C~(D) is Stein. This follows at
once from the fact that on a Stein manifold we can prescribe the
values of a holomorphic function on any finite set.
b) Suppose now th at D is an open subset of a prcjective
algebraic manifold X of dim; X = n , Given em integer d ,
+
o !; d ~ n, we can consider the space Cd(D) of positive compact
d-dimensional cycles of D i.e. the free ~belian monoid gener-
ated by irreducible compact analytic subsets of D of dimensio~
C~(D) = fL niA i I n i Eo:IN almost all zero, Ai compact
irreducible analytic subset of
D 0 f dim lC Ai =d ~
It is known that C~(D) carries a complex structure of a weakly
normal complex space.
Examples.
1.' D = X = Fn(lC), d = n-1-
C~_l (F n(lC» = ~(a:) V lPn +2 (D:) U F +
n
(D:) U •••
( 2 )-1 ( 33 )-1
2. X = Fn(!I:) I D = ~(a:) - ~ point} then

0;_1(8) = En u a:(n22)-1 u a:(n~3)-1 u •••

d d
Let us consider the cohomology group H (D, n ). Every cohomol-
ogy class S on it can be represented by a a-closed differentiaJ
form 4> d,d of type (d, d) modulo a of forms of type (d, d-l).
+
Given C b Cd(D) we can consider (by virtue of a theorem of
Lelong) the integral
S(c) = ~ d,d. f
c
One verifies th at this integral is independent of the choice of
the representative ~ of the class 5 ~
 -12 7-

A. An d r eotti

As analogue of proposition (7.4.1) we then h ave the fo11aWi~g

Theorem (7.4.2).
a) The function ; ( c ) 12!: c a variable on C~(D) ~
holomorphic function so th at we get a linear map
d
r« : H (D, n. d
)O '"
-p H (C
,C
d(D),L7).

b) !! D is strictly d-Levi-convex then for any di ver gent

sequence 1C) c C~(D) there exists a class t; E Hd(D, .nd )
such that
= <:0.

In particular C~(D) is holomorphically convex.

c) If D in addition is d-complete then given c l' c 2 ~
+ d d
Cd(D)!!!h c l '" c 2 there exists a class !; e H (D, ..Q. )
such that ;(c l) '" S(c 2).

In particular C~(D) is holomorphicallY complete.

We limit ourself to a v~ry brief skp.tch of the proof (cf. T9]).
(d) First one shows that the :unction S(c) is a continuous
function of c (this is the mo st di f f i cul t part).
(~) Since C~(D) is weakly normal it is enough to show that
+
~(c) iB holomorphic at non-singul ar p~ints of Cd(D). This is
essentially done via Morera's theorem. Loosely speaking, if h
+
is an analytic I-dimensional disc in Cd(D), if -y = d 4 and if
F
rr~ Cd(D)
+
is the fibered space of d-dimensional cycles over
+

r
their parameter space Cd(D), we h ave

=
: n-l(y)
f t/>d,d" dt =
17 -
1
(.6)
d( d> 1\ dt) =

17'-l(A)
h,;"'dt = 0
 -128-

A . An d r e otti

where t denotes the variable on ~ and c(t) = H-l(t). The

last equality sign is for reason of degree as n -1(,1) is a
d+l dimensional space.
eY) If D is strictly d-Levi convex then one constructs
the class e, using the local non-extendable cohomology classes
as we did in (Levi-problem)d •
+
(8) The separation of points on Cd(D) when D is
d-complete follows by an exact sequence argument from the
vanishing of Hd+\D,!L d) and the existence of a .:'i-closed (d,d)
form which h as non-vanishing integral over every cycle
+
c e Cd(D) (this is for instance the d-th exterior power of the
exterior form of a Kahler metric on X).
c) Consider, as an application, the following situation. Let
f : X ~ Y be ~ proper holomorphic map between the projective
al.gebr- ed,c manifold X onto the projective al gebraic v ariety Y.
Let A be a compact irreducibl e analytic subset of X and B
a compact irreducible analytic subset of Y such that
f-l(B) = A and f: X - A ~ Y - B is an isomorphism.
Set a = dim t A, b = d im~ B and assume th at a > b.
Then there eXists a neighborhood W of C~( A) in C~(X) which
is holomorphically c o~vex.
+ + + +
In fact f
induces a map f b (Cb(X) - Cb(A» -7 (Cb (Y) - Cb(B».
The cycles B, 2B, 3B,... are isol ated points of C~(Y) thus
n~lnB has a holomorphically convex ne ighborhood U in C~(Y).
One then verifies th at W = C~( A) U f-~(U) has the reqUired
property.
Now the eXistence of a neighborhoo l W of th ~t sort is certainly
garantied by the existe nce of a nei ghborhood N(A) of A in X
which i s strictly b-convex, by virtue of the previous theorem.
 -129-

A. Andr eotti

Chapter VIII. Duality theorems on complex manifolds.

8.1 Preliminaries.
v
a) Cech homology ¥d t h value in a cosheaf.
As in the case of poincare duality on topological manifolds,
it is better unlerstood as a duality between cohomology and
homology, so in the case of complex manifolds duality should be
a pairing between cohomology and homology.
v
For this reason we develo n here Cech homolo gy theory as a pre-
paration for duality theorems (c f. r 7 J which we follow in this
exposition).
Precosheaves (cf. [18J). A precosheaf on a topological sp ace
X is a covariant functor from the category of open sets U c X
to abelian groups, i.e. f?r every open set U c X an abelian
group [) (U) is given and if V c U are open a homomorphism
iV d) (V) ~ ~ (U)
u
is given, such that if WcVeU are open subsets of X we
have, V W W
i U0 i V= i U •

A precosheaf is called a coshe af if for every open set Jl. C X

and every open covering ?( = ~ ui 1 i' I of .Q., the following
sequence is exact.

where 00 is defined by

and

Example. Set &(U) = continuous functions on U with compact

support in U. Define the "extension maps" as the n atural in-
elusions S(V) c ~(U). We obtain in this way a precosheat and
also a cosheaf.
 -130-

A . Andreotti

v
Cach homology. Given a precosheaf . S = t .,)(U), iV
uJ and
an open covering 7...{ = l Ui~ i~I of X, one defines the groups

cS (U:t (j""'4)
.... q
and the homomorphisms

by
aq-l1Jt. i 1 =L(_l)h
""J. O• • • q)

for g ={gi 0"·. i q)I ~ C (U, $).'

q .

We h ave aq-l 0 q = 0 for all q 2. 1 and we put d -1 = o.

We thus get a complex with differential operator of degree -1
and an augmentation

eX : C
o(
21, S) ""7 $ (X) .

We define Hq(U, $) the q-th homology group of this complex,

Ker o
H (0(,,$) = 9-1
q Im Q
q
If V =~ Vj1, jEJ is a refinement of U, to each choice of

the refinement function 1": J '7' I th ere is aascc Lat ed a

homomorphism

r * : Cq('Zf; £ ) -'I Cq ('2{, ~ ) for every q 2:. O.

which is compatible with the differential operator, Thus '1"*

induces a homomorphism
V u. :
'3' Hq( V, {)' ) '7' Hq <l{, S)

which, as one verifies, is independent from the choice of the

refinement function. One can then d e f i n e

H (X,
<::
c) = lim
~
q U
 -1 31-

A . Andreotti

Remark. We may not wish to use all open subsets of X but only
the open sets of a particular class 1?(. This can be don e pro-
vided 7J( is stable by finite intersections and contains arbi-
trarily fine coverings 0 f X.

b) A cosheaf J = f cS (U), iV u? is called flabby if

i UX r ' <5 (U) -?> $ (X)

1s injective for all open sets U.

For example let I be a soft sheaf on X. On every open set

U c, X consider the group r k(U; I) of sections of l:., compact-
ly supported in U and set .s (U) =fk ( u, I). De fine i\ in

the obvious way. We thus obtain a flabby cosheaf on X,

$ = Z~ (U), 1 Vu}.' For every cosheaf of t his sort and for

every locally finite covering U = } U i""f hI of X, we have

Hq(U, S) =0 if q ~ 1.

~. Define the sheaf i by

ui
O • ~ .i q

as the sheaf which has for

fiber at a point x the direct sum of the fibers of the sheaf

!U at the same point.
i •••1
O q

We have an exact sequence of sheaves

ci ci _ ci d.
q q l 1 o
(*) ••• ~ ! q ~ ~q-l -7 ••• ~ ~O r. ~ 0
where
 -132-

A. Andreotti

The exactness tollows trom the tact that at each point x the
homology of that complex is the homology of a simplex with co-
e tticient in 1" x"
Now the sheaves ~ are soft. Taking sections with compact
q
supports in the sequence (*) we get an exact sequence as the
functor r k is exact on so ft sheaves. We thus get exactness of
the sequence
••• .., C (1A , S ) .., Cq_l('U,$) -7 ••• -i> C ('l{ , c3 ) ~ t3(X)~ 0
q O
and this proves our contention.
One can prove that any fi abby cosheaf is of the sort described
above.
We have tor cosheaves the corresponding s t at ement to the
Leray theorem:
Let -u. = fUi ~ i'I be a locally finite covering of X ~
~ cS be a cosheaf on X with the following property
for every open covering z/' = 1. Vj"5 j ~J we h ave

= H (U
q i
(\ U . n
J
V- , ~) = =0 for every
q> 0
then the natural homomorphism
Hq(1(,J)~ Hq (x, 3 )
is an isomorphism.
A covering ~ of this s ort will be call ed a Leray covering
for the cosheaf $. As a consequence of the Leray theorem, we
mention th e folloWing fact:
~ S>, S , ~II be cosheaves on X and let
Z/ =Z Ui~ i6 I be a Leray covering for J' de an d ,S
~ II.

SU'ppose th at for each open set U = U.1. . , .J.. homoraornhisms

0 q
are given such th at the sequence
h U' ku hU kU
0 "7 S '(U) -;> J ( U) -7 '0' II(U ) ..,. 0
 -133 -

A. Andreotti

is exact, and compatible with the extension maps. Then one has
an exact homology seguence
h. k. , d
~ H (X,
l
SI) -7 H (X,
l
S) _ H (X,
l
S II) ~
h. k.
.s
HO(X, I) _;> Ho(X'.') -7 Ho(X' .,) II) -? 0
-7'
Note that exact sequences do not commute in gen eral with inverse
limits, thus the Leray theorem is essential to replace here
holomogy on the covering U by that of X.
8.2. eech homology on complex manifolds.
The following lemma is a consequence of the Hahn - Banach
theorem (e r, [47J).
~ (8.2.1). ~
A ~ B .!,. C
be a seguence of locally convex topological vector spaces over ~

and continuous linear maps u , v !i!!!. v 0 u = o , !!ll

t tv
A' __u B' 4- C'
be the corresponding sequence of dugl spaces and transposed maps.
Then t u tv
o = 0 and we have a natural algebraic homomorphism
t
u
d: :;;K.;;;.e::,r""":,,,_ ....:;. Hom cont (~ £).
1m tv Im u '
If v is a topological homomorphism then 0 is an isomorphism.

{3 = b ' + t v ( C' ) with b ' o u = o.

For every k £ ~:r ~ k = b + u(A) with v(b) = 0, we define

~ ( k)= bl(b). This does not de pend on the choice of the repre-
sent ative b ' and b and thus it defines a linear map
Ker v
-
1m 7u
t. For every c > 0 the set Sl- b /0 BI I b l (b) I <£. 1
is open in B and u(A) saturated. The r e f or e o( ~) is
continuous.
Conversely given A ~ Hom cont ( ~ 1m u'
1&) i t lifts to a
continuous line ar map J,. I Ker v -3' t. By Hahn - Banach we can
/\
extend A' to a cont i nuous line ar map .A' B .:, t. Since
A' I 1m u = 0 t hen 1· I 1m u = 0
/'
th er e f or e J..' e Ker
t
u,
 - 13 4 -

A. An d r e otti

t -v
and thus it d e f i n e s an element of Ker u If A I is another
Im tv
extension 0 fA' to B "
then). I - ;., ;" 0 on Ke r v. Then
" - '"A'
AI defines a linear map of v (B) int o !C. If v is
topolo Gical, t hen the map of v(B) into a: is. continuous and
by Hahn - Banach can be extended to a map y : C .~ lei linear
and continuous. This shows th at i.' - ~ , = tY(}J). Therefore
" ..v
.AI and.A I define the same element rCA.) in Ker t u
Im tv
One then verifies that 0 and T are e ach oth ers inverse.

Let X be a complex manifold. We consider only those open

set U c.c X which are open sets of holomorphy. This collection
/1( of sets if stable by finite intersections and contains
arbitrarily fine coverings.
Let & be the structure she af of X and l et ~ be any
coherent sheaf. For every U" /1(1 we can find a surjection
f}P :; ".f -7 0(1) from which we derive (by theorem B of H.
Cartan and J.P. Serre) a surjection [(UI S P ) : . r(U,j=')~.
The space r (U, f) p) has the structure of a .pace of Frechet-
Schwartz. One verifies that there i s a unique str ucture of a
space of Freehet-Schwartz on [(U, 1=') which makes the mapd.
continuous. Moreover this structure on r(u, ~) is independent
of the presentation e p ~ 'J'~ 0 we have chosen. (See also
[28]). For any U ~ 7?( we set

:(.(U) = Hom cont (£(U, j(), lC)

rU
As tr ansposed s of the restriction maps
v :[ (V,ll .., f(V,j='),
for VC U, V, U" 7J(which are continuous) we get extension
maps

(1) This fact can be proved but it is not obvious (c r , r 22]).

I t is obvious for sU f fi c i e nt ly small U's, ' indeed it is part
of the definition of c oherence.
 A Andreotti

and therefore z":r.(U), i\5 is a precosheaf.

Proposition (8.2.2). For any coherent sheaf 1, 5. ~

o-r' 0-;-'

cosheaf. Moreover for any .fl l&. 7?( and any countable locally
finite covering U = f ui } iEI of 0., !llh. U c. JJ(, we have

Hq(U~~) =° .!2.!: q) 0.
v
~. The augmented Cech conplex
0-> [(fI.,rr) -+ Co(1.(,C;) ~ Cl ('2(,"J') 7 ...
is a complex of Frechet spaces (as V is countable) and con-
tinuous maps. By theorem B this complex is acyclic, i.e.
the sequence is exact. By the lemma (8.2.1) the dual sequence
is exact. But that is the sequence of the augmented homology
complex.
°~ "{(0-) Co(V, T.) ~ Cl ("2(, T.) ~ ...
+-

In particular any countable locally finite covering "Z( c;»( is

a Leray covering for the dual coshe aves ~. of coherent sheaves
T.
8.3. Duality between cohomology and homology.
This duality results by comparison of the two sequ~nces
S S
(I) cq-1Ctl, 1') q::; Cq(t{, T) --$ cq+l('Z(, '.F)

(II)

where:F is a coherent sheaf and '2( c. 77{ is a countable

finite covering of X.
In (I) the spaces are spaces of frechet-Schwartz and the
maps are continuous, in (II) the spaces are strong duals of
spaces of Frechet-Schwartz and the maps as transposeds of the
previous ones are continuous. Moreover each sequence is the
dual of the other as the spaces of Frechet-Schwartz and their
strong duals are reflexive spaces.
 -T3"6 -

A. Andreotti

By application of the duality lemma (8.2.1) we obtain

Theorem (8.3.1)
( a) II cS q is a topologic al homomorphism then

H (X,
q
t.) = Hom cont (H
q(X,1"),
lC).

Moreover Hq+l(X , "f) !!!2. Hq (X , 'I)

•
are separated.
-

(b) If d q_1 is a topological homomorphism then

Hq(X, ~) = Hom cont (Hq(X, '1"'.), lC) •

Moreover H (X, j:'.) and Hq(X,"J") are separated.

q_ 1

Note th at the assumption in (a) is equivalent to the separation

of Hq+l(X,;t) and that the assumption in (b) is equivalent
to the sepu-ation 0 f H
q_ 1
(X , cr.) . These conditions, are

certainly ,s a t i s f i ed if Hq+l(X,'T) or, respectively,

H _ (X,
q l
'1.) are finite-dimensional.

v
8.4. eech homology and the functor EXT.

a) Let T , Jj- be sheaves of e-- -modules on X. Then the

sheaf Jlom f" (:f ,f) is defined as the she af associated to
the preshe af

U -r Hom
e)u
and y are coherent so is the sheaf PI om6' (1, f).
is a family of sup norta, we set

HOM ~ (U,:r, = <!' (U , CJB ~ r 110m <:1" , )•

A shea! Jj of &'-mo<lules on X is called injective if
HOM(X;.,t!) is an exact functor on sheaves of 8' -modules Le.
for any exact sequence of she aves of ~ -modules
o -r :y;: I ~:r ~ 'J " -r 0
0--;> HOM (X;:t''', ·::) --rHQ!.1 (X:".t, ;J')-:> HOM (X; 'rl,-J)-;> 0

is exact.
 -137 ~

A. Andreotti

We have the following facts:

i)
an injective sheaf is flabby
ii) if {j is injective, for any sheaf of &- -modules T
#,om ~ (j:',j') is flabby
iii) for any sheaf;- of C! -modules one can find an injective
resolution (.) 0"7;' -;> ft
o -?> ;;:1 ~ ;;;2 ~ ••• 1. e. the
sequence is exact and every ~ i' i ~ 0, is an
injective sheaf.
These facts follow directly from the definitions and the possi-
bility to imbed every module in an jnjective module.
Applying the functor ;Yom t7 ('~r, to the sequence (.)
we get a complex of sheaves and homomorphisms

o-?>;Yome- (7,f.o~ ~;(om E/T';l)~;l/om& (T,fl2)?

The q-th cohomology group of this complex is denoted by

t xt
q
tY ('7';').
This is a s he a f of &- -modules and one verifies it is independent
of the choice of the resolution (.).
Moreover if 7 and Y-
are coherent then E xt
q
t1
(j'f.) is
a coherent sheaf. This can be seen as follows:
Let
••• 7~ 2 ~d' 1 7'/0 - 'Y-;. 0 (**)

be a resolution of ':f be locally free she aves(l). ApplYing

(1)
On any open set of holomorphy U e X we h ave such a resolu-Co-

tion, as any coherent sheaf on X is the quotient sheaf on U

of a free · sheaf by th e theorem of eoen (cf. [ 22J). However since
here we need only this resolution locally one can invoque the
theorem of syzygies of Hilbert to derive the eXistence in a
sUffici ently s mall neighborhood of a given point x ~ X of a
finite free resolution:
o -e- 6' Pd -7 61 Pd-l..",..•• ",t ) PO -,> ,1 -;> 0 where d s; diIDxX (c f. [.;1> •
 -138-

A Andr-eotti

the functor ;fom ( .,~) we g e t anoth ~r complex

t9
o ~ ;rom & ( ;to';; ) ?' ,;t(om(9 (,J;
"
1'9)-. /r om&
.?f'P
(~2'}) -;> • • •

whose q:-th cohomology group is again (. xt ~ ('J' 'X ), as it

follows by a standard spectral sequence argument. By this
construction G xt~(T ,~) is a coherent sheaf: '
q
In particular if ::J-
is locally free we get t xt (7, t;) = 0
Y --'I l' 19 7-
if q>O as we can take 000=j'Xl=N2= ••• =O.

Note th at in any case because

the sequence

o '-71fom t9 ('T'ff. ) .., ;t( o~ (-r .j'0) -; ;fom& (7 ,f1)

is exact as the functor ;t(om6l C; t , . ) is left exact.
Applying to the resolution (*) the functor HOM ~ (X; 'J', .)
we obtain a complex whose q-th cohomology group is denoted by

EXT~ (X; 'J' ,J).

For q = 0 we have analogously to the previous case

EXT ~(X; 1,X) = HeM 1>(X;'J',;).

The spectral sequence of the double complex K ={KP,q, d}, where
P,
K q = ['4> (X; 110m c/ o(;p'j q»
and d is induced by the ma 's of the resolution (*) and (**),
leads to a spectral sequence connecting the global with the
local extension functors
-- 'J
EXTn (X; j- t. ) ( n = p+q)

where

b) The connection of homology ani the functor EXT is

established by the following

~ (8.4.1). ~ X be a complex .anifold of pure dimension

n
n. Let fL be the sheaf of germs of holomorphic n-forms on_X.
 -1 39-

A. An d reotti

For any open set of holomorphy U C~ X and for any coherent

~ 'T sa X one has
1*(U) =. EX~ (U;T, ~'1. n )
the suffix k denoting the family of compact supports.

~.'
(<1) Let "J' be a locally free sheaf on X and let U be
an open set of holomorphy in X. The sheaf T
can be consider-
ed as the sheaf of germs of holomorphic sections of a holomorphic
vector bundle E, .~ =
(E). Let E* denote the dual
bundle of E; if E is defined by the transition functions
fg i j 5 then E* is defined by the transition functions sI t gij
-llI •

Let /i
r,B(E) denote the sheaf of germs of C oJ' forms of
type (r,s) with values in E and let ;XU,V (E*) denote the
sheaf of germs of forms with distribution coefficients, of type
(u, v), with value in E*.
Since U is an open set of holomorphy, the sequence
o~f(u, SeE»~ ~ (U, AO'O(E»!. (U,JiO,l(E» ~ •••

••• ~ [(U,Ao,n(E» ~ °
is an exact aequence of s paces of Frechet-Schwartz and contin-
uous maps. By the duality lemma (8.2.1) the dual sequence is
also exact. But this is the sequence
° ~ (U,19 (E»' ':i k(u,1C n,n(E*»! k(U,7(n,n-l(E*»!
••• ~rk(U,1(n,o(E*» e: 0.
Now for any vector bundle E*, denoting by ~n(E*) the sheaf
of germs of holomorphic n-forms with values in E*, we have
in the exact sequence of she aves
°4 Jl.n(E*} ~ it n ' O( E* ) ~ ::tn,l(E*)-r
A so ft resolution 0 f .a. n( E*) • Therefore, since E* can be any
holomorphic vector bundle and hence ~n(E*) any locally free
sheaf, we get:
 -140-

A . An d r e ott i

for any locally free s heaf J = j2n(E*) = 1(om &('1, Jl. n )

o if r ~ n
H~(U,f) =
{ Hom cont (f(U,CI(E», t) i f r =n.

($) Let 'Y and :; be locally free then we h ave

(U;'1,~) =0
»: )
ElC~ if P ~ n
k

ElCrfl (Uj
k

In fact the spectral sequence converging to liZlCT:(U;T,; ) has

the term EP'2 q = H~(U; CHt q (~,;(» (p+q=s). Thus EP '2 Q = 0

if q ~ 0 as jF is locally free. Mo r e o v e r

Ext~ (7 ,f..) = ;fom{; ( } ,;,) is also locally free, hence by

(c{) we get EP '2
0
= 0 i f P ~ n and

= l{ (Uj .1fom,9 Cj\~ ».

n o
E
z
degener ated so th at we h ave
But the spectral sequence is
ElC~(U; T::i ) = EPzo •
This proves our contention.

( Y) Suppose now th ~t "'J' is coherent and j locally free.

The we h av e
ElC~ ( Uj T ,f. ) = 0 if p < n

and an exact sequence

o -? EXT~(U;T'I) -> H~(U; /10m() C.z o. f) . .,. ~(Uj Jf'om CJ (£ '1'; ».

In fact consider the double complex KP , q = r k(U, :t,fc>mcd" (..&p ,fq)}

has for cohomology grouns the groups EXT*k(U; '1 'J ).
For an y p the se quence of she aves

-0 7.Jf omt9 «6 p' /I )~ ;:{om& ()S p' ;/o ) ~;;f om£9 c.); p'; 1) -T •••
is exact as ~ is locally fr ee and provides an inject ive
~ p ~ .,
resolution of :/(omc: ( ,,(;P'i-). Taking cohomolo gy with respect
to t he differe ntial co ming from th e r esolution (- ) we get
 -141-

A. Andreotti

Efiq=H~(u,;fom§~p';»=o if ql-n

as ffomt9(rPp'f) is locally free.

It follows that EX~ (U;T i) = 0 if P < n and

EXTn~1 (U;~'Ji) is the l-th cohomology group of the complex:

(0) H~(U; i/om ~ (rP 0' 1» ~ H~(U; .;rom & (06' 1 ,f) -7

~ ~(u;..fom61(;: l'1) -;> •••

In particular we get the exact sequence

(1) 0 ~ EXT~(U; j"f) 7 H~(u;;fomt' (~o'f) ~ H~(U;#om& rLl '1»

(8) Consider the exact sequence of spaces of Frechet-
Schwartz obtained by applying the functor r to ( •• )

••• .."r(U,06l)--- f(U,et'o)~ f(U,:(>-7 o .

Exactness follows from the assumption th at U is an open set

of holomorphy. By the duality lemma and (a. ) we get an exact
sequence

(2) 0 ~ T.(U)-'!- H~(Ut.tom6'UO,jLn»~H~(u,ifom6'(.,tl'£\.n)

Comparing (0), (1) and (2) we obtain

1"'.(u) = EXT~(U;i',S1.n) and EXT~(U;j,nn) = 0 if q I- n ,

Given T and y. coherent she avee , define a precosheaf by

U '7' EXT~(U; T.;...) for U €;J1(. We denote this precosheaf ~
f xt~( ·T.; ). We do not need to verify i t is a cosheaf,

Lemma (8. 4.2). There exists a spectral sequence

!!ill E-~·q = Hp(U ,E xt~( 1, ~».

where 2{ c ?1{ is a locally finite covering of X.
 - 142 -

A. An dr eotti

~. Consider the doubl e complex

K-P,q = Cp(?(,J(omk( ·~';!q». I f we t ake coho mology with
respect to the differential coming from (*) we get
I E-i' Cl = C ( 2{, { xt~(
p
7, )
~) and then taking homology with
v .
respect to the Cech differenti dl we get
IE-P
2Cl = H~<Z{, Ext~( T,; )) .
Now we remark that !fomk('f, fq) (U) = [k(u;;fom ('T, fq».
Thus we have in '!/fomk( '.f 19
q) a fl abby cosheaf as 7/ om (f If. ~
is soft, ~ being injective. Th€refore takin g first homology
.r q v
with respect to the Cech di f f er ent i al we get
"E-i' q = 0 if P ~ 0

Taking now the cohomology with respect to the differential

coming from (.) we get th at the spectral sequence degenerates
having as total cohomology the groups

El{T~(X;1\~ ).
We apply this lemma to $. =S1.n • Then [xt~( 1', S\.. n) =0 if

q ~ n and txt:('J'I ..n.n) ='1. as i t follo ws from lemma

(8.4.1). Therefore we get the following
Theerem (8.4.3). ~ X be a complex manifo~d of pure
dimension n and let ~ be any coherent sheaf on X. !h!a.
we have
Hp(X, '1.) ~ El{~-p(X;"'f, .n,n).
The combination of this theorem (8.4.3) with theorem (~,3.l)
is what is usually called the "duality theorem"; in this form
it is due to Malgrange and Serre (cf. [39], [4V], [44]).
 - 14 3 -

A. Andreotti

Remarks ..
1.' If j! is locally free 1.e. the sheaf of germs of
holomorphic sections of a holomorphic vector bundle
E, or =
&(E), then
Ext ll (r',S1
n)
=0 if II ~ 0 and E xt~ ('1,-St n) =
=if o.m & (oj, fl, n) =Jl.n (E* ) ,
E* being the dual bundle of E. Therefore
ElC~-p (X;&(E),J1,n) = a:-p(X, Jl,n(E*)
so that theorem (8.4.3) gives
Hp(x,B (E)*) ~ ~-p(X, .n,n(E*».
In particular if E is a line bundle, E* = E- l and we get in
this case
~(x,9 (E).) :: a:-p(X, An(E-l».,

2. If '1" =~ (E) is locally free and i f X is a compact

manifold then the theorem (8.3.1) can be applied without
restrictions and we get
HP(X,6'(E» = Hom (~-p(X" j1.n(E·», I)
as the cohomology groups being finite-dimensional any linear
map into ~ is automatically continuous.

8.5. Divisors and Riemann-Roch theorem.

a) Let X be a connected complex maaifold of complex
dimension n. The sheaf &.
of germs of never vanishing
holomorphic functions as a sheaf of multiplicative groups, can
be considered as a subsheaf of the following sheaves:
the sheaf ~* of germs of non identically zero meromorphic
functions,
the sheaf () o· of germs of non identically zero homolorphic
functions.'
 -144-

A . Andreotti

sheaf &:
While the sheaf ~* is a sheaf of ~ultiplicative groups, the
is a sheaf of multiplicative .onoids.
We thus get two exact sequences of sheaves

(1) °~ tJ * ~?Jf.* -> fJ ~ °

(2) 0..,. f)*~ lJ*~f) -",0
o •
where the sheaves)9 and 1). are defined by the exactness
of the sequences. The shea! j) is called the sheaf of germs
of (meromorphic) divisors. The sheaf ~. is called the ~
of germs of holomorphic or positive divisors.
The elements of HO(X,,v) are called meromorphic divisor!! on
X and the elements of HO(X,)}.) are called holomorphic or
positive divisors on X.
An element D E HO(X,0) (resp.. D E HOex,D.» is given on a
sUfficiently fine open covering "Z/ = [Ui~i~I of X by a
collection ffil of meromorphic (resp. holomorphic) functions,
not identically zero, on each Ui, and such that

~j : Ui {\ U j T Ie * s, j E I

is holomorphic and never zero.

The functions fgijt are transition functions of a holomorphic
1
line bundle {DJ and represent the element <feD) EO H (X,o ),
,(\*

where 8 is the connecting homomorphism of the exact cohomology

sequences associated to (1) or (2).
In every meromorphic divisor D we can distinguish the
O-part DO' and the polar part D"", these are holomorphic
divisors and we usually write

Correspondingly we have
lD} =lD~} .fD",}-l.
The study of general (i.e. meromorphic) divisors can thus be
reduced to the study of holomorphic divisors.
 -145-

A. Andr eott i

Consider the sp ace HO(X,CJ(D» where C)(D) is the sheaf

of germs of holomorphic sections of the bundle [D]. To every
s HO(X,~(D» there corresponds a holomorphic divisor, this
in its turn determines s up to multiplication by a global never
vanishing holomorphic function.
If X i s compact (or pseunoconc ave) then the divisor of a
section s ~ HO(X,C/(D» determines the section up to multipli-
cation by a non-zero constant. The set of positive divisors
corresponding to elements of HO(X, ~) ( D » ) is call ed the linear
system of D and is denoted by \DI. Its elements correspond
one-to-one to the point of the projective sp ace
(HO(X, f) ( D» - {of) / a:* •
We attribute to I DI the dimensi on of this sp ace Thus

dim IDI = dima: HO(X,[5(D» - 1 •

The problems we want to mnsider is th e followin g: given on
X a divisor D compute dim IDI. Note th at if we are able to
establish that dim IDI 2 °
then we h ave proved th at the holo-
norphf,c line bundle £ D'5 admits a holomorphic section di fferent
from the O-section.

b) Let us suppose now th at X is comprc t and d im~ = 1.

Then every divisor D is given by a finit , sum D = 1 niPi
wi th n G 1l and Pi being the divisors as s ,' ~ i at e d t o poi nt s
i
of X. A divisor is positive i f f al l n are /~ . The integer
i
ln is c alled the degree of the divisor D.
i
Assume first th at D = L niPi is a pos i t i ve divi s or . In this
case the bun dle l D] has cert ainly a holomorphic s ect i on s t. 0,
the section corres ponding to th e divisor itself. Gi ven s we
h ave a sheaf ho momorphism tS ..., [J (D) gi ven by mul t i pl i cat i on
by s. The homomorphism is cert ainl y inj ective and the quotient
she af is concentr ated in the points Pi and gi ven at e ach po int
Pi by th e vector sp ace v(n Pi ) of dimens i on n i r epr esent-
i,
i ng t he Taylor expans ion at Pi of holomorphic f un ~tions,
trunc ated at t he orier n -1.
i
 - 14 6 -

A . An d r e otti

In other words we h ave an exact sequence of she aves:

° ~ !J s -+ & (D) ~ 11. V (n i ' Pi) -7 °.

This gives the exact cohomology sequence

° -¥' HO(X,6' ) -? HO(X,Cl (D»-7> a:

~n.
~

~ Hl(X,D ) -t' Hl(X ,6) (D» ~ °

as
1
H (X, u V(ni ,Pi» "= ° the sheaf having O-dimensional
support. Therefore we get, in particular:

dim~ HO(X,lJ(D» - dima: Hl(X, 6'(D» = deg(D) + dima: HO(x,D) -

- dima: Hl(X,~ ).
Now:
HO(X,e ) = II: as X is comp act thus dima: HO(X,6J) =1.
_1
I1(X,19) is finite-dimensional and, by duality, ~ H (X, _<1 1 ), °
the space of holomorphic differentials on X. Its dimension
is e al.Ledtthe genus g(X) of X.

r(X,C)(D» if finite-dimensional and, by duality

v HO(X, Jt l(_D» = HO(X, e (K-D» where l K~ is the line

~undle associated to the sheaf Jt 1. The bundle ~ K1.. is
called the canonical bundle, and dima: HO(X, [) (K-D» = i(D)
is called the speciality index of the divisor D.

We thus have the following formula (Riemann - Roch t heorem)

(1) dim IDj = deg (D) - g(x) + i(D)

This formula can be extended to any divisor D = DO - D (f' even

if not positiv~ maintaining for i(D) the meaning of
dima: HO(X,e (K-D».

Indeed let s denote the section, corresponding to D ~ of

HO(X,6>(D",». We have an exact sequence of sheaves

0-> e(D) -') 19(D o ) -<> Jl V(Pi,ni) ~ °

where DO' =Z niPi.
 -147-

A. Andreotti

Therefore

dima: HO(X, e(D

o
» - dima: H (X,l'J (DO) = dima: HO(X, e (D)) -

- dima: Hl(X,{)(D)) + deg(D,..,,)

Thus using for Do the formula already e st ablished we get

deg (Do) - g(X) = dim I D \ -i(D) + deg (Dc")'

From this the assertion follows as we have deg(D) = deg (Do) -

- deg (D ,..).

c) we add a few remarks to the Riemann-Roch theorem in

dimension one.

The genus g(X) tl X eguals t of the first Betti number

b of X :
1(X)
(1) 2g(X) b (X).
1
=
First we show that b ~ 2g(X). From the exact sequence
1(X)
07 a:_.>ed~&l-"O
Since HO(X, a:) ~ a:, HO(X, &) ~ e, we get an injective map

°-r HO(X,JI.. 1) --,) H1(X, a:) •

This map associates to every holomorphic I-form its cohomology

class as a closed form.
Now HO(X, J1,1) and HO(X, ji 1) can .b e considered as subspaces
of H
1(X,
er. .
Their int~rsection is reduced to [01. In fact
if a e HO(, Jl 1 ) , t3 Eo HO(X,..ii
1) and if ft =t3 +dg where
g is a C ~ fun ction on X, by reason of bidegree we get
« = as- Thus '3 a g = ° i ..e. g is (pluri-)harmonic. By the
maxi.urn principle , must ve constant hence a. = 0.
1
Therefore dima: H ( X, a:) ~ 2 dima: HO(X, Jtl) i.e. bl(X) ~ 2g(X).

We show now that b ~ 2g(X). From the exact sequence

1(X)
0.., a: ~ G It e ~ ?1'~ °
l(X,t9
we get an injection 0.., Hl ( X, a:) -7 H ) It Hl(X, &).
 -148-

A. And reot ti

In fact HO(X, an ~ e, HO(X, (}) ~ e, HO(xdf) ~ a: (by the

maximum principle). Therefore
dima: Hl(X, '~ ) ~ 2 dima: Hl(X,b!) i.e. bl(X) ~ 2g(X).

If we apply the Riemann-Roch theorem to D = ° we get

(2) di m )KI = g(X) - 1
If' we apply the same theorem to D = K we get
0) deg (K) = 2g(X) - 2.
To do this we have to know th at t
K 1 comes from a divisor.
Now if g( X) ~ 1 this follows from ( 2). If g(X) then =°
for any positive divisor D we have dim I D\ = deg (D) + i(D).
But then necessarily i (D) = 0.
There eXi sts on X a r ational function with a sing],e pole of
first order. This function extablishes an isomorphism of X
onto the Riemann sph ere and on this manifold on e verifies immed-
iately that K = -2p, p being a point of X. In p articular
for a divisor D !1!h. deg(D) > 2g-2 we must h ave
i(D) = O.
As an exercise one can show now th at every compact manifold
X of complex dimension one admits a porjective imbedding i.e.
is projective algebraic. Indeed if .0' ..• ' St is a basis of
HO(X, e (D» th e map X -? Ft(l:) d e f i ned by
x ~ (sO(x), ••• , St(x» is holomorphic everywhere. If
( D) > 2g t hen one verifi es by means of the Rieman n-Roch t heorem
th at t he map is one-to-one and biholomorphic.
d) Let X be a co nnec t ed c omp act man i fo ld of dima: X = 2.
Let D be a holomorphic divisor on X which will be s upp os ed
o f "multi pl ic ity" on e a t e ac h point and non- s i ngul ar .
We h ave now an exact s equenc e

0"";' O ~ 8 ( D ) ....,. 6) ( D) / D -7 °
where s is a section of f DJ co rres pon di ng to t he di vi s or D.
 - 149 -

A. Andreotti

We get an exact cohomology sequence:

0--7 HO(X,l) ) ~ HO(X, & (D» ~ HO(D,e (D) I D) -j'

~ ~(x, e) j.... !?-(x, e (D» ~ Hl(D, () (D) I D) -;>

2(X,
~ ~(x, CJ) ~ H t9 (D» -7 0.

We have

dim(CI)HOCD, 6l(D)/ D) - dim£!?-(D, D (D)/ D) ::: deg ~D~ID

- genus of (D) + 1
dim~H2(x,~ ) ::: dim£ HO(X,Jl 2) ::: Pg(X) ::: geometric genus of X
dim~Hl(X,e) ::: hO,l

dim~H2(x,r:9(D» ::: dima: HO(X, S(K-D» where tKS denotes the

bundle corresDonding to the sheaf of holomorphic 2-forms
S!.2 ::: () (K). This dimension is denoted by i(D) and called the
speciality index of D.
By the same argument used for dimension one we get now
dim I DI 4 deg 1 Dr)D - genus (D) + (p (X) - ho,l) - i(D) + 1.
g .

This is Castelnuovo's theorem. The l i f f er ence of the left and

right side is dimll: Hl(X,8(D» which is called the
"superabundance".
If X is Kahler then q::: hO,l ::: hl,O ::: dima: HO(X,s&l) ::: number
of linearly independent holomorp hic l-forms, and
p (X) - q ::: p (X) is called the arithmetic genus.
g a
The inequality can be extablished without the r estrictive
as s umpt i ons we h ave made on D. In particular for a multiple
lD of D we get the inequ ality
dim 11DI .2 1 2 deg iDS}n - (l(~-l) deg ~ D}ln + 1 genus (D) -

- 1+1) + ( p (X) _hO,l) - i( l D) + 1.

g

If D is positi ve and 1 large enough and pos i t i ve , then

iClDL ::: 0.
 - J:? O-

A. An d r eo tti

2•
Therefore if deg~DsjD > 0, dim IID) grows like 1
Therefore

II X contains a d i vis or 0 with des OSlo> ~ t. °

trans. degree 1<.
(X) =
2. 'One could s how th at X i s in this
case a projective algebraic v ariety as it is al ways the case
f or a complex surface X with 7f ( X) 0 f transcendence
degree 2. (c r , [21]).
 - 151-

A . Andreotti

Chapter IX. The H. Lewy problem.

9.1. Preliminaries.
To simplify the exposition we restrict ourselves to the
space a: n although the results will hold on any complex manifold
with only formal changes of notation.
Let U be an open ,... ct in a: n and let p U ~ m be a . C <>D
function. We define
U+ = ~ Z E- U I p(z) 2. O}
u- = fz ~ uI p (z) s::. OS

S = (z " uI p(z) =O}

and we will assume d p # °
on S, so that S is a smooth
hypersurface. On U we consider the Dolbeault complex
C* (U) = f CO,O(U) ~ CO,l(U) ~ CO,2(U) -;> ••• S
where CO,s(U) denotes the space of C ~ forms on U of type
(O,s) and where a is the exterior differentiation with re-
spect to antiholomorphic coordinates.
Analogously we define the spaces Co,s(Vr, resp. Co,s(U-), as
the spaces of those forms of type (O,S) on fl+, resp fl-,
h aving CO" coefficients with all partial deriviatives continu-
ous up to the boundary S. In this way we obtain two similar
complexes, c*(u+) and C*(U~).
Define
c' O,s(U) = ~ ~ "" CO,s(U) I 4> = pd. + ~ p/l (3 ,d .. Co,s(U),

(3 E.. CO,s-l(U) s.
We have atj0s(U)c;J°,s+l(u), therefore
;j*(U) = .lL
S
U O,s(U) is a sUbcomplex of C*(U) and indeed a
di ffernetial ideal.
in a similar way one defines the sUbcomplexes <1*(U:) of C*(U:).
 -152-

A. Andreotti

Finally one defines the quotient complex

O*(S) = tQO,O(S)
by the exact sequence
o -:p j"* ( U) ~ c* ( U) -;> Q*(5) --? O.

The quotient complex is denoted by Q* (S) as each one of its

spaces is concentrated on S. The operator 55 is by defini-
tion induced by the operator a- on C* (U) and (J * (U).
One could define the quotient complex also for the inclusions
/1*
CJ
+ * + '
(U-) c C (U-) but one obtains in this W;Jy the same complex
Q* (5) as only the values on S of the coefficients of the forms
considered are of importance.
We thus can consider four types of cohomology groups

H*(U) the cohomology of C* (U)

H*(U!) the cohomology of c*(u!)
H*(S) the cohomology of Q* (5).
Note tbat while the cohomology H*(U) is the cohomology of U
with values in the sheaf e
the same is not true for H*(U!).

Remark.
We have .{jO,O(U) =~t>o,o : U~ a: I ~ °'°15 = °5 thus QO,o(S)
represents the space of C <P functions on S. For u <c Q ' (S) °°
the condition ~S u = °
is a necessary condition for u to be
the tr~ce on S of a function ~ € Co,o(U+) (or 'it. €. CO,O(U-»
which is holomorphic in fl+ (or fl-). Indeed if U is a
C d' extension of u
to U- (or u") we have
~ 11 =° on U+ (or U-) and therefore :;; U = fer.. for some
a c CO,l(U), i.e. as
u = 0.
In general calling the im ~ge of a form of CO,s(U) on
QO,s(S) the tr ace of that form on S, we can st ate th at for
u E Q0 , s the condi tion u = as °
is a nec ess ary condition for
 - 15 3-

A Andreotti

u to be the trace on S of a form U e CO,s(u+) with <l u" = -

Indeed let u be any form of CO,s(U+) h aving u as trace on S.
N °
Then
.-
U = u + p<1-+ a p 1\ (.3
N
for some Q and ~ in c*(u+) •
Thus

i.e.
Same argulllent :Eor U-.

Example.
2
Take U = t , zl = Xl + iX 2 , z2 = x3 + iX as co ordinates
4
on a: 2 • Let

p 5 X
2
4 - (x + x
2)
= 2i (Z2-- Z2) - Izl l
2

Then S is the product of the paraboloid x = Xl2 + x 22 in

4
m 3 where Xl' x 2 ' x 4 are coordinates, by the x 3-axis.

At each point . dZI and ap = - 2i dZ2 - ~ d~

1 form a basis
for the (O,l)-forms. Thus we have
QO'O(S) = C""(S) = C 6' functions on S
QO,l(S) "" C"'(S) /\ dZl
QO,2(S) = o,
Thus i'
Q*(S) - - tC"(S) ~ C (S) /I dZI ~ ° r.
To compute as' by its definition, we h ave to do the following
- given u G C~(S) extend (in any way) to a C ~ function u
on a: 2 • Due to the shape of S we may assume U in:iependent
of the variable X ;
4
- compute au;
 -154-

A . An dreotti

- compute _'1J
au;
'V
c\i
=
_1.1\
(1U ~ dZ + dZ
l 2
zl z2
'\) N

- ~
dZi. - ~ (2i~p + 2iZi.dz
l)
aZl o z2

- "restrict" the form thus obtained to 5 i.e. compute modulo

tJ 0,1 (1: 2 » to get

N
as u is .independent 0 f x and ~, can be taken
4
can be taken as coordin ates on 5.
In conclusion the complex Q* (5) is isomorphic to the complex
on lR 3 where ~ = xl + 1X 2 ' and x
3
are coordinates

where
L =~ - iZ l ~
oZl 0%3

The operator L is the operator of H. Lewy, of r 371.

9.2. Mayer-Vietrois sequence.
A C <::P function on U is called fiat on 5 if i t
a)
vanishes on 5 with all of its partial de r i v at i v es . 5et
'1 0 , s ( U) = N € CO,s(U) I all coefficients of eP are flat
on 5 S.
 -105-

A , Andreotti

We have a 1°'s(u) c, 'j O,s+l(U) therefore '1'. (u) = JL'1

a
0 , 8 ( U)

is another aUbcomplex of •
C (U) and in fact a subcomplex of
u
~*
(U). The quotient complex
•
C (S) =C tI
(U)/'J (U)
•
is
concentrated on S and is obtained by restricting to S the
coe fficients 0 f the C rjJ forms on U.
We have

Co,a(S) =t !: act (x) dZa: ••• dZ a

a:l( •••~tt.8 l···as i. a

Eo Co,o(S) for all (a'l •••O:s) •

0-7 C...(U) -p c" (u+) It C· (U-) ~ c" (S) .....:p ° is an exact sequence
from which we get a cohomology sequence
° ~ HO,o(U) 7' HO,o(U+) It HO,o(U-) -» HO,o(C· (S» ~
(1)

To connect the cohomology groups HO,s(C·(S» with the groups

HO,s(S) = Ho,s(Q·(S» we may use the following exact sequence

° -?> v (U)/3=' * (U)

J-/* -"? C* (U)/'1- * (U) -7 Q* (S) -;> °
i.e.
*
O~tJ (U)/ 'Y- * (U).....:pC* (S). '--i' Q* (S) -i> 0.

