On Two Classical Theorems of Algebraic Topology 1971

Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

The American Mathematical Monthly

ISSN: 0002-9890 (Print) 1930-0972 (Online) Journal homepage: https://www.tandfonline.com/loi/uamm20

On Two Classical Theorems of Algebraic Topology

W. M. Boothby

To cite this article: W. M. Boothby (1971) On Two Classical Theorems of Algebraic Topology, The
American Mathematical Monthly, 78:3, 237-249, DOI: 10.1080/00029890.1971.11992736

To link to this article: https://doi.org/10.1080/00029890.1971.11992736

Published online: 11 Apr 2018.

Submit your article to this journal

Article views: 1

View related articles

Citing articles: 1 View citing articles

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=uamm20
ON TWO CLASSICAL THEOREMS OF ALGEBRAIC TOPOLOGY
W. M. BOOTHBY, Washington University, St. Louis, Missouri

Among the most beautiful and frequently quoted results of elementary


algebraic topology are the following two theorems, the first due to L. E. J.
Brouwer and the second to Brouwer and H. Poincare.
THEOREM I. (Brouwer Fixed Point Theorem) Each continuous map of the
solid unit ball into itself has a fixed point.
THEOREM 11. Any continuous field of vectors tangent to an even dimensional
sphere must be zero at some point.
For theorems which have stimulated so much further research, beginning
with the work of H. Hopf and continuing to the present, and whose content is
so clear and easy to state, they are surprisingly difficult to prove, even in the
simplest cases-the unit disk and the ordinary 2-sphere. Proofs are custom-
arily given in standard courses in algebraic topology, but only after a fairly
extensive theory is developed. In a brief, but very readable and elegant book
[3], J. Milnor gave relatively simple proofs, based in part on a very original
approach due to M. Hirsch [4]. Since this book goes into many generaliza-
tions of these theorems, it introduces and uses some techniques of differential
topology which we wish to avoid, for example Sard's theorem. In this paper
it is our purpose to make use of the fact that several advanced calculus texts
discuss both manifolds and exterior differential forms in the presentation of
multiple integrals and Stokes' theorem, and thus to rewrite the standard clas-
sical proofs using these techniques to avoid overt use of algebraic topology.
In fact the present treatment goes very little beyond material to be found in
the texts of either Fleming [6] or Devinatz [7] or in the more specialized book
of Flanders [5], and the proofs of these theorems become, then, a somewhat
delicate exercise in advanced calculus.
1. Preliminaries. Although we suppose the reader to be acquainted with
those portions of the advanced calculus texts mentioned above which deal
with manifolds and maps, differential forms, Stokes' theorem, and the Poin-
care lemma, we shall nevertheless briefly discuss some of these topics by way
of review. The manifolds-always supposed differentiable-which we shall
need are particularly simple ones: Euclidean n-space Rn represented as n-
tuples of real numbers-hence furnished with a fixed coordinate system cover-
ing the entire space; its submanifold sn-l (the n-1 dimensional sphere con-

Prof. Boothby received his Michigan Ph.D. under W. Kaplan. Before his present position at
Washington University, he was on the staff of Northwestern University, and he spent a post-
doctoral fellowship of the American Swiss Foundation at the E.T.H., Zurich. Since then he has
held an NSF fellowship at the I.A.S., Princeton, and a sabbatical year at the University of Geneva,
supported both by the NSF and the Amer. Swiss Found. His main research is in differential geom-
etry and Lie groups. Editor.
237
238 W. M. BOOTHBY [March

