Professional Documents
Culture Documents
Topología para Comenzar
Topología para Comenzar
Beginner’s Course
in Topology
Springer-Verlag
Berlin Heidelberg New York Tokyo
NUNC C O G NOSCO EX PARTE
https://archive.org/details/beginnerscourseiOOOOrokh
Universitext
D.B. Fuks V.A. Rokhlin
Beginner’s Course
in Topology
Geometric Chapters
Springer-Verlag
Berlin Heidelberg New York Tokyo 1984
Dmitrij Borisovich Fuks
Laboratory of Bioorganic Chemistry
Moscow State University, Moscow, USSR
Andrei lacob
Department of Theoretical Mathematics. The Weizmann Institute of
Science. Rehovot 76100, Israel
The Authors
Contents
Chapter 1
TOPOLOGICAL SPACES ................................................................................................................ 5
§ 1. FUNDAMENTAL CONCEPTS.......................................
1. Topologies............................................. 5
2. M e t r i c s ............................................... 8
3. S u b s p a c e s ............................................. 10
4. Continuous M a p s ....................................... 11
5. Separation Axioms ..................................... 15
6. Countability Axioms ................................ 19
7. Compactness........................................... 21
§2. CONSTRUCTIONS............................................. 26
1. Sums................................................... 26
2. Products............................................... 27
3. Q u o t i e n t s ............................................. 31
4. G l u e i n g ............................................... 34
5. Projective Spaces .................................... 38
6. More Special Constructions............................ 42
7. Spaces of Continuous M a p s ............................ 46
8. The Case of Pointed Spaces............................. 49
9. E x e r c i s e s ............................................. 55
§3. HOMOTOPIES................................................. 56
Chapter 2
CELLULAR S P A C E S ........................................................................................................................... 81
1. Fundamental Concepts.................................... 81
2. Glueing Cellular Spaces from Balls.................... 86
VIII
1. Euclidean Simplices................................... 99
2. Simplicial Spaces and Simplicial Maps................. 101
3. Simplicial Schemes .................................... 105
4. Polyhedra.............................................. 106
5. Simplicial Constructions ............................. 107
6. Stars. Links. Regular Neighborhoods................... 114
7. Simplicial Approximation of Continuous Maps........... 118
8. Exercises.............................................. 119
Chapter 3
SMOOTH MANIFOLDS ............................................... 133
Chapter 4
BUNDLES.......................................................... 265
Chapter 5
H0M0T0PY GROUPS................................................. 373
BIBLIOGRAPHY ................................................... 5 06
INDEX............................................................ 508
GLOSSARY OF SYMBOLS............................................. 517
Set-Theoretical Terms and Notations Used in this
Book, but not Generally Adopted
Maps
f = rel cp , cp = abs f .
Quotients
the saturation of A.
If p and q are partitions of X and Y, respectively,
then each map f: X + Y which transforms elements of p into elements
of q induces a map X/p Y/q, which takes each element A of p
into the element of q which contains f(A). This map is denoted by
fact f . In particular, factf is defined when q is the partition
of Y into its points, and f is constant on the elements of p.
Thus, for each map f : XY constant on the elements of the partition
p of X we have the corresponding map fact f : X/p Y.
Given a map f: X -+ Y, the partition of X into the nonempty
preimages of the points of Y is denoted by zer(f). The corresponding
map fact f : X/zer(f) Y is injective and is called the injective
factor (or injective quotient) of the map f.
Sums
Products
<X 1 " x i- V
This map is called the i-th projection and is denoted by pr± .
Given maps f-^ X± Y ^ . . . , ^ : Xn + Yn , the rule
defines a map of the product X^ * ... x xn into the product
Y^ x ... x Yn' caH ed the product of the maps and denoted
1. Topologies
sets.
3. A neighborhood of a point of a topological space is any
open set containing the given point. A neighborhood of a subset of a
topological space is any open set containing the given subset.
Derived Concepts
sets such that any open set can be represented as a union of sets from
this collection. Equivalently, a collection r of open sets is a base
if for any open set U and any point x £ U there is V G F such that
x € V c U.
A base completely determines the topology: the open sets are
exactly those which can be expressed as union of elements of the base.
Covers
2. Metrics
ball and the unit sphere of IRn , i.e., the ball and sphere with center
(0 P. ..,0) and radius 1, are simply called the n-dimensional ball Dn
and the (n-1)-dimensional sphere Sn ^ . In particular, is just a
point, S - a pair of points, and S_1 = 0. Moreover, we set Dn = 0
for n < -1 and Sn = 0 for n £ -2.
3. Subspaces
and a = 1, the cube is called the unit cube ln . The unit segment
1 .
I is also denoted by I.
Fundamental Covers
Clearly, the claim is true for both open and finite closed
covers. Now suppose that Г is a locally finite closed cover of a
space X. Consider a cover Д of X consisting of open sets which
intersect only a finite number of elements of Г. Since Д is
fundamental, it is enough to check that, given any set U £ Л, the
cover of U with elementsU П В, В £ Г, is fundamental. But this
results from the fact that the latter cover is finite and closed.
4. Continuous Maps
is continuous.
Continuity at a Point
Retractions
Numerical Functions
5. Separation Axioms
. Any point and any closed set not containing this point
have disjoint neighborhoods. An equivalent formulation:
every neighborhood of an arbitrary point contains the closure of a
neighborhood of this point.
Urysohn Functions
property (1), we see that the last intersection is just nr .>r Clcp(r’);
Extension Theorems
6. Countability Axioms
uI’
I = UII \ U?I CIV.-L and Vn’ = Vn \ U1?
I Cl U.1 . The sets U = U?1 U*n and
21
oo
V = U1 Vn are ° P en and clearly disjoint. Since Cl LL fl B = 0 and
Cl n A = 0, we have U => A and V => B.
7. Compactness
time compact (see 2) and discrete. But this contradicts the fact that
Y is infinite.
Local Compactness
Information: Paracompactness
§2. CONSTRUCTIONS
1. Sums
or T4 , then their sum satisfies the same axiom. The same hold for
the first axiom of countability, and also for the properties of local
compactness and metrizability [if X are metric spaces, one can define
a metric on J j_X^ by the formulas: dist (in^ (x) ,in^ , (x 1) ) = 1 when
v i v1 or when v = v’ and dist(x,x') ^ 1 ; dist (inv (x) ,inv (x1 )) =
= dist (x,x1) when dist(x,x') < 1]. If each X^ has a countable base
and M is countable, then J |_X has a countable base too. Similarly,
when M is countable and each X^ is separable, J [x is separable
too. Finally, J [x is compact whenever all X are compact and M
is finite.
2. Products
Properties of Products
8. If
S 1f...,S are dense sets in the spaces X „ ,...,X ,
I n c 1 n
then S1 x... x sn is obviously densein X^ x ... x x . Consequently,
a product of separable spaces is separable.
If now
r„,...,r are bases of X„ ,...,X at the points
i n I n
x.,...,x ,then the sets U. x ... x u with U. £ T U £ F ,
i n 1 n l i n n
form a base of X^ x ... x x^ at the point (x^ ,...,x^) . Consequently,
a product of first countable spaces is first countable.
An Application:
A Method for Constructing Continuous Maps
is continuous.
Information
3. Quotients
is a homeomorphism.
Closed Partitions
Open Partitions
4. Glueing
xv X JE2U X/p
Unions
if k1 = k,
^kk' id Xk'
[ab ((¡)^, o ... o rf)k )] - 1 , if k 1 > k,
f,
X,
xk
vk
k +1
k+1 k +1
are commutative. Then the rule f(imm, (x)) = imm, (f, (x)) defines a
K K K
continuous map f: lim(Xk ,<t>k ) -+■ limiX^,^) (see 3.2); f is called
Attaching
in2 (X^ ^cp(C)), and the sets in^ ((p ^ (x) ) U in2 (x) with x £ cp(C). The
f2 (cp(x) ) , if x €C ,
5. Projective Spaces
(E x ... x (E,I H X ... X IH, and (Ea x ... x (Ca, are denoted by CDU r
H , and (Ea . Since every complex number is a pair of real numbers,
every quternion a quadruplet of real numbers, and every Caylely number
an octuplet of real numbers, one can naturally identify (En , IHn , and
(Ea with 1R , 3R^n ,and !R^n ,respectively.In particular, the
former are endowed with natural topologies and metrics. The vector
operations in (E , IHn , and (Can (addition of vectors and left or
right multiplication by scalars) are continuous in these topologies.
One may also identify the points of 3RPn with the lines of
IRn 1 which pass through the point 0 = (0,...,0) [the line passing
through the points x and -x corresponds to the pair of points
x ,-x E Sn ] . The set of all these lines, equipped with the angular
metric (i.e., the distance between two lines is defined as the angle
between them which is less than tt/2 ) , is a metric space, and the above
natural map of lRPn onto this space is clearly a homeomorphism. This
provides a third description of the real projective space. The fourth,
a coordinate description, can be obtained if one remarks that every line
passing through 0 is uniquely determined by any of its nonzero points,
and that the coordinates of any two nonzero points of such a line are
proportional. This enables us to interpret the points of lRPn as
classes of proportional nonzero real sequences (x^,...,x^+^); the
point determined by the sequence (x^,...,x^+^) is denoted by
(x^:...:xn+^), and the numbers x ^ ,...,xn+^ are called its homogeneous
coordinates. The description of the topology of lRPn in terms of the
homogeneous coordinates is plain.
oo k °° k
D = lim D , S = lim S ,
and
HRp” = l i m l P k , (EP°° = lim (EPk , 3HP°° = l i m K P k .
jc A 1 r . . . fA
(pr(x,x n ,t)) = ((1 -t)jc A A .
(x),pr(x n ,t)),
i n I n-1
where the last arrow denotes the canonical homeomorphism, and the r-th
arrow, with r £ k, denotes the composite canonical homeomorphism
m m m m m +m +1
s * ... * s s * ... *s n * s n n - ... ^
“ '1 iun m - —i m -
D x ... x D -*■ con S x ... x con S n
m1 - 1 0 m _1 n
S *D * . . . * S n * D -*■
m i_1 n i~ 1 n m _1
-* COn(S * D * ... * S * D * s n ) ->-
ml - ^ 0 mn - 1 - ^ n m -1
-*■ con S X con D X ... X con S X con D x con S n -*
m. m m
D x l x . . . x D n x I x D n ->
m1 1 m m m,+...+m +n-1
D X D x ... x D n x D x D n - >- D n
n»1 m m1 m
Th er e fo re , the jo in S * . . . * S , the product D x ... x d n,
X, ^ 2 — cyl £ x2
equals f.
46
C (pr ,id Y) [C (X/p ,A; Y,B) ] = C (X ,pr~ 1 (A) ;Y ,B) fl C(pr,idY) [C(X/p;Y)].
pr2 (Q) in X and Y are also compact (see 1.7.8). Moreover, they
are Hausdorff spaces together with X and Y, and hence normal
(see 1.7.5). Consequently, there exist open subsets of pr^ (Q)
and V'q of pr0
2 (0) such that
A Surprising Application
f x id . id x q
X x Y — ---- — >X' x Y ------ X' x Y'
PIt (Xq ), the product of the pointed spaces (X^ ,x^ ),..., (X^x^) has
the natural base point (x^,...,x ), and the space of continuous maps
from X into a pointed topological space (Y,Yq ) contains the constant
map const: X + Y, x yQ , and hence has the natural base point const.
Other constructions such as the sum, suspension, and join need more
serious modifications.
We shall describe these modified constructions below, and also
introduce a new one - the tensor product of pointed spaces. In every
case, pointed spaces produce pointed spaces, and base point-preserving
maps again produce base point-preserving maps. We remark that the maps
factf , C(f,g), and f^ * ... x fR preserve base points whenever the
initial maps have this property.
We use the symbol bp as a general notation for the base
points.
xv , if v ’ / v,
prv (immv ,(x)
x, if v 1 = V,
continuous, with f (xQ ) = y 0 ' then the maP fact con f : con(X,xQ)
con(Y,yQ) is well defined and continuous, and we denote it simply by
con f .
Equivalently, one may describe con(X,xQ) as the quotient
space of the cylinder X x I by (X x 0) U (x^ x I).
7. For any two pointed spaces (X^x^) and (X2 ,x2), the
bouquet (su(X^,x^) ,bp) V (su(X2 , ^ 2 ) 'bp) is canonically homeomorphic
to su((X1 /x 1) V (X2 /X2 )/bp).
m m
8. con(S ,ort^) , su(S ,ort^) , and (Sm ,ort1) * (Sn ,ort )
First base
9. Since
con(Sm ,ort^) = (Sm ,ort^) ® (1,0) (see 4), and
A
su(Sm ,ort1) = (S (Ort^ 0 (S jort.j) (see 5), the homeomorphisms
con(Sm ,ort1) -> Dm+1 and su(Sm ,ort.|) -> Sm+1 , defined in 8 , lead for
n ^ 1 to the canonical homeomorphisms
54
and
Indeed,
under the mapping $ h- <j)U is just C((X,x0) ® (Y,y0 ),pr(A x B) ,bp;Z ,C ,zQ),
which shows that <J> h - $ is continuous. Assume now that X and Y
are Hausdorff and compact, and consider the mapping
C (pr,id_Z) ^ c ^x x Y ,Z).
9. Exercises
9 ? ? 2 ^
x + ... + x, ^ x, + . .. + x in the standard coordinates of IR ,
1 k k +1 n
1c xi k
is homeomorphic to D x S
3.Let M, Xy , and $ , be as in 4.2, and let X denote
the union ofthe spaces X defined by the homeomorphisms ‘f’yy •• Show
that the maps immy : Xy ^ x are topological embeddings whenever M
has only two elements, but when M has three elements this is not
necessarily so.
4. Show that for n i 1 the spaces C (1,0,1;Sn ,ort^,ort2 )
and C(I,0 ,1 ;Sn ,ort1 ,ort1) are homeomorphic.
00
5. Let T be the set of all real sequences with
the topology defined by the prebase consisting of all sets of the form
{{x.}“ | a < x < b}. Further, let S be the quotient space of T ^ 0
1 1 s
(where 0 = {x^=0 }” ) by its partition into rays, i.e., into the sets
§3. HOMOTOPIES
1. General Definitions
Stationary Homotopies
Contractible Spaces
Deformation Retractions
Relative Homotopies
2. Paths
_1
called the inverse of s and denoted by s . Given two paths s1 and
with s 1 (1 ) = s 2 (0 ), the formula
s^ (2 t ) , for t £ 1 /2 ,
s2 (2 t- 1 ), for t ^ 1 /2 ,
Connectedness
k-Connectedness
r
6. The following properties of a continuous map f: S
with r ^ 0 are equivalent:
(i)
!
l ^ (ii) (iii)
V ^>"7
(iv)
and
64
r+ 1
and some continuous map g: D -* X (see 2.3.4 and 1.7.9), and it is
clear that g = f.
sr
The implication (iv) => (i) is trivial.
4. Local Properties
5. Borsuk Pairs
X x I -P-* ( X x O ) U (A x I) -iL, y,
where G is defined by
f(x) , if t = 0,
G(x,t)
F (x ,t) , if x G A,
p: A x I -> [ A x 0] U [ (Afl B) x I]
defines a retraction X x i (x x 0) U (B x i) by
'p(x, t) , if x e A,
(X,t)
(x,t), if x e b.
g: [(X x 0) U (A x i) u (X x 1 )] x i x
by
f x, if t1 = 0 ,
g((x,t1 ),t2) = |f(x,(1-t2 )t1), if x E A,
I f(P(X),1-t2), if t 1 = 1,
ab H :X x i (x x0) U (A x i) is a retraction.
6 . CNRS-spaces
This is a resultof 2.
CM tt
V >A — ► X.
Then dist ((J)(x) ,x) < c for x £ V, and <Mx) = x for x £ A. Hence
the inclusion V X is A-homotopic to <j), and since (j)(V) = A, we
can apply Theorem 5.10.
Information
7. Homotopy Properties
of Topological Constructions
Products
Quotients
X X I Pr X idI (X/p) x
F I G
y ------ EE-------- , Y/,
Attachings
imm. lmm. (1 )
X, u X,
2 op 1 X2 V X,
PROOF. Let $: C x i X2 be a homotopy from cp to cp1 ,
and let a: X^ x 1 + (x^ x 0) U (C x 1 ) be a retraction. Define the
maps h: (X^ x 0) U (C x I) + X 2 U X^ and h !: (X ^ x 0) U (C x I) +
* X2 V X1 b*
imm1 (x) , if t = 0 ,
h(x,t) =
imm2 ° <j>(x,t), if x £ C,
75
and
imm (x) , if t = 0,
h ’ (x,t) =
iiran2 o <!>(x,1-t) , if x £ C.
It isclear that all thesemaps are continuous and that the diagram (1)
is commutative. Moreover, it is readily seen that the map
( X 2 U<p V X1 ^ X 2 Utp' X1 9 ive n by
\ A q
\ b \ b3 b.
w \ \\
\\ \ \
\ P2 \ Q b
\De
Qa 'P'
At Aj A^
Fig. 3 Fig. 4
fa c t (id XJ—
I If)
1
: Y U X
tp Y ' U_
f °ip X
is also a homotopy equivalence.
imm (x), if t = 0,
g(x,t)
imm2 o F(ip(x),t), if x G C
and
imm1 (x) , if t1 = 0 ,
g '((x,t1),t2)
imm2 o G ((ip (x) ,t 1 ) ,t2) , if x G C.
(imm1 (x) ,t) = g ' (p (x,t) ,1) , (imm2 (y ') ,t) = imm2 (F '(y ',t ))
and
(imm1 (x),t) = g '(p(x,1),t), (imm2 (y'),t) = imm2 (fof'(y)),
t i pr (x ^ ,x 2 ,2 - 3 x ) , if 1/3 £ x £ 2/3,
8. Exercises
00
1. Show that the sphere S is contractible.
1. Fundamental Concepts
Subspaces
Compact Subsets
finite subspace which contains it; the union of these subspace is again
a finite subspace and contains the above neighborhood.
Cellular Maps
and id Dn , respectively.
or KPr mpn
D2r (cpH
Embedding Theorems
V =
ryi 1 Yn - 1 Yn
(2 p-1 )j°cha + (2 - 2 p) 0 , . ..,0 , ^ ------------ +2 k ,1
PJ
if p 5 1/2 ,
q+n +1
This yields a continuous map J :X -+ IR
and since the plane y^ = 0 ,. y = 0 ,y „ = 1 contains no lines
Jq q+n +1
parallel to IR^, J is injective. Hence, J is an embedding (see
1.1.7.10).
V7,k. =
= {x
{x e
e U
\,_ | dist (x,ip. (x) ) < Dist(x, IRq )} ,
91
{[(Uk = 1 C l V k ) n Fr V ] U (U^ = 1 Ak )} D w,
( 1)
[(Uk = 1 F r V k ) fl C I V fl W] v Uk= 1 <j>k (Sn_1) ,
Y = (Uk=1Cl Vk ) n Cl V n W.
if x e uk = i 4>k (sn 1 ) ,
(x) , if X E ci vk n (W \ V) ,
x h» (x) , if X E w " uk = i V
T (x) , if X E Y,
Connectedness. Components
5. Cellular Constructions
Cellular Product
dim e dim e
with the product cha x cha : D x d X x x^ of
e e 1 2
dim e
arbitrary characteristic maps cha : D -> x and
dime 2 e1 1
cha : D X9 . When the decompositions p and p0 are rigged,
2 1 Z
p ^ x p2 takes on a canonical rigging.
Attaching
Limits
6. Exercises
2. Show that the sphere S°° and the ball D°° are homeo-
morphic to cellular spaces.
1. Euclidean Simplices
centric
— . coordinates
—.— -- -- —. . of x and is denoted by bar a (x).
Obviously, a face Esi B of the simplex Esi A is defined
in the barycentric coordinates of EsiA by the equations bar^ (x) = 0
for a E A \ B. Moreover, if x E Esi B , then the coordinates bar a (x)
computed in EsiA and EsiB coincide for all a E B.
The point of EsiA having all barycentric coordinates equal,
i.e., equal to 1/(r+1), is the center of the simplex EsiA .
Topological Simplices
x E X is denoted by si x .
According to 1.2.4.3, a triangulation transforms the given set
into a topological space, and 1 .2 .4.1 shows that the supports of the
simplices of the triangulation yield a fundamental cover of this space.
Since the intersection of two simplices in the triangulation is closed
in each of them, the simplices in the triangulation keep the same
topology when considered as subspaces of this topological space (see
1 .2.4.2) .
Let a be a vertex of the simplicial space X. Then the
a-th barycentric coordinate bar (x) is well defined for any point x
a
belonging to any simplex which has a as one of its vertices (see 1 . 2
and 1.6), and we obtain a continuous function bar : X -> IR if we set
a
bar (x) = 0 for those points x G X contained in simplices which do
a
not have a as a vertex.
bar_ is called the a-th barycentric function.
a
Given two arbitrary distinct points x,y E X, there obviously is a
vertex a such that bar_ (x) f bar_ (y). Consequently, every simplicial
a a
space is Hausdorff.
Subspaces
Simplicial Maps
3. Simplicial Schemes
4. Polyhedra
positive integers j-,'••• »j2n +2 such that ^ < ... < j2n +2 * 2 q +1 and
x-j = X' for j f j, j2nt2. Since
¡? :i <- n — i|:] i 1 «}
5. Simplicial Constructions
Barycentric Subdivision
Simplicial Products
nondecreasing sequence Y-j (x^ ,x2) , .. .,y^ +r (x^ ,x2) . Further, let
^ jk + r (k+1)
k =0 l =i -k[Yk+l + 1 (X1 ,X2 ) " Yk + 1 (x1 ,x2 )] (ak'bl ) = (xi'x 2 ]
Jk
Limits
dim X + dim X 9 + 1
where and T2 are opposite faces of the simplex T t
and <J> and c|>2 are simplicial homeomorphisms (the equality
dim X^ + dim X 2 + 1
T = T * T2 stands for the simplicial homeomorphism
established in 1.1). Now one may canonically triangulate the cellular
join X 1 *c X 2 of two arbitrary simplicial spaces X1 and X2 : its
simplices are the images of the simplices of X1 and X2 under the
inclusions X 1 ->■ X 1 *c X 2 and X 2 -> X *c X , and also the images of
113
con X = X * and su X = X *
for any topological space X (see 1.2.6 .8 ), we see that the simplicial
join construction transforms the cone and the suspension over an
arbitrary simplicial space into simplicial spaces.
a2 < cj)(a^ ) - Themaps in^ andin2 define twomaps sch X^ sch Scyl f
and sch X 2 sch Scyl f, and hence two simplicial embeddings,
X Scyl f and X 9 -> Scyl f . The images of these embeddings are called
i z
the (lower and upper) bases of the cylinder Scylf and can be
identified with X1 and X 2 - Moreover, the map M 1 x (0 U 1) M 1J_[_M2
defined by (a,0 ) in 1 (a) ,(a,1 ) in2 ° <j>(a), induces acertain map
sch (X„ x I) -* sch Scyl f, and hence a simplicial map X 1 * I -> Scyl f .