~ (9.2.1). For any choi ce of U and S the seguence

j'o,o(U) a ~O,l(U)
, °(U)
.o is exact.
o~
-:r ~
1 o , 1 ( U)
~.
(Q) Let u €: -d O,O(U)
Assume th .\t the coefficients of ju are fl at on S. Then
.}pqllS =0.
 -156-

A..Andreotti

This means that q'l =pet 2 for some a E: COiO(U) thus

2 '
u =f 2
Q'2. But then

;1' Q'21 s = 0 .
thus o: 2 = pc( 3 for some C( 3 E; Co,O(U), thus u = p3dy
Continuing in this w~ we see that u must be flat on 5 i.e.
we have exactness at d 0,°1'1°,°.
«(3 ) To treat the general case we will make use 0 f the
following fact
Given on S a seguence
O'
f f
l, 2
f , 2! C CI' func tions,
there exists a C d' func tion F 2!! U such that

ok F
d pk
I5 = fk
for k = 0, 1, 2, ..•
This can be derived as a particular case from the Whitney
extension theorem. Also direct proofs are available.(l)
Let f e {jo,s(U), s 21. Then f = I'd. + ap /I (:, for some
and (3. Using the above remark, we can find a form
r> 1 E. CO,s-l(U) such that

131 1
5
= 131 5
akp
(*) _l[ =0 for k = 1, 2, 3, ...
S ap k
Thus we can write f = (f - a p" 13 1) + "5p /I ( \ = pGJ. + a p " 13 1
as the coefficient of f - Jp )\ 13 1 vanish on S while (31

satisfies the conditions (*). Let now u ~ ~ O t s( U) and

assume that the coefficients of ~u are fl at on S. Write
u = pQ'l + 3p /I 6
1
with 13 1 satisfying (*).

(1) In particular it follows from this lemm a th at the space

CO,O(S) can be identified with the space [ (S) ~ p } of formal
power series in p with C ~ coefficients on S,
 -157-

A .An dne ntti

Then
u - a(p(\) = peal - 81\).
Set y 1 =« i - 3~1· By the assumption we get ~e .1\ '(11s = °thus
Il=P~2+~f'/I(32 with 13 2 satisfying (*). Then

u -
-
o(I'f31 + tf' 132 ) 2
=p 2 -
«((2 - h~2)·

Set r 2 =q2 - t~ e 2· By the assumption a t'/\ y 2( =

S
° thus
--(2 = PQ'3 + ap /I 11 3 with (33 satisfying (*). Then

Proceeding in this way we construct a formal power series in

om 1.
c; = 1 ilt'3 m
wi th fl m satisfying (*) and with the property that

_ m+l pk m+l
u - a (~ i t 13 k)
1
=P Y m+l

Using the remark made of the beginning, we can find

f eo OO,8-1(U) such that

() k' k~
T t Is = ~ =0 for k = 0, 1, 2,
~f'
e
Set v = f then (l f f: -~ ,
ap J.f ° s-l (U)
. and we have that u - ov
has flat coefficients on S. This proves exactness at
~ 0,8(U)/):'0,8(U).

This lemma tells us that the cohomology of vJ-l * (U)/'T * (U) is

zero in any dimension and therefore the cohomology sequence of
(2) gives a set of isomorphism
HO,s(O*(S» ~ HO's(~*(S) = HO,s(S).
 -158-

A. Andreotti

Introducing this result in the sequence (1) we obtain an exact

sequence

which is called the Mayer-Vietoris sequence for U and S.

b) The previous considerations can be repe ated replacing the
space of C <P functions on U, u-+ with the space of those C
functions with compact support on U, U! respectively. This
leads to an exact sequence

-,. H°itl(U)....". H°itl(U+) • H°itl(U-) --i' H°itl(S) -

(Mayer-Vietoris sequence with compact supports).

9.3. Bochner theorem.

Let X be any (n-2)-complete connected manifold of complex
dimension n ~ 2, for instance a Stein manifold. Let S be any
connected closed cot' hypersurface i n X such that
X-S =f- U f+ is the union of two connected open sets f- and
f+ t of which say f- is relatively compact.
Theorem (9.3.1). Under t he above assumptions, any C ~ function
f 2a s s atisfYin~ the compatibility condition asf = is °
the trace on S £f...1: C '" function on X- which is holomornhic
in f-.
~. The Mayer-Vietoris sequence with comp act supports gives
 - 1 59 -

A.... Andrec.tH

Now HOkO(X) = HOkO(X+) = °

as X and X+ are connected and
and not compaet.
Moreover HO'k O(X- ) =
HOkO(X-) as X- ~s · compact. By the duality
theorem we get
HOkl(X) ~ Hom cont (Hn-l(X, n),~) =°
by the assumption th at n ~ 2 and X is (n-2)-complete.
Since Hn-l(X, Jl. n ) and Hn(X, J\,n) are finite-dimens io·nal, i,

the maps 8 are topological homomorphis~.

Hence Hn _l (X,.n~) ~ Hom cont (Hn-l(X, J1. n ) , a:)

nn
~ Hom (rt..n-), (X,JL), ~).

And

Here E* is the dual bundle of t he canonical bundle E.

Hence the restriction map
is surjective.
FOr X =~n this t heorem is due to Bochner, Fichera, Martinelli
(cf. [16,1 [24,1 (41]).
Note th :d no assumption is ma le on the shape or convexity of S.
9.4. Riemann-Hilbert and Cauchy problem.
These problems are ge ner ali zations of a pr obl em concerning
holomorphic functions in several v ari ables f irst considered and
solved by Hans Lewy (cf. [37~ [381).
As s um e th at U is an open s et o f holomorphy in ~n. Then
·t he Mayer-Vi etoris seq uence (with closed su pnorts) sp lits in
the s hort ex act se quences
o ·~ HO,O(U) ~ HO,O(U+) i HO, O(U- ) ---'f HO, O(S ) ...". °
 • 1 .6 0~

A. An d r e otti

This shows th :3.t

(a~ Every cohomology class on S can be wr i t t e n as a jump
of a cohomology class on U+ and a cohomology class on
U- i.e. the so-called Riemann-Hilbert problem is always
solvable for an open set .0 f hoLomo r phy ,
Moreover if s >U the solution is unique while for
s = °
it is determined up to addition to the functions
on U+ and U- of a global holomorphic function on U.
(b) Let us agree to say th at the Cauchy-problem is solvable
from the side of U+ if HO,s(U+) ~ HO,s(S) is sur-
jective. Then we reali ze that if U is an open set of
holomorphy the solvability of Cauchy-problem for s > °
is equivalent to the vanishing theorem
HO,s(U-) = °
and in this case the solution is uni que.
If s = ° the Cauchy-problem is solvable if and only if
HO,O(U) ~ HO,O(U-)

is surjective i.e., loosely speaking, is

contained in the envelope of holomorphy The
solution will be unique if for instance,
connected.

Example.
Let us consider the situation of the example given at the end of
section 9.1.' Here U = 1 x
4
+" Xl2 + x 22 ~ is an elementary con-
vex set while U is the closure of the complement in ~2. The
boundary S of U+ is strongly Levi-convex. Let a E S and
let J1 be any neighborhood of a which is also a domain of
holomorphy. Writing for fI.,.ft+,.n. - and SJ1, = nOS the Mayer-
Vietoris sequence, we get the exact sequence
 -161-

A_Andreotti

We have
i) HO,l(Jl) = 0, HO,2(.Q) = 0, as n is a domain of
holomorphy.
11) Ho,l(Jl+) = 0.
This fact is a consequence of the regularity theorem of
Kohn and Nirenberg (cf. next section). It would be desir-
able to obtain a direct proof in this special case.
iii) Since U- has a pseudoconcave boundary one realizes that
for every point a €o S we can find a fundamental sequence
of neighborhoods Jt. of a, which are domains of holo.
morphy and such that
HO ,0 (-J\) ~ HO, °(Jl-)
is surjective (and an isomorphism).
Making use of this information the sequence of Mayer-Vietoris
gives us the following isomorphism

HO,l(JC) ~ Ho,l(s.rJ

The first tells us that given any ell> functions on s..Jl,

satisfying the compatibility condition asu ~ exists a°
~t~h~e~r~e-=~~~=­
C 1» function u !.!! Jt+ which is holomorphic in Jl+ and such
!.!l!l
11 Is = u.
(use is made of the assumption iii)).
 -162-

A . A n d reotti

In connection with the second isomorphism (which is valid even

if the assumption iii) is not satisfied) we remark th at
iv) 01
H' (If) and in fact dim H01
I. 0
-
' (A ) = '-, provided.J~
1\

lC
is sufficiently small. In fact by a loc al change of holo-
morphic coordinates at a we may assume th at s..n. is
strongly elementary convex (1.4, exerci se 2). Then the
statement is a st~aightforward consequence of lemma (7. 2.2).
An immediate consequence of this fact and the isomorphism
established above is the following theorem first proved by
H. Lewy [37].

Given on lR 3 the equation

Lu !! ~ iz ..L1!
1 aX3
= f
oZl
for any point a ~ lR 3 we can find a fundamental sequence of
neighborhcjods wiJ such th at for infinitely many f Eo C"'( w >1)
equation does not admi t any solution u E. C"'(w y ) .

9.5. Cauchy-problem as a vanishing theorem for coho mology.

Let us now consider the Levi form restricted to th e analytic
tangent plane of the hypersurface S.
Using the methods of proof of the vanishing theorem (7.1.1) and
the regularization theorem of Kohn and Nirenberg, (s ee (3~,
[321 and r
5J) we obtain the following result
Theorem (9.5.1). For any point Zo ~ S at which the Levi form
~ p positive and q negative eigenvalues on the analytic
tangent plane to S at zo' we can find a fundamental sequence
of neighborhoods l U,)'v~N tl zo' all domains of holomorphY,
such that
8 > n-q-l
•
or
[
o < s < p
 -16 3-

A. Andreotti

Analogously one can find a similar fundamental seguence of neigh-

borhoods ~tU " veW
such that
s ) n-p-l

2!
°
[

< s < q
Moreover, if P > 0, we can select the seguence ~ U } in such a
v
Way that the restriction
HO,O(U-J) ~ HO'O(U~)

is surjective, Le. U'l! is in the "envelope of holomorphY" of

+
U)I' AnalogouslY, if q). 0, we can select the seguence i. U'" 5 !B.
such a w!Y that the restriction

HO,O(U ~ HO'O(U~)
lI)
is surjective, Le. Uy is in the "envelope of holomorphY" of U~,

According to the remarks made in the previous sections this

theorem tells us when locally the Cauchy problem for cohomology
classes is solvable. Two special cases will serve as an
illustration
Case 1. Assume th at the Levi-form is non de ge ner a t e with

° <, p < q = n-l-p

Marking only the cohomology groups which are ( possible) different
from zero, the situation is illustr ated by the follo win g picture

u+ HO,o(U+) HO,P(U+)
ls !S
s HO,o(S) HO'P(S) HO,q(S)

6151 ~~51->
-
pI
U HO,o(U-) Ho,q(U-)

Moreover one can sho w by applic ation of Lemm a (7.2.2) as we

did in th e generaliz ation 0 f Levi-problem to coho mology ~l2<Sses
 -16~-

A. Andreotti

(point l:,) in the proof) that the cohomology groups we have

marked are all infinite-dimensional.
Case 2. Assume that the Levi-form is non-degenerate, n is odd
and
°<: P = q = Tn-l
then with the same conventions the situation is illustrated by
the follewing picture
u+ HO,O(U+) HO,P(U+)
P J,

S HO'O(S) HO,P(S) ~ HO,P(U+)

°
Again the groups marked in the picture are all infinite-dimensional
In this case in dimension p the Cauchy problem is not solvable
from either side, only the Riemann-Hilbert problem is solvable in
that dimension.
Remark. In both cases in dimensions p and q we are in the
presence of systems of first order partial differential equations
Lu = f on S which for infinitely many C rP functions f sat-
isfying the integrability conditions have no solution u of
class C-:
9.6. Non-validity of Poincare lemma for the complex f R~ (S), JS •
On S we can consider for any s the sheaf QO,s defined by
the or es he a f
JL _ QO's(Jl.)
We thus get a complex of she aves s,* = 11 rl's with differential
-
as . os ...
2J
operator QO'O ~
QO,l -4 QO,2 ~
IN" 1.'V'
""'"
It is natural to ask of t his se quence of she aves is exact. Indeed
in th at case<it woul d provide a soft resolution of th e she af
.9(S) of germs o f C"" functions f sat i s f yi ng th e :restricted
Cauchy-Riemann equ ations is f = 0.
 -165-

A Andnaott,'

The answer to this question is in general negative as does show

the following

Theorem (9.6.1). ~ S be a locally closed hypersurface in

a: n and let z o e S be a point at which t he Levi
.
form on the
analytic tangent space of S ~ Zo . ~ p positive and q
negative eigenvalues and is non-degenerate (so that p+q = n-l).
Then in the complex

the Poincar~ lemma is not valid in dimensions p

holds in any other dimension.

~. From theorem (9.5.1) and the Mayer-Vietoris sequence we

deduce that there exists a fundamental sequence of neighborhoods
W
v of Zo in S, v = 1, 2, 3, ••• , such that Ho,s(4J y ) = °
if s f. p,q. Thus the Poincare lemma for is valid inas
dimensions different from p and q.
We have to show that this is no longer true in dimensions p
and q. Let w be any neighborhood of Zo in S. First we
remark that
QO,p(w) ~ C"'(w)a:(p)

where (p) = (n;l).Indeed we can select a baSis for the space

of n
(0,1) forms in a neighborhood Jl, of Zo in a: , Jl. n S = w,
of the form YJ 1 t • 0 • ,'I n-l' ap'
Then
a
°loooop

£. C"'(w) 5•
As such QO,p(u) has a natural structure of a Frechet space.
 - 166-

A . An dr-eotti

Given w we can select a fundamental sequence of neighborhoods

()J C co,v = 1,2,3, ••• , of Zo in S such that aP(",-,J) # as °
we have remarked in section 9.5.
Set for LV = w or Wv'

Since d is a differential operator, it is continuous for the

S
topology of Q*(w). Therefore ZP(w) is a closed subspace of
QO'p(w) and therefore a Frechet space. On its turn
BP(w) =QO,p-1(~) / Zp-l(w) inherits a quotient structure of a
Frechet space.
Consider for every v the following set of continuous maps
r
Zp(to.!) ~ZP(Wy)

ti
'\!

BP( w)

where i v is the natural injection and r v the restriction map.

Set
E"\J = 1: (0, /3) E Zp(u;» <. BP(w) I r)cd = i )M 1.
Then E vasa closed subspace 0 t ZpC.. : ) '" BP (w 11 ) has the struc-
ture of a Frechet space and we c an co mpl ete t !1e :'1r evi ous maps with
the following commut at Lve diagr am
r..;
ZPCu.) _ ..- T ZPC w y.)
-
I'j 1'i

E~ - 7 BPC"",v)
(j
1
With j an d d cont inuous. We h ave j C E ~ ) = r: ivC BP ( ~v» '
Thus t his s pac e is a con tinuous i motGe b y .-J. l ine ar map of a
Frec het sp ace. By the BCUlach open mappin g theorem we mnst h ave
on e o f th ese t wo ~ r o pe r t ie s
 -167-

A_ A n d r eotti

1) either r-} ivCBP(UJ~» = Zp(iA)

i1) or elser~l iv(BP(w v) is of first category in zP( w).

Now as we have remarked in section 9.5. we can construct for
each v an e'Lemerrt ~\I E> Zp(w) such that r~ (4)v) ¢. i /BPC,l 1y»'
This rules out possibility i) for ev 7ry v. Hence

is of first category.

Therefore there exists an element g € ZP(w) such that for

every v

i.e. such that the equation

3S }JV = g

cannot be solved in w)) although the integrability condition

~S g = 0 is satisfied in the whole of fA) •
This shows that p nincar~ lemma cannot hold in dimension p. For
dimension q the argument is the same.

Elcample. In the particular case 0 f the Lewy operator Lu = f

in lR 3 it follo ws that given a €:iR 3 there exists a neigh-
borhood lV of a and f <c: CO'(...,) wuch that for any ...J \J c. w
the equation Lu =
f cannot be solved.

9.7. Global results.

We illustrate the type of global results one can obtain by
the following example. For the proofs we refer to [5J.
Let X be a compact connected manifold of complex dimension
n , Let p: X ~ lR be a C 0' function and assume that

s = fx .. X I p (x) = Os
is a smooth hypersurface (dp # 0 on S) dividing X into the
two regions
X- = ~x E X I p(x) ~ o I, X+ =[x E X I fl (x) :: 0 S•
 -168-

A. Andreotti

We assume that the Levi form of f restricted to the analytic

tangent plane to S is nowhere degener~te and has p positive
and q negative eigenvalues (p+q = n-l).
Then one can prove that

dim HO,s(X-) < ~ if s ~ q

t
dim HO,q(X-) = <:p
t

and similarly

dim HO,P(X+) = Of! •

t
Let us agree to say that the global Riemann-Hilbert nroblem is
almost alWayS solvable in dimension r i f f

has finite-dimensional kernel and cokernel.

Similarly we agree to say th at the Cauchy-problem is almost
alWayS solvable in dimension r from the side X+ iff

has finite-dimensional kernel and cokernel.

A straightforward appl i c at i on of the May er - Vie t or i s sequence in

connection with the description of t he groups Ho , r (X! ) given
above le ads to the follo wing conclusion:
 -169-

A . Andreotti

!! p # q the Cauchy problem is of interest(l) only in dimen-

~ p from the side X+ and in dimension q from the side
X- and it 18 almost alWayS solvable. If p = q = 2 n-l (n !!!.!!ll
be odd) then the Cauchy problem is of interest in dimension p
bu' not almost alWayS solvable on either side, while the
Riemann-Hilbert problem is of interest in dimension p ~
almost alWayS solvable.

It is worth noticing th at these considerations can be extended

to more general complexes of partial differential operators~
The operator ~s was first introduced in 'a different form
in [35]. In this ' exposition we have followed [41 and [71.

(1)
In the sense that it leads to maps between infinite-
dimensional spaces.
 -170 -

A . Andreotti

BIBLIOGRAPHY.
(1) Andreotti, A.: Theoremes de dependance algebrique sur les
espaces complexes pseudo-concaves. Bull.
Soc. Math. France 91 (1963), 1-38.
[2] Andreotti, A. . and Grauert, H.: Theoremes de finitude pour
la cohomologie des espaces complexes. Bull.
Soc. Math. France 90 (1962), 193-259.
[3J Andreotti, A. and Grauert, H.: Algebraische Korper von
automorphen Funktionen. Nachr. Ak. Wiss.
Gottingen (1961), 39-48.
[41 Andreotti, A. and Hill, C.D.: E.E.Levi convexity and the
Hans Lewy problem. Part I: Reduction to
vanishing theorems. Ann. Sc. Norm. Sup.
Pisa 26 (1972), 325-363.
[5] . Alilireott:t, -.A. and Hill, C.D.: E.E. Levi convexity and
the Hans Lewy problem. Part II: Vanishing
theorems. Ann. Sc. Norm. Sup. Pisa 26
(1972), 747-806.
[61 Andreotti, A. and Huckleberry, A.: Pseudoconc ave Lie
groups. Compositio Mathematica 25 (1972) ,
109-115.
[7 J Andreotti, A. and Kas, A. : Duality theorems for complex
spaces. Ann. Sc. Norm. Sup. Pisa, to appear.
[8] Andreotti, A. and Norgu et, F.: Probleme de Levi et
convexite holomorphe pour les classes de
cohomologie. Ann. Sc. · Norm. Sup. Pis a,
s. 3, 20 (1966), 197-241.
[91 Andreotti, A. and Norguet, F.: La convexite holomorphe
dans l'espace analytique des cycles d'une
variete algebrique. Ann . Sc. Norm. Sup.
Pisa, s. 3, 21 (1967), 31-82.
 -17 1-

A. Andreotti

[10] Andreotti, A. and Yum-Tong Siu: Projective embedding ~

pseudoconcave spaces. Ann. Sc. Nor m. Sup.
Pisa, s. 3, 24 (1970), 231-278.
[1~ Andreotti, A. and stoll, W.: Analytic and algebraic
dependence of meromorphic functions.
~ecture notes in Mathematics 234. Springel I

Berlin, 1971.
[12J Andreotti, A. and Tomassini, G.: Some remarks on pseudo-
concave manifolds. Essays in topology and
related topics. M~moires dedies aG. de
Rham, 1970, 8.5-104.
[13J Andreotti, A. and Vesentini, E.: Sopra un teorema di
Kodaira. Ann. Sc. Norm. Sup. Pisa, s. 3,
~.5 (1961), 283-309.
[14] Andreotti, A. and Vesentini, E.: Car1eman estim ates for
the Laplace-Beltrami equation on complex
manifolds. Publications Mathematiques
I.H.E.S. 2.5 (196.5), 81- 130 .
[1.5] Barth, W.: Der Abstand von einer algebraischen Mann-
igfaltigkeit im komp1ex-projectiven Raum.
Math. Ann. 187 (1970), 1.50-162.
[161 Bochner, S.: Analytic and meromorphic continuation by
means of Green's formula. Ann. of Math.
44 (1943), 6.52-673.
[17] Borel, A.: Pseudo-concaVite et groupes arithm~tiques.
Essays in topology and related topics.
Memoires dedies a G. de Rham, 1970, 70-84.
118] Bredon, G.: Sheaf theory. Me. Graw-Hill series i n
higher mathematics, 1967.
 -172-

A. Andreotti

[19J Cartan, Ii.: Quotient d t un espace analytique par un

groupe d1automorphismes. Algebraic geometry
and topology. A symposium in honor of S.
Lefschetz. Princeton U.P. 1957, 90-102;
[20J Cartan, H.: Quotients of complex analytic spaces, Con-
tributions to function theory. Tata Insti-
tute of Fundamental Research, BombaY, 1960,
1-15.
(21) ChOW, W.L. and Kodaira, K.: On analytic surfaces with
two independent meromorphic runc t.Lona,' Proc.
Nat. Acad. Sci. U.S.A. 38 (1952), 319-325.
[22J Coen, S·: Sul rango dei fasci coerenti, Boll. U.M.I.
22 (1967), 373-382.
[23] Docquier, F. ' and Grauert, H.: Levisches Problem und
Rungescher Satz fur Teilgebiete Steinscher
Mannigfaltigkeiten. Math. Ann. 140 (1960),
94-123.
[24] Fichera, G.: Caratterizzazione della traccia, su1la
frontiera di un campo, di una funzione ana-
1itica di pi~ variabi1i comp1esse. At t i
Accad. Naz. Lincei. Rend. C1. Sci. Fia. Mat.
Nat. 22 (1957), 706-715.
[25J Godement, R.: Topo1ogi e algebrique et theorie des fais-
ceaux. Hermann, Paris, 1958.
[26) Grauert, H.: On Levi's problem and t he imbedding of real-
analytic manifolds. Ann. of Math 68 (1858)
460-472.
[27J Grauert, H. l Uber Modifikationen und exzeptionel1e
analytische Mengen. Math. Ann. 14 6 (1962) ,
331- 368 .
 - 1 7 3-

A. An d r e otti

[28] Gunning, R. : and Rossi, H.: Analytic Functions of

Several Complex Variables. Prentice-Hall
Inc. Englewood Cl if f s , N. J . , 1965.
f29l Herve, M.: Several complex variables. Local theory.
Tata Institute of Fund amental Research,
Bombay, Oxford U,.P., 1963.
[3OJ Hirzebruch, F.: Topological metho ds in al gebraic geometry.
Springer, Berlin, 1966.
[31J Hormander, L.: An introduction to complex analysis in
several variabl es. Van Nostrand, Prince-
ton, N.J., 1966.
(32) Hormander, L.: L2 estimates and existence theorems for
the a operator . Ac t a Mat h . 113 (1965),
89-152.
[33J Kodaira, K. : On Kahler varieties of restricted type
(an intrinsic characterization of alge-
braic varieties). Ann. of Math. 60 (1954),
28-48.
[34J Kohn, J.J. : and Nirenberg, L.: Non-coercive boundary
value problems. Comm. Pure Appl. Math.
18 (1965), 443- 492 .
[35J Kohn, J.J. and Rossi, H.: On the extension of ho1o-
morphic functions from the boundary of a
complex manifold. Ann. of Math. 81
(1965), 451-472.
[361 Levi, E.E.: Studii sui punti singolari essenziali
delle funzioni analitiche di due 0 piu
variabi1i. Opere, Cremonese, Roma,
1958, 187-213.
 -174-

A. Andreotti

1"37) Lewy, H.: An example 0 f a smooth linear partial

differential equation without solution.
Ann. of Math. 66(1957), 155-158.
[381 Lewy, H.: On hulls of holomorphy. Co mm. Pure Appl.
l1ath. 13 (1960) ', 587-591.
[39) Malgrange , B.: Systemes differe ntiels a coefficients
constants. Sam. Bourbaki 1962, 265.
[40] Ku1hmann, N.: Uber holomorphe Abbildungen komplexer
Raume. Archiv Math. 15 (1964), 81-90.
[411 Martinelli, E.: Sopra un teorema di F. Severi nella teoria
delle funzioni analitiche di piu vari abili
complesse. Rend. Mat. e Appl. 20 (1961),
81-96.
[42J Narasimhan, R.: Introduction to the theory of analytic
spaces. Lecture notes in Mathematics 25.
Springer, Berlin, 19 66.
[431 Narasimhan, R.: Several complex variabl es. U. of Chicago
Press, 1971.
[441 Ramis, J.P. and Ruget, G.: Complexe dualisant et
theoremes de dUalite en geometrie analyti-
que complexe. Publications Mathematiques
I.H.E.S. 38 (1970), 77-91.
t45J Remmert, R.: Projektionen analytischer Mengen. Math.
Ann. 130 (1956), 410-441.
[46] Serre, J.P.: Fonctions automorphes, quelques majorations
dans le cas ou X/G est compact. Sam.
H. cartan, 1953-54. Benjamin 1957.
[47] Serre, J.P.: Un theoreme de dualite. Comm. Math.
Helvetici 29 (1955), 9-26.
 -1 75 _

A. An d r eotti

[481 Siegel, C.L.: Analytic functions of several complex

variabl es. Lectures delivered at t he
Insti tute ' f or advanced study, 19Lf8-49.
Notes by P. Bateman.
[49) Siegel, C.L.: Einfuhrung in die Theorie der Modul fo r men
n l ten Grqdes. Math. Ann . 116 (1939),
617-657.
[50] Siegel, C.L.: Meromorphe Funktionen auf kompakten
analytischen Mannigfal tigkeiten. Nachr ,
Ak. Wiss. Gottingen, 19 55, 71-77.
[51J Weil, A.: Variates Kahleriennes. Hermann, Paris,
1958.
r521 Zariski , 0.: Some results in the arithmetic theory of
algebraic varieties. Am. J. Hath. 61
(1939), 249-294.
[53J Zariski , 0.: Sur la normalite analytique des variates
normales. Ann. Inst. Fourier 2 (1950),
161-164.
 CENTRO INTERNAZIONALE MATEMATICO ESTIVO
(C . 1. M . E. )

pROPAGATION OF SINGULARI TIES

FOR THE CAUCHY-RIEMA NN E QU AT ION S

J.J . KOHN

Cor so t e n ut o a Bressanone dal 3 al 12 gi u gno 19 7 3

 - 17 9 -

J . J . Kohn
Introduction.

These .lectures are intended as an i ntro duction to the

study of several complex variables fro m the po i nt of view
of partial differential equations. More specifically he r e
we take the approach of the calculus of variations known as
the a- Neumann problem. Mos t of the material covered here
is contained in Folland and Kohn, [ 4 1, Hor ma nde r ( 11 ] and
in the more recent work of the author (see [ 161 , 117] and
L201). we consistently use the Laplace operator as in
Kohn ['14l , since we believe that this method is particularly
sUitable for the study of re~ularity and for t he study of
the induced Cauchy-Riemann equations. Our main emphasis is
in finding regular solutions of the inhomo~enious Cauchy-
Riemann equations. We wish to call attention to the ex-
tensive research on this problem by different methods from
the ones mentioned above (see Ramirez [ 29] , Gr a u e r t and
Lieb [8 J, Kerzman [ 13], 0vrelid [ 28], Henkin [ 9], Folland
and Stein [ 5] ). It would take us too far afield to present
these matters here. Another closely related SUbject which
we cannot take up here is the theory of approximations by
holomorphic functions (see R. Nirenberg and O. Wells [ 27] ,
R. Nirenberg [26]. Hormander and Wermer ["12 ] , e tc , },
 - 180-

J . J . Kohn

Lecture 1. The a- problem and Hartog's theorem

The purpose of these lectures is to serve as an intro-

duction to the use of the methods of partial differential
equations in the theory of several complex variables. Let
zl"",zn be the coordinate functions in (n and let
x J = Re (z J)' Yj = Im(z j)'
for a function u we define the derivatives u z and Uz by:
. j j

(1.1) : and Uz = !(ux +iU y)'

j J j

The Cauchy-Riemann equations are then the equations

(1. 2) Uz
j
=0 for j = l, ••• ,n.
A function satisfying the system (2) is called holo-
morphic and the theory of several complex variables consists
of the study of these functions. A classical theorem states
that a function is holomorphic if and only if it can be
represented locally as a power series in the coordinates
zl, ... zn (see, for example, Hormander [111 theorem 2.2.6).
Here we will prove the existence of globally defined holo-
morphic functions, these functions will not be constructed
by piecing together local solutions of (2) but rather by
studying the inhomogeneous Cauchy-Riemann equations. More
..
precisely, given functions o(l""o(n we wish to solve the
eq~ations

(1..) ) j = l, ... ,n.

Furthermore, we will want to investigate the dependence of

 -1 81-

.T. .J. Ka h n

FUrthermore, we wlll want to lnvestlgate the dependence of

the solutlon u on the OC j • For example, ln the dlscusslon
of Hartog's theorem glven below, the crucial step ls to
find a compactly supported u when the ~j are compactly
supported. Slnce the operators ln (1) have constant co-
efficlents we have UZjz k = uzkz j and hence the followlng
compatlbl1ity condltlons are necessary for the exlstence of
a solutlon u of ()):

(1. 4) ()(k-
Zj' for all k, j = 1,2, •• • ;n.
The equatlons ()) and (4) are best expressed ln terms
of dlfferentlal forms settln~

c( =lcl.dz ,
J j

and

we have
(1.)') dU = 0(.

(1.4' )

The problem of flndlng u satlsfyln~ ()') wlth glven

that satlsfles (4') ls called the a- problem.
The followlng classlcal theorem due to Hartog lndlcates
the profound dlfference between the theory of several com-
plex varlables and one complex varlable. The proof ls an
example of how the a- problem can be used to prove exlstence
theorems of holomorphlc functlons.
 - 182-

J . J. K ohn

Theorem (Hartog). If 1);> 0 and

and if n ~ 2, then every ho1omorphic function defined on

U has a unique ho1omorphic extension to the ball

That is, if h is a ho1omorphic function on U then there

exists a ho1omorphic function ""h on B such that
=
'V
h h on U

In case n = 1 the function h(z) = Z -1 shows that,

in the above theorem, the restriction n ~ 2 is necessary
Ehrenpreis in [ 3] showed that the required extension
property can be obtainedfram the existence of a compactly
supported solution u whenever 0/. has compact support.
'r he argument is as follows: let f be a Coe function
which equals 1 when Liz j \ 2 ~ 1- 0/4 and equals 0
if L I zjl 2:::::: 1-&/2 we set

(1. 5) for 1- a <.. Ll Z \

j
2 <:
elsewhere

Clearly ci.. satisfies (3) and has compact support.

If there eXists u with compact support satisfying (3)
then, first of all by (3) u is holomorphic in the domain
'2I Z
j
\2 > 1_0/2,

and hence by analytic continuation u is zero in this domain.

 -18 3-

J . J . Kohn
,-,J
In B we deflne h by
(1.6 ) h = ph - u,
...,
Then h ls the requlred solutlon slnce, by (3) and (5)
lt ls holomorphlc and slnce
"V
h =h ln an non-empty open
subset of u and hence throughout u.
It remalns to be shown that when n ~ 2 then there
exlst u E (:([:n) whenever the o(j E CoOQU[.n).
Followlng Hormander (see [11) 2.3), we deduce thls from
the followlng classlcal solutlon of (3) ln one varlable.
Namely, if J1. C ([ 1 and b.Q, the boundary of Jl ls
smooth and lf u E: ("'" ( (l) then

(1.7) n(z) =~ ~ ~'l:~z d't' + rio)~ ~~-z d'l:' A d<t-.

bn ~
Thls formula ls easlly derlved by uslng Stoke's theorem on
the domaln obtalned by removlng a dlsc of radlus t and
center Z from Jl and then lettlng E. ~ O. It then
follows that if eX fer it) and 1f we deflne u by

(1.8 ) u(z) = -..L

211'1
fS ~
't" - Z
d'l A df
.n.
= _1_
2nl

1
=m
dlfferentlatlng and changlng varlables agaln, we obtaln
 -184 -

J . J . Kahn

(1.9 )' uz(z) = 1

2m. rsc o(:r ('! + z) d'r"dr
'T

= 1
"21iT f5
..n.
~d-r
'l-Z "
dT

Since c<.. =0 on b.1't, applying (1.1 to 0<. we obtain

(1. 10 ).

It is clear that if a solution of (10) has compact

support then the integral of 0( is zero. In the case of
n z, 2 , however, we obtain a solution u of () with
compact support whenever the 0( j have compact support and
satisfy (4). The desired u is then defined by:

(1.11)
~ IS
C
0<.1 ('I: 'Z2'··· ':?:n)
't' - Zl
d 'L" d:e

= oI.1('t'"+

TOree that u satisfies () we first note that by

(10) we have further, using (4) and we
obtain
(1 .12) . =

= 1
~
 -185-

J . J. Kahn

In particular we see that u is holomorphic outside of

K = U supp (o(j). From (11) we see that u is 0 if I z2\
is sufficiently large and hence, by analytic continuation,
u. =0 in the unbounded component of the compliment of K;
therefore u has compact support.
The above argument can be used to prove the following
generalization of Hartog's theorem, due to Bochner (see ':2 ': ).
Theorem. If n > 2 and if Jl C <C ->\.. is a bounded con-
nected open set then every holomorphic function on 11 has
.n.. ; where,
r-- ,--
a unique holomorphic extension to .0.. is the
union of JlL and the bounded components of the compliment
of SL.
If we suppose that b.a, the boundary of Il , is smooth
and that a function h is defined on b JL it is natural
r-...
to ask, when does there exist a smooth function h defined
on it. such that h =h on bll and h is holomorphic
in .0.. The obvious necessary condition on h is that it
satisfies the so-called "tangential Cauchy-Riemann equations~

i.e. those linear combinations of the d which are tan-

JYj
gential to b Q. More precisely, if r is a real-valued
smooth function defined in a neighborhood of b _Q such that
dr,. ':: and if a ••• ,an are functions .s uc h. that,;on , -8L1
1,

(1.13) La j r-
Zj
0,

and i f Uf is any smooth extension of h, then,

( 1.1.' L a j f zj ::;
° on bn.
 -18 6-

J . J . Kahn

An equlvalent way of wrltlng thls ls:

(1. 14.'.).-

Theorem. If .fLc. a:n, n >-. 2, is a domain with a smooth

boundary bJl, if the complement of Jl is connected and
lf ..Q is bounded. Then a smooth function h on b Sl.
can be extended to a smooth functlon h on n. such that
h ls holomorphlc ln J1, if and only if h satisfies (14)
f9:rall al""'~ that satisfy (13).
PoHowing Hormander en] (theorem 2.3.2') we can
prove the above by first constructing a smooth extension H
such that

( 1.1.5') j = It •.. t n •

Slnce the a -problem can be solved with a compactly

supported u for 0( j E Col (([ n). we set d. = a (j'H)
where p =1 in a small neighborhood of bSl and vanishes
outside a sllghtly larger neighborhood. Uslng o(=-:au
with a compactly supported u we obtaln the desired exten-
..v
sion h = pH - u as in the previous theorem. The function
H can be constructed by starting wlth any smooth extension
r of h and noting that (14) implles that:

(1.1E! ) near bit

rewritlng this we have

(1.17)
 -1 87-

J . J . K ahn

which implies that

and hence

(1. 18}
setting

H :: 2
(1.19' f - fo r - f 1 r /2
we obtain

(1. 20)
as required.
It should be mentioned that the tangential Cauchy-
Riemann equations (14) have been studied extensively (see
Boc hne r ( 2) ,Lewy [2 21 • Kohn and Rossi [ n1
Kohn [15] , Andr eot t i and ~ il ~ [-1] • . e.tc. ).
In fact, the famous example of Lewy of an equation
without solutions is one of these. We will return to this
equation later.
Our main concern will be to study (3) on a domain
without any restriction on the support of cc , Suppose, for
example, that the c( j E L 2 cD.. ) then ..we wish to find L
u~L2(rt) satisfying (3), by this we mean that we want to
find a sequence of smooth uy defined on..fl such that
in L2 eQ ) we have u:: 11m u y and o(j :: 11m u yzj •
Suppose that there exists a point P E. b{l and a holo-
morphic function f defined in a nei~hborhood 11 of P
such that f (P) :: 0 and f ~ 0 in U(\JL - {p J. Then
 -1 88-

.T . J. Kohn

we claim that if a solution of ()) eXists in L then

2(fL)
there exists a holomorphic function h defined on Jl, suet:
that h cannot be continued analytically over the point P.
That is, if V is an open set containing P then there ex-
ists no holomorphic function g defined on V such that
g =h on V (\ n. To see this let f e C;( u ) such that
p = 1 in a neighborhood of P. Now we define F by:
= J pf-· in N
Vf"IiL
(1. 21) F
lOin ~- U

and we chose N so large that F~L2(5'L). We set 0<,,-:= 3F

and note that 0( Jf L 2 (rt) (in fact, e{ J =0 in a neigh-
borhood of P) • So if there exists
u E L2 (.~\) such that
au=cX then we define the desired h by:
(1. 22.) h :: F - u

Now h $ L (0-) but for any neighborhood W of P

2
hE L2 LCl - W) and thus h cannot be continued past P.
The problem of finding a holomorphic function h on
which cannot be continued past a point P€:. bfl is called
the Levi problem for fl at P. A holomorphic function f
defined in a ne1;;ghborhood of P E. b rt. which has the proper-
ties hypothesized above is called a local holomorphic se-
parating function at P. We have thus proven the follOWing:

Theorem. If the a - problem has a solution UE:: L

2
CQ )
whenever the 'cf.. J E:: L2 (Q ) and satisfies (4) and i f there
eXists a holomorphic separating function at P E- bSl then
the Levi oroblem has a solution at P.
 -1 89 -

J . J . Kah n

It ls easy to see that a holomorphlc separatlng funct-

10n exlsts at P lf the demaln Sl ls convex ln a neigh-
borhood of P (In that case there ls a llnear holomorphlc
separatlng functlon). Let (Lt' be the domaln glven by the
lnequalltles

(1. 2J)

and.fl.2 ls the ball of radlus ~ and center a,O, •.. ,O).

Let iL = n lV ..0 2 then the orlgln is ttl b..a and _Q
ls convex ln a nelghborhood of the orlgln. Thus we see that
toe equatlon (J) cannot be solved ln
L2( for lf lt n)
were posslble to solve lt we would obtaln also a solutlon of
Levl's problem at the orlgln and this would contradlct
Hartog's theorem. A modlficatlon of the above argument
shows that for the above JI.. the range of the operator a
, u z2 ,. . . , uz
n
) ls not closed ln any of the

natural topologles (l.e. Lp' dlstrlbutlon) and thus the

~ - problem cannot be solved ln any satlsfactory sense.
 -1 90-

J . J . Ko h n

Lecture 2. Pseudo-convexity

nc tr n
We will now study the properties of jL ~ which
insure that there is a local holomorphic separating function
at PE bCl. As was pointed out in the previous lecture it
is enough that.fl be convex in a neighbor!;!ood of P.
However, the notion "of holomorphic separating function is
invariant under holomorphic transformations whereas the "no-
tion of convexity is not. We will introduce the property
of "pseudo-convexity" by, roughly speakin~, isolating that
feature of convexity which is invariant under holomorphic
transformations.
We assume that blL is smooth; that is, that there
eXists a real-valued function r defined in a neighborhood
of bil such that dr '" 0 and r =0 on b...C1. We will
fix the sign of r so that:
(:3.1)0 r > 0 outside of n and r < o in D.. •
The domain n is convex if the Hessian of r (1.e.

the matrix of second oartial derivatives) is non-negative

acting as a quadratic form on the tangent vectors to blL;
if it is positive - definite then fL is strictly convex.
In IR2 this reduces to the familiar fact that the graph
of a function is convex if the second derivative is non -
positive, the general case may also be deduced from this by
taking intersections with planes. In terms of the complex
n
coordlnate in r'L. this condition is expressed by
 -191-

J . J . Kahn

wherever

(2.3)

In other words we are restricting the 2n ~ 2n quad-

.::: ) (l:)
ratic form
rz z (P)
i j

( rz z (P)
i j
n

To those vectors which are real (i.e. b = a) and

j j
tangential to bJl at P. It is clear that this condition
is not invariant under holomorphic transformations. However,
observe that the condition implies that

(2.4) > 0,

whenever

° for P E b.a.

Note that the above is invariant under holomorphic

transformations and, in fact, we will express this last con-
dition in invariant form. Let ~l,O(biL) be the subspace
of the complex tangent vectors at P € b ft which are of type
(1,0); that is, they annihilate the anti-holomorphic func-
tions. In terms of coordinates Tp 1,0(bfL) if and
only if it can be expressed as
'0
( 2 • 6P .~ L = L a j az j , where lr
zi
(P)a i = 0.
 -192-

J . J. Kahn

For each P ~ bit we have the herml tlan form

(2.7) x :r p r 1.0 (b.S1.) X T , 1.0 (bi\.) ~ c

p p '

by
(2.8). ;\
p
(L. L') = «oar) • 'p
L;\L'> •

where <: ,> denotes the contraction between contravarl-

"\ I
ant and covarlant tensors. The form A ls called the
p
Levl form and the condltlon glven by (4) and (5) ls equlval-
ent to
0
A. ,p
(L, L) ~ o for all L ~
1>
1. (bD.,).

Thls condltlon ls clearly lnvarlant and we say that 1l

ls pseudo-convex 1 f (9) ls satlsfied for all }>l E. b.Q and

that lt is strongly pseudo-convex lf the Levl form ls posl-

tlve deflnlte at all P E bfl.
Conversely. l f i l ls strongly pseudo-convex and lf
p~ b{l then there exlsts a holomorphlc coordlnate system
on a nelghborhood U of P such that JL {\ u ls strlctly
convex. To prove thls we flrst note that strong pseudo-
convexlty ls lndependent of the choice of r; provlded, of
course. that r satlsfy (1). Then we note that for any
coordlnate system r
can be chosen so that (r z (p) ) ls z
1 j
a posltlve deflnlte hermltlan form on all n - t .uples .
(a 1 ••••• a n). 1.e. wlthout the restrlctlon (5). To achleve
this we set
(2.10)
't.H
r = e - 1.
 -19 3-

J . J . Kohn

where R is any fixed function satisfying ( 1) and T is

a sufficiently large number. Then we have

(2.11~ _

we , e compos e an arbitrary n-tuples as follows:

(2.12)

where

so that (al •••• ,a n) satisfies (5). Thus we have

(2.13) . = t: L ~ z i zj (P) 9 1a j + '['2

2

+ 0(" 'fl (2. la i , 2)~.

The error term is bounded by:
(2.14) - (large const. ~fi\ 2 + small const. L: lai l 2).

Choosing T sufficiently large this is smaller than

the first two terms on the ri ght of (13) - since, by hy po-
thesis, the first term is larger than ''C const. L. I ail 2.
Therefore. we have
(2.15)- In z i zj (P)b;'b j > const."[ I b
i\
2

as desired.
We are now ready to prove the following classical re-
sult.
Theorem. If i1. is a strongly pseudo-convex domain and
 -194-

J . J . Kohn

P!C: bil. then there exists a neighborhood U of P And a

holomorphic coordinate system whose do~ain contains

such that U f\ft is strictly convex with respect to the
linear structure given by these coordinates.
Proof: It suffices to find a coordinate system such that
the form (2) is positive-definite even without restriction
(3). Let u1, ••• ,u
n
be any holomorphic coordinate system
with origin at P and let T be a function satisfying (1)
with (~u U (0» positive definite. Expanding ~ in a
i j
Taylor series, we have
(2.16) r = 2 qe ([ r u (O)u i + r u (O)UiU j )
i ui J
+ lr U 11 (O)uiU j +
i j
((Ilu i \ ) 3).

Setting

(2.17 ) Z
i
= ui for i = 1, ••• ,n-1, and

zn = 2 2 'ru ( P) u ,.. 2l:r ( P)uiu j

i i uiu j

we have
(2.18) r = Be ( z ) +
n
L r zi zj (0 ) Zi Z
.J
,+

(0 )
Since r is positive definite we also have ,r _ (0)
u U ZiZj
i j

positive definite and thu s b Y (1 8) the ~essian is also

positive definite.
The followin ~ classical theore~ shows that in case the
Levi form is identically zero the d oma i n is al so locally
convex. rhe proof of t hi s t.he or-en is less elementary than
t he one a bove.
 - 1 95 -

J. J. Kohn

Theorem. If r is real, analytic and

If in B nei ~hbor hoo d U of P ~ bil the Levi-
form is identically zero (i.e. on u n blL) then there ex-
ists a coordinate neighborhood V of P with holo morphic
coordinates zl''''' zn on V , such that the set IT/ (\ b fl
consists of the points for which
The two theorems above make it seem plausible that
whenever a domain is pseudo-convex then in a neighborhood
of each boundary point these exist coordinates with respect
to which it is convex. However, this is not true as is
shown in the following example (see Kohn and Nirenberg [14])
Let fL C( 2 so that near the origin the function ~ is
given by:
2 8 1S 6
(2.19) r =: Re (w) + [z ] lWl 2 + /z{ +"7 IZ!2 "l.e (z ).

