Professional Documents
Culture Documents
Smooth
Manifolds
Smooth Manifolds
Rajnikant Sinha
Smooth Manifolds
123
Rajnikant Sinha
Department of Mathematics
Magadh University
Bodh Gaya, Bihar
India
v
vi Preface
on, throughout the rest part of the course, the deeper understanding of the subject is
lost.
I have benefited from a number of advanced textbooks on the subject, and some
unpublished works of my own and my distinguished friends. Some of these sources
are mentioned in the bibliography. However, in an introductory book such as this, it
is not possible to mention all literatures and all the people who have contributed to
the understanding of this exciting field.
It is hoped that our readers (active and passive both) will find that I have
substantially fulfilled the objective of bridging the gap between university curric-
ulum and more advanced texts, and that they will enjoy the reading this book as
much as I did while writing it.
Rajnikant Sinha
Contents
1 Differentiable Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Topological Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Smooth Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Examples of Smooth Manifolds . . . . . . . . . . . . . . . . . . . . . . . . 14
2 Tangent Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1 Smooth Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2 Algebra of Smooth Functions . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3 Smooth Germs on Smooth Manifolds . . . . . . . . . . . . . . . . . . . . 54
2.4 Derivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.5 Cotangent Spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.6 Tangent Space as a Dual Space . . . . . . . . . . . . . . . . . . . . . . . . 91
2.7 Contravariant Vectors and Covariant Vectors. . . . . . . . . . . . . . . 108
2.8 Tangent Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
vii
viii Contents
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 485
About the Author
ix
Chapter 1
Differentiable Manifolds
The pace of this chapter is rather slow for the simple reason that the definition of
smooth manifold is itself forbidding for many unfamiliar readers. Although many
young university students feel comfortable with the early parts of multivariable
calculus, if one feels uncomfortable, then he should go first through some parts of
Chap. 3. How verifications are done in various examples of smooth manifolds is a
crucial thing to learn for a novice, so this aspect has been dealt here with some more
detail.
uU : U ! uU ðU Þ
such that uU ðUÞ is an open subset of Rm : Here, the ordered pair ðU; uU Þ is called a
coordinate chart of M: ðU; uU Þ is also simply denoted by ðU; uÞ:
Definition Let M be an m-dimensional topological manifold. Let ðU; uU Þ be a
coordinate chart of M: So
uU : U ! uU ðU Þ ð Rm Þ:
ðuU ð xÞÞi ui :
Proof of 1 Let us take any x in domðuV ðuU Þ1 Þ: We want to show that x is in
uU ðU \ VÞ:
Since x is in domðuV ðuU Þ1 Þ, by the definition of domain, there exists y such
that ðx; yÞ is in uV ðuU Þ1 : Since ðx; yÞ is in uV ðuU Þ1 , by the definition of
composition, there exists z such that ðx; zÞ is in ðuU Þ1 ; and ðz; yÞ is in uV : Since
ðx; zÞ is in ðuU Þ1 ; ðz; xÞ is in uU : Since ðz; xÞ is in uU ; and uU : U ! uU ðUÞ;
uU ðzÞ ¼ x; and z is in U: Since ðz; yÞ is in uV ; and uV : V ! uV ðVÞ; uV ðzÞ ¼ y;
and z is in V: Here, z is in U; and z is in V; so z is in U \ V: Since z is in U \ V; and
x ¼ uU ðzÞ; x is in uU ðU \ VÞ: Thus, (see Fig. 1.1),
dom uV ðuU Þ1 uU ðU \ V Þ:
ðy; wÞ is in uV ; ðz; wÞ is in uV ðuU Þ1 ; and hence, z is in domðuV ðuU Þ1 Þ: Since
ðy; zÞ is in uU ; and uU is a function, uU ðyÞ ¼ z: Since uU ðyÞ ¼ z; and uU ðyÞ ¼ x;
x ¼ z: Since x ¼ z; and z is in domðuV ðuU Þ1 Þ; x is in domðuV ðuU Þ1 Þ: Thus,
uU ðU \ V Þ dom uV ðuU Þ1 :
h
Proof of 2 On using 1, we have domðuU ðuV Þ1 Þ ¼ uV ðV \ UÞ: So
1
LHS ¼ ran uV ðuU Þ1 ¼ dom uV ðuU Þ1
1
¼ dom ðuU Þ1 ðuV Þ1
¼ dom uU ðuV Þ1
¼ uV ðV \ U Þ
¼ uV ðU \ V Þ ¼ RHS:
h
4 1 Differentiable Manifolds
and
uU ðuV Þ1 : uV ðU \ V Þ ! uU ðU \ V Þ
are continuous.
Since ðU; uU Þ is a coordinate chart of M; uU is a homeomorphism, and hence,
ðuU Þ1 is continuous. Since ðV; uV Þ is a coordinate chart of M; uV is a homeo-
morphism, and hence, uV is continuous. Since ðuU Þ1 ; uV are continuous, their
composite uV ðuU Þ1 is continuous.
Similarly, uU ðuV Þ1 is continuous. Hence, uV ðuU Þ1 is a homeomorphism.
Similarly, uU ðuV Þ1 is a homeomorphism. h
1.2 Smooth Manifolds 5
or
ðf i ; gi are C1 for every i ¼ 1; . . .; m; whenever U \ V is a nonempty setÞ:
or
ðf i ; gi are C2 for every i ¼ 1; . . .; m; whenever U \ V is a nonempty setÞ:
Similar definitions for ðU; uU Þ and ðV; uV Þ are C 3 -compatible can be supplied,
etc.
It is clear that if ðU; uU Þ and ðV; uV Þ are C 3 -compatible, then ðU; uU Þ and
ðV; uV Þ are C 2 -compatible, etc.
Definition Let M be an m-dimensional topological manifold. Let r be a positive
integer. Let
and
1 1
uU ðuW Þ1 y1 ; . . .; ym g11 y1 ; . . .; ym ; . . .; gm
1 y ; . . .; y
m
¼ x ; . . .; xm :
8 1 Differentiable Manifolds
Similarly, f2i ; gi2 are Cr for every i ¼ 1; . . .; m, where for every ðx1 ; . . .; xm Þ in
uV ðV \ WÞ;
uW ðuV Þ1 x1 ; . . .; xm f21 x1 ; . . .; xm ; . . .; f2m x1 ; . . .; xm ¼ y1 ; . . .; ym
;
2 uW ðV \ W Þ
and
1 1
uV ðuW Þ1 y1 ; . . .; ym g12 y1 ; . . .; ym ; . . .; gm
2 y ; . . .; y
m
¼ x ; . . .; xm :
Hence,
f 1 x1 ; . . .; xm ¼ g12 f11 x1 ; . . .; xm ; . . .; f1m x1 ; . . .; xm :
Case II: when U \ V is a nonempty set. For every ðx1 ; . . .; xm Þ in uU ðU \ VÞ; put
uV ðuU Þ1 x1 ; . . .; xm f 1 x1 ; . . .; xm ; . . .; f m x1 ; . . .; xm ¼ y1 ; . . .; ym ;
1. A is an atlas on M;
e ; u Þ is a coordinate chart of M, but not a
2. A is maximal (in the sense that if ð U e
U
member of A; then there exists ðU; uU Þ in A; such that ð U e ; u Þ and ðU; uU Þ
e
U
are not C1 -compatible),
then we say that A is a C1 -differentiable structure on M; and the pair ðM; AÞ is
called a C 1 -differentiable manifold. C1 -differentiable structure is also called
smooth structure, and C 1 -differentiable manifold is also called smooth manifold.
uU ðuV Þ1 y1 ; . . .; ym g1 y1 ; . . .; ym ; . . .; gm y1 ; . . .; ym ¼ x1 ; . . .; xm :
and
1 1
uU ðuW Þ1 y1 ; . . .; ym g11 y1 ; . . .; ym ; . . .; gm
1 y ; . . .; y
m
¼ x ; . . .; xm :
Similarly, f2i ; gi2 are C 1 for every i ¼ 1; . . .; m, where for every ðx1 ; . . .; xm Þ in
uV ðV \ WÞ;
uW ðuV Þ1 x1 ; . . .; xm f21 x1 ; . . .; xm ; . . .; f2m x1 ; . . .; xm
¼ y1 ; . . .; ym 2 uW ðV \ W Þ;
and
1 1
uV ðuW Þ1 y1 ; . . .; ym g12 y1 ; . . .; ym ; . . .; gm
2 y ; . . .; y
m
¼ x ; . . .; xm :
Hence,
f 1 x1 ; . . .; xm ¼ g12 f11 x1 ; . . .; xm ; . . .; f1m x1 ; . . .; xm :
Example 1.6 (i) Let G be a nonempty open subset of Rm : Let us take G for M:
Since Rm is a Hausdorff second countable topological space, G with the induced
topology is a Hausdorff second countable topological space. For every x in G; G is
an open neighborhood of x: Since the identity mapping
IdG : G ! G
given by
IdG ð xÞ ¼ x
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
x1 ; . . .xmþ1 : x1 ; . . .xmþ1 2 Rmþ1 and ðx1 Þ2 þ þ ðxmþ1 Þ2 ¼ 1 :
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
S1 x1 ; x2 : x1 ; x2 2 R2 and ðx1 Þ2 þðx2 Þ2 ¼ 1 ;
1 2 3 1 2 3 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
S2 x ; x ; x : x ; x ; x 2 R3 and ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 ¼ 1 ; etc.
Now we shall try to prove that the unit circle S1 is a 1-dimensional topological
manifold. For this purpose, let us take any ða1 ; a2 Þ in S1 : We have to find an open
neighborhood G of ða1 ; a2 Þ, and a homeomorphism u : G ! uðGÞ such that uðGÞ
is an open subset of R1 : Since ð0; 0Þ does not lie in S1 ; and ða1 ; a2 Þ is in S1 ; both of
a1 ; a2 cannot be 0. So two cases arise: either a1 6¼ 0 or a2 6¼ 0:
Case I: when a1 6¼ 0
Now two subcases arise: either 0\a1 or a1 \0:
Subcase I: when 0\a1 . Let us take
x1 ; x2 : x1 ; x2 2 S1 and 0\x1
and fðx1 ; x2 Þ : ðx1 ; x2 Þ 2 R2 and 0\x1 gð¼ ð0; 1Þ ð1; 1ÞÞ is open in R2 ;
fðx1 ; x2 Þ : ðx1 ; x2 Þ 2 S1 and 0\x1 gð¼ G1 Þ is open in S1 relative to the induced
topology. Thus, G1 is open in S1 : Since ða1 ; a2 Þ 2 G1 ; and G1 is open in S1 ; G1 is
an open neighborhood of ða1 ; a2 Þ: Since
16 1 Differentiable Manifolds
G1 ¼ x1 ; x2 : x1 ; x2 2 S1 and 0\x1
1 2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ x ; x : ðx1 Þ2 þðx2 Þ2 ¼ 1 and 0\x1 ;
it is clear that fx2 : ðx1 ; x2 Þ 2 G1 g is equal to the open interval ð1; 1Þ:
Let us define a function
p1 : G1 ! x2 : x1 ; x2 2 G
Clearly, p1 is a 1–1 function from G1 onto the open interval ð1; 1Þ: Further,
since projection map is a continuous and open map,
p1 : G1 ! ð1; 1Þ
Subcase II: when a2 \0. Proceeding as above, there exist an open neighborhood
1 2 1 2
G4 x ; x : x ; x 2 S1 and x2 \0
Thus, we see that in all cases, there exist an open neighborhood G of ða1 ; a2 Þ,
and a homeomorphism u : G ! uðGÞ; where uðGÞ is an open subset of R1 :
Hence, by the definition of topological manifold, S1 is a 1-dimensional topo-
logical manifold. Here, we get four coordinate charts ðG1 ; p1 Þ; ðG2 ; p2 Þ; ðG3 ; p3 Þ;
ðG4 ; p4 Þ: Let us take
is C 1 :
By Lemma 1.1, p1 ðp1 3 Þ is a homeomorphism from open subset p3 ðG1 \ G3 Þ
of R1 onto open subset p1 ðG1 \ G3 Þ of R1 : Now let us take any p3 ðxÞ in
p3 ðG1 \ G3 Þ; where x is in G1 \ G3 : Since x is in G1 \ G3 ; and G1 \ G3 ¼
fðx1 ; x2 Þ : ðx1 ; x2 Þ 2 S1 ; 0\x1 and 0\x2 g; there exist real numbers x1 ; x2 such
that x ¼ ðx1 ; x2 Þ 2 S1 ; 0\x1 and 0\x2 . Since
1 2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
x ; x 2 S1 ¼ x1 ; x2 : x1 ; x2 2 R2 and ðx1 Þ2 þðx2 Þ2 ¼ 1 ;
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
and 0\x2 ; x2 ¼ 1 ðx1 Þ2 : Since 0\x1 ; 0\x2 ; and ðx1 Þ2 þ ðx2 Þ2 ¼ 1; it follows
that 0\x1 \1; and 0\x2 \1: Hence, t p3 ðxÞ ¼ p3 ðx1 ; x2 Þ ¼ x1 2 ð0; 1Þ: Now it
is clear that p3 ðG1 \ G3 Þ is equal to the open interval ð0; 1Þ:
Further,
p1 p1
3 ðtÞ ¼ p1 p1
3 ðp3 ð xÞÞ ¼ p1 p1 3 ðp3 ð xÞÞ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ p1 ð xÞ ¼ p1 x1 ; x2 ¼ x2 ¼ 1 ðx1 Þ2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi
¼ 1 ðp3 ðx1 ; x2 ÞÞ2 ¼ 1 ðp3 ð xÞÞ2 ¼ 1 t2
or, qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
p1 p1
3 ð t Þ ¼ 1 ðt Þ2
0 1 t
p1 p1
3 ðtÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffi ð2tÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffi
2 1t 2 1 t2
or,
0 t
p1 p1
3 ðtÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffi ðt 2 ð0; 1ÞÞ:
1 t2
1.3 Examples of Smooth Manifolds 19
0 t
p1 p1
3 ðtÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffi :
1 t2
and t 7! pffiffiffiffiffiffiffi
t
1t2
is a continuous function,
p1 p1
3 : p3 ðG1 \ G3 Þ ! p1 ðG1 \ G3 Þ
is C 2 ; etc. Thus,
p1 p1
3 : p3 ðG1 \ G3 Þ ! p1 ðG1 \ G3 Þ
is C 1 : Thus, we have shown that the pair ðG1 ; p1 Þ; ðG3 ; p3 Þ are C1 -compatible.
Similarly, we can show that all other pairs of members in A are C1 -compatible.
This proves 2. Hence, by Theorem 1.2, there exists a unique C 1 -differentiable
structure B on S1 which contains A: Thus, the ordered pair ðS1 ; BÞ is an example of
a smooth manifold of dimension 1.
Similarly, we can construct a C 1 -differentiable structure on S2 ; etc. In short, the
unit sphere Sm is a smooth manifold of dimension m.
Example 1.8 Let us define a relation over R2 fð0; 0Þg as follows:
For every ðx1 ; x2 Þ; ðy1 ; y2 Þ in R fð0; 0Þg; by
2
1 2 1 2
x ;x y ;y
We shall try to show that is an equivalence relation over R2 fð0; 0Þg; that is,
1. ðx1 ; x2 Þ ðx1 ; x2 Þ for every ðx1 ; x2 Þ in R2 fð0; 0Þg;
2. if ðx1 ; x2 Þ ðy1 ; y2 Þ; then ðy1 ; y2 Þ ðx1 ; x2 Þ;
3. if ðx1 ; x2 Þ ðy1 ; y2 Þ; and ðy1 ; y2 Þ ðz1 ; z2 Þ then ðx1 ; x2 Þ ðz1 ; z2 Þ:
20 1 Differentiable Manifolds
1 2
x1 ; x2 y1 ; y2 : y1 ; y2 x ;x
1 2 1 2
¼ y ; y : y ; y ¼ t x1 ; x2 for some real t
or,
x1 ; x2 ¼ t x1 ; x2 : t 6¼ 0 :
or,
P1 ¼ t x1 ; x2 : t 6¼ 0 : x1 ; x2 6¼ ð0; 0Þ :
The topology of R2 fð0; 0Þg is the induced topology of the standard topology of
R2 ; and the topology of the quotient space ðR2 fð0; 0ÞgÞ= ð¼ P1 Þ is the
quotient topology. Thus, a subset G of P1 is open means the set
[ x1 ; x2 : x1 ; x2 2 G
or
[ t x1 ; x2 : t 6¼ 0 : x1 ; x2 2 G [fð0; 0Þg
is open in R2 :
Since R2 is a Hausdorff space, its subspace R2 fð0; 0Þg is a Hausdorff space,
and hence, its quotient space ðR2 fð0; 0ÞgÞ= ð¼ P1 Þ is a Hausdorff space.
Since R2 is a second countable space, its subspace R2 fð0; 0Þg is a second
countable space, and hence, its quotient space ðR2 fð0; 0ÞgÞ= ð¼ P1 Þ is a
second countable space. Put
1 2
U1 x ;x : x1 6¼ 0 :
O
O
u1 : U1 ! R
x2
u1 x1 ; x2 :
x1
Since ½ðx1 ; x2 Þ ¼ ½ðy1 ; y2 Þ ; ðx1 ; x2 Þ ðy1 ; y2 Þ: Further since ðx1 ; x2 Þ; ðy1 ; y2 Þ are
nonzero members of R2 ; there exists a nonzero real number t such that
x1 ; x2 ¼ t y1 ; y2 ¼ ty1 ; ty2 :
x2 y2
1
¼ u1 x1 ; x2 ¼ u 1 y1 ; y2 ¼ 1:
x y
Case I: when x2 6¼ 0
2 2 2 2
Since xx1 ¼ yy1 ; and x2 6¼ 0; y2 6¼ 0; and hence, x1 ¼ xy2 y1 : So ðx1 ; x2 Þ ¼ xy2 ðy1 ; y2 Þ:
1.3 Examples of Smooth Manifolds 23
is a continuous function.
For 4. It is clear that
u1 ðU1 Þ ¼ ð1; þ1Þ ¼ R ;
u1
1 ðmÞ ¼ ½ð1; mÞ :
u2 : U2 ! R
x1
u2 x1 ; x2 ;
x2
Further
u2 ðu1 Þ1 : u1 ðU1 \ U2 Þ ! u2 ðU1 \ U2 Þ
Now we want to show that the Cartesian product M N; with the product
topology, is a ð2 þ 3Þ-dimensional topological manifold. For this purpose, we must
prove that
1. M N; with the product topology, is a Hausdorff topological space,
2. for every ðx; yÞ in M N; there exists an open neighborhood of ðx; yÞ in M N
which is homeomorphic to some open subset of the Euclidean space R2þ3 (see
Fig 1.3).
For 1: Since M is a smooth manifold, the topology of M is Hausdorff. Similarly
the topology of N is Hausdorff. Since the topologies of M and N are Hausdorff,
their product topology is Hausdorff. This proves 1.
For 2: Let us take any ðx; yÞ in M N; where x is in M; and y is in N: Since x is
in M; and M is a 2-dimensional smooth manifold with differentiable structure A;
there exists a coordinate chart ðU; uU Þ 2 A such that x is in U: Similarly, there
exists a coordinate chart ðV; wV Þ 2 B such that y is in V: Since ðU; uU Þ 2 A;
and A is a differentiable structure on M; uU is a homeomorphism from the open
subset U of M onto the open subset uU ðUÞ of R2 : Similarly, wV is a homeo-
morphism from the open subset V of N onto the open subset wV ðVÞ of R3 :
Since x is in U; and U is open in M; U is an open neighborhood of x: Similarly,
V is an open neighborhood of y: Since U is an open neighborhood of x; and V is
an open neighborhood of y; the Cartesian product U V is an open neighborhood
of ðx; yÞ in M N: Since uU ðUÞ is open in R2 ; and wV ðVÞ is open in R3 ; their
Cartesian product ðuU ðUÞÞ ðwV ðVÞÞ is open in R2 R3 ð¼ R2þ3 Þ:
Now let us define a function
1: ðuU wV Þ is 1 1;
2: ðuU wV Þ is onto;
3: ðuU wV Þ is continuous,
4: ðuU wV Þ1 is continuous:
For 1: Let ðuU wV Þðx; yÞ ¼ ðuU wV Þðz; wÞ; where x; z are in U; and y; w are
in V: We have to show that ðx; yÞ ¼ ðz; wÞ; that is, x ¼ z; and y ¼ w: Since
ðuU ð xÞ; wV ð yÞÞ ¼ ðuU wV Þðx; yÞ ¼ ðuU wV Þðz; wÞ ¼ ðuU ðzÞ; wV ðwÞÞ;
is onto. For this purpose, let us take any ðs; tÞ in ðuU ðUÞÞ ðwV ðVÞÞ: Now we
have to find a member of U V; whose ðuU wV Þ-image is ðs; tÞ: Since ðs; tÞ is
in ðuU ðUÞÞ ðwV ðVÞÞ; s is in uU ðUÞ; and t is in wV ðVÞ: Since s is in uU ðUÞ;
there exists an element x in U such that s ¼ uU ðxÞ. Similarly, there exists an
element y in V such that t ¼ wV ðyÞ: Since x is in U; and y is in V; ðx; yÞ is in
U V: Further
ðuU wV Þðx; yÞ ¼ ðuU ð xÞ; wV ð yÞÞ ¼ ðs; tÞ:
This proves 2.
For 3: Since uU is a homeomorphism, uU is a continuous function, and hence,
x 7! uU ðxÞ is continuous. Since, in the product topology, ðx; yÞ 7! x is contin-
uous, and x 7! uU ðxÞ is continuous, their composite ðx; yÞ 7! uU ðxÞ is continu-
ous. Similarly, ðx; yÞ 7! wV ðyÞ is continuous. Since ðx; yÞ 7! uU ðxÞ is continuous,
and ðx; yÞ 7! wV ðyÞ is continuous, with respect to product topology,
ðx; yÞ 7! ðuU ðxÞ; wV ðyÞÞð¼ ðuU wV Þðx; yÞÞ is continuous, and hence,
is an open mapping. For this purpose, let us take any open subset G H of
U V; where G is open in U; and H is open in V: We have to prove that
ðuU wV ÞðG HÞ is open in ðuU ðUÞÞ ðwV ðVÞÞ. By the definition of
ðuU wV Þ;
of coordinate charts of M N by C:
We want to show that C has the following two properties:
1: [fU V : ðU V; ðuU wV ÞÞ 2 Cg ¼ M N;
2: all pairs of members in C are Cr -compatible, for every r ¼ 1; 2; 3; . . .:
is C 1 ; that is,
28 1 Differentiable Manifolds
^ V \V
uU^ wV^ ðuU wV Þ1 : ðuU wV Þ U \ U ^
! uU^ wV^ U \ U ^ V \V^
^
uU ðU \ UÞð R2 Þ;
uU^ ðuU Þ1 ^ ;
x1 ; x2 f 1 x1 ; x2 ; f 2 x1 ; x2 ¼ y1 ; y2 2 uU^ U \ U
and
1 1 2 1 1 2 2 1 2 1 2
uU uU^ y ;y g y ;y ;g y ;y ¼ x ;x :
and
1 1 2 3 1 1 2 3 2 1 2 3 3 1 2 3
wV wV^ y ;y ;y k y ;y ;y ;k y ;y ;y ;k y ;y ;y
¼ x1 ; x2 ; x3 :
Hence,
ðuU wV Þ1 x1 ; x2 ; x3 ; x4 ; x5 ¼ ðx; x Þ;
or
1.3 Examples of Smooth Manifolds 29
u U ð xÞ ¼ x1 ; x2 ;
and
wV ðx Þ ¼ x3 ; x4 ; x5 :
Now
uU^ wV^ ðuU wV Þ1 x1 ; x2 ; x3 ; x4 ; x5
¼ uU^ wV^ ðuU wV Þ1 x1 ; x2 ; x3 ; x4 ; x5
¼ uU^ wV^ ðx; x Þ
¼ uU^ ðuU Þ1 uU ð xÞ; wV^ ðwV Þ1 wV ðx Þ
¼ uU^ ðuU Þ1 uU ð xÞ; wV^ ðwV Þ1 wV ðx Þ
¼ uU^ ðuU Þ1 ðuU ð xÞÞ; wV^ ðwV Þ1 ðwV ðx ÞÞ
¼ uU^ ðuU Þ1 x1 ; x2 ; wV^ ðwV Þ1 x3 ; x4 ; x5
¼ uU^ ðuU Þ1 x1 ; x2 ; wV^ ðwV Þ1 x3 ; x4 ; x5
¼ f 1 x 1 ; x 2 ; f 2 x 1 ; x 2 ; h1 x 3 ; x 4 ; x 5 ; h2 x 3 ; x 4 ; x 5 ; h3 x 3 ; x 4 ; x 5
¼ f 1 x 1 ; x 2 ; f 2 x 1 ; x 2 ; h1 x 3 ; x 4 ; x 5 ; h2 x 3 ; x 4 ; x 5 ; h3 x 3 ; x 4 ; x 5 :
^ ðV \ VÞÞð
Thus, for every ðx1 ; x2 ; x3 ; x4 ; x5 Þ in ðuU wV ÞððU \ UÞ ^ R2þ3 Þ;
uU^ wV^ ðuU wV Þ1 x1 ; x2 ; x3 ; x4 ; x5
¼ f 1 x 1 ; x 2 ; f 2 x 1 ; x 2 ; h1 x 3 ; x 4 ; x 5 ; h2 x 3 ; x 4 ; x 5 ; h3 x 3 ; x 4 ; x 5 :
for x ¼ ðx1 ; x2 ; x3 ; x4 ; x5 Þ;
1
DF ð xÞ ¼ D f 1 a ð xÞ ¼ Df 1 ðað xÞÞ ððDaÞð xÞÞ
or,
h 1 1 i 1 0 0 0 0
of of
DF 1 ð xÞ ¼ ox1 ðað xÞÞ ox2 ð a ð x Þ Þ
0 1 00 0
h 1 1 i
of of
¼ ox1 ðað xÞÞ ox 2 ð a ð x Þ Þ 0 0 0 :
30 1 Differentiable Manifolds
Hence,
DF 1 ¼ ðD1 f 1 Þ a ðD2 f 1 Þ a 0 0 0 :
x1 ; x2 ; x3 ; x4 ; x5 !7 h2 x3 ; x4 ; x5 ; x1 ; x2 ; x3 ; x4 ; x5 7! h3 x3 ; x4 ; x5
x ; x ; x ; x ; x 7! h x ; x ; x
are C1 ; and
uU^ wV^ ðuU wV Þ1 x1 ; x2 ; x3 ; x4 ; x5
¼ f 1 x 1 ; x 2 ; f 2 x 1 ; x 2 ; h1 x 3 ; x 4 ; x 5 ; h2 x 3 ; x 4 ; x 5 ; h3 x 3 ; x 4 ; x 5 ;
^ V \V
uU^ wV^ ðuU wV Þ1 : ðuU wV Þ U \ U ^
! uU^ wV^ U \ U ^ V \V^
is C 1 :
Similarly, we can prove that ðuU^ wV^ Þ ðuU wV Þ1 is C 2 ; etc. Hence, the
pair ðU V; ðuU wV ÞÞ; ðU ^ V;
^ ðu ^ w ^ ÞÞ of members in C are Cr -compatible,
U V
for every r ¼ 1; 2; 3; . . .: This proves 2.
Hence, by Theorem 1.2, there exists a unique C 1 -differentiable structure D on
M N which contains C: Thus, the ordered pair ðM N; DÞ is an example of a
smooth manifold of dimension ð2 þ 3Þ. Similarly, if M is an m-dimensional
smooth manifold, and N is an n-dimensional smooth manifold, then the Cartesian
product M N; with the product topology, becomes an ðm þ nÞ-dimensional
smooth manifold. In short, we say that the product manifold M N is an ðm þ nÞ-
dimensional smooth manifold.
Chapter 2
Tangent Spaces
Above dimension three, Euclidean space loses its quality of being visible. So, in order
to investigate higher-dimensional smooth manifolds, we need to generalize the
concept of tangent line in the case of curves in two-dimensional Euclidean space, and
tangent plane in the case of surfaces in three-dimensional Euclidean space, using
abstract tools of algebra, and analysis. This generalized concept has been named the
tangent space. Among many other concepts that facilitate the enquiry into the
properties of higher-dimensional smooth manifolds as we shall see later, the notion of
tangent space is foremost. It is possible to introduce tangent space in various ways.
Because each method has its own merit, we shall introduce it in multiple ways in this
chapter for deeper understanding. Remember that in the process of generalization,
unlike curves and surfaces, smooth manifolds are devoid of any ambient space. It was
in this sense a challenging task for early mathematicians.
ðuU ðuV Þ1 Þ is C 1 at the point uV ðpÞ in Rm , and ðf ðuU Þ1 Þ is C1 at the point
uU ðpÞ in Rm , their composite ðf ðuU Þ1 Þ ðuU ðuV Þ1 Þð¼ ðf ðuV Þ1 ÞÞ is
C 1 at the point uV ðpÞ in Rm : Thus, we have shown that ðf ðuV Þ1 Þ : uV ðVÞ ! R
is C 1 at the point uV ðpÞ in Rm : h
Definition Let M be an m-dimensional smooth manifold. Let f : M ! R be any
function. By f is C 1 on M (or f is smooth on M), we mean that f is C 1 at every
point p in M. The set of all smooth functions f : M ! R on M is denoted by
C 1 ðMÞ.
1
ðuU ðuU Þ1 Þ is C 1 at the point u b ðpÞ in Rm , wV ðf ðuU Þ Þ is C 1 at the point
U
uU ðpÞ in Rm , and ðwb ðwV Þ1 Þ is C1 at the point wV ðf ðpÞÞ in Rn , their composite
V
ðwb ðwV Þ1 Þ ðððwV f Þ ðuU Þ1 Þ ðuU ðu b Þ1 ÞÞ ð¼ wb ðf ðu b Þ1 ÞÞ is C 1
V U V U
at the point u b ðpÞ in Rm .
U
b \ f 1 ð V
Thus, we have shown that wb ðf ðu b Þ1 Þ : u b ð U b ÞÞ ! w ð V
b Þ is
V U U b
V
1
C at the point u b ðpÞ in R .m
h
U
vw ðg f Þ ðuU Þ1 : uU U \ ðg f Þ1 ðW Þ ! vw ðW Þ
is C 1 at the point ðuU wV Þðp; qÞð¼ ðuU ðpÞ; wV ðqÞÞÞ in R2þ3 . Let us take any
ðx1 ; x2 ; x3 ; x4 ; x5 Þ in ðuU wV ÞðU VÞ. So, there exist u in U, and v in V such that
ðuU ðuÞ; wV ðvÞÞ ¼ ðuU wV Þðu; vÞ ¼ x1 ; x2 ; x3 ; x4 ; x5 :
Hence,
ðuU wV Þ1 x1 ; x2 ; x3 ; x4 ; x5 ¼ ðu; vÞ;
36 2 Tangent Spaces
or
uU ðuÞ ¼ x1 ; x2 ; wV ðvÞ ¼ x3 ; x4 ; x5 :
So
uU p1 ðuU wV Þ1 x1 ; x2 ; x3 ; x4 ; x5 ¼ ðuU p1 Þ ðuU wV Þ1 x1 ; x2 ; x3 ; x4 ; x5
¼ ðuU p1 Þðu; vÞ ¼ uU ðp1 ðu; vÞÞ ¼ uU ðuÞ
¼ x1 ; x2 :
f ðIdG Þ1 ¼ f IdG ¼ f , and uU ðUÞ ¼ IdG ðUÞ ¼ U ¼ G. Since uU ðUÞ ¼ G; and
f ðuU Þ1 ¼ f , and f ðuU Þ1 : uU ðUÞ ! Rk is smooth in the sense of ordinary
calculus, f : G ! Rk is smooth in the sense of ordinary calculus.
ðf þ gÞð xÞ f ð xÞ þ gð xÞ:
For every f in C1 ðMÞ, and for every real t, we define ðtf Þ : M ! R as follows:
for every x in M,
It is clear that the set C 1 ðMÞ, together with vector addition, and scalar multi-
plication defined as above, constitutes a real linear space. In short, C 1 ðMÞ is a real
linear space.
For every f, g in C 1 ðMÞ, we define ðf gÞ : M ! R as follows: for every x in M,
and
0
RHS ¼ ðwV c1 Þ0 ð0Þ ¼ wV ðuU Þ1 uU c1 ð0Þ
0
¼ wV ðuU Þ1 ðuU c1 Þ ð0Þ
0
¼ wV ðuU Þ1 ððuU c1 Þð0ÞÞ ðuU c1 Þ0 ð0Þ :
where
½½c c1 2 Cp ðM Þ : c1 c :
Intuitively, the Lemma 2.10 says that in Cp ðMÞ we will not distinguish between
c1 and c whenever c1 c.
40 2 Tangent Spaces
uU : ðCp ðMÞ=Þ ! Rm
is the function defined as follows: for every ½½c in Cp ðMÞ= where c is in Cp ðMÞ,
c1 ðtÞ tv þ uU ð pÞ:
Put
For 1: Here,
LHS ¼ cð0Þ ¼ ðuU Þ1 c1 ð0Þ ¼ ðuU Þ1 ðc1 ð0ÞÞ ¼ ðuU Þ1 ð0v þ uU ð pÞÞ
¼ ðuU Þ1 ðuU ð pÞÞ ¼ p ¼ RHS:
This proves 1.
For 2: By the definition of c1 , the function c1 : ð1; 1Þ ! Rm is continuous.
Since c1 : ð1; 1Þ ! Rm is continuous, and 0 is in the open interval ð1; 1Þ, c1
is continuous at 0. Since ðU; uU Þ is a coordinate chart of M, ðuU Þ1 : uU ðUÞ !
U is a 1–1, onto, continuous function, and uU ðUÞ is an open subset of Rm . Since
p 2 U, uU ðpÞ 2 uU ðUÞ. Since c1 ð0Þ ¼ 0v þ uU ðpÞ ¼ uU ðpÞ 2 uU ðUÞ, and
ðuU Þ1 : uU ðUÞ ! U is continuous, ðuU Þ1 : uU ðUÞ ! U is continuous at the
point c1 ð0Þ. Since c1 ð0Þ 2 uU ðUÞ, and uU ðUÞ is an open subset of Rm , uU ðUÞ
is an open neighborhood of c1 ð0Þ. Since c1 : ð1; 1Þ ! Rm is continuous,
uU ðUÞ is an open neighborhood of c1 ð0Þ, there exists d [ 0 such that d\1 and,
for every t in ðd; dÞ, we have c1 ðtÞ 2 uU ðUÞ, and hence, cðtÞ ¼
ððuU Þ1 c1 ÞðtÞ ¼ ðuU Þ1 ðc1 ðtÞÞ 2 U. Since cðtÞ 2 U for every t in ðd; dÞ, it
follows that ðd; dÞ is a subset of the domðcÞ. Hence, 0 is an interior point of the
domain of the function c. This proves 2.
For 3: In 2, we have seen that c is defined over ðd; dÞ. Now, it remains to be
proved that c is a smooth map from ðd; dÞ to M; that is, c is C 1 at every point
t in ðd; dÞ. Here, c is defined over ðd; dÞ, so cðtÞ is meaningful for every t in
ðd; dÞ. Since cðtÞ is meaningful for every t in ðd; dÞ, and
cðtÞ ¼ ðuU Þ1 c1 ðtÞ ¼ ðuU Þ1 ðc1 ðtÞÞ ¼ ðuU Þ1 ðtv þ uU ð pÞÞ;
ðuU Þ1 ðtv þ uU ðpÞÞ is meaningful for every t in ðd; dÞ. Since ðU; uU Þ is a
coordinate chart of M, ðuU Þ1 : uU ðUÞ ! U. Since ðuU Þ1 : uU ðUÞ ! U, and
ðuU Þ1 ðtv þ uU ðpÞÞ is meaningful for every t in ðd; dÞ, ðuU Þ1 ðtv þ uU ðpÞÞ
is in U for every t in ðd; dÞ. Since ðuU Þ1 ðtv þ uU ðpÞÞ is in U for every t in
ðd; dÞ, and cðtÞ ¼ ðuU Þ1 ðtv þ uU ðpÞÞ, cðtÞ is in U for every t in ðd; dÞ:
Thus, c maps ðd; dÞ to U:
Now, it remains to be proved that c is a smooth map from ðd; dÞ to M; that is,
c is C 1 at every point t in ðd; dÞ: For this purpose, let us fix any t in ðd; dÞ:
Since c1 is C1 at t; c ¼ ðuU Þ1 c1 ; c maps ðd; dÞ to U; and t is in ðd; dÞ;
uU c is C 1 at t: Since uU c is C1 at t; and ðU; uU Þ is an admissible coordinate
chart of M satisfying cðtÞ 2 U, by Theorem 2.2, c is C1 at t in ðd; dÞ: This
proves 3. From 1, 2, and 3, we find that c is a parametrized curve in M through p;
that is, c is in Cp ðMÞ.
42 2 Tangent Spaces
Since uU : ðCp ðMÞ=Þ ! Rm is 1–1 onto, ðuU Þ1 : Rm ! ðCp ðMÞ=Þ exists,
and is 1–1 onto. This fact allows us to define a binary operation + over Cp ðMÞ=
as follows:
For every ½½c; ½½c1 in Cp ðMÞ= where c; c1 are in Cp ðMÞ; by ½½c þ ½½c1 , we
mean
Next, for every ½½c in Cp ðMÞ= where c is in Cp ðMÞ; and for every real t; by
t½½c, we mean
Hence, uU ð½½c þ ½½c1 Þ ¼ uU ð½½cÞ þ uU ð½½c1 Þ; and uU ðt½½cÞ ¼ tðuU ð½½cÞÞ:
Now, we want to verify that Cp ðMÞ= is a real linear space:
1. þ is associative: Take any ½½c; ½½c1 ; ½½c2 in Cp ðMÞ= where c; c1 ; c2 are in
Cp ðMÞ: We have to show that ð½½c þ ½½c1 Þ þ ½½c2 ¼ ½½c þ ð½½c1 þ ½½c2 Þ:
LHS ¼ ð½½c þ ½½c1 Þ þ ½½c2 ¼ ðuU Þ1 ðuU ð½½cÞ þ uU ð½½c1 ÞÞ þ ½½c2
¼ ðuU Þ1 uU ðuU Þ1 ðuU ð½½cÞ þ uU ð½½c1 ÞÞ þ uU ð½½c2 Þ
¼ ðuU Þ1 ððuU ð½½cÞ þ uU ð½½c1 ÞÞ þ uU ð½½c2 ÞÞ
¼ ðuU Þ1 ðuU ð½½cÞ þ ðuU ð½½c1 Þ þ uU ð½½c2 ÞÞÞ;
RHS ¼ ½½c þ ð½½c1 þ ½½c2 Þ ¼ ½½c þ ðuU Þ1 ðuU ð½½c1 Þ þ uU ð½½c2 ÞÞ
¼ ðuU Þ1 uU ð½½cÞ þ uU ðuU Þ1 ðuU ð½½c1 Þ þ uU ð½½c2 ÞÞ
¼ ðuU Þ1 ðuU ð½½cÞ þ ðuU ð½½c1 Þ þ uU ð½½c2 ÞÞÞ:
serves the purpose of zero element. For this purpose, let us take any ½½c in
Cp ðMÞ= where c is in Cp ðMÞ: We have to show that
5. Take any ½½c; ½½c1 in Cp ðMÞ= where c; c1 are in Cp ðMÞ: Take any real number t:
We have to prove that
LHS ¼ tð½½c þ ½½c1 Þ ¼ ðuU Þ1 ðtðuU ð½½c þ ½½c1 ÞÞÞ
¼ ðuU Þ1 t uU ðuU Þ1 ðuU ð½½cÞ þ uU ð½½c1 ÞÞ
¼ ðuU Þ1 ðtðuU ð½½cÞ þ uU ð½½c1 ÞÞÞ;
RHS ¼ t½½c þ t½½c1 ¼ ðuU Þ1 ðtðuU ð½½cÞÞÞ þ ðuU Þ1 t uU ½½c1 :
44 2 Tangent Spaces
ðuU Þ1 ðtðuU ð½½cÞ þ uU ð½½c1 ÞÞÞ ¼ ðuU Þ1 ðtðuU ð½½cÞÞÞ
þ ðuU Þ1 ðtðuU ð½½c1 ÞÞÞ
that is,
tðuU ð½½cÞ þ uU ð½½c1 ÞÞ ¼ uU ðuU Þ1 ðtðuU ð½½cÞÞÞ þ ðuU Þ1 ðtðuU ð½½c1 ÞÞÞ :
RHS ¼ uU ðuU Þ1 ðtðuU ð½½cÞÞÞ þ ðuU Þ1 ðtðuU ð½½c1 ÞÞÞ
¼ uU ðuU Þ1 ðtðuU ð½½cÞÞÞ þ uU ðuU Þ1 ðtðuU ð½½c1 ÞÞÞ
¼ tðuU ð½½cÞÞ þ tðuU ð½½c1 ÞÞ
¼ tðuU ð½½cÞ þ uU ð½½c1 ÞÞ ¼ LHS:
6. Take any ½½c in Cp ðMÞ= where c is in Cp ðMÞ: Take any real numbers s; t: We
have to prove that
ðuU Þ1 ðsðuU ð½½cÞÞ þ tðuU ð½½cÞÞÞ ¼ ðuU Þ1 ðuU ðs½½cÞ þ uU ðt½½cÞÞ;
that is,
7. Take any ½½c in Cp ðMÞ= where c is in Cp ðMÞ: Take any real numbers s; t: We
have to prove that
ðstÞ½½c ¼ sðt½½cÞ:
2.2 Algebra of Smooth Functions 45
RHS ¼ sðt½½cÞ ¼ ðuU Þ1 ðsðuU ðt½½cÞÞÞ ¼ ðuU Þ1 ðsðtðuU ð½½cÞÞÞÞ
¼ ðuU Þ1 ððstÞðuU ð½½cÞÞÞ:
1½½c ¼ ½½c:
LHS ¼ 1½½c ¼ ðuU Þ1 ð1ðuU ð½½cÞÞÞ ¼ ðuU Þ1 ðuU ð½½cÞÞ ¼ ½½c ¼ RHS: h
uU ð½½c þ ½½c1 Þ ¼ uU ð½½cÞ þ uU ð½½c1 Þ; and uU ðt½½cÞ ¼ tðuU ð½½cÞÞ;
so uU is an isomorphism between real linear space Cp ðMÞ= and the real linear
space Rm : Further since the dimension of Rm is m; Cp ðMÞ= is a real linear space
of dimension m:
Conclusion: Let M be an m-dimensional smooth manifold. Let p be an element
of M: Then, Cp ðMÞ= is isomorphic to Rm for some vector addition, and scalar
multiplication over Cp ðMÞ=:
Definition Let M be an m-dimensional smooth manifold. Let p be an element of M:
The set Cp ðMÞ= is denoted by Tp M or Tp ðMÞ (see Fig. 2.1).
Note 2.14 Let M be a 2-dimensional smooth manifold. Let p 2 M: Let ðU; uU Þ be
any admissible coordinate chart of M satisfying p 2 U: Let us define functions
u1 : U ! R; u2 : U ! R
uU ð xÞ u1 ð xÞ; u2 ð xÞ :
(In short, we say that u1 ; u2 are the component functions of uU :) We shall try to
prove that u1 is in Cp1 ðMÞ: By the definition of Cp1 ðMÞ; it remains to be proved that
for every admissible coordinate chart ðV; uV Þ of M satisfying p 2 V,
u1 ðuV Þ1 : uV ðU \ V Þ ! R
So ðu1 ðuV Þ1 Þ; ðu2 ðuV Þ1 Þ are the component functions of ðuU ðuV Þ1 Þ:
Next, since ðuU ðuV Þ1 Þ is C 1 at the point uV ðpÞ in R2 ; its component functions
ðu1 ðuV Þ1 Þ; ðu2 ðuV Þ1 Þ are C 1 at the point uV ðpÞ in R2 : This proves that u1 is
in Cp1 ðMÞ: Similarly, u2 is in Cp1 ðMÞ:
Theorem 2.15 Let M be an m-dimensional smooth manifold. Let p be an element
of M:
Let be a relation over Cp1 ðMÞ defined as follows: for every f and g in
1
Cp ðMÞ; by f g; we shall mean that there exists an open neighborhood H of p
such that f ðxÞ ¼ gðxÞ for every x in H: Then, is an equivalence relation over
Cp1 ðMÞ:
Proof Here, we must prove
1. for every f in Cp1 ðMÞ; f f ;
2. if f g, then g f ;
3. if f g; and g h, then f h:
2.2 Algebra of Smooth Functions 47
For 1. Let us take any f in Cp1 ðMÞ: So, by the definition of Cp1 ðMÞ; dom f is an
open neighborhood of p in M: Since dom f is an open neighborhood of p in M;
and f ðxÞ ¼ f ðxÞ for every x in dom f , by the definition of ; f f : This proves 1.
For 2. Let f g where f and g are in Cp1 ðMÞ: So, by the definition of ; there
exists an open neighborhood H of p such that f ðxÞ ¼ gðxÞ for every x in H:
Hence, gðxÞ ¼ f ðxÞ for every x in H; where H is an open neighborhood of p:
Therefore, by the definition of , g f , this proves 2.
For 3. Let f g; and g h; where f ; g; h are in Cp1 ðMÞ: Since f g, by the
definition of ; there exists an open neighborhood H of p such that f ðxÞ ¼ gðxÞ
for every x in H: Similarly, there exists an open neighborhood K of p such that
gðxÞ ¼ hðxÞ for every x in K: Since H; K are open neighborhoods of p; H \ K is
an open neighborhood of p: Also we have f ðxÞ ¼ hðxÞ for every H \ K: Hence,
by the definition of ; f h: This proves 3.
Hence, is an equivalence relation over Cp1 ðMÞ: h
Definition Let M be an m-dimensional smooth manifold. Let p be an element of M:
From Theorem 2.15, the relation is an equivalence relation over Cp1 ðMÞ: The
1
quotient set Cp ðMÞ= ; of all equivalence classes is denoted by F p ðMÞ (or, simply
F p Þ: Thus,
n o
F p ðM Þ ½ f : f 2 Cp1 ðM Þ
where
n o
½ f g : g 2 Cp1 ðM Þ and g f :
Intuitively, Theorem 2.15 says that in Cp1 ðMÞ we will not distinguish between f
and g whenever f g:
Definition Let M be an m-dimensional smooth manifold. Let p be an element of M:
Let f ; g be in Cp1 ðMÞ: By the sum f þ g, we mean the function whose domain is
ðdom f Þ \ ðdom gÞ; and for every x in ðdom f Þ \ ðdom gÞ;
ðf þ gÞð xÞ f ð xÞ þ gð xÞ:
By the product f g, we mean the function whose domain is ðdom f Þ \ðdom gÞ;
and for every x in ðdom f Þ \ðdom gÞ;
ðf gÞð xÞ f ð xÞ gð xÞ:
For any real t; by the scalar multiple tf , we mean the function whose domain is
dom f ; and for every x in dom f ;
48 2 Tangent Spaces
and
g ðuU Þ1 : uU ððdom gÞ \ U Þ ! R;
and, for every x in dom ðf þ gÞ ðuU Þ1 ;
f ðuU Þ1 þ g ðuU Þ1 ð xÞ ¼ f ðuU Þ1 ð xÞ þ g ðuU Þ1 ð xÞ
¼ f ðuU Þ1 ð xÞ þ g ðuU Þ1 ð xÞ
¼ ðf þ gÞ ðuU Þ1 ð xÞ
¼ ðf þ gÞ ðuU Þ1 ð xÞ
so,
f ðuU Þ1 þ g ðuU Þ1 ¼ ðf þ gÞ ðuU Þ1 :
Since ðf ðuU Þ1 Þ þ ðg ðuU Þ1 Þ ¼ ðf þ gÞ ðuU Þ1 ; and ðf ðuU Þ1 Þ þ
ðg ðuU Þ1 Þ is C 1 at the point uU ðpÞ in Rm ; ðf þ gÞ uU Þ1 is C1 at the point
uU ðpÞ in Rm : This proves 2. Hence, f þ g is in Cp1 ðMÞ: h
½f þ ½g ½f þ g;
t½f ½tf ;
½f ½g ½f g:
Let us define multiplication operation over Cp1 ðMÞ= as follows: for every f ; g
in Cp1 ðMÞ;
½f ½g ½f g:
2.2 Algebra of Smooth Functions 51
Then,
1. for every f ; g; h in Cp1 ðMÞ; ð½f ½gÞ½h ¼ ½f ð½g½hÞ;
2. for every f ; g in Cp1 ðMÞ; ½f ½g ¼ ½g½f ;
3. for every f ; g; h in Cp1 ðMÞ; ð½f þ ½gÞ½h ¼ ½f ½h þ ½g½h;
4. for every real t; and for every f ; g in Cp1 ðMÞ; tð½f ½gÞ ¼ ðt½f Þ½g:
In short, F p ðMÞ is an algebra.
Proof The conditions for linear space remain to be verified.
1. + is associative: Let us take any f ; g; h in Cp1 ðMÞ: We have to prove that
This proves 2.
52 2 Tangent Spaces
3. Existence of negative element: Let us take any f in Cp1 ðMÞ: So, by the definition
of Cp1 ðMÞ; f is a real-valued function f whose dom f is an open neighborhood
of p in M; and for every admissible coordinate chart ðU; uU Þ of M satisfying
p 2 U,
f ðuU Þ1 : uU ððdom f Þ \ U Þ ! R
Thus, we have shown that ðf Þ is in Cp1 ðMÞ: Now, it remains to be showed
that ½ðf Þ þ ½f ¼ ½0 ¼ ½f þ ½ðf Þ. Here,
This proves 3.
4. + is commutative: Let us take any f ; g in Cp1 ðMÞ: We have to prove that
½ f þ ½g ¼ ½g þ ½ f :
2.2 Algebra of Smooth Functions 53
Here
This proves 4.
5. For every real s; t; and f in Cp1 ðMÞ; ðs þ tÞ½f ¼ s½f þ t½f :
Here
7. For every real t; and for every f ; g in Cp1 ðMÞ; tð½f þ ½gÞ ¼ t½f þ t½g :
Here,
LHS ¼ tð½ f þ ½gÞ ¼ t½ðf þ gÞ ¼ ½tðf þ gÞ ¼ ½ðtf Þ þ ðtgÞ
¼ ½ðtf Þ þ ½ðtgÞ ¼ t½ f þ t½g ¼ RHS:
Thus, we have shown that quotient set Cp1 ðMÞ= is a real linear space.
Now, we want to prove that F p ðMÞ is an algebra.
1. Multiplication is associative:
2. Multiplication is commutative:
4. LHS ¼ tð½f ½gÞ ¼ t½f g ¼ ½tðf gÞ ¼ ½ðtf Þ g ¼ ½tf ½g ¼ ðt½f Þ½g ¼ RHS: h
54 2 Tangent Spaces
Proof
1: Since c is in Cp ðMÞ, by the definition of Cp ðMÞ; there exists a real number
d [ 0 such that c : ðd; dÞ ! M, cð0Þ ¼ p; and c is a smooth map from
ðd; dÞ to M: Since cis a smooth map from ðd; dÞ to M; c is a continuous
map.
Since f is in Cp1 ðMÞ; f : ðdom f Þ ! R and dom f is an open neighborhood of p
in M: Since cð0Þ ¼ p; and p is in dom f ; 0 is an element of the domain of f c:
Since f is in Cp1 ðMÞ; f is continuous on some open neighborhood U of p. Since
c : ðd; dÞ ! M is a continuous map, cð0Þ ¼ p; and U is an open neighborhood of
p; there exists e [ 0 such that e\d and, for every t in ðe; eÞ, we have cðtÞ 2 U
ð ðdom f ÞÞ; and hence, ðf cÞðtÞ ¼ f ðcðtÞÞ 2 R: It follows that ðe; eÞ is a subset
of domðf cÞ: Hence, 0 is an interior point of the domain of the function f c: This
proves 1.
2: Since 0 is an interior point of the domain of the real-valued function f c; it is
meaningful to write
ð f c Þ ð 0 þ t Þ ð f c Þ ð 0Þ
lim ;
t!0 t
is C1 at the point uU ðpÞ in Rm : Observe that the domain of ðf ðuU Þ1 Þ; that is,
uU ððdom f Þ \ UÞ is an open neighborhood of uU ðpÞ ð¼ uU ðcð0ÞÞ ¼ ðuU cÞð0ÞÞ
in Rm : Since c is a smooth map from ðd; dÞ to M; and 0 2 ðd; dÞ; c is C 1 at 0 in
R: Since c is C1 at 0 in R; cð0Þ ¼ p; and ðU; uU Þ is an admissible coordinate chart
of M satisfying p 2 U; ðuU cÞ : ðd; dÞ ! Rm is C1 at the point 0 in R: Since
ðuU cÞ : ðd; dÞ ! Rm
is C 1 at the point 0;
f ðuU Þ1 : uU ððdom f Þ \ U Þ ! R
is C 1 at the point ðuU cÞð0Þ in Rm ; and the domain of ðf ðuU Þ1 Þ is an open
neighborhood of ðuU cÞð0Þ in Rm ; the composite function ðf ðuU Þ1 Þ
ðuU cÞð¼ f cÞ is C 1 at the point 0 in R: Hence,
dðf cÞðtÞ
dt t¼0
that is,
ð f c Þ ð 0 þ t Þ ð f c Þ ð 0Þ
lim
t!0 t
Since f g, there exists an open neighborhood H of pð¼ cð0ÞÞ such that f ðxÞ ¼
gðxÞ for every x in H: Since f ðxÞ ¼ gðxÞ for every x in H; and cð0Þ is in H;
Clearly, e [ 0: If t is in the open interval ðe; eÞ; then t is in the open interval
ðe2 ; e2 Þ; and hence, ðf cÞðtÞ is meaningful. Similarly, if t is in the open interval
ðe; eÞ; then ðg cÞðtÞ is meaningful. If t is in the open interval ðe; eÞ; then t is in
the open interval ðe1 ; e1 Þ; and hence, cðtÞ is in H: Since, for every t in ðe; eÞ; cðtÞ
is in H; and since f ðxÞ ¼ gðxÞ for every x in H; for every t in ðe; eÞ;
Hence,
ð f c Þ ð 0 þ t Þ ð f c Þ ð 0Þ
lim ¼
t!0 t
ðf cÞðtÞ ðf cÞð0Þ ðf cÞðtÞ ðf cÞð0Þ
lim ¼ lim
t!0 t t ! 0; t
t in ðe; eÞ
ð g c Þ ð t Þ ð g c Þ ð 0Þ
¼ lim
t ! 0; t
t in ðe; eÞ
ð g c Þ ð 0 þ t Þ ð g c Þ ð 0Þ
¼ lim
t ! 0; t
t in ðe; eÞ
ðg cÞð0 þ tÞ ðg cÞð0Þ
¼ lim :
t!0 t
This proves 3. h
Note 2.25 Let M be an m-dimensional smooth manifold. Let p be an element of M:
Let c be in Cp ðMÞ: Let ½f be in F p ðMÞ where f is in Cp1 ðMÞ: Let ½g be in F p ðMÞ
where g is in Cp1 ðMÞ: If ½f ¼ ½g, then f g; and hence,
2.3 Smooth Germs on Smooth Manifolds 57
ð f c Þ ð 0 þ t Þ ð f c Þ ð 0Þ
lim
t!0 t
or,
dð f c Þ ð t Þ
c; ½ f :
dt t¼0
2. Here
Note 2.27 Let M be an m-dimensional smooth manifold. Let p 2 M: Let ½½c; ½½c1
be in Tp Mð¼ Cp ðMÞ=Þ where c; c1 are in Cp ðMÞ: Let ½f ; ½f1 be in F p ðMÞ where
f ; f1 are in Cp1 ðMÞ: Let ½½c ¼ ½½c1 ; and ½f ¼ ½f1 : We shall try to show that
dð f c Þ ð t Þ dðf1 c1 ÞðtÞ
¼ :
dt t¼0 dt t¼0
Proof
1: LHS ¼ ½½cð½f þ ½f1 Þ ¼ ½½cð½f þ f1 Þ ¼ c; ½f þ f1 ¼ c þ ½f1
¼ c; ½f þ c; ½f1 ¼ c½f þ c½f1 ¼ RHS:
2: LHS ¼ ½½cðt½f Þ ¼ ½½cð½tf Þ ¼ c; t½f ¼ t c; ½f ¼ tð½½c½f Þ ¼ RHS:
dððf f1 Þ cÞðtÞ dððf f1 ÞðcðtÞÞÞ
3: LHS ¼ ½½cð½f ½f1 Þ ¼ ½½cð½f f1 Þ ¼ jt¼0 ¼ jt¼0
dt dt
dððf ðcðtÞÞ f1 ðcðtÞÞÞÞ dððððf cÞðtÞÞ ððf1 cÞðtÞÞÞÞ
¼ jt¼0 ¼ jt¼0
dt dt
dððf cÞðtÞÞ dððf1 cÞðtÞÞ h
¼ jt¼0 ððf1 cÞð0ÞÞ þ ððf cÞð0ÞÞ jt¼0
dt dt
¼ ð½½c½f Þððf1 cÞð0ÞÞ þ ððf cÞð0ÞÞð½½c½f Þ
¼ ð½½c½f Þððf1 ðcð0ÞÞÞÞ þ ðf ðcð0ÞÞÞð½½c½f1 Þ
¼ ð½½c½f Þððf1 ðpÞÞÞ þ ðf ðpÞÞð½½c½f1 Þ ¼ RHS: h
2.4 Derivations
D½ c ð½ f Þ ½½c½ f
3. ðtDÞð½f ½f1 Þ ¼ ððtDÞ½f Þðf1 ðpÞÞ þ ðf ðpÞÞððtDÞ½f1 Þ for every ½f ; ½f1 in F p ðMÞ
where f ; f1 are in Cp1 ðMÞ:
For 1: LHS ¼ ðtDÞð½f þ ½f1 Þ ¼ tðDð½f þ ½f1 ÞÞ
¼ tðD½f þ D½f1 Þ ¼ tðD½f Þ þ tðD½f1 Þ ¼ ðtDÞ½f þ ðtDÞ½f1 ¼ RHS:
For 2: LHS ¼ ðtDÞðs½f Þ ¼ tðDðs½f ÞÞ ¼ tðsðD½f ÞÞ
¼ sðtðD½f ÞÞ ¼ sððtDÞ½f Þ ¼ RHS:
For 3: LHS ¼ ðtDÞð½f ½f1 Þ ¼ tðDð½f ½f1 ÞÞ ¼ tððD½f Þðf1 ðpÞÞ þ ðf ðpÞÞðD½f1 ÞÞ
¼ tðD½f Þðf1 ðpÞÞ þ ðf ðpÞÞtðD½f1 Þ
¼ ððtDÞ½f Þðf1 ðpÞÞ þ ðf ðpÞÞððtDÞ½f1 Þ ¼ RHS:
This proves that tD is in Dp ðMÞ:
Now, we shall try to verify that Dp ðMÞ is a real linear space.
1. + is associative: Let us take any D; D1 ; D2 in Dp ðMÞ: We have to prove that
ðD þ D1 Þ þ D2 ¼ D þ ðD1 þ D2 Þ; that is, for every ½f in F p ðMÞ where f is in
Cp1 ðMÞ; ððD þ D1 Þ þ D2 Þ½f ¼ ðD þ ðD1 þ D2 ÞÞ½f :
LHS ¼ ððD þ D1 Þ þ D2 Þ½ f ¼ ðD þ D1 Þ½ f þ D2 ½ f ¼ ðD½ f þ D1 ½ f Þ þ D2 ½ f
D½ f þ ðD1 ½ f þ D2 ½ f Þ ¼ D½ f þ ðD1 þ D2 Þ½ f ¼ ðD þ ðD1 þ D2 ÞÞ½ f ¼ RHS:
2. Existence of zero element: Let us define the constant function 0 : F p ðMÞ ! R
as follows: for every ½f in F p ðMÞ; 0½f 0: We want to prove that 0 is in
Dp ðMÞ: For this purpose, we must prove
1. 0ð½f þ ½f1 Þ ¼ 0½f þ 0½f1 for every ½f ; ½f1 in F p ðMÞ where f ; f1 are in
Cp1 ðMÞ;
2. 0ðt½f Þ ¼ tð0½f Þ for every ½f in F p ðMÞ where f is in Cp1 ðMÞ; and for every
real t;
3. 0ð½f ½f1 Þ ¼ ð0½f Þðf1 ðpÞÞ þ ðf ðpÞÞð0½f1 Þ; for every ½f ; ½f1 in F p ðMÞ where f ;
f1 are in Cp1 ðMÞ:
For 1: LHS ¼ 0ð½f þ ½f1 Þ ¼ 0 ¼ 0 þ 0 ¼ 0½f þ 0½f1 ¼ RHS:
For 2: LHS ¼ 0ðt½f Þ ¼ 0 ¼ t 0 ¼ tð0½f Þ ¼ RHS:
For 3: LHS ¼ 0ð½f ½f1 Þ ¼ 0ð½f f1 Þ ¼ 0 ¼ 0ðf1 ðpÞÞ þ ðf ðpÞÞ0 ¼ ð0½f Þðf1 ðpÞÞ þ
ðf ðpÞÞð0½f1 Þ ¼ RHS:
Thus, we have shown that 0 is in Dp ðMÞ: Now, it remains to be showed that
0 þ D ¼ D for every D in Dp ðMÞ; that is, for every ½f in F p ðMÞ where f is in
Cp1 ðMÞ; ð0 þ DÞ½f ¼ D½f :
LHS ¼ ð0 þ DÞ½ f ¼ 0½ f þ D½ f ¼ 0 þ D½ f ¼ D½ f ¼ RHS:
Hence, 0 serves the purpose of zero element in Dp ðMÞ:
3. Existence of negative element: Let us take any D in Dp ðMÞ: Now, let us define
ðDÞ : F p ðMÞ ! R as follows: For every ½f in F p ðMÞ; ðDÞ½f ðD½f Þ:
We want to prove that ðDÞ is in Dp ðMÞ: For this purpose, we must prove
62 2 Tangent Spaces
Thus, we have shown that ðDÞ is in Dp ðMÞ: Now, it remains to be showed that
ðDÞ þ D ¼ 0; that is, for every ½f in F p ðMÞ where f is in Cp1 ðMÞ;
ððDÞ þ DÞ½f ¼ 0½f :
This proves 4.
5. For every real s; t; and D in Dp ðMÞ; we have to prove that ðs þ tÞD ¼ sD þ tD;
that is, for every ½f in F p ðMÞ where f is in Cp1 ðMÞ; ððs þ tÞDÞ½f ¼
ðsD þ tDÞ½f : Here
LHS ¼ ððs þ tÞDÞ½ f ¼ ðs þ tÞðD½ f Þ ¼ sðD½ f Þ þ tðD½ f Þ ¼ ðsDÞ½ f þ ðtDÞ½ f
¼ ðsD þ tDÞ½ f ¼ RHS:
6. For every real s; t; and D in Dp ðMÞ; we have to prove that ðstÞD ¼ sðtDÞ; that is,
for every ½f in F p ðMÞ where f is in Cp1 ðMÞ; ððstÞDÞ½f ¼ ðsðtDÞÞ½f : Here,
LHS ¼ ððstÞDÞ½ f ¼ ðstÞðD½ f Þ ¼ sðtðD½ f ÞÞ ¼ sððtDÞ½ f Þ ¼ ðsðtDÞÞ½ f ¼ RHS:
2.4 Derivations 63
7. For every real t; and for every D; D1 in Dp ðMÞ; we have to prove that
tðD þ D1 Þ ¼ tD þ tD1 ; that is, for every ½f in F p ðMÞ where f is in Cp1 ðMÞ;
ðtðD þ D1 ÞÞ½f ¼ ðtD þ tD1 Þ½f : Here
8. For every D in Dp ðMÞ; we have to show that 1D ¼ D; that is, for every ½f in
F p ðMÞ where f is in Cp1 ðMÞ; ð1DÞ½f ¼ D½f : Here,
Dð1Þ ¼ 0: Now,
u1 : U ! R; u2 : U ! R;
Thus,
D1 u1 ðuU Þ1 ðuU ð pÞÞ ¼ 1:
2.4 Derivations 65
Next
D2 u1 ðuU Þ1 ðuU ð pÞÞ
¼ D2 u1 ðuU Þ1 u1 ð pÞ; u2 ð pÞ
u1 ðuU Þ1 ðu1 ð pÞ; u2 ð pÞ þ tÞ u1 ðuU Þ1 ðu1 ð pÞ; u2 ð pÞÞ
¼ lim
t!0
t
1
u ðuU Þ
1
ððu ð pÞ; u ð pÞÞ þ ð0; tÞÞ u1 ðuU Þ1 ðu1 ð pÞ; u2 ð pÞÞ
1 2
¼ lim
t
t!0
1
u ðuU Þ
1
ðuU ð pÞ þ ð0; tÞÞ u1 ðuU Þ1 ðuU ð pÞÞ
¼ lim
t!0
t
1
u ðuU Þ
1
ðuU ð pÞ þ ð0; tÞÞ u1 ð pÞ
¼ lim
t!0
t
1
u ðuU Þ ðuU ð pÞ þ ð0; tÞÞ u1 ð pÞ
1
¼ lim
t!0 t
u1 ð qÞ u1 ð pÞ
¼ lim
t!0 t
Thus,
D2 u1 ðuU Þ1 ðuU ð pÞÞ ¼ 0:
Similarly, ðD2 ðu2 ðuU Þ1 ÞÞðuU ðpÞÞ ¼ 1; and ðD1 ðu2 ðuU Þ1 ÞÞðuU ðpÞÞ ¼ 0:
Definition Let M be a 2-dimensional smooth manifold. Let p be an element of M:
Let ðU; uU Þ be any admissible coordinate chart of M satisfying p 2 U: Let u1 ; u2 be
the component functions of uU : The function ouo1 jp : F p ðMÞ ! R is defined as
follows: For every ½f in F p ðMÞ;
66 2 Tangent Spaces
!
o
½ f D1 f ðuU Þ1 ðuU ð pÞÞ:
ou1 p
o
The function ou2 jp : F p ðMÞ ! R is defined as follows: For every ½f in F p ðMÞ;
!
o
1
½ f D 2 f ð u Þ ðuU ð pÞÞ:
ou2 p U
o
Note 2.33 We shall prove that ou1 jp is in Dp ðMÞ: For this purpose, we must prove
1. ðouo1 jp Þð½f þ ½f1 Þ ¼ ðouo1 jp Þ½f þ ðouo1 jp Þ½f1 for every ½f ; ½f1 in F p ðMÞ where f ; f1
are in Cp1 ðMÞ;
2. ðouo1 jp Þðt½f Þ ¼ tððouo1 jp Þ½f Þ for every ½f in F p ðMÞ where f is in Cp1 ðMÞ; and for
every real t;
3. ðouo1 jp Þð½f ½f1 Þ ¼ ððouo1 jp Þ½f Þðf1 ðpÞÞ þ ðf ðpÞÞððouo1 jp Þ½f1 Þ; for every ½f ; ½f1 in
F p ðMÞ where f ; f1 are in Cp1 ðMÞ
! !
o o
For 1: LHS ¼ ð ½ f þ ½ f 1 Þ ¼ ð½f þ f1 Þ ¼ D1 ðf þ f1 Þ ðuU Þ1 ðuU ð pÞÞ
ou p
1 ou p
1
1
¼ D1 f ðuU Þ þ f1 ðuU Þ1 ðuU ð pÞÞ ¼ D1 f ðuU Þ1 ðuU ð pÞÞ
! !
o o
1
þ D1 f1 ðuU Þ ðuU ð pÞÞ ¼ ½f þ ½f1 ¼ RHS:
ou1 p ou1 p
!
o
For 2: LHS ¼ ðt½ f Þ ¼ D1 ðtf Þ ðuU Þ1 ðuU ð pÞÞ ¼ D1 t f ðuU Þ1 ðuU ð pÞÞ
ou1 p
! !
o
¼ t D1 f ðuU Þ1 ðuU ð pÞÞ ¼ t ½f ¼ RHS:
ou1
! !
p
o o
For 3: LHS ¼ ou1 ð½ f ½f1 Þ ¼ ð½f f1 Þ ¼ D1 ðf f1 Þ ðuU Þ1 ðuU ð pÞÞ
p ou p1
1
¼ D1 f ðuU Þ f1 ðuU Þ1 ðuU ð pÞÞ
1
¼ D1 f ðuU Þ ðuU ð pÞÞ f1 ðuU Þ1 ðuU ð pÞÞ
þ f ðuU Þ1 ðuU ð pÞÞ D1 f1 ðuU Þ1 ðuU ð pÞÞ
¼ D1 f ðuU Þ1 ðuU ð pÞÞ ðf1 ð pÞÞ þ ðf ð pÞÞ D1 f1 ðuU Þ1 ðuU ð pÞÞ
! ! ! !
o o
¼ ½ f ð f1 ð p Þ Þ þ ð f ð p Þ Þ ½f1 ¼ RHS:
ou1 p ou1 p
2.4 Derivations 67
o o
This proves that ou1 jp is in Dp ðMÞ: Similarly, ou2 jp is in Dp ðMÞ:
Note 2.34 We shall prove that ouo1 jp ; ouo2 jp are linearly independent. For this purpose,
let t1 ðouo1 jp Þ þ t2 ðouo2 jp Þ ¼ 0: We have to show that t1 ¼ t2 ¼ 0: Since t1 ðouo1 jp Þ þ
t2 ðouo2 jp Þ ¼ 0; so
! !! ! ! ! !
o o o o
0¼0 u 1
¼ t1 þ t2 u 1
¼ t1 u 1
þ t2 u 1
ou1 p ou2 p ou1 p ou2 p
¼ t1 ð1Þ þ t2 ð0Þ ¼ t1 :
Thus, t1 ¼ 0: Similarly, t2 ¼ 0: Thus, we have shown that ouo1 jp ; ouo2 jp are linearly
independent.