1 S I S
Clearly, together with the inclusion X 2 -+■ Scyl f , this simplicial map
114
St s 1 c s t s , st s' c st s , lk s 1 c Iks .
s t (s,X) = st (s,X *) n X
115
Regular Neighborhoods
8. Exercises
1. Cellular Pairs
r— 1
where e E cell X \ cell A, t £ I, and y E S . Since the functions
r r
hr are continuous and each of them extends the preceding one, together
they yield a continuous function X ^ I. This function is called the
characteristic function of the pair (X,A), and the neighborhood of A
consisting of all points of X where the characteristic function is
less than 1 is called the neat neighborhood of the subspace A.
Obviously, the characteristic function of the pair (X,A)
vanishes on A, and only on A; hence, every subspace of a cellular
space is distinguishable.
if x € U
r- 1 '
Fr (x,t)
cha (((1 —t)i + t)y), if x = cha (ty ),
if 0 g t £ 2 r,
Each homotopy Gr extends the preceding one, and together they yield
an A-homotopy U x i -> u from id U to a map which takes U into A.
The compression of the last map to a map U A is the desired strong
deformation retraction.
into B, for each cell e £ cell X \ cell A. Now take advantage of the
q q
k-connectedness of the pair (Y,B) to deduce that, given any cell
e £ cell^X \ cell^A, there is an S^ 1-homotopy x i -y y from
he to a map whose image is a subset of B. We put
contractible.
t in imm
D ----- | I D ---- -— > X ; (1 )
y J— lv GM v
k +1 s
(iii) g(D \ U ± _ 1 Int d^) c= A'; (iv) for k ^ 1, the point of the
a^(ort^), and the segment joining this point with ort^ is entirely
k +1 s
contained in D \ Uj= 1 Int ^ = 1/---/S).
, _ (abh ) ' 1 ,
K -a- -> h (a) ----- y---- * ba a - (2)
y V
This theorem tells us that there is an m such that the maps (2) admit
simplicial approximations when one replaces by bamK ^ . We let
denote the canonical homotopy from (2 ) to the above simplicial
approximation. Since t is not a face of any other simplex,
F^ ((L D K ) x I) D I n t t = 0, and so together the homotopies F^
define a homotopy F: ( L D ( U K ) ) x I ^ X \ U h (Inti) . By 1.3 and
y y y y
1.3.5.9, F extends to a homotopy G: L x I X \ U h (Int t ) of the
y y
r+1 —1
map ab f : L X \ U^h^(Inti) , stationary on B = D \ Uyf (h^(Intx)).
It is evident that the composite maps
F h.
K x
X iI -— L * a yla -> X
v
and
L x I -JL_» x ^ U h (Int t )
V y
ab f (abh ) ' 1
ba K ----- — »h (ct) ----- y > ba a
X, if x € A' ,
Pt (x) =
° f (h^ 1 (x) ,t) , if x € h (Int D^ +1 ) ,
(i) f _ 1 = f;
126
(imm1 (x, t^ ,t2) ,t) h- imm1 (x,t1 ,tt2) [x £ skekX, t,t 1 ,t2 £I],
•<
if y G Y, and t £ 2
J k
(y /1 )
F i ( F 1 +1 ( ... F k (x,1 ) ...,1),2 t-1 ) , if y € Y. and
J k
O' if k = 0,
Yk =
U c (y) , if k > 0,
y e A k'Ak-i
uu
and Y = Uk=Q Yk . Obviously, Y is a one-dimensional subspace of X
containing skeQX, and the cover {Yk } of Y satisfies the conditions
of Lemma 4. Therefore, YQ is a strong deformation retract of Y,
i.e., Y is contractible.
note that f_ 1 (bp) = [(X ,x.j) *c (x2 ,x2 >] U [ ( x ^ x ^ *c (X2 ,x2 )], and
{fr *
. skerX Yr }r_Q, with the following four properties:
5. Exercises
§1 . FUNDAMENTAL CONCEPTS
1. Topological Manifolds
in each context, from confusing the interior and boundary points, and
the interior part and boundary defined here with the interior and
boundary points and the interior part and boundary of a set in a
topological space.) Clearly, the interior of X is a dense open set,
whereas the boundary of X is closed.
If each point of a topological space has a neighborhood homeo-
morphic to an open subset of IRn or , then obviously it already is
an n-dimensional locally Euclidean space. Consequently, every open
subset of an n-dimensional locally Euclidean space is also an
n-dimensional locally Euclidean space. In particular, the interior of
an n-dimensional Euclidean space is an n-dimensional locally Euclidean
space without boundary. Moreover, the interior and boundary of an open
subset U of a locally Euclidean space X are given by int U =
= U n int X and d\J = U fl 3X.
Since a locally Euclidean space is locally connected, its
components are open (see 1 .3.4.3), and hence also closed.
Obvious examples of n-dimensional locally Euclidean spaces
are 3Rn , 3R^, Sn , and Dn . It is clear that 3nRn = 0 and 3Sn = 0.
Furthermore, all the boundary points of the half space lie in the
limiting hyperplane ^, consisting of the points (x ,.../XR ) such
that x^ = 0 , and all the boundary points of the ball D lie in the
limiting sphere Sn 1.
n n2 n i +n2
3. Since the product HR x 3R is homeomorphic to 3R ,
we see that the product X^ x of two locally Euclidean spaces X^
and X^ of dimensions n^ and , and without boundary, is an
(n i+n2)-dimensional locally Euclidean space. This is true in general,
i.e., a product X x ... x x
of boundaryless locally Euclidean spaces
s 1
X i,...,X
s of dimensions n i ,...,n
s , is an (n I+...+n
s )-dimensional
boundaryless locally Euclidean space. Turning to locally Euclidean
spaces with boundary, note that the formula ((x ,...,x ),(y ,...,y ))
1 nl 1 n2
(x ,. ..,x fyw...fy ), which gives the canonical homeomorphism
1 1 2
n1 n2 n l+ n 2 ni n2 n i+n?
1R x hr ]R r also defines a homeomorphism 3R x ]R 3R
for n 1 > 0. Similarly, the formula ((x , .. .,x ) , (y , ...,y ))
n1 n2
ni n2 n i+n?
** (y-i /•••fY ,...,x ) defines a homeomorphism nR x ]R -* 1R
2 1 "
2 2
(-2 x^y^,x^ - y^,X2 ,...rxn ,y2 '#’’'yn ^ defines a homeomorphism
R1 n2 n l+ n 2
x ]R_ -+ nR_ for n 1 > 0, n 2 > 0. Thus, each of the products
135
n ^1 n2 ^ n. +n^
1R_ x IR , JR x 3R_ , and IR x IR_ is homeomorphic to HR- . We
conclude that given locally Euclidean spaces and X^ of dimensions
n^ and n2 , the product X^ x X^ is an (n^+n2)-dimensional locally
Euclidean space. In general, the product X^ x ... x X g of arbitrary
locally Euclidean soaces of dimensions n., ...,n is an (n.+...+n )-
" 1 s 1 s
dimensional locally Euclidean space.
n1 n2 n1 n2 nl n2 n1 n?
3R x IR , HR x HR_ , IR_ x IR or 3R_ x JR_ ,and composing it
withone of the homeomorphisms exhibited in 3, we obtain a homeomorphism
n 1+n 2 n 1+n2
of x u2 onto !IR or 3R_ . We denote this composition by
nl
cp and analyse the four possible cases. If cp^ (U^ ) = IR and ^2^U2^ =
n2 n 1+n2
- IR , then cp(U^ x U2> - 3R and x^ , x 2 , and (x^ are aH
nl n?
interior points. If cp1 (U1 ) = 3R and <P2 ^U2* = 3R- ' then
n l+ n 2 n i+ n ? “ 1
cp(U^ x U2> = 3R_ , and cp(x^ ,x2l £ 3R^ if and only if
n2 _1
^ 2 ^x 2^ ^ ^1 “ Thus (x ,x ) is an interior (boundary) point if and
137
ni
while x2 is an interior point. Finally, if ip^ (U^ ) = ]R_ and
n n +n - 1
(P2 (U2) = 3R_ , then tp(U1 x v ) =®1 z if and only if
n i" 1 n2 _ 1
cp^ix^) € ]R1 or cp2 (x2> € 3R.j . That is to say, (x1 ,x2) is a
boundary point if and only if at least one of the points ,x2 is
boundary. Our conclusion is that in all cases (x^/X^) is an interior
(boundary point) if x^ and x^ areinterior points (respectively,
if x^ or x^ are boundary points).
Manifolds
One-dimensional Manifolds
This map is injective and continuous, and hence monotone and obviously
-1 -1
increasing (if it were to decrease, the points cp (a) and ip (b)
would haveno disjoint heighborhoods in X). Thus,
]R1 .
In case (ii), <P(U fl V) = (^>°,a^) U (a2 '00) an<^
i|j(U fl V) = (-0°,^) U ( b 2 ,°o) , for some a ^ a ^ ^ ^ (a1 < a2 , b 1 < b2 >,
and we may assume that the composite homeomorphism
ip(U n V) -- ► U n V Ip (U n V)
maps (-^a^) onto (b2 ,°°), and (a2 ,°o) onto (-°°,b,j). Both functions
(-0° ^ ) + (b2/») and (a2 /°°) + (-oo,^), which represent compressions
of this composite homeomorphism, are increasing (if, for example, the
-1 -1
first were to decrease, then the points cp (a^) and ip (b2) would
have no disjoint neighborhoods in X) . We can thuswrite
morphism.
2. Differentiable Structures
1.
Recall that a real function defined on an open subset of
n r r
IR is of class C (or a C -function) if it has continuous partial
derivatives of all orders up to and including r. The definition implies
that 0 < r <°°, that Cis the class of all continuous functions,
CO
and that C is the class of all functions which have continuous partial
derivatives of all orders. In addition, we say that the real analytic
functions are of class Ca (orCa-functions). It isconvenient to
consider a > 00 and thusencompass all theclasseslisted above by the
inequality 0 < r £ a.
Obviously, these definitions can be extended to real functions
defined on an open subset of the half space IR^- To do this, we consider
the derivatives with respect to the first coordinate at the points of the
boundary hyperplane IR^ to be left derivatives, and analyticity at such
points is understood as the existence of an analytic continuation to an
open set in IRn . We further extend the definitions to maps of an open
subset of IRn or of IR^, into any subset of IRq : such a map is of
•p
r r
C -structures and C -spaces
ab “ 1 ab ib
cp(suppcp n supp ip) --- ^— > supp tp n supp ^ iMsuppcp il supp (j)
and
ab , - 1 ab
^(supptp n supp ip) — -— > supp cp n s u p p \p -— cpfsupptp n supp ij),
r r
which are inverses of one another, are of class C (i.e., C -diffeo-
morphisms for r > 1 and homeomorphisms for r = 0). This condition
is trivially satisfied whenever supp cp n supp iii = 0 . If
supp cp nsupp if;^ 0,then the C -compatibility of the charts tp and ip
implies the equality of their dimensions. In fact, this equality results
also from the C^-compatibility of cp and \p, as shown by 1.4.
A collection of charts is an n-dimensional Cr-atlas of the
set X
if these charts cover X, are n-dimensional, and each two of
r r r
them are C -compatible. Two C -atlases of X are C -equivalent if
their union is again a C -atlas. This is clearly an equivalence
relation, and the equivalence classes of n-dimensional Cr-atlases of
the set X are called n-dimensional Cr-structures. The Cr-structures
with r > 0 are called differentiable structures.
Clearly, if 0 £ q £ r, then each n-dimensional Cr-atlas is
143
n1 n1 n? no
such that Imcp^ = IR or ]R_ (respectively, Im ip^ = ]R or H ),
then the composition of x tp2 with one of the homeomorphisms
145
1 2 n l+ n 2 n1 n2 n i+n?
]R x ]R ]R and 3R_ x nR -> HR_ , defined in 1.3, provides
an (n^+n2 )“dimensional chart on x X 2 - If = then the charts
constructed as above form a Cr-atlas of the set Xx X~, with
r = min(r 1 ,r2), and hence define a C -structure on X x X„. Thus the
rl
product of the n 1-dimensional C -space X1 and the n 2-dimensional
2
C -space X2 with 3X 2 = 0 is an (n^ +n2>-dimensional Cr-space, where
r is the smallest of
the numbers r. and r„.In general, theproduct
r .
Smooth Maps
>r ^r
8 . A continuous map f of a C -space X into a C -space
r r
Y isof class C , or a C -map, if for any chart cpon X and any
chart on Y, the composite map
- 1
-1 , ,x , ab cp ^ j-— 1 , lXabf , ib - .
cp (supp cp n f (supp ip )) supp cp n f (suppiji) >supp ip Im ijj
-
XT • •
is an open subset or the boundary of a C-space X, then the inclusion
A ->■ X is a Cr-map. If A is open in X or A = 9x, and B is open
in Y or B = 9Y, then the compression A B of any C -map X -+ Y
nr
is a C -map.
^1 ^1
9. A map f of a C " -space X into a C -space Y is a
diffeomorphism if it is invertible and both f and f ^ are smooth.
The space Y is said to be diffeomorphic to the space X if there is
a diffeomorphism X -* Y, and Cr-diffeomorphic to X, if there is a
C -diffeomorphism X -> Y.
r
Of course, the identity map of a C " -space with r ^ 1 is a
Cr-diffeomorphism. Also, the composition of two Cr -diffeomorphisms is
r r
a C -diffeomorphism, and the inverse of a C -diffeomorphism is a
Cr-diffeomorphism itself, as we may easily see from 2 (iii). Therefore,
the property of being C -diffeomorphic is an equivalence relation.
Using 2 (i), we conclude that nonempty diffeomorphic spaces
have the same dimension.
10. Let
f . : X. -+ Y. ,. ..,f : X Y be Cr-maps with r ^ 1,
1 1 1 m m m ^
where no more than one of the spaces X^,...,Xm , andno more than one
of the spaces Y^,...,Y , has a boundary. Then
Subspaces
C -space X, and consider the maps ab tp: A fl supp tp -+ tp(A fl supp tp)
corresponding to all charts tp on X such that tp(A D supp tp) =
k k
= Im cp 0 3R or Im cp fl 3R_. Each such map is a k-dimensional chart on
A, and together they yield a k-dimensional Cr-atlas of A. The
27
C -structure defined by this atlas transforms A into a k-dimensional
r r
C -space. The topology of this C -space obviously coincides with the
relative topology.
When we extend the Cr-structure of a space to a C^-structure
with q ^ 1, its subspaces remain subspaces. The subspaces of Cgx
q 2T
are called C -subspaces of the original C -space X.
The codimension of a subspace is the difference between the
dimension of the ambient space and that of the subspace.
27
It is evident that the open subsets of a C -space (with
r ^ 1 ) are among its subspaces; in particular, we cite its interior
27
and its components. Warning: if not empty, the boundary of a C -space
is not a subspace.
It is readily seen that if A is a subspace of codimension
0 of a Cr-space X, then int A = Int A fl int X .
27
A subspace of a C -space is neat if itis closed as a subset
and its boundary is contained in the boundary of the space. Note that
every neat subspace of codimension 0 is made of whole components of
the ambient space.
27
C -manifolds
r r T
14. AC -space is a C -manifold, or a manifold of class C ,
if it is a topological manifold, i.e., a second-countable, Hausdorff
space. A manifold of class C° is simply a topological manifold; see 5.
The manifolds of class Cr with r ^ 1 arecalled smooth, or
differentiable. The manifolds of class Ca are called (real) analytic.
27
Since any C -structure is defined by one of its atlases, it
is interesting to discover those properties of an atlas of a Cr-space
which guarantee that the space is Hausdorff and second-countable. Here
we formulate only two obvious conditions: if each pair of points is
covered by a chart of the atlas or by two disjoint charts of the atlas,
then the space is Hausdorff; if the atlas is countable, then the space
is second-countable.
There is no need to check that a space is Hausdorff and
second-countable if the differentiable structure is introduced on a set
which is already a topological manifold, and the topology defined by
the differentiable structure coincides with the initial one. If the
differentiable structure is defined by an atlas then the two
topologies agree if and only if supp cp^ are open and cp^ are homeo-
morphisms; see 5.
A smooth manifold is closed if it is compact and has no
boundary; cf. 1.10. Warning: at the present time the equality
3X = 3(C°X) is not proven (see 6 ). Therefore, we should be careful
to distinguish between the smooth manifold X being closed and the
topological manifold C^X being closed.
17.
Finally, we note that every real, n-dimensional vector
27"
space has a natural C -structure for any r (0 < r £ a ) , which makes
27*
it into an n-dimensional C -manifold. This structure is defined by
the linear charts i.e., by the linear maps onto IRn .
3. Orientations
PROOF. Let cp^ E Catl (int X^) and cp2 E Catl (int X 2) be
n1 n2
charts with Im ^ = 1R and Im cp2 = 1R , respectively. Clearly,
155
X ^2
^ Catl(int(X^ x x2)), cp2 x € Catl(int(X2 x x^ )) , and the
n +n n +n2
local representative IR -*IR of the canonicaldif feomorphism
X1 xX 2 X2 * X 1 relative to these charts is given by
(x1 .... Xn 1 +n 2> " (xn 1 +1' • • " xn 1 +n 2 'x 1 " - - ' xn 1) • Therefore, the
n1 n2
Jacobian of this local representative is (-1)
with id(Xi+^ x ... x Xg), while the second is the product of the
canonical diffeomorphism
(X1 X ... X Xi_1) X 3Xi -> ax^^ X (X1 x ... X Xi _ 1 )
n 1 + “ *+ni- 1
explains the factor (-1 )
is meaningful [and is the linear map IRn ]Rn whose matrix is the
Jacobi matrix of the map loc (tp,^) id at the point cp(x)]. We denote
this differential by d^icp,^).
Now consider the real vector space of all the maps
Atl X-> IRn (with the natural operations) . It is clear that those maps
v:Atl^X -* HR which satisfy the relation v(i¡j ) = d (cp,ij;)v (cp) for any
two charts cp,ip £ Atlxx yield a subspace of this vector space, i.e.,
they form a real vector space. The latter is called the tangent space
to X at the point x and is denoted by Tang X. The maps in Tang X
x x
are called tangent vectors at the point x (on X ) , or vectors
tangent to X at x .
Obviously, a tangent vector is completely defined byits
value on an arbitrary chart of Atl X, and for each chart <p £ Atl X
n x x
and each vector u £ 3R there is a vector v £ Tang X with v(cp) = u.
„ X
Therefore, the mapping tp# : TangxX + 3R , tp# (v) = v(ip), defined for a
157
pr 1 (x) = Tang^X.
The set Tang X has a natural topology whichmakes it a
topological manifold. Furthermore, for r ^ 2, Tang X has a natural
differentiable structure. In order to describe these structures, we
define, for each chart cp E Atl X , the map tn cp : pr 1 (suppcp) ->
+ Im cp x ]Rn , tncp(v) = (cp ° pr (v) ,cp^ (v) ) . If we identify JRU x JRU
with HR2n, and consider Im cp x nRn to be an open subset of lR^n or
]R2n, then we may interpret tn cp as a 2n-dimensional chart on TangX.
Obviously, if ]p E Atl X is another chart, the composition
/
= tn cp (pr (suppcp)\ fl pr " 1 / ,
(suppijj)) ab(tncp) 1 >
----------
-1 z v -1 , -x ab tn :d
pr (suppcp) fl pr (suppy) -----
bj “ l j (al '•*•'an }
and t ,...,£
are the coordinate functions of the map loc (cp,i[i) id
1n
(D. denotes the partial derivative with respect to the i-th coordinate)
l r— 1
Formula (2) shows that the charts tn cp and tnare C compatible
(we set C° °’ 1 = C°° and Ca _ 1 = Ca ) .
Moreover, the charts tn cp ,
r— 1
tp £ Atl X , cover Tang X , and thus yield a C-atlas ofthe set
Tang X . This atlas has a countable subatlas (since AtlX has such a
subatlas). Furthermore, for any two vectors of TangX, it has either
158
r >r
3. Let f
be a C -map of an m-dimensional C ' -manifold X
>£•
into an n-dimensional C ' -manifold Y, r s 1. For a point x £ X and
two charts tp £ Atl X and \p £ Atlf , .Y, we let d (f;ip,^) denote
x r \X ) x
the differential of the map loc(tp,i{j)f at the point tp(x), regarded
as the linear map 3R -*■ ]Rn whose matrix is the Jacobi matrix of
loc (cprvp)f at tp(x) . If tp' £ Atl X and \p'£ Atl... .Y are two
x r f(x)
other ch a rts , then dX(f ;tp' , ijj' ) = d -I. IXj
. (ij) ) ° d X (f;tp,^) ° dX (tp',tp).
Combining this relation with the equalitiesd (ip',<p) = tp„ o (tp■)— 1
x w w
1 _
(4>p 1 o d x ( f ; tp ’ ) o ^ o dx (f;tp,i^) o tp ,
i.e., the linear map ijj” 1 o d^. (f ;tp,^ ) ° tp# : TangxX -»■ Tangf(x)Y does
not depend upon the choice of the charts tp and ip . This linear
map is called the differential of the map f atthe point x and is
159
pr pr
X
is clearly commutative.