sisting of all (xl, . 'Xn) with.L: ~ = 1); and the "cylinder" R xsn-t, which
is an n-dimensional su bmanifold of Rn+l = R X Rn. These manifolds are orient-
able, which means that they possess a covering by coordinate neighborhoods
with the property that whenever two of these neighborhoods overlap, the
differentiable functions giving one set of coordinates in terms of the other
have positive J acobian. If a connected manifold is orientable, then there are
two disjoint classes of such coverings; a choice of one of these classes is said
to orient the manifold. Note that the space Rn is automatically oriented by
the fixed coordinate system associated with it. In the sequel manifold will
always mean connected, orientable, and differentiable manifold.
Throughout this paper we shall use the term differentiable to mean C""
(indefinitely differentiable). Thus a real valued function j(x1, · · · , Xn) on an
open subset UCRn is differentiable if it has continuous partial derivatives of
all orders at every point of U; manifolds are assumed to be covered by coordi-
nate neighborhoods for which change of coordinates on overlapping neighbor-
hoods is C""; and mappings from one manifold to another, unless otherwise
stated, will have expressions in local coordinates which are indefinitely dif-
ferentiable. In particular, any real-valued function on an open subset U of a
manifold M is differentiable if its restriction to each local coordinate neighbor-
hood on M is C"" when expressed in terms of the local coordinates, i.e., as a
function on an open subset of Rn(n=dim M).
These definitions should be fairly familiar, but less familiar is the defini-
tion of a differentiable function f on an arbitrary subset A of a manifold M.
In this case we say thatf is differentiable if it can be extended to a differen-
tiable function!* defined on an open subset Usuch that ACUCM: this re-
duces it to the familiar case above. But there is a difficulty; even a trivial case
-A consisting of a single point of Rn-shows that the values of the deriva-
tives may depend on the extension and thus have no meaning for the function
f itself. To avoid this problem we restrict the arbitrariness of A: we shall say
that a connected subset A of a manifold M is a domain if it is the closure of its
interior, i.e., A= V, where V= Int A is an open subset of M. If f is a differen-
tiable function on a domain A, then any two extensions which are differen-
tiable on an open set U containing A must have identical derivatives of all
orders on the open set VC U and thus on V=A. It follows that the derivatives
of all orders are well-defined, continuous functions on A. In particular, the
restrictions of local coordinates of M to a domain A are differentiable func-
tions on A, hence coordinates on A. Therefore the same covering which orients
M orients any domain A of M. If A CM and BCN are domains of manifolds
M and N, then differentiability of a map F:A~B is defined as for manifolds,
i.e., in terms of local coordinates.
Many of the domains we use are even more restricted. A regular domain D
of a manifold M is a domain which is compact, connected, and whose boun-
dary, denoted aD, is an n-1 dimensional submanifold which divides M into
two components. Examples are: (i) the unit interval I= {xERIO;;:ix;;:i1} in
1971] ON TWO CLASSICAL THEOREMS OF ALGEBRAIC TOPOLOGY 239

I
the manifold M= R, (ii) the unit ball Dn = { (xh · · ·, x,.) ER" LX:= 1} in
the manifold M= R", and (iii) M itself if M is compact-in this case oM is
empty. Regular domains are the domains of integration in the treatment of
Stokes' theorem found in [6]; it is important for this theorem that D is ori-
ented by the restrictions to D of the covering by coordinate neighborhoods
which orients M-moreover, this orientation of D induces an orientation on
oD. Thus the fixed orientation of R" by the single system of coordinates which
covers it also orients D" and its boundary, the unit sphere S"- 1 • Not all the
domains which we consider are regular; it will also be necessary for us to use
spaces which consist of the cartesian product IX D of the unit interval and
a regular domain D of a manifold M, for example, the solid cylinder IX D 2•
Spaces of this type are domains of R X M, a manifold; they differ from regular
domains only in that their boundary is not quite a submanifold. Their use
presents no new difficulties.
Since differentiable functions, coordinates, and maps have meaning for a
domain A of a manifold M, we may also consider differential forms and their
exterior derivatives on A just as we do on a manifold. The forms of degree q
are denoted by IV(A) and the totality of all forms of all degrees by 1\(A).
This set forms an associative algebra over the real numbers (products are
denoted with a wedge /\) and each 1\q(A) is a linear subspace. The 0-forms
are just the differentiable functions on A, and 1\q(A) for q>n=dim M con-
tains only the 0-form (0 of the algebra). The exterior derivative d is a linear
map whose iterate d 2 = 0 and which takes each q-form to a q+ 1 form. In par-
ticular, it assigns to each function (0-form) its differential and to each n-form
the only possible n+l form, namely 0. If aE/\r(A) and fJEA•(A), we have
ai\{J = ( -l)r+•[JI\a and d(ai\{J) = dai\{J + (-1)ral\d{:J.
Thus 1\(A) is not commutative, nor is d a homomorphism. However, these
properties do imply that the kernel of the linear map d, i.e., those forms "'
such that dw = 0, is a subalgebra and that the image of d, i.e., those forms 8
such that 8 = d7J for some 7J, is an ideal within it (forms in the kernel are called
closed and those in the image exact). Finally we mention the fact that any dif-
ferentiable map F:A~B of domains induces an algebra homomorphism
F*: 1\ (B)~ 1\ (A) in the opposite direction. It has the properties (i) F* o d
=d o F*, (ii) under composition of maps: (F o G)*= G* o F*, and (iii) the
identity map induces the identity isomorphism.
A form "' on A is completely determined by its expressions in the local
coordinates, which we now discuss briefly. In local coordinates or on an open
subset of R", the differentials dx1, · · · , dx,. of the coordinate functions form
a set of generators for the algebra, and wE 1\ q(U) may be written
with dx; 1\ dx1 = - dx; 1\ dx;.