Lemma 2.35 Let M be a 2-dimensional smooth manifold. Let p be an element of
M: Let ðU; uU Þ be any admissible coordinate chart of M satisfying p 2 U: Let us
define functions
u1 : U ! R; u2 : U ! R
If uU ðpÞ ¼ ð0; 0Þ, then fouo1 jp ; ouo2 jp g is a basis of the real linear space Dp ðMÞ:
Proof From the above discussion, it remains to be proved that fouo1 jp ; ouo2 jp g gen-
erates the whole space Dp ðMÞ: For this purpose, let us take any D in Dp ðMÞ: We
have to find real numbers t1 ; t2 such that D ¼ t1 ðouo1 jp Þ þ t2 ðouo2 jp Þ; that is, for every
½f in F p ðMÞ where f is in Cp1 ðMÞ; we have
! !!
o o
D ½ f ¼ t1 þ t2 ½ f :
ou1 p ou2 p
Let us take any ½f in F p ðMÞ where f is in Cp1 ðMÞ: It suffices to prove that
! !!
o o
D½ f ¼ D u1 þ D u2 ½ f ;
ou1 p ou2 p
that is,
! ! ! !
o o
D½ f ¼ D u1 ½f þ D u 2
½f :
ou1 p ou2 p
Let us note that for every uU ðqÞð¼ ðu1 ðqÞ; u2 ðqÞÞÞ in uU ðUÞ; where q is in U;
68 2 Tangent Spaces
f ðqÞ f ð pÞ ¼ f ðqÞ f ðuU Þ1 ð0; 0Þ
¼ f ðuU Þ1 u1 ðqÞ; u2 ðqÞ f ðuU Þ1 ð0; 0Þ
¼ f ðuU Þ1 u1 ðqÞ; u2 ðqÞ f ðuU Þ1 ð0; 0Þ
t¼1
¼ f ðuU Þ1 u1 ðqÞ t; u2 ðqÞ t
t¼0
Z1
d
¼ f ðuU Þ1 u1 ðqÞ t; u2 ðqÞ t dt
dt
0
Z 1
d 1
¼ D1 f ðuU Þ1 u1 ðqÞ t; u2 ðqÞ t u ð qÞ t
dt
0
d 2
þ D2 f ðuU Þ1 u1 ðqÞ t; u2 ðqÞ t u ðqÞ t dt
dt
Z 1 1
¼ D1 f ðuU Þ1 u1 ðqÞ t; u2 ðqÞ t u ð qÞ
0
2
þ D2 f ðuU Þ1 u1 ðqÞ t; u2 ðqÞ t u ðqÞ dt
Z 1
¼ u1 ðqÞ D1 f ðuU Þ1 u1 ðqÞ t; u2 ðqÞ t dt
0
Z1
þ u2 ðqÞ D2 f ðuU Þ1 u1 ðqÞ t; u2 ðqÞ t dt
0
Z 1
¼ u1 ðqÞ D1 f ðuU Þ1 t u1 ðqÞ; u2 ðqÞ dt
0
Z1
þ u ðqÞ 2
D2 f ðuU Þ1 t u1 ðqÞ; u2 ðqÞ dt
0
Z 1
¼ u1 ðqÞ D1 f ðuU Þ1 ðtðuU ðqÞÞÞ dt
0
Z1
þ u2 ðqÞ D2 f ðuU Þ1 ðtðuU ðqÞÞÞ dt
0
Z1 ! !
o
¼ u ðqÞ 1
½ f dt
ou1 ðuU Þ1 ðtðuU ðqÞÞÞ
0
Z1 ! !
o
þ u ðqÞ 2
½ f dt
ou2 ðuU Þ1 ðtðuU ðqÞÞÞ
0
¼ u1 ðqÞ ðg1 ðqÞÞ þ u2 ðqÞ ðg2 ðqÞÞ;
2.4 Derivations 69
where
Z1 ! !
o
g1 ð qÞ ½ f dt;
ou1 ðuU Þ1 ðtðuU ðqÞÞÞ
0
and
Z1 ! !
o
g2 ðqÞ ½ f dt
ou2 ðuU Þ1 ðtðuU ðqÞÞÞ
0
for every q in U: So
f ¼ u1 g1 þ u2 g2 þ f ð pÞ:
Next
D½ f ¼ D f ð pÞ þ u1 g1 þ u2 g2 ¼ D½f ð pÞ þ D u1 g1 þ D u2 g2
¼ 0 þ D u1 g1 þ D u2 g2 ¼ D u1 g1 þ D u2 g2
¼ D u1 ðg1 ð pÞÞ þ u1 ð pÞ ðD½g1 Þ þ D u2 ðg2 ð pÞÞ þ u2 ð pÞ ðD½g2 Þ
¼ D u1 ðg1 ð pÞÞ þ ð0ÞðD½g1 Þ þ D u2 ðg2 ð pÞÞ þ ð0ÞðD½g2 Þ
¼ D u1 ðg1 ð pÞÞ þ D u2 ðg2 ð pÞÞ
0 1 ! ! 1
Z
1 @ o
¼ D u ½ f dtA
ou1 ðuU Þ1 ðtðuU ð pÞÞÞ
0
0 1 ! ! 1
Z
2 @ o
þ D u ½ f dtA
ou2 ðuU Þ1 ðtðuU ð pÞÞÞ
0
0 1 ! ! 1
Z
o
¼ D u1 @ ½ f dtA
ou1 ðuU Þ1 ðtð0;0ÞÞ
0
0 1 ! ! 1
Z
o
þ D u2 @ ½ f dtA
ou2 ðuU Þ1 ðtð0;0ÞÞ
0
70 2 Tangent Spaces
0 ! ! 1
Z1
o
¼ D u1 @ ½ f dtA
ou1 ðuU Þ1 ð0;0Þ
0
0 ! ! 1
Z1
o
þ D u2 @ ½ f dtA
ou2 ðuU Þ1 ð0;0Þ
0
0 ! ! 1 0 1 ! ! 1
Z1 Z
o o
¼ D u1 @ ½ f dtA þ D u2 @ ½ f dtA
ou1 p ou2 p
0 0
0 ! ! Z1 1 0 ! ! Z1 1
o o
¼ D u1 @ ½f 1dtA þ D u2 @ ½f 1dtA
ou1 p ou2 p
0 0
! ! ! !
o o
¼ D u1 ½ f 1 þ D u2 ½f 1
ou1 p ou2 p
! ! ! !
o o
¼ D u1 ½f þ D u 2
½ f ¼ RHS: h
ou1 p ou2 p
h
Lemma 2.36 Let M be a 2-dimensional smooth manifold. Let p be an element of
M: Then, Dp ðMÞ is a 2-dimensional real linear space.
Proof Since M is a 2-dimensional smooth manifold, and p is in M; there exists an
admissible coordinate chart ðU; uU Þ of M satisfying p 2 U: Here, we can find
admissible coordinate chart ðU; wU Þ of M satisfying wU ðxÞ ¼ uU ðxÞ uU ðpÞ for
every x in U: Hence, wU ðpÞ ¼ 0: Let us define functions
u1 : U ! R; u2 : U ! R
Then, by Lemma 2.35, fouo1 jp ; ouo2 jp g is a basis of the real linear space Dp ðMÞ:
Hence, the dimension of Dp ðMÞ is 2: h
Then, Hp ðMÞ is a linear subspace of the real linear space F p ðMÞ: In short, we
say that Hp is a linear subspace of F p :
Proof Let us try to prove that Hp ðMÞ is nonempty. Let us recall that ½0 2 F p ðMÞ;
where 0 denotes the constant function zero defined on M: Since, for every c in
Cp ðMÞ,
c; ½ f þ ½g ¼ c; ½ f þ c; ½g ¼ 0 þ 0 ¼ 0:
c; a½ f ¼a c; ½ f ¼ að0Þ ¼ 0:
So, by the definition of Hp ðMÞ; a½f 2 Hp ðMÞ: This proves 2. Hence, Hp ðMÞ is
a linear subspace of the real linear space F p ðMÞ: h
72 2 Tangent Spaces
Fp
½ f þ Hp : ½ f 2 F p :
Hp
Fp
Here, vector addition, and scalar multiplication over Hp are defined as follows:
For every ½f ; ½g 2 F p ; and for every real t;
½ f þ Hp þ ½g þ Hp ð½ f þ ½gÞ þ Hp ¼ ½f þ g þ Hp ;
t ½ f þ Hp ðt½ f Þ þ Hp ¼ ½tf þ Hp :
F
The quotient space Hpp is denoted by Tp ðMÞ (or, simply Tp ) and is called the
F
cotangent space of M at p: Since Hpp is a real linear space, the cotangent space Tp of
M at p is a real linear space. Intuitively, in Tp , we will not distinguish between ½f
and ½g whenever ½f ½g is in Hp :
Definition Let M be an m-dimensional smooth manifold. Let p be an element of M:
Let ½f 2 F p : ½f þ Hp is denoted by ½~f ðor; ½f Þ or ðdf Þp ; and is called the
cotangent vector on M at p determined by the term ½f : Thus, we can write
Tp ¼ ½~f : ½ f 2 F p
or,
n o
Tp ¼ ðdf Þp : ½ f 2 F p :
½ f þ½g ¼ ½f þ g
t½ f ¼ ½tf
or,
2.5 Cotangent Spaces 73
(
ðdf Þp þðdgÞp ¼ ðd ðf þ gÞÞp
t ðdf Þp ¼ ðd ðtf ÞÞp :
ðuU cÞ : ðe; eÞ ! Rm
for every t in ðe; eÞ: Since f is in Cp1 ðMÞ, by the definition of Cp1 ðMÞ; dom f is an
open neighborhood of p in M; and
74 2 Tangent Spaces
f ðuU Þ1 : uU ððdom f Þ \ U Þ ! R
is C 1 at the point uU ðpÞ ð¼ uU ðcð0ÞÞ ¼ ðuU cÞð0ÞÞ in Rm ; and hence, the partial
derivatives ðD1 ðf ðuU Þ1 ÞÞððuU cÞð0ÞÞ; . . .; ðDm ðf ðuU Þ1 ÞÞððuU cÞð0ÞÞ
exist. Now,
dðf cÞðtÞ d f ðuU Þ1 ðuU cÞ ðtÞ
LHS ¼ ¼
dt t¼0 dt
t¼0
2 3
dF 1 ðtÞ
6 7
h i6 dt 7
6 7
¼ D1 f ðuU Þ1 ððuU cÞð0ÞÞ ... Dm f ðuU Þ1 ððuU cÞð0ÞÞ 6 ... 7
6 7
4 dF ðtÞ 5
m
dt
2 3 t¼0
dF1 ðtÞ
6 7
h i6 dt 7
6 7
¼ D1 f ðuU Þ1 ðuU ð pÞÞ ... Dm f ðuU Þ1 ðuU ð pÞÞ 6 .. 7
6. 7
4 dF ðtÞ 5
m
dt
2 3 t¼0
dF1 ðtÞ
6 dt 7
6 7 dF1 ðtÞ dFm ðtÞ
6 7
¼ ½ 0 . . . 0 6 ... 7 ¼ 0 þ ... þ 0 ¼ 0 ¼ RHS:
6 7 dt dt
4 dF ðtÞ 5 t¼0
m
dt t¼0
h
Proof (only if part): Let ½f 2 Hp ðMÞ: We have to show that
D1 f ðuU Þ1 ðuU ð pÞÞ ¼ ¼ Dm f ðuU Þ1 ðuU ð pÞÞ ¼ 0:
If not, otherwise, let ðD1 ðf ðuU Þ1 ÞÞðuU ðpÞÞ (for simplicity) be nonzero. We
have to arrive at a contradiction. Let us define a function c1 : ð1; 1Þ ! Rm as
follows: For every t in the open interval ð1; 1Þ;
c1 ðtÞ ðt; 0; . . .; 0Þ þ uU ð pÞ:
Put
c ðuU Þ1 c1 :
For 1: Here,
LHS ¼ cð0Þ ¼ ðuU Þ1 c1 ð0Þ ¼ ðuU Þ1 ðc1 ð0ÞÞ ¼ ðuU Þ1 ðð0; 0; . . .; 0Þ þ uU ð pÞÞ
¼ ðuU Þ1 ðuU ð pÞÞ ¼ p ¼ RHS:
This proves 1.
For 2: By the definition of c1 ; the function c1 : ð1; 1Þ ! Rm is continuous.
Since c1 : ð1; 1Þ ! Rm is continuous, and 0 is in the open interval ð1; 1Þ; c1
is continuous at 0.
Since ðU; uU Þ is a coordinate chart of M; ðuU Þ1 : uU ðUÞ ! U is a 1–1; onto,
continuous function, and uU ðUÞ is an open subset of Rm : Since p 2 U; uU ðpÞ 2
uU ðUÞ: Since c1 ð0Þ ¼ ð0; 0; . . .; 0Þ þ uU ðpÞ ¼ uU ðpÞ 2 uU ðUÞ; and ðuU Þ1 :
uU ðUÞ ! U is continuous, ðuU Þ1 : uU ðUÞ ! U is continuous at the point
c1 ð0Þ: Since c1 ð0Þ 2 uU ðUÞ; and uU ðUÞ is an open subset of Rm ; uU ðUÞ is an
open neighborhood of c1 ð0Þ: Since c1 : ð1; 1Þ ! Rm is continuous, uU ðUÞ is an
open neighborhood of c1 ð0Þ; there exists d [ 0 such that d\1 and, for every t in
ðd; dÞ we have c1 ðtÞ 2 uU ðUÞ; and hence, cðtÞ ¼ ððuU Þ1 c1 ÞðtÞ ¼
ðuU Þ1 ðc1 ðtÞÞ 2 U: Since cðtÞ 2 U for every t in ðd; dÞ; it follows that ðd; dÞ is
a subset of the domðcÞ: Hence, 0 is an interior point of the domain of the function
c: This proves 2.
For 3: In 2, we have seen that c is defined over ðd; dÞ: Now, it remains to be
proved that c is a smooth map from ðd; dÞ to M; that is, c is C 1 at every point t
in ðd; dÞ:
Here, c is defined over ðd; dÞ; so cðtÞ is meaningful for every t in ðd; dÞ:
Since cðtÞ is meaningful for every t in ðd; dÞ; and
cðtÞ ¼ ðuU Þ1 c1 ðtÞ ¼ ðuU Þ1 ðc1 ðtÞÞ ¼ ðuU Þ1 ððt; 0; . . .; 0Þ þ uU ð pÞÞ;
ðuU Þ1 ððt; 0; . . .; 0Þ þ uU ðpÞÞ is meaningful for every t in ðd; dÞ: Since ðU; uU Þ
is a coordinate chart of M; ðuU Þ1 : uU ðUÞ ! U: Since ðuU Þ1 : uU ðUÞ ! U;
and ðuU Þ1 ððt; 0; . . .; 0Þ þ uU ðpÞÞ is meaningful for every t in ðd; dÞ;
ðuU Þ1 ððt; 0; . . .; 0Þ þ uU ðpÞÞ is in U for every t in ðd; dÞ: Since
ðuU Þ1 ððt; 0; . . .; 0Þ þ uU ðpÞÞ is in U for every t in ðd; dÞ; and cðtÞ ¼
ðuU Þ1 ððt; 0; . . .; 0Þ þ uU ðpÞÞ; cðtÞ is in U for every t in ðd; dÞ: Thus, c maps
ðd; dÞ to U:
Now, it remains to be proved that c is a smooth map from ðd; dÞ to M; that is, c
is C 1 at every point t in ðd; dÞ: For this purpose, let us fix any t in ðd; dÞ: From
the definition of c1 ; c1 is C 1 at t: Since t is in ðd; dÞ; and c maps ðd; dÞ to U;
cðtÞ is in U: Since c1 is C 1 at t; c ¼ ðuU Þ1 c1 ; c maps ðd; dÞ to U; and t is in
ðd; dÞ; uU c is C1 at t: Since uU c is C1 at t; and ðU; uU Þ is an admissible
76 2 Tangent Spaces
uU ð pÞ ða1 ; . . .; am Þ:
Hence,
dðf c Þðt Þ d f ðuU Þ1 ðuU cÞ ðtÞ
0¼ c; ½ f ¼ ¼
dt t¼0 dt
t¼0
2 3
dðt þ a1 Þ
6 dt 7
6 7
6 7
h i6 dða2 Þ 7
6 dt 7
¼ D1 f ðuU Þ1 ððuU cÞð0ÞÞ Dm f ð u U Þ 1
ððu U c Þð0 ÞÞ 6 7
6 . 7
6 . 7
6 . 7
4 dða Þ 5
m
dt t¼0
3 2
1
h i6 0 7
6 7
¼ D1 f ðuU Þ1 ðuU ð pÞÞ Dm f ðuU Þ1 ðuU ð pÞÞ 6 .. 7
4 . 5
0 t¼0
1 1
¼ D 1 f ðu U Þ ðuU ð pÞÞ 1 þ 0 þ þ 0 ¼ D1 f ðuU Þ ðuU ð pÞÞ 6¼ 0;
which is a contradiction. h
Theorem 2.40 Let M be a m-dimensional smooth manifold. Let p be an element of
M: Let f 1 ; f 2 2 Cp1 ðMÞ: Let G be an open neighborhood of ðf 1 ðpÞ; f 2 ðpÞÞ in R2 : Let
F : G ! R be a smooth function. Then, there exists f in Cp1 ðMÞ such that for every
x in dom f ;
f ð xÞ ¼ F f 1 ð xÞ; f 2 ð xÞ :
2.5 Cotangent Spaces 77
Proof For this purpose, by the definition of Cp1 ðMÞ; we must find a function f such
that
1. dom f is an open neighborhood of p in M;
2. f ðxÞ ¼ Fðf 1 ðxÞ; f 2 ðxÞÞ for every x in dom f ; and
3. for every admissible coordinate chart ðU; uU Þ of M satisfying p 2 U,
f ðuU Þ1 : uU ððdom f Þ \ U Þ ! R
g : V1 ! R2
f :V !R
f ðuU Þ1 : uU ðV \ U Þ ! R
Then, f g:
Proof By Theorem 2.40, the existence of f ; g are guaranteed. Since f is in Cp1 ðMÞ;
dom f is an open neighborhood of p in M: Similarly, dom g is an open neighbor-
hood of p in M: Since dom f is an open neighborhood of p in M; and dom g is an
open neighborhood of p in M; their intersection ðdom f Þ \ ðdom gÞ is an open
neighborhood of p in M: Next, let us take any x in ðdom f Þ \ðdom gÞ: It remains to
be showed that f ðxÞ ¼ gðxÞ: Since x is in ðdom f Þ \ðdom gÞ; x is in dom f : Since x is
indom f ; f ðxÞ ¼ Fðf 1 ðxÞ; f 2 ðxÞÞ: Similarly, gðxÞ ¼ Fðf 1 ðxÞ; f 2 ðxÞÞ: Since f ðxÞ ¼
Fðf 1 ðxÞ; f 2 ðxÞÞ and gðxÞ ¼ Fðf 1 ðxÞ; f 2 ðxÞÞ; f ðxÞ ¼ gðxÞ: h
Theorem 2.42 Let M be an m-dimensional smooth manifold. Let p be an element
of M: Let f 1 ; f 2 2 Cp1 ðMÞ: Let G be an open neighborhood of ðf 1 ðpÞ; f 2 ðpÞÞ in R2 :
Let F : G ! R be a smooth function. Then, there exists a unique ½f in F p ðMÞ;
where f is in Cp1 ðMÞ; such that for every x in dom f ;
f ð xÞ ¼ F f 1 ð xÞ; f 2 ð xÞ :
Also
½ f ¼ ðD1 F Þ f 1 ð pÞ; f 2 ð pÞ f 1 þ ðD2 F Þ f 1 ð pÞ; f 2 ð pÞ f 2 :
Proof Existence: By Theorem 2.40, there exists f in Cp1 ðMÞ such that for every x
in dom f ;
f ð xÞ ¼ F f 1 ð xÞ; f 2 ð xÞ :
80 2 Tangent Spaces
Further, since f is in Cp1 ðMÞ; the C1 -germ ½f is in F p ðMÞ: This proves the
existence part of the theorem.
Uniqueness: For this purpose, let ½f ; ½g be in F p ðMÞ; where f ; g are in Cp1 ðMÞ;
such that for every x in dom f ;
f ð xÞ ¼ F f 1 ð xÞ; f 2 ð xÞ
Since
ðD1 F Þ f 1 ð pÞ; f 2 ð pÞ f 1 þ ðD2 F Þ f 1 ð pÞ; f 2 ð pÞ f 2
¼ ðD1 F Þ f 1 ð pÞ; f 2 ð pÞ f 1 þ ðD2 F Þ f 1 ð pÞ; f 2 ð pÞ f 2
¼ ðD1 F Þ f 1 ð pÞ; f 2 ð pÞ f 1 þ ðD2 F Þ f 1 ð pÞ; f 2 ð pÞ f 2
that is,
f þ Hp ¼ ðD1 F Þ f 1 ð pÞ; f 2 ð pÞ f 1 þ ðD2 F Þ f 1 ð pÞ; f 2 ð pÞ f 2 þ Hp ;
that is,
ðD1 F Þ f 1 ð pÞ; f 2 ð pÞ f 1 þ ðD2 F Þ f 1 ð pÞ; f 2 ð pÞ f 2 f 2 Hp ;
For this purpose, let us take any c in Cp ðMÞ. Since c is in Cp ðMÞ; there exists a
real number d [ 0 such that c is defined on the open interval ðd; dÞ; and cð0Þ ¼ p:
We have to prove that
ðD1 F Þ f 1 ð pÞ; f 2 ð pÞ c; f 1 þ ðD2 F Þ f 1 ð pÞ; f 2 ð pÞ c; f 2
¼ c; ½ f :
LHS ¼ ðD1 F Þ f 1 ð pÞ; f 2 ð pÞ c; f 1 þ ðD2 F Þ f 1 ð pÞ; f 2 ð pÞ c; f 2
dðf 1 cÞðtÞ dðf 2 cÞðtÞ
¼ ðD1 F Þ f 1 ð pÞ; f 2 ð pÞ þ ðD2 F Þ f 1 ð pÞ; f 2 ð pÞ
dt t¼0 dt t¼0
1 dð f 1
ð c ðt ÞÞ Þ d ðf 2
ð c ðt Þ ÞÞ
¼ ðD1 F Þ f ð pÞ; f 2 ð pÞ þ ðD2 F Þ f 1 ð pÞ; f 2 ð pÞ :
dt t¼0 dt t¼0
dðf cÞðtÞ dðf ðcðtÞÞÞ dðF ðf 1 ðcðtÞÞ; f 2 ðcðtÞÞÞÞ
RHS ¼ c; ½ f ¼ ¼ ¼
dt t¼0 dt t¼0 dt t¼0
dðF ððf 1 cÞðtÞ; ðf 2 cÞðtÞÞÞ
¼
dt t¼0
82 2 Tangent Spaces
dðf 1 cÞðtÞ
¼ ðD1 F Þ f 1 c ð0Þ; f 2 c ð0Þ
dt t¼0
1 2 dðf 2 cÞðtÞ
þ ðD2 F Þ f c ð0Þ; f c ð0Þ
dt t¼0
1 d ð f 1
ð cð t Þ Þ Þ
¼ ðD1 F Þ f ðcð0ÞÞ; f 2 ðcð0ÞÞ
dt t¼0
1 d ð f 2
ðcðtÞÞÞ
þ ðD2 F Þ f ðcð0ÞÞ; f ðcð0ÞÞ
2
dt t¼0
1 dð f 1
ð c ð t Þ Þ Þ dðf 2 ðcðtÞÞÞ
¼ ðD1 F Þ f ð pÞ; f 2 ð pÞ þ ðD2 F Þ f 1 ð pÞ; f 2 ð pÞ :
dt t¼0 dt t¼0
oF
Since ðD1 FÞðf 1 ðpÞ; f 2 ðpÞÞ is classically written as ox 1 ðf ðpÞ; f ðpÞÞ; etc., we can
1 2
write:
1 2 oF 1 1 oF 1 2
dF f ; f p ¼ ðdf Þp ¼ f ð pÞ; f ð pÞ
2
df p þ f ð pÞ; f ð pÞ
2
df p :
ox1 ox1
Note 2.44 Since the multiplication operation of real numbers is a smooth function
from R2 to R, in Theorem 2.42, we can take multiplication operation in place of F:
oF oF
1 ðf ðpÞ; f ðpÞÞ becomes f ðpÞ; and ox2 ðf ðpÞ; f ðpÞÞ becomes f ðpÞ: Hence,
1 2 2 1 2 1
So ox
we get
1 2 oF 1 1 oF 1 2
d f f p ¼ ðdf Þp ¼ f ð pÞ; f ð pÞ
2
df p þ f ð pÞ; f ð pÞ
2
df p
ox1 ox2
2 1 1 2
¼ f ð pÞ df p þ f ð pÞ df p
or,
1 2 2 1 1 2
d f f p ¼ f ð pÞ df p þ f ð pÞ df p :
Also
½f ¼ ðD1 F Þ f 1 ð pÞ; f 2 ð pÞ; f 3 ð pÞ f 1 þ ðD2 F Þ f 1 ð pÞ; f 2 ð pÞ; f 3 ð pÞ f 2
þ ðD3 F Þ f 1 ð pÞ; f 2 ð pÞ; f 3 ð pÞ f 3 :
Proof Its proof is quite similar to the proof of Theorem 2.42, etc. h
Theorem 2.47 Let M be a 2-dimensional smooth manifold. Let p be an element of
M: Let ðU; uU Þ be an admissible coordinate chart of M satisfying p 2 U: Let u1 ; u2
be the component functions of uU : Then,
1. u1 ; u2 are in Cp1 ðMÞ;
2. ½u1 ; ½u2 are linearly independent in the real linear space Tp ðMÞ;
3. for every ½f in Tp ðMÞ; where f is in Cp1 ðMÞ;
½f ¼ D1 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u1 þ D2 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u2 :
In other words,
! ! ! !
o o
ðdf Þp ¼ ½f du 1
þ ½f du2 :
ou1 p
p ou2 p
p
Proof
1. This fact has been shown earlier.
2. In 1, we have seen that u1 is in Cp1 ðMÞ: Since u1 is in Cp1 ðMÞ, the C 1 -germ ½f
is in F p ðMÞ; and hence, ½u1 ð¼ ½u1 þ Hp Þ is in the cotangent space Tp of M
at p: Similarly, ½u2 ð¼ ½u2 þ Hp Þ is in the cotangent space Tp of M at p (see
Fig. 2.2).
We have to show that ½u1 ; ½u2 are linearly independent. For this purpose, let
a1 ½u1 þ a2 ½u2 ¼ 0; where a1 ; a2 are real numbers. We have to show that a1 ¼
0; and a2 ¼ 0. Since 0 ¼ a1 ½u1 þ a2 ½u2 ¼ ½a1 u1 þ ½a2 u2 ¼
½a1 u þ a2 u ; ½a1 u þ a2 u is in Hp ; and hence, by the definition of Hp ;
1 2 1 2
0¼ c; a1 u1 þ a2 u2 ¼ c; a1 u1 þ a2 u2 ¼ c; a1 u1 þ a2 u2
dð u1 c Þ ð t Þ dðu2 cÞðtÞ
¼ a1 c; u1 þ a2 c; u2 ¼ a1 þa2 ;
dt t¼0 dt t¼0
that is,
dð u1 c Þ ð t Þ dð u2 c Þ ð t Þ
a1 þa2 ¼ 0. . .ð Þ
dt t¼0 dt t¼0
c1 : R ! R 2
c : ðe; eÞ ! M
We shall try to see that c is in Cp ðMÞ. By the definition of Cp ðMÞ; we must prove
that
1. cð0Þ ¼ p;
2. c is a parametrized curve in the manifold M; that is, c : ðe; eÞ ! M is a smooth
map from ðe; eÞ to M; that is, c is C 1 at every point p of ðe; eÞ:
For 1: Here, LHS ¼ cð0Þ ¼ ðuU Þ1 ðc1 ð0ÞÞ ¼ ðuU Þ1 ðu1 ðpÞ þ 0; u2 ðpÞÞ ¼
ðuU Þ1 ðu1 ðpÞ; u2 ðpÞÞ ¼ ðuU Þ1 ðuU ðpÞÞ ¼ p ¼ RHS:
For 2: For this purpose, let us take any t0 in ðe; eÞ: We have to show that
c : ðe; eÞ ! M is C1 at t0 :
We want to apply Theorem 2.2. Since t0 is in ðe; eÞ, by the definition of
c; cðt0 Þð¼ ðuU Þ1 ðc1 ðt0 ÞÞÞ is in U; and hence, ðU; uU Þ is an admissible coordi-
nate chart of M satisfying cðt0 Þ 2 U: Since uU c ¼ uU ððuU Þ1 c1 Þ ¼
ðuU ðuU Þ1 Þ c1 ¼ c1 ; and c1 is a smooth function, uU c is a smooth func-
tion. Since c : ðe; eÞ ! M; t0 is in ðe; eÞ; ðU; uU Þ is an admissible coordinate
chart of M satisfying cðt0 Þ 2 U; and uU c is a smooth function so, by Theorem
2.2, c is C 1 at t0 in ðe; eÞ: This proves 2. Thus, we have shown that c is in
Cp ðMÞ. Hence, from ð Þ;
d u1 ðuU Þ1 c1 ðtÞ d u2 ðuU Þ1 c1 ðtÞ
0 ¼ a1 þa2
dt dt
t¼0 t¼0
d u1 ðuU Þ1 c1 ðtÞ d u2 ðuU Þ1 c1 ðtÞ
¼ a1 þa2
dt dt
t¼0
2
t¼0
3
h i d ðu1 ð pÞ þ tÞ
¼ a1 D 1 u ðuU Þ
1 1
ðc1 ð0ÞÞ D 2 u ðuU Þ
1 1
ðc1 ð0ÞÞ 4 dt 5
d ðu2 ð pÞÞ
dt t¼0
86 2 Tangent Spaces
2 3
h i d ðu1 ð pÞ þ tÞ
þ a2 D1 u2 ðuU Þ 1
ðc1 ð0ÞÞ D2 u2 ðuU Þ 1
ðc1 ð0ÞÞ 4 dt 5
d 2
dt ðu ð pÞÞ t¼0
h i 1
1 1
¼ a1 D1 u1 ðuU Þ ðc1 ð0ÞÞ D2 u1 ðuU Þ ðc1 ð0ÞÞ
0
h i 1
1 1
þ a2 D1 u ðuU Þ
2
ðc1 ð0ÞÞ D 2 u ð uU Þ
2
ðc1 ð0ÞÞ
0
¼ a1 D1 u1 ðuU Þ1 ðc1 ð0ÞÞ þ a2 D1 u2 ðuU Þ1 ðc1 ð0ÞÞ
¼ a1 D1 u1 ðuU Þ1 ðuU ð pÞÞ þ a2 D1 u2 ðuU Þ1 ðuU ð pÞÞ ¼ a1 ð1Þ þ a2 ð0Þ ¼ a1 :
Thus, a1 ¼ 0: Similarly, a2 ¼ 0: Thus, we have shown that ½u1 ; ½u2 are lin-
early independent. This proves 2.
3. Let us take any ½f ; where f is in Cp1 ðMÞ:
This proves 3.
4. From 2,3, we find that f½u1 ; ½u2 g is a basis of the real linear space Tp ðMÞ:
This proves 4.
2.5 Cotangent Spaces 87
5. From 4, f½u1 ; ½u2 g is a basis of the real linear space Tp ðMÞ; and the number of
elements in the basis is 2; so the dimension of Tp ðMÞ is 2: This proves 5. h
As above, we can prove the following theorem.
Theorem 2.48 Let M be an m-dimensional smooth manifold. Let p be an element
of M: Let ðU; uU Þ be an admissible coordinate chart of M satisfying p 2 U: Let us
define m functions
u1 : U ! R; . . .; um : U ! R
Then,
1. u1 ; . . .; um are in Cp1 ðMÞ:
2. ½u1 ; . . .; ½um are linearly independent in the real linear space Tp ðMÞ:
3. For every ½f in Tp ðMÞ; where f is in Cp1 ðMÞ;
½f ¼ D1 f ðuU Þ1 u1 ð pÞ; . . .; um ð pÞ u1 þ
þ Dm f ðuU Þ1 u1 ð pÞ; . . .; um ð pÞ ½um :
In other words,
! ! ! !
o o
ðdf Þp ¼ ½f du 1
þ þ ½f ðdum Þp :
ou1 p
p oum p
0¼ c; ½ f ½g ¼ c; ½ f c; ½g
or,
c; ½ f ¼ c; ½g :
We shall denote the quotient set Cp ðMÞ= by Cp ðMÞ (or, simply, Cp ). Thus,
Cp ðM Þ ½c : c 2 Cp ðM Þ :
c; ½ f ¼ c0 ; ½ f :
dð f c Þ ð t Þ
h½c; ½ f i ¼ c; ½ f ¼ c; ½ f ¼
dt t¼0
f ðcð0 þ tÞÞ f ðcð0ÞÞ
¼ lim ;
t!0 t
u1 : U ! R; u2 : U ! R
d 1
h½c; ½ f i ¼ D1 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u c ðt Þ
dt
d
t¼0
1
þ D2 f ðuU Þ u ð pÞ; u ð pÞ
1 2
u c ðt Þ
2
:
dt t¼0
90 2 Tangent Spaces
In other words,
! ! ! !
o d 1 o d 2
h½c; ½ f i ¼ ½f u c ðt Þ þ ½f u c ðt Þ ;
ou1 p dt t¼0 ou2 p dt t¼0
or
! !
o d 1
h½c; ½ f i ¼ ½f u c ðt Þ
ou1 dt
p
! ! t¼0
o d 2
þ ½ f u c ð t Þ :
ou2 p dt t¼0
Proof Since c is in Cp ðMÞ, by the definition of Cp ðMÞ, there exists a real number
d [ 0 such that c is defined on the open interval ðd; dÞ; cð0Þ ¼ p; and c is a
smooth map from ðd; dÞ to M: Also, c is continuous at 0: Since c is continuous at
0; U is an open neighborhood of pð¼ cð0ÞÞ, there exists e [ 0 such that e\d; and
for every t in the open interval ðe; eÞ we have cðtÞ 2 U: So, for every t in ðe; eÞ;
ðuU cÞðtÞ ¼ uU ðcðtÞÞ ¼ u1 ðcðtÞÞ; u2 ðcðtÞÞ ¼ u1 c ðtÞ; u2 c ðtÞ :
Now
dð f c Þ ð t Þ
LHS ¼ h½c; ½ f i ¼ c; ½ f ¼ c; ½ f ¼
dt
t¼0
1
d f ðuU Þ uU c ðtÞ
¼ jt¼0
dt
d f ðuU Þ1 ðuU cÞ ðtÞ
¼
dt
h t¼0
i
1
¼ D1 f ðuU Þ ð ð u U c Þ ð 0Þ Þ D2 f ðuU Þ1 ðuU cÞð0Þ
2 3
d ð u1 c Þ ð t Þ
4 dt 5
d ð u2 c Þ ð t Þ
dt t¼0
2.5 Cotangent Spaces 91
h i
¼ D1 f ðuU Þ1 ðu1 ð pÞ; u2 ð pÞÞ D2 f ðuU Þ1 ðu1 ð pÞ; u2 ð pÞÞ
2 3
d ð u1 c Þ ð t Þ
4 dt 5
d ð u2 c Þ ð t Þ
dt
t¼0
1 d 1
¼ D1 f ðuU Þ u ð pÞ; u ð pÞ
1 2
u c ðt Þ
dt
d
t¼0
þ D2 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u2 c ðtÞ ¼ RHS:
dt t¼0
h
Note 2.53 Here, we will digress slightly from the current topic. Let V be any real
linear space. Let V be the collection of all linear functionals of V: We know that
V is also a real linear space under pointwise vector addition, and pointwise scalar
multiplication operations. V is called the dual space of V: Let n be the dimension
of V: Let fe1 ; . . .; en g be a basis of V: Let us define a function e1 : V ! R as
follows: For every real t1 ; . . .; tn
e1 t1 e1 þ þ tn en ¼ t1 :
etc. We shall try to show that fe1 ; . . .; en g constitutes a basis of V : For this
purpose, we must prove
1. e1 ; . . .; en are linearly independent, that is, t1 e1 þ þ tn e1 ¼ 0 implies
t1 ¼ 0; . . .; tn ¼ 0:
2. fe1 ; . . .; en g generates V ; that is, for every f in V ; there exist reals t1 ; . . .; tn
such that t1 e1 þ þ tn en ¼ f :
For 1: Let t1 e1 þ þ tn e1 ¼ 0: So,
0 ¼ 0 e1 ¼ t1 e1 þ þ tn e1 e1 ¼ t1 e1 e1 þ þ tn en e1
¼ t1 1 þ t2 0 þ þ tn 0 ¼ t1 :
Hence, t1 ¼ 0: Similarly, t2 ¼ 0; . . .; tn ¼ 0:
92 2 Tangent Spaces
Here,
1
LHS ¼ f e e1 þ þ ðf ðen ÞÞen ðe1 Þ ¼ f e1 e1 ðe1 Þ þ þ ðf ðen ÞÞen ðe1 Þ
¼ f e1 1 þ f e2 0 þ þ ðf ðen ÞÞ 0 ¼ f e1 ¼ RHS;
etc. Thus, we have shown that fe1 ; . . .; en g constitutes a basis of V : The basis
fe1 ; . . .; en g of the dual space V is called the dual basis of the basis fe1 ; . . .; en g:
Observe that
1. ej ðei Þ ¼ dij :
2. dimðV Þ ¼ n ¼ dimðVÞ:
Now, let us consider the mapping m : V ! V defined as follows: For every real
t1 ; . . .; tn
m t1 e1 þ þ tn en ¼ t1 e1 þ þ tn en :
t1 e1 þ þ tn en ¼ f :
Now
m t1 e1 þ þ tn en ¼ t1 e1 þ þ tn en ¼ f ;
u1 : U ! R; . . .; um : U ! R
By Theorem 2.48, f½u1 ; . . .; ½um g is a basis of the real linear space Tp ðMÞ:
Let us define the function ½u1 : Tp ðMÞ ! R as follows: For every real
t1 ; . . .; tm ;
½u1 t1 u1 þ þ tm ½um t1 :
½u2 t1 u1 þ þ tm ½um t2 ;
etc. We know that f½u1 ; . . .; ½um g constitutes a basis of the dual space T p ðMÞ of
Tp ðMÞ. f½u1 ; . . .; ½um g is called the dual basis of the basis f½u1 ; . . .; ½um g
Clearly,
½ ui uj ¼ dij ;
Also,
dim T p ðM Þ ¼ dim Tp ðM Þ ¼ m:
½c : Tp ðM Þ ! R;
½c ð½ f Þ h½c; ½ f i:
Then, ½c is a linear functional on the real linear space Tp ðMÞ; that is,
f½c : c 2 Cp ðMÞg is a subset of the dual space T p ðMÞ of Tp ðMÞ:
Proof Here, we must prove:
For every real a; b; and for every ½f ; ½g in Tp ðMÞ; where f ; g are in Cp1 ðMÞ;
LHS ¼ ½c ða½ f þb½g Þ ¼ ½c ð½af þ½bg Þ ¼ ½c ð½af þ bg Þ ¼ h½c; ½af þ bg i
¼ c; ½af þ bg ¼ c; ½af þ bg ¼ c; ½af þ ½bg ¼ c; a½ f þ b½g
¼ a c; ½ f þ b c; ½g ¼ a c; ½ f þb c; ½g ¼ ha½c; ½ f i þ bh½c; ½g i
¼ að½c ð½ f ÞÞ þ bð½c ð½g ÞÞ ¼ RHS: h
h
Lemma 2.55 Let M be a 2-dimensional smooth manifold. Let p be an element of
M: Let ðU; uU Þ be an admissible coordinate chart ðU; uU Þ of M such that p is in U:
2.6 Tangent Space as a Dual Space 95
c1 : R ! R 2
Then
1. ðuU Þ1 c1 is in Cp ðMÞ
2. for every f in Cp1 ðMÞ;
h i
T ð½ f Þ ¼ h ðuU Þ1 c1 ; ½ f i:
Proof
1: Clearly, c1 is a smooth function. Here, c1 is continuous, c1 ð0Þ ¼
ððTð½u1 ÞÞ 0; ðTð½u2 ÞÞ 0Þ þ uU ðpÞ ¼ uU ðpÞ; and uU ðUÞ is an open
neighborhood of uU ðpÞ; so there exists e [ 0 such that for every t in the open
interval ðe; eÞ; we have c1 ðtÞ in uU ðUÞ; and hence, ðuU Þ1 ðc1 ðtÞÞ is in U:
Now, let us define a function
c : ðe; eÞ ! M
So,
We shall try to see that c is in Cp ðMÞ: By the definition of Cp ðMÞ; we must prove
that
1. cð0Þ ¼ p;
2. c is a parametrized curve in the manifold M; that is, c : ðe; eÞ ! M is a smooth
map from ðe; eÞ to M; that is, c is C 1 at every point of ðe; eÞ:
96 2 Tangent Spaces
For 1: Here LHS ¼ cð0Þ ¼ ðuU Þ1 ðc1 ð0ÞÞ ¼ ðuU Þ1 ðuU ðpÞÞ ¼ p ¼ RHS:
This proves 1.
For 2: For this purpose, let us take any t0 in ðe; eÞ: We have to show that
c : ðe; eÞ ! M is C 1 at t0 : We want to apply Theorem 2.2. Since t0 is in
ðe; eÞ, by the definition of c; cðt0 Þð¼ ðuU Þ1 ðc1 ðt0 ÞÞÞ is in U; and hence,
ðU; uU Þ is an admissible coordinate chart of M satisfying cðt0 Þ 2 U: Since
uU c ¼ uU ððuU Þ1 c1 Þ ¼ ðuU ðuU Þ1 Þ c1 ¼ c1 ; and c1 is a smooth
function, uU c is a smooth function. Since c : ðe; eÞ ! M; t0 is in ðe; eÞ;
ðU; uU Þ is an admissible coordinate chart of M satisfying cðt0 Þ 2 U; and uU c
is a smooth function, by Theorem 2.2, c is C 1 at t0 in ðe; eÞ: This proves 2.
Thus, we have shown that c is in Cp ðMÞ. This proves 1.
2: Since ½f is in Tp ðMÞ so, by Theorem 2.47,
½f ¼ D1 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u1 þ D2 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u2 ;
and hence,
1 1 2 1
LHS ¼ T f ¼ T D1 f uU u p ;u p u
1 1 2 2
þ D2 f uU u p ;u p u
1 1 2 1
¼ D1 f uU u p ;u p T u
1 1 2 2
þ D2 f uU u p ;u p T u :
and
d u2 ðuU Þ1 c1 ðtÞ
¼ T u2 :
dt
t¼0
Here,
d u1 ðuU Þ1 c1 ðtÞ d u1 ðuU Þ1 c1 ðtÞ
LHS ¼ ¼
dt dt
t¼0
2 t¼0
3
h i d ððT ð½u1 ÞÞt þ u1 ð pÞÞ
¼ D1 u1 ðuU Þ1 ðc1 ð0ÞÞ D 2 u ð uU Þ 1 1
ðc1 ð0ÞÞ 4 dt 5
d ððT ð½u2 ÞÞt þ u2 ð pÞÞ
dt t¼0
h i T ð½u1 Þ
1 1
¼ D1 u ðuU Þ
1
ðc1 ð0ÞÞ D 2 u ð uU Þ
1
ðc1 ð0ÞÞ
T ð½u2 Þ
1
¼ D1 u1 ðuU Þ ðuU ð pÞÞ T u1 þ D2 u1 ðuU Þ1 ðuU ð pÞÞ T u2
1
¼ ð1Þ T u1 þ ð0Þ T u2 ¼ T u ¼ RHS:
Similarly,
d u2 ðuU Þ1 c1 ðtÞ
¼ T u2 : h
dt
t¼0
h
Theorem 2.56 Let M be a 2-dimensional smooth manifold. Let p be an element of
M: Let T be a linear functional on Tp ðMÞ: Then, there exists c in Cp ðMÞ such that
for every ½f in Tp ðMÞ;
T ð½ f Þ ¼ h½c; ½ f i:
Let
c1 : R ! R 2
98 2 Tangent Spaces
Then, by Lemma 2.55, ðuU Þ1 c1 is in Cp ðMÞ and for every ½f in Tp ðMÞ;
h i
T ð½ f Þ ¼ h ðuU Þ1 c1 ; ½ f i: h
h
Theorem 2.57 Let M be a 2-dimensional smooth manifold. Let p be an element of
M: For every c in Cp ðMÞ we define a function
½c : Tp ðM Þ ! R;
½c ð½ f Þ h½c; ½ f i:
Then, the mapping ½c 7! ½c is a 1–1 mapping from f½c : c 2 Cp ðMÞg onto
T p ðMÞ:
Proof Here, we must prove
1. ½c 7! ½c is a well-defined mapping from f½c : c 2 Cp ðMÞg to T p ðMÞ;
2. if ½c ¼ ½c0 , where c; c0 2 Cp ðMÞ, then ½c ¼ ½c0 ;
3. for every T in T p ðMÞ; there exists c in Cp ðMÞ such that ½c ¼ T:
For 1: Let ½c ¼ ½c0 ; that is, c c0 : We have to show that ½c ¼ ½c0 ; that is, for
every ½f in Tp ðMÞ; ½c ð½f Þ ¼ ½c0 ð½f Þ: For this purpose, let us take any
½f in Tp ðMÞ: Since c c0 ; and ½f is in Tp ðMÞ, by the definition of ;
c; ½f ¼ c0 ; ½f : Now
c; ½ f ¼ c0 ; ½ f :
2.6 Tangent Space as a Dual Space 99
For 3: Let us take any T in T p ðMÞ: Since T is in T p ðMÞ; and T p ðMÞ is dual
space of the real linear space Tp ðMÞ; T is a linear functional on Tp ðMÞ; and
hence, by Theorem 2.56, there exists c in Cp ðMÞ such that for every ½f in
Tp ðMÞ;
T ð½ f Þ ¼ h½c; ½ f i:
½c : Tp ðM Þ ! R;
½c ð½ f Þ h½c; ½ f i:
Then, the mapping ½c 7! ½c is a 1–1 mapping from f½c : c 2 Cp ðMÞg onto
T p ðMÞ:
Proof Its proof is quite similar to the proof of Theorem 2.57. h
Definition Let M be an m-dimensional smooth manifold. Let p be an element of
M. Let g be the mapping ½c 7! ½c from f½c : c 2 Cp ðMÞg to T p ðMÞ; as defined in
Theorem 2.58. By Theorem 2.58, g is a 1–1 mapping from f½c : c 2 Cp ðMÞg onto
T p ðMÞ; and hence, g1 is a 1–1 mapping from T p ðMÞ onto f½c : c 2 Cp ðMÞg:
Define vector addition, and scalar multiplication over the set f½c : c 2 Cp ðMÞg
as follows: For every c; c0 in Cp ðMÞ; and for every real t,
and
and
gðt½cÞ ¼ tðgð½cÞÞ:
In other words,
and
ðt½cÞ ¼ tð½cÞ ;
because g is the mapping ½c 7! ½c from f½c : c 2 Cp ðMÞg to T p ðMÞ, as defined in
Theorem 2.58.
Since g : ½c 7! ½c is an isomorphism from real linear space f½c : c 2 Cp ðMÞg
onto real linear space T p ðMÞ, g1 : ½c 7! ½c is also an isomorphism. Also,
dim ½c : c 2 Cp ðM Þ ¼ dim T p ðM Þ :
Further, since dimðT p ðMÞÞ ¼ dimðTp ðMÞÞ ¼ ðthe dimension of the manifoldÞ,
so
dim ½c : c 2 Cp ðM Þ ¼ ðthe dimension of the manifoldÞ:
Proof of “only if” part: Let ½c ¼ ½c0 . Since ½c ¼ ½c0 , c c0 : Let us define 2
functions
u1 : U ! R; u2 : U ! R
By Theorem 2.48, ½u1 ; ½u2 are in Tp ðMÞ: Since c c0 ; and ½u1 ; ½u2 are
in Tp ðMÞ, by the definition of , c; ½u1 ¼ c0 ; ½u1 ; and
c; ½u2 ¼ c0 ; ½u2 : Now
We have to show that ½c ¼ ½c0 , that is, c c0 ; that is, for every ½f in Tp ðMÞ;
c; ½f ¼ c0 ; ½f : For this purpose, let us take any ½f in Tp ðMÞ: We
have to show that c; ½f ¼ c0 ; ½f : Now, since
102 2 Tangent Spaces
dððu1 cÞðtÞÞ dððu2 cÞðtÞÞ
c; u1 ; c; u2 ¼ ;
dt t¼0 dt t¼0
dððu1 cÞðtÞÞ dððu2 cÞðtÞÞ
¼ ;
dt dt t¼0
1
dðu ðcðtÞÞÞ dðu ðcðtÞÞÞ
2
dðu1 ðcðtÞÞ; u2 ðcðtÞÞÞ
¼ ; ¼
dt dt t¼0 dt t¼0
dðuU ðcðtÞÞÞ
¼
dt t¼0
dðuU cÞðtÞ dðuU c0 ÞðtÞ
¼ ¼
dt t¼0 dt t¼0
dðuU ðc0 ðtÞÞÞ dðu1 ðc0 ðtÞÞ; u2 ðc0 ðtÞÞÞ
¼ ¼
dt t¼0 dt t¼0
1 0
dðu ðc ðtÞÞÞ dðu2 ðc0 ðtÞÞÞ
¼ ;
dt dt t¼0
dððu c ÞðtÞÞ dððu2 c0 ÞðtÞÞ
1 0
¼ ;
dt dt t¼0
dððu1 c0 ÞðtÞÞ dððu2 c0 ÞðtÞÞ
¼ ;
dt t¼0 dt t¼0
0 1
¼ c; u ; c0 ; u2 ;
so,
c; u1 ¼ c 0 ; u1 and c; u2 ¼ c 0 ; u2 :
Now,
LHS ¼ c; ½ f
¼ c; D1 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u1 þ D2 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u2
¼ c; D1 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u1 þ D2 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u2
¼ c; D1 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u1 þ D2 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u2
¼ D1 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ c; u1 þ D2 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ c; u2
¼ D1 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ c 0 ; u1 þ D2 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ c 0 ; u2
¼ c0 ; D1 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u1 þ D2 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u2
¼ c0 ; D1 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u1 þ D2 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u2
¼ c0 ; D1 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u1 þ D2 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ u2
¼ c0 ; ½ f ¼ RHS:
h
2.6 Tangent Space as a Dual Space 103
3. For every ½f in Tp ðMÞ; for every c in Cp ðMÞ, and for every real s,
hs½c; ½ f i ¼ sh½c; ½ f i:
LHS ¼ h½c þ ½c0 ; ½f i ¼ h½c00 ; ½f i where ½c00 ½c þ ½c0 for some c00 in Cp ðMÞ:
Since ½c00 ½c þ ½c0 ; ð½c00 Þ ¼ ð½c þ ½c0 Þ ¼ ð½cÞ þ ð½c0 Þ ; and hence,
h½c00 ; ½ f i ¼ ð½c00 Þ ð½ f Þ ¼ ð½cÞ þð½c0 Þ ð½ f Þ
¼ ðð½cÞ Þð½ f Þ þ ð½c0 Þ ð½ f Þ ¼ h½c; ½ f i þ h½c0 ; ½ f i:
104 2 Tangent Spaces
Thus, we have shown that h½c þ ½c0 ; ½f i ¼ h½c; ½f i þ ½c0 ; ½f : This proves 2.
For 3: Let us take any ½f in Tp ðMÞ: Let us take any c in Cp ðMÞ; and any real s:
We have to show that hs½c; ½f i ¼ sh½c; ½f i:
LHS ¼ hs½c; ½f i ¼ h½c0 ; ½f i where ½c0 s½c for some c0 in Cp ðMÞ:
Since ½c0 ¼ s½c; ð½c0 Þ ¼ ðs½cÞ ¼ sðð½cÞ Þ; and hence,
So
!
o
h½k1 ; ðdf Þp i ¼ ½ f :
ou1 p
Similarly, we have
!
o
h½k2 ; ðdf Þp i ¼ ½ f :
ou2 p
!
o o
h i ; ðdf Þp i ¼ ½ f :
ou p oui p
Also, we have
o
h i ; du j p i ¼ dij :
ou p
are the components of the tangent vector ½c at p, with respect to the natural basis
( )
o o
;
ou1 p ou2 p
so
d 1 d 2
½c ¼ u c ðt Þ ½ k1 þ u c ðt Þ ½ k2 :
dt t¼0 dt t¼0
Further, since
d 1 d 2
gð½cÞ ¼ ½c ¼ u c ðt Þ ½ k1 þ u c ðt Þ ½k2
dt dt
t¼0
t¼0
d 1 d 2
¼g u c ðt Þ ½ k1 þ u c ðt Þ ½ k2 ;
dt t¼0 dt t¼0
Further, since f½k1 ; ½k2 g is a basis of the tangent space Tp ðMÞ, that is,
( ! !)
o o
;
ou1 p ou2 p
are the components of the tangent vector ½c at p, with respect to the natural basis.h
Note 2.64 As above, for m-dimensional smooth manifold M, we get the formula:
! !
d 1 o d m o
½ c ¼ u c ðt Þ þ þ ð u cÞ ð t Þ :
dt t¼0 ou1 p dt t¼0 oum p
Further, since f½k1 ; . . .; ½km g is a basis of the tangent space Tp ðMÞ; that is,
( ! !)
o o
; . . .;
ou1 p oum p
108 2 Tangent Spaces
are the components of the tangent vector ½c at p, with respect to the natural basis.
Note 2.65 When there is no confusion, we generally avoid writing the lower index
p of the tangent vectors, cotangent vectors, etc.
Theorem 2.66 Let M be an m-dimensional smooth manifold. Let p be an element
of M. Let ðdf Þp ; ðdgÞp be in Tp ; where f ; g are in Cp1 . Let X be in Tp M. Let s; t be
any real numbers. Then,
1. hX; ðdðsf þ tgÞÞp i ¼ s hX; ðdf Þp i þ t hX; ðdgÞp i;
2. hX; ðdðf gÞÞp i ¼ f ðpÞ hX; ðdgÞp i þ gðpÞ hX; ðdf Þp i:
Proof
1: By Theorem 2.45, hX; ðdðsf þ tgÞÞp i ¼ hX; s ððdf Þp Þ þ t ððdgÞp Þi: Further,
since the mapping ð½c; ½f Þ 7! h½c; ½f i from the real linear space Tp Tp to
R is linear in the second variable, hX; s ððdf Þp Þ þ t ððdgÞp Þi ¼ s hX; ðdf Þp þ
t X; ðdgÞp i; and hence,
This proves 1.
2: By Theorem 2.45, hX; ðdðf gÞÞp i ¼ hX; ðgðpÞÞðdf Þp þ ðf ðpÞÞðdgÞp i: Further,
since the mapping ð½c; ½f Þ 7! h½c; ½f i from the real linear space Tp Tp to
R is linear in the second variable, hX; ðgðpÞÞðdf Þp þ ðf ðpÞÞðdgÞp i ¼ ðgðpÞÞ
hX; ðdf Þp i þ ðf ðpÞÞ hX; ðdgÞp i; and hence,
This proves 2. h
where
!
ou01 01 1
o
D 1 u ð u Þ u1 ð pÞ; u2 ð pÞ ¼ u01 ;
ou1 U
ou1 p
and
!
ou01 01 1
o
D 2 u ð u Þ u1 ð pÞ; u2 ð pÞ ¼ u01 :
ou2 U
ou2 p
Proof Here,
d 1 d 2
u c ðt Þ ; u c ðt Þ
dt t¼0 dt t¼0
are the components of the tangent vector ½c at p, with respect to the natural basis
( )
o o
;
ou1 p ou2 p
are the components of the tangent vector ½c at p, with respect to the natural basis
( )
o o
;
ou01 p ou02 p
of the tangent space Tp M under local coordinate system ðu0i Þ. Since c is in Cp ðMÞ,
cð0Þ ¼ p. We have to prove that
01 01
d 01 ou d 1 ou d 2
u c ðt Þ ¼ u c ð t Þ þ u c ð t Þ :
dt t¼0 ou1 dt t¼0 ou2 dt t¼0
d 01 d 0 1
LHS ¼ u c ðt Þ ¼ uU c ðt Þ
dt dt
t¼0
1
t¼0
d 0 1
¼ uU ðu U Þ uU c ðt Þ
dt
1 t¼0
d 0 1
¼ uU ðu U Þ ðuU cÞ ðt Þ
dt
1 t¼0
d 1
¼ u0U ðuU Þ ðuU cÞ ð t Þ
dt
1
t¼0
1
0 1 0 1
¼ D 1 uU ðu U Þ ððu U c Þð0 ÞÞ D 2 u U ðuU Þ ððuU cÞð0ÞÞ
2 3
d ðu c Þ1 ðt Þ
6 dt U
7
4 t¼0 5
d ðu c Þ2 ðt Þ
dt U
1
t¼0
1
1
¼ D1 u0U ðuU Þ ðuU ðcð0ÞÞÞ D2 u0U ðuU Þ1 ðuU ðcð0ÞÞÞ
2 3
d ðu c Þ1 ðt Þ
6 dt U
7
4 t¼0 5
d ðu c Þ2 ðt Þ
dt U
1
t¼0
1
1
¼ 0
D 1 uU ðu U Þ ðu U ð p ÞÞ D2 u0U ðuU Þ1 ðu U ð p ÞÞ
2 3
d ðu c Þ1 ðt Þ
6 dt U
7
4 t¼0 5
d ðu c Þ2 ðt Þ
dt U
1
t¼0
1
1
¼ 0
D 1 uU ðu U Þ ðu U ð p ÞÞ D2 u0U ðuU Þ1 ðu U ð p ÞÞ
2 3
d ð u Þ 1
c ð t Þ
6 dt U
7
4 t¼0 5
d ð u Þ 2
c ð t Þ
dt U
t¼0
2.7 Contravariant Vectors and Covariant Vectors 111
1 1
¼ D1 u0U ðuU Þ1 ðuU ð pÞÞ D2 u0U ðuU Þ1 ðu U ð p ÞÞ
2 3
d ðu U Þ1 ðc ðt ÞÞ
6 dt 7
4 t¼0 5
d ð u Þ 2
ð c ð t Þ Þ
dt U
1
t¼0
1
1
¼ 0
D 1 u U ðu U Þ ðuU ð pÞÞ D2 u0U ðuU Þ1 ðu U ð p ÞÞ
2 3
d ððu1 ÞðcðtÞÞÞ
6 dt 7
4 t¼0 5
d ððu2 ÞðcðtÞÞÞ
dt
1
t¼0
1
1
¼ 0
D 1 u U ðu U Þ ðuU ð pÞÞ D2 u0U ðuU Þ1 ðu U ð p ÞÞ
2 3
d ðu1 cÞðtÞ
6 dt t¼0 7
4 5
d ðu2 cÞðtÞ
dt
t¼0
1 1
¼ D1 u0U ðuU Þ1 ðu1 ð pÞ; u2 ð pÞÞ D2 u0U ðuU Þ1 ðu1 ð pÞ; u2 ð pÞÞ
2 3
d ðu1 cÞðtÞ
6 dt 7
4 t¼0 5
d ðu2 cÞðtÞ
dt
t¼0
1
0 1 d 1
¼ D 1 u U ðu U Þ u1 ð pÞ; u2 ð pÞ u c ðt Þ
dt
1
t¼0
1 d
þ D2 u0U ðuU Þ u1 ð pÞ; u2 ð pÞ u 2 c ðt Þ :
dt t¼0
01
ou01 d 1 ou d 2
RHS ¼ u c ð t Þ þ u c ð t Þ
ou1 dt ou2 dt
d t¼0
t¼0
01 1
¼ D1 u ðuU Þ u ð pÞ; u ð pÞ
1 2
u c ðt Þ
1
dt
t¼0
01 1 d 2
þ D2 u ðuU Þ u ð pÞ; u ð pÞ
1 2
u c ðt Þ
dt
1
t¼0
0 1 d 1
¼ D1 uU ðuU Þ u ð pÞ; u ð pÞ
1 2
u c ðt Þ
dt
1
t¼0
1 d
þ D2 u0U ðuU Þ u1 ð pÞ; u2 ð pÞ u2 c ð t Þ :
dt t¼0
Note 2.69 Similarly, under the same conditions as in theorem 34, we get
02 02
d 02 ou d 1 ou d 2
u c ðt Þ ¼ u c ðt Þ þ u c ðt Þ :
dt t¼0 ou1 dt t¼0 ou2 dt t¼0
In short, we write
0i 0i
d 0i ou d 1 ou d 2
u c ðt Þ ¼ u c ðt Þ þ u c ð t Þ ;
dt t¼0 ou1 dt t¼0 ou2 dt t¼0
or
2 0i
X
d 0i ou d
u c ðt Þ ¼ u j c ðt Þ ;
dt t¼0 j¼1
ou j dt t¼0
where
!
of o
1
D1 f u0U u01 ð pÞ; u02 ð pÞ ¼ ½ f ;
ou01 ou01 p
!
of o
1
D1 f ðu Þ u1 ð pÞ; u2 ð pÞ ¼ ½ f ;
ou 1 U
ou1 p
!
of o
1
D2 f ðu Þ u1 ð pÞ; u2 ð pÞ ¼ ½ f :
ou2 U
ou1 p
are the components of the cotangent vector ðdf Þp at p, with respect to the natural
basis fðdu1 Þp ; ðdu2 Þp g of the cotangent space Tp under local coordinate system ðui Þ.
Also,
of 1 of 1 01
D1 f u0U u01 ð pÞ; u02 ð pÞ ; D2 f u0U u ð pÞ; u02 ð pÞ
ou01 p ou02 p
are the components of the cotangent vector ðdf Þp at p, with respect to the natural
basis fðdu01 Þp ; ðdu02 Þp g of the cotangent space Tp under local coordinate system
ðu0i Þ. We have to prove that
2
of ou1 of ou of
¼ þ :
ou01 ou01 ou1 ou01 ou2
of
1
LHS ¼ 01
¼ D1 f u0U u01 ð pÞ; u02 ð pÞ
ou
1
¼ D1 f ðuU Þ1 uU u0U u01 ð pÞ; u02 ð pÞ
1 01
¼ D1 f ðuU Þ1 uU u0U u ð pÞ; u02 ð pÞ
¼ 1st column of the matrix product
h 1 01
D1 f ðuU Þ1 uU u0U u ð pÞ; u02 ð pÞ
1 01 i
D2 f ðuU Þ1 uU u0U u ð pÞ; u02 ð pÞ
2
12
3
0 1 1 01 02
0 1 1 01 02
6 D 1 u U u U ðu ð p Þ; u ð pÞ Þ D 2 u U u U ðu ð pÞ; u ð p ÞÞ 7
6 7
6 2 7
4 0 1 2 01 02
0 1 01 02
5
D 1 uU uU ðu ð pÞ; u ð pÞÞ D2 u U u U ðu ð pÞ; u ð pÞÞ
1 01
¼ D1 f ðuU Þ1 uU u0U u ð pÞ; u02 ð pÞ
1 1 01
D1 uU u0U u ð pÞ; u02 ð pÞ
1 01
þ D2 f ðuU Þ1 uU u0U u ð pÞ; u02 ð pÞ
1 2 01
D1 uU u0U u ð pÞ; u02 ð pÞ
1 0 1 1 0
¼ D1 f ðuU Þ1 uU u0U uU ð p Þ D1 uU u0U uU ð pÞ
1 0 1 2 0
þ D2 f ðuU Þ1 uU u0U uU ð pÞ D1 uU u0U uU ð pÞ
1 1 1
0
¼ D1 f ðuU Þ1 uU u0U u0U ð pÞ D1 uU u0U uU ð pÞ
1 1 2
0
þ D2 f ðuU Þ1 uU u0U u0U ð pÞ D1 uU u0U uU ð pÞ
114 2 Tangent Spaces
2
ou1 of ou of
RHS ¼ 01
þ 01
ou ou 1 ou ou2
1 01 of
¼ D1 u1 u0U u ð pÞ; u02 ð pÞ
ou1
of
1
þ D1 u2 u0U u01 ð pÞ; u02 ð pÞ
ou2
0 1 1 01 of
¼ D1 uU uU u ð pÞ; u02 ð pÞ
ou1
1 2 01 of
þ D1 uU u0U u ð pÞ; u02 ð pÞ
ou2
1 1 01
¼ D1 uU u0U u ð pÞ; u02 ð pÞ
D1 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ
0 1 2 01 02
þ D1 uU uU u ð pÞ; u ð pÞ
D2 f ðuU Þ1 u1 ð pÞ; u2 ð pÞ
0 1 1 0
¼ D1 uU uU uU ð pÞ D1 f ðuU Þ1 ðuU ð pÞÞ
0 1 2 0
þ D1 uU uU u U ð pÞ D2 f ðuU Þ1 ðuU ð pÞÞ :
Note 2.71 Similarly, under the same conditions as in Theorem 2.70, we get
2
of ou1 of ou of
¼ þ :
ou02 ou02 ou1 ou02 ou2
In short, we write
X2
of ou j of
¼ ;
ou0i j¼1
ou0i ou j
1
Reason: Since f is in CFðpÞ ðNÞ, f is a real-valued function such that dom f is an
open neighborhood of FðpÞ in N. Since F : M ! N is a smooth map, and p is in M,
F : M ! N is continuous at p. Since F : M ! N is continuous at p, and dom f is an
open neighborhood of FðpÞ in N, F 1 ðdom f Þ is an open neighborhood of p in
M. Since domðf FÞ ¼ F 1 ðdom f Þ, and F 1 ðdom f Þ is an open neighborhood of
p in M, domðf FÞ is an open neighborhood of p in M. Now, it remains to be
proved that f F is C1 at p.
