DIFFERENTIABLE

MANIFOLDS
AND
mUADRATIC FORMS
F. Hirzebruch and W. D. Neumann
Mathematisches Institut der Universitat Bonn
Bonn, Germany

and
S.S.Koh
Department of Mathematics
West Chester State College
West Chester, Pennsylvania

Appendix 11 by w. SCHARLAU

MARCEL DEKKER, INC.

New York

1971

These Lecture Notes in Mathematics are produced directly from the author's
typewritten notes. They are intended to make available to a wide audience new
developments in mathematical research and teaching that would normally be
restricted to the author's classes and associates.
The publishers feel that this series will provide rapid, wide distribution of
important material at a low price.

COPYRIGHT 1971 by MARCEL DEKKER, INC.

ALL RIGHTS RESERVED

No part of this work may be reproduced or utilized in any form or by

any means, electronic or mechanical, including photocopying, microfilm, and recording, or by any information storage and retrieval system,
without permission in writing from the publisher.

MARCEL DEKKER, INC.

95 Madison Avenue, New York, New York 10016

LIBRARY OF_CONGRESS CATALOG CARD NUMBER

ISBN NO. 0-8247-1318-4

PRINTED IN THE UNITED STATES OF AMERICA

70-176304

PREFACE

The original notes on "Differentiable manifolds and quadratic

forms" where written by Sebastian S. Koh on the basis of lectures and
seminars which I held in 1962 at Brandeis University and at the University of California in Berkeley. They were a fairly faithful version
of the actual lectures, which explains their informal character.
The Marcel Dekker publishing company suggested to me that these
notes appear in their Lecture Notes series. Although the material is
in some respects outdated, a demand for the notes still seems to
exist, so I decided to follow this suggestion.
W.D. Neumann accepted the job of updating the notes and giving
them a form which does justice to the rather more permanent character
of such a Lecture Notes series. This was done by reordering and rewriting a lot of the material and adding two appendices to indicate
developements in the field of differentiable manifolds (Appendix I)
and quadratic forms (Appendix 1I ). Apart from one or two further minor
additions, the notes have remained otherwise essentially unchanged
from the original mimeographed Berkeley version.
W.D. Neumann has, with his own initiative and ideas, brought the
task of editing these notes to a successful conclusion. I would like
to thank him for this effort, and also W. Scharlau for the composition
of Appendix 1I.
F. HIRZEBRUCH

Bonn, 30th June 1971

iii

COliTENTS

1.

Quadratic forms

2.

The Grothendieck ring

17

3.

Certain arithmetical properties of quadratic forms

25

4.

Integral unimodular quadratic forms

32

5.

Quadratic forms over 2Z ;

36

6.

The quadratic form of a 4k-dimensional manifold

42

7.

An application of Rohlin' s theorem, 1-'-- invariants

46

8.

Plumbing

56

9.

Complex manifolds of complex dimension 2

75

the genus of integral forms

10. A theorem of Kervaire and Milnor

References

Appendix I
References
Appendix lI, Grothendieck and Witt rings, by W. Scharlau

87
92

94
105
108

References

114

Author index

117

Subject index

119

1. QUADRATIC FORMS.

Let

A be an integral domain. A lattice over

is a finitely generated free unitary

A-lattice

(or A-lattice)

A-module. The terms base, rank,

etc., will be used in the usual way. In particular

considered as an

A itself can be

A-lattice of rank 1. The dual

V' .. Hom (V ,A) of an

A
A-lattice, and there is a bilinear pairing

V is again an

V'XV ~ A

de fined by

< x', x> : = x' (x) , called the Kronecker produc t.

We define a quadratic form

f= (f,V)

over

to be a symmetric

bilinear map
f: V

where

V is an

XV~

A,

A-lattice. To such a form

there corresponds a

linear map

f:

--+

V'

called the correlation associated with

f,

given by

< ~ (x) ,y > = f(x,y)

for all
lation
(1.2)

x,ye: V.
~

The form

is called non-degenerate if its corre-

is injective. This is equivalent to the condition

f(x,y)=O

for all

ye:V

=>

x=O.

2
Let

V'

dual base of
and

be a base in

(e i )1=1, _. ,r

and

d ~J
..

<ei,e j > =

characterised by

the

(e:).
1 ,_.,r
~
~=
for each

Then the coefficients of the representation

are given by ~ij = f(ei,e ) .

j
is caJ.led the matrix of f

Lo<
. .e~J
j

'P(e.)

~J

The symmetric matrix

M = M = (O(ij)
f
(e ) ,
i

with respect to the base

and its

rank, which'is independant of the base chosen, is called the

of

We shall only consider non-degenerate quadratic forms, so the

rank of a form

is the same as the rank of

will be called an

rank

(f,V)

column matrix, e.g.

where

form of

= (x1 '

to express each element of

...

,x )
r

denotes transposition, then

r-ary form (unary, binary, and so on).

(e )
i

I f we use the base

V.

and
f

= (Y1'

...

as a
'Yr)

is given by

f(x,y) = x t My =
If

(e.). 1
~

~=

,r

is another base for

applying the invertible matrix

P,

M=

tive to the new base is clearly

determinant

det f

of

to be

obtained from

M of

then the matrix

P~

detM ,
f

domain

then

of a form
Let

and

det f

= (g,W)

are said to be equivaJ.ent, written

ish an isomorphism
x,ye: V.

f = (f,V)

u: V~ W with

Such an isomorphism
(f,V)
and

rela-

is well deA.

over the same

f _ g,

if there ex-

f(x,y) = g(u(x),u(y

for all

is called an isometry. An isometry

to itself is called an automorph of

M
g

by

Hence if we define the

fined up to multiplication by the square of a unit in

Two quadratic forms

(e )
i

be the matrices of

and

f.
with respect to

3
some bases. Then clearly
LEMMA (1.3):

The quadratic forms

and

and only if there exists an invertible matrix

i.e.

and

Let
V1

f1

and

V S V
1
2

M
g

and

are congruent matrices.

f2

(e ,
1

the matrices

f11& f 2

... ,ek ;ek + 1 , ...

M, M , M ,
1
2

= ptMgP

M
f

IJ

is the quadratic form on

f _ f 1$ f 2 '
decomposes into

,e )
s

( :1

and

then

V1 SV 2 '

are related by

~J

we shall say that

f1

is given for

f, f , f2'
1

of

with

defined by

If a base

If

are equivalent if

be quadratic forms defined on the A-lattices

Their ~

V2 '

f1

splits off from

or that

Thus a form decomposes into unary

forms if and only if its matrix is equivalent to a diagonal matrix.

A quadratic form is called non-singular if its correlation

cp :

V ~ V' is an isomorphism. This means that the matrix

is invertible, or equivalently,

det f

is a unit in

A.

M of

If

is

a field, the properties of being non-singular and of being non-degenerate coincide. If the restriction of
is non-singular, we say that

on a sublattice

is non-singular on

V
1

V
1

of

Notice

that the sum of non-singular forms is again non-singular.

LEI'iMA (1.4):

If

which is non-singular on

is a quadratic form defined on

V1~V2

V ,
1

of

then the restriction

f1

to

4
V1

splits off from

~:

V1 61V ,
2

where

M1

f.
(e 1 , ,ek ; ek + , ,e )
1
s
is of the form

With respect to a base

the matrix

M of

is, by assumption, invertible. If

of

is the invertible

matrix
p

(: _~'Lt)

:;

then
ptMP =

r:

_LM- 1L t + N
1

which, in view of (1.3), proves the lemma.

11

From now on we shall assume that the integral domain

local domain , that is,

A has a unique maximal ideal

A is a
Then

1tf.

has the following trivial but useful properties.

(i).

The set

Ca). If

0<.,

A'"

fJ

of all units of

are uni ts and !

and -< + ~

are units and

,2
0(

A-m.

A coincides:.with
are non-units in
and

S'l

A then

o(p

are non-units.

The following theorem generalizes the classical diagonalization

theorem for symmetric matrices.
THOOREM (1.5):

Every non-singular qUadratic form

over a

local domain

A decomposes into unary and binary forms. If, in addi-

tion,

then

2 s A'"

decomposes into unary forms.

5
Proof:

Let

M = (D( .) be the matrix of

~J

some base. If a diagonal entry D(ii

with respect to

is a unit, the corresponding

unary form splits off by (1.4). We can thus assume that no diagonal
entry is in
0<1i e: A*

Since

A*.

for some

det f

is a unit,

OC'11 ~ A*

implies that

so we lose nothing by assuming 0(12 s A*

i,

But then the matrix

(1.6)

o(ii e:

~11~22

has determinant

~12

e: A*

m,

0(12

= 0(21 e: A* ,

and is hence invertible. The

first assertion now follows from (1.4) by a trivial induction. To

prove the second statement we need only show that the matrix (1.6) is
congruent to one whose diagonal entries are units, for we can then
decompose the corresponding binary form by (1.4). Indeed the as sumption

2 e: A*

shows that

is invertible, and

IJ

then gives the required congruence.

Consider now a quadratic form

that

f(a,a) =

0(

e: A* ,

then the linear map

u (x) :=
a

is an involution, that is
is an automorph of

f = (f,V)

u u

a a

2f(x,a)
0<

= id

If
u

ae:V

V--.V

is such
given by

a - x

It is easy to check that it

and leaves the element

fixed. Such

automorphs are also known as reflections.

LEMMA (1.7)

Let

2 e: A*.

suppose

f = (f,V)

x,y e: V

If

then there is an automorph

Since

a= x+y

In the first case

f1e h -

then

which interchanges

= -u a

e(

e: A*

and

= 2(f(x,x) + f(y,y

y.

e: A*,

a= x-y

or

and in the second case

2 e: A*

Suppose

be quadratic forms over

f 2 EP h

and

u = u

JJ

THEOREM (1.8) (Wi tt)

of

f(x,x) = f(y,y) =

must be satisfied by either

is the desired automorph.

and

satisfy

f(x-y,x-y) + f(x+y,x+y)

f(a,a) e: A*

the relation

be a quadratic form over

f1 -

with

and let

f1

non-singular. If

f2

This may be regarded as the "cancellation law" for forms.

Since

decomposes into unary forms by (1.5), we lose

nothing by assuming that

which

f1'

ely. Then

f2

and

W has rank

is an isometry

itself is unary. Let the lattices on

are defined be
1;

let

u: V1 EPW ~V2ew.

= (f &h)(w,w) '" (f EPh)(u(w),u(w,

1
2

V ,
1

= w

which carries
restriction of

The composite
w

vu

into
to

w
V
1

W respectiv-

and

be a base. By assumption there

Then

(f E9h)(w,w)
2

= h(w,w)

and this is a unit since

is non-singular. (1.7) gives an automorph

v(u(w

V
2

vu: V1 $W

of

f2&h

~V2EPW

It follows that

vu(V )
1

is the required isometry

=
h

with

is an isometry
;;;

V
2

so the

f 1 - f2

JI

We shall now investigate the classification (up to equivalence)

of non-singular quadratic forms in some simple cases. The simplest
case is obviously that any non-singular form decomposes into unary

7
forms. In view of theorem (1.5) we therefore state:
this section

A is a local ring with

in the rest of

2 e: A*

For notational convenience we denote the multiplicatively written

cyclic group of order
2
{ x J x e: A*

Consider the subgroup

n
of the multiplicative group of units A* of

by

A.

Clearly, an equivalence class of non-singular unary forms over

be identified with an element of
A*2

(e.g. if

= e ),

A*/A*2.

In particular, if

A can
A*

then the "diagonalization theorem" (1.5)

shows that non-singular quadratic forms over

A are classified by

their rank alone.

Assume now

and let the two cosets in

e:

of

A*.

A*/A*2

be represented by elements

Then by (1.5), any non-singular form

over

and

A is

equivalent to one of the form

(We follow the standard convention of denoting a form

pression for

f(x,x) ). Let

by the ex-

be another non-singular form over

and
2
+ e:y r+s

We may assume that

that

f..., g

k$,r.

By the cancellation theorem (1.8) we see

i f and only i f

In particular if

A = 1R,

we may take

(1.11) holds if and only if

e:"' -1,

m-s = r-k =

o.

and the relation

This gives us

Sylvester's law of inertia:

the number

does not depend on the diagonalization of

classify quadratic forms over

(called the ~ of
f.

f)

Clearly rank and index

JR.

Recall that the determinant of a form

is determined up to the

square of a unit. Hence for a non-singular form

determines a well defined element

DET f e: A*/A*2.

over

A,

det f

Now assume

satisfies in addition to (1.10) the condition:

the

matr~ces

an d

(10 0)
l'

are congruen t over

Then, with the notation as above,

f..., g

if and only if

is even;

(oe: 0e:)

DET f = DET g

in other words

THEOREM (1.13):

If

m- s

r-k

We have shown:

satisfies the additional conditions

(1.10) and (1.12), then two non-singular quadratic forms

and

are equivalent if and only if they have the same rank and same DET. 11
Condition (1.12) can be put in the more convenient form:

e:x + e:y

= 1

has a solution in

A.

Indeed, (1.12) ~ (1.12)'

is trivial, and conversely if

is a solution of (1.12)',

then

(x,y)

We have been working under the general assumption that

local domain in which

domains are now in order.

2

is such a domain (with

= (a,b)

is a

is a unit. Perhaps a few examples of such

Any field of characteristic different from

m=

{O}).

For instance

m,

a).

~,

b).

~,

C,

the fields of rational, real and complex numbers.

the field of p-adic numbers, where

is a (rational)

prime.

c).

IF

= ~p:iZ,

where

the finite field of rational integers modulo p ,

is an odd prime.

As examples of local domains which are not fields we have

d).

:iZ , Q{p) ,

the ring of p-adic integers and the ring of rational

p-adic integers (p

an odd prime).

We may regard

as

Q{p)

b ~ 0 (mod p )

which consists of all fractions

Q,

Notice that the maximal ideal tn

a/b

E:

:iZ

is generated by the single element

Q wi th

:iZ

E:

:iZ
p

in

In fact

principal ideal domain in which each ideal is of the form

one can talk of congruences modulo
tion that

pt

in

:iZ.
p

We make the conven-

:iZoo = Qeo = 1R.

The examples

= 1R

We now consider the case

cyclic group of even order

Legendre symbol

(qJp)

A= C

and

= IFp

(p an odd prime). Then

p-1,

for
(qJp)

E:

have already been deal t wi th.

A*/A*2 ~ C
2

so
A*

A*

is the

We define the

by

=1

(qJp) = -1
From (1.13) we have:

COROLLARY (1.15)
ratic forms
and only if

!3:.2.!?! :

and

If
g

A = IF

(p an odd prime), then two quad-

of the same rank over

A are equivalent if

(det f Jp) = (det g Jp)

We need only show that (1.12)' holds in

:iZ.
p

This

10
follows from the following lemma.
LEMMA (1.16)
in

If

0(

,fS

2
2
e: IF; then o<x + Py = 1

is soluble

[ 0, 1,

of

IF
p

Let
The maps

i, j :

H~

be the subset

IF

, (p-1) /2 }

lF
p

given by

j(y) = 1-Py2
are both injective. Hence their image sets have at least one element

IJ

in common, proving the lemma.

A = 1lIp

To carry the above results over to the case

, the ring

of p-adic integers, we need the following lemma.

LEMMA (1.17):

Let

f;; (f,V)

be a quadratic form over

not necessarily non-degenerate. If there exists an

c e: 111*

where
p is odd or
Proof:

p = 2,

x e: V

then there is an

(O<1}

Recall that a sequence

ges to a. limi t in

111

if

(0(

n+ 1 -

with

if

is odd and

This is meaningful, since

admissible even when

n -+ 0 as

0< )

f(x,x)

p = 2,

since

f(x,'X) = c

n-+

oD

(see e.g. Van

t

f(x,x) - c '" p u ,

3 if p = 2

Define

is invertible and the factor

t

~ 1.

1 2t 2
u f(x,x)

= i;P

according as

of p-adic numbers conver-

Der Waerden: Algebra I, 5te Auflage, p255). Put

where

x e: V with

w = 1 or 3

and

1lIp

Then

(mod pt+1 )

'21

is

11
By this means we construct a sequence
in a sublattice of rank
element

Sf

Let 111-

in

and

in

x= x 'x ,x ,
o 1 2

such that

f(x,x) = c .

p.

converges to an

1I

This proves the lemma.

221.

be the maximal ideal in

prime field of characteristic

[ xi

of elements

Then

the

If

JT : 221 --+ IF
p
p

is the projection, then for each

COROLLARY (1.18) :

Proof:
that the map
phism. For

(1.17), with

8,

For

0(

ITo(

is in IF
p

taken as the unary form

221*/221*2--+ IF*/lF*2 ~ C
induced by 1T
p
P
P P
2
p = 2 the same argument gives
2212/221 2 -

We have

implies

is an isomor-

(2Z/82Z)* / (22I/82Z) *2 ~ C xC ,
2
2
modulo

e: 221; ,

is an odd prime then

If

0(

and

(221/8221) *

where

221/8221

its group of units.

is the ring of integers

IJ

e: 221* (p odd) we can define the Legendre symbol by

p

(o<Jp) := (TTco(Jp)
Then the above argument shows that
Co( Jp) = -1

(o(Jp) = 1

if

is a square and

0(

otherwise.

A further corollary of (1.17) is:

COROLLARY (1.19)
soluble in

If

0(

,{1 e: 221;

(p odd) then o(x +{1Y

= 1

is

221
p

is soluble in
the corollary follows by (1. 17)

JI

2Z

by (1.16), so

12
As in the case of IF

(1.13) now yields:

COROLLARY (1.20):
of the same rank over
DET f

;; DET g ,

Two non- singular quadrati c forms

:iZ

that is,

(det f Jp) = (det g Jp)

other local rings. For the fields

.-

and

(p odd) are equivalent i f and only if

It is of interest to study when O(x + {Jx2 ;; 1

1

bol defined as follows:

for

0(,

P e:

JJ

has a solution in-

this leads to the Hilbert

p=2,3,5, ,or

sy~

p=oa

is soluble in

._{ .1

otherwise.

-1

We list some properties of Hilbert symbols that we will need later.

For the proofs of these properties and further properties see for instance B.W. Jones [11], p27ff.
1.

(~r2,pc2) p

2.

(o,,B)p

= (~,P) p

= (3,O<)p
(0<,(3) (0<,)$) = (O(,p~)
. p
p
p

(0<,-0<)

Remark:

=1

Property 1

states essentially that the Hilbert symbol il:\'

a map

and properties 2 and 3

state that this map is symmetric and "bilin-

ear". In fact if one writes

not hard to see that it is an

~/~2

F 2-lattice and the above map becomes

a non-degenerate quadratic form over

given by property

as an additive group, then it is

F2

in Jones (loc. cit.).

- the non-degeneracy is

13
The calculation of the Hilbert symbol is given by the following
properties:

5.

(o<,P)QD

= 1

unless 0<

6.

write

(0<,(3)2

Here

(2\0<)

and

0<

f3 in the form

= (-1 \ p)ab(o< 1 \ p) b( P1 ip)a

= (2\0( 1 )b(2\ fl 1 )a(_1)(o(1-1)(,s1-1)/4
is defined to be

is of course defined as
Observe that for

p 0<1 ' P 111 '

then

with 0(1 ' t1 e: :lZo;,

(o<,P)

fl are both negative.

and

p ~ 2,

(-1) (0<2 -1)/8;

or

-1

for

according as

the case

=b =0

if

p ~ 2

if

=2

x e: :lZo2'

(_1)x

x:; 0 or 1 (mod 2).

of property 6

is just

corollary (1.19). A final important property of the Hilbert symbol is

the Hilbert product formula:

7.