Let be the coordinate functions of the map
loc (cp, f. Then the local representative loc (tn cp,tn ^ )df of the
differential df is given in coordinates by the formula
(a 1 ' “ ‘'am;u i ' * *.'U^) ^ (b-j / •••fb^; v^ ,.. .,v ) , where
bj = V ai V '
j = 1 ,...,n.
v j = ^i = 1 Di'ej ( a 1 am )ui
r—1
Therefore, df is of class C
Of course, d(h o f) = dh ° df for any smooth map h of Y
into a third smooth manifold, and df = id(Tang X) when X = Y and
f = id X . Also, if f is a diffeomorphism, then df is a diffeo-
morphism for r ^ 2 and a homeomorphism for r = 1.
In the special case when X is an open subset of nRm or of
3R™, and Y is an opensubset of nRn or of IR^, in addition to the
differential d f: TangX + Tang,. ,.Y one has the classical
differential of the map f, i.e., the linear map IR + JR whose matrix
is the Jacobi matrix of f at the point x. Let us identify the spaces
Tang^X and 3Rm (Tan9f(x )Y anc^ v^a t^ie l^near isomorphism
(id X) ^ :Tang^X 3Rm (respectively, (id Y) ^ : T a n g ^ ^ Y + 3Rn ) . Then it
is clear that d^f becomes the classical differential.
Vector Fields
both maps are of class Cr . Thus the total manifold of vectors tangent
to a Cr-parallelizable n-dimensional smooth manifold X is Cr-diffeo-
morphic to the product X x ]Rn .
•£
1. A map f: X Y of smooth manifolds is a C -embedding
if f (X) is a Cr-submanifold of Y and ab f : X f (X) is a
Cr-diffeomorphism. For example, the inclusion of a submanifold into
its ambient C ^ r -manifold is a Cr -embedding. Since every map
f: X -* Y can be written as the composition of its compression
ab f : X + f(X)
with the inclusion f (X) + Y, a Cr-embedding is
27
really a map of class C .
The -embeddings are also termed differentiable embeddings.
Using Theorem 3 we shall prove below, we can see that a differentiable
embedding which is of class Cr is a Cr-embedding. Moreover, it is
evident that differentiable embeddings are topological embeddings. The
latter are sometimes called C^-embeddings.
A differentiable embedding f : X -+ Y is neat if f (X) is a
162
Immersions
JC
C -embedding h: W Im cp , such that, for i = 1 , .. .,m, one has:
f(N) — '
b ^ -» iMf(N)) -- ► W — -* h(W) - (abtP) 1 » N ,
(see 4).
Submersions
the pair (Imcp,Im ip) coincides with one of the pairs (IRm ,IRn ) ,
QR™ ,IRn ) , or 0R^,3R^) , and in each case cp(x) = 0 and ij-(f(x)) = 0;
the corresponding local representative loc(cp,\JJ)f can be described,
in the first case, as the projection of the product HRn * 3Rm n onto
its first factor, in the second case - as the projection of the product
JR^ n x IRn onto its second factor, and in the third case - as the
projection of the product 2R^ x nRm n onto its first factor.
and
iK(z)
(z) = --- - ^ —2--- 7 (j =
D e - | \p j ( z )
6. Complex Structures
Complex Manifolds
7. Exercises
1. Stiefel Manifolds
1. We denote by
JRV(n,k), or simply by V(n,k) (0 £ k £ n ) ,
k n
the set of linear isometric maps 1R ->■1R . Such a map is uniquely
k
determined by the images of the vectors ort^,...,ort^ E 3R , i.e., by
yn . v .v . = 6 .. (1 £i £ j £ k) . (1 )
L s=^ si sj id J
nk
We may regard a matrix livsill as a P°i-nt of 3R , if we
index its entries v . in dictionary order. Thus V(n,k) becomes
si
nk
the subset of 3R defined by the equations (1). An easy computation
shows that the gradients of the left-hand sides of (1 ) do not vanish
and are pairwise orthogonal on this subset. Hence V(n,k) is an
172
a nk
[nk-k(k+1)/2]-dimensional C -submanifold without boundary of 3R
(see 1.2.12). V(n,k) is called the Stiefel manifold.
Clearly, V(n,0)reduces to a point, V(n,1) is just the
sphere Sn \ and V(n,2) is thesubmanifold of all vectors of unit
length in Tang Sn 1 .
2. The points of V(n,n) are orthogonal transformations of
3Rn , or orthogonal matrices of order n, and V(n,n) is usually
denoted by 0(n). The composition of transformation (multiplication of
matrices) induces a group structure on 0(n). The subgroup of 0(n)
consisting of all matrices with determinant +1 is denoted by SO(n) .
The two sets SO(n) and 0(n) \ SO(n) are open in 0(n), and hence
are Ca-manifolds. They are actually Ca-diffeomorphic: multiplication
by an arbitrary matrix from 0 (n) \ SO(n) establishes a diffeomorphism.
Moreover, the manifold SO(n) is canonically Ca-diffeomorphic to
V(n,n-1): a matrix from V(n,n-1) is carried by this diffeomorphism
into a matrix from SO(n) through the addition of a column; that is
to say, we complete each orthonormal (n-1)-frame in IRn to a positive
orthonormal n-frame.
We further note that SO (2) = V(2,1) = s \ and the group
structure on S O (2) agrees with the group structure on the circle S ,
considered as the multiplicative group of complex numbers of modulus 1.
a k —cr k
Finally, the inclusion IR M ->•]R
induces a C -submersion
k n
V(n,k) -*■ V (n,k-q) , which transforms each map cp: 3R -*•1R into the
k —cr in k n
composite map ]R M -- > ]R ■+tr , i.e., each frame (v ,...,v ) is
I K
taken into the frame (v^,...,vk_q )• Using Theorem 1.5.9, it is readily
seen that this is indeed a submersion, because given any frame
.0 0, , .
(v^,...,vk ) € V(n,k) we have an explicit construction of the neighbor
hood V of the frame (v°,...,vk_q ) in V(n,k-q) and of the Ca-map
g : V ->■ V (n,k) which are required by this theorem. In fact, one can
take as V the set of frames (v 1,...,vk_q ) in V(n,k-q) such that
4.
We see from equations (1) that the set V(n,k) is bounded
nk
and closed in 1R . Therefore, V(n,k) is a closed manifold.
The manifold V(n,n-1) = SO(n) is connected: each matrix of
SO(n) can be expressed as cu (^ ,...,(j)^)c ^ where
cos -sin
sin cos
CO S -sm
U ( (¡)
1'
sin cos
Y-n 1 v . v . = 6 .. (1 £ i £ j £ k).
S = 1 SI
SI SI
SJ 111
J
, v ^ „nk m 2 nk
These equations show that (CV(nfk) is a subset of (E - IR . Now
2
we separate their real and imaginary parts and obtain k real equations
174
such that the gradients of their left-hand sides do not vanish and are
2
pairwise orthogonal on CEV(n,k) . Thus (EV(n,k) is a (2nk-k )-dimen
sional Ca-submanifold without boundary of IR^n^ , called the complex
Stiefel manifold.
Warning: (EV(n,k) is nota complex manifold in the sense of
1.6.5. The present definition doesnot equip it with a complex
structure; in fact, such a structure does not exist in general, since
CDV(n,k) is odd-dimensional for k odd.
Clearly, (EV (n,0) reduces to a point, and CCV(n,1) is just
the sphere S^n The points of the manifold CCV(n,n) are unitary
transformations of CCn , and CCV(n,n) is usually denoted by U(n) .
Like 0(n), U(n) is a group under the composition operation o . The
subgroup of U(n) consisting of all matrices with determinant 1 is
denoted by SU(n) and is a Ca-submanifold of U(n), canonically
diffeomorphic to (EV(n,n-1) (cf. 2).
The manifold U(n) is canonically diffeomorphic to
1 1
SU(n) x S : this diffeomorphism takes each pair (u,z) E SU(n) x S
into the matrix obtained from u by multiplying its first row by z.
Warning: this diffeomorphism is notagroup isomorphism between the
direct product of groups SU(n) x s and U(n).
The Ca-embeddings (EV(n,k) (EV(n+q,k) and
(EV(n,k) (EV (n +q,k+q) , and the Ca-submer sion CEV(n,k) CEV(n,k-q) are
defined exactly as in the real case. Moreover, since each linear
k n
isometric map (E + (E may be regarded as a linear isometric map
2k 2n a
H ->•3R , there is a canonical C -embedding (EV(n,k) -*■HRV (2n ,2k ) .
The manifolds (EV(n,k) are compact. Moreover, they are all
connected. Indeed, (EV(n,n) = U (n) is connected, because every unitary
matrix can be expressed as cu ( ^ , ...,<j>n )c , where u (<j>^ , .. .,<J> ) is
the diagonal matrix with entries exp (i<J).j) , ...,exp (i<|> ) , ... ,4>n £ ir)
and c is a unitary matrix, and then joined to the identity matrix
by the path tn-cu((1 - t) <|>1 ,...,(1 - t) $ ) . The manifolds (EV(n,k)
with k < n are connected as continuous images of IV(n,n).
¿i=1 uiv i-> In this way, we obtain: the C -manifold without boundary
HV(n,k) (0 ^ k ^ n) of dimension 4nk-(2k2 -k), called the
quaternionic Stiefel manifold; the Ca-embeddings HV(n,k) +HV(n+q,k)
HV (n ,k ) + KV(n+q,k+q) , and MV(n,k) -v (EV(2n,2k);and the Ca-submersion
HV(n,k) + HV(n,k-q)
Clearly,' HV(n,0) reduces to a point, and UV(n,1) is just
4n— 1
S . The manifold HV(n,n) is usually denoted by Sp(n); its points
are the linear isometric transformations of H n , and the composition
of transformations makes Sp(n) into a group.
All the manifolds living) are compact and connected. Since
the proof of connectedness along the lines in 4 or 5 requires the
normal form of amatrix from Sp(n), which is less known than the
normal forms for SO(n) and U(n), we remark that a different proof
of connectedness is given in Chapter 5 (see 5.2.7.4).
7. Let
IRV* (n,k), or simply V'(n,k), 0 £ k £ n, denote
k n
the set of linear monomorphisms 1R -*•1R . Alternatively, one may
describe V'(n,k) as the set of nondegenerate k-frames in 3Rn , or as
the set of the real nxk-matrices of rank k. Clearly, this set is open
nk
in the space IRof all real nxk-matrices, and hence V 1 (n,k) is an
nk-dimensional Ca-manifold containing V(n,k) as a submanifold.
Let T(k^IR), or simply T(k), be the set of real upper
triangular matrices of order k with positive diagonal entries.
Trivially, T(k) is open in the space of all upper triangular matrices
of order k, n R ^ ^ +1^ 2 , and is actually diffeomorphic to +
If we orthogonalize a given nondegenerate k-frame in lRn via the
standard procedure, then the matrix corresponding to this frame takes
the form ut, where u £ V(n,k) and t £ T(k). Moreover, it is
obvious that this representation is unique and defines a diffeomorphism
V'(n,k) -* V(n,k) x T(k), transforming V(n,k) into the fiber
V(n,k) x E, where E is the identity matrix. Therefore, the manifold
V'(n,k) is Ca-diffeomorphic to V(n,k) x i R ^ ^ + 1^//^ / and V(n,k) is
its strong deformation retract. In particular, V*(n,k) is connected
for k < n, and V f(n,n) has two components.
Again, V f(n,0) reduces to a point, while V*(n,1) coincides
with nRn ^ 0. The manifold V*(n,n) is usually denoted by GL (n ,1R)
and its points are the nondegenerate linear transformations of IRn or,
equivalently, the nondegenerate matrices of order n. The composition
176
and ]R^ ^ ~*~n > IR^ ^ nRn , and so we obtain the Ca-embeddings
V 1 (n,k) V 1 (n+q,k) , V'(n,k) -+ V'(n+q,k+q), and the Ca-submersion
V ' (n,k) -> V' (n,k-q) , respectively; cf. 3.
8.
We let (EV' (n ,k ) (0 £ k £ n) denote the set of linear
k n
monomorphisms (E -+ (E . Alternatively, (EV(n,k) is the set of non
degenerate k-frames in (En ,
or the set of complex nxk-matrices of rank
nk
k. Again, it it clear that this is an open set in the space (E of
all complex nxk-matrices. Thus (EV!(n,k) is a 2nk-dimensional
Ca-manifold containing (EV(n,k) as a submanifold.
Repeating what was said in 7 (with obvious modifications),
we obtain a Ca-diffeomorphism (EV* (n,k) CEV(n,k) x T(k,(E), where
T(k,(E) is the manifold of complex upper triangular matrices of order k
with positive diagonal elements. Clearly, T(k,(E) is Ca-diffeomorphic
k2 k2
to IR , and hence (EV* (n,k) is diffeomorphic to (EV(n,k) * IR . In
particular, the manifolds CEV*(n,k) are all connected. (EV'(n,0)
reduces to a point, while (EV! (n,1) is just (En \ 0. Usually the
manifold (EV’(n,n) is denoted by GL(n,(E); it consists of all non
degenerate linear transformations of (En , and is a group with group
operation o. There are also the natural Ca-embeddings
(EV 1 (n,k) -* (EV 1 (n+q,k) and (EV'(n,k) (EV 1 (n+q,k+q) , as well as the
natural Ca-submersion (EV1 (n,k) (EV *(n,k-q) .
2. Grassman Manifolds
u i = e i + £s = 1 C isGs (i = c is €IR)-
This construction yields a map cp(e,e) from the set U into the space
k (n-k)
HR of real kx(n-k)-matrices. Obviously, <p(e,e) is invertible,
and the collection of maps <p(e,E) obtained from all the pairs e,e
is a Ca-atlas of the set G(n,k). Since any two planes in G(n,k) are
contained in some set , the topology defined by the atlas (<p(e,e)}
is Hausdorff. Moreover, this atlas contains finite subatlases (for
example, the subatlas consisting of the charts cp(e,e) where e and
c are subframes of the standard frame ort^,...,ort of 3Rn ), and
hence the above topology has a countable base. Thus, the atlas (<p(e,e)}
transforms G(n,k) into a k(n-k)-dimensional Ca-manifold, called the
Grassman manifold.
Clearly, G(n,0) and G(n,n) both reduce to points, and
G(n,1), as a set and a topological space, is identical to lRPn So
we see that the space ]RPn 1 has a Ca-structure compatible with its
topology, and hence lRpn is an (n-1)-dimensional Ca-manifold, We
remark that this (^-structure may be described directly and
conveniently as follows: for k = 1, the finite atlas of the manifold
G(n,k) given above consists of n charts cp^ : -> IRn 1 f...,
co : U -> nRn \ defined in homogeneous coordinates by the formulas
n n
U . = { (x : ...:x ) ] x . t 0},
i 1 n i
cp ( e 1 ,£ 1 ) ,
such that j ^ ( e f )= j £ (e > + 1- Thenit: is readily seen that
the coordinate c^ with m= -i + 1 does not vanish on the
intersection of the supportsof the two charts, but takes all the
remaining real values on this intersection. A simple computation shows
that
(c. Co - c. Co ) c n \ if i * s m,
is im lm Is Im
if i = 1, s m,
c ls°lm '
c!
is if i t I, s = m,
cimC£m'
if i -
I, s = m,
,
and thus the Jacobian J (cp(e,c ),cp(e ',c 1 )) [see 1.3.1] is equal to
(-1)nc ^ . If n is odd, this Jacobian takes both positive and negative
values; hence the manifold G(n,k) is not orientable for such n. For
k[j (e)+. ..+jk (e)]
n even, the formula tp(e,e) (-1 ) defines a map of
the atlas under consideration into S°, which satisfies the
compatibility condition in 1.3.1: this condition obviously holds for
contiguous charts, which in turn implies the compatibility for non
contiguous charts. Hence, for n even G(n,k) is orientable.
'(At) 1det A 0 0
0 (At )”1detA . 0
0 0 .. (At) 1det A
^k(n-k)
0 .......... a
+ + rk + + = 0 .
9. The maps
( y 1#y 2 ) h- (2(1 - p 2 ) 1 / 2 y 2 , 2 y 2y r 1 - 2 |y 2 I2 ) /
2 2 1/2
where y ,y2 € (Ea and P = ( | | + |y2 ! ) <the first map is just
DS, but we do notuse this fact explicitly). These three maps yield
16 16 16 16
a map F : D S x S x S , and one can easily verify that
16 2
zer(F) = zer[pr: D (EaP ].
q - shi (cp(ort )) q ort 2 - shi (cp(ort2 ))q ort 3 - shi (cp(ort3 ) ) q ort4 ,
4. Exercises
Y E lRG(n,k) into the matrix of the composite map ]Rn — y —1-n-» 3Rn
(where pr is the orthogonal projection) is a Ca-embedding. Show that
2
the same is true for the map (EG(n,k) ■+ (E which takes each plane
y E (EG(n,k) into the matrix of the composite map (En Pr > y - (En .
1 d - t2)kdt = 1
lim ,
k^°° k -6
where
■1 9 1c
(1 - t ) dt .
-1
1 (1 _ t 2 ,kdt , 2. (1 - M 1 - 6'
0 < 1 L
ak J- 6
1 - 4
The left inequality is plain, while the right one follows from the
estimates
2 V
(1 - t )Kdt = 2 (1 - t2)kdt < 2 ( 1 - 6 ) (1 - 6 2)k
ak -
-6
and
6/2
(1 - t2)kdt > 6 ( 1 - 4 p)k .
- 6/2
1 n — ,
? v n
1 - (1 - tz)Kdt
4M
for all k > K (see 2). We next show that |[p^(f)](x) - f(x) | < g
for x E In and k > K.
Write [p^(f)](x) - f(x) as
n
n (f(t1+x1,...,tn +xn ) - f(x1,...,xn ))1 fn - tj)kdt1...dtn
n j =1
and then replace the integrand by its absolute value. Now divide the
new integral into two integrals, one over the cube [— 6 ,6 ]n and one
over its complement [—1,1]n \ [-6 ,6 ]n . We obtain
|f (t1+x1,...,tn+xn ) - f (X l ,...,xn ) I £ 2M
and
(1 - t2)kdt,...dt
n j 1 n
,n j= 1
*k [-1 ,1 ]n^t- 6 ,6 ]
n n
(1 - tj) dt 1 ...dtn n (1 - t?
t.))^dt 1 .-.dt
n j 1 n
n j= 1 n j= 1
‘k [-1,1] k [-6 ,6 ]
Therefore, |[p^(f)](x) - f (x) | < e .
[p^ (D^f)] / • • •/ ) =
_1_
n f < t , ............‘ „ i f t : f r « ’ - (tj-x i 2)kd t 1. . . d t n . ( 3)
n 1 3 = 1 J J
ak i
Since
91 . I \ j j 9x . I ' j j
1 3=1 J J 1 J=^ J J
2. Singular Values
mn r , r~ / ,s + 1 ]
[2 b (a/n/m) iq = cmn-q (s+ 1 ) ,
nowhere dense.
We proceed inductively on n. For n = 0 there is nothing
to prove; hence it is enough to show that the theorem holds for
n = k +1 if it holds for n = k.
We start with the special case F^ = 0. Since C is compact,
it suffices to find, for each point x E C, a neighborhood V in U
such that f(C n V) is nowhere dense. D, Assume that
f (x) t 0
K 1 ? k +1
(we can always achieve this by reindexing the coordinates in nR and
k +1
]r") and consider the map g: U -> JR defined by the formula
g(w) w ^abf^q.
It transforms each point from g(W) into a point which has the same
last coordinate. In particular, for any real number u, h can be
compressed to a map
k k
If we identify g (W) fl [3R x u] with its orthogonal projection on JR
in the standard fashion, and JR^ 1 x u - with its orthogonal projection
Q—1 oo k cf— 1
on JR, we obtain a C -map h^ of an open subset of JR into JRM
Clearly, the Jacobi matrix of hu at the point (y1 If-*-/yvK.) is
obtained from the Jacobi matrix of the map h at the point
(y^l *•••r r u ) by deleting the last column and the last row, which has
the form 0,...,0,1. Therefore, the rank of the first matrix is less
than q-1 if and only if the rank of the second matrix is less than q,
i.e., if (y^ ,...,y^. ,u) E g(F n W). Applying the induction hypothesis
to h , we deduce that the intersection of h(g(C n CIV)) with each
hyperplane JR^ 1 x u is nowhere dense in IR^ 1 x u. But if this is
the case, h(g(C fl Cl V) ) has no interior points in 3R^ and we need
only note that this set is closed and coincides with f(C fl Cl V) .
Now let us turn to a second special case: C cF and
Fs+1 = ^ (for some s). Again, it is enough to exhibit for each point
x E C a neighborhood V in U such that the set f (C D V ) is nowhere
dense. Let cp be a derivative of order s of one of the functions f.
3
193
that one can take any neighborhood of xwith compact closure included
k
in W for V. To do this, note that g(C) c: IR andrestrict the map
k cr
(4) to a map h: g (W) n 3R HR . Clearly, all the derivatives of order
£ s of the coordinate functions of h vanish on g(C n W). Using
the induction hypothesis, it is evident that h carries the compact
parts of the set g(C n W) into nowhere dense sets. Finally, we
observe that g (C fl CIV) is a compact part of g(C n W) , and that
h (g (C fl CIV)) is just f (C fl Cl V) .
At last, wecome to the general case. According to Lemma 2,
there exists a numberr such that the set f(C fl F^) is nowhere
dense. We shall prove statement 3 by induction on r, i.e., assuming
that f (C fl F ) is nowhere dense, we show that, for r = 1, f (C ) is
nowheredense, and for r > 1, f (C ni-s nowhere dense.
Let G be an open nonempty subset of 3Rq . Since C n F^ is
compact and f (C fl F )is nowhere dense, the set C n has a
Now we are back to one of the cases covered by the first part
of the proof (namely, in the first case for r = 1 , and in the second
one for r > 1). Therefore, we conclude that f(C') does not cover
G ^ f(Cl N) . Consequently, if r = 1 the set f(C) does not cover
G, while if r > 1 the set f(C n F ) does not cover G. This
completes the proof, because f(C) and f (C n are closed.
2
1. Let f be a real C -function defined on an open subset
of nRn . A critical point y of f is nondegenerate if the second
differential of f at y (considered as a quadratic form) has rank n.