Then the properties mentioned above imply that


240 W. M. BOOTHBY [March

They also imply that whenever a map F:A~B, A and B domains of manifolds
of dimensions m and n respectively, is expressed in local coordinates, say with
ya=ya(x1, · · ·, x,.), a=1, ···,m, then F*:l\q(B)~IV(A) is computed in
these coordinates by replacing each dy, by its corresponding expression as a
differential of a function of X1, • • • , x,. and regarding the coefficients as com-
posite functions of the x-coordinates through the y's. For example, if q = 1,
then
F*[ .'E b;(y)dy;] = .'E .'E b;(y(x)) ay·
- ' dx;.
i i OXJ

An important special case of the map F* induced by a differentiable map F is


the inclusion (or identity) map J of a subset into the set. Examples of inclusion
we need are a domain ACM into M, a submanifold N of M into N (especially,
for us, S"- 1 into R"), and of a coordinate neighborhood U of M into the mani-
fold M. In these cases we usually adopt the notation WA, wa, wu, etc., for the
image J*w of a form won the ambient manifold M. The form thus obtained
is called the restriction of the original form to the subset and it will be used
below to obtain forms on both D" and S"- 1 from forms on R", or even, in the
S"- 1 case, from forms on D", since S"- 1 is a submanifold of D" as well as of R".
In computation with forms, one makes repeated use of the fact that a form
on ACM is determined by its restrictions to each neighborhood of a covering
of coordinate neighborhoods of M. In this case, as already seen, we allow
ourselves to use w, rather than wu, wv, etc., for these restrictions.
2. One-parameter families of differential forms. Let D be a domain of a
manifold M, possibly all of M. We need to consider a family of q-forms w1 on
D parametrized by the variable t with 0 ~ t ~ 1. In local coordinates X1, • • • , x,.
of a coordinate neighborhood U of M the forms would be written

the coefficients b;1 ... ; 9 being differentiable functions of t, X1, • • • , x,.. It is


natural in this context to consider the domain IXD in the manifold RXM.
Each coordinate neighborhood U of M determines a coordinate neighborhood
RXU of RXM with local coordinates t, X1, • • · , x,., where X1, • • · , x,. are
the local coordinates in U. We use only coordinates of this kind on RXM and
its domain IXD. Let t be a fixed element of the unit interval I and define the
map J,: D~IX D by J,(p) = (t, p) and let w be a q-form on IX D. Then
J:
w, = w defines, for each 0 ~ t ~ 1, a q-form on D. The collection w1 for all
tEI is a one-parameter family of q-forms on D. Conversely, any one-param-
eter family we on D, given in local coordinates as described above, determines
an won IXD of which it is the image under the maps J,. The latter definition
1971] ON TWO CLASSICAL THEOREMS OF ALGEBRAIC TOPOLOGY 241

is more precise since it does not involve local coordinates and thus relieves us
of the necessity of checking independence of coordinates.
More generally, consider an arbitrary q-form won IX D. Using local coor-
dinates as described above, its expression on IX U will have the form

(*) w= .L: a, dt 1\ dx 11 1\ · · · 1\ dx 10 _ 1
1 ••• 10 _ 1 + ,L: b~o 1 ••• ~;0 dx~o 1 1\ · · · 1\ dxk.·

In general, the coefficients depend on t as well as x 1 , • • • , x,.. In fact t, the


first entry of the pair (t, p)ERXM, is a real-valued function defined on all
of the manifold RXM; hence its differential dt is a globally defined 1-form.
It follows that each q-form w on IX D splits into the sum of a form which
contains dt as factor and one which does not-the decomposition above does
not depend on local coordinates. Note in passing that because t is constant
in the definition of J 11 J,* (dt) = 0 and more generally
J,*(dt 1\ 8) = J't(dt) 1\ Jt(8) = 0.
We use these remarks to define a linear operator d: /\•(IXD)-+/\'~- 1 (D)
which plays a crucial role in the theory. Let w be a q-form on IX D given in
local coordinates by the expression (*) above, then
1

dw = .L: [ Jo a,1 ••• , 0 _ 1dt] dx,, 1\ · · · (\ dx;9 _ 1•

Since only the coordinates X1, • • • , x,. of the coordinate neighborhood U of


M are involved on the right, this gives a form on D in local coordinates.
(Note that it is zero if dt is not a factor of w.) The verification that this map-
ping is independent of local coordinates is straightforward and may be found
in the references cited, so we do not give it here. The most important property
of the operator d is embodied in the following lemma, in which wo, w1 denote
J.0*-w, J*- . 1
1 w respective y:

