D-Math FS 18

Differential Geometry II
Prof. Dr. D.A. Salamon March 27, 2018

Solution 6

1. A parallelizable manifold M is one where the tangent bundle T M is isomorphic to the trivial bundle M × Rm . In other
words, M carries m vector fields that are linearly independent at every point of M .

a) Prove that S 3 is parallelizable.

b) Prove that S 7 is parallelizable.

c) Prove that S 2n is not parallelizable for all n ≥ 1.

Hint: For a) and b), see S 3 , S 7 as the unit quaternions resp. unit octonions. It is a hard theorem that S 0 , S 1 , S 3
and S 7 are the only parallelizable spheres.
Solution:

a) We can see S 3 as the set of unit quaternions: Define

H = {z = x + yi + vj + wk : (x, y, v, w) ∈ R4 }
with relations ij = k , i2 = j2 = k2 = −1 and
S 3 = {z = x + yi + vj + wk ∈ H : z z̄ = x2 + y 2 + v 2 + w2 = 1}.
Here the conjugate is defined as z̄ = x − yi − vj − wk). We can define three linearly
independent vector fields X, Y, Z on S 3 by
X(z) = iz, Y (z) = jz and Z(z) = kz.
This statement is implied by proving that {z, X(z), Y (z), Z(z)} is an orthonormal
basis of R4 for every z ∈ S 3 . In the standard basis of R4 we get
x −y −v −w
 
y x w −v 
M (z) := (z, X(z), Y (z), Z(z)) =  .
 
 v −w x y 
w v −y x

Then we see that M > M = 1 which indeed means that M ∈ O(4). (As M (pN ) = 1,
this even gives SO(4).)
Note that the quaternion multiplication gives S 3 the structure of a Lie group and
every Lie group is parallelizable. (Simply take a basis of left invariant vector fields.
This is exactly what we did by taking X, Y, Z. )

b) We can see S 7 as the set of unit octonions: The octonians are defined as
O = {z = z1 + z2 l : z1 , z2 ∈ H}
with relations l2 = −1, il = −li, jl = −lj and kl = −lk. Then
S 7 = {z = z1 + z2 l ∈ O : zz = (z1 + z2 l)(z̄1 − z2 l) = |z1 |2 + |z2 |2 = 1}.
(Note that z2 l = lz̄2 .) We can define seven linearly independent vector fields on S 7
by
X1 (z) = iz, X2 (z) = jz, X3 (z) = kz, X4 (z) = lz,
X5 (z) = ilz, X6 (z) = jlz, and X7 (z) = klz.

1
D-Math FS 18
Differential Geometry II
Prof. Dr. D.A. Salamon March 27, 2018

This statement is implied by proving that

{z, X1 (z).X2 (z), X3 (z), X4 (z), X5 (z), X6 (z), X7 (z)}

is an orthonormal basis for every z ∈ S 7 . In the standard basis of R8 we get

M (z) := (z, X1 (z), X2 (z), X3 (z), X4 (z), X5 (z), X6 (z), X7 (z))

 
x1 −y1 −v1 −w1 −x2 −y2 −v2 −w2
 y1

x1 w1 −v1 −y2 x2 w2 −v2 

−w1 −v2 −w2
 
 v1 x1 y1 x2 y2 
 
w
 1 v1 −y1 x1 −w2 v2 −y2 x2 
= .

 x2
 y2 v2 w2 x1 −y1 −v1 −w1 

y
 2 −x2 w2 −v2 y1 x1 −w1 v1 

 v2 −w2 −x2 y2 v1 w1 x1 −y1 
 

w2 v2 −y2 −x2 w1 −v1 y1 x1

Then we see that M > M = 1 which indeed means that M ∈ O(8). (As M (pN ) = 1,
this even gives SO(8).)
Note that S 7 with octonian multiplication is not a Lie group since associativity fails.

c) We already know χ(S 2n ) = 2. So by Poincaré–Hopf, there are no non-vanishing

vector fields. Therefore S 2n cannot be parallelizable.
2. Let M be a compact manifold.

a) Prove that there exist a finite collection X1 , X2 , . . . , Xk ∈ Vect(M ) such that {X1 (p), X2 (p), · · · , Xk (p)} spans Tp M
at every p ∈ M .

b) Prove that the number k from a) can be chosen such that k ≤ 2m, where m = dim(M ).
Hint: For b): Start with any large number of vector fields X1 , . . . , Xk and consider the map F : Tp M → Rk whose
j-th coordinate is given by Fj (p, v) := hv, Xj (p)i. What does Sard’s theorem tell us when k > 2m and how can you
use this to construct k − 1 vector fields Y1 , . . . , Yk−1 which still span Tp M at every p ∈ M ?