Since complex dimension 2 the spacer-1,O(bjl) is one-

~
dimensional the Levi form 1 K 1 matrix. In the above ex-
ample it can be shown that the Levi form is larger than
const. (/W1 2 + / z I 6 ) near the origin. Thus, by suitably
extending r we have Jl pseudo-convex and strongly
pseudo-convex everywhere except at (0,0). Now if there
were holomorphic coordinates on a neighborhood OJ of the
origin relative to which u /\ .n.. is convex, then we could
find a linear function h such that He h(O,O) = ° and
the zeros of ae h which are in U remain outs~de of J1
That this is not possible in this example 1s shown by the
following result.
 -196 -

J .J. kahn

Theorem. If h is a holomor?hic function which is defined

on a neighborhood U of the origin and 1f h(O,O) = °then
there exist points (z1,wl)' (z2'w 2) E ;J such that
= h(z2'w 2) = 0, > <
h (Zl,w1) 'T(z l , wl )
°and r (z2'w 2) 0;

where ~~ is defined by (19).

The proof of this theorem depends on the following
;Lemma.
Lemma. For each a E. ([' the function f, defined by
(2.20) fA(Z) = \z\8 + ~5 iz\2 '1 e (z6 ) + l1e(az 8),

changes sign in every ne Lghbo rhood of the origin.

Proof: since for t ~ 0, fa (t z) = t Bfa (z) it suffices to
that the function

changes sign. Then we obtain

So·2rf Ie: (e)I de .

(2.22)
.!.l rr = S
2rr
2; (O)e
.J) 1'"
. d£> ~
7 0

Since ~2"'?; (9)d~ = 2 IT, we see that i f does not chan!1;e

o
( 211
sign then ._
D
I,,:. ( fd)! d = 2 11 which contradicts the above

inequa 11 tv ,
Proof of the theorem: Suppose h is a holomorphic function
defined in a neighborhood of (0,0) with h(O,O) = 0. If
h(O,w) !!! ° then r, (O,w) : \e (w) which changes sign as re-
qu i r-ed , If h(Z ,0) !! °then h(Z,O) = fo<z) which changes
sign by the above lemma. Otherwise some of the zeroes of
h can be Deranretriz ed RS f011ows:
 -197-

J . J .Kohn

(2.23) · ' z = q

w = with I'l 0.
P

That is, for every '! we have h(z,w) = ° when z and

w are defln~d by (2). Evaluating r on these zeroes,
we obtain

(2.24)r =

Now, if P < 8q then the first term on the right

controls the sign of:T . for I'SI small and hence f' changes
sign. If p = 8q, then the first three terms control the
sign and so r changes sign by the lemma) wi th a=a • If
P
p> 8q, then the second and third terms control the sign
and the lemma can be applied with a = 0.
At this point we wish to discuss invariants of pseudo-
convex domains which are not strongly pseudo-convex. These
invariants play an important role in the study of boundary
regUlarity of the ~ - problem (see Kohn [16] ). Let L
be a C OO
vector field defined in a neighborhood of
PE-b.Q. We suppose that L is of degree (1,0) and that
for each i,O(bR), 1.e.
~

(2.25):.:
and

L(r) =
° on

Denote by L the conjuge of L (defined by

 -198-

J . J . Kohn

IL(u) = ' t ,U) , in terms of local coordinates

(2.2q) L = La J ~
aJ
For each non-negative integer k we define a space of
vector 'f i e l ds i (rL..) on 'u as follows

(2.27.1 tQ- ;(L)

= {fL + a; 1-J
oe
Le. the space of all f ~ + giT- with t , lI;E( ('{);

(2.2~) ;f(L) = :t. k -1 (L) + [1° (L ), t: k -1 (L U,

that is the space of all vector fields of the form

where IJ
,
W' €- oJ...
'f"k-1(L)
'
r ~
J:.. 0 (L) and

r V' v' r .
We say that L is of finite type at P if for some
~ we have

( 2 • 30 ) . . ; ( : (L ) T/,O(o"J)) + r O,·1(b1U,
!"
where t'~ (U denotes the vector space obtained by evalu-
ating alFthe vector fields in 1 k (L) at ? 'r he lowest
integer it for which DO) is true is called the order of
L at .' and i f DO) is not ture for any k then we say
that L is of infinite order at 2 . l'he following hold :
(A) If .it is pseudo-convex, then it is strongly
pseudo-convex if and only if for each P c- b 51 an d each

non-zero L defined in a neighborhood of L is of

 -199-

J . J . Kahn

order 1 at P •
(B) If U is open and all vector fields on U are
of infinite order at f' E u n b n, then the Levi form is
identically zero on
The basis for the above properties is the following ex-
pression for the Levi form in terms of vector fields
with values o n r 1,O(b..Q).
(2.)1) <d~r , Lt.L> <ar , [L ,E l >
This formula follows from the fact that
and the classical expression for the exterior derivative:

<~ ~'r, L 1\ r:)= -~,a r, u/\ L >

= -d< d r ,I) + [«ar , L ) + <()r, [L,rJ)
<~ 'r, r > =
and we have
(1, 0) and [ of dell;ree
° since ar
further
is of degree
(0,1) ,

(2.)2) - . <ar,c>= < dr , e) = L (r) = 0.

Lemma: If ve<fr(bSU, Le. VCr) =° on bfl, then

v G r 1, °(bn) + r": 1 ( bn ) if and on ly i f
p P
(2•) ) ) <(~ r ) 0' 'J ) = 0

Proof: gy ()2) and by considerations of degree it is clear

that all r O, l ( bfU satisfy (J). On
p
the other hand

(2.)41' . dim ( r 1 , O(.b{l ) +

~
r ?, l (bfU 2n - 2
t'

dim q. Q.' [t bfU 2n - 1

p
 -200 -

J. J . Kahn

and since (or)p is considered as a linear functional on

et: Tp ( b .0. ) and is non-trivial (if it were trivial it would
be a multiple of dr ) we see that the subspace of ( Tr(b~)

of those V satisfying ())) is of dimension 2n - 2 which

concludes the proof.
The above properties are then e~sy consequences of ()1),
the lemma and the definition of the Levi form (8).
We conclude the present discussion of order with the
following lemma.
Lemma. If every non-zero local vector field on b1L, with
values in Tp1,0(bft), is of finite order then there exists
no non-trivial connected analytic vari : contained in bSl.
Proof: Suppose V is such an analytic variety and that
P e V is a regular point of V. Then there is a non-zero
vector field L of degree (1,0) def1ned. :\1n a neighborhood
U of P such that

LQ,cri,O(bD.) for ~ € i.J () b ~1.. and such

that L E. TQ+O(yj (f ot Q 6 U() V. Then, clearly
Q

; [ :( L ) c cr TQ (V ) for all Q E U() V and

since crTQ(V) r
~
1, °(V ) + T O,l(V
.~
we have

~TQ. ('I{ ) C T 1,O(b .(L) + TQ. O,l(b.fU and

Q

therefore, L is of inflni te order for all Q e Un v.

To conclude this lecture we wish to mention some classi-
cal properties of pseudo-convexity. We refer to Hormander
tlQJ section 2.6 for .s ~v s t e ma t i c treatment of these.
 -201-

J . J. Kohn

Def1nit1on. A real-valued runc t ron u E COO (Sl) is called

plurisubharmon1c if the quadrat1c form
(5) 2. u zi zj Siaj
1s pos1tive def1n1te.
The follow1ng theorem (see Hormander theorems
2.6.7, 2.6.11 and 2.612) shows how the notion of plurisubhar-
mon1clty leads to generalizing pseudo-convexity to domains
whose boundary is not smooth.
Theorem. If Jrl is an open set in ~n with a smooth
boundary bfl then the following are equivalent:

(a) There exists a pluri-subharmonic function u on

n such that for every JR.
C E.

f! f = { ZEn\U(Z)< c}
is compact.

(b) Sl is pseudo-convex, 1.e. (4), (5) are satisfied.

As is standard we extend the definition of pseudo-coo-

vex to domains satisfying condition (Ja). This condition
says essentially that J( can be exhausted by strongly
pseudo-convex domains. This need not be ture for pseudo-
convex dqmains in complex manifolds. From condition (a) and
from the solution of the Levi problem for strongly pseudo-
convex domains, it can be deduced that a domain 1l in (0
is a domain of holomorphy (i.e. there eXists a holomorphic
function on 1l. which cannot be continued to a larger do-
ma.Ln ) if_ and only if .0. is pseudo-convex.
 _202 -

J . J . K ahn

Finally we wish to c a l l att ention to the f o l lo w 1n ~ re-

sult (see Ho r ma n :ie r r ll] t he o r-e m 2 . 6.'1 3) wht c.h shows t hat
t r '1'l'
if a domain in \L w1th a s mooth boundarY is not ps eudo-
convex then it is not a do main of holomor ph y .
Theorem. Suppose that nc ([ :n has a boundary of class
c4 , i.e. the function r is of class c4 • Su ppo s e t hat for
some FE bfl. and some (al"" ,an) with

we have

then there exists a neighborhood iJ U of P s uc h t hat

if u € C 4 ( U) a nd satisfies t he tan ge ntial Ca u c hy - Ri ema nn
equations {( l . ) on V·n b it.. then there exf s t s a function
v E.. C 1 ( Y ) such t hat v = u on V () b .Il, a nd 0 v = 0 on
'Y+ = {
7'
<i E V· Ir ( .:< ) .;. 0 ~ • In particular, any holo-
morphic function on u- = [ ~ uI r ( ~ ) < o} can be
extended to y+ .
 - 2OJ-

J . J . K ahn
Lecture 3. Formulatlon of the a-Neumann problem.

We return now to the study of the operator a from the

polnt of vIew of L2• Denotlng byap,q(fD the forms of type
(p,q) whlch are C00 i n J2,
- that is restrictions of C<IO forms on
a neighborhood of n, we have

with 82 .. 0. On these spaces we define ~ inner products as

follows:
(3.2) (u,v) = SUVdV for u,v~a:>,O({l)
.a..
('P,\f') :II LI 'PiV;idV for 'f,tEO..0, 1 (11)

(9,0-) .. 2..n..~ei jifi jdV for

e, <rE-"0,2"
u.. (4L).

The corresponding norms are denoted by II II. We denote

by Ll,q(Q) the completion ofap,q<!2) under the L2-norm.
Let T and S denote the L2-closures of the operators a in
(3.1) and T*, S* the adjoints of T, S then we have

O,O,('\)~ 0,1(f'\\h 0,2«('\\

L2 ~L 1--- ~ .1""~~ .l&.J.
T* S*

These operators are defined 001 their domains (not on

the whole space at the beginning of the arrow). The domain
of T, denoted by Dom (T), is defined as follows:

0.4) Dom (T) = {UE-L20'°<n)\3{uj~ with Uj~~'O(Q),

u = 11m u j and ~Uj} Cauchy in L20,1(Q)j.

 - 204-

J. J . Kohn

By taklng \jIE<i>' 1 (.rL) wl th oompaot support and wfCClClan

we have

(3.5) (aw,lIJ) = (w, - L'P1z i.

1

Therefore, lf uEDom (T) and 1Uj} , {VjJ are two sequencet

wlth l1mlt u and such that both {au ~ and {;~VJ\ are Cauchy;
j
thus, settlng w = uJ - V
j
ln (3.5), we obtaln

(3.6 )

so that for u"= Dam (T) we oan deflne Tu by

and lt ls lndependent of the sequence.

Slmllarly, we deflne Dom (S) and Sf for all ~~Dom (S).

The domaln of T*, denoted by Dam (T*), ls def1ned by:

(3.8) Dom (T*) ={lf~L20,1(Q)l3 c > 0 w1th

\(Tu,cp)\ ~ cl\ull for all u~ Dom (T)}

to deflne T* on Dom (T*) we see from (3.8) that the map

u ~ (Tu,lp) ls a bounded l1near functlonal on Dom (T) and
o 0
thus has a un1que extens10n to L2 ' (fl), slnoe Dam (T) is
dense. This extenslon has a unique representatlve whloh ls
by .defin1tion T* , thus we have

(u, 'I'*I,f» = ('ru, If) ,

 - 2 05-

J . J . Kahn

If 1~Dom (T*) ls dlfferentlable then we can apply ().5)

wl th I/J =If and w wl th compac t support and we 0 btaln

(J.1O)

Th1s can be thought of as a complex d1vergence corres-

pond1ng to the complex gradlent and complex curl whlch are
g1ven by ~ on func t10ns and on (0,1) -forms respec t1 ve ly,
2
S1nce ~ = 0 we have

(J.ll) ST =0 and T*S* = O.

Th1s means that lf uE Dom (T) then Tu€ Dom (S) and STu 0

and s1m1larly I f ~ ~, s* and T*.

The following ls a class1cal result 1n H1lbert space

theory (see Hormander [111 theorem 1.1.1).

Theorem. If A and B are HUbert spaces and T : A ~B ls a

closed densely def1ned operator, then the follow1ng are

equ1valent:

(a) ~(T), the range of T, i s closed

(b) There 1s a constant C such that

(J.12) liull
A
f c(lruji B , uEDom (T) n[<1«T*)],

when L J denotes closure.

(0 ) <R(T* ) 1s closed

(d) There 1s a constant C such that

 - 206-

J . J . K ahn

Proof. Since (T*) is the ortho gonal complement of

11( T) = { h f. Dom ( 'r )\ Th = o}

we have

is aclosed one-to-one map hence by thecl o ~ed g r a ph theorem

(a) ===> (b) and similarly (since T = T**), (b)~(d). We
wish to prove (b)==}(d) . From (b) we obtain:

\(If' Tu)B\ = I(T*'f' u ) A\ ~ I\T*lfIlA llu tlAf. cIlT*tIlA11ru\lB '

iff Dam (T*) and u f Dam (T) n [61(T*)]

hence

which implies ().12') since

\\ 'fIls = i nf 1(tp , IP ) BI
lI ~llB

The proof is then complete since (b) ' :::;> (a) and (d) ~ (c)
are clear.

Now let C be a third Hilbert space and we assume that

we have closed operators densely defined T : A ~ Band
S : B ~ C such that
 -2 0 7-

J . J . K oh n

(3.1) ST = 0

We define 2{C.B by

(3.14) ~ = 1( (T*) n'l1.. (S )

then we have:

Theorem. A necessary and sufficient condition thatR(T) and

(R(S) be both closed is that

(3.15 )

for all 'f'~ Dom (T*) () Dom (S) 1fi th If> .1<K..

Proof. First we have the weak ortho~onal decomposition

formula

(J. 16 )

which is an immediate consequence of ().1). From ().1)

we also conclude that (iRJTU C ;}f(S) thus if

gE Dom (T*)n [(R(TD then Sg = 0

and ().15) reduces to ().12'). Similarly, since

T*S* = 0

we obtain ().12) as a consequence of ().15). For an arbi-

trary

If' £ ~ wi th tp .l Jt
 - 20 8-

J. J . Kahn

we use the orthogonal decomposition ().16) and obtain

f = 1f + ~2 •
1

To prove ().15) it suffices to show

. 2 .2 . 2 · . 2
1I<p It
1 B
+ nep 2II B
~ cons t , (\\T*'f'
. 1 A
II · +)15 '1f, ~
T2 .
),

which is the sum of ().12) and ().12').

The inequality ().15) is often obtained as a conse-

quence of the following:

Theorem. Suppose that whenever {If 1iS a sequence

k

'f
k
Eo Dom (T*) n Dom (S)

such that Ulp U is bounded. 1f

k B

lim T*r.p = 0 in A and 1f lim :3 (IJ = 0 in C

k Tk

there exists a subsequence which converges in B. Then () .~5)

holds and 1t is finite dimensional.

~: The hypothesis implies that the unit sphere ina( is

compact, hence ~ is finite dimensional. If () .15) did not
hold then we could find a sequence

.@ t=- Dom (T*) (\ Dom (5)

k
 - 2 09 -

J. J . Kohn

with ek 1. X such that

2 2
liek 11 B + \\ s t> k 1\C ).

Setting

we obtain

2 2 1
llT*lf II
k A
+ II S<pk(I C <. K

hence thereexists a subsequence of { ~ 3 which converges to

an element If .1. X but since tp € Je andk \I If Ii = 1 we obtain a
contradict1on.

Corollary. If the norm 1l¥,11 + ' 1lT*!p1l + \Islj)11 defined

B A B

on Dam (T*) n Dom (S) is compact then ().15) holds and

is finite dimensional.

once we know that (3.15) is satisfied then we can re-

move the brackets in (3.16) and thus obtain

(J.1?)

which is called the orthogonal decomposition. We shall make

~his more explicit by introducing the laplacian
 -210-

J . J . Kahn

(3.18) Lrp = TT*.p + S*Stp for tpEDom (T*)n Dom (8)

with T*t? E:Dom (T) and Slf E Dom (8*). Then

(3.19) It = 'Yl(L) ,
since clearlyj(C?1(L) and if LV' =0 then
2
(3.20 ) (L'f .....)
. B
+ l\stp\\ = 0
C

so that Y'~X.

The following theorem is essentially due to Gaffney

[ 6 ].

'rh eor em. The operator L, defined by (3.18) where Sand T

satisfy (3.13)i5 self-adjoint.

Proof: By a theorem of von Neumann,

-1 -1
(I + TT*) and (I + S*8)

are bounded and self-adjoint we set

-1 -1
t1 = (I + 'r T* ) + (I + 8*S) - I

which is bounded and self-adjoint, we shall prove the theo-

rem by showing that R= (L + I) -1 =. Firat,
-1 -1
(I + TT*) - I = (I - (I + TT*» (I + TT*)
-1
-TT* (I + TT*)

which shows that

 -211-

J . J. Kohn

-1
(j{ (I + T'l'*) C Dom (T'1'*).
-1
s1m1larly (}((1 + 3*3) C. Dom (S*8) and we have

-1 -1
R = (I + S*S) - TT*(1 + TT*)

so since ST =0 we have d(un c Dam (S* S ) and

-1
S*SR = S*5(1 + S*S)

Slm1larly (}(.(R) C Dom (TT*) and

-1
TT*R = TT*(1 + TT*)

so that cR.(R) c Dom (L) and

-1 -1
(L + 1)3 = TT*(1 + TT*) + S*S(1 + S*S) + 3 = I
. 2
Finally L + I 1s injective. s1nce ((L + 1)/f.tp)':::; l\epli.
-1
and therefore. R = (L + I) •

We def1ne the spaces ~ and ,1'by:

(J.21) :i) = Dom ( T* ) () Dom (S)

();"22)

and the herm1 tlan form r+ Jrr:

q : '- ><.
c
~ iL by

(J.22' )

rhen (3.15) can be wr1tten as

2
(J.?) \\f\\ :So co ns t , q (,p.t) for
~
1ff-5.
 - 2.12 -

J . J . Kahn

Now observe that if ~ 1s

dense in B then, setting

.~. = {~~B \ fl Je) I

we have 1'dense in .~. Since i f y~~ there ext s cs l' 'E l)

which is close to;r and y I = ~" + -e where ~t-£ and
''I' l ¥ h e n c e 1'''e-1 and henceis close to .0' •
lf

The following result is often called the Friedrich

representation theorem.

Theorem. If r c ~ is dense and y: is a Hibbert space with thE

norm q, a Hilbert space with the norm i\ \\ and i f (3.2)
is satisfied then there eXists a densely defined F :~-7 ~
such that Dom (F) C ~ and

(3 .24 )

and all 'P(:1[ Furthermore, F is self-adjoint, onto and

Proof: Given ~E .~ then consider the function T"" a:-.~~

given by ~(I;#i) = (~,lji) from (3.2) we have

so that T is a bounded functional and hence there exists a

unique K(c()E;1'which represents T«.,_ that is:

(3.26-)
 -21 3-

J ..J . Kahn

for all Iff-9=: Us1ng (J.23) and (J.26) we have

2
\\KIi\\ S canst. q (1<:0(, K'J"_) ~ const. \ k, Ko<.)\

f, const. ilocl\ HKo(lI

hence

\\K-t(\ ~ const. ll""- \l.

Furthermore, for <:/. , p<=-5we have
q(&X, K~)
•
= (~, KP)
I
= q(Kp, •
K~) = (~, K~) = (K~,p).

Therefore, K : ~ --=r J'

1s one-one, bounded and self-
adjo1nt. Sett1ng Dom (F)= ~(K) we define F = K- 1. a nd 1t
has the des1red properties.

Defin1ng F by setting
1

(J. 28) Dom (F ) Dam (F) +Je,

1

F F on Dom (F) and F = 0 on Je.

1 1

we obta1n that F is a self-adjo1nt densely def1ned o?erator

1
We cla1m that Dom (F = Dom (L) and that F = L. It is
1) l

clear that Dom (L) C Dom (F l) and that F l = L on Dom (L)

therefore 'Dom (L*)~ Dam (F l*) and s1nce both Land Fl are
self-adjoint, we have Dam (F 1) = Dam (L). Thus we conclude
thato-t(L) =q ani from (J.l?) we can conclude that

d?( Si') ", J( , .::,' -;) a nd :R.('r) = O« T'r*>.

 -214-

J. J . K ahn

Further we define the operator N B '-.7 B such that

~. 0 on J{.
N =
L K on 5'
Denoting by H : 8 -7X the orthogonal projection into
~ we obtain the orthogonal decomposition of

D. 30) cJ.. =. TT*No( + S*SNv( + Ho<.

NoW i f Sa{ =0 we have from (3.30)

(3.31) SS*SN~ = 0

and thus
2
(SS*SNoe, SNoe.) = ~S*SN,( II = 0

and hence S*SNol. = 0 and

r/.. = TT*No( + Hli

Thus we see that the necessary and sufficient condition to

solve the equation d.. = Tu is that So( = 0 and H4( = O. Furth-
ermore, the solution u = T*N~ is the unique solution ortho-
gonal to n( T) •

These facts are summarized in the following theorem.

Theo~.. Given that A, Band C are Hilbert spaces and

T : A -7 B, S : B ~C are densely defined closed operators
such that ST = 0 and setting £J = Dom ('I'*) (\ Dom (S),

;K='Il(1'* ) rJ '11(s) and 'j : {rr /l) \ tr 1 df.1

 -215-

J . J . K ahn

we assume that PJ is dense in B and that

.for all If W. Then the space B spli ts into an orthogonal

decomposition

Furthermore, the operator L= TT* + S*S whos'e domain is

is self-adjoint and has a closed range and there eXists a

unique bounded self-adjoint operator N : B -7 Dom (L)
such that

D.}2 ) LN ~ I - Hand HN = NH = O.
where H B4 Jf is the orthogonal projection onto -;.e. It
then follows that each o:eR has the ortho;z:onal decomposition
(}.}O) and that the necessary and sufficient condition for
the existance of a solution u satisfying 'ru =~ is that
S =0 and ~J.Je. then u = 'l'*N"" . It also follows that If
P : A -7 7'l(T) is the orthogonal projection onto neT) then

(3.33) P = I - I'*,'H.

Proof: All that remains to be proven is (}.}3), that is,we

must show that the onerator P defined by (3.33) is the or-
thogonal projection of A onto n (T) . It suffices to show
I'P = 0 a rid J« ? -
/C"'
I) 1.
71 (r) ,
 - 2 16 -

J. J . K ahn

now
TP = T - TT*NT = T - LNT = T - T - HT = 0

since HT - O. Flnally P - I = -T*NT and hence

rR-(p:- I) C O{(T*) .i. ll(T).

The above Hllbert space set-up ·ca n be applied whenever

we complete the spaces~,q with a Hllbert norm, elther llke
the ones lntroduced ln 0.2) or with welghts. The same se t
up works for domalns in complex manifolds. In that case we
put a hermltlan metrlc on the manlfold thls lnduces an lnner
product on the forms at each point, the lntegral of thls In-
ner product ls then the L 2- lnner product. We shall also
lntroduce welg~ts here.

Deflnltlon. If [Lc iii , M a complex hermltlan manifold, then

suppose that ~ ls a non-negatlve function, we form the lnner
product

0.)4)

where ~,t~a:;q(1J~> denotes the lnner product deflned at

each polnt and dV denotes the volume element. Denotlng by
p,q p,q
L (Q,A) the completlon ofC[ under the norm assoclated
wlth ().)4) and by T and S the closure of the operators
,..,p,q-l p,q P,q (1 Q, q + 1
a : {A.
/y
"3:(J:, ~ LA- and ~ LA.- 'lie then de-
flne the operator L as above and we say that the ~-NeumaIUl

problem for forms of type (p,q) on [L and weight A ls to

 -21 7-

J.J .Kohn

prove the exlstence of an operator N as above.

As shown ln the theorem, the exlstence of N then lm-

plles that the range of als closed ln L2 so that N does not
exlst ln general. we shall solve the a-Neuma nn problem on
certaln types of domalns by means of establlshlng estlmate
(3.16) and thls estlmate ln turn wlll follow from stronger
estlmates. Further, we wlll lnvestlgate varlous klnds of
regularlty propertles for N whlch wlll automatlcally yleld
regularlty results for the a - problem and for the projectlon
on holomorphlc functlons (3.33).
 -218_

J . J . Kohn
Lecture 4. The basic a priori estimates.

We will proceed to solve the ~-Neumann problem as set

up in the definition of Lecture) in the spaces ~iven by
(J.4). We will assume that n. has a smooth boundary defined
by a function r as in Lecture 2. First we wish to find the
smooth elements in.1)= Dom (T*) J' Dom (5). Denote by ~ the
space of 'f'e.a which are C'" i-n 11 , Le.Z> = ~n&,q. Then
1!fE:~ and 'II~ o..p,q we obtain, by integration by parts:
->.
(4.1 ) (oyJ, i}-IJ)f (e =
(~).
jl
ID ,
T
311J)
/'
dV

= S<,eY.(e
JL
-~), YJ> S d 'v - e -;\
bSt
.(€(~,dr)'f" ~> d\/
= (e~ j'(e->''f) ,'V) - ' (~-~ <G"(~,dr).,,'tI> av ,
(>.) {~..

where ~ ~ CCJII(.fi) , A ~ 0, ~ denotes the formal adjoint of a

and (7'(S';dr) denotes the symbol of ~ evaluated on dr.
Here we assume that r has been normalized so that

(4.2) <dr, dr) = 1 on b11..

Before proceedin~ we will illustrate the formula (4.1)

in the case p = 0, q = 1,11. C (: n and A = 0; 1 t reduces

(4.) (-2,jz ,u) +XJrf

b1l z j
U dS •
j j

S1nce the boundary term in (4.1) vanishes when ~ has com-

oact su npo r t 1n..o... we have
 -219 -

J . J ; Kohn

(4.4)

for all 'f'!'EQP,q-l wlth compact support and slnce the set of
these ls dense, we conclude that

>- -).
=e j":(e 'f) for

Then, slnce (\f,aI/J) = (T*\f1''jI) for 'fE;.~ and

(>-., (>.)

~€ ap,q-lwe conclude that the boundary term ln (4.1) must

vanish for al'l lfi<i:.aP , Q- l and hence we obtain

(4.6 ) cr (J',dr)cp = 0 on brr lf \f7f:t).

Conversely 1t ls clear that 'fE:Dom (T*) whenever f((lP,q

and satlsfles (4.6). Therefore, since Dam (5) ~ ctp,q,
we conclude that J!J conslsts of all f "o~,q for which
(4.6) holds.

It wl11 be useful to express the ooerators ~ and Jr

ln terms of a special basis of vector fields ln neighbor-
hoods of boundary points. Let U be a small neighborhood
of PEbSl. and on U choose an o r-thcno rma L set of vector-
fields of type (1.0), L , ••• ,L such that
1 2

L (r) o if j < nand L (r) = 1

j n
1 n
Let w , ... ,~ be the dual basls of forms of type (1,0) on
U.I'hen 1.f o/EQP,q on u(j Sl , If ca n be expressed by
 - 2 20-

J . J . Kahn

(4 . 8)

where
I = (i , ••• , i ) w1 th 1 <. 1 < 1 <. •• < 1 < n
1 p 1 2 p

J = (J • • • ,J ) w1th 1 ~ J < <. J <: n

1 q 1 q

and
1 i i
IJ J P 1 j
co = (..,. 1\ • '11 6J 1\ WIt.."1\ os q .

'!ben on u()ll. we have

au =~E (u )c;;J
J
JIJ
. ~ en
. T
= c:~ Ej (toT )w It.. W - + •••
1J

and
IH
(4.10) = 2: ... L (If J. + •••
jfH - J I<j H)

where the H run over ordered (q-1)-tuples, <j H) re pre-

sent ordered q-tuple whose elements arecomb1nat1ons ot the
~ and the elements ofH.The dots represent l1near
IJ

From (4.10) 1t 1s 1mmed1ate that the cond1t1on (4 .6)

on un bfi is equivalent to

(4.11) to
T IJ
=0 on U bJ2.. whenever
(\
nE-J.
 - 221-

J. J . Ka h n

The fact that the operators T* and 3 are of the

form given above and that the boundary conditions are of the
form (4.11) show that the argument of Lax a nd Phillips (see
f'31] can be applied as in Hormander [l l) ) (proposition
1.2.4) to prove the following:

Proposi tion. :l> is dense in ~ unde'r the norm

222
II T*f\\ .» + If\! + \ If\\ \\3 (A) o..} •
This proposition enables us to prove estimates for elements
in ~ from which we can then deduce the crucial inequality
(3.15) of Lecture 3 for e l.emen t s in2\. Let (8 k) be the
ij
n x n matrices defined by
n k n k _
(4.12) LL,r:J=2:a L +;> b L
i j k-1 ij k ~1 ij k

k k
then a = -b
ij ij

Let (c ) be the (n-1) ~ (n-1) matrix defined by

ij
n
(4 .13) c a for 1 c: r , j c: n-1 "
ij ij

It then follows from the formula for the Levi-form ((2.31)

of Lecture 2) that C is the Levi-form in terms of our
ij
basis.

The following formula is at the root of all the

estimates which we will derive here. If If, E j) and the
 -222-

J . J . Kahn
support of If lies in JlnU ' then
2 2 2
(4.14) KT*lf" + US<p1l = \\IfU _ +
(~) (A) (A)Z

-A
+2-), S
.n,
If if
[jkJ I<jK> I<kK)-
e 4V

+ L5 e If'
- If. e
-)..dS
bn. jk !-<jk') I<:kK,>

where
2 2 2
(4.15) IIIfH -
(?-)z
= '2\\LID \\
j J IJ ('>I,)
+ \\ If\\
(/\)

and the a are defined by:

Qtt)
i
(4.16 ) A[jk:"\ = L L(?\) + ~a L (",).
'J j k jk i

Observe that the norm defined by (4.15) is equivalent to

(4.17)

We will derive formula (4.14) only in toe case p = 0.

q = 1 and A = 0, to obtain th e gene r a l case one proceeds
in exactly the same manner, the calculations are then some-
what more complicated.

From (4. 9) we obtai n

 - 22 3-

J. J . Kohn

(4.18)
.
~alf'\\ = L
2
I\L If - 4.. If \
l2 + 0(\\4>1\ \\If\\ )'
j<k j k k j z

where ~\4'Il- = lllfn _.

z (O )z

Then
2 2
(4.19) ~ III: If - I: 'f
j<.k jk kj
II = 211~
j,k
If' R -
j k
L
j,k
( ~ If '
j k
L If)
k j

and

(4.20 ) (I:~' ,I: tf )

j k k j

To justify (4.20) we observe t hat if u,V cQO(lln U) then,

o
by inte ~ration by part s

(4.21) (u , L v) = - ( L u , v) +
k k
SL
bJl.k
(r )uvdS

+ o (\\u I! Il v )/ ) ,

t he boundary term t hus appears only whe n k = n since

Lk(r) =~kn. 'r he boundary does not appear in (4.20) since
.
Ife-:D implies t ha t (r If ) . i s z e r o on biL; be ca use t hen
j n

tfn = 0 on bSL so the t e rm va nishes when j = n a nd i f

j <. n then r lf = 0 on b..D.. s in ce L is t.ange n tia l.

j n j

Furthe r, we ha ve
 - 224-

J . J. Koh n

(4.22) (L Lip ,If) = ([L .LJ'f .tp) + (L L If'.Ip)

kjkj k .1kj jkkj

(a Ltp.lp)-(Ltp.L-r)
kj n k j k k j j

+ 0( 1lIf\l_\IIf\\).
z

where again no boundary term appears and we use the fact

that

(4.2) (fL If • If ) = 0 q\\?\L \\tp\\)

k k j z

since the boundary term appears only when k = nand

~n =0 on bll. Observe that
2
(4.2!j.) ',\J <f II = L (LkIfk •Lip)
j j
+ 0 <IIIfL /I If II
Z
).
and that

(4.25) (c L
kj n k
If. '4' )
j
=Sf\
bJ(
c
kj k j
l.P, ijidS + OQ\f\L~f\\)·
Z

Summing on k and j and combining the above we obtain,

I

for tfe,Z)
.0,1
n cOl:lo (U,,5l)
(4.26)
2
1\~~11 + \~lf\\
2
= I\lfll-
2
+
S c If ~ dS
Z bll kj k j

+0 ( II <p/l-Illf/l
Z
)•
 - 22 5 -

J . J . Kahn

which is the desired formula (4.14) when p = 0, A= 0

and q :: 1.

The following theorems give estimates which lead to the

solution of the ~-Neumann problem.

Theorem. If PebCL then the follow~ng are equivalent

(a) There exists a neighborhood U of P such that

..p,q
for all ~ e;2). with coefficients in

(b) The Levi form at P has ei~her at least n-q

positive eigenvalues or at least q+l negative eigenvalues.

In case i t is strongly pseudo-convex, i.e. the Levi

form has n-1 positive eigenvalues, the inequality (4.27)
for q? 1 is an immediate consequence of (4.14) with
A :: O. For q =n being in~p,n means that f vanishes
on the boundary and hence again (4.27) is a consequenoe of
.
(4.14). In case q:: 0 we have ,2)p,q = oY,O and J-cr= 0,
thus (4.14) with A= 0 is replaces by
2 2
(4.28 ) k\.alfJ\ :: lIlfll_ + 0 (lICf\lll'f>ll )
z z

for nP'O
~€ VL. • Condition (b) for q =0 says that the
Levi form has at least one negative eigenvalue so that for
some i <. n we have c 11 (P) 4 0, we can suppose that
 -2 2 6-

J . J . K ahn

C
11
(P) < O. For any function uEC oo (lil"\ U )
0 I I
we have

2 2
(4.29) ~II: ull = -«(L ,1]u,u) + \\L u\1 + O(\\u\Ln u\\)
, 1 1 1 1 z

. b{\.
S c 11 lu \2 dS + O(\lu \\ lIu ll)
Z

tak1ng U small enough so that ell <. ~Cll (P) in U we

obta1n

(4.30)

and from th1s (4.27) follows 1n case q = O. A~ a final

1llustrat10n of how (4.27) 1s der1ved consider the Gase
q = 1 assum1ng that the Lev1 form has at least two negat1ve
e1genvalues • .' Choose a basis L 1 , ••• ,L such that the Levi
n

form 1s d1agonal ~t P and assume that Ci 1 ( P ) < 0

for 1-=. 1 ~ m, CU( P» 0 for m < 1< m' and

C11 (P) = 0 for m' z, 1 < n-1 where m > 2. Choo s e U

small enough so that \ ei i - Cii(P)\< ~ on U. Then

2
(4.31) 1\I:(.j711 2 t III:
i j 1j
lfl\2 -
1 j
u-sij )ln
b
C 11f\ 2dS
11 j

+ 0 <I1'fL lIyll) ,
z

We use th1s formula when 1 ~ t , j < m' and set

 - 227-

J . J . Kahn

£ =£ and f =t when i ~ j sUbstituting in (4.26)

11 ij

we then obta1n
2 2
(4.32) + \\J-crll ~ const.lllfll +
z

+ 5 f f..' (-t L.. c

bSL j=l b'j<m 11
+€ c
jj
) \lp \ 2
j

+ L c
jtm' jj ~Tj
IIIJI I2 + ,... C If' M
17j 1j i.'·'I"jJ
L dS

By choosing E and l) small enough we get the coeffic1ents of

2
I~JI 1n the boundary integral pos1t1ve and bounded away
from zero in U. On the other hand, the last term 1n the
boundary 1ntegral 1s bounded by cons t , L I4>J \2 where the
constant can be choosen as small as we please by making U
small. Thus we obtain the inequality (4.27) in case p = 0,
q = 1, and the Levi form has at least two negat1ve eigen-
values. To establish (4.27) for arbitrary q when the
Lev1 form has at least q + 1 ne~ative ei~envalues we pro-
ceed in the same way.

The proof that (4.27) implies (b) is given in

H~rmander [io]

Proposition. If condition (b) of the above theorem is

satisf1ed then there exists a constant C > 0 and a
neighborhood W of bll
 -228-

J . J . Kahn

(4.33)

~p,q -
for all 'l ~ 0, and If E- o (\ COO (fl 'II U) ,
o
sr
where ~ =e for some fixed large s.

For the proof of this proposition see H~rmander[llJ •

To establish (4.33) in the case that SL is strongly
pseudo-convex, choose s sufficiently large so that
(~ -) is positive- definite as in (2.10)' of section 2; it
"ZiZJ

will then be positive definite in a neighborhood W of bM.

Then from (4.16) we see that !I[jk] is positive definite
and then (4.33) follows from (4.14) by use of a partition
of unity.

Finally the same argument also yields the following

result.

Proposition. If ~E-c""(li) is strongly pluri-subharmonic

in a neighborhood W of bSl.. and i f n.. is pseudo-convex
then (4.33) holds for all 't? 0 and lff,'b,P,q
•
n c0
C>O ( fl (\ U)
when q ~ 1.

In case there eXists a I\ 'E-C~-

(.(L) which is strongly
pluri-subharmonic throu~hout tl, then the integral in (4.33)
can be all taken overIL 80'- thtt't we- have
 - 229-

J. J. Kahn

(4.34)

Hence by the results of the previous lecture we obtain the

following result.

Theorem. If there exists a strongly plur1-subharmon1c func-

t i.on 1n c" <ri. )and 1f .D- 1s pseudo-convex w1 th a smooth
-
b oun dary th en th e a-Neumann pro bl em L p, q ( ~l
f\ , n-"\
.~~
)
has a
a
solut10n for q ~ 1 and ~ suff1c1ently large.

The condition that there ex1sts ).~ C""(J1) which 1s

pluri-subharmonic 1s satisf1ed in ~n, b, taking
2 2
\=\z\ ="Llzli
j

for example. It is also sat1sfied for fL in a Stein

manifold, by taking

r. N
=2:\h \ ,
1 k
2

where hl, ••• ,h n are holomorphic functions that separate

points. Thus we obtain the operator N, for '[ C, and ->
for each 1r and L2 solution of the a-problem
given by T*NO(. Now we have, if SO( = 0:
2
(4.35) I\r*No<!1 = (rr*N«, No£)
(T/,) ('tA)

(0(, Noc)
(to£ )

(cont)nued O~ next Da~e)

 ~ 230-

J . J . Kahn

2
c: C Ito(~,
-'t'-C (1:'>..).

If IJ ls the mlnlmum of the least elgenvalue of

(Az !.") we can ohoose C = q (J -1. In q:n we can select the
1 .1
coord1nates so that the orl~ln "tlesln Sl then settlng
~ =0 d1ameter of..Q. (l.e. ~ = sup' \p-Q\, P, Q ~ SU
ohoos1ng .~_ = 1 z' 2, we have C =q and
_'1:'&2 _ Izl 2
(4.36) .~ < e' ~ 1 for zESt.

Chooslng er = q + ~ -2 we obtaln

Henoe 1n en we obta1n the result of Hbrmander ( [10J

that there ex1sts an L2 solutlon of the a-problem whlch
ls bounded as 1n (4.37).

The above theorem, and also estlmate (4.37) can then

be genera11zed to pseudo-oonvex doma1ns whose boundary ls
not smooth by oonstructing an appropr1ate sequence corres-
pond1ng to the exhaustnon of Jl = Un. (see Hermande r- [10]).
c
 -231-

J .J.Kohn

Lecture 5. Pseudo-differential operators.

In this lecture we will give a brief review of Sobolev

spaces and pseudo-differential operators. If nee lR n
and if s is a non-negative integer then for ue c"<il)
we define
2
(5.1) !\ull = L \\D""u \\ 2
s \otl~S

and let H (it) denote the completion of c~(Ii) under

s
this norm. Hs (.0.. ) is called a Sobolev space and II u Us is

called the Sobolev s-norm.

These spaces have the following properties:

A. H C£UCH LCl) if s > t,

s t

further if UEH we have

s

and i f t < t.;: s then for any i)o 0 there eXists a con-
o
stant C(l) such that

for all ufH (fL).

s
B. If u~H (fl). "11= 1.2 ••••
~. s
 -2 32 -

J . J . K ahn

and 1f II u)\ -<. C independently of -V then there exists a

subsequence tUv 1 which conver~es in H (il) whenever
J t

t < e,

C. If fl has a smooth boundary and if s >k + tn then

k _
Hell> C C (0-)',
s

this follows from the inequality

for all u~C~«(l). A corollary of this property is that

u~C""(.rL) 1f and only if u~nH CO...>.
s
By dual1 ty we define 1\ \\ and H for negative in-
s s
tegers as follows. If UECOO(D.. ) and if s <0 then we
define 1\ ull by:
s
(u,v)
(5.5) \\u II = sup
s II v l\_s

where tbe sup is taken over all vEC~(li). From (5.5)

follows the generalized Schwarz inequality

(5.5' ) ,\(u , v )\ S \Iull t\vlI

s -s

whioh is very useful.

H (.Q.) is then again defined as the completion of
s
 - 2 33 -

.T . J . Kohn

c" Cn.) under Itu II and the properties A, B still hold as

s
well as (5.2) and (5.).

For UEC
OO
en.) the above norms can be expressed in
o
terms of the Fourier transform as follows:
s
(5.6) II U\I !".J IIA U\t ,
s

I\s n
where / \ co. CQ..) ~ c'' (~) is defined by
o
~ 2 i
!\ u("S) =
A
(1 + log,) u(~) ,

I'
where v denotes the Fourier transform of v which is
given by
C _lx· 5
(5.8 ) ~ ('~) = ..) e v ( x ) dx •

Formula (5.6) can be used as a definition of s-norms

for arbitrary s€~,of course it then applies only to compactl~
supported functions.

Formulas (5.7) and (5.8) make sense for distribution

With compact support. If u is a distribution on II then
we ~ay that u is locally in H
s <iU if

s
f\ ($u) (; L
2
U1.) for all :rE- Goa
Q
(fU.

loc
We denote bV q (S\..) the set of all which are
s
locally in !ts (Sl ) . rhe n ir UE ~! Lo c (D-) we have
s
 -234-

J . J . Kahn

Su Eo H (.D...) for all ;r E- C oa CIt.).

s o
n n
If P:C 00 (lR ) --7' Cr><>(IR) 18 a l1near operator and 1f
o
U c R~ we say tha~ P 1s of order m on U 1f for every
n
8tH and every ~~ C 04 (Ii) there ex1sts C such that
o

lIS Pull ~ cl\ u] for all u e-c00 (U) •

s s+m 0

If P 1s of order m for every m then we say P 1s of

order -00.
n
It then follows that ao( E:' C""(lR )

m
1s of order m and that I\ 1s of order m.

60 n n
Defln1 t1on. A funct10n PE-C ua x Iii ) 1s called
n
symbol order m 1f for every compact KC'cR and for every
pa1r of mult1-1ndices ~ , ~ there ex1sts a constant C
(depend1ng on K,C< and ~) such that

(5.10) SUp/Do(D0!p(x,~)1 -c C(l + Isl)m-l~J .

K x ~

n n
The operator P:Coo(lq )~C""(n defined by
o

(5.11) Pu(x)

where d~ d~, ••• ,df and x '3 =I x f 1s called t he

n j j
 -2 35-

J . J . Kahn
pseudo-differential operator with symbol p.

In the above example ~ a~D~ 1s the pseudo-differ-

\ot:.l6;m

~ a (x) Sol.., where

g:
ential operator with symbol
I" 1= 'W\. 0(

-gp(.= ~«'1, ••• nand I\'TtI. is the pseudo-differential

operator with symbol (1 ~\sI2)m/2 •

The following are the properties of pseudo-differential

operators that will be most useful to us.

(A) If p 1s a symbol of order m then P defined by

(5.11) is an operator of order m.

(B) If P is a pseudo-differential operator of order m

then its adjoint p* is also a pseudo-differential operator
9f order m.

(0) If P and Q are pseudo-differential operators of

orders a and b then P + Q is a pseudo-differential
operator of order max(a,b). Further the composition PQ,
modulo operators of order-~, is a pseudo-differential
operator of order a + b and the commu tatoz u

[p,Ql = PQ - QP

i~ of order a + b - 1.

The meaning of "PQ modulo operators of order -~" shoul~

be interpreted as follows. If W is any proper neighborhood

of the diagonal in \HnJ(IRn, (proper -means that if UC.IR n
 .ass.