The index of the second differential of f at y (i.e., the number
of negative squares in the diagonal representation of this form) is
called the index of the point y and is denoted by ind^y.
2
We remark that if cp is a C -diffeomorphism of an open
subset U of IRn onto another open subset of nRn and y is a non
degenerate critical point of f: U 3R, then cp(y) is a nondegenerate
critical point of the function f o cp ^ : cp(U) -* HR, and ind _>|<p(y) =
= ind^y. Both conclusions remain true in the more general
1 -1
situation where cp is only of class C / but the function f ° cp
2
is of class C -
Now consider the function JRn -> HR defined as
f (tx) = X£ = 1 x.D.f(tx) .
f(x) = ^ =1 x.f.( X )
r —1
in some neighborhood of 0, where f^ are C -functions.
Differentiating, we obtain D^f(x) = x^D^f^(x), since f^ (0) =
= ... = fn (0). Again we apply Lemma 2 and write, in a neighborhood
VQ of 0,
r —2
with C -functions f^j. Therefore, for x € Vq ,
fix) ■
V W V W v0 *
p p p- 1 1 0
ab cp ab cp9 ab cp ab ip ab tt
V ---- ^ cp1 (V) -----1 cp2 (cp1 (V) ) -----i ... ----- > Vr ---- ► n (vn ) ,
x h- ±x 2A ± ... ± x 2 + v).
n \
. h..(x)x.x.
1 q i ,D=q + 1 ID ID
where h . . = h.. and h - .1 (0) ^ 0 . Now consider the subset L of
ID D1 q+ 1 ,q+1
t (V ) consisting of all the points x where h 1 (x) f 0 and
q q +\ tq +I
has the same sign as h 1 1 (0) , and then define : L -> 3Rn as
q + I,q + l
£ = X .
, v>
+ V
hs —/q+ 1 (x)
' 1 -----
q+1 s'>q+1 S hq+q>q+ 1 (x) ‘
q+ 1 ,q + 1
APPROXIMATIONS
^r-
1. Let X and X1 be C -manifolds
(0 £ r £ a ) . Wedenote
27 r
by C (X,X 1 ) the set of all C -maps X X 1 . If r £ 03, we equip
r . r r
C (X,X') with the C -topology which makes C (X,X') into a topological
space, as follows. Given two arbitrary charts tp G AtlX and
tp1 E Atl X 1 , a sequence of nonnegative integers r^ ,...,r with
n = dim X and r^ + ...+r^ < r, a compact subset A of Im cp , and an
n*
open subset A' of IR , where n* = dim X 1 ,consider the subset of
27
C (X ,X 1) consisting of all maps f such that
r r
[D .. .D nloc (tp,tp1 )f ] (A) CI A 1.
1 n
one of the orthogonal projections IRn -> !Rn , ]R^ -+ nRn (see 1.5.7).
Now cover X with a finite number of sets U = cp 1 (Int Dn ) , say
X X
ux ,...,Ux
„ , and denote by ti the--- ^
subset of C (X,X ') consisting of
XI "s
all maps f such that f (Cl U ) c suppip1 and the left n'^n' -minor
Xi xi
of the Jacobi matrix of the map loc (ip ,cp' )f has no zeros on Dn
x i xi
The intersection U 1 fl ... fl U is the desired neighborhood of the map
V
4.If X is compact, then the set Embr (X,X') is open in
Cr (X,X’) (1 ^ r £ oo) .
3R , or IR^
(see 1.5.1). Now cover X with a finite
n ~1
number of sets U = tp (Int D ) , say U , ...,U . Let U. be the
x ^x 2 x ' x 1
27 1 s
subset of C (X,Xf) consisting of all maps f such that
f (Cl U ) c: supp cp' and, if we symmetrize the upper nxn-part of the
xi xi
Jacobi matrix of loc (cpX .,cpx
* ,)f and take all the principal minors,
1 1
they are all positive on the ball Dn . (The principal minors are the
left-upper minors; the symmetrized matrix is half the sum of the matrix
with its transpose.) Finally, denote by U that part of Cr (X,Xf)
consisting of all maps such that the preimage of any point of X1 lies
in one of the sets U . Let us show that the intersection
x.
l
1/ = 11^ fl . . . H n U is a neighborhood of f^ with the necessary
property. It is clear that f^ £ 1/ and that all the sets IL are
open. Hence it suffices to verify that: (i) U is open, and
(ii) the maps in IL are injective.
To prove (i), note that U is the preimage of the set
^r
1. Every compact C -manifold, 1 £ r £ °°, admits a
C -embedding in a Euclidean space of sufficiently high dimension.
j x <y> =
(0,0) , if y 6 X supp ip
X
n
Here 6 : HR + IR is given by 3 (t^,..., t^) = t^a(t^, ...,t^) . Covering
In fo rm a t i o n
Ux G X d T (x,p) and Ux £ X [dT (x,p) ^ s T <x »P)i are ca l l e d the tube (or
203
the tubular neighborhood) and the open tube (or the open tubular
neighborhood) of radius p of the transversalization t , and are
denoted by Tub^_p and tub^_p , respectively.
A tube Tub^_p is said to be neat ifthere is a a > p such
that: (i) the open balls dT (x ,g ) \ sT (x,cr) , x € X, are pairwise
disjoint and the open tube tubT<J they form is a neighborhood of
j(X) in 3Rq ; (ii) the map of this neighborhood onto X, which
transforms all the points of dT (x,cr) \ sT (x,a) into x, is smooth.
The restrictions of the last map to TubTp or tub^_p (which obviously
do not depend on the choice of a) are called projections and are
denoted by PrT •
If Tub^p is neat, then all the tubes Tub p 1 with p' < p
are obviously neat.
Warning: it may happen that the normal transversalization
does not have a neat tube, or even a tube such that the balls d^_(x,p)
are disjoint; see exercise 11.4.
r cr
Tub^_p is a neat tube, then it is a C -submanifold of TR with
int (Tub^p) = tubTP, and prT : TubTP -* X is a Cr-submersion.
PROOF. Let a > p be such that the conditions (i) and (ii)
in 2 are satisfied. As 3 shows, the map ab nat : tu a -* tub a is
-1
invertible and its inverse ab nat : tub a ■> tu a is obviously given
-1
by y ** (pr (y),y - j ° prT (y)). This formula shows that ab nat is
smooth and thus a C -diffeomorphism. Now it is evident that the
properties of Tub^p and the projection pr^: Tub^p -> X which we have
to verify are consequences of the properties established in 3 for their
models Tu p and ab pr. : Tu p X.
t 1 T
5. Every smooth transversalization has a neat tube.
(see 1.2.7.3). We construct the neat tube Tub^.p 1 with p' £ e/2
and choose 6 to be the smallest of the two numbers e/ 2 and
Dist(j ' (X'), IRq \ tub^,p '); note that 6 > 0 (see 1.1.7.15). Then,
according to 1.1.5.12, the composition
where ab j = [abj : X ->■ TubT (p/2)] (see 1.1). Since it takes Cr-maps
into Cr-maps, it suffices to prove that the preimage 1/ of U under
this mapping intersects Cr (Tubx (p/2),tubx ,p ’). But V is open and
contains the restriction to Tubx <p/2) of the composition
[abg: Tub^(p/2) tub^ ,p '] £ 1/. Finally, apply Theorem 3.1.4 to the
coordinate functions of (1 ) to deduce that there exists a map g with
polynomial coordinate functions whose partial derivatives have the
property above. Thus I/ (1 Cr (Tub^. (p/2) ,tub^ ,p ') / 0.
Supplements to Theorem 2
max dist (j* o f(x),jf o g(x)) < Dist (j 1 (X) , JRg \ tub ,p 1 ) .
xex T
One may check directly that <2> is continuous and 0(f)=f; hence
—i —1
the set 0 (U) is open and nonempty. Now Theorem 2 shows that $ (U)
t* r
contains a C -map. Finally, we note that $ takes C -maps into
27
C -extensions of <\>.
>r
9. Suppose that X and X' are closed C' -manifolds and
is a submanifold of X which is itself closed as a manifold. Let
(5 ; A -> X ' be a Cr-map. If 0 ,< s < r ^ then that part of
CS (X,X’)
t S
consisting of the C -extensions of <(> is dense in the part of C (X,X’)
208
s
consisting of all the C -extensions of d>.
x h- ^ if j (x) £ TubTP,
Collars
Glueing
and
¥(k(z,t)) = (<Mz),-t)
z £ C, t £ [0,1)
V (k ’ (z,t) ) = (ip (z) ,t)
t
In the proof that follows we let f^ and denote the
composite maps
f prl
x X [-1 ,1 ] — ^ X' X [-1 ,1 ] U X’
and
f .......... pr2
X x [-1,1] X ’ x [-1,1] [-1,1]
respectively.
For a start, assume that for some positive e the map f^
is constant on each set x x [-c,e], x e X. In this case, fix positive
numbers 6 and r\, such that 6 £ min(e,1/2) and for any x E X the
derivative of the function t ^ f^ix^) is not less than n on the
intervals [-6,0) and (0,6]. [The existence of such 6 and n
results from the continuity of the functions X x [-1,0] ->1R and
X x [0,1] HR, given by the formula (x,t) h- 8 f 2 ( x , t ) 9 t |T =t (for
t = 0 one takes the left derivative for the first function and the
righ derivative for the second one), and the positivity of both
functions on X x 0. Since X is compact, these functions are bounded
from below by a positive constant on X x 0, and hence on X x [-6 ,6 ]
for some 6 > 0.] To proceed further, pick a C -function
a: [-6 ,6 ] I such that a(t) = 0 for |t| < 6/4 and a(t) = 1 for
Itl ^ 6/2. It is not hard to verify that the formula
defines a map g: X x [-1,1] -> X ' x [-1,1] with the desired properties.
In the general case, choose 0 < £ 1/2 such that the map
(jj : X -*■ X ’, defined as <J)t (x) = f^Xjt), is a dif feomorphism for all
ItI 4 z . [The existence of such an e.
is a consequence of: the
r
continuous dependence of on t in the C-topology, the fact that
^ is a dif feomorphism (recall that ab f : X x 0 - * X * x0 is a
diffeomorphism), and the fact that Diff (X,Xf) is open in C (X,X*)
212
C’
^1
ab F ab F 1
C 21
Cutting
f (y /1 ) , if t £ 0,
F (y ,t)
Yr (-1 )^ r + 1 f(y,-kt), if t ^ 0
¿k =0 * u k +1
is also of class Cr .
215
and this is valid if we interpret the sum as the M (r+1 )-th difference"
of the integral function k computed at k = -1 .
3 (LEMMA) ^r
Let X be a collared C ' -manifold. If
1 £ r < «, thenevery function in C (X,JR) extends to a function in
Cr (dopp X ,JR) .
( 0, if x € dopp X \ (X U ¿ ( d X x [0,1/r]
ip (x) = j<t>(x) , if x GX
F(£_ 1 (x)), if x € K d X x [0,1/r]),
dist (j 1 o f (x) ,g (cop (x) )) < Dist (j ’ (X’) , ]Rq ^ tubT ,P ') •
if x 6 Cl U ,
define a Cr-map h: dopp X -*■ 3Rq which extends j' ° f and satisfies
h(doppX) c tub^ , p '. Finally, the composite map
ab h pr ,
dopp X ----- * tub^ ,p 1 --- -— * X'
PROOF. The m a p p i n g
Pri
1 9X *,
> 9X' x I
1 •
P 2
max dist(j' o f(y),j' o g (y) ) < Dist (j '(9X ') ¿IR^ \ tub ,p'),
y e 3x T
and define $ as
[0 (g)] (k(y,t )) -
= k ' (prT , (j ' (f 1 (y,t)) + a (t/e) [j ' (f (y) ) - j ' (g (y) )]) ,f2 ( y, t) ) ,
if y e 3X, t € [0,e].
7. General Position
Technicalities
g st (cpt )_1
> Im cp^ -------- ► supp cp^
together define a Cr-map g: X -> X' [ (iv) shows that the sets
-1
supp ips D fg (supp t p p cover X and clearly the maps gg^_ are
compatible on the intersections of these sets]. g is the limit
(in C (X,X') of the sequence f^f^..., and (i)-(iii) show that
221
g e w n (n“ = 1 u±).
differ by less than 2at the points of <PS (KS n fs <K^.) ) • Obviously,
G is open and € G. Therefore, G fl ? 0 and, since Uk is
dense, G n ^ ^ 0• Choose to be any nonempty open set
readily seen that the objects l/^, f^, and n(k) satisfy conditions
(i)- (vi) for i = k.
dense set V in 1RQ 1 such that for each vector v E V, the map
Q1
X1 defined by f^(x^) + v is transverse to f .
ql ql 1
such that Im tp. = HR or 3R_ , cp1 (x1 ) - 0, Cl supp (p.. is
Ix ^ IX ^ I 'j
X ' ^ f _ (Cl supp i>9 .) , 14>' ■]~1 (Int Dq ) , for any given j = 1 ,. ..,1.
-1 ^1
Finally, cover by a finite number of the sets ip1x .(Int D ), say
maps h 1 :X ■+ X ' such that:h 1 (Cl supp ^ 1 ±) <= X ' \ f 2 (Cl supp ^ 2 j )
whenever g.] (Cl supp ^ <= X ' \f2 (Cl supp \p2 ^ ) , and h 1 (Cl supp ^ ± ) <=
c X ' \ (tjj'.) "*(Int ) whenever (Cl supp <= X ' \ (^^ ) 1 (Int Dq ) .
oo
That V is a neighborhood of g^ in C (X^,X') is plain.
loc(<K )tu_1
Im tp . '-JLJULtV-.jRq'
1lt
gv (x1 } =
(^4 )_ 1 ( ^ (h^_ 1 (x )) + a(<K (h^ - 1 ( x J D v ) ,
Jt t Jt ' '
-1 qi -1 ^9
and gv and f2 are transverse on
\p ' (D '), ip ! (D ) for r < t.
r
On the other hand, the definition of V shows that for v £ V,9 qTr
225
-i q-, _ ^ ^2
and f2 are transverse on (D ), ^ (D ), and hence one may
t t t
take h^ = g^ for any v £ U H V.
q' q2 q' q2
QR ¡JR_ /0), or GR_ /3R_ /0),with corresponding local coordinates
ip , (x) = 0, and in the case (ii) - by the same equations and the
inequality ip^ (x) < 0. Let us verify that for all cases (i)-(iv) , the
above equations and inequality satisfy the independence conditions
displayed in 1.2.12.
In case (i) , we have to show that ...... ,(|i . are
ip
q2 q
independent at the point x 1. This follows from the equality
q1
dim (1j_^^ +^Ker (d^ iJj^) = q f-q2 , which in turn follows from the trivial
inclusion dx ^f1 (nq=q^ + 1Ker(dx ^ j)) C Tangf (j£ ) X2 and the equality
definition of transversality).
In case (ii) , we have to showthat ilJ^.ip „ , .. .,ip . are
r1 rq2+1 'rq ’
are independent at x .
The existence of such a function is equivalent
i _
£■
4. Let A be a closed subset of a closed C -manifold X,
and let U be a neighborhood of A. _If 1 £ r £ °°, then thereis in
U a compact submanifold B of codimension 0 such that A c int B .
and the sets f(X \ U^) are closed. Finally, since X' is connected,
A' 7^0, and B' i X', we have A' = X' and B' = 0, i.e., f is
invertible.
r Q
4 (LEMMA). Suppose X is a compact C-submanifold of M 1,
1 £r £00, U is a neighborhood of X in , and X' is an open
subset of a closed Ca-manifold admitting a Ca-embedding in Euclidean
space. If f E Cr (U,X’), then every neighborhood in Cr (X,X’) ojf
the restriction f|w contains the restriction of some map belonging to
■
------ — ■
—■ x
ca (U,X*) .
Cr (dT (Xq ,p/2) /T (xQ)) • Hence fjTub (p/2) has in Cr (TubT (p/2),G1(q,n))
a neighborhood 1/ such that for each g £ 1/ one has g(d (xQ ,p/2)) c: V
r ^
and tt o [ab g : d^(xQ,p/2) V] is a C -embedding whose image contains
0. Since G'(q,n) is an open subset of a closed Ca-manifold (see
2.2.10), Lemma 4 guarantees the existence of a Ca-map
h: tub^p -+ G' (q,n) whose restriction to Tub^ (p/2) belongs to (J f11/.
i -1
Set Y = th|Tuk (p/2)-^ (G (q,n) ) . By virtue of Theorem 8.2, Y is a
a point.
t ' of j', and a neat tube Tub^,pf. Let 1/ c Cr (B,X') be the set
of all maps g such that
max dist (j ' ° f (x) ,j ' (g (x) )) < Dist (j '(X ') , IRq ' v tuh>T ,p ') .
x£B
f(x), if x £ X \ B.
Y _JL 3X ax'
is a C -diffeomorphism.
Auxiliary Manifolds
>27
2. Suppose X is a closed n-dimensional C' -manifold with
1 < r £ a, and let j: X IRqbe anembedding of class Cr . Also,
let m be a positive integer such that 0 < m < q. We shall need two
constructions.
The first construction: we denote by Aux^ or, more
specifically by Aux^(j;m), that subset of Tang X x G(q,m) consisting
of the pairs (u,y) such that <dj (u) ,dj (u)> = 1 and dj (u) G y.
Further, let aux1 : Aux1 ^ G ’(q,m) be the map defined by
aux1 (u,y) = j(pr(u)) + y, where pr = [pr: Tang X + X]. Aux1 is a
234
>if
3 (LEMMA). Let X be a closed n-dimensional C ' -manifold,
q 27
1 £ r £ 00, and let j : X ]R^ be an embedding of class C together
j7
with a C -transversalization
------------------------------------------------------ t : X G(q,q-n) and a neat tube Tub T p.
is a Cr-embedding and hg (tubTP) => TubT (p/2) (see 9.5). For g € 1/,
let i denote the composition
- 1
h • (ab h ) 1 . pr
x — -J--> TubT (p/2) ----- 2--- ► h~ '(TubT (p/2) ) TubTP ---^ X
4. Suppose X
is a compact n-dimensional C' -manifold,
^r
1 £ r £ 00, and X1 is a closed n-dimensional C -manifold. Then for
n' ^ 2n, Immr (X,X') is dense in Cr (X,X*), and for n* ^ 2n +1 ,
Embr (X,X ') is dense in Cr (X,X').
t ' , (x ' ) = g ' (x ') - [ J ’ o <5 ' (g ' ) ] (x ' ) is a transversalization of the
embedding J' ° 1 (g ’) ;
(iii) some neat tube of this transversalization contains
TubT , (p'/2).
Now consider the mapping V: W -+ Cm ( X , X ' ) given by
[T '(g')](x) = pr , (J(x)). We see that Ÿ’ is continuous and
V
236
r (x') = f. Hence, (f) 1 (U) is open in C°° (X ',G ’(q '+q, q '+q-n 1) )
- *]
and nonempty. This fact together with Theorem 7.7 show that (H") (U)
contains a map h transverse to each of the maps
We shall presently show that r(h) is an immersion for n' 52n, and
a differentiable embedding for n' ^ 2n+1. This will complete the proof,
since ¥ 1 (h) £ U .
The inequality n' ^ 2n is equivalent to
dim X ' + dim Aux1 (J;q'+q-n') < dim G 1 (q '+q,q 1+q-n ') ;
hence for n' ^ 2n the fact that aux^ and h are transverse means
that h(X') does not intersect Imaux^ , i.e., none of the planes
h(x'), x' £ X', contains a line tangent to J(X). The latter, in
turn, means that the differential d (pr ,|T #v x) of the restriction
y Th '
Pr i I-r/v\ is a monomorphism for any point y £ J (X) . Thus pr i|T/vx/
|J(x ) ^h I
and also ¥ 1 (h), are immersions.
Similarly, the inequality n' > 2n+1 is equivalent to
d i m X 1 + dim Aux^ (J; q '+q-n ') < dim G ' (q '+q ,q 1+q-n 1) ;
hence for n* ^ 2n+1 the fact that aux^ and h are transverse means
that h(X') does not intersect Im aux^ , i.e., none of the planes
h(x'), x' £ X 1, intersects J(X) at more than one point. This says
that the restriction pr , |T ,v v is injective. We see at once that
Th I
y ' (h) is also an injective map, and since it is an immersion, ¥ 1 (h)
is a differentiable embedding (see 1.5.4).
>£•
5. Every compact n-dimensional C' -manifold with 1 £ r £
can be Cr-immersed in IR2n and Cr-embedded in 3R2n + 1.
11 . Exercises
>37
1. Show that for any C -manifolds X and X', with
0 s? r £ , the space Cr (X,X') has a countable base.
2. Let X
be a compact C ^ -manifold, 1 £ r£ and let
^r
X' be an arbitrary C -manifold. Show that the set
Submr (X,X') 0(X,X1) is open in C^(X,X').
CO CO
f- £ C _ (X„,X1). Show that the subset of C.(X1,XI) consisting of all
2 3 2. ' _ co d 1
mapstransverse to f2 is dense in Cg(X^,X').
1. Morse Functions
00
2. A compact C -manifold X is called a cobordism if its
boundary 3X is the disjoint union of two parts, 3^X and 3^X, each
consisting of whole components of 3X.Those two parts are termed the
beginning and the end of the cobordism X.Each of them may be empty?
when both are empty, X is closed. In general, given a compact
C°°-manifold, one can transform it into a cobordism in 21 ways, where
1 is the number of components of 3X. Among these cobordisms, there is
one without beginning (3QX = 0, ^ X = 3X) and one without end
(3QX = 3X, 31X = 0) .
Two cobordisms, X and X', are said to be diffeomorphic if
oo
there is a diffeomorphism (and hence a C -diffeomorphism) f: X X1
such that f(3^X) = 9q X' and f (3^X) = 3^X!.