Solution:

a) We show the following local statement first: For every p ∈ M exists an open neigh-
(p) (p)
borhood p ∈ Up and vector field X1 , . . . , Xm ∈ Vect(M ) such that
(p) (p)
X1 (q), . . . Xm (q) is a basis for Tq M for every q ∈ Up .

Indeed, let ϕp : Vp → Ωp be a chart defined on an open neighborhood p ∈ Vp ⊂ M

with image Ωp ⊂ Rm . Choose a bounded open subset ϕp (p) ∈ Wp ⊂ Ωp with
W p ⊂ Ωp and a cut-off function ρp : Wp → [0, 1] with ρ(x) = 1 for all x ∈ Wp and
such that supp(ρp ) ⊂ Ωp is compact (i.e. ρp vanishes near the boundary). Define

Yj : Ωp → Rm , Yj (x) := ρ(x)ej
(p)
where e1 , . . . , em denotes the standard basis of Rm . We define the vector fields Xj
as the pullback of Yj :
(
(p) 0 q∈/ Vp
Xj (q) := −1
dϕ(q) Yj (ϕ(q)) q ∈ Vp

2
D-Math FS 18
Differential Geometry II
Prof. Dr. D.A. Salamon March 27, 2018

These are clearly smooth vector fields on M and yield a basis of Tq M for all q ∈
Up := ϕ−1 (Wp ).
By compactness of M , there exists finitely many points p1 , . . . , pN such that Up1 , . . . , UpN
(p )
cover M . Then {Xj i } with j = 1, . . . , m and i = 1, . . . , N is a finite collection of
vector fields spanning Tp M at every p ∈ M .

b) Let X1 , . . . , Xk ∈ Vect(M ) be any finite number of vector fields which span Tp M at

every point p ∈ M . Choose a Riemannian metric (or embedding M ⊂ Rn ) to define
the map

F : T M → Rk , F (p, v) = (hv, X1 (p)i, hv, X2 (p)i, . . . , hv, Xk (p)i) .

Suppose k > 2m. Then it follows from Sard’s theorem that F is not surjective, since
dim(T M ) = 2m. Choose ξ ∈ Rk \Image(F ). Note that F is homogeneous in the
sense that F (p, tv) = tF (p, v) for all (p, v) ∈ T M and t ∈ R. This yields the stronger
assertion Rξ ∩ Image(F ) = {0}. We may assume (after relabelling if necessary) that
ξk 6= 0 and define

Yj (p) := ξk Xj (p) − ξi Xk (p), for j = 1, . . . , k − 1.

We claim that Y1 , . . . , Yk−1 span Tp M at every point p ∈ M . If not, then there exists
0 6= v ∈ Tp M such that

0 = hYj (p), vi = ξk hXi (p), vi − ξi hXk (p), vi = ξk Fi (p, v) − ξi Fk (p, v)

and hence
!
ξ1 ξ2 ξk−1
F (p, v) = Fk (p, v), Fk (p, v), . . . , Fk (p, v), Fk (p, v)
ξk ξk ξk
Fk (p, v)
= (ξ1 , . . . , ξk ) .
ξk

Since Rξ ∩ Image(F ) = {0}, it follows F (p, v) = 0 and hence v = 0. This contradic-

tions our assumption and shows that Y1 , . . . , Yk−1 indeed span Tp M .
Repeating the procedure described above, we can decrease the number of vector
fields successively until k ≤ 2m. For k = 2m the argument ceases to work, since F
might be surjective. Nevertheless, in many examples one can do better, e.g. for any
closed hypersurface M m ⊂ Rm+1 the coordinate vector fields project to k = m + 1
vector fields spanning Tp M at every point. Can you find an example where k = 2m
vector fields are needed?
3. Let M be a connected manifold without boundary and let p, q ∈ M . Let K ⊂ M be a compact set containing p, q ∈ K ◦ in
its non empty, connected interior. Then there exists an isotopy ψt : M → M for t ∈ [0, 1] such that ψ0 = id, ψ1 (p) = q and

[
supp({ψt }t∈[0,1] ) := {x ∈ M : ψt (x) 6= x} ⊂ K.
0≤t≤1

We call this an isotopy with compact support in K.

Hint: This is a slight generalization of the homogenisation lemma from the lecture. Have a careful look at the
proof and check how it can be adopted to encompass this case.