J . J. K ahn
is relatively compact then •
U' = {(x,y)eW\XfUJ

and un = {(x,Y)fW1YfU ~ ,

are relat1vely compact) , then there ex1sts an operator

Po of order -00 such that 1f Uemn 1s relatively compact,
then sett1ng

then for any UEC 00 (R n ) such that

o
=0
tV
U = 0 on U (P + P )(u) on U.
O·

Thus P + Po has a un1que extens10n to an operator of

so that (P + PO)Q makes sense. Further, 1f P'O is an-

other such operator then

(P + P')Q - (P + P )Q
o 0
1s of order -OQ,

(D) The 1nequal1ty (5.9) and the d1scuss10n above show

that a pseudo-d1fferent1al operator P
can be extended,
m
.odulo operators of order -00
n)
to UH s loc(R • A crucial

property of pseudo-different1al operators is that they are

pseudo-local. We . say that an operator P is pseudo-
local if, whenever u is distr1bution in the domain of
P then
 - 2 37-

J . J . Kahn

(5.12) sing supp(Pu) c: sing supp(u) ,

where the complement of sing supp(v) is defined as those

ee
.po i nt s which have neighborhoods on which v is in C •

Definition. A pseudo-differential operator P of order m

is called elliptic if for every compact K there exists
C such that

(5.13) il p (x, ~ ) \ ~ cIs \m for I~ I large,

where p is the symbol of P.

Theorem. If P is an elliptic pseudo-differential operator

of order m then there exists a unique (modulo operators
of order -~) pseudo-differential operator Q such that both
PQ - I and QP - I are of order -~. Further, this Q is
elliptic of ~order -me

In view of (5.12) we have the following:

Corollary. If P is elliptic and u is a distribution

such that Pu =f then sing supp(u)x: sing supp(f), in
particular U€C~(Sn) whenever fEC~(an).

Proposition. If P is elliptic of order m and if

n n n
UfH (1Ft •. ) and Pu~H (IR ) then UE H (S).
-N s s+rn

Further we have
 -238-

J . J . Kahn

(5.14) Hull ~ C(~Pu\l + \\u\\ )

s+m s -N
n
for all u~H (~).
s+m
n
Conversely, if (5.14) holds for all uec " (fi ) and
o
some fixed sand N with -N<s<m then P is elliptic
of order m,

The estimate (5.14) follows immed1ately from the

theorem and (5.9) since

u = QPu + Ku ,

w1th Q of order -m and K of order -~.

Proposit10n. If P 1s a pseudo-different1al operator of

order m > 0 and 1f
2 2
(5.15) <
II u \i h - C ( Pu , u) + II u II

then P is elliptic.

Proof: It suff1ces to prove (5.14) with s = 0 and N = O.

We have
2 m/2 2 m/2 m/2 2
(5.16) /lull
m
=1\'/\ ull
m/2
'S: C(p!\ u, (\ u) + \\u,1
m/2

, {[ m/2J f\m/2 \
~c \( ?, !\ u, u) +

1\t>uIII\ ull + IIul)2 '1

In m/Z J c on t , next pa ze
 - 2 39 -

J . J . Ka h n

2 2
< C{\\Ul\ + La r ge c ons t , iiPu\\
- m-~ -

2 2
+ small const. nuli + I\ui! •
11 m/2

The desired estimate is then obtained by use of (5.).

In case P = ::[ ao< 0"- ellipti city is equivalent to

\oq~m

the condition
ol
L,
\ol.l=m
a.J "0 if 5 " 0

All the above results can be generalized for determined

sys tems. If P represents a kxk matrix of pseudo-
differential operators and u represents a k-tuple of
functions. Defining 1\ Ull by
s
2 k 2
t s. 18 ) /lull = ~ \lu /I
s j=l j s

Then ellipticity can be defined by requiring (5.1 4) to hold

for some -N < s < m. For a matrix of differential operators
this is equivalent to requiring that the matrix of the prin-
cipal parts be non-singular. All the above theorems the~

hold for the case of such operators.

We conclude with the observation that all the above

results can be "localized" in a natural manner 1.e. the
notion of ellipticity can be defined on any open subset of
Rn and the natural analogues of the above results hold. In
 - 240 -

J. J . Kohn

particular we have:

Theorem. If P is a differential operator defined on

UClRn which is elliptic and of order m, 1.e.

P = L ao<D'" with a ol ~ C"'(U)

,\ .q ~m

and (5.11) holds then for

J, S'E: C""(U)
0
with S = 1

in a neighborhood of the support of ! we have

1I5 u \l ~ C(\\)"'Pu\l + \I)'ul\ )

s+m s -N

for all UtC""(U), where C depends on -S, $' ,s and N. It

then follows that sing supp (u) C sing supp Pu.

Proof. Let )1 E:- - : (U) be such taat $1 =1 in a neighbor-

hood of the support of s and 5' =1 in a nei ghborhood of

the support of 3 1· Then let p' be an elliptic differen-
tial operator on Rn which is eqU.al to P in a neighbor-
hood of the support of 5'0. Let Q be a pseudo-differ-
ential operator such that QP' = I + K where K is of
order -C'\} and Q of order -me rnen

J'u =~P.ru + K(.:s- u )

taking s + m norms we obtain

 -241-

J. J. Kahn

(5.20) 1\:Su\l ~ 1\ p:sull + cons tvll uil

s+m s-N

~ \\SPuI\ +1\ \:),"s] u\\ + const.\l-Sull

s s -N

since the support of [p,,51 is contained in the support of

5 and since [p,~J is an operator of order m-1 we have

(5.21) II\[p,;s]ull = 1I1p,!]5 un 'S const.II,~ uj(

s 1 s 1 s+m-l

so we obtain

(5.22) j\:sull ~ II! ull + \lrPuI\ + const.ll~ulf

s+m 1 s+m-1 s -N.

Repeating the same argument with s replaced by s -1

and s by .)1 and 'S1 by S2 where

"5 E: CDo(U) With ~ = 1

2 0 2

I
in a neighborhood of the support of "S and:5 = 1 in a
1
neighborhood of the support of "S. After k steps we
2
obtain

U:S~II ~const·(Il:rPuI\ +\l3 ull

s+m k s k s+m-k

+ II 5 ull
k - N.

Choosing k =N+ s + m concludes the proof of (5.16).

 -242-

J. J . Kohn

It rema1ns to show that 1f Pul V 1s C 6d then u}

V
1s

OIl
C, where V 1s an open subset of V. To do th1s we choose
the ! and :i' 1n Co (V) • Then us1lfll; a standard smooth1ng
operator 1n (5.19) 1t can be shown that 'SuE-H for every
s+m
s and hence 1n C~.

Th1s same result holds for when P 1s an el11pt1c

system.
 -243-

.L.J... Kohn

Lecture 6. Inter10r regular1ty and ex1stence theorems.

As 1n Lecture J we set

(6.1 ) (T*¥l,T*IjI) + (5." 51/1)

("A) eX)

for If ,'fEe~. From (4.14) we deduce the follow1ng

Lemma. '!bere ex1sts C;> 0 such that

2 2
(6.2) IIfll ~C ql\(f,'f) +lIepll
1
p,q
for all ~~ e.t w1 th If = 0 on b5l, where C depends
on /t.

Proof. S1nce ~ =0 on bil we , ha ve

2 _ 2
(6.)) 11111/
T Z
= L~'L1If IJ
1\

also
2 2
\\ 10 \\ <. const.l\f/l
T Z - (,,>-)z

and
2 2 2 2
(6.4) "'flll ~ const.(IICfIl + Il'fL ~ const.lltpll ...
1 z z . (),)z

S1nce the .bounda r y 1ntegral 1s zero we obta1n (6.1) from

(6.4) and (4.14).

From th1s lemma we see that the operator T*T + ss*

.1s elliptic.,. ' of second or.d.e.r and hanaa we conclude that
 -244 -

J . J . Kahn

whenever the a-Neumann problem has a solution N on

L~,q(.Q,r.) then No( is smooth on any open subset on which
~ is smooth. Hence also the solution of the a- 'problem
given by T*No(, when o<l,tp,q and SOl = O,is -smooth wher.-e
<X.. is smooth. Under these circumstances if "S and
'S' EC;Cn) we have, by (5.16) with'

u = No(. and P = T*T + Ss*

(6.5)

'nleorem. If Il. is pseUdo-convex and if in a neighbor-

hood W of b.Q there is a strongly pluri-subharmonic
function x and q ~ 1; or if the Levi form has ei~her

at least n - q positive eigenvalues or q + 1 negative

eigenvalues then for sufficiently large 1:' the a -Neumann
problem has a solution on LP,q (il, 1:: A), and the space
2
xp,q is finite dimensional and consists of elements which
are in COO on Q .
Furthermore, if W= 1L then
(tp,q : I 0 for q~ 1.

Proofs The case W =Jl follows from (4.)4). To establish

the general case it suff1ces:to'show (by the results of
Lecture » that if l\lf"ll is bounded, 4'v..L dtp,q and

that if
11m q (<t>v ,'fol ) = 0
't'i\

then there is a subsequence such that

 -245-

J. J. Kohn

in L~,q. To do this, choose a function J 1e C~(1L) such

that ~1 = 1 in a neighbrohood V of bSL and is zero in

Jl - W. Then, if r is sufficiently large, we obtain the
following inequality as a consequence of (4.33).

(6.6 ) 1I.:s"1'P1I
2
s const.q ('t',\)
.
(Jr.-If', Jd)

~ canst. (q (-t-).) (If,~) + II ~ If' II~

• p,q .
for 'f'.~:b where SEC ~ (.Q.) and ! = 1 on .n.. - V.The.
000

second inequality is obtained by noting that

(6.7)

<:: cons t ,
-
C\IStp" + 1\50\iD\1 )

since
.
ls,s)
1
is a matrix whose components have supports in

J0- - V and hence are bounded by const.IS \. A similar

o
calculation for T*'1 establishes (6.6). Further, from
(6.2) we obtain
2 2
(6.8) \t:r~l\ ~ const. q (lfl,If) + lIS' 5&'11
o 1 (0.) 0

where S' f C"" Ln.) and 5' = 1 on the support of S •

o 0 0 0

Combining (6. 6) and (6.8) yields

 -240 -

J . J . Kahn

Apply1ng (6.8) to the sequence {<f" ~ .we have to conclude

that It j 0 Lf'elll 1s bounded and thus there exists a sUbsequence
tfv
J
such that .r0 <Pv converges 1n L~,q (fl.) thus we see
J
tha t for every r G Co" Ul.) there ex1sts a subsequence Lfv
J
p,q
such that S'lf'v converges in L (Jl). setting r =5 •
J 2 o
and applying (6.9) to the differences I.fv - LPlJ we see that
i J
a Cauchy sequence and hence converges in

Therefore Cf V converges to e E= L~,q(..o. ) and

J
so it also converges to e Since

qt")- (G.e) =0 and 8.Ct:,q we conclude that e=0 as

required.

Th1s solution of the ~ -Neumann problem can be applied

to prove existence of holomorph1c funct10ns as 1nd1cated in
the first lecture. It gives the follow1ng result.

Theorem. Under the hypothesis of the prev10us theorem and

1f there eX1sts P~b~ such that at P there exists a
local holomorphic separat1ng funct10n, then there ex1sts a
holomorph1c runc t ton on SL which cannot be continued to
any domain which contains P in the 1nterior.

~: . As 1n the first lecture we construct a function F

which is not 1n L cn.. ) but such that 01.. = <l F is in
2
 -247-

J . J . Kohn

L (il ) , what is more the oon~truotion oan be so arranged

2
that if F~ is a small translation of F in the direotion
normal to b n. then d.. = lim a F" • '!hus we see that
80(=0 and that d...l.dt°,l. Then

'9l.. = T(T*Nod and T*No< I:: L (nJ thus h =F - T*N~

2

is the desired holomorphio funotion.

Another applioation of the a-Neomann problem is to

prove the Newlander-Nirenberg theorem. To do this. we first

define almost oomplex struoture.

Definition. If Sl is a differentiable manifold and CT

denotes the oomplexif1ed tangent bundle then an almost-
oonplex struoture on Jl is given by sub-bundle T1• O oftT
wlthth~ following property. If ~.1 denotes the conJu-
gate of T1,0 then
1.0 0.1 1.0 0.1
t:T =T ~T i.e. T (\ T =0
1.0 0.1
(:T T + r

If Jrl is a conplex manifold then the underlying al-

most-complex structure is given by the vectors of type (1.0).
It is clear that if there is a complex struoture associated
with a given almost-complex structure then it is unique.

Definition. If ..fL
is almost-complex with t he r almost-com-
plex structure giTen by T1• O is called integrable if for
any two local vector fields Land L' with values in
 -248-

J . J . Kahn
1,0 'I 1,0
T the commutatox [L, L'j also has values in T
1,0
Clearly if T is the almost-complex structure
associated to a complex structure then it is integrable.
Conversely we have
1,0
Theorem. (New1ander, Nirenberg [ 25J i. If T gives
an integrable almost-complex structure on st then Jl has
a complex structure such that T1,0 gives the associated
almost-complex structure.

First observe that this theorem is strictly local, that

is, it suffices to show that given a point P ~ n. there
existr function h 1, ••• ,h n defined in a neighborhood U of
P, such that dh 1, ••• ,dh n are linearly independent and such
that for every vector field L with values in T1,0 we
have

(6.10) L(h ) = L(n ).~O for J = It ••• ,n.

j j

Then hi, ••• ,h

are a complex coordinate system and it is
n
clear that a function u defined on an open subset of U
satisfies the equations L(u) = 0 for all L with values
in r 1 , 0 if and only if u satisfies the Cauchy-Riemann
equations with respect to the coordinates h 1, ••• ,h n•

On an almost-complex manifold the space of form (l has

a natural bi~radation
 -249-

J . J . Kohn

1nduced by the decomposItion ~T = T1 , 0 + TO,l. Denote

by TT : 0. ~~,q the correspond1ng projection mapp i nga ,
p,q

We define the operators d and a- by

(6.11 ) a = 2-
p,q p+hq
rr d IT
p,q

~ = L-
p,q
IT
p,q+l
d TT
p,q

Proposition. For an almost complex structure the following

are equivalent.

(a) integrable

(b) For every form ~ of type (0,1) we have

Erdlf = °
2,0

(c) If <f is a (O,l)-form then d =«)~ + a If>

°
2
(d) If u is a function ~ u::

= a = ° and p'3 :::

2 2
(e) d =~ + a d -a d •
Proof: By a standard form~la for an exterior derivative we
have

(6.12) (d1, L"L') = L«~, L'» - L' «If, L»

- {If' [L, L~/ ,

choosing Land L' to be any two vector vields with
values in T1,0 we have
 -250-

J. J . Kahn

since If is of degree (0,1). '!hus we see that 1T 1. = 0

2,0
if and only of L, L' is of type (1,0) ; which shows
that (a) is equivalent to (b). Clearly (c) is equiv-
alent to (b) since for forms ~ of degree (0,1) we
have by (6 • 11 ) Cl If = 11 dlf
1,1 .
and 3l.f = TT dr so that (c)
0,2
holds i f and only i f dlf = o. By (6.11) we have
2,0
2
(lu= ofT d u hence i f L and L' are vector fields in
2,0
T1 , 0 then

(6.13)

since -au = . du we obtain

1,0

(6.14 )

this is zero if and only if [L,L'] is of type (1,0). So

that (e) is equivalent to (a). Finally t he conditions
(c) and (e) and their conjugates extend to forms of all
types and hence are equivalent to (f).

Corollary. On an integrable almost-complex manifold a

function .u satisfies the equation ~u = 0 if and only if
du = 'Ou and this in turn 1s equ i va l.e n t to ru = a for all
L of type (1,0).
 -2 51-

J . J . K ahn

If JL ls a domaln ln an lntegrable almost-complex

manlfold then we can put thls manlfold a hermitlan metric,
1.e. a metrlc whose lnner product lnduces a
Aermlt14~ ;

hermltlan lnner product on (T such that T1,0 ls ortho-

gona.l to TO, 1. '!ben all our results ell the a-Neumann
problem generallze wlth exactly the same proofs.
2n
Let UC ~ be an open subset on whlch ls deflned an
lntegrable almost-complex structure • . We choose real coor-
dlnates x 1 , ••• ,x 2n such that the ball
2n 2
B={xe:R2n \
:2: Ix I « 1
1 j

ls con talned ln U. For each t €: LO, iJ we define the map

by Let fJJ
1
t •••
n
,6.1 be a

basls for the forms of type (1,0) 1.e.

j 21"\. j k
(6.15 ) b,) = L.. a (x)dx
k=1 k

for each tE(O,l] we set

j -1:f..* j
CAl = t ~ ·w
t ~.

these define an lntegrable almost-complex structure on B

for each t~ (0.1) we shall denote thls by B • In terms of
t
ooordinates we have
j 2n j k
(6.16 ) t.J = '> a (tx)dx
t j(;1 k
 - 252 -

J . J . K ah n

We denote by d ,
t
.a·t the operators associated with

B. Note that B has a complex structure, in fact setting

t °\
j j k
(6.17) = La (O)x
°
Z

k
j j j - j
we have w = dz =d z so that (} z =° and the

dz
j °
are independent.
° ° Now
° B
°°
is strongly pseudo-convex
° j 2 °
and the function ~ =LIz \ is strongly plurisubharmonic
. °
on B. Hence for small t, B is strongly pseudo-convex
. 0' t

and ~ is strongly plurisubharmonic (in the sense that

is positive definite). Choosing a hermitian metric

° when
which varies smoothly with t we conclude that there exists
. 1,
a ~ <0 a fixe.d rt such that (4.)4) holds in 'J)
t

t ~ ~ with a constant independent of t i.e.

(6.18) ~lfll
2 ~ const.
(l\T*lfl\
2 + \\5 \f\\
2 \.;
t ('t.q t t ('t>') t t ('t:f.)

.1,0
for all If ~ 1J where II denotes the norm in
t t

the space LO,l(B ,tA) with respect to the hermitian metric

2 t

on B. Hence, when the a-Neumann problem on

t
°
L ' 1 (B ,'r,x)
"jO,l
has a solution and dU = 0. Hence if
2 t t

~ and~' have supports in the interior of Band J' = 1

 ~253-

J . J . Kohn

On support of j then (6.5) holds independently of t, i.e.

(6.19) liN 0<11 <;. const. HI'S'olll + \\"t.\I)

t s+2 s .
for all of.. G L 0,1 (B) and t <.. 0. Further, we have
2 t -

aU
(6.20 ) a u(x) = Lb
k
(tx) _ _ W
1

t i axk
where
k j j
(6.21) ~b (tx) a (tx) =~
k 1 k 1

Hence 1f U,,=C(B) we have for any 0(

o () o"'-
ClI_
(6.22) 11m u = d u
t~O t 0
j
uniformly. Now we define the function z by
t
j j j
(6.23 ) z = z TiN ~ z
t 0 t t t 0

- j
and we have d Z = O.
t t
In (6.19) choose 3", 3 and S' with supports in thE
o
interior of B such that ) =1 in a ne1ghborhood of the
o
orig1n, S' = 1 1n a neighborhood of the support of S a n d
o
5' =1 1n a neighborhood of the support of J; then
 - 254-

J . J . Ka hn

J J J
(6.24 ) 115 (z - z yu ~ I,\T*J':N "§ z ~
o t 0 s tOttOs

~I\"SN "3 zJ U
t t 0 s+l
<... const. Ij'~z\\ l - t'OJ s +
- J\
+HJz\.
t 0

Now for s large we have

(6.25) sup
'"
~
\
D~(Z
j j t
- z )( ~ const.R! (z - z J j
)U
V \O(,~ 1 tOO t t s

henoe, since the right of (6.24) goes to zero as t -70 we

see that if t = t is sufficiently small then on V the
o
1 n
gradients of z , ••• ,z are linearly independent. Thus
to to
8 admits holomorphic coordinates in a neighborhood of the
to
origin and hence so does 8, in fact the coordinates on
1
j -1
8 are given by z (xt ), which oroves the theorem.
1
to 0
 -255-

J . J . Kohn
Lecture Z. Boundary regularity.
In this lecture we will assume.n.. is a domain in a
complex manifold with a C~ boundary which is pseudo-convex
and such that there exists a strongly piurisubharmonic func-
tion in a neighborhood of b 11 • We wish to discuss smooth-
ness of solutions of the a-problem and of the 3-Neumann prob-
lem in the ca~sed domain Jl, i.e. up to and including the
boundary. We will restrict our attention to the a_problem
for functions or equivalently the ~-Neumann problem on (0,1)-
forms. A nautral question to ask is, given a (Ot1)-form
=~v when does there exist a solution u of~ =a such that

(7.1) sing supp (w) sing supp ( i.

It is easy to see that every solution u has this

property in .n.. ; however, we wish to interpret the above
in .n. . The problem is more delicate there for i f u is a
solution and h is a.holomorphic function then u + h is
also a solution which in general will not be smooth on b(l.
The following example shows that it is not always poss-
ible to find a solution satisfying (7.1). Let Il c ~ 2 such
that in a neighborhood U of (0,0) we have

(7.2) 11 c, U= (zl,z2)~U
I Re ( 2: 2 ) .<- 0
Let r e CS( U I ) , wlth U' a neighborhood of (0,0) and il' U

such that, r = 1 on a neighborhood

_..E.... U
V of (0,0) and let
(7.3) .{ = J(z2)= z2
 -256-

J .J .Kohn

Then the support of ,'- is contained in W = (U' - V)

We will show that every solution of 3u = ~ 1s unbounded
in .!L - W'. where WIt is a neighborhood of iii. For
simplicity we suppose that U and U' are so chosen that
for some posit1ve numbers A and ~ 0 the set

(7.4) f(z ,z )
1 2
I \z I·~
1
A, 1m(z )
2
= 0, -~ ~
0
Re(z )
2
<. 0 ~

is contained in n and the set

S" (z ,z )
(. 1 2
I I z 1I = A, 1m (z )
2
= 0, - ~ < He(z ) <. 0
0- 2-

.J'
does not intersect U'. Let h =u - ~2' then h 1s holo-
morphic; consider the restriction of h to the line
zi =- ~ with 0< a< ~O we have
(z , -'f,)
h(z ,-g) = u(z ,-~) + ~1~__
1 1 ~

If u is bounded by K in ~ -W' then h(zl'-~) is bounded

by K on the circle zl = A, Re(z2) = -0; on the other hand
1
h(O,-b) 2. S -K

which is a contrad1ction, since h(O,-$) is an average of

h(zl'-~) on IZl/ = A.
The theory of interior regularity of elliptic operators
extends to boundary regularity for the so-called coercive
boundary value problems. Suppose Q is an integer-differ-
ential form on k-tuples of functions on .fL expressed by:
 -257-

J . J . Kohn

where u = (u , • •• , u ), v = (v , ••• , v) and

1 k 1 1 k
u ,v a 1 j E C (jt.). »: h
,0 -
Suppose that B 1s a k matr1x of
j j'
Cl>4 funot1ons on bfL such that B 1s of constant rank.

'!'h en 1f

k .:2 ~ k 2 )
Z \Iii II ~ cons t , Q(u,u) + L.. \\u \\
j=l j 1 j=l j

for all U wi th BU =0 on b Ij, , we say that the Q re-

str1cted to the space of k-tuples satlsfY1n~ Bu =0 1s

ooero1ve.

Theorem. Suppose Q w1th the boundary cond1t1on Bu =0

1s coerc1ve. Let

03 = {u \ Bu = 0 on
r--:
and let a3 denote the complet1on of c5 under the norm ~1ven
the r1ght s1de of (7.7). Then 1f UE- cr; and sat1sf1es by

Q(u,v) = (f,v)

for ~ll Then s i.ng supp(u) C s Lng supp(f)

and 1f

3, :S' E C""(0.), .s'u L <..~U and S' t; H s (J2.)

o 2

then
5 u~H ea) and
s +"2.
 - 258 -

J . J . Kahn

n1 ulls+2 'So canst.(ll.:s"f11 + IIs'u

s
U) .

The ~-Neumann problem 1s coerc1ve (i.e. (7.7) holds

when Q =q act1ng on b) only when q = n-1. To see this

note that (~.1~) 1mp11es that

(7.10) q(lj>,lf) $ cons t , (~lfl\_ +

2 5 I 'PI. 2
n dS + Ulfll )
2
z b~l.

hence 1f (7.7) holds then

2 2 S 2 2
1\.lf.1I S cons t , (IItfl\_ + IfI dS + Illfll )
1 z bU,.

and th1s is true if and only if ~ = 0 on . bit.

We do however, have the following global regularity

result (see Kohn 117] ).

Theorem. Suppose b Jl. is Coo, Jl pseudo-convex and if

there ex1sts a function A wh1ch 1s C"" 1n a neighborhood
of bit. Then for every integer k there ex1sts a number
~k such that if N g1ves the solut10n of the ~-Neumann
problem 1n p,q
L (il,~~) w1th q ~ 1
2

then
p.q k _
N (9- ) C C (.Q) if I'r
."
>- Itk

e
Corollary. Under the samle hypotheses on n if ocE ap.q
q 2 1, and if au = t::J.. 1n L2 then for each k there ex-
 - 259-

J . J . Kogn
k _
ists u EC (fU such that ~u =~.
k k

Here we give a brief outline of the proof of the above

theorem. First we set

(7.12)

Now for each p,q

and each of €:- L (iU
2
q.,p,q
there exists a unique lfrE-.u such that
(fA p, q
for each tj; f- <U •

We wish to show that given an integer s there exists

a number A such that '0 ~ H (n) whenever -r :> A and
s '-r s l- - s
that in fact

(7.14 ) )\lRt: II s ~ const. Hc<l!S •

To prove that l.f. E H (ll), we first prove ·(7.14) under

'l: s

the assumption that <fr: is smooth. For PEb.D. choose a

"boundary coordinate system" in a neighborhood U of P
that is, we choose functions x 1 •••• ,x 2n-l . such that
1
dX •••• ,dX 2n- 1,dr are ' . UlJ f; ~ r l y., independent t.hrougnou t
U we say that x 1, •••• x 2n- 1 are the tangential coordinates
and x 2n = r is the normal coordinate. Note that if
lf~j) and the support of ip lies in U (I {l then for any
index r:I.. 0:'f' f- V. where
 - 260-

J. J . Kchn

01. 101- I
D = (-L) -~----
b

and
tX ~ . 0( IJ
D4'= L. D" W
b IJ b IJ
where
~ IJ
= Ltp W
IJ

starting with the inequal1 ty (4,;3')

2
(7.16 ) -r IIlfll ~ C Qoc(lp,tP) ,
(t"~ )

sUbstituting D
01.
rtf., for I.f in (7.16) we obtain
b or

We wish to estimate the right-hand side of (7.17) by using

(7.1;) with an appropriate \f f: i>. Integration by parts
gives
(7.18)

+ ol!"L II D~!'f. 'i\l

1'~I~s b (TA) '!:

+ Cit' L..\\ DTj ' cp.'[" If

('1 <.s

where the 4 "through m-tuples, :r'f:C~(U(\?L) with .:5' = 1

runs 0
 -261 -

J . J . Kahn

on support of S. Further, we have by (7.12) and integra-

tion by parts

= (D.,(5tl, D" S .\ + 0 ( c.-l\o<: II 1\ Cf....\\ )

b b 'Pc ~ s . ~ s-l

The fact} b~ is non-characteristic (which is a ~on­

sequence of the ellipticity of the "laplacian") implies:

Assuming (7.14) is true for s - 1 in place of s, we

obtain from (7.18), (7.19) and (7.20) With ~ sUfficiently
large

Ttle fact that tp-r; and cY-. are connected by an elliptic

system 1.e. E-c(<f-t) = 0( to solve for the normal der1vates
in U as follows
~ KL
(7.22) =L a r:J.
IJ KL
 -262-

J.J. Kahn
where F is an operator of first order. Differentiation
IJ

of (7.22) and use of (7.21) the yield

2 2
(7.23) 1I"!4'~" ~ cons t , \lcLll
("t,q s

The desired estimate (7.14) is t~en obtalned using a

partition of unity w1th (7.23) for nelghborhoods of boundary
po1nts and (6. 5 ) for neighborhoods of interior points. To

~ctualll l>!'Ove that CP" E H (St) we use the method of

s

"ellipt1c regularization" (see Kohn and Nlrenberg lt9] ).

Cons1der the form

where r >0, !"f'L

the l.) ) are a parti tlon of unity and '1', """t>,p,q .
v
~en, clearly Q is coericlve 1.e. (7.7) is satisfied.

Hence, 1f we denote By
'" ""p,q
Z
the completion of
:l5 p,q
~ ..... p,q
under Q~ then the unlque <f't'E:~ whlch satlsfies

. p,q
for all Iff- .3 , is in C t>O (Ii) and hence Now,
the same derivat10n is that of (7.14) yields

where the constant is independent of ~ when ~ is small.

 -2 6 3-

.J . J. Kohn
p,q
Definition. We say that the a-Neumann problem in L (ii)
2
is !-subelliptic at PfbJIL if there exists a neighbor~
hood U of P such that

(7.29) v: 'I'll
b
2
~ canst. Q(V,'f)

for all lfE-ilP,q (\ Co (Urn. )

where

and

The estimate (7.29) is equivalent to

2n t -1 2
(7.)0) L \\ f\ ~"Sif II s canst. Q(tr./f) ,
j=l b a xj .

in particular if £ =1 then Q is coercive (i.e. (7.7)

holds). If (-ell ipticity holds and if '£. <;:. 1 t hen necess-
arly ' £' s ti

Theorem. If ..Q is such that at each Pfb11. the 3"-

Neumann on LP,q (fl "
) is 'i.-subell1ptic then it is solnable
and (7.27) holds.

From the theory of Sobolev spaces it follows that the

norm given on the left side of (7.30) is compact in
LP,q(Un.n.) from this it follows that the ~-Neumann problem
2 p,q
has a solution in L (fl). The pseudo-Iocalness depends
2
 -264-

J. Jc.Kohn
It then follows that there ls a sequence 0 v ...-" 0 such

that the arlthemetlc means of the sequence (Cf; v J are a

Cauchy sequence ln Hs and slnce they converge to ~~ ln

L thls lmplles that lfrr f Hs • 'ro pass from the smoothness

2
of f/f1 to the smoothness of N't(l( reQ~lres a functlonal an-
alytlc argwu.nt gl ven ln Kohn ' [171 •

We now return to the questlon when the a-Neumann pro-

blem has boundary regularlty. More preclsely. we want to
show that. under certaln clrcumstances. the operator N ls
pseudo-local; ln the sense that

slng supp N ~ C slng supp 0<..

where the slngular support 1.s consldered ln .D..

If U ls a boundary coordlnate nel~hborhood as above,

wlth boundary coordlnates x 1 ••••• x 2n - 1 • x 2n = r then
s .., 2n
sell( and u~C""(U(l.R) we define !\ u E C (lR ). where

I
o b -

IR_2n = {(x
1 ••••• x 2n ) x
2n "S. 0 J •

~ 2 ¥or-' >
/\ u«( .r) = (1 + I ~ I ) "u(~.r"
b

where S= (~ ••••• S2n_1) and ~(~,r) denotes the partlal

Fourler transform deflned by

(v(f.r)
\.- = 2f -lx·f
1e v(x.r)dx
~n-

1 2n-1
where and dx ,- dx • •••• dx
 -265-

J. H . Kahn

on the follow1ng: suppose ~~b is the solution of

(7.31)

Then if "S ,:S'fC;(U(\n.) with :S' =1 on the support

u- 2
of :5" and 1f ! k e Hs rn ) then J\ b )'fE- Hs~2 (Jl) and we
have
H.-2
(7.)2) 11/\b S<p II s+2 s const.I\3"'cXll •
There e~flmt~es ;arl!J:Obta1ned ,b y cr epl ac ing ~ 1n ; (7 ;jO) ,"'With
Ar:; then proceed1ng by induction on m with the same
type of argument as in the previous theorem.

Theorem. If the Levi-form at ptbSl either has at least

n-q pos1t1ve e1genvalues or q+1 negat1ve e1genvalues
then the a~Neumann problem 1n L~,q(Jl) 1s t-subel11pt1c
at P.

~: Due to the results of the prev10us lecture 1t

suffices to show that

(7.33)
t 2 (2 J
2
!t(\bu\l 6 const. l\ulI~ + b.f\. [u ] as + ~u\\
2)

for all UE-C;(U (\ iL).

LetL1, ••• ,L be a basis of the vectorf1elds in T1,O

n
on U such that LJ(r) = 0 and Ln(r) = 1 throughout U.
It 1s then eas11y verif1ed that 1f for each r, Pr 1s
2n 1
pseudo-differential operator on nl - which var1es
 - 26 6-

J . J . Kahn
smoothly w1 th 'r (1. e. 1ts symbol p+-(J\, '3) e c <lQ as a
rune mon r , x and ~ ) then on U C \R.2n the operator
[Ln,PrJ acts as a pseudo-d1fferent1al operator on
r = const. of the same order at Pro Now 1t suff1ces to
bound \\ /\ -t ...:Lu { by the r1gh t-hand s 1de of (7.33) w1 th
b axj
d
j < 2n. S1nce these are l1near combinat1ons the
aXj
vectors Ll' ••• ,Ln_ l ,L l , · · · ,Ln _ l and N = Ln - Ln 1t

therefore suff1ces to bound

1\ 1\ -f L u\{ ,\Ill -i L u~ 2
b j b j

and for j< n and V\ -'NU\\2. The second of these 1s

1mmed1ate.
-I 2 o 0 0
(7. 34) \\ f\ L u 1\ = (L U,P u) = (u,P Lu) + (uQ u)
j j j '
2
~ cons t , (~u II ",ul\- + I\ull
z

where po, QO are operators whose restr1ct1ons to r =

const. are pseudo-d1fferent1al operators of order O.
F1nally:
-~ 2 0
(7.35 ) 111\ Nul\ = (L -L
n n
)u, P u)

in view 'of the above remark this is bounded by the right

side of (7.33).
 -267-

J .J.Kohn
The followlng theorem ls proven ln Kohn 1.16].

Theorem. If J(L1S pseudo-convex and lf ln a nelghborhood

of PfbSl each non-zero vector fleld of degree (1,0)
and values ln T1,0(blL) on bil ls of flnlte type at p
and lf the Levl form ls dlagonallzable ln a nelghborhood of
P (l.e. ln C) the c =c S throughout a nelghborhood)
lJ 11 lJ
then there eXlsts £ ~ ° such that the F-Neumann problem
on LP,q({l) with q ~ ° ls €-subelllptlc at P.
 -26H-

.1. .1.Kohn
Lecture 8. The induced Cauchy-Riemann equations.

In this lecture we will take up systematically the

euqations on b U. which arise in the extension problem
discussed in lecture 1. We define "

(8.1 ) on
.,p,q
Observe that if q =0 then o/€ ~ is equivalent to

(8.2) If=~r;\~+re near bn.,

p,q-1 p,q
where ljJfa and ee&.. From {8.2) it follows that
p,q+1
(8.3)

Therefore we have the following diagram

p,q+1 p,q+1 . d3 p,q+1
(8.4) O-"~ ~ Cl ~ ., 0

~ i
~p,q ~,q
d r f ~b
p,q
o -'7 --'" , ---" B ~ 0

where
rn.p,q I1P,q p,q
UJ =l./I.. I~

and a"b is the mapping induced by a. In case p = q = 0

0,0
the space S is naturally identified with C~(bJl) and
the ~ f = 0 is necessary for extending f to a holo-
b
morphic function.

We define the following cohomology groups:

 -269-

J . J . Kohn

(8.6)

then (8.4) 1nduces a sequence

p,q p,q p,q p,q+1
(8.7) H (~) ~ H (Q.) ~ H «(f, )- - - j " H (~ )

0,0 0,1
1l1ustrate, we define the map H
To ($) -", H (~).
0,0
If f~H (n), f 1s represented by a funct10n 1n -C~(bJl)
wh1ch we w111 also denote by f and wh1ch has the property
rv "" 0,1
1f f
1s an extens10n of f then afe ~ the cohomology
'" 0,1
class" of 'iff 1n H (~) then 151ves the 1mage of f under
0,0 0,1
the map H (~) ~ H (,\:). It 1sclear that th1s 1mage is
,.....
1ndependent of the extension f and that f can be ex-
tended to a holomorph1c funct10n on JL if and only 1f
0,1
is cohomologous to 0 1n H (~); that 1s there eX1sts
0,0
gE If. such that ar
= ;-g. The des1red extens10n then 1s

f - . g.

p,q p,q p,q

Proposit10n. H (~ ) '.:t H (Q), where H U{) 1s the
o o
cohomology of forms wh1ch van1sh on b~.

r p,q
Proof: If If ~ -e then from (8.2) we see that
 - 2-'W-

J . J . Kahn
and that
p,q n-p,n-q
(8.11) F(J{' )C t;,

From (8.9) it follows that F is an isomorphism and from

;,,P q
(8.10) it follows that i f Iff:- J(. , then
Thus we aF'f = o.
p,q
have proved that F induces an isom?rphis~between H (Q)
n-p,n-q
and H (~).

Theorem. If bJL is connected and if the Levi form has at

least one positive eigenvalue at each pE-bSl and i f
f€-C()<l(biU with ~ f
b
=0 then there exists a function

hEC (ll. )
oo
such that h = f on bSl. and h is holomorphic
in Jl .
tv 0tJ
Proof: Let f'-C (J\.) be an extension of f. as we have
seen the desired h exists if and only if -'"
af is cohomo-
0,1
olgous to zero 1n H (~). By the previous theorem
0.1 n,n-1 n.n-1
H (~) ~ J{ . Let I.f' ••••• f. be a basis for df
1 k
so
86 at 1s cohomologous to zero if and only 1f

(8.12) fri;..1/1j
jL
= if'f
bR j
= 0 for j = 1 ••••• k,

Thus we see that the space

(8.13 ) "'{ f EC"" (b.R )[ abf =0}

tf =h on b.a • h coonE) hol.}
2 k+l
has dimens10n ~ k. Thus f, f , ••• ,f have linearly
dependent cosets 1n ~ so that 1f
 -2 71 -

J . J . Kohn
showing that ~ is cohomologous to a form that vanishes on
b It , Further, ~cy = ~(rCa<p+ e) which concludes the
proof.

Proposition. represents a cohomology class in

p,q
H (~) and ~ represents a cohomology class in
n-p, n-q
H (It) then depends only on the cohomology
classes of ~ and ~

" .o.fP , q- 1
Proofl If I.f> = ae with t7<>.." then we have
~I1A~ = d((1,,'/I) and hence

"(8.8 ) J ~e
bn.. A
~ = Ie '4J = a
bIt A

n,n-1
since e/\'t'f~ the same calculation yields the result
when ~ = a-f.
Theorem. If at each point PEb..o. the Levi-form has either
n-q positive or q+1 negative eigenvalues then
n-p,n-l
H (~) is finite dimensional and is 1somorphic to
P, q
H (OJ •
Proof: By our previous results the condition on the Levi-
form implies that problem is ~-subelliptic and hence the
")pp,q
space ~ is finite dimensional and all its elements are
p;q nn-p,n-q
in C on Jl.. We define a map F: 0.,.
(.>< ---" LA- by

It is easy to verify that

(a .10)
 -272-

J . J . Kahn
m j
(8.14) P(f) = ~ a f w1 th a E. ([ , a ~ 0 ,
j=l j j m

then has a holomorph1c extens10n F to fl

P(f)
2 k+l
Similarly, Ff. F f ••••• F f are linearly dependent and
hence there 1s a polygon1al Q so that Q(F)f has an
holomorphic extens10n C;. If f 1s constant there 1s
nothing to prove. If' f 1s not constant then it takes on
infinitely many values and thus F and Q(F) taka,on ,1nfl-
nitely many .values and ,bence c the set of zeroes of Q(F) is
. .
tb"1n• Then ofro III '
(8.15) P(f) = p(~)

on bJl.- {zeroes of Q(F)} we conclude that F = .. ~

wr
holds on ·"'~.- i ze roes of Q(F)j. Thus we know that q/Q(F)
is locally bounded and hence can be extended to holomorph1c
function on II which g1ves the desired extension of f.
p,q
From (8.7) and from the identification of H (00
n-p.n-q P.q
with H (~) we see that H (~) is finite dimension-
P.q n-p.n-q-l
al whenever H (a) and H (~) are finite dimen-
siona1. Thus the following condition implies the finite
P.q
dimensionality of H (£): The Levi-form has e1ther at
least max (n-q. q+l) non-zero eigenvalues of the same s1gn
~r at least mtn (n-q. q+l) non-zero eigenvalues of opposite
signs. In fact this depends on the "internal structure" of
b.fl which can be defined abstractly as follows.

Def1nition. If X is a real C~ manifold. we say that X

 -275

J . J. Kohn
has a partiallY almost-complex structure of codimension k
1,0
if there exists a sub-bundle T (X)CcfT(X) such that;
0,1 - 1,6
setting T ( X) = T' ( X), we have
1,0 0,1
(a) T (X) () T (X) = 0

1,0
(b) dim T(X) =2 dim T (X) + k

(c) · If Land L' are local vector fields with values

1,0
in T (X) then eL, L'j has also values 1n
1,0
T ; . (X)

Under these circumstances we define the space of exter-

1,0 0,1
ior forms Q on T (X) + T (X), on this space this sum:
b
induces a bigradation

(8.16 )

and we define the operators

(8 • 17) d
b
'. op,q ~ I1 P+ l,q and
b
-, Iv(.
b
p,q
-0 b : (;i
1>-
-:r abp,q+l
by setting

1,0
(8.18 ) <d u,L) = LU, u a function L~T (X)
b

and extending this to forms in the usual way. These opera-

tors have the same formal properties as a and d. Re-
stricting ourselves to the ~~e of co-dimension 1 we can
def'ine the Levi-form at Pe X as follows. Take a real l-form
defined in a neighborhood of P, which annihilates
 -274-

J . J . Kahn
1,0 0,1 ,
T (X ') + T Of.) the Levi-form, then is given by

(8.19)
1,0 ,
for L~T (X). Since at each point the space of such
~
is one-dimensional the numbers of non-zero eigenvalues and
the numbers of eigenvalues of the s~me and opposite signs
are invariants. NOW we can put a hermitian metric on
1,0 0,1
(T(X) which makes T (X) orthogonal to T (X) and we
obtain the adjoint of ,.J-
b
of i"
b
. By methods analogous

as for the a-Neumann problem we can prove the following

result (see Kohn In]).
Theorem. Let X be a compact partially complex manifold
of co-dimension 1 with dim 'X = 2n-l and if the Levi-form
(8.19) has either at least max (n-q, q+l) eigenvalues of
the same sign or at least min(n-q,q+l) eigenvalues of
opposite signs. Then, setting

(8.20)

where {j is the adjoint of ~ and

b b

(8.21 ) ob =.va
b b
+~J
b b
'

then there exists a completely continuous self-adjoint

operator
p,q ~ p,q v
N L (,1\\) -') L (I'c) such that N
b 2 ' 2 b
 -27 5-

J . .J . Kohn
is pseudo-local and

(8.22 ) C/N =I+H

b b b
p,q
where H
b
is the othogonal projection of L
2
on on
o,q p,q . !? , q
Xb . Furthermore, i f Ii E. L on, d 0(
b
= 0 and 0(.1 de

then «=:a(,J"N.,t).
b b b

The theory indicated in the above theorem is based on

the estimate
2 2 2 2
(8.2)) IIlflll. <.:. const.(\\ ~ 'fll + ,I\~~l\ + (tr ll )
l!" b b

p,q
for 'ff- ab

Observe that dim x =) and q = 1 the conditions in

the above theorem cannot be satisfied because then the Levi-
form is 1 x 1. In this case can be represented
2
locally by single first order operator. In fact, i f X. eC
is given by
2
(8.24 ) r = Im (z ) + lzI
1 2

then to the equation au = f

b
is the Lewy equation which

has no solutions for" ¥lost" functions f.

The following natural oroblem was posed by H. Lewy.

 - 216 -

J . J . Kohn

Suppose L ls a complex vector field ln tR? , 1.e.

) 'iJ
~ )
(8.25) L = 1:- a ax ae.e (fR.)
J=l J J J

do there exlst non-trlvlal local solutlons of the equatlon

Lu = O. Recently, L. Nlrenberg (see ['3 2] ) has found an
example of such a vector fleld for whlch the only lo cal
solutlon of Lu = 0 are u = const. What is lUore,ln
Nirenberg's example the vector flelds L, 1. and LL, LJ
are llnearly independent. It ls stlll an open questlon
whether on X whlch satlsfles the condltlons on the Levl-form
for q = 1 glven ln the above theorem the equatlon
au = 0
b
has non-trlvlal local solutlons.

To conclude these lectures wewlsh to polnt out an

appllcatlon of these results due to Kerzman (see [30]).
Namely, lf H : L (Il) ~
2
If 0,0 denotes the orthogonal pro-
Jection map onto the space of holomorphlc functlons, and lf
N ls pseudo-local then H ls also pseudo-local. Now H
can be expressed as

(8.25)

Then the pseudo-locallty of H lmplles that

K €. [h'([i ><o f[ - d1a'g (b O ,.. bn) } .

 - 2 77 -

J . J . Ka hn

REFERENCES

(1] A. ANDREOTTI, and C.D. HI LL, several articles to appear

in Ann. Scuola Nor m. Sup. Pisa.

[2J BOCHNER, S. "Analytic and meromorphic continuation by

means of Green's formula," Ann. Math. (2) 44, 652~673

(1943).

[31. EHRENPREIS, L. "A new proof and extens ion of Hartog's

theorem," BulL Amer. Math. Soc. 67, 5007-509 (1961).

(41 FOLLAND, G.B. and KOHN, J. J. "The Neumann problem for

the Cauchy-Riemann complex," Ann. of Math. Study Vol.
75, Princeton Univ. Press, 1972.

[ 5 J FOLLAND, G.B. and STEIN, E.M. "Parametrices and esti-

mates for the a b
on strongly pseudo-convex boundaries,"
Bull. Amer. Math. Soc. (to appear).

[61 GAFFNEY, f·r~P , , "H ll be r t space methods in the theory of

harmonic integrals," trans. Amer. Math. Soc. 78(1955)
426 - 444.

[7] GRA{JERT, H,0 "Bemerkenswerte pseudo.konvese Manni,dAl-

-t-
;tigkei ten," Math. Z. 81 (1963) ~ 377 - 391.