LEMMA1. Let D be a domain of a manifold M (possibly all of M), w be a


q-form on I X D, and dw the (q -1)-form on D defined above. Then we have
ddw+ddw=wl-wo.
The proof, reduced to local coordinates, is an application of Leibniz rule
for differentiating under the integral together with the elementary properties
of differential forms summarized in the previous section. It is proved in detail
in the references, e.g., p. 27 ff. of (5] or formula (7-20) on p. 280 of (6], and
we do not repeat it here. The present statement is somewhat more general in
that D may be either a manifold, including an open subset of a manifold-
which is also a manifold-or a more general domain of the type discussed in
the previous section in connection with differentiability. The formal details
are not changed by this increase in generality. Just as in the references (5],
(6 ], and (7 ], we give as first application Poincare's lemma, but in our case
for the closed unit ball D" as well as the open unit ball.
242 W. M. BOOTHBY [March

COROLLARY (Poincare's Lemma). ror q > 0 each closed q-form on the closed
ball D" CR" is exact. The same statement holds for its interior, the open ball B".
Proof. Let 8 be a q-form on D" with dO= 0. We define H: I X D"-+ D" by
H(t, x1, · · · , x,.) = (tx1, · · · , tx,.); then H(O, x1o • · · , x,.) = (0, · · · , 0) and
H(1, x1, · · · , x,.) = (x1, · · · , x,.). If we let~ =H*O and adopt our earlier no-
tation: (H*O),=Jio for the 1-parameter family of forms defined by 8 on D",
then we see that (H*8) 0 =0 and (H*O)t=O. Since dH*O=H*dO=O, we have
d~=O. By Lemma 1, dd~=O; it follows that 8 is exact as claimed.

REMARK. From the proof it is clear that the theorem holds for any star-
shaped open set or domain of R".
3. A differentiable version of the theorems. In this section we apply the
preceding lemma, Stokes' theorem, and the facts mentioned earlier about
the behavior of differential forms with respect to mappings in order to estab-
lish coo versions of Theorems I and I I. The following lemma is used in both
proofs:
LEMMA 2. Let {} be the (n -l)-form on Sn-l obtained by restricting to 5"- 1
the form

"
ft = :E (-1)i-lx;dx1 1\ · · · 1\ dx1-1 1\ dxi+1 1\ · · · 1\ dx,.
i-1

defined on all of R". Then D is closed but not exact.


Proof. The form {} is closed since dD has degree n which is greater than the
dimension of S"- 1 ' thus d{} = 0. If we suppose there exists an (n- 2)-form 8
on S"- 1 such that D=dO, then applying Stokes' theorem, we obtain

Jfs..-1 D= Jfsn-1 d8=fas..-18=0 '


because oS"- 1 is empty. However, S"- 1 =oD", and another application of
Stokes' theorem shows that

f {} =
Jsn-1
f aD"
n = JfD" an = n JfD" dx1 1\ . . . 1\ dx,.,
which is clearly not zero. Thus our assumption that{} is exact leads to a con-
tradiction.
The next lemma is the standard one used in the classical proof. In its
continuous version it is of interest in its own right; it says that a solid ball
cannot be retracted onto its boundary.
LEMMA 3. There is no differentiable map F: D"-+S"- 1 such that on the boun-
dary F is the identity, i.e., such that F(x) = x for every xES"- 1 •
1971] ON TWO CLASSICAL THEOREMS OF ALGEBRAIC TOPOLOGY 243

Proof. We suppose that such a map exists and obtain a contradiction. Let
0 be the form on S"- 1 defined in Lemma 2 and let J:S"- 1---+D" be the inclusion
map. Then Fo J:S"- 1---+S"- 1 is just the identity map of S"- 1• Thus (Fo J)*
=J* oF*, implies J*(F*O) =0. Now F*Q is a form on D" and since dF*O
= F*dQ = 0, it is closed. According to the Corollary to Lemma 1 it is exact,
i.e., we have F*O = d8 for some form 8 on D". This means that 0 = J* ( F*O)
= J*dO = dJ*O so that 0 is also exact. But we have proved in Lemma 1 that
this is not the case. It follows that no such map F exists.
From this lemma we now obtain by the usual method ( [3] or [8]) the dif-
ferentiable version of Theorem I.
THEOREM 1'. If G: D"---+D" is a differentiable map, then there is an xED"
suck that G(x) =.z:.