Since F : M ! N is a smooth map, and p is in M, F is C 1 at p. Since f is in
1
CFðpÞ ðNÞ, f is a real-valued function such that dom f is an open neighborhood of
FðpÞ in N, and f is C1 at FðpÞ in N. Since the F-image of domðf FÞ is contained
in dom f, F is C 1 at p, and f is C 1 at FðpÞ, the composite function f F is C1 at p.
Hence, f F is in Cp1 ðMÞ.
Theorem 2.73 Let M be any m-dimensional smooth manifold, and let N be any n-
dimensional smooth manifold. Let F : M ! N be any smooth map, and let p be in
M. Let
F : TF ð pÞ ! Tp
be the function defined as follows: For every ðdf ÞFðpÞ in TFðpÞ , where f is in
1
CFðpÞ ðNÞ,
F ðdf ÞF ð pÞ ¼ ðd ðf F ÞÞp :
116 2 Tangent Spaces
Then, F is linear.
1
Proof For every f in CFðpÞ ðNÞ, from the above note, f F is in Cp1 ðMÞ, and hence,
1
ðdðf FÞÞp is in Tp . Since, for every f in CFðpÞ ðNÞ, F ððdf ÞFðpÞ Þ ð ðdðf FÞÞp Þ is
1
in Tp , F : TFðpÞ ! Tp is a function. Next, let us take any f ; g in CFðpÞ ðNÞ, and any
real s; t. We have to show that
F ðd ðsf þ tgÞÞF ð pÞ ¼ s F ðdf ÞF ð pÞ þ t F ðdgÞF ð pÞ :
h
Note 2.74 Let M be any m-dimensional smooth manifold, and N be any n-dimen-
sional smooth manifold. Let p be in M. Let F : M ! N be any smooth map. Fix any
c in Tp ðMÞ, where c is in Cp ðMÞ. As in Theorem 2.73, F : TFðpÞ ! Tp . So for every
1
ðdf ÞFðpÞ in TFðpÞ where f is in CFðpÞ ðNÞ, F ððdf ÞFðpÞ Þ is in Tp . Since F ððdf ÞFðpÞ Þ is in
Tp and, ½c is in Tp ðMÞ, h½c; F ððdf ÞFðpÞ Þi is a real number. We shall try to show that
the mapping ðdf ÞFðpÞ 7! h½c; F ððdf ÞFðpÞ Þi from real linear space TFðpÞ to R is linear,
1
that is, for every f ; g in CFðpÞ ðNÞ, and for every real s; t;
h½c; F s ðdf ÞF ð pÞ þ t ðdgÞF ð pÞ i ¼ sh½c; F ðdf ÞF ð pÞ i
þ th½c; F ðdgÞF ð pÞ i:
By Theorem 2.73,
LHS ¼ h½c; F s ðdf ÞF ð pÞ þ t ðdgÞF ð pÞ i
¼ h½c; s F ðdf ÞF ð pÞ þ t F ðdgÞF ð pÞ i:
so, LHS = RHS. So, ðdf ÞFðpÞ 7! h½c; F ððdf ÞFðpÞ Þi is a linear functional on the real
linear space TFðpÞ , and hence, the mapping ðdf ÞFðpÞ 7! h½c; F ððdf ÞFðpÞ Þi is a
member of T FðpÞ . Let us denote the mapping ðdf ÞFðpÞ 7! h½c; F ððdf ÞFðpÞ Þi by
b ð½cÞ. Thus, for every ½c in Tp ðMÞ, F
F b ð½cÞ is in T FðpÞ . Since F
b ð½cÞ is in T FðpÞ ,
b ð½cÞ; that is, for
there exists a unique F ð½cÞ in TFðpÞ ðNÞ such that ðF ð½cÞÞ ¼ F
1
every ðdf ÞFðpÞ in TFðpÞ where f is in CFðpÞ ðNÞ,
ðF ð½cÞÞ ðdf ÞF ð pÞ ¼ h½c; F ðdf ÞF ð pÞ i:
Since for every ½c in Tp ðMÞ, F ð½cÞ is a unique member of TFðpÞ ðNÞ,
F : Tp ðMÞ ! TFðpÞ ðNÞ:
1
We have seen that, for every ðdf ÞFðpÞ in TFðpÞ where f is in CFðpÞ ðNÞ, and for every
½c in Tp ðMÞ where c is in Cp ðMÞ,
ðF ð½cÞÞ ðdf ÞF ð pÞ ¼ h½c; F ðdf ÞF ð pÞ i:
Theorem 2.75 Let M be any m-dimensional smooth manifold, and let N be any n-
dimensional smooth manifold. Let F : M ! N be any smooth map, and let p be in
M. Then, the tangent map F : Tp ðMÞ ! TFðpÞ ðNÞ induced by F is linear.
Proof Let us take any ½c1 ; ½c2 in Tp ðMÞ, and any real s; t. We have to show that
LHS ¼ ðF ðs½c1 þ t½c2 ÞÞ ðdf ÞF ð pÞ ¼ hs½c1 þ t½c2 ; F ðdf ÞF ð pÞ i
¼ sh½c1 ; F ðdf ÞF ð pÞ i þ th½c2 ; F ðdf ÞF ð pÞ i
¼ s ðF ð½c1 ÞÞ ðdf ÞF ð pÞ þ t ðF ð½c2 ÞÞ ðdf ÞF ð pÞ
¼ ðsððF ð½c1 ÞÞ ÞÞ ðdf ÞF ð pÞ þ ðtððF ð½c2 ÞÞ ÞÞ ðdf ÞF ð pÞ
¼ ðsððF ð½c1 ÞÞ Þ þ tððF ð½c2 ÞÞ ÞÞ ðdf ÞF ð pÞ
¼ ðsðF ð½c1 ÞÞ þ tðF ð½c2 ÞÞÞ ðdf ÞF ð pÞ ¼ RHS:
Hence, LHS = RHS. Thus, we have shown that, for every ðdf ÞFðpÞ in TFðpÞ ,
ðF ðs½c1 þ t½c2 ÞÞ ðdf ÞF ð pÞ ¼ ðsðF ð½c1 ÞÞ þ tðF ð½c2 ÞÞÞ ðdf ÞF ð pÞ :
Hence,
gðF ðs½c1 þ t½c2 ÞÞ ¼ ðF ðs½c1 þ t½c2 ÞÞ ¼ ðsðF ð½c1 ÞÞ þ tðF ð½c2 ÞÞÞ
¼ gðsðF ð½c1 ÞÞ þ tðF ð½c2 ÞÞÞ
so,
gðF ðs½c1 þ t½c2 ÞÞ ¼ gðsðF ð½c1 ÞÞ þ tðF ð½c2 ÞÞÞ:
Tp ðMÞ determined by ðU; uU Þ. Since F : TFðpÞ ! Tp , and ðdv1 ÞFðpÞ ; ðdv2 ÞFðpÞ ;
ðdv3 ÞFðpÞ are in TFðpÞ , F ððdv1 ÞFðpÞ Þ; F ððdv2 ÞFðpÞ Þ; F ððdv3 ÞFðpÞ Þ are in Tp . Also, by
the definition of F ,
F dv1 F ð pÞ ¼ d v1 F p :
Thus,
2
X 1
i
F dv1 F ð pÞ
¼ Di wV F ðuU Þ1 u1 ð pÞ; u2 ð pÞ d u p:
i¼1
Similarly,
2
X 2
i
1
F dv 2
F ð pÞ
¼ Di wV F ðuU Þ u ð pÞ; u ð pÞ
1 2
d u p;
i¼1
and
2
X 3
i
F dv3 F ð pÞ
¼ Di wV F ðuU Þ1 u1 ð pÞ; u2 ð pÞ d u p:
i¼1
120 2 Tangent Spaces
In short, we write
X2 a
F ðdva ÞF ð pÞ ¼ Di wV F ðuU Þ1 u1 ð pÞ; u2 ð pÞ d ui p
i¼1
or
X 2
a oF a i
F ðdv ÞF ð pÞ ¼ d u p;
i¼1
oui p
where
a a
oF 1
Di w F ðu Þ u1 ð pÞ; u2 ð pÞ :
ou p
i V U
X
2
oF a
F ðdva Þ ¼ dui :
i¼1
oui
X
m
oF a
F ðdva Þ ¼ dui :
i¼1
oui
Note 2.77 Let M be any m-dimensional smooth manifold, and N be any n-dimen-
sional smooth manifold. Let p be in M. Let F : M ! N be any smooth map. Let
ðU; uU Þ be an admissible coordinate chart of M satisfying p 2 U, and let ðV; wV Þ be
an admissible coordinate chart of N satisfying FðpÞ 2 V. Let u1 ; . . .; um be the
component functions of uU , and let v1 ; . . .; vn be the component functions of wV .
Xm
oF a o
¼ ; d u j pi
j¼1
ou p ou p
j i
X oF a j oF a
m
¼ d ¼ :
j¼1
ou j p i oui p
Thus,
!!
o oF a
a
F ðdv ÞF ð pÞ ¼ :
oui p oui p
!!
Xn
o oF b o
F ¼ j ;
oui p b¼1
oui p ovb F ð pÞ
1
that is, for every ðdf ÞFðpÞ in TFðpÞ , where f is in CFðpÞ ðNÞ,
!! !
Xn
o oF b o
F ðdf ÞF ð pÞ ¼ j ð df Þ F ð pÞ :
oui p b¼1
oui p ovb F ð pÞ
Here,
!!
o o
LHS ¼ F ðdf ÞF ð pÞ ¼ ; F ð df ÞF ð pÞ
oui p oui p
!
n
X
o 1 a
¼ i
;F Da f ðwV Þ v ð pÞ; . . .; v ð pÞ
1 n
ðdv ÞF ð pÞ
p a¼1
n
o X 1 a
¼ ; Da f ðwV Þ v ð pÞ; . . .; v ð pÞ F ðdv ÞF ð pÞ
1 n
oui p a¼1
n
X 1 o
a
¼ Da f ðwV Þ v ð pÞ; . . .; v ð pÞ
1 n
; F ðdv ÞF ð pÞ
a¼1
oui p
!! !
Xn
o
¼ Da f ðwV Þ1 v1 ð pÞ; . . .; vn ð pÞ F ð dv a
Þ F ð pÞ
a¼1
oui p
n
X oF a
¼ Da f ðwV Þ1 v1 ð pÞ; . . .; vn ð pÞ ;
a¼1
oui p
122 2 Tangent Spaces
and
! !!
n
X
oF b o
RHS ¼ ðdf ÞF ð pÞ
b¼1
oui p ovb F ð pÞ
! !
Xn
oF b o
¼ ðdf ÞF ð pÞ
b¼1
oui p ovb F ð pÞ
!
Xn
oF b o
¼
b¼1
oui p ovb F ð pÞ
!!
n
X
1
Da f ðwV Þ v ð pÞ; . . .; v ð pÞ ðdva ÞF ð pÞ
1 n
a¼1
n
X n
X
oF b
¼ Da f ðwV Þ1 v1 ð pÞ; . . .; vn ð pÞ
b¼1
oui p a¼1
! !!
o
a
ðdv ÞF ð pÞ
ovb F ð pÞ
!
Xn Xn o
oF b 1 a
¼ Da f ðwV Þ v ð pÞ; . . .; v ð pÞ
1 n
; ðdv ÞF ð pÞ
b¼1
oui p a¼1 ovb F ð pÞ
!
Xn Xn a
oF b 1
¼ Da f ðwV Þ v ð pÞ; . . .; v ð pÞ db
1 n
b¼1
oui p a¼1
n
X oF b 1
¼ D b f ð w Þ v 1
ð p Þ; . . .; v n
ð p Þ
b¼1
oui p V
Xn
oF a
¼ Da f ðwV Þ1 v1 ð pÞ; . . .; vn ð pÞ :
a¼1
ou p
i
so,
!! !!
n
X
o oF b o
g F ¼g :
oui p b¼1
oui p ovb F ð pÞ
In short, we write
X
o n
oF a o
F ¼ :
oui a¼1
oui ova
Note 2.78 Let M and N be any 3-dimensional smooth manifolds. Let p be in M. Let
F : M ! N be any smooth map. Let ðU; uU Þ be an admissible coordinate chart of
M satisfying p 2 U, and let ðV; wV Þ be an admissible coordinate chart of N satis-
fying FðpÞ 2 V. Let u1 ; u2 ; u3 be the component functions of uU , and let v1 ; v2 ; v3
be the component functions of wV . Here,
( )
o o o
; ;
ou1 p ou2 p ou3 p
is the natural basis of the tangent space Tp ðMÞ under local coordinate system ðui Þ, and
( )
o o o
; 2 ; 3
ov1 F ð pÞ ov F ð pÞ ov F ð pÞ
is the natural basis of the tangent space TFðpÞ ðNÞ under local coordinate system ðvi Þ.
Also, by Theorem 2.75, the tangent map F is a linear transformation from linear
space Tp ðMÞ to linear space TFðpÞ ðNÞ. From the above note,
! !
X3
o oF j o
F ¼ :
oui p j¼1
oui p ov j F ð pÞ
is
2 3
oF 1 oF 1 oF 1
6 ou p ou2 ou3
7
1
6 oF 2 2 p 2 p 7
6 ou1 oF oF 7 ;
6 ou2 ou3 7
4 3 p 3 p 3 p 5
oF oF oF
ou1 p ou2 p ou3 p
33
1 j
where ðoF
j
oui Þp stands for ðDi ððwV ðF ðuU Þ ÞÞ ÞÞðu ðpÞ; u ðpÞ; u ðpÞÞ:
1 2 2
6 oF 2 2 p 2 p 7
det6
6 ou1 p
oF
ou2
oF
ou3
7 6¼ 0:
7
4 3 3 p 3 p 5
oF oF oF
ou1 p ou2 p ou3 p
6 oF 2 2 p 2 p 7
det6
6 ou1 p
oF
ou2
oF
ou3
7 6¼ 0;
7
4 3 3 p 3 p 5
oF oF oF
ou1 p ou2 p ou3 p
where
j j
oF 1
¼ Di i R 3 F ð i E Þ ðiE ð pÞÞ ¼ Di F j ðiE ð pÞÞ ¼ Di F j ð pÞ:
ou p
i
In the very definition of smooth manifold, there is a statement like “…for every
point p in the topological space M there exist an open neighborhood U of p, an open
subset V of Euclidean space Rn ; and a homeomorphism u from U onto V …”. This
phrase clearly indicates that homeomorphism u can act as a messenger in carrying
out the well-developed multivariable differential calculus of Rn into the realm of
unornamented topological space M. So the role played by multivariable differential
calculus in the development of the theory of smooth manifolds is paramount. Many
theorems of multivariable differential calculus are migrated into manifolds like their
generalizations. Although most students are familiar with multivariable differential
calculus from their early courses, but the exact form of theorems we need here are
generally missing from their collection of theorems. So this is somewhat long
chapter on multivariable differential calculus that is pertinent here for a good
understanding of smooth manifolds.
BA : Rn ! Rk
ðBAÞðxÞ BðAðxÞÞ:
Thus, we have seen that for every (t1, t2, t3) in R3 satisfying jðt1 ; t2 ; t3 Þj 1; we
have
So, ðjAðð1; 0; 0ÞÞj þ jAðð0; 1; 0ÞÞj þ jAðð0; 0; 1ÞÞjÞ is an upper bound of the set
jAððt1 ; t2 ; t3 ÞÞj : ðt1 ; t2 ; t3 Þ is in R3 ; and jðt1 ; t2 ; t3 Þj 1 :
Theorem 3.2
1. If A is in LðRn ; Rm Þ; then for every x in Rn ;
A : Rn ! Rm
is uniformly continuous.
3. If A, B are in LðRn ; Rm Þ; then
kA þ Bk k Ak þ kBk:
dðA; BÞ kA Bk;
Proof
1: Let A be in LðRn ; Rm Þ: Let x be in Rn :
Case I: when x = 0
128 3 Multivariable Differential Calculus
Case II: when x ≠ 0. Since x ≠ 0, jxj is a nonzero real number, and hence, jxj
1
Thus, we see that in all cases, for every ɛ > 0, there exists δ > 0 such that for
every x, y in Rn satisfying jx yj\d; we have jAðxÞ AðyÞj\e: So, by the
definition of uniform continuity, A is uniformly continuous. This proves 2.
3: Let A, B be in LðRn ; Rm Þ: Let us take any x in Rn satisfying jxj 1: Now,
This proves 3.
3.1 Linear Transformations 129
jðtAÞðxÞj ¼ jtðAðxÞÞj ¼ jtjjAðxÞj jtjðk Akj xjÞ ¼ ðjtjk AkÞj xj ðjtjk AkÞ 1
¼ jtjk Ak:
So, jtjkAk is an upper bound of the set fjðtAÞðxÞj : x is in Rn ; and jxj 1g; and
hence,
This proves 4.
5: Here, we must prove
i. For every A in LðRn ; Rm Þ;
kA Ak ¼ 0:
130 3 Multivariable Differential Calculus
kA Bk ¼ kB Ak:
kA Bk kA C k þ kC Bk:
For iv: By 3,
This proves 6.
7: Let us take any A in LðRn ; Rm Þ and B in LðRm ; Rk Þ: Let us take any x in Rn
satisfying jxj 1: Since
This proves 7. h
3.1 Linear Transformations 131
Theorem 3.3 Let X be the set of all invertible (i.e., 1–1 and onto) members of
LðRn Þ:
1. If A is in X; then kA1 k is nonzero.
2. If A is in X, then the open sphere Sð1=kA1 kÞ ðAÞ is contained in X; that is, if A is in
X, B is in LðRn Þ; and kB Ak\ kA11 k, then B is in X.
3. X is an open subset of LðRn Þ:
4. The mapping A 7! A1 from X to X is continuous.
Proof
1: Let us take any A in X: We claim that kA1 k is nonzero. If not, otherwise, let
kA1 k ¼ 0: We have to arrive at a contradiction. Since A is in X, by the definition
of X; A−1 exists and A−1 is in X: Also, AA−1 = I, where I denotes the identity
transformation from Rn to Rn : For every x in Rn ; 0 jA1 ðxÞj kA1 kjxj ¼ 0
jxj ¼ 0; so jA1 ðxÞj ¼ 0; and hence, A−1(x) = 0. Therefore,
and hence, 1 = 0, a contradiction. So, our claim is true, that is, kA1 k is
nonzero. This proves 1.
2: Let A be in X; and let B be in LðRn Þ: Since A is in X, by 1, kA1 k is nonzero,
and hence, kA11 k is a real number. Also, by the definition of norm kk, kA11 k is a
positive real number. Let
1
kB Ak\ 1 :
kA k
Thus, jx yj\jx yj; which gives a contradiction. This proves that B is 1–1.
Ontoness: Let us take any y in Rn : Since B : Rn ! Rn is 1–1, B
((1, 0, …, 0)), …, B((0, …, 0, 1)) are n distinct elements of Rn : Now, we shall
show that B((1, 0, …, 0)), …, B((0, …, 0, 1)) are linearly independent. For this
purpose, let t1(B((1, 0, …, 0))) + ⋯ + tn(B((0, …, 0, 1))) = 0. We have to show
that t1 = 0, …, tn = 0. Here,
Thus, B((s1, …, sn)) = y, where (s1, …, sn) is in Rn : This proves that B is onto.
This proves 2.
3: Let us take any A in X. Since A is in X, by 1, kA1 k is nonzero. Now by 2, the
open sphere fB : B 2 LðRn Þ; and kB Ak\ kA11 kg is contained in X. Hence, by
the definition of open set, X is an open subset of LðRn Þ: This proves 3.
4: Let us take any A in X. We have to prove that the mapping B 7! B1 is
continuous at A Since A is in X, by 1, kA1 k is nonzero. If B is in LðRn Þ
satisfying kB Ak\ kA11 k ; then by 2, B is in X, and hence, B−1 exists. Now, it
is enough to prove that limB!A kB1 A1 k exists, and its value is 0. Here,
1
B A1 ¼ A1 A B1 A1 BB1 ¼ A1 AB1 A1 BB1
¼ A1 AB1 BB1 ¼ A1 ðA BÞB1 A1 kA BkB1
¼ A1 kA Bk sup B1 ðxÞ : x is in Rn ; and j xj 1
or,
1
B A1 A1 kA Bk
1
sup B ðxÞ : x is in Rn ; and j xj 1 ðÞ:
3.1 Linear Transformations 133
so,
1
B ðxÞ 1 A1 kB Ak A1 :
Hence,
1
A
sup B1 ðxÞ : x is in Rn and j xj 1 :
1 kA1 kkB Ak
and
1
lim A1 1 þ 1
¼0
B!A 1 kA kkB Ak
so, limB!A kB1 A1 k exists, and its value is 0. This shows that the mapping
B 7! B1 from X to X is continuous at A. This proves 4. h
Definition Let {e1, e2, e3} be any basis of real linear space R3 ; and let {f1, f2} be
any basis of real linear space R2 : Let A be in LðR3 ; R2 Þ: Let
The matrix
is denoted by [A].
Similar definition can be supplied for A in LðRn ; Rm Þ:
Theorem 3.4 Let M be the collection of all 2 × 3 matrices with real entries. Let
{e1, e2, e3} be a basis of real linear space R3 ; and let {f1, f2} be a basis of real
linear space R2 : Then, the mapping A 7! ½A from LðR3 ; R2 Þ to M is 1–1 and onto.
Proof 1–1ness: Let [A] = [B], where
Aðt1 e1 þ t2 e2 þ t3 e3 Þ ¼ Bðt1 e1 þ t2 e2 þ t3 e3 Þ:
For this purpose, let us take any t1e1 + t2e2 + t3e3 in R3 : Since [A] = [B], aij = bij
for every i = 1, 2, and for every j = 1, 2, 3. Further,
3.1 Linear Transformations 135
so A(e1) = a11f1 + a21f2 = b11f1 + b21f2 = B(e1). Similarly, A(e2) = B(e2) and A
(e3) = B(e3).
So,
We shall show that A is in LðR3 ; R2 Þ: For this purpose, let us take any
x ≡ s1e1 + s2e2 + s3e3, y ≡ t1e1 + t2e2 + t3e3, and any real number t. We have to
prove that
1. Aðx þ yÞ ¼ AðxÞ þ AðyÞ;
2. AðtxÞ ¼ tðAðxÞÞ:
For 1: LHS ¼ Aðx þ yÞ ¼ Aððs1 e1 þ s2 e2 þ s3 e3 Þ þ ðt1 e1 þ t2 e2 þ t3 e3 ÞÞ
¼ Aððs1 þ t1 Þe1 þ ðs2 þ t2 Þe2 þ ðs3 þ t3 Þe3 Þ
¼ ðs1 þ t1 Þða11 f1 þ a21 f2 Þ þ ðs2 þ t2 Þða12 f1 þ a22 f2 Þ
þ ðs3 þ t3 Þða13 f1 þ a23 f2 Þ:
and
and
Since
Aðe1 Þ ¼ Að1e1 þ 0e2 þ 0e3 Þ
¼ 1ða11 f1 þ a21 f2 Þ þ 0ða12 f1 þ a22 f2 Þ þ 0ða13 f1 þ a23 f2 Þ;
Similarly,
Aðe2 Þ ¼ a12 f1 þ a22 f2 ;
Aðe3 Þ ¼ a13 f1 þ a23 f2 :
Hence,
a11 a12 a13
½ A ¼ :
a21 a22 a23 h
h
Note 3.5 The result similar to Theorem 3.4 can be proved as above for LðR ; R Þ: n m
Theorem 3.6 Let {e1, e2, e3, e4} be any basis of real linear space R4 ; let {f1, f2, f3}
be any basis of real linear space R3 ; and let {g1, g2} be any basis of real linear
space R2 : Let A be in LðR4 ; R3 Þ; and let B be in LðR3 ; R2 Þ: Then,
3.1 Linear Transformations 137
and
Hence,
ðBAÞðe1 Þ ¼ BðAðe1 ÞÞ ¼ Bða11 f1 þ a21 f2 þ a31 f3 Þ ¼ a11 ðBðf1 ÞÞ þ a21 ðBðf2 ÞÞ þ a31 ðBðf3 ÞÞ
¼ a11 ðb11 g1 þ b21 g2 Þ þ a21 ðb12 g1 þ b22 g2 Þ þ a31 ðb13 g1 þ b23 g2 Þ
¼ ða11 b11 þ a21 b12 þ a31 b13 Þg1 þ ða11 b21 þ a21 b22 þ a31 b23 Þg2 :
Now,
2 3
a11 a12 a13 a14
b11 b12 b13 6 7
½B½ A ¼ 4 a21 a22 a23 a24 5
b21 b22 b23 23
a31 a32 a33 a34 34
b11 a11 þ b12 a21 þ b13 a31
¼
b21 a11 þ b22 a21 þ b23 a31 24
a11 b11 þ a21 b12 þ a31 b13
¼
a11 b21 þ a21 b22 þ a31 b23 24
Hence, first column of [BA] is the same as the first column of [B][A]. Similarly,
second column of [BA] is the same as the second column of [B][A], third column of
[BA] is the same as the third column of [B][A], and fourth column of [BA] is the
same as the fourth column of [B][A]. This shows that [BA] = [B][A]. h
138 3 Multivariable Differential Calculus
Note 3.7 The result similar to Theorem 3.6 can be proved as above for A in
LðRn ; Rm Þ and B in LðRm ; Rp Þ:
Note 3.8 Let A be in LðR3 ; R2 Þ: Let
relative to the basis {(1, 0, 0), (0, 1, 0), (0, 0, 1)} of R3 and the basis {(1, 0), (0, 1)}
of R2 : Let (t1, t2, t3) be any member of R3 satisfying jðt1 ; t2 ; t3 Þj 1: Since A is a
linear transformation from R3 to R2 ;
and hence,
jAððt1 ; t2 ; t3 ÞÞj ¼ jððt1 ; t2 ; t3 Þ ða11 ; a12 ; a13 Þ; ðt1 ; t2 ; t3 Þ ða21 ; a22 ; a23 ÞÞj
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ jðt1 ; t2 ; t3 Þ ða11 ; a12 ; a13 Þj2 þjðt1 ; t2 ; t3 Þ ða21 ; a22 ; a23 Þj2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðjðt1 ; t2 ; t3 Þjjða11 ; a12 ; a13 ÞjÞ2 þðjðt1 ; t2 ; t3 Þjjða21 ; a22 ; a23 ÞjÞ2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð1jða11 ; a12 ; a13 ÞjÞ2 þð1jða21 ; a22 ; a23 ÞjÞ2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ jða11 ; a12 ; a13 Þj2 þjða21 ; a22 ; a23 Þj2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ ða11 Þ2 þða12 Þ2 þða13 Þ2 þða21 Þ2 þða22 Þ2 þða23 Þ2 :
relative to the basis {(1, 0, 0), (0, 1, 0), (0, 0, 1)} of R3 and the basis {(1, 0), (0, 1)}
of R2 ; then
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
!
u 3
uX X 2
k Ak t
2
aij :
j¼1 i¼1
The result similar to this conclusion can be proved as above for Rn in place of R3
and Rm in place of R2 :
Theorem 3.9 Let X be a metric space. Let s be the 1–1, onto function A 7! ½A of
Theorem 3.4, relative to the basis {(1, 0, 0), (0, 1, 0), (0, 0, 1)} of R3 and the basis
{(1, 0), (0, 1)} of R2 : If, for i = 1, 2, and j = 1, 2, 3, each aij : X ! R is
continuous, then the mapping
a11 ðxÞ a12 ðxÞ a13 ðxÞ
x 7! s1
a21 ðxÞ a22 ðxÞ a23 ðxÞ 23
where
a11 ðxÞ a12 ðxÞ a13 ðxÞ
Ax s1 2 L R3 ; R2 ;
a21 ðxÞ a22 ðxÞ a23 ðxÞ
1 a11 ðyÞ a12 ðyÞ a13 ðyÞ
Ay s 2 L R3 ; R2 :
a21 ðyÞ a22 ðyÞ a23 ðyÞ
Since
a11 ðxÞ a12 ðxÞ a13 ðxÞ
Ax ¼ s1 ;
a21 ðxÞ a22 ðxÞ a23 ðxÞ
and hence,
Ax ðð1; 0; 0ÞÞ ¼ ða11 ðxÞÞð1; 0Þ þ ða21 ðxÞÞð0; 1Þ ¼ ða11 ðxÞ; a21 ðxÞÞ;
Ax ðð0; 1; 0ÞÞ ¼ ða12 ðxÞÞð1; 0Þ þ ða22 ðxÞÞð0; 1Þ ¼ ða12 ðxÞ; a22 ðxÞÞ;
Ax ðð0; 0; 1ÞÞ ¼ ða13 ðxÞÞð1; 0Þ þ ða23 ðxÞÞð0; 1Þ ¼ ða13 ðxÞ; a23 ðxÞÞ:
Similarly,
1 a11 ðxÞ a12 ðxÞ a13 ðxÞ
x 7! s
a21 ðxÞ a22 ðxÞ a23 ðxÞ 23
from X to LðR3 ; R2 Þ is continuous at x. For this purpose, let us take any ɛ > 0. Since
each aij : X ! R is continuous at x, there exists δ > 0 such that for every y in the
open neighborhood Bδ(x) of x, we have jaij ðxÞ aij ðyÞj\ 6e ðfor every i ¼ 1;
2; and j ¼ 1; 2; 3Þ: Hence, for every y in the open neighborhood Bδ(x) of x, we have
from (*),
1 a11 ðxÞ a12 ðxÞ a13 ðxÞ a11 ðyÞ a12 ðyÞ a13 ðyÞ
s s1
a21 ðxÞ a22 ðxÞ a23 ðxÞ a21 ðyÞ a22 ðyÞ a23 ðyÞ
ðja11 ðxÞ a11 ðyÞj þ ja21 ðxÞ a21 ðyÞjÞþ_ ðja12 ðxÞ a12 ðyÞj þ ja22 ðxÞ a22 ðyÞjÞ
e e e e e e
þ ðja13 ðxÞ a13 ðyÞj þ ja23 ðxÞ a23 ðyÞjÞ \ þ þ þ þ þ ¼ e: h
6 6 6 6 6 6
h
Note 3.10 The result similar to Theorem 3.9 can be proved as above for R in place n
of R3 and Rm in place of R2 :
3.2 Differentiation
and
1
lim ððf1 ðx þ hÞ; f2 ðx þ hÞÞ ðf1 ðxÞ; f2 ðxÞÞ BðhÞÞ ¼ 0;
h!0 j hj
then A = B.
Proof We have to prove that A = B, that is, for every h in R3 ; AðhÞ BðhÞ ¼ 0:
For this purpose, let us take any h* ≡ (h1, h2, h3) in R3 :
It is sufficient to prove that |(A − B)(h*)| = 0.
Case I: when h ¼ 0
142 3 Multivariable Differential Calculus
Similarly,
1
lim ððf1 ðx þ th Þ; f2 ðx þ th ÞÞ ðf1 ðxÞ; f2 ðxÞÞ Bðth ÞÞ ¼ 0:
t!0 jth j
Hence,
1
lim ððf1 ðx þ th Þ; f2 ðx þ th ÞÞ ðf1 ðxÞ; f2 ðxÞÞ Aðth ÞÞ
t!0 jth j
1
ððf1 ðx þ th Þ; f2 ðx þ th ÞÞ ðf1 ðxÞ; f2 ðxÞÞ Bðth ÞÞ
jth j
Since jh1 j jðB AÞðh Þj ¼ 0, LHS ¼ jðA BÞðh Þj ¼ jðB AÞðh Þj ¼ 0 ¼ RHS:
Thus, we have shown that in all cases, LHS = RHS. h
3.2 Differentiation 143
Note 3.12 The result similar to Theorem 3.11 can be proved as above for Rn in
place of R3 and Rm in place of R2 : According to Theorem 3.11, if there exists A in
LðR3 ; R2 Þ such that
1
lim ððf1 ðx þ hÞ; f2 ðx þ hÞÞ ðf1 ðxÞ; f2 ðxÞÞ AðhÞÞ ¼ 0;
h!0 jhj
f 0 ðxÞ ¼ A:
Since r is continuous at 0,
1
RHS ¼ 0 ¼ rð0Þ ¼ lim rðhÞ ¼ lim ðf ðx þ hÞ f ðxÞ AðhÞÞ ¼ LHS:
h!0 h!0 jhj
relative to the basis {(1, 0, 0), (0, 1, 0), (0, 0, 1)} of R3 and the basis {(1, 0), (0, 1)}
of R2 : f 0 ðxÞ ¼ A if and only if
1
lim ðf1 ðx þ hÞ f1 ðxÞ ðh1 ; h2 ; h3 Þ ða11 ; a12 ; a13 ÞÞ ¼ 0;
ðh1 ;h2 ;h3 Þh!0 jhj
1
lim ðf2 ðx þ hÞ f2 ðxÞ ðh1 ; h2 ; h3 Þ ða21 ; a22 ; a23 ÞÞ ¼ 0:
ðh1 ;h2 ;h3 Þh!0 jhj
3.2 Differentiation 145
Proof
f 0 ðxÞ ¼ A means
1
ð0; 0Þ ¼ 0 ¼ lim ðf ðx þ hÞ f ðxÞ AðhÞÞ
h!0 jhj
1
¼ lim ððf1 ðx þ hÞ; f2 ðx þ hÞÞ ðf1 ðxÞ; f2 ðxÞÞ AðhÞÞ :
ðh1 ;h2 ;h3 Þh!0 jhj
Since
AðhÞ ¼ Aððh1 ; h2 ; h3 ÞÞ
¼ h1 ðAðð1; 0; 0ÞÞÞ þ h2 ðAðð0; 1; 0ÞÞÞ þ h3 ðAðð0; 0; 1ÞÞÞ
¼ h1 ða11 ; a21 Þ þ h2 ða12 ; a22 Þ þ h3 ða13 ; a23 Þ
¼ ðh1 a11 þ h2 a12 þ h3 a13 ; h1 a21 þ h2 a22 þ h3 a23 Þ
¼ ððh1 ; h2 ; h3 Þ ða11 ; a12 ; a13 Þ; ðh1 ; h2 ; h3 Þ ða21 ; a22 ; a23 ÞÞ;
1
lim ðf1 ðx þ hÞ f1 ðxÞ ðh1 ; h2 ; h3 Þ ða11 ; a12 ; a13 ÞÞ ¼ 0;
ðh1 ;h2 ;h3 Þh!0 jhj
1
lim ðf2 ðx þ hÞ f2 ðxÞ ðh1 ; h2 ; h3 Þ ða21 ; a22 ; a23 ÞÞ ¼ 0: h
ðh1 ;h2 ;h3 Þh!0 jhj
h
Note 3.15 The result similar to Theorem 3.14 can be proved as above for R in n
So, f is continuous at x.
Case II: when kAk ¼ 0
Here, kAk ¼ 0, so A = 0. Now, if h is in fh : x þ h 2 Eg satisfying
0\jhj\minf1; dg, then
1
jðf ðx þ hÞ f ðxÞÞ 0j ¼ jhj ðf ðx þ hÞ f ðxÞ 0ðhÞÞ
jhj
1
¼ j hj jðf ðx þ hÞ f ðxÞ 0ðhÞÞj
j hj
1 e
1 jðf ðx þ hÞ f ðxÞ AðhÞÞj \1 \e:
j hj 2
So, f is continuous at x.
Thus, we see that in all cases, f is continuous at x. h
3.2 Differentiation 147
lim tðhÞ ¼ 0:
h!0
1 1
jðf 0 ðxÞÞðhÞ þ jhjðrðhÞÞj ðjðf 0 ðxÞÞðhÞj þ jjhjðrðhÞÞjÞ
j hj j hj
1 1
¼ ðjðf 0 ðxÞÞðhÞj þ jhjjrðhÞjÞ ðkf 0 ðxÞkjhj þ jhjjrðhÞjÞ ¼ kf 0 ðxÞk þ jrðhÞj:
j hj j hj
Since limh!0 rðhÞ ¼ 0; and limk!0 sðkÞ ¼ 0; limh!0 sððf 0 ðxÞÞðhÞ þ jhjðrðhÞÞÞ ¼
sððf 0 ðxÞÞð0Þ þ j0jðrð0ÞÞÞ ¼ sð0 þ 0Þ ¼ sð0Þ ¼ 0; and hence, there exists d3 [ 0
such that for every h in fh : x þ h 2 Eg satisfying 0\jhj\d3 , we have
jsððf 0 ðxÞÞðhÞ þ jhjðrðhÞÞÞj\ 2ðkf 0 ðxÞkþ1Þ
e
:
Hence, for every h in fh : x þ h 2 Eg satisfying 0\jhj\minfd1 ; d2 ; d3 g, we
have
1
jðf 0 ðxÞÞðhÞ þ jhjðrðhÞÞjðsððf 0 ðxÞÞðhÞ þ jhjðrðhÞÞÞÞ
jhj
1
¼ jðf 0 ðxÞÞðhÞ þ jhjðrðhÞÞjjsððf 0 ðxÞÞðhÞ þ jhjðrðhÞÞÞj
jhj
e e
ðkf 0 ðxÞk þ 1Þjsððf 0 ðxÞÞðhÞ þ jhjðrðhÞÞÞj\ðkf 0 ðxÞk þ 1Þ ¼ :
2ðkf 0 ðxÞk þ 1Þ 2
Further,
1
jtðhÞ 0j ¼ jtðhÞj ¼ ðg0 ðf ðxÞÞÞðrðhÞÞ þ jðf 0 ðxÞÞðhÞ þ jhjðrðhÞÞj
j hj
ðsððf 0 ðxÞÞðhÞ þ jhjðrðhÞÞÞÞj jðg0 ðf ðxÞÞÞðrðhÞÞj
1 e e
þ jðf 0 ðxÞÞðhÞ þ jhjðrðhÞÞjðsððf 0 ðxÞÞðhÞ þ jhjðrðhÞÞÞÞ\ þ ¼ e:
j hj 2 2
This proves that limh!0 tðhÞ ¼ 0: h
Then, all the partial derivatives ðD1 f1 ÞðxÞ; ðD2 f1 ÞðxÞ; ðD3 f1 ÞðxÞ; ðD1 f2 ÞðxÞ; ðD2 f2 Þ
ðxÞ; ðD3 f2 ÞðxÞ exist.
Also,
where
ðrf1 ÞðxÞ ððD1 f1 ÞðxÞ; ðD2 f1 ÞðxÞ; ðD3 f1 ÞðxÞÞ;
ðrf2 ÞðxÞ ððD1 f2 ÞðxÞ; ðD2 f2 ÞðxÞ; ðD3 f2 ÞðxÞÞ:
relative to the basis fð1; 0; 0Þ; ð0; 1; 0Þ; ð0; 0; 1Þg of R3 and the basis fð1; 0Þ; ð0; 1Þg
of R2 : By Theorem 3.14,
1
lim ðf1 ðx þ hÞ f1 ðxÞ ðh1 ; h2 ; h3 Þ ða11 ; a12 ; a13 ÞÞ ¼ 0:
ðh1 ;h2 ;h3 Þh!0 jhj
Hence,
1
lim ðf1 ððx1 ; x2 ; x3 Þ þ ð0; t; 0ÞÞ f1 ððx1 ; x2 ; x3 ÞÞ ð0; t; 0Þ ða11 ; a12 ; a13 ÞÞ ¼0
t!0 jð0; t; 0Þj
or,
1
lim ðf1 ððx1 ; x2 þ t; x3 ÞÞ f1 ððx1 ; x2 ; x3 ÞÞ ta12 Þ ¼ 0
t!0 jtj
or,
1
lim ðf1 ððx1 ; x2 þ t; x3 ÞÞ f1 ððx1 ; x2 ; x3 ÞÞ ta12 Þ ¼ 0
t!0 t
3.2 Differentiation 151
or,
1
lim ðf1 ððx1 ; x2 þ t; x3 ÞÞ f1 ððx1 ; x2 ; x3 ÞÞÞ a12 ¼ 0
t!0 t
or,
f1 ððx1 ; x2 þ t; x3 ÞÞ f1 ððx1 ; x2 ; x3 ÞÞ
ðD2 f1 ÞðxÞ ¼ lim ¼ a12 :
t!0 t
relative to the basis fð1; 0; 0Þ; ð0; 1; 0Þ; ð0; 0; 1Þg of R3 and the basis fð1; 0Þ; ð0; 1Þg
of R2 : Also,
ðf 0 ðxÞÞðð1; 0; 0ÞÞ ¼ ððD1 f1 ÞðxÞ; ðD1 f2 ÞðxÞÞ;
ðf 0 ðxÞÞðð0; 1; 0ÞÞ ¼ ððD2 f1 ÞðxÞ; ðD2 f2 ÞðxÞÞ;
ðf 0 ðxÞÞðð0; 0; 1ÞÞ ¼ ððD3 f1 ÞðxÞ; ðD3 f2 ÞðxÞÞ:
h
Note 3.21 The result similar to Theorem 3.20 can be proved as above for R in n
where
(Here, ðDu f1 ÞðxÞ is called the directional derivative of f1 in the direction of u, etc.)
Proof Since f is differentiable at x; f 0 ðxÞ is in LðR3 ; R2 Þ: Let
relative to the basis fð1; 0; 0Þ; ð0; 1; 0Þ; ð0; 0; 1Þg of R3 and the basis fð1; 0Þ; ð0; 1Þg
of R2 : By Theorem 3.14,
1
lim ðf1 ðx þ hÞ f1 ðxÞ ðh1 ; h2 ; h3 Þ ða11 ; a12 ; a13 ÞÞ ¼ 0:
ðh1 ;h2 ;h3 Þh!0 jhj
Hence,
1
lim ðf1 ððx1 ; x2 ; x3 Þ þ tðu1 ; u2 ; u3 ÞÞ f1 ððx1 ; x2 ; x3 ÞÞ
t!0 j t ð u1 ; u2 ; u3 Þ j
ðtðu1 ; u2 ; u3 ÞÞ ða11 ; a12 ; a13 ÞÞ ¼ 0
or,
1
lim ðf1 ððx1 þ tu1 ; x2 þ tu2 ; x3 þ tu3 ÞÞ f1 ððx1 ; x2 ; x3 ÞÞ
t!0 jtjjðu1 ; u2 ; u3 Þj
tððu1 ; u2 ; u3 Þ ða11 ; a12 ; a13 ÞÞÞ ¼ 0
or,
1
lim ðf1 ððx1 þ tu1 ; x2 þ tu2 ; x3 þ tu3 ÞÞ f1 ððx1 ; x2 ; x3 ÞÞ
t!0 jt j 1
tððu1 ; u2 ; u3 Þ ða11 ; a12 ; a13 ÞÞÞ ¼ 0
3.2 Differentiation 153
or,
1
lim ðf1 ððx1 þ tu1 ; x2 þ tu2 ; x3 þ tu3 ÞÞ f1 ððx1 ; x2 ; x3 ÞÞ
t!0 t
tððu1 ; u2 ; u3 Þ ða11 ; a12 ; a13 ÞÞÞ ¼ 0
or,
f1 ððx1 þ tu1 ; x2 þ tu2 ; x3 þ tu3 ÞÞ f1 ððx1 ; x2 ; x3 ÞÞ
lim
t!0 t
ððu1 ; u2 ; u3 Þ ða11 ; a12 ; a13 ÞÞ ¼ 0
or,
f1 ððx1 þ tu1 ; x2 þ tu2 ; x3 þ tu3 ÞÞ f1 ððx1 ; x2 ; x3 ÞÞ
ðDu f1 ÞðxÞ ¼ lim
t!0 t
¼ ðu1 ; u2 ; u3 Þ ððD1 f1 ÞðxÞ; ðD2 f1 ÞðxÞ; ðD3 f1 ÞðxÞÞ
¼ ðu1 ; u2 ; u3 Þ ððrf1 ÞðxÞÞ ¼ u ððrf1 ÞðxÞÞ
Thus,
ðDu f1 ÞðxÞ ¼ u ððrf1 ÞðxÞÞ:
Similarly,
ðDu f2 ÞðxÞ ¼ u ððrf2 ÞðxÞÞ: h
h
Note 3.23 The result similar to Theorem 3.22 can be proved as above for Rn in
place of R3 and Rm in place of R2 :
Theorem 3.24 Let E be an open subset of Rn : Let f : E ! R: Let x ðx1 ; . . .; xn Þ
be in E Let f be differentiable at x. Then,
0 1
maxfðDu f ÞðxÞ : u 2 Rn and juj ¼ 1g ¼ @D f AðxÞ:
1
jðrf ÞðxÞj
ððrf ÞðxÞÞ
Hence,
0 1
ðDu f ÞðxÞ @D f AðxÞ:
1
jðrf ÞðxÞj
ððrf ÞðxÞÞ
h
Theorem 3.25 Let a\b: Let c : ða; bÞ ! R2 be any function differentiable on the
open interval ða; bÞ: Let G be an open subset of R2 ; and G contains the range of c:
Let f : G ! R be differentiable on G Then, for every t in the open interval ða; bÞ;
0 d
ðf cÞ ðtÞ ¼ ððrf ÞðcðtÞÞÞ cðtÞ :
dt
Hence,
ðf cÞ0 ðtÞ ¼ ½ðf 0 ðcðtÞÞÞðc0 ðtÞÞ ¼ ½ðf 0 ðcðtÞÞÞ½ðc0 ðtÞÞ
2d 3
c1 ðtÞ
6 7
¼ ½ ðD1 f ÞðcðtÞÞ ðD2 f ÞðcðtÞÞ 12 4 dt 5
d
c ðtÞ
dt 2 21
d d
¼ ððD1 f ÞðcðtÞÞÞ c ðtÞ þ ððD2 f ÞðcðtÞÞÞ c ðtÞ :
dt 1 dt 2
3.2 Differentiation 155
So,
0 d d
LHS ¼ ðf cÞ ðtÞ ¼ ððD1 f ÞðcðtÞÞÞ c ðtÞ þ ððD2 f ÞðcðtÞÞÞ c ðtÞ
dt 1 dt 2
d d
¼ ððD1 f ÞðcðtÞÞ; ðD2 f ÞðcðtÞÞÞ c ðtÞ; c2 ðtÞ
dt 1 dt
d
¼ ððD1 f ÞðcðtÞÞ; ðD2 f ÞðcðtÞÞÞ ðc1 ðtÞ; c2 ðtÞÞ
dt
d
¼ ððD1 f ÞðcðtÞÞ; ðD2 f ÞðcðtÞÞÞ cðtÞ
dt
d
¼ ððrf ÞðcðtÞÞÞ cðtÞ ¼ RHS:
dt
h
Note 3.26 The result similar to Theorem 3.25 can be proved as above for R in m
place of R2 :
Note 3.27 Let a be in Rn : The function t 7! ta from R to Rn is denoted by a,
provided that there is no confusion. Here,
Proof of Lemma
Case I: when f ðaÞ ¼ f ðbÞ. The lemma is trivial in this case.
Case II: when f ðaÞ 6¼ f ðbÞ. Since f ðaÞ 6¼ f ðbÞ; 0\jf ðbÞ f ðaÞj; and hence,
jf ðbÞf ðaÞj ðf ðbÞ f ðaÞÞ is in R : Put c jf ðbÞf ðaÞj ðf ðbÞ f ðaÞÞ: Clearly, jcj ¼ 1:
1 n 1
g : ½a; b ! R
gðxÞ c ðf ðxÞÞ:
g0 ðxÞ ¼ c ðf 0 ðxÞÞ:
Hence, by the Lagrange’s mean value theorem, there exists a real number t in
ða; bÞ such that
1
jf ðbÞ f ðaÞj ¼ ðf ðbÞ f ðaÞÞ ðf ðbÞ f ðaÞÞ ¼ c ðf ðbÞ f ðaÞÞ
jf ðbÞ f ðaÞj
¼ c ðf ðbÞÞ c ðf ðaÞÞ ¼ gðbÞ gðaÞ ¼ ðb aÞg0 ðtÞ ¼ ðb aÞðc ðf 0 ðtÞÞÞ
¼ c ððb aÞðf 0 ðtÞÞÞ jc ððb aÞðf 0 ðtÞÞÞj jcjjðb aÞðf 0 ðtÞÞj
¼ 1jðb aÞðf 0 ðtÞÞj ¼ ðb aÞjf 0 ðtÞj:
c : ½0; 1 ! R3
c0 ðtÞ ¼ b a:
Hence, by Lemma 3.29 and Theorem 3.2, there exists a real number t in ð0; 1Þ
such that
Thus,
so
jðD1 f2 ÞðyÞ ðD1 f2 ÞðxÞj jððD1 f1 ÞðyÞ ðD1 f1 ÞðxÞ; ðD1 f2 ÞðyÞ ðD1 f2 ÞðxÞÞj
¼ jðf 0 ðyÞ f 0 ðxÞÞðð1; 0; 0ÞÞj
sup jðf 0 ðyÞ f 0 ðxÞÞðtÞj : t 2 R3 ; jtj 1
¼ kf 0 ðyÞ f 0 ðxÞk\e:
Hence,
jðD1 f2 ÞðyÞ ðD1 f2 ÞðxÞj\e:
and
1
lim ðf2 ðða1 ; a2 ; a3 Þ
(ii) ðh1 ;h2 ;h3 Þh!0 jðh1 ; h2 ; h3 Þj
that is,
1
0¼ lim ðf1 ðða1 þ h1 ; a2 þ h2 ; a3 þ h3 ÞÞ f1 ðða1 ; a2 ; a3 ÞÞ
ðh1 ;h2 ;h3 Þh!0 jðh1 ; h2 ; h3 Þj
ðh1 ððD1 f1 Þðða1 ; a2 ; a3 ÞÞÞ þ h2 ððD2 f1 Þðða1 ; a2 ; a3 ÞÞÞ þ h3 ððD3 f1 Þðða1 ; a2 ; a3 ÞÞÞÞÞ
g1 : ½a1 d; a1 þ d ! R
Now, we want to apply Lagrange’s mean value theorem on g1. For fixed t0 in
½a1 d; a1 þ d; we have
160 3 Multivariable Differential Calculus
Since
limt!0 g1 ðt0 þtÞg
t
1 ðt0 Þ
¼ ðD1 f1 Þððt0 ; a2 þ h2 ; a3 þ h3 ÞÞ; ðt0 ; a2 þ h2 ; a3 þ h3 Þ is in E,
and D1 f1 : E ! R exists, g1 is differentiable at t0. Further,
f1 ðða1 þ h1 ; a2 þ h2 ; a3 þ h3 ÞÞ f1 ðða1 ; a2 þ h2 ; a3 þ h3 ÞÞ
¼ g1 ð a1 þ h1 Þ g1 ð a1 Þ
¼ ðða1 þ h1 Þ a1 Þg0 ðc1 Þ ¼ h1 ðg0 ðc1 ÞÞ
¼ h1 ððD1 f1 Þððc1 ; a2 þ h2 ; a3 þ h3 ÞÞÞ:
Thus, we see that there exists a real number c1 lying between a1 þ h1 and a1
such that
f1 ðða1 þ h1 ; a2 þ h2 ; a3 þ h3 ÞÞ ¼ f1 ðða1 ; a2 þ h2 ; a3 þ h3 ÞÞ
þ h1 ððD1 f1 Þððc1 ; a2 þ h2 ; a3 þ h3 ÞÞÞ:
Similarly, there exists a real number c2 lying between a2 þ h2 and a2 such that
Again, there exists a real number c3 lying between a3 þ h3 and a3 such that
f1 ðða1 þ h1 ; a2 þ h2 ; a3 þ h3 ÞÞ f1 ðða1 ; a2 ; a3 ÞÞ
¼ h1 ð ð D 1 f 1 Þ ð ð c 1 ; a2 þ h2 ; a3 þ h3 Þ Þ Þ
þ h2 ððD2 f1 Þðða1 ; c2 ; a3 þ h3 ÞÞÞ þ h3 ððD3 f1 Þðða1 ; a2 ; c3 ÞÞÞ
or,
3.2 Differentiation 161
So,
1
¼ ðf1 ðða1 þ h1 ; a2 þ h2 ; a3 þ h3 ÞÞ f1 ðða1 ; a2 ; a3 ÞÞ
jðh1 ; h2 ; h3 Þj
ðh1 ððD1 f1 Þðða1 ; a2 ; a3 ÞÞÞ þ h2 ððD2 f1 Þðða1 ; a2 ; a3 ÞÞÞ þ h3 ððD3 f1 Þðða1 ; a2 ; a3 ÞÞÞÞÞ
1
¼ ðh1 ððD1 f1 Þððc1 ; a2 þ h2 ; a3 þ h3 ÞÞ ðD1 f1 Þðða1 ; a2 ; a3 ÞÞÞ
jðh1 ; h2 ; h3 Þj
þ h2 ððD2 f1 Þðða1 ; c2 ; a3 þ h3 ÞÞ ðD2 f1 Þðða1 ; a2 ; a3 ÞÞÞ
þ h3 ððD3 f1 Þðða1 ; a2 ; c3 ÞÞ ðD3 f1 Þðða1 ; a2 ; a3 ÞÞÞÞ
jh1 j
jðD1 f1 Þððc1 ; a2 þ h2 ; a3 þ h3 ÞÞ ðD1 f1 Þðða1 ; a2 ; a3 ÞÞj
jðh1 ; h2 ; h3 Þj
jh2 j
þ jðD2 f1 Þðða1 ; c2 ; a3 þ h3 ÞÞ ðD2 f1 Þðða1 ; a2 ; a3 ÞÞj
jðh1 ; h2 ; h3 Þj
jh3 j
þ jðD3 f1 Þðða1 ; a2 ; c3 ÞÞ ðD3 f1 Þðða1 ; a2 ; a3 ÞÞj
jðh1 ; h2 ; h3 Þj
¼ jðD1 f1 Þððc1 ; a2 þ h2 ; a3 þ h3 ÞÞ ðD1 f1 Þðða1 ; a2 ; a3 ÞÞj
þ jðD2 f1 Þðða1 ; c2 ; a3 þ h3 ÞÞ ðD2 f1 Þðða1 ; a2 ; a3 ÞÞj
þ jðD3 f1 Þðða1 ; a2 ; c3 ÞÞ ðD3 f1 Þðða1 ; a2 ; a3 ÞÞj:
Since
jðc1 ; a2 þ h2 ; a3 þ h3 Þ ða1 ; a2 ; a3 Þj\d , by ðÞ;
162 3 Multivariable Differential Calculus
e
jðD1 f1 Þððc1 ; a2 þ h2 ; a3 þ h3 ÞÞ ðD1 f1 Þðða1 ; a2 ; a3 ÞÞj\ :
3
Similarly,
e
jðD2 f1 Þðða1 ; c2 ; a3 þ h3 ÞÞ ðD2 f1 Þðða1 ; a2 ; a3 ÞÞj\ ;
3
and
e
jðD3 f1 Þðða1 ; a2 ; c3 ÞÞ ðD3 f1 Þðða1 ; a2 ; a3 ÞÞj\ :
3
Hence,
1
jðh ; h ; h Þj ðf1 ðða1 þ h1 ; a2 þ h2 ; a3 þ h3 ÞÞ f1 ðða1 ; a2 ; a3 ÞÞ
1 2 3
ðh1 ððD1 f1 Þðða1 ; a2 ; a3 ÞÞÞ þ h2 ððD2 f1 Þðða1 ; a2 ; a3 ÞÞÞ þ h3 ððD3 f1 Þðða1 ; a2 ; a3 ÞÞÞÞÞ
kf 0 ðyÞ f 0 ðxÞk\e:
Here,
and
so
ðD1 f1 ÞðyÞ ðD1 f1 ÞðxÞ ðD2 f1 ÞðyÞ ðD2 f1 ÞðxÞ ðD3 f1 ÞðyÞ ðD3 f1 ÞðxÞ
½f 0 ðyÞ f 0 ðxÞ ¼ ;
ðD1 f2 ÞðyÞ ðD1 f2 ÞðxÞ ðD2 f2 ÞðyÞ ðD2 f2 ÞðxÞ ðD3 f3 ÞðyÞ ðD3 f3 ÞðxÞ 23
and hence,
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
!ffi v ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u 3 u 3 2 2 !
uX X 2 2 uX X e
0 0
kf ðyÞ f ðxÞk t Dj fi ðyÞ Dj fi ðxÞ \ t pffiffiffiffiffiffiffiffiffiffiffi ¼ e:
j¼1 i¼1 j¼1 i¼1 23
h
Note 3.34 The result similar to Theorem 3.33 can be proved as above for Rn in
place of R3 and Rm in place of R2 :
Also,
d ðf ðf ðaÞÞ; f ðf ðf ðaÞÞÞÞ cðd ðf ðaÞ; f ðf ðaÞÞÞÞ cðcðd ða; f ðaÞÞÞÞ ¼ c2 ðd ða; f ðaÞÞÞ;
so
Similarly,
etc. Now,
Similarly,
1
d ðf ðf ðf ðaÞÞÞ; f ðf ðf ðf ðf ðf ðf ðaÞÞÞÞÞÞÞÞ c3 ðd ða; f ðaÞÞÞ;
1c
lim an ¼ b:
n!1
the set of all invertible (i.e., 1–1 and onto) members of LðR3 Þ: Since f 0 ðaÞ is
invertible, f 0 ðaÞ is in X Since f 0 ðaÞ is in X, by Theorem 3.3, the open sphere
S 1 ðf 0 ðaÞÞ;
kðf 0 ðaÞÞ1 k
1
;
ðf ðaÞÞ1
0
is contained in X
Since f is a C1 function, by Theorem 3.33, f 0 : E ! LðR3 Þ is continuous. Since
f 0 : E ! LðR3 Þ is continuous at a, and
S 1 ðf 0 ðaÞÞ
2 kðf 0 ðaÞÞ1 k
is an open neighborhood of f 0 ðaÞ in X, there exists a real number r > 0 such that the
open sphere Sr ðaÞ is contained in E, and for every x in Sr ðaÞ; we have f 0 ðxÞ in
S 1 ðf 0 ðaÞÞ:
2 kðf 0 ðaÞÞ1 k
S 1 ðf 0 ðaÞÞ;
2 kðf 0 ðaÞÞ1 k
so
1
kf 0 ðxÞ f 0 ðaÞk\
2 ðf ðaÞÞ1
0
or,
1 1 1
I ðf 0 ðaÞÞ ðf 0 ðxÞÞ ¼ ðf 0 ðaÞÞ ðf 0 ðxÞÞ I ¼ ðf 0 ðaÞÞ ðf 0 ðxÞÞ I
1 1
¼ ðf 0 ðaÞÞ ðf 0 ðxÞÞ ðf 0 ðaÞÞ ðf 0 ðaÞÞ
1 1
¼ ðf 0 ðaÞÞ ðf 0 ðxÞ f 0 ðaÞÞ ¼ ðf 0 ðaÞÞ ðf 0 ðxÞ f 0 ðaÞÞ
1 1
ðf 0 ðaÞÞ kf 0 ðxÞ f 0 ðaÞk\ :
2
1
1
I ðf 0 ðaÞÞ ðf 0 ðxÞÞ ðÞ
2
uk : Sr ðaÞ ! R3
Let
2 3
h i a11 a12 a13
1
ðf 0 ðaÞÞ 4 a21 a22 a23 5
a31 a32 a33 33
or,
uk ðxÞ ¼ ðx1 ða11 ðf1 ðxÞÞ þ a12 ðf2 ðxÞÞ þ a13 ðf3 ðxÞÞÞ þ k1 ; x2
ða21 ðf1 ðxÞÞ þ a22 ðf2 ðxÞÞ þ a23 ðf3 ðxÞÞÞ þ k2 ; x3
ða31 ðf1 ðxÞÞ þ a32 ðf2 ðxÞÞ þ a33 ðf3 ðxÞÞÞ þ k3 Þ
168 3 Multivariable Differential Calculus
so,
u0k ðxÞ ðð1; 0; 0ÞÞ ¼ ðD1 ðx1 ða11 ðf1 ðxÞÞ þ a12 ðf2 ðxÞÞ þ a13 ðf3 ðxÞÞÞ þ k1 Þ;
D1 ðx2 ða21 ðf1 ðxÞÞ þ a22 ðf2 ðxÞÞ þ a23 ðf3 ðxÞÞÞ þ k2 Þ;
D1 ðx3 ða31 ðf1 ðxÞÞ þ a32 ðf2 ðxÞÞ þ a33 ðf3 ðxÞÞÞ þ k3 ÞÞ
¼ ð1 ða11 ððD1 f1 ÞðxÞÞ þ a12 ððD1 f2 ÞðxÞÞ þ a13 ððD1 f3 ÞðxÞÞÞ;
ða21 ððD1 f1 ÞðxÞÞ þ a22 ððD1 f2 ÞðxÞÞ þ a23 ððD1 f3 ÞðxÞÞÞ;
ða31 ððD1 f1 ÞðxÞÞ þ a32 ððD1 f2 ÞðxÞÞ þ a33 ððD1 f3 ÞðxÞÞÞÞ:
Also,
1 1
I ðf 0 ðaÞÞ ðf 0 ðxÞÞ ðð1; 0; 0ÞÞ ¼ I ðð1; 0; 0ÞÞ ðf 0 ðaÞÞ ððf 0 ðxÞÞðð1; 0; 0ÞÞÞ
1
¼ ð1; 0; 0Þ ðf 0 ðaÞÞ ððD1 f1 ÞðxÞ; ðD1 f2 ÞðxÞ; ðD1 f3 ÞðxÞÞ
1
¼ ð1; 0; 0Þ ððD1 f1 ÞðxÞÞ ðf 0 ðaÞÞ ðð1; 0; 0ÞÞ
1 1
þ ððD1 f2 ÞðxÞÞ ðf 0 ðaÞÞ ðð0; 1; 0ÞÞ þ ððD1 f3 ÞðxÞÞ ðf 0 ðaÞÞ ðð0; 0; 1ÞÞ
¼ ð1; 0; 0Þ ðððD1 f1 ÞðxÞÞða11 ; a21 ; a31 Þ þ ððD1 f2 ÞðxÞÞða12 ; a22 ; a32 Þ
þ ððD1 f3 ÞðxÞÞða13 ; a23 ; a33 ÞÞ
¼ ð1 ða11 ððD1 f1 ÞðxÞÞ þ a12 ððD1 f2 ÞðxÞÞ þ a13 ððD1 f3 ÞðxÞÞÞ;
ða21 ððD1 f1 ÞðxÞÞ þ a22 ððD1 f2 ÞðxÞÞ þ a23 ððD1 f3 ÞðxÞÞÞ;
ða31 ððD1 f1 ÞðxÞÞ þ a32 ððD1 f2 ÞðxÞÞ þ a33 ððD1 f3 ÞðxÞÞÞÞ:
Similarly,
1
ðuk 0 ðxÞÞðð0; 1; 0ÞÞ ¼ I ðf 0 ðaÞÞ ðf 0 ðxÞÞ ðð0; 1; 0ÞÞ;
1
ðuk 0 ðxÞÞðð0; 0; 1ÞÞ ¼ I ðf 0 ðaÞÞ ðf 0 ðxÞÞ ðð0; 0; 1ÞÞ:
1
kuk 0 ðxÞk :
2
Therefore, 12 is an upper bound of the set fkuk 0 ðxÞk : x 2 Sr ðaÞg; and hence,
fkuk 0 ðxÞk : x 2 Sr ðaÞg is bounded above.
Now, we want to apply Theorem 3.28. Since Sr ðaÞ is an open convex subset of
R ; uk : Sr ðaÞ ! R3 is differentiable on Sr ðaÞ; and the set fkuk 0 ðxÞk : x 2 Sr ðaÞg is
3
For 3: We have to show that f ðSr ðaÞÞ is an open subset of R3 and is connected.