If

a, b e: Q*,

then

TT(a,b) = 1

(a, b) p

=1

for almost all primes

, where the product is over all

=2,3,5,

and

,00.

Our next goal is to define the Hasse-Minkowski symbol for a quadratic form over

through a diagonalization of its matrix. An inva-

riant definition will be given in 2.

Let

nal entries 0<1' ,O<r.

toC )

denote the diagonal matrix with diago-

A quadratic form is said to be diagonal if

its matrix has been diagonalized, i.e.

LEMMA (1.21)

field

K.

Then

Let

f,g

may

be carried into

is of the form

be equivalent diagonal forms over a

g

by successive applica-

tion of binary transformations such that at each stage the form remains
diagonal.

]4
~

Let

and

be represented by the matrices

,,er)

is by induction on

r.

The case

=2

M=

respectively. Our proof

is trivial. Using the cancel-

lation theorem (1.8) and the induction hypothesis we see that for
r> 2

it suffices to show that

diag("" l ' 'c(r)

diag(p1,l(2' '~r)

into a form

can be transformed

by binary transformations such that

at each stage the resulting form is diagonal.

First note that any permutation of the

can be realized by

~i's

such a sequence of binary transformations. Now by assumption there is

' 1 ar ma tr1X
'
a non-s1ngu

/31

= L~'f~1
i 1 1

=(

JD

W1' th

ij )

RtMR -_ H.

S'1nce

~ 0,

we may, by permuting the o(i's (and hence the

2
2
2
fij's) if necessary, assume that 0(1 f 11 'c(1f11+~2f21 '
L<><ifi1

are all non-zero. Now the binary transformation with matrix

U=

carries

M into

goes into
terms.

M'

UtMU =

diag{(o(/~1 + o(2f~1)' )(2,0(3' 'o(r)

= diag{(O<1f~1 + D(2f'~1) ,0(3'

,,2

0(1 = 0(1)11+

t><.,2

2>21

,..(2 =

~
~3'

~2)

M',

using f;1

'1

1n pace

=1

'>11

.~:tI.e

by permuting

det U = <>i'1f;1 + D(2~~1

U is non-singular since

now apply the same process to

,o<'r'

which

~ O.

We can

f21 = 1'31 '

'21'

0<

l'

p(

2'

Continued iteration finally leads to the desired matrix

Now let
f

be a quadratic form over

by (1.5). Let the matrix of

define

be

Mf

~.

We may diagonalize

= diag(c(1'

'~r)

We

15

(1.22)

c (f) :=

which is either
on

f,

or

:=
p (0<1' ,0<)
r

-1.

IT.~ < J.(0<.,0(.)

~
JP

We must show that

c (f)

and not on the particular diagonalization

Indeed, for
Clearly, for

,Pr)

is another diagonalization of

r =1

we define

c p (f) :=
1
.

c (f) = 1 or -1

r =2

or has not a solution;

depends only

H
f
f

That is, if

then we claim

and there is nothing to prove

f(x,x) = 1

according as

has

this property is obviously independant of the

diagonalization. For the general case observe that by properties

and

3 of the Hilbert symbols

By property 1 of Hilbert symbols and the fact that c (0<1' ,0()

is unchanged if we permute the 0<. 's

,0<)
r

is invariant

under binary transformations provided that the resulting form is

again diagonal. Our claim thus follows from (1.21).
The symbol

c (f)

is called the Hasse-Minkowski symbol.

COROLLARY (1.23)
\.

Let

and

f2

be quadratic forms over

ThEm

A quadratic form
over

f1

over

111 clP
p
1>'

since

1lIp

can always be regarded as a form

thus the symbol

c (f)
p

is also defined.

By (1.19) and the definition of the Hilbert symbol

LEMMA (1.24):
111
p

then

If

f
c (f)
P

is a non-singular quadratic form over

=1

11

16
c (f)

By the same token we may consider

f

is a quadratic form over

Q.

for

p = 2,3,5,

,00

if

The Hilbert product formula (property

7) of the Hilbert symbol gives

LEMMA (1.25):
then

c (f) = 1
p

If

is a non-degenerate quadratic form over

for almost all

ITc

(f)

2,3,5,

,00

p P

where

ranges over

and

=1
11

2.

Let

THE GROTHENDIECK RING.

B be a commutative semigroup with operation $ .

be the free abelian group with base

and

A the subgroup of

b 1 Eli1b

- b1 - b
2

B and operation denoted by

+,

generated by all elements of the form

(b ,b e: B).
1 2

Grothendieck group of

Let

The group

G(B):- F/A

is called the

B.

The obvious canonical map

j: B --;. G(B)
is a semigroup homomorphism and has the universal property:

any semi-

group homomorphism

h: B--;.G

uniquely over

that is, there is a unique group homomorphism

j i

of

B into an abelian group

factors

g: G(B) --;. G such that the diagram

commutes.
EXERCISE (2.1)

The natural map

j:

and only if the cancellation law holds in

B~

G(B)

is injective if

B.

The name "Grothendieck group" is in honour of Grothendieck, who

used the above construction to define a group

17

Kw(X)

for any algebraic

18

manifold

X,

starting from a semigroup of analytic sheaves over

Borel, Serre; Le th~oreme de Riemann-Roch (d' apres Groth-

(see e.g.

endieck), Bull. Math. Soc. Franc., 86(1958),97- 136). It was this example which first made the importance of the above construction so
apparant. A closely related example is the following.
EXAMPLE (2.2):

Let

X be a topological space and

the semi-

group of equivalence classes of real (respectively complex) vector

bundles over

X,

with Whitney sum as

usually denoted by

KO (X)

(resp.

e.

Then the group

G(B)

is

..

K(X. We shall not discuss these

examples further here (see for instance Atiyah, Hirzebruch; Riemann-Roch theorems for differentiable manifolds, Bull. A.M.S., 65 (1956) ,
276- 281).

As the reader will have guessed, the example which interests us

here is the following.
EXAMPLE (2.3);
FO(A)

Let

A be an integral domain and let r(A)

and

be respectively the sets of equivalence classes of non-degener-

ate and non-singular quadratic forms over

semigroups with respect to the sum
G(~ (A

and

G(~ 0

(A

A.

These are commutative

defined in 1. The groups

will be denoted by

G(A)

Recall that the cancellation law holds in

and

r o (A)

Go (A)
if

A is suit-

ably restricted (see (1.8. Thus by (2.1), one can expect the calculation of

Go(A)

to yield a complete classification of non-singular

quadratic forms over

if

A in such cases. Observe also that

A is a field.
The rank of quadratic forms induces a homomorphism

19
called the augmentation. If
A-lattices
form over

V1

and

V
2

V1 A V2

f1

and

f2

are forms defined on the

we define their product

f1f2

to be the

characterised by

This operation induces a ring structure in

G(A)

and

rk: G(A)~~

is then a ring homomorphism. This ring will be referred to as the

Grothendieck ring of quadratic forms over
apply of course equally well to
Remark:

A.

The above remarks

Go(A).

The groups of example (2.2) are also in a natural way

augmented rings, see for instance Atiyah - Hirzebruch, loc. ci t.

The ring
f = x2

form

e;(1) = 1 .
rk

G(A)

has a unit element

represented by the unary

e;: ~ --;. G(A)

There is a ring homomorphism

rkoe; = id,

Clearly

so if

(l(A)

defined by

denotes the kernel of

then the short exact sequence of rings

splits, and
(l (A)
o

G(A) = (l(A)

is the kernel of

Ell ~.

where

Similarly

rk: G (A) ~ ~
0

The determinant of forms induces a group homomorphism

DET:Go(A)~A*/A*2

DET
G (A)
o
DET

since

restricted to the subgroup

= (l
to

(A) E9 ~

(l (A)
o

DET(f E9f ) = DET(f )DET(f )

1
1
2
2
~

=Im E:

Since

in the representation

is trivial, it follows that the restriction of

is already an epimorphism. Hence we have a short

exact sequence of groups

20
where

L(A)

is the kernel of

DET

restricted to

(A)

This sequ-

ence does not always split, as is shown by the examples discussed below.
Suppose now that

A is a local domain and

cancellation law (1.8) holds in

r o (A),

2 e: A*.

so by (2.1)

Then the

11 (A) eG (A)
0

Since we can identify an equivalence class of non-singular unary forms

with an element of

A*/A*2,
G (A)
o

generates

we have

A*/A *2 eG (A).

additively by the "diagonalization theorem"

(1.5). It follows that the elements of the form

a E: A*/A*2

Go (A)

generate

EXAMPLES (2.5):
~:

G(m)

a-1 E: ~ (A)

with

additively (here and in the following

always denote operations in

and

Furthermore,

~ 1li51li

G (A) ).
o

as a group.

G(e)

~ 1li.

A*/A*2 = [ 1,-1} ~ C Every quadratic

2
form may be written uniquely as 0(+.1190<-(-1), so G(m)~1li51li as
If

A=m then

a group. Note that it follows that

by

(-1) - 1,

in

1li.

For

G(m) ~ 1li

as a group, generated

however the ring structure is not the usual structure

A= C

the assertion is trivial.

11

Our aim now is to use the above comments to calculate

p

G(~)

a rational prime.
If

(1.18),

is an odd prime, then

1li;/1li;2 ~ C

Now for any element of

able power of
( 1,e:,p,pe: }

is in

We hence know that

e:-1 ,p-1

and

pE:-1,

~/~2 ~ C2 x C2

Indeed, by

let its elements be represented by

~,
1lip ,

~(~)

[ 1,E:}

the product of this element with a suitso

~/~2

is represented by

is generated additively by the elements

and hence also by the elements

e:-1 ,p-1

and

21

= (s-1)(p-1).

(pe:-1) - (p-1) - (e:-1)

We must investigate the relations

between these elements.

LEHHA. (2.6):

For

,p

0<

<=>

(O<,P)p = 1
(o<,P)

=1

E:

~/~2

we have:

= 1 +o<.{$

c<+p

means that c::o<x + f3x

1
2

in

G(\)

=1

. has a solution in

\.

A simple calculation shows that this is equivalent to saying

that

(0(0

,...~l

(1
0
OO(P

and

Now by property
(e:lp)

= -1

and

modulo 4

are congruent, which proves the lemma.

6 of the Hilbert symbol, (s,e:) p 1 , (e:,p) p

(p,p)

Hence in

= (-1Ip) = 1

or -1

2s

s + p ~ 1 + e:p

For

2p ~ 2

in

G(\)

p:: 1 (mod 4)

this gives

2(e:-1)(p-1) = 0 , (s-1)(p-1) ~
( e:-1) - (p-1

G(~)

=2

e:+p ~ 1 + sp

2p = 2

=1 + 1

p = 1 or 3

piii3 (mod 4 )

2s 2

we have

G(\)

p=1 (mod 4 )

where

according as

11

= e:ps

-1 -1

= 1

o.

2(e:-1)
Also

=0

, 2(p-1)

=0

DET(s-1)(p-1) = DETe:p-1)-

Hence by the exact sequence (2.4)

111

ED

a(\)

-=

111

E9

111/2111
(s-1 )

E9

111/2111
(p-1)

If)

111/2111
(s-1)(p-1)

where the element under a group shows a generator of that group, and
the last summand

111/2111

is

L(\)

22
p" 3 (mod 4 )

For

G(\)

we have

111

<H\)

111

If)

111/2111
(e-1)

111/4111
(p-1)

Ell!

Indeed, property 3 of the Hil bert symbol shows that (ep,p)

(e,p) (p,p)

-2(p-1).

= 1.

Thus

Therefore

~(\)

is generated by

4(p-1) = -2(e-1)(p-1) = 0

Also

The element

2(p-1)

so

C ;
2

C
2

C
2

sent

C ,
2

and

represented by

by

(p-1)

(e-1)(p-1) =
(p-1)

has order

As we know,

{1,e,f,d'}.

respectively modulo

property 6 of the Hilbert symbol gives

alone.
in G(\)

1112/11122 ~
~~2 ~

Then

(1,e,S,d' ,2,2s,2f,2sS' }.

and

L(\) ~ 1li/21li

so

p = 2.

let it be represented by

and

(e-1)

L(\) ,

generates

Finally, we come to the case

C
2

whence

ep + p = 1 + s ,

We can repre-

8,

and then

(s,s)2-- 1 , (e'~)2=1 ,
.. _-/

(e,2)2=-1 , (\"~)2=1 , (~,2)2=-1 , (2,2)2=1.

we obtain (where

now denotes the element

e+~
2~

Writing

for

1+1 e G(~2) )

= 1 + ef
= 2

2t = 2

e + e15 = 1+t
e~

+ 1 = 1 + d~

Using an argument similar to that used before it follows that

....=

111
1

and

L(~) ~ 111/2111,

the reader.

111/2111 19

111/2111

(f -1)

(t-1)

generated by

2(e-1).

EP

111/4111
(e-1)

We leave the details to

23
We conclude this section by giving an invariant definition of the
Hasse-Minkowski symbol. For each finite prime

let

be the map defined by

c' (a) = a - DETa - (rk a)'1 + 1
P

L(\)

Identify

111/2111,

with

c (a)
p

For

p=oo

(i.e.

c'{a) = 0 or 1 ,

so

and define

= (_1)c~{a)

\=m),write

asG{m)

as

a=ot1+/X-{-1) ,

and define

c... (a)
We shall show that if

a = f

= { _1)0(- (o<- -

1)/2

is a quadratic form, then this definition

of

c (f)

coincides with the previous one.

If

is a finite prime then trivially

c'{fEPg) = f+g-DETfDETg-rkf -rkg +1
p

c'f+c'g = f-DETf-rkf +1+g-DETg-rkg +1

p
P
where we write

rk a

for

{rk a)1.

c' (fS g) - c' f - c' g

Hence

= DETf + DETg -

DETfDETg - 1

By lemma (2.6)
(DETf,DETg)
where

RHS

= {_1)RHS

is the right hand side of (2.7). Thus

(_1)c {feg) -cpf-cpg = (DETf,DETg)

that is,

24
(2.8)

c (f)c (g){DETf,DETg)
p
p
p

Since this newly defined

c (f)

coincides with the one defined in 1

for unary forms, and since every form over

decomposes,into unary

forms, the desired result follows by (2.8) and (1.23). For

p=

00

the

proof is similar; alternatively by direct computation from the definition of 1, using property 5 of the Hilbert symbol.
An immediate consequence of the above definition and the exact

sequences of this section is

COROLLARY (2.9): An element
pletely determined by

rk a

DET a

of
and

G{~)

c (a).
p

the equivalence class of a quadratic form over

~,

finite, is com-

In particular,
p

determined by its rank, DET, and Hasse-Minkowski symbol.

finite, is

11

3. CERTAIN ARITHMETICAL PROPERTIES OF QUADRATIC FORMS.

Let

AcB

be an inclusion of integral domains. Then clearly any

quadratic form over

can be considered as a quadratic form over B.

Formaly, this goes as follows. Consider

(f , V)

is a form over

A,

denote by

is a B-lattice whose rank over

A)

characterized by

write

for

fB

according as

fB

as an

A-module. If

the form over

(which

over

We shall simply.

when there is no possibility of confusion.

A

will be called integral, rational or

A = 111, Q or

m.

Let

be a real quadratic form

w~se matrix has been diagonalized according to (1.5), and let

and

V ~ AB

is the same as the rank of

..rB(XGllb,yGl/c):= bcf(x,y).

A quadratic form over

~

~+

denote the number of positive and negative diagonal entries

0<-

respectively. The integers

~+

~-

and

depend only on the form

by Sylvester's law of inertia, which we proved in 1.

Clearly
-r(f)

of

0(+

+0(-

= rk f ,

the rank of

f.

Define the signature

by

:=

By what we said above,

0(+

and

D(

0(

and hence also

0(

~(f)

ned also for integral and rational forms and forms over

are defi-

Q(p) = Qn 1lIp ,

the ring of rational p-adic integers.

~ quadratic form

f (x, x)

lE

0 (mod 2 )

f = (f, V)

for all

E:

over
V,

25

111 or

otherwise

1112 is called ~
f

is called odd.

if

26
Since

(3.1)

f(x+y,x+y)

= f(x,x)

+ f(y,y)+ 2f(x,y) ,

is even if and only if the diagonal entries in its matrix are all

even.
Consider a quadratic form

= (f,V)

determinant. Then there exists an element

over

1li or

we: V,

1li2 with odd

in general not

unique, with
f(x,x)
For if
w

e 1 , ... ,e

= ~ wi e i

f(x,w)

(mod 2 ), for all

is a base of

'

!I

and we write

xe:V.
x = 2:xiei

'

then

f(x,x)

(mod 2 )

and
f(x,w) = Lf(e.,e'>x.w .
i,j
1
J 1 J
Hence relation (3.2) is equivalent to

= 2:
f(ei,e.)w.
j
J J
and since
of

det f ~O (mod 2 ) ,

we can solve for the coefficients

modulo 2

Such an element
f.

(mod 2 ) ,

Clearly

e: V

will be called a characteristic element 01

is another characteristic element of

only if
~

for sOlle

z e: V.

= w+ 2z

One then has

f(~,'t.f)

= f(w,w) + 4f(w,z) + 4f(z,z),

if ant

27
so by (3.2)
f(w,w)

(mod 8 )

f(w,w)

mod 8 of course means modulo the ideal

Let

THIDREM (3.5):

2 1112

111 or

be a quadratic form over

1112

with odd determinant. Then

(3.6)

f(w,w)-r-detf +1!! 0

where

= rk f

~:

is the rank of f

Observe that

(3.6) is meaningful, since

fined up to the square of a unit, so

1li c 1li ,
2

Sinc e

2212 ,

a form over

1li

det f

modulo 4

=(f, V)

ment, and

be the given form. If

a = det f

where

det f

is de-

is well defined

can be considered as a form over

111
2

so it suffices to prove the theorem for

Let

a:x.

(mod4) ,

w= 1

is odd. Thus

=1

then

is the form

is a characteristic ele-

(3.6) is clearly satisfied.

Suppose now

is an even form of rank

r = 2.

Then

has

matrix
2a c )
( c 2b
with
w

=0

odd since

4ab + c

Thus

fi

f,

andsince

f(w,w) - r - det f + 1 =

+ 1 ;: 0 (mod 4)

(3.5) is proved for unary and for even binary forms.

How assume

ran r 1

c 2 ;:1 (mod4),

is odd. Thus

is a characteristic element of

o- 2 -

of

detf

,J

and
for

f = f 61 f2
1

is a decomposition of

respectively. If
i

=1,2,

then

w
i

= -161 w2

into forms of

is a characteristic element
is a characteristic element

28
of

f.

Hence
f(w,w) - r - det f + 1 =

f1(w1,w1)-r1-detf1+1 +
+f2(w2,w2)-r2-detf2+1 +

+ det f1 + det f2 - det f 1 -det f2 -1

But

det f1

and

det f2

are odd, so

= -(detf 1 -1)(detf 2 -1)

detf 1 +detf -detf -detf -1

1
2
2

= 0

(mod 4)

Hence, if we assume that the theorem holds for forms of rank

then it holds for

f.

<r

To complete the proof we need only show that

every form of odd determinant may be written as a sum of unary forms

and binary even forms with odd determinants_ This needs a slight modification of the proof of (1.5) and is left to the reader.
For

a e: 111 ,
2

we defined

is even or odd. For a form

(_1)a
over

as

or

-1

11

according as

1112 wi th odd determinant we can

thus define

C (f) := (-1) (f(w,w) - r - det f + 1)/4

2

Recall that the Hasse-Minkowski symbol

form over

1112

is also defined for a

221
2

~:

over

c (:)
2

If

f1

and

f2

and hence also

f = f 1 9f.

are forms

with odd determinants, then by (3.7) and property 6 of the

Hilbert symbol

Thus by (1.23)

and

c.

behave in the same way on sums of forms, so

29
(3.8) is proved once we prove it for unary and even binary forms.
(3.5).