3
D-Math FS 18
Differential Geometry II
Prof. Dr. D.A. Salamon March 27, 2018

Solution: Look at the proof of the homogenisation lemma given in class or in Milnor’s
book at page 22. In this proof, we constructed isotopies Ft supported on a small ball in a
chart around any point which sends the origin to any desired point in this ball. Thus say
that two interior points x, y ∈ K ◦ are ‘isotopic’ in K if there is an isotopy supported in
K which sends x to y. By the previous construction in the chart, we see that all points
near y ∈ K ◦ are ‘isotopic’ to y in K as long as we choose the small ball in the interior
of K. "In other words, each isotopy class of points in the interior of K is an open set
and the interior of K is partitioned into disjoint open isotopy classes. But the interior of
K is connected; hence there can only be one isotopy class." (quote from p:23 of Milnor’s
book.)
4. Let M m be a compact, connected manifold without boundary where m ≥ 2.

a) Let ϕ : U → Rm be a chart defined on U ⊂ M with B1 (0) ⊂ ϕ(U ). Prove that there exists X ∈ Vect(M ) which does
not vanish outside of ϕ−1 (B1 (0)) and has only isolated non-degenerate zeros in ϕ−1 (B1 (0)).

b) Let YP: B1 (0) → Rm be a vector field with isolated non-degenerate zeros. Suppose that Y (x) 6= 0 for x ∈ ∂B1 (0)
and x∈Y −1 (0)
ι(Y, p) = 0. Then there exists a nowhere vanishing vector field Z : B1 (0) → Rm which agrees with
Y near the boundary.

c) Prove that if χ(M ) = 0 then there is a vector field X ∈ Vect(M ) with no zeroes.
Hint: For a): Start with any vector field X with isolated non-degenerate zeros. Use the homogeneity lemma of the
previous exercise to move all the zeros into a ball inside a coordinate chart of X.
For b): The Gauss map ∂B1 (0) → S m−1 defined by x 7→ Y (x)/kY (x)k has degree zero under the given assumptions.
Hence it is homotopic to a constant map by the Hopf degree theorem.
For c): Combine a) and b) and make sure that all your modifications produce a smooth vector field.

Solution:

a) Let X be a vector field with isolated non-degenerate zeroes X −1 (0) = {p1 , . . . , pk }

and ϕ : U → Rm be a chart defined on U ⊂ M with B1 (0) ⊂ ϕ(U ). Let ` =
#X −1 (0) ∩ ϕ−1 (B1 (0)). If ` = k, we are done. Else pick q ∈ ϕ−1 (B1 (0)) with X(q) 6=
/ ϕ−1 (B1 (0)) and choose small open neighbourhoods
0, i ∈ {1, . . . , k} such that pi ∈
pj ∈ Vj for j 6= i such that
K := M \(V1 ∪ · · · ∪ Vj−1 ∪ Vj+1 ∪ · · · ∪ Vk )
is a compact connected subset with pj , q ∈ K ◦ . By the previous exercise, there exists
an isotoply {ψt } supported in K with ψ1 (pj ) = q. Then Y = ψ∗ X is a vector field on
M such that Y −1 (0) = {p1 , . . . , pk , q} \ {pi }. Hence, #Y −1 (0) ∩ ϕ−1 (B1 (0)) = ` + 1.
We can repeat this argument until all the zeroes are in ϕ−1 (B1 (0)).

b) Since Y has only isolated zeros, it follows from the Hopf lemma that the signed count
m−1
→
P
x∈Y −1 (0) ι(Y, p) agrees with the degree of the Gauss map f : ∂B1 (0) = S
m−1
S defined by x 7→ Y (x)/kY (x)k (see Exercise 2 on Exercise Sheet 4). Hence this
has degree zero and it is homotopic to a constant map by the Hopf degree theorem.
Say Ht : S m−1 → S m−1 such that H1 = f and H0 ≡ pN . Take  > 0 such that
B1 (0) \ B1− (0) ∩ Y −1 (0) = ∅. Choose a cut off function ρ : R → R such that ρ = 0
near 1 − , ρ = 1 near 1 and ρ is strictly increasing. Define Z ∈ Vect(B1 (0)) by

Z(p)= pN for p ∈ B1− (0),
Z(rσ) = (ρ(r)|Y (σ)| + (1 − ρ(r)))H
ρ(r) (σ) for rσ ∈ B1 (0) \ B1− (0),

where r ∈ [1 − , 1), σ ∈ S m−1 . This is a smooth vector field which is non vanishing
and agrees with Y near the boundary.