Since Sr ðaÞ is connected, and f is continuous, f ðSr ðaÞÞ is connected. Now, we
want to show that f ðSr ðaÞÞ is an open subset of R3 : For this purpose, let us take
any f ðbÞ in f ðSr ðaÞÞ; where b is in Sr ðaÞ: We have to find a real number e [ 0
such that the sphere
S e ðf ðbÞÞ
2 kðf 0 ðaÞÞ1 k
is contained in f ðSr ðaÞÞ; that is, we have to find a real number e [ 0 such that if
e
jy f ðbÞj\
2 ðf ðaÞÞ1
0
Observe that
1 1
uððf 0 ðaÞÞ1 ÞðyÞ ðbÞ b ¼ b ðf 0 ðaÞÞ ðf ðbÞÞ þ ðf 0 ðaÞÞ ðyÞ b
1 1 1
¼ ðf 0 ðaÞÞ ðyÞ ðf 0 ðaÞÞ ðf ðbÞÞ ¼ ðf 0 ðaÞÞ ðy f ðbÞÞ
e e
1 1
ðf 0 ðaÞÞ jy f ðbÞj ðf 0 ðaÞÞ ¼ :
2 ðf ðaÞÞ
0 1 2
Thus,
e
uððf 0 ðaÞÞ1 ÞðyÞ ðbÞ b :
2
Hence, by Theorem 3.35, there exists a unique x in Se ½bð Sr ðaÞÞ such that
uððf 0 ðaÞÞ1 ÞðyÞ ðx Þ ¼ x : Hence, jx aj\r; and
1 1
x ¼ uððf 0 ðaÞÞ1 ÞðyÞ ðx Þ ¼ x ðf 0 ðaÞÞ ðf ðx ÞÞ þ ðf 0 ðaÞÞ ðyÞ:
Therefore,
1 1 1
ðf 0 ðaÞÞ ð0Þ ¼ 0 ¼ ðf 0 ðaÞÞ ðf ðx ÞÞ þ ðf 0 ðaÞÞ ðyÞ
0 1
¼ ðf ðaÞÞ ðy f ðx ÞÞ:
For 4: We first prove that the 1–1 function f−1 from f ðSr ðaÞÞ onto Sr ðaÞ is
differentiable at every point of f ðSr ðaÞÞ: For this purpose, let us take any f(x) in
f ðSr ðaÞÞ; where x is in Sr ðaÞ: Since f : E ! R3 is a C1 function, by Theorem
3.33, f is differentiable at every point of E. Since f is differentiable at every point
of E, and x is in Sr ðaÞ ð EÞ; f 0 ðxÞ exists. Observe that if t is in Sr ðaÞ; then
1
f 0 ðtÞ; ðf 0 ðtÞÞ 2 X: ð0 Þ
Since x is in Sr ðaÞ; from ð0 Þ; f 0 ðxÞ; ðf 0 ðxÞÞ1 exist. Now, we will try to prove that
f−1 is differentiable at f(x) and ðf 1 Þ0 ðf ðxÞÞ ¼ ðf 0 ðxÞÞ1 ; that is,
1 1
1
lim f ðf ðxÞ þ k Þ f 1 ðf ðxÞÞ ðf 0 ðxÞÞ ðkÞ ¼ 0;
k!0 jk j
that is,
1 1 1 0 1
lim f ðyÞ f ðf ðxÞÞ ðf ðxÞÞ ðy f ðxÞÞ ¼ 0;
y ! f ðxÞ jy f ðxÞj
y 2 f ðSr ðaÞÞ
that is,
1 1 1
lim f ðyÞ f 1 ðf ðxÞÞ ðf 0 ðxÞÞ ðy f ðxÞÞ ¼ 0;
y ! f ðxÞ jy f ðxÞj
y 2 f ðSr ðaÞÞ
that is,
1 1 1
lim f ðyÞ x ðf 0 ðxÞÞ ðy f ðxÞÞ ¼ 0:
y ! f ðxÞ j y f ðxÞj
y 2 f ðSr ðaÞÞ
For this purpose, let us take a sequence ff ðan Þg in f ðSr ðaÞÞ such that each an is
in Sr ðaÞ; each f ðan Þ is different from f ðxÞ; and limn!1 f ðan Þ ¼ f ðxÞ: We have to
prove that
1 1 1
lim f ðf ðan ÞÞ x ðf 0 ðxÞÞ ðf ðan Þ f ðxÞÞ ¼ 0;
n!1 jf ðan Þ f ðxÞj
that is,
1 1
lim an x ðf 0 ðxÞÞ ðf ðan Þ f ðxÞÞ ¼ 0:
n!1 jf ðan Þ f ðxÞj
and hence,
1
0 jan xj 2ðf 0 ðaÞÞ jf ðan Þ f ðxÞj ðn ¼ 1; 2; 3; Þ:
This, together with limn!1 f ðan Þ ¼ f ðxÞ; implies that limn!1 jan xj ¼ 0:
Thus,
lim an ¼ x:
n!1
Since each f ðan Þ is different from f ðxÞ; x 6¼ an : Since f 0 ðxÞ exists, each an is in
Sr ðaÞ; x is in Sr ðaÞ; each an is different from x, and limn!1 an ¼ x;
1
lim jf ðan Þ f ðxÞ ððf 0 ðxÞÞðan xÞÞj ¼ 0:
n!1 jan xj
Since
1 1
0 an x ðf 0 ðxÞÞ ðf ðan Þ f ðxÞÞ
jf ðan Þ f ðxÞj
1
1 1
2ðf 0 ðaÞÞ ðf 0 ðxÞÞ jf ðan Þ f ðxÞ ððf 0 ðxÞÞðan xÞÞj ;
jan xj
174 3 Multivariable Differential Calculus
and
1
lim jf ðan Þ f ðxÞ ððf 0 ðxÞÞðan xÞÞj ¼ 0;
n!1 jan xj
so,
1 1
lim an x ðf 0 ðxÞÞ ðf ðan Þ f ðxÞÞ ¼ 0:
n!1 jf ðan Þ f ðxÞj
for every x in Sr ðaÞ: Hence, f−1 is differentiable on the open subset f ðSr ðaÞÞ of R3 :
Since f−1 is differentiable on the open subset f ðSr ðaÞÞ of R3 ; f−1 is continuous on
the open subset f ðSr ðaÞÞ of R3 : Now, we want to prove that the mapping ðf 1 Þ0 :
f ðSr ðaÞÞ ! LðR3 Þ is continuous. Let us take any y in f ðSr ðaÞÞ: Since y is in
f ðSr ðaÞÞ; there exists x in Sr ðaÞ; such that f ðxÞ ¼ y: Since f ðxÞ ¼ y 2 f ðSr ðaÞÞ;
f 1 ðyÞ ¼ x: Since x is in Sr ðaÞ, by ð00 Þ;
0 0 1 1
f 1 ðyÞ ¼ f 1 ðf ðxÞÞ ¼ ðf 0 ðxÞÞ ¼ f 0 f 1 ðyÞ :
0 1
f 1 ðyÞ ¼ f 0 f 1 ðyÞ ;
or,
0
f 1 ðyÞ f 0 f 1 ðyÞ ¼ I;
Hence,
1
ðD2 ðg3 ÞÞðyÞ ¼
ð D
1 1 ð f Þ Þ ð f 1
ðyÞ Þ ð D2 1 ðf
ðf ÞÞ 1
ðyÞÞ ðD3 ðf1 ÞÞðf 1 ðyÞÞ
ðD1 ðf2 ÞÞðf 1 ðyÞÞ ðD2 ðf2 ÞÞðf 1 ðyÞÞ ðD3 ðf2 ÞÞðf 1 ðyÞÞ
ðD1 ðf3 ÞÞðf 1 ðyÞÞ ðD2 ðf3 ÞÞðf 1 ðyÞÞ ðD3 ðf3 ÞÞðf 1 ðyÞÞ
ðD1 ðf1 ÞÞðf 1 ðyÞÞ ðD2 ðf1 ÞÞðf 1 ðyÞÞ
ðD1 ðf3 ÞÞðf 1 ðyÞÞ ðD2 ðf3 ÞÞðf 1 ðyÞÞ
ðD1 ðf1 ÞÞðf 1 ðyÞÞ ðD2 ðf1 ÞÞðf 1 ðyÞÞ
ðD ðf ÞÞðf 1 ðyÞÞ ðD ðf ÞÞðf 1 ðyÞÞ
¼ 1
1 3
1
2 3
1
:
ðD1 ðf1 ÞÞðf ðyÞÞ ðD2 ðf1 ÞÞðf ðyÞÞ ðD3 ðf1 ÞÞðf ðyÞÞ
1 1 1
ðD1 ðf2 ÞÞðf ðyÞÞ ðD2 ðf2 ÞÞðf ðyÞÞ ðD3 ðf2 ÞÞðf ðyÞÞ
ðD1 ðf3 ÞÞðf 1 ðyÞÞ ðD2 ðf3 ÞÞðf 1 ðyÞÞ ðD3 ðf3 ÞÞðf 1 ðyÞÞ
176 3 Multivariable Differential Calculus
where
ðD1 ðf1 ÞÞðf 1 ðyÞÞ ðD2 ðf1 ÞÞðf 1 ðyÞÞ
num D1
ðD1 ðf3 ÞÞðf 1 ðyÞÞ ðD2 ðf3 ÞÞðf 1 ðyÞÞ
ðD1 ðf1 ÞÞðf 1 ðyÞÞ ðD2 ðf1 ÞÞðf 1 ðyÞÞ ðD3 ðf1 ÞÞðf 1 ðyÞÞ
ðD1 ðf2 ÞÞðf 1 ðyÞÞ ðD2 ðf2 ÞÞðf 1 ðyÞÞ ðD3 ðf2 ÞÞðf 1 ðyÞÞ
ðD1 ðf3 ÞÞðf 1 ðyÞÞ ðD2 ðf3 ÞÞðf 1 ðyÞÞ ðD3 ðf3 ÞÞðf 1 ðyÞÞ
ðD1 ðf1 ÞÞðf 1 ðyÞÞ ðD2 ðf1 ÞÞðf 1 ðyÞÞ
ðD1 ðf3 ÞÞðf 1 ðyÞÞ ðD2 ðf3 ÞÞðf 1 ðyÞÞ
0 1
ðD1 ðf1 ÞÞðf 1 ðyÞÞ ðD2 ðf1 ÞÞðf 1 ðyÞÞ ðD3 ðf1 ÞÞðf 1 ðyÞÞ
B C
@D1 ðD1 ðf2 ÞÞðf 1 ðyÞÞ ðD2 ðf2 ÞÞðf 1 ðyÞÞ ðD3 ðf2 ÞÞðf 1 ðyÞÞ A
ðD1 ðf3 ÞÞðf ðyÞÞ ðD2 ðf3 ÞÞðf ðyÞÞ ðD3 ðf3 ÞÞðf ðyÞÞ
1 1 1
ðD11 ðf1 ÞÞðf 1 ðyÞÞ ðD2 ðf1 ÞÞðf 1 ðyÞÞ ðD1 ðf1 ÞÞðf 1 ðyÞÞ ðD12 ðf1 ÞÞðf 1 ðyÞÞ
¼ þ
ðD11 ðf3 ÞÞðf 1 ðyÞÞ ðD2 ðf3 ÞÞðf 1 ðyÞÞ ðD1 ðf3 ÞÞðf 1 ðyÞÞ ðD12 ðf3 ÞÞðf 1 ðyÞÞ
ðD1 ðf1 ÞÞðf 1 ðyÞÞ ðD2 ðf1 ÞÞðf 1 ðyÞÞ ðD3 ðf1 ÞÞðf 1 ðyÞÞ
ðD1 ðf2 ÞÞðf 1 ðyÞÞ ðD2 ðf2 ÞÞðf 1 ðyÞÞ ðD3 ðf2 ÞÞðf 1 ðyÞÞ
ðD1 ðf3 ÞÞðf 1 ðyÞÞ ðD2 ðf3 ÞÞðf 1 ðyÞÞ ðD3 ðf3 ÞÞðf 1 ðyÞÞ
ðD1 ðf1 ÞÞðf 1 ðyÞÞ ðD2 ðf1 ÞÞðf 1 ðyÞÞ
ðD1 ðf3 ÞÞðf 1 ðyÞÞ ðD2 ðf3 ÞÞðf 1 ðyÞÞ
0
ðD11 ðf1 ÞÞðf 1 ðyÞÞ ðD2 ðf1 ÞÞðf 1 ðyÞÞ ðD3 ðf1 ÞÞðf 1 ðyÞÞ
B
@ ðD11 ðf2 ÞÞðf 1 ðyÞÞ ðD2 ðf2 ÞÞðf 1 ðyÞÞ ðD3 ðf2 ÞÞðf 1 ðyÞÞ
ðD11 ðf3 ÞÞðf 1 ðyÞÞ ðD2 ðf3 ÞÞðf 1 ðyÞÞ ðD3 ðf3 ÞÞðf 1 ðyÞÞ
ðD1 ðf1 ÞÞðf 1 ðyÞÞ ðD12 ðf1 ÞÞðf 1 ðyÞÞ ðD3 ðf1 ÞÞðf 1 ðyÞÞ
þ ðD1 ðf2 ÞÞðf 1 ðyÞÞ ðD12 ðf2 ÞÞðf 1 ðyÞÞ ðD3 ðf2 ÞÞðf 1 ðyÞÞ
ðD1 ðf3 ÞÞðf 1 ðyÞÞ ðD12 ðf3 ÞÞðf 1 ðyÞÞ ðD3 ðf3 ÞÞðf 1 ðyÞÞ
1
ðD1 ðf1 ÞÞðf 1 ðyÞÞ ðD2 ðf1 ÞÞðf 1 ðyÞÞ ðD13 ðf1 ÞÞðf 1 ðyÞÞ
C
ðD1 ðf2 ÞÞðf 1 ðyÞÞ ðD2 ðf2 ÞÞðf 1 ðyÞÞ ðD13 ðf2 ÞÞðf 1 ðyÞÞ A;
ðD1 ðf3 ÞÞðf 1 ðyÞÞ ðD2 ðf3 ÞÞðf 1 ðyÞÞ ðD13 ðf3 ÞÞðf 1 ðyÞÞ
3.3 Inverse Function Theorem 177
and
ðD1 ðf1 ÞÞðf 1 ðyÞÞ ðD2 ðf1 ÞÞðf 1 ðyÞÞ ðD3 ðf1 ÞÞðf 1 ðyÞÞ 2
denom ðD1 ðf2 ÞÞðf 1 ðyÞÞ ðD2 ðf2 ÞÞðf 1 ðyÞÞ ðD3 ðf2 ÞÞðf 1 ðyÞÞ :
ðD1 ðf3 ÞÞðf 1 ðyÞÞ ðD2 ðf3 ÞÞðf 1 ðyÞÞ ðD3 ðf3 ÞÞðf 1 ðyÞÞ
It follows that ðD12 ðg3 ÞÞðyÞ exists. So D12 ðg3 Þ : f ðSr ðaÞÞ ! R:
Next, since each of the nine second-order partial derivatives Djk ðfi Þ : Sr ðaÞ ! R
of fi is continuous, and f−1is continuous on f ðSr ðaÞÞ; by the above formula for num
and denom, we find that num and denom are continuous functions on f ðSr ðaÞÞ; and
num Þ is continuous on f ðS ðaÞÞ:
hence, D12 ðg3 Þð¼ denom r
Similarly, all the other Djk ðgi Þ : f ðSr ðaÞÞ ! R are continuous. This proves that
f−1 from f(U) onto U is a C2 function, etc. h
Note 3.37 The result similar to Theorem 3.36 can be proved as above for Rn in
place of R3 : This theorem is known as the inverse function theorem.
Theorem 3.38 Let E be an open subset of Rn : Let f : E ! Rn be a function. If
1. f is a C1 function,
2. f 0 ðaÞ is invertible for every a in E,
then f is an open mapping (i.e., f-image of every open subset of E in an open set).
Proof Let us take any open subset G of E. We have to prove that f(G) is an open
set. For this purpose, let us take any f(a) in f(G) where a is in G. Since a 2 G E,
by the given assumption, f 0 ðaÞ is invertible. Now, by the inverse function theorem,
there exists an open neighborhood U of a such that U is contained in G and f(U) is
open in R3 : Since a 2 U G; f ðaÞ 2 f ðUÞ f ðGÞ: Since f ðaÞ 2 f ðUÞ; and f(U) is
open in R3 ; f(U) is an open neighborhood of f(a). Since f(U) is an open neigh-
borhood of f(a) and f ðUÞ f ðGÞ; f(a) is an interior point of f(G), and hence, f(G) is
an open set. h
Ax : R3 ! R3
Ay : R2 ! R3
Ay ðaÞ Að0; 0; 0; a1 ; a2 Þ:
Note 3.39 Similar notations and results as above can be supplied for Rn in place of
R3 and Rm in place of R2 :
Note 3.40 Let A 2 LðR3þ2 ; R3 Þ; and let
2 3
a11 a12 a13 a14 a15
½ A ¼ 4 a21 a22 a23 a24 a25 5 :
a31 a32 a33 a34 a35 35
So,
Hence,
2 3
a11 a12 a13
½Ax ¼ 4 a21 a22 a23 5 :
a31 a32 a33 33
Next,
Hence,
2 3
a14 a15
Ay ¼ 4 a24 a25 5 :
a34 a35 32
then
2 3 2 3
a11 a12 a13 a14 a15
½Ax ¼ 4 a21 a22 a23 5 and Ay ¼ 4 a24 a25 5 :
a31 a32 a33 33 a34 a35 32
and hence,
1
ðAx Þ1 ðð1; 0; 0ÞÞ ¼ ðA11 ; A12 ; A13 Þ;
D
1
ðAx Þ1 ðð0; 1; 0ÞÞ ¼ ðA21 ; A22 ; A23 Þ;
D
1
ðAx Þ1 ðð0; 0; 1ÞÞ ¼ ðA31 ; A32 ; A33 Þ:
D
3.4 Implicit Function Theorem 181
then
1
ðAx Þ1 ðð1; 0; 0ÞÞ ¼ ðA11 ; A12 ; A13 Þ;
D
1
ðAx Þ1 ðð0; 1; 0ÞÞ ¼ ðA21 ; A22 ; A23 Þ;
D
1
ðAx Þ1 ðð0; 0; 1ÞÞ ¼ ðA31 ; A32 ; A33 Þ:
D
and Ax is invertible,
0 ¼ ðAx Þ1 ð0Þ ¼ ðAx Þ1 Ax ðhÞ þ Ay ðkÞ
¼ ðAx Þ1 ðAx ðhÞÞ þ ðAx Þ1 Ay ðkÞ
¼ ðAx Þ1 Ax ðhÞ þ ðAx Þ1 Ay ðkÞ ¼ IðhÞ þ ðAx Þ1 Ay ðkÞ
¼ h þ ðAx Þ1 Ay ðkÞ :
182 3 Multivariable Differential Calculus
It follows that
h¼ ðAx Þ1 Ay ðkÞ :
Similarly,
h1 ¼ ðAx Þ1 Ay ðkÞ ;
which contradicts the assumption h 6¼ h1 : Further, we have shown that for a given
k, the value of h is ðððAx Þ1 ÞðAy ðkÞÞÞ: h
Note 3.42 The result similar to Theorem 3.41 can be proved as above for Rn in
place of R3 and Rm in place of R2 :
Theorem 3.43 Let E be an open subset of R3þ2 : Let f : E ! R3 : Let a 2 R3 ; b 2
R2 ; and ða; bÞ 2 E: If
(i) f is a C 1 function,
(ii) ðf 0 ðða; bÞÞÞx is invertible,
(iii) f ðða; bÞÞ ¼ 0;
then there exist an open neighborhood Uð EÞ of ða; bÞ in R3þ2 ; an open neigh-
borhood W of b in R2 ; and a function
g : W ! R3
such that
1. for every y in W, ðgðyÞ; yÞ 2 U;
2. for every y in W, f ððgðyÞ; yÞÞ ¼ 0;
3. gðbÞ ¼ a;
4. g is a C 1 function,
5. for every y in W, g0 ðy Þ ¼ ððf 0 ððgðy Þ; y ÞÞÞx Þ1 ðf 0 ððgðy Þ; y ÞÞÞy
6. g0 ðbÞ ¼ ððf 0 ðða; bÞÞÞx Þ1 ðf 0 ðða; bÞÞÞy
7. if f is a C 2 function, then g is a C 2 function, etc.
F : E ! R3þ2
F ððx; yÞÞ ðf ððx; yÞÞ; yÞ ¼ ððf1 ððx; yÞÞ; f2 ððx; yÞÞ; f3 ððx; yÞÞÞ; ðy1 ; y2 ÞÞ
¼ ðf1 ððx; yÞÞ; f2 ððx; yÞÞ; f3 ððx; yÞÞ; y1 ; y2 Þ:
ðx; yÞ 7! f1 ððx; yÞÞ; ðx; yÞ 7! f2 ððx; yÞÞ; ðx; yÞ 7! f3 ððx; yÞÞ; ðx; yÞ 7! y1 ;
ðx; yÞ 7! y2
from E to R:
We want to apply the inverse function theorem on F. For this purpose, we first
prove
(a) F is a C1 function,
(b) F 0 ðða; bÞÞ is invertible.
For (a): Since f is a C 1 function, for i ¼ 1; 2; 3 and j ¼ 1; 2; 3; 4; 5ð¼ 3 þ 2Þ,
each function Dj fi : E ! R exists and is continuous. Also, ðx; yÞ 7! y1 and
ðx; yÞ 7! y2 are smooth functions. This shows that F is a C1 function.
For (b): Here,
and
so,
2 3
ðD1 f1 Þðða; bÞÞ ðD2 f1 Þðða; bÞÞ ðD3 f1 Þðða; bÞÞ
ðf 0 ðða; bÞÞÞx ¼ 4 ðD1 f2 Þðða; bÞÞ ðD2 f2 Þðða; bÞÞ ðD3 f2 Þðða; bÞÞ 5:
ðD1 f3 Þðða; bÞÞ ðD2 f3 Þðða; bÞÞ ðD3 f3 Þðða; bÞÞ
is continuous. Further, since det½ðf 0 ðða; bÞÞÞx is nonzero, there exists a neighborhood
U of ða; bÞ such that det½ðf 0 ððx; yÞÞÞx is nonzero for every ðx; yÞ in U : It follows that
ðf 0 ððx; yÞÞÞx is invertible for every ðx; yÞ in U : Now, by the inverse function theorem
and Theorem 3.38, there exists an open neighborhood U of ða; bÞ such that
(1) U is contained in E,
(2) F is 1–1 on U,
(3) FðUÞ is open in R3þ2 ;
(4) the 1–1 function F 1 from FðUÞ onto U is a C 1 function,
(5) U is contained in U :
Put
W y : y 2 R2 and for some x in R3 ðx ; yÞ 2 U; f ððx ; yÞÞ ¼ 0 :
Thus,
W ¼ y : y 2 R2 and ð0; yÞ 2 FðUÞ :
Now, we will show that W is an open subset of R2 : For this purpose, let us take
any y0 2 W: Since y0 2 W, ð0; y0 Þ 2 FðUÞ: Since ð0; y0 Þ 2 FðUÞ; and FðUÞ is
3.4 Implicit Function Theorem 185
For this purpose, let us take any y 2 W: The existence of x is clear from the
definition of W.
Uniqueness: If not, otherwise, let there exist x ; x in R3 such that ðx ; yÞ 2
U; f ððx ; yÞÞ ¼ 0; ðx ; yÞ 2 U; f ððx ; yÞÞ ¼ 0; x 6¼ x : We have to arrive at a
contradiction. Here, x 6¼ x ; so ðx ; yÞ 6¼ ðx ; yÞ: Since ðx ; yÞ 6¼ ðx ; yÞ; ðx ; yÞ 2
U; ðx ; yÞ 2 U; and F is 1–1 on U, ð0; yÞ ¼ ðf ðx ; yÞ; yÞ ¼ Fððx ; yÞÞ 6¼
Fððx ; yÞÞ ¼ ðf ððx ; yÞÞ; yÞ ¼ ð0; yÞ: Thus, ð0; yÞ 6¼ ð0; yÞ; which is a contradiction.
Thus, we have shown that if y 2 W, then there exists a unique x in R3 such that
ðx ; yÞ 2 U; f ððx ; yÞÞ ¼ 0:
This proves 1, 2.
Further, since b 2 W; ðgðbÞ; bÞ 2 U; and f ðða; bÞÞ ¼ 0 ¼ f ððgðbÞ; bÞÞ: Since
f ðða; bÞÞ ¼ f ððgðbÞ; bÞÞ; Fðða; bÞÞ ¼ ðf ðða; bÞÞ; bÞ ¼ ðf ððgðbÞ; bÞÞ; bÞ ¼ FððgðbÞ; bÞÞ:
Since Fðða; bÞÞ ¼ FððgðbÞ; bÞÞ; ða; bÞ 2 U; ðgðbÞ; bÞ 2 U; and F is 1–1 on U,
ða; bÞ ¼ ðgðbÞ; bÞ; and hence, gðbÞ ¼ a:
This proves 3.
4: Let g1 : W ! R; g2 : W ! R; g3 : W ! R be the component functions of
g, that is, gððy1 ; y2 ÞÞ ¼ ðg1 ððy1 ; y2 ÞÞ; g2 ððy1 ; y2 ÞÞ; g3 ððy1 ; y2 ÞÞÞ for every ðy1 ; y2 Þ in
Wð R2 Þ: Let
F 1 ððx; yÞÞ ¼ ðh1 ððx; yÞÞ; h2 ððx; yÞÞ; h3 ððx; yÞÞ; h4 ððx; yÞÞ; h5 ððx; yÞÞÞ
Now,
Further, since y 7! ð0; y Þ and ðx; yÞ 7! ðD5 h1 Þððx; yÞÞ are continuous functions,
their composite function y 7! ðD5 h1 Þðð0; y ÞÞð¼ ðD2 g1 Þððy ÞÞÞ is continuous.
Thus, we have shown that D2 g1 : W ! R exists and is continuous. Similarly, all
other Dj gi : W ! R exist and are continuous.
3.4 Implicit Function Theorem 187
This proves 4.
5: Let us fix any y ðy1 ; y2 Þ in W. Since y ¼ ðy1 ; y2 Þ is in W, ðgðy Þ; y Þ 2
U; f ððgðy Þ; y ÞÞ ¼ 0: Let us define a function
hy : W ! R3þ2
hy ðyÞ ðgðyÞ; yÞ ¼ ðg1 ððy1 ; y2 ÞÞ; g2 ððy1 ; y2 ÞÞ; g3 ððy1 ; y2 ÞÞ; y1 ; y2 Þ:
Hence,
0
ð0; 0; 0Þ ¼ 0ðð1; 0ÞÞ ¼ ðf 0 ððgðyÞ; yÞÞÞ hy ðyÞ ðð1; 0ÞÞ
0
¼ ðf 0 ððgðyÞ; yÞÞÞ hy ðyÞ ðð1; 0ÞÞ
¼ ðf 0 ððgðyÞ; yÞÞÞðððD1 g1 ÞðyÞ; ðD1 g2 ÞðyÞ; ðD1 g3 ÞðyÞ; 1; 0ÞÞ
¼ ððD1 g1 ÞðyÞÞððf 0 ððgðyÞ; yÞÞÞðð1; 0; 0; 0; 0ÞÞÞ
þ ððD1 g2 ÞðyÞÞððf 0 ððgðyÞ; yÞÞÞðð0; 1; 0; 0; 0ÞÞÞ
þ ððD1 g3 ÞðyÞÞððf 0 ððgðyÞ; yÞÞÞðð0; 0; 1; 0; 0ÞÞÞ
þ 1ðf 0 ððgðyÞ; yÞÞÞðð0; 0; 0; 1; 0ÞÞ þ 0
¼ ððD1 g1 ÞðyÞÞððD1 f1 ÞððgðyÞ; yÞÞ; ðD1 f2 ÞððgðyÞ; yÞÞ; ðD1 f3 ÞððgðyÞ; yÞÞÞ
þ ððD1 g2 ÞðyÞÞððD2 f1 ÞððgðyÞ; yÞÞ; ðD2 f2 ÞððgðyÞ; yÞÞ; ðD2 f3 ÞððgðyÞ; yÞÞÞ
þ ððD1 g3 ÞðyÞÞððD3 f1 ÞððgðyÞ; yÞÞ; ðD3 f2 ÞððgðyÞ; yÞÞ; ðD3 f3 ÞððgðyÞ; yÞÞÞ
þ ððD4 f1 ÞððgðyÞ; yÞÞ; ðD4 f2 ÞððgðyÞ; yÞÞ; ðD4 f3 ÞððgðyÞ; yÞÞÞ
188 3 Multivariable Differential Calculus
or,
ð0; 0; 0Þ ¼ ððD1 g1 ÞðyÞÞððD1 f1 ÞððgðyÞ; yÞÞ; ðD1 f2 ÞððgðyÞ; yÞÞ; ðD1 f3 ÞððgðyÞ; yÞÞÞ
þ ððD1 g2 ÞðyÞÞððD2 f1 ÞððgðyÞ; yÞÞ; ðD2 f2 ÞððgðyÞ; yÞÞ; ðD2 f3 ÞððgðyÞ; yÞÞÞ
þ ððD1 g3 ÞðyÞÞððD3 f1 ÞððgðyÞ; yÞÞ; ðD3 f2 ÞððgðyÞ; yÞÞ; ðD3 f3 ÞððgðyÞ; yÞÞÞ
þ ððD4 f1 ÞððgðyÞ; yÞÞ; ðD4 f2 ÞððgðyÞ; yÞÞ; ðD4 f3 ÞððgðyÞ; yÞÞÞ:
0 ¼ ððD1 g1 Þðy ÞÞððD1 fi Þððgðy Þ; y ÞÞÞ þ ððD1 g2 Þðy ÞÞððD2 fi Þððgðy Þ; y ÞÞÞ
þ ððD1 g3 Þðy ÞÞððD3 fi Þððgðy Þ; y ÞÞÞ þ ðD4 fi Þððgðy Þ; y ÞÞ ðÞ:
that is,
that is,
(i) ðððf 0 ððgðy Þ; y ÞÞÞx Þðg0 ðy ÞÞÞðð1; 0ÞÞ ¼ ððf 0 ððgðy Þ; y ÞÞÞy Þðð1; 0ÞÞ;
and
(ii) ðððf 0 ððgðy Þ; y ÞÞÞx Þðg0 ðy ÞÞÞðð0; 1ÞÞ ¼ ððf 0 ððgðy Þ; y ÞÞÞy Þðð0; 1ÞÞ:
For (i):
LHS ¼ ðf 0 ððgðy Þ; y ÞÞÞx ðg0 ðy ÞÞ ðð1; 0ÞÞ ¼ ðf 0 ððgðy Þ; y ÞÞÞx ððg0 ðy ÞÞðð1; 0ÞÞÞ
¼ ðf 0 ððgðy Þ; y ÞÞÞx ðððD1 g1 Þðy Þ; ðD1 g2 Þðy Þ; ðD1 g3 Þðy ÞÞÞ
¼ ðf 0 ððgðy Þ; y ÞÞÞðððD1 g1 Þðy Þ; ðD1 g2 Þðy Þ; ðD1 g3 Þðy Þ; 0; 0ÞÞ
¼ ððD1 g1 Þðy ÞÞððD1 f1 Þððgðy Þ; y ÞÞ; ðD1 f2 Þððgðy Þ; y ÞÞ; ðD1 f3 Þððgðy Þ; y ÞÞÞ
þ ððD1 g2 Þðy ÞÞððD2 f1 Þððgðy Þ; y ÞÞ; ðD2 f2 Þððgðy Þ; y ÞÞ; ðD2 f3 Þððgðy Þ; y ÞÞÞ
þ ððD1 g3 Þðy ÞÞððD3 f1 Þððgðy Þ; y ÞÞ; ðD3 f2 Þððgðy Þ; y ÞÞ; ðD3 f3 Þððgðy Þ; y ÞÞÞ
¼ ððD4 f1 Þððgðy Þ; y ÞÞ; ðD4 f2 Þððgðy Þ; y ÞÞ; ðD4 f3 Þððgðy Þ; y ÞÞÞ;
3.4 Implicit Function Theorem 189
by ðÞ: Now,
RHS ¼ ðf 0 ððgðy Þ; y ÞÞÞy ðð1; 0ÞÞ ¼ ðf 0 ððgðy Þ; y ÞÞÞy ðð1; 0ÞÞ
¼ ððf 0 ððgðy Þ; y ÞÞÞðð0; 0; 01; 0ÞÞÞ
¼ ððD4 f1 Þððgðy Þ; y ÞÞ; ðD4 f2 Þððgðy Þ; y ÞÞ; ðD4 f3 Þððgðy Þ; y ÞÞÞ
¼ ððD4 f1 Þððgðy Þ; y ÞÞ; ðD4 f2 Þððgðy Þ; y ÞÞ; ðD4 f3 Þððgðy Þ; y ÞÞÞ:
This proves 6.
7: Let f be a C2 function. We have to prove that g : W ! R3 is a C 2 function,
that is, each of the 12 second-order partial derivatives Djk gi : W ! R exists and is
continuous. We will try to prove that D12 g3 : W ! R exists and is continuous. Let
us take any y in W.
By 5,
1
g0 ðyÞ ¼ ðf 0 ððgðyÞ; yÞÞÞx ðf 0 ððgðyÞ; yÞÞÞy
so,
where
ðD1 f1 ÞððgðyÞ; yÞÞ ðD2 f1 ÞððgðyÞ; yÞÞ ðD3 f1 ÞððgðyÞ; yÞÞ
D ðD1 f2 ÞððgðyÞ; yÞÞ ðD2 f2 ÞððgðyÞ; yÞÞ ðD3 f2 ÞððgðyÞ; yÞÞ ;
ðD1 f3 ÞððgðyÞ; yÞÞ ðD2 f3 ÞððgðyÞ; yÞÞ ðD3 f3 ÞððgðyÞ; yÞÞ
and Aij denotes the cofactor of ðDj fi ÞððgðyÞ; yÞÞ in the above determinant. Hence,
1
ðD1 g3 ÞðyÞ ¼ ðððD4 f1 ÞððgðyÞ; yÞÞÞA13 þ ððD4 f2 ÞððgðyÞ; yÞÞÞA23 þ ððD4 f3 ÞððgðyÞ; yÞÞÞA33 Þ:
D
g : W ! R2
such that
1. for every x in W, ðx; gðxÞÞ 2 U;
2. for every x in W, f ðx; gðxÞÞ ¼ 0;
3. gðaÞ ¼ b;
4. g is a C 1 function,
5. for every x in W, g0 ðx Þ ¼ ððf 0 ððx ; gðx ÞÞÞÞy Þ1 ðf 0 ððx ; gðx ÞÞÞÞx
6. g0 ðaÞ ¼ ððf 0 ðða; bÞÞÞy Þ1 ðf 0 ðða; bÞÞÞx
7. if f is a C 2 function, then g is a C 2 function, etc.
3.5 Constant Rank Theorem (Easy Version) 191
Here, i1 is called the first injection of factor into product. Similarly, by the
second injection i2 of factor into product, we mean the function i2 : Rn ! Rm Rn
defined as follows: For every y in Rn
p1 ðx; yÞ x:
Here, p1 is called the first projection from product onto factor. Similarly, by the
second projection p2 from product onto factor, we mean the function p2 :
Rm Rn ! Rn defined as follows: For every x in Rm and for every y in Rn
p2 ðx; yÞ y:
ðf 0 ða1 ; a2 ; a3 ; b1 ; b2 ÞÞ i2 : R2 ! R2
then
(i) aðW1 Þ is an open neighborhood of ða; bÞ in f 1 ð0; 0Þ;
(ii) a is a homeomorphism from W1 onto aðW1 Þ;
(iii) a ¼ u s1 and a1 ¼ p1 on their common domain,
(iv) a; a1 are smooth functions.
it follows that
and hence, ðf 0 ða1 ; a2 ; a3 ; b1 ; b2 ÞÞy is invertible. Now, by the implicit function the-
orem, there exist an open neighborhood W1 of ða1 ; a2 ; a3 Þ in R3 ; an open neigh-
borhood W2 of ðb1 ; b2 Þ in R2 , and a smooth function ðg1 ; g2 Þ g : W1 ! R2 such
that W1 W2 U and for every ðx1 ; x2 ; x3 Þ in W1 ;
and
h2 ðx1 ; x2 ; x3 ÞÞ ¼ ð0; 0Þ; where h ðh1 ; h2 Þ: We have to prove that g ¼ h; that is, for
every ðx1 ; x2 ; x3 Þ in W1 ;
gðx1 ; x2 ; x3 Þ ¼ hðx1 ; x2 ; x3 Þ:
Since
so,
and hence,
Hence, g ¼ h:
This proves 2(c).
(d) (i): For every ðx1 ; x2 ; x3 Þ in W1 ; f ðx1 ; x2 ; x3 ; g1 ðx1 ; x2 ; x3 Þ; g2 ðx1 ; x2 ; x3 ÞÞ ¼
ð0; 0Þ; so
Further, since
so
(Reason: For this purpose, let us take any aðx1 ; x2 ; x3 Þ 2 LHS; where ðx1 ; x2 ; x3 Þ
is in W1 : Here,
ðw1 ; w2 ; w3 ; t1 ; t2 Þ ¼ wðx1 ; x2 ; x3 ; y1 ; y2 Þ
¼ ðx1 ; x2 ; x3 ; f1 ðx1 ; x2 ; x3 ; y1 ; y2 Þ; f2 ðx1 ; x2 ; x3 ; y1 ; y2 ÞÞ
¼ ðx1 ; x2 ; x3 ; 0; 0Þ:
ðx1 ; x2 ; x3 Þ ¼ ðy1 ; y2 ; y3 Þ:
For (B): Since g : W1 ! R2 is a smooth function, and
a ¼ u s1 : Next, since
ðf 0 ða; bÞÞ i2 : Rp ! Rp
then
(i) aðW1 Þ is an open neighborhood of ða; bÞ in f 1 ð0Þ;
(ii) a is a homeomorphism from W1 onto aðW1 Þ;
(iii) a ¼ u s1 and a1 ¼ p1 on their common domain,
(iv) a; a1 are smooth functions.
e x
1
if 0\x;
f ðxÞ
0 if x 0:
Thus, f ðxÞ is positive if and only if x is positive, and f ðxÞ is never negative. We
shall first try to prove by induction (a): For every x [ 0; and for every nonnegative
integer k, the kth derivative f ðkÞ ðxÞ is of the form ðp2k ð1xÞÞex for some polynomial
1
p2k ðyÞ of degree 2k in y. Let us denote the statement, f ðkÞ ðxÞ is of the form p2k ð1xÞex
1
For 1: Pð1Þ means f 0 ðxÞ is of the form p2 ð1xÞex for some polynomial p2 ðyÞ of
1
p2k ðyÞ of degree 2k in y. We have to prove that Pðk þ 1Þ is true, that is, f ðkþ1Þ ðxÞ is
of the form p2kþ2 ð1xÞex for some polynomial p2kþ2 ðyÞ of degree 2k þ 2 in y. Here,
1
0
d d 1 1
f ðkþ1Þ ðxÞ ¼ f ðkÞ ðxÞ ¼ f ðkÞ ðxÞ ¼ p2k e x
dx dx x
1 1 1 1 1x 1
¼ p02k e x þ p
2k e
x x2 x x2
1 1 1 1
¼ p2k p02k e x 2
x x x
1 1
¼ q2kþ2 e x;
x
200 3 Multivariable Differential Calculus
where q2kþ2 ðyÞ ðp2k ðyÞ p02k ðyÞÞy2 ; which is a polynomial of degree 2k þ 2
in y. This proves 2.
Hence, by the principle of mathematical induction, PðkÞ is true for every positive
integer k. This proves (a). Now, we shall try to prove (b): f is C1 on R and that
f ðkÞ ð0Þ ¼ 0 for all k 0: Here, for x\0; f 0 ðxÞ ¼ 0; and for 0\x; f 0 ðxÞ ¼ ddx ex ¼
1
1 1x
x2 e : Next,
and
f ð0 þ eÞ f ð0Þ f ðeÞ 0
lim ¼ lim
e!0 e e!0 e
e[0 e[0
e e
1
f ðeÞ
¼ lim ¼ lim
e!0 e e!0 e
e[0 e[0
t 1
¼ lim ¼ lim ¼ 0:
t!1 et t!1 et
Hence,
f ð0 þ hÞ f ð0Þ
lim ¼ 0;
h!0 h
0 if x 0;
f 0 ðxÞ ¼ 1 1x
x2 e if 0\x:
Now,
f 0 ð0 eÞ f 0 ð0Þ f 0 ðeÞ 0 f 0 ðeÞ 0
lim ¼ lim ¼ lim ¼ lim
e!0 e e!0 e e!0 e e!0 e
e[0 e[0 e[0 e[0
¼ 0;
and
1 1e
f 0 ð0 þ eÞ f 0 ð0Þ f 0 ðeÞ 0 f 0 ðeÞ e2 e 1 t3 3t2
lim ¼ lim ¼ lim ¼ lim ¼ lim 1 ¼ lim t ¼ lim
e!0 e e!0 e e!0 e e!0 e e ! 0 e ee
3 t!1 e t!1 et
e[0 e[0 e[0 e[0 e[0
3 2t 321
¼ lim ¼ lim ¼ 0:
t!1 et t!1 et
Hence,
f 0 ð0 þ hÞ f 0 ð0Þ
lim ¼ 0;
h!0 h
x x x
Now,
¼ 0;
202 3 Multivariable Differential Calculus
and
2
f 00 ð0 þ eÞ f 00 ð0Þ f 00 ðeÞ 0 f 00 ðeÞ þ e14 e e
1
e3
lim ¼ lim ¼ lim ¼ lim
e!0 e e!0 e e!0 e e!0 e
e[0 e[0 e[0 e[0
2 1 1 2t4 þ t5 5!
¼ lim þ 1 ¼ lim ¼ lim t ¼ 0:
e!0 e 4 e ee t!1
5 et t!1 e
e[0
Hence,
f 00 ð0 þ hÞ f 00 ð0Þ
lim ¼ 0;
h!0 h
0 if x 0;
f 000 ðxÞ ¼
þ 6 þ x16 ex
1
6
x4 x5
if 0\x:
Similarly, f ð4Þ ð0Þ ¼ 0; f 5 ð0Þ ¼ 0; etc. This completes the proof of (b). Thus,
f : R ! R is a smooth function.
Now, let us define a function g : R3 ! R as follows: For every x ðx1 ; x2 ; x3 Þ
in R3 ;
f e2 ðx1 Þ2 þðx2 Þ2 þðx3 Þ2
gðxÞ :
f e2 ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 þ f ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 14 e2
and hence,
0 1
f e2 ðx1 Þ2 þðx2 Þ2 þðx3 Þ2
gðxÞ@¼ A is 0:
f e2 ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 þ f ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 14 e2
and hence,
0
f e2 ðx1 Þ2 þðx2 Þ2 þðx3 Þ2
gð x Þ @ ¼
f e2 ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 þ f ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 14 e2
1
f e 2 ð x1 Þ 2 þ ð x2 Þ 2 þ ð x3 Þ 2
¼ A is 1:
f e2 ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 þ 0
Finally, for every x ðx1 ; x2 ; x3 Þ satisfying ðx1 Þ2 þ ðx2 Þ2 þ ðx3 Þ2 \e2 ; we have
e ððx1 Þ2 þ ðx2 Þ2 þ ðx3 Þ2 Þ is positive, and hence, f ðe2 ððx1 Þ2 þ ðx2 Þ2 þ ðx3 Þ2 ÞÞ
2
is positive. Thus, for every x ðx1 ; x2 ; x3 Þ satisfying ðx1 Þ2 þ ðx2 Þ2 þ ðx3 Þ2 \e2 ;
0 1
f e2 ðx1 Þ2 þðx2 Þ2 þðx3 Þ2
gðxÞ@¼ A
f e2 ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 þ f ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 14 e2
Note 3.51 We shall try to prove that there exists a smooth function h : R ! R
such that
1. hðtÞ ¼ 1 if t 1;
2. 0\hðtÞ\1 if 1\t\2;
3. hðtÞ ¼ 0 if 2 t:
In the proof of Theorem 3.50, we have seen that the function f : R ! R is defined
as follows: For every t in R;
e t
1
if 0\t
f ðtÞ
0 if t 0;
e2t
1
hðtÞ ¼ ¼ 1:
e2t þ 0
1
0\et1 ¼ f ðt 1Þ:
1
3.6 Smooth Bump Functions 205
If 1\t\2; then 0\2 t; and hence, 0\e2t ¼ f ð2 tÞ: It follows that for
1
1\t\2;
e2t
1
Note 3.52 We shall try to prove that there exists a function H : R3 ! ½0; 1 such
that
1. H is smooth,
2. HðxÞ ¼ 1; if x is in the closed ball B1 ½0ð¼ fx : x 2 R3 ; jxj 1gÞ;
3. the closure ðH 1 ðð0; 1ÞÞ of H 1 ðð0; 1Þ is equal to the closed ball
B2 ½0ð¼ fx : x 2 R3 ; jxj 2gÞ:
Let us define the function H : R3 ! ½0; 1 as follows: For every x ðx1 ; x2 ; x3 Þ
in R3 ;
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
HðxÞ h ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 ;
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
For 1: Since x ðx1 ; x2 ; x3 Þ 7! ð ðx1 Þ2 þ ðx2 Þ2 þ ðx3 Þ2 Þ is a smooth function
over R3 fð0; 0; 0Þg; and h : R ! ½0; 1 is a smooth function, their com-
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
posite x 7! hð ðx1 Þ2 þ ðx2 Þ2 þ ðx3 Þ2 Þð¼ HðxÞÞ is a smooth function from
206 3 Multivariable Differential Calculus
and
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2 2
h ð0 tÞ þð0Þ þ ð0Þ h ð0Þ2 þ ð0Þ2 þ ð0Þ2
lim
t!0 t
t[0
hðtÞ hð0Þ 11
¼ lim ¼ lim ¼ lim 0 ¼ 0:
t!0 t t ! 0 t t!0
t[0 t[0 t[0
Since
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
h ð0 þ tÞ2 þð0Þ2 þ ð0Þ2 h ð0Þ2 þ ð0Þ2 þ ð0Þ2
lim ¼0
t!0 t
t[0 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2 2
h ð0 tÞ þð0Þ þ ð0Þ h ð0Þ2 þ ð0Þ2 þ ð0Þ2
¼ lim ;
t!0 t
t[0
ðD1 HÞð0; 0; 0Þ exists, and its value is 0. Also, for ðx1 ; x2 ; x3 Þ 6¼ ð0; 0; 0Þ;
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1
ðD1 H Þðx1 ; x2 ; x3 Þ ¼ h0 ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð2x1 þ 0 þ 0Þ
2 ðx1 Þ þðx2 Þ2 þðx3 Þ2
2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
x1
¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi h0 ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 :
2 2 2
ðx1 Þ þðx2 Þ þðx3 Þ
3.6 Smooth Bump Functions 207
Thus,
8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
< pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 0
x1
h ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 if ðx1 ; x2 ; x3 Þ 6¼ ð0; 0; 0Þ
ðD1 H Þðx1 ; x2 ; x3 Þ ¼ ðx1 Þ2 þðx2 Þ2 þðx3 Þ2
:
0 if ðx1 ; x2 ; x3 Þ ¼ ð0; 0; 0Þ:
f ð2 t Þ e x
1
if 0\x;
hðtÞ ¼ : Also f ðxÞ and f 0 ðxÞ
f ð 2 t Þ þ f ð t 1Þ 0 if x 0;
0 if x 0;
¼ 1 1x
x 2 e if 0\x:
So
and hence,
ðf 0 ð2 0ÞÞð1Þðf ð2 0Þ þ f ð0 1ÞÞ ðf ð2 0ÞÞðf 0 ð2 0Þ þ f 0 ð0 1ÞÞ
h0 ð0Þ ¼
ðf ð2 0Þ þ f ð0 1ÞÞ2
ðf ð2ÞÞð1Þðf ð2Þ þ f ð1ÞÞ ðf ð2ÞÞðf 0 ð2Þ þ f 0 ð1ÞÞ
0
¼
ðf ð2Þ þ f ð1ÞÞ2
ðf ð2ÞÞð1Þðf ð2Þ þ 0Þ ðf ð2ÞÞðf 0 ð2Þ þ 0Þ
0
0
¼ ¼ ¼ 0:
ðf ð2Þ þ 0Þ2 1 2
e2
and h0 ð0Þ ¼ 0;
lim ðD1 H Þðx1 ; x2 ; x3 Þ
ðx1 ;x2 ;x3 Þ!ð0;0;0Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
x1
¼ lim qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi h0 ðx1 Þ2 þðx2 Þ2 þðx3 Þ2 ¼ 0 ¼ ðD1 H Þð0; 0; 0Þ:
ðx1 ;x2 ;x3 Þ!ð0;0;0Þ
ðx1 Þ þðx2 Þ2 þðx3 Þ2
2
208 3 Multivariable Differential Calculus
and
1 1 a ¼ 1 1 jaj ¼ 1 1 2 ¼ 2 1 2 ð1; 2Þ:
2n 2n 2n n
Since
1 1 a 2 ð1; 2Þ;
2n
1 1
H 1
a ¼h 1 a 2 ð0; 1Þ ð0; 1;
2n 2n
and hence,
1
1 a 2 H 1 ðð0; 1Þ:
2n
3.6 Smooth Bump Functions 209
and
1
lim 1 a ¼ a; so a 2 H 1 ðð0; 1Þ :
n!1 2n
As above, we can show that there exists a function H : Rn ! ½0; 1 such that
1. H is smooth,
2. HðxÞ ¼ 1; if x is in the closed ball B1 ½0ð¼ fx : x 2 Rn ; jxj 1gÞ;
3. the closure ðH 1 ðð0; 1ÞÞ of H 1 ðð0; 1Þ is equal to the closed ball
B2 ½0ð¼ fx : x 2 Rn ; jxj 2gÞ:
then we say that H is a smooth bump function corresponding to closed set F and
open set G. By Note 3.52, there exists a smooth bump function corresponding to
closed set B1 ½0 and open set B3 ð0Þ:
For simplicity, let us take 1 for j. Now, since x 2 B12dx ðx1 Þ; h1 ðxÞ ¼ 1; and
1
hence, rðxÞ ¼ 1 ðð1 1Þð1 h2 ðxÞÞ ð1 hn ðxÞÞÞ ¼ 1:
4. Since F is nonempty, there exists a in F. Hence, rðaÞ ¼ 0: Thus, the function r
assumes 0. Similarly, r assumes 1. Since r : R3 ! ½0; 1; and r is smooth, r is
continuous. Since r : R3 ! ½0; 1 is continuous, r assumes 0, and r assumes 1,
r assumes all values of ½0; 1: Hence, r : R3 ! ½0; 1 is onto. h
Note 3.54 The result similar to Theorem 3.53 can be proved as above for Rn in
place of R3 :
Theorem 3.55 Let a be in R3 : Let U be an open neighborhood of a in R3 : Let
f : U ! R be any function. If f is smooth, then there exist an open neighborhood V
of a and a function F : R3 ! R such that
1. V is contained in U,
2. for every x in V, FðxÞ ¼ f ðxÞ;
3. for every x 62 U; FðxÞ ¼ 0;
4. F is smooth.
are disjoint sets. Since Be=2 ðaÞ is an open set, R3 Be=2 ðaÞ is a closed set. Now, by
Theorem 3.53, there exists a function r : R3 ! ½0; 1 such that
1. r is smooth,
2. for every x in R3 Be=2 ðaÞ; rðxÞ ¼ 0;
3. for every x in Be=3 ½a; rðxÞ ¼ 1;
4. r is onto.
Let us take Be=3 ðaÞ for V. Let us define a function F : R3 ! R as follows: For
every x in R3 ;
f ðxÞrðxÞ if x 2 U;
FðxÞ
0 if x 62 U:
Note 3.56 The result similar to Theorem 3.55 can be proved as above for Rn in
place of R3 :
Hence,
2 3
ðD1 F1 Þð0; 0; 0Þ ðD2 F1 Þð0; 0; 0Þ ðD3 F1 Þð0; 0; 0Þ
6 7
detðG0 ð0; 0; 0ÞÞ ¼ det4 ðD1 F2 Þð0; 0; 0Þ ðD2 F2 Þð0; 0; 0Þ ðD3 F2 Þð0; 0; 0Þ 5
0 0 1
ðD1 F1 Þð0; 0; 0Þ ðD2 F1 Þð0; 0; 0Þ
¼ det 6¼ 0:
ðD1 F2 Þð0; 0; 0Þ ðD2 F2 Þð0; 0; 0Þ
Since detðG0 ð0; 0; 0ÞÞ is nonzero, G0 ð0; 0; 0Þ is invertible. Now, we can apply the
inverse function theorem on G : A0 ! R3 : There exists an open neighborhood A1
of ð0; 0; 0Þ such that
1. A1 is contained in A0 ;
2. G is 1–1 on A1 (i.e., if x and y are in A1 ; and GðxÞ ¼ GðyÞ; then x ¼ y),
3. GðA1 Þ is open in R3 ;
4. the 1–1 function G1 from GðA1 Þ onto A1 is a C 2 function.
Now, since ðF1 ð0; 0; 0Þ; F2 ð0; 0; 0Þ; F3 ð0; 0; 0Þ; F4 ð0; 0; 0ÞÞ ¼ Fð0; 0; 0Þ ¼
ð0; 0; 0; 0Þ; each Fi ð0; 0; 0Þ ¼ 0: Next, since ð0; 0; 0Þ is in A0 ; and G : A0 ! R3 ;
Gð0; 0; 0Þ ¼ ðF1 ð0; 0; 0Þ; F2 ð0; 0; 0Þ; 0Þ ¼ ð0; 0; 0Þ: Thus, Gð0; 0; 0Þ ¼ ð0; 0; 0Þ:
Since G is 1–1 on A1 ; Gð0; 0; 0Þ ¼ ð0; 0; 0Þ; and ð0; 0; 0Þ is in A1 ; G1 ð0; 0; 0Þ ¼
ð0; 0; 0Þ: Now, ðF G1 Þð0; 0; 0Þ ¼ FðG1 ð0; 0; 0ÞÞ ¼ Fð0; 0; 0Þ ¼ ð0; 0; 0; 0Þ:
Thus, ðF G1 Þð0; 0; 0Þ ¼ ð0; 0; 0; 0Þ: Since A1 is contained in A0 ; and F : A0 ! B0 ;
ðF G1 ÞðGðA1 ÞÞ ¼ FðA1 Þ B0 : Thus, F G1 : GðA1 Þ ! B0 . For every
ðx1 ; x2 ; x3 Þ in GðA1 Þ, there exists ðy1 ; y2 ; y3 Þ in A1 such that ðx1 ; x2 ; x3 Þ ¼
Gðy1 ; y2 ; y3 Þ ¼ ðF1 ðy1 ; y2 ; y3 Þ; F2 ðy1 ; y2 ; y3 Þ; y3 Þ. Further, since G is 1–1 on A1 ;
so
2 3
100
0
0 6¼ 1 ¼ det4 0 1 0 5 ¼ det G0 G1 ðxÞ G1 ðxÞ
001
0
¼ det G0 G1 ðxÞ det G1 ðxÞ :
It follows that detððG1 Þ0 ðxÞÞ 6¼ 0; and hence, ðG1 Þ0 ðxÞ is invertible for every
x in GðA1 Þ: Since for every x ðx1 ; x2 ; x3 Þ in GðA1 Þ; ðF G1 Þ0 ðxÞ ¼
ðF 0 ðG1 ðxÞÞÞ ððG1 Þ0 ðxÞÞ; and ðG1 Þ0 ðxÞ is invertible,
2 3
1 0 0
60 7
6 1 0 7
rank6 7
4 ðD1 ðF3 G ÞÞðx1 ; x2 ; x3 Þ ðD2 ðF3 G ÞÞðx1 ; x2 ; x3 Þ ðD3 ðF3 G ÞÞðx1 ; x2 ; x3 Þ 5
1 1 1
ðD1 ðF4 G1 ÞÞðx1 ; x2 ; x3 Þ ðD2 ðF4 G1 ÞÞðx1 ; x2 ; x3 Þ ðD3 ðF4 G1 ÞÞðx1 ; x2 ; x3 Þ
0
0
¼ rank F G1 ðxÞ ¼ rank F 0 G1 ðxÞ G1 ðxÞ ¼ rank F 0 G1 ðxÞ ¼ 2:
D3 F3 G1 ðx1 ; x2 ; x3 Þ
2 3
1 0 0
6 7
¼ det4 0 1 0 5 ¼ 0;
1 1 1
ðD1 ðF3 G ÞÞðx1 ; x2 ; x3 Þ ðD2 ðF3 G ÞÞðx1 ; x2 ; x3 Þ ðD3 ðF3 G ÞÞðx1 ; x2 ; x3 Þ
and
D3 F4 G1 ðx1 ; x2 ; x3 Þ
2 3
1 0 0
6 7
¼ det4 0 1 0 5 ¼ 0:
1 1 1
ðD1 ðF4 G ÞÞðx1 ; x2 ; x3 Þ ðD2 ðF4 G ÞÞðx1 ; x2 ; x3 Þ ðD3 ðF4 G ÞÞðx1 ; x2 ; x3 Þ
216 3 Multivariable Differential Calculus
It follows that for every x ðx1 ; x2 ; x3 Þ in GðA1 Þ; F3 G1 and F4 G1 are
functions of x1 ; x2 only.
Since Gð0; 0; 0Þ ¼ ð0; 0; 0Þ; A1 is an open neighborhood of ð0; 0; 0Þ; and GðA1 Þ
is open in R3 ; GðA1 Þ is an open neighborhood of ð0; 0; 0Þ; and hence, GðA1 Þ R is
an open neighborhood of ð0; 0; 0; 0Þ in R4 : Let us define a function T : GðA1 Þ
R ! R4 as follows: For every ðy1 ; y2 ; y3 ; y4 Þ in GðA1 Þ R;
F1 G1 ð0; 0; 0Þ; F2 G1 ð0; 0; 0Þ; F3 G1 ð0; 0; 0Þ; F4 G1 ð0; 0; 0Þ
¼ F1 G1 ð0; 0; 0Þ ; F2 G1 ð0; 0; 0Þ ; F3 G1 ð0; 0; 0Þ ; F4 G1 ð0; 0; 0Þ
¼ F G1 ð0; 0; 0Þ ¼ F G1 ð0; 0; 0Þ ¼ ð0; 0; 0; 0Þ;
so
T 0 ðy1 ; y2 ; y3 ; y4 Þ
2
1 0
60
6 1
¼6
4 D1 ðy3 þ ðF3 G1 Þðy1 ; y2 ; y3 ÞÞ D2 ðy3 þ ðF3 G1 Þðy1 ; y2 ; y3 ÞÞ
D1 ðy4 þ ðF4 G1 Þðy1 ; y2 ; y3 ÞÞ D2 ðy4 þ ðF4 G1 Þðy1 ; y2 ; y3 ÞÞ
3
0 0
7
0 0 7
7
D3 ðy3 þ ðF3 1
G Þðy1 ; y2 ; y3 ÞÞ D4 ðy3 þ ðF3 G Þðy1 ; y2 ; y3 ÞÞ 5
1
2
1 0
60
6 1
¼6
4 0 þ D1 ðF3 G1 Þðy1 ; y2 ; y3 Þ 0 þ D2 ðF3 G1 Þðy1 ; y2 ; y3 Þ
0 þ D1 ðF3 G1 Þðy1 ; y2 ; y3 Þ 0 þ D2 ðF4 G1 Þðy1 ; y2 ; y3 Þ
3
0 0
7
0 0 7
7
1
1 þ D3 ðF3 G Þðy1 ; y2 ; y3 Þ 0 5
0 þ D3 ðF4 G1 Þðy1 ; y2 ; y3 Þ 1 þ D4 ðF4 G1 Þðy1 ; y2 ; y3 Þ
2
1 0
60
6 1
¼6
4 D1 ðF3 G1 Þðy1 ; y2 ; y3 Þ D2 ðF3 G1 Þðy1 ; y2 ; y3 Þ
D1 ðF3 G1 Þðy1 ; y2 ; y3 Þ D2 ðF4 G1 Þðy1 ; y2 ; y3 Þ
3
0 0
7
0 0 7
7
1
1 þ D3 ðF3 G Þðy1 ; y2 ; y3 Þ 0 5
D3 ðF4 G1 Þðy1 ; y2 ; y3 Þ 1 þ D4 ðF4 G1 Þðy1 ; y2 ; y3 Þ
2 3
1 0 0 0
60 7
6 1 0 0 7
¼6 7
4 D1 ðF3 G1 Þðy1 ; y2 ; y3 Þ D2 ðF3 G1 Þðy1 ; y2 ; y3 Þ 1 þ 0 0 5
D1 ðF3 G1 Þðy1 ; y2 ; y3 Þ D2 ðF4 G1 Þðy1 ; y2 ; y3 Þ 0 1þ0
2 3
1 0 0 0
60 0 07
6 1 7
¼6 7
4 D1 ðF3 G1 Þðy1 ; y2 ; y3 Þ D2 ðF3 G1 Þðy1 ; y2 ; y3 Þ 1 05
D1 ðF3 G1 Þðy1 ; y2 ; y3 Þ D2 ðF4 G1 Þðy1 ; y2 ; y3 Þ 0 1
Hence,
and
It follows that
and
4. FðAÞ B;
5. For every x ðx1 ; x2 ; x3 Þ in U, ðH F G1 Þðx1 ; x2 ; x3 Þ ¼ ðx1 ; x2 ; 0; 0Þ:
^
FðxÞ FðxÞ b ¼ ðTb F ÞðxÞ;
(vii) G : A ! U is C 2 ;
(viii) G1 : U ! A is C 2 ;
(ix) H : B ! V is 1–1,
(x) H : B ! V is onto,
(xi) H : B ! V is C 2 ;
(xii) H 1 : V ! B is C2 ;
(xiii) A is contained in A0 ;
(xiv) B is contained in B0 ;
(xv) FðAÞ B;
(xvi) For every x ðx1 ; x2 ; x3 Þ in U, ðH F G1 Þðx1 ; x2 ; x3 Þ ¼ ðx1 ; x2 ; 0; 0Þ:
For (i): Since A ^ is an open neighborhood of ð0; 0; 0Þ; Að¼ AÞ ^ is an open
neighborhood of ð0; 0; 0Þ.
For (ii): Since B^ is an open neighborhood of 0, Bð¼ B ^ þ bÞ is an open neigh-
borhood of b.
For (iii): Since U^ is an open subset of R3 ; Uð¼ UÞ ^ is an open subset of R3 .
^ ^
For (iv): Since V is an open subset of R ; Vð¼ VÞ is an open subset of R4 .
4
^ is contained in A
For (xiii): Since A ^0; A
^ ¼ A; and A^ 0 ¼ A0 ; A is contained in A0.
^ ^ ^
For (xiv): Since B is contained in B0 ; B ¼ B b; and B ^ 0 ¼ B0 b; B − b is
contained in B0 b; and hence, B is contained in B0.
For (xv): Let us take any x in A. We have to prove that F(x) is in B. Here, x is in
A, and A ¼ A, ^ so x is in A:
^ Since x is in A;
^ and Fð ^
^ AÞ ^ ¼ B b; FðxÞ b ¼
B
^
FðxÞ 2 B b; and hence, FðxÞ is in B.
3.7 The Constant Rank Theorem in Rn 223
^
FðxÞ Fðx þ aÞ ¼ ðF Ta ÞðxÞ;
^0, F
Thus, for every x in A ^ 0 ; then x þ a is in A0, and
^ 0 ðxÞ ¼ F 0 ðxÞ: If x is in A
^ 0 0
hence, F ðxÞð¼ F ðx þ aÞÞ has rank 2. Thus, we see that all the conditions of
224 3 Multivariable Differential Calculus
Put A A^ þ a; B B;
^ U U; ^ and V V:
^ Let us define a function G : A ! U
^
as follows: For every x in A, GðxÞ Gðx aÞ ¼ ðG ^ Ta ÞðxÞ: Let us define a
function H : B ! V as follows: For every y in B, HðyÞ HðyÞ:^ Hence, G ¼
^ Ta ; and H ¼ H:
G ^ Further, G^ ¼ G Ta :
It remains to prove the following:
(i) A is an open neighborhood of a,
(ii) B is an open neighborhood of b,
(iii) U is an open subset of R3 ;
(iv) V is an open subset of R4 ;
(v) G : A ! U is 1–1,
(vi) G : A ! U is onto,
(vii) G : A ! U is C 2 ;
(viii) G1 : U ! A is C 2 ;
(ix) H : B ! V is 1–1,
(x) H : B ! V is onto,
(xi) H : B ! V is C 2 ;
(xii) H 1 : V ! B is C2 ;
(xiii) A is contained in A0 ;
(xiv) B is contained in B0 ;
(xv) FðAÞ B;
(xvi) For every x ðx1 ; x2 ; x3 Þ in U, ðH F G1 Þðx1 ; x2 ; x3 Þ ¼ ðx1 ; x2 ; 0; 0Þ:
For (i): Since A^ is an open neighborhood of ð0; 0; 0Þ; Að¼ A^ þ aÞ is an open
neighborhood of a.
^ is an open neighborhood of b, Bð¼ BÞ
For (ii): Since B ^ is an open neighborhood
of b.
For (iii): Since U ^ is an open subset of R3 .
^ is an open subset of R3 ; Uð¼ UÞ
For (iv): Since V^ is an open subset of R ; Vð¼ VÞ
4 ^ is an open subset of R4 .
For (v): Since G ^ :A^!U ^ is 1–1, and Ta : A ^þa!A ^ is 1–1, G ^ Ta :
^
Aþa!U ^ is 1–1, and hence, G^ Ta : A ! U is 1–1.
3.7 The Constant Rank Theorem in Rn 225
For (vi): Since G ^ :A^!U ^ is onto, and Ta : A ^þa!A ^ is onto, G ^ Ta :
^ ^ ^
A þ a ! U is onto, and hence, G Ta : A ! U is onto.
For (vii): Since G ^ :A^!U ^ is C 2 ; and Ta : A ^þa!A ^ is C2 ; G ^ Ta :
^ ^ ^
A þ a ! U isC , and hence, G Ta : A ! U is C .
2 2
3. H is a Ck diffeomorphism,
4. FðAÞ B;
5. For every x ðx1 ; x2 ; . . .; xn Þ in U, ðH F G1 Þðx1 ; x2 ; . . .; xn Þ ¼ ðx1 ; x2 ; . . .
xk ; 0; . . .; 0Þ:
|fflfflffl{zfflfflffl}
mk
Proof Let us take any x in U and any t satisfying 0 t 1: Now, since U is star-
shaped with respect to point p, ð1 tÞp þ tx is in S, and hence, f ðð1 tÞp þ txÞ is a
real number. Here,
d d
f ðð1 tÞp þ txÞ ¼ f p1 þ t x1 p1 ; p2 þ t x2 p2
dt dt
of d 1
¼ ð ð1 t Þp þ tx Þ p þ t x 1
p 1
ox1 dt
of d 2
þ ð ð 1 t Þp þ tx Þ p þ t x 2
p 2
ox2 dt
of
¼ ðð1 tÞp þ txÞ x1 p1
ox1
of
þ ðð1 tÞp þ txÞ x2 p2 ;
ox2
3.8 Taylor’s Theorem 227
hence,
Z1of 2 Z
1 of
x 1 p1 ð ð 1 t Þp þ tx Þ dt þ x 2
p ðð 1 t Þp þ tx Þ dt
0
ox1 0
ox2
Z1 of of
¼ ðð1 tÞp þ txÞ x p þ1 1
ðð1 tÞp þ txÞ x p 2 2
dt
0
ox1 ox2
t¼1
¼ f ðð1 tÞp þ txÞjt¼0 ¼ f ðxÞ f ðpÞ:
Now,
where
Z1 of Z1 of
g1 ðxÞ ðð1 tÞp þ txÞ dt; and g2 ðxÞ ðð1 tÞp þ txÞ dt:
0
ox1 0
ox2
of
It remains to prove that g1 ; g2 are C1 on U, and gi ðpÞ ¼ ox i ðpÞði ¼ 1; 2Þ:
Here,
og1 o Z 1 of
¼ ðð1 tÞp þ txÞ dt
ox1 ox1 0 ox1
Z o of
¼ 0 1
ðð1 tÞp þ txÞ dt
ox1 ox1
Z o of 1
¼ 0 1
p þ t x p ;p þ t x p
1 1 2 2 2
dt
ox1 ox1
Z o of 1 oð p1 þ t ð x 1 p1 Þ Þ
¼ 10 p þ t x 1
p 1
; p 2
þ t x 2
p 2
ox1 ox1 ox1
o of 1 oð p þ t ð x 1 p1 Þ Þ
1
þ p þ t x 1
p 1
; p 2
þ t x 2
p 2
dt
ox2 ox1 ox2
Z o of 1
¼ 10 p þ t x 1 p1 ; p2 þ t x 2 p2 t
ox ox1
1
o of 1
þ p þ t x 1 p1 ; p2 þ t x 2 p2 0 dt
ox ox1
2
!!
Z o2 f
¼ 10 p1 þ t x 1 p1 ; p2 þ t x 2 p2 t dt
oðx1 Þ2
!
Z o2 f
¼ 0t1
ðð1 tÞp þ txÞ dt:
oð x 1 Þ 2
228 3 Multivariable Differential Calculus
Thus,
!
og1 Z1 o2 f
¼ t ðð1 tÞp þ txÞ dt:
ox1 0 oðx1 Þ2
og1 o2 g
Similarly, ; ; etc. can be obtained. This proves that g1 is C 1 on U.
ox2 oðx1 Þ2
Similarly, g2 is C1 on U. Now,
Z1 of Z1 of of Z1
g1 ðpÞ ¼ ð ð 1 t Þp þ tpÞ dt ¼ ðpÞ dt ¼ ðpÞ dt
ox 1 ox 1 ox 1
0 0 0
of of
¼ ðpÞ 1 ¼ 1 ðpÞ:
ox 1 ox
of of
Thus, g1 ðpÞ ¼ ox 1 ðpÞ: Similarly, g2 ðpÞ ¼ ox2 ðpÞ: h
Note 3.63 The result similar to above is also valid for R ; R ; . . .:
3 4
of
f ðx; yÞ ¼ f ð0; 0Þ þ x ð0; 0Þ þ xg11 ðx; yÞ þ yg12 ðx; yÞ
ox
of
þy ð0; 0Þ þ xg21 ðx; yÞ þ yg22 ðx; yÞ
oy
of of
¼ f ð0; 0Þ þ ð0; 0Þ x þ ð0; 0Þ y
ox oy
þ ðg11 ðx; yÞÞx2 þ ðg12 ðx; yÞ þ g21 ðx; yÞÞxy þ ðg22 ðx; yÞÞy2 :
Since g12 ðx; yÞ; g21 ðx; yÞ are C1 functions, g12 ðx; yÞ þ g21 ðx; yÞ is also a C 1
function. h
Chapter 4
Topological Properties of Smooth
Manifolds
In real analysis, its deep theorems (like Lagrange’s mean value theorem, funda-
mental theorem of integral calculus, etc.) are quite difficult to prove without the help
of topological theorems (like Weierstrass theorem, Heine–Borel theorem, etc.) for
real line. Similar is the situation in complex analysis. Here, in the study of smooth
manifolds, in order to prove its deep rooted theorems, we will need its topological
properties vigorously. So, for smoothening later work, it is better to collect all the
results, to be needed in future, relating to topological properties of smooth mani-
folds at one place. This chapter aims at this end and is largely self-contained. The
pace of this chapter is moderate so that the topic is well assimilated.
2 3
ðD1 f 1 ÞðaÞ ðD2 f 1 ÞðaÞ ðD3 f 1 ÞðaÞ
4
det ðD1 f 2 ÞðaÞ ðD2 f 2 ÞðaÞ ðD3 f 2 ÞðaÞ 5 6¼ 0;
ðD1 f 3 ÞðaÞ ðD2 f 3 ÞðaÞ ðD3 f 3 ÞðaÞ 33
exists, and hence, f 0 ðaÞ is invertible. Hence, by the Theorem 3.36, there exists a
connected open neighborhood U of a such that
1. U is contained in E,
2. f is 1–1 on U (that is, if x, y are in U, and f ðxÞ ¼ f ðyÞ; then x = y),
3. f ðUÞ is connected and open in R3 ;
4. the 1–1 function f 1 from f ðUÞ onto U is a C 1 function,
5. if f is a C2 function, then f 1 from f ðUÞ onto U is a C 2 function, etc.