We use the notation of the proof of

=a

det f

If

since

e: 1li*2
f

and we choose

is an even

bina~y

2 - 1 ;;; 0 (mod 8 ).
~

congruent over

If

w=1

a = b=0

to one of the form

c (f) = 1
2

one of

a,

and

b,
~2

congruent over

say

c2 (f)

so

form we choose

4 of the Hilbert symbol,

a

If

is unary then

1 = c (f)
2

w=O

so

then the matrix of

diag(~,-~),

=C2 (f).

is

so by property

We may thus assume that

is non-zero. Then the matrix of

is

to

so
c (f)
2

= (2a,(4ab- c2 )/2a)2
= (2a,(4ab-c 2 )/2a)2(2a,-2a)2

2
= (2a, c -4ab) 2
It is now a routine matter to check that

6 of the Hilbert symbol and the fact that

as

0(

= :t1 or :1:3 (mod 8).

c (f) =
2

c 2 (f)

using property

(210<) = 1 or -1

according

11

The lecturer is indepted to J.W.S. Cassels for a helpful letter

containing the above theorem and proof. See also Cassels [7 ].
Now let
signature

be a quadratic form over

1" = 't'(f)

THEOREM (3.9):

Q(2) = 1li2 n ~.

Then the

is defined.
For a non-singular quadratic form f

over

Q(2) ,

30
~:

Let

sec tion. Then

(-1)

1:"

...+

and

=0(+ -

be defined as at the beginning of this

0(-

0(-

= c<+ + 0<-

and clearly

sign det f

(3.8) we must consider

In view of theorem

f(w,w)-L-detf +signdetf _ f(w,w)-r-detf +1

4
_1:"+ r + sign det f - 1

4
2oC- + (-1)

0(-

- 1

4
(mod 2 )

Since

= (_1)0(-(0<- -

c .... (f)

Now suppose

1)/2,

the theorem is proved.

is a unimodular quadratic form, that is a non-

- singular integral form. Then

det f =:1:1

Furthermore, by lemma (1.24),

c (r) = 1

by lemma (1.25)

c2(f)c~(f) = 1.

THOOREM (3.10):

If

With

so
if

det f - sign det f

P

(3.9)

=0

is an odd prime. Hence

this proves

is a unimodular form then

f(w,w) -'t' :; 0
If

11

11

(mod 8 ) .

is in addition an even form, then

w= 0

is a characteris-

tic element, so
COROLLARY (3.11):

If

is an even unimodular form then

'L(f) :: 0

(mod 8 ) .

11

We remark that this last result is the best possible in the sense
that there actually exists a quadratic form satisfying the hypothesis
with

't" = 8.

Indeed let

E8

be the following graph:

31
Construct an integral matrix

M=

(~ij)

J1ij

=2
=1

i f the vertices

f1 ij

=0

otherwise.

.Mii

by the formulae

1~iS8

v. and
~
joined by an edge in E8

v.

are

Thus

2 1 0 0 0 0 0 0
1 2 1 000 0 0
01210000
001 2 1 000
00012101
00001210
00000120
000 0 100 2

The corresponding even quadratic form is easily seen to be unimodular

with

1:'

= 8.

Until we give a more general definition of "quadratic

forms of graphs" in 8, we shall refer to this quadratic form as "the"

quadratic form associated with the graph

E8.

4. INTEGRAL UNIK:>DULAR QUADRATIC FORMS.

This section is based both on the original lectures and on the

article by J.P. Serre [24]. The proofs of the statements on the first
two pages can be found both in Serrels article and in Jones [11].
Let
rank

= (f,V)

n.

be a non-degenerate integral quadratic form of

The set of non-negative integers

a minimum which we denote by

zero form or

min f

e , , en
1

min f

=0

we sq f

is a

of the lattice

the form

is given by
f

If

If

has

represents zero.

Wi th respect to a base
f

{If(x,x)11 xsV}

La ..

= . . 1J X.X
1 j
1,J

is not a zero form one can always choose the base

(e ) of V
i
such that the expression (4.1) is Hermite reduced. This term is defi-

ned inductively as follows:

1)

2).

If

= a 11 x 12

has rank 1 , then

has rank n , the expression (4.1) is Hermite reduced if

a)

1a111 ~ 21 a 1 j 1

b)

1a 11 1 =min

c).

a 11 f =

for

is Hermite reduced.

>1 ,

(~a1ixi)2+ f 1 (x 2 ,

reduced of rank

n-1

32

. ,x )
n

where

is Hermite

33
The following theorem gives a bound for
expression for

in an Hermite reduced

f.

THOOREM (4.2):
n

la11i

If

is a non-degenerate integral form of rank

then
min f <

COROLLARY (4.3):

::::'5
or -

If

(4/3) (n-1) /21 det f 11/n

is an integral unimodular form of rank

and is not a zero form, then it is equivalent to either

2:x~

2:x~. 11
~

In particular an indefinite unimodular form of rank ::::. 5

~5

sents zero. For forms of rank

repre-

the condition unimodular may be

dropped;
THOOREM (4.4) (Meyer):

form of rank

Every non-degenerate indefinite integral

represents zero.

We now turn to study the Grothendieck ring

An integral unimodular form
by

f )

r o (111)
and

-1

~ (111)

defined in 2.

represents an element (also denoted

in the semigroup

(c.f.(2.3. The elements in

represented by the forms

f.

_x

and

respectively. The element in

will be denoted by

G (111)
o

G (111)
o

will be denoted by
represented by

E:

This differs very slightly from our

conventions in 2.
LEMMA (4.5):

Every indefini,te odd unimodular form can be decom-

posed into the form

~:

Let

1ad-1) 6lg,

=(f, V)

with

non-singular.

be the given form. Br (4.3) and (4.4)

34
is a zero form.
f{x,x) = o.
maps

x e: V

Let

Since

is non-singular, the correlation

into an indivisible element

ists aye: V

with

z e: V

choose a

with

f{z,z);;; 1 (mod 2)

visible in

by

V,

one of

and

Hence there exis even, we

(this is possible since

Then

and

fey' ,y')

f{y,y)

f
is

is odd, say

e = y-{m+1)x
2

are indi-

The lemma hence

11
If

is an indefinite odd unimodular form then

(8

"t{-1)

By lemma (4.5)

1E9g

e = y-mx
1

'P of

f{y,y)

f{e ,e ) = 1 = -f{e ,e ) .
1 1
2 2

is equivalent to 0(+-1
~:

If

We may hence assume that

as

THEOREM (4.6):

V'.

in

y'= z+ (1-f{x,zy.

The elements

follows from (1.4).

f{x,y') = 1 .

f{y,y) = 2m+1.

~(x)

<<f'{x),y> = f{x,y) = 1

is odd) and replace

odd and

be an indivisible element satisfying

(-1)E9g

1 $ (-1) g.

Since

has to be indefinite, the theorem follows

11

by a trivial induction.
COROLLARY (4.7):

is equivalent .to

Two indefinite odd unimodular forms are equiv-

11

alent if they have the same rank and same index.

G (111)
o

We are now ready to calculate

THEOREM (4.8)
~:

For

f e: ~o (111),

either

indefinite. ~ (4.6) it follows that

so

and

=::r

generate

G (1lI)

1. f

or

(-1). f

f = ~+1 + ~- .(=1)

in

G (111)
o

Since the homomorphism

induced by the inclusion

these generators into free generators of

11

is odd and

111(8111,

11I c lR,

maps

the theorem follows

35
Corollary (3.10), which states that

Remark (4.9) :
:: 0 (mod 8)
(4.8).

h: F (?l) ~ 1lI/8?l
o

The map

(mod 8)

h: G

for any unimodular form

f(w,w)-"t'(f)

now follows easily frcm

defined by

is a semigroup homomorphism, so it induces a homomorphism

(?l) ~ ?l/81l1.

identically on
generators

'1

(3.10) is equivalent to saying that

(111)

but this is clear since

=1

and

of

vanishes

vanishes on the

(111)

The structure of integral unimodular indefinite odd forms is

completely determined by theorem

(4.6). The structure of unimodular

indefinite even forms can also be determined (see Serre [24J):

two

such forms are equivalent if and only if they have the same rank and
same index. Combining this with

THEOREM (4.10) :

(4.7) we have

T\-'o unimodular indefinite quadratic forms are

equivalent if and only if they have the same rank, index and type
(even or odd).

11

As for definite forms, our knowledge is meagre. We have for

example:

THEOREM (4.11)
rank ~ 8

Two unimodular quadratic forms of the same

are equivalent if they have the same signature and type.

The last theorem is no longer true for rank > 8.

let

be the form associated with

f' = f ~ 1 .

Then

f' and

E8

g = 1~ 1~

as defined in
$

(9

For example,

3, and let

times) are of the

same rank, signature and type. But they are not equivalent as

f' (x,x) = 1

has only

solutions but

g(x,x) = 1

has

18.

5. QUADRATIC FORMS OVER lZ'

THE GENUS OF INTEGRAL FORMS.
p'

So far we have considered exclusively quadratic forms which are

non-degenerate. This is not an essential restriction;
(f,V)

1':

is any quadratic form, the kernel

V ~ V'

of

is a direct summand of

non-degenerate form
DETf

by

de't f

Rad(V)

l'

on

V/Rad(V)

DET f

and

totally isotropic (that is

V,

for if

of the correlation
and

We shall denote

induces a
det1'

respectively, and agree that i f

f(x,y)

=0

for all

f =

x,ye: V)

and
f

is

then

DETf=1.
THIDREM

(5.1): Every non-degenerate quadratic form

over

lZp may b,a decomposed into

where

fi

(i = 0, ... ,k)

COROLLARY (5.3):
(p

-t 2)

is a non-singular form of rank

r i::! 0

Every non-degenerate quadratic form over

lZp

decomposes into unary forms.

Proofs:

The corollary follows immediately from the theorem and

(1.5). To prove (5.1) let

(i, j

be the matrix of

= 1,2, ,r)

with respect to some base. If every O(ij

36

is

37
divisible by

in

M = ptM"

write
that

7Zp

where

we say that
M'

= (0<: .)

is divisible by

is not divisible by

~J

We can
Notice

M' need not be invertible.

a).

If not all the diagonal entries of

M' are divisible by

p,

then some of them are units. Applying (1.4), we may split off these
entries.
b).

If all the diagonal entries of

M' are divisible by

0<:. ,

least one entry off the diagonal, say

~J

p,

then at

is a unit. But then the

minor

is non-singular and hence splits off.

~

where

repeated application of a) and b) we get

f

is non-singular and the matrix of

t
An obvious induction completes the proof. 11
The ranks

r.

of the forms

by the equivalence class of

f.

f.

is divisible by

p.

in (5.2) are uniquely determined

Indeed, by the structure theorem for

finitely generated modules over a prinCipal ideal domain (c.f.

Bourbaki [6]
'f' : V ~ V'
S;;;

elm -t 7Zp

In

7Zp

Ch 1lII, 4, th 2), the cokernel of the correlation

the only non-zero ideals are of the form

appears in

invariant definition of the

o

for

i~1.

Since

cr:2 S

and hence by
pt7Z

is clearly

r.

-t ct1 s:

f.

The number

rt '
Lr.

giving an

= rankf

is then also determined.

Each

p -t 2

(a}

where

are ideals uniquely determined by 1',

of times

f 7Zp /Otj

is isomorphic to

fi

we define

in (5.2) is non- singular, so

't i

=(det fi I p)

for each

det f i
i

so

is a uni t.

For

38

In other words, if we identify

~i

= DET f i

det f

if

det f. {22I*2

f;

22I*/2Z*2

V \

't i

We can consider

in

V Cl/ \

U = V+/V

f.

we give an invariant definit-

as a quadratic form over

V CV Cl/ \ '

and let

V+

be

i. e.

'

V+ = {xe:VCl/\lf(X,y) e: 22Ip
Then

then

only depend on the equi va-

t.

(see beginning of 3). Consider

0

(1,-1},

they are called the Minkowski invariants of

f = (f, V).

the "dual"

with

To prove the invariance of the

ion. Let

i
~

We shall show that the

lence class of

e: 221*2
P

if

is a finite abelian

for all

ye:V}

p-group, and

induces a bilin-

ear pairing

Filter the group

where

by

(x e: U1pix;o O}.

U.

quotient

W.~ = U./U.
1
~~-

It is clear that every element of the

so we may consider

p ,

is of order

a vec tor spac e over the fini te field

The above pairing

IFp

as
t

in-

duces a bilinear pairing

1
f: : W. X Wi --+ 1i 221 / i 1 221 ':' IF
~
~
p
PpP
P
f~

Therefore
Identify
)f i =

is a quadratic form over

IF */IF *2
p
p

nET f~

(i ~ 1)

with

{1,-1}

Indeed let

so

nET f~

f.=(f.,V.)
~

nET f:

and

lFp

;t 1

is defined.

We claim that

he as in (5.2). Then

39

f.

is just

't i

the

reduced modulo p

and for

and

det f

to

it follows since

Observe that

a second invariant definition of the

The invariance of

DETf:

;0

to

Alternatively, it is easily seen that

ll'1' t2 ,

mined by

'6.

so
i~1

is hence proved for

to(f) = ~1(Pf)

f~

It is now easy to see that

and

r.

is deter-

= rk1:,
~

giving

r .
~

An immedia t e cons equenc e of (1. 20) is:

The integers

THEOREM (5.4):

and

'6i

(i

~O)

constitute a

complete set of invariants for the equivalence classes of non-degenerate quadratic forms over

2Z

(p 12).

11

The following lemma is not hard to prove (c.f. Jones [11J p91).
For

pl2

it is an immediate consequence of (1.20) and (2.9).

Let

over

2Z
p

For

p ;i 2

are equivalent over

f
f

and
and

\.

For

be non-singular quadratic forms

are equi valent if and only if they

=2

they are equivalent if and only

if they have the same type (even or odd) and are equivalent over

~2'

11

We now return to integral quadratic forms. Two such forms

and
f

are called semi-equivalent or said to have the same genus if

is equivalent to
THOOREM (5.6):

over

2Zp

for

p = 2,3,

,DO

,00

Two integral even quadratic forms

are semi-equivalent if they are equivalent over

2Zp

for

and

p = 3,5,

40
Proof:
g

In virtue of lemma (5.5) we need only show that

f_ g

~2.

are equivalent over

Furthermore (1.25) implies

Since

and

detf:; detg
Thus by (2.9)

c. (f)
2

over

=c 2 (g)

lZ.oo

and

rk f = rk g

implies

and hence

c 2 (f)

=c (g)
2

are even forms, the last equality implies

(mod8), and hence

f_ g

ilt

over

THEOREM (5.7):

(c.f. proof of (1.18

11

DETf = DETg

The rank, index and type (even or odd) are a

complete system of invariants for the genus of unimodular forms.

Proof:

If

and

are unimodular forms with the same rank,

c (f) = c (g).
2
2

index and type, then by (1.24) and (1.25)

clearly

det f = det g ,

it follows by (2.9) that

The theorem thus follows by (1.20) and (5.5).

Remark (5.8):
denote the invariants

r.

VI/t

and let

lZ
p

V+

U: = V+ IV

ilt2

11

by

of

and

and can also be defined directly, with-

Namely, consider

be the "dual" of

VCVI/t

V+:= (xe:Viltlf(x,y)e:lZ
Then

over

be an integral quadratic form. We

and

1.

These are invariants of

out reference to

f = (f,V)

Let

f- g

Since

as a form defined on
given by

for all

ye:V}

I det f I ,

is a finite abelian group of order

and

induces a bilinear pairing

L: U X U~ Qj2Z

For any odd prime

U

p,

this pairing restricted to the p-component of

is essentially the pairing

Thus if one puts

U~p)
1.

= (u e: U

considered earlier in this section.

I piu = O}

and

W~p)
1.

U~p)
/U~p)
1.
1.-1

then

41
L

induces a bilinear pairing

which can be used to define

1f~p)
~

and

6. THE QUADRATIC FORM OF A 4k-DIMENSIONAL MANIFOLD.

Unless otherwise specified, by a manifold we mean a connected

compact orientable differentiable manifold with or without boundary
together with a given orientation. Thus all manifolds under consideration are oriented.
Let

M be a

4k-dimensional manifold (briefly: 4k-manifold).

The homology is finitely generated, so

~-lattice.

V:= H

2k (M,lZ)/Torsion

is a

The intersection numbers of cycles induces a quadratic

form

SM

is called the quadratic form of

is called the signature of

~

and its signature

M and will also be written

~(SM)

~(M)

can also be defined by means of the cup product in cohomology

as follows. The Poincare duality for an n-manifold may be expressed by

an isomorphism

for t::..tch

=0, ... ,n.

Denote the elements in

a,b, ... e:

H2k(M,~)

In particular for

2k (M,~ M;~)

=4k

, i

=2k,

which correspond to

under this isomorphism by eX,p,_

42

we have

respectively.

43
Since
o(v~

we may consider vp

e: H4k (M,dMjlZ) = 7Z ,

as an integer. This integer is precisely the

intersection number of

and

[9 J p156). We may thus regard

H2lt (M, ~ Mj 2Z) / Torsion

Exercise (6.1):
boundary then

SM

into

(see for instance Hilton and Wylie

SM

as the pairing defined on

by cup product.

7Z

Show that if

M is a 4k-manifold with empty

is unimodular (see for instance Milnor [14J).

Many of the results stated below have higher dimensional analogues

but our main concern for a while will be 4-manifolds.

Let

M be a 4-manifold with empty boundary and let

be its cohomology ring with coefficients in

form of

M over

7Z/22Z

7Z/2lZ.

H*(Mj7Z/27Z)

The quadratic

is again given by cup product:

S(x,y)

=xvy

w = w (M)
2
2
x e: H2 (M;7Z/27Z). We

In this case there exists a unique characteristic element

in

H2(M;2Z/22Z)

with

xvx

= xvw2

for each

remark that (for instance by the Wu formula)

middle Stiefel Whitney class of

w
2

is actually the

M.

The following theorems are known (Rohlin [21J, Borel Hirzebruch

[4 J, see also Kervaire Milnor [12J and [13J).
THIDREM (6.2) (Rohlin) :
w2

=0

If

M is a 4-manifold with

.)M = t;1

and

then
-r(M)

!!

(mod 16 )

We remind the reader that all manifolds under consideration are

differentiable. We do not know whether (6.2) holds in general for
oriented topological 4-manifolds.

44
Let

11":

H2(Mi7Z)

~ H2(M;7Z/27Z)

be the reduction modu1o 2,

that is the homomorphism induced by the coefficient epimorphism

If

THEOREM (6.3) (Borel Hirzebruch)

71'd = w ,
2

then

SM(d, d) ;;; 1:'(M) (mod 8) ,

quadratic form of

.!:!:.2..21.:

where

SM

is such that

is the integral

The assumption of (2.3) implies that

istic element of
11'.

de: H2(M;7Z)

~,

is a character-

as cup product commutes with the homomorphism

Hence (6.3) follows from (6.1) and (3.10).

A weaker version of (6.2), namely

11

-r:(M);;; 0 (mod 8)

is of course

a special case of (6.3).

The short exac t sequence

~ 7Z ~ lZ --+ 7Z/2lZ ~ 0

of

coefficients induces an exact sequence

2
2
2
1\'"
2
~ H (M;7Z) ~H (Mj7Z) ---+H (M;7Z/2lZ) --+
where the number

over an arrow means multiplication by

M has no 2-torsion (that is, the groups

i
H (Mj7Z)

2.