[al ______ and LIES I., "Das Ramirezsche Integral und die
Gleichung af = im Bereich der beschrankten Formen,
Rice University St~dies, (to appear).
 -278-

J . J . Kahn

{9] HENKIN,G"Integral representations of holomorphic

functions in strongly pseudocomvex domains and certain
applications," Mat. Sbornik 78 (120): 4(1969), 611-632
(Russian), English translation in Math. of the U.S.S.R.
April 1969, 7(4), 597-616.

(10J HOl1l'1ANDER, L., "Estimates and existence theorems for

the i operator,"Acta Math. 113 (1965),89 - 152.

[11] , "An introduction to complex analysis in sev-

eral variables, Van Nostrand, Princeton, 1966.

(12) , and 'wERI'lER, J., "Uniform approximation on

compact subsets in n", l1ath. scand , 23 (1968), 5-21.

(1:;] KERZMAN, N., "Holder and LP estimates for solutions of

au =f in strongly pseudoconvex domains, Comm. Pure
APpl. Math. 24(1971), 301 - 379.

[14] KOHN, J.J. "Harmonic integrals on strongly pseudo-

convex manifolds," I, Ann. of Math. 78(1963),112-148;
ibid. 79 (1964), 450-472.

, "Boundarles of complex manifolds," Proc. Con-

ference on Complex Manlfolds (Minneapolls), Springer-
Verlag, New York, 1965.

, "Boundary behavior of on weakly pseudo-

convex manlfolds of dimenslon two," J. Diff. Geom.
Vol 6, 523 - 542 (1972).
 -279-

J . J . Kohn

{17] KOHN, J.J., "G<lobal regular1ty for:a on weakly pseudo-

convex man1folds," Trans. Amer. Math. Soc. (to appear).

[18J 1\ ' and NIRENBERG, L. "An algebra of pseudo -

d1fferent1al operators," Comm. Pure Appl. Math. 18
(1965), 269 - )05.

[191 , and NIRENBERG, L., "Non-coerc1ve boundary

value ' pr obl ems , Comm. Pure APPl. Math. 19 (1965)
443 - 492.

[20) .,_ , and NIRENBERG, L., "A',pseudo-convex doma1n not

adm1tt1ng a ho10morph1c support funct10n," Math. Ann.
201, 265 - 268 (197)).

[2ti , and ROSSI, H., "On the extens10n of holomor-

ph1c funct10ns from the boundary of a complex man1fold ,
Ann. of Math. 81 (1965), 451 - 472.

f221 LEWY, H., "On the local charact~r of the solut10ns of

an atyp1cal l1near d1fferent1al equat10n 1n three
var1ables and a related theorem for regular funct10ns
of two complex var1ables," Ann. of Math. 64 (1956),
514 - 522.

[231 , "An example of a smooth 11near part1al d1ffer.

ent1al equat10n w1thout solut1on, Ann. of Math. 66
(1957), 155 - 158.

(24 J LIEB, 1., "E1n Approx1ma tlonssa tz auf streng pseudo-

konvexen Geb1eten, Math. Annalen 184 (1969), 55-60.
 -280-

J . J. Kahn

[25] NEWLANDER, A. and NIRENBERG, L., "Complex analytic

coordinates in almost-complex manifolds," Ann. of Math.
65 (1957), 391 - 404.

[26J NIRENBERG, R., "On the H. Lewy extension phenomenon,"

Trans. Amer. Math. Soc., (to appear).

, and WELLS, R.O., "APproximation theorems on

differentiable sUbmanifolds of a complex manifold,"
Trans. Amer. Math. Soc. 142 (1969), 15 -36.

[281 0VRELID, N., "Integral representation formulas and

LP estimates for the equation -u = f," Math. Scand.
29 (1971), 137 - 160.

[29J RAMIREZ, E., "Ein Divisionsproblem in der komplexen

Analysis mit einer Anwendung auf Randintegraldar-
stellung," Math. Annalen 184 (1970), 172 - 187.

[3 0] KERZMAN, N. ,"The Bergmann kernel function: differen-

tiability at the boundary," Math. Ann. 195 (1972)
149-158.

[311 LAX, P.D. and PHILLIPS, R.S., "Local boundary condi-

tions for dissipative symmetric operators." Comm.
Pure APpl. Math. 13 (1960), 427 - 455.

[32] NIRE~BERG, L., "Lectures on linear partial differ-

ential equations," Mimeographed notes. Courant
Institute (1973).
 CENTRO INTERNAZIONALE MATEMATICO ESTIVO
(C . I. M. E .)

THE MIXED CASE OF THE DIRECT IMAGE

THEOREM AND ITS APPLICATIONS

YUM-TONG SID

Corso tenuto a Bressanone dal 3 al 12 giugno 1973

 THE MIIED CASE OF THE DIRECT IMAGE THEOREM
AND ITS APPLICATIONS

Yum-Tong Siu 1

J o. Introduction

In these lectures we will discuss the so-called mixed

case of the direct image theorem and its applications. The
starting point of the direct image theorem is the following
finiteness theorem.

(0.1) Theorem (Cartan-Serre)· If ~ is a coherent analytic

sheaf on a compact complex space X, then H~(X,7) is
finite-dimensional over ~ for ~ >
•
0 •

The proof is obtained by Schwartz's finiteness theore~

on the perturbation of a surjective operator between Frechet

spaces by a compact operator (see [8, VIII.A.19] and [6]).

In 1960 Grauert proved the following parametrized ver-

sion of the above theorem which is known as the proper Case
of the direct image theorem [6].

(0.2) Theorem (Grauert). If n: X ---> S is a proper

1 PartiallY supported by a National Science Fbundation Grant

and a Sloan Fellowship
 -284-

Y-T .Si u

holomorphic map of complex spaces and '1 is a coherent ana-

R n* 7 of
V
lytic sheaf on X, then the v th direct image
'] under n is coherent on S for v>= 0 •

Grauert's proof uses the power series method. The

idea is to expand a v -dimensional cohomology class in a
power series in the variables of S (after reducing the gen-
eral case to the case where S is an open subset of a number
space). The coefficients in the power series expansion may
not be cocycles, but, by using descending induction on ])
Grauert showed that they can be approximated by cocycles.
Then he used induction on dim S and applied the induction
hypothesis to the approximating cocycles to get the coherence
of the ~th direct image.

About ten years later Knorr [13] and Narasimhan [18]

gave simplified presentations of Grauert's original proof.

Recently Kiehl [10] used nuclear and homotopy opera-

tors and a form of Schwartz's finiteness theorem to obtain a
new proof for an important special case of Grauert's theorem.
Then Forster-Knorr [3] and Kiehl-Verdier [12] succeeded in
obtaining new proofs of Grauert's theorem along such lines.
Their proofs make use of descending induction on JJ , but
does not u~~ induction on dim S. This opens the way to
generalizing Grauert's theorem to relative-analytic spaces
and such generalizations were carried out by Kiehl [11],
Fbrster-Knorr [4] and Houzel [9].
 Y-T . Si u
In 1962 Andreotti-Grauert (1] generalized the theorem

df Cartan-Serre in another direction. They introduced the

concepts of strongly pseudo convexity and pseudo concaVity and
proved finiteness theorems for spaces which are strongly
pseudo convex or pseudoconcave. A function ~ on a complex
space X is said to be strongly p-pseudoconvex if for every
x ~ X there exists an embedding T of an open neighborhood
U of x onto a complex subspace of an open subset G of
,N and there exists a real-valued cf function ~ on G
such tha t f .. yo -r and the N x N hermitian matrix

( dZa21J1
dZ
\
)
has at least N _ P + I . positive eigenvalues at
i j

every point of G (where zl' ••• ' zN are the coordinates

of (N).

(0.3 ) Theorem (Andreotti-Grauert). Suppose X is a complex

space and f: X -> (a*,b* ) C IR is a proper map· Sup-
~ a* < a# < b# < b* such that l' is strongly p-pseudo-
convex on If> b#} and is strongly q-pseudoconvex on

I ~ < a#} and I 'f' ~ b} '" I ~< b}- for b# < b < b* -
Suppose t is a coherent analytic sheaf on X such that
)!
codh? S;' r on IT< a#}- . Then, for p ~ )J < r-q , H (X,"])

is finite-dimensional over (: and

"1'
H (X, "]) ---.-) .• ; I a
)I .)
< 'f < b ] , __ } is an isomorphism for
a* .~ a < a # and b # < b ~ b* -

(For the definition of codh"] , see (A.I) of the Appendix- )

 - 286-

Y - T. Siu

A holomorphic map n: X - > S of complex spaces is

called stronglY (p,q)-pseudoconvex-pseudoconcave if there
exists a if map f: X - > (a*, b*) C 1R. and there exists
a* < a# < b# < b* such that

i) nl (a ~ l' ~ bJ is proper for a* < a < b < b* •

ii) {f ~ b 1 = {SO < b J- for b# < b < b* •

iii) fj' is strongly p-pseudoconvex on

iv) ep is strongly q-pseudoconvex on

We introduce the following notations. Fbr

a* ~ a < b ~ b* '
b
Xa [a < 1 < b J

v b "I
For a coherent analytic sheaf "] on X , R (n a )*T denotes
R)I(n: )*(1I x: ) •
The so-called mixed case of the direct image theorem
is the following parametrized version of the theorem of
Andreotti-Grauert.

(0.4) Con jecture. Suppose n ; X - > S is a strongly

(p,q)-pseudoconvex-pseudoconcave holomorphic map (giv~n 'with

l' and a* < a# < b# < b*) and "l is a coherent analytic
sheaf on X such that dim S ~ n and codh 7 ~ r £E
.
If < a#J. R n* "l
,)
Then, for p ~ v < r -q-n, 1S coherent on
,) vb u-r
S and Rn*"J-> R (n a )* 7 is an isomorphism for
 - 2 87-

Y- T . Si u

This conjecture so far has not been completely proved.

The special case {'f < ~ 1 = Rf is called the pure pseudo con-
vex case. The special case {'f > b# 1 = f5 is called the pure
pseudo concave case. Partial results for these two pure cases
were obtained by Knorr [14] and Siu (24,25]. Recently the
pure pseudo convex case was completely proved by Siegfried
[21] by using the methods of the new proofs of Grauert's the-
orem and the pure pseudo concave case was completely proved by
Ramis-Ruget [19] by using the methods of the new proofs of
Grauert's theorem together with duality. Unfortunately these
methods cannot be applied to the mixed case, because any in-
duction on the dimension of the direct image is impossible.
A partial result on the mixed case was obtained by Siu [26].
m these lectures we will prove the following improved par-
tial result of the mixed case which is good enough for the
known applications.

(0.5) Main Theorem. Suppose n: X --> S is a strongly

(p,q)-pseudoconvex-pseudoconcave holomorphic map (given with
If and ~ < a# < b# < b*) and 7 is a coherent analytic
sheaf on X such that dim S ~ n and codh 1 ~ r .Q.!1

{'f<~l . Suppose codh(£)

.
'1? n
S,n(x) x
for x<;:X. Let
,
a* < a < a < a# and Then the following
conclusions hold.

i) (Rv(n~)*7)s is finitely generated over (OS,s for

 -28 8-

Y .-T . Siu

s ~ S and P:i- v < r-q-n •

v
ii)
,J
R n*o1
~
-> R (n a
b
)*7 is an isomorphism for
p < ,; < r-q-n •
b'
iii) RP(n , ) * '1
a
-> RP (n b)
a *'1 is an isomorphism.
-
iv) p < r-q-2n ,~ R'" (n~ )*1 is coherent on s
for P:i-)) < r-q -n-l •

Fbr the applications, only conclusions i) and iii) of

the Main Theorem are needed. The Main Theorem will be proved
by the power series method. If we couple the power series
method with the methods of new proofs of Grauert's theorem
and duality, we can improve conclusions iii) and iV) to the
following, but we will not discuss it in these lectures.

(0.6) Theorem- Under the assumptions of (0.5)~ if

lIn*
p < r-q-2n , then, for p ~]1 < r-q-n , R '1 is coherent on
S and R,) n* 7 --> R'" (n~ )/1 is an isomorphism .f or
a* ~ a < a# and b# < b ~ b* •

The Main Theorem will be applied to the following:

i) extending coherent analytic sheaves

11) blowing down strongly I-pseudo convex maps
iii) blowing down relative exceptional sets
iV) obtaining a criterion for the projectivity of a map
v) extending families of complex spaces.

(For applications ii), iii) and iv), the pure pseudo convex
case of the direct image theorem suffices.)
 -289-

Y-T .~u

Fbr coherent sheaf extension, we will not obtain the

best known result of extension from Hartogs' figures [23]~

~e will only obtain the result of extension from ring domains

(22] (which implies the extension across subvarieties
[29,5]). The proof of coherent sheaf extension by means of
the direct image theorem is not the simplest approach. A
very simple proof of the extension from Hartogs' figures was
given in [27] which does not use the power series method of
Grauert and does not use the method of privileged sets of
Douady.

The smooth case of the local result on blowing down

strongly I-pseudoconvex maps was obtained by Markoe-Rossi
[17] and the general case of the complete result waS obtained
in (25J. The results on relative exceptional sets and pro-
jectivity criterion were due to Knorr-Schneider [15]. The
special case of the result on extending families of complex
spaces where the parameter space is a single point was
obtained by Rossi (20) and the general case was due to Ling [16l

Now we give here a brief sketch of the main ideas of

the proof of the Main Theorem. In the actual proof, for
technical reasons, we use sheaf systems to construct com-
plexes of Banach bundles to calculate the direct image
sheaves, but, here in this sketch, for simplicity, we compro-
mise the accuracy by calculating the direct image sheaves by
the usual ~ech coch.in complex. In this sketch there are
also other compromises of accuracy in some minor points for
 -2M-

Y - T . Siu

the sake of simplicity. The proof of the Main Theorem has

three key steps.

The first key step is the existence of privileged sets

for a coherent sheaf, i .e. if s. nt9P --> l)q is a sheaf-
homomorphism on an open neighborhood G of 0 in ([n which
is part of a finite resolution of the, given sheaf (wh~re nO
is the structure sheaf of C n), then there exists an open
polydisc neighborhood P of 0 in G satisfying the fol-
lowing. If s ~ r(p, noq) . i s bounded (in a suitable
sense) and the germ of s at 0 belongs to the image of e , then
s is the image of a bounded section v of over P

and v can be so chosen that s ~> v is continuous

C-linear. The existence of privileged sets can be proved in
three ways (some of which are valid only for certain senses
of boundedness). The first proof by Cartan uses the Weier-
strass preparation and division theorems and it is usually
used in the proof of the Closure-of-M:>dules Theorem [$, II. n].
The second proof by Grauert uses power series expansion and
it is used in Grauert's original proof of the proper case of
the direct image theorem. The third proof by Douady uses
holomorphic Banach bundles and it is used in his solution of
the module problem [2]. In these lectures, we present Dou-
ady's proof, because it works for all senses of boundedness
needed for our purpose and because the idea of Grauert's
proof occurs in the second key step of the proof of the Main
'lheorem and we will see it there 8.I\YWay. '!he existence of priTi-
leged sets gives rise to 'lheorem B with bounds and Leray's
 -21]1-

Y-T . Siu
theorem with bounds, which are used, together with the bump-
ing techniques of Andreotti-Grauert [1], to construct com-
plexes of Banach bundles to calculate the direct image
sheaves.

The second key step is the analog of the Hauptlemma .

of Grauert's original proof of the proper Case. Suppose S
is an open neighborhood of 0 in 4: n. Let A (f) be the
open polydisc in «n centered at 0 with polyradius f •
Roughly the analog of the Hauptlemma can be described as
follows. SUppose 'l.t '" {U i} , l{ '" {Vj } be suitable collec-
tions of Stein open subsets of X such that each Vj is
relatively compact in some Ui • Let

rUin n- l ( .6 (p») }

{V n n- l (A(p»)} •
j

Then there exist

for some fO satisfying the following. Fbr p sufficiently

small (in a suitable sense), every l, ~ l(?Jl(f} ' "1) can be
written as

t, '" ~ a l,(i) + 67
i'"l i

when restricted to ~(f) , where

 -292-

Y-T . Siu

ai G: r(A(p}, nf!))

? G: c"-l(\(f}, 7) •

1tt>reover the bounds (in a sui table sense) of ai and 7 are

dominated by a constant times the bound of ~ • we will look
upon this as a generalization of the existence of privileged
sets. Instead of lifting bounded sections in the map

induced by 8 , in this case we consider the lifting of

bounded sections in the map

defined by

-> ~ a. l,(i} + 6?
i=l ~

(restricted to l(P}) •

It turns out that the ideas of Grauert's proof of the exis-

tence of privileged sets can be generalized to this case when
we apply Leray's theorem with bounds. For this we have to
use induction on dim S, but we do not need - any descending in-
duction on -V. The reason why we Can avoid this descending
induction 'on v which is so essential in Grauert' s original
proof of the proper case is that we look upon the analog of
the Hauptlemma as the generalization of tpe existence of
privileged sets and we use both the surjectivity and the in-
 - 29 3-

Y- T . Siu

jectivity statements of Leray's theorem with bounds, whereas

Grauert did not use the technique for proving the existence
of privileged sets in his proof of the Hauptlemma and he
used only the surjectivity statement of Leray1s theorem with
bounds. Our approach is simpler and gives the best possible
result in the finite generation of the stalks of the direct
image sheaves.

After the above two key steps there is still one ob-
stacle to proving the coherence of the direct image sheaves.
Suppose S is an open neighborhood of 0 in C n• To get
the coherence of the direct image sheaves by induction on n,
we need the following statement on global isomorphism. There
exists an open polydisc neighborhood U of 0 in S such
that

is an isomorphism for all sheaves -§

of the form "l/(t n-c)m7,
where t••• , t n are the coordinates of (( n. The third
l,
key step is to obtain this isomorphism statement. Suppose
U:a lUi} is a Stein open covering of X. Fbr any open sub-
set Dof S let (,t(D) :a lUi n n-I(D)}. The presheaves
v
D 1--> B (1t ( D) , ~ )

D 1--> eV ( U(D) ,-9)

)/ oJ
define sheaves 13 (§) , ~ (9) on s . We derive the isomor-
phism statement by constructing a sheaf-homomorphism
 - 294-

Y . T . Siu

-e (7 )
V
- > ~v-l(7) on U which is a right inverse of s »

This right inverse gives rise to a right inverse

13"(9) _>fb l1- l C§ ) of () for sheaves .g of the form
7j'ctn-c)mj. The existence of a right inverse
f>v(§) _>~"-l(§) of () implies the vanishing of
~(U, ~)IC~)) which, together with t.he coherence of R"Vn* -g ,
v+l .P. _U-l .P.
R n*~, ••• , K n*~, yields the isomorphism

The construction of a right inverse S :11("]) - > ~v-l (7) of

() is based on the generalization of the following observa-
tion. For an open polydisc G in a: n , a continuous lC-lin-
ear map

is induced by a sheaf-homomorphism on G if

and only if ~ is linear over the polynomial ring

~[tl' ••• , tnJ· We show that, under the additional assump-
)1+1 )I+n_l "1
tion of the vanishing of (R n* '1 )0' ••• , (R n*)O'
the lifting

in the analog of Grauert's Hauptlemma can be done in such a

way that it is linear over the polynomial ring
([t l, ••• , tnJ. When we have the finite generation of
v
(R n* "])0' ••• , (R
l1+n CO
n* trr~)O over no' we can apply the
above argument to a complex of the form '& h (']) @ n([)
Ph
in-
 -29 5 -

Y. T . Siu

stead of to ~A('1) and obtain the isomorphism

for ));;; p. :i- v+ n. It is the third key step that makes the
additional assumption of p < r - q - 2n necessary, because
we need some room to get a right inverse of 6. It is also
the arguments of this step that necessitate the introduction
of complexes of Banach bundles for the calculation of the
direct image sheaves, although such an introduction stream-
lines the presentation elsewhere as well.

In these lectures some tedious details, which are ob-

vious and can easily be filled in, are left out, especially
in § 3, § 4, and § 5 • Deta Us of this nature can be found in
[13, 18, 24]. There is an appendix at the end which deals
with homological codimension, flatness, and gap-sheaves.
Consult the appendix when these concepts are mentioned or
their properties are used.

Now we list the notations we will use in these

lectures.

~ the set of all positive integers

~o N U IO}

~* ~ V ICll}

IR+ the set of all positive numbers

nfD the structure sheaf of ([n •

 -296-

Y- T .Siu

The components of a c;: q:m are denoted by a ••• , am.

l,
m
For a, b ~ IR ,by a < b we mean a i < b i for 1 ~ i ~ m
and by a ~ b we mean ai ~ bi for 1 ~ i ~ m •

n occupies a special position in these lectures. The

coordinates of ([ n are always denoted by
([ t] denotes the polynomial ring a: [tl, ••• J t
n
].

to r: II"'n
'=. \L. ,p r:
'=.
fRn
+ ,and ~ ~ v IlJn •
in 1'rI0

.D.(tO , p) the open polydisc in a: n with

center to and po1yradius e
(tpt
O
r .. o lJ o lJ
(tl-tl ) 1 ••• (tn-tn)n
PI Pn
11> I v1 + ••• + ))
n

II )1
dt 1 ()t n
1 n

When to .. 0 , we denote .t d to, f) simply by Ll (p) •

When pa (1, ---,1) ,we denote .6(P) simply by .6.. In

general, for b ~ IR~ , ~N(b) denotes the open polydisc in

(N with center 0 and po1yradius b. Fbr 0 ~ a < b in
IR.N ,

GN(a,b) .. {z ~ ,6N(b) Ilzil > ai for some 1 ~ i ~ N} •

 - 29 7-

Y - T . Siu

The closure of a set G is denoted by G- /loU

G
denotes the sup norm on G • If G is an open subset of
«N , r 2.(G, JD) denotes the set of all L
2
holomorphic
L
functions on G.

The stalk of a sheaf ~ at a point s is denoted by

7s ' rr U is an open neighborhood of sand f ~ nU, "]) ,
then fs denotes the germ of f at s.

A complex space may have nonzero nilpotent elements in

its structure sheaf. The structure sheaf of a complex space
X is denoted by Ox • For x~X , ...... X,x means the maximum
ideal of the local ring oX,x and sometimes (when no confu-
sion can arise) it also means the ideal sheaf for the subva-
riety [x] of X. Suppose n: X --> S is a holomorphic
map of complex spaces and s ~ S.
Then MN S,s means also
the ideal sheaf on X generated by the inverse image of

S,s when no confusion can arise • For a coherent analytic

n.
..,.,."

sheaf "] on X, R t denotes the yth direct image of

V

under n. Ii.' Y is an open subset of X and CT' = nlY ,

Il
then R" cr. (71 Y) is simply denoted by R (1""* "].

Suppose U· lUi J and )( = (Vj J are collections of

open subsets of a complex space X and? is a coherent ana-
lytic sheaf on X. Vl«)( means that each Ui is rela-
tively compact in some VT(i)' For every t, ~ eliot, '1)
we can define T *l, ~ ev nt, "J) by means of the index map "t •

•* l, is also denoted by C, I U. For

V
l" 7 ~ e (1(', "l) , we say
that t; = 7 on ?il if T*l," T*? IViI denotes Vi Ui. For
 -298-

a complex space Y, Y x 7Jl. deno tes {Y x lIt} •

When we have a sheaf-homomorphism e : '1 --> -g of

analytic sheaves on a complex space X, we sometimes use the
same symbol e to deno te also the maps

r (X, "] ) -> r (X, .q )

. 7/J7-> -g/J§
(where jr is an ideal sheaf on X) and other similar maps
induced by e when no confusion can arise.
 - 2.99-

Y .-T . Siu

Table of Contents

§o Introduction 1

Part I Construction of Complexes of Banach Bundles 18

h Privileged Polydiscs 18
§2 Semi-norms on Unreduced Spaces 31
§3 Theorem B with Bounds 40
§4 Leray t s Theorem wi th Bounds 48
§5 Extension of Cohomology Classes 55
§6 Sheaf Systems 65

Part II The Power Series Method 75

§7 Finite Generation with fuunds 75
§a Right Inverses of Coboundary Maps 107
§9 Global Isomorphism 111
§ 10 Proof of Coherence 119

Part III Applications 135

§ll Coherent Sheaf Extension 135
912 Blow-downs 141
§13 Relative Exceptional Sets 155
§14 Projectivity Criterion 157
§15 Extension of Complex Spaces 160

Appendix 164

References 178
 - 30 0 -

Y-T . Siu

PART I <XlNSTRUCTION OF <XlMPLEXES OF BANACH BUNDLES

§l Privileged Polydiscs

O
(1.1) Suppose D· A(t ,P) and Eo is a Banach space.
Define B(D,EQ) as one of the following:

i) the set of all EO-vaJ,ued uniformly bounded hoIonor--

phic functions on D,

ii) the set of all EO-valued uniformly continuous holo-

morphic functions on D,

iii) the set of all EO-valued holomorphic functions

on D with

where 1I 0ll
is the norm of EO
Eb
In any of these three cases B(D,E ) is a Banach space.
o
B(D,~) is simply denoted by B(D). If U is an open
neighborhood of n- and E is the trivial bundle on U
with fiber EO' we denote B(D,E o) also by B(D,E). Sup-

pose '1 is a coherent analytic sheaf on an open neighborhood

There exists an exact sequence
 - 30 1-

Y - T . Siu

m~
0-> n1:1 m~
-> ... -> n1:1 -> nVYO ->7-> 0

on an open neighborhood of DI.

Definition. D is an 7 -privileged neighborhood i f

(a) the induced sequence

p P
o _> B(D) m _> B(D) 0

is split exact,

(b) Coker CL --> "lois injective.

t
When (a) is satisfied, one defines B(D,']) as CokerCL.

This privilegedness is said to be in the sense of

Cartan, Douady, or Grauert according as B(D) has the mean-
ing of i), ii), or iii).

The definition of privilegedness and B(D, 7) is

independent of the choice of the resolution of 7, because,
by using Theorem B of Cartan-Oka, we can easily prove that
any two finite free resolution of 7 on a Stein open neigh-
borhood of n- become isomorphic finite free resolutions
after we apply to each of them a finite number of modifica-
tions [8, Def. VI.F.I], i.e. after we apply to each of
them a finite number of times the process of replacing it by
its direct sum with some finite free resolution of the zero
sheaf which has only two nonzero terms ( of. [8 , p.202,
 - 30 2-

Yo-To Si u

(1.2) For Banach spacesEO' FO we denote by L(EO' FO}

the Banach space of all continuous linear maps from EO to
F •
O
Suppose S is an open subset of ([ nand E is a
holomorphic Banach bundle on S with fiber EO. For s ~ S
we denote by Es the fiber of E at s . For any open sub-
set U of S we denote by EI U the restrictio n of E to
U•

Suppose F is a holomorphic Banach bundle on S with

fiber FO• A map '(: E --> F is called a bundle-homomor-
phism if for every open subset U of S for which there are
trivializations

a: El U ~ > U x EO

there exists a holomorphic map A(' ) from U to L( EO,FO}

such t hat

for s~U and x ~ EO • For any open subset U of S we

denote by tlu the bundle-homomorphism

El u - > Flu

induced by '( . For s ~ S we denote by 'Is the map

 - 30 3-

Y-T . Siu
induced by 't •

Suppose
e
o _> E(m) _> E(m-l) _' _> •.• _> E(O)

is a complex of bundle-h9momorphisms of holomorphic Banach

bundles on S. If for some So ~ S the sequence

-> . .. _>

is split exact, then there exists an open neighborhood U of

So in S such that the sequence

is split exact.

To prove this, it suffices to prove the case where

m· land E(l), E(O) are both trivial bundles. Let H be

the closed subspace of which complements

Let

tr : E(l) ~ (S X H) -> E(O)

be the bundle-homomorphism induced by e and the inclusion

map

S x H '--> S x (H EB Im es) = E(
0)
•
o

(j is an isomorphism. Since the invertible elements of

So
 -304-

Y -T . Si u

L(E~l)(f)H, E~O)) f'orm an open subset, there exists an open

o 0
neighborhood U of' So such that a: I U is a bundle-isomor-
phism (i.e. (~IU)-l is a bundle-homomorphism).

(1.3) Suppose S (respectively Q ) is an open subset of'

{.n (respectively ([, N) and 7 is a coherent analytic sheaf" on
S x .Q. For

we denote by 7(s) the sheaf'

where ••• , t
t are the coordinates of' {n. 7(s) can
l, n
be regarded in a natural way as a sheaf' on n.
For p ~ 1 , we denote by B(S2. 'n+P) the trivial
bundle on ([ n whose f'ib'er is B(.Q)P.

Let n: S dl -> S be the natural projection. Sup-

pose s~ S and '1 is n-f'lat at {s}x.Q and '1 admits a

f'inite f'ree resolution

on S x.Q. Suppose z ~.Q and G C C.Q is an open polydisc

centered at z which is an '1(s)-privileged neighborhood of'

z. Then there exists an open neighborhood U of' s in S
such that, f'or any open polydisc DC U centered at s,
 - 30 5 -

Y - T . Siu

D x G is an 1-privileged neighborhood of (s , z) •

To prove this, consider the following sequence of bun-

dle-homomorphisms induced by (*):

Since G is 1(s)-priyileged and since by the n-flatness of

"1 at {s} x Q the sequence

I
o -> (rlm(s) -> ... ->
n+N'" n+NI:J
((ll
(s) -> ":1
.1(s) -> 0

induced by (*) is exact, we conclude that the sequence (#),

when restricted to the singleton {s}, is split exact. By
(1.2), on some open neighborhood of s in S, (#) is split
exact and Coker a is a holomorphic Banach bundle. Observe
that, for any bounded open polydisc D centered at s,

B(D, B(G'n+NvPi)) is naturally topologically isomorphic to

B(D)( G'n+Jl i ) (0 ~ i :;:. m). Hence, when D is contained

in a sufficiently small neighborhood U of s in S,

...

is split exact and B(D XG, "J) is topologically isomorphic

to B(D, Coker a). To show the injectivity of
 -- 30 6-

Y -T.Siu

B(DlCG,'1) -> '1(S,Z) ,

it suffices to show the injectivity of

(where V(Coker~) is the sheaf of germs of holomorphic sec-

tions of the bundle Coker~) and, by induction on n, it
suffices to do the case n ~ 1. Take

f ~ Ker ~ •

Suppose f is nonzero. There exists a maximum nonnegative

integer k such that

with

By the n-flatness of "J,

g ~ Ker ~ •

Since G is an F(s)-privileged neighborhood of z, it fol-

lows that g(s) ~ 0 , contradicting the maximality of k •

(1.4) Suppose 7 is a coherent analytic sheaf on an open

subset S2. of a:;N. IdentHy(N with the subset 0 x (N of
{N+l and extend ' 7 trivially to a coherent analytic sheaf
,..,
'1 on ~ x .Q. If GCC.Q is an open polydisc centered at
z and if G is an '1 -privileged neighborhood of z , ~,
 - 307 -

Y:-T . Siu
for any bounded open disc DC ([ centered at 0 , D x G is
""
.2.!! "I-privileged neighborhood of (0, z) •

Tb prove this, we can assume without loss of general-

ity that there exists an exact sequence

on Q. Define
""
Q.
O _> (0Pm (9
~l ~l
(QPm-l m>
-

by
N
Q. == (Q.l,Wj
l

C
(_ljJ-'w )
Q.. (1 < j ~ m)
J
Q.j-l

where w is the coordinate of c an element of N+lJP is

represented as a column vector, and Q.. is regarded as a
J
matrix of holomorphic functions which are considered as func-
tions on ([ x.Q independent of w • The sequence is exact on
(( x Q. Let
p.
-> B(G, lJ J) (1 ~ j ~ m)

be a continuous linear map, which, when composed with the map

 - 308-

Y-T .Siu

P 'l
B (G, N(D J- )

induced by a. , gives rise to a projection

J

A corresponding

(where p
-1
= 0) can be given by

,
where a is independent of t and one denotes also by ~j

the map

it induces. Hence

""
B( D )( G, "1) ~ B( G, ., )

and the result follows.

(1.5) Suppose D is a bounded 0 pen polydisc in ([n cen-

tered at to and suppose

0 ·-> I ' ->~->-

17"7 ":r"
-> 0

is an exact sequence of coherent analytic sheaves on ' an open

,
neighborhood of D-. If D is an 7 -privileged and 7 "-
privileged neighborhood of t o ,then D is an '1 -privileged
 Y - T. Siu

neighborhood of to.

Tb prove this, we take finite free resolutions on n-:

, a
, ,
P m PI' a I Po'
0-> nI[)m_> -> n
->
(9
n
f!) -) '1 - ) 0
I

" a "m " a" "

0-> n
(!)
Pm
-> /API
-> nI;J ->
I
n
//,\PO
I;J ->1 ->
"
°.
We can construct the following commutative diagram

o 000
J, ~ , J, t
,,,Pm PI
0-> nI;J -> -> nV -> n~lo -> 7' -> 0

~lf\P m
m ->
a t
tf\PI
a
1
~
,,,PO
1
tI7
0-> nv -> nI;J -> nI;J -> iT -> 0

~ n
,,,Pm _ ...
1" P
l P
n I
~
O -> nI;J v -> n(!) I -> nf!) 0 -> "1" _> 0
~
o
1 01
10 0
where
, II

i) v",Pj = ,I\Pj
v w mPj
l!'I
~ and, except in the last column,the
n n n
vertical maps are the natural injections and projec-
tions. (a~
11) a j is of the form oJ

being a sheaf-homomorphism).
 - 310-

Y-T . Siu
Let
t
p.
B(D, nlD J)

be a continuous linear map which, when composed with the map

, t
,.."
( p.
a. .: B D, nl[) J) -> B(D, n(!lj-l)
J

induced by
t
a..
J
, gives rise to a projection

Let
•
~~: B(D, n(~lj-l) ->

be a similar map. Then a corresponding

can be defined by

Since clearly

o -> B(D, '7

, ) -> B(D,7) -> " ->
B(D,'1 ) 0

is exact, the result follows.

Before we state the principal theorem on privilegad

polydiscs, we have to introduce a terminology. Suppose Sf
is a statement depending on f (; IR~. We say that Sf holds
for f sufficiently strictly small, if there exist
 - 3 1 1-

Y - T. Siu

(.,)1 (; R+ and positive-valued functions (.0i (fl' ••• , Pi-l) on

1R +i-l (1 < ].
. ~ n) sue h t h at Sr holds for f satisfying

(1 < i ~ n) •

(1.6) Theorem. Suppose '7 is a coherent analytic sheaf on

an open neighborhood U of 0 in «. n. Then, for f suf-
ficiently strictly small, ~(p) is an 7-privileged neigh-
borhood of O.

Proof. Use induction on n. By shrinking U , we can

assume that there exists a nonnegative integer d such that
t n "i s not a zero-divisor of t~ 70 for to (; U Then
t
~: .. t~7iS n-flat at U n {tn .. OJ , where nO: ([n -> tC
is the projection onto the last coordinate. By induction

hypothesis, when (Pl, ••• ,Pn-l) is sUffici~ntly strictly small,

the polydisc GCa:: n- 1 with polyradius (f , ••• , fn-l) and l
centered at 0 is relatively compact in U n {t n .. OJ and
is a privileged neighborhood of 0 for the coherent analytic
sheaf

on U n {t n .. OJ. By (1.3), when an open disc DC 4: cen-

tered at 0 is sufficientiy small, D x Gee U and D x G is
a ~ -privileged neighborhood of O. By using (1.4) and
 - 3.12 -

Y. T . Siu

applying (1.5) to the exact sequences

0-> (1 ~ j < d)

0-> -> 7-

we conclude that D x G. is an t -privileged neighborhood of

 -313-

Y-T. Siu

§2 Semi-norms on Unreduced Spaces

(2.1) Lemma. Suppose ~ is a coherent analytic sheaf on a

complex space X and Dee x is an open set. Then there
exists a nonnegative integer p such that, i f fG: r(U,"])
for some open subset U of D and f for
xG:......",P+17x
X,X
x G: U , then f = 0 •

Proof. Take :x;..G:

-u
X. Let n
0 = i Qi be the primary decompo-

si tion of the zero submodule of 7x . Let ki be the di-

o
mension of the radical of Qi· Qi is the stalk at Xo
of a coherent analytic subsheaf ~i of ~ defined on some
open neighborhood of x O• Then

dimx Supp o/~ ~ ki •

i

((~i) [k i -1] '7)x • (!2i)X

for x = Xo (see (A.7) of the Appendix). Hence these two

conditions hold for x in some open neighborhood of xo •
Since we need only prove the lemma for each 7'/~.' we can
~

assume without loss of generality that X has pure dimension

k and

We can also assume the following:

i) X is a subspace of ~.
 -314-

Y-T.Siu

ii) The projection ~: ~ --> D onto the first k

coordinates makes X an analytic cover over D.

iii) The reduction of X is defined by the ideal sheaf J1

generated by holomorphic functions gk+l' ••• , gn
on A.

iv) The unreduced structure of X is defined by J' .t .

We are going to prove that it suffices to set p = L. Let
n = ~Ix. Let S be the sUbvariety of X outside which
t l, ••• , t k, gk+l' ••• , gn form a local coordinates system.
Let T be the set where ROn* 7 is not locally free. Since

it suffices to prove that f = 0 on U - n-l(TUn(S)). There

f
x
b.1Wn~+1
-Jl.,X
7x

means that

1; t
la.l=.! 1
0. .
1 ••• ...

Since
a.
g n = 0
n

= ~k = 0, one has
 -- 3 15-

Y. T. Siu

fX '"

in '7x . Let W be an open neighborhood of n(x) in D

,
such that n-1(W) nu is closed in n-1(W) . Let s be the
element of r(n-l(W},"l) which equa~s f on n-l(w)nu
and equals 0 on n- 1 (W) - U • The element f* of
,
r (W, ROn* 7-) corresponding to t satisfies

(y ~ W) •

It follows that f* '" 0 on w- T • Q. E. D.

(2.2) SUppose V is a subvariety of an open subset G of

(n and p is a nonnegative integer. Define Jrv(p) as the
sheaf of germs of holomorphic functions on G whose partial
derivatives of order ~ p vanish identically on V. Then
JV(p) is a coherent ideal sheaf on G.

We prove by induction on p. The case p '" 0 is

the theorem of Cartan-Oka. We can assume without loss of
generality that

for some holomorphic functions f l, ••• , f k on G. Let i

be the number of n-tuples of nonnegative integers a with
lal '" p. Define
 -316 -

Y-T .Siu

by the matrix

and define

by the row vector

Let
(OJ.
n

be the quotient map. Then

Fbr, every g ~ J'V(P-l)x Can be written as

g '"

and, for 1~1 p,

k
Ifg '" E a. (Iff.)
~ x •
i"'l a

It follows that g ~J1v(P)x if and only if, when restricted

to V,
k ~
E a. (D f.) '" 0 (I~I = p) •
i"'l ~ ~ x
 - 317-

Y-T .Siu

(2.3) Suppose V is a subvariety of an open subset G of

n
~ ,p is a nonnegative integer, and

r (V ,
n
C9 /.1 (p))
'I" V
•

Suppose L is an open subset of U. We define a semi-norm

P II f ilL as follows. For x ~ U , there exists

IV
for some open neighborhood n of x such that f induces
rlno v . Let

(leLl ~ p) •

Then

{feL (x) }
/eLl ~ p
N

is independent of the choice of f. Define

p [r II L = sUP{/feL(X)llleL/ ~ p, x~ L} •

The semi-norms pll' ilL 2!! r'(v, l)/lv{ p ) ) define a Frechet

space structure when L runs through all relatively compact
open subsets of V •

We prove this by induction on p • Let {f]I } be a

Cauchy sequence in rtv. nl9/fy(p)) . It suffices to show
that every point of V admits an open neighborhood U in V

such that {f~IU} converges to some element of

r(U, nl9/JV{p)). We can assume that G is Stein and
 Y - T. Si u

Jy(p-l)

for some holomorphic functions sl' 000, sk on G 0 Con-

sider the following commutative diagram of quotient maps

Y\ /7
n'!J!Jy(P-l)

By induction hypothesis, 7(f~) converges to some element

f' of reV, nV/<1y(P-l)) 0 By applying the open mapping
theorem to the map

induced by y, we can find

such that

,
f

Since it suffices to show that fv - ~(gv) converges in

r(y, n~;I5V(p)) ,we can assume without loss of generality
that ?(fv) = 0 0 For some holomorphic functions aVi on G
 - 31 9-

Y -T . Siu

From the definition of plH L ' i t follows that, for lal = p,

is a Cauchy sequence in ]) • By the Closure-of-Modules

T heo rem [ 8 , II •D. 3 ] there exist holomorphic functions ai
on G such that, for /(11 = p ,

converges to

k
as ]) --> (]). Then converges to ep(Ea ~s.) in
i=l ~ ~

(2.4 ) Suppc ~e X is a complex space. Define the reduction

order of X as the smallest nonnegative integer PO such
that, if

for some open subset U of X and

 -320-

Y-T . Siu

for all x ~ U , then f · O. A complex space is reduced if

and only if its reduction order is zero. By (2.1) a rela-
tively compact open subset of a complex space has finite re-
duction order.

SUppose the reduction order of X is ~ P< 00. Let U

be a relatively compact open subset of a Stein open subset '"U
of X. Fbr an element f of r (U, taX) define IIfllU as
follows. Imbed ""U as a subspace Y of an open subset G
,
of (N by a holomorphic map ~ • Let f be the element of
r(~(U), lOy) corresponding to f. Define
,
Ilfl~ .. inflpl/r*II!(u) /f* ~ r(~(U), nV/5y(p)) induces f } •

Two different choices of U or its embedding give rise to

equivalent semi-norms. (However, semi-norms defined by dif-
ferent p's may not be equfva Lentv ) 'When we write II·"u'
we assume that a fixed U and a fixed ~ (and a fixed p)
are chosen. Let rb(U, lOX) denote the set of all
f ~ r'(u, (9X) with IIfllu < 00 •

SUppose t ~ r( ~(tO 'f) )( u, CD n ). Let

« xX

be its power series expansion in t. The Grauert norm

Ilfll 0 of f is defined as sup 11r11 11 u • Suppose
U,t ,p 11

~ .. lUi} is a collection of open subsets of X such that

Ui is relatively compact in some Stein open subset of X.
 - 3 21 -

Y . T . Siu

For

l,={t,. i} ~ eft (A(tO,olX 7Jl, I{) )

~O··· k ' 4 : n )( X

define

11E.11•• °
VI, t, P
=
i
O
sup
' ••• ,i
k
/lSi i
0· •• k
°
II U. (\ ••• nU. ,t 'f
~O ~k

When n = ° J II E, 1/
'tk, t
°'f is simply denoted by II~lIVl . Let

~ (ll, (OX l denote the set of all E, ~ eft (a, 0x l with

I!E,II < CXl •
VL

For a holomorphic function g on A(tO, pl with

power series expansion

define
 ~ 322 -

Y - T . Siu

u. Theorem B wi th Bounds

C3.1) Lemma. Suppose X is a complex subspace of an open

subset Q of lC N and Gl C C G2 C C.Q are open subsets
wi th ~ Stein. Let Hi a Xn Gi Ci .. 1,2). Then there
exists a continuous map

1J' : r (ACtO, f) x ~, ID.

([
n
)( X
) -> r (ACtO, f) x Gl, n+ NID )

linear over ([ t] such that

i) e"f is the restriction map from

is the quotient map.

11)
1l'1C~) 1IGl' to 'f' :; c 11E,11~ ,to 'f' where C is a constant

independent of ~, to, and f.

Proof. By considering the power series expansion, we observe
that, it suffices to prove the special case where n" 0 •

G"
, If
Take open subsets Gl C C Gee G C Gz with
, , If If
Stein. Let H .. X(\G and H .. XnG • Take
t,G;: r C~, <ax) • There exists ?G;: fCG' , (0x) such that

where C is independent of ~ and comes from applying the

open mapping theorem to
 - 32 3-

Y -T. Siu

,
Let? be the projection of ? onto the orthogonal comple-
ment of the kernel of
, ,
r 2 (G , N(f) -> r(H , (9X) •
L

Define 'Y'(~) = ? ' . Then 311 (s) is independent of the choice

of 7 and satisfies the required conditions.

(3·2 ) Lemma- Suppose X is a Stein complex space and

Ul C C U2 C c
x are open subsets with U2 Stein. Then the
restriction map r : r b (U2, <OX) -> rb(U l, (0x) factors
through a Hilbert space.

Proof. U2 can be identified with a complex subspace of a

Stein 0 pen subset Q of some ce N. Take open subsets

U1 CGl C C ~ C C ~ CeQ with ~ Stein. Let

~= U n
2 ~. From (3.1) we obtain a continuous linear map

such that the composite map of

is r. Hence r factors through the Hilbert space

 - 32 4-

Y -T . Siu

(3·3) ~. Suppose X is a Stein complex space and

Gl C C G:2 C C X are open subsets with G:2 Stein. Let

Vi,,· {uJaI) j be a finite Stein open covering of GlJ ('lJ" 1,2)

with UI1) C C UJ2). Then for k ~ 1 there exists a conti-

nuous map

linear over ct t] such that bf the restriction

~ag:g;:,.,r.:;.ee=..;s;;.,.,;.w;.;:i;..::t~h~:.:..=..~;...::;.:===

map and II T(E,) 11 , 0 ~ CIlE,1! . 0 ~ C is a con-

7-1.,t ,p U2 ,t 'f
stant independent of c; , to, ~ f .
Proof. By considering the power series expansion in t , we
observe that it suffices to prove the special case where
n • 0 •

Take Stein open subsets Gl C C Gee

t
Gee
"
G:2 •
, t
Take a finite Stein open covering 1t .. lUi] (respectively
VLI• t
{Ui])" ,
of G (respectively
"
G) such that

UJ1) CC U~ CC U~ CC uI2) •
Take ~ (; z~tt2' <OX) • There exists ~ (; rf - 1 (Vl,' (0X)

such that

°7 • s; Ill'
1/7/1 vl .< c I E. Ilvt "
where C is independent of t: and comes from applying the
 -325-

Y - T Siu
open mapping theorem to

k n
-> Z (tt , (OX) •

By (3.2) the restriction map

factors ~hrough a Hilbert space H. Let 7r be the pro-

jection of 7 onto the orthogonal complement of the kernel
of the composite map

Define ~(~) = 7r Ittl • Then f(~) is independent of the

choice of 7 and satisfies the requirements. Q. E. D.