240
cp ► im cp -- > 3R ,
supp cp --
(t^,...,^) ^ , if x £ 9^X ;
point of f ;
(t
I n
f (x) - t 2
1
- .. . - t2
k
+
k +1
+ ... + t2 ,
n' —
if X
To prove the first three cases we need only remark that the
function X 3R, defined by y b» f (y) - x for x £ int X U 9^X
An Existence Theorem
oo y\
6 (LEMMA). Given any C -function f: D + 1R, there exists
an open dense subset A of lRn such that for a £ A the function
Dn 1R defined by
4.7.4 implies that itis also dense in 3Rn . Moreover, i’t is evident
that if x£ Dn is a critical point of the function (1), then
[grad f] (x) = a, andthe matrix of the second-order partial derivatives
of (1) at x is precisely the matrix of the differential d^gradf
242
Pick charts (p ,... ,cp £ Atl X such that Imip = ... = imcp =3Rn ,
i s I s
n = dim X , and the sets (p^ (Int Dn ) ,.. .,(p~1 (Int Dn ) cover
function on X.
Let 9q = 9 and assume that for some k ^ 1 functions
enjoying the required properties are already constructed for i < k.
For each point a £ 3Rn , the formula
Standard Cobordisms
O, 1 4-2 ,2 2 , 2. ,
8 { T6 ~ + \ + 1 + ••• + tn )) •
\ /
V ( S ° * V 1)
[V\y(S0* \ n i @ 2 ’))\*/
A5 As B5 B8 A? A8 B^
Fig. 5 (n = 2, k = 1)
245
(v,t) * t
and
and as
246
2,1/2
lp((t1 V ' (tk +1 V ’ " (t1 (T6 + fck + 1 + ••• + tn )
+ t2)1/2 t t )
V T 6 + fck + 1 + V ,fc k + 1 ' • • • ' V
on the part
,^k-1 ^n-k.
cp (S x d )
V x I ir, (v,t) t ,
n— 1
If k = n, then cp embeds S into V, the image of
this embedding being one of the components of V (see 1.5.1). Again,
E(n,k) is the ball ^-Dn and el is a homeomorphism of empty sets.
Here the similarity dilatation of the ball E(n,k) transforms El(V,cp)
into the sum ([V\cp(Sn ^)] x i)| |Dn , with
and
31{ ([V ^ (p(Sn 1)] X I) I |pn } = ini ([V ^ cp(Sn 1 )] X 1) ,
D m, (t1 tn ) "
_i
coincides with cp . In order to accomplish this, let us fix:
OO
a collaring k: 9X * I -*■ X; a C -diffeomorphism f: X' + 9QX' x I
such that f (x 1) = (x',0) for all x' € 9QX'; and an increasing
C°°-function a: I + I, such that a(t) = 1/2 + t/3 for t g 3/4 and
a(t) = t for t ^ 7/8. Further, using the function 6: I I,
8 (t) = (1 - t)/2, set
j (x ) = x , if xex^kO^xtoj)),
and
2 2 2
t, + ... + t ^ 4c ,
k+1 n '
and
249
TE
16 ‘ - ‘Î
1 - • '• - k * * 1k+1, * • ■• * t2
n * h16 ■
for some c > 0. Let
_ t2 - - t2 + t2 + + t2)
2 16 r 1 ••• rk k +1 ••• V
c
2 2
£ G
[Here we use the inequalities -r-p < f (x) <1 - ^ , which are immediate
lb lb
consequences of the inclusion A c Imip .]
CO
The C -structure of Y is fixed by the atlas consisting of
oo
arbitrary atlases of the C -manifolds
Sk 1 x 2 Dn k x 0.
and
^k-1 _n-k „
S x 2D x 1
The functions
250
-1 2 2 -1
f ([f (x) - fg,f(x) + ^g]) ^ cp (A B)
R 2
Sk 1 X Int (2Dn k ) X I m,
in ^k-1 _ , ,n n-k. in
►S x Int (2D ) x i ------ > y
k—1 n —k —1
and its compression s S x d g (1/2). To complete the proof
of the theorem we shall presently verify that the cobordism X is
dif feomorphic to El (g ^ (1/2) ,\j;) .
As a preliminary step, we find a number 6 , 0 < 6 < 1/2,
OO
and a C -diffeomorphism
H: g” 1 ( j ) * - « 4 + + 6]) '
such that
H ((u,v,-l) ,t) = (u ,v ,t )
By Theorem 4.5.8, one can find n, 0 < n < 1/2, such that
OO —1 1 1 “1 1
there is a C -submersion it: g ((- - n,^- + n) ) + g (j) which is the
—1 1 oo n _V
identity map on g (-j) • Fix a C -function a : IR 3R, equal to 1
TT (y) , if y £ Sk 1 X int(2Dn k ) X I,
p (y)
tt (u ,v ,-1+ (t - 1) (1 - a (v) ) ) , if y = (u,v,t) ,
Further, define
G: g n^ ^ g ^ x ^ + n)
- 1 1
Finally, to show that the cobordisms X and El (g i-^) /^)
are diffeomorphic, cut each of them into three cobordisms: X into the
2 2 2
cobordisms f ^([0,f(x) - *^|-] ) , f_ 1 ([f(x) -^|-,f(x) + ]) , and
Theorem 6, the first and the third cobordisms in each of these triples
are trivial, and hence the second cobordism is diffeomorphic to the
entire cobordism.Therefore, it is enough to exhibit a diffeomorphism
2 2
mo"1 ([-1 - 6,1 +6]) + f_1 ([f (x) - ^§-/f (x) + £§-] ) .
A t t a c h i n g Handles
Dn by a diffeomorphism of Sn ^ .
AHomotopy Corollary
PROOF. The discussion in 1.2 implies that one may assume that
the given manifold is a cobordism with an empty beginning. Therefore,
all we have to show is that if an n-dimensional cobordism is homotopy
equivalent to a finite cellular space of dimension £ n, then it
retains this property after we attach to it an arbitrary handle; see 10.
But from Lemma 12 it follows that attaching a handle of index k to a
k
cobordism X has the same homotopy effect as attaching D to X by
k- ^
some embedding f: S X. Now replace X by a finite cellular space
Y of dimension £ n with the same homotopy type, replace the map f
by its composition with a homotopy equivalence X Y, and subsequently
k-1
replace this composition by a homotopic cellular map g: S Y (see
2.3.2.4). By Theorem 1.3.7.8, the cobordism which results by attaching
a handle of index k to X is homotopy equivalent to the space
k k
Y Ug D ; according to 2.1.5.5, Y D is a finite cellular space of
dimension £ n.
Spherical Modifications
oo
14. Let V be a closed n-dimensional C -manifold, and let
k— 1 n—k + 1 00
tp: S x D ->• V b e a C -embedding. Fix arbitrary collars on
Y ^ cp(Sk_1 x IntDn_k + ^) and Dk x Sn k , and then glue these manifolds
k—1 n —k k —1 n —k
by the dif feomorphism ab tp : S xs +tp(S xs ) of the
254
boundary of the second onto the boundary of the first. We say that the
glued manifold is obtained from V by a spherical modification along
the embedding cp.* The number k is the index of the modification.
PROOF. Since 9^X ' = 9 ^El (9^X ,cp) , we actually claim that
9 ^El (9^X ,cp)is obtained from 9^X by a spherical modification along
cp. Recall that El(9^X,cp) is the result of glueing the spaces
3. Two-dimensional Manifolds
Model Surfaces
1 1
is diffeomorphic to the cylinder S x D ; we call the latter a handle
(do not confuse with the "handles" in Subsection 2 !).
3
A Möbius strip is a submanifold of IR produced by the motion
1
of a segment of length 1 whose middle glides along a circle S m
such a manner that the segment remains normal to the circle and turns
uniformly through a total angle (see Fig. 6). Every Möbius strip
tt
3
is a nonorientable compact submanifold of IR with boundary diffeo-
1
morphi c to S .
Fig. 9
Auxiliary Propositions
F ig . 10
259
Cellular Decompositions
of the Closed Model Surfaces
1
In the case of a sphere with h crosscaps, we represent S
as the contour of a regular polygon with first vertex ort^ and 2h
edges, which are arranged successively as
4. Exercises
5.
Consider the manifold obtained by glueing two copies of
1 2 1 1
the solid torus S x d by a diffeomorphism of its boundary S x S ,
given by the formula (z^z^) (z^ Z^/Z^ z^) , where a,b,c,d are
1. General Definitions
tl E ' tl K
Pr K ' (1 )
bs E bs E,
trivial, but not standard trivial bundle, these homeomorphisms are not
canonical any longer. If the base of a locally trivial bundle is
connected, then its fibers are also mutually homeomorphic; indeed, the
set of the points of the base having fibers homeomorphic to a given
fiber is open, and the sets of this type form a partition of the base
(see 1.3.3.5). On the other hand, the example of the locally trivial
bundle ( (B x F)J |_(B' x F'),pr^J |_pr -j ^BJ |
_B ') * where B, F, B ' , and
F* are arbitrary topological spaces, demonstrates that in a locally
trivial bundle the fibers over points situated in different components
of the base are not necessarily homeomorphic. Moreover, we see that
there are locally trivial bundles which are not trivial.
A nontrivial, locally trivial bundle may have a connected
base; see 5 and 6.
Coverings
3. Serre Bundles
face of Ir whose points have the last coordinate equal to zero; see
1.2.5.7.)
The requirement that pr E, ° f = f appearing in Serre's
condition is fundamental in the theory of bundles, and is encountered
also when f and f are defined on spaces other than cubes. If two
maps, f~: X tl K and f: X -> bs £ , satisfy this last requirement,
We remark that the first bundle has both empty and nonempty
fibers, while the second has a single connected fiber, the others being
not connected. As we shall see later (see 5.4.3.6), such features of
a bundle are not compatible with Serre 1s condition vfrien the base is connected.
pr £ ° f . -i = f I,, and f . „
1-1 Iwi - 1 1-1 ^ - 1 = f 0' Set gi = f |Qi and
n “1
is continuous, covers f, and equals f^ on I
This example shows also that in a Serre bundle with connected
base there can be nonhomeomorphic fibers. Actually, there are Serre
bundles with connected base and in which some fibers are not even
homotopy equivalent, being instead equivalent in a certain weaker sense
(see 5.4.4.3 and 5.4.3.6) .
272
f (cha (x)), if t = 0,
e
^0,e(x't)
(cha^ (x) ,t) , if x E Sr ^
r 27” 1
defines a continuous map <J>n : (D x 0) U (S x I) -*tl £ .
u ,e
Obviously, the pairs (Dr x I,(Dr x 0) U (Sr ^ x i) ) and (ir + \ i r )
are homeomorphic , and pr ^ o ^ j
^ = ^e (Drx0)u (S17-^xi)# Consequently,
r
there is a continuous map (J> : D x I -> tl £ covering $ and
^ ~ ^ r—1
extending <\>n . Since <f> (x,t) =F (cha (x),t) for all x ES ,
u ,e e r— i e
the maps <j>~ corresponding to all r-cells from X \ A together with
F~_^ yield a continuous map F~: (A U ske^X) x I + tl £ , and it is
Let £
be the given bundle, and let s be a path joining
-1 -1
two given points of bs £ . Set Fq = pr £ (s (0 )) and F^ = pr E, (s (1 ))
Now consider two homotopies : F^ x I -> bs £ and : F^ x i bs £
J~(x,2t), if t £ 1/2,
j(x,t) =
J1 (fQ (x) ,2t-1) , if t * 1/2,
and
Pr i
tl C
Z — — » tl ad K in -> tl Ç x C (I,bs Ç) pr.
C (I,bs Ç) ,
5. Exercises
§2. A DIGRESSION:
1. Topological Groups
continuity of the map (g,h) h- gh at the point (g0 ,hQ) means that
for every neighborhood W of the point gQh 0 one can find neighborhoods
U and V of gQ and hQ such that UV c w. Similarly, the continuity
-1
of the map (g,h) g h means that for every neighborhood W of the
point g^ h^ there are neighborhoods U and V of g^ and h^ such
that U_1V <= w.
Clearly, every group becomes a topological group if it is
equipped with the discrete topology.
Direct Products
simply S^. However, this is not a normal subgroup. The coset space
S'Vs^ is canonically homeomorphic to S^:
this canonical homeomorphism
3 2
is provided by the injective factor of the Hopf map S S (obviously,
282
3 1 3
zer(S ,S ) = zer (S s 2) ) .
2. Groups of Homeomorphisms
Groups of Diffeomorphisms
3. Actions
Continuous Actions
Examples
are the injective factors of the maps GL (n, IR) -> V'(n,k),
GL(n,(C) + (CV' (n,k) , and GL(n, H) 3HV’ (n,k) , defined in Subsection
3.2.1 (see 3.2.1.7, 3.2.1.8, and 3.2.1.9). The isotropy subgroup of
GL+ (n, IR) at the point [ (x^,...,x^) h- (0 , ...,0 ,x ,. ..,x^)] E V ’(n,k)
is GL+ (n-k , IR) .
All the right actions are free. The corresponding orbit
spaces, V ’ (n,k)/GL (k, IR) , V 1 (n ,k) /GL+ (k , IR) , (CV 1 (n,k) /GL (k ,(C) , and
]HV1 (n,k)/GL(k, ih) , are canonically homeomorphic to the Grassman
manifolds G(n,k),G+ (n,k), £CG(n,k), and 3HG(n,k); the corresponding
canonical homeomorphisms are the injective factors of the maps
V 1 (n,k ) +G(n,k), V 1 (n ,k) G (n,k) , (CV!(n,k)->(CG(n,k),and
H V 1 (n,k) -*IHG(n,k), defined in Subsection 3.2.2 (see 3.2.2.3, 3.2.2.7,
and 3 .2 .2 .8) .
14. GL (n, IR) and its subgroups GL+ (n, IR) , 0(n), and
S0(n) obviously act continuously from the left on the Grassman
manifolds G(n,k), G (n,k). Similarly, GL(n,(C) and its subgroups
U(n) and SU(n) act continuously from the left on (EG(n,k), while
GL(n, IH) and Sp(n) act continuously from the left on IHG(n,k).
For k odd,the actions of 0(n) and SO(n) on G + (n,k) are effective.
The noneffectiveness kernels of the actions GL (n, IR) x G+ (n,k) -> G + (n,k)
and GL+ (n,k) x G+ (n,k) G
+ (n,k) for k odd consist of scalar
matrices with positive diagonal elements. If we exclude the trivial
cases k = 0 and k = n, the noneffectivennes kernels of the remaining
actions consists of all scalar matrices contained in the corresponding
group. The only intransitive actions are GL(n, IR) x G+ (n,0) G+ (n,0) ,
GL+ (n, IR) x G+ (n,0) G + (n,0), 0(n) x G + (n,0) G + (n,0),*
SO (n) x G (n,0) -+G + (n,0), GL+ (n, IR) x G + (n,n) G + (n,n), and
SO (n) x g
+
(n,n) -+G+ (n,n) .Take the plane x =I 0,...,x n ,
Js. = 0
294
2iti k i /m 2iTikl ~ / m
(k, (z1 ,z2 , ...) ) h- (z 1e ,z2e 1 ,...)
OO
[We identify JR^j with 1R^ via the map shi: JR^ ->■1R^; see 3.2.3.1.]
3 3 3
The noneffectiveness kernel of the action S x jr ]R is obviously
S°, and now it is clear that the effective action of the factor group
S^/S^ = IRP^ on ]R^ becomes the standard action of S0(3) on IR^
(see 11) under the canonical identification of the spaces IRP^ and
S0(3) (see 3.2.3.1) .
4. Exercises
1. Show that for any smooth manifold X the first and the
second topologies on Top X coincide.
3 3
space S /H, where H is the subgroup of S consisting of the
quaternions ±ort^, i o r t ^ ±ort^, ±ort^.
2 2
8. Consider the action Z>2 x TP -*(CP , where the nonzero
element of takes each point (z :z :z )into (z : z n : z 0 ) . Show
^ 2 i z J a 1 z 3
that the orbit space (EP / Z 2 is homeomorphic to S .
2. Steenrod Bundles
where <J> and \fj are some maps from bs £ into the structure group G.
Moreover, if we use the topologizing homeomorphism bs£ x g +
-1
then MH (f) and MH (f) become the maps bs ^ x G ^ b s ^ x g given
by
Principal Bundles
homeomorphisms.
11. This construction of the free right action of G on the
total space of a principal bundle with structure group G can be
partially reversed. Assume that the topological group G acts
continuously and freely from the right on the topological space X, and
consider the bundle (X,pr,X/G).Its fibers (orbits) carry natural
_ 1
G-structures: the marked homeomorphisms G pr (b) (b € X/G) are
_ 1
given by g xg, x £ pr (b). Since to distinct pointsx correspond
distinct homeomorphisms g xg, we obtain also an invertible map of X
onto the set of marked homeomorphisms, and thus we get a topology on the
last set. Therefore, (X,pr,X/G) is a G-bundle.
To explain why we called this last construction a partial
inversion of the original one, apply it now to the right action
tl^ x G ->■ tl £ described in 10; the resulting bundle is exactly £.
More precisely, the injective factor of the projection pr C maps
tl £/G onto bs C , and together with id(tl^) forms a G-isomorphism
(tU,pr,tl £/G) +
3. Associated Bundles
(ii) The invertible map MH(£) MH (asso (^,F 1)) that we used
to topologize MH(asso (£,F 1)) , is a G-map with respect to the right
canonical actions of G on MH(£) and MH(asso(£, F 1)). As a corollary,
we may state that, given an arbitrary effective G-space F", the product
of the above invertible G-map with id F" is a G-map MH(£) x f"
-> MH (asso (^ ,F1) ) x F", where G acts from the right on MH(£) x F"
_ -i
and MH(asso(£,F')) x F" by g(a,y) = (ag,g 'y). The resulting homeo-
morphism (MH(£) x F")/G -»• (MH (asso (£ ,F ') ) x F")/G, together with
id bs £ , define an F"-equivalence asso(£,F") asso (asso (£ ,F') ,F" ) .
Therefore, the bundles asso(asso(C,F'),F") and asso(£,F") are
canonically F"-equivalent.
and
i -1 g-1
► (pr f •£’) (6(F) ) -2--- .
tlh(x) = tl adf(Bv
x (gv
x (a"1
x (x) )) ) ,
(aX x SX ,g
X ) £ s ° pr £ (x) .
Let us remark that when bs £ ' = bs £ and f = id bs £ , an
F-map h: £ -> £ 1 with bs h = f is simply an F-equivalence (see 2.8).
Therefore, when £ and £' have the same base, the above construction
yields a one-to-one correspondence between the F-equivalences £ -> £ 1
and the sections of the bundle Fibr(£,£';id bs £) .
4. Ehresmann-Feldbau Bundles
Y x X —f X ld > X x G -> X (1 )
F-equivalence).
In fact, let £ and be Steenrod F-bundles. We need only
prove the continuity of the map M H (f ) corresponding to the given
W-F-map f: £ + and we may assume that £ and are standard
trivial bundles. In this case, tl £ = bs £ x F, tl £ 1 = bs £ 1 x F,
and tl f is given by (b,y) ^ (bs f (b) ,<f>(b)y ) , where <p is some map
of bs£ into the structure group G. At the same time,if we use the
canonical homeomorphisms bs £ x g -> MH(£) and bs £ 'x G -> MH ( ^' )
(see 2.2), then MH(f) becomes the map bs ^ xg bs £ ' x g ,
(b,g) (bs f (b) ,<j>(b) g) . The continuity of tl f implies the
continuity of the composition pr^ o tlf : b s £ x p - ^ F , which equals
We remark that the last assertion is also true for all coverings in the
broad sense with connected bases.
5. Exercises
h 1 : n1 = (B1 x F fpr1/B1) + I |B
and 1
h 2 : n 2 = (B 2 x F , p r 1 , B 2 ) -► £ | B »
f * [abf' : n2 |
over W = In .
Nn
4- Let and £ be Steenrod bundles with common standard
fiber F and common base B, and let A be a retract of B. If £
and E, ^ are F-trivial, then for every F-equivalence h: ^ ^2 |A
(Cl e. xI) n (Cl e„ x i), and that for each cell e € cell. „X ^ cell. „A,
1 Z 1+1 1+1
tlh C and tlh._L agree over (Cle x i) n C.. Since the sets C. and
Cle x I, e £ celli+1X \ cell^ + 1A, constitute a fundamental cover of
Ci+1 ' WB use 3 *2 *^ to conclude that h^ and hg , e £ ce l l ^ X \cell.+1A,
form together an F-map U -*• L | . We take this map for h. „
l|Li+1 ^ ,Ci+1 1+1
and note that it obviously extends h ^ ; moreover, by 3.2.7, h~ 1-| is
315
also an F-map.
To check the rest, i.e., that the sequence {h^ : ^ U ^ ^Ic ^
Ii 'i
converges to an F~equivalence £ it is enough to remark that
the sets constitute a fundamental cover of X x I, and then apply
Theorem 3.2.7 to the sequences {h^} and { h ^ }.
f ’: Stee(B,F) +Stee(B',F).
i
If B'is a cellular space, Theorem 5 shows that f"depends only on
the homotopy class of f. In particular, if B and B' are both
i
cellular andf is a homotopy equivalence, then f ‘ is invertible.
The extension of the structure group, which transforms the
-space F into the effective G-space F, defines a mapping
asso
Stee (B ,F) ------ > Stee(B,F'
f*
f•
2. Universal Bundles
7i(idB,g) n. induz(B,£)
7T(B ,bs O
induz(B,£) asso
Stee(B,F)
We remark also that both (i 1) and (ii1) (and hence (i) and
(ii)) are consequences of the following condition: given an arbitrary
F-bundle £ with cellular base and an arbitrary subspace A of bs £ ,
every F-map 5 |A * £ extends to an F-map £ -* £. To see that this
condition implies (i1), it suffices to take A = 0. To see that it
implies(ii'), take the F-bundle £ x(I,id 1,1) , thesubspace
A = (bsKx o) U (bs K x 1) of bs(5 x (I,id 1,1) )= bs 5 x I, and
take as the F-map that must be extended
g: £ x (l,idl,l) |( b s £ x o) U (b s £ x 1) ^ ^ '
with
bs g (b,0) = bs hQ (b) , bsg(b,1) = b s h <l (b) [b e bs C ] ,
invertible.
