Math 598 Mar 2, 20051

Geometry and Topology II

Spring 2005, PSU

Lecture Notes 9

3 Some Topics from Differential Topology

3.1 Regular points and values; Fundamental Theorem
of Algebra
Let f : M → N be a smooth map. We say that p ∈ M is a regular point of
f provided that rank(dfp ) = dim(N ); otherwise we say that p is a critical
point or a singular point. We say q ∈ N is a regular value of f provided that
every p ∈ f −1 (q) is a regular point. If q is not a regular value of f , then we
say that it is a critical value or singular value.

Exercise 1. Show that if f : M m → N n has rank k at some point p ∈ M ,

then it has rank k on a neighborhood of p.

Proposition 2. If f : M m → N n is a smooth map, q is a regular value of f ,

and f −1 (q) = ∅, then f −1 (q) is a smooth (m − n)-dimensional submanifold
of M .

Proof. By definition, every p ∈ f −1 (q) is a regular point. Thus, f has con-

stant rank n on f −1 (q). This implies that f has constant rank n on an
open neighborhood of U of f −1 (q). Since f : U → N has constant rank n, it
now follows, as we had proved earlier, that f −1 (q) is an (m − n)-dimensional
submanifold of U , and therefore of M .

Exercise 3. Show that if f : M → N is a smooth map, dim(M ) = dim(N ),

M is compact, and q is a regular value of f , then f −1 (q) consists of a finite
number of points. Further, show that if we denote the number of these
points by #f −1 (q), then #f −1 (q) is locally constant, i.e., there is an open
neighborhood U of q in N , such that #f −1 (q
) = #f −1 (q), for all q
∈ U .
1
Last revised: March 8, 2005

1
 The following proof demonstrates the elegant and powerful utility of basic
techniques of Differential topology.

Theorem 4 (Fundamental Theorem of Algebra). Every nonconstant com-

plex polynomial has a zero.

Proof(After Milnor). Let P : R2 → R2 be a complex polynomial, and π+ : S2 −

{(0, 0, 1)} → R2 be the stereographic projection from the north pole. Define
f : S2 → S2 by f ((0, 0, 1)) := (0, 0, 1), and

f (p) := (π+ )−1 ◦ P ◦ π+ (p),

if p = (0, 0, 1).
We claim that f is smooth. This is obvious on S2 − {(0, 0, 1)} where f
is the composition of three smooth functions. To see that f is smooth in a
neighborhood of (0, 0, 1) as well, let π− : S2 − {(0, 0, −1)} → R2 be the stere-
ographic projection from the south pole, U ⊂ R2 be a small neighborhood
of the origin o ∈ R2 , and define Q : U → R2 by

Q(z) := π− ◦ f ◦ (π− )−1 (z).

Note that if U is sufficiently small f ◦ (π− )−1 (z) is close to f ◦ (π− )−1 (o) =
(0, 0, 1) for all z ∈ U . In particular, if U is sufficiently small, f ◦ (π− )−1 (z) =
(0, 0, −1) for all z ∈ U . Thus Q is well defined. Secondly, note that π+ ◦
(π− )−1 is inversion with respect to the unit
circle in R2 , i.e., π+ ◦(π− )−1 (z) =
z/z = 1/z. This yields that, if P (z) = ni=0 ai z i , with an = o, then
2

Q(z) = π− ◦ (π+ )−1 ◦ P ◦ π+ ◦ (π− )−1 (z)

= ((π+ ) ◦ (π− )−1 )−1 ◦ P (1/z)
1
=
P (1/z)
zn
=
n n−i
.
i=0 ai z

Thus Q is smooth on U , which yields that f is smooth near (0, 0, 1).

Next note that dfp

is singular, if and only of π+ (p) is a root of the complex

n i−1
polynomial P (z) = i=1 ai iz (see the next exercise). But a complex
polynomial has only finitely many roots unless it is identically zero. So,
since by assumption an = o, we conclude that the set of regular values of f

2
consists of S2 minus a finite number of points. In particular, the set of regular
values of f is connected and open. So the locally constant function #f −1 is
constant on the set of regular values of f . Since the number of singular points
of f are finite, #f −1 cannot be zero everywhere, so it is zero nowhere on the
set of regular values of f . This yields that f is onto. In particular, there
exists p ∈ S2 such that f (p) = (0, 0, −1). So P (π+ (p)) = π+ (f (p)) = o.
Exercise 5. Let P : R2 → R2 be a complex polynomial. Show that

dPz (w) = P
(z)w.