If

have no elements

of even order), then (6.4) leads to a short exact sequence

de: H2 (Mj7Z)

so there exists a
d,

considered as elements of

of characteristic elements for

2-torsion, then
may state

w
2

=0

with

1Td .. w
2

The set of all such

2
H (M;7Z)/Torsion,
SM.

is just the set

We conclude that if

if and only if the form

SM

M has no

is even. Thus we

45
THIDREM (6.2a):
BM

is even, then

If

M is a 4-manifold with no 2-torsion and

"t"(M) ;;; 0 (mod 16).

11

Thus the quadratic form associated with

occur as the quadratic form of a 4-manifold

Ea
~ith

given in 3 cannot
no 2-torsion.

The following theorem uf Kervaire and Milnor sharpens the results

of (6.2) and (6.3).
THEOREM (6.5):
d

In the statement of (6.3), i f the dual class of

can be represented by a differentiably imbedded 2-sphere in

then

BM(d,d);;; 1:(M) (mod 16)

The proof of this theorem and some examples will be given in 10.

7. AN APPLICATION OF ROHLIN'S THECREI"'I,A-INVARIA}r'l'S.

If
fold

that is
G -t 0 ,

is an abelian group, then a G-hoillology k-sphere is a mani-

with

II X = t;1

H. (X;G) 'it G
~

this implies

which has the same G-homology as the k-sphere;

for

i = O,k,

dim X

=k

and

H. (X;G) = 0 otherwise. I f
~

We intend to define an invariant ft

for lZ/27Z - homology

3-spheres, but first we need a lemma.

LEMMA (7.1)

Any lZ/27Z -homology 3-sphere bounds a 4-manifold

with

a).

H1 (Y;7Z)

b).

By

has no 2-torsion;

is an even quadratic form.

The proof will be postponed to the end of this section.

Now let

be a

lZ/27Z -homology 3-sphere and

be as in lemma

(7.1). Define
- 1:'(y)

}'(X) := ~

that is

reduc ed modulo

THIDREM (7.2)

morphism type of

~(X)

E:

QV7Z ,

1.

is an invariant of the oriented diffeo-

X.

For this reason we call .fI-(X)

46

the

l!-invariant of

X.

This

47
is a special case of the ft-invariants studied by Eells and Kuiper in

[ 8 ]. Before we prove (7.2) we need

LEMNA (7.3):

If

is a

7Z/27Z -homology k-sphere then

a).

Hi (X;7Z)

b)

Xis a

~ -homology sphere;

c).

7Z -homology sphere (we remind topologists that

2

is a

is a torsion group of odd order for

i #'O,k ;

7Z2 de-

notes the ring of 2-adic integers).

~:

Consider the exact sequence

2
~ H. 1 (X;7Z/27Z) ~ H. (X;7Z) ~H. (X;7Z) ~
~+

induced by the exact sequence of coefficients

If

i #,O,k

then

H. (X;?l) ~ H. (X;?l)
~

is an epimorphism, so since

the groups involved are finitely generated, a) follows. b) is an obvious consequence of a), so it remains to prove c).
7Z2

Recall that the ring

ideal domain and

Hi (X; 7Z )
2

is fini tely generated. Hence

m
Hi (X; 7Z 2 ) ':' ~ 7Z.(a k

(7.5)
where

of 2-adic integers is a principal

cr1 ~ ()(2 ~

~ am c 7Z2

are ideals in

7Z2

Bourbaki [6 ] ch7, 4, th2). But the only ideals in

ideal and the ideals
7Z2 or

2 7Z

j
j
7Z/2 7Z ':' 7Z/2 7Z.
2

one can replace

7Z by

(see for instance

7Z2

are the zero

so each summand in (7.5) is of the form

However, using the exact sequence

7Z2 in (7.4), showing that

48

Hi (Xj7Z 2 )

Hi (Xj2Z 2 )

Hi (Xj7Z 2 ) = 0 for

is epic for

i;l O,k.

I t follows that

11

i;l O,k.

We now return to the proof of (7.2).

Proof (7.2):

It clearly suffices to show that

dant of the choice of

Y.

given by lemma (7.1). Let

mon boundary
reversed.

X.

Here

Let

Y1

and

M=Y 1 U-Y2
-Y 2

is indepen-

j4.(X)

Y2 be two 4-manifolds as
pasted together along the com-

of course means

Y2 with orientation
M has, after smoothing if necessary, a differentiable

structure compatible with those on

reader to Milnor

Y1 and -Y 2 We refer the

for details about pasting and smoothing.

[17], [18]

We claim that

M has the following properties:

1).

M has no 2-torsionj

2)

SM

3).

S~ EI1-S~ = S~
where S~ is defined as in 3.
Y1
Y2
M
Before we prove these properties, observe that they suffice to

is even;

prove (7.2). For by property 3), -r(M) = '"(Y1 ) - 't'(Y 2 )

and properties
1) and 2) together with Rohlin's theorem (6.2a) imply that 't'(M) =0
(mod 16 ). Theorem (7.2) then follows.
To prove 1), 2) and 3) above, apply the 14ayer-Vietoris sequence
to

M = Y1 U -Y 2

~H.

to obtain
(X)

~H.

(Y 1 ) $H. (Y2)
~

~H.

(M)

~H.

~-

1(X)

where the coefficient group is not specified.

By (7.3) we know that

X is a

~-homology

sphere, so (7.6)

gives
0~H2(Y1;~)

It

H2(Y2j~) ~H2(Mj~) ~O

49
Taking orientations into account, it is not hard to see that
proving property 3).
Next we take integral coefficients in (7.6) to obtain

HO(X,7Z) = 0

where

is the reduced homology group. Since

Y1

and

Y
2

have no 2-torsion and

H (Xj7Z)
1

follows that

has no 2-component. Poincare duality of homology

H1 (Mj7Z)

is a torsion group (by (7.3a, it

groups (see for instance [23] p245, Satz m) now proves property 1).
Finally, we take coefficients

in (7.6) and apply (7e3c) to

7Z2

obtain

o ~ H2 (Y1 j7Z 2 ) ID H2 (Y 2 j7Z 2 ) --=-+ H2 (Mj 7Z 2 ) ---+0

just as in (7.7). Since

Tor(odd torsion,7Z ) = 0 ,
2

ficient theorem gives that

H (Y j7Z )

Recall that

the universal coef-

=H2 (Yi j7Z) 7Z2

and

s'f2 is defined on
~

The terms on the right hand side are even by assumption, so

even. This implies that

SM

itself is even, proving property 2) and

completing the proof of (7.2).

11

As an example we shall apply the ,,-invariant to lens spaces.

Let

be the standard 3-sphere, with canonical orientation inherited from

e2

Let

and

is

be integers with greatest common divisor

50
(n, q)

=1

for each
S3.

Define an operation of the group

~ Z/nZ.

Then

Z/n:1Z

on

by

acts freely and differentiably on

s3/(:1Z/nZ) =: L(n,q)

The quotient space (space of orbits)

is,

by definition, the lens space of type (n,q), with orientation and

differentiable structure inherited from
as universal covering space and

S3.

Z/nZ

Since

L(n,q)

has

S3

as group of covering trans-

formations,

Thus the Betti numbers of

coefficient is
L(n,q)

L(n,q)

at dimension

are

1,0,0,1
If

and the only torsion

is odd it follows that

is a 1li/2Z -homology sphere.

Clearly

L(n,q)

so one may assume

only depends on the residue of

0 < q < n.

LEMMA (7.9)

modulo

n,

Furthermore

With respect to the canonical orientations

L(n,q) = -L(n,n-q)

Proof:

The differentiable orientation reversing involution

on

S3

carries

L(n,q)

onto -L(n,n-q) ,

since

11
Since clearly
X,

}L(-X)

= -fl(X)

for any 1li/2Z -homology 3-sphere

we need only consider lens spaces

values. Recall that a fraction

continued frac tion :

n/q> 1

L(n,q)

where

takes even

can be expanded into a finite

51
n/q = b

-~1_ _ __

2
-

bs
(notation)

where each

bi

that if

is odd and

is an integer with

this type wi th each

bi

~2

Ib.1
~

It is easy to prove

is even, there is a unique expansion of

p + = P+ (n, q)

even. Deno te by

the number of positive and negative

b. 's
~

and

-=

in this unique

expansion.
RECIPE (7.10) :

( (n,q
,.ML

=p

+ -p

16

A proof will be given in 8. As examples we have:

7/6

= [2,2,2,2,2,2]

7/2 = [4,2]
Therefore,
~(L(7,6

= 3/8

~(L(7,2

= 1/8

j-<-(L(7,1

= -r(L(7,6 = -3/8

= 5/8

(mod 1 ).

Recall (J.H.C. Whitehead [28])

THIDREM (7.11):

Two lens spaces

L(n,q)

same homotopy type if and only if either

residue modulo
Thus

n.

L(7,1)

or

L(n,q')
-qq'

have the

is a quadratic

11
and

L(7,2)

have the same homotopy type; but they

are not diffeomorphic since !'(L(7, 1) )

-r(L(7,2

qq'

and

differs from r(L(7, 2

and

52
Now let
call that

X1

X1

and

and

X
2

X
2

are called h-cobordant if and only if there

exists an n+1 -manifold

and

X1

and

X
2

be two n-manifolds without boundaries. Re-

W such that

~W

= X1 U -X

(disjoint union)

are both deformation retracts of

THEOREM (Z.12):

Let

X
1

and

X
2

W.

be h-cobordant 1li/27Z -homology

3-spheres. Then .f-(X 1 ) t(X2 )

Proof:

Let

(Z.1), and let

WU Y2

= ~Y.

X.

(i

= 1,2)

I
\
\

is a deformation retract of

Y ,
2

SN = SY2

so

(Z.1) with

~N

=X1

W,

On the other hand,

1:(N)

It follows that

L(Z,1)

Observe that the set 't3

= - 't'(1Yi

be

is homotopy equivalent
is a manifold as in

and

(mod 1 ) == r(X 2 )

L(Z,2)

and

X
2

11

are not even h-cobordant.

of all 7Z/27Z -homology 3-spheres is

X1~X2

closed under connected sum (the connected sum

X1

Hence

= - 1'b (mod 1 )

n-manifolds

are as in lemma

() (D

to

X2

-X2 X2

X
2

Y.

W be the manifold of the h-cobordism. Let

pasted along

Since

where the

of two

is obtained by cutting a small open disc

out of each and pasting together along the resulting boundaries

Sn-1 ).

With respect to the operation

*,

03

is a commutative

53
semigroup with identity. One easily proves the following
THIDREM (Z.13):

As a map

If3 ~ CV1li,

)I..

is a semigroup

homomorphism which maps the identity onto the identity.

The operation
r3

11

is compatible with the h-cobordism relation in

(see Milnor [1Z] lemma 2.3). Hence

is even a homomorphism of

)I..

the semigroup of h-cobordism classes in

03

into

Q,/7Z.

An easy consequence of the definition of )I.. and the Alexander

duality theorem is
Exercise (7.14):
is embeddable in

JR4

Let

be a 14I27Z -homology 3-sphere. I f

then )A(X) = 0

The spherical dodecahedral space is a classical example of a

14I21li -homology 3-sphere (see for example Seifert and Threlfall [23]

p218). The )A-invariant of this space, as we shall show in 8, is

1/2,

so this space is not embeddable in

JR4.

We now give the promised proof of lemma (Z.1). Recall the well-known fact that every 3-manifold is parallelizable (Stiefel [26]).
Therefore, by Milnor [1Z] (see also M.W. Hirsch [10J), every 3-manifold wi th empty boundary bounds a simply connected 1I-manifold. Let
X

be a

1li/27Z -homology 3-sphere and

fold wi th
Let

~Y::

X.

since

a simply connected 11 -mani-

Then (Z.1a) is trivially satisfied. We verify b).

M be the double of

the common boundary

X.

Y,

We claim

that is
w (M)
2

M = Y U -Y

=O.

Indeed

pasted along
w (Y)
2

=0

is a 1I-manifold, so the claim follows immediately from the

54
Mayer-Vietoris sequence with coefficients in

~/2~

and the naturality of the Btiefel-Whitney classes. It follows, as in

6, that

BM

is even. On the other hand

(see proof of (7.2. It follows that

By

is even, as was to be

11

shown.

We end this seution with a digression. Let

~y

4k-manifold with boundary

1).

=X

be an oriented

and assume

only has homology in dimensions

and

2k

and this is

torsion-free (integer homology);

~y

2).

= X is a rational homology sphere.

(These assumptions may be weakened). Consider

(coefficients in ~ ). Let
of

where

is defined on

V = H (y) ,
2k

V.

so the quadratic form

V+ = (x e: V lit I Sy(x,y) e: ~ for all

correlation for
group, so
H2k(y,X)

By

By,

and

ye: V}.

cokern(f) ~ H _ (X)
2k 1

is non-degenerate. Furthermore

is essentially

(5.8),

Define, as in remark

V+

and

H2k-1(X)

Now

cP

= V+/V

By
,

is the

is a finite torsion

Hom(H

2k

(y) ,~)

can be identified with

55
U.

As in (5.8), the quadratic form

5y

L: U xU~

induces a bilinear pairing

CV7Z

which can be used to define the invariants

~~p)
~

and

r~p)
~

of

5y ,

p an odd prime. However, as the informed reader may already have observed,

gives the linking numbers of homology classes in

5eifert and Threlfall [23J 77). Thus

(see

and consequently also

L ,

are invariants of the oriented homotopy type of

Theorem (5.6) yields the following consequence: If

Y1

and

X.
Y2

are 4k-manifolds satisfYing 1) and 2) above and the quadratic forms

51

and

52

of

equivalent over
JY1

and

JY

Y1

and

1li ....

= lR ,

Y2

are even with odd determinants and

then

51

and

52

have the same genus if

have the same oriented homotopy type.

8. PLUMBING.

We shall first describe plumbing in arbitrary dimensions before

going into more detail in the case which interests us here, namely
plumbing of 2-disc bundles over

S2.

Recall (Steenrod [25] p 96) that principal

Sn,

and hence also n-disc bundles over

SO(n) ,

Sn

SO(n)

bundles over

with structure group

are classified up to equivalence by the elements of

1L _ SO(n)
n 1

This classification is as follows.

Sn

Sn = (x E: :nf+1 I Ixl = 1 }
mn+1 Sn is the union of the two discs Dn = (x E: snl
xn+1 ~O }
+
n
} , which intersect in their common boundary
Dn
_ =(
X E: S I xn+ 1 ~ 0
Consider

as the unit sphere

n l xn+ 1 = 0 }
Sn-1 = (
X E: S
Sn,

If

is a principal

SO(n)

in
and

bundle over

let

be trivialisations of

E restricted to each of the discs. Then the

map

is of the form

( t, x) -

f: Sn-1 ~ SO(n).

( t, !( t) x)

where!

The homotopy class

ifying element of the bundle.

56

[f]

is a map

E:TC _ S0(n)
n 1

is the class-

57

Let

'1

= (E1'P1'S~)

n-disc bundLes over

Sn.

= (E2'P2'S~)

and

;2

Let

D~ c S~
~

be two oloien:;ed

be embedded n-discs in the

base spaces and let

f.: D~X Dn~E.ID~
~

be trivialisations of the restricted bundles

To plumo

j1

and

~2

and identify the points

E.ID~
~

we take the disjoint union of

f 1 (x,y)

and

f (y,x)
2

for
E1

i = 1,2
and

E2

for each

n
(x,y) e: D XDn

We obtain an orientable manifold with boundary, and this manifold is

differentiable except along

f (Sn-1 X sn-1) , where it has a "cor1

ner". By Milnor [17] this corner can be smoothed, and the smoothing
is essentially unique. We give a brief idea of how this can be done:
A point

x e: f 1 ( Sn-1 XSn-1)

has a neighbourhood which looks like

(Sn-1 XS n- 1 ) X Cm xlR U lRxlR )

+

ative half line in

lR.

'

where

lR
+

denotes the non-neg-

The second factor may be "straightened ll by

the transformation
26+1t'
2B+1I:
(r cos(t}) ,r sin(e t---+ (r cos -3,r sin -3-)

which is differentiable except at

If
ion from

=0

is even, the plumbed manifold inherits the same orientatE1

and

E2

and is thus canonically oriented. If

is

58

odd one must make a choice.

We now want to describe how to plumb several bundles together
according to a "tree". A!!:!!

is a finite contractible 1-dimen-

sional simplicial complex.

Suppose we have a tree
is, to each vertex
Called the weight of

j.~

n-disc bundle
1't'n_1S0 (n)

v.

To each vertex

= (Ei ,p., S~)

~

We then plumb

joined by an edge of
Vi '

of

Vi

(T,m ) weighted in nn_1S0(n). That

i
T is assigned an element m. e: 1t"n_1S0(n)

v.

T we choose an

of

classified by the element

to

ji

~j

whenever

I f several edges of

v.

of

Vj

are

meet in one vertex

we choose the necessary embeddings of the disc

and

m.

n
D

in

S~
~

to

be disjoint. A theorem of Thom (Milnor [16J) assures that the result

of plumbing is independant of the choice of these embedded discs.
We denote the resulting manifold after smoothing by
Example:

Plumbing trivial 1-disc bundles according to

gives:

P(T,m.)
~

59
From now on we restrict
Y

= P(T,mi )

to be even, say

=2k.

Let

be the 4k-dimensional oriented manifold obtained by

plumbing according to the weighted tree

(T,m ) ,
i

We wish to calculate the homology of

and

and let

= aY

Denote the bundles which have been plumbed by

i = 1, ... , s .
S2k" S

2k

has as deformation retract the one-point union

v VS

2k

of

homology groups are

(s

times).

copies of

Ho (Y;2Z) ~ 2Z

H2k (Y; 2Z)

S2k,

so the only non-zero

H (Y;2Z) ~ 2Z$ $ 2Z
2k

and

where

has a basis

homology class represented by the zero section

Ji

Denote the euler number of the bundle

only depends on the classit,ying element

m
i

sn
by

is the

e(m ) ,
i

n2k_1SO(2k).

since it
There

are many equivalent definitions of the euler number, for instance,

considered as a map

e: 1'C _ SO(2k) ~ 2Z,

2k 1

P.: "'2k_1 S0 (2k) ~ 1L'2k_1S2k-1

map

=2Z

-P.,

is just

where

is the map induced by the fibre

SO(2k) ~ SO (2k)/SO(2k-1) = S2k-1

(see proof of theorem 2 in

Milnor [15] or theorem 35.12 in Steenrod [25]). The "classical" definition is essentially that

e(m.)

number of the zero section

S2k C E.

is equal to the self-intersection

Using this definition the

following theorem is obvious.

THEOREM (8.1):

The matrix of the quadratic form

with respect to the basis

a , ,as
1

of

H (Y)
2k

Sy

is given by

with
1

if

i!j

and

v.

and

connected by an edge in

v. are
J
T ,

0(
~J

0
e(m. )
~

if

i!j

if

i =j

of

otherwise,

11

60
Since the above quadratic form depends only on the tree
(T,e(mi
(T,e(mi

weighted in

lZ,

we cal.l it the quadratic form of

With theorem (8.1) we are in a position to cal.culate the homolos

of

X. JY.

Namely let

be the linear map given by the matrix

THIDREM (8.2) :

H.~ (XjlZ) =0

M above.

for

i~0,2k-1,2k,4k-1,

and

2k_ 1 (X; lZ) ':' Coker f

Ker f

Proof: Consider the exact homology sequence with integral.

coefficients

~H.(y,X)~Hi 1(X)~H.

i ~ 2k,4k.