(3.4) Proposition. Suppose X is a Stein complex space and

Gl CC ~ C C! are open subsets with ~ Stein. Let

'l-t)) = {uJ)))} be a finite Stein open covering of - GJJ ()J = 1,2)

with U~l) C C u~2). Let a: ([ll -> f()~ be a sheaf-

1. 1. - AX! AxX
homomorphism such that Coker a is flat with respect to
AX X -> A. Then there exists an open neighborhood .Q. of
o in A satisfying the following:

O,
(a) for A(t fl C .Q, there exists

'Y':r(~(tO'f) x ~,Im a) -> r(A(tO,P) x Gl'

linear over ([,[ t] such that d.Y; agrees with the

restriction map and 1/ 'J!(7) II G °,f ~ c 11711 ,

0;
,p
l 't ~ t
 - 326 -

Y -T . Siu

(b) for A(tO, f) C .Q and k ~ 1 there exists

<p: zk (A(tO,1') x 'lilz ' 1m a) -> c;k-l(.A(tO, f) x Vll ' 1m a)

linear over a: l t] such that [) 'f agrees with the

restriction map and II f(E,) II
1\ ,to:i-
,p .
C Ill-I! °
Vl. 2 ,t 'f

where C is a constant, independent of l;, to, and f .

Proof. Consider first the special case where X is a poly-
disc. We can assume without loss of generality that there
exists an exact sequence

PI a Po
-> (9 -> (f)
.AxX AxX

Let (a)m (respectively (b)m) denote (a) (respectively

(b» for the case L ~ m • By using (1.6) and (1.3) to ob-
tain local solutions of (a) and by piecing together these
local solutions by Cech cohomology, we conclude that (a)l
holds and that Cb)m_l implies (a)m. From (3.3), we con-
clude that Calm and (b)m~l imply 1b)m. Hence the special
case follows by induction on m.

Fbr the general case, we prove (a) first. We can

assume without loss of generality that

i) X is a complex subspace of an open polydisc p., and

ii) there exists a commutative diagram of sheaf-homomor-

phisms
 - 32 7-

Y T . Siu

a. f!JPO
->
A~P

jquot.
~
a. Po
->
({)AXX

N
such that Coker a. is isomorphic to Coker a. under
the quotient map-

Then (a) follows from (3.1) and the special case. For (b)
let m be a positive integer such that no more than m mem-
bers of vs. can intersect. We can assume without loss of
generality that we have an exact sequence

p
(0 m ->
AxX

rhen (b) follows from (a) and (3.3).

(3.5) When the flatness condition on Coker a. is dropped in

(3.4), the conclusions of (3·4) remain valid with the follow-
ing modifications.

i) 'Y' and 'f are linear over <C , but may not be linear
over <C[t].

ii) Fbr a fixed to, f has to be sufficiently strictly

small.

iii) C may depend on to and p.

However, i f is not a zero-divisor of any stalk of
 - 328 -

Y -T. Siu

Coker a ,then C can be chosen to be independent of fn •

(3.6) Suppose X is a Stein complex space, 7 is a coherent

analytic sheaf on Ax X, and 'f : oP -> "J is a sheaf-
AXX

epimorphism. Let A(t O'P) C C A and U be a relatively

compact open subset of X. For f 6: r(A(t O, f) x u, "}) de-
fine the Grauert norm as the infimum of

111:II 0 where
U,t 'f

A different choice of rp would give an equivalent semi-norm.

When we use such a Grauert norm. we assume that a fixed 'f
is chosen.

Suppose Vl "" .[ Ui } is a collection of relatively com-

pact open subsets of I . Fbr

define

When to = 0 ,II~ II 0 is simply denoted by 11~IIVl •

'lil, t 'f 'f
(J.7J Proposition (Theorem B with Bounds). Suppose I, Gv
 - 329 -

Y -T. Siu

U" (JI .. 1,2) are as in (3.4). Suppose '1 is a coherent

analytic sheaf on A lC I flat with respect to n:.a ll I -> .c.
and suppose a.: (OP -> 7 is a sheaf-epimorphism. Then
.AliI

there exists an open neighborhood.Q of 0 in .6. satis-

fying the following:

(a) for A(t O,pl C n there exists

linear over ([ [ t] such that a.1j! equals the restric-

tion map and /rr( 7)/I Gl't0 ,r ~ c /I? /I 0
G.2 ,t ,p

(b) for A(t O ,pl C.Q there exists

linear over a: [t] such that 6 rp equals the restric-

tion map and Ilr(~)11 0 ~ Cll~" 0 ;

l\ ,t 'f 1712 , t ,f

C is a constant independent of Z, , t o , and f .

(1h!m '1 is not n-flat, for a fixed to, ([ -linear 'f and
'f exist for p
sufficiently strictly small but may not be
, 0
linear over <[ Le l and C may depend on t and f. If
t n - to
n
is not a zero-divisor of any stalk of
.
'1 , C .£1!!!
be chosen to be independent of fn •j

Proof. Follows from (1.6), (1.3), and (3.4).

 - 33 0 -

Y -T . Si u

§4. Leray's Theorem with Ibunds

(4.1) First we examine the diagram-chasing proof of the

usual Leray's theorem without bounds. Suppose 1 is a co-
herent analytic sheaf o~ a complex space X and

{U(l}11 c; A

{Vi} ·' r:
~ '"'- I

are Stein open coverings of X such that 1( refines Vi

with an index map ~ : I ---> A. Leray's theorem states that
the restriction map

H
L
(tt, 7) -> 1.
H (1(,1)

is an isomorphism. Define cf"v (tt, l() as the set of all

such that

is skew-symmetric in (la, ••• , I1

p and skew-symmetric in
i O' ••• , i)l. Define
 - 33 1 -

Y ~T . Siu

6
1:
ef'v (Vt, 1.() -> ef+l,v (1.t, l( )

62 : ef'» (U, l() -> ef,Y+l (11, l( )

e1 : c~
or ,F} -> CO ,v (l7l, l()

8 : efClJl ,F)
2 -> ef,O.(7Ji, I()

as follows:

. Iuo
.. c,.~O· •• ~)l a.
('IV .
~O
n···()v.~).1

Consider the following commutative diagram:

 - 3.32 -

Y-T. Siu

a a a

0->
1
rex, "1 } -> CO (Lt, 7
1)
-2-> cl(Vl,
1"1 } _6_> ...

J e2 'If1 e2'iI1
0-> CO (1(, '1) ->
1
9
co, °(lJl, 1(')
6
..J:...>
61
cl,0 (1t, J.( ) ->
6J
el
0-:>- cl (1(', 'l) ->
62

CO , 1
1 6
62/
(a, )() -> cl (a,1 ,1
'If
l( )
61
->

6/
'V
62/
'If
62 J
A sequence

is called a zigzag seguence if

f*d (; zJ(l(, "l)

f (; cv,l-v-l(lt,l()
JJ ,f -))-1
, i
f I. , * (; z (lot, "l)

The proof of Leray's theorem consists in showing that the

 - 333 -

Y-T. Siu

correspondence

(cohomo.l rvy class of f*,t) <;--> (cohomology class of f t ,* )

(where f*,£ and f£,*

are the end-terms of a zigzag se-
quence) defines an isomorphism between H1(')(, '1) and
Hi <7.t, "1 ) which agree s with the map Hi ('1..t, "1) -> Wi (l(, '7 )
defined by restriction. This is shown in the following
three steps.

a) Fbr every f*,L ~ zl{l(,"J) one can construct by the

Theorem B of Cartan-Oka a zigzag sequence with f*,£ as the
first term. Likewise, for every f l,* ~ Zl (Vl, "] ) there is
a zigzag sequence with f l ,* as the last term.

bl If f*,£ ' f )1.1-)1-1 (0 ~)J <.£-1), f£,* is a zigzag

sequence, then f*,£ ~ Bl-l('l(, "l) if and only i f

f,£,* ~ B1-l(lt, "1 ) . For the "if" part, let

with

Construct inductively

(0 ~ )J ~.£ -2 )

such that
 - 334 -

Y - T . Siu

e1g * , £- 1 + f
0,£-1

&1 g )./-1,£ -li-l + f

li,l-1I-1

The construction is possible, because of the Theorem B of

Cartan-Oka and the following equations:

&1 f v-l,£-)l-2 + 6 f
2 1I,}->l-1
0

Finally, since

it follows that f 1,* '" &(-gt-l,*) • The proof of the "only

if" part is analogous.

c) The correspondence agrees with the restriction map,

because, if
 - ~~5 -

Y- T . Siu

the sequence f *, .e' f lJ,!-v-l (O~V<£},f£,* defined by

(f* ;.).
, ~O···~
.

is a zigzag sequence.

(4.2) Proposition (Leray's Theorem with Bounds). Suppose

X is a complex space and 7 is a coherent analytic sheaf on

DxX flat with respect to n ; .6x X -> A. Suppose Vi., l( ,
, 1('
a, are finite collections of Stein open subsets of X,
each of which is relatively compact in some (but in general
not the same) Stein open subset of X.
, ,
(a) (Surjectivity). Suppose 1{ < < 7Jl,)( < < )(, and
lUI C C 11(1. Then there exists an open neighborhood .Q of
° in t::. satisfying the following. For A(tO,PlCQ and
! ~ 1 there exists

(Cf,1j'l: zl(D(tO'fjXI(, 1)
-> Z£ (.6(tO ,f) x Vl,1) ffiC f - l (A(tO 'f) x)(', "1)

linear over ([ l t] such that

 - 336 -

Y-T . Siu

Eo = 9' (l,) + 6 y(l, ) 0 n A ( t o , pl x 1('

Max(IIPU,)
~ Il,~ Vl., t
0
'f
' Ilr(~l II", 't ,
° )\~
t ,f
c 11l,11)(,to , f

where C is a constant independent of !" to, and f.

, ,
[b ] (Injectivity). Suppose?il < ~ 7..t, 1« < tt; and
lUI C C /l(l. Then there exists an open neighborhood Q of
° in A satisfying the following. f2£ A(tO,fl C.Q and
L~ 1 there exists a map e from

[(2,'7l c;: Zi (A(tO'fll< vr.' ,7)ec£-1 (A(tO,Plx\(,1) ~= 67 on

A( t~ pl xl(]

!Q. cl - 1 (A ( t O' f ) x 17l, 7 ) linear over ([,[t] such that

6e(!,,?) .. !, on L::!.(tO ,fl x 7J1.

/le(!,,?) °~
J/,t""t,p c Max (II!'I/,
n,t,e
° ' 11711If,t°,'f' ),
C is a constant independent of !" 7, °
t , and f

(When '1 is not n-flat, for a fixed to, ([, -linear T' 1f,
e exist for p sufficiently strictly small but may not be
linear over 4: [t] and C maY depend on to and f. If
t n - t~ is not a zero-divisor of any stalk of 7 , then
can be chosen to be independent of fn .)

Proof. Fbllows the Same line as in (4.1) except that (3.7)(b)

(Theorem B with bounds) is used instead of the Theorem B of
Cartan-Oka· Q. E. D.
 - 331 -

Y-T . Siu

~. Extension of Cohomology Classes

(5.l) Andreotti-Grauert [1] proved that a k-dimensional co-

homology class with coefficients in a coherent analytic sheaf
7 can be extended across a strongly k-pseudo convex boundary
and across a strongly (r-k-l}-pseudoconcave boundar y if
codh? ~ r . We need the corresponding result with bounds.
So we are going to examine one key point ~f Andreotti-
Grauertts result in such a way as can be carried over to the
situation with bounds. The precise statement of the situa-
tion with bounds is given in (5.2).

Suppose X is a complex space and 7 is a coherent

analytic sheaf on ~)( X. Suppose . 1. ~ 0 and Xl' DeC X
are open subsets. Assume the following.

H£ (A x (Xl" D) , '1) .. 0 in case t ~ 1

Wt(AX D,t) -> Hl(A x(Xl" D), 7) is surjective
in case /, .. 0

I+l(AxD,'.1) -> I+l(AX(Xln D),"l) is injective

Let ~ .. Xl U D. From the Mayer-Vietoris sequence of '1 on

(A x Xl) U (A x D) it follows that the restriction map

is surjective for )I .. 1. and is injective for ]) = 1 + 1

Such an argument cannot be carried over to the case with
bounds. So we look at the conclusion in Rnother way. Intro-
 - 338 -

Y - T. Siu

duce the following additional assumption

With this additional assumption, it is easy to see that one

can choose a Stein open covering ttl (respectively tt 2 ) of
Xl (respecively D) such that tt
12
: .a ttl' U2 covers

Xl f"I D and

(* )

Let ~.a ViI fl1J[2. We are going to show, in a way that Can be
carried over to the case with bounds, that

~ ~
H (~xU,7) -> H (Ax~1,7)

is surjective for v ... 1. and is injective for v » 1. + 1 •

Let us consider surjectivity first. The case 1, ... 0

is clear. Assume 1. ~ 1. Suppose ; <;: Z.l(A x U l , 7) •
Since H1(A x 'Zk
12
,"1) .. 0 , there exists 7 <;: Cl-l(,A)C Vt 12 ,"1)
such that

l,1.6 x V!12 67·

Extend 7 trivi ally to ? <;: Cl-l (.6. U l , 7) Since ~ - 6"7
x

is 0 on A x 'U12 ' by (*), ~ - 6;; can be trivially extended

to some element of Zl(Alltt,7).

Now let us consider injectivity. Suppose

l,<;: ZL+l(6)CU,7) such that
,
67
 Y.-T. Siu

for some ?' ~ CI(06 x vt1 , 7). Then there exists

7" ~ C1. (A II z,t2/1) such that

~ ~ 67" on A x tt12 •

Then

o on Ax tt •
12

In Case L ~ 1 , there exists t; ~ Cl - l (A x ~12' 1) such that

6(, = 7, -?" on A x 1.t

12
•

,.. I. 1
Extend l; trivially to l; ~ C - (Ax 'lJl ,
2
"1). Define
7~ 1
C (4 x u ,1) by
,
C· 7= 7
7"
on
,..
+ 6 t; on
Ax~
1

A x 1.t •
2

Then l, = 6? on A x 1.t. In case R. = 0 , (7' - 7") IA x Vl12

Can be extended to e <; Z£ (A x 1..t 2 , '1). Define
? ~ Ct (A x tt , '1) by

Then !, = 67 on A x Vl •

Now we consider the strongly p-pseudoconvex Case.

Assume that there exist

i) a biholomorphic map r embedding an 0 pen neighborhood

 - 340-

Y- T . Si u

Ii c C X of D onto a complex subspace of an open poly-

disc P of a: N
11) real-valued cf functions 11 ~ 12 on X such that
a) Xl .. {erl < OJ
b) D .. D(\ {Cf2 < O}
c) SUpp ('PI - f2) C C D
d) (J)i-",-l
) 0 . t h e restriction of a
~s cf function 1i on
P (i" 1,2) whose restriction to pn({x}xa: N- p+ l )
is strongly plurisubharmonic for every x <;: lt P- l
e) there exists an exact sequence

(I) . .
,..Pm ....PO 0 ..,
o -> n+NI:.r -> ... -> n+NU -> R (.('1/AX D) -> 0

on L:::. x P , where ::; : A x D -> A x G is defined by the

identity map of .A and '( •

Then, by an intermediate result of Andreotti-Grauert [1,

p.217, Prop. 11],

It follows that (t),t is satisfied for l6> p. In consid-

ering the situation with bounds, one has to apply the open
mapping theorem to

( 26> p)

for a Stein open covering )( of P n Uf'"l < 0 J. in a way

analogous to the proof of (3.3) and one has to apply (3·4 )(a)
to the sheaf-homomorphisms of (#).
 -341-

Y-T .Siu
For the strongly p-pseudoconcave case, we can assume
that there exist t ,and ~l' f2 as in the p-pseudoconvex
case such that

Xl'" {<f2 > o}

Xl U D '" {'fl > oI

and conditions c), d), e) of the p-pseudoconvex case are
satisfied, where m ~ n + N - codh'" on Ax D • and,
moreover, the Hartogs' figure

•
'"
is contained in p(H f 2 > O}, where P '" P I X P II with
P'C<c P and PIlC<c N-P , ¢." H{CCpI S H"CCP" are open
polydiscs. Again, by the proof of an intermediate result of
Andreotti-Grauert [1, p.222, Prop. 12],

is surjective for t < N - p. Hence the first two condi-

tions of (t)t are satisfied for i « n+N-m-p. To obtain the
third condition of (t)£ for 1. '" n + N - m- p - 1, we argue as
....
follows. For every given e > 0, we can assume that t, 'fi ,
HI ,H" are so chosen that, for some open polydisc
H1' C H'2 C C. P ' • t h e Hartogs I f igure

Q2: '" (pi X (p" _ H")) U (H2 x p ll )

contains P n{ ~l > E.} (cr , [1, pp , 219-220]) • It suffices to

show that the restriction map

 ·342-

Y - T. Siu

N-p" N-p
"
II II
is injective. Suppose P = TT P. and H TT H
j
with
j=l J j=l

j-l N-p
I
""
U .. P x TT P x (P - HjP
j j TT PJJ " " (1 ~ j ~ N-p)
/pI It ~"j+l

lt1 .. {U~l) , U
1'
... , UN_pI
1%2 .. {U~2) , U '
1
... , UN_pI
An element C,i <;: HN-P ('l~.' NlD) (i = 1,2) is represented by a
holomorphic function U n ••• n UN_ p,
fi
1
on U~ i) () and

~i .. 0 if and only if the holomorphic function

2: f(i) za p+1
ap+l'··· ,aN"< - 1 a +
p 1 • • .a N p+l

on U~i) (I U1 (\ ••• n UN_ p can be extended to a holomorphic .

function g.
1
on U
1
(I ••• n UN-p where

GO

f.
1
= 2:
apfol' ••• ,aN" _CD

is the Laurent series expansion in the last N- P coordi-

nates zpfol' ••• ' zN of a:: •N
Hence 1: is injective. When
 - 34 3-

Y -T . Siu

~i = 0 , the function fi when regarded as an element of

ZN-p(Ui, N~) is the coboundary of

(i)
{h . . } c;: CN-p-l(Lt., NO)
JO·· ·IN_p_l ~

where
(i)
h = gi
1 •.. (N-p)
(i) a p+ l
h 1\
01 •• • v ••• (N-p)
2: f
(i)
z
? 0 a p+ l
a p-+: l • • .a N p+l

ap+L l ? 0
a p+)I :i- -1

It follows that, if fl is the restriction of f2 and fl

is the coboundary of some

and if

sl ••• (N_p_l)

then

(2 )
and f2 is the coboundary of [h . j }. Hence, if
JO··· N-p-l
the sup norms of Sj j and f2 are all :i- e , then
0··· N-p-l
 - 344-

Y- T . Siu

(2 )
the sup norms of h .
j 0··· IN_p_l
are ~ Ce, where C is a

constant depending only on P" and H" Now, to get the

situation with bounds, one need only apply the open mapping
theorem to

(t < N-p)
,..,
for a Stein open covering ~ of P (l {Cf2 > O} in a. way
analogous to the proof of (3.3) and apply (3.4)(a) to the
/

sheaf-homomorphisms of (#).

(5.2) Proposition (Extension of Cohomology Classes with

Ebunds). Suppose X is a complex space,
1: X -:> (a*,b*) C (-lXl,lD) is a cf function,
a* < a#< b#< b* ' p,q" 1 , r" 0 , and "1 is a coherent
analytic sheaf on 6 x X such that

i) { 9' < b ]" = {1 ~ b ] for b# < b < b*

il) Cf is strongly p-pseudoconvex on {'f > b#} and

strongly q -pseudo convex on {" < a#}

iii) the natural projection (Supp1) n ~x{a ~ 'f~ b})--->

is proper for a* < a < b < b*

iv) "] ii flat with respect to the pro-
jection n: 6 x X -> 1:::.

For a* < a < b < b* let x~ = {a < " < b}. Suppose
 - 345 -

Y - T . Siu

n. c c c: is an 0 pen neighborhood of 0 and Vi, 1.(, Vi.' , )('

are finite collections of Stein open subsets of X , each of
which is relatively compact in some (but in general not the
same) Stein open subset of X. Then, for ~ < a l < a# and
b# < b l < b* ' there exist a* < a2 < al and b l < b 2 <b*
(independent of ill which satisfy the following.
(a) (Surjectivity). Suppose a2 ~ '"
a < a ~ al '
bl ~ b < 2 ~ b2
'" ' 1.(' "
<.<" ~,
, ' < < '1/'
(.11., 1(
I1(' ICC Xb
a
b ,., b -
IVlj C C IN , and (n x I ) n Supp'i C C .Q x \)e\' Then there
a - a
exists an open neighborhood .Q . of ° in.Q'" with the fol-
lowing property. For ~(tO,P) cn and pSi. <r - q - n
there exists a map

(rr ,1J') : Z t ( A (to, f ) x l( ,"1)

-> Zl(A(tO'fpc?Jl.,'1)(BC£-l(A(tO,f)lCl(" ,7)

linear over a:. [t] such that

E.,= '1'([,) + 6'Y'(~) on A(tO,P)X i

Max (1Irr([,)II U, t °, r ,IIJV(E,)/L,
" ,t°'F )~
.
CIIE,1I)( ,t°, f

where C is a constant independent of 1" °

t , and f •
, ,
(b) (Injectivity). Suppose 'Vi < < ~ , 1« < 'lk ,
jttj CC I~ , (SixX b ) n Supp"lCC QX\\fI, and
.. a a
(.Qx X~) () Supp"} CC Qxlu'l. Then there exists an open neigh-
° -
a
borhood Q. !2!. .!.!:!..Q with the following property. For
A(tO,f) C n and p < £. ~ r - q - n there exists a map f)
 ,- 34'6-

Y - T , Siu

from

{(~,7) c;: ZL(A(tO,P)XU' ,"l)E9C l - l(A(tO,P)xl.(,"l) ~= 67 on

A( to " f) Xl(}

to Cl-1(A(tO,p)xVl,"l) linear over ([[t] such that

where C is a constant independent of .; ,7 , to, and f .

(When? is not n-flat, for a fixed t
°,([ -linear 'f ' 1j ,
() exist for f sufficiently strictly small but may not be
linear over c [t] and C may depend on to and f. If
t n - t~ is not a zero-divisor of any stalk of OF , then C
can be chosen to be independent of fn .)
 - 3'4 7-

Y,.T. Siu

§6. Sheaf Systems

The treatment given here is a simplified version of

the sheaf systems introduced by Forster-Knorr [3].

(6.1) is a collection of open subsets

of a complex space X. Let Av be the set of all
(i O ' ••• , iJl) with 1 $ iO < < iJ1 ~ k and let
A'" U A •
v~ 0 v
For a.". (i O' ... , ilJ) G: A , denote

U. (1 ••• nu. by Ua.. For f\'" (jO' ••• , jll.) G: A , a. C f\

~O ~11 ,..

means {iO' ••• , iv} C {jo' ••• , jp.l •

A sheaf system on £1 is defined as consisting of

i) a coherent analytic sheaf ~a. on Ua. for a. G: A

ii) a sheaf-homomorphism Yf\a.: §a.IU p ->§p for a. C ~

such that Ya.a. is the identity map of -9a. and

for a. C ~ ct.
A sheaf system (§a.' f~a.) is said to be free if each
§a. is isomorphic to (O~a. I Ua. for some Pa.·

A morphism from a sheaf system (§a.' 1j'~a.) to a sheaf

system (~, t~a.) is defined as a collection (Ba.) of
sheaf-homomorphisms ea.: -ga. --> Wa. such that
e~ • Y~a. = J pa. 0 (Ba. I U~) for a. C p. The kernel of a mor-ph i sm,
an exact sequence of sheaf systems, and an epimorphism of
 - 34 8 -

Y -T . Si u

sheaf systems are understood in the most obvious sense.

t t k'
Suppose 1,t {Ui}i=l is a collection of open subsets
of X which refines Vl by an index map 1: ': {I, ••• , k I} ->
I
(1, ••• ,k} , i.e. Ui C Ul: (i) • For a sheaf system
.p...
'1 (.f1.
~a'
1U
T ~a
) on "'90,
C/L
define a sheaf system

1:*§ = (~a" J ~ 'a tl on

..,'11 t
VL as follows.

i) '¢fa,::o ~(atdU~t

ii) J~ta': 9+a t ->W~t equals ~(~')-r(a'dU;'

where "'t"(iO' ••• , i
v)
means (T(i
o ) ' ••• , 1:(iv)) • -c*-§ is
also denoted by ~I~' .

Suppose 7 is a coherent analytic sheaf on

.Q: • lUI. We can regard Vi as a refinement of Q with
the unique index map a- and obtain (1"'* 7 . In such a
case, we identify 0; with the sheaf system a-*7 •

(6.2) ~. Suppose Ui(l ~ i ~ k) is a relatively com-

pact open subset of a Stein open subset '"
Ui of a complex
k ~
space X and 1 is a coherent analytic sheaf on U U .
i::ol i
Then there exists an exact sequence

... ->1(, m -> ... ->R.1 ->R.a ->;r

"..., -> 0

of sheaf systems on ?Jl: ::0 lUi }~::Ol such that each R.. m is free.
 - 34&-

Y -T. Siu
, " N
Proof. Take Stein open subsets U. C CU. C C U C CU.
~ ~ i ~
~~ , k ,~" k
(1 ~ i ~ k) and let V~ = !UiJi=l and ~ = {UiJi=l. It
suffices to show that, for an} '. """Rf system -§ = (§a' '!f~a)

on 1t." , there exist a free sheaf system ;/ = (.Ja , cr~a) on

1Jl' and.an epimorphism f) : .J - ,->.g Itt'. We construct them
as follows. Fix a multi-index a O' It suffices to con-
(a) ( (a) (a) ,
struct a free sheaf system J 0 = .J a 0 ,cr~a 0 ) on ~
(a) (a), (a )
and a morphism e 0 R/ 0 -> §IVl such that BI¥ O
(a ) 0
is surjective, because we can set ~ = $Jl 0 and obtain
a
O
There exists a sheaf-epimorphism
on Define

;/(a o ) p ,
a = (OX1ua for a
O
C a

,
a
(a.O )
0 for a
O
et a
tr (a O )
~a
PI U~'
the identity map of (Ox for a
O
cae ~

o:(a o )
~a
= 0 for a
O
C/; a

e(a O )
a = Yaa 0 (7 I Ua )
a for a
O
C a

(a )
ea O 0 for "o et a •

The construction is complete.

 - 3 50 -

Y - T . Si u

(6.3) Suppose ~. {Uilf=l is a collection of open subsets

of a complex space X and.q = (§a.' 'Y'fla.) is a sheaf system
on VL. Introduce the cocha1n group

= IT r'(u J -§ )
a.(;A a. a.
- "1
(where At is as in (6.1)) and define the coboundary map

by

where

fl •
;:...
••• , Ji J

SUppose 7 is a coherent analytic sheaf on X and

... _>R,m -> ... _>R. l _>R.0 ->'7 -> 0

is an exact sequence of sheaf systems on £n , where each

~m is free. Consider the following commutative diagram.
 - 3'5 1-

Y - T . Si u

o o o o

-> ...

-> ...

-> ...

->

... ...

where the horizontal maps are coboundary maps and the verti-
cal maps are defined by the morphisms of (*). Let

IT
OJ
cl+JJ (Vi, e" , .
)I =0

An element ~ of CR(a) is given by (~l+~ ,~):=o with

~ R+v ~
">R+ l.l ,~ ~ c (If, 1{, j. Define

d : C1 (Vi) -> Cf.+ 1 (Vi)

by
 - 3 52 -

Y - T . Si u

where

Define

by

(6.4) Proposition. With the notations of (6.)), if each Ui

is Stein, then the map e*: H~C' (11») -> Hl~. (Lt,1J) in-
duced by () is an isomorphism for all £
(This proposition can easily be proved by a spectral sequence
argument. However we prefer to present a more elementary
proof, because it can be carried over to t~ case with bounda , )
,., I
Proof. (a) (Surjectivityj. Take l, ~ Z (Vl,?). By Theorem

B of Cart an-Dk a , one can construct, by induction on 11 ,

~ r: Cl + )) ~i) :i- < lD)

"1.+)) ,v (0
(, ..,
~ V( "' )J

such that

d.l,OE,l,O ,. ~

? 1) •

Then

and 8!," l, •
 - 353-

Y-T . Siu

(b) (Injectivi ty) • Suppose

E, :0 (SL+v,11 )::00 c;: ZJ (t!)

is mapped to 0 in Hi (c'(~, 7)) Then there exists

~ c;: C2- I(Vl,'1) such that

By Theorem B of Cartan-Oka, one can construct, by induction

on v ,

(0 ~ v < 00)
such that

d£_I,O 7,-1 , 0 t;

d£_I+J', 117R._I+1J,11 :0 (-It ~l-I+)l,)I-l + 5.t-2+ 1J,"J-17R.-2+ 11,))_1

()I~l) ,

because

It follows that

n{ : .
( 7£-1+)),11)1:00
)00 c;: Cl - l (Vl)

satisfies a? = l, . Q. E. D.
 - 3S4-

Y -T . Siu

(6.5) For a given strongly (p,q)-pseudoconvex-pseudoconcave

holomo~phic map and for a given coherent analytic s heaf
on the domain space, we are going to use the results and §5
and §6 to construct a sequence of complexes of Banach bun-
dles which Can be used to calculate certain direct images of
the sheaf.
b ,.,
Suppose I, 1 ",. p, q , r, Ia,Q, ai' bi (i = 1,2)
are as in (5.2). We arrive at such a situation by consider-
ing the graph of the strongly (p,q)-pseudoconvex-pseudocon-
cave holomorphic map and the trivial extension of the.co-
herent analytic sheaf to the product after its transplanta-
tion to the graph.

Let m ~ 1. Choose

Choose finite collections

Vlj «Vlj + 1 « 7.Jl (1 ~ j < m)

of Stein open subsets of I such that

~ d. ~ d +
n Supp7'cc.QxIUjICCQxxc j +1 (0
N
i) (.QXI J) ~ j < m)
cj j 1

ii) there exists an exact sequence

1
... -> II
~ -> ... ->~ -> ~0 -> '1-> 0
,., ,.. ,}
of sheaf systems on Q)C 1.t where each 1? is free.
 - 35 5-

Y-T . Siu

By (6.3) and (6.4), for 1 ~ j ~ m and for any Stein open

subset Q '"
of n , we have a complex

oJ
(constructed from the co chain groups of It IQ x 1tj ,
o ~ v < al) whose 1. th cohomology group is isomorphic to
H't(Qx~j,l) (O~.e<al). By letting Q vary, we obtain a
complex of sheaves

""
on J2 whose cohomology sheaves are isomorphic to the direct
image of 71nx /Vljl under the projection.Q lC I~jl ->.Q .
Now we turn to the situation with bounds. Fix 1 ~ j ~ m •

Let 1tj = {Ui}~=l. Let At (0 ~ t < al) and Ua

be as in (6.1). Define the subgroup ~(QxUj,Jt)
i ~
C (Q x Vi., Jl) as follows. We say that an element
J
J. oJ
f = (fa) c; C (Q x Vlj , R)
a c; At

belongs to C;(Q x z.tj , R) i f and only i f fa c; r (Q x Ua ,(R/Vlj)cx)

satisfies

for all a ~ At. Let

 - 356 - .

Y - T . oru

As in the case without bounds, we have a complex

As Q varies, we obtain a complex of sheaves

~
on ~,. Since each on"
~ is free, this complex of sheaves is
naturally isomorphic to

o -> ocE?)
J
-> (9ci1:)
J
-> ... -> VCEJ~) -> ...

where V(E~) is the sheaf of germs of holomorphic sections ofa

t'" 1 .
trivial Banach bundle E j on Iz , Let 1:'j be the i t h cohomology
,., dj ...
sheaf of (#)j. Let 7rj : .Q.xXCj - ) Q be the natural project-
ion map. We have natural maps defined by restrictions:

').It.
>'f'J
_> Rt Cn.) *;r
"?
-> ')/1.
,.,..J- 1 -> R
l. Cn . 1) * ...,
;r •
J J-

For p ~ l < r - q - n , by the bumping techniques of Andre-

otti-Grauert [1] and by C5.2), one concludes that the maps

RlCn j ): l - > R£Cn j_l ).'7

1. L
9:tj -> ~j-l
satisfy certain conditions of surjectivity and injectivity.
lliese conditions can be translated into relations between ~jI
and Rf Cn j )*1 . llie coherence of RlCnj);:t will be investi-
gated by working with Wi. lliese statements will be made
precise and presented in detail in Part II.
 - 35.7-

Y-T.Siu

PART II THE POWER SERIES METHOD

§7 Finite Generation with Bounds

(7.1) In (6.5) we constructed a sequence of complexes of

trivial Banach bundles and stated that the direct images,
whose coherence we are interested in, are related to the co-
homology sheaves of the complexes. NOw we are going to con-
sider abstractly a sequence of complexes of holomorphic Ban-
ach bundles satisfying certain conditions and derive conclu-
sions concerning the finite generation of the cohomology
sheaves of the complexes.

First we introduce some notations. Suppose B is a

trivial bundle with a Banach space (F, II-U F) as fiber. Let
~(B) be the sheaf of germs of holomorphic sections of B.
For to ~ A and d ~N~ let

When to = ° ,~(B)(tO,d) is simply denoted by ~(B)(d) •

By identifying ~(B)(tO,d) with its zeroth direct image
under the natural projection from .6. to

we can regard V(B)(tO,d) as an analytic sheaf over the re-

duced space A* . (O(B) (to ,d) is the sheaf of germs of holo-
morphic sections of a trivial bundle over A * whose fiber is
 - 358 -

Y -T .Siu

the direct sum of TT d. copies of F'. For

d.foo 1
1

with power series expansion

v
f)l
(t P to)

define

[r liB 0 d ..
tt t tf

Denote by B(to tdtf} the set of all

f <; r(A(tOtflt O(B)(tOtdl)

with
[r liB, to ,d 'f
When d .. (00 , ••• tOOl
°,dtf)
, sre

is simply denoted by B(pl •

Now we introduce a formulation in terms of abstract

complexes of holomorphic Banach bundles. Let

6 )1-1 6)1
••• --> e:- l
->
a.
Et:
a.
- > ~+l_>

••• - > ~-l_>

1 )1+1
lr~+l,a
"\I
)l
1
+1 ~+l-> Fu+l->
 · 359 -

Y-T . Siu

be a commutative diagram of trivial Banach bundles and bun-

dIe-homomorphisms on f:::. wi th
al ~ a ~ a2 for s ome integers aI' a 2; that is, we have a
sequence of complexes. Some statements concerning these com-
plexes which we will consider later are true only for
al + c ~ a ~ a2 - c where c is a positive integer depend-
ing only on n (and other given numbers). Such a restric-
tion will be clear from the proofs and will not be explicitly
stated. We will be interested in the behavior of these com-
plexes in a neighborhood of O. So we will sometimes re-
place A by some sui table A( fP) •
For t
o~ A and d ~N*
n ~
let ~a[t ,d]
0
denote the
vt h cohomology sheaf of the complex

...-> (D(Fb,)/-1 )(t0 ,d ) -> ~

(Q(~){t
O .
,d) -> (Q(Ea
1I+1 0
) (t ,d) ->"0
})
When to .. O· , ¥~[ to ,d] is simply denoted by 9?a[d] , and,
when d > (0), ••• , 0)) , it is simply denoted by '#a" •
For a < ~ let