More precisely if θz : Tz R2 → R2 denotes the standard isomorphism, then

dPz (w) = θz−1 (P
(z)θz (w)). In particular z is a singular point of P if and only
if z is a root of P
(z)

3.2 Manifolds with boundary

Next we generalize the last proposition, concerning the inverse image of reg-
ular values, to manifolds with boundary. Recall that M is a manifold with
boundary if every point of M has an open neighborhood which is homeomor-
phic to an open subset of the half space

H n := {(x1 , . . . , xn ) ∈ Rn | xn ≥ 0}.

We define the interior, int H n , and boundary, ∂H n , as the subsets of H n

where xn > 0 and xn = 0 respectively. We say that p is an interior point of
M if an open neighborhood of p is homeomorphic to an open neighborhood
in int H n ; otherwise, we say that p is a boundary point.
Exercise 6. Show that if M n is an n-manifold with ∂M = ∅, then ∂M is
an (n − 1)-manifold without boundary.
Similar to the case of manifolds (without boundary), we say that a man-
ifold with boundary is smooth if it admits an atlas such that all the local
charts in that atlas are C ∞ compatible, i.e., if (U, φ) and (V, ψ) are local
charts of M then φ ◦ ψ −1 : ψ(U ∩ V ) → Rn is smooth. Further, it is im-
portant to recall that if A ⊂ M is any subset, then f : A → N is smooth
provided that for every p ∈ A there exists an open neighborhood U of p
in M and a smooth function fp : U → N such that f |U ∩A = fp . For any
p ∈ M , we may define Tp M as the equivalence class of all the half curves

3
which originate or end at p. More explicitly, define Curvesp M as the set of
curves α : [0, ) → M and α : (−, 0] → M such α(0) = p. We say that a pair
of curves α, β ∈ Curvesp M are equivalent provided that there exists a local
chart (U, φ) centered at p such that (φ ◦ α)
(0) = (ψ ◦ β)
(0). Then Tp M is
defined as the collection of equivalence classes Curvesp M/ ∼.

Exercise 7. Check that the above definition for Tp M coincides with the one
we had given earlier whenever p ∈ int M .

Exercise 8. Show that if M n is a manifold with boundary, p ∈ M , and

(U, φ) is any local chart centered at p, then the mapping

Tp M  [α] −→ (φ ◦ α)
(0) ∈ Rn

is a bijection. Thus we may use this map to endow Tp M with the structure
of an n-dimensional vector space.

The differential of any smooth map f : M → N , where M is a manifold

with boundary, is defined as before.

Exercise 9. Let f : H m → M be a smooth map and p ∈ ∂H. Show that if

U is any open neighborhood of p in Rn and f˜: U → M is any smooth map
such that f˜ = f on U ∩ H m , then dfp = df˜p .

Exercise 10. Let M m be a smooth manifold with boundary and f : M m →

R be a smooth map which has 0 as a regular value. Then f −1 ([0, ∞)) is a
smooth m-manifold with boundary and ∂f −1 ([0, ∞)) = f −1 (0).

Our main aim in this section is to prove:

Theorem 11. Let M m be a manifold with boundary and f : M m → N n be a

smooth map. Suppose that q ∈ N is a regular value of both f and f |∂M , and
f −1 (q) = ∅, then f −1 (q) is a smooth m − n manifold with boundary, and
 
∂ f −1 (q) = f −1 (q) ∩ ∂M.

To prove the above theorem we need a couple of more exercises:

Exercise 12. Show that to prove the above theorem it is enough to consider
the case of M m = H m .

4
Exercise 13. Show that if M is a manifold (without boundary) and f : M →
N is a smooth map which has q as a regular value, then, for every p ∈ f −1 (q),
Tp f −1 (q) is the null space of dfp .