~-i(y) = Hom(H _ i (Y),lZ),

4k

Hence

1(Y)~

H.(Y,X) ~ H4k- i (y),

By Poincare-Lefschetz dual.ity,
has no torsion,

~-

H.(X)
= 0
~

for

and since

which vanishes for

i ~ O,2k-1,2k,4k-1.

In the

middle dimensions we obtain the commutative diagram with exact row

is the correlation associated with

so the theorem follows.

11

SY'

and hence has matrix

61
det M ~ 0

In particular if

I det M I

finite of order

Thus if

2Z/22Z - homology sphere and if

k > 1

sphere. In fact for

then

H2k (X)
det M

det M =:1:1

= 0 and H2k _ 1 (X)

is odd,
then

is a

is a 2Z -homology

X is simply

it is not hard to see that

connected, so in this case

det M

homotopy sphere. However for

= 1

implies that

k = 1,

dim Y

that is

is

is even a

=4,

this is

not true in general. We shall now discuss this 4-dimensional case in

more detail.

Plumbing Bundles over

The principal
'"'1S0(2) :;; 2Z,
generator in

S2.

SO(2)

bundles over

S2

are classified by

S3~S2 corresponding to a

with the Hopf fibration

2Z (see for instance Steenrod [25] p99 and p105; here

and in the following we identify

SO(2)

with

S1).

By the remarks

preceding theorem (8.1), the euler number of bundles gives an isomorphism

e: "'1SO(2)

of a principal

2Z,

SO(2)

so we can identify the classifying element

bundle over

identification of 1'(1 SO (2)

with

S2
2Z

with its euler number (this

is the negative of the usual

identification given by degrees of maps).

The trees we need for plumbing are hence trees
in

2Z,

and the quadratic form of the manifold

to the quadratic form of the tree

(T,m )
i

weighted

Y '" p(T,m )
i

is equal

(T,m.)
~

As examples we mention the following trees which arise in the

classification theory of simple Lie algebras (see for instance
Seminaire Sophus Lie, 1

annea, Exp.13). These are the only trees,

62
when weighted by

-2,

whose quadratic forms are negative definite.

v~----'v~~t---v"'; - - - - - - -

--V":

v~------------v:

'.

In the following table each vertex of the tree

by

is weighted

-2

TABLE
T

H1 (ap(T); 221)

22I/(s+1) 221

Ds

1't'1 (dP(T

{ ",/2",. "'/2"'
221/4221

.... }

, s odd

=: FT

s+1

D'

s-2

221/3221

T'

E7

221/2221

w'

E8

I'

FT

s/16 (s even)

E6

The groups

)o({Clp( T) )

6/16

8/16

are the only finite subgroups of

S3 (. unit

quaternions), as can be seen as follows: the finite subgroups of

SO(3)

are known to be the cyclic groups

Cn ,

the dihedral groups

63
Dn (not to be confused with the tree
octahedral and icosahedral groups
double

uoverin~

subgroups of

SO(3~,

of

S3

D ), and the tetrahedral,

n
T, W and I Regarding ~3 as a

one deducds easily

are the cyclic groups

en

~nat

the only finite

and the binary groups

D~

, T', W' and I' obtained by lifting the subgroups

of

Du' T ,Wand

SO(3)

We sketch a proof of the statements in the table. The homology

groups

H1 (JP(T

(s~,
a

can be checked by theorem (8.2). For

T = E6

T = E8

and

it follows that the manifold

~P(T)

is

2Z/22Z - homology sphere, so the ,ft-invariant is defined. The values

of the f4-invariant are obvious, since the quadratic form of

is

negative definite and even for each of the trees listed and is equal
to the quadratic form of
~1dP(T)

P(T).

Finally the fundamental groups

can be computed directly by finding explicit generators and

relations, however these groups can be obtained much more easily from
the general results of

von Randow [20]

which we shall now describe

briefly.
An oriented 3-manifold will be called a Seifert manifold if it

can be fibred, with e~eptional fibres, over

(Seifert [22]). To such a manifold

X,

S2

with fibre

S1

Seifert associated a system

of integers

known as the Seifert invariants, where

ional fibres and the integers
O<,Bi<co(i
Expand

for each

0<./(0<. _ A.)
~

'~i

is the number of except-

are coprime and satisfy

i=1, , r .

~i/(~i - ~i)

~i

r~

into a continued fraction

"l (1) ]
si

64
and let

Then

be the star-shaped tree weighted by the

= ~P(T).

Von Randow's proof involves a special construction

of lens spaces via plumbing, which we shall study later in this section.
Remark:

Von Randow's orientation conventions for Seifert mani-

folds and lens spaces are opposite to ours, so his weighted trees are
the negatives of those here. We discuss various orientation conventions
occurring in the literature in the appendix.

Let us return to the fundamental groups of the table. For any

finite subgroup

of

s3,

the coset space

S3/F

is a Seifert

manifold (the fibration is given by the action of a subgroup

S3

by left multiplication). We claim that each manifold

listed in the table is in fact diffeomorphic to

the spherical dodecahedral space
(-1; (5,1),(3,1),(2,1

S3/I'

S /FT

S1

oP(T)
For instance,

has Seifert inVariants

(see Seifert 1:22J). Here

5/(5-1)

= 5/4

3/(3-1)

= 3/2 =

[2,2J

2/(2-1)

= 2/1

= [2,2,2,2J

r .. 3

of

and

65
By the above, we get the star-shaped tree:

-2

-2

-2

-2

-2

-2

-2

L2
But this is

Ea

hedral space if
I'.

~P(Ea)

in other words

Ea

is weighted by

-2.

is the spherical dodeca-

n1~P(Ea)

In particular

The other cases in the table may be checked in the same way.

To investigate plumbing in more detail, we need a precise description of ~he bundles involved. We shall identify
so that the operation in

S1

by

m = (X m,p m,s2)
represented by a map

~~,

S2

with euler number

is classified by the element in

s()

Tr S1

with

is written additively.

S1 bundle over

Denote the principal

S1

of degree

obtained as the union of two trivial

-m.

Hence

S1 bundles over the disc

as follows:
X

where

= D2 X S 1 U D2 X S1
f

f: d(D2XS1)=S1XS1~S1XS1= ~(D2XS1)

is the map

f: (x,y) t--+ (-x,y-mx)

(the first entry
each

D2XS1

-x

on the right assures that the orientation of

is compatible with that of

X). This map

m

mayalso

be represented by the matrix

(-1 0)
-m

Note that the above is a well known description of the lens

66
space

L(-m,1) ,

denotes

so

Xm = L(-m,1).

-L(m,1) = L(m,m-1)

S3~S2,

from the unit sphere in

patible with the

L(-m,1)

is by observing that

represented as the map

(z1,z2)

~[z1,z2]

~2 to the complex projective line, is com-

2Z/d - action (7.8) on

S1 bundle

S3

when

q = 1.

Using the

it is not hard to see that the in-

S3/(2Z/m2Z) = L(m, 1) --+ S2

The principal

is negative,

X = L(-m,1)
m

fact that the Hopf map is ~(-1)'

duced map

-m

in standard notation.

Another way of seeing that

the Hopf map

If

j(m)

~(-m).

is just

can be identified with the boun-

dary of the associated 2-disc bundle over

which is the bundle

we need for plumbing. However as we have seen, in many cases we are

interested in the manifold
T

dP(T)

rather than

is a weighted tree. The assignment

called plumbing;
operator
1).

If

~P: T~ ~P(T)

m2

will also be

it may be defined directly without help of the

We describe this in three steps:

has just one vertex and this is weighted by

Let

be the tree

Let

embedded in the base spaces of the bundles

~(

itself, where

m,

define

= Xm

~P(T)

2).

P.

P(T)

2
) = (X ,p,S)
m2

D2

be 2-discs

2
j(m1) = (X ,p,S)
m1

and

X I Di
mi

be trivialisations of the restricted bundles

the restricted bundle

ffii

and

respectively and let

f. : D. X S

D1

X I (S2 - Int D. )
mi
~

XmilDi

Denote by

and consider the com-

posite map
f:
where

,)X~1

f- 1

~(Xm1ID1)~)(D1XS1)~d(D2XS1)~dX~2

t: d(D1XS1)=S1XS1~S1XS1 = J{D2XS1)

is defined by

67

t(x,y) = (y,x)

3).

Then

peT)

= X' U x'

after smoothing.

m1 f m2

The procedure for general weighted trees is now clear.

The map

Denote by

above is given by the matrix

the annulus obtained by cutting a small open disc from

the center of the disc

f),.

can be identified with

[0,1]

81

In the notation of 2) above, it is clear by our previous comments that

we can write

x'

m1

m2

where

1
1
= D2 X 8 U AX 8
f1

l::.x 8

1
f.: 8 1 X 8 1 ~ 8 X 8 1

1 U D2 X 8 1
f2

is gi ven by the matrix

(-1 0)
-m
i

for

i = 1,2

Thus
~P(T)

1
1
= D2 X 8 UAX81U~X81 U D2 X 8
t

f1

f2

1
1
= D2 X 8 U D2 X 8
f

pt

where

f = f2ot of1

is given by the matrix

0). J.( -1-m 0)1

(-1
-m 2 1

68
More generally let us consider the weighted tree

..m1

m2

(As,m ) 1 ~ i ~ s
i

..

ms _1

ms

----~.~ ------------.~----

where the

m
i

are integers. If we denote by

S1 XS1~S1 XS1

f.

the map

with matrix

(-1 0)
-m. 1
~

then it is clear that

dP(AS,m )
i

D2 X S1 U I1x S 1 UI1XS 1 U~xS1 U

t
t
f1
f2

'"

U 6x S1 U D2 X S 1
f
t
s

(8.3)
where

f=fot-f
S

s-1

1
U D2 X S
f

D2 X S1

...
J.(-1 0)

et

we see that

Observing that - tf.

-m.

-mi
( -1

has matrix

1)
0

has matrix

-q p.)
(n q

:=

-m

101)

-m
( -1

0). (-mS_ 1 1)

(-1
s

-1

(8.3) is thus the well-known description of the lens space L(n,q)

as the union of two solid tori. By multiplying each aide
from the left by

we get the hat.dier equation

'"

(-ms 1). r-mS-1 1)

-1 0
-1
0

(-m 1 1)
-1 0

0; (8.4)

69
so we have proveu
THIDREM (8.6)
~P{A

Then
(8.5).

Let

(A ,m.)
s ~

,m.) = L{n,q),
~

be the weighted tree as above.

where

and

are defined by equation

11

The matrix of (8.5) has determinant

right hand side has determinant 1 .

since each factor on the

This shows

np + qq' 1
In particular
qq'

is

1 (mod n ) .

and

are coprime, as they should be, and also

If we reverse the order of the tree and plumb accor-

ding to the reversed tree we get

L{n,q'),

so

L{n,q) = L{n,q')

This is a classical result on lens spaces (see Seifert and Threlfall

1:23] p215 SatzlI).

For given coprimes

such that
~1

= q,

~P{A )

and

0 < q < n,

q,

= L{n,q)

we can find a tree

To see this we let

AO

=n

and we use the euclidean algorithm to obtain

AO
A1

= a1~1 = a 2A2 -

A2

o ~)..2 <).,1

A,3

o ~A3 <'\2

This is equivalent to saying that

continued fraction

n/q

= [a 1 ,a 2 ,

n/q

can be expanded into the

70
Using (8.7) we get by induction

(a 1)
s
-1 0

(8.8)

(a. 1)

-1 0

so

(:~ ~) (:~~) =
Comparing this with theorem (8.6) and the remarks following this
theorem we obtain
THEOREM (8.9) :

where the

The lens space

P(A )

4-manifold
tree

(A ,m.)
s

If

As

a.

are given by (8.7), then

L(n,q)

(A ,m.)
s

is the tree with weights

L(n,q) = JP(A ) .11
s

obtained in this theorem bounds a

whose quadratic form is the same as that of the

with

m. = -a .
~

As the integers

a~

may not be even,

this quadratic form cannot be used to compute ",(L(n,q

when the

latter is defined. This situation can be remedied. By 7, the

variant jL(n, q

is only defined for odd

may assume without loss of generality that

and

n
q

~-in-

and in this case we

is even. For such

(8.7) may be modified to yield

""1 = q
(8.10)

Ib s 1>0
where each

b
i

is even. This, by the way, proves the assertion pre-

ceding recipe (7.10). Now let

= (A ,n.)
s ~

with weights

n.~

= -b.

71
As above, one can easily show that

JP(A ) ~ L(n,q) ,
s

(8.8) must be modified slightly if

As = -1.

L(n,q)

bounds a 4-manifold

even, and I'-(L(n,q

P(A )
s

and

p-

In this construction

whose quadratic form

= (p+ - p-)/16 (mod 1),

are the number of positive and negative

To prove this, we must show that

quadratic form

Sy

= p- -

r(Y)

Sy

is

-"t'(Y)/16 (mod 1)

can be calculated as

In (7~ 10) we stated that jA-(L(n,q

p+

even though

p+.

where

respectively

Hence, since the

is represented by the matrix

-b
1

(8.11)

-b

Mc

1
1 -b

we need only prove the following lemma.

LEMMA (8.12):
lb.

I > 1.

Then

Proof:

Let

1:"(M)

= p- -

If

c
i

I b. I > 1
~

be a matrix of the form (8.11) with

p+

An easy induction shows that

to the diagonal matrix

(8.13)

diag(-c ,
1

,-c)
s

is congruent over

where

= b. _ _1!.-_ _ _.,....
~
b i + 1 - _..:!.bs

for all

obviously proves the lemma.

then clearly

sign c

an integer prime to

= sign b i

which

11

We end this section with a digression. Let

and

lR

Suppose that

be an odd integer
is

72
the prime decomposition of
via the Legendre symbol

It is clear that if

(qIPi)

and

quadratic residue modulo

2Z/n2Z ) ,

n.

q'

The Jacobi symbol

(qln)

is defined

by the formula

are integers such that

(that is,

qq' = x

qq'

is a

has a solution in

then
(qln) = (q'ln)

By theorem (7.11) it follows that

(qln)

homotopy type invariant for lens spaces

is an orientation preserving
L{n,q)

that the h-cobordism invariant ,,(L{n, q

with odd

n.

Notice

is also defined for such

spaces.
THEOREM (8.14):
16jd:L{n,q

If

is odd and

=- Q (mod 4)

+ (n-1)

coprime to

then

and

(qln) = (_1)4J<L(n,q) + {n-1)/4

Proof:

Suppose we have proved the theorem for

even. Then

using the identities

(-1In) = (_1){n-1)/2

(8.15)
and

L{n, q) = - L{n,n-q)
the theorem also follows for
to the case that
Expand

n/q

odd. Hence we can restrict ourselves

is even.

into the continued fraction

accordingto{8.1Q),thatis
We have shown that

L{n,q)

Ib l>1
i

and

bi

bounds a 4-manifold

n/q = [~, ,bs ]

even for

Y peA )
s

i=1, ,s.
whose

73

quadratic form

Sy

may be described as follows. Let

= -Sy

then

s
2
Lc.y.

where the rational numbers

i=1

~ ~

c.

are given by (8.13). We have

j-tL(n,q) = -"t'(Sy)/16

1:'(s)/16

(mod 1)

We now compute the Hasse-Minkowski symbol

primes

p.

assume

does not divide

is a prime factor of

= :tn

we have that

then
n,

=1
p =Pi

c (S)
p

say

'

where

Since

(q,n)

=
=1

in the form

and

=:tq,

n.

by (1.24). We hence

ptq.

where
det S2

for all odd

is the prime decomposition of

Write

Note first that by theorem (8.2) and the fact that

det S
Hence if

c (S)

so since

cp(S)

It follows that

ptq ,

Also, since

S1

is a

= c p (S1)c p (S2)(detS 1 ,detS 2 )p

= (n/q,q.sign de tS 2 )p

Since

and

(n, q)

=1

we may write

erty 6 (1) of the Hilbert symbol

and by prop-

74
fl

(q.sign det S21p) ~

~.
#.
(qlp) ~(signdetS2Ip) ~

Hence
-1--1.
cp(S)
p odd pr~me

t
IT

i::1

fJ

(S.

(qlp.) ~(signdetS2Ip.) ~
~

(qln) (sign det S21n)

We may assume

n/ q > 0 ,

so

sign det S2 = sign det S.

Applying lemma

(1.25), we get
(8.16)

c (S)c",,(S)
2

On the other hand, since

(8.17)

(qln) (sign det Sin)

is even, theorem (3.9) gives

c (S)c",,(S) = (-1) (1:'(S) + det S - sign det S)/4

2

Combining (8.15), (8.16) and (8.17) now easily proves the theorem.

11

The lecturer is again indepted to the letter of J.W.S. Cassels,

mentioned in conection with theorem (3.8), for essentially the above
proof.

9. COMPLEX MANIFOLDS OF COMPLEX DIMENSION 2.

In this section we indicate the connection between plumbing and

the resolution of singularities of 2-dimensional complex varieties.
We shall express complex dimension by superscripts in parentheses.
Let

M = M(n)

be a complex manifold, not necessarily compact. A com-

plex analytic subset is said to have maximal dimension if it has complex co dimension

N is given locally

Recall that such a subset

as the set of zeros of a holomorphic function; that is, to each point

pe: N

there is an open neighbourhood

phic function
ponent of

defined on

p,

otherwise

in

and a holomor-

U and not identically zero on any com-

NnU = {qe:Ulf(q) =O}.

such that

called regular if the ahove

about

U of

can be chosen as a

The point

pe:N

is

function

coo~dinate

is called singular.

The complex analytic subset

N is called irreducible if it is

not t.he union of two smaller non-empty closed analytic subsets. A

divisor

on

is a formal linear combination

D

=~
"'"n.N.
~ ~

of irreducible closed analytic subsets of maximal dimension in

with integer coefficients.

A divisor

D can also be defined by an indexed family

of meromorphic functions, where each

75

fi

{fi}i e: I

is defined an an open

76
subset

U
i

of

(a)

{uilieI

(b)

fi

(c)

fi/fj

and
is an open covering of

M,

is not identic~ly zero on any component of

is holomorphic and has no zeros on

Ui '

and

uinu j

We express this by

D _ {f }. Note that condition (c) implies that

i
the locus of zeros and poles of the family {f. } is well defined. This
~

locus, together with the multiplicities of its irreducible components

(positive multiplicities for zeros, negative for poles) describes the
divisor

D according to the previous definition.

If, in particular, each function in the family

{f }
i

is holomor-

phia, the ccrresponding divisor

D _ {f } is called non-negative, and

i
is called positive if in addition the locus of zeros ND is not empty.
Perhaps the simplest non-negative divisor is one given by a single
globally defined holomorphic function, i.e.
divisor we simply wri te
To each divisor
bundle

[D]

over

=(f)

and

D _ {fi}ieI

ND

D _ {f}.

= If =0 I

For such a

there is an associated complex line

M given by the transition functions

The characteristic class

class

c ([D])
1

c (D) of D is defined to be the Chern

1
of the bundle [D]. If D
(f), then clearly [D]

is trivial and hence

c 1 (D)

=0

Recall that the usual singular homology

compact supports. Let

H.

H.

is a theory with

denote a homology theory with closed sup-

ports (see for instance Borel Moore [5]. A singular theory of this
type, defined on the category of locally compact spaces and proper
maps, is given by allowing infinite, but locally finite chains). The
main property of

K.

that we need here is that there is a Poincare

77

duality isomorphism

for any (not necessarily compact) oriented m-manifold

M without

boundary.
Using this, one can generalize the intersection numbers of cycles
to the case where one cycle may have non-compact support. Namely if
x e:

K . (Mj2Z)

and

m-~

y e:

H~

(M;2Z) ,
x.y

define

= 6(x) (y)

that is, the result of evaluating .the cohomology class

homology class

y.