4. If the F-bundles £ and C' are universal and bs 5 and
bs £ ' are cellular spaces, then bs g : bs £ bs £ ' is a homotopy
equivalence for any F-map g : £ -* C 1•
Classifying Spaces
From Theorem 3 it follows that every space which has the same
homotopy type as a classifying space of G is itself a classifying
space of G. Moreover, it results from Theorem 4 that any two cellular
classifying spaces of G have the same homotopy type.
k-Universal Bundles
i i
the bundles and f *£ are F-equivalent, are homotopic. Equivalent
conditions are: a 1) given any Steenrod F-bundle £ with cellular base
of dimension ^ k, there is an F-map £ £ and b') for any Steenrod
F-bundle £ with cellular base of dimension £ k-1 and any F-maps
g0'g l : ^ ^' bs g0 and bs g 1are homotoP ic-
In all these formulations k is a positive integer, and the
universal bundles are sometimes termed ^-universal. Every k-universal
bundle is obviously 1-universal for 1 £ k. Moreover, Theorems 2.3.2.4
and 2.3.2.5 show that the restriction of a k-universal bundle with
cellular base to a subspace of the base which contains its 1-skeleton
is 1-universal, for any 1 £ k.
Theorems 3 and 5 have immediate analogs for k-universal
bundles: every bundle induced from a k-universal bundle by a homotopy
equivalence is k-universal; a bundle aasociated with a k-universal bundle
is itself k-universal.
Mi G is Locally Trivial
Mi G is Universal
4. Mi G is universal.
,r
f = h o [ab cha : S a U ske bs Ç ] : cha *Ç ^ MiG
e r e r e
sr
Further, note that for any cells e„,e0 E cell ,bs r \ cell „A,
1 2 r+1 r+1
tl h and tl h agree over Cl e fl Cl e0 , and that for any cell
1 2
e E cell^^bs £ \ cell^^A, tl h^ and tl h^ agree over
A Promise
* ( B , b s Cl > ------ . , ( B , b s t ]
5. Exercises
(j / ft) ^ j ° ft [j £ Embr (X, 3Rq ) , <J> G Diffrx] , and the limit right
action of Diff X on
lim (Embr (X , 3Rq ) ,ab Cr (id X ,in) : Embr (X,TOq ) -> Embr (X, lRq+1)) .
Show that (lim Embr (X,lRq ) ,pr, [lim Embr (X,Kq )]/DiffrX) is a
universal Diffrx-bundle, while (Embr (X, ]Rq ),pr,Embr (X, lRq )/DiffrX)
is a ( q - 2 n - 1 )-universal DiffrX-bundle, for any q 5 2n+1.
324
1. General Definitions
]R x tl £ tl K , (X ,x ) b Xx,
and
325
are continuous.
?|u" Define a map, h': (U x nRn ,pr^ ,U) -> £, linear on fibers, by
Maps
Vector Fields
2. Constructions
Subbundles
pr^ denotes the projection of the fiber b x ]Rn onto its subspace
g(b)]. Therefore, is a k-dimensional real vector bundle. We call
it the subbundle of t associated with g.
7.
For real vector or Euclidean bundles, the orientation of
-1 -1 -1 -1
each of the three spaces, pr £ (b) , pr n (b) , pr £ (b)/pr n (b) ,
is uniquely determined by the orientations of the other two (see
3.1.3.10). Consequently, the orientability of two out of the three
bundles, £, r\, £/n/ implies the orientability of the third, and
given orientations of two of them canonically determine the orientation
of the third.
Sums
Complexification
2. Set
The first of these maps (as the canonical map G+ (m,n) -> G(m,n)
with m < 00) is a two-sheeted covering projection; the second is the
composition of the first and the third; and all these maps, except the
first, are embeddings.
= 03(1) + ... + 03 (n ) •
We have to produce a characteristic map, cha^^: + G(°°,n).
Fix u) and, for points u,v £ sw(n)+n-1 such fchat u + v ^ o , denote
by r(u,v) e SO(w (n)+n) the orthogonal transformation which takes u
into v and keeps fixed all the vectors of n which are ortho
gonal to u and v. Further, let denote the co(i) -dimensional
V [ r ( O r t 0 ) ( 1 ) + 1 ' U 1 ) ] ( U2 ) ...........
(1 )
[r (Ort /-V.-/U ) o... O r(ort , . 17U ! )](u )
U)(1)+1 1 co(n-1 )+n-1 n-1 n
(x1 '• ••,XU)(1 ) ,0,XU)(1)+1 ' •• •,X03(2) ,0'** •'°'xw(i-1 )+1 '•• •'
x (1 - x V /2>
Xw(i), n ¿1 x j' '•
(see 2) are cellular. The image of the first contains skenG (°°,n+q) ,
the image of the second - ske G+ (cc,n+q), and the image of the third -
r *
g = f oad[abcha : S -*• A U ske bs g] : cha *f Gra GL (n, TR)
^e r e r e
S
n *
extends to a GLIR -map h ^ : °ha^£ -+ Gra GL(n,]R) , and it is clear that
tlh is constant on the elements of the partition zerftlad cha ).
^ r+1 r+1 e
Applying Theorem 3.2.6 (with B = D and p = [ab cha^ : D + Cle] ),
are obviously V' (m,n) , V 1(m,n) , (CV* (m,n) , and V(m,n) , V(m,n) ,
(CV(m,n). The corresponding projections are the maps described in
3.2. 2.3 and 3.2.2.7:
The Bundles
asso (Gra 0(1) ,0(1) ) and as so (Gra U (1 ) ,U (1 ))
The meaning of the left-hand side was explained in 4.3.2, while in the
right-hand side the elements g. of 0(1) or U(1) are thought of as
numbers (the following inclusions are used: V^l) cnR°°, (CV(°°,1) c (E°°,
0(1) = S° c ]R, and U (1) = S1 c (C) .
(4). Indeed,the bundles Gra GL (n,IR) and GraO(n) have the same
base, G(°°,n), and the same is true for GraGL+ (n,]R) and Gra SO (n) ,
with the base G (°°,n), and for Gra GL(n;(C) and GraU(n) , with the
base (EG (°°,n) . It is obvious that in all three cases the identity map
of the base is classifying. Therefore, the mappings
4. For the inclusions (7) and (8) there are also obvious
classifying maps: for both left inclusions, such a map is the canonical
embedding G(°°,n-s) G(°°,n), for both middle inclusions - the canonical
embedding G+ (°°,n-s) -+ G (<»,n) , and for both last inclusions - the
canonical embedding (EG(°°,n-s) -> (EG(°°,n) (see 3.2). Identifying
G(°°,n-s), G+ (°°,n-s), and (EG(°°,n-s) with their images under these
embeddings, and using 3.5, we can write:
are invertible for any cellular B with dimB $ n-s, and surjective
for any cellular B with dimB = n-s. In exactly the same manner the
inclusion G+ (~,n-s) => sken_sG+ (°°,n) leads to the invertibility
(surjectivity) of the mappings
5. Exercises
^r l pr£
tl £ x . .. x tl £ --- -— > tl £ ■ ^ > bs £ .
tl £ X ... X tl £ !— » tl £ ■■Pr 5 , bs £ .
1. Fundamental Concepts
i r
induced, then corr (p (bs cp)‘£ is also a C -map.
27
5. Let F
be a C -manifold with r ^ 1. According to 2.3.10,
r r
F is an effective Diff F-space, and thus every C -bundle £ whose
fibers are Cr-diffeomorphic to F is a W-F-bundle (see 3.1.3 and 3.2.1).
Moreover, Ç is clearly locally W-F-trivial, i.e., it is a Ehresmann-
-Feldbau bundle. However, the procedure that enabled us in Subsection
3.4 to turn Ehresmann-Feldbau bundles into Steenrod bundles does not
work here: as we already had the ocassion to note (see 3.5.2), when
27
dim F > 0, the natural action of Diff F on F is not topologically
effective. Nevertheless, the set MH(Ç) carries a natural topology,
transferred from Cr (F,tlÇ) with the aid of the injective mapping
MH(£) ** Cr (F,tl Ç ) which takes each dif feomorphism a € MH(Ç) into
the map [in: a(F) -> tl fj ° a. It is clear that with this topology on
MH(Ç), £ becomes a Steenrod F-bundle.
where d^ (x,p) and (x,p) are the ball, and respectivelythe sphere,
with center j(x) and radius p in the plane j(x) + t ( x ) .
Equivalently,
TUV = V b s £ (y x Tubi p)
pr z 1 (y)
and
tub p = U (y x tub P ).
T yebs £ ■* t
pr Г 1 (y)
3. Letandr £ let
«>, £ and £ 1 be C -bundles with
closed total spaces tl£ and tl £ f, and closed bases bs £ and bs £ '.
Then Cr (£,£') is dense in Cs (£,£') for any s < r. The same holds
true for r = a, provided that tl£, tl £1, bs £ , and bs £ 1 admit
Ca-embeddings in Euclidean spaces.
Pick a Cr-embedding j 1: tl £ 1 +
PPOOF. a £ 1-transver-
x*
salization t 1 of j 1 of class С , and a neat tube Tub t *
,p'.
,s s 1
Let U denote the subset of C (tl£,tl£!) x c (bs£,bs£*) consisting
of all the pairs (F: tl £ + tl £ 1, f : bs £ + bs £ 1) such that
(f (pr £ (x) ) ,j 1 (F (x) )) £ tubT ,p * for all x £ tl £ . Then U is open
352
^r
If two C -bundles with closed total manifolds are
1 r
C -equivalent and r £ then they are C -equivalent. The same holds
true for r = a, provided that the total manifolds admit Ca-embeddings
in Euclidean spaces.
>27
7. I_f r £ 00, then given any C ^ -bundle £ with closed
tl 5 and bs £ , the space Secr£ is dense in SecS£ , for any s < r.
The same holds true for r = a, provided that tl£ and bs£ admit
Ca-embeddings in Euclidean spaces.
Fundamental Concepts
27
2. £ is an n-dimensional real vector C -bundle (1 £ r £ a)
27
if it is both an n-dimensional real vector bundle and a C -bundle,
and these two structures are compatible, meaning that the restriction
of £ over a small enough neighborhood of an arbitrary point of bs £
is Cr-GLIRn -trivial (i.e., is Cr-GLIRn -equivalent to a standard trivial
bundle, where, of course, a Cr-GLIRn -equivalence is just a GLIRn -equi
valence which is simultaneously a (^-equivalence) . The Euclidean,
27
complex vector, and Hermitian C -bundles are similarly defined.
^r
Products of vector, Euclidean, or Hermitian C -bundles, as
well as bundles induced (in particular, obtained by reducing) such
>r
bundles, are again C ' -bundles of the same kind, provided the conditions
imposed by the corresponding definitions from Subsection 1 (see 1.2)
are fulfilled.
We add that the statements and proofs of Theorems 3.2.8 and
3.2.9 carry over, with obvious modifications, to vector, Euclidean, and
Hermitian C^r-bundles. Here we formulate only the Cr-GLnRn -version of
Theorem 3.2.9: let f: Ç n be a Cr-GLmn -map, where Ç and n are
C"'r-GL3Rn -bundles, and suppose that the bundle (bs f) !ti is neatly
induced (see 1.1.); then corrf is a Cr-GLnRn -equivalence.
Since the proofs of the real and complex cases differ only in
some obvious details, we shall prove only the former. Choose, for every
point b € bs K , a chart E Atlbbs £ such that
n
and 0 outside 2D Finally, define H tl £ IR by
— 1 ” 1
a (ip (pr £ (x) )) pr otl Tb (x) , if x £ pr£ (suppi^),
i i i
H± (x) =
-1
0, if x t pr £ (supp ),
l
n
where pr„ = [pr„ : supp ipb . x nR IRn ] , and let H: tl £ IRnx ...x]Rn =
1
= IR be given by H(x) = (H^ (x),... , (x)). Clearly, H is a C -map
and its restrictions H are linear monomorphisms for all
pr £ 1 (b)
b E bs t . Set m = s n and f: bs £ -> G (m,n) , f(b) H (pr £ 1 (b)
i
To verify that the bundles £ and f *Gra(m,GL(n,JR) ) are
Cr-GLIRn -equivalent, it is enough to produce a Cr-GLnRn -map
<f>: £ Gra (m,GL (n,IR) ) with bs (j> = f. Such a ^ is defined by
tl cj) : tl £ -> tl Gra (m,GL (n, ]R) ) , tl cj)(x ) = (f (pr £ (x) ,H (x) ) (recall
that tl Gra (m,GL (n,IR) ) = {(y,y) € G(m,n) x ]Rn | y £ y}).
An Application
27
6. ijE 1 £ r £ °°, then every real (complex) vector C -bun
with compact base has a Euclidean (respectively, Hermitian) C -metric.
>r
7. Given two real or complex vector C ' -bundles, £ and £ !,
and a C S-map f : bs £ ' -* bs £ with r ^ s ^ 0, we let LS (£ ,£ 1;f )
denote, in Theorems 8 and 10 below, the set of all linear CS-maps
cf>: £ 1 + £ such that bs <p = f. If s ^ a, then LS (£,£!;f) inherits
s
a natural topology as a subset of C (£',£).
When dim £ = dim £ ' = n, bs £ = bs £ ' , and f = id, LS (£,£';f)
contains the set of all CS-GL3Rn -equivalences £' -> £ in the real case,
and the set of all CS-GL(Cn-equivalences £' -+ £ in the conplex case. In both
cases this subset is open for any s / a.
Notice that among the spaces L S (£,£';f) we find SecS£,
0 £ s £ r (see 2.1). More precisely, Secs£ is canonically homeo-
morphic to L (£ ,£ ';f) , where £' is the standard trivial bundle
(bs £ x ]R ,pr^,bs £) , and f = id bs £; the canonical homeomorphism
357
s s
L (£,£'; f) Sec £ takes each map <}>: -> £ into the section
b ^ tl <Mb,1) .
Since the proofs of the real and complex cases differ only in
some obvious details, we shall again prove only the former. By Theorem
5, we may assume that £ = g *Gra (m,-GL (n,HR) ) and = g' !Gra (m' ,GL (n' ,]R),
where g: bs £ -> G(m,n) and g' : bs £ 1 -*• G(m',n') are some Cr-maps.
Then we can identify tl £ with the Cr-submanif old { (b,y ) £ bs £ x i " 1)
y = g (b) } of bs £ x 3Rn and, similarly, tl £ ' = { (b',y') € bs F,1 x iRm |
y' = g(b')}. The orthogonal projections of 3Rm onto its subspaces
g(b) with b £ bs £
combine to define a Cr-map p: bs £ x ]Rm tl £,
x in1
and a C -map p' : bs x ]R £* is similarlydefined. Let Abe the
m* m
Euclidean space of all linear maps IR IR (which is the same as the
space of all real (m><m1)-matrices; c f . 3.2. 1.1. or 3.2. 1.7), and
consider the mappings
{ [$(<(>) ] (b') } (y' ) = [pr2 : bs £ X 3Rm + 3Rm ] (tl <j> o p' (b* ,y' ) )
and
Constructions
Tangent Bundles
Normal Bundles
5. Degree
Further, let y,z € Y besuch that deg^g and degzh are defined
(see 1). If the maps
But
Indeed, since relg and relh are homotopic, h ^ (Z) = g ^ (Z), and
thus (1) follows from that part of the lemma which we have already
proved.
367
_ oo
3 (LEMMA). For any compact C -manifolds X and Y, the set
oo
C g (X,Y) n C (X, int X;Y, int Y) is d e n s e in C (X,3X;Y,3Y) .
k 1 (x) = x, if x £ X ^ k( 3 X x I),
and
'l1 ( 1 (z, t) ) = l(z, (1 — 6)t + <5) , if z £ 3Y , t £ I,
1 (y) = y, if y £ Y V 1 ( 3Y x I).
Now define h2 by
368
h3 (X ^ k O X x [0,6/6)) c int Y ;
Applications
(Cf. 3.1.1.4.)
6. Exercises
ab pr E, x id I pr K
9bs K x I ------ *■ bs K
commutes.
372
>r
4. Show that if r £ °°, then for every C ' -bundle £ with
compact bs £ and tl F3, Secr£ is dense in SecS£ for any s < r.
(This generalizes Theorem 2.7 for r f a.)
>r
5. Show that every C ' -bundle £ with compact bs £ and
r 00 _.
tl£ is C -isomorphic to a C -bundle (cf. 2.8).
11 . Show that for odd n > 1 the degree of any map Sn -+ IRpn
is even.
-1
For r > 0 and <p £ Sph^(X,XQ), the spheroid <p , called
- 1
the inverse of <p , is defined by <p (t ^ ,t^ ,...,t^) = <p ( 1-t ^ ,t^ /.../1^)
1+1
— ,t2 ,...,tr if 0 * t1 £ ~ ^ r
lf _1+t
_ ,^ tl <„ —2+t (2 )
X if ^ « t 1 £ 1,
2ti
—— t" t" if 0 £ t_ s? 1+t
1+t 2' ' *•' r
(3)
if i t, s 1,
V ¥
and
1-t
if 0 £ t„ £
xo ' 1 v 2 '
(4)
2t -1+t
1-t
The Case r = 0
The Case r = 1
We have to verify that the products (p\p and ¡¡j(p are homotopic
for any spheroids <j),ip £ Sph (X,Xq ), r > 1. Consider the following
r
three homotopies I x I X:
( 2 t l ~1 , (1+t) t 2 - t , t 3 , . . . , t r ) , if J sc t 1 « 1, i: t 2 ^ 1,
0' otherwise;
O' otherwise
and
( ( t 1 , t 2 , t 3 , . . . , t r ) ,t) H-
0' otherwise.
Second homotopy
wm
?
9 lfP 9 m
? i 9
B 9
y y y / ?
W //r
9
iü \ 9 w !Ü1
F ir s t hom otopy Third hom otopy
Fig. 13 (r = 2)
and clearly f^ takes homotopic spheroids into homotopic ones, and takes
the constant spheroid into the constant one. Moreover, f„ (M) =
ttr
= f n (ijj) for r > 0. Therefore, f„ defines a homomorphism
fr r tf r fipr
t\ ^ (X ,X q ) tt^X^Xq) for each r > 0, called the homomorphism induced
(see 7) .
378
A Multiplication Theorem
7Tr (X x y, (x Q ,y Q ) )
TTr (X'X0 ) X TTr (Y,Y0 :
(f x g)
(iii) if s and s
are homotopic paths, then T = T
I s s
Condition(iii) shows that we may write T instead of T ,
a s
where a is the homotopy class of the path s. Moreover, from (i)-(iii)
it follows that all the homomorphisms T (T ) are actually isomor-
1 — s o ._
(X,{G },{T }), where allthe G are equal to some fixed group G,
X s x
and all the homomorphisms T are the identical automorphism of G.
By Theorem 4, a local system of groups on a connected
topological space X with base point x^ is simple if and only if the
induced action of tt (X,xn ) on G
is the identical action. In
x0
particular, a local system is simple whenever (X,xQ ) is trivial or
the groupsare all isomorphic to 7L^ .
3. Local Systems of
Homotopy Groups of a Topological Space
'0 along the homotopic paths s and s'. To show that the spheroids
381
and <j)^ are homo topic, where <f>1 (11 ,...,t^) = h ((t ,...,tr ) ,t )
and <J>^ (t^ ,. . .,t^) = h ' ((t^ , .. .,t^) ,t) , pick some homotopies,
rr
f: I x I + X, from (pQ to (p^, and g: I x I X, from s to s 1,
g(u,v), if y £ Fr Ir
IRr
H(y,u ,v) =
h(y,v ) , if u = 0,
h ' (y ,v) , if u = 1
JO
(where y E I , u E I,v E I). There exists a homeomorphism
k: Ir x i -> k such that k(y,u) = (y,u,1) for all (y,u) E Fr(ir x i);
for example, take the inverse of the homeomorphism k ^ : K -*ir x x,
TTr (X )
383
4. R e lat iv e H o m o to py Grou ps
1. Set J r_1 Ir = Fr
Int • G i v e n any t o p o l o g i c a l
IRr 3Rr
pair (X,A) w ith base p o int xQ € A and any p o si ti ve i n t eg er r, we
f* (ab f) *
Local Systems of
Homotopy Groups of a Topological Pair
T T
s > s
TTr (X,A,s(1) ) -- ^-- > 7Tr_1 (A,s(1 ))
-1
7. If a ,3 £ ^ 2 (X/A,Xq), then a Ba = T 0a 3.
Now set
T (too) = T (to)T. (a) for all a £ tt.. (X, s (0) ) and u £ tt (X, A, s (0) ) .
s s in°s 1 1
Furthermore,given any continuous map f: (X,A,Xq ) (X',A',Xq ), the
homomorphismf*: tt^(X,A,Xq )tt^ (X ',A ',X q ) is a
[f* : tt1 (X,xQ)tt ^ (X' ,Xq) ]-map, i.e., f*(wo) = f*(to)f*(a) for all
io £ tt.
j (X,A,Xq ) and a£tt^ (X,Xq ) . In particular, if we consider the
T^to = a)(in*o) ,
h.
i-1 G.
i i+1
••• Gi-i l G i+i
H. H
H±-i l i+1
h! h! .
h i-i i 1+1
Gi-i G!l Gi+i
-
This example has a universal character: every short exact sequence,
1 F £ G 2 H + 1iscanonically isomorphic to a sequence of thistype,
namely, to 1 --> Im f — in— > G - Pr— >G/lm f --- >1; the canonical
isomorphism is obviously {id 1 ,ab f : F -* Im f ,id G,h ^ pr (g~1 (h) ) ,id 1 } .
Splitting
Five Lemma
9 • 11
h h^ h h
G 1 ---- » G2 > G3 ‘ G4 — ^ S
'1 <!>>
g ; — 1 GJ, G. G. h4 G,
= <t>3 o h 2 (b) = <t3 (a) = eQ ,, there is c e G| such that h]j (c) = <t>2 (b).
other hand, h^j o ^ (d) = <{>2 o h^ (d) , and hence 4>2 (b) = <j>2 « h^ (d) .