By the conclusions 1, 2, 3, and 4, we find the f 1 is a C 1 function on f ðUÞ:
Now, it remains to prove that f 1 is smooth on f ðUÞ; that is, f 1 is a Cn function
for every n ¼ 2; 3; 4; . . .: Since f is a smooth function, f is a C 2 function, and
hence by conclusion 5, f 1 is a C 2 function. Again, since f is a smooth function,
f is a C 3 function, and hence by conclusion 5, f 1 is a C3 function, etc. Hence,
f 1 is smooth on f ðUÞ: h
Note 4.2 The result similar to Theorem 4.1 can be proved as above for Rn in place
of R3 :
Theorem 4.3 Let E be an open subset of R3 : Let f : E ! R3 : Let a be in E. If
(i) f is a smooth function,
(ii) the tangent map f of f at a is an isomorphism,
then there exists an open neighborhood U of a, and an open neighborhood V of
f ðaÞ such that the restriction of f to U is a diffeomorphism from U onto V.
Proof By the Note 2.77, the tangent map f of f at a is an isomorphism if and only if
2 3
ðD1 f 1 ÞðaÞ ðD2 f 1 ÞðaÞ ðD3 f 1 ÞðaÞ
det4 ðD1 f 2 ÞðaÞ ðD2 f 2 ÞðaÞ ðD3 f 2 ÞðaÞ 5 6¼ 0:
ðD1 f 3 ÞðaÞ ðD2 f 3 ÞðaÞ ðD3 f 3 ÞðaÞ
4.1 Constant Rank Theorem 233
This shows that we can apply Theorem 4.1. By Theorem 4.1, there exists an
open neighborhood U of a such that
1. U is contained in E,
2. f ðUÞ is open in R3 ;
3. f has a smooth inverse on f ðUÞ:
This shows that f ðUÞ is an open neighborhood of f ðaÞ such that the restriction of
f to U is a diffeomorphism from U onto f ðUÞ: h
Note 4.4 The result similar to Theorem 4.3 can be proved as above for Rn in place
of R3 :
Theorem 4.5 Let M and N be 3-dimensional smooth manifolds. Let f : M ! N:
Let p be in M. If
(i) f is a smooth map,
(ii) the tangent map f : Tp ðMÞ ! Tf ðpÞ ðNÞ is an isomorphism,
then there exists an open neighborhood U of p in M such that
1. V f ðUÞ is an open neighborhood of f ðpÞ in N,
2. the restriction of f to U is a diffeomorphism from U onto V.
2 1 1
1 1
6 D 1 wV f ð u U Þ ð uU ð pÞ Þ D 2 w V f ð u U Þ ðuU ð pÞÞ
6
6 2 2
6
det6 D1 wV f ðuU Þ1 ðuU ð pÞÞ D2 wV f ðuU Þ1 ðuU ð pÞÞ
6
6
4 3 3
D1 wV f ðuU Þ1 ðuU ð pÞÞ D2 wV f ðuU Þ1 ðuU ð pÞÞ
1 3
D3 wV f ðuU Þ1 ðu U ð p ÞÞ 7
7
2 7
1 7
D3 wV f ðuU Þ ðuU ð pÞÞ 7 6¼ 0;
7
3 7
1
5
D3 wV f ðuU Þ ðu U ð p ÞÞ
and hence, the determinant of the Jacobian matrix of wV ðf ðuU Þ1 Þ at uU ðpÞ is
nonzero. Hence, by Theorem 4.1, there exists an open neighborhood W of uU ðpÞ
such that
1. W is contained in uU ðU \ U1 Þ;
2. ðwV ðf ðuU Þ1 ÞÞðWÞ is open in R3 ;
3. wV ðf ðuU Þ1 Þ has a smooth inverse on ðwV ðf ðuU Þ1 ÞÞðWÞ:
Since W is contained in uU ðU \ U1 Þ; and uU is 1–1, ðuU Þ1 ðWÞ is contained in
U \ U1 : Since uU ðpÞ is in W, p is in ðuU Þ1 ðWÞ: Since uU is a homeomorphism,
and W is an open subset of open set uU ðUÞ; ðuU Þ1 ðWÞ is open. Thus, ðuU Þ1 ðWÞ
is an open neighborhood of p in M.
For 1: We have to prove that f ððuU Þ1 ðWÞÞ is open in N. Since ðwV ðf
ðuU Þ1 ÞÞðWÞ is open, and wV is a homeomorphism, ðwV Þ1 ððwV ðf ðuU Þ1 ÞÞ
ðWÞÞð¼ f ððuU Þ1 ðWÞÞÞ is open.
For 2: Since wV ðf ðuU Þ1 Þ is 1–1, ðwV Þ1 is 1–1 and uU is 1–1, their com-
posite ðwV Þ1 ðwV ðf ðuU Þ1 ÞÞ uU ð¼ f Þ is 1–1. Since f is 1–1, the function
f 1 exists. It remains to prove that f 1 is smooth. Since ðwV ðf ðuU Þ1 ÞÞ1
ð¼ uU f 1 ðwV Þ1 Þ is smooth, ðuU Þ1 is smooth, and wV is smooth, their
composite function ðuU Þ1 ðuU f 1 ðwV Þ1 Þ wV ð¼ f 1 Þ is smooth. h
Note 4.6 The result similar to Theorem 4.5 can be proved as above for Rn in place
of R3 :
Theorem 4.7 Let M be an m-dimensional differentiable manifold, N be an n-
dimensional smooth manifold, and F : N ! M be a smooth function. Let p be in N.
Let ðU; uU Þ be an admissible coordinate chart in N satisfying p 2 U; ðV; wV Þ be an
admissible coordinate chart in M satisfying FðpÞ 2 V; and FðUÞ V: Let ð U b;u Þ
b U
4.1 Constant Rank Theorem 235
b ; ðV
be an admissible coordinate chart in N satisfying p 2 U b ; w Þ be an admissible
b
V
coordinate chart in M satisfying FðpÞ 2 V b ; and Fð U
bÞ V b : Then, the rank of
ðwV F ðuU Þ1 Þ0 ðuU ðpÞÞ and the rank of ðwb F ðu b Þ1 Þ0 ðu b ðpÞÞ are equal.
V U U
Proof Here,
1 0 1 0
wb F u b u b ð pÞ ¼ wb F ðuU Þ1 uU u b u b ð pÞ
V U U V U U
1 0
¼ wb ðwV Þ1 wV F ðuU Þ1 uU u b u b ð pÞ
V U U
1 0
¼ wb ðwV Þ1 wV F ðuU Þ1 uU u b u b ð pÞ
V U U
0 1
¼ wb ðwV Þ1 wV F ðuU Þ1 uU u b u b ð pÞ
V U U
0 1 1 0
wV F ðuU Þ1 uU u b u b ð pÞ uU u b u b ð pÞ
U U U U
0 1
¼ wb ðwV Þ1 wV F ðuU Þ1 uU u b u b ð pÞ
V U U
0 1 0
wV F ðuU Þ1 ðuU ð pÞÞ uU u b u b ð pÞ
U U
0 0
1 1
¼ w b ðwV Þ ððwV F Þð pÞÞ wV F ðuU Þ ðuU ð pÞÞ
V
1 0
uU u b u b ð pÞ :
U U
0 0
wb ðwV Þ1 ððwV F Þð pÞÞ wV F ðuU Þ1 ðuU ð pÞÞ
V
and the rank of ððwV F ðuU Þ1 Þ0 ðuU ðpÞÞÞ are equal. Similarly the rank of
0 0
wb ðwV Þ1 ððwV F Þð pÞÞ wV F ðuU Þ1 ðuU ð pÞÞ
V
1 0
uU u b u b ð pÞ
U U
236 4 Topological Properties of Smooth Manifolds
Hence, rank of
0 0
wb ðwV Þ1 ððwV F Þð pÞÞ wV F ðuU Þ1 ðuU ð pÞÞ
V
Now, we want to apply Theorem 3.59. Here, uU ðUÞ acts for A0 ; uU ðpÞ acts for
a, wV ðVÞ acts for B0, wV ðFðpÞÞ acts for b, wV ðF ðuU Þ1 Þ acts for F. Now, we
must verify the following conditions of the theorem:
1. uU ðUÞ is an open neighborhood of uU ðpÞ in R3 ;
2. wV ðVÞ is an open neighborhood of wV ðFðpÞÞ in R4 ;
3. wV ðF ðuU Þ1 Þ : uU ðUÞ ! wV ðVÞ is a C 1 function,
4. for every x in uU ðUÞ; the linear transformation ðwV ðF ðuU Þ1 ÞÞ0 ðxÞ has
rank 2,
5. ðwV ðF ðuU Þ1 ÞÞðuU ðpÞÞ ¼ wV ðFðpÞÞ:
For 1: Since ðU; uU Þ is an admissible coordinate chart in M satisfying p 2 U;
uU ðUÞ is an open neighborhood of uU ðpÞ in R3 :
For 2: Since ðV; wV Þ is an admissible coordinate chart in N satisfying FðpÞ 2 V;
wV ðVÞ is an open neighborhood of wV ðFðpÞÞ in R4 :
For 3: Since F : M ! N is a smooth function, ðU; uU Þ is an admissible coor-
dinate chart in M satisfying p 2 U; and ðV; wV Þ an admissible coordinate chart
in N satisfying FðpÞ 2 V; wV ðF ðuU Þ1 Þ : uU ðUÞ ! wV ðVÞ is a C 1
function.
For 4: Let us take any uU ðqÞ in uU ðUÞ where q is in U. We have to prove that the
rank of ðwV F ðuU Þ1 Þ0 ðuU ðqÞÞ is 2. Since 2 is the rank of F : M ! N; and p is
in N, the rank of F at p is 2, and hence, the rank of ðwV F ðuU Þ1 Þ0 ðuU ðpÞÞ is 2.
For 5: Here, LHS ¼ ðwV ðF ðuU Þ1 ÞÞðuU ðpÞÞ ¼ ðwV FÞðpÞ ¼ wV ðFðpÞÞ ¼
RHS: Hence, by Theorem 3.59, there exists an open neighborhood A of uU ðpÞ in
R3 ; an open neighborhood B of ðwV FÞðpÞ in R4 ; an open subset U b of R3 ; an
open subset V b of R ; a function G
4 b :A!U b ; and a function H
b :B!V b such that
¼ ðx1 ; x2 ; 0; 0Þ ¼ RHS:
4.1 Constant Rank Theorem 239
For (vii): Since wV ððF ðuU Þ1 ÞðAÞÞ ¼ ðwV ðF ðuU Þ1 ÞÞðAÞ B; and wV
is 1–1, FððuU Þ1 ðAÞÞ ðwV Þ1 ðBÞ: h
uU ðUÞð Rm Þ: Then ððuU Þ1 ðGÞ; wðuU Þ1 ðGÞ Þ is an admissible coordinate chart of
M, where wðuU Þ1 ðGÞ denotes the restriction of uU on ðuU Þ1 ðGÞ:
Further, we may assume uU ðUÞ ¼ ðe; eÞ ðe; eÞ ðe; eÞ, and wV ðVÞ ¼
ðe; eÞ ðe; eÞ ðe; eÞ ðe; eÞ for some e [ 0: In short, uU ðUÞ ¼ Ce3 ð0Þ;
and wV ðVÞ ¼ Ce4 ð0Þ for some e [ 0:
Proof By Theorem 4.8, there exist admissible coordinate chart ðU ; uU Þ in
M satisfying p 2 U ; and admissible coordinate chart ðV ; wV Þ in N satisfying
FðpÞ 2 V such that for every ðx1 ; x2 ; x3 Þ in uU ðU Þ;
wV F ðuU Þ1 ðx1 ; x2 ; x3 Þ ¼ ðx1 ; x2 ; 0; 0Þ:
uU : U ! uU ðU Þ is a homeomorphism. Since uU : U ! uU ðU Þ is a
homeomorphism, and the translation
so vW ðuU Þ1 ¼ ðvW ðuU Þ1 Þ TuU ðpÞ : Since ðW; vW Þ; and ðU ; uU Þ are
admissible coordinate charts in M, vW ðuU Þ1 is C 1 . Since vW ðuU Þ1 is C1 ;
and translation TuU ðpÞ is C1 , their composite ðvW ðuU Þ1 Þ TuU ðpÞ ð¼ vW
ðuU Þ1 Þ is C1 , and hence, vW ðuU Þ1 is C1 . Similarly, uU ðvW Þ1 is C1 :
For (ii): Its proof is similar to that of (i).
For (iii): uU ðpÞ ¼ ðuU uU ðpÞÞðpÞ ¼ uU ðpÞ uU ðpÞ ¼ ð0; 0; 0Þ: Thus
uU ðpÞ ¼ ð0; 0; 0Þ:
For (iv): Let us take any ðx1 ; x2 ; x3 Þ in uU ðUÞ: We have to prove that
ðwV F ðuU Þ1 Þðx1 ; x2 ; x3 Þ ¼ ðx1 ; x2 ; 0; 0Þ: Here, ðuU Þ1 ¼ ðuU uU
ðpÞÞ1 ¼ ðTuU ðpÞ uU Þ1 ¼ ðuU Þ1 TuU ðpÞ : Since uU ¼ uU uU ðpÞ;
uU ðUÞ ¼ uU ðUÞ uU ðpÞ ¼ uU ðU Þ uU ðpÞ: Now, since ðx1 ; x2 ; x3 Þ is in
uU ðUÞ; ðx1 þ ðuU ðpÞÞ1 ; x2 þ ðuU ðpÞÞ2 ; x3 þ ðuU ðpÞÞ3 Þ is in uU ðU Þ; and
hence,
wV F ðuU Þ1 x1 þ ðuU ð pÞÞ1 ; x2 þ ðuU ð pÞÞ2 ; x3 þ ðuU ð pÞÞ3
¼ x1 þ ðuU ð pÞÞ1 ; x2 þ ðuU ð pÞÞ2 ; 0; 0 :
4.1 Constant Rank Theorem 243
Now,
LHS ¼ wV F ðuU Þ1 ðx1 ; x2 ; x3 Þ ¼ wV F ðuU Þ1 uU ð pÞ ðx 1 ; x 2 ; x 3 Þ
¼ wV ðuU ð pÞÞ1 ; ðuU ð pÞÞ2 ; 0; 0 F ðuU Þ1 TuU ð pÞ ðx1 ; x2 ; x3 Þ
¼ TððuU ð pÞÞ1 ;ðuU ð pÞÞ2 ;0; 0Þ wV F ðuU Þ1 TuU ð pÞ ðx1 ; x2 ; x3 Þ
¼ TððuU ð pÞÞ1 ;ðuU ð pÞÞ2 ;0; 0Þ wV F ðuU Þ1 TuU ð pÞ ðx1 ; x2 ; x3 Þ
¼ TððuU ð pÞÞ1 ;ðuU ð pÞÞ2 ;0; 0Þ wV F ðuU Þ1
x1 þ ðuU ð pÞÞ1 ; x2 þ ðuU ð pÞÞ2 ; x3 þ ðuU ð pÞÞ3
¼ TððuU ð pÞÞ1 ;ðuU ð pÞÞ2 ;0; 0Þ x1 þ ðuU ð pÞÞ1 ; x2 þ ðuU ð pÞÞ2 ; 0; 0
¼ ðx1 ; x2 ; 0; 0Þ ¼ RHS:
For (v): Since p 2 U; uU ðpÞ 2 uU ðUÞ; and ððuU ðpÞÞ1 ; ðuU ðpÞÞ2 ; ðuU ðpÞÞ2 Þ ¼
uU ðpÞ ¼ ð0; 0; 0Þ so
ð0; 0; 0; 0Þ ¼ ðuU ð pÞÞ1 ; ðuU ð pÞÞ2 ; 0; 0 ¼ wV F ðuU Þ1 ðuU ð pÞÞ
¼ ðwV F Þð pÞ ¼ wV ðF ð pÞÞ:
Further, we may assume uU ðUÞ ¼ Cem ð0Þ; and wV ðVÞ ¼ Cen ð0Þ for some e [ 0:
Proof Its proof is quite similar to the proof of Theorem 4.12. h
Lemma 4.14 Let M be an m-dimensional smooth manifold, N be an n-dimensional
smooth manifold, and F : N ! M be a 1–1 smooth function. Let O1 be the topology
of N. Let A1 be the differential structure on N. Put O fFðGÞ : G 2 O1 g and
A fðFðUÞ; uU F 1 Þ : ðU; uU Þ 2 A1 g: Then,
I. O is a topology over FðNÞ;
II. A determines a unique differential structure on FðNÞ, and
III. F is a diffeomorphism from N onto FðNÞ:
Proof of I
(i) Since / 2 O1 ; Fð/Þð¼ /Þ is in O; and hence, / 2 O: Next, since
N 2 O1 ; so FðNÞis in O:
(ii) Let FðGÞ 2 O; where G 2 O1 : Let FðHÞ 2 O; where H 2 O1 : We have to
prove that FðGÞ \ FðHÞ 2 O: Since G 2 O1 ; H 2 O1 ; and O1 is a topology,
G \ H 2 O1 ; and hence, FðG \ HÞ 2 O: Since F : N ! M is 1–1,
FðG \ HÞ ¼ FðGÞ \ FðHÞ: Since FðG \ HÞ ¼ FðGÞ \ FðHÞ; and
FðG \ HÞ 2 O; FðGÞ \ FðHÞ 2 O:
(iii) Let FðGi Þ 2 O; where Gi 2 O1 for every index i in index set I. We have to
prove that [i2I FðGi Þ 2 O: Since for every index i, Gi 2 O1 ; and O1 is a
topology, [i2I Gi 2 O1 ; and hence, Fð[i2I Gi Þð¼ [i2I FðGi ÞÞ 2 O: Thus,
[i2I FðGi Þ 2 O: Hence, O is a topology over FðNÞ: h
mðg; hÞ gh:
iðgÞ g1 :
We shall try to show that F is a smooth mapping. For this purpose, let us fix any
h0 in M. We have to prove that F is smooth at h0 in M.
For this purpose, let us take any admissible coordinate chart ðV; wV Þ of M sat-
isfying h0 2 V; and any admissible coordinate chart ðU; uU Þ of G satisfying a 2 U:
Here, ðU; uU Þ is an admissible coordinate chart of G satisfying a 2 U; ðV; wV Þ is an
admissible coordinate chart of M satisfying h0 2 V; and Fðh0 Þ ¼ ða; h0 Þ 2 U V;
ðU V; ðuU wV ÞÞ is an admissible coordinate chart of G M satisfying Fðh0 Þ 2
U V: We have to prove that each of the 2 component functions of the mapping
4.2 Lie Groups 247
ðuU wV Þ F ðwV Þ1 : wV V \ F 1 ðU V Þ ! ðuU wV ÞðU V Þ
ðr; sÞ 7! p1 ðððuU wV Þ ðF ðwV Þ1 ÞÞðr; sÞÞ ¼ p1 ðuU ðaÞ; wV ðyÞÞ ¼ uU ðaÞ; which is
a constant function. Hence, ðr; sÞ 7! p1 ðððuU wV Þ ðF ðwV Þ1 ÞÞðr; sÞÞ is a
smooth function. Next, since ðr; sÞ 7! p2 ðððuU wV Þ ðF ðwV Þ1 ÞÞðr; sÞÞ ¼
p2 ðuU ðaÞ; wV ðyÞÞ ¼ wV ðyÞ ¼ ðr; sÞ is the identity function, ðr; sÞ 7! p2 ðððuU
wV Þ ðF ðwV Þ1 ÞÞðr; sÞÞ is a smooth function. Thus, we have shown that F is a
smooth mapping, that is, h 7! ða; hÞ is a smooth mapping.
Similarly, h 7! ðh; aÞ is a smooth mapping.
Note 4.16 Let G; H; K be smooth manifolds. Let f : H ! K be a smooth mapping.
Let F : G H ! G K be the mapping defined as follows: for every ðx; yÞ in
G H;
We shall try to show that F is a smooth mapping. For simplicity, let each of G,H,
K have dimension 2. Let us fix any ðx0 ; y0 Þ in G H: We have to prove that F is
smooth at ðx0 ; y0 Þ:
For this purpose, let us take any admissible coordinate chart ðU V; ðuU wV ÞÞ
of the product manifold G H; where ðU; uU Þ is an admissible coordinate chart of
G satisfying x0 2 U; and ðV; wV Þ is an admissible coordinate chart of H satisfying
y0 2 V: Next, let us take any admissible coordinate chart ðU V; ~ ðuU wV~ ÞÞ of
~
G K; where ðV; wV~ Þ is an admissible coordinate chart of K satisfying f ðy0 Þ 2 V: ~
~ ~ ~
Here, since x0 2 U; and f ðy0 Þ 2 V; Fðx0 ; y0 Þ ¼ ðx0 ; f ðy0 ÞÞ 2 U V: Thus, ðU V;
ðuU wV~ ÞÞ is an admissible coordinate chart of G K satisfying Fðx0 ; y0 Þ 2 U V:~
Now, we have to prove that each of the 4 component functions of the mapping
~
uU wV~ F ðuU wV Þ1 : ðuU wV Þ ðU V Þ \ F 1 U V
! uU wV~ U V ~
248 4 Topological Properties of Smooth Manifolds
so
ðr; s; t; uÞ 7! p1 uU wV~ F ðuU wV Þ1 ðr; s; t; uÞ
¼ p1 r; s; wV~ f ðwV Þ1 ðt; uÞ ¼ r;
is C 1 at the point wV ðy0 Þ in R2 ; that is, ðt; uÞ 7! p1 ðwV~ ðf ðwV Þ1 Þðt; uÞÞ; and
ðt; uÞ 7! p2 ðwV~ ðf ðwV Þ1 Þðr; sÞÞ are C1 at the point wV ðy0 Þ in R2 : Since
ðr; s; t; uÞ 7! ðt; uÞ is smooth, and ðt; uÞ 7! p1 ðwV~ ðf ðwV Þ1 Þðt; uÞÞ is smooth,
their composite ðr; s; t; uÞ 7! p1 ðwV~ ðf ðwV Þ1 Þðt; uÞÞ is smooth, and hence,
ðr; s; t; uÞ 7! p3 uU wV~ F ðuU wV Þ1 ðr; s; t; uÞ
mðg; hÞ gh:
For (i): Since h 7! ðe; hÞ is a smooth mapping, and ðg; hÞ 7! gh1 is a smooth
mapping, their composite h 7! eh1 ¼ h1 ¼ iðhÞ is a smooth mapping, and
hence, i is a smooth mapping.
For (ii): Since h 7! h1 is a smooth mapping, ðg; hÞ 7! ðg; h1 Þ is a smooth
mapping. Since ðg; hÞ 7! ðg; h1 Þ is a smooth mapping, and ðg; hÞ 7! gh1 is a
smooth mapping, their composite ðg; hÞ 7! gðh1 Þ1 ¼ gh ¼ mðghÞ is a smooth
mapping, and hence, m is a smooth mapping. h
Example 4.21 Let Mð2 3; RÞ be the collection of all 2 × 3 matrices with real
entries, that is,
abc
Mð2 3; RÞ : a; b; c; d; e; f 2 R :
def
a b c a b1 c1 a þ a1 b þ b1 c þ c1
þ 1 :
d e f d1 e1 f1 d þ d1 e þ e1 f þ f1
a b c ta tb tc
t :
d e f td te tf
1 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0
; ; ; ; ; :
0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 1
Clearly, the real linear space R23 is isomorphic to the real linear space
Mð2 3; RÞ: Since R23 is isomorphic to Mð2 3; RÞ; we will not distinguish
between
252 4 Topological Properties of Smooth Manifolds
a b c
and ða; b; c; d; e; f Þ:
d e f
x1 x2 x1 x2
GLð2; RÞ : 2 Mð2; RÞ andx1 x4 x2 x3 6¼ 0 :
x3 x4 x3 x4
x x
It is clear that ½ 1 2 7! ðx1 ; x2 ; x3 ; x4 Þ is in 1–1 correspondence from
x3 x4
GLð2; RÞ onto fðx1 ; x2 ; x3 ; x4 Þ : ðx1 ; x2 ; x3 ; x4 Þ 2 R4 and x1 x4 x2 x3 6¼ 0g: Let us
define a function f : R4 ! R as follows: for every ðx1 ; x2 ; x3 ; x4 Þ in R4
f ð x1 ; x2 ; x3 ; x4 Þ x1 x4 x2 x3 :
we have
a b e f
IdGLð2;RÞ IdGLð2;RÞ ;
c d g h
a b e f
¼ IdGLð2;RÞ ; IdGLð2;RÞ
c d g h
a b e f a b e f
¼ ; ¼ IdGLð2;RÞGLð2;RÞ ; ;
c d g h c d g h
a b e f ae þ bg af þ bh
:
c d g h ce þ dg cf þ dh
we have
a b e f a b e f ae þ bg af þ bh
m ; ¼ :
c d g h c d g h ce þ dg cf þ dh
a b
Let us define i : GLð2; RÞ ! GLð2; RÞ as follows: for every in
c d
GLð2; RÞ;
254 4 Topological Properties of Smooth Manifolds
2 3
d b
a b 1 d b 6 ad bc ad bc 7
i ¼ 4 c 5:
c d ad bc c a a
ad bc ad bc
We have to show:
(i) m is a smooth map from (4 + 4)-dimensional product manifold GLð2; RÞ
GLð2; RÞ to 4-dimensional smooth manifold GLð2; RÞ,
(ii) i is a smooth map from 4-dimensional smooth manifold GLð2; RÞ to 4-
dimensional smooth manifold GLð2; RÞ.
a b e f
; in GLð2; RÞ GLð2; RÞ:
c d g h
For this purpose, let us take the admissible coordinate chart ðGLð2; RÞ
GLð2; RÞ; IdGLð2;RÞGLð2;RÞ Þ of the product manifold GLð2; RÞ GLð2; RÞ: Here,
a b e f
; 2 GLð2; RÞ GLð2; RÞ:
c d g h
Next, let us take the admissible coordinate chart ðGLð2; RÞ; IdGLð2;RÞ Þ of
GLð2; RÞ: Clearly,
a b e f ae þ bg af þ bh
m ; ¼ 2 GLð2; RÞ:
c d g h ce þ dg cf þ dh
Now, we have to prove that each of the 4 component functions of the mapping
1
IdGLð2;RÞ m IdGLð2;RÞGLð2;RÞ : IdGLð2;RÞGLð2;RÞ
ðGLð2; RÞ GLð2; RÞÞ \ m1 ðGLð2; RÞÞ ! IdGLð2;RÞ ðGLð2; RÞÞ
is C 1 at the point
a b e f a b e f
IdGLð2;RÞGLð2;RÞ ; ¼ ; in R8 :
c d g h c d g h
4.2 Lie Groups 255
is C 1 at the point
a b e f
; ;
c d g h
that is,
x1 x2 y1 y2 x1 x2 y1 y2
7! x1 y1 þ x2 y3 ; 7! x1 y2 þ x2 y4 ;
x3 x4 y3 y4 x3 x4 y3 y4
x1 x2 y1 y2 x1 x2 y1 y2
7! x3 y1 þ x4 y3 ; 7! x3 y2 þ x4 y4
x3 x4 y3 y4 x3 x4 y3 y4
a b
in GLð2; RÞ:
c d
a b
:
c d
For this purpose, let us take the admissible coordinate chart ðGLð2; RÞ; IdGLð2;RÞ Þ
of the smooth manifold GLð2; RÞ: Here,
a b
2 GLð2; RÞ:
c d
256 4 Topological Properties of Smooth Manifolds
Next, let us take the admissible coordinate chart ðGLð2; RÞ; IdGLð2;RÞ Þ of the
smooth manifold GLð2; RÞ: Clearly,
2 3
d b
a b 1 d b 6 bc ad bc 7
i ¼ ¼ 4 adc 5 2 GLð2; RÞ:
c d ad bc c a a
ad bc ad bc
Now, we have to prove that each of the 4 component functions of the mapping
1
IdGLð2;RÞ i IdGLð2;RÞ : IdGLð2;RÞ ðGLð2; RÞÞ \ i1 ðGLð2; RÞÞ
! IdGLð2;RÞ ðGLð2; RÞÞ
is C 1 at the point
a b a b
IdGLð2;RÞ ¼ in R4 :
c d c d
i : GLð2; RÞ ! GLð2; RÞ
is C 1 at the point
a b
;
c d
that is,
x1 x2 x4 x1 x2 x2
7! ; 7! ;
x3 x4 x1 x4 x2 x3 x3 x4 x1 x4 x2 x3
x1 x2 x3 x1 x2 x1
7! ; 7!
x3 x4 x1 x4 x2 x3 x3 x4 x1 x4 x2 x3
are smooth. Clearly, all these four functions are smooth. Thus, we have shown that
i is a smooth map. Hence, GLð2; RÞ is a 4-dimensional Lie group. Similarly,
4.2 Lie Groups 257
(1) Closure law: Let us take any T1 ; T2 in GLðVÞ: We have to show that T1 T2 is
in GLðVÞ; that is,
(a) T1 T2 is 1–1 from V to V,
(b) T1 T2 maps V onto V,
(c) for every x; y in V, and for every s; t in R; ðT1 T2 Þðsx þ tyÞ ¼
sððT1 T2 ÞðxÞÞ þ tððT1 T2 ÞðyÞÞ:
LHS ¼ T 1 ðsx þ tyÞ ¼ T 1 ðsðT ðx1 ÞÞ þ tðT ðy1 ÞÞÞ ¼ T 1 ðT ðsx1 þ ty1 ÞÞ ¼ sx1 þ ty1
¼ s T 1 ð xÞ þ t T 1 ð yÞ ¼ RHS:
258 4 Topological Properties of Smooth Manifolds
ða11 t1 þ a12 t2 Þe1 þ ða21 t1 þ a22 t2 Þe2 ¼ t1 ða11 e1 þ a21 e2 Þ þ t2 ða12 e1 þ a22 e2 Þ
¼ t1 ðT ðe1 ÞÞ þ t2 ðT ðe2 ÞÞ ¼ T ðt1 e1 þ t2 e2 Þ
¼ e1 ¼ 1e1 þ 0e2 :
a11 t1 þ a12 t2 ¼ 1
a21 t1 þ a22 t2 ¼ 0
has a unique solution for ðt1 ; t2 Þ; and hence, a11 a22 a12 a21 6¼ 0: Since
a11 a22 a12 a21 6¼ 0;
a11 a12
is in GLð2; RÞ:
a21 a22
a11 a12
gð T Þ :
a21 a22
We shall try to show that g is an isomorphism from group GLðVÞ onto group
GLð2; RÞ: For this purpose, we must show:
(i) g is 1–1,
(ii) g maps GLðVÞ onto GLð2; RÞ;
(iii) for every S; T in GLðVÞ; gðS TÞ ¼ ðgðSÞÞðgðTÞÞ:
For (i): Let gðSÞ ¼ gðTÞ; where Sðe1 Þ ¼ a11 e1 þ a21 e2 ; Sðe2 Þ ¼ a12 e1 þ a22 e2 ;
Tðe1 Þ ¼ b11 e1 þ b21 e2 ; and Tðe2 Þ ¼ b12 e1 þ b22 e2 : We have to show that S ¼
T; that is, Sðe1 Þ ¼ Tðe1 Þ; and Sðe2 Þ ¼ Tðe2 Þ: Since
a11 ¼ b11 ; and a21 ¼ b21 ; and hence, Sðe1 Þ ¼ a11 e1 þ a21 e2 ¼ b11 e1 þ b21 e2 ¼
Tðe1 Þ: Thus, Sðe1 Þ ¼ Tðe1 Þ: Similarly, Sðe2 Þ ¼ Tðe2 Þ:
4.2 Lie Groups 259
For (ii): We have to show that g : GLðVÞ ! GLð2; RÞ is onto. For this purpose,
let us take any
a11 a12
in GLð2; RÞ:
a21 a22
a11 a12
gð T Þ ¼ :
a21 a22
It is clear that T is linear. Also Tðe1 Þ ¼ Tð1e1 þ 0e2 Þ ¼ ða11 1 þ a12 0Þe1 þ
ða21 1 þ a22 0Þe2 ¼ a11 e1 þ a21 e2 : Thus, Tðe1 Þ ¼ a11 e1 þ a21 e2 : Similarly, Tðe2 Þ ¼
a a12
a12 e1 þ a22 e2 : Now, since det 11 ¼ a11 a22 a12 a21 6¼ 0; T is invertible.
a21 a22
Since T is an invertible linear transformation from V onto V, T is in GLðVÞ: Since
Tðe1 Þ ¼ a11 e1 þ a21 e2 ; and Tðe2 Þ ¼ a12 e1 þ a22 e2 ;
a11 a12
gð T Þ ¼ :
a21 a22
a11 a12
gð SÞ ¼ ;
a21 a22
b11 b12
gð T Þ ¼ ;
b21 b22
Thus,
ðS T Þðe1 Þ ¼ ða11 b11 þ a12 b21 Þe1 þ ða21 b11 þ a22 b21 Þe2 :
Next,
Thus,
ðS T Þðe2 Þ ¼ ða11 b12 þ a12 b22 Þe1 þ ða21 b12 þ a22 b22 Þe2 :
It follows that
a11 b11 þ a12 b21 a11 b12 þ a12 b22 a11 a12 b11 b12
LHS ¼ gðS T Þ ¼ ¼
a21 b11 þ a22 b21 a21 b12 þ a22 b22 a21 a22 b21 b22
¼ ðgðSÞÞðgðT ÞÞ ¼ RHS:
is C 1 at the point IdRR ða; bÞð¼ ða; bÞÞ in R2 : Let us observe that IdR ðf
ðIdRR Þ1 Þ ¼ f ; ðIdRR ÞððR RÞ \ f 1 ðRÞÞ ¼ ðR RÞ \ f 1 ðRÞ ¼ ðR RÞ \
ðR RÞ ¼ R R; and IdR ðRÞ ¼ R; so we have to prove that the function
4.2 Lie Groups 261
f :RR!R
is C1 at the point ða; bÞ: But this is known to be true. Thus, we have shown that
ðx; yÞ 7! x y is a smooth map. This proves that R is a Lie group under addition.
Similarly, R2 is a Lie group under addition, R3 is a Lie group under addition, etc.
Since each ordered pair ðx; yÞ of real numbers x; y can be identified with the
pffiffiffiffiffiffiffi
complex number x þ 1y; and R2 is a Lie group under addition, clearly C is a
Lie group under addition.
Example 4.24 By R we shall mean the set of all nonzero real numbers. We know
that R is a group under multiplication as binary operation. Since the elements of
R can be identified, in a natural way, with the elements of GLð1; RÞ; and
GLð1; RÞ; is a 1-dimensional Lie group, R is also a 1-dimensional Lie group under
multiplication.
Next, since R is a 1-dimensional Lie group, R is a smooth manifold. Since R
is a smooth manifold, and fx : x 2 R and 0\xg is an open subset of R ;
fx : x 2 R and 0\xg is a smooth manifold. We know that fx : x 2 R and 0\xg is
a subgroup of the multiplicative group R : Thus, fx : x 2 R and 0\xg is a 1-
dimensional smooth manifold that is also a group. Since the function ðx; yÞ 7! xy is
smooth so, by Lemma 4.19, fx : x 2 R and 0\xg is a 1-dimensional Lie group
under multiplication.
Example 4.25 By C we shall mean the set of all nonzero complex numbers. We
know that C is a group under multiplication as binary operation. Since the ele-
ments of R can be identified, in a natural way, with the elements of GLð1; CÞ; and
GLð1; CÞ is a 2-dimensional Lie group, C is also a 2-dimensional Lie group under
multiplication.
Example 4.26 We know that the unit circle S1 ð¼ fðx; yÞ : ðx; yÞ 2 R2 and
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
x2 þ y2 ¼ 1gÞ is a 1-dimensional smooth manifold. Observe that we can identify
pffiffiffiffiffiffiffi
ðx; yÞ with the complex number x þ 1y: Thus, we can identify the unit circle S1
with fz : z 2 C and jzj ¼ 1g: Further, we know that fz : z 2 C and jzj ¼ 1g is a
multiplicative group. Thus, unit circle S1 is a 1-dimensional smooth manifold that is
also a group. Since the function ðz; wÞ 7! wz is smooth, by Lemma 4.19, the circle S1
is a 1-dimensional Lie group. The Lie group S1 is called the circle group.
Example 4.27 Let G1 and G2 be Lie groups. Since G1 and G2 are Lie groups, G1
and G2 are groups. We know that the Cartesian product G1 G2 is also a group
under the binary operation given by ðg1 ; h1 Þðg2 ; h2 Þ ðg1 g2 ; h1 h2 Þ: Since G1 and
G2 are Lie groups, G1 and G2 are smooth manifolds, and hence, their Cartesian
product G1 G2 is a smooth manifold (called the product manifold of G1 and G2 ).
Thus, G1 G2 is a 1-dimensional smooth manifold that is also a group.
262 4 Topological Properties of Smooth Manifolds
Example 4.31 We have seen that C is a Lie group under addition, and C is a Lie
group under multiplication. Consider the function exp : C ! C defined as follows:
for every z in C;
pffiffiffiffiffiffiffi
expðzÞ ez ¼ eReðzÞ cosðImðzÞÞ þ 1 sinðImðzÞÞ :
Since
pffiffiffiffiffiffiffi pffiffiffiffiffiffiffi
ðs; tÞ 7! cosð2psÞ þ 1 sinð2psÞ; cosð2ptÞ þ 1 sinð2ptÞ
¼ ððcosð2psÞ; sinð2psÞÞ; ðcosð2ptÞ; sinð2ptÞÞÞ
¼ ðcosð2psÞ; sinð2psÞ; cosð2ptÞ; sinð2ptÞÞ
is a Lie group under matrix multiplication, and R is a Lie group under ordinary
multiplication. Consider the function det : GLð2; RÞ 7! R defined as follows: For
every
x1 x2
2 Mð2; RÞ;
x3 x4
we have
x1 x2
det x1 x4 x2 x3 :
x3 x4
Since
x1 x2
ð x1 ; x2 ; x3 ; x4 Þ ¼ 7! ðx1 x4 x2 x3 Þ
x3 x4
For 1: Let us take any x in S. Since the singleton set {x} is connected subset of
X, and x 2 fxg S, by the definition of *, x x:
For 2: Let x y: Now, by the definition of *, there exists a connected subset
C of X such that C S; x 2 C; and y 2 C: Thus, C is a connected subset of
X such that C S; y 2 C; and x 2 C; and hence, by the definition of *, y x:
For 3: Let x y; and y z: Since x y; so by the definition of *, there exists a
connected subset C of X such that C S; x 2 C; and y 2 C: Similarly, there exists
a connected subset C1 of X such that C1 S; y 2 C1 ; and z 2 C1 : Since C is a
connected subset of X, C1 is a connected subset of X, and y 2 C \ C1 ; C [ C1 is a
connected subset of X. Also x 2 C C [ C1 ; and z 2 C1 C [ C1 : Since C S;
and C1 S; so C [ C1 S: It follows, by the definition of *, that x z:
Thus, we have shown that * is an equivalence relation over S. Hence, S is
partitioned into equivalence classes. Here, each equivalence class is called a
component of S.
Now, we shall try to show that each component of S is a connected subset of X.
Let us take any component [a] of S, where a is in S, and ½a
fx : x 2 S; and x ag: We claim that [a] is connected. If not, otherwise, let [a] be
not connected, that is, there exist open subsets G1 ; G2 of X such that G1 \½a is
nonempty, G2 \½a is nonempty, G1 \ G2 \½a is empty, and ½a G1 [ G2 :
We have to arrive at a contradiction. Since G1 \½a is nonempty, there exists b in
G1 \½a : Similarly, there exists c in G2 \½a : Here, b and c are in [a], and [a] is an
266 4 Topological Properties of Smooth Manifolds
For (1): Let us take any x in S. By the definition of *, we must find a path f in
S from x to x. For this purpose, consider the constant function f : ½0; 1 ! S
defined by
f ðt Þ x
for every t in [0,1]. Since every constant function is a continuous function, and
f is a constant function, f is a continuous function. Further, by the definition of f,
f ð0Þ ¼ x and f ð1Þ ¼ x. Hence, by the definition of path, f is a path in S from x to
x. This proves (1).
For (2): Let x y. So, by the definition of *, there exists a path f in S from x to
y. Since f is a path in S from x to y so, by the definition of path, f is a function
from the closed interval [0,1] to S such that
(i) f is continuous,
(ii) f ð0Þ ¼ x and f ð1Þ ¼ y.
(10 ) h is continuous,
(20 ) hð0Þ ¼ x and hð1Þ ¼ z.
Since x y, by the definition of *, there exists a path in S from x to y, and
hence, there exists a function f : ½0; 1 ! S such that
(100 ) f is continuous,
(200 ) f ð0Þ ¼ x and f ð1Þ ¼ y.
Since y z, by the definition of *, there exists a path in S from y to z, and
hence, there exists a function g : ½0; 1 7! S such that
(1000 ) g is continuous,
(2000 ) gð0Þ ¼ y and gð1Þ ¼ z. Let us define a function h : ½0; 1 7! X as follows:
f ð2tÞ if 0 t\ 12
hð t Þ
gð2t 1Þ if 12 t 1:
Clearly, the range of h is the union of the range of f, and the range of g. Further,
since f : ½0; 1 ! S; and g : ½0; 1 ! S; the range of h is contained S, and hence
h : ½0; 1 7! S:
For ð10 Þ: we have to prove that h is continuous over [0, 1], that is, h is continuous at
all points of [0, 1]. By the definition of h over 0 t\ 12, h is the composite function
of two continuous functions, namely t 7! 2t and f so, h is continuous at all points of
ft : 0 t\ 12g: Similarly, h is continuous at all points of ft : 12 \t 1g: So, it
remains to prove that h is continuous at 12 : For this purpose, we should calculate the
left-hand limit of h at 12, the right-hand limit of h at 12 and hð12Þ. If all the three values
are equal, then we can say that h is continuous at 12. In short, we have to prove:
1
lim hðtÞ ¼ lim þ hðtÞ ¼ h :
t!ð2Þ
1
t!ð12Þ 2
Thus, we have shown that h is continuous over [0,1]. This proves ð10 Þ
For ð20 Þ: Here hð0Þ ¼ f ð2ð0ÞÞ ¼ f ð0Þ ¼ x, and hð1Þ ¼ gð2ð1Þ 1Þ ¼ gð1Þ ¼ z.
This proves ð20 Þ: Thus, we have shown: If x y and y z, then x z: This
proves (3).
Hence, * is an equivalence relation over S. Hence, S is partitioned into
equivalence classes. Here, each equivalence class is called a path component of S.
4.3 Locally Path Connected Spaces 269
where y 0 a stands for: There exists a path in S from y to a. For this purpose, let us
take any y in LHS. We have to prove that y is in RHS, that is, y a: By the
definition of *, we should find a connected subset A of X such that A S; and y,
a are in A. Since y is in LHS, y 0 a. Since y 0 a, by the definition of 0 , there
exists a path f in S from y to a. Since f is a path in S from y to a, f is a function from
closed interval [0,1] to S satisfying
(I) f is continuous,
(II) f ð0Þ ¼ y and f ð1Þ ¼ a. Since f : ½0; 1 ! S is continuous, and [0,1] is con-
nected, so the f-image set f ð½0; 1 Þ of [0, 1] is a connected subset of X. Since
0 is in [0, 1], f ð0Þ is in f ð½0; 1 Þ. Similarly, f ð1Þ is in f ð½0; 1 Þ: Further,
f ð0Þ ¼ y and f ð1Þ ¼ a: Thus, f ð½0; 1 Þ is a connected subset of X such that
f ð½0; 1 Þ S; and y, a are in f ð½0; 1 Þ:
Note 4.40 Let X be a topological space. We shall try to show: If X is path con-
nected, then X is connected. If not, otherwise, let X be not connected. We have to
arrive at a contradiction. Since X is not connected, there exist open subsets G1 ; G2
of X such that G1 is nonempty, G2 is nonempty, G1 \ G2 is empty, and X ¼
G1 [ G2 : Since G1 is nonempty, there exists a in G1 : Similarly, there exists b in G2
Now, since X is path connected, there exists a continuous function f : ½0; 1 ! X
such that f ð0Þ ¼ a; and f ð1Þ ¼ b: Since f : ½0; 1 ! X is continuous, and [0, 1] is
connected, the image set f ð½0; 1 Þ is connected. Here, a is in G1 ; and a ¼ f ð0Þ 2
f ð½0; 1 Þ; so a 2 G1 \ f ð½0; 1 Þ; and hence, G1 \ f ð½0; 1 Þ is nonempty. Similarly,
G2 \ f ð½0; 1 Þ is nonempty. Next, since G1 \ G2 is empty, G1 \ G2 \ f ð½0; 1 Þ is
empty. Since G1, G2 are open subsets of X such that G1 \ f ð½0; 1 Þ is nonempty,
G2 \ f ð½0; 1 Þ is nonempty, G1 \ G2 \ f ð½0; 1 Þ is empty, and f ð½0; 1 Þ X ¼
G1 [ G2 ; f ð½0; 1 Þ is not connected, a contradiction.
Definition Let X be a topological space. If for every x in X, and for every open
neighborhood U of x, there exists an open neighborhood V of x such that
(i) x 2 V U;
(ii) V is a connected subset of X,
then we say that X is a locally connected space.
Definition Let X be a topological space. If for every x in X, and for every open
neighborhood U of x, there exists an open neighborhood V of x such that
(i) x 2 V U;
(ii) V is a path connected subset of X,
then we say that X is a locally path connected space.
4.3 Locally Path Connected Spaces 271
½a fx : x 2 G and x ag
for some a 2 G, where x a stands for: There exists a connected subset A of X such
that A G; x 2 A; and a 2 A: We have to show that [a] is an open subset of X, that
is, every point of [a] is an interior point of [a]. For this purpose, let us take any
x 2 ½a . We must find an open neighborhood V of x such that V ½a . Since x 2 ½a
and ½a G, x 2 G. Since x 2 G and G is open in X, G is an open neighborhood of
x. Since X is a locally connected space, and G is an open neighborhood of x, there
exists an open neighborhood V of x such that
(I) x 2 V G;
(II) V is a connected subset of X.
We claim that V ½a :
If not, otherwise, let V 6 ½a : So, there exists b in V such that b 62 ½a . Since [a] is
an equivalence class of a, and b 62 ½a , b ¿ a: Here, b 2 V G so b 2 G: Similarly,
x 2 G. Further, since V is a connected subset of X, V G; and x, b are in V, by the
definition of *, x b: Since x b; and * is an equivalence relation, ½x ¼ ½b :
Since x 2 ½a ; and * is an equivalence relation, ½x ¼ ½a : Since ½x ¼ ½b ; and
½x ¼ ½a , ½a ¼ ½b . Since ½a ¼ ½b , a b. This is a contradiction. So, our claim is
true, that is, V ½a :
Clearly, if X is locally connected space, then each component of X is open.
Note 4.42 Let X be a topological space. We shall try to prove: If X is locally path
connected space, and G is a nonempty open subset of X, then each path component
of G is open. Let X be a locally path connected space, and let G be an open subset of
X. We have to show that each path component of G is open in X. By the definition
of path component, every path component of G is of the form
½a fx : x 2 G and x ag
B fy 2 Rn : jy uð xÞj\eg
B is an open subset of Ue 1 , u1 ðBÞ is an open subset of U1. Let us take u1 ðBÞ for
V in (I) and (II). So, we must prove:
(I) x 2 u1 ðBÞ U;
(II) u1 ðBÞ is a path connected subset of M.
For (i): Let M be path connected. We have to prove that M is connected. Since
M is path connected, M is a path component. Since M is a topological manifold,
M is locally path connected. Since M is locally path connected, and M is a path
component, M is a component. Since M is a component, M is connected.
Converse: Let M be connected. We have to prove that M is path connected.
Since M is connected, M is a component. Since M is a topological manifold,
M is locally path connected. Since M is locally path connected, and M is a
component, M is a path component. Since M is a path component, M is path
connected.
This completes the proof of (i).
For (ii): Since M is a topological manifold, M is locally path connected. Since
M is locally path connected, the collection of all path components of M is equal
to the collection of all components of M. This proves (ii).
For (iii): We have to prove that M has countable many components. If not,
otherwise, let M has uncountable many components. We have to arrive at a
contradiction.
Let C be a component of M. Since M is a topological manifold, M is locally path
connected. Since M is locally path connected, and C is a component of M, C is a
path component. Since M is locally path connected, and C is a path component, C is
open in M. Since C is open in M, and a is in C, C is an open neighborhood of a in
M. Since M is a topological manifold, M is a second countable space. Since M is a
second countable space, there exists a countable collection C of open subsets of
M such that for every neighborhood U of x, there exists a member G of C such that
x 2 G U: Now, since C is an open neighborhood of a, there exists a member G of
C such that x 2 G C:
Thus, we see that corresponding to each component C, there exists a member
G of C such that G is contained in C. Since M is partitioned into components, each
component contains a member of C, and the collection of all components is
uncountable, the collection C is uncountable. This is a contradiction. This proves
that M has countable many components.
276 4 Topological Properties of Smooth Manifolds
Now, let us take a component C. We have to prove that C is open and connected.
Since M is a topological manifold, M is locally path connected. Since M is locally
path connected and C is a component of M, C is a path component. Since M is
locally path connected, and C is a path component, C is open in M. Since C is a
component, C is connected. Hence, C is open and connected.
Lastly, we have to show that C is a topological manifold of dimension n, where
topology on C is the induced topology of M. Since C is an open subset of topo-
logical space M, under induced topology of M, open subsets of C are exactly those
subsets of C which are open in M. By the definition of topological manifold, we
must prove:
1. C is a Hausdorff space,
2. C is a second countable space,
e1
3. for every x in C, there exist a neighborhood U1 of x in C, and an open subset U
of Euclidean space R such that U1 and U
n e 1 are homeomorphic:
G1 fGU : U 2 B1 g:
B is a basis of Rn : Since the set of all rational numbers is countable, the collection B
is countable. Let B1 be the collection of all Br ðxÞ 2 B such that the closure ðBr ðxÞÞ
of Br ðxÞ is contained in U e 1 : We shall try to show that B1 is a basis of U e 1:
For this purpose, let us take any open neighborhood V of y such that V ~ ~ is
contained in U e 1 : Since V ~ is an open neighborhood of y, and B is a basis of R ; there n
exists an open ball Bry ðxy Þ in B such that y 2 Bry ðxy Þ V; ~ where ry is a positive
rational, and each coordinate of xy is a rational number. Since y 2 Bry ðxy Þ;
jy xy j\ry : There exists a positive rational number sy such that jy xy j\sy \ry :
Hence, y 2 Bsy ðxy Þ ðBsy ðxy ÞÞ fz : jz xy j sy g Bry ðxy Þ V ~U e 1 : Since
sy is a positive rational, and each coordinate of xy is a rational number, Bsy ðxy Þ is in
B: Since Bsy ðxy Þ is in B; and ðBsy ðxy ÞÞ U e 1 ; Bsy ðxy Þ is in B1 . Also y 2 Bsy ðxy Þ
~ and Bsy ðxy Þ is an open neighborhood of y, B1 is a basis of U
V; e 1 : Thus, we have
shown that B1 is a basis of U e 1:
Since B is countable, and B1 is contained in B; B1 is countable. Thus, B1 is a
countable basis of U e 1 : Let us take any Br ðxÞ in B1 : Since Br ðxÞ is in B1 ; and B1 is a
e
basis of U 1 ; Br ðxÞ is an open subset of the open set U e 1 : Since Br ðxÞ is an open
subset of U e 1 ; and u1 : U1 ! U e 1 is a homeomorphism, u1 e
1 ðBr ðxÞÞ is open in U 1 :
1 e e
Since u1 ðBr ðxÞÞ is open in U 1 ; and u1 : U1 ! U 1 is a homeomorphism, the
restriction u1 ju1 ðBr ðxÞÞ : u11 ðBr ðxÞÞ ! Br ðxÞ is a homomorphism. Thus, the
1
homeomorphism. Let B2 be the collection of all Br ðxÞ 2 B such that the closure
e2
ðBr ðxÞÞ of Br ðxÞ is contained in U
As above, fðu1
2 ðBr ðxÞÞ; u2 ju1 ðBr ðxÞÞ Þ : Br ðxÞis in B 2 g is a countable collection
2
of coordinate charts of M such that
(a) fu1
2 ðBr ðxÞÞ : Br ðxÞis in B2 g is a basis of U2 ;
(b) for each Br ðxÞin B2 ; ðu2 ju1 ðBr ðxÞÞ Þðu1
2 ðBr ðxÞÞÞ is an open ball in R ;
n
2
(c) for each Br ðxÞin B2 ; the closure ðu1 1
2 ðBr ðxÞÞÞ of u2 ðBr ðxÞÞ is a compact
subset of U2 ; etc.
Since the countable union of countable sets is countable, the collection
[1 1
i¼1 fðui ðBr ðxÞÞ; ui ju1 Þ
: Br ðxÞis in Bi g of coordinates charts is countable.
i ðBr ðxÞÞ
It remains to prove that
I. [1 1
i¼1 fui ðBr ðxÞÞ : Br ðxÞis in Bi g is a basis of M,
II. for each i ¼ 1; 2; 3; . . .; and for each Br ðxÞ in Bi ; ðui ju1 ðBr ðxÞÞ Þðu1
i ðBr ðxÞÞÞ is
i
an open ball in R ; n
III. for each i ¼ 1; 2; 3; . . .; and for each Br ðxÞ in Bi ; the closure ðu1 i ðBr ðxÞÞÞ of
1
ui ðBr ðxÞÞ is a compact subset of M, etc.
For I: Let G be any open neighborhood of a in M. Since fU1 ; U2 ; U3 ; . . .g is a cover
of M, there exists a positive integer i such that a 2 Ui : Since Ui \ G is an open
neighborhood of a, Ui \ G is contained in Ui ; and fu1 i ðBr ðxÞÞ : Br ðxÞ is in Bi g is
a basis of Ui ; there exists Br ðxÞ in Bi such that a 2 u1 i ðBr ðxÞÞ Ui \ G G:
This shows that [1 i¼1 fu1
i ðBr ðxÞÞ : Br ðxÞ is in B i g is a basis of M.
For II: This is clear.
For III: This is clear. h
uk ðUk Þ ! Uk is continuous, and Br ½uk ðpÞ is compact, u1 k ðBr ½uk ðpÞ Þ is compact.
Since M is topological manifold, the topology of M is Hausdorff. Since the topology
of M is Hausdorff, and u1 1
k ðBr ½uk ðpÞ Þ is compact, uk ðBr ½uk ðpÞ Þ is closed in M.
Clearly, uk ðBr ðuk ðpÞÞÞ is an open neighborhood of p, and u1
1
k ðBr ðuk ðpÞÞÞ
u1
k ðB ½u
r k ðpÞ Þ: Let us take u 1
k ðB r ðu k ðpÞÞÞ for V. Thus, V is an open neighbor-
1 1
hood of p, and V uk ðBr ½uk ðpÞ Þ: Since V uk ðBr ½uk ðpÞ Þ; and
u1 1
k ðBr ½uk ðpÞ Þ is closed in M, V uk ðBr ½uk ðpÞ Þ Uk U: Thus, V U:
Proof Let C be the collection of all admissible coordinate charts ðU; uÞ of M. Since
M is a smooth manifold, fU : ðU; uÞ 2 Cg is an open cover of M. Since M is a
smooth manifold, M is second countable. Since M is a second countable space, and
fU : ðU; uÞ 2 Cg is an open cover of M, there exists a countable subcover
fU1 ; U2 ; U3 ; . . .g of fU : ðU; uÞ 2 Cg:
Here, U1 2 fU1 ; U2 ; U3 ; . . .g fU : ðU; uÞ 2 Cg; U1 2 fU : ðU; uÞ 2 Cg; and
hence, there exists a function u1 ; and an open subset U e 1 of Rn such that
u1 : U1 ! U e 1 , and ðU1 ; u1 Þ 2 C: Let B be the collection of all open balls Br ðxÞ
with center x and radius r such that r is a positive rational, and each coordinate of
x is a rational number. We know that B is a basis of Rn : Since the set of all rational
numbers is countable, the collection B is countable. Let B1 be the collection of all
Br ðxÞ 2 B such that the closure ðBr ðxÞÞ of Br ðxÞ is contained in U e 1:
e
We shall try to show that B1 is a basis of U 1 : For this purpose, let us take any
open neighborhood V ~ of y such that V ~ is contained in U e 1 : Since V ~ is an open
neighborhood of y, and B is a basis of Rn ; there exists an open ball Bry ðxy Þ in B such
that y 2 Bry ðxy Þ V; ~ where ry is a positive rational, and each coordinate of xy is a
rational number. Since y 2 Bry ðxy Þ; jy xy j\ry : There exists a positive rational
number sy such that jy xy j\sy \ry : Hence, y 2 Bsy ðxy Þ ðBsy ðxy ÞÞ
fy : jy xy j sy g Bry ðxy Þ V ~U e 1 : Since sy is a positive rational, and each
coordinate of xy is a rational number, Bsy ðxy Þ is in B: Since Bsy ðxy Þ is in B; and
ðBsy ðxy ÞÞ U e 1 ; Bsy ðxy Þ is in B1 . Also y 2 Bsy ðxy Þ V;~ and Bsy ðxy Þ is an open
neighborhood of y, B1 is a basis of U 1 : e
Thus, we have shown that B1 is a basis of U e 1 : Since B is countable, and B1 is
contained in B; B1 is countable. Thus, B1 is a countable basis of U e 1 : Let us take any
4.4 Smooth Manifold as Paracompact Space 281
Br ðxÞ is a homomorphism. Thus, the ordered pair ðu1 1 ðBr ðxÞÞ; u1 ju1 ðBr ðxÞÞ Þ is a
1
coordinate chart of U1 :
Now, we want to show that ðu1 1 ðBr ðxÞÞ; u1 ju1 ðBr ðxÞÞ Þ 2 C: For this purpose, let
1
us take any ðW; wÞ 2 C: Now, we must prove that
u1 ju1 ðBr ðxÞÞ w1 : w u1 1 ðBr ð xÞÞ \ W
1
! u1 ju1 ðBr ðxÞÞ u11 ð B r ð xÞ Þ \ W ;
1
and
1
w u1 ju1 ðBr ðxÞÞ : u1 ju1 ðBr ðxÞÞ u1
1 ðBr ð xÞÞ \ W
1 1
are C 1 : Let us take any ðy1 ; . . .; yn Þ 2 wððu1 1 ðBr ðxÞÞÞ \ WÞ: So, there exists p in
ðu1
1 ðB r ðxÞÞÞ \ W such that wðpÞ ¼ ðy 1 ; . . .; yn Þ; and hence w1 ðy1 ; . . .; yn Þ ¼ p;
and p 2 u1 1 ðBr ðxÞÞ: Now,
u1 ju1 ðBr ðxÞÞ w1 ðy1 ; . . .; yn Þ
1
so ðu1 ju1 ðBr ðxÞÞ Þ w1 ¼ ðu1 w1 Þjwððu1 ðBr ðxÞÞÞ \ WÞ : Since ðU1 ; u1 Þ 2 C; ðW; wÞ 2
1 1
w1 is C 1 : Thus, we have shown that ððu1 ju1 ðBr ðxÞÞ Þ w1 Þ : wððu1
1 ðBr ðxÞÞÞ
1
1
\WÞ ! ðu1 ju1 ðBr ðxÞÞ Þððu1 1
1 ðBr ðxÞÞÞ \ WÞis C : Similarly, ðw ðu1 ju1 ðBr ðxÞÞ Þ Þ :
1 1
fU : U 2 U and U \ G is nonemptyg
is a finite set, then we say that U is locally finite. In other words, U is locally finite
means for every x in X, there exists an open neighborhood G of x such that all
members of U except finite many are “outside” G. In short, if for every x in X, there
exists an open neighborhood G of x such that G intersects only finite many
members of U; then we say that U is locally finite. Clearly, every subcollection of a
locally finite collection is locally finite.
Note 4.50 Let X be a topological space. Let U be an open cover of X. Let each
member of U intersects only finite many members of U: We shall try to show that U
is locally finite. For this purpose, let us take any x in X. Since x is in X, and U is a
cover of X, there exists G in U such that x is in G. Since G is in U; and U is an open
cover of X, G is open. Since G is open, and x is in G, G is an open neighborhood of
x. Since G is in U; and each member of U intersects only finite many members of U;
G intersects only finite many members of U: This proves that U is locally finite.
Lemma 4.51 Let M be a topological manifold. There exists a countable collection
fG1 ; G2 ; G3 ; . . .g of open sets such that
1. fG1 ; G2 ; G3 ; . . .g is a cover of M,
2. fG1 ; G2 ; G3 ; . . .g is locally finite,
3. for each n ¼ 1; 2; 3; . . .; the closure G
n of Gn is compact.
Put G1 B1 ; G2 B2 ; G3 B3 ðB
1 Þ; G4 B4 ðB2 Þ; G5 B5 ðB3 Þ; etc.
Clearly fG1 ; G2 ; G3 ; . . .g is a countable collection of open sets (see Fig. 4.1).
It remains to prove that
1. fG1 ; G2 ; G3 ; . . .g is a cover of M,
2. fG1 ; G2 ; G3 ; . . .g is locally finite,
3. for each n ¼ 1; 2; 3; . . .; the closure G
n of Gn is compact.
is a finite set.
Since p 2 Wp ; fGk : p 2 Gk g ¼ fGk : p 2 Gk \ Wp g fGk : Gk \ Wp is nonemptyg:
Since fGk : p 2 Gk g fGk : Gk \ Wp is nonemptyg; and fGk : Gk \ Wp is nonemptyg
is finite, fGk : p 2 Gk g is finite. Since fG1 ; G2 ; G3 ; . . .g is a cover of M, and p 2 M;
fGk : p 2 Gk g is nonempty. Since fGk : p 2 Gk g is a nonempty finite collection of
open neighborhoods of p, \p2Gk Gk is an open neighborhood of p.
286 4 Topological Properties of Smooth Manifolds
Similarly, there exists finite many p21 ; p22 ; . . .; p2k2 2 G 2 such that G2
Up21 [ Up22 [ [ Up2k2 ; etc. Clearly, ðUp11 ; wp11 Þ; ðUp12 ; wp12 Þ; . . .; ðUp1k1 ; wp1k1 Þ,
ðUp21 ; wp21 Þ; ðUp22 ; wp22 Þ; . . .; ðUp2k2 ; wp2k Þ; ðUp31 ; wp31 Þ; ðUp32 ; wp32 Þ; . . .; ðUp3k3 ; wp3k Þ; . . .
2 3
constitute a countable collection of admissible coordinate charts of M:
1 1 1
For 1: Since G1 G
1 ðwp11 Þ ðB1 ð0ÞÞ [ðwp12 Þ ðB1 ð0ÞÞ [ [ðwp1k1 Þ
1 1
ðB1 ð0ÞÞ Up11 [ Up12 [ [ Up1k1 ; G2 G
2 ðwp21 Þ ðB1 ð0ÞÞ [ðwp22 Þ ðB1 ð0ÞÞ
1
[ [ðwp2k2 Þ ðB1 ð0ÞÞ Up21 [ Up22 [ [ Up2k2 ; . . .; and fG1 ; G2 ; G3 ; g is
a cover of M; fUp11 ; Up12 ; . . .; Up1k1 ; Up21 ; Up22 ; . . .; Up2k2 ; Up31 ; Up32 ; . . .; Up3k3 ; . . .g
is an open cover of M:
For 2: We have to prove that fUp11 ; Up12 ; . . .; Up1k1 ; Up21 ; Up22 ; . . .; Up2k2 ;
Up31 ; Up32 ; . . .; Up3k3 ; . . .g is locally finite.
Since fG1 ; G2 ; G3 ; . . .g is a cover of M; and p11 2 M; there exists a positive
integer n11 such that p11 2 Gn11 : Since p11 2 Gn11 ; Up11 Gn11 : Similarly, there
exists a positive integer n12 such that Up12 Gn12 ; etc.
Since fUp11 ; Up12 ; . . .; Up1k1 ; Up21 ; Up22 ; . . .; Up2k2 ; Up31 ; Up32 ; . . .; Up3k3 ; . . .g is a
cover of M; and each Upij is contained in Gnij ; fGn11 ; Gn12 ; . . .; Gn1k1 ; Gn21 ; Gn22 ; . . .;
Gn2k2 ; Gn31 ; Gn32 ; . . .; Gn3k3 ; . . .g is a cover of M: Since fG1 ; G2 ; G3 ; . . .g is locally
finite, and fGn11 ; Gn12 ; . . .; Gn1k1 ; Gn21 ; Gn22 ; . . .; Gn2k2 ; Gn31 ; Gn32 ; . . .; Gn3k3 ; . . .g
fG1 ; G2 ; G3 ; . . .g; fGn11 ; Gn12 ; . . .; Gn1k1 ; Gn21 ; Gn22 ; . . .; Gn2k2 ; Gn31 ; Gn32 ; . . .; Gn3k3 ; . . .g is
locally finite. Since fGn11 ; Gn12 ; . . .; Gn1k1 ; Gn21 ; Gn22 ; . . .; Gn2k2 ; Gn31 ; Gn32 ; . . .;
Gn3k3 ; . . .g is locally finite, and each Upij is contained in Gnij ;
fUp11 ; Up12 ; . . .; Up1k1 ; Up21 ; Up22 ; . . .; Up2k2 ; Up31 ; Up32 ; . . .; Up3k3 ; . . .g is locally finite.
For 3: We have to show that the open cover fUp11 ; Up12 ; . . .; Up1k1 ;
Up21 ; Up22 ; . . .; Up2k2 ; Up31 ; Up32 ; . . .; Up3k3 ; . . .g is a refinement of the open cover X :
Here Up11 ð\p11 2Gk Gk Þ \ Wp11 \ Xp11 \ Vp11 Xp11 2 X : Thus, Up11 Xp11 2 X :
Similarly, Up12 Xp12 2 X ; etc.
For 4: This is clear.
For 5: We have to show that
n o
w1 1 1 1 1 1
p11 ðB1 ð0ÞÞ; wp12 ðB1 ð0ÞÞ; . . .; wp1k ðB1 ð0ÞÞ; wp21 ðB1 ð0ÞÞ; wp22 ðB1 ð0ÞÞ; . . .; wp2k ðB1 ð0ÞÞ; . . .
1 2
covers M.