A(x)

on the

Intersection of cycles is then defined via the

homology classes they represent. One can check that this gives the
usual definition in the compact case.
Returning to the complex manifold
written as

= ~niNi'

if

D is a divisor

then each irreducible component

Ni

has

a fundamental homology class in the sense of Borel and Haeflinger [1],

so

D represents a homology class

h(D) e:

H2n _2 (Mj2Z).

Borel and

Haeflinger proved:
LEMMA (9.1):
h(D) e:

If

D is a positive divisor, then the class

X2n_ 2 (M;2Z) is dual to '!;he characteristic class

11

:a2(Mj2Z).

This clearly follows also for arbitrary divisors, but we shall

not need this.
Now suppose
so

h(D)

= (t).

Then by the above, .oh(D)

= 0 , so certainly h(D). x

considering

D as a cycle on

= 0

for any

M we have shown:

= c 1 (D)

x e: H (Mj 2Z).
2

= 0 ,

Hence

78
If

COROLLARY (9.2):
D.x

D = (f),

vanis.hes for any cycle

Now let

with compact support in

be a point in the complex manifold

plex line elements at

Cp(n-1)

then the intersection number

M(n).

11

The com-

in

form a complex projective space

One can "replace"

in

complex manifold

M.

"'pM;

M by

Cp(n-1)

to

obtain a new

this process is known as a '" - process,

blowing-up operation or quadratic transformation.

To make this process precise we take a coordinate neighbourhood
U

of

in

centered at

given by

M with local coefficients

p

(Le.

z(p) =0).

CP(u) = <z1 (u), ,zn (u

r
be the graph of

f,

z= (z1' ,zn): U~Cn

There is an obvious map

(homogeneous coordinates). Let

c U X Cp(n-1)

and
K : = {p} X CP (n-1) c U X O:P (n-1)
p

is a non-singular analytic subset of

N=T'UK
P
Indeed, if w ' ,w
are homogeneous. coordinates in
1
n

Then

be the open subset of

let
cover

U X Cp(n-1).

U X Cp(n-1)

given by

By normalizing so that

w.

=1

Cp(n-1) ,
The
we can

consider

to be local coordinates in
N

is given by

Vi

With respect to these coordinates

79

w1 ' ,wi _ 1 ,zi'w + 1 , ,w n

i

SO

give a system of local coordinates

N n Vi

in

Using the projection T'~u- {p}

r;;;

N- K
p

in the disjoint union of

It is not hard to see that

d' M
p

we identify

M - {p}

and

depends only on

U- {p}

to obtain

M and

with

<5 M

P
and not

on the various choices used in the construction. There is a projection

11: : if M~M

which maps
M-

onto

and maps

biholomorphically onto

ifM-K

{p}

Complex Dimension 2
In the following
Let

M will always have complex dimension

pe: M with local coordinates

on a neighbourhood

of

in

z1' z2
M.

(centred at

p)

2.
defined

In the complex manifold

these coordinates are replaced by two systems

u,v

and

u,~

6 M
P
such

that

as one can see by

(9.3). The subspace Kp

locally by the equations

K
p

=0

and

=0

of

t:J' M is clearly given

p

It is also clear that

is a 2-sphere embedded in d' M and hence represents

compact support in if M

cycle with

80
LEMMA (9.5) :

Proof:
ing,

The self-intersection number

Consider the function

is a function defined on

a function on

~;1 (U)

given in the

(u,v)-chart by

z1

on

K oK

P P

n- 1 (U)
p

-1.

(Strictly speak-

U only. When we consider

we actually mean the function

u

is

and in the

z10TCp

as
which is

(u,v)-chart by llV

This is a very convenient abuse of notation which we have already used

in (9.4) and shall use again in (9.7).)
The divisor
in other words
lu=OloK

1 z1 = 01

in 1"C- 1 (U)
P

1z1 = 01 = K + lu= 01
p

is clearly

+1

consists

and

I'u = 01

Now the intersection number

so applying (9.2) to the open manifold

gives
0= Iz 1 =01oK = K oK + lu=ol"K = K oK +1
p
P P
P
P P
This clearly proves the lemma.
Remark (9.6):
M=

and consider

The

If we ignore the complex analytic structure on

blowin~up

11

as a

C~-manifold,

then

0" M Pick a point P2 in

P1
up at this point to give ()'
M:= (j' 6' M
P1 P 2
P2 P1

K1

and

K2

process may be iterated. Let

M and consider

then

M is diffeomorphic to

p= P1

be a point in

K
TC;~ (p) and blow
P1
by K. ;
Denote

~i

intersect at exactly one point and the projection

:re

:=Tt" on:
: O'.p p M~ M maps K1 U K2 onto p. We now olow
P1 P 2
P2 P1
1 2
up at a point P3 e: K1 U K , and so on. We finally reach a complex
2

manifold

1t

= cfP1 ... P

K1 U U Ks.

M and projection Ti':'H~M with ;::r1(p) =

s
Notice that Ki and Kj (i I j) are either disjoint

or intersect at one point regularly.

81
We can construct a weighted tree
v l'

K.~

by joining

t Vs

K.

and

v.

and

tion numbers of the

the open mani fold
in

vertices

by an edge in

v.
J

K.~

in

T
v.

i f and only if

with the self-

This tree is called the dual

;::t1(p)

weighted tree of the "spherical space"

intersect, and weighting each vertex

-intersection number of

of

with

The self-intersec-

can easily be computed by applying (9.2) to

K.

1i- 1 (u)

where

is a coordinate neighbourhood

M.

We are now ready to investigate how the resolution of singular"~iemann

ities in the

surface" of an algebroid function by means of

the blowing-up process is related to the plumbing operation studied

in

8.

An example should suffice to clarify the situation.

Example (9.7):
ction

Let

= (z~ + z~) 1/2

M = (1:2

and let

defined on

M.

singularity at

= or -1( P1)

of the Riemann surface onto

by blowing
Let

M up at

= z31 + z42

P1

of

or:

where
M.

that is, the complex

has a non-uniformizable

Rf~M

is the projection

Then the locus

Iw= 01
f

is the complex analy-

occurs. Blow up

if M consider the local coordinates

u,v

q'P1M

as a function on

in the proof of (9.5. The divisor

expressed by

lu 3 (1+uv 4 ) =01

=01

is

P1
given by (9.4). We consider

1~3('f+lr3)

f;

P1 = (0,0)

We want to resolve this singularity

tic subset along which the branching of

and in

be the algebroid fun-

The origin

the only non-uniformizable singularity of

2-dimensional ''Riemann surface"

Iw=OI

in c1'p1M

and

M at
u,~

(see remark
is clearly

in the (u,v)-chart and by

in the (lr,'f)-chart. It thus has two irreducible

82

componen ts, namely

1u

=01

and

which occurs with multiplicity

is a curve which meets

K1

Iv=OI
3,

give the 2-sphere K1

~1 '
4
3
and 1(1+uv )=01 = l(v+u )=01
(u,v) = (0,0).

in the point

We represent

this divisor by the diagram

3 __

-------1-(~IV-+-U-3-=

__
0_1_

- K1

where the integers

----

and

represent the multiplicities of the

irreducible components indicated. Now the non-uniformizable singularity

of

1 2
f = w / (considered as a function on

where

(u,v) = (0,0).

M) is at the point P2
P1
Since this point does not appear in the (u,v)6

-chart, we need only consider the restriction of

-chart, that is we only consider
at
u ,v
1 1

u= u 1

satisfying

w to the (u,v)-

v 3 (v + u3 )
take local coordinates

and in
,v= u 1v
1

(for reasons similar to those

given above, the other chart may be discarded). In this chart

I w = 01

or in a diagram:

is given by

Iv 1 + u 1 = 01

Next let

be a point where

and consider the charts

(u 2 ,v 2 )

and

(u 1 'v 1) = (0,0).

(u2 'v2 )

in

Blow up at

6p1P~3M

P3

satis-

83
is given by

Iw= 01

and

respectively.

IU 2 + v 2

Finally we blow up at
manifold
by

IU

M = 6'

P1 P 2 P 3P4

12 3(
v3 v3 + 1) = 01
3

=01

P4

where

(u ,v )
2 2

= (0,0)
Iw = 01

in which the divisor

8 12
11r3 ~ 3 (1 + 1r ) = 0 I
3

and

to obtain a
is expressed

In a diagram this

is
1

The function
f

= w1/2

occurs only along

K1

in

and

L .. I v3 + 1 = 01

are odd. In the Riemann surface

is lifted to a curve
out.

K4'

K4

is now uniformizable. Branching of

Rf

of the function

with two branches

is a two-fold branched cover of

points (the intersection points with

a 2-sphere;

K;

covers

K1

where the multiplicities

once and

K1

K;
K4

and

and

f
L'

on

M,

K4

sticking

with just two branch

and is henc e again

L' covers

once. Thus the

84

divisor

D =

I f= 01

Rf

in

can be expressed by the diagram

1
I

11
1
1
1
1

K'4

I
1

L'

where

and

K;

the loci except

space

I
I

K'

cover

K
2

AJ.l

L' are 2-spheres. The dual graph of the the spherical

K; U UKG

is just

E6

(with vertices renamed)

KioKi

We now use (9.2) to compute the self-intersection numbers

(i

=1,

,6).

Note that if

K:

and

(i ~ j)

K'
j

then they intersect in one point regularly, so

K L' = +1.

K:.K'. = +1
~
J

= 6 + 3K;'K;
o = K4D = 3 + 1 + 4 + 4 + 6K4K4,
= K;.D

K;.K; = K4oK4

showing that

K:K: = -2
~

= -2.

-2;

also for the remaining curves. Thus the dual

K

= K; U

let us denote this tree by

It is now clear that if

the Riemann surface

Rf

The reader will have no trouble in

weighted tree of the union of 2-spheres

weighted by

AJ.so

We have for example

Therefore

are not disjoint,

UK6

is

(E6 ;-2)

V is a tubular neighbourhood of

of (the modified)

E6

then

in

V is diffeo-

85

morphic to

P(E ;-2)

6
~V ~ s3/T,

particu1ar

We have investigated this space in 8. In

T'

where

is the binary tetrahedral group.

4) 1/2
f = (3
z1 + z2

We now return to the unmodified function

(1:2
onto

be the projection of the Riemann surface

and let

C2

B be the unit ball in

1
f- (O) = q

neighbourhood of the singu1ar point

R
f

is the set of points

Then
of

is a

R
f

dary of this neighbourhood is clearly diffeomorphic to

Note that for this

(I:

xC =

a: 3

The boun-

~V ~ S3/T,.

is isomorphic to the graph of

( z1,z2,z3) e:

on

satisfying

f,

that

3
4
2
z1 + z2 = z3

We have thus calcu1ated the diffeomorphism type of a "neighbourhood

boundary" of the singu1ar point
Exercise (5.8): Let
/

z; +

z~

, /

z1 (z~ +

z~)

q= 0

M = (1:2.

of this surface in
Show that for

(n ~ 2) , / z; -

z~

(n ~ 2) ,

f ..

a:3

/z1(z~+ z~)

the dual trees

obtained by the preceding process are respectively

An_1'

every tree weighted by

-2.

In general, resolving a singularity of an algebroid function

amounts to replacing the singular

defined on a complex manifold

point

in the "Riemann surface" of

K ,K ,
1 2

pair

(di~Ki = 2)

,Kn
i,j

(i

j)

the curves

by a finite number of curves

of various genera, such that for any

K.~

and

are either disjoint or

intersect in one point regularly. Thus the intersection relations give

rise to a dual weighted graph, where each vertex corresponds to a
curve; furthermore this graph is connected. The quadratic form associated with this dual graph (in the same way as that associated with a
tree) may be represented by the intersection matrix
where

0(. .
~J

We have.:

M = (~ij) ,

86
THEOREM (9.9)

The quadratic form

defined above is negative

definite.
This is essentially a classical theorem (see Du Val [27J). The
proof we are about to give is due to D. MUmford [19J.
Proof: Let

1 = f - f(p)

where

is the singular point of

m.

the Riemann surface. Then the multiplicity

the divisor

11=01

of the curve

is a positive integer for each

in

i=1, ,n

Define a quadratic form

2:.o(~ .x. Xj

S' _

~J

~,J

Clearly
if and only if

is. negative definite

S' is so. Notice that

(9.10)

if

i~j

In virtue of (9.2) we have

2:o<~. = C[m.K.)om.K. So 0
i
~J
i
~ ~
J J

(9.11)

(j

= 1, ,n) ,

and

LO<:
j
i
~

(9.12)
Write

< 0

for some

j.

S' in the form

L<><i'.x.x.=

. j

~,

J ~ J

L(Lo(j)X~- LO<i'(x.-x.)2
j

'<j

It is then clear that (9.10) and (9.11) imply that

S is negative

semidefinite. To prove definiteness equate the above expression to

zero. Then considering the first term on the right hand side, (9.12)
shows that

x. =0
J

for some

Considering the second term, the

fact that the dual graph is connected proves that

= x n .11

10. A THOOREM OF KERVAIRE AND MILNOR.

In this section we prove theorem (6.5) which was stated without

proof in 6. This theorem and the proof given here are due to Kervaire
and Milnor [13J.
Let

M be a compact oriented differentiable 4 -manifold without

boundary. Let

2
H (Mj2Z)

re:

and let

2
de:H (Mj?li)

duality

~ H2 (Mj?li!2?li)

,be such that

be the reduction modulo

1rd=W (M).
2

corresponds to an element

Under Poincare

be: H (Mj?li).
2

Theorem (6.5)

states that
THEOREM:
ding

f:

If

S2~M

is congruent to
Proof:
f(S2)
-bundle

in

then the self-intersection number

~(M)

bob = (d"d)[M]

modulo 16

First let us consider the case where

bb = -1.

M has normal bundle associated to the principal

J(_1)

defined in 8. In other words

tubular neighbourhood
normal

can be represented by a differentiable embed-

S 1-bundle of

M1 = (M - Int A) U e 4
is attached to

~A

whose boundary

f(S2)
t

where

= S3

~A t

f(S2)

M a

S3 ~ S2.

is a 4-cell whose boundary

by the identity map. Then

87

SO(2)-

considered as the

is the Hopf fibration

e4

has in

Then

fJe 4

Put

= S3

88
Clearly
-reM)

w2(~1)

=0

and hence

1:(M );;: 0 (mod 16)

1

= 1:'(M1 ) + 1:'(_o:p(2 = 't"(M1 ) -

1,

by (6.2). Since

the theorem is verified for this

particular case.
bob = s

In the general case we may assume that

the orientation of
copies of

_CP(2)

M if necessary. Let

0,

P1' ,P s+1

by reversing
be

s+1

and let

Using the natural isomorphism

Let

where

denotes a generator of
cc

= bob

H (P ;221)

s+1
+ Lg.og.
i=1 ~ ~

Then

= -1

Using the hypothesis that

embedding of

it follows easily that

S2

in

can be represented by a differentiable

ted by a differentiable embedding of

S2

in

M'.

can be represen-

Since

is

clearly dual to a cohomology class whose reduction modulo 2 is

w (Mr) ,
2

the special case of the theorem that we have just proved

shows that
-r(M') - -1
Since

-r(M') = -reM) - (s+1) ,

we deduce that

1:"(M) ;:

as was to be proved.

11

(mod16)

(mod 16 )

89
EXAMPLE (10.1):

Let

M = S2 X S2

standard generators. Then

2
H (M;:iZ) = 2Z19:iZ
SM

of

and

D(

,p e: H2 (M;2Z) be the

together form a basis of

and with respect to this basis the quadratic form

M is given by the matrix

The reduction modulo

w (M)
2

and let

if and only if

case and let

of an element
m

and

m= 2m' , n = 2n'

mC<+ np e: H2(M;2Z)

is equal to

are both even. Assume this is the

Using the relation

it follows that the dual homology class of

mc<+ np

can be represented

by a differentiably embedded 2-sphere only i f

8m'n';: 0 (mod 16) ;

2<>(+ 213

is not representable

that is

m'n'

is even. In particular

by a differentiably embedded 2-sphere in

EXAMPLE (10.2)

Let

M = G:p(2)

M.

and let

g e: H2 (M;:iZ)

standard generator. The dual homology class of

be the

is obviously rep-

resented by a differentiably embedded 2-sphere ; however the dual

class of

3g, for example, is not. Indeed, an element

nge: H2(M;:iZ)

satisfies n(ng) = w (M) if and only if n is odd. Consider the

2
2
element (2k+1)ge: H 0i;2Z). If its dual homology class is representable by a differentiably embedded 2-sphere then
that is

k 5 0 or 3 (mod

(2k+1)2

=1

(mod16);

4)

On the other hand, the dual homology class of every element

can be represented by a combinatorially embedded 2-sphere in
..
p(2).
~

T0 see this let

Cp(2) ,

and let

zo,z1,z2

P(zo,z1,z2)

ng

M=

be homogeneous coordinates in

be a homogeneous polynomial of (total)

90
degree
G:p(2)

in

zo' z1 and z2'

This polynomial defines a divisor of

in the obvious fashion (e.g. in the chart where

the function

~o

consider

P(1,z1/zo,z2/zo' and the homology class of this

divisor is represented by the algebraic curve

Q(zo,z1,z2)

Zo

P = O.

If

Q=

is another homogeneous polynomial of the same degree,

p/Q is globally defined on

then the meromorphic function

, so

as a divisor it gives the zero homology class. In other words

and

Q represent the same homology class.

Since the dual class of
Zo = 0,
to

for

ng.

of degree

is represented by the projective line

the class represented by

P=0

is dual

In particular let

The curve

P=O

has just one singularity, a cusp at

(1,0,0).

Using

the classical Plucker formula

genus = (n-1) (n-2)/2 - local terms
one sees that this curve has genus zero (the multiplicity
curve at the cusp is

n-1 ,

in the 4-manifold

Cp(2)

P=O,

-1)/2
P P

so the local term is

(n-1)(n-2)/2 ). Thus the curve

m (m

of the

considered as a real surface

is a 2-sphere and is clearly combinator-

ially embedded. Another way of seeing this is by checking that the

map of

maps

c:J. 1 ) into

CP(2) given in homogeneous coordinates by

bijectively onto the curve

P= 0

We have thus represented the dual class of

ng,

n~

1 ,

by a

combinatorially embedded 2-sphere. This is trivially also possible for

91
n

=0

and for

n ::; -1

one c an clearly do it by reversing the ori en-

tation of the embedded sphere.

Remark (10.3):
8

X8

Kervaire and Milnor proved also for the manifold

that any 2-dimensional homology class can be represented by a

combinatorial embedding of

REFERENCES

[1J

A. Borel, A. Haeflinger, La classe d'homologie fundamentale d'un

espace analytique, Bull. Soc. Math. de France 2 (1961)
461-513.

[2J

A. Borel, F. Hirzebruch, Characteristic classes and homogeneous

spaces I, Am. J. Math. 80 (1958) 459-538.
ditto IT, Am. J. Math. .:!. (1959) 315-382.

OJ

m,

[4J

------

[5J

A. Borel, J. Moore, Homology theory for locally compact spaces,

Mich. Math. J. 1 (1960) 137- 1 59

[6J

N. Bourbaki, Algebre, Hermann, Paris, 1952.

[7J

J.W.S. Cassels, Uber die Aquivalenz 2-adischer quadratischer

Formen, Comm. Math. Helv. 21 (1962/63) 61-64.