-1 -1
^4 ° 04 ° h 3 (a) = eG , i.e., ((>4 o h 3 <a) £ Ker h^ = Imh^ . Let
= (<fr4 o <f, " 1 c h 3 (a)) (h3 o((,3 (b) ) _1 = 4>4 ((0 4 1 ° h^ (a) ) (h3 (b) )_1 ) =eG ,,
_ ^
there is c £ G^ such that h^ (c) = ai^ib)) , and hence there is
77- Sequences
where:
V n , II2 are sets with an identity (distinguished element),
morphism (B *+ a - 1 3 a; '
0 = hi ° Pi for all i ^ 0;
yp3 (a) = p2 (B)p3 (a) = p2 (Bp3 (a)) = p2 (3p3 <a)B 1 B) = p2 (-Bp3 <a)B 1)S =
Splitting of TT-Sequences
Ba B_ 1 = [ B C ( p 3 ( B ~ 1 ) ) ] U ( p 3 ( 3) ) a B _ 1 ] =
_1
permuting the factors is permissible because 3c(p3 (6 )) e K e r p 3 = Imp^
°° 00
19. Let =o and i /Pi i=0 be exact Tr-sequences,
... tt2 (A,x 0 ) -ln* > ti2 (X,X q ) ■ rel* > tt2 (X,A, x q ) — it 1 (A,xQ ) ln*^
(7)
TI1 (X,xQ) -rel* > ir1 (X,A,xq) — itq (A,xQ ) in* > ir0 (X,x0 )
Here, all the terms, except for the last six, are Abelian
groups, all the terms, except for the last three, are groups, the last
three termsare sets with an identity, all themaps,except for the
last three, are group homomorphisms, and the last three maps are homo
morphisms of sets with identity. By 3.3 and 4.4, tt^(X,Xq ) acts from
the right on TTr (X,xQ), while tt^(A,x 0 ) acts from the right on
iTr (A,XQ) and 7Tr (X,A,XQ), and all these are group-actions.
Furthermore, the homomorphisms in*, rel*, and 9 are compatible
with these actions, as required in Definition 5.11 (see 3.6, 4.5, and
3.4), and a Ba = T ^ 3 for all a,B £ tt2 (X,A,xq ) (see 4.7). From
4.8 and 4.9 it follows that tt^(X ,x 0 ) acts from the right on tt^(X,A,Xq
in such a manner that rel*(co)a = rel* (wa) for all oo, a £ tt (X,xQ).
Therefore, (7) is a 7T-sequence, called the homotopy sequence of the
pair (X,A) with base point x ^ .
T Ь (аза) = T S (w)T.
1П оS
(a) [a E тг| (X,s(0)), ш € тт] (X,A, s (0 ))] , deduced in
4.8, it follows that the vertical homomorphisms are compatible with the
actions of the fundamental groups.
rel* /T, ^ x
... 7Tr (A,x0 ) -- n * > TTr (X,x0 ) *-» тг (X,A,x..)
"r - 1 (A' V
3 o Yr = idSphr (A,xQ) .
is homotopic to the constant loop. Since (abf)* : tTq (A,Xq ) tTq (A',Xq )
is an isomorphism and (abf)* : (A,w (1 )) -* tt^ (A ',f (w^ (1 )))isan
epimorphism, there is a path s: I A such that s(0) = w^(1),
s(1) =w 2 (1) / and the path ab f os: I + A* is homotopic to s'. For
such achoice of s, f takes the loop
... tt2 (A, B ,xQ ) -— ■*-> 2 (X ,B ,xQ ) - rel*^ tt2 (X,A ,xQ ) —
(11 )
tt1 (A ,B ,xQ ) ^ (X,B,x0 ) rel* » TT1 (X, A,xq )
<f>(t^/...ft^)f if — ^'
^ (t^ / •• • tr+i — 1 t
Let us prove the last part of the theorem. Suppose that the
spheroids 4>0 € Sphr (FQ ,x0 ) and <{>., G Sph^F^x.,) are fiber homotopic
404
along the path u: I + tl £ , and that the same holds for the spheroids
^ e Sphr (F0 ,xQ) , I^ e Sphr (F1 ,x1), and the path v: I tl E, . Further,
suppose that the paths pr £ ° u >Pr £ ° v: I bs E, are homotopic, and
that (J)Q and are freely homotopic along a path w: I + FQ . The
last means that there is a fiber homotopy from <J>0 to \J/Q along the
path wQ = [in: FQ tlEj o w. It is clear that the loop
V : 71 r (Pr £ (s(0)),s (0)) -+ 7T^ (pr E, 1 (S(1)) ,s (1)) along some path
ip £ Sphr (tl £ ,pr £ ~ 1 (B) ,Xq ) such that pr^ ^ (ip) = <J>.
2) pr £* is monomorphic.
1 _
Let ip £ Sphr (tl £ ,pr £ (B) ,Xq ) , and suppose that the spheroid
prt^(ip) £ Sphr (bs E, ,B,b0) is homotopic to the constant spheroid.
406
we see that [pr £* : tt (tl^/F^x^) -> tt^ (bs £ ,b^)] © rel* is nothing
else but pr E,* : tt (tl£,xQ) ■> TTr (bs^,b0 ) . Finally, if we attach the
homotopy group TT^(bs^,b0) to the right of the resulting sequence by
means of the homomorphism pr : ^(tl^x^) -* tTq (bs £ ,b^) , we obtain
the sequence
tt1 (Fq ,x0) > tt1 (tl ?/XQ ) P— tt1 (bs S/bQ ) — (12^
T
Trr (tl£,v(0 )) --- 2 -- » Trr (ti C,v( 1 ))
is free and its orbits coincide with the fibers of £• Since u(t) and
v(t) lie in the same fiber, for each t €I there is a unique g(t) € G
such that u(t)g(t) = v(t). Therefore, if h : ir x I tl £ is a free
homotopy from <t>Q e Sphr (tl £ ,u (0 )) to <|> € Sph (tl £ ,u(1)) along u,
then (y,t) h(y,t)g(t) yields a free homotopy from the spheroid
y ** ^giyigQ to the spheroid y c|>.j(y)g along v.
... 1T2 (G,e) in* > it2 (tl S,x0 ) pr •*"»TT2 (bs £,bQ ) — (G,e) -in*->
(bQ = pr £ (xQ ) ) .
416
of the ball Dr , and we let kug^ denote the elementof tt r (Dr ,Sr ~ 1,ort )
that it defines.
Obviously, 9 (ID) = IS, whence 9(kug^) = sph^_^.
from the pointed space (Sr ,ort1) into the pointed space (X,x ), and
( Y 1 * Y 2 ' • • * , y r ) =
(14;
2
^ (y1 ' Y 2 '*““ 'yr * ' if i m y i ^ °'
(Dr ,Sr _ 1 ,ort ) -> (X,A, xQ ) , and define ID^: Sph°(X,A,xQ) + Sphr (X,A,xQ )
u
by ID (<J)) = <p 0 ID. This map is invertible, and 1.3.7. 6 implies that
two maps, (p,ip € Sph°(X,A,x ), are homotopic,if and only if the
n r nU jj.
spheroids ID (<t>) and ID (rp) are homotopic. If r 5 2, then ID
This leads to two maps, Cub,Buc: Sphr +s (X,xQ) -> Sphr (Sphg (X,xQ),const) .
These maps are invertible and both they and their inverses take
homotopic spheroids into homotopic ones. If r > 0, then the
420
Y ■ (T Y ^ I T T2 )...(T V '
1 2 m
where y € irr (X,A,x0) and y^ 6 irr (X, A ^ o t ^ (ort1) ) are the elements
represented by the spheroids g and g ° 6 Sph^(X,A, go-t ^ (ort^ ) ) ,
of* IR .
Indeed, g*: (Dr ,C ,ort ) -+ iTr (X,A,XQ) takes the class of the spheroid
r
rel id D into y, while it takes the class of the spheroid
translated along £., T y.. into
1
s. 1
l
Second remark: for a given m, it suffices to prove the
theorem for a standard choice of d^,...,d^, namely, for the balls of
Now apply the theorem, first for the case of two factors, and
then for the case of m- 1 factors, to conclude that
’. V i V , = s- (17)
and
m- I m
take (X,A,Xq) = (Dr ,C,ort1), g = id, and the balls d ^ ,...,d^_ 2 ,d ) .
At last, (16), (17), and (18) yield (15).
12. Exercises
in* : Tir (pr5 1 (pr ^ (x) ) ,pr 5 1(pr £ (x) )flE,x) + TTr (tl£,E,x)
is an isomorphism.
commutes.
Let us add two important, yet obvious remarks. Firstly, if
f: (X,xQ) + (Y,yQ) is continuous, then the diagram
f (su f ) *
Secondly, since (su (Sn ,ort^ ) ,bp) = (Sn+\ort^), our construction, when
applied to spheres, yields a homomorphism Tr^S^ort^ + Trr + 1 (Sn + 1 ,ort 1 ) .
The main theorem of this subsection, Theorem 4, is devoted precisely to
this homomorphism.
(see 1.10.6), and hence the homomorphism su: TTr (X,xQ) nr + 1 (su(X,xQ),bp)
may be defined as su = cub o lp^ (to check this facts is routine).
-1
TTr (X,xQ) ---- > TTr + 1 (con (X,xQ ) ,X,xQ) Pr* >7T
r + 1 (su (X,xQ) ,bp)
(9: TTr+>|(con (X,Xq ) ,X,Xq ) TTr (X,XQ) is invertible because the cone is
contractible; see 1.6.5).
4. su: Trr (Sn ,ort^) -> tt^ +^ (Sn+\ ort^ ) is an isomorphism for
r £ 2 n- 2 and an epimorphism for r = 2 n- 1 .
iM t-j , . . •, t r ) = <p ( t 1 , . . . , t r , 1 / 2 ) .
427
The formula
((t^,...,t^+^),t) ^
pr (^(t^,..., t^ ,— ) ,t^+ ^) 1 -t
if £ t
2 r +1 ‘ T*
t --t
(t t r_+4_ ) 1 +t
\ L , , , . . . , L-^_ ,
'1 ' ■
i
1 -t
), if £ t £ 1,
2 " "r+1
2T+ 1
homotopic to (J),and we need only remark that <j>^ o rel G : (I x I,
Fr Ir+1 * ! ) - * (Sn+1,ort^) is a homotopy from tf^ to a spheroid (J>2
... 7Tr (Sn ,ort1) TTr + 1 (sn+1 ,ort1 ) TTr+2 (Sn+ 2 ,ort1 ) -2H-* ...
n 'i k + 2 , with a single group, called the stable group of the series
{iTn+k (Sn ,ort.^ ) } ; we denote it by Stab(k).
1 1
2. If r > 1, (S ) is trivial. tt^ (S ,ort^ ) is an
infinite cyclic group with generator sph^.
1
The proof uses the covering (3R,hel,S ) [see 4.1.2.6 ].
First, notice that hel(0) = ort 1 and hel 1 (ort^) = TL . Since the line
H is contractible, its homotopy groups are trivial, and hence, by
1 1
1.8.12 so is 7Tr (S ,ort ^ ) for any r > 1, and A: tt^ (S ^ort^) ■>
7tq (ZZ,0 ) = 2Z is invertible. Moreover, dR,hel,S^) is obviously a
principal bundle with structure group 7L, so that A is a group homo
morphism (see 1.9.15). Therefore, A is a group isomorphism, and it
remains to observe that A(sph^) = - 1 .
1 ~
u: I -+ S with origin ort , the covering path u : I 1R with origin
0 ends at an integer. Consequently, u~ is homotopic to one of the
paths u~, and thus u is homotopic to one of the loops u^. In other
words, the classes (sph )n with n £ 7L are pairwise distinct and
1
exhaust tt ^ (S ,ort^).
Corollaries
2 1 2 1
4. The pair (D ,S ) is simple. tt (D , S ) is trivial for
2 1 r
any r / 2. ^2 (D ,S ,ort^) is an infinite cyclic group with generator
kug2 .
These are all consequences of Theorem 2 and of the fact that
2 2 1 1
D is contractible, which implies that 0: tt^. (D ,S ,ort^ ) -> TTr _^ (S ,ort^)
is an isomorphism for any r ^ 1 (see 1.6.5).
6.
For n i* 1 , su: irn (Sn ,ort.]) irn + 1 (Sn + 1 ,ort1 ) is an
isomorphism and su(sph^) = sph^+^.
Since tt^(S ) - 7L (see 2), su: tt^ ( s \ ort^ ) -* ^ (S2 ,ort^) cannot have
a nontrivial kernel. The equality sufsph^) = sph^+^ is obvious.
3 2
10. If r ^ 3, the homomorphism pr*: tt (S ) tt (S )
3 2 r r
induced by the Hopf map pr: S S is an isomorphism.In particular,
2
tt^ (S ) is canonically isomorphic to 7L , and is generated by
pr*(sph^), i.e., by the class of the Hopf map itself.
3 2
of the homotopv sequence of the Hopf bundle (S ,pr,S ).
7 4
11. If r ^ 1, the homomorphism pr*: tt„ (S ) -* tt_ (S )
7 4 7 r
induced by the Hopf map p r : S -> S mapstt^(S ) isomorphically onto
7
a subgroup of tt (S ) which has a direct complement isomorphic to
----- 3 r 4 3
tt^ ^ (S ). In particular, tt^(S ) = 7L 0 tt^ (S ) .
15 8
12. If r 5 1, the homomorphism pr*: tt^(S ) -+ tt^IS )
T5 S 15r
induced by the Hopf map pr : S -+ S maps it (S ) isomorphically
8
onto a subgroup of tt^(S ) which has a direct complement isomorphic to
r g
ttr_ 1 (S ). In particular, ti^(S ) ~ 7L © tt^(S ) .
2 2 1/2
X (x^,...,x ) = (xlf...,x , ( 1 - x 1 - ... - xc ) ,0 ,...,0 ).
q
,2 q - 1
Its restriction to Sq - 1 is simply the inclusion Sq - 1 while
pr o Xf where pr is the Hopf map p r : S2q 1 is simply
D S : Dq -+ Sq . Therefore, the diagram
(s’ -1)
1.
3 tt , /(S
„q-1
■ )
r- 1
equals id tt^_^ (Sq ^) To complete the proof, notice that
■1
rel DS* o 9 - su (see 1.2) and 9 ° (rel pr*) - 1 see 1 .8.4) .
14. It follows from 13 and 1.8.7 that su. tt r-(S3)
, -.
v i„
, (S4 )
7 8 5 6
and su: ) -*■ ) are isomorphisms, that is, in the two series
433
(a1 + a2) o su 6 = “ •) 0 su 3 + a 2 o su 8 .
In particular,
(k sphp) o su 8 = k su 8
for anv
------------------------ i -
8 £ 7iq - 1
„ (Sp 1) and integer k.
------------------------------------------- ---------
<+>1 0 p 1 (if;(x) ,2 t) , if 0 £ t £ 1 /2 ,
p 2 (x,t) h-
(f>2 O p (lj;(x) ,2t-1 ) , if 1/2 t £ 1.
n n+k , ..kl n 0 n+ 1
su a o su a = (-1 ) su 8 ° su a .
Using the right and left distributivities (see 1 and 2),. it is not hard
to reduce this equality to the form
435
2
and now we note that [n(n+k) + (n+k) (n+ 1 )] - [n +n(n +l)] = kl (mod 2 ) .
oo
4. Vie set Stab = k§^Stab(k) and identify each group
Stab(k) with its image under the natural embedding Stab(k) -* Stab.
The operation o transforms Stab into a ring: if a € and
B € ^H+k+i (sI1) '
then su(a o B) = su a ° su B (see 1 ). Therefore, °
is well defined as a distributive multiplication Stab(k) x Stab(l)
-+ Stab(k+1) (see 1 and 2), and can be extended bidistributively to a
multiplication Stab x Stab -+ Stab. It results from 1, 2, and 3 that
the ring Stab is associative, has the identity sph = sph 1 = sph = ..
kl
and is skew-commutative, meaning that B ° a = (-1 ) a o B for any
a £ Stab(k) and B £ Stab(l).
An Application
(see 1 ), and we note that the right-hand side of the last equality is
0 (according to 2 ).
3. Among the groups 7Tr (Sn ) which have been computed are
all 7rn+]c(sn) with k £ 22, and all Stab(k) with k £ 37. The group
Table 1
Groups Generators
Table 2
k Stab(k) k Stab(k)
8 7L2 © S 2 12 0
10 2Z2 14 s6©s2
11 15 S 480 0 S 2
S 504
Table 3
438
(-id S )* /r,n , N
---------- 7Tr (S ,ort ^ )
r (Sn t -ort 1i) = 7Tr (Sn ,ort i) = ttr (Sn ) (see 2.5), then we see that T rp
is theidentity if and only if (-id Sn )* : TT^(Sn ) -+ TT^(Sn ) is the
Stabilization
irr (Sp(1)) -> 7Tr (Sp (2) ) + irr (Sp (3) ) -> ...,
stabilizes: the first one, starting with i\^ (SO (r+2)) , the second one,
withTTr (U ([r+2)/2] )) , and the third one, with tt(Sp( [(r+2)/4] ) ) . The
groups 7Tr (SO(n)) with n > r+2, TTr (U(n)) with n 5 [(r+2)/2], and
rrr (Sp (n) ) with n > [(r+2)/4] are said to be stable, and are denoted
by irr (SO), fTr (u), and ir^Sp), respectively. By Theorem 1,
tt1 (SO) s 2Z , tt2 (SO) = 0, and ir3 (SO) s (a © ZZ)/ (cyclic subgroup).
By Theorem 2, (U) = ZZ , tt2 (u) = 0, and tt (U) s a . Finally, by
Theorem 3, (Sp) = 0, Tt2 (Sp) = 0, and tt3 (Sp) = ZZ .
The notations tt^JSO), tt^ (U) , and TTr (Sp) have also a
direct meaning: they represent the ordinary r-th homotopy groups of the
441
Information
r 1 2 3 4 5 6 7 8 r 1 2
TTr (Sp) 0 0 2Z 0 7L 0
*2 *2
TT2n^U ^n ^ " ffin! ' TT4n +2^S p ^n ^ “ S (2n+1)! for n even' and
71 a l0 (Sp(n) ) = Z20 r / 0 for n odd. For details and references
4n+ z Z [ [zn +1 ) !J
for the proofs, see [7].
The fact that the Stiefel manifolds are simple may be seen
from the equalities V(n,k) = SO (n)/SO (n-k) , £CV(n,k) = U(n)/U(n-k),
and IIV(n,k) = Sp(n)/Sp(n-k) (see 4.2.3.12 and 1.9.7). To prove the
rest, use the homotopy sequences of the bundles (V(n+ 1 ,k+ 1 ),p r ,S ),
((CV(n+1,k+1) ,pr,S2n+1) , and (HV(n +1 ,k +1) ,p r ,S4n +3 ) described in
4.6 .1.4, taking the inclusions TR^ -+ lRn , -► (Cn , and lHn
as base points (in the respective total spaces). In the real case, we
take advantage, in addition, of the fact that for n odd, the bundle
(V(n+1,2),p r ,Sn ) admits a section (see 3.1.4.7); for n odd and
k = 1, this ensures that the first of the aforementioned homotopy
sequences splits from the left at the terms tt (V (n+1 ,k+1 ) ) (see 1.8.8)
-s- ... ->■ 7Tn K, (V(n,k)) all the maps are isomorphisms,
~
except for the firs-
which is an isomorphism for n-k even and an epimorphism for n-k odd.
The first claim results from Theorems 7.2 and 7.5, and the
homotopy sequence of the bundle (V(n,k),pr,G (n,k)), defined in
k n
4.6 .1.4, with the inclusion IR -+ 3R as base point The second claim
results from the commutativity of the diagram
A
Tir (G+ (n,k) (so(k))
in. in* = id
(see 1 .8 .6 ]
9. Exercises
t_,1 m imm
D ------- | | D U— X,
J— -V V '
Dk + 1 » d. Dk + 1 J2_ X
1
When
X ^ A c ske, all these assertions follow from 1, 2,
k+ 1
and 3. The general case is reduced to this special situation by
Theorem 2.3 .2 .4 .
An Additional Theorem
[ Ti 1 (
A ,x Q ) * TT ( B , X Q) ] /vk(X,A,B,x0) ----- ► tt n (X,xQ )
a ^ a ^ e ^ . - . a g B g a g ^ ; 1.
S i V i V - ' - V V g V - 1-
452
Fig. 15 (g = 3)
(3:
Imm
) r ., TT (X ,x ) TTr (B,bp)
yCM r y' y
and the right vertical map is the homomorphism induced by the map
f y , (Yy*,) [y *y,€ X ’y,, y ' € M ’] .
a (y 1 )
454
= ilLyEM
£гм(Рг
“ V
, ° у
)** Ь )
у
} vEM = {а v }vEM
тт (X ,x ) = 0 for 1 £ s £ к .
s у у у
If 2 si r «: m, then
are isomorphisms.
(This theorem will be generalized in the next section; see
4.3.1 .)
PROOF. Suppose first that M is finite. By Theorem 2.3.3.2
and the fact that Imm is natural (see 1), we may assume that
ske, X
ky u
reduces to the point x for all p € M. In this case, B => ske „X,
y m+ 1
where X is the cellular product of the spaces X^. By Theorem
2.3.2.2, in*: 7T^(B,bp) TT^(X,bp) is an isomorphism for r £ m.
According to 1.1.9, in* ° Imm: ©y€MTTr (X ,x ) + 7ir (B,bp) is an
isomorphism for any r. Therefore, Imm is an isomorphism for r £ m.
In the general case, 2 shows that for any a c TTr (B,bp) and
r £ m, there is a finite subbouquet B* = V M ,(Xy ,xy) of B such that
a lies in the image of the homomorphism 7T^(B',bp)7T^(B,bp) induced
by the natural embedding B 1 -+ B. Using 1, this homomorphism is part of
the commutative diagram
© ..IT (X ,X ) >TY(B,bp) .
y€M r y y r c
□ — 1— >□
aI I6
n —s
456
Pr<
where abs pr* and rel* are also epimorphic, and Ker(abspr*) is
the smallest normal subgroup of Tr^(X,xQ) which contains Ker rel* =
= Im in* .
(This theorem will be generalized in the next section; see
4.3.14.)