FIG. 1.

Proof. If there is no such point, then x aud G(x) determine a line; let F(x)
be its point of intersection with S"- 1 as we move along the line in the direc-
tion G(x) to x. For xES"- 1 this implies F(x) =x; so that F:D"---+S"- 1 and F
is the identity on S"- 1 • Moreover, treating x, G(x), etc., as vectors, F(x) is
given by [3]
F(x) = x + >.(x)u(x),
where u(x) is the unit vector (x-G(x))/llx-G(x)ll and }.(x) is the scalar de-
fined by }.(::r:) = -::r:· u+ [1-x·x+(x· u)2]t' 2 ; the dot indicates inner product.
It is easily checked from the geometry (Fig. 1) that the term in brackets is
strictly positive so that }.(x) is C"". Thus F: D"---+S"- 1 is a differentiable map
which is the identity on S"- 1 , contradicting Lemma 3. Hence no such G(x)
exists.
We conclude this section by using Lemma 2 to give the differentiable ver-
sion of Theorem I I.
THEOREM 11'. There is no smooth, nowhere vanishing field of vectors tangent
to an even dimensional sphere.
244 W. M. BOOTHBY [March

Proof. Again following the classical proof of Alexandroff-Hopf [8], we


first note that if the unit sphere sn- 1 has a nonvanishing, differentiable, tan-
gent vector field, then it has a differentiable field of unit tangent vectors:
simply divide each vector by its length. Given a field of unit tangent vectors
u(x) to sn-1, with u(x) tangent at x, we have x·x=1, U·x=O, and u·u=l.
The map H:IXS"- 1-+S"- 1 given by the formula (in vector notation)
H(t, x) = (sin 1rt)u - (cos 1rt)x
coincides with the identity map J: x-+x when t = 1 and with the antipodal map
A :x-+-x (which takes each xES"- 1 to the point at the opposite end of its
diameter) when t = 0. Let A: R"-+R" be the reflection in the origin, i.e.,
A :x-+-x. If ft is the (n-1)-form of Lemma 1, then A*ft = ( -1)" ft. Let J:S"- 1
-+R" be the inclusion map; we have defined n = J*ft. Obviously J o A =A o J
so that A *!J=A * o J*ft= ( -1)"J*ft and A *!J= ( -1)" n.
Now consider the form H*Q on IXS"- 1 ; we have dH*!J=H*d!J=O, (H*Q)t
=!J, and (H*!J)o =A *!J. We assume n is odd, which means that A •n = -n. Com-
bining these facts and applying Lemma 1, we have
dd(H*n) = !J- A*Q = 2!:2.
But this means that!J is exact: !J=d(!dH*Q), contrary to Lemma 2.
REMARK. When n is even, say n = 2m, there always exists a differentia-
ble, nowhere-vanishing vector field tangent to S"- 1 • In fact the vector u at
x= (x 11 x 2, • • · , X2m), xES"- 1, whose components are (x2, -x11 X4, -xa, · · ·,
X2m, -X2m-1) is a unit vector with u·x=O; hence it is tangent at X to sn- 1• We
shall discuss some further cases in Section 5.
4. The proof of Theorems I and 11. For the continuous case of Theorems
I and II, one needs an approximation theorem. This may be proved directly
(see Appendix) or by appeal to the Weierstrass Approximation Theorem, which
is proved in [7] but only for the case of an interval on the real line.
APPROXIMATION LEMMA. Let K be a compact subset of R" and fa continuous,
real-valued function defined on K. Then for any E > 0 there is a C"' function g (x)
I
on R" such that IJ(x) -g(x) <Efor all xEK.
In fact, according to the Weierstrass Approximation Theorem [9, p. 133]
there is a polynomial g(x) with the required property. For our purpose any C""
function g(x) will suffice, and we are only interested in the case K = D". An
easy, standard proof of the lemma in just the generality we require is given in
an appendix. Using this lemma (forK= D") we derive from Theorems I' and
II', the Theorems I and II of the introduction.
Proof of Theorem I. As in [3] suppose that F: D"-+D" is a continuous map
without fixed points. Using compactness we choose E>O so that 11 F(x) -xll >3E
on D". Let Go(x) be a differentiable E-approximation to F(x) on D"; the existence
of Go(x) may be seen by applying the preceding lemma to each of the eo-
1971] ON TWO CLASSICAL THEOREMS OF ALGEBRAIC TOPOLOGY 245

ordinate functions of F(x) = (ft(x), · · · , f,.(x)). If xED", then Go(x) may be


outside D", but we must have jjGo(x)li <1+e. Thus G(x) =1/(l+e)Go(x) is a
differentiable map of D" into D", and for xED" it must satisfy
1 1
IIF(x)- G(x)!l = IIF(x)- - -Go(x)ll
l+e
=--lleF(x)
1+e
+ F(x)- Go(x)!l
~ e!IF(x)!l + iiF(x) - Go(x)!l < 2e.