[8J

J. Eells, N.H. Kuiper, An invariant for certain smooth manifolds,

Annali di Mat. pura et apple 60 (1963) 93-110.

[9J

P.J. Hilton, S. Wylie, Homology theory, Cambridge, 1960.

[10J

M.W. Hirsch, The imbedding of bounding manifolds in euclidean

space, Ann. of Math. 74 (1961) 494-497.

[11J

B.W. Jones, The arithmetic theory of quadratic forms, Carus

Math. Monograph No. 10, New York, 1950.

[12J

M.A. Kervaire, J. Milnor, Bernoulli numbers, homotopy groups,

and a theorem of Rohlin, Proc. Internat. Congr. of Math.
(Edinburgh 1958) 454-458.

[13J
[14J

ditto

Am.

J. Math. 82 (1960) Lr91-504.

On 2-spheres in a 4-manifold, Proc. Nat. Acad. Sci.

U.S.A. 49 (1961) 1651-1657.
J. Milnor, On simply connected 4-manifolds, Symp. Int. de Topol.
Algebraica, Mexico 1956, (Mexico 1958) 122-128.

[15J

Some consequences of a theorem of Bott, Ann. Math. 68

(1958) 444-449.

[16J

Differentiable structures (mimeographed), Princeton 1960.

92

93
[17J

------

Differentiable manifolds which are homotopy spheres

(mimeographed), Princeton, 1959.

[18J

------

A procedure for killing homotopy groups of differentiable

manifolds, Proc. Symposia in Pure Math. 2, A.M.S.
(1961) 39-55.

[19J

D. Mumford, The topology of normal singularities of an algebraic

surface and a criterion for simplicity, Publ. Math. No. 9
I.H.E.S., Paris, 1961.

[2OJ

R. von Randow, Zur Topologie von dreidimensionalen Baummannigfaltigkeiten, Bonner Math. Schriften 14, Bonn, 1962.

[21J

V.A. Rohlin, A new result in the theory of 4-dimensional manifolds (russian), Dokl. Akad. Nauk. SSSR 84 (1952) 221224.

[22J

H. Seifert, Topologie dreidimensionaler gefaserter Raume, Acta

Math. 60,(1933) 147-238.

[23J

H. Seifert, W. Threlfall, Lehrbuch der Topologie, Leipzig, 1934.

[24J

J.P. Serre, Formes bilineares symetriques entieres a discriminant

1, Sem. H. Cartan, 14 e annee (1961/62), Exp. 14.

[25J

N. Steenrod, The topology of fibre bundles, Princeton, 1951.

[26J

E. Stiefel, Richtungsfelder und Fernparallelismus in Mannigfaltigkeiten, Comm. Math. Helv. ~ (1936) 3-51.

[27J

P. Du Val, On absolute and non-absolute singularities of algebraic surfaces, Revue de la Faculte de Science de l'Universite d'Istanbul (A) 91 (1944) 159-215.

[28J

J.H.C. Whitehead, On incidence matrices, nuclei and homotopy

type, Ann. of Math. 42 (1941) 1197-1237.

APPENDIX I.

In this appendix we comment on the subject matter of these notes

from a 1971 viewpoint, thus giving an indication of newer developements in the field.

1. Definition of

~adratic

Nowadays the space

Forms.

V on which a bilinear form is defined is

usually allowed to be a finitely generated projective (rather than

free)

A -module, in order to assure that a direct summand of

again of the same type. If

V is

A is a principal ideal domain, as is the

case for all rings considered in these notes, then finitely generated
projective A -modules are free, so this is no change.
If

V is a finitely generated projective A -module then a

quadratic form

on

is defined as a map
q:

such that

q (0( x) "' 0<. q (x)

is bilinear;

(0{

E:

V~A

and

A)

b (x,y) := q(x+y) - q(x) - q(y)

q

is called non-singular or non-degenerate according

as the associated bilinear form

is non-singular or non-degener-

ate. A form as defined in 1 is simply called a symmetric bilinear

94

95
form.

If

is a unit in

A,

then the two definitions are essen-

tially the same, since a quadratic form

bilinear form

(1)

b(x,y)

determines a symmetric

and vice versa by

= b q (x,y)

q(x+y) - q(x) - q(y)

q(x) = ~1 b(x,x)

(2)

However if

2 t A*

then the theory

f quadratic forms and the theory

of symmetric bilinear forms diverge.

Example:

If

= 221

or

2212

(2-adic integers) then (1) and (2)

define a one-one correspondence between quadratic forms and even

symmetric bilinear forms, so over

2Z and

2Z2 the theory of quadratic

forms is essentially the theory of even bilinear symmetric forms.

In view of the above comments, when we remarked in 1 that the

Hilbert symbol is a non-degenerate quadratic form over

F2 '

we

"meant" symmetric bilinear form, and in 4 it was the Grothendieck

ring of non-singular integral symmetric bilinear forms (not non-singular quadratic forms) which was calculated.

2. The Grothendieck and Witt Groups.

B,y the above comments, the ring

G (A)

defined in 2 would

nowadays be called the Grothendieck. group of non-singular symmetric

bilinear forms, which we shall now denote by KU (A).
o

The analogously

defined Grothendieck group of non-singular quadratic forms is generally

denoted by

GU (A)
o

Assigning to a quadratic form

the associated

96
symmetric bilinear form

leads to a map

GU (A) ~ KU (A) ,

which is an isomorphism if
The quotient groups of

2 e: A*
KU (A)

and

obtained by setting

hyperbolic forms (see Appendix lI) equal to zero are called the Witt
groups and denoted by

B(A)

and

W(A)

respectively. The theory of

Grothendieck and Witt groups has taken enormous strides since these
notes were first written. In particular many connections with other
fields such as algebraic number theory and algebraic K-theory have
developed. We refer the reader to Appendix II

(by W. Scharlau) for

more information.

3. Integral Forms.

In 4 we stated that little is known about the classification of

definite unimodular integral forms. This is still true, though a little
more is known now. We refer the reader to the charming book "Cours
d1arithmetique ll by J.P. Serre [29J

for a brief discussion, see also

Niemeyer [22J.

4. Symbols.

In 2 we used the Hilbert symbol to calculate the Grothendieck

there denoted by
More generally if

G(~).

is a field, a symbol on

is a

97
bimultiplicative symmetric map

F*
into an abelian group
for

1.

Hasse - Vlitt

xF*~

C
2

satisfying

(a ,b) "' 1 ,

and (a,1-a) "' 1

Given such a symbol one defines a Hasse - Minkowski

invariant

c(f)

via a diagonalization of the form

or
f

by

c(diag(a , , a : = I T . < . (a.,a.)

1
n
~
J
~
J
The methods of 2 can then be generalized. For instance it is not
hard to see that every Hasse-Minkowski invariant factors over the map
c': G(F) ~ L(F)

defined in the same way as

in 2,

so if this

map is itself a Hasse-llinkowski invariant then the result of corollary

(2.9) applies.
There is a universal symbol which turns out to be given by the
algebraic K - group

K (F) ,
2

which leads to a universal Hasse-Minkowski

invariant (see Milnor [20J). However

K2

is generally difficult to

calculate, so one often works with a concrete symbol with values in

the Brauer group of

F, given by setting

the quaternionic algebra of

(a, b),

(a,b)

equal to the class of

see for instance VI. Scharlau

[27]

5. Signature and the f/.-Invari an t

In the proof of the invariance of the

)J-- invariant we proved a

special case of the following "addi tivi ty property" of signature. If

M is a compact oriented manifold obtained by pasting two compact

98
oriented manifolds

M1

and

M2

together along boundary components,

then

This property was first observed by S.P. Novikov;

a short proof (also

in the equi varian t case) can be found in Atiyah Singer [3 J , see also
Jani.ch [14J.

This property was fundamental in the definition of the

r- invar-

iant. In a similar way, any additive invariant defined on some set of

manifolds and zero on closed manifolds defines an invariant for
boundaries of manifolds in the set, so long as the set is closed under
the operation of pasting manifolds along their boundaries. In conjunction with the Atiyah - Singer index theorem [3 J this has proved useful
in obtaining interesting invariants for manifolds with group actions
or other structure. See for instance [2J, [10J, [11J, [19J, [25J.
The minor application of the

r- invariant

to detecting h-cobord-

ism type of lens spaces in special cases (7) has been superceded by
a complete h-cobordism classification (Atiyah Bott [ 1 J p479). It turns
out to be the same as the diffeomorphism classification. A general
reference for this sort of classification problem is C.T.C. Wall's
book [3OJ

and the literature quoted there.

6. Plumbing and Spheres.

As remarked in 9, the boundary of a manifold

= P(T,m.)
~

obtained by plumbing n-disc bundles over the n-sphere (n>2) is a

99
homotopy sphere if and only if the intersection form
equal to the form of the weighted tree
%1.

(T, e(m.
~

SY'

which is

has determinant

A lot is known about smooth manifolds which are homotopy

k:a:. 5

k-spheres for

([ 16J, [7 J, [18J). They are all homeomorphic to

the standard sphere

Sk

but not necessarily diffeomorphic. All the

possible differentiable structures on the k -sphere form a fini te

group EJi

with respect to the connected sum operation

#=.

Those

spheres which bound a stably parallelizable manifold represent a cyclic

bP + c @k
k 1

subgroup
2

for

which has order

for

k;;; 0 (mod 2 ),

or

k '" -1 (mod 4 ), and rapidly increasing order for increasing

k _ 1 (mod 4).
Clearly, if one only plumbs stably trivial bundles, then
P(T,m.)
~

is stably parallelizable, so

an element of

bP

2n

Sy

of the tangent disc bundle of

A2

generator of

bP

if a sphere, represents

In this case the element is easily determined

from the intersection form

(n even) or

dY,

for instance if one plumbs copies

Sn

(n ~ 3)

according to the tree

(n odd), the resulting sphere

2n

Y =

dY

E8

represents a

A convenient reference for plumbing of spheres

is Hirzebruch Mayer [12J

7.

Plumbing Seifert Manifolds.

In

we described

von Randow's result on obtaining Seifert

manifolds by plumbing. To apply this to calculating the

one must alter it slightly. The Seifert manifold

~-invariant

with invariants

100
is a

1li/22Z - homology sphere if and only if

is odd, as follows immediately from Seifert's results [28J. For any

Seifert fibration one can drop the condition
Sei fert invariants and al ter each fJ i
taniously alters

modulo

0 < f3i < o(i

0(.

on the

if one simul-

to keep the above expression (1) constant, the

resulting IIgeneralized ll Seifert invariant set still determines the

Seifert fibration in a well defined way [21J. If one now makes the
restriction

O<{3.~ <20<.~ , von Randow's result can be applied to the

generalized invariants with no formal change to obtain alternative

methods of obtaining

X by plumbing. In particular if

2Z/22Z - homology sphere and one o(i

represent

X as the boundary

X is a

is even one can use this to

X = dY

with even intersection form (i.e. the

of a manifold
m
i

Y = P(T,m.)
~

are even), which can hence

be used to calculate jt(X)

8. Resolution of Singularities and Plumbing.

In 9 we illustrated the uniformization of singularities of

algebroid functions in two variables by means of the example

z~ + z~

The relatively naive method of this example - blowing up

"nasty" points (points where the "Riemann surface" has a non-uniformizable singularity) until none are left - will work for instance for
functions of the form

f(z1,z2) with

breaks down for the function

(z1~-q)1/n

a polynomial, but already

(O<q<n, n>2, (n,q)

=1).

However this function can be uniformized by replacing the singular

point of its Riemann surface by a "spherical space" (c.f. p.81)

101

of theorem (8.9). Further-

having dual graph equal to the tree (A ,m.)

s

more, it turns out that essentially the above "naive method" can be
used to reduce the general problem to this latter case, giving a general method of uniformizing such singularities. For details see
Hirzebruch [9 J.
As we indicated in

broid function

9, uniformizing the singularity of the alge-

z~ + z~

singularity of the variety

is equivalent to resolving the isolated

z~ + z~ = z; in (:3.

clearly isomorphic to

This variety is

More generally one can con-

X(a 1 ,a ,a ) given by z~1 + z~2+ z;3 = 0 in \1:3

2 3
which has an isolated singularity at zero. This singularity

sider the variety

(ai >1),

has been resolved when the

are pairwise coprime by Hirzebruch

i
Jwch [11J and in the general case by Orlik and . Wagreich [24 J. In
the pairwise coprime case the resul t is that the singularity can be
resolved by inserting a spherical space

s =

behaviour is given by the following dual tree

Here the integers

and

Us..

'"

whose intersection

(T,-bi):

are determined by the relations

j

, b s J
J

102
where the

f3 k

are given by

0 <fk < ~

Let L(a ,a ,a ) = x(a ,a ,a )

1 2 3
1 2 3
sphere in

~3.

and

ns5

where

S5

is the unit

The result quoted above shows that if the

are

L(a 1 ,a2 ,a ) is given by plumbing as ~(a1,a2,a3)

3
= dP(T,-br). This can also be seen as follows: L:'(a 1 ,a2 ,a ) has a
3
natural S1-action
pairwise coprime,

1
(t e: S ) ,

which gives ~(a1,a2,a3)

Seifert invariants

c~

the structure of a Seifert space. The

be calculated directly [21J, and one can then

apply von Randow's results quoted in these notes. By part 7 of this

appendix one can in this way also obtain alternative representations

cases (namely if one

a.

is even) one can do this with even

mi

We mention two applications: firstly one can calculate the

fL- invariant j.(.2":(a ,a ,a )

1 2 3

in many cases where it is defined.

Secondly and more interestingly, 2.:(a ,a ,a )

1 2 3
sphere if the

are pairwise coprime. Hence by theorem (8.2) the

quadratic form of the tree

1

is a :iZ - homology

(T,m.)
~

mentioned above has determinant

and can be used to plumb in higher dimensions to give homotopy

spheres (see Hirzebruch [10J and Mayer [19J for an application to

involutions on spheres). Of course if some

mi

are odd one needs

bundles with odd euler characteristic for this, which only exist in
dimensions

2, 4 and 8

103
necessarily pairwise coprime) one can "replace" the singular point by

~(a1,a2,a3)/s1,

which is a complex curve by Brieskorn and Van de

Ven [6 J. One is then only left with. singular points equivalent to

singularities of the form'

n
z3

= z1z2n-q

which can be resolved as des-

cribed at the beginning of this section. One thus also gets a star-shaped dual tree, but the central curve usually has higher genus and
the tree more branches than in the coprime case. The details of this
method are a bi t messy and Orlik and W.agreich (loc. ci t.) use a
similar but more economical method to resolve in fact all 2-dimensional
normal singularities which have an effective

C*-action.

In higher dimensions one has the analogous variety

'~+1)

X(a 1 ,

in

a: n+1

givenby

z~1+ +z::'1 1 =o

(ai >1),

which has an isolated singularity at zero. These singularities have

been studied extensively by Brieskorn and others;
the neighbourhood boundary Z(a1 , '~+1)

it turns out that

= X(a 1 ,

,an +1 ) ns2n+1

of such a singularity is often a homotopy sphere, and that all elements

of

bP2n

can be represented by such spheres [4 J, see also Hirzebruch

and Mayer [12J.

We mention that theorem (9.9) has a converse due to Grauert

([8 J p.367). Suppose that K = llK.~

is a union of compact irreducible

curves in a non-singular complex surface

connected and the bilinear form
of the

Ki

collapsing

is negative definite;
K

M(2)

such that

K is

defined by the intersection numbers

then the space

M' obtained by

to a point is a complex surface with at most an isolated

normal singularity at this point. Thus the classification of 2-dimensional normal singularities is essentially reduced to the determination
of the dual graphs which can occur, together with the possible complex
structures in a neighbourhood of such a union of curves

K.

This is

104
a very difficult problem and is not solved in general. A connected
account will appear in the notes by Henry Laufer [17J ; see also
Brieskorn [5 J for a discussion of the case of rational si ngul ari ties.

9. Orientation Conventions.

As mentioned in 8, von

Randow's orientation conventions for'

Seifert spaces and lens spaces are opposite to those adopted here, so
his weighted trees are actually the negatives of the trees we gave when
quoting his results on Seifert spaces and lens spaces.
In these notes we have tried to follow Raymond's orientation
conventions of [26J and [23J. This is also the convention adopted in
[21J. In the original mimeographed version of these notes (apart from
a sign error at one point) and in Orlik Wagreich [24J the opposite
convention has been adopted for Seifert spaces (but not for lens
spaces). This leads to compatible results on obtaining the manifolds
~(a1,a2,a3)

by plumbing, but negative values to the above for Seifert

invariants (except for lens spaces).

10. Representing Homology Classes by Embedded Spheres.

Using the Atiyah Singer equivariant signature theorem, W.C.

Hsiang and R.H. Szczarba [13J

have obtained results on representing

integral homology classes by differentiably embedded surfaces in

4-manifolds, among other things greatly improving the results of

105

examples (10.1) and (10.2). For

H2 (S2 X S2)

they show a class in

can be represented by a differentiably embedded 2-sphere

only i f it is dual to
coprime. For
classes of

2
2
S XS

~p(2)

me( + nf3

(notation of (10.1

classes in T(n(M2n)

can be so represented.

These two examples are both simply connected, so

n> 2

m and

they have complete results, namely only the dual

0, g, 2g

In dimensions

with

H2 (M) =

~2(M)

Kervaire [15J has complete results on representing

by differentiably embedded n-spheres.

REFERENCES
[1J

M.F. Atiyah, R. Bott, A Lefschetz fixed point formula for elliptic' complexes, IT Applications, Ann. of Math. 88 (1968)
451-491.
-

[2J

M.F. Atiyah, F. Hirzebruch, Spin-manifolds and group actions,

Essays on Tbpology and Related Tbpics, 17-28, Springer
Verlag, 1970.

[3J

M.F. Atiyah, I.M. Singer, The index of elliptic operators DI,

Ann. of Math .1 (1968) 546-604.

[4J

E. Brieskorn, Beispiele zur Differentialtopologie von Singularitaten, Inv. Math. ~ (1966) 1-14.
Rationale Singularitaten komplexer Flachen, Inv. Math.
~ (1968) 336-358.

[6J

E. Brieskorn, A. Van de Ven, Some complex structures on products

of homotopy spheres, Tbpology L (1968) 389-393.

[7J

W. Browder, The Kervaire invariant of framed manifolds and its

generalization, Ann. of Math. 90 (1969) 157-186.

[8J

H. Grauert, Uber Modifikation und exzeptionelle analytische

Mengen, Math. Ann. ~ (1962) 331-368.

[9J

F. Hirzebruch, Uber vierdimensionale Riemannsche Flachen mehrdeutiger analytischer Funktionen von zwei komplexen
Veranderlichen, Math. Ann. 126 (1953) 1-22.

[10J

------

Involutionen auf Mannigfaltigkeiten, Proc. Conf. on

Transf. Groups, New Orleans 1967, Springer 1968, 148-166.

106

[11J

F. Hirzebruch, K. Janich, Involutions and singularities, Algebraic Geometry, Papers presented at the Bombay Coll. 1968
Oxford University Press, 1969, 219-240.

[12J

F. Hirzebruch, K.H. Mayer, O(n)-l-'lannigfaltigkeiten, exotische

Spharen und Singularitaten, Springer Lecture Notes 57,
1968.
-

[13J

W.C. Hsiang, R.H. Szczarba,

Math., A.M.S.

[14J

K. Janich, Charakterisierung der Signatur von Mannigfaltigkeiten

durch eine Additivitatseigenschaft, Inv. Math. (1968)
35-40.

[15J

M.A. Kervaire, Geometric and algebraic intersection numbers,

Comm. Math. Helv. 39 (1968) 271-280.