Proof of Theorem 11. We know that f −1 (q) ∩ int M is a smooth manifold,

because int M is a manifold (without boundary) and q is a regular value of
f |int M . Thus it remains to consider points p ∈ f −1 (q) ∩ ∂M .
We may suppose M = H m . Then, by definition of smoothness, there
exists, for every p ∈ ∂M = ∂H, an open neighborhood V of p in Rm and
a smooth function f˜: V → N such that f˜ = f on V ∩ H m . Since, by a
previous exercise, df˜p = dfp , it follows that p is a regular point of f˜. Thus,
after replacing V by a smaller neighborhood of p, which we again denote
by V , we may assume that f˜ as no critical points in V . In particular, it
follows that q is a regular value of f˜. So f˜−1 (q) is a smooth submanifold
of V . Now define g : f˜−1 (q) → R by g(x1 , . . . , xn ) = xn . Then g(p) = 0
and V ∩ f −1 (q) = H ∩ f˜−1 (q) = g −1 ([0, ∞)). Thus to complete the proof it
suffices to show that 0 is a regular value of g. Suppose not. Then Tp f˜−1 (q)
is equal to the null space of dgp . But the null space of dgp is a subset of
Tp ∂H. Thus, since Tp f˜−1 (q) and Tp ∂H have the same dimension, it follows
that Tp f˜−1 (q) = Tp ∂H. But Tp f˜−1 (q) is the null space of df˜p which is equal
to the null space of dfp . So the null space of dfp is equal to Tp ∂H, which
contradicts the assumption that p is a regular point of f |∂H .

3.3 Sard’s Theorem, and Brouwer’s Fixed Point The-

orem
Let f : M → N be a smooth map. Sard’s theorem states that almost ev-
ery point q ∈ N is a regular value of f , where “almost every”, or “almost
all”, means except for a set of measure zero. This theorem, whose proof
we postpone for the time being, has great many applications, including the
Brouwer’s fixed point theorem which we prove below.

Exercise 14. Let Γ ⊂ R2 be a smooth simple closed curve, i.e., the image of
a smooth embedding of S1 . For any unit vector u ∈ S1 , the height function
hu : Γ → R is defined as hu (p) = p, u. Use Sard’s theorem to show that,
for almost all u ∈ S2 , hu has a finite number of critical points (Hint: It is
enough to show that for almost every u ∈ S1 there exist only finitely many
tangent lines of Γ which are orthogonal to u).

5
 Our main aim in this section is to use Sard’s theorem to show that
Theorem 15 (Brouwer). For n ≥ 2, any continuous map f : B n → B n
has a fixed point, where B n denotes the n-dimensional closed unit ball in Rn .
The proof is by contradiction and requires the following lemmas:
Lemma 16. If there exists a continuous map f : B n → B n without fixed
points, then there exists a smooth map f˜: B n → B n without fixed points.
Proof. If f has no fixed points, then, since B n is compact, there exists an
 > 0 such that f (p) − p >  for all p ∈ B n . By Wierstrauss approximation
theorem, there exists a smooth map f : B n → Rn such that f (p) − f (p) <
/2 for all p ∈ B n . Let
1
f˜ := f.
1 + /2
Then f˜: B n → B n , because, by the triangle inequality,

f (p) ≤ f (p) + f (p) − f (p) < 1 + /2,

which yields that f˜ ≤ 1. Further note that, again by the triangle inequality,

f˜(p) − f (p) ≤ f˜(p) − f (p) + f (p) − f (p) < /2 + /2 = .

Thus, since f (p) − p > , it follows that

f˜(p) − p ≥ f (p) − p − f˜(p) − f (p) > 0.

So f˜ has no fixed points.

Lemma 17. If there exists a smooth map f : B n → B n without fixed points,
then there exists a smooth map r : B n → Sn−1 such that r(p) = p for all
p ∈ Sn−1 .
Proof. Consider the ray which starts from f (p) and passes through p. Let
r(p) be the intersection of this ray with Sn−1 . To see that r is smooth, note
that the ray may be parametrized by
 
(t) := f (p) + t p − f (p) ,

where t ≥ 0. Solving  (t) = 1 for t and substituting the solution back in (t)
gives an explicit expression for r(p), which one may check to be smooth.

6
Exercise 18. Find the explicit expression for r in the above lemma and
show that r is smooth.

Now we are ready to prove the main result of this section:

Proof of Brouwer’s Theorem. Let r be as in the previous lemma. By Sard’s
theorem, r has a regular value p ∈ Sn−1 . Then r−1 (p) is a 1-dimensional
manifold with boundary which contains p, since r(p) = p. Further recall
that ∂r−1 (p) = r−1 (p) ∩ Sn−1 . But r−1 (p) ∩ Sn−1 = {p} because r is one-to-
one on Sn−1 . Thus ∂r−1 (p) = {p}. But r−1 (p), being the closed subset of a
compact space, is compact, and each component of a compact 1-dimensional
manifold with boundary must have either zero or two boundary points. So
we have a contradiction.
Note that the above proof uses the fact that every compact connected
one dimensional manifold with boundary is either homeomorphic to S1 or
the interval [0, 1]. Can you prove this fact?