By Theorem I' we see that G(x) must have a fixed point. However we have the
following inequalities:
I!G(x) - xll = 11 (F(x) - x) - (F(x) - G(x))ll
~ IIF(x) - xll -IIF(x) - G(x)ll
~ 3e- 2e =e.
This is an obvious contradiction, hence no such F can exist.
Proof of Theorem II. We consider the unit sphere S"- 1 with n-1 even and
suppose that u(x) is a continuous field of tangent vectors which is never zero.
Thus, dividing each vector by its length, we may assume in fact that for each
xES"- 1 the tangent vector u(x) has unit length, i.e., we may assume x· u =0,
and 11 ull = 1 =l!xl!. For x = (x1, · · • , x,.) ER" we let r 2 = L x~ = llxl! 2 ; then we
may extend the component functions u1(x), · · · , u,.(x) of u(x), which are de-
fined only on S"-1, to continuous functions ut(x) on D", in fact, all of R" by
setting u*(O) =0 and u*(x) =ru(x/r) for r.eO. Let v(x) = (v1 (x), · · ·, v,.(x)) be
a differentiable E-approximation to u*(x) on D" with e>O small enough so that
w(x) = v(x) - (x · v(x)) x is never zero on S"- 1• The vector w(x) for x ES"- 1 is
just the orthogonal projection of v(x) onto the tangent hyperplane to S"- 1 at x;
it vanishes only if v(x) is parallel to x, i.e., orthogonal to u(x). Clearly it is
possible to choose E so small that this does not occur. Since w(x) is differentiable,
tangent to S"- 1 , and not zero for any xES"- 1, we have a contradiction to
Theorem 11'. Thus no vector field u(x) of the type assumed is possible, and
Theorem 11 is established.
5. Related topics. In this section we shall try to give the general reader some
insight into the reasons for the importance of Theorems I and 11 by mentioning
a few applications, related theorems, and generalizations. To begin with, we
observe that any space X homeomorphic to D" also has the fixed point property
with respect to continuous maps. Indeed, if H:X~D" is a homeomorphism and
F:x~x is continuous, then Ho F o H- 1 : D"~D· has a fixed point x 0 ; it
follows that y 0 =H- 1 (x 0) is fixed under F. In particular, a compact convex sub-
set of R" will have the fixed point property; using the Brouwer theorem (The-
orem I), it is possible--by a very short and straightforward line of argument-to
extend this result to each continuous map of a closed, convex set of a Banach
space into a compact subset of itself (Schauder-Mazur theorem). This general-
246 W. M. BOOTH BY [March

ized fixed point theorem has been used to prove existence theorems for boundary
value problems of differential equations. This will not surprise anyone who
has seen proofs of the Inverse Function Theorem and of the Existence Theorem
for systems of ordinary differential equations which use the Contracting Map-
ping Theorem as in [9] and [10 ]. (This latter theorem is a very simple fixed point
theorem which asserts that any continuous map F:X-+X of a complete metric
space which contracts distances between pairs of points must have a unique fixed
point; although it is not as strong as Theorem I, it is much easier to prove.)
A very nice account of the Schauder theorem is found in Bers lecture notes [11 ] ,
where he gives as an example the following application (C 0 denotes continuous
and C", k > 0, k times continuously differentiable functions):
(5.1) Let L>O be a real number andj(x1, X2, xa) a bounded continuous func-
tion on R 3 ; then there exists a junction x(t) of class C2 which is a solution of the
boundary value problem:
x'(t) = J(t, x, x') and x(O) = 0 = x(L).
To see how this follows from a fixed point theorem one considers the Banach
space C0 of continuous functions y(t) such that y(O) =0 =y(L) and its subspaces
C1 and C2 of functions of class C 1 and C2 respectively, with suitable norms in all
cases. Define H:C1-+C0 by H[x(t)]=J(t, y(t), y'(t)) and G:Co-+C2 by the for-
mula