PROOF OF THE CASE k ^ 1. Suppose first that X \ A consists
only of (k+1)-cells and, as a consequence, X/A is a bouquet of (k+1)-
-dimensional spheres. For each cell e £ X \ A, consider the homotopy
class of its characteristic map (viewed as a spheroid of the pair (X,A)),
and translate it to xn , denoting the resulting element of ti, (X,A,x )
0 ^ r k+1 0
by a^. Set Be = pr*(ae ).By Theorem 2.4, the classes T0ae with
a £ t^(A,x0 ) form a system of generators of Trk+^ (X,A,xQ ) , and by
Theorem 5.6, the classes B0 form a system of independent generators
of the Abelian group 7Tk+ ^ (X/A,pr (x ) ) . Moreover, it is obvious that
Pr* (Tga ) = Pr*(a) for all a € Trk+ 1 (X,A, x Q) and o e u (A,x ) , and
these facts will suffice to complete the proof of the theorem for
k > 1. Since Se = pr*(ote )generate Trk + 1 (X/A,pr (x )) , pr* is epi
morphic. Further, since
pr*(T o cx) _
= pr.
- * (a), we have (Tact)cx- 1 G Ker L
pr,.
*
Let us show that the ratios (Toa)a generate Ker pr* . If k > 1
457
X ( e , a)
and the class £ T1 l 1, (e ,a) (T a
a e
) [with only a finite number of
nonzero integers A( e, a ) j belongs to Ker pr* , then £ A( e; a) = 0
for any cell e (because Y [Y X(e,a)]B pr* (a) o: and thus
Le La e
“1 -iÀ (e ,a )
■
5 = T1
# , [ (T a a e ) a e '] When k = 1, this argument is valid
1( e , a)
only after we factor (X,A,Xq ) by its commutator subgroup, and it
k +1
only demonstrates that every element of Ker pr* is a product of the
above form multiplied by some commutators. However, since in
rr2 (X,A,xq) each commutator y ^ 6y6 ^ equals (T 9 Y 6) 6 1 (see 1.4.7),
-1
we obtain again the desired decomposition of £ into ratios (T a)a
a
In the general situation, we first transform (X,A) into a
k-connected pair, removing those components of X which do not contain
x and then replace it by a homotopy equivalent pair (X ’,A *) such
0'
that ske, X 1 c= A ’ (see 1.4.6 and 2.3.3.1). Thus, we may assume that
k
ske, X c A. Now set Y = A U ske, 1X and consider the commutative
k K +1
diagram
l =in*
Tk + 1 (Y,A,x q )
P ’ = pr, P = prv
= [in*: TTk + 1 (Y,x0 ) ^k+ 1 (X,x0 )] and pr* = [pr* : ^k+ 1 (Y,x0 ) "
-*• TTk + 1 (Y/A,pr (xQ) )] . Since the diagram
in*
Tk + 1 (X'x 0 )
rel^ rel.
commutes and pr* = p ’ ° [rel* : Trk+ 1 (Y,xQ) •+ Trk+ ^ (Y,A, xQ)] , we see that
pr* (Ker in*) c p ’ (Ker i) , whence Ker i ’ c p ’(Keri).
458
sphere of C by the ball that it bounds and then attach the resulting
bouquet to X. by g to obtain X.. Finally, set x = inutu (X. .)
1^ I ~ 1 1 ¿J_~l
and 4k = imm2 . The fact that (X^x^) satisfies (i), (ii) for
k = i, and (Jk satisfies (iii) for k = i- 1 , is a consequence of
2.3.
2. Given an arbitrary group it and arbitrary Abelian grou
2 , tt3, ..., there exists aconnected cellular space X such that
TTr (X) s 7Tr (r = 1 ,2, ...) .
1 k
group of tt , ((S ,ort ) V (S ,ort ) ,bp) by its subgroup generated by the
K. I I
elements 2a - a
. (see 6.2). Consequently, tt, (X) is isomorphic to
n+ 1n K
the additive group of binary rational numbers.
9. Exercises
2
1 Consider the subset of (C defined by the equation
p g where p and q are comprime integers, and intersect it with
X1 X2'
,3
the sphere S'". Show that the fundamental group of the complement of
this intersection in S^
is isomorphic to the group with generators
a^,a2, which are connected by the relation p g
= a 2-
(The above intersections (with various p,q) are all homeo-
morphic to a circle and lie on the torus |x1 1 = v/2 /2 , <2 ! = /2 / 2
3
which is contained in S ; they are called torus knots. A torus knot
and the torus which carries it are depicted in Fig. 16 as they lie in
3 3 ,
IR = S \ b p .)
2
2. Consider the submanifold of (CP defined by the equation
X1 2 3 = ' Wlth m a positive integer (cf. 3.5.4.2). Show that
the fundamental group of the complement of this submanifold is
isomorphic to .
space of the tangent bundle of a sphere with g handles, and show that
its fundamental group is isomorphic to the group with generators
a 1»'* **'aa'k
y 1i'’’ ’ g which are connected by the relations
a b a 1b 1 ...a b a 1b 1 = d 2 2g
1 1 1 1 g g g g
and
a 1d = da , .. .,a d = da^, b1 d = d b 1 ,...,bd=db .
l l g g 1 1 g g
,2 -h
C 1 C 1 * **ch°h ' d
and
-1 “ 1
c.d = d c . , . . . ,c, d = d c, .
I I n n
1. Fundamental Concepts
is a weak homotopy equivalence and that the converse is not true. The
first fact was established in 1.3.7? for the second, see 3.5.
The composition of two weak homotopy equivalences is obviously
a weak homotopy equivalence.
2. Let f : X -> Y be a weak homotopy equivalence. Then for
any cellular pair (K,L) and continuous maps <j>: K -* Y and ijj: L X
with f ° ip = <$>I there is a continuous map y: K + X such that
,rTi_T" |L ■ ..
x |IT
l = ip and f o y is L-homotopic to <p. — The converse is also true;
moreover, if f: X -* Y is continuous and has the property that for any
continuous maps, <j>: Dr Y and ip: S r ^ •+ X (r ^ 0) with f ° ip =
1 , there is a continuous map x* D -+ X such that x „ = ip and f °x
sr 1 Sr - 1 —
r —'I
S -homotopic to (p, then f is a weak homotopy equivalence.
k-Equivalence
pair (K,L) withK \ L c: ske^K and continuous maps (j>: K -> Y and
4*: L -> X with f o\p = (j)|t there is a continuous map y : K X such
,Li • —
that x |L = ^ and f © ^ is L-homotopic to c(>. The converse is also
true; moreover, if f : X -> Y is continuous and has the property that
for any continuous maps cf).: Dr Y and ip: Sr~1 -> X (0 £ r £ k) with
f ° $ = <f> r_ 1 there is a continuous map y : Dr -> X such that
S r- 1
X r _i
sr = ^ and f ° x is S -homotopic to <f>, then f is a
k-equivalence.
Let Z denote the resulting space and, to avoid confusion, denote the
maps imm^ : X -+ Z, imm2 : Y Z, and imm^ : C x i z by a, 8, and y.
First, we want to show that the formulas
----------------------------------- - z 1
3. Cellular Approximations
of Topological Spaces
f! #a € ^ ( X , x 0) (Sa = * *•
*'
'i'
(X, A) (X*,A' )
An Application:
Generalization of Theorems 3.3.6, 3.5.5, and 3.6.2
the pairs (X ,x ) be cellular and instead ask only that they be Borsuk
y y
pairs. That this is possible is guaranteed by Theorem 8, the discussion
of bouquets in 10, an the commutativity of diagram (3) in 3.5.1.
14. Theorem 3.6.2 and its corollary 3.6.3 are valid not on
for cellular pairs, but also for arbitrary Borsuk pairs. This
generalization follows from Theorem 8 and the last statement on
quotients in 10.
4. Exercises
s„m+n-1 . „m+n-1
> ^
0
B (m,n) B (n,m)
480
where the vertical maps are the attaching spheroids which represent the
classes wd(m,n) and wd (n,m) (see 1), while the upper horizontal map
is given by (x., xm+n) h- (*m + r xm+n 'X 1 ' '’'' V (and itS de^ree
is (-1)mn).
4. If_ m = 1, then
wd(m,n)
' = imm~.(sph
2* ^ n )[T.lmra^(sph^)^imm0
2*. (sphn )]
—1
In particular, wd(1,1) = c^ct^c^ a , where (as in Subsection 3.1),
,a2 denote the elements imm^ * (sph.|) ,imm2* (sph ^) £ tt^ (B (1 ,1 ) ,bp) .
bp, if x^ £ 1//2,
(1 )
imm2 » DS (/2 (x2 , .. .,xn+1 )) , if x1 5 1 / Æ
along the path t imm. ° DS(1-2t). Now it remains to observe that the
' _1
class of (1) is imm2*(sphn ), the class of (2) is [imir^* (sph^) ] ,
and the class of the above path is imm^*(sph^).
(id Sm Vy)*(w d (m ,n )) =
k k?
= [(id Sm V imm^)* (wd (m,n) )] [(idSm V imm2 )* (wd (m, n) )] , (3)
Pr -|* pr
2 . /Cn I )
(S ,ort. %
^m+n_-, (B(m,n),bP ) tt
m+n-1 1
7rm+n- 1 (Sm'ort 1 )
su SU su
,_m+1 pr 1* ^ 21 n +1
. .
r ort 1 )
^m+n 'ort 1 >
— tt , (B (m+1 ,n + 1) ,bp) ---
m+n m+n (S
tt
given by pr^* and Pr2*- Finally, recall that the last homomorphism
is an isomorphism (see 3.5.5).
[6,a] = ( - 1 ) mn [ a , 3] .
483
= h * ( (-1)mnwd(m,n)) = (-1) mn [a , 8 ]
(see 1.3).
5.
For any■*- a £ tmt (X,x„)
------ u and
--- B £ tt_(X,
n x „),
u the
----product
-------
[a,B] belongs to the kernel of the homomorphism su: TTm+n_1 (X,xQ ) -+
TT (su (X,xn ) ,bp) .
m+n u
This is a corollary of 1.7.
7T1 ( (S1 ,or 1 1 ) V (S1 ,ort 1) V (S1 ,ort1 ) ,bp) , then [[0^ , 0^ ] , ^ ] =
_1 _^ ^ ^ ^
= a^a2a^a2 a2 , whereas [0^ , [a2 , ]] =
“1 —1 1 1 1
= a 3 a 2 a 3 a2 aia2a3a2 a3 a1 ‘ A second example can be found in 4 .2 below,
3. Applications
products [inun1* (a) ,imm2* (3) ] with a £ irk+1(X,x0 ) and 3 € irl+ 1 <Y,y0 )
T imm^*(a)
t mm 1 * <a > , i m m z * (B) ]
with a,a £ it (X,xQ) and 8 £ tt1+1 (Y,yQ) (see 3.2.4); finally, when
k = 1 = 0, Ker in* is generated by the classes
The class
2. [sph^sph^] has infinite order in ^ n - l ^ 3^ '
for every even positive integer n. In particular, the groups
2k
TT4k-1 ^ ^ with k ^ 1 are infinite.
(Cf. Subsections 2.2 and 2.4.)
PROOF. Since
Now assuming that [sphn ,sph ] has finite order, the class
wd(n,n) would have finite order too, contradicting 1.2.
3.
The kernel of su: tr„ „ (Sn ) -* tu (S2n + 1 ) is infinite
------------- zn~I Zn -----------
for every positive integer n .
4. Exercises
2. Show that if
and
3 = Y = i m m 2 * ( s p h 2 ) € tt2 (B (1 ,2) ,bp) ,
then
3. S h o w that
for any a £ Trm (X,x0 ), 3 £ TTn (X,x0 ), an d Y € TTp (X,x0 ) wit h m > 1,
n > 1, and p > 1.
1 i
\J;#M i G . Furthermore, tl Mi G and tl(<|>-£) are «»-connected, and the
latter implies the ^-connectedness of tl(^'MiG) . Applying Lemma
1.5.19 (to the homomorphism from the homotopy sequence of the bundle
i
\p *Mi G into the homotopy sequence of the bundle M i G induced by the
i
map ad : ifj’Mi G ->■ Mi G ) , we see that is a weak homotopy
equivalence. Thus, £ is weakly homotopy equivalent to MiG , and so
£ is universal (see 2).
n 1
is a universal (7L © ... © '&m )-bundle .In particular, S is
2. T h e o ry of C o v e r in g s
The H ie r a r c h y of C o v e r in g s
(pr E,1 © s ' ) (pr K ' ° $ 2 ) 1 belongs to gr 5 1 (x^) , and hence to gr 5 (xQ ) .
and U' = Vj il (pr £' ) ^(pr£(U)). Obviously, x' £ U*. Let us check
that F(U1) c u.
Let y' £ U' . Since ab pr £ ': V' W is a homeomorphism,
thereis a path v ': I tl £ ’ such that v ' (I ) c V ' and v ' (0 ) = x ',
v ' (1 ) = y '. Define v: I tl £ by
pick a path u': I tl £ ' with the property that u'(0) = x^ , u'( 1 ) =
= x', and consider the path u :I -*tl £with u(0) = xQ , u(1) = F(x’),
and ucovers the path pr £ 1 o u'. Clearly, (u'v')(0) = xQ,
(u'v ') (1) = y', while the path uv covers p r £ 1 °(u'v') and
(uv) (0) = X q . Therefore, (uv)(1) = F(v'), and hence F(y’)= v(1) =
= (pr £ |y )~1 (pr ^'(y1)) £ V fl pr (pr (U'))c U.
The G ro u p of A u t o m o r p h is m s of a C o v e r in g
_ , ab ev _ (fact ab A) , rt r/ *
Aut ------- > Reg ---------------> Nr (gr E, (xQ ))/gr £ (xQ )
R e g u la r C o v e r in g s
Existence of Coverings
a loop with origin b^, the path which covers it and has origin b^
ends at Xq if and only if the homotopy class of the given loop belongs
tO 7T.
17. The previous theorem completes the theory of coverings
with a fixed base. Combined with Theorem 7, it establishes a one-to-one
correspondence between the equivalence classes of coverings over a
connected, locally connected, and semilocally simply connected space B
with base point b^ and the classes of conjugate subgroups of tt^ (B,b^).
It transforms the hierarchy of coverings into the usual, set-theoretic
hierarchy of subgroups, and the normal subgroups correspond to the
regular coverings. The covering corresponding to the trivial subgroup
has a simply connected total space. Since every covering is subordinate
to this one, it is called universal. (Warning: do not confuse this
universality with the notion of universality defined in the theory of
Steenrod bundles. Note, however, that among the universal Steenrod
bundles one finds also universal coverings, namely the universal
principal bundles with discrete structure groups; see 1.5 and cf. 18).
An Application:
Classifying Spaces of Discrete Groups
Covering Maps
tl Hj (tl Ç,XQ ]
7T1 (tl Ç ' ,X Q ]
pr Ç*
it follows that f* (gr Ç ' (x^)) c gr Ç (xQ) . Now let us prove the convers
Consider the bundle f ‘Ç together with the map ad f : f 'Ç -* Ç. Let
y q £ (pr f !Ç) _1 (pr Ç ' (x^) ) be such that tladf(y^) = xQ . Further, let
Y' be the component of tl(fÇ) containing y^, and let p' be the
restriction of pr fX to Y'. Obviously, (Y',p',bs^) is a coverin
tl ad f and f combine to define a map (Y',p',bsÇ') Ç, and
tl ad f I is injective on every fiber of (Y',p',bsÇ') . In the diagra
Y'
pJl ^, .... ^ > A ,, .,-1 (pr
. Ç' ^¿)) ,y^)
(Y '/Yq V<P'
ab tl ad fj ab tl ad f
.-1
(tl Ç,xQ) Pr £*■>tt 1 (bs ^,pr (xQ) ) îTq (pr Ç (pr Ç (xQ) ) / xQ]
the rows are exact, while the left vertical homomorphism is monomorphic.
i _
Consequently, f* (Im(pr£*)) = Im p* , whence Imp* gr F,1 (x^) . Sine
bs£' is locally connected, the last inclusion shows that the covering
(Y',p',bs£’) with base point y^ is subordinate to the covering
with base point x^. If is such a subordination, then
(tl adf L.) tl ip' and f combine to define the desired map <t>.
Finally, we claim that <p is unique. Indeed, suppose
(py. £ is another map with bs = f and tl^fx^) = x^. Then
for every point x ' £ tl £' and every path s': I -> tl £' with s ’ (0]
= Xq , s'O) = x ', the paths tl s and tl (p cover
f o pr o s' and have the same origin x^. Therefore, in these
circumstances tl <j> ° s * = tl ^ ° s', and hence tl <j>(x ') = t l <p ° s ' (1
= tl s ' (1 ) = tl c()1 (x' )
1
20. Suppose
is a covering with base point £
e t u , x^
Y is a locally connected space with base point y^, and f (Y,y0)
+ (bs £,pr £ (xQ)) is continuous. If f* (tt 1 (Y,yQ)) c gr U x Q) [in
particular, if Y is simply connected], then there is a map
F: (Y,yn) (tl£,xn) which covers f.
3. Orientations
5. Exercises
10. Let £ be
a covering with bs 5a topological group.
1 -
19.
Show that for n ^ 2 the number of pairwise GL(Cn -non-
n 5 -
equivalent GL(E -bundles over S does not exceed 2.
[2] Adams, J. F., Atiyah, M. F.: K-theory and the Hopf invariant,
0. J. Math., Oxf. II. Ser. V7, 31-38 (1966).
[7] Fuks, D. B.: Homotopic topology, Itogi Nauki Tekh., Ser. Algebra
Topologiya Geom., 71-122 (1969); English. Transl.: J. Sov. Math.
1, No. 3, 333-362 (1 973) .
[8] Grauert, H.: On Levi1s problem and the imbedding of real analytic
manifolds, Ann. Math., II. Ser. 6J3, 460-472 (1958).
[11] Kelley, J. L.: General Topology, Van Nostrand, New York, 1955;
Reprint Graduate Texts in Math., Springer-Verlag, New York,
Heidelberg, Berlin, 1975.
[12] Kervaire, M.: A manifold which does not admit any differentiable
structure, Comment. Math. Helv. 3_4, 257-270 (1960).
A-homotopy 5 7
a-th barycentric coordinate 99
a-th barycentric function 1 0 2
C*>atlas 142
C -bundle 34 6
Complex vector — 354
Euclidean — 354
Hermitian — 354
Real vector — 354
C^r-bundle 347
C^-embedding 161
C -equivalence 34 7
Cr-function 141
C -isomorphism 34 7
C -manifold 149
Cr-map 141, 145
Cr-parallelizable 160
C>-space 143
-space 143
C -structure 142
Cr~subspace 147
C -topology 197
F-bundle 300
Locally F-trivial — 300
F-trivial — 300
Weak — 299
F-isomorphism 298, 300
F-map 300
F-structure 297
G-equivalence 299
G-map 286, 289
G-space 288
Effective — 288
Right - 290
G-structure 298
H-space 413
Glossary of Symbols
Con -46 El - 2 4 4
con -42, 51 el - 244
conj - 328 Embr - 197
const — 50 Esi — 99
cop - 13 7 ev _ 4 9 3
corr - 267 ext - 325
CP11 - 39 fact _ 3
oo
115 " 41 Fibr - 317
csc - 114 pr - 6
Cub - 419 G(n,k) - 177
cub - 420 G (°°,k) - 337
(CV(n,k) - 173 G ' (n,k) - 185
I V (00,k ) - 342 G + (n,k) - 178
(CV' (n,k) - 176 G+ (°°,n ) - 337
(EV' (°°,k) - 342 GF - 445
Cy1 _ 45 GL (n,CC) - 176
“ 335 GL(n,M) - 176
d " 159 GL (n,3R) - 175
3 - 133,143, 385 GL+ (n,3R) - 176
D. - 157 n
1 GL +!IR - 324
°n “9 GL(En - 324
dx - 156, 159 GL3Rn - 324
D00 - 41 gr - 490
d T (x,p) - 202 Gra “ 340
deg - 364 9 rad " 241
diag - 4 IH - 38
Diff - 197 m n - 39
00
Dist - 9 3H - 41
dist - 8 M * - 281
-,]RП -
о
8 supp - 142
OO
Ж - 41 Symm - 282
]R* - 281 (e.g. Afc) - 181
3R^ - 133 Ts - 378 , 381, 386, 4 0 4
3R^ ” 1 - 134 T r - 100
IR* - 281 Tang - 156, 157
IRÇ - 329 tang - 36 0
Reg - 493 TD - 100
rel - 2
TG - 319
IRG (n ,к ) - 177
ti - 265
H G ' (n,к) - 185
tn - 157
Ж Р П - 39 Top - 282
1RP°° - 41 Tu - 203
rt - 45 tu - 203
3RV (n ,к ) - 171
Tub - 2 03
3RV' (n,k) - 175
tub - 203
Sn - 9 и - 441
OO
S - 41 U(n) - 174
S T (x,p) - 202
Uin - 324
Scyl - 113
V (n,к) - 171
SecS - 350
V (°°,k) - 342
sch - 105 V ’ (n,k) - 175
shi - 185
V'(°°,k) - 342
Si - 102
vk - 44 9
si - 1 0 2
Wd - 4 79
ske - 85
wd - 4 79
SO - 441
ZZ - 281
S O (n) - 172
7L - 294
Б ОШ 11 - 324 m
zer - 3, 278
Sp - 441
Л - 407
Sp(n) - 175
TT - 57
Sph - 373, 384
ïï - 373 , 384
sph - 416 r
V
Sph° - 416, 417 (e.g. <f>V ) - 47
Л
St - 114 (e.g. ф л ) - 47
U
st - 114 (e.g. фи ) - 54
n
Stab - 429 (e .g. 'P ) - 54
1 -
L ¡r-Ц
Stee - 315 (e.g. Y / 5 ) - 178 , 332
0 - 1
L I - 3
X - 1 , 2<
V - 50
* 42, !
-
0 51 -
0 - 334
U - 38
Ф
X - 94
c
* - 97,
с
0 - 98
с
X - 110
s
★ - 113
s
I. M. James
General
Topology and
Homotopy
Theory
1984. 3 figures. Approx. 459 ms-pages.
ISBN 3-540-90970-2
D .H .L u e c k in g , L .A .R u b e l
Complex Analysis:
A Functional
Analysis Approach
Universitext
ISBN 3-540-13577-4
ISBN 0-387-13577-4