1 1 1
Since G1 G
1 ðwp11 Þ ðB1 ð0ÞÞ [ðwp12 Þ ðB1 ð0ÞÞ [ [ðwp1k1 Þ ðB1 ð0ÞÞ;
1 1 1
G2 G
2 ðwp21 Þ ðB1 ð0ÞÞ [ðwp22 Þ ðB1 ð0ÞÞ [ [ðwp2k2 Þ ðB1 ð0ÞÞ; . . .; and
288 4 Topological Properties of Smooth Manifolds
ð0ÞÞ; w1 1
p21 ðB1 ð0ÞÞ; wp22 ðB1 ð0ÞÞ; ; w1
p2k ðB1 ð0ÞÞ; g is a cover of M: h
2
Note 4.54 (Pasting Lemma) Let X and Y be topological spaces. Let A and B be
subsets of X such that the union of A and B is X: Let f : A ! Y be a continuous
function, where the topology of A is the induced topology of X: Let g : B ! Y be a
continuous function, where the topology of B is the induced topology of X: Let
f ðxÞ ¼ gðxÞ for every x in A \ B. Let F : X ! Y be a function defined as follows:
f ð xÞ if x is in A
F ð xÞ
gð xÞ if x is in B:
f ð xÞ if x is in A
F ð xÞ
gð xÞ if x is in B;
Now, since U is locally finite, there exists an open neighborhood Ga of a such that
all but finite many U 2 U are outside Ga : For every U 2 U, we have a 62 U ; there
exists an open neighborhood HU of a such that HU contains no point of U ; and
hence, HU contains no point of U: Since all but finite many U 2 U are outside Ga ;
the finite intersection ð\Ga \ U6¼; HU Þ \ Ga is an open neighborhood of a: Since for
every U 2 U; HU contains no point of U; and all but finite many U 2 U are outside
Ga ; ð\Ga \ U6¼; HU Þ \ Ga contains no point of [fU : U 2 Ug: It follows that a is not
in ð[fU : U 2 UgÞ ; a contradiction. h
Definition Let X be a topological space. Let f : X ! Rk be any function. The
closure fx : f ðxÞ 6¼ 0g of fx : f ðxÞ ¼
6 0g is denoted by supp f ; and is called the
support of f :
Lemma 4.56 Let M be a 3-dimensional smooth manifold. Let fXi gi2I be an open
cover of M: Then, there exists a family fwi gi2I of functions wi : M ! ½0; 1 such
that
1. for each i 2 I; wi is smooth,
2. for each i 2 I; supp wi is contained in Xi ;
3. fsupp wi gi2I is locally finite,
P
4. i2I wi ¼ 1:
Proof By the Lemma 4.52, there exists a countable collection fðU1 ; u1 Þ; ðU2 ;
u2 Þ; ðU3 ; u3 Þ; . . .g of admissible coordinate charts of M such that
i. fU1 ; U2 ; U3 ; . . .g covers M;
ii. fU1 ; U2 ; U3 ; . . .g is locally finite,
iii. fU1 ; U2 ; U3 ; . . .g is a refinement of fXi gi2I ;
iv. each un ðUn Þ is equal to the open ball B3 ð0Þ;
v. fu1 1 1
1 ðB1 ð0ÞÞ; u2 ðB1 ð0ÞÞ; u3 ðB1 ð0ÞÞ; . . .g covers M.
Un ¼ u1
n ðB3 ð0ÞÞ u1
n ðB2 ½0 Þ:
If x 2 Un u1 1
n ðB2 ½0 Þ; then x 62 un ðB2 ½0 Þ; and hence, un ðxÞ 62 B2 ½0 ¼
1 1
supp H ¼ ðH ðR f0gÞÞ H ðR f0gÞ: It follows that if x 2 Un
290 4 Topological Properties of Smooth Manifolds
u1 1
n ðB2 ½0 Þ; then un ðxÞ 62 H ðR f0gÞ; and hence, Hðun ðxÞÞ ¼ 0: This shows
that the function fn : M ! ½0; 1 defined as follows: For every x in M;
H ðun ð xÞÞ if x 2 Un ;
fn ð xÞ
0 6 u1
if x 2 n ðB2 ½0 Þ;
ðH un Þð xÞ if x 2 Un ;
f n ð xÞ ¼
0 if x 2 M ðun 1 ðB2 ½0 ÞÞ
is a well-defined function, Un and M ðu1 n ðB2 ½0 ÞÞ are open sets, the restriction
fn jUn ð¼ H un Þ is smooth, and the restriction fn jMðu1 n ðB2 ½0 ÞÞ
ð¼ 0Þ is smooth, fn is
smooth.
Let us fix any a in M: Since fU1 ; U2 ; U3 ; . . .g is a locally finite cover of M, and a is
in M; there exists an open neighborhood Ga of a; and a positive integer N such that
k N implies Uk is outside Ga : For k N; Uk ¼ u1 k ðB3 ð0ÞÞ u1
k ðB2 ½0 Þ; and Uk is
1
outside Ga ; uk ðB2 ½0 Þ is outside Ga for every k N: Since for every k N; Ga is
outside u1 k ðB2 ½0 Þ; by the definition of fk ; for every x in the open neighborhood Ga of
a; and k N; we have fk ðxÞ ¼ 0: Now, since a 2 Ga ; fN ðaÞ ¼ 0 ¼ fNþ1 ðaÞ ¼
fNþ2 ðaÞ ¼ : It follows that f1 ðaÞ þ f2 ðaÞ þ f3 ðaÞ þ is a real number.
Since fu1
1 ðB 1 ð0ÞÞ; u 1
2 ðB1 ð0ÞÞ; u 1
3 ðB1 ð0ÞÞ; . . .g covers M; and a is in M; there
exists a positive integer N0 such that a 2 u1 N0 ðB 1 ð0ÞÞ u 1
N0 ðB3 ð0ÞÞ ¼ UN0 ; and
hence uN0 ðaÞ 2 B1 ð0Þ B1 ½0 : Since uN0 ðaÞ 2 B1 ½0 ; a 2 UN0 ; and HðxÞ ¼ 1 for
every x in the closed ball B1 ½0 ; fN0 ðaÞ ¼ HðuN0 ðaÞÞ ¼ 1:
Since f1 : M ! ½0; 1 ; f2 : M ! ½0; 1 ; f3 : M ! ½0; 1 ; . . .; and fN0 ðaÞ ¼ 1;
f1 ðaÞ þ f2 ðaÞ þ f3 ðaÞ þ fN0 ðaÞ ¼ 1:
Further, since 1 f1 ðaÞ þ f2 ðaÞ þ f3 ðaÞ þ ; and f1 ðaÞ þ f2 ðaÞ þ f3 ðaÞ þ is
a real number,
4.5 Partitions of Unity Theorem 291
fn ðaÞ
2 ½0; 1 :
f1 ðaÞ þ f2 ðaÞ þ f3 ðaÞ þ
f n ð xÞ
gn ð x Þ
f 1 ð xÞ þ f 2 ð xÞ þ f 3 ð xÞ þ
is well defined. Further, we have shown that for every a in M; there exists an open
neighborhood Ga of a; and a positive integer N such that for every x in the open
neighborhood Ga ;
f n ð xÞ
gn ð x Þ ¼ :
f1 ð xÞ þ f2 ð xÞ þ þ f N ð xÞ
f n ð xÞ
x 7! ð ¼ gn ð x Þ Þ
f 1 ð xÞ þ f 2 ð xÞ þ þ f N ð xÞ
f N ð xÞ 0
gN ð x Þ ¼ ¼ ¼ 0;
f 1 ð xÞ þ f 2 ð xÞ þ þ fN ð xÞ f1 ð xÞ þ f2 ð xÞ þ þ f N ð xÞ
fNþ1 ð xÞ 0
gNþ1 ð xÞ ¼ ¼ ¼ 0;
f 1 ð xÞ þ f 2 ð xÞ þ þ fN ð xÞ f1 ð xÞ þ f2 ð xÞ þ þ f N ð xÞ
gNþ1 ðxÞ ¼ 0; . . .:
g1 ð x Þ þ g2 ð x Þ þ g3 ð x Þ þ
1 1
¼ f 1 ð xÞ þ f2 ð xÞ
f1 ð xÞ þ f2 ð xÞ þ f3 ð xÞ þ f1 ð xÞ þ f2 ð xÞ þ f3 ð xÞ þ
1
þ f 3 ð xÞ þ
f 1 ð xÞ þ f 2 ð xÞ þ f 3 ð xÞ þ
1
¼ ðf1 ð xÞ þ f2 ð xÞ þ f3 ð xÞ þ Þ ¼ 1:
f1 ð xÞ þ f2 ð xÞ þ f3 ð xÞ þ
In short, g1 þ g2 þ g3 þ ¼ 1:
292 4 Topological Properties of Smooth Manifolds
is well defined.
For 1: Case I: When Pfm : iðmÞ ¼ i g is nonempty. Let us take any a in M: We
have to prove that iðmÞ¼i gm ð¼ wi Þ is smooth at a: We have seen that there
exist an open neighborhood Ga of a; and positiveP integers m1 ; m2 ; . . .; mk such
that i ¼ iðm1 Þ ¼ iðm2 Þ ¼ ¼ iðmk Þ; and ð iðmÞ¼i gm ÞjGa ¼ ðgm1 þ gm2
þ þ gmk ÞjGa : Since each gn is smooth, ðgm1 þ gm2 þ þ gmk ÞjGa is smooth
P
at a: Since ðgm1 þ gm2 þ þ gmk ÞjGa is smooth at a; and ð iðmÞ¼i gm ÞjGa ¼
P
ðgm1 þ gm2 þ þ gmk ÞjGa ; ð iðmÞ¼i gm ÞjGa is smooth at a: This shows that
P
wi ð¼ iðmÞ¼i gm Þ is smooth.
Case 2: When fm : iðmÞ ¼ i gis the empty set. Since the constant function 0 is
smooth, and wi ¼ 0; wi is smooth. So, in all cases, wi is smooth.
For 2: Case I: When fm : iðmÞ ¼ i g is nonempty. We have to prove that
supp wi Xi : Here
0 1 00 11 1
X X
supp wi ¼ supp @ gm A ¼ @ @ gm A ð R f0gÞ A
iðmÞ¼i iðmÞ¼i
8 0 1 9 8 9
< X = < X =
¼ x : 0\@ gm Að xÞ ¼ x : 0\ ð gm ð x Þ Þ
: iðmÞ¼i
; : iðmÞ¼i
;
8 9
< X 1
=
¼ x : 0\ f m ð xÞ
: iðmÞ¼i
f 1 ð xÞ þ f 2 ð xÞ þ f 3 ð xÞ þ ;
4.5 Partitions of Unity Theorem 293
8 9
< 1 X =
¼ x : 0\ ðfm ð xÞÞ
: f1 ð xÞ þ f2 ð xÞ þ f3 ð xÞ þ iðmÞ¼i
;
8 9
< X =
¼ x : 0\ fm ð xÞ ¼ [iðmÞ¼i fx : 0\fm ð xÞg Xi :
: iðmÞ¼i
;
X X X X
LHS ¼ wi ¼ wi ¼ wi þ wi
i2I i 2I i 2I i 2I
fm:iðmÞ¼i gis non empty fm:iðmÞ¼i gis empty set
X X X
¼ wi þ 0¼ wi
i 2I i 2I i 2I
fm:iðmÞ¼i gis non empty fm:iðmÞ¼i gis empty set fm:iðmÞ¼i gis non empty
0 1
X X
¼ @ gm A ¼ g1 þ g2 þ g3 þ ¼ 1 ¼ RHS: h
i 2I iðmÞ¼i
fm:iðmÞ¼i gis non empty
Note 4.57 Clearly, the above lemma is also valid for m-dimensional smooth
manifold M.
Definition Let M be a topological space. Let fXi gi2I be an open cover of M: Let
fwi gi2I be a family of functions wi : M ! ½0; 1 satisfying the following conditions:
1. for each i 2 I; wi is continuous,
2. for each i 2 I; supp wi is contained in Xi ;
3. fsupp wi gi2I is locally finite,
P
4. wi ¼ 1:
i2I
294 4 Topological Properties of Smooth Manifolds
Then, we say that fwi gi2I is a partition of unity subordinate to fXi gi2I : Since
fsupp wi gi2I is locally
P finite, for P every x in M; there are only finite many nonzero
wi ðxÞ’s in the sum i2I wi ðxÞ; i2I wi ¼ 1 is meaningful. Further, if M is a smooth
manifold, and each wi is smooth, then we say that fwi gi2I is a smooth partition of
unity subordinate to fXi gi2I : Now, the Lemma 4.56 can be abbreviated as: “If M is
an n-dimensional smooth manifold, and fXi gi2I is an open cover of M; then there
exists a smooth partition of unity subordinate to fXi gi2I :” This result is known as
the partitions of unity theorem.
Definition Let M be a topological space. Let A be a closed subset of M: Let U be
an open subset of M such that A U: Let w : M ! ½0; 1 be a continuous function.
If
1. for every x in A, wðxÞ ¼ 1;
2. supp w U;
then we say that w is a bump function for A supported in U:
¼ 0 Fp ðtÞ ¼ 0:
Vp \ U
supp wp ’s are outside Gx : Since all but finite many supp wp ’s are outside Gx ; there
P
exists p1 ; p2 ; . . .; pn 2 A such that ð p2A vp ÞjGx ¼ ðvp1 þ vp2 þ þ vpn ÞjGx : Since
vp1 ; vp2 ; . . .; vpn are smooth, and Gx is an open neighborhood of x;
ðvp1 þ vp2 þ þ vpn ÞjGx is smooth on Gx : Since ðvp1 þ vp2 þ þ vpn ÞjGx is
P P
smooth on Gx ; and ð p2A vp ÞjGx ¼ ðvp1 þ vp2 þ þ vpn ÞjGx ; ð p2A vp ÞjGx is
P
smooth on Gx : It follows that y 7! ð p2A vp ÞðyÞ is smooth.
Now, we can define a function ~f : M ! Rk as follows:
X
~f vp :
p2A
Clearly, ~f is smooth.
For 1: Let us fix any x 2 A: We have to show that ~f ðxÞ ¼ f ðxÞ:
Since fy : wðyÞ 6¼ 0g supp w M A; x 2 A fy : wðyÞ ¼ 0g; and hence
wðxÞ ¼ 0: Since supp wp Vp \ U; M ðVp \ UÞ M ðsupp wp Þ: If x 62
Vp \ U; then x 2 M ðVp \ UÞ M ðsupp wp Þ; and hence vp ðxÞ ¼ 0: Also if
x 62 Vp \ U then x 62 supp wp ¼ fy : wp ðyÞ 6¼ 0g fy : wp ðyÞ 6¼ 0g; and hence
wp ðxÞ ¼ 0: Thus, if x 62 Vp \ U; then vp ðxÞ ¼ 0 ¼ wp ðxÞ: Now,
!
X X X X
~f ð xÞ ¼ vp ðxÞ ¼ vp ðxÞ ¼ vp ðxÞ þ vp ð xÞ
p2A fp:p2Ag fp:p2A;x2Vp \ U g fp:p2A;x62Vp \ U g
X X X
¼ vp ð xÞ þ 0¼ vp ð xÞ
fp:p2A;x2Vp \ U g fp:p2A;x62Vp \ U g fp:p2A;x2Vp \ U g
X X
¼ wp Fp
Vp \ U
ð xÞ ¼
Vp \ U
wp Vp \ U
ð xÞ Fp Vp \ U ð xÞ
fp:p2A;x2Vp \ U g fp:p2A;x2Vp \ U g
X X
¼ wp V \U
ð xÞ Fp ðxÞ ¼ wp V \ U ð xÞ ðf ðxÞÞ
p p
fp:p2A;x2Vp \ U g fp:p2A;x2Vp \ U g
0 1 0 1
X X
B C B C
¼@ wp V \ U ð xÞAðf ðxÞÞ ¼ @ wp ð xÞAðf ð xÞÞ
p
fp:p2A;x2Vp \ U g fp:p2A;x2Vp \ U g
0 1
B X X C
¼@ wp ð x Þ þ 0Aðf ðxÞÞ
fp:p2A;x2Vp \ U g fp:p2A;x62Vp \ U g
0 1 0 1
B X X C X
¼@ wp ð x Þ þ wp ðxÞAðf ð xÞÞ ¼ @ wp ðxÞAðf ð xÞÞ
f p:p2A;x2V p \ U g fp:p2A;x6 2Vp \ U g f p:p2A g
00 1 1
X
¼ @@ wp AðxÞAðf ð xÞÞ ¼ ðð1 wÞðxÞÞðf ðxÞÞ ¼ ð1 wð xÞÞðf ðxÞÞ ¼ ð1 0Þðf ðxÞÞ ¼ f ðxÞ:
fp:p2Ag
4.5 Partitions of Unity Theorem 297
Thus, ~f ðxÞ ¼ f ðxÞ: So, for any x 2 A we have ~f ðxÞ ¼ f ðxÞ: Thus, ~f jA ¼ f : This
proves 1.
For 2: Let us take any y 2 M such that ~f ðyÞ 6¼ 0: Here,
!
X X X X
0 6¼ ~f ðyÞ ¼ vp ð yÞ ¼ vp ðyÞ ¼ vp ð yÞ þ vp ð yÞ
p2A fp:p2Ag fp:p2A;y2Vp \ U g fp:p2A;y62Vp \ U g
X X X
¼ vp ð yÞ þ 0¼ vp ðyÞ
fp:p2A;y2Vp \ U g fp:p2A;y62Vp \ U g fp:p2A;y2Vp \ U g
X X
¼ w p Vp \ U Fp ð yÞ ¼
Vp \ U
wp Vp \ U ð yÞ Fp Vp \ U ðyÞ
fp:p2A;y2Vp \ U g fp:p2A;y2Vp \ U g
X
¼ wp ðyÞ Fp ð xÞ :
fp:p2A;y2Vp \ U g
P
Since fp:p2A;y2Vp \ Ug ðwp ðyÞÞðFp ðxÞÞ 6¼ 0; there exists p 2 A such that wp ðyÞ 6¼ 0:
It follows that fy : ~f ðyÞ 6¼ 0g [p2A fy : wp ðyÞ 6¼ 0g; and hence, supp ~f ¼ fy : ~f ðyÞ
6¼ 0g ð[p2A fy : wp ðyÞ 6¼ 0gÞ : Since fsupp wp : p 2 Ag is locally finite, ð[p2A
fy : wp ðyÞ 6¼ 0gÞ ¼ [p2A ðfy : wp ðyÞ 6¼ 0g Þ: It follows that supp ~f [p2A ðfy :
wp ðyÞ 6¼ 0g Þ ¼ [p2A ðsupp wp Þ [p2A ðVp \ UÞ ¼ ð[p2A Vp Þ \ U U: Thus,
supp ~f U: This proves 2. h
uðpÞ 62 oH3 ; uðpÞ 2 G; and G oH3 is open in H3 ; there exists an open neigh-
borhood HuðpÞ of uðpÞ in H3 such that HuðpÞ G oH3 G; and hence
HuðpÞ \ oH3 ¼ ;: Since HuðpÞ \ oH3 ¼ ;; and HuðpÞ is open in H3 ; HuðpÞ is open in
R3 : Since u is a homeomorphism from U onto G; G is open in H3 ; HuðpÞ G; and
HuðpÞ is an open neighborhood of uðpÞ in H3 ; u1 ðHuðpÞ Þ is an open neighborhood
of p in M; and uju1 ðHuðpÞ Þ is a homeomorphism from u1 ðHuðpÞ Þ onto HuðpÞ : Thus,
the ordered pair ðu1 ðHuðpÞ Þ; uju1 ðHuðpÞ Þ Þ is a chart for M: Now, since ðuju1 ðHuðpÞ Þ Þ
ðu1 ðHuðpÞ ÞÞ ¼ HuðpÞ ; and HuðpÞ is open in R3 ; the chart ðu1 ðHuðpÞ Þ; uju1 ðHuðpÞ Þ Þ
is an interior chart for M: Since ðu1 ðHuðpÞ Þ; uju1 ðHuðpÞ Þ Þ is an interior chart for M;
and p 2 u1 ðHuðpÞ Þ; p is an interior point of M; and hence p 2 Int M; a
contradiction.
For 2: Its proof is postponed right now. This result is known as the topological
invariance of the boundary.
Similar results can be supplied for 4-dimensional topological manifolds with
boundary, etc.
Note 4.61 Let M be a topological n-manifold with boundary. We want to show that
Int M is an open subset of M. Here,
For this purpose, let us take any p 2 Int M: We have to find an open neigh-
borhood Up such that if q 2 Up then q 2 Int M: Since p 2 Int M; p 2
M; there exists a chart ðU; uÞ of M such that p 2 U; and uðUÞ is open in Rn :
Since ðU; uÞ is a chart, U is open in M: Since U is open in M; and p 2 U; U is an
open neighborhood of p: Next, let us take any q 2 U: Here, ðU; uÞ is a chart of M
such that q 2 U; and uðUÞ is open in Rn : Hence, by definition of the Int M;
q 2 Int M:
This proves that Int M is an open subset of M:
Note 4.62 Clearly, a topological n-manifold M “in the original sense” is a topo-
logical n-manifold M with boundary.
Next, let us fix any p 2 oM: Since p 2 oM; p is a boundary point of M; and hence,
there exists a boundary chart ðU; uÞ of M such that p 2 U; and uðpÞ 2 oH3 ¼
fðx1 ; x2 ; 0Þ : ðx1 ; x2 ; 0Þ 2 R3 g: Since ðU; uÞ is a boundary chart of M, U is open in M;
uðUÞ is open in H3 ; and u : U ! uðUÞ is a homeomorphism. Since p 2 U; uðpÞ 2
uðUÞ: Since uðpÞ 2 uðUÞ; and uðUÞ is open in H3 ; uðUÞ is an open neighborhood of
uðpÞ in H3 : Since uðUÞ is an open neighborhood of uðpÞ in H3 ; and uðpÞ 2 oH3
H3 ; uðUÞ \ oH3 is an open neighborhood of uðpÞ in oH3 : Since u : U ! uðUÞ is a
homeomorphism, and uðUÞ \ oH3 uðUÞ; the restriction uju1 ðuðUÞ \ oH3 Þ is a
homeomorphism from u1 ðuðUÞ \ oH3 Þ onto uðUÞ \ oH3 : Since uðpÞ 2 uðUÞ; and
uðpÞ 2 oH3 ; uðpÞ 2 uðUÞ \ oH3 ; and hence p 2 u1 ðuðUÞ \ oH3 Þ:
Now, we shall try to prove:
(a) u1 ðuðUÞ \ oH3 Þ oM;
(b) u1 ðuðUÞ \ oH3 Þ is open in oM;
(c) uðUÞ \ oH3 is open in oH3 :
For a: Let us take any a 2 u1 ðuðUÞ \ oH3 Þ: Since a 2 u1 ðuðUÞ \ oH3 Þ; and
u : U ! uðUÞ; a 2 U: Since a 2 u1 ðuðUÞ \ oH3 Þ; uðaÞ 2 uðUÞ \ oH3 ; and
hence, uðaÞ 2 oH3 : Since ðU; uÞ is a boundary chart of M; a 2 U; and uðaÞ 2
oH3 ; a is a boundary point of M; and hence a 2 oM:
For b: First of all, we want to prove that u1 ðuðUÞ \ oH3 Þ ¼ U \ oM: For this
purpose, let us take any a in u1 ðuðUÞ \ oH3 Þ: Since a 2 u1 ðuðUÞ \ oH3 Þ; and
u : U ! uðUÞ; a 2 U: Hence, u1 ðuðUÞ \ oH3 Þ U: Since u1 ðuðUÞ \ oH3 Þ
U; and from (a), u1 ðuðUÞ \ oH3 Þ oM; u1 ðuðUÞ \ oH3 Þ U \ oM:
Next, let us take any b in U \ oM: Since b is inU \ oM; b 2 U; and hence,
uðbÞ 2 uðUÞ: Clearly, uðbÞ 2 oH3 :
(Reason: If not, otherwise, let uðbÞ 62 oH3 : We have to arrive at a contradiction.
Since b is in U \ oM; b is in U: Since b is in U; and u : U ! uðUÞ; uðbÞ 2 uðUÞ:
Since uðbÞ 2 uðUÞ; and uðUÞ is open in H3 ; uðbÞ 2 H3 ¼ Int H3 [ oH3 :Since
uðbÞ 2 Int H3 [ oH3 ;and uðbÞ 62 oH3 ; uðbÞ 2 Int H3 ¼ fðx1 ; x2 ; x3 Þ : ðx1 ; x2 ; x3 Þ
2 R3 and 0\x3 g: Since uðbÞ 2 fðx1 ; x2 ; x3 Þ : ðx1 ; x2 ; x3 Þ 2 R3 and 0\x3 g; uðbÞ 2
uðUÞ; and uðUÞ is open in H3 ; there exists a real number r [ 0 such that the open
ball Br ðuðbÞÞ is contained in uðUÞ \ Int H3 : It follows that Br ðuðbÞÞ is open in
uðUÞ: Since Br ðuðbÞÞ is open in uðUÞ; and u : U ! uðUÞ is a homeomorphism,
u1 ðBr ðuðbÞÞÞ is open in U; and the restriction uju1 ðBr ðuðbÞÞÞ is a homeomorphism
from u1 ðBr ðuðbÞÞÞ onto Br ðuðbÞÞ: Since uðbÞ 2 Br ðuðbÞÞ; b 2 u1 ðBr ðuðbÞÞÞ:
Since u1 ðBr ðuðbÞÞÞ is open in U; and U is open in M; u1 ðBr ðuðbÞÞÞ is open in
M: Since u1 ðBr ðuðbÞÞÞ is open in M; and b 2 u1 ðBr ðuðbÞÞÞ; u1 ðBr ðuðbÞÞÞ is
an open neighborhood of b in M: It follows that the ordered pair
302 4 Topological Properties of Smooth Manifolds
ðu1 ðBr ðuðbÞÞÞ; uju1 ðBr ðuðbÞÞÞ Þ is a chart of M: Also since b 2 u1 ðBr ðuðbÞÞÞ;
ðuju1 ðBr ðuðbÞÞÞ Þðu1 ðBr ðuðbÞÞÞÞ ¼ Br ðuðbÞÞ; and Br ðuðbÞÞ is open in R3 ;
ðu1 ðBr ðuðbÞÞÞ; uju1 ðBr ðuðbÞÞÞ Þ is an interior chart of M; and hence b 2 Int M:
Since b 2 Int M, by the topological invariance of the boundary, b 62 oM; and hence,
b 62 U \ oM; a contradiction.)
Since uðbÞ 2 oH3 ; and uðbÞ 2 uðUÞ; uðbÞ 2 uðUÞ \ oH3 ; and hence, b 2
u ðuðUÞ \ oH3 Þ: Thus, U \ oM u1 ðuðUÞ \ oH3 Þ:
1
Thus, we have shown that u1 ðuðUÞ \ oH3 Þ ¼ U \ oM: Since U is open in M;
and oM M; U \ oMð¼ u1 ðuðUÞ \ oH3 ÞÞ is open in oM; and hence,
u1 ðuðUÞ \ oH3 Þ is open in oM: This proves (b).
For c: Since uðUÞ is open in H3 ; and oH3 H3 ; uðUÞ \ oH3 is open in oH3 :
This proves (c).
Since the mapping p3 : oH3 ! R2 defined by p3 ðx1 ; x2 ; 0Þ ðx1 ; x2 Þ is a
homeomorphism, and uju1 ðuðUÞ \ oH3 Þ is a homeomorphism from u1 ðuðUÞ \ oH3 Þ
onto uðUÞ \ oH3 ; their composite p3 ðuju1 ðuðUÞ \ oH3 Þ Þ is a homeomorphism from
u1 ðuðUÞ \ oH3 Þ onto p3 ðuðUÞ \ oH3 Þ: Since uðUÞ \ oH3 is open in oH3 ; and
p3 : oH3 ! R2 is a homeomorphism, p3 ðuðUÞ \ oH3 Þ is open in R2 : Here, p3
ðuju1 ðuðUÞ \ oH3 Þ Þ is a homeomorphism from u1 ðuðUÞ \ oH3 Þ onto p3 ðuðUÞ
\oH3 Þ; u1 ðuðUÞ \ oH3 Þ is open in oM; p 2 u1 ðuðUÞ \ oH3 Þ; and p3 ðuðUÞ \
oH3 Þ is open in R2 ; the ordered pair ðu1 ðuðUÞ \ oH3 Þ;Þp3 ðuju1 ðuðUÞ \ oH3 Þ Þ is
an internal chart of oM; where p 2 u1 ðuðUÞ \ oH3 Þ: Hence, oM is a topological
2-manifold without boundary.
This completes the proof of 2.
Similar results can be supplied for 4-dimensional topological manifolds with
boundary, etc.
Definition Let A be a nonempty subset of Rn : Let f : A ! Rm : By f is smooth we
mean: For every x in A; there exist an open neighborhood Vx of x in Rn ; and a
function fx : Vx ! Rm such that fx jA \ Vx ¼ f jA \ Vx ; and fx is smooth.
Definition Let G be a nonempty subset of Hn : Let G be open in Hn : Let f : G !
Rm : By f is smooth we mean: For every x in G; there exist an open neighborhood Vx
of x in Rn ; and a function fx : Vx ! Rm such that fx jG \ Vx ¼ f jG \ Vx ; and fx is smooth
in the usual sense.
Note 4.65 Let G be a nonempty subset of Hn : Let G be open in Hn : Let f : G !
Rm : Let f be smooth.
1. p maps M e onto M;
2. for every p in M; there exists a connected open neighborhood U of p such that
for each component C of the open set p1 ðUÞ; the restriction pjC : C ! U is a
homeomorphism,
Clearly, R3a becomes a real linear space. Here, the zero vector is ð0; 0; 0Þa ; and the
negative vector of va is ðvÞa : It is easy to see that the mapping v 7! va is an
isomorphism from R3 onto R3a ; and hence, R3 and R3a are isomorphic.
Note 5.1 Since members of a smooth manifold M may not be added, the above
definition of tangent space cannot be generalized. Observe that C1 ðR3 Þ is a real
linear space under pointwise vector addition and scalar multiplication. Also, cor-
responding to each geometric tangent vector va in R3 at a, we can define a function
Dv ja : C1 ðR3 Þ ! R as follows: For every f 2 C1 ðR3 Þ;
d f ða þ tvÞ f ðaÞ
Dv ja ðf Þ ðDv Þðf ðaÞÞ ¼ ðf ða þ tvÞÞ ¼ lim :
dt t¼0
t!0 t
Also,
Dðv1 ;v2 ;v3 Þ a ðf Þ ¼ v1 Dð1;0;0Þ a ðf Þ þ v2 Dð0;1;0Þ a ðf Þ þ v3 Dð0;0;1Þ a ðf Þ
f ða þ tð1; 0; 0ÞÞ f ðaÞ f ða þ tð0; 1; 0ÞÞ f ðaÞ
¼ v1 lim þ v2 lim
t!0 t t!0 t
f ð a þ t ð 0; 0; 1 Þ Þ f ðaÞ
þ v3 lim
t!0 t
¼ v ððD1 f ÞðaÞÞ þ v ððD2 f ÞðaÞÞ
1 2
of 2 of 3 of
þ v3 ððD3 f ÞðaÞÞ ¼ v1 ðaÞ þ v ðaÞ þ v ðaÞ
ox1 ox2 ox3
X
3
of of
¼ vi ðaÞ ¼ vi ðaÞ :
i¼1
ox i ox i
Thus,
i of
Dðv1 ;v2 ;v3 Þ a ðf Þ ¼ v ðaÞ :
oxi
It follows that
of of
Dð1;0;0Þ a ðf Þ ¼ 1 ðaÞ; Dð0;1;0Þ a ðf Þ ¼ 2 ðaÞ;
ox ox
of
and
Dð0;0;1Þ a ðf Þ ¼ 3 ðaÞ:
ox
for every f, g in C 1 ðR3 Þ; then we say that w is a derivation at a. Let us denote the
collection of all derivations at a by Ta R3 : Clearly, Ta R3 is a real linear space under
pointwise vector addition and scalar multiplication. It is clear that if va 2 R3a ; then
the mapping Dv ja : C1 ðR3 Þ ! R; defined as above, is a derivation at a. Thus, for
every va 2 R3a ; Dv ja 2 Ta R3 :
5.1 Pointwise Pushforward 309
gðva Þ Dv ja :
Then, g is a linear isomorphism from real linear space R3a onto real linear space
Ta R3 :
Reason: Let va ; wa 2 R3a ; where v ðv1 ; v2 ; v3 Þ; and w ðw1 ; w2 ; w3 Þ: Let s; t 2 R:
We have to prove that gðsðva Þ þ tðwa ÞÞ ¼ sðgðva ÞÞ þ tðgðwa ÞÞ; that is, for every f in
C 1 ðR3 Þ; ðgðsðva Þ þ tðwa ÞÞÞ ðf Þ ¼ ðsðgðva ÞÞ þ tðgðwa ÞÞðf Þ.
LHS ¼ ðgðsðva Þ þ tðwa ÞÞÞðf Þ ¼ g ðsvÞa þðtwÞa ðf Þ
¼ g ðsv þ twÞa ðf Þ ¼ DðsvþtwÞ a ðf Þ
of
¼ Dðsv1 þtw1 ;sv2 þtw2 ;sv3 þtw3 Þ a ðf Þ ¼ svi þ twi ðaÞ
oxi
of of
¼ s vi ðaÞ þ t wi ðaÞ ¼ s Dv ja ðf Þ þ t Dw ja ðf Þ
ox i ox i
and p1 2 C 1 ðR3 Þ;
310 5 Immersions, Submersions, and Embeddings
of 2 of 3 of
v1 ¼ v1 ð1Þ þ v2 ð0Þ þ v3 ð0Þ ¼ v1 ðaÞ þ v ðaÞ þ v ðaÞ
ox1 ox2 ox3
of
¼ Dðv1 ;v2 ;v3 Þ a ðp1 Þ ¼ Dðw1 ;w2 ;w3 Þ a ðp1 Þ ¼ w1 ðaÞ
ox1
of of
þ w2 ðaÞ þ w3 ðaÞ
ox2 ox3
¼ w1 ð1Þ þ w2 ð0Þ þ w3 ð0Þ ¼ w1 ;
RHS ¼ wðf Þ ¼ w f ðaÞ þ g1 p1 a1 þ g2 p2 a2 þ g3 p3 a3
¼ wðf ðaÞÞ þ ðwðg1 ÞÞ p1 a1 ðaÞ þ w p1 a1 ðg1 ðaÞÞ
þ ðwðg2 ÞÞ p2 a2 ðaÞ þ w p2 a2 ðg2 ðaÞÞ
þ ðwðg3 ÞÞ p3 a3 ðaÞ þ w p3 a3 ðg1 ðaÞÞ
¼ 0 þ ðwðg1 ÞÞ a1 a1 þ ðwðp1 Þ 0Þðg1 ðaÞÞ
þ ðwðg2 ÞÞ a2 a2 þ ðwðp2 Þ 0Þðg2 ðaÞÞ
þ ðwðg3 ÞÞ a3 a3 þ ðwðp3 Þ 0Þðg3 ðaÞÞ
¼ ðwðp1 ÞÞðg1 ðaÞÞ þ ðwðp2 ÞÞðg2 ðaÞÞ þ ðwðp3 ÞÞðg3 ðaÞÞ
of
¼ ðwðpi ÞÞðgi ðaÞÞ ¼ ðwðpi ÞÞ ðaÞ
oxi
¼ LHS:
5.1 Pointwise Pushforward 311
Thus, we have shown that g is a linear isomorphism from real linear space R3a onto
real linear space Ta R3 : Hence, R3a is isomorphic to Ta R3 :
Since R3a is a linear space of dimension 3, and R3a is isomorphic to Ta R3 ; Ta R3 is
of dimension 3. Further, since fð1; 0; 0Þa ; ð0; 1; 0Þa ; ð0; 0; 1Þa g is a basis of R3a ; and
g is a linear isomorphism from R3a onto Ta R3 ; fgðð1; 0; 0Þa Þ; gðð0; 1; 0Þa Þ;
gðð0; 0; 1Þa Þg is a basis of Ta R3 ; that is,
n o
Dð1;0;0Þ a ; Dð0;1;0Þ a ; Dð0;0;1Þ a
is a basis of Ta R3 : Similarly,
o o o
; ; . . .; n
ox1 a ox2 a ox a
is a basis of Ta Rn :
Note 5.4 Since R3a is the geometric tangent space to R3 at a, and R3a is isomorphic
to Ta R3 ; Ta R3 can be thought of as the geometric tangent space to R3 at a. Simi-
larly, for every a in Rk ; Rka is isomorphic to Ta Rk . The interesting thing is that
Ta R3 ; but not R3a ; can be generalized on a smooth manifold.
Definition Let M be an m-dimensional smooth manifold. Let p 2 M: Let v :
C 1 ðMÞ ! R be a linear function. If v satisfies the Leibnitz rule:
for every f, g in C 1 ðMÞ; then we say that v is a derivation at p. The collection of all
derivations at p is denoted by Tp M: Clearly, Tp M is a real linear space under
pointwise vector addition and scalar multiplication. Here, the real linear space Tp M
is called the tangent space to M at p and the members of Tp M are called tangent
vectors at p.
Note 5.5 Let M be an m-dimensional smooth manifold. Let p 2 M: Let v 2 Tp M:
Let f ; g 2 C1 ðMÞ: As above, it is easy to show:
1. If f is a constant function, then vðf Þ ¼ 0:
2. If f ðpÞ ¼ gðpÞ ¼ 0, then vðf gÞ ¼ 0:
We shall try to show that ððdFp ÞðvÞÞ 2 TFðpÞ N; that is, ðdFp ÞðvÞ is a derivation at
FðpÞ: Observe that ðdFp ÞðvÞ : C1 ðNÞ ! R is linear.
(Reason: Let f ; g 2 C 1 ðNÞ; and let s, t be any real. We must prove:
dFp ðvÞ ðsf þ tgÞ ¼ s dFp ðvÞ ð f Þ þ t dFp ðvÞ ðgÞ :
LHS ¼ ððdFp ÞðvÞÞðsf þ tgÞ ¼ vððsf þ tgÞ FÞ ¼ vðsðf FÞ þ tðg FÞÞ ¼ sðvðf FÞÞþ
tðvðg FÞÞ ¼ sðððdFp ÞðvÞÞðf ÞÞ þ tðððdFp ÞðvÞÞðgÞÞ ¼ RHS:Þ
Now, we shall try to prove that ðdFp ÞðvÞ : C 1 ðNÞ ! R satisfies the Leibnitz
rule:
dFp ðvÞ ðf gÞ ¼ dFp ðvÞ ðf Þ ðgðFðpÞÞÞ þ ðf ðFðpÞÞÞ dFp ðvÞ ðgÞ
dFp ðsv1 þ tv2 Þ ðf Þ ¼ s dFp ðv1 Þ þ t dFp ðv2 Þ ðf Þ:
LHS ¼ dFp ðsv1 þ tv2 Þ ðf Þ ¼ ðsv1 þ tv2 Þðf F Þ
¼ s ð v1 ð f F Þ Þ þ t ð v2 ð f F Þ Þ
¼ s dFp ðv1 Þ ðf Þ þ t dFp ðv2 Þ ðf Þ
¼ s dFp ðv1 Þ ðf Þ þ t dFp ðv2 Þ ðf Þ
¼ s dFp ðv1 Þ þ t dFp ðv2 Þ ðf Þ ¼ RHS:
1
d F 1 FðpÞ : TFðpÞ N ! Tp M; and dFp : TFðpÞ N ! Tp M:
Now, we want to prove that dðF 1 ÞFðpÞ ¼ ðdFp Þ1 ; that is, ðdðF 1 ÞFðpÞ Þ
ðdFp Þ ¼ IdTp M :
LHS ¼ d F 1 FðpÞ dFp ¼ d F 1 F p ¼ dðIdM Þp ¼ IdTp M ¼ RHS:
dim Rm uðpÞ : Since dim Tp M ¼ dim RuðpÞ ; and dim RuðpÞ ¼ m; dim Tp M ¼ m:
m m
Since v 7! ðdðp1 Þðp;qÞ ÞðvÞ is linear, and v 7! ðdðp2 Þðp;qÞ ÞðvÞ is linear,
v 7! ððdðp1 Þðp;qÞ ÞðvÞ; ðdðp2 Þðp;qÞ ÞðvÞÞ is linear, and hence, a : Tðp;qÞ ðM NÞ !
Tp M Tp N is linear.
Similarly, we can show that let M be an m-dimensional smooth manifold, N be
an n-dimensional smooth manifold, and ðp; qÞ 2 M N: Define a : Tðp;qÞ
ðM NÞ ! Tp M Tq N as follows: For every v in Tðp;qÞ ðM NÞ; aðvÞ
ððdðp1 Þðp;qÞ ÞðvÞ; ðdðp2 Þðp;qÞ ÞðvÞÞ: Then, α is linear, etc.
is a basis of TuðpÞ Rm ;
320 5 Immersions, Submersions, and Embeddings
( !
1 o 1
dip d u1 uðpÞ d^iuðpÞ ; dip d u1 uðpÞ d^iuðpÞ
ox uðpÞ
1
! !)
o 1 o
1
; . . .; dip d u uðpÞ d^iuðpÞ
ox2 uðpÞ oxm uðpÞ
!
1
1 o
dip d u d^iuðpÞ
uðpÞ ox1 uðpÞ
!!
1
1 o
¼ dip d u d^iuðpÞ
uðpÞ ox1 uðpÞ
! !!
o 1 o
¼ dip d u 1
¼ dip d u uðpÞ
uðpÞ ox1 uðpÞ ox1 uðpÞ
!
o
¼ d u1 uðpÞ :
ox1 uðpÞ
Thus,
! !
1 o 1 o
dip d u1 uðpÞ d^iuðpÞ ¼ d u uðpÞ :
ox1 uðpÞ ox1 uðpÞ
Similarly,
! !
1 o 1 o
dip d u 1
d^iuðpÞ ¼ d u uðpÞ ; etc;
uðpÞ ox2 uðpÞ ox2 uðpÞ
Thus,
( ! ! !)
o 1 o o
d u1 uðpÞ ; d u uðpÞ 1
; . . .; d u uðpÞ
ox1 uðpÞ ox2 uðpÞ oxm uðpÞ
!
o
d u1 uðpÞ
oxi uðpÞ
5.2 Open Submanifolds 321
o
at p is denoted by oxi p . Thus,
( )
o o o
; ; . . .; m
ox1 p ox2 p ox p
is a basis of Tp M:
Conclusion: Let M be an m-dimensional smooth manifold. Let ðU; uÞ be an
admissible coordinate chart of M. Then, for every p 2 U;
( )
o o o
; ; . . .; m
ox1 p ox2 p ox p
o
is a basis of Tp M; where oxi p stands for
!
o
d u1 uðpÞ :
oxi uðpÞ
Here, we say that ðv1 ; v2 ; . . .; vm Þ are the components of v with respect to coordinate
basis.
Note 5.18 Let U be an open subset of Rn ; V be an open subset of Rm ; and p 2 U:
Let F : U ! V be smooth, where F ðF 1 ; F 2 ; . . .; F m Þ:
Since U is a nonempty open subset of the n-dimensional smooth manifold Rn ;
U is also an n-dimensional smooth manifold. Similarly, V is an m-dimensional
smooth manifold. It follows that dFp : Tp U ! TFðpÞ V is a linear map from real
linear space Tp U to real linear space TFðpÞ V: Here,
322 5 Immersions, Submersions, and Embeddings
!
o o o
; ; . . .; n
ox1 p ox2 p ox p
is the coordinate basis of TFðpÞ V: Now, since dFp : Tp U ! TFðpÞ V; for every
f 2 C 1 ðVÞ;
!! !
o o
dFp ðf Þ ¼ ðf F Þ
ox1 p ox1 p
¼ 1st column of ½ðD1 f ÞðFðpÞÞ
2 3
ðD1 F 1 ÞðpÞ ðDn F 1 ÞðpÞ
6. .. . 7
ðDm f ÞðFðpÞÞ 6
4 .. . .. 7
5
ðD1 F ÞðpÞ ðDn F ÞðpÞ
m m
¼ ððD1 f ÞðFðpÞÞÞ D1 F 1 ðpÞ
þ þ ððDm f ÞðFðpÞÞÞððD1 F m ÞðpÞÞ
Xm i
of oF
¼ ðFðpÞÞ ðpÞ
i¼1
oy i ox1
X
m
oF i of
¼ ðpÞ ðFðpÞÞ
i¼1
ox1 oyi
X i ! !!
m
oF o
¼ ðpÞ f
i¼1
ox1 oyi FðpÞ
X i !! !
m
oF o
¼ ðpÞ f
i¼1
ox1 oyi FðpÞ
! !
X m
oF i o
¼ ðpÞ f
i¼1
ox1 oyi FðpÞ
5.2 Open Submanifolds 323
and hence,
! X m !
o oF i o
dFp ¼ ðpÞ :
ox1 p i¼1
ox1 oyi FðpÞ
Similarly,
! m !
o X oF i o
dFp ¼ ðpÞ ; etc:
ox2 p i¼1
ox2 oyi FðpÞ
and
!
o o o
; ; . . .; m
oy1 FðpÞ oy2 FðpÞ oy FðpÞ
is the coordinate basis of TFðpÞ V; the matrix representation of linear map dFp is the
following m n matrix:
2 3
oF 1 oF 1 oF 1
ox1 ðpÞ ox2 ðpÞ oxn ðpÞ
6 oF 2 7
6 1 ðpÞ oF 2
ðpÞ oF 2
ðpÞ 7
6 ox ox2 oxn 7:
4 5
oF m oF m
oF
m
ox1 ðpÞ ox2 ðpÞ oxn ðpÞ
Let us recall that this matrix is the matrix representation of the total derivative
DFðpÞ : Rn ! Rm : Thus, we find that in this case, differential dFp : Tp Rn !
TFðpÞ Rm corresponds to the total derivative DFðpÞ : Rn ! Rm ; tangent space Tp Rn
corresponds Rn ; and tangent space TFðpÞ Rm corresponds Rm :
Note 5.19 Let M be an m-dimensional smooth manifold, and N be an n-dimen-
sional smooth manifold. Let p 2 M: Let F : M ! N be a smooth mapping. Let
ðU; uÞ be an admissible coordinate chart of M such that p 2 U: Let ðV; wÞ be an
admissible coordinate chart of N such that FðpÞ 2 V: Since F : M ! N is a smooth
mapping, ðw F u1 Þ : uðU \ F 1 ðVÞÞ ! wðVÞ is smooth. Put ðF ^ 1; F
^ 2 ; . . .;
^ ÞF
F n ^ w F u : Thus, F
1 ^ : uðU \ F ðVÞÞ ! wðVÞ is smooth, that is, each
1
1
pi ðw F u Þ is smooth, where pi denotes the ith projection function from Rn
to R:
324 5 Immersions, Submersions, and Embeddings
2 oF^ 1 ^1
oF ^1
oF
3
ox1 ðuðpÞÞ ox2 ðuðpÞÞ oxm ðuðpÞÞ
6 ^2 7
6 oF1 ðuðpÞÞ
^2
oF
ðuðpÞÞ
^2
oF
ðuðpÞÞ 7
6 ox ox2 oxm 7:
4 5
^n
oF ^n
oF ^n
ox1 ðuðpÞÞ ox2 ðuðpÞÞ ooxFm ðuðpÞÞ
!
o o o
; ; . . .; m
ox1 p ox2 p ox p
Also,
!
o o o
; ; . . .; n
oy1 FðpÞ oy2 FðpÞ oy FðpÞ
Thus,
! n ^j !
o X oF o
dFp ¼ ðuðpÞÞ :
oxi p j¼1
oxi oy j FðpÞ
and
!
o o o
; ; . . .; n
oy1 FðpÞ oy2 FðpÞ oy FðpÞ
is the coordinate basis of TFðpÞ N; the matrix representation of linear map dFp is the
following n m matrix:
2 3
^1
oF ^1
oF ^1
oF
ox1 ðuðpÞÞ ox2 ðuðpÞÞ oxm ðuðpÞÞ
6 7
6 oF^ 2 ^2 ^2 7
6 ox1 ðuðpÞÞ oF
ðuðpÞÞ ooxFm ðuðpÞÞ 7
6 ox2 7
4 5
oF^n ^n
oF ^n
ox1 ðuðpÞÞ ox2 ðuðpÞÞ ooxFm ðuðpÞÞ
2 3
oðp1 ðwFu1 ÞÞ oðp1 ðwFu1 ÞÞ oðp1 ðwFu1 ÞÞ
ðuðpÞÞ ðuðpÞÞ ðuðpÞÞ
6 ox1 ox2 oxm 7
6 7
6 o p wFu1 ÞÞ 7
¼ 6 ð 2 ð ox1
oðp2 ðwFu1 ÞÞ oðp2 ðwFu1 ÞÞ
ðuðpÞÞ ðuðpÞÞ ð uðpÞ Þ 7:
6 ox2 oxm 7
4 5
oðpn ðwFu1 ÞÞ oðpn ðwFu1 ÞÞ oðpn ðwFu1 ÞÞ
ox 1 ð uðpÞ Þ ox2 ðuðpÞÞ oxm ðuðpÞÞ
It follows that the matrix representation of dFp is the same as the matrix repre-
sentation of dðw F u1 ÞuðpÞ :
Let
!
o o o
; ; . . .; m
o~x1 p o~x2 p o~x p
of Tp M: Let ð~v1 ; ~v2 ; . . .; ~vm Þ be the components of v with respect to coordinate basis
!
o o o
; ; . . .; m
o~x1 p o~x2 p o~x p
of Tp M: Thus,
0 1 0 1
!
1@ o A 2@ o A o
~v
1
þ ~v
2
þ þ ~vm
m ¼v
ox p ox p ox p
! ! !
o o o
¼ v1 þ v2 þ þ vm :
ox1 p
ox2 p oxm p
Since ðU; uÞ and ðV; wÞ are admissible coordinate charts of M such that p 2 U \ V;
w u1 : uðU \ VÞ ! wðU \ VÞ is smooth.
Here, with the hope that there will be no confusion, for every i ¼ 1; 2; . . .; m; the
ith component of the “transition map” w u1 is generally denoted by ~xi : Thus, ~xi
has two meanings: the ith component of wðpÞ; and the ith component function of
w u1 : The situation indicates which meaning is to be attached with. Here, for
every x 2 uðU \ VÞ; ðw u1 ÞðxÞ ¼ ð~x1 ðxÞ; ~x2 ðxÞ; . . .; ~xm ðxÞÞ:
328 5 Immersions, Submersions, and Embeddings
Thus, for j ¼ 1; 2; . . .; m;
m i !
o X o~x o
¼ ðuðpÞÞ :
ox j p i¼1 ox j o~xi p
Now,
! ! ! !
X
m
o o o o
~v
i
¼ ~v1
þ ~v 2
þ þ ~v m
¼v
i¼1
o~xi p o~x1 p o~x2 p o~xm p
! ! !
o o o
¼v 1
þv 2
þ þ v m
ox1 p ox2 p oxm p
! !!
X X X m i
m
o m
o~x o
¼ v j
¼ v j
ðuðpÞÞ
j¼1
ox j p j¼1 i¼1
ox j o~xi p
m i !!!
X m X o~x o
¼ ðuðpÞÞ v j
j¼1 i¼1
ox j o~xi p
m i !!!
X m X o~x o
¼ ð uðpÞ Þ v j
i¼1 j¼1
ox j o~xi p
X X m i ! !!
m
o~x o
¼ ðuðpÞÞ v j
i¼1 j¼1
ox j o~xi p
X X m i ! !
m
o~x o
¼ ðuðpÞÞ v j
;
i¼1 j¼1
ox j o~xi p
so
X ! X Xm i ! !
m
o m
o~x o
~v
i
¼ ðuðpÞÞ v j
;
i¼1
o~xi p i¼1 j¼1
ox j o~xi p
and hence
m i
X
o~x
~vi ¼ ð uðpÞ Þ v j:
j¼1
ox j
In short,
o~xi
~v ¼
i
ðuðpÞÞ v j :
ox j
330 5 Immersions, Submersions, and Embeddings
Lemma 5.21 Let M be a nonempty set. Let fUi gi2I be a nonempty family of subsets
of M. Let fui gi2I be a family of maps ui from Ui to Rm : If
1. for each i 2 I; ui is a 1–1 mapping from Ui onto an open subset ui ðUi Þ of Rm ;
2. for each i; j 2 I; ui ðUi \ Uj Þ; and uj ðUi \ Uj Þ are open in Rm ;
3. if Ui \ Uj 6¼ ;; then the map uj u1 i : ui ðUi \ Uj Þ ! uj ðUi \ Uj Þ is smooth,
4. there exists a countable subcollection of fUi gi2I that covers M;
5. if p; q are distinct points of M; then either ðthere exists i 2 I such that p; q 2
Ui Þ; or ðthere exist i; j 2 I such that p 2 Ui ; q 2 Uj ; and Ui \ Uj ¼ ;Þ; then
there exists a unique smooth differential structure A over M such that for each
i 2 I; ðUi ; ui Þ 2 A:
Proof First of all, we shall try to show that the collection fu1 k ðGÞ : k 2 I;
and G is open in Rm g is closed with respect to finite intersection.
For this purpose, let us take any u1 1 1
i ðG1 Þ; uj ðG2 Þ 2 fuk ðGÞ : k 2 I;
and G is open in R g; where i; j 2 I; G1 ; G2 are open in R :
m m
1
fuk ðGÞ : k 2 I; and G is open in R g: m
ðui u1 1
j ÞðG2 Þ is open in R : Since G1 \ ðui uj ÞðG2 Þ is open in R ;
m m
u1 1 1
i ðG1 \ ðui uj ÞðG2 ÞÞ 2 fuk ðGÞ : k 2 I; and G is open in R g: Since fuk
m 1
1 1 1 1 1 1
ððui uj ÞðG2 ÞÞ ¼ ui ðG1 Þ \ uj ðG2 Þ; ui ðG1 Þ \ uj ðG2 Þ 2 fuk ðGÞ : k 2 I;
and G is open in Rm g:
5.3 Tangent Bundles 331
with respect to finite intersection. Let O be the collection of arbitrary unions of the
members of fu1 i ðGÞ : i 2 I; and G is open in R g:
m
; 2 O:
(b) Since for each i 2 I; ui maps from Ui to Rm ; u1
i ðR Þ ¼ Ui : Now, since R is
m m
1 1
open in R ; fUi : i 2 Ig ¼ fui ðR Þ : i 2 Ig fui ðGÞ : i 2 I; and G
m m
Here, ð[fu1 1
i ðGi1 Þ : i 2 I; and Gi1 is open in R gÞ \ ð[fui ðGi2 Þ : i 2 I; and Gi2
m
m 1 1
is open in R gÞ ¼ [fuj ðGj1 Þ \ uk ðGk2 Þ : ðj; kÞ 2 I I; Gj1 ; Gk2 are open in Rm g:
Since fu1 i ðGÞ : i 2 I; and G is open in R g is closed with respect to finite
m
intersection, fuj ðGj1 Þ \ uk ðGk2 Þ : ðj; kÞ 2 I I; Gj1 ; Gk2 are open in Rm g fu1
1 1
i
ðGÞ : i 2 I; and G is open in Rm g; and hence, [fu1 j ðG j1 Þ \ u 1
k ðG k2 Þ : ðj; kÞ 2
I I; Gj1 ; Gk2 are open in Rm g 2 O: It follows that ð[fu1 i ðG i1 Þ : i 2 I;
1
and Gi1 is open in R gÞ \ ð[fui ðGi2 Þ : i 2 I; and Gi2 is open in R gÞ 2 O:
m m
logical space ðM; OÞ. Now, we want to prove that for each i 2 I; Ui 2 O: For this
purpose, let us fix any i0 2 I: We have to prove that Ui0 2 O: Clearly, fu1 i ðGÞ :
i 2 I; and G is open in Rm g O; and Ui0 ¼ u1 i0 ðR m
Þ 2 fu 1
i ðGÞ : i 2 I; and G
is open in R g; so Ui0 2 O:
m
fu1
j ðGÞ \ Ui : j 2 I; and G is open in R g is a basis of Ui : Now, it suffices to
m
ðui u1 1
j ÞðGÞ \ ui ðUi Þ is open in ui ðUi Þ: Since ðui uj ÞðGÞ \ ui ðUi Þ ð¼ ui
1 1
ðuj ðGÞ \ Ui ÞÞ is open in ui ðUi Þ; ui ðuj ðGÞ \Ui Þ is open in ui ðUi Þ:
! !!
o o
~ p; v
u 1
þ þ v m
ðp1 uÞðpÞ; . . .; ðpm uÞðpÞ; v1 ; . . .; vm :
ox p
1 ox p
m
~ is well defined.
First of all, observe that u
(Reason:Let ðp; v1 ðoxo1 p Þ þ þ vm ðoxom p ÞÞ ¼ ðq; w1 ðoxo1 q Þ þ þ wm ðoxom q ÞÞ where
‘
ðp; v1 ð o1 Þ þ þ vm ð om ÞÞ 2
ox p ox p
Tr M;
r2U
and
! !!
o o a
q; w1 þ þ w m 2 Tr M:
ox1 q
oxm q r2U
Since
! !! ! !!
o o o o
p; v 1
þ þ v m
¼ q; w 1
þ þ wm
;
ox1 p oxm p ox1 q oxm q
p ¼ q and
! ! ! !
o o o o
v 1
þ þ v m
¼w 1
þ þ w m
:
ox1 p oxm p ox1 q oxm q
5.3 Tangent Bundles 335
Since
! ! ! !
o o o o
v 1
þ þ v m
¼w 1
þ þ w m
;
ox1 p oxm p ox1 p oxm p
and
!
o o o
; ; . . .; m
ox1 p ox2 p ox p
! !! ! !!
o o o o
~ p; v
u 1
þ þ vm
~ q; w
¼u 1
þ þ wm
:
ox1 p oxm p ox1 q oxm q
Here,
! !!
o o
ðp1 uÞðpÞ; . . .; ðpm uÞðpÞ; v ; . . .; v
1 m
~ p; v
¼u 1
þ þ v m
ox1 p oxm p
! !!
o o
~ q; w1
¼u þ þ wm
ox1 q oxm q
¼ ðp1 uÞðqÞ; . . .; ðpm uÞðqÞ; w1 ; . . .; wm ;
336 5 Immersions, Submersions, and Embeddings
and hence, for every i ¼ 1; . . .; m; ðpi uÞðpÞ ¼ ðpi uÞðqÞ; and vi ¼ wi : Since for
every i ¼ 1; . . .; m; ðpi uÞðpÞ ¼ ðpi uÞðqÞ; uðpÞ ¼ uðqÞ: Since uðpÞ ¼ uðqÞ;
and u is 1–1, p ¼ q: Since for every i ¼ 1; . . .; m; vi ¼ wi ;
! ! ! !
o o o o
v 1
þ þ v m
¼w 1
þ þ w m
:
ox1 p oxm p ox1 p oxm p
Since
! ! ! !
o o o o
v 1
þ þ v m
¼w1
þ þ w m
; and p ¼ q;
ox1 p oxm p ox1 p oxm p
! ! ! !
o o o o
so v1 þ þ v m
¼w1
þ þ w m
:
ox1 p oxm p ox1 q oxm q
It follows that
! !! ! !!
o o
o o
p; v1 þ þ v m
¼ q; w1 þ þ wm
:
ox1 p oxm p
ox1 q oxm q
‘
Thus, we have shown that u ~ : r2U Tr M ! uðUÞ Rm is 1–1. Now, we want to
‘
show that u~ : r2U Tr M ! uðUÞ Rm is onto. For this purpose, let us take any
ððp1 uÞðpÞ; . . .; ðpm uÞðpÞ; v1 ; . . .; vm Þ 2 uðUÞ Rm ; where p 2 U: Here, p 2 U;
and
! !
o o
v 1
þ þ v m
2 Tp M;
ox1 p oxm p
so
! !
o o a
v 1
þ þ v m
2 Tr M:
ox p
1 ox p
m
r2U
Also,
! !!
o o
~ v1
u þ þ vm ¼ ðp1 uÞðpÞ; . . .; ðpm uÞðpÞ; v1 ; . . .; vm :
ox p
1 ox p
m
‘
This shows that u~ : r2U Tr M ! uðUÞ Rm is a 1–1 onto mapping.
Now, we want to show that uðUÞ Rm is open in R2m : Since ðU; uÞ 2 A; and
A is a smooth structure of M; uðUÞ is open in Rm ; and hence, uðUÞ Rm is open
in Rm Rm ð¼ R2m Þ:
5.3 Tangent Bundles 337
! !! a
o o
p; v1
þ þ v m
2 Tr M;
ox1 p oxm p r2U
! !!
o o
~ p; v1
u þ þ vm
ðp1 uÞðpÞ; . . .; ðpm uÞðpÞ; v1 ; . . .; vm
ox1 p oxm p
‘ ‘
It follows that u~ ð r2U \ V Tr MÞ ¼ uðU \ VÞ Rm ; and hence, u~ ðð r2U Tr MÞ
‘
\ð r2V Tr MÞÞ ¼ uðU \ VÞ Rm : Since ðU; uÞ 2 A; and ðV; wÞ 2 A; uðU \ VÞ
is open in Rm ; and hence, uðU \ VÞ Rm is open in R2m : This shows that
‘ ‘ ~ ‘
~ ðð r2U Tr MÞ \ð r2V Tr MÞÞ is open in R2m : Similarly, wðð
u r2U Tr MÞ\
‘
ð r2V Tr MÞÞ is open in R :2m
‘ ‘
3. Let us take any ðU; uÞ 2 A; and ðV; wÞ 2 A: Let ð r2U Tr MÞ \ ð r2V Tr MÞ 6¼
;: We have to prove that
! !! ! !!
a a a a
~ u
w ~ 1
~
:u Tr M \ Tr M ~
!w Tr M \ Tr M
r2U r2V r2U r2V
and let !
o o o
; ; . . .; m
o~x1 p o~x2 p o~x p
Also,
i !
o o~x o
¼ ðuðpÞÞ :
ox j p ox j o~xi p
‘
~ð
Here, for every ðx1 ; . . .; xm ; v1 ; . . .; vm Þ 2 u r2U\V Tr MÞ; we have
1 1
~ u
w ~ 1 x1 ; . . .; xm ; v1 ; . . .; vm ¼ w ~ u ~ x ; . . .; xm ; v1 ; . . .; vm
!
1 1 o
¼w ~ u x ; . . .; x ; v
1 m
ox1 u1 ðx1 ;...;xm Þ
!!
o
þ þ vm
oxm u1 ðx1 ;...;xm Þ
!!
¼w ~ u1 x1 ; . . .; xm ; vi o
oxi u1 ðx1 ;...;xm Þ
j !!!
~
1 1
i o~x 1 o
¼ w u x ; . . .; x ; v m
x ; . . .; x m
oxi o~x j u1 ðx1 ;...;xm Þ
j !!
~
1 1
i o~ x 1 o
¼ w u x ; . . .; x ; v m
x ; . . .; x m
oxi o~x j u1 ðx1 ;...;xm Þ
1 1 1 1
¼ ðp1 wÞ u x ; ; x ; . . .; ðpm wÞ u x ; . . .; xm ;
m
1 m
o~x 1 i o~ x 1
vi x ; . . .; x m
; . . .; v x ; . . .; x m
oxi oxi
1
1
¼ p1 w u x ; . . .; x ; . . .; pm w u1 x1 ; . . .; xm ;
m
1 m
o~x 1 o~x 1
vi x ; . . .; xm ; . . .; vi x ; . . .; xm
oxi oxi
¼ p1 w u1 x1 ; . . .; xm ; . . .; pm w u1 x1 ; . . .; xm ;
! !
oðw u1 Þ 1 1 m
1
i oð w u Þ
vi x ; . . .; x m
; . . .; v x 1
; . . .; x m
oxi oxi
1 m 1
¼ w u1 x1 ; . . .; xm ; . . .; w u1 x ; . . .; xm ;
! !
oðw u1 Þ 1 1 m
1
i oð w u Þ
vi x ; . . .; x m
; . . .; v x 1
; . . .; x m
:
oxi oxi
5.3 Tangent Bundles 339
Thus,
1 m 1
~ u
w ~ 1 x1 ; . . .; xm ; v1 ; . . .; vm ¼ w u1 x1 ; . . .; xm ; . . .; w u1 x ; . . .; xm ;
!
oðw u1 Þ 1 1 m
1
i oð w u Þ
v i
x ; . . .; x m
; . . .; v x 1
; . . .; x m
Þ:
oxi oxi
~ u
is smooth, and hence, w ~ 1 is smooth.
4. Since fMg is an open cover of M, by the Lemma 4.52, there exists a countable
collection fðU1 ; u1 Þ; ðU2 ; u2 Þ; ðU3 ; u3 Þ; g of admissible coordinate charts of
M such that fU1 ; U2 ; U3 ; g covers M. Here, each ðUk ; u‘ k Þ is an admissible
coordinate charts of M, so each ðUk ; uk Þ 2 A, and hence, f r2Uk Tr Mgk2N is a
‘
countable subcollection of f r2U Tr MgðU;uÞ2A :
‘ ‘
Now, we want to prove that f r2Uk Tr Mgk2N covers T M: Since
1 1
‘ r
r2M
fU1 ; U2 ; U3 ; . . .g covers M, M ¼ [k¼1 Uk ; and hence, [k¼1 ð q2Uk Tq MÞ ¼
‘ ‘ ‘ ‘ ‘
q2[1 Tq M ¼ q2M Tq M ¼ r2M Tr M: It follows that f q2U1 Tq M; q2U2
‘ Uk
k¼1 ‘
Tq M; q2U3 Tq M; . . .g covers r2M Tr M:
‘
5. Let ðp; vÞ 2 TM; ðq; wÞ 2 r2M Tr M; and ðp; vÞ 6¼ ðq; wÞ: Since ðp; vÞ 2
‘
r2M Tr M; v 2 Tp M: Similarly, w 2 Tq M:
Here, dF is called the global differential of F (or, global tangent map of F).
Theorem 5.24 Let M be an m-dimensional smooth manifold, N be an n-dimen-
sional smooth manifold, and F : M ! N be smooth. Then, dF : TM ! TN is
smooth.
‘
Proof Let us take any ðp; vÞ 2 TMð¼ r2M Tr MÞ where p 2 M; and ðp; vÞ 2 Tp M:
Let us take any admissible coordinate chart ðU; uÞ of M such that p 2 U: For every
q 2 U; let
5.3 Tangent Bundles 341
!
o o o
; ; . . .; m
ox1 q ox2 q ox q
Thus,
~ Fu
w ~ 1 x1 ; . . .; xm ; v1 ; . . .; vm
¼ p01 w F u1 x1 ; . . .; xm ; . . .; p0n w F u1 x1 ; . . .; xm ;
0 0
o p1 w F u1 1 i o pn w F u
1
vi x ; . . .; xm
; . . .; v x1
; . . .; xm
oxi oxi
~ Fu
is smooth, and hence, w ~ 1 is smooth. h
Hence, LHS ¼ RHS: Thus, we have shown that dðG FÞ ¼ ðdGÞ ðdFÞ:
Note 5.26 Let M be a smooth manifold. Here, IdM : M ! M is given by IdM ðxÞ ¼
x for every x in M. Clearly, IdM is a smooth function. Hence, by Theorem 5.24,
dðIdM Þ : TM ! TM is smooth. Here, IdTM : TM ! TM is a smooth function. ‘ We
shall try to show that dðIdM Þ ¼ IdTM ; that is, for every ðp; vÞ 2 TMð¼ r2U Tr MÞ;
where p 2 M; and v 2 Tp M; ðdðIdM ÞÞðp; vÞ ¼ ðIdTM Þðp; vÞ:
LHS ¼ ðdðIdM ÞÞðp; vÞ ¼ IdM ðpÞ; dðIdM Þp ðvÞ ¼ p; dðIdM Þp ðvÞ
¼ p; dðIdM Þp ðvÞ ¼ p; IdTp M ðvÞ ¼ ðp; vÞ ¼ ðIdTM Þðp; vÞ ¼ RHS:
basis vector of Tt0 R; dtd is the standard coordinate basis vector of T ðt e; t þ eÞ:
t0 0 0
t0
d is in T ðt e; t þ eÞ; and dc : T ðt e; t þ eÞ ! T M;
Since dt t0 0 0 t0 t0 0 0 cðt0 Þ
t0
d Þ 2 T M: Here, ðdc Þð d Þ is denoted by c0 ðt Þ; and is called the
ðdct0 Þðdt cðt0 Þ t0 dt 0
t0 t0
velocity of c at t0 : Since
!
d
c0 ðt0 Þ ¼ dct0 2 Tcðt0 Þ M; so c0 ðt0 Þ 2 Tcðt0 Þ M:
dtt0
Since c0 ðt0 Þ 2 Tcðt0 Þ M; c0 ðt0 Þ is a function from C 1 ðMÞ to R: Hence, for every
f 2 C 1 ðMÞ; we have
! !!