4
D-Math FS 18
Differential Geometry II
Prof. Dr. D.A. Salamon March 27, 2018

c) Let ϕ : U → Rm be a chart defined on U ⊂ M with B1 (0) ⊂ ϕ(U ). Take as in a)

X ∈ Vect(M ) which does not vanish outside of ϕ−1 (B1 (0)) and has only isolated
non-degenerate zeros in ϕ−1 (B1 (0)). Then Y = ϕ∗ X|B1 (0) fulfills the assumptions in
b) due to Poincaré–Hopf theorem and χ(M ) = 0. Therefore we can find a nowhere
vanishing vector field Z : B1 (0) → Rm which agrees with Y near the boundary. Now
define the vector X̃ ∈ Vect(M ) by

X(p), / ϕ−1 (B1 (0)),
for p ∈
X̃(p) =
(ϕ−1 )∗ Z, for p ∈ ϕ−1 (B1 (0)).

By construction, X̃ is smooth and has no zeroes.

5. Let f : M → R be a Morse function as in Exercise 2, Sheet 3. The Morse Lemma asserts that for every p ∈ Crit(f ),
there is a chart ϕp : Up ⊂ M → Rm with ϕp (p) = 0 such that

f ◦ ϕ−1 2 2 2 2
p (x) = f (p) − x1 − · · · − xk + xk+1 + · · · + xm .

Here k := µ(f, p) is determined as the dimension of the negative eigenspace of the Hessian Hp f and it is called the Morse
index of f at p.
Let g be a Riemannian metric on M . The gradient vector field X = ∇g f of a smooth function f : M → R is defined by

g(X(p), p̂) = df (p)p̂, for all p ∈ M and p̂ ∈ Tp M .

Prove the following properties for the gradient field of a Morse function f .

a) For every metric g the gradiant X = ∇g f is a vector field with isolated, non-degenerate zeroes.

b) For every p ∈ Crit(f ) and any two Riemannian metrics g1 , g2 the indices ι(∇g0 f, p) = ι(∇g1 f, p) agree.

c) For every p ∈ Crit(f ) it holds ι(∇g f, p) = (−1)µ(f,p) .

Solution:

a) The linearization of the gradient vector field ∇f at a critical point p is the linear
map
∇(∇f )(p) : Tp M → Tp M, p̂ →
7 ∇p̂ (∇f )(p).
We verify that h∇p̂ (∇f ), ·i = Hp f (p̂, ·) is related to the Hessian defined in Exercise
2 of Exercise Sheet 3. Indeed,

Hp f (X, Y ) = LX df (p)Y − df (p)∇X Y

= LX h∇f (p), Y (p)i − h∇f (p), ∇X Y (p)i
= h∇X ∇f (p), Y (p)i.

Here we used that the Levi-Civita connection ∇ is Riemannian. Hence Hp f is

nondegenerate if and only if ∇(∇f )(p) is an isomorphism. This shows that every
zero is non-degenerate.
Non-degeneracy implies isolated. This follows from the following more general state-
ment. Surjectivity of ∇(∇f )(p) at every point p ∈ M with ∇f (p) = 0 is equivalent
to the statement that ∇f : M → T M intersects the zero section M × {0} ⊂ T M
transversely. Therefore the intersection {p ∈ M | ∇f (p) = 0} is a 0-dimensional
manifold and hence a discrete set of points.

5
D-Math FS 18
Differential Geometry II
Prof. Dr. D.A. Salamon March 27, 2018

b) The set of Riemannian metrics is convex. Hence gt := (1 − t)g0 + tg1 is a Riemannian

metric for 0 ≤ t ≤ 1. This defines a homotopy of the gradient vector fields defined
by ∇gt f . The zeros for each of these vector field are the same, since ∇gt f (p) = 0
if and only if df (p) = 0 and the later statement does not depend on the metric. It
follows from the homotopy invariance of the degree, that all the zeros have the same
index for ∇g0 f and ∇g1 f .

c) Define h : Rm → R by

h(x) = f (p) − x21 − · · · − x2k + x2k+1 + · · · + x2m .

Its gradient for the standard metric is the vector field

∇h(x) : Rm → Rm : ∇h(x) := 2(−x1 , . . . , −xk , xk+1 , . . . , xm ).

Using the Morse Lemma, the index of ∇f at a critical point p of index k := µ(f, p)
agrees with the index of ∇h at the origin. This is defined as degree of the Gauss
map S m−1 → S m−1 given by

(x1 , . . . , xm ) 7→ (−x1 , . . . , −xk , xk+1 , . . . , xm ).