0/ y
r~ ,~-l r~_1,~_2

For to ~ A and d ~ N~ let

When d" (0), is simply denoted by

 Y - T . Si u

11 ~ .,)
Consider the following conditions (E) , (M) , (F) ,
~
(B)n •

(Er (Quasi- epimorphism with fuunds). There exists a con-

stant C with the following property. Fbr a < (:I ,
~~~(tO,d'f) with 6;;=0 there exist

;"" : with

such that
i)

ii)

iii) l' 0 and 1f 0 are linear over 4: [t]

a,(:I,t ,d,p a,(:I,t ,d,f

~
(M) (Quasi-monomorphism with fuunds). There exists a con-
stant C with the following property. For a < (:I ,
e; (; ~
• 0
(t , d , f) and ? (; Ft3 ~l 0
( t , d , p) satisfying

and
oJ ~
r(:la"" = 6(:1v-I ? there exists

~ : epa,(:I,t 0 ,d,p ( .."~ ?) r:

~
Fo"-l( t 0 ,d'f )
il.+l

such that

i)
 -- 3,61- .

Y- T. Siu

iii) l' 0 is bilinear over It [t] •

a.,P ,t ,d'f

( F).J ( Finite-dimensionality along the Fi b

ers.
) Fbr t °
I~ ~

and d ~Nn there exists a commutative diagram of continuous

linear maps

...-> ~-l(tO,d'f) -> Eh(t

~ 0
,chp) -> EttHI (t ,d,p) ->. ... °
1 1 1
)1+1
...-> F1>-l -> F-J -> ->...
F 0
°
a.,t ,d,p a,t ,d,p
° a,t ,d,p

t 1 1
...-> Eb.+l °
1>-l( t ,d 'f) -> Eb.+l(t
)I
"+1 0 °
,d,p) -> Ea.+l(t ,d,p) ->...

where

i) the composite vertical maps are

11) f'l' °
a.,t ,d,p
-> Ff
a+l
(to,d,p) factors through a Hilbert

space

iii) the middle row is a complex of Frechet spaces

iv) "
dim H (F . ° )< IX)
<I: a,t ,d,p

v) HY(F'
a.,t ,d'f
° )- > H)/(F'a+l,t°,d,p ) is bijective
 -3 62 -

Y - T . Siu

~
(B)n (Finite Generation with Ebunds)-

a) Im(~
~

~,t
° --->~
~

~+l,t
0)
m
is finitely generated over n~tO

for t
o c= A_

b) Let t, (1 ) , ••• ,

(1 ~ i ~ k) and let A be the V O-submodule

n t
of

11
f:J 0 - . Then, for f sufficiently strictly small,
~+l,t

there exists a constant C satisfying the following.

f
If ~ . c;: ..~ (to 'f) with 6~E," ° such that the image of
~ Ln '>..l
..,...,.,. b e I ongs to A , then there exist
~+l,tO

a (1) , •• _, a(k) c;: r(A(tO'f)' n([))

? c;: <;i(tO ,r)

such that
 - 363- -

Y -T . Siu

(7.2) Proposition. ( E))I , (M))/+l ,

(B)n-l
)J \I
~ (B)n •

Proof. It suffices to prove the case to .. 0. Let

n
d '" (CXl , ••• ,00,1) G N
*
"N~[iIJo
°
~
'Wa.,
III

J}

H«+l :::
°- 1 W + l ,0)
Im(~ (J,
II II
rt

H)l
11+1
.. Im(~a. ,° Wa.+l,0) .II
--;>
--'Ii

"7
Since Ha.+l is finitely generated over n_lJo' there exist

t,(l) ,... , (for some p°)

with o~ E,( i) = ° (1 ~ i ~ k) such that they generate

over n-l(f)0· A ssume °

f'n
f.n < ~.
For

= ···,Pn-l)

f* = ••• ,fn-l' fn) °

In this proof clearly we can assume without loss of general-
ity that 0; and r:+l,r are norm-nonincreasing. In other
proofs where we can also make such an assumption without loss
of generality, it will not be explicitly mentioned. Since
to= 0, we will drop to from all notations if no confusion can
arise. We break up the proof into four parts.
 -364 - -

Y -T . Si u

(I) First we prove the following.

Fbr f sufficiently strictly small, there exists C,

such that, i f ~ <;: <. (p ) and 6~ E," a , then there
exists 9 <;: ~+ 1 (p*) satisfying

,. a
(* )
"11 "11

1/ 8 110.+1'f * ,.< , n f) C_
f 1lJ;llo. 'Io

where is the power series expansion

of ~ in t
n
•

By "
(B) n-l for
f sufficiently strictly small there exist

(1)
a <;: r(~(p*), n(!J) independent of tn

7 <;: ~~i(f*) independent of tn

<;: ." *
l; ~+ 1 (p )

such that

"711~~i, r
 - 36.5-

Y- T.Siu

t >"-1
(-!!) .
(Xl

Let ~*.. 1: ~ Then

>"=1 >.. Pn

Define

Then

Hence

M:>reover,

where C" is a constant depending on f . It follows that

f

Therefore (*) is proved.

(II) Next we prove the following.

 - 36 6 -

Y - T . S iu

For f sufficiently strictly small, there exists D p

Y
such tha t , i f c; E
>I
Eo.(p) and 0o.l,= a and m~ 1, then

there exists ~ *
em ~ E~+l(P) such that

o~lX.+l (e m + r~+l a. ~em ~

)" -m
n)
(t '\ = a
'A A fn )
(t)

is the power series expan~

sion of E, in tn.

We are going to construct em by induction on m. The

ease m=1 has been done in (*) of (1). Suppose we r~ve

~
(
2:
00
~
,-m J + e
tn
1:'" or
rlt+l,cx. h"'m A (fn) m

y y
Then 0ll+l L or a • Let e ,,; 1lE,II Cl,p Then

< e + Ilemll:+l'f <

or e + fnD e •
IJoe 11:+l,p '" rr

We assume that fn 1s so small that 2 • So

v
1I"t11~+l'F ~ 2e •
 - 36 7-

Y -T. Siu

is independent of tn. By

applying (*) of (I) to 1: , we obtain '" <;;:

;V ~ (F* )
~+2 such
that
~ ~ o
~a+2 ,a+l (I"" + 1:
r a + 2,a+l *) =

y
1I;tlla+2,p* ;f fnCr/e •
Let

(X) t A-m-l
l,# = E ~ ( n)
A=m+l A fn

We are going to apply (M) " +1 to

For this purpose we need estimates on the norms of the two

expressions in parentheses.

First we are going to obtain an estimate on the norm

of ~~ ~#. Since

by defining
 - 368-

Y-T . Siu

f'or tn =I 0 , we Can regard 6~ E.# as an element of'

~+l(f*l. Because

it f'ollows that

e •

NOw we are going to obtain an estimate on the norm of'

(r~+2 ,a. c,# - r~+2 ,a.-+l"r* - 1-). From the def'inition of' "r ,

~ 2e + e + PDr
De ~ 4e

(where we have defined

 - 369 -

Y -T .Siu

-1
(;~) (LO - r~+l,o.C,m - em)

for tn f O}. It follows that

' \I

(;n)(r~+l'o.S# - 1:*) ,
n o.+l,p*

~
Pn
-0 4e •
Pn
Hence

)1+1
By (M) ,there exists such that

where C' is the constant coming from (M}~+l . This fin-

ishes the proof of (t) if we put

~
(III) Now we apply (B) n-l to get the finite generation of
~
~+ lover lJ o • Let
 -3 7 0 -

Y - T . Si u

be the power series expansion of em in tn Since

and
11

~ emO + r ,) + ~ 11
a 1 ,a m a + 1 'P*
:i-

v
by applying (B)n_l to the n_lOO-submodule
11 -11 -v
Im (Ha + 2 -> ~+2) of ~a+2,o generated by the images of
,)
ra+l,at,
(i)
(1 :i- i ~ k) , we can find, for f sufficiently
strictly small,

(i) (; r(A(f* l, nlD )

am independent of tn

'7 m (; ~';}(f* l independent of tn

11 *
l;m (; ~+2 (f )
such that
 - 37 1-

Y - T . Siu

I a (i) I <
=
" f n D + l)e
Cf5(
m f* jO

" r Dr
117mll::~ ,p* <
= C,o ( n + 1) e

" (f D + l)e ,
IIt; ml(+2,p* ~ C_
f n jO

C"
)l
where p is the constant from (B)n_l •

From (#) it follows that

< "
C"' C_( fnD- + l)e
IIt;ml(+2 'f* '" f f

where C'" is a constant.

m
By multiplying (#) by
(~~) and summing over m ,
we obtain

where

e""
III t m
a
(i) E a (i) (...E.)
m=O m Pn

III t m
E ~ (
m-O mPn
n)
 - 3 72 -

Y -T .Siu

It follows that

lie11:+ 1 'f ~ f n Dpe

I a {i} I "
C-(fD_+ 1)e
~
r ~
f n f'

II? Il a+ 2 , f " fn D
< cpt + 1)e
>2
1
-;
Ill; Ila +2 .f ~ CC '" Cf{fnDf
..., " + 1 )e
where '" is a constant.
C

1 }e •

~
By (E) ,we can find

such that
 -373-

Y -T . Siu

A ~
where e is the constant from (E)· • Choose D
In so sma 11
that

"(
e fn D r ,It"e_(~
-Bee ) 1
2l" +
r fn r n
D
f
+ 1) ~-2

Define

Fi (i)
(~) a

11 1 ....
~(~) r a';3 ,a+27 + ?
,...,
'f(~) ~ .
,.
Let '¥ denote qlo ••• o'J! (). times). Then

" (f
e_ D_ + 1) e + e •
~ 2
r n r

This is almost but not exactly the result we want concerning

finite generation, because the equation is in Ea + instead
3
of Eh+l' We are going to remedy it by using (E)" first
before using the argument to get the equation.

By (E)" there exist ~(i) ~ Ell (pO) (1 ~ i ~ k)

a-2
. y A(i)_
with
° such that they. generate
°a_2 E.. -

Im(f-b:-l -> Ha_ 1) over

" )I
n-l °
(We retain f
° and
I[) k
 .- 3 7.4-

Y-T , Siu

simply to avoid the introduction of more symbols and such a

retention clearly does not result in any loss of 'gener al i t y . )
·
By app1 y~ng (E)" to ~~ ,we can find
/\ ))

t, ~ Fu-2 (f) with

/\ 11
7~ ~+l (f)

such that

11~1I:-2 ,P ~ Ce
11911:+ 1 , p ~ '"Ce •

/\ t(i)
By repeating the preceding argument with t, , ~ (instead
of E" ~(i)), we obtain (for p sufficiently strictly small)

such that

II 1\
r J: ..
a.+1 ,a.-2 ">

(where again " ' D

Cp are retained simply to avoid more sym-
f
bols and such a retention does not result in any loss of gen-
 - 375-

Y-T. Siu

erality). It follows that

~
Hence ~+l is finitely generated over nOO. We have actu-
ally proved more than this, namely, we have shown that the
finite generation is with bounds when generators are chosen
in a certain way.
,)
(IV) We are going to prove the full strength of (B)n b) by
invoking the existence of privileged polydiscs (in the sense
of Grauert). Suppose ~(il ~ ~(fo) (1 ~ i ~ k) with
6~(i) .. 0. Let A be the n<OO-submodule of ~:+1,0 gener-
ated by ~(l), ••• , ~(k) •
By (E)" and t h e finite generation
n_l<OO ' (after shrinking pO) there exist
(1 ~ i ~.£) with 6,)11-2 l; (il "" ° such that
»
generate Im(Hll_ l -> -HL1). Let

f: nOO
t -> Pill+l,O
v / A

be induced by the 1 xI matrix

,) (l) )l (1) )
( r"'+1,"'_21"
'" '" .., , ••• , r"'+l '" 21"
'" 1"'-'"
(an element of being represented by a column i-vector).
There exist (after shrinking pO)

-- 1 l I· •
(v(i • I
veil)
f, (
I::.
r (A (0
P) I (1 ~ i ~ m)
 - 3 76-

Y-T .Siu

such that the germs of y(l) ••••• y(m) at 0 generate

Ker f . There exist (after shrinking pO)

j
u i ) ~ r (A( pO), n(9)
(l ~ i ~ m, 1 ~ j £)
<;i(
~
K(i) ~ pO)

such that

(1 ~ i ~ mj •

Let 1': n(!)m -> oJ,

n
on A(pO) be defined by the matrix
( y(i})
j l~i~m, l~j~£·

Now. take .; ~ Et: (f) wi th 6~ ~ = ° such that the

image of .; in 'N~+ I ,0 belongs to A. Let e "" 11.;/1:. f
By (III), for p sufficiently strictly small, there exist

b ~ r(A(p), neD)
i

n ~ <;i(f)
such that

Ihi If *
C-e
r
lin 11~;i.f *
C_e
f
 --37-7-

Y -T . Siu

where c~ is a constant depending on f Since the image

v
of E. in 'N-a+l,o belongs to A, it follows that

By (1.6), for p sufficiently strictly small, A(p) is a

(Coker p ) -privileged neighborhood of 0 in the sense of

Grauert. Hence, for such f ' there exist

(1 ~ i :i- m)

such that

~ c .(vl(i), ••• , v(i»)

i""l a 1,

where c~ is a .co ns t a n t depending on f. It follows that

k
t ( ~ c u(j»)r-;
- l j i a+l,a
~(i) + 611 - 1( ~ .4l:(j) +
a+1 j""l J
c n) .
j
i-I

This concludes the proof of ( B)" • Q. E. D.

n

Notice that, if tn is not a zero-divisor of

(Coker ~)O ' then all the constants involved in the proof are
independent of fn •

(7.3) Observe that, by the open mapping theorem,

(F) " ===* (B)O.
)1
ltbreover, by the condition of factorization

through a Hilbert space given in "

(F) ,in the statement of
 - 378 -

Y - T . Siu

{B)O " we can choose a

(i)
and ? so that the map

is linear over ([ [t] •

By induction on n , it follows from (7.2 ) that

11
(E) , (M)HI, (Ff ===} (B)" • Under the assumption of
n
~
HI 11
( E) (M) ,and (F) , we are going to show, by induction

on n that the natural map £) from

to

is an isomorphism. We can assume without loss of generality

that to ~ O. The case n = 0 follows immediat~ly from

II ~
(F) • The surjectivity of £) follows from (E) • Suppose
e(~*) ~ 0 for some

y
, t.,* is the image of some C, ~ -aI-(l_l,O
~
By ( E) Take arbi-
. n
trarily m ~ l. Let d m = (CD, ••• , CD,m) ~~ *. Let
-l, ~ 9f(l_l [dm.]O be
~
the ima ge of~. Since the image of C, in
":I.')) m
l'l"~+l [d]O is 0 by induction hypothesis the image of ~
II m
in P:t(l [d]O is 0 From the cohomology sequence
 _ 37 9 -

Y-T .Siu

of the short exact sequence

°-> (O(~)o ->(Q(~)O -> (Q(()O(d m) ->

a-
0

(where o: is defined by multiplication by t~), it follows

th&t the image of t, "

in . Wa. ,0
m v
belongs to tn'H a. ,0· Hence

Since Im(~a.oJ,0 -> ata.+l

"
,0) is finitely generated over l)o'
it follows from the arbitrariness of m, that ~* = °.
From now on, whenever (E)" ,(M) v+l ,and (F/, are
satisfied, we denote ':I./~
Im(~a.
'11"
--->wa.+l) by ':1./".
~ ~ "is in-
dependent of the choice of a. •

(7.4) Let us investigate under which circumstances, in the

statement of (Brn , we can choose a(i) and ? so that the

¥-p
C, f--> ( a (1) , ••• , a (k ) , 7 )

is linear over a: [t]. Looking at the proof of (7.2), one

easily sees that this is the case if

i)
~
(B) n- I
.
has the correspond1ng property of
[[tl, ••• , tn_l]-linearity for the n_IOO-submodule
" )j -oJ
Im(Ha._l -> ~-l) of "a.-l,O •
 - 3.80 -

V -T o Si u

ii) Coker Cf is locally free at o.

For 1 ~ t ~ n let

1, ••• , 1)
'----y--J
n-t
Denote

Im(~[d(t)] ->~+l[d(.e)])

by ~~[d(t)] which by (7.3) is independent of the choice of

~ .
y
The above condition ii) is satisfied if A = Ha + 1 •
By induction on n , condition ii) is satisfied if the nat-
~ (.l ~ ( .
ural maps from 9f [d )]0 to 9:1- [d £-1)]0 are surjective
for 1 < l ~ n. (A by-product of this surjectivity condition
is that all the constants i n the proof of (7.2) are indepen-
dent of p , because of the last sentence of (7.2)). From
the exact sequence

(where 0- is defined 'by multiplication by tt)' we obtain

the exact sequence

->~:(d(t)) ->~:(d(l))

->1J~(d(1-1)) ->Pirl(d (t)) _> •••

Hence the surjectivity condition just mentioned is satisfied

if p.j~ = 0 for v< p. < ~ + n. Of course, in general , this
 - 381-

Y - T . Siu

last condition is not satisfied. Under the assumption of

(E)~, (M)~, (F)~ for certain ~'s, we are going to modify
the complexes so that the new complexes satisfy this last
condition.

v P
(7.5) Suppose L ",,6x ([ ~ and
-;

->L
-;-1 )1" v+l - >
->L ->L •••

is a complex in which the maps of (holomorphic) bundle-homo-

morphisms. Let

be a commutative diagram of bundle-homomorphisms such that

for a < ~ •

Note that, to have ~: ' it suffices to have and then

V
define ~~
a as r aa ~~
a Let the complex
l l

be the mapping cone of the above commutat ive diagram; that is,
Ny
Ea '"

and
 - 38 2 -

Y - T .Si u

is given by

Define

by

We are going to prove the following three statements

(after a possible replacement of .6. by any open polydisc re-
latively compact in ~).

(M/+ l
a) The complexes Et: satisfy ( E{' and ~ the
comp.Lexas
--.J

Ft: satisfy ( E)" .

b) The complexes ( satisfy (M)" :::=::} the complexes
"'.
~ satisfy (M)" .
c) The complexes ( satisfy (Ft ::::::::} the complexes
N.

Fu satisfy ( F)" .
Statement c) is clear. Let us prove b): Suppose

and
 _ 383 ~

Y-T .Siu

for some l, ED f
, ,
(; ~- (t,d'f)
N~ 1 ° and a < 13. From the defi-
, N~ "'''-1
nltion of 6a ,6 we have
13

i) 6a
~
s; + ( -1 )
)/+1 )1+1
O""a f = °
.• )I 1 ' ~ ~ I
ii) r;aE. .. 6 - ~
13
+ (-1) (J 13f

,
a
~

iii) f .. t

It follows from ~ rr~ .. 0"")1+ 1

6a a a
cf and i) , iii) that

°.
Y v+l ~ ,
6 ( E, + (-1)
a
o:a f )

By applying (Mt of the complexes

t
E to the equation (ob-

tained from ii) )

.• ,)+1 ~ , J,/-l~'
" (~+
r13a "'>
(-1) 0""a f ) 6 '-,
13

we obtain

such that

(-1) )l + 1 ()a" f , ,) II
a,tO,d,p
,l, I 'II" °
-1
13,t ,d'f
) .
So we have
 - 38 4 -

Y -T. Siu

The requirement on the estimation of the norm of ~" w

lI> f'
, ,
~

the norms of .; lB f and .; E9 f is clearly satisfied.

Now we prove a) , Suppose a < p

N~ 0
c,EDf ~ Ep(t ,d,p) with 7,;(E,ED f) 0, i.e.

6'"J 1; + (_l)H 10-;+ If

{ p '" 0

all+lf = 0

From the first equation, we obtain

~ c:?'+ 1 y v
rp ,a-l a-I f '" c5 p (-1) C,

~+l •
By applying property (M) of the complexes E
t to this
)1+1 "11+1 )1+1::\Y+l
equation, (since 6a_lCT"a_lf '" CT"a_l Q f = 0), we obtain

such that
y )1+ 1 ~ ,
r
a,a-lCT"a-I f = 6 >
a~

It . follows that

By applying property (Et of the complexes E( to

)1+ 1" ,
l, + (-1) r pa t, ,we 0 btain
 - 385 -

Y- T . Siu

with

such that

'!hen
, 0
(f, +f,) (9 f c;: "'~
Eh(t ,d'f) with '6:((E,' + l,) EB f)= 0

?$O c;: i;-l

+1
l,
r~+ 1 ,a (l, + ~)$ f) + 6;~f(7$
",11 ~11 ('
0) •
r~+l,~( Sf)
,
The requirement on the estimation of the norms of (l, + l, ) Q) f
and ? EB 0 by the norm of C; $ f is clearly satisfied.

(7.6) Proposition. Suppose, for p:;;: v :i- s , the properties

(Et, (M))I+l, (F)v hold for the complexes ~. Then there
exists a complex

of trivial vector bundles of finite rank on ~(fO) (where

pO c;: IR~ and A( fO) C A) in which the maps are holomorphic
bundle-homomorphisms and there exists a commutative diagram
 - 386 -

Y -T .Siu

x:
()~ / \ 0"'~+1
\} ~

~ -> (+1
r~+1,a

of complex-homomorphisms on A{fO) such that the mapping

of 0-" satisfy
- a

for p ~ >J ~ s •

Proof. We are going to prove by descending induction on p

that, for p ~ p ~ s + 1 , there exist a complex

of trivial vector bundles of finite rank on some ~(fO ) and

a commutative diagram

N.
0"""
such that the mapping cones pEb, of pa satisfy

for P ~ V~ s "
 - 387 -

Y-T . Si u

The case # a s + 1 is trivial, because one Can set

#L• a 0. To go from the step # to the step # - 1 , we ob-
serve that, by (7.5), #E~ has properties (E)#-~(M)#, and
(F)#-l. By (7.2), #~ has property (B)~-l.

A: .. Im (H\I(CO(#~)o) -> H)I((9(#~+l)O))

is finitely generated over nmO. One can find (after
shrinking pO)

c,(l) , ... ,
whose images in A generate A over nCOO. Define

)/
LV
#_lL #
(# ~ ~ ~ s)
# 'i/
~
0
#-1
rr~
#-1 u
.. #
rr~
(l

Let

be defined by ~(l), ••• , ~(k). Define

( - 1 )·#-1 r au
#-11:'
1
1

1l-1
all - 1
 - 388 -

Y-T.Si u

Then the complex ~_lL· and the map

a-. satisfy the
~-la.
requirement. Q.E.D.
 - 38 9 -

Y - T . Si u

§s. Right Inverses of Coboundary Maps

As in § 7, suppose
v
... _~ ~-l _~ ~ 6a ~ <+1 _~

(a ~ a ~ (2) is a commutative diagram of trivial Banach

l
bundles and bundle-homomorphisms on A with 6,)6)1-1
a a =0 •
We use the same notations as in§7. Fix p. We assume that

i) the complexes ( satisfy (E))I, (M))I+l, and (Ft

for p ~ v :i- p + n •
~ . ~

11 ) Im(~,O -~ ~a+l,OJ .. 0 for p ~ v ~ p + nand

all a •

l
By the results of§7, (B)~ and (B)t hold with an addi-
tional statement of <C[t]-linearity. So, for p sufficient-
~
ly strictly small and for ~ .. p, p + 1 , we have maps !l!a
from

to ~;i(f) linear over a:[ t] such that

~
ra+l,a
~ .. v-I t)l E,
6a+ l a

11~:~lt:~,p ..
)l

< C II~/la 'f

 - 390 -

Y -1' . S iu

where C is a constant. It follows that, for l:, c;. ((p J

we have

By replacing LJ. by A( p) for some sufficiently strictly

small p ,we assume that ~~ and ~~+l are defined f or

p" (1, ••• ,1) and, from now on,2~ and 9?~+1 denote the

maps for that particular f·

(8.2) We are going to define

such that

(* )

on (0((-1) •

Fbr ~(tO, p) C D. and S c; ((to 'f) with power series

expansion

define

where ~~ is regarded as an element of ~(l, ••• , 1 ) . Be-

 - 391 -

Y -T. Siu

cause of the norm estimates for ~g+l' ~g+l, we have

'fg;(; r.g~~(tO ,P) •

To verify that this definition can g ive us a sheaf-

homomorphism, we have to prove its compatibility with re.

I , 0
s tr-Lc tdo ns- Suppose ~(t ,P)CC ~(t 'f). Let
f,' (; r.g (t r , f' ) be the restriction of z, Then we have the

power series expansion

,
E, =

Hence

It follows from the <C -linearity of ~g+ i ' ~g+ 1 , 6g and their
norm estimates that

, ,
equals the restriction of ~~ t, to ~(t ,r) .
It remains to verify the identity (*j. For

~(tO'f) CC Ll.., take

 _ 392 -

Y -T. Siu

Let

Then
oA
E(E
A 'I ~
~
A A-T, ,.
) (t -f t )

and, by definition of 1£ g ,

Since ~g+l' ~tl both are linear over ([[t] and have norm
estimates, it follows that

Hence

and (*) follows.

 - 3 93 -

Y-T . Si u

12· Global Isomorphism

(9.1) Suppose Eti is a sequence of complexes of trivial

Banach bundles as in §7. Fix two integers s ~ p. Assume
that, for p:;' oJ ~ S and a < P , the natural map from
II ..,
Im(~" -> ~Cl+l)
II
to Im (~p -> W- P+1) is an isomorphism.
Let

(p :;. v :;. s) •

Assume that 9:jP, NP+ 1 , ... , ,., :1/5 - 1

are coherent on LJ.. Let

&"-1
Im(C9(~-l) ~> (0«))
Ker((O(~) 6Cl> (O(Eh+ l ))

and, for any open polydisc Q C LJ., let

H~(.Q)

Suppose that there exists, for every Cl J a sheaf-homomor-

phism

such that
P- l
6a+2 ep
(l
=

We are going to prove the f ollowing two statements f or any

open polydisc n C A.
 - 3 94 -

Y - T . Siu

Let

Consider the following two statements.

1))1 If(S'2, E:) -> If (Sl, S:+V_P+2) has zero image for
k ~ 1 •
» ))
2)
)I
If(Sl, 'tal - > ff(n, ~a+)l-p+2) has zero image for
k ~ 1 •

First, let us show that 1) ==} 2) for p ~ v < s

» JJ
The commutative diagram with exact rows

0-> S"a -> ~)la -> ,#->0

."

(* )
)l

~
1 . ,..~
~ II
,)

o - > Sa+v_p+2 ->~a+}1-p+2 - > '#->0

yields the commutative diagram with exact rows

If ($1, e:) -> ff(Q, ~:) -> If (Q, W")

1 1
"
II
vi') .
i ...
ff (Q., S a+1I-p+2) - > Hk(Q, ~C1+V-P+2) - > ff (Q,

Since 9-i~ is coherent, ff(Q, W'll) = 0 for k ~ 1 • The

!result follows.
 - 399 -

Y-T. Siu

Next, we want to show that 2)~ ~ 1) for any «,

~+1
The commutative diagram with exact rows

0-> ->0

yields the commutative diagram with exact rows

If(.Q,(9(~)) -> If(Q,B~+l) -> If+1 (Q, 9.~

a)

1 1 /0-
'"
-> ~+1 (n, ~la+1I-p+3)
k(Q,<9«+1I_P+3 ~ 1>+1
H )) -> (S2,6a+1I -p+3 ) •

Ihe result follows from ~(.Q, (!)(E~+)I_P+3)) .. 0 [k S;' 1) and

the fact that 0- factors through the map

Now we are ready to prove i) and ii). '!he existence

0", e, together with the vanishing of ~(.Q.,(ry( (+2)) .. 0
"
for k;;;- 1 , implies that lip holds. Since 1)
II
==? 2)II
for p ~ y < sand 2)11 ===} 1),,1 for any ).l , we have 1))/
for p ~ ~ ~ sand 2 )~ for p ~ v < s. '!he diagram (*) s
yields the commutative diagram with exact rows
 - 3 96 -

Y - T . Si u

r(Q, ~~) -> r(.Q,W S ) -> W-(Q,S~)

,.. 1 T
II
S
1
(0, 't~+s-P+2) -> r(Q, '# ) -> Hl(Q, B~+s_p+2 )

By l)s, T is surjective and i) follo~s. Tb prove ii),

we have to distinguish between the case s = p and the case
s > p. The case s· p follows immediately from the exis-
tence of e~+l. Suppose s > p. The diagram (t)~;i
yields the commutatiye diagram with exact rows

->

Suppose an element t. of r(Q, ~~) is mapped to 0 in

r(Q, *s). Then

Since 1- factors through the map

it follows from 2)s_1 that there exists

such that
 - 39'7 -

Y -T. Siu

Hence ii) is proved.

(9.2) SUppose ~ is a sequence of complexes of trivial

Banach bundles as in §7. Fix two integers s ~ p • Assume
that the complexes ~ satisfy (Et, (M)lI+ l, and (Ft for
p ~ y ~ Max(s,p+n). By (7.6) (after replacing L::. by some
~(PO)) there exists a complex L' of trivial vector bundles
of finite rank on A and there exists a commutative diagram

of complex-homomorphisms on L::. such that the -mappi ng cones

~ of o-~ satisfy

for p ~ y :? s . By the results of §e, (after replacing ~

by some A(PO)) there exists a sheaf-hpmomorphism

such that

are as in (7.5).
 - 39 8 -

Y -T . Siu

For any object derived from E~ " we put a '" on top

of its symbol to denote the corresponding object derived from

~. For any open polydisc Q C A and for to ~ b. and
d ~N~ let

For to ~.6 and d ~ ~~ , '£~ induces a sheaf-homomorphism

8~(tO,d) from

~~(tO ,d): Im((O(E;~-I) (to ,d) -> (0(F&)(t O ,d))

to (0(~;~)(tO,d) such that

... 0
on 6~(t ,d). By applying (9.1) to the complexes of bundles
associated to ~(~}(tO,d) , we obtain the following. If
.... 0 .... I 0
~p[ t ,d], ••• , Ws - [t ,d] are coherent on b. for all
to ~ A and all d ~ ~~ with d
n
"! <Xl , then, for p ~ v ~ s ,

""v 0 ..." 0
i) ~ (SL,t ,d) -> r(Q., JH t ,d]) is surjective

0 .. " 0
ii)
...,~
Ker ( Ha, (Q.,t ,d) -> rtc, 'N [t ,d]))
0 0 )
Ker ( ~(Q,t,d)
",, ~ -~
C -> ~+)l_p+3(n,t ,d)

for any open polydisc Q C A and for all to ~ A and all

d ~N~ with dn f <Xl •

Since ~(~}(tO,d) is the mapping cone of

 - .399 -

Y -T . Siu

and f(Q, (9(~) (to ,d)) is the mapping cone of

we have the following two long exact sequences:

->~a.[t
11 °
...,~
,d]->9rJ.a.[t ,d] °
- > P\lHl((a(Ll (to ,d)) - > 9J:+ l [ to ,d] _>

-> I\z. (.Q,t

j) °,dl - > ""~fb. (Sl,t°,dl
- > l+l(r(Q, (!J(L) (to ,d))) - > th+l(Q,tO ,dl -> ...

From the first long exact sequence and (M)~+l (p ~ v < s) ,

it follows that

9>/-P(OCL)(tO,d)) ->~.P[tO,d] ->WP[tO,d]->

~p+l((D(r:)(tO,d)) _ > ••• _>,#s-l[tO,d] '_ >

w,s-l[tO,d] ->W-S(O(Ll(tO,d)) ->9f.s[tO,d]

is exact. From the sharp form of the Five-Lemma, we conclude

the following. If ~p[ to ,d], ••• , Ws [ to ,d] are coherent on
4 for all to G;: A and all d G;:f\ll ~ with d n" CD , then

i) for p ~ JJ v
~ s , ~(.Q,t ° y
,d) - > f(Q,W[t 0 ,d]) is
surjective

il) for p < v ~ s , Ker(~(.Q,tO,dl - > r(n,W[tO,d]))

C Ker(F{ (Q,tO ,d) - > ~+)I-P+3 (Q,tO ,dl) for any open
 - .4 0 0 -

Y -T . Siu

polydisc Q C A and for all t

o G;: A and all
d
n
=I CD •
 - 401 -

Y -T . Siu

§10. Proof of Coherence

(10.1) Suppose ~ is a sequence of complexes of trivial

Banach bundles ~s in §7. Fix two integers p, s such that
s ~ p + n. Assume that the complexes ~ satisfy (Et,
(M))l+l , (F)'J for p < ." =
= < s. We use t h e notations of '§ 9 .
We assume the following two conditions for p ~ )I ~ s , a, < ~,

to (; ,6. , d (; N ~ and for 8ny 0 pen polydisc .Q. C ,6. •

a) rm({(Q,tO,d) -> ~+l(n.,tO,d))

rm (~(Q,tO ,d) -> ~+l (Q,tO ,d))

b) Ker ( Ib.)l+1 (S1.,t ,d) ° -> )l+ 1

H~ (S1,t-0 ,d) )

C Ker ( fh)l+ 1 (Q,t ° .d ) -> ')1+ 1 (Q,t ,d)

Ha,+l ° ) •

~
We are going to prove by induction on n that ~ is coher-

ent on ,6. for P:i-'" < s. The case n = ° is trivial.

Assume n ~ 1. The induction hypothesis states that
91[ to ,d] is coherent for p ~ v< s , to (; b.., and d ~N~
with d ~ m. Since coherence is a local property, by (9·2)
n
w~thout loss of generality we can assume the following for
any open polydisc Q C.6 and for all to (; ,6. and all
d (; N~ with d n ~ co •

i) For p s»< v
s , Ha,(Sl,t ,d) ° v
-> r(S1., '9Ht ° ,d]j is
surjective.
 _ ~0 2 _

Y - T. Si u

ii) For p < y < s , Ker(~(Q,tO .d ) -> r(Q:~/[tO,d]l)

C Ker ( l\t~ ( Sl,t0 ,dl -> ~ (0)
'b.+)I-p+3 .Q.,t .d ) •

As in (7.3) we use the following notation. Fbr m ~ 1 , let

(10.2) Lemma.Suppose p ~ ~ < s and t n is not a zero-

m )1+1 'n
divisor for t n '1+0 • If f < P in lR+ and p is suffi-
ciently strictly small, then for to ~A(f) and A~~O

0 A+ m)
1m
r (
l\t~ ( .6.( f , ),t,d -> Het.;+ l (A(P),t0 ,d A)~)

is contained in

Proof. Consider the following commutative diagram

~+l(~(P))

I
H: (.a.( p' »

which comes from the commutative diagram

 - 403-

Y -T . Siu

°-~(Q(~+l) ...2....>(0(~+1) -~(!)(~+l)(tO,d>') -~ °

b AI r~+l ,0.1r,.. 11\1

°-~ (Q(() 2...> (D(~) -~ (0(~) (to ,d>.+m) -~ °

where

i) I- is the natural map

ii) a is defined by multiplication by (t n - t n )

°>.+m
m
iii) b is defined by multiplication by (t n - to)
n
x
iv) c is defined by multiplication by (t n _ to)
n

Let

1 :
~L\!+l
l'f"0
_
y >'f' °
..... ::l/.)1+l

be defined by multiplication by (t n - t)
n
m• ° Let

be induced by
v+l
r 0.+ I ,0. By (Bl +I
Y
n. (applied to A = 0), for

f sufficiently strictly small, Ker J C Ker g • Since t n

m ,#V+ I , i t follows that
is not a zero-divisor of
tn °
(* ) Ker 'f C Ker 1)'

for t~ = ° When t~ = ° ,both ~ and 1jJ are isomor-

phisms and, hence, (*) trivially holds.
 - 404 _

Y -T . Siu

One has

a
, t
===> gf't a ~ hr a

because h = gf. It follows that "te = a and Im eC Imo- •

The following lemma is a strengthened form of (10.2)

in codimension ~ 1. Its proof is similar to that of (10.2).
Its consequence (10.5) will be needed only for the proof of
the coherence of ~ p •

(10.3) Lemma. Suppose s ~ p+2 , 1 ~ P, < n ,

d,t+l' ... , d n <;: ~,nCC.6. is an open polydisc, and
to ~ .Q. Fbr A <;:N* let

eA = ••• , d )
n

Then there exists m <;: No such that, for A <;: ~,

Im (~(Q,tO ,e A+m) -> ~+3 (Q,tO ,e A)) is contained in

Im (~+3 (n,tO ,em) -> ~+3 (Q,tO ,eA.))

Proof. Since ¥P+l[tO,em] is coherent on D., by consider-

ing the increasing sequence of subsheaves consisting of the
kernels of the sheaf-homomorphisms 9:J-p+l[ to ,em] _> 'H P+ l [ to ,em]

defined by multiplication by (t - t~)m as m varies, we

1
conclude that there exists m ~ ~O such that t - tJ is
1
not a zero-divisor of (t.f - t~.)m W- p + l [ to ,em]x for x <; n.
 - 4.05 -

Y - T . Siu

Consider the following commutative diagram

p
HG:+3(Sl, t
° Clll o: p
,e ) -;> H~+3(.n, t
°~ "'l' p+l
,e ) ~ Ha.+3(Q, t
° Cll
,e )

1 B T f
i h

H~(Q,tO,eCO) ~ H~(n,tO,e~+m) ~ Hrlm.,tO,e ClO

) ~ Hfl(Q,t O, ella .

which comes from the commutative diagram

. °
o ~(O(E:u.+J)(t,eClD) -C4 (9(E.u+
3)
t ,eClll)
(0

I
I:- h.

b r~+J,~ I
O~<a(E~)(tO,eOO) ~> (9(E~)(tO,eClD) ~(9(E~)(tO,e~+m)---7'O

where

i) ~ is the natural map

1I,+m
ii) a is defined by multiplication by ( t£ - to)
1.
m
iii) b is defined by multiplication by (tt - to)
1
x
iV) c is defined by multiplication by (t _ to)
1 'i

Let

a m
be defined by multiplication by (t.( - tl, ) • Let
 - 406 _

Y-T. Si u

g.• fb:p+1 (,..,a,t °,elI) , ->

J ~+3
p+1 (n,t U,ern )

be induced by
p+l
r a + 3,a . By (10.1) ii) , Ker J C Ker g • By

the choice of m , Ker f C Ker'lf . The conclusion follows

from repeating verbatim the last paragraph of the proof of
(10.2) •

(10.4) Lemma. Suppose s" p+2 , 0 ;;;;: 1 <n ,

d.e+ l' ... , d
n
<;: N , .Q C C A is an open polydisc, and to <;:Q.

For x <;: N* lI), ~,dl+l' ••• , d n) <;: ~~ • Then

let e"'· (lI), •• q
L ' ~(
for (f ••• , f.l) <;: N there exists (gl' ••• , g.t.) <;: l't
l,
such that, i f t, <;: ~ (Q, to and the image of C. in.e,
9tp [ t O, elI) ]
t
° belongs to f
1=1
(t
i
- t9)giS¥P[tO,elI)]
~ t
°' then

the image of ~ in ~+3.e (5"2, to ,elI)) belongs to

l
~ (t i - ti
'0 )fi ~3R
P (,.., lI)
~"t ,e )
°
i"l

Proof. We prove by induction on 1. • The case .£ = ° is

trivial. By (10.3) there exists m <;:N O such that

is contained in

Let gi." f + m. By induction hypothesis there exists

t
(gl' ••• , gl-1) <;:~.t-l such that, i f the image of ~ in
 _ 407 -

Y -T. Siu

9IlP[ to, e gl]

t
° belongs to

then the image ~ of ~ in P

~+3 <.e-l) (Q,t ,e gl ) ° belongs to

Observe that, if the image of ~ in ~p[tO,eOO]

t
° belongs to

then the image of l, ",p

in ~[t ,e
° gJ. ] ° belongs to
t

Hence

-1,= .l-l
E (t. _ t?) i ~.
f
i=l 1. 1. 1.

for some -l,i ~ f\i+ 3 C€ -1) (n, t °,e gl ) . By (10·3) the image of

l,i in P
~+3l(Q,t ,e ° ~ ) equals the image in P
~+3£(Q,t ,e ° ~)
of some l,i ~ f\i+ (Q,tO,eOOj (1 ~ i <.£J. It follows that
3t
.
the amage C; * s
of <-, in P
~+3i oJ"
(t"'l °
t ,e)
00 satisfies
 - 40 8 -

Y-T.Siu

~
* 1-1 a f.
_ E (t . _ t.) l~. ~
.... i=l 1 1 1

(l0.5) Lemma. If s ~ p+2 , then for p ~IR~ sufficiently

strictly small the following holds. Fbr tO~t:.(f) and
f~~n there exists g~Nn such that, i f !, ~ H~(A(f))
nag·
and the image of t, 'in ¥p0 belongs to t.) l'9lo'
E (t. -
t i=l 1 1 t

then the image of E, in ~+3n_2(L:!.(p)) belongs to

n a f.
E (t. - t]..) 1~+3n_2(~(f))
i-I 1

Proof. Since ~g+l is finitely generated over nOO ' there

exists m~No such that tn is not a zero-divisor for
t~ ,.,g+l. gn" f n + m. By (10.4) (applied to
Let
£ - n - 1), there exists (gl' ••• , gn-l) ~ ~n-l such that i f
the image of E, in ~p[ to ,d gn] a belongs to
t

then the image e; of E, in

p (
~+3 (n-l) A( f) ,ta,d gn) belongs

to

n-l a fi p ( a gn)
E (t. - til ~+3 (n-l) A(p) ,t .d •
i=l ].

Observe that, if the image of ~ belongs to

 - 409 -

Y - T . Si u

gn]
then the image of ~ in WP[ to ,d 0 belongs to
t

Hence,

- p ( 0 gn) By (10.2), for p

for some c;i ~ I-\i+3 (n-l) 06((') ,t ,d •
sufficiently strictly small, the image of ~i in
f
~+3n_2(A(P),tO,d n) equals the image in
f
~+3n_2(o6(f),tO,d n) of some ~i ~ ~+3n_2(.6(p)) • It

follows that the image t,* of l, in ~+3n_2(A(('))

satisfies

~
* n-l
_ E (t . _ t. )
0 f.
1~. ~
i=l 1 1 1

(10. 6) Theorem.
. coherent on
9/~ 1S ~ for p ~ p < s •

>I
Proof. It suffices to prove that 11 is coh erent at o.
We break up the proof into three parts.
~
(Ii We first show that 9J is of finite type at o.
 - 410 -

Y - T . Si u

Take e,
<f
It
< f
"'
in 1R~ and to <;: A( l) By

for f sufficiently strictly small,

Im (~(A( fit)) -> H:+ I (A( f' n) generates a finitely

(* ) {
generated r(A( p' J , n(O) -modu Le,

(()
Since 1/- 0~+l is finitely generated over nO" there exists
tm 'AL + 1
N lI
m <;:I~O such that tn is not a zero-divisor of n ~O •
By (10.2), for f " sufficiently strictly small,

'J rn 0 m+l) y 0 1 ~
f ) ,t -> F\t(A( P ) .e
It
Im ( Fb._l(A( ,d ,d )) is con-
(# ) {
'J " 01
Im ( lb:(A( f )) -> F\t(.6( P ) ,t ,d )
)l It
tained in

By (10.1) i), we have the surjectivity of

It follows from the coherence of W [to ,dm+l] and the

Theor em A of Cartan-Oka that

,
;.s surjective. From (#) and (t) we conclude that

Since W~o is finitely gener a t ed over nVO' by Nakayama's

t
Lemma,
 -4 11-

Y - T . Si u
v ,
It follows from (*) that W IA( p ) is generated by a fi nite
number of elements ~l' ••• , ~ k ~ ~ ( pI). Let
~i ~r(A(f''')~W") be induced by ~i (l:i- i < k) •

Next we prove that the relation sheaf 16 of

-, -,
~lIA( f ), ••• , ';kIA( f ) is coherent at, 0. We distinguish
,
between the Case v > p and the case y. p. Take f <f
in IR~ •

(II) The case ~ < p. Suppose

~ Rt °
°
k
for some to ~A(P ) ; that is Ea ·(';·)O=O. Let "M
i=l ~ ~ t t

denote the maximal ideal of n~

t
° as well as the ideal sheaf

on (n for the reduced subspace {to]. By taking the A

th

partial sum of the power series of a at to , we obtain

i
bi ~ r(4("" ), n<D) such that

(1 :i- i ~ k) •

It follows that

Since 9/-"[ to ,d A+ m] is coherent on A , by the Theorem B of

Cartan-Oka
 - 41 2 -

Y - T . Si u

equals

Hence the image of

i:&l
k -
~ b .~.
~ ~
in r (A(('
I t ), 9+-oJ[t 0 ,d ;\+ m])
belongs to

r(~(f" )'IWvO)I~(P
;\ ( " ),9Ht
II 0 ,d ;\+m)
] •
t

k
By (10.1) i) and ii), the image of ~ b.~ . in
i:al ~ ~

oJ (It
~+~-p+3 A(f ),t ,d
0 ;\+m)
belongs to the image of

r(A(p"), 1N>,A )
O
l (A(('") ,to ,d A+m)
t

in
II (It
~+II-p+3 .A(p ), t
0
,d
A+m) By (10.2), for
I
p suffi-
k
ciently strictly small, the image of ~ b.~ . in
i=l ~ ~

)l (lOA)
~+)I-P+3 A(p ),t ,d belongs to the image of

r(A( p' j, """AO) ~+l(~( pi))

t
II 'OA
in ~+)I_P+3(A( f ), t ,d ) . It follows that the image of
k ~ I
~ b. t,. in Ib.+1I_P+3(A( f)) belongs to
i"'l ~ ~

y
By ( B)n' f or f s uffici ently strictly s mall , th e image of
k k
~ b·S· equa l s th e i mag e of ~ c .z, in
i"'l ~ ~ i "'l 1. 1.
 - 413 -

Y -T . Si u

(1 :i i :i k) •

Since

and

it follows that

1t 0 C fWIIhO f!)kO + nifor(,6(,,), R) •

t tnt t I

Since h is arbitrary,

Hence R is coherent on ,6(P)

(IIIl The Case ~ = p. The only difference between this

case and the previous case is that, when ~ = p , (10.1) ii)
no longer implies that, for open polydisc n. C ~, to (;: ~
and d G;;:N~ with d
n
f !Xl ,

Ke r ( ~ (Q, to ,d) -> r(.Q, W"[ to ,d] l)

(**) { "0" 0
C Ker(%. (Q,t ,d) -> lb.+11_P+3(Q,t ,d)) .

When n = 1 , (**) is clearly satisfied for 'Y = p. So we

, n ~IP+l
can assume that n? 2 Since s? ~n is coherent
 - 414 _

Y-T . Siu

on b,. by (II). Now we modify the argument of (II) to avoid

the use of (**) when ~ = p. We pick up the argument of
(II) at the point where

k - "A p
L b.~. ~ r(A(f i. AJoIy 0W ) •
i=l J. 1 t

II
By (10.5), for p sufficiently strictly small, there exists
I
A depending on A such that
I
i) A -> co as A -> co
k p ("
ii) the image of E b .t,. in H<t+3n_2 ~(p )) belongs to
i=l 1 J.

r(~(f" )'N,\0)~+3n-2
A' P (It
A(f )) •

I
By (B)~, for p sufficiently strictly small, the image of
k y I k
i~lb/)i in fh+ 3n-2(A(f)) equals the image of L c . ~. in
i=l J. J.

~+3n-2(~(f')) for some

,As in (II), we conclude that

1<- 0 C !IN'v
A'
0 n(0 0
k
+ n(f)
k
Or\,~(
(
pI ), 'n)
II\" •
t t t t
,
Since A -> co as A -> co , it follows from the arbi-
trariness of A that

k I
I!) Or(~(p i,f?) •
n t
 - 4 15 _

Y - T . Si u

Hence R is coherent on ~(f ) •

(10.7) Proof of Main Theorem.

Fbr every So ~ S one can find a proper holomorphic

map o: with finite fibers from an open neighborhood U of
So into an open subset G of ~ n. We have

It is easy to see that an analytic sheaf ~ on U is coher-

ent if and only if RO~*~ is coherent on G. MOreover, if
M is an ~S ,s -module for some s ~ U , then

codh O M = codh lD M
S,s n a-(s)

where M is regarded naturally as an ncD0- ( s ) -modu.Le - In

particular, if codh lD M;;;- n , then M is a flat nWo-(s )-
S,s
module (see (A.$)-(A .12) of the Appendix). Hence for the
proof of the Main Theorem we can assume without loss of gen-
erality that S = A and '7 is n-flat. MO r eover , in the
course of the proof, any replacement of ~ by ~ (f ) (with
P~IR~) does not result in any loss of generality.
In (6.5) we have constructed a sequence of complexes

~ of trivial Banach bundles on A. By (5.2 ) and the re-

sults of Andreotti-Grauert [1] these complexes ~ satisfy

(El, (M)I+I, (Ft , and (l0.1) a), b) for p~ » < r - q - n,
By (l0 . 6 ), W" is coherent on b. . for p ~ )I < r - q - n_ L
 - 41 6 -

Y -T. Siu

By (7.3) and the bumpin~ techniques of Andreotti-Grauert 1

it is easy to see that

for p < r-q-n , a* ~ a < a# ' and

~ y b# < b < b*. MOre-
over for p < ~ < r-q-n ,
 - 417 _

Y - T . Si u

PART III APPLICATIONS

§ll. Coherent Sheaf Extension

For the definition and properties of gap-sheaves 7 [n]

used here, refer to the Appendix.

(11.1) Theorem (Coherent Sheaf Extension on Ring Domains).

Suppose 0 ~ a < b in JRN , D is an open subset of ([n , "J
is a coherent analytic sheaf on D X GN(a,b) such that
'][ntl] .. '1. Then "1 extends uniquely to a coherent anal-
ytic sheaf 7' .Q.U D x AN (b) such that '7 [ntl] = 7' .
...
Proof. The uniqueness of ? follows from the extension
theory of sections of gap-sheaves (see (A-le) of the Appen-
dix) •
...,
For the existence of ? , we consider first the spe-
cial case where

i) D is bounded and Stein

11) codh? S? n + 3

iii) ~ is flat with respect to the natural projection

n: D X GN(a,b) ---> D •

For m~ ~ sufficiently large, there exist

o< a < ~ in lR

a < a" < b < b in RN

 Y - T . Si u

such that

GN(a ' ,b)

, CC {z <;: <I: NI ex < EN 112m
z, < p} CC GN(a,b) •
i=l ~

For £ > a sufficiently small, D x GN (a ' ,b , ) is contained in

{(t,z) <;: DxceNlex + £ < £/t/ 2 + ~ Iz .1 2m < p-£} •

i =l ~

Let
,
ex = (l + 2"£

(l " = (l + £

, e
P P - 2"
P" P - £

£ltj2 + N
E 11
z,
2m •
~(t,z)
i=l ~

, ,
Take (l
< PI < ex and P < P2 < P . Let

x {(t ,z) <;: Dxa:Nlex < cy(t,z) < P}

Xi Xn{<p <PiJ (i =1,2)

n ~I x

rt
i n!X i (i = 1,2) .
Fix t
a <;: D • For any a a
(t , z ) <;: ~ - Xl ' there exists a
linear funct i on f(t, z) such that f(tO,zO) =a and f(t,z)
 - 41 9 -

Y - T . Siu

is nowhere zero on Xl - Let e: "} -> '1 on Xz be de-

fined by multiplication by f - Since f is nowhere zero on
Xl ' Supp Ker '8 and Supp Coker e are subvarieties of the
Stein space

and hence are Stein- From the cohomology sequence of the

short exact sequence
e
0-> Kere -> 7-> Ime -> 0 ,

we conclude that 8 induces an isomorphism

l".
., . (R1 n._)
2"7 0 -> ( R1 n.
2 (Im 8) ) 0-
t t

Consider the following exact sequence

1 2"7
(R n• .T) 0
t

coming from the short exact sequence

o -> Im e c-> "1 -> Coker e -> 0 •

We are going to prove that ~ is surjective- It suffices to

show that Ker 7 .. O. Consider the following commutative
diagram
 - 4~O -

Y -T . Siu

1 2 92
(R n;l) 0 ->
t

where 1
9 , e
2 are induced by and g is the restriction e
map- Since 82 .. 1 c.
and C, is an isomorphism it suffices
to show that Ker 9 2 .. 0 - Since is an isomorphism on e
Xl ,81 is an isomorphism- By applying the Main Theorem to

the map n: X -> D together with the function po and the

sheaf 7 on X, we conclude that g is an isomorphism-
Hence Ker e2 = 0 and C, is surjective- Since Supp CokerE
is Stein, the image of the natural map

(ROn~(Coker e)) 0 -> (Coker e) 0 0

t (t ,z )

generates (Coker 9) 0 0 over n+N({) 0 0 Since t,

(t ,z ) (t ,z )
is surjective, by Nakayama's Lemma, the image of the natural
map
•

generates "1 0 0 By letting zO vary, for some open

(t ,z )
f
neighborhood U of to in D we can find a sheaf-epimor-
phism (J :n+P -> t on
 - 421 -

Y -T. Siu

By applying the Same argument to Ker c: instead of 7 , for

some open neighborhood U" of to in U' we can find a
sheaf-epimorphism

-r : n+~ -> Ker 0-

on

{(t,z) ~ U" xa: Nj"

a < 1(t,z) < P" } •

By Hartogs' Theorem, or extends to a sheaf-homomorphism

T: n+~q -> n+Nif

on f (t,z) G U• x a: N I l'(t , z) < {3'}. Then

Coker~ / )[n+l]
(
.. /O[n+Z]Cokerl:

extends '1 I U x GN (a' ,b)

fl , and, hence, extends "J I"
U)C GN (a, b ' )
(see (A.IS) of the Appendix). By the arbitrariness of to
and the uniqueness of extension, the special Case follows.

Fbr the general case we use induction on n. Let S

be the set of points of D x GN(a,b) where codh"J:i- n + Z •
Let

TI .. {(t,z) ~ sjdim(t,Z)s n({t} x GN(a,b)) ? I} •

N
Let TZ be the set of points of DXG (a,b) where '1 is not
n-flat. Take a < a" < b < b in IR N • Let

By applying the special Case to '1 I U X GN (a ' ,b , ) for bounded

 - 422' -

Y-T , Siu

Stein open subsets U of D-A, we conclude that

"11 (D-A) X aN(a,b) can be extended to a coherent analytic
sheaf '1 on (D-A) x AN(b) satisfying 7[n+l] = :;. Since
dim S :f n , rank ~l Tl U T2 < n (cf (A. 1)) of the Appendix).