0 d d
ðf cÞ ðt0 Þ ¼ ð f cÞ ¼ dct0 ðf Þ ¼ ðc0 ðt0 ÞÞðf Þ:
dtt0 dtt0
Thus, for every f 2 C 1 ðMÞ; we have ðc0 ðt0 ÞÞðf Þ ¼ ðf cÞ0 ðt0 Þ:
Here, cðt0 Þ 2 M; and M is an m-dimensional smooth manifold, so there exists an
admissible coordinate chart ðU; uÞ of M such that cðt0 Þ 2 U: Let
!
o o
; . . .; m
ox1 cðt0 Þ ox cðt0 Þ
Thus,
!
0 dðpi ðu cÞÞ o
c ðt0 Þ ¼ ðt 0 Þ :
dt oxi cðt0 Þ
In the case of Euclidean space Rm for M, the above formula takes the form:
!
0 d ð pi c Þ o
c ðt0 Þ ¼ ðt0 Þ ;
dt oxi cðt0 Þ
Here,
!
o o
; . . .; m
ox1 p ox p
348 5 Immersions, Submersions, and Embeddings
so
d ð pi ð u c Þ Þ
ð0Þ ¼ vi :
dt
Hence, c0 ð0Þ ¼ v: h
Note 5.30 Let M be an m-dimensional smooth manifold, N be an n-dimensional
smooth manifold, and F : M ! N be a smooth map. Let c : ðt0 e; t0 þ eÞ ! M be
a smooth map.
Since c : ðt0 e; t0 þ eÞ ! M is a smooth map, and F : M ! N is a smooth
map, their composite F c : ðt0 e; t0 þ eÞ ! N is a smooth map, and hence,
ðF cÞ0 ðt0 Þ 2 TðFcÞðt0 Þ N ¼ TFðcðt0 ÞÞ N: Since c : ðt0 e; t0 þ eÞ ! M is a smooth
map, c0 ðt0 Þ 2 Tcðt0 Þ M: Since c : ðt0 e; t0 þ eÞ ! M; cðt0 Þ 2 M: Since cðt0 Þ
2 M; and F : M ! N is smooth, dFcðt0 Þ : Tcðt0 Þ M ! TFðcðt0 ÞÞ N. Since dFcðt0 Þ :
5.4 Smooth Immersion 349
Tcðt0 Þ M ! TFðcðt0 ÞÞ N; and c0 ðt0 Þ 2 Tcðt0 Þ M; ðdFcðt0 Þ Þðc0 ðt0 ÞÞ 2 TFðcðt0 ÞÞ N: Thus,
ðdFcðt0 Þ Þðc0 ðt0 ÞÞ and ðF cÞ0 ðt0 Þ are elements of TFðcðt0 ÞÞ N: We shall try to show that
ðdFcðt0 Þ Þðc0 ðt0 ÞÞ ¼ ðF cÞ0 ðt0 Þ:
! !
d d
0
RHS ¼ ðF cÞ ðt0 Þ ¼ dðF cÞt0 ¼ dFcðt0 Þ dct0
dtt0 dtt0
!!
d
¼ dFcðt0 Þ dct0 ¼ dFcðt0 Þ ðc0 ðt0 ÞÞ ¼ LHS:
dt t0
This proves that ðdFp ÞðvÞ ¼ ðF cÞ0 ð0Þ: In short, we write ðdFcð0Þ Þ
ðc0 ð0ÞÞ ¼ ðF cÞ0 ð0Þ:
Definition Let M be an m-dimensional smooth manifold, N be an n-dimensional
smooth manifold, and F : M ! N be a smooth map. Let p 2 M:
It follows that dFp : Tp M ! TFðpÞ N: We know that dFp is a linear map from real
linear space Tp M to real linear space TFðpÞ N: Here, ðdFp ÞðTp MÞ is a linear subspace
of TFðpÞ N: We know that dim Tp M ¼ m; and dim TFðpÞ N ¼ n: Since ðdFp ÞðTp MÞ is
a linear subspace of TFðpÞ N; dimððdFp ÞðTp MÞÞ dim TFðpÞ N ¼ n: Since dFp is a
linear map from real linear space Tp M to real linear space TFðpÞ N; dimððdFp Þ
ðTp MÞÞ dim Tp M ¼ m: Since dimððdFp ÞðTp MÞÞ m; and dimððdFp ÞðTp MÞÞ
n; dimððdFp ÞðTp MÞÞ minfm; ng: Here, dimððdFp ÞðTp MÞÞ is called the rank of
F at p.
It follows that the rank of F at p is r if and only if the matrix representation of
linear map dFp has rank r. Clearly, the rank of F at p is less than or equal to
minfm; ng: Observe that the linear map dFp : Tp M ! TFðpÞ N is onto if and only if
dimððdFp ÞðTp MÞÞ ¼ dim TFðpÞ Nð¼ nÞ: Thus, the linear map dFp : Tp M ! TFðpÞ N
is onto if and only if the rank of F at p is n.
350 5 Immersions, Submersions, and Embeddings
Observe that the linear map dFp : Tp M ! TFðpÞ N is 1–1 if and only if
dimððdFp ÞðTp MÞÞ ¼ dim Tp Mð¼ mÞ: Thus, the linear map dFp : Tp M ! TFðpÞ N is
1–1 if and only if the rank of F at p is m.
Definition Let M be an m-dimensional smooth manifold, N be an n-dimensional
smooth manifold, and F : M ! N be a smooth map. Let k be a positive integer. If
for every p 2 M; the rank of F at p is k, then we write rank F ¼ k; and say that
F has constant rank k. Let M be an m-dimensional smooth manifold, N be an n-
dimensional smooth manifold, and F : M ! N be a smooth map. Here, the linear
map dFp : Tp M ! TFðpÞ N is onto for every p 2 M if and only if the rank of F at p is
n for every p 2 M: Thus, the linear map dFp : Tp M ! TFðpÞ N is onto for every
p 2 M if and only if rank F ¼ n:
Note 5.32 Let M be an m-dimensional smooth manifold, N be an n-dimensional
smooth manifold, and F : M ! N be a smooth map. Here, the linear map dFp :
Tp M ! TFðpÞ N is 1–1 for every p 2 M if and only if the rank of F at p is m for
every p 2 M: Thus, the linear map dFp : Tp M ! TFðpÞ N is 1–1 for every p 2 M if
and only if rank F ¼ m:
Definition Let M be an m-dimensional smooth manifold, N be an n-dimensional
smooth manifold, and F : M ! N be a smooth map. If for every p 2 M; the linear
map dFp : Tp M ! TFðpÞ N is onto, then we say that F is a smooth submersion. Thus,
F is a smooth submersion if and only if rank F ¼ n:
Definition Let M be an m-dimensional smooth manifold, N be an n-dimensional
smooth manifold, and F : M ! N be a smooth map. If for every p 2 M; the linear
map dFp : Tp M ! TFðpÞ N is 1–1, then we say that F is a smooth immersion. Thus,
F is a smooth immersion if and only if rank F ¼ m:
Note 5.33 Let Mð2 3; RÞ be the collection of all 2 3 matrices with real entries,
that is,
a b c
Mð2 3; RÞ : a; b; c; d; e; f 2 R :
d e f
a b c a b1 c1 a þ a1 b þ b1 c þ c 1
þ 1 :
d e f d1 e1 f1 d þ d1 e þ e1 f þ f1
a b c ta tb tc
t :
d e f td te tf
Clearly, Mð2 3; RÞ becomes a real linear space of dimension ð2 3Þ: Here, one
of the basis of Mð2 3; RÞ is
1 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0
; ; ; ; ; :
0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 1
Clearly, the real linear space R23 is isomorphic to the real linear space
Mð2 3; RÞ: Since R23 is isomorphic to Mð2 3; RÞ; we will not distinguish
between
a b c
and ða; b; c; d; e; f Þ:
d e f
a b c
2 Mð2 3; RÞ
d e f
such that
a b c
rank ¼ 2:
d e f
If
a b c
2 M2 ð2 3; RÞ;
d e f
then
a b c
rank ¼ 2;
d e f
that is, if
a b c
2 M2 ð2 3; RÞ
d e f
352 5 Immersions, Submersions, and Embeddings
then
b c a c a b
det 6¼ 0 or det 6¼ 0 or det 6¼ 0 :
e f d f d e
a b c
2 M2 ð2 3; RÞ:
d e f
It follows that
b c a c a b
det 6¼ 0 or det 6¼ 0 or det 6¼ 0:
e f d f d e
Case I: when
b c
det 6¼ 0
e f
Here,
b c
bf ce ¼ det 6¼ 0:
e f
x2 x3
f ðx1 ; x2 ; x3 ; x4 ; x5 ; x6 Þ x2 x6 x5 x3 ¼ det :
x5 x6
b c
f ða; b; c; d; e; f Þ ¼ det :
e f
x2 x3
det ¼ f ðx1 ; x2 ; x3 ; x4 ; x5 ; x6 Þ 2 ðf ða; b; c; d; e; f Þ e; f ða; b; c; d; e; f Þ þ eÞ
x5 x6
b c b c
¼ det e; det þ e 63 0:
e f e f
and hence,
x1 x2 x3
rank ¼2
x4 x5 x6
is an open neighborhood of
a b c
in Mð2 3; RÞ:
d e f
Since
x1 x2 x3
rank ¼2
x4 x5 x6
x1 x2 x3
: ð x1 ; x2 ; x3 ; x4 ; x5 ; x6 Þ 2 U M2 ð2 3; RÞ:
x4 x5 x6
Since
x1 x2 x3
: ð x1 ; x2 ; x3 ; x4 ; x5 ; x6 Þ 2 U
x4 x5 x6
is an open neighborhood of
a b c
in Mð2 3; RÞ;
d e f
354 5 Immersions, Submersions, and Embeddings
and
x1 x2 x3
: ð x1 ; x2 ; x3 ; x4 ; x5 ; x6 Þ 2 U M2 ð2 3; RÞ;
x4 x5 x6
a b c
d e f
a b c
d e f
!
o 1 o
d u uðqÞ :
oxi q oxi uðqÞ
Here, the matrix representation of linear map dFp is the following n m matrix:
2
oðp1 ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ
6 ðuðpÞÞ ðuðpÞÞ
6 ox1 ox2
6 oðp ðw F u1 ÞÞ 1
oðp2 ðw F u ÞÞ
6 2
6 ðuðpÞÞ ðuðpÞÞ
6 ox1 ox2
6... ...
6
4 oðpn ðw F u1 ÞÞ oðpn ðw F u1 ÞÞ
ðuðpÞÞ ðuðpÞÞ
ox1 3 ox
2
1
oðp1 ðw F u ÞÞ
... ðuðpÞÞ 7
oxm 7
1
oðp2 ðw F u ÞÞ 7
7
... ðuðpÞÞ 7:
ox m 7
... ... 7
7
oðpn ðw F u1 ÞÞ 5
... ð uðpÞÞ
ox m
Since the linear map dFp : Tp M ! TFðpÞ N is onto, and dim TFðpÞ N ¼ n; the rank of
the matrix representation of linear map dFp is n; that is,
2
oðp1 ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ
6 ðuðpÞÞ ðuðpÞÞ
6 ox1 ox2
6 1 1
6 oðp2 ðw F u ÞÞ oðp2 ðw F u ÞÞ
rank6 ðuðpÞÞ ðuðpÞÞ
6 ox1 ox2
6... ...
6
4 oðpn ðw F u1 ÞÞ oðpn ðw F u1 ÞÞ
ðuðpÞÞ ðuðpÞÞ
ox1 3 ox
2
1
oðp1 ðw F u ÞÞ
... ðuðpÞÞ 7
oxm 7
oðp2 ðw F u1 ÞÞ 7
7
... ðuðpÞÞ 7
oxm 7
... ... 7
7
1
oðpn ðw F u ÞÞ 5
... ðuðpÞÞ
ox m
356 5 Immersions, Submersions, and Embeddings
is n. Since the rank of F at p is less than or equal to minfm; ng; and the rank of F at
p is n, n minfm; ng m; and hence, n m: Here, n m; and
2
oðp1 ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ
6 ð uðpÞÞ ðuðpÞÞ
6 ox1 ox2
6 oðp ðw F u1 ÞÞ 1
oðp2 ðw F u ÞÞ
6 2
rank6 ðuðpÞÞ ðuðpÞÞ
6 ox1 ox2
6... . . .
6
4 oðpn ðw F u1 ÞÞ oðpn ðw F u1 ÞÞ
ðuðpÞÞ ðuðpÞÞ
ox1
3 ox2
oðp1 ðw F u1 ÞÞ
... ðuðpÞÞ 7
oxm 7
1
oðp2 ðw F u ÞÞ 7
7
... ðuðpÞÞ 7
ox m 7
... ... 7
7
oðpn ðw F u1 ÞÞ 5
... ðuðpÞÞ
oxm
is n, so
2
oðp1 ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ
6 ðuðpÞÞ ðuðpÞÞ
6 ox 1 ox2
6 oðp ðw F u1 ÞÞ oðp2 ðw F u1ÞÞ
6 2
6 ðuðpÞÞ ðuðpÞÞ
6 ox1 ox2
6... ...
6
4 oðpn ðw F u1 ÞÞ oðpn ðw F u1 ÞÞ
ðuðpÞÞ ðuðpÞÞ
ox 1
3 ox
2
1
oðp1 ðw F u ÞÞ
... ðuðpÞÞ 7
oxm 7
oðp2 ðw F u1 ÞÞ 7
7
... ð uðpÞ Þ 7
oxm 7
... ... 7
7
1
oðpn ðw F u ÞÞ 5
... ð uðpÞ Þ
oxm
oðpi ðw F u1 ÞÞ
q 7! ðuðqÞÞ
ox j
2
oðp1 ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ
6 ðuðqÞÞ ðuðqÞÞ
6 ox1 ox2
6 oðp ðw F u1 ÞÞ 1
oðp2 ðw F u ÞÞ
6 2
rank6 ðuðqÞÞ ðuðqÞÞ
6 ox1 ox2
6... ...
6
4 oðpn ðw F u1 ÞÞ oðpn ðw F u1 ÞÞ
ðuðqÞÞ ðuðqÞÞ
ox1 3 ox2
oðp1 ðw F u1 ÞÞ
... ðuðqÞÞ 7
oxm 7
oðp2 ðw F u1 ÞÞ 7
7
... ð uðqÞ Þ 7 ¼ n:
oxm 7
... ... 7
7
oðpn ðw F u1 ÞÞ 5
... ðuðqÞÞ
ox m
It follows that for every q 2 W; the linear map dFq : Tq M ! TFðqÞ N is onto. h
Theorem 5.35 Let M be an m-dimensional smooth manifold, N be an n-dimen-
sional smooth manifold, and F : M ! N be a smooth map. Let p 2 M: If dFp :
Tp M ! TFðpÞ N is 1–1, then there exists an open neighborhood W of p such that for
every q 2 W; the linear map dFq : Tq M ! TFðqÞ N is 1–1.
Proof Here, p 2 M; and M is an m-dimensional smooth manifold, so there exists an
admissible coordinate chart ðU; uÞ of M such that p 2 U: For every q 2 U;
!
o o o
; ; . . .; m
ox1 q ox2 q ox q
Here, the matrix representation of linear map dFp is the following n m matrix:
5.5 Inverse Function Theorem for Manifolds 359
2
oðp1 ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ
6 ðuðpÞÞ ðuðpÞÞ
6 ox 1 ox2
6 oðp ðw F u1 ÞÞ oðp2 ðw F u1 ÞÞ
6 2
6 ðuðpÞÞ ðuðpÞÞ
6 ox1 ox2
6... . . .
6
4 oðpn ðw F u1 ÞÞ oðpn ðw F u1 ÞÞ
ðuðpÞÞ ðuðpÞÞ
ox 1
3 ox2
oðp1 ðw F u1 ÞÞ
... ðuðpÞÞ 7
oxm 7
1
oðp2 ðw F u ÞÞ 7
7
... ð uðpÞ Þ 7:
oxm 7
... ... 7
7
1
oðpn ðw F u ÞÞ 5
... ðuðpÞÞ
oxm
Since the linear map dFp : Tp M ! TFðpÞ N is 1–1, and dim Tp M ¼ m; the rank of
the matrix representation of linear map dFp is m; that is,
2
oðp1 ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ
6 ðuðpÞÞ ðuðpÞÞ
6 ox1 ox2
6
6 oðp2 ðw F u1 ÞÞ oðp2 ðw F u1 ÞÞ
rank6 ðuðpÞÞ ðuðpÞÞ
6 ox1 ox2
6... . . .
6
4 oðpn ðw F u1 ÞÞ oðpn ðw F u1 ÞÞ
ðuðpÞÞ ðuðpÞÞ
ox1
3 ox2
oðp1 ðw F u1 ÞÞ
... ðuðpÞÞ 7
oxm 7
1
oðp2 ðw F u ÞÞ 7
7
... ðuðpÞ Þ 7
ox m 7
... ... 7
7
1
oðpn ðw F u ÞÞ 5
... ðuðpÞÞ
ox m
is m. Since the rank of F at p is less than or equal to minfm; ng; and the rank of F at
p is m, m minfm; ng n; and hence, m n: Here, m n; and
2
oðp1 ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ
6 ðuðpÞÞ ðuðpÞÞ
6 ox1 ox2
6 oðp ðw F u1 ÞÞ oðp2 ðw F u1 ÞÞ
6 2
rank6 ðuðpÞÞ ðuðpÞÞ
6 ox1 ox2
6... . . .
6
4 oðpn ðw F u1 ÞÞ oðpn ðw F u1 ÞÞ
ðuðpÞÞ ðuðpÞÞ
ox1 3 ox
2
1
oðp1 ðw F u ÞÞ
... ðuðpÞÞ 7
oxm 7
1
oðp2 ðw F u ÞÞ 7
7
... ðuðpÞÞ 7
ox m 7
... ... 7
7
oðpn ðw F u1 ÞÞ 5
... ðuðpÞ Þ
ox m
360 5 Immersions, Submersions, and Embeddings
is m, so
2
oðp1 ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ
6 ðuðpÞÞ ðuðpÞÞ
6 ox 1 ox2
6 oðp ðw F u1 ÞÞ oðp2 ðw F u1 ÞÞ
6 2
6 ð uðpÞ Þ ðuðpÞÞ
6 ox1 ox2
6... . . .
6
4 oðpn ðw F u1 ÞÞ oðpn ðw F u1 ÞÞ
ð uðpÞ Þ ðuðpÞÞ
ox1 3 ox2
oðp1 ðw F u1 ÞÞ
... ðuðpÞÞ 7
oxm 7
oðp2 ðw F u1 ÞÞ 7
7
... ð uðpÞ Þ 7
oxm 7
... ... 7
7
1
oðpn ðw F u ÞÞ 5
... ð uðpÞ Þ
oxm
oðpi ðw F u1 ÞÞ
q 7! ðuðqÞÞ
ox j
2
oðp1 ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ
6 ð uðqÞ Þ ðuðqÞÞ
6 ox1 ox2
6 oðp ðw F u1 ÞÞ oðp2 ðw F u1 ÞÞ
6 2
rank6 ðuðqÞÞ ðuðqÞÞ
6 ox 1 ox2
6... ...
6
4 oðpn ðw F u1 ÞÞ oðpn ðw F u1 ÞÞ
ðuðqÞÞ ðuðqÞÞ
ox1 3 ox2
oðp1 ðw F u1 ÞÞ
... ðuðqÞÞ 7
oxm 7
1 7
oðp2 ðw F u ÞÞ 7
... ð uðqÞ Þ 7 ¼ m:
oxm 7
... ... 7
7
1
oðpn ðw F u ÞÞ 5
... ð uðqÞ Þ
oxm
It follows that for every q 2 W; the linear map dFq : Tq M ! TFðqÞ N is 1–1. h
Theorem 5.36 Let M be an m-dimensional smooth manifold, N be an m-dimen-
sional smooth manifold, and F : M ! N be a smooth map. Let p 2 M: If dFp :
Tp M ! TFðpÞ N is invertible (i.e., 1–1 and onto); then, there exists a connected open
neighborhood U0 of p, and a connected open neighborhood V0 of FðpÞ such that
FjU0 : U0 ! V0 is a diffeomorphism.
Proof Here, p 2 M; and M is an m-dimensional smooth manifold, so there exists an
admissible coordinate chart ðU; uÞ of M such that p 2 U: For every q 2 U; let
!
o o o
; ; . . .; m
ox1 q ox2 q ox q
!
o 1 o
d w wðrÞ :
oyi r oyi wðrÞ
Here, the matrix representation of linear map dFp is the following m m matrix:
2
oðp1 ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ
6 ðuðpÞÞ ðuðpÞÞ
6 ox 1 ox2
6 oðp ðw F u1 ÞÞ oðp2 ðw F u1 ÞÞ
6 2
6 ð uðpÞ Þ ðuðpÞÞ
6 ox1 ox2
6... ...
6
4 oðpm ðw F u1 ÞÞ oðpm ðw F u1 ÞÞ
ðuðpÞÞ ðuðpÞÞ
ox1 3 ox2
oðp1 ðw F u1 ÞÞ
... ðuðpÞÞ 7
oxm 7
oðp2 ðw F u1 ÞÞ 7
7
... ðuðpÞÞ 7:
ox m 7
... ... 7
7
oðpm ðw F u ÞÞ1 5
... ðuðpÞ Þ
oxm
Since the linear map dFp : Tp M ! TFðpÞ N is 1–1 onto, and dim Tp M ¼ m;
2
oðp1 ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ
6 ð uðpÞ Þ ðuðpÞÞ
6 ox1 ox2
6 oðp ðw F u1 ÞÞ oðp2 ðw F u1 ÞÞ
6 2
det6 ðuðpÞÞ ðuðpÞÞ
6 ox 1 ox2
6... ...
6
4 oðpn ðw F u1 ÞÞ oðpn ðw F u1 ÞÞ
ðuðpÞÞ ðuðpÞÞ
ox1 3 ox2
oðp1 ðw F u1 ÞÞ
... ðuðpÞÞ 7
oxm 7
oðp2 ðw F u1 ÞÞ 7
7
... ð uðpÞ Þ 7 6¼ 0:
oxm 7
... ... 7
7
1
oðpn ðw F u ÞÞ 5
... ðuðpÞÞ
ox m
Put U0 u1 ðUÞ;^ and V0 ðF u1 ÞðUÞ: ^ We want to prove that U0 is con-
nected and is an open neighborhood of p in M. Since U ^ is connected, U
^ is contained
1 1 1 ^
in uðU \ F ðVÞÞ; and u is continuous, u ðUÞð¼ U0 Þ is connected, and hence,
U0 is connected. Since U ^ is open, U
^ contained in uðU \ F 1 ðVÞÞð uðUÞÞ; and
uðUÞ is open, U ^ is open in uðUÞ; and hence, u1 ðUÞ ^ is open in U. Since u1 ðUÞ ^ is
1 ^
open in U, and U is open in M, u ðUÞð¼ U0 Þ is open in M, and hence, U0 is open
in M. Since uðpÞ 2 U; ^
^ p 2 u1 ðUÞð¼ U0 Þ; and hence, p 2 U0 :
Since u ; F are continuous, F u1 is continuous. Since F u1 is continu-
1
We have to show that F(U) is open in N, that is, each member of F(U) is an
interior point of F(U), that is, for every a in U, there exists an open neighborhood
V of F(a) such that V FðUÞ.
For this purpose, let us take any a in U. Since F : M ! N is a local diffeo-
morphism, and a is in M, there exists an open neighborhood U1 of a such that
FðU1 Þ is open in N, and the restriction FjU1 : U1 ! FðU1 Þ is a diffeomorphism.
Since U; U1 are open neighborhoods of a, U \ U1 is an open neighborhood of a,
and hence, U \ U1 is open in U1 . Since FjU1 : U1 ! FðU1 Þ is a diffeomorphism,
FjU1 : U1 ! FðU1 Þ is a homeomorphism. Since FjU1 : U1 ! FðU1 Þ is a homeo-
morphism, and U \ U1 is open in U1 , FjU1 ðU \ U1 Þð¼ FðU \ U1 ÞÞ is open in the
open set FðU1 Þ, and hence, FðU \ U1 Þ is open in the open set FðU1 Þ. Since
FðU \ U1 Þ is open in the open set FðU1 Þ, FðU \ U1 Þ is open in N. Since a is in
U \ U1 , F(a) is in FðU \ U1 Þ. Thus, FðU \ U1 Þ is an open neighborhood of F(a).
Also, FðU \ U1 Þ FðUÞ.
Note 5.39 Let M, N, P be smooth manifolds. Let F : M ! N; and G : N ! P be
local diffeomorphisms. We shall try to show that their composite G F : M ! P is
a local diffeomorphism.
Let us take any a 2 M. We have to find an open neighborhood U of a such that
ðG FÞðUÞ is open in P, and the restriction ðG FÞjU : U ! ðG FÞðUÞ is a
diffeomorphism.
Since F : M ! N is a local diffeomorphism, and a 2 M, there exists an open
neighborhood U1 of a such that FðU1 Þ is open in N, and the restriction FjU1 :
U1 ! FðU1 Þ is a diffeomorphism. Since FðaÞ 2 N, and G : N ! P is a local dif-
feomorphism, there exists an open neighborhood V of F(a) such that G(V) is open
in P, and the restriction GjV : V ! GðVÞ is a diffeomorphism. Since FðU1 Þ is an
open neighborhood of F(a), and V is an open neighborhood of F(a), FðU1 Þ \ V is
an open neighborhood of F(a). Here, FðU1 Þ \ V is an open neighborhood of F(a) in
FðU1 Þ, and FjU1 : U1 ! FðU1 Þ is a diffeomorphism, ðFjU1 Þ1 ðFðU1 Þ \ VÞ is an
open neighborhood of a, and FjðFj 1
ðFðU1 Þ\VÞ : ðFjU1 Þ1 ðFðU1 Þ \ VÞ ! FðU1 Þ \
U1 Þ
(Reason: Let ðy; xÞ 2 LHS. So ðy; xÞ 2 ðFjU Þ1 . Since ðy; xÞ 2 ðFjU Þ1 ,
ðx; yÞ 2 FjU . Since ðx; yÞ 2 FjU , x 2 U, and ðx; yÞ 2 F. Since ðx; yÞ 2 F, ðy; xÞ
2 F 1 . Since ðx; yÞ 2 F, FðxÞ ¼ y. Since
x 2 U, y ¼ FðxÞ 2 FðUÞ. Since ðy; xÞ 2
F 1 , and y 2 FðUÞ, ðy; xÞ 2 ðF 1 ÞFðUÞ ¼ RHS. Thus, LHS RHS. Next, let
ðy; xÞ 2 RHS. So ðy; xÞ 2 ðF 1 Þ FðUÞ
. Since ðy; xÞ 2 ðF 1 Þ , y 2 FðUÞ, and
FðUÞ
ðy; xÞ 2 F 1 . Since y 2 FðUÞ, there exists x1 2 U such that Fðx1 Þ ¼ y. Since
ðy; xÞ 2 F 1 , ðx; yÞ 2 F, and hence, FðxÞ ¼ y. Since FðxÞ ¼ y, and Fðx1 Þ ¼ y,
FðxÞ ¼ Fðx1 Þ. Since FðxÞ ¼ Fðx1 Þ and F is 1–1, x ¼ x1 . Since x ¼ x1 , and x1 2 U,
x 2 U. Since x 2 U, and ðx; yÞ 2 F, ðx; yÞ 2 FjU , and hence, ðy; xÞ 2
ðFjU Þ1 ¼ LHS. Thus, RHS LHS. Hence, RHS LHS:Þ
Since
F : M ! N is a diffeomorphism, F 1 : N ! Mis smooth, and hence,
ðF 1 ÞFðUÞ : FðUÞ ! U is smooth. Since ðF 1 ÞFðUÞ : FðUÞ ! U is smooth, and
ðFj Þ1 ¼ ðF 1 Þ
U FðUÞ
, ðFj Þ1 : FðUÞ ! U is smooth.
U
Proof For this purpose, let us take any p 2 M. We have to find an open neigh-
borhood U of p in M such that F(U) is open in N, and the restriction FjU : U !
FðUÞ is a diffeomorphism.
Since p 2 M, and F : M ! N is a smooth immersion, the linear map dFp :
Tp M ! TFðpÞ N is 1–1. Since p 2 M, and F : M ! Nis a smooth submersion, the
linear map dFp : Tp M ! TFðpÞ N is onto. Since the linear map dFp : Tp M ! TFðpÞ N
is 1–1 onto, m ¼ dim M ¼ dim Tp M ¼ dimðTFðpÞ NÞ ¼ n.
Since m = n, and N is an n-dimensional smooth manifold, N is an m-dimensional
smooth manifold. Since M, N are m-dimensional smooth manifolds, F : M ! N is a
smooth map, and dFp : Tp M ! TFðpÞ N is 1–1 onto, by Theorem 5.36, there exists
an open neighborhood U of p, and an open neighborhood V of F(p) such that
FjU : U ! V is a diffeomorphism. Since FjU : U ! V is a diffeomorphism, FjU :
U ! V is 1–1 onto, and hence, FðUÞ ¼ ðFjU ÞðUÞ ¼ V. Since FðUÞ ¼ V, and V is
open, F(U) is open. Since FjU : U ! V is a diffeomorphism, and FðUÞ ¼ V, FjU :
U ! FðUÞ is a diffeomorphism. Thus, U is an open neighborhood of p, F(U) is
open in N, and FjU : U ! FðUÞ is a diffeomorphism. h
Theorem 5.46 Let M, N be m-dimensional smooth manifolds. Let F : M ! N be a
smooth immersion. Then, F : M ! N is a local diffeomorphism.
Proof We first try to show that F : M ! N is a smooth submersion. For this
purpose, let us take any p 2 M. We have to show that the linear map dFp : Tp M !
TFðpÞ N is onto.
Since p 2 M, and F : M ! N is a smooth immersion, the linear map dFp :
Tp M ! TFðpÞ N is 1–1. Since dim Tp M ¼ dim M ¼ m ¼ dim N ¼ dimðTFðpÞ NÞ,
dim Tp M ¼ dimðTFðpÞ NÞ. Since dim Tp M ¼ dimðTFðpÞ NÞ, and the linear map dFp :
Tp M ! TFðpÞ N is 1–1, dFp : Tp M ! TFðpÞ N is onto. Since F : M ! N is a smooth
immersion and smooth submersion, by Theorem 5.45, F : M ! N is a local
diffeomorphism. h
Theorem 5.47 Let M, N be m-dimensional smooth manifolds. Let F : M ! N be a
smooth submersion. Then, F : M ! N is a local diffeomorphism.
Proof We first try to show that F : M ! N is a smooth immersion. For this pur-
pose, let us take any p 2 M. We have to show that the linear map dFp : Tp M !
TFðpÞ N is 1–1.
Since p 2 M, and F : M ! N is a smooth submersion, the linear map dFp :
Tp M ! TFðpÞ N is onto. Since dim Tp M ¼ dim M ¼ m ¼ dim N ¼ dimðTFðpÞ NÞ,
dim Tp M ¼ dimðTFðpÞ NÞ. Since dim Tp M ¼ dimðTFðpÞ NÞ, and the linear map dFp :
Tp M ! TFðpÞ N is onto, dFp : Tp M ! TFðpÞ N is 1–1. Since F : M ! N is a smooth
submersion and smooth immersion, by Theorem 5.45, F : M ! N is a local dif-
feomorphism. h
Note 5.48 Before going ahead, let us recall the constant rank Theorem 4.9:
Let M be an m-dimensional smooth manifold, N be an n-dimensional smooth
manifold, F : M ! N be a smooth map, and r be the rank of F. Then, for every p in
5.6 Shrinking Lemma 371
For every q 2 U,
!
o o o
; ; . . .; m
ox1 q ox2 q ox q
For every r 2 V,
o o o
; ; . . .; n
oy1 r oy2 r oy r
We know that for every q 2 U, the matrix representation of linear map dFq is the
following n m matrix:
2
oðp1 ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ
6 ðuðqÞÞ ðuðqÞÞ
6 ox 1 ox2
6 oðp ðw F u1 ÞÞ 1
oðp2 ðw F u ÞÞ
6 2
6 ðuðqÞÞ ðuðqÞÞ
6 ox1 ox2
6... . . .
6
4 oðpn ðw F u1 ÞÞ oðpn ðw F u1 ÞÞ
ðuðqÞÞ ðuðqÞÞ
ox 1
3 ox
2
1
oðp1 ðw F u ÞÞ
... ðuðqÞÞ 7
oxm 7
1
oðp2 ðw F u ÞÞ 7
7
... ð uðqÞ Þ 7:
oxm 7
... ... 7
7
oðpn ðw F u1 ÞÞ 5
... ðuðqÞÞ
oxm
5.6 Shrinking Lemma 373
It follows that for every q 2 U, the rank of the matrix representation of the linear
map dFq is r. Since for every q 2 U, the rank of F at q is equal to the rank of the
matrix representation of the linear map dFq , and the rank of the matrix represen-
tation of the linear map dFq is r, the rank of F at q is r.
Thus, we have shown that, corresponding to each p 2 M, there exist a positive
integer r and an open neighborhood Up;r of p such that for every q 2 Up;r , the rank of
F at q is r. Observe that f[p2M Up;r : r ¼ 1; 2; . . .; minfm; ngg is a finite partition of
M into open sets. Since f[p2M Up;r : r ¼ 1; 2; . . .; minfm; ngg is a finite partition
of M into open sets, and M is connected, f[p2M Up;r : r ¼ 1; 2; . . .; minfm; ngg is a
singleton set. It follows that for every q 2 M, the rank of F at q is a constant. Thus,
F has constant rank.
2 ) 1: Let F has constant rank r. By the constant rank theorem, for every p in
M, there exist an admissible coordinate chart ðU; uÞ in M satisfying p 2 U and an
admissible coordinate chart ðV; wÞ in N satisfying FðpÞ 2 V such that FðUÞ V,
and for every ðx1 ; . . .; xr ; xrþ1 ; . . .; xm Þ in uðUÞ,
0 1
w F u1 ðx1 ; . . .; xr ; xrþ1 ; . . .; xm Þ ¼ @x1 ; . . .; xr ; 0; . . .; 0 A: h
|fflfflffl{zfflfflffl}
nr
h
Definition Let X be a topological space. If for every x 2 X, there exist an open
neighborhood Ux of x and a compact set C such that Ux C, then we say that X is a
locally compact space. Clearly, every compact space is locally compact.
Note 5.52 Let M be an n-dimensional smooth manifold. We shall try to show that
M is locally compact space. For this purpose, let us take any p 2 M. By Lemma
4.49 of Chap. 4, there exists a countable collection fðU1 ; u1 Þ; ðU2 ; u2 Þ;
ðU3 ; u3 Þ; . . .g of admissible coordinate charts of M such that
1. fU1 ; U2 ; U3 ; . . .g is a basis of M,
2. each ui ðUi Þ is an open ball in Rn ,
3. each closure Ui of Ui is a compact subset of M.
Since fU1 ; U2 ; U3 ; . . .g is a basis of M, fU1 ; U2 ; U3 ; . . .g is a cover of M, and
hence, there exists a positive integer k such that p 2 Uk . By 3, Uk is a compact
subset of M. Further, Uk Uk . Hence, M is locally compact.
Note 5.53 Let X be a Hausdorff topological space. Then, we shall try to show that
the following statements are equivalent:
1. X is locally compact.
2. For every x 2 X, there exists an open neighborhood Ux of x such that Ux is
compact.
3. There exists a basis B of X such that for every U 2 B; U is compact.
3 ) 2: Let B be a basis of X such that for every U 2 B; U is compact. We
have to prove 2. For this purpose, let us take any x 2 X. Since x 2 X, and B is a
374 5 Immersions, Submersions, and Embeddings
N satisfying FðpÞ 2 VFðpÞ , such that FðUp Þ VFðpÞ , and for every ðx1 ; . . .; xr ;
xrþ1 ; . . .; xm Þ in up ðUp Þ,
0 1
B C
wFðpÞ F u1 ðx1 ; . . .; xr ; xrþ1 ; . . .; xm Þ ¼ @x1 ; . . .; xr ; 0; . . .; 0A
p |fflfflffl{zfflfflffl}
ð1 Þnr
2 wFðpÞ VFðpÞ :
2 |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
R R f0g f0g :
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
r ð1 Þnr
0 1
B C
FðqÞ 2 VFðpÞ \ w1 R R f0g f0g A :
FðpÞ @|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
r ð1 Þnr
378 5 Immersions, Submersions, and Embeddings
It follows that
0 1
B C
F Wp VFðpÞ \ w1 R R f0g f0g A :
FðpÞ @|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
r ð1 Þnr
Clearly,
0 1
B C
VFðpÞ \ w1 R R f0g f0g A
FðpÞ @|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
r ð1 Þnr
and
0 1
B C
R R f0g f0g A
@|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
r ð1 Þnr
0 0 11
B B CC
wFðpÞ VFðpÞ ; @VFðpÞ \ w1 R R f0g f0g AA VFðpÞ ;
FðpÞ @|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
r ð1 Þnr
and
0 0 11
B B CC
wFðpÞ @VFðpÞ \ w1 R R f0g f0g A A
FðpÞ @|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
r ð1 Þnr
Since
0 1
B C
@a1 ; . . .; ar ; arþ1 ; . . .; am A ¼ uðpÞ 2 uðUÞ;
|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
ð1 Þmr
Since
0 1 0 1
B C B C
Ba1 ; . . .; ar ; arþ1 ; . . .; am e C; Ba1 ; . . .; ar ; arþ1 ; . . .; am þ e C
@ A @ A
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl2} |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl2}
ð1 Þmr ð1 Þmr
Since
0 1
B eC
w F u1 B
@a1 ; . . .; ar ; arþ1 ; . . .; am 2A
C
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
ð1 Þmr
0 1
B eC
¼ w F u1 B
@a1 ; . . .; ar ; arþ1 ; . . .; am þ 2A
C
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
ð1 Þmr
and w F u1 is 1 1;
0 1 0 1
B C B C
Ba1 ; . . .; ar ; arþ1 ; . . .; am e C ¼ Ba1 ; . . .; ar ; arþ1 ; . . .; am þ e C;
@ A @ A
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl2} |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl2}
ð1 Þmr ð1 Þmr
which is a contradiction. h
382 5 Immersions, Submersions, and Embeddings
of N, and the topology of GðFðMÞÞ is the subspace topology of G(N). Since the
topology of GðFðMÞÞ is the subspace topology of GðNÞ; and the topology of G
(N) is the subspace topology of P, the topology of GðFðMÞÞ as the subspace
topology of G(N) and the topology of G(F(M)) as the subspace topology of
P are the same. It follows that G|F(M) is a homeomorphism from F(M) onto G(F
(M)), where the topology of F(M) is the subspace topology of N, and the
topology of G(F(M)) is the subspace topology of P. Since F is a homeomor-
phism from M onto F(M), where F(M) has the subspace topology of N, and
GjFðMÞ is a homeomorphism from F(M) onto G(F(M)), where the topology of F
(M) is the subspace topology of N, and the topology of G(F(M)) is the subspace
topology of P, the composite ðGjFðMÞ Þ F is a homeomorphism from M onto
GðFðMÞÞð¼ ðG FÞðMÞÞ; where the topology of ðG FÞðMÞ is the subspace
topology of P. Further, since ðGjFðMÞ Þ F ¼ G F; G F is a homeomorphism
from M onto (G F)(M), where the topology of (G F)(M) is the subspace
topology of P.
For 1: Clearly, FjU : U ! N is a smooth immersion, that is, for every q 2 U; the
rank of F|U at q is m.
(Reason: For this purpose, let us take any q 2 U: Since F : M ! N is a smooth
immersion, and q 2 M, the linear map dFq : Tq M ! TFðqÞ N is 1–1. Since U is an
open neighborhood of q in M, Tq U and Tq M are essentially the same. Since Tq U
and Tq M are essentially the same, and dFq : Tq M ! TFðqÞ N is 1–1, dðFjU Þq :
Tq U ! TFðqÞ N is 1–1. Since the linear map dðFjU Þq : Tq U ! TFðqÞ N is 1–1,
ðrank of FjU at qÞ ¼ dimððdðFjU Þq ÞðTq UÞÞ ¼ dimðTq UÞ ¼ dimðTq MÞ ¼ dim M ¼
m: Hence, rank of FjU at q is m.)
For 2: First of all, we shall try to show that FjU : U ! N is a topological
embedding. Since FjU1 : U1 ! N is 1–1, and U U1 , FjU : U ! N is 1–1.
Since F : M ! N is smooth, F : M ! N is continuous, and hence, the
restriction FjU : U ! N is continuous. Further, since U is compact, and N is
a Hausdorff space, by closed map Lemma 5.66, FjU : U ! N is a topological
embedding. Since FjU : U ! N is a topological embedding, and U U ; the
388 5 Immersions, Submersions, and Embeddings
restriction ðFjU ÞU : U ! N is a topological embedding. Since ðFjU ÞU : U !
N is a topological embedding, and ðFjU ÞU ¼ FjU ; FjU : U ! N is a topological
embedding. h
Hence,
ða1 ; . . .; an Þ ¼ w p u1 ða1 ; . . .; an ; anþ1 ; . . .; am Þ
¼ w p u1 ðuðpÞÞ ¼ wðpðpÞÞ
is an interior point of
w p u1 ðða1 e; a1 þ eÞ ðan e; an þ eÞ
ðanþ1 e; anþ1 þ eÞ ðam e; am þ eÞÞÞÞ:
5.7 Global Rank Theorem 389
Since w; u1 ; and all projection maps pi are smooth, r is smooth. Here, we want to
prove that p r ¼ IdV ; that is;for every q 2 V ; pðrðqÞÞ ¼ q.
LHS ¼ pðrðqÞÞ ¼ p u1 ðp1 ðwðqÞÞ; . . .; pn ðwðqÞÞ; anþ1 ; . . .; am Þ
¼ p u1 ðp1 ðwðqÞÞ; . . .; pn ðwðqÞÞ; anþ1 ; . . .; am Þ
¼ w1 w p u1 ðp1 ðwðqÞÞ; . . .; pn ðwðqÞÞ; anþ1 ; . . .; am Þ
¼ w1 ðp1 ðwðqÞÞ; . . .; pn ðwðqÞÞÞ ¼ w1 ðwðqÞÞ ¼ q ¼ RHS:
390 5 Immersions, Submersions, and Embeddings
LHS ¼ rðpðpÞÞ
¼ u1 p1 w p u1 ðuðpÞÞ ; . . .; pn w p u1 ðuðpÞÞ ; anþ1 ; . . .; am
¼ u1 p1 w p u1 ða1 ; . . .; an ; anþ1 ; . . .; am Þ ; . . .; pn w p u1
ða1 ; . . .; an ; anþ1 ; . . .; am ÞÞ; anþ1 ; . . .; am Þ
¼ u1 ðp1 ða1 ; . . .; an Þ; . . .; pn ða1 ; . . .; an Þ; anþ1 ; . . .; am Þ
¼ u1 ða1 ; . . .; an ; anþ1 ; . . .; am Þ ¼ u1 ðuðpÞÞ ¼ p ¼ RHS: h
Since
dpp drpðpÞ ¼ IdTpðpÞ N ; so dpp : Tp M ! TpðpÞ N
is onto. h
Note 5.78 If we combine the Theorems 5.76, and 5.77, then we get the result: Let
M be an m-dimensional smooth manifold, N be an n-dimensional smooth manifold,
and p : M ! N be a smooth map.
p : M ! N is a smooth submersion if and only if for every p 2 M, there exist an
open neighborhood V of π(p) in N and a smooth map r : V ! M such that p r ¼
IdV and rðpðpÞÞ ¼ p.
Theorem 5.79 Let M be an m-dimensional smooth manifold, N be an n-dimen-
sional smooth manifold, and p : M ! N be a smooth submersion. Then, p : M !
N is an open map.
Proof Let G be a nonempty open subset of M. We have to show that π(G) is open
in N.
For this purpose, let us take any pðpÞ 2 pðGÞ; where p 2 G: Since p 2 M; and
p : M ! N is a smooth submersion, by Theorem 5.76, there exist an open
5.7 Global Rank Theorem 391
Proof Existence: We first try to show that F p1 is a function. For this purpose,
let ðx; yÞ 2 F p1 ; and ðx; zÞ 2 F p1 : We have to show that y ¼ z:
Since ðx; yÞ 2 F p1 ; there exists u such that ðx; uÞ 2 p1 ; and ðu; yÞ 2 F:
Since ðx; uÞ 2 p1 ; ðu; xÞ 2 p: Since ðx; zÞ 2 F p1 ; there exists v such that
ðx; vÞ 2 p1 ; and ðv; zÞ 2 F: Since ðx; vÞ 2 p1 ; ðv; xÞ 2 p: Since ðv; xÞ 2 p; pðvÞ ¼
x: Similarly, pðuÞ ¼ x: It follows that pðuÞ ¼ pðvÞ: Since pðuÞ ¼ pðvÞ; and F is
constant on fibers of π, FðuÞ ¼ FðvÞ: Since ðv; zÞ 2 F; FðvÞ ¼ z: Since ðu; yÞ 2 F;
FðuÞ ¼ y: Since FðuÞ ¼ y; FðvÞ ¼ z; and FðuÞ ¼ FðvÞ; y ¼ z: This proves that
F p1 is a function.
Since π : M ! N is onto, and F : M ! P; dom ðF p1 Þ ¼ N; and
ran ðF p1 Þ P: Since F p1 is a function, dom ðF p1 Þ ¼ N; and
ran ðF p1 Þ P; we can write F p1 : N ! P:
Put F^ F p1 : Since F^ ¼ F p1 ; and F p1 : N ! P; F ^ : N ! P: Now,
^ 1 1 1
F p ¼ ðF p Þ p ¼ F ðp pÞ: Clearly, F ðp pÞ ¼ F:
5.8 Properly Embedded 393
(Reason: Let ðx; yÞ 2 LHS ¼ F ðp1 pÞ: So there exist u, v such that
ðx; uÞ 2 p; ðu; vÞ 2 p1 ; and ðv; yÞ 2 F: Since ðu; vÞ 2 p1 ; ðv; uÞ 2 p: It follows
that pðvÞ ¼ u; pðxÞ ¼ u; and hence, pðxÞ ¼ pðvÞ: Since pðxÞ ¼ pðvÞ; and F is
constant on fibers of π, FðxÞ ¼ FðvÞ: Since ðv; yÞ 2 F; FðvÞ ¼ y: Since FðvÞ ¼ y;
and FðxÞ ¼ FðvÞ; FðxÞ ¼ y; and hence, ðx; yÞ 2 F ¼ RHS: Hence, LHS RHS:
Next, let ðx; yÞ 2 RHS ¼ F: Since ðx; yÞ 2 F; and F : M ! P; x 2 M; and y 2 P:
Since x 2 M; and p : M ! N; there exists u 2 N such that ðx; uÞ 2 p: Since ðx; uÞ 2
p; ðu; xÞ 2 p1 : Since ðx; uÞ 2 p; ðu; xÞ 2 p1 ; and ðx; yÞ 2 F; ðx; yÞ 2
F ðp1 pÞ ¼ LHS: Hence, RHS LHS: Since LHS RHS; and RHS
LHS; LHS ¼ RHS:) Thus, F ^ p ¼ F:
1
Since F p : N ! P; and F ^ ¼ F p1 ; F
^ : N ! P: It remains to be showed
^ ^
that F is smooth. Since F p ¼ F; and F is a smooth map, F ^ p is smooth. Since
^ : N ! P is a map, and F
F ^ p is smooth, by Theorem 5.81, F ^ is smooth.
This completes the proof of existence part.
Uniqueness: Let F1 : N ! P be a smooth mapping such that F1 p ¼ F; and let
F2 : N ! P be a smooth mapping such that F2 p ¼ F: We have to show that
F1 ¼ F2 :
Since p : M ! N is onto, p p1 ¼ IdN : Since F1 p ¼ F; F p1 ¼
ðF1 pÞ p1 ¼ F1 ðp p1 Þ ¼ F1 IdN ¼ F1 : Thus, F1 ¼ F p1 : Similarly,
F2 ¼ F p1 : It follows that F1 ¼ F2 : This proves the uniqueness part. h
Theorem 5.83 Let M be an m-dimensional smooth manifold, N be an n-dimen-
sional smooth manifold, and P be a p-dimensional smooth manifold. Let p1 : M !
N be a smooth submersion onto map. Let p2 : M ! P be a smooth submersion onto
map. Let p1 be constant on fibers of p2 ; and p2 be constant on fibers of p1 : Then,
there exists a unique diffeomorphism F : N ! P such that F p1 ¼ p2 :
Proof Existence: We first try to show that p2 p1 1 is a function. For this purpose,
let ðx; yÞ 2 p2 p11 ; and ðx; zÞ 2 p2 p 1
1 : We have to show that y ¼ z:
Since ðx; yÞ 2 p2 p1 1 ; there exists z such that ðx; zÞ 2 p1
1 ; and ðz; yÞ 2 p2 :
1 1
Since ðx; zÞ 2 p1 ; ðz; xÞ 2 p1 : Since ðx; zÞ 2 p2 p1 ; there exists w such that
ðx; wÞ 2 p1 1
1 ; and ðw; zÞ 2 p2 : Since ðx; wÞ 2 p1 ; ðw; xÞ 2 p1 : Since ðw; zÞ 2 p2 ;
p2 ðwÞ ¼ z: Similarly, p2 ðzÞ ¼ y; p1 ðwÞ ¼ x; and p1 ðzÞ ¼ x: Since p1 ðwÞ ¼ x ¼
p1 ðzÞ; and p2 is constant on fibers of p1 ; p2 ðwÞ ¼ p2 ðzÞ: Since z ¼ p2 ðwÞ ¼
p2 ðzÞ ¼ y; y ¼ z: This proves that p2 p1 1 is a function.
Since p1 : M ! N is onto, and p2 : M ! P is onto, dom ðp2 p1 1 Þ ¼ N; and
1 1 1
ran ðp2 p1 Þ P: Since p2 p1 is a function, dom ðp2 p1 Þ ¼ N; and
ran ðp2 p1 1
1 Þ P; we can write p2 p1 : N ! P:
Put F p2 p1 : Since F ¼ p2 p1
1 1
1 ; and p2 p1 : N ! P; F : N ! P:
Now, F p1 ¼ ðp2 p1 1 1
1 Þ p1 ¼ p2 ðp1 p1 Þ: Clearly, p2 ðp1 p1 Þ ¼ p2 :
1
(Reason: Let ðx; yÞ 2 LHS ¼ p2 ðp1 p1 Þ: So there exist u; v such that
ðx; uÞ 2 p1 ; ðu; vÞ 2 p1 1
1 ; and ðv; yÞ 2 p2 : Since ðu; vÞ 2 p1 ; ðv; uÞ 2 p1 : It follows
394 5 Immersions, Submersions, and Embeddings
that p1 ðvÞ ¼ u; and p1 ðxÞ ¼ u; and hence, p1 ðvÞ ¼ p1 ðxÞ: Since p1 ðvÞ ¼ p1 ðxÞ; and
p2 is constant on fibers of p1 ; p2 ðvÞ ¼ p2 ðxÞ: Since ðv; yÞ 2 p2 ; p2 ðvÞ ¼ y: Since
p2 ðvÞ ¼ y; and p2 ðvÞ ¼ p2 ðxÞ; p2 ðxÞ ¼ y; and hence, ðx; yÞ 2 p2 ¼ RHS: Hence,
LHS RHS: Next, let ðx; yÞ 2 RHS ¼ p2 : Since ðx; yÞ 2 p2 and p2 : M ! P;
x 2 M; and y 2 P: Since x 2 M; and p1 : M ! N; there exists u 2 N such that
ðx; uÞ 2 p1 : Since ðx; uÞ 2 p1 ; ðu; xÞ 2 p1 1
1 : Since ðx; uÞ 2 p1 ; ðu; xÞ 2 p1 ; and
ðx; yÞ 2 p2 ; ðx; yÞ 2 p2 ðp11 p1 Þ ¼ LHS: Hence, RHS LHS: Since LHS
RHS; and RHS LHS; LHS ¼ RHS:) Thus, F p1 ¼ p2 :
Since p2 p1 1
1 : N ! P; and F ¼ p2 p1 ; F : N ! P: Now, we want to show
that F is 1–1, that is, p2 p1 is 1–1, that is, if ðx; zÞ 2 p2 p1
1
1 ; and
ðy; zÞ 2 p2 p1
1 , then x ¼ y: For this purpose, let ðx; zÞ 2 p2 p 1
1 ; and ðy; zÞ 2
1
p2 p1 : We have to show that x ¼ y:
Since ðx; zÞ 2 p2 p1 1
1 ; there exists u such that ðx; uÞ 2 p1 ; and ðu; zÞ 2 p2 :
1 1
Since ðx; uÞ 2 p1 ; ðu; xÞ 2 p1 : Since ðy; zÞ 2 p2 p1 ; there exists v such that
ðy; vÞ 2 p1 1
1 ; and ðv; zÞ 2 p2 : Since ðy; vÞ 2 p1 ; ðv; yÞ 2 p1 : Since ðv; yÞ 2 p1 ;
p1 ðvÞ ¼ y: Similarly, p1 ðuÞ ¼ x; p2 ðuÞ ¼ z; and p2 ðvÞ ¼ z: Since p2 ðuÞ ¼ z ¼
p2 ðvÞ; and p1 is constant on fibers of p2 ; p1 ðuÞ ¼ p1 ðvÞ: Since x ¼ p1 ðuÞ ¼
p1 ðvÞ ¼ y; x ¼ y: Thus, we have shown that F : N ! P is 1–1.
Now, we want to show that F : N ! P is onto. For this purpose, let us take any
c 2 P: We have to find y 2 N such that ðy; cÞ 2 Fð¼ p2 p1 1 Þ:
Since p2 : M ! P is onto, and c 2 P; there exists z 2 M such that ðz; cÞ 2 p2 :
Since z 2 M; and p1 : M ! N; there exists y 2 N such that ðz; yÞ 2 p1 ; and hence,
ðy; zÞ 2 p1 1 1
1 : Since ðy; zÞ 2 p1 ; and ðz; cÞ 2 p2 ; ðy; cÞ 2 p2 p1 : Thus, we have
shown that F : N ! P is onto.
Now, we want to show that F : N ! P is smooth. Since F p1 ¼ p2 ; and p2 is
smooth, F p1 is smooth. Since p1 : M ! N is a smooth submersion onto map,
and F p1 is smooth, by Theorem 25, F is smooth.
Lastly, we have to show that F 1 : P ! N is smooth. Since F p1 ¼ p2 ; and
F is 1–1 onto, F 1 p2 ¼ p1 : Since F 1 p2 ¼ p1 ; and p1 is smooth, F 1 p2 is
smooth. Since p2 : M ! P is a smooth submersion onto map, and F 1 p2
is smooth, by Theorem 5.81, F 1 is smooth. Thus, we have shown that F : N ! P
is a diffeomorphism such that F p1 ¼ p2 :
This completes the proof of existence part.
Uniqueness: Let F1 : N ! P be a smooth mapping such that F1 p1 ¼ p2 ; and
let F2 : N ! P be a smooth mapping such that F2 p1 ¼ p2 : We have to show that
F1 ¼ F2 :
Since p1 : M ! N is onto, p1 p1 1 ¼ IdN : Since F1 p1 ¼ p2 ; p2 p1 ¼
1
1 1 1
ðF1 p1 Þ p1 ¼ F1 ðp1 p1 Þ ¼ F1 IdN ¼ F1 : Thus, F1 ¼ p2 p1 : Simi-
larly, F2 ¼ p2 p1 1 : It follows that F1 ¼ F2 : This proves the uniqueness part. h
5.8 Properly Embedded 395
Note 5.84 Before going ahead, let us recall the definition of open submanifold of M:
Let M be an m-dimensional smooth manifold, whose topology is O; and differential
structure is A: Let G be a nonempty open subset of M. Let O1 be the subspace
topology over G inherited from M, that is, O1 ¼ fG1 : G1 G; and G1 2 Og:
Since M is an m-dimensional smooth manifold, M is a Hausdorff space. Since
M is a Hausdorff space, and G is a subspace of M, G with the subspace topology is a
Hausdorff space. Since M is an m-dimensional smooth manifold, M is a second
countable space. Since M is a second countable space, and G is a subspace of M,
G with the subspace topology is a second countable space. Put
For (2): Let us take any ðU; uÞ; ðV; wÞ 2 AG : We have to show that ðU; uÞ
andðV; wÞ are C1 compatible.
Since ðU; uÞ 2 AG , by the definition of AG ; ðU; uÞ 2 A: Similarly, ðV; wÞ 2 A:
Since ðU; uÞ 2 A; ðV; wÞ 2 A; and A is a differential structure, ðU; uÞ and ðV; wÞ
are C 1 compatible. Thus, we have shown that AG is an atlas on G. Hence, AG
determines a unique smooth structure on G. Thus, the open subset G of smooth
manifold M becomes an m-dimensional smooth manifold. Here, G is called an open
submanifold of M:
Note 5.85 Let M be an m-dimensional smooth manifold with smooth structure A:
Let G be a nonempty open subset of M. In the open submanifold G of M; the
smooth structure of G is determined by the atlas fðU; uÞ : ðU; uÞ 2 A; and U Gg
of G. Thus, G becomes an m-dimensional smooth manifold.
396 5 Immersions, Submersions, and Embeddings
We shall try to show that the inclusion map i : G ! M (that is, for every x in G,
iðxÞ x) is a smooth embedding.
Since i : G ! M is the identity mapping IdG ; and IdG is smooth, i : G ! M is
smooth.
Now, we want to show that i : G ! M is a smooth immersion, that is, for every
p 2 G; the linear map dip : Tp G ! TiðpÞ M is 1–1. For this purpose, let us fix any
p 2 G: We have to show that the linear map dip : Tp G ! TiðpÞ M is 1–1. Clearly,
Tp G ¼ Tp M; and TiðpÞ M ¼ Tp M: So we have to show that the linear map dip :
Tp M ! Tp M is 1–1. Here, i ¼ IdG ; so we have to show that the linear map
dðIdG Þp : Tp M ! Tp M is 1–1.
Since dðIdG Þp ¼ IdTp G ¼ IdTp M ; and IdTp M : Tp M ! Tp M is 1–1, dðIdG Þp :
Tp M ! Tp M is 1–1. Thus, we have shown that i : G ! M is a smooth immersion.
Also, clearly i : G ! M is 1–1.
Since G is open in M, i : G ! M is an open map. Since i : G ! M is a 1–1
immersion, and open, by Note 5.69, i : G ! M is a smooth embedding.
Conclusion: Let M be an m-dimensional smooth manifold. Let G be a nonempty
subset of M.
If G is open in M, then there exists an m-dimensional smooth structure on G such
that the inclusion map i : G ! M is a smooth embedding. Now, we want to prove
its converse part:
Let M be an m-dimensional smooth manifold. Let G be a nonempty subset of
M. Let G be endowed with the subspace topology of M.
Suppose there exists an m-dimensional smooth structure A on G such that the
inclusion map i : G ! M is a smooth embedding. We shall try to show that G is
open in M.
Since i : G ! M is a smooth embedding, i : G ! M is a smooth immersion.
Since G; M are m-dimensional smooth manifolds, and i : G ! M is a smooth
immersion, by Theorem 5.46, i : G ! M is a local diffeomorphism. Since i : G !
M is a local diffeomorphism, by Note 5.38, i : G ! M is an open map. Since
i : G ! M is an open map, Gð¼ iðGÞÞ is an open subset of M.
Definition Let M be an m-dimensional smooth manifold. Let S be a nonempty
subset of M. Let S be endowed with the subspace topology of M. Let k be a
nonnegative integer.
If there exists an (m − k)-dimensional smooth structure A on S; such that the
inclusion map i : S ! M is a smooth embedding, then we say that S is an embedded
submanifold of M with codimension k. Here, we also say that M is an ambient
manifold for S.
From the Note 5.85, we can write:
Let M be an m-dimensional smooth manifold. Let S be a nonempty subset of
M. Let S be endowed with the subspace topology of M. S is open in M if and only if
S is an embedded submanifold of M with codimension 0. In this sense, we say that
the open submanifolds of M are exactly the embedded submanifolds of M with
codimension 0.
5.8 Properly Embedded 397
From the Note 5.86, we see that AF is a collection of coordinate charts of F(N).
Now, we shall try to show that AF is an n-dimensional atlas on F(N), that is,
1. [fFðUÞ : ðU; uÞ 2 Bg ¼ FðNÞ;
2. all pairs of members in AF are C1 compatible.
Clearly,
2 3
1 0 0 0
60 1 0 07
6 .. 7
6 7
6 . 7
6 7
60 0 0 17
rank 6 7 ¼ m;
60 0 0 07
6 7
60 0 0 07
6 .. 7
4 . 5
0 0 0 0
For this purpose, let us fix any q 2 G: We have to show that the linear map
dFq : Tq M ! TFðqÞ ðM NÞ is 1–1. Since q 2 Gð MÞ; and M is an m-dimensional
smooth manifold, there exists an admissible coordinate chart ðU; uÞ of M such that
q 2 U: Since q 2 U; f ðqÞ 2 N: Since f ðqÞ 2 N; and N is an n-dimensional smooth
manifold, there exists an admissible coordinate chart ðV; wÞ of N such that f ðqÞ 2
V: For every r 2 U;
o o o
; ; . . .; m
ox1 ox2 r ox
r r
Clearly,
2 3
1 0 0
60 7
6 1 0 7
6. .. .. 7
6. 7
6. . . 7
6 7
60 0 7
6 1 7
6 oðp ðw F u1 ÞÞ oðp1 ðw F u1 ÞÞ 1
oð p 1 ð w F u Þ Þ 7
6 1
ðuðqÞÞ ðuðqÞÞ ð uðqÞ Þ 7
rank6
6 ox1 ox2 oxm
7 ¼ m;
7
6 1 1 1 7
6 oð p 2 ð w F u Þ Þ oð p 2 ð w F u Þ Þ oð p 2 ð w F u Þ Þ 7
6 ðuðqÞÞ ðuðqÞÞ ðuðqÞÞ 7
6 ox1 ox2 oxm 7
6 7
6. .. .. 7
6 .. . . 7
6 7
4 oðp ðw F u1 ÞÞ oðpn ðw F u1 ÞÞ 1
oð p n ð w F u Þ Þ 5
n
ðuðqÞÞ ðuðqÞÞ ð uðqÞ Þ
ox1 ox2 oxm
j2 w u1 uððU\V Þ\SÞ ðj1 Þ1 : ðj1 uÞððU \ V Þ \ SÞ
! ðj2 wÞððU \ V Þ \ SÞ
is smooth, that is, ðw u1 ÞuððU\VÞ\SÞ j1 is smooth. Here, U \ V 6¼ ;: Since
U \ V 6¼ ;; and ðU; uÞ; ðV; wÞ are admissible coordinate charts of M; w u1 :
410 5 Immersions, Submersions, and Embeddings
uðU \ VÞ ! wðU \ VÞ is smooth, and hence, its restriction ðw u1 ÞuððU\VÞ\SÞ
is smooth. Since ðw u1 ÞuððU\VÞ\SÞ is smooth, and j1 is smooth, their
composite ðw u1 ÞuððU\VÞ\SÞ j1 is smooth.
For b: Since S has the subspace topology inherited from M; i : S ! M is a
topological embedding.
For c: Let us take any p 2 S: We have to show that the linear map dip : Tp S !
TiðpÞ M is 1–1, that is, the linear map dip : Tp S ! Tp M is 1–1.
Since p 2 S; there exists ðU \ S; j ðujU\S ÞÞ 2 A; where ðU; uÞ is an admis-
sible coordinate chart of M, and p 2 U \ S: Here, ðU; uÞ is an admissible coordi-
nate chart of M satisfying iðpÞ 2 U: For every q 2 U \ S;
!
o o
;
ox1 q ox2 q
For every r 2 U;
o o o o o
; ; ; ;
oy1 r oy2 r oy3 r oy4 r oy5 r
Also clearly, j1 : R2 ! R2 fa3 g fa4 g fa5 g is such that for every ðx1 ; x2 Þ in
R2 ; j1 ðx1 ; x2 Þ ¼ ðx1 ; x2 ; a3 ; a4 ; a5 Þ: Hence, for every q 2 U \ S; the matrix rep-
resentation of linear map diq is the following 5 × 2 matrix:
5.9 Regular Level Sets 411
2 3
1 0
60 17
6 7
60 07
6 7:
40 05
0 0
Since the
2 3
1 0
60 17
6 7
rank 6
60 077
40 05
0 0
Now, since
! !!
o o
diq ; diq
ox1 q ox2 q
By the constant rank Theorem 5.48, for every p 2 U1 ðcÞðthat is; UðpÞ ¼ cÞ
there exist admissible coordinate chart ðU; uÞ in M satisfying p 2 U; and admis-
sible coordinate chart ðV; wÞ in N satisfying UðpÞ 2 V; such that UðUÞ V; and
for every ðx1 ; x2 ; x3 Þ in uðUÞ;
w U u1 ðx1 ; x2 ; x3 Þ ¼ ðx1 ; x2 ; 0; 0Þ:
that is,
u U1 ðcÞ \ U ¼ fðx1 ; x2 ; x3 Þ 2 uðUÞ : x1 ¼ ðp1 wÞðcÞ; x2 ¼ ðp2 wÞðcÞg:
Hence, x1 ¼ ðp1 wÞðcÞ; and x2 ¼ ðp2 wÞðcÞ: Thus, U1 ðcÞ \ U u1 ðfðx1 ;
x2 ; x3 Þ 2 uðUÞ : x1 ¼ ðp1 wÞðcÞ; x2 ¼ ðp2 wÞðcÞgÞ: Next, let q 2 u1 ðfðx1 ; x2 ;
x3 Þ 2 uðUÞ : x1 ¼ ðp1 wÞðcÞ; x2 ¼ ðp2 wÞðcÞgÞ: We have to show that q 2 U1
ðcÞ \ U; that is, UðqÞ ¼ c; and q 2 U: Since q 2 u1 ðfðx1 ; x2 ; x3 Þ 2 uðUÞ : x1 ¼
ðp1 wÞðcÞ; x2 ¼ ðp2 wÞðcÞgÞ U; q 2 U:
Since uðqÞ 2 uðUÞ; ððp1 wÞðcÞ; ðp2 wÞðcÞ; ðp3 wÞðcÞ; ðp4 wÞðcÞÞ ¼ wðcÞ
¼ wðUðqÞÞ ¼ ðw U u1 ÞðuðqÞÞ ¼ ððp1 uÞðqÞ; ðp2 uÞðqÞ; 0; 0Þ; and hence,
ðp3 wÞðcÞ ¼ 0 ¼ ðp4 wÞðcÞ: Since q 2 u1 ðfðx1 ; x2 ; x3 Þ 2 uðUÞ : x1 ¼ ðp1 wÞ
ðcÞ; x2 ¼ ðp2 wÞðcÞgÞ; uðqÞ 2 fðx1 ; x2 ; x3 Þ 2 uðUÞ : x1 ¼ ðp1 wÞðcÞ; x2 ¼ ðp2
wÞðcÞg; and hence, ðp1 uÞðqÞ ¼ ðp1 wÞðcÞ; and ðp2 uÞðqÞ ¼ ðp2 wÞðcÞ:
Since ðp1 uÞðqÞ ¼ ðp1 wÞðcÞ; ðp2 uÞðqÞ ¼ ðp2 wÞðcÞ; ðp3 wÞ ðcÞ ¼ 0 ¼
ðp4 wÞðcÞ; and wðUðqÞÞ ¼ ððp1 uÞðqÞ; ðp2 uÞðqÞ; 0; 0Þ; wðUðqÞÞ ¼ ððp1 wÞ
ðcÞ; ðp2 wÞ ðcÞ; 0; 0Þ ¼ ððp1 wÞðcÞ; ðp2 wÞðcÞ; ðp3 wÞ ðcÞ; ðp4 wÞðcÞÞ ¼
wðcÞ: Since wðUðqÞÞ ¼ wðcÞ; UðqÞ ¼ c:
Thus, uðU1 ðcÞ \ UÞ ¼ fðx1 ; x2 ; x3 Þ 2 uðUÞ : x1 ¼ ðp1 wÞðcÞ; x2 ¼ ðp2 wÞ
ðcÞg: Hence, the level set U1 ðcÞ satisfies the local 2-slice condition. Now, by Note
5.97, if U1 ðcÞ has the subspace topology inherited from M; then U1 ðcÞ is an
embedded submanifold of M. Since U : M ! N is a smooth map, U : M ! N is
continuous. Since U : M ! N is continuous, and fcg is a closed subset of N;
U1 ðcÞ is a closed subset of M. Since U1 ðcÞ is a closed subset of M, and U1 ðcÞ is
414 5 Immersions, Submersions, and Embeddings
If U1 ðcÞ G; then we say that c is a regular value of U; and U1 ðcÞ is a
regular value set. If c is not a regular value of U; then we say that c is a critical
value of U: If c is a regular value of U; then U1 ðcÞ is called a regular level set.