[16J

M.A. Kervaire, J. Milnor, Groups of homotopy spheres I, Ann. of

Math. 77 (1963) 504-537.

[17J

H. Laufer, LectUre notes, Princeton University Press, to appear.

[18J

M. Mahowald, The order of the image of the J-homomorphism, Bull.

A.M.S. 76 (1970) 1310-1313.

[19J

K.H. Mayer, Fixpunktfreie Involutionen auf 7-Spharen, Math. Ann.

122 (1970) 250-258.

[20J

J. Hilnor, Algebraic K-theory and quadratic forms, Inv. Hath

(1970) 318-344.

[21J

\-l.D.

[22J

H. V. Niemeyer, Definite quadratische Formen der Dimension 24

und Discriminante 1, Dissertation, Gottingen, 1968.

[23J

P. Orlik, F. Raymond, Actions of SU(2) on 3-manifolds, Proc.

Conf. on Transf. Groups, New Orleans 1967, Springer,
1968, 297-318.

[24J

P. Orlik, P. Wagreich, Isolated singularities of algebraic surfaces with C* action, Ann. of Math. 22 (1971) 205-228.

[25J

E. Ossa, Cobordismustheorie von fixpunktfreien und semifreien

S1_ Mannigfaltigkeiten, Dissertation, Bonn, 1969.

[26J

F. Raymond, Classification of actions of the circle on 3-manifolds, Trans. A.M.S. 121 (1968) 51-78.

[27J

W. Scharlau, ~adratic forms, ~eens Paper on Pure and Appl.

Math. No. 22, Kingston 1969.

To appear in Proc. Symposia in Pure

.2

Neumann, S 1-actions and the 0( - invariant of their involutions, Dissertation Bonn, 1969, to appear as Bonner
Math. Schriften 44.

-,

- .. ~'Y_I"
C
__d,

_ .............
C',,. '....,
................

'.8.a,.... , ' 'a ....

"'lllnl ............. . .

..., _
lite 'It ..................1'1 .. .....
DI1 ..... ..-a.
~I"_"'''''''''''''''''''''

APPENDIX II

GROTHENDIECK AND \'IITT RINGS

W. Scharlau
Mathematisches Institut,
Universitat Munster, Munster, Germany.

1. Generalities.
Let

be a commutative ring. A non-singular symmetric bilinear

space over

projective

R-module

b:

MxM~

is a pair

(M,b)

consisting of a finitely generated

M and a non-singular symmetric bilinear form

The Grothendieck group of non-singular symmetric

bilinear spaces will be denoted by

KUo(R)

A non-singular quadratic space over

sisting of a finitely generated projective
ratic form

q:

M~

is a pair
R-module

(M,q)

M and a quad-

q(X+y) - q(x) - q(y)

is non-singular. The Grothendieck group will be denoted by

There is an obvious natural homomorphism
an isomorphism if

con-

whose associated bilinear form

b (x,y):=
q

is a unit in

GUo(R) ~ KUo(R)

GUo(R)
which is

R.

The theory of non-singular quadratic spaces can be developed

analogously to the developement of algebraic K-theory. This has been
done by H. Bass [ 2 J, A. Roy [16J, and above all A. Bak [ 1 J. The main

108

109

results are the analogons of the stability and cancellation theorems

of Bass and Serre. The theory can be developed in a more general setting by considering a ring

R with involution instead of just a ring;

one must then consider hermitian forms.

An important role is played by the hyperbolic forms, defined as

follows.
Let

M be a finitely generated projective R-module and

b:MXM~R

any symmetric bilinear form, respectively

q:M~R

any quadratic form, not necessarily non-singular. The space

or ]H{M, q)

is defined as follows:

the underlying module is

m{M,b)
MED M*

and the form is respectively

(x,f) X (y,g)

b{x,y) + g{x) + fey)

(x, f)

q{x) + f{x)

or

Both forms are non-singular. The hyperbolic forms are the special case

One checks that

m{M,b) E9mC}-1,-b)

~ m{M,o)

E,lJH{M,-b)

m{M,q) EPJH{M,-q)

~ m{M,o)

E9JH{M,-q)

and

so the subgroup of
m{M,b) or
]H{M,o)

KU (R)

m{M, q)

or

generated respectively by the

is already generated by the hyperbolic forms

The quotient group by this subgroup is called the Witt group and
denoted by

B{R)

and

Remark: KU (R)
o

W{R)
and

respectively.
B{R)

are rings via tensor product of forms

110
and

GU (R)

is in a natural. way a

KU (R) - module and

W(R)

B(R) - module.

2. Dedekind Domains.
From now on let

be a Dedekind ring. We first consider the

local. case, that is,

is a discrete val.uation ring. Let

uniformizing element, that is,

(Tr) ==

'1

char (F)

symmetric bilinear form

over

be a

is the maximal. ideal.. Let

be the quotient field and assume

b

T'C

12.

by non-singular

can be diagonal.ized, and by

multiplying the diagonal. coefficients by suitable squares one can

write

with

ai,b

in the form

e: R - , .

The form

well defined element of

'd 2 (b):

B(R/f)

= diag(b 1 ,

,bn)

gives a

and one obtains in this way exact

sequences

O~

B(R)

B(F)

~B(R/l)-+O

and
0 - + KU (R)

~KU

(F) LB(R/1O) - + 0

0 0 1 '

d2

(which depends on the choice of

class form.

The element

defined in the Wi tt group

~ ) is cal.led the second residue

d (b) : = diag(a 1 ,
B(R/1)

,~)

is al.so well

and is cal.led the first residue

class form.
THIDREM:

If

is complete and

canonical. isomorphisms

char(R/1)

then there exist

111
B(R) ';;: B(R/~)
Furthermore the above exact sequences split by means of the first
residue class form, so one has direct sum representations

Li terature:

T. A. Springer [22J. This theorem can be generalized

to complete semilocal rings: C.T.C. Wall [24J.

See also Knebusch

[5J, Scharlau [21J.

Now let

be an arbitrary Dedekind ring with quotient field

char(F) ~ 2

and

THIDREM:

(not a very essential restriction).

There is an exact sequence

o ~ B(R) ~ B(F) --4

d

where the map

1J E:

B(R/f)

Max{R)

is given by the second residue class forms.

~ : Knebusch [5J, Knebusch Scharlau [8J, Milnor [12J,

Frohlich OJ.
THEOREM:

where

GU (R)
o

There is a canonical exact sequence

denotes the ideal class group of

assigns to an ideal

ot

of

and the map

the element

C/C2~

JH(ot) - E(R)

of

GU (R).
o
Proof:

Kneser [9J, Knebusch [5J, Frohlich [3J.

For symmetric bilinear forms the situation is similar but there

are difficulties at the dyadic prime places (c.f. Knebusch loco cit.)

112
3. Algebraic Number Fields.
The calculation of the Witt group has been done so far only for a
few fields. One of the main results is
THEOREM (Hasse - Minkowski)

Let

be an algebraic number field.

Then the canonical map

is injective, where

ranges over all finite and infinite prime

places.
The image of this map can also be determined.
Li terature: O. T. 0 I Meara [13J, Wi tt [26J.
Let

R be the ring of integers in the algebraic number field F.

Then the first exact sequence of the preceding page can be extended to
an exact sequence

o ~ B(R) ---;. B(F) -4.L.L.

1E: Max(R)

B(R!y) ~ C/C2~O

detailed investigation of this situation, taking into account also

the infinite primes and the quadratic reciprocity law, can be found
in [8J.
Example:

The second residue class forms give an exact sequence

O~

B(2Z)

B(Q)

.LL

B(2Z/p2Z)

~O

p=2, 3, ...
On the other hand

so

B(~)

can be calculated directly as

B(2Z) 'at 2Z (application to the invariant b(w,w)! - see remark (5.9)

113
of the notes).
FUrther Literature: Frohlich [3J, Milnor [12J, Scharlau [aDJ,
C.T.C. Wall [25J.

4. FUnction fields.
A similar situation to the case
function fields

exists also for rational

F = k(t)

THEORE){ (Milnor [11]):

The second residue class forms yield an

exact sequence

o ~ B(k) --+ B(F) ~.l...L

B(k[tJ/p) --+0

p irred.

On the other hand,

B(F)--.llB(k[tJ/p).

B(k[tJ)
Since
B(k)

also lies in the kernel of

B(k)cB(k[tJ) ,

= B(k[tJ)

it follows that

This has also been proved by Harder (unpublished).

Little is known for arbitrary function fields in one variable.
The known results mainly concern the quadratic reCiprocity law, see
for instance Geyer Harder Knebusch and Scharlau [4J, Scharlau [21J.
However if ""the field of constants is finite or equal to

then

the theory is about as complete as in the algebraic number field case

5. Characteristic Classes.
One can define invariants for quadratic forms which are in a

114
certain way analogous to characteristic classes for vector bundles.
This can for instance be done by means of Galois cohomology in a very
similar way to how it is done for vector bundles, see Scharlau [17J,
Springer [23J. The definitions vary a little according to whether one
works with the Witt group or the Grothendieck group.
Literature: Witt [26J, O'Meara [13J, Scharlau [17J,[18J, Milnor
[11J.

6. Pfister Theory.
Let

be a field of characteristic

2.

The theorems of

Pfister and subsequent investigations tell us a certain amount about

the structure of
THEOREM:

B(F)

= W(F).

The torsion subgroup

W-z:(F)

of

W(F)

is a 2-group

and is precisely the kernel of the homomorphism

W(F) ~ IT (W(F...

where

W(F.. )

~ 2Z

runs through the real closures of

(observe that

by Sylvester' s law of inertia). In particular if

formally real then

Literature:

W(F)

is not

is a 2-group.

Pfister [14J, [15J, Lorenz [10J, Scharlau [18J,[19J,

Knebusch Scharlau [7J, Knebusch Rosenberg Ware [6J.

REFERENCES
[1J

A. Bak, The stable structure of quadratic modules, Preprint 1970.

[2J

H. Bass, Lectures on topics in algebraic K-theory, Tata Inst.

Fund. Res. Bombay, 1967.

115

[3J

A. Frohlich, On the K-theory of unimodular forms over rings of

algebraic integers, Preprint 1970.

[4J

W.D. Geyer, G. Harder, M. Knebusch, W. Scharlau, Ein Residuensatz fUr symmetrische Bilinearformen, Inv. Math. 11

(1970) 319-328.

--

[5J

M. Knebusch, Grothendieck- und Wittringe von nichtausgearteten

symmetrischen Bilinearformen, Sitzungsber. Heidelberger
Akad. Wiss. 1969/70, 3. Abh., Springer-Verlag 1970.

[6J

M. Knebusch, A. Rosenberg, R. Ware, Structure of Witt rings,

quotients of abelian group rings, and orderings of
fields, Bull. A.M.S o 11 (1971) 205-210.

[7J

M. Knebusch, W. Scharlau, Uber das Verhalten der Wittgruppe bei

KOrpererweiterungen, Math. Ann., to appear.

[8J

Quadratische Formen und quadratische Reziprozitatsgesetze

uber algebraischen Zahlkorpern, 11ath. Zeitschr., to
appear 1971.

[9J

M. Kneser, Unpublished manuscript, 1962.

[10J

F. Lorenz, Quadratische Formen uber Korpern, Springer Lecture

Notes, 1970.

[11J

J. Milnor, Algebraic K-theory and quadratic forms, Inv. Math.

[12J
[13J

(1970) 318-344.

Symmetric inner product spaces over a Dedekind domain,

Preprint, 1970.
O.T. O'Meara, Introduction to quadratic forms, 2nd. Ed., Springer

1971.
[14J

A. Pfister, Quadratische Formen in beliebigen Korpern, Inv. Math.

1 (1966) 116-132.

[15J

Quadratic forms, Cambridge lecture notes 1967.

[16J

A. Roy, Cancellation of quadratic forms over commutative rings,

J. Algebra 1Q (1968) 286-298.

[17J

W. Scharlau, Quadratische Formen und Galois-Cohomologie, Inv.

Math. ~ (1967) 238-264.

[18J

Quadratic forms, QueedS paper on pure and applied math.

No. 22, Kingston 1969.

[19J

Induction theorems and the structure of the Witt group,

Inv. Math. 11 (1970) 37-44.

[2OJ

Quadratic reciprocity laws, J. Number Theory, to appear.

116
[21J

W. Scharlau, Klassifikation hermitescher Formen uber lokalen

Korpern, Math. Ann. 186 (1970) 201-208.

[22J

T. A. Springer, ~adratic forms over a field with a discrete

valuation, Indagationes Math. 17 (1955) 352-362.

[23J
[24J
[25J
[26J

On the equivalence of quadratic forms, Proc. Acad.

Amsterdam 62 (1959) 241-253.
C.T.C. Wall, On the classification of hermitian forms nI,
complete semilocal rings, Preprint May 1971.
ditto :& , Global rings, Preprint June 1971.

E. Witt, Theorie der quadratischen Formen in beliebigen Korpern,

J. reine u. angew. Math. 176, (1937) 31-44.

AUTHOR INDEX
Numbers in square brackets are reference numbers and indicate
that an author's work is referred to although his name is not
cited in the text. Numbers that are underlined show the page
on which a complete reference is listed.
A

Atiyah, M.F.,
.1Q2

~,

19, 98, 98[2J,

Hilton, P.J., 43, 92

Hirsch, M.W., 53, ~
Hirzebruch, F., 18, 19, 43, 92,
98[2,10,11J,99, 101, 102,"
103,

Bu, A., 108, 114

Bass, H., 108, 114
Borel, A., 18, 43, 76, 77, ~
Bott, R., 98, .:!Q2:
Bourbaki, N., 37, 47, ,2g
Brieskorn, E., 103, 103[4J, 104,

1Q2

Browder, W., 99[7J, 1Q2

C

Cassels, J.W.S.,

1Q2, 106

Hsiang, W.C., 104, 106

~,

74, 92

J8n!ch, K., 98, 98[11J, 101, 106

Jones, B.W., 12, 32, 39, 92
K

Kervaire, M.A., 43, 87, 92,

99[16J, 105, 106
-Knebusch, M., 111~12[8J, 113,
114, ill
Kneser, M. 111, ill
Kuiper, N.H. 47, 92

Du Val,

P., 86,

22

Laufer, H., 104, 106

Lorenz, F., 114, 115

Eells, J., 47, ,2g,

FrOhlich, A., 111, 113,

G

Geyer, W.D., 113, ill

Grauert, H., 103, 1Q2

ill

Mahowald, M., 99[18J, 106

Mayer, K.H., 98[19J, 9~102,
103, 106
Milnor, J:-i+3, 48, 53, 57, 58,
59, 87, 92, 22, 97, 99[16J,
106, 111, 113, 114, ill
Moore, J., 76, 92
Mumford, D., 86, 22

Haeflinger, A., 77, ,2g,

Harder, G., 113, ill

Neumann, W.D., 100[21J, 102[21J

104[21J, 106

117

118

Niemeyer, H.V., 96, 106

Serre, J.P., 18, 32, 35,

106

o
O'Meara, O.T., 112, 114, 112
Orlik, P., 101, 103, 104[231,

104, 106

Ossa, E.,~[25J, 106

22, 96,

Singer, I.M., 98, 122

Springer, T.A., 111, 114, 116
Steenrod, N., 56, 59, 61, 22
Stiefel, E., 53, 22
Szc zarba, R. H., 104, 106
T

Pfister, A., 114,

122

Threlfall, W., 49[23J, 53, 55,

69, 93

Randow, R. von, 63, ~, 99, 104

Raymond, F., 104, ~
Rohlin, V.A., 43, 22
Rosenberg, A., 114, 122
Ror, A., 108, 122
S

Scharlau, W., 97, 106, 111, 112

[8J, 113, 114, 115, 116

Seifert, H., 49[23J, 53, 55,

63, 64, 69,

~,

100, 106

Van Der Waerden, B.L., 10

Van de Ven, A., 103, 105
W

Wagreich, P., 101, 103, 104, 106

Wall, C.T.C., 98, 106, 111, 113,

116

Ware-;-R., 114, 122

Whitehead, J.H.C., 51, 22
Witt, E., 6, 112, 114, 116
WYlie, S., 43, ~

SUBJECT INDEX

Additivity property of signature, 98

Algebraic number field, 112
Algebroid function, 81, 100
Augmentation, 19
Automorph, 2

Equivalence of forms, 2
Euler number, 59, 61
Even form, 25, 30
F

Form, quadratic, 1, 94, 108

symmetric bilinear, 94, 108
Function field, 113
Fundamental class, 77

Base of a lattice, 1
Blowing up, 78
Binary form, 2
Binary polyhedral groups, 63
Brauer group, 97

Genus of a form, 39
Grothendieck group, 17, 95, 108
of analytic sheaves, 18
of vector bundles, 18
Grothendieck ring, 19

Cancellation theorem for

forms, 6
Characteristic class of a
divisor, 76
of a form, 113
Characteristic element of a
form, 26
Classifying element of a
bundle, 56
Closed supports, homology, 76
Compact supports, homology, 76
Complex variety, 75, 101
Connected sum, 52
Correlation, 1

Hasse Minkowski invariant, 15,

23, 28, 97
Hasse Witt invariant, 97
h-cobordism, 52, 98
Hermite reduced form, 32
Hilbert symbol, 12
Hilbert product formula, 13
Homotopy sphere, 61, 98, 103
Hopf map, 61
HYperbolic form, 96, 109

Dedekind domain, 110

Definite form, 35
DET, 8
Determinant, 2
Diagonal form, 13
Diagonalization theorem, 4
Dihedral group, 62
Divisor, 75
Dual weighted tree, 81

Icosahedral group, 63
Indefinite form, 33
Index of a form, 8
Integral form, 25, 32, 96
Intersection number of cycles,
42, 77
Irreducible variety, 75
Isometry, 2

119

120

Quadratic reciprocity law, 112

Quadratic space, 108

Jacobi symbol, 72

<

K-group, algebraic, 97
Kronecker product, 1
K-theory, 18
K-theory, algebraic, 108
L

Lattice over a ring, 1

Legendre symbol, 9
Lens space, 49, 68, 98
Lie algebras, classification
of, 61
Linking numbers, 55
Local domain, 4, 8
M

Manifold, 42
Matrix of a form, 2
Minkowski invariant, 38
J{- invariant, 46, 62, 70, 97
N

Non-degenerate form, 1
Non-singular form, 3

o
Octahedral group, 63
Odd form, 25, 33
Orientation conventions, 104
P

Pfister theory, 114

Plucker formula, 90
Plumbing, 56, 66, 98
in dimension 4, 61
Polyhedral groups, 63
Principal bundle, 56

Rank 0 f a form, 2
of a lattice, 1
Rational form, 25
Real form, 25
Reflection, 6
Regular point of a complex
variety, 75
Residue class form, 110
Riemann surface, 85, 100
S

Seifert invariant, 63, 99

Seifert manifold, 63, 99
Semi-equivalence, 39
Singularity, singular point,
75, 100
Signature of a form, 25
of a manifold, 42, 97
Sphere, homology-, 46
homotopy-, 61, 98, 103
Spherical dodecahedral space,
53, 64
Spherical space, 81
Stiefel Whitney class, 43
Sylvester's law of inertia, 8,
114
Symbol, 96
Symmetric bilinear form, 94, 108
bilinear space, 108
a"- process, 78
T

TYPe of a form, 35
Tree, 58
Tetrahedral group, 63
U

Unary form, 2
Unimodular form, 30, 32

v-

Quadratic form, 1, 94, 108

of a graph, 85
of a manifold, 42
of a tree, 31, 60

Weighted tree, 58
Witt group, 95, 109
Zero form, 32