Since A = f! when n = 0 , the case n = 0 is proved. Take
arbitrarily t o ~ A. After a coordinates transformation, we
can assume that to = 0 and there exist 0 < a < ~ in ~
and '( > 0 n
in IR - l such that Lln- l (1') x al(a,~) is dis-
joint from A. By induction hypothesis, the sheaf on

1\
which agrees with 7 on

and agrees with ? on

can be extended to a coherent analytic sheaf 7* on

Lln- l (,t) x Al (~ ) x d (b) satisfying (,:1*) [n+ 1] = 1*. The gen-
eral case now follows from the arbitrariness of to and the
uniqueness of extension.
 - .4 2 .3 -

Y -T. Siu

§12 • Blow-downs

(12.1) A holomarphic map n: X ---> S is said to be strongly

I-pseudo convex if there exist a cf function
~: X ---> (~,c*) C (~,ooi and a real number c < c* such

that

i) n/[f:i- c} is proper for c < c*

ii) 'f is strongly l-pseudoconvex on [Cf> c#} •

(When the additional condition [SO:i- c} = [1' < c}- for

c < c* is added, this definition agrees with a special case
of strongly (p,q)-pseudoconvex-pseudoconcave maps.)

Fbr :r <;: r(X,lOx) and x <;: X let :rex) denote the

image of the germ of f at x under the natural map

Ox ,x -> <9x ,x/"~X ,x a: .

We are going to prove the following result concerning blowing
down. If' n: X ---> S is strongly I-pseudo convex and S
is Stein, then X is holomorphically convex (that is, for
every discrete sequence [x~} in X there exists
f <;: r(X,(!)X) such that f(x~) -> 00 as )l -> 00). Once we
have the holornorphic convexity of X, we can blow down X
by the Reduction Theorem of Remmert (whose generalization to
the unreduced case can be proved in a way analogous to the
reduced case [30J).
 - 424 -

Y -T . Siu

The result on blowing down will be proved by using the finite

generation of (Rln;Oxl s for s <;: Sand c# < c < c*. For
such a finite generation, it suffices to consider the case
where S = A and s '" o. Strictly speaking this finite
generation does not follow from the Main Theorem, because Ox
in general is not n-flat. However, this can be obtained
from the argument used in proving the Main Theorem. In the

proof of the Main Theorem, the ~latness is needed to get a

sequence of complexes ~ satisfying (E{, (Ml~+l, (F)
for p ~ ~ < r-q-n. Such a sequence of complexes is needed
for getting right inverses of coboundary maps (§8) and
global isomorphisms (99) which, in turn, are essential for
proving the coherence of the direct image sheaves under con-
sideration. However, when only the finite generation of the
stalks of the direct image sheaves under consideration is
needed, it can be proved directly by the arguments of §7
without using the sequence of complexes, 'provided that i is
n-flat on {~ < a#} . Of course, some modifications are
needed and, when » '" p , one can only obtain the finite gen-
eration of (R)l (n~i/J)o for a* < a < a# and b# < b < b* •
As before, we replace X by the graph of n • We prove the
finite generation by induction on n. By replacing X by
X~ with a* < a < a# and b# < b < b* ' we can assume with-
out loss of generality that there exists m <;:N O such that
tn is not a zero-divisor for any stalk of t~t. BYapply-
ing the induction hypothesis to ? It~ "] and considering the
exact sequence
 - 42 5 _

Y - T . Siu

we Can without loss of generality replace 7 by t~? and

assume that m = O. Use the notations of (6.5). Now, to
get the finite generation, one need only replace ~(f) by

and use the parenthetical statement at the end of (5.2). We

note that the case of the strongly l-pseudoconvex map corre-
spends to the case where p = 1 and { 1 < 8fI} = RJ. Hence
no flatness of ~X is needed for the finite generation of
(Rln;COx}s •

(12.2) Lemma. If n: X ---> S is a strongly l-pseudoconvex

holomorphic map and S is a single point, then X is holo-
morphically convex.

Proof. Take c# < c < c* and take a discrete sequence {x~}

in {T > c}. Let.J be the ideal-sheaf of the subvariety

{x~}. The exact sequence

yields the commutative diagram with exact rows

r(X,<Ox)
7 rtx, ~5) -> Hl(X,5) -> HI (X,<OX)
--->
0-1
~ 1
---> Hl(X c ,COX}
'l:
HI (Xc ,9)
 - 426 -

Y -T . Siu

Since S = CD
X on XC , 1: is an isomorphism. By the results

of Andreotti-Grauert [1], o: is an isomorphism. It follows

that ? is surjective. There exists f~ r(X,OX) such that
f'{x)/ ) -> (J) •

(12·3 ) Lemma· Suppose n: X -> S is a strongly I-pseudo-

convex map- Then for s c;: S and c# < c < c* there exists
k~N such that

(ROn;~)s -> (ROn; (lOX/.w.s,s(OX)) s

has the same image as

Proof. Use induction on dimsS. For some open neighborhood

U of s in S, there exists f c;: ,(U,lOS) such that
f(s) ,. 0 and fs is not a zero-divisor of ~S,s Since
l
(R n;1Js is finitely generated over ~S,s ' there exists

m ~NO such that ·

ii) fx (when f is naturally regarded as an element of

r(n-l(u),lOx» is not a zero-divisor of ~~ for
xc;:xcnn-l(U) •

The commutative diagram

 - 427 -

Y - T . Siu

o -> ?m+l(9 _> 19. _> 19 /?m+lCQ -> 0

X X X X

1 II 1
o -> f~ -> (Ox -> CDx/f~ -> a

yields the following commutative diagram

We are going to show that ba ~ O. Consider the following

commutative diagram

where

i) ~ and P are defined by multiplication by ~l

ii) e is induced by the inclusion map ~'9x C-> tOx

 - 428 _

Y -T. Siu

iii} 0- is defined by multiplication by ~

iv} ~ is defined by multiplication by f.

By the choice of m , ~ is an isomorphism and Ker a C KerO-.

It follows that

b(Ker c}

C b (Ker 'ta-e~ -l) b (Ker b) o.

Hence ba = 0 • It follows that 1m E, = 1m 7 .

By induction hypothesis, there exists kEN such that

and

have the same image. Then k satisfies the requirement.

(12.4 ) Lemma. Suppose n: x - > S is a strongly I-pseudo-

convex holomorphic map. s <;;: S , and K is a compact subset
of n-l(s} . Then there exists an open neighborhood G of
K in X such that (G,VxIG) is holomorphically convex.

Proof. yhoose c# < c < c* such that K C Xc. There

exists k <;;:N satisfying the condition of (12.3). By (12.2).
(n-l(s)'~x/~~,s~XJ is holomorphically convex· There exist
 - 429 -

Y - T . Siu

and an 0 pen neighborhood U of K in n-1( s) tI XC such

that K is contained in

By the choice of k , for some open neighborhood D of s

in S, there exist

such that fi and gi have the Same image in

r(n-l(s),~x/~s,s~x) • Let W be an open neighborhood of K
in XC such that ~ () n -1 (s) CU. There exists a Stein
open neighborhood Q of s in D such that the restriction
o'f rt to

is a proper map into Q. It follows that

is a holomorphically convex neighborhood of K in X. Q.E.D.

(12.5) Lemma. Suppose n: X-> S is a strongly I-pseudo-

convex holomorphic map, S is Stein, and c# < c < c*. Then
X ;s Stein if and only if for every compact subset K of XC
there exists a strongly plurisubharmonic function on an open
neighborhood of K.
 - 43lJ -

Y-T . Siu

E!:2.2.f. Only the "if" part requires a proof. S is the coun-

table union of relatively compact Stein open subsets Sk sucl
that Sk CC Sk+l and the restriction map

has dense image. It suffices to show that n-l(Sk) is Stein

for each k. Let Yk be a strongly plurisubharmonic exhau~

tion function on Sk. Take c# < e < b < a < c*. By

assumption there exists a strongly plurisubharmonic function

e on n -l( Sk+l) n Xa •
Choose a nonnegative C"..2
function
~ on X whose support is contained in X a and which is
identically 1 on Xb • There exists a if function ~ on
(_ro,c*) such that

i) Supp o: C (c#, c* )

ii) the first and second derivatives of (J are ~ 0 and

are > 0 on (e,c*)
__> ro as ;\ ->
iii) 0-(;\) c* froa the left.

For some positive number A, the function

is a strongly plurisubharmonic exhaustion function on

n -I( Sk). Q. E. D•

(12.6) Lemma. Suppose D is an open subset of ([n ,

[V J.. J.J. { I i s a locally finite open covering of D and
~
L.
J.
 - 4 31 -

Y -T. Siu

{i ~ II is a rf nonnegative funct ion on D with support

in Ui such that all first-order derivatives of ~i van ish

on the zero-set of l:: "to > 0 on D• Let 1)1 be
i ~ I ~
a strongly plurisubharmonic function on D. Suppose G is
an open subset of a: N and (Ii (1 ~ I) is a strongly pluri-

subharmonic function on U
i
x G. Let K be a compact subset
of DxG and let f: D x G--> D be the natural projection.

Then there exists Al ~ lR and a function 1\: lR --> IR ~

~
Mr.
l:: ("t • • fie ~ + B{Y· f)
i ~ I ~

is strongly plurisubharmonic on some open neighborhood of K

Proof. For a rf function h on an open subset .Q of a: m

and for x ~ Q and a ~~m ,let L(h;x,a) denote

m
l::
( a2h) (x)a.a_ .
i,j=ldZ .dZ'. ~J
~ J

and let a(h;x,a) denote

m
l:: -
ah (x/a .
i=ldz ~. i

Let ""
-r.~ '" '1:'. D f and Y''" '" 'Y' D f and let s be the unit sphere
~

in <e n+ N • It suffices to show that for fixed x ~ K and

, ,
a ~ S there exist A (x s a ] ~ IR and B (. ,x,a) ~ lR such tha~
 - 4 32 -

Y-T . Siu
, ,
if A? A (x,a) and B? B. (A,x,a) , then

Direct computation shows that

+ ].
The first bracketed term is > 0 when A > 0 When the
first n components of a are all zero, the second bracket-
ed term is o. When the first n components of a are not
all zero, there exists B* (.) ~ lR such that the second
bracketed term is > 0 when B? B*(Al. The third bracket-
ed term is at least as great as

where the only nonzero terms are those with ~i(X) t 0 and
O(<T'i ;x,a) to. Therefore there exists A* ~ lR · such that
J > 0 when A? A* Q.E.D.
 - 43 3 -

Y - T . Siu

(12.7J Lemma. Suppose n: X--> Sis a strongly l-oseudo·

convex holomorphic map and S is Stein. If every point of
S has an open neighborhood such that n-l(U) is Stein, then
X is Stein.

Proof. Let K be an arbitrary compact subset of X. By

(12.5), it suffices to prove the existence of a strongly plu-
risubharmonic function on some open neighborhood of K. By
embedding an open neighborhood of n(K} as a subspace of
~n , we can assume that S a (n. Cover n(K) by a finite
number of open balls Bi such that n-l(Bi) is Stein. For
each i , C hoose a nonnega ti ve coo f t ron
unc '
i on
Bi 'C

with compact support such that the first-order derivatives of

'C
i vanish whenever 'C
i vanishes and ET. > 0 on n(K) •
i ~

Since every point of K has an open neighborhood D which

can be embedded into (n+N by a (nonproper) holomorphic map
whose composite with the natural projection (n+N --> <c n
equals nlD, by using the fact that avery strongly plurisub-
harmonic function on a subspace can be extended locally to a
strongly plurisubharmonic function on the ambient space, we
conclude from (12.6) that, if )V (respectively ~i) is a
strongly plurisubharmonic function on S (respectively
n- l( Bi)), then

is strongly plurisubharmonic on an open neighborhood of K

for some A,B> 0 •
 _ 4 34 _

Y - T . Siu

(12.8) Theorem. If n: X ---> S is a strongly I-pseudo-

convex holomorphic map and S is Stein, then X is holo-
morphically convex.

Proof. By (12.4), every point s of S admits an open

neighborhood U such that

is holomorphically convex for some cH < c < c*. Let

7: G ---> R be the Remmert quotient of G (that is,

i) R is Stein
ii) ? is a proper surjective holomorphic map
iii) R07*<OG = (DR
iv )
?- l (x ) is connected for every x I~ R) •

Since no compact irreducible positive-dimensional subvariety

of G can intersect {~> cHl by virtue of the maximum
principle for strongly plurisubharmonic functions, ? mapS
Gn { rr> cH 1 biholomorphically onto its image. Hence we can
piece n-l(U) n {CJl> cHl and R through 7 and obtain a
c~mplex
'"
space
R. n induces a strongly l-pseudoconvex
,..,
holomprphic map R ---> U. Since there exists a strongly
plurisubharmonic function on R, by (12.4), R is Stein. By
the uniqueness of Remmert quotients, as s varies, we can
piece together all the Remmert quotients R together and
,
form a complex space X n induces a strongly l-pseudo-
, ,
convex map n : X ---> S. Since, for every s ~ S ,
 - 435 -

Y - T . Siu
'1
=
N ,
(n }- (U) R is Stein, by (12.7) X is Stein. Since
,
there is a pro per holomorphic map a- from X to X ,it
follows that X is holomorphically convex.

(12.9) Theorem. Suppose n ; X --> S is a stronglY

I-pseudoconvex holomorphic map and ? is a coherent analytic
sheaf on X. Then for v ~ 1

)I
i) R n* "} is coherent on S
Y
ii) R)l n* 1--> R n; '] is an isomorphicm for c# < c < c*

iii} HY(n-l(U),"j} -> r(U,R,)n*"J) is an isomorphism for

any Stein open subset U of S.

, ,
Proof. Letn, X ,~ be as in the proof of (12.8). Since
a- is proper, by Grauert's direct image theo~em (for the pro-
per case), for J.t ~ 0 , RJ.to; "J is coherent on X' and

,
is an isomorphism for any Stein open subset W of X
Since (j" maps { f > c# J biholomorphically onto its image,
for y ~ 1

By applying Grauert's direct image theorem (for the proper

case} to I lI
n' Supp R 0-* 1, it follows that, for »~ 1 ,

RO(n'}*(Rv~*1) is coherent and

 - 436 -

Y-T . Siu

for every Stein open subset U of S. It follows that

 - 437 -

Y- T. Siu

§13. Relative Exceptional Sets

(13·1) Suppose n: X ---> S is a holomorphic map. A sub-

variety A of X is said to be proper nowhere discrete
above S if nlA is proper and every fiber of nlA is pos-
itive-dimensional at any of its points. A subvariety A. of
X which is proper nowhere discrete above S is said to be
exceptional relative to S if there exists a commutative
diagram of holomorphic maps

i
X -> Y

s
such that

i) ~ is proper
ii) every fiber of ~1!(A) has dimension ~ 0

iii) ~ maps X-A biholomorphically onto y- (A)

o .
iv) R ~*(DX = COy •

The following result on relative exceptional sets is

a consequence of (12.8).

(13.2) Theorem. Suppose n: X --> S is a holomorphic map

and A is a subvariety of X which is proper nowhere dis-
crete above S. Then A is exceptional relative to S if
and only if for every s ~ S there exist an open neighborhood
 - 438 -

Y -T . Siu

U of s and an open neighborhood W of Atln-l(U) in

n-l(U) such that nlW: w---> U is strongly I-pseudoconvex
and Atln-l(U) is maximum among all subvarieties of W
which are proper nowhere discrete above U.
 - 1 $9 -

Y -T. Siu

§14. Projectivity Criterion

(14.1) Suppose n: X --> S is a proper holomorphic map and

p: V --> X is a holomorphic vector bundle. V is said to
be weakly negative relative to S if for every s ~ S .t here
exist an open neighborhood U of s and an open neighbor-
hood D of the zero-section of Vln-l(U) such that
no plD: D --> U is strongly I-pseudoconvex.

Let N be the zero-section of V and J be the

ideal-sheaf of N • We identify N with X and consider
Jk/Jk+" 1 as an Ox-sheaf. It is easy to see that, where V
is a line bundle, Jk/Jk+l equals (Q((L*)k) where L* is
the dual of L.

(14.2) Theorem. Suppose p: V ---> X is a weakly negative

holomorphic vector bundle relative to S. If "1 is a co-
herent analytic sheaf on X and K is a compact subset of
S , then there exists kO ~NO such that R~n*(7@~jk/jk+lJ
is zero on K for '}? 1 and k? kO

Proof. The case where S is a single point was proved by

Grauert [7]. The proof of the general case is completely
analogous to that of the special case. It fol lows fro m the
coherence of R')1 [rt 0 pi DJ* [p*']) and the fact that
 - HO _

Y-T. Siu

is a subsheaf of *
R~ (n. p I D)*(p "]) ,where )I ~ 1, r~
k \:.1'0'
and U,D are as in (14.1). Q.E.D.

(14.3) Theorem. If there exists a weakly negative holomor-

phic line bundle p: L ---> X relative to S and S is

Stein, then for every relatively compact open subset T of
S there exists a holomorphic embedding o: of n- l (T) into
PNx T such that the composite of ~ and the natural projec-
tion
--
PN x T ---> T equals
-
n.

Proof. Let l. = (0 (L* J . Let 52 be a relatively compact

Stein open neighborhood of ~ in S. Take two distinct
points x,y of X. Consider the two exact sequences

°->/Wo2X ,x'OIl. k ->W.X .xfl)£k ->

101
(ANY
/2
X ,x AMX ,x )~.c
k -> °,
where """'"X ,x,y = I#..X ,x {\ I\MX ,y • By (14.2), for k sufficient-

ly large, both Rln*('MI ,x,yfl} <Ivr k) and Rln.( ......... 2 ~J}) Van-
x ~',x x
ish on Q. It follows that the two maps

ROn*£k -> ROn* (((Ox/"Mx,x,y)~ £,k)

ROn*(w.,x,x~Lk) -> ROn* ((........x,x/......i,x)@.,(,k)

are both surjective on Q. Since Q is Stein, it follows

 Y -T. Si u

that there exist enough holomorphic sections of L* over

n-1(Q) to construc t an embedding ~ . Q. E. D.
 - 4 42 -

Y - T . Siu

§15. Extension of Complex Spaces

(15.1 ) Theorem. Suppose X is a complex space with

(O[n+l] = ~X and S is a Stein space of dimension ~ n •
X
Suppose n: X ---> S is a holomorphic map and
~: X ---> (a*,b*) C (~ro,m) is a strongly plurisubharmonic
function such that. nl la ~ ~ ~ b} is proper for
a* < a < b < b*. Then there exist uniquely a Stein space ""
X

and a holomorphic map n: X---> S such that

i) (9~n+l] .. (9X

ii) X is an open subset of X which intersects every

branch of X
iii} n = ~Ix

iv) the restriction of n to ""

X- l!f>a} is proper for

Proof (sketch). First we prove the case n = O. It follows

from ~~l] = Ox that, outside a subvariety of dimension ~0
,
in X, codh (Ox ~ 3. We can choose a* < a < b < b < b*
such that codh (Ox ~ 3 on I <p" a}. For a* ~ c < d ~ b*
let x~ .. Ic < f < d}. Let j be an arbitrary coherent
ideal-sheaf on X such that ~ • ~x on X~'. Consider
the following commutative diagram
 - 443 -

Y -T Siu

Hl(X~,,1) ->

a-IVI
;;
~(X~' ,,1) ->

coming from

o ->S -> Ox -> lDxlJ -> 0 •

By the results of Andreotti-Grauert [1], cr is an isomorphism.

Since J = CD
X on Xa
b' , 1: is an isomorphism. It follows
that e is an epimorphism. From the arbitrariness of , S
we conclude that there exist a holomorphic map

such that, for some 0 <a <p <i in IR N and some

, ,
a < a < b

i) f maps Cl(GN(a,p)) biholomorphically onto a sub-

space V of GN(a,Pl

ii) X:' c f-l(~N(a})

iii) x~, C f-l(G N(p ,n) .

~

By (11.1), V can be extended to a complex subspace V of

N
6. (P l satisfying (9~1] =([)r;. By gluinp; f-l(GN(a,f\l) and
,...,
V by means of f , we can piece together V and
N
C l (G (a, l')) U X
b
*
to form a complex space X. By (12·5 l ,
bl
X is Stein. X satisfies (0~xl] = (!)i and extends X
b
* By
bl
 - 44 4 -

Y -T . Siu

the property of gap-sheaves (cf. [A.16] of the Appendix), any

complex space satisfying these three conditions is isomorphic
to X and the ·i somor phi s m is unique. Choose arbitrarily
< b" < b, b"
a
,By repeating the preceding argument with
A
(instead of b ), we obtain a complex space X (instead of
X). Since X extends Xb* , in particular
bIt
X extends
b
X *
b'
and is therefore uniquely isomorphic to '"
X. From the arbi-
trariness of b" ,we conclude that X is an open subset
of '"X.

Now, consider the case n > o. The uniqueness of X

is a consequence of the properties of gap-sheaves (cf. [A.16]
of the Appendix). Because of (12.7), it suffices to prove
the existence of X locally (with respect to S). By local-
ly representing S as an analytic cover, we can assume with-
out loss of generality that S = A. First consider the
special Case where Ox is n-flat and codh ~X ~ n+) By
1 b 11'1
the finite generation of (R (na)*~X)s for a* < a < b < b*
and s ~ S , we use the methods of §12 and of the case n =0
""
to get the local existence of X.

Fbr the general case, as in the case of sheaf exten-

sion, we use a ring domain to avoid the bad set and appeal to
induction on n. MOre precisely, we choose 0 < ~ < ~ in
Rand Y ~ IRt 1 and a* < a < b < b*, such that

i) on n-I(An- I("() x Gl(~,~)) n X~ ~X is n-flat and

codh lOX ~ n+)
 - ·445 -

Y-T. Siu

ii) the extension n: X-> An-l(n x Gl(<l,~) of

n-l(An-l('t) x Gl(<l,~)) (\ x~ satisfies codh lOX ~ n+2 •

Now we apply the induction hypothesis to the holomorphic map

which is induced by the composites of the natural projection

cp-l(r) x d (~) -> A n- l (r) with n and with ~ •
 - 446 -

Y -T. Siu

APPENDIX

This appendix contains materials on homological c ~di­

mension, flatness, and gap-sheaves which are used in these

lecture notes. For more details, see [28].

(A.l) Suppose M is a finitely generated module over a

local ring (R,~) An M-seguence is a sequence
f
l,
... , f k ~ I\Nv such that fj is not a zero-divisor of

/~';'l < .
Pi .E fiM for
~"l
1
= J. ;i. k All maximal M-sequences have the

Same length. Define the homological codimension of Mover

R (denoted by codhRM or simply by codh Ml as the common
length of all maximal M-sequences. If S ---> R is an epi-
morphism of local rings, then codhSM = codhRM when M is
regarded as over S When R is regular of dimens ion n ,
codh M agrees with the maximum of n-l such that there
exists an exact sequence

o -> It - > ... -> II -> lO -> M -> 0 •

Suppose , is a coherent analytic sheaf on a complex

space X. We define codh"J as the function

x -> codh», 1...

I:IX,x X

Let Sk(7l den ote the set of points of X where

codh J. ;i. k •
 - 447 -

Y - T . Si u

(A.2) Lemma (Frenkel ). Suppos e D i s a Ste in doma in i n

,
D is a non empty Stein subdomai n of D. If
N .
o ~ a < b in R I then, f or 1 ~ ~ < N- l ,

This lemma is proved by Laurent series ex pansi on. For de~

tails, s ee [1, pp.217-219]. As a corollary, we hav e the

f ollowing.

(A·3i Proposition (Scheja). Suppose A is a subvariety of

dimension ~d in a complex space X and ? is a coherent
analytic sheaf on X with codh"l ;;;- r . Then ¥~A"J = 0 for
o ~ ).I < r-d . He nce
~
H (X,"] )
.
--->
~
H (X-A,?) is bij ective
for 0 ~ v < r-d- l and in j ective for )I = r-d-l •

We can assume without l oss of g ener a l i ty that X is

an 0 pen subs et of cI: n. When "J = n([) and A i s regular, i t
is a direct consequence of Frenke l's lemma, and, when the
,
singular set A of A is nonempty, it f ollows from the
long exact sequence

When "J r n'!) , we use a local finite free resolution of "J .

Now we define relative gap-sheaves with respect t o a
subvariety. Suppose A is a subvariety of a complex space
 - 448 _

Y -T. Siu

X and 1 C -1 are coherent analytic sheaves on X. Define

the sheaf 7 [A]1 by the presheaf

U ...-> {s~r(u,.~)I(sIU-A) ~r(U-A,7)}·

The following is a consequence of the Nullstellensatz.

(A.4) Proposition. If.5 is the ideal sheaf of A , then

Q)

1[A1 = U ("l:J k ) and is therefore coherent, where

-1 k=l 1
is the subsheaf of ~ whose stalk at x is the set
s ~~x such that J~s C tx •

(A·5 ) Proposition. is a subvariety of dimension < m-

=

Proof. We can assume without loss of generaltiy that 7 is

defined on an open subset of ~n. By taking a local finite
free resolution of ? and considering the rank of the matrix
defining the extreme left sheaf-homomorphism of the resolu-
tion, we see eas ily that Sm(?) is a subvariety. For the
dimension estimate, the Case m= a is obtained by consider-
ing ,Ia[ {x}]? for x ~ So (':1). The general case follows
from induction on m and considering the quotient of 7 by
the subsheaf generated by a holomorphic function whose germ
at some x ~ Sm(7) is not a zero-divisor of 'x

Now we define the dt h relative gap-sheaf. Suppose

iJ C 1. are coherent analytic sh eaves on a complex space X.
Define the subsheaf "}[dJ.~Of 1- by the presheaf
 - 449 -

Y -T. Si u

I
U 1--> {S c;: r(U ,$) (s] U-A) c;: r(U-A,':1) for some subvariety

A of U of dimension ~ dJ •

(A. 6) Proposition. "1[dJ~="1lSd(-§/7)J~ Hence '][dJ~ is

coherent and dim Supp(7[dJgI11 ~ d •

Proof. Follows from (A.3l and (A.5).

(A.7 l Lemma. Suppose "1 C.fj are coherent analytic sheaves

on a complex space X and x c;: X such that 7x is a pri-
mary submodule o f ~x whose radical P is of dimension d.

Then di mx Supp..gl1 ~ d and (1[d_l J§ lx = ?x •

Proof. For some open ne ighborhood U of x in X there

exists a coherent ideal sheaf J on U whose stalk at x
is P • There exists k c;: ~ such that pkg C ~
x
. Hence
Jk-lj C "J on some open neighborhood and

Let Y = su PP('7[d_ l J9 1:fl and l e t ! be the ideal sheaf of

Y. Since dim Y < d , there exists f c;: Jx-P. By the
Nullstellensatz,

f or s ome 1. c;: ~. Sin ce f Cl P , i t f ollows t hat

 - 4 50 -

Y - T . Si u

(A.B) Proposition. Suppose ? is a coherent a nal yt i c sheaf

on a complex space X, x (;. X , a nd f (;. r( X,(!)X) · Then fx
is not a zero~divisor of 7x i f and only if
dim x V(f) n Supp O[k] 7 < k f or a ll k (;. N* ' wher e
V(f ) = Supp <aX/ f t9x •

Proof. Suppose fx i s a zero-d ivisor. Then f x sx

° fo r
some s (;. re U,?) with Sx f °
, where U is an open ne igh-
borhood of x. Let k = dim x Supp s . Then
Supp s C V(f) () Supp e lk] 1 .
Suppose f is not a zero-div isor. By considering
x
the kernel of the sheaf-homomorphism 7 --->7 defined by
multiplication by f , we conc lude that, f or s ome op en neigh-
borhood U of x , fy is not a zero-divisor of 7y for
y~U. If dimxV(f ) n Supp 0l k]?= k f or some k , then,
for some open subset W of U ,

which, because of the Nullstellensatz, contradicts fy being

a non-zero-divisor of 7y for y (;. u •

(A·9) Proposition. Suppose 7 is a coherent analyt ic sheaf

on a complex space X. Then the d-dimensional component of
Supp Oed] ':1 equals the d-dimensi onal component of Sd (7) •
Equivalently, f or x (;. X , dim x Supp O[d ]1 < d i f and only

if dim x Sd ('1) < d •

 -451_

Y-T . Siu

Proof. We prove the equivalent statement. The "if" part

follows from

For the "only if" part, we assume that dim Supp O[d]7<d
x
and dimx Sd(t) = d. There exists an open subset U of
X - Supp Old] "] such that dim U" Sd ('I) = d. After replac- .
ing U by a smaller open subset, we Can assume that there
exists f ~ r(U,~) such that V(fj: = Supp (OX/fOX contains
UnSd (7 ) and V(f) does not contain any k-dimensional

branch of un Supp 0lk] "J for any k > d By (A.$) , f

y is
not a zero-divisor of "Jy for y~U. Hence

Sd ("1) n U = Sd_l (7/f"1) n U , contradicting dim U n Sd ("]) = d .

(A.IO) Corollary. 0ld]"]= ° i f and only i f

dim ~+l ('1) ~ k 42.£ k < d •

(A.II) Corollary. For

f ~ r(X,~X) , f is not a zero-di-
x
visor vf 7x i f and only i f dimx (Supp lOx/fC9x) Sk ("]) < k n
for k;;;- °.
(A.12) Suppose rt : X - > Y is a holomorphic map of complex

spaces and "] is a coherent analytic sheaf on X . 7- is

said t o be n-flat at x~X if tx is a flat Oy,n (X) -mod-

ule. t is said t o be n- flat on (or at) a subset G of X
 - 452 -

Y- T. Si u

if 7 is n-flat at every point of G.

When Y = 4:n , '7 is n-flat at x if and only if

to is not a zero-divisor for "JjjEl(t._t?j'Z, where
t r j x i=l 1. 1. x
(t~, ••• , t~) = n(xj •

(A.l) j Proposition. Suppose "] is a coherent analytic

sheaf on a complex space X and n ; X -> <en is a holomor-
phic map. Let Z be the set of all points of xG:X such
that 7x is not n-flat. Then Z is a subvariety of X
and the rank of nlZ is <n .
Proof. By (A. H) ,

Hence Z is a subvariety of X• Let Tk be the set of all

co
x G: Sk (1) such that rank x n I Sk (7') < n • Since Z C V Tk "
k=O
it follows that rank nlZ < n • Q. E. D.

Now we define absolute gap-sheaves. Suppose "1 is a

coherent analytic sheaf on a complex space X. The d t h

absolute gap-sheaf 7[d] of ? is defined by the pre sheaf

U t---> ind lim r( U-A,"1)

A G:Ol.d(Uj

where ~d(U) is the directed set of all subvarieties of U

of dimension ~ d •
 - 4 53 -

Y-T. Siu

(A.14) Proposition. Suppose 7 is a coherent analytic

sheaf on a complex space X. Then the following three con-
ditions are eguivalent.

i) 7[d] is a coherent on X.

ii) dim Supp O[d+I]7 ~ d •

iii) dim Sd+l (1) ~ d •

Proof. The equivalence of ii) and iii) is (A.9J.

To show iii) ~ i), we can assume that X is an

open subset of ~n and there exist exact sequences

(* ) o -> K _> n[/n-d-2 -> ...

PI Po
-> n(9 -> n(9 ->7-> 0

(** )
Il"Iql qo
-> n
v -> n<0 -> ~/n(f<:, (0)
nl:.! n
-> 0

on X. Let A = Sd+ I (1)· For a coherent analytic sheaf 1-

on X, let R~1 be the ~th direct image of ~lx-A under
the inclusion map X-A C-> X. By (A.3) and (*),

1[d] ::::: R~? ~ R~-d-2.k • By applying ~nO (., nO) to (**),

we obtain an exact sequence

qo ql
o -> J( -> n(0 -> n([) ->

on X-A, because K is locally free on X-A. It follows

 - 4;34 -

Y - T. Siu

from (A. 3) that

is coherent.

To show i) ~ ii ), we suppose dim Supp O[d+lJ,]= d+l

and are going to derive a contradiction. Let n be the unit
open I-disc. Without loss of generality we Can aSsume the
following.

X = nn

Let A = nd x 0 and let '1 be the zero th direct image of

O[d+lJ7Ind+lx O-A under the inclusion map

nd+1 x O-A C--> nd+1 x O. Since

'1 (O[d+l]7) [AJ(7[dJ)

~ is coherent. Since the sheaf-homomorphism ~ ---> ~ de-

fined by multiplication by t d+ I is 0 , it follows from
Nakayama's Lemma that 10 = 0 . Let {x» lV=l
co
be a sequence
in nd+lx O-A approaching 0 Since nd+lxo_A is Stein,
there exists

such that Sx f 0 f or al l V. It f ol l ows tha t s defi nes

 - 4 5 5-

Y -T.Siu

a non-zero element of 10 ' contradi cting 1 0 0

(A.15) Proposition. Suppose '7 is a coherent analytic

sheaf on a complex space X. Then th e natural s heaf- homo-
morph ism"] > 7 Ld ] i s an isomor phis m i f and only i f
dim 1k+2(7) ~ k for a ll k < d •

Proof. i) The "if" part. Us e induction on d. The case

d = -1 is trivial. Suppose A is a subvarie ty of d imension
~ d in an open subset U of X Since codh' ~ d+2 on
U - (Ansd+l('])) , i t follows from (A.3) that

Since dim Sd+l(7) ~ d-l , by i nduc t i on hypothesis

r(U,"]) ~ r(U-(Ansd+l(':t)), "]).

Hence "] ~ "J[d] •

ii) The "only if" part. From t he defi niti on of "J [d] , we
conclude that Old] 7' = 0 • By (A.14 ), dim Supp O[d+l]1 ~ d.

Hence 0[d+l]7= O. For x ~ X there exists f ~r( U,OX )

for some open neighborhood U of x such that Supp 0X/f~
contains Sd+l(?l but contains no k-dimensional branch of
U n Supp O[k] '] for any k > d+1. Let -§ = "]If"J. Then

O[d ]1 = 0 and

dim un Sk+2 (7) ~ dim 1k+l (~) ~ k

for k < d • Q. E. D.
 - 456 -

Y -T . Siu

(A.16) Proposition. Suppose a ~ a < b in (RN D is a

,
domain in q:n , and D is a nonempty open subset of D •

If "} is a coherent analytic sheaf on Dx~N(b) such that

1[n-l] :z ';I, then the restriction map

is an isomorphism.

Proof. For any open subset U of D, the restriction map

is injective, because the support V of any nonzero element

s of Ker a. would be a subvariety of UX 6,N (b) disjoint
from U x aN(a,b) and, by considering ({xJ x AN(b)) (\ V for
x ~ U , we conclude that dim V ~ n, contradicting O[nJo;= 0.
In particular, e is injective.

We are going to prove the surjectivity of e by in-

duction on n. Take s t;. r(D x aN(a,bl) U (D' x AN(b}) ,7).
We have to show that s t;. Im e•
Consider first the special case where codh 1? n+l
on D x ~ (b). Let.Q be the largest open subset of D
such that s!QXaN(a,bJ extends to an element of
r(Q x ~N (b) ,1). To show that Q is closed in D take an
,
element x of the boundary of Q in D Let pep be
,
nonempty open polydiscs in D such that x t;. P and P C Q.

There exists an exact sequence

 - 4 57 -

Y -T. Siu

Pl ·..PO
... -> n+ Nl[) -> n+ Nt} -> "1 -> 0

on It follows from

o (l :;i. v< N-l)

that P C Q • Hence.Q is closed in D and Q = D •

I
Now consider the general case. Take a < b < b in
IRN • Let n: D xllN(bj -> D be the natural projection.
Let A = n(Sn(1l n(DxAN(b'))) By the special case,
s/(D-A)xGN(a,blj can be extended to Sl ~ r(D-A)XllN(b ' ),7) .
To finish the proof, it suffices to show that, for any given
x ~ A , there exists an open neighborhood U of x in D
such that sIUxGN(a,b) extends to an element of
r( U x llN(b) ,"]). Since dim Sn (1) :;i. n-2 , we can assume with-
out loss of generality that x = 0 and there exist 0 < t
2
in IR n - 2 and 0 < (1 < 13 in lR such that
6 n-2 ('() x t:;P (13) C D and 6 n-2 (¥) x (a ,13) if
is disjoint from
A· By induction hypothesis, the restriction map

rc.b.n - 2 (n x .6.2 (13 ) x AN(bj ,']l ->

r((.Dr.n - 2 (n x i (13) x aN(a,b») U (An-2 (n x if (a,f3) x AN(bl) ,i)
is surjective. Hence the element of
 - 4 58 -

Y-T . Siu

with s' on .6,n-2(t) xcf(a.,~)X6N(b') can be extended t o an

(Ad7) Lemma. Suppose 7, -fj are coherent analytic sheaves

on a complex space X. If "1 = 7 [n] , then
(1Jr"..(9 (-§,'])) [n] = ~(!) (~' "]) •
X X

Proof. Because of the local nature, we can assume that there

exists an exact sequence

Ot -> (0~ ->.fj -> a

on X. Hence

is exact on X. The result follows from the fact that

lJ
'?k1tC(9 ({!)X,"J) is isomorphic to the direct sum of V copies
X
of '7. Q.E.D.

( A• I 0d ) Proposl.. t ao
. n- Su ppose a =< < a '< b
a = l.'n lRN , D l.' S

an open subset of a: n , and :t i is a coherent analytic sheaf

on DXaN(a,b) such that 1i = ':li [ n ] (i=1,2). Then
every sheaf-isoJllOrphism 71 -> '72 on D x GN(a' ,b) extends
uniquely to a sheaf-isomorphism 71 -> "12 £!! D x GN(a,b) •

l!££f. We can assume without loss of generality that

,
aj = aj for 2 ~ j ~ n. By (A.17) and (A.16), both re-
 - 4 59 -

Y -T. Siu

striction maps

are bijective.
 - 460-

Y - T . Siu

REFERENCES

[1] A. Andreotti and H. Grauert, "Theoremes de finitude

pour La cohomologie des aspace s complexes", Bull. Soc.
Math. France 90 (1962), 193-259.

[2] A. Ibuady, "Le probH:!me des modules pour les sous-

espaces analytiques compacts d'un espace analytique
donne", Ann. Inst . Fourier (Grenoble) 16 (1966),
1-95·

[3] O. Forster and K. Knorr, "Ein Beweis des Grauertschen

Bi1dgarbensatz nach Ideen von B. Ma1grange",
Manuscripta Math. i (1971), 19-44.

[4] , "Re1ativ-analytische R8.ume und die

Koharenz von Bildgarben", Invent. Math. 16 (1972) ,
113-160.

[5] J. Frisch and G. Guenot, "Prolongement de faisceaux

analytiques coher-ent s'", Invent. Math. 7 (1969) ,
3 21-343.
=
[6] H. Grauert, EinTheorem der ana1ytischen Garbentheorie
und die MDdu1ra.ume komp1exer Strukturen, I.H.E.S.
No. i (1960). Berichtigung. I . H. E. S. No. 16 (1963),
35-36.

[7] , "Uber MDdifikationen und exzeptionel1e

ana1ytische Mengen", Math. Ann. 146 (1962), 331-368.
 - 4 61 -

Y - T , Siu

[8] R. C. Gunning and H. Rossi, Analytic FUnctions of Sev-

eral Complex Variables, Englewood Cliffs, N.J.:
Prentice-Hall, 1965.

[9] C. Houzel, "Espaces analytiques relatifs et theoreme

de finitude", Preprint, Nice, 1972.

[10] R. Kiehl, "Variationen iiber- den Kohar-enz aat z von

Grauert", Pre print , Frankfurt, 1970.

[11] , "Relativ analytische Raume", Invent. Math.

16 (1972),40-112.

[12] and J. -L. Verdier, "Ein einfacher Beweis

des Koharenzsatzes von Grauert", Math. Ann'. 195
(1971), 24-50.

[13] K. Knorr, "Der Grauertsche Projektionssatz", Invent.

Math. 12 (1971), 118-172.

[14] _ _ _ _ _ , "Noch ein Theorem der analytischen Garben-

theorie", Preprint, Regensburg, 1970.

[15] and M. Schneider, "Relativexzeptionelle

analytische Mengen", Math. Ann. 193 (1971), 23 8-254.

[16] H.-S. Ling, "Extending families of pseudo con cave com-

plex spaces", Math. Ann. 204 (1973), 13-48.

[17] A. Markoe and H. Rossi, "Families of strongly pseudo-

convex manifolds", Proc. Coof. Several Complex Vari-
ab l e s (Park City, Ut a h , 1970) pp. 182-207, Lecture
No t e s in Math. 184, Be r l i n -He i d e l be rg - New York :
Spri ng er 197 1.
 - 4 62 -

Y -T. Siu

[18] R. Narasimhan, Grauert's Theorem on Direct Images of

Coherent Sheaves, seminaire de Mathematiques Super-
ieures, '1969, MOntreal: Universite de MOntreal, 1971.

[19] J.-P. Ramis and G. Ruget, "Residue et dualite", Pre-

print, Strasbourg, 1973.

[20] H. Rossi, "Attaching analytic spaces to an analytic

space along a pseudo concave boundary", Proc. Conf.
Complex Analysis (Minneapolis 1964), pp. 242-256,
Berlin-Heidelberg-New York: Springer, 1965.

[21J P. Siegfried, "Un theoreme de finitude pour les mor-

phismes q-convexes', Preprint, Geneva, 1972.

[22] Y.-T. Siu, "Extending coherent analytic sheaves", Ann.

of Math. 90 (1969), 108-143.

[23 ] _____ , " A Hartogs type extension theorem for co-

herent analytic sheaves", Ann. of Math • .21 (1971),
166-188.

[24] , "A pseudo concave generalization of Grau-

ert's direct image theorem I,II", Ann. Scuola Norm.
Sup. Pisa 24 (1970), 278-330; 439-489·

[25 ] _____ , "The l-convex generalization of Grauert's

direct image theorem", Math. Ann. 190 (1971), 203- 214.

[26] , "A pseudoconvex-pseudoconcave generaliza-

tion of Grauert's direct image theorem", Ann. Scuola
Norm. Sup. Pisa 25 (1972), 649-664.
 - 463 -

Y - T . Siu

[27J Y.-T. Siu, Techniques of Extension of Analytic Objects,

University of Paris VII Lecture Notes, 1972 (to appear
in the Marcel Dekker Lecture Notes Series).
[28] and G. Trautmann, Gap-sheaves and Extension
of Coherent Analytic Subsheaves, Lecture Notes in Math.
172, Ber l i n- He i de l berg- New York: Springer 1971.
[29] G. Tr a u t ma nn , "Ein Kontinuitatssatz fUr die Fortsetzung
koharenter analytischer Garben", Arch. Math. 18
(1967), 188-196.
[30] K.-W. Wiegmann, "Uber Quotienten holomorph-knonvexer
komplexer Raume", Math. Z. 97 (1967), 2.51-2.58.

Department of Mathematics
Yale University
New Haven, Connecticut 06.520
U.S.A.
EDITOR[ALE GRAFIC A-ROM A
CENTRO INTERNAZIONALE MATEMATICO ESTIVO
(C.I.M.E.)
INTERNATIONAL MATHEMATICAL SUMMER CENTER

VOLUMES ON CJ.M.E. SESSIONS

1960
I. Sistemi dinamicl e teoremi ergodici. (Les-
sons of P. R. Ha lmos, E . Hopf and J . L.
Ma ssera) $ 4.-

1961
1. Geometria del calcolo delle variazionl. -
Secon d volume. Direct ed by Prof . Enrico
Bompiani and V. V. Wagner . " 5.-
3. Onde superflciall, (Lessons of H . Brem-
m er, L. Fel sen, F. J . Zu cker, etc.) ,. 10.-

1962
3. Magnetofluidodlnamica. (Lessons of C.
Agostinelli, V. Ferrario , R. Nardini, A. G.
Pach olczy .and U Schmidt) ,. 7.-

1963
3. Proprleta di media e teoreml dl confron-
to In fisica matematica. (Lessons of B.
Col eman , J . Se rrin, H. Ziegler, C. Agosti-
nelli , D. Gr affi and G. Grioli) " 6.-

1964
1. Rel atlvlta generale. Dir ected by Prof. Car-
lo Cattan eo. (Lessons of L. Bel, J. Ehlers,
G. Carica to, G. Fer rarese and L Mariot) " 6.-
2. Dinamica dei gas rarefatti. Dir ected by
Prof. Carlo Ferr ari " 10.-
3. Equazloni differenziali non-llnearl. Direc-
ted by Prof. Guido Stampacchi a " 8.-
4. Questlonl dl analisl numerica: Directed
by Prof. Aida Ghizzetti ,. 10.-

1965
I. Non Linear Continuum Theories. Direc-
ted by Prof. C. Truesd ell . It 8.-
3. Mathematical Optimization in Economics.
Directed b y Prof. Bruno de Finetti ,,7.-
1966
I. Calculus of Variations, Classical and Mo-
dern. Direct ed by Prof. R. Conti » 8:-
2. Economia matematica. Directed by Prof.
Bruno de Fin etti . . . . . . It 4.-
3. Classi caratteristiche e questloni connes-
se. Directed by Prof. E . C. Ma rtinelli ,. 6.-
4. Some -aspects of diffusion theory. Direc-
ted by Prof. Pignedoli ,. 13.-
 1967
1. Modern Questions of Celestial Mechanics.
Directed by Prof. G. Colombo . $ 7.-
2. Numerical Analysis of Partial Differential
Equations. Directed by Prof. J. L. Lions » 12.-
3. Geometry of homogeneous bounded do-
mains. Directed by Prof. Vesentini » 8.-

1968
1. Controllability and Observabllity. Direc-
ted by Prof. G. Evangclisti » 8.-
2. Pseudo-differential Operators. Directed by
Prof. L. Nirenberg It 10.-
3. Aspects of Mathematical Logic. Directed
by Prof. E . Casari It 8.-
1969
1. Potential Theory. Directed by Prof. M.
Brelot " 8.-
2. Non-linear Continuum Theories in Mecha-
nics and Physics and their applications.
Directed by Prof. R. S. Rivlin . " 10.-
3. Questions on Algebraic Varieties. Directed
by Prof. E. Marchionna . » 10.-

1970
1. Relativistic Fluid Dynamics. Directed by
Prof. C. Cattaneo » 13.-
2. Theory of Group Representations and
Fourier Analysis. Directed by Prof. F.
Gherardelli » 10.-
3. Functional Equations and Inequalities. Di-
rected by Prof. B. Forte. » 13.-
4. Problems in non-linear Analysis. Directed
by Prof. G. Prodi » 16.-
1971
1. Stereodynamics. Directed by Prof. G.
Grioli » 10.-
2. Constructive Aspects of Functional Ana-
lysis. Directed by Prof. G. Geymonat.
(Being printed) » 25.-
3. Categories and Commutative Algebra.
Directed by Prof. P. Salmon » 10.-
1972
1. Non-linear Mechanics. Directed by Prof.
D. Graffi » 14.-
2. Finite geometric structures and their ap-
plications. Directed by Prof. A. Barlotti » 9.-
3. Geometric measure theory and minimal
surfaces. Directed by Prof. G. Bombieri. » 7.-
1973
1. Complex Analysis. Directed by Prof. F.
Ghcrardelli . . . . . . .. . » 14.-
2. New variational techniques in mathema-
EDIZIONI CREMONESE tical physics. Directed by Prof. G. Capriz
and Prof. G. Stampacchia » 10.-
Via della Croce, 17 3. Spectral Analysis. Directed by Prof. J.
00187 ROMA (Italia) Cecconi » 9.-