This is a composition of k reflections and hence has degree (−1)k . Hence, it follows
ι(∇g f, p) = (−1)µ(f,p) for any metric g by part b).
6. a) Show that fn : RP n → R defined by
Pn
j=1
jx2j
fn ([x0 : x1 : . . . : xn ]) := Pn 2
j=0
xj

is a Morse function on RP n . Determine all the critical points of fn and their Morse index.

b) Show that χ(RP n ) = 0 when n is odd and χ(RP n ) = 1 when n is even.

c) Show that gn : CP n → R defined by

Pn
j=1
j|zj |2
gn ([z0 : z1 : . . . : zn ]) := Pn .
2
|zj |
j=0

is a Morse function on CP n . Determine all the critical points of gn and their Morse index.

d) Show that χ(CP n ) = n + 1 for n ≥ 1.

Hint: Use the gradient vector fields of the Morse functions to compute the Euler characteristic in part b) and d)
(see Exercise 5 above)

Solution:

a) We write fn in the standard charts for RP n . Define hi : Rn → R by

hi (x1 , . . . , xn ) := fn ([x1 : · · · : xi−1 : 1 : xi : · · · : xn ])

P  P 
i
j=1 (j − 1)x2j + i + j=i+1 jx2j
= Pn
1+ j=1 x2j
The partial derivatives ∂k hi is for k ≤ i given by
 Pn  hP  P i
i
1+ j=1 x2j 2(k − 1)xj − 2xk j=1 (j − 1)x2j + i + j=i+1 jx2j
∂k hi (x) = Pn 2 2
(1 + j=1 xj )

6
D-Math FS 18
Differential Geometry II
Prof. Dr. D.A. Salamon March 27, 2018

and for k > i by

 Pn  hP  P i
2 i 2
1+ j=1 xj 2kxj − 2xk j=1 (j − 1)xj +i+ j=i+1 jx2j
∂k hi (x) = Pn 2 2
.
(1 + j=1 xj )

It follows directly that 0 is a critical value of hi . Inductively, one can show that
∂n hi (x) = 0 implies xn = 0 and then ∂n−1 hi (x) = 0 yields xn−1 = 0, etc. Therefore
0 is the only critical value of hi .
When calculating the second partial derivatives, we see that most terms vanish at
the origin, and get as final expression
(
2(k − 1)δk` − 2iδk` for k ≤ i,
∂` ∂k hi (0) =
2kδk` − 2iδk` for k > i.
Hence H0 (hi ) = diag (−2i, −2i + 2, . . . , −2, 2, . . . , 2n − 2j). This shows that the
origin is a nondegenerate critical point for hi with Morse index j − 1.
For the function fn : RP n → R it follows that the critical points are
p0 = [1 : 0 : . . . : 0], p1 = [0 : 1 : 0 : . . . : 0], ..., pn = [0 : . . . : 0 : 1].
They are all non-degenerate and have Morse index µ(fn , pi ) = i.

b) It follows from Exercise 5 and part a) that

m m
(
n
X
µ(fn ,pi )
X
i 1 for n even,
χ(RP ) = (−1) = (−1) =
i=0 i=0
0 for n odd.

Note that this is half the Euler characteristic of S n , which hints at another proof
using the Poincaré–Hopf theorem.

c) We proceed similarly to part a). To make things more explicit, we write zj =

xj + iyj and use real notation. In local coordinates the function gn is represented by
h̃i : R2n → Rn with
P  P 
i
j=1 (j − 1)(x2j + yj2 ) + i + 2 2
j=i+1 j(xj + yj )
h̃i (x, y) = Pn
1+ j=1 x2j + yj2
q q 
Since h̃i (x, y) = hi x21 + y12 , . . . , x2n + yn2 , it follows from the chain rule and part
a) that the only critical point of h̃i is the origin. The second partial derivatives are:
(
2(k − 1) − 2i for k ≤ i,
∂xk xk hi (0) = ∂yk yk hi (0) =
2k − 2i for k > i.
and all other mixed partial derivative vanish. Hence the Hessian is again a diagonal
matrix with precisely 2i negative entries.
It follows that gn : CP n → R has the critical points
p0 = [1 : 0 : . . . : 0], p1 = [0 : 1 : 0 : . . . : 0], ..., pn = [0 : . . . : 0 : 1].
They are all non-degenerate and have Morse index µ(fn , pi ) = 2i.

7
D-Math FS 18
Differential Geometry II
Prof. Dr. D.A. Salamon March 27, 2018

d) It follows from Exercise 5 and part c) that

m m
χ(CP n ) = (−1)µ(gn ,pi ) = (−1)2i = n + 1.
X X

i=0 i=0