G[y(t)] = f otf•o y(u)du dv - -t


L
fLf• y(u)du dv.
o o
Then F= Go H maps C1 into C2C C1 and it is easy to check, using the Funda-
mental Theorem of Calculus, that x(t) is a fixed point of F if and only if x(t)
is the desired solution. Other, similar applications are given in the reference
cited [11 ].
In quite a different direction Theorem I can serve as a cornerstone for dimen-
sion theory as may be seen from Brouwer's original work or from the book of
that title by Hurewicz and Wallman [12]. As is well known, Brouwer was the
first to prove that the dimension of R" had topological meaning by showing that
there is no homeomorphism of an open subset of R" into R"' if m<n. In their
book Hurewicz and Wallman derive both the fact that R" is n-dimensional (in
a topologically invariant sense) and the Lebesgue tiling theorem very easily
and directly from the following statement:
(5.2) Consider I,.= {xER!Ix;l ~1}, a cube, and let C;, c:
be the opposite
+
faces given by x; = 1, -1 respectively. Let B, be a closed subset of I,. separating
c:
C; and for each i = 1, · · · , n. Then there is a point common to all of the B;.
We shall reproduce the proof of [12 ]. Let u,, Ul be the components of I,.- B,
and letf,(x) be the continuous functions on I,. given by f;(x) = ±d(x, B;), the
distance of x from B, with sign depending on whether x is in u, or U/. Then by
1971] ON TWO CLASSICAL THEOREMS OF ALGEBRAIC TOPOLOGY 247

Theorem I the map F:In~I,. defined for x= (x1, · · · , x,.) by


F(x1, · · · , x,.) = (x1 + f;(x), · · · , x,. + f,.(x))
has a fixed point, which must, then, lie in the intersection of the B;.
Finally, we mention that Lemma 3, which was the key to proving Theorem I,
may be interpreted as stating that any continuous map F: Dn~D" which
leaves the boundary pointwise fixed must be onto. It is possible to relax the
condition on the boundary, that is to require that the map Fcarry the boundary
onto itself in an essential way (i.e., not be deformable to a constant map) and
still get the same conclusion. This leads directly to the concept of the Brouwer
degree of a map, an important generalization, which is beautifully developed in
[3 ], using an approach which does not require homology-as does the classical
treatment of [8].
Theorem 11 also has been a well-spring of research in algebraic topology,
beginning with the work of Brouwer and Poincare, then H. Hopf and his stu-
dents, and continuing to the present day. For example, having answered the
question for spheres one might ask exactly which closed differentiable manifolds
admit continuous nonvanishing fields of tangent vectors. This question is an-
swered as a special case of the Poincare-Hopf Theorem, whose full statement
requires the definition of the index of a vector field at an isolated zero of the
field-it is an integer which measures the nature of the singularity. This goes
beyond our scope--we refer again to [3 ]-but we can state the following conse-
quence of the theorem:
(5.3) A manifold M has a continuous, nonvanishing field of tangent vectors if
and only if its Euler characteristic vanishes.
We will not define the Euler characteristic except to say that in the case of
Sn- 1 it has the value 2 for n-1 even and zero for n-1 odd-in fact it is zero for
any odd dimensional manifold. In the case of surfaces it is zero only for the torus.
Therefore the torus is the only closed surface which can have a nonvanishing
field of tangent vectors. (Visualizing a closed, orientable surface as a 2-sphere
with handles attached, it can be shown that the Euler characteristic is 2(1-g),
g being the number of handles, see [13, Ch. VI].)
Returning to the case of spheres, a second question which arises naturally
is as follows. Given an integer k, 1 ~k ~n-1, do there exist k vector fields,
u1, · · · , u~: tangent to sn-I which not only do not vanish but are even linearly
independent at each point? This was a subject of investigation for over thirty
years with definitive results being given by J. F. Adams in 1962 [1]. He proved
that the maximum number of linearly independent vector fields on S"- 1 was the
number which previously had been shown to exist by Hurwitz-Radon-Eckmann;
that is, Adams showed that their result was the best possible. The number
p(n) -1 of independent vector fields on S"- 1 may be defined as follows: write
n as an odd integer times a power of 2, i.e., n = (2a+1)2b, and set b =c+4d with
0 ~ c ~ 3. Then the integers a, b, c, d depend only on n, and p(n) is given by the
248 W. M. BOOTHBY [March

formula p(n) =2•+8d. The proof that this many vector fields actually exist
depends on purely algebraic results-basically from linear algebra. We shall
give a hint of how this very geometric theorem is related to algebra by showing
that in the case of sa the maximum possible number of independent vector fields
can be defined, namely 3. This occurs only in two other cases: S 1 and SI, as can
be checked rather easily by finding all solutions of the equation p(n) -1 =n-1.
For the case of 5 1, let X= (xi, x2, Xa, X4) be a point of the unit sphere 5 3 in R'.
If we consider it as a vector, then we are required to find three unit vectors
u;(x), i = 1, 2, 3, which are orthogonal to x, hence tangent to S 3 at x, and which
are linearly independent for each x. The following three vectors which are mu-
tually orthogonal clearly satisfy this requirement:

The interesting fact is that if we identify R 4 with the algebra of quaternions by


the correspondence (x1, X2, xa, X4)~x1+x2i+x;J+x4k, then these four mutually
orthogonal unit vectors are just x, u1 = ix, u2 = jx, and u 3 = kx. Using an algebra
of dimension 8, the Cayley algebra, precisely the same type of construction
gives the 7 vector fields on 5 7• Conversely, one has used topological theorems
of the type we have been considering to prove the nonexistence of certain types
of algebras over the real numbers.
Appendix.
PROPOSITION. For any ~>0 and xER" there is a C"' function f/>~(y) which is
non-negative, is zero outside the ~-sphere IIYII
< ~ around the origin and satisfies
]Bn 4J~(y)dy1 ' ' ' dyn = 1.
Proof. We shall assume that the standard example of C"' function given in
many calculus texts, e.g., [6, p. 51], is indeed C"', and begin from there. The func-
tion referred to is
0, t<O
1/t(t) = { e-llt\ t > 0.
Then the composite function f/>~(y) = kt/;(~ 2 - L; ~). where k is a positive con-
stant so chosen that

f R" rf>&(y)tlyl ... dy,. = 1,

has all the properties claimed.


Proof of the Approximation Lemma forK= D". Letf(x) be a continuous func-
tion on K and extend it to a continuous function on all of Rn by defining it along
each radial line outside the boundary sphere to have the same value that it
takes on the unit sphere; that is, when llxll > 1, wedefinef(x) to bef(x/llxii). We
remark that this is the only place in the proof at which we use the assumption
1971] ON TWO CLASSICAL THEOREMS OF ALGEBRAIC TOPOLOGY 249

that K is the unit ball; were we to concede the fact that a continuous function/
on an arbitrary compact set KCR" can be extended to a continuous function on
all of R", then the remainder of the proof would establish the lemma in that case
also.
Now, given E>O, we choose ~>0 so xEK= D" (a compact set) and <~ IIYII
implies IJ(y+x) -J(x) I <E. Define g(x) by

g(x) = J R"
J(y + x)tf>&(y)dy1 · · · dy,..
For each fixed x this integral is finite since cf>&(y) = 0 if IIYII
~ ~. i.e., we could
take the region of integration to be the closed ~-ball around x. Moreover,

g(x) - j(x) = f R" [J(y + x) - J(x) ]tf>&(y)dy1 · · · dy,.

smce

J R" tf>&(y)j(x)dy1 · · · dy,. = J(x) J R" t/>&(y)dy1 · · · dy,. = j(x).

I I
Thus g(x) -J(x) <EJ cf>&(y)dy1 · · · dy,. =E.
On the other hand, if we change the variable of integration, letting z=y+x,
we have g(x) = JR" f(z )c/>&(z- x)dzt · · · dz,.. From this expression, the rule for
differentiation under the integral sign, and the fact that cp, is C"", it follows that
g is also C"".
References
1. J. F. Adams, Vector fields on spheres, Bull. Amer. Math. Soc., 68 (1962) 39-41.
2. R. H. Bing, The elusive fixed point property, this MONTHLY, 76 (1969) 119-131.
3. J. Milnor, Topology from the Differentiable Viewpoint, University of Virginia Press,
Charlottesville, 1965.
4. M. Hirsch, A proof of the nonretractability of a cell onto its boundary, Proc. Amer. Math.
Soc., 16 (1963) 364-365.
5. H. Flanders, Differential Forms, Academic Press, New York, 1963.
6. W. H. Fleming, Functions of Several Variables, Addison-Wesley, Reading, Mass., 1965.
7. A. Devinatz, Advanced Calculus, Holt, Rinehart & Winston, New York, 1968.
8. Alexandroff-Hopf, Topologie, Springer Verlag, Berlin, 1935.
9. J. Dieudonne, Foundations of Modern Analysis, Academic Press, New York, 1960.
10. Loomis and Sternberg, Advanced Calculus, Addison-Wesley, Reading, Mass., 1968.
11. L. Bers, Topology (Lecture Notes), New York University, 1957.
12. Witold Hurewicz and Henry Wallman, Dimension Theory, Princeton University Press,
1948.
13. David Hilbert and S. Cohn-Vossen, Geometry and the Imagination, Chelsea, New York,
1952.

You might also like