Clearly, if U1 ðcÞ ¼ ;; then c is a regular value of U:
Note 5.104 Let M be an m-dimensional smooth manifold, N be an n-dimensional
smooth manifold, and U : M ! N be a smooth map. Letc 2 N: Let U1 ðcÞ be a
nonempty regular level set. We shall try to show that U1 ðcÞ is a properly
embedded submanifold of M, and its codimension is n.
Let G be the set of all regular points of U: We know that G is an open subset of
M. Let O1 be the subspace topology over G inherited from M, that is, O1 ¼
fG1 : G1 G; and G1 2 Og: Let A be the differential structure of M. Put
Since ðUjG Þ1 ðcÞ ¼ U1 ðcÞ; and the level set ðUjG Þ1 ðcÞ is a properly embed-
ded submanifold of G with its codimension n, U1 ðcÞ is a properly embedded
submanifold of G with its codimension n. Since U1 ðcÞ is an embedded subman-
ifold of G, the inclusion map i1 : U1 ðcÞ ! G is a smooth embedding. Since G is
an embedded submanifold of M, the inclusion map i2 : G ! M is a smooth
embedding. Since i1 : U1 ðcÞ ! G is a smooth embedding, and i2 : G ! M is a
smooth embedding, their composite i2 i1 : U1 ðcÞ ! M is a smooth embedding.
Since i2 i1 ¼ i; where idenotes the inclusion map from U1 ðcÞ to M; and i2 i1 :
U1 ðcÞ ! M is a smooth embedding, the inclusion map i : U1 ðcÞ ! M is a
smooth embedding. This shows that U1 ðcÞ is an embedded submanifold of M.
Now, we want to show that U1 ðcÞ is a properly embedded submanifold of
M. For this purpose, let us take any compact set K in M. We have to show that
ði2 i1 Þ1 ðKÞ is compact in U1 ðcÞ: Here, ði2 i1 Þ1 ðKÞ ¼ ðði1 Þ1 ði2 Þ1 ÞðKÞ ¼
ði1 Þ1 ðK \ GÞ ¼ ðK \ GÞ \ U1 ðcÞ ¼ K \ U1 ðcÞ: Since K is compact in M, and
U1 ðcÞ is closed in M, K \ U1 ðcÞ is compact in M. Since K \ U1 ðcÞ is compact
in M, and K \ U1 ðcÞ U1 ðcÞ; K \ U1 ðcÞð¼ ði2 i1 Þ1 ðKÞÞ is compact in
U1 ðcÞ; and hence ði2 i1 Þ1 ðKÞ is compact in U1 ðcÞ:
Thus, U1 ðcÞ is a properly embedded submanifold of M with codimension n.
O fFðVÞ : V is open in N g:
F(p) in S such that ι(U) is open in M, and the restriction ijU : U ! iðUÞ is a
diffeomorphism, and hence, ðijU Þ1 : iðUÞ ! U is smooth. Since U is an open
neighborhood of F(p) in S, and F : N ! S is continuous, there exists an open
neighborhood V of p in N such that F(V) ⊂ U. Since V is open in N, V is an open
submanifold of N, and hence, V is an embedded submanifold of N. Since V is an
embedded submanifold of N, V is a smooth submanifold of N. Since V is a smooth
submanifold of N, and F : N ! M is a smooth map, by Note 5.111, FjV : V ! M is
a smooth map. Since FjV : V ! M is a smooth map, ðijU Þ1 : iðUÞ ! U is smooth,
and ðFjV ÞðVÞ ¼ FðVÞ U ¼ iðUÞ; their composite ððijU Þ1 Þ ðFjV Þ : V ! U is
smooth. Since ððijU Þ1 Þ ðFjV Þ : V ! U is smooth, and ððijU Þ1 Þ ðFjV Þ ¼ FjV ;
FjV : V ! U is smooth. Since FjV : V ! U is smooth, and V is an open neigh-
borhood of p in N, F : N ! S is smooth at the point p. It follows that F : N ! S is
smooth.
Note 5.113 Let M be an m-dimensional smooth manifold, N be an n-dimensional
smooth manifold, and S be a nonempty subset of M. Let S be an embedded sub-
manifold of M with codimension k. Let F : N ! M be a smooth map, and FðNÞ
S: Let F : N ! S be continuous. We shall try to show F : N ! S is smooth.
Since S is an embedded submanifold of M with codimension k, by Note 5.105,
S is a smooth submanifold of M with codimension k. Since F : N ! M is a smooth
map, F : N ! M is continuous. Since S is an embedded submanifold of M, the
topology over S is the subspace topology inherited from M. Since F : N ! M is
continuous, FðNÞ S M; and the topology over S is the subspace topology
inherited from M, F : N ! S is continuous. It follows, from Note 5.112, that F :
N ! S is smooth.
Note 5.114 Let M be an m-dimensional smooth manifold. Let S be a nonempty
subset of M. Suppose that S satisfies local k-slice condition.
Since S satisfies local k-slice condition, by Note 5.97, S is an embedded sub-
manifold of M. Since S is an embedded submanifold of M, the topology over S is
the subspace topology inherited from M, and there exists a smooth structure A on
S with respect to which S becomes a k-dimensional smooth manifold such that the
inclusion map i : S ! M is a smooth embedding.
Now, we want to show that such a differential structure A is unique. For this
purpose, let A1 be a smooth structure on S with respect to which S becomes a k-
dimensional smooth manifold such that the inclusion map i : S ! M is a smooth
embedding. Next, let A2 be a smooth structure on S with respect to which
S becomes a k-dimensional smooth manifold such that the inclusion map i : S ! M
is a smooth embedding. We have to prove that A1 ¼ A2 :
Let O be the topology over M; and let B be the smooth structure of M: Since the
inclusion map i : S ! M is a smooth embedding from smooth manifold ðS; A1 Þ to
ðM; BÞ; the inclusion map i : S ! M is a smooth map from smooth manifold
ðS; A1 Þ to ðM; BÞ: Since the inclusion map i : S ! M is a smooth map from smooth
manifold ðS; A1 Þ to ðM; BÞ; ðS; A2 Þ is an embedded submanifold of M; and i :
S ! S is continuous, by Lemma 5.113, i : ðS; A1 Þ ! ðS; A2 Þ is smooth. Similarly,
5.10 Smooth Submanifolds 421
Here,
!
o o
; . . .; k
ox1 p ox p
for every t 2 ð0 e; 0 þ eÞ, we have ðtv1 ; . . .; tvk Þ 2 uðUÞ, and hence, for every
t 2 ðe; eÞ, we have u1 ðtv1 ; . . .; tvk Þ 2 U. Now, let us define c : ðe; eÞ !
Uð S MÞ as follows: For every t 2 ðe; eÞ; cðtÞ u1 ðtv1 ; . . .; tvk Þ.
Here, ðU; uÞ is an admissible coordinate chart of S, so U is open in S. Here,
cð0Þ ¼ u1 ð0v1 ; . . .; 0vk Þ ¼ u1 ð0Þ ¼ u1 ðuðpÞÞ ¼ p. Thus, cð0Þ ¼ p. We shall
try to show that γ as a function from ðe; eÞ to smooth manifold S is smooth.
Since ðU; uÞ is an admissible coordinate chart of S, φ is a diffeomorphism from
U onto uðUÞ, and hence, u1 is smooth. Since u1 is smooth, and t 7! ðtv1 ; . . .; tvk Þ
is smooth, their composite t 7! u1 ðtv1 ; . . .; tvk Þð¼ cðtÞÞ is smooth, and hence, γ as
a mapping from ðe; eÞ to S is smooth.
Now, we want to show that γ is smooth as a map from smooth manifold ðe; eÞ
to smooth manifold M. For this purpose, let us take any t0 2 ðe; eÞ. Here,
cðt0 Þ ¼ u1 ðt0 v1 ; . . .; t0 vk Þ 2 U.
Since S is an embedded submanifold of M, there exists a smooth structure A on
S such that the inclusion map i : S ! M is a smooth embedding (i.e., i : S ! M is
smooth, and smooth immersion.). Let ðW; wÞ be any admissible coordinate chart of
M such that cðt0 Þ 2 W. We have to show that w c ðIdðe;eÞ Þ1 is smooth.
Since i : S ! M is smooth, ðU; uÞ is an admissible coordinate chart of S satis-
fying cðt0 Þ 2 U, and ðW; wÞ is any admissible coordinate chart of M such that
ðiðcðt0 ÞÞ ¼Þcðt0 Þ 2 W, w u1 is smooth. Further, since t 7! ðtv1 ; . . .; tvk Þ is
smooth, t 7! ðw u1 Þðtv1 ; . . .; tvk Þ is smooth. For every t 2 ðe; eÞ,
1 1
w c Idðe;eÞ ðtÞ ¼ ðw cÞ Idðe;eÞ ðtÞ ¼ ðw cÞðtÞ ¼ wðcðtÞÞ
¼ w u1 tv1 ; . . .; tvk ¼ w u1 tv1 ; . . .; tvk ;
so
d ð pi ð u c Þ Þ
ð0Þ ¼ vi :
dt
Hence, c0 ð0Þ ¼ v. h
424 5 Immersions, Submersions, and Embeddings
Further, we may assume uðUÞ ¼ Cek ð0Þ, and wðVÞ ¼ Cem ð0Þ for some e [ 0. Since
ðU; uÞ is an admissible coordinate chart in S satisfying p 2 U, U is an open
neighborhood of p in S, and hence, there exists an open neighborhood W of p in
M such that W \ S ¼ U. Since W is an open neighborhood of p in M, and V is an
open neighborhood of p in M, their intersection V \ W is an open neighborhood of
p in M. Since V \ Wð VÞ is an open neighborhood of p in M, and ðV; wÞ is an
admissible coordinate chart in M satisfying p 2 V, ðV \ W; wjV\W Þ is an admissible
coordinate chart in M satisfying p 2 V \ W. It suffices to show that
08 0 10 1
1 <
ðV \ W Þ \ S ¼ wjV \ W @ @x1 ; . . .; xk ; 0; . . .; 0A:@x1 ; . . .; xk ; 0; . . .; 0A
: |fflfflffl{zfflfflffl} |fflfflffl{zfflfflffl}
mk mk
2 wjV \ W ðV \ W Þ ;
5.11 Tangent Space to a Submanifold 425
that is,
80 10 1
<
wjV \ W ððV \ W Þ \ SÞ ¼ @x1 ; . . .; xk ; 0; . . .; 0A:@x1 ; . . .; xk ; 0; . . .; 0 A
: |fflfflffl{zfflfflffl} |fflfflffl{zfflfflffl}
mk mk
2 wjV \ W ðV \ W Þ ;
that is,
80 1 0 1 9
< =
wððV \ W Þ \ SÞ ¼ @x1 ; . . .; xk ; 0; . . .; 0A : @x1 ; . . .; xk ; 0; . . .; 0A 2 wðV \ W Þ ;
: |fflfflffl{zfflfflffl} |fflfflffl{zfflfflffl} ;
mk mk
that is,
80 1 0 1 9
< =
wðU \ V Þ ¼ @x1 ; . . .; xk ; 0; . . .; 0 A : @x1 ; . . .; xk ; 0; . . .; 0 A 2 wðV \ W Þ :
: |fflfflffl{zfflfflffl} |fflfflffl{zfflfflffl} ;
mk mk
Thus,
0 1 80 1
<
LHS ¼ wðqÞ ¼ @y1 ; . . .; yk ; 0; . . .; 0 A 2 @x1 ; . . .; xk ; 0; . . .; 0A :
|fflfflffl{zfflfflffl} : |fflfflffl{zfflfflffl}
mk mk
0 1 9
=
@x1 ; . . .; xk ; 0; . . .; 0A 2 wðV \ W Þ ¼ RHS:
|fflfflffl{zfflfflffl} ;
mk
426 5 Immersions, Submersions, and Embeddings
where q 2 V \ W. We have to show that wðqÞ 2 wðU \ VÞ. It suffices to show that
q 2 U. Here, q 2 V \ W, so
0 1
@y1 ; . . .; yk ; 0; . . .; 0 A ¼ wðqÞ 2 wðVÞ ¼ C m ð0Þ
|fflfflffl{zfflfflffl} e
mk
¼ ðe; eÞ ðe; eÞ ðe; eÞ ðe; eÞ ;
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
k mk
and hence,
that is,
that is,
80 1 0 1 9
< =
wðV \ SÞ ¼ @x1 ; . . .; xk ; 0; . . .; 0A : @x1 ; . . .; xk ; 0; . . .; 0A 2 wðVÞ :
: |fflfflffl{zfflfflffl} |fflfflffl{zfflfflffl} ;
mk mk
is a basis of ðdiq ÞðTq SÞ, where i : S ! M denotes the inclusion map of S. Now,
since A is unique, A ¼ B. Since p 2 V,
!
o o o o
; . . .; k ; kþ1 ; . . . m
oy1 p oy p oy p oy p
428 5 Immersions, Submersions, and Embeddings
is a basis of the real linear space Tp M. Now, since v 2 Tp M, there exist real
numbers v1 ; . . .; vk ; vkþ1 ; . . .; vm such that
! ! ! !
o o o o
v¼v 1
þ þ v k
þv kþ1
þ þ v m
:
oy1 p oyk p oykþ1 p oym p
It follows that
0 1
ðpkþ1 wÞðxÞ ¼ pkþ1 ðwðxÞÞ ¼ pkþ1 @x1 ; . . .; xk ; 0; . . .; 0A ¼ 0:
|fflfflffl{zfflfflffl}
mk
Now, since
! !!
o o
ðpkþ1 wÞjW ¼ d w1 wðpÞ ðpkþ1 wÞjW
oy p
i oy wðpÞ
i
! !
o o
¼ ðpkþ1 wÞjW w1 ¼ ðpkþ1 Þ
oyi wðpÞ oyi wðpÞ
0 if i 6¼ k þ 1
¼
1 if i ¼ k þ 1;
0 ¼ v1 ð0Þ þ þ vk ð0Þ þ vkþ1 ð1Þ þ þ vm ð0Þ ðpÞ ¼ vkþ1 ð1ðpÞÞ ¼ vkþ1 1
¼ vkþ1 :
Thus, vkþ1 ¼ 0.
Similarly, vkþ2 ¼ 0; ; vm ¼ 0. Hence,
! ! ! !
o o o o
v¼v 1
þ þ v k
þ0 þ þ 0
oy1 p oyk p oykþ1 p oym p
! ! !
o o o o
¼ v1 þ þ v k
2 span ; . . .; k ¼ dip Tp S :
oy p
1 oy p
k oy p
1 oy p
Thus, v 2 ðdip ÞðTp SÞ. Now, since we do not distinguish between ðdip ÞðTp SÞ and
Tp S, we can write v 2 Tp S. h
Chapter 6
Sard’s Theorem
The root of Sard’s theorem lies in real analysis. It is one of the deep results in real
analysis. How this result can be generalized into the realm of smooth manifold
theory is only a later development. Its importance in manifold theory is for a
definite reason. Upon applying Sard’s theorem, Whitney was able to prove a
startling property about smooth manifolds: For every smooth manifold, ambient
space can be constructed, etc. For pedagogical reasons, we have given its proof in
step-by-step manner using Taylor’s inequality. Largely, this chapter is self-
contained.
For this purpose, let us fix any real c. Since A is a nonempty compact subset of
R2 ; clearly fs : ðc; sÞ 2 Ag is a compact subset of R: (See the proof of Lemma 6.1)
Let us define a function fc : fs : ðc; sÞ 2 Ag ! R as follows: for every s 2
fs : ðc; sÞ 2 Ag; fc ðsÞ f ðc; sÞ: Since fc is the composite of continuous function
s 7! ðc; sÞ; and continuous function f : A ! R; fc : fs : ðc; sÞ 2 Ag ! R is a con-
tinuous function. Also, the domain of fc ; that is, fs : ðc; sÞ 2 Ag is compact. It
follows, by Lemma 6.3, that l2 ðfðx; fc ðxÞÞ : x 2 fs : ðc; sÞ 2 AggÞ ¼ 0: Now, it
suffices to show that fðs; tÞ : ðc; s; tÞ 2 fðx; y; f ðx; yÞÞ : ðx; yÞ 2 Agg ¼ fðx; fc ðxÞÞ :
x 2 fs : ðc; sÞ 2 Agg:
Let us take any ðg; 1Þ 2 LHS. It follows that ðc; g; 1Þ 2 fðx; y; f ðx; yÞÞ : ðx; yÞ
2 Ag; and hence, ðc; gÞ 2 A; and 1 ¼ f ðc; gÞ ¼ fc ðgÞ: Since ðc; gÞ 2 A; g 2
fs : ðc; sÞ 2 Ag; and hence, ðg; 1Þ ¼ ðg; fc ðgÞÞ 2 RHS: Thus, LHS RHS: Next,
let us take any ðg; 1Þ 2 RHS: It follows that there exists a real x such that ðg; 1Þ ¼
ðx; fc ðxÞÞ: Since ðg; 1Þ ¼ ðx; fc ðxÞÞ; g ¼ x; and 1 ¼ fc ðxÞ ¼ f ðc; xÞ: Since 1 ¼ f ðc; xÞ;
and f : A ! R; ðc; xÞ 2 A; and hence, ðc; g; 1Þ ¼ ðc; x; 1Þ ¼ ðc; x; f ðc; xÞÞ 2
fðx; y; f ðx; yÞÞ : ðx; yÞ 2 Ag: Since ðc; g; 1Þ 2 fðx; y; f ðx; yÞÞ : ðx; yÞ 2 Ag; ðg; 1Þ 2
LHS: Thus, RHS LHS: Since RHS LHS; and LHS RHS; LHS ¼ RHS: h
Note 6.5 The Lemma 6.4 can be generalized as follows:
Let n 2 f2; 3; 4; . . .g: Let A be a nonempty compact subset of Rn1 : Let f :
A ! R be continuous. For every k ¼ 1; . . .; n; let lk denote the k-dimensional
measure over Rk : Then, ln ðfðx; f ðxÞÞ : x 2 AgÞ ¼ 0.
Lemma 6.6 Let X be a second countable, locally compact Hausdorff space. Then,
there exists a sequence fKn g of compact subsets of X such that
1. fKn g is a cover of X;
2. K1 ðK2 Þo K2 ðK3 Þo K3 ðK4 Þo K4 :
( )!
[
1
0 ln ðfðx; f ð xÞÞ : x 2 AgÞ ln ðx; f ð xÞÞ : x 2 Kn
n¼1
!
[
1 X
1
¼ ln fðx; f ð xÞÞ : x 2 Kn g ln ðfðx; f ð xÞÞ : x 2 Kn gÞ
n¼1 n¼1
X
1 n o X1
¼ ln x; f jKn ð xÞ : x 2 Kn ¼ 0 ¼ 0:
n¼1 n¼1
A4 A1 A2
aþS¼ x; y; x y : ðx; yÞ 2 R2 :
A1 A3 A3
such that
6.1 Measure Zero in Manifolds 439
ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
p1 epðnÞ ; p1 þ epðnÞ p2 epðnÞ ; p2 þ epðnÞ
2 2 2 2
ðnÞ 1 ðnÞ 1
p3 epðnÞ ; p3 þ epðnÞ : n ¼ 1; 2; 3; . . .
2 2
and hence,
1
ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
A¼A\ p1 epðnÞ ; p1 þ epðnÞ p2 epðnÞ ; p2 þ epðnÞ
[
n¼1 2 2 2 2
ðnÞ 1 ðnÞ 1
p3 epðnÞ ; p3 þ epðnÞ
2 2
1 ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
¼ [ A \ p1 epðnÞ ; p1 þ epðnÞ p2 epðnÞ ; p2 þ epðnÞ
n¼1 2 2 2 2
ðnÞ 1 ðnÞ 1
p3 epðnÞ ; p3 þ epðnÞ :
2 2
Hence,
1 ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
FðAÞ ¼ F [ A \ p1 epðnÞ ; p1 þ epðnÞ p2 epðnÞ ; p2 þ epðnÞ
n¼1 2 2 2 2
ðnÞ 1 ðnÞ 1
p3 epðnÞ ; p3 þ epðnÞ
2 2
1 ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
¼ [ F A \ p1 epðnÞ ; p1 þ epðnÞ p2 epðnÞ ; p2 þ epðnÞ
n¼1 2 2 2 2
ðnÞ 1 ðnÞ 1
p3 epðnÞ ; p3 þ epðnÞ :
2 2
440 6 Sard’s Theorem
It follows that
1 ðnÞ 1 ðnÞ 1
0 l3 ðFðAÞÞ ¼ l3 [ F A \ p1 epðnÞ ; p1 þ epðnÞ
n¼1 2 2
ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
p2 epðnÞ ; p2 þ epðnÞ p3 epðnÞ ; p3 þ epðnÞ
2 2 2 2
X1
ðnÞ 1 ðnÞ 1
l3 F A \ p1 epðnÞ ; p1 þ epðnÞ
n¼1
2 2
ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
p2 epðnÞ ; p2 þ epðnÞ p3 epðnÞ ; p3 þ epðnÞ
2 2 2 2
X1
ðnÞ 1 ðnÞ 1
¼ l3 F A \ p1 epðnÞ ; p1 þ epðnÞ
n¼1
2 2
ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
p2 epðnÞ ; p2 þ epðnÞ p3 epðnÞ ; p3 þ epðnÞ
2 2 2 2
is continuous. Since
is a compact subset of the domain of the continuous function ðFpðnÞ Þ0 , its image
under ðFpðnÞ Þ0 is compact in the metric space LðR3 ; R3 Þ and hence is bounded. Thus,
there exists a positive real number C such that
n 0 3 3 3 3
ðnÞ ðnÞ ðnÞ ðnÞ
sup FpðnÞ ð xÞ : x 2 p1 epðnÞ ; p1 þ epðnÞ p2 epðnÞ ; p2 þ epðnÞ
4 4 4 4
ðnÞ 3 ðnÞ 3
p3 epðnÞ ; p3 þ epðnÞ C:
4 4
Since
ðnÞ 3 ðnÞ 3 ðnÞ 3 ðnÞ 3
FpðnÞ : p1 epðnÞ ; p1 þ epðnÞ p2 epðnÞ ; p2 þ epðnÞ
4 4 4 4
ðnÞ 3 ðnÞ 3
p3 epðnÞ ; p3 þ epðnÞ
4 4
! R3
Since
ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
A\ p1 epðnÞ ; p1 þ epðnÞ p2 epðnÞ ; p2 þ epðnÞ
2 2 2 2
ðnÞ 1 ðnÞ 1
p3 epðnÞ ; p3 þ epðnÞ A;
2 2
and
ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
0 l3 A \ p1 epðnÞ ; p1 þ epðnÞ p2 epðnÞ ; p2 þ epðnÞ
2 2 2 2
ðnÞ 1 ðnÞ 1
p3 epðnÞ ; p3 þ epðnÞ l3 ðAÞ ¼ 0;
2 2
442 6 Sard’s Theorem
so
ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
l3 A \ p1 epðnÞ ; p1 þ epðnÞ p2 epðnÞ ; p2 þ epðnÞ
2 2 2 2
ðnÞ 1 ðnÞ 1
p3 epðnÞ ; p3 þ epðnÞ ¼ 0:
2 2
Since
ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
l3 A \ p1 epðnÞ ; p1 þ epðnÞ p2 epðnÞ ; p2 þ epðnÞ
2 2 2 2
ðnÞ 1 ðnÞ 1
p3 epðnÞ ; p3 þ epðnÞ ¼ 0;
2 2
and
X
1
d
l3 ðBdk Þ\ pffiffiffi 3 :
k¼1 2 3C
so,
3
l3 FpðnÞ ðBdk Þ 2 diam FpðnÞ ðBdk Þ
pffiffiffi 3 pffiffiffi 3 pffiffiffi 3
2C 3dk ¼ 2 3C ðdk Þ3 ¼ 2 3C l3 ðBdk Þ:
6.1 Measure Zero in Manifolds 443
It follows that
ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
0 l3 FpðnÞ A \ p1 epðnÞ ; p1 þ epðnÞ p2 epðnÞ ; p2 þ epðnÞ
2 2 2 2
ðnÞ 1 ðnÞ 1 1
p3 epðnÞ ; p3 þ epðnÞ l3 FpðnÞ [ Bdk
2 2 k¼1
X 1 pffiffiffi 3
1 1 X
¼ l3 [ FpðnÞ ðBdk Þ l3 FpðnÞ ðBdk Þ 2 3C l3 ðBdk Þ
k¼1
k¼1 k¼1
pffiffiffi 3 X
1
¼ 2 3C l3 ðBdk Þ\d:
k¼1
It follows that
ðnÞ 1 ðnÞ 1 ðnÞ 1 ðnÞ 1
l3 FpðnÞ A \ p1 epðnÞ ; p1 þ epðnÞ p2 epðnÞ ; p2 þ epðnÞ
2 2 2 2
ðnÞ 1 ðnÞ 1
p3 epðnÞ ; p3 þ epðnÞ ¼ 0: h
2 2
which is a contradiction. h
Lemma 6.14 Let M be an m-dimensional smooth manifold. Let fAn g be a col-
lection of nonempty subsets of M: For every k ¼ 1; . . .; m; let lk denote the k-
S
dimensional measure over Rk : Let each An has measure zero in M: Then, 1 n¼1 An
has measure zero in M:
Proof For this purpose, let us takeS any admissible coordinate chart ðU; uÞ S1for M:
We have to prove that lm ðuðð 1 A Þ \ UÞÞ ¼ 0: Here, 0 l ðuðð n¼1 An Þ
S n¼1S n
P1 m
\UÞÞ ¼ lm ðuð 1 ðA
n¼1 n \ UÞÞÞ ¼ l m ð 1
n¼1 ðuðA n \ UÞÞÞ l
n¼1 m ðuðAn \ UÞÞ;
so it suffices to show that for each positive integer n; lm ðuðAn \ UÞÞ ¼ 0: If not,
otherwise, let there exists a positive integer n such that 0\lm ðuðAn \ UÞÞ: We
have to arrive at a contradiction. Since An has measure zero in M; and ðU; uÞ is an
admissible coordinate chart for M, lm ðuðAn \ UÞÞ ¼ 0; a contradiction. h
Lemma 6.15 Let M be an m-dimensional smooth manifold. Let A be a nonempty
subset of M: For every k ¼ 1; . . .; m; let lk denote the k-dimensional measure over
Rk : Let A has measure zero in M: Then, Ac ð M AÞ is a dense subset of M:
Proof We have to show that Ac is dense in M; that is, ðAc Þ ¼ M; that is,
ððAc Þ Þc ð¼ Ao Þ ¼ ;: If not, otherwise, let Ao 6¼ ;: We have to arrive at a contra-
diction. Since Ao 6¼ ;; there exists a point p 2 A such that p is an interior point of A:
Since p is an interior point of A; there exists an open neighborhood G of p in M
such that p 2 G A: Since p 2 A M; and M is an m-dimensional smooth
manifold, there exists an admissible coordinate chart ðU; uÞ for M such that p 2 U:
Clearly, U \ Gð UÞ is an open neighborhood of p: Since U \ Gð UÞ is an open
neighborhood of p; and ðU; uÞ is an admissible coordinate chart for M;
ðU \ G; ujU\G Þ is an admissible coordinate chart for M: Here, uðpÞ is an interior
point of the subset ðujU\G ÞðU \ GÞð¼ uðU \ GÞÞ of Rm ; and hence, lm ðu
6.1 Measure Zero in Manifolds 445
This is a contradiction. h
Lemma 6.16 Let M be an n-dimensional smooth manifold, N be an n-dimensional
smooth manifold, F : M ! N be a smooth map, and A be a nonempty subset of M:
Let A has measure zero in M: Then, FðAÞ has measure zero in N:
Proof We have to prove that FðAÞ has measure zero in N: For this purpose, let us
take an admissible coordinate chart ðV; wÞ for N such that ðFðAÞÞ \ V 6¼ ;: We
have to show that ln ðwððFðAÞÞ \ VÞÞ ¼ 0; where ln denotes the n-dimensional
measure over Rn : Clearly, FðA \ F 1 ðVÞÞ ¼ ðFðAÞÞ \ V:
(Reason: LHS ¼ FðA \ F 1 ðVÞÞ ðFðAÞÞ \ ðFðF 1 ðVÞÞÞ ðFðAÞÞ \ V ¼
RHS: So, LHS RHS: Next, let us take any q 2 RHS ¼ ðFðAÞÞ \ V: It follows that
there exists p 2 A such that FðpÞ ¼ q 2 V: Since FðpÞ 2 V; p 2 F 1 ðVÞ: Since
p 2 F 1 ðVÞ; and p 2 A; p 2 A \ F 1 ðVÞ; and hence, q ¼ FðpÞ 2 FðA \ F 1 ðVÞÞ ¼
LHS: Hence, RHS LHS).
Since M is an n-dimensional smooth manifold, by Lemma 4.47, there exists a
countable collection fðU1 ; u1 Þ; ðU2 ; u2 Þ; ðU3 ; u3 Þ; . . .g of admissible coordinate
charts for M such that fU1 ; U2 ; U3 ; . . .g is a basis of M:
Let us take any p 2 A \ F 1 ðVÞ: Since p 2 A \ F 1 ðVÞ; p 2 F 1 ðVÞ; and hence,
FðpÞ 2 V: Thus, V is an open neighborhood of FðpÞ in N: Since F : M ! N is a
smooth map, F : M ! N is continuous. Since F : M ! N is continuous, and V is an
open neighborhood of FðpÞ; F 1 ðVÞ is an open neighborhood of p in M: Since F 1 ðVÞ
is an open neighborhood of p in M; and fU1 ; U2 ; U3 ; . . .g is a basis of M; there exists a
positive integer kðpÞ such that p 2 UkðpÞ F 1 ðVÞ; and hence, A \ UkðpÞ A \
F 1 ðVÞ: Since p 2 A \ F 1 ðVÞ; p 2 A: Since p 2 A; and p 2 UkðpÞ ; p 2 A \ UkðpÞ : It
follows that for every p 2 A \ F 1 ðVÞ; there exists a positive integer kðpÞ such that
p 2 A \ UkðpÞ A \ F 1 ðVÞ, and hence, A \ F 1 ðVÞ ¼ [fA \ UkðpÞ : p 2 A\
F 1 ðVÞg: Now, ðFðAÞÞ \ V ¼ FðA \ F 1 ðVÞÞ ¼ Fð[fA \ UkðpÞ : p 2 A \ F 1 ðVÞgÞ ¼
[fFðA \ UkðpÞ Þ : p 2 A \ F 1 ðVÞg; and hence wððFðAÞÞ \ VÞ ¼ wð[fFðA \ UkðpÞ Þ :
p 2 A \ F 1 ðVÞgÞ ¼ [fwðFðA \ UkðpÞ ÞÞ : p 2 A \ F 1 ðVÞg ¼ [fðw FÞðA [ U
kðpÞÞ : p 2 A \ F 1 ðVÞg. Thus, wððFðAÞÞ \ VÞ is the union of countable many sets
ðw FÞðA \ UkðpÞ Þ; where p 2 A \ F 1 ðVÞ: Here, fkðpÞ : p 2 A \ F 1 ðVÞg
f1; 2; 3; . . .g; so fkðpÞ : p 2 A \ F 1 ðVÞg can be written as fk1 ; k2 ; k3 ; . . .g: It fol-
lows that
Ukð pÞ : p 2 A \ F 1 ðV Þ ¼ fUk1 ; Uk2 ; Uk3 ; . . .g;
446 6 Sard’s Theorem
and hence,
[1
[ ðw F Þ A \ Ukð pÞ : p 2 A \ F 1 ðV Þ ¼ ðw F Þ A \ Ukj :
j¼1
It follows that
ln ðwððFðAÞÞ \ V ÞÞ ¼ ln [ ðw F Þ A \ Ukð pÞ : p 2 A \ F 1 ðV Þ
!
[
1 X
1
¼ ln ð w F Þ A \ U kj ln ðw F Þ A \ Ukj
j¼1 j¼1
1
X
1
¼ ln w F ukj ukj A \ Ukj :
j¼1
1 1 1 1
a1 e; a1 þ e an e; an þ e
2 2 2 2
will be denoted by Ce ½a: Here, Ce ðaÞ will be called an open cube of side length e;
and Ce ½a will be called a closed cube of side length e:
Let G be a nonempty open subset of Rn : Let 0\e: We shall try toSshow that
there exists a countable collection fUn g of open cubes such that 1 n¼1 Un ¼
S1
G; n¼1 ðUn Þ ¼ G; and side length of each Un is strictly less than e:
Let us take any a ða1 ; . . .; an Þ 2 G: Since a 2 G; and G is open in Rn ; there
exists a positive rational number r such that Cr ðaÞ G; and r\e: Now, since the
6.1 Measure Zero in Manifolds 447
and hence, a 2 C2r ðsÞ: Now, clearly C2r ½s Cr ðAÞ: (Reason: Let us take any x
ðx1 ; . . .; xn Þ 2 C2r ½s: It follows that for every i ¼ 1; . . .; n; jxi si j 4r : Now, for
every i ¼ 1; . . .; n;
i
x ai xi si þ si ai r þ si ai \ r þ r ¼ r ;
4 4 4 2
and hence, x 2 Cr ðaÞ:Þ Thus, for every a 2 G; there exist a positive rational number
r; and s 2 Rn such that all coordinates of s are rational numbers, and
a 2 C2r ðsÞ C2r ðsÞ ¼ C2r ½s Cr ðAÞ G:
Now, since the collection of all open cubes of the form C2r ðsÞ; where r is a positive
rational number, and all coordinates of sð2 Rn Þ are rational numbers, is aScountable
set, there exists a countable collection fUn g of open cubes such that 1 n¼1 Un ¼
S
G; 1 n¼1 ðUn Þ
¼ G; and side length of each Un is strictly less than e: h
Note 6.18 Let G be a nonempty open subset of Rn : Let 0\e: We shall try to show
that there exist a countable collection fUn g of open S1 cubes, and S a countable col-
1
lection fVn g of open cubes such that Un ¼ G, n¼1 ðUn Þ ¼ G,
S1
n¼1
n¼1 Vn ¼ G, Un ðUn Þ Vn ðn ¼ 1; 2; 3; . . .Þ, and side length of each Vn is
strictly less than e:
Let us take any a ða1 ; . . .; an Þ 2 G: Since a 2 G; and G is open in Rn ; there
exists a positive rational number r such that Cr ðaÞ G; and r\e: Now, since the
collection fs : s 2 Rn ; and all coordinates of s are rational numbersg is dense in
Rn ; there exists s ðs1 ; . . .; sn Þ 2 Rn such that s 2 C4r ðaÞ; and all coordinates of s
are rational numbers. Since s 2 C4r ðaÞ; for every i ¼ 1; . . .; n;
i
a s i ¼ s i ai \ r ;
8
and hence, a 2 C4r ðsÞ: Now, clearly C4r ½s C2r ðaÞ: (Reason: Let us take any x
ðx1 ; . . .; xn Þ 2 C4r ½s: It follows that for every i ¼ 1; . . .; n; jxi si j 8r : Now, for
every i ¼ 1; . . .; n;
i
x ai xi si þ si ai r þ si ai \ r þ r ¼ r ;
8 8 8 4
and hence, x 2 C2r ðAÞ:) Also, it is clear that C2r ½s Cr ðAÞ: (Reason: Let us take any
x ðx1 ; . . .; xn Þ 2 C2r ½s: It follows that for every i ¼ 1; . . .; n; jxi si j 4r : Now,
for every i ¼ 1; . . .; n;
448 6 Sard’s Theorem
i
x ai x i s i þ s i ai r þ s i ai \ r þ r \ r ;
4 4 8 2
and hence, x 2 Cr ðAÞ:) Thus, for every a 2 G; there exist a positive rational number
r; and s 2 Rn such that all coordinates of s are rational numbers, and
a 2 C4r ðsÞ C4r ðsÞ ¼ C4r ½s C2r ½s Cr ðAÞ G:
Now, since the collection of all open cubes of the form C4r ðsÞ; where r is a positive
rational number, and all coordinates of sð2 Rn Þ are rational numbers, is a countable
set, there exist a countable collection S fUn g of open Scubes, and a countable
S1 col-
lection fVn g of open cubes such that 1 n¼1 U n ¼ G; 1
n¼1 ðUn Þ
¼ G, n¼1 Vn ¼
G, Un ðUn Þ Vn ðn ¼ 1; 2; 3; . . .Þ, and side length of each Vn is strictly less
than e: h
¼
6.2 Taylor’s Inequality 449
Proof Let us take any t satisfying 0 t 1: Now, since S is star shaped with
respect to point p; and x 2 S; ð1 tÞp þ tx is in S; and hence, f ðð1 tÞp þ txÞ is a
real number. Here,
d d
f ðð1 tÞp þ txÞ ¼ f p1 þ t x1 p1 ; p2 þ t x2 p2
dt dt
of d 1 1
¼ ð ð1 t Þp þ txÞ p þ t x p 1
ox1 dt
of d 2 2
þ ðð1 tÞp þ txÞ p þt x p 2
ox2 dt
of
¼ ðð1 tÞp þ txÞ x1 p1
ox 1
of 2
þ ðð 1 t Þp þ tx Þ x p2 ;
ox 2
Hence,
Z1
of 2 1 of
2 Z
x1 p 1 ðð 1 t Þp þ tx Þ dt þ x p ðð 1 t Þp þ tx Þ dt
0
ox1 0
ox2
Z1 of 1 of 2
¼ ð ð1 t Þp þ txÞ x p 1
þ ðð 1 t Þp þ tx Þ x p 2
dt
0
ox1 ox2
¼ f ðð1 tÞp þ txÞjt¼1
t¼0 ¼ f ð xÞ f ð pÞ:
Now,
!
of Z1
f ð x Þ ¼ f ð pÞ þ x 1 p 1
ðð1 tÞp þ txÞ dt
ox1
!
0
2 Z1 of
þ x p 2
ðð1 tÞp þ txÞ dt :
0
ox2
d
þððD12 f Þðð1 tÞp þ txÞÞ ð1 tÞp þ tx
2 2
dt
dt
t¼1
þ x2 p2 ðtððD2 f Þðð1 tÞp þ txÞÞÞjt¼0 dt
Z1 d
t ððD21 f Þðð1 tÞp þ txÞÞ ð1 tÞp1 þ tx1
0
dt
d
þððD22 f Þðð1 tÞp þ txÞÞ ð1 tÞp þ tx
2 2
dt
dt
Z1
¼ f ð pÞ þ x1 p1 ððD1 f Þð xÞ t ððD11 f Þðð1 tÞp þ txÞÞ x1 p1
0
þððD12 f Þðð1 tÞp þ txÞÞ x p2 dt 2
Z1
þ x2 p2 ððD2 f Þð xÞ t ððD21 f Þðð1 tÞp þ txÞÞ x1 p1
0
þððD22 f Þðð1 tÞp þ txÞÞ x2 p2 dt
¼ f ð pÞ þ x1 p1 ððD1 f Þð xÞÞ þ x2 p2 ððD2 f Þð xÞÞ
2 Z1
x1 p1 tðððD11 f Þðð1 tÞp þ txÞÞÞdt
0
Z1
2 x1 p 1
x 2 p2 tððD12 f Þðð1 tÞp þ txÞÞdt
0
2 Z1
x 2 p2 tðððD22 f Þðð1 tÞp þ txÞÞÞdt
0
Z1
¼ f ð pÞ þ x1 p1 ðD1 f Þð pÞ þ x1 p1 ðD11 f Þðð1 tÞp þ txÞdt
0
!
Z1
þ x2 p2 ðD12 f Þðð1 tÞp þ txÞdt
0
Z1
þ x2 p2 ðD2 f Þð pÞ þ x1 p1 ðD21 f Þðð1 tÞp þ txÞdt
0
!!
Z1
þ x2 p2 ðD22 f Þðð1 tÞp þ txÞdt
0
1
1 2Z
x1 p tðððD11 f Þðð1 tÞp þ txÞÞÞdt
0
Z1
2 x1 p1 x2 p2 tððD12 f Þðð1 tÞp þ txÞÞdt
0
1
2 2Z
x2 p tðððD22 f Þðð1 tÞp þ txÞÞÞdt
0
6.2 Taylor’s Inequality 451
¼ f ð pÞ þ x1 p1 ðD1 f Þð pÞ þ x2 p2 ðD2 f Þð pÞ
2 Z1
þ x1 p1 ððD11 f Þðð1 tÞp þ txÞ tðððD11 f Þðð1 tÞp þ txÞÞÞÞdt
0
Z1
þ 2 x1 p1 x2 p2 ððD12 f Þðð1 tÞp þ txÞ tððD12 f Þðð1 tÞp þ txÞÞÞdt
0
1
2 2Z
þ x p 2
ððD22 f Þðð1 tÞp þ txÞ tðððD22 f Þðð1 tÞp þ txÞÞÞÞdt
0
1 1 of 2
2 of
¼ f ð pÞ þ x p1 ð p Þ þ x p ð pÞ
1! ox1 ox2
!
1 1 1
1 2Z o2 f
þ x p ð1 t Þ ðð1 tÞp þ txÞ dt
1! 0 oðx1 Þ2
Z1 o2 f
þ 2 x1 p 1 x2 p 2 ð1 tÞ 1 2 ðð1 tÞp þ txÞ dt
0
ox ox
! !
2 1
2 2Z o f
2
þ x p ð1 tÞ ðð1 tÞp þ txÞ dt
0 oðx2 Þ2
0 1
1@ X i
¼ f ð pÞ þ x pi ððDi f Þð pÞÞA
1! i2f1;2g
0 1
1@ X i Z1
þ x pi x j p j ð1 tÞDij ðð1 tÞp þ txÞ dtA:
1! ði;jÞ2f1;2gf1;2g 0
Similarly,
0 1
1@ X i
f ð x Þ ¼ f ð pÞ þ x pi ððDi f Þð pÞÞA
1! i2f1;2g
0 1
1@ X i
þ x pi x j p j Dij f ð pÞ A
2! 2
ði;jÞ2f1;2g
0 1
1@ X i j k 1
þ x p x p x p
i j k Z
ð1 tÞ Dijk ðð1 tÞp þ txÞ dtA
2
2! 3 0
ði;j;k Þ2f1;2g
¼
Note 6.20 Let p 2 U Rn : Let U be an open set. Let S U; and S be star shaped
with respect to point p ðp1 ; . . .; pn Þ 2 S. Let f : U ! R be C 1 on U: Let x
ðx1 ; . . .; xn Þ 2 S: Let all the third-order partial derivatives of f be bounded in
absolute value by a constant M on S: Then,
8 0 1
<
X
f ðxÞ f ð pÞ þ 1 @ xi pi ððDi f Þð pÞÞA
:
1! i2f1;...;ng
0 19
1@ X i =
þ x pi x j p j Dij f ð pÞ A
2! ;
ði;jÞ2f1;...;ng2
0 1
1 X i Z1
¼ @ x p x p x p
i j j k k
ð1 tÞ Dijk ðð1 tÞp þ txÞ dt
2 A
2! ði;j;kÞ2f1;...;ng3 0
X Z1
1 i
x pi x j p j xk pk ð1 tÞ2 Dijk ðð1 tÞp þ txÞ dt
2! 3
0
ði;j;k Þ2f1;...;ng
X i 1
¼
1 x pi x j p j xk pk Z ð1 tÞ2 Dijk ðð1 tÞp þ txÞ dt
2! 3
0
ði;j;k Þ2f1;...;ng
1 X i 1
x pi x j p j xk pk Z j1 tj2 Dijk ðð1 tÞp þ txÞdt
2! 0
ði;j;k Þ2f1;...;ng3
1 X i 1
x pi x j p j xk pk Z 1Dijk ðð1 tÞp þ txÞdt
2! 0
ði;j;k Þ2f1;...;ng3
1 X i 1
x pi x j p j xk pk Z Mdt
2! 0
ði;j;k Þ2f1;...;ng3
1 X i
¼ x pi x j p j xk pk M
2!
ði;j;k Þ2f1;...;ng3
1 X
jx pjjx pjjx pjM
2!
ði;j;k Þ2f1;...;ng3
1 X 1
¼ M j x pj3 ¼ ðn n n Þ M j x pj 3
2! 3!
ði;j;k Þ2f1;...;ng3
1 3
¼ n M j x pj 3 :
2!
6.2 Taylor’s Inequality 453
Thus
8 0 1
< X
1
f ð xÞ f ð pÞ þ @ xi pi ððDi f Þð pÞÞA
:
1! i2f1;...;ng
0 19
1@ X i =
þ x p x p
i j j
Dij f ð pÞ A
2! 2
;
ði;jÞ2f1;...;ng
1
n3 jx pj3 M;
2!
etc.
Proof Put
8 2 3 9
< ðD1 F1 Þð xÞ =
C x : x 2 G; and rank4 ðD1 F2 Þð xÞ 5 ¼ 0 :
: ;
ðD1 F3 Þð xÞ
454 6 Sard’s Theorem
Thus,
and hence,
2 !!!!3
1 R
0 l3 ðF ðC1 \ Ek ÞÞ l3 B 1 ðFðaÞÞ 2 3 K
K ðNR Þ
2
3 1! 1! N
! 0 as N ! 1:
S1 of R ; there
m
Proof Since G is a nonempty open subset S exists acountable collection
fCn g of open cubes in R such that n¼1 Cn ¼ G, 1
m
n¼1 ðCn Þ ¼ G; and for each
n ¼ 1; 2; 3; . . .; the side length Rn of Cn is \1: Now,
6.3 Sard’s Theorem on Rn 457
08 2 3 91
>
> ðD1 f1 Þð xÞ ðDm f1 Þð xÞ >
>
B< 6. .. . 7 =C
B
f ðC Þ ¼ f @ x : x 2 G; and rank4 . 6 . ... 7 \n C
> 5 > A
>
: >
;
ðD1 fn Þð xÞ ðDm fn Þð xÞ
08 2 3 91
>
> ðD1 f1 Þð xÞ ðDm f1 Þð xÞ >
>
B< [
1 6. .. . 7 =C
B
¼ f@ x : x 2
ðCn Þ ; and rank4 . 6 . ... 7 \n C
> 5 > A
>
: n¼1 >
;
ðD1 fn Þð xÞ ðDm fn Þð xÞ
0 8 2 3 91
>
> ðD1 f1 Þð xÞ ðDm f1 Þð xÞ >
>
B[ 1 < 6. .. . 7 =C
¼ fB@ > x : x 2 ð Cn Þ
; and rank 6
4. . . .
.
7
5\n C
A
n¼1>
>
>
: ;
ðD1 fn Þð xÞ ðDm fn Þð xÞ
08 2 3 91
>
> ðD1 f1 Þð xÞ ðDm f1 Þð xÞ >
>
[ B
1 < 6. .. . 7 = C
¼ fB@> x : x 2 ð C n Þ
; and rank 6
4. . . .
.
7
5 \n C
A
> >
>
n¼1 : ;
ðD1 fn Þð xÞ ðDm fn Þð xÞ
[1
¼ f ðC \ðCn Þ Þ:
n¼1
S1
Since f ðCÞ ¼ n¼1 f ðC \ ðCn Þ Þ;
!
[
1
X
1
0 lðf ðC ÞÞ ¼ l f ðC \ ðCn Þ Þ lðf ðC \ ðCn Þ ÞÞ:
n¼1 n¼1
1
jf ð yÞ ðf ð xÞ þ ðf 0 ð xÞÞðy xÞÞj\ nm3 M 2 :
N
1
jf ð yÞ ðf ðAÞ þ ðf 0 ðAÞÞðy aÞÞj\ nm3 M 2 :
N
2 3 2 3 2 3
f 1 ð yÞ f1 ð xÞ f1 ð yÞ f1 ð xÞ
6 . 7 6. 7 6 . 7
jf ð yÞ f ð xÞj ¼ 64 ..
7 6.
5 4.
7 ¼ 6 .
5 4 .
7
5
f n ð yÞ fn ð xÞ fn ð yÞ fn ð xÞ
jf1 ð yÞ f1 ð xÞj þ þ jfn ð yÞ fn ð xÞj
1 pffiffiffiffi R1 1 1 pffiffiffiffi R1 1
m1 m M0 þ þ m1 m M0
0! N 0! N
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
n
1 1
¼ nm 3=2
M0 R1 \nm3=2 M0 :
N N
and hence,
n1 !
1 3 1
lð f ð C \ E k Þ Þ 3=2
2 nm M0 2 nm M 2
N N
3ðnþ1Þ
1
¼ 2n nn m 2 ðM0 Þn1 M :
N nþ1
460 6 Sard’s Theorem
It follows that
X
lðf ðC \ ðC1 Þ ÞÞ flðf ðC \ Ek ÞÞ : C \ Ek 6¼ ;; 1 k N m g
3ðnþ1Þ
1
2n nn m 2 ðM0 Þn1 M N m ! 0 as N ! 1;
N nþ1
Ci fx: x 2 G; and all 1st; 2nd; . . .; ith order partial derivatives of f at x are zerog:
Then,
1. C4 C3 C2 C1 C G;.
2. each Ci and C are closed in G:
Proof
1. This is clear from the definitions of Ci and C:
2. Since f : G ! Rn is smooth, each Dj fi : G ! R is continuous. Since each Dj fi :
G ! R is continuous, and f0g is closed in R; each ðDj fi Þ1 ðf0gÞ is closed in G:
It follows that
\ 1
Dj fi ðf0gÞ ¼ x : x 2 G; and 0 ¼ Dj fi ð xÞ for every j 2 f1; . . .; kg;
j 2 f1; . . .; k g
i 2 f1; . . .; ng
and for every i 2 f1; . . .; ngg ¼ C1 Þ
8 2 3 9
>
> ðD1 f1 Þð xÞ ðDk f1 ÞðxÞ >
>
< 6. .. . 7 =
C ¼ x : x 2 G; and rank6 4 .
. . .
.
7\n
5
>
> >
>
: ;
ðD1 fn Þð xÞ ðDk fn ÞðxÞ
8 2 3 9
>
> ðDi1 f1 ÞðxÞ ðDin f1 ÞðxÞ >
>
\ < 6. .. . 7 =
¼ 6
x : x 2 G; and det4 .. . .. 7 ¼ 0 :
> 5 >
i1 ; . . .; in 2 f1; . . .; k g; >
:
ðDi1 fn ÞðxÞ ðDin fn ÞðxÞ
>
;
i1 ; . . .; in are distinct
Ci fx : x 2 G; and all 1st; 2nd; . . .; ith order partial derivatives of f at x are zerog:
Then,
ln f C½m ¼ 0;
n
Proof Since G is a nonemptyS open subset ofSRm ; there exists a countable collection
fUl g of open cubes such that 1 l¼1 Ul ¼ G,
1
l¼1 ðUl Þ ¼ G; and side length of each
Ul is strictly less than 1: It follows that
1 1
C½m ¼ C½m \ G ¼ C½m \ [ ðUl Þ ¼ [ C½m \ ðUl Þ ;
n n n l¼1 l¼1 n
and hence,
!
1
[ 1
[
f C½m ¼ f C½m \ ðUl Þ ¼ f C½m \ ðUl Þ :
n n n
l¼1 l¼1
Thus,
!
1
[ X
1
0 ln f C½m ¼ ln f C½m \ ðUl Þ ln f C½m \ ðUl Þ ;
n n n
l¼1 l¼1
Let us fix any positive integer N [ 1: Let us subdivide the closed cube ðUl Þ
into N m closed cubes E1 ; E2 ; . . .; EN m such that each Ei is of length
N ðside length of Ul Þ: For simplicity, let the side length of Ul be 1. Thus, ðUl Þ ¼
1
E1 [ E2 [ [ EN m : It follows that
¼ C½m \ Ek : C½m \ Ek 6¼ ;; 1 k N m :
n n
Hence,
[n o
f C½m \ ðUl Þ ¼ f C½m \ Ek : C½m \ Ek 6¼ ;; 1 k N m
[n o
n n n
¼ f C½m \ Ek : C½m \ Ek 6¼ ;; 1 k N m ;
n n
6.3 Sard’s Theorem on Rn 463
and hence,
[n o
0 ln f C½m \ ðUl Þ ¼ ln f C½m \ Ek : C½m \ Ek 6¼ ;; 1 k N m
Xn o
n n n
ln f C½m \ Ek : C½m \ Ek 6¼ ;; 1 k N :
m
n n
n ! n !
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
n
!
1
¼ n m m½ n þ1 jx yj½ n þ1 M
m m
n !
m !
1 ½mnþ1 pffiffiffiffi 1 ½ n þ1
n m m m M :
n !
N
and hence,
m !!!n
1 pffiffiffiffi 1 ½ n þ1
ln f C½m \ Ek 2 n m m½ n þ1
m
m M :
n !
n N
464 6 Sard’s Theorem
Thus,
m !!!n !
1 ½mnþ1 pffiffiffiffi 1 ½ n þ1
0 ln f C½m \ ðUl Þ N m 2 n m m m M
n !
n N
! 0 as N ! 1;
because
hmi m hmi
m\ þ1 n that is; \ þ1 :
n n n
Thus,
ln f C½m \ ðUl Þ ¼ 0: h
n
Then, ln ðf ðCÞÞ ¼ 0:
Proof (Induction on m): By Lemma 6.23, the theorem is true for m ¼ 1:
Case I: when m n. By Lemma 6.23, ln ðf ðCÞÞ ¼ 0:
Case II: when n\m: For i ¼ 1; 2; 3; . . .; put
and hence,
f ðCÞ ¼ f ðC C1 Þ [ ðC1 C2 Þ [ ðC2 C3 Þ [ [ C½m1 C½m [ C½m
n n n
It follows that
ln ðf ðC C1 ÞÞ þ ln ðf ðC1 C2 ÞÞ þ ln ðf ðC2 C3 ÞÞ
þ þ ln f C½m1 C½m þ ln f C½m :
n n n
ðD1 f1 ; f2 ; f3 Þ F 1 u1 ; u2 ; . . .; um
¼ u1 ; f2 F 1 u1 ; u2 ; . . .; um ; f3 F 1 u1 ; u2 ; . . .; um :
It follows that
2 3
1 0 0
6 7
0 ¼ det4 ðD1 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD2 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD3 ðf2 ðF 1 ÞÞÞðF ð xÞÞ 5
ðD1 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD2 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD3 ðf3 ðF 1 ÞÞÞðF ð xÞÞ
ðD2 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD3 ðf2 ðF 1 ÞÞÞðF ð xÞÞ
¼ det ;
ðD2 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD3 ðf3 ðF 1 ÞÞÞðF ð xÞÞ
2 3
1 0 0
6 7
0 ¼ det4 ðD1 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD2 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD4 ðf2 ðF 1 ÞÞÞðF ð xÞÞ 5
ðD1 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD2 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD4 ðf3 ðF 1 ÞÞÞðF ð xÞÞ
ðD2 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD4 ðf2 ðF 1 ÞÞÞðF ð xÞÞ
¼ det ;
ðD2 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD4 ðf3 ðF 1 ÞÞÞðF ð xÞÞ
ðD2 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD3 ðf2 ðF 1 ÞÞÞðF ð xÞÞ . . ðDm ðf2 ðF 1 ÞÞÞðF ðxÞÞ
rank . \2:
ðD2 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD3 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðDm ðf3 ðF 1 ÞÞÞðF ðxÞÞ
and hence,
l2 f2 F 1 0; y2 ; . . .; ym ; f3 F 1 0; y2 ; . . .; ym
: 0; y2 ; . . .; ym 2 F ððC1 C2 Þ \ ðVa Þ Þ ¼ 0:
u2 ; . . .; um Þ 2 FðUa Þ;
f F 1 u1 ; u2 ; . . .; um ¼ u1 ; f2 F 1 u1 ; u2 ; . . .; um ;
f3 F 1 u1 ; u2 ; . . .; um :
It follows that
2 3
1 0 0
6 7
0 ¼ det4 ðD1 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD2 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD3 ðf2 ðF 1 ÞÞÞðF ð xÞÞ 5
ðD1 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD2 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD3 ðf3 ðF 1 ÞÞÞðF ð xÞÞ
ðD2 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD3 ðf2 ðF 1 ÞÞÞðF ð xÞÞ
¼ det ;
ðD2 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD3 ðf3 ðF 1 ÞÞÞðF ð xÞÞ
2 3
1 0 0
6 7
0 ¼ det4 ðD1 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD2 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD4 ðf2 ðF 1 ÞÞÞðF ð xÞÞ 5
ðD1 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD2 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD4 ðf3 ðF 1 ÞÞÞðF ð xÞÞ
ðD2 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD4 ðf2 ðF 1 ÞÞÞðF ð xÞÞ
¼ det ;
ðD2 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD4 ðf3 ðF 1 ÞÞÞðF ð xÞÞ
ðD2 ðf2 ðF 1 ÞÞÞðF ð xÞÞ ðD3 ðf2 ðF 1 ÞÞÞðF ð xÞÞ . . ðDm ðf2 ðF 1 ÞÞÞðF ðxÞÞ
rank . \2:
ðD2 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðD3 ðf3 ðF 1 ÞÞÞðF ð xÞÞ ðDm ðf3 ðF 1 ÞÞÞðF ðxÞÞ
fc y2 ; . . .; ym ¼ f2 F 1 c; y2 ; . . .; ym ; f3 F 1 c; y2 ; . . .; ym :
and hence,
474 6 Sard’s Theorem
1
f2 x ; . . .; xm ; f3 x1 ; . . .; xm : x1 ; . . .; xm 2 ðC C1 Þ \ ðVa Þ ;
and f1 x1 ; . . .; xm ¼ c
f2 F 1 c; y2 ; . . .; ym ; f3 F 1 c; y2 ; . . .; ym
: c; y2 ; . . .; ym 2 F ððC C1 Þ \ ðVa Þ Þ :
c; x2 ; . . .; xm ¼ f1 x1 ; . . .; xm ; x2 ; . . .; xm ¼ F x1 ; . . .; xm
2 F ððC C1 Þ \ ðVa Þ Þ;
1
f2 x ; . . .; xm ; f3 x1 ; . . .; xm
¼ f2 F 1 c; x2 ; . . .; xm ; f3 F 1 c; x2 ; . . .; xm
¼ f2 F 1 c; x2 ; . . .; xm ; f3 F 1 c; x2 ; . . .; xm 2 RHS:
[
l3 ðf ðC C1 ÞÞ ¼ l3 ff ððC C1 Þ \ ðVa Þ Þ : a 2 ðC C1 Þg
X X
l3 ðf ððC C1 Þ \ðVa Þ ÞÞ ¼ 0 ¼ 0:
countable countable
Thus, l3 ðf ðC C1 ÞÞ ¼ 0: h
6.4 Sard’s Theorem on Manifolds 475
ln ðwððf ðC ÞÞ \ V ÞÞ ¼ 0;
0 ln ðwððf ðCÞÞ \ V ÞÞ
¼ ln ðwððf ððC \ U1 Þ [ ðC \ U2 Þ [ ðC \ U3 Þ [ ÞÞ \ V ÞÞ
¼ ln ðwðððf ðC \ U1 ÞÞ [ ðf ðC \ U2 ÞÞ [ ðf ðC \ U3 ÞÞ [ Þ \ V ÞÞ
¼ ln ðwðððf ðC \ U1 ÞÞ \ V Þ [ ððf ðC \ U2 ÞÞ \ V Þ [ ððf ðC \ U3 ÞÞ \ V Þ [ ÞÞ
¼ ln ðwððf ðC \ U1 ÞÞ \ V Þ [ wððf ðC \ U2 ÞÞ \ V Þ [ wððf ðC \ U3 ÞÞ \ V Þ [ Þ
ln ðwððf ðC \ U1 ÞÞ \ V ÞÞ þ ln ðwððf ðC \ U2 ÞÞ \ V ÞÞ
þ ln ðwððf ðC \ U3 ÞÞ \ V ÞÞ þ :
Since
0 ln ðwððf ðCÞÞ \ V ÞÞ
ln ðwððf ðC \ U1 ÞÞ \ V ÞÞ
þ ln ðwððf ðC \ U2 ÞÞ \ V ÞÞ
þ ln ðwððf ðC \ U3 ÞÞ \ V ÞÞ þ ;
ln ðwððf ðC \ Uk ÞÞ \ V ÞÞ ¼ 0:
ln ðwððf ðC \ U1 ÞÞ \ V ÞÞ ¼ 0:
Since u1 is a diffeomorphism from open set U1 onto open set u1 ðU1 Þ; and w is a
diffeomorphism from open set V onto the open set wðVÞ; the collection of all critical
points of f in U1 is equal to ðu1 Þ1 ðthe collection of all critical points of ðw f
ðu1 Þ1 Þin u1 ðU1 ÞÞ. Since f : M ! N is a smooth mapping, w f ðu1 Þ1 is smooth.
Since w f ðu1 Þ1 is smooth, by Lemma 6.26, 0 ¼ ln ððw f ðu1 Þ1 Þ ðthe
collection of all critical points of ðw f ðu1 Þ1 Þ in u1 ðU1 ÞÞÞ ¼ ln ððw f Þ
ððu1 Þ1 ðthe collection of all critical points of ðw f ðu1 Þ1 Þin u1 ðU1 ÞÞÞÞ ¼ ln
ððw f Þðcollection of all critical points of f in U1 ÞÞ ¼ ln ðwðf ðcollection of all
critical points of f in U1 ÞÞÞ ¼ ln ðwððf ðC \ U1 ÞÞ \ VÞÞ: Thus, ln ðwððf ðC\ U1 ÞÞ \
VÞÞ ¼ 0: Similarly,
This is the smallest chapter of this book, because it contains only two theorems
which are due to Whitney. These theorems have three serious reasons to study.
Firstly, in its proof the celebrated Sard’s theorem got an application. Secondly, the
statement of Whitney embedding theorem was contrary to the common belief that a
smooth manifold may not have any ambient space. Thirdly, in its proof, Whitney
used almost all tools of smooth manifolds developed at that time. Fortunately, in
this chapter we have all the prerequisite for its proof in the special case of compact
smooth manifolds. For the general case, which is more difficult, one can find its
proof somewhere else.
Since u1 1 1
1 ðB1 ð0ÞÞ; u2 ðB1 ð0ÞÞ; u3 ðB1 ð0ÞÞ; . . . is an open cover of M, and M is
compact, there exists a positive integer k such that u1 1
1 ðB1 ð0ÞÞ; . . .; uk ðB1 ð0ÞÞ is
1
an open cover of M. For every i ¼ 1; . . .; k; ui is continuous, and B1 ½0 is com-
pact, so u1 1
i ðB1 ½0Þ is compact. ui ðB1 ½0Þ is compact in M, and M is Hausdorff, so
ui ðB1 ½0Þ is a closed subset of M. Since u1
1
i ðB1 ½0Þ is a closed subset of M,
u1 1 1
i ðB2 ð0ÞÞ is an open subset of M, and ui ðB1 ½0Þ ui ðB2 ð0ÞÞ, by Lemma
4.58, there exists a smooth function wi : M ! ½0; 1 such that
a. for every x in u1
i ðB1 ½0Þ; wi ðxÞ ¼ 1;
b. supp wi u1i ðB2 ð0ÞÞ:
1
w2 ð pÞu22 ð pÞ if p 2 u
2 1ðB2 ½0Þ c
^ 22 ð pÞ
u
0 if p 2 u2 ðB2 ½0Þ
F ð pÞ ððu ^ 1m ð pÞÞ; . . .; ðu
^ 11 ð pÞ; . . .; u ^ k1 ð pÞ; . . .; u
^ km ð pÞÞ; ðw1 ð pÞ; . . .; wk ð pÞÞÞ:
Since p 2 M ¼ u1 1
1 ðB1 ð0ÞÞ [ [ uk ðB1 ð0ÞÞ; there exists a positive integer l
such that 1 l k; and p 2 u1 l ðB1 ð0ÞÞ: It follows that for every x 2
u1
l ðB1 ð0ÞÞð u 1
l ðB 2 ½0ÞÞ; we have wl ðxÞ ¼ 1; and ð^ ^ lm ðxÞÞ ¼
ul1 ðxÞ; . . .; u
ðwl ðxÞul1 ðxÞ; . . .; wl ðxÞulm ðxÞÞ ¼ ð1ul1 ðxÞ; . . .; 1ulm ðxÞÞ ¼ ul ðxÞ: For simplicity,
let l ¼ 1: Thus, for every x 2 u1 1 ðB1 ð0ÞÞ; we have ð^ ^ 1m ðxÞÞ ¼ u1 ðxÞ:
u11 ðxÞ; . . .; u
Here, 0 ¼ ðdFp ÞðvÞ ¼ ððdð^ u1 Þp ÞðvÞ; . . .; ðdð^uk Þp ÞðvÞ; ðdðw1 Þp ÞðvÞ; . . .; ðdðwk Þp ÞðvÞÞ ¼
ððdðu1 Þp ÞðvÞ; . . .; ðdð^uk Þp ÞðvÞ; ðdðw1 Þp ÞðvÞ; . . .; ðdðwk Þp ÞðvÞÞ; so ðdðu1 Þp ÞðvÞ ¼ 0:
Since u1 is a diffeomorphism, dðu1 Þp is an isomorphism. Since dðu1 Þp is an iso-
morphism, and ðdðu1 Þp ÞðvÞ ¼ 0; v ¼ 0:
Since F : M ! RN is 1–1, F : M ! RN is a smooth immersion, and M is
compact, by Note 5.72, F : M ! RN is a smooth embedding. h
Note 7.2 The above Lemma 7.1 is known as the compact Whitney embedding in
RN :
!
ðv aÞ
Pa ðvÞ v a
jaj2
ð f ð pÞ f ð qÞ Þ a
ð f ð pÞ f ð qÞ Þ a ¼ Pa ðf ð pÞ f ðqÞÞ ¼ 0:
jaj2
Since
ð f ð pÞ f ð qÞ Þ a
ð f ð pÞ f ð qÞ Þ a ¼ 0;
jaj2
Thus, a 2 aðM M RÞ; which is a contradiction. So our claim is true, that is,
p ¼ q: Thus, ðPa f Þ : M ! RN1 is 1–1.
For 2: Let us take any p 2 M: Let ðdðPa f Þp ÞðwÞ ¼ 0 where w 2 Tp M: We
have to show that w ¼ 0: If not, otherwise, let w 6¼ 0: We have to arrive at a
contradiction. Since f : M ! RN is a smooth embedding (and hence, f : M !
RN is an immersion), and p 2 M; the linear map dfp : Tp M ! Tf ðpÞ ðRN Þð¼ RN Þ
is 1–1. Since the linear map dfp : Tp M ! RN is 1–1, and 0 6¼ w 2 Tp M;
ðdfp ÞðwÞ 6¼ 0: Here,
7.2 Compact Whitney Embedding in R2mþ1 483
dfp ðwÞ a
dfp ðwÞ 2
a ¼ Pa dfp ðwÞ ¼ Pa dfp ðwÞ
j aj
¼ d ðPa ÞF ð pÞ dfp ðwÞ
¼ d ðPa f Þp ðwÞ ¼ 0; so dfp ðwÞ
dfp ðwÞ a
¼ a:
jaj2
Since
dfp ðwÞ a
dfp ðwÞ ¼ a;
jaj2
Note 7.4 The Lemma 7.3 is known as the compact Whitney embedding in R2mþ1 :